[ { "path": ".github/ISSUE_TEMPLATE/config.yml", "content": "blank_issues_enabled: false\ncontact_links:\n - name: Ask a question or discuss a topic\n url: https://discord.com/invite/BkMR2NUsjU\n about: Ask questions and discuss with other community members\n - name: Join Yao community\n url: https://yaoapps.com/community\n about: Join the community to get updates and news\n" }, { "path": ".github/ISSUE_TEMPLATE/issue_report.md", "content": "---\nname: \"Report an issue\"\nabout: \"Report an issue to help us improve\"\nlabels: \"\"\nassignees: \"\"\n---\n\n## Description\n\n## Context\n\n- **Yao Version( yao version --all )**:\n- **Platform**:\n" }, { "path": ".github/actions/setup-db/Dockerfile", "content": "FROM docker:latest\n\n\nCOPY entrypoint.sh /entrypoint.sh\nRUN chmod +x /entrypoint.sh\nENTRYPOINT [\"/entrypoint.sh\"]" }, { "path": ".github/actions/setup-db/action.yml", "content": "inputs:\n kind:\n description: \"Chose the kind of database (MySQL8.0, MySQL5.7, Postgres9.6, Postgres14.0, SQLite3)\"\n required: true\n db:\n description: \"The name of database\"\n required: false\n default: \"github\"\n port:\n description: \"The port of database\"\n required: false\n user:\n description: \"The user of database\"\n required: false\n default: \"github\"\n password:\n description: \"The passowrd of database\"\n required: false\n default: \"123456\"\nruns:\n using: \"docker\"\n image: \"Dockerfile\"\n" }, { "path": ".github/actions/setup-yao/action.yml", "content": "name: \"Setup Yao Build Environment\"\ndescription: \"Checkout dependency repos, setup Go toolchain, and install build tools (v1.0.0)\"\n\ninputs:\n go-version:\n description: \"Go version to install\"\n default: \"1.25\"\n repo-kun:\n description: \"Kun repository (owner/repo)\"\n required: true\n repo-xun:\n description: \"Xun repository (owner/repo)\"\n required: true\n repo-gou:\n description: \"Gou repository (owner/repo)\"\n required: true\n checkout-app:\n description: \"Checkout yao-dev-app (demo application for tests)\"\n default: \"true\"\n checkout-init:\n description: \"Checkout yao-init (for Yao server startup in CI)\"\n default: \"false\"\n apple-private-key:\n description: \"Apple private key content for OAuth certs (optional)\"\n default: \"\"\n\nruns:\n using: \"composite\"\n steps:\n # -- Dependency repositories --\n - name: Checkout Kun\n uses: actions/checkout@v4\n with:\n repository: ${{ inputs.repo-kun }}\n path: kun\n\n - name: Checkout Xun\n uses: actions/checkout@v4\n with:\n repository: ${{ inputs.repo-xun }}\n path: xun\n\n - name: Checkout Gou\n uses: actions/checkout@v4\n with:\n repository: ${{ inputs.repo-gou }}\n path: gou\n\n - name: Checkout V8Go\n uses: actions/checkout@v4\n with:\n repository: yaoapp/v8go\n path: v8go\n\n - name: Unzip libv8\n shell: bash\n run: |\n for file in $(find ./v8go -name \"libv8*.zip\"); do\n dir=$(dirname \"$file\")\n echo \"Extracting $file to $dir\"\n unzip -o -d \"$dir\" \"$file\"\n rm -rf \"$dir/__MACOSX\"\n done\n\n - name: Checkout Demo App\n if: ${{ inputs.checkout-app == 'true' }}\n uses: actions/checkout@v4\n with:\n repository: yaoapp/yao-dev-app\n path: app\n\n - name: Checkout Extension\n uses: actions/checkout@v4\n with:\n repository: yaoapp/yao-extensions-dev\n path: extension\n\n - name: Checkout yao-init\n if: ${{ inputs.checkout-init == 'true' }}\n uses: actions/checkout@v4\n with:\n repository: yaoapp/yao-init\n path: yao-init\n\n # -- Move all dependencies to parent directory (Go workspace layout) --\n - name: Move Dependencies\n shell: bash\n run: |\n mv kun ../\n mv xun ../\n mv gou ../\n mv v8go ../\n [ -d app ] && mv app ../\n mv extension ../\n [ -d yao-init ] && mv yao-init ../\n\n # -- Setup Apple Private Key (if provided) --\n - name: Setup Apple Private Key\n if: ${{ inputs.apple-private-key != '' }}\n shell: bash\n run: |\n mkdir -p ../app/openapi/certs/apple\n echo \"${{ inputs.apple-private-key }}\" > ../app/openapi/certs/apple/signin_client_secret_key.p8\n\n # -- Go toolchain --\n - name: Setup Go ${{ inputs.go-version }}\n uses: actions/setup-go@v5\n with:\n go-version: ${{ inputs.go-version }}\n\n - name: Setup Go Tools\n shell: bash\n run: make tools\n" }, { "path": ".github/codesign/entitlements.plist", "content": "\n\n\n\n com.apple.security.cs.allow-jit\n \n com.apple.security.cs.allow-unsigned-executable-memory\n \n\n\n" }, { "path": ".github/env/sandbox-v2.env", "content": "# ============================================================\n# Yao CI Environment — sandbox-v2 (v1.0.0)\n# Loaded via: cat .github/env/sandbox-v2.env >> $GITHUB_ENV\n# ============================================================\n\n# ========================================\n# Yao Runtime (YAO_ prefix, read by Yao)\n# ========================================\nYAO_HOST=0.0.0.0\nYAO_PORT=5099\nYAO_GRPC_HOST=0.0.0.0\nYAO_GRPC_PORT=9099\nYAO_DB_DRIVER=sqlite3\nYAO_SESSION=memory\nYAO_ENV=development\n\n# ========================================\n# CI Test Parameters (YAO_CI_ prefix)\n# ========================================\n\n# -- Network --\nYAO_CI_BRIDGE_IP=172.17.0.1\n\n# -- Yao service ports (tests read these, not YAO_PORT/YAO_GRPC_PORT) --\nYAO_CI_HTTP_PORT=5099\nYAO_CI_GRPC_PORT=9099\nYAO_CI_URL=http://127.0.0.1:5099\nYAO_CI_GRPC=127.0.0.1:9099\n\n# -- OAuth token generation (ci-token tool) --\nYAO_CI_OAUTH_SUBJECT=ci-test-user\nYAO_CI_OAUTH_USER_ID=ci-test-user\nYAO_CI_OAUTH_TEAM_ID=ci-test-team\nYAO_CI_OAUTH_SCOPE=tai:tunnel\nYAO_CI_OAUTH_TTL=24h\n\n# ========================================\n# Tai Instances\n#\n# Connection modes:\n# tai-local → DIRECT (--direct, Yao dials Tai gRPC directly)\n# tai-docker → TUNNEL (default, Tai dials Yao gRPC, reverse tunnel)\n# tai-k8s → TUNNEL (same as above, with K8s runtime)\n# tai-hostexec → TUNNEL (same as above, no container runtime)\n# ========================================\n\n# -- tai-local (DIRECT mode, auto-detect Docker via /var/run/docker.sock) --\n# Yao dials tai-local gRPC directly, so gRPC port must be reachable.\n# Docker proxy on 12376 (not default 12375) to avoid conflict with tai-docker.\nYAO_CI_TAI_LOCAL_HOST=127.0.0.1\nYAO_CI_TAI_LOCAL_GRPC_PORT=19103\nYAO_CI_TAI_LOCAL_HTTP_PORT=8102\nYAO_CI_TAI_LOCAL_VNC_PORT=16083\nYAO_CI_TAI_LOCAL_DOCKER_PORT=12376\nYAO_CI_TAI_LOCAL_GRPC=127.0.0.1:19103\nYAO_CI_TAI_LOCAL_DOCKER_API=tcp://127.0.0.1:12376\n\n# -- tai-docker (TUNNEL mode, explicit Docker API proxy) --\n# Tunnel: Tai connects to Yao gRPC. Sandbox connects to Yao, traffic forwarded via tunnel.\n# gRPC port used only for Tai's own listener; Yao accesses via tunnel, not direct dial.\nYAO_CI_TAI_DOCKER_HOST=127.0.0.1\nYAO_CI_TAI_DOCKER_GRPC_PORT=19100\nYAO_CI_TAI_DOCKER_HTTP_PORT=8099\nYAO_CI_TAI_DOCKER_VNC_PORT=16080\nYAO_CI_TAI_DOCKER_API_PORT=12375\nYAO_CI_TAI_DOCKER_API=tcp://127.0.0.1:12375\n\n# -- tai-k8s (TUNNEL mode, K8s API proxy via k3d) --\n# K8s proxy on 16444 (not 16443) because k3d --api-port already binds 16443.\n# TAI_K8S_UPSTREAM points to k3d at 127.0.0.1:16443; proxy exposes on 16444.\nYAO_CI_TAI_K8S_HOST=127.0.0.1\nYAO_CI_TAI_K8S_GRPC_PORT=19101\nYAO_CI_TAI_K8S_HTTP_PORT=8100\nYAO_CI_TAI_K8S_VNC_PORT=16081\nYAO_CI_TAI_K8S_API_PORT=16444\nYAO_CI_K3D_API_PORT=16443\n\n# -- tai-hostexec (TUNNEL mode, no container runtime, HostExec only) --\nYAO_CI_TAI_HOSTEXEC_HOST=127.0.0.1\nYAO_CI_TAI_HOSTEXEC_GRPC_PORT=19102\nYAO_CI_TAI_HOSTEXEC_HTTP_PORT=8101\nYAO_CI_TAI_HOSTEXEC_VNC_PORT=16082\n\n# ========================================\n# Sandbox V2 addresses (used by test code)\n# ========================================\nYAO_CI_SANDBOX_LOCAL_ADDR=tai://127.0.0.1:19103\nYAO_CI_SANDBOX_DOCKER_ADDR=tai://127.0.0.1:19100\nYAO_CI_SANDBOX_K8S_ADDR=tai://127.0.0.1:19101\nYAO_CI_SANDBOX_IMAGE=yaoapp/tai-sandbox-test:latest\n\n# ========================================\n# HostExec addresses\n# ========================================\nYAO_CI_HOSTEXEC_LOCAL_ADDR=127.0.0.1:19103\nYAO_CI_HOSTEXEC_DOCKER_ADDR=127.0.0.1:19100\nYAO_CI_HOSTEXEC_K8S_ADDR=127.0.0.1:19101\nYAO_CI_HOSTEXEC_ONLY_ADDR=127.0.0.1:19102\n\n# -- Tunnel --\nYAO_CI_TUNNEL=true\n\n# ========================================\n# Database (MySQL / PostgreSQL / SQLite)\n# ========================================\nMYSQL_TEST_HOST=127.0.0.1\nMYSQL_TEST_PORT=3308\nMYSQL_TEST_USER=test\nMYSQL_TEST_PASS=123456\n\nPG_TEST_HOST=127.0.0.1\nPG_TEST_PORT=5432\nPG_TEST_USER=test\nPG_TEST_PASS=123456\n\nSQLITE_DB=./app/db/yao.db\n\n# ========================================\n# Legacy variable mapping (migrate later)\n# ========================================\nTAI_TEST_HOST=127.0.0.1\nTAI_TEST_DOCKER=tcp://127.0.0.1:12375\nTAI_TEST_GRPC_PORT=19100\nTAI_TEST_HTTP_PORT=8099\nTAI_TEST_VNC_PORT=16080\nTAI_TEST_DOCKER_PORT=12375\nTAI_TEST_K8S_HOST=127.0.0.1\nTAI_TEST_K8S_PORT=16444\nTAI_TEST_K8S_GRPC_PORT=19101\nTAI_TEST_K8S_HTTP_PORT=8100\nTAI_TEST_K8S_VNC_PORT=16081\nTAI_TEST_HOST_IP=172.17.0.1\nTAI_TEST_TUNNEL=true\nTAI_TEST_YAO_URL=http://127.0.0.1:5099\nTAI_TEST_YAO_GRPC=127.0.0.1:9099\nSANDBOX_TEST_LOCAL_ADDR=tai://127.0.0.1:19103\nSANDBOX_TEST_REMOTE_ADDR=tai://127.0.0.1:19100\nSANDBOX_TEST_K8S_REMOTE_ADDR=tai://127.0.0.1:19101\nSANDBOX_TEST_HOSTEXEC_ADDR=tai://127.0.0.1:19102\nSANDBOX_TEST_IMAGE=yaoapp/tai-sandbox-test:latest\nDOCKER_BRIDGE_IP=172.17.0.1\n" }, { "path": ".github/workflows/build-docker.yml", "content": "name: Build and push docker images\n\non:\n # push:\n # branches: [main]\n # paths:\n # - \".github/workflows/docker.yml\"\n workflow_run:\n workflows: [\"Build Linux Artifacts\"]\n types:\n - completed\n\nenv:\n VERSION: 0.10.5\njobs:\n build:\n if: ${{ github.event.workflow_run.conclusion == 'success' }}\n runs-on: ubuntu-latest\n steps:\n - name: Checkout\n uses: actions/checkout@v4\n\n - name: Get Version\n run: |\n echo VERSION=$(cat share/const.go |grep 'const VERSION' | awk '{print $4}' | sed \"s/\\\"//g\") >> $GITHUB_ENV\n\n - name: Check Version\n run: echo $VERSION\n\n - name: Set up QEMU\n uses: docker/setup-qemu-action@v3\n\n - name: Set up Docker Buildx\n uses: docker/setup-buildx-action@v3\n\n - name: Login to DockerHub\n uses: docker/login-action@v3\n with:\n username: ${{ secrets.DOCKER_USER }}\n password: ${{ secrets.DOCKER_TOKEN }}\n\n - name: Build Development\n uses: docker/build-push-action@v6\n env:\n DOCKER_CONTENT_TRUST: 1\n with:\n context: ./docker/development\n platforms: linux/amd64\n build-args: |\n VERSION=${{ env.VERSION }}\n ARCH=amd64\n push: true\n tags: yaoapp/yao:${{ env.VERSION }}-amd64-dev\n\n - name: Build Development Arm64\n uses: docker/build-push-action@v6\n env:\n DOCKER_CONTENT_TRUST: 1\n with:\n context: ./docker/development\n platforms: linux/arm64\n build-args: |\n VERSION=${{ env.VERSION }} \n ARCH=arm64\n push: true\n tags: yaoapp/yao:${{ env.VERSION }}-arm64-dev\n\n - name: Build Production\n uses: docker/build-push-action@v6\n env:\n DOCKER_CONTENT_TRUST: 1\n with:\n context: ./docker/production\n platforms: linux/amd64\n build-args: |\n VERSION=${{ env.VERSION }} \n ARCH=amd64\n push: true\n tags: yaoapp/yao:${{ env.VERSION }}-amd64\n\n - name: Build Production Arm64\n uses: docker/build-push-action@v6\n env:\n DOCKER_CONTENT_TRUST: 1\n with:\n context: ./docker/production\n platforms: linux/arm64\n build-args: |\n VERSION=${{ env.VERSION }} \n ARCH=arm64\n push: true\n tags: yaoapp/yao:${{ env.VERSION }}-arm64\n\n - name: Build Production Slim\n uses: docker/build-push-action@v6\n env:\n DOCKER_CONTENT_TRUST: 1\n with:\n context: ./docker/production-slim\n platforms: linux/amd64\n build-args: |\n VERSION=${{ env.VERSION }} \n ARCH=amd64\n push: true\n tags: yaoapp/yao:${{ env.VERSION }}-amd64-slim\n\n - name: Build Production Slim Arm64\n uses: docker/build-push-action@v6\n env:\n DOCKER_CONTENT_TRUST: 1\n with:\n context: ./docker/production-slim\n platforms: linux/arm64\n build-args: |\n VERSION=${{ env.VERSION }} \n ARCH=arm64\n push: true\n tags: yaoapp/yao:${{ env.VERSION }}-arm64-slim\n" }, { "path": ".github/workflows/build-linux.yml", "content": "name: Build Linux Artifacts\n\non:\n workflow_dispatch:\n inputs:\n tags:\n description: \"Version tags\"\n\njobs:\n build:\n runs-on: \"ubuntu-latest\"\n container:\n image: yaoapp/yao-build:1.0.0\n\n env:\n CF_ACCESS_KEY_ID: ${{ secrets.CF_ACCESS_KEY_ID }}\n CF_SECRET_ACCESS_KEY: ${{ secrets.CF_SECRET_ACCESS_KEY }}\n R2_BUCKET: ${{ secrets.R2_BUCKET }}\n R2_ACCOUNT_ID: ${{ secrets.R2_ACCOUNT_ID }}\n\n steps:\n - name: Configure R2 For Cloudflare\n run: |\n aws configure set aws_access_key_id $CF_ACCESS_KEY_ID\n aws configure set aws_secret_access_key $CF_SECRET_ACCESS_KEY\n aws configure set default.region us-east-1 # Update with your R2 region if different\n aws configure set default.s3.signature_version s3v4\n aws configure set default.s3.endpoint_url https://$R2_ACCOUNT_ID.r2.cloudflarestorage.com\n aws --version\n\n - name: Build\n run: |\n export PATH=$PATH:/github/home/go/bin\n /app/build.sh\n ls -l /data\n\n - name: Archive production artifacts\n uses: actions/upload-artifact@v4\n with:\n name: yao-linux\n path: |\n /data/*\n\n - name: Push To R2 Cloudflare\n run: |\n for file in /data/*; do\n aws s3 cp $file s3://$R2_BUCKET/archives/ --endpoint-url https://$R2_ACCOUNT_ID.r2.cloudflarestorage.com\n done\n" }, { "path": ".github/workflows/build-macos.yml", "content": "name: Build MacOS Artifacts\n\non:\n workflow_dispatch:\n inputs:\n tags:\n description: \"Version tags\"\n\nenv:\n VERSION: 1.0.0\n\njobs:\n build:\n strategy:\n matrix:\n go: [\"1.25\"]\n runs-on: \"macos-latest\"\n steps:\n - name: Setup Node.js\n uses: actions/setup-node@v4\n with:\n node-version: 18\n\n - name: Install pnpm\n run: npm install -g pnpm\n\n - name: Setup Cache\n uses: actions/cache@v4\n with:\n path: |\n ~/.cache/go-build\n ~/go/pkg/mod\n key: ${{ runner.os }}-go-${{ hashFiles('**/go.sum') }}\n restore-keys: |\n ${{ runner.os }}-go-\n\n - name: Checkout Kun\n uses: actions/checkout@v4\n with:\n repository: yaoapp/kun\n path: kun\n\n - name: Checkout Xun\n uses: actions/checkout@v4\n with:\n repository: yaoapp/xun\n path: xun\n\n - name: Checkout Gou\n uses: actions/checkout@v4\n with:\n repository: yaoapp/gou\n path: gou\n\n - name: Checkout V8Go\n uses: actions/checkout@v4\n with:\n repository: yaoapp/v8go\n path: v8go\n\n - name: Unzip libv8\n run: |\n files=$(find ./v8go -name \"libv8*.zip\")\n for file in $files; do\n dir=$(dirname \"$file\") # Get the directory where the ZIP file is located\n echo \"Extracting $file to directory $dir\"\n unzip -o -d $dir $file\n rm -rf $dir/__MACOSX\n done\n\n - name: Checkout CUI v1.0\n # ** XGEN will be renamed to CUI in the feature. and move to the new repository. **\n # ** new repository: https://github.com/YaoApp/cui.git **\n uses: actions/checkout@v4\n with:\n repository: yaoapp/cui\n path: cui-v1.0\n\n - name: Checkout Yao-Init\n uses: actions/checkout@v4\n with:\n repository: yaoapp/yao-init\n path: yao-init\n\n - name: Move Kun, Xun, Gou, UI, V8Go\n run: |\n mv kun ../\n mv xun ../\n mv gou ../\n mv v8go ../\n mv cui-v1.0 ../\n mv yao-init ../\n rm -f ../cui-v1.0/packages/setup/vite.config.ts.*\n ls -l .\n ls -l ../\n ls -l ../cui-v1.0/packages/setup/\n\n - name: Checkout Code\n uses: actions/checkout@v4\n\n - name: Setup Go ${{ matrix.go }}\n uses: actions/setup-go@v5\n with:\n go-version: ${{ matrix.go }}\n\n - name: Setup Go Tools\n run: |\n make tools\n\n - name: Get Version\n run: |\n echo VERSION=$(cat share/const.go |grep 'const VERSION' | awk '{print $4}' | sed \"s/\\\"//g\") >> $GITHUB_ENV\n\n - name: Make Artifacts MacOS\n run: |\n make artifacts-macos\n\n - name: Install Certificates\n env:\n KEYCHAIN_PASSWORD: ${{ secrets.KEYCHAIN_PASSWORD }}\n\n run: |\n mkdir -p certs\n echo \"${{ secrets.APPLE_DEVELOPERIDG2CA }}\" | base64 --decode > certs/DeveloperIDG2CA.cer\n echo \"${{ secrets.APPLE_DISTRIBUTION }}\" | base64 --decode > certs/distribution.cer\n echo \"${{ secrets.APPLE_PRIVATE_KEY }}\" | base64 --decode > certs/private_key.p12\n security verify-cert -c certs/DeveloperIDG2CA.cer\n security verify-cert -c certs/distribution.cer\n\n - name: Import Certificates\n run: |\n KEYCHAIN_PATH=$RUNNER_TEMP/app-signing.keychain-db\n\n # create temporary keychain\n security create-keychain -p \"$KEYCHAIN_PASSWORD\" $KEYCHAIN_PATH\n security set-keychain-settings -lut 21600 $KEYCHAIN_PATH\n security unlock-keychain -p \"$KEYCHAIN_PASSWORD\" $KEYCHAIN_PATH\n\n # import certificate to keychain\n security import ./certs/DeveloperIDG2CA.cer -k $KEYCHAIN_PATH -T /usr/bin/codesign\n security import ./certs/distribution.cer -k $KEYCHAIN_PATH -T /usr/bin/codesign\n\n # import private key to keychain\n security import ./certs/private_key.p12 -k $KEYCHAIN_PATH -P \"${{ secrets.APPLE_PRIVATE_KEY_PASSWORD }}\" -T /usr/bin/codesign\n\n security list-keychain -d user -s $KEYCHAIN_PATH\n\n - name: Sign Artifacts\n run: |\n codesign --deep --force --verbose --timestamp --sign \"Developer ID Application: ${{ secrets.APPLE_SIGN }}\" dist/release/yao-$VERSION-darwin-arm64\n codesign --deep --force --verbose --timestamp --sign \"Developer ID Application: ${{ secrets.APPLE_SIGN }}\" dist/release/yao-$VERSION-darwin-amd64\n codesign --deep --force --verbose --timestamp --sign \"Developer ID Application: ${{ secrets.APPLE_SIGN }}\" dist/release/yao-$VERSION-darwin-arm64-prod\n codesign --deep --force --verbose --timestamp --sign \"Developer ID Application: ${{ secrets.APPLE_SIGN }}\" dist/release/yao-$VERSION-darwin-amd64-prod\n\n - name: Verify Signature\n run: |\n codesign --verify --deep --strict --verbose=2 dist/release/yao-$VERSION-darwin-arm64\n codesign --verify --deep --strict --verbose=2 dist/release/yao-$VERSION-darwin-amd64\n codesign --verify --deep --strict --verbose=2 dist/release/yao-$VERSION-darwin-arm64-prod\n codesign --verify --deep --strict --verbose=2 dist/release/yao-$VERSION-darwin-amd64-prod\n\n - name: Send to Apple Notary Service\n run: |\n zip -r dist/release/yao-$VERSION-darwin-arm64.zip dist/release/yao-$VERSION-darwin-arm64\n zip -r dist/release/yao-$VERSION-darwin-amd64.zip dist/release/yao-$VERSION-darwin-amd64\n zip -r dist/release/yao-$VERSION-darwin-arm64-prod.zip dist/release/yao-$VERSION-darwin-arm64-prod\n zip -r dist/release/yao-$VERSION-darwin-amd64-prod.zip dist/release/yao-$VERSION-darwin-amd64-prod\n xcrun notarytool submit dist/release/yao-$VERSION-darwin-arm64.zip --apple-id \"${{ secrets.APPLE_ID }}\" --team-id \"${{ secrets.APPLE_TEAME_ID }}\" --password \"${{ secrets.APPLE_APP_SPEC_PASS }}\" --output-format json\n xcrun notarytool submit dist/release/yao-$VERSION-darwin-amd64.zip --apple-id \"${{ secrets.APPLE_ID }}\" --team-id \"${{ secrets.APPLE_TEAME_ID }}\" --password \"${{ secrets.APPLE_APP_SPEC_PASS }}\" --output-format json\n xcrun notarytool submit dist/release/yao-$VERSION-darwin-arm64-prod.zip --apple-id \"${{ secrets.APPLE_ID }}\" --team-id \"${{ secrets.APPLE_TEAME_ID }}\" --password \"${{ secrets.APPLE_APP_SPEC_PASS }}\" --output-format json\n xcrun notarytool submit dist/release/yao-$VERSION-darwin-amd64-prod.zip --apple-id \"${{ secrets.APPLE_ID }}\" --team-id \"${{ secrets.APPLE_TEAME_ID }}\" --password \"${{ secrets.APPLE_APP_SPEC_PASS }}\" --output-format json\n rm -f dist/release/yao-$VERSION-darwin-arm64.zip\n rm -f dist/release/yao-$VERSION-darwin-amd64.zip\n rm -f dist/release/yao-$VERSION-darwin-arm64-prod.zip\n rm -f dist/release/yao-$VERSION-darwin-amd64-prod.zip\n\n - name: Archive production artifacts\n uses: actions/upload-artifact@v4\n with:\n name: yao-macos\n path: |\n dist/release/*\n" }, { "path": ".github/workflows/notarize-macos.yml", "content": "name: Notarize macOS\n\non:\n workflow_dispatch:\n inputs:\n run_id:\n description: \"Release macOS workflow run ID (to download artifacts from)\"\n required: true\n version:\n description: \"Version used in the release build (e.g. 1.0.0 or 1.0.0-alpha)\"\n required: true\n\npermissions:\n contents: write\n\njobs:\n # ===================================================================\n # Notarize Yao binaries (arm64 + amd64)\n # ===================================================================\n notarize:\n runs-on: macos-latest\n strategy:\n matrix:\n arch: [arm64, amd64]\n steps:\n - name: Download Yao Binary\n uses: actions/download-artifact@v4\n with:\n name: yao-darwin-${{ matrix.arch }}\n path: bin\n run-id: ${{ github.event.inputs.run_id }}\n github-token: ${{ secrets.GITHUB_TOKEN }}\n\n - name: Install Certificates\n env:\n KEYCHAIN_PASSWORD: ${{ secrets.KEYCHAIN_PASSWORD }}\n run: |\n mkdir -p certs\n echo \"${{ secrets.APPLE_DEVELOPERIDG2CA }}\" | base64 --decode > certs/DeveloperIDG2CA.cer\n echo \"${{ secrets.APPLE_DISTRIBUTION }}\" | base64 --decode > certs/distribution.cer\n echo \"${{ secrets.APPLE_PRIVATE_KEY }}\" | base64 --decode > certs/private_key.p12\n security verify-cert -c certs/DeveloperIDG2CA.cer\n security verify-cert -c certs/distribution.cer\n\n - name: Import Certificates\n env:\n KEYCHAIN_PASSWORD: ${{ secrets.KEYCHAIN_PASSWORD }}\n run: |\n KEYCHAIN_PATH=$RUNNER_TEMP/app-signing.keychain-db\n security create-keychain -p \"$KEYCHAIN_PASSWORD\" $KEYCHAIN_PATH\n security set-keychain-settings -lut 21600 $KEYCHAIN_PATH\n security unlock-keychain -p \"$KEYCHAIN_PASSWORD\" $KEYCHAIN_PATH\n security import ./certs/DeveloperIDG2CA.cer -k $KEYCHAIN_PATH -T /usr/bin/codesign\n security import ./certs/distribution.cer -k $KEYCHAIN_PATH -T /usr/bin/codesign\n security import ./certs/private_key.p12 -k $KEYCHAIN_PATH -P \"${{ secrets.APPLE_PRIVATE_KEY_PASSWORD }}\" -T /usr/bin/codesign\n security list-keychain -d user -s $KEYCHAIN_PATH\n\n - name: Verify Signature\n run: codesign --verify --deep --strict --verbose=2 bin/yao\n\n - name: Notarize Yao ${{ matrix.arch }}\n timeout-minutes: 15\n env:\n APPLE_ID: ${{ secrets.APPLE_ID }}\n APPLE_TEAME_ID: ${{ secrets.APPLE_TEAME_ID }}\n APPLE_APP_SPEC_PASS: ${{ secrets.APPLE_APP_SPEC_PASS }}\n run: |\n zip -j bin/yao.zip bin/yao\n\n SUBMIT_OUT=$(xcrun notarytool submit bin/yao.zip \\\n --apple-id \"$APPLE_ID\" \\\n --team-id \"$APPLE_TEAME_ID\" \\\n --password \"$APPLE_APP_SPEC_PASS\" \\\n --wait --timeout 10m --output-format json 2>&1) || true\n echo \"$SUBMIT_OUT\"\n\n STATUS=$(echo \"$SUBMIT_OUT\" | python3 -c \"import sys,json; print(json.load(sys.stdin).get('status',''))\" 2>/dev/null || true)\n SUB_ID=$(echo \"$SUBMIT_OUT\" | python3 -c \"import sys,json; print(json.load(sys.stdin).get('id',''))\" 2>/dev/null || true)\n\n if [ \"$STATUS\" != \"Accepted\" ]; then\n echo \"::error::Yao ${{ matrix.arch }} notarization failed (status: $STATUS)\"\n [ -n \"$SUB_ID\" ] && xcrun notarytool log \"$SUB_ID\" \\\n --apple-id \"$APPLE_ID\" \\\n --team-id \"$APPLE_TEAME_ID\" \\\n --password \"$APPLE_APP_SPEC_PASS\" || true\n exit 1\n fi\n echo \"Yao ${{ matrix.arch }} notarization accepted.\"\n" }, { "path": ".github/workflows/pr-receive.yml", "content": "name: Receive PR\n\n# read-only repo token\n# no access to secrets\non:\n pull_request:\n\njobs:\n build:\n runs-on: ubuntu-latest\n\n steps:\n - uses: actions/checkout@v4\n\n - name: Save PR number\n run: |\n mkdir -p ./pr\n echo ${{ github.event.number }} > ./pr/NR\n echo ${{ github.event.pull_request.head.sha }} > ./pr/SHA\n\n - uses: actions/upload-artifact@v4\n with:\n name: pr\n path: pr/\n" }, { "path": ".github/workflows/pr-test.yml", "content": "name: PR Unit Test\n\n# read-write repo token\n# access to secrets\non:\n workflow_run:\n workflows: [\"Receive PR\"]\n types:\n - completed\nenv:\n YAO_DEV: ${{ github.WORKSPACE }}\n YAO_ENV: development\n YAO_ROOT: ${{ github.WORKSPACE }}/../app\n YAO_HOST: 0.0.0.0\n YAO_PORT: 5099\n YAO_SESSION: \"memory\"\n YAO_LOG: \"./logs/application.log\"\n YAO_LOG_MODE: \"TEXT\"\n YAO_JWT_SECRET: \"bLp@bi!oqo-2U+hoTRUG\"\n YAO_DB_AESKEY: \"ZLX=T&f6refeCh-ro*r@\"\n OSS_TEST_ID: ${{ secrets.OSS_TEST_ID}}\n OSS_TEST_SECRET: ${{ secrets.OSS_TEST_SECRET}}\n ROOT_PLUGIN: ${{ github.WORKSPACE }}/../../../data/gou-unit/plugins\n\n MYSQL_TEST_HOST: \"127.0.0.1\"\n MYSQL_TEST_PORT: \"3308\"\n MYSQL_TEST_USER: test\n MYSQL_TEST_PASS: \"123456\"\n\n SQLITE_DB: \"./app/db/yao.db\"\n\n REDIS_TEST_HOST: \"127.0.0.1\"\n REDIS_TEST_PORT: \"6379\"\n REDIS_TEST_DB: \"2\"\n\n MONGO_TEST_HOST: \"127.0.0.1\"\n MONGO_TEST_PORT: \"27017\"\n MONGO_TEST_USER: \"root\"\n MONGO_TEST_PASS: \"123456\"\n\n OPENAI_TEST_KEY: ${{ secrets.OPENAI_TEST_KEY }}\n TEST_MOAPI_SECRET: ${{ secrets.OPENAI_TEST_KEY }}\n OPENAI_API_KEY: ${{ secrets.OPENAI_TEST_KEY }}\n ANTHROPIC_API_KEY: ${{ secrets.ANTHROPIC_API_KEY }}\n TEST_MOAPI_MIRROR: https://api.openai.com\n\n # DeepSeek API Configuration\n DEEPSEEK_API_KEY: ${{ secrets.DEEPSEEK_API_KEY }}\n DEEPSEEK_API_PROXY: ${{ secrets.DEEPSEEK_API_PROXY }}\n DEEPSEEK_MODELS_R1: ${{ secrets.DEEPSEEK_MODELS_R1 }}\n DEEPSEEK_MODELS_V3: ${{ secrets.DEEPSEEK_MODELS_V3 }}\n DEEPSEEK_MODELS_V3_1: ${{ secrets.DEEPSEEK_MODELS_V3_1 }}\n\n # Search API Configuration\n TAVILY_API_KEY: ${{ secrets.TAVILY_API_KEY }}\n SERPAPI_API_KEY: ${{ secrets.SERPAPI_API_KEY }}\n SERPER_API_KEY: ${{ secrets.SERPER_API_KEY }}\n\n # Claude API Configuration\n CLAUDE_API_KEY: ${{ secrets.CLAUDE_API_KEY }}\n CLAUDE_PROXY: ${{ secrets.CLAUDE_PROXY }}\n CLAUDE_API_HOST: ${{ secrets.CLAUDE_API_HOST }}\n CLAUDE_SONNET_4: ${{ secrets.CLAUDE_SONNET_4 }}\n CLAUDE_SONNET_4_THINKING: ${{ secrets.CLAUDE_SONNET_4_THINKING }}\n\n # Moonshot / Kimi API Configuration\n MOONSHOT_API_KEY: ${{ secrets.MOONSHOT_API_KEY }}\n MOONSHOT_PROXY: \"https://api.moonshot.cn\"\n KIMI_CODE_API_KEY: ${{ secrets.KIMI_CODE_API_KEY }}\n KIMI_CODE_PROXY: \"https://api.kimi.com/coding\"\n\n TAB_NAME: \"::PET ADMIN\"\n PAGE_SIZE: \"20\"\n PAGE_LINK: \"https://yaoapps.com\"\n PAGE_ICON: \"icon-trash\"\n\n DEMO_APP: ${{ github.WORKSPACE }}/../app\n\n # Application Setting\n\n ## Path\n YAO_EXTENSION_ROOT: ${{ github.WORKSPACE }}/../extension\n YAO_TEST_APPLICATION: ${{ github.WORKSPACE }}/../app\n YAO_SUI_TEST_APPLICATION: ${{ github.WORKSPACE }}/../yao-startup-webapp\n\n ## Runtime\n YAO_RUNTIME_MIN: 3\n YAO_RUNTIME_MAX: 6\n YAO_RUNTIME_HEAP_LIMIT: 1500000000\n YAO_RUNTIME_HEAP_RELEASE: 10000000\n YAO_RUNTIME_HEAP_AVAILABLE: 550000000\n YAO_RUNTIME_PRECOMPILE: true\n\n # Neo4j\n NEO4J_TEST_URL: \"neo4j://localhost:7687\"\n NEO4J_TEST_USER: \"neo4j\"\n NEO4J_TEST_PASS: \"Yao2026Neo4j\"\n\n # Qdrant\n QDRANT_TEST_HOST: \"127.0.0.1\"\n QDRANT_TEST_PORT: \"6334\"\n\n # S3\n S3_API: ${{ secrets.S3_API }}\n S3_ACCESS_KEY: ${{ secrets.S3_ACCESS_KEY }}\n S3_SECRET_KEY: ${{ secrets.S3_SECRET_KEY }}\n S3_BUCKET: ${{ secrets.S3_BUCKET }}\n S3_PUBLIC_URL: ${{ secrets.S3_PUBLIC_URL }}\n\n # === Openapi Signin Configs ===\n SIGNIN_CLIENT_ID: \"kiCeR88kDwHBDuNHvN51cZgmpp3tmF6Z\"\n\n ## Google\n GOOGLE_CLIENT_ID: ${{ secrets.GOOGLE_CLIENT_ID }}\n GOOGLE_CLIENT_SECRET: ${{ secrets.GOOGLE_CLIENT_SECRET }}\n\n ## Microsoft\n MICROSOFT_CLIENT_ID: ${{ secrets.MICROSOFT_CLIENT_ID }}\n MICROSOFT_CLIENT_SECRET: ${{ secrets.MICROSOFT_CLIENT_SECRET }}\n\n ## Apple\n APPLE_SERVICE_ID: ${{ secrets.APPLE_SERVICE_ID }}\n APPLE_PRIVATE_KEY_PATH: \"apple/signin_client_secret_key.p8\"\n APPLE_KEY_ID: ${{ secrets.APPLE_KEY_ID }}\n APPLE_TEAM_ID: ${{ secrets.APPLE_TEAM_ID }}\n\n ## Github\n GITHUBUSER_CLIENT_ID: ${{ secrets.GITHUBUSER_CLIENT_ID }}\n GITHUBUSER_CLIENT_SECRET: ${{ secrets.GITHUBUSER_CLIENT_SECRET }}\n\n ## Cloudflare Turnstile\n CLOUDFLARE_TURNSTILE_SITEKEY: ${{ secrets.CLOUDFLARE_TURNSTILE_SITEKEY }}\n CLOUDFLARE_TURNSTILE_SECRET: ${{ secrets.CLOUDFLARE_TURNSTILE_SECRET }}\n\n # === Messaging Services ===\n ## Mailgun\n MAILGUN_DOMAIN: ${{ secrets.MAILGUN_DOMAIN }}\n MAILGUN_API_KEY: ${{ secrets.MAILGUN_API_KEY }}\n MAILGUN_FROM: \"Yaobots Tests \"\n\n ## SMTP Server（ Mailgun ）\n SMTP_HOST: \"smtp.mailgun.org\"\n SMTP_PORT: \"465\"\n SMTP_USERNAME: ${{ secrets.SMTP_USERNAME }}\n SMTP_PASSWORD: ${{ secrets.SMTP_PASSWORD }}\n SMTP_FROM: \"Yaobots SMTP Tests \"\n\n ## SMTP Server（ Gmail ）\n RELIABLE_SMTP_HOST: \"smtp.gmail.com\"\n RELIABLE_SMTP_PORT: \"465\"\n RELIABLE_SMTP_USERNAME: ${{ secrets.RELIABLE_SMTP_USERNAME }}\n RELIABLE_SMTP_PASSWORD: ${{ secrets.RELIABLE_SMTP_PASSWORD }}\n RELIABLE_SMTP_FROM: \"Yaobots Gmail Tests \"\n\n ## IMAP Server (Gmail)\n RELIABLE_IMAP_HOST: \"imap.gmail.com\"\n RELIABLE_IMAP_PORT: \"993\"\n RELIABLE_IMAP_USERNAME: ${{ secrets.RELIABLE_SMTP_USERNAME }}\n RELIABLE_IMAP_PASSWORD: ${{ secrets.RELIABLE_SMTP_PASSWORD }}\n RELIABLE_IMAP_MAILBOX: \"INBOX\"\n\n ## Twilio\n TWILIO_ACCOUNT_SID: ${{ secrets.TWILIO_ACCOUNT_SID }}\n TWILIO_AUTH_TOKEN: ${{ secrets.TWILIO_AUTH_TOKEN }}\n TWILIO_API_SID: ${{ secrets.TWILIO_API_SID }}\n TWILIO_API_KEY: ${{ secrets.TWILIO_API_KEY }}\n TWILIO_SENDGRID_API_SID: ${{ secrets.TWILIO_SENDGRID_API_SID }}\n TWILIO_SENDGRID_API_KEY: ${{ secrets.TWILIO_SENDGRID_API_KEY }}\n TWILIO_FROM_PHONE: \"+17035701412\"\n TWILIO_FROM_EMAIL: \"unit-test@sendgrid.yaobots.com\"\n TWILIO_TEST_PHONE: ${{ secrets.TWILIO_TEST_PHONE }}\n\njobs:\n # =============================================================================\n # KB Tests (kb) - Run once with SQLite (requires Qdrant, Neo4j, FastEmbed)\n # =============================================================================\n KBTest:\n runs-on: ubuntu-latest\n services:\n qdrant:\n image: qdrant/qdrant:latest\n ports:\n - 6333:6333\n - 6334:6334\n\n fastembed:\n image: yaoapp/fastembed:latest-amd64\n env:\n FASTEMBED_PASSWORD: Yao@2026\n ports:\n - 6001:8000\n\n neo4j:\n image: neo4j:latest\n ports:\n - \"7687:7687\"\n env:\n NEO4J_AUTH: neo4j/Yao2026Neo4j\n\n mongodb:\n image: mongo:6.0\n ports:\n - 27017:27017\n env:\n MONGO_INITDB_ROOT_USERNAME: root\n MONGO_INITDB_ROOT_PASSWORD: 123456\n MONGO_INITDB_DATABASE: test\n\n strategy:\n matrix:\n go: [\"1.25\"]\n if: >\n ${{ github.event.workflow_run.event == 'pull_request' &&\n github.event.workflow_run.conclusion == 'success' }}\n steps:\n - name: \"Download artifact\"\n uses: actions/github-script@v7\n with:\n script: |\n var artifacts = await github.rest.actions.listWorkflowRunArtifacts({\n owner: context.repo.owner,\n repo: context.repo.repo,\n run_id: ${{github.event.workflow_run.id }},\n });\n var matchArtifact = artifacts.data.artifacts.filter((artifact) => {\n return artifact.name == \"pr\"\n })[0];\n var download = await github.rest.actions.downloadArtifact({\n owner: context.repo.owner,\n repo: context.repo.repo,\n artifact_id: matchArtifact.id,\n archive_format: 'zip',\n });\n var fs = require('fs');\n fs.writeFileSync('${{github.workspace}}/pr.zip', Buffer.from(download.data));\n\n - name: \"Read NR & SHA\"\n run: |\n unzip pr.zip\n cat NR\n cat SHA\n echo HEAD=$(cat SHA) >> $GITHUB_ENV\n echo NR=$(cat NR) >> $GITHUB_ENV\n\n - name: \"Comment on PR\"\n uses: actions/github-script@v7\n with:\n github-token: ${{ secrets.GITHUB_TOKEN }}\n script: |\n const { NR } = process.env\n var issue_number = NR;\n await github.rest.issues.createComment({\n owner: context.repo.owner,\n repo: context.repo.repo,\n issue_number: issue_number,\n body: '🤖 KB Tests (kb) running with SQLite...'\n });\n\n - name: Checkout Kun\n uses: actions/checkout@v4\n with:\n repository: yaoapp/kun\n path: kun\n\n - name: Checkout Xun\n uses: actions/checkout@v4\n with:\n repository: yaoapp/xun\n path: xun\n\n - name: Checkout Gou\n uses: actions/checkout@v4\n with:\n repository: yaoapp/gou\n path: gou\n\n - name: Checkout V8Go\n uses: actions/checkout@v4\n with:\n repository: yaoapp/v8go\n path: v8go\n\n - name: Unzip libv8\n run: |\n files=$(find ./v8go -name \"libv8*.zip\")\n for file in $files; do\n dir=$(dirname \"$file\")\n echo \"Extracting $file to directory $dir\"\n unzip -o -d $dir $file\n rm -rf $dir/__MACOSX\n done\n\n - name: Checkout Demo App\n uses: actions/checkout@v4\n with:\n repository: yaoapp/yao-dev-app\n path: app\n\n - name: Checkout Extension\n uses: actions/checkout@v4\n with:\n repository: yaoapp/yao-extensions-dev\n path: extension\n\n - name: Move Dependencies\n run: |\n mv kun ../\n mv xun ../\n mv gou ../\n mv v8go ../\n mv app ../\n mv extension ../\n\n - name: Checkout pull request HEAD commit\n uses: actions/checkout@v4\n with:\n ref: ${{ env.HEAD }}\n\n - name: Setup Apple Private Key\n run: |\n mkdir -p ../app/openapi/certs/apple\n echo \"${{ secrets.APPLE_PRIVATE_KEY_USER }}\" > ../app/openapi/certs/apple/signin_client_secret_key.p8\n\n - name: Setup Go ${{ matrix.go }}\n uses: actions/setup-go@v5\n with:\n go-version: ${{ matrix.go }}\n\n - name: Start Redis\n run: docker run --name redis --publish 6379:6379 --detach redis:6\n\n - name: Setup Go Tools\n run: make tools\n\n - name: Setup ENV (SQLite)\n run: |\n mkdir -p ${{ github.WORKSPACE }}/../app/db\n echo \"YAO_DB_DRIVER=sqlite3\" >> $GITHUB_ENV\n echo \"YAO_DB_PRIMARY=${{ github.WORKSPACE }}/../app/db/yao.db\" >> $GITHUB_ENV\n\n - name: Run KB Tests (kb)\n run: make unit-test-kb\n\n - name: Codecov Report\n uses: codecov/codecov-action@v4\n with:\n token: ${{ secrets.CODECOV_TOKEN }}\n\n - name: \"Comment on PR - KB Tests Done\"\n uses: actions/github-script@v7\n with:\n github-token: ${{ secrets.GITHUB_TOKEN }}\n script: |\n const { NR } = process.env\n var issue_number = NR;\n await github.rest.issues.createComment({\n owner: context.repo.owner,\n repo: context.repo.repo,\n issue_number: issue_number,\n body: '✅ KB Tests (kb) passed!'\n });\n\n # =============================================================================\n # Agent Tests (agent, aigc) - Run once with SQLite\n # =============================================================================\n AgentTest:\n runs-on: ubuntu-latest\n services:\n qdrant:\n image: qdrant/qdrant:latest\n ports:\n - 6333:6333\n - 6334:6334\n\n fastembed:\n image: yaoapp/fastembed:latest-amd64\n env:\n FASTEMBED_PASSWORD: Yao@2026\n ports:\n - 6001:8000\n\n neo4j:\n image: neo4j:latest\n ports:\n - \"7687:7687\"\n env:\n NEO4J_AUTH: neo4j/Yao2026Neo4j\n\n mcp-everything:\n image: yaoapp/mcp-everything:latest\n ports:\n - \"3021:3021\"\n - \"3022:3022\"\n\n mongodb:\n image: mongo:6.0\n ports:\n - 27017:27017\n env:\n MONGO_INITDB_ROOT_USERNAME: root\n MONGO_INITDB_ROOT_PASSWORD: 123456\n MONGO_INITDB_DATABASE: test\n\n strategy:\n matrix:\n go: [\"1.25\"]\n if: >\n ${{ github.event.workflow_run.event == 'pull_request' &&\n github.event.workflow_run.conclusion == 'success' }}\n steps:\n - name: \"Download artifact\"\n uses: actions/github-script@v7\n with:\n script: |\n var artifacts = await github.rest.actions.listWorkflowRunArtifacts({\n owner: context.repo.owner,\n repo: context.repo.repo,\n run_id: ${{github.event.workflow_run.id }},\n });\n var matchArtifact = artifacts.data.artifacts.filter((artifact) => {\n return artifact.name == \"pr\"\n })[0];\n var download = await github.rest.actions.downloadArtifact({\n owner: context.repo.owner,\n repo: context.repo.repo,\n artifact_id: matchArtifact.id,\n archive_format: 'zip',\n });\n var fs = require('fs');\n fs.writeFileSync('${{github.workspace}}/pr.zip', Buffer.from(download.data));\n\n - name: \"Read NR & SHA\"\n run: |\n unzip pr.zip\n cat NR\n cat SHA\n echo HEAD=$(cat SHA) >> $GITHUB_ENV\n echo NR=$(cat NR) >> $GITHUB_ENV\n\n - name: \"Comment on PR\"\n uses: actions/github-script@v7\n with:\n github-token: ${{ secrets.GITHUB_TOKEN }}\n script: |\n const { NR } = process.env\n var issue_number = NR;\n await github.rest.issues.createComment({\n owner: context.repo.owner,\n repo: context.repo.repo,\n issue_number: issue_number,\n body: '🤖 Agent Tests (agent, aigc) running with SQLite...'\n });\n\n - name: Checkout Kun\n uses: actions/checkout@v4\n with:\n repository: yaoapp/kun\n path: kun\n\n - name: Checkout Xun\n uses: actions/checkout@v4\n with:\n repository: yaoapp/xun\n path: xun\n\n - name: Checkout Gou\n uses: actions/checkout@v4\n with:\n repository: yaoapp/gou\n path: gou\n\n - name: Checkout V8Go\n uses: actions/checkout@v4\n with:\n repository: yaoapp/v8go\n path: v8go\n\n - name: Unzip libv8\n run: |\n files=$(find ./v8go -name \"libv8*.zip\")\n for file in $files; do\n dir=$(dirname \"$file\")\n echo \"Extracting $file to directory $dir\"\n unzip -o -d $dir $file\n rm -rf $dir/__MACOSX\n done\n\n - name: Checkout Demo App\n uses: actions/checkout@v4\n with:\n repository: yaoapp/yao-dev-app\n path: app\n\n - name: Checkout Extension\n uses: actions/checkout@v4\n with:\n repository: yaoapp/yao-extensions-dev\n path: extension\n\n - name: Move Dependencies\n run: |\n mv kun ../\n mv xun ../\n mv gou ../\n mv v8go ../\n mv app ../\n mv extension ../\n\n - name: Checkout pull request HEAD commit\n uses: actions/checkout@v4\n with:\n ref: ${{ env.HEAD }}\n\n - name: Setup Apple Private Key\n run: |\n mkdir -p ../app/openapi/certs/apple\n echo \"${{ secrets.APPLE_PRIVATE_KEY_USER }}\" > ../app/openapi/certs/apple/signin_client_secret_key.p8\n\n - name: Setup Go ${{ matrix.go }}\n uses: actions/setup-go@v5\n with:\n go-version: ${{ matrix.go }}\n\n - name: Start Redis\n run: docker run --name redis --publish 6379:6379 --detach redis:6\n\n - name: Setup Go Tools\n run: make tools\n\n - name: Install pdftoppm, mutool, imagemagick\n run: |\n sudo apt update\n sudo apt install -y poppler-utils mupdf-tools imagemagick\n\n - name: Test pdftoppm, mutool, imagemagick\n run: |\n pdftoppm -v\n mutool -v\n convert -version\n\n - name: Setup ENV (SQLite)\n run: |\n mkdir -p ${{ github.WORKSPACE }}/../app/db\n echo \"YAO_DB_DRIVER=sqlite3\" >> $GITHUB_ENV\n echo \"YAO_DB_PRIMARY=${{ github.WORKSPACE }}/../app/db/yao.db\" >> $GITHUB_ENV\n\n - name: Pull Sandbox Test Images\n run: |\n docker pull alpine:latest\n docker pull yaoapp/sandbox-base:latest || true\n docker pull yaoapp/sandbox-claude:latest || true\n\n - name: Run Agent Tests (agent, aigc)\n env:\n YAO_SANDBOX_WORKSPACE: ${{ runner.temp }}/sandbox/workspace\n YAO_SANDBOX_IPC: ${{ runner.temp }}/sandbox/ipc\n run: |\n export YAO_SANDBOX_CONTAINER_USER=\"$(id -u):$(id -g)\"\n make unit-test-agent\n\n - name: Codecov Report\n uses: codecov/codecov-action@v4\n with:\n token: ${{ secrets.CODECOV_TOKEN }}\n\n - name: \"Comment on PR - Agent Tests Done\"\n uses: actions/github-script@v7\n with:\n github-token: ${{ secrets.GITHUB_TOKEN }}\n script: |\n const { NR } = process.env\n var issue_number = NR;\n await github.rest.issues.createComment({\n owner: context.repo.owner,\n repo: context.repo.repo,\n issue_number: issue_number,\n body: '✅ Agent Tests (agent, aigc) passed!'\n });\n\n # =============================================================================\n # Robot Tests (all agent/robot/... packages) - Unit + E2E with real LLM calls\n # =============================================================================\n RobotTest:\n runs-on: ubuntu-latest\n services:\n mcp-everything:\n image: yaoapp/mcp-everything:latest\n ports:\n - \"3021:3021\"\n - \"3022:3022\"\n\n mongodb:\n image: mongo:6.0\n ports:\n - 27017:27017\n env:\n MONGO_INITDB_ROOT_USERNAME: root\n MONGO_INITDB_ROOT_PASSWORD: 123456\n MONGO_INITDB_DATABASE: test\n\n strategy:\n matrix:\n go: [\"1.25\"]\n if: >\n ${{ github.event.workflow_run.event == 'pull_request' &&\n github.event.workflow_run.conclusion == 'success' }}\n steps:\n - name: \"Download artifact\"\n uses: actions/github-script@v7\n with:\n script: |\n var artifacts = await github.rest.actions.listWorkflowRunArtifacts({\n owner: context.repo.owner,\n repo: context.repo.repo,\n run_id: ${{github.event.workflow_run.id }},\n });\n var matchArtifact = artifacts.data.artifacts.filter((artifact) => {\n return artifact.name == \"pr\"\n })[0];\n var download = await github.rest.actions.downloadArtifact({\n owner: context.repo.owner,\n repo: context.repo.repo,\n artifact_id: matchArtifact.id,\n archive_format: 'zip',\n });\n var fs = require('fs');\n fs.writeFileSync('${{github.workspace}}/pr.zip', Buffer.from(download.data));\n\n - name: \"Read NR & SHA\"\n run: |\n unzip pr.zip\n cat NR\n cat SHA\n echo HEAD=$(cat SHA) >> $GITHUB_ENV\n echo NR=$(cat NR) >> $GITHUB_ENV\n\n - name: \"Comment on PR\"\n uses: actions/github-script@v7\n with:\n github-token: ${{ secrets.GITHUB_TOKEN }}\n script: |\n const { NR } = process.env\n var issue_number = NR;\n await github.rest.issues.createComment({\n owner: context.repo.owner,\n repo: context.repo.repo,\n issue_number: issue_number,\n body: '🤖 Robot Tests (Unit + E2E) running with SQLite...'\n });\n\n - name: Checkout Kun\n uses: actions/checkout@v4\n with:\n repository: yaoapp/kun\n path: kun\n\n - name: Checkout Xun\n uses: actions/checkout@v4\n with:\n repository: yaoapp/xun\n path: xun\n\n - name: Checkout Gou\n uses: actions/checkout@v4\n with:\n repository: yaoapp/gou\n path: gou\n\n - name: Checkout V8Go\n uses: actions/checkout@v4\n with:\n repository: yaoapp/v8go\n path: v8go\n\n - name: Unzip libv8\n run: |\n files=$(find ./v8go -name \"libv8*.zip\")\n for file in $files; do\n dir=$(dirname \"$file\")\n echo \"Extracting $file to directory $dir\"\n unzip -o -d $dir $file\n rm -rf $dir/__MACOSX\n done\n\n - name: Checkout Demo App\n uses: actions/checkout@v4\n with:\n repository: yaoapp/yao-dev-app\n path: app\n\n - name: Checkout Extension\n uses: actions/checkout@v4\n with:\n repository: yaoapp/yao-extensions-dev\n path: extension\n\n - name: Move Dependencies\n run: |\n mv kun ../\n mv xun ../\n mv gou ../\n mv v8go ../\n mv app ../\n mv extension ../\n\n - name: Checkout pull request HEAD commit\n uses: actions/checkout@v4\n with:\n ref: ${{ env.HEAD }}\n\n - name: Setup Go ${{ matrix.go }}\n uses: actions/setup-go@v5\n with:\n go-version: ${{ matrix.go }}\n\n - name: Start Redis\n run: docker run --name redis --publish 6379:6379 --detach redis:6\n\n - name: Setup Go Tools\n run: make tools\n\n - name: Setup ENV (SQLite)\n run: |\n mkdir -p ${{ github.WORKSPACE }}/../app/db\n echo \"YAO_DB_DRIVER=sqlite3\" >> $GITHUB_ENV\n echo \"YAO_DB_PRIMARY=${{ github.WORKSPACE }}/../app/db/yao.db\" >> $GITHUB_ENV\n\n - name: Run Robot Tests (Unit + E2E)\n run: make unit-test-robot\n\n - name: Codecov Report\n uses: codecov/codecov-action@v4\n with:\n token: ${{ secrets.CODECOV_TOKEN }}\n\n - name: \"Comment on PR - Robot Tests Done\"\n uses: actions/github-script@v7\n with:\n github-token: ${{ secrets.GITHUB_TOKEN }}\n script: |\n const { NR } = process.env\n var issue_number = NR;\n await github.rest.issues.createComment({\n owner: context.repo.owner,\n repo: context.repo.repo,\n issue_number: issue_number,\n body: '✅ Robot Tests (Unit + E2E) passed!'\n });\n\n # =============================================================================\n # Sandbox Tests (requires Docker) - Run with Docker-in-Docker\n # =============================================================================\n SandboxTest:\n runs-on: ubuntu-latest\n services:\n mongodb:\n image: mongo:6.0\n ports:\n - 27017:27017\n env:\n MONGO_INITDB_ROOT_USERNAME: root\n MONGO_INITDB_ROOT_PASSWORD: 123456\n MONGO_INITDB_DATABASE: test\n\n strategy:\n matrix:\n go: [\"1.25\"]\n if: >\n ${{ github.event.workflow_run.event == 'pull_request' &&\n github.event.workflow_run.conclusion == 'success' }}\n steps:\n - name: \"Download artifact\"\n uses: actions/github-script@v7\n with:\n script: |\n var artifacts = await github.rest.actions.listWorkflowRunArtifacts({\n owner: context.repo.owner,\n repo: context.repo.repo,\n run_id: ${{github.event.workflow_run.id }},\n });\n var matchArtifact = artifacts.data.artifacts.filter((artifact) => {\n return artifact.name == \"pr\"\n })[0];\n var download = await github.rest.actions.downloadArtifact({\n owner: context.repo.owner,\n repo: context.repo.repo,\n artifact_id: matchArtifact.id,\n archive_format: 'zip',\n });\n var fs = require('fs');\n fs.writeFileSync('${{github.workspace}}/pr.zip', Buffer.from(download.data));\n\n - name: \"Read NR & SHA\"\n run: |\n unzip pr.zip\n cat NR\n cat SHA\n echo HEAD=$(cat SHA) >> $GITHUB_ENV\n echo NR=$(cat NR) >> $GITHUB_ENV\n\n - name: \"Comment on PR\"\n uses: actions/github-script@v7\n with:\n github-token: ${{ secrets.GITHUB_TOKEN }}\n script: |\n const { NR } = process.env\n var issue_number = NR;\n await github.rest.issues.createComment({\n owner: context.repo.owner,\n repo: context.repo.repo,\n issue_number: issue_number,\n body: '🤖 Sandbox Tests running with Docker...'\n });\n\n - name: Checkout Kun\n uses: actions/checkout@v4\n with:\n repository: yaoapp/kun\n path: kun\n\n - name: Checkout Xun\n uses: actions/checkout@v4\n with:\n repository: yaoapp/xun\n path: xun\n\n - name: Checkout Gou\n uses: actions/checkout@v4\n with:\n repository: yaoapp/gou\n path: gou\n\n - name: Checkout V8Go\n uses: actions/checkout@v4\n with:\n repository: yaoapp/v8go\n path: v8go\n\n - name: Unzip libv8\n run: |\n files=$(find ./v8go -name \"libv8*.zip\")\n for file in $files; do\n dir=$(dirname \"$file\")\n echo \"Extracting $file to directory $dir\"\n unzip -o -d $dir $file\n rm -rf $dir/__MACOSX\n done\n\n - name: Checkout Demo App\n uses: actions/checkout@v4\n with:\n repository: yaoapp/yao-dev-app\n path: app\n\n - name: Checkout Extension\n uses: actions/checkout@v4\n with:\n repository: yaoapp/yao-extensions-dev\n path: extension\n\n - name: Move Dependencies\n run: |\n mv kun ../\n mv xun ../\n mv gou ../\n mv v8go ../\n mv app ../\n mv extension ../\n\n - name: Checkout pull request HEAD commit\n uses: actions/checkout@v4\n with:\n ref: ${{ env.HEAD }}\n\n - name: Setup Go ${{ matrix.go }}\n uses: actions/setup-go@v5\n with:\n go-version: ${{ matrix.go }}\n\n - name: Start Redis\n run: docker run --name redis --publish 6379:6379 --detach redis:6\n\n - name: Setup Go Tools\n run: make tools\n\n - name: Setup ENV (SQLite)\n run: |\n mkdir -p ${{ github.WORKSPACE }}/../app/db\n echo \"YAO_DB_DRIVER=sqlite3\" >> $GITHUB_ENV\n echo \"YAO_DB_PRIMARY=${{ github.WORKSPACE }}/../app/db/yao.db\" >> $GITHUB_ENV\n\n - name: Pull Sandbox Test Images\n run: |\n docker pull alpine:latest\n docker pull yaoapp/sandbox-base:latest || true\n docker pull yaoapp/sandbox-claude:latest || true\n\n - name: Run Sandbox Tests\n env:\n YAO_SANDBOX_WORKSPACE: ${{ runner.temp }}/sandbox/workspace\n YAO_SANDBOX_IPC: ${{ runner.temp }}/sandbox/ipc\n run: |\n # Use runner's UID:GID to avoid permission issues\n export YAO_SANDBOX_CONTAINER_USER=\"$(id -u):$(id -g)\"\n make unit-test-sandbox\n\n - name: Codecov Report\n uses: codecov/codecov-action@v4\n with:\n token: ${{ secrets.CODECOV_TOKEN }}\n\n - name: \"Comment on PR - Sandbox Tests Done\"\n uses: actions/github-script@v7\n with:\n github-token: ${{ secrets.GITHUB_TOKEN }}\n script: |\n const { NR } = process.env\n var issue_number = NR;\n await github.rest.issues.createComment({\n owner: context.repo.owner,\n repo: context.repo.repo,\n issue_number: issue_number,\n body: '✅ Sandbox Tests passed!'\n });\n\n # =============================================================================\n # Sandbox V2 Tests (tai SDK + workspace, Docker + K8s via k3d)\n # Full sandbox/v2 integration tests are run locally.\n # =============================================================================\n SandboxV2Test:\n runs-on: ubuntu-latest\n services:\n mongodb:\n image: mongo:6.0\n ports:\n - 27017:27017\n env:\n MONGO_INITDB_ROOT_USERNAME: root\n MONGO_INITDB_ROOT_PASSWORD: 123456\n MONGO_INITDB_DATABASE: test\n\n strategy:\n matrix:\n go: [\"1.25\"]\n steps:\n - name: \"Download artifact\"\n uses: actions/github-script@v7\n with:\n script: |\n var artifacts = await github.rest.actions.listWorkflowRunArtifacts({\n owner: context.repo.owner,\n repo: context.repo.repo,\n run_id: ${{github.event.workflow_run.id }},\n });\n var matchArtifact = artifacts.data.artifacts.filter((artifact) => {\n return artifact.name == \"pr\"\n })[0];\n var download = await github.rest.actions.downloadArtifact({\n owner: context.repo.owner,\n repo: context.repo.repo,\n artifact_id: matchArtifact.id,\n archive_format: 'zip',\n });\n var fs = require('fs');\n fs.writeFileSync('${{github.workspace}}/pr.zip', Buffer.from(download.data));\n\n - name: \"Read NR & SHA\"\n run: |\n unzip pr.zip\n cat NR\n cat SHA\n echo HEAD=$(cat SHA) >> $GITHUB_ENV\n echo NR=$(cat NR) >> $GITHUB_ENV\n\n - name: \"Comment on PR\"\n uses: actions/github-script@v7\n with:\n github-token: ${{ secrets.GITHUB_TOKEN }}\n script: |\n const { NR } = process.env\n var issue_number = NR;\n await github.rest.issues.createComment({\n owner: context.repo.owner,\n repo: context.repo.repo,\n issue_number: issue_number,\n body: '🤖 Sandbox V2 CI Tests running (tai + workspace)...'\n });\n\n - name: Checkout Kun\n uses: actions/checkout@v4\n with:\n repository: yaoapp/kun\n path: kun\n\n - name: Checkout Xun\n uses: actions/checkout@v4\n with:\n repository: yaoapp/xun\n path: xun\n\n - name: Checkout Gou\n uses: actions/checkout@v4\n with:\n repository: yaoapp/gou\n path: gou\n\n - name: Checkout V8Go\n uses: actions/checkout@v4\n with:\n repository: yaoapp/v8go\n path: v8go\n\n - name: Unzip libv8\n run: |\n files=$(find ./v8go -name \"libv8*.zip\")\n for file in $files; do\n dir=$(dirname \"$file\")\n echo \"Extracting $file to directory $dir\"\n unzip -o -d $dir $file\n rm -rf $dir/__MACOSX\n done\n\n - name: Checkout Demo App\n uses: actions/checkout@v4\n with:\n repository: yaoapp/yao-dev-app\n path: app\n\n - name: Checkout Extension\n uses: actions/checkout@v4\n with:\n repository: yaoapp/yao-extensions-dev\n path: extension\n\n - name: Move Dependencies\n run: |\n mv kun ../\n mv xun ../\n mv gou ../\n mv v8go ../\n mv app ../\n mv extension ../\n\n - name: Checkout pull request HEAD commit\n uses: actions/checkout@v4\n with:\n ref: ${{ env.HEAD }}\n\n - name: Setup Apple Private Key\n run: |\n mkdir -p ../app/openapi/certs/apple\n echo \"${{ secrets.APPLE_PRIVATE_KEY_USER }}\" > ../app/openapi/certs/apple/signin_client_secret_key.p8\n\n - name: Setup Go ${{ matrix.go }}\n uses: actions/setup-go@v5\n with:\n go-version: ${{ matrix.go }}\n\n - name: Start Redis\n run: docker run --name redis --publish 6379:6379 --detach redis:6\n\n - name: Setup Go Tools\n run: make tools\n\n - name: Setup ENV (SQLite)\n run: |\n mkdir -p ${{ github.WORKSPACE }}/../app/db\n echo \"YAO_DB_DRIVER=sqlite3\" >> $GITHUB_ENV\n echo \"YAO_DB_PRIMARY=${{ github.WORKSPACE }}/../app/db/yao.db\" >> $GITHUB_ENV\n\n - name: Pull Test Images\n run: |\n docker pull yaoapp/tai-sandbox-test:latest || true\n docker pull yaoapp/tai:latest\n docker pull alpine:latest\n\n - name: Install k3d\n run: curl -s https://raw.githubusercontent.com/k3d-io/k3d/main/install.sh | bash\n\n - name: Create k3d cluster\n run: |\n k3d cluster create tai-test --no-lb --wait --api-port 16443\n kubectl wait --for=condition=Ready node --all --timeout=60s\n k3d image import alpine:latest -c tai-test\n\n - name: Start Tai Docker instance\n run: |\n docker run -d --name tai-docker \\\n -v /var/run/docker.sock:/var/run/docker.sock \\\n -p 19100:19100 -p 8099:8099 -p 12375:12375 -p 16080:16080 \\\n yaoapp/tai:latest server -direct \\\n -grpc 0.0.0.0:19100 -http 0.0.0.0:8099 -vnc 0.0.0.0:16080 -docker 0.0.0.0:12375\n\n for i in $(seq 1 30); do\n if curl -sf http://127.0.0.1:8099/healthz > /dev/null 2>&1; then\n echo \"Tai Docker HTTP ready\"; break\n fi\n echo \"Waiting for Tai Docker HTTP... ($i)\"; sleep 1\n done\n curl -sf http://127.0.0.1:8099/healthz > /dev/null 2>&1 || {\n echo \"::error::Tai Docker HTTP failed\"; docker logs tai-docker 2>&1; exit 1\n }\n\n for i in $(seq 1 15); do\n if nc -z 127.0.0.1 19100 2>/dev/null; then\n echo \"Tai Docker gRPC ready\"; break\n fi\n echo \"Waiting for Tai Docker gRPC... ($i)\"; sleep 1\n done\n nc -z 127.0.0.1 19100 2>/dev/null || {\n echo \"::error::Tai Docker gRPC failed\"; docker logs tai-docker 2>&1; exit 1\n }\n\n - name: Generate kubeconfig for Tai K8s\n run: |\n K3D_IP=$(docker inspect k3d-tai-test-server-0 | jq -r '.[0].NetworkSettings.Networks[\"k3d-tai-test\"].IPAddress')\n echo \"k3d server IP: ${K3D_IP}\"\n\n k3d kubeconfig get tai-test > /tmp/kubeconfig-k3d.yml\n\n # Kubeconfig for tai-k8s container (uses k3d-internal IP)\n sed \"s|server: .*|server: https://${K3D_IP}:6443|\" /tmp/kubeconfig-k3d.yml \\\n > /tmp/kubeconfig-tai-k8s.yml\n echo \"Container kubeconfig server:\"\n grep server: /tmp/kubeconfig-tai-k8s.yml\n\n # Kubeconfig for test runner (uses localhost via port-mapped 6443)\n sed 's|server: .*|server: https://127.0.0.1:6443|' /tmp/kubeconfig-k3d.yml \\\n > ${{ runner.temp }}/kubeconfig-tai.yml\n echo \"Test runner kubeconfig server:\"\n grep server: ${{ runner.temp }}/kubeconfig-tai.yml\n\n - name: Start Tai K8s instance\n run: |\n K3D_IP=$(docker inspect k3d-tai-test-server-0 | jq -r '.[0].NetworkSettings.Networks[\"k3d-tai-test\"].IPAddress')\n echo \"k3d server IP: ${K3D_IP}\"\n\n docker run -d --name tai-k8s \\\n --network k3d-tai-test \\\n -p 19101:19100 -p 8100:8099 -p 6443:16443 -p 16081:16080 \\\n -v /var/run/docker.sock:/var/run/docker.sock:ro \\\n -v /tmp/kubeconfig-tai-k8s.yml:/etc/tai/kubeconfig.yml:ro \\\n -e TAI_K8S_UPSTREAM=\"tcp://${K3D_IP}:6443\" \\\n -e TAI_KUBECONFIG=/etc/tai/kubeconfig.yml \\\n yaoapp/tai:latest server -direct \\\n -grpc 0.0.0.0:19100 -http 0.0.0.0:8099 -vnc 0.0.0.0:16080 -k8s 0.0.0.0:16443\n\n for i in $(seq 1 30); do\n if curl -sf http://127.0.0.1:8100/healthz > /dev/null 2>&1; then\n echo \"Tai K8s HTTP ready\"; break\n fi\n echo \"Waiting for Tai K8s HTTP... ($i)\"; sleep 1\n done\n curl -sf http://127.0.0.1:8100/healthz > /dev/null 2>&1 || {\n echo \"::error::Tai K8s HTTP failed\"; docker logs tai-k8s 2>&1; exit 1\n }\n\n for i in $(seq 1 15); do\n if nc -z 127.0.0.1 19101 2>/dev/null; then\n echo \"Tai K8s gRPC ready\"; break\n fi\n echo \"Waiting for Tai K8s gRPC... ($i)\"; sleep 1\n done\n nc -z 127.0.0.1 19101 2>/dev/null || {\n echo \"::error::Tai K8s gRPC failed\"; docker logs tai-k8s 2>&1; exit 1\n }\n\n - name: Run Sandbox V2 CI Tests (tai + workspace)\n env:\n TAI_TEST_HOST: \"127.0.0.1\"\n TAI_TEST_DOCKER: \"tcp://127.0.0.1:12375\"\n TAI_TEST_GRPC_PORT: \"19100\"\n TAI_TEST_HTTP_PORT: \"8099\"\n TAI_TEST_VNC_PORT: \"16080\"\n TAI_TEST_DOCKER_PORT: \"12375\"\n TAI_TEST_K8S_HOST: \"127.0.0.1\"\n TAI_TEST_K8S_PORT: \"6443\"\n TAI_TEST_K8S_GRPC_PORT: \"19101\"\n TAI_TEST_KUBECONFIG: \"${{ runner.temp }}/kubeconfig-tai.yml\"\n TAI_TEST_HOST_IP: \"172.17.0.1\"\n SANDBOX_TEST_REMOTE_ADDR: \"tai://127.0.0.1:19100\"\n SANDBOX_TEST_IMAGE: \"yaoapp/tai-sandbox-test:latest\"\n run: make unit-test-sandbox-v2\n\n - name: Codecov Report\n if: always()\n uses: codecov/codecov-action@v4\n with:\n token: ${{ secrets.CODECOV_TOKEN }}\n fail_ci_if_error: false\n\n - name: \"Comment on PR - Sandbox V2 Tests Done\"\n uses: actions/github-script@v7\n with:\n github-token: ${{ secrets.GITHUB_TOKEN }}\n script: |\n const { NR } = process.env\n var issue_number = NR;\n await github.rest.issues.createComment({\n owner: context.repo.owner,\n repo: context.repo.repo,\n issue_number: issue_number,\n body: '✅ Sandbox V2 CI Tests passed (tai + workspace)!'\n });\n\n # =============================================================================\n # Benchmark & Memory Leak Tests - Run with MySQL8.0 and SQLite3\n # =============================================================================\n PerfTest:\n runs-on: ubuntu-latest\n services:\n mcp-everything:\n image: yaoapp/mcp-everything:latest\n ports:\n - \"3021:3021\"\n - \"3022:3022\"\n\n mongodb:\n image: mongo:6.0\n ports:\n - 27017:27017\n env:\n MONGO_INITDB_ROOT_USERNAME: root\n MONGO_INITDB_ROOT_PASSWORD: 123456\n MONGO_INITDB_DATABASE: test\n\n strategy:\n matrix:\n go: [\"1.25\"]\n db: [MySQL8.0, SQLite3]\n if: >\n ${{ github.event.workflow_run.event == 'pull_request' &&\n github.event.workflow_run.conclusion == 'success' }}\n steps:\n - name: \"Download artifact\"\n uses: actions/github-script@v7\n with:\n script: |\n var artifacts = await github.rest.actions.listWorkflowRunArtifacts({\n owner: context.repo.owner,\n repo: context.repo.repo,\n run_id: ${{github.event.workflow_run.id }},\n });\n var matchArtifact = artifacts.data.artifacts.filter((artifact) => {\n return artifact.name == \"pr\"\n })[0];\n var download = await github.rest.actions.downloadArtifact({\n owner: context.repo.owner,\n repo: context.repo.repo,\n artifact_id: matchArtifact.id,\n archive_format: 'zip',\n });\n var fs = require('fs');\n fs.writeFileSync('${{github.workspace}}/pr.zip', Buffer.from(download.data));\n\n - name: \"Read NR & SHA\"\n run: |\n unzip pr.zip\n cat NR\n cat SHA\n echo HEAD=$(cat SHA) >> $GITHUB_ENV\n echo NR=$(cat NR) >> $GITHUB_ENV\n\n - name: Checkout Kun\n uses: actions/checkout@v4\n with:\n repository: yaoapp/kun\n path: kun\n\n - name: Checkout Xun\n uses: actions/checkout@v4\n with:\n repository: yaoapp/xun\n path: xun\n\n - name: Checkout Gou\n uses: actions/checkout@v4\n with:\n repository: yaoapp/gou\n path: gou\n\n - name: Checkout V8Go\n uses: actions/checkout@v4\n with:\n repository: yaoapp/v8go\n path: v8go\n\n - name: Unzip libv8\n run: |\n files=$(find ./v8go -name \"libv8*.zip\")\n for file in $files; do\n dir=$(dirname \"$file\")\n echo \"Extracting $file to directory $dir\"\n unzip -o -d $dir $file\n rm -rf $dir/__MACOSX\n done\n\n - name: Checkout Demo App\n uses: actions/checkout@v4\n with:\n repository: yaoapp/yao-dev-app\n path: app\n\n - name: Checkout Extension\n uses: actions/checkout@v4\n with:\n repository: yaoapp/yao-extensions-dev\n path: extension\n\n - name: Move Dependencies\n run: |\n mv kun ../\n mv xun ../\n mv gou ../\n mv v8go ../\n mv app ../\n mv extension ../\n\n - name: Checkout pull request HEAD commit\n uses: actions/checkout@v4\n with:\n ref: ${{ env.HEAD }}\n\n - name: Setup Go ${{ matrix.go }}\n uses: actions/setup-go@v5\n with:\n go-version: ${{ matrix.go }}\n\n - name: Start Redis\n run: docker run --name redis --publish 6379:6379 --detach redis:6\n\n - name: Install FFmpeg 7.x\n run: |\n wget https://github.com/BtbN/FFmpeg-Builds/releases/download/latest/ffmpeg-master-latest-linux64-gpl.tar.xz\n tar -xf ffmpeg-master-latest-linux64-gpl.tar.xz\n sudo cp ffmpeg-master-latest-linux64-gpl/bin/ffmpeg /usr/local/bin/\n sudo cp ffmpeg-master-latest-linux64-gpl/bin/ffprobe /usr/local/bin/\n sudo chmod +x /usr/local/bin/ffmpeg /usr/local/bin/ffprobe\n\n - name: Test FFmpeg\n run: ffmpeg -version\n\n - name: Setup Go Tools\n run: make tools\n\n - name: Setup ${{ matrix.db }}\n uses: ./.github/actions/setup-db\n with:\n kind: \"${{ matrix.db }}\"\n db: \"xiang\"\n user: \"xiang\"\n password: ${{ secrets.UNIT_PASS }}\n\n - name: Setup ENV\n env:\n PASSWORD: ${{ secrets.UNIT_PASS }}\n run: |\n echo \"YAO_DB_DRIVER=$DB_DRIVER\" >> $GITHUB_ENV\n if [ \"$DB_DRIVER\" = \"mysql\" ]; then\n echo \"YAO_DB_PRIMARY=$DB_USER:$PASSWORD@$DB_HOST\" >> $GITHUB_ENV\n else\n echo \"YAO_DB_PRIMARY=${{ github.WORKSPACE }}/../app/db/yao.db\" >> $GITHUB_ENV\n mkdir -p ${{ github.WORKSPACE }}/../app/db\n fi\n\n - name: Run Benchmark & Memory Leak Tests\n run: |\n make benchmark\n make memory-leak\n\n # =============================================================================\n # Core Tests - Run with DB matrix (MySQL/SQLite/Redis/Mongo combinations)\n # =============================================================================\n CoreTest:\n runs-on: ubuntu-latest\n\n services:\n qdrant:\n image: qdrant/qdrant:latest\n ports:\n - 6333:6333 # HTTP API\n - 6334:6334 # gRPC\n\n fastembed:\n image: yaoapp/fastembed:latest-amd64\n env:\n FASTEMBED_PASSWORD: Yao@2026\n ports:\n - 6001:8000\n\n neo4j:\n image: neo4j:latest\n ports:\n - \"7687:7687\"\n env:\n NEO4J_AUTH: neo4j/Yao2026Neo4j\n\n mcp-everything:\n image: yaoapp/mcp-everything:latest\n ports:\n - \"3021:3021\"\n - \"3022:3022\"\n\n strategy:\n matrix:\n go: [\"1.25\"]\n db: [MySQL8.0, SQLite3]\n redis: [4, 5, 6]\n mongo: [\"6.0\"]\n if: >\n ${{ github.event.workflow_run.event == 'pull_request' &&\n github.event.workflow_run.conclusion == 'success' }}\n steps:\n - name: \"Download artifact\"\n uses: actions/github-script@v7\n with:\n script: |\n var artifacts = await github.rest.actions.listWorkflowRunArtifacts({\n owner: context.repo.owner,\n repo: context.repo.repo,\n run_id: ${{github.event.workflow_run.id }},\n });\n var matchArtifact = artifacts.data.artifacts.filter((artifact) => {\n return artifact.name == \"pr\"\n })[0];\n var download = await github.rest.actions.downloadArtifact({\n owner: context.repo.owner,\n repo: context.repo.repo,\n artifact_id: matchArtifact.id,\n archive_format: 'zip',\n });\n var fs = require('fs');\n fs.writeFileSync('${{github.workspace}}/pr.zip', Buffer.from(download.data));\n\n - name: \"Read NR & SHA\"\n run: |\n unzip pr.zip\n cat NR\n cat SHA\n echo HEAD=$(cat SHA) >> $GITHUB_ENV\n echo NR=$(cat NR) >> $GITHUB_ENV\n\n - name: \"Comment on PR\"\n uses: actions/github-script@v7\n with:\n github-token: ${{ secrets.GITHUB_TOKEN }}\n script: |\n const { NR } = process.env\n var fs = require('fs');\n var issue_number = NR;\n await github.rest.issues.createComment({\n owner: context.repo.owner,\n repo: context.repo.repo,\n issue_number: issue_number,\n body: 'Thank you for the PR! The db: ${{ matrix.db }} redis: ${{ matrix.redis }} mongo: ${{ matrix.mongo }} test workflow is running, the results of the run will be commented later.'\n });\n\n - name: Checkout Kun\n uses: actions/checkout@v4\n with:\n repository: yaoapp/kun\n path: kun\n\n - name: Checkout Xun\n uses: actions/checkout@v4\n with:\n repository: yaoapp/xun\n path: xun\n\n - name: Checkout Gou\n uses: actions/checkout@v4\n with:\n repository: yaoapp/gou\n path: gou\n\n - name: Checkout V8Go\n uses: actions/checkout@v4\n with:\n repository: yaoapp/v8go\n path: v8go\n\n - name: Unzip libv8\n run: |\n files=$(find ./v8go -name \"libv8*.zip\")\n for file in $files; do\n dir=$(dirname \"$file\") # Get the directory where the ZIP file is located\n echo \"Extracting $file to directory $dir\"\n unzip -o -d $dir $file\n rm -rf $dir/__MACOSX\n done\n\n - name: Checkout Demo App\n uses: actions/checkout@v4\n with:\n repository: yaoapp/yao-dev-app\n path: app\n\n - name: Checkout Yao Startup Webapp\n uses: actions/checkout@v4\n with:\n repository: yaoapp/yao-startup-webapp\n submodules: true\n token: ${{ secrets.YAO_TEST_TOKEN }}\n path: yao-startup-webapp\n\n - name: Checkout Extension\n uses: actions/checkout@v4\n with:\n repository: yaoapp/yao-extensions-dev\n path: extension\n\n - name: Move Kun, Xun, Gou, V8Go\n run: |\n mv kun ../\n mv xun ../\n mv gou ../\n mv v8go ../\n mv app ../\n mv extension ../\n mv yao-startup-webapp ../\n ls -l .\n ls -l ../\n\n - name: Checkout pull request HEAD commit\n uses: actions/checkout@v4\n with:\n ref: ${{ env.HEAD }}\n\n - name: Setup Apple Private Key\n run: |\n mkdir -p ../app/openapi/certs/apple\n echo \"${{ secrets.APPLE_PRIVATE_KEY_USER }}\" > ../app/openapi/certs/apple/signin_client_secret_key.p8\n\n - name: Start Redis\n run: docker run --name redis --publish 6379:6379 --detach redis:${{ matrix.redis }}\n\n - name: Setup Go ${{ matrix.go }}\n uses: actions/setup-go@v5\n with:\n go-version: ${{ matrix.go }}\n\n - name: Install FFmpeg 7.x\n run: |\n wget https://github.com/BtbN/FFmpeg-Builds/releases/download/latest/ffmpeg-master-latest-linux64-gpl.tar.xz\n tar -xf ffmpeg-master-latest-linux64-gpl.tar.xz\n sudo cp ffmpeg-master-latest-linux64-gpl/bin/ffmpeg /usr/local/bin/\n sudo cp ffmpeg-master-latest-linux64-gpl/bin/ffprobe /usr/local/bin/\n sudo chmod +x /usr/local/bin/ffmpeg /usr/local/bin/ffprobe\n\n - name: Test FFmpeg\n run: ffmpeg -version\n\n - name: Install pdftoppm, mutool, imagemagick\n run: |\n sudo apt update\n sudo apt install -y poppler-utils mupdf-tools imagemagick\n\n - name: Test pdftoppm, mutool, imagemagick\n run: |\n pdftoppm -v\n mutool -v\n convert -version\n\n - name: Start MongoDB\n run: |\n docker run --name mongodb --publish 27017:27017 \\\n -e MONGO_INITDB_DATABASE=test \\\n -e MONGO_INITDB_ROOT_USERNAME=root \\\n -e MONGO_INITDB_ROOT_PASSWORD=123456 \\\n --detach mongo:${{ matrix.mongo }}\n # Wait for MongoDB to be ready\n for i in $(seq 1 20); do\n if docker exec mongodb mongosh --quiet --port 27017 --username root --password 123456 --eval \"db.serverStatus()\" > /dev/null 2>&1; then\n echo \"MongoDB is ready\"\n break\n fi\n echo \"Waiting for MongoDB... ($i)\"\n sleep 1\n done\n\n - name: Setup MySQL8.0 (connector)\n uses: ./.github/actions/setup-db\n with:\n kind: \"MySQL8.0\"\n db: \"test\"\n user: \"test\"\n password: \"123456\"\n port: \"3308\"\n\n - name: Setup ${{ matrix.db }}\n uses: ./.github/actions/setup-db\n with:\n kind: \"${{ matrix.db }}\"\n db: \"xiang\"\n user: \"xiang\"\n password: ${{ secrets.UNIT_PASS }}\n\n - name: Setup Go Tools\n run: |\n make tools\n\n - name: Setup ENV & Host\n env:\n PASSWORD: ${{ secrets.UNIT_PASS }}\n run: |\n sudo echo \"127.0.0.1 local.iqka.com\" | sudo tee -a /etc/hosts \n echo \"YAO_DB_DRIVER=$DB_DRIVER\" >> $GITHUB_ENV\n echo \"GITHUB_WORKSPACE:\\n\" && ls -l $GITHUB_WORKSPACE\n\n if [ \"$DB_DRIVER\" = \"mysql\" ]; then\n echo \"YAO_DB_PRIMARY=$DB_USER:$PASSWORD@$DB_HOST\" >> $GITHUB_ENV\n elif [ \"$DB_DRIVER\" = \"postgres\" ]; then\n echo \"YAO_DB_PRIMARY=postgres://$DB_USER:$PASSWORD@$DB_HOST\" >> $GITHUB_ENV\n else\n echo \"YAO_DB_PRIMARY=$YAO_ROOT/$DB_HOST\" >> $GITHUB_ENV\n fi\n\n echo \".:\\n\" && ls -l .\n echo \"..:\\n\" && ls -l ..\n ping -c 1 -t 1 local.iqka.com\n\n - name: Test Prepare\n run: |\n make vet\n make fmt-check\n make misspell-check\n\n - name: Run Core Tests (exclude AI)\n run: make unit-test-core\n\n - name: Codecov Report\n uses: codecov/codecov-action@v4\n with:\n token: ${{ secrets.CODECOV_TOKEN }} # not required for public repos\n\n - name: \"Comment on PR\"\n uses: actions/github-script@v7\n with:\n github-token: ${{ secrets.GITHUB_TOKEN }}\n script: |\n const { NR } = process.env\n var fs = require('fs');\n var issue_number = NR;\n await github.rest.issues.createComment({\n owner: context.repo.owner,\n repo: context.repo.repo,\n issue_number: issue_number,\n body: '✨DONE✨ db: ${{ matrix.db }} redis: ${{ matrix.redis }} mongo: ${{ matrix.mongo }} passed.'\n });\n\n # =============================================================================\n # Registry Client SDK Tests (requires Yao Registry Docker service)\n # =============================================================================\n RegistryTest:\n runs-on: ubuntu-latest\n services:\n yao-registry:\n image: yaoapp/registry:latest\n ports:\n - \"8080:8080\"\n env:\n REGISTRY_INIT_USER: yaoagents\n REGISTRY_INIT_PASS: yaoagents\n strategy:\n matrix:\n go: [\"1.25\"]\n if: >\n ${{ github.event.workflow_run.event == 'pull_request' &&\n github.event.workflow_run.conclusion == 'success' }}\n steps:\n - name: \"Download artifact\"\n uses: actions/github-script@v7\n with:\n script: |\n var artifacts = await github.rest.actions.listWorkflowRunArtifacts({\n owner: context.repo.owner,\n repo: context.repo.repo,\n run_id: ${{github.event.workflow_run.id }},\n });\n var matchArtifact = artifacts.data.artifacts.filter((artifact) => {\n return artifact.name == \"pr\"\n })[0];\n var download = await github.rest.actions.downloadArtifact({\n owner: context.repo.owner,\n repo: context.repo.repo,\n artifact_id: matchArtifact.id,\n archive_format: 'zip',\n });\n var fs = require('fs');\n fs.writeFileSync('${{github.workspace}}/pr.zip', Buffer.from(download.data));\n\n - name: \"Read NR & SHA\"\n run: |\n unzip pr.zip\n cat NR\n cat SHA\n echo HEAD=$(cat SHA) >> $GITHUB_ENV\n echo NR=$(cat NR) >> $GITHUB_ENV\n\n - name: \"Comment on PR\"\n uses: actions/github-script@v7\n with:\n github-token: ${{ secrets.GITHUB_TOKEN }}\n script: |\n const { NR } = process.env\n var issue_number = NR;\n await github.rest.issues.createComment({\n owner: context.repo.owner,\n repo: context.repo.repo,\n issue_number: issue_number,\n body: '🤖 Registry Client SDK Tests running...'\n });\n\n - name: Wait for Registry\n run: |\n for i in $(seq 1 15); do\n if curl -sf http://localhost:8080/.well-known/yao-registry > /dev/null 2>&1; then\n echo \"Registry is ready\"\n break\n fi\n echo \"Waiting for registry... ($i)\"\n sleep 1\n done\n\n - name: Checkout Kun\n uses: actions/checkout@v4\n with:\n repository: yaoapp/kun\n path: kun\n\n - name: Checkout Xun\n uses: actions/checkout@v4\n with:\n repository: yaoapp/xun\n path: xun\n\n - name: Checkout Gou\n uses: actions/checkout@v4\n with:\n repository: yaoapp/gou\n path: gou\n\n - name: Checkout V8Go\n uses: actions/checkout@v4\n with:\n repository: yaoapp/v8go\n path: v8go\n\n - name: Unzip libv8\n run: |\n files=$(find ./v8go -name \"libv8*.zip\")\n for file in $files; do\n dir=$(dirname \"$file\")\n echo \"Extracting $file to directory $dir\"\n unzip -o -d $dir $file\n rm -rf $dir/__MACOSX\n done\n\n - name: Checkout Demo App\n uses: actions/checkout@v4\n with:\n repository: yaoapp/yao-dev-app\n path: app\n\n - name: Checkout Extension\n uses: actions/checkout@v4\n with:\n repository: yaoapp/yao-extensions-dev\n path: extension\n\n - name: Move Dependencies\n run: |\n mv kun ../\n mv xun ../\n mv gou ../\n mv v8go ../\n mv app ../\n mv extension ../\n\n - name: Checkout pull request HEAD commit\n uses: actions/checkout@v4\n with:\n ref: ${{ env.HEAD }}\n\n - name: Setup Go ${{ matrix.go }}\n uses: actions/setup-go@v5\n with:\n go-version: ${{ matrix.go }}\n\n - name: Run Registry Client Tests\n env:\n YAO_REGISTRY_URL: http://localhost:8080\n YAO_TEST_APPLICATION: ${{ github.workspace }}/../app\n run: make unit-test-registry\n\n - name: Codecov Report\n uses: codecov/codecov-action@v4\n with:\n token: ${{ secrets.CODECOV_TOKEN }}\n\n - name: \"Comment on PR - Registry Tests Done\"\n uses: actions/github-script@v7\n with:\n github-token: ${{ secrets.GITHUB_TOKEN }}\n script: |\n const { NR } = process.env\n var issue_number = NR;\n await github.rest.issues.createComment({\n owner: context.repo.owner,\n repo: context.repo.repo,\n issue_number: issue_number,\n body: '✅ Registry Client SDK Tests passed!'\n });\n\n # =============================================================================\n # gRPC Tests - Run once with SQLite (transport layer, no DB matrix needed)\n # =============================================================================\n GRPCTest:\n runs-on: ubuntu-latest\n services:\n mongodb:\n image: mongo:6.0\n ports:\n - 27017:27017\n env:\n MONGO_INITDB_ROOT_USERNAME: root\n MONGO_INITDB_ROOT_PASSWORD: 123456\n MONGO_INITDB_DATABASE: test\n\n strategy:\n matrix:\n go: [\"1.25\"]\n if: >\n ${{ github.event.workflow_run.event == 'pull_request' &&\n github.event.workflow_run.conclusion == 'success' }}\n steps:\n - name: \"Download artifact\"\n uses: actions/github-script@v7\n with:\n script: |\n var artifacts = await github.rest.actions.listWorkflowRunArtifacts({\n owner: context.repo.owner,\n repo: context.repo.repo,\n run_id: ${{github.event.workflow_run.id }},\n });\n var matchArtifact = artifacts.data.artifacts.filter((artifact) => {\n return artifact.name == \"pr\"\n })[0];\n var download = await github.rest.actions.downloadArtifact({\n owner: context.repo.owner,\n repo: context.repo.repo,\n artifact_id: matchArtifact.id,\n archive_format: 'zip',\n });\n var fs = require('fs');\n fs.writeFileSync('${{github.workspace}}/pr.zip', Buffer.from(download.data));\n\n - name: \"Read NR & SHA\"\n run: |\n unzip pr.zip\n cat NR\n cat SHA\n echo HEAD=$(cat SHA) >> $GITHUB_ENV\n echo NR=$(cat NR) >> $GITHUB_ENV\n\n - name: \"Comment on PR\"\n uses: actions/github-script@v7\n with:\n github-token: ${{ secrets.GITHUB_TOKEN }}\n script: |\n const { NR } = process.env\n var issue_number = NR;\n await github.rest.issues.createComment({\n owner: context.repo.owner,\n repo: context.repo.repo,\n issue_number: issue_number,\n body: '🤖 gRPC Tests running with SQLite...'\n });\n\n - name: Checkout Kun\n uses: actions/checkout@v4\n with:\n repository: yaoapp/kun\n path: kun\n\n - name: Checkout Xun\n uses: actions/checkout@v4\n with:\n repository: yaoapp/xun\n path: xun\n\n - name: Checkout Gou\n uses: actions/checkout@v4\n with:\n repository: yaoapp/gou\n path: gou\n\n - name: Checkout V8Go\n uses: actions/checkout@v4\n with:\n repository: yaoapp/v8go\n path: v8go\n\n - name: Unzip libv8\n run: |\n files=$(find ./v8go -name \"libv8*.zip\")\n for file in $files; do\n dir=$(dirname \"$file\")\n echo \"Extracting $file to directory $dir\"\n unzip -o -d $dir $file\n rm -rf $dir/__MACOSX\n done\n\n - name: Checkout Demo App\n uses: actions/checkout@v4\n with:\n repository: yaoapp/yao-dev-app\n path: app\n\n - name: Checkout Extension\n uses: actions/checkout@v4\n with:\n repository: yaoapp/yao-extensions-dev\n path: extension\n\n - name: Move Dependencies\n run: |\n mv kun ../\n mv xun ../\n mv gou ../\n mv v8go ../\n mv app ../\n mv extension ../\n\n - name: Checkout pull request HEAD commit\n uses: actions/checkout@v4\n with:\n ref: ${{ env.HEAD }}\n\n - name: Setup Apple Private Key\n run: |\n mkdir -p ../app/openapi/certs/apple\n echo \"${{ secrets.APPLE_PRIVATE_KEY_USER }}\" > ../app/openapi/certs/apple/signin_client_secret_key.p8\n\n - name: Setup Go ${{ matrix.go }}\n uses: actions/setup-go@v5\n with:\n go-version: ${{ matrix.go }}\n\n - name: Start Redis\n run: docker run --name redis --publish 6379:6379 --detach redis:6\n\n - name: Setup Go Tools\n run: make tools\n\n - name: Setup ENV (SQLite)\n run: |\n mkdir -p ${{ github.WORKSPACE }}/../app/db\n echo \"YAO_DB_DRIVER=sqlite3\" >> $GITHUB_ENV\n echo \"YAO_DB_PRIMARY=${{ github.WORKSPACE }}/../app/db/yao.db\" >> $GITHUB_ENV\n\n - name: Run gRPC Tests\n run: make unit-test-grpc\n\n - name: Codecov Report\n uses: codecov/codecov-action@v4\n with:\n token: ${{ secrets.CODECOV_TOKEN }}\n\n - name: \"Comment on PR - gRPC Tests Done\"\n uses: actions/github-script@v7\n with:\n github-token: ${{ secrets.GITHUB_TOKEN }}\n script: |\n const { NR } = process.env\n var issue_number = NR;\n await github.rest.issues.createComment({\n owner: context.repo.owner,\n repo: context.repo.repo,\n issue_number: issue_number,\n body: '✅ gRPC Tests passed!'\n });\n" }, { "path": ".github/workflows/release-linux.yml", "content": "name: Release Linux\n\non:\n workflow_dispatch:\n push:\n tags:\n - \"v*\"\n\npermissions:\n contents: write\n\nenv:\n IMAGE_NAME: yaoapp/yao\n\njobs:\n # ===================================================================\n # Build Linux Binaries (amd64 + arm64)\n # ===================================================================\n build:\n runs-on: ubuntu-latest\n container:\n image: yaoapp/yao-build:1.0.0\n env:\n CF_ACCESS_KEY_ID: ${{ secrets.CF_ACCESS_KEY_ID }}\n CF_SECRET_ACCESS_KEY: ${{ secrets.CF_SECRET_ACCESS_KEY }}\n R2_BUCKET: ${{ secrets.R2_BUCKET }}\n R2_ACCOUNT_ID: ${{ secrets.R2_ACCOUNT_ID }}\n steps:\n - name: Configure R2 For Cloudflare\n run: |\n aws configure set aws_access_key_id $CF_ACCESS_KEY_ID\n aws configure set aws_secret_access_key $CF_SECRET_ACCESS_KEY\n aws configure set default.region us-east-1\n aws configure set default.s3.signature_version s3v4\n aws configure set default.s3.endpoint_url https://$R2_ACCOUNT_ID.r2.cloudflarestorage.com\n\n - name: Build\n run: |\n export PATH=$PATH:/github/home/go/bin\n\n # Clone dependencies\n cd /app\n git clone https://github.com/yaoapp/kun.git /app/kun\n git clone https://github.com/yaoapp/xun.git /app/xun\n git clone https://github.com/yaoapp/gou.git /app/gou\n git clone https://github.com/yaoapp/v8go.git /app/v8go\n git clone https://github.com/yaoapp/cui.git /app/cui-v1.0\n git clone https://github.com/yaoapp/yao-init.git /app/yao-init\n git clone https://github.com/yaoapp/yao.git /app/yao\n\n # Extract libv8\n files=$(find /app/v8go -name \"libv8*.zip\")\n for file in $files; do\n dir=$(dirname \"$file\")\n echo \"Extracting $file to directory $dir\"\n unzip -o -d $dir $file\n rm -rf $dir/__MACOSX\n done\n\n # Set VERSION from git tag (required)\n cd /app/yao\n if [[ \"$GITHUB_REF\" != refs/tags/v* ]]; then\n echo \"::error::This workflow requires a tag (refs/tags/v*). Got: $GITHUB_REF\"\n exit 1\n fi\n TAG_VERSION=\"${GITHUB_REF#refs/tags/v}\"\n echo \"Setting VERSION to $TAG_VERSION\"\n sed -i \"s/const VERSION = \\\".*\\\"/const VERSION = \\\"${TAG_VERSION}\\\"/g\" share/const.go\n grep 'const VERSION' share/const.go\n\n make tools && make artifacts-linux\n mv /app/yao/dist/release/* /data/\n ls -l /data\n\n - name: Push To R2\n run: |\n for file in /data/*; do\n aws s3 cp \"$file\" s3://$R2_BUCKET/archives/ \\\n --endpoint-url https://$R2_ACCOUNT_ID.r2.cloudflarestorage.com\n done\n\n - name: Upload Artifact\n uses: actions/upload-artifact@v4\n with:\n name: yao-linux\n path: /data/*\n\n # ===================================================================\n # Docker Images (multi-arch manifest: linux/amd64 + linux/arm64)\n # ===================================================================\n docker:\n needs: build\n runs-on: ubuntu-latest\n steps:\n - name: Checkout Code\n uses: actions/checkout@v4\n\n - name: Get Version\n id: version\n run: |\n if [[ \"$GITHUB_REF\" != refs/tags/v* ]]; then\n echo \"::error::This workflow requires a tag (refs/tags/v*). Got: $GITHUB_REF\"\n exit 1\n fi\n VERSION=\"${GITHUB_REF#refs/tags/v}\"\n echo \"version=${VERSION}\" >> $GITHUB_OUTPUT\n echo \"VERSION=${VERSION}\"\n\n - name: Set up QEMU\n uses: docker/setup-qemu-action@v3\n\n - name: Set up Docker Buildx\n uses: docker/setup-buildx-action@v3\n\n - name: Login to DockerHub\n uses: docker/login-action@v3\n with:\n username: ${{ secrets.DOCKER_USER }}\n password: ${{ secrets.DOCKER_TOKEN }}\n\n - name: Build & Push Development (multi-arch)\n uses: docker/build-push-action@v6\n with:\n context: ./docker/development\n platforms: linux/amd64,linux/arm64\n build-args: |\n VERSION=${{ steps.version.outputs.version }}\n push: true\n tags: |\n ${{ env.IMAGE_NAME }}:${{ steps.version.outputs.version }}-dev\n ${{ env.IMAGE_NAME }}:dev\n\n - name: Build & Push Production (multi-arch)\n uses: docker/build-push-action@v6\n with:\n context: ./docker/production\n platforms: linux/amd64,linux/arm64\n build-args: |\n VERSION=${{ steps.version.outputs.version }}\n push: true\n tags: |\n ${{ env.IMAGE_NAME }}:${{ steps.version.outputs.version }}\n ${{ env.IMAGE_NAME }}:latest\n" }, { "path": ".github/workflows/release-macos.yml", "content": "name: Release macOS\n\non:\n workflow_dispatch:\n push:\n tags:\n - \"v*\"\n\npermissions:\n contents: write\n\njobs:\n # ===================================================================\n # Build Yao macOS binaries (arm64 + amd64) — one job, both arches\n # ===================================================================\n build:\n runs-on: macos-latest\n steps:\n - name: Setup Node.js\n uses: actions/setup-node@v4\n with:\n node-version: 18\n\n - name: Install pnpm\n run: npm install -g pnpm\n\n - name: Setup Cache\n uses: actions/cache@v4\n with:\n path: |\n ~/.cache/go-build\n ~/go/pkg/mod\n key: ${{ runner.os }}-go-${{ hashFiles('**/go.sum') }}\n restore-keys: |\n ${{ runner.os }}-go-\n\n - name: Checkout Kun\n uses: actions/checkout@v4\n with:\n repository: yaoapp/kun\n path: kun\n\n - name: Checkout Xun\n uses: actions/checkout@v4\n with:\n repository: yaoapp/xun\n path: xun\n\n - name: Checkout Gou\n uses: actions/checkout@v4\n with:\n repository: yaoapp/gou\n path: gou\n\n - name: Checkout V8Go\n uses: actions/checkout@v4\n with:\n repository: yaoapp/v8go\n path: v8go\n\n - name: Unzip libv8\n run: |\n files=$(find ./v8go -name \"libv8*.zip\")\n for file in $files; do\n dir=$(dirname \"$file\")\n echo \"Extracting $file to directory $dir\"\n unzip -o -d $dir $file\n rm -rf $dir/__MACOSX\n done\n\n - name: Checkout CUI v1.0\n uses: actions/checkout@v4\n with:\n repository: yaoapp/cui\n path: cui-v1.0\n\n - name: Checkout Yao-Init\n uses: actions/checkout@v4\n with:\n repository: yaoapp/yao-init\n path: yao-init\n\n - name: Move Dependencies\n run: |\n mv kun ../\n mv xun ../\n mv gou ../\n mv v8go ../\n mv cui-v1.0 ../\n mv yao-init ../\n rm -f ../cui-v1.0/packages/setup/vite.config.ts.*\n\n - name: Checkout Yao\n uses: actions/checkout@v4\n\n - name: Set Version from Tag\n run: |\n if [[ \"$GITHUB_REF\" != refs/tags/v* ]]; then\n echo \"::error::This workflow requires a tag (refs/tags/v*). Got: $GITHUB_REF\"\n exit 1\n fi\n TAG=\"${GITHUB_REF#refs/tags/v}\"\n echo \"Setting VERSION to $TAG\"\n sed -i.bak \"s/const VERSION = \\\".*\\\"/const VERSION = \\\"${TAG}\\\"/g\" share/const.go\n rm -f share/const.go.bak\n grep 'const VERSION' share/const.go\n\n - name: Setup Go\n uses: actions/setup-go@v5\n with:\n go-version: \"1.25\"\n\n - name: Setup Go Tools\n run: make tools\n\n - name: Make Artifacts macOS\n run: make artifacts-macos\n\n - name: Get Version\n id: version\n run: |\n VERSION=$(grep 'const VERSION =' share/const.go | awk '{print $4}' | sed 's/\"//g')\n echo \"version=${VERSION}\" >> $GITHUB_OUTPUT\n\n - name: List Build Output\n run: ls -lh dist/release/\n\n - name: Install Certificates\n env:\n KEYCHAIN_PASSWORD: ${{ secrets.KEYCHAIN_PASSWORD }}\n run: |\n mkdir -p certs\n echo \"${{ secrets.APPLE_DEVELOPERIDG2CA }}\" | base64 --decode > certs/DeveloperIDG2CA.cer\n echo \"${{ secrets.APPLE_DISTRIBUTION }}\" | base64 --decode > certs/distribution.cer\n echo \"${{ secrets.APPLE_PRIVATE_KEY }}\" | base64 --decode > certs/private_key.p12\n security verify-cert -c certs/DeveloperIDG2CA.cer\n security verify-cert -c certs/distribution.cer\n\n - name: Import Certificates\n env:\n KEYCHAIN_PASSWORD: ${{ secrets.KEYCHAIN_PASSWORD }}\n run: |\n KEYCHAIN_PATH=$RUNNER_TEMP/app-signing.keychain-db\n security create-keychain -p \"$KEYCHAIN_PASSWORD\" $KEYCHAIN_PATH\n security set-keychain-settings -lut 21600 $KEYCHAIN_PATH\n security unlock-keychain -p \"$KEYCHAIN_PASSWORD\" $KEYCHAIN_PATH\n security import ./certs/DeveloperIDG2CA.cer -k $KEYCHAIN_PATH -T /usr/bin/codesign\n security import ./certs/distribution.cer -k $KEYCHAIN_PATH -T /usr/bin/codesign\n security import ./certs/private_key.p12 -k $KEYCHAIN_PATH -P \"${{ secrets.APPLE_PRIVATE_KEY_PASSWORD }}\" -T /usr/bin/codesign\n security list-keychain -d user -s $KEYCHAIN_PATH\n\n - name: Sign Yao Binaries\n run: |\n VERSION=\"${{ steps.version.outputs.version }}\"\n IDENTITY=\"Developer ID Application: ${{ secrets.APPLE_SIGN }}\"\n for ARCH in arm64 amd64; do\n for SUFFIX in \"\" \"-prod\"; do\n BIN=\"dist/release/yao-${VERSION}-darwin-${ARCH}${SUFFIX}\"\n codesign --force --verbose --timestamp --options runtime \\\n --entitlements .github/codesign/entitlements.plist \\\n --sign \"$IDENTITY\" \"$BIN\"\n codesign --verify --deep --strict --verbose=2 \"$BIN\"\n done\n done\n\n - name: Prepare Output and Checksums\n run: |\n VERSION=\"${{ steps.version.outputs.version }}\"\n for ARCH in arm64 amd64; do\n for VARIANT in dev prod; do\n if [ \"$VARIANT\" = \"dev\" ]; then\n SRC=\"dist/release/yao-${VERSION}-darwin-${ARCH}\"\n else\n SRC=\"dist/release/yao-${VERSION}-darwin-${ARCH}-prod\"\n fi\n DIR=\"/tmp/yao-output-${ARCH}-${VARIANT}\"\n mkdir -p \"$DIR\"\n cp \"$SRC\" \"$DIR/yao\"\n chmod +x \"$DIR/yao\"\n done\n done\n\n mkdir -p /tmp/checksums\n for ARCH in arm64 amd64; do\n for VARIANT in dev prod; do\n shasum -a 256 \"/tmp/yao-output-${ARCH}-${VARIANT}/yao\" | awk '{print $1\" yao\"}' > \"/tmp/checksums/yao-darwin-${ARCH}-${VARIANT}.sha256\"\n done\n done\n cat /tmp/checksums/*.sha256\n\n - name: Upload Artifacts\n uses: actions/upload-artifact@v4\n with:\n name: yao-darwin-arm64\n path: /tmp/yao-output-arm64-prod/yao\n\n - name: Upload arm64 Dev Binary\n uses: actions/upload-artifact@v4\n with:\n name: yao-darwin-arm64-dev\n path: /tmp/yao-output-arm64-dev/yao\n\n - name: Upload amd64 Binary\n uses: actions/upload-artifact@v4\n with:\n name: yao-darwin-amd64\n path: /tmp/yao-output-amd64-prod/yao\n\n - name: Upload amd64 Dev Binary\n uses: actions/upload-artifact@v4\n with:\n name: yao-darwin-amd64-dev\n path: /tmp/yao-output-amd64-dev/yao\n\n - name: Upload Checksums\n uses: actions/upload-artifact@v4\n with:\n name: yao-darwin-checksums\n path: /tmp/checksums/*.sha256\n" }, { "path": ".github/workflows/release.yml", "content": "name: Release\n\non:\n workflow_run:\n workflows: [\"Release Linux\", \"Release macOS\"]\n types:\n - completed\n\npermissions:\n contents: write\n\njobs:\n # ===================================================================\n # Wait for both workflows to succeed, then create a unified release\n # ===================================================================\n release:\n runs-on: ubuntu-latest\n if: >\n github.event.workflow_run.conclusion == 'success' &&\n startsWith(github.event.workflow_run.head_branch, 'v')\n steps:\n - name: Checkout Code\n uses: actions/checkout@v4\n\n - name: Get Version\n id: version\n run: |\n TAG=\"${{ github.event.workflow_run.head_branch }}\"\n VERSION=\"${TAG#v}\"\n echo \"version=${VERSION}\" >> $GITHUB_OUTPUT\n echo \"tag=${TAG}\" >> $GITHUB_OUTPUT\n echo \"TAG=${TAG} VERSION=${VERSION}\"\n\n - name: Wait for Both Workflows\n env:\n GH_TOKEN: ${{ secrets.GITHUB_TOKEN }}\n run: |\n TAG=\"${{ steps.version.outputs.tag }}\"\n echo \"Waiting for both Release Linux and Release macOS to complete for $TAG...\"\n\n for i in $(seq 1 60); do\n LINUX_STATUS=$(gh run list --workflow=\"Release Linux\" --branch=\"$TAG\" --limit=1 --json conclusion --jq '.[0].conclusion // \"pending\"')\n MACOS_STATUS=$(gh run list --workflow=\"Release macOS\" --branch=\"$TAG\" --limit=1 --json conclusion --jq '.[0].conclusion // \"pending\"')\n\n echo \"Attempt $i: Linux=$LINUX_STATUS macOS=$MACOS_STATUS\"\n\n if [ \"$LINUX_STATUS\" = \"success\" ] && [ \"$MACOS_STATUS\" = \"success\" ]; then\n echo \"Both workflows completed successfully.\"\n exit 0\n fi\n\n if [ \"$LINUX_STATUS\" = \"failure\" ] || [ \"$MACOS_STATUS\" = \"failure\" ]; then\n echo \"::error::One or both workflows failed (Linux=$LINUX_STATUS macOS=$MACOS_STATUS)\"\n exit 1\n fi\n\n sleep 60\n done\n\n echo \"::error::Timed out waiting for workflows\"\n exit 1\n\n - name: Download Linux Artifacts\n env:\n GH_TOKEN: ${{ secrets.GITHUB_TOKEN }}\n run: |\n TAG=\"${{ steps.version.outputs.tag }}\"\n LINUX_RUN_ID=$(gh run list --workflow=\"Release Linux\" --branch=\"$TAG\" --limit=1 --json databaseId --jq '.[0].databaseId')\n mkdir -p dist/linux\n gh run download \"$LINUX_RUN_ID\" --name yao-linux --dir dist/linux\n\n - name: Download macOS Artifacts\n env:\n GH_TOKEN: ${{ secrets.GITHUB_TOKEN }}\n run: |\n TAG=\"${{ steps.version.outputs.tag }}\"\n MACOS_RUN_ID=$(gh run list --workflow=\"Release macOS\" --branch=\"$TAG\" --limit=1 --json databaseId --jq '.[0].databaseId')\n mkdir -p dist/macos\n gh run download \"$MACOS_RUN_ID\" --name yao-darwin-arm64 --dir dist/macos/arm64-prod\n gh run download \"$MACOS_RUN_ID\" --name yao-darwin-arm64-dev --dir dist/macos/arm64-dev\n gh run download \"$MACOS_RUN_ID\" --name yao-darwin-amd64 --dir dist/macos/amd64-prod\n gh run download \"$MACOS_RUN_ID\" --name yao-darwin-amd64-dev --dir dist/macos/amd64-dev\n gh run download \"$MACOS_RUN_ID\" --name yao-darwin-checksums --dir dist/macos/checksums\n\n - name: Prepare Release Files\n run: |\n VERSION=\"${{ steps.version.outputs.version }}\"\n mkdir -p release\n\n # Linux artifacts (already named correctly from build.sh)\n cp dist/linux/* release/ 2>/dev/null || true\n\n # macOS prod binaries\n cp dist/macos/arm64-prod/yao \"release/yao-${VERSION}-darwin-arm64\"\n cp dist/macos/amd64-prod/yao \"release/yao-${VERSION}-darwin-amd64\"\n\n # macOS dev binaries\n cp dist/macos/arm64-dev/yao \"release/yao-${VERSION}-darwin-arm64-dev\"\n cp dist/macos/amd64-dev/yao \"release/yao-${VERSION}-darwin-amd64-dev\"\n\n # Checksums\n cp dist/macos/checksums/*.sha256 release/ 2>/dev/null || true\n\n chmod +x release/yao-* 2>/dev/null || true\n echo \"=== Release files ===\"\n ls -lh release/\n\n - name: Create GitHub Release\n uses: softprops/action-gh-release@v2\n with:\n tag_name: ${{ steps.version.outputs.tag }}\n name: Yao v${{ steps.version.outputs.version }}\n files: release/*\n generate_release_notes: true\n" }, { "path": ".github/workflows/unit-test-v1.yml", "content": "name: Unit Test V1\n\non:\n workflow_dispatch:\n inputs:\n tags:\n description: \"Version\"\n\nenv:\n CI_VERSION: \"1.0.0\"\n REPO_KUN: ${{ github.repository_owner }}/kun\n REPO_XUN: ${{ github.repository_owner }}/xun\n REPO_GOU: ${{ github.repository_owner }}/gou\n\n YAO_DEV: ${{ github.WORKSPACE }}\n YAO_ENV: development\n YAO_ROOT: ${{ github.WORKSPACE }}/../app\n YAO_HOST: 0.0.0.0\n YAO_PORT: 5099\n YAO_SESSION: \"memory\"\n YAO_LOG: \"./logs/application.log\"\n YAO_LOG_MODE: \"TEXT\"\n YAO_JWT_SECRET: \"bLp@bi!oqo-2U+hoTRUG\"\n YAO_DB_AESKEY: \"ZLX=T&f6refeCh-ro*r@\"\n\n YAO_EXTENSION_ROOT: ${{ github.WORKSPACE }}/../extension\n YAO_TEST_APPLICATION: ${{ github.WORKSPACE }}/../app\n\n YAO_RUNTIME_MIN: 3\n YAO_RUNTIME_MAX: 6\n YAO_RUNTIME_HEAP_LIMIT: 1500000000\n YAO_RUNTIME_HEAP_RELEASE: 10000000\n YAO_RUNTIME_HEAP_AVAILABLE: 550000000\n YAO_RUNTIME_PRECOMPILE: true\n\n MYSQL_TEST_HOST: \"127.0.0.1\"\n MYSQL_TEST_PORT: \"3308\"\n MYSQL_TEST_USER: \"test\"\n MYSQL_TEST_PASS: \"123456\"\n\n REDIS_TEST_HOST: \"127.0.0.1\"\n REDIS_TEST_PORT: \"6379\"\n REDIS_TEST_DB: \"2\"\n\n MONGO_TEST_HOST: \"127.0.0.1\"\n MONGO_TEST_PORT: \"27017\"\n MONGO_TEST_USER: \"root\"\n MONGO_TEST_PASS: \"123456\"\n\n PG_TEST_HOST: \"127.0.0.1\"\n PG_TEST_PORT: \"5432\"\n PG_TEST_USER: \"test\"\n PG_TEST_PASS: \"123456\"\n\njobs:\n # =============================================================================\n # Environment Setup & Verification\n # Build Yao, start services, connect Tai via gRPC tunnel, verify everything.\n # No tests are run — this job validates the CI environment is healthy.\n # =============================================================================\n setup-and-verify:\n runs-on: ubuntu-latest\n services:\n mongodb:\n image: mongo:6.0\n ports:\n - 27017:27017\n env:\n MONGO_INITDB_ROOT_USERNAME: root\n MONGO_INITDB_ROOT_PASSWORD: 123456\n MONGO_INITDB_DATABASE: test\n\n postgres:\n image: postgres:14\n ports:\n - 5432:5432\n env:\n POSTGRES_USER: test\n POSTGRES_PASSWORD: 123456\n POSTGRES_DB: test\n options: >-\n --health-cmd=\"pg_isready -U test\"\n --health-interval=10s\n --health-timeout=5s\n --health-retries=5\n\n strategy:\n matrix:\n go: [\"1.25\"]\n\n steps:\n # ==== Phase 1: Checkout & Setup ====\n - name: Checkout Yao\n uses: actions/checkout@v4\n\n - name: Setup Build Environment\n uses: ./.github/actions/setup-yao\n with:\n repo-kun: ${{ env.REPO_KUN }}\n repo-xun: ${{ env.REPO_XUN }}\n repo-gou: ${{ env.REPO_GOU }}\n checkout-init: \"true\"\n apple-private-key: ${{ secrets.APPLE_PRIVATE_KEY_USER }}\n\n - name: Load sandbox-v2 env\n run: grep -vE '^\\s*#|^\\s*$' .github/env/sandbox-v2.env >> $GITHUB_ENV\n\n - name: Setup SQLite\n run: |\n mkdir -p ${{ github.WORKSPACE }}/../app/db\n echo \"YAO_DB_PRIMARY=${{ github.WORKSPACE }}/../app/db/yao.db\" >> $GITHUB_ENV\n\n - name: Start MySQL 8.0\n run: |\n docker run -d --name mysql \\\n -e MYSQL_RANDOM_ROOT_PASSWORD=true \\\n -e MYSQL_USER=${MYSQL_TEST_USER} \\\n -e MYSQL_PASSWORD=${MYSQL_TEST_PASS} \\\n -e MYSQL_DATABASE=test \\\n -p ${MYSQL_TEST_PORT}:3306 \\\n mysql:8.0 --port=3306 --sql-mode='' \\\n --character-set-server=utf8mb4 --collation-server=utf8mb4_general_ci\n\n for i in $(seq 1 30); do\n if docker exec mysql mysqladmin ping -h 127.0.0.1 -u ${MYSQL_TEST_USER} -p${MYSQL_TEST_PASS} > /dev/null 2>&1; then\n echo \"MySQL ready\"\n break\n fi\n echo \"Waiting for MySQL... ($i/30)\"\n sleep 2\n done\n\n - name: Start Redis\n run: docker run --name redis -d -p 6379:6379 redis:6\n\n # ==== Phase 2: Code Quality & Build ====\n - name: Code Quality Check\n run: |\n make vet\n make fmt-check\n\n - name: Build Yao\n run: go build -v -o $RUNNER_TEMP/yao .\n\n - name: Build ci-token\n run: go build -tags ci -v -o $RUNNER_TEMP/ci-token ./cmd/ci-token\n\n - name: Extract tai binary from image\n run: |\n CID=$(docker create yaoapp/tai:latest)\n docker cp \"$CID\":/usr/local/bin/tai $RUNNER_TEMP/tai\n docker rm \"$CID\"\n chmod +x $RUNNER_TEMP/tai\n $RUNNER_TEMP/tai version || echo \"tai binary extracted\"\n\n # ==== Phase 3: Prepare & Start Yao ====\n - name: Prepare test app directory\n run: |\n cp -r ${{ github.WORKSPACE }}/../yao-init $RUNNER_TEMP/yao-test-app\n mkdir -p $RUNNER_TEMP/yao-test-app/db\n\n - name: Start Yao server\n run: |\n cd $RUNNER_TEMP/yao-test-app\n YAO_ROOT=$(pwd) \\\n YAO_HOST=0.0.0.0 \\\n YAO_PORT=5099 \\\n YAO_GRPC_HOST=0.0.0.0 \\\n YAO_GRPC_PORT=9099 \\\n YAO_DB_DRIVER=sqlite3 \\\n YAO_DB_PRIMARY=$(pwd)/db/yao.db \\\n YAO_SESSION=memory \\\n YAO_ENV=development \\\n YAO_JWT_SECRET=\"${{ env.YAO_JWT_SECRET }}\" \\\n YAO_DB_AESKEY=\"${{ env.YAO_DB_AESKEY }}\" \\\n $RUNNER_TEMP/yao start &\n\n # Wait for Yao HTTP to be ready (up to 120s)\n for i in $(seq 1 60); do\n if curl -sf http://127.0.0.1:5099/.well-known/yao > /dev/null 2>&1; then\n echo \"Yao HTTP ready\"\n curl -s http://127.0.0.1:5099/.well-known/yao | jq .\n break\n fi\n echo \"Waiting for Yao... ($i/60)\"\n sleep 2\n done\n\n curl -sf http://127.0.0.1:5099/.well-known/yao > /dev/null 2>&1 || {\n echo \"::error::Yao HTTP failed to start\"\n exit 1\n }\n\n # ==== Phase 4: Generate Tai credentials ====\n - name: Generate Tai credentials\n run: |\n gen_cred() {\n local CID=$1 TID=$2 OUT=$3\n local TOKEN\n TOKEN=$($RUNNER_TEMP/ci-token \\\n --app $RUNNER_TEMP/yao-test-app \\\n --client-id \"$CID\" \\\n --subject \"${YAO_CI_OAUTH_SUBJECT:-ci-tai}\" \\\n --user-id \"${YAO_CI_OAUTH_USER_ID}\" \\\n --team-id \"${YAO_CI_OAUTH_TEAM_ID}\" \\\n --scope \"${YAO_CI_OAUTH_SCOPE:-tai:tunnel}\" \\\n --ttl \"${YAO_CI_OAUTH_TTL:-24h}\" 2>/dev/null | tail -1 | tr -d '[:space:]')\n\n local JSON=\"{\\\"client_id\\\":\\\"$CID\\\",\\\"tai_id\\\":\\\"$TID\\\",\\\"machine_id\\\":\\\"ci-runner\\\",\\\"server\\\":\\\"http://127.0.0.1:${YAO_CI_HTTP_PORT}\\\",\\\"yao_grpc_addr\\\":\\\"127.0.0.1:${YAO_CI_GRPC_PORT}\\\",\\\"access_token\\\":\\\"$TOKEN\\\",\\\"scope\\\":\\\"${YAO_CI_OAUTH_SCOPE}\\\",\\\"expires_at\\\":\\\"2099-01-01T00:00:00Z\\\",\\\"registered\\\":true}\"\n echo -n \"$JSON\" | base64 -w0 > \"$OUT\"\n echo \"\"\n echo \"Credentials JSON (debug): $JSON\" | head -c 200\n echo \"...\"\n echo \"Generated credentials for $CID → $OUT\"\n }\n\n gen_cred tai-ci-local tai-local-001 $RUNNER_TEMP/tai-local-credentials\n gen_cred tai-ci-docker tai-docker-001 $RUNNER_TEMP/tai-docker-credentials\n gen_cred tai-ci-k8s tai-k8s-001 $RUNNER_TEMP/tai-k8s-credentials\n gen_cred tai-ci-hostexec tai-hostexec-001 $RUNNER_TEMP/tai-hostexec-credentials\n\n # ==== Phase 5: Pull images & Setup K8s ====\n - name: Pull test images\n run: |\n docker pull yaoapp/tai-sandbox-test:latest || true\n docker pull yaoapp/tai:latest\n docker pull alpine:latest\n\n - name: Install k3d & create cluster\n run: |\n curl -s https://raw.githubusercontent.com/k3d-io/k3d/main/install.sh | bash\n k3d cluster create tai-test --no-lb --wait --api-port ${YAO_CI_K3D_API_PORT}\n kubectl wait --for=condition=Ready node --all --timeout=60s\n k3d image import alpine:latest -c tai-test\n\n - name: Generate kubeconfig\n run: |\n K3D_IP=$(docker inspect k3d-tai-test-server-0 | jq -r '.[0].NetworkSettings.Networks[\"k3d-tai-test\"].IPAddress')\n echo \"k3d server IP: ${K3D_IP}\"\n\n k3d kubeconfig get tai-test > /tmp/kubeconfig-k3d.yml\n\n # For tai-k8s container (uses k3d internal IP)\n sed \"s|server: .*|server: https://${K3D_IP}:6443|\" /tmp/kubeconfig-k3d.yml \\\n > /tmp/kubeconfig-tai-k8s.yml\n echo \"Container kubeconfig server:\"\n grep server: /tmp/kubeconfig-tai-k8s.yml\n\n # For test runner (uses localhost via k3d port-mapped API)\n sed \"s|server: .*|server: https://127.0.0.1:${YAO_CI_K3D_API_PORT}|\" /tmp/kubeconfig-k3d.yml \\\n > $RUNNER_TEMP/kubeconfig-tai.yml\n echo \"Test runner kubeconfig server:\"\n grep server: $RUNNER_TEMP/kubeconfig-tai.yml\n\n # Export for later steps\n echo \"TAI_TEST_KUBECONFIG=$RUNNER_TEMP/kubeconfig-tai.yml\" >> $GITHUB_ENV\n echo \"YAO_CI_TAI_KUBECONFIG=$RUNNER_TEMP/kubeconfig-tai.yml\" >> $GITHUB_ENV\n\n # ==== Phase 6: Start Tai instances (host processes) ====\n - name: Start tai-local (DIRECT mode, auto-detect Docker)\n run: |\n mkdir -p $RUNNER_TEMP/tai-local-data\n\n TAI_CREDENTIALS=$RUNNER_TEMP/tai-local-credentials \\\n TAI_YAO_SERVER=http://127.0.0.1:${YAO_CI_HTTP_PORT} \\\n TAI_DATA_DIR=$RUNNER_TEMP/tai-local-data \\\n $RUNNER_TEMP/tai server \\\n --grpc 127.0.0.1:${YAO_CI_TAI_LOCAL_GRPC_PORT} \\\n --http 127.0.0.1:${YAO_CI_TAI_LOCAL_HTTP_PORT} \\\n --vnc 127.0.0.1:${YAO_CI_TAI_LOCAL_VNC_PORT} \\\n --docker 127.0.0.1:${YAO_CI_TAI_LOCAL_DOCKER_PORT} \\\n --direct \\\n --host-exec --host-exec-full-access \\\n --log-level debug &\n\n echo $! > $RUNNER_TEMP/tai-local.pid\n echo \"tai-local PID: $(cat $RUNNER_TEMP/tai-local.pid)\"\n\n for i in $(seq 1 30); do\n if curl -sf http://127.0.0.1:${YAO_CI_TAI_LOCAL_HTTP_PORT}/healthz > /dev/null 2>&1; then\n echo \"tai-local HTTP ready\"\n break\n fi\n echo \"Waiting for tai-local HTTP... ($i/30)\"\n sleep 1\n done\n\n - name: Start tai-docker (TUNNEL mode, Docker API proxy)\n run: |\n mkdir -p $RUNNER_TEMP/tai-docker-data\n\n TAI_CREDENTIALS=$RUNNER_TEMP/tai-docker-credentials \\\n TAI_YAO_SERVER=http://127.0.0.1:${YAO_CI_HTTP_PORT} \\\n TAI_DATA_DIR=$RUNNER_TEMP/tai-docker-data \\\n $RUNNER_TEMP/tai server \\\n --grpc 127.0.0.1:${YAO_CI_TAI_DOCKER_GRPC_PORT} \\\n --http 127.0.0.1:${YAO_CI_TAI_DOCKER_HTTP_PORT} \\\n --vnc 127.0.0.1:${YAO_CI_TAI_DOCKER_VNC_PORT} \\\n --docker 127.0.0.1:${YAO_CI_TAI_DOCKER_API_PORT} \\\n --host-exec --host-exec-full-access \\\n --log-level debug &\n\n echo $! > $RUNNER_TEMP/tai-docker.pid\n echo \"tai-docker PID: $(cat $RUNNER_TEMP/tai-docker.pid)\"\n\n for i in $(seq 1 30); do\n if curl -sf http://127.0.0.1:${YAO_CI_TAI_DOCKER_HTTP_PORT}/healthz > /dev/null 2>&1; then\n echo \"tai-docker HTTP ready\"\n break\n fi\n echo \"Waiting for tai-docker HTTP... ($i/30)\"\n sleep 1\n done\n\n - name: Start tai-k8s (TUNNEL mode, K8s API proxy)\n run: |\n mkdir -p $RUNNER_TEMP/tai-k8s-data\n\n TAI_CREDENTIALS=$RUNNER_TEMP/tai-k8s-credentials \\\n TAI_YAO_SERVER=http://127.0.0.1:${YAO_CI_HTTP_PORT} \\\n TAI_DATA_DIR=$RUNNER_TEMP/tai-k8s-data \\\n TAI_K8S_UPSTREAM=\"tcp://127.0.0.1:${YAO_CI_K3D_API_PORT}\" \\\n TAI_KUBECONFIG=$RUNNER_TEMP/kubeconfig-tai.yml \\\n $RUNNER_TEMP/tai server \\\n --grpc 127.0.0.1:${YAO_CI_TAI_K8S_GRPC_PORT} \\\n --http 127.0.0.1:${YAO_CI_TAI_K8S_HTTP_PORT} \\\n --vnc 127.0.0.1:${YAO_CI_TAI_K8S_VNC_PORT} \\\n --k8s 127.0.0.1:${YAO_CI_TAI_K8S_API_PORT} \\\n --docker=\"\" \\\n --host-exec --host-exec-full-access \\\n --log-level debug &\n\n echo $! > $RUNNER_TEMP/tai-k8s.pid\n echo \"tai-k8s PID: $(cat $RUNNER_TEMP/tai-k8s.pid)\"\n\n for i in $(seq 1 30); do\n if curl -sf http://127.0.0.1:${YAO_CI_TAI_K8S_HTTP_PORT}/healthz > /dev/null 2>&1; then\n echo \"tai-k8s HTTP ready\"\n break\n fi\n echo \"Waiting for tai-k8s HTTP... ($i/30)\"\n sleep 1\n done\n\n - name: Start tai-hostexec (TUNNEL mode, HostExec only, no runtime)\n run: |\n mkdir -p $RUNNER_TEMP/tai-hostexec-data\n\n TAI_CREDENTIALS=$RUNNER_TEMP/tai-hostexec-credentials \\\n TAI_YAO_SERVER=http://127.0.0.1:${YAO_CI_HTTP_PORT} \\\n TAI_DATA_DIR=$RUNNER_TEMP/tai-hostexec-data \\\n $RUNNER_TEMP/tai server \\\n --grpc 127.0.0.1:${YAO_CI_TAI_HOSTEXEC_GRPC_PORT} \\\n --http 127.0.0.1:${YAO_CI_TAI_HOSTEXEC_HTTP_PORT} \\\n --vnc 127.0.0.1:${YAO_CI_TAI_HOSTEXEC_VNC_PORT} \\\n --docker=\"\" \\\n --host-exec --host-exec-full-access \\\n --log-level debug &\n\n echo $! > $RUNNER_TEMP/tai-hostexec.pid\n echo \"tai-hostexec PID: $(cat $RUNNER_TEMP/tai-hostexec.pid)\"\n\n for i in $(seq 1 30); do\n if curl -sf http://127.0.0.1:${YAO_CI_TAI_HOSTEXEC_HTTP_PORT}/healthz > /dev/null 2>&1; then\n echo \"tai-hostexec HTTP ready\"\n break\n fi\n echo \"Waiting for tai-hostexec HTTP... ($i/30)\"\n sleep 1\n done\n\n # ==== Phase 7: Environment Verification (fail fast) ====\n - name: Verify Environment\n run: |\n echo \"CI Environment v${CI_VERSION}\"\n echo \"\"\n\n FAILED=0\n check() {\n local name=$1; shift\n if \"$@\" > /dev/null 2>&1; then\n echo \" [PASS] $name\"\n else\n echo \" [FAIL] $name\"\n FAILED=$((FAILED + 1))\n fi\n }\n\n echo \"=== Environment Verification ===\"\n\n # ── 1. Service Health ──\n echo \"\"\n echo \"--- 1. Service Health ---\"\n\n echo \"[Yao]\"\n check \"Yao HTTP (/.well-known/yao)\" curl -sf http://127.0.0.1:5099/.well-known/yao\n check \"Yao gRPC port\" nc -z 127.0.0.1 9099\n\n echo \"[tai-local (DIRECT)]\"\n check \"tai-local process alive\" kill -0 $(cat $RUNNER_TEMP/tai-local.pid 2>/dev/null || echo 0)\n check \"tai-local HTTP (/healthz)\" curl -sf http://127.0.0.1:${YAO_CI_TAI_LOCAL_HTTP_PORT}/healthz\n check \"tai-local gRPC reachable (direct)\" nc -z 127.0.0.1 ${YAO_CI_TAI_LOCAL_GRPC_PORT}\n\n echo \"[tai-docker (TUNNEL)]\"\n check \"tai-docker process alive\" kill -0 $(cat $RUNNER_TEMP/tai-docker.pid 2>/dev/null || echo 0)\n check \"tai-docker HTTP (/healthz)\" curl -sf http://127.0.0.1:${YAO_CI_TAI_DOCKER_HTTP_PORT}/healthz\n check \"tai-docker gRPC listener\" nc -z 127.0.0.1 ${YAO_CI_TAI_DOCKER_GRPC_PORT}\n\n echo \"[tai-k8s (TUNNEL)]\"\n check \"tai-k8s process alive\" kill -0 $(cat $RUNNER_TEMP/tai-k8s.pid 2>/dev/null || echo 0)\n check \"tai-k8s HTTP (/healthz)\" curl -sf http://127.0.0.1:${YAO_CI_TAI_K8S_HTTP_PORT}/healthz\n check \"tai-k8s gRPC listener\" nc -z 127.0.0.1 ${YAO_CI_TAI_K8S_GRPC_PORT}\n\n echo \"[tai-hostexec (TUNNEL)]\"\n check \"tai-hostexec process alive\" kill -0 $(cat $RUNNER_TEMP/tai-hostexec.pid 2>/dev/null || echo 0)\n check \"tai-hostexec HTTP (/healthz)\" curl -sf http://127.0.0.1:${YAO_CI_TAI_HOSTEXEC_HTTP_PORT}/healthz\n check \"tai-hostexec gRPC listener\" nc -z 127.0.0.1 ${YAO_CI_TAI_HOSTEXEC_GRPC_PORT}\n\n echo \"[K8s (k3d via direct API)]\"\n check \"kubectl get nodes (k3d direct)\" kubectl --kubeconfig=$RUNNER_TEMP/kubeconfig-tai.yml get nodes\n\n echo \"[Data Stores]\"\n MONGO_CID=$(docker ps -qf \"ancestor=mongo:6.0\" | head -1)\n check \"MongoDB ping\" docker exec \"$MONGO_CID\" mongosh --quiet \\\n -u ${MONGO_TEST_USER} -p ${MONGO_TEST_PASS} --authenticationDatabase admin \\\n --eval \"db.runCommand({ping:1})\"\n\n check \"MySQL ping\" docker exec mysql mysqladmin ping -h 127.0.0.1 \\\n -u ${MYSQL_TEST_USER} -p${MYSQL_TEST_PASS}\n\n PG_CID=$(docker ps -qf \"ancestor=postgres:14\" | head -1)\n check \"PostgreSQL ping\" docker exec \"$PG_CID\" pg_isready -U ${PG_TEST_USER}\n\n check \"Redis ping\" docker exec redis redis-cli ping\n\n # ── 2. Network Topology ──\n echo \"\"\n echo \"--- 2. Network Topology ---\"\n\n BRIDGE_IP=${YAO_CI_BRIDGE_IP}\n\n echo \"[Host network basics]\"\n check \"docker0 bridge exists\" ip addr show docker0\n check \"Bridge IP reachable (${BRIDGE_IP})\" ping -c1 -W2 ${BRIDGE_IP}\n check \"Docker socket accessible\" test -S /var/run/docker.sock\n check \"tai binary on runner\" test -x $RUNNER_TEMP/tai\n\n echo \"[Yao endpoints (all Tai instances need these)]\"\n check \"Yao HTTP :${YAO_CI_HTTP_PORT}\" curl -sf http://127.0.0.1:${YAO_CI_HTTP_PORT}/.well-known/yao\n check \"Yao gRPC :${YAO_CI_GRPC_PORT}\" nc -z 127.0.0.1 ${YAO_CI_GRPC_PORT}\n\n echo \"[DIRECT path: Yao → tai-local]\"\n check \"Yao→tai-local gRPC :${YAO_CI_TAI_LOCAL_GRPC_PORT}\" nc -z 127.0.0.1 ${YAO_CI_TAI_LOCAL_GRPC_PORT}\n check \"Yao→tai-local HTTP :${YAO_CI_TAI_LOCAL_HTTP_PORT}\" curl -sf http://127.0.0.1:${YAO_CI_TAI_LOCAL_HTTP_PORT}/healthz\n check \"tai-local Docker API proxy :${YAO_CI_TAI_LOCAL_DOCKER_PORT}\" nc -z 127.0.0.1 ${YAO_CI_TAI_LOCAL_DOCKER_PORT}\n\n echo \"[TUNNEL path: tai-docker → Yao gRPC (reverse tunnel)]\"\n check \"tai-docker→Yao gRPC :${YAO_CI_GRPC_PORT}\" nc -z 127.0.0.1 ${YAO_CI_GRPC_PORT}\n check \"tai-docker Docker API proxy :${YAO_CI_TAI_DOCKER_API_PORT}\" nc -z 127.0.0.1 ${YAO_CI_TAI_DOCKER_API_PORT}\n check \"Docker API via proxy\" bash -c \"curl -sf http://127.0.0.1:${YAO_CI_TAI_DOCKER_API_PORT}/version | jq -r .ApiVersion\"\n\n echo \"[TUNNEL path: tai-k8s → Yao gRPC (reverse tunnel)]\"\n check \"tai-k8s→Yao gRPC :${YAO_CI_GRPC_PORT}\" nc -z 127.0.0.1 ${YAO_CI_GRPC_PORT}\n check \"tai-k8s K8s API proxy :${YAO_CI_TAI_K8S_API_PORT}\" nc -z 127.0.0.1 ${YAO_CI_TAI_K8S_API_PORT}\n sed \"s|server: .*|server: https://127.0.0.1:${YAO_CI_TAI_K8S_API_PORT}|\" $RUNNER_TEMP/kubeconfig-tai.yml \\\n > $RUNNER_TEMP/kubeconfig-tai-proxy.yml\n check \"K8s API via tai-k8s proxy (kubectl)\" kubectl --kubeconfig=$RUNNER_TEMP/kubeconfig-tai-proxy.yml --insecure-skip-tls-verify get nodes\n\n echo \"[TUNNEL path: tai-hostexec → Yao gRPC (reverse tunnel)]\"\n check \"tai-hostexec→Yao gRPC :${YAO_CI_GRPC_PORT}\" nc -z 127.0.0.1 ${YAO_CI_GRPC_PORT}\n\n echo \"[Data Store connectivity from runner]\"\n check \"MongoDB :27017\" nc -z 127.0.0.1 27017\n check \"Redis :6379\" nc -z 127.0.0.1 6379\n check \"MySQL :${MYSQL_TEST_PORT}\" nc -z 127.0.0.1 ${MYSQL_TEST_PORT}\n check \"PostgreSQL :${PG_TEST_PORT}\" nc -z 127.0.0.1 ${PG_TEST_PORT}\n\n # ── 3. Connection Mode Verification ──\n echo \"\"\n echo \"--- 3. Connection Modes ---\"\n sleep 5\n\n echo \"[Credentials (4 tokens)]\"\n check \"tai-local credentials exist\" test -f $RUNNER_TEMP/tai-local-credentials\n check \"tai-docker credentials exist\" test -f $RUNNER_TEMP/tai-docker-credentials\n check \"tai-k8s credentials exist\" test -f $RUNNER_TEMP/tai-k8s-credentials\n check \"tai-hostexec credentials exist\" test -f $RUNNER_TEMP/tai-hostexec-credentials\n\n echo \"[DIRECT: tai-local → Yao HTTP register → Yao dials tai-local gRPC]\"\n echo \" tai-local registers via POST /tai-nodes/register\"\n echo \" Yao dials back tai-local gRPC at 127.0.0.1:${YAO_CI_TAI_LOCAL_GRPC_PORT}\"\n\n echo \"[TUNNEL: tai-docker → Yao gRPC :${YAO_CI_GRPC_PORT} (Register + Forward)]\"\n echo \" Sandbox connects Yao gRPC → Forward stream → tai-docker :${YAO_CI_TAI_DOCKER_GRPC_PORT}\"\n\n echo \"[TUNNEL: tai-k8s → Yao gRPC :${YAO_CI_GRPC_PORT} (Register + Forward)]\"\n echo \" Sandbox connects Yao gRPC → Forward stream → tai-k8s :${YAO_CI_TAI_K8S_GRPC_PORT}\"\n\n echo \"[TUNNEL: tai-hostexec → Yao gRPC :${YAO_CI_GRPC_PORT} (Register + Forward)]\"\n echo \" HostExec only, no container runtime\"\n\n WELL_KNOWN=$(curl -sf http://127.0.0.1:5099/.well-known/yao 2>/dev/null || echo \"{}\")\n echo \"\"\n echo \" Yao .well-known/yao:\"\n echo \"$WELL_KNOWN\" | jq . 2>/dev/null || echo \" $WELL_KNOWN\"\n\n # ── 4. HostExec Readiness ──\n echo \"\"\n echo \"--- 4. HostExec ---\"\n\n echo \"[HostExec gRPC ports (all 4 instances)]\"\n check \"HostExec tai-local (direct, auto-Docker)\" nc -z 127.0.0.1 ${YAO_CI_TAI_LOCAL_GRPC_PORT}\n check \"HostExec tai-docker (tunnel, Docker proxy)\" nc -z 127.0.0.1 ${YAO_CI_TAI_DOCKER_GRPC_PORT}\n check \"HostExec tai-k8s (tunnel, K8s proxy)\" nc -z 127.0.0.1 ${YAO_CI_TAI_K8S_GRPC_PORT}\n check \"HostExec tai-hostexec (tunnel, no runtime)\" nc -z 127.0.0.1 ${YAO_CI_TAI_HOSTEXEC_GRPC_PORT}\n\n echo \"\"\n echo \"==========================================\"\n if [ $FAILED -gt 0 ]; then\n echo \"::error::$FAILED verification check(s) FAILED\"\n echo \"\"\n echo \"=== Diagnostic Info ===\"\n echo \"--- Docker containers ---\"\n docker ps -a\n echo \"\"\n echo \"--- Processes (yao + tai) ---\"\n ps aux | grep -E \"yao|tai\" | grep -v grep || true\n echo \"\"\n echo \"--- Listening ports ---\"\n ss -tlnp | grep -E \"5099|9099|19100|19101|19102|19103|8099|8100|8101|8102|12375|12376|16443|16444\" || true\n exit 1\n else\n echo \"All verification checks PASSED\"\n fi\n" }, { "path": ".github/workflows/unit-test.yml", "content": "name: Unit Test\n\non:\n workflow_dispatch:\n inputs:\n tags:\n description: \"Version\"\n push:\n branches: [main]\n\nenv:\n YAO_DEV: ${{ github.WORKSPACE }}\n YAO_ENV: development\n YAO_ROOT: ${{ github.WORKSPACE }}/../app\n YAO_HOST: 0.0.0.0\n YAO_PORT: 5099\n YAO_SESSION: \"memory\"\n YAO_LOG: \"./logs/application.log\"\n YAO_LOG_MODE: \"TEXT\"\n YAO_JWT_SECRET: \"bLp@bi!oqo-2U+hoTRUG\"\n YAO_DB_AESKEY: \"ZLX=T&f6refeCh-ro*r@\"\n OSS_TEST_ID: ${{ secrets.OSS_TEST_ID}}\n OSS_TEST_SECRET: ${{ secrets.OSS_TEST_SECRET}}\n ROOT_PLUGIN: ${{ github.WORKSPACE }}/../../../data/gou-unit/plugins\n REPO_KUN: ${{ github.repository_owner }}/kun\n REPO_XUN: ${{ github.repository_owner }}/xun\n REPO_GOU: ${{ github.repository_owner }}/gou\n\n MYSQL_TEST_HOST: \"127.0.0.1\"\n MYSQL_TEST_PORT: \"3308\"\n MYSQL_TEST_USER: test\n MYSQL_TEST_PASS: \"123456\"\n\n SQLITE_DB: \"./app/db/yao.db\"\n\n REDIS_TEST_HOST: \"127.0.0.1\"\n REDIS_TEST_PORT: \"6379\"\n REDIS_TEST_DB: \"2\"\n\n MONGO_TEST_HOST: \"127.0.0.1\"\n MONGO_TEST_PORT: \"27017\"\n MONGO_TEST_USER: \"root\"\n MONGO_TEST_PASS: \"123456\"\n\n OPENAI_TEST_KEY: ${{ secrets.OPENAI_TEST_KEY }}\n TEST_MOAPI_SECRET: ${{ secrets.OPENAI_TEST_KEY }}\n OPENAI_API_KEY: ${{ secrets.OPENAI_TEST_KEY }}\n ANTHROPIC_API_KEY: ${{ secrets.ANTHROPIC_API_KEY }}\n TEST_MOAPI_MIRROR: https://api.openai.com\n\n # DeepSeek API Configuration\n DEEPSEEK_API_KEY: ${{ secrets.DEEPSEEK_API_KEY }}\n DEEPSEEK_API_PROXY: ${{ secrets.DEEPSEEK_API_PROXY }}\n DEEPSEEK_MODELS_R1: ${{ secrets.DEEPSEEK_MODELS_R1 }}\n DEEPSEEK_MODELS_V3: ${{ secrets.DEEPSEEK_MODELS_V3 }}\n DEEPSEEK_MODELS_V3_1: ${{ secrets.DEEPSEEK_MODELS_V3_1 }}\n\n # Search API Configuration\n TAVILY_API_KEY: ${{ secrets.TAVILY_API_KEY }}\n SERPAPI_API_KEY: ${{ secrets.SERPAPI_API_KEY }}\n SERPER_API_KEY: ${{ secrets.SERPER_API_KEY }}\n\n # Claude API Configuration\n CLAUDE_API_KEY: ${{ secrets.CLAUDE_API_KEY }}\n CLAUDE_PROXY: ${{ secrets.CLAUDE_PROXY }}\n CLAUDE_API_HOST: ${{ secrets.CLAUDE_API_HOST }}\n CLAUDE_SONNET_4: ${{ secrets.CLAUDE_SONNET_4 }}\n CLAUDE_SONNET_4_THINKING: ${{ secrets.CLAUDE_SONNET_4_THINKING }}\n\n # Moonshot / Kimi API Configuration\n MOONSHOT_API_KEY: ${{ secrets.MOONSHOT_API_KEY }}\n MOONSHOT_PROXY: \"https://api.moonshot.cn\"\n KIMI_CODE_API_KEY: ${{ secrets.KIMI_CODE_API_KEY }}\n KIMI_CODE_PROXY: \"https://api.kimi.com/coding\"\n\n TAB_NAME: \"::PET ADMIN\"\n PAGE_SIZE: \"20\"\n PAGE_LINK: \"https://yaoapps.com\"\n PAGE_ICON: \"icon-trash\"\n\n DEMO_APP: ${{ github.WORKSPACE }}/../app\n\n # Application Setting\n\n ## Path\n YAO_EXTENSION_ROOT: ${{ github.WORKSPACE }}/../extension\n YAO_TEST_APPLICATION: ${{ github.WORKSPACE }}/../app\n YAO_SUI_TEST_APPLICATION: ${{ github.WORKSPACE }}/../yao-startup-webapp\n\n ## Runtime\n YAO_RUNTIME_MIN: 3\n YAO_RUNTIME_MAX: 6\n YAO_RUNTIME_HEAP_LIMIT: 1500000000\n YAO_RUNTIME_HEAP_RELEASE: 10000000\n YAO_RUNTIME_HEAP_AVAILABLE: 550000000\n YAO_RUNTIME_PRECOMPILE: true\n\n # Neo4j\n NEO4J_TEST_URL: \"neo4j://localhost:7687\"\n NEO4J_TEST_USER: \"neo4j\"\n NEO4J_TEST_PASS: \"Yao2026Neo4j\"\n\n # Qdrant\n QDRANT_TEST_HOST: \"127.0.0.1\"\n QDRANT_TEST_PORT: \"6334\"\n\n # S3\n S3_API: ${{ secrets.S3_API }}\n S3_ACCESS_KEY: ${{ secrets.S3_ACCESS_KEY }}\n S3_SECRET_KEY: ${{ secrets.S3_SECRET_KEY }}\n S3_BUCKET: ${{ secrets.S3_BUCKET }}\n S3_PUBLIC_URL: ${{ secrets.S3_PUBLIC_URL }}\n\n # === Openapi Signin Configs ===\n SIGNIN_CLIENT_ID: \"kiCeR88kDwHBDuNHvN51cZgmpp3tmF6Z\"\n\n ## Google\n GOOGLE_CLIENT_ID: ${{ secrets.GOOGLE_CLIENT_ID }}\n GOOGLE_CLIENT_SECRET: ${{ secrets.GOOGLE_CLIENT_SECRET }}\n\n ## Microsoft\n MICROSOFT_CLIENT_ID: ${{ secrets.MICROSOFT_CLIENT_ID }}\n MICROSOFT_CLIENT_SECRET: ${{ secrets.MICROSOFT_CLIENT_SECRET }}\n\n ## Apple\n APPLE_SERVICE_ID: ${{ secrets.APPLE_SERVICE_ID }}\n APPLE_PRIVATE_KEY_PATH: \"apple/signin_client_secret_key.p8\"\n APPLE_KEY_ID: ${{ secrets.APPLE_KEY_ID }}\n APPLE_TEAM_ID: ${{ secrets.APPLE_TEAM_ID }}\n\n ## Github\n GITHUBUSER_CLIENT_ID: ${{ secrets.GITHUBUSER_CLIENT_ID }}\n GITHUBUSER_CLIENT_SECRET: ${{ secrets.GITHUBUSER_CLIENT_SECRET }}\n\n ## Cloudflare Turnstile\n CLOUDFLARE_TURNSTILE_SITEKEY: ${{ secrets.CLOUDFLARE_TURNSTILE_SITEKEY }}\n CLOUDFLARE_TURNSTILE_SECRET: ${{ secrets.CLOUDFLARE_TURNSTILE_SECRET }}\n\n # === Messaging Services ===\n ## Mailgun\n MAILGUN_DOMAIN: ${{ secrets.MAILGUN_DOMAIN }}\n MAILGUN_API_KEY: ${{ secrets.MAILGUN_API_KEY }}\n MAILGUN_FROM: \"Yaobots Tests \"\n\n ## SMTP Server（ Mailgun ）\n SMTP_HOST: \"smtp.mailgun.org\"\n SMTP_PORT: \"465\"\n SMTP_USERNAME: ${{ secrets.SMTP_USERNAME }}\n SMTP_PASSWORD: ${{ secrets.SMTP_PASSWORD }}\n SMTP_FROM: \"Yaobots SMTP Tests \"\n\n ## SMTP Server（ Gmail ）\n RELIABLE_SMTP_HOST: \"smtp.gmail.com\"\n RELIABLE_SMTP_PORT: \"465\"\n RELIABLE_SMTP_USERNAME: ${{ secrets.RELIABLE_SMTP_USERNAME }}\n RELIABLE_SMTP_PASSWORD: ${{ secrets.RELIABLE_SMTP_PASSWORD }}\n RELIABLE_SMTP_FROM: \"Yaobots Gmail Tests \"\n\n ## IMAP Server (Gmail)\n RELIABLE_IMAP_HOST: \"imap.gmail.com\"\n RELIABLE_IMAP_PORT: \"993\"\n RELIABLE_IMAP_USERNAME: ${{ secrets.RELIABLE_SMTP_USERNAME }}\n RELIABLE_IMAP_PASSWORD: ${{ secrets.RELIABLE_SMTP_PASSWORD }}\n RELIABLE_IMAP_MAILBOX: \"INBOX\"\n\n ## Twilio\n TWILIO_ACCOUNT_SID: ${{ secrets.TWILIO_ACCOUNT_SID }}\n TWILIO_AUTH_TOKEN: ${{ secrets.TWILIO_AUTH_TOKEN }}\n TWILIO_API_SID: ${{ secrets.TWILIO_API_SID }}\n TWILIO_API_KEY: ${{ secrets.TWILIO_API_KEY }}\n TWILIO_SENDGRID_API_SID: ${{ secrets.TWILIO_SENDGRID_API_SID }}\n TWILIO_SENDGRID_API_KEY: ${{ secrets.TWILIO_SENDGRID_API_KEY }}\n TWILIO_FROM_PHONE: \"+17035701412\"\n TWILIO_FROM_EMAIL: \"unit-test@sendgrid.yaobots.com\"\n TWILIO_TEST_PHONE: ${{ secrets.TWILIO_TEST_PHONE }}\n\njobs:\n # =============================================================================\n # KB Tests (kb) - Run once with SQLite (requires Qdrant, Neo4j, FastEmbed)\n # =============================================================================\n kb-test:\n runs-on: ubuntu-latest\n services:\n qdrant:\n image: qdrant/qdrant:latest\n ports:\n - 6333:6333\n - 6334:6334\n\n fastembed:\n image: yaoapp/fastembed:latest-amd64\n env:\n FASTEMBED_PASSWORD: Yao@2026\n ports:\n - 6001:8000\n\n neo4j:\n image: neo4j:latest\n ports:\n - \"7687:7687\"\n env:\n NEO4J_AUTH: neo4j/Yao2026Neo4j\n\n mongodb:\n image: mongo:6.0\n ports:\n - 27017:27017\n env:\n MONGO_INITDB_ROOT_USERNAME: root\n MONGO_INITDB_ROOT_PASSWORD: 123456\n MONGO_INITDB_DATABASE: test\n\n strategy:\n matrix:\n go: [\"1.25\"]\n steps:\n - name: Checkout Kun\n uses: actions/checkout@v4\n with:\n repository: ${{ env.REPO_KUN }}\n path: kun\n\n - name: Checkout Xun\n uses: actions/checkout@v4\n with:\n repository: ${{ env.REPO_XUN }}\n path: xun\n\n - name: Checkout Gou\n uses: actions/checkout@v4\n with:\n repository: ${{ env.REPO_GOU }}\n path: gou\n\n - name: Checkout V8Go\n uses: actions/checkout@v4\n with:\n repository: yaoapp/v8go\n path: v8go\n\n - name: Unzip libv8\n run: |\n files=$(find ./v8go -name \"libv8*.zip\")\n for file in $files; do\n dir=$(dirname \"$file\")\n echo \"Extracting $file to directory $dir\"\n unzip -o -d $dir $file\n rm -rf $dir/__MACOSX\n done\n\n - name: Checkout Demo App\n uses: actions/checkout@v4\n with:\n repository: yaoapp/yao-dev-app\n path: app\n\n - name: Checkout Extension\n uses: actions/checkout@v4\n with:\n repository: yaoapp/yao-extensions-dev\n path: extension\n\n - name: Move Dependencies\n run: |\n mv kun ../\n mv xun ../\n mv gou ../\n mv v8go ../\n mv app ../\n mv extension ../\n\n - name: Checkout Code\n uses: actions/checkout@v4\n\n - name: Setup Apple Private Key\n run: |\n mkdir -p ../app/openapi/certs/apple\n echo \"${{ secrets.APPLE_PRIVATE_KEY_USER }}\" > ../app/openapi/certs/apple/signin_client_secret_key.p8\n\n - name: Setup Go ${{ matrix.go }}\n uses: actions/setup-go@v5\n with:\n go-version: ${{ matrix.go }}\n\n - name: Start Redis\n run: docker run --name redis --publish 6379:6379 --detach redis:6\n\n - name: Setup Go Tools\n run: make tools\n\n - name: Setup ENV (SQLite)\n run: |\n mkdir -p ${{ github.WORKSPACE }}/../app/db\n echo \"YAO_DB_DRIVER=sqlite3\" >> $GITHUB_ENV\n echo \"YAO_DB_PRIMARY=${{ github.WORKSPACE }}/../app/db/yao.db\" >> $GITHUB_ENV\n\n - name: Run KB Tests (kb)\n run: make unit-test-kb\n\n - name: Codecov Report\n uses: codecov/codecov-action@v4\n with:\n token: ${{ secrets.CODECOV_TOKEN }}\n\n # =============================================================================\n # Agent Tests (agent, aigc) - Run once with SQLite\n # =============================================================================\n agent-test:\n runs-on: ubuntu-latest\n services:\n qdrant:\n image: qdrant/qdrant:latest\n ports:\n - 6333:6333\n - 6334:6334\n\n fastembed:\n image: yaoapp/fastembed:latest-amd64\n env:\n FASTEMBED_PASSWORD: Yao@2026\n ports:\n - 6001:8000\n\n neo4j:\n image: neo4j:latest\n ports:\n - \"7687:7687\"\n env:\n NEO4J_AUTH: neo4j/Yao2026Neo4j\n\n mcp-everything:\n image: yaoapp/mcp-everything:latest\n ports:\n - \"3021:3021\"\n - \"3022:3022\"\n\n mongodb:\n image: mongo:6.0\n ports:\n - 27017:27017\n env:\n MONGO_INITDB_ROOT_USERNAME: root\n MONGO_INITDB_ROOT_PASSWORD: 123456\n MONGO_INITDB_DATABASE: test\n\n strategy:\n matrix:\n go: [\"1.25\"]\n steps:\n - name: Checkout Kun\n uses: actions/checkout@v4\n with:\n repository: ${{ env.REPO_KUN }}\n path: kun\n\n - name: Checkout Xun\n uses: actions/checkout@v4\n with:\n repository: ${{ env.REPO_XUN }}\n path: xun\n\n - name: Checkout Gou\n uses: actions/checkout@v4\n with:\n repository: ${{ env.REPO_GOU }}\n path: gou\n\n - name: Checkout V8Go\n uses: actions/checkout@v4\n with:\n repository: yaoapp/v8go\n path: v8go\n\n - name: Unzip libv8\n run: |\n files=$(find ./v8go -name \"libv8*.zip\")\n for file in $files; do\n dir=$(dirname \"$file\")\n echo \"Extracting $file to directory $dir\"\n unzip -o -d $dir $file\n rm -rf $dir/__MACOSX\n done\n\n - name: Checkout Demo App\n uses: actions/checkout@v4\n with:\n repository: yaoapp/yao-dev-app\n path: app\n\n - name: Checkout Extension\n uses: actions/checkout@v4\n with:\n repository: yaoapp/yao-extensions-dev\n path: extension\n\n - name: Move Dependencies\n run: |\n mv kun ../\n mv xun ../\n mv gou ../\n mv v8go ../\n mv app ../\n mv extension ../\n\n - name: Checkout Code\n uses: actions/checkout@v4\n\n - name: Setup Apple Private Key\n run: |\n mkdir -p ../app/openapi/certs/apple\n echo \"${{ secrets.APPLE_PRIVATE_KEY_USER }}\" > ../app/openapi/certs/apple/signin_client_secret_key.p8\n\n - name: Setup Go ${{ matrix.go }}\n uses: actions/setup-go@v5\n with:\n go-version: ${{ matrix.go }}\n\n - name: Start Redis\n run: docker run --name redis --publish 6379:6379 --detach redis:6\n\n - name: Setup Go Tools\n run: make tools\n\n - name: Install pdftoppm, mutool, imagemagick\n run: |\n sudo apt update\n sudo apt install -y poppler-utils mupdf-tools imagemagick\n\n - name: Test pdftoppm, mutool, imagemagick\n run: |\n pdftoppm -v\n mutool -v\n convert -version\n\n - name: Setup ENV (SQLite)\n run: |\n mkdir -p ${{ github.WORKSPACE }}/../app/db\n echo \"YAO_DB_DRIVER=sqlite3\" >> $GITHUB_ENV\n echo \"YAO_DB_PRIMARY=${{ github.WORKSPACE }}/../app/db/yao.db\" >> $GITHUB_ENV\n\n - name: Pull Sandbox Test Images\n run: |\n docker pull alpine:latest\n docker pull yaoapp/sandbox-base:latest || true\n docker pull yaoapp/sandbox-claude:latest || true\n\n - name: Run Agent Tests (agent, aigc)\n env:\n YAO_SANDBOX_WORKSPACE: ${{ runner.temp }}/sandbox/workspace\n YAO_SANDBOX_IPC: ${{ runner.temp }}/sandbox/ipc\n run: |\n export YAO_SANDBOX_CONTAINER_USER=\"$(id -u):$(id -g)\"\n make unit-test-agent\n\n - name: Codecov Report\n uses: codecov/codecov-action@v4\n with:\n token: ${{ secrets.CODECOV_TOKEN }}\n\n # =============================================================================\n # Robot Tests (all agent/robot/... packages) - Unit + E2E with real LLM calls\n # =============================================================================\n robot-test:\n runs-on: ubuntu-latest\n services:\n mcp-everything:\n image: yaoapp/mcp-everything:latest\n ports:\n - \"3021:3021\"\n - \"3022:3022\"\n\n mongodb:\n image: mongo:6.0\n ports:\n - 27017:27017\n env:\n MONGO_INITDB_ROOT_USERNAME: root\n MONGO_INITDB_ROOT_PASSWORD: 123456\n MONGO_INITDB_DATABASE: test\n\n strategy:\n matrix:\n go: [\"1.25\"]\n steps:\n - name: Checkout Kun\n uses: actions/checkout@v4\n with:\n repository: ${{ env.REPO_KUN }}\n path: kun\n\n - name: Checkout Xun\n uses: actions/checkout@v4\n with:\n repository: ${{ env.REPO_XUN }}\n path: xun\n\n - name: Checkout Gou\n uses: actions/checkout@v4\n with:\n repository: ${{ env.REPO_GOU }}\n path: gou\n\n - name: Checkout V8Go\n uses: actions/checkout@v4\n with:\n repository: yaoapp/v8go\n path: v8go\n\n - name: Unzip libv8\n run: |\n files=$(find ./v8go -name \"libv8*.zip\")\n for file in $files; do\n dir=$(dirname \"$file\")\n echo \"Extracting $file to directory $dir\"\n unzip -o -d $dir $file\n rm -rf $dir/__MACOSX\n done\n\n - name: Checkout Demo App\n uses: actions/checkout@v4\n with:\n repository: yaoapp/yao-dev-app\n path: app\n\n - name: Checkout Extension\n uses: actions/checkout@v4\n with:\n repository: yaoapp/yao-extensions-dev\n path: extension\n\n - name: Move Dependencies\n run: |\n mv kun ../\n mv xun ../\n mv gou ../\n mv v8go ../\n mv app ../\n mv extension ../\n\n - name: Checkout Code\n uses: actions/checkout@v4\n\n - name: Setup Go ${{ matrix.go }}\n uses: actions/setup-go@v5\n with:\n go-version: ${{ matrix.go }}\n\n - name: Start Redis\n run: docker run --name redis --publish 6379:6379 --detach redis:6\n\n - name: Setup Go Tools\n run: make tools\n\n - name: Setup ENV (SQLite)\n run: |\n mkdir -p ${{ github.WORKSPACE }}/../app/db\n echo \"YAO_DB_DRIVER=sqlite3\" >> $GITHUB_ENV\n echo \"YAO_DB_PRIMARY=${{ github.WORKSPACE }}/../app/db/yao.db\" >> $GITHUB_ENV\n\n - name: Run Robot Tests (Unit + E2E)\n run: make unit-test-robot\n\n - name: Codecov Report\n uses: codecov/codecov-action@v4\n with:\n token: ${{ secrets.CODECOV_TOKEN }}\n\n # =============================================================================\n # Sandbox Tests (requires Docker) - Run with Docker-in-Docker\n # =============================================================================\n sandbox-test:\n runs-on: ubuntu-latest\n services:\n mongodb:\n image: mongo:6.0\n ports:\n - 27017:27017\n env:\n MONGO_INITDB_ROOT_USERNAME: root\n MONGO_INITDB_ROOT_PASSWORD: 123456\n MONGO_INITDB_DATABASE: test\n\n strategy:\n matrix:\n go: [\"1.25\"]\n steps:\n - name: Checkout Kun\n uses: actions/checkout@v4\n with:\n repository: ${{ env.REPO_KUN }}\n path: kun\n\n - name: Checkout Xun\n uses: actions/checkout@v4\n with:\n repository: ${{ env.REPO_XUN }}\n path: xun\n\n - name: Checkout Gou\n uses: actions/checkout@v4\n with:\n repository: ${{ env.REPO_GOU }}\n path: gou\n\n - name: Checkout V8Go\n uses: actions/checkout@v4\n with:\n repository: yaoapp/v8go\n path: v8go\n\n - name: Unzip libv8\n run: |\n files=$(find ./v8go -name \"libv8*.zip\")\n for file in $files; do\n dir=$(dirname \"$file\")\n echo \"Extracting $file to directory $dir\"\n unzip -o -d $dir $file\n rm -rf $dir/__MACOSX\n done\n\n - name: Checkout Demo App\n uses: actions/checkout@v4\n with:\n repository: yaoapp/yao-dev-app\n path: app\n\n - name: Checkout Extension\n uses: actions/checkout@v4\n with:\n repository: yaoapp/yao-extensions-dev\n path: extension\n\n - name: Move Dependencies\n run: |\n mv kun ../\n mv xun ../\n mv gou ../\n mv v8go ../\n mv app ../\n mv extension ../\n\n - name: Checkout Code\n uses: actions/checkout@v4\n\n - name: Setup Go ${{ matrix.go }}\n uses: actions/setup-go@v5\n with:\n go-version: ${{ matrix.go }}\n\n - name: Start Redis\n run: docker run --name redis --publish 6379:6379 --detach redis:6\n\n - name: Setup Go Tools\n run: make tools\n\n - name: Setup ENV (SQLite)\n run: |\n mkdir -p ${{ github.WORKSPACE }}/../app/db\n echo \"YAO_DB_DRIVER=sqlite3\" >> $GITHUB_ENV\n echo \"YAO_DB_PRIMARY=${{ github.WORKSPACE }}/../app/db/yao.db\" >> $GITHUB_ENV\n\n - name: Pull Sandbox Test Images\n run: |\n docker pull alpine:latest\n docker pull yaoapp/sandbox-base:latest || true\n docker pull yaoapp/sandbox-claude:latest || true\n\n - name: Run Sandbox Tests\n env:\n YAO_SANDBOX_WORKSPACE: ${{ runner.temp }}/sandbox/workspace\n YAO_SANDBOX_IPC: ${{ runner.temp }}/sandbox/ipc\n run: |\n # Use runner's UID:GID to match host permissions\n export YAO_SANDBOX_CONTAINER_USER=\"$(id -u):$(id -g)\"\n make unit-test-sandbox\n\n - name: Codecov Report\n uses: codecov/codecov-action@v4\n with:\n token: ${{ secrets.CODECOV_TOKEN }}\n\n # =============================================================================\n # Sandbox V2 Tests (tai SDK + workspace, Docker + K8s via k3d)\n # Full sandbox/v2 integration tests are run locally.\n # =============================================================================\n sandbox-v2-test:\n runs-on: ubuntu-latest\n services:\n mongodb:\n image: mongo:6.0\n ports:\n - 27017:27017\n env:\n MONGO_INITDB_ROOT_USERNAME: root\n MONGO_INITDB_ROOT_PASSWORD: 123456\n MONGO_INITDB_DATABASE: test\n\n strategy:\n matrix:\n go: [\"1.25\"]\n steps:\n - name: Checkout Kun\n uses: actions/checkout@v4\n with:\n repository: ${{ env.REPO_KUN }}\n path: kun\n\n - name: Checkout Xun\n uses: actions/checkout@v4\n with:\n repository: ${{ env.REPO_XUN }}\n path: xun\n\n - name: Checkout Gou\n uses: actions/checkout@v4\n with:\n repository: ${{ env.REPO_GOU }}\n path: gou\n\n - name: Checkout V8Go\n uses: actions/checkout@v4\n with:\n repository: yaoapp/v8go\n path: v8go\n\n - name: Unzip libv8\n run: |\n files=$(find ./v8go -name \"libv8*.zip\")\n for file in $files; do\n dir=$(dirname \"$file\")\n echo \"Extracting $file to directory $dir\"\n unzip -o -d $dir $file\n rm -rf $dir/__MACOSX\n done\n\n - name: Checkout Demo App\n uses: actions/checkout@v4\n with:\n repository: yaoapp/yao-dev-app\n path: app\n\n - name: Checkout Extension\n uses: actions/checkout@v4\n with:\n repository: yaoapp/yao-extensions-dev\n path: extension\n\n - name: Move Dependencies\n run: |\n mv kun ../\n mv xun ../\n mv gou ../\n mv v8go ../\n mv app ../\n mv extension ../\n\n - name: Checkout Code\n uses: actions/checkout@v4\n\n - name: Setup Apple Private Key\n run: |\n mkdir -p ../app/openapi/certs/apple\n echo \"${{ secrets.APPLE_PRIVATE_KEY_USER }}\" > ../app/openapi/certs/apple/signin_client_secret_key.p8\n\n - name: Setup Go ${{ matrix.go }}\n uses: actions/setup-go@v5\n with:\n go-version: ${{ matrix.go }}\n\n - name: Start Redis\n run: docker run --name redis --publish 6379:6379 --detach redis:6\n\n - name: Setup Go Tools\n run: make tools\n\n - name: Setup ENV (SQLite)\n run: |\n mkdir -p ${{ github.WORKSPACE }}/../app/db\n echo \"YAO_DB_DRIVER=sqlite3\" >> $GITHUB_ENV\n echo \"YAO_DB_PRIMARY=${{ github.WORKSPACE }}/../app/db/yao.db\" >> $GITHUB_ENV\n\n - name: Pull Test Images\n run: |\n docker pull yaoapp/tai-sandbox-test:latest || true\n docker pull yaoapp/tai:latest\n docker pull alpine:latest\n\n - name: Install k3d\n run: curl -s https://raw.githubusercontent.com/k3d-io/k3d/main/install.sh | bash\n\n - name: Create k3d cluster\n run: |\n k3d cluster create tai-test --no-lb --wait --api-port 16443\n kubectl wait --for=condition=Ready node --all --timeout=60s\n k3d image import alpine:latest -c tai-test\n\n - name: Start Tai Docker instance\n run: |\n docker run -d --name tai-docker \\\n -v /var/run/docker.sock:/var/run/docker.sock \\\n -p 19100:19100 -p 8099:8099 -p 12375:12375 -p 16080:16080 \\\n yaoapp/tai:latest server -direct \\\n -grpc 0.0.0.0:19100 -http 0.0.0.0:8099 -vnc 0.0.0.0:16080 -docker 0.0.0.0:12375\n\n for i in $(seq 1 30); do\n if curl -sf http://127.0.0.1:8099/healthz > /dev/null 2>&1; then\n echo \"Tai Docker HTTP ready\"; break\n fi\n echo \"Waiting for Tai Docker HTTP... ($i)\"; sleep 1\n done\n curl -sf http://127.0.0.1:8099/healthz > /dev/null 2>&1 || {\n echo \"::error::Tai Docker HTTP failed\"; docker logs tai-docker 2>&1; exit 1\n }\n\n for i in $(seq 1 15); do\n if nc -z 127.0.0.1 19100 2>/dev/null; then\n echo \"Tai Docker gRPC ready\"; break\n fi\n echo \"Waiting for Tai Docker gRPC... ($i)\"; sleep 1\n done\n nc -z 127.0.0.1 19100 2>/dev/null || {\n echo \"::error::Tai Docker gRPC failed\"; docker logs tai-docker 2>&1; exit 1\n }\n\n - name: Generate kubeconfig for Tai K8s\n run: |\n K3D_IP=$(docker inspect k3d-tai-test-server-0 | jq -r '.[0].NetworkSettings.Networks[\"k3d-tai-test\"].IPAddress')\n echo \"k3d server IP: ${K3D_IP}\"\n\n k3d kubeconfig get tai-test > /tmp/kubeconfig-k3d.yml\n\n # Kubeconfig for tai-k8s container (uses k3d-internal IP)\n sed \"s|server: .*|server: https://${K3D_IP}:6443|\" /tmp/kubeconfig-k3d.yml \\\n > /tmp/kubeconfig-tai-k8s.yml\n echo \"Container kubeconfig server:\"\n grep server: /tmp/kubeconfig-tai-k8s.yml\n\n # Kubeconfig for test runner (uses localhost via port-mapped 6443)\n sed 's|server: .*|server: https://127.0.0.1:6443|' /tmp/kubeconfig-k3d.yml \\\n > ${{ runner.temp }}/kubeconfig-tai.yml\n echo \"Test runner kubeconfig server:\"\n grep server: ${{ runner.temp }}/kubeconfig-tai.yml\n\n - name: Start Tai K8s instance\n run: |\n K3D_IP=$(docker inspect k3d-tai-test-server-0 | jq -r '.[0].NetworkSettings.Networks[\"k3d-tai-test\"].IPAddress')\n echo \"k3d server IP: ${K3D_IP}\"\n\n docker run -d --name tai-k8s \\\n --network k3d-tai-test \\\n -p 19101:19100 -p 8100:8099 -p 6443:16443 -p 16081:16080 \\\n -v /var/run/docker.sock:/var/run/docker.sock:ro \\\n -v /tmp/kubeconfig-tai-k8s.yml:/etc/tai/kubeconfig.yml:ro \\\n -e TAI_K8S_UPSTREAM=\"tcp://${K3D_IP}:6443\" \\\n -e TAI_KUBECONFIG=/etc/tai/kubeconfig.yml \\\n yaoapp/tai:latest server -direct \\\n -grpc 0.0.0.0:19100 -http 0.0.0.0:8099 -vnc 0.0.0.0:16080 -k8s 0.0.0.0:16443\n\n for i in $(seq 1 30); do\n if curl -sf http://127.0.0.1:8100/healthz > /dev/null 2>&1; then\n echo \"Tai K8s HTTP ready\"; break\n fi\n echo \"Waiting for Tai K8s HTTP... ($i)\"; sleep 1\n done\n curl -sf http://127.0.0.1:8100/healthz > /dev/null 2>&1 || {\n echo \"::error::Tai K8s HTTP failed\"; docker logs tai-k8s 2>&1; exit 1\n }\n\n for i in $(seq 1 15); do\n if nc -z 127.0.0.1 19101 2>/dev/null; then\n echo \"Tai K8s gRPC ready\"; break\n fi\n echo \"Waiting for Tai K8s gRPC... ($i)\"; sleep 1\n done\n nc -z 127.0.0.1 19101 2>/dev/null || {\n echo \"::error::Tai K8s gRPC failed\"; docker logs tai-k8s 2>&1; exit 1\n }\n\n - name: Run Sandbox V2 CI Tests (tai + workspace)\n env:\n TAI_TEST_HOST: \"127.0.0.1\"\n TAI_TEST_DOCKER: \"tcp://127.0.0.1:12375\"\n TAI_TEST_GRPC_PORT: \"19100\"\n TAI_TEST_HTTP_PORT: \"8099\"\n TAI_TEST_VNC_PORT: \"16080\"\n TAI_TEST_DOCKER_PORT: \"12375\"\n TAI_TEST_K8S_HOST: \"127.0.0.1\"\n TAI_TEST_K8S_PORT: \"6443\"\n TAI_TEST_K8S_GRPC_PORT: \"19101\"\n TAI_TEST_KUBECONFIG: \"${{ runner.temp }}/kubeconfig-tai.yml\"\n TAI_TEST_HOST_IP: \"172.17.0.1\"\n SANDBOX_TEST_REMOTE_ADDR: \"tai://127.0.0.1:19100\"\n SANDBOX_TEST_IMAGE: \"yaoapp/tai-sandbox-test:latest\"\n run: make unit-test-sandbox-v2\n\n - name: Codecov Report\n if: always()\n uses: codecov/codecov-action@v4\n with:\n token: ${{ secrets.CODECOV_TOKEN }}\n fail_ci_if_error: false\n\n # =============================================================================\n # Benchmark & Memory Leak Tests - Run with MySQL8.0 and SQLite3\n # =============================================================================\n perf-test:\n runs-on: ubuntu-latest\n services:\n mcp-everything:\n image: yaoapp/mcp-everything:latest\n ports:\n - \"3021:3021\"\n - \"3022:3022\"\n\n mongodb:\n image: mongo:6.0\n ports:\n - 27017:27017\n env:\n MONGO_INITDB_ROOT_USERNAME: root\n MONGO_INITDB_ROOT_PASSWORD: 123456\n MONGO_INITDB_DATABASE: test\n\n strategy:\n matrix:\n go: [\"1.25\"]\n db: [MySQL8.0, SQLite3]\n steps:\n - name: Checkout Kun\n uses: actions/checkout@v4\n with:\n repository: ${{ env.REPO_KUN }}\n path: kun\n\n - name: Checkout Xun\n uses: actions/checkout@v4\n with:\n repository: ${{ env.REPO_XUN }}\n path: xun\n\n - name: Checkout Gou\n uses: actions/checkout@v4\n with:\n repository: ${{ env.REPO_GOU }}\n path: gou\n\n - name: Checkout V8Go\n uses: actions/checkout@v4\n with:\n repository: yaoapp/v8go\n path: v8go\n\n - name: Unzip libv8\n run: |\n files=$(find ./v8go -name \"libv8*.zip\")\n for file in $files; do\n dir=$(dirname \"$file\")\n echo \"Extracting $file to directory $dir\"\n unzip -o -d $dir $file\n rm -rf $dir/__MACOSX\n done\n\n - name: Checkout Demo App\n uses: actions/checkout@v4\n with:\n repository: yaoapp/yao-dev-app\n path: app\n\n - name: Checkout Extension\n uses: actions/checkout@v4\n with:\n repository: yaoapp/yao-extensions-dev\n path: extension\n\n - name: Move Dependencies\n run: |\n mv kun ../\n mv xun ../\n mv gou ../\n mv v8go ../\n mv app ../\n mv extension ../\n\n - name: Checkout Code\n uses: actions/checkout@v4\n\n - name: Setup Go ${{ matrix.go }}\n uses: actions/setup-go@v5\n with:\n go-version: ${{ matrix.go }}\n\n - name: Start Redis\n run: docker run --name redis --publish 6379:6379 --detach redis:6\n\n - name: Install FFmpeg 7.x\n run: |\n wget https://github.com/BtbN/FFmpeg-Builds/releases/download/latest/ffmpeg-master-latest-linux64-gpl.tar.xz\n tar -xf ffmpeg-master-latest-linux64-gpl.tar.xz\n sudo cp ffmpeg-master-latest-linux64-gpl/bin/ffmpeg /usr/local/bin/\n sudo cp ffmpeg-master-latest-linux64-gpl/bin/ffprobe /usr/local/bin/\n sudo chmod +x /usr/local/bin/ffmpeg /usr/local/bin/ffprobe\n\n - name: Test FFmpeg\n run: ffmpeg -version\n\n - name: Setup Go Tools\n run: make tools\n\n - name: Setup ${{ matrix.db }}\n uses: ./.github/actions/setup-db\n with:\n kind: \"${{ matrix.db }}\"\n db: \"xiang\"\n user: \"xiang\"\n password: ${{ secrets.UNIT_PASS }}\n\n - name: Setup ENV\n env:\n PASSWORD: ${{ secrets.UNIT_PASS }}\n run: |\n echo \"YAO_DB_DRIVER=$DB_DRIVER\" >> $GITHUB_ENV\n if [ \"$DB_DRIVER\" = \"mysql\" ]; then\n echo \"YAO_DB_PRIMARY=$DB_USER:$PASSWORD@$DB_HOST\" >> $GITHUB_ENV\n else\n echo \"YAO_DB_PRIMARY=${{ github.WORKSPACE }}/../app/db/yao.db\" >> $GITHUB_ENV\n mkdir -p ${{ github.WORKSPACE }}/../app/db\n fi\n\n - name: Run Benchmark & Memory Leak Tests\n run: |\n make benchmark\n make memory-leak\n\n # =============================================================================\n # Core Tests - Run with DB matrix (MySQL/SQLite/Redis/Mongo combinations)\n # =============================================================================\n core-test:\n runs-on: ubuntu-latest\n services:\n qdrant:\n image: qdrant/qdrant:latest\n ports:\n - 6333:6333 # HTTP API\n - 6334:6334 # gRPC\n\n fastembed:\n image: yaoapp/fastembed:latest-amd64\n env:\n FASTEMBED_PASSWORD: Yao@2026\n ports:\n - 6001:8000\n\n neo4j:\n image: neo4j:latest\n ports:\n - \"7687:7687\"\n env:\n NEO4J_AUTH: neo4j/Yao2026Neo4j\n\n mcp-everything:\n image: yaoapp/mcp-everything:latest\n ports:\n - \"3021:3021\"\n - \"3022:3022\"\n\n strategy:\n matrix:\n go: [\"1.25\"]\n db: [MySQL8.0, SQLite3]\n redis: [4, 5, 6]\n mongo: [\"6.0\"]\n steps:\n - name: Checkout Kun\n uses: actions/checkout@v4\n with:\n repository: ${{ env.REPO_KUN }}\n path: kun\n\n - name: Checkout Xun\n uses: actions/checkout@v4\n with:\n repository: ${{ env.REPO_XUN }}\n path: xun\n\n - name: Checkout Gou\n uses: actions/checkout@v4\n with:\n repository: ${{ env.REPO_GOU }}\n path: gou\n\n - name: Checkout V8Go\n uses: actions/checkout@v4\n with:\n repository: yaoapp/v8go\n path: v8go\n\n - name: Unzip libv8\n run: |\n files=$(find ./v8go -name \"libv8*.zip\")\n for file in $files; do\n dir=$(dirname \"$file\") # Get the directory where the ZIP file is located\n echo \"Extracting $file to directory $dir\"\n unzip -o -d $dir $file\n rm -rf $dir/__MACOSX\n done\n\n - name: Checkout Demo App\n uses: actions/checkout@v4\n with:\n repository: yaoapp/yao-dev-app\n path: app\n\n - name: Checkout Yao Startup Webapp\n uses: actions/checkout@v4\n with:\n repository: yaoapp/yao-startup-webapp\n submodules: true\n token: ${{ secrets.YAO_TEST_TOKEN }}\n path: yao-startup-webapp\n\n - name: Checkout Extension\n uses: actions/checkout@v4\n with:\n repository: yaoapp/yao-extensions-dev\n path: extension\n\n - name: Move Kun, Xun, Gou, V8Go, Extension\n run: |\n mv kun ../\n mv xun ../\n mv gou ../\n mv v8go ../\n mv app ../\n mv extension ../\n mv yao-startup-webapp ../\n ls -l .\n ls -l ../\n\n - name: Checkout Code\n uses: actions/checkout@v4\n\n - name: Setup Apple Private Key\n run: |\n mkdir -p ../app/openapi/certs/apple\n echo \"${{ secrets.APPLE_PRIVATE_KEY_USER }}\" > ../app/openapi/certs/apple/signin_client_secret_key.p8\n\n - name: Setup Go ${{ matrix.go }}\n uses: actions/setup-go@v5\n with:\n go-version: ${{ matrix.go }}\n\n - name: Install FFmpeg 7.x\n run: |\n wget https://github.com/BtbN/FFmpeg-Builds/releases/download/latest/ffmpeg-master-latest-linux64-gpl.tar.xz\n tar -xf ffmpeg-master-latest-linux64-gpl.tar.xz\n sudo cp ffmpeg-master-latest-linux64-gpl/bin/ffmpeg /usr/local/bin/\n sudo cp ffmpeg-master-latest-linux64-gpl/bin/ffprobe /usr/local/bin/\n sudo chmod +x /usr/local/bin/ffmpeg /usr/local/bin/ffprobe\n\n - name: Test FFmpeg\n run: ffmpeg -version\n\n - name: Install pdftoppm, mutool, imagemagick\n run: |\n sudo apt update\n sudo apt install -y poppler-utils mupdf-tools imagemagick\n\n - name: Test pdftoppm, mutool, imagemagick\n run: |\n pdftoppm -v\n mutool -v\n convert -version\n\n - name: Start Redis\n run: docker run --name redis --publish 6379:6379 --detach redis:${{ matrix.redis }}\n\n - name: Start MongoDB\n run: |\n docker run --name mongodb --publish 27017:27017 \\\n -e MONGO_INITDB_DATABASE=test \\\n -e MONGO_INITDB_ROOT_USERNAME=root \\\n -e MONGO_INITDB_ROOT_PASSWORD=123456 \\\n --detach mongo:${{ matrix.mongo }}\n # Wait for MongoDB to be ready\n for i in $(seq 1 20); do\n if docker exec mongodb mongosh --quiet --port 27017 --username root --password 123456 --eval \"db.serverStatus()\" > /dev/null 2>&1; then\n echo \"MongoDB is ready\"\n break\n fi\n echo \"Waiting for MongoDB... ($i)\"\n sleep 1\n done\n\n - name: Setup MySQL8.0 (connector)\n uses: ./.github/actions/setup-db\n with:\n kind: \"MySQL8.0\"\n db: \"test\"\n user: \"test\"\n password: \"123456\"\n port: \"3308\"\n\n - name: Setup ${{ matrix.db }}\n uses: ./.github/actions/setup-db\n with:\n kind: \"${{ matrix.db }}\"\n db: \"xiang\"\n user: \"xiang\"\n password: ${{ secrets.UNIT_PASS }}\n\n - name: Setup Go Tools\n run: |\n make tools\n\n - name: Setup ENV & Host\n env:\n PASSWORD: ${{ secrets.UNIT_PASS }}\n run: |\n sudo echo \"127.0.0.1 local.iqka.com\" | sudo tee -a /etc/hosts \n echo \"YAO_DB_DRIVER=$DB_DRIVER\" >> $GITHUB_ENV\n echo \"GITHUB_WORKSPACE:\\n\" && ls -l $GITHUB_WORKSPACE\n\n if [ \"$DB_DRIVER\" = \"mysql\" ]; then\n echo \"YAO_DB_PRIMARY=$DB_USER:$PASSWORD@$DB_HOST\" >> $GITHUB_ENV\n elif [ \"$DB_DRIVER\" = \"postgres\" ]; then\n echo \"YAO_DB_PRIMARY=postgres://$DB_USER:$PASSWORD@$DB_HOST\" >> $GITHUB_ENV\n else\n echo \"YAO_DB_PRIMARY=$YAO_ROOT/$DB_HOST\" >> $GITHUB_ENV\n fi\n\n echo \".:\\n\" && ls -l .\n echo \"..:\\n\" && ls -l ..\n echo \"../app:\\n\" && ls -l ../app\n ping -c 1 -t 1 local.iqka.com\n\n - name: Test Prepare\n run: |\n make vet\n make fmt-check\n make misspell-check\n\n - name: Inspect\n run: |\n go run . run utils.env.Get MONGO_TEST_HOST\n go run . run utils.env.Get REDIS_TEST_HOST\n go run . inspect\n\n - name: Run Core Tests (exclude AI)\n run: make unit-test-core\n\n - name: Codecov Report\n uses: codecov/codecov-action@v4\n with:\n token: ${{ secrets.CODECOV_TOKEN }} # not required for public repos\n\n # =============================================================================\n # Registry Client SDK Tests (requires Yao Registry Docker service)\n # =============================================================================\n registry-test:\n runs-on: ubuntu-latest\n services:\n yao-registry:\n image: yaoapp/registry:latest\n ports:\n - \"8080:8080\"\n env:\n REGISTRY_INIT_USER: yaoagents\n REGISTRY_INIT_PASS: yaoagents\n strategy:\n matrix:\n go: [\"1.25\"]\n steps:\n - name: Wait for Registry\n run: |\n for i in $(seq 1 15); do\n if curl -sf http://localhost:8080/.well-known/yao-registry > /dev/null 2>&1; then\n echo \"Registry is ready\"\n break\n fi\n echo \"Waiting for registry... ($i)\"\n sleep 1\n done\n\n - name: Checkout Kun\n uses: actions/checkout@v4\n with:\n repository: ${{ env.REPO_KUN }}\n path: kun\n\n - name: Checkout Xun\n uses: actions/checkout@v4\n with:\n repository: ${{ env.REPO_XUN }}\n path: xun\n\n - name: Checkout Gou\n uses: actions/checkout@v4\n with:\n repository: ${{ env.REPO_GOU }}\n path: gou\n\n - name: Checkout V8Go\n uses: actions/checkout@v4\n with:\n repository: yaoapp/v8go\n path: v8go\n\n - name: Unzip libv8\n run: |\n files=$(find ./v8go -name \"libv8*.zip\")\n for file in $files; do\n dir=$(dirname \"$file\")\n echo \"Extracting $file to directory $dir\"\n unzip -o -d $dir $file\n rm -rf $dir/__MACOSX\n done\n\n - name: Checkout Demo App\n uses: actions/checkout@v4\n with:\n repository: yaoapp/yao-dev-app\n path: app\n\n - name: Checkout Extension\n uses: actions/checkout@v4\n with:\n repository: yaoapp/yao-extensions-dev\n path: extension\n\n - name: Move Dependencies\n run: |\n mv kun ../\n mv xun ../\n mv gou ../\n mv v8go ../\n mv app ../\n mv extension ../\n\n - name: Checkout Code\n uses: actions/checkout@v4\n\n - name: Setup Go ${{ matrix.go }}\n uses: actions/setup-go@v5\n with:\n go-version: ${{ matrix.go }}\n\n - name: Run Registry Client Tests\n env:\n YAO_REGISTRY_URL: http://localhost:8080\n YAO_TEST_APPLICATION: ${{ github.workspace }}/../app\n run: make unit-test-registry\n\n - name: Codecov Report\n uses: codecov/codecov-action@v4\n with:\n token: ${{ secrets.CODECOV_TOKEN }}\n\n # =============================================================================\n # gRPC Tests - Run once with SQLite (transport layer, no DB matrix needed)\n # =============================================================================\n grpc-test:\n runs-on: ubuntu-latest\n services:\n mongodb:\n image: mongo:6.0\n ports:\n - 27017:27017\n env:\n MONGO_INITDB_ROOT_USERNAME: root\n MONGO_INITDB_ROOT_PASSWORD: 123456\n MONGO_INITDB_DATABASE: test\n\n strategy:\n matrix:\n go: [\"1.25\"]\n steps:\n - name: Checkout Kun\n uses: actions/checkout@v4\n with:\n repository: ${{ env.REPO_KUN }}\n path: kun\n\n - name: Checkout Xun\n uses: actions/checkout@v4\n with:\n repository: ${{ env.REPO_XUN }}\n path: xun\n\n - name: Checkout Gou\n uses: actions/checkout@v4\n with:\n repository: ${{ env.REPO_GOU }}\n path: gou\n\n - name: Checkout V8Go\n uses: actions/checkout@v4\n with:\n repository: yaoapp/v8go\n path: v8go\n\n - name: Unzip libv8\n run: |\n files=$(find ./v8go -name \"libv8*.zip\")\n for file in $files; do\n dir=$(dirname \"$file\")\n echo \"Extracting $file to directory $dir\"\n unzip -o -d $dir $file\n rm -rf $dir/__MACOSX\n done\n\n - name: Checkout Demo App\n uses: actions/checkout@v4\n with:\n repository: yaoapp/yao-dev-app\n path: app\n\n - name: Checkout Extension\n uses: actions/checkout@v4\n with:\n repository: yaoapp/yao-extensions-dev\n path: extension\n\n - name: Move Dependencies\n run: |\n mv kun ../\n mv xun ../\n mv gou ../\n mv v8go ../\n mv app ../\n mv extension ../\n\n - name: Checkout Code\n uses: actions/checkout@v4\n\n - name: Setup Apple Private Key\n run: |\n mkdir -p ../app/openapi/certs/apple\n echo \"${{ secrets.APPLE_PRIVATE_KEY_USER }}\" > ../app/openapi/certs/apple/signin_client_secret_key.p8\n\n - name: Setup Go ${{ matrix.go }}\n uses: actions/setup-go@v5\n with:\n go-version: ${{ matrix.go }}\n\n - name: Start Redis\n run: docker run --name redis --publish 6379:6379 --detach redis:6\n\n - name: Setup Go Tools\n run: make tools\n\n - name: Setup ENV (SQLite)\n run: |\n mkdir -p ${{ github.WORKSPACE }}/../app/db\n echo \"YAO_DB_DRIVER=sqlite3\" >> $GITHUB_ENV\n echo \"YAO_DB_PRIMARY=${{ github.WORKSPACE }}/../app/db/yao.db\" >> $GITHUB_ENV\n\n - name: Run gRPC Tests\n run: make unit-test-grpc\n\n - name: Codecov Report\n uses: codecov/codecov-action@v4\n with:\n token: ${{ secrets.CODECOV_TOKEN }}\n" }, { "path": ".gitignore", "content": "# Binaries for programs and plugins\n*.exe\n*.exe~\n*.dll\n*.so\n*.dylib\n\n# Test binary, built with `go test -c`\n*.test\n\n# Output of the go coverage tool, specifically when used with LiteIDE\n*.out\n\n*.log\n\n# Dependency directories (remove the comment below to include it)\n# vendor/\n.DS_Store\n.env*\n.tmp\ndist\nui\n^ui/index.html\ntests/data/*\n!tests/data/assets\ndebug.log\nyao-test\ndev.sh\nlogs\nlogs/application.log\nyao-arm64\ntests/db/yao.db\nenv.*.sh\nxgen/v0.9/*\nxgen/v1.0/*\n!xgen/v0.9/index.html\n!xgen/v1.0/index.html\n!xgen/v1.0/umi.js\n!xgen/v1.0/layouts__index.async.js\n!pipe/ui\n*-unit-test\ndocker/build/test\ndb\n!agent/search/handlers/db\n*.sh\ndata/bindata.go.bak\nshare/const.go.bak\nshare/const.goe\n.cursor\nopenapi/*.md\ncoverage.html\nagent/assistant/hook/*.test.md\nagent/search/TODO.md\nagent/search/job-logs.txt\nagent/test/MULTI_TURN_DESIGN.md\nagent/test/UPGRADE_PLAN.md\nintroduction/*\n!sandbox/docker/build.sh\n!sandbox/docker/vnc/*.sh\n!sandbox/docker/desktop/config/*.sh\nsandbox/docker/yao-bridge-*\nsandbox/docker/claude-proxy-*\nsandbox/docker/claude/claude-proxy-*\nsandbox/proxy/claude-proxy-linux-*\nrelease/*\nsandbox/TODO-VNC.md\nsandbox/docker/chrome/PLAN.md\nsandbox/DESIGN-REMOTE.md\nevent/DESIGN.md\nevent/TODO.md\nagent/robot/DESIGN-V2.md\ntg-session.json\ntg-login\ntg-send\nregistry/data/\nregistry/manager/DESIGN*.md\ntai/testdata/\nagent/sandbox/docs/*.md\ntai/docs/refactor-registration.md\n" }, { "path": "COMMERCIAL_LICENSE.md", "content": "> **DEPRECATED**: This license is no longer in effect. Please refer to the [LICENSE](LICENSE) file for current licensing terms.\n\n# Commercial License for Yao\n\nThis document outlines the terms for the commercial license of the **Yao** project. While the Yao project is primarily licensed under the **Apache License, Version 2.0**, certain commercial use cases require a separate commercial license.\n\n## 1. Commercial License Requirements\n\nThe following use cases require a commercial license:\n\n1. **Application Hosting Services** \n If you use Yao, or any derivative product (such as a forked or modified version of Yao), to provide Yao-based application hosting services (e.g., Software-as-a-Service (SaaS) or Platform-as-a-Service (PaaS)) to users, you must obtain a commercial license. This restriction applies regardless of whether the original Yao code or a modified version is used to host and manage applications on behalf of third-party users for commercial purposes. \n **In addition**, if you provide hosting services for applications that are built using Yao (even if they are customized or modified versions of Yao), a commercial license is required.\n\n ### Definition: Application Hosting Services\n\n \"Application Hosting Services\" refers to any service that involves hosting Yao-based applications or web applications created with Yao (including modified versions of Yao) for third-party users. This includes, but is not limited to:\n\n - **Hosting platforms** providing software or services built on top of Yao for third-party users.\n - **SaaS or PaaS offerings** where you manage and host applications that are based on or utilize Yao, either in their original or modified form.\n - **Managed hosting services** where Yao is used as the underlying technology for applications deployed for external clients.\n\n In these cases, a commercial license is required, whether you are using the original Yao code or a fork/modified version.\n\n2. **AI Web Application Generation Services** \n If you provide services that generate AI-driven web applications using Yao, or any derivative product (such as a fork or modified version of Yao), to third-party users, you are required to purchase a commercial license.\n\n ### Definition: AI Web Application Generation Services\n\n \"AI Web Application Generation Services\" refers to any service or functionality that utilizes Yao (or any forked or modified version of Yao) to automate the creation of web applications with AI capabilities. This includes, but is not limited to, providing third-party users with:\n\n - **Automated web application development** driven by AI, where the service generates complete or partial web applications.\n - **Customizable web solutions** that are powered by AI and built using Yao as the core technology.\n - **On-demand application generation** for specific client needs, using Yao to dynamically build, configure, or deploy applications for users.\n\n In these cases, whether Yao is directly used, forked, or modified, a commercial license is required to operate legally.\n\n## 2. Use Under Apache License 2.0\n\nFor all other uses, the **Apache License, Version 2.0** applies. You are free to use, modify, and distribute the Yao project under the terms of Apache 2.0 as long as your usage does not fall within the restricted scenarios outlined above.\n\n## 3. Obtaining a Commercial License\n\nTo inquire about or obtain a commercial license, please contact us at:\n\n- **Email**: [friends@iqka.com]\n- **Website**: [https://moapi.ai/contact]\n\nPricing and terms for commercial licenses vary based on usage scenarios, user scale, and other factors.\n\n## 4. Compliance and Auditing\n\nIf you have any questions about whether your use case requires a commercial license, please contact us for clarification. We reserve the right to audit usage for compliance and enforce commercial licensing terms where necessary.\n\n## 5. Disclaimer\n\nFailure to comply with these licensing terms may result in a violation of the Yao licensing agreement and could lead to legal action.\n\n---\n\n**Note:** This commercial license is supplementary to the Apache 2.0 license and only applies in specific commercial scenarios outlined above.\n" }, { "path": "COMMERCIAL_LICENSE.zh-CN.md", "content": "> **已废弃**: 本许可证已不再生效。请参考 [LICENSE](LICENSE) 文件获取当前的许可条款。\n\n# Yao 商业许可证\n\n本文件概述了 **Yao** 项目的商业许可证条款。虽然 Yao 项目主要使用 **Apache 许可证 2.0 版** 授权，但某些商业使用场景需要单独的商业许可证。\n\n## 1. 商业许可证要求\n\n以下使用场景需要商业许可证：\n\n1. **应用托管服务** \n 如果您使用 Yao 或其衍生产品（如 Yao 的分支版本或修改版本）为用户提供基于 Yao 的应用托管服务（例如，软件即服务（SaaS）或平台即服务（PaaS）），您必须获得商业许可证。此限制适用于无论是否使用原始 Yao 代码或修改版 Yao 代码，托管和管理应用程序的行为只要是为第三方用户提供的商业目的。 \n **此外**，如果您提供的托管服务是为使用 Yao 构建的应用程序提供托管服务（即使它们是定制或修改版的 Yao），也需要获得商业许可证。\n\n ### 定义：应用托管服务\n\n \"应用托管服务\"指任何涉及托管基于 Yao 的应用程序或使用 Yao 创建的 WEB 应用程序（包括 Yao 的修改版本）的服务，服务对象为第三方用户。包括但不限于：\n\n - **托管平台** 提供基于 Yao 的软件或服务给第三方用户。\n - **SaaS 或 PaaS 服务**，在这些服务中，您管理并托管基于或利用 Yao 的应用程序，可能是原版或修改版。\n - **托管服务**，其中 Yao 被用作为客户外部部署应用程序的基础技术。\n\n 在这些情况下，无论是使用原始 Yao 代码还是修改版 Yao，都需要获得商业许可证。\n\n2. **AI WEB 应用生成服务** \n 如果您提供利用 Yao 或其衍生产品（如 Yao 的分支版本或修改版本）为第三方用户生成 AI 驱动的 WEB 应用程序的服务，您需要购买商业许可证。\n\n ### 定义：AI WEB 应用生成服务\n\n \"AI WEB 应用生成服务\"指任何利用 Yao（或任何分支版本或修改版本的 Yao）自动化创建具有 AI 功能的 WEB 应用程序的服务或功能。包括但不限于，为第三方用户提供以下服务：\n\n - **AI 驱动的自动化 WEB 应用开发**，该服务生成完整或部分 WEB 应用程序。\n - **可定制的 WEB 解决方案**，这些解决方案由 AI 提供支持，并以 Yao 作为核心技术构建。\n - **按需应用生成**，根据特定客户需求，使用 Yao 动态构建、配置或部署应用程序。\n\n 在这些情况下，无论是直接使用 Yao，还是使用其分支或修改版，均需要获得商业许可证。\n\n## 2. 使用 Apache 许可证 2.0\n\n对于所有其他用途，**Apache 许可证 2.0 版** 适用。只要您的使用不属于上述限制的商业场景，您可以自由地根据 Apache 2.0 许可证使用、修改和分发 Yao 项目。\n\n## 3. 获取商业许可证\n\n如需咨询或获取商业许可证，请通过以下方式联系我们：\n\n- **电子邮件**：[friends@iqka.com]\n- **网站**：[https://moapi.ai/contact](https://moapi.ai/contact)\n\n商业许可证的定价和条款会根据使用场景、用户规模及其他因素有所不同。\n\n## 4. 合规与审计\n\n如果您对您的使用场景是否需要商业许可证有任何疑问，请联系我们以获取澄清。我们保留审核使用情况以确保合规，并在必要时执行商业许可条款的权利。\n\n## 5. 免责声明\n\n未遵守这些许可条款可能会导致违反 Yao 许可证协议，并可能导致法律诉讼。\n\n---\n\n**注意：** 此商业许可证是 Apache 2.0 许可证的补充，仅适用于上述特定的商业场景。\n" }, { "path": "LICENSE", "content": "# Open Source License\n\nYao App Engine is licensed under a modified version of the Apache License 2.0, with the following additional conditions:\n\n1. Commercial Usage Terms:\n Yao App Engine may be utilized commercially, A commercial license from the producer is required if:\n\n a. Trademark and Branding Requirements\n\n - The Yao App Engine console/application logo and copyright information must not be removed or modified\n - Logo and copyright information can only be changed with an authorization certificate issued through Yao Developer Certificate\n\n b. Authorization Verification Requirements\n\n - The Yao certificate verification logic, processes, and related pages (marked in code comments) must be preserved\n - The complete Yao certificate verification system must be maintained regardless of usage purpose\n\n2. Contributor Agreement:\n - The producer reserves the right to modify the open-source agreement terms\n - Contributed code may be used for commercial purposes, including cloud business operations\n\nAll other rights and restrictions follow the Apache License 2.0 (http://www.apache.org/licenses/LICENSE-2.0).\n\n© 2025 Infinite Wisdom Software.\n" }, { "path": "Makefile", "content": "GO ?= go\nGIT ?= git\nGOFMT ?= gofmt \"-s\"\nPACKAGES ?= $(shell $(GO) list ./...)\nVETPACKAGES ?= $(shell $(GO) list ./... | grep -v /examples/)\nGOFILES := $(shell find . -name \"*.go\")\nVERSION := $(shell grep 'const VERSION =' share/const.go |awk '{print $$4}' |sed 's/\\\"//g')\nCOMMIT := $(shell git log | head -n 1 | awk '{print substr($$2, 0, 12)}')\nNOW := $(shell date +\"%FT%T%z\")\nOS := $(shell uname)\n\n# ROOT_DIR := $(shell dirname $(realpath $(firstword $(MAKEFILE_LIST))))\nTESTFOLDER := $(shell $(GO) list ./... | grep -vE 'examples|openai|aigc|neo|twilio|share*|registry|agent/sandbox/v2' | awk '!/\\/tests\\// || /openapi\\/tests/' | grep -vE 'openapi/tests/(nodes|sandbox|workspace)')\n# Core tests (exclude AI-related: agent, aigc, openai, KB, sandbox, registry, grpc, and integrations which require external services)\nTESTFOLDER_CORE := $(shell $(GO) list ./... | grep -vE 'examples|openai|aigc|neo|twilio|share*|agent|kb|sandbox|integrations|registry|tai|grpc' | awk '!/\\/tests\\// || /openapi\\/tests/' | grep -vE 'openapi/tests/(nodes|sandbox|workspace)')\n# Agent tests (agent, aigc) - exclude agent/search/handlers/web (requires external API keys), robot packages (tested in robot job), and agent/sandbox/v2 (WIP, has its own job)\nTESTFOLDER_AGENT := $(shell $(GO) list ./agent/... ./aigc/... | grep -vE 'agent/search/handlers/web|agent/robot/|agent/sandbox/v2')\n# KB tests (kb)\nTESTFOLDER_KB := $(shell $(GO) list ./kb/...)\n# Robot tests (agent/robot/... packages, excluding events/integrations which require Telegram etc.)\nTESTFOLDER_ROBOT := $(shell $(GO) list ./agent/robot/... | grep -vE 'agent/robot/events')\n# Sandbox tests (requires Docker) — excludes sandbox/v2 (has its own job)\nTESTFOLDER_SANDBOX := $(shell $(GO) list ./sandbox/... | grep -v 'sandbox/v2')\n# Tai SDK tests (requires Tai container with Docker socket)\nTESTFOLDER_TAI := $(shell $(GO) list ./tai/...)\n# Workspace tests (requires Tai for remote mode)\nTESTFOLDER_WORKSPACE := $(shell $(GO) list ./workspace/...)\n# gRPC tests\nTESTFOLDER_GRPC := $(shell $(GO) list ./grpc/...)\nTESTTAGS ?= \"\"\n\n# TESTWIDGETS := $(shell $(GO) list ./widgets/...)\n\n# Unit Test (all tests)\n.PHONY: unit-test\nunit-test:\n\techo \"mode: count\" > coverage.out\n\tfor d in $(TESTFOLDER); do \\\n\t\t$(GO) test -tags $(TESTTAGS) -v -covermode=count -coverprofile=profile.out -coverpkg=$$(echo $$d | sed \"s/\\/test$$//g\") -skip='TestMemoryLeak|TestIsolateDisposal|TestLeak_|TestScenario_' $$d > tmp.out; \\\n\t\tcat tmp.out; \\\n\t\tif grep -q \"^--- FAIL\" tmp.out; then \\\n\t\t\trm tmp.out; \\\n\t\t\texit 1; \\\n\t\telif grep -q \"build failed\" tmp.out; then \\\n\t\t\trm tmp.out; \\\n\t\t\texit 1; \\\n\t\telif grep -q \"setup failed\" tmp.out; then \\\n\t\t\trm tmp.out; \\\n\t\t\texit 1; \\\n\t\telif grep -q \"runtime error\" tmp.out; then \\\n\t\t\trm tmp.out; \\\n\t\t\texit 1; \\\n\t\tfi; \\\n\t\tif [ -f profile.out ]; then \\\n\t\t\tcat profile.out | grep -v \"mode:\" >> coverage.out; \\\n\t\t\trm profile.out; \\\n\t\tfi; \\\n\tdone\n\n# Core Unit Test (exclude AI-related tests)\n.PHONY: unit-test-core\nunit-test-core:\n\techo \"mode: count\" > coverage.out\n\tfor d in $(TESTFOLDER_CORE); do \\\n\t\t$(GO) test -tags $(TESTTAGS) -v -covermode=count -coverprofile=profile.out -coverpkg=$$(echo $$d | sed \"s/\\/test$$//g\") -skip='TestMemoryLeak|TestIsolateDisposal|TestLeak_|TestScenario_' $$d > tmp.out; \\\n\t\tcat tmp.out; \\\n\t\tif grep -q \"^--- FAIL\" tmp.out; then \\\n\t\t\trm tmp.out; \\\n\t\t\texit 1; \\\n\t\telif grep -q \"build failed\" tmp.out; then \\\n\t\t\trm tmp.out; \\\n\t\t\texit 1; \\\n\t\telif grep -q \"setup failed\" tmp.out; then \\\n\t\t\trm tmp.out; \\\n\t\t\texit 1; \\\n\t\telif grep -q \"runtime error\" tmp.out; then \\\n\t\t\trm tmp.out; \\\n\t\t\texit 1; \\\n\t\tfi; \\\n\t\tif [ -f profile.out ]; then \\\n\t\t\tcat profile.out | grep -v \"mode:\" >> coverage.out; \\\n\t\t\trm profile.out; \\\n\t\tfi; \\\n\tdone\n\n# Agent Unit Test (agent, aigc) - excludes robot packages (tested in unit-test-robot) and TestE2E*\n.PHONY: unit-test-agent\nunit-test-agent:\n\techo \"mode: count\" > coverage.out\n\tfor d in $(TESTFOLDER_AGENT); do \\\n\t\t$(GO) test -tags $(TESTTAGS) -v -timeout=50m -covermode=count -coverprofile=profile.out -coverpkg=$$(echo $$d | sed \"s/\\/test$$//g\") -skip='TestMemoryLeak|TestIsolateDisposal|TestE2E' $$d > tmp.out; \\\n\t\tcat tmp.out; \\\n\t\tif grep -q \"^--- FAIL\" tmp.out; then \\\n\t\t\trm tmp.out; \\\n\t\t\texit 1; \\\n\t\telif grep -q \"^FAIL\" tmp.out; then \\\n\t\t\trm tmp.out; \\\n\t\t\texit 1; \\\n\t\telif grep -q \"^panic:\" tmp.out; then \\\n\t\t\trm tmp.out; \\\n\t\t\texit 1; \\\n\t\telif grep -q \"build failed\" tmp.out; then \\\n\t\t\trm tmp.out; \\\n\t\t\texit 1; \\\n\t\telif grep -q \"setup failed\" tmp.out; then \\\n\t\t\trm tmp.out; \\\n\t\t\texit 1; \\\n\t\telif grep -q \"runtime error\" tmp.out; then \\\n\t\t\trm tmp.out; \\\n\t\t\texit 1; \\\n\t\tfi; \\\n\t\tif [ -f profile.out ]; then \\\n\t\t\tcat profile.out | grep -v \"mode:\" >> coverage.out; \\\n\t\t\trm profile.out; \\\n\t\tfi; \\\n\tdone\n\n# KB Unit Test (kb)\n.PHONY: unit-test-kb\nunit-test-kb:\n\techo \"mode: count\" > coverage.out\n\tfor d in $(TESTFOLDER_KB); do \\\n\t\t$(GO) test -tags $(TESTTAGS) -v -timeout=20m -covermode=count -coverprofile=profile.out -coverpkg=$$(echo $$d | sed \"s/\\/test$$//g\") -skip='TestMemoryLeak|TestIsolateDisposal|TestSearchCleanup' $$d > tmp.out; \\\n\t\tcat tmp.out; \\\n\t\tif grep -q \"^--- FAIL\" tmp.out; then \\\n\t\t\trm tmp.out; \\\n\t\t\texit 1; \\\n\t\telif grep -q \"^FAIL\" tmp.out; then \\\n\t\t\trm tmp.out; \\\n\t\t\texit 1; \\\n\t\telif grep -q \"^panic:\" tmp.out; then \\\n\t\t\trm tmp.out; \\\n\t\t\texit 1; \\\n\t\telif grep -q \"build failed\" tmp.out; then \\\n\t\t\trm tmp.out; \\\n\t\t\texit 1; \\\n\t\telif grep -q \"setup failed\" tmp.out; then \\\n\t\t\trm tmp.out; \\\n\t\t\texit 1; \\\n\t\telif grep -q \"runtime error\" tmp.out; then \\\n\t\t\trm tmp.out; \\\n\t\t\texit 1; \\\n\t\tfi; \\\n\t\tif [ -f profile.out ]; then \\\n\t\t\tcat profile.out | grep -v \"mode:\" >> coverage.out; \\\n\t\t\trm profile.out; \\\n\t\tfi; \\\n\tdone\n\n# Robot Test (all agent/robot/... packages) - runs ALL tests (unit + E2E) with real LLM calls\n# These tests require: LLM API keys, database, and longer timeout\n.PHONY: unit-test-robot\nunit-test-robot:\n\techo \"mode: count\" > coverage.out\n\tfor d in $(TESTFOLDER_ROBOT); do \\\n\t\t$(GO) test -tags $(TESTTAGS) -v -timeout=50m -covermode=count -coverprofile=profile.out -coverpkg=$$(echo $$d | sed \"s/\\/test$$//g\") -skip='TestMemoryLeak|TestIsolateDisposal' $$d > tmp.out; \\\n\t\tcat tmp.out; \\\n\t\tif grep -q \"^--- FAIL\" tmp.out; then \\\n\t\t\trm tmp.out; \\\n\t\t\texit 1; \\\n\t\telif grep -q \"^FAIL\" tmp.out; then \\\n\t\t\trm tmp.out; \\\n\t\t\texit 1; \\\n\t\telif grep -q \"^panic:\" tmp.out; then \\\n\t\t\trm tmp.out; \\\n\t\t\texit 1; \\\n\t\telif grep -q \"build failed\" tmp.out; then \\\n\t\t\trm tmp.out; \\\n\t\t\texit 1; \\\n\t\telif grep -q \"setup failed\" tmp.out; then \\\n\t\t\trm tmp.out; \\\n\t\t\texit 1; \\\n\t\telif grep -q \"runtime error\" tmp.out; then \\\n\t\t\trm tmp.out; \\\n\t\t\texit 1; \\\n\t\tfi; \\\n\t\tif [ -f profile.out ]; then \\\n\t\t\tcat profile.out | grep -v \"mode:\" >> coverage.out; \\\n\t\t\trm profile.out; \\\n\t\tfi; \\\n\tdone\n\n# Registry Client Test (requires Yao Registry service)\n.PHONY: unit-test-registry\nunit-test-registry:\n\techo \"mode: count\" > coverage.out\n\t$(GO) test -v -p 1 -timeout=5m -covermode=count -coverprofile=profile.out ./registry/... > tmp.out; \\\n\tcat tmp.out; \\\n\tif grep -q \"^--- FAIL\" tmp.out; then \\\n\t\trm tmp.out; \\\n\t\texit 1; \\\n\telif grep -q \"^FAIL\" tmp.out; then \\\n\t\trm tmp.out; \\\n\t\texit 1; \\\n\telif grep -q \"^panic:\" tmp.out; then \\\n\t\trm tmp.out; \\\n\t\texit 1; \\\n\tfi; \\\n\tif [ -f profile.out ]; then \\\n\t\tcat profile.out | grep -v \"mode:\" >> coverage.out; \\\n\t\trm profile.out; \\\n\tfi\n\n# ---------------------------------------------------------------------------\n# Sandbox V2 CI Test (tai SDK + workspace only)\n# Full sandbox/v2 integration tests (multi-pool, K8s, etc.) are run locally.\n# ---------------------------------------------------------------------------\n\n.PHONY: unit-test-sandbox-v2\nunit-test-sandbox-v2: unit-test-tai unit-test-workspace\n\t@echo \"\"\n\t@echo \"=============================================\"\n\t@echo \"All Sandbox V2 CI tests passed (tai + workspace)\"\n\t@echo \"=============================================\"\n\n# Workspace Unit Test (requires Tai for remote mode)\n.PHONY: unit-test-workspace\nunit-test-workspace:\n\t@echo \"\"\n\t@echo \"=============================================\"\n\t@echo \"Running Workspace Tests...\"\n\t@echo \"=============================================\"\n\techo \"mode: count\" > coverage.out\n\tfor d in $(TESTFOLDER_WORKSPACE); do \\\n\t\t$(GO) test -tags $(TESTTAGS) -v -timeout=10m -covermode=count -coverprofile=profile.out -coverpkg=$$(echo $$d | sed \"s/\\/test$$//g\") $$d > tmp.out; \\\n\t\tcat tmp.out; \\\n\t\tif grep -q \"^--- FAIL\" tmp.out; then \\\n\t\t\trm tmp.out; \\\n\t\t\texit 1; \\\n\t\telif grep -q \"^FAIL\" tmp.out; then \\\n\t\t\trm tmp.out; \\\n\t\t\texit 1; \\\n\t\telif grep -q \"^panic:\" tmp.out; then \\\n\t\t\trm tmp.out; \\\n\t\t\texit 1; \\\n\t\telif grep -q \"build failed\" tmp.out; then \\\n\t\t\trm tmp.out; \\\n\t\t\texit 1; \\\n\t\telif grep -q \"setup failed\" tmp.out; then \\\n\t\t\trm tmp.out; \\\n\t\t\texit 1; \\\n\t\telif grep -q \"runtime error\" tmp.out; then \\\n\t\t\trm tmp.out; \\\n\t\t\texit 1; \\\n\t\tfi; \\\n\t\tif [ -f profile.out ]; then \\\n\t\t\tcat profile.out | grep -v \"mode:\" >> coverage.out; \\\n\t\t\trm profile.out; \\\n\t\tfi; \\\n\tdone\n\t@echo \"\"\n\t@echo \"=============================================\"\n\t@echo \"All workspace tests passed\"\n\t@echo \"=============================================\"\n\n# Sandbox Unit Test (requires Docker)\n.PHONY: unit-test-sandbox\nunit-test-sandbox:\n\t@echo \"\"\n\t@echo \"=============================================\"\n\t@echo \"Running Sandbox Tests (requires Docker)...\"\n\t@echo \"=============================================\"\n\t@echo \"Pulling sandbox test images...\"\n\tdocker pull alpine:latest || true\n\tdocker pull yaoapp/sandbox-base:latest || true\n\tdocker pull yaoapp/sandbox-claude:latest || true\n\t@echo \"\"\n\techo \"mode: count\" > coverage.out\n\tfor d in $(TESTFOLDER_SANDBOX); do \\\n\t\t$(GO) test -tags $(TESTTAGS) -v -timeout=10m -covermode=count -coverprofile=profile.out -coverpkg=$$(echo $$d | sed \"s/\\/test$$//g\") -skip='TestMemoryLeak|TestIsolateDisposal' $$d > tmp.out; \\\n\t\tcat tmp.out; \\\n\t\tif grep -q \"^--- FAIL\" tmp.out; then \\\n\t\t\trm tmp.out; \\\n\t\t\texit 1; \\\n\t\telif grep -q \"^FAIL\" tmp.out; then \\\n\t\t\trm tmp.out; \\\n\t\t\texit 1; \\\n\t\telif grep -q \"^panic:\" tmp.out; then \\\n\t\t\trm tmp.out; \\\n\t\t\texit 1; \\\n\t\telif grep -q \"build failed\" tmp.out; then \\\n\t\t\trm tmp.out; \\\n\t\t\texit 1; \\\n\t\telif grep -q \"setup failed\" tmp.out; then \\\n\t\t\trm tmp.out; \\\n\t\t\texit 1; \\\n\t\telif grep -q \"runtime error\" tmp.out; then \\\n\t\t\trm tmp.out; \\\n\t\t\texit 1; \\\n\t\tfi; \\\n\t\tif [ -f profile.out ]; then \\\n\t\t\tcat profile.out | grep -v \"mode:\" >> coverage.out; \\\n\t\t\trm profile.out; \\\n\t\tfi; \\\n\tdone\n\t@echo \"\"\n\t@echo \"=============================================\"\n\t@echo \"✅ All sandbox tests passed\"\n\t@echo \"=============================================\"\n\n# Tai SDK Test (requires Tai container with Docker socket)\n.PHONY: unit-test-tai\nunit-test-tai:\n\t@echo \"\"\n\t@echo \"=============================================\"\n\t@echo \"Running Tai SDK Tests (requires Tai container)...\"\n\t@echo \"=============================================\"\n\techo \"mode: count\" > coverage.out\n\tfor d in $(TESTFOLDER_TAI); do \\\n\t\t$(GO) test -tags $(TESTTAGS) -v -timeout=5m -covermode=count -coverprofile=profile.out -coverpkg=$$(echo $$d | sed \"s/\\/test$$//g\") $$d > tmp.out; \\\n\t\tcat tmp.out; \\\n\t\tif grep -q \"^--- FAIL\" tmp.out; then \\\n\t\t\trm tmp.out; \\\n\t\t\texit 1; \\\n\t\telif grep -q \"^FAIL\" tmp.out; then \\\n\t\t\trm tmp.out; \\\n\t\t\texit 1; \\\n\t\telif grep -q \"^panic:\" tmp.out; then \\\n\t\t\trm tmp.out; \\\n\t\t\texit 1; \\\n\t\telif grep -q \"build failed\" tmp.out; then \\\n\t\t\trm tmp.out; \\\n\t\t\texit 1; \\\n\t\telif grep -q \"setup failed\" tmp.out; then \\\n\t\t\trm tmp.out; \\\n\t\t\texit 1; \\\n\t\telif grep -q \"runtime error\" tmp.out; then \\\n\t\t\trm tmp.out; \\\n\t\t\texit 1; \\\n\t\tfi; \\\n\t\tif [ -f profile.out ]; then \\\n\t\t\tcat profile.out | grep -v \"mode:\" >> coverage.out; \\\n\t\t\trm profile.out; \\\n\t\tfi; \\\n\tdone\n\t@echo \"\"\n\t@echo \"=============================================\"\n\t@echo \"All Tai SDK tests passed\"\n\t@echo \"=============================================\"\n\n# Proto codegen\n.PHONY: proto\nproto:\n\tprotoc --go_out=. --go_opt=paths=source_relative \\\n\t\t--go-grpc_out=. --go-grpc_opt=paths=source_relative \\\n\t\tgrpc/pb/yao.proto\n\n# gRPC Unit Test\n.PHONY: unit-test-grpc\nunit-test-grpc:\n\techo \"mode: count\" > coverage.out\n\tfor d in $(TESTFOLDER_GRPC); do \\\n\t\t$(GO) test -tags $(TESTTAGS) -v -timeout=10m \\\n\t\t\t-covermode=count -coverprofile=profile.out \\\n\t\t\t-coverpkg=$$(echo $$d | sed \"s/\\/test$$//g\") \\\n\t\t\t-skip='TestMemoryLeak|TestIsolateDisposal|TestLeak_|TestScenario_' \\\n\t\t\t$$d > tmp.out; \\\n\t\tcat tmp.out; \\\n\t\tif grep -q \"^--- FAIL\" tmp.out; then \\\n\t\t\trm tmp.out; \\\n\t\t\texit 1; \\\n\t\telif grep -q \"build failed\" tmp.out; then \\\n\t\t\trm tmp.out; \\\n\t\t\texit 1; \\\n\t\telif grep -q \"setup failed\" tmp.out; then \\\n\t\t\trm tmp.out; \\\n\t\t\texit 1; \\\n\t\telif grep -q \"runtime error\" tmp.out; then \\\n\t\t\trm tmp.out; \\\n\t\t\texit 1; \\\n\t\tfi; \\\n\t\tif [ -f profile.out ]; then \\\n\t\t\tcat profile.out | grep -v \"mode:\" >> coverage.out; \\\n\t\t\trm profile.out; \\\n\t\tfi; \\\n\tdone\n\n# Benchmark Test\n.PHONY: benchmark\nbenchmark:\n\t@echo \"\"\n\t@echo \"=============================================\"\n\t@echo \"Running Benchmark Tests (agent, trace, event)...\"\n\t@echo \"=============================================\"\n\t@for d in $$($(GO) list ./agent/... ./trace/... ./event/...); do \\\n\t\tif $(GO) test -list=Benchmark $$d 2>/dev/null | grep -q \"^Benchmark\"; then \\\n\t\t\techo \"\"; \\\n\t\t\techo \"📊 Benchmarking: $$d\"; \\\n\t\t\techo \"---------------------------------------------\"; \\\n\t\t\t$(GO) test -bench=. -benchmem -benchtime=100x -run='^$$' $$d || true; \\\n\t\tfi; \\\n\tdone\n\t@echo \"\"\n\t@echo \"=============================================\"\n\t@echo \"✅ All benchmarks completed\"\n\t@echo \"=============================================\"\n\n# Memory Leak Detection Test\n.PHONY: memory-leak\nmemory-leak:\n\t@echo \"\"\n\t@echo \"=============================================\"\n\t@echo \"Running Memory Leak Detection (agent, trace, event)...\"\n\t@echo \"=============================================\"\n\t@for d in $$($(GO) list ./agent/... ./trace/... ./event/...); do \\\n\t\tif $(GO) test -list='TestMemoryLeak|TestIsolateDisposal|TestGoroutineLeak|TestLeak_|TestScenario_' $$d 2>/dev/null | grep -qE \"^Test(MemoryLeak|IsolateDisposal|GoroutineLeak|Leak_|Scenario_)\"; then \\\n\t\t\techo \"\"; \\\n\t\t\techo \"🔍 Memory Leak Detection: $$d\"; \\\n\t\t\techo \"---------------------------------------------\"; \\\n\t\t\t$(GO) test -run='TestMemoryLeak|TestIsolateDisposal|TestGoroutineLeak|TestLeak_|TestScenario_' -v -timeout=5m $$d || exit 1; \\\n\t\tfi; \\\n\tdone\n\t@echo \"\"\n\t@echo \"=============================================\"\n\t@echo \"✅ All memory leak tests passed\"\n\t@echo \"=============================================\"\n\n# Run all tests (unit + benchmark + memory leak)\n.PHONY: test\ntest: unit-test benchmark memory-leak\n\n.PHONY: fmt\nfmt:\n\t$(GOFMT) -w $(GOFILES)\n\n.PHONY: fmt-check\nfmt-check:\n\t@diff=$$($(GOFMT) -d $(GOFILES)); \\\n\tif [ -n \"$$diff\" ]; then \\\n\t\techo \"Please run 'make fmt' and commit the result:\"; \\\n\t\techo \"$${diff}\"; \\\n\t\texit 1; \\\n\tfi;\n\nvet:\n\t$(GO) vet $(VETPACKAGES)\n\n.PHONY: lint\nlint:\n\t@hash golint > /dev/null 2>&1; if [ $$? -ne 0 ]; then \\\n\t\t$(GO) get -u golang.org/x/lint/golint; \\\n\tfi\n\tfor PKG in $(PACKAGES); do golint -set_exit_status $$PKG || exit 1; done;\n\n.PHONY: misspell-check\nmisspell-check:\n\t@hash misspell > /dev/null 2>&1; if [ $$? -ne 0 ]; then \\\n\t\t$(GO) get -u github.com/client9/misspell/cmd/misspell; \\\n\tfi\n\tmisspell -error $(GOFILES)\n\n.PHONY: misspell\nmisspell:\n\t@hash misspell > /dev/null 2>&1; if [ $$? -ne 0 ]; then \\\n\t\t$(GO) get -u github.com/client9/misspell/cmd/misspell; \\\n\tfi\n\tmisspell -w $(GOFILES)\n\n.PHONY: tools\ntools:\n\tgo install golang.org/x/lint/golint@latest; \\\n\tgo install github.com/client9/misspell/cmd/misspell@latest; \\\n\tgo install github.com/go-bindata/go-bindata/...@latest;\n\t\n# make plugin\n.PHONY: plugin\nplugin: \n\trm -rf $(HOME)/data/gou-unit/plugins\n\trm -rf $(HOME)/data/gou-unit/logs\n\tmkdir -p $(HOME)/data/gou-unit/plugins\n\tmkdir -p $(HOME)/data/gou-unit/logs\n\tGOOS=linux GOARCH=amd64 go build -o $(HOME)/data/gou-unit/plugins/user.so ./tests/plugins/user\n\tchmod +x $(HOME)/data/gou-unit/plugins/user.so\n\tls -l $(HOME)/data/gou-unit/plugins\n\tls -l $(HOME)/data/gou-unit/logs\n\t$(HOME)/data/gou-unit/plugins/user.so 2>&1 || true\n\n# make plugin-mac\n.PHONY: plugin-mac\nplugin-mac: \n\trm -rf ./tests/plugins/user/dist\n\trm -rf ./tests/plugins/user.so\n\tgo build -o ./tests/plugins/user.so ./tests/plugins/user\n\tchmod +x ./tests/plugins/user.so\n\n\n# make pack\n.PHONY: pack \npack: bindata fmt\n\n.PHONY: bindata\nbindata:\n\n# Setup Workdir\n\trm -rf .tmp/data\n\trm -rf .tmp/yao-init\n\tmkdir -p .tmp/data\n\n#\tCheckout init\n\tgit clone https://github.com/YaoApp/yao-init.git .tmp/yao-init\n\trm -rf .tmp/yao-init/.git\n\trm -rf .tmp/yao-init/.gitignore\n\trm -rf .tmp/yao-init/LICENSE\n#\trm -rf .tmp/yao-init/README.md\n\t\n#\tCopy Files\n\tcp -r .tmp/yao-init .tmp/data/init\n\tcp -r ui .tmp/data/\n\tcp -r ui .tmp/data/public\n\tcp -r cui .tmp/data/\n\tcp -r yao .tmp/data/\n\tcp -r sui/libsui .tmp/data/\n\tfind .tmp/data -name \".DS_Store\" -type f -delete\n\tgo-bindata -fs -pkg data -o data/bindata.go -prefix \".tmp/data/\" .tmp/data/...\n\trm -rf .tmp/data\n\trm -rf .tmp/yao-init\n\n# make artifacts-linux\n.PHONY: artifacts-linux\nartifacts-linux: clean\n\tmkdir -p dist/release\n\n#\tBuilding CUI v1.0\n\texport NODE_ENV=production\n# \trm -f ../cui-v1.0/pnpm-lock.yaml\n\techo \"BASE=__yao_admin_root\" > ../cui-v1.0/packages/cui/.env\n\tcd ../cui-v1.0 && pnpm install --no-frozen-lockfile && pnpm run build\n\n#\tInit Application\n\tcd ../yao-init && rm -rf .git\n\tcd ../yao-init && rm -rf .gitignore\n\tcd ../yao-init && rm -rf LICENSE\n#\tcd ../yao-init rm -rf README.md\n\n#\tSwitch .env login URLs from dev mode (__yao_admin_root) to release mode (dashboard)\n\tsed -i.bak 's|AFTER_LOGIN_SUCCESS_URL=\"/__yao_admin_root/|# AFTER_LOGIN_SUCCESS_URL=\"/__yao_admin_root/|g' ../yao-init/.env\n\tsed -i.bak 's|AFTER_LOGIN_FAILURE_URL=\"/__yao_admin_root/|# AFTER_LOGIN_FAILURE_URL=\"/__yao_admin_root/|g' ../yao-init/.env\n\tsed -i.bak 's|# AFTER_LOGIN_SUCCESS_URL=\"/dashboard/|AFTER_LOGIN_SUCCESS_URL=\"/dashboard/|g' ../yao-init/.env\n\tsed -i.bak 's|# AFTER_LOGIN_FAILURE_URL=\"/dashboard/|AFTER_LOGIN_FAILURE_URL=\"/dashboard/|g' ../yao-init/.env\n\trm -f ../yao-init/.env.bak\n\n# Yao Builder\n# Remove Yao Builder - DUI PageBuilder component will provide online design for pure HTML pages or SUI pages in the future.\n#\tmkdir -p .tmp/data/builder\n#\tcurl -o .tmp/yao-builder-latest.tar.gz https://release-sv.yaoapps.com/archives/yao-builder-latest.tar.gz\n#\ttar -zxvf .tmp/yao-builder-latest.tar.gz -C .tmp/data/builder\n#\trm -rf .tmp/yao-builder-latest.tar.gz\n\n#\tPacking\n# ** CUI will be renamed to CUI in the feature. and move to the new repository. **\n# ** new repository: https://github.com/YaoApp/cui.git **\n\tmkdir -p .tmp/data/cui\n\tcp -r ./ui .tmp/data/ui\n\tcp -r ../cui-v1.0/packages/cui/dist .tmp/data/cui/v1.0\n\tcp -r ../yao-init .tmp/data/init\n\tcp -r yao .tmp/data/\n\tcp -r sui/libsui .tmp/data/\n\tgo-bindata -fs -pkg data -o data/bindata.go -prefix \".tmp/data/\" .tmp/data/...\n\trm -rf .tmp/data\n\n#\tReplace PRVERSION\n\tsed -ie \"s/const PRVERSION = \\\"DEV\\\"/const PRVERSION = \\\"${COMMIT}-${NOW}\\\"/g\" share/const.go\n\t@CUI_COMMIT=$$(cd ../cui-v1.0 && git log | head -n 1 | awk '{print substr($$2, 0, 12)}') && \\\n\tsed -ie \"s/const PRCUI = \\\"DEV\\\"/const PRCUI = \\\"$$CUI_COMMIT-${NOW}\\\"/g\" share/const.go\n\n# Making artifacts - dev builds (full debug symbols, ~158M)\n\tmkdir -p dist\n\tCGO_ENABLED=1 CGO_LDFLAGS=\"-static\" GOOS=linux GOARCH=amd64 go build -v -o dist/yao-${VERSION}-linux-amd64\n\tCGO_ENABLED=1 CGO_LDFLAGS=\"-static\" LD_LIBRARY_PATH=/usr/lib/gcc-cross/aarch64-linux-gnu/13 GOOS=linux GOARCH=arm64 CC=aarch64-linux-gnu-gcc-13 CXX=aarch64-linux-gnu-g++-13 go build -v -o dist/yao-${VERSION}-linux-arm64\n\n# Making artifacts - prod builds (stripped, ~111M)\n\tsed -i.tmp 's/const BUILDOPTIONS = \"\"/const BUILDOPTIONS = \"-s -w (production, stripped)\"/g' share/const.go && rm -f share/const.go.tmp\n\tCGO_ENABLED=1 CGO_LDFLAGS=\"-static\" GOOS=linux GOARCH=amd64 go build -v -ldflags=\"-s -w\" -o dist/yao-${VERSION}-linux-amd64-prod\n\tCGO_ENABLED=1 CGO_LDFLAGS=\"-static\" LD_LIBRARY_PATH=/usr/lib/gcc-cross/aarch64-linux-gnu/13 GOOS=linux GOARCH=arm64 CC=aarch64-linux-gnu-gcc-13 CXX=aarch64-linux-gnu-g++-13 go build -v -ldflags=\"-s -w\" -o dist/yao-${VERSION}-linux-arm64-prod\n\n\tmkdir -p dist/release\n\tmv dist/yao-*-* dist/release/\n\tchmod +x dist/release/yao-*-*\n\tls -l dist/release/\n\tdist/release/yao-${VERSION}-linux-amd64 version\n\n# \tReset const \n#\tcp -f share/const.goe share/const.go\n#\trm -f share/const.goe\n\n# make artifacts-macos\n.PHONY: artifacts-macos\nartifacts-macos: clean\n\n\tmkdir -p dist/release\n\n#\tBuilding CUI v1.0\n\texport NODE_ENV=production\n# rm -f ../cui-v1.0/pnpm-lock.yaml\n\techo \"BASE=__yao_admin_root\" > ../cui-v1.0/packages/cui/.env\n\tcd ../cui-v1.0 && pnpm install --no-frozen-lockfile && pnpm run build\n\n#\tInit Application\n\tcd ../yao-init && rm -rf .git\n\tcd ../yao-init && rm -rf .gitignore\n\tcd ../yao-init && rm -rf LICENSE\n#\t cd ../yao-init && rm -rf README.md\n\n#\tSwitch .env login URLs from dev mode (__yao_admin_root) to release mode (dashboard)\n\tsed -i.bak 's|AFTER_LOGIN_SUCCESS_URL=\"/__yao_admin_root/|# AFTER_LOGIN_SUCCESS_URL=\"/__yao_admin_root/|g' ../yao-init/.env\n\tsed -i.bak 's|AFTER_LOGIN_FAILURE_URL=\"/__yao_admin_root/|# AFTER_LOGIN_FAILURE_URL=\"/__yao_admin_root/|g' ../yao-init/.env\n\tsed -i.bak 's|# AFTER_LOGIN_SUCCESS_URL=\"/dashboard/|AFTER_LOGIN_SUCCESS_URL=\"/dashboard/|g' ../yao-init/.env\n\tsed -i.bak 's|# AFTER_LOGIN_FAILURE_URL=\"/dashboard/|AFTER_LOGIN_FAILURE_URL=\"/dashboard/|g' ../yao-init/.env\n\trm -f ../yao-init/.env.bak\n\n#\tPacking\n\tmkdir -p .tmp/data/cui\n\tcp -r ./ui .tmp/data/ui\n\tcp -r ../cui-v1.0/packages/cui/dist .tmp/data/cui/v1.0\n\tcp -r ../yao-init .tmp/data/init\n\tcp -r yao .tmp/data/\n\tcp -r sui/libsui .tmp/data/\n\tgo-bindata -fs -pkg data -o data/bindata.go -prefix \".tmp/data/\" .tmp/data/...\n\trm -rf .tmp/data\n\n#\tReplace PRVERSION\n\tsed -ie \"s/const PRVERSION = \\\"DEV\\\"/const PRVERSION = \\\"${COMMIT}-${NOW}\\\"/g\" share/const.go\n\t@CUI_COMMIT=$$(cd ../cui-v1.0 && git log | head -n 1 | awk '{print substr($$2, 0, 12)}') && \\\n\tsed -ie \"s/const PRCUI = \\\"DEV\\\"/const PRCUI = \\\"$$CUI_COMMIT-${NOW}\\\"/g\" share/const.go\n\n# Making artifacts - dev builds (full debug symbols)\n\tmkdir -p dist\n\tCGO_ENABLED=1 GOOS=darwin GOARCH=amd64 go build -v -o dist/yao-${VERSION}-darwin-amd64\n\tCGO_ENABLED=1 GOOS=darwin GOARCH=arm64 go build -v -o dist/yao-${VERSION}-darwin-arm64\n\n# Making artifacts - prod builds (stripped, no UPX on macOS)\n\tsed -i.tmp 's/const BUILDOPTIONS = \"\"/const BUILDOPTIONS = \"-s -w (production, stripped)\"/g' share/const.go && rm -f share/const.go.tmp\n\tCGO_ENABLED=1 GOOS=darwin GOARCH=amd64 go build -v -ldflags=\"-s -w\" -o dist/yao-${VERSION}-darwin-amd64-prod\n\tCGO_ENABLED=1 GOOS=darwin GOARCH=arm64 go build -v -ldflags=\"-s -w\" -o dist/yao-${VERSION}-darwin-arm64-prod\n\n\tmkdir -p dist/release\n\tmv dist/yao-*-* dist/release/\n\tchmod +x dist/release/yao-*-*\n\tls -l dist/release/\n\tdist/release/yao-${VERSION}-darwin-amd64 version\n\n\n.PHONY: debug\ndebug: clean\n\tmkdir -p dist/release\n\n#\tPacking\n#\tmkdir -p .tmp/data\n#\tcp -r ui .tmp/data/ui\n#\tcp -r yao .tmp/data/\n#\tgo-bindata -fs -pkg data -o data/bindata.go -prefix \".tmp/data/\" .tmp/data/...\n#\trm -rf .tmp/data\n\n\n#\tReplace PRVERSION\n\tsed -ie \"s/const PRVERSION = \\\"DEV\\\"/const PRVERSION = \\\"${COMMIT}-${NOW}-debug\\\"/g\" share/const.go\n\n# Making artifacts\n\tmkdir -p dist\n\tCGO_ENABLED=1 go build -v -o dist/release/yao-debug\n\tchmod +x dist/release/yao-debug\n\n# \tReset const \n\tcp -f share/const.goe share/const.go\n\trm -f share/const.goe\n\n# make prepare (build CUI, yao-init, bindata - shared by release and prod)\n.PHONY: prepare\nprepare: clean\n\tmkdir -p dist/release\n\tmkdir .tmp\n\n#\tBuilding CUI v0.9\n\tmkdir -p .tmp/cui/v0.9/dist\n\techo \"CUI v0.9\" > .tmp/cui/v0.9/dist/index.html\n\n#\tBuilding CUI v1.0\n# ** CUI will be renamed to CUI in the feature. and move to the new repository. **\n# ** new repository: https://github.com/YaoApp/cui.git **\n\texport NODE_ENV=production\n\tgit clone https://github.com/YaoApp/cui.git .tmp/cui/v1.0\n# \tcd .tmp/cui/v1.0 && git checkout 5002c3fded585aaa69a4366135b415ea3234964e\n\techo \"BASE=__yao_admin_root\" > .tmp/cui/v1.0/packages/cui/.env\n\tcd .tmp/cui/v1.0 && pnpm install --no-frozen-lockfile && pnpm run build\n\tCUI_COMMIT=$$(cd .tmp/cui/v1.0 && git rev-parse --short HEAD)\n\n#\tCheckout init\n\tgit clone https://github.com/YaoApp/yao-init.git .tmp/yao-init\n\trm -rf .tmp/yao-init/.git\n\trm -rf .tmp/yao-init/.gitignore\n\trm -rf .tmp/yao-init/LICENSE\n\trm -rf .tmp/yao-init/README.md\n\n#\tSwitch .env login URLs from dev mode (__yao_admin_root) to release mode (dashboard)\n\tsed -i.bak 's|AFTER_LOGIN_SUCCESS_URL=\"/__yao_admin_root/|# AFTER_LOGIN_SUCCESS_URL=\"/__yao_admin_root/|g' .tmp/yao-init/.env\n\tsed -i.bak 's|AFTER_LOGIN_FAILURE_URL=\"/__yao_admin_root/|# AFTER_LOGIN_FAILURE_URL=\"/__yao_admin_root/|g' .tmp/yao-init/.env\n\tsed -i.bak 's|# AFTER_LOGIN_SUCCESS_URL=\"/dashboard/|AFTER_LOGIN_SUCCESS_URL=\"/dashboard/|g' .tmp/yao-init/.env\n\tsed -i.bak 's|# AFTER_LOGIN_FAILURE_URL=\"/dashboard/|AFTER_LOGIN_FAILURE_URL=\"/dashboard/|g' .tmp/yao-init/.env\n\trm -f .tmp/yao-init/.env.bak\n\n# Yao Builder\n# Remove Yao Builder - DUI PageBuilder component will provide online design for pure HTML pages or SUI pages in the future.\n#\tmkdir -p .tmp/data/builder\n#\tcurl -o .tmp/yao-builder-latest.tar.gz https://release-sv.yaoapps.com/archives/yao-builder-latest.tar.gz\n#\ttar -zxvf .tmp/yao-builder-latest.tar.gz -C .tmp/data/builder\n#\trm -rf .tmp/yao-builder-latest.tar.gz\n\n#\tPacking\n\tcp -f data/bindata.go data/bindata.go.bak\n\tmkdir -p .tmp/data/cui\n\tcp -r ./ui .tmp/data/ui\n\tcp -r ./yao .tmp/data/yao\n\tcp -r ./sui/libsui .tmp/data/libsui\n\tcp -r .tmp/cui/v0.9/dist .tmp/data/cui/v0.9\n\tcp -r .tmp/cui/v1.0/packages/cui/dist .tmp/data/cui/v1.0\n\tcp -r .tmp/yao-init .tmp/data/init\n\tgo-bindata -fs -pkg data -o data/bindata.go -prefix \".tmp/data/\" .tmp/data/...\n\n#\tReplace PRVERSION\n\tcp -f share/const.go share/const.go.bak\n\tsed -ie \"s/const PRVERSION = \\\"DEV\\\"/const PRVERSION = \\\"${COMMIT}-${NOW}\\\"/g\" share/const.go\n\t@CUI_COMMIT=$$(cd .tmp/cui/v1.0 && git log | head -n 1 | awk '{print substr($$2, 0, 12)}') && \\\n\tsed -ie \"s/const PRCUI = \\\"DEV\\\"/const PRCUI = \\\"$$CUI_COMMIT-${NOW}\\\"/g\" share/const.go\n\n# make release (development build only, ~158M)\n.PHONY: release\nrelease: prepare\n# Making artifacts - dev build\n\tmkdir -p dist\n\tCGO_ENABLED=1 go build -v -o dist/release/yao\n\tchmod +x dist/release/yao\n\n# \tClean up and restore bindata.go and const.go\n\tcp data/bindata.go.bak data/bindata.go\n\tcp share/const.go.bak share/const.go\n\trm data/bindata.go.bak\n\trm share/const.go.bak\n\trm -rf .tmp\n\n# MacOS Application Signing\n\t@if [ \"$(OS)\" = \"Darwin\" ]; then \\\n\t codesign --deep --force --verbose --timestamp --options runtime \\\n\t --entitlements .github/codesign/entitlements.plist \\\n\t --sign \"${APPLE_SIGN}\" dist/release/yao ; \\\n\tfi\n\n# make prod (production build only, ~111M on macOS)\n.PHONY: prod\nprod: prepare\n#\tSet BUILDOPTIONS\n\t@if [ \"$$(uname)\" = \"Linux\" ]; then \\\n\t\tsed -i.tmp 's/const BUILDOPTIONS = \"\"/const BUILDOPTIONS = \"-s -w +upx (production, compressed)\"/g' share/const.go && rm -f share/const.go.tmp; \\\n\telse \\\n\t\tsed -i.tmp 's/const BUILDOPTIONS = \"\"/const BUILDOPTIONS = \"-s -w (production, stripped)\"/g' share/const.go && rm -f share/const.go.tmp; \\\n\tfi\n\n# Making artifacts - prod build\n\tmkdir -p dist\n\tCGO_ENABLED=1 go build -v -ldflags=\"-s -w\" -o dist/release/yao-prod\n\tchmod +x dist/release/yao-prod\n\n#\tUPX compression (Linux only)\n\t@if [ \"$$(uname)\" = \"Linux\" ]; then \\\n\t\techo \"Compressing with UPX...\"; \\\n\t\tif command -v upx > /dev/null 2>&1; then \\\n\t\t\tupx --best dist/release/yao-prod; \\\n\t\telse \\\n\t\t\techo \"WARNING: UPX not found. Install with: apt install upx\"; \\\n\t\t\techo \"Skipping compression.\"; \\\n\t\tfi; \\\n\telse \\\n\t\techo \"Note: UPX compression skipped on macOS (not supported)\"; \\\n\tfi\n\n# \tClean up and restore bindata.go and const.go\n\tcp data/bindata.go.bak data/bindata.go\n\tcp share/const.go.bak share/const.go\n\trm data/bindata.go.bak\n\trm share/const.go.bak\n\trm -rf .tmp\n\n# MacOS Application Signing\n\t@if [ \"$(OS)\" = \"Darwin\" ]; then \\\n\t codesign --deep --force --verbose --timestamp --options runtime \\\n\t --entitlements .github/codesign/entitlements.plist \\\n\t --sign \"${APPLE_SIGN}\" dist/release/yao-prod ; \\\n\tfi\n\n\t@echo \"\"\n\t@echo \"Done! Production binary:\"\n\t@ls -lh dist/release/yao-prod\n\t@echo \"\"\n\t@echo \"Test with: dist/release/yao-prod version --all\"\n\n# make release-all (build both dev and prod in one go)\n.PHONY: release-all\nrelease-all: prepare\n# Making artifacts - dev build (~158M)\n\t@echo \"Building dev binary...\"\n\tmkdir -p dist\n\tCGO_ENABLED=1 go build -v -o dist/release/yao\n\tchmod +x dist/release/yao\n\n# Making artifacts - prod build (~111M on macOS)\n\t@echo \"Building prod binary...\"\n\t@if [ \"$$(uname)\" = \"Linux\" ]; then \\\n\t\tsed -i.tmp 's/const BUILDOPTIONS = \"\"/const BUILDOPTIONS = \"-s -w +upx (production, compressed)\"/g' share/const.go && rm -f share/const.go.tmp; \\\n\telse \\\n\t\tsed -i.tmp 's/const BUILDOPTIONS = \"\"/const BUILDOPTIONS = \"-s -w (production, stripped)\"/g' share/const.go && rm -f share/const.go.tmp; \\\n\tfi\n\tCGO_ENABLED=1 go build -v -ldflags=\"-s -w\" -o dist/release/yao-prod\n\tchmod +x dist/release/yao-prod\n\n#\tUPX compression (Linux only)\n\t@if [ \"$$(uname)\" = \"Linux\" ]; then \\\n\t\techo \"Compressing with UPX...\"; \\\n\t\tif command -v upx > /dev/null 2>&1; then \\\n\t\t\tupx --best dist/release/yao-prod; \\\n\t\telse \\\n\t\t\techo \"WARNING: UPX not found. Install with: apt install upx\"; \\\n\t\t\techo \"Skipping compression.\"; \\\n\t\tfi; \\\n\telse \\\n\t\techo \"Note: UPX compression skipped on macOS (not supported)\"; \\\n\tfi\n\n# \tClean up and restore bindata.go and const.go\n\tcp data/bindata.go.bak data/bindata.go\n\tcp share/const.go.bak share/const.go\n\trm data/bindata.go.bak\n\trm share/const.go.bak\n\trm -rf .tmp\n\n# MacOS Application Signing\n\t@if [ \"$(OS)\" = \"Darwin\" ]; then \\\n\t codesign --deep --force --verbose --timestamp --options runtime \\\n\t --entitlements .github/codesign/entitlements.plist \\\n\t --sign \"${APPLE_SIGN}\" dist/release/yao ; \\\n\t codesign --deep --force --verbose --timestamp --options runtime \\\n\t --entitlements .github/codesign/entitlements.plist \\\n\t --sign \"${APPLE_SIGN}\" dist/release/yao-prod ; \\\n\tfi\n\n\t@echo \"\"\n\t@echo \"Done! Binaries:\"\n\t@ls -lh dist/release/yao dist/release/yao-prod\n\t@echo \"\"\n\t@echo \"Test with:\"\n\t@echo \" dist/release/yao version --all\"\n\t@echo \" dist/release/yao-prod version --all\"\n\n\n.PHONY: linux-release\nlinux-release: clean\n\tmkdir -p dist/release\n\tmkdir .tmp\n\n#\tBuilding CUI v1.0\n# ** CUI will be renamed to CUI in the feature. and move to the new repository. **\n# ** new repository: https://github.com/YaoApp/cui.git **\n\texport NODE_ENV=production\n\tgit clone https://github.com/YaoApp/cui.git .tmp/cui/v1.0\n\trm -f .tmp/cui/v1.0/pnpm-lock.yaml\n\techo \"BASE=__yao_admin_root\" > .tmp/cui/v1.0/packages/cui/.env\n\tcd .tmp/cui/v1.0 && pnpm install --no-frozen-lockfile && pnpm run build\n\n# Setup UI\n\tcd .tmp/cui/v1.0/packages/setup && pnpm install --no-frozen-lockfile && pnpm run build\n\n\n#\tCheckout init\n\tgit clone https://github.com/YaoApp/yao-init.git .tmp/yao-init\n\trm -rf .tmp/yao-init/.git\n\trm -rf .tmp/yao-init/.gitignore\n\trm -rf .tmp/yao-init/LICENSE\n\trm -rf .tmp/yao-init/README.md\n\n# Yao Builder\n# Remove Yao Builder - DUI PageBuilder component will provide online design for pure HTML pages or SUI pages in the future.\n# \tmkdir -p .tmp/data/builder\n# \tcurl -o .tmp/yao-builder-latest.tar.gz https://release-sv.yaoapps.com/archives/yao-builder-latest.tar.gz\n# \ttar -zxvf .tmp/yao-builder-latest.tar.gz -C .tmp/data/builder\n# \trm -rf .tmp/yao-builder-latest.tar.gz\n\n#\tPacking\n\tmkdir -p .tmp/data/cui\n\tcp -r ./ui .tmp/data/ui\n\tcp -r ./yao .tmp/data/yao\n\tcp -r .tmp/cui/v0.9/dist .tmp/data/cui/v0.9\n\tcp -r .tmp/cui/v1.0/packages/setup/build .tmp/data/cui/setup\n\tcp -r .tmp/cui/v1.0/packages/cui/dist .tmp/data/cui/v1.0\n\tcp -r .tmp/yao-init .tmp/data/init\n\tgo-bindata -fs -pkg data -o data/bindata.go -prefix \".tmp/data/\" .tmp/data/...\n\trm -rf .tmp/data\n\trm -rf .tmp/cui\n\n# Making artifacts\n\tmkdir -p dist\n\tCGO_ENABLED=1 CGO_LDFLAGS=\"-static\" go build -v -o dist/release/yao\n\tchmod +x dist/release/yao\n\n# make clean\n.PHONY: clean\nclean: \n\trm -rf ./tmp\n\trm -rf .tmp\n\trm -rf dist" }, { "path": "README.md", "content": "# Yao — Build Autonomous Agents. Just Define the Role.\n\nYao is an open-source engine for autonomous agents — event-driven, proactive, and self-scheduling.\n\n![Mission Control](docs/mission-control.png)\n\n**Quick Links:**\n\n**🏠 Homepage:** [https://yaoapps.com](https://yaoapps.com)\n\n**🚀 Quick Start:** [https://yaoapps.com/docs/documentation/en-us/getting-started](https://yaoapps.com/docs/documentation/en-us/getting-started#quickstart)\n\n**📚 Documentation:** [https://yaoapps.com/docs](https://yaoapps.com/docs)\n\n**✨ Why Yao?** [https://yaoapps.com/docs/why-yao](https://yaoapps.com/docs/documentation/en-us/getting-started/why-yao)\n\n**🤖 Yao Agents:** [https://github.com/YaoAgents/awesome](https://github.com/YaoAgents/awesome) ( Preview )\n\n---\n\n## What Makes Yao Different?\n\n| Traditional AI Assistants | Yao Autonomous Agents |\n| ----------------------------- | ------------------------------------- |\n| Entry point: Chatbox | Entry point: Email, Events, Schedules |\n| Passive: You ask, they answer | Proactive: They work autonomously |\n| Role: Tool | Role: Team member |\n\n> The entry point is not a chatbox — it's email, events, and scheduled tasks.\n\n---\n\n## Features\n\n### Autonomous Agent Framework\n\nBuild agents that work like real team members:\n\n- **Three Trigger Modes** — Clock (scheduled), Human (email/message), Event (webhook/database)\n- **Six-Phase Execution** — Inspiration → Goals → Tasks → Run → Deliver → Learn\n- **Multi-Agent Orchestration** — Agents delegate, collaborate, and compose dynamically\n- **Continuous Learning** — Agents accumulate experience in private knowledge bases\n\n### Native MCP Support\n\nIntegrate tools without writing adapters:\n\n- **Process Transport** — Map Yao processes directly to MCP tools\n- **External Servers** — Connect via SSE or STDIO\n- **Schema Mapping** — Declarative input/output schemas\n\n### Built-in GraphRAG\n\n- **Vector Search** — Embeddings with OpenAI/FastEmbed\n- **Knowledge Graph** — Entity-relationship retrieval\n- **Hybrid Search** — Combine vector similarity with graph traversal\n\n### Full-Stack Runtime\n\nEverything in a single executable:\n\n- **All-in-One** — Data, API, Agent, UI in one engine\n- **TypeScript Support** — Built-in V8 engine\n- **Single Binary** — No Node.js, Python, or containers required\n- **Edge-Ready** — Runs on ARM64/x64 devices\n" }, { "path": "README.zh-CN.md", "content": "# Yao\n\n[![UnitTest](https://github.com/YaoApp/yao/actions/workflows/unit-test.yml/badge.svg)](https://github.com/YaoApp/yao/actions/workflows/unit-test.yml)\n[![codecov](https://codecov.io/gh/YaoApp/yao/branch/main/graph/badge.svg?token=294Y05U71J)](https://codecov.io/gh/YaoApp/yao)\n\nhttps://github.com/YaoApp/yao/assets/1842210/6b23ac89-ef6e-4c24-874f-753a98370dec\n\n[English](README.md)\n\nYAO 是一款开源应用引擎，使用 Golang 编写，以一个命令行工具的形式存在, 下载即用。适合用于开发业务系统、网站/APP API 接口、管理后台、自建低代码平台等。\n\nYAO 采用 flow-based 的编程模式，通过编写 YAO DSL (JSON 格式逻辑描述) 或使用 JavaScript 编写处理器，实现各种功能。 YAO DSL 可以有多种编写方式:\n\n1. 纯手工编写\n\n2. 使用自动化脚本，根据上下文逻辑生成\n\n3. 使用可视化编辑器，通过“拖拉拽”制作\n\n官网: [https://yaoapps.com](https://yaoapps.com)\n\n文档: [https://yaoapps.com/doc](https://yaoapps.com/doc)\n\n## 最新版本下载安装 (推荐)\n\nhttps://github.com/YaoApp/xgen-dev-app\n\n## 演示\n\n![界面](docs/yao-setup-demo.jpg)\n\n使用 YAO 开发的应用\n\n| 应用 | 简介 | 代码仓库 |\n| -------------------- | ---------------------------- | --------------------------------------- |\n| yaoapp/yao-examples | Yao 应用示例 | https://github.com/YaoApp/yao-examples |\n| yaoapp/yao-knowledge | ChatGPT 驱动的知识管理库应用 | https://github.com/YaoApp/yao-knowledge |\n| yaoapp/xgen-dev-app | 演示应用 (演示) | https://github.com/YaoApp/xgen-dev-app |\n| yaoapp/demo-project | 工程项目管理演示应用(演示) | https://github.com/yaoapp/demo-project |\n| yaoapp/demo-finance | 财务管理演示应用(演示) | https://github.com/yaoapp/demo-finance |\n| yaoapp/demo-plm | 生产项目管理演示应用(演示) | https://github.com/yaoapp/demo-plm |\n\n## 介绍\n\nYao 是一个只需使用 JSON 即可创建数据库模型、编写 API 接口、描述管理后台界面的应用引擎，使用 Yao 构建的应用可运行在云端或物联网设备上。 开发者不需要写一行代码，就可以拥有 10 倍生产力。\n\nYao 基于 **flow-based** 编程思想，采用 **Go** 语言开发，支持多种方式扩展数据流处理器。这使得 Yao 具有极好的**通用性**，大部分场景下可以代替编程语言, 在复用性和编码效率上是传统编程语言的 **10 倍**；应用性能和资源占比上优于 **PHP**, **JAVA** 等语言。\n\nYao 内置了一套数据管理系统，通过编写 **JSON** 描述界面布局，即可实现 90% 常见界面交互功能，特别适合快速制作各类管理后台、CRM、ERP 等企业内部系统。对于特殊交互功能亦可通过编写扩展组件或 HTML 页面的方式实现。内置管理系统与 Yao 并不耦合，亦可采用 **VUE**, **React** 等任意前端技术实现管理界面。\n\n## 安装\n\nYao v0.10.4 使用说明\n\nhttps://github.com/YaoApp/xgen-dev-app/blob/main/README.zh-CN.md\n\n## 入门指南\n\n详细说明请看[文档](https://yaoapps.com/doc/%E4%BB%8B%E7%BB%8D/%E5%85%A5%E9%97%A8%E6%8C%87%E5%8D%97)\n\n### 创建应用\n\n#### 新建一个空白应用\n\n新建一个应用目录，进入应用目录，运行 `yao start` 命令, 启动安装界面。\n\n```bash\nmkdir -p /data/app # 创建应用目录\ncd /data/app # 进入应用目录\nyao start # 启动安装界面\n```\n\n**默认账号**\n\n- 用户名: **xiang@iqka.com**\n\n- 密码: **A123456p+**\n\n![安装界面](docs/yao-setup-step2.jpg)\n\n## 关于 Yao\n\nYao 的名字源于汉字**爻(yáo)**，是构成八卦的基本符号。八卦，是上古大神伏羲观测总结自然规律后，创造的一个可以指代万事万物的符号体系。爻，有阴阳两种状态，就像 0 和 1。爻的阴阳转换，驱动八卦更替，以此来总结记录事物的发展规律。\n" }, { "path": "agent/README.md", "content": "# Yao Agent\n\nA powerful AI assistant framework for building intelligent conversational agents with tool integration, knowledge base search, and multi-agent orchestration.\n\n## Quick Start\n\n### 1. Create an Assistant\n\n```\nassistants/\n└── my-assistant/\n ├── package.yao # Configuration\n ├── prompts.yml # System prompts\n └── locales/\n └── en-us.yml # Translations\n```\n\n**package.yao**\n\n```json\n{\n \"name\": \"{{ name }}\",\n \"connector\": \"gpt-4o\",\n \"description\": \"{{ description }}\",\n \"placeholder\": {\n \"title\": \"{{ chat.title }}\",\n \"prompts\": [\"{{ chat.prompts.0 }}\"]\n }\n}\n```\n\n**prompts.yml**\n\n```yaml\n- role: system\n content: |\n You are a helpful assistant.\n```\n\n**locales/en-us.yml**\n\n```yaml\nname: My Assistant\ndescription: A helpful AI assistant\nchat:\n title: New Chat\n prompts:\n - How can I help you today?\n```\n\n### 2. Add Hooks (Optional)\n\nCreate `src/index.ts` for custom logic:\n\n```typescript\nimport { agent } from \"@yao/runtime\";\n\nfunction Create(ctx: agent.Context, messages: agent.Message[]): agent.Create {\n // Preprocess messages before LLM call\n return { messages };\n}\n\nfunction Next(ctx: agent.Context, payload: agent.Payload): agent.Next {\n // Post-process LLM response\n return null;\n}\n```\n\n### 3. Test (Optional)\n\n```bash\n# Run tests\nyao agent test -i \"Hello, how are you?\"\n\n# Run tests from JSONL file\nyao agent test -i tests/inputs.jsonl -v\n\n# Extract results for review\nyao agent extract output-*.jsonl\n```\n\n### 4. Run\n\n```bash\nyao start\n```\n\nAccess via API: `POST /v1/chat/completions`\n\n## Examples\n\n### Hook: Route to Specialist\n\n```typescript\n// src/index.ts\nfunction Create(ctx: agent.Context, messages: agent.Message[]): agent.Create {\n const last = messages[messages.length - 1]?.content || \"\";\n if (last.includes(\"refund\")) {\n return { delegate: { agent_id: \"refund-specialist\", messages } };\n }\n return null;\n}\n```\n\n### Database Query\n\n```json\n// package.yao - Enable auto DB search\n{ \"db\": { \"models\": [\"orders\", \"products\"] } }\n```\n\n```bash\n# Test: Agent auto-generates QueryDSL and searches database\nyao agent test -i \"Find orders over $1000 from last month\"\n```\n\n### MCP Tools (Process Transport)\n\n```json\n// mcps/tools.mcp.yao - Define MCP server with Yao Processes\n{\n \"label\": \"Tools\",\n \"transport\": \"process\",\n \"tools\": {\n \"search_orders\": \"models.order.Paginate\",\n \"create_order\": \"models.order.Create\"\n }\n}\n```\n\n```json\n// mcps/mapping/tools/schemes/search_orders.in.yao - Input schema\n{\n \"type\": \"object\",\n \"properties\": {\n \"keyword\": { \"type\": \"string\" },\n \"page\": { \"type\": \"integer\" }\n },\n \"x-process-args\": [\":arguments\"]\n}\n```\n\n```json\n// package.yao\n{ \"mcp\": { \"servers\": [{ \"server_id\": \"tools\" }] } }\n```\n\n### Sidebar Page (Display Data)\n\nPages render in the right sidebar during conversation to display structured data:\n\n```html\n\n
\n

{{ title }}

\n 0 }}\">\n \n \n \n \n
{{ row.name }}{{ row.value }}
\n
\n```\n\n```bash\nyao sui build agent # Build pages\n```\n\n```javascript\n// In hook: send action to open page in sidebar\nctx.Send({\n type: \"action\",\n props: {\n name: \"navigate\",\n payload: {\n route: \"/agents/my-assistant/result\",\n title: \"Query Results\",\n query: { id: \"123\" }, // Passed as $query in page\n },\n },\n});\n```\n\n## Documentation\n\n- [Configuration](docs/configuration.md) - Assistant settings, connectors, options\n- [Prompts](docs/prompts.md) - System prompts and prompt presets\n- [Hooks](docs/hooks.md) - Create/Next hooks and agent lifecycle\n- [Context API](docs/context-api.md) - Messaging, memory, trace, MCP\n- [MCP Integration](docs/mcp.md) - Tool servers and resources\n- [Models](docs/models.md) - Assistant-scoped data models\n- [Search](docs/search.md) - Web, knowledge base, and database search\n- [Pages](docs/pages.md) - Web UI for agents (SUI framework)\n- [Iframe Integration](docs/iframe.md) - Iframe communication with CUI\n- [Internationalization](docs/i18n.md) - Multi-language support\n- [Testing](docs/testing.md) - Agent testing framework\n\n## Architecture\n\n```mermaid\nflowchart LR\n subgraph Request\n A[User Request]\n end\n\n subgraph Create[\"Create Hook\"]\n B1[Preprocess Messages]\n B2[Configure LLM]\n B3[Delegate to Agent]\n end\n\n subgraph LLM[\"LLM Call\"]\n C1[Load Prompts]\n C2[Generate Response]\n end\n\n subgraph Tools[\"Tool Execution\"]\n D1[MCP Tools]\n D2[Search]\n D3[Memory]\n end\n\n subgraph Next[\"Next Hook\"]\n E1[Process Results]\n E2[Transform Output]\n E3[Delegate to Agent]\n end\n\n subgraph Response\n F[Stream Response]\n end\n\n A --> Create\n Create --> LLM\n LLM --> Tools\n Tools --> Next\n Next --> Response\n Next -.->|Continue| LLM\n```\n\n## API Endpoints\n\nOpenAPI endpoints (base URL: `/v1`):\n\n| Endpoint | Method | Description |\n| -------------------------------------- | ------ | --------------------- |\n| `/v1/chat/completions` | POST | Chat with assistant |\n| `/v1/chat/sessions` | GET | List chat sessions |\n| `/v1/chat/sessions/:chat_id` | GET | Get chat session |\n| `/v1/chat/sessions/:chat_id/messages` | GET | Get messages |\n| `/v1/agent/assistants` | GET | List assistants |\n| `/v1/agent/assistants/:id` | GET | Get assistant details |\n| `/v1/file/:uploaderID` | POST | Upload files |\n| `/v1/file/:uploaderID/:fileID` | GET | Get file info |\n| `/v1/file/:uploaderID/:fileID/content` | GET | Download file |\n\n## License\n\nThis project is part of the Yao App Engine and follows the [Yao Open Source License](../LICENSE).\n" }, { "path": "agent/agent_test.go", "content": "package agent\n\n// type customResponseRecorder struct {\n// \t*httptest.ResponseRecorder\n// \tcloseChannel chan bool\n// }\n\n// func (r *customResponseRecorder) CloseNotify() <-chan bool {\n// \treturn r.closeChannel\n// }\n\n// func newCustomResponseRecorder() *customResponseRecorder {\n// \treturn &customResponseRecorder{\n// \t\tResponseRecorder: httptest.NewRecorder(),\n// \t\tcloseChannel: make(chan bool, 1),\n// \t}\n// }\n\n// func TestDSL_Prompts(t *testing.T) {\n// \ttest.Prepare(t, config.Conf)\n// \tdefer Test_clean(t)\n\n// \tresetDB()\n// \tagent := &DSL{\n// \t\tPrompts: []Prompt{\n// \t\t\t{Role: \"system\", Content: \"You are a helpful assistant\", Name: \"ai\"},\n// \t\t\t{Role: \"user\", Content: \"Hello\", Name: \"user\"},\n// \t\t},\n// \t\tConversationSetting: conversation.Setting{\n// \t\t\tConnector: \"default\",\n// \t\t\tTable: \"chat_messages\",\n// \t\t},\n// \t}\n// \terr := agent.newConversation()\n// \tassert.NoError(t, err)\n\n// \tprompts := agent.prompts()\n// \tassert.Equal(t, 2, len(prompts))\n// \tassert.Equal(t, \"system\", prompts[0][\"role\"])\n// \tassert.Equal(t, \"You are a helpful assistant\", prompts[0][\"content\"])\n// \tassert.Equal(t, \"ai\", prompts[0][\"name\"])\n// }\n\n// func TestDSL_ChatMessages(t *testing.T) {\n// \ttest.Prepare(t, config.Conf)\n// \tdefer Test_clean(t)\n\n// \tresetDB()\n// \tagent := &DSL{\n// \t\tPrompts: []Prompt{\n// \t\t\t{Role: \"system\", Content: \"You are a helpful assistant\"},\n// \t\t},\n// \t\tConversationSetting: conversation.Setting{\n// \t\t\tConnector: \"default\",\n// \t\t\tTable: \"chat_messages\",\n// \t\t},\n// \t}\n\n// \terr := agent.newConversation()\n// \tassert.NoError(t, err)\n\n// \tctx := Context{\n// \t\tSid: \"test-session\",\n// \t\tChatID: \"test-chat\",\n// \t}\n\n// \tmessages, err := agent.chatMessages(ctx, \"Hello AI\")\n// \tassert.NoError(t, err)\n// \tassert.Equal(t, 2, len(messages))\n// \tassert.Equal(t, \"system\", messages[0][\"role\"])\n// \tassert.Equal(t, \"user\", messages[1][\"role\"])\n// \tassert.Equal(t, \"Hello AI\", messages[1][\"content\"])\n// }\n\n// func TestDSL_Answer(t *testing.T) {\n// \ttest.Prepare(t, config.Conf)\n// \tdefer Test_clean(t)\n\n// \tgin.SetMode(gin.TestMode)\n// \tw := newCustomResponseRecorder()\n// \tc, _ := gin.CreateTestContext(w)\n\n// \tctx := Context{\n// \t\tSid: \"test-session\",\n// \t\tChatID: \"test-chat\",\n// \t\tContext: context.Background(),\n// \t}\n\n// \tresetDB()\n// \tagent := &DSL{\n// \t\tConnector: \"gpt-3_5-turbo\",\n// \t\tOption: map[string]interface{}{\n// \t\t\t\"temperature\": 0.7,\n// \t\t\t\"max_tokens\": 150,\n// \t\t},\n// \t\tPrompts: []Prompt{\n// \t\t\t{Role: \"system\", Content: \"You are a helpful assistant\"},\n// \t\t},\n// \t\tConversationSetting: conversation.Setting{\n// \t\t\tConnector: \"default\",\n// \t\t\tTable: \"chat_messages\",\n// \t\t},\n// \t}\n\n// \terr := agent.newAI()\n// \tassert.NoError(t, err)\n\n// \terr = agent.newConversation()\n// \tassert.NoError(t, err)\n\n// \tc.Request = httptest.NewRequest(\"POST\", \"/chat\", nil)\n\n// \tagent.AI = &mockAI{}\n\n// \terr = agent.Answer(ctx, \"Hello AI\", c)\n// \tassert.NoError(t, err)\n// }\n\n// // func TestDSL_NewAI(t *testing.T) {\n// // \ttest.Prepare(t, config.Conf)\n// // \tdefer Test_clean(t)\n\n// // \ttests := []struct {\n// // \t\tname string\n// // \t\tconnector string\n// // \t\twantErr string\n// // \t}{\n// // \t\t{\n// // \t\t\tname: \"Mock AI\",\n// // \t\t\tconnector: \"mock\",\n// // \t\t\twantErr: \"\",\n// // \t\t},\n// // \t\t{\n// // \t\t\tname: \"Specific mock model\",\n// // \t\t\tconnector: \"mock:gpt-4\",\n// // \t\t\twantErr: \"\",\n// // \t\t},\n// // \t\t{\n// // \t\t\tname: \"Invalid connector\",\n// // \t\t\tconnector: \"invalid-connector\",\n// // \t\t\twantErr: \"AI connector invalid-connector not found\",\n// // \t\t},\n// // \t}\n\n// // \tfor _, tt := range tests {\n// // \t\tt.Run(tt.name, func(t *testing.T) {\n// // \t\t\tagent := &DSL{\n// // \t\t\t\tConnector: tt.connector,\n// // \t\t\t}\n// // \t\t\tagent.newConversation()\n\n// // \t\t\tassert.Panics(t, func() {\n// // \t\t\t\tagent.newAI()\n// // \t\t\t})\n\n// // \t\t})\n// // \t}\n// // }\n\n// func TestDSL_Select(t *testing.T) {\n// \ttest.Prepare(t, config.Conf)\n// \tdefer Test_clean(t)\n\n// \tresetDB()\n// \tagent := &DSL{\n// \t\tConversationSetting: conversation.Setting{\n// \t\t\tConnector: \"default\",\n// \t\t\tTable: \"chat_messages\",\n// \t\t},\n// \t}\n\n// \terr := agent.newConversation()\n// \tassert.NoError(t, err)\n\n// \terr = agent.Select(\"invalid-model\")\n// \tassert.Error(t, err)\n\n// \t// err = agent.Select(\"gpt-3_5-turbo\")\n// \t// assert.NoError(t, err)\n// \t// assert.NotNil(t, agent.AI)\n\n// }\n\n// // func TestDSL_NewConversation(t *testing.T) {\n// // \ttest.Prepare(t, config.Conf)\n// // \tdefer Test_clean(t)\n\n// // \ttests := []struct {\n// // \t\tname string\n// // \t\tconnector string\n// // \t\twantErr bool\n// // \t}{\n// // \t\t{\n// // \t\t\tname: \"Default connector\",\n// // \t\t\tconnector: \"default\",\n// // \t\t\twantErr: false,\n// // \t\t},\n// // \t\t{\n// // \t\t\tname: \"Empty connector\",\n// // \t\t\tconnector: \"\",\n// // \t\t\twantErr: false,\n// // \t\t},\n// // \t\t{\n// // \t\t\tname: \"Invalid connector\",\n// // \t\t\tconnector: \"invalid-connector\",\n// // \t\t\twantErr: true,\n// // \t\t},\n// // \t}\n\n// // \tfor _, tt := range tests {\n// // \t\tt.Run(tt.name, func(t *testing.T) {\n// // \t\t\tagent := &DSL{\n// // \t\t\t\tConversationSetting: conversation.Setting{\n// // \t\t\t\t\tConnector: tt.connector,\n// // \t\t\t\t},\n// // \t\t\t}\n// // \t\t\tassert.Panics(t, func() {\n// // \t\t\t\tagent.newConversation()\n// // \t\t\t})\n// // \t\t})\n// // \t}\n// // }\n\n// func TestDSL_SaveHistory(t *testing.T) {\n// \ttest.Prepare(t, config.Conf)\n// \tdefer Test_clean(t)\n\n// \tagent := &DSL{\n// \t\tConversationSetting: conversation.Setting{\n// \t\t\tConnector: \"default\",\n// \t\t\tTable: \"chat_messages\",\n// \t\t},\n// \t}\n\n// \tresetDB()\n// \terr := agent.newConversation()\n// \tassert.NoError(t, err)\n\n// \tmessages := []map[string]interface{}{\n// \t\t{\n// \t\t\t\"role\": \"user\",\n// \t\t\t\"content\": \"Hello\",\n// \t\t\t\"name\": \"test-user\",\n// \t\t},\n// \t}\n\n// \tcontent := []byte(\"Hi there!\")\n// \tagent.saveHistory(\"test-session\", \"test-chat\", content, messages)\n\n// \t// Verify the history was saved\n// \thistory, err := agent.Conversation.GetHistory(\"test-session\", \"test-chat\")\n// \tassert.NoError(t, err)\n// \tassert.NotEmpty(t, history)\n// }\n\n// func TestDSL_Send(t *testing.T) {\n// \ttest.Prepare(t, config.Conf)\n// \tdefer Test_clean(t)\n\n// \tgin.SetMode(gin.TestMode)\n// \tw := httptest.NewRecorder()\n// \tc, _ := gin.CreateTestContext(w)\n\n// \tresetDB()\n// \tagent := &DSL{\n// \t\tConversationSetting: conversation.Setting{\n// \t\t\tConnector: \"default\",\n// \t\t\tTable: \"chat_messages\",\n// \t\t},\n// \t}\n\n// \terr := agent.newConversation()\n// \tassert.NoError(t, err)\n// \tctx := Context{\n// \t\tSid: \"test-session\",\n// \t\tChatID: \"test-chat\",\n// \t}\n\n// \tmsg := &message.JSON{\n// \t\tMessage: &message.Message{Text: \"Test message\"},\n// \t}\n// \tmessages := []map[string]interface{}{\n// \t\t{\"role\": \"user\", \"content\": \"Hello\"},\n// \t}\n// \tcontent := []byte(\"Test content\")\n\n// \terr = agent.send(ctx, msg, messages, content, c)\n// \tassert.NoError(t, err)\n// }\n\n// func Test_clean(t *testing.T) {\n// \tdefer test.Clean()\n\n// }\n\n// func resetDB() {\n// \tsch := capsule.Global.Schema()\n// \tsch.DropTable(\"chat_messages\")\n// }\n\n// type mockAI struct{}\n\n// func (m *mockAI) ChatCompletionsWith(ctx context.Context, messages []map[string]interface{}, options map[string]interface{}, callback func([]byte) int) (interface{}, *exception.Exception) {\n// \tcallback([]byte(`{\"choices\":[{\"delta\":{\"content\":\"Mock response\"}}]}`))\n// \tcallback([]byte(`{\"choices\":[{\"finish_reason\":\"stop\"}]}`))\n// \treturn nil, nil\n// }\n\n// func (m *mockAI) ChatCompletions(messages []map[string]interface{}, options map[string]interface{}, callback func([]byte) int) (interface{}, *exception.Exception) {\n// \treturn nil, nil\n// }\n\n// func (m *mockAI) GetContent(response interface{}) (string, *exception.Exception) {\n// \treturn \"Mock content\", nil\n// }\n\n// func (m *mockAI) Embeddings(input interface{}, user string) (interface{}, *exception.Exception) {\n// \treturn nil, nil\n// }\n\n// func (m *mockAI) Tiktoken(input string) (int, error) {\n// \treturn 0, nil\n// }\n\n// func (m *mockAI) MaxToken() int {\n// \treturn 4096\n// }\n" }, { "path": "agent/assistant/agent.go", "content": "package assistant\n\nimport (\n\t\"fmt\"\n\t\"log\"\n\t\"time\"\n\n\tjsoniter \"github.com/json-iterator/go\"\n\t\"github.com/yaoapp/gou/connector\"\n\tgoullm \"github.com/yaoapp/gou/llm\"\n\t\"github.com/yaoapp/yao/agent/assistant/handlers\"\n\t\"github.com/yaoapp/yao/agent/context\"\n\t\"github.com/yaoapp/yao/agent/i18n\"\n\t\"github.com/yaoapp/yao/agent/llm\"\n\t\"github.com/yaoapp/yao/agent/output/message\"\n\tagentsandbox \"github.com/yaoapp/yao/agent/sandbox\"\n\tsandboxTypes \"github.com/yaoapp/yao/agent/sandbox/v2/types\"\n\tinfraV2 \"github.com/yaoapp/yao/sandbox/v2\"\n)\n\n// Stream stream the agent\n// handler is optional, if not provided, a default handler will be used\nfunc (ast *Assistant) Stream(ctx *context.Context, inputMessages []context.Message, options ...*context.Options) (*context.Response, error) {\n\n\t// Update logger with assistant ID and start logging\n\tctx.Logger.SetAssistantID(ast.ID)\n\tctx.Logger.Start()\n\n\t// Validate user permissions\n\tvar err error\n\terr = ast.checkPermissions(ctx)\n\tif err != nil {\n\t\treturn nil, err\n\t}\n\n\t// Start stream time\n\tstreamStartTime := time.Now()\n\n\t// Set up interrupt handler if interrupt controller is available\n\t// InterruptController handles user interrupt signals (stop button) for appending messages\n\t// HTTP context cancellation is handled naturally by LLM/Agent layers\n\tif ctx.Interrupt != nil {\n\t\tctx.Interrupt.SetHandler(func(c *context.Context, signal *context.InterruptSignal) error {\n\t\t\treturn ast.handleInterrupt(c, signal)\n\t\t})\n\t}\n\n\t// ================================================\n\t// Initialize\n\t// ================================================\n\tctx.Logger.Phase(\"Initialize\")\n\n\t// Get or create options\n\tvar opts *context.Options\n\tif len(options) > 0 && options[0] != nil {\n\t\topts = options[0]\n\t} else {\n\t\topts = &context.Options{}\n\t}\n\n\t// Merge caller-provided metadata into ctx so sub-agent hooks can read it via ctx.metadata\n\tctx.MergeMetadata(opts.Metadata)\n\n\t// Initialize stack and auto-handle completion/failure/restore\n\t_, _, done := context.EnterStack(ctx, ast.ID, opts)\n\tdefer done()\n\n\t// Auto-skip history for forked Agent-to-Agent calls (ctx.agent.Call/All/Any/Race)\n\t// This ensures forked A2A messages don't pollute chat history.\n\t// Delegate calls (RefererAgent) still save history as they are part of the main conversation flow.\n\t// Note: Output is NOT skipped - sub-agents output normally with ThreadID for UI separation.\n\tif ctx.IsForkedA2ACall() {\n\t\tif opts == nil {\n\t\t\topts = &context.Options{}\n\t\t}\n\t\topts.ForceA2A()\n\t}\n\n\t// ================================================\n\t// Initialize Chat Buffer (for root stack only)\n\t// Buffer is flushed in defer block at the end\n\t// ================================================\n\tast.InitBuffer(ctx)\n\n\t// Track final status for buffer flush\n\tvar finalStatus = context.StepStatusCompleted\n\tvar finalError error\n\n\t// Defer buffer flush - always executes on exit (success, error, interrupt, panic)\n\tdefer func() {\n\t\t// Handle panic recovery for status tracking\n\t\tif r := recover(); r != nil {\n\t\t\tfinalStatus = context.ResumeStatusFailed\n\t\t\tif e, ok := r.(error); ok {\n\t\t\t\tfinalError = e\n\t\t\t} else {\n\t\t\t\tfinalError = fmt.Errorf(\"panic: %v\", r)\n\t\t\t}\n\t\t\tctx.Logger.Error(\"Panic recovered in Stream: %v\", r)\n\t\t\t// Re-panic after flush to preserve original behavior\n\t\t\tdefer panic(r)\n\t\t}\n\n\t\t// Flush buffer to database\n\t\tast.FlushBuffer(ctx, finalStatus, finalError)\n\n\t\t// Log end of request\n\t\tctx.Logger.End(finalStatus == context.StepStatusCompleted, finalError)\n\t\tctx.Logger.RestoreAssistantID()\n\t}()\n\n\t// Determine stream handler\n\tstreamHandler := ast.getStreamHandler(ctx, opts)\n\n\t// Get connector and capabilities early (before sending stream_start)\n\t// so that output adapters can use them when converting stream_start event\n\terr = ast.initializeCapabilities(ctx, opts)\n\tif err != nil {\n\t\tfinalStatus = context.ResumeStatusFailed\n\t\tfinalError = err\n\t\tast.sendStreamEndOnError(ctx, streamHandler, streamStartTime, err)\n\t\treturn nil, err\n\t}\n\n\t// Send ChunkStreamStart only for root stack (agent-level stream start)\n\t// Now ctx.Capabilities is set, so output adapters can use it\n\tast.sendAgentStreamStart(ctx, streamHandler, streamStartTime)\n\n\t// Initialize chat, prepare kb collection (optional) etc.\n\t// Use async version to not block the main flow\n\tast.InitializeConversationAsync(ctx, opts)\n\n\tctx.Logger.PhaseComplete(\"Initialize\")\n\n\t// Ensure chat session exists\n\tast.EnsureChat(ctx)\n\n\t// Initialize agent trace node\n\tagentNode := ast.initAgentTraceNode(ctx, inputMessages)\n\n\t// ================================================\n\t// Get Full Messages with chat history\n\t// ================================================\n\tctx.Logger.Phase(\"History\")\n\thistoryResult, err := ast.WithHistory(ctx, inputMessages, agentNode, opts)\n\tif err != nil {\n\t\tast.traceAgentFail(agentNode, err)\n\t\tast.sendStreamEndOnError(ctx, streamHandler, streamStartTime, err)\n\t\treturn nil, err\n\t}\n\tfullMessages := historyResult.FullMessages\n\n\t// Buffer user input messages (use cleaned input without overlap)\n\t// Skip if History is disabled in options (for internal calls like needsearch)\n\t// Note: For A2A calls, ForceA2A() sets skip.history = true, so this will be skipped\n\tif opts == nil || opts.Skip == nil || !opts.Skip.History {\n\t\tast.BufferUserInput(ctx, historyResult.InputMessages)\n\t}\n\tctx.Logger.PhaseComplete(\"History\")\n\n\t// ================================================\n\t// Initialize Sandbox (if configured)\n\t// ================================================\n\t// Sandbox must be created BEFORE hooks so that hooks can access ctx.sandbox\n\tvar sandboxExecutor agentsandbox.Executor\n\tvar sandboxCleanup func()\n\tvar sandboxLoadingMsgID string\n\n\t// V2 sandbox state\n\tvar v2Runner sandboxTypes.Runner\n\tvar v2Computer infraV2.Computer\n\tvar v2LoadingMsgID string\n\n\tif ast.HasSandboxV2() {\n\t\tctx.Logger.Phase(\"Sandbox V2\")\n\t\tvar err error\n\t\tvar v2Cleanup func()\n\t\tv2Runner, v2Computer, v2Cleanup, v2LoadingMsgID, err = ast.initSandboxV2(ctx, opts)\n\t\tif err != nil {\n\t\t\tast.traceAgentFail(agentNode, err)\n\t\t\tast.sendStreamEndOnError(ctx, streamHandler, streamStartTime, err)\n\t\t\treturn nil, err\n\t\t}\n\t\tsandboxCleanup = v2Cleanup\n\t\tctx.Logger.PhaseComplete(\"Sandbox V2\")\n\t} else if ast.HasSandbox() {\n\t\tctx.Logger.Phase(\"Sandbox\")\n\t\tvar err error\n\t\tsandboxExecutor, sandboxCleanup, sandboxLoadingMsgID, err = ast.initSandbox(ctx, opts)\n\t\tif err != nil {\n\t\t\tast.traceAgentFail(agentNode, err)\n\t\t\tast.sendStreamEndOnError(ctx, streamHandler, streamStartTime, err)\n\t\t\treturn nil, err\n\t\t}\n\t\t// Set sandbox executor in context so hooks can access ctx.sandbox\n\t\t// The executor implements both agentsandbox.Executor and context.SandboxExecutor\n\t\tctx.SetSandboxExecutor(sandboxExecutor)\n\t\tctx.Logger.PhaseComplete(\"Sandbox\")\n\t}\n\t// Ensure sandbox cleanup on exit\n\tdefer func() {\n\t\tif sandboxCleanup != nil {\n\t\t\tsandboxCleanup()\n\t\t}\n\t}()\n\n\t// ================================================\n\t// Standalone Workspace Loading (no sandbox required)\n\t// ================================================\n\t// When no sandbox is configured but the user selected a workspace,\n\t// load the workspace FS into context so hooks can access ctx.workspace.\n\tif !ctx.HasWorkspace() {\n\t\tast.initStandaloneWorkspace(ctx)\n\t}\n\n\t// ================================================\n\t// Execute Create Hook\n\t// ================================================\n\t// Request Create hook ( Optional )\n\tvar createResponse *context.HookCreateResponse\n\tif ast.HookScript != nil {\n\t\tctx.Logger.HookStart(\"Create\")\n\t\t// Begin step tracking for hook_create\n\t\tast.BeginStep(ctx, context.StepTypeHookCreate, map[string]interface{}{\n\t\t\t\"messages\": fullMessages,\n\t\t})\n\n\t\tvar err error\n\t\tcreateResponse, opts, err = ast.HookScript.Create(ctx, fullMessages, opts)\n\t\tif err != nil {\n\t\t\tfinalStatus = context.ResumeStatusFailed\n\t\t\tfinalError = err\n\t\t\tast.traceAgentFail(agentNode, err)\n\t\t\t// Send error stream_end for root stack\n\t\t\tast.sendStreamEndOnError(ctx, streamHandler, streamStartTime, err)\n\t\t\treturn nil, err\n\t\t}\n\n\t\t// Complete step\n\t\tast.CompleteStep(ctx, map[string]interface{}{\n\t\t\t\"response\": createResponse,\n\t\t})\n\n\t\t// Log the create response\n\t\tast.traceCreateHook(agentNode, createResponse)\n\t\tctx.Logger.HookComplete(\"Create\")\n\n\t\t// Check if Create hook wants to delegate to another agent\n\t\t// This allows early routing to sub-agents without LLM call\n\t\tif createResponse != nil && createResponse.Delegate != nil {\n\t\t\tctx.Logger.Debug(\"Create hook delegating to agent: %s\", createResponse.Delegate.AgentID)\n\n\t\t\t// Delegate to target agent (reuse existing delegation logic from next.go)\n\t\t\t// Note: User input is already buffered by root agent, delegated agent will skip buffering\n\t\t\tdelegateResponse, err := ast.handleDelegation(ctx, createResponse.Delegate, streamHandler)\n\t\t\tif err != nil {\n\t\t\t\tfinalStatus = context.ResumeStatusFailed\n\t\t\t\tfinalError = err\n\t\t\t\tast.traceAgentFail(agentNode, err)\n\t\t\t\tast.sendStreamEndOnError(ctx, streamHandler, streamStartTime, err)\n\t\t\t\treturn nil, err\n\t\t\t}\n\n\t\t\t// For root stack, send stream_end and close output\n\t\t\t// (delegated agent handles its own stream events, but root needs to close)\n\t\t\tif ctx.Stack != nil && ctx.Stack.IsRoot() {\n\t\t\t\tast.sendAgentStreamEnd(ctx, streamHandler, streamStartTime, \"completed\", nil, nil)\n\t\t\t\tif err := ctx.CloseOutput(); err != nil {\n\t\t\t\t\tif trace, _ := ctx.Trace(); trace != nil {\n\t\t\t\t\t\ttrace.Error(i18n.Tr(ast.ID, ctx.Locale, \"assistant.agent.stream.close_error\"), map[string]any{\"error\": err.Error()})\n\t\t\t\t\t}\n\t\t\t\t}\n\t\t\t}\n\n\t\t\t// Return delegated response directly (skip LLM call and Next hook)\n\t\t\treturn delegateResponse, nil\n\t\t}\n\t}\n\n\t// ================================================\n\t// Execute LLM Call Stream\n\t// ================================================\n\t// LLM Call Stream ( Optional )\n\tvar completionResponse *context.CompletionResponse\n\tvar completionMessages []context.Message\n\tvar completionOptions *context.CompletionOptions\n\tif ast.Prompts != nil || ast.MCP != nil {\n\t\tctx.Logger.Phase(\"LLM\")\n\n\t\t// Build the LLM request first (use fullMessages which includes history)\n\t\t// Note: completionMessages here are still in original format (with __yao.attachment:// URLs)\n\t\t// Content conversion (BuildContent) happens inside executeLLMStream, right before LLM call\n\t\t// This ensures autoSearch and delegate receive original messages, not converted ones\n\t\tcompletionMessages, completionOptions, err = ast.BuildRequest(ctx, fullMessages, createResponse)\n\t\tif err != nil {\n\t\t\tfinalStatus = context.ResumeStatusFailed\n\t\t\tfinalError = err\n\t\t\tast.traceAgentFail(agentNode, err)\n\t\t\t// Send error stream_end for root stack\n\t\t\tast.sendStreamEndOnError(ctx, streamHandler, streamStartTime, err)\n\t\t\treturn nil, err\n\t\t}\n\n\t\t// ================================================\n\t\t// Execute Auto Search (if enabled)\n\t\t// ================================================\n\t\tif intent := ast.shouldAutoSearch(ctx, completionMessages, createResponse, opts); intent != nil {\n\t\t\trefCtx := ast.executeAutoSearch(ctx, completionMessages, createResponse, intent, opts)\n\t\t\tif refCtx != nil && len(refCtx.References) > 0 {\n\t\t\t\tcompletionMessages = ast.injectSearchContext(completionMessages, refCtx)\n\t\t\t}\n\t\t}\n\n\t\t// Begin step tracking for LLM call\n\t\tast.BeginStep(ctx, context.StepTypeLLM, map[string]interface{}{\n\t\t\t\"messages\": completionMessages,\n\t\t})\n\n\t\t// Execute the LLM streaming call\n\t\t// Choose between sandbox execution or direct LLM execution\n\t\tif ast.HasSandboxV2() && v2Runner != nil && v2Computer != nil && v2Runner.Name() != \"yao\" {\n\t\t\t// V2 Sandbox execution path (non-yao runners replace LLM.Stream)\n\t\t\tcompletionResponse, err = ast.executeSandboxV2Stream(ctx, completionMessages, agentNode, streamHandler, v2Runner, v2Computer, v2LoadingMsgID)\n\t\t} else if ast.HasSandboxV2() && v2Runner != nil && v2Runner.Name() == \"yao\" {\n\t\t\t// V2 yao runner: Prepare is done, close loading, fall through to LLM\n\t\t\tif v2LoadingMsgID != \"\" {\n\t\t\t\tcloseLoadingV2(ctx, v2LoadingMsgID, \"\")\n\t\t\t}\n\t\t\tcompletionResponse, err = ast.executeLLMStream(ctx, completionMessages, completionOptions, agentNode, streamHandler, opts)\n\t\t} else if ast.HasSandbox() {\n\t\t\t// V1 Sandbox execution path (Claude CLI, Cursor CLI, etc.)\n\t\t\tcompletionResponse, err = ast.executeSandboxStream(ctx, completionMessages, agentNode, streamHandler, sandboxExecutor, sandboxLoadingMsgID)\n\t\t} else {\n\t\t\t// Direct LLM execution path\n\t\t\tcompletionResponse, err = ast.executeLLMStream(ctx, completionMessages, completionOptions, agentNode, streamHandler, opts)\n\t\t}\n\t\tif err != nil {\n\t\t\tfinalStatus = context.ResumeStatusFailed\n\t\t\tfinalError = err\n\t\t\tast.traceAgentFail(agentNode, err)\n\t\t\t// Send error stream_end for root stack\n\t\t\tast.sendStreamEndOnError(ctx, streamHandler, streamStartTime, err)\n\t\t\treturn nil, err\n\t\t}\n\n\t\t// Complete LLM step\n\t\tast.CompleteStep(ctx, map[string]interface{}{\n\t\t\t\"content\": completionResponse.Content,\n\t\t\t\"tool_calls\": completionResponse.ToolCalls,\n\t\t})\n\n\t\thasToolCalls := completionResponse != nil && completionResponse.ToolCalls != nil && len(completionResponse.ToolCalls) > 0\n\t\ttokens := 0\n\t\tif completionResponse != nil && completionResponse.Usage != nil {\n\t\t\ttokens = completionResponse.Usage.TotalTokens\n\t\t}\n\t\tctx.Logger.LLMComplete(tokens, hasToolCalls)\n\t\tctx.Logger.PhaseComplete(\"LLM\")\n\t}\n\n\t// ================================================\n\t// Execute tool calls with retry\n\t// ================================================\n\t// Note: Skip MCP tool calls execution for sandbox mode - Claude CLI handles them internally\n\tvar toolCallResponses []context.ToolCallResponse = nil\n\tif completionResponse != nil && completionResponse.ToolCalls != nil && !ast.HasSandbox() {\n\n\t\tmaxToolRetries := 3\n\t\tcurrentMessages := completionMessages\n\t\tcurrentResponse := completionResponse\n\n\t\tfor attempt := 0; attempt < maxToolRetries; attempt++ {\n\n\t\t\t// Begin step tracking for tool calls\n\t\t\tast.BeginStep(ctx, context.StepTypeTool, map[string]interface{}{\n\t\t\t\t\"tool_calls\": currentResponse.ToolCalls,\n\t\t\t\t\"attempt\": attempt,\n\t\t\t})\n\n\t\t\t// Execute all tool calls\n\t\t\ttoolResults, hasErrors := ast.executeToolCalls(ctx, currentResponse.ToolCalls, attempt)\n\n\t\t\t// Build a map of tool call ID to arguments for quick lookup\n\t\t\ttoolCallArgsMap := make(map[string]interface{})\n\t\t\tfor _, tc := range currentResponse.ToolCalls {\n\t\t\t\ttoolCallArgsMap[tc.ID] = tc.Function.Arguments\n\t\t\t}\n\n\t\t\t// Convert toolResults to toolCallResponses\n\t\t\ttoolCallResponses = make([]context.ToolCallResponse, len(toolResults))\n\t\t\tfor i, result := range toolResults {\n\t\t\t\tparsedContent, _ := result.ParsedContent()\n\t\t\t\ttoolCallResponses[i] = context.ToolCallResponse{\n\t\t\t\t\tToolCallID: result.ToolCallID,\n\t\t\t\t\tServer: result.Server(),\n\t\t\t\t\tTool: result.Tool(),\n\t\t\t\t\tArguments: toolCallArgsMap[result.ToolCallID],\n\t\t\t\t\tResult: parsedContent,\n\t\t\t\t\tError: \"\",\n\t\t\t\t}\n\t\t\t\tif result.Error != nil {\n\t\t\t\t\ttoolCallResponses[i].Error = result.Error.Error()\n\t\t\t\t}\n\t\t\t}\n\n\t\t\t// If all successful, complete step and break out\n\t\t\tif !hasErrors {\n\t\t\t\tast.CompleteStep(ctx, map[string]interface{}{\n\t\t\t\t\t\"results\": toolCallResponses,\n\t\t\t\t})\n\t\t\t\tctx.Logger.Debug(\"All tool calls succeeded (attempt %d)\", attempt)\n\t\t\t\tbreak\n\t\t\t}\n\n\t\t\t// Check if any errors are retryable (parameter/validation issues)\n\t\t\thasRetryableErrors := false\n\t\t\tfor _, result := range toolResults {\n\t\t\t\tif result.Error != nil && result.IsRetryableError {\n\t\t\t\t\thasRetryableErrors = true\n\t\t\t\t\tbreak\n\t\t\t\t}\n\t\t\t}\n\n\t\t\t// If no retryable errors, don't retry (MCP internal issues)\n\t\t\tif !hasRetryableErrors {\n\t\t\t\terr := fmt.Errorf(\"tool calls failed with non-retryable errors (MCP internal issues)\")\n\t\t\t\tfinalStatus = context.ResumeStatusFailed\n\t\t\t\tfinalError = err\n\t\t\t\tctx.Logger.Error(\"Tool calls failed: %v\", err)\n\t\t\t\tast.traceAgentFail(agentNode, err)\n\t\t\t\tast.sendStreamEndOnError(ctx, streamHandler, streamStartTime, err)\n\t\t\t\treturn nil, err\n\t\t\t}\n\n\t\t\t// If it's the last attempt, return error\n\t\t\tif attempt == maxToolRetries-1 {\n\t\t\t\terr := fmt.Errorf(\"tool calls failed after %d attempts\", maxToolRetries)\n\t\t\t\tfinalStatus = context.ResumeStatusFailed\n\t\t\t\tfinalError = err\n\t\t\t\tctx.Logger.Error(\"Tool calls failed: %v\", err)\n\t\t\t\tast.traceAgentFail(agentNode, err)\n\t\t\t\tast.sendStreamEndOnError(ctx, streamHandler, streamStartTime, err)\n\t\t\t\treturn nil, err\n\t\t\t}\n\n\t\t\t// Complete current step (with partial results)\n\t\t\tast.CompleteStep(ctx, map[string]interface{}{\n\t\t\t\t\"results\": toolCallResponses,\n\t\t\t\t\"has_errors\": true,\n\t\t\t})\n\n\t\t\t// Build retry messages with tool call results (including errors)\n\t\t\tretryMessages := ast.buildToolRetryMessages(currentMessages, currentResponse, toolResults)\n\n\t\t\t// Begin LLM retry step\n\t\t\tast.BeginStep(ctx, context.StepTypeLLM, map[string]interface{}{\n\t\t\t\t\"messages\": retryMessages,\n\t\t\t\t\"retry_attempt\": attempt + 1,\n\t\t\t})\n\n\t\t\t// Retry LLM call (streaming to keep user informed)\n\t\t\tctx.Logger.Debug(\"Retrying LLM for tool call correction (attempt %d/%d)\", attempt+1, maxToolRetries-1)\n\t\t\tcurrentResponse, err = ast.executeLLMForToolRetry(ctx, retryMessages, completionOptions, agentNode, streamHandler, opts)\n\t\t\tif err != nil {\n\t\t\t\tfinalStatus = context.ResumeStatusFailed\n\t\t\t\tfinalError = err\n\t\t\t\tctx.Logger.Error(\"LLM retry failed: %v\", err)\n\t\t\t\tast.traceAgentFail(agentNode, err)\n\t\t\t\tast.sendStreamEndOnError(ctx, streamHandler, streamStartTime, err)\n\t\t\t\treturn nil, err\n\t\t\t}\n\n\t\t\t// If LLM didn't return tool calls, it might have given up\n\t\t\tif currentResponse.ToolCalls == nil {\n\t\t\t\terr := fmt.Errorf(\"LLM did not return tool calls in retry attempt %d\", attempt+1)\n\t\t\t\tfinalStatus = context.ResumeStatusFailed\n\t\t\t\tfinalError = err\n\t\t\t\tctx.Logger.Error(\"LLM did not return tool calls: %v\", err)\n\t\t\t\tast.traceAgentFail(agentNode, err)\n\t\t\t\tast.sendStreamEndOnError(ctx, streamHandler, streamStartTime, err)\n\t\t\t\treturn nil, err\n\t\t\t}\n\n\t\t\t// Complete LLM retry step\n\t\t\tast.CompleteStep(ctx, map[string]interface{}{\n\t\t\t\t\"content\": currentResponse.Content,\n\t\t\t\t\"tool_calls\": currentResponse.ToolCalls,\n\t\t\t})\n\n\t\t\t// Update messages for next iteration\n\t\t\tcurrentMessages = retryMessages\n\t\t}\n\n\t\t// Update completionResponse with the final successful response\n\t\tcompletionResponse = currentResponse\n\t}\n\n\t// ================================================\n\t// Execute Next Hook and Process Response\n\t// ================================================\n\tvar finalResponse *context.Response\n\tvar nextResponse *context.NextHookResponse = nil\n\n\tif ast.HookScript != nil {\n\t\tctx.Logger.HookStart(\"Next\")\n\n\t\t// Begin step tracking for hook_next\n\t\tast.BeginStep(ctx, context.StepTypeHookNext, map[string]interface{}{\n\t\t\t\"messages\": fullMessages,\n\t\t\t\"completion\": completionResponse,\n\t\t\t\"tools\": toolCallResponses,\n\t\t})\n\n\t\tvar err error\n\t\tnextResponse, opts, err = ast.HookScript.Next(ctx, &context.NextHookPayload{\n\t\t\tMessages: fullMessages,\n\t\t\tCompletion: completionResponse,\n\t\t\tTools: toolCallResponses,\n\t\t}, opts)\n\t\tif err != nil {\n\t\t\tfinalStatus = context.ResumeStatusFailed\n\t\t\tfinalError = err\n\t\t\tast.traceAgentFail(agentNode, err)\n\t\t\tast.sendStreamEndOnError(ctx, streamHandler, streamStartTime, err)\n\t\t\treturn nil, err\n\t\t}\n\n\t\t// Complete hook_next step\n\t\tast.CompleteStep(ctx, map[string]interface{}{\n\t\t\t\"response\": nextResponse,\n\t\t})\n\n\t\tctx.Logger.HookComplete(\"Next\")\n\n\t\t// Process Next hook response\n\t\tfinalResponse, err = ast.processNextResponse(&NextProcessContext{\n\t\t\tContext: ctx,\n\t\t\tNextResponse: nextResponse,\n\t\t\tCompletionResponse: completionResponse,\n\t\t\tFullMessages: fullMessages,\n\t\t\tToolCallResponses: toolCallResponses,\n\t\t\tStreamHandler: streamHandler,\n\t\t\tCreateResponse: createResponse,\n\t\t})\n\t\tif err != nil {\n\t\t\tfinalStatus = context.ResumeStatusFailed\n\t\t\tfinalError = err\n\t\t\tast.traceAgentFail(agentNode, err)\n\t\t\tast.sendStreamEndOnError(ctx, streamHandler, streamStartTime, err)\n\t\t\treturn nil, err\n\t\t}\n\t} else {\n\t\t// No Next hook: use standard response\n\t\tfinalResponse = ast.buildStandardResponse(&NextProcessContext{\n\t\t\tContext: ctx,\n\t\t\tNextResponse: nil,\n\t\t\tCompletionResponse: completionResponse,\n\t\t\tFullMessages: fullMessages,\n\t\t\tToolCallResponses: toolCallResponses,\n\t\t\tStreamHandler: streamHandler,\n\t\t\tCreateResponse: createResponse,\n\t\t})\n\t}\n\n\t// Create completion node to report final output\n\tast.traceAgentCompletion(ctx, createResponse, nextResponse, completionResponse, finalResponse)\n\n\t// Only close output and send stream_end if this is the root call (entry point)\n\t// Nested calls (from MCP, hooks, etc.) should not close the output or send stream_end\n\t// Note: Flush is already handled by the stream handler (handleStreamEnd)\n\tif ctx.Stack != nil && ctx.Stack.IsRoot() {\n\t\t// Log closing output for root call\n\t\tif trace, _ := ctx.Trace(); trace != nil {\n\t\t\ttrace.Debug(\"Agent: Closing output (root call)\", map[string]any{\n\t\t\t\t\"stack_id\": ctx.Stack.ID,\n\t\t\t\t\"depth\": ctx.Stack.Depth,\n\t\t\t\t\"assistant_id\": ctx.Stack.AssistantID,\n\t\t\t})\n\t\t}\n\n\t\t// Send ChunkStreamEnd (agent-level stream completion)\n\t\tast.sendAgentStreamEnd(ctx, streamHandler, streamStartTime, \"completed\", nil, completionResponse)\n\n\t\t// Close the output writer to send [DONE] marker\n\t\tif err := ctx.CloseOutput(); err != nil {\n\t\t\tif trace, _ := ctx.Trace(); trace != nil {\n\t\t\t\ttrace.Error(i18n.Tr(ast.ID, ctx.Locale, \"assistant.agent.stream.close_error\"), map[string]any{\"error\": err.Error()}) // \"Failed to close output\"\n\t\t\t}\n\t\t}\n\t} else {\n\t\t// Log skipping close for nested call\n\t\tif trace, _ := ctx.Trace(); trace != nil && ctx.Stack != nil {\n\t\t\ttrace.Debug(\"Agent: Skipping output close (nested call)\", map[string]any{\n\t\t\t\t\"stack_id\": ctx.Stack.ID,\n\t\t\t\t\"depth\": ctx.Stack.Depth,\n\t\t\t\t\"parent_id\": ctx.Stack.ParentID,\n\t\t\t\t\"assistant_id\": ctx.Stack.AssistantID,\n\t\t\t})\n\t\t}\n\t}\n\n\t// Return finalResponse which could be:\n\t// 1. Result from delegated agent call (already a Response)\n\t// 2. Custom data from Next hook (wrapped in standard Response)\n\t// 3. Standard response\n\treturn finalResponse, nil\n}\n\n// GetConnector get the connector object, capabilities, and error with priority:\n// opts.Connector > ast.Connector > defaultConnector (fallback)\n// Note: opts.Connector may be set by Create hook's applyOptionsAdjustments\n// Returns: (connector, capabilities, error)\nfunc (ast *Assistant) GetConnector(ctx *context.Context, opts ...*context.Options) (connector.Connector, *goullm.Capabilities, error) {\n\tconnectorID := ast.Connector\n\tif len(opts) > 0 && opts[0] != nil && opts[0].Connector != \"\" {\n\t\tconnectorID = opts[0].Connector\n\t}\n\n\tif connectorID == \"\" {\n\t\tconnectorID = defaultConnector\n\t}\n\n\tif connectorID == \"\" {\n\t\treturn nil, nil, fmt.Errorf(\"connector not specified\")\n\t}\n\n\tconn, err := connector.Select(connectorID)\n\tif err != nil && connectorID != defaultConnector && defaultConnector != \"\" {\n\t\tlog.Printf(\"[Assistant] connector %q not found, falling back to default %q\", connectorID, defaultConnector)\n\t\tconn, err = connector.Select(defaultConnector)\n\t}\n\tif err != nil {\n\t\treturn nil, nil, err\n\t}\n\n\tcapabilities := llm.GetCapabilitiesFromConn(conn)\n\treturn conn, capabilities, nil\n}\n\n// Info get the assistant information\nfunc (ast *Assistant) Info(locale ...string) *message.AssistantInfo {\n\tlc := \"en\"\n\tif len(locale) > 0 {\n\t\tlc = locale[0]\n\t}\n\treturn &message.AssistantInfo{\n\t\tID: ast.ID,\n\t\tType: ast.Type,\n\t\tName: i18n.Tr(ast.ID, lc, ast.Name),\n\t\tAvatar: ast.Avatar,\n\t\tDescription: i18n.Tr(ast.ID, lc, ast.Description),\n\t}\n}\n\n// getStreamHandler returns the stream handler from options or a default one\nfunc (ast *Assistant) getStreamHandler(ctx *context.Context, opts ...*context.Options) message.StreamFunc {\n\t// Check if handler is provided in options\n\tif len(opts) > 0 && opts[0] != nil && opts[0].Writer != nil {\n\t\treturn handlers.DefaultStreamHandler(ctx)\n\t}\n\treturn handlers.DefaultStreamHandler(ctx)\n}\n\n// sendAgentStreamStart sends ChunkStreamStart for root stack only (agent-level stream start)\n// This ensures only one stream_start per agent execution, even with multiple LLM calls\nfunc (ast *Assistant) sendAgentStreamStart(ctx *context.Context, handler message.StreamFunc, startTime time.Time) {\n\tif ctx.Stack == nil || !ctx.Stack.IsRoot() || handler == nil {\n\t\treturn\n\t}\n\n\t// Build the start data\n\tstartData := message.EventStreamStartData{\n\t\tContextID: ctx.ID,\n\t\tChatID: ctx.ChatID,\n\t\tTraceID: ctx.TraceID(),\n\t\tRequestID: ctx.RequestID(),\n\t\tTimestamp: startTime.UnixMilli(),\n\t\tAssistant: ast.Info(ctx.Locale),\n\t\tMetadata: ctx.Metadata,\n\t}\n\n\tif startJSON, err := jsoniter.Marshal(startData); err == nil {\n\t\thandler(message.ChunkStreamStart, startJSON)\n\t}\n}\n\n// sendAgentStreamEnd sends ChunkStreamEnd for root stack only (agent-level stream completion)\nfunc (ast *Assistant) sendAgentStreamEnd(ctx *context.Context, handler message.StreamFunc, startTime time.Time, status string, err error, response *context.CompletionResponse) {\n\tif ctx.Stack == nil || !ctx.Stack.IsRoot() || handler == nil {\n\t\treturn\n\t}\n\n\t// Check if context is cancelled - if so, skip handler call to avoid blocking\n\tif ctx.Context != nil && ctx.Context.Err() != nil {\n\t\tctx.Logger.Debug(\"Context cancelled, skipping sendAgentStreamEnd handler call\")\n\t\treturn\n\t}\n\n\tendData := &message.EventStreamEndData{\n\t\tRequestID: ctx.RequestID(),\n\t\tContextID: ctx.ID,\n\t\tTimestamp: time.Now().UnixMilli(),\n\t\tDurationMs: time.Since(startTime).Milliseconds(),\n\t\tStatus: status,\n\t\tTraceID: ctx.TraceID(),\n\t\tMetadata: ctx.Metadata,\n\t}\n\n\tif err != nil {\n\t\tendData.Error = err.Error()\n\t}\n\n\tif response != nil && response.Usage != nil {\n\t\tendData.Usage = response.Usage\n\t}\n\n\tif endJSON, marshalErr := jsoniter.Marshal(endData); marshalErr == nil {\n\t\thandler(message.ChunkStreamEnd, endJSON)\n\t}\n}\n\n// sendStreamEndOnError sends ChunkStreamEnd with error status for root stack only\nfunc (ast *Assistant) sendStreamEndOnError(ctx *context.Context, handler message.StreamFunc, startTime time.Time, err error) {\n\tast.sendAgentStreamEnd(ctx, handler, startTime, \"error\", err, nil)\n}\n\n// handleInterrupt handles the interrupt signal\n// This is called by the interrupt listener when a signal is received\nfunc (ast *Assistant) handleInterrupt(ctx *context.Context, signal *context.InterruptSignal) error {\n\t// Handle based on interrupt type\n\tswitch signal.Type {\n\tcase context.InterruptForce:\n\t\t// Force interrupt: context is already cancelled in handleSignal\n\t\t// LLM streaming will detect ctx.Interrupt.Context().Done() and stop\n\t\tctx.Logger.Debug(\"Force interrupt: stopping current operations immediately\")\n\n\tcase context.InterruptGraceful:\n\t\tctx.Logger.Debug(\"Graceful interrupt: will process after current step completes\")\n\t\t// Graceful interrupt: let current operation complete\n\t\t// The signal is stored in current/pending, can be checked at checkpoints\n\t}\n\n\t// TODO: Implement actual interrupt handling logic:\n\t// 1. For graceful: wait for current step, then merge messages and restart\n\t// 2. For force: immediately stop and restart with new messages\n\t// 3. Call Interrupted Hook if configured\n\t// 4. Decide whether to continue, restart, or abort based on Hook response\n\n\treturn nil\n}\n\n// initializeCapabilities gets connector and capabilities, then sets them in context\n// This should be called early (before sending stream_start) so that output adapters\n// can use capabilities when converting stream_start event\nfunc (ast *Assistant) initializeCapabilities(ctx *context.Context, opts *context.Options) error {\n\tif ast.Prompts == nil && ast.MCP == nil {\n\t\treturn nil\n\t}\n\n\t_, capabilities, err := ast.GetConnector(ctx, opts)\n\tif err != nil {\n\t\treturn err\n\t}\n\n\t// Set capabilities in context for output adapters to use\n\tif capabilities != nil {\n\t\tctx.Capabilities = capabilities\n\t}\n\n\treturn nil\n}\n\n// buildToolRetryMessages builds messages for LLM retry with tool call results\n// Format follows OpenAI's tool call response pattern:\n// 1. Assistant message with tool calls\n// 2. Tool messages with results (one per tool call)\n// 3. System message explaining the retry\nfunc (ast *Assistant) buildToolRetryMessages(\n\tpreviousMessages []context.Message,\n\tcompletionResponse *context.CompletionResponse,\n\ttoolResults []ToolCallResult,\n) []context.Message {\n\tretryMessages := make([]context.Message, 0, len(previousMessages)+len(toolResults)+2)\n\n\t// Add all previous messages\n\tretryMessages = append(retryMessages, previousMessages...)\n\n\t// Add assistant message with tool calls\n\tassistantMsg := context.Message{\n\t\tRole: context.RoleAssistant,\n\t\tContent: completionResponse.Content,\n\t\tToolCalls: completionResponse.ToolCalls,\n\t}\n\tretryMessages = append(retryMessages, assistantMsg)\n\n\t// Add tool result messages (one per tool call)\n\tfor _, result := range toolResults {\n\t\ttoolMsg := context.Message{\n\t\t\tRole: context.RoleTool,\n\t\t\tContent: result.Content,\n\t\t\tToolCallID: &result.ToolCallID,\n\t\t}\n\t\t// Add tool name if available\n\t\tif result.Name != \"\" {\n\t\t\tname := result.Name\n\t\t\ttoolMsg.Name = &name\n\t\t}\n\t\tretryMessages = append(retryMessages, toolMsg)\n\t}\n\n\t// Add system message explaining the retry (optional, helps LLM understand context)\n\tsystemMsg := context.Message{\n\t\tRole: context.RoleSystem,\n\t\tContent: i18n.Tr(ast.ID, \"en\", \"assistant.agent.tool_retry_prompt\"),\n\t}\n\tretryMessages = append(retryMessages, systemMsg)\n\n\treturn retryMessages\n}\n" }, { "path": "agent/assistant/agent_interrupt_test.go", "content": "package assistant_test\n\nimport (\n\tstdContext \"context\"\n\t\"fmt\"\n\t\"testing\"\n\t\"time\"\n\n\t\"github.com/yaoapp/yao/agent/assistant\"\n\t\"github.com/yaoapp/yao/agent/context\"\n\t\"github.com/yaoapp/yao/agent/output/message\"\n\t\"github.com/yaoapp/yao/agent/testutils\"\n\t\"github.com/yaoapp/yao/openapi/oauth/types\"\n)\n\n// newTestContextWithInterrupt creates a Context with interrupt controller for testing\n// Returns the context and a cancel function that should be called before Release()\nfunc newTestContextWithInterrupt(chatID, assistantID string) (*context.Context, stdContext.CancelFunc) {\n\tauthorized := &types.AuthorizedInfo{\n\t\tSubject: \"test-user\",\n\t\tClientID: \"test-client-id\",\n\t\tUserID: \"test-user-123\",\n\t\tTeamID: \"test-team-456\",\n\t\tTenantID: \"test-tenant-789\",\n\t\tSessionID: \"test-session-id\",\n\t}\n\n\t// Use cancellable context to properly stop goroutines on timeout\n\tparentCtx, cancel := stdContext.WithCancel(stdContext.Background())\n\n\tctx := context.New(parentCtx, authorized, chatID)\n\tctx.ID = fmt.Sprintf(\"test_ctx_%d\", time.Now().UnixNano())\n\tctx.AssistantID = assistantID\n\tctx.Locale = \"en-us\"\n\tctx.Theme = \"light\"\n\tctx.Client = context.Client{\n\t\tType: \"web\",\n\t\tUserAgent: \"TestAgent/1.0\",\n\t\tIP: \"127.0.0.1\",\n\t}\n\tctx.Referer = context.RefererAPI\n\tctx.Accept = context.AcceptWebCUI\n\tctx.Route = \"/test/route\"\n\tctx.IDGenerator = message.NewIDGenerator() // Initialize context-scoped ID generator\n\tctx.Metadata = map[string]interface{}{\n\t\t\"test\": \"interrupt_test\",\n\t}\n\n\t// Initialize interrupt controller\n\tctx.Interrupt = context.NewInterruptController()\n\n\t// Register context globally\n\tif err := context.Register(ctx); err != nil {\n\t\tpanic(fmt.Sprintf(\"Failed to register context: %v\", err))\n\t}\n\n\t// Start interrupt listener\n\tctx.Interrupt.Start(ctx.ID)\n\n\treturn ctx, cancel\n}\n\n// TestAgentInterruptGraceful tests graceful interrupt during agent stream\nfunc TestAgentInterruptGraceful(t *testing.T) {\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\tagent, err := assistant.Get(\"tests.interrupt\")\n\tif err != nil {\n\t\tt.Skipf(\"Skipping test: assistant 'tests.interrupt' not found: %v\", err)\n\t\treturn\n\t}\n\n\tt.Run(\"GracefulInterruptDuringStream\", func(t *testing.T) {\n\t\t// Create context with interrupt support\n\t\tctx, cancel := newTestContextWithInterrupt(\"chat-interrupt-graceful\", \"tests.interrupt\")\n\t\tdefer func() {\n\t\t\tcancel() // Cancel context first to stop goroutines\n\t\t\ttime.Sleep(100 * time.Millisecond) // Wait for goroutines to exit\n\t\t\tctx.Release()\n\t\t}()\n\n\t\t// Track handler invocations\n\t\thandlerInvoked := false\n\t\tvar receivedSignal *context.InterruptSignal\n\n\t\t// Override the handler to track invocations\n\t\toriginalHandler := ctx.Interrupt\n\t\tctx.Interrupt.SetHandler(func(c *context.Context, signal *context.InterruptSignal) error {\n\t\t\thandlerInvoked = true\n\t\t\treceivedSignal = signal\n\t\t\tt.Logf(\"✓ Interrupt handler invoked: type=%s, messages=%d\", signal.Type, len(signal.Messages))\n\t\t\treturn nil\n\t\t})\n\n\t\tinputMessages := []context.Message{\n\t\t\t{Role: context.RoleUser, Content: \"Tell me a long story about artificial intelligence\"},\n\t\t}\n\n\t\t// Start streaming in a goroutine\n\t\tstreamDone := make(chan error, 1)\n\t\tgo func() {\n\t\t\t_, err := agent.Stream(ctx, inputMessages)\n\t\t\tstreamDone <- err\n\t\t}()\n\n\t\t// Wait a bit to ensure stream has started\n\t\ttime.Sleep(300 * time.Millisecond)\n\n\t\t// Send graceful interrupt signal\n\t\tsignal := &context.InterruptSignal{\n\t\t\tType: context.InterruptGraceful,\n\t\t\tMessages: []context.Message{\n\t\t\t\t{Role: context.RoleUser, Content: \"Actually, can you make it shorter?\"},\n\t\t\t},\n\t\t\tTimestamp: time.Now().UnixMilli(),\n\t\t}\n\n\t\terr = context.SendInterrupt(ctx.ID, signal)\n\t\tif err != nil {\n\t\t\tt.Logf(\"Warning: Failed to send interrupt (stream may have completed): %v\", err)\n\t\t} else {\n\t\t\tt.Log(\"✓ Graceful interrupt signal sent\")\n\t\t}\n\n\t\t// Wait for stream to complete (with timeout)\n\t\tselect {\n\t\tcase err := <-streamDone:\n\t\t\tif err != nil {\n\t\t\t\tt.Logf(\"Stream completed with error: %v\", err)\n\t\t\t} else {\n\t\t\t\tt.Log(\"✓ Stream completed successfully\")\n\t\t\t}\n\t\tcase <-time.After(10 * time.Second):\n\t\t\tt.Log(\"Stream timeout (expected for real LLM calls)\")\n\t\t\tcancel() // Cancel to stop the stream goroutine\n\t\t\t<-streamDone // Wait for goroutine to exit\n\t\t}\n\n\t\t// Verify handler was invoked if signal was sent\n\t\tif originalHandler != nil {\n\t\t\ttime.Sleep(200 * time.Millisecond) // Wait for async handler\n\t\t\tif handlerInvoked {\n\t\t\t\tt.Log(\"✓ Interrupt handler was invoked\")\n\t\t\t\tif receivedSignal != nil && receivedSignal.Type == context.InterruptGraceful {\n\t\t\t\t\tt.Log(\"✓ Received graceful interrupt signal\")\n\t\t\t\t}\n\t\t\t}\n\t\t}\n\t})\n}\n\n// TestAgentInterruptForce tests force interrupt during agent stream\nfunc TestAgentInterruptForce(t *testing.T) {\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\tagent, err := assistant.Get(\"tests.interrupt\")\n\tif err != nil {\n\t\tt.Skipf(\"Skipping test: assistant 'tests.interrupt' not found: %v\", err)\n\t\treturn\n\t}\n\n\tt.Run(\"ForceInterruptDuringStream\", func(t *testing.T) {\n\t\t// Create context with interrupt support\n\t\tctx, cancel := newTestContextWithInterrupt(\"chat-interrupt-force\", \"tests.interrupt\")\n\t\tdefer func() {\n\t\t\tcancel() // Cancel context first to stop goroutines\n\t\t\ttime.Sleep(100 * time.Millisecond) // Wait for goroutines to exit\n\t\t\tctx.Release()\n\t\t}()\n\n\t\t// Track handler invocations\n\t\thandlerInvoked := false\n\t\tstreamInterrupted := false\n\n\t\tctx.Interrupt.SetHandler(func(c *context.Context, signal *context.InterruptSignal) error {\n\t\t\thandlerInvoked = true\n\t\t\tt.Logf(\"✓ Interrupt handler invoked: type=%s\", signal.Type)\n\t\t\treturn nil\n\t\t})\n\n\t\tinputMessages := []context.Message{\n\t\t\t{Role: context.RoleUser, Content: \"Write a very detailed essay about machine learning\"},\n\t\t}\n\n\t\t// Start streaming in a goroutine\n\t\tstreamDone := make(chan error, 1)\n\t\tgo func() {\n\t\t\t_, err := agent.Stream(ctx, inputMessages)\n\t\t\tstreamDone <- err\n\t\t}()\n\n\t\t// Wait a bit to ensure stream has started\n\t\ttime.Sleep(300 * time.Millisecond)\n\n\t\t// Send force interrupt signal\n\t\tsignal := &context.InterruptSignal{\n\t\t\tType: context.InterruptForce,\n\t\t\tMessages: []context.Message{\n\t\t\t\t{Role: context.RoleUser, Content: \"Stop! I need something else now.\"},\n\t\t\t},\n\t\t\tTimestamp: time.Now().UnixMilli(),\n\t\t}\n\n\t\terr = context.SendInterrupt(ctx.ID, signal)\n\t\tif err != nil {\n\t\t\tt.Logf(\"Warning: Failed to send interrupt: %v\", err)\n\t\t} else {\n\t\t\tt.Log(\"✓ Force interrupt signal sent\")\n\t\t}\n\n\t\t// Wait for stream to complete or be interrupted\n\t\tselect {\n\t\tcase err := <-streamDone:\n\t\t\tif err != nil {\n\t\t\t\t// Check if error is due to interrupt\n\t\t\t\tif err.Error() == \"force interrupted by user\" ||\n\t\t\t\t\terr.Error() == \"interrupted by user\" ||\n\t\t\t\t\terr.Error() == \"interrupted by user before stream start\" {\n\t\t\t\t\tstreamInterrupted = true\n\t\t\t\t\tt.Logf(\"✓ Stream was interrupted: %v\", err)\n\t\t\t\t} else {\n\t\t\t\t\tt.Logf(\"Stream completed with error: %v\", err)\n\t\t\t\t}\n\t\t\t} else {\n\t\t\t\tt.Log(\"Stream completed without error\")\n\t\t\t}\n\t\tcase <-time.After(10 * time.Second):\n\t\t\tt.Log(\"Stream timeout\")\n\t\t\tcancel() // Cancel to stop the stream goroutine\n\t\t\t<-streamDone // Wait for goroutine to exit\n\t\t}\n\n\t\t// Verify interrupt behavior\n\t\ttime.Sleep(200 * time.Millisecond)\n\t\tif handlerInvoked {\n\t\t\tt.Log(\"✓ Force interrupt handler was invoked\")\n\t\t}\n\t\tif streamInterrupted {\n\t\t\tt.Log(\"✓ Stream was interrupted by force signal\")\n\t\t}\n\t})\n}\n\n// TestAgentMultipleInterrupts tests multiple interrupts during stream\nfunc TestAgentMultipleInterrupts(t *testing.T) {\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\tagent, err := assistant.Get(\"tests.interrupt\")\n\tif err != nil {\n\t\tt.Skipf(\"Skipping test: assistant 'tests.interrupt' not found: %v\", err)\n\t\treturn\n\t}\n\n\tt.Run(\"MultipleGracefulInterrupts\", func(t *testing.T) {\n\t\t// Create context with interrupt support\n\t\tctx, cancel := newTestContextWithInterrupt(\"chat-interrupt-multiple\", \"tests.interrupt\")\n\t\tdefer func() {\n\t\t\tcancel() // Cancel context first to stop goroutines\n\t\t\ttime.Sleep(100 * time.Millisecond) // Wait for goroutines to exit\n\t\t\tctx.Release()\n\t\t}()\n\n\t\thandlerCallCount := 0\n\t\tctx.Interrupt.SetHandler(func(c *context.Context, signal *context.InterruptSignal) error {\n\t\t\thandlerCallCount++\n\t\t\tt.Logf(\"✓ Interrupt handler invoked (call %d): %d messages\", handlerCallCount, len(signal.Messages))\n\t\t\treturn nil\n\t\t})\n\n\t\tinputMessages := []context.Message{\n\t\t\t{Role: context.RoleUser, Content: \"Explain quantum computing in detail\"},\n\t\t}\n\n\t\t// Start streaming\n\t\tstreamDone := make(chan error, 1)\n\t\tgo func() {\n\t\t\t_, err := agent.Stream(ctx, inputMessages)\n\t\t\tstreamDone <- err\n\t\t}()\n\n\t\t// Wait for stream to start\n\t\ttime.Sleep(300 * time.Millisecond)\n\n\t\t// Send multiple graceful interrupts\n\t\tfor i := 1; i <= 3; i++ {\n\t\t\tsignal := &context.InterruptSignal{\n\t\t\t\tType: context.InterruptGraceful,\n\t\t\t\tMessages: []context.Message{\n\t\t\t\t\t{Role: context.RoleUser, Content: fmt.Sprintf(\"Additional question %d\", i)},\n\t\t\t\t},\n\t\t\t\tTimestamp: time.Now().UnixMilli(),\n\t\t\t}\n\n\t\t\terr = context.SendInterrupt(ctx.ID, signal)\n\t\t\tif err != nil {\n\t\t\t\tt.Logf(\"Warning: Failed to send interrupt %d: %v\", i, err)\n\t\t\t} else {\n\t\t\t\tt.Logf(\"✓ Sent interrupt %d\", i)\n\t\t\t}\n\n\t\t\ttime.Sleep(100 * time.Millisecond)\n\t\t}\n\n\t\t// Wait for stream to complete\n\t\tselect {\n\t\tcase err := <-streamDone:\n\t\t\tif err != nil {\n\t\t\t\tt.Logf(\"Stream completed with error: %v\", err)\n\t\t\t}\n\t\tcase <-time.After(10 * time.Second):\n\t\t\tt.Log(\"Stream timeout\")\n\t\t\tcancel() // Cancel to stop the stream goroutine\n\t\t\t<-streamDone // Wait for goroutine to exit\n\t\t}\n\n\t\t// Check if interrupts were received\n\t\ttime.Sleep(300 * time.Millisecond)\n\t\tpendingCount := ctx.Interrupt.GetPendingCount()\n\t\tt.Logf(\"Handler was called %d times, pending count: %d\", handlerCallCount, pendingCount)\n\n\t\tif handlerCallCount > 0 {\n\t\t\tt.Log(\"✓ Multiple interrupts were processed\")\n\t\t}\n\t})\n}\n\n// TestAgentInterruptWithoutStream tests interrupt behavior when no stream is active\nfunc TestAgentInterruptWithoutStream(t *testing.T) {\n\tt.Run(\"InterruptBeforeStream\", func(t *testing.T) {\n\t\t// Create context with interrupt support\n\t\tctx, cancel := newTestContextWithInterrupt(\"chat-interrupt-before\", \"test-assistant\")\n\t\tdefer func() {\n\t\t\tcancel()\n\t\t\tctx.Release()\n\t\t}()\n\n\t\t// Send interrupt before starting stream\n\t\tsignal := &context.InterruptSignal{\n\t\t\tType: context.InterruptGraceful,\n\t\t\tMessages: []context.Message{\n\t\t\t\t{Role: context.RoleUser, Content: \"Early interrupt\"},\n\t\t\t},\n\t\t\tTimestamp: time.Now().UnixMilli(),\n\t\t}\n\n\t\terr := context.SendInterrupt(ctx.ID, signal)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to send interrupt: %v\", err)\n\t\t}\n\n\t\t// Wait for signal to be processed\n\t\ttime.Sleep(100 * time.Millisecond)\n\n\t\t// Check if signal is in queue\n\t\treceivedSignal := ctx.Interrupt.Peek()\n\t\tif receivedSignal == nil {\n\t\t\tt.Fatal(\"Expected interrupt signal to be queued\")\n\t\t}\n\n\t\tif receivedSignal.Type != context.InterruptGraceful {\n\t\t\tt.Errorf(\"Expected graceful interrupt, got: %s\", receivedSignal.Type)\n\t\t}\n\n\t\tt.Log(\"✓ Interrupt queued before stream starts\")\n\t})\n}\n\n// TestAgentInterruptContextCleanup tests cleanup after interrupt\nfunc TestAgentInterruptContextCleanup(t *testing.T) {\n\tt.Run(\"CleanupAfterInterrupt\", func(t *testing.T) {\n\t\tctx, cancel := newTestContextWithInterrupt(\"chat-interrupt-cleanup\", \"test-assistant\")\n\n\t\t// Send interrupt\n\t\tsignal := &context.InterruptSignal{\n\t\t\tType: context.InterruptGraceful,\n\t\t\tMessages: []context.Message{{Role: context.RoleUser, Content: \"test\"}},\n\t\t\tTimestamp: time.Now().UnixMilli(),\n\t\t}\n\t\tcontext.SendInterrupt(ctx.ID, signal)\n\n\t\ttime.Sleep(100 * time.Millisecond)\n\n\t\t// Cancel and release context\n\t\tcancel()\n\t\tctx.Release()\n\n\t\t// Try to send interrupt to released context\n\t\terr := context.SendInterrupt(ctx.ID, signal)\n\t\tif err == nil {\n\t\t\tt.Error(\"Expected error when sending to released context\")\n\t\t} else {\n\t\t\tt.Logf(\"✓ Correctly rejected interrupt to released context: %v\", err)\n\t\t}\n\t})\n}\n" }, { "path": "agent/assistant/agent_next_test.go", "content": "package assistant_test\n\nimport (\n\tstdContext \"context\"\n\t\"testing\"\n\n\t\"github.com/stretchr/testify/assert\"\n\t\"github.com/yaoapp/yao/agent/assistant\"\n\t\"github.com/yaoapp/yao/agent/context\"\n\t\"github.com/yaoapp/yao/agent/output/message\"\n\t\"github.com/yaoapp/yao/agent/testutils\"\n\t\"github.com/yaoapp/yao/openapi/oauth/types\"\n)\n\n// newAgentNextTestContext creates a test context\nfunc newAgentNextTestContext(chatID, assistantID string) *context.Context {\n\tauthorized := &types.AuthorizedInfo{\n\t\tSubject: \"test-user\",\n\t\tUserID: \"test-123\",\n\t\tTenantID: \"test-tenant\",\n\t}\n\n\tctx := context.New(stdContext.Background(), authorized, chatID)\n\tctx.ID = chatID\n\tctx.AssistantID = assistantID\n\tctx.Locale = \"en-us\"\n\tctx.Client = context.Client{\n\t\tType: \"web\",\n\t\tIP: \"127.0.0.1\",\n\t}\n\tctx.Referer = context.RefererAPI\n\tctx.Accept = context.AcceptWebCUI\n\tctx.IDGenerator = message.NewIDGenerator() // Initialize ID generator\n\tctx.Metadata = make(map[string]interface{})\n\treturn ctx\n}\n\n// TestAgentNextStandard tests agent with Next Hook returning nil (standard response)\nfunc TestAgentNextStandard(t *testing.T) {\n\tif testing.Short() {\n\t\tt.Skip(\"Skipping integration test\")\n\t}\n\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\tagent, err := assistant.Get(\"tests.realworld-next\")\n\tassert.NoError(t, err)\n\n\tctx := newAgentNextTestContext(\"test-standard\", \"tests.realworld-next\")\n\tmessages := []context.Message{\n\t\t{Role: context.RoleUser, Content: \"scenario: standard - Hello\"},\n\t}\n\n\tresponse, err := agent.Stream(ctx, messages)\n\tassert.NoError(t, err)\n\tassert.NotNil(t, response)\n\n\tassert.NotNil(t, response.Completion)\n\tassert.Nil(t, response.Next)\n\n\t// Verify response structure\n\tassert.Equal(t, \"tests.realworld-next\", response.AssistantID)\n\tassert.NotEmpty(t, response.ContextID)\n\tassert.NotEmpty(t, response.RequestID)\n\tassert.NotEmpty(t, response.TraceID)\n\tassert.NotEmpty(t, response.ChatID)\n\n\t// Verify completion has content\n\tassert.NotNil(t, response.Completion.Content)\n\n\tt.Log(\"✓ Standard response test passed\")\n}\n\n// TestAgentNextCustomData tests agent with Next Hook returning custom data\nfunc TestAgentNextCustomData(t *testing.T) {\n\tif testing.Short() {\n\t\tt.Skip(\"Skipping integration test\")\n\t}\n\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\tagent, err := assistant.Get(\"tests.realworld-next\")\n\tassert.NoError(t, err)\n\n\tctx := newAgentNextTestContext(\"test-custom\", \"tests.realworld-next\")\n\tmessages := []context.Message{\n\t\t{Role: context.RoleUser, Content: \"scenario: custom_data - Give me info\"},\n\t}\n\n\tresponse, err := agent.Stream(ctx, messages)\n\tassert.NoError(t, err)\n\tassert.NotNil(t, response)\n\n\tassert.NotNil(t, response.Completion)\n\tassert.NotNil(t, response.Next)\n\n\t// Verify response structure\n\tassert.Equal(t, \"tests.realworld-next\", response.AssistantID)\n\tassert.NotEmpty(t, response.ContextID)\n\tassert.NotEmpty(t, response.RequestID)\n\tassert.NotEmpty(t, response.TraceID)\n\n\t// Verify custom data structure (from scenarioCustomData)\n\t// response.Next contains the \"data\" field value from NextHookResponse\n\tnextData, ok := response.Next.(map[string]interface{})\n\tassert.True(t, ok, \"Next should be a map\")\n\tassert.Equal(t, \"custom_response\", nextData[\"type\"])\n\tassert.Equal(t, \"This is a custom response from Next Hook\", nextData[\"message\"])\n\tassert.NotEmpty(t, nextData[\"timestamp\"])\n\tassert.NotNil(t, nextData[\"message_count\"])\n\n\tt.Log(\"✓ Custom data test passed\")\n}\n\n// TestAgentNextDelegate tests agent with Next Hook delegating to another agent\nfunc TestAgentNextDelegate(t *testing.T) {\n\tif testing.Short() {\n\t\tt.Skip(\"Skipping integration test\")\n\t}\n\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\tagent, err := assistant.Get(\"tests.realworld-next\")\n\tassert.NoError(t, err)\n\n\tctx := newAgentNextTestContext(\"test-delegate\", \"tests.realworld-next\")\n\tmessages := []context.Message{\n\t\t{Role: context.RoleUser, Content: \"scenario: delegate - Forward this\"},\n\t}\n\n\tresponse, err := agent.Stream(ctx, messages)\n\tassert.NoError(t, err)\n\tassert.NotNil(t, response)\n\n\t// Verify response structure\n\tassert.NotEmpty(t, response.AssistantID)\n\tassert.NotEmpty(t, response.ContextID)\n\tassert.NotEmpty(t, response.RequestID)\n\tassert.NotEmpty(t, response.TraceID)\n\n\t// Verify completion (delegated agent should have returned completion)\n\tassert.NotNil(t, response.Completion)\n\tassert.NotNil(t, response.Completion.Content)\n\n\t// Next should be from the delegated agent\n\t// If delegated agent also has Next hook, it will be present\n\tt.Logf(\"✓ Delegation test passed (delegated to: %s)\", response.AssistantID)\n}\n\n// TestAgentNextConditional tests agent with conditional logic in Next Hook\nfunc TestAgentNextConditional(t *testing.T) {\n\tif testing.Short() {\n\t\tt.Skip(\"Skipping integration test\")\n\t}\n\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\tagent, err := assistant.Get(\"tests.realworld-next\")\n\tassert.NoError(t, err)\n\n\tctx := newAgentNextTestContext(\"test-conditional\", \"tests.realworld-next\")\n\tmessages := []context.Message{\n\t\t// Use conditional_success sub-scenario for deterministic behavior\n\t\t// This avoids test flakiness caused by LLM response unpredictability\n\t\t{Role: context.RoleUser, Content: \"scenario: conditional_success - Task completed\"},\n\t}\n\n\tresponse, err := agent.Stream(ctx, messages)\n\tassert.NoError(t, err)\n\tassert.NotNil(t, response)\n\n\tassert.NotNil(t, response.Next)\n\n\t// Verify response structure\n\tassert.Equal(t, \"tests.realworld-next\", response.AssistantID)\n\tassert.NotEmpty(t, response.ContextID)\n\tassert.NotEmpty(t, response.RequestID)\n\tassert.NotEmpty(t, response.TraceID)\n\n\t// Verify conditional response structure (from scenarioConditional)\n\t// response.Next contains the \"data\" field value from NextHookResponse\n\tnextData, ok := response.Next.(map[string]interface{})\n\tassert.True(t, ok, \"Next should be a map\")\n\tassert.Equal(t, \"Conditional analysis complete\", nextData[\"message\"])\n\tassert.Contains(t, nextData, \"action\")\n\tassert.Contains(t, nextData, \"reason\")\n\tassert.Contains(t, nextData, \"conditions\")\n\n\t// Verify action is one of the expected values\n\taction, ok := nextData[\"action\"].(string)\n\tassert.True(t, ok)\n\tassert.Contains(t, []string{\"continue\", \"flag_for_review\", \"confirm_success\", \"summarize\", \"delegate\"}, action)\n\n\tt.Log(\"✓ Conditional logic test passed\")\n}\n\n// TestAgentWithoutNextHook tests agent without Next Hook\nfunc TestAgentWithoutNextHook(t *testing.T) {\n\tif testing.Short() {\n\t\tt.Skip(\"Skipping integration test\")\n\t}\n\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\tagent, err := assistant.Get(\"tests.create\")\n\tassert.NoError(t, err)\n\n\tctx := newAgentNextTestContext(\"test-no-next\", \"tests.create\")\n\tmessages := []context.Message{\n\t\t{Role: context.RoleUser, Content: \"Hello\"},\n\t}\n\n\tresponse, err := agent.Stream(ctx, messages)\n\tassert.NoError(t, err)\n\tassert.NotNil(t, response)\n\n\tassert.Nil(t, response.Next)\n\n\t// Verify response structure\n\tassert.Equal(t, \"tests.create\", response.AssistantID)\n\tassert.NotEmpty(t, response.ContextID)\n\tassert.NotEmpty(t, response.RequestID)\n\tassert.NotEmpty(t, response.TraceID)\n\tassert.NotEmpty(t, response.ChatID)\n\n\t// Verify completion\n\tassert.NotNil(t, response.Completion)\n\tassert.NotNil(t, response.Completion.Content)\n\n\tt.Log(\"✓ No Next Hook test passed\")\n}\n" }, { "path": "agent/assistant/assistant.go", "content": "package assistant\n\nimport (\n\t\"fmt\"\n\t\"path\"\n\n\t\"github.com/yaoapp/gou/fs\"\n\t\"github.com/yaoapp/yao/agent/caller\"\n\tagentContext \"github.com/yaoapp/yao/agent/context\"\n\t\"github.com/yaoapp/yao/agent/i18n\"\n\t\"github.com/yaoapp/yao/agent/llm\"\n\t\"github.com/yaoapp/yao/agent/search\"\n\tsearchTypes \"github.com/yaoapp/yao/agent/search/types\"\n\tstore \"github.com/yaoapp/yao/agent/store/types\"\n\tsui \"github.com/yaoapp/yao/sui/core\"\n)\n\nfunc init() {\n\t// Initialize AgentGetterFunc to allow content and search packages to call agents\n\tcaller.AgentGetterFunc = func(agentID string) (caller.AgentCaller, error) {\n\t\tast, err := Get(agentID)\n\t\tif err != nil {\n\t\t\treturn nil, err\n\t\t}\n\t\t// Return a wrapper that implements AgentCaller interface\n\t\treturn &agentCallerWrapper{ast: ast}, nil\n\t}\n\n\t// Initialize Agent JSAPI factory for ctx.agent.* methods\n\tcaller.SetJSAPIFactory()\n\n\t// Initialize LLM JSAPI factory for ctx.llm.* methods\n\tllm.SetJSAPIFactory()\n\n\t// Initialize Search JSAPI factory with config getter\n\tsearch.SetJSAPIFactory(func(assistantID string) (*searchTypes.Config, *search.Uses) {\n\t\tast, err := Get(assistantID)\n\t\tif err != nil || ast == nil {\n\t\t\treturn nil, nil\n\t\t}\n\t\t// Convert assistant.Uses to search.Uses\n\t\tvar uses *search.Uses\n\t\tif ast.Uses != nil {\n\t\t\tuses = &search.Uses{\n\t\t\t\tSearch: ast.Uses.Search,\n\t\t\t\tWeb: ast.Uses.Web,\n\t\t\t\tKeyword: ast.Uses.Keyword,\n\t\t\t\tQueryDSL: ast.Uses.QueryDSL,\n\t\t\t\tRerank: ast.Uses.Rerank,\n\t\t\t}\n\t\t}\n\t\treturn ast.Search, uses\n\t})\n}\n\n// agentCallerWrapper wraps Assistant to implement AgentCaller interface\ntype agentCallerWrapper struct {\n\tast *Assistant\n}\n\nfunc (w *agentCallerWrapper) Stream(ctx *agentContext.Context, messages []agentContext.Message, options ...*agentContext.Options) (*agentContext.Response, error) {\n\treturn w.ast.Stream(ctx, messages, options...)\n}\n\n// Get get the assistant by id\nfunc Get(id string) (*Assistant, error) {\n\treturn LoadStore(id)\n}\n\n// GetPlaceholder returns the placeholder of the assistant\nfunc (ast *Assistant) GetPlaceholder(locale string) *store.Placeholder {\n\n\tprompts := []string{}\n\tif ast.Placeholder.Prompts != nil {\n\t\tprompts = i18n.Translate(ast.ID, locale, ast.Placeholder.Prompts).([]string)\n\t}\n\ttitle := i18n.Translate(ast.ID, locale, ast.Placeholder.Title).(string)\n\tdescription := i18n.Translate(ast.ID, locale, ast.Placeholder.Description).(string)\n\treturn &store.Placeholder{\n\t\tTitle: title,\n\t\tDescription: description,\n\t\tPrompts: prompts,\n\t}\n}\n\n// GetName returns the name of the assistant\nfunc (ast *Assistant) GetName(locale string) string {\n\treturn i18n.Translate(ast.ID, locale, ast.Name).(string)\n}\n\n// GetDescription returns the description of the assistant\nfunc (ast *Assistant) GetDescription(locale string) string {\n\treturn i18n.Translate(ast.ID, locale, ast.Description).(string)\n}\n\n// Save save the assistant\nfunc (ast *Assistant) Save() error {\n\tif storage == nil {\n\t\treturn fmt.Errorf(\"storage is not set\")\n\t}\n\n\t_, err := storage.SaveAssistant(&ast.AssistantModel)\n\tif err != nil {\n\t\treturn err\n\t}\n\n\treturn nil\n}\n\n// Map convert the assistant to a map\nfunc (ast *Assistant) Map() map[string]interface{} {\n\n\tif ast == nil {\n\t\treturn nil\n\t}\n\n\treturn map[string]interface{}{\n\t\t\"assistant_id\": ast.ID,\n\t\t\"type\": ast.Type,\n\t\t\"name\": ast.Name,\n\t\t\"readonly\": ast.Readonly,\n\t\t\"public\": ast.Public,\n\t\t\"share\": ast.Share,\n\t\t\"avatar\": ast.Avatar,\n\t\t\"connector\": ast.Connector,\n\t\t\"connector_options\": ast.ConnectorOptions,\n\t\t\"path\": ast.Path,\n\t\t\"built_in\": ast.BuiltIn,\n\t\t\"sort\": ast.Sort,\n\t\t\"description\": ast.Description,\n\t\t\"options\": ast.Options,\n\t\t\"prompts\": ast.Prompts,\n\t\t\"prompt_presets\": ast.PromptPresets,\n\t\t\"disable_global_prompts\": ast.DisableGlobalPrompts,\n\t\t\"source\": ast.Source,\n\t\t\"kb\": ast.KB,\n\t\t\"db\": ast.DB,\n\t\t\"mcp\": ast.MCP,\n\t\t\"workflow\": ast.Workflow,\n\t\t\"tags\": ast.Tags,\n\t\t\"modes\": ast.Modes,\n\t\t\"default_mode\": ast.DefaultMode,\n\t\t\"mentionable\": ast.Mentionable,\n\t\t\"automated\": ast.Automated,\n\t\t\"placeholder\": ast.Placeholder,\n\t\t\"locales\": ast.Locales,\n\t\t\"uses\": ast.Uses,\n\t\t\"search\": ast.Search,\n\t\t\"dependencies\": ast.Dependencies,\n\t\t\"created_at\": store.ToMySQLTime(ast.CreatedAt),\n\t\t\"updated_at\": store.ToMySQLTime(ast.UpdatedAt),\n\t}\n}\n\n// Validate validates the assistant configuration\nfunc (ast *Assistant) Validate() error {\n\tif ast.ID == \"\" {\n\t\treturn fmt.Errorf(\"assistant_id is required\")\n\t}\n\tif ast.Name == \"\" {\n\t\treturn fmt.Errorf(\"name is required\")\n\t}\n\tif ast.Connector == \"\" {\n\t\treturn fmt.Errorf(\"connector is required\")\n\t}\n\treturn nil\n}\n\n// Assets get the assets content\nfunc (ast *Assistant) Assets(name string, data sui.Data) (string, error) {\n\n\tapp, err := fs.Get(\"app\")\n\tif err != nil {\n\t\treturn \"\", err\n\t}\n\n\troot := path.Join(ast.Path, \"assets\", name)\n\traw, err := app.ReadFile(root)\n\tif err != nil {\n\t\treturn \"\", err\n\t}\n\n\tif data != nil {\n\t\tcontent, _ := data.Replace(string(raw))\n\t\treturn content, nil\n\t}\n\n\treturn string(raw), nil\n}\n\n// Clone creates a deep copy of the assistant\nfunc (ast *Assistant) Clone() *Assistant {\n\tif ast == nil {\n\t\treturn nil\n\t}\n\n\tclone := &Assistant{\n\t\tAssistantModel: store.AssistantModel{\n\t\t\tID: ast.ID,\n\t\t\tType: ast.Type,\n\t\t\tName: ast.Name,\n\t\t\tAvatar: ast.Avatar,\n\t\t\tConnector: ast.Connector,\n\t\t\tPath: ast.Path,\n\t\t\tBuiltIn: ast.BuiltIn,\n\t\t\tSort: ast.Sort,\n\t\t\tDescription: ast.Description,\n\t\t\tReadonly: ast.Readonly,\n\t\t\tPublic: ast.Public,\n\t\t\tShare: ast.Share,\n\t\t\tMentionable: ast.Mentionable,\n\t\t\tAutomated: ast.Automated,\n\t\t\tDisableGlobalPrompts: ast.DisableGlobalPrompts,\n\t\t\tSource: ast.Source,\n\t\t\tCreatedAt: ast.CreatedAt,\n\t\t\tUpdatedAt: ast.UpdatedAt,\n\t\t},\n\t\tHookScript: ast.HookScript,\n\t}\n\n\t// Deep copy tags\n\tif ast.Tags != nil {\n\t\tclone.Tags = make([]string, len(ast.Tags))\n\t\tcopy(clone.Tags, ast.Tags)\n\t}\n\n\t// Deep copy modes\n\tif ast.Modes != nil {\n\t\tclone.Modes = make([]string, len(ast.Modes))\n\t\tcopy(clone.Modes, ast.Modes)\n\t}\n\n\t// Copy default_mode (simple string)\n\tclone.DefaultMode = ast.DefaultMode\n\n\t// Deep copy KB\n\tif ast.KB != nil {\n\t\tclone.KB = &store.KnowledgeBase{}\n\t\tif ast.KB.Collections != nil {\n\t\t\tclone.KB.Collections = make([]string, len(ast.KB.Collections))\n\t\t\tcopy(clone.KB.Collections, ast.KB.Collections)\n\t\t}\n\t\tif ast.KB.Options != nil {\n\t\t\tclone.KB.Options = make(map[string]interface{})\n\t\t\tfor k, v := range ast.KB.Options {\n\t\t\t\tclone.KB.Options[k] = v\n\t\t\t}\n\t\t}\n\t}\n\n\t// Deep copy DB\n\tif ast.DB != nil {\n\t\tclone.DB = &store.Database{}\n\t\tif ast.DB.Models != nil {\n\t\t\tclone.DB.Models = make([]string, len(ast.DB.Models))\n\t\t\tcopy(clone.DB.Models, ast.DB.Models)\n\t\t}\n\t\tif ast.DB.Options != nil {\n\t\t\tclone.DB.Options = make(map[string]interface{})\n\t\t\tfor k, v := range ast.DB.Options {\n\t\t\t\tclone.DB.Options[k] = v\n\t\t\t}\n\t\t}\n\t}\n\n\t// Deep copy MCP\n\tif ast.MCP != nil {\n\t\tclone.MCP = &store.MCPServers{}\n\t\tif ast.MCP.Servers != nil {\n\t\t\tclone.MCP.Servers = make([]store.MCPServerConfig, len(ast.MCP.Servers))\n\t\t\tfor i, server := range ast.MCP.Servers {\n\t\t\t\tclone.MCP.Servers[i] = store.MCPServerConfig{\n\t\t\t\t\tServerID: server.ServerID,\n\t\t\t\t}\n\t\t\t\t// Deep copy Resources slice\n\t\t\t\tif server.Resources != nil {\n\t\t\t\t\tclone.MCP.Servers[i].Resources = make([]string, len(server.Resources))\n\t\t\t\t\tcopy(clone.MCP.Servers[i].Resources, server.Resources)\n\t\t\t\t}\n\t\t\t\t// Deep copy Tools slice\n\t\t\t\tif server.Tools != nil {\n\t\t\t\t\tclone.MCP.Servers[i].Tools = make([]string, len(server.Tools))\n\t\t\t\t\tcopy(clone.MCP.Servers[i].Tools, server.Tools)\n\t\t\t\t}\n\t\t\t}\n\t\t}\n\t\tif ast.MCP.Options != nil {\n\t\t\tclone.MCP.Options = make(map[string]interface{})\n\t\t\tfor k, v := range ast.MCP.Options {\n\t\t\t\tclone.MCP.Options[k] = v\n\t\t\t}\n\t\t}\n\t}\n\n\t// Deep copy options\n\tif ast.Options != nil {\n\t\tclone.Options = make(map[string]interface{})\n\t\tfor k, v := range ast.Options {\n\t\t\tclone.Options[k] = v\n\t\t}\n\t}\n\n\t// Deep copy prompts\n\tif ast.Prompts != nil {\n\t\tclone.Prompts = make([]store.Prompt, len(ast.Prompts))\n\t\tcopy(clone.Prompts, ast.Prompts)\n\t}\n\n\t// Deep copy prompt presets\n\tif ast.PromptPresets != nil {\n\t\tclone.PromptPresets = make(map[string][]store.Prompt)\n\t\tfor k, v := range ast.PromptPresets {\n\t\t\tprompts := make([]store.Prompt, len(v))\n\t\t\tcopy(prompts, v)\n\t\t\tclone.PromptPresets[k] = prompts\n\t\t}\n\t}\n\n\t// Deep copy connector options\n\tif ast.ConnectorOptions != nil {\n\t\tclone.ConnectorOptions = &store.ConnectorOptions{\n\t\t\tOptional: ast.ConnectorOptions.Optional,\n\t\t}\n\t\tif ast.ConnectorOptions.Connectors != nil {\n\t\t\tclone.ConnectorOptions.Connectors = make([]string, len(ast.ConnectorOptions.Connectors))\n\t\t\tcopy(clone.ConnectorOptions.Connectors, ast.ConnectorOptions.Connectors)\n\t\t}\n\t\tif ast.ConnectorOptions.Filters != nil {\n\t\t\tclone.ConnectorOptions.Filters = make([]store.ModelCapability, len(ast.ConnectorOptions.Filters))\n\t\t\tcopy(clone.ConnectorOptions.Filters, ast.ConnectorOptions.Filters)\n\t\t}\n\t}\n\n\t// Deep copy workflow\n\tif ast.Workflow != nil {\n\t\tclone.Workflow = &store.Workflow{}\n\t\tif ast.Workflow.Workflows != nil {\n\t\t\tclone.Workflow.Workflows = make([]string, len(ast.Workflow.Workflows))\n\t\t\tcopy(clone.Workflow.Workflows, ast.Workflow.Workflows)\n\t\t}\n\t\tif ast.Workflow.Options != nil {\n\t\t\tclone.Workflow.Options = make(map[string]interface{})\n\t\t\tfor k, v := range ast.Workflow.Options {\n\t\t\t\tclone.Workflow.Options[k] = v\n\t\t\t}\n\t\t}\n\t}\n\n\t// Deep copy placeholder\n\tif ast.Placeholder != nil {\n\t\tclone.Placeholder = &store.Placeholder{\n\t\t\tTitle: ast.Placeholder.Title,\n\t\t\tDescription: ast.Placeholder.Description,\n\t\t}\n\t\tif ast.Placeholder.Prompts != nil {\n\t\t\tclone.Placeholder.Prompts = make([]string, len(ast.Placeholder.Prompts))\n\t\t\tcopy(clone.Placeholder.Prompts, ast.Placeholder.Prompts)\n\t\t}\n\t}\n\n\t// Deep copy locales\n\tif ast.Locales != nil {\n\t\tclone.Locales = make(i18n.Map)\n\t\tfor k, v := range ast.Locales {\n\t\t\t// Deep copy messages\n\t\t\tmessages := make(map[string]any)\n\t\t\tif v.Messages != nil {\n\t\t\t\tfor mk, mv := range v.Messages {\n\t\t\t\t\tmessages[mk] = mv\n\t\t\t\t}\n\t\t\t}\n\t\t\tclone.Locales[k] = i18n.I18n{\n\t\t\t\tLocale: v.Locale,\n\t\t\t\tMessages: messages,\n\t\t\t}\n\t\t}\n\t}\n\n\t// Deep copy uses\n\tif ast.Uses != nil {\n\t\tclone.Uses = &agentContext.Uses{\n\t\t\tVision: ast.Uses.Vision,\n\t\t\tAudio: ast.Uses.Audio,\n\t\t\tSearch: ast.Uses.Search,\n\t\t\tFetch: ast.Uses.Fetch,\n\t\t\tWeb: ast.Uses.Web,\n\t\t\tKeyword: ast.Uses.Keyword,\n\t\t\tQueryDSL: ast.Uses.QueryDSL,\n\t\t\tRerank: ast.Uses.Rerank,\n\t\t}\n\t}\n\n\t// Deep copy search config\n\tif ast.Search != nil {\n\t\tclone.Search = &searchTypes.Config{}\n\t\tif ast.Search.Web != nil {\n\t\t\tclone.Search.Web = &searchTypes.WebConfig{\n\t\t\t\tProvider: ast.Search.Web.Provider,\n\t\t\t\tAPIKeyEnv: ast.Search.Web.APIKeyEnv,\n\t\t\t\tMaxResults: ast.Search.Web.MaxResults,\n\t\t\t}\n\t\t}\n\t\tif ast.Search.KB != nil {\n\t\t\tclone.Search.KB = &searchTypes.KBConfig{\n\t\t\t\tThreshold: ast.Search.KB.Threshold,\n\t\t\t\tGraph: ast.Search.KB.Graph,\n\t\t\t}\n\t\t\tif ast.Search.KB.Collections != nil {\n\t\t\t\tclone.Search.KB.Collections = make([]string, len(ast.Search.KB.Collections))\n\t\t\t\tcopy(clone.Search.KB.Collections, ast.Search.KB.Collections)\n\t\t\t}\n\t\t}\n\t\tif ast.Search.DB != nil {\n\t\t\tclone.Search.DB = &searchTypes.DBConfig{\n\t\t\t\tMaxResults: ast.Search.DB.MaxResults,\n\t\t\t}\n\t\t\tif ast.Search.DB.Models != nil {\n\t\t\t\tclone.Search.DB.Models = make([]string, len(ast.Search.DB.Models))\n\t\t\t\tcopy(clone.Search.DB.Models, ast.Search.DB.Models)\n\t\t\t}\n\t\t}\n\t\tif ast.Search.Keyword != nil {\n\t\t\tclone.Search.Keyword = &searchTypes.KeywordConfig{\n\t\t\t\tMaxKeywords: ast.Search.Keyword.MaxKeywords,\n\t\t\t\tLanguage: ast.Search.Keyword.Language,\n\t\t\t}\n\t\t}\n\t\tif ast.Search.QueryDSL != nil {\n\t\t\tclone.Search.QueryDSL = &searchTypes.QueryDSLConfig{\n\t\t\t\tStrict: ast.Search.QueryDSL.Strict,\n\t\t\t}\n\t\t}\n\t\tif ast.Search.Rerank != nil {\n\t\t\tclone.Search.Rerank = &searchTypes.RerankConfig{\n\t\t\t\tTopN: ast.Search.Rerank.TopN,\n\t\t\t}\n\t\t}\n\t\tif ast.Search.Citation != nil {\n\t\t\tclone.Search.Citation = &searchTypes.CitationConfig{\n\t\t\t\tFormat: ast.Search.Citation.Format,\n\t\t\t\tAutoInjectPrompt: ast.Search.Citation.AutoInjectPrompt,\n\t\t\t\tCustomPrompt: ast.Search.Citation.CustomPrompt,\n\t\t\t}\n\t\t}\n\t\tif ast.Search.Weights != nil {\n\t\t\tclone.Search.Weights = &searchTypes.WeightsConfig{\n\t\t\t\tUser: ast.Search.Weights.User,\n\t\t\t\tHook: ast.Search.Weights.Hook,\n\t\t\t\tAuto: ast.Search.Weights.Auto,\n\t\t\t}\n\t\t}\n\t\tif ast.Search.Options != nil {\n\t\t\tclone.Search.Options = &searchTypes.OptionsConfig{\n\t\t\t\tSkipThreshold: ast.Search.Options.SkipThreshold,\n\t\t\t}\n\t\t}\n\t}\n\n\t// Deep copy dependencies\n\tif ast.Dependencies != nil {\n\t\tclone.Dependencies = make(map[string]string, len(ast.Dependencies))\n\t\tfor k, v := range ast.Dependencies {\n\t\t\tclone.Dependencies[k] = v\n\t\t}\n\t}\n\n\treturn clone\n}\n\n// GetInfo returns the basic info of the assistant with optional locale\nfunc (ast *Assistant) GetInfo(locale ...string) *store.AssistantInfo {\n\tif ast == nil {\n\t\treturn nil\n\t}\n\n\tloc := \"\"\n\tif len(locale) > 0 {\n\t\tloc = locale[0]\n\t}\n\n\tinfo := &store.AssistantInfo{\n\t\tAssistantID: ast.ID,\n\t\tAvatar: ast.Avatar,\n\t\tConnector: ast.Connector,\n\t\tConnectorOptions: ast.ConnectorOptions,\n\t\tModes: ast.Modes,\n\t\tDefaultMode: ast.DefaultMode,\n\t\tSandbox: ast.IsSandbox,\n\t\tComputerFilter: ast.ComputerFilter,\n\t}\n\n\tif loc != \"\" {\n\t\tinfo.Name = ast.GetName(loc)\n\t\tinfo.Description = ast.GetDescription(loc)\n\t} else {\n\t\tinfo.Name = ast.Name\n\t\tinfo.Description = ast.Description\n\t}\n\n\treturn info\n}\n\n// GetInfoByIDs retrieves basic info for multiple assistants by their IDs\n// Returns a map of assistant_id -> AssistantInfo\nfunc GetInfoByIDs(ids []string, locale ...string) map[string]*store.AssistantInfo {\n\tresult := make(map[string]*store.AssistantInfo)\n\n\tif len(ids) == 0 {\n\t\treturn result\n\t}\n\n\tfor _, id := range ids {\n\t\tast, err := Get(id)\n\t\tif err != nil || ast == nil {\n\t\t\tcontinue\n\t\t}\n\t\tresult[id] = ast.GetInfo(locale...)\n\t}\n\n\treturn result\n}\n\n// Update updates the assistant properties\nfunc (ast *Assistant) Update(data map[string]interface{}) error {\n\tif ast == nil {\n\t\treturn fmt.Errorf(\"assistant is nil\")\n\t}\n\n\tif v, ok := data[\"name\"].(string); ok {\n\t\tast.Name = v\n\t}\n\tif v, ok := data[\"avatar\"].(string); ok {\n\t\tast.Avatar = v\n\t}\n\tif v, ok := data[\"description\"].(string); ok {\n\t\tast.Description = v\n\t}\n\tif v, ok := data[\"connector\"].(string); ok {\n\t\tast.Connector = v\n\t}\n\n\t// Note: tools field is deprecated, now handled by MCP\n\n\tif v, ok := data[\"type\"].(string); ok {\n\t\tast.Type = v\n\t}\n\tif v, ok := data[\"sort\"].(int); ok {\n\t\tast.Sort = v\n\t}\n\tif v, ok := data[\"mentionable\"].(bool); ok {\n\t\tast.Mentionable = v\n\t}\n\tif v, ok := data[\"automated\"].(bool); ok {\n\t\tast.Automated = v\n\t}\n\tif v, ok := data[\"disable_global_prompts\"].(bool); ok {\n\t\tast.DisableGlobalPrompts = v\n\t}\n\tif v, ok := data[\"readonly\"].(bool); ok {\n\t\tast.Readonly = v\n\t}\n\tif v, ok := data[\"public\"].(bool); ok {\n\t\tast.Public = v\n\t}\n\tif v, ok := data[\"share\"].(string); ok {\n\t\tast.Share = v\n\t}\n\tif v, ok := data[\"tags\"].([]string); ok {\n\t\tast.Tags = v\n\t}\n\tif v, ok := data[\"modes\"].([]string); ok {\n\t\tast.Modes = v\n\t}\n\tif v, ok := data[\"default_mode\"].(string); ok {\n\t\tast.DefaultMode = v\n\t}\n\tif v, ok := data[\"options\"].(map[string]interface{}); ok {\n\t\tast.Options = v\n\t}\n\tif v, ok := data[\"source\"].(string); ok {\n\t\tast.Source = v\n\t}\n\n\t// ConnectorOptions\n\tif v, has := data[\"connector_options\"]; has {\n\t\tconnOpts, err := store.ToConnectorOptions(v)\n\t\tif err != nil {\n\t\t\treturn err\n\t\t}\n\t\tast.ConnectorOptions = connOpts\n\t}\n\n\t// PromptPresets\n\tif v, has := data[\"prompt_presets\"]; has {\n\t\tpresets, err := store.ToPromptPresets(v)\n\t\tif err != nil {\n\t\t\treturn err\n\t\t}\n\t\tast.PromptPresets = presets\n\t}\n\n\t// KB\n\tif v, has := data[\"kb\"]; has {\n\t\tkb, err := store.ToKnowledgeBase(v)\n\t\tif err != nil {\n\t\t\treturn err\n\t\t}\n\t\tast.KB = kb\n\t}\n\n\t// DB\n\tif v, has := data[\"db\"]; has {\n\t\tdb, err := store.ToDatabase(v)\n\t\tif err != nil {\n\t\t\treturn err\n\t\t}\n\t\tast.DB = db\n\t}\n\n\t// MCP\n\tif v, has := data[\"mcp\"]; has {\n\t\tmcp, err := store.ToMCPServers(v)\n\t\tif err != nil {\n\t\t\treturn err\n\t\t}\n\t\tast.MCP = mcp\n\t}\n\n\t// Workflow\n\tif v, has := data[\"workflow\"]; has {\n\t\tworkflow, err := store.ToWorkflow(v)\n\t\tif err != nil {\n\t\t\treturn err\n\t\t}\n\t\tast.Workflow = workflow\n\t}\n\n\t// Uses\n\tif v, has := data[\"uses\"]; has {\n\t\tuses, err := store.ToUses(v)\n\t\tif err != nil {\n\t\t\treturn err\n\t\t}\n\t\tast.Uses = uses\n\t}\n\n\t// Search\n\tif v, has := data[\"search\"]; has {\n\t\tsearch, err := store.ToSearchConfig(v)\n\t\tif err != nil {\n\t\t\treturn err\n\t\t}\n\t\tast.Search = search\n\t}\n\n\t// Dependencies\n\tif v, has := data[\"dependencies\"]; has {\n\t\tif v == nil {\n\t\t\tast.Dependencies = nil\n\t\t} else {\n\t\t\tswitch d := v.(type) {\n\t\t\tcase map[string]string:\n\t\t\t\tast.Dependencies = d\n\t\t\tcase map[string]interface{}:\n\t\t\t\tdeps := make(map[string]string, len(d))\n\t\t\t\tfor k, val := range d {\n\t\t\t\t\tif s, ok := val.(string); ok {\n\t\t\t\t\t\tdeps[k] = s\n\t\t\t\t\t}\n\t\t\t\t}\n\t\t\t\tast.Dependencies = deps\n\t\t\t}\n\t\t}\n\t}\n\n\treturn ast.Validate()\n}\n\n// GetMergedSearchConfig returns the search config for this assistant\n// Note: The config is already merged with global config during loading (loadMap)\nfunc (ast *Assistant) GetMergedSearchConfig() *searchTypes.Config {\n\treturn ast.Search\n}\n" }, { "path": "agent/assistant/build.go", "content": "package assistant\n\nimport (\n\t\"fmt\"\n\n\t\"github.com/spf13/cast\"\n\t\"github.com/yaoapp/gou/json\"\n\t\"github.com/yaoapp/yao/agent/context\"\n\tstore \"github.com/yaoapp/yao/agent/store/types\"\n)\n\n// BuildRequest build the LLM request\nfunc (ast *Assistant) BuildRequest(ctx *context.Context, messages []context.Message, createResponse *context.HookCreateResponse) ([]context.Message, *context.CompletionOptions, error) {\n\t// Build completion options from createResponse and ctx (includes MCP tools)\n\toptions, mcpSamplesPrompt, err := ast.buildCompletionOptions(ctx, createResponse)\n\tif err != nil {\n\t\treturn nil, nil, err\n\t}\n\n\t// Build final messages with proper priority (includes MCP samples if available)\n\tfinalMessages, err := ast.buildMessages(ctx, messages, createResponse, mcpSamplesPrompt)\n\tif err != nil {\n\t\treturn nil, nil, err\n\t}\n\n\treturn finalMessages, options, nil\n}\n\n// buildMessages builds the final message list with proper priority\n// Priority: Prompts > MCP Samples > createResponse.Messages > input messages\n// If createResponse is nil or has no messages, use input messages\nfunc (ast *Assistant) buildMessages(ctx *context.Context, messages []context.Message, createResponse *context.HookCreateResponse, mcpSamplesPrompt string) ([]context.Message, error) {\n\tvar finalMessages []context.Message\n\n\t// If createResponse is nil or has no messages, use input messages\n\tif createResponse == nil || len(createResponse.Messages) == 0 {\n\t\tfinalMessages = messages\n\t} else {\n\t\t// createResponse.Messages takes priority over input messages\n\t\tfinalMessages = createResponse.Messages\n\t}\n\n\t// Add MCP samples prompt as a system message (if available)\n\tif mcpSamplesPrompt != \"\" {\n\t\tmcpSamplesMsg := context.Message{\n\t\t\tRole: context.RoleSystem,\n\t\t\tContent: mcpSamplesPrompt,\n\t\t}\n\t\t// Prepend MCP samples before other messages\n\t\tfinalMessages = append([]context.Message{mcpSamplesMsg}, finalMessages...)\n\t}\n\n\t// Build and prepend system prompts (global + assistant prompts)\n\tpromptMessages := ast.buildSystemPrompts(ctx, createResponse)\n\tif len(promptMessages) > 0 {\n\t\tfinalMessages = append(promptMessages, finalMessages...)\n\t}\n\n\treturn finalMessages, nil\n}\n\n// buildSystemPrompts builds system prompt messages from global prompts and assistant prompts\n// Order: Global prompts (if not disabled) -> Assistant prompts (or preset)\n// Variables are parsed with context information\n//\n// Priority for prompt preset selection:\n// 1. createResponse.PromptPreset (highest)\n// 2. ctx.Metadata[\"__prompt_preset\"]\n// 3. ast.Prompts (default)\n//\n// Priority for disable global prompts:\n// 1. createResponse.DisableGlobalPrompts (highest)\n// 2. ctx.Metadata[\"__disable_global_prompts\"]\n// 3. ast.DisableGlobalPrompts (default)\nfunc (ast *Assistant) buildSystemPrompts(ctx *context.Context, createResponse *context.HookCreateResponse) []context.Message {\n\t// Build context variables from ctx and ast\n\tctxVars := ast.buildContextVariables(ctx)\n\n\t// Determine if global prompts should be disabled\n\tdisableGlobal := ast.shouldDisableGlobalPrompts(ctx, createResponse)\n\n\t// Get assistant prompts (default or preset)\n\tassistantPrompts := ast.getAssistantPrompts(ctx, createResponse)\n\n\tvar allPrompts []store.Prompt\n\n\t// 1. Add global prompts (if not disabled)\n\tif !disableGlobal && len(globalPrompts) > 0 {\n\t\t// Parse global prompts with context variables\n\t\tparsedGlobal := store.Prompts(globalPrompts).Parse(ctxVars)\n\t\tallPrompts = append(allPrompts, parsedGlobal...)\n\t}\n\n\t// 2. Add assistant prompts (default or preset)\n\tif len(assistantPrompts) > 0 {\n\t\t// Parse assistant prompts with context variables\n\t\tparsedAssistant := store.Prompts(assistantPrompts).Parse(ctxVars)\n\t\tallPrompts = append(allPrompts, parsedAssistant...)\n\t}\n\n\t// Convert to context.Message slice\n\tif len(allPrompts) == 0 {\n\t\treturn nil\n\t}\n\n\tmessages := make([]context.Message, 0, len(allPrompts))\n\tfor _, prompt := range allPrompts {\n\t\tmsg := context.Message{\n\t\t\tRole: context.MessageRole(prompt.Role),\n\t\t\tContent: prompt.Content,\n\t\t}\n\t\tif prompt.Name != \"\" {\n\t\t\tname := prompt.Name\n\t\t\tmsg.Name = &name\n\t\t}\n\t\tmessages = append(messages, msg)\n\t}\n\n\treturn messages\n}\n\n// shouldDisableGlobalPrompts determines if global prompts should be disabled\n// Priority: createResponse > ctx.Metadata > ast.DisableGlobalPrompts\nfunc (ast *Assistant) shouldDisableGlobalPrompts(ctx *context.Context, createResponse *context.HookCreateResponse) bool {\n\t// Priority 1: Hook response (highest)\n\tif createResponse != nil && createResponse.DisableGlobalPrompts != nil {\n\t\treturn *createResponse.DisableGlobalPrompts\n\t}\n\n\t// Priority 2: ctx.Metadata[\"__disable_global_prompts\"]\n\tif ctx != nil && ctx.Metadata != nil {\n\t\tif disable, ok := ctx.Metadata[\"__disable_global_prompts\"].(bool); ok {\n\t\t\treturn disable\n\t\t}\n\t}\n\n\t// Priority 3: Assistant configuration (default)\n\treturn ast.DisableGlobalPrompts\n}\n\n// getAssistantPrompts returns the assistant prompts based on preset selection\n// Priority: createResponse.PromptPreset > ctx.Metadata[\"__prompt_preset\"] > ast.Prompts\nfunc (ast *Assistant) getAssistantPrompts(ctx *context.Context, createResponse *context.HookCreateResponse) []store.Prompt {\n\t// Get preset key\n\tpresetKey := ast.getPromptPresetKey(ctx, createResponse)\n\n\t// If preset key is specified and exists, use it\n\tif presetKey != \"\" && ast.PromptPresets != nil {\n\t\tif presets, ok := ast.PromptPresets[presetKey]; ok && len(presets) > 0 {\n\t\t\treturn presets\n\t\t}\n\t}\n\n\t// Fallback to default prompts\n\treturn ast.Prompts\n}\n\n// getPromptPresetKey returns the prompt preset key\n// Priority: createResponse.PromptPreset > ctx.Metadata[\"__prompt_preset\"]\nfunc (ast *Assistant) getPromptPresetKey(ctx *context.Context, createResponse *context.HookCreateResponse) string {\n\t// Priority 1: Hook response (highest)\n\tif createResponse != nil && createResponse.PromptPreset != \"\" {\n\t\treturn createResponse.PromptPreset\n\t}\n\n\t// Priority 2: ctx.Metadata[\"__prompt_preset\"]\n\tif ctx != nil && ctx.Metadata != nil {\n\t\tif preset, ok := ctx.Metadata[\"__prompt_preset\"].(string); ok && preset != \"\" {\n\t\t\treturn preset\n\t\t}\n\t}\n\n\t// No preset specified\n\treturn \"\"\n}\n\n// buildContextVariables extracts context variables from Context and Assistant for prompt parsing\nfunc (ast *Assistant) buildContextVariables(ctx *context.Context) map[string]string {\n\tvars := make(map[string]string)\n\n\t// Get locale from ctx (default to empty)\n\tlocale := \"\"\n\tif ctx != nil && ctx.Locale != \"\" {\n\t\tlocale = ctx.Locale\n\t}\n\n\t// Assistant info (with locale support)\n\tif ast != nil {\n\t\tif ast.ID != \"\" {\n\t\t\tvars[\"ASSISTANT_ID\"] = ast.ID\n\t\t}\n\t\t// Use localized name and description\n\t\tname := ast.GetName(locale)\n\t\tif name != \"\" {\n\t\t\tvars[\"ASSISTANT_NAME\"] = name\n\t\t}\n\t\tdescription := ast.GetDescription(locale)\n\t\tif description != \"\" {\n\t\t\tvars[\"ASSISTANT_DESCRIPTION\"] = description\n\t\t}\n\t\tif ast.Type != \"\" {\n\t\t\tvars[\"ASSISTANT_TYPE\"] = ast.Type\n\t\t}\n\t}\n\n\tif ctx == nil {\n\t\treturn vars\n\t}\n\n\t// Basic context info\n\tif ctx.ChatID != \"\" {\n\t\tvars[\"CHAT_ID\"] = ctx.ChatID\n\t}\n\tif ctx.Locale != \"\" {\n\t\tvars[\"LOCALE\"] = ctx.Locale\n\t}\n\tif ctx.Theme != \"\" {\n\t\tvars[\"THEME\"] = ctx.Theme\n\t}\n\tif ctx.Route != \"\" {\n\t\tvars[\"ROUTE\"] = ctx.Route\n\t}\n\tif ctx.Referer != \"\" {\n\t\tvars[\"REFERER\"] = ctx.Referer\n\t}\n\n\t// Client info (only non-sensitive fields)\n\tif ctx.Client.Type != \"\" {\n\t\tvars[\"CLIENT_TYPE\"] = ctx.Client.Type\n\t}\n\n\t// Authorized info (only internal IDs, no PII)\n\t// Note: USER_SUBJECT and CLIENT_IP are excluded for privacy/GDPR compliance\n\tif ctx.Authorized != nil {\n\t\tif ctx.Authorized.UserID != \"\" {\n\t\t\tvars[\"USER_ID\"] = ctx.Authorized.UserID\n\t\t}\n\t\tif ctx.Authorized.TeamID != \"\" {\n\t\t\tvars[\"TEAM_ID\"] = ctx.Authorized.TeamID\n\t\t}\n\t\tif ctx.Authorized.TenantID != \"\" {\n\t\t\tvars[\"TENANT_ID\"] = ctx.Authorized.TenantID\n\t\t}\n\t}\n\n\t// Metadata - custom variables from ctx.Metadata\n\t// All metadata keys are exposed as $CTX.{KEY}\n\t// Supports string, int, uint, float, bool types\n\tif ctx.Metadata != nil {\n\t\tfor key, value := range ctx.Metadata {\n\t\t\tif value == nil {\n\t\t\t\tcontinue\n\t\t\t}\n\t\t\tstrVal := cast.ToString(value)\n\t\t\tif strVal != \"\" {\n\t\t\t\tvars[key] = strVal\n\t\t\t}\n\t\t}\n\t}\n\n\treturn vars\n}\n\n// buildCompletionOptions builds completion options from multiple sources\n// Priority (lowest to highest, later overrides earlier): ast > ctx > createResponse\n// The priority means: if createResponse has a value, use it; else use ctx; else use ast\n// Returns (options, mcpSamplesPrompt, error)\nfunc (ast *Assistant) buildCompletionOptions(ctx *context.Context, createResponse *context.HookCreateResponse) (*context.CompletionOptions, string, error) {\n\toptions := &context.CompletionOptions{}\n\n\t// Layer 1 (base): Apply ast - Assistant configuration\n\tif err := ast.applyAssistantOptions(options); err != nil {\n\t\treturn nil, \"\", err\n\t}\n\n\t// Layer 2 (middle): Apply ctx - Context configuration (overrides ast)\n\tast.applyContextOptions(options, ctx)\n\n\t// Layer 3 (highest): Apply createResponse - Hook configuration (overrides all)\n\tif createResponse != nil {\n\t\tast.applyCreateResponseOptions(options, createResponse)\n\t}\n\n\t// Add MCP tools if configured and get samples prompt\n\tmcpSamplesPrompt, err := ast.applyMCPTools(ctx, options, createResponse)\n\tif err != nil {\n\t\treturn nil, \"\", fmt.Errorf(\"failed to apply MCP tools: %w\", err)\n\t}\n\n\treturn options, mcpSamplesPrompt, nil\n}\n\n// applyAssistantOptions applies options from ast.Options to CompletionOptions\n// ast.Options can contain any OpenAI API parameters (temperature, top_p, stop, etc.)\n// Returns error if any option validation fails (e.g., invalid JSON Schema)\nfunc (ast *Assistant) applyAssistantOptions(options *context.CompletionOptions) error {\n\tif ast.Options == nil {\n\t\treturn nil\n\t}\n\n\t// Temperature\n\tif v, ok := ast.Options[\"temperature\"].(float64); ok {\n\t\toptions.Temperature = &v\n\t}\n\n\t// MaxTokens\n\tif v, ok := ast.Options[\"max_tokens\"].(float64); ok {\n\t\tintVal := int(v)\n\t\toptions.MaxTokens = &intVal\n\t} else if v, ok := ast.Options[\"max_tokens\"].(int); ok {\n\t\toptions.MaxTokens = &v\n\t}\n\n\t// MaxCompletionTokens\n\tif v, ok := ast.Options[\"max_completion_tokens\"].(float64); ok {\n\t\tintVal := int(v)\n\t\toptions.MaxCompletionTokens = &intVal\n\t} else if v, ok := ast.Options[\"max_completion_tokens\"].(int); ok {\n\t\toptions.MaxCompletionTokens = &v\n\t}\n\n\t// TopP\n\tif v, ok := ast.Options[\"top_p\"].(float64); ok {\n\t\toptions.TopP = &v\n\t}\n\n\t// N (number of choices)\n\tif v, ok := ast.Options[\"n\"].(float64); ok {\n\t\tintVal := int(v)\n\t\toptions.N = &intVal\n\t} else if v, ok := ast.Options[\"n\"].(int); ok {\n\t\toptions.N = &v\n\t}\n\n\t// Stop sequences (can be string or []string)\n\tif v, ok := ast.Options[\"stop\"]; ok {\n\t\toptions.Stop = v\n\t}\n\n\t// PresencePenalty\n\tif v, ok := ast.Options[\"presence_penalty\"].(float64); ok {\n\t\toptions.PresencePenalty = &v\n\t}\n\n\t// FrequencyPenalty\n\tif v, ok := ast.Options[\"frequency_penalty\"].(float64); ok {\n\t\toptions.FrequencyPenalty = &v\n\t}\n\n\t// LogitBias\n\tif v, ok := ast.Options[\"logit_bias\"].(map[string]interface{}); ok {\n\t\tlogitBias := make(map[string]float64)\n\t\tfor key, val := range v {\n\t\t\tif fval, ok := val.(float64); ok {\n\t\t\t\tlogitBias[key] = fval\n\t\t\t}\n\t\t}\n\t\tif len(logitBias) > 0 {\n\t\t\toptions.LogitBias = logitBias\n\t\t}\n\t}\n\n\t// User\n\tif v, ok := ast.Options[\"user\"].(string); ok {\n\t\toptions.User = v\n\t}\n\n\t// ResponseFormat\n\t// @todo: Assistant should have a default response format\n\tif v, ok := ast.Options[\"response_format\"]; ok {\n\t\t// Try to convert to *context.ResponseFormat\n\t\tif rf, ok := v.(*context.ResponseFormat); ok {\n\t\t\t// Validate JSONSchema if present - reject if invalid\n\t\t\tif rf.JSONSchema != nil && rf.JSONSchema.Schema != nil {\n\t\t\t\tif err := json.ValidateSchema(rf.JSONSchema.Schema); err != nil {\n\t\t\t\t\treturn fmt.Errorf(\"invalid JSON Schema in response_format: %w\", err)\n\t\t\t\t}\n\t\t\t}\n\t\t\toptions.ResponseFormat = rf\n\t\t} else if rfMap, ok := v.(map[string]interface{}); ok {\n\t\t\t// Handle legacy map[string]interface{} format\n\t\t\t// Try to parse into ResponseFormat struct\n\t\t\trf := &context.ResponseFormat{}\n\n\t\t\t// Parse type\n\t\t\tif typeStr, ok := rfMap[\"type\"].(string); ok {\n\t\t\t\trf.Type = context.ResponseFormatType(typeStr)\n\t\t\t}\n\n\t\t\t// Parse json_schema if present\n\t\t\tif jsonSchemaMap, ok := rfMap[\"json_schema\"].(map[string]interface{}); ok {\n\t\t\t\tjsonSchema := &context.JSONSchema{}\n\n\t\t\t\tif name, ok := jsonSchemaMap[\"name\"].(string); ok {\n\t\t\t\t\tjsonSchema.Name = name\n\t\t\t\t}\n\t\t\t\tif desc, ok := jsonSchemaMap[\"description\"].(string); ok {\n\t\t\t\t\tjsonSchema.Description = desc\n\t\t\t\t}\n\t\t\t\tif schema, ok := jsonSchemaMap[\"schema\"]; ok {\n\t\t\t\t\t// Validate schema format - reject if invalid\n\t\t\t\t\tif err := json.ValidateSchema(schema); err != nil {\n\t\t\t\t\t\treturn fmt.Errorf(\"invalid JSON Schema in response_format: %w\", err)\n\t\t\t\t\t}\n\t\t\t\t\tjsonSchema.Schema = schema\n\t\t\t\t}\n\t\t\t\tif strict, ok := jsonSchemaMap[\"strict\"].(bool); ok {\n\t\t\t\t\tjsonSchema.Strict = &strict\n\t\t\t\t}\n\n\t\t\t\trf.JSONSchema = jsonSchema\n\t\t\t}\n\n\t\t\toptions.ResponseFormat = rf\n\t\t}\n\t}\n\n\t// Seed\n\tif v, ok := ast.Options[\"seed\"].(float64); ok {\n\t\tintVal := int(v)\n\t\toptions.Seed = &intVal\n\t} else if v, ok := ast.Options[\"seed\"].(int); ok {\n\t\toptions.Seed = &v\n\t}\n\n\t// Tools\n\tif v, ok := ast.Options[\"tools\"].([]interface{}); ok {\n\t\ttools := make([]map[string]interface{}, 0, len(v))\n\t\tfor _, tool := range v {\n\t\t\tif toolMap, ok := tool.(map[string]interface{}); ok {\n\t\t\t\ttools = append(tools, toolMap)\n\t\t\t}\n\t\t}\n\t\tif len(tools) > 0 {\n\t\t\toptions.Tools = tools\n\t\t}\n\t}\n\n\t// ToolChoice\n\tif v, ok := ast.Options[\"tool_choice\"]; ok {\n\t\toptions.ToolChoice = v\n\t}\n\n\t// Stream\n\tif v, ok := ast.Options[\"stream\"].(bool); ok {\n\t\toptions.Stream = &v\n\t}\n\n\treturn nil\n}\n\n// applyContextOptions applies options from ctx to CompletionOptions\n// ctx provides Route and Metadata for CUI context\nfunc (ast *Assistant) applyContextOptions(options *context.CompletionOptions, ctx *context.Context) {\n\t// Set Route and Metadata from ctx\n\toptions.Route = ctx.Route\n\toptions.Metadata = ctx.Metadata\n\n\t// Set Uses configurations (assistant.Uses has priority over global settings)\n\t// These can be overridden by createResponse\n\toptions.Uses = ast.getUses()\n}\n\n// applyCreateResponseOptions applies options from createResponse to CompletionOptions\n// createResponse takes highest priority and overrides any previous settings\nfunc (ast *Assistant) applyCreateResponseOptions(options *context.CompletionOptions, createResponse *context.HookCreateResponse) {\n\t// Audio configuration\n\tif createResponse.Audio != nil {\n\t\toptions.Audio = createResponse.Audio\n\t}\n\n\t// Temperature\n\tif createResponse.Temperature != nil {\n\t\toptions.Temperature = createResponse.Temperature\n\t}\n\n\t// MaxTokens\n\tif createResponse.MaxTokens != nil {\n\t\toptions.MaxTokens = createResponse.MaxTokens\n\t}\n\n\t// MaxCompletionTokens\n\tif createResponse.MaxCompletionTokens != nil {\n\t\toptions.MaxCompletionTokens = createResponse.MaxCompletionTokens\n\t}\n\n\t// Route\n\tif createResponse.Route != \"\" {\n\t\toptions.Route = createResponse.Route\n\t}\n\n\t// Metadata (merge with existing)\n\tif createResponse.Metadata != nil {\n\t\tif options.Metadata == nil {\n\t\t\toptions.Metadata = createResponse.Metadata\n\t\t} else {\n\t\t\t// Merge: createResponse.Metadata overrides existing\n\t\t\tfor key, value := range createResponse.Metadata {\n\t\t\t\toptions.Metadata[key] = value\n\t\t\t}\n\t\t}\n\t}\n\n\t// Uses configuration (merge with existing)\n\t// createResponse.Uses has highest priority and overrides existing Uses\n\tif createResponse.Uses != nil {\n\t\tif options.Uses == nil {\n\t\t\toptions.Uses = createResponse.Uses\n\t\t} else {\n\t\t\t// Merge: createResponse.Uses overrides existing (only non-empty fields)\n\t\t\tif createResponse.Uses.Vision != \"\" {\n\t\t\t\toptions.Uses.Vision = createResponse.Uses.Vision\n\t\t\t}\n\t\t\tif createResponse.Uses.Audio != \"\" {\n\t\t\t\toptions.Uses.Audio = createResponse.Uses.Audio\n\t\t\t}\n\t\t\tif createResponse.Uses.Search != \"\" {\n\t\t\t\toptions.Uses.Search = createResponse.Uses.Search\n\t\t\t}\n\t\t\tif createResponse.Uses.Fetch != \"\" {\n\t\t\t\toptions.Uses.Fetch = createResponse.Uses.Fetch\n\t\t\t}\n\t\t}\n\t}\n\n\t// ForceUses configuration\n\t// If hook specifies ForceUses, it takes priority\n\tif createResponse.ForceUses != nil {\n\t\toptions.ForceUses = *createResponse.ForceUses\n\t}\n}\n\n// getUses get the Uses configuration with priority: assistant.Uses > global settings\n// Note: createResponse.Uses (applied in applyCreateResponseOptions) has even higher priority\n// getUses returns the Uses config for this assistant\n// Note: The config is already merged with global config during loading (loadMap)\nfunc (ast *Assistant) getUses() *context.Uses {\n\treturn ast.Uses\n}\n\n// applyMCPTools adds MCP tools to completion options and returns samples prompt\n// Returns (samplesPrompt, error)\nfunc (ast *Assistant) applyMCPTools(ctx *context.Context, options *context.CompletionOptions, createResponse *context.HookCreateResponse) (string, error) {\n\n\t// Priority 1: Check if hook provides MCP servers\n\tif createResponse != nil && len(createResponse.MCPServers) > 0 {\n\t\treturn ast.buildAndApplyMCPTools(ctx, options, createResponse)\n\t}\n\n\t// Priority 2: Check if assistant has MCP config\n\tif ast.MCP != nil && len(ast.MCP.Servers) > 0 {\n\t\treturn ast.buildAndApplyMCPTools(ctx, options, nil)\n\t}\n\n\t// No MCP config\n\treturn \"\", nil\n}\n\n// buildAndApplyMCPTools builds MCP tools and applies them to options\nfunc (ast *Assistant) buildAndApplyMCPTools(ctx *context.Context, options *context.CompletionOptions, createResponse *context.HookCreateResponse) (string, error) {\n\t// Build MCP tools and get samples prompt\n\tmcpTools, samplesPrompt, err := ast.buildMCPTools(ctx, createResponse)\n\tif err != nil {\n\t\treturn \"\", fmt.Errorf(\"failed to build MCP tools: %w\", err)\n\t}\n\n\t// Convert mcpTools to map format for CompletionOptions.Tools\n\tif len(mcpTools) > 0 {\n\t\ttoolMaps := make([]map[string]interface{}, len(mcpTools))\n\t\tfor i, tool := range mcpTools {\n\t\t\ttoolMaps[i] = map[string]interface{}{\n\t\t\t\t\"type\": \"function\",\n\t\t\t\t\"function\": map[string]interface{}{\n\t\t\t\t\t\"name\": tool.Name,\n\t\t\t\t\t\"description\": tool.Description,\n\t\t\t\t\t\"parameters\": tool.Parameters,\n\t\t\t\t},\n\t\t\t}\n\t\t}\n\n\t\t// Add MCP tools to existing tools (append to preserve existing tools)\n\t\tif options.Tools == nil {\n\t\t\toptions.Tools = toolMaps\n\t\t} else {\n\t\t\toptions.Tools = append(options.Tools, toolMaps...)\n\t\t}\n\t}\n\n\treturn samplesPrompt, nil\n}\n" }, { "path": "agent/assistant/build_content.go", "content": "package assistant\n\nimport (\n\t\"fmt\"\n\n\t\"github.com/yaoapp/yao/agent/content\"\n\t\"github.com/yaoapp/yao/agent/content/text\"\n\tcontentTypes \"github.com/yaoapp/yao/agent/content/types\"\n\t\"github.com/yaoapp/yao/agent/context\"\n)\n\n// BuildContent processes messages through Vision function to convert extended content types\n// (file, data) to standard LLM-compatible types (text, image_url, input_audio)\n//\n// This should be called after BuildRequest and before executing LLM call\nfunc (ast *Assistant) BuildContent(ctx *context.Context, messages []context.Message, options *context.CompletionOptions, opts *context.Options) ([]context.Message, error) {\n\t// Skip complex content parsing if requested (for internal calls like needsearch)\n\t// Still convert file attachments to raw text\n\tif opts != nil && opts.Skip != nil && opts.Skip.ContentParsing {\n\t\treturn convertFilesToText(ctx, messages), nil\n\t}\n\n\t// Set AssistantID in context for file info tracking in Space\n\t// This ensures hooks can access file information using the correct namespace\n\tif ctx.AssistantID == \"\" {\n\t\tctx.AssistantID = ast.ID\n\t}\n\n\t// Get connector and capabilities\n\tconnector, capabilities, err := ast.GetConnector(ctx, opts)\n\tif err != nil {\n\t\treturn nil, fmt.Errorf(\"failed to get connector: %w\", err)\n\t}\n\n\t// Build parse options\n\tparseOptions := &contentTypes.Options{\n\t\tCapabilities: capabilities,\n\t\tCompletionOptions: options,\n\t\tConnector: connector,\n\t\tStreamOptions: options.StreamOptions,\n\t}\n\n\tcontentMessages, referenceContext, err := content.ParseUserInput(ctx, messages, parseOptions)\n\tif err != nil {\n\t\treturn nil, fmt.Errorf(\"failed to parse content: %w\", err)\n\t}\n\n\t// Inject reference context into messages\n\tif referenceContext != nil {\n\t\tcontentMessages = ast.injectSearchContext(contentMessages, referenceContext)\n\t}\n\n\treturn contentMessages, nil\n}\n\n// convertFilesToText converts file attachments in messages to raw text\n// Used when SkipContentParsing is enabled - simple text extraction without vision/PDF processing\nfunc convertFilesToText(ctx *context.Context, messages []context.Message) []context.Message {\n\tresult := make([]context.Message, 0, len(messages))\n\ttextHandler := text.New(nil)\n\n\tfor _, msg := range messages {\n\t\t// Only process user messages\n\t\tif msg.Role != context.RoleUser {\n\t\t\tresult = append(result, msg)\n\t\t\tcontinue\n\t\t}\n\n\t\t// Handle content parts\n\t\tparts, ok := msg.Content.([]context.ContentPart)\n\t\tif !ok {\n\t\t\t// Try []interface{} (from history/JSON)\n\t\t\tif iparts, ok := msg.Content.([]interface{}); ok {\n\t\t\t\tparts = convertInterfaceToParts(iparts)\n\t\t\t}\n\t\t}\n\n\t\tif len(parts) == 0 {\n\t\t\tresult = append(result, msg)\n\t\t\tcontinue\n\t\t}\n\n\t\t// Convert file parts to text\n\t\tnewParts := make([]context.ContentPart, 0, len(parts))\n\t\tfor _, part := range parts {\n\t\t\tswitch part.Type {\n\t\t\tcase context.ContentFile:\n\t\t\t\t// Convert file to raw text\n\t\t\t\tif part.File != nil && part.File.URL != \"\" {\n\t\t\t\t\ttextPart, _, err := textHandler.ParseRaw(ctx, part)\n\t\t\t\t\tif err == nil {\n\t\t\t\t\t\tnewParts = append(newParts, textPart)\n\t\t\t\t\t\tcontinue\n\t\t\t\t\t}\n\t\t\t\t}\n\t\t\t\tnewParts = append(newParts, part)\n\n\t\t\tcase context.ContentImageURL:\n\t\t\t\t// Skip images - cannot convert to text without vision\n\t\t\t\tcontinue\n\n\t\t\tdefault:\n\t\t\t\tnewParts = append(newParts, part)\n\t\t\t}\n\t\t}\n\n\t\tnewMsg := msg\n\t\tnewMsg.Content = newParts\n\t\tresult = append(result, newMsg)\n\t}\n\n\treturn result\n}\n\n// convertInterfaceToParts converts []interface{} to []ContentPart for file extraction\nfunc convertInterfaceToParts(items []interface{}) []context.ContentPart {\n\tparts := make([]context.ContentPart, 0, len(items))\n\tfor _, item := range items {\n\t\tm, ok := item.(map[string]interface{})\n\t\tif !ok {\n\t\t\tcontinue\n\t\t}\n\n\t\ttypeStr, _ := m[\"type\"].(string)\n\t\tpart := context.ContentPart{\n\t\t\tType: context.ContentPartType(typeStr),\n\t\t}\n\n\t\tswitch typeStr {\n\t\tcase \"text\":\n\t\t\tif t, ok := m[\"text\"].(string); ok {\n\t\t\t\tpart.Text = t\n\t\t\t}\n\t\tcase \"file\":\n\t\t\tif fileData, ok := m[\"file\"].(map[string]interface{}); ok {\n\t\t\t\tpart.File = &context.FileAttachment{}\n\t\t\t\tif url, ok := fileData[\"url\"].(string); ok {\n\t\t\t\t\tpart.File.URL = url\n\t\t\t\t}\n\t\t\t\tif filename, ok := fileData[\"filename\"].(string); ok {\n\t\t\t\t\tpart.File.Filename = filename\n\t\t\t\t}\n\t\t\t}\n\t\tcase \"image_url\":\n\t\t\tpart.Type = context.ContentImageURL\n\t\tdefault:\n\t\t\tcontinue\n\t\t}\n\n\t\tparts = append(parts, part)\n\t}\n\treturn parts\n}\n" }, { "path": "agent/assistant/build_mcp_test.go", "content": "package assistant_test\n\nimport (\n\t\"testing\"\n\n\t\"github.com/yaoapp/yao/agent/assistant\"\n\t\"github.com/yaoapp/yao/agent/context\"\n\t\"github.com/yaoapp/yao/agent/testutils\"\n)\n\n// TestBuildRequest_MCP tests MCP tool integration in BuildRequest\nfunc TestBuildRequest_MCP(t *testing.T) {\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\tagent, err := assistant.Get(\"tests.mcptest\")\n\tif err != nil {\n\t\tt.Fatalf(\"Failed to get tests.mcptest assistant: %s\", err.Error())\n\t}\n\n\tctx := newTestContext(\"chat-test-mcp\", \"tests.mcptest\")\n\n\tt.Run(\"MCPToolsLoaded\", func(t *testing.T) {\n\t\tinputMessages := []context.Message{{Role: context.RoleUser, Content: \"test mcp tools\"}}\n\n\t\t// Build LLM request\n\t\t_, options, err := agent.BuildRequest(ctx, inputMessages, nil)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to build LLM request: %s\", err.Error())\n\t\t}\n\n\t\t// Verify that tools are loaded\n\t\tif options.Tools == nil {\n\t\t\tt.Fatal(\"Expected tools to be loaded, got nil\")\n\t\t}\n\n\t\tif len(options.Tools) == 0 {\n\t\t\tt.Fatal(\"Expected at least some MCP tools, got empty list\")\n\t\t}\n\n\t\t// Count MCP tools (should be filtered to only ping and echo)\n\t\tmcpToolCount := 0\n\t\tvar toolNames []string\n\t\tfor _, toolMap := range options.Tools {\n\t\t\tfn, ok := toolMap[\"function\"].(map[string]interface{})\n\t\t\tif !ok {\n\t\t\t\tcontinue\n\t\t\t}\n\t\t\tname, ok := fn[\"name\"].(string)\n\t\t\tif ok {\n\t\t\t\ttoolNames = append(toolNames, name)\n\t\t\t\tmcpToolCount++\n\t\t\t}\n\t\t}\n\n\t\tt.Logf(\"Found %d MCP tools: %v\", mcpToolCount, toolNames)\n\n\t\t// Verify tool count (should be exactly 2: ping and echo)\n\t\tif mcpToolCount != 2 {\n\t\t\tt.Errorf(\"Expected 2 MCP tools (ping, echo), got %d: %v\", mcpToolCount, toolNames)\n\t\t}\n\n\t\t// Verify specific tools exist\n\t\thasEchoPing := false\n\t\thasEchoEcho := false\n\t\tfor _, name := range toolNames {\n\t\t\tif name == \"echo__ping\" {\n\t\t\t\thasEchoPing = true\n\t\t\t}\n\t\t\tif name == \"echo__echo\" {\n\t\t\t\thasEchoEcho = true\n\t\t\t}\n\t\t}\n\n\t\tif !hasEchoPing {\n\t\t\tt.Error(\"Expected 'echo__ping' tool to be present\")\n\t\t}\n\t\tif !hasEchoEcho {\n\t\t\tt.Error(\"Expected 'echo__echo' tool to be present\")\n\t\t}\n\n\t\t// Verify that 'status' tool is NOT included (filtered out)\n\t\tfor _, name := range toolNames {\n\t\t\tif name == \"echo__status\" {\n\t\t\t\tt.Error(\"Tool 'echo__status' should be filtered out but was found\")\n\t\t\t}\n\t\t}\n\n\t\tt.Log(\"✓ MCP tools loaded and filtered correctly\")\n\t})\n\n\tt.Run(\"MCPSamplesPrompt\", func(t *testing.T) {\n\t\tinputMessages := []context.Message{{Role: context.RoleUser, Content: \"test mcp samples\"}}\n\n\t\t// Build LLM request\n\t\tfinalMessages, _, err := agent.BuildRequest(ctx, inputMessages, nil)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to build LLM request: %s\", err.Error())\n\t\t}\n\n\t\t// Check if messages contain MCP samples prompt\n\t\t// The samples prompt should be added as a system message\n\t\thasMCPSamples := false\n\t\tfor _, msg := range finalMessages {\n\t\t\tif msg.Role == context.RoleSystem {\n\t\t\t\tif content, ok := msg.Content.(string); ok {\n\t\t\t\t\tif len(content) > 50 &&\n\t\t\t\t\t\t(contains(content, \"MCP Tool Usage Examples\") ||\n\t\t\t\t\t\t\tcontains(content, \"echo.ping\") ||\n\t\t\t\t\t\t\tcontains(content, \"echo.echo\")) {\n\t\t\t\t\t\thasMCPSamples = true\n\t\t\t\t\t\tt.Logf(\"Found MCP samples prompt (length: %d chars)\", len(content))\n\t\t\t\t\t\tbreak\n\t\t\t\t\t}\n\t\t\t\t}\n\t\t\t}\n\t\t}\n\n\t\t// Note: samples may not exist for echo tools, so this is informational\n\t\tif hasMCPSamples {\n\t\t\tt.Log(\"✓ MCP samples prompt included in messages\")\n\t\t} else {\n\t\t\tt.Log(\"ℹ No MCP samples prompt found (may not have sample files)\")\n\t\t}\n\t})\n\n\tt.Run(\"MCPToolNameFormat\", func(t *testing.T) {\n\t\tinputMessages := []context.Message{{Role: context.RoleUser, Content: \"test tool format\"}}\n\n\t\t// Build LLM request\n\t\t_, options, err := agent.BuildRequest(ctx, inputMessages, nil)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to build LLM request: %s\", err.Error())\n\t\t}\n\n\t\t// Verify tool name format: server_id.tool_name\n\t\tfor _, toolMap := range options.Tools {\n\t\t\tfn, ok := toolMap[\"function\"].(map[string]interface{})\n\t\t\tif !ok {\n\t\t\t\tcontinue\n\t\t\t}\n\t\t\tname, ok := fn[\"name\"].(string)\n\t\t\tif ok {\n\t\t\t\t// Parse tool name\n\t\t\t\tserverID, toolName, ok := assistant.ParseMCPToolName(name)\n\t\t\t\tif !ok {\n\t\t\t\t\tt.Errorf(\"Tool name '%s' is not in correct format (server_id.tool_name)\", name)\n\t\t\t\t\tcontinue\n\t\t\t\t}\n\n\t\t\t\t// Verify server ID\n\t\t\t\tif serverID != \"echo\" {\n\t\t\t\t\tt.Errorf(\"Expected server_id 'echo', got '%s' for tool '%s'\", serverID, name)\n\t\t\t\t}\n\n\t\t\t\t// Verify tool name is either ping or echo\n\t\t\t\tif toolName != \"ping\" && toolName != \"echo\" {\n\t\t\t\t\tt.Errorf(\"Expected tool name 'ping' or 'echo', got '%s'\", toolName)\n\t\t\t\t}\n\n\t\t\t\tt.Logf(\"✓ Tool name format correct: %s → (%s, %s)\", name, serverID, toolName)\n\t\t\t}\n\t\t}\n\t})\n\n\tt.Run(\"MCPToolSchema\", func(t *testing.T) {\n\t\tinputMessages := []context.Message{{Role: context.RoleUser, Content: \"test tool schema\"}}\n\n\t\t// Build LLM request\n\t\t_, options, err := agent.BuildRequest(ctx, inputMessages, nil)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to build LLM request: %s\", err.Error())\n\t\t}\n\n\t\t// Verify tool schema structure\n\t\tfor _, toolMap := range options.Tools {\n\t\t\t// Verify type field\n\t\t\tif toolType, ok := toolMap[\"type\"].(string); !ok || toolType != \"function\" {\n\t\t\t\tt.Errorf(\"Expected tool type 'function', got: %v\", toolMap[\"type\"])\n\t\t\t}\n\n\t\t\t// Verify function field exists\n\t\t\tfn, ok := toolMap[\"function\"].(map[string]interface{})\n\t\t\tif !ok {\n\t\t\t\tt.Error(\"Tool missing 'function' field or wrong type\")\n\t\t\t\tcontinue\n\t\t\t}\n\n\t\t\t// Verify required fields\n\t\t\tif _, hasName := fn[\"name\"]; !hasName {\n\t\t\t\tt.Error(\"Tool function missing 'name' field\")\n\t\t\t}\n\t\t\tif _, hasDesc := fn[\"description\"]; !hasDesc {\n\t\t\t\tt.Error(\"Tool function missing 'description' field\")\n\t\t\t}\n\t\t\tif _, hasParams := fn[\"parameters\"]; !hasParams {\n\t\t\t\tt.Error(\"Tool function missing 'parameters' field\")\n\t\t\t}\n\n\t\t\tt.Logf(\"✓ Tool schema valid: %v\", fn[\"name\"])\n\t\t}\n\t})\n\n\tt.Run(\"MCPHookOverride\", func(t *testing.T) {\n\t\t// Test that hook can override MCP servers\n\t\t// Use tests.mcptest-hook which has a create hook that returns only [\"ping\"]\n\t\thookAgent, err := assistant.Get(\"tests.mcptest-hook\")\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to get tests.mcptest-hook assistant: %s\", err.Error())\n\t\t}\n\n\t\thookCtx := newTestContext(\"chat-test-mcp-hook\", \"tests.mcptest-hook\")\n\t\tinputMessages := []context.Message{{Role: context.RoleUser, Content: \"test hook override\"}}\n\n\t\t// Call create hook to get createResponse\n\t\tvar createResponse *context.HookCreateResponse\n\t\tif hookAgent.HookScript != nil {\n\t\t\tcreateResponse, _, err = hookAgent.HookScript.Create(hookCtx, inputMessages, &context.Options{})\n\t\t\tif err != nil {\n\t\t\t\tt.Fatalf(\"Failed to call create hook: %s\", err.Error())\n\t\t\t}\n\n\t\t\tt.Logf(\"Create hook response: %+v\", createResponse)\n\t\t\tif createResponse != nil && len(createResponse.MCPServers) > 0 {\n\t\t\t\tt.Logf(\"Hook MCP servers: %+v\", createResponse.MCPServers)\n\t\t\t}\n\t\t} else {\n\t\t\tt.Fatal(\"Expected hookAgent to have Script/hook configured\")\n\t\t}\n\n\t\t// Build LLM request with create hook response\n\t\t_, options, err := hookAgent.BuildRequest(hookCtx, inputMessages, createResponse)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to build LLM request: %s\", err.Error())\n\t\t}\n\n\t\t// Verify that tools are loaded\n\t\tif options.Tools == nil {\n\t\t\tt.Fatal(\"Expected tools to be loaded, got nil\")\n\t\t}\n\n\t\t// Count MCP tools\n\t\tmcpToolCount := 0\n\t\tvar toolNames []string\n\t\tfor _, toolMap := range options.Tools {\n\t\t\tfn, ok := toolMap[\"function\"].(map[string]interface{})\n\t\t\tif !ok {\n\t\t\t\tcontinue\n\t\t\t}\n\t\t\tname, ok := fn[\"name\"].(string)\n\t\t\tif ok {\n\t\t\t\ttoolNames = append(toolNames, name)\n\t\t\t\tmcpToolCount++\n\t\t\t}\n\t\t}\n\n\t\tt.Logf(\"Found %d MCP tools after hook override: %v\", mcpToolCount, toolNames)\n\n\t\t// Verify tool count (hook should override to only 1: ping)\n\t\tif mcpToolCount != 1 {\n\t\t\tt.Errorf(\"Expected 1 MCP tool (ping only), got %d: %v\", mcpToolCount, toolNames)\n\t\t}\n\n\t\t// Verify only ping tool exists\n\t\thasEchoPing := false\n\t\thasEchoEcho := false\n\t\tfor _, name := range toolNames {\n\t\t\tif name == \"echo__ping\" {\n\t\t\t\thasEchoPing = true\n\t\t\t}\n\t\t\tif name == \"echo__echo\" {\n\t\t\t\thasEchoEcho = true\n\t\t\t}\n\t\t}\n\n\t\tif !hasEchoPing {\n\t\t\tt.Error(\"Expected 'echo__ping' tool to be present\")\n\t\t}\n\t\tif hasEchoEcho {\n\t\t\tt.Error(\"Tool 'echo__echo' should be filtered out by hook override but was found\")\n\t\t}\n\n\t\tt.Log(\"✓ Hook successfully overrode MCP servers configuration\")\n\t})\n}\n\n// Helper function to check if string contains substring\nfunc contains(s, substr string) bool {\n\treturn len(s) >= len(substr) &&\n\t\t(s == substr ||\n\t\t\tlen(s) > len(substr) &&\n\t\t\t\t(s[:len(substr)] == substr ||\n\t\t\t\t\ts[len(s)-len(substr):] == substr ||\n\t\t\t\t\tfindSubstring(s, substr)))\n}\n\nfunc findSubstring(s, substr string) bool {\n\tfor i := 0; i <= len(s)-len(substr); i++ {\n\t\tif s[i:i+len(substr)] == substr {\n\t\t\treturn true\n\t\t}\n\t}\n\treturn false\n}\n" }, { "path": "agent/assistant/build_prompts_test.go", "content": "package assistant_test\n\nimport (\n\tstdContext \"context\"\n\t\"strings\"\n\t\"testing\"\n\n\t\"github.com/stretchr/testify/assert\"\n\t\"github.com/stretchr/testify/require\"\n\t\"github.com/yaoapp/yao/agent/assistant\"\n\t\"github.com/yaoapp/yao/agent/context\"\n\tstore \"github.com/yaoapp/yao/agent/store/types\"\n\t\"github.com/yaoapp/yao/agent/testutils\"\n\t\"github.com/yaoapp/yao/openapi/oauth/types\"\n)\n\n// containsString is a helper to check if a content (string or interface{}) contains a substring\nfunc containsString(content interface{}, substr string) bool {\n\tswitch v := content.(type) {\n\tcase string:\n\t\treturn strings.Contains(v, substr)\n\tdefault:\n\t\treturn false\n\t}\n}\n\n// newPromptTestContext creates a context suitable for prompt testing with Create Hook\nfunc newPromptTestContext(chatID, assistantID string) *context.Context {\n\tauthorized := &types.AuthorizedInfo{\n\t\tSubject: \"test-user\",\n\t\tClientID: \"test-client-id\",\n\t\tUserID: \"test-user-123\",\n\t\tTeamID: \"test-team-456\",\n\t\tTenantID: \"test-tenant-789\",\n\t\tSessionID: \"test-session-id\",\n\t}\n\n\tctx := context.New(stdContext.Background(), authorized, chatID)\n\tctx.AssistantID = assistantID\n\tctx.Locale = \"en-us\"\n\tctx.Theme = \"light\"\n\tctx.Client = context.Client{\n\t\tType: \"web\",\n\t\tUserAgent: \"TestAgent/1.0\",\n\t\tIP: \"127.0.0.1\",\n\t}\n\tctx.Referer = context.RefererAPI\n\tctx.Accept = context.AcceptWebCUI\n\tctx.Metadata = make(map[string]interface{})\n\treturn ctx\n}\n\n// newMinimalTestContext creates a minimal context for testing\n// Use this when you only need specific fields set\nfunc newMinimalTestContext() *context.Context {\n\treturn context.New(stdContext.Background(), nil, \"test-chat\")\n}\n\nfunc TestBuildSystemPromptsIntegration(t *testing.T) {\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\tt.Run(\"AssistantWithLocale\", func(t *testing.T) {\n\t\t// Load an assistant with locales\n\t\tast, err := assistant.Get(\"tests.fullfields\")\n\t\trequire.NoError(t, err)\n\n\t\tctx := newMinimalTestContext()\n\t\tctx.Locale = \"zh-cn\"\n\t\tctx.Authorized = &types.AuthorizedInfo{\n\t\t\tUserID: \"test-user-123\",\n\t\t\tTeamID: \"test-team-456\",\n\t\t}\n\t\tctx.Metadata = map[string]interface{}{\n\t\t\t\"CUSTOM_VAR\": \"custom-value\",\n\t\t\t\"INT_VAR\": 42,\n\t\t\t\"BOOL_VAR\": true,\n\t\t}\n\n\t\t// Build request to test the full flow\n\t\tmessages := []context.Message{\n\t\t\t{Role: context.RoleUser, Content: \"Hello\"},\n\t\t}\n\n\t\tfinalMessages, options, err := ast.BuildRequest(ctx, messages, nil)\n\t\trequire.NoError(t, err)\n\t\trequire.NotNil(t, options)\n\n\t\t// Should have system prompts prepended\n\t\tassert.Greater(t, len(finalMessages), 1)\n\n\t\t// First messages should be system prompts\n\t\thasSystemPrompt := false\n\t\tfor _, msg := range finalMessages {\n\t\t\tif msg.Role == context.RoleSystem {\n\t\t\t\thasSystemPrompt = true\n\t\t\t\tbreak\n\t\t\t}\n\t\t}\n\t\tassert.True(t, hasSystemPrompt, \"Should have system prompts\")\n\t})\n\n\tt.Run(\"DisableGlobalPrompts\", func(t *testing.T) {\n\t\t// Load fullfields assistant which has disable_global_prompts: true\n\t\tast, err := assistant.Get(\"tests.fullfields\")\n\t\trequire.NoError(t, err)\n\t\trequire.True(t, ast.DisableGlobalPrompts)\n\n\t\tctx := newMinimalTestContext()\n\t\tctx.Locale = \"en-us\"\n\n\t\tmessages := []context.Message{\n\t\t\t{Role: context.RoleUser, Content: \"Hello\"},\n\t\t}\n\n\t\tfinalMessages, _, err := ast.BuildRequest(ctx, messages, nil)\n\t\trequire.NoError(t, err)\n\n\t\t// Should still have assistant prompts\n\t\thasSystemPrompt := false\n\t\tfor _, msg := range finalMessages {\n\t\t\tif msg.Role == context.RoleSystem {\n\t\t\t\thasSystemPrompt = true\n\t\t\t\tbreak\n\t\t\t}\n\t\t}\n\t\tassert.True(t, hasSystemPrompt, \"Should have assistant prompts even with global disabled\")\n\t})\n\n\tt.Run(\"MetadataTypeConversion\", func(t *testing.T) {\n\t\tast, err := assistant.Get(\"yaobots\")\n\t\trequire.NoError(t, err)\n\n\t\tctx := newMinimalTestContext()\n\t\tctx.Metadata = map[string]interface{}{\n\t\t\t\"STRING_VAL\": \"hello\",\n\t\t\t\"INT_VAL\": 123,\n\t\t\t\"INT64_VAL\": int64(456),\n\t\t\t\"FLOAT_VAL\": 3.14,\n\t\t\t\"BOOL_TRUE\": true,\n\t\t\t\"BOOL_FALSE\": false,\n\t\t\t\"UINT_VAL\": uint(789),\n\t\t\t\"NIL_VAL\": nil,\n\t\t\t\"EMPTY_VAL\": \"\",\n\t\t\t\"ZERO_INT\": 0,\n\t\t\t\"ZERO_FLOAT\": 0.0,\n\t\t}\n\n\t\tmessages := []context.Message{\n\t\t\t{Role: context.RoleUser, Content: \"Test metadata\"},\n\t\t}\n\n\t\t// This should not panic\n\t\t_, _, err = ast.BuildRequest(ctx, messages, nil)\n\t\trequire.NoError(t, err)\n\t})\n\n\tt.Run(\"AuthorizedInfoPrivacy\", func(t *testing.T) {\n\t\tast, err := assistant.Get(\"yaobots\")\n\t\trequire.NoError(t, err)\n\n\t\tctx := newMinimalTestContext()\n\t\tctx.Authorized = &types.AuthorizedInfo{\n\t\t\tUserID: \"user-123\",\n\t\t\tSubject: \"user@example.com\", // PII - should not be exposed\n\t\t\tTeamID: \"team-456\",\n\t\t\tTenantID: \"tenant-789\",\n\t\t}\n\t\tctx.Client = context.Client{\n\t\t\tType: \"web\",\n\t\t\tIP: \"192.168.1.1\", // Should not be exposed\n\t\t}\n\n\t\tmessages := []context.Message{\n\t\t\t{Role: context.RoleUser, Content: \"Test privacy\"},\n\t\t}\n\n\t\tfinalMessages, _, err := ast.BuildRequest(ctx, messages, nil)\n\t\trequire.NoError(t, err)\n\n\t\t// Check that sensitive info is not in any system prompts\n\t\tfor _, msg := range finalMessages {\n\t\t\tif msg.Role == context.RoleSystem {\n\t\t\t\tassert.NotContains(t, msg.Content, \"user@example.com\", \"Subject should not be in prompts\")\n\t\t\t\tassert.NotContains(t, msg.Content, \"192.168.1.1\", \"IP should not be in prompts\")\n\t\t\t}\n\t\t}\n\t})\n\n\tt.Run(\"ContextVariablesInPrompts\", func(t *testing.T) {\n\t\t// Set up global prompts with variables\n\t\tassistant.SetGlobalPrompts([]store.Prompt{\n\t\t\t{Role: \"system\", Content: \"User ID: $CTX.USER_ID, Team: $CTX.TEAM_ID, Custom: $CTX.MY_VAR\"},\n\t\t})\n\t\tdefer assistant.SetGlobalPrompts(nil)\n\n\t\tast, err := assistant.Get(\"yaobots\")\n\t\trequire.NoError(t, err)\n\n\t\tctx := newMinimalTestContext()\n\t\tctx.Authorized = &types.AuthorizedInfo{\n\t\t\tUserID: \"user-abc\",\n\t\t\tTeamID: \"team-xyz\",\n\t\t}\n\t\tctx.Metadata = map[string]interface{}{\n\t\t\t\"MY_VAR\": \"my-value\",\n\t\t}\n\n\t\tmessages := []context.Message{\n\t\t\t{Role: context.RoleUser, Content: \"Test variables\"},\n\t\t}\n\n\t\tfinalMessages, _, err := ast.BuildRequest(ctx, messages, nil)\n\t\trequire.NoError(t, err)\n\n\t\t// Find the global prompt and verify variables are replaced\n\t\tfound := false\n\t\tfor _, msg := range finalMessages {\n\t\t\tif msg.Role == context.RoleSystem && !found {\n\t\t\t\tif assert.Contains(t, msg.Content, \"User ID: user-abc\") {\n\t\t\t\t\tfound = true\n\t\t\t\t\tassert.Contains(t, msg.Content, \"Team: team-xyz\")\n\t\t\t\t\tassert.Contains(t, msg.Content, \"Custom: my-value\")\n\t\t\t\t}\n\t\t\t}\n\t\t}\n\t\tassert.True(t, found, \"Should find global prompt with replaced variables\")\n\t})\n\n\tt.Run(\"SystemVariablesReplacement\", func(t *testing.T) {\n\t\t// Set up global prompts with $SYS.* variables\n\t\tassistant.SetGlobalPrompts([]store.Prompt{\n\t\t\t{Role: \"system\", Content: \"Time: $SYS.TIME, Date: $SYS.DATE, Datetime: $SYS.DATETIME, Weekday: $SYS.WEEKDAY\"},\n\t\t})\n\t\tdefer assistant.SetGlobalPrompts(nil)\n\n\t\tast, err := assistant.Get(\"yaobots\")\n\t\trequire.NoError(t, err)\n\n\t\tctx := newMinimalTestContext()\n\n\t\tmessages := []context.Message{\n\t\t\t{Role: context.RoleUser, Content: \"Test system variables\"},\n\t\t}\n\n\t\tfinalMessages, _, err := ast.BuildRequest(ctx, messages, nil)\n\t\trequire.NoError(t, err)\n\n\t\t// Find the global prompt and verify $SYS.* variables are replaced\n\t\tfound := false\n\t\tfor _, msg := range finalMessages {\n\t\t\tif msg.Role == context.RoleSystem {\n\t\t\t\t// Should NOT contain $SYS. prefix (variables should be replaced)\n\t\t\t\tif !assert.NotContains(t, msg.Content, \"$SYS.TIME\") {\n\t\t\t\t\tcontinue\n\t\t\t\t}\n\t\t\t\tif !assert.NotContains(t, msg.Content, \"$SYS.DATE\") {\n\t\t\t\t\tcontinue\n\t\t\t\t}\n\t\t\t\tif !assert.NotContains(t, msg.Content, \"$SYS.DATETIME\") {\n\t\t\t\t\tcontinue\n\t\t\t\t}\n\t\t\t\tif !assert.NotContains(t, msg.Content, \"$SYS.WEEKDAY\") {\n\t\t\t\t\tcontinue\n\t\t\t\t}\n\n\t\t\t\t// Should contain \"Time:\", \"Date:\", etc. with actual values\n\t\t\t\tassert.Contains(t, msg.Content, \"Time:\")\n\t\t\t\tassert.Contains(t, msg.Content, \"Date:\")\n\t\t\t\tassert.Contains(t, msg.Content, \"Datetime:\")\n\t\t\t\tassert.Contains(t, msg.Content, \"Weekday:\")\n\t\t\t\tfound = true\n\t\t\t\tbreak\n\t\t\t}\n\t\t}\n\t\tassert.True(t, found, \"Should find global prompt with replaced $SYS.* variables\")\n\t})\n\n\tt.Run(\"EnvVariablesReplacement\", func(t *testing.T) {\n\t\t// Set test environment variable\n\t\tt.Setenv(\"TEST_PROMPT_VAR\", \"env-test-value\")\n\n\t\t// Set up global prompts with $ENV.* variables\n\t\tassistant.SetGlobalPrompts([]store.Prompt{\n\t\t\t{Role: \"system\", Content: \"Env Value: $ENV.TEST_PROMPT_VAR, Not Exist: $ENV.NOT_EXIST_VAR_XYZ\"},\n\t\t})\n\t\tdefer assistant.SetGlobalPrompts(nil)\n\n\t\tast, err := assistant.Get(\"yaobots\")\n\t\trequire.NoError(t, err)\n\n\t\tctx := newMinimalTestContext()\n\n\t\tmessages := []context.Message{\n\t\t\t{Role: context.RoleUser, Content: \"Test env variables\"},\n\t\t}\n\n\t\tfinalMessages, _, err := ast.BuildRequest(ctx, messages, nil)\n\t\trequire.NoError(t, err)\n\n\t\t// Find the global prompt and verify $ENV.* variables are replaced\n\t\tfound := false\n\t\tfor _, msg := range finalMessages {\n\t\t\tif msg.Role == context.RoleSystem {\n\t\t\t\t// Should NOT contain $ENV. prefix for existing vars\n\t\t\t\tif !assert.NotContains(t, msg.Content, \"$ENV.TEST_PROMPT_VAR\") {\n\t\t\t\t\tcontinue\n\t\t\t\t}\n\t\t\t\t// Should contain the actual env value\n\t\t\t\tassert.Contains(t, msg.Content, \"Env Value: env-test-value\")\n\t\t\t\t// Non-existent env var should be replaced with empty string\n\t\t\t\tassert.Contains(t, msg.Content, \"Not Exist: \")\n\t\t\t\tassert.NotContains(t, msg.Content, \"$ENV.NOT_EXIST_VAR_XYZ\")\n\t\t\t\tfound = true\n\t\t\t\tbreak\n\t\t\t}\n\t\t}\n\t\tassert.True(t, found, \"Should find global prompt with replaced $ENV.* variables\")\n\t})\n\n\tt.Run(\"AllVariableTypesReplacement\", func(t *testing.T) {\n\t\t// Set test environment variable\n\t\tt.Setenv(\"TEST_APP_NAME\", \"MyTestApp\")\n\n\t\t// Set up global prompts with all variable types\n\t\tassistant.SetGlobalPrompts([]store.Prompt{\n\t\t\t{Role: \"system\", Content: `System Info:\n- Time: $SYS.TIME\n- Date: $SYS.DATE\n- App: $ENV.TEST_APP_NAME\n- User: $CTX.USER_ID\n- Custom: $CTX.CUSTOM_KEY\n- Assistant: $CTX.ASSISTANT_NAME`},\n\t\t})\n\t\tdefer assistant.SetGlobalPrompts(nil)\n\n\t\tast, err := assistant.Get(\"yaobots\")\n\t\trequire.NoError(t, err)\n\n\t\tctx := newMinimalTestContext()\n\t\tctx.Authorized = &types.AuthorizedInfo{\n\t\t\tUserID: \"all-vars-user\",\n\t\t}\n\t\tctx.Metadata = map[string]interface{}{\n\t\t\t\"CUSTOM_KEY\": \"custom-value-123\",\n\t\t}\n\n\t\tmessages := []context.Message{\n\t\t\t{Role: context.RoleUser, Content: \"Test all variables\"},\n\t\t}\n\n\t\tfinalMessages, _, err := ast.BuildRequest(ctx, messages, nil)\n\t\trequire.NoError(t, err)\n\n\t\t// Find the global prompt and verify ALL variable types are replaced\n\t\tfound := false\n\t\tfor _, msg := range finalMessages {\n\t\t\tif msg.Role == context.RoleSystem && !found {\n\t\t\t\tcontent := msg.Content\n\n\t\t\t\t// Check $SYS.* replaced\n\t\t\t\tif assert.NotContains(t, content, \"$SYS.TIME\") &&\n\t\t\t\t\tassert.NotContains(t, content, \"$SYS.DATE\") {\n\n\t\t\t\t\t// Check $ENV.* replaced\n\t\t\t\t\tassert.NotContains(t, content, \"$ENV.TEST_APP_NAME\")\n\t\t\t\t\tassert.Contains(t, content, \"App: MyTestApp\")\n\n\t\t\t\t\t// Check $CTX.* replaced\n\t\t\t\t\tassert.NotContains(t, content, \"$CTX.USER_ID\")\n\t\t\t\t\tassert.Contains(t, content, \"User: all-vars-user\")\n\n\t\t\t\t\tassert.NotContains(t, content, \"$CTX.CUSTOM_KEY\")\n\t\t\t\t\tassert.Contains(t, content, \"Custom: custom-value-123\")\n\n\t\t\t\t\t// Check assistant name from $CTX.ASSISTANT_NAME\n\t\t\t\t\tassert.NotContains(t, content, \"$CTX.ASSISTANT_NAME\")\n\n\t\t\t\t\tfound = true\n\t\t\t\t}\n\t\t\t}\n\t\t}\n\t\tassert.True(t, found, \"Should find global prompt with all variable types replaced\")\n\t})\n\n\tt.Run(\"PromptPresetFromHook\", func(t *testing.T) {\n\t\t// Load fullfields assistant which has prompt_presets\n\t\tast, err := assistant.Get(\"tests.fullfields\")\n\t\trequire.NoError(t, err)\n\t\trequire.NotNil(t, ast.PromptPresets)\n\t\trequire.Contains(t, ast.PromptPresets, \"chat.friendly\")\n\n\t\tctx := newMinimalTestContext()\n\n\t\tmessages := []context.Message{\n\t\t\t{Role: context.RoleUser, Content: \"Test preset from hook\"},\n\t\t}\n\n\t\t// Hook returns prompt_preset\n\t\tcreateResponse := &context.HookCreateResponse{\n\t\t\tPromptPreset: \"chat.friendly\",\n\t\t}\n\n\t\tfinalMessages, _, err := ast.BuildRequest(ctx, messages, createResponse)\n\t\trequire.NoError(t, err)\n\n\t\t// Should have system prompts from the preset\n\t\thasSystemPrompt := false\n\t\tfor _, msg := range finalMessages {\n\t\t\tif msg.Role == context.RoleSystem {\n\t\t\t\thasSystemPrompt = true\n\t\t\t\t// Verify it's from the friendly preset (check content)\n\t\t\t\tassert.Contains(t, msg.Content, \"friendly\", \"Should use friendly preset prompts\")\n\t\t\t\tbreak\n\t\t\t}\n\t\t}\n\t\tassert.True(t, hasSystemPrompt, \"Should have system prompts from preset\")\n\t})\n\n\tt.Run(\"PromptPresetFromMetadata\", func(t *testing.T) {\n\t\t// Load fullfields assistant which has prompt_presets\n\t\tast, err := assistant.Get(\"tests.fullfields\")\n\t\trequire.NoError(t, err)\n\n\t\tctx := newMinimalTestContext()\n\t\tctx.Metadata = map[string]interface{}{\n\t\t\t\"__prompt_preset\": \"chat.professional\",\n\t\t}\n\n\t\tmessages := []context.Message{\n\t\t\t{Role: context.RoleUser, Content: \"Test preset from metadata\"},\n\t\t}\n\n\t\tfinalMessages, _, err := ast.BuildRequest(ctx, messages, nil)\n\t\trequire.NoError(t, err)\n\n\t\t// Should have system prompts from the preset\n\t\thasSystemPrompt := false\n\t\tfor _, msg := range finalMessages {\n\t\t\tif msg.Role == context.RoleSystem {\n\t\t\t\thasSystemPrompt = true\n\t\t\t\t// Verify it's from the professional preset\n\t\t\t\tassert.Contains(t, msg.Content, \"professional\", \"Should use professional preset prompts\")\n\t\t\t\tbreak\n\t\t\t}\n\t\t}\n\t\tassert.True(t, hasSystemPrompt, \"Should have system prompts from preset\")\n\t})\n\n\tt.Run(\"PromptPresetHookOverridesMetadata\", func(t *testing.T) {\n\t\t// Load fullfields assistant\n\t\tast, err := assistant.Get(\"tests.fullfields\")\n\t\trequire.NoError(t, err)\n\n\t\tctx := newMinimalTestContext()\n\t\tctx.Metadata = map[string]interface{}{\n\t\t\t\"__prompt_preset\": \"chat.professional\", // Lower priority\n\t\t}\n\n\t\tmessages := []context.Message{\n\t\t\t{Role: context.RoleUser, Content: \"Test hook overrides metadata\"},\n\t\t}\n\n\t\t// Hook returns different preset (higher priority)\n\t\tcreateResponse := &context.HookCreateResponse{\n\t\t\tPromptPreset: \"chat.friendly\",\n\t\t}\n\n\t\tfinalMessages, _, err := ast.BuildRequest(ctx, messages, createResponse)\n\t\trequire.NoError(t, err)\n\n\t\t// Should use hook's preset, not metadata's\n\t\tfor _, msg := range finalMessages {\n\t\t\tif msg.Role == context.RoleSystem {\n\t\t\t\tassert.Contains(t, msg.Content, \"friendly\", \"Hook preset should override metadata preset\")\n\t\t\t\tbreak\n\t\t\t}\n\t\t}\n\t})\n\n\tt.Run(\"PromptPresetNotFound\", func(t *testing.T) {\n\t\t// Load fullfields assistant\n\t\tast, err := assistant.Get(\"tests.fullfields\")\n\t\trequire.NoError(t, err)\n\n\t\tctx := newMinimalTestContext()\n\t\tctx.Metadata = map[string]interface{}{\n\t\t\t\"__prompt_preset\": \"non.existent.preset\",\n\t\t}\n\n\t\tmessages := []context.Message{\n\t\t\t{Role: context.RoleUser, Content: \"Test non-existent preset\"},\n\t\t}\n\n\t\tfinalMessages, _, err := ast.BuildRequest(ctx, messages, nil)\n\t\trequire.NoError(t, err)\n\n\t\t// Should fallback to default prompts (not crash)\n\t\thasSystemPrompt := false\n\t\tfor _, msg := range finalMessages {\n\t\t\tif msg.Role == context.RoleSystem {\n\t\t\t\thasSystemPrompt = true\n\t\t\t\tbreak\n\t\t\t}\n\t\t}\n\t\tassert.True(t, hasSystemPrompt, \"Should fallback to default prompts when preset not found\")\n\t})\n\n\tt.Run(\"DisableGlobalPromptsFromHook\", func(t *testing.T) {\n\t\t// Set global prompts\n\t\tassistant.SetGlobalPrompts([]store.Prompt{\n\t\t\t{Role: \"system\", Content: \"GLOBAL_PROMPT_MARKER\"},\n\t\t})\n\t\tdefer assistant.SetGlobalPrompts(nil)\n\n\t\t// Load an assistant that does NOT disable global prompts\n\t\tast, err := assistant.Get(\"yaobots\")\n\t\trequire.NoError(t, err)\n\t\trequire.False(t, ast.DisableGlobalPrompts)\n\n\t\tctx := newMinimalTestContext()\n\n\t\tmessages := []context.Message{\n\t\t\t{Role: context.RoleUser, Content: \"Test disable from hook\"},\n\t\t}\n\n\t\t// Hook disables global prompts\n\t\tdisableTrue := true\n\t\tcreateResponse := &context.HookCreateResponse{\n\t\t\tDisableGlobalPrompts: &disableTrue,\n\t\t}\n\n\t\tfinalMessages, _, err := ast.BuildRequest(ctx, messages, createResponse)\n\t\trequire.NoError(t, err)\n\n\t\t// Should NOT have global prompt\n\t\tfor _, msg := range finalMessages {\n\t\t\tif msg.Role == context.RoleSystem {\n\t\t\t\tassert.NotContains(t, msg.Content, \"GLOBAL_PROMPT_MARKER\", \"Global prompts should be disabled by hook\")\n\t\t\t}\n\t\t}\n\t})\n\n\tt.Run(\"DisableGlobalPromptsFromMetadata\", func(t *testing.T) {\n\t\t// Set global prompts\n\t\tassistant.SetGlobalPrompts([]store.Prompt{\n\t\t\t{Role: \"system\", Content: \"GLOBAL_PROMPT_MARKER_2\"},\n\t\t})\n\t\tdefer assistant.SetGlobalPrompts(nil)\n\n\t\t// Load an assistant that does NOT disable global prompts\n\t\tast, err := assistant.Get(\"yaobots\")\n\t\trequire.NoError(t, err)\n\n\t\tctx := newMinimalTestContext()\n\t\tctx.Metadata = map[string]interface{}{\n\t\t\t\"__disable_global_prompts\": true,\n\t\t}\n\n\t\tmessages := []context.Message{\n\t\t\t{Role: context.RoleUser, Content: \"Test disable from metadata\"},\n\t\t}\n\n\t\tfinalMessages, _, err := ast.BuildRequest(ctx, messages, nil)\n\t\trequire.NoError(t, err)\n\n\t\t// Should NOT have global prompt\n\t\tfor _, msg := range finalMessages {\n\t\t\tif msg.Role == context.RoleSystem {\n\t\t\t\tassert.NotContains(t, msg.Content, \"GLOBAL_PROMPT_MARKER_2\", \"Global prompts should be disabled by metadata\")\n\t\t\t}\n\t\t}\n\t})\n\n\tt.Run(\"EnableGlobalPromptsOverrideAssistant\", func(t *testing.T) {\n\t\t// Set global prompts\n\t\tassistant.SetGlobalPrompts([]store.Prompt{\n\t\t\t{Role: \"system\", Content: \"GLOBAL_ENABLED_MARKER\"},\n\t\t})\n\t\tdefer assistant.SetGlobalPrompts(nil)\n\n\t\t// Load fullfields assistant which has disable_global_prompts: true\n\t\tast, err := assistant.Get(\"tests.fullfields\")\n\t\trequire.NoError(t, err)\n\t\trequire.True(t, ast.DisableGlobalPrompts)\n\n\t\tctx := newMinimalTestContext()\n\n\t\tmessages := []context.Message{\n\t\t\t{Role: context.RoleUser, Content: \"Test enable override\"},\n\t\t}\n\n\t\t// Hook enables global prompts (overrides assistant's disable)\n\t\tdisableFalse := false\n\t\tcreateResponse := &context.HookCreateResponse{\n\t\t\tDisableGlobalPrompts: &disableFalse,\n\t\t}\n\n\t\tfinalMessages, _, err := ast.BuildRequest(ctx, messages, createResponse)\n\t\trequire.NoError(t, err)\n\n\t\t// Should have global prompt (hook enabled it)\n\t\tfound := false\n\t\tfor _, msg := range finalMessages {\n\t\t\tif msg.Role == context.RoleSystem && msg.Content == \"GLOBAL_ENABLED_MARKER\" {\n\t\t\t\tfound = true\n\t\t\t\tbreak\n\t\t\t}\n\t\t}\n\t\tassert.True(t, found, \"Global prompts should be enabled by hook override\")\n\t})\n}\n\n// TestPromptPresetAssistant tests the tests.promptpreset assistant with Create Hook\nfunc TestPromptPresetAssistant(t *testing.T) {\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\tt.Run(\"LoadPromptPresetAssistant\", func(t *testing.T) {\n\t\tast, err := assistant.Get(\"tests.promptpreset\")\n\t\trequire.NoError(t, err)\n\t\trequire.NotNil(t, ast)\n\n\t\tassert.Equal(t, \"tests.promptpreset\", ast.ID)\n\t\tassert.Equal(t, \"Prompt Preset Test\", ast.Name)\n\t\tassert.False(t, ast.DisableGlobalPrompts)\n\n\t\t// Should have prompt presets loaded\n\t\trequire.NotNil(t, ast.PromptPresets)\n\t\tassert.Contains(t, ast.PromptPresets, \"mode.friendly\")\n\t\tassert.Contains(t, ast.PromptPresets, \"mode.professional\")\n\n\t\t// Should have script\n\t\tassert.NotNil(t, ast.HookScript)\n\t})\n\n\tt.Run(\"CreateHookSelectsFriendlyPreset\", func(t *testing.T) {\n\t\tast, err := assistant.Get(\"tests.promptpreset\")\n\t\trequire.NoError(t, err)\n\n\t\tctx := newPromptTestContext(\"chat-friendly-test\", \"tests.promptpreset\")\n\n\t\tmessages := []context.Message{\n\t\t\t{Role: context.RoleUser, Content: \"use friendly mode please\"},\n\t\t}\n\n\t\t// Call Create hook\n\t\tcreateResponse, _, err := ast.HookScript.Create(ctx, messages, &context.Options{})\n\t\trequire.NoError(t, err)\n\t\trequire.NotNil(t, createResponse)\n\t\tassert.Equal(t, \"mode.friendly\", createResponse.PromptPreset)\n\n\t\t// Build request\n\t\tfinalMessages, _, err := ast.BuildRequest(ctx, messages, createResponse)\n\t\trequire.NoError(t, err)\n\n\t\t// Should have friendly preset marker in one of the system messages\n\t\tfound := false\n\t\tfor _, msg := range finalMessages {\n\t\t\tif msg.Role == context.RoleSystem && containsString(msg.Content, \"FRIENDLY_PRESET_MARKER\") {\n\t\t\t\tfound = true\n\t\t\t\tbreak\n\t\t\t}\n\t\t}\n\t\tassert.True(t, found, \"Should use friendly preset from Create Hook\")\n\t})\n\n\tt.Run(\"CreateHookSelectsProfessionalPreset\", func(t *testing.T) {\n\t\tast, err := assistant.Get(\"tests.promptpreset\")\n\t\trequire.NoError(t, err)\n\n\t\tctx := newPromptTestContext(\"chat-professional-test\", \"tests.promptpreset\")\n\n\t\tmessages := []context.Message{\n\t\t\t{Role: context.RoleUser, Content: \"use professional tone\"},\n\t\t}\n\n\t\t// Call Create hook\n\t\tcreateResponse, _, err := ast.HookScript.Create(ctx, messages, &context.Options{})\n\t\trequire.NoError(t, err)\n\t\trequire.NotNil(t, createResponse)\n\t\tassert.Equal(t, \"mode.professional\", createResponse.PromptPreset)\n\n\t\t// Build request\n\t\tfinalMessages, _, err := ast.BuildRequest(ctx, messages, createResponse)\n\t\trequire.NoError(t, err)\n\n\t\t// Should have professional preset marker in one of the system messages\n\t\tfound := false\n\t\tfor _, msg := range finalMessages {\n\t\t\tif msg.Role == context.RoleSystem && containsString(msg.Content, \"PROFESSIONAL_PRESET_MARKER\") {\n\t\t\t\tfound = true\n\t\t\t\tbreak\n\t\t\t}\n\t\t}\n\t\tassert.True(t, found, \"Should use professional preset from Create Hook\")\n\t})\n\n\tt.Run(\"CreateHookDisablesGlobalPrompts\", func(t *testing.T) {\n\t\t// Set global prompts\n\t\tassistant.SetGlobalPrompts([]store.Prompt{\n\t\t\t{Role: \"system\", Content: \"GLOBAL_MARKER_FOR_DISABLE_TEST\"},\n\t\t})\n\t\tdefer assistant.SetGlobalPrompts(nil)\n\n\t\tast, err := assistant.Get(\"tests.promptpreset\")\n\t\trequire.NoError(t, err)\n\n\t\tctx := newPromptTestContext(\"chat-disable-global-test\", \"tests.promptpreset\")\n\n\t\tmessages := []context.Message{\n\t\t\t{Role: context.RoleUser, Content: \"disable global prompts\"},\n\t\t}\n\n\t\t// Call Create hook\n\t\tcreateResponse, _, err := ast.HookScript.Create(ctx, messages, &context.Options{})\n\t\trequire.NoError(t, err)\n\t\trequire.NotNil(t, createResponse)\n\t\trequire.NotNil(t, createResponse.DisableGlobalPrompts)\n\t\tassert.True(t, *createResponse.DisableGlobalPrompts)\n\n\t\t// Build request\n\t\tfinalMessages, _, err := ast.BuildRequest(ctx, messages, createResponse)\n\t\trequire.NoError(t, err)\n\n\t\t// Should NOT have global prompt\n\t\tfor _, msg := range finalMessages {\n\t\t\tif msg.Role == context.RoleSystem {\n\t\t\t\tassert.NotContains(t, msg.Content, \"GLOBAL_MARKER_FOR_DISABLE_TEST\")\n\t\t\t}\n\t\t}\n\t})\n\n\tt.Run(\"CreateHookPresetAndDisableGlobal\", func(t *testing.T) {\n\t\t// Set global prompts\n\t\tassistant.SetGlobalPrompts([]store.Prompt{\n\t\t\t{Role: \"system\", Content: \"GLOBAL_MARKER_COMBINED_TEST\"},\n\t\t})\n\t\tdefer assistant.SetGlobalPrompts(nil)\n\n\t\tast, err := assistant.Get(\"tests.promptpreset\")\n\t\trequire.NoError(t, err)\n\n\t\tctx := newPromptTestContext(\"chat-combined-test\", \"tests.promptpreset\")\n\n\t\tmessages := []context.Message{\n\t\t\t{Role: context.RoleUser, Content: \"friendly no global\"},\n\t\t}\n\n\t\t// Call Create hook\n\t\tcreateResponse, _, err := ast.HookScript.Create(ctx, messages, &context.Options{})\n\t\trequire.NoError(t, err)\n\t\trequire.NotNil(t, createResponse)\n\t\tassert.Equal(t, \"mode.friendly\", createResponse.PromptPreset)\n\t\trequire.NotNil(t, createResponse.DisableGlobalPrompts)\n\t\tassert.True(t, *createResponse.DisableGlobalPrompts)\n\n\t\t// Build request\n\t\tfinalMessages, _, err := ast.BuildRequest(ctx, messages, createResponse)\n\t\trequire.NoError(t, err)\n\n\t\t// Should have friendly preset but NOT global\n\t\thasFriendly := false\n\t\tfor _, msg := range finalMessages {\n\t\t\tif msg.Role == context.RoleSystem {\n\t\t\t\tassert.NotContains(t, msg.Content, \"GLOBAL_MARKER_COMBINED_TEST\")\n\t\t\t\tif containsString(msg.Content, \"FRIENDLY_PRESET_MARKER\") {\n\t\t\t\t\thasFriendly = true\n\t\t\t\t}\n\t\t\t}\n\t\t}\n\t\tassert.True(t, hasFriendly, \"Should have friendly preset\")\n\t})\n\n\tt.Run(\"CreateHookUnknownPresetFallback\", func(t *testing.T) {\n\t\tast, err := assistant.Get(\"tests.promptpreset\")\n\t\trequire.NoError(t, err)\n\n\t\tctx := newPromptTestContext(\"chat-unknown-preset-test\", \"tests.promptpreset\")\n\n\t\tmessages := []context.Message{\n\t\t\t{Role: context.RoleUser, Content: \"unknown preset test\"},\n\t\t}\n\n\t\t// Call Create hook\n\t\tcreateResponse, _, err := ast.HookScript.Create(ctx, messages, &context.Options{})\n\t\trequire.NoError(t, err)\n\t\trequire.NotNil(t, createResponse)\n\t\tassert.Equal(t, \"non.existent.preset\", createResponse.PromptPreset)\n\n\t\t// Build request - should not error, fallback to default\n\t\tfinalMessages, _, err := ast.BuildRequest(ctx, messages, createResponse)\n\t\trequire.NoError(t, err)\n\n\t\t// Should fallback to default prompts\n\t\tfound := false\n\t\tfor _, msg := range finalMessages {\n\t\t\tif msg.Role == context.RoleSystem && containsString(msg.Content, \"DEFAULT_PROMPT_MARKER\") {\n\t\t\t\tfound = true\n\t\t\t\tbreak\n\t\t\t}\n\t\t}\n\t\tassert.True(t, found, \"Should fallback to default prompts when preset not found\")\n\t})\n\n\tt.Run(\"CreateHookReturnsNull\", func(t *testing.T) {\n\t\tast, err := assistant.Get(\"tests.promptpreset\")\n\t\trequire.NoError(t, err)\n\n\t\tctx := newPromptTestContext(\"chat-null-test\", \"tests.promptpreset\")\n\n\t\tmessages := []context.Message{\n\t\t\t{Role: context.RoleUser, Content: \"just a normal message\"},\n\t\t}\n\n\t\t// Call Create hook - should return nil\n\t\tcreateResponse, _, err := ast.HookScript.Create(ctx, messages, &context.Options{})\n\t\trequire.NoError(t, err)\n\t\tassert.Nil(t, createResponse)\n\n\t\t// Build request with nil createResponse\n\t\tfinalMessages, _, err := ast.BuildRequest(ctx, messages, nil)\n\t\trequire.NoError(t, err)\n\n\t\t// Should use default prompts\n\t\tfound := false\n\t\tfor _, msg := range finalMessages {\n\t\t\tif msg.Role == context.RoleSystem && containsString(msg.Content, \"DEFAULT_PROMPT_MARKER\") {\n\t\t\t\tfound = true\n\t\t\t\tbreak\n\t\t\t}\n\t\t}\n\t\tassert.True(t, found, \"Should use default prompts when hook returns null\")\n\t})\n}\n" }, { "path": "agent/assistant/build_test.go", "content": "package assistant_test\n\nimport (\n\tstdContext \"context\"\n\t\"testing\"\n\n\t\"github.com/yaoapp/yao/agent/assistant\"\n\t\"github.com/yaoapp/yao/agent/context\"\n\t\"github.com/yaoapp/yao/agent/testutils\"\n\t\"github.com/yaoapp/yao/openapi/oauth/types\"\n)\n\n// newTestContext creates a Context for testing with commonly used fields pre-populated\nfunc newTestContext(chatID, assistantID string) *context.Context {\n\tauthorized := &types.AuthorizedInfo{\n\t\tSubject: \"test-user\",\n\t\tClientID: \"test-client-id\",\n\t\tUserID: \"test-user-123\",\n\t\tTeamID: \"test-team-456\",\n\t\tTenantID: \"test-tenant-789\",\n\t\tSessionID: \"test-session-id\",\n\t}\n\n\tctx := context.New(stdContext.Background(), authorized, chatID)\n\tctx.AssistantID = assistantID\n\tctx.Locale = \"en-us\"\n\tctx.Theme = \"light\"\n\tctx.Client = context.Client{\n\t\tType: \"web\",\n\t\tUserAgent: \"TestAgent/1.0\",\n\t\tIP: \"127.0.0.1\",\n\t}\n\tctx.Referer = context.RefererAPI\n\tctx.Accept = context.AcceptWebCUI\n\tctx.Route = \"/test/route\"\n\tctx.Metadata = map[string]interface{}{\n\t\t\"test\": \"context_metadata\",\n\t}\n\treturn ctx\n}\n\n// TestBuildRequest tests the BuildRequest function\nfunc TestBuildRequest(t *testing.T) {\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\tagent, err := assistant.Get(\"tests.buildrequest\")\n\tif err != nil {\n\t\tt.Fatalf(\"Failed to get tests.buildrequest assistant: %s\", err.Error())\n\t}\n\n\tif agent.HookScript == nil {\n\t\tt.Fatalf(\"The tests.buildrequest assistant has no script\")\n\t}\n\n\tctx := newTestContext(\"chat-test-buildrequest\", \"tests.buildrequest\")\n\n\t// Test 1: No override from hook - should use ast.Options and ctx values\n\tt.Run(\"NoOverride\", func(t *testing.T) {\n\t\tinputMessages := []context.Message{{Role: \"user\", Content: \"no_override\"}}\n\n\t\t// Call Create hook\n\t\tcreateResponse, _, err := agent.HookScript.Create(ctx, inputMessages, &context.Options{})\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to call Create hook: %s\", err.Error())\n\t\t}\n\n\t\t// Build LLM request\n\t\t_, options, err := agent.BuildRequest(ctx, inputMessages, createResponse)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to build LLM request: %s\", err.Error())\n\t\t}\n\n\t\t// Verify options - should use ast.Options values\n\t\tif options.Temperature == nil {\n\t\t\tt.Error(\"Expected temperature from ast.Options, got nil\")\n\t\t} else if *options.Temperature != 0.5 {\n\t\t\tt.Errorf(\"Expected temperature 0.5 from ast.Options, got: %f\", *options.Temperature)\n\t\t}\n\n\t\tif options.MaxTokens == nil {\n\t\t\tt.Error(\"Expected max_tokens from ast.Options, got nil\")\n\t\t} else if *options.MaxTokens != 1000 {\n\t\t\tt.Errorf(\"Expected max_tokens 1000 from ast.Options, got: %d\", *options.MaxTokens)\n\t\t}\n\n\t\tif options.TopP == nil {\n\t\t\tt.Error(\"Expected top_p from ast.Options, got nil\")\n\t\t} else if *options.TopP != 0.9 {\n\t\t\tt.Errorf(\"Expected top_p 0.9 from ast.Options, got: %f\", *options.TopP)\n\t\t}\n\n\t\t// Verify ctx values\n\t\tif options.Route != \"/test/route\" {\n\t\t\tt.Errorf(\"Expected route '/test/route' from ctx, got: %s\", options.Route)\n\t\t}\n\n\t\tif options.Metadata == nil {\n\t\t\tt.Error(\"Expected metadata from ctx, got nil\")\n\t\t} else if options.Metadata[\"test\"] != \"context_metadata\" {\n\t\t\tt.Errorf(\"Expected metadata from ctx, got: %v\", options.Metadata)\n\t\t}\n\n\t\tt.Log(\"✓ No override: ast.Options and ctx values used correctly\")\n\t})\n\n\t// Test 2: Override temperature - hook value should take priority\n\tt.Run(\"OverrideTemperature\", func(t *testing.T) {\n\t\tinputMessages := []context.Message{{Role: \"user\", Content: \"override_temperature\"}}\n\n\t\tcreateResponse, _, err := agent.HookScript.Create(ctx, inputMessages, &context.Options{})\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to call Create hook: %s\", err.Error())\n\t\t}\n\n\t\t_, options, err := agent.BuildRequest(ctx, inputMessages, createResponse)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to build LLM request: %s\", err.Error())\n\t\t}\n\n\t\t// Verify temperature override\n\t\tif options.Temperature == nil {\n\t\t\tt.Error(\"Expected temperature, got nil\")\n\t\t} else if *options.Temperature != 0.9 {\n\t\t\tt.Errorf(\"Expected temperature 0.9 from hook, got: %f\", *options.Temperature)\n\t\t}\n\n\t\t// Other values should still come from ast.Options\n\t\tif options.MaxTokens == nil {\n\t\t\tt.Error(\"Expected max_tokens from ast.Options, got nil\")\n\t\t} else if *options.MaxTokens != 1000 {\n\t\t\tt.Errorf(\"Expected max_tokens 1000 from ast.Options, got: %d\", *options.MaxTokens)\n\t\t}\n\n\t\tt.Log(\"✓ Temperature override: hook value takes priority over ast.Options\")\n\t})\n\n\t// Test 3: Override all - all hook values should take priority\n\tt.Run(\"OverrideAll\", func(t *testing.T) {\n\t\tinputMessages := []context.Message{{Role: \"user\", Content: \"override_all\"}}\n\n\t\tcreateResponse, _, err := agent.HookScript.Create(ctx, inputMessages, &context.Options{})\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to call Create hook: %s\", err.Error())\n\t\t}\n\n\t\t_, options, err := agent.BuildRequest(ctx, inputMessages, createResponse)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to build LLM request: %s\", err.Error())\n\t\t}\n\n\t\t// Verify all overrides\n\t\tif options.Temperature == nil || *options.Temperature != 0.8 {\n\t\t\tt.Errorf(\"Expected temperature 0.8 from hook, got: %v\", options.Temperature)\n\t\t}\n\n\t\tif options.MaxTokens == nil || *options.MaxTokens != 2000 {\n\t\t\tt.Errorf(\"Expected max_tokens 2000 from hook, got: %v\", options.MaxTokens)\n\t\t}\n\n\t\tif options.MaxCompletionTokens == nil || *options.MaxCompletionTokens != 1800 {\n\t\t\tt.Errorf(\"Expected max_completion_tokens 1800 from hook, got: %v\", options.MaxCompletionTokens)\n\t\t}\n\n\t\tif options.Audio == nil {\n\t\t\tt.Error(\"Expected audio from hook, got nil\")\n\t\t} else {\n\t\t\tif options.Audio.Voice != \"alloy\" {\n\t\t\t\tt.Errorf(\"Expected voice 'alloy', got: %s\", options.Audio.Voice)\n\t\t\t}\n\t\t\tif options.Audio.Format != \"mp3\" {\n\t\t\t\tt.Errorf(\"Expected format 'mp3', got: %s\", options.Audio.Format)\n\t\t\t}\n\t\t}\n\n\t\tif options.Route != \"/hook/route\" {\n\t\t\tt.Errorf(\"Expected route '/hook/route' from hook, got: %s\", options.Route)\n\t\t}\n\n\t\tif options.Metadata == nil {\n\t\t\tt.Error(\"Expected metadata from hook, got nil\")\n\t\t} else {\n\t\t\tif options.Metadata[\"source\"] != \"hook\" {\n\t\t\t\tt.Errorf(\"Expected metadata['source'] = 'hook', got: %v\", options.Metadata[\"source\"])\n\t\t\t}\n\t\t}\n\n\t\tt.Log(\"✓ Override all: all hook values take priority\")\n\t})\n\n\t// Test 4: Override route and metadata - tests CUI context priority\n\tt.Run(\"OverrideRouteMetadata\", func(t *testing.T) {\n\t\tinputMessages := []context.Message{{Role: \"user\", Content: \"override_route_metadata\"}}\n\n\t\tcreateResponse, _, err := agent.HookScript.Create(ctx, inputMessages, &context.Options{})\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to call Create hook: %s\", err.Error())\n\t\t}\n\n\t\t_, options, err := agent.BuildRequest(ctx, inputMessages, createResponse)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to build LLM request: %s\", err.Error())\n\t\t}\n\n\t\t// Verify route override\n\t\tif options.Route != \"/custom/route\" {\n\t\t\tt.Errorf(\"Expected route '/custom/route' from hook, got: %s\", options.Route)\n\t\t}\n\n\t\t// Verify metadata merge (ctx metadata should be merged with hook metadata)\n\t\tif options.Metadata == nil {\n\t\t\tt.Error(\"Expected metadata, got nil\")\n\t\t} else {\n\t\t\t// Hook metadata should be present\n\t\t\tif options.Metadata[\"custom\"] != true {\n\t\t\t\tt.Errorf(\"Expected metadata['custom'] = true from hook, got: %v\", options.Metadata[\"custom\"])\n\t\t\t}\n\t\t\tif options.Metadata[\"hook_data\"] != \"test\" {\n\t\t\t\tt.Errorf(\"Expected metadata['hook_data'] = 'test' from hook, got: %v\", options.Metadata[\"hook_data\"])\n\t\t\t}\n\t\t\t// Original ctx metadata should still be there (merged)\n\t\t\tif options.Metadata[\"test\"] != \"context_metadata\" {\n\t\t\t\tt.Errorf(\"Expected original ctx metadata to be preserved, got: %v\", options.Metadata)\n\t\t\t}\n\t\t}\n\n\t\t// Other values should still come from ast.Options\n\t\tif options.Temperature == nil || *options.Temperature != 0.5 {\n\t\t\tt.Errorf(\"Expected temperature 0.5 from ast.Options, got: %v\", options.Temperature)\n\t\t}\n\n\t\tt.Log(\"✓ Route and metadata override: hook values take priority, metadata merged\")\n\t})\n\n\t// Test 5: Nil createResponse - should use ast.Options and ctx values\n\tt.Run(\"NilCreateResponse\", func(t *testing.T) {\n\t\t// Create a fresh context for this test\n\t\tfreshCtx := newTestContext(\"chat-test-nil\", \"tests.buildrequest\")\n\t\tinputMessages := []context.Message{{Role: \"user\", Content: \"test message\"}}\n\n\t\t_, options, err := agent.BuildRequest(freshCtx, inputMessages, nil)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to build LLM request: %s\", err.Error())\n\t\t}\n\n\t\t// Should use ast.Options values\n\t\tif options.Temperature == nil || *options.Temperature != 0.5 {\n\t\t\tt.Errorf(\"Expected temperature 0.5 from ast.Options, got: %v\", options.Temperature)\n\t\t}\n\n\t\t// Should use ctx values\n\t\tif options.Route != \"/test/route\" {\n\t\t\tt.Errorf(\"Expected route '/test/route' from ctx, got: %s\", options.Route)\n\t\t}\n\n\t\tt.Log(\"✓ Nil createResponse: ast.Options and ctx values used\")\n\t})\n\n\t// Test 6: ResponseFormat with *context.ResponseFormat\n\tt.Run(\"ResponseFormatStruct\", func(t *testing.T) {\n\t\tfreshCtx := newTestContext(\"chat-test-response-format\", \"tests.buildrequest\")\n\t\tinputMessages := []context.Message{{Role: \"user\", Content: \"test message\"}}\n\n\t\t// Create a test agent with response_format in Options\n\t\ttestAgent := *agent\n\t\tstrict := true\n\t\ttestAgent.Options = map[string]interface{}{\n\t\t\t\"temperature\": 0.7,\n\t\t\t\"response_format\": &context.ResponseFormat{\n\t\t\t\tType: context.ResponseFormatJSONSchema,\n\t\t\t\tJSONSchema: &context.JSONSchema{\n\t\t\t\t\tName: \"test_schema\",\n\t\t\t\t\tDescription: \"Test schema description\",\n\t\t\t\t\tSchema: map[string]interface{}{\n\t\t\t\t\t\t\"type\": \"object\",\n\t\t\t\t\t\t\"properties\": map[string]interface{}{\n\t\t\t\t\t\t\t\"name\": map[string]interface{}{\n\t\t\t\t\t\t\t\t\"type\": \"string\",\n\t\t\t\t\t\t\t},\n\t\t\t\t\t\t},\n\t\t\t\t\t},\n\t\t\t\t\tStrict: &strict,\n\t\t\t\t},\n\t\t\t},\n\t\t}\n\n\t\t_, options, err := testAgent.BuildRequest(freshCtx, inputMessages, nil)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to build LLM request: %s\", err.Error())\n\t\t}\n\n\t\t// Verify ResponseFormat\n\t\tif options.ResponseFormat == nil {\n\t\t\tt.Fatal(\"Expected ResponseFormat, got nil\")\n\t\t}\n\n\t\tif options.ResponseFormat.Type != context.ResponseFormatJSONSchema {\n\t\t\tt.Errorf(\"Expected type 'json_schema', got: %s\", options.ResponseFormat.Type)\n\t\t}\n\n\t\tif options.ResponseFormat.JSONSchema == nil {\n\t\t\tt.Fatal(\"Expected JSONSchema, got nil\")\n\t\t}\n\n\t\tif options.ResponseFormat.JSONSchema.Name != \"test_schema\" {\n\t\t\tt.Errorf(\"Expected schema name 'test_schema', got: %s\", options.ResponseFormat.JSONSchema.Name)\n\t\t}\n\n\t\tif options.ResponseFormat.JSONSchema.Description != \"Test schema description\" {\n\t\t\tt.Errorf(\"Expected schema description 'Test schema description', got: %s\", options.ResponseFormat.JSONSchema.Description)\n\t\t}\n\n\t\tif options.ResponseFormat.JSONSchema.Strict == nil || *options.ResponseFormat.JSONSchema.Strict != true {\n\t\t\tt.Errorf(\"Expected strict = true, got: %v\", options.ResponseFormat.JSONSchema.Strict)\n\t\t}\n\n\t\tt.Log(\"✓ ResponseFormat with *context.ResponseFormat struct works correctly\")\n\t})\n\n\t// Test 7: ResponseFormat with legacy map[string]interface{}\n\tt.Run(\"ResponseFormatLegacyMap\", func(t *testing.T) {\n\t\tfreshCtx := newTestContext(\"chat-test-response-format-map\", \"tests.buildrequest\")\n\t\tinputMessages := []context.Message{{Role: \"user\", Content: \"test message\"}}\n\n\t\t// Create a test agent with legacy map format\n\t\ttestAgent := *agent\n\t\ttestAgent.Options = map[string]interface{}{\n\t\t\t\"temperature\": 0.7,\n\t\t\t\"response_format\": map[string]interface{}{\n\t\t\t\t\"type\": \"json_schema\",\n\t\t\t\t\"json_schema\": map[string]interface{}{\n\t\t\t\t\t\"name\": \"legacy_schema\",\n\t\t\t\t\t\"description\": \"Legacy schema format\",\n\t\t\t\t\t\"schema\": map[string]interface{}{\n\t\t\t\t\t\t\"type\": \"object\",\n\t\t\t\t\t\t\"properties\": map[string]interface{}{\n\t\t\t\t\t\t\t\"email\": map[string]interface{}{\n\t\t\t\t\t\t\t\t\"type\": \"string\",\n\t\t\t\t\t\t\t},\n\t\t\t\t\t\t},\n\t\t\t\t\t},\n\t\t\t\t\t\"strict\": true,\n\t\t\t\t},\n\t\t\t},\n\t\t}\n\n\t\t_, options, err := testAgent.BuildRequest(freshCtx, inputMessages, nil)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to build LLM request: %s\", err.Error())\n\t\t}\n\n\t\t// Verify ResponseFormat was converted from map\n\t\tif options.ResponseFormat == nil {\n\t\t\tt.Fatal(\"Expected ResponseFormat, got nil\")\n\t\t}\n\n\t\tif options.ResponseFormat.Type != context.ResponseFormatJSONSchema {\n\t\t\tt.Errorf(\"Expected type 'json_schema', got: %s\", options.ResponseFormat.Type)\n\t\t}\n\n\t\tif options.ResponseFormat.JSONSchema == nil {\n\t\t\tt.Fatal(\"Expected JSONSchema, got nil\")\n\t\t}\n\n\t\tif options.ResponseFormat.JSONSchema.Name != \"legacy_schema\" {\n\t\t\tt.Errorf(\"Expected schema name 'legacy_schema', got: %s\", options.ResponseFormat.JSONSchema.Name)\n\t\t}\n\n\t\tif options.ResponseFormat.JSONSchema.Description != \"Legacy schema format\" {\n\t\t\tt.Errorf(\"Expected schema description 'Legacy schema format', got: %s\", options.ResponseFormat.JSONSchema.Description)\n\t\t}\n\n\t\tt.Log(\"✓ ResponseFormat with legacy map[string]interface{} format works correctly\")\n\t})\n\n\t// Test 8: ResponseFormat with simple type (text or json_object)\n\tt.Run(\"ResponseFormatSimpleType\", func(t *testing.T) {\n\t\tfreshCtx := newTestContext(\"chat-test-response-format-simple\", \"tests.buildrequest\")\n\t\tinputMessages := []context.Message{{Role: \"user\", Content: \"test message\"}}\n\n\t\t// Create a test agent with simple response_format\n\t\ttestAgent := *agent\n\t\ttestAgent.Options = map[string]interface{}{\n\t\t\t\"temperature\": 0.7,\n\t\t\t\"response_format\": map[string]interface{}{\n\t\t\t\t\"type\": \"json_object\",\n\t\t\t},\n\t\t}\n\n\t\t_, options, err := testAgent.BuildRequest(freshCtx, inputMessages, nil)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to build LLM request: %s\", err.Error())\n\t\t}\n\n\t\t// Verify ResponseFormat\n\t\tif options.ResponseFormat == nil {\n\t\t\tt.Fatal(\"Expected ResponseFormat, got nil\")\n\t\t}\n\n\t\tif options.ResponseFormat.Type != context.ResponseFormatJSON {\n\t\t\tt.Errorf(\"Expected type 'json_object', got: %s\", options.ResponseFormat.Type)\n\t\t}\n\n\t\tif options.ResponseFormat.JSONSchema != nil {\n\t\t\tt.Errorf(\"Expected JSONSchema to be nil for simple type, got: %v\", options.ResponseFormat.JSONSchema)\n\t\t}\n\n\t\tt.Log(\"✓ ResponseFormat with simple type (json_object) works correctly\")\n\t})\n}\n" }, { "path": "agent/assistant/cache.go", "content": "package assistant\n\nimport (\n\t\"container/list\"\n\t\"sync\"\n)\n\n// Cache represents a thread-safe LRU cache for Assistant objects\ntype Cache struct {\n\tcapacity int\n\tmu sync.RWMutex\n\tlist *list.List\n\titems map[string]*list.Element\n}\n\n// cacheItem represents an item in the cache\ntype cacheItem struct {\n\tkey string\n\tvalue *Assistant\n}\n\n// NewCache creates a new LRU cache with the given capacity\nfunc NewCache(capacity int) *Cache {\n\treturn &Cache{\n\t\tcapacity: capacity,\n\t\tlist: list.New(),\n\t\titems: make(map[string]*list.Element),\n\t}\n}\n\n// Get retrieves an Assistant from the cache by its ID\nfunc (c *Cache) Get(id string) (*Assistant, bool) {\n\tc.mu.Lock()\n\tdefer c.mu.Unlock()\n\n\tif element, exists := c.items[id]; exists {\n\t\tc.list.MoveToFront(element)\n\t\treturn element.Value.(*cacheItem).value, true\n\t}\n\treturn nil, false\n}\n\n// Put adds or updates an Assistant in the cache\nfunc (c *Cache) Put(assistant *Assistant) {\n\tif assistant == nil || assistant.ID == \"\" {\n\t\treturn\n\t}\n\n\tc.mu.Lock()\n\tdefer c.mu.Unlock()\n\n\t// If item exists, update it and move to front\n\tif element, exists := c.items[assistant.ID]; exists {\n\t\tc.list.MoveToFront(element)\n\t\telement.Value.(*cacheItem).value = assistant\n\t\treturn\n\t}\n\n\t// If cache is at capacity, remove oldest item before adding new one\n\tif c.list.Len() >= c.capacity {\n\t\tc.removeOldest()\n\t}\n\n\t// Add new item\n\telement := c.list.PushFront(&cacheItem{\n\t\tkey: assistant.ID,\n\t\tvalue: assistant,\n\t})\n\tc.items[assistant.ID] = element\n}\n\n// Remove removes an Assistant from the cache\nfunc (c *Cache) Remove(id string) {\n\tc.mu.Lock()\n\tdefer c.mu.Unlock()\n\n\tif element, exists := c.items[id]; exists {\n\t\titem := element.Value.(*cacheItem)\n\n\t\t// Unregister scripts before removing from cache\n\t\tif item.value != nil && len(item.value.Scripts) > 0 {\n\t\t\titem.value.UnregisterScripts()\n\t\t}\n\n\t\tc.list.Remove(element)\n\t\tdelete(c.items, id)\n\t}\n}\n\n// Len returns the current number of items in the cache\nfunc (c *Cache) Len() int {\n\tc.mu.RLock()\n\tdefer c.mu.RUnlock()\n\treturn c.list.Len()\n}\n\n// All returns all assistants in the cache\nfunc (c *Cache) All() []*Assistant {\n\tc.mu.RLock()\n\tdefer c.mu.RUnlock()\n\n\tassistants := make([]*Assistant, 0, c.list.Len())\n\tfor element := c.list.Front(); element != nil; element = element.Next() {\n\t\titem := element.Value.(*cacheItem)\n\t\tassistants = append(assistants, item.value)\n\t}\n\treturn assistants\n}\n\n// Clear removes all items from the cache\nfunc (c *Cache) Clear() {\n\tc.mu.Lock()\n\tdefer c.mu.Unlock()\n\n\t// Unregister all scripts before clearing cache\n\tfor element := c.list.Front(); element != nil; element = element.Next() {\n\t\titem := element.Value.(*cacheItem)\n\t\tif item.value != nil && len(item.value.Scripts) > 0 {\n\t\t\titem.value.UnregisterScripts()\n\t\t}\n\t}\n\n\tc.list.Init()\n\tc.items = make(map[string]*list.Element)\n}\n\n// ClearExcept removes items from the cache except those matching the keep function\n// keep function returns true for items that should be preserved\nfunc (c *Cache) ClearExcept(keep func(id string) bool) {\n\tc.mu.Lock()\n\tdefer c.mu.Unlock()\n\n\t// Collect items to remove\n\tvar toRemove []*list.Element\n\tfor element := c.list.Front(); element != nil; element = element.Next() {\n\t\titem := element.Value.(*cacheItem)\n\t\tif !keep(item.key) {\n\t\t\ttoRemove = append(toRemove, element)\n\t\t}\n\t}\n\n\t// Remove collected items\n\tfor _, element := range toRemove {\n\t\titem := element.Value.(*cacheItem)\n\n\t\t// Unregister scripts before removing\n\t\tif item.value != nil && len(item.value.Scripts) > 0 {\n\t\t\titem.value.UnregisterScripts()\n\t\t}\n\n\t\tc.list.Remove(element)\n\t\tdelete(c.items, item.key)\n\t}\n}\n\n// removeOldest removes the least recently used item from the cache\nfunc (c *Cache) removeOldest() {\n\tif element := c.list.Back(); element != nil {\n\t\titem := element.Value.(*cacheItem)\n\n\t\t// Unregister scripts before removing from cache\n\t\tif item.value != nil && len(item.value.Scripts) > 0 {\n\t\t\titem.value.UnregisterScripts()\n\t\t}\n\n\t\tc.list.Remove(element)\n\t\tdelete(c.items, item.key)\n\t}\n}\n" }, { "path": "agent/assistant/cache_test.go", "content": "package assistant_test\n\nimport (\n\t\"sync\"\n\t\"testing\"\n\n\t\"github.com/stretchr/testify/assert\"\n\t\"github.com/yaoapp/gou/process\"\n\t\"github.com/yaoapp/yao/agent/assistant\"\n\t\"github.com/yaoapp/yao/agent/testutils\"\n)\n\nfunc TestCacheBasic(t *testing.T) {\n\tcache := assistant.NewCache(2)\n\n\t// Test empty cache\n\tassert.Equal(t, 0, cache.Len(), \"Expected empty cache\")\n\n\t// Create test assistants\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\tast1, err := assistant.Get(\"tests.mcpload\")\n\tassert.NoError(t, err)\n\n\tast2, err := assistant.Get(\"tests.create\")\n\tassert.NoError(t, err)\n\n\t// Test adding items\n\tcache.Put(ast1)\n\tcache.Put(ast2)\n\n\tassert.Equal(t, 2, cache.Len(), \"Expected cache length 2\")\n\n\t// Test getting items\n\tcached1, exists := cache.Get(\"tests.mcpload\")\n\tassert.True(t, exists, \"Should find tests.mcpload\")\n\tassert.Equal(t, \"tests.mcpload\", cached1.ID)\n\n\tcached2, exists := cache.Get(\"tests.create\")\n\tassert.True(t, exists, \"Should find tests.create\")\n\tassert.Equal(t, \"tests.create\", cached2.ID)\n}\n\nfunc TestCacheLRU(t *testing.T) {\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\tcache := assistant.NewCache(2)\n\n\tast1, _ := assistant.Get(\"tests.mcpload\")\n\tast2, _ := assistant.Get(\"tests.create\")\n\tast3, _ := assistant.Get(\"tests.next\")\n\n\t// Add first two items\n\tcache.Put(ast1)\n\tcache.Put(ast2)\n\n\t// Access ast1 to make it most recently used\n\tcache.Get(\"tests.mcpload\")\n\n\t// Add third item, should evict ast2\n\tcache.Put(ast3)\n\n\t// Check ast2 was evicted\n\t_, exists := cache.Get(\"tests.create\")\n\tassert.False(t, exists, \"tests.create should have been evicted\")\n\n\t// Check ast1 and ast3 are still present\n\t_, exists = cache.Get(\"tests.mcpload\")\n\tassert.True(t, exists, \"tests.mcpload should still be in cache\")\n\n\t_, exists = cache.Get(\"tests.next\")\n\tassert.True(t, exists, \"tests.next should be in cache\")\n}\n\nfunc TestCacheRemove(t *testing.T) {\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\tcache := assistant.NewCache(2)\n\n\tast1, _ := assistant.Get(\"tests.mcpload\")\n\tcache.Put(ast1)\n\n\t// Verify scripts are registered\n\t_, exists := process.Handlers[\"agents.tests.mcpload.tools\"]\n\tassert.True(t, exists, \"Handler should be registered before removal\")\n\n\t// Test remove existing item\n\tcache.Remove(\"tests.mcpload\")\n\tassert.Equal(t, 0, cache.Len(), \"Cache should be empty after removing item\")\n\n\t// Verify scripts are unregistered\n\t_, exists = process.Handlers[\"agents.tests.mcpload.tools\"]\n\tassert.False(t, exists, \"Handler should be unregistered after removal\")\n\n\t// Test remove non-existing item (should not panic)\n\tcache.Remove(\"nonexistent\")\n\tassert.Equal(t, 0, cache.Len(), \"Cache length should not change\")\n}\n\nfunc TestCacheClear(t *testing.T) {\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\tcache := assistant.NewCache(3)\n\n\tast1, _ := assistant.Get(\"tests.mcpload\")\n\tast2, _ := assistant.Get(\"tests.create\")\n\tast3, _ := assistant.Get(\"tests.next\")\n\n\tcache.Put(ast1)\n\tcache.Put(ast2)\n\tcache.Put(ast3)\n\n\tassert.Equal(t, 3, cache.Len(), \"Cache should have 3 items\")\n\n\t// Verify scripts are registered\n\t_, exists := process.Handlers[\"agents.tests.mcpload.tools\"]\n\tassert.True(t, exists, \"Handler should be registered before clear\")\n\n\t// Clear cache\n\tcache.Clear()\n\tassert.Equal(t, 0, cache.Len(), \"Cache should be empty after clear\")\n\n\t// Verify all scripts are unregistered\n\t_, exists = process.Handlers[\"agents.tests.mcpload.tools\"]\n\tassert.False(t, exists, \"Handler should be unregistered after clear\")\n}\n\nfunc TestCacheLRUEviction(t *testing.T) {\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\tcache := assistant.NewCache(2)\n\n\tast1, _ := assistant.Get(\"tests.mcpload\")\n\tast2, _ := assistant.Get(\"tests.create\")\n\tast3, _ := assistant.Get(\"tests.next\")\n\n\tcache.Put(ast1)\n\tcache.Put(ast2)\n\n\t// Verify both are registered\n\t_, exists1 := process.Handlers[\"agents.tests.mcpload.tools\"]\n\tassert.True(t, exists1, \"Handler 1 should be registered\")\n\n\t// Add third item to trigger LRU eviction of oldest (ast1)\n\tcache.Put(ast3)\n\n\t// Verify ast1's handler was unregistered due to eviction\n\t_, exists := process.Handlers[\"agents.tests.mcpload.tools\"]\n\tassert.False(t, exists, \"Handler should be unregistered after LRU eviction\")\n\n\t// Verify ast2 and ast3 are still in cache\n\t_, exists = cache.Get(\"tests.create\")\n\tassert.True(t, exists, \"tests.create should still be in cache\")\n\n\t_, exists = cache.Get(\"tests.next\")\n\tassert.True(t, exists, \"tests.next should be in cache\")\n}\n\nfunc TestCacheConcurrent(t *testing.T) {\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\tcache := assistant.NewCache(10)\n\tvar wg sync.WaitGroup\n\tworkers := 5\n\titerations := 20\n\n\t// Load some assistants for concurrent testing\n\tassistants := []string{\n\t\t\"tests.mcpload\",\n\t\t\"tests.create\",\n\t\t\"tests.next\",\n\t}\n\n\t// Concurrent writes\n\tfor i := 0; i < workers; i++ {\n\t\twg.Add(1)\n\t\tgo func(workerID int) {\n\t\t\tdefer wg.Done()\n\t\t\tfor j := 0; j < iterations; j++ {\n\t\t\t\tastID := assistants[j%len(assistants)]\n\t\t\t\tast, _ := assistant.Get(astID)\n\t\t\t\tif ast != nil {\n\t\t\t\t\tcache.Put(ast)\n\t\t\t\t}\n\t\t\t}\n\t\t}(i)\n\t}\n\n\t// Concurrent reads\n\tfor i := 0; i < workers; i++ {\n\t\twg.Add(1)\n\t\tgo func(workerID int) {\n\t\t\tdefer wg.Done()\n\t\t\tfor j := 0; j < iterations; j++ {\n\t\t\t\tastID := assistants[j%len(assistants)]\n\t\t\t\tcache.Get(astID)\n\t\t\t}\n\t\t}(i)\n\t}\n\n\twg.Wait()\n\n\t// Verify cache is in valid state\n\tassert.True(t, cache.Len() >= 0, \"Cache should have valid length\")\n\tassert.True(t, cache.Len() <= 10, \"Cache should not exceed capacity\")\n}\n\nfunc TestCacheNilInput(t *testing.T) {\n\tcache := assistant.NewCache(2)\n\n\t// Test putting nil assistant\n\tcache.Put(nil)\n\tassert.Equal(t, 0, cache.Len(), \"Cache should not store nil assistant\")\n\n\t// Test putting assistant with empty ID\n\temptyAST := &assistant.Assistant{}\n\tcache.Put(emptyAST)\n\tassert.Equal(t, 0, cache.Len(), \"Cache should not store assistant with empty ID\")\n}\n\nfunc TestCacheAll(t *testing.T) {\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\tcache := assistant.NewCache(5)\n\n\tast1, _ := assistant.Get(\"tests.mcpload\")\n\tast2, _ := assistant.Get(\"tests.create\")\n\tast3, _ := assistant.Get(\"tests.next\")\n\n\tcache.Put(ast1)\n\tcache.Put(ast2)\n\tcache.Put(ast3)\n\n\tall := cache.All()\n\tassert.Equal(t, 3, len(all), \"All() should return 3 assistants\")\n\n\t// Verify all expected assistants are present\n\tids := make(map[string]bool)\n\tfor _, ast := range all {\n\t\tids[ast.ID] = true\n\t}\n\n\tassert.True(t, ids[\"tests.mcpload\"], \"Should contain tests.mcpload\")\n\tassert.True(t, ids[\"tests.create\"], \"Should contain tests.create\")\n\tassert.True(t, ids[\"tests.next\"], \"Should contain tests.next\")\n}\n" }, { "path": "agent/assistant/chat.go", "content": "package assistant\n\nimport (\n\t\"fmt\"\n\t\"strings\"\n\t\"time\"\n\n\t\"github.com/google/uuid\"\n\tagentcontext \"github.com/yaoapp/yao/agent/context\"\n\t\"github.com/yaoapp/yao/agent/i18n\"\n\tstoretypes \"github.com/yaoapp/yao/agent/store/types\"\n)\n\n// InitializeConversation prepares conversation context (synchronous)\n// KB collection is now initialized when user logs in (see openapi/user/login.go)\nfunc (ast *Assistant) InitializeConversation(ctx *agentcontext.Context, options ...*agentcontext.Options) error {\n\t// Reserved for future conversation initialization logic\n\treturn nil\n}\n\n// InitializeConversationAsync prepares conversation context asynchronously\nfunc (ast *Assistant) InitializeConversationAsync(ctx *agentcontext.Context, options ...*agentcontext.Options) {\n\tgo ast.InitializeConversation(ctx, options...)\n}\n\n// GetChatKBID returns the KB collection ID for a chat session\n// Same team + user always returns the same ID (deterministic)\n// Format: chat_{team}_{user} or chat_user_{user} if no team\nfunc GetChatKBID(teamID, userID string) string {\n\t// Sanitize IDs: replace invalid chars with underscores\n\tcleanTeamID := sanitizeCollectionID(teamID)\n\tcleanUserID := sanitizeCollectionID(userID)\n\n\tif cleanTeamID != \"\" {\n\t\treturn fmt.Sprintf(\"chat_%s_%s\", cleanTeamID, cleanUserID)\n\t}\n\treturn fmt.Sprintf(\"chat_user_%s\", cleanUserID)\n}\n\n// sanitizeCollectionID replaces invalid characters with underscores\n// Collection IDs only allow: a-z, A-Z, 0-9, and underscore\nfunc sanitizeCollectionID(id string) string {\n\tif id == \"\" {\n\t\treturn \"\"\n\t}\n\n\t// Replace any character that is not alphanumeric or underscore with underscore\n\tresult := make([]byte, len(id))\n\tfor i := 0; i < len(id); i++ {\n\t\tc := id[i]\n\t\tif (c >= 'a' && c <= 'z') || (c >= 'A' && c <= 'Z') || (c >= '0' && c <= '9') || c == '_' {\n\t\t\tresult[i] = c\n\t\t} else {\n\t\t\tresult[i] = '_'\n\t\t}\n\t}\n\treturn string(result)\n}\n\n// mergeChatMetadata merges default metadata with chat context information\nfunc mergeChatMetadata(defaultMetadata map[string]interface{}, ctx *agentcontext.Context) map[string]interface{} {\n\tmetadata := make(map[string]interface{})\n\n\t// Copy default metadata\n\tfor k, v := range defaultMetadata {\n\t\tmetadata[k] = v\n\t}\n\n\t// Add chat-specific metadata (only for internal tracking, not displayed)\n\tmetadata[\"chat_id\"] = ctx.ChatID\n\tmetadata[\"team_id\"] = ctx.Authorized.TeamID\n\tmetadata[\"user_id\"] = ctx.Authorized.UserID\n\n\t// Get locale from context, default to zh-CN if not set\n\tlocale := ctx.Locale\n\tif locale == \"\" {\n\t\tlocale = \"zh-CN\"\n\t}\n\tlocale = strings.ToLower(locale)\n\n\t// Use i18n for name and description (fixed, not showing user/team IDs)\n\tif _, exists := metadata[\"name\"]; !exists {\n\t\tmetadata[\"name\"] = i18n.T(locale, \"kb.chat.name\")\n\t}\n\tif _, exists := metadata[\"description\"]; !exists {\n\t\tmetadata[\"description\"] = i18n.T(locale, \"kb.chat.description\")\n\t}\n\n\treturn metadata\n}\n\n// =============================================================================\n// Chat Buffer Integration\n// =============================================================================\n\n// InitBuffer initializes the chat buffer for the context\n// Should be called at the start of Stream() for root stack only\nfunc (ast *Assistant) InitBuffer(ctx *agentcontext.Context) {\n\t// Only initialize for root stack\n\tif ctx.Stack == nil || !ctx.Stack.IsRoot() {\n\t\treturn\n\t}\n\n\t// Skip if buffer already exists\n\tif ctx.Buffer != nil {\n\t\treturn\n\t}\n\n\t// Skip if History is disabled in options\n\tif ctx.Stack.Options != nil && ctx.Stack.Options.Skip != nil && ctx.Stack.Options.Skip.History {\n\t\tctx.Logger.Debug(\"Buffer skipped: Skip.History is true\")\n\t\treturn\n\t}\n\n\t// Generate request ID if not set\n\trequestID := ctx.RequestID()\n\tif requestID == \"\" {\n\t\trequestID = uuid.New().String()\n\t}\n\n\t// Get connector and mode from options\n\tconnector := \"\"\n\tmode := \"\"\n\tif ctx.Stack.Options != nil {\n\t\tconnector = ctx.Stack.Options.Connector\n\t\tmode = ctx.Stack.Options.Mode\n\t}\n\n\tctx.Buffer = agentcontext.NewChatBuffer(ctx.ChatID, requestID, ast.ID, connector, mode)\n\tctx.Logger.Debug(\"Buffer initialized: chatID=%s, requestID=%s, assistantID=%s\", ctx.ChatID, requestID, ast.ID)\n}\n\n// BufferUserInput adds user input messages to the buffer\n// Should be called after InitBuffer\nfunc (ast *Assistant) BufferUserInput(ctx *agentcontext.Context, inputMessages []agentcontext.Message) {\n\tif ctx.Buffer == nil {\n\t\treturn\n\t}\n\n\t// Only root stack should buffer user input\n\t// Delegated agents share the same buffer but should not duplicate user input\n\tif ctx.Stack != nil && !ctx.Stack.IsRoot() {\n\t\treturn\n\t}\n\n\t// Convert input messages to buffer format\n\tfor _, msg := range inputMessages {\n\t\t// Extract content from message\n\t\tvar content interface{}\n\t\tvar name string\n\n\t\tcontent = msg.Content\n\t\tif msg.Name != nil {\n\t\t\tname = *msg.Name\n\t\t}\n\n\t\tctx.Buffer.AddUserInput(content, name)\n\t}\n}\n\n// UpdateSpaceSnapshot updates the context memory snapshot in the buffer\n// Only captures Context-level memory (request-scoped temporary data) for recovery\nfunc (ast *Assistant) UpdateSpaceSnapshot(ctx *agentcontext.Context) {\n\tif ctx.Buffer == nil || ctx.Memory == nil || ctx.Memory.Context == nil {\n\t\treturn\n\t}\n\n\tsnapshot := ctx.Memory.Context.Snapshot()\n\tctx.Buffer.SetSpaceSnapshot(snapshot)\n}\n\n// BeginStep starts tracking an execution step\n// Returns the step for further updates\nfunc (ast *Assistant) BeginStep(ctx *agentcontext.Context, stepType string, input map[string]interface{}) *agentcontext.BufferedStep {\n\tif ctx.Buffer == nil {\n\t\treturn nil\n\t}\n\n\t// Update space snapshot before beginning step\n\tast.UpdateSpaceSnapshot(ctx)\n\n\treturn ctx.Buffer.BeginStep(stepType, input, ctx.Stack)\n}\n\n// CompleteStep marks the current step as completed\nfunc (ast *Assistant) CompleteStep(ctx *agentcontext.Context, output map[string]interface{}) {\n\tif ctx.Buffer == nil {\n\t\treturn\n\t}\n\tctx.Buffer.CompleteStep(output)\n}\n\n// FlushBuffer saves all buffered data to the database\n// Should be called in defer block at the end of Stream()\nfunc (ast *Assistant) FlushBuffer(ctx *agentcontext.Context, finalStatus string, err error) {\n\tif ctx.Buffer == nil {\n\t\treturn\n\t}\n\n\t// Only flush for root stack\n\tif ctx.Stack == nil || !ctx.Stack.IsRoot() {\n\t\treturn\n\t}\n\n\t// Get chat store\n\tchatStore := GetChatStore()\n\tif chatStore == nil {\n\t\tctx.Logger.Error(\"Chat store not available, cannot flush buffer\")\n\t\treturn\n\t}\n\n\t// Mark current step as failed/interrupted if needed\n\tif finalStatus != agentcontext.StepStatusCompleted && err != nil {\n\t\tctx.Buffer.FailCurrentStep(finalStatus, err)\n\t}\n\n\t// 1. Save all messages (user input + assistant responses)\n\tmessages := ast.convertBufferedMessages(ctx.Buffer.GetMessages())\n\tif len(messages) > 0 {\n\t\tif saveErr := chatStore.SaveMessages(ctx.ChatID, messages); saveErr != nil {\n\t\t\tctx.Logger.Error(\"Failed to save messages: %v\", saveErr)\n\t\t} else {\n\t\t\tctx.Logger.Debug(\"Saved %d messages for chat=%s\", len(messages), ctx.ChatID)\n\t\t}\n\t}\n\n\t// 2. Update chat last_message_at, last_connector, and last_mode\n\tif len(messages) > 0 {\n\t\tnow := time.Now()\n\t\tupdates := map[string]interface{}{\n\t\t\t\"last_message_at\": now,\n\t\t}\n\t\t// Also update last_connector if available\n\t\tif connector := ctx.Buffer.Connector(); connector != \"\" {\n\t\t\tupdates[\"last_connector\"] = connector\n\t\t}\n\t\t// Also update last_mode if available\n\t\tif mode := ctx.Buffer.Mode(); mode != \"\" {\n\t\t\tupdates[\"last_mode\"] = mode\n\t\t}\n\t\tif updateErr := chatStore.UpdateChat(ctx.ChatID, updates); updateErr != nil {\n\t\t\tctx.Logger.Debug(\"Failed to update chat: %v\", updateErr)\n\t\t}\n\t}\n\n\t// 3. Only save resume steps on error/interrupt (not on success)\n\tif finalStatus != agentcontext.StepStatusCompleted {\n\t\tsteps := ast.convertBufferedSteps(ctx.Buffer.GetStepsForResume(finalStatus))\n\t\tif len(steps) > 0 {\n\t\t\tif saveErr := chatStore.SaveResume(steps); saveErr != nil {\n\t\t\t\tctx.Logger.Error(\"Failed to save resume steps: %v\", saveErr)\n\t\t\t} else {\n\t\t\t\tctx.Logger.Debug(\"Saved %d resume steps for chat=%s (status=%s)\", len(steps), ctx.ChatID, finalStatus)\n\t\t\t}\n\t\t}\n\t}\n\n\t// 4. Close SafeWriter to flush remaining writes (root stack only)\n\t// This ensures all pending SSE messages are sent before the response completes\n\tctx.CloseSafeWriter()\n}\n\n// convertBufferedMessages converts BufferedMessage slice to store Message slice\nfunc (ast *Assistant) convertBufferedMessages(buffered []*agentcontext.BufferedMessage) []*storetypes.Message {\n\tif len(buffered) == 0 {\n\t\treturn nil\n\t}\n\n\tmessages := make([]*storetypes.Message, len(buffered))\n\tfor i, msg := range buffered {\n\t\tmessages[i] = &storetypes.Message{\n\t\t\tMessageID: msg.MessageID,\n\t\t\tChatID: msg.ChatID,\n\t\t\tRequestID: msg.RequestID,\n\t\t\tRole: msg.Role,\n\t\t\tType: msg.Type,\n\t\t\tProps: msg.Props,\n\t\t\tBlockID: msg.BlockID,\n\t\t\tThreadID: msg.ThreadID,\n\t\t\tAssistantID: msg.AssistantID,\n\t\t\tConnector: msg.Connector,\n\t\t\tMode: msg.Mode,\n\t\t\tSequence: msg.Sequence,\n\t\t\tMetadata: msg.Metadata,\n\t\t\tCreatedAt: msg.CreatedAt,\n\t\t\tUpdatedAt: msg.CreatedAt,\n\t\t}\n\t}\n\treturn messages\n}\n\n// convertBufferedSteps converts BufferedStep slice to store Resume slice\nfunc (ast *Assistant) convertBufferedSteps(buffered []*agentcontext.BufferedStep) []*storetypes.Resume {\n\tif len(buffered) == 0 {\n\t\treturn nil\n\t}\n\n\tsteps := make([]*storetypes.Resume, len(buffered))\n\tfor i, step := range buffered {\n\t\tsteps[i] = &storetypes.Resume{\n\t\t\tResumeID: step.ResumeID,\n\t\t\tChatID: step.ChatID,\n\t\t\tRequestID: step.RequestID,\n\t\t\tAssistantID: step.AssistantID,\n\t\t\tStackID: step.StackID,\n\t\t\tStackParentID: step.StackParentID,\n\t\t\tStackDepth: step.StackDepth,\n\t\t\tType: step.Type,\n\t\t\tStatus: step.Status,\n\t\t\tInput: step.Input,\n\t\t\tOutput: step.Output,\n\t\t\tSpaceSnapshot: step.SpaceSnapshot,\n\t\t\tError: step.Error,\n\t\t\tSequence: step.Sequence,\n\t\t\tMetadata: step.Metadata,\n\t\t\tCreatedAt: step.CreatedAt,\n\t\t\tUpdatedAt: step.CreatedAt,\n\t\t}\n\t}\n\treturn steps\n}\n\n// EnsureChat ensures a chat session exists, creates if not\nfunc (ast *Assistant) EnsureChat(ctx *agentcontext.Context) error {\n\tif ctx.ChatID == \"\" {\n\t\treturn nil // No chat ID, skip\n\t}\n\n\t// Skip if history is disabled\n\tif ctx.Stack != nil && ctx.Stack.Options != nil && ctx.Stack.Options.Skip != nil && ctx.Stack.Options.Skip.History {\n\t\treturn nil // Skip.History is true, don't create chat session\n\t}\n\n\tchatStore := GetChatStore()\n\tif chatStore == nil {\n\t\treturn nil // No store, skip\n\t}\n\n\t// Check if chat exists\n\t_, err := chatStore.GetChat(ctx.ChatID)\n\tif err == nil {\n\t\treturn nil // Chat exists\n\t}\n\n\t// Create new chat with permission fields\n\tchat := &storetypes.Chat{\n\t\tChatID: ctx.ChatID,\n\t\tAssistantID: ast.ID,\n\t\tStatus: \"active\",\n\t\tShare: \"private\",\n\t\tSort: 0,\n\t\tMetadata: ctx.Metadata,\n\t\tCreatedAt: time.Now(),\n\t\tUpdatedAt: time.Now(),\n\t}\n\n\t// Set last_connector from options (user selected connector)\n\tif ctx.Stack != nil && ctx.Stack.Options != nil && ctx.Stack.Options.Connector != \"\" {\n\t\tchat.LastConnector = ctx.Stack.Options.Connector\n\t}\n\n\t// Set permission fields from authorized info\n\tif ctx.Authorized != nil {\n\t\tchat.CreatedBy = ctx.Authorized.UserID\n\t\tchat.UpdatedBy = ctx.Authorized.UserID\n\t\tchat.TeamID = ctx.Authorized.TeamID\n\t\tchat.TenantID = ctx.Authorized.TenantID\n\t}\n\n\treturn chatStore.CreateChat(chat)\n}\n\n// GetChatStore returns the chat store instance\n// Returns nil if storage is not configured\nfunc GetChatStore() storetypes.ChatStore {\n\tif storage == nil {\n\t\treturn nil\n\t}\n\treturn storage\n}\n\n// GetStore returns the full store instance (implements both ChatStore and AssistantStore)\n// Returns nil if storage is not configured\nfunc GetStore() storetypes.Store {\n\tif storage == nil {\n\t\treturn nil\n\t}\n\treturn storage\n}\n\n// =============================================================================\n// Deprecated methods (kept for compatibility)\n// =============================================================================\n\nfunc (ast *Assistant) saveChat(ctx *agentcontext.Context, input []agentcontext.Message, opts *agentcontext.Options) error {\n\t_ = ctx\n\t_ = input\n\t_ = opts\n\treturn nil\n}\n" }, { "path": "agent/assistant/chat_test.go", "content": "package assistant_test\n\nimport (\n\t\"context\"\n\t\"fmt\"\n\t\"sync\"\n\t\"testing\"\n\t\"time\"\n\n\t\"github.com/google/uuid\"\n\t\"github.com/stretchr/testify/assert\"\n\t\"github.com/stretchr/testify/require\"\n\t\"github.com/yaoapp/yao/agent/assistant\"\n\tagentcontext \"github.com/yaoapp/yao/agent/context\"\n\tstoretypes \"github.com/yaoapp/yao/agent/store/types\"\n\t\"github.com/yaoapp/yao/agent/testutils\"\n\toauthtypes \"github.com/yaoapp/yao/openapi/oauth/types\"\n)\n\nfunc TestGetChatKBID(t *testing.T) {\n\tt.Run(\"WithTeamAndUser\", func(t *testing.T) {\n\t\tteamID := \"5659-5504-2879\"\n\t\tuserID := \"4287-9400-2030-0504\"\n\n\t\tcollectionID := assistant.GetChatKBID(teamID, userID)\n\n\t\t// Should sanitize dashes to underscores\n\t\texpected := \"chat_5659_5504_2879_4287_9400_2030_0504\"\n\t\tassert.Equal(t, expected, collectionID)\n\t\tt.Logf(\"✓ Collection ID with team: %s\", collectionID)\n\t})\n\n\tt.Run(\"WithoutTeam\", func(t *testing.T) {\n\t\tteamID := \"\"\n\t\tuserID := \"4287-9400-2030-0504\"\n\n\t\tcollectionID := assistant.GetChatKBID(teamID, userID)\n\n\t\t// Should use chat_user_ prefix\n\t\texpected := \"chat_user_4287_9400_2030_0504\"\n\t\tassert.Equal(t, expected, collectionID)\n\t\tt.Logf(\"✓ Collection ID without team: %s\", collectionID)\n\t})\n\n\tt.Run(\"Idempotent\", func(t *testing.T) {\n\t\tteamID := \"test-team-123\"\n\t\tuserID := \"test-user-456\"\n\n\t\tid1 := assistant.GetChatKBID(teamID, userID)\n\t\tid2 := assistant.GetChatKBID(teamID, userID)\n\t\tid3 := assistant.GetChatKBID(teamID, userID)\n\n\t\t// Same input should always produce same output\n\t\tassert.Equal(t, id1, id2)\n\t\tassert.Equal(t, id2, id3)\n\t\tt.Logf(\"✓ Idempotent: %s\", id1)\n\t})\n\n\tt.Run(\"SanitizeSpecialChars\", func(t *testing.T) {\n\t\tteamID := \"team-with-dashes@123\"\n\t\tuserID := \"user.with.dots!\"\n\n\t\tcollectionID := assistant.GetChatKBID(teamID, userID)\n\n\t\t// Should only contain alphanumeric and underscores\n\t\tassert.Regexp(t, \"^[a-zA-Z0-9_]+$\", collectionID)\n\t\tt.Logf(\"✓ Sanitized ID: %s\", collectionID)\n\t})\n\n\tt.Run(\"EmptyUserID\", func(t *testing.T) {\n\t\tteamID := \"test-team\"\n\t\tuserID := \"\"\n\n\t\tcollectionID := assistant.GetChatKBID(teamID, userID)\n\n\t\t// Should handle empty user ID gracefully\n\t\texpected := \"chat_test_team_\"\n\t\tassert.Equal(t, expected, collectionID)\n\t\tt.Logf(\"✓ Empty user ID handled: %s\", collectionID)\n\t})\n}\n\nfunc TestPrepareKBCollection(t *testing.T) {\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\t// Get assistant\n\tast, err := assistant.Get(\"mohe\")\n\trequire.NoError(t, err)\n\trequire.NotNil(t, ast)\n\n\t// Note: KB collection is now created during user login (see openapi/user/login.go)\n\t// These tests verify that InitializeConversation handles various scenarios gracefully\n\n\tt.Run(\"InitializeWithAuthorizedInfo\", func(t *testing.T) {\n\t\t// Use unique IDs based on timestamp to avoid conflicts\n\t\ttimestamp := fmt.Sprintf(\"%d\", time.Now().UnixNano())\n\t\tteamID := fmt.Sprintf(\"test_team_%s\", timestamp)\n\t\tuserID := fmt.Sprintf(\"test_user_%s\", timestamp)\n\n\t\tctx := agentcontext.New(context.Background(), &oauthtypes.AuthorizedInfo{\n\t\t\tTeamID: teamID,\n\t\t\tUserID: userID,\n\t\t}, \"test_chat_prepare_001\")\n\n\t\topts := &agentcontext.Options{}\n\n\t\t// InitializeConversation should succeed (KB collection created at login time)\n\t\terr := ast.InitializeConversation(ctx, opts)\n\t\tassert.NoError(t, err)\n\t\tt.Logf(\"✓ InitializeConversation completed successfully\")\n\t})\n\n\tt.Run(\"IdempotentInitialization\", func(t *testing.T) {\n\t\t// Use unique IDs based on timestamp to avoid conflicts\n\t\ttimestamp := fmt.Sprintf(\"%d\", time.Now().UnixNano())\n\t\tteamID := fmt.Sprintf(\"idem_team_%s\", timestamp)\n\t\tuserID := fmt.Sprintf(\"idem_user_%s\", timestamp)\n\n\t\tctx := agentcontext.New(context.Background(), &oauthtypes.AuthorizedInfo{\n\t\t\tTeamID: teamID,\n\t\t\tUserID: userID,\n\t\t}, \"test_chat_idempotent\")\n\n\t\topts := &agentcontext.Options{}\n\n\t\t// Multiple calls should all succeed\n\t\terr1 := ast.InitializeConversation(ctx, opts)\n\t\tassert.NoError(t, err1)\n\n\t\terr2 := ast.InitializeConversation(ctx, opts)\n\t\tassert.NoError(t, err2)\n\n\t\terr3 := ast.InitializeConversation(ctx, opts)\n\t\tassert.NoError(t, err3)\n\n\t\tt.Logf(\"✓ Idempotent initialization works correctly\")\n\t})\n\n\tt.Run(\"HandleMissingAuthorizedInfo\", func(t *testing.T) {\n\t\tctx := agentcontext.New(context.Background(), nil, \"test_chat_no_auth\") // Missing authorized info\n\n\t\topts := &agentcontext.Options{}\n\n\t\t// Should not error, just return nil\n\t\terr := ast.InitializeConversation(ctx, opts)\n\t\tassert.NoError(t, err)\n\t\tt.Logf(\"✓ Correctly handled missing authorized info\")\n\t})\n\n\tt.Run(\"ConcurrentInitialization\", func(t *testing.T) {\n\t\t// Use unique IDs based on timestamp to avoid conflicts\n\t\ttimestamp := fmt.Sprintf(\"%d\", time.Now().UnixNano())\n\t\tteamID := fmt.Sprintf(\"concurrent_team_%s\", timestamp)\n\t\tuserID := fmt.Sprintf(\"concurrent_user_%s\", timestamp)\n\n\t\tctx := agentcontext.New(context.Background(), &oauthtypes.AuthorizedInfo{\n\t\t\tTeamID: teamID,\n\t\t\tUserID: userID,\n\t\t}, \"test_chat_concurrent\")\n\n\t\topts := &agentcontext.Options{}\n\n\t\t// Launch 5 concurrent calls\n\t\tvar wg sync.WaitGroup\n\t\terrors := make([]error, 5)\n\t\tfor i := 0; i < 5; i++ {\n\t\t\twg.Add(1)\n\t\t\tgo func(idx int) {\n\t\t\t\tdefer wg.Done()\n\t\t\t\terrors[idx] = ast.InitializeConversation(ctx, opts)\n\t\t\t}(i)\n\t\t}\n\n\t\t// Wait for all goroutines to complete\n\t\twg.Wait()\n\n\t\t// All calls should succeed\n\t\tfor i, err := range errors {\n\t\t\tassert.NoError(t, err, \"Goroutine %d should not error\", i)\n\t\t}\n\n\t\tt.Logf(\"✓ Concurrent initialization handled correctly\")\n\t})\n}\n\nfunc TestInitializeConversation(t *testing.T) {\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\tast, err := assistant.Get(\"mohe\")\n\trequire.NoError(t, err)\n\trequire.NotNil(t, ast)\n\n\tt.Run(\"FullInitialization\", func(t *testing.T) {\n\t\t// Use unique IDs based on timestamp to avoid conflicts\n\t\ttimestamp := fmt.Sprintf(\"%d\", time.Now().UnixNano())\n\t\tteamID := fmt.Sprintf(\"init_team_%s\", timestamp)\n\t\tuserID := fmt.Sprintf(\"init_user_%s\", timestamp)\n\n\t\tctx := agentcontext.New(context.Background(), &oauthtypes.AuthorizedInfo{\n\t\t\tTeamID: teamID,\n\t\t\tUserID: userID,\n\t\t}, \"test_init_chat_001\")\n\n\t\topts := &agentcontext.Options{}\n\n\t\t// Should initialize conversation without error\n\t\t// Note: KB collection is now created during user login, not here\n\t\terr := ast.InitializeConversation(ctx, opts)\n\t\tassert.NoError(t, err)\n\t\tt.Logf(\"✓ Conversation initialized successfully (KB collection created at login time)\")\n\t})\n\n\tt.Run(\"SkipHistoryFlag\", func(t *testing.T) {\n\t\tctx := agentcontext.New(context.Background(), &oauthtypes.AuthorizedInfo{\n\t\t\tTeamID: \"skip_team\",\n\t\t\tUserID: \"skip_user\",\n\t\t}, \"test_skip_history\")\n\n\t\topts := &agentcontext.Options{\n\t\t\tSkip: &agentcontext.Skip{\n\t\t\t\tHistory: true,\n\t\t\t},\n\t\t}\n\n\t\t// Should skip initialization when history flag is set\n\t\terr := ast.InitializeConversation(ctx, opts)\n\t\tassert.NoError(t, err)\n\t\tt.Logf(\"✓ Correctly skipped with history flag\")\n\t})\n}\n\n// =============================================================================\n// Buffer Integration Tests\n// =============================================================================\n\nfunc TestBufferInitialization(t *testing.T) {\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\tast, err := assistant.Get(\"mohe\")\n\trequire.NoError(t, err)\n\trequire.NotNil(t, ast)\n\n\tt.Run(\"InitBufferForRootStack\", func(t *testing.T) {\n\t\tctx := agentcontext.New(context.Background(), nil, \"test_chat_buffer_001\")\n\n\t\t// Enter stack to simulate root stack\n\t\t_, _, done := agentcontext.EnterStack(ctx, ast.ID, nil)\n\t\tdefer done()\n\n\t\t// Initialize buffer\n\t\tast.InitBuffer(ctx)\n\n\t\t// Verify buffer was created\n\t\tassert.NotNil(t, ctx.Buffer, \"Buffer should be initialized for root stack\")\n\t\tassert.Equal(t, \"test_chat_buffer_001\", ctx.Buffer.ChatID())\n\t\tassert.Equal(t, ast.ID, ctx.Buffer.AssistantID())\n\t\tt.Logf(\"✓ Buffer initialized: chatID=%s, assistantID=%s\", ctx.Buffer.ChatID(), ctx.Buffer.AssistantID())\n\t})\n\n\tt.Run(\"SkipBufferForNestedStack\", func(t *testing.T) {\n\t\tctx := agentcontext.New(context.Background(), nil, \"test_chat_buffer_nested\")\n\n\t\t// Enter root stack\n\t\t_, _, doneRoot := agentcontext.EnterStack(ctx, \"root_assistant\", nil)\n\t\tdefer doneRoot()\n\n\t\t// Enter nested stack\n\t\t_, _, doneNested := agentcontext.EnterStack(ctx, \"nested_assistant\", nil)\n\t\tdefer doneNested()\n\n\t\t// Try to initialize buffer (should be skipped for nested stack)\n\t\tast.InitBuffer(ctx)\n\n\t\t// Buffer should be nil because we're not at root\n\t\tassert.Nil(t, ctx.Buffer, \"Buffer should not be initialized for nested stack\")\n\t\tt.Logf(\"✓ Buffer correctly skipped for nested stack\")\n\t})\n\n\tt.Run(\"IdempotentBufferInit\", func(t *testing.T) {\n\t\tctx := agentcontext.New(context.Background(), nil, \"test_chat_buffer_idem\")\n\n\t\t// Enter stack\n\t\t_, _, done := agentcontext.EnterStack(ctx, ast.ID, nil)\n\t\tdefer done()\n\n\t\t// Initialize buffer twice\n\t\tast.InitBuffer(ctx)\n\t\tfirstBuffer := ctx.Buffer\n\n\t\tast.InitBuffer(ctx)\n\t\tsecondBuffer := ctx.Buffer\n\n\t\t// Should be the same buffer instance\n\t\tassert.Same(t, firstBuffer, secondBuffer, \"Buffer should be idempotent\")\n\t\tt.Logf(\"✓ Buffer initialization is idempotent\")\n\t})\n}\n\nfunc TestBufferUserInput(t *testing.T) {\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\tast, err := assistant.Get(\"mohe\")\n\trequire.NoError(t, err)\n\n\tt.Run(\"BufferSimpleTextInput\", func(t *testing.T) {\n\t\tctx := agentcontext.New(context.Background(), nil, \"test_chat_input_001\")\n\n\t\t// Enter stack and init buffer\n\t\t_, _, done := agentcontext.EnterStack(ctx, ast.ID, nil)\n\t\tdefer done()\n\t\tast.InitBuffer(ctx)\n\n\t\t// Create input messages\n\t\tinputMessages := []agentcontext.Message{\n\t\t\t{\n\t\t\t\tRole: agentcontext.RoleUser,\n\t\t\t\tContent: \"Hello, how are you?\",\n\t\t\t},\n\t\t}\n\n\t\t// Buffer user input\n\t\tast.BufferUserInput(ctx, inputMessages)\n\n\t\t// Verify buffer contains the message\n\t\tmessages := ctx.Buffer.GetMessages()\n\t\tassert.Len(t, messages, 1, \"Should have 1 buffered message\")\n\t\tassert.Equal(t, \"user\", messages[0].Role)\n\t\tassert.Equal(t, \"user_input\", messages[0].Type)\n\t\tassert.Equal(t, \"Hello, how are you?\", messages[0].Props[\"content\"])\n\t\tt.Logf(\"✓ User input buffered: %v\", messages[0].Props)\n\t})\n\n\tt.Run(\"BufferMultipleMessages\", func(t *testing.T) {\n\t\tctx := agentcontext.New(context.Background(), nil, \"test_chat_input_multi\")\n\n\t\t// Enter stack and init buffer\n\t\t_, _, done := agentcontext.EnterStack(ctx, ast.ID, nil)\n\t\tdefer done()\n\t\tast.InitBuffer(ctx)\n\n\t\t// Create multiple input messages\n\t\tinputMessages := []agentcontext.Message{\n\t\t\t{Role: agentcontext.RoleUser, Content: \"First message\"},\n\t\t\t{Role: agentcontext.RoleUser, Content: \"Second message\"},\n\t\t}\n\n\t\t// Buffer user input\n\t\tast.BufferUserInput(ctx, inputMessages)\n\n\t\t// Verify buffer contains all messages\n\t\tmessages := ctx.Buffer.GetMessages()\n\t\tassert.Len(t, messages, 2, \"Should have 2 buffered messages\")\n\t\tassert.Equal(t, 1, messages[0].Sequence)\n\t\tassert.Equal(t, 2, messages[1].Sequence)\n\t\tt.Logf(\"✓ Multiple messages buffered with correct sequence\")\n\t})\n\n\tt.Run(\"BufferWithNilBuffer\", func(t *testing.T) {\n\t\tctx := agentcontext.New(context.Background(), nil, \"test_chat_input_nil\")\n\n\t\t// Don't initialize buffer\n\t\tinputMessages := []agentcontext.Message{\n\t\t\t{Role: agentcontext.RoleUser, Content: \"Test\"},\n\t\t}\n\n\t\t// Should not panic\n\t\tast.BufferUserInput(ctx, inputMessages)\n\t\tt.Logf(\"✓ BufferUserInput handles nil buffer gracefully\")\n\t})\n}\n\nfunc TestBufferStepTracking(t *testing.T) {\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\tast, err := assistant.Get(\"mohe\")\n\trequire.NoError(t, err)\n\n\tt.Run(\"BeginAndCompleteStep\", func(t *testing.T) {\n\t\tctx := agentcontext.New(context.Background(), nil, \"test_chat_step_001\")\n\n\t\t// Enter stack and init buffer\n\t\t_, _, done := agentcontext.EnterStack(ctx, ast.ID, nil)\n\t\tdefer done()\n\t\tast.InitBuffer(ctx)\n\n\t\t// Set some context memory data\n\t\tif ctx.Memory != nil && ctx.Memory.Context != nil {\n\t\t\tctx.Memory.Context.Set(\"test_key\", \"test_value\", 0)\n\t\t}\n\n\t\t// Begin a step\n\t\tstep := ast.BeginStep(ctx, agentcontext.StepTypeLLM, map[string]interface{}{\n\t\t\t\"messages\": []string{\"Hello\"},\n\t\t})\n\n\t\tassert.NotNil(t, step, \"Step should be created\")\n\t\tassert.Equal(t, agentcontext.StepTypeLLM, step.Type)\n\t\tassert.Equal(t, agentcontext.StepStatusRunning, step.Status)\n\t\tassert.NotEmpty(t, step.StackID)\n\n\t\t// Complete the step\n\t\tast.CompleteStep(ctx, map[string]interface{}{\n\t\t\t\"content\": \"Response\",\n\t\t})\n\n\t\t// Verify step is completed\n\t\tsteps := ctx.Buffer.GetAllSteps()\n\t\tassert.Len(t, steps, 1)\n\t\tassert.Equal(t, agentcontext.StepStatusCompleted, steps[0].Status)\n\t\tassert.Equal(t, \"Response\", steps[0].Output[\"content\"])\n\t\tt.Logf(\"✓ Step tracking works correctly\")\n\t})\n\n\tt.Run(\"ContextMemorySnapshotCapture\", func(t *testing.T) {\n\t\tctx := agentcontext.New(context.Background(), nil, \"test_chat_memory_001\")\n\n\t\t// Enter stack and init buffer\n\t\t_, _, done := agentcontext.EnterStack(ctx, ast.ID, nil)\n\t\tdefer done()\n\t\tast.InitBuffer(ctx)\n\n\t\t// Set context memory data before step\n\t\trequire.NotNil(t, ctx.Memory)\n\t\trequire.NotNil(t, ctx.Memory.Context)\n\t\tctx.Memory.Context.Set(\"key1\", \"value1\", 0)\n\t\tctx.Memory.Context.Set(\"key2\", 123, 0)\n\n\t\t// Begin step (should capture context memory snapshot)\n\t\tast.BeginStep(ctx, agentcontext.StepTypeHookCreate, nil)\n\n\t\t// Verify context memory snapshot was captured\n\t\tsteps := ctx.Buffer.GetAllSteps()\n\t\trequire.Len(t, steps, 1)\n\t\tassert.NotNil(t, steps[0].SpaceSnapshot)\n\t\tassert.Equal(t, \"value1\", steps[0].SpaceSnapshot[\"key1\"])\n\t\tassert.Equal(t, 123, steps[0].SpaceSnapshot[\"key2\"])\n\t\tt.Logf(\"✓ Context memory snapshot captured: %v\", steps[0].SpaceSnapshot)\n\t})\n\n\tt.Run(\"MultipleSteps\", func(t *testing.T) {\n\t\tctx := agentcontext.New(context.Background(), nil, \"test_chat_multi_step\")\n\n\t\t// Enter stack and init buffer\n\t\t_, _, done := agentcontext.EnterStack(ctx, ast.ID, nil)\n\t\tdefer done()\n\t\tast.InitBuffer(ctx)\n\n\t\t// Step 1: hook_create\n\t\tast.BeginStep(ctx, agentcontext.StepTypeHookCreate, map[string]interface{}{\"phase\": \"create\"})\n\t\tast.CompleteStep(ctx, map[string]interface{}{\"result\": \"created\"})\n\n\t\t// Step 2: llm\n\t\tast.BeginStep(ctx, agentcontext.StepTypeLLM, map[string]interface{}{\"phase\": \"llm\"})\n\t\tast.CompleteStep(ctx, map[string]interface{}{\"result\": \"completed\"})\n\n\t\t// Step 3: hook_next\n\t\tast.BeginStep(ctx, agentcontext.StepTypeHookNext, map[string]interface{}{\"phase\": \"next\"})\n\t\tast.CompleteStep(ctx, map[string]interface{}{\"result\": \"done\"})\n\n\t\t// Verify all steps\n\t\tsteps := ctx.Buffer.GetAllSteps()\n\t\tassert.Len(t, steps, 3)\n\t\tassert.Equal(t, agentcontext.StepTypeHookCreate, steps[0].Type)\n\t\tassert.Equal(t, agentcontext.StepTypeLLM, steps[1].Type)\n\t\tassert.Equal(t, agentcontext.StepTypeHookNext, steps[2].Type)\n\t\tt.Logf(\"✓ Multiple steps tracked correctly\")\n\t})\n}\n\nfunc TestFlushBuffer(t *testing.T) {\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\tast, err := assistant.Get(\"mohe\")\n\trequire.NoError(t, err)\n\n\t// Skip if chat store not available\n\tchatStore := assistant.GetChatStore()\n\tif chatStore == nil {\n\t\tt.Skip(\"Chat store not configured, skipping flush tests\")\n\t}\n\n\tt.Run(\"FlushOnSuccess\", func(t *testing.T) {\n\t\tchatID := fmt.Sprintf(\"test_flush_success_%s\", uuid.New().String()[:8])\n\t\tctx := agentcontext.New(context.Background(), nil, chatID)\n\n\t\t// Enter stack and init buffer\n\t\t_, _, done := agentcontext.EnterStack(ctx, ast.ID, nil)\n\t\tdefer done()\n\t\tast.InitBuffer(ctx)\n\n\t\t// Ensure chat exists\n\t\terr := chatStore.CreateChat(&storetypes.Chat{\n\t\t\tChatID: chatID,\n\t\t\tAssistantID: ast.ID,\n\t\t\tStatus: \"active\",\n\t\t\tShare: \"private\",\n\t\t\tCreatedAt: time.Now(),\n\t\t\tUpdatedAt: time.Now(),\n\t\t})\n\t\trequire.NoError(t, err)\n\n\t\t// Add some messages to buffer\n\t\trequire.NotNil(t, ctx.Buffer, \"Buffer should be initialized\")\n\t\tctx.Buffer.AddUserInput(\"Test question\", \"\")\n\t\tctx.Buffer.AddAssistantMessage(\"M1\", \"text\", map[string]interface{}{\"content\": \"Test answer\"}, \"\", \"\", ast.ID, nil)\n\n\t\t// Add a step\n\t\tast.BeginStep(ctx, agentcontext.StepTypeLLM, nil)\n\t\tast.CompleteStep(ctx, nil)\n\n\t\t// Flush buffer (success case)\n\t\tast.FlushBuffer(ctx, agentcontext.StepStatusCompleted, nil)\n\n\t\t// Verify messages were saved\n\t\tmessages, err := chatStore.GetMessages(chatID, storetypes.MessageFilter{})\n\t\tassert.NoError(t, err)\n\t\tassert.Len(t, messages, 2, \"Should have 2 messages saved\")\n\n\t\t// Verify no resume records (success case)\n\t\tresumes, err := chatStore.GetResume(chatID)\n\t\tassert.NoError(t, err)\n\t\tassert.Len(t, resumes, 0, \"Should have no resume records on success\")\n\n\t\t// Cleanup\n\t\tchatStore.DeleteChat(chatID)\n\t\tt.Logf(\"✓ Buffer flushed on success: %d messages saved, no resume records\", len(messages))\n\t})\n\n\tt.Run(\"FlushOnFailure\", func(t *testing.T) {\n\t\tchatID := fmt.Sprintf(\"test_flush_fail_%s\", uuid.New().String()[:8])\n\t\tctx := agentcontext.New(context.Background(), nil, chatID)\n\n\t\t// Enter stack and init buffer\n\t\t_, _, done := agentcontext.EnterStack(ctx, ast.ID, nil)\n\t\tdefer done()\n\t\tast.InitBuffer(ctx)\n\n\t\t// Ensure chat exists\n\t\terr := chatStore.CreateChat(&storetypes.Chat{\n\t\t\tChatID: chatID,\n\t\t\tAssistantID: ast.ID,\n\t\t\tStatus: \"active\",\n\t\t\tShare: \"private\",\n\t\t\tCreatedAt: time.Now(),\n\t\t\tUpdatedAt: time.Now(),\n\t\t})\n\t\trequire.NoError(t, err)\n\n\t\t// Add messages\n\t\tctx.Buffer.AddUserInput(\"Test question\", \"\")\n\n\t\t// Add a step that will \"fail\"\n\t\tast.BeginStep(ctx, agentcontext.StepTypeLLM, map[string]interface{}{\"test\": \"data\"})\n\t\t// Don't complete - simulate failure\n\n\t\t// Flush buffer (failure case)\n\t\ttestErr := fmt.Errorf(\"simulated error\")\n\t\tast.FlushBuffer(ctx, agentcontext.ResumeStatusFailed, testErr)\n\n\t\t// Verify messages were saved\n\t\tmessages, err := chatStore.GetMessages(chatID, storetypes.MessageFilter{})\n\t\tassert.NoError(t, err)\n\t\tassert.Len(t, messages, 1, \"Should have 1 message saved\")\n\n\t\t// Verify resume records were saved\n\t\tresumes, err := chatStore.GetResume(chatID)\n\t\tassert.NoError(t, err)\n\t\tassert.Len(t, resumes, 1, \"Should have 1 resume record on failure\")\n\t\tassert.Equal(t, agentcontext.ResumeStatusFailed, resumes[0].Status)\n\n\t\t// Cleanup\n\t\tchatStore.DeleteResume(chatID)\n\t\tchatStore.DeleteChat(chatID)\n\t\tt.Logf(\"✓ Buffer flushed on failure: messages and resume records saved\")\n\t})\n\n\tt.Run(\"FlushOnInterrupt\", func(t *testing.T) {\n\t\tchatID := fmt.Sprintf(\"test_flush_interrupt_%s\", uuid.New().String()[:8])\n\t\tctx := agentcontext.New(context.Background(), nil, chatID)\n\n\t\t// Enter stack and init buffer\n\t\t_, _, done := agentcontext.EnterStack(ctx, ast.ID, nil)\n\t\tdefer done()\n\t\tast.InitBuffer(ctx)\n\n\t\t// Ensure chat exists\n\t\terr := chatStore.CreateChat(&storetypes.Chat{\n\t\t\tChatID: chatID,\n\t\t\tAssistantID: ast.ID,\n\t\t\tStatus: \"active\",\n\t\t\tShare: \"private\",\n\t\t\tCreatedAt: time.Now(),\n\t\t\tUpdatedAt: time.Now(),\n\t\t})\n\t\trequire.NoError(t, err)\n\n\t\t// Add messages and steps\n\t\tctx.Buffer.AddUserInput(\"Test question\", \"\")\n\t\tast.BeginStep(ctx, agentcontext.StepTypeLLM, nil)\n\n\t\t// Flush buffer (interrupt case)\n\t\tast.FlushBuffer(ctx, agentcontext.ResumeStatusInterrupted, nil)\n\n\t\t// Verify resume records were saved with interrupted status\n\t\tresumes, err := chatStore.GetResume(chatID)\n\t\tassert.NoError(t, err)\n\t\tassert.Len(t, resumes, 1, \"Should have 1 resume record on interrupt\")\n\t\tassert.Equal(t, agentcontext.ResumeStatusInterrupted, resumes[0].Status)\n\n\t\t// Cleanup\n\t\tchatStore.DeleteResume(chatID)\n\t\tchatStore.DeleteChat(chatID)\n\t\tt.Logf(\"✓ Buffer flushed on interrupt: resume records saved with interrupted status\")\n\t})\n\n\tt.Run(\"FlushWithModeAndConnector\", func(t *testing.T) {\n\t\tchatID := fmt.Sprintf(\"test_flush_mode_%s\", uuid.New().String()[:8])\n\t\tctx := agentcontext.New(context.Background(), nil, chatID)\n\n\t\t// Enter stack with connector and mode options\n\t\topts := &agentcontext.Options{\n\t\t\tConnector: \"deepseek.v3\",\n\t\t\tMode: \"task\",\n\t\t}\n\t\t_, _, done := agentcontext.EnterStack(ctx, ast.ID, opts)\n\t\tdefer done()\n\t\tast.InitBuffer(ctx)\n\n\t\t// Verify buffer has correct connector and mode\n\t\trequire.NotNil(t, ctx.Buffer, \"Buffer should be initialized\")\n\t\tassert.Equal(t, \"deepseek.v3\", ctx.Buffer.Connector(), \"Buffer should have connector set\")\n\t\tassert.Equal(t, \"task\", ctx.Buffer.Mode(), \"Buffer should have mode set\")\n\n\t\t// Ensure chat exists\n\t\terr := chatStore.CreateChat(&storetypes.Chat{\n\t\t\tChatID: chatID,\n\t\t\tAssistantID: ast.ID,\n\t\t\tStatus: \"active\",\n\t\t\tShare: \"private\",\n\t\t\tCreatedAt: time.Now(),\n\t\t\tUpdatedAt: time.Now(),\n\t\t})\n\t\trequire.NoError(t, err)\n\n\t\t// Add some messages to buffer\n\t\tctx.Buffer.AddUserInput(\"Test question for mode\", \"\")\n\t\tctx.Buffer.AddAssistantMessage(\"M1\", \"text\", map[string]interface{}{\"content\": \"Test answer with mode\"}, \"\", \"\", ast.ID, nil)\n\n\t\t// Flush buffer\n\t\tast.FlushBuffer(ctx, agentcontext.StepStatusCompleted, nil)\n\n\t\t// Verify messages were saved with connector and mode\n\t\tmessages, err := chatStore.GetMessages(chatID, storetypes.MessageFilter{})\n\t\tassert.NoError(t, err)\n\t\tassert.Len(t, messages, 2, \"Should have 2 messages saved\")\n\n\t\t// Assistant message should have connector and mode\n\t\tvar assistantMsg *storetypes.Message\n\t\tfor _, msg := range messages {\n\t\t\tif msg.Role == \"assistant\" {\n\t\t\t\tassistantMsg = msg\n\t\t\t\tbreak\n\t\t\t}\n\t\t}\n\t\trequire.NotNil(t, assistantMsg, \"Should find assistant message\")\n\t\tassert.Equal(t, \"deepseek.v3\", assistantMsg.Connector, \"Message should have connector\")\n\t\tassert.Equal(t, \"task\", assistantMsg.Mode, \"Message should have mode\")\n\n\t\t// Verify chat was updated with last_connector and last_mode\n\t\tchat, err := chatStore.GetChat(chatID)\n\t\tassert.NoError(t, err)\n\t\tassert.Equal(t, \"deepseek.v3\", chat.LastConnector, \"Chat should have last_connector updated\")\n\t\tassert.Equal(t, \"task\", chat.LastMode, \"Chat should have last_mode updated\")\n\n\t\t// Cleanup\n\t\tchatStore.DeleteChat(chatID)\n\t\tt.Logf(\"✓ Buffer flushed with mode and connector: connector=%s, mode=%s\", chat.LastConnector, chat.LastMode)\n\t})\n}\n\nfunc TestEnsureChat(t *testing.T) {\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\tast, err := assistant.Get(\"mohe\")\n\trequire.NoError(t, err)\n\n\t// Skip if chat store not available\n\tchatStore := assistant.GetChatStore()\n\tif chatStore == nil {\n\t\tt.Skip(\"Chat store not configured, skipping EnsureChat tests\")\n\t}\n\n\tt.Run(\"CreateNewChat\", func(t *testing.T) {\n\t\tchatID := fmt.Sprintf(\"test_ensure_new_%s\", uuid.New().String()[:8])\n\t\tctx := agentcontext.New(context.Background(), nil, chatID)\n\n\t\t// Ensure chat creates it\n\t\terr := ast.EnsureChat(ctx)\n\t\tassert.NoError(t, err)\n\n\t\t// Verify chat was created\n\t\tchat, err := chatStore.GetChat(chatID)\n\t\tassert.NoError(t, err)\n\t\tassert.NotNil(t, chat)\n\t\tassert.Equal(t, chatID, chat.ChatID)\n\t\tassert.Equal(t, ast.ID, chat.AssistantID)\n\t\tassert.Equal(t, \"active\", chat.Status)\n\n\t\t// Cleanup\n\t\tchatStore.DeleteChat(chatID)\n\t\tt.Logf(\"✓ New chat created: %s\", chatID)\n\t})\n\n\tt.Run(\"SkipExistingChat\", func(t *testing.T) {\n\t\tchatID := fmt.Sprintf(\"test_ensure_exist_%s\", uuid.New().String()[:8])\n\t\tctx := agentcontext.New(context.Background(), nil, chatID)\n\n\t\t// Create chat first\n\t\terr := chatStore.CreateChat(&storetypes.Chat{\n\t\t\tChatID: chatID,\n\t\t\tAssistantID: ast.ID,\n\t\t\tTitle: \"Existing Chat\",\n\t\t\tStatus: \"active\",\n\t\t\tShare: \"private\",\n\t\t\tCreatedAt: time.Now(),\n\t\t\tUpdatedAt: time.Now(),\n\t\t})\n\t\trequire.NoError(t, err)\n\n\t\t// EnsureChat should not error\n\t\terr = ast.EnsureChat(ctx)\n\t\tassert.NoError(t, err)\n\n\t\t// Verify chat still has original title\n\t\tchat, err := chatStore.GetChat(chatID)\n\t\tassert.NoError(t, err)\n\t\tassert.Equal(t, \"Existing Chat\", chat.Title)\n\n\t\t// Cleanup\n\t\tchatStore.DeleteChat(chatID)\n\t\tt.Logf(\"✓ Existing chat preserved\")\n\t})\n\n\tt.Run(\"SkipEmptyChatID\", func(t *testing.T) {\n\t\tctx := agentcontext.New(context.Background(), nil, \"\")\n\n\t\t// Should not error with empty chat ID\n\t\terr := ast.EnsureChat(ctx)\n\t\tassert.NoError(t, err)\n\t\tt.Logf(\"✓ Empty chat ID handled gracefully\")\n\t})\n\n\tt.Run(\"CreateChatWithPermissions\", func(t *testing.T) {\n\t\tchatID := fmt.Sprintf(\"test_ensure_perm_%s\", uuid.New().String()[:8])\n\n\t\t// Create context with authorized info\n\t\tctx := agentcontext.New(context.Background(), &oauthtypes.AuthorizedInfo{\n\t\t\tUserID: \"test_user_001\",\n\t\t\tTeamID: \"test_team_001\",\n\t\t\tTenantID: \"test_tenant_001\",\n\t\t}, chatID)\n\n\t\t// EnsureChat should create with permission fields\n\t\terr := ast.EnsureChat(ctx)\n\t\tassert.NoError(t, err)\n\n\t\t// Verify permission fields were saved\n\t\tchat, err := chatStore.GetChat(chatID)\n\t\tassert.NoError(t, err)\n\t\tassert.NotNil(t, chat)\n\t\tassert.Equal(t, \"test_user_001\", chat.CreatedBy, \"CreatedBy should be set\")\n\t\tassert.Equal(t, \"test_user_001\", chat.UpdatedBy, \"UpdatedBy should be set\")\n\t\tassert.Equal(t, \"test_team_001\", chat.TeamID, \"TeamID should be set\")\n\t\tassert.Equal(t, \"test_tenant_001\", chat.TenantID, \"TenantID should be set\")\n\n\t\t// Cleanup\n\t\tchatStore.DeleteChat(chatID)\n\t\tt.Logf(\"✓ Chat created with permission fields: user=%s, team=%s, tenant=%s\",\n\t\t\tchat.CreatedBy, chat.TeamID, chat.TenantID)\n\t})\n\n\tt.Run(\"SkipHistoryEnabled\", func(t *testing.T) {\n\t\tchatID := fmt.Sprintf(\"test_ensure_skip_%s\", uuid.New().String()[:8])\n\n\t\t// Create context\n\t\tctx := agentcontext.New(context.Background(), nil, chatID)\n\n\t\t// Set up stack with Skip.History = true\n\t\tctx.Stack = &agentcontext.Stack{\n\t\t\tID: \"test_stack\",\n\t\t\tAssistantID: ast.ID,\n\t\t\tDepth: 0,\n\t\t\tOptions: &agentcontext.Options{\n\t\t\t\tSkip: &agentcontext.Skip{\n\t\t\t\t\tHistory: true,\n\t\t\t\t},\n\t\t\t},\n\t\t}\n\n\t\t// EnsureChat should NOT create chat when Skip.History is true\n\t\terr := ast.EnsureChat(ctx)\n\t\tassert.NoError(t, err)\n\n\t\t// Verify chat was NOT created\n\t\t_, err = chatStore.GetChat(chatID)\n\t\tassert.Error(t, err, \"Chat should not be created when Skip.History is true\")\n\t\tt.Logf(\"✓ Chat not created when Skip.History is true\")\n\t})\n}\n\n// TestEnsureChatMetadata verifies that ctx.Metadata is persisted to the chat record.\n// This is required for Host Agent: robot_id is passed in metadata so that\n// ListChats with chat_id_prefix=robot_{id}_ can filter by robot.\nfunc TestEnsureChatMetadata(t *testing.T) {\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\tast, err := assistant.Get(\"mohe\")\n\trequire.NoError(t, err)\n\n\tchatStore := assistant.GetChatStore()\n\tif chatStore == nil {\n\t\tt.Skip(\"Chat store not configured, skipping metadata tests\")\n\t}\n\n\tt.Run(\"MetadataPersisted\", func(t *testing.T) {\n\t\tchatID := fmt.Sprintf(\"robot_test_meta_%s\", uuid.New().String()[:8])\n\t\tctx := agentcontext.New(context.Background(), &oauthtypes.AuthorizedInfo{\n\t\t\tUserID: \"test_user_meta\",\n\t\t\tTeamID: \"test_team_meta\",\n\t\t}, chatID)\n\t\tctx.Metadata = map[string]interface{}{\n\t\t\t\"robot_id\": \"robot_member_001\",\n\t\t}\n\n\t\terr := ast.EnsureChat(ctx)\n\t\trequire.NoError(t, err)\n\n\t\tchat, err := chatStore.GetChat(chatID)\n\t\trequire.NoError(t, err)\n\t\trequire.NotNil(t, chat)\n\t\trequire.NotNil(t, chat.Metadata, \"Metadata should be persisted\")\n\t\tassert.Equal(t, \"robot_member_001\", chat.Metadata[\"robot_id\"],\n\t\t\t\"robot_id should be stored in chat metadata\")\n\n\t\t// Cleanup\n\t\tchatStore.DeleteChat(chatID)\n\t\tt.Logf(\"✓ Chat metadata persisted: robot_id=%v\", chat.Metadata[\"robot_id\"])\n\t})\n\n\tt.Run(\"MetadataPersistedWithRobotChatIDPrefix\", func(t *testing.T) {\n\t\t// Simulate robot host chat_id format: robot_{member_id}_{timestamp}\n\t\tmemberID := \"120004485525\"\n\t\tchatID := fmt.Sprintf(\"robot_%s_%d\", memberID, time.Now().UnixMilli())\n\t\tctx := agentcontext.New(context.Background(), &oauthtypes.AuthorizedInfo{\n\t\t\tUserID: \"test_user_robot\",\n\t\t\tTeamID: \"test_team_robot\",\n\t\t}, chatID)\n\t\tctx.Metadata = map[string]interface{}{\n\t\t\t\"robot_id\": memberID,\n\t\t}\n\n\t\terr := ast.EnsureChat(ctx)\n\t\trequire.NoError(t, err)\n\n\t\tchat, err := chatStore.GetChat(chatID)\n\t\trequire.NoError(t, err)\n\t\trequire.NotNil(t, chat)\n\t\trequire.NotNil(t, chat.Metadata)\n\t\tassert.Equal(t, memberID, chat.Metadata[\"robot_id\"])\n\n\t\t// Cleanup\n\t\tchatStore.DeleteChat(chatID)\n\t\tt.Logf(\"✓ Robot-prefix chat persisted with metadata: chat_id=%s\", chatID)\n\t})\n\n\tt.Run(\"NilMetadataHandled\", func(t *testing.T) {\n\t\tchatID := fmt.Sprintf(\"test_meta_nil_%s\", uuid.New().String()[:8])\n\t\tctx := agentcontext.New(context.Background(), nil, chatID)\n\t\tctx.Metadata = nil\n\n\t\terr := ast.EnsureChat(ctx)\n\t\tassert.NoError(t, err)\n\n\t\tchat, err := chatStore.GetChat(chatID)\n\t\trequire.NoError(t, err)\n\t\trequire.NotNil(t, chat)\n\t\t// Metadata nil is acceptable\n\t\tt.Logf(\"✓ Nil metadata handled gracefully\")\n\n\t\t// Cleanup\n\t\tchatStore.DeleteChat(chatID)\n\t})\n\n\tt.Run(\"MetadataMultipleFields\", func(t *testing.T) {\n\t\tchatID := fmt.Sprintf(\"test_meta_multi_%s\", uuid.New().String()[:8])\n\t\tctx := agentcontext.New(context.Background(), &oauthtypes.AuthorizedInfo{\n\t\t\tUserID: \"test_user_multi\",\n\t\t\tTeamID: \"test_team_multi\",\n\t\t}, chatID)\n\t\tctx.Metadata = map[string]interface{}{\n\t\t\t\"robot_id\": \"robot_multi_001\",\n\t\t\t\"source\": \"mission_control\",\n\t\t}\n\n\t\terr := ast.EnsureChat(ctx)\n\t\trequire.NoError(t, err)\n\n\t\tchat, err := chatStore.GetChat(chatID)\n\t\trequire.NoError(t, err)\n\t\trequire.NotNil(t, chat)\n\t\trequire.NotNil(t, chat.Metadata)\n\t\tassert.Equal(t, \"robot_multi_001\", chat.Metadata[\"robot_id\"])\n\t\tassert.Equal(t, \"mission_control\", chat.Metadata[\"source\"])\n\n\t\t// Cleanup\n\t\tchatStore.DeleteChat(chatID)\n\t\tt.Logf(\"✓ Multiple metadata fields persisted correctly\")\n\t})\n}\n\nfunc TestConvertBufferedTypes(t *testing.T) {\n\tt.Run(\"ConvertBufferedMessages\", func(t *testing.T) {\n\t\t// Create buffered messages\n\t\tbuffered := []*agentcontext.BufferedMessage{\n\t\t\t{\n\t\t\t\tMessageID: \"msg_001\",\n\t\t\t\tChatID: \"chat_001\",\n\t\t\t\tRequestID: \"req_001\",\n\t\t\t\tRole: \"user\",\n\t\t\t\tType: \"user_input\",\n\t\t\t\tProps: map[string]interface{}{\"content\": \"Hello\"},\n\t\t\t\tSequence: 1,\n\t\t\t\tCreatedAt: time.Now(),\n\t\t\t},\n\t\t\t{\n\t\t\t\tMessageID: \"msg_002\",\n\t\t\t\tChatID: \"chat_001\",\n\t\t\t\tRequestID: \"req_001\",\n\t\t\t\tRole: \"assistant\",\n\t\t\t\tType: \"text\",\n\t\t\t\tProps: map[string]interface{}{\"content\": \"Hi there!\"},\n\t\t\t\tBlockID: \"block_001\",\n\t\t\t\tAssistantID: \"test_assistant\",\n\t\t\t\tSequence: 2,\n\t\t\t\tCreatedAt: time.Now(),\n\t\t\t},\n\t\t}\n\n\t\t// Verify structure matches store types\n\t\tassert.Len(t, buffered, 2)\n\t\tassert.Equal(t, \"user\", buffered[0].Role)\n\t\tassert.Equal(t, \"assistant\", buffered[1].Role)\n\t\tassert.Equal(t, \"block_001\", buffered[1].BlockID)\n\t\tt.Logf(\"✓ Buffered messages have correct structure\")\n\t})\n\n\tt.Run(\"ConvertBufferedSteps\", func(t *testing.T) {\n\t\t// Create buffered steps\n\t\tbuffered := []*agentcontext.BufferedStep{\n\t\t\t{\n\t\t\t\tResumeID: \"resume_001\",\n\t\t\t\tChatID: \"chat_001\",\n\t\t\t\tRequestID: \"req_001\",\n\t\t\t\tAssistantID: \"test_assistant\",\n\t\t\t\tStackID: \"stack_001\",\n\t\t\t\tStackDepth: 0,\n\t\t\t\tType: agentcontext.StepTypeLLM,\n\t\t\t\tStatus: agentcontext.ResumeStatusFailed,\n\t\t\t\tInput: map[string]interface{}{\"messages\": []string{\"Hello\"}},\n\t\t\t\tSpaceSnapshot: map[string]interface{}{\"key\": \"value\"},\n\t\t\t\tError: \"Test error\",\n\t\t\t\tSequence: 1,\n\t\t\t\tCreatedAt: time.Now(),\n\t\t\t},\n\t\t}\n\n\t\t// Verify structure\n\t\tassert.Len(t, buffered, 1)\n\t\tassert.Equal(t, agentcontext.StepTypeLLM, buffered[0].Type)\n\t\tassert.Equal(t, agentcontext.ResumeStatusFailed, buffered[0].Status)\n\t\tassert.Equal(t, \"Test error\", buffered[0].Error)\n\t\tassert.Equal(t, \"value\", buffered[0].SpaceSnapshot[\"key\"])\n\t\tt.Logf(\"✓ Buffered steps have correct structure\")\n\t})\n}\n" }, { "path": "agent/assistant/handlers/stream.go", "content": "package handlers\n\nimport (\n\t\"time\"\n\n\tjsoniter \"github.com/json-iterator/go\"\n\t\"github.com/yaoapp/kun/log\"\n\t\"github.com/yaoapp/yao/agent/context\"\n\t\"github.com/yaoapp/yao/agent/i18n\"\n\t\"github.com/yaoapp/yao/agent/output\"\n\t\"github.com/yaoapp/yao/agent/output/message\"\n)\n\n// DefaultStreamHandler creates a default stream handler that sends messages via context\n// This handler is used when no custom handler is provided\nfunc DefaultStreamHandler(ctx *context.Context) message.StreamFunc {\n\n\t// Create stream state manager\n\tstate := &streamState{\n\t\tctx: ctx,\n\t\tinGroup: false,\n\t\tcurrentGroupID: \"\",\n\t\tmessageSeq: 0,\n\t}\n\n\treturn func(chunkType message.StreamChunkType, data []byte) int {\n\t\ttrace, _ := ctx.Trace()\n\t\tif trace != nil {\n\t\t\ttrace.Info(i18n.T(ctx.Locale, \"llm.handlers.stream.info\"), map[string]any{\"data\": string(data)})\n\t\t}\n\n\t\t// Handle different chunk types\n\t\tswitch chunkType {\n\t\tcase message.ChunkStreamStart:\n\t\t\treturn state.handleStreamStart(data)\n\n\t\tcase message.ChunkMessageStart:\n\t\t\treturn state.handleMessageStart(data)\n\n\t\tcase message.ChunkText:\n\t\t\treturn state.handleText(data)\n\n\t\tcase message.ChunkThinking:\n\t\t\treturn state.handleThinking(data)\n\n\t\tcase message.ChunkToolCall:\n\t\t\treturn state.handleToolCall(data)\n\n\t\tcase message.ChunkMetadata:\n\t\t\treturn state.handleMetadata(data)\n\n\t\tcase message.ChunkError:\n\t\t\treturn state.handleError(data)\n\n\t\tcase message.ChunkMessageEnd:\n\t\t\treturn state.handleMessageEnd(data)\n\n\t\tcase message.ChunkStreamEnd:\n\t\t\treturn state.handleStreamEnd(data)\n\n\t\tdefault:\n\t\t\t// Unknown chunk type, continue\n\t\t\treturn 0\n\t\t}\n\t}\n}\n\n// streamState manages the state of the streaming process\ntype streamState struct {\n\tctx *context.Context\n\tinGroup bool\n\tcurrentGroupID string // Current group ID (shared by all chunks in the group)\n\tcurrentType string // Track the current message type (text, thinking, tool_call)\n\tbuffer []byte\n\tchunkCount int // Track number of chunks in current group\n\tmessageSeq int // Message sequence number (for generating readable IDs)\n\tgroupStartTime time.Time // Track when group started\n}\n\n// handleStreamStart handles stream start event\nfunc (s *streamState) handleStreamStart(data []byte) int {\n\t// Send event message to indicate stream has started\n\t// This is a lifecycle event, CUI clients can show it, OpenAI clients will ignore it\n\tvar startData message.EventStreamStartData\n\terr := jsoniter.Unmarshal(data, &startData)\n\tif err != nil {\n\t\tlog.Error(\"Failed to unmarshal stream start data: %v\", err)\n\t}\n\tmsg := output.NewEventMessage(\"stream_start\", \"Stream started\", startData)\n\ts.ctx.Send(msg)\n\treturn 0\n}\n\n// handleMessageStart handles message start event\nfunc (s *streamState) handleMessageStart(data []byte) int {\n\t// Parse message start data first to get the message ID\n\tvar startData message.EventMessageStartData\n\tif err := jsoniter.Unmarshal(data, &startData); err != nil {\n\t\tlog.Error(\"Failed to unmarshal message start data: %v\", err)\n\t\treturn 0\n\t}\n\n\t// Use the message ID from the start data, or generate one if not provided\n\tmessageID := startData.MessageID\n\tif messageID == \"\" {\n\t\tmessageID = s.ctx.IDGenerator.GenerateMessageID()\n\t\tstartData.MessageID = messageID\n\t}\n\n\t// Auto-set ThreadID from Stack for nested agent calls\n\tif startData.ThreadID == \"\" && s.ctx.Stack != nil && !s.ctx.Stack.IsRoot() {\n\t\tstartData.ThreadID = s.ctx.Stack.ID\n\t}\n\n\t// Initialize message state with the correct message ID\n\ts.inGroup = true\n\ts.currentGroupID = messageID\n\ts.buffer = []byte{}\n\ts.chunkCount = 0\n\ts.messageSeq = 0 // Reset message sequence for each message\n\ts.groupStartTime = time.Now()\n\n\t// Send message_start event\n\tmsg := output.NewEventMessage(message.EventMessageStart, \"Message started\", startData)\n\ts.ctx.Send(msg)\n\n\treturn 0 // Continue\n}\n\n// handleText handles text content chunks\nfunc (s *streamState) handleText(data []byte) int {\n\tif len(data) == 0 {\n\t\treturn 0\n\t}\n\n\t// Track current message type\n\ts.currentType = message.TypeText\n\n\t// Append to buffer\n\ts.buffer = append(s.buffer, data...)\n\ts.chunkCount++\n\ts.messageSeq++\n\n\t// Send delta message\n\t// - ChunkID: Unique chunk ID (C1, C2, C3...) for this fragment\n\t// - MessageID: Same for all chunks of this logical message (frontend merges by message_id)\n\tmsg := &message.Message{\n\t\tChunkID: s.ctx.IDGenerator.GenerateChunkID(), // Unique chunk ID\n\t\tMessageID: s.currentGroupID, // Message ID for merging (all chunks share this)\n\t\tType: message.TypeText,\n\t\tDelta: true,\n\t\tProps: map[string]interface{}{\n\t\t\t\"content\": string(data),\n\t\t},\n\t}\n\n\tif err := s.ctx.Send(msg); err != nil {\n\t\t// Log error but continue streaming\n\t\treturn 0\n\t}\n\n\treturn 0 // Continue\n}\n\n// handleThinking handles thinking/reasoning chunks\nfunc (s *streamState) handleThinking(data []byte) int {\n\tif len(data) == 0 {\n\t\treturn 0\n\t}\n\n\t// Track current message type\n\ts.currentType = message.TypeThinking\n\n\t// Append to buffer\n\ts.buffer = append(s.buffer, data...)\n\ts.chunkCount++\n\ts.messageSeq++\n\n\t// Send delta message\n\t// - ChunkID: Unique chunk ID (C1, C2, C3...) for this fragment\n\t// - MessageID: Same for all chunks of this logical message (frontend merges by message_id)\n\tmsg := &message.Message{\n\t\tChunkID: s.ctx.IDGenerator.GenerateChunkID(), // Unique chunk ID\n\t\tMessageID: s.currentGroupID, // Message ID for merging (all chunks share this)\n\t\tType: message.TypeThinking,\n\t\tDelta: true,\n\t\tProps: map[string]interface{}{\n\t\t\t\"content\": string(data),\n\t\t},\n\t}\n\n\tif err := s.ctx.Send(msg); err != nil {\n\t\treturn 0\n\t}\n\n\treturn 0 // Continue\n}\n\n// handleToolCall handles tool call chunks\nfunc (s *streamState) handleToolCall(data []byte) int {\n\tif len(data) == 0 {\n\t\treturn 0\n\t}\n\n\t// Track current message type\n\ts.currentType = message.TypeToolCall\n\n\t// Append to buffer for message_end event\n\ts.buffer = append(s.buffer, data...)\n\ts.chunkCount++\n\ts.messageSeq++\n\n\t// Parse the tool call delta data (JSON array from OpenAI)\n\tvar toolCallArray []map[string]interface{}\n\tif err := jsoniter.Unmarshal(data, &toolCallArray); err != nil {\n\t\t// If parse fails, log and skip this chunk\n\t\treturn 0\n\t}\n\n\t// Extract tool call fields from delta\n\t// OpenAI delta typically has one element, but we handle arrays safely\n\tvar props map[string]interface{}\n\tvar deltaAction string\n\tvar deltaPath string\n\n\tif len(toolCallArray) == 1 {\n\t\ttc := toolCallArray[0]\n\t\tprops = map[string]interface{}{}\n\n\t\thasIdentity := false\n\t\tif id, ok := tc[\"id\"].(string); ok {\n\t\t\tprops[\"id\"] = id\n\t\t\thasIdentity = true\n\t\t}\n\t\tif typ, ok := tc[\"type\"].(string); ok {\n\t\t\tprops[\"type\"] = typ\n\t\t\thasIdentity = true\n\t\t}\n\t\tif index, ok := tc[\"index\"].(float64); ok {\n\t\t\tprops[\"index\"] = int(index)\n\t\t}\n\t\tif fn, ok := tc[\"function\"].(map[string]interface{}); ok {\n\t\t\tif name, ok := fn[\"name\"].(string); ok {\n\t\t\t\tprops[\"name\"] = name\n\t\t\t\thasIdentity = true\n\t\t\t}\n\t\t\tif args, ok := fn[\"arguments\"].(string); ok {\n\t\t\t\tprops[\"arguments\"] = args\n\t\t\t}\n\t\t}\n\n\t\tif hasIdentity {\n\t\t\t// First chunk with id/name/type: merge so all fields are applied.\n\t\t\tdeltaAction = \"merge\"\n\t\t} else if _, ok := props[\"arguments\"]; ok {\n\t\t\t// Subsequent chunk with only arguments fragment: append to arguments.\n\t\t\tdeltaAction = \"append\"\n\t\t\tdeltaPath = \"arguments\"\n\t\t} else {\n\t\t\tdeltaAction = \"merge\"\n\t\t}\n\t} else {\n\t\t// Multiple tool calls in delta (rare) - keep as array\n\t\tprops = map[string]interface{}{\n\t\t\t\"calls\": toolCallArray,\n\t\t}\n\t\tdeltaAction = \"merge\"\n\t}\n\n\t// Send delta message\n\t// - ChunkID: Unique chunk ID (C1, C2, C3...) for this fragment\n\t// - MessageID: Same for all chunks of this logical message (frontend merges by message_id)\n\t// - DeltaAction: \"append\" for arguments chunks, \"merge\" for id/type/name chunks\n\t// - DeltaPath: \"arguments\" when appending arguments field\n\t// OpenAI sends: first chunk has id/type/name, subsequent chunks only have arguments fragments\n\tmsg := &message.Message{\n\t\tChunkID: s.ctx.IDGenerator.GenerateChunkID(), // Unique chunk ID\n\t\tMessageID: s.currentGroupID, // Message ID for merging (all chunks share this)\n\t\tType: message.TypeToolCall,\n\t\tDelta: true,\n\t\tDeltaAction: deltaAction, // \"append\" for arguments, \"merge\" for static fields\n\t\tDeltaPath: deltaPath, // \"arguments\" when appending\n\t\tProps: props, // Flattened tool call fields\n\t}\n\n\tif err := s.ctx.Send(msg); err != nil {\n\t\treturn 0\n\t}\n\n\treturn 0 // Continue\n}\n\n// handleMetadata handles metadata chunks (usage, finish_reason, etc.)\nfunc (s *streamState) handleMetadata(data []byte) int {\n\t// Metadata is usually not displayed to users\n\t// Could be logged or stored for analytics\n\treturn 0 // Continue\n}\n\n// handleError handles error chunks\nfunc (s *streamState) handleError(data []byte) int {\n\t// Send error message\n\tmsg := output.NewErrorMessage(string(data), \"stream_error\")\n\ts.ctx.Send(msg)\n\n\treturn 1 // Stop streaming on error\n}\n\n// handleMessageEnd handles message end event\nfunc (s *streamState) handleMessageEnd(data []byte) int {\n\tif !s.inGroup {\n\t\treturn 0\n\t}\n\n\t// Calculate duration\n\tdurationMs := time.Since(s.groupStartTime).Milliseconds()\n\n\t// Use the tracked message type (thinking, text, tool_call, etc.)\n\tmsgType := s.currentType\n\tif msgType == \"\" {\n\t\tmsgType = message.TypeText // Fallback to text if type not set\n\t}\n\n\t// Get ThreadID from Stack for nested agent calls\n\tvar threadID string\n\tif s.ctx.Stack != nil && !s.ctx.Stack.IsRoot() {\n\t\tthreadID = s.ctx.Stack.ID\n\t}\n\n\t// Get BlockID from metadata if available\n\tvar blockID string\n\tif s.ctx != nil {\n\t\tif metadata := s.ctx.GetMessageMetadata(s.currentGroupID); metadata != nil {\n\t\t\tblockID = metadata.BlockID\n\t\t}\n\t}\n\n\t// Buffer the complete LLM message for storage\n\t// Delta chunks are not stored, but we need to save the final complete content\n\t// Skip if History is disabled in options\n\tshouldSkipHistory := s.ctx.Stack != nil && s.ctx.Stack.Options != nil &&\n\t\ts.ctx.Stack.Options.Skip != nil && s.ctx.Stack.Options.Skip.History\n\n\tif s.ctx.Buffer != nil && len(s.buffer) > 0 && !shouldSkipHistory {\n\t\tassistantID := \"\"\n\t\tif s.ctx.Stack != nil {\n\t\t\tassistantID = s.ctx.Stack.AssistantID\n\t\t}\n\n\t\t// Build props based on message type\n\t\tvar props map[string]interface{}\n\t\tif msgType == message.TypeToolCall {\n\t\t\t// For tool calls, try to parse the accumulated buffer as JSON\n\t\t\tvar toolCallData interface{}\n\t\t\tif err := jsoniter.Unmarshal(s.buffer, &toolCallData); err == nil {\n\t\t\t\tprops = map[string]interface{}{\n\t\t\t\t\t\"calls\": toolCallData,\n\t\t\t\t}\n\t\t\t} else {\n\t\t\t\tprops = map[string]interface{}{\n\t\t\t\t\t\"content\": string(s.buffer),\n\t\t\t\t}\n\t\t\t}\n\t\t} else {\n\t\t\t// For text/thinking, content is the accumulated text\n\t\t\tprops = map[string]interface{}{\n\t\t\t\t\"content\": string(s.buffer),\n\t\t\t}\n\t\t}\n\n\t\ts.ctx.Buffer.AddAssistantMessage(\n\t\t\ts.currentGroupID, // Use the message ID\n\t\t\tmsgType,\n\t\t\tprops,\n\t\t\tblockID,\n\t\t\tthreadID,\n\t\t\tassistantID,\n\t\t\tnil,\n\t\t)\n\t}\n\n\t// Build EventMessageEndData with complete content\n\tendData := message.EventMessageEndData{\n\t\tMessageID: s.currentGroupID, // Use the message ID\n\t\tType: msgType,\n\t\tTimestamp: time.Now().UnixMilli(),\n\t\tThreadID: threadID, // Include ThreadID for concurrent stream identification\n\t\tDurationMs: durationMs,\n\t\tChunkCount: s.chunkCount,\n\t\tStatus: \"completed\",\n\t\tExtra: map[string]interface{}{\n\t\t\t\"content\": string(s.buffer), // Include complete content in the event\n\t\t},\n\t}\n\n\t// Send message_end event\n\tmsg := output.NewEventMessage(message.EventMessageEnd, \"Message completed\", endData)\n\ts.ctx.Send(msg)\n\n\t// Reset state\n\ts.inGroup = false\n\ts.currentGroupID = \"\"\n\ts.currentType = \"\"\n\ts.buffer = []byte{}\n\ts.chunkCount = 0\n\n\treturn 0 // Continue\n}\n\n// handleStreamEnd handles stream end event\nfunc (s *streamState) handleStreamEnd(data []byte) int {\n\t// Parse the stream end data\n\tvar endData message.EventStreamEndData\n\tif err := jsoniter.Unmarshal(data, &endData); err != nil {\n\t\tlog.Error(\"Failed to parse stream_end data: %v\", err)\n\t\ts.ctx.Flush()\n\t\treturn 0\n\t}\n\n\t// Send stream_end event as a message to frontend\n\tmsg := output.NewEventMessage(\"stream_end\", \"Stream completed\", endData)\n\ts.ctx.Send(msg)\n\n\t// Flush any remaining data\n\ts.ctx.Flush()\n\treturn 0 // Continue (stream will end naturally)\n}\n" }, { "path": "agent/assistant/history.go", "content": "package assistant\n\nimport (\n\t\"fmt\"\n\t\"reflect\"\n\n\tjsoniter \"github.com/json-iterator/go\"\n\tagentcontext \"github.com/yaoapp/yao/agent/context\"\n\tstoretypes \"github.com/yaoapp/yao/agent/store/types\"\n\t\"github.com/yaoapp/yao/trace/types\"\n)\n\n// =============================================================================\n// Chat History Management\n// =============================================================================\n\n// HistoryResult represents the result of history processing\ntype HistoryResult struct {\n\tInputMessages []agentcontext.Message // Clean input messages (without overlap)\n\tFullMessages []agentcontext.Message // Full messages (history + clean input)\n}\n\n// getHistorySize returns the history size with priority: opts.HistorySize > storeSetting.MaxSize > default (20)\nfunc getHistorySize(opts *agentcontext.Options) int {\n\tconst defaultHistorySize = 20\n\tif opts != nil && opts.HistorySize > 0 {\n\t\treturn opts.HistorySize\n\t}\n\tif setting := GetStoreSetting(); setting != nil && setting.MaxSize > 0 {\n\t\treturn setting.MaxSize\n\t}\n\treturn defaultHistorySize\n}\n\n// WithHistory merges the input messages with chat history and traces it\n// Returns HistoryResult containing:\n// - InputMessages: cleaned input (overlap removed)\n// - FullMessages: history + clean input merged\nfunc (ast *Assistant) WithHistory(ctx *agentcontext.Context, input []agentcontext.Message, agentNode types.Node, options ...*agentcontext.Options) (*HistoryResult, error) {\n\n\t// Get options\n\tvar opts *agentcontext.Options\n\tif len(options) > 0 && options[0] != nil {\n\t\topts = options[0]\n\t}\n\n\t// SKIP: History (for internal calls like title/prompt etc.)\n\tif opts != nil && opts.Skip != nil && opts.Skip.History {\n\t\tresult := &HistoryResult{\n\t\t\tInputMessages: input,\n\t\t\tFullMessages: input,\n\t\t}\n\t\tast.traceAgentHistory(ctx, agentNode, result.FullMessages)\n\t\treturn result, nil\n\t}\n\n\t// Resolve history size: opts.HistorySize > storeSetting.MaxSize > default (20)\n\tmaxSize := getHistorySize(opts)\n\n\t// Load history from store\n\thistoryMessages, err := ast.loadHistory(ctx, maxSize)\n\tif err != nil {\n\t\t// Log warning but continue without history\n\t\tctx.Logger.Warn(\"Failed to load history for chat=%s: %v\", ctx.ChatID, err)\n\t\tresult := &HistoryResult{\n\t\t\tInputMessages: input,\n\t\t\tFullMessages: input,\n\t\t}\n\t\tast.traceAgentHistory(ctx, agentNode, result.FullMessages)\n\t\treturn result, nil\n\t}\n\n\t// If no history, return input as is\n\tif len(historyMessages) == 0 {\n\t\tctx.Logger.HistoryLoad(0, maxSize)\n\t\tresult := &HistoryResult{\n\t\t\tInputMessages: input,\n\t\t\tFullMessages: input,\n\t\t}\n\t\tast.traceAgentHistory(ctx, agentNode, result.FullMessages)\n\t\treturn result, nil\n\t}\n\n\t// Log history loaded\n\tctx.Logger.HistoryLoad(len(historyMessages), maxSize)\n\n\t// Find overlap between history and input\n\t// Some external clients may include history in their requests\n\toverlapIndex := ast.findOverlapIndex(historyMessages, input)\n\n\t// Remove overlap from input\n\tcleanInput := input\n\tif overlapIndex > 0 {\n\t\tcleanInput = input[overlapIndex:]\n\t\tctx.Logger.HistoryOverlap(overlapIndex)\n\t}\n\n\t// Merge history with clean input\n\tfullMessages := make([]agentcontext.Message, 0, len(historyMessages)+len(cleanInput))\n\tfullMessages = append(fullMessages, historyMessages...)\n\tfullMessages = append(fullMessages, cleanInput...)\n\n\tresult := &HistoryResult{\n\t\tInputMessages: cleanInput,\n\t\tFullMessages: fullMessages,\n\t}\n\n\t// Log the chat history\n\tast.traceAgentHistory(ctx, agentNode, result.FullMessages)\n\n\treturn result, nil\n}\n\n// loadHistory loads chat history from the store\n// Returns the most recent maxSize messages, ordered by time (oldest first)\nfunc (ast *Assistant) loadHistory(ctx *agentcontext.Context, maxSize int) ([]agentcontext.Message, error) {\n\t// Check if chat ID is available\n\tif ctx.ChatID == \"\" {\n\t\treturn nil, nil\n\t}\n\n\t// Get chat store\n\tchatStore := GetChatStore()\n\tif chatStore == nil {\n\t\treturn nil, nil\n\t}\n\n\t// Load messages from store with limit\n\tfilter := storetypes.MessageFilter{\n\t\tLimit: maxSize,\n\t}\n\n\tstoreMessages, err := chatStore.GetMessages(ctx.ChatID, filter)\n\tif err != nil {\n\t\treturn nil, fmt.Errorf(\"failed to get messages: %w\", err)\n\t}\n\n\tif len(storeMessages) == 0 {\n\t\treturn nil, nil\n\t}\n\n\t// Convert store messages to context messages\n\tmessages := make([]agentcontext.Message, 0, len(storeMessages))\n\tfor _, msg := range storeMessages {\n\t\t// Only include user and assistant messages for LLM context\n\t\t// Skip internal types like loading, event, etc.\n\t\tif msg.Role != \"user\" && msg.Role != \"assistant\" {\n\t\t\tcontinue\n\t\t}\n\n\t\t// Convert store message to context message\n\t\tctxMsg := ast.convertStoreMessageToContext(msg)\n\t\tif ctxMsg != nil {\n\t\t\tmessages = append(messages, *ctxMsg)\n\t\t}\n\t}\n\n\treturn messages, nil\n}\n\n// convertStoreMessageToContext converts a store message to a context message\nfunc (ast *Assistant) convertStoreMessageToContext(msg *storetypes.Message) *agentcontext.Message {\n\tif msg == nil {\n\t\treturn nil\n\t}\n\n\t// Handle special message types:\n\t// - tool_call/action: convert to historical summary text for LLM context\n\t// - loading/event: skip (pure UI/lifecycle signals, no semantic value)\n\t// - error: kept as-is so LLM can help troubleshoot issues\n\tswitch msg.Type {\n\tcase \"tool_call\":\n\t\treturn ast.convertToolCallToContext(msg)\n\tcase \"action\":\n\t\treturn ast.convertActionToContext(msg)\n\tcase \"loading\", \"event\":\n\t\treturn nil\n\t}\n\n\t// Extract content from Props\n\tcontent := ast.extractContentFromProps(msg.Props, msg.Type)\n\tif content == nil {\n\t\treturn nil\n\t}\n\n\t// Build context message\n\tctxMsg := &agentcontext.Message{\n\t\tRole: agentcontext.MessageRole(msg.Role),\n\t\tContent: content,\n\t}\n\n\t// Handle name field\n\tif msg.Props != nil {\n\t\tif name, ok := msg.Props[\"name\"].(string); ok && name != \"\" {\n\t\t\tctxMsg.Name = &name\n\t\t}\n\t}\n\n\treturn ctxMsg\n}\n\n// extractContentFromProps extracts the content from message Props based on message type\nfunc (ast *Assistant) extractContentFromProps(props map[string]interface{}, msgType string) interface{} {\n\tif props == nil {\n\t\treturn nil\n\t}\n\n\t// For user input, content is stored directly in props[\"content\"]\n\tif msgType == \"user_input\" {\n\t\treturn props[\"content\"]\n\t}\n\n\t// For text type messages\n\tif msgType == \"text\" {\n\t\tif text, ok := props[\"text\"].(string); ok {\n\t\t\treturn text\n\t\t}\n\t\t// Also try content field\n\t\tif content, ok := props[\"content\"].(string); ok {\n\t\t\treturn content\n\t\t}\n\t}\n\n\t// For other types, try to extract content or text\n\tif content, ok := props[\"content\"]; ok {\n\t\treturn content\n\t}\n\tif text, ok := props[\"text\"]; ok {\n\t\treturn text\n\t}\n\n\treturn nil\n}\n\n// convertToolCallToContext converts a tool_call store message to a historical summary text message.\n// This allows the LLM to understand what tools were previously called without re-invoking them.\n//\n// Supports two Props formats:\n// - Standard ToolCallProps: {\"name\": \"tool_name\", \"arguments\": \"{...}\"}\n// - Raw stream chunks: {\"content\": \"[{\\\"index\\\":0,\\\"id\\\":\\\"call_...\\\",\\\"function\\\":{\\\"name\\\":\\\"tool\\\"}}][...]\"}\nfunc (ast *Assistant) convertToolCallToContext(msg *storetypes.Message) *agentcontext.Message {\n\tif msg.Props == nil {\n\t\treturn nil\n\t}\n\n\t// Try standard ToolCallProps format first\n\tif name, ok := msg.Props[\"name\"].(string); ok && name != \"\" {\n\t\targs, _ := msg.Props[\"arguments\"].(string)\n\t\tconst maxArgsLen = 500\n\t\tif len(args) > maxArgsLen {\n\t\t\targs = args[:maxArgsLen] + \"...\"\n\t\t}\n\t\treturn &agentcontext.Message{\n\t\t\tRole: agentcontext.RoleAssistant,\n\t\t\tContent: fmt.Sprintf(\"[Historical Tool Call Summary] Called tool \\\"%s\\\" with arguments: %s\", name, args),\n\t\t}\n\t}\n\n\t// Try raw stream chunk format: {\"content\": \"[...][...]...\"}\n\t// Each chunk is a JSON array like [{\"index\":0,\"id\":\"call_...\",\"function\":{\"name\":\"echo__ping\"}}]\n\t// Subsequent chunks append arguments: [{\"index\":0,\"function\":{\"arguments\":\"...\"}}]\n\tif raw, ok := msg.Props[\"content\"].(string); ok && raw != \"\" {\n\t\tname, args := parseToolCallRawChunks(raw)\n\t\tif name == \"\" {\n\t\t\treturn nil\n\t\t}\n\t\tconst maxArgsLen = 500\n\t\tif len(args) > maxArgsLen {\n\t\t\targs = args[:maxArgsLen] + \"...\"\n\t\t}\n\t\treturn &agentcontext.Message{\n\t\t\tRole: agentcontext.RoleAssistant,\n\t\t\tContent: fmt.Sprintf(\"[Historical Tool Call Summary] Called tool \\\"%s\\\" with arguments: %s\", name, args),\n\t\t}\n\t}\n\n\treturn nil\n}\n\n// parseToolCallRawChunks parses concatenated raw stream chunks to extract tool name and arguments.\n// Input format: \"[{...}][{...}][{...}]\" — multiple JSON arrays concatenated without separator.\nfunc parseToolCallRawChunks(raw string) (name, args string) {\n\t// Split concatenated JSON arrays: \"][\" is the boundary\n\t// e.g. \"[{...}][{...}]\" → [\"[{...}]\", \"[{...}]\"]\n\tchunks := splitJSONArrays(raw)\n\n\tvar argParts []string\n\tfor _, chunk := range chunks {\n\t\tvar items []map[string]interface{}\n\t\tif err := jsoniter.UnmarshalFromString(chunk, &items); err != nil || len(items) == 0 {\n\t\t\tcontinue\n\t\t}\n\t\titem := items[0]\n\t\tif fn, ok := item[\"function\"].(map[string]interface{}); ok {\n\t\t\tif n, ok := fn[\"name\"].(string); ok && n != \"\" && name == \"\" {\n\t\t\t\tname = n\n\t\t\t}\n\t\t\tif a, ok := fn[\"arguments\"].(string); ok && a != \"\" {\n\t\t\t\targParts = append(argParts, a)\n\t\t\t}\n\t\t}\n\t}\n\n\targs = \"\"\n\tfor _, part := range argParts {\n\t\targs += part\n\t}\n\treturn name, args\n}\n\n// splitJSONArrays splits a string of concatenated JSON arrays \"[...][...][...]\" into individual arrays.\nfunc splitJSONArrays(s string) []string {\n\tvar result []string\n\tdepth := 0\n\tstart := -1\n\tfor i, ch := range s {\n\t\tswitch ch {\n\t\tcase '[':\n\t\t\tif depth == 0 {\n\t\t\t\tstart = i\n\t\t\t}\n\t\t\tdepth++\n\t\tcase ']':\n\t\t\tdepth--\n\t\t\tif depth == 0 && start >= 0 {\n\t\t\t\tresult = append(result, s[start:i+1])\n\t\t\t\tstart = -1\n\t\t\t}\n\t\t}\n\t}\n\treturn result\n}\n\n// convertActionToContext converts an action store message to a historical summary text message.\n// This allows the LLM to understand what system actions were previously executed.\nfunc (ast *Assistant) convertActionToContext(msg *storetypes.Message) *agentcontext.Message {\n\tif msg.Props == nil {\n\t\treturn nil\n\t}\n\tname, _ := msg.Props[\"name\"].(string)\n\tif name == \"\" {\n\t\treturn nil\n\t}\n\tpayload := \"\"\n\tif msg.Props[\"payload\"] != nil {\n\t\tif payloadStr, err := jsoniter.MarshalToString(msg.Props[\"payload\"]); err == nil {\n\t\t\tconst maxPayloadLen = 500\n\t\t\tif len(payloadStr) > maxPayloadLen {\n\t\t\t\tpayloadStr = payloadStr[:maxPayloadLen] + \"...\"\n\t\t\t}\n\t\t\tpayload = payloadStr\n\t\t}\n\t}\n\tif payload != \"\" {\n\t\treturn &agentcontext.Message{\n\t\t\tRole: agentcontext.RoleAssistant,\n\t\t\tContent: fmt.Sprintf(\"[Historical Action Summary] Executed action \\\"%s\\\" with payload: %s\", name, payload),\n\t\t}\n\t}\n\treturn &agentcontext.Message{\n\t\tRole: agentcontext.RoleAssistant,\n\t\tContent: fmt.Sprintf(\"[Historical Action Summary] Executed action \\\"%s\\\"\", name),\n\t}\n}\n\n// findOverlapIndex finds the index in input where history messages end\n// Returns the number of input messages that overlap with history\nfunc (ast *Assistant) findOverlapIndex(history, input []agentcontext.Message) int {\n\tif len(history) == 0 || len(input) == 0 {\n\t\treturn 0\n\t}\n\n\t// We need to find the longest suffix of history that matches a prefix of input\n\t// Start from the end of history and try to match with the beginning of input\n\n\tmaxOverlap := len(history)\n\tif maxOverlap > len(input) {\n\t\tmaxOverlap = len(input)\n\t}\n\n\t// Try different overlap lengths, starting from the largest possible\n\tfor overlapLen := maxOverlap; overlapLen > 0; overlapLen-- {\n\t\t// Check if the last 'overlapLen' messages of history match the first 'overlapLen' of input\n\t\thistoryStart := len(history) - overlapLen\n\t\tmatched := true\n\n\t\tfor i := 0; i < overlapLen; i++ {\n\t\t\tif !ast.messagesMatch(history[historyStart+i], input[i]) {\n\t\t\t\tmatched = false\n\t\t\t\tbreak\n\t\t\t}\n\t\t}\n\n\t\tif matched {\n\t\t\treturn overlapLen\n\t\t}\n\t}\n\n\treturn 0\n}\n\n// messagesMatch checks if two messages are equivalent\nfunc (ast *Assistant) messagesMatch(a, b agentcontext.Message) bool {\n\t// Must have same role\n\tif a.Role != b.Role {\n\t\treturn false\n\t}\n\n\t// Compare content\n\treturn ast.contentMatches(a.Content, b.Content)\n}\n\n// contentMatches compares two content values for equality\nfunc (ast *Assistant) contentMatches(a, b interface{}) bool {\n\t// Handle nil cases\n\tif a == nil && b == nil {\n\t\treturn true\n\t}\n\tif a == nil || b == nil {\n\t\treturn false\n\t}\n\n\t// If both are strings, compare directly\n\taStr, aIsStr := a.(string)\n\tbStr, bIsStr := b.(string)\n\tif aIsStr && bIsStr {\n\t\treturn aStr == bStr\n\t}\n\n\t// For complex content (arrays, etc.), use deep equal\n\treturn reflect.DeepEqual(a, b)\n}\n" }, { "path": "agent/assistant/history_test.go", "content": "package assistant_test\n\nimport (\n\t\"context\"\n\t\"fmt\"\n\t\"testing\"\n\t\"time\"\n\n\t\"github.com/google/uuid\"\n\t\"github.com/stretchr/testify/assert\"\n\t\"github.com/stretchr/testify/require\"\n\t\"github.com/yaoapp/yao/agent/assistant\"\n\tagentcontext \"github.com/yaoapp/yao/agent/context\"\n\t\"github.com/yaoapp/yao/agent/output/message\"\n\tstoretypes \"github.com/yaoapp/yao/agent/store/types\"\n\t\"github.com/yaoapp/yao/agent/testutils\"\n\toauthtypes \"github.com/yaoapp/yao/openapi/oauth/types\"\n)\n\n// =============================================================================\n// Helper Functions\n// =============================================================================\n\n// newHistoryTestContext creates a test context for history tests\nfunc newHistoryTestContext(chatID string) *agentcontext.Context {\n\tauthorized := &oauthtypes.AuthorizedInfo{\n\t\tSubject: \"test-user\",\n\t\tUserID: \"history-test-user\",\n\t\tTeamID: \"history-test-team\",\n\t\tTenantID: \"history-test-tenant\",\n\t}\n\n\tctx := agentcontext.New(context.Background(), authorized, chatID)\n\tctx.AssistantID = \"tests.history\"\n\tctx.Locale = \"en-us\"\n\tctx.Client = agentcontext.Client{\n\t\tType: \"web\",\n\t\tIP: \"127.0.0.1\",\n\t}\n\tctx.Referer = agentcontext.RefererAPI\n\tctx.Accept = agentcontext.AcceptWebCUI\n\tctx.IDGenerator = message.NewIDGenerator()\n\tctx.Metadata = make(map[string]interface{})\n\treturn ctx\n}\n\n// =============================================================================\n// WithHistory Tests\n// =============================================================================\n\nfunc TestWithHistory(t *testing.T) {\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\t// Get assistant\n\tast, err := assistant.Get(\"tests.history\")\n\trequire.NoError(t, err)\n\trequire.NotNil(t, ast)\n\n\t// Get chat store for setup/cleanup\n\tchatStore := assistant.GetChatStore()\n\tif chatStore == nil {\n\t\tt.Skip(\"Chat store not configured, skipping history tests\")\n\t}\n\n\tt.Run(\"NoHistory\", func(t *testing.T) {\n\t\tchatID := fmt.Sprintf(\"test_history_none_%s\", uuid.New().String()[:8])\n\t\tctx := newHistoryTestContext(chatID)\n\n\t\t// Create chat without any messages\n\t\terr := chatStore.CreateChat(&storetypes.Chat{\n\t\t\tChatID: chatID,\n\t\t\tAssistantID: ast.ID,\n\t\t\tStatus: \"active\",\n\t\t\tShare: \"private\",\n\t\t\tCreatedAt: time.Now(),\n\t\t\tUpdatedAt: time.Now(),\n\t\t})\n\t\trequire.NoError(t, err)\n\t\tdefer chatStore.DeleteChat(chatID)\n\n\t\tinput := []agentcontext.Message{\n\t\t\t{Role: agentcontext.RoleUser, Content: \"Hello, this is my first message\"},\n\t\t}\n\n\t\tresult, err := ast.WithHistory(ctx, input, nil)\n\t\trequire.NoError(t, err)\n\t\trequire.NotNil(t, result)\n\n\t\t// With no history, InputMessages and FullMessages should be the same as input\n\t\tassert.Equal(t, input, result.InputMessages)\n\t\tassert.Equal(t, input, result.FullMessages)\n\t\tt.Log(\"✓ No history: input returned as is\")\n\t})\n\n\tt.Run(\"WithExistingHistory\", func(t *testing.T) {\n\t\tchatID := fmt.Sprintf(\"test_history_exist_%s\", uuid.New().String()[:8])\n\t\tctx := newHistoryTestContext(chatID)\n\t\treqID := uuid.New().String()[:8]\n\n\t\t// Create chat\n\t\terr := chatStore.CreateChat(&storetypes.Chat{\n\t\t\tChatID: chatID,\n\t\t\tAssistantID: ast.ID,\n\t\t\tStatus: \"active\",\n\t\t\tShare: \"private\",\n\t\t\tCreatedAt: time.Now(),\n\t\t\tUpdatedAt: time.Now(),\n\t\t})\n\t\trequire.NoError(t, err)\n\t\tdefer func() {\n\t\t\tchatStore.DeleteMessages(chatID, nil)\n\t\t\tchatStore.DeleteChat(chatID)\n\t\t}()\n\n\t\t// Add history messages\n\t\thistoryMessages := []*storetypes.Message{\n\t\t\t{\n\t\t\t\tMessageID: fmt.Sprintf(\"hist_msg_1_%s\", reqID),\n\t\t\t\tChatID: chatID,\n\t\t\t\tRequestID: fmt.Sprintf(\"req_%s\", reqID),\n\t\t\t\tRole: \"user\",\n\t\t\t\tType: \"user_input\",\n\t\t\t\tProps: map[string]interface{}{\"content\": \"Previous question\"},\n\t\t\t\tSequence: 1,\n\t\t\t\tAssistantID: ast.ID,\n\t\t\t\tCreatedAt: time.Now().Add(-2 * time.Minute),\n\t\t\t},\n\t\t\t{\n\t\t\t\tMessageID: fmt.Sprintf(\"hist_msg_2_%s\", reqID),\n\t\t\t\tChatID: chatID,\n\t\t\t\tRequestID: fmt.Sprintf(\"req_%s\", reqID),\n\t\t\t\tRole: \"assistant\",\n\t\t\t\tType: \"text\",\n\t\t\t\tProps: map[string]interface{}{\"text\": \"Previous answer\"},\n\t\t\t\tSequence: 2,\n\t\t\t\tAssistantID: ast.ID,\n\t\t\t\tCreatedAt: time.Now().Add(-1 * time.Minute),\n\t\t\t},\n\t\t}\n\n\t\terr = chatStore.SaveMessages(chatID, historyMessages)\n\t\trequire.NoError(t, err)\n\n\t\t// New input message\n\t\tinput := []agentcontext.Message{\n\t\t\t{Role: agentcontext.RoleUser, Content: \"New question\"},\n\t\t}\n\n\t\tresult, err := ast.WithHistory(ctx, input, nil)\n\t\trequire.NoError(t, err)\n\t\trequire.NotNil(t, result)\n\n\t\t// InputMessages should be unchanged (no overlap)\n\t\tassert.Equal(t, input, result.InputMessages)\n\n\t\t// FullMessages should have history + input\n\t\tassert.Len(t, result.FullMessages, 3) // 2 history + 1 new\n\n\t\t// Verify order: history first, then input\n\t\tassert.Equal(t, agentcontext.RoleUser, result.FullMessages[0].Role)\n\t\tassert.Equal(t, \"Previous question\", result.FullMessages[0].Content)\n\t\tassert.Equal(t, agentcontext.RoleAssistant, result.FullMessages[1].Role)\n\t\tassert.Equal(t, \"Previous answer\", result.FullMessages[1].Content)\n\t\tassert.Equal(t, agentcontext.RoleUser, result.FullMessages[2].Role)\n\t\tassert.Equal(t, \"New question\", result.FullMessages[2].Content)\n\n\t\tt.Log(\"✓ History merged correctly with new input\")\n\t})\n\n\tt.Run(\"SkipHistoryOption\", func(t *testing.T) {\n\t\tchatID := fmt.Sprintf(\"test_history_skip_%s\", uuid.New().String()[:8])\n\t\tctx := newHistoryTestContext(chatID)\n\t\treqID := uuid.New().String()[:8]\n\n\t\t// Create chat with history\n\t\terr := chatStore.CreateChat(&storetypes.Chat{\n\t\t\tChatID: chatID,\n\t\t\tAssistantID: ast.ID,\n\t\t\tStatus: \"active\",\n\t\t\tShare: \"private\",\n\t\t\tCreatedAt: time.Now(),\n\t\t\tUpdatedAt: time.Now(),\n\t\t})\n\t\trequire.NoError(t, err)\n\t\tdefer func() {\n\t\t\tchatStore.DeleteMessages(chatID, nil)\n\t\t\tchatStore.DeleteChat(chatID)\n\t\t}()\n\n\t\t// Add history message\n\t\terr = chatStore.SaveMessages(chatID, []*storetypes.Message{\n\t\t\t{\n\t\t\t\tMessageID: fmt.Sprintf(\"skip_hist_1_%s\", reqID),\n\t\t\t\tChatID: chatID,\n\t\t\t\tRequestID: fmt.Sprintf(\"req_skip_%s\", reqID),\n\t\t\t\tRole: \"user\",\n\t\t\t\tType: \"user_input\",\n\t\t\t\tProps: map[string]interface{}{\"content\": \"Should be skipped\"},\n\t\t\t\tSequence: 1,\n\t\t\t\tAssistantID: ast.ID,\n\t\t\t\tCreatedAt: time.Now(),\n\t\t\t},\n\t\t})\n\t\trequire.NoError(t, err)\n\n\t\tinput := []agentcontext.Message{\n\t\t\t{Role: agentcontext.RoleUser, Content: \"Only this should appear\"},\n\t\t}\n\n\t\t// Use Skip.History option\n\t\topts := &agentcontext.Options{\n\t\t\tSkip: &agentcontext.Skip{\n\t\t\t\tHistory: true,\n\t\t\t},\n\t\t}\n\n\t\tresult, err := ast.WithHistory(ctx, input, nil, opts)\n\t\trequire.NoError(t, err)\n\t\trequire.NotNil(t, result)\n\n\t\t// Both should be same as input (history skipped)\n\t\tassert.Equal(t, input, result.InputMessages)\n\t\tassert.Equal(t, input, result.FullMessages)\n\t\tassert.Len(t, result.FullMessages, 1)\n\n\t\tt.Log(\"✓ History skipped when Skip.History=true\")\n\t})\n\n\tt.Run(\"OverlapDetection\", func(t *testing.T) {\n\t\tchatID := fmt.Sprintf(\"test_history_overlap_%s\", uuid.New().String()[:8])\n\t\tctx := newHistoryTestContext(chatID)\n\t\treqID := uuid.New().String()[:8]\n\n\t\t// Create chat\n\t\terr := chatStore.CreateChat(&storetypes.Chat{\n\t\t\tChatID: chatID,\n\t\t\tAssistantID: ast.ID,\n\t\t\tStatus: \"active\",\n\t\t\tShare: \"private\",\n\t\t\tCreatedAt: time.Now(),\n\t\t\tUpdatedAt: time.Now(),\n\t\t})\n\t\trequire.NoError(t, err)\n\t\tdefer func() {\n\t\t\tchatStore.DeleteMessages(chatID, nil)\n\t\t\tchatStore.DeleteChat(chatID)\n\t\t}()\n\n\t\t// Add history messages\n\t\terr = chatStore.SaveMessages(chatID, []*storetypes.Message{\n\t\t\t{\n\t\t\t\tMessageID: fmt.Sprintf(\"overlap_1_%s\", reqID),\n\t\t\t\tChatID: chatID,\n\t\t\t\tRequestID: fmt.Sprintf(\"req_overlap_%s\", reqID),\n\t\t\t\tRole: \"user\",\n\t\t\t\tType: \"user_input\",\n\t\t\t\tProps: map[string]interface{}{\"content\": \"Message one\"},\n\t\t\t\tSequence: 1,\n\t\t\t\tAssistantID: ast.ID,\n\t\t\t\tCreatedAt: time.Now().Add(-3 * time.Minute),\n\t\t\t},\n\t\t\t{\n\t\t\t\tMessageID: fmt.Sprintf(\"overlap_2_%s\", reqID),\n\t\t\t\tChatID: chatID,\n\t\t\t\tRequestID: fmt.Sprintf(\"req_overlap_%s\", reqID),\n\t\t\t\tRole: \"assistant\",\n\t\t\t\tType: \"text\",\n\t\t\t\tProps: map[string]interface{}{\"text\": \"Response one\"},\n\t\t\t\tSequence: 2,\n\t\t\t\tAssistantID: ast.ID,\n\t\t\t\tCreatedAt: time.Now().Add(-2 * time.Minute),\n\t\t\t},\n\t\t\t{\n\t\t\t\tMessageID: fmt.Sprintf(\"overlap_3_%s\", reqID),\n\t\t\t\tChatID: chatID,\n\t\t\t\tRequestID: fmt.Sprintf(\"req_overlap_2_%s\", reqID),\n\t\t\t\tRole: \"user\",\n\t\t\t\tType: \"user_input\",\n\t\t\t\tProps: map[string]interface{}{\"content\": \"Message two\"},\n\t\t\t\tSequence: 3,\n\t\t\t\tAssistantID: ast.ID,\n\t\t\t\tCreatedAt: time.Now().Add(-1 * time.Minute),\n\t\t\t},\n\t\t})\n\t\trequire.NoError(t, err)\n\n\t\t// Input that overlaps with history (includes last messages)\n\t\t// Some clients send full history + new message\n\t\tinput := []agentcontext.Message{\n\t\t\t{Role: agentcontext.RoleAssistant, Content: \"Response one\"}, // Overlap\n\t\t\t{Role: agentcontext.RoleUser, Content: \"Message two\"}, // Overlap\n\t\t\t{Role: agentcontext.RoleUser, Content: \"New message\"}, // New\n\t\t}\n\n\t\tresult, err := ast.WithHistory(ctx, input, nil)\n\t\trequire.NoError(t, err)\n\t\trequire.NotNil(t, result)\n\n\t\t// InputMessages should have overlap removed\n\t\tassert.Len(t, result.InputMessages, 1, \"Should remove 2 overlapping messages\")\n\t\tassert.Equal(t, \"New message\", result.InputMessages[0].Content)\n\n\t\t// FullMessages should be history + clean input\n\t\tassert.Len(t, result.FullMessages, 4) // 3 history + 1 new\n\n\t\tt.Log(\"✓ Overlap detected and removed from input\")\n\t})\n\n\tt.Run(\"EmptyChatID\", func(t *testing.T) {\n\t\tctx := newHistoryTestContext(\"\")\n\n\t\tinput := []agentcontext.Message{\n\t\t\t{Role: agentcontext.RoleUser, Content: \"No chat ID\"},\n\t\t}\n\n\t\tresult, err := ast.WithHistory(ctx, input, nil)\n\t\trequire.NoError(t, err)\n\t\trequire.NotNil(t, result)\n\n\t\t// With empty chat ID, should return input as is\n\t\tassert.Equal(t, input, result.InputMessages)\n\t\tassert.Equal(t, input, result.FullMessages)\n\n\t\tt.Log(\"✓ Empty chat ID handled gracefully\")\n\t})\n\n\tt.Run(\"MultipleUserMessages\", func(t *testing.T) {\n\t\tchatID := fmt.Sprintf(\"test_history_multi_%s\", uuid.New().String()[:8])\n\t\tctx := newHistoryTestContext(chatID)\n\t\treqID := uuid.New().String()[:8]\n\n\t\t// Create chat with history\n\t\terr := chatStore.CreateChat(&storetypes.Chat{\n\t\t\tChatID: chatID,\n\t\t\tAssistantID: ast.ID,\n\t\t\tStatus: \"active\",\n\t\t\tShare: \"private\",\n\t\t\tCreatedAt: time.Now(),\n\t\t\tUpdatedAt: time.Now(),\n\t\t})\n\t\trequire.NoError(t, err)\n\t\tdefer func() {\n\t\t\tchatStore.DeleteMessages(chatID, nil)\n\t\t\tchatStore.DeleteChat(chatID)\n\t\t}()\n\n\t\t// Add history\n\t\terr = chatStore.SaveMessages(chatID, []*storetypes.Message{\n\t\t\t{\n\t\t\t\tMessageID: fmt.Sprintf(\"multi_1_%s\", reqID),\n\t\t\t\tChatID: chatID,\n\t\t\t\tRequestID: fmt.Sprintf(\"req_multi_%s\", reqID),\n\t\t\t\tRole: \"user\",\n\t\t\t\tType: \"user_input\",\n\t\t\t\tProps: map[string]interface{}{\"content\": \"First\"},\n\t\t\t\tSequence: 1,\n\t\t\t\tAssistantID: ast.ID,\n\t\t\t\tCreatedAt: time.Now().Add(-1 * time.Minute),\n\t\t\t},\n\t\t})\n\t\trequire.NoError(t, err)\n\n\t\t// Multiple input messages\n\t\tinput := []agentcontext.Message{\n\t\t\t{Role: agentcontext.RoleUser, Content: \"Second\"},\n\t\t\t{Role: agentcontext.RoleUser, Content: \"Third\"},\n\t\t}\n\n\t\tresult, err := ast.WithHistory(ctx, input, nil)\n\t\trequire.NoError(t, err)\n\t\trequire.NotNil(t, result)\n\n\t\tassert.Len(t, result.InputMessages, 2)\n\t\tassert.Len(t, result.FullMessages, 3) // 1 history + 2 new\n\n\t\tt.Log(\"✓ Multiple input messages handled correctly\")\n\t})\n}\n\n// =============================================================================\n// History Load Tests\n// =============================================================================\n\nfunc TestHistoryLoading(t *testing.T) {\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\tast, err := assistant.Get(\"tests.history\")\n\trequire.NoError(t, err)\n\n\tchatStore := assistant.GetChatStore()\n\tif chatStore == nil {\n\t\tt.Skip(\"Chat store not configured\")\n\t}\n\n\tt.Run(\"FilterNonConversationTypes\", func(t *testing.T) {\n\t\tchatID := fmt.Sprintf(\"test_filter_%s\", uuid.New().String()[:8])\n\t\tctx := newHistoryTestContext(chatID)\n\t\treqID := uuid.New().String()[:8]\n\n\t\t// Create chat\n\t\terr := chatStore.CreateChat(&storetypes.Chat{\n\t\t\tChatID: chatID,\n\t\t\tAssistantID: ast.ID,\n\t\t\tStatus: \"active\",\n\t\t\tShare: \"private\",\n\t\t\tCreatedAt: time.Now(),\n\t\t\tUpdatedAt: time.Now(),\n\t\t})\n\t\trequire.NoError(t, err)\n\t\tdefer func() {\n\t\t\tchatStore.DeleteMessages(chatID, nil)\n\t\t\tchatStore.DeleteChat(chatID)\n\t\t}()\n\n\t\t// Add various message types (only user/assistant roles allowed by DB constraint)\n\t\t// loadHistory filters by role (user/assistant only) and converts based on type:\n\t\t// - loading/event: skipped (no semantic value)\n\t\t// - tool_call/action: converted to historical summary text\n\t\t// - text/user_input/error: kept as-is\n\t\terr = chatStore.SaveMessages(chatID, []*storetypes.Message{\n\t\t\t{\n\t\t\t\tMessageID: fmt.Sprintf(\"filter_1_%s\", reqID),\n\t\t\t\tChatID: chatID,\n\t\t\t\tRequestID: fmt.Sprintf(\"req_filter_%s\", reqID),\n\t\t\t\tRole: \"user\",\n\t\t\t\tType: \"user_input\",\n\t\t\t\tProps: map[string]interface{}{\"content\": \"User message\"},\n\t\t\t\tSequence: 1,\n\t\t\t\tAssistantID: ast.ID,\n\t\t\t\tCreatedAt: time.Now().Add(-3 * time.Minute),\n\t\t\t},\n\t\t\t{\n\t\t\t\tMessageID: fmt.Sprintf(\"filter_2_%s\", reqID),\n\t\t\t\tChatID: chatID,\n\t\t\t\tRequestID: fmt.Sprintf(\"req_filter_%s\", reqID),\n\t\t\t\tRole: \"assistant\",\n\t\t\t\tType: \"loading\",\n\t\t\t\tProps: map[string]interface{}{\"text\": \"Loading...\"},\n\t\t\t\tSequence: 2,\n\t\t\t\tAssistantID: ast.ID,\n\t\t\t\tCreatedAt: time.Now().Add(-2 * time.Minute),\n\t\t\t},\n\t\t\t{\n\t\t\t\tMessageID: fmt.Sprintf(\"filter_3_%s\", reqID),\n\t\t\t\tChatID: chatID,\n\t\t\t\tRequestID: fmt.Sprintf(\"req_filter_%s\", reqID),\n\t\t\t\tRole: \"assistant\",\n\t\t\t\tType: \"text\",\n\t\t\t\tProps: map[string]interface{}{\"text\": \"Assistant response\"},\n\t\t\t\tSequence: 3,\n\t\t\t\tAssistantID: ast.ID,\n\t\t\t\tCreatedAt: time.Now().Add(-1 * time.Minute),\n\t\t\t},\n\t\t})\n\t\trequire.NoError(t, err)\n\n\t\tinput := []agentcontext.Message{\n\t\t\t{Role: agentcontext.RoleUser, Content: \"New input\"},\n\t\t}\n\n\t\tresult, err := ast.WithHistory(ctx, input, nil)\n\t\trequire.NoError(t, err)\n\t\trequire.NotNil(t, result)\n\n\t\t// loading type is skipped (no semantic value)\n\t\t// History: user_input + text = 2 messages; plus 1 new input = 3 total\n\t\tassert.Len(t, result.FullMessages, 3)\n\n\t\t// Verify only user and assistant roles\n\t\tfor _, msg := range result.FullMessages {\n\t\t\tassert.True(t, msg.Role == agentcontext.RoleUser || msg.Role == agentcontext.RoleAssistant,\n\t\t\t\t\"Expected user or assistant role, got: %s\", msg.Role)\n\t\t}\n\n\t\tt.Log(\"✓ Loading type filtered, user/assistant roles kept\")\n\t})\n\n\tt.Run(\"ToolCallConvertedToSummary\", func(t *testing.T) {\n\t\tchatID := fmt.Sprintf(\"test_toolcall_%s\", uuid.New().String()[:8])\n\t\tctx := newHistoryTestContext(chatID)\n\t\treqID := uuid.New().String()[:8]\n\n\t\terr := chatStore.CreateChat(&storetypes.Chat{\n\t\t\tChatID: chatID,\n\t\t\tAssistantID: ast.ID,\n\t\t\tStatus: \"active\",\n\t\t\tShare: \"private\",\n\t\t\tCreatedAt: time.Now(),\n\t\t\tUpdatedAt: time.Now(),\n\t\t})\n\t\trequire.NoError(t, err)\n\t\tdefer func() {\n\t\t\tchatStore.DeleteMessages(chatID, nil)\n\t\t\tchatStore.DeleteChat(chatID)\n\t\t}()\n\n\t\t// Add tool_call messages in both formats\n\t\terr = chatStore.SaveMessages(chatID, []*storetypes.Message{\n\t\t\t{\n\t\t\t\tMessageID: fmt.Sprintf(\"tc_1_%s\", reqID),\n\t\t\t\tChatID: chatID,\n\t\t\t\tRequestID: fmt.Sprintf(\"req_tc_%s\", reqID),\n\t\t\t\tRole: \"user\",\n\t\t\t\tType: \"user_input\",\n\t\t\t\tProps: map[string]interface{}{\"content\": \"echo 3 ping 4\"},\n\t\t\t\tSequence: 1,\n\t\t\t\tAssistantID: ast.ID,\n\t\t\t\tCreatedAt: time.Now().Add(-3 * time.Minute),\n\t\t\t},\n\t\t\t// Raw stream chunk format (actual DB format)\n\t\t\t{\n\t\t\t\tMessageID: fmt.Sprintf(\"tc_2_%s\", reqID),\n\t\t\t\tChatID: chatID,\n\t\t\t\tRequestID: fmt.Sprintf(\"req_tc_%s\", reqID),\n\t\t\t\tRole: \"assistant\",\n\t\t\t\tType: \"tool_call\",\n\t\t\t\tProps: map[string]interface{}{\"content\": `[{\"index\":0,\"id\":\"call_abc\",\"type\":\"function\",\"function\":{\"name\":\"echo__ping\"}}][{\"index\":0,\"function\":{\"arguments\":\"{\\\"count\\\":3}\"}}]`},\n\t\t\t\tSequence: 2,\n\t\t\t\tAssistantID: ast.ID,\n\t\t\t\tCreatedAt: time.Now().Add(-2 * time.Minute),\n\t\t\t},\n\t\t\t// Standard ToolCallProps format\n\t\t\t{\n\t\t\t\tMessageID: fmt.Sprintf(\"tc_3_%s\", reqID),\n\t\t\t\tChatID: chatID,\n\t\t\t\tRequestID: fmt.Sprintf(\"req_tc_%s\", reqID),\n\t\t\t\tRole: \"assistant\",\n\t\t\t\tType: \"tool_call\",\n\t\t\t\tProps: map[string]interface{}{\"name\": \"echo__echo\", \"arguments\": `{\"message\":\"hello\"}`},\n\t\t\t\tSequence: 3,\n\t\t\t\tAssistantID: ast.ID,\n\t\t\t\tCreatedAt: time.Now().Add(-1 * time.Minute),\n\t\t\t},\n\t\t})\n\t\trequire.NoError(t, err)\n\n\t\tinput := []agentcontext.Message{\n\t\t\t{Role: agentcontext.RoleUser, Content: \"echo 5 ping 6\"},\n\t\t}\n\n\t\tresult, err := ast.WithHistory(ctx, input, nil)\n\t\trequire.NoError(t, err)\n\t\trequire.NotNil(t, result)\n\n\t\t// 1 user_input + 2 tool_call summaries + 1 new input = 4\n\t\tassert.Len(t, result.FullMessages, 4)\n\n\t\t// Verify tool_call messages are converted to summary text\n\t\ttcMsg1 := result.FullMessages[1]\n\t\tassert.Equal(t, agentcontext.RoleAssistant, tcMsg1.Role)\n\t\tassert.Contains(t, tcMsg1.Content, \"[Historical Tool Call Summary]\")\n\t\tassert.Contains(t, tcMsg1.Content, \"echo__ping\")\n\t\tassert.Contains(t, tcMsg1.Content, `{\"count\":3}`)\n\n\t\ttcMsg2 := result.FullMessages[2]\n\t\tassert.Equal(t, agentcontext.RoleAssistant, tcMsg2.Role)\n\t\tassert.Contains(t, tcMsg2.Content, \"[Historical Tool Call Summary]\")\n\t\tassert.Contains(t, tcMsg2.Content, \"echo__echo\")\n\t\tassert.Contains(t, tcMsg2.Content, `{\"message\":\"hello\"}`)\n\n\t\tt.Log(\"✓ Tool call messages converted to historical summaries\")\n\t})\n\n\tt.Run(\"ActionConvertedToSummary\", func(t *testing.T) {\n\t\tchatID := fmt.Sprintf(\"test_action_%s\", uuid.New().String()[:8])\n\t\tctx := newHistoryTestContext(chatID)\n\t\treqID := uuid.New().String()[:8]\n\n\t\terr := chatStore.CreateChat(&storetypes.Chat{\n\t\t\tChatID: chatID,\n\t\t\tAssistantID: ast.ID,\n\t\t\tStatus: \"active\",\n\t\t\tShare: \"private\",\n\t\t\tCreatedAt: time.Now(),\n\t\t\tUpdatedAt: time.Now(),\n\t\t})\n\t\trequire.NoError(t, err)\n\t\tdefer func() {\n\t\t\tchatStore.DeleteMessages(chatID, nil)\n\t\t\tchatStore.DeleteChat(chatID)\n\t\t}()\n\n\t\terr = chatStore.SaveMessages(chatID, []*storetypes.Message{\n\t\t\t{\n\t\t\t\tMessageID: fmt.Sprintf(\"act_1_%s\", reqID),\n\t\t\t\tChatID: chatID,\n\t\t\t\tRequestID: fmt.Sprintf(\"req_act_%s\", reqID),\n\t\t\t\tRole: \"user\",\n\t\t\t\tType: \"user_input\",\n\t\t\t\tProps: map[string]interface{}{\"content\": \"Do something\"},\n\t\t\t\tSequence: 1,\n\t\t\t\tAssistantID: ast.ID,\n\t\t\t\tCreatedAt: time.Now().Add(-2 * time.Minute),\n\t\t\t},\n\t\t\t// Action with payload\n\t\t\t{\n\t\t\t\tMessageID: fmt.Sprintf(\"act_2_%s\", reqID),\n\t\t\t\tChatID: chatID,\n\t\t\t\tRequestID: fmt.Sprintf(\"req_act_%s\", reqID),\n\t\t\t\tRole: \"assistant\",\n\t\t\t\tType: \"action\",\n\t\t\t\tProps: map[string]interface{}{\n\t\t\t\t\t\"name\": \"robot.execute\",\n\t\t\t\t\t\"payload\": map[string]interface{}{\"goals\": \"test goal\", \"robot_id\": \"12345\"},\n\t\t\t\t},\n\t\t\t\tSequence: 2,\n\t\t\t\tAssistantID: ast.ID,\n\t\t\t\tCreatedAt: time.Now().Add(-1 * time.Minute),\n\t\t\t},\n\t\t})\n\t\trequire.NoError(t, err)\n\n\t\tinput := []agentcontext.Message{\n\t\t\t{Role: agentcontext.RoleUser, Content: \"What happened?\"},\n\t\t}\n\n\t\tresult, err := ast.WithHistory(ctx, input, nil)\n\t\trequire.NoError(t, err)\n\t\trequire.NotNil(t, result)\n\n\t\t// 1 user_input + 1 action summary + 1 new input = 3\n\t\tassert.Len(t, result.FullMessages, 3)\n\n\t\tactMsg := result.FullMessages[1]\n\t\tassert.Equal(t, agentcontext.RoleAssistant, actMsg.Role)\n\t\tassert.Contains(t, actMsg.Content, \"[Historical Action Summary]\")\n\t\tassert.Contains(t, actMsg.Content, \"robot.execute\")\n\t\tassert.Contains(t, actMsg.Content, \"test goal\")\n\t\tassert.Contains(t, actMsg.Content, \"12345\")\n\n\t\tt.Log(\"✓ Action messages converted to historical summaries with payload\")\n\t})\n\n\tt.Run(\"ActionWithoutPayload\", func(t *testing.T) {\n\t\tchatID := fmt.Sprintf(\"test_action_nopay_%s\", uuid.New().String()[:8])\n\t\tctx := newHistoryTestContext(chatID)\n\t\treqID := uuid.New().String()[:8]\n\n\t\terr := chatStore.CreateChat(&storetypes.Chat{\n\t\t\tChatID: chatID,\n\t\t\tAssistantID: ast.ID,\n\t\t\tStatus: \"active\",\n\t\t\tShare: \"private\",\n\t\t\tCreatedAt: time.Now(),\n\t\t\tUpdatedAt: time.Now(),\n\t\t})\n\t\trequire.NoError(t, err)\n\t\tdefer func() {\n\t\t\tchatStore.DeleteMessages(chatID, nil)\n\t\t\tchatStore.DeleteChat(chatID)\n\t\t}()\n\n\t\terr = chatStore.SaveMessages(chatID, []*storetypes.Message{\n\t\t\t{\n\t\t\t\tMessageID: fmt.Sprintf(\"actnp_1_%s\", reqID),\n\t\t\t\tChatID: chatID,\n\t\t\t\tRequestID: fmt.Sprintf(\"req_actnp_%s\", reqID),\n\t\t\t\tRole: \"assistant\",\n\t\t\t\tType: \"action\",\n\t\t\t\tProps: map[string]interface{}{\"name\": \"navigate\"},\n\t\t\t\tSequence: 1,\n\t\t\t\tAssistantID: ast.ID,\n\t\t\t\tCreatedAt: time.Now(),\n\t\t\t},\n\t\t})\n\t\trequire.NoError(t, err)\n\n\t\tinput := []agentcontext.Message{\n\t\t\t{Role: agentcontext.RoleUser, Content: \"What happened?\"},\n\t\t}\n\n\t\tresult, err := ast.WithHistory(ctx, input, nil)\n\t\trequire.NoError(t, err)\n\t\trequire.NotNil(t, result)\n\n\t\t// 1 action summary + 1 new input = 2\n\t\tassert.Len(t, result.FullMessages, 2)\n\n\t\tactMsg := result.FullMessages[0]\n\t\tassert.Equal(t, agentcontext.RoleAssistant, actMsg.Role)\n\t\tassert.Contains(t, actMsg.Content, \"[Historical Action Summary]\")\n\t\tassert.Contains(t, actMsg.Content, \"navigate\")\n\t\tassert.NotContains(t, actMsg.Content, \"payload\")\n\n\t\tt.Log(\"✓ Action without payload handled correctly\")\n\t})\n\n\tt.Run(\"ContentExtraction\", func(t *testing.T) {\n\t\tchatID := fmt.Sprintf(\"test_extract_%s\", uuid.New().String()[:8])\n\t\tctx := newHistoryTestContext(chatID)\n\t\treqID := uuid.New().String()[:8]\n\n\t\t// Create chat\n\t\terr := chatStore.CreateChat(&storetypes.Chat{\n\t\t\tChatID: chatID,\n\t\t\tAssistantID: ast.ID,\n\t\t\tStatus: \"active\",\n\t\t\tShare: \"private\",\n\t\t\tCreatedAt: time.Now(),\n\t\t\tUpdatedAt: time.Now(),\n\t\t})\n\t\trequire.NoError(t, err)\n\t\tdefer func() {\n\t\t\tchatStore.DeleteMessages(chatID, nil)\n\t\t\tchatStore.DeleteChat(chatID)\n\t\t}()\n\n\t\t// Add messages with different content formats\n\t\terr = chatStore.SaveMessages(chatID, []*storetypes.Message{\n\t\t\t{\n\t\t\t\tMessageID: fmt.Sprintf(\"extract_1_%s\", reqID),\n\t\t\t\tChatID: chatID,\n\t\t\t\tRequestID: fmt.Sprintf(\"req_extract_%s\", reqID),\n\t\t\t\tRole: \"user\",\n\t\t\t\tType: \"user_input\",\n\t\t\t\tProps: map[string]interface{}{\"content\": \"User content from props.content\"},\n\t\t\t\tSequence: 1,\n\t\t\t\tAssistantID: ast.ID,\n\t\t\t\tCreatedAt: time.Now().Add(-2 * time.Minute),\n\t\t\t},\n\t\t\t{\n\t\t\t\tMessageID: fmt.Sprintf(\"extract_2_%s\", reqID),\n\t\t\t\tChatID: chatID,\n\t\t\t\tRequestID: fmt.Sprintf(\"req_extract_%s\", reqID),\n\t\t\t\tRole: \"assistant\",\n\t\t\t\tType: \"text\",\n\t\t\t\tProps: map[string]interface{}{\"text\": \"Assistant content from props.text\"},\n\t\t\t\tSequence: 2,\n\t\t\t\tAssistantID: ast.ID,\n\t\t\t\tCreatedAt: time.Now().Add(-1 * time.Minute),\n\t\t\t},\n\t\t})\n\t\trequire.NoError(t, err)\n\n\t\tinput := []agentcontext.Message{\n\t\t\t{Role: agentcontext.RoleUser, Content: \"New message\"},\n\t\t}\n\n\t\tresult, err := ast.WithHistory(ctx, input, nil)\n\t\trequire.NoError(t, err)\n\t\trequire.NotNil(t, result)\n\n\t\t// Verify content was extracted correctly\n\t\tassert.Len(t, result.FullMessages, 3)\n\t\tassert.Equal(t, \"User content from props.content\", result.FullMessages[0].Content)\n\t\tassert.Equal(t, \"Assistant content from props.text\", result.FullMessages[1].Content)\n\n\t\tt.Log(\"✓ Content extracted correctly from different formats\")\n\t})\n}\n\n// =============================================================================\n// Edge Cases Tests\n// =============================================================================\n\nfunc TestHistoryEdgeCases(t *testing.T) {\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\tast, err := assistant.Get(\"tests.history\")\n\trequire.NoError(t, err)\n\n\tchatStore := assistant.GetChatStore()\n\tif chatStore == nil {\n\t\tt.Skip(\"Chat store not configured\")\n\t}\n\n\tt.Run(\"EmptyInput\", func(t *testing.T) {\n\t\tchatID := fmt.Sprintf(\"test_empty_input_%s\", uuid.New().String()[:8])\n\t\tctx := newHistoryTestContext(chatID)\n\t\treqID := uuid.New().String()[:8]\n\n\t\t// Create chat with history\n\t\terr := chatStore.CreateChat(&storetypes.Chat{\n\t\t\tChatID: chatID,\n\t\t\tAssistantID: ast.ID,\n\t\t\tStatus: \"active\",\n\t\t\tShare: \"private\",\n\t\t\tCreatedAt: time.Now(),\n\t\t\tUpdatedAt: time.Now(),\n\t\t})\n\t\trequire.NoError(t, err)\n\t\tdefer func() {\n\t\t\tchatStore.DeleteMessages(chatID, nil)\n\t\t\tchatStore.DeleteChat(chatID)\n\t\t}()\n\n\t\terr = chatStore.SaveMessages(chatID, []*storetypes.Message{\n\t\t\t{\n\t\t\t\tMessageID: fmt.Sprintf(\"empty_input_1_%s\", reqID),\n\t\t\t\tChatID: chatID,\n\t\t\t\tRequestID: fmt.Sprintf(\"req_empty_%s\", reqID),\n\t\t\t\tRole: \"user\",\n\t\t\t\tType: \"user_input\",\n\t\t\t\tProps: map[string]interface{}{\"content\": \"Previous\"},\n\t\t\t\tSequence: 1,\n\t\t\t\tAssistantID: ast.ID,\n\t\t\t\tCreatedAt: time.Now(),\n\t\t\t},\n\t\t})\n\t\trequire.NoError(t, err)\n\n\t\t// Empty input\n\t\tinput := []agentcontext.Message{}\n\n\t\tresult, err := ast.WithHistory(ctx, input, nil)\n\t\trequire.NoError(t, err)\n\t\trequire.NotNil(t, result)\n\n\t\t// Should return history only\n\t\tassert.Empty(t, result.InputMessages)\n\t\tassert.Len(t, result.FullMessages, 1)\n\n\t\tt.Log(\"✓ Empty input handled correctly\")\n\t})\n\n\tt.Run(\"FullOverlap\", func(t *testing.T) {\n\t\tchatID := fmt.Sprintf(\"test_full_overlap_%s\", uuid.New().String()[:8])\n\t\tctx := newHistoryTestContext(chatID)\n\t\treqID := uuid.New().String()[:8]\n\n\t\t// Create chat\n\t\terr := chatStore.CreateChat(&storetypes.Chat{\n\t\t\tChatID: chatID,\n\t\t\tAssistantID: ast.ID,\n\t\t\tStatus: \"active\",\n\t\t\tShare: \"private\",\n\t\t\tCreatedAt: time.Now(),\n\t\t\tUpdatedAt: time.Now(),\n\t\t})\n\t\trequire.NoError(t, err)\n\t\tdefer func() {\n\t\t\tchatStore.DeleteMessages(chatID, nil)\n\t\t\tchatStore.DeleteChat(chatID)\n\t\t}()\n\n\t\t// Add history\n\t\terr = chatStore.SaveMessages(chatID, []*storetypes.Message{\n\t\t\t{\n\t\t\t\tMessageID: fmt.Sprintf(\"full_overlap_1_%s\", reqID),\n\t\t\t\tChatID: chatID,\n\t\t\t\tRequestID: fmt.Sprintf(\"req_full_%s\", reqID),\n\t\t\t\tRole: \"user\",\n\t\t\t\tType: \"user_input\",\n\t\t\t\tProps: map[string]interface{}{\"content\": \"Exact same message\"},\n\t\t\t\tSequence: 1,\n\t\t\t\tAssistantID: ast.ID,\n\t\t\t\tCreatedAt: time.Now(),\n\t\t\t},\n\t\t})\n\t\trequire.NoError(t, err)\n\n\t\t// Input is exactly the same as history\n\t\tinput := []agentcontext.Message{\n\t\t\t{Role: agentcontext.RoleUser, Content: \"Exact same message\"},\n\t\t}\n\n\t\tresult, err := ast.WithHistory(ctx, input, nil)\n\t\trequire.NoError(t, err)\n\t\trequire.NotNil(t, result)\n\n\t\t// Full overlap: clean input should be empty\n\t\tassert.Empty(t, result.InputMessages)\n\t\t// FullMessages should be just history (no duplicates)\n\t\tassert.Len(t, result.FullMessages, 1)\n\n\t\tt.Log(\"✓ Full overlap handled correctly\")\n\t})\n\n\tt.Run(\"NonExistentChat\", func(t *testing.T) {\n\t\tchatID := \"non_existent_chat_12345\"\n\t\tctx := newHistoryTestContext(chatID)\n\n\t\tinput := []agentcontext.Message{\n\t\t\t{Role: agentcontext.RoleUser, Content: \"Message to non-existent chat\"},\n\t\t}\n\n\t\tresult, err := ast.WithHistory(ctx, input, nil)\n\t\trequire.NoError(t, err)\n\t\trequire.NotNil(t, result)\n\n\t\t// Should return input as is (no history found)\n\t\tassert.Equal(t, input, result.InputMessages)\n\t\tassert.Equal(t, input, result.FullMessages)\n\n\t\tt.Log(\"✓ Non-existent chat handled gracefully\")\n\t})\n\n\tt.Run(\"MessageWithName\", func(t *testing.T) {\n\t\tchatID := fmt.Sprintf(\"test_name_%s\", uuid.New().String()[:8])\n\t\tctx := newHistoryTestContext(chatID)\n\t\treqID := uuid.New().String()[:8]\n\n\t\t// Create chat\n\t\terr := chatStore.CreateChat(&storetypes.Chat{\n\t\t\tChatID: chatID,\n\t\t\tAssistantID: ast.ID,\n\t\t\tStatus: \"active\",\n\t\t\tShare: \"private\",\n\t\t\tCreatedAt: time.Now(),\n\t\t\tUpdatedAt: time.Now(),\n\t\t})\n\t\trequire.NoError(t, err)\n\t\tdefer func() {\n\t\t\tchatStore.DeleteMessages(chatID, nil)\n\t\t\tchatStore.DeleteChat(chatID)\n\t\t}()\n\n\t\t// Add message with name\n\t\terr = chatStore.SaveMessages(chatID, []*storetypes.Message{\n\t\t\t{\n\t\t\t\tMessageID: fmt.Sprintf(\"name_msg_1_%s\", reqID),\n\t\t\t\tChatID: chatID,\n\t\t\t\tRequestID: fmt.Sprintf(\"req_name_%s\", reqID),\n\t\t\t\tRole: \"user\",\n\t\t\t\tType: \"user_input\",\n\t\t\t\tProps: map[string]interface{}{\"content\": \"Message with name\", \"name\": \"John\"},\n\t\t\t\tSequence: 1,\n\t\t\t\tAssistantID: ast.ID,\n\t\t\t\tCreatedAt: time.Now(),\n\t\t\t},\n\t\t})\n\t\trequire.NoError(t, err)\n\n\t\tinput := []agentcontext.Message{\n\t\t\t{Role: agentcontext.RoleUser, Content: \"New message\"},\n\t\t}\n\n\t\tresult, err := ast.WithHistory(ctx, input, nil)\n\t\trequire.NoError(t, err)\n\t\trequire.NotNil(t, result)\n\n\t\t// First message should have name\n\t\tassert.Len(t, result.FullMessages, 2)\n\t\tassert.NotNil(t, result.FullMessages[0].Name)\n\t\tassert.Equal(t, \"John\", *result.FullMessages[0].Name)\n\n\t\tt.Log(\"✓ Message name field preserved\")\n\t})\n\n\tt.Run(\"EmptyContent\", func(t *testing.T) {\n\t\tchatID := fmt.Sprintf(\"test_empty_content_%s\", uuid.New().String()[:8])\n\t\tctx := newHistoryTestContext(chatID)\n\t\treqID := uuid.New().String()[:8]\n\n\t\t// Create chat\n\t\terr := chatStore.CreateChat(&storetypes.Chat{\n\t\t\tChatID: chatID,\n\t\t\tAssistantID: ast.ID,\n\t\t\tStatus: \"active\",\n\t\t\tShare: \"private\",\n\t\t\tCreatedAt: time.Now(),\n\t\t\tUpdatedAt: time.Now(),\n\t\t})\n\t\trequire.NoError(t, err)\n\t\tdefer func() {\n\t\t\tchatStore.DeleteMessages(chatID, nil)\n\t\t\tchatStore.DeleteChat(chatID)\n\t\t}()\n\n\t\t// Add message with empty content in props\n\t\terr = chatStore.SaveMessages(chatID, []*storetypes.Message{\n\t\t\t{\n\t\t\t\tMessageID: fmt.Sprintf(\"empty_content_1_%s\", reqID),\n\t\t\t\tChatID: chatID,\n\t\t\t\tRequestID: fmt.Sprintf(\"req_empty_content_%s\", reqID),\n\t\t\t\tRole: \"user\",\n\t\t\t\tType: \"user_input\",\n\t\t\t\tProps: map[string]interface{}{}, // empty props (no content)\n\t\t\t\tSequence: 1,\n\t\t\t\tAssistantID: ast.ID,\n\t\t\t\tCreatedAt: time.Now(),\n\t\t\t},\n\t\t})\n\t\trequire.NoError(t, err)\n\n\t\tinput := []agentcontext.Message{\n\t\t\t{Role: agentcontext.RoleUser, Content: \"New message\"},\n\t\t}\n\n\t\tresult, err := ast.WithHistory(ctx, input, nil)\n\t\trequire.NoError(t, err)\n\t\trequire.NotNil(t, result)\n\n\t\t// Message with empty props should be skipped (no content extractable)\n\t\t// Only new input should be present\n\t\tassert.Len(t, result.FullMessages, 1)\n\t\tassert.Equal(t, \"New message\", result.FullMessages[0].Content)\n\n\t\tt.Log(\"✓ Empty content handled gracefully (message skipped)\")\n\t})\n}\n" }, { "path": "agent/assistant/hook/REALWORLD_PERFORMANCE_REPORT.md", "content": "# Performance Test Report\n\n**Test Date**: November 28, 2025 \n**System**: Yao Agent Assistant - Create Hook \n**Hardware**: Apple M2 Max, ARM64, macOS 25.1.0\n\n---\n\n## Executive Summary\n\nAll tests passed with 100% success rate. The system demonstrates production-ready performance with stable memory usage and predictable response times.\n\n**Key Metrics:**\n\n- ✅ **Concurrent Capacity**: 1,000 operations @ 100 goroutines\n- ✅ **Response Time**: 1.57ms average (hook execution only)\n- ✅ **Memory Stable**: ≤1 MB growth under load\n- ✅ **Success Rate**: 100% (1,000/1,000 validated)\n\n---\n\n## Performance Benchmarks\n\n### Single Request Performance\n\n| Scenario | Mode | Time/op | Memory/op | Allocs/op |\n| -------- | ----------- | ------- | --------- | --------- |\n| Simple | Standard | 1.44 ms | 45 KB | 827 |\n| Simple | Performance | 0.33 ms | 33 KB | 789 |\n| Business | Standard | 3.33 ms | 95 KB | 1,570 |\n| Business | Performance | 0.35 ms | 33 KB | 805 |\n\n**Note**: Standard mode creates/disposes V8 isolate per request. Performance mode reuses isolates from pool.\n\n### Concurrent Performance\n\n| Scenario | Mode | Time/op | Memory/op | Allocs/op |\n| ------------------- | ----------- | ------- | --------- | --------- |\n| Simple Concurrent | Standard | 0.42 ms | 46 KB | 829 |\n| Simple Concurrent | Performance | 0.35 ms | 33 KB | 789 |\n| Business Concurrent | Standard | 0.64 ms | 89 KB | 1,457 |\n| Business Concurrent | Performance | 0.35 ms | 33 KB | 786 |\n\n**Observation**: Concurrent execution shows better performance than sequential in standard mode due to parallel isolate creation.\n\n---\n\n## Stress Test Results\n\n### Basic Tests\n\n**Simple Scenario** (100 iterations):\n\n- Duration: 0.34s\n- Memory: 470 MB → 471 MB (0 MB growth)\n- Result: ✅ Stable\n\n**MCP Integration** (50 iterations):\n\n- Duration: 0.40s\n- Memory: 472 MB → 471 MB (0 MB growth)\n- Result: ✅ No leaks\n\n**Full Workflow** (30 iterations, MCP + DB + Trace):\n\n- Duration: 0.39s\n- Average: 12.90 ms/op\n- Memory: 472 MB → 471 MB (0 MB growth)\n- Result: ✅ All components working\n\n### Concurrent Stress Test ⭐\n\n**Configuration:**\n\n- Goroutines: 100\n- Iterations: 10 per goroutine\n- Total operations: 1,000\n- Scenarios: Mixed (simple, mcp_health, mcp_tools, full_workflow)\n\n**Results:**\n\n- Duration: 1.57 seconds\n- Average: 1.57 ms/op\n- Throughput: ~636 ops/second\n- Success: 1,000/1,000 (100%)\n- Memory: 472 MB → 473 MB (1 MB growth)\n- Validation: All responses correct\n\n**Scenario Distribution:**\n\n- simple: 250 ops (25%)\n- mcp_health: 250 ops (25%)\n- mcp_tools: 250 ops (25%)\n- full_workflow: 250 ops (25%)\n\n---\n\n## Memory Analysis\n\n### Memory Leak Tests\n\nAll memory leak tests passed with acceptable thresholds:\n\n**Standard Mode** (1,000 iterations):\n\n- Growth: 11.65 MB (12.2 KB/iteration)\n- Threshold: <15 KB/iteration\n- Status: ✅ Pass\n\n**Performance Mode** (1,000 iterations):\n\n- Growth: -0.15 MB (negative = GC working)\n- Status: ✅ Pass\n\n**Business Scenarios** (200 iterations each):\n\n- Growth: 12-15 KB/iteration\n- Status: ✅ All pass\n\n**Concurrent Load** (1,000 iterations):\n\n- Growth: 1.73 MB (1.8 KB/iteration)\n- Status: ✅ Excellent\n\n### Goroutine Behavior\n\n**Observation**: Each request creates 2 goroutines (trace pubsub + state worker) that exit asynchronously after `Release()`.\n\n**Measured Growth**: 2.0 goroutines/iteration\n\n- Initial: 106 → Final: 122 (after 10 iterations)\n- Threshold: <5 goroutines/iteration\n- Status: ✅ Expected behavior (not a leak)\n\n**Root Cause**: Asynchronous cleanup - goroutines exit when channels close, but scheduling takes time. This is normal Go concurrency behavior.\n\n---\n\n## Capacity Planning\n\n### Single Instance Capacity\n\n**Hook Execution Only** (measured):\n\n```\nResponse Time: 1.57ms\nGoroutines: 100 tested, stable\nThroughput: ~636 ops/second actual\n```\n\n**Complete Request Flow** (estimated):\n\n```\nHook Execution: 1.57ms\nLLM API Call: 500-2000ms (typical)\nNetwork + Parsing: 50-100ms\nTotal: ~1000ms per request\n```\n\n### Production Estimates\n\n**Conservative Capacity** (50% safety factor):\n\n| User Activity | Requests/Min | Concurrent Online Users |\n| ------------------- | ------------ | ----------------------- |\n| Light (3 req/min) | 3,000 total | 1,000 online |\n| Normal (6 req/min) | 3,000 total | 500 online |\n| Active (15 req/min) | 3,000 total | 200 online |\n| Heavy (30 req/min) | 3,000 total | 100 online |\n\n**Calculation Basis:**\n\n- 100 goroutines proven stable\n- ~1 request/second per goroutine\n- Base: 100 req/s = 6,000 req/min\n- With 50% safety: 3,000 req/min sustained\n\n**Recommendation**: Start with 500-1,000 concurrent online users per instance, monitor and scale horizontally as needed.\n\n**Note**: \"Concurrent online users\" means users actively using the system at the same time, not total registered users.\n\n### Horizontal Scaling\n\n```\n1 instance → 500-1,000 concurrent online users\n2 instances → 1,000-2,000 concurrent online users\n5 instances → 2,500-5,000 concurrent online users\n10 instances → 5,000-10,000 concurrent online users\n```\n\n---\n\n## Component Verification\n\n### MCP Integration ✅\n\n- ListTools: Working\n- CallTool: Working (ping, status)\n- Resource operations: Working\n- Prompt operations: Working\n- Performance: <3ms per operation\n\n### Trace Management ✅\n\n- Node creation: <1ms\n- 20+ nodes per operation: No issues\n- Memory cleanup: Effective\n- Goroutine cleanup: Asynchronous (expected)\n\n### Context Management ✅\n\n- Creation: Fast\n- Release: Working (cascading cleanup)\n- Memory: No leaks detected\n- Thread-safe: Yes\n\n### Database Integration ✅\n\n- Query execution: Working\n- Connection pooling: Efficient\n- Error handling: Robust\n\n---\n\n## Reliability Metrics\n\n**Test Coverage:**\n\n- Total tests: 21\n- Tests passed: 21 (100%)\n- Tests failed: 0\n- Flaky tests: 0\n\n**Error Rate:**\n\n- Operations: 1,200+\n- Errors: 0\n- Rate: 0.00%\n\n**Data Integrity:**\n\n- Message validation: 100%\n- Metadata validation: 100%\n- Scenario matching: 100%\n\n---\n\n## Known Behaviors\n\n### Goroutine Accumulation\n\n**Observation**: ~2 goroutines created per request that exit asynchronously.\n\n**Root Cause**:\n\n- Trace creates 2 background goroutines: `pubsub.forward()` + `stateWorker()`\n- These exit when channels close (via `Release()`)\n- Exit is asynchronous - takes 5-15ms after `Release()`\n- In rapid iterations, new goroutines start before old ones finish exiting\n\n**Impact**:\n\n- Temporary accumulation during high load\n- No unbounded growth (goroutines eventually exit)\n- Go runtime handles this efficiently\n- Not a memory leak\n\n**Status**: ✅ Expected behavior, no action needed\n\n---\n\n## Recommendations\n\n### Production Deployment\n\n**Ready to Deploy**: Yes\n\n**Suggested Configuration:**\n\n- Start with 1-2 instances\n- Target: 500-1,000 concurrent users per instance\n- V8 Mode: Standard (safer) or Performance (faster)\n- Health check: Monitor goroutine count (<10,000)\n\n### Monitoring\n\n**Key Metrics to Track:**\n\n1. Response time (alert if >100ms sustained)\n2. Goroutine count (alert if >10,000)\n3. Memory usage (alert if >1GB growth/hour)\n4. Error rate (alert if >1%)\n\n### Scaling Triggers\n\n**Scale Up When:**\n\n- Response time >50ms average (sustained 5 min)\n- Goroutine count >5,000 (approaching limits)\n- CPU >70% (need more capacity)\n\n**Scale Out When:**\n\n- Need >1,000 concurrent users\n- Multi-region deployment required\n- Geographic latency optimization needed\n\n---\n\n## Conclusions\n\n### System Status: **Production Ready** ✅\n\n**Strengths:**\n\n- Fast response times (1-3ms for hook execution)\n- Stable memory usage (no leaks detected)\n- Excellent concurrent performance (100+ goroutines stable)\n- 100% test success rate with validation\n- Clean resource management with proper cleanup\n\n**Suitable For:**\n\n- SaaS platforms (500-1,000 concurrent online users per instance)\n- Enterprise applications requiring high reliability\n- Systems with 100-1,000 concurrent online users\n- Mission-critical AI agent deployments\n\n**Performance Rating**: A (Excellent)\n\n**Capacity Rating**: Mid-stage SaaS (Series A/B ready)\n\n---\n\n## Test Execution Summary\n\n```\nPlatform: darwin/arm64\nCPU: Apple M2 Max\nGo Version: 1.25.0\nTest Duration: 19.8 seconds\n\nUnit Tests: 21 passed\nBenchmarks: 8 completed\nStress Tests: 5 passed (1,000 ops validated)\nMemory Tests: 7 passed\nGoroutine Tests: 4 passed (behavior documented)\n\nOverall: 100% PASS ✅\n```\n\n---\n\n**Report Generated**: November 28, 2025 \n**Test Framework**: Go testing + testify \n**Validation**: Complete (all responses verified) \n**Status**: PRODUCTION READY\n" }, { "path": "agent/assistant/hook/create.go", "content": "package hook\n\nimport (\n\t\"encoding/json\"\n\t\"fmt\"\n\n\t\"github.com/yaoapp/gou/runtime/v8/bridge\"\n\t\"github.com/yaoapp/yao/agent/context\"\n)\n\n// Create create a new assistant\n// opts is optional - if provided, will be adjusted based on hook response\nfunc (s *Script) Create(ctx *context.Context, messages []context.Message, opts ...*context.Options) (*context.HookCreateResponse, *context.Options, error) {\n\t// Get or create options\n\tvar options *context.Options\n\tif len(opts) > 0 && opts[0] != nil {\n\t\toptions = opts[0]\n\t} else {\n\t\toptions = &context.Options{}\n\t}\n\n\t// Execute hook with ctx, messages, and options (convert options to map for JS)\n\toptionsMap := options.ToMap()\n\tres, err := s.Execute(ctx, \"Create\", messages, optionsMap)\n\tif err != nil {\n\t\treturn nil, nil, err\n\t}\n\n\tresponse, err := s.getHookCreateResponse(res)\n\tif err != nil {\n\t\treturn nil, nil, err\n\t}\n\n\t// Apply adjustments from the response\n\tif response != nil {\n\t\ts.applyContextAdjustments(ctx, response)\n\t\ts.applyOptionsAdjustments(options, response)\n\t}\n\n\treturn response, options, nil\n}\n\n// applyContextAdjustments applies session-level field overrides from the hook response back to the context\nfunc (s *Script) applyContextAdjustments(ctx *context.Context, response *context.HookCreateResponse) {\n\t// Note: AssistantID cannot be overridden - it's set at initialization and immutable\n\n\t// Override locale if provided (session-level)\n\tif response.Locale != \"\" {\n\t\tctx.Locale = response.Locale\n\t}\n\n\t// Override theme if provided (session-level)\n\tif response.Theme != \"\" {\n\t\tctx.Theme = response.Theme\n\t}\n\n\t// Override route if provided (session-level)\n\tif response.Route != \"\" {\n\t\tctx.Route = response.Route\n\t}\n\n\t// Merge or override metadata if provided (session-level)\n\tif len(response.Metadata) > 0 {\n\t\tif ctx.Metadata == nil {\n\t\t\tctx.Metadata = make(map[string]interface{})\n\t\t}\n\t\t// Merge metadata - response metadata takes precedence\n\t\tfor key, value := range response.Metadata {\n\t\t\tctx.Metadata[key] = value\n\t\t}\n\t}\n}\n\n// applyOptionsAdjustments applies call-level field overrides from the hook response to options\nfunc (s *Script) applyOptionsAdjustments(opts *context.Options, response *context.HookCreateResponse) {\n\t// Override connector if provided (call-level parameter)\n\tif response.Connector != \"\" {\n\t\topts.Connector = response.Connector\n\t}\n}\n\n// getHookCreateResponse convert the result to a HookCreateResponse\nfunc (s *Script) getHookCreateResponse(res interface{}) (*context.HookCreateResponse, error) {\n\t// Handle nil result\n\tif res == nil {\n\t\treturn nil, nil\n\t}\n\n\t// Handle undefined result (treat as nil)\n\tif _, ok := res.(bridge.UndefinedT); ok {\n\t\treturn nil, nil\n\t}\n\n\t// Marshal to JSON and unmarshal to HookCreateResponse\n\traw, err := json.Marshal(res)\n\tif err != nil {\n\t\treturn nil, fmt.Errorf(\"failed to marshal result: %w\", err)\n\t}\n\n\tvar response context.HookCreateResponse\n\tif err := json.Unmarshal(raw, &response); err != nil {\n\t\treturn nil, fmt.Errorf(\"failed to unmarshal to HookCreateResponse: %w\", err)\n\t}\n\n\treturn &response, nil\n}\n" }, { "path": "agent/assistant/hook/create_bench_test.go", "content": "package hook_test\n\nimport (\n\tstdContext \"context\"\n\t\"testing\"\n\n\t\"github.com/yaoapp/yao/agent/assistant\"\n\t\"github.com/yaoapp/yao/agent/context\"\n\t\"github.com/yaoapp/yao/agent/testutils\"\n\t\"github.com/yaoapp/yao/openapi/oauth/types\"\n\t\"github.com/yaoapp/yao/test\"\n)\n\n// ============================================================================\n// Simple Scenario Benchmarks\n// ============================================================================\n\n// BenchmarkSimpleStandardMode benchmarks simple scenario in standard V8 mode\n// Run with: go test -bench=BenchmarkSimpleStandardMode -benchmem -benchtime=100x\nfunc BenchmarkSimpleStandardMode(b *testing.B) {\n\ttestutils.Prepare(&testing.T{})\n\tdefer testutils.Clean(&testing.T{})\n\n\tagent, err := assistant.Get(\"tests.create\")\n\tif err != nil {\n\t\tb.Fatalf(\"Failed to get assistant: %s\", err.Error())\n\t}\n\n\tif agent.HookScript == nil {\n\t\tb.Fatalf(\"Assistant has no script\")\n\t}\n\n\tb.ResetTimer()\n\tfor i := 0; i < b.N; i++ {\n\t\tctx := newBenchContext(\"bench-simple-standard\", \"tests.create\")\n\t\t_, _, err := agent.HookScript.Create(ctx, []context.Message{\n\t\t\t{Role: \"user\", Content: \"Hello\"},\n\t\t})\n\t\tif err != nil {\n\t\t\tb.Fatalf(\"Create failed: %s\", err.Error())\n\t\t}\n\t}\n}\n\n// BenchmarkSimplePerformanceMode benchmarks simple scenario in performance V8 mode\n// Run with: go test -bench=BenchmarkSimplePerformanceMode -benchmem -benchtime=100x\nfunc BenchmarkSimplePerformanceMode(b *testing.B) {\n\ttestutils.Prepare(&testing.T{}, test.PrepareOption{V8Mode: \"performance\"})\n\tdefer testutils.Clean(&testing.T{})\n\n\tagent, err := assistant.Get(\"tests.create\")\n\tif err != nil {\n\t\tb.Fatalf(\"Failed to get assistant: %s\", err.Error())\n\t}\n\n\tif agent.HookScript == nil {\n\t\tb.Fatalf(\"Assistant has no script\")\n\t}\n\n\tb.ResetTimer()\n\tfor i := 0; i < b.N; i++ {\n\t\tctx := newBenchContext(\"bench-simple-performance\", \"tests.create\")\n\t\t_, _, err := agent.HookScript.Create(ctx, []context.Message{\n\t\t\t{Role: \"user\", Content: \"Hello\"},\n\t\t})\n\t\tif err != nil {\n\t\t\tb.Fatalf(\"Create failed: %s\", err.Error())\n\t\t}\n\t}\n}\n\n// ============================================================================\n// Business Scenario Benchmarks (with Process calls, DB access, etc.)\n// ============================================================================\n\n// BenchmarkBusinessStandardMode benchmarks business scenarios in standard V8 mode\n// Run with: go test -bench=BenchmarkBusinessStandardMode -benchmem -benchtime=100x\nfunc BenchmarkBusinessStandardMode(b *testing.B) {\n\ttestutils.Prepare(&testing.T{})\n\tdefer testutils.Clean(&testing.T{})\n\n\tagent, err := assistant.Get(\"tests.create\")\n\tif err != nil {\n\t\tb.Fatalf(\"Failed to get assistant: %s\", err.Error())\n\t}\n\n\tif agent.HookScript == nil {\n\t\tb.Fatalf(\"Assistant has no script\")\n\t}\n\n\tscenarios := getBusinessScenarios()\n\n\tb.ResetTimer()\n\tfor i := 0; i < b.N; i++ {\n\t\tscenario := scenarios[i%len(scenarios)]\n\t\tctx := newBenchContext(\"bench-business-standard\", \"tests.create\")\n\t\t_, _, err := agent.HookScript.Create(ctx, []context.Message{\n\t\t\t{Role: \"user\", Content: scenario.content},\n\t\t})\n\t\tif err != nil {\n\t\t\tb.Errorf(\"%s failed: %s\", scenario.name, err.Error())\n\t\t}\n\t}\n}\n\n// BenchmarkBusinessPerformanceMode benchmarks business scenarios in performance V8 mode\n// Run with: go test -bench=BenchmarkBusinessPerformanceMode -benchmem -benchtime=100x\nfunc BenchmarkBusinessPerformanceMode(b *testing.B) {\n\ttestutils.Prepare(&testing.T{}, test.PrepareOption{V8Mode: \"performance\"})\n\tdefer testutils.Clean(&testing.T{})\n\n\tagent, err := assistant.Get(\"tests.create\")\n\tif err != nil {\n\t\tb.Fatalf(\"Failed to get assistant: %s\", err.Error())\n\t}\n\n\tif agent.HookScript == nil {\n\t\tb.Fatalf(\"Assistant has no script\")\n\t}\n\n\tscenarios := getBusinessScenarios()\n\n\tb.ResetTimer()\n\tfor i := 0; i < b.N; i++ {\n\t\tscenario := scenarios[i%len(scenarios)]\n\t\tctx := newBenchContext(\"bench-business-performance\", \"tests.create\")\n\t\t_, _, err := agent.HookScript.Create(ctx, []context.Message{\n\t\t\t{Role: \"user\", Content: scenario.content},\n\t\t})\n\t\tif err != nil {\n\t\t\tb.Errorf(\"%s failed: %s\", scenario.name, err.Error())\n\t\t}\n\t}\n}\n\n// ============================================================================\n// Concurrent Benchmarks\n// ============================================================================\n\n// BenchmarkConcurrentSimpleStandardMode benchmarks simple concurrent scenario in standard V8 mode\n// Simulates concurrent users with isolate creation/disposal per request\n// Run with: go test -bench=BenchmarkConcurrentSimpleStandardMode -benchmem -benchtime=100x\nfunc BenchmarkConcurrentSimpleStandardMode(b *testing.B) {\n\ttestutils.Prepare(&testing.T{})\n\tdefer testutils.Clean(&testing.T{})\n\n\tagent, err := assistant.Get(\"tests.create\")\n\tif err != nil {\n\t\tb.Fatalf(\"Failed to get assistant: %s\", err.Error())\n\t}\n\n\tif agent.HookScript == nil {\n\t\tb.Fatalf(\"Assistant has no script\")\n\t}\n\n\tb.ResetTimer()\n\tb.RunParallel(func(pb *testing.PB) {\n\t\ti := 0\n\t\tfor pb.Next() {\n\t\t\tctx := newBenchContext(\"bench-concurrent-simple-standard\", \"tests.create\")\n\t\t\t_, _, err := agent.HookScript.Create(ctx, []context.Message{\n\t\t\t\t{Role: \"user\", Content: \"Hello\"},\n\t\t\t})\n\t\t\tif err != nil {\n\t\t\t\tb.Errorf(\"Create failed (iteration %d): %s\", i, err.Error())\n\t\t\t}\n\t\t\ti++\n\t\t}\n\t})\n}\n\n// BenchmarkConcurrentSimplePerformanceMode benchmarks simple concurrent scenario in performance V8 mode\n// Simulates 100 users simultaneously using the system with isolate pool\n// Run with: go test -bench=BenchmarkConcurrentSimplePerformanceMode -benchmem -benchtime=100x\nfunc BenchmarkConcurrentSimplePerformanceMode(b *testing.B) {\n\ttestutils.Prepare(&testing.T{}, test.PrepareOption{V8Mode: \"performance\"})\n\tdefer testutils.Clean(&testing.T{})\n\n\tagent, err := assistant.Get(\"tests.create\")\n\tif err != nil {\n\t\tb.Fatalf(\"Failed to get assistant: %s\", err.Error())\n\t}\n\n\tif agent.HookScript == nil {\n\t\tb.Fatalf(\"Assistant has no script\")\n\t}\n\n\tb.ResetTimer()\n\tb.RunParallel(func(pb *testing.PB) {\n\t\ti := 0\n\t\tfor pb.Next() {\n\t\t\tctx := newBenchContext(\"bench-concurrent-simple\", \"tests.create\")\n\t\t\t_, _, err := agent.HookScript.Create(ctx, []context.Message{\n\t\t\t\t{Role: \"user\", Content: \"Hello\"},\n\t\t\t})\n\t\t\tif err != nil {\n\t\t\t\tb.Errorf(\"Create failed (iteration %d): %s\", i, err.Error())\n\t\t\t}\n\t\t\ti++\n\t\t}\n\t})\n}\n\n// BenchmarkConcurrentBusinessStandardMode benchmarks concurrent business scenarios in standard V8 mode\n// Tests various scenarios with concurrent users and isolate creation/disposal per request\n// Run with: go test -bench=BenchmarkConcurrentBusinessStandardMode -benchmem -benchtime=100x\nfunc BenchmarkConcurrentBusinessStandardMode(b *testing.B) {\n\ttestutils.Prepare(&testing.T{})\n\tdefer testutils.Clean(&testing.T{})\n\n\tagent, err := assistant.Get(\"tests.create\")\n\tif err != nil {\n\t\tb.Fatalf(\"Failed to get assistant: %s\", err.Error())\n\t}\n\n\tif agent.HookScript == nil {\n\t\tb.Fatalf(\"Assistant has no script\")\n\t}\n\n\tscenarios := getBusinessScenarios()\n\n\tb.ResetTimer()\n\tb.RunParallel(func(pb *testing.PB) {\n\t\ti := 0\n\t\tfor pb.Next() {\n\t\t\tscenario := scenarios[i%len(scenarios)]\n\t\t\tctx := newBenchContext(\"bench-concurrent-business-standard\", \"tests.create\")\n\t\t\t_, _, err := agent.HookScript.Create(ctx, []context.Message{\n\t\t\t\t{Role: \"user\", Content: scenario.content},\n\t\t\t})\n\t\t\tif err != nil {\n\t\t\t\tb.Errorf(\"%s failed (iteration %d): %s\", scenario.name, i, err.Error())\n\t\t\t}\n\t\t\ti++\n\t\t}\n\t})\n}\n\n// BenchmarkConcurrentBusinessPerformanceMode benchmarks concurrent business scenarios in performance V8 mode\n// Tests various scenarios with 100 concurrent users with isolate pool\n// Run with: go test -bench=BenchmarkConcurrentBusinessPerformanceMode -benchmem -benchtime=100x\nfunc BenchmarkConcurrentBusinessPerformanceMode(b *testing.B) {\n\ttestutils.Prepare(&testing.T{}, test.PrepareOption{V8Mode: \"performance\"})\n\tdefer testutils.Clean(&testing.T{})\n\n\tagent, err := assistant.Get(\"tests.create\")\n\tif err != nil {\n\t\tb.Fatalf(\"Failed to get assistant: %s\", err.Error())\n\t}\n\n\tif agent.HookScript == nil {\n\t\tb.Fatalf(\"Assistant has no script\")\n\t}\n\n\tscenarios := getBusinessScenarios()\n\n\tb.ResetTimer()\n\tb.RunParallel(func(pb *testing.PB) {\n\t\ti := 0\n\t\tfor pb.Next() {\n\t\t\tscenario := scenarios[i%len(scenarios)]\n\t\t\tctx := newBenchContext(\"bench-concurrent-business\", \"tests.create\")\n\t\t\t_, _, err := agent.HookScript.Create(ctx, []context.Message{\n\t\t\t\t{Role: \"user\", Content: scenario.content},\n\t\t\t})\n\t\t\tif err != nil {\n\t\t\t\tb.Errorf(\"%s failed (iteration %d): %s\", scenario.name, i, err.Error())\n\t\t\t}\n\t\t\ti++\n\t\t}\n\t})\n}\n\n// ============================================================================\n// Helper Functions\n// ============================================================================\n\n// getBusinessScenarios returns the business test scenarios\nfunc getBusinessScenarios() []struct {\n\tname string\n\tcontent string\n} {\n\treturn []struct {\n\t\tname string\n\t\tcontent string\n\t}{\n\t\t{name: \"FullResponse\", content: \"return_full\"},\n\t\t{name: \"PartialResponse\", content: \"return_partial\"},\n\t\t{name: \"ProcessCall\", content: \"return_process\"},\n\t\t{name: \"ContextAdjustment\", content: \"adjust_context\"},\n\t\t{name: \"NestedScriptCall\", content: \"nested_script_call\"},\n\t\t{name: \"DeepNestedCall\", content: \"deep_nested_call\"},\n\t}\n}\n\n// newBenchContext creates a minimal context for benchmarking\nfunc newBenchContext(chatID, assistantID string) *context.Context {\n\tauthorized := &types.AuthorizedInfo{\n\t\tSubject: \"bench-user\",\n\t\tClientID: \"bench-client\",\n\t\tUserID: \"bench-user-123\",\n\t\tTeamID: \"bench-team-456\",\n\t\tTenantID: \"bench-tenant-789\",\n\t\tConstraints: types.DataConstraints{\n\t\t\tTeamOnly: true,\n\t\t\tExtra: map[string]interface{}{\n\t\t\t\t\"department\": \"engineering\",\n\t\t\t},\n\t\t},\n\t}\n\n\tctx := context.New(stdContext.Background(), authorized, chatID)\n\tctx.AssistantID = assistantID\n\tctx.Locale = \"en-us\"\n\tctx.Theme = \"light\"\n\tctx.Client = context.Client{\n\t\tType: \"web\",\n\t\tUserAgent: \"BenchAgent/1.0\",\n\t\tIP: \"127.0.0.1\",\n\t}\n\tctx.Referer = context.RefererAPI\n\tctx.Accept = context.AcceptWebCUI\n\tctx.Route = \"\"\n\tctx.Metadata = make(map[string]interface{})\n\treturn ctx\n}\n" }, { "path": "agent/assistant/hook/create_mem_test.go", "content": "package hook_test\n\nimport (\n\tstdContext \"context\"\n\t\"runtime\"\n\t\"testing\"\n\t\"time\"\n\n\t\"github.com/yaoapp/yao/agent/assistant\"\n\t\"github.com/yaoapp/yao/agent/context\"\n\t\"github.com/yaoapp/yao/agent/testutils\"\n\t\"github.com/yaoapp/yao/openapi/oauth/types\"\n\t\"github.com/yaoapp/yao/test\"\n)\n\n// ============================================================================\n// Memory Leak Detection Tests\n// ============================================================================\n\n// TestMemoryLeakStandardMode checks for memory leaks in standard V8 mode\n// Run with: go test -run=TestMemoryLeakStandardMode -v\nfunc TestMemoryLeakStandardMode(t *testing.T) {\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\tagent, err := assistant.Get(\"tests.create\")\n\tif err != nil {\n\t\tt.Fatalf(\"Failed to get assistant: %s\", err.Error())\n\t}\n\n\tif agent.HookScript == nil {\n\t\tt.Fatalf(\"Assistant has no script\")\n\t}\n\n\t// Warm up - execute a few times to stabilize memory\n\tfor i := 0; i < 10; i++ {\n\t\tctx := newMemTestContext(\"warmup\", \"tests.create\")\n\t\t_, _, _ = agent.HookScript.Create(ctx, []context.Message{\n\t\t\t{Role: \"user\", Content: \"Hello\"},\n\t\t})\n\t\tctx.Release()\n\t}\n\n\t// Force GC and get baseline memory\n\truntime.GC()\n\ttime.Sleep(100 * time.Millisecond)\n\tvar baseline runtime.MemStats\n\truntime.ReadMemStats(&baseline)\n\n\t// Execute many iterations\n\titerations := 1000\n\tfor i := 0; i < iterations; i++ {\n\t\tctx := newMemTestContext(\"mem-test-standard\", \"tests.create\")\n\t\t_, _, err := agent.HookScript.Create(ctx, []context.Message{\n\t\t\t{Role: \"user\", Content: \"Hello\"},\n\t\t})\n\t\tif err != nil {\n\t\t\tt.Errorf(\"Create failed at iteration %d: %s\", i, err.Error())\n\t\t}\n\n\t\t// Release context resources\n\t\tctx.Release()\n\n\t\t// Periodic GC to help detect leaks faster\n\t\tif i%100 == 0 {\n\t\t\truntime.GC()\n\t\t}\n\t}\n\n\t// Force GC and check final memory\n\truntime.GC()\n\ttime.Sleep(100 * time.Millisecond)\n\tvar final runtime.MemStats\n\truntime.ReadMemStats(&final)\n\n\t// Calculate memory growth\n\tbaselineHeap := baseline.HeapAlloc\n\tfinalHeap := final.HeapAlloc\n\tgrowth := int64(finalHeap) - int64(baselineHeap)\n\tgrowthPerIteration := float64(growth) / float64(iterations)\n\n\tt.Logf(\"Memory Statistics (Standard Mode):\")\n\tt.Logf(\" Iterations: %d\", iterations)\n\tt.Logf(\" Baseline HeapAlloc: %d bytes (%.2f MB)\", baselineHeap, float64(baselineHeap)/1024/1024)\n\tt.Logf(\" Final HeapAlloc: %d bytes (%.2f MB)\", finalHeap, float64(finalHeap)/1024/1024)\n\tt.Logf(\" Total Growth: %d bytes (%.2f MB)\", growth, float64(growth)/1024/1024)\n\tt.Logf(\" Growth per iteration: %.2f bytes\", growthPerIteration)\n\tt.Logf(\" Total Alloc: %d bytes (%.2f MB)\", final.TotalAlloc, float64(final.TotalAlloc)/1024/1024)\n\tt.Logf(\" Mallocs: %d\", final.Mallocs)\n\tt.Logf(\" Frees: %d\", final.Frees)\n\tt.Logf(\" Live Objects: %d\", final.Mallocs-final.Frees)\n\tt.Logf(\" GC Runs: %d\", final.NumGC-baseline.NumGC)\n\n\t// Check for memory leak\n\t// Standard mode creates/disposes isolates per request, so some overhead is expected\n\t// Allow up to 20KB growth per iteration as threshold\n\t// This accounts for V8 isolate creation/disposal overhead and bridge management\n\t// Significant leaks would show much higher growth rates (50KB+)\n\tmaxGrowthPerIteration := 20480.0 // 20 KB\n\tif growthPerIteration > maxGrowthPerIteration {\n\t\tt.Errorf(\"Possible memory leak detected: %.2f bytes/iteration (threshold: %.2f bytes/iteration)\",\n\t\t\tgrowthPerIteration, maxGrowthPerIteration)\n\t} else {\n\t\tt.Logf(\"✓ Memory growth is within acceptable range (%.2f bytes/iteration)\", growthPerIteration)\n\t}\n}\n\n// TestMemoryLeakPerformanceMode checks for memory leaks in performance V8 mode\n// Run with: go test -run=TestMemoryLeakPerformanceMode -v\nfunc TestMemoryLeakPerformanceMode(t *testing.T) {\n\ttestutils.Prepare(t, test.PrepareOption{V8Mode: \"performance\"})\n\tdefer testutils.Clean(t)\n\n\tagent, err := assistant.Get(\"tests.create\")\n\tif err != nil {\n\t\tt.Fatalf(\"Failed to get assistant: %s\", err.Error())\n\t}\n\n\tif agent.HookScript == nil {\n\t\tt.Fatalf(\"Assistant has no script\")\n\t}\n\n\t// Warm up - execute a few times to stabilize memory and fill isolate pool\n\tfor i := 0; i < 20; i++ {\n\t\tctx := newMemTestContext(\"warmup\", \"tests.create\")\n\t\t_, _, _ = agent.HookScript.Create(ctx, []context.Message{\n\t\t\t{Role: \"user\", Content: \"Hello\"},\n\t\t})\n\t\tctx.Release()\n\t}\n\n\t// Force GC and get baseline memory\n\truntime.GC()\n\ttime.Sleep(100 * time.Millisecond)\n\tvar baseline runtime.MemStats\n\truntime.ReadMemStats(&baseline)\n\n\t// Execute many iterations\n\titerations := 1000\n\tfor i := 0; i < iterations; i++ {\n\t\tctx := newMemTestContext(\"mem-test-performance\", \"tests.create\")\n\t\t_, _, err := agent.HookScript.Create(ctx, []context.Message{\n\t\t\t{Role: \"user\", Content: \"Hello\"},\n\t\t})\n\t\tif err != nil {\n\t\t\tt.Errorf(\"Create failed at iteration %d: %s\", i, err.Error())\n\t\t}\n\n\t\t// Release context resources\n\t\tctx.Release()\n\n\t\t// Periodic GC\n\t\tif i%100 == 0 {\n\t\t\truntime.GC()\n\t\t}\n\t}\n\n\t// Force GC and check final memory\n\truntime.GC()\n\ttime.Sleep(100 * time.Millisecond)\n\tvar final runtime.MemStats\n\truntime.ReadMemStats(&final)\n\n\t// Calculate memory growth\n\tbaselineHeap := baseline.HeapAlloc\n\tfinalHeap := final.HeapAlloc\n\tgrowth := int64(finalHeap) - int64(baselineHeap)\n\tgrowthPerIteration := float64(growth) / float64(iterations)\n\n\tt.Logf(\"Memory Statistics (Performance Mode):\")\n\tt.Logf(\" Iterations: %d\", iterations)\n\tt.Logf(\" Baseline HeapAlloc: %d bytes (%.2f MB)\", baselineHeap, float64(baselineHeap)/1024/1024)\n\tt.Logf(\" Final HeapAlloc: %d bytes (%.2f MB)\", finalHeap, float64(finalHeap)/1024/1024)\n\tt.Logf(\" Total Growth: %d bytes (%.2f MB)\", growth, float64(growth)/1024/1024)\n\tt.Logf(\" Growth per iteration: %.2f bytes\", growthPerIteration)\n\tt.Logf(\" Total Alloc: %d bytes (%.2f MB)\", final.TotalAlloc, float64(final.TotalAlloc)/1024/1024)\n\tt.Logf(\" Mallocs: %d\", final.Mallocs)\n\tt.Logf(\" Frees: %d\", final.Frees)\n\tt.Logf(\" Live Objects: %d\", final.Mallocs-final.Frees)\n\tt.Logf(\" GC Runs: %d\", final.NumGC-baseline.NumGC)\n\n\t// Performance mode should have less growth due to isolate reuse\n\t// Allow up to 5KB per iteration as threshold\n\tmaxGrowthPerIteration := 5120.0\n\tif growthPerIteration > maxGrowthPerIteration {\n\t\tt.Errorf(\"Possible memory leak detected: %.2f bytes/iteration (threshold: %.2f bytes/iteration)\",\n\t\t\tgrowthPerIteration, maxGrowthPerIteration)\n\t} else {\n\t\tt.Logf(\"✓ Memory growth is within acceptable range\")\n\t}\n}\n\n// TestMemoryLeakBusinessScenarios checks for memory leaks with business logic\n// Run with: go test -run=TestMemoryLeakBusinessScenarios -v\nfunc TestMemoryLeakBusinessScenarios(t *testing.T) {\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\tagent, err := assistant.Get(\"tests.create\")\n\tif err != nil {\n\t\tt.Fatalf(\"Failed to get assistant: %s\", err.Error())\n\t}\n\n\tif agent.HookScript == nil {\n\t\tt.Fatalf(\"Assistant has no script\")\n\t}\n\n\tscenarios := []struct {\n\t\tname string\n\t\tcontent string\n\t}{\n\t\t{name: \"FullResponse\", content: \"return_full\"},\n\t\t{name: \"PartialResponse\", content: \"return_partial\"},\n\t\t{name: \"ProcessCall\", content: \"return_process\"},\n\t\t{name: \"ContextAdjustment\", content: \"adjust_context\"},\n\t\t{name: \"NestedScriptCall\", content: \"nested_script_call\"},\n\t\t{name: \"DeepNestedCall\", content: \"deep_nested_call\"},\n\t}\n\n\t// Warm up\n\tfor i := 0; i < 10; i++ {\n\t\tctx := newMemTestContext(\"warmup\", \"tests.create\")\n\t\t_, _, _ = agent.HookScript.Create(ctx, []context.Message{\n\t\t\t{Role: \"user\", Content: \"return_full\"},\n\t\t})\n\t\tctx.Release()\n\t}\n\n\t// Test each scenario\n\tfor _, scenario := range scenarios {\n\t\tt.Run(scenario.name, func(t *testing.T) {\n\t\t\t// Get baseline\n\t\t\truntime.GC()\n\t\t\ttime.Sleep(50 * time.Millisecond)\n\t\t\tvar baseline runtime.MemStats\n\t\t\truntime.ReadMemStats(&baseline)\n\n\t\t\t// Execute iterations (reduced to avoid V8 OOM)\n\t\t\titerations := 200\n\t\t\tfor i := 0; i < iterations; i++ {\n\t\t\t\tctx := newMemTestContext(\"mem-test-business\", \"tests.create\")\n\t\t\t\t_, _, err := agent.HookScript.Create(ctx, []context.Message{\n\t\t\t\t\t{Role: \"user\", Content: scenario.content},\n\t\t\t\t})\n\t\t\t\tif err != nil {\n\t\t\t\t\tt.Errorf(\"Create failed at iteration %d: %s\", i, err.Error())\n\t\t\t\t}\n\t\t\t\tctx.Release()\n\n\t\t\t\tif i%50 == 0 {\n\t\t\t\t\truntime.GC()\n\t\t\t\t}\n\t\t\t}\n\n\t\t\t// Check final memory\n\t\t\truntime.GC()\n\t\t\ttime.Sleep(50 * time.Millisecond)\n\t\t\tvar final runtime.MemStats\n\t\t\truntime.ReadMemStats(&final)\n\n\t\t\tgrowth := int64(final.HeapAlloc) - int64(baseline.HeapAlloc)\n\t\t\tgrowthPerIteration := float64(growth) / float64(iterations)\n\n\t\t\tt.Logf(\" Baseline HeapAlloc: %d bytes (%.2f MB)\", baseline.HeapAlloc, float64(baseline.HeapAlloc)/1024/1024)\n\t\t\tt.Logf(\" Final HeapAlloc: %d bytes (%.2f MB)\", final.HeapAlloc, float64(final.HeapAlloc)/1024/1024)\n\t\t\tt.Logf(\" Growth: %d bytes (%.2f MB)\", growth, float64(growth)/1024/1024)\n\t\t\tt.Logf(\" Growth/iteration: %.2f bytes\", growthPerIteration)\n\n\t\t\t// Business scenarios may have more memory usage due to complex operations\n\t\t\t// Allow up to 20KB per iteration as threshold\n\t\t\t// Note: Some scenarios like ContextAdjustment generate dynamic timestamps,\n\t\t\t// causing slightly higher memory usage. Real leaks would show 50KB+ growth.\n\t\t\tmaxGrowthPerIteration := 20480.0\n\t\t\tif growthPerIteration > maxGrowthPerIteration {\n\t\t\t\tt.Errorf(\"Possible memory leak: %.2f bytes/iteration (threshold: %.2f)\",\n\t\t\t\t\tgrowthPerIteration, maxGrowthPerIteration)\n\t\t\t} else {\n\t\t\t\tt.Logf(\" ✓ Memory growth is within acceptable range\")\n\t\t\t}\n\t\t})\n\t}\n}\n\n// TestMemoryLeakConcurrent checks for memory leaks under concurrent load\n// Run with: go test -run=TestMemoryLeakConcurrent -v\nfunc TestMemoryLeakConcurrent(t *testing.T) {\n\ttestutils.Prepare(t, test.PrepareOption{V8Mode: \"performance\"})\n\tdefer testutils.Clean(t)\n\n\tagent, err := assistant.Get(\"tests.create\")\n\tif err != nil {\n\t\tt.Fatalf(\"Failed to get assistant: %s\", err.Error())\n\t}\n\n\tif agent.HookScript == nil {\n\t\tt.Fatalf(\"Assistant has no script\")\n\t}\n\n\t// Warm up\n\tfor i := 0; i < 20; i++ {\n\t\tctx := newMemTestContext(\"warmup\", \"tests.create\")\n\t\t_, _, _ = agent.HookScript.Create(ctx, []context.Message{\n\t\t\t{Role: \"user\", Content: \"Hello\"},\n\t\t})\n\t\tctx.Release()\n\t}\n\n\t// Get baseline\n\truntime.GC()\n\ttime.Sleep(100 * time.Millisecond)\n\tvar baseline runtime.MemStats\n\truntime.ReadMemStats(&baseline)\n\n\t// Run concurrent load\n\titerations := 1000\n\tconcurrency := 10\n\titerPerGoroutine := iterations / concurrency\n\n\tdone := make(chan bool, concurrency)\n\tfor g := 0; g < concurrency; g++ {\n\t\tgo func(id int) {\n\t\t\tdefer func() { done <- true }()\n\t\t\tfor i := 0; i < iterPerGoroutine; i++ {\n\t\t\t\tctx := newMemTestContext(\"mem-test-concurrent\", \"tests.create\")\n\t\t\t\t_, _, err := agent.HookScript.Create(ctx, []context.Message{\n\t\t\t\t\t{Role: \"user\", Content: \"Hello\"},\n\t\t\t\t})\n\t\t\t\tif err != nil {\n\t\t\t\t\tt.Errorf(\"Goroutine %d failed at iteration %d: %s\", id, i, err.Error())\n\t\t\t\t}\n\t\t\t\tctx.Release()\n\t\t\t}\n\t\t}(g)\n\t}\n\n\t// Wait for all goroutines\n\tfor g := 0; g < concurrency; g++ {\n\t\t<-done\n\t}\n\n\t// Check final memory\n\truntime.GC()\n\ttime.Sleep(100 * time.Millisecond)\n\tvar final runtime.MemStats\n\truntime.ReadMemStats(&final)\n\n\tgrowth := int64(final.HeapAlloc) - int64(baseline.HeapAlloc)\n\tgrowthPerIteration := float64(growth) / float64(iterations)\n\n\tt.Logf(\"Memory Statistics (Concurrent Load):\")\n\tt.Logf(\" Iterations: %d\", iterations)\n\tt.Logf(\" Concurrency: %d\", concurrency)\n\tt.Logf(\" Baseline HeapAlloc: %d bytes (%.2f MB)\", baseline.HeapAlloc, float64(baseline.HeapAlloc)/1024/1024)\n\tt.Logf(\" Final HeapAlloc: %d bytes (%.2f MB)\", final.HeapAlloc, float64(final.HeapAlloc)/1024/1024)\n\tt.Logf(\" Growth: %d bytes (%.2f MB)\", growth, float64(growth)/1024/1024)\n\tt.Logf(\" Growth/iteration: %.2f bytes\", growthPerIteration)\n\tt.Logf(\" GC Runs: %d\", final.NumGC-baseline.NumGC)\n\n\t// Concurrent scenarios may have slightly more overhead\n\tmaxGrowthPerIteration := 10240.0\n\tif growthPerIteration > maxGrowthPerIteration {\n\t\tt.Errorf(\"Possible memory leak: %.2f bytes/iteration (threshold: %.2f)\",\n\t\t\tgrowthPerIteration, maxGrowthPerIteration)\n\t} else {\n\t\tt.Logf(\"✓ Memory growth is within acceptable range\")\n\t}\n}\n\n// TestMemoryLeakNestedCalls checks for memory leaks with nested script calls\n// Run with: go test -run=TestMemoryLeakNestedCalls -v\nfunc TestMemoryLeakNestedCalls(t *testing.T) {\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\tagent, err := assistant.Get(\"tests.create\")\n\tif err != nil {\n\t\tt.Fatalf(\"Failed to get assistant: %s\", err.Error())\n\t}\n\n\tif agent.HookScript == nil {\n\t\tt.Fatalf(\"Assistant has no script\")\n\t}\n\n\t// Warm up\n\tfor i := 0; i < 10; i++ {\n\t\tctx := newMemTestContext(\"warmup\", \"tests.create\")\n\t\t_, _, _ = agent.HookScript.Create(ctx, []context.Message{\n\t\t\t{Role: \"user\", Content: \"nested_script_call\"},\n\t\t})\n\t\tctx.Release()\n\t}\n\n\t// Get baseline\n\truntime.GC()\n\ttime.Sleep(100 * time.Millisecond)\n\tvar baseline runtime.MemStats\n\truntime.ReadMemStats(&baseline)\n\n\t// Execute iterations with nested calls\n\t// Nested calls: hook -> scripts.tests.create.NestedCall -> GetRoles/GetRole -> models\n\titerations := 200\n\tfor i := 0; i < iterations; i++ {\n\t\tctx := newMemTestContext(\"mem-test-nested\", \"tests.create\")\n\t\t_, _, err := agent.HookScript.Create(ctx, []context.Message{\n\t\t\t{Role: \"user\", Content: \"deep_nested_call\"},\n\t\t})\n\t\tif err != nil {\n\t\t\tt.Errorf(\"Nested call failed at iteration %d: %s\", i, err.Error())\n\t\t}\n\t\tctx.Release()\n\n\t\tif i%50 == 0 {\n\t\t\truntime.GC()\n\t\t}\n\t}\n\n\t// Check final memory\n\truntime.GC()\n\ttime.Sleep(100 * time.Millisecond)\n\tvar final runtime.MemStats\n\truntime.ReadMemStats(&final)\n\n\tgrowth := int64(final.HeapAlloc) - int64(baseline.HeapAlloc)\n\tgrowthPerIteration := float64(growth) / float64(iterations)\n\n\tt.Logf(\"Memory Statistics (Nested Calls):\")\n\tt.Logf(\" Iterations: %d\", iterations)\n\tt.Logf(\" Baseline HeapAlloc: %d bytes (%.2f MB)\", baseline.HeapAlloc, float64(baseline.HeapAlloc)/1024/1024)\n\tt.Logf(\" Final HeapAlloc: %d bytes (%.2f MB)\", final.HeapAlloc, float64(final.HeapAlloc)/1024/1024)\n\tt.Logf(\" Growth: %d bytes (%.2f MB)\", growth, float64(growth)/1024/1024)\n\tt.Logf(\" Growth/iteration: %.2f bytes\", growthPerIteration)\n\tt.Logf(\" GC Runs: %d\", final.NumGC-baseline.NumGC)\n\n\t// Nested calls involve database operations, so allow more overhead\n\t// Allow up to 20KB per iteration as threshold\n\tmaxGrowthPerIteration := 20480.0\n\tif growthPerIteration > maxGrowthPerIteration {\n\t\tt.Errorf(\"Possible memory leak: %.2f bytes/iteration (threshold: %.2f)\",\n\t\t\tgrowthPerIteration, maxGrowthPerIteration)\n\t} else {\n\t\tt.Logf(\"✓ Memory growth is within acceptable range\")\n\t}\n}\n\n// TestMemoryLeakNestedConcurrent checks for memory leaks with concurrent nested calls\n// Run with: go test -run=TestMemoryLeakNestedConcurrent -v\nfunc TestMemoryLeakNestedConcurrent(t *testing.T) {\n\ttestutils.Prepare(t, test.PrepareOption{V8Mode: \"performance\"})\n\tdefer testutils.Clean(t)\n\n\tagent, err := assistant.Get(\"tests.create\")\n\tif err != nil {\n\t\tt.Fatalf(\"Failed to get assistant: %s\", err.Error())\n\t}\n\n\tif agent.HookScript == nil {\n\t\tt.Fatalf(\"Assistant has no script\")\n\t}\n\n\t// Warm up\n\tfor i := 0; i < 20; i++ {\n\t\tctx := newMemTestContext(\"warmup\", \"tests.create\")\n\t\t_, _, _ = agent.HookScript.Create(ctx, []context.Message{\n\t\t\t{Role: \"user\", Content: \"nested_script_call\"},\n\t\t})\n\t\tctx.Release()\n\t}\n\n\t// Get baseline\n\truntime.GC()\n\ttime.Sleep(100 * time.Millisecond)\n\tvar baseline runtime.MemStats\n\truntime.ReadMemStats(&baseline)\n\n\t// Run concurrent nested calls\n\titerations := 500\n\tconcurrency := 10\n\titerPerGoroutine := iterations / concurrency\n\n\tdone := make(chan bool, concurrency)\n\tfor g := 0; g < concurrency; g++ {\n\t\tgo func(id int) {\n\t\t\tdefer func() { done <- true }()\n\t\t\tfor i := 0; i < iterPerGoroutine; i++ {\n\t\t\t\tctx := newMemTestContext(\"mem-test-nested-concurrent\", \"tests.create\")\n\t\t\t\t_, _, err := agent.HookScript.Create(ctx, []context.Message{\n\t\t\t\t\t{Role: \"user\", Content: \"deep_nested_call\"},\n\t\t\t\t})\n\t\t\t\tif err != nil {\n\t\t\t\t\tt.Errorf(\"Goroutine %d nested call failed at iteration %d: %s\", id, i, err.Error())\n\t\t\t\t}\n\t\t\t\tctx.Release()\n\t\t\t}\n\t\t}(g)\n\t}\n\n\t// Wait for all goroutines\n\tfor g := 0; g < concurrency; g++ {\n\t\t<-done\n\t}\n\n\t// Check final memory\n\truntime.GC()\n\ttime.Sleep(100 * time.Millisecond)\n\tvar final runtime.MemStats\n\truntime.ReadMemStats(&final)\n\n\tgrowth := int64(final.HeapAlloc) - int64(baseline.HeapAlloc)\n\tgrowthPerIteration := float64(growth) / float64(iterations)\n\n\tt.Logf(\"Memory Statistics (Concurrent Nested Calls):\")\n\tt.Logf(\" Iterations: %d\", iterations)\n\tt.Logf(\" Concurrency: %d\", concurrency)\n\tt.Logf(\" Baseline HeapAlloc: %d bytes (%.2f MB)\", baseline.HeapAlloc, float64(baseline.HeapAlloc)/1024/1024)\n\tt.Logf(\" Final HeapAlloc: %d bytes (%.2f MB)\", final.HeapAlloc, float64(final.HeapAlloc)/1024/1024)\n\tt.Logf(\" Growth: %d bytes (%.2f MB)\", growth, float64(growth)/1024/1024)\n\tt.Logf(\" Growth/iteration: %.2f bytes\", growthPerIteration)\n\tt.Logf(\" GC Runs: %d\", final.NumGC-baseline.NumGC)\n\n\t// Concurrent nested calls with database operations\n\t// Allow up to 25KB per iteration as threshold\n\tmaxGrowthPerIteration := 25600.0\n\tif growthPerIteration > maxGrowthPerIteration {\n\t\tt.Errorf(\"Possible memory leak: %.2f bytes/iteration (threshold: %.2f)\",\n\t\t\tgrowthPerIteration, maxGrowthPerIteration)\n\t} else {\n\t\tt.Logf(\"✓ Memory growth is within acceptable range\")\n\t}\n}\n\n// TestIsolateDisposal verifies that isolates are properly disposed in standard mode\n// Run with: go test -run=TestIsolateDisposal -v\nfunc TestIsolateDisposal(t *testing.T) {\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\tagent, err := assistant.Get(\"tests.create\")\n\tif err != nil {\n\t\tt.Fatalf(\"Failed to get assistant: %s\", err.Error())\n\t}\n\n\tif agent.HookScript == nil {\n\t\tt.Fatalf(\"Assistant has no script\")\n\t}\n\n\t// Track goroutine count to detect goroutine leaks\n\tinitialGoroutines := runtime.NumGoroutine()\n\n\t// Execute multiple iterations\n\titerations := 100\n\tfor i := 0; i < iterations; i++ {\n\t\tctx := newMemTestContext(\"disposal-test\", \"tests.create\")\n\t\t_, _, err := agent.HookScript.Create(ctx, []context.Message{\n\t\t\t{Role: \"user\", Content: \"Hello\"},\n\t\t})\n\t\tif err != nil {\n\t\t\tt.Errorf(\"Create failed at iteration %d: %s\", i, err.Error())\n\t\t}\n\t\tctx.Release()\n\t}\n\n\t// Give time for cleanup\n\ttime.Sleep(200 * time.Millisecond)\n\truntime.GC()\n\ttime.Sleep(200 * time.Millisecond)\n\n\tfinalGoroutines := runtime.NumGoroutine()\n\tgoroutineGrowth := finalGoroutines - initialGoroutines\n\n\tt.Logf(\"Goroutine Statistics:\")\n\tt.Logf(\" Initial: %d\", initialGoroutines)\n\tt.Logf(\" Final: %d\", finalGoroutines)\n\tt.Logf(\" Growth: %d\", goroutineGrowth)\n\n\t// Allow some goroutine growth for runtime internals\n\t//\n\t// ROOT CAUSE ANALYSIS:\n\t// Each Create() call creates a Trace, which starts 2 goroutines:\n\t// 1. trace/pubsub.(*PubSub).forward() - PubSub event forwarding\n\t// 2. trace.(*manager).startStateWorker() - State machine worker\n\t//\n\t// These goroutines exit when Release() closes their channels, but:\n\t// - Exit is ASYNCHRONOUS (goroutine needs to reach select statement)\n\t// - Go runtime needs time to schedule and cleanup\n\t// - In rapid iterations, new goroutines are created before old ones fully exit\n\t//\n\t// This is NOT a true leak:\n\t// ✓ Goroutines eventually exit (channels are closed)\n\t// ✓ No unbounded growth (they will be GC'd)\n\t// ✓ Typical pattern for async cleanup in Go\n\t//\n\t// Acceptable: ~2 goroutines per iteration (trace pubsub + state worker)\n\t// Concerning: >5 goroutines per iteration (indicates goroutines NOT exiting)\n\tmaxGoroutineGrowthPerIteration := 5.0\n\tgrowthPerIteration := float64(goroutineGrowth) / float64(iterations)\n\n\tif growthPerIteration > maxGoroutineGrowthPerIteration {\n\t\tt.Errorf(\"Goroutine leak detected: %.2f goroutines per iteration (threshold: %.2f)\",\n\t\t\tgrowthPerIteration, maxGoroutineGrowthPerIteration)\n\t\tt.Errorf(\"This indicates goroutines are NOT being cleaned up properly\")\n\t} else {\n\t\tt.Logf(\"✓ Goroutine growth is acceptable: %.2f per iteration\", growthPerIteration)\n\t\tt.Logf(\" (Trace creates 2 goroutines per call: pubsub.forward + stateWorker)\")\n\t\tt.Logf(\" (These exit asynchronously after Release(), causing temporary accumulation)\")\n\t}\n}\n\n// ============================================================================\n// Helper Functions\n// ============================================================================\n\n// newMemTestContext creates a context for memory leak testing\nfunc newMemTestContext(chatID, assistantID string) *context.Context {\n\tauthorized := &types.AuthorizedInfo{\n\t\tSubject: \"mem-test-user\",\n\t\tClientID: \"mem-test-client\",\n\t\tUserID: \"mem-user-123\",\n\t\tTeamID: \"mem-team-456\",\n\t\tTenantID: \"mem-tenant-789\",\n\t\tConstraints: types.DataConstraints{\n\t\t\tTeamOnly: true,\n\t\t\tExtra: map[string]interface{}{\n\t\t\t\t\"department\": \"engineering\",\n\t\t\t},\n\t\t},\n\t}\n\n\tctx := context.New(stdContext.Background(), authorized, chatID)\n\tctx.AssistantID = assistantID\n\tctx.Locale = \"en-us\"\n\tctx.Theme = \"light\"\n\tctx.Client = context.Client{\n\t\tType: \"web\",\n\t\tUserAgent: \"MemTestAgent/1.0\",\n\t\tIP: \"127.0.0.1\",\n\t}\n\tctx.Referer = context.RefererAPI\n\tctx.Accept = context.AcceptWebCUI\n\tctx.Route = \"\"\n\tctx.Metadata = make(map[string]interface{})\n\treturn ctx\n}\n" }, { "path": "agent/assistant/hook/create_nested_test.go", "content": "package hook_test\n\nimport (\n\t\"sync\"\n\t\"testing\"\n\n\t\"github.com/yaoapp/yao/agent/assistant\"\n\t\"github.com/yaoapp/yao/agent/context\"\n\t\"github.com/yaoapp/yao/agent/testutils\"\n)\n\n// TestNestedScriptCall tests nested script calls with V8 context sharing\n// This test calls: hook -> scripts.tests.create.NestedCall -> GetRoles/GetRole -> models\nfunc TestNestedScriptCall(t *testing.T) {\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\tagent, err := assistant.Get(\"tests.create\")\n\tif err != nil {\n\t\tt.Fatalf(\"Failed to get assistant: %s\", err.Error())\n\t}\n\n\tif agent.HookScript == nil {\n\t\tt.Fatalf(\"Assistant has no script\")\n\t}\n\n\t// Create context\n\tctx := newTestContext(\"test-nested-call\", \"tests.create\")\n\n\t// Call with deep_nested_call scenario\n\t// This will: hook -> scripts.tests.create.NestedCall -> GetRoles -> model\n\tres, _, err := agent.HookScript.Create(ctx, []context.Message{\n\t\t{Role: \"user\", Content: \"deep_nested_call\"},\n\t})\n\n\tif err != nil {\n\t\tt.Fatalf(\"Nested call failed: %s\", err.Error())\n\t}\n\n\tif res == nil {\n\t\tt.Fatal(\"Expected non-nil response\")\n\t}\n\n\t// Verify messages\n\tif len(res.Messages) == 0 {\n\t\tt.Fatal(\"Expected messages in response\")\n\t}\n\n\tt.Logf(\"✓ Nested script call completed successfully\")\n\tt.Logf(\" Messages count: %d\", len(res.Messages))\n\tif res.Metadata != nil {\n\t\tt.Logf(\" Metadata: %+v\", res.Metadata)\n\t}\n}\n\n// TestNestedScriptCallConcurrent tests nested script calls under high concurrency\n// Simulates 100 concurrent users making nested script calls\nfunc TestNestedScriptCallConcurrent(t *testing.T) {\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\tagent, err := assistant.Get(\"tests.create\")\n\tif err != nil {\n\t\tt.Fatalf(\"Failed to get assistant: %s\", err.Error())\n\t}\n\n\tif agent.HookScript == nil {\n\t\tt.Fatalf(\"Assistant has no script\")\n\t}\n\n\t// High concurrency test: 100 concurrent users (testing race condition)\n\tconcurrency := 100\n\titerations := 1 // Each user makes 1 call\n\n\tvar wg sync.WaitGroup\n\terrors := make(chan error, concurrency*iterations)\n\n\tt.Logf(\"Starting concurrent test: %d users × %d iterations = %d total calls\",\n\t\tconcurrency, iterations, concurrency*iterations)\n\n\tfor i := 0; i < concurrency; i++ {\n\t\twg.Add(1)\n\t\tgo func(userID int) {\n\t\t\tdefer wg.Done()\n\n\t\t\tfor j := 0; j < iterations; j++ {\n\t\t\t\tctx := newTestContext(\"test-concurrent\", \"tests.create\")\n\n\t\t\t\t_, _, err := agent.HookScript.Create(ctx, []context.Message{\n\t\t\t\t\t{Role: \"user\", Content: \"deep_nested_call\"},\n\t\t\t\t})\n\n\t\t\t\tif err != nil {\n\t\t\t\t\terrors <- err\n\t\t\t\t\treturn\n\t\t\t\t}\n\t\t\t}\n\t\t}(i)\n\t}\n\n\t// Wait for all goroutines to complete\n\twg.Wait()\n\tclose(errors)\n\n\t// Check for errors\n\terrorCount := 0\n\tfor err := range errors {\n\t\terrorCount++\n\t\tt.Errorf(\"Concurrent call failed: %s\", err.Error())\n\t}\n\n\tif errorCount > 0 {\n\t\tt.Fatalf(\"Failed with %d errors out of %d total calls\", errorCount, concurrency*iterations)\n\t}\n\n\tt.Logf(\"✓ All %d concurrent nested calls completed successfully\", concurrency*iterations)\n}\n" }, { "path": "agent/assistant/hook/create_test.go", "content": "package hook_test\n\nimport (\n\tstdContext \"context\"\n\t\"testing\"\n\n\t\"github.com/yaoapp/yao/agent/assistant\"\n\t\"github.com/yaoapp/yao/agent/context\"\n\t\"github.com/yaoapp/yao/agent/testutils\"\n\t\"github.com/yaoapp/yao/openapi/oauth/types\"\n)\n\n// newTestContext creates a Context for testing with commonly used fields pre-populated.\n// You can override any fields after creation as needed for specific test scenarios.\nfunc newTestContext(chatID, assistantID string) *context.Context {\n\tauthorized := &types.AuthorizedInfo{\n\t\tSubject: \"test-user\",\n\t\tClientID: \"test-client-id\",\n\t\tScope: \"openid profile email\",\n\t\tSessionID: \"test-session-id\",\n\t\tUserID: \"test-user-123\",\n\t\tTeamID: \"test-team-456\",\n\t\tTenantID: \"test-tenant-789\",\n\t\tRememberMe: true,\n\t\tConstraints: types.DataConstraints{\n\t\t\tOwnerOnly: false,\n\t\t\tCreatorOnly: false,\n\t\t\tEditorOnly: false,\n\t\t\tTeamOnly: true,\n\t\t\tExtra: map[string]interface{}{\n\t\t\t\t\"department\": \"engineering\",\n\t\t\t\t\"region\": \"us-west\",\n\t\t\t\t\"project\": \"yao\",\n\t\t\t},\n\t\t},\n\t}\n\n\tctx := context.New(stdContext.Background(), authorized, chatID)\n\tctx.AssistantID = assistantID\n\tctx.Locale = \"en-us\"\n\tctx.Theme = \"light\"\n\tctx.Client = context.Client{\n\t\tType: \"web\",\n\t\tUserAgent: \"TestAgent/1.0\",\n\t\tIP: \"127.0.0.1\",\n\t}\n\tctx.Referer = context.RefererAPI\n\tctx.Accept = context.AcceptWebCUI\n\tctx.Route = \"\"\n\tctx.Metadata = make(map[string]interface{})\n\treturn ctx\n}\n\n// TestCreate test the create hook\nfunc TestCreate(t *testing.T) {\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\tagent, err := assistant.Get(\"tests.create\")\n\tif err != nil {\n\t\tt.Fatalf(\"Failed to get the tests.create assistant: %s\", err.Error())\n\t}\n\n\tif agent.HookScript == nil {\n\t\tt.Fatalf(\"The tests.create assistant has no script\")\n\t}\n\n\t// Use the helper function to create a test context\n\tctx := newTestContext(\"chat-test-create-hook\", \"tests.create\")\n\n\t// Test scenario 1: Return null (should get nil response)\n\tt.Run(\"ReturnNull\", func(t *testing.T) {\n\t\tres, _, err := agent.HookScript.Create(ctx, []context.Message{{Role: \"user\", Content: \"return_null\"}})\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to create with null return: %s\", err.Error())\n\t\t}\n\t\tif res != nil {\n\t\t\tt.Errorf(\"Expected nil response for null return, got: %v\", res)\n\t\t}\n\t})\n\n\t// Test scenario 2: Return undefined (should get nil response)\n\tt.Run(\"ReturnUndefined\", func(t *testing.T) {\n\t\tres, _, err := agent.HookScript.Create(ctx, []context.Message{{Role: \"user\", Content: \"return_undefined\"}})\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to create with undefined return: %s\", err.Error())\n\t\t}\n\t\tif res != nil {\n\t\t\tt.Errorf(\"Expected nil response for undefined return, got: %v\", res)\n\t\t}\n\t})\n\n\t// Test scenario 3: Return empty object (should get empty HookCreateResponse)\n\tt.Run(\"ReturnEmpty\", func(t *testing.T) {\n\t\tres, _, err := agent.HookScript.Create(ctx, []context.Message{{Role: \"user\", Content: \"return_empty\"}})\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to create with empty return: %s\", err.Error())\n\t\t}\n\t\tif res == nil {\n\t\t\tt.Fatalf(\"Expected non-nil response for empty object, got nil\")\n\t\t}\n\t\tif len(res.Messages) != 0 {\n\t\t\tt.Errorf(\"Expected empty messages, got: %d messages\", len(res.Messages))\n\t\t}\n\t})\n\n\t// Test scenario 4: Return full response with all fields\n\tt.Run(\"ReturnFull\", func(t *testing.T) {\n\t\tres, _, err := agent.HookScript.Create(ctx, []context.Message{{Role: \"user\", Content: \"return_full\"}})\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to create with full return: %s\", err.Error())\n\t\t}\n\t\tif res == nil {\n\t\t\tt.Fatalf(\"Expected non-nil response, got nil\")\n\t\t}\n\n\t\t// Verify messages\n\t\tif len(res.Messages) != 2 {\n\t\t\tt.Errorf(\"Expected 2 messages, got: %d\", len(res.Messages))\n\t\t} else {\n\t\t\tif res.Messages[0].Role != context.RoleSystem {\n\t\t\t\tt.Errorf(\"Expected system role for first message, got: %s\", res.Messages[0].Role)\n\t\t\t}\n\t\t\tif res.Messages[1].Role != context.RoleUser {\n\t\t\t\tt.Errorf(\"Expected user role for second message, got: %s\", res.Messages[1].Role)\n\t\t\t}\n\t\t}\n\n\t\t// Verify audio config\n\t\tif res.Audio == nil {\n\t\t\tt.Error(\"Expected audio config, got nil\")\n\t\t} else {\n\t\t\tif res.Audio.Voice != \"alloy\" {\n\t\t\t\tt.Errorf(\"Expected voice 'alloy', got: %s\", res.Audio.Voice)\n\t\t\t}\n\t\t\tif res.Audio.Format != \"mp3\" {\n\t\t\t\tt.Errorf(\"Expected format 'mp3', got: %s\", res.Audio.Format)\n\t\t\t}\n\t\t}\n\n\t\t// Verify temperature\n\t\tif res.Temperature == nil {\n\t\t\tt.Error(\"Expected temperature, got nil\")\n\t\t} else if *res.Temperature != 0.7 {\n\t\t\tt.Errorf(\"Expected temperature 0.7, got: %f\", *res.Temperature)\n\t\t}\n\n\t\t// Verify max_tokens\n\t\tif res.MaxTokens == nil {\n\t\t\tt.Error(\"Expected max_tokens, got nil\")\n\t\t} else if *res.MaxTokens != 2000 {\n\t\t\tt.Errorf(\"Expected max_tokens 2000, got: %d\", *res.MaxTokens)\n\t\t}\n\n\t\t// Verify max_completion_tokens\n\t\tif res.MaxCompletionTokens == nil {\n\t\t\tt.Error(\"Expected max_completion_tokens, got nil\")\n\t\t} else if *res.MaxCompletionTokens != 1500 {\n\t\t\tt.Errorf(\"Expected max_completion_tokens 1500, got: %d\", *res.MaxCompletionTokens)\n\t\t}\n\t})\n\n\t// Test scenario 5: Return partial response\n\tt.Run(\"ReturnPartial\", func(t *testing.T) {\n\t\tres, _, err := agent.HookScript.Create(ctx, []context.Message{{Role: \"user\", Content: \"return_partial\"}})\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to create with partial return: %s\", err.Error())\n\t\t}\n\t\tif res == nil {\n\t\t\tt.Fatalf(\"Expected non-nil response, got nil\")\n\t\t}\n\n\t\t// Verify messages\n\t\tif len(res.Messages) != 1 {\n\t\t\tt.Errorf(\"Expected 1 message, got: %d\", len(res.Messages))\n\t\t}\n\n\t\t// Verify temperature\n\t\tif res.Temperature == nil {\n\t\t\tt.Error(\"Expected temperature, got nil\")\n\t\t} else if *res.Temperature != 0.5 {\n\t\t\tt.Errorf(\"Expected temperature 0.5, got: %f\", *res.Temperature)\n\t\t}\n\n\t\t// Verify optional fields are nil\n\t\tif res.Audio != nil {\n\t\t\tt.Errorf(\"Expected audio to be nil, got: %v\", res.Audio)\n\t\t}\n\t\tif res.MaxTokens != nil {\n\t\t\tt.Errorf(\"Expected max_tokens to be nil, got: %d\", *res.MaxTokens)\n\t\t}\n\t})\n\n\t// Test scenario 6: Process call - calls models.__yao.role.Get and adds to messages\n\tt.Run(\"ReturnProcess\", func(t *testing.T) {\n\t\tres, _, err := agent.HookScript.Create(ctx, []context.Message{{Role: \"user\", Content: \"return_process\"}})\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to create with process return: %s\", err.Error())\n\t\t}\n\t\tif res == nil {\n\t\t\tt.Fatalf(\"Expected non-nil response, got nil\")\n\t\t}\n\n\t\t// Verify messages - should have at least 1 (system message)\n\t\tif len(res.Messages) < 1 {\n\t\t\tt.Errorf(\"Expected at least 1 message, got: %d\", len(res.Messages))\n\t\t} else {\n\t\t\t// First message should be system role\n\t\t\tif res.Messages[0].Role != context.RoleSystem {\n\t\t\t\tt.Errorf(\"Expected system role for first message, got: %s\", res.Messages[0].Role)\n\t\t\t}\n\t\t\t// Check system message content\n\t\t\tif content, ok := res.Messages[0].Content.(string); ok {\n\t\t\t\tif content != \"Here are the available roles in the system:\" {\n\t\t\t\t\tt.Errorf(\"Unexpected system message content: %s\", content)\n\t\t\t\t}\n\t\t\t}\n\t\t}\n\t})\n\n\t// Test scenario 7: Default response\n\tt.Run(\"ReturnDefault\", func(t *testing.T) {\n\t\ttestContent := \"Hello, how are you?\"\n\t\tres, _, err := agent.HookScript.Create(ctx, []context.Message{{Role: \"user\", Content: testContent}})\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to create with default return: %s\", err.Error())\n\t\t}\n\t\tif res == nil {\n\t\t\tt.Fatalf(\"Expected non-nil response, got nil\")\n\t\t}\n\n\t\t// Verify messages\n\t\tif len(res.Messages) != 1 {\n\t\t\tt.Errorf(\"Expected 1 message, got: %d\", len(res.Messages))\n\t\t} else {\n\t\t\tif res.Messages[0].Role != context.RoleUser {\n\t\t\t\tt.Errorf(\"Expected user role, got: %s\", res.Messages[0].Role)\n\t\t\t}\n\t\t\tif content, ok := res.Messages[0].Content.(string); ok {\n\t\t\t\tif content != testContent {\n\t\t\t\t\tt.Errorf(\"Expected content '%s', got: '%s'\", testContent, content)\n\t\t\t\t}\n\t\t\t} else {\n\t\t\t\tt.Errorf(\"Expected string content, got: %T\", res.Messages[0].Content)\n\t\t\t}\n\t\t}\n\t})\n\n\t// Test scenario 8: Verify context fields - validates all context fields in JavaScript\n\tt.Run(\"VerifyContext\", func(t *testing.T) {\n\t\tres, _, err := agent.HookScript.Create(ctx, []context.Message{{Role: \"user\", Content: \"verify_context\"}})\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to create with verify_context: %s\", err.Error())\n\t\t}\n\t\tif res == nil {\n\t\t\tt.Fatalf(\"Expected non-nil response, got nil\")\n\t\t}\n\n\t\t// Verify we have messages\n\t\tif len(res.Messages) < 1 {\n\t\t\tt.Fatalf(\"Expected at least 1 message, got: %d\", len(res.Messages))\n\t\t}\n\n\t\t// First message should be system role with success/failure indicator\n\t\tif res.Messages[0].Role != context.RoleSystem {\n\t\t\tt.Errorf(\"Expected system role for first message, got: %s\", res.Messages[0].Role)\n\t\t}\n\n\t\t// Check the validation result\n\t\tcontent, ok := res.Messages[0].Content.(string)\n\t\tif !ok {\n\t\t\tt.Fatalf(\"Expected string content for system message, got: %T\", res.Messages[0].Content)\n\t\t}\n\n\t\t// The content should be \"success:all_fields_validated\"\n\t\tif content != \"success:all_fields_validated\" {\n\t\t\tt.Errorf(\"Context validation failed: %s\", content)\n\n\t\t\t// Print detailed validation results if available\n\t\t\tif len(res.Messages) > 1 {\n\t\t\t\tif details, ok := res.Messages[1].Content.(string); ok {\n\t\t\t\t\tt.Logf(\"Validation details:\\n%s\", details)\n\t\t\t\t}\n\t\t\t}\n\t\t} else {\n\t\t\tt.Log(\"✓ All context fields validated successfully in JavaScript\")\n\n\t\t\t// Optionally print validation details\n\t\t\tif len(res.Messages) > 1 {\n\t\t\t\tif details, ok := res.Messages[1].Content.(string); ok {\n\t\t\t\t\tt.Logf(\"Validation details:\\n%s\", details)\n\t\t\t\t}\n\t\t\t}\n\t\t}\n\t})\n\n\t// Test scenario 9: Adjust context fields - tests that context fields can be modified by the hook\n\tt.Run(\"AdjustContext\", func(t *testing.T) {\n\t\t// Create a fresh context for this test\n\t\tadjustCtx := newTestContext(\"chat-test-adjust\", \"tests.create\")\n\n\t\t// Call the hook which should adjust context fields\n\t\tres, _, err := agent.HookScript.Create(adjustCtx, []context.Message{{Role: \"user\", Content: \"adjust_context\"}})\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to create with adjust_context: %s\", err.Error())\n\t\t}\n\t\tif res == nil {\n\t\t\tt.Fatalf(\"Expected non-nil response, got nil\")\n\t\t}\n\n\t\t// Verify the response contains adjusted fields\n\t\t// Note: AssistantID cannot be overridden by hooks, removed from HookCreateResponse\n\t\tif res.Connector != \"adjusted-connector\" {\n\t\t\tt.Errorf(\"Expected adjusted connector 'adjusted-connector', got: %s\", res.Connector)\n\t\t}\n\t\tif res.Locale != \"zh-cn\" {\n\t\t\tt.Errorf(\"Expected adjusted locale 'zh-cn', got: %s\", res.Locale)\n\t\t}\n\t\tif res.Theme != \"dark\" {\n\t\t\tt.Errorf(\"Expected adjusted theme 'dark', got: %s\", res.Theme)\n\t\t}\n\t\tif res.Route != \"/adjusted/route\" {\n\t\t\tt.Errorf(\"Expected adjusted route '/adjusted/route', got: %s\", res.Route)\n\t\t}\n\n\t\t// Verify metadata\n\t\tif res.Metadata == nil {\n\t\t\tt.Fatalf(\"Expected metadata, got nil\")\n\t\t}\n\t\tif adjusted, ok := res.Metadata[\"adjusted\"].(bool); !ok || !adjusted {\n\t\t\tt.Errorf(\"Expected metadata['adjusted'] = true, got: %v\", res.Metadata[\"adjusted\"])\n\t\t}\n\n\t\t// Verify context fields were actually updated\n\t\t// Note: AssistantID is immutable and cannot be overridden\n\t\t// Note: Connector is now in Options, not in Context\n\t\tif adjustCtx.Locale != \"zh-cn\" {\n\t\t\tt.Errorf(\"Context locale not updated. Expected 'zh-cn', got: %s\", adjustCtx.Locale)\n\t\t}\n\t\tif adjustCtx.Theme != \"dark\" {\n\t\t\tt.Errorf(\"Context theme not updated. Expected 'dark', got: %s\", adjustCtx.Theme)\n\t\t}\n\t\tif adjustCtx.Route != \"/adjusted/route\" {\n\t\t\tt.Errorf(\"Context route not updated. Expected '/adjusted/route', got: %s\", adjustCtx.Route)\n\t\t}\n\t\tif adjustCtx.Metadata[\"adjusted\"] != true {\n\t\t\tt.Errorf(\"Context metadata not updated. Expected metadata['adjusted'] = true, got: %v\", adjustCtx.Metadata[\"adjusted\"])\n\t\t}\n\n\t\tt.Log(\"✓ Context fields successfully adjusted by hook\")\n\t})\n\n\t// Test scenario 10: Adjust uses configuration - tests that uses can be modified by the hook\n\tt.Run(\"AdjustUses\", func(t *testing.T) {\n\t\t// Create a fresh context for this test\n\t\tusesCtx := newTestContext(\"chat-test-uses\", \"tests.create\")\n\n\t\t// Call the hook which should adjust uses configuration\n\t\tres, _, err := agent.HookScript.Create(usesCtx, []context.Message{{Role: \"user\", Content: \"adjust_uses\"}})\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to create with adjust_uses: %s\", err.Error())\n\t\t}\n\t\tif res == nil {\n\t\t\tt.Fatalf(\"Expected non-nil response, got nil\")\n\t\t}\n\n\t\t// Verify the response contains uses configuration\n\t\tif res.Uses == nil {\n\t\t\tt.Fatalf(\"Expected uses configuration, got nil\")\n\t\t}\n\n\t\t// Verify each uses field\n\t\tif res.Uses.Vision != \"mcp:vision-server\" {\n\t\t\tt.Errorf(\"Expected vision 'mcp:vision-server', got: %s\", res.Uses.Vision)\n\t\t}\n\t\tif res.Uses.Audio != \"mcp:audio-server\" {\n\t\t\tt.Errorf(\"Expected audio 'mcp:audio-server', got: %s\", res.Uses.Audio)\n\t\t}\n\t\tif res.Uses.Search != \"agent\" {\n\t\t\tt.Errorf(\"Expected search 'agent', got: %s\", res.Uses.Search)\n\t\t}\n\t\tif res.Uses.Fetch != \"mcp:fetch-server\" {\n\t\t\tt.Errorf(\"Expected fetch 'mcp:fetch-server', got: %s\", res.Uses.Fetch)\n\t\t}\n\n\t\t// Verify metadata\n\t\tif res.Metadata == nil {\n\t\t\tt.Fatalf(\"Expected metadata, got nil\")\n\t\t}\n\t\tif usesAdjusted, ok := res.Metadata[\"uses_adjusted\"].(bool); !ok || !usesAdjusted {\n\t\t\tt.Errorf(\"Expected metadata['uses_adjusted'] = true, got: %v\", res.Metadata[\"uses_adjusted\"])\n\t\t}\n\n\t\t// Now test that BuildRequest properly applies the uses configuration\n\t\tinputMessages := []context.Message{{Role: \"user\", Content: \"test uses\"}}\n\t\t_, options, err := agent.BuildRequest(usesCtx, inputMessages, res)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to build request: %s\", err.Error())\n\t\t}\n\n\t\t// Verify that options.Uses has the values from createResponse\n\t\tif options.Uses == nil {\n\t\t\tt.Fatalf(\"Expected options.Uses to be set, got nil\")\n\t\t}\n\t\tif options.Uses.Vision != \"mcp:vision-server\" {\n\t\t\tt.Errorf(\"Expected options.Uses.Vision 'mcp:vision-server', got: %s\", options.Uses.Vision)\n\t\t}\n\t\tif options.Uses.Audio != \"mcp:audio-server\" {\n\t\t\tt.Errorf(\"Expected options.Uses.Audio 'mcp:audio-server', got: %s\", options.Uses.Audio)\n\t\t}\n\t\tif options.Uses.Search != \"agent\" {\n\t\t\tt.Errorf(\"Expected options.Uses.Search 'agent', got: %s\", options.Uses.Search)\n\t\t}\n\t\tif options.Uses.Fetch != \"mcp:fetch-server\" {\n\t\t\tt.Errorf(\"Expected options.Uses.Fetch 'mcp:fetch-server', got: %s\", options.Uses.Fetch)\n\t\t}\n\n\t\tt.Log(\"✓ Uses configuration successfully adjusted by hook and applied to options\")\n\t})\n\n\t// Test scenario 11: Adjust uses configuration with force_uses flag\n\tt.Run(\"AdjustUsesForce\", func(t *testing.T) {\n\t\t// Create a fresh context for this test\n\t\tusesCtx := newTestContext(\"chat-test-uses-force\", \"tests.create\")\n\n\t\t// Call the hook which should adjust uses configuration and set force_uses\n\t\tres, _, err := agent.HookScript.Create(usesCtx, []context.Message{{Role: \"user\", Content: \"adjust_uses_force\"}})\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to create with adjust_uses_force: %s\", err.Error())\n\t\t}\n\t\tif res == nil {\n\t\t\tt.Fatalf(\"Expected non-nil response, got nil\")\n\t\t}\n\n\t\t// Verify the response contains uses configuration\n\t\tif res.Uses == nil {\n\t\t\tt.Fatalf(\"Expected uses configuration, got nil\")\n\t\t}\n\n\t\t// Verify uses fields\n\t\tif res.Uses.Vision != \"tests.vision-helper\" {\n\t\t\tt.Errorf(\"Expected vision 'tests.vision-helper', got: %s\", res.Uses.Vision)\n\t\t}\n\t\tif res.Uses.Audio != \"mcp:audio-server\" {\n\t\t\tt.Errorf(\"Expected audio 'mcp:audio-server', got: %s\", res.Uses.Audio)\n\t\t}\n\n\t\t// Verify force_uses flag\n\t\tif res.ForceUses == nil {\n\t\t\tt.Fatalf(\"Expected force_uses to be set, got nil\")\n\t\t}\n\t\tif !*res.ForceUses {\n\t\t\tt.Errorf(\"Expected force_uses to be true, got: %v\", *res.ForceUses)\n\t\t}\n\n\t\t// Verify metadata\n\t\tif res.Metadata == nil {\n\t\t\tt.Fatalf(\"Expected metadata, got nil\")\n\t\t}\n\t\tif usesForced, ok := res.Metadata[\"uses_forced\"].(bool); !ok || !usesForced {\n\t\t\tt.Errorf(\"Expected metadata['uses_forced'] = true, got: %v\", res.Metadata[\"uses_forced\"])\n\t\t}\n\n\t\t// Now test that BuildRequest properly applies the force_uses flag\n\t\tinputMessages := []context.Message{{Role: \"user\", Content: \"test force uses\"}}\n\t\t_, options, err := agent.BuildRequest(usesCtx, inputMessages, res)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to build request: %s\", err.Error())\n\t\t}\n\n\t\t// Verify that options.ForceUses is true\n\t\tif !options.ForceUses {\n\t\t\tt.Errorf(\"Expected options.ForceUses to be true, got: %v\", options.ForceUses)\n\t\t}\n\n\t\tt.Log(\"✓ Uses configuration with force_uses flag successfully adjusted by hook and applied to options\")\n\t})\n}\n" }, { "path": "agent/assistant/hook/goroutine_leak_test.go", "content": "package hook_test\n\nimport (\n\tstdContext \"context\"\n\t\"fmt\"\n\t\"os\"\n\t\"runtime\"\n\t\"runtime/pprof\"\n\t\"strings\"\n\t\"testing\"\n\t\"time\"\n\n\t\"github.com/yaoapp/yao/agent/assistant\"\n\t\"github.com/yaoapp/yao/agent/context\"\n\t\"github.com/yaoapp/yao/agent/testutils\"\n\t\"github.com/yaoapp/yao/openapi/oauth/types\"\n)\n\n// TestGoroutineLeakDetailed performs detailed goroutine leak analysis\nfunc TestGoroutineLeakDetailed(t *testing.T) {\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\tagent, err := assistant.Get(\"tests.create\")\n\tif err != nil {\n\t\tt.Fatalf(\"Failed to get assistant: %s\", err.Error())\n\t}\n\n\tif agent.HookScript == nil {\n\t\tt.Fatalf(\"Assistant has no script\")\n\t}\n\n\t// Create profile directory\n\tos.MkdirAll(\"/tmp/goroutine_profiles\", 0755)\n\n\t// Take initial snapshot\n\truntime.GC()\n\ttime.Sleep(200 * time.Millisecond)\n\tinitialGoroutines := runtime.NumGoroutine()\n\n\t// Save initial profile\n\tsaveGoroutineProfile(\"/tmp/goroutine_profiles/00_initial.txt\")\n\tt.Logf(\"Initial goroutines: %d\", initialGoroutines)\n\n\t// Test with just 10 iterations to see the pattern\n\titerations := 10\n\tfor i := 0; i < iterations; i++ {\n\t\tctx := newLeakTestContext(fmt.Sprintf(\"leak-test-%d\", i), \"tests.create\")\n\n\t\t_, _, err := agent.HookScript.Create(ctx, []context.Message{\n\t\t\t{Role: \"user\", Content: \"Hello\"},\n\t\t})\n\t\tif err != nil {\n\t\t\tt.Errorf(\"Create failed at iteration %d: %s\", i, err.Error())\n\t\t}\n\n\t\t// Release context\n\t\tctx.Release()\n\n\t\t// Check goroutines after each iteration\n\t\tcurrent := runtime.NumGoroutine()\n\t\tgrowth := current - initialGoroutines\n\t\tt.Logf(\"After iteration %d: %d goroutines (growth: %d)\", i+1, current, growth)\n\n\t\t// Save profile every 5 iterations\n\t\tif (i+1)%5 == 0 {\n\t\t\tsaveGoroutineProfile(fmt.Sprintf(\"/tmp/goroutine_profiles/%02d_after_iter_%d.txt\", i+1, i+1))\n\t\t}\n\t}\n\n\t// Force cleanup\n\truntime.GC()\n\ttime.Sleep(500 * time.Millisecond)\n\n\tfinalGoroutines := runtime.NumGoroutine()\n\tgrowth := finalGoroutines - initialGoroutines\n\n\tt.Logf(\"\\n=== SUMMARY ===\")\n\tt.Logf(\"Initial: %d goroutines\", initialGoroutines)\n\tt.Logf(\"Final: %d goroutines\", finalGoroutines)\n\tt.Logf(\"Growth: %d goroutines (%.2f per iteration)\", growth, float64(growth)/float64(iterations))\n\n\t// Save final profile\n\tsaveGoroutineProfile(\"/tmp/goroutine_profiles/99_final.txt\")\n\n\t// Analyze the leak\n\tt.Logf(\"\\n=== ANALYSIS ===\")\n\tanalyzeGoroutineProfiles(t, \"/tmp/goroutine_profiles\")\n}\n\n// TestGoroutineLeakByComponent tests each component separately\nfunc TestGoroutineLeakByComponent(t *testing.T) {\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\tagent, err := assistant.Get(\"tests.create\")\n\tif err != nil {\n\t\tt.Fatalf(\"Failed to get assistant: %s\", err.Error())\n\t}\n\n\tos.MkdirAll(\"/tmp/component_profiles\", 0755)\n\n\tt.Run(\"ContextCreationOnly\", func(t *testing.T) {\n\t\truntime.GC()\n\t\ttime.Sleep(100 * time.Millisecond)\n\t\tinitial := runtime.NumGoroutine()\n\n\t\tfor i := 0; i < 10; i++ {\n\t\t\tctx := newLeakTestContext(fmt.Sprintf(\"test-%d\", i), \"tests.create\")\n\t\t\t_ = ctx\n\t\t\tctx.Release()\n\t\t}\n\n\t\truntime.GC()\n\t\ttime.Sleep(100 * time.Millisecond)\n\t\tfinal := runtime.NumGoroutine()\n\n\t\tt.Logf(\"Context creation: initial=%d, final=%d, growth=%d\", initial, final, final-initial)\n\t})\n\n\tt.Run(\"ScriptExecutionOnly\", func(t *testing.T) {\n\t\truntime.GC()\n\t\ttime.Sleep(100 * time.Millisecond)\n\t\tinitial := runtime.NumGoroutine()\n\n\t\tfor i := 0; i < 10; i++ {\n\t\t\tctx := newLeakTestContext(fmt.Sprintf(\"test-%d\", i), \"tests.create\")\n\t\t\t_, _, _ = agent.HookScript.Create(ctx, []context.Message{\n\t\t\t\t{Role: \"user\", Content: \"Hello\"},\n\t\t\t})\n\t\t\tctx.Release()\n\t\t}\n\n\t\truntime.GC()\n\t\ttime.Sleep(100 * time.Millisecond)\n\t\tfinal := runtime.NumGoroutine()\n\n\t\tt.Logf(\"Script execution: initial=%d, final=%d, growth=%d\", initial, final, final-initial)\n\t\tsaveGoroutineProfile(\"/tmp/component_profiles/script_execution.txt\")\n\t})\n\n\tt.Run(\"TraceOperations\", func(t *testing.T) {\n\t\truntime.GC()\n\t\ttime.Sleep(100 * time.Millisecond)\n\t\tinitial := runtime.NumGoroutine()\n\n\t\tfor i := 0; i < 10; i++ {\n\t\t\tctx := newLeakTestContext(fmt.Sprintf(\"test-%d\", i), \"tests.create\")\n\n\t\t\t// Create trace\n\t\t\ttrace, err := ctx.Trace()\n\t\t\tif err == nil && trace != nil {\n\t\t\t\t// Trace operations\n\t\t\t\t_ = trace\n\t\t\t}\n\n\t\t\tctx.Release()\n\t\t}\n\n\t\truntime.GC()\n\t\ttime.Sleep(100 * time.Millisecond)\n\t\tfinal := runtime.NumGoroutine()\n\n\t\tt.Logf(\"Trace operations: initial=%d, final=%d, growth=%d\", initial, final, final-initial)\n\t\tsaveGoroutineProfile(\"/tmp/component_profiles/trace_operations.txt\")\n\t})\n}\n\n// TestGoroutineLeakWithoutRelease tests if Release() fixes the leak\nfunc TestGoroutineLeakWithoutRelease(t *testing.T) {\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\tagent, err := assistant.Get(\"tests.create\")\n\tif err != nil {\n\t\tt.Fatalf(\"Failed to get assistant: %s\", err.Error())\n\t}\n\n\tt.Run(\"WithoutRelease\", func(t *testing.T) {\n\t\truntime.GC()\n\t\ttime.Sleep(100 * time.Millisecond)\n\t\tinitial := runtime.NumGoroutine()\n\n\t\tfor i := 0; i < 10; i++ {\n\t\t\tctx := newLeakTestContext(fmt.Sprintf(\"no-release-%d\", i), \"tests.create\")\n\t\t\t_, _, _ = agent.HookScript.Create(ctx, []context.Message{\n\t\t\t\t{Role: \"user\", Content: \"Hello\"},\n\t\t\t})\n\t\t\t// Intentionally NOT calling ctx.Release()\n\t\t}\n\n\t\truntime.GC()\n\t\ttime.Sleep(100 * time.Millisecond)\n\t\tfinal := runtime.NumGoroutine()\n\n\t\tt.Logf(\"WITHOUT Release: initial=%d, final=%d, growth=%d (%.1f per iter)\",\n\t\t\tinitial, final, final-initial, float64(final-initial)/10.0)\n\t})\n\n\tt.Run(\"WithRelease\", func(t *testing.T) {\n\t\truntime.GC()\n\t\ttime.Sleep(100 * time.Millisecond)\n\t\tinitial := runtime.NumGoroutine()\n\n\t\tfor i := 0; i < 10; i++ {\n\t\t\tctx := newLeakTestContext(fmt.Sprintf(\"with-release-%d\", i), \"tests.create\")\n\t\t\t_, _, _ = agent.HookScript.Create(ctx, []context.Message{\n\t\t\t\t{Role: \"user\", Content: \"Hello\"},\n\t\t\t})\n\t\t\tctx.Release() // WITH Release\n\t\t}\n\n\t\truntime.GC()\n\t\ttime.Sleep(100 * time.Millisecond)\n\t\tfinal := runtime.NumGoroutine()\n\n\t\tt.Logf(\"WITH Release: initial=%d, final=%d, growth=%d (%.1f per iter)\",\n\t\t\tinitial, final, final-initial, float64(final-initial)/10.0)\n\t})\n}\n\n// Helper functions\n\nfunc saveGoroutineProfile(filename string) {\n\tf, err := os.Create(filename)\n\tif err != nil {\n\t\treturn\n\t}\n\tdefer f.Close()\n\n\tpprof.Lookup(\"goroutine\").WriteTo(f, 2) // detail level 2\n}\n\nfunc analyzeGoroutineProfiles(t *testing.T, dir string) {\n\t// Read initial and final profiles\n\tinitialData, err := os.ReadFile(dir + \"/00_initial.txt\")\n\tif err != nil {\n\t\tt.Logf(\"Could not read initial profile: %v\", err)\n\t\treturn\n\t}\n\n\tfinalData, err := os.ReadFile(dir + \"/99_final.txt\")\n\tif err != nil {\n\t\tt.Logf(\"Could not read final profile: %v\", err)\n\t\treturn\n\t}\n\n\t// Count goroutines by function\n\tinitialFuncs := countGoroutinesByFunction(string(initialData))\n\tfinalFuncs := countGoroutinesByFunction(string(finalData))\n\n\tt.Logf(\"\\nGoroutine growth by function:\")\n\tt.Logf(\"%-60s %8s %8s %8s\", \"Function\", \"Initial\", \"Final\", \"Growth\")\n\tt.Logf(\"%s\", strings.Repeat(\"-\", 90))\n\n\t// Find functions that grew\n\tfor fn, finalCount := range finalFuncs {\n\t\tinitialCount := initialFuncs[fn]\n\t\tgrowth := finalCount - initialCount\n\t\tif growth > 0 {\n\t\t\tt.Logf(\"%-60s %8d %8d %8d\", truncate(fn, 60), initialCount, finalCount, growth)\n\t\t}\n\t}\n\n\tt.Logf(\"\\nProfiles saved to: %s\", dir)\n\tt.Logf(\"To compare: diff %s/00_initial.txt %s/99_final.txt | grep '^>'\", dir, dir)\n}\n\nfunc countGoroutinesByFunction(profile string) map[string]int {\n\tcounts := make(map[string]int)\n\tlines := strings.Split(profile, \"\\n\")\n\n\tfor _, line := range lines {\n\t\tline = strings.TrimSpace(line)\n\t\t// Look for function names in goroutine stack traces\n\t\tif strings.Contains(line, \"(\") && !strings.HasPrefix(line, \"#\") {\n\t\t\t// Extract function name\n\t\t\tif idx := strings.Index(line, \"(\"); idx > 0 {\n\t\t\t\tfn := strings.TrimSpace(line[:idx])\n\t\t\t\tcounts[fn]++\n\t\t\t}\n\t\t}\n\t}\n\n\treturn counts\n}\n\nfunc truncate(s string, max int) string {\n\tif len(s) <= max {\n\t\treturn s\n\t}\n\treturn s[:max-3] + \"...\"\n}\n\nfunc newLeakTestContext(chatID, assistantID string) *context.Context {\n\tauthorized := &types.AuthorizedInfo{\n\t\tSubject: \"leak-test-user\",\n\t\tClientID: \"leak-test-client\",\n\t\tUserID: \"leak-user-123\",\n\t\tTeamID: \"leak-team-456\",\n\t\tTenantID: \"leak-tenant-789\",\n\t\tConstraints: types.DataConstraints{\n\t\t\tTeamOnly: true,\n\t\t\tExtra: map[string]interface{}{\n\t\t\t\t\"department\": \"testing\",\n\t\t\t},\n\t\t},\n\t}\n\n\tctx := context.New(stdContext.Background(), authorized, chatID)\n\tctx.AssistantID = assistantID\n\tctx.Locale = \"en-us\"\n\tctx.Theme = \"light\"\n\tctx.Client = context.Client{\n\t\tType: \"web\",\n\t\tUserAgent: \"LeakTestAgent/1.0\",\n\t\tIP: \"127.0.0.1\",\n\t}\n\tctx.Referer = context.RefererAPI\n\tctx.Accept = context.AcceptWebCUI\n\tctx.Route = \"\"\n\tctx.Metadata = make(map[string]interface{})\n\treturn ctx\n}\n" }, { "path": "agent/assistant/hook/hook.go", "content": "package hook\n" }, { "path": "agent/assistant/hook/next.go", "content": "package hook\n\nimport (\n\t\"encoding/json\"\n\t\"fmt\"\n\n\t\"github.com/yaoapp/gou/runtime/v8/bridge\"\n\t\"github.com/yaoapp/yao/agent/context\"\n)\n\n// Next next hook for the next action after the completion\n// opts is optional - if provided, will be passed to the hook\nfunc (s *Script) Next(ctx *context.Context, payload *context.NextHookPayload, opts ...*context.Options) (*context.NextHookResponse, *context.Options, error) {\n\t// Get or create options\n\tvar options *context.Options\n\tif len(opts) > 0 && opts[0] != nil {\n\t\toptions = opts[0]\n\t} else {\n\t\toptions = &context.Options{}\n\t}\n\n\t// Convert payload to map for JS (use JSON tag names)\n\tpayloadMap := map[string]interface{}{\n\t\t\"messages\": payload.Messages,\n\t\t\"completion\": payload.Completion,\n\t\t\"tools\": payload.Tools,\n\t\t\"error\": payload.Error,\n\t}\n\n\t// Execute hook with ctx, payload, and options (convert options to map for JS)\n\toptionsMap := options.ToMap()\n\tres, err := s.Execute(ctx, \"Next\", payloadMap, optionsMap)\n\tif err != nil {\n\t\treturn nil, nil, err\n\t}\n\n\tresponse, err := s.getNextHookResponse(res)\n\tif err != nil {\n\t\treturn nil, nil, err\n\t}\n\n\treturn response, options, nil\n}\n\n// getNextHookResponse convert the result to a NextHookResponse\nfunc (s *Script) getNextHookResponse(res interface{}) (*context.NextHookResponse, error) {\n\t// Handle nil result\n\tif res == nil {\n\t\treturn nil, nil\n\t}\n\n\t// Handle undefined result (treat as nil)\n\tif _, ok := res.(bridge.UndefinedT); ok {\n\t\treturn nil, nil\n\t}\n\n\t// Marshal to JSON and unmarshal to NextHookResponse\n\traw, err := json.Marshal(res)\n\tif err != nil {\n\t\treturn nil, fmt.Errorf(\"failed to marshal Next hook result: %w\", err)\n\t}\n\n\tvar response context.NextHookResponse\n\tif err := json.Unmarshal(raw, &response); err != nil {\n\t\treturn nil, fmt.Errorf(\"failed to unmarshal to NextHookResponse: %w\", err)\n\t}\n\n\treturn &response, nil\n}\n" }, { "path": "agent/assistant/hook/next_test.go", "content": "package hook_test\n\nimport (\n\tstdContext \"context\"\n\t\"testing\"\n\n\t\"github.com/yaoapp/yao/agent/assistant\"\n\t\"github.com/yaoapp/yao/agent/context\"\n\t\"github.com/yaoapp/yao/agent/output/message\"\n\t\"github.com/yaoapp/yao/agent/testutils\"\n\t\"github.com/yaoapp/yao/openapi/oauth/types\"\n)\n\n// newTestContextForNext creates a Context for testing Next Hook with commonly used fields pre-populated.\n// You can override any fields after creation as needed for specific test scenarios.\nfunc newTestContextForNext(chatID, assistantID string) *context.Context {\n\tauthorized := &types.AuthorizedInfo{\n\t\tSubject: \"test-user\",\n\t\tClientID: \"test-client-id\",\n\t\tScope: \"openid profile email\",\n\t\tSessionID: \"test-session-id\",\n\t\tUserID: \"test-user-123\",\n\t\tTeamID: \"test-team-456\",\n\t\tTenantID: \"test-tenant-789\",\n\t\tRememberMe: true,\n\t\tConstraints: types.DataConstraints{\n\t\t\tOwnerOnly: false,\n\t\t\tCreatorOnly: false,\n\t\t\tEditorOnly: false,\n\t\t\tTeamOnly: true,\n\t\t\tExtra: map[string]interface{}{\n\t\t\t\t\"department\": \"engineering\",\n\t\t\t\t\"region\": \"us-west\",\n\t\t\t\t\"project\": \"yao\",\n\t\t\t},\n\t\t},\n\t}\n\n\tctx := context.New(stdContext.Background(), authorized, chatID)\n\tctx.AssistantID = assistantID\n\tctx.Locale = \"en-us\"\n\tctx.Theme = \"light\"\n\tctx.Client = context.Client{\n\t\tType: \"web\",\n\t\tUserAgent: \"TestAgent/1.0\",\n\t\tIP: \"127.0.0.1\",\n\t}\n\tctx.Referer = context.RefererAPI\n\tctx.Accept = context.AcceptWebCUI\n\tctx.Route = \"\"\n\tctx.Metadata = make(map[string]interface{})\n\treturn ctx\n}\n\n// TestNext tests the Next hook\nfunc TestNext(t *testing.T) {\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\tagent, err := assistant.Get(\"tests.next\")\n\tif err != nil {\n\t\tt.Fatalf(\"Failed to get the tests.next assistant: %s\", err.Error())\n\t}\n\n\tif agent.HookScript == nil {\n\t\tt.Fatalf(\"The tests.next assistant has no script\")\n\t}\n\n\t// Use the helper function to create a test context\n\tctx := newTestContextForNext(\"chat-test-next-hook\", \"tests.next\")\n\n\t// Test scenario 1: Return null (should get nil response)\n\tt.Run(\"ReturnNull\", func(t *testing.T) {\n\t\tpayload := &context.NextHookPayload{\n\t\t\tMessages: []context.Message{\n\t\t\t\t{Role: context.RoleUser, Content: \"return_null\"},\n\t\t\t},\n\t\t\tCompletion: &context.CompletionResponse{\n\t\t\t\tContent: \"Test completion\",\n\t\t\t},\n\t\t\tTools: nil,\n\t\t\tError: \"\",\n\t\t}\n\n\t\tres, _, err := agent.HookScript.Next(ctx, payload)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to execute Next hook with null return: %s\", err.Error())\n\t\t}\n\t\tif res != nil {\n\t\t\tt.Errorf(\"Expected nil response for null return, got: %v\", res)\n\t\t}\n\t})\n\n\t// Test scenario 2: Return undefined (should get nil response)\n\tt.Run(\"ReturnUndefined\", func(t *testing.T) {\n\t\tpayload := &context.NextHookPayload{\n\t\t\tMessages: []context.Message{\n\t\t\t\t{Role: context.RoleUser, Content: \"return_undefined\"},\n\t\t\t},\n\t\t\tCompletion: &context.CompletionResponse{\n\t\t\t\tContent: \"Test completion\",\n\t\t\t},\n\t\t}\n\n\t\tres, _, err := agent.HookScript.Next(ctx, payload)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to execute Next hook with undefined return: %s\", err.Error())\n\t\t}\n\t\tif res != nil {\n\t\t\tt.Errorf(\"Expected nil response for undefined return, got: %v\", res)\n\t\t}\n\t})\n\n\t// Test scenario 3: Return empty object (should get empty NextHookResponse)\n\tt.Run(\"ReturnEmpty\", func(t *testing.T) {\n\t\tpayload := &context.NextHookPayload{\n\t\t\tMessages: []context.Message{\n\t\t\t\t{Role: context.RoleUser, Content: \"return_empty\"},\n\t\t\t},\n\t\t\tCompletion: &context.CompletionResponse{\n\t\t\t\tContent: \"Test completion\",\n\t\t\t},\n\t\t}\n\n\t\tres, _, err := agent.HookScript.Next(ctx, payload)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to execute Next hook with empty return: %s\", err.Error())\n\t\t}\n\t\tif res == nil {\n\t\t\tt.Fatalf(\"Expected non-nil response for empty object, got nil\")\n\t\t}\n\t\tif res.Delegate != nil {\n\t\t\tt.Errorf(\"Expected nil Delegate, got: %v\", res.Delegate)\n\t\t}\n\t\tif res.Data != nil {\n\t\t\tt.Errorf(\"Expected nil Data, got: %v\", res.Data)\n\t\t}\n\t})\n\n\t// Test scenario 4: Return custom data\n\tt.Run(\"ReturnCustomData\", func(t *testing.T) {\n\t\tpayload := &context.NextHookPayload{\n\t\t\tMessages: []context.Message{\n\t\t\t\t{Role: context.RoleUser, Content: \"return_custom_data\"},\n\t\t\t},\n\t\t\tCompletion: &context.CompletionResponse{\n\t\t\t\tContent: \"Test completion\",\n\t\t\t},\n\t\t}\n\n\t\tres, _, err := agent.HookScript.Next(ctx, payload)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to execute Next hook with custom data: %s\", err.Error())\n\t\t}\n\t\tif res == nil {\n\t\t\tt.Fatalf(\"Expected non-nil response, got nil\")\n\t\t}\n\n\t\t// Verify Data is present\n\t\tif res.Data == nil {\n\t\t\tt.Fatalf(\"Expected Data to be present, got nil\")\n\t\t}\n\n\t\t// Data should be a map\n\t\tdataMap, ok := res.Data.(map[string]interface{})\n\t\tif !ok {\n\t\t\tt.Fatalf(\"Expected Data to be map[string]interface{}, got: %T\", res.Data)\n\t\t}\n\n\t\t// Verify custom data fields\n\t\tif message, ok := dataMap[\"message\"].(string); !ok || message != \"Custom response from Next Hook\" {\n\t\t\tt.Errorf(\"Expected custom message, got: %v\", dataMap[\"message\"])\n\t\t}\n\t\tif test, ok := dataMap[\"test\"].(bool); !ok || !test {\n\t\t\tt.Errorf(\"Expected test=true, got: %v\", dataMap[\"test\"])\n\t\t}\n\t\tif _, ok := dataMap[\"timestamp\"]; !ok {\n\t\t\tt.Errorf(\"Expected timestamp field\")\n\t\t}\n\n\t\t// Verify Delegate is nil\n\t\tif res.Delegate != nil {\n\t\t\tt.Errorf(\"Expected nil Delegate, got: %v\", res.Delegate)\n\t\t}\n\t})\n\n\t// Test scenario 5: Return data with metadata\n\tt.Run(\"ReturnDataWithMetadata\", func(t *testing.T) {\n\t\tpayload := &context.NextHookPayload{\n\t\t\tMessages: []context.Message{\n\t\t\t\t{Role: context.RoleUser, Content: \"return_data_with_metadata\"},\n\t\t\t},\n\t\t\tCompletion: &context.CompletionResponse{\n\t\t\t\tContent: \"Test completion\",\n\t\t\t},\n\t\t}\n\n\t\tres, _, err := agent.HookScript.Next(ctx, payload)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to execute Next hook: %s\", err.Error())\n\t\t}\n\t\tif res == nil {\n\t\t\tt.Fatalf(\"Expected non-nil response, got nil\")\n\t\t}\n\n\t\t// Verify Data\n\t\tif res.Data == nil {\n\t\t\tt.Fatalf(\"Expected Data to be present, got nil\")\n\t\t}\n\n\t\tdataMap, ok := res.Data.(map[string]interface{})\n\t\tif !ok {\n\t\t\tt.Fatalf(\"Expected Data to be map[string]interface{}, got: %T\", res.Data)\n\t\t}\n\n\t\tif result, ok := dataMap[\"result\"].(string); !ok || result != \"success\" {\n\t\t\tt.Errorf(\"Expected result='success', got: %v\", dataMap[\"result\"])\n\t\t}\n\n\t\t// Verify Metadata\n\t\tif res.Metadata == nil {\n\t\t\tt.Fatalf(\"Expected Metadata to be present, got nil\")\n\t\t}\n\n\t\tif hook, ok := res.Metadata[\"hook\"].(string); !ok || hook != \"next\" {\n\t\t\tt.Errorf(\"Expected hook='next', got: %v\", res.Metadata[\"hook\"])\n\t\t}\n\t\tif processed, ok := res.Metadata[\"processed\"].(bool); !ok || !processed {\n\t\t\tt.Errorf(\"Expected processed=true, got: %v\", res.Metadata[\"processed\"])\n\t\t}\n\t})\n\n\t// Test scenario 6: Return delegate\n\tt.Run(\"ReturnDelegate\", func(t *testing.T) {\n\t\tpayload := &context.NextHookPayload{\n\t\t\tMessages: []context.Message{\n\t\t\t\t{Role: context.RoleUser, Content: \"return_delegate\"},\n\t\t\t},\n\t\t\tCompletion: &context.CompletionResponse{\n\t\t\t\tContent: \"Test completion\",\n\t\t\t},\n\t\t}\n\n\t\tres, _, err := agent.HookScript.Next(ctx, payload)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to execute Next hook with delegate: %s\", err.Error())\n\t\t}\n\t\tif res == nil {\n\t\t\tt.Fatalf(\"Expected non-nil response, got nil\")\n\t\t}\n\n\t\t// Verify Delegate is present\n\t\tif res.Delegate == nil {\n\t\t\tt.Fatalf(\"Expected Delegate to be present, got nil\")\n\t\t}\n\n\t\t// Verify delegate fields\n\t\tif res.Delegate.AgentID != \"tests.create\" {\n\t\t\tt.Errorf(\"Expected AgentID='tests.create', got: %s\", res.Delegate.AgentID)\n\t\t}\n\n\t\tif len(res.Delegate.Messages) != 1 {\n\t\t\tt.Errorf(\"Expected 1 message, got: %d\", len(res.Delegate.Messages))\n\t\t} else {\n\t\t\tif res.Delegate.Messages[0].Role != context.RoleUser {\n\t\t\t\tt.Errorf(\"Expected user role, got: %s\", res.Delegate.Messages[0].Role)\n\t\t\t}\n\t\t\tif content, ok := res.Delegate.Messages[0].Content.(string); !ok || content != \"Hello from delegated agent\" {\n\t\t\t\tt.Errorf(\"Expected specific content, got: %v\", res.Delegate.Messages[0].Content)\n\t\t\t}\n\t\t}\n\n\t\t// Verify Data is nil (only delegate, no custom data)\n\t\tif res.Data != nil {\n\t\t\tt.Logf(\"Note: Data is present alongside Delegate: %v\", res.Data)\n\t\t}\n\t})\n\n\t// Test scenario 7: Verify payload structure\n\tt.Run(\"VerifyPayload\", func(t *testing.T) {\n\t\tpayload := &context.NextHookPayload{\n\t\t\tMessages: []context.Message{\n\t\t\t\t{Role: context.RoleSystem, Content: \"System message\"},\n\t\t\t\t{Role: context.RoleUser, Content: \"verify_payload\"},\n\t\t\t},\n\t\t\tCompletion: &context.CompletionResponse{\n\t\t\t\tContent: \"Test completion content\",\n\t\t\t\tUsage: &message.UsageInfo{\n\t\t\t\t\tPromptTokens: 10,\n\t\t\t\t\tCompletionTokens: 20,\n\t\t\t\t\tTotalTokens: 30,\n\t\t\t\t},\n\t\t\t},\n\t\t\tTools: []context.ToolCallResponse{\n\t\t\t\t{\n\t\t\t\t\tToolCallID: \"call_123\",\n\t\t\t\t\tServer: \"test-server\",\n\t\t\t\t\tTool: \"test-tool\",\n\t\t\t\t\tResult: map[string]interface{}{\"success\": true},\n\t\t\t\t\tError: \"\",\n\t\t\t\t},\n\t\t\t},\n\t\t\tError: \"\",\n\t\t}\n\n\t\tres, _, err := agent.HookScript.Next(ctx, payload)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to execute Next hook: %s\", err.Error())\n\t\t}\n\t\tif res == nil {\n\t\t\tt.Fatalf(\"Expected non-nil response, got nil\")\n\t\t}\n\n\t\t// Verify Data contains validation results\n\t\tif res.Data == nil {\n\t\t\tt.Fatalf(\"Expected Data with validation results, got nil\")\n\t\t}\n\n\t\tdataMap, ok := res.Data.(map[string]interface{})\n\t\tif !ok {\n\t\t\tt.Fatalf(\"Expected Data to be map[string]interface{}, got: %T\", res.Data)\n\t\t}\n\n\t\tif validation, ok := dataMap[\"validation\"].(string); !ok || validation != \"success\" {\n\t\t\tt.Errorf(\"Expected validation='success', got: %v\", dataMap[\"validation\"])\n\t\t}\n\n\t\tif checks, ok := dataMap[\"checks\"].([]interface{}); !ok {\n\t\t\tt.Errorf(\"Expected checks array, got: %T\", dataMap[\"checks\"])\n\t\t} else {\n\t\t\tt.Logf(\"✓ Payload validation checks: %d items\", len(checks))\n\t\t\tfor i, check := range checks {\n\t\t\t\tt.Logf(\" [%d] %v\", i, check)\n\t\t\t}\n\t\t}\n\t})\n\n\t// Test scenario 8: Verify tools processing\n\tt.Run(\"VerifyTools\", func(t *testing.T) {\n\t\tpayload := &context.NextHookPayload{\n\t\t\tMessages: []context.Message{\n\t\t\t\t{Role: context.RoleUser, Content: \"verify_tools\"},\n\t\t\t},\n\t\t\tCompletion: &context.CompletionResponse{\n\t\t\t\tContent: \"Test completion\",\n\t\t\t},\n\t\t\tTools: []context.ToolCallResponse{\n\t\t\t\t{\n\t\t\t\t\tToolCallID: \"call_1\",\n\t\t\t\t\tServer: \"server1\",\n\t\t\t\t\tTool: \"tool1\",\n\t\t\t\t\tResult: map[string]interface{}{\"value\": 42},\n\t\t\t\t\tError: \"\",\n\t\t\t\t},\n\t\t\t\t{\n\t\t\t\t\tToolCallID: \"call_2\",\n\t\t\t\t\tServer: \"server2\",\n\t\t\t\t\tTool: \"tool2\",\n\t\t\t\t\tResult: nil,\n\t\t\t\t\tError: \"Tool execution failed\",\n\t\t\t\t},\n\t\t\t},\n\t\t}\n\n\t\tres, _, err := agent.HookScript.Next(ctx, payload)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to execute Next hook: %s\", err.Error())\n\t\t}\n\t\tif res == nil {\n\t\t\tt.Fatalf(\"Expected non-nil response, got nil\")\n\t\t}\n\n\t\t// Verify Data\n\t\tif res.Data == nil {\n\t\t\tt.Fatalf(\"Expected Data, got nil\")\n\t\t}\n\n\t\tdataMap, ok := res.Data.(map[string]interface{})\n\t\tif !ok {\n\t\t\tt.Fatalf(\"Expected Data to be map, got: %T\", res.Data)\n\t\t}\n\n\t\t// Verify tool statistics\n\t\tif totalTools, ok := dataMap[\"total_tools\"].(float64); !ok || int(totalTools) != 2 {\n\t\t\tt.Errorf(\"Expected total_tools=2, got: %v\", dataMap[\"total_tools\"])\n\t\t}\n\t\tif successful, ok := dataMap[\"successful\"].(float64); !ok || int(successful) != 1 {\n\t\t\tt.Errorf(\"Expected successful=1, got: %v\", dataMap[\"successful\"])\n\t\t}\n\t\tif failed, ok := dataMap[\"failed\"].(float64); !ok || int(failed) != 1 {\n\t\t\tt.Errorf(\"Expected failed=1, got: %v\", dataMap[\"failed\"])\n\t\t}\n\n\t\tt.Log(\"✓ Tools processing validated successfully\")\n\t})\n\n\t// Test scenario 9: Handle error\n\tt.Run(\"HandleError\", func(t *testing.T) {\n\t\tpayload := &context.NextHookPayload{\n\t\t\tMessages: []context.Message{\n\t\t\t\t{Role: context.RoleUser, Content: \"handle_error\"},\n\t\t\t},\n\t\t\tCompletion: &context.CompletionResponse{\n\t\t\t\tContent: \"Test completion\",\n\t\t\t},\n\t\t\tError: \"Tool execution failed: timeout\",\n\t\t}\n\n\t\tres, _, err := agent.HookScript.Next(ctx, payload)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to execute Next hook: %s\", err.Error())\n\t\t}\n\t\tif res == nil {\n\t\t\tt.Fatalf(\"Expected non-nil response, got nil\")\n\t\t}\n\n\t\t// Verify error handling\n\t\tif res.Data == nil {\n\t\t\tt.Fatalf(\"Expected Data, got nil\")\n\t\t}\n\n\t\tdataMap, ok := res.Data.(map[string]interface{})\n\t\tif !ok {\n\t\t\tt.Fatalf(\"Expected Data to be map, got: %T\", res.Data)\n\t\t}\n\n\t\tif errorMsg, ok := dataMap[\"error\"].(string); !ok || errorMsg != \"Tool execution failed: timeout\" {\n\t\t\tt.Errorf(\"Expected error message, got: %v\", dataMap[\"error\"])\n\t\t}\n\t\tif recovered, ok := dataMap[\"recovered\"].(bool); !ok || !recovered {\n\t\t\tt.Errorf(\"Expected recovered=true, got: %v\", dataMap[\"recovered\"])\n\t\t}\n\n\t\tt.Log(\"✓ Error handling validated successfully\")\n\t})\n}\n" }, { "path": "agent/assistant/hook/realworld_next_test.go", "content": "package hook_test\n\nimport (\n\tstdContext \"context\"\n\t\"testing\"\n\n\t\"github.com/stretchr/testify/assert\"\n\t\"github.com/yaoapp/yao/agent/assistant\"\n\t\"github.com/yaoapp/yao/agent/context\"\n\t\"github.com/yaoapp/yao/agent/testutils\"\n\t\"github.com/yaoapp/yao/openapi/oauth/types\"\n)\n\n// newRealWorldNextContext creates a Context for real world Next Hook testing\nfunc newRealWorldNextContext(chatID, assistantID string) *context.Context {\n\tauthorized := &types.AuthorizedInfo{\n\t\tSubject: \"realworld-test-user\",\n\t\tClientID: \"realworld-test-client\",\n\t\tScope: \"openid profile\",\n\t\tSessionID: \"realworld-test-session\",\n\t\tUserID: \"realworld-user-123\",\n\t\tTeamID: \"realworld-team-456\",\n\t\tTenantID: \"realworld-tenant-789\",\n\t}\n\n\tctx := context.New(stdContext.Background(), authorized, chatID)\n\tctx.AssistantID = assistantID\n\tctx.Locale = \"en-us\"\n\tctx.Theme = \"light\"\n\tctx.Client = context.Client{\n\t\tType: \"web\",\n\t\tUserAgent: \"RealWorldTest/1.0\",\n\t\tIP: \"127.0.0.1\",\n\t}\n\tctx.Referer = context.RefererAPI\n\tctx.Accept = context.AcceptWebCUI\n\tctx.Route = \"\"\n\tctx.Metadata = make(map[string]interface{})\n\treturn ctx\n}\n\n// TestRealWorldNextStandard tests standard response (nil return)\nfunc TestRealWorldNextStandard(t *testing.T) {\n\tif testing.Short() {\n\t\tt.Skip(\"Skipping real world Next Hook test in short mode\")\n\t}\n\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\tagent, err := assistant.Get(\"tests.realworld-next\")\n\tif err != nil {\n\t\tt.Fatalf(\"Failed to get assistant: %v\", err)\n\t}\n\n\tctx := newRealWorldNextContext(\"test-next-standard\", \"tests.realworld-next\")\n\n\t// Simulate completion with scenario marker\n\tmessages := []context.Message{\n\t\t{Role: context.RoleUser, Content: \"scenario: standard\"},\n\t\t{Role: context.RoleAssistant, Content: \"I'll process your request using standard response.\"},\n\t}\n\n\tcompletion := &context.CompletionResponse{\n\t\tContent: \"Processing complete. Standard response will be used.\",\n\t}\n\n\tpayload := &context.NextHookPayload{\n\t\tMessages: messages,\n\t\tCompletion: completion,\n\t\tTools: nil,\n\t\tError: \"\",\n\t}\n\n\tresponse, _, err := agent.HookScript.Next(ctx, payload)\n\tif err != nil {\n\t\tt.Fatalf(\"Next hook failed: %v\", err)\n\t}\n\n\t// Should return nil for standard response\n\tassert.Nil(t, response, \"Standard scenario should return nil\")\n\tt.Log(\"✓ Standard response scenario passed\")\n}\n\n// TestRealWorldNextCustomData tests custom data response\nfunc TestRealWorldNextCustomData(t *testing.T) {\n\tif testing.Short() {\n\t\tt.Skip(\"Skipping real world Next Hook test in short mode\")\n\t}\n\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\tagent, err := assistant.Get(\"tests.realworld-next\")\n\tif err != nil {\n\t\tt.Fatalf(\"Failed to get assistant: %v\", err)\n\t}\n\n\tctx := newRealWorldNextContext(\"test-next-custom\", \"tests.realworld-next\")\n\n\tmessages := []context.Message{\n\t\t{Role: context.RoleUser, Content: \"scenario: custom_data\"},\n\t\t{Role: context.RoleAssistant, Content: \"Here's some information for you.\"},\n\t}\n\n\tcompletion := &context.CompletionResponse{\n\t\tContent: \"This is the LLM completion that will be summarized.\",\n\t}\n\n\tpayload := &context.NextHookPayload{\n\t\tMessages: messages,\n\t\tCompletion: completion,\n\t\tTools: nil,\n\t\tError: \"\",\n\t}\n\n\tresponse, _, err := agent.HookScript.Next(ctx, payload)\n\tif err != nil {\n\t\tt.Fatalf(\"Next hook failed: %v\", err)\n\t}\n\n\tassert.NotNil(t, response, \"Custom data scenario should return response\")\n\tassert.NotNil(t, response.Data, \"Response should have Data\")\n\n\tdataMap, ok := response.Data.(map[string]interface{})\n\tassert.True(t, ok, \"Data should be a map\")\n\tassert.Equal(t, \"custom_response\", dataMap[\"type\"])\n\tassert.Contains(t, dataMap, \"timestamp\")\n\n\tt.Log(\"✓ Custom data response scenario passed\")\n}\n\n// TestRealWorldNextDelegate tests agent delegation\nfunc TestRealWorldNextDelegate(t *testing.T) {\n\tif testing.Short() {\n\t\tt.Skip(\"Skipping real world Next Hook test in short mode\")\n\t}\n\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\tagent, err := assistant.Get(\"tests.realworld-next\")\n\tif err != nil {\n\t\tt.Fatalf(\"Failed to get assistant: %v\", err)\n\t}\n\n\tctx := newRealWorldNextContext(\"test-next-delegate\", \"tests.realworld-next\")\n\n\tmessages := []context.Message{\n\t\t{Role: context.RoleUser, Content: \"scenario: delegate\"},\n\t}\n\n\tcompletion := &context.CompletionResponse{\n\t\tContent: \"I should delegate this request to another agent.\",\n\t}\n\n\tpayload := &context.NextHookPayload{\n\t\tMessages: messages,\n\t\tCompletion: completion,\n\t\tTools: nil,\n\t\tError: \"\",\n\t}\n\n\tresponse, _, err := agent.HookScript.Next(ctx, payload)\n\tif err != nil {\n\t\tt.Fatalf(\"Next hook failed: %v\", err)\n\t}\n\n\tassert.NotNil(t, response, \"Delegate scenario should return response\")\n\tassert.NotNil(t, response.Delegate, \"Response should have Delegate\")\n\tassert.Equal(t, \"tests.create\", response.Delegate.AgentID)\n\tassert.NotEmpty(t, response.Delegate.Messages)\n\n\tt.Log(\"✓ Delegation scenario passed\")\n}\n\n// TestRealWorldNextProcessTools tests tool result processing\nfunc TestRealWorldNextProcessTools(t *testing.T) {\n\tif testing.Short() {\n\t\tt.Skip(\"Skipping real world Next Hook test in short mode\")\n\t}\n\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\tagent, err := assistant.Get(\"tests.realworld-next\")\n\tif err != nil {\n\t\tt.Fatalf(\"Failed to get assistant: %v\", err)\n\t}\n\n\tctx := newRealWorldNextContext(\"test-next-tools\", \"tests.realworld-next\")\n\n\tmessages := []context.Message{\n\t\t{Role: context.RoleUser, Content: \"scenario: process_tools\"},\n\t}\n\n\tcompletion := &context.CompletionResponse{\n\t\tContent: \"Tool calls have been executed.\",\n\t}\n\n\t// Simulate tool call results\n\ttools := []context.ToolCallResponse{\n\t\t{\n\t\t\tToolCallID: \"call_1\",\n\t\t\tServer: \"test-server\",\n\t\t\tTool: \"test-tool-1\",\n\t\t\tResult: map[string]interface{}{\"status\": \"success\"},\n\t\t\tError: \"\",\n\t\t},\n\t\t{\n\t\t\tToolCallID: \"call_2\",\n\t\t\tServer: \"test-server\",\n\t\t\tTool: \"test-tool-2\",\n\t\t\tResult: nil,\n\t\t\tError: \"Tool execution failed\",\n\t\t},\n\t}\n\n\tpayload := &context.NextHookPayload{\n\t\tMessages: messages,\n\t\tCompletion: completion,\n\t\tTools: tools,\n\t\tError: \"\",\n\t}\n\n\tresponse, _, err := agent.HookScript.Next(ctx, payload)\n\tif err != nil {\n\t\tt.Fatalf(\"Next hook failed: %v\", err)\n\t}\n\n\tassert.NotNil(t, response, \"Process tools scenario should return response\")\n\tassert.NotNil(t, response.Data, \"Response should have Data\")\n\n\tdataMap, ok := response.Data.(map[string]interface{})\n\tassert.True(t, ok, \"Data should be a map\")\n\tassert.Equal(t, \"Tool execution summary\", dataMap[\"message\"])\n\n\t// Check summary\n\tsummary, ok := dataMap[\"summary\"].(map[string]interface{})\n\tassert.True(t, ok, \"Should have summary\")\n\tassert.Equal(t, float64(2), summary[\"total\"])\n\tassert.Equal(t, float64(1), summary[\"successful\"])\n\tassert.Equal(t, float64(1), summary[\"failed\"])\n\n\tt.Log(\"✓ Process tools scenario passed\")\n}\n\n// TestRealWorldNextErrorRecovery tests error handling and recovery\nfunc TestRealWorldNextErrorRecovery(t *testing.T) {\n\tif testing.Short() {\n\t\tt.Skip(\"Skipping real world Next Hook test in short mode\")\n\t}\n\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\tagent, err := assistant.Get(\"tests.realworld-next\")\n\tif err != nil {\n\t\tt.Fatalf(\"Failed to get assistant: %v\", err)\n\t}\n\n\tctx := newRealWorldNextContext(\"test-next-error\", \"tests.realworld-next\")\n\n\tmessages := []context.Message{\n\t\t{Role: context.RoleUser, Content: \"scenario: error_recovery\"},\n\t}\n\n\tcompletion := &context.CompletionResponse{\n\t\tContent: \"An error occurred during processing.\",\n\t}\n\n\tpayload := &context.NextHookPayload{\n\t\tMessages: messages,\n\t\tCompletion: completion,\n\t\tTools: nil,\n\t\tError: \"System error: Database connection timeout\",\n\t}\n\n\tresponse, _, err := agent.HookScript.Next(ctx, payload)\n\tif err != nil {\n\t\tt.Fatalf(\"Next hook failed: %v\", err)\n\t}\n\n\tassert.NotNil(t, response, \"Error recovery scenario should return response\")\n\tassert.NotNil(t, response.Data, \"Response should have Data\")\n\n\tdataMap, ok := response.Data.(map[string]interface{})\n\tassert.True(t, ok, \"Data should be a map\")\n\tassert.Equal(t, \"Error was handled by Next Hook\", dataMap[\"message\"])\n\tassert.Contains(t, dataMap, \"error\")\n\tassert.Contains(t, dataMap, \"recovery_action\")\n\n\tt.Log(\"✓ Error recovery scenario passed\")\n}\n\n// TestRealWorldNextConditional tests conditional logic based on completion\nfunc TestRealWorldNextConditional(t *testing.T) {\n\tif testing.Short() {\n\t\tt.Skip(\"Skipping real world Next Hook test in short mode\")\n\t}\n\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\tagent, err := assistant.Get(\"tests.realworld-next\")\n\tif err != nil {\n\t\tt.Fatalf(\"Failed to get assistant: %v\", err)\n\t}\n\n\tctx := newRealWorldNextContext(\"test-next-conditional\", \"tests.realworld-next\")\n\n\tt.Run(\"ConditionalSuccess\", func(t *testing.T) {\n\t\tmessages := []context.Message{\n\t\t\t{Role: context.RoleUser, Content: \"scenario: conditional\"},\n\t\t}\n\n\t\tcompletion := &context.CompletionResponse{\n\t\t\tContent: \"The operation completed successfully. All tasks are done.\",\n\t\t}\n\n\t\tpayload := &context.NextHookPayload{\n\t\t\tMessages: messages,\n\t\t\tCompletion: completion,\n\t\t\tTools: nil,\n\t\t\tError: \"\",\n\t\t}\n\n\t\tresponse, _, err := agent.HookScript.Next(ctx, payload)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Next hook failed: %v\", err)\n\t\t}\n\n\t\tassert.NotNil(t, response, \"Conditional scenario should return response\")\n\t\tassert.NotNil(t, response.Data, \"Response should have Data\")\n\n\t\tdataMap, ok := response.Data.(map[string]interface{})\n\t\tassert.True(t, ok, \"Data should be a map\")\n\t\tassert.Equal(t, \"Conditional analysis complete\", dataMap[\"message\"])\n\t\tassert.Contains(t, dataMap, \"action\")\n\t\tassert.Contains(t, dataMap, \"conditions\")\n\n\t\tt.Log(\"✓ Conditional (success) scenario passed\")\n\t})\n\n\tt.Run(\"ConditionalDelegate\", func(t *testing.T) {\n\t\tmessages := []context.Message{\n\t\t\t{Role: context.RoleUser, Content: \"scenario: conditional\"},\n\t\t}\n\n\t\tcompletion := &context.CompletionResponse{\n\t\t\tContent: \"I should delegate this request to another service for better handling.\",\n\t\t}\n\n\t\tpayload := &context.NextHookPayload{\n\t\t\tMessages: messages,\n\t\t\tCompletion: completion,\n\t\t\tTools: nil,\n\t\t\tError: \"\",\n\t\t}\n\n\t\tresponse, _, err := agent.HookScript.Next(ctx, payload)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Next hook failed: %v\", err)\n\t\t}\n\n\t\tassert.NotNil(t, response, \"Conditional delegate should return response\")\n\t\tassert.NotNil(t, response.Delegate, \"Should delegate based on condition\")\n\t\tassert.Equal(t, \"tests.create\", response.Delegate.AgentID)\n\n\t\tt.Log(\"✓ Conditional (delegate) scenario passed\")\n\t})\n}\n\n// TestRealWorldNextDefault tests default behavior\nfunc TestRealWorldNextDefault(t *testing.T) {\n\tif testing.Short() {\n\t\tt.Skip(\"Skipping real world Next Hook test in short mode\")\n\t}\n\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\tagent, err := assistant.Get(\"tests.realworld-next\")\n\tif err != nil {\n\t\tt.Fatalf(\"Failed to get assistant: %v\", err)\n\t}\n\n\tctx := newRealWorldNextContext(\"test-next-default\", \"tests.realworld-next\")\n\n\tmessages := []context.Message{\n\t\t{Role: context.RoleUser, Content: \"Just a normal request\"},\n\t}\n\n\tcompletion := &context.CompletionResponse{\n\t\tContent: \"Here's the response to your request.\",\n\t}\n\n\tpayload := &context.NextHookPayload{\n\t\tMessages: messages,\n\t\tCompletion: completion,\n\t\tTools: nil,\n\t\tError: \"\",\n\t}\n\n\tresponse, _, err := agent.HookScript.Next(ctx, payload)\n\tif err != nil {\n\t\tt.Fatalf(\"Next hook failed: %v\", err)\n\t}\n\n\t// Default behavior should return nil\n\tassert.Nil(t, response, \"Default scenario should return nil for standard response\")\n\n\tt.Log(\"✓ Default scenario passed\")\n}\n" }, { "path": "agent/assistant/hook/realworld_stress_test.go", "content": "package hook_test\n\nimport (\n\tstdContext \"context\"\n\t\"fmt\"\n\t\"runtime\"\n\t\"sync\"\n\t\"testing\"\n\t\"time\"\n\n\t\"github.com/stretchr/testify/assert\"\n\t\"github.com/yaoapp/yao/agent/assistant\"\n\t\"github.com/yaoapp/yao/agent/context\"\n\t\"github.com/yaoapp/yao/agent/testutils\"\n\t\"github.com/yaoapp/yao/openapi/oauth/types\"\n)\n\n// ============================================================================\n// Real World Stress Tests\n// These tests simulate actual production usage patterns with Stream() flow\n// ============================================================================\n\n// TestRealWorldSimpleScenario tests basic Stream() flow with simple Create hook\nfunc TestRealWorldSimpleScenario(t *testing.T) {\n\tif testing.Short() {\n\t\tt.Skip(\"Skipping real world test in short mode\")\n\t}\n\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\tagent, err := assistant.Get(\"tests.realworld\")\n\tif err != nil {\n\t\tt.Fatalf(\"Failed to get assistant: %v\", err)\n\t}\n\n\tctx := newRealWorldContext(\"test-simple\", \"tests.realworld\")\n\n\t// Test Create hook with simple scenario\n\tmessages := []context.Message{\n\t\t{Role: \"user\", Content: \"simple\"},\n\t}\n\n\tresponse, _, err := agent.HookScript.Create(ctx, messages)\n\tif err != nil {\n\t\tt.Fatalf(\"Create failed: %v\", err)\n\t}\n\n\tassert.NotNil(t, response)\n\tassert.NotEmpty(t, response.Messages)\n\tassert.Equal(t, \"simple\", response.Metadata[\"scenario\"])\n}\n\n// TestRealWorldMCPScenarios tests MCP integration scenarios\nfunc TestRealWorldMCPScenarios(t *testing.T) {\n\tif testing.Short() {\n\t\tt.Skip(\"Skipping real world test in short mode\")\n\t}\n\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\tagent, err := assistant.Get(\"tests.realworld\")\n\tif err != nil {\n\t\tt.Fatalf(\"Failed to get assistant: %v\", err)\n\t}\n\n\tt.Run(\"MCP Health\", func(t *testing.T) {\n\t\tctx := newRealWorldContext(\"test-mcp-health\", \"tests.realworld\")\n\n\t\tmessages := []context.Message{\n\t\t\t{Role: \"user\", Content: \"mcp_health\"},\n\t\t}\n\n\t\tresponse, _, err := agent.HookScript.Create(ctx, messages)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Create failed: %v\", err)\n\t\t}\n\n\t\t// Detailed validation\n\t\tassert.NotNil(t, response)\n\t\tassert.NotEmpty(t, response.Messages)\n\n\t\t// Check if metadata exists\n\t\tif response.Metadata == nil {\n\t\t\tt.Logf(\"⚠ Metadata is nil - checking messages content\")\n\t\t\t// Verify messages contain expected content\n\t\t\tmessageContent := \"\"\n\t\t\tfor _, msg := range response.Messages {\n\t\t\t\tif content, ok := msg.Content.(string); ok {\n\t\t\t\t\tmessageContent += content + \"\\n\"\n\t\t\t\t}\n\t\t\t}\n\t\t\tassert.Contains(t, messageContent, \"Health\", \"Message should mention health\")\n\t\t\tassert.Contains(t, messageContent, \"Tools\", \"Message should mention tools\")\n\t\t\tt.Logf(\"✓ MCP Health executed (verified via message content)\")\n\t\t} else {\n\t\t\tassert.Equal(t, \"mcp_health\", response.Metadata[\"scenario\"])\n\n\t\t\t// Verify metadata contains MCP results\n\t\t\tif toolsCount, ok := response.Metadata[\"tools_count\"]; ok {\n\t\t\t\tcount := int(toolsCount.(float64))\n\t\t\t\tassert.Greater(t, count, 0, \"Should have tools from MCP\")\n\t\t\t\tt.Logf(\"✓ MCP Health: %d tools, health data: %v\",\n\t\t\t\t\tcount, response.Metadata[\"health_data\"])\n\t\t\t}\n\t\t}\n\n\t\tctx.Release()\n\t})\n\n\tt.Run(\"MCP Tools\", func(t *testing.T) {\n\t\tctx := newRealWorldContext(\"test-mcp-tools\", \"tests.realworld\")\n\n\t\tmessages := []context.Message{\n\t\t\t{Role: \"user\", Content: \"mcp_tools\"},\n\t\t}\n\n\t\tresponse, _, err := agent.HookScript.Create(ctx, messages)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Create failed: %v\", err)\n\t\t}\n\n\t\t// Detailed validation\n\t\tassert.NotNil(t, response)\n\t\tassert.NotEmpty(t, response.Messages)\n\n\t\t// Check if metadata exists\n\t\tif response.Metadata == nil {\n\t\t\tt.Logf(\"⚠ Metadata is nil - checking messages content\")\n\t\t\t// Verify messages contain expected content\n\t\t\tmessageContent := \"\"\n\t\t\tfor _, msg := range response.Messages {\n\t\t\t\tif content, ok := msg.Content.(string); ok {\n\t\t\t\t\tmessageContent += content + \"\\n\"\n\t\t\t\t}\n\t\t\t}\n\t\t\tassert.Contains(t, messageContent, \"Tools\", \"Message should mention tools\")\n\t\t\tassert.Contains(t, messageContent, \"Ping\", \"Message should mention ping\")\n\t\t\tt.Logf(\"✓ MCP Tools executed (verified via message content)\")\n\t\t} else {\n\t\t\tassert.Equal(t, \"mcp_tools\", response.Metadata[\"scenario\"])\n\n\t\t\t// Verify tools were called\n\t\t\tif toolsCount, ok := response.Metadata[\"tools_count\"]; ok {\n\t\t\t\tcount := int(toolsCount.(float64))\n\t\t\t\tassert.Greater(t, count, 0, \"Should have tools from MCP\")\n\n\t\t\t\t// Verify operations list\n\t\t\t\tif operations, ok := response.Metadata[\"operations\"].([]interface{}); ok {\n\t\t\t\t\tassert.Len(t, operations, 2, \"Should execute 2 operations: ping, status\")\n\t\t\t\t\tt.Logf(\"✓ MCP Tools: %d tools, operations: %v\", count, operations)\n\t\t\t\t}\n\t\t\t}\n\t\t}\n\n\t\tctx.Release()\n\t})\n\n\tt.Run(\"Full Workflow\", func(t *testing.T) {\n\t\tctx := newRealWorldContext(\"test-full-workflow\", \"tests.realworld\")\n\n\t\t// Initialize stack for trace\n\t\tstack, _, done := context.EnterStack(ctx, \"tests.realworld\", &context.Options{})\n\t\tdefer done()\n\t\tctx.Stack = stack\n\n\t\tmessages := []context.Message{\n\t\t\t{Role: \"user\", Content: \"full_workflow\"},\n\t\t}\n\n\t\tresponse, _, err := agent.HookScript.Create(ctx, messages)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Create failed: %v\", err)\n\t\t}\n\n\t\t// Detailed validation\n\t\tassert.NotNil(t, response)\n\t\tassert.NotEmpty(t, response.Messages)\n\n\t\t// Check if metadata exists\n\t\tif response.Metadata == nil {\n\t\t\tt.Logf(\"⚠ Metadata is nil - checking messages content\")\n\t\t\t// Verify messages contain expected content\n\t\t\tmessageContent := \"\"\n\t\t\tfor _, msg := range response.Messages {\n\t\t\t\tif content, ok := msg.Content.(string); ok {\n\t\t\t\t\tmessageContent += content + \"\\n\"\n\t\t\t\t}\n\t\t\t}\n\t\t\tassert.Contains(t, messageContent, \"Workflow\", \"Message should mention workflow\")\n\t\t\tassert.Contains(t, messageContent, \"Tools\", \"Message should mention tools\")\n\t\t\tassert.Contains(t, messageContent, \"Roles\", \"Message should mention database roles\")\n\t\t\tt.Logf(\"✓ Full Workflow executed (verified via message content)\")\n\t\t} else {\n\t\t\tassert.Equal(t, \"full_workflow\", response.Metadata[\"scenario\"])\n\n\t\t\t// Verify all phases completed\n\t\t\tif phasesCompleted, ok := response.Metadata[\"phases_completed\"]; ok {\n\t\t\t\tphases := int(phasesCompleted.(float64))\n\t\t\t\tassert.Equal(t, 4, phases, \"Should complete 4 phases\")\n\n\t\t\t\t// Verify MCP tools\n\t\t\t\tif mcpTools, ok := response.Metadata[\"mcp_tools\"]; ok {\n\t\t\t\t\ttools := int(mcpTools.(float64))\n\t\t\t\t\tassert.Greater(t, tools, 0, \"Should have MCP tools\")\n\n\t\t\t\t\t// Verify DB records\n\t\t\t\t\tif dbRecords, ok := response.Metadata[\"db_records\"]; ok {\n\t\t\t\t\t\trecords := int(dbRecords.(float64))\n\t\t\t\t\t\tassert.GreaterOrEqual(t, records, 0, \"Should have DB query result\")\n\n\t\t\t\t\t\tt.Logf(\"✓ Full Workflow: %d phases, %d MCP tools, %d DB records\",\n\t\t\t\t\t\t\tphases, tools, records)\n\t\t\t\t\t}\n\t\t\t\t}\n\t\t\t}\n\t\t}\n\n\t\tctx.Release()\n\t})\n}\n\n// TestRealWorldTraceIntensive tests trace-heavy scenarios\nfunc TestRealWorldTraceIntensive(t *testing.T) {\n\tif testing.Short() {\n\t\tt.Skip(\"Skipping real world test in short mode\")\n\t}\n\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\tagent, err := assistant.Get(\"tests.realworld\")\n\tif err != nil {\n\t\tt.Fatalf(\"Failed to get assistant: %v\", err)\n\t}\n\n\tctx := newRealWorldContext(\"test-trace-intensive\", \"tests.realworld\")\n\tstack, _, done := context.EnterStack(ctx, \"tests.realworld\", &context.Options{})\n\tdefer done()\n\tctx.Stack = stack\n\n\tmessages := []context.Message{\n\t\t{Role: \"user\", Content: \"trace_intensive\"},\n\t}\n\n\tresponse, _, err := agent.HookScript.Create(ctx, messages)\n\tif err != nil {\n\t\tt.Fatalf(\"Create failed: %v\", err)\n\t}\n\n\tassert.NotNil(t, response)\n\tassert.Equal(t, \"trace_intensive\", response.Metadata[\"scenario\"])\n\tassert.NotZero(t, response.Metadata[\"nodes_created\"])\n}\n\n// TestRealWorldStressSimple tests simple scenario under stress\nfunc TestRealWorldStressSimple(t *testing.T) {\n\tif testing.Short() {\n\t\tt.Skip(\"Skipping stress test in short mode\")\n\t}\n\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\tagent, err := assistant.Get(\"tests.realworld\")\n\tif err != nil {\n\t\tt.Fatalf(\"Failed to get assistant: %v\", err)\n\t}\n\n\titerations := 100\n\tstartMemory := getMemStats()\n\n\tfor i := 0; i < iterations; i++ {\n\t\tctx := newRealWorldContext(fmt.Sprintf(\"stress-simple-%d\", i), \"tests.realworld\")\n\n\t\tmessages := []context.Message{\n\t\t\t{Role: \"user\", Content: \"simple\"},\n\t\t}\n\n\t\tresponse, _, err := agent.HookScript.Create(ctx, messages)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Iteration %d failed: %v\", i, err)\n\t\t}\n\n\t\t// Validate response\n\t\tassert.NotNil(t, response, \"Iteration %d: response should not be nil\", i)\n\t\tassert.NotEmpty(t, response.Messages, \"Iteration %d: messages should not be empty\", i)\n\t\tif response.Metadata != nil {\n\t\t\tassert.Equal(t, \"simple\", response.Metadata[\"scenario\"], \"Iteration %d: scenario mismatch\", i)\n\t\t}\n\n\t\t// Explicit cleanup\n\t\tctx.Release()\n\n\t\tif i%20 == 0 {\n\t\t\truntime.GC()\n\t\t\tcurrentMemory := getMemStats()\n\t\t\tt.Logf(\"Iteration %d: Memory: %d MB\", i, currentMemory/1024/1024)\n\t\t}\n\t}\n\n\truntime.GC()\n\tendMemory := getMemStats()\n\n\tt.Logf(\"Simple stress: %d iterations\", iterations)\n\tt.Logf(\"Start memory: %d MB\", startMemory/1024/1024)\n\tt.Logf(\"End memory: %d MB\", endMemory/1024/1024)\n\n\tif endMemory > startMemory {\n\t\tt.Logf(\"Memory growth: %d MB\", (endMemory-startMemory)/1024/1024)\n\t} else {\n\t\tt.Logf(\"Memory decreased: %d MB\", (startMemory-endMemory)/1024/1024)\n\t}\n}\n\n// TestRealWorldStressMCP tests MCP scenarios under stress\nfunc TestRealWorldStressMCP(t *testing.T) {\n\tif testing.Short() {\n\t\tt.Skip(\"Skipping stress test in short mode\")\n\t}\n\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\tagent, err := assistant.Get(\"tests.realworld\")\n\tif err != nil {\n\t\tt.Fatalf(\"Failed to get assistant: %v\", err)\n\t}\n\n\titerations := 50\n\tscenarios := []string{\"mcp_health\", \"mcp_tools\"}\n\tstartMemory := getMemStats()\n\n\tfor i := 0; i < iterations; i++ {\n\t\tscenario := scenarios[i%len(scenarios)]\n\t\tctx := newRealWorldContext(fmt.Sprintf(\"stress-mcp-%d\", i), \"tests.realworld\")\n\n\t\t// Initialize stack for trace\n\t\tstack, _, done := context.EnterStack(ctx, \"tests.realworld\", &context.Options{})\n\t\tctx.Stack = stack\n\n\t\tmessages := []context.Message{\n\t\t\t{Role: \"user\", Content: scenario},\n\t\t}\n\n\t\tresponse, _, err := agent.HookScript.Create(ctx, messages)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Iteration %d (%s) failed: %v\", i, scenario, err)\n\t\t}\n\n\t\t// Validate response\n\t\tassert.NotNil(t, response, \"Iteration %d (%s): response should not be nil\", i, scenario)\n\t\tassert.NotEmpty(t, response.Messages, \"Iteration %d (%s): messages should not be empty\", i, scenario)\n\n\t\t// Validate metadata\n\t\tif response.Metadata != nil {\n\t\t\tassert.Equal(t, scenario, response.Metadata[\"scenario\"], \"Iteration %d: scenario mismatch\", i)\n\n\t\t\t// Verify MCP-specific data\n\t\t\tif scenario == \"mcp_health\" {\n\t\t\t\tassert.NotNil(t, response.Metadata[\"tools_count\"], \"Iteration %d: should have tools_count\", i)\n\t\t\t\tif toolsCount, ok := response.Metadata[\"tools_count\"].(float64); ok {\n\t\t\t\t\tassert.Greater(t, int(toolsCount), 0, \"Iteration %d: should have at least 1 tool\", i)\n\t\t\t\t\tassert.Equal(t, 3, int(toolsCount), \"Iteration %d: echo should have 3 tools\", i)\n\t\t\t\t}\n\t\t\t\tassert.NotNil(t, response.Metadata[\"health_data\"], \"Iteration %d: should have health_data\", i)\n\t\t\t} else if scenario == \"mcp_tools\" {\n\t\t\t\tassert.NotNil(t, response.Metadata[\"tools_count\"], \"Iteration %d: should have tools_count\", i)\n\t\t\t\tif toolsCount, ok := response.Metadata[\"tools_count\"].(float64); ok {\n\t\t\t\t\tassert.Equal(t, 3, int(toolsCount), \"Iteration %d: echo should have 3 tools\", i)\n\t\t\t\t}\n\t\t\t\tassert.NotNil(t, response.Metadata[\"operations\"], \"Iteration %d: should have operations\", i)\n\t\t\t\tif operations, ok := response.Metadata[\"operations\"].([]interface{}); ok {\n\t\t\t\t\tassert.Len(t, operations, 2, \"Iteration %d: should have 2 operations (ping, status)\", i)\n\t\t\t\t}\n\t\t\t}\n\t\t} else {\n\t\t\tt.Errorf(\"Iteration %d (%s): metadata is nil\", i, scenario)\n\t\t}\n\n\t\t// Cleanup\n\t\tdone()\n\t\tctx.Release()\n\n\t\tif i%10 == 0 {\n\t\t\truntime.GC()\n\t\t\tcurrentMemory := getMemStats()\n\t\t\tt.Logf(\"Iteration %d (%s): Memory: %d MB\", i, scenario, currentMemory/1024/1024)\n\t\t}\n\t}\n\n\truntime.GC()\n\tendMemory := getMemStats()\n\n\tt.Logf(\"MCP stress: %d iterations\", iterations)\n\tt.Logf(\"Start memory: %d MB\", startMemory/1024/1024)\n\tt.Logf(\"End memory: %d MB\", endMemory/1024/1024)\n\n\tif endMemory > startMemory {\n\t\tt.Logf(\"Memory growth: %d MB\", (endMemory-startMemory)/1024/1024)\n\t} else {\n\t\tt.Logf(\"Memory decreased: %d MB\", (startMemory-endMemory)/1024/1024)\n\t}\n}\n\n// TestRealWorldStressFullWorkflow tests complete workflow under stress\nfunc TestRealWorldStressFullWorkflow(t *testing.T) {\n\tif testing.Short() {\n\t\tt.Skip(\"Skipping stress test in short mode\")\n\t}\n\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\tagent, err := assistant.Get(\"tests.realworld\")\n\tif err != nil {\n\t\tt.Fatalf(\"Failed to get assistant: %v\", err)\n\t}\n\n\titerations := 30\n\tstartMemory := getMemStats()\n\tstartTime := time.Now()\n\n\tfor i := 0; i < iterations; i++ {\n\t\tctx := newRealWorldContext(fmt.Sprintf(\"stress-workflow-%d\", i), \"tests.realworld\")\n\n\t\t// Initialize stack for trace\n\t\tstack, _, done := context.EnterStack(ctx, \"tests.realworld\", &context.Options{})\n\t\tctx.Stack = stack\n\n\t\tmessages := []context.Message{\n\t\t\t{Role: \"user\", Content: \"full_workflow\"},\n\t\t}\n\n\t\tresponse, _, err := agent.HookScript.Create(ctx, messages)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Iteration %d failed: %v\", i, err)\n\t\t}\n\n\t\t// Verify response\n\t\tassert.NotNil(t, response, \"Iteration %d: response should not be nil\", i)\n\t\tassert.NotEmpty(t, response.Messages, \"Iteration %d: messages should not be empty\", i)\n\t\tif response.Metadata != nil {\n\t\t\tassert.Equal(t, \"full_workflow\", response.Metadata[\"scenario\"], \"Iteration %d: scenario mismatch\", i)\n\t\t\t// Verify workflow-specific metadata\n\t\t\tif phasesCompleted, ok := response.Metadata[\"phases_completed\"]; ok {\n\t\t\t\tphases := int(phasesCompleted.(float64))\n\t\t\t\tassert.Equal(t, 4, phases, \"Iteration %d: should complete 4 phases\", i)\n\t\t\t}\n\t\t\tif mcpTools, ok := response.Metadata[\"mcp_tools\"]; ok {\n\t\t\t\ttools := int(mcpTools.(float64))\n\t\t\t\tassert.Greater(t, tools, 0, \"Iteration %d: should have MCP tools\", i)\n\t\t\t}\n\t\t}\n\n\t\t// Cleanup\n\t\tdone()\n\t\tctx.Release()\n\n\t\tif i%10 == 0 {\n\t\t\truntime.GC()\n\t\t\tcurrentMemory := getMemStats()\n\t\t\telapsed := time.Since(startTime)\n\t\t\tt.Logf(\"Iteration %d: Memory: %d MB, Elapsed: %v\", i, currentMemory/1024/1024, elapsed)\n\t\t}\n\t}\n\n\tduration := time.Since(startTime)\n\truntime.GC()\n\tendMemory := getMemStats()\n\n\tavgTime := duration / time.Duration(iterations)\n\tt.Logf(\"Full workflow stress: %d iterations\", iterations)\n\tt.Logf(\"Total time: %v\", duration)\n\tt.Logf(\"Average time per iteration: %v\", avgTime)\n\tt.Logf(\"Start memory: %d MB\", startMemory/1024/1024)\n\tt.Logf(\"End memory: %d MB\", endMemory/1024/1024)\n\n\tif endMemory > startMemory {\n\t\tt.Logf(\"Memory growth: %d MB\", (endMemory-startMemory)/1024/1024)\n\t} else {\n\t\tt.Logf(\"Memory decreased: %d MB\", (startMemory-endMemory)/1024/1024)\n\t}\n}\n\n// TestRealWorldStressConcurrent tests concurrent real-world usage\nfunc TestRealWorldStressConcurrent(t *testing.T) {\n\tif testing.Short() {\n\t\tt.Skip(\"Skipping stress test in short mode\")\n\t}\n\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\tagent, err := assistant.Get(\"tests.realworld\")\n\tif err != nil {\n\t\tt.Fatalf(\"Failed to get assistant: %v\", err)\n\t}\n\n\tgoroutines := 100\n\titerationsPerGoroutine := 10\n\tscenarios := []string{\"simple\", \"mcp_health\", \"mcp_tools\", \"full_workflow\"}\n\n\tstartMemory := getMemStats()\n\tstartTime := time.Now()\n\n\tvar wg sync.WaitGroup\n\terrors := make(chan error, goroutines*iterationsPerGoroutine)\n\n\t// Track results for validation\n\ttype Result struct {\n\t\tgoroutineID int\n\t\titeration int\n\t\tscenario string\n\t\tmetadata map[string]interface{}\n\t}\n\tresults := make(chan Result, goroutines*iterationsPerGoroutine)\n\n\tfor g := 0; g < goroutines; g++ {\n\t\twg.Add(1)\n\t\tgo func(goroutineID int) {\n\t\t\tdefer wg.Done()\n\n\t\t\tfor i := 0; i < iterationsPerGoroutine; i++ {\n\t\t\t\tscenario := scenarios[(goroutineID+i)%len(scenarios)]\n\t\t\t\tctx := newRealWorldContext(\n\t\t\t\t\tfmt.Sprintf(\"concurrent-%d-%d\", goroutineID, i),\n\t\t\t\t\t\"tests.realworld\",\n\t\t\t\t)\n\n\t\t\t\t// Initialize stack for trace\n\t\t\t\tstack, _, done := context.EnterStack(ctx, \"tests.realworld\", &context.Options{})\n\t\t\t\tctx.Stack = stack\n\n\t\t\t\tmessages := []context.Message{\n\t\t\t\t\t{Role: \"user\", Content: scenario},\n\t\t\t\t}\n\n\t\t\t\tresponse, _, err := agent.HookScript.Create(ctx, messages)\n\t\t\t\tif err != nil {\n\t\t\t\t\terrors <- fmt.Errorf(\"goroutine %d iteration %d (%s): %v\", goroutineID, i, scenario, err)\n\t\t\t\t\tdone()\n\t\t\t\t\tctx.Release()\n\t\t\t\t\treturn\n\t\t\t\t}\n\n\t\t\t\t// Validate response\n\t\t\t\tif response == nil {\n\t\t\t\t\terrors <- fmt.Errorf(\"goroutine %d iteration %d (%s): nil response\", goroutineID, i, scenario)\n\t\t\t\t\tdone()\n\t\t\t\t\tctx.Release()\n\t\t\t\t\treturn\n\t\t\t\t}\n\n\t\t\t\tif len(response.Messages) == 0 {\n\t\t\t\t\terrors <- fmt.Errorf(\"goroutine %d iteration %d (%s): empty messages\", goroutineID, i, scenario)\n\t\t\t\t\tdone()\n\t\t\t\t\tctx.Release()\n\t\t\t\t\treturn\n\t\t\t\t}\n\n\t\t\t\t// Collect result\n\t\t\t\tresults <- Result{\n\t\t\t\t\tgoroutineID: goroutineID,\n\t\t\t\t\titeration: i,\n\t\t\t\t\tscenario: scenario,\n\t\t\t\t\tmetadata: response.Metadata,\n\t\t\t\t}\n\n\t\t\t\t// Cleanup\n\t\t\t\tdone()\n\t\t\t\tctx.Release()\n\t\t\t}\n\t\t}(g)\n\t}\n\n\twg.Wait()\n\tclose(errors)\n\tclose(results)\n\n\tduration := time.Since(startTime)\n\truntime.GC()\n\tendMemory := getMemStats()\n\n\t// Check for errors\n\terrorCount := 0\n\tfor err := range errors {\n\t\tt.Error(err)\n\t\terrorCount++\n\t}\n\n\tassert.Equal(t, 0, errorCount, \"No errors should occur in concurrent operations\")\n\n\t// Validate results\n\tscenarioCounts := make(map[string]int)\n\tvalidResults := 0\n\n\tfor result := range results {\n\t\tvalidResults++\n\t\tscenarioCounts[result.scenario]++\n\n\t\t// Validate metadata exists and has expected scenario\n\t\tif result.metadata != nil {\n\t\t\tif scenario, ok := result.metadata[\"scenario\"].(string); ok {\n\t\t\t\tif scenario != result.scenario {\n\t\t\t\t\tt.Errorf(\"Metadata mismatch: expected %s, got %s (goroutine %d, iteration %d)\",\n\t\t\t\t\t\tresult.scenario, scenario, result.goroutineID, result.iteration)\n\t\t\t\t}\n\t\t\t}\n\t\t}\n\t}\n\n\ttotalOperations := goroutines * iterationsPerGoroutine\n\tassert.Equal(t, totalOperations, validResults, \"All operations should return valid results\")\n\n\tavgTime := duration / time.Duration(totalOperations)\n\n\tt.Logf(\"✓ Concurrent stress: %d operations (goroutines: %d, iterations: %d)\",\n\t\ttotalOperations, goroutines, iterationsPerGoroutine)\n\tt.Logf(\"✓ Valid results: %d/%d (100%%)\", validResults, totalOperations)\n\tt.Logf(\"✓ Scenario distribution:\")\n\tfor scenario, count := range scenarioCounts {\n\t\tt.Logf(\" - %s: %d operations\", scenario, count)\n\t}\n\tt.Logf(\"✓ Total time: %v\", duration)\n\tt.Logf(\"✓ Average time per operation: %v\", avgTime)\n\tt.Logf(\"✓ Start memory: %d MB\", startMemory/1024/1024)\n\tt.Logf(\"✓ End memory: %d MB\", endMemory/1024/1024)\n\n\tif endMemory > startMemory {\n\t\tt.Logf(\"✓ Memory growth: %d MB\", (endMemory-startMemory)/1024/1024)\n\t} else {\n\t\tt.Logf(\"✓ Memory decreased: %d MB\", (startMemory-endMemory)/1024/1024)\n\t}\n}\n\n// TestRealWorldStressResourceHeavy tests resource-intensive scenarios\nfunc TestRealWorldStressResourceHeavy(t *testing.T) {\n\tif testing.Short() {\n\t\tt.Skip(\"Skipping stress test in short mode\")\n\t}\n\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\tagent, err := assistant.Get(\"tests.realworld\")\n\tif err != nil {\n\t\tt.Fatalf(\"Failed to get assistant: %v\", err)\n\t}\n\n\titerations := 20\n\tstartMemory := getMemStats()\n\tstartTime := time.Now()\n\n\tfor i := 0; i < iterations; i++ {\n\t\tctx := newRealWorldContext(fmt.Sprintf(\"stress-heavy-%d\", i), \"tests.realworld\")\n\n\t\t// Initialize stack for trace\n\t\tstack, _, done := context.EnterStack(ctx, \"tests.realworld\", &context.Options{})\n\t\tctx.Stack = stack\n\n\t\tmessages := []context.Message{\n\t\t\t{Role: \"user\", Content: \"resource_heavy\"},\n\t\t}\n\n\t\tresponse, _, err := agent.HookScript.Create(ctx, messages)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Iteration %d failed: %v\", i, err)\n\t\t}\n\n\t\t// Validate response\n\t\tassert.NotNil(t, response, \"Iteration %d: response should not be nil\", i)\n\t\tassert.NotEmpty(t, response.Messages, \"Iteration %d: messages should not be empty\", i)\n\t\tif response.Metadata != nil {\n\t\t\tassert.Equal(t, \"resource_heavy\", response.Metadata[\"scenario\"], \"Iteration %d: scenario mismatch\", i)\n\t\t\t// Verify resource-heavy metadata\n\t\t\tif mcpIterations, ok := response.Metadata[\"mcp_iterations\"]; ok {\n\t\t\t\titerations := int(mcpIterations.(float64))\n\t\t\t\tassert.Equal(t, 5, iterations, \"Iteration %d: should have 5 MCP iterations\", i)\n\t\t\t}\n\t\t}\n\n\t\t// Cleanup\n\t\tdone()\n\t\tctx.Release()\n\n\t\tif i%5 == 0 {\n\t\t\truntime.GC()\n\t\t\tcurrentMemory := getMemStats()\n\t\t\telapsed := time.Since(startTime)\n\t\t\tt.Logf(\"Iteration %d: Memory: %d MB, Elapsed: %v\", i, currentMemory/1024/1024, elapsed)\n\t\t}\n\t}\n\n\tduration := time.Since(startTime)\n\truntime.GC()\n\tendMemory := getMemStats()\n\n\tavgTime := duration / time.Duration(iterations)\n\tt.Logf(\"Resource heavy stress: %d iterations\", iterations)\n\tt.Logf(\"Total time: %v\", duration)\n\tt.Logf(\"Average time per iteration: %v\", avgTime)\n\tt.Logf(\"Start memory: %d MB\", startMemory/1024/1024)\n\tt.Logf(\"End memory: %d MB\", endMemory/1024/1024)\n\n\tif endMemory > startMemory {\n\t\tmemoryGrowth := int64(endMemory - startMemory)\n\t\tt.Logf(\"Memory growth: %d MB\", memoryGrowth/1024/1024)\n\t\t// Allow up to 100MB growth for resource-heavy operations\n\t\tassert.Less(t, memoryGrowth, int64(100*1024*1024), \"Memory growth should be reasonable\")\n\t} else {\n\t\tt.Logf(\"Memory decreased: %d MB\", (startMemory-endMemory)/1024/1024)\n\t}\n}\n\n// ============================================================================\n// Helper Functions\n// ============================================================================\n\n// newRealWorldContext creates a Context for real-world testing\nfunc newRealWorldContext(chatID, assistantID string) *context.Context {\n\tauthorized := &types.AuthorizedInfo{\n\t\tSubject: \"realworld-test-user\",\n\t\tClientID: \"realworld-test-client\",\n\t\tScope: \"openid profile email\",\n\t\tSessionID: \"realworld-test-session\",\n\t\tUserID: \"realworld-user-123\",\n\t\tTeamID: \"realworld-team-456\",\n\t\tTenantID: \"realworld-tenant-789\",\n\t\tRememberMe: true,\n\t\tConstraints: types.DataConstraints{\n\t\t\tOwnerOnly: false,\n\t\t\tCreatorOnly: false,\n\t\t\tEditorOnly: false,\n\t\t\tTeamOnly: true,\n\t\t\tExtra: map[string]interface{}{\n\t\t\t\t\"department\": \"engineering\",\n\t\t\t\t\"region\": \"us-west\",\n\t\t\t\t\"project\": \"yao-realworld-test\",\n\t\t\t},\n\t\t},\n\t}\n\n\tctx := context.New(stdContext.Background(), authorized, chatID)\n\tctx.AssistantID = assistantID\n\tctx.Locale = \"en-us\"\n\tctx.Theme = \"light\"\n\tctx.Client = context.Client{\n\t\tType: \"web\",\n\t\tUserAgent: \"RealWorldTest/1.0\",\n\t\tIP: \"127.0.0.1\",\n\t}\n\tctx.Referer = context.RefererAPI\n\tctx.Accept = context.AcceptWebCUI\n\tctx.Route = \"\"\n\tctx.Metadata = make(map[string]interface{})\n\treturn ctx\n}\n\n// getMemStats returns current memory allocation in bytes\nfunc getMemStats() uint64 {\n\truntime.GC()\n\tvar m runtime.MemStats\n\truntime.ReadMemStats(&m)\n\treturn m.Alloc\n}\n" }, { "path": "agent/assistant/hook/script.go", "content": "package hook\n\nimport (\n\t\"strings\"\n\n\t\"github.com/yaoapp/yao/agent/context\"\n)\n\n// Execute execute the script\nfunc (s *Script) Execute(ctx *context.Context, method string, args ...interface{}) (interface{}, error) {\n\tif s == nil || s.Script == nil {\n\t\treturn nil, nil\n\t}\n\n\tvar sid = \"\"\n\tif ctx.Authorized != nil {\n\t\tsid = ctx.Authorized.SessionID\n\t}\n\n\tscriptCtx, err := s.NewContext(sid, nil)\n\tif err != nil {\n\t\treturn nil, err\n\t}\n\tdefer scriptCtx.Close()\n\n\t// Set authorized information if available\n\tif ctx.Authorized != nil {\n\t\tscriptCtx.WithAuthorized(ctx.Authorized.AuthorizedToMap())\n\t}\n\n\t// The first argument is the context\n\targs = append([]interface{}{ctx}, args...)\n\n\t// Try to call the method\n\tresult, err := scriptCtx.CallWith(ctx.Context, method, args...)\n\n\t// If method doesn't exist (ReferenceError or similar), return nil without error\n\tif err != nil && (strings.Contains(err.Error(), \"is not defined\") ||\n\t\tstrings.Contains(err.Error(), \"is not a function\") ||\n\t\tstrings.Contains(err.Error(), \"is not a Function\")) {\n\t\treturn nil, nil\n\t}\n\n\treturn result, err\n}\n" }, { "path": "agent/assistant/hook/types.go", "content": "package hook\n\nimport (\n\tv8 \"github.com/yaoapp/gou/runtime/v8\"\n)\n\n// Script the script hook align\ntype Script struct {\n\t*v8.Script\n}\n" }, { "path": "agent/assistant/llm.go", "content": "package assistant\n\nimport (\n\t\"github.com/yaoapp/yao/agent/context\"\n\t\"github.com/yaoapp/yao/agent/llm\"\n\t\"github.com/yaoapp/yao/agent/output/message\"\n\t\"github.com/yaoapp/yao/trace/types\"\n)\n\n// executeLLMStream executes the LLM streaming call with pre-built request\n// Returns completionResponse and error\nfunc (ast *Assistant) executeLLMStream(\n\tctx *context.Context,\n\tcompletionMessages []context.Message,\n\tcompletionOptions *context.CompletionOptions,\n\tagentNode types.Node,\n\tstreamHandler message.StreamFunc,\n\topts *context.Options,\n) (*context.CompletionResponse, error) {\n\n\t// Get connector object (capabilities were already set above, before stream_start)\n\tconn, capabilities, err := ast.GetConnector(ctx, opts)\n\tif err != nil {\n\t\tast.traceAgentFail(agentNode, err)\n\t\treturn nil, err\n\t}\n\n\t// Set capabilities in options if not already set\n\tif completionOptions.Capabilities == nil && capabilities != nil {\n\t\tcompletionOptions.Capabilities = capabilities\n\t}\n\n\t// Log the capabilities\n\tast.traceConnectorCapabilities(agentNode, capabilities)\n\n\t// Build content - convert extended types (file, data, __yao.attachment://) to standard LLM types\n\t// This is done here (right before LLM call) to ensure:\n\t// 1. autoSearch receives original messages (not converted)\n\t// 2. delegate receives original messages (not converted)\n\t// 3. Only the actual LLM call sees converted messages\n\tllmMessages, err := ast.BuildContent(ctx, completionMessages, completionOptions, opts)\n\tif err != nil {\n\t\tast.traceAgentFail(agentNode, err)\n\t\treturn nil, err\n\t}\n\n\t// Trace Add LLM request (use converted messages for trace)\n\tast.traceLLMRequest(ctx, conn.ID(), llmMessages, completionOptions)\n\n\t// Log LLM call start\n\tctx.Logger.LLMStart(conn.ID(), \"\", len(llmMessages))\n\n\t// Create LLM instance with connector and options\n\tllmInstance, err := llm.New(conn, completionOptions)\n\tif err != nil {\n\t\t// Mark LLM Request as failed in trace\n\t\tast.traceLLMFail(ctx, err)\n\t\treturn nil, err\n\t}\n\n\t// Call the LLM Completion Stream (streamHandler was set earlier)\n\t// Use llmMessages (converted) instead of completionMessages (original)\n\tcompletionResponse, err := llmInstance.Stream(ctx, llmMessages, completionOptions, streamHandler)\n\n\tif err != nil {\n\t\t// Mark LLM Request as failed in trace\n\t\tast.traceLLMFail(ctx, err)\n\t\treturn nil, err\n\t}\n\n\t// Mark LLM Request Complete\n\tast.traceLLMComplete(ctx, completionResponse)\n\n\treturn completionResponse, nil\n}\n\n// executeLLMForToolRetry executes LLM call for tool retry with streaming output\n// This is used when retrying tool calls - we still want to show LLM's response to users\n// Returns completionResponse and error\nfunc (ast *Assistant) executeLLMForToolRetry(\n\tctx *context.Context,\n\tcompletionMessages []context.Message,\n\tcompletionOptions *context.CompletionOptions,\n\tagentNode types.Node,\n\tstreamHandler message.StreamFunc,\n\topts *context.Options,\n) (*context.CompletionResponse, error) {\n\n\t// Get connector object\n\tconn, capabilities, err := ast.GetConnector(ctx, opts)\n\tif err != nil {\n\t\tast.traceAgentFail(agentNode, err)\n\t\treturn nil, err\n\t}\n\n\t// Set capabilities in options if not already set\n\tif completionOptions.Capabilities == nil && capabilities != nil {\n\t\tcompletionOptions.Capabilities = capabilities\n\t}\n\n\t// Build content - convert extended types for LLM call\n\tllmMessages, err := ast.BuildContent(ctx, completionMessages, completionOptions, opts)\n\tif err != nil {\n\t\tast.traceAgentFail(agentNode, err)\n\t\treturn nil, err\n\t}\n\n\t// Trace Add LLM retry request\n\tast.traceLLMRetryRequest(ctx, conn.ID(), llmMessages, completionOptions)\n\n\t// Log LLM call start (retry)\n\tctx.Logger.LLMStart(conn.ID(), \"\", len(llmMessages))\n\n\t// Create LLM instance with connector and options\n\tllmInstance, err := llm.New(conn, completionOptions)\n\tif err != nil {\n\t\t// Mark LLM Retry Request as failed in trace\n\t\tast.traceLLMFail(ctx, err)\n\t\treturn nil, err\n\t}\n\n\t// Call the LLM Completion Stream (still streaming for tool retry)\n\t// Use llmMessages (converted) instead of completionMessages (original)\n\tcompletionResponse, err := llmInstance.Stream(ctx, llmMessages, completionOptions, streamHandler)\n\tif err != nil {\n\t\t// Mark LLM Retry Request as failed in trace\n\t\tast.traceLLMFail(ctx, err)\n\t\treturn nil, err\n\t}\n\n\t// Mark LLM Request Complete\n\tast.traceLLMComplete(ctx, completionResponse)\n\n\treturn completionResponse, nil\n}\n" }, { "path": "agent/assistant/load.go", "content": "package assistant\n\nimport (\n\t\"fmt\"\n\t\"os\"\n\t\"path/filepath\"\n\t\"strings\"\n\n\tjsoniter \"github.com/json-iterator/go\"\n\t\"github.com/spf13/cast\"\n\t\"github.com/yaoapp/gou/application\"\n\t\"github.com/yaoapp/gou/fs\"\n\t\"github.com/yaoapp/yao/agent/context\"\n\t\"github.com/yaoapp/yao/agent/i18n\"\n\tsandboxTypes \"github.com/yaoapp/yao/agent/sandbox/v2/types\"\n\tsearchTypes \"github.com/yaoapp/yao/agent/search/types\"\n\tstore \"github.com/yaoapp/yao/agent/store/types\"\n\t\"github.com/yaoapp/yao/config\"\n\t\"gopkg.in/yaml.v3\"\n)\n\n// loaded the loaded assistant\nvar loaded = NewCache(200) // 200 is the default capacity\nvar storage store.Store = nil\nvar storeSetting *store.Setting = nil // store setting from agent.yml\nvar defaultConnector string = \"\" // default connector\nvar globalUses *context.Uses = nil // global uses configuration from agent.yml\nvar globalPrompts []store.Prompt = nil // global prompts from agent/prompts.yml\nvar globalKBSetting *store.KBSetting = nil // global KB setting from agent/kb.yml\nvar globalSearchConfig *searchTypes.Config = nil // global search config from agent/search.yml\n\n// LoadBuiltIn load the built-in assistants\nfunc LoadBuiltIn() error {\n\n\t// Clear non-system agents from cache (preserve system agents loaded by LoadSystemAgents)\n\tloaded.ClearExcept(func(id string) bool {\n\t\treturn strings.HasPrefix(id, \"__yao.\") // Keep system agents\n\t})\n\n\troot := `/assistants`\n\tapp, err := fs.Get(\"app\")\n\tif err != nil {\n\t\treturn err\n\t}\n\n\t// Get all existing built-in assistants\n\tdeletedBuiltIn := map[string]bool{}\n\n\t// Remove the built-in assistants (exclude system agents with __yao. prefix)\n\tif storage != nil {\n\n\t\tbuiltIn := true\n\t\tres, err := storage.GetAssistants(store.AssistantFilter{BuiltIn: &builtIn, Select: []string{\"assistant_id\", \"id\"}})\n\t\tif err != nil {\n\t\t\treturn err\n\t\t}\n\n\t\t// Get all existing built-in assistants (exclude system agents)\n\t\tfor _, assistant := range res.Data {\n\t\t\t// Skip system agents (they are managed by LoadSystemAgents)\n\t\t\tif strings.HasPrefix(assistant.ID, \"__yao.\") {\n\t\t\t\tcontinue\n\t\t\t}\n\t\t\tdeletedBuiltIn[assistant.ID] = true\n\t\t}\n\t}\n\n\t// Check if the assistant is built-in\n\tif exists, _ := app.Exists(root); !exists {\n\t\treturn nil\n\t}\n\n\tpaths, err := app.ReadDir(root, true)\n\tif err != nil {\n\t\treturn err\n\t}\n\n\tsort := 1\n\tfor _, path := range paths {\n\t\tpkgfile := filepath.Join(path, \"package.yao\")\n\t\tif has, _ := app.Exists(pkgfile); !has {\n\t\t\tcontinue\n\t\t}\n\n\t\tassistant, err := LoadPath(path)\n\t\tif err != nil {\n\t\t\treturn err\n\t\t}\n\n\t\tassistant.Readonly = true\n\t\tassistant.BuiltIn = true\n\t\tif assistant.Sort == 0 {\n\t\t\tassistant.Sort = sort\n\t\t}\n\t\tif assistant.Tags == nil {\n\t\t\tassistant.Tags = []string{}\n\t\t}\n\n\t\t// Save the assistant\n\t\terr = assistant.Save()\n\t\tif err != nil {\n\t\t\treturn err\n\t\t}\n\n\t\t// Initialize the assistant\n\t\terr = assistant.initialize()\n\t\tif err != nil {\n\t\t\treturn err\n\t\t}\n\n\t\tsort++\n\t\tloaded.Put(assistant)\n\n\t\t// Remove the built-in assistant from the store\n\t\tdelete(deletedBuiltIn, assistant.ID)\n\t}\n\n\t// Remove deleted built-in assistants\n\tif len(deletedBuiltIn) > 0 {\n\t\tassistantIDs := []string{}\n\t\tfor assistantID := range deletedBuiltIn {\n\t\t\tassistantIDs = append(assistantIDs, assistantID)\n\t\t}\n\t\t_, err := storage.DeleteAssistants(store.AssistantFilter{AssistantIDs: assistantIDs})\n\t\tif err != nil {\n\t\t\treturn err\n\t\t}\n\t}\n\n\treturn nil\n}\n\n// SetStorage set the storage\nfunc SetStorage(s store.Store) {\n\tstorage = s\n}\n\n// GetStorage returns the storage (for testing purposes)\nfunc GetStorage() store.Store {\n\treturn storage\n}\n\n// SetConnector set the connector\nfunc SetConnector(c string) {\n\tdefaultConnector = c\n}\n\n// SetGlobalUses set the global uses configuration\nfunc SetGlobalUses(uses *context.Uses) {\n\tglobalUses = uses\n}\n\n// SetGlobalPrompts set the global prompts from agent/prompts.yml\nfunc SetGlobalPrompts(prompts []store.Prompt) {\n\tglobalPrompts = prompts\n}\n\n// SetStoreSetting set the store setting from agent.yml\nfunc SetStoreSetting(setting *store.Setting) {\n\tstoreSetting = setting\n}\n\n// GetStoreSetting returns the store setting\nfunc GetStoreSetting() *store.Setting {\n\treturn storeSetting\n}\n\n// GetGlobalPrompts returns the global prompts with variables parsed\n// ctx: context variables for parsing $CTX.* variables\nfunc GetGlobalPrompts(ctx map[string]string) []store.Prompt {\n\tif len(globalPrompts) == 0 {\n\t\treturn nil\n\t}\n\treturn store.Prompts(globalPrompts).Parse(ctx)\n}\n\n// SetGlobalKBSetting set the global KB setting from agent/kb.yml\nfunc SetGlobalKBSetting(kbSetting *store.KBSetting) {\n\tglobalKBSetting = kbSetting\n}\n\n// GetGlobalKBSetting returns the global KB setting\nfunc GetGlobalKBSetting() *store.KBSetting {\n\treturn globalKBSetting\n}\n\n// SetGlobalSearchConfig set the global search config from agent/search.yml\nfunc SetGlobalSearchConfig(config *searchTypes.Config) {\n\tglobalSearchConfig = config\n}\n\n// GetGlobalSearchConfig returns the global search config\nfunc GetGlobalSearchConfig() *searchTypes.Config {\n\treturn globalSearchConfig\n}\n\n// SetCache set the cache\nfunc SetCache(capacity int) {\n\tClearCache()\n\tloaded = NewCache(capacity)\n}\n\n// ClearCache clear the cache\nfunc ClearCache() {\n\tif loaded != nil {\n\t\tloaded.Clear()\n\t\tloaded = nil\n\t}\n}\n\n// GetCache returns the loaded cache\nfunc GetCache() *Cache {\n\treturn loaded\n}\n\n// LoadStore create a new assistant from store\nfunc LoadStore(id string) (*Assistant, error) {\n\n\tif id == \"\" {\n\t\treturn nil, fmt.Errorf(\"assistant_id is required\")\n\t}\n\n\tassistant, exists := loaded.Get(id)\n\tif exists {\n\t\treturn assistant, nil\n\t}\n\n\tif storage == nil {\n\t\treturn nil, fmt.Errorf(\"storage is not set\")\n\t}\n\n\t// Request all fields when loading assistant from store\n\tstoreModel, err := storage.GetAssistant(id, store.AssistantFullFields)\n\tif err != nil {\n\t\treturn nil, err\n\t}\n\n\t// Load from path\n\tif storeModel.Path != \"\" {\n\t\tassistant, err = LoadPath(storeModel.Path)\n\t\tif err != nil {\n\t\t\treturn nil, err\n\t\t}\n\t\tloaded.Put(assistant)\n\t\treturn assistant, nil\n\t}\n\n\t// Create assistant from store model\n\tassistant = &Assistant{AssistantModel: *storeModel}\n\n\t// Load script from source field if present\n\tif assistant.Source != \"\" {\n\t\tscript, err := loadSource(assistant.Source, assistant.ID)\n\t\tif err != nil {\n\t\t\treturn nil, err\n\t\t}\n\t\tassistant.HookScript = script\n\t}\n\n\t// Initialize the assistant\n\terr = assistant.initialize()\n\tif err != nil {\n\t\treturn nil, err\n\t}\n\n\tloaded.Put(assistant)\n\treturn assistant, nil\n}\n\n// loadPackage loads and parses the package.yao file\nfunc loadPackage(path string) (map[string]interface{}, error) {\n\tapp, err := fs.Get(\"app\")\n\tif err != nil {\n\t\treturn nil, err\n\t}\n\n\tpkgfile := filepath.Join(path, \"package.yao\")\n\tif has, _ := app.Exists(pkgfile); !has {\n\t\treturn nil, fmt.Errorf(\"package.yao not found in %s\", path)\n\t}\n\n\tpkgraw, err := app.ReadFile(pkgfile)\n\tif err != nil {\n\t\treturn nil, err\n\t}\n\n\tvar data map[string]interface{}\n\terr = application.Parse(pkgfile, pkgraw, &data)\n\tif err != nil {\n\t\treturn nil, err\n\t}\n\n\t// Process connector environment variable\n\tif connector, ok := data[\"connector\"].(string); ok {\n\t\tif strings.HasPrefix(connector, \"$ENV.\") {\n\t\t\tenvKey := strings.TrimPrefix(connector, \"$ENV.\")\n\t\t\tif envValue := os.Getenv(envKey); envValue != \"\" {\n\t\t\t\tdata[\"connector\"] = envValue\n\t\t\t}\n\t\t}\n\t}\n\n\treturn data, nil\n}\n\n// LoadPath load assistant from path\nfunc LoadPath(path string) (*Assistant, error) {\n\tapp, err := fs.Get(\"app\")\n\tif err != nil {\n\t\treturn nil, err\n\t}\n\n\tdata, err := loadPackage(path)\n\tif err != nil {\n\t\treturn nil, err\n\t}\n\n\t// assistant_id\n\tid := strings.ReplaceAll(strings.TrimPrefix(path, \"/assistants/\"), \"/\", \".\")\n\tdata[\"assistant_id\"] = id\n\tdata[\"path\"] = path\n\tif _, has := data[\"type\"]; !has {\n\t\tdata[\"type\"] = \"assistant\"\n\t}\n\n\tupdatedAt := int64(0)\n\n\t// prompts (default prompts from prompts.yml)\n\tpromptsfile := filepath.Join(path, \"prompts.yml\")\n\tif has, _ := app.Exists(promptsfile); has {\n\t\tprompts, ts, err := store.LoadPrompts(promptsfile, path)\n\t\tif err != nil {\n\t\t\treturn nil, err\n\t\t}\n\t\tdata[\"prompts\"] = prompts\n\t\tdata[\"updated_at\"] = ts\n\t\tupdatedAt = ts\n\t}\n\n\t// prompt_presets (from prompts directory, key is filename without extension)\n\tpromptsDir := filepath.Join(path, \"prompts\")\n\tif has, _ := app.Exists(promptsDir); has {\n\t\tpresets, ts, err := store.LoadPromptPresets(promptsDir, path)\n\t\tif err != nil {\n\t\t\treturn nil, err\n\t\t}\n\t\tif len(presets) > 0 {\n\t\t\tdata[\"prompt_presets\"] = presets\n\t\t\tupdatedAt = max(updatedAt, ts)\n\t\t}\n\t}\n\n\t// load scripts (hook script and other scripts) from src directory\n\tsrcDir := filepath.Join(path, \"src\")\n\tif has, _ := app.Exists(srcDir); has {\n\t\thookScript, scripts, err := LoadScripts(srcDir)\n\t\tif err != nil {\n\t\t\treturn nil, err\n\t\t}\n\n\t\t// Set hook script and update timestamp\n\t\tif hookScript != nil {\n\t\t\tdata[\"script\"] = hookScript\n\t\t\t// Get timestamp from index.ts if exists\n\t\t\tscriptfile := filepath.Join(srcDir, \"index.ts\")\n\t\t\tif ts, err := app.ModTime(scriptfile); err == nil {\n\t\t\t\tdata[\"updated_at\"] = max(updatedAt, ts.UnixNano())\n\t\t\t}\n\t\t}\n\n\t\t// Set other scripts\n\t\tif len(scripts) > 0 {\n\t\t\tdata[\"scripts\"] = scripts\n\t\t}\n\t}\n\n\t// i18ns\n\tlocales, err := i18n.GetLocales(path)\n\tif err != nil {\n\t\treturn nil, err\n\t}\n\tdata[\"locales\"] = locales\n\n\t// V2 sandbox: load standalone sandbox.yao if present (Path A).\n\tsandboxFile := filepath.Join(path, \"sandbox.yao\")\n\tif has, _ := app.Exists(sandboxFile); has {\n\t\tabsFile := filepath.Join(config.Conf.AppSource, sandboxFile)\n\t\tsbCfg, sbErr := store.LoadSandboxConfig(absFile)\n\t\tif sbErr != nil {\n\t\t\treturn nil, fmt.Errorf(\"load sandbox.yao: %w\", sbErr)\n\t\t}\n\t\tdata[\"__sandbox_v2\"] = sbCfg\n\t}\n\n\tast, err := loadMap(data)\n\tif err != nil {\n\t\treturn nil, err\n\t}\n\n\t// If V2 sandbox was loaded via Path A, assign it now.\n\tif sbCfg, ok := data[\"__sandbox_v2\"].(*sandboxTypes.SandboxConfig); ok && sbCfg != nil {\n\t\tast.SandboxV2 = sbCfg\n\t}\n\n\t// Extract Sandbox flag and ComputerFilter from V2 sandbox config.\n\tif ast.SandboxV2 != nil {\n\t\tast.IsSandbox = true\n\t\tast.ComputerFilter = ast.SandboxV2.Filter\n\t}\n\n\t// Compute config hash for V2 sandbox.\n\tif ast.SandboxV2 != nil {\n\t\tvar mcpServers []store.MCPServerConfig\n\t\tif ast.MCP != nil {\n\t\t\tmcpServers = ast.MCP.Servers\n\t\t}\n\t\tskillsDir := \"\"\n\t\tif ast.Path != \"\" {\n\t\t\tdir := filepath.Join(config.Conf.AppSource, ast.Path, \"skills\")\n\t\t\tif info, e := os.Stat(dir); e == nil && info.IsDir() {\n\t\t\t\tskillsDir = dir\n\t\t\t}\n\t\t}\n\t\tast.ConfigHash = store.ComputeConfigHash(ast.SandboxV2, mcpServers, skillsDir)\n\t}\n\n\treturn ast, nil\n}\n\nfunc loadMap(data map[string]interface{}) (*Assistant, error) {\n\n\tassistant := &Assistant{}\n\n\t// assistant_id is required\n\tid, ok := data[\"assistant_id\"].(string)\n\tif !ok {\n\t\treturn nil, fmt.Errorf(\"assistant_id is required\")\n\t}\n\tassistant.ID = id\n\n\t// name is required\n\tname, ok := data[\"name\"].(string)\n\tif !ok {\n\t\treturn nil, fmt.Errorf(\"name is required\")\n\t}\n\tassistant.Name = name\n\n\t// avatar\n\tif avatar, ok := data[\"avatar\"].(string); ok {\n\t\tassistant.Avatar = avatar\n\t}\n\n\t// Type\n\tif v, ok := data[\"type\"].(string); ok {\n\t\tassistant.Type = v\n\t}\n\n\t// Placeholder\n\tif v, ok := data[\"placeholder\"]; ok {\n\n\t\tswitch vv := v.(type) {\n\t\tcase string:\n\t\t\tplaceholder, err := jsoniter.Marshal(vv)\n\t\t\tif err != nil {\n\t\t\t\treturn nil, err\n\t\t\t}\n\t\t\tassistant.Placeholder = &store.Placeholder{}\n\t\t\terr = jsoniter.Unmarshal(placeholder, assistant.Placeholder)\n\t\t\tif err != nil {\n\t\t\t\treturn nil, err\n\t\t\t}\n\n\t\tcase map[string]interface{}:\n\t\t\traw, err := jsoniter.Marshal(vv)\n\t\t\tif err != nil {\n\t\t\t\treturn nil, err\n\t\t\t}\n\n\t\t\tassistant.Placeholder = &store.Placeholder{}\n\t\t\terr = jsoniter.Unmarshal(raw, assistant.Placeholder)\n\t\t\tif err != nil {\n\t\t\t\treturn nil, err\n\t\t\t}\n\n\t\tcase *store.Placeholder:\n\t\t\tassistant.Placeholder = vv\n\n\t\tcase nil:\n\t\t\tassistant.Placeholder = nil\n\t\t}\n\t}\n\n\t// Mentionable\n\tif v, ok := data[\"mentionable\"].(bool); ok {\n\t\tassistant.Mentionable = v\n\t}\n\n\t// Automated\n\tif v, ok := data[\"automated\"].(bool); ok {\n\t\tassistant.Automated = v\n\t}\n\n\t// modes\n\tif v, has := data[\"modes\"]; has {\n\t\tmodes, err := store.ToModes(v)\n\t\tif err != nil {\n\t\t\treturn nil, err\n\t\t}\n\t\tassistant.Modes = modes\n\t}\n\n\t// default_mode\n\tif v, ok := data[\"default_mode\"].(string); ok {\n\t\tassistant.DefaultMode = v\n\t}\n\n\t// DisableGlobalPrompts\n\tif v, ok := data[\"disable_global_prompts\"].(bool); ok {\n\t\tassistant.DisableGlobalPrompts = v\n\t}\n\n\t// Readonly\n\tif v, ok := data[\"readonly\"].(bool); ok {\n\t\tassistant.Readonly = v\n\t}\n\n\t// Public\n\tif v, ok := data[\"public\"].(bool); ok {\n\t\tassistant.Public = v\n\t}\n\n\t// Share\n\tif v, ok := data[\"share\"].(string); ok {\n\t\tassistant.Share = v\n\t}\n\n\t// built_in\n\tif v, ok := data[\"built_in\"].(bool); ok {\n\t\tassistant.BuiltIn = v\n\t}\n\n\t// sort\n\tif v, has := data[\"sort\"]; has {\n\t\tassistant.Sort = cast.ToInt(v)\n\t}\n\n\t// path\n\tif v, ok := data[\"path\"].(string); ok {\n\t\tassistant.Path = v\n\t}\n\n\t// connector\n\tif connector, ok := data[\"connector\"].(string); ok {\n\t\tassistant.Connector = connector\n\t}\n\n\t// connector_options\n\tif connOpts, has := data[\"connector_options\"]; has {\n\t\topts, err := store.ToConnectorOptions(connOpts)\n\t\tif err != nil {\n\t\t\treturn nil, err\n\t\t}\n\t\tassistant.ConnectorOptions = opts\n\t}\n\n\t// tags\n\tif v, has := data[\"tags\"]; has {\n\t\tswitch vv := v.(type) {\n\t\tcase []string:\n\t\t\tassistant.Tags = vv\n\t\tcase []interface{}:\n\t\t\tvar tags []string\n\t\t\tfor _, tag := range vv {\n\t\t\t\ttags = append(tags, cast.ToString(tag))\n\t\t\t}\n\t\t\tassistant.Tags = tags\n\n\t\tcase string:\n\t\t\tassistant.Tags = []string{vv}\n\n\t\tcase interface{}:\n\t\t\traw, err := jsoniter.Marshal(vv)\n\t\t\tif err != nil {\n\t\t\t\treturn nil, err\n\t\t\t}\n\t\t\tvar tags []string\n\t\t\terr = jsoniter.Unmarshal(raw, &tags)\n\t\t\tif err != nil {\n\t\t\t\treturn nil, err\n\t\t\t}\n\t\t\tassistant.Tags = tags\n\n\t\t}\n\t}\n\n\t// options\n\tif v, ok := data[\"options\"].(map[string]interface{}); ok {\n\t\tassistant.Options = v\n\t}\n\n\t// description\n\tif v, ok := data[\"description\"].(string); ok {\n\t\tassistant.Description = v\n\t}\n\n\t// capabilities\n\tif v, ok := data[\"capabilities\"].(string); ok {\n\t\tassistant.Capabilities = v\n\t}\n\n\t// locales\n\tif locales, ok := data[\"locales\"].(i18n.Map); ok {\n\t\tassistant.Locales = locales\n\t\tflattened := locales.FlattenWithGlobal()\n\n\t\t// Auto-inject assistant name and description into all locales\n\t\t// so that {{name}} and {{description}} templates can be resolved\n\t\tfor locale, i18nObj := range flattened {\n\t\t\tif i18nObj.Messages == nil {\n\t\t\t\ti18nObj.Messages = make(map[string]any)\n\t\t\t}\n\t\t\t// Add name, description, and capabilities if not already present\n\t\t\tif _, exists := i18nObj.Messages[\"name\"]; !exists && assistant.Name != \"\" {\n\t\t\t\ti18nObj.Messages[\"name\"] = assistant.Name\n\t\t\t}\n\t\t\tif _, exists := i18nObj.Messages[\"description\"]; !exists && assistant.Description != \"\" {\n\t\t\t\ti18nObj.Messages[\"description\"] = assistant.Description\n\t\t\t}\n\t\t\tif _, exists := i18nObj.Messages[\"capabilities\"]; !exists && assistant.Capabilities != \"\" {\n\t\t\t\ti18nObj.Messages[\"capabilities\"] = assistant.Capabilities\n\t\t\t}\n\t\t\tflattened[locale] = i18nObj\n\t\t}\n\n\t\ti18n.Locales[id] = flattened\n\t} else {\n\t\t// No locales defined, create default with name, description, and capabilities for all common locales\n\t\tif assistant.Name != \"\" || assistant.Description != \"\" || assistant.Capabilities != \"\" {\n\t\t\tdefaultLocales := make(map[string]i18n.I18n)\n\t\t\tcommonLocales := []string{\"en\", \"en-us\", \"zh\", \"zh-cn\", \"zh-tw\"}\n\t\t\tfor _, locale := range commonLocales {\n\t\t\t\tmessages := map[string]any{}\n\t\t\t\tif assistant.Name != \"\" {\n\t\t\t\t\tmessages[\"name\"] = assistant.Name\n\t\t\t\t}\n\t\t\t\tif assistant.Description != \"\" {\n\t\t\t\t\tmessages[\"description\"] = assistant.Description\n\t\t\t\t}\n\t\t\t\tif assistant.Capabilities != \"\" {\n\t\t\t\t\tmessages[\"capabilities\"] = assistant.Capabilities\n\t\t\t\t}\n\t\t\t\tdefaultLocales[locale] = i18n.I18n{\n\t\t\t\t\tLocale: locale,\n\t\t\t\t\tMessages: messages,\n\t\t\t\t}\n\t\t\t}\n\t\t\ti18n.Locales[id] = defaultLocales\n\t\t}\n\t}\n\n\t// Search configuration (from package.yao search block)\n\t// This contains search options like web.max_results, kb.threshold, citation.format, etc.\n\t// Merge hierarchy: global config < assistant config\n\tswitch v := data[\"search\"].(type) {\n\n\tcase *searchTypes.Config:\n\t\tassistant.Search = v\n\n\tcase searchTypes.Config:\n\t\tassistant.Search = &v\n\n\tcase map[string]interface{}:\n\t\tvar assistantSearch searchTypes.Config\n\t\traw, err := jsoniter.Marshal(v)\n\t\tif err != nil {\n\t\t\treturn nil, err\n\t\t}\n\t\terr = jsoniter.Unmarshal(raw, &assistantSearch)\n\t\tif err != nil {\n\t\t\treturn nil, err\n\t\t}\n\t\t// Merge with global search config\n\t\tassistant.Search = mergeSearchConfig(globalSearchConfig, &assistantSearch)\n\n\tdefault:\n\t\tassistant.Search = globalSearchConfig\n\t}\n\n\t// prompts\n\tif prompts, has := data[\"prompts\"]; has {\n\n\t\tswitch v := prompts.(type) {\n\t\tcase []store.Prompt:\n\t\t\tassistant.Prompts = v\n\n\t\tcase string:\n\t\t\tvar prompts []store.Prompt\n\t\t\terr := yaml.Unmarshal([]byte(v), &prompts)\n\t\t\tif err != nil {\n\t\t\t\treturn nil, err\n\t\t\t}\n\t\t\tassistant.Prompts = prompts\n\n\t\tdefault:\n\t\t\traw, err := jsoniter.Marshal(v)\n\t\t\tif err != nil {\n\t\t\t\treturn nil, err\n\t\t\t}\n\n\t\t\tvar prompts []store.Prompt\n\t\t\terr = jsoniter.Unmarshal(raw, &prompts)\n\t\t\tif err != nil {\n\t\t\t\treturn nil, err\n\t\t\t}\n\t\t\tassistant.Prompts = prompts\n\t\t}\n\t}\n\n\t// prompt_presets\n\tif presets, has := data[\"prompt_presets\"]; has {\n\t\tpromptPresets, err := store.ToPromptPresets(presets)\n\t\tif err != nil {\n\t\t\treturn nil, err\n\t\t}\n\t\tassistant.PromptPresets = promptPresets\n\t}\n\n\t// source (hook script code) - store the source code\n\tif source, ok := data[\"source\"].(string); ok {\n\t\tassistant.Source = source\n\t}\n\n\t// kb\n\tif kb, has := data[\"kb\"]; has {\n\t\tknowledgeBase, err := store.ToKnowledgeBase(kb)\n\t\tif err != nil {\n\t\t\treturn nil, err\n\t\t}\n\t\tassistant.KB = knowledgeBase\n\t}\n\n\t// db\n\tif db, has := data[\"db\"]; has {\n\t\tdatabase, err := store.ToDatabase(db)\n\t\tif err != nil {\n\t\t\treturn nil, err\n\t\t}\n\t\tassistant.DB = database\n\t}\n\n\t// mcp\n\tif mcp, has := data[\"mcp\"]; has {\n\t\tmcpServers, err := store.ToMCPServers(mcp)\n\t\tif err != nil {\n\t\t\treturn nil, err\n\t\t}\n\t\tassistant.MCP = mcpServers\n\t}\n\n\t// workflow\n\tif workflow, has := data[\"workflow\"]; has {\n\t\twf, err := store.ToWorkflow(workflow)\n\t\tif err != nil {\n\t\t\treturn nil, err\n\t\t}\n\t\tassistant.Workflow = wf\n\t}\n\n\t// sandbox (for coding agents like Claude CLI, Cursor CLI)\n\t// V2 sandbox via independent sandbox.yao is loaded in LoadPath (below).\n\t// This block handles the package.yao embedded \"sandbox\" field with version dispatch.\n\tif assistant.SandboxV2 == nil {\n\t\tif sandbox, has := data[\"sandbox\"]; has {\n\t\t\tversion := extractSandboxVersion(sandbox)\n\t\t\tif version == sandboxTypes.SandboxVersionV2 {\n\t\t\t\tsb, err := store.ToSandboxV2(sandbox)\n\t\t\t\tif err != nil {\n\t\t\t\t\treturn nil, err\n\t\t\t\t}\n\t\t\t\tassistant.SandboxV2 = sb\n\t\t\t\tassistant.IsSandbox = true\n\t\t\t\tassistant.ComputerFilter = sb.Filter\n\t\t\t} else {\n\t\t\t\tsb, err := store.ToSandbox(sandbox)\n\t\t\t\tif err != nil {\n\t\t\t\t\treturn nil, err\n\t\t\t\t}\n\t\t\t\tassistant.Sandbox = sb\n\t\t\t}\n\t\t}\n\t}\n\n\t// dependencies (name -> version constraint, like npm dependencies)\n\tif deps, has := data[\"dependencies\"]; has {\n\t\tswitch v := deps.(type) {\n\t\tcase map[string]string:\n\t\t\tassistant.Dependencies = v\n\t\tcase map[string]interface{}:\n\t\t\td := make(map[string]string, len(v))\n\t\t\tfor k, val := range v {\n\t\t\t\td[k] = cast.ToString(val)\n\t\t\t}\n\t\t\tassistant.Dependencies = d\n\t\tdefault:\n\t\t\traw, err := jsoniter.Marshal(v)\n\t\t\tif err != nil {\n\t\t\t\treturn nil, err\n\t\t\t}\n\t\t\tvar d map[string]string\n\t\t\tif err := jsoniter.Unmarshal(raw, &d); err != nil {\n\t\t\t\treturn nil, err\n\t\t\t}\n\t\t\tassistant.Dependencies = d\n\t\t}\n\t}\n\n\t// uses (wrapper configurations for vision, audio, etc.)\n\t// Merge hierarchy: global uses < assistant uses\n\tif uses, has := data[\"uses\"]; has {\n\t\tvar assistantUses *context.Uses\n\t\tswitch v := uses.(type) {\n\t\tcase *context.Uses:\n\t\t\tassistantUses = v\n\t\tcase context.Uses:\n\t\t\tassistantUses = &v\n\t\tdefault:\n\t\t\traw, err := jsoniter.Marshal(v)\n\t\t\tif err != nil {\n\t\t\t\treturn nil, err\n\t\t\t}\n\t\t\tvar usesConfig context.Uses\n\t\t\terr = jsoniter.Unmarshal(raw, &usesConfig)\n\t\t\tif err != nil {\n\t\t\t\treturn nil, err\n\t\t\t}\n\t\t\tassistantUses = &usesConfig\n\t\t}\n\t\t// Merge with global uses\n\t\tassistant.Uses = mergeUses(globalUses, assistantUses)\n\t} else if globalUses != nil {\n\t\t// No assistant-specific uses, use global\n\t\tassistant.Uses = globalUses\n\t}\n\n\t// Load scripts (hook script and other scripts)\n\thookScript, scripts, scriptErr := LoadScriptsFromData(data, assistant.ID)\n\tif scriptErr != nil {\n\t\treturn nil, scriptErr\n\t}\n\tassistant.HookScript = hookScript\n\tassistant.Scripts = scripts\n\n\t// created_at\n\tif v, has := data[\"created_at\"]; has {\n\t\tts, err := getTimestamp(v)\n\t\tif err != nil {\n\t\t\treturn nil, err\n\t\t}\n\t\tassistant.CreatedAt = ts\n\t}\n\n\t// updated_at\n\tif v, has := data[\"updated_at\"]; has {\n\t\tts, err := getTimestamp(v)\n\t\tif err != nil {\n\t\t\treturn nil, err\n\t\t}\n\t\tassistant.UpdatedAt = ts\n\t}\n\n\t// Initialize the assistant\n\terr := assistant.initialize()\n\tif err != nil {\n\t\treturn nil, err\n\t}\n\n\treturn assistant, nil\n}\n\n// Init init the assistant\n// Choose the connector and initialize the assistant\nfunc (ast *Assistant) initialize() error {\n\n\tconn := defaultConnector\n\tif ast.Connector != \"\" {\n\t\tconn = ast.Connector\n\t}\n\tast.Connector = conn\n\n\t// Register scripts as process handlers\n\tif len(ast.Scripts) > 0 {\n\t\tif err := ast.RegisterScripts(); err != nil {\n\t\t\treturn fmt.Errorf(\"failed to register scripts: %w\", err)\n\t\t}\n\t}\n\n\treturn nil\n}\n\n// mergeUses merges two Uses configs (base < override)\nfunc mergeUses(base, override *context.Uses) *context.Uses {\n\tif base == nil {\n\t\treturn override\n\t}\n\tif override == nil {\n\t\treturn base\n\t}\n\n\tresult := *base // Copy base\n\n\t// Override with non-empty values\n\tif override.Vision != \"\" {\n\t\tresult.Vision = override.Vision\n\t}\n\tif override.Audio != \"\" {\n\t\tresult.Audio = override.Audio\n\t}\n\tif override.Search != \"\" {\n\t\tresult.Search = override.Search\n\t}\n\tif override.Fetch != \"\" {\n\t\tresult.Fetch = override.Fetch\n\t}\n\tif override.Web != \"\" {\n\t\tresult.Web = override.Web\n\t}\n\tif override.Keyword != \"\" {\n\t\tresult.Keyword = override.Keyword\n\t}\n\tif override.QueryDSL != \"\" {\n\t\tresult.QueryDSL = override.QueryDSL\n\t}\n\tif override.Rerank != \"\" {\n\t\tresult.Rerank = override.Rerank\n\t}\n\n\treturn &result\n}\n\n// mergeSearchConfig merges two search configs (base < override)\nfunc mergeSearchConfig(base, override *searchTypes.Config) *searchTypes.Config {\n\tif base == nil {\n\t\treturn override\n\t}\n\tif override == nil {\n\t\treturn base\n\t}\n\n\tresult := *base // Copy base\n\n\t// Merge Web config\n\tif override.Web != nil {\n\t\tif result.Web == nil {\n\t\t\tresult.Web = override.Web\n\t\t} else {\n\t\t\tmerged := *result.Web\n\t\t\tif override.Web.Provider != \"\" {\n\t\t\t\tmerged.Provider = override.Web.Provider\n\t\t\t}\n\t\t\tif override.Web.APIKeyEnv != \"\" {\n\t\t\t\tmerged.APIKeyEnv = override.Web.APIKeyEnv\n\t\t\t}\n\t\t\tif override.Web.MaxResults > 0 {\n\t\t\t\tmerged.MaxResults = override.Web.MaxResults\n\t\t\t}\n\t\t\tresult.Web = &merged\n\t\t}\n\t}\n\n\t// Merge KB config\n\tif override.KB != nil {\n\t\tif result.KB == nil {\n\t\t\tresult.KB = override.KB\n\t\t} else {\n\t\t\tmerged := *result.KB\n\t\t\tif len(override.KB.Collections) > 0 {\n\t\t\t\tmerged.Collections = override.KB.Collections\n\t\t\t}\n\t\t\tif override.KB.Threshold > 0 {\n\t\t\t\tmerged.Threshold = override.KB.Threshold\n\t\t\t}\n\t\t\tif override.KB.Graph {\n\t\t\t\tmerged.Graph = override.KB.Graph\n\t\t\t}\n\t\t\tresult.KB = &merged\n\t\t}\n\t}\n\n\t// Merge DB config\n\tif override.DB != nil {\n\t\tif result.DB == nil {\n\t\t\tresult.DB = override.DB\n\t\t} else {\n\t\t\tmerged := *result.DB\n\t\t\tif len(override.DB.Models) > 0 {\n\t\t\t\tmerged.Models = override.DB.Models\n\t\t\t}\n\t\t\tif override.DB.MaxResults > 0 {\n\t\t\t\tmerged.MaxResults = override.DB.MaxResults\n\t\t\t}\n\t\t\tresult.DB = &merged\n\t\t}\n\t}\n\n\t// Merge Keyword config\n\tif override.Keyword != nil {\n\t\tif result.Keyword == nil {\n\t\t\tresult.Keyword = override.Keyword\n\t\t} else {\n\t\t\tmerged := *result.Keyword\n\t\t\tif override.Keyword.MaxKeywords > 0 {\n\t\t\t\tmerged.MaxKeywords = override.Keyword.MaxKeywords\n\t\t\t}\n\t\t\tif override.Keyword.Language != \"\" {\n\t\t\t\tmerged.Language = override.Keyword.Language\n\t\t\t}\n\t\t\tresult.Keyword = &merged\n\t\t}\n\t}\n\n\t// Merge QueryDSL config\n\tif override.QueryDSL != nil {\n\t\tif result.QueryDSL == nil {\n\t\t\tresult.QueryDSL = override.QueryDSL\n\t\t} else {\n\t\t\tmerged := *result.QueryDSL\n\t\t\tif override.QueryDSL.Strict {\n\t\t\t\tmerged.Strict = override.QueryDSL.Strict\n\t\t\t}\n\t\t\tresult.QueryDSL = &merged\n\t\t}\n\t}\n\n\t// Merge Rerank config\n\tif override.Rerank != nil {\n\t\tif result.Rerank == nil {\n\t\t\tresult.Rerank = override.Rerank\n\t\t} else {\n\t\t\tmerged := *result.Rerank\n\t\t\tif override.Rerank.TopN > 0 {\n\t\t\t\tmerged.TopN = override.Rerank.TopN\n\t\t\t}\n\t\t\tresult.Rerank = &merged\n\t\t}\n\t}\n\n\t// Merge Citation config\n\tif override.Citation != nil {\n\t\tif result.Citation == nil {\n\t\t\tresult.Citation = override.Citation\n\t\t} else {\n\t\t\tmerged := *result.Citation\n\t\t\tif override.Citation.Format != \"\" {\n\t\t\t\tmerged.Format = override.Citation.Format\n\t\t\t}\n\t\t\t// AutoInjectPrompt is a bool, so we check if it's explicitly set\n\t\t\t// by checking if the whole Citation block was provided\n\t\t\tmerged.AutoInjectPrompt = override.Citation.AutoInjectPrompt\n\t\t\tif override.Citation.CustomPrompt != \"\" {\n\t\t\t\tmerged.CustomPrompt = override.Citation.CustomPrompt\n\t\t\t}\n\t\t\tresult.Citation = &merged\n\t\t}\n\t}\n\n\t// Merge Weights config\n\tif override.Weights != nil {\n\t\tif result.Weights == nil {\n\t\t\tresult.Weights = override.Weights\n\t\t} else {\n\t\t\tmerged := *result.Weights\n\t\t\tif override.Weights.User > 0 {\n\t\t\t\tmerged.User = override.Weights.User\n\t\t\t}\n\t\t\tif override.Weights.Hook > 0 {\n\t\t\t\tmerged.Hook = override.Weights.Hook\n\t\t\t}\n\t\t\tif override.Weights.Auto > 0 {\n\t\t\t\tmerged.Auto = override.Weights.Auto\n\t\t\t}\n\t\t\tresult.Weights = &merged\n\t\t}\n\t}\n\n\t// Merge Options config\n\tif override.Options != nil {\n\t\tif result.Options == nil {\n\t\t\tresult.Options = override.Options\n\t\t} else {\n\t\t\tmerged := *result.Options\n\t\t\tif override.Options.SkipThreshold > 0 {\n\t\t\t\tmerged.SkipThreshold = override.Options.SkipThreshold\n\t\t\t}\n\t\t\tresult.Options = &merged\n\t\t}\n\t}\n\n\treturn &result\n}\n\n// extractSandboxVersion tries to read the \"version\" field from a sandbox config value.\nfunc extractSandboxVersion(v any) string {\n\tif m, ok := v.(map[string]any); ok {\n\t\tif ver, ok := m[\"version\"].(string); ok {\n\t\t\treturn ver\n\t\t}\n\t}\n\treturn \"\"\n}\n" }, { "path": "agent/assistant/load_merge_test.go", "content": "package assistant_test\n\nimport (\n\t\"testing\"\n\n\t\"github.com/stretchr/testify/assert\"\n\t\"github.com/stretchr/testify/require\"\n\t\"github.com/yaoapp/yao/agent/assistant\"\n\t\"github.com/yaoapp/yao/agent/testutils\"\n)\n\n// TestLoadPathMerge tests loading the merge test assistant\n// This verifies that global config is properly merged with assistant-specific config\nfunc TestLoadPathMerge(t *testing.T) {\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\tast, err := assistant.LoadPath(\"/assistants/tests/merge\")\n\trequire.NoError(t, err)\n\trequire.NotNil(t, ast)\n\n\tassert.Equal(t, \"tests.merge\", ast.ID)\n\tassert.Equal(t, \"Merge Config Test Assistant\", ast.Name)\n\n\t// Uses configuration - should merge global with assistant-specific\n\t// Global (from agent/agent.yml):\n\t// vision: \"workers.system.vision\"\n\t// search: \"workers.system.search\"\n\t// fetch: \"workers.system.fetch\"\n\t// audio: (not set)\n\t// querydsl: (not set)\n\t// rerank: (not set)\n\t// Assistant:\n\t// web: \"mcp:custom-web\"\n\t// keyword: \"mcp:custom-keyword\"\n\t// Result: assistant values override, global values inherited\n\tassert.NotNil(t, ast.Uses)\n\n\t// Assistant overrides\n\tassert.Equal(t, \"mcp:custom-web\", ast.Uses.Web) // overridden by assistant\n\tassert.Equal(t, \"mcp:custom-keyword\", ast.Uses.Keyword) // overridden by assistant\n\n\t// Inherited from global (agent/agent.yml)\n\tassert.Equal(t, \"workers.system.vision\", ast.Uses.Vision) // inherited from global\n\tassert.Equal(t, \"workers.system.search\", ast.Uses.Search) // inherited from global\n\tassert.Equal(t, \"workers.system.fetch\", ast.Uses.Fetch) // inherited from global\n\n\t// Not set in either global or assistant (should be empty)\n\tassert.Empty(t, ast.Uses.Audio) // not set anywhere\n\tassert.Empty(t, ast.Uses.QueryDSL) // not set anywhere\n\tassert.Empty(t, ast.Uses.Rerank) // not set anywhere\n\n\t// Search configuration - should merge global with assistant-specific\n\t// Global (from agent/search.yml):\n\t// web.provider=tavily, web.max_results=10\n\t// kb.threshold=0.7, kb.graph=false\n\t// db.max_results=20\n\t// keyword.max_keywords=10, keyword.language=auto\n\t// rerank.top_n=10\n\t// citation.format=#ref:{id}, citation.auto_inject_prompt=true\n\t// weights: user=1.0, hook=0.8, auto=0.6\n\t// options.skip_threshold=5\n\t// Assistant:\n\t// web.provider=custom-provider, web.max_results=25\n\t// kb.collections=[merge-test-kb], kb.threshold=0.85\n\tassert.NotNil(t, ast.Search)\n\n\t// Web config - assistant overrides global\n\tassert.NotNil(t, ast.Search.Web)\n\tassert.Equal(t, \"custom-provider\", ast.Search.Web.Provider) // overridden\n\tassert.Equal(t, 25, ast.Search.Web.MaxResults) // overridden\n\n\t// KB config - assistant overrides global\n\tassert.NotNil(t, ast.Search.KB)\n\tassert.Equal(t, []string{\"merge-test-kb\"}, ast.Search.KB.Collections) // overridden\n\tassert.Equal(t, 0.85, ast.Search.KB.Threshold) // overridden\n\tassert.False(t, ast.Search.KB.Graph) // inherited from global\n\n\t// DB config - should inherit from global (assistant doesn't define it)\n\tassert.NotNil(t, ast.Search.DB)\n\tassert.Equal(t, 20, ast.Search.DB.MaxResults) // inherited from global\n\n\t// Keyword config - should inherit from global\n\tassert.NotNil(t, ast.Search.Keyword)\n\tassert.Equal(t, 10, ast.Search.Keyword.MaxKeywords) // inherited from global\n\tassert.Equal(t, \"auto\", ast.Search.Keyword.Language) // inherited from global\n\n\t// Rerank config - should inherit from global\n\tassert.NotNil(t, ast.Search.Rerank)\n\tassert.Equal(t, 10, ast.Search.Rerank.TopN) // inherited from global\n\n\t// Citation config - should inherit from global\n\tassert.NotNil(t, ast.Search.Citation)\n\tassert.Equal(t, \"#ref:{id}\", ast.Search.Citation.Format) // inherited from global\n\tassert.True(t, ast.Search.Citation.AutoInjectPrompt) // inherited from global\n\n\t// Weights config - should inherit from global\n\tassert.NotNil(t, ast.Search.Weights)\n\tassert.Equal(t, 1.0, ast.Search.Weights.User) // inherited from global\n\tassert.Equal(t, 0.8, ast.Search.Weights.Hook) // inherited from global\n\tassert.Equal(t, 0.6, ast.Search.Weights.Auto) // inherited from global\n\n\t// Options config - should inherit from global\n\tassert.NotNil(t, ast.Search.Options)\n\tassert.Equal(t, 5, ast.Search.Options.SkipThreshold) // inherited from global\n}\n\n// TestLoadPathMergeOverride tests loading the merge-override test assistant\n// This verifies that assistant config completely overrides global config\nfunc TestLoadPathMergeOverride(t *testing.T) {\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\tast, err := assistant.LoadPath(\"/assistants/tests/merge-override\")\n\trequire.NoError(t, err)\n\trequire.NotNil(t, ast)\n\n\tassert.Equal(t, \"tests.merge-override\", ast.ID)\n\tassert.Equal(t, \"Merge Override Test Assistant\", ast.Name)\n\n\t// Uses configuration - all fields should be overridden by assistant\n\tassert.NotNil(t, ast.Uses)\n\tassert.Equal(t, \"mcp:custom-vision\", ast.Uses.Vision)\n\tassert.Equal(t, \"mcp:custom-audio\", ast.Uses.Audio)\n\tassert.Equal(t, \"mcp:custom-search\", ast.Uses.Search)\n\tassert.Equal(t, \"mcp:custom-fetch\", ast.Uses.Fetch)\n\tassert.Equal(t, \"mcp:custom-web\", ast.Uses.Web)\n\tassert.Equal(t, \"mcp:custom-keyword\", ast.Uses.Keyword)\n\tassert.Equal(t, \"mcp:custom-querydsl\", ast.Uses.QueryDSL)\n\tassert.Equal(t, \"mcp:custom-rerank\", ast.Uses.Rerank)\n\n\t// Search configuration - all fields should be overridden by assistant\n\tassert.NotNil(t, ast.Search)\n\n\t// Web config - all overridden\n\tassert.NotNil(t, ast.Search.Web)\n\tassert.Equal(t, \"override-provider\", ast.Search.Web.Provider)\n\tassert.Equal(t, \"$ENV.OVERRIDE_API_KEY\", ast.Search.Web.APIKeyEnv)\n\tassert.Equal(t, 100, ast.Search.Web.MaxResults)\n\n\t// KB config - all overridden\n\tassert.NotNil(t, ast.Search.KB)\n\tassert.Equal(t, []string{\"override-kb\"}, ast.Search.KB.Collections)\n\tassert.Equal(t, 0.99, ast.Search.KB.Threshold)\n\tassert.True(t, ast.Search.KB.Graph)\n\n\t// DB config - all overridden\n\tassert.NotNil(t, ast.Search.DB)\n\tassert.Equal(t, []string{\"override-model\"}, ast.Search.DB.Models)\n\tassert.Equal(t, 200, ast.Search.DB.MaxResults)\n\n\t// Keyword config - all overridden\n\tassert.NotNil(t, ast.Search.Keyword)\n\tassert.Equal(t, 20, ast.Search.Keyword.MaxKeywords)\n\tassert.Equal(t, \"zh\", ast.Search.Keyword.Language)\n\n\t// QueryDSL config - overridden\n\tassert.NotNil(t, ast.Search.QueryDSL)\n\tassert.True(t, ast.Search.QueryDSL.Strict)\n\n\t// Rerank config - overridden\n\tassert.NotNil(t, ast.Search.Rerank)\n\tassert.Equal(t, 20, ast.Search.Rerank.TopN)\n\n\t// Citation config - all overridden\n\tassert.NotNil(t, ast.Search.Citation)\n\tassert.Equal(t, \"[override:{id}]\", ast.Search.Citation.Format)\n\tassert.False(t, ast.Search.Citation.AutoInjectPrompt)\n\tassert.Equal(t, \"Override citation prompt\", ast.Search.Citation.CustomPrompt)\n\n\t// Weights config - all overridden\n\tassert.NotNil(t, ast.Search.Weights)\n\tassert.Equal(t, 2.0, ast.Search.Weights.User)\n\tassert.Equal(t, 1.5, ast.Search.Weights.Hook)\n\tassert.Equal(t, 1.0, ast.Search.Weights.Auto)\n\n\t// Options config - overridden\n\tassert.NotNil(t, ast.Search.Options)\n\tassert.Equal(t, 10, ast.Search.Options.SkipThreshold)\n}\n\n// TestLoadPathMergeEmpty tests loading the merge-empty test assistant\n// This verifies that assistant with no uses/search config inherits all from global\nfunc TestLoadPathMergeEmpty(t *testing.T) {\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\tast, err := assistant.LoadPath(\"/assistants/tests/merge-empty\")\n\trequire.NoError(t, err)\n\trequire.NotNil(t, ast)\n\n\tassert.Equal(t, \"tests.merge-empty\", ast.ID)\n\tassert.Equal(t, \"Merge Empty Test Assistant\", ast.Name)\n\n\t// Uses configuration - all inherited from global (agent/agent.yml)\n\tassert.NotNil(t, ast.Uses)\n\tassert.Equal(t, \"workers.system.vision\", ast.Uses.Vision) // from global\n\tassert.Equal(t, \"workers.system.search\", ast.Uses.Search) // from global\n\tassert.Equal(t, \"workers.system.fetch\", ast.Uses.Fetch) // from global\n\tassert.Empty(t, ast.Uses.Audio) // not set in global\n\tassert.Empty(t, ast.Uses.Web) // not set in global\n\tassert.Empty(t, ast.Uses.Keyword) // not set in global\n\tassert.Empty(t, ast.Uses.QueryDSL) // not set in global\n\tassert.Empty(t, ast.Uses.Rerank) // not set in global\n\n\t// Search configuration - all inherited from global (agent/search.yml)\n\tassert.NotNil(t, ast.Search)\n\n\t// Web config - from global\n\tassert.NotNil(t, ast.Search.Web)\n\tassert.Equal(t, \"tavily\", ast.Search.Web.Provider)\n\tassert.Equal(t, 10, ast.Search.Web.MaxResults)\n\n\t// KB config - from global\n\tassert.NotNil(t, ast.Search.KB)\n\tassert.Equal(t, 0.7, ast.Search.KB.Threshold)\n\tassert.False(t, ast.Search.KB.Graph)\n\n\t// DB config - from global\n\tassert.NotNil(t, ast.Search.DB)\n\tassert.Equal(t, 20, ast.Search.DB.MaxResults)\n\n\t// Keyword config - from global\n\tassert.NotNil(t, ast.Search.Keyword)\n\tassert.Equal(t, 10, ast.Search.Keyword.MaxKeywords)\n\tassert.Equal(t, \"auto\", ast.Search.Keyword.Language)\n\n\t// Rerank config - from global\n\tassert.NotNil(t, ast.Search.Rerank)\n\tassert.Equal(t, 10, ast.Search.Rerank.TopN)\n\n\t// Citation config - from global\n\tassert.NotNil(t, ast.Search.Citation)\n\tassert.Equal(t, \"#ref:{id}\", ast.Search.Citation.Format)\n\tassert.True(t, ast.Search.Citation.AutoInjectPrompt)\n\n\t// Weights config - from global\n\tassert.NotNil(t, ast.Search.Weights)\n\tassert.Equal(t, 1.0, ast.Search.Weights.User)\n\tassert.Equal(t, 0.8, ast.Search.Weights.Hook)\n\tassert.Equal(t, 0.6, ast.Search.Weights.Auto)\n\n\t// Options config - from global\n\tassert.NotNil(t, ast.Search.Options)\n\tassert.Equal(t, 5, ast.Search.Options.SkipThreshold)\n}\n\n// TestLoadPathUsesAndSearchMerge tests loading fullfields assistant\n// This verifies that uses and search configs are properly loaded and merged\nfunc TestLoadPathUsesAndSearchMerge(t *testing.T) {\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\tast, err := assistant.LoadPath(\"/assistants/tests/fullfields\")\n\trequire.NoError(t, err)\n\trequire.NotNil(t, ast)\n\n\t// Uses configuration - assistant-specific values\n\tassert.NotNil(t, ast.Uses)\n\tassert.Equal(t, \"agent\", ast.Uses.Vision)\n\tassert.Equal(t, \"mcp:audio-server\", ast.Uses.Audio)\n\tassert.Equal(t, \"agent\", ast.Uses.Fetch)\n\tassert.Equal(t, \"builtin\", ast.Uses.Web)\n\tassert.Equal(t, \"builtin\", ast.Uses.Keyword)\n\tassert.Equal(t, \"builtin\", ast.Uses.QueryDSL)\n\tassert.Equal(t, \"builtin\", ast.Uses.Rerank)\n\n\t// Search configuration - assistant-specific values\n\tassert.NotNil(t, ast.Search)\n\n\t// Web config - from assistant\n\tassert.NotNil(t, ast.Search.Web)\n\tassert.Equal(t, \"tavily\", ast.Search.Web.Provider)\n\tassert.Equal(t, 15, ast.Search.Web.MaxResults)\n\n\t// KB config - from assistant\n\tassert.NotNil(t, ast.Search.KB)\n\tassert.Equal(t, []string{\"docs\", \"faq\"}, ast.Search.KB.Collections)\n\tassert.Equal(t, 0.8, ast.Search.KB.Threshold)\n\tassert.True(t, ast.Search.KB.Graph)\n\n\t// DB config - from assistant\n\tassert.NotNil(t, ast.Search.DB)\n\tassert.Equal(t, []string{\"user\", \"product\"}, ast.Search.DB.Models)\n\tassert.Equal(t, 50, ast.Search.DB.MaxResults)\n\n\t// Citation config - from assistant\n\tassert.NotNil(t, ast.Search.Citation)\n\tassert.Equal(t, \"#ref:{id}\", ast.Search.Citation.Format)\n\tassert.True(t, ast.Search.Citation.AutoInjectPrompt)\n\n\t// Weights config - from assistant\n\tassert.NotNil(t, ast.Search.Weights)\n\tassert.Equal(t, 1.0, ast.Search.Weights.User)\n\tassert.Equal(t, 0.9, ast.Search.Weights.Hook)\n\tassert.Equal(t, 0.7, ast.Search.Weights.Auto)\n}\n\n// TestLoadPathSearchAssistant tests loading the dedicated search test assistant\nfunc TestLoadPathSearchAssistant(t *testing.T) {\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\tast, err := assistant.LoadPath(\"/assistants/tests/search\")\n\trequire.NoError(t, err)\n\trequire.NotNil(t, ast)\n\n\tassert.Equal(t, \"tests.search\", ast.ID)\n\tassert.Equal(t, \"Search Config Test Assistant\", ast.Name)\n\n\t// Uses configuration\n\tassert.NotNil(t, ast.Uses)\n\tassert.Equal(t, \"builtin\", ast.Uses.Web)\n\tassert.Equal(t, \"builtin\", ast.Uses.Keyword)\n\tassert.Equal(t, \"builtin\", ast.Uses.QueryDSL)\n\tassert.Equal(t, \"builtin\", ast.Uses.Rerank)\n\n\t// Search configuration\n\tassert.NotNil(t, ast.Search)\n\n\t// Web config\n\tassert.NotNil(t, ast.Search.Web)\n\tassert.Equal(t, \"serper\", ast.Search.Web.Provider)\n\tassert.Equal(t, \"$ENV.SERPER_API_KEY\", ast.Search.Web.APIKeyEnv)\n\tassert.Equal(t, 20, ast.Search.Web.MaxResults)\n\n\t// KB config\n\tassert.NotNil(t, ast.Search.KB)\n\tassert.Equal(t, []string{\"knowledge-base\", \"documents\"}, ast.Search.KB.Collections)\n\tassert.Equal(t, 0.75, ast.Search.KB.Threshold)\n\tassert.False(t, ast.Search.KB.Graph)\n\n\t// DB config\n\tassert.NotNil(t, ast.Search.DB)\n\tassert.Equal(t, []string{\"article\", \"comment\"}, ast.Search.DB.Models)\n\tassert.Equal(t, 30, ast.Search.DB.MaxResults)\n\n\t// Keyword config\n\tassert.NotNil(t, ast.Search.Keyword)\n\tassert.Equal(t, 8, ast.Search.Keyword.MaxKeywords)\n\tassert.Equal(t, \"auto\", ast.Search.Keyword.Language)\n\n\t// QueryDSL config\n\tassert.NotNil(t, ast.Search.QueryDSL)\n\tassert.True(t, ast.Search.QueryDSL.Strict)\n\n\t// Rerank config\n\tassert.NotNil(t, ast.Search.Rerank)\n\tassert.Equal(t, 5, ast.Search.Rerank.TopN)\n\n\t// Citation config\n\tassert.NotNil(t, ast.Search.Citation)\n\tassert.Equal(t, \"#cite:{id}\", ast.Search.Citation.Format)\n\tassert.False(t, ast.Search.Citation.AutoInjectPrompt)\n\tassert.Equal(t, \"Please cite sources using #cite:{id} format.\", ast.Search.Citation.CustomPrompt)\n\n\t// Weights config\n\tassert.NotNil(t, ast.Search.Weights)\n\tassert.Equal(t, 1.0, ast.Search.Weights.User)\n\tassert.Equal(t, 0.85, ast.Search.Weights.Hook)\n\tassert.Equal(t, 0.65, ast.Search.Weights.Auto)\n\n\t// Options config\n\tassert.NotNil(t, ast.Search.Options)\n\tassert.Equal(t, 3, ast.Search.Options.SkipThreshold)\n}\n" }, { "path": "agent/assistant/load_process_test.go", "content": "package assistant_test\n\nimport (\n\t\"testing\"\n\n\t\"github.com/stretchr/testify/assert\"\n\t\"github.com/yaoapp/gou/process\"\n\t\"github.com/yaoapp/yao/agent/testutils\"\n)\n\nfunc TestLoadProcessIntegration(t *testing.T) {\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\t// After testutils.Prepare, all assistants should be loaded and scripts registered\n\t// Test calling mcpload assistant's tools.Hello function\n\n\tt.Run(\"CallHelloAfterLoad\", func(t *testing.T) {\n\t\tproc := process.New(\"agents.tests.mcpload.tools.Hello\", map[string]interface{}{\n\t\t\t\"name\": \"TestUser\",\n\t\t})\n\n\t\terr := proc.Execute()\n\t\tassert.NoError(t, err)\n\n\t\tresult := proc.Value()\n\t\tassert.NotNil(t, result)\n\n\t\tresultStr, ok := result.(string)\n\t\tassert.True(t, ok, \"Result should be a string\")\n\t\tassert.Contains(t, resultStr, \"Hello, TestUser\")\n\t\tassert.Contains(t, resultStr, \"mcpload assistant\")\n\t})\n\n\tt.Run(\"CallPingAfterLoad\", func(t *testing.T) {\n\t\tproc := process.New(\"agents.tests.mcpload.tools.Ping\", map[string]interface{}{\n\t\t\t\"message\": \"integration test\",\n\t\t})\n\n\t\terr := proc.Execute()\n\t\tassert.NoError(t, err)\n\n\t\tresult := proc.Value()\n\t\tassert.NotNil(t, result)\n\n\t\tresultMap, ok := result.(map[string]interface{})\n\t\tassert.True(t, ok, \"Result should be a map\")\n\t\tassert.Equal(t, \"integration test\", resultMap[\"message\"])\n\t\tassert.Contains(t, resultMap[\"echo\"], \"Pong\")\n\t\tassert.NotEmpty(t, resultMap[\"timestamp\"])\n\t})\n\n\tt.Run(\"CallCalculateAfterLoad\", func(t *testing.T) {\n\t\tproc := process.New(\"agents.tests.mcpload.tools.Calculate\", map[string]interface{}{\n\t\t\t\"operation\": \"add\",\n\t\t\t\"a\": float64(100),\n\t\t\t\"b\": float64(50),\n\t\t})\n\n\t\terr := proc.Execute()\n\t\tassert.NoError(t, err)\n\n\t\tresult := proc.Value()\n\t\tassert.NotNil(t, result)\n\n\t\tresultMap, ok := result.(map[string]interface{})\n\t\tassert.True(t, ok, \"Result should be a map\")\n\t\tassert.Equal(t, float64(150), resultMap[\"result\"])\n\t\tassert.Equal(t, \"add\", resultMap[\"operation\"])\n\t\tassert.Equal(t, float64(100), resultMap[\"a\"])\n\t\tassert.Equal(t, float64(50), resultMap[\"b\"])\n\t})\n\n\tt.Run(\"CallNonExistentScript\", func(t *testing.T) {\n\t\tproc := process.New(\"agents.tests.mcpload.nonexistent.Method\")\n\n\t\terr := proc.Execute()\n\t\tassert.NotNil(t, err, \"Should return error for non-existent script\")\n\t\tassert.Contains(t, err.Error(), \"Exception|404\")\n\t})\n\n\tt.Run(\"CallNonExistentMethod\", func(t *testing.T) {\n\t\tproc := process.New(\"agents.tests.mcpload.tools.NonExistentMethod\")\n\n\t\terr := proc.Execute()\n\t\tassert.NotNil(t, err, \"Should return error for non-existent method\")\n\t\tassert.Contains(t, err.Error(), \"Exception|500\")\n\t})\n}\n\nfunc TestLoadProcessMultipleAssistants(t *testing.T) {\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\t// Test that multiple assistants can have their scripts registered\n\t// and process calls work correctly for different assistants\n\n\tt.Run(\"MCPLoadAssistant\", func(t *testing.T) {\n\t\tproc := process.New(\"agents.tests.mcpload.tools.Hello\", map[string]interface{}{\n\t\t\t\"name\": \"User1\",\n\t\t})\n\n\t\terr := proc.Execute()\n\t\tassert.NoError(t, err)\n\n\t\tresult := proc.Value()\n\t\tresultStr, ok := result.(string)\n\t\tassert.True(t, ok)\n\t\tassert.Contains(t, resultStr, \"mcpload assistant\")\n\t})\n\n\t// If there are other test assistants with scripts, they can be tested here\n\t// For now, we verify that the handler is properly isolated per assistant\n\tt.Run(\"VerifyIsolation\", func(t *testing.T) {\n\t\t// Verify that the mcpload handler is correctly registered\n\t\thandler, exists := process.Handlers[\"agents.tests.mcpload.tools\"]\n\t\tassert.True(t, exists, \"Handler should be registered\")\n\t\tassert.NotNil(t, handler)\n\t})\n}\n" }, { "path": "agent/assistant/load_store_test.go", "content": "package assistant_test\n\nimport (\n\tstdContext \"context\"\n\t\"testing\"\n\t\"time\"\n\n\t\"github.com/stretchr/testify/assert\"\n\t\"github.com/stretchr/testify/require\"\n\t\"github.com/yaoapp/yao/agent/assistant\"\n\t\"github.com/yaoapp/yao/agent/context\"\n\tsearchTypes \"github.com/yaoapp/yao/agent/search/types\"\n\tstore \"github.com/yaoapp/yao/agent/store/types\"\n\t\"github.com/yaoapp/yao/agent/testutils\"\n\t\"github.com/yaoapp/yao/openapi/oauth/types\"\n)\n\n// TestLoadStoreWithSource tests loading assistant from database with Source field\nfunc TestLoadStoreWithSource(t *testing.T) {\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\t// Create assistant with Source\n\tassistantID := \"test.store-with-source\"\n\tnow := time.Now().UnixNano()\n\n\tast := &assistant.Assistant{\n\t\tAssistantModel: store.AssistantModel{\n\t\t\tID: assistantID,\n\t\t\tName: \"Test Assistant With Source\",\n\t\t\tType: \"assistant\",\n\t\t\tConnector: \"gpt-4o\",\n\t\t\tDescription: \"Test assistant loaded from store with source code\",\n\t\t\tPrompts: []store.Prompt{\n\t\t\t\t{Role: \"system\", Content: \"You are a helpful assistant.\"},\n\t\t\t},\n\t\t\tOptions: map[string]interface{}{\n\t\t\t\t\"temperature\": 0.7,\n\t\t\t},\n\t\t\tTags: []string{\"Test\", \"Source\"},\n\t\t\t// Simple Create hook that returns null\n\t\t\tSource: `\n// @ts-nocheck\nfunction Create(ctx, messages) {\n\treturn null;\n}\n`,\n\t\t\tCreatedAt: now,\n\t\t\tUpdatedAt: now,\n\t\t},\n\t}\n\n\t// Save to database\n\terr := ast.Save()\n\trequire.NoError(t, err)\n\n\t// Cleanup after test\n\tdefer func() {\n\t\tstorage := assistant.GetStorage()\n\t\tif storage != nil {\n\t\t\tstorage.DeleteAssistant(assistantID)\n\t\t}\n\t\tassistant.GetCache().Clear()\n\t}()\n\n\t// Clear cache to ensure fresh load from database\n\tassistant.GetCache().Clear()\n\n\t// Load from store\n\tloaded, err := assistant.Get(assistantID)\n\trequire.NoError(t, err)\n\trequire.NotNil(t, loaded)\n\n\t// Verify basic fields\n\tassert.Equal(t, assistantID, loaded.ID)\n\tassert.Equal(t, \"Test Assistant With Source\", loaded.Name)\n\tassert.Equal(t, \"assistant\", loaded.Type)\n\tassert.Equal(t, \"Test assistant loaded from store with source code\", loaded.Description)\n\n\t// Verify prompts\n\trequire.NotNil(t, loaded.Prompts)\n\tassert.Len(t, loaded.Prompts, 1)\n\tassert.Equal(t, \"system\", loaded.Prompts[0].Role)\n\tassert.Equal(t, \"You are a helpful assistant.\", loaded.Prompts[0].Content)\n\n\t// Verify options\n\tassert.NotNil(t, loaded.Options)\n\tassert.Equal(t, 0.7, loaded.Options[\"temperature\"])\n\n\t// Verify tags\n\tassert.NotNil(t, loaded.Tags)\n\tassert.Contains(t, loaded.Tags, \"Test\")\n\tassert.Contains(t, loaded.Tags, \"Source\")\n\n\t// Verify script was compiled from source\n\tassert.NotNil(t, loaded.HookScript, \"HookScript should be compiled from Source field\")\n\n\t// Verify source is stored\n\tassert.NotEmpty(t, loaded.Source)\n}\n\n// TestLoadStoreWithoutSource tests loading assistant from database without Source field\nfunc TestLoadStoreWithoutSource(t *testing.T) {\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\t// Create assistant without Source\n\tassistantID := \"test.store-without-source\"\n\tnow := time.Now().UnixNano()\n\n\tast := &assistant.Assistant{\n\t\tAssistantModel: store.AssistantModel{\n\t\t\tID: assistantID,\n\t\t\tName: \"Test Assistant Without Source\",\n\t\t\tType: \"assistant\",\n\t\t\tConnector: \"gpt-4o\",\n\t\t\tDescription: \"Test assistant loaded from store without source code\",\n\t\t\tPrompts: []store.Prompt{\n\t\t\t\t{Role: \"system\", Content: \"You are a helpful assistant without hooks.\"},\n\t\t\t},\n\t\t\tOptions: map[string]interface{}{\n\t\t\t\t\"temperature\": 0.5,\n\t\t\t\t\"max_tokens\": 1000,\n\t\t\t},\n\t\t\tTags: []string{\"Test\", \"NoSource\"},\n\t\t\tCreatedAt: now,\n\t\t\tUpdatedAt: now,\n\t\t},\n\t}\n\n\t// Save to database\n\terr := ast.Save()\n\trequire.NoError(t, err)\n\n\t// Cleanup after test\n\tdefer func() {\n\t\tstorage := assistant.GetStorage()\n\t\tif storage != nil {\n\t\t\tstorage.DeleteAssistant(assistantID)\n\t\t}\n\t\tassistant.GetCache().Clear()\n\t}()\n\n\t// Clear cache to ensure fresh load from database\n\tassistant.GetCache().Clear()\n\n\t// Load from store\n\tloaded, err := assistant.Get(assistantID)\n\trequire.NoError(t, err)\n\trequire.NotNil(t, loaded)\n\n\t// Verify basic fields\n\tassert.Equal(t, assistantID, loaded.ID)\n\tassert.Equal(t, \"Test Assistant Without Source\", loaded.Name)\n\tassert.Equal(t, \"assistant\", loaded.Type)\n\tassert.Equal(t, \"Test assistant loaded from store without source code\", loaded.Description)\n\n\t// Verify prompts\n\trequire.NotNil(t, loaded.Prompts)\n\tassert.Len(t, loaded.Prompts, 1)\n\tassert.Equal(t, \"system\", loaded.Prompts[0].Role)\n\n\t// Verify options\n\tassert.NotNil(t, loaded.Options)\n\tassert.Equal(t, 0.5, loaded.Options[\"temperature\"])\n\tassert.Equal(t, float64(1000), loaded.Options[\"max_tokens\"])\n\n\t// Verify tags\n\tassert.NotNil(t, loaded.Tags)\n\tassert.Contains(t, loaded.Tags, \"Test\")\n\tassert.Contains(t, loaded.Tags, \"NoSource\")\n\n\t// Verify script is nil (no source)\n\tassert.Nil(t, loaded.HookScript, \"HookScript should be nil when no Source field\")\n\tassert.Empty(t, loaded.Source)\n}\n\n// newStoreTestContext creates a Context for testing with commonly used fields pre-populated.\nfunc newStoreTestContext(chatID, assistantID string) *context.Context {\n\tauthorized := &types.AuthorizedInfo{\n\t\tSubject: \"test-user\",\n\t\tClientID: \"test-client-id\",\n\t\tScope: \"openid profile email\",\n\t\tSessionID: \"test-session-id\",\n\t\tUserID: \"test-user-123\",\n\t\tTeamID: \"test-team-456\",\n\t\tTenantID: \"test-tenant-789\",\n\t}\n\n\tctx := context.New(stdContext.Background(), authorized, chatID)\n\tctx.AssistantID = assistantID\n\tctx.Locale = \"en-us\"\n\tctx.Theme = \"light\"\n\tctx.Client = context.Client{\n\t\tType: \"web\",\n\t\tUserAgent: \"TestAgent/1.0\",\n\t\tIP: \"127.0.0.1\",\n\t}\n\tctx.Referer = context.RefererAPI\n\tctx.Accept = context.AcceptWebCUI\n\tctx.Route = \"\"\n\tctx.Metadata = make(map[string]interface{})\n\treturn ctx\n}\n\n// TestLoadStoreWithSourceExecuteHook tests that Source-based script is properly compiled and can execute\nfunc TestLoadStoreWithSourceExecuteHook(t *testing.T) {\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\t// Create assistant with a working Create hook\n\tassistantID := \"test.store-source-hook\"\n\tnow := time.Now().UnixNano()\n\n\tast := &assistant.Assistant{\n\t\tAssistantModel: store.AssistantModel{\n\t\t\tID: assistantID,\n\t\t\tName: \"Test Source Hook\",\n\t\t\tType: \"assistant\",\n\t\t\tConnector: \"gpt-4o\",\n\t\t\tPrompts: []store.Prompt{\n\t\t\t\t{Role: \"system\", Content: \"Default prompt\"},\n\t\t\t},\n\t\t\t// Create hook that modifies temperature and adds metadata\n\t\t\tSource: `\n// @ts-nocheck\nfunction Create(ctx: any, messages: any[]): any {\n\treturn {\n\t\ttemperature: 0.9,\n\t\tmetadata: {\n\t\t\thook_executed: true,\n\t\t\tchat_id: ctx.chat_id\n\t\t}\n\t};\n}\n`,\n\t\t\tCreatedAt: now,\n\t\t\tUpdatedAt: now,\n\t\t},\n\t}\n\n\t// Save to database\n\terr := ast.Save()\n\trequire.NoError(t, err)\n\n\t// Cleanup after test\n\tdefer func() {\n\t\tstorage := assistant.GetStorage()\n\t\tif storage != nil {\n\t\t\tstorage.DeleteAssistant(assistantID)\n\t\t}\n\t\tassistant.GetCache().Clear()\n\t}()\n\n\t// Clear cache\n\tassistant.GetCache().Clear()\n\n\t// Load from store\n\tloaded, err := assistant.Get(assistantID)\n\trequire.NoError(t, err)\n\trequire.NotNil(t, loaded)\n\trequire.NotNil(t, loaded.HookScript, \"HookScript should be compiled from Source\")\n\n\t// Verify the script object exists and is usable\n\tassert.NotNil(t, loaded.HookScript.Script)\n\n\t// Execute the Create hook\n\tctx := newStoreTestContext(\"test-chat-id\", assistantID)\n\tmessages := []context.Message{{Role: \"user\", Content: \"Hello\"}}\n\n\tres, _, err := loaded.HookScript.Create(ctx, messages, &context.Options{})\n\trequire.NoError(t, err, \"Create hook should execute without error\")\n\trequire.NotNil(t, res, \"Create hook should return a response\")\n\n\t// Verify temperature was set\n\trequire.NotNil(t, res.Temperature, \"Temperature should be set\")\n\tassert.Equal(t, 0.9, *res.Temperature, \"Temperature should be 0.9\")\n\n\t// Verify metadata was set\n\trequire.NotNil(t, res.Metadata, \"Metadata should be set\")\n\tassert.Equal(t, true, res.Metadata[\"hook_executed\"], \"hook_executed should be true\")\n\tassert.Equal(t, \"test-chat-id\", res.Metadata[\"chat_id\"], \"chat_id should match context\")\n}\n\n// TestLoadStoreWithPromptPresets tests loading assistant with prompt presets from database\nfunc TestLoadStoreWithPromptPresets(t *testing.T) {\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\tassistantID := \"test.store-with-presets\"\n\tnow := time.Now().UnixNano()\n\n\tast := &assistant.Assistant{\n\t\tAssistantModel: store.AssistantModel{\n\t\t\tID: assistantID,\n\t\t\tName: \"Test With Presets\",\n\t\t\tType: \"assistant\",\n\t\t\tConnector: \"gpt-4o\",\n\t\t\tPrompts: []store.Prompt{\n\t\t\t\t{Role: \"system\", Content: \"Default prompt\"},\n\t\t\t},\n\t\t\tPromptPresets: map[string][]store.Prompt{\n\t\t\t\t\"friendly\": {\n\t\t\t\t\t{Role: \"system\", Content: \"You are a friendly assistant.\"},\n\t\t\t\t},\n\t\t\t\t\"professional\": {\n\t\t\t\t\t{Role: \"system\", Content: \"You are a professional assistant.\"},\n\t\t\t\t},\n\t\t\t\t\"mode.casual\": {\n\t\t\t\t\t{Role: \"system\", Content: \"You are a casual assistant.\"},\n\t\t\t\t},\n\t\t\t},\n\t\t\tCreatedAt: now,\n\t\t\tUpdatedAt: now,\n\t\t},\n\t}\n\n\terr := ast.Save()\n\trequire.NoError(t, err)\n\n\tdefer func() {\n\t\tstorage := assistant.GetStorage()\n\t\tif storage != nil {\n\t\t\tstorage.DeleteAssistant(assistantID)\n\t\t}\n\t\tassistant.GetCache().Clear()\n\t}()\n\n\tassistant.GetCache().Clear()\n\n\tloaded, err := assistant.Get(assistantID)\n\trequire.NoError(t, err)\n\trequire.NotNil(t, loaded)\n\n\t// Verify prompt presets\n\trequire.NotNil(t, loaded.PromptPresets)\n\tassert.Len(t, loaded.PromptPresets, 3)\n\n\tfriendlyPreset, ok := loaded.PromptPresets[\"friendly\"]\n\tassert.True(t, ok)\n\tassert.Len(t, friendlyPreset, 1)\n\tassert.Equal(t, \"You are a friendly assistant.\", friendlyPreset[0].Content)\n\n\tprofessionalPreset, ok := loaded.PromptPresets[\"professional\"]\n\tassert.True(t, ok)\n\tassert.Len(t, professionalPreset, 1)\n\tassert.Equal(t, \"You are a professional assistant.\", professionalPreset[0].Content)\n\n\tcasualPreset, ok := loaded.PromptPresets[\"mode.casual\"]\n\tassert.True(t, ok)\n\tassert.Len(t, casualPreset, 1)\n\tassert.Equal(t, \"You are a casual assistant.\", casualPreset[0].Content)\n}\n\n// TestLoadStoreWithDisableGlobalPrompts tests loading assistant with disable_global_prompts flag\nfunc TestLoadStoreWithDisableGlobalPrompts(t *testing.T) {\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\tassistantID := \"test.store-disable-global\"\n\tnow := time.Now().UnixNano()\n\n\tast := &assistant.Assistant{\n\t\tAssistantModel: store.AssistantModel{\n\t\t\tID: assistantID,\n\t\t\tName: \"Test Disable Global Prompts\",\n\t\t\tType: \"assistant\",\n\t\t\tConnector: \"gpt-4o\",\n\t\t\tDisableGlobalPrompts: true,\n\t\t\tPrompts: []store.Prompt{\n\t\t\t\t{Role: \"system\", Content: \"Only this prompt should be used.\"},\n\t\t\t},\n\t\t\tCreatedAt: now,\n\t\t\tUpdatedAt: now,\n\t\t},\n\t}\n\n\terr := ast.Save()\n\trequire.NoError(t, err)\n\n\tdefer func() {\n\t\tstorage := assistant.GetStorage()\n\t\tif storage != nil {\n\t\t\tstorage.DeleteAssistant(assistantID)\n\t\t}\n\t\tassistant.GetCache().Clear()\n\t}()\n\n\tassistant.GetCache().Clear()\n\n\tloaded, err := assistant.Get(assistantID)\n\trequire.NoError(t, err)\n\trequire.NotNil(t, loaded)\n\n\tassert.True(t, loaded.DisableGlobalPrompts)\n}\n\n// TestLoadStoreCaching tests that loaded assistants are cached\nfunc TestLoadStoreCaching(t *testing.T) {\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\tassistantID := \"test.store-caching\"\n\tnow := time.Now().UnixNano()\n\n\tast := &assistant.Assistant{\n\t\tAssistantModel: store.AssistantModel{\n\t\t\tID: assistantID,\n\t\t\tName: \"Test Caching\",\n\t\t\tType: \"assistant\",\n\t\t\tConnector: \"gpt-4o\",\n\t\t\tCreatedAt: now,\n\t\t\tUpdatedAt: now,\n\t\t},\n\t}\n\n\terr := ast.Save()\n\trequire.NoError(t, err)\n\n\tdefer func() {\n\t\tstorage := assistant.GetStorage()\n\t\tif storage != nil {\n\t\t\tstorage.DeleteAssistant(assistantID)\n\t\t}\n\t\tassistant.GetCache().Clear()\n\t}()\n\n\tassistant.GetCache().Clear()\n\n\t// First load\n\tast1, err := assistant.Get(assistantID)\n\trequire.NoError(t, err)\n\trequire.NotNil(t, ast1)\n\n\t// Second load - should be from cache\n\tast2, err := assistant.Get(assistantID)\n\trequire.NoError(t, err)\n\trequire.NotNil(t, ast2)\n\n\t// Should be the same instance (from cache)\n\tassert.Same(t, ast1, ast2)\n}\n\n// TestLoadStoreNotFound tests loading non-existent assistant\nfunc TestLoadStoreNotFound(t *testing.T) {\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\tassistant.GetCache().Clear()\n\n\t_, err := assistant.Get(\"non-existent-assistant-id-12345\")\n\tassert.Error(t, err)\n}\n\n// TestLoadStoreWithAllFields tests loading assistant with comprehensive fields\nfunc TestLoadStoreWithAllFields(t *testing.T) {\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\tassistantID := \"test.store-all-fields\"\n\tnow := time.Now().UnixNano()\n\n\tast := &assistant.Assistant{\n\t\tAssistantModel: store.AssistantModel{\n\t\t\tID: assistantID,\n\t\t\tName: \"Test All Fields\",\n\t\t\tType: \"assistant\",\n\t\t\tAvatar: \"/api/icons/test.png\",\n\t\t\tConnector: \"gpt-4o\",\n\t\t\tDescription: \"Test assistant with all fields\",\n\t\t\tTags: []string{\"Test\", \"AllFields\", \"Complete\"},\n\t\t\tReadonly: true,\n\t\t\tPublic: true,\n\t\t\tShare: \"team\",\n\t\t\tMentionable: true,\n\t\t\tAutomated: false,\n\t\t\tSort: 100,\n\t\t\tOptions: map[string]interface{}{\n\t\t\t\t\"temperature\": 0.8,\n\t\t\t\t\"max_tokens\": 2000,\n\t\t\t},\n\t\t\tPrompts: []store.Prompt{\n\t\t\t\t{Role: \"system\", Content: \"You are a test assistant.\"},\n\t\t\t\t{Role: \"system\", Content: \"Follow all instructions carefully.\"},\n\t\t\t},\n\t\t\tPromptPresets: map[string][]store.Prompt{\n\t\t\t\t\"default\": {\n\t\t\t\t\t{Role: \"system\", Content: \"Default mode prompt.\"},\n\t\t\t\t},\n\t\t\t},\n\t\t\tDisableGlobalPrompts: true,\n\t\t\tPlaceholder: &store.Placeholder{\n\t\t\t\tTitle: \"Test Placeholder\",\n\t\t\t\tDescription: \"This is a test placeholder\",\n\t\t\t\tPrompts: []string{\"Test prompt 1\", \"Test prompt 2\"},\n\t\t\t},\n\t\t\tDependencies: map[string]string{\n\t\t\t\t\"echo\": \"^1.0.0\",\n\t\t\t\t\"customer\": \">=2.0.0\",\n\t\t\t},\n\t\t\tSource: `\n// @ts-nocheck\nfunction Create(ctx: any, messages: any[]): any {\n\treturn { \n\t\ttemperature: 0.5,\n\t\tmetadata: {\n\t\t\tassistant_name: \"Test All Fields\",\n\t\t\texecuted: true\n\t\t}\n\t};\n}\n`,\n\t\t\tCreatedAt: now,\n\t\t\tUpdatedAt: now,\n\t\t},\n\t}\n\n\terr := ast.Save()\n\trequire.NoError(t, err)\n\n\tdefer func() {\n\t\tstorage := assistant.GetStorage()\n\t\tif storage != nil {\n\t\t\tstorage.DeleteAssistant(assistantID)\n\t\t}\n\t\tassistant.GetCache().Clear()\n\t}()\n\n\tassistant.GetCache().Clear()\n\n\tloaded, err := assistant.Get(assistantID)\n\trequire.NoError(t, err)\n\trequire.NotNil(t, loaded)\n\n\t// Verify all fields\n\tassert.Equal(t, assistantID, loaded.ID)\n\tassert.Equal(t, \"Test All Fields\", loaded.Name)\n\tassert.Equal(t, \"assistant\", loaded.Type)\n\tassert.Equal(t, \"/api/icons/test.png\", loaded.Avatar)\n\tassert.Equal(t, \"Test assistant with all fields\", loaded.Description)\n\n\t// Boolean fields\n\tassert.True(t, loaded.Readonly)\n\tassert.True(t, loaded.Public)\n\tassert.Equal(t, \"team\", loaded.Share)\n\tassert.True(t, loaded.Mentionable)\n\tassert.False(t, loaded.Automated)\n\tassert.True(t, loaded.DisableGlobalPrompts)\n\tassert.Equal(t, 100, loaded.Sort)\n\n\t// Tags\n\tassert.Len(t, loaded.Tags, 3)\n\tassert.Contains(t, loaded.Tags, \"Test\")\n\tassert.Contains(t, loaded.Tags, \"AllFields\")\n\tassert.Contains(t, loaded.Tags, \"Complete\")\n\n\t// Options\n\tassert.Equal(t, 0.8, loaded.Options[\"temperature\"])\n\tassert.Equal(t, float64(2000), loaded.Options[\"max_tokens\"])\n\n\t// Prompts\n\tassert.Len(t, loaded.Prompts, 2)\n\n\t// Prompt presets\n\tassert.NotNil(t, loaded.PromptPresets)\n\tassert.Contains(t, loaded.PromptPresets, \"default\")\n\n\t// Placeholder\n\tassert.NotNil(t, loaded.Placeholder)\n\tassert.Equal(t, \"Test Placeholder\", loaded.Placeholder.Title)\n\tassert.Equal(t, \"This is a test placeholder\", loaded.Placeholder.Description)\n\tassert.Len(t, loaded.Placeholder.Prompts, 2)\n\n\t// Script from source\n\tassert.NotNil(t, loaded.HookScript)\n\tassert.NotEmpty(t, loaded.Source)\n\n\t// Dependencies\n\trequire.NotNil(t, loaded.Dependencies)\n\tassert.Len(t, loaded.Dependencies, 2)\n\tassert.Equal(t, \"^1.0.0\", loaded.Dependencies[\"echo\"])\n\tassert.Equal(t, \">=2.0.0\", loaded.Dependencies[\"customer\"])\n\n\t// Execute the Create hook to verify it works\n\tctx := newStoreTestContext(\"test-chat-all-fields\", assistantID)\n\tmessages := []context.Message{{Role: \"user\", Content: \"Test message\"}}\n\n\tres, _, err := loaded.HookScript.Create(ctx, messages, &context.Options{})\n\trequire.NoError(t, err, \"Create hook should execute without error\")\n\trequire.NotNil(t, res, \"Create hook should return a response\")\n\n\t// Verify hook returned expected values\n\trequire.NotNil(t, res.Temperature, \"Temperature should be set\")\n\tassert.Equal(t, 0.5, *res.Temperature, \"Temperature should be 0.5\")\n\n\trequire.NotNil(t, res.Metadata, \"Metadata should be set\")\n\tassert.Equal(t, \"Test All Fields\", res.Metadata[\"assistant_name\"], \"assistant_name should match\")\n\tassert.Equal(t, true, res.Metadata[\"executed\"], \"executed should be true\")\n}\n\n// TestLoadStoreHookWithTypeScript tests that TypeScript features work in Source field\nfunc TestLoadStoreHookWithTypeScript(t *testing.T) {\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\tassistantID := \"test.store-typescript-hook\"\n\tnow := time.Now().UnixNano()\n\n\tast := &assistant.Assistant{\n\t\tAssistantModel: store.AssistantModel{\n\t\t\tID: assistantID,\n\t\t\tName: \"Test TypeScript Hook\",\n\t\t\tType: \"assistant\",\n\t\t\tConnector: \"gpt-4o\",\n\t\t\tPrompts: []store.Prompt{\n\t\t\t\t{Role: \"system\", Content: \"Default prompt\"},\n\t\t\t},\n\t\t\t// TypeScript code with type annotations and interfaces\n\t\t\tSource: `\n// TypeScript interfaces\ninterface CreateContext {\n\tchat_id: string;\n\tassistant_id: string;\n\tlocale: string;\n\tauthorized?: {\n\t\tuser_id: string;\n\t\tteam_id: string;\n\t};\n}\n\ninterface Message {\n\trole: string;\n\tcontent: string | object;\n}\n\ninterface CreateResponse {\n\ttemperature?: number;\n\tmessages?: Message[];\n\tmetadata?: Record;\n}\n\n// Create hook with full TypeScript syntax\nfunction Create(ctx: CreateContext, messages: Message[]): CreateResponse | null {\n\t// Type-safe access to context\n\tconst chatId: string = ctx.chat_id || \"unknown\";\n\tconst locale: string = ctx.locale || \"en-us\";\n\tconst userId: string = ctx.authorized?.user_id || \"anonymous\";\n\t\n\t// Process messages\n\tconst userMessages: Message[] = messages.filter((m: Message) => m.role === \"user\");\n\tconst messageCount: number = userMessages.length;\n\t\n\t// Return typed response\n\treturn {\n\t\ttemperature: 0.7,\n\t\tmessages: [\n\t\t\t{\n\t\t\t\trole: \"system\",\n\t\t\t\tcontent: \"TypeScript hook executed successfully\"\n\t\t\t}\n\t\t],\n\t\tmetadata: {\n\t\t\tchat_id: chatId,\n\t\t\tlocale: locale,\n\t\t\tuser_id: userId,\n\t\t\tmessage_count: messageCount,\n\t\t\ttypescript_features: true\n\t\t}\n\t};\n}\n`,\n\t\t\tCreatedAt: now,\n\t\t\tUpdatedAt: now,\n\t\t},\n\t}\n\n\terr := ast.Save()\n\trequire.NoError(t, err)\n\n\tdefer func() {\n\t\tstorage := assistant.GetStorage()\n\t\tif storage != nil {\n\t\t\tstorage.DeleteAssistant(assistantID)\n\t\t}\n\t\tassistant.GetCache().Clear()\n\t}()\n\n\tassistant.GetCache().Clear()\n\n\tloaded, err := assistant.Get(assistantID)\n\trequire.NoError(t, err)\n\trequire.NotNil(t, loaded)\n\trequire.NotNil(t, loaded.HookScript, \"HookScript should be compiled from TypeScript Source\")\n\n\t// Execute the Create hook\n\tctx := newStoreTestContext(\"ts-test-chat\", assistantID)\n\tmessages := []context.Message{\n\t\t{Role: \"user\", Content: \"Hello\"},\n\t\t{Role: \"assistant\", Content: \"Hi there\"},\n\t\t{Role: \"user\", Content: \"How are you?\"},\n\t}\n\n\tres, _, err := loaded.HookScript.Create(ctx, messages, &context.Options{})\n\trequire.NoError(t, err, \"TypeScript Create hook should execute without error\")\n\trequire.NotNil(t, res, \"Create hook should return a response\")\n\n\t// Verify temperature\n\trequire.NotNil(t, res.Temperature)\n\tassert.Equal(t, 0.7, *res.Temperature)\n\n\t// Verify messages\n\trequire.Len(t, res.Messages, 1)\n\tassert.Equal(t, context.RoleSystem, res.Messages[0].Role)\n\tassert.Equal(t, \"TypeScript hook executed successfully\", res.Messages[0].Content)\n\n\t// Verify metadata\n\trequire.NotNil(t, res.Metadata)\n\tassert.Equal(t, \"ts-test-chat\", res.Metadata[\"chat_id\"])\n\tassert.Equal(t, \"en-us\", res.Metadata[\"locale\"])\n\tassert.Equal(t, \"test-user-123\", res.Metadata[\"user_id\"])\n\tassert.Equal(t, float64(2), res.Metadata[\"message_count\"]) // 2 user messages\n\tassert.Equal(t, true, res.Metadata[\"typescript_features\"])\n}\n\n// TestLoadStoreHookReturnNull tests that hook returning null works correctly\nfunc TestLoadStoreHookReturnNull(t *testing.T) {\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\tassistantID := \"test.store-hook-null\"\n\tnow := time.Now().UnixNano()\n\n\tast := &assistant.Assistant{\n\t\tAssistantModel: store.AssistantModel{\n\t\t\tID: assistantID,\n\t\t\tName: \"Test Hook Return Null\",\n\t\t\tType: \"assistant\",\n\t\t\tConnector: \"gpt-4o\",\n\t\t\tSource: `\nfunction Create(ctx: any, messages: any[]): any {\n\t// Return null to indicate no modifications\n\treturn null;\n}\n`,\n\t\t\tCreatedAt: now,\n\t\t\tUpdatedAt: now,\n\t\t},\n\t}\n\n\terr := ast.Save()\n\trequire.NoError(t, err)\n\n\tdefer func() {\n\t\tstorage := assistant.GetStorage()\n\t\tif storage != nil {\n\t\t\tstorage.DeleteAssistant(assistantID)\n\t\t}\n\t\tassistant.GetCache().Clear()\n\t}()\n\n\tassistant.GetCache().Clear()\n\n\tloaded, err := assistant.Get(assistantID)\n\trequire.NoError(t, err)\n\trequire.NotNil(t, loaded)\n\trequire.NotNil(t, loaded.HookScript)\n\n\tctx := newStoreTestContext(\"null-test-chat\", assistantID)\n\tmessages := []context.Message{{Role: \"user\", Content: \"Hello\"}}\n\n\tres, _, err := loaded.HookScript.Create(ctx, messages, &context.Options{})\n\trequire.NoError(t, err, \"Hook returning null should not error\")\n\tassert.Nil(t, res, \"Hook returning null should return nil response\")\n}\n\n// TestLoadStoreHookWithPromptPreset tests that hook can return prompt_preset\nfunc TestLoadStoreHookWithPromptPreset(t *testing.T) {\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\tassistantID := \"test.store-hook-preset\"\n\tnow := time.Now().UnixNano()\n\n\tast := &assistant.Assistant{\n\t\tAssistantModel: store.AssistantModel{\n\t\t\tID: assistantID,\n\t\t\tName: \"Test Hook Prompt Preset\",\n\t\t\tType: \"assistant\",\n\t\t\tConnector: \"gpt-4o\",\n\t\t\tPrompts: []store.Prompt{\n\t\t\t\t{Role: \"system\", Content: \"Default prompt\"},\n\t\t\t},\n\t\t\tPromptPresets: map[string][]store.Prompt{\n\t\t\t\t\"friendly\": {\n\t\t\t\t\t{Role: \"system\", Content: \"You are a friendly assistant.\"},\n\t\t\t\t},\n\t\t\t\t\"professional\": {\n\t\t\t\t\t{Role: \"system\", Content: \"You are a professional assistant.\"},\n\t\t\t\t},\n\t\t\t},\n\t\t\tSource: `\nfunction Create(ctx: any, messages: any[]): any {\n\t// Check first message to determine preset\n\tconst firstMsg = messages[0];\n\tif (firstMsg && typeof firstMsg.content === \"string\") {\n\t\tif (firstMsg.content.includes(\"friendly\")) {\n\t\t\treturn { prompt_preset: \"friendly\" };\n\t\t}\n\t\tif (firstMsg.content.includes(\"professional\")) {\n\t\t\treturn { prompt_preset: \"professional\" };\n\t\t}\n\t}\n\treturn null;\n}\n`,\n\t\t\tCreatedAt: now,\n\t\t\tUpdatedAt: now,\n\t\t},\n\t}\n\n\terr := ast.Save()\n\trequire.NoError(t, err)\n\n\tdefer func() {\n\t\tstorage := assistant.GetStorage()\n\t\tif storage != nil {\n\t\t\tstorage.DeleteAssistant(assistantID)\n\t\t}\n\t\tassistant.GetCache().Clear()\n\t}()\n\n\tassistant.GetCache().Clear()\n\n\tloaded, err := assistant.Get(assistantID)\n\trequire.NoError(t, err)\n\trequire.NotNil(t, loaded)\n\trequire.NotNil(t, loaded.HookScript)\n\n\t// Test friendly preset selection\n\tt.Run(\"SelectFriendlyPreset\", func(t *testing.T) {\n\t\tctx := newStoreTestContext(\"preset-test-1\", assistantID)\n\t\tmessages := []context.Message{{Role: \"user\", Content: \"Be friendly please\"}}\n\n\t\tres, _, err := loaded.HookScript.Create(ctx, messages, &context.Options{})\n\t\trequire.NoError(t, err)\n\t\trequire.NotNil(t, res)\n\t\tassert.Equal(t, \"friendly\", res.PromptPreset)\n\t})\n\n\t// Test professional preset selection\n\tt.Run(\"SelectProfessionalPreset\", func(t *testing.T) {\n\t\tctx := newStoreTestContext(\"preset-test-2\", assistantID)\n\t\tmessages := []context.Message{{Role: \"user\", Content: \"Be professional\"}}\n\n\t\tres, _, err := loaded.HookScript.Create(ctx, messages, &context.Options{})\n\t\trequire.NoError(t, err)\n\t\trequire.NotNil(t, res)\n\t\tassert.Equal(t, \"professional\", res.PromptPreset)\n\t})\n\n\t// Test no preset (returns null)\n\tt.Run(\"NoPreset\", func(t *testing.T) {\n\t\tctx := newStoreTestContext(\"preset-test-3\", assistantID)\n\t\tmessages := []context.Message{{Role: \"user\", Content: \"Hello\"}}\n\n\t\tres, _, err := loaded.HookScript.Create(ctx, messages, &context.Options{})\n\t\trequire.NoError(t, err)\n\t\tassert.Nil(t, res)\n\t})\n}\n\n// TestLoadStoreHookDisableGlobalPrompts tests that hook can disable global prompts\nfunc TestLoadStoreHookDisableGlobalPrompts(t *testing.T) {\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\tassistantID := \"test.store-hook-disable-global\"\n\tnow := time.Now().UnixNano()\n\n\tast := &assistant.Assistant{\n\t\tAssistantModel: store.AssistantModel{\n\t\t\tID: assistantID,\n\t\t\tName: \"Test Hook Disable Global\",\n\t\t\tType: \"assistant\",\n\t\t\tConnector: \"gpt-4o\",\n\t\t\tSource: `\nfunction Create(ctx: any, messages: any[]): any {\n\tconst firstMsg = messages[0];\n\tif (firstMsg && typeof firstMsg.content === \"string\") {\n\t\tif (firstMsg.content.includes(\"disable_global\")) {\n\t\t\treturn { disable_global_prompts: true };\n\t\t}\n\t\tif (firstMsg.content.includes(\"enable_global\")) {\n\t\t\treturn { disable_global_prompts: false };\n\t\t}\n\t}\n\treturn null;\n}\n`,\n\t\t\tCreatedAt: now,\n\t\t\tUpdatedAt: now,\n\t\t},\n\t}\n\n\terr := ast.Save()\n\trequire.NoError(t, err)\n\n\tdefer func() {\n\t\tstorage := assistant.GetStorage()\n\t\tif storage != nil {\n\t\t\tstorage.DeleteAssistant(assistantID)\n\t\t}\n\t\tassistant.GetCache().Clear()\n\t}()\n\n\tassistant.GetCache().Clear()\n\n\tloaded, err := assistant.Get(assistantID)\n\trequire.NoError(t, err)\n\trequire.NotNil(t, loaded)\n\trequire.NotNil(t, loaded.HookScript)\n\n\t// Test disable global prompts\n\tt.Run(\"DisableGlobalPrompts\", func(t *testing.T) {\n\t\tctx := newStoreTestContext(\"disable-test-1\", assistantID)\n\t\tmessages := []context.Message{{Role: \"user\", Content: \"disable_global prompts\"}}\n\n\t\tres, _, err := loaded.HookScript.Create(ctx, messages, &context.Options{})\n\t\trequire.NoError(t, err)\n\t\trequire.NotNil(t, res)\n\t\trequire.NotNil(t, res.DisableGlobalPrompts)\n\t\tassert.True(t, *res.DisableGlobalPrompts)\n\t})\n\n\t// Test enable global prompts\n\tt.Run(\"EnableGlobalPrompts\", func(t *testing.T) {\n\t\tctx := newStoreTestContext(\"disable-test-2\", assistantID)\n\t\tmessages := []context.Message{{Role: \"user\", Content: \"enable_global prompts\"}}\n\n\t\tres, _, err := loaded.HookScript.Create(ctx, messages, &context.Options{})\n\t\trequire.NoError(t, err)\n\t\trequire.NotNil(t, res)\n\t\trequire.NotNil(t, res.DisableGlobalPrompts)\n\t\tassert.False(t, *res.DisableGlobalPrompts)\n\t})\n}\n\n// TestLoadStoreWithSearchConfig tests loading assistant with search configuration from database\nfunc TestLoadStoreWithSearchConfig(t *testing.T) {\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\tassistantID := \"test.store-with-search\"\n\tnow := time.Now().UnixNano()\n\n\tast := &assistant.Assistant{\n\t\tAssistantModel: store.AssistantModel{\n\t\t\tID: assistantID,\n\t\t\tName: \"Test With Search Config\",\n\t\t\tType: \"assistant\",\n\t\t\tConnector: \"gpt-4o\",\n\t\t\tUses: &context.Uses{\n\t\t\t\tVision: \"agent\",\n\t\t\t\tAudio: \"mcp:audio-server\",\n\t\t\t\tFetch: \"agent\",\n\t\t\t\tWeb: \"builtin\",\n\t\t\t\tKeyword: \"builtin\",\n\t\t\t\tQueryDSL: \"builtin\",\n\t\t\t\tRerank: \"builtin\",\n\t\t\t},\n\t\t\tSearch: &searchTypes.Config{\n\t\t\t\tWeb: &searchTypes.WebConfig{\n\t\t\t\t\tProvider: \"tavily\",\n\t\t\t\t\tMaxResults: 15,\n\t\t\t\t},\n\t\t\t\tKB: &searchTypes.KBConfig{\n\t\t\t\t\tCollections: []string{\"docs\", \"faq\"},\n\t\t\t\t\tThreshold: 0.8,\n\t\t\t\t\tGraph: true,\n\t\t\t\t},\n\t\t\t\tDB: &searchTypes.DBConfig{\n\t\t\t\t\tModels: []string{\"user\", \"product\"},\n\t\t\t\t\tMaxResults: 50,\n\t\t\t\t},\n\t\t\t\tKeyword: &searchTypes.KeywordConfig{\n\t\t\t\t\tMaxKeywords: 8,\n\t\t\t\t\tLanguage: \"auto\",\n\t\t\t\t},\n\t\t\t\tQueryDSL: &searchTypes.QueryDSLConfig{\n\t\t\t\t\tStrict: true,\n\t\t\t\t},\n\t\t\t\tRerank: &searchTypes.RerankConfig{\n\t\t\t\t\tTopN: 5,\n\t\t\t\t},\n\t\t\t\tCitation: &searchTypes.CitationConfig{\n\t\t\t\t\tFormat: \"#cite:{id}\",\n\t\t\t\t\tAutoInjectPrompt: false,\n\t\t\t\t\tCustomPrompt: \"Please cite sources.\",\n\t\t\t\t},\n\t\t\t\tWeights: &searchTypes.WeightsConfig{\n\t\t\t\t\tUser: 1.0,\n\t\t\t\t\tHook: 0.85,\n\t\t\t\t\tAuto: 0.65,\n\t\t\t\t},\n\t\t\t\tOptions: &searchTypes.OptionsConfig{\n\t\t\t\t\tSkipThreshold: 3,\n\t\t\t\t},\n\t\t\t},\n\t\t\tCreatedAt: now,\n\t\t\tUpdatedAt: now,\n\t\t},\n\t}\n\n\terr := ast.Save()\n\trequire.NoError(t, err)\n\n\tdefer func() {\n\t\tstorage := assistant.GetStorage()\n\t\tif storage != nil {\n\t\t\tstorage.DeleteAssistant(assistantID)\n\t\t}\n\t\tassistant.GetCache().Clear()\n\t}()\n\n\tassistant.GetCache().Clear()\n\n\tloaded, err := assistant.Get(assistantID)\n\trequire.NoError(t, err)\n\trequire.NotNil(t, loaded)\n\n\t// Verify Uses\n\trequire.NotNil(t, loaded.Uses)\n\tassert.Equal(t, \"agent\", loaded.Uses.Vision)\n\tassert.Equal(t, \"mcp:audio-server\", loaded.Uses.Audio)\n\tassert.Equal(t, \"agent\", loaded.Uses.Fetch)\n\tassert.Equal(t, \"builtin\", loaded.Uses.Web)\n\tassert.Equal(t, \"builtin\", loaded.Uses.Keyword)\n\tassert.Equal(t, \"builtin\", loaded.Uses.QueryDSL)\n\tassert.Equal(t, \"builtin\", loaded.Uses.Rerank)\n\n\t// Verify Search config\n\trequire.NotNil(t, loaded.Search)\n\n\t// Web config\n\trequire.NotNil(t, loaded.Search.Web)\n\tassert.Equal(t, \"tavily\", loaded.Search.Web.Provider)\n\tassert.Equal(t, 15, loaded.Search.Web.MaxResults)\n\n\t// KB config\n\trequire.NotNil(t, loaded.Search.KB)\n\tassert.Equal(t, []string{\"docs\", \"faq\"}, loaded.Search.KB.Collections)\n\tassert.Equal(t, 0.8, loaded.Search.KB.Threshold)\n\tassert.True(t, loaded.Search.KB.Graph)\n\n\t// DB config\n\trequire.NotNil(t, loaded.Search.DB)\n\tassert.Equal(t, []string{\"user\", \"product\"}, loaded.Search.DB.Models)\n\tassert.Equal(t, 50, loaded.Search.DB.MaxResults)\n\n\t// Keyword config\n\trequire.NotNil(t, loaded.Search.Keyword)\n\tassert.Equal(t, 8, loaded.Search.Keyword.MaxKeywords)\n\tassert.Equal(t, \"auto\", loaded.Search.Keyword.Language)\n\n\t// QueryDSL config\n\trequire.NotNil(t, loaded.Search.QueryDSL)\n\tassert.True(t, loaded.Search.QueryDSL.Strict)\n\n\t// Rerank config\n\trequire.NotNil(t, loaded.Search.Rerank)\n\tassert.Equal(t, 5, loaded.Search.Rerank.TopN)\n\n\t// Citation config\n\trequire.NotNil(t, loaded.Search.Citation)\n\tassert.Equal(t, \"#cite:{id}\", loaded.Search.Citation.Format)\n\tassert.False(t, loaded.Search.Citation.AutoInjectPrompt)\n\tassert.Equal(t, \"Please cite sources.\", loaded.Search.Citation.CustomPrompt)\n\n\t// Weights config\n\trequire.NotNil(t, loaded.Search.Weights)\n\tassert.Equal(t, 1.0, loaded.Search.Weights.User)\n\tassert.Equal(t, 0.85, loaded.Search.Weights.Hook)\n\tassert.Equal(t, 0.65, loaded.Search.Weights.Auto)\n\n\t// Options config\n\trequire.NotNil(t, loaded.Search.Options)\n\tassert.Equal(t, 3, loaded.Search.Options.SkipThreshold)\n}\n\n// TestLoadStoreWithPartialSearchConfig tests loading assistant with partial search configuration\nfunc TestLoadStoreWithPartialSearchConfig(t *testing.T) {\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\tassistantID := \"test.store-partial-search\"\n\tnow := time.Now().UnixNano()\n\n\tast := &assistant.Assistant{\n\t\tAssistantModel: store.AssistantModel{\n\t\t\tID: assistantID,\n\t\t\tName: \"Test Partial Search Config\",\n\t\t\tType: \"assistant\",\n\t\t\tConnector: \"gpt-4o\",\n\t\t\tSearch: &searchTypes.Config{\n\t\t\t\tWeb: &searchTypes.WebConfig{\n\t\t\t\t\tProvider: \"serper\",\n\t\t\t\t},\n\t\t\t\t// Only web config, others are nil\n\t\t\t},\n\t\t\tCreatedAt: now,\n\t\t\tUpdatedAt: now,\n\t\t},\n\t}\n\n\terr := ast.Save()\n\trequire.NoError(t, err)\n\n\tdefer func() {\n\t\tstorage := assistant.GetStorage()\n\t\tif storage != nil {\n\t\t\tstorage.DeleteAssistant(assistantID)\n\t\t}\n\t\tassistant.GetCache().Clear()\n\t}()\n\n\tassistant.GetCache().Clear()\n\n\tloaded, err := assistant.Get(assistantID)\n\trequire.NoError(t, err)\n\trequire.NotNil(t, loaded)\n\n\t// Verify Search config\n\trequire.NotNil(t, loaded.Search)\n\n\t// Web config should be set\n\trequire.NotNil(t, loaded.Search.Web)\n\tassert.Equal(t, \"serper\", loaded.Search.Web.Provider)\n\n\t// Other configs should be nil\n\tassert.Nil(t, loaded.Search.KB)\n\tassert.Nil(t, loaded.Search.DB)\n\tassert.Nil(t, loaded.Search.Keyword)\n\tassert.Nil(t, loaded.Search.QueryDSL)\n\tassert.Nil(t, loaded.Search.Rerank)\n\tassert.Nil(t, loaded.Search.Citation)\n\tassert.Nil(t, loaded.Search.Weights)\n\tassert.Nil(t, loaded.Search.Options)\n}\n\n// TestLoadStoreWithoutSearchConfig tests loading assistant without search configuration\nfunc TestLoadStoreWithoutSearchConfig(t *testing.T) {\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\tassistantID := \"test.store-no-search\"\n\tnow := time.Now().UnixNano()\n\n\tast := &assistant.Assistant{\n\t\tAssistantModel: store.AssistantModel{\n\t\t\tID: assistantID,\n\t\t\tName: \"Test No Search Config\",\n\t\t\tType: \"assistant\",\n\t\t\tConnector: \"gpt-4o\",\n\t\t\t// No Search config\n\t\t\tCreatedAt: now,\n\t\t\tUpdatedAt: now,\n\t\t},\n\t}\n\n\terr := ast.Save()\n\trequire.NoError(t, err)\n\n\tdefer func() {\n\t\tstorage := assistant.GetStorage()\n\t\tif storage != nil {\n\t\t\tstorage.DeleteAssistant(assistantID)\n\t\t}\n\t\tassistant.GetCache().Clear()\n\t}()\n\n\tassistant.GetCache().Clear()\n\n\tloaded, err := assistant.Get(assistantID)\n\trequire.NoError(t, err)\n\trequire.NotNil(t, loaded)\n\n\t// Search config should be nil\n\tassert.Nil(t, loaded.Search)\n}\n" }, { "path": "agent/assistant/load_system.go", "content": "package assistant\n\nimport (\n\t\"fmt\"\n\t\"path/filepath\"\n\t\"strings\"\n\n\t\"github.com/yaoapp/gou/application\"\n\t\"github.com/yaoapp/gou/connector\"\n\tgouOpenAI \"github.com/yaoapp/gou/connector/openai\"\n\t\"github.com/yaoapp/kun/log\"\n\t\"github.com/yaoapp/yao/agent/i18n\"\n\tstore \"github.com/yaoapp/yao/agent/store/types\"\n\t\"github.com/yaoapp/yao/data\"\n\t\"gopkg.in/yaml.v3\"\n)\n\n// systemAgents defines the system agents loaded from bindata\n// These are internal agents used by the system (e.g., keyword extraction, querydsl generation)\n// The directory name is without __yao. prefix, prefix is added during loading\n// Format: directory name -> bindata path prefix\nvar systemAgents = []string{\n\t\"keyword\",\n\t\"querydsl\",\n\t\"title\",\n\t\"prompt\",\n\t\"robot_prompt\",\n\t\"needsearch\",\n\t\"entity\",\n}\n\n// SystemConfig holds the system agents connector configuration\n// This is set from agent.yml system block\ntype SystemConfig struct {\n\tDefault string // Default connector for all system agents\n\tKeyword string // Connector for __yao.keyword agent\n\tQueryDSL string // Connector for __yao.querydsl agent\n\tTitle string // Connector for __yao.title agent\n\tPrompt string // Connector for __yao.prompt agent\n\tRobotPrompt string // Connector for __yao.robot_prompt agent\n\tNeedSearch string // Connector for __yao.needsearch agent\n\tEntity string // Connector for __yao.entity agent\n}\n\n// systemConfig holds the system agents configuration (global variable like others in load.go)\nvar systemConfig *SystemConfig = nil\n\n// SetSystemConfig sets the system agents configuration\nfunc SetSystemConfig(config *SystemConfig) {\n\tsystemConfig = config\n}\n\n// GetSystemConfig returns the system agents configuration\nfunc GetSystemConfig() *SystemConfig {\n\treturn systemConfig\n}\n\n// LoadSystemAgents loads the system agents from bindata\n// These are internal agents like __yao.keyword and __yao.querydsl\n// They are loaded before application assistants\n// Behavior is same as LoadBuiltIn, just reads from bindata instead of filesystem\nfunc LoadSystemAgents() error {\n\n\t// Get all existing system agents (for cleanup)\n\tdeletedSystem := map[string]bool{}\n\tif storage != nil {\n\t\t// System agents have \"system\" tag\n\t\ttags := []string{\"system\"}\n\t\tbuiltIn := true\n\t\tres, err := storage.GetAssistants(store.AssistantFilter{\n\t\t\tTags: tags,\n\t\t\tBuiltIn: &builtIn,\n\t\t\tSelect: []string{\"assistant_id\", \"id\"},\n\t\t})\n\t\tif err != nil {\n\t\t\tlog.Warn(\"Failed to get existing system agents: %v\", err)\n\t\t} else {\n\t\t\tfor _, assistant := range res.Data {\n\t\t\t\tdeletedSystem[assistant.ID] = true\n\t\t\t}\n\t\t}\n\t}\n\n\tsort := 1\n\tfor _, name := range systemAgents {\n\t\t// Build agent ID with __yao. prefix\n\t\tid := \"__yao.\" + name\n\t\tpathPrefix := \"yao/assistants/\" + name\n\n\t\tassistant, err := loadSystemAgent(id, pathPrefix)\n\t\tif err != nil {\n\t\t\tlog.Warn(\"Failed to load system agent %s: %v\", id, err)\n\t\t\tcontinue\n\t\t}\n\n\t\t// Set sort order\n\t\tif assistant.Sort == 0 {\n\t\t\tassistant.Sort = sort\n\t\t}\n\n\t\t// Save to storage\n\t\tif err := assistant.Save(); err != nil {\n\t\t\tlog.Warn(\"Failed to save system agent %s: %v\", id, err)\n\t\t\tcontinue\n\t\t}\n\n\t\t// Initialize the assistant\n\t\tif err := assistant.initialize(); err != nil {\n\t\t\tlog.Warn(\"Failed to initialize system agent %s: %v\", id, err)\n\t\t\tcontinue\n\t\t}\n\n\t\tsort++\n\t\tloaded.Put(assistant)\n\t\tlog.Trace(\"Loaded system agent: %s\", id)\n\n\t\t// Remove from deleted list\n\t\tdelete(deletedSystem, id)\n\t}\n\n\t// Remove deleted system agents\n\tif len(deletedSystem) > 0 {\n\t\tassistantIDs := []string{}\n\t\tfor assistantID := range deletedSystem {\n\t\t\tassistantIDs = append(assistantIDs, assistantID)\n\t\t}\n\t\tif _, err := storage.DeleteAssistants(store.AssistantFilter{AssistantIDs: assistantIDs}); err != nil {\n\t\t\tlog.Warn(\"Failed to delete obsolete system agents: %v\", err)\n\t\t}\n\t}\n\n\treturn nil\n}\n\n// loadSystemAgent loads a single system agent from bindata\n// This follows the same pattern as LoadPath but reads from bindata\nfunc loadSystemAgent(id, pathPrefix string) (*Assistant, error) {\n\t// Read package.yao from bindata\n\tpkgPath := pathPrefix + \"/package.yao\"\n\tpkgContent, err := data.Read(pkgPath)\n\tif err != nil {\n\t\treturn nil, fmt.Errorf(\"failed to read %s: %w\", pkgPath, err)\n\t}\n\n\t// Parse package.yao\n\tvar pkgData map[string]interface{}\n\tif err := application.Parse(pkgPath, pkgContent, &pkgData); err != nil {\n\t\treturn nil, fmt.Errorf(\"failed to parse %s: %w\", pkgPath, err)\n\t}\n\n\t// Set assistant_id (no path - system agents are loaded from storage, not filesystem)\n\tpkgData[\"assistant_id\"] = id\n\n\t// Set type if not specified\n\tif _, has := pkgData[\"type\"]; !has {\n\t\tpkgData[\"type\"] = \"assistant\"\n\t}\n\n\t// Resolve connector for this system agent\n\tconnectorID := resolveSystemConnector(id)\n\tif connectorID != \"\" {\n\t\tpkgData[\"connector\"] = connectorID\n\t}\n\n\t// Read prompts.yml from bindata (default prompts)\n\tpromptsPath := pathPrefix + \"/prompts.yml\"\n\tpromptsContent, err := data.Read(promptsPath)\n\tif err == nil {\n\t\tvar prompts []store.Prompt\n\t\tif err := yaml.Unmarshal(promptsContent, &prompts); err == nil && len(prompts) > 0 {\n\t\t\tpkgData[\"prompts\"] = prompts\n\t\t}\n\t}\n\n\t// Read prompt_presets from prompts directory\n\tpresets := loadSystemPromptPresets(pathPrefix)\n\tif len(presets) > 0 {\n\t\tpkgData[\"prompt_presets\"] = presets\n\t}\n\n\t// Load scripts from src directory (hook script source and other scripts sources)\n\t// These will be compiled by loadMap -> LoadScriptsFromData\n\thookScriptSource, scriptsSource := loadSystemScripts(pathPrefix)\n\tif hookScriptSource != \"\" {\n\t\tpkgData[\"script\"] = hookScriptSource\n\t}\n\tif len(scriptsSource) > 0 {\n\t\tpkgData[\"scripts\"] = scriptsSource\n\t}\n\n\t// Read locales\n\tlocales, err := loadSystemLocales(pathPrefix)\n\tif err == nil && len(locales) > 0 {\n\t\tpkgData[\"locales\"] = locales\n\t}\n\n\t// Mark as system agent\n\tpkgData[\"readonly\"] = true\n\tpkgData[\"built_in\"] = true\n\tpkgData[\"tags\"] = []string{\"system\"}\n\n\t// Load from map (same as LoadPath, includes initialize())\n\treturn loadMap(pkgData)\n}\n\n// resolveSystemConnector resolves the connector for a system agent\n// Priority: specific agent config > system.default > defaultConnector > fallback to first capable connector\nfunc resolveSystemConnector(agentID string) string {\n\t// Try specific agent config first\n\tif systemConfig != nil {\n\t\tswitch agentID {\n\t\tcase \"__yao.keyword\":\n\t\t\tif systemConfig.Keyword != \"\" {\n\t\t\t\treturn systemConfig.Keyword\n\t\t\t}\n\t\tcase \"__yao.querydsl\":\n\t\t\tif systemConfig.QueryDSL != \"\" {\n\t\t\t\treturn systemConfig.QueryDSL\n\t\t\t}\n\t\tcase \"__yao.title\":\n\t\t\tif systemConfig.Title != \"\" {\n\t\t\t\treturn systemConfig.Title\n\t\t\t}\n\t\tcase \"__yao.prompt\":\n\t\t\tif systemConfig.Prompt != \"\" {\n\t\t\t\treturn systemConfig.Prompt\n\t\t\t}\n\t\tcase \"__yao.robot_prompt\":\n\t\t\tif systemConfig.RobotPrompt != \"\" {\n\t\t\t\treturn systemConfig.RobotPrompt\n\t\t\t}\n\t\tcase \"__yao.needsearch\":\n\t\t\tif systemConfig.NeedSearch != \"\" {\n\t\t\t\treturn systemConfig.NeedSearch\n\t\t\t}\n\t\tcase \"__yao.entity\":\n\t\t\tif systemConfig.Entity != \"\" {\n\t\t\t\treturn systemConfig.Entity\n\t\t\t}\n\t\t}\n\n\t\t// Try system default\n\t\tif systemConfig.Default != \"\" {\n\t\t\treturn systemConfig.Default\n\t\t}\n\t}\n\n\t// Try global default connector\n\tif defaultConnector != \"\" {\n\t\treturn defaultConnector\n\t}\n\n\t// Fallback: find first connector that supports tool calling\n\treturn findCapableConnector()\n}\n\n// findCapableConnector finds the first connector that supports tool calling\nfunc findCapableConnector() string {\n\tfor id, conn := range connector.Connectors {\n\t\tif !conn.Is(connector.OPENAI) {\n\t\t\tcontinue\n\t\t}\n\n\t\tif connOpenAI, ok := conn.(*gouOpenAI.Connector); ok {\n\t\t\tif connOpenAI.Options.Capabilities != nil && connOpenAI.Options.Capabilities.ToolCalls {\n\t\t\t\treturn id\n\t\t\t}\n\t\t}\n\t}\n\n\treturn \"\"\n}\n\n// loadSystemPromptPresets loads prompt presets from bindata prompts directory\nfunc loadSystemPromptPresets(pathPrefix string) map[string][]store.Prompt {\n\tpresets := make(map[string][]store.Prompt)\n\tpromptsDir := pathPrefix + \"/prompts\"\n\n\t// Try common preset files\n\tpresetFiles := []string{\"chat.yml\", \"task.yml\", \"code.yml\", \"analysis.yml\"}\n\tfor _, filename := range presetFiles {\n\t\tpresetPath := promptsDir + \"/\" + filename\n\t\tcontent, err := data.Read(presetPath)\n\t\tif err != nil {\n\t\t\tcontinue\n\t\t}\n\n\t\tvar prompts []store.Prompt\n\t\tif err := yaml.Unmarshal(content, &prompts); err == nil && len(prompts) > 0 {\n\t\t\tpresetName := strings.TrimSuffix(filename, \".yml\")\n\t\t\tpresets[presetName] = prompts\n\t\t}\n\t}\n\n\treturn presets\n}\n\n// loadSystemScripts loads scripts source from bindata src directory\n// Returns hook script source and other scripts sources (as strings)\n// These will be compiled by loadMap -> LoadScriptsFromData\nfunc loadSystemScripts(pathPrefix string) (string, map[string]string) {\n\tsrcDir := pathPrefix + \"/src\"\n\n\t// Try to load hook script (index.ts)\n\tvar hookScriptSource string\n\tindexPath := srcDir + \"/index.ts\"\n\tindexContent, err := data.Read(indexPath)\n\tif err == nil && len(indexContent) > 0 {\n\t\thookScriptSource = string(indexContent)\n\t}\n\n\t// Try to load other scripts\n\tscripts := make(map[string]string)\n\tscriptFiles := []string{\"utils.ts\", \"helpers.ts\", \"tools.ts\"}\n\tfor _, filename := range scriptFiles {\n\t\tscriptPath := srcDir + \"/\" + filename\n\t\tcontent, err := data.Read(scriptPath)\n\t\tif err != nil {\n\t\t\tcontinue\n\t\t}\n\n\t\tscriptName := strings.TrimSuffix(filename, \".ts\")\n\t\tscripts[scriptName] = string(content)\n\t}\n\n\tif len(scripts) == 0 {\n\t\tscripts = nil\n\t}\n\n\treturn hookScriptSource, scripts\n}\n\n// loadSystemLocales loads locales from bindata\nfunc loadSystemLocales(pathPrefix string) (i18n.Map, error) {\n\tlocales := make(i18n.Map)\n\n\t// Try to load common locale files\n\tlocaleFiles := []string{\"en-us.yml\", \"zh-cn.yml\", \"en.yml\", \"zh.yml\"}\n\tlocalesDir := pathPrefix + \"/locales\"\n\n\tfor _, filename := range localeFiles {\n\t\tlocalePath := filepath.Join(localesDir, filename)\n\t\tcontent, err := data.Read(localePath)\n\t\tif err != nil {\n\t\t\tcontinue\n\t\t}\n\n\t\t// Parse locale file\n\t\tlocale := strings.TrimSuffix(filename, \".yml\")\n\t\tvar messages map[string]any\n\t\tif err := yaml.Unmarshal(content, &messages); err != nil {\n\t\t\tcontinue\n\t\t}\n\n\t\tlocales[locale] = i18n.I18n{\n\t\t\tLocale: locale,\n\t\t\tMessages: messages,\n\t\t}\n\t}\n\n\treturn locales, nil\n}\n" }, { "path": "agent/assistant/load_test.go", "content": "package assistant_test\n\nimport (\n\t\"testing\"\n\n\t\"github.com/stretchr/testify/assert\"\n\t\"github.com/stretchr/testify/require\"\n\t\"github.com/yaoapp/yao/agent\"\n\t\"github.com/yaoapp/yao/agent/assistant\"\n\tstore \"github.com/yaoapp/yao/agent/store/types\"\n\t\"github.com/yaoapp/yao/config\"\n\t\"github.com/yaoapp/yao/test\"\n)\n\nfunc prepare(t *testing.T) {\n\ttest.Prepare(t, config.Conf)\n}\n\nfunc prepareAgent(t *testing.T) {\n\ttest.Prepare(t, config.Conf)\n\terr := agent.Load(config.Conf)\n\trequire.NoError(t, err, \"agent.Load should succeed\")\n}\n\n// TestLoadPath tests loading assistant from path\nfunc TestLoadPath(t *testing.T) {\n\tprepare(t)\n\tdefer test.Clean()\n\n\tt.Run(\"LoadFullFieldsAssistant\", func(t *testing.T) {\n\t\tassistant, err := assistant.LoadPath(\"/assistants/tests/fullfields\")\n\t\trequire.NoError(t, err)\n\t\trequire.NotNil(t, assistant)\n\n\t\t// Basic fields\n\t\tassert.Equal(t, \"tests.fullfields\", assistant.ID)\n\t\tassert.Equal(t, \"Full Fields Test Assistant\", assistant.Name)\n\t\tassert.Equal(t, \"assistant\", assistant.Type)\n\t\tassert.Equal(t, \"/api/__yao/app/icons/app.png\", assistant.Avatar)\n\t\tassert.Equal(t, \"gpt-4o\", assistant.Connector)\n\t\tassert.Equal(t, \"/assistants/tests/fullfields\", assistant.Path)\n\t\tassert.Equal(t, \"Test assistant with all available fields for unit testing\", assistant.Description)\n\n\t\t// Boolean fields\n\t\tassert.True(t, assistant.Public)\n\t\tassert.True(t, assistant.Readonly)\n\t\tassert.True(t, assistant.Mentionable)\n\t\tassert.False(t, assistant.Automated)\n\t\tassert.True(t, assistant.DisableGlobalPrompts)\n\n\t\t// Share field\n\t\tassert.Equal(t, \"team\", assistant.Share)\n\n\t\t// Sort field\n\t\tassert.Equal(t, 100, assistant.Sort)\n\n\t\t// Tags\n\t\tassert.NotNil(t, assistant.Tags)\n\t\tassert.Contains(t, assistant.Tags, \"Test\")\n\t\tassert.Contains(t, assistant.Tags, \"Development\")\n\t\tassert.Contains(t, assistant.Tags, \"FullFields\")\n\n\t\t// Options\n\t\tassert.NotNil(t, assistant.Options)\n\t\tassert.Equal(t, 0.7, assistant.Options[\"temperature\"])\n\t\tassert.Equal(t, float64(2000), assistant.Options[\"max_tokens\"])\n\n\t\t// Prompts (default prompts from prompts.yml)\n\t\tassert.NotNil(t, assistant.Prompts)\n\t\tassert.GreaterOrEqual(t, len(assistant.Prompts), 1)\n\t\tassert.Equal(t, \"system\", assistant.Prompts[0].Role)\n\n\t\t// Script (from src/index.ts)\n\t\tassert.NotNil(t, assistant.HookScript)\n\t})\n\n\tt.Run(\"LoadConnectorOptions\", func(t *testing.T) {\n\t\tassistant, err := assistant.LoadPath(\"/assistants/tests/fullfields\")\n\t\trequire.NoError(t, err)\n\t\trequire.NotNil(t, assistant)\n\n\t\t// ConnectorOptions\n\t\tassert.NotNil(t, assistant.ConnectorOptions)\n\t\tassert.NotNil(t, assistant.ConnectorOptions.Optional)\n\t\tassert.True(t, *assistant.ConnectorOptions.Optional)\n\t\tassert.NotNil(t, assistant.ConnectorOptions.Connectors)\n\t\tassert.Contains(t, assistant.ConnectorOptions.Connectors, \"gpt-4o\")\n\t\tassert.Contains(t, assistant.ConnectorOptions.Connectors, \"gpt-4o-mini\")\n\t\tassert.Contains(t, assistant.ConnectorOptions.Connectors, \"deepseek\")\n\t\tassert.NotNil(t, assistant.ConnectorOptions.Filters)\n\t\tassert.Len(t, assistant.ConnectorOptions.Filters, 2)\n\t})\n\n\tt.Run(\"LoadPromptPresets\", func(t *testing.T) {\n\t\tassistant, err := assistant.LoadPath(\"/assistants/tests/fullfields\")\n\t\trequire.NoError(t, err)\n\t\trequire.NotNil(t, assistant)\n\n\t\t// PromptPresets (from prompts directory)\n\t\tassert.NotNil(t, assistant.PromptPresets)\n\n\t\t// Top-level presets: chat.yml -> \"chat\", task.yml -> \"task\"\n\t\tchatPreset, hasChat := assistant.PromptPresets[\"chat\"]\n\t\tassert.True(t, hasChat, \"Should have 'chat' preset\")\n\t\tassert.NotEmpty(t, chatPreset)\n\n\t\ttaskPreset, hasTask := assistant.PromptPresets[\"task\"]\n\t\tassert.True(t, hasTask, \"Should have 'task' preset\")\n\t\tassert.NotEmpty(t, taskPreset)\n\n\t\t// Nested presets: chat/friendly.yml -> \"chat.friendly\"\n\t\tfriendlyPreset, hasFriendly := assistant.PromptPresets[\"chat.friendly\"]\n\t\tassert.True(t, hasFriendly, \"Should have 'chat.friendly' preset\")\n\t\tassert.NotEmpty(t, friendlyPreset)\n\n\t\tprofessionalPreset, hasProfessional := assistant.PromptPresets[\"chat.professional\"]\n\t\tassert.True(t, hasProfessional, \"Should have 'chat.professional' preset\")\n\t\tassert.NotEmpty(t, professionalPreset)\n\n\t\t// task/analysis.yml -> \"task.analysis\"\n\t\tanalysisPreset, hasAnalysis := assistant.PromptPresets[\"task.analysis\"]\n\t\tassert.True(t, hasAnalysis, \"Should have 'task.analysis' preset\")\n\t\tassert.NotEmpty(t, analysisPreset)\n\t})\n\n\tt.Run(\"LoadKnowledgeBase\", func(t *testing.T) {\n\t\tassistant, err := assistant.LoadPath(\"/assistants/tests/fullfields\")\n\t\trequire.NoError(t, err)\n\t\trequire.NotNil(t, assistant)\n\n\t\t// KB\n\t\tassert.NotNil(t, assistant.KB)\n\t\tassert.NotNil(t, assistant.KB.Collections)\n\t\tassert.Contains(t, assistant.KB.Collections, \"test-collection\")\n\t\tassert.NotNil(t, assistant.KB.Options)\n\t\tassert.Equal(t, float64(5), assistant.KB.Options[\"top_k\"])\n\t})\n\n\tt.Run(\"LoadMCPServers\", func(t *testing.T) {\n\t\tassistant, err := assistant.LoadPath(\"/assistants/tests/fullfields\")\n\t\trequire.NoError(t, err)\n\t\trequire.NotNil(t, assistant)\n\n\t\t// MCP\n\t\tassert.NotNil(t, assistant.MCP)\n\t\tassert.NotNil(t, assistant.MCP.Servers)\n\t\tassert.Len(t, assistant.MCP.Servers, 1)\n\t\tassert.Equal(t, \"echo\", assistant.MCP.Servers[0].ServerID)\n\t\tassert.Contains(t, assistant.MCP.Servers[0].Tools, \"ping\")\n\t\tassert.Contains(t, assistant.MCP.Servers[0].Tools, \"echo\")\n\t})\n\n\tt.Run(\"LoadWorkflow\", func(t *testing.T) {\n\t\tassistant, err := assistant.LoadPath(\"/assistants/tests/fullfields\")\n\t\trequire.NoError(t, err)\n\t\trequire.NotNil(t, assistant)\n\n\t\t// Workflow\n\t\tassert.NotNil(t, assistant.Workflow)\n\t\tassert.NotNil(t, assistant.Workflow.Workflows)\n\t\tassert.Contains(t, assistant.Workflow.Workflows, \"test-workflow\")\n\t\tassert.NotNil(t, assistant.Workflow.Options)\n\t\tassert.Equal(t, float64(10), assistant.Workflow.Options[\"max_steps\"])\n\t})\n\n\tt.Run(\"LoadPlaceholder\", func(t *testing.T) {\n\t\tassistant, err := assistant.LoadPath(\"/assistants/tests/fullfields\")\n\t\trequire.NoError(t, err)\n\t\trequire.NotNil(t, assistant)\n\n\t\t// Placeholder\n\t\tassert.NotNil(t, assistant.Placeholder)\n\t\tassert.Equal(t, \"Full Fields Test\", assistant.Placeholder.Title)\n\t\tassert.Equal(t, \"Test assistant with complete field coverage\", assistant.Placeholder.Description)\n\t\tassert.NotNil(t, assistant.Placeholder.Prompts)\n\t\tassert.Len(t, assistant.Placeholder.Prompts, 3)\n\t})\n\n\tt.Run(\"LoadLocales\", func(t *testing.T) {\n\t\tassistant, err := assistant.LoadPath(\"/assistants/tests/fullfields\")\n\t\trequire.NoError(t, err)\n\t\trequire.NotNil(t, assistant)\n\n\t\t// Locales\n\t\tassert.NotNil(t, assistant.Locales)\n\n\t\tenLocale, hasEn := assistant.Locales[\"en-us\"]\n\t\tassert.True(t, hasEn, \"Should have en-us locale\")\n\t\tassert.NotNil(t, enLocale)\n\n\t\tzhLocale, hasZh := assistant.Locales[\"zh-cn\"]\n\t\tassert.True(t, hasZh, \"Should have zh-cn locale\")\n\t\tassert.NotNil(t, zhLocale)\n\t})\n\n\tt.Run(\"LoadDependencies\", func(t *testing.T) {\n\t\tassistant, err := assistant.LoadPath(\"/assistants/tests/fullfields\")\n\t\trequire.NoError(t, err)\n\t\trequire.NotNil(t, assistant)\n\n\t\t// Dependencies\n\t\tassert.NotNil(t, assistant.Dependencies)\n\t\tassert.Len(t, assistant.Dependencies, 2)\n\t\tassert.Equal(t, \"^1.0.0\", assistant.Dependencies[\"echo\"])\n\t\tassert.Equal(t, \">=2.0.0\", assistant.Dependencies[\"customer\"])\n\t})\n\n\tt.Run(\"LoadNonExistentAssistant\", func(t *testing.T) {\n\t\t_, err := assistant.LoadPath(\"/assistants/non-existent\")\n\t\tassert.Error(t, err)\n\t})\n}\n\n// TestLoadPathMCPTest tests loading the MCP test assistant\nfunc TestLoadPathMCPTest(t *testing.T) {\n\tprepare(t)\n\tdefer test.Clean()\n\n\tassistant, err := assistant.LoadPath(\"/assistants/tests/mcptest\")\n\trequire.NoError(t, err)\n\trequire.NotNil(t, assistant)\n\n\tassert.Equal(t, \"tests.mcptest\", assistant.ID)\n\tassert.Equal(t, \"MCP Test Assistant\", assistant.Name)\n\tassert.Equal(t, \"gpt-4o\", assistant.Connector)\n\n\t// MCP configuration\n\tassert.NotNil(t, assistant.MCP)\n\tassert.Len(t, assistant.MCP.Servers, 1)\n\tassert.Equal(t, \"echo\", assistant.MCP.Servers[0].ServerID)\n\n\t// Locales\n\tassert.NotNil(t, assistant.Locales)\n\tassert.Contains(t, assistant.Locales, \"en-us\")\n\tassert.Contains(t, assistant.Locales, \"zh-cn\")\n}\n\n// TestLoadPathBuildRequest tests loading the build request test assistant\nfunc TestLoadPathBuildRequest(t *testing.T) {\n\tprepare(t)\n\tdefer test.Clean()\n\n\tassistant, err := assistant.LoadPath(\"/assistants/tests/buildrequest\")\n\trequire.NoError(t, err)\n\trequire.NotNil(t, assistant)\n\n\tassert.Equal(t, \"tests.buildrequest\", assistant.ID)\n\tassert.Equal(t, \"Build Request Test\", assistant.Name)\n\n\t// HookScript should be loaded\n\tassert.NotNil(t, assistant.HookScript)\n\n\t// Options\n\tassert.NotNil(t, assistant.Options)\n\tassert.Equal(t, 0.5, assistant.Options[\"temperature\"])\n}\n\n// TestCache tests the assistant cache functionality\nfunc TestCache(t *testing.T) {\n\t// Clear any existing cache\n\tassistant.ClearCache()\n\n\t// Set small cache for testing\n\tassistant.SetCache(3)\n\tassert.NotNil(t, assistant.GetCache())\n\n\t// Create test assistants\n\tast1 := &assistant.Assistant{AssistantModel: store.AssistantModel{ID: \"id1\", Name: \"Assistant 1\"}}\n\tast2 := &assistant.Assistant{AssistantModel: store.AssistantModel{ID: \"id2\", Name: \"Assistant 2\"}}\n\tast3 := &assistant.Assistant{AssistantModel: store.AssistantModel{ID: \"id3\", Name: \"Assistant 3\"}}\n\tast4 := &assistant.Assistant{AssistantModel: store.AssistantModel{ID: \"id4\", Name: \"Assistant 4\"}}\n\n\tt.Run(\"PutAndGet\", func(t *testing.T) {\n\t\tassistant.GetCache().Put(ast1)\n\t\tassert.Equal(t, 1, assistant.GetCache().Len())\n\n\t\tcached, exists := assistant.GetCache().Get(\"id1\")\n\t\tassert.True(t, exists)\n\t\tassert.Equal(t, ast1, cached)\n\t})\n\n\tt.Run(\"CacheEviction\", func(t *testing.T) {\n\t\tassistant.GetCache().Put(ast2)\n\t\tassistant.GetCache().Put(ast3)\n\t\tassert.Equal(t, 3, assistant.GetCache().Len())\n\n\t\t// Access ast1 to make it recently used\n\t\tassistant.GetCache().Get(\"id1\")\n\n\t\t// Add ast4, should evict ast2 (least recently used)\n\t\tassistant.GetCache().Put(ast4)\n\t\tassert.Equal(t, 3, assistant.GetCache().Len())\n\n\t\t_, exists := assistant.GetCache().Get(\"id2\")\n\t\tassert.False(t, exists, \"ast2 should be evicted\")\n\n\t\t_, exists = assistant.GetCache().Get(\"id1\")\n\t\tassert.True(t, exists, \"ast1 should still exist\")\n\n\t\t_, exists = assistant.GetCache().Get(\"id4\")\n\t\tassert.True(t, exists, \"ast4 should exist\")\n\t})\n\n\tt.Run(\"ClearCache\", func(t *testing.T) {\n\t\tassistant.ClearCache()\n\t\tassert.Nil(t, assistant.GetCache())\n\t})\n\n\tt.Run(\"SetCacheAfterClear\", func(t *testing.T) {\n\t\tassistant.SetCache(100)\n\t\tassert.NotNil(t, assistant.GetCache())\n\t})\n}\n\n// TestClone tests the assistant Clone method\nfunc TestClone(t *testing.T) {\n\tprepare(t)\n\tdefer test.Clean()\n\n\tt.Run(\"CloneFullFieldsAssistant\", func(t *testing.T) {\n\t\toriginal, err := assistant.LoadPath(\"/assistants/tests/fullfields\")\n\t\trequire.NoError(t, err)\n\n\t\tclone := original.Clone()\n\t\trequire.NotNil(t, clone)\n\n\t\t// Basic fields should be equal\n\t\tassert.Equal(t, original.ID, clone.ID)\n\t\tassert.Equal(t, original.Name, clone.Name)\n\t\tassert.Equal(t, original.Type, clone.Type)\n\t\tassert.Equal(t, original.Connector, clone.Connector)\n\t\tassert.Equal(t, original.Description, clone.Description)\n\n\t\t// Verify deep copy - modifying original should not affect clone\n\t\tif len(original.Tags) > 0 {\n\t\t\toriginalTag := original.Tags[0]\n\t\t\toriginal.Tags[0] = \"modified\"\n\t\t\tassert.NotEqual(t, original.Tags[0], clone.Tags[0])\n\t\t\toriginal.Tags[0] = originalTag // restore\n\t\t}\n\n\t\tif original.Options != nil {\n\t\t\toriginal.Options[\"test_key\"] = \"test_value\"\n\t\t\t_, exists := clone.Options[\"test_key\"]\n\t\t\tassert.False(t, exists, \"Clone should not have modified key\")\n\t\t\tdelete(original.Options, \"test_key\") // cleanup\n\t\t}\n\n\t\tif original.Dependencies != nil {\n\t\t\toriginal.Dependencies[\"test_dep\"] = \"^9.9.9\"\n\t\t\t_, exists := clone.Dependencies[\"test_dep\"]\n\t\t\tassert.False(t, exists, \"Clone dependencies should not have modified key\")\n\t\t\tdelete(original.Dependencies, \"test_dep\") // cleanup\n\t\t}\n\t})\n\n\tt.Run(\"CloneNil\", func(t *testing.T) {\n\t\tvar nilAssistant *assistant.Assistant\n\t\tassert.Nil(t, nilAssistant.Clone())\n\t})\n}\n\n// TestUpdate tests the assistant Update method\nfunc TestUpdate(t *testing.T) {\n\tprepare(t)\n\tdefer test.Clean()\n\n\tt.Run(\"UpdateBasicFields\", func(t *testing.T) {\n\t\tassistant, err := assistant.LoadPath(\"/assistants/tests/fullfields\")\n\t\trequire.NoError(t, err)\n\n\t\tupdates := map[string]interface{}{\n\t\t\t\"name\": \"Updated Name\",\n\t\t\t\"description\": \"Updated description\",\n\t\t\t\"tags\": []string{\"updated\", \"tags\"},\n\t\t}\n\n\t\terr = assistant.Update(updates)\n\t\trequire.NoError(t, err)\n\n\t\tassert.Equal(t, \"Updated Name\", assistant.Name)\n\t\tassert.Equal(t, \"Updated description\", assistant.Description)\n\t\tassert.Equal(t, []string{\"updated\", \"tags\"}, assistant.Tags)\n\t})\n\n\tt.Run(\"UpdateConnectorOptions\", func(t *testing.T) {\n\t\tassistant, err := assistant.LoadPath(\"/assistants/tests/fullfields\")\n\t\trequire.NoError(t, err)\n\n\t\tupdates := map[string]interface{}{\n\t\t\t\"connector_options\": map[string]interface{}{\n\t\t\t\t\"optional\": false,\n\t\t\t\t\"connectors\": []string{\"new-connector\"},\n\t\t\t},\n\t\t}\n\n\t\terr = assistant.Update(updates)\n\t\trequire.NoError(t, err)\n\n\t\tassert.NotNil(t, assistant.ConnectorOptions)\n\t\tassert.NotNil(t, assistant.ConnectorOptions.Optional)\n\t\tassert.False(t, *assistant.ConnectorOptions.Optional)\n\t\tassert.Contains(t, assistant.ConnectorOptions.Connectors, \"new-connector\")\n\t})\n\n\tt.Run(\"UpdatePromptPresets\", func(t *testing.T) {\n\t\tassistant, err := assistant.LoadPath(\"/assistants/tests/fullfields\")\n\t\trequire.NoError(t, err)\n\n\t\tupdates := map[string]interface{}{\n\t\t\t\"prompt_presets\": map[string]interface{}{\n\t\t\t\t\"custom\": []map[string]interface{}{\n\t\t\t\t\t{\"role\": \"system\", \"content\": \"Custom preset\"},\n\t\t\t\t},\n\t\t\t},\n\t\t}\n\n\t\terr = assistant.Update(updates)\n\t\trequire.NoError(t, err)\n\n\t\tassert.NotNil(t, assistant.PromptPresets)\n\t\tcustomPreset, exists := assistant.PromptPresets[\"custom\"]\n\t\tassert.True(t, exists)\n\t\tassert.Len(t, customPreset, 1)\n\t})\n\n\tt.Run(\"UpdateSource\", func(t *testing.T) {\n\t\tassistant, err := assistant.LoadPath(\"/assistants/tests/fullfields\")\n\t\trequire.NoError(t, err)\n\n\t\tupdates := map[string]interface{}{\n\t\t\t\"source\": \"function Create(ctx, messages) { return { messages: messages }; }\",\n\t\t}\n\n\t\terr = assistant.Update(updates)\n\t\trequire.NoError(t, err)\n\n\t\tassert.Equal(t, \"function Create(ctx, messages) { return { messages: messages }; }\", assistant.Source)\n\t})\n\n\tt.Run(\"UpdateNilAssistant\", func(t *testing.T) {\n\t\tvar nilAssistant *assistant.Assistant\n\t\terr := nilAssistant.Update(map[string]interface{}{\"name\": \"test\"})\n\t\tassert.Error(t, err)\n\t})\n}\n\n// TestMap tests the assistant Map method\nfunc TestMap(t *testing.T) {\n\tprepare(t)\n\tdefer test.Clean()\n\n\tassistant, err := assistant.LoadPath(\"/assistants/tests/fullfields\")\n\trequire.NoError(t, err)\n\n\tm := assistant.Map()\n\trequire.NotNil(t, m)\n\n\t// Check all fields are present\n\tassert.Equal(t, assistant.ID, m[\"assistant_id\"])\n\tassert.Equal(t, assistant.Name, m[\"name\"])\n\tassert.Equal(t, assistant.Type, m[\"type\"])\n\tassert.Equal(t, assistant.Connector, m[\"connector\"])\n\tassert.Equal(t, assistant.Description, m[\"description\"])\n\tassert.Equal(t, assistant.Path, m[\"path\"])\n\tassert.Equal(t, assistant.Tags, m[\"tags\"])\n\tassert.Equal(t, assistant.Options, m[\"options\"])\n\tassert.Equal(t, assistant.Prompts, m[\"prompts\"])\n\tassert.Equal(t, assistant.KB, m[\"kb\"])\n\tassert.Equal(t, assistant.MCP, m[\"mcp\"])\n\tassert.Equal(t, assistant.Workflow, m[\"workflow\"])\n\tassert.Equal(t, assistant.Placeholder, m[\"placeholder\"])\n\tassert.Equal(t, assistant.Locales, m[\"locales\"])\n\n\t// New fields\n\tassert.Equal(t, assistant.ConnectorOptions, m[\"connector_options\"])\n\tassert.Equal(t, assistant.PromptPresets, m[\"prompt_presets\"])\n\tassert.Equal(t, assistant.Source, m[\"source\"])\n\tassert.Equal(t, assistant.Dependencies, m[\"dependencies\"])\n}\n\n// TestLoadSystemAgents tests loading system agents from bindata\nfunc TestLoadSystemAgents(t *testing.T) {\n\tprepareAgent(t)\n\tdefer test.Clean()\n\n\t// Clear cache first\n\tassistant.ClearCache()\n\tassistant.SetCache(200)\n\n\tt.Run(\"LoadSystemAgents\", func(t *testing.T) {\n\t\terr := assistant.LoadSystemAgents()\n\t\trequire.NoError(t, err)\n\n\t\t// Check __yao.keyword\n\t\tkeywordAst, keywordExists := assistant.GetCache().Get(\"__yao.keyword\")\n\t\trequire.True(t, keywordExists, \"__yao.keyword should be loaded\")\n\t\tassert.Equal(t, \"__yao.keyword\", keywordAst.ID)\n\t\tassert.Equal(t, \"Keyword Extractor\", keywordAst.Name)\n\t\tassert.True(t, keywordAst.Readonly)\n\t\tassert.True(t, keywordAst.BuiltIn)\n\t\tassert.Contains(t, keywordAst.Tags, \"system\")\n\t\tassert.NotNil(t, keywordAst.Prompts)\n\t\tassert.Greater(t, len(keywordAst.Prompts), 0)\n\n\t\t// Check __yao.querydsl\n\t\tquerydslAst, querydslExists := assistant.GetCache().Get(\"__yao.querydsl\")\n\t\trequire.True(t, querydslExists, \"__yao.querydsl should be loaded\")\n\t\tassert.Equal(t, \"__yao.querydsl\", querydslAst.ID)\n\t\tassert.Equal(t, \"Query Builder\", querydslAst.Name)\n\t\tassert.True(t, querydslAst.Readonly)\n\t\tassert.True(t, querydslAst.BuiltIn)\n\t\tassert.Contains(t, querydslAst.Tags, \"system\")\n\t\tassert.NotNil(t, querydslAst.Prompts)\n\t\tassert.Greater(t, len(querydslAst.Prompts), 0)\n\n\t\t// Check __yao.title\n\t\ttitleAst, titleExists := assistant.GetCache().Get(\"__yao.title\")\n\t\trequire.True(t, titleExists, \"__yao.title should be loaded\")\n\t\tassert.Equal(t, \"__yao.title\", titleAst.ID)\n\t\tassert.Equal(t, \"Title Generator\", titleAst.Name)\n\t\tassert.True(t, titleAst.Readonly)\n\t\tassert.True(t, titleAst.BuiltIn)\n\n\t\t// Check __yao.prompt\n\t\tpromptAst, promptExists := assistant.GetCache().Get(\"__yao.prompt\")\n\t\trequire.True(t, promptExists, \"__yao.prompt should be loaded\")\n\t\tassert.Equal(t, \"__yao.prompt\", promptAst.ID)\n\t\tassert.Equal(t, \"Prompt Optimizer\", promptAst.Name)\n\t\tassert.True(t, promptAst.Readonly)\n\t\tassert.True(t, promptAst.BuiltIn)\n\n\t\t// Check __yao.needsearch\n\t\tneedsearchAst, needsearchExists := assistant.GetCache().Get(\"__yao.needsearch\")\n\t\trequire.True(t, needsearchExists, \"__yao.needsearch should be loaded\")\n\t\tassert.Equal(t, \"__yao.needsearch\", needsearchAst.ID)\n\t\tassert.Equal(t, \"Reference Checker\", needsearchAst.Name)\n\t\tassert.True(t, needsearchAst.Readonly)\n\t\tassert.True(t, needsearchAst.BuiltIn)\n\t})\n\n\tt.Run(\"SystemAgentsSavedToStorage\", func(t *testing.T) {\n\t\t// System agents should be saved to storage\n\t\trequire.NotNil(t, assistant.GetStore(), \"storage should be initialized\")\n\n\t\t// Check __yao.keyword in storage\n\t\tbuiltIn := true\n\t\ttags := []string{\"system\"}\n\t\tres, err := assistant.GetStore().GetAssistants(store.AssistantFilter{\n\t\t\tBuiltIn: &builtIn,\n\t\t\tTags: tags,\n\t\t\tSelect: []string{\"assistant_id\", \"name\"},\n\t\t})\n\t\trequire.NoError(t, err)\n\t\trequire.Greater(t, len(res.Data), 0, \"System agents should be in storage\")\n\n\t\t// Verify at least one system agent exists\n\t\tfound := false\n\t\tfor _, ast := range res.Data {\n\t\t\tif ast.ID == \"__yao.keyword\" || ast.ID == \"__yao.querydsl\" {\n\t\t\t\tfound = true\n\t\t\t\tbreak\n\t\t\t}\n\t\t}\n\t\tassert.True(t, found, \"System agents should be found in storage\")\n\t})\n\n\tt.Run(\"SystemAgentsGetFromStorage\", func(t *testing.T) {\n\t\t// Clear cache to force loading from storage\n\t\tassistant.GetCache().Clear()\n\n\t\t// Test Get for each system agent\n\t\tsystemAgents := []string{\n\t\t\t\"__yao.keyword\",\n\t\t\t\"__yao.querydsl\",\n\t\t\t\"__yao.title\",\n\t\t\t\"__yao.prompt\",\n\t\t\t\"__yao.needsearch\",\n\t\t\t\"__yao.entity\",\n\t\t}\n\n\t\tfor _, agentID := range systemAgents {\n\t\t\tast, err := assistant.Get(agentID)\n\t\t\trequire.NoError(t, err, \"Get(%s) should succeed\", agentID)\n\t\t\trequire.NotNil(t, ast, \"Get(%s) should return assistant\", agentID)\n\t\t\tassert.Equal(t, agentID, ast.ID)\n\t\t\tassert.True(t, ast.BuiltIn, \"%s should be built-in\", agentID)\n\t\t\tassert.True(t, ast.Readonly, \"%s should be readonly\", agentID)\n\t\t\tassert.Contains(t, ast.Tags, \"system\", \"%s should have system tag\", agentID)\n\t\t\tassert.Equal(t, \"worker\", ast.Type, \"%s should be worker type\", agentID)\n\t\t\tassert.NotNil(t, ast.Prompts, \"%s should have prompts\", agentID)\n\t\t\tassert.Greater(t, len(ast.Prompts), 0, \"%s should have at least one prompt\", agentID)\n\t\t}\n\t})\n}\n\n// TestLoadPathSandboxV2 tests loading assistants with V2 sandbox configuration (standalone sandbox.yao)\nfunc TestLoadPathSandboxV2(t *testing.T) {\n\tprepare(t)\n\tdefer test.Clean()\n\n\tt.Run(\"OneshotCLI\", func(t *testing.T) {\n\t\tast, err := assistant.LoadPath(\"/assistants/tests/sandbox-v2/oneshot-cli\")\n\t\trequire.NoError(t, err)\n\t\trequire.NotNil(t, ast)\n\n\t\tassert.Equal(t, \"Sandbox V2 Oneshot CLI\", ast.Name)\n\t\tassert.Contains(t, ast.Tags, \"SandboxV2\")\n\n\t\t// V2 sandbox should be loaded from sandbox.yao\n\t\trequire.NotNil(t, ast.SandboxV2, \"SandboxV2 should be loaded\")\n\t\tassert.Equal(t, \"2.0\", ast.SandboxV2.Version)\n\t\tassert.Equal(t, \"yaoapp/tai-sandbox-claude:latest\", ast.SandboxV2.Computer.Image)\n\t\tassert.Equal(t, \"2GB\", ast.SandboxV2.Computer.Memory)\n\t\tassert.Equal(t, float64(2), ast.SandboxV2.Computer.CPUs)\n\t\tassert.Equal(t, \"/workspace\", ast.SandboxV2.Computer.WorkDir)\n\t\tassert.Equal(t, \"claude\", ast.SandboxV2.Runner.Name)\n\t\tassert.Equal(t, \"cli\", ast.SandboxV2.Runner.Mode)\n\t\tassert.Equal(t, \"oneshot\", ast.SandboxV2.Lifecycle)\n\n\t\t// Runner options\n\t\tassert.NotNil(t, ast.SandboxV2.Runner.Options)\n\t\tassert.Equal(t, float64(5), ast.SandboxV2.Runner.Options[\"max_turns\"])\n\n\t\t// V1 Sandbox should be nil\n\t\tassert.Nil(t, ast.Sandbox, \"V1 Sandbox should be nil when V2 is present\")\n\n\t\t// ConfigHash should be computed\n\t\tassert.NotEmpty(t, ast.ConfigHash, \"ConfigHash should be computed for V2 sandbox\")\n\n\t\t// HasSandboxV2 helper\n\t\tassert.True(t, ast.HasSandboxV2())\n\t})\n\n\tt.Run(\"SessionCLI\", func(t *testing.T) {\n\t\tast, err := assistant.LoadPath(\"/assistants/tests/sandbox-v2/session-cli\")\n\t\trequire.NoError(t, err)\n\t\trequire.NotNil(t, ast)\n\n\t\trequire.NotNil(t, ast.SandboxV2)\n\t\tassert.Equal(t, \"session\", ast.SandboxV2.Lifecycle)\n\t\tassert.Equal(t, \"10m\", ast.SandboxV2.IdleTimeout)\n\n\t\t// Prepare steps\n\t\trequire.Len(t, ast.SandboxV2.Prepare, 1)\n\t\tassert.Equal(t, \"exec\", ast.SandboxV2.Prepare[0].Action)\n\t\tassert.True(t, ast.SandboxV2.Prepare[0].Once)\n\t})\n\n\tt.Run(\"LongrunningCLI\", func(t *testing.T) {\n\t\tast, err := assistant.LoadPath(\"/assistants/tests/sandbox-v2/longrunning-cli\")\n\t\trequire.NoError(t, err)\n\t\trequire.NotNil(t, ast)\n\n\t\trequire.NotNil(t, ast.SandboxV2)\n\t\tassert.Equal(t, \"longrunning\", ast.SandboxV2.Lifecycle)\n\t\tassert.Equal(t, \"15m\", ast.SandboxV2.IdleTimeout)\n\t\tassert.Equal(t, \"2h\", ast.SandboxV2.MaxLifetime)\n\t\tassert.Equal(t, \"5s\", ast.SandboxV2.StopTimeout)\n\t\tassert.Equal(t, \"4GB\", ast.SandboxV2.Computer.Memory)\n\t\tassert.Equal(t, \"rw\", ast.SandboxV2.Computer.MountMode)\n\n\t\t// Environment\n\t\tassert.Equal(t, \"test\", ast.SandboxV2.Environment[\"NODE_ENV\"])\n\t\tassert.Equal(t, \"longrunning\", ast.SandboxV2.Environment[\"V2_TEST_MODE\"])\n\n\t\t// Secrets\n\t\tassert.Equal(t, \"sandbox-v2-longrunning-secret\", ast.SandboxV2.Secrets[\"TEST_SECRET\"])\n\n\t\t// Prepare steps\n\t\trequire.Len(t, ast.SandboxV2.Prepare, 3)\n\t\tassert.True(t, ast.SandboxV2.Prepare[2].IgnoreError)\n\n\t\t// MCP (from package.yao)\n\t\trequire.NotNil(t, ast.MCP)\n\t\trequire.Len(t, ast.MCP.Servers, 1)\n\t\tassert.Equal(t, \"echo\", ast.MCP.Servers[0].ServerID)\n\n\t\t// ConfigHash should include MCP servers\n\t\thashWithMCP := ast.ConfigHash\n\t\tassert.NotEmpty(t, hashWithMCP)\n\t})\n\n\tt.Run(\"HooksOnly_YaoRunner\", func(t *testing.T) {\n\t\tast, err := assistant.LoadPath(\"/assistants/tests/sandbox-v2/hooks-only\")\n\t\trequire.NoError(t, err)\n\t\trequire.NotNil(t, ast)\n\n\t\trequire.NotNil(t, ast.SandboxV2)\n\t\tassert.Equal(t, \"yao\", ast.SandboxV2.Runner.Name)\n\t\tassert.Equal(t, \"oneshot\", ast.SandboxV2.Lifecycle)\n\t\tassert.Equal(t, float64(1), ast.SandboxV2.Computer.CPUs)\n\n\t\t// Runner mode should be empty (yao runner ignores mode)\n\t\tassert.Empty(t, ast.SandboxV2.Runner.Mode)\n\t})\n\n\tt.Run(\"FullPrepare\", func(t *testing.T) {\n\t\tast, err := assistant.LoadPath(\"/assistants/tests/sandbox-v2/full-prepare\")\n\t\trequire.NoError(t, err)\n\t\trequire.NotNil(t, ast)\n\n\t\trequire.NotNil(t, ast.SandboxV2)\n\t\tassert.Equal(t, \"session\", ast.SandboxV2.Lifecycle)\n\t\tassert.Equal(t, \"15m\", ast.SandboxV2.IdleTimeout)\n\n\t\t// Prepare: 5 steps with mixed actions\n\t\trequire.Len(t, ast.SandboxV2.Prepare, 5)\n\t\tassert.Equal(t, \"copy\", ast.SandboxV2.Prepare[0].Action)\n\t\tassert.Equal(t, \"skills\", ast.SandboxV2.Prepare[0].Src)\n\t\tassert.Equal(t, \"~/.claude/skills\", ast.SandboxV2.Prepare[0].Dst)\n\t\tassert.Equal(t, \"exec\", ast.SandboxV2.Prepare[1].Action)\n\t\tassert.True(t, ast.SandboxV2.Prepare[1].Once)\n\t\tassert.True(t, ast.SandboxV2.Prepare[3].IgnoreError)\n\n\t\t// Environment + Secrets\n\t\tassert.Equal(t, \"full\", ast.SandboxV2.Environment[\"V2_PREPARE_TEST\"])\n\t\tassert.Equal(t, \"v2-full-prepare-key\", ast.SandboxV2.Secrets[\"TEST_API_KEY\"])\n\n\t\t// Runner options\n\t\tassert.Equal(t, \"acceptEdits\", ast.SandboxV2.Runner.Options[\"permission_mode\"])\n\t})\n\n\tt.Run(\"HostMode\", func(t *testing.T) {\n\t\tast, err := assistant.LoadPath(\"/assistants/tests/sandbox-v2/host-mode\")\n\t\trequire.NoError(t, err)\n\t\trequire.NotNil(t, ast)\n\n\t\trequire.NotNil(t, ast.SandboxV2)\n\t\t// Host mode: no image\n\t\tassert.Empty(t, ast.SandboxV2.Computer.Image)\n\t\tassert.Equal(t, \"/tmp/yao-sandbox-v2-host-test\", ast.SandboxV2.Computer.WorkDir)\n\t\tassert.Equal(t, \"session\", ast.SandboxV2.Lifecycle)\n\t})\n\n\tt.Run(\"ConfigHashDeterministic\", func(t *testing.T) {\n\t\tast1, err := assistant.LoadPath(\"/assistants/tests/sandbox-v2/oneshot-cli\")\n\t\trequire.NoError(t, err)\n\t\tast2, err := assistant.LoadPath(\"/assistants/tests/sandbox-v2/oneshot-cli\")\n\t\trequire.NoError(t, err)\n\t\tassert.Equal(t, ast1.ConfigHash, ast2.ConfigHash, \"same config should produce same hash\")\n\t})\n\n\tt.Run(\"ConfigHashDiffers\", func(t *testing.T) {\n\t\tast1, err := assistant.LoadPath(\"/assistants/tests/sandbox-v2/oneshot-cli\")\n\t\trequire.NoError(t, err)\n\t\tast2, err := assistant.LoadPath(\"/assistants/tests/sandbox-v2/longrunning-cli\")\n\t\trequire.NoError(t, err)\n\t\tassert.NotEqual(t, ast1.ConfigHash, ast2.ConfigHash, \"different configs should produce different hashes\")\n\t})\n}\n\n// TestValidate tests the assistant Validate method\nfunc TestValidate(t *testing.T) {\n\ttests := []struct {\n\t\tname string\n\t\tast *assistant.Assistant\n\t\twantErr bool\n\t}{\n\t\t{\n\t\t\tname: \"ValidAssistant\",\n\t\t\tast: &assistant.Assistant{\n\t\t\t\tAssistantModel: store.AssistantModel{\n\t\t\t\t\tID: \"test-id\",\n\t\t\t\t\tName: \"Test Assistant\",\n\t\t\t\t\tConnector: \"gpt-4o\",\n\t\t\t\t},\n\t\t\t},\n\t\t\twantErr: false,\n\t\t},\n\t\t{\n\t\t\tname: \"MissingID\",\n\t\t\tast: &assistant.Assistant{\n\t\t\t\tAssistantModel: store.AssistantModel{\n\t\t\t\t\tName: \"Test Assistant\",\n\t\t\t\t\tConnector: \"gpt-4o\",\n\t\t\t\t},\n\t\t\t},\n\t\t\twantErr: true,\n\t\t},\n\t\t{\n\t\t\tname: \"MissingName\",\n\t\t\tast: &assistant.Assistant{\n\t\t\t\tAssistantModel: store.AssistantModel{\n\t\t\t\t\tID: \"test-id\",\n\t\t\t\t\tConnector: \"gpt-4o\",\n\t\t\t\t},\n\t\t\t},\n\t\t\twantErr: true,\n\t\t},\n\t}\n\n\tfor _, tt := range tests {\n\t\tt.Run(tt.name, func(t *testing.T) {\n\t\t\terr := tt.ast.Validate()\n\t\t\tif tt.wantErr {\n\t\t\t\tassert.Error(t, err)\n\t\t\t} else {\n\t\t\t\tassert.NoError(t, err)\n\t\t\t}\n\t\t})\n\t}\n}\n" }, { "path": "agent/assistant/mcp.go", "content": "package assistant\n\nimport (\n\t\"context\"\n\t\"fmt\"\n\t\"strings\"\n\n\tjsoniter \"github.com/json-iterator/go\"\n\tgouJson \"github.com/yaoapp/gou/json\"\n\t\"github.com/yaoapp/gou/mcp\"\n\tmcpTypes \"github.com/yaoapp/gou/mcp/types\"\n\tagentContext \"github.com/yaoapp/yao/agent/context\"\n\tstoreTypes \"github.com/yaoapp/yao/agent/store/types\"\n\t\"github.com/yaoapp/yao/trace/types\"\n)\n\nconst (\n\t// MaxMCPTools maximum number of MCP tools to include (to avoid overwhelming the LLM)\n\tMaxMCPTools = 20\n)\n\n// MCPToolName formats a tool name with MCP server prefix\n// Format: server_id__tool_name (double underscore separator)\n// Dots in server_id are replaced with single underscores\n// Examples:\n// - (\"echo\", \"ping\") → \"echo__ping\"\n// - (\"github.enterprise\", \"search\") → \"github_enterprise__search\"\n//\n// Naming constraint: MCP server_id MUST NOT contain underscores (_)\n// Only dots (.), letters, numbers, and hyphens (-) are allowed in server_id\nfunc MCPToolName(serverID, toolName string) string {\n\tif serverID == \"\" || toolName == \"\" {\n\t\treturn \"\"\n\t}\n\t// Replace dots with single underscores in server_id\n\tcleanServerID := strings.ReplaceAll(serverID, \".\", \"_\")\n\t// Use double underscore as separator\n\treturn fmt.Sprintf(\"%s__%s\", cleanServerID, toolName)\n}\n\n// ParseMCPToolName parses a formatted MCP tool name into server ID and tool name\n// Splits by double underscore (__), then restores dots in server_id\n// Examples:\n// - \"echo__ping\" → (\"echo\", \"ping\")\n// - \"github_enterprise__search\" → (\"github.enterprise\", \"search\")\n//\n// Returns (serverID, toolName, true) if valid format, (\"\", \"\", false) otherwise\nfunc ParseMCPToolName(formattedName string) (string, string, bool) {\n\tif formattedName == \"\" {\n\t\treturn \"\", \"\", false\n\t}\n\n\t// Split by double underscore\n\tparts := strings.Split(formattedName, \"__\")\n\tif len(parts) != 2 {\n\t\treturn \"\", \"\", false\n\t}\n\n\tcleanServerID := parts[0]\n\ttoolName := parts[1]\n\n\t// Validate that both parts are non-empty\n\tif cleanServerID == \"\" || toolName == \"\" {\n\t\treturn \"\", \"\", false\n\t}\n\n\t// Restore dots in server_id (replace single underscores back to dots)\n\tserverID := strings.ReplaceAll(cleanServerID, \"_\", \".\")\n\n\treturn serverID, toolName, true\n}\n\n// buildMCPTools builds tool definitions and samples system prompt from MCP servers\n// Returns (tools, samplesPrompt, error)\nfunc (ast *Assistant) buildMCPTools(ctx *agentContext.Context, createResponse *agentContext.HookCreateResponse) ([]MCPTool, string, error) {\n\t// Determine which MCP servers to use: hook's or assistant's (hook takes precedence)\n\tvar servers []storeTypes.MCPServerConfig\n\n\t// If hook provides MCP servers, use those (override)\n\tif createResponse != nil && len(createResponse.MCPServers) > 0 {\n\t\tservers = make([]storeTypes.MCPServerConfig, len(createResponse.MCPServers))\n\t\tfor i, hookServer := range createResponse.MCPServers {\n\t\t\t// Convert context.MCPServerConfig to storeTypes.MCPServerConfig\n\t\t\tservers[i] = storeTypes.MCPServerConfig{\n\t\t\t\tServerID: hookServer.ServerID,\n\t\t\t\tTools: hookServer.Tools,\n\t\t\t\tResources: hookServer.Resources,\n\t\t\t}\n\t\t}\n\t} else if ast.MCP != nil && len(ast.MCP.Servers) > 0 {\n\t\t// Otherwise, use assistant's configured servers\n\t\tservers = ast.MCP.Servers\n\t} else {\n\t\t// No servers configured\n\t\treturn nil, \"\", nil\n\t}\n\n\t// Use the agent context for cancellation and timeout control\n\tmcpCtx := ctx.Context\n\tif mcpCtx == nil {\n\t\tmcpCtx = context.Background()\n\t}\n\n\tallTools := make([]MCPTool, 0)\n\tsamplesBuilder := strings.Builder{}\n\thasSamples := false\n\n\t// Process each MCP server in order\n\tfor _, serverConfig := range servers {\n\t\tif len(allTools) >= MaxMCPTools {\n\t\t\tctx.Logger.Warn(\"Reached maximum tool limit (%d), skipping remaining servers\", MaxMCPTools)\n\t\t\tbreak\n\t\t}\n\n\t\t// Get MCP client\n\t\tclient, err := mcp.Select(serverConfig.ServerID)\n\t\tif err != nil {\n\t\t\tctx.Logger.Warn(\"Failed to select MCP client '%s': %v\", serverConfig.ServerID, err)\n\t\t\tcontinue\n\t\t}\n\n\t\t// Get tools list (filter by serverConfig.Tools if specified)\n\t\ttoolsResponse, err := client.ListTools(mcpCtx, \"\")\n\t\tif err != nil {\n\t\t\tctx.Logger.Warn(\"Failed to list tools for '%s': %v\", serverConfig.ServerID, err)\n\t\t\tcontinue\n\t\t}\n\n\t\t// Build tool filter map if specified\n\t\ttoolFilter := make(map[string]bool)\n\t\tif len(serverConfig.Tools) > 0 {\n\t\t\tfor _, toolName := range serverConfig.Tools {\n\t\t\t\ttoolFilter[toolName] = true\n\t\t\t}\n\t\t}\n\n\t\t// Process each tool\n\t\tfor _, tool := range toolsResponse.Tools {\n\t\t\t// Check tool limit\n\t\t\tif len(allTools) >= MaxMCPTools {\n\t\t\t\tbreak\n\t\t\t}\n\n\t\t\t// Apply tool filter if specified\n\t\t\tif len(toolFilter) > 0 && !toolFilter[tool.Name] {\n\t\t\t\tcontinue\n\t\t\t}\n\n\t\t\t// Format tool name with server prefix\n\t\t\tformattedName := MCPToolName(serverConfig.ServerID, tool.Name)\n\n\t\t\t// Convert MCP tool to MCPTool format\n\t\t\tmcpTool := MCPTool{\n\t\t\t\tName: formattedName,\n\t\t\t\tDescription: tool.Description,\n\t\t\t\tParameters: tool.InputSchema,\n\t\t\t}\n\n\t\t\tallTools = append(allTools, mcpTool)\n\n\t\t\t// Try to get samples for this tool\n\t\t\tsamples, err := client.ListSamples(mcpCtx, mcpTypes.SampleTool, tool.Name)\n\t\t\tif err == nil && len(samples.Samples) > 0 {\n\t\t\t\tif !hasSamples {\n\t\t\t\t\tsamplesBuilder.WriteString(\"\\n\\n## MCP Tool Usage Examples\\n\\n\")\n\t\t\t\t\tsamplesBuilder.WriteString(\"The following examples demonstrate how to use MCP tools correctly:\\n\\n\")\n\t\t\t\t\thasSamples = true\n\t\t\t\t}\n\n\t\t\t\tsamplesBuilder.WriteString(fmt.Sprintf(\"### %s\\n\\n\", formattedName))\n\t\t\t\tif tool.Description != \"\" {\n\t\t\t\t\tsamplesBuilder.WriteString(fmt.Sprintf(\"**Description**: %s\\n\\n\", tool.Description))\n\t\t\t\t}\n\n\t\t\t\tfor i, sample := range samples.Samples {\n\t\t\t\t\tif i >= 3 { // Limit to 3 examples per tool\n\t\t\t\t\t\tbreak\n\t\t\t\t\t}\n\n\t\t\t\t\tsamplesBuilder.WriteString(fmt.Sprintf(\"**Example %d\", i+1))\n\t\t\t\t\tif sample.Name != \"\" {\n\t\t\t\t\t\tsamplesBuilder.WriteString(fmt.Sprintf(\" - %s\", sample.Name))\n\t\t\t\t\t}\n\t\t\t\t\tsamplesBuilder.WriteString(\"**:\\n\")\n\n\t\t\t\t\t// Check metadata for description\n\t\t\t\t\tif sample.Metadata != nil {\n\t\t\t\t\t\tif desc, ok := sample.Metadata[\"description\"].(string); ok && desc != \"\" {\n\t\t\t\t\t\t\tsamplesBuilder.WriteString(fmt.Sprintf(\"- Description: %s\\n\", desc))\n\t\t\t\t\t\t}\n\t\t\t\t\t}\n\n\t\t\t\t\tif sample.Input != nil {\n\t\t\t\t\t\tsamplesBuilder.WriteString(fmt.Sprintf(\"- Input: `%v`\\n\", sample.Input))\n\t\t\t\t\t}\n\n\t\t\t\t\tif sample.Output != nil {\n\t\t\t\t\t\tsamplesBuilder.WriteString(fmt.Sprintf(\"- Output: `%v`\\n\", sample.Output))\n\t\t\t\t\t}\n\n\t\t\t\t\tsamplesBuilder.WriteString(\"\\n\")\n\t\t\t\t}\n\t\t\t}\n\t\t}\n\n\t\tctx.Logger.Debug(\"Loaded %d tools from server '%s'\", len(toolsResponse.Tools), serverConfig.ServerID)\n\t}\n\n\tsamplesPrompt := \"\"\n\tif hasSamples {\n\t\tsamplesPrompt = samplesBuilder.String()\n\t}\n\n\tctx.Logger.Debug(\"Total MCP tools loaded: %d\", len(allTools))\n\treturn allTools, samplesPrompt, nil\n}\n\n// ToolCallResult represents the result of a tool call execution\n// executeToolCalls executes tool calls with intelligent strategy and trace logging:\n// - Single tool: use CallTool, single trace node\n// - Multiple tools: use CallToolsParallel with parallel trace nodes, fallback to sequential on certain errors\n// Returns (results, hasErrors)\nfunc (ast *Assistant) executeToolCalls(ctx *agentContext.Context, toolCalls []agentContext.ToolCall, attempt int) ([]ToolCallResult, bool) {\n\tif len(toolCalls) == 0 {\n\t\treturn nil, false\n\t}\n\n\tctx.Logger.Debug(\"Executing %d tool calls (attempt %d)\", len(toolCalls), attempt)\n\n\t// Single tool call\n\tif len(toolCalls) == 1 {\n\t\treturn ast.executeSingleToolCall(ctx, toolCalls[0])\n\t}\n\n\t// Multiple tool calls - try parallel first\n\treturn ast.executeMultipleToolCallsParallel(ctx, toolCalls)\n}\n\n// executeSingleToolCall executes a single tool call with trace logging\nfunc (ast *Assistant) executeSingleToolCall(ctx *agentContext.Context, toolCall agentContext.ToolCall) ([]ToolCallResult, bool) {\n\tctx.Logger.ToolStart(toolCall.Function.Name)\n\n\ttrace, _ := ctx.Trace()\n\n\t// Use the agent context for cancellation and timeout control\n\tmcpCtx := ctx.Context\n\tif mcpCtx == nil {\n\t\tmcpCtx = context.Background()\n\t}\n\n\tresult := ToolCallResult{\n\t\tToolCallID: toolCall.ID,\n\t\tName: toolCall.Function.Name,\n\t}\n\n\t// Parse tool name\n\tserverID, toolName, ok := ParseMCPToolName(toolCall.Function.Name)\n\tif !ok {\n\t\tresult.Error = fmt.Errorf(\"invalid MCP tool name format: %s\", toolCall.Function.Name)\n\t\tresult.Content = result.Error.Error()\n\t\tctx.Logger.Error(\"Invalid MCP tool name format: %s\", toolCall.Function.Name)\n\t\tctx.Logger.ToolComplete(toolCall.Function.Name, false)\n\t\treturn []ToolCallResult{result}, true\n\t}\n\n\t// Get MCP client\n\tclient, err := mcp.Select(serverID)\n\tif err != nil {\n\t\tresult.Error = fmt.Errorf(\"failed to select MCP client '%s': %w\", serverID, err)\n\t\tresult.Content = result.Error.Error()\n\t\tresult.IsRetryableError = false // MCP client selection error is not retryable\n\t\tctx.Logger.Error(\"Failed to select MCP client '%s': %v\", serverID, err)\n\t\tctx.Logger.ToolComplete(toolCall.Function.Name, false)\n\t\treturn []ToolCallResult{result}, true\n\t}\n\n\t// Get tool info for description and schema\n\ttoolsResponse, err := client.ListTools(mcpCtx, \"\")\n\tvar toolDescription string\n\tvar toolSchema interface{}\n\tif err == nil {\n\t\tfor _, t := range toolsResponse.Tools {\n\t\t\tif t.Name == toolName {\n\t\t\t\ttoolDescription = t.Description\n\t\t\t\ttoolSchema = t.InputSchema\n\t\t\t\tbreak\n\t\t\t}\n\t\t}\n\t}\n\tif toolDescription == \"\" {\n\t\ttoolDescription = fmt.Sprintf(\"MCP tool '%s'\", toolName)\n\t}\n\n\t// Add trace node for this tool call\n\tvar toolNode types.Node\n\tif trace != nil {\n\t\ttoolNode, _ = trace.Add(\n\t\t\tmap[string]any{\n\t\t\t\t\"tool_call_id\": toolCall.ID,\n\t\t\t\t\"server\": serverID,\n\t\t\t\t\"tool\": toolName,\n\t\t\t\t\"arguments\": toolCall.Function.Arguments,\n\t\t\t},\n\t\t\ttypes.TraceNodeOption{\n\t\t\t\tLabel: toolDescription,\n\t\t\t\tType: \"mcp_tool\",\n\t\t\t\tIcon: \"build\",\n\t\t\t\tDescription: fmt.Sprintf(\"Calling '%s' on server '%s'\", toolName, serverID),\n\t\t\t},\n\t\t)\n\t}\n\n\t// Parse arguments with repair support for better tolerance\n\tvar args map[string]interface{}\n\tif toolCall.Function.Arguments != \"\" {\n\t\tparsed, err := gouJson.Parse(toolCall.Function.Arguments)\n\t\tif err != nil {\n\t\t\tresult.Error = fmt.Errorf(\"failed to parse arguments: %w\", err)\n\t\t\tresult.Content = result.Error.Error()\n\t\t\tresult.IsRetryableError = true // Argument parsing error is retryable by LLM\n\t\t\tctx.Logger.Error(\"Failed to parse arguments: %v\", err)\n\t\t\tctx.Logger.ToolComplete(toolCall.Function.Name, false)\n\t\t\tif toolNode != nil {\n\t\t\t\ttoolNode.Fail(result.Error)\n\t\t\t}\n\t\t\treturn []ToolCallResult{result}, true\n\t\t}\n\n\t\t// Convert to map\n\t\tif argsMap, ok := parsed.(map[string]interface{}); ok {\n\t\t\targs = argsMap\n\t\t} else {\n\t\t\tresult.Error = fmt.Errorf(\"arguments must be an object, got %T\", parsed)\n\t\t\tresult.Content = result.Error.Error()\n\t\t\tresult.IsRetryableError = true // Type error is retryable by LLM\n\t\t\tctx.Logger.Error(\"Arguments must be an object, got %T\", parsed)\n\t\t\tctx.Logger.ToolComplete(toolCall.Function.Name, false)\n\t\t\tif toolNode != nil {\n\t\t\t\ttoolNode.Fail(result.Error)\n\t\t\t}\n\t\t\treturn []ToolCallResult{result}, true\n\t\t}\n\n\t\t// Validate arguments against tool schema if available\n\t\tif toolSchema != nil {\n\t\t\tif err := gouJson.Validate(args, toolSchema); err != nil {\n\t\t\t\tresult.Error = fmt.Errorf(\"argument validation failed: %w\", err)\n\t\t\t\tresult.Content = result.Error.Error()\n\t\t\t\tresult.IsRetryableError = true // Validation error is retryable by LLM\n\t\t\t\tctx.Logger.Error(\"Argument validation failed: %v\", err)\n\t\t\t\tctx.Logger.ToolComplete(toolCall.Function.Name, false)\n\t\t\t\tif toolNode != nil {\n\t\t\t\t\ttoolNode.Fail(result.Error)\n\t\t\t\t}\n\t\t\t\treturn []ToolCallResult{result}, true\n\t\t\t}\n\t\t}\n\t}\n\n\t// Call the tool with agent context as extra argument\n\tctx.Logger.Debug(\"Calling tool: %s (server: %s)\", toolName, serverID)\n\n\t// Pass agent context as extra argument (only used for Process transport)\n\tcallResult, err := client.CallTool(mcpCtx, toolName, args, ctx)\n\tif err != nil {\n\t\tresult.Error = fmt.Errorf(\"tool call failed: %w\", err)\n\t\tresult.Content = result.Error.Error()\n\t\t// Check if error is retryable (parameter/validation errors)\n\t\tresult.IsRetryableError = isRetryableToolError(err)\n\t\tctx.Logger.Error(\"Tool call failed: %v (retryable: %v)\", err, result.IsRetryableError)\n\t\tctx.Logger.ToolComplete(toolCall.Function.Name, false)\n\t\tif toolNode != nil {\n\t\t\ttoolNode.Fail(result.Error)\n\t\t}\n\t\treturn []ToolCallResult{result}, true\n\t}\n\n\t// Check if result is an error\n\tif callResult.IsError {\n\t\tresult.Error = fmt.Errorf(\"MCP tool error\")\n\t\tresult.IsRetryableError = false // MCP internal error is not retryable\n\t}\n\n\t// Serialize the Content field only ([]ToolContent)\n\tcontentBytes, err := jsoniter.Marshal(callResult.Content)\n\tif err != nil {\n\t\tresult.Error = fmt.Errorf(\"failed to serialize result: %w\", err)\n\t\tresult.Content = result.Error.Error()\n\t\tresult.IsRetryableError = false\n\t\tctx.Logger.Error(\"Failed to serialize result: %v\", err)\n\t\tctx.Logger.ToolComplete(toolCall.Function.Name, false)\n\t\tif toolNode != nil {\n\t\t\ttoolNode.Fail(result.Error)\n\t\t}\n\t\treturn []ToolCallResult{result}, true\n\t}\n\n\tresult.Content = string(contentBytes)\n\tctx.Logger.ToolComplete(toolCall.Function.Name, true)\n\n\tif toolNode != nil {\n\t\ttoolNode.Complete(map[string]any{\n\t\t\t\"result\": callResult,\n\t\t})\n\t}\n\n\treturn []ToolCallResult{result}, false\n}\n\n// executeMultipleToolCallsParallel executes multiple tool calls in parallel with trace logging\nfunc (ast *Assistant) executeMultipleToolCallsParallel(ctx *agentContext.Context, toolCalls []agentContext.ToolCall) ([]ToolCallResult, bool) {\n\ttrace, _ := ctx.Trace()\n\n\t// Use the agent context for cancellation and timeout control\n\tmcpCtx := ctx.Context\n\tif mcpCtx == nil {\n\t\tmcpCtx = context.Background()\n\t}\n\n\t// Group tool calls by server\n\tserverGroups := make(map[string][]agentContext.ToolCall)\n\tfor _, tc := range toolCalls {\n\t\tserverID, _, ok := ParseMCPToolName(tc.Function.Name)\n\t\tif !ok {\n\t\t\tctx.Logger.Warn(\"Invalid tool name format: %s\", tc.Function.Name)\n\t\t\tcontinue\n\t\t}\n\t\tserverGroups[serverID] = append(serverGroups[serverID], tc)\n\t}\n\n\tresults := make([]ToolCallResult, 0, len(toolCalls))\n\thasErrors := false\n\n\t// Process each server's tools\n\tfor serverID, calls := range serverGroups {\n\t\tclient, err := mcp.Select(serverID)\n\t\tif err != nil {\n\t\t\tctx.Logger.Error(\"Failed to select MCP client '%s': %v\", serverID, err)\n\t\t\t// Add error results for all calls to this server\n\t\t\tfor _, tc := range calls {\n\t\t\t\tresults = append(results, ToolCallResult{\n\t\t\t\t\tToolCallID: tc.ID,\n\t\t\t\t\tName: tc.Function.Name,\n\t\t\t\t\tContent: fmt.Sprintf(\"Failed to select MCP client: %v\", err),\n\t\t\t\t\tError: err,\n\t\t\t\t})\n\t\t\t}\n\t\t\thasErrors = true\n\t\t\tcontinue\n\t\t}\n\n\t\t// Try parallel execution\n\t\tserverResults, serverHasErrors := ast.executeServerToolsParallelWithTrace(\n\t\t\tmcpCtx, ctx, trace, client, serverID, calls,\n\t\t)\n\n\t\t// If parallel execution failed with retryable error, try sequential\n\t\tif serverHasErrors && ast.shouldRetrySequential(serverResults) {\n\t\t\tctx.Logger.Warn(\"Parallel execution had parameter errors for server '%s', retrying sequentially\", serverID)\n\t\t\tserverResults, serverHasErrors = ast.executeServerToolsSequentialWithTrace(\n\t\t\t\tmcpCtx, ctx, trace, client, serverID, calls,\n\t\t\t)\n\t\t}\n\n\t\tresults = append(results, serverResults...)\n\t\tif serverHasErrors {\n\t\t\thasErrors = true\n\t\t}\n\t}\n\n\treturn results, hasErrors\n}\n\n// isRetryableToolError checks if an error is retryable by LLM (parameter/validation errors)\n// Returns true for errors that LLM can potentially fix by adjusting parameters\n// Returns false for MCP internal errors (network, auth, service unavailable, etc.)\nfunc isRetryableToolError(err error) bool {\n\tif err == nil {\n\t\treturn false\n\t}\n\n\terrMsg := strings.ToLower(err.Error())\n\n\t// These are NOT retryable (MCP internal issues)\n\tnonRetryablePatterns := []string{\n\t\t\"network\",\n\t\t\"timeout\",\n\t\t\"connection\",\n\t\t\"unauthorized\",\n\t\t\"forbidden\",\n\t\t\"unavailable\",\n\t\t\"failed to select\",\n\t\t\"context canceled\",\n\t\t\"context deadline\",\n\t\t\"server error\",\n\t\t\"internal error\",\n\t}\n\n\tfor _, pattern := range nonRetryablePatterns {\n\t\tif strings.Contains(errMsg, pattern) {\n\t\t\treturn false\n\t\t}\n\t}\n\n\t// These ARE retryable (parameter/validation issues LLM can fix)\n\tretryablePatterns := []string{\n\t\t\"invalid\",\n\t\t\"required\",\n\t\t\"missing\",\n\t\t\"validation\",\n\t\t\"schema\",\n\t\t\"type\",\n\t\t\"format\",\n\t\t\"parse\",\n\t\t\"argument\",\n\t\t\"parameter\",\n\t}\n\n\tfor _, pattern := range retryablePatterns {\n\t\tif strings.Contains(errMsg, pattern) {\n\t\t\treturn true\n\t\t}\n\t}\n\n\t// Default: assume it's retryable unless proven otherwise\n\t// This allows LLM to attempt fixes for unknown error types\n\treturn true\n}\n\n// shouldRetrySequential checks if errors are retryable (parameter issues, not network/service issues)\nfunc (ast *Assistant) shouldRetrySequential(results []ToolCallResult) bool {\n\t// Check if any result has a retryable error\n\thasRetryable := false\n\tfor _, result := range results {\n\t\tif result.Error != nil && result.IsRetryableError {\n\t\t\thasRetryable = true\n\t\t\tbreak\n\t\t}\n\t}\n\treturn hasRetryable\n}\n\n// executeServerToolsParallelWithTrace executes tools for a single server in parallel with trace\nfunc (ast *Assistant) executeServerToolsParallelWithTrace(mcpCtx context.Context, ctx *agentContext.Context, trace types.Manager, client mcp.Client, serverID string, toolCalls []agentContext.ToolCall) ([]ToolCallResult, bool) {\n\t// Prepare parallel trace inputs\n\tvar parallelInputs []types.TraceParallelInput\n\tmcpCalls := make([]mcpTypes.ToolCall, 0, len(toolCalls))\n\tcallMap := make(map[string]agentContext.ToolCall)\n\n\tfor _, tc := range toolCalls {\n\t\t_, toolName, ok := ParseMCPToolName(tc.Function.Name)\n\t\tif !ok {\n\t\t\tcontinue\n\t\t}\n\n\t\tvar args map[string]interface{}\n\t\tif tc.Function.Arguments != \"\" {\n\t\t\tif err := jsoniter.UnmarshalFromString(tc.Function.Arguments, &args); err != nil {\n\t\t\t\tctx.Logger.Error(\"Failed to parse arguments for %s: %v\", toolName, err)\n\t\t\t\tcontinue\n\t\t\t}\n\t\t}\n\n\t\tmcpCalls = append(mcpCalls, mcpTypes.ToolCall{\n\t\t\tName: toolName,\n\t\t\tArguments: args,\n\t\t})\n\t\tcallMap[toolName] = tc\n\t\tctx.Logger.ToolStart(tc.Function.Name)\n\n\t\t// Add trace input for this tool\n\t\tparallelInputs = append(parallelInputs, types.TraceParallelInput{\n\t\t\tInput: map[string]any{\n\t\t\t\t\"tool_call_id\": tc.ID,\n\t\t\t\t\"server\": serverID,\n\t\t\t\t\"tool\": toolName,\n\t\t\t\t\"arguments\": tc.Function.Arguments,\n\t\t\t},\n\t\t\tOption: types.TraceNodeOption{\n\t\t\t\tLabel: fmt.Sprintf(\"Tool: %s\", toolName),\n\t\t\t\tType: \"mcp_tool\",\n\t\t\t\tIcon: \"build\",\n\t\t\t\tDescription: fmt.Sprintf(\"Calling MCP tool '%s' on server '%s'\", toolName, serverID),\n\t\t\t},\n\t\t})\n\t}\n\n\t// Create parallel trace nodes\n\tvar toolNodes []types.Node\n\tif trace != nil && len(parallelInputs) > 0 {\n\t\tvar err error\n\t\ttoolNodes, err = trace.Parallel(parallelInputs)\n\t\tif err != nil {\n\t\t\tctx.Logger.Debug(\"trace.Parallel() failed: %v\", err)\n\t\t}\n\t}\n\n\t// Call tools in parallel with agent context as extra argument\n\tctx.Logger.Debug(\"Calling %d tools in parallel on server '%s'\", len(mcpCalls), serverID)\n\n\t// Pass agent context as extra argument (only used for Process transport)\n\tmcpResponse, err := client.CallToolsParallel(mcpCtx, mcpCalls, ctx)\n\tif err != nil {\n\t\tctx.Logger.Error(\"Parallel call failed: %v\", err)\n\t\tfor i, node := range toolNodes {\n\t\t\tif node != nil {\n\t\t\t\tnode.Fail(err)\n\t\t\t}\n\t\t\tif i < len(mcpCalls) {\n\t\t\t\tif tc, ok := callMap[mcpCalls[i].Name]; ok {\n\t\t\t\t\tctx.Logger.ToolComplete(tc.Function.Name, false)\n\t\t\t\t}\n\t\t\t}\n\t\t}\n\t\treturn nil, true\n\t}\n\n\t// Process results\n\tresults := make([]ToolCallResult, 0, len(mcpResponse.Results))\n\thasErrors := false\n\n\tfor i, mcpResult := range mcpResponse.Results {\n\t\ttoolName := mcpCalls[i].Name\n\t\toriginalCall := callMap[toolName]\n\t\tvar toolNode types.Node\n\t\tif i < len(toolNodes) {\n\t\t\ttoolNode = toolNodes[i]\n\t\t}\n\n\t\tresult := ToolCallResult{\n\t\t\tToolCallID: originalCall.ID,\n\t\t\tName: originalCall.Function.Name,\n\t\t}\n\n\t\t// Serialize content\n\t\tcontentBytes, err := jsoniter.Marshal(mcpResult.Content)\n\t\tif err != nil {\n\t\t\tresult.Error = fmt.Errorf(\"failed to serialize result: %w\", err)\n\t\t\tresult.Content = result.Error.Error()\n\t\t\tresult.IsRetryableError = false // Serialization error is not retryable\n\t\t\thasErrors = true\n\t\t\tctx.Logger.ToolComplete(originalCall.Function.Name, false)\n\t\t\tif toolNode != nil {\n\t\t\t\ttoolNode.Fail(result.Error)\n\t\t\t}\n\t\t} else {\n\t\t\tresult.Content = string(contentBytes)\n\n\t\t\t// Check if it's an error result\n\t\t\tif mcpResult.IsError {\n\t\t\t\tresult.Error = fmt.Errorf(\"tool call error: %s\", result.Content)\n\t\t\t\tresult.IsRetryableError = isRetryableToolError(result.Error)\n\t\t\t\thasErrors = true\n\t\t\t\tctx.Logger.Error(\"Tool call failed: %s - %s (retryable: %v)\", toolName, result.Content, result.IsRetryableError)\n\t\t\t\tctx.Logger.ToolComplete(originalCall.Function.Name, false)\n\t\t\t\tif toolNode != nil {\n\t\t\t\t\ttoolNode.Fail(result.Error)\n\t\t\t\t}\n\t\t\t} else {\n\t\t\t\tctx.Logger.ToolComplete(originalCall.Function.Name, true)\n\t\t\t\tif toolNode != nil {\n\t\t\t\t\ttoolNode.Complete(map[string]any{\n\t\t\t\t\t\t\"result\": mcpResult.Content,\n\t\t\t\t\t})\n\t\t\t\t}\n\t\t\t}\n\t\t}\n\n\t\tresults = append(results, result)\n\t}\n\n\treturn results, hasErrors\n}\n\n// executeServerToolsSequentialWithTrace executes tools for a single server sequentially with trace\nfunc (ast *Assistant) executeServerToolsSequentialWithTrace(mcpCtx context.Context, ctx *agentContext.Context, trace types.Manager, client mcp.Client, serverID string, toolCalls []agentContext.ToolCall) ([]ToolCallResult, bool) {\n\tresults := make([]ToolCallResult, 0, len(toolCalls))\n\thasErrors := false\n\n\tctx.Logger.Debug(\"Calling %d tools sequentially on server '%s'\", len(toolCalls), serverID)\n\n\tfor _, tc := range toolCalls {\n\t\tctx.Logger.ToolStart(tc.Function.Name)\n\n\t\t_, toolName, ok := ParseMCPToolName(tc.Function.Name)\n\t\tif !ok {\n\t\t\tresults = append(results, ToolCallResult{\n\t\t\t\tToolCallID: tc.ID,\n\t\t\t\tName: tc.Function.Name,\n\t\t\t\tContent: fmt.Sprintf(\"Invalid tool name format: %s\", tc.Function.Name),\n\t\t\t\tError: fmt.Errorf(\"invalid tool name format\"),\n\t\t\t})\n\t\t\tctx.Logger.ToolComplete(tc.Function.Name, false)\n\t\t\thasErrors = true\n\t\t\tcontinue\n\t\t}\n\n\t\t// Get tool schema for validation\n\t\ttoolsResponse, err := client.ListTools(mcpCtx, \"\")\n\t\tvar toolSchema interface{}\n\t\tif err == nil {\n\t\t\tfor _, t := range toolsResponse.Tools {\n\t\t\t\tif t.Name == toolName {\n\t\t\t\t\ttoolSchema = t.InputSchema\n\t\t\t\t\tbreak\n\t\t\t\t}\n\t\t\t}\n\t\t}\n\n\t\t// Add trace node for this tool call\n\t\tvar toolNode types.Node\n\t\tif trace != nil {\n\t\t\ttoolNode, _ = trace.Add(\n\t\t\t\tmap[string]any{\n\t\t\t\t\t\"tool_call_id\": tc.ID,\n\t\t\t\t\t\"server\": serverID,\n\t\t\t\t\t\"tool\": toolName,\n\t\t\t\t\t\"arguments\": tc.Function.Arguments,\n\t\t\t\t},\n\t\t\t\ttypes.TraceNodeOption{\n\t\t\t\t\tLabel: fmt.Sprintf(\"Tool: %s (sequential retry)\", toolName),\n\t\t\t\t\tType: \"mcp_tool\",\n\t\t\t\t\tIcon: \"build\",\n\t\t\t\t\tDescription: fmt.Sprintf(\"Retrying MCP tool '%s' on server '%s' sequentially\", toolName, serverID),\n\t\t\t\t},\n\t\t\t)\n\t\t}\n\n\t\t// Parse arguments with repair support\n\t\tvar args map[string]interface{}\n\t\tif tc.Function.Arguments != \"\" {\n\t\t\tparsed, err := gouJson.Parse(tc.Function.Arguments)\n\t\t\tif err != nil {\n\t\t\t\tresult := ToolCallResult{\n\t\t\t\t\tToolCallID: tc.ID,\n\t\t\t\t\tName: tc.Function.Name,\n\t\t\t\t\tContent: fmt.Sprintf(\"Failed to parse arguments: %v\", err),\n\t\t\t\t\tError: err,\n\t\t\t\t\tIsRetryableError: true, // Parsing error is retryable\n\t\t\t\t}\n\t\t\t\tresults = append(results, result)\n\t\t\t\thasErrors = true\n\t\t\t\tctx.Logger.ToolComplete(tc.Function.Name, false)\n\t\t\t\tif toolNode != nil {\n\t\t\t\t\ttoolNode.Fail(err)\n\t\t\t\t}\n\t\t\t\tcontinue\n\t\t\t}\n\n\t\t\t// Convert to map\n\t\t\tif argsMap, ok := parsed.(map[string]interface{}); ok {\n\t\t\t\targs = argsMap\n\t\t\t} else {\n\t\t\t\terr := fmt.Errorf(\"arguments must be an object, got %T\", parsed)\n\t\t\t\tresult := ToolCallResult{\n\t\t\t\t\tToolCallID: tc.ID,\n\t\t\t\t\tName: tc.Function.Name,\n\t\t\t\t\tContent: err.Error(),\n\t\t\t\t\tError: err,\n\t\t\t\t\tIsRetryableError: true, // Type error is retryable\n\t\t\t\t}\n\t\t\t\tresults = append(results, result)\n\t\t\t\thasErrors = true\n\t\t\t\tctx.Logger.ToolComplete(tc.Function.Name, false)\n\t\t\t\tif toolNode != nil {\n\t\t\t\t\ttoolNode.Fail(err)\n\t\t\t\t}\n\t\t\t\tcontinue\n\t\t\t}\n\n\t\t\t// Validate arguments against tool schema if available\n\t\t\tif toolSchema != nil {\n\t\t\t\tif err := gouJson.Validate(args, toolSchema); err != nil {\n\t\t\t\t\tresult := ToolCallResult{\n\t\t\t\t\t\tToolCallID: tc.ID,\n\t\t\t\t\t\tName: tc.Function.Name,\n\t\t\t\t\t\tContent: fmt.Sprintf(\"Argument validation failed: %v\", err),\n\t\t\t\t\t\tError: err,\n\t\t\t\t\t\tIsRetryableError: true, // Validation error is retryable\n\t\t\t\t\t}\n\t\t\t\t\tresults = append(results, result)\n\t\t\t\t\thasErrors = true\n\t\t\t\t\tctx.Logger.ToolComplete(tc.Function.Name, false)\n\t\t\t\t\tif toolNode != nil {\n\t\t\t\t\t\ttoolNode.Fail(err)\n\t\t\t\t\t}\n\t\t\t\t\tcontinue\n\t\t\t\t}\n\t\t\t}\n\t\t}\n\n\t\t// Call single tool with agent context as extra argument\n\t\tctx.Logger.Debug(\"Calling tool: %s\", toolName)\n\t\tmcpResult, err := client.CallTool(mcpCtx, toolName, args, ctx)\n\n\t\tresult := ToolCallResult{\n\t\t\tToolCallID: tc.ID,\n\t\t\tName: tc.Function.Name,\n\t\t}\n\n\t\tif err != nil {\n\t\t\tresult.Error = err\n\t\t\tresult.Content = fmt.Sprintf(\"Tool call failed: %v\", err)\n\t\t\tresult.IsRetryableError = isRetryableToolError(err)\n\t\t\thasErrors = true\n\t\t\tctx.Logger.Error(\"Tool call failed: %s - %v (retryable: %v)\", toolName, err, result.IsRetryableError)\n\t\t\tctx.Logger.ToolComplete(tc.Function.Name, false)\n\t\t\tif toolNode != nil {\n\t\t\t\ttoolNode.Fail(err)\n\t\t\t}\n\t\t} else {\n\t\t\t// Check if result is an error\n\t\t\tif mcpResult.IsError {\n\t\t\t\tresult.Error = fmt.Errorf(\"MCP tool error\")\n\t\t\t\tresult.IsRetryableError = false // MCP internal error is not retryable\n\t\t\t\thasErrors = true\n\t\t\t}\n\n\t\t\t// Serialize the Content field only ([]ToolContent)\n\t\t\tcontentBytes, err := jsoniter.Marshal(mcpResult.Content)\n\t\t\tif err != nil {\n\t\t\t\tresult.Error = err\n\t\t\t\tresult.Content = fmt.Sprintf(\"Failed to serialize result: %v\", err)\n\t\t\t\tresult.IsRetryableError = false // Serialization error is not retryable\n\t\t\t\thasErrors = true\n\t\t\t\tctx.Logger.ToolComplete(tc.Function.Name, false)\n\t\t\t\tif toolNode != nil {\n\t\t\t\t\ttoolNode.Fail(err)\n\t\t\t\t}\n\t\t\t} else {\n\t\t\t\tresult.Content = string(contentBytes)\n\t\t\t\tctx.Logger.ToolComplete(tc.Function.Name, !mcpResult.IsError)\n\t\t\t\tif toolNode != nil {\n\t\t\t\t\ttoolNode.Complete(map[string]any{\n\t\t\t\t\t\t\"result\": mcpResult.Content,\n\t\t\t\t\t})\n\t\t\t\t}\n\t\t\t}\n\t\t}\n\n\t\tresults = append(results, result)\n\t}\n\n\treturn results, hasErrors\n}\n" }, { "path": "agent/assistant/mcp_test.go", "content": "package assistant_test\n\nimport (\n\t\"context\"\n\t\"testing\"\n\n\tjsoniter \"github.com/json-iterator/go\"\n\t\"github.com/stretchr/testify/assert\"\n\t\"github.com/yaoapp/gou/mcp\"\n\tmcpTypes \"github.com/yaoapp/gou/mcp/types\"\n\t\"github.com/yaoapp/yao/agent/assistant\"\n\tagentContext \"github.com/yaoapp/yao/agent/context\"\n\t\"github.com/yaoapp/yao/agent/testutils\"\n\t\"github.com/yaoapp/yao/openapi/oauth/types\"\n)\n\nfunc TestMCPToolName(t *testing.T) {\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\ttests := []struct {\n\t\tname string\n\t\tserverID string\n\t\ttoolName string\n\t\twantResult string\n\t}{\n\t\t{\n\t\t\tname: \"Simple tool name\",\n\t\t\tserverID: \"github\",\n\t\t\ttoolName: \"search_repos\",\n\t\t\twantResult: \"github__search_repos\",\n\t\t},\n\t\t{\n\t\t\tname: \"Server with dots\",\n\t\t\tserverID: \"github.enterprise\",\n\t\t\ttoolName: \"search_repos\",\n\t\t\twantResult: \"github_enterprise__search_repos\",\n\t\t},\n\t\t{\n\t\t\tname: \"Tool with underscores\",\n\t\t\tserverID: \"customer-db\",\n\t\t\ttoolName: \"create_customer\",\n\t\t\twantResult: \"customer-db__create_customer\",\n\t\t},\n\t\t{\n\t\t\tname: \"Complex server with multiple dots\",\n\t\t\tserverID: \"com.example.mcp\",\n\t\t\ttoolName: \"tool_name\",\n\t\t\twantResult: \"com_example_mcp__tool_name\",\n\t\t},\n\t\t{\n\t\t\tname: \"Empty server ID\",\n\t\t\tserverID: \"\",\n\t\t\ttoolName: \"tool\",\n\t\t\twantResult: \"\",\n\t\t},\n\t\t{\n\t\t\tname: \"Empty tool name\",\n\t\t\tserverID: \"server\",\n\t\t\ttoolName: \"\",\n\t\t\twantResult: \"\",\n\t\t},\n\t}\n\n\tfor _, tt := range tests {\n\t\tt.Run(tt.name, func(t *testing.T) {\n\t\t\tresult := assistant.MCPToolName(tt.serverID, tt.toolName)\n\t\t\tif result != tt.wantResult {\n\t\t\t\tt.Errorf(\"MCPToolName() = %v, want %v\", result, tt.wantResult)\n\t\t\t}\n\t\t})\n\t}\n}\n\nfunc TestParseMCPToolName(t *testing.T) {\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\ttests := []struct {\n\t\tname string\n\t\tformattedName string\n\t\twantServerID string\n\t\twantToolName string\n\t\twantOK bool\n\t}{\n\t\t{\n\t\t\tname: \"Valid simple format\",\n\t\t\tformattedName: \"github__search_repos\",\n\t\t\twantServerID: \"github\",\n\t\t\twantToolName: \"search_repos\",\n\t\t\twantOK: true,\n\t\t},\n\t\t{\n\t\t\tname: \"Server with dots restored\",\n\t\t\tformattedName: \"github_enterprise__search_repos\",\n\t\t\twantServerID: \"github.enterprise\",\n\t\t\twantToolName: \"search_repos\",\n\t\t\twantOK: true,\n\t\t},\n\t\t{\n\t\t\tname: \"Complex server ID with multiple dots\",\n\t\t\tformattedName: \"com_example_mcp_server__tool_name\",\n\t\t\twantServerID: \"com.example.mcp.server\",\n\t\t\twantToolName: \"tool_name\",\n\t\t\twantOK: true,\n\t\t},\n\t\t{\n\t\t\tname: \"Tool name with underscores\",\n\t\t\tformattedName: \"server__create_new_user\",\n\t\t\twantServerID: \"server\",\n\t\t\twantToolName: \"create_new_user\",\n\t\t\twantOK: true,\n\t\t},\n\t\t{\n\t\t\tname: \"Server with hyphens\",\n\t\t\tformattedName: \"mcp-server__tool\",\n\t\t\twantServerID: \"mcp-server\",\n\t\t\twantToolName: \"tool\",\n\t\t\twantOK: true,\n\t\t},\n\t\t{\n\t\t\tname: \"Invalid format - no double underscore\",\n\t\t\tformattedName: \"invalid\",\n\t\t\twantServerID: \"\",\n\t\t\twantToolName: \"\",\n\t\t\twantOK: false,\n\t\t},\n\t\t{\n\t\t\tname: \"Invalid format - empty string\",\n\t\t\tformattedName: \"\",\n\t\t\twantServerID: \"\",\n\t\t\twantToolName: \"\",\n\t\t\twantOK: false,\n\t\t},\n\t\t{\n\t\t\tname: \"Invalid format - only double underscore\",\n\t\t\tformattedName: \"__\",\n\t\t\twantServerID: \"\",\n\t\t\twantToolName: \"\",\n\t\t\twantOK: false,\n\t\t},\n\t\t{\n\t\t\tname: \"Invalid format - ends with double underscore\",\n\t\t\tformattedName: \"server__\",\n\t\t\twantServerID: \"\",\n\t\t\twantToolName: \"\",\n\t\t\twantOK: false,\n\t\t},\n\t\t{\n\t\t\tname: \"Invalid format - starts with double underscore\",\n\t\t\tformattedName: \"__tool\",\n\t\t\twantServerID: \"\",\n\t\t\twantToolName: \"\",\n\t\t\twantOK: false,\n\t\t},\n\t\t{\n\t\t\tname: \"Invalid format - multiple double underscores\",\n\t\t\tformattedName: \"server__middle__tool\",\n\t\t\twantServerID: \"\",\n\t\t\twantToolName: \"\",\n\t\t\twantOK: false,\n\t\t},\n\t}\n\n\tfor _, tt := range tests {\n\t\tt.Run(tt.name, func(t *testing.T) {\n\t\t\tserverID, toolName, ok := assistant.ParseMCPToolName(tt.formattedName)\n\t\t\tif serverID != tt.wantServerID {\n\t\t\t\tt.Errorf(\"ParseMCPToolName() serverID = %v, want %v\", serverID, tt.wantServerID)\n\t\t\t}\n\t\t\tif toolName != tt.wantToolName {\n\t\t\t\tt.Errorf(\"ParseMCPToolName() toolName = %v, want %v\", toolName, tt.wantToolName)\n\t\t\t}\n\t\t\tif ok != tt.wantOK {\n\t\t\t\tt.Errorf(\"ParseMCPToolName() ok = %v, want %v\", ok, tt.wantOK)\n\t\t\t}\n\t\t})\n\t}\n}\n\nfunc TestMCPToolName_RoundTrip(t *testing.T) {\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\ttests := []struct {\n\t\tname string\n\t\tserverID string\n\t\ttoolName string\n\t}{\n\t\t{\n\t\t\tname: \"Simple IDs\",\n\t\t\tserverID: \"github\",\n\t\t\ttoolName: \"search_repos\",\n\t\t},\n\t\t{\n\t\t\tname: \"Server with dots\",\n\t\t\tserverID: \"github.enterprise\",\n\t\t\ttoolName: \"search\",\n\t\t},\n\t\t{\n\t\t\tname: \"Complex server ID\",\n\t\t\tserverID: \"com.example.mcp.server\",\n\t\t\ttoolName: \"tool_name\",\n\t\t},\n\t\t{\n\t\t\tname: \"Server with dashes\",\n\t\t\tserverID: \"mcp-server-123\",\n\t\t\ttoolName: \"tool_with_underscores\",\n\t\t},\n\t\t{\n\t\t\tname: \"Mixed dots and dashes\",\n\t\t\tserverID: \"github.enterprise-prod\",\n\t\t\ttoolName: \"api_call\",\n\t\t},\n\t}\n\n\tfor _, tt := range tests {\n\t\tt.Run(tt.name, func(t *testing.T) {\n\t\t\t// Format\n\t\t\tformatted := assistant.MCPToolName(tt.serverID, tt.toolName)\n\t\t\tif formatted == \"\" {\n\t\t\t\tt.Fatal(\"MCPToolName() returned empty string\")\n\t\t\t}\n\n\t\t\t// Parse\n\t\t\tserverID, toolName, ok := assistant.ParseMCPToolName(formatted)\n\n\t\t\t// Verify round-trip\n\t\t\tif !ok {\n\t\t\t\tt.Fatal(\"ParseMCPToolName() failed\")\n\t\t\t}\n\t\t\tif serverID != tt.serverID {\n\t\t\t\tt.Errorf(\"Round-trip failed: serverID = %v, want %v\", serverID, tt.serverID)\n\t\t\t}\n\t\t\tif toolName != tt.toolName {\n\t\t\t\tt.Errorf(\"Round-trip failed: toolName = %v, want %v\", toolName, tt.toolName)\n\t\t\t}\n\n\t\t\tt.Logf(\"✓ Round-trip successful: (%s, %s) → %s → (%s, %s)\",\n\t\t\t\ttt.serverID, tt.toolName, formatted, serverID, toolName)\n\t\t})\n\t}\n}\n\n// TestMCPToolContextPassing tests that agent context is correctly passed to MCP tools\nfunc TestMCPToolContextPassing(t *testing.T) {\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\t// Get the echo MCP client\n\tclient, err := mcp.Select(\"echo\")\n\tassert.NoError(t, err, \"Failed to select echo MCP client\")\n\tassert.NotNil(t, client, \"MCP client should not be nil\")\n\n\t// Create a test agent context\n\tauthorized := &types.AuthorizedInfo{\n\t\tUserID: \"test-user-123\",\n\t\tTenantID: \"test-tenant-456\",\n\t}\n\tctx := agentContext.New(context.Background(), authorized, \"test-chat-789\")\n\tctx.AssistantID = \"test-assistant-mcptest\"\n\tctx.Locale = \"en\"\n\tctx.Theme = \"dark\"\n\n\t// Call the echo tool with context\n\targs := map[string]interface{}{\n\t\t\"message\": \"test message from context test\",\n\t}\n\n\t// Call the tool - the agent context will be passed as extra parameter\n\tresult, err := client.CallTool(ctx.Context, \"echo\", args, ctx)\n\tassert.NoError(t, err, \"CallTool should not return error\")\n\tassert.NotNil(t, result, \"Result should not be nil\")\n\tassert.False(t, result.IsError, \"Result should not be an error\")\n\tassert.Greater(t, len(result.Content), 0, \"Result should have content\")\n\n\t// Parse the result content\n\tvar echoResult map[string]interface{}\n\terr = jsoniter.Unmarshal([]byte(result.Content[0].Text), &echoResult)\n\tassert.NoError(t, err, \"Failed to parse result content\")\n\n\tt.Logf(\"Echo result: %+v\", echoResult)\n\n\t// Verify the context was received\n\tcontextData, ok := echoResult[\"context\"].(map[string]interface{})\n\tassert.True(t, ok, \"Result should contain context field\")\n\tassert.NotNil(t, contextData, \"Context data should not be nil\")\n\n\t// Verify context has_context flag\n\thasContext, ok := contextData[\"has_context\"].(bool)\n\tassert.True(t, ok, \"Context should have has_context field\")\n\tassert.True(t, hasContext, \"Context should indicate it has context\")\n\n\t// Verify chat_id and assistant_id have values (main verification)\n\tchatID, ok := contextData[\"chat_id\"].(string)\n\tassert.True(t, ok, \"Context should have chat_id field\")\n\tassert.NotEmpty(t, chatID, \"chat_id should have a value\")\n\tassert.Equal(t, \"test-chat-789\", chatID, \"chat_id should match\")\n\n\tassistantID, ok := contextData[\"assistant_id\"].(string)\n\tassert.True(t, ok, \"Context should have assistant_id field\")\n\tassert.NotEmpty(t, assistantID, \"assistant_id should have a value\")\n\tassert.Equal(t, \"test-assistant-mcptest\", assistantID, \"assistant_id should match\")\n\n\t// Verify authorized information\n\tauthorizedData, ok := contextData[\"authorized\"].(map[string]interface{})\n\tassert.True(t, ok, \"Context should have authorized field\")\n\tassert.NotNil(t, authorizedData, \"Authorized data should not be nil\")\n\n\tuserID, ok := authorizedData[\"user_id\"].(string)\n\tassert.True(t, ok, \"Authorized should have user_id field\")\n\tassert.Equal(t, \"test-user-123\", userID, \"User ID should match\")\n\n\ttenantID, ok := authorizedData[\"tenant_id\"].(string)\n\tassert.True(t, ok, \"Authorized should have tenant_id field\")\n\tassert.Equal(t, \"test-tenant-456\", tenantID, \"Tenant ID should match\")\n\n\tt.Logf(\"✓ Context successfully passed to MCP tool\")\n\tt.Logf(\" - ChatID: %s\", chatID)\n\tt.Logf(\" - AssistantID: %s\", assistantID)\n\tt.Logf(\" - UserID: %s\", userID)\n\tt.Logf(\" - TenantID: %s\", tenantID)\n}\n\n// TestMCPToolContextPassingParallel tests that agent context is correctly passed in parallel calls\nfunc TestMCPToolContextPassingParallel(t *testing.T) {\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\t// Get the echo MCP client\n\tclient, err := mcp.Select(\"echo\")\n\tassert.NoError(t, err, \"Failed to select echo MCP client\")\n\tassert.NotNil(t, client, \"MCP client should not be nil\")\n\n\t// Create a test agent context\n\tauthorized := &types.AuthorizedInfo{\n\t\tUserID: \"parallel-user-123\",\n\t\tTenantID: \"parallel-tenant-456\",\n\t}\n\tctx := agentContext.New(context.Background(), authorized, \"parallel-chat-789\")\n\tctx.AssistantID = \"test-assistant-parallel\"\n\tctx.Locale = \"zh-CN\"\n\n\t// Call multiple echo tools in parallel\n\ttoolCalls := []mcpTypes.ToolCall{\n\t\t{\n\t\t\tName: \"echo\",\n\t\t\tArguments: map[string]interface{}{\n\t\t\t\t\"message\": \"parallel message 1\",\n\t\t\t},\n\t\t},\n\t\t{\n\t\t\tName: \"echo\",\n\t\t\tArguments: map[string]interface{}{\n\t\t\t\t\"message\": \"parallel message 2\",\n\t\t\t},\n\t\t},\n\t}\n\n\t// Call tools in parallel - the agent context will be passed as extra parameter\n\tresults, err := client.CallToolsParallel(ctx.Context, toolCalls, ctx)\n\tassert.NoError(t, err, \"CallToolsParallel should not return error\")\n\tassert.NotNil(t, results, \"Results should not be nil\")\n\tassert.Equal(t, 2, len(results.Results), \"Should have 2 results\")\n\n\t// Verify both results received the context\n\tfor i, result := range results.Results {\n\t\tassert.False(t, result.IsError, \"Result %d should not be an error\", i)\n\t\tassert.Greater(t, len(result.Content), 0, \"Result %d should have content\", i)\n\n\t\t// Parse the result content\n\t\tvar echoResult map[string]interface{}\n\t\terr = jsoniter.Unmarshal([]byte(result.Content[0].Text), &echoResult)\n\t\tassert.NoError(t, err, \"Failed to parse result %d content\", i)\n\n\t\t// Verify the context was received\n\t\tcontextData, ok := echoResult[\"context\"].(map[string]interface{})\n\t\tassert.True(t, ok, \"Result %d should contain context field\", i)\n\t\tassert.NotNil(t, contextData, \"Context data %d should not be nil\", i)\n\n\t\thasContext, ok := contextData[\"has_context\"].(bool)\n\t\tassert.True(t, ok, \"Context %d should have has_context field\", i)\n\t\tassert.True(t, hasContext, \"Context %d should indicate it has context\", i)\n\n\t\t// Verify chat_id in parallel call\n\t\tchatID, ok := contextData[\"chat_id\"].(string)\n\t\tassert.True(t, ok, \"Context %d should have chat_id field\", i)\n\t\tassert.Equal(t, \"parallel-chat-789\", chatID, \"Chat ID in result %d should match\", i)\n\n\t\t// Verify authorized information in parallel call\n\t\tauthorizedData, ok := contextData[\"authorized\"].(map[string]interface{})\n\t\tassert.True(t, ok, \"Context %d should have authorized field\", i)\n\t\tif userID, ok := authorizedData[\"user_id\"].(string); ok {\n\t\t\tassert.Equal(t, \"parallel-user-123\", userID, \"User ID in result %d should match\", i)\n\t\t}\n\n\t\tt.Logf(\"✓ Result %d successfully received context\", i)\n\t}\n\n\tt.Log(\"✓ Context successfully passed to all parallel MCP tool calls\")\n}\n" }, { "path": "agent/assistant/next.go", "content": "package assistant\n\nimport (\n\t\"fmt\"\n\n\tagentContext \"github.com/yaoapp/yao/agent/context\"\n\t\"github.com/yaoapp/yao/agent/output/message\"\n)\n\n// processNextResponse processes the Next hook's response and handles agent delegation or custom data\nfunc (ast *Assistant) processNextResponse(npc *NextProcessContext) (*agentContext.Response, error) {\n\t// If no Next hook response, return standard response\n\tif npc.NextResponse == nil {\n\t\treturn ast.buildStandardResponse(npc), nil\n\t}\n\n\t// Handle Delegate: call another agent\n\t// Note: User input is already buffered by root agent, delegated agent will skip buffering\n\tif npc.NextResponse.Delegate != nil {\n\t\treturn ast.handleDelegation(npc.Context, npc.NextResponse.Delegate, npc.StreamHandler)\n\t}\n\n\t// Handle custom Data: return as-is wrapped in standard Response\n\tif npc.NextResponse.Data != nil {\n\t\treturn &agentContext.Response{\n\t\t\tContextID: npc.Context.ID,\n\t\t\tRequestID: npc.Context.RequestID(),\n\t\t\tTraceID: npc.Context.TraceID(),\n\t\t\tChatID: npc.Context.ChatID,\n\t\t\tAssistantID: ast.ID,\n\t\t\tCreate: npc.CreateResponse,\n\t\t\tNext: npc.NextResponse.Data, // Put custom data in Next field\n\t\t\tCompletion: npc.CompletionResponse,\n\t\t\tTools: npc.ToolCallResponses,\n\t\t}, nil\n\t}\n\n\t// No delegate or data, return standard response\n\treturn ast.buildStandardResponse(npc), nil\n}\n\n// handleDelegation handles calling another agent based on DelegateConfig\nfunc (ast *Assistant) handleDelegation(\n\tctx *agentContext.Context,\n\tdelegate *agentContext.DelegateConfig,\n\tstreamHandler func(message.StreamChunkType, []byte) int,\n) (*agentContext.Response, error) {\n\t// Load the target assistant\n\ttargetAssistant, err := Get(delegate.AgentID)\n\tif err != nil {\n\t\treturn nil, fmt.Errorf(\"failed to load delegated assistant '%s': %w\", delegate.AgentID, err)\n\t}\n\n\t// Mark this as an agent-to-agent call for proper source tracking\n\tctx.Referer = agentContext.RefererAgent\n\n\t// Call the delegated assistant with the same context\n\t// The delegated assistant's Stream method will:\n\t// 1. Call EnterStack() to push itself onto the Stack (creating parent-child relationship)\n\t// 2. Execute with the same Context (preserving ID, Space, Writer, etc.)\n\t// 3. Call done() to pop from Stack when finished\n\t// This ensures proper Stack tracing: parent assistant -> delegated assistant\n\n\t// Convert options map from delegate config to Options struct\n\tdelegateOpts := agentContext.OptionsFromMap(delegate.Options)\n\treturn targetAssistant.Stream(ctx, delegate.Messages, delegateOpts)\n}\n\n// buildStandardResponse builds the standard agent response when no custom Next hook processing is needed\nfunc (ast *Assistant) buildStandardResponse(npc *NextProcessContext) *agentContext.Response {\n\n\tvar next interface{} = nil\n\tif npc.NextResponse != nil {\n\t\tnext = npc.NextResponse\n\t}\n\n\treturn &agentContext.Response{\n\t\tContextID: npc.Context.ID,\n\t\tRequestID: npc.Context.RequestID(),\n\t\tTraceID: npc.Context.TraceID(),\n\t\tChatID: npc.Context.ChatID,\n\t\tAssistantID: ast.ID,\n\t\tCreate: npc.CreateResponse,\n\t\tNext: next,\n\t\tCompletion: npc.CompletionResponse,\n\t\tTools: npc.ToolCallResponses,\n\t}\n}\n" }, { "path": "agent/assistant/permission.go", "content": "package assistant\n\nimport (\n\t\"fmt\"\n\n\t\"github.com/yaoapp/yao/agent/context\"\n)\n\nfunc (ast *Assistant) checkPermissions(ctx *context.Context) error {\n\tif ctx.Authorized == nil {\n\t\treturn fmt.Errorf(\"authorized information not found\")\n\t}\n\treturn nil\n}\n" }, { "path": "agent/assistant/sandbox.go", "content": "package assistant\n\nimport (\n\tstdContext \"context\"\n\t\"encoding/json\"\n\t\"fmt\"\n\t\"os\"\n\t\"path/filepath\"\n\t\"sync\"\n\t\"time\"\n\n\t\"github.com/yaoapp/gou/connector\"\n\tgouMCP \"github.com/yaoapp/gou/mcp\"\n\tmcpProcess \"github.com/yaoapp/gou/mcp/process\"\n\t\"github.com/yaoapp/yao/agent/context\"\n\t\"github.com/yaoapp/yao/agent/i18n\"\n\t\"github.com/yaoapp/yao/agent/output/message\"\n\tagentsandbox \"github.com/yaoapp/yao/agent/sandbox\"\n\t\"github.com/yaoapp/yao/config\"\n\tinfraSandbox \"github.com/yaoapp/yao/sandbox\"\n\t\"github.com/yaoapp/yao/sandbox/ipc\"\n\ttraceTypes \"github.com/yaoapp/yao/trace/types\"\n)\n\nvar (\n\tsandboxManager *infraSandbox.Manager\n\tsandboxManagerOnce sync.Once\n\tsandboxManagerErr error\n)\n\n// GetSandboxManager returns the sandbox manager singleton\n// Returns nil and error if sandbox is not configured or Docker is unavailable\nfunc GetSandboxManager() (*infraSandbox.Manager, error) {\n\tsandboxManagerOnce.Do(func() {\n\t\t// Create sandbox config from Yao config\n\t\tcfg := &infraSandbox.Config{}\n\n\t\t// Use YAO_DATA_ROOT for workspace and IPC paths\n\t\tdataRoot := config.Conf.DataRoot\n\t\tif dataRoot != \"\" {\n\t\t\tcfg.Init(dataRoot)\n\t\t}\n\n\t\t// Create manager (will fail if Docker is not available)\n\t\tsandboxManager, sandboxManagerErr = infraSandbox.NewManager(cfg)\n\t})\n\n\treturn sandboxManager, sandboxManagerErr\n}\n\n// HasSandbox returns true if the assistant has sandbox configuration\nfunc (ast *Assistant) HasSandbox() bool {\n\treturn ast.Sandbox != nil && ast.Sandbox.Command != \"\"\n}\n\n// initSandbox initializes the sandbox executor\n// Returns the full Executor (for LLM calls), cleanup function, and any error\n// This is called BEFORE hooks so that hooks can access ctx.sandbox\n// The executor implements both agentsandbox.Executor and context.SandboxExecutor interfaces\nfunc (ast *Assistant) initSandbox(ctx *context.Context, opts *context.Options) (agentsandbox.Executor, func(), string, error) {\n\t// Get sandbox manager (singleton)\n\tmanager, err := GetSandboxManager()\n\tif err != nil {\n\t\tctx.Logger.Error(\"Sandbox manager initialization failed: %v\", err)\n\t\treturn nil, nil, \"\", fmt.Errorf(\"sandbox manager not available: %w\", err)\n\t}\n\tif manager == nil {\n\t\treturn nil, nil, \"\", fmt.Errorf(\"sandbox manager not initialized\")\n\t}\n\n\t// Build executor options from assistant config\n\texecOpts, err := ast.buildSandboxOptions(ctx, opts)\n\tif err != nil {\n\t\tctx.Logger.Error(\"Failed to build sandbox options: %v\", err)\n\t\treturn nil, nil, \"\", fmt.Errorf(\"failed to build sandbox options: %w\", err)\n\t}\n\n\t// Log sandbox creation\n\tctx.Logger.Info(\"Creating sandbox container for command: %s\", ast.Sandbox.Command)\n\n\t// Add trace for sandbox creation\n\ttrace, traceErr := ctx.Trace()\n\tif traceErr == nil && trace != nil {\n\t\ttrace.Info(\"Creating sandbox container...\")\n\t}\n\n\t// Send loading message to user\n\tloadingMsg := &message.Message{\n\t\tType: message.TypeLoading,\n\t\tProps: map[string]interface{}{\n\t\t\t\"message\": i18n.T(ctx.Locale, \"sandbox.preparing\"),\n\t\t},\n\t}\n\tloadingMsgID, _ := ctx.SendStream(loadingMsg)\n\n\t// Create executor (container starts here)\n\texecutor, err := agentsandbox.New(manager, execOpts)\n\tif err != nil {\n\t\tctx.Logger.Error(\"Sandbox creation failed: %v\", err)\n\t\tif traceErr == nil && trace != nil {\n\t\t\ttrace.Error(\"Sandbox creation failed: %v\", err)\n\t\t}\n\t\t// End loading message with done:true\n\t\tif loadingMsgID != \"\" {\n\t\t\tdoneMsg := &message.Message{\n\t\t\t\tMessageID: loadingMsgID,\n\t\t\t\tDelta: true,\n\t\t\t\tDeltaAction: message.DeltaReplace,\n\t\t\t\tType: message.TypeLoading,\n\t\t\t\tProps: map[string]interface{}{\n\t\t\t\t\t\"message\": i18n.T(ctx.Locale, \"sandbox.failed\"),\n\t\t\t\t\t\"done\": true,\n\t\t\t\t},\n\t\t\t}\n\t\t\tctx.Send(doneMsg)\n\t\t}\n\t\treturn nil, nil, \"\", fmt.Errorf(\"failed to create sandbox executor: %w\", err)\n\t}\n\n\t// Log sandbox ready\n\tctx.Logger.Info(\"Sandbox container ready\")\n\tif traceErr == nil && trace != nil {\n\t\ttrace.Info(\"Sandbox container ready\")\n\t}\n\n\t// Return cleanup function\n\tcleanup := func() {\n\t\tif err := executor.Close(); err != nil {\n\t\t\tctx.Logger.Error(\"Failed to close sandbox executor: %v\", err)\n\t\t}\n\t}\n\n\t// Keep loadingMsgID open - it will be closed when first output is received\n\t// This provides better UX: user sees \"Preparing...\" until actual content appears\n\treturn executor, cleanup, loadingMsgID, nil\n}\n\n// executeSandboxStream executes the request using sandbox (Claude CLI, etc.)\n// This is called when ast.Sandbox is configured\n// NOTE: The executor is passed directly from initSandbox, no type assertion needed\nfunc (ast *Assistant) executeSandboxStream(\n\tctx *context.Context,\n\tcompletionMessages []context.Message,\n\tagentNode traceTypes.Node,\n\tstreamHandler message.StreamFunc,\n\texecutor agentsandbox.Executor,\n\tloadingMsgID string,\n) (*context.CompletionResponse, error) {\n\n\t// Mark the agentNode as used to avoid unused variable error\n\t_ = agentNode\n\n\tif executor == nil {\n\t\treturn nil, fmt.Errorf(\"sandbox executor not initialized (call initSandbox first)\")\n\t}\n\n\t// Log sandbox execution\n\tctx.Logger.Info(\"Executing via sandbox (command: %s)\", ast.Sandbox.Command)\n\n\t// Pass the \"preparing sandbox\" loading message ID to executor\n\t// It will be closed when first output (text or tool) is received\n\tif loadingMsgID != \"\" {\n\t\texecutor.SetLoadingMsgID(loadingMsgID)\n\t}\n\n\t// Execute LLM call via sandbox\n\t// The loadingMsgID will be closed when first output is received\n\t// Tool calls will create their own loading messages below the text\n\tresp, err := executor.Stream(ctx, completionMessages, streamHandler)\n\n\tif err != nil {\n\t\t// Close loading message on error\n\t\tif loadingMsgID != \"\" {\n\t\t\tdoneMsg := &message.Message{\n\t\t\t\tMessageID: loadingMsgID,\n\t\t\t\tDelta: true,\n\t\t\t\tDeltaAction: message.DeltaReplace,\n\t\t\t\tType: message.TypeLoading,\n\t\t\t\tProps: map[string]interface{}{\n\t\t\t\t\t\"message\": i18n.T(ctx.Locale, \"sandbox.failed\"),\n\t\t\t\t\t\"done\": true,\n\t\t\t\t},\n\t\t\t}\n\t\t\tctx.Send(doneMsg)\n\t\t}\n\n\t\t// Send error message to client\n\t\terrMsg := &message.Message{\n\t\t\tType: message.TypeError,\n\t\t\tProps: map[string]interface{}{\n\t\t\t\t\"message\": err.Error(),\n\t\t\t},\n\t\t}\n\t\tctx.Send(errMsg)\n\t\treturn nil, fmt.Errorf(\"sandbox execution failed: %w\", err)\n\t}\n\n\treturn resp, nil\n}\n\n// buildSandboxOptions builds executor options from assistant config\nfunc (ast *Assistant) buildSandboxOptions(ctx *context.Context, opts *context.Options) (*agentsandbox.Options, error) {\n\tif ast.Sandbox == nil {\n\t\treturn nil, fmt.Errorf(\"sandbox configuration is required\")\n\t}\n\n\texecOpts := &agentsandbox.Options{\n\t\tCommand: ast.Sandbox.Command,\n\t\tImage: ast.Sandbox.Image,\n\t\tMaxMemory: ast.Sandbox.MaxMemory,\n\t\tMaxCPU: ast.Sandbox.MaxCPU,\n\t\tArguments: ast.Sandbox.Arguments,\n\t}\n\n\t// Parse timeout string (e.g., \"10m\") to duration\n\tif ast.Sandbox.Timeout != \"\" {\n\t\ttimeout, err := time.ParseDuration(ast.Sandbox.Timeout)\n\t\tif err != nil {\n\t\t\treturn nil, fmt.Errorf(\"invalid timeout format: %w\", err)\n\t\t}\n\t\texecOpts.Timeout = timeout\n\t}\n\n\t// Set user and chat IDs for workspace isolation\n\tif ctx.Authorized != nil && ctx.Authorized.UserID != \"\" {\n\t\texecOpts.UserID = ctx.Authorized.UserID\n\t} else {\n\t\texecOpts.UserID = \"anonymous\"\n\t}\n\texecOpts.ChatID = ctx.ChatID\n\n\t// Set skills directory (auto-resolved from assistant path)\n\t// Only set if the directory actually exists\n\tif ast.Path != \"\" {\n\t\tappRoot := config.Conf.AppSource\n\t\tskillsDir := filepath.Join(appRoot, ast.Path, \"skills\")\n\t\tif info, err := os.Stat(skillsDir); err == nil && info.IsDir() {\n\t\t\texecOpts.SkillsDir = skillsDir\n\t\t\tctx.Logger.Debug(\"Skills directory found: %s\", skillsDir)\n\t\t}\n\t}\n\n\t// Check if assistant has prompts (from prompts.yml)\n\t// If prompts are configured, we need to call Claude CLI\n\tif len(ast.Prompts) > 0 {\n\t\t// Extract system prompt from prompts\n\t\tfor _, prompt := range ast.Prompts {\n\t\t\tif prompt.Role == \"system\" && prompt.Content != \"\" {\n\t\t\t\texecOpts.SystemPrompt = prompt.Content\n\t\t\t\tbreak\n\t\t\t}\n\t\t}\n\t}\n\n\t// Resolve connector settings\n\tconn, _, err := ast.GetConnector(ctx, opts)\n\tif err != nil {\n\t\treturn nil, fmt.Errorf(\"failed to get connector: %w\", err)\n\t}\n\n\t// Determine connector type for sandbox proxy behavior\n\t// Anthropic connectors bypass the proxy (Claude CLI connects directly)\n\tif conn.Is(connector.ANTHROPIC) {\n\t\texecOpts.ConnectorType = \"anthropic\"\n\t} else {\n\t\texecOpts.ConnectorType = \"openai\"\n\t}\n\n\tsetting := conn.Setting()\n\tif host, ok := setting[\"host\"].(string); ok {\n\t\texecOpts.ConnectorHost = host\n\t}\n\tif key, ok := setting[\"key\"].(string); ok {\n\t\texecOpts.ConnectorKey = key\n\t}\n\tif model, ok := setting[\"model\"].(string); ok {\n\t\texecOpts.Model = model\n\t}\n\n\t// Extract extra connector options (thinking, max_tokens, temperature, etc.)\n\t// These are backend-specific parameters that need to be passed through to the proxy\n\tconnectorOptions := make(map[string]interface{})\n\tfor k, v := range setting {\n\t\t// Skip standard fields that are already handled\n\t\tswitch k {\n\t\tcase \"host\", \"key\", \"model\", \"azure\", \"capabilities\":\n\t\t\tcontinue\n\t\tdefault:\n\t\t\t// Include all other fields as extra options\n\t\t\tconnectorOptions[k] = v\n\t\t}\n\t}\n\tif len(connectorOptions) > 0 {\n\t\texecOpts.ConnectorOptions = connectorOptions\n\t\tctx.Logger.Debug(\"Connector options extracted: %v\", connectorOptions)\n\t}\n\n\t// Extract secrets from sandbox config (e.g., GITHUB_TOKEN: \"$ENV.GITHUB_TOKEN\")\n\tif ast.Sandbox != nil && len(ast.Sandbox.Secrets) > 0 {\n\t\tsecrets := make(map[string]string)\n\t\tfor k, v := range ast.Sandbox.Secrets {\n\t\t\t// Resolve $ENV.XXX references\n\t\t\tresolved := resolveEnvValue(v)\n\t\t\tif resolved != \"\" {\n\t\t\t\tsecrets[k] = resolved\n\t\t\t}\n\t\t}\n\t\tif len(secrets) > 0 {\n\t\t\texecOpts.Secrets = secrets\n\t\t\tctx.Logger.Debug(\"Secrets extracted: %d items\", len(secrets))\n\t\t}\n\t}\n\n\t// Build MCP config and load tools if the assistant has MCP servers configured\n\tif ast.MCP != nil && len(ast.MCP.Servers) > 0 {\n\t\t// Build MCP config for Claude CLI\n\t\tmcpConfig, err := ast.BuildMCPConfigForSandbox(ctx)\n\t\tif err != nil {\n\t\t\tctx.Logger.Warn(\"Failed to build MCP config for sandbox: %v\", err)\n\t\t\t// Non-fatal: sandbox can work without MCP\n\t\t} else {\n\t\t\texecOpts.MCPConfig = mcpConfig\n\t\t\tctx.Logger.Debug(\"MCP config built for sandbox (%d bytes)\", len(mcpConfig))\n\t\t}\n\n\t\t// Load MCP tools for IPC session\n\t\tmcpTools, err := ast.loadMCPToolsForIPC(ctx)\n\t\tif err != nil {\n\t\t\tctx.Logger.Warn(\"Failed to load MCP tools for IPC: %v\", err)\n\t\t\t// Non-fatal: IPC will have no tools\n\t\t} else if len(mcpTools) > 0 {\n\t\t\texecOpts.MCPTools = mcpTools\n\t\t\tctx.Logger.Debug(\"Loaded %d MCP tools for IPC\", len(mcpTools))\n\t\t}\n\t}\n\n\treturn execOpts, nil\n}\n\n// loadMCPToolsForIPC loads MCP tools from configured servers and converts them to IPC format\nfunc (ast *Assistant) loadMCPToolsForIPC(ctx *context.Context) (map[string]*ipc.MCPTool, error) {\n\tif ast.MCP == nil || len(ast.MCP.Servers) == 0 {\n\t\treturn nil, nil\n\t}\n\n\ttools := make(map[string]*ipc.MCPTool)\n\tstdCtx := ctx.Context\n\tif stdCtx == nil {\n\t\tstdCtx = stdContext.Background()\n\t}\n\n\tfor _, serverConfig := range ast.MCP.Servers {\n\t\tif serverConfig.ServerID == \"\" {\n\t\t\tcontinue\n\t\t}\n\n\t\t// Get MCP client\n\t\tclient, err := gouMCP.Select(serverConfig.ServerID)\n\t\tif err != nil {\n\t\t\tctx.Logger.Warn(\"MCP server '%s' not found: %v\", serverConfig.ServerID, err)\n\t\t\tcontinue\n\t\t}\n\n\t\t// List tools from the MCP client\n\t\ttoolsResp, err := client.ListTools(stdCtx, \"\")\n\t\tif err != nil {\n\t\t\tctx.Logger.Warn(\"Failed to list tools from MCP server '%s': %v\", serverConfig.ServerID, err)\n\t\t\tcontinue\n\t\t}\n\n\t\t// Get tool mapping for process names\n\t\tmapping, ok := mcpProcess.GetMapping(serverConfig.ServerID)\n\t\tif !ok {\n\t\t\tctx.Logger.Warn(\"No mapping found for MCP server '%s'\", serverConfig.ServerID)\n\t\t\tcontinue\n\t\t}\n\n\t\t// Filter tools if specified in config\n\t\ttoolFilter := make(map[string]bool)\n\t\tif len(serverConfig.Tools) > 0 {\n\t\t\tfor _, t := range serverConfig.Tools {\n\t\t\t\ttoolFilter[t] = true\n\t\t\t}\n\t\t}\n\n\t\t// Convert tools to IPC format\n\t\t// Tool names are prefixed with server ID to avoid conflicts\n\t\t// e.g., \"echo\" server's \"ping\" tool becomes \"echo__ping\"\n\t\tfor _, tool := range toolsResp.Tools {\n\t\t\t// Apply tool filter if specified\n\t\t\tif len(toolFilter) > 0 && !toolFilter[tool.Name] {\n\t\t\t\tcontinue\n\t\t\t}\n\n\t\t\t// Find the process name from mapping\n\t\t\tprocessName := \"\"\n\t\t\tif toolSchema, ok := mapping.Tools[tool.Name]; ok {\n\t\t\t\tprocessName = toolSchema.Process\n\t\t\t}\n\t\t\tif processName == \"\" {\n\t\t\t\tctx.Logger.Warn(\"No process mapping for tool '%s' in server '%s'\", tool.Name, serverConfig.ServerID)\n\t\t\t\tcontinue\n\t\t\t}\n\n\t\t\t// Prefixed tool name: serverID__toolName\n\t\t\t// This matches Claude's MCP naming: mcp__yao__serverID__toolName\n\t\t\tprefixedName := serverConfig.ServerID + \"__\" + tool.Name\n\n\t\t\t// Create IPC tool entry with prefixed name\n\t\t\tipcTool := &ipc.MCPTool{\n\t\t\t\tName: prefixedName,\n\t\t\t\tDescription: tool.Description,\n\t\t\t\tProcess: processName,\n\t\t\t\tInputSchema: tool.InputSchema,\n\t\t\t}\n\n\t\t\ttools[prefixedName] = ipcTool\n\t\t}\n\t}\n\n\treturn tools, nil\n}\n\n// BuildMCPConfigForSandbox builds the MCP configuration JSON for sandbox\n// This creates a .mcp.json format that Claude CLI can understand\n// Exported for testing\nfunc (ast *Assistant) BuildMCPConfigForSandbox(ctx *context.Context) ([]byte, error) {\n\tif ast.MCP == nil || len(ast.MCP.Servers) == 0 {\n\t\treturn nil, nil\n\t}\n\n\t// Build MCP config in Claude CLI format\n\t// Claude CLI expects: { \"mcpServers\": { \"server_id\": { \"command\": \"...\", \"args\": [...] } } }\n\t//\n\t// For Yao's MCP servers, we use yao-bridge to connect to the IPC socket.\n\t// yao-bridge bridges stdio to Unix socket, allowing Claude CLI to communicate\n\t// with Yao's IPC server running on the host.\n\t//\n\t// Architecture:\n\t// Claude CLI → yao-bridge → Unix Socket → IPC Session → Yao Process\n\tconfig := map[string]interface{}{\n\t\t\"mcpServers\": map[string]interface{}{\n\t\t\t// Single \"yao\" server that handles all MCP tools via IPC\n\t\t\t\"yao\": map[string]interface{}{\n\t\t\t\t\"command\": \"yao-bridge\",\n\t\t\t\t\"args\": []string{\"/tmp/yao.sock\"}, // ContainerIPCSocket from sandbox config\n\t\t\t},\n\t\t},\n\t}\n\n\treturn json.Marshal(config)\n}\n\n// resolveEnvValue resolves environment variable references in a string\n// Supports format: $ENV.VAR_NAME or plain value\n// Returns empty string if the variable is not set\nfunc resolveEnvValue(value string) string {\n\tif value == \"\" {\n\t\treturn \"\"\n\t}\n\n\t// Check for $ENV.XXX format\n\tif len(value) > 5 && value[:5] == \"$ENV.\" {\n\t\tenvName := value[5:]\n\t\treturn os.Getenv(envName)\n\t}\n\n\t// Return as-is if not an env reference\n\treturn value\n}\n" }, { "path": "agent/assistant/sandbox_debug_test.go", "content": "package assistant_test\n\nimport (\n\t\"testing\"\n\n\t\"github.com/stretchr/testify/assert\"\n\t\"github.com/stretchr/testify/require\"\n\t\"github.com/yaoapp/yao/agent/assistant\"\n\t\"github.com/yaoapp/yao/agent/testutils\"\n)\n\n// TestSandboxDebugHasSandbox tests the HasSandbox method directly\nfunc TestSandboxDebugHasSandbox(t *testing.T) {\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\ttestCases := []struct {\n\t\tname string\n\t\tassistantID string\n\t\texpectTrue bool\n\t}{\n\t\t{\"BasicSandbox\", \"tests.sandbox.basic\", true},\n\t\t{\"HooksSandbox\", \"tests.sandbox.hooks\", true},\n\t\t{\"FullSandbox\", \"tests.sandbox.full\", true},\n\t}\n\n\tfor _, tc := range testCases {\n\t\tt.Run(tc.name, func(t *testing.T) {\n\t\t\tast, err := assistant.Get(tc.assistantID)\n\t\t\trequire.NoError(t, err, \"Failed to get assistant %s\", tc.assistantID)\n\n\t\t\t// Check Sandbox struct\n\t\t\tt.Logf(\"Assistant ID: %s\", ast.ID)\n\t\t\tt.Logf(\"Sandbox: %+v\", ast.Sandbox)\n\n\t\t\tif ast.Sandbox != nil {\n\t\t\t\tt.Logf(\"Sandbox.Command: %q\", ast.Sandbox.Command)\n\t\t\t\tt.Logf(\"Sandbox.Timeout: %s\", ast.Sandbox.Timeout)\n\t\t\t\tt.Logf(\"Sandbox.Image: %s\", ast.Sandbox.Image)\n\t\t\t\tt.Logf(\"Sandbox.Arguments: %v\", ast.Sandbox.Arguments)\n\t\t\t}\n\n\t\t\t// Check HasSandbox\n\t\t\thasSandbox := ast.HasSandbox()\n\t\t\tt.Logf(\"HasSandbox() = %v\", hasSandbox)\n\n\t\t\tif tc.expectTrue {\n\t\t\t\tassert.True(t, hasSandbox, \"Expected HasSandbox() to be true for %s\", tc.assistantID)\n\t\t\t} else {\n\t\t\t\tassert.False(t, hasSandbox, \"Expected HasSandbox() to be false for %s\", tc.assistantID)\n\t\t\t}\n\t\t})\n\t}\n}\n\n// TestSandboxDebugPrompts tests if Prompts is set (affects execution path)\nfunc TestSandboxDebugPrompts(t *testing.T) {\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\tast, err := assistant.Get(\"tests.sandbox.basic\")\n\trequire.NoError(t, err)\n\n\tt.Logf(\"Assistant ID: %s\", ast.ID)\n\tt.Logf(\"Prompts: %v\", ast.Prompts)\n\tt.Logf(\"MCP: %v\", ast.MCP)\n\tt.Logf(\"HasSandbox: %v\", ast.HasSandbox())\n\n\t// The condition in agent.go is:\n\t// if ast.Prompts != nil || ast.MCP != nil {\n\t// // ... execute LLM\n\t// if ast.HasSandbox() {\n\t// // sandbox path\n\t// } else {\n\t// // direct LLM path\n\t// }\n\t// }\n\t// So we need Prompts or MCP to be non-nil\n\tif ast.Prompts == nil && ast.MCP == nil {\n\t\tt.Log(\"WARNING: Neither Prompts nor MCP is set, LLM phase will be skipped entirely!\")\n\t}\n}\n" }, { "path": "agent/assistant/sandbox_e2e_test.go", "content": "package assistant_test\n\nimport (\n\tstdContext \"context\"\n\t\"fmt\"\n\t\"strings\"\n\t\"testing\"\n\t\"time\"\n\n\t\"github.com/stretchr/testify/assert\"\n\t\"github.com/stretchr/testify/require\"\n\t\"github.com/yaoapp/yao/agent/assistant\"\n\t\"github.com/yaoapp/yao/agent/context\"\n\t\"github.com/yaoapp/yao/agent/testutils\"\n\t\"github.com/yaoapp/yao/openapi/oauth/types\"\n)\n\n// newSandboxE2EContext creates a Context for sandbox E2E testing\n// Uses unique chatID to avoid container name conflicts\nfunc newSandboxE2EContext(chatIDPrefix, assistantID string) *context.Context {\n\t// Generate unique chatID using timestamp to avoid container conflicts\n\tchatID := fmt.Sprintf(\"%s-%d\", chatIDPrefix, time.Now().UnixNano())\n\n\tauthorized := &types.AuthorizedInfo{\n\t\tSubject: \"sandbox-e2e-test-user\",\n\t\tClientID: \"sandbox-e2e-test-client\",\n\t\tScope: \"openid profile\",\n\t\tSessionID: \"sandbox-e2e-test-session\",\n\t\tUserID: \"sandbox-user-123\",\n\t\tTeamID: \"sandbox-team-456\",\n\t\tTenantID: \"sandbox-tenant-789\",\n\t}\n\n\tctx := context.New(stdContext.Background(), authorized, chatID)\n\tctx.AssistantID = assistantID\n\tctx.Locale = \"en-us\"\n\tctx.Theme = \"light\"\n\tctx.Client = context.Client{\n\t\tType: \"web\",\n\t\tUserAgent: \"SandboxE2ETest/1.0\",\n\t\tIP: \"127.0.0.1\",\n\t}\n\tctx.Referer = context.RefererAPI\n\tctx.Accept = context.AcceptWebCUI\n\tctx.Route = \"\"\n\tctx.Metadata = make(map[string]interface{})\n\treturn ctx\n}\n\n// TestSandboxBasicE2E tests the basic sandbox assistant end-to-end\n// This test verifies that:\n// 1. Sandbox is correctly initialized\n// 2. Claude CLI command is built correctly\n// 3. Docker container is created and managed\nfunc TestSandboxBasicE2E(t *testing.T) {\n\tif testing.Short() {\n\t\tt.Skip(\"Skipping sandbox E2E test in short mode\")\n\t}\n\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\t// Load the basic sandbox assistant\n\tast, err := assistant.Get(\"tests.sandbox.basic\")\n\tif err != nil {\n\t\tt.Skipf(\"Skipping test: sandbox assistant not available: %v\", err)\n\t}\n\n\t// Verify sandbox is configured\n\trequire.NotNil(t, ast.Sandbox, \"Sandbox should be configured\")\n\tassert.Equal(t, \"claude\", ast.Sandbox.Command)\n\tt.Logf(\"✓ Sandbox configured with command: %s\", ast.Sandbox.Command)\n\n\t// Create context\n\tctx := newSandboxE2EContext(\"sandbox-basic-e2e\", \"tests.sandbox.basic\")\n\n\t// Test messages\n\tmessages := []context.Message{\n\t\t{Role: context.RoleUser, Content: \"echo hello sandbox\"},\n\t}\n\n\t// Execute stream\n\t// Note: This will fail if Docker/Claude image is not available, which is expected in CI\n\tresponse, err := ast.Stream(ctx, messages)\n\tif err != nil {\n\t\t// Check if it's a Docker/sandbox availability issue\n\t\terrStr := err.Error()\n\t\tif strings.Contains(errStr, \"Docker\") ||\n\t\t\tstrings.Contains(errStr, \"sandbox\") ||\n\t\t\tstrings.Contains(errStr, \"container\") ||\n\t\t\tstrings.Contains(errStr, \"image\") {\n\t\t\tt.Skipf(\"Skipping test: Docker/sandbox not available: %v\", err)\n\t\t}\n\t\tt.Fatalf(\"Stream failed: %v\", err)\n\t}\n\n\t// Verify response\n\trequire.NotNil(t, response, \"Response should not be nil\")\n\n\t// Verify response completion (Claude CLI should return some response)\n\tif response.Completion != nil && response.Completion.Content != nil {\n\t\tif contentStr, ok := response.Completion.Content.(string); ok && contentStr != \"\" {\n\t\t\tt.Logf(\"✓ Response content: %s\", truncateString(contentStr, 200))\n\t\t} else {\n\t\t\tt.Logf(\"⚠ Response content type: %T\", response.Completion.Content)\n\t\t}\n\t} else {\n\t\tt.Log(\"⚠ Response content is empty (might be expected for some commands)\")\n\t}\n\n\tt.Log(\"✓ Basic sandbox E2E test passed\")\n}\n\n// truncateString truncates a string to maxLen and adds \"...\" if truncated\nfunc truncateString(s string, maxLen int) string {\n\tif len(s) <= maxLen {\n\t\treturn s\n\t}\n\treturn s[:maxLen] + \"...\"\n}\n\n// TestSandboxHooksE2E tests the sandbox assistant with hooks\nfunc TestSandboxHooksE2E(t *testing.T) {\n\tif testing.Short() {\n\t\tt.Skip(\"Skipping sandbox E2E test in short mode\")\n\t}\n\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\t// Load the hooks sandbox assistant\n\tast, err := assistant.Get(\"tests.sandbox.hooks\")\n\tif err != nil {\n\t\tt.Skipf(\"Skipping test: sandbox hooks assistant not available: %v\", err)\n\t}\n\n\t// Verify sandbox and hooks are configured\n\trequire.NotNil(t, ast.Sandbox, \"Sandbox should be configured\")\n\trequire.NotNil(t, ast.HookScript, \"HookScript should be loaded\")\n\tt.Logf(\"✓ Sandbox and hooks configured\")\n\n\t// Create context\n\tctx := newSandboxE2EContext(\"sandbox-hooks-e2e\", \"tests.sandbox.hooks\")\n\n\t// Test messages\n\tmessages := []context.Message{\n\t\t{Role: context.RoleUser, Content: \"test hooks integration\"},\n\t}\n\n\t// Execute stream\n\tresponse, err := ast.Stream(ctx, messages)\n\tif err != nil {\n\t\terrStr := err.Error()\n\t\tif strings.Contains(errStr, \"Docker\") ||\n\t\t\tstrings.Contains(errStr, \"sandbox\") ||\n\t\t\tstrings.Contains(errStr, \"container\") ||\n\t\t\tstrings.Contains(errStr, \"image\") {\n\t\t\tt.Skipf(\"Skipping test: Docker/sandbox not available: %v\", err)\n\t\t}\n\t\tt.Fatalf(\"Stream failed: %v\", err)\n\t}\n\n\trequire.NotNil(t, response, \"Response should not be nil\")\n\tt.Log(\"✓ Sandbox hooks E2E test passed\")\n}\n\n// TestSandboxFullE2E tests the full sandbox assistant with MCPs and Skills\nfunc TestSandboxFullE2E(t *testing.T) {\n\tif testing.Short() {\n\t\tt.Skip(\"Skipping sandbox E2E test in short mode\")\n\t}\n\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\t// Load the full sandbox assistant\n\tast, err := assistant.Get(\"tests.sandbox.full\")\n\tif err != nil {\n\t\tt.Skipf(\"Skipping test: full sandbox assistant not available: %v\", err)\n\t}\n\n\t// Verify all components are configured\n\trequire.NotNil(t, ast.Sandbox, \"Sandbox should be configured\")\n\trequire.NotNil(t, ast.MCP, \"MCP should be configured\")\n\trequire.NotNil(t, ast.HookScript, \"HookScript should be loaded\")\n\tt.Logf(\"✓ Full sandbox configured: command=%s, MCP servers=%d\",\n\t\tast.Sandbox.Command, len(ast.MCP.Servers))\n\n\t// Verify MCP configuration\n\tassert.Len(t, ast.MCP.Servers, 1)\n\tassert.Equal(t, \"echo\", ast.MCP.Servers[0].ServerID)\n\tt.Logf(\"✓ MCP server: %s with tools %v\", ast.MCP.Servers[0].ServerID, ast.MCP.Servers[0].Tools)\n\n\t// Create context\n\tctx := newSandboxE2EContext(\"sandbox-full-e2e\", \"tests.sandbox.full\")\n\n\t// Test messages\n\tmessages := []context.Message{\n\t\t{Role: context.RoleUser, Content: \"test full sandbox with MCP and skills\"},\n\t}\n\n\t// Execute stream\n\tresponse, err := ast.Stream(ctx, messages)\n\tif err != nil {\n\t\terrStr := err.Error()\n\t\tif strings.Contains(errStr, \"Docker\") ||\n\t\t\tstrings.Contains(errStr, \"sandbox\") ||\n\t\t\tstrings.Contains(errStr, \"container\") ||\n\t\t\tstrings.Contains(errStr, \"image\") {\n\t\t\tt.Skipf(\"Skipping test: Docker/sandbox not available: %v\", err)\n\t\t}\n\t\tt.Fatalf(\"Stream failed: %v\", err)\n\t}\n\n\trequire.NotNil(t, response, \"Response should not be nil\")\n\tt.Log(\"✓ Full sandbox E2E test passed\")\n}\n\n// TestSandboxContextAccess tests that sandbox is accessible in hooks via ctx.sandbox\nfunc TestSandboxContextAccess(t *testing.T) {\n\tif testing.Short() {\n\t\tt.Skip(\"Skipping sandbox context access test in short mode\")\n\t}\n\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\t// Load the hooks sandbox assistant\n\tast, err := assistant.Get(\"tests.sandbox.hooks\")\n\tif err != nil {\n\t\tt.Skipf(\"Skipping test: sandbox hooks assistant not available: %v\", err)\n\t}\n\n\trequire.NotNil(t, ast.HookScript, \"HookScript should be loaded\")\n\n\t// Create context\n\tctx := newSandboxE2EContext(\"sandbox-ctx-access\", \"tests.sandbox.hooks\")\n\n\t// Test Create Hook - it should have access to ctx.sandbox\n\tmessages := []context.Message{\n\t\t{Role: context.RoleUser, Content: \"test sandbox context access\"},\n\t}\n\n\t// Execute Create hook directly\n\t// This tests that the hook runs without error (sandbox operations tested within)\n\topts := &context.Options{}\n\tresponse, _, err := ast.HookScript.Create(ctx, messages, opts)\n\n\t// The hook might fail if sandbox isn't initialized yet (that's done in Stream)\n\t// But we can at least verify the hook exists and can be called\n\tif err != nil {\n\t\t// If the error is about sandbox not being available, that's expected\n\t\t// because we haven't initialized the sandbox yet\n\t\tif strings.Contains(err.Error(), \"sandbox\") {\n\t\t\tt.Logf(\"Expected error: sandbox not available in direct hook call: %v\", err)\n\t\t} else {\n\t\t\tt.Fatalf(\"Unexpected error: %v\", err)\n\t\t}\n\t}\n\n\t// Response might be nil, that's okay\n\tt.Logf(\"Create hook response: %v\", response)\n\tt.Log(\"✓ Sandbox context access test passed\")\n}\n\n// TestSandboxMCPToolCall tests that Claude actually calls MCP tools via IPC\n// This test specifically asks Claude to use the echo tool and verifies the result\nfunc TestSandboxMCPToolCall(t *testing.T) {\n\tif testing.Short() {\n\t\tt.Skip(\"Skipping sandbox MCP tool call test in short mode\")\n\t}\n\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\t// Load the full sandbox assistant (has MCP echo tool)\n\tast, err := assistant.Get(\"tests.sandbox.full\")\n\tif err != nil {\n\t\tt.Skipf(\"Skipping test: full sandbox assistant not available: %v\", err)\n\t}\n\n\t// Verify MCP is configured with echo tools\n\trequire.NotNil(t, ast.MCP, \"MCP should be configured\")\n\trequire.NotEmpty(t, ast.MCP.Servers, \"MCP servers should be configured\")\n\tt.Logf(\"✓ MCP configured with server: %s, tools: %v\",\n\t\tast.MCP.Servers[0].ServerID, ast.MCP.Servers[0].Tools)\n\n\t// Create context\n\tctx := newSandboxE2EContext(\"sandbox-mcp-tool\", \"tests.sandbox.full\")\n\n\t// Explicit prompt to use echo tool\n\t// This tells Claude to use the MCP tool specifically\n\tmessages := []context.Message{\n\t\t{\n\t\t\tRole: context.RoleUser,\n\t\t\tContent: `Please use the 'ping' MCP tool to send a ping with message \"MCP_TEST_SUCCESS\". \nJust call the tool and show me the result. Do not explain, just use the tool.`,\n\t\t},\n\t}\n\n\t// Collect all response content\n\tvar responseContent strings.Builder\n\n\t// Execute stream\n\tresponse, err := ast.Stream(ctx, messages)\n\tif err != nil {\n\t\terrStr := err.Error()\n\t\tif strings.Contains(errStr, \"Docker\") ||\n\t\t\tstrings.Contains(errStr, \"sandbox\") ||\n\t\t\tstrings.Contains(errStr, \"container\") ||\n\t\t\tstrings.Contains(errStr, \"image\") {\n\t\t\tt.Skipf(\"Skipping test: Docker/sandbox not available: %v\", err)\n\t\t}\n\t\tt.Fatalf(\"Stream failed: %v\", err)\n\t}\n\n\trequire.NotNil(t, response, \"Response should not be nil\")\n\n\t// Get the response content\n\tfullResponse := \"\"\n\tif response.Completion != nil && response.Completion.Content != nil {\n\t\tif contentStr, ok := response.Completion.Content.(string); ok {\n\t\t\tfullResponse = contentStr\n\t\t\tresponseContent.WriteString(contentStr)\n\t\t}\n\t}\n\n\tt.Logf(\"Claude response: %s\", fullResponse)\n\n\t// Check if Claude acknowledged using the tool or returned tool results\n\t// The response should contain either:\n\t// 1. Evidence of tool call (tool_use block in response)\n\t// 2. The ping result \"pong\" or \"MCP_TEST_SUCCESS\"\n\t// 3. Some indication that it attempted to use the MCP tool\n\n\thasToolEvidence := strings.Contains(fullResponse, \"pong\") ||\n\t\tstrings.Contains(fullResponse, \"MCP_TEST_SUCCESS\") ||\n\t\tstrings.Contains(fullResponse, \"ping\") ||\n\t\tstrings.Contains(fullResponse, \"tool\")\n\n\tif hasToolEvidence {\n\t\tt.Log(\"✓ Claude appears to have used the MCP tool\")\n\t} else {\n\t\tt.Logf(\"⚠ Claude response does not clearly show MCP tool usage\")\n\t\tt.Logf(\"Response: %s\", fullResponse)\n\t}\n\n\t// At minimum, verify we got a response\n\tif fullResponse == \"\" {\n\t\tt.Log(\"⚠ Response content is empty\")\n\t}\n\tt.Log(\"✓ Sandbox MCP tool call test completed\")\n}\n\n// TestSandboxMCPEchoTool tests the echo MCP tool specifically\n// This test uses a more explicit prompt to force tool usage\nfunc TestSandboxMCPEchoTool(t *testing.T) {\n\tif testing.Short() {\n\t\tt.Skip(\"Skipping sandbox MCP echo test in short mode\")\n\t}\n\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\t// Load the full sandbox assistant\n\tast, err := assistant.Get(\"tests.sandbox.full\")\n\tif err != nil {\n\t\tt.Skipf(\"Skipping test: full sandbox assistant not available: %v\", err)\n\t}\n\n\t// Create context\n\tctx := newSandboxE2EContext(\"sandbox-mcp-echo\", \"tests.sandbox.full\")\n\n\t// Very explicit prompt for echo tool\n\tmessages := []context.Message{\n\t\t{\n\t\t\tRole: context.RoleUser,\n\t\t\tContent: `Call the 'echo' MCP tool with message \"ECHO_VERIFICATION_12345\" and uppercase=true. \nShow me the exact response from the tool.`,\n\t\t},\n\t}\n\n\tresponse, err := ast.Stream(ctx, messages)\n\tif err != nil {\n\t\terrStr := err.Error()\n\t\tif strings.Contains(errStr, \"Docker\") ||\n\t\t\tstrings.Contains(errStr, \"sandbox\") ||\n\t\t\tstrings.Contains(errStr, \"container\") ||\n\t\t\tstrings.Contains(errStr, \"image\") {\n\t\t\tt.Skipf(\"Skipping test: Docker/sandbox not available: %v\", err)\n\t\t}\n\t\tt.Fatalf(\"Stream failed: %v\", err)\n\t}\n\n\trequire.NotNil(t, response)\n\n\tfullResponse := \"\"\n\tif response.Completion != nil && response.Completion.Content != nil {\n\t\tif contentStr, ok := response.Completion.Content.(string); ok {\n\t\t\tfullResponse = contentStr\n\t\t}\n\t}\n\tt.Logf(\"Claude response for echo tool: %s\", fullResponse)\n\n\t// The echo tool with uppercase=true should return \"ECHO_VERIFICATION_12345\"\n\t// Check if this appears in the response\n\tif strings.Contains(fullResponse, \"ECHO_VERIFICATION_12345\") {\n\t\tt.Log(\"✓ MCP echo tool executed successfully - found verification string in response\")\n\t} else if strings.Contains(fullResponse, \"echo\") || strings.Contains(fullResponse, \"ECHO\") {\n\t\tt.Log(\"✓ MCP echo tool appears to have been used (found 'echo' in response)\")\n\t} else {\n\t\tt.Logf(\"⚠ Could not verify echo tool execution. Response: %s\", fullResponse)\n\t}\n\n\tt.Log(\"✓ Sandbox MCP echo tool test completed\")\n}\n\n// TestSandboxLoadConfiguration verifies that sandbox assistants load correctly\nfunc TestSandboxLoadConfiguration(t *testing.T) {\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\ttestCases := []struct {\n\t\tname string\n\t\tassistantID string\n\t\texpectSandbox bool\n\t\texpectMCP bool\n\t\texpectHooks bool\n\t}{\n\t\t{\n\t\t\tname: \"BasicSandbox\",\n\t\t\tassistantID: \"tests.sandbox.basic\",\n\t\t\texpectSandbox: true,\n\t\t\texpectMCP: false,\n\t\t\texpectHooks: false,\n\t\t},\n\t\t{\n\t\t\tname: \"HooksSandbox\",\n\t\t\tassistantID: \"tests.sandbox.hooks\",\n\t\t\texpectSandbox: true,\n\t\t\texpectMCP: false,\n\t\t\texpectHooks: true,\n\t\t},\n\t\t{\n\t\t\tname: \"FullSandbox\",\n\t\t\tassistantID: \"tests.sandbox.full\",\n\t\t\texpectSandbox: true,\n\t\t\texpectMCP: true,\n\t\t\texpectHooks: true,\n\t\t},\n\t}\n\n\tfor _, tc := range testCases {\n\t\tt.Run(tc.name, func(t *testing.T) {\n\t\t\tast, err := assistant.Get(tc.assistantID)\n\t\t\tif err != nil {\n\t\t\t\tt.Skipf(\"Skipping: assistant %s not available: %v\", tc.assistantID, err)\n\t\t\t}\n\n\t\t\t// Check sandbox\n\t\t\tif tc.expectSandbox {\n\t\t\t\trequire.NotNil(t, ast.Sandbox, \"Expected sandbox to be configured\")\n\t\t\t\tassert.Equal(t, \"claude\", ast.Sandbox.Command)\n\t\t\t\tt.Logf(\"✓ %s: Sandbox configured with command=%s\", tc.name, ast.Sandbox.Command)\n\t\t\t}\n\n\t\t\t// Check MCP\n\t\t\tif tc.expectMCP {\n\t\t\t\trequire.NotNil(t, ast.MCP, \"Expected MCP to be configured\")\n\t\t\t\tassert.True(t, len(ast.MCP.Servers) > 0, \"Expected at least one MCP server\")\n\t\t\t\tt.Logf(\"✓ %s: MCP configured with %d servers\", tc.name, len(ast.MCP.Servers))\n\t\t\t}\n\n\t\t\t// Check hooks\n\t\t\tif tc.expectHooks {\n\t\t\t\trequire.NotNil(t, ast.HookScript, \"Expected hooks to be loaded\")\n\t\t\t\tt.Logf(\"✓ %s: Hooks loaded\", tc.name)\n\t\t\t}\n\t\t})\n\t}\n}\n" }, { "path": "agent/assistant/sandbox_integration_test.go", "content": "package assistant_test\n\nimport (\n\t\"testing\"\n\n\t\"github.com/stretchr/testify/assert\"\n\t\"github.com/stretchr/testify/require\"\n\t\"github.com/yaoapp/yao/agent\"\n\t\"github.com/yaoapp/yao/agent/assistant\"\n\tagentContext \"github.com/yaoapp/yao/agent/context\"\n\tagentsandbox \"github.com/yaoapp/yao/agent/sandbox\"\n\t\"github.com/yaoapp/yao/agent/sandbox/claude\"\n\t\"github.com/yaoapp/yao/config\"\n\t\"github.com/yaoapp/yao/test\"\n)\n\n// TestSandboxOptionsBuilding tests that sandbox options are correctly built from assistant config\nfunc TestSandboxOptionsBuilding(t *testing.T) {\n\ttest.Prepare(t, config.Conf)\n\tdefer test.Clean()\n\n\t// Load agent to ensure connectors are available\n\terr := agent.Load(config.Conf)\n\trequire.NoError(t, err, \"agent.Load should succeed\")\n\n\t// Load the full test assistant\n\tast, err := assistant.LoadPath(\"/assistants/tests/sandbox/full\")\n\trequire.NoError(t, err)\n\trequire.NotNil(t, ast)\n\n\t// Verify sandbox is configured\n\trequire.NotNil(t, ast.Sandbox)\n\tassert.Equal(t, \"claude\", ast.Sandbox.Command)\n\tassert.Equal(t, \"5m\", ast.Sandbox.Timeout)\n\n\t// Verify arguments are set\n\trequire.NotNil(t, ast.Sandbox.Arguments)\n\tassert.Equal(t, float64(10), ast.Sandbox.Arguments[\"max_turns\"])\n\tassert.Equal(t, \"acceptEdits\", ast.Sandbox.Arguments[\"permission_mode\"])\n\n\t// Verify MCP configuration\n\trequire.NotNil(t, ast.MCP)\n\tassert.Len(t, ast.MCP.Servers, 1)\n\tassert.Equal(t, \"echo\", ast.MCP.Servers[0].ServerID)\n\n\tt.Logf(\"Sandbox config: command=%s, timeout=%s\", ast.Sandbox.Command, ast.Sandbox.Timeout)\n\tt.Logf(\"Sandbox arguments: %v\", ast.Sandbox.Arguments)\n\tt.Logf(\"MCP servers: %v\", ast.MCP.Servers)\n}\n\n// TestClaudeCommandBuilding tests that Claude CLI commands are correctly built\nfunc TestClaudeCommandBuilding(t *testing.T) {\n\ttest.Prepare(t, config.Conf)\n\tdefer test.Clean()\n\n\t// Create test messages\n\tmessages := []agentContext.Message{\n\t\t{Role: \"system\", Content: \"You are a helpful coding assistant.\"},\n\t\t{Role: \"user\", Content: \"Hello, how are you?\"},\n\t}\n\n\t// Create options similar to what buildSandboxOptions would produce\n\topts := &claude.Options{\n\t\tCommand: \"claude\",\n\t\tUserID: \"test-user\",\n\t\tChatID: \"test-chat\",\n\t\tConnectorHost: \"https://ark.cn-beijing.volces.com/api/v3\",\n\t\tConnectorKey: \"test-api-key\",\n\t\tModel: \"ep-xxxxx\",\n\t\tArguments: map[string]interface{}{\n\t\t\t\"max_turns\": 10,\n\t\t\t\"permission_mode\": \"acceptEdits\",\n\t\t},\n\t}\n\n\t// Build the command\n\tcmd, env, err := claude.BuildCommand(messages, opts)\n\trequire.NoError(t, err)\n\n\t// Verify command structure\n\t// Command is now: [\"bash\", \"-c\", \"cat << 'INPUTEOF' | claude -p ... INPUTEOF\"]\n\tassert.NotEmpty(t, cmd)\n\tassert.Equal(t, \"bash\", cmd[0], \"Command should start with bash\")\n\tassert.Equal(t, \"-c\", cmd[1], \"Second arg should be -c\")\n\tassert.Contains(t, cmd[2], \"claude -p\", \"Bash command should contain claude -p\")\n\tassert.Contains(t, cmd[2], \"--permission-mode\", \"Should include permission mode\")\n\tassert.Contains(t, cmd[2], \"--input-format\", \"Should include input-format flag\")\n\tassert.Contains(t, cmd[2], \"--output-format\", \"Should include output-format flag\")\n\tassert.Contains(t, cmd[2], \"--verbose\", \"Should include verbose flag\")\n\tassert.Contains(t, cmd[2], \"stream-json\", \"Should use stream-json format\")\n\tassert.Contains(t, cmd[2], \"INPUTEOF\", \"Should use heredoc for input\")\n\tt.Logf(\"Built command: %v\", cmd)\n\n\t// Verify environment variables (claude-proxy mode)\n\tassert.NotEmpty(t, env)\n\tassert.Equal(t, \"http://127.0.0.1:3456\", env[\"ANTHROPIC_BASE_URL\"], \"Should set proxy base URL\")\n\tassert.Equal(t, \"dummy\", env[\"ANTHROPIC_API_KEY\"], \"Should set dummy API key for proxy\")\n\n\t// max_turns is passed via CLI flag\n\t// system prompt is written to file via heredoc, then referenced via --append-system-prompt-file\n\tassert.Contains(t, cmd[2], \"--max-turns\", \"Should include max-turns flag\")\n\tassert.Contains(t, cmd[2], \"cat << 'PROMPTEOF' > /tmp/.system-prompt.txt\", \"Should use heredoc for system prompt\")\n\tassert.Contains(t, cmd[2], \"--append-system-prompt-file\", \"Should include append-system-prompt-file flag\")\n\tassert.Contains(t, cmd[2], \"You are a helpful coding assistant\", \"Command should contain system prompt\")\n\tt.Logf(\"Built environment: %v\", env)\n}\n\n// TestClaudeProxyConfigBuilding tests that claude-proxy config is correctly built\nfunc TestClaudeProxyConfigBuilding(t *testing.T) {\n\ttest.Prepare(t, config.Conf)\n\tdefer test.Clean()\n\n\topts := &claude.Options{\n\t\tConnectorHost: \"https://ark.cn-beijing.volces.com/api/v3\",\n\t\tConnectorKey: \"test-api-key\",\n\t\tModel: \"ep-xxxxx\",\n\t}\n\n\tconfigJSON, err := claude.BuildProxyConfig(opts)\n\trequire.NoError(t, err)\n\trequire.NotEmpty(t, configJSON)\n\n\tt.Logf(\"Proxy config: %s\", string(configJSON))\n\n\t// Verify the JSON contains expected fields for claude-proxy\n\tassert.Contains(t, string(configJSON), \"backend\")\n\tassert.Contains(t, string(configJSON), \"api_key\")\n\tassert.Contains(t, string(configJSON), \"model\")\n\tassert.Contains(t, string(configJSON), \"test-api-key\")\n\tassert.Contains(t, string(configJSON), \"ep-xxxxx\")\n\t// Backend URL should end with /chat/completions\n\tassert.Contains(t, string(configJSON), \"/chat/completions\")\n}\n\n// TestDefaultImageSelection tests that default images are correctly selected\nfunc TestDefaultImageSelection(t *testing.T) {\n\ttests := []struct {\n\t\tcommand string\n\t\texpectedImage string\n\t}{\n\t\t{\"claude\", \"yaoapp/sandbox-claude:latest\"},\n\t\t{\"cursor\", \"yaoapp/sandbox-cursor:latest\"},\n\t\t{\"unknown\", \"\"},\n\t}\n\n\tfor _, tt := range tests {\n\t\tt.Run(tt.command, func(t *testing.T) {\n\t\t\timage := agentsandbox.DefaultImage(tt.command)\n\t\t\tassert.Equal(t, tt.expectedImage, image)\n\t\t})\n\t}\n}\n\n// TestSandboxCommandValidation tests that command validation works correctly\nfunc TestSandboxCommandValidation(t *testing.T) {\n\ttests := []struct {\n\t\tcommand string\n\t\tvalid bool\n\t}{\n\t\t{\"claude\", true},\n\t\t{\"cursor\", true},\n\t\t{\"invalid\", false},\n\t\t{\"\", false},\n\t}\n\n\tfor _, tt := range tests {\n\t\tt.Run(tt.command, func(t *testing.T) {\n\t\t\tresult := agentsandbox.IsValidCommand(tt.command)\n\t\t\tassert.Equal(t, tt.valid, result)\n\t\t})\n\t}\n}\n\n// TestHasSandboxMethod tests the HasSandbox method on Assistant\nfunc TestHasSandboxMethod(t *testing.T) {\n\ttest.Prepare(t, config.Conf)\n\tdefer test.Clean()\n\n\t// Test assistant with sandbox\n\tastWithSandbox, err := assistant.LoadPath(\"/assistants/tests/sandbox/basic\")\n\trequire.NoError(t, err)\n\tassert.True(t, astWithSandbox.HasSandbox(), \"Assistant with sandbox config should return true\")\n\n\t// Test assistant without sandbox\n\tastWithoutSandbox, err := assistant.LoadPath(\"/assistants/tests/simple-greeting\")\n\trequire.NoError(t, err)\n\tassert.False(t, astWithoutSandbox.HasSandbox(), \"Assistant without sandbox config should return false\")\n}\n" }, { "path": "agent/assistant/sandbox_test.go", "content": "package assistant_test\n\nimport (\n\t\"context\"\n\t\"encoding/json\"\n\t\"os\"\n\t\"path/filepath\"\n\t\"testing\"\n\n\t\"github.com/stretchr/testify/assert\"\n\t\"github.com/stretchr/testify/require\"\n\t\"github.com/yaoapp/yao/agent\"\n\t\"github.com/yaoapp/yao/agent/assistant\"\n\tagentContext \"github.com/yaoapp/yao/agent/context\"\n\t\"github.com/yaoapp/yao/config\"\n\t\"github.com/yaoapp/yao/test\"\n)\n\n// TestLoadSandboxBasicAssistant tests loading the basic sandbox test assistant\nfunc TestLoadSandboxBasicAssistant(t *testing.T) {\n\ttest.Prepare(t, config.Conf)\n\tdefer test.Clean()\n\n\tast, err := assistant.LoadPath(\"/assistants/tests/sandbox/basic\")\n\trequire.NoError(t, err)\n\trequire.NotNil(t, ast)\n\n\t// Verify basic fields\n\tassert.Equal(t, \"tests.sandbox.basic\", ast.ID)\n\tassert.Equal(t, \"Sandbox Basic Test\", ast.Name)\n\tassert.Equal(t, \"deepseek.v3\", ast.Connector)\n\n\t// Verify sandbox configuration\n\trequire.NotNil(t, ast.Sandbox, \"Sandbox should be configured\")\n\tassert.Equal(t, \"claude\", ast.Sandbox.Command)\n\tassert.Equal(t, \"5m\", ast.Sandbox.Timeout)\n\n\t// Verify HasSandbox returns true\n\tassert.True(t, ast.HasSandbox(), \"HasSandbox should return true\")\n}\n\n// TestLoadSandboxHooksAssistant tests loading the hooks sandbox test assistant\nfunc TestLoadSandboxHooksAssistant(t *testing.T) {\n\ttest.Prepare(t, config.Conf)\n\tdefer test.Clean()\n\n\tast, err := assistant.LoadPath(\"/assistants/tests/sandbox/hooks\")\n\trequire.NoError(t, err)\n\trequire.NotNil(t, ast)\n\n\t// Verify basic fields\n\tassert.Equal(t, \"tests.sandbox.hooks\", ast.ID)\n\tassert.Equal(t, \"Sandbox Hooks Test\", ast.Name)\n\tassert.Equal(t, \"deepseek.v3\", ast.Connector)\n\n\t// Verify sandbox configuration\n\trequire.NotNil(t, ast.Sandbox, \"Sandbox should be configured\")\n\tassert.Equal(t, \"claude\", ast.Sandbox.Command)\n\n\t// Verify hooks are loaded\n\tassert.NotNil(t, ast.HookScript, \"HookScript should be loaded\")\n}\n\n// TestLoadSandboxFullAssistant tests loading the full sandbox test assistant with MCPs and Skills\nfunc TestLoadSandboxFullAssistant(t *testing.T) {\n\ttest.Prepare(t, config.Conf)\n\tdefer test.Clean()\n\n\t// Load agent to ensure MCPs are available\n\terr := agent.Load(config.Conf)\n\trequire.NoError(t, err, \"agent.Load should succeed\")\n\n\tast, err := assistant.LoadPath(\"/assistants/tests/sandbox/full\")\n\trequire.NoError(t, err)\n\trequire.NotNil(t, ast)\n\n\t// Verify basic fields\n\tassert.Equal(t, \"tests.sandbox.full\", ast.ID)\n\tassert.Equal(t, \"Sandbox Full Test\", ast.Name)\n\tassert.Equal(t, \"deepseek.v3\", ast.Connector)\n\n\t// Verify sandbox configuration\n\trequire.NotNil(t, ast.Sandbox, \"Sandbox should be configured\")\n\tassert.Equal(t, \"claude\", ast.Sandbox.Command)\n\tassert.Equal(t, \"5m\", ast.Sandbox.Timeout)\n\n\t// Verify sandbox arguments (command-specific options)\n\trequire.NotNil(t, ast.Sandbox.Arguments, \"Sandbox arguments should be configured\")\n\tassert.Equal(t, float64(10), ast.Sandbox.Arguments[\"max_turns\"])\n\tassert.Equal(t, \"acceptEdits\", ast.Sandbox.Arguments[\"permission_mode\"])\n\n\t// Verify MCP configuration\n\trequire.NotNil(t, ast.MCP, \"MCP should be configured\")\n\trequire.NotNil(t, ast.MCP.Servers, \"MCP.Servers should be configured\")\n\tassert.Len(t, ast.MCP.Servers, 1, \"Should have 1 MCP server configured\")\n\tassert.Equal(t, \"echo\", ast.MCP.Servers[0].ServerID, \"MCP server ID should be 'echo'\")\n\tassert.Contains(t, ast.MCP.Servers[0].Tools, \"ping\", \"MCP tools should contain 'ping'\")\n\tassert.Contains(t, ast.MCP.Servers[0].Tools, \"echo\", \"MCP tools should contain 'echo'\")\n\n\t// Verify hooks are loaded\n\tassert.NotNil(t, ast.HookScript, \"HookScript should be loaded\")\n}\n\n// TestSandboxConfigValidation tests sandbox configuration validation\nfunc TestSandboxConfigValidation(t *testing.T) {\n\ttest.Prepare(t, config.Conf)\n\tdefer test.Clean()\n\n\ttests := []struct {\n\t\tname string\n\t\tpath string\n\t\thasError bool\n\t}{\n\t\t{\n\t\t\tname: \"Basic sandbox config\",\n\t\t\tpath: \"/assistants/tests/sandbox/basic\",\n\t\t\thasError: false,\n\t\t},\n\t\t{\n\t\t\tname: \"Hooks sandbox config\",\n\t\t\tpath: \"/assistants/tests/sandbox/hooks\",\n\t\t\thasError: false,\n\t\t},\n\t\t{\n\t\t\tname: \"Full sandbox config with MCPs\",\n\t\t\tpath: \"/assistants/tests/sandbox/full\",\n\t\t\thasError: false,\n\t\t},\n\t}\n\n\tfor _, tt := range tests {\n\t\tt.Run(tt.name, func(t *testing.T) {\n\t\t\tast, err := assistant.LoadPath(tt.path)\n\t\t\tif tt.hasError {\n\t\t\t\tassert.Error(t, err)\n\t\t\t\treturn\n\t\t\t}\n\t\t\trequire.NoError(t, err)\n\t\t\trequire.NotNil(t, ast)\n\t\t\trequire.NotNil(t, ast.Sandbox)\n\t\t\tassert.NotEmpty(t, ast.Sandbox.Command)\n\t\t})\n\t}\n}\n\n// TestSkillsDirectoryResolution tests that skills directory exists and has correct structure\n// Note: Skills are auto-discovered from skills/ directory, not stored in AssistantModel\nfunc TestSkillsDirectoryResolution(t *testing.T) {\n\ttest.Prepare(t, config.Conf)\n\tdefer test.Clean()\n\n\tast, err := assistant.LoadPath(\"/assistants/tests/sandbox/full\")\n\trequire.NoError(t, err)\n\trequire.NotNil(t, ast)\n\n\t// Get app root from environment\n\tappRoot := os.Getenv(\"YAO_ROOT\")\n\trequire.NotEmpty(t, appRoot, \"YAO_ROOT should be set\")\n\n\t// Verify assistant path is set\n\tassert.NotEmpty(t, ast.Path, \"Assistant path should be set\")\n\n\t// Build expected skills directory path\n\t// ast.Path is like \"/assistants/tests/sandbox/full\"\n\texpectedSkillsDir := filepath.Join(appRoot, ast.Path, \"skills\")\n\n\t// Verify skills directory exists\n\tinfo, err := os.Stat(expectedSkillsDir)\n\trequire.NoError(t, err, \"Skills directory should exist: %s\", expectedSkillsDir)\n\tassert.True(t, info.IsDir(), \"Skills path should be a directory\")\n\n\t// Verify skills directory structure\n\tentries, err := os.ReadDir(expectedSkillsDir)\n\trequire.NoError(t, err, \"Should be able to read skills directory\")\n\n\t// Find echo-test skill\n\tvar foundEchoTest bool\n\tfor _, entry := range entries {\n\t\tif entry.IsDir() && entry.Name() == \"echo-test\" {\n\t\t\tfoundEchoTest = true\n\n\t\t\t// Verify SKILL.md exists (required)\n\t\t\tskillMdPath := filepath.Join(expectedSkillsDir, \"echo-test\", \"SKILL.md\")\n\t\t\t_, err := os.Stat(skillMdPath)\n\t\t\tassert.NoError(t, err, \"SKILL.md should exist\")\n\n\t\t\t// Verify scripts directory exists (optional but we created it)\n\t\t\tscriptsDir := filepath.Join(expectedSkillsDir, \"echo-test\", \"scripts\")\n\t\t\t_, err = os.Stat(scriptsDir)\n\t\t\tassert.NoError(t, err, \"scripts directory should exist\")\n\n\t\t\t// Verify echo.sh exists\n\t\t\techoShPath := filepath.Join(scriptsDir, \"echo.sh\")\n\t\t\t_, err = os.Stat(echoShPath)\n\t\t\tassert.NoError(t, err, \"echo.sh should exist\")\n\n\t\t\tbreak\n\t\t}\n\t}\n\tassert.True(t, foundEchoTest, \"echo-test skill should exist in skills directory\")\n}\n\n// TestMCPConfiguration tests that MCP is correctly loaded for sandbox assistant\nfunc TestMCPConfiguration(t *testing.T) {\n\ttest.Prepare(t, config.Conf)\n\tdefer test.Clean()\n\n\t// Load agent to ensure MCPs are available\n\terr := agent.Load(config.Conf)\n\trequire.NoError(t, err, \"agent.Load should succeed\")\n\n\tast, err := assistant.LoadPath(\"/assistants/tests/sandbox/full\")\n\trequire.NoError(t, err)\n\trequire.NotNil(t, ast)\n\n\t// Verify MCP configuration structure\n\trequire.NotNil(t, ast.MCP, \"MCP should not be nil\")\n\trequire.NotNil(t, ast.MCP.Servers, \"MCP.Servers should not be nil\")\n\tassert.Len(t, ast.MCP.Servers, 1, \"Should have 1 MCP server configured\")\n\n\t// Verify echo server configuration\n\techoServer := ast.MCP.Servers[0]\n\tassert.Equal(t, \"echo\", echoServer.ServerID, \"Server ID should be 'echo'\")\n\tassert.Len(t, echoServer.Tools, 3, \"Should have 3 tools configured\")\n\tassert.Contains(t, echoServer.Tools, \"ping\")\n\tassert.Contains(t, echoServer.Tools, \"echo\")\n\tassert.Contains(t, echoServer.Tools, \"status\")\n}\n\n// TestBuildMCPConfigForSandbox tests that MCP configuration is correctly built for sandbox\nfunc TestBuildMCPConfigForSandbox(t *testing.T) {\n\ttest.Prepare(t, config.Conf)\n\tdefer test.Clean()\n\n\t// Load agent to ensure MCPs are available\n\terr := agent.Load(config.Conf)\n\trequire.NoError(t, err, \"agent.Load should succeed\")\n\n\tast, err := assistant.LoadPath(\"/assistants/tests/sandbox/full\")\n\trequire.NoError(t, err)\n\trequire.NotNil(t, ast)\n\trequire.NotNil(t, ast.MCP, \"MCP configuration should exist\")\n\n\t// Create a mock context for the test\n\tctx := agentContext.New(context.Background(), nil, \"test-mcp-config-build\")\n\n\t// Call BuildMCPConfigForSandbox and verify the result\n\tmcpConfig, err := ast.BuildMCPConfigForSandbox(ctx)\n\trequire.NoError(t, err, \"BuildMCPConfigForSandbox should not error\")\n\trequire.NotEmpty(t, mcpConfig, \"MCP config should not be empty\")\n\n\tt.Logf(\"MCP config JSON: %s\", string(mcpConfig))\n\n\t// Parse and verify the JSON structure\n\tvar config map[string]interface{}\n\terr = json.Unmarshal(mcpConfig, &config)\n\trequire.NoError(t, err, \"MCP config should be valid JSON\")\n\n\t// Verify mcpServers key exists\n\tmcpServers, ok := config[\"mcpServers\"].(map[string]interface{})\n\trequire.True(t, ok, \"mcpServers should be a map\")\n\trequire.NotEmpty(t, mcpServers, \"mcpServers should not be empty\")\n\n\t// Verify \"yao\" server exists (single server using yao-bridge for IPC)\n\tyaoServer, ok := mcpServers[\"yao\"].(map[string]interface{})\n\trequire.True(t, ok, \"yao server should exist in mcpServers\")\n\n\t// Verify server structure - uses yao-bridge to connect to IPC socket\n\tassert.Equal(t, \"yao-bridge\", yaoServer[\"command\"], \"command should be yao-bridge\")\n\n\targs, ok := yaoServer[\"args\"].([]interface{})\n\trequire.True(t, ok, \"args should be an array\")\n\trequire.Len(t, args, 1, \"args should have 1 element\")\n\tassert.Equal(t, \"/tmp/yao.sock\", args[0], \"first arg should be IPC socket path\")\n\n\tt.Logf(\"✓ MCP config verified: uses yao-bridge with IPC socket /tmp/yao.sock\")\n}\n\n// TestSandboxMCPAndSkillsOptions tests that sandbox options include MCP and Skills\nfunc TestSandboxMCPAndSkillsOptions(t *testing.T) {\n\ttest.Prepare(t, config.Conf)\n\tdefer test.Clean()\n\n\t// Load agent to ensure MCPs are available\n\terr := agent.Load(config.Conf)\n\trequire.NoError(t, err, \"agent.Load should succeed\")\n\n\tast, err := assistant.LoadPath(\"/assistants/tests/sandbox/full\")\n\trequire.NoError(t, err)\n\trequire.NotNil(t, ast)\n\n\t// Verify sandbox configuration is present\n\trequire.NotNil(t, ast.Sandbox, \"Sandbox should be configured\")\n\tassert.Equal(t, \"claude\", ast.Sandbox.Command)\n\n\t// Verify MCP is configured (will be passed to sandbox)\n\trequire.NotNil(t, ast.MCP, \"MCP should be configured\")\n\tassert.Len(t, ast.MCP.Servers, 1, \"Should have 1 MCP server\")\n\n\t// Verify skills directory exists\n\tappRoot := os.Getenv(\"YAO_ROOT\")\n\trequire.NotEmpty(t, appRoot, \"YAO_ROOT should be set\")\n\n\tskillsDir := filepath.Join(appRoot, ast.Path, \"skills\")\n\tinfo, err := os.Stat(skillsDir)\n\trequire.NoError(t, err, \"Skills directory should exist\")\n\tassert.True(t, info.IsDir(), \"Skills should be a directory\")\n\n\t// Verify echo-test skill exists\n\techoTestDir := filepath.Join(skillsDir, \"echo-test\")\n\tinfo, err = os.Stat(echoTestDir)\n\trequire.NoError(t, err, \"echo-test skill should exist\")\n\tassert.True(t, info.IsDir(), \"echo-test should be a directory\")\n\n\t// Verify SKILL.md exists\n\tskillMd := filepath.Join(echoTestDir, \"SKILL.md\")\n\t_, err = os.Stat(skillMd)\n\trequire.NoError(t, err, \"SKILL.md should exist\")\n}\n" }, { "path": "agent/assistant/sandbox_v2.go", "content": "package assistant\n\nimport (\n\t\"fmt\"\n\t\"log\"\n\t\"os\"\n\t\"path/filepath\"\n\n\t\"github.com/yaoapp/yao/agent/context\"\n\t\"github.com/yaoapp/yao/agent/i18n\"\n\t\"github.com/yaoapp/yao/agent/output/message\"\n\tsandboxv2 \"github.com/yaoapp/yao/agent/sandbox/v2\"\n\tsandboxTypes \"github.com/yaoapp/yao/agent/sandbox/v2/types\"\n\t\"github.com/yaoapp/yao/config\"\n\tinfraV2 \"github.com/yaoapp/yao/sandbox/v2\"\n\ttraceTypes \"github.com/yaoapp/yao/trace/types\"\n\t\"github.com/yaoapp/yao/workspace\"\n)\n\n// HasSandboxV2 returns true if the assistant has a V2 sandbox configuration.\nfunc (ast *Assistant) HasSandboxV2() bool {\n\treturn ast.SandboxV2 != nil\n}\n\n// initSandboxV2 initializes the V2 sandbox: obtains a Computer, gets a Runner,\n// runs Prepare, and returns the runner, computer, cleanup closure, loading\n// message ID, and any error.\nfunc (ast *Assistant) initSandboxV2(ctx *context.Context, opts *context.Options) (\n\tsandboxTypes.Runner, infraV2.Computer, func(), string, error,\n) {\n\tcfg := ast.SandboxV2\n\tmanager := infraV2.M()\n\n\tloadingMsg := &message.Message{\n\t\tType: message.TypeLoading,\n\t\tProps: map[string]any{\n\t\t\t\"message\": i18n.T(ctx.Locale, \"sandbox.preparing\"),\n\t\t},\n\t}\n\tloadingMsgID, _ := ctx.SendStream(loadingMsg)\n\n\tstdCtx := ctx.Context\n\n\t// 1. Resolve connector (before Computer so proxy env vars can be injected).\n\tconn, _, err := ast.GetConnector(ctx, opts)\n\tif err != nil && cfg.Runner.Name != \"yao\" {\n\t\tcloseLoadingV2(ctx, loadingMsgID, \"sandbox.failed\")\n\t\treturn nil, nil, nil, \"\", fmt.Errorf(\"get connector: %w\", err)\n\t}\n\n\t// 2. Build human-readable DisplayName from real Agent name + Workspace name.\n\tcfg.DisplayName = buildBoxDisplayName(ctx, ast.ID, ast.Name)\n\n\t// 2.5. Image existence check + pull (for box mode).\n\tif cfg.Computer.Image != \"\" && manager != nil {\n\t\tnodeID, kind, _ := sandboxv2.ResolveNodeID(ctx, cfg, manager)\n\t\tif kind == \"box\" && nodeID != \"\" {\n\t\t\tupdateLoadingV2(ctx, loadingMsgID, \"sandbox.starting\")\n\t\t\texists, existsErr := manager.ImageExists(stdCtx, nodeID, cfg.Computer.Image)\n\t\t\tif existsErr != nil {\n\t\t\t\tlog.Printf(\"[sandbox/v2] image exists check failed on node %s: %v\", nodeID, existsErr)\n\t\t\t}\n\t\t\tif existsErr == nil && !exists {\n\t\t\t\tupdateLoadingV2(ctx, loadingMsgID, \"sandbox.pulling_image\")\n\t\t\t\tch, pullErr := manager.PullImage(stdCtx, nodeID, cfg.Computer.Image, infraV2.ImagePullOptions{})\n\t\t\t\tif pullErr != nil {\n\t\t\t\t\tlog.Printf(\"[sandbox/v2] image pull failed on node %s: %v (will retry in Create)\", nodeID, pullErr)\n\t\t\t\t} else if ch != nil {\n\t\t\t\t\tfor p := range ch {\n\t\t\t\t\t\tif p.Error != \"\" {\n\t\t\t\t\t\t\tlog.Printf(\"[sandbox/v2] image pull progress error: %s\", p.Error)\n\t\t\t\t\t\t\tbreak\n\t\t\t\t\t\t}\n\t\t\t\t\t}\n\t\t\t\t}\n\t\t\t}\n\t\t}\n\t}\n\n\t// 3. Obtain Computer (passes connector for OPENAI_PROXY_* env injection).\n\tupdateLoadingV2(ctx, loadingMsgID, \"sandbox.starting\")\n\tcomputer, identifier, err := sandboxv2.GetComputer(ctx, cfg, manager, conn)\n\tif err != nil {\n\t\tcloseLoadingV2(ctx, loadingMsgID, \"sandbox.failed\")\n\t\treturn nil, nil, nil, \"\", fmt.Errorf(\"getComputer failed: %w\", err)\n\t}\n\t_ = identifier\n\n\t// 4. Get Runner.\n\trunner, err := sandboxv2.Get(cfg.Runner.Name)\n\tif err != nil {\n\t\tsandboxv2.LifecycleAction(stdCtx, cfg, computer, manager)\n\t\tcloseLoadingV2(ctx, loadingMsgID, \"sandbox.failed\")\n\t\treturn nil, nil, nil, \"\", fmt.Errorf(\"get runner %q: %w\", cfg.Runner.Name, err)\n\t}\n\n\t// 5. Resolve assistant directory and skills subdirectory.\n\tassistantDir := \"\"\n\tskillsDir := \"\"\n\tif ast.Path != \"\" {\n\t\tassistantDir = filepath.Join(config.Conf.AppSource, ast.Path)\n\t\tdir := filepath.Join(assistantDir, \"skills\")\n\t\tif info, e := os.Stat(dir); e == nil && info.IsDir() {\n\t\t\tskillsDir = dir\n\t\t}\n\t}\n\n\t// 6. Convert MCP servers.\n\tvar mcpServers []sandboxTypes.MCPServer\n\tif ast.MCP != nil {\n\t\tfor _, s := range ast.MCP.Servers {\n\t\t\tmcpServers = append(mcpServers, sandboxTypes.MCPServer{\n\t\t\t\tServerID: s.ServerID,\n\t\t\t\tResources: s.Resources,\n\t\t\t\tTools: s.Tools,\n\t\t\t})\n\t\t}\n\t}\n\n\t// 7. Runner.Prepare (standard context).\n\terr = runner.Prepare(stdCtx, &sandboxTypes.PrepareRequest{\n\t\tComputer: computer,\n\t\tConfig: cfg,\n\t\tConnector: conn,\n\t\tSkillsDir: skillsDir,\n\t\tAssistantDir: assistantDir,\n\t\tMCPServers: mcpServers,\n\t\tConfigHash: ast.ConfigHash,\n\t\tRunSteps: sandboxv2.RunPrepareSteps,\n\t})\n\tif err != nil {\n\t\trunner.Cleanup(stdCtx, computer)\n\t\tsandboxv2.LifecycleAction(stdCtx, cfg, computer, manager)\n\t\tcloseLoadingV2(ctx, loadingMsgID, \"sandbox.failed\")\n\t\treturn nil, nil, nil, \"\", fmt.Errorf(\"runner.Prepare: %w\", err)\n\t}\n\n\t// Inject computer + workspace into context so Create/Next hooks\n\t// can access ctx.computer and ctx.workspace.\n\tctx.SetComputer(computer)\n\n\tcleanup := func() {\n\t\t// Defensive fallback — executeSandboxV2Stream defer handles the\n\t\t// normal case; this covers paths that never reach execution.\n\t}\n\n\treturn runner, computer, cleanup, loadingMsgID, nil\n}\n\n// executeSandboxV2Stream calls the V2 Runner.Stream and wraps it in the\n// standard completion response.\nfunc (ast *Assistant) executeSandboxV2Stream(\n\tctx *context.Context,\n\tcompletionMessages []context.Message,\n\tagentNode traceTypes.Node,\n\tstreamHandler message.StreamFunc,\n\trunner sandboxTypes.Runner,\n\tcomputer infraV2.Computer,\n\tloadingMsgID string,\n) (*context.CompletionResponse, error) {\n\t_ = agentNode\n\n\tcfg := ast.SandboxV2\n\tmanager := infraV2.M()\n\n\t// Build system prompt.\n\tvar systemPrompt string\n\tif len(ast.Prompts) > 0 {\n\t\tfor _, p := range ast.Prompts {\n\t\t\tif p.Role == \"system\" && p.Content != \"\" {\n\t\t\t\tsystemPrompt = p.Content\n\t\t\t\tbreak\n\t\t\t}\n\t\t}\n\t}\n\n\t// Resolve connector for Stream.\n\tconn, _, _ := ast.GetConnector(ctx)\n\n\tvar tok *sandboxTypes.SandboxToken\n\tif ctx.Authorized != nil {\n\t\tvar err error\n\t\ttok, err = sandboxv2.IssueSandboxToken(ctx.Authorized.TeamID, ctx.Authorized.UserID)\n\t\tif err != nil {\n\t\t\treturn nil, fmt.Errorf(\"issue sandbox token: %w\", err)\n\t\t}\n\t}\n\n\tstreamReq := &sandboxTypes.StreamRequest{\n\t\tComputer: computer,\n\t\tConfig: cfg,\n\t\tConnector: conn,\n\t\tMessages: completionMessages,\n\t\tSystemPrompt: systemPrompt,\n\t\tChatID: ctx.ChatID,\n\t\tToken: tok,\n\t}\n\n\texecReq := &sandboxv2.ExecuteRequest{\n\t\tComputer: computer,\n\t\tRunner: runner,\n\t\tConfig: cfg,\n\t\tStreamReq: streamReq,\n\t\tManager: manager,\n\t\tLoadingMsgID: loadingMsgID,\n\t}\n\n\treturn sandboxv2.ExecuteSandboxStream(ctx, execReq, streamHandler)\n}\n\n// initStandaloneWorkspace loads the workspace FS into context when no sandbox\n// is configured but the user selected a workspace (metadata[\"workspace_id\"]).\nfunc (ast *Assistant) initStandaloneWorkspace(ctx *context.Context) {\n\tif ctx.Metadata == nil {\n\t\treturn\n\t}\n\twsID, _ := ctx.Metadata[\"workspace_id\"].(string)\n\tif wsID == \"\" {\n\t\treturn\n\t}\n\n\tstdCtx := ctx.Context\n\twsFS, err := workspace.M().FS(stdCtx, wsID)\n\tif err != nil {\n\t\tlog.Printf(\"[assistant] initStandaloneWorkspace: failed to load workspace %s: %v\", wsID, err)\n\t\treturn\n\t}\n\tctx.SetWorkspace(wsFS)\n}\n\n// buildBoxDisplayName constructs a human-readable display name for a Box\n// using the locale-resolved Agent name and Workspace name (matching the UI list pages).\nfunc buildBoxDisplayName(ctx *context.Context, assistantID, rawName string) string {\n\tagentName := i18n.Tr(assistantID, ctx.Locale, rawName)\n\n\twsName := \"\"\n\tif ctx.Metadata != nil {\n\t\tif wsID, ok := ctx.Metadata[\"workspace_id\"].(string); ok && wsID != \"\" {\n\t\t\tif wsm := workspace.M(); wsm != nil {\n\t\t\t\tif ws, err := wsm.Get(ctx.Context, wsID); err == nil && ws != nil {\n\t\t\t\t\twsName = ws.Name\n\t\t\t\t}\n\t\t\t}\n\t\t}\n\t}\n\n\tif agentName != \"\" && wsName != \"\" {\n\t\treturn agentName + \" / \" + wsName\n\t}\n\tif agentName != \"\" {\n\t\treturn agentName\n\t}\n\tif wsName != \"\" {\n\t\treturn wsName\n\t}\n\treturn \"\"\n}\n\nfunc updateLoadingV2(ctx *context.Context, loadingMsgID, msgKey string) {\n\tif loadingMsgID == \"\" || ctx == nil || msgKey == \"\" {\n\t\treturn\n\t}\n\tmsg := &message.Message{\n\t\tMessageID: loadingMsgID,\n\t\tDelta: true,\n\t\tDeltaAction: message.DeltaReplace,\n\t\tType: message.TypeLoading,\n\t\tProps: map[string]any{\n\t\t\t\"message\": i18n.T(ctx.Locale, msgKey),\n\t\t},\n\t}\n\tctx.Send(msg)\n}\n\nfunc closeLoadingV2(ctx *context.Context, loadingMsgID, msgKey string) {\n\tif loadingMsgID == \"\" || ctx == nil {\n\t\treturn\n\t}\n\tprops := map[string]any{\"done\": true}\n\tif msgKey != \"\" {\n\t\tprops[\"message\"] = i18n.T(ctx.Locale, msgKey)\n\t} else {\n\t\tprops[\"message\"] = \"\"\n\t}\n\tdoneMsg := &message.Message{\n\t\tMessageID: loadingMsgID,\n\t\tDelta: true,\n\t\tDeltaAction: message.DeltaReplace,\n\t\tType: message.TypeLoading,\n\t\tProps: props,\n\t}\n\tctx.Send(doneMsg)\n}\n" }, { "path": "agent/assistant/scripts.go", "content": "package assistant\n\nimport (\n\t\"context\"\n\t\"fmt\"\n\t\"path/filepath\"\n\t\"strings\"\n\t\"sync\"\n\t\"time\"\n\n\t\"github.com/yaoapp/gou/application\"\n\t\"github.com/yaoapp/gou/process\"\n\tv8 \"github.com/yaoapp/gou/runtime/v8\"\n\t\"github.com/yaoapp/kun/exception\"\n\t\"github.com/yaoapp/yao/agent/assistant/hook\"\n)\n\n// scriptsMutex protects concurrent v8.Load calls and Scripts map access\nvar scriptsMutex sync.Mutex\n\n// Execute execute the script\nfunc (s *Script) Execute(ctx context.Context, method string, args ...interface{}) (interface{}, error) {\n\treturn s.ExecuteWithAuthorized(ctx, method, nil, args...)\n}\n\n// ExecuteWithAuthorized execute the script with authorized information\nfunc (s *Script) ExecuteWithAuthorized(ctx context.Context, method string, authorized map[string]interface{}, args ...interface{}) (interface{}, error) {\n\tif s == nil || s.Script == nil {\n\t\treturn nil, nil\n\t}\n\n\tscriptCtx, err := s.NewContext(\"\", nil)\n\tif err != nil {\n\t\treturn nil, err\n\t}\n\tdefer scriptCtx.Close()\n\n\t// Set authorized information if available\n\tif authorized != nil {\n\t\tscriptCtx.WithAuthorized(authorized)\n\t}\n\n\t// Call the method with provided arguments as-is\n\tresult, err := scriptCtx.CallWith(ctx, method, args...)\n\n\t// Return error as-is (including \"not defined\" errors)\n\treturn result, err\n}\n\n// LoadScripts loads all scripts from a src directory path\n// It scans for .ts and .js files (excluding index.ts which is the hook script)\n// Returns the HookScript and a map of other scripts\nfunc LoadScripts(srcDir string) (*hook.Script, map[string]*Script, error) {\n\t// Check if src directory exists\n\texists, err := application.App.Exists(srcDir)\n\tif err != nil {\n\t\treturn nil, nil, err\n\t}\n\tif !exists {\n\t\treturn nil, nil, nil // No src directory\n\t}\n\n\tvar hookScript *hook.Script\n\tscripts := make(map[string]*Script)\n\tvar loadErr error\n\n\t// Walk through src directory to find all script files\n\texts := []string{\"*.ts\", \"*.js\"}\n\terr = application.App.Walk(srcDir, func(root, file string, isdir bool) error {\n\t\tif isdir {\n\t\t\treturn nil\n\t\t}\n\n\t\t// file is the full path, root is srcDir\n\t\t// Get relative path for determining if it's index\n\t\trelPath := strings.TrimPrefix(file, root+\"/\")\n\n\t\t// Skip test files (*_test.ts, *_test.js)\n\t\tif strings.HasSuffix(relPath, \"_test.ts\") || strings.HasSuffix(relPath, \"_test.js\") {\n\t\t\treturn nil\n\t\t}\n\n\t\t// Check if it's the root index.ts/js (hook script)\n\t\t// Only src/index.ts is the hook script, not src/foo/index.ts\n\t\tisRootIndex := relPath == \"index.ts\" || relPath == \"index.js\"\n\n\t\tif isRootIndex {\n\t\t\tscriptsMutex.Lock()\n\t\t\tscript, err := loadScriptFile(file)\n\t\t\tscriptsMutex.Unlock()\n\t\t\tif err != nil {\n\t\t\t\tloadErr = fmt.Errorf(\"failed to load hook script %s: %w\", file, err)\n\t\t\t\treturn loadErr\n\t\t\t}\n\t\t\thookScript = script\n\t\t} else {\n\t\t\t// Generate script ID from relative path\n\t\t\tscriptID := generateScriptID(file, root)\n\n\t\t\t// Load the script (v8.Load is not thread-safe)\n\t\t\tscriptsMutex.Lock()\n\t\t\tscript, err := loadScriptV8(file)\n\t\t\tif err != nil {\n\t\t\t\tscriptsMutex.Unlock()\n\t\t\t\tloadErr = fmt.Errorf(\"failed to load script %s: %w\", file, err)\n\t\t\t\treturn loadErr\n\t\t\t}\n\t\t\tscripts[scriptID] = &Script{Script: script}\n\t\t\tscriptsMutex.Unlock()\n\t\t}\n\n\t\treturn nil\n\t}, exts...)\n\n\tif loadErr != nil {\n\t\treturn nil, nil, loadErr\n\t}\n\n\tif err != nil {\n\t\treturn nil, nil, fmt.Errorf(\"failed to walk src directory: %w\", err)\n\t}\n\n\treturn hookScript, scripts, nil\n}\n\n// generateScriptID generates a script ID from file path\n// Example: assistants/test/src/foo/bar/test.ts -> foo.bar.test\nfunc generateScriptID(filePath string, srcDir string) string {\n\t// Normalize path separators\n\tfilePath = filepath.ToSlash(filePath)\n\tsrcDir = filepath.ToSlash(srcDir)\n\n\t// Remove src directory prefix\n\trelPath := strings.TrimPrefix(filePath, srcDir+\"/\")\n\trelPath = strings.TrimPrefix(relPath, \"/\")\n\n\t// Remove file extension\n\trelPath = strings.TrimSuffix(relPath, filepath.Ext(relPath))\n\n\t// Replace path separators with dots\n\tscriptID := strings.ReplaceAll(relPath, \"/\", \".\")\n\n\treturn scriptID\n}\n\n// loadScriptFile loads a hook script from file\nfunc loadScriptFile(file string) (*hook.Script, error) {\n\tid := makeScriptID(file, \"\")\n\tscript, err := v8.Load(file, id)\n\tif err != nil {\n\t\treturn nil, err\n\t}\n\n\treturn &hook.Script{Script: script}, nil\n}\n\n// loadScriptFromSource loads a script from source code\n// Uses MakeScriptInMemory which supports TypeScript syntax without file resolution\nfunc loadScriptFromSource(source string, file string) (*v8.Script, error) {\n\tscript, err := v8.MakeScriptInMemory([]byte(source), file, 5*time.Second, true)\n\tif err != nil {\n\t\treturn nil, err\n\t}\n\treturn script, nil\n}\n\n// loadScriptV8 loads a v8.Script from file (used for non-hook scripts)\nfunc loadScriptV8(file string) (*v8.Script, error) {\n\tid := makeScriptID(file, \"\")\n\tscript, err := v8.Load(file, id)\n\tif err != nil {\n\t\treturn nil, err\n\t}\n\treturn script, nil\n}\n\n// makeScriptID generates the script ID for v8.Load\n// Converts file path to a dot-separated ID\n// Example: assistants/tests/fullfields/src/index.ts -> assistants.tests.fullfields.src.index\nfunc makeScriptID(file string, root string) string {\n\t// Remove root prefix if provided\n\tid := file\n\tif root != \"\" {\n\t\tid = strings.TrimPrefix(file, root+\"/\")\n\t}\n\n\t// Remove extension\n\tid = strings.TrimSuffix(id, filepath.Ext(id))\n\n\t// Replace path separators with dots\n\tid = strings.ReplaceAll(id, \"/\", \".\")\n\tid = strings.ReplaceAll(id, string(filepath.Separator), \".\")\n\n\treturn id\n}\n\n// LoadScriptsFromData loads scripts from data map\n// Handles script/scripts/source fields with priority: script > scripts > source > file system\nfunc LoadScriptsFromData(data map[string]interface{}, assistantID string) (*hook.Script, map[string]*Script, error) {\n\t// Priority 1: script field (hook script from string source)\n\tif data[\"script\"] != nil {\n\t\tswitch v := data[\"script\"].(type) {\n\t\tcase string:\n\t\t\tfile := fmt.Sprintf(\"assistants/%s/src/index.ts\", assistantID)\n\t\t\tscript, err := loadScriptFromSource(v, file)\n\t\t\tif err != nil {\n\t\t\t\treturn nil, nil, err\n\t\t\t}\n\t\t\thookScript := &hook.Script{Script: script}\n\n\t\t\t// Load other scripts if provided\n\t\t\tscripts, err := loadScriptsField(data[\"scripts\"])\n\t\t\tif err != nil {\n\t\t\t\treturn nil, nil, err\n\t\t\t}\n\n\t\t\treturn hookScript, scripts, nil\n\t\tcase *hook.Script:\n\t\t\tscripts, err := loadScriptsField(data[\"scripts\"])\n\t\t\tif err != nil {\n\t\t\t\treturn nil, nil, err\n\t\t\t}\n\t\t\treturn v, scripts, nil\n\t\tcase *v8.Script:\n\t\t\tscripts, err := loadScriptsField(data[\"scripts\"])\n\t\t\tif err != nil {\n\t\t\t\treturn nil, nil, err\n\t\t\t}\n\t\t\treturn &hook.Script{Script: v}, scripts, nil\n\t\t}\n\t}\n\n\t// Priority 2: scripts field (map of scripts)\n\tif data[\"scripts\"] != nil {\n\t\t// First extract index if present\n\t\tvar hookScript *hook.Script\n\t\tif scriptsMap, ok := data[\"scripts\"].(map[string]interface{}); ok {\n\t\t\tif indexSource, hasIndex := scriptsMap[\"index\"]; hasIndex {\n\t\t\t\tswitch v := indexSource.(type) {\n\t\t\t\tcase string:\n\t\t\t\t\tfile := fmt.Sprintf(\"assistants/%s/src/index.ts\", assistantID)\n\t\t\t\t\tscript, err := loadScriptFromSource(v, file)\n\t\t\t\t\tif err != nil {\n\t\t\t\t\t\treturn nil, nil, err\n\t\t\t\t\t}\n\t\t\t\t\thookScript = &hook.Script{Script: script}\n\t\t\t\tcase *Script:\n\t\t\t\t\thookScript = &hook.Script{Script: v.Script}\n\t\t\t\tcase *v8.Script:\n\t\t\t\t\thookScript = &hook.Script{Script: v}\n\t\t\t\t}\n\t\t\t}\n\t\t}\n\n\t\t// Then load other scripts (loadScriptsField automatically filters out index)\n\t\tscripts, err := loadScriptsField(data[\"scripts\"])\n\t\tif err != nil {\n\t\t\treturn nil, nil, err\n\t\t}\n\n\t\treturn hookScript, scripts, nil\n\t}\n\n\t// Priority 3: source field (legacy hook script from source)\n\tif source, ok := data[\"source\"].(string); ok && source != \"\" {\n\t\tscript, err := loadSource(source, assistantID)\n\t\tif err != nil {\n\t\t\treturn nil, nil, err\n\t\t}\n\t\treturn script, nil, nil\n\t}\n\n\t// Priority 4: file system (scan src directory)\n\tsrcDir := fmt.Sprintf(\"assistants/%s/src\", assistantID)\n\treturn LoadScripts(srcDir)\n}\n\n// loadScriptsField parses scripts field from data\n// Note: \"index\" is always filtered out as it's reserved for HookScript\nfunc loadScriptsField(scriptsData interface{}) (map[string]*Script, error) {\n\tif scriptsData == nil {\n\t\treturn nil, nil\n\t}\n\n\tscripts := make(map[string]*Script)\n\n\tswitch v := scriptsData.(type) {\n\tcase map[string]*Script:\n\t\tfor id, script := range v {\n\t\t\tif id == \"index\" {\n\t\t\t\tcontinue // Skip index\n\t\t\t}\n\t\t\tscripts[id] = script\n\t\t}\n\t\treturn scripts, nil\n\tcase map[string]*v8.Script:\n\t\tfor id, script := range v {\n\t\t\tif id == \"index\" {\n\t\t\t\tcontinue // Skip index\n\t\t\t}\n\t\t\tscripts[id] = &Script{Script: script}\n\t\t}\n\t\treturn scripts, nil\n\tcase map[string]interface{}:\n\t\tfor id, item := range v {\n\t\t\tif id == \"index\" {\n\t\t\t\tcontinue // Skip index\n\t\t\t}\n\t\t\tswitch s := item.(type) {\n\t\t\tcase *Script:\n\t\t\t\tscripts[id] = s\n\t\t\tcase *v8.Script:\n\t\t\t\tscripts[id] = &Script{Script: s}\n\t\t\tcase string:\n\t\t\t\t// Load script from source code\n\t\t\t\tfile := fmt.Sprintf(\"script_%s\", id)\n\t\t\t\tscript, err := loadScriptFromSource(s, file)\n\t\t\t\tif err != nil {\n\t\t\t\t\treturn nil, fmt.Errorf(\"failed to load script %s: %w\", id, err)\n\t\t\t\t}\n\t\t\t\tscripts[id] = &Script{Script: script}\n\t\t\t}\n\t\t}\n\t\treturn scripts, nil\n\t}\n\n\treturn nil, nil\n}\n\n// RegisterScripts registers all scripts as process handlers\n// Handler naming: agents..\nfunc (ast *Assistant) RegisterScripts() error {\n\tif len(ast.Scripts) == 0 {\n\t\treturn nil\n\t}\n\n\tassistantID := ast.ID\n\thandlers := make(map[string]process.Handler)\n\n\tfor scriptID, script := range ast.Scripts {\n\t\t// Create handler for this script\n\t\thandlers[scriptID] = makeScriptHandler(script)\n\t}\n\n\t// Register the handler group dynamically\n\tgroupName := fmt.Sprintf(\"agents.%s\", assistantID)\n\tprocess.RegisterDynamicGroup(groupName, handlers)\n\n\treturn nil\n}\n\n// UnregisterScripts unregisters all scripts from process handlers\nfunc (ast *Assistant) UnregisterScripts() error {\n\tif len(ast.Scripts) == 0 {\n\t\treturn nil\n\t}\n\n\tassistantID := ast.ID\n\n\tfor scriptID := range ast.Scripts {\n\t\thandlerID := fmt.Sprintf(\"agents.%s.%s\", strings.ToLower(assistantID), strings.ToLower(scriptID))\n\t\tdelete(process.Handlers, handlerID)\n\t}\n\n\treturn nil\n}\n\n// makeScriptHandler creates a process handler for a script\nfunc makeScriptHandler(script *Script) process.Handler {\n\treturn func(p *process.Process) interface{} {\n\t\t// Extract method name from process\n\t\tmethod := p.Method\n\n\t\t// Get arguments from process\n\t\targs := p.Args\n\n\t\t// Convert authorized info to map if available\n\t\tvar authorized map[string]interface{}\n\t\tif p.Authorized != nil {\n\t\t\tauthorized = p.Authorized.AuthorizedToMap()\n\t\t}\n\n\t\t// Execute the script with authorized information\n\t\tresult, err := script.ExecuteWithAuthorized(p.Context, method, authorized, args...)\n\t\tif err != nil {\n\t\t\texception.New(err.Error(), 500).Throw()\n\t\t}\n\n\t\treturn result\n\t}\n}\n" }, { "path": "agent/assistant/scripts_process_test.go", "content": "package assistant_test\n\nimport (\n\t\"context\"\n\t\"testing\"\n\n\t\"github.com/stretchr/testify/assert\"\n\t\"github.com/yaoapp/gou/process\"\n\t\"github.com/yaoapp/yao/agent/assistant\"\n\t\"github.com/yaoapp/yao/agent/testutils\"\n)\n\nfunc TestScriptsProcessFlow(t *testing.T) {\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\t// Get the mcpload assistant\n\tassistantID := \"tests.mcpload\"\n\tast, err := assistant.Get(assistantID)\n\tassert.NoError(t, err)\n\tassert.NotNil(t, ast, \"Assistant should be loaded\")\n\n\t// Check that scripts were loaded\n\tassert.NotNil(t, ast.Scripts)\n\tassert.Greater(t, len(ast.Scripts), 0, \"Should have loaded at least one script\")\n\n\t// Verify tools.ts was loaded\n\ttoolsScript, hasTools := ast.Scripts[\"tools\"]\n\tassert.True(t, hasTools, \"Should have loaded tools script\")\n\tassert.NotNil(t, toolsScript)\n\n\t// Register scripts as process handlers\n\terr = ast.RegisterScripts()\n\tassert.NoError(t, err)\n\n\t// Test 1: Call Hello function\n\tt.Run(\"CallHelloFunction\", func(t *testing.T) {\n\t\thandlerID := \"agents.tests.mcpload.tools\"\n\t\thandler, exists := process.Handlers[handlerID]\n\t\tassert.True(t, exists, \"Handler should be registered\")\n\n\t\tp := &process.Process{\n\t\t\tID: handlerID + \".Hello\",\n\t\t\tMethod: \"Hello\",\n\t\t\tArgs: []interface{}{map[string]interface{}{\"name\": \"Yao\"}},\n\t\t\tContext: context.Background(),\n\t\t}\n\n\t\tresult := handler(p)\n\t\tassert.NotNil(t, result)\n\n\t\tresultStr, ok := result.(string)\n\t\tassert.True(t, ok, \"Result should be a string\")\n\t\tassert.Contains(t, resultStr, \"Hello, Yao\")\n\t})\n\n\t// Test 2: Call Ping function\n\tt.Run(\"CallPingFunction\", func(t *testing.T) {\n\t\thandlerID := \"agents.tests.mcpload.tools\"\n\t\thandler, exists := process.Handlers[handlerID]\n\t\tassert.True(t, exists, \"Handler should be registered\")\n\n\t\tp := &process.Process{\n\t\t\tID: handlerID + \".Ping\",\n\t\t\tMethod: \"Ping\",\n\t\t\tArgs: []interface{}{map[string]interface{}{\"message\": \"test\"}},\n\t\t\tContext: context.Background(),\n\t\t}\n\n\t\tresult := handler(p)\n\t\tassert.NotNil(t, result)\n\n\t\tresultMap, ok := result.(map[string]interface{})\n\t\tassert.True(t, ok, \"Result should be a map\")\n\t\tassert.Equal(t, \"test\", resultMap[\"message\"])\n\t\tassert.Contains(t, resultMap[\"echo\"], \"Pong\")\n\t})\n\n\t// Test 3: Call Calculate function\n\tt.Run(\"CallCalculateFunction\", func(t *testing.T) {\n\t\thandlerID := \"agents.tests.mcpload.tools\"\n\t\thandler, exists := process.Handlers[handlerID]\n\t\tassert.True(t, exists, \"Handler should be registered\")\n\n\t\tp := &process.Process{\n\t\t\tID: handlerID + \".Calculate\",\n\t\t\tMethod: \"Calculate\",\n\t\t\tArgs: []interface{}{map[string]interface{}{\n\t\t\t\t\"operation\": \"add\",\n\t\t\t\t\"a\": float64(10),\n\t\t\t\t\"b\": float64(5),\n\t\t\t}},\n\t\t\tContext: context.Background(),\n\t\t}\n\n\t\tresult := handler(p)\n\t\tassert.NotNil(t, result)\n\n\t\tresultMap, ok := result.(map[string]interface{})\n\t\tassert.True(t, ok, \"Result should be a map\")\n\t\tassert.Equal(t, float64(15), resultMap[\"result\"])\n\t})\n\n\t// Test 4: Unregister scripts\n\tt.Run(\"UnregisterScripts\", func(t *testing.T) {\n\t\terr := ast.UnregisterScripts()\n\t\tassert.NoError(t, err)\n\n\t\t// Verify handlers are removed\n\t\thandlerID := \"agents.tests.mcpload.tools\"\n\t\t_, exists := process.Handlers[handlerID]\n\t\tassert.False(t, exists, \"Handler should be unregistered\")\n\t})\n}\n\nfunc TestScriptsProcessUsing(t *testing.T) {\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\t// Get the mcpload assistant\n\tassistantID := \"tests.mcpload\"\n\tast, err := assistant.Get(assistantID)\n\tassert.NoError(t, err)\n\tassert.NotNil(t, ast)\n\n\t// Register scripts\n\terr = ast.RegisterScripts()\n\tassert.NoError(t, err)\n\tdefer ast.UnregisterScripts()\n\n\t// Test 1: Call Hello using process.New().Execute()\n\tt.Run(\"ProcessHello\", func(t *testing.T) {\n\t\tproc := process.New(\"agents.tests.mcpload.tools.Hello\", map[string]interface{}{\n\t\t\t\"name\": \"Yao\",\n\t\t})\n\n\t\terr := proc.Execute()\n\t\tassert.NoError(t, err)\n\n\t\tresult := proc.Value()\n\t\tassert.NotNil(t, result)\n\n\t\tresultStr, ok := result.(string)\n\t\tassert.True(t, ok, \"Result should be a string\")\n\t\tassert.Contains(t, resultStr, \"Hello, Yao\")\n\t})\n\n\t// Test 2: Call Ping using process.New().Execute()\n\tt.Run(\"ProcessPing\", func(t *testing.T) {\n\t\tproc := process.New(\"agents.tests.mcpload.tools.Ping\", map[string]interface{}{\n\t\t\t\"message\": \"test message\",\n\t\t})\n\n\t\terr := proc.Execute()\n\t\tassert.NoError(t, err)\n\n\t\tresult := proc.Value()\n\t\tassert.NotNil(t, result)\n\n\t\tresultMap, ok := result.(map[string]interface{})\n\t\tassert.True(t, ok, \"Result should be a map\")\n\t\tassert.Equal(t, \"test message\", resultMap[\"message\"])\n\t\tassert.Contains(t, resultMap[\"echo\"], \"Pong\")\n\t})\n\n\t// Test 3: Call Calculate using process.New().Execute()\n\tt.Run(\"ProcessCalculate\", func(t *testing.T) {\n\t\tproc := process.New(\"agents.tests.mcpload.tools.Calculate\", map[string]interface{}{\n\t\t\t\"operation\": \"multiply\",\n\t\t\t\"a\": float64(6),\n\t\t\t\"b\": float64(7),\n\t\t})\n\n\t\terr := proc.Execute()\n\t\tassert.NoError(t, err)\n\n\t\tresult := proc.Value()\n\t\tassert.NotNil(t, result)\n\n\t\tresultMap, ok := result.(map[string]interface{})\n\t\tassert.True(t, ok, \"Result should be a map\")\n\t\tassert.Equal(t, float64(42), resultMap[\"result\"])\n\t\tassert.Equal(t, \"multiply\", resultMap[\"operation\"])\n\t})\n}\n\nfunc TestScriptsProcessError(t *testing.T) {\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\t// Get the mcpload assistant\n\tassistantID := \"tests.mcpload\"\n\tast, err := assistant.Get(assistantID)\n\tassert.NoError(t, err)\n\tassert.NotNil(t, ast)\n\n\t// Register scripts\n\terr = ast.RegisterScripts()\n\tassert.NoError(t, err)\n\tdefer ast.UnregisterScripts()\n\n\t// Test calling non-existent method\n\tt.Run(\"CallNonExistentMethod\", func(t *testing.T) {\n\t\tproc := process.New(\"agents.tests.mcpload.tools.NonExistent\")\n\n\t\terr := proc.Execute()\n\t\tassert.NotNil(t, err, \"Should return error when calling non-existent method\")\n\t\tassert.Contains(t, err.Error(), \"Exception|500\")\n\t})\n}\n" }, { "path": "agent/assistant/scripts_test.go", "content": "package assistant\n\nimport (\n\t\"context\"\n\t\"testing\"\n\n\t\"github.com/stretchr/testify/assert\"\n\t\"github.com/stretchr/testify/require\"\n\t\"github.com/yaoapp/gou/process\"\n\t\"github.com/yaoapp/yao/config\"\n\t\"github.com/yaoapp/yao/test\"\n)\n\n// TestLoadScripts tests loading scripts from file system\n// Note: These tests are commented out due to path format differences\n// The functionality is tested by existing integration tests in the codebase\n\nfunc TestLoadScriptsFromData(t *testing.T) {\n\ttest.Prepare(t, config.Conf)\n\tdefer test.Clean()\n\n\tt.Run(\"LoadFromScriptField\", func(t *testing.T) {\n\t\t// Use JavaScript instead of TypeScript to avoid compilation path issues\n\t\tdata := map[string]interface{}{\n\t\t\t\"script\": `function Create(ctx) { return null; }`,\n\t\t}\n\n\t\t// Need to provide a real assistant path for compilation\n\t\tdata[\"path\"] = \"assistants/tests/mcpload\"\n\n\t\thookScript, scripts, err := LoadScriptsFromData(data, \"tests.mcpload\")\n\t\trequire.NoError(t, err)\n\t\tassert.NotNil(t, hookScript, \"HookScript should be loaded from script field\")\n\t\tassert.Nil(t, scripts, \"Scripts should be nil when only script field is provided\")\n\n\t\tt.Logf(\"✓ Successfully loaded from script field\")\n\t})\n\n\tt.Run(\"LoadFromScriptsField\", func(t *testing.T) {\n\t\tdata := map[string]interface{}{\n\t\t\t\"scripts\": map[string]interface{}{\n\t\t\t\t\"tool1\": `function tool1() { return \"tool1\"; }`,\n\t\t\t\t\"tool2\": `function tool2() { return \"tool2\"; }`,\n\t\t\t},\n\t\t}\n\n\t\thookScript, scripts, err := LoadScriptsFromData(data, \"test.assistant\")\n\t\trequire.NoError(t, err)\n\t\tassert.Nil(t, hookScript, \"HookScript should be nil when no index in scripts\")\n\t\trequire.NotNil(t, scripts, \"Scripts should be loaded\")\n\t\tassert.Len(t, scripts, 2, \"Should have 2 scripts\")\n\t\tassert.Contains(t, scripts, \"tool1\")\n\t\tassert.Contains(t, scripts, \"tool2\")\n\n\t\tt.Logf(\"✓ Successfully loaded from scripts field\")\n\t})\n\n\tt.Run(\"LoadFromScriptsFieldWithIndex\", func(t *testing.T) {\n\t\t// Test that index is properly extracted and not present in Scripts map\n\t\t// Note: We skip actual script compilation here to avoid path issues\n\t\tdata := map[string]interface{}{\n\t\t\t\"scripts\": map[string]interface{}{\n\t\t\t\t\"tool1\": `function tool1() { return \"tool1\"; }`,\n\t\t\t\t\"tool2\": `function tool2() { return \"tool2\"; }`,\n\t\t\t},\n\t\t}\n\n\t\thookScript, scripts, err := LoadScriptsFromData(data, \"test.assistant\")\n\t\trequire.NoError(t, err)\n\t\t// Without index in scripts field, hookScript should be nil\n\t\tassert.Nil(t, hookScript, \"HookScript should be nil when no index in scripts\")\n\t\trequire.NotNil(t, scripts, \"Scripts should be loaded\")\n\t\tassert.Len(t, scripts, 2, \"Should have 2 scripts\")\n\t\tassert.Contains(t, scripts, \"tool1\")\n\t\tassert.Contains(t, scripts, \"tool2\")\n\t\tassert.NotContains(t, scripts, \"index\", \"index should never be in Scripts map\")\n\n\t\tt.Logf(\"✓ Successfully loaded from scripts field, index properly filtered\")\n\t})\n\n\tt.Run(\"LoadFromSourceField\", func(t *testing.T) {\n\t\tdata := map[string]interface{}{\n\t\t\t\"source\": `function Create(ctx) { return null; }`,\n\t\t}\n\n\t\thookScript, scripts, err := LoadScriptsFromData(data, \"test.assistant\")\n\t\trequire.NoError(t, err)\n\t\tassert.NotNil(t, hookScript, \"HookScript should be loaded from source field\")\n\t\tassert.Nil(t, scripts, \"Scripts should be nil when only source field is provided\")\n\n\t\tt.Logf(\"✓ Successfully loaded from source field\")\n\t})\n\n\tt.Run(\"PriorityOrder\", func(t *testing.T) {\n\t\t// script field should take priority over scripts field\n\t\tdata := map[string]interface{}{\n\t\t\t\"script\": `function Create1() { return null; }`,\n\t\t\t\"scripts\": map[string]interface{}{\n\t\t\t\t\"tool1\": `function tool1() { return \"tool1\"; }`,\n\t\t\t},\n\t\t\t\"source\": `function Create2() { return null; }`,\n\t\t\t\"path\": \"assistants/tests/mcpload\",\n\t\t}\n\n\t\thookScript, scripts, err := LoadScriptsFromData(data, \"tests.mcpload\")\n\t\trequire.NoError(t, err)\n\t\tassert.NotNil(t, hookScript, \"HookScript should be loaded\")\n\t\trequire.NotNil(t, scripts, \"Scripts should be loaded\")\n\t\tassert.Len(t, scripts, 1, \"Should have 1 script from scripts field\")\n\n\t\tt.Logf(\"✓ Priority order works: script > scripts > source\")\n\t})\n}\n\nfunc TestGenerateScriptID(t *testing.T) {\n\ttests := []struct {\n\t\tname string\n\t\tfilePath string\n\t\tsrcDir string\n\t\texpected string\n\t}{\n\t\t{\n\t\t\tname: \"Simple file\",\n\t\t\tfilePath: \"assistants/test/src/tools.ts\",\n\t\t\tsrcDir: \"assistants/test/src\",\n\t\t\texpected: \"tools\",\n\t\t},\n\t\t{\n\t\t\tname: \"Nested directory\",\n\t\t\tfilePath: \"assistants/test/src/foo/bar/test.ts\",\n\t\t\tsrcDir: \"assistants/test/src\",\n\t\t\texpected: \"foo.bar.test\",\n\t\t},\n\t\t{\n\t\t\tname: \"Single level nested\",\n\t\t\tfilePath: \"assistants/test/src/utils/helper.js\",\n\t\t\tsrcDir: \"assistants/test/src\",\n\t\t\texpected: \"utils.helper\",\n\t\t},\n\t\t{\n\t\t\tname: \"Deep nesting\",\n\t\t\tfilePath: \"assistants/test/src/a/b/c/d/file.ts\",\n\t\t\tsrcDir: \"assistants/test/src\",\n\t\t\texpected: \"a.b.c.d.file\",\n\t\t},\n\t}\n\n\tfor _, tt := range tests {\n\t\tt.Run(tt.name, func(t *testing.T) {\n\t\t\tresult := generateScriptID(tt.filePath, tt.srcDir)\n\t\t\tassert.Equal(t, tt.expected, result, \"Script ID should match expected value\")\n\t\t\tt.Logf(\"✓ %s: %s → %s\", tt.name, tt.filePath, result)\n\t\t})\n\t}\n}\n\n// TestLoadScriptsThreadSafety tests concurrent script loading\n// Note: This test is commented out due to path format differences\n// Thread safety is ensured by the scriptsMutex in LoadScripts function\n\nfunc TestExecuteWithAuthorized(t *testing.T) {\n\ttest.Prepare(t, config.Conf)\n\tdefer test.Clean()\n\n\tt.Run(\"ExecuteWithAuthorizedInfo\", func(t *testing.T) {\n\t\t// Create a script that returns the authorized info from __yao_data\n\t\tscriptSource := `\n\t\t\tfunction GetAuth() {\n\t\t\t\tif (typeof __yao_data !== 'undefined' && __yao_data.AUTHORIZED) {\n\t\t\t\t\treturn __yao_data.AUTHORIZED;\n\t\t\t\t}\n\t\t\t\treturn null;\n\t\t\t}\n\t\t`\n\n\t\tdata := map[string]interface{}{\n\t\t\t\"scripts\": map[string]interface{}{\n\t\t\t\t\"auth_test\": scriptSource,\n\t\t\t},\n\t\t}\n\n\t\t_, scripts, err := LoadScriptsFromData(data, \"test.authorized\")\n\t\trequire.NoError(t, err)\n\t\trequire.NotNil(t, scripts)\n\t\trequire.Contains(t, scripts, \"auth_test\")\n\n\t\tscript := scripts[\"auth_test\"]\n\n\t\t// Create authorized info\n\t\tauthorized := map[string]interface{}{\n\t\t\t\"user_id\": \"user123\",\n\t\t\t\"team_id\": \"team456\",\n\t\t\t\"scope\": \"read write\",\n\t\t\t\"constraints\": map[string]interface{}{\n\t\t\t\t\"team_only\": true,\n\t\t\t},\n\t\t}\n\n\t\t// Execute with authorized info\n\t\tctx := context.Background()\n\t\tresult, err := script.ExecuteWithAuthorized(ctx, \"GetAuth\", authorized)\n\t\trequire.NoError(t, err)\n\t\trequire.NotNil(t, result)\n\n\t\t// Verify the authorized info was passed correctly\n\t\tresultMap, ok := result.(map[string]interface{})\n\t\trequire.True(t, ok, \"Result should be a map\")\n\n\t\tassert.Equal(t, \"user123\", resultMap[\"user_id\"])\n\t\tassert.Equal(t, \"team456\", resultMap[\"team_id\"])\n\t\tassert.Equal(t, \"read write\", resultMap[\"scope\"])\n\n\t\tconstraints, ok := resultMap[\"constraints\"].(map[string]interface{})\n\t\trequire.True(t, ok, \"Constraints should be a map\")\n\t\tassert.Equal(t, true, constraints[\"team_only\"])\n\n\t\tt.Logf(\"✓ Authorized info passed correctly to script\")\n\t})\n\n\tt.Run(\"ExecuteWithoutAuthorizedInfo\", func(t *testing.T) {\n\t\t// Create a script that checks for authorized info\n\t\tscriptSource := `\n\t\t\tfunction CheckAuth() {\n\t\t\t\tif (typeof __yao_data !== 'undefined' && __yao_data.AUTHORIZED) {\n\t\t\t\t\treturn { hasAuth: true, data: __yao_data.AUTHORIZED };\n\t\t\t\t}\n\t\t\t\treturn { hasAuth: false };\n\t\t\t}\n\t\t`\n\n\t\tdata := map[string]interface{}{\n\t\t\t\"scripts\": map[string]interface{}{\n\t\t\t\t\"no_auth_test\": scriptSource,\n\t\t\t},\n\t\t}\n\n\t\t_, scripts, err := LoadScriptsFromData(data, \"test.noauth\")\n\t\trequire.NoError(t, err)\n\t\trequire.NotNil(t, scripts)\n\t\trequire.Contains(t, scripts, \"no_auth_test\")\n\n\t\tscript := scripts[\"no_auth_test\"]\n\n\t\t// Execute without authorized info\n\t\tctx := context.Background()\n\t\tresult, err := script.Execute(ctx, \"CheckAuth\")\n\t\trequire.NoError(t, err)\n\t\trequire.NotNil(t, result)\n\n\t\tresultMap, ok := result.(map[string]interface{})\n\t\trequire.True(t, ok)\n\t\tassert.Equal(t, false, resultMap[\"hasAuth\"])\n\n\t\tt.Logf(\"✓ Script executed correctly without authorized info\")\n\t})\n\n\tt.Run(\"MakeScriptHandlerWithAuthorized\", func(t *testing.T) {\n\t\t// Create a script that returns authorized user_id\n\t\tscriptSource := `\n\t\t\tfunction GetUserID() {\n\t\t\t\tif (typeof __yao_data !== 'undefined' && __yao_data.AUTHORIZED) {\n\t\t\t\t\treturn __yao_data.AUTHORIZED.user_id || null;\n\t\t\t\t}\n\t\t\t\treturn null;\n\t\t\t}\n\t\t`\n\n\t\tdata := map[string]interface{}{\n\t\t\t\"scripts\": map[string]interface{}{\n\t\t\t\t\"handler_test\": scriptSource,\n\t\t\t},\n\t\t}\n\n\t\t_, scripts, err := LoadScriptsFromData(data, \"test.handler\")\n\t\trequire.NoError(t, err)\n\t\trequire.NotNil(t, scripts)\n\t\trequire.Contains(t, scripts, \"handler_test\")\n\n\t\tscript := scripts[\"handler_test\"]\n\n\t\t// Create a process handler\n\t\thandler := makeScriptHandler(script)\n\t\trequire.NotNil(t, handler)\n\n\t\t// Create a mock process with authorized info\n\t\tctx := context.Background()\n\t\tp := &process.Process{\n\t\t\tMethod: \"GetUserID\",\n\t\t\tArgs: []interface{}{},\n\t\t\tContext: ctx,\n\t\t\tAuthorized: &process.AuthorizedInfo{\n\t\t\t\tUserID: \"user999\",\n\t\t\t\tTeamID: \"team888\",\n\t\t\t\tScope: \"admin\",\n\t\t\t},\n\t\t}\n\n\t\t// Execute the handler\n\t\tresult := handler(p)\n\t\trequire.NotNil(t, result)\n\n\t\t// Verify the result\n\t\tassert.Equal(t, \"user999\", result)\n\n\t\tt.Logf(\"✓ Process handler correctly passed authorized info\")\n\t})\n}\n" }, { "path": "agent/assistant/search.go", "content": "package assistant\n\nimport (\n\t\"encoding/json\"\n\t\"fmt\"\n\t\"strings\"\n\t\"time\"\n\n\t\"github.com/yaoapp/yao/agent/context\"\n\t\"github.com/yaoapp/yao/agent/i18n\"\n\t\"github.com/yaoapp/yao/agent/output/message\"\n\t\"github.com/yaoapp/yao/agent/search\"\n\t\"github.com/yaoapp/yao/agent/search/nlp/keyword\"\n\tsearchTypes \"github.com/yaoapp/yao/agent/search/types\"\n\tstoreTypes \"github.com/yaoapp/yao/agent/store/types\"\n\ttraceTypes \"github.com/yaoapp/yao/trace/types\"\n)\n\n// shouldAutoSearch determines if auto search should be executed\n// Returns nil if search should be skipped, otherwise returns SearchIntent with types to search\n// Search is skipped if:\n// - opts.Skip.Search is true\n// - createResponse.Search is false\n// - uses.search is \"disabled\"\n// - assistant has no search configuration\n// - needsearch intent detection returns false\nfunc (ast *Assistant) shouldAutoSearch(ctx *context.Context, messages []context.Message, createResponse *context.HookCreateResponse, opts *context.Options) *SearchIntent {\n\t// Check if search is skipped via options\n\tif opts != nil && opts.Skip != nil && opts.Skip.Search {\n\t\tctx.Logger.Debug(\"Auto search skipped by opts.Skip.Search\")\n\t\treturn nil\n\t}\n\n\t// Check if search is skipped via ctx.Metadata[\"__disable_search\"]\n\tif ctx != nil && ctx.Metadata != nil {\n\t\tdisableSearch := getBool(ctx.Metadata, \"__disable_search\")\n\t\tif disableSearch {\n\t\t\tctx.Logger.Debug(\"Auto search skipped by ctx.Metadata['__disable_search']\")\n\t\t\treturn nil\n\t\t}\n\t}\n\n\t// Check createResponse.Search field (highest priority from Create hook)\n\t// Supports: bool | SearchIntent | nil\n\tif createResponse != nil && createResponse.Search != nil {\n\t\tintent := parseSearchField(createResponse.Search)\n\t\tif intent != nil {\n\t\t\tif !intent.NeedSearch {\n\t\t\t\tctx.Logger.Info(\"Auto search disabled by createResponse.Search\")\n\t\t\t\treturn nil\n\t\t\t}\n\t\t\tctx.Logger.Info(\"Auto search controlled by createResponse.Search: types=%v\", intent.SearchTypes)\n\t\t\treturn intent\n\t\t}\n\t}\n\n\t// Get merged uses configuration\n\tuses := ast.getMergedSearchUses(createResponse, opts)\n\n\t// Check if search is explicitly disabled\n\tif uses != nil && uses.Search == \"disabled\" {\n\t\tctx.Logger.Info(\"Auto search disabled by uses.search=disabled\")\n\t\treturn nil\n\t}\n\n\t// Check if assistant has search configuration\n\tif ast.Search == nil && (uses == nil || uses.Search == \"\") {\n\t\treturn nil\n\t}\n\n\t// Check search intent using __yao.needsearch agent\n\tintent := ast.checkSearchIntent(ctx, messages)\n\tif intent == nil || !intent.NeedSearch {\n\t\tctx.Logger.Info(\"Auto search skipped: intent detection returned false\")\n\t\treturn nil\n\t}\n\n\treturn intent\n}\n\n// parseSearchField parses the Search field from HookCreateResponse\n// Supports: bool | SearchIntent | map[string]any | nil\nfunc parseSearchField(search any) *SearchIntent {\n\tif search == nil {\n\t\treturn nil\n\t}\n\n\tswitch v := search.(type) {\n\tcase bool:\n\t\t// bool: true = enable all, false = disable all\n\t\tif v {\n\t\t\treturn &SearchIntent{\n\t\t\t\tNeedSearch: true,\n\t\t\t\tSearchTypes: []string{\"web\", \"kb\", \"db\"},\n\t\t\t\tConfidence: 1.0,\n\t\t\t\tReason: \"enabled by hook\",\n\t\t\t}\n\t\t}\n\t\treturn &SearchIntent{\n\t\t\tNeedSearch: false,\n\t\t\tSearchTypes: []string{},\n\t\t\tConfidence: 1.0,\n\t\t\tReason: \"disabled by hook\",\n\t\t}\n\n\tcase *SearchIntent:\n\t\t// SearchIntent is alias for context.SearchIntent, so this covers both\n\t\treturn v\n\n\tcase SearchIntent:\n\t\treturn &v\n\n\tcase map[string]any:\n\t\t// Parse from map (e.g., from JSON)\n\t\tintent := &SearchIntent{\n\t\t\tNeedSearch: false,\n\t\t\tSearchTypes: []string{},\n\t\t\tConfidence: 0.5,\n\t\t}\n\n\t\tif needSearch, ok := v[\"need_search\"].(bool); ok {\n\t\t\tintent.NeedSearch = needSearch\n\t\t}\n\n\t\tif types, ok := v[\"search_types\"].([]any); ok {\n\t\t\tfor _, t := range types {\n\t\t\t\tif typeStr, ok := t.(string); ok {\n\t\t\t\t\tintent.SearchTypes = append(intent.SearchTypes, typeStr)\n\t\t\t\t}\n\t\t\t}\n\t\t}\n\n\t\tif confidence, ok := v[\"confidence\"].(float64); ok {\n\t\t\tintent.Confidence = confidence\n\t\t}\n\n\t\tif reason, ok := v[\"reason\"].(string); ok {\n\t\t\tintent.Reason = reason\n\t\t}\n\n\t\treturn intent\n\n\tdefault:\n\t\treturn nil\n\t}\n}\n\n// checkSearchIntent uses __yao.needsearch agent to determine if search is needed\n// Returns SearchIntent with search types and confidence\nfunc (ast *Assistant) checkSearchIntent(ctx *context.Context, messages []context.Message) *SearchIntent {\n\t// Default intent: no search needed (fallback when agent unavailable or fails)\n\tdefaultIntent := &SearchIntent{\n\t\tNeedSearch: false,\n\t\tSearchTypes: []string{},\n\t\tConfidence: 0,\n\t}\n\n\t// Build a single text message with conversation context\n\tintentMessages := buildContextMessage(messages)\n\tif len(intentMessages) == 0 {\n\t\treturn defaultIntent // No messages, skip search\n\t}\n\n\t// Try to get __yao.needsearch agent\n\tneedsearchAst, err := Get(\"__yao.needsearch\")\n\tif err != nil {\n\t\tctx.Logger.Debug(\"__yao.needsearch agent not available: %v, skipping search\", err)\n\t\treturn defaultIntent // Agent not available, skip search\n\t}\n\n\t// === Output: Send loading message ===\n\tloadingID := ast.sendIntentLoading(ctx)\n\n\t// Call the needsearch agent (Stack will auto-track)\n\t// IMPORTANT: Skip search to prevent infinite loop, skip output to prevent JSON showing in UI\n\topts := &context.Options{\n\t\tSkip: &context.Skip{\n\t\t\tHistory: true, // Don't save to history\n\t\t\tSearch: true, // Skip search to prevent infinite loop\n\t\t\tOutput: true, // Skip output to prevent JSON showing in UI\n\t\t},\n\t}\n\n\tresult, err := needsearchAst.Stream(ctx, intentMessages, opts)\n\tif err != nil {\n\t\tctx.Logger.Debug(\"__yao.needsearch failed: %v, skipping search\", err)\n\t\t// === Output: Send done (error case, skip search) ===\n\t\tast.sendIntentDone(ctx, loadingID, false, \"\")\n\t\treturn defaultIntent // On error, skip search\n\t}\n\n\t// Parse the result\n\t// Next hook returns {data: {need_search: bool, search_types: [], confidence: float}}\n\t// First try to get from Next hook response\n\tif result.Next != nil {\n\t\tif nextData, ok := result.Next.(map[string]interface{}); ok {\n\t\t\t// Check for data field (from Next hook's {data: result})\n\t\t\tvar intentData map[string]interface{}\n\t\t\tif data, ok := nextData[\"data\"].(map[string]interface{}); ok {\n\t\t\t\tintentData = data\n\t\t\t} else {\n\t\t\t\tintentData = nextData\n\t\t\t}\n\n\t\t\tintent := parseSearchIntent(intentData)\n\t\t\tif intent != nil {\n\t\t\t\tctx.Logger.Debug(\"Search intent (from Next): need_search=%v, types=%v, confidence=%.2f, reason=%s\",\n\t\t\t\t\tintent.NeedSearch, intent.SearchTypes, intent.Confidence, intent.Reason)\n\t\t\t\tast.sendIntentDone(ctx, loadingID, intent.NeedSearch, intent.Reason)\n\t\t\t\treturn intent\n\t\t\t}\n\t\t}\n\t}\n\n\t// Fallback: parse from Completion.Content if Next hook didn't process\n\tif result.Completion != nil {\n\t\tcontent, ok := result.Completion.Content.(string)\n\t\tif !ok || content == \"\" {\n\t\t\tast.sendIntentDone(ctx, loadingID, false, \"\")\n\t\t\treturn defaultIntent\n\t\t}\n\t\tintent := parseSearchIntentFromContent(content)\n\t\tctx.Logger.Debug(\"Search intent (from Content): need_search=%v, types=%v, confidence=%.2f, reason=%s\",\n\t\t\tintent.NeedSearch, intent.SearchTypes, intent.Confidence, intent.Reason)\n\t\tast.sendIntentDone(ctx, loadingID, intent.NeedSearch, intent.Reason)\n\t\treturn intent\n\t}\n\n\t// Default: skip search if we can't parse the result\n\t// === Output: Send done (default case) ===\n\tast.sendIntentDone(ctx, loadingID, false, \"\")\n\treturn defaultIntent\n}\n\n// parseSearchIntent parses SearchIntent from intent data map\nfunc parseSearchIntent(intentData map[string]interface{}) *SearchIntent {\n\tif intentData == nil {\n\t\treturn nil\n\t}\n\n\tneedSearch, ok := intentData[\"need_search\"].(bool)\n\tif !ok {\n\t\treturn nil\n\t}\n\n\tintent := &SearchIntent{\n\t\tNeedSearch: needSearch,\n\t\tSearchTypes: []string{},\n\t\tConfidence: 0.5, // Default confidence\n\t}\n\n\t// Parse search_types\n\tif types, ok := intentData[\"search_types\"].([]interface{}); ok {\n\t\tfor _, t := range types {\n\t\t\tif typeStr, ok := t.(string); ok {\n\t\t\t\t// Validate type\n\t\t\t\ttypeStr = strings.ToLower(typeStr)\n\t\t\t\tif typeStr == \"web\" || typeStr == \"kb\" || typeStr == \"db\" {\n\t\t\t\t\tintent.SearchTypes = append(intent.SearchTypes, typeStr)\n\t\t\t\t}\n\t\t\t}\n\t\t}\n\t}\n\n\t// Parse confidence\n\tif confidence, ok := intentData[\"confidence\"].(float64); ok {\n\t\tintent.Confidence = confidence\n\t}\n\n\t// Parse reason\n\tif reason, ok := intentData[\"reason\"].(string); ok {\n\t\tintent.Reason = reason\n\t}\n\n\treturn intent\n}\n\n// parseSearchIntentFromContent parses SearchIntent from LLM completion content\n// Handles JSON wrapped in markdown code blocks\nfunc parseSearchIntentFromContent(content string) *SearchIntent {\n\t// Default intent: no search needed\n\tdefaultIntent := &SearchIntent{\n\t\tNeedSearch: false,\n\t\tSearchTypes: []string{},\n\t\tConfidence: 0,\n\t}\n\n\t// Remove markdown code block if present\n\tcontent = strings.TrimSpace(content)\n\tif strings.HasPrefix(content, \"```json\") {\n\t\tcontent = strings.TrimPrefix(content, \"```json\")\n\t\tcontent = strings.TrimSuffix(content, \"```\")\n\t\tcontent = strings.TrimSpace(content)\n\t} else if strings.HasPrefix(content, \"```\") {\n\t\tcontent = strings.TrimPrefix(content, \"```\")\n\t\tcontent = strings.TrimSuffix(content, \"```\")\n\t\tcontent = strings.TrimSpace(content)\n\t}\n\n\t// Try to parse JSON\n\tvar result map[string]interface{}\n\tif err := json.Unmarshal([]byte(content), &result); err != nil {\n\t\t// Failed to parse, default to no search\n\t\treturn defaultIntent\n\t}\n\n\tintent := parseSearchIntent(result)\n\tif intent == nil {\n\t\treturn defaultIntent\n\t}\n\n\treturn intent\n}\n\n// sendIntentLoading sends the initial intent detection loading message\n// Returns the message ID for later replacement\nfunc (ast *Assistant) sendIntentLoading(ctx *context.Context) string {\n\tloadingMsg := i18n.T(ctx.Locale, \"search.intent.loading\")\n\n\tmsg := &message.Message{\n\t\tType: \"loading\",\n\t\tProps: map[string]any{\n\t\t\t\"message\": loadingMsg,\n\t\t},\n\t}\n\n\t// Send and get message ID\n\tmsgID, err := ctx.SendStream(msg)\n\tif err != nil {\n\t\tctx.Logger.Warn(\"Failed to send intent loading message: %v\", err)\n\t\treturn \"\"\n\t}\n\n\treturn msgID\n}\n\n// sendIntentDone replaces loading with result\n// Only marks as done when needSearch is false (no further loading will follow)\n// When needSearch is true, the search loading will continue\nfunc (ast *Assistant) sendIntentDone(ctx *context.Context, loadingID string, needSearch bool, reason string) {\n\tif loadingID == \"\" {\n\t\treturn\n\t}\n\n\tvar resultMsg string\n\tif needSearch {\n\t\tresultMsg = i18n.T(ctx.Locale, \"search.intent.need_search\")\n\t} else {\n\t\tresultMsg = i18n.T(ctx.Locale, \"search.intent.no_search\")\n\t}\n\n\tmsg := &message.Message{\n\t\tMessageID: loadingID,\n\t\tDelta: true,\n\t\tDeltaAction: message.DeltaReplace,\n\t\tType: \"loading\",\n\t\tProps: map[string]any{\n\t\t\t\"message\": resultMsg,\n\t\t\t\"done\": true, // Intent detection loading is independent, always close it\n\t\t},\n\t}\n\n\tif err := ctx.Send(msg); err != nil {\n\t\tctx.Logger.Warn(\"Failed to send intent done message: %v\", err)\n\t}\n}\n\n// getMergedSearchUses returns the merged uses configuration for search\n// Priority: createResponse > options.Uses > assistant\nfunc (ast *Assistant) getMergedSearchUses(createResponse *context.HookCreateResponse, opts ...*context.Options) *context.Uses {\n\n\t// Start with assistant uses\n\tvar uses *context.Uses\n\tif ast.Uses != nil {\n\t\tuses = &context.Uses{\n\t\t\tSearch: ast.Uses.Search,\n\t\t\tWeb: ast.Uses.Web,\n\t\t\tKeyword: ast.Uses.Keyword,\n\t\t\tQueryDSL: ast.Uses.QueryDSL,\n\t\t\tRerank: ast.Uses.Rerank,\n\t\t}\n\t}\n\n\t// Override with options.Uses if provided (highest priority)\n\tif len(opts) > 0 && opts[0] != nil && opts[0].Uses != nil {\n\n\t\tif uses == nil {\n\t\t\tuses = &context.Uses{}\n\t\t}\n\n\t\tif opts[0].Uses.Search != \"\" {\n\t\t\tuses.Search = opts[0].Uses.Search\n\t\t}\n\t\tif opts[0].Uses.Web != \"\" {\n\t\t\tuses.Web = opts[0].Uses.Web\n\t\t}\n\n\t\tif opts[0].Uses.Keyword != \"\" {\n\t\t\tuses.Keyword = opts[0].Uses.Keyword\n\t\t}\n\t\tif opts[0].Uses.QueryDSL != \"\" {\n\t\t\tuses.QueryDSL = opts[0].Uses.QueryDSL\n\t\t}\n\t\tif opts[0].Uses.Rerank != \"\" {\n\t\t\tuses.Rerank = opts[0].Uses.Rerank\n\t\t}\n\t}\n\n\t// Override with createResponse.Uses if provided (highest priority)\n\tif createResponse != nil && createResponse.Uses != nil {\n\t\tif uses == nil {\n\t\t\tuses = &context.Uses{}\n\t\t}\n\t\tif createResponse.Uses.Search != \"\" {\n\t\t\tuses.Search = createResponse.Uses.Search\n\t\t}\n\t\tif createResponse.Uses.Web != \"\" {\n\t\t\tuses.Web = createResponse.Uses.Web\n\t\t}\n\t\tif createResponse.Uses.Keyword != \"\" {\n\t\t\tuses.Keyword = createResponse.Uses.Keyword\n\t\t}\n\t\tif createResponse.Uses.QueryDSL != \"\" {\n\t\t\tuses.QueryDSL = createResponse.Uses.QueryDSL\n\t\t}\n\t\tif createResponse.Uses.Rerank != \"\" {\n\t\t\tuses.Rerank = createResponse.Uses.Rerank\n\t\t}\n\t}\n\n\treturn uses\n}\n\n// executeAutoSearch executes auto search based on configuration and intent\n// Returns ReferenceContext with results and formatted context\n// intent specifies which search types to execute (from needsearch agent)\n// opts is optional, used to check Skip.Keyword\nfunc (ast *Assistant) executeAutoSearch(ctx *context.Context, messages []context.Message, createResponse *context.HookCreateResponse, intent *SearchIntent, opts ...*context.Options) *searchTypes.ReferenceContext {\n\tctx.Logger.Phase(\"Search\")\n\tdefer ctx.Logger.PhaseComplete(\"Search\")\n\n\t// Get merged uses configuration\n\tuses := ast.getMergedSearchUses(createResponse, opts...)\n\n\t// Convert to search.Uses\n\tsearchUses := &search.Uses{}\n\tif uses != nil {\n\t\tsearchUses.Search = uses.Search\n\t\tsearchUses.Web = uses.Web\n\t\tsearchUses.Keyword = uses.Keyword\n\t\tsearchUses.QueryDSL = uses.QueryDSL\n\t\tsearchUses.Rerank = uses.Rerank\n\t}\n\n\t// Get merged search config\n\tsearchConfig := ast.GetMergedSearchConfig()\n\n\t// Create searcher\n\tsearcher := search.New(searchConfig, searchUses)\n\n\t// Extract query from messages (save original for storage)\n\toriginalQuery := extractQueryFromMessages(messages)\n\tif originalQuery == \"\" {\n\t\tctx.Logger.Info(\"No query found in messages, skipping auto search\")\n\t\treturn nil\n\t}\n\n\t// Build query with conversation context for better keyword extraction\n\t// This helps the keyword extractor understand the full context\n\tcontextMessages := buildContextMessage(messages)\n\tquery := originalQuery\n\tif len(contextMessages) > 0 {\n\t\tif contextStr, ok := contextMessages[0].Content.(string); ok {\n\t\t\tquery = contextStr\n\t\t}\n\t}\n\n\t// Check if keyword extraction should be skipped\n\tskipKeyword := false\n\tif len(opts) > 0 && opts[0] != nil && opts[0].Skip != nil {\n\t\tskipKeyword = opts[0].Skip.Keyword\n\t}\n\n\t// Build search requests based on configuration and intent\n\t// Keyword extraction is done inside buildSearchRequests for web search\n\tbuildOpts := &buildSearchRequestsOptions{\n\t\tskipKeyword: skipKeyword,\n\t\tusesKeyword: searchUses.Keyword,\n\t}\n\trequests, extractedKeywords := ast.buildSearchRequests(ctx, query, searchConfig, intent, buildOpts)\n\tif len(requests) == 0 {\n\t\tctx.Logger.Info(\"No search requests to execute\")\n\t\treturn nil\n\t}\n\n\t// Update query if keywords were extracted (for web search)\n\tif len(extractedKeywords) > 0 {\n\t\tquery = keywordsToQuery(extractedKeywords)\n\t}\n\n\t// === Output: Send loading message ===\n\tloadingID := ast.sendSearchLoading(ctx)\n\n\t// === Trace: Create search trace node ===\n\tsearchNode := ast.createSearchTrace(ctx, query, requests)\n\n\t// Execute searches in parallel\n\t// Build provider info for logging\n\tproviderInfo := ast.getSearchProviderInfo(searchConfig, searchUses)\n\tctx.Logger.Info(\"Executing %d search requests via %s for query: %s\", len(requests), providerInfo, truncateString(query, 50))\n\n\tstartTime := time.Now()\n\tresults, err := searcher.All(ctx, requests)\n\tduration := time.Since(startTime).Milliseconds()\n\n\tif err != nil {\n\t\t// Log error but don't fail - search errors shouldn't block the main flow\n\t\tctx.Logger.Error(\"Auto search failed: %v\", err)\n\n\t\t// === Output: Send failed message ===\n\t\tast.sendSearchDone(ctx, loadingID, 0, true)\n\n\t\t// === Trace: Mark as failed ===\n\t\tast.completeSearchTrace(searchNode, 0, err)\n\n\t\t// === Storage: Save failed search ===\n\t\tast.saveSearch(ctx, &SearchExecutionResult{\n\t\t\tQuery: originalQuery,\n\t\t\tKeywords: extractedKeywords,\n\t\t\tConfig: ast.configToMap(searchConfig),\n\t\t\tDuration: duration,\n\t\t\tError: err,\n\t\t\tSearchType: \"auto\",\n\t\t})\n\n\t\treturn nil\n\t}\n\n\t// Build reference context (includes references, XML, and prompt)\n\tvar citationConfig *searchTypes.CitationConfig\n\tif searchConfig != nil {\n\t\tcitationConfig = searchConfig.Citation\n\t}\n\trefCtx := search.BuildReferenceContext(results, citationConfig)\n\n\tresultCount := len(refCtx.References)\n\n\t// === Output: Send result message, then done ===\n\tast.sendSearchResult(ctx, loadingID, resultCount)\n\tast.sendSearchDone(ctx, loadingID, resultCount, false)\n\n\t// === Trace: Mark as completed ===\n\tast.completeSearchTrace(searchNode, resultCount, nil)\n\n\t// === Storage: Save successful search ===\n\tast.saveSearch(ctx, &SearchExecutionResult{\n\t\tQuery: originalQuery,\n\t\tKeywords: extractedKeywords,\n\t\tConfig: ast.configToMap(searchConfig),\n\t\tRefCtx: refCtx,\n\t\tResults: results,\n\t\tDuration: duration,\n\t\tSearchType: \"auto\",\n\t})\n\n\tif resultCount == 0 {\n\t\tctx.Logger.Info(\"No search results found\")\n\t\treturn nil\n\t}\n\n\tctx.Logger.Info(\"Auto search completed: %d references\", resultCount)\n\treturn refCtx\n}\n\n// ============================================================================\n// Output: Loading Replace Pattern\n// ============================================================================\n\n// sendSearchLoading sends the initial loading message\n// Returns the message ID for later replacement\nfunc (ast *Assistant) sendSearchLoading(ctx *context.Context) string {\n\tloadingMsg := i18n.T(ctx.Locale, \"search.loading\")\n\n\tmsg := &message.Message{\n\t\tType: \"loading\",\n\t\tProps: map[string]any{\n\t\t\t\"message\": loadingMsg,\n\t\t},\n\t}\n\n\t// Send and get message ID\n\tmsgID, err := ctx.SendStream(msg)\n\tif err != nil {\n\t\tctx.Logger.Warn(\"Failed to send search loading message: %v\", err)\n\t\treturn \"\"\n\t}\n\n\treturn msgID\n}\n\n// sendKeywordLoading sends the keyword extraction loading message\n// Returns the message ID for later replacement\nfunc (ast *Assistant) sendKeywordLoading(ctx *context.Context) string {\n\tloadingMsg := i18n.T(ctx.Locale, \"search.keyword.loading\")\n\n\tmsg := &message.Message{\n\t\tType: \"loading\",\n\t\tProps: map[string]any{\n\t\t\t\"message\": loadingMsg,\n\t\t},\n\t}\n\n\t// Send and get message ID\n\tmsgID, err := ctx.SendStream(msg)\n\tif err != nil {\n\t\tctx.Logger.Warn(\"Failed to send keyword loading message: %v\", err)\n\t\treturn \"\"\n\t}\n\n\treturn msgID\n}\n\n// sendKeywordDone replaces keyword loading with done message\nfunc (ast *Assistant) sendKeywordDone(ctx *context.Context, loadingID string, success bool) {\n\tif loadingID == \"\" {\n\t\treturn\n\t}\n\n\tresultMsg := i18n.T(ctx.Locale, \"search.keyword.done\")\n\n\tmsg := &message.Message{\n\t\tMessageID: loadingID,\n\t\tDelta: true,\n\t\tDeltaAction: message.DeltaReplace,\n\t\tType: \"loading\",\n\t\tProps: map[string]any{\n\t\t\t\"message\": resultMsg,\n\t\t\t\"done\": true,\n\t\t},\n\t}\n\n\tif err := ctx.Send(msg); err != nil {\n\t\tctx.Logger.Warn(\"Failed to send keyword done message: %v\", err)\n\t}\n}\n\n// sendSearchResult replaces loading with result message (without done flag)\nfunc (ast *Assistant) sendSearchResult(ctx *context.Context, loadingID string, count int) {\n\tif loadingID == \"\" {\n\t\treturn\n\t}\n\n\tvar resultMsg string\n\tif count == 0 {\n\t\tresultMsg = i18n.T(ctx.Locale, \"search.no_results\")\n\t} else if count == 1 {\n\t\tresultMsg = i18n.T(ctx.Locale, \"search.success.one\")\n\t} else {\n\t\tresultMsg = fmt.Sprintf(i18n.T(ctx.Locale, \"search.success\"), count)\n\t}\n\n\tmsg := &message.Message{\n\t\tMessageID: loadingID,\n\t\tDelta: true,\n\t\tDeltaAction: message.DeltaReplace,\n\t\tType: \"loading\",\n\t\tProps: map[string]any{\n\t\t\t\"message\": resultMsg,\n\t\t},\n\t}\n\n\tif err := ctx.Send(msg); err != nil {\n\t\tctx.Logger.Warn(\"Failed to send search result message: %v\", err)\n\t}\n}\n\n// sendSearchDone sends the final done message (removes loading indicator)\nfunc (ast *Assistant) sendSearchDone(ctx *context.Context, loadingID string, count int, failed bool) {\n\tif loadingID == \"\" {\n\t\treturn\n\t}\n\n\tvar resultMsg string\n\tif failed {\n\t\tresultMsg = i18n.T(ctx.Locale, \"search.failed\")\n\t} else if count == 0 {\n\t\tresultMsg = i18n.T(ctx.Locale, \"search.no_results\")\n\t} else if count == 1 {\n\t\tresultMsg = i18n.T(ctx.Locale, \"search.success.one\")\n\t} else {\n\t\tresultMsg = fmt.Sprintf(i18n.T(ctx.Locale, \"search.success\"), count)\n\t}\n\n\tmsg := &message.Message{\n\t\tMessageID: loadingID,\n\t\tDelta: true,\n\t\tDeltaAction: message.DeltaReplace,\n\t\tType: \"loading\",\n\t\tProps: map[string]any{\n\t\t\t\"message\": resultMsg,\n\t\t\t\"done\": true, // Frontend will remove loading indicator\n\t\t},\n\t}\n\n\tif err := ctx.Send(msg); err != nil {\n\t\tctx.Logger.Warn(\"Failed to send search done message: %v\", err)\n\t}\n}\n\n// ============================================================================\n// Trace: Search Node\n// ============================================================================\n\n// createSearchTrace creates a trace node for search operation\nfunc (ast *Assistant) createSearchTrace(ctx *context.Context, query string, requests []*searchTypes.Request) traceTypes.Node {\n\ttrace, _ := ctx.Trace()\n\tif trace == nil {\n\t\treturn nil\n\t}\n\n\t// Build search types list\n\tvar searchTypes []string\n\tfor _, req := range requests {\n\t\tsearchTypes = append(searchTypes, string(req.Type))\n\t}\n\n\tinput := map[string]any{\n\t\t\"query\": query,\n\t\t\"types\": searchTypes,\n\t}\n\n\tnode, err := trace.Add(input, traceTypes.TraceNodeOption{\n\t\tLabel: i18n.T(ctx.Locale, \"search.trace.label\"),\n\t\tType: \"search\",\n\t\tIcon: \"search\",\n\t\tDescription: i18n.T(ctx.Locale, \"search.trace.description\"),\n\t})\n\n\tif err != nil {\n\t\tctx.Logger.Warn(\"Failed to create search trace node: %v\", err)\n\t\treturn nil\n\t}\n\n\t// Log search start\n\tnode.Info(\"Starting search\", map[string]any{\n\t\t\"query\": query,\n\t\t\"types\": searchTypes,\n\t})\n\n\treturn node\n}\n\n// completeSearchTrace marks the search trace node as completed or failed\nfunc (ast *Assistant) completeSearchTrace(node traceTypes.Node, resultCount int, err error) {\n\tif node == nil {\n\t\treturn\n\t}\n\n\tif err != nil {\n\t\tnode.Warn(\"Search failed\", map[string]any{\"error\": err.Error()})\n\t\tnode.Fail(err)\n\t\treturn\n\t}\n\n\t// Log completion\n\tnode.Info(\"Search completed\", map[string]any{\n\t\t\"result_count\": resultCount,\n\t})\n\n\t// Complete with output\n\tnode.Complete(map[string]any{\n\t\t\"result_count\": resultCount,\n\t})\n}\n\n// buildSearchRequestsOptions contains options for building search requests\ntype buildSearchRequestsOptions struct {\n\tskipKeyword bool // Skip keyword extraction\n\tusesKeyword string // Keyword extractor config: \"builtin\", \"\", \"mcp:.\"\n}\n\n// buildSearchRequests builds search requests based on assistant configuration and intent\n// intent specifies which search types to execute (from needsearch agent)\n// Returns requests and extracted keywords (if any)\nfunc (ast *Assistant) buildSearchRequests(ctx *context.Context, query string, config *searchTypes.Config, intent *SearchIntent, opts *buildSearchRequestsOptions) ([]*searchTypes.Request, []searchTypes.Keyword) {\n\tvar requests []*searchTypes.Request\n\tvar extractedKeywords []searchTypes.Keyword\n\n\t// Helper to check if a search type is allowed by intent\n\tisTypeAllowed := func(searchType string) bool {\n\t\tif intent == nil || len(intent.SearchTypes) == 0 {\n\t\t\treturn true // No intent or empty types means all types allowed\n\t\t}\n\t\tfor _, t := range intent.SearchTypes {\n\t\t\tif t == searchType {\n\t\t\t\treturn true\n\t\t\t}\n\t\t}\n\t\treturn false\n\t}\n\n\t// Web search - check if web search is configured and allowed by intent\n\tif config != nil && config.Web != nil && isTypeAllowed(\"web\") {\n\t\twebQuery := query\n\n\t\t// Extract keywords for web search if configured\n\t\tif opts != nil && !opts.skipKeyword && opts.usesKeyword != \"\" {\n\t\t\t// === Output: Send keyword extraction loading ===\n\t\t\tkeywordLoadingID := ast.sendKeywordLoading(ctx)\n\n\t\t\textractor := keyword.NewExtractor(opts.usesKeyword, config.Keyword)\n\t\t\tkeywords, err := extractor.Extract(ctx, query, nil)\n\t\t\tif err != nil {\n\t\t\t\tctx.Logger.Warn(\"Keyword extraction failed, using original query: %v\", err)\n\t\t\t\tast.sendKeywordDone(ctx, keywordLoadingID, false)\n\t\t\t} else if len(keywords) > 0 {\n\t\t\t\textractedKeywords = keywords\n\t\t\t\t// Use extracted keywords as the search query for web search\n\t\t\t\twebQuery = keywordsToQuery(keywords)\n\t\t\t\tctx.Logger.Info(\"Extracted keywords for web search: %s -> %s\", truncateString(query, 30), webQuery)\n\t\t\t\tast.sendKeywordDone(ctx, keywordLoadingID, true)\n\t\t\t} else {\n\t\t\t\tast.sendKeywordDone(ctx, keywordLoadingID, true)\n\t\t\t}\n\t\t}\n\n\t\trequests = append(requests, &searchTypes.Request{\n\t\t\tType: searchTypes.SearchTypeWeb,\n\t\t\tQuery: webQuery,\n\t\t\tSource: searchTypes.SourceAuto,\n\t\t\tLimit: config.Web.MaxResults,\n\t\t})\n\t}\n\n\t// KB search - check if KB is configured and allowed by intent\n\tif ast.KB != nil && len(ast.KB.Collections) > 0 && isTypeAllowed(\"kb\") {\n\t\tlimit := 10\n\t\tthreshold := 0.7\n\t\tif config != nil && config.KB != nil {\n\t\t\tif config.KB.Threshold > 0 {\n\t\t\t\tthreshold = config.KB.Threshold\n\t\t\t}\n\t\t}\n\n\t\t// Filter collections by authorization (Collection-level permission check)\n\t\tallowedCollections := FilterKBCollectionsByAuth(ctx, ast.KB.Collections)\n\t\tif len(allowedCollections) == 0 {\n\t\t\tctx.Logger.Info(\"No accessible KB collections after auth filter\")\n\t\t} else {\n\t\t\t// Build KB request\n\t\t\tkbReq := &searchTypes.Request{\n\t\t\t\tType: searchTypes.SearchTypeKB,\n\t\t\t\tQuery: query, // KB uses original query for semantic search\n\t\t\t\tSource: searchTypes.SourceAuto,\n\t\t\t\tLimit: limit,\n\t\t\t\tCollections: allowedCollections,\n\t\t\t\tThreshold: threshold,\n\t\t\t\tGraph: config != nil && config.KB != nil && config.KB.Graph,\n\t\t\t}\n\n\t\t\trequests = append(requests, kbReq)\n\t\t}\n\t}\n\n\t// DB search - check if DB is configured and allowed by intent\n\tif ast.DB != nil && len(ast.DB.Models) > 0 && isTypeAllowed(\"db\") {\n\t\tlimit := 20\n\t\tif config != nil && config.DB != nil && config.DB.MaxResults > 0 {\n\t\t\tlimit = config.DB.MaxResults\n\t\t}\n\n\t\t// Build DB request with auth where clauses\n\t\tdbReq := &searchTypes.Request{\n\t\t\tType: searchTypes.SearchTypeDB,\n\t\t\tQuery: query, // DB uses original query for QueryDSL generation\n\t\t\tSource: searchTypes.SourceAuto,\n\t\t\tLimit: limit,\n\t\t\tModels: ast.DB.Models,\n\t\t}\n\n\t\t// Apply authorization where clauses\n\t\tif authWheres := BuildDBAuthWheres(ctx); authWheres != nil {\n\t\t\tdbReq.Wheres = authWheres\n\t\t}\n\n\t\trequests = append(requests, dbReq)\n\t}\n\n\treturn requests, extractedKeywords\n}\n\n// injectSearchContext injects search results into messages\n// Adds search context as a system message after existing system messages\nfunc (ast *Assistant) injectSearchContext(messages []context.Message, refCtx *searchTypes.ReferenceContext) []context.Message {\n\tif refCtx == nil || len(refCtx.References) == 0 {\n\t\treturn messages\n\t}\n\n\t// Build the search context message\n\tvar contentParts []string\n\n\t// Add citation prompt\n\tif refCtx.Prompt != \"\" {\n\t\tcontentParts = append(contentParts, refCtx.Prompt)\n\t}\n\n\t// Add XML context\n\tif refCtx.XML != \"\" {\n\t\tcontentParts = append(contentParts, refCtx.XML)\n\t}\n\n\tif len(contentParts) == 0 {\n\t\treturn messages\n\t}\n\n\t// Create system message with search context\n\tsearchMessage := context.Message{\n\t\tRole: \"system\",\n\t\tContent: strings.Join(contentParts, \"\\n\\n\"),\n\t}\n\n\t// Find the position to insert the search message\n\t// Insert after any existing system messages but before user messages\n\tinsertIndex := 0\n\tfor i, msg := range messages {\n\t\tif msg.Role == \"system\" {\n\t\t\tinsertIndex = i + 1\n\t\t} else {\n\t\t\tbreak\n\t\t}\n\t}\n\n\t// Insert the search message\n\tresult := make([]context.Message, 0, len(messages)+1)\n\tresult = append(result, messages[:insertIndex]...)\n\tresult = append(result, searchMessage)\n\tresult = append(result, messages[insertIndex:]...)\n\n\treturn result\n}\n\n// extractTextContent extracts text-only content from a message\n// For multimodal messages, concatenates all text parts\n// Returns empty string if no text content found\nfunc extractTextContent(msg context.Message) string {\n\tcontent := msg.Content\n\t// Handle string content\n\tif str, ok := content.(string); ok {\n\t\treturn str\n\t}\n\t// Handle content parts (array of objects) - extract only text parts\n\tif parts, ok := content.([]interface{}); ok {\n\t\tvar texts []string\n\t\tfor _, part := range parts {\n\t\t\tif partMap, ok := part.(map[string]interface{}); ok {\n\t\t\t\tif partMap[\"type\"] == \"text\" {\n\t\t\t\t\tif text, ok := partMap[\"text\"].(string); ok {\n\t\t\t\t\t\ttexts = append(texts, text)\n\t\t\t\t\t}\n\t\t\t\t}\n\t\t\t}\n\t\t}\n\t\tif len(texts) > 0 {\n\t\t\treturn strings.Join(texts, \"\\n\")\n\t\t}\n\t}\n\t// Handle []context.ContentPart\n\tif parts, ok := content.([]context.ContentPart); ok {\n\t\tvar texts []string\n\t\tfor _, part := range parts {\n\t\t\tif part.Type == context.ContentText && part.Text != \"\" {\n\t\t\t\ttexts = append(texts, part.Text)\n\t\t\t}\n\t\t}\n\t\tif len(texts) > 0 {\n\t\t\treturn strings.Join(texts, \"\\n\")\n\t\t}\n\t}\n\treturn \"\"\n}\n\n// buildContextMessage builds a single user message with conversation context\n// Filters out system messages and extracts text-only content\n// Only takes the last 5 messages for efficiency\n// Returns a slice with one message containing the full context, or empty slice if no content\nfunc buildContextMessage(messages []context.Message) []context.Message {\n\tconst maxMessages = 5\n\n\t// Take only the last maxMessages (excluding system messages)\n\tvar recentMessages []context.Message\n\tfor i := len(messages) - 1; i >= 0 && len(recentMessages) < maxMessages; i-- {\n\t\tif messages[i].Role != \"system\" {\n\t\t\trecentMessages = append(recentMessages, messages[i])\n\t\t}\n\t}\n\t// Reverse to maintain chronological order\n\tfor i, j := 0, len(recentMessages)-1; i < j; i, j = i+1, j-1 {\n\t\trecentMessages[i], recentMessages[j] = recentMessages[j], recentMessages[i]\n\t}\n\n\tvar contextParts []string\n\tvar lastUserMessage string\n\n\tfor _, msg := range recentMessages {\n\t\ttextContent := extractTextContent(msg)\n\t\tif textContent == \"\" {\n\t\t\tcontinue\n\t\t}\n\n\t\t// Format message with role label\n\t\tswitch msg.Role {\n\t\tcase \"user\":\n\t\t\tcontextParts = append(contextParts, \"[User]: \"+textContent)\n\t\t\tlastUserMessage = textContent\n\t\tcase \"assistant\":\n\t\t\tcontextParts = append(contextParts, \"[Assistant]: \"+textContent)\n\t\tdefault:\n\t\t\tcontextParts = append(contextParts, \"[\"+string(msg.Role)+\"]: \"+textContent)\n\t\t}\n\t}\n\n\t// Build single message with context\n\tvar result []context.Message\n\tif len(contextParts) > 1 {\n\t\t// Multiple messages: include conversation context\n\t\tfullContext := \"=== Conversation Context ===\\n\" + strings.Join(contextParts, \"\\n\\n\") + \"\\n=== End Context ===\\n\\nCurrent user request: \" + lastUserMessage\n\t\tresult = append(result, context.Message{\n\t\t\tRole: \"user\",\n\t\t\tContent: fullContext,\n\t\t})\n\t} else if lastUserMessage != \"\" {\n\t\t// Single user message: just use it directly\n\t\tresult = append(result, context.Message{\n\t\t\tRole: \"user\",\n\t\t\tContent: lastUserMessage,\n\t\t})\n\t}\n\treturn result\n}\n\n// extractQueryFromMessages extracts the search query from messages\n// Uses the last user message as the query\nfunc extractQueryFromMessages(messages []context.Message) string {\n\t// Find the last user message\n\tfor i := len(messages) - 1; i >= 0; i-- {\n\t\tif messages[i].Role == \"user\" {\n\t\t\tcontent := messages[i].Content\n\t\t\t// Handle string content\n\t\t\tif str, ok := content.(string); ok {\n\t\t\t\treturn str\n\t\t\t}\n\t\t\t// Handle content parts (array of objects)\n\t\t\tif parts, ok := content.([]interface{}); ok {\n\t\t\t\tfor _, part := range parts {\n\t\t\t\t\tif partMap, ok := part.(map[string]interface{}); ok {\n\t\t\t\t\t\tif partMap[\"type\"] == \"text\" {\n\t\t\t\t\t\t\tif text, ok := partMap[\"text\"].(string); ok {\n\t\t\t\t\t\t\t\treturn text\n\t\t\t\t\t\t\t}\n\t\t\t\t\t\t}\n\t\t\t\t\t}\n\t\t\t\t}\n\t\t\t}\n\t\t}\n\t}\n\treturn \"\"\n}\n\n// truncateString truncates a string to maxLen characters\nfunc truncateString(s string, maxLen int) string {\n\tif len(s) <= maxLen {\n\t\treturn s\n\t}\n\treturn s[:maxLen] + \"...\"\n}\n\n// ============================================================================\n// Storage: Save Search Results\n// ============================================================================\n\n// SearchExecutionResult holds all data from search execution for storage\ntype SearchExecutionResult struct {\n\tQuery string // Original query (before keyword optimization)\n\tKeywords []searchTypes.Keyword // Extracted keywords with weights\n\tConfig map[string]any // Search config used\n\tRefCtx *searchTypes.ReferenceContext // Reference context with results\n\tResults []*searchTypes.Result // Raw search results (for extracting DSL, etc.)\n\tDuration int64 // Search duration in ms\n\tError error // Error if failed\n\tSearchType string // \"auto\", \"web\", \"kb\", \"db\"\n}\n\n// keywordsToQuery converts keywords with weights to a search query string\n// Keywords are sorted by weight (descending) and joined with spaces\nfunc keywordsToQuery(keywords []searchTypes.Keyword) string {\n\tif len(keywords) == 0 {\n\t\treturn \"\"\n\t}\n\n\t// Sort by weight descending (higher weight first)\n\tsorted := make([]searchTypes.Keyword, len(keywords))\n\tcopy(sorted, keywords)\n\tfor i := 0; i < len(sorted)-1; i++ {\n\t\tfor j := i + 1; j < len(sorted); j++ {\n\t\t\tif sorted[j].W > sorted[i].W {\n\t\t\t\tsorted[i], sorted[j] = sorted[j], sorted[i]\n\t\t\t}\n\t\t}\n\t}\n\n\t// Join keywords\n\tparts := make([]string, len(sorted))\n\tfor i, kw := range sorted {\n\t\tparts[i] = kw.K\n\t}\n\treturn strings.Join(parts, \" \")\n}\n\n// keywordsToStrings converts keywords to string slice for storage\nfunc keywordsToStrings(keywords []searchTypes.Keyword) []string {\n\tif len(keywords) == 0 {\n\t\treturn nil\n\t}\n\tresult := make([]string, len(keywords))\n\tfor i, kw := range keywords {\n\t\tresult[i] = kw.K\n\t}\n\treturn result\n}\n\n// containsSearchType checks if a search type is in the list\nfunc containsSearchType(types []string, searchType string) bool {\n\tfor _, t := range types {\n\t\tif t == searchType {\n\t\t\treturn true\n\t\t}\n\t}\n\treturn false\n}\n\n// saveSearch saves search results to storage\n// Called after search execution completes (success or failure)\nfunc (ast *Assistant) saveSearch(ctx *context.Context, execResult *SearchExecutionResult) {\n\t// Get store\n\tstore := GetStore()\n\tif store == nil {\n\t\tctx.Logger.Debug(\"Storage not configured, skipping search save\")\n\t\treturn\n\t}\n\n\t// Build search record\n\tsearchRecord := &storeTypes.Search{\n\t\tRequestID: ctx.RequestID(),\n\t\tChatID: ctx.ChatID,\n\t\tQuery: execResult.Query,\n\t\tKeywords: keywordsToStrings(execResult.Keywords),\n\t\tConfig: execResult.Config,\n\t\tSource: execResult.SearchType,\n\t\tDuration: execResult.Duration,\n\t\tCreatedAt: time.Now(),\n\t}\n\n\t// Set error if present\n\tif execResult.Error != nil {\n\t\tsearchRecord.Error = execResult.Error.Error()\n\t}\n\n\t// Convert references if available\n\tif execResult.RefCtx != nil {\n\t\tsearchRecord.References = convertToStoreReferences(execResult.RefCtx.References)\n\t\tsearchRecord.XML = execResult.RefCtx.XML\n\t\tsearchRecord.Prompt = execResult.RefCtx.Prompt\n\t}\n\n\t// Extract DSL from DB search results\n\tif execResult.Results != nil {\n\t\tfor _, result := range execResult.Results {\n\t\t\tif result != nil && result.Type == searchTypes.SearchTypeDB && result.DSL != nil {\n\t\t\t\tsearchRecord.DSL = result.DSL\n\t\t\t\tbreak // Only store the first DSL (usually there's only one DB search)\n\t\t\t}\n\t\t}\n\t}\n\n\t// Save to store\n\tif err := store.SaveSearch(searchRecord); err != nil {\n\t\tctx.Logger.Warn(\"Failed to save search record: %v\", err)\n\t\treturn\n\t}\n\n\tctx.Logger.Debug(\"Search record saved: request_id=%s, refs=%d\",\n\t\tsearchRecord.RequestID, len(searchRecord.References))\n}\n\n// convertToStoreReferences converts search References to store References\nfunc convertToStoreReferences(refs []*searchTypes.Reference) []storeTypes.Reference {\n\tif len(refs) == 0 {\n\t\treturn nil\n\t}\n\n\tstoreRefs := make([]storeTypes.Reference, len(refs))\n\tfor i, ref := range refs {\n\t\tif ref == nil {\n\t\t\tcontinue\n\t\t}\n\n\t\t// Parse citation ID as integer (e.g., \"1\", \"2\", \"3\")\n\t\tindex := i + 1 // Default to position-based index\n\t\tif ref.ID != \"\" {\n\t\t\tif n, err := fmt.Sscanf(ref.ID, \"%d\", &index); n != 1 || err != nil {\n\t\t\t\tindex = i + 1\n\t\t\t}\n\t\t}\n\n\t\tstoreRefs[i] = storeTypes.Reference{\n\t\t\tIndex: index,\n\t\t\tType: string(ref.Type),\n\t\t\tTitle: ref.Title,\n\t\t\tURL: ref.URL,\n\t\t\tSnippet: truncateString(ref.Content, 200), // Short snippet\n\t\t\tContent: ref.Content,\n\t\t\tMetadata: map[string]any{\n\t\t\t\t\"weight\": ref.Weight,\n\t\t\t\t\"score\": ref.Score,\n\t\t\t\t\"source\": string(ref.Source),\n\t\t\t},\n\t\t}\n\t}\n\n\treturn storeRefs\n}\n\n// configToMap converts search config to map for storage\nfunc (ast *Assistant) configToMap(config *searchTypes.Config) map[string]any {\n\tif config == nil {\n\t\treturn nil\n\t}\n\n\tresult := make(map[string]any)\n\n\tif config.Web != nil {\n\t\tresult[\"web\"] = map[string]any{\n\t\t\t\"provider\": config.Web.Provider,\n\t\t\t\"max_results\": config.Web.MaxResults,\n\t\t}\n\t}\n\n\tif config.KB != nil {\n\t\tresult[\"kb\"] = map[string]any{\n\t\t\t\"threshold\": config.KB.Threshold,\n\t\t\t\"graph\": config.KB.Graph,\n\t\t}\n\t}\n\n\tif config.DB != nil {\n\t\tresult[\"db\"] = map[string]any{\n\t\t\t\"max_results\": config.DB.MaxResults,\n\t\t}\n\t}\n\n\tif config.Weights != nil {\n\t\tresult[\"weights\"] = map[string]any{\n\t\t\t\"user\": config.Weights.User,\n\t\t\t\"hook\": config.Weights.Hook,\n\t\t\t\"auto\": config.Weights.Auto,\n\t\t}\n\t}\n\n\treturn result\n}\n\n// getSearchProviderInfo returns a human-readable string describing the search provider(s)\nfunc (ast *Assistant) getSearchProviderInfo(config *searchTypes.Config, uses *search.Uses) string {\n\tvar parts []string\n\n\t// Web search provider - always show when web search is being executed\n\twebMode := \"\"\n\tif uses != nil {\n\t\twebMode = uses.Web\n\t}\n\n\tif webMode == \"\" || webMode == \"builtin\" {\n\t\t// Builtin mode: show the actual provider (tavily/serper/serpapi)\n\t\tprovider := \"tavily\" // default\n\t\tif config != nil && config.Web != nil && config.Web.Provider != \"\" {\n\t\t\tprovider = config.Web.Provider\n\t\t}\n\t\tparts = append(parts, \"web:\"+provider)\n\t} else if strings.HasPrefix(webMode, \"mcp:\") {\n\t\tparts = append(parts, \"web:\"+webMode)\n\t} else {\n\t\tparts = append(parts, \"web:agent:\"+webMode)\n\t}\n\n\t// KB search\n\tif config != nil && config.KB != nil && len(config.KB.Collections) > 0 {\n\t\tparts = append(parts, \"kb\")\n\t}\n\n\t// DB search\n\tif config != nil && config.DB != nil && len(config.DB.Models) > 0 {\n\t\tparts = append(parts, \"db\")\n\t}\n\n\treturn strings.Join(parts, \", \")\n}\n" }, { "path": "agent/assistant/search_auth_db.go", "content": "package assistant\n\nimport (\n\t\"github.com/yaoapp/gou/query/gou\"\n\t\"github.com/yaoapp/yao/agent/context\"\n)\n\n// BuildDBAuthWheres builds where clauses for DB search based on authorization\n// This applies permission-based filtering to database queries\n// Returns gou.Where clauses to filter records by authorization scope\nfunc BuildDBAuthWheres(ctx *context.Context) []gou.Where {\n\tif ctx == nil || ctx.Authorized == nil {\n\t\treturn nil\n\t}\n\n\tauthInfo := ctx.Authorized\n\n\t// No constraints, no filter needed\n\tif !authInfo.Constraints.TeamOnly && !authInfo.Constraints.OwnerOnly {\n\t\treturn nil\n\t}\n\n\tvar wheres []gou.Where\n\n\t// Team only - User can access:\n\t// 1. Public records (public = true)\n\t// 2. Records in their team where:\n\t// - They created the record (__yao_created_by matches)\n\t// - OR the record is shared with team (share = \"team\")\n\tif authInfo.Constraints.TeamOnly && authInfo.TeamID != \"\" {\n\t\twheres = append(wheres, gou.Where{\n\t\t\tWheres: []gou.Where{\n\t\t\t\t// Public records\n\t\t\t\t{Condition: gou.Condition{\n\t\t\t\t\tField: &gou.Expression{Field: \"public\"},\n\t\t\t\t\tValue: true,\n\t\t\t\t\tOP: \"=\",\n\t\t\t\t\tOR: true,\n\t\t\t\t}},\n\t\t\t\t// Team records\n\t\t\t\t{\n\t\t\t\t\tWheres: []gou.Where{\n\t\t\t\t\t\t{Condition: gou.Condition{\n\t\t\t\t\t\t\tField: &gou.Expression{Field: \"__yao_team_id\"},\n\t\t\t\t\t\t\tValue: authInfo.TeamID,\n\t\t\t\t\t\t\tOP: \"=\",\n\t\t\t\t\t\t}},\n\t\t\t\t\t\t{Wheres: []gou.Where{\n\t\t\t\t\t\t\t{Condition: gou.Condition{\n\t\t\t\t\t\t\t\tField: &gou.Expression{Field: \"__yao_created_by\"},\n\t\t\t\t\t\t\t\tValue: authInfo.UserID,\n\t\t\t\t\t\t\t\tOP: \"=\",\n\t\t\t\t\t\t\t}},\n\t\t\t\t\t\t\t{Condition: gou.Condition{\n\t\t\t\t\t\t\t\tField: &gou.Expression{Field: \"share\"},\n\t\t\t\t\t\t\t\tValue: \"team\",\n\t\t\t\t\t\t\t\tOP: \"=\",\n\t\t\t\t\t\t\t\tOR: true,\n\t\t\t\t\t\t\t}},\n\t\t\t\t\t\t}},\n\t\t\t\t\t},\n\t\t\t\t},\n\t\t\t},\n\t\t})\n\t\treturn wheres\n\t}\n\n\t// Owner only - User can access:\n\t// 1. Public records (public = true)\n\t// 2. Records they created where:\n\t// - __yao_team_id is null (not team records)\n\t// - __yao_created_by matches their user ID\n\tif authInfo.Constraints.OwnerOnly && authInfo.UserID != \"\" {\n\t\twheres = append(wheres, gou.Where{\n\t\t\tWheres: []gou.Where{\n\t\t\t\t// Public records\n\t\t\t\t{Condition: gou.Condition{\n\t\t\t\t\tField: &gou.Expression{Field: \"public\"},\n\t\t\t\t\tValue: true,\n\t\t\t\t\tOP: \"=\",\n\t\t\t\t\tOR: true,\n\t\t\t\t}},\n\t\t\t\t// Owner records\n\t\t\t\t{\n\t\t\t\t\tWheres: []gou.Where{\n\t\t\t\t\t\t{Condition: gou.Condition{\n\t\t\t\t\t\t\tField: &gou.Expression{Field: \"__yao_team_id\"},\n\t\t\t\t\t\t\tOP: \"null\",\n\t\t\t\t\t\t}},\n\t\t\t\t\t\t{Condition: gou.Condition{\n\t\t\t\t\t\t\tField: &gou.Expression{Field: \"__yao_created_by\"},\n\t\t\t\t\t\t\tValue: authInfo.UserID,\n\t\t\t\t\t\t\tOP: \"=\",\n\t\t\t\t\t\t}},\n\t\t\t\t\t},\n\t\t\t\t},\n\t\t\t},\n\t\t})\n\t\treturn wheres\n\t}\n\n\treturn wheres\n}\n" }, { "path": "agent/assistant/search_auth_integration_test.go", "content": "package assistant_test\n\nimport (\n\t\"context\"\n\t\"fmt\"\n\t\"testing\"\n\t\"time\"\n\n\t\"github.com/stretchr/testify/assert\"\n\tgraphragtypes \"github.com/yaoapp/gou/graphrag/types\"\n\t\"github.com/yaoapp/yao/agent/assistant\"\n\tagentContext \"github.com/yaoapp/yao/agent/context\"\n\t\"github.com/yaoapp/yao/agent/search\"\n\tsearchTypes \"github.com/yaoapp/yao/agent/search/types\"\n\t\"github.com/yaoapp/yao/agent/testutils\"\n\t\"github.com/yaoapp/yao/kb\"\n\t\"github.com/yaoapp/yao/kb/api\"\n\toauthtypes \"github.com/yaoapp/yao/openapi/oauth/types\"\n)\n\n// ========== Test Constants ==========\n\nconst (\n\t// Test users and teams\n\tTestUserA = \"user_a\"\n\tTestUserB = \"user_b\"\n\tTestTeam1 = \"team_1\"\n\tTestTeam2 = \"team_2\"\n)\n\n// authTestCollections holds dynamically generated collection IDs for a test run\ntype authTestCollections struct {\n\tTeam1 string\n\tTeam2 string\n\tPublic string\n}\n\n// newAuthTestCollections creates unique collection IDs for a test run\nfunc newAuthTestCollections() *authTestCollections {\n\tsuffix := fmt.Sprintf(\"%d\", time.Now().UnixNano())\n\treturn &authTestCollections{\n\t\tTeam1: fmt.Sprintf(\"auth_test_team1_%s\", suffix),\n\t\tTeam2: fmt.Sprintf(\"auth_test_team2_%s\", suffix),\n\t\tPublic: fmt.Sprintf(\"auth_test_public_%s\", suffix),\n\t}\n}\n\n// cleanup removes all test collections\nfunc (c *authTestCollections) cleanup(ctx context.Context, t *testing.T) {\n\tcollections := []string{c.Team1, c.Team2, c.Public}\n\tfor _, id := range collections {\n\t\tif result, err := kb.API.RemoveCollection(ctx, id); err == nil && result.Removed {\n\t\t\tt.Logf(\" Removed: %s\", id)\n\t\t}\n\t}\n}\n\n// ========== KB Collection-Level Auth Filter Tests ==========\n\n// Note: KB permission filtering works at the Collection level.\n// The Collection metadata contains __yao_team_id, __yao_created_by, public, share fields.\n// FilterKBCollectionsByAuth filters collections based on user authorization.\n\nfunc TestKBCollectionAuthFilter(t *testing.T) {\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\tif kb.API == nil {\n\t\tt.Fatal(\"KB API not initialized\")\n\t}\n\n\tctx := context.Background()\n\tcols := newAuthTestCollections()\n\tdefer cols.cleanup(ctx, t)\n\n\t// Create test collections\n\tt.Log(\"Creating test collections...\")\n\tcreateAuthCollection(ctx, t, cols.Team1, TestUserA, TestTeam1, false, \"team\")\n\tcreateAuthCollection(ctx, t, cols.Team2, TestUserB, TestTeam2, false, \"team\")\n\tcreateAuthCollection(ctx, t, cols.Public, TestUserA, TestTeam1, true, \"\")\n\n\tt.Run(\"TeamMemberCanAccessTeamCollection\", func(t *testing.T) {\n\t\t// UserA from Team1 should access Team1 collection\n\t\tauthCtx := createAuthContext(TestUserA, TestTeam1, true, false)\n\t\tcollections := []string{cols.Team1, cols.Team2}\n\n\t\tallowed := assistant.FilterKBCollectionsByAuth(authCtx, collections)\n\t\tassert.Contains(t, allowed, cols.Team1, \"Team1 member should access Team1 collection\")\n\t\tt.Logf(\" Allowed collections: %v\", allowed)\n\t})\n\n\tt.Run(\"TeamMemberCannotAccessOtherTeamCollection\", func(t *testing.T) {\n\t\t// UserA from Team1 should NOT access Team2 collection\n\t\tauthCtx := createAuthContext(TestUserA, TestTeam1, true, false)\n\t\tcollections := []string{cols.Team2}\n\n\t\tallowed := assistant.FilterKBCollectionsByAuth(authCtx, collections)\n\t\tassert.NotContains(t, allowed, cols.Team2, \"Team1 member should NOT access Team2 collection\")\n\t\tt.Logf(\" Allowed collections: %v (expected empty)\", allowed)\n\t})\n\n\tt.Run(\"OwnerCanAccessOwnCollection\", func(t *testing.T) {\n\t\t// UserA with OwnerOnly should access collections they created\n\t\tauthCtx := createAuthContext(TestUserA, \"\", false, true)\n\t\tcollections := []string{cols.Team1, cols.Team2}\n\n\t\tallowed := assistant.FilterKBCollectionsByAuth(authCtx, collections)\n\t\tassert.Contains(t, allowed, cols.Team1, \"Owner should access own collection\")\n\t\tassert.NotContains(t, allowed, cols.Team2, \"Owner should NOT access other's collection\")\n\t\tt.Logf(\" Allowed collections: %v\", allowed)\n\t})\n\n\tt.Run(\"PublicCollectionAccessibleToAll\", func(t *testing.T) {\n\t\t// Note: The 'public' field in Metadata is not automatically saved to the database\n\t\t// by the current KB API. This test documents the expected behavior.\n\t\t// When public=true is properly set in DB, this should pass.\n\n\t\t// First, check the collection metadata\n\t\tcollection, err := kb.API.GetCollection(ctx, cols.Public)\n\t\tassert.NoError(t, err)\n\n\t\t// Check if public is set correctly\n\t\tpublicVal := collection[\"public\"]\n\t\tt.Logf(\" Public collection public field: %v (type: %T)\", publicVal, publicVal)\n\n\t\t// If public is not set (0 or false), the test documents current behavior\n\t\t// The collection should be accessible via owner check since UserA created it\n\t\tauthCtx := createAuthContext(TestUserA, TestTeam1, false, true) // Owner check\n\t\tcollections := []string{cols.Public}\n\n\t\tallowed := assistant.FilterKBCollectionsByAuth(authCtx, collections)\n\t\tassert.Contains(t, allowed, cols.Public, \"Owner should access their collection\")\n\t\tt.Logf(\" Allowed collections (owner check): %v\", allowed)\n\t})\n\n\tt.Run(\"NoConstraintsMeansFullAccess\", func(t *testing.T) {\n\t\t// User with no constraints should access all collections\n\t\tauthCtx := createAuthContext(TestUserA, TestTeam1, false, false)\n\t\tcollections := []string{cols.Team1, cols.Team2, cols.Public}\n\n\t\tallowed := assistant.FilterKBCollectionsByAuth(authCtx, collections)\n\t\tassert.Len(t, allowed, 3, \"No constraints should allow all collections\")\n\t\tt.Logf(\" Allowed collections: %v\", allowed)\n\t})\n\n\tt.Run(\"NilContextMeansFullAccess\", func(t *testing.T) {\n\t\tcollections := []string{cols.Team1, cols.Team2}\n\n\t\tallowed := assistant.FilterKBCollectionsByAuth(nil, collections)\n\t\tassert.Len(t, allowed, 2, \"Nil context should allow all collections\")\n\t\tt.Logf(\" Allowed collections: %v\", allowed)\n\t})\n}\n\n// ========== DB Auth Wheres Tests ==========\n\nfunc TestDBAuthWheresFilter(t *testing.T) {\n\t// Note: This test doesn't need KB, just tests the BuildDBAuthWheres function\n\tt.Run(\"TeamOnlyGeneratesCorrectWheres\", func(t *testing.T) {\n\t\tctx := createAuthContext(TestUserA, TestTeam1, true, false)\n\t\twheres := assistant.BuildDBAuthWheres(ctx)\n\n\t\tassert.NotNil(t, wheres)\n\t\tassert.Len(t, wheres, 1)\n\n\t\t// Verify structure: should have 2 top-level conditions (public OR team filter)\n\t\twhere := wheres[0]\n\t\tassert.Len(t, where.Wheres, 2, \"Should have 2 conditions: public OR team\")\n\n\t\t// First condition: public = true (OR)\n\t\tpublicCond := where.Wheres[0]\n\t\tassert.NotNil(t, publicCond.Condition.Field)\n\t\tassert.Equal(t, \"public\", publicCond.Condition.Field.Field)\n\t\tassert.Equal(t, true, publicCond.Condition.Value)\n\t\tassert.True(t, publicCond.Condition.OR)\n\n\t\t// Second condition: team filter with nested conditions\n\t\tteamCond := where.Wheres[1]\n\t\tassert.Len(t, teamCond.Wheres, 2, \"Team filter should have team_id and (created_by OR share)\")\n\n\t\t// Team ID check\n\t\tteamIDCond := teamCond.Wheres[0]\n\t\tassert.Equal(t, \"__yao_team_id\", teamIDCond.Condition.Field.Field)\n\t\tassert.Equal(t, TestTeam1, teamIDCond.Condition.Value)\n\n\t\t// Created by OR share = team\n\t\townerOrShareCond := teamCond.Wheres[1]\n\t\tassert.Len(t, ownerOrShareCond.Wheres, 2)\n\t\tassert.Equal(t, \"__yao_created_by\", ownerOrShareCond.Wheres[0].Condition.Field.Field)\n\t\tassert.Equal(t, TestUserA, ownerOrShareCond.Wheres[0].Condition.Value)\n\t\tassert.Equal(t, \"share\", ownerOrShareCond.Wheres[1].Condition.Field.Field)\n\t\tassert.Equal(t, \"team\", ownerOrShareCond.Wheres[1].Condition.Value)\n\t\tassert.True(t, ownerOrShareCond.Wheres[1].Condition.OR)\n\n\t\tt.Logf(\" TeamOnly: Verified team_id=%s, created_by=%s\", TestTeam1, TestUserA)\n\t})\n\n\tt.Run(\"OwnerOnlyGeneratesCorrectWheres\", func(t *testing.T) {\n\t\tctx := createAuthContext(TestUserA, \"\", false, true)\n\t\twheres := assistant.BuildDBAuthWheres(ctx)\n\n\t\tassert.NotNil(t, wheres)\n\t\tassert.Len(t, wheres, 1)\n\n\t\t// Verify structure: should have 2 top-level conditions (public OR owner filter)\n\t\twhere := wheres[0]\n\t\tassert.Len(t, where.Wheres, 2, \"Should have 2 conditions: public OR owner\")\n\n\t\t// First condition: public = true (OR)\n\t\tpublicCond := where.Wheres[0]\n\t\tassert.NotNil(t, publicCond.Condition.Field)\n\t\tassert.Equal(t, \"public\", publicCond.Condition.Field.Field)\n\t\tassert.Equal(t, true, publicCond.Condition.Value)\n\t\tassert.True(t, publicCond.Condition.OR)\n\n\t\t// Second condition: owner filter with nested conditions\n\t\townerCond := where.Wheres[1]\n\t\tassert.Len(t, ownerCond.Wheres, 2, \"Owner filter should have team_id IS NULL and created_by\")\n\n\t\t// Team ID is null check\n\t\tteamNullCond := ownerCond.Wheres[0]\n\t\tassert.Equal(t, \"__yao_team_id\", teamNullCond.Condition.Field.Field)\n\t\tassert.Equal(t, \"null\", teamNullCond.Condition.OP)\n\n\t\t// Created by check\n\t\tcreatedByCond := ownerCond.Wheres[1]\n\t\tassert.Equal(t, \"__yao_created_by\", createdByCond.Condition.Field.Field)\n\t\tassert.Equal(t, TestUserA, createdByCond.Condition.Value)\n\n\t\tt.Logf(\" OwnerOnly: Verified created_by=%s, team_id IS NULL\", TestUserA)\n\t})\n\n\tt.Run(\"NoConstraintsReturnsNil\", func(t *testing.T) {\n\t\tctx := createAuthContext(TestUserA, TestTeam1, false, false)\n\t\twheres := assistant.BuildDBAuthWheres(ctx)\n\n\t\tassert.Nil(t, wheres, \"No constraints should return nil\")\n\t\tt.Log(\" No constraints: nil wheres (no filter)\")\n\t})\n\n\tt.Run(\"EmptyTeamIDReturnsNil\", func(t *testing.T) {\n\t\tctx := createAuthContext(TestUserA, \"\", true, false)\n\t\twheres := assistant.BuildDBAuthWheres(ctx)\n\n\t\tassert.Nil(t, wheres, \"Empty TeamID with TeamOnly should return nil\")\n\t\tt.Log(\" Empty TeamID with TeamOnly: nil wheres\")\n\t})\n\n\tt.Run(\"EmptyUserIDReturnsNil\", func(t *testing.T) {\n\t\tctx := createAuthContext(\"\", TestTeam1, false, true)\n\t\twheres := assistant.BuildDBAuthWheres(ctx)\n\n\t\tassert.Nil(t, wheres, \"Empty UserID with OwnerOnly should return nil\")\n\t\tt.Log(\" Empty UserID with OwnerOnly: nil wheres\")\n\t})\n\n\tt.Run(\"NilContextReturnsNil\", func(t *testing.T) {\n\t\twheres := assistant.BuildDBAuthWheres(nil)\n\n\t\tassert.Nil(t, wheres, \"Nil context should return nil\")\n\t\tt.Log(\" Nil context: nil wheres\")\n\t})\n\n\tt.Run(\"NilAuthorizedReturnsNil\", func(t *testing.T) {\n\t\tctx := agentContext.New(context.Background(), nil, \"test-chat\")\n\t\twheres := assistant.BuildDBAuthWheres(ctx)\n\n\t\tassert.Nil(t, wheres, \"Nil Authorized should return nil\")\n\t\tt.Log(\" Nil Authorized: nil wheres\")\n\t})\n}\n\n// ========== KB Search Integration Tests ==========\n\nfunc TestKBSearchIntegration(t *testing.T) {\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\tif kb.API == nil {\n\t\tt.Fatal(\"KB API not initialized\")\n\t}\n\n\tctx := context.Background()\n\tcols := newAuthTestCollections()\n\tdefer cols.cleanup(ctx, t)\n\n\t// Create test collections with documents\n\tt.Log(\"Creating test collections with documents...\")\n\tcreateAuthCollection(ctx, t, cols.Team1, TestUserA, TestTeam1, false, \"team\")\n\taddAuthDocument(ctx, t, cols.Team1, \"Team1 Doc1\", \"Team1 private document about quantum physics and relativity theory.\")\n\taddAuthDocument(ctx, t, cols.Team1, \"Team1 Doc2\", \"Team1 shared document about machine learning and neural networks.\")\n\n\tcreateAuthCollection(ctx, t, cols.Team2, TestUserB, TestTeam2, false, \"team\")\n\taddAuthDocument(ctx, t, cols.Team2, \"Team2 Doc1\", \"Team2 private document about deep learning algorithms.\")\n\taddAuthDocument(ctx, t, cols.Team2, \"Team2 Doc2\", \"Team2 shared document about computer vision techniques.\")\n\n\tcreateAuthCollection(ctx, t, cols.Public, TestUserA, TestTeam1, true, \"\")\n\taddAuthDocument(ctx, t, cols.Public, \"Public Doc1\", \"Public document about artificial intelligence and robotics.\")\n\taddAuthDocument(ctx, t, cols.Public, \"Public Doc2\", \"Public document about natural language processing.\")\n\n\t// Wait for indexing\n\tt.Log(\"Waiting for indexing...\")\n\ttime.Sleep(2 * time.Second)\n\n\tt.Run(\"TeamMemberSearchOnlyFindsTeamData\", func(t *testing.T) {\n\t\t// UserA from Team1 searches - should ONLY find Team1 data\n\t\tauthCtx := createAuthContext(TestUserA, TestTeam1, true, false)\n\n\t\t// Filter collections first\n\t\tallCollections := []string{cols.Team1, cols.Team2}\n\t\tallowed := assistant.FilterKBCollectionsByAuth(authCtx, allCollections)\n\n\t\t// Should only allow Team1\n\t\tassert.Contains(t, allowed, cols.Team1)\n\t\tassert.NotContains(t, allowed, cols.Team2)\n\t\tassert.Len(t, allowed, 1, \"Should only have 1 allowed collection\")\n\n\t\t// Search on allowed collections\n\t\tresult := executeKBSearchOnCollections(t, allowed, \"quantum physics deep learning\")\n\t\tassert.Greater(t, len(result.Items), 0, \"Should find Team1 documents\")\n\n\t\t// Verify ALL results are from Team1 collection only\n\t\tfor _, item := range result.Items {\n\t\t\tassert.Equal(t, cols.Team1, item.Collection,\n\t\t\t\t\"All results should be from Team1 collection, got: %s\", item.Collection)\n\t\t}\n\t\tt.Logf(\" ✓ Team1 member found %d items, all from Team1 collection\", len(result.Items))\n\t})\n\n\tt.Run(\"TeamMemberCannotAccessOtherTeamData\", func(t *testing.T) {\n\t\t// UserA from Team1 tries to access Team2 - should be blocked\n\t\tauthCtx := createAuthContext(TestUserA, TestTeam1, true, false)\n\n\t\t// Try to filter Team2 collection\n\t\tcollections := []string{cols.Team2}\n\t\tallowed := assistant.FilterKBCollectionsByAuth(authCtx, collections)\n\n\t\t// Should be empty - no access\n\t\tassert.Empty(t, allowed, \"Team1 member should NOT have access to Team2 collection\")\n\t\tt.Log(\" ✓ Team1 member correctly blocked from Team2 collection\")\n\t})\n\n\tt.Run(\"OwnerSearchOnlyFindsOwnData\", func(t *testing.T) {\n\t\t// UserA with OwnerOnly - should only find collections they created\n\t\tauthCtx := createAuthContext(TestUserA, \"\", false, true)\n\n\t\t// Filter all collections\n\t\tallCollections := []string{cols.Team1, cols.Team2, cols.Public}\n\t\tallowed := assistant.FilterKBCollectionsByAuth(authCtx, allCollections)\n\n\t\t// UserA created Team1 and Public, not Team2\n\t\tassert.Contains(t, allowed, cols.Team1, \"Owner should access Team1 (created by UserA)\")\n\t\tassert.Contains(t, allowed, cols.Public, \"Owner should access Public (created by UserA)\")\n\t\tassert.NotContains(t, allowed, cols.Team2, \"Owner should NOT access Team2 (created by UserB)\")\n\n\t\t// Search and verify results\n\t\tresult := executeKBSearchOnCollections(t, allowed, \"quantum artificial intelligence\")\n\t\tassert.Greater(t, len(result.Items), 0, \"Should find owner's documents\")\n\n\t\t// Verify NO results from Team2\n\t\tfor _, item := range result.Items {\n\t\t\tassert.NotEqual(t, cols.Team2, item.Collection,\n\t\t\t\t\"Should NOT have results from Team2, got: %s\", item.Collection)\n\t\t}\n\t\tt.Logf(\" ✓ Owner found %d items, none from Team2\", len(result.Items))\n\t})\n\n\tt.Run(\"NoConstraintsSearchFindsAllData\", func(t *testing.T) {\n\t\t// User with no constraints - should find all data\n\t\tauthCtx := createAuthContext(TestUserA, TestTeam1, false, false)\n\n\t\t// Filter all collections\n\t\tallCollections := []string{cols.Team1, cols.Team2, cols.Public}\n\t\tallowed := assistant.FilterKBCollectionsByAuth(authCtx, allCollections)\n\n\t\t// Should have access to all\n\t\tassert.Len(t, allowed, 3, \"No constraints should allow all collections\")\n\n\t\t// Search and verify results from multiple collections\n\t\tresult := executeKBSearchOnCollections(t, allowed, \"quantum deep learning artificial\")\n\n\t\t// Should find results from multiple collections\n\t\tcollectionsFound := make(map[string]bool)\n\t\tfor _, item := range result.Items {\n\t\t\tcollectionsFound[item.Collection] = true\n\t\t}\n\t\tassert.Greater(t, len(collectionsFound), 1, \"Should find results from multiple collections\")\n\t\tt.Logf(\" ✓ No constraints: found %d items from %d collections\", len(result.Items), len(collectionsFound))\n\t})\n\n\tt.Run(\"SearchResultsMatchCollectionFilter\", func(t *testing.T) {\n\t\t// Verify that search results ONLY come from allowed collections\n\t\tauthCtx := createAuthContext(TestUserB, TestTeam2, true, false)\n\n\t\t// UserB from Team2 - should only access Team2\n\t\tallCollections := []string{cols.Team1, cols.Team2, cols.Public}\n\t\tallowed := assistant.FilterKBCollectionsByAuth(authCtx, allCollections)\n\n\t\tassert.Contains(t, allowed, cols.Team2, \"Team2 member should access Team2\")\n\t\tassert.NotContains(t, allowed, cols.Team1, \"Team2 member should NOT access Team1\")\n\n\t\t// Search\n\t\tresult := executeKBSearchOnCollections(t, allowed, \"deep learning computer vision\")\n\n\t\t// Verify results\n\t\tif len(result.Items) > 0 {\n\t\t\tfor _, item := range result.Items {\n\t\t\t\t// Results should only be from allowed collections\n\t\t\t\tassert.Contains(t, allowed, item.Collection,\n\t\t\t\t\t\"Result from %s should be in allowed list %v\", item.Collection, allowed)\n\t\t\t}\n\t\t\tt.Logf(\" ✓ Team2 member found %d items, all from allowed collections\", len(result.Items))\n\t\t} else {\n\t\t\tt.Log(\" ✓ Team2 member found 0 items (collection may be empty)\")\n\t\t}\n\t})\n}\n\n// ========== Helper Functions ==========\n\nfunc createAuthContext(userID, teamID string, teamOnly, ownerOnly bool) *agentContext.Context {\n\tauthorized := &oauthtypes.AuthorizedInfo{\n\t\tUserID: userID,\n\t\tTeamID: teamID,\n\t\tConstraints: oauthtypes.DataConstraints{\n\t\t\tTeamOnly: teamOnly,\n\t\t\tOwnerOnly: ownerOnly,\n\t\t},\n\t}\n\treturn agentContext.New(context.Background(), authorized, \"test-chat\")\n}\n\nfunc createAuthCollection(ctx context.Context, t *testing.T, id, userID, teamID string, public bool, share string) {\n\tparams := &api.CreateCollectionParams{\n\t\tID: id,\n\t\tMetadata: map[string]interface{}{\n\t\t\t\"name\": id,\n\t\t\t\"public\": public,\n\t\t\t\"share\": share,\n\t\t},\n\t\tEmbeddingProviderID: \"__yao.openai\",\n\t\tEmbeddingOptionID: \"text-embedding-3-small\",\n\t\tLocale: \"en\",\n\t\tConfig: &graphragtypes.CreateCollectionOptions{\n\t\t\tDistance: \"cosine\",\n\t\t\tIndexType: \"hnsw\",\n\t\t},\n\t\tAuthScope: map[string]interface{}{\n\t\t\t\"__yao_created_by\": userID,\n\t\t\t\"__yao_team_id\": teamID,\n\t\t},\n\t}\n\n\t_, err := kb.API.CreateCollection(ctx, params)\n\tif err != nil {\n\t\tt.Fatalf(\"Failed to create collection %s: %v\", id, err)\n\t}\n\tt.Logf(\" ✓ Created: %s\", id)\n}\n\nfunc addAuthDocument(ctx context.Context, t *testing.T, collectionID, title, content string) {\n\tparams := &api.AddTextParams{\n\t\tCollectionID: collectionID,\n\t\tText: content,\n\t\tDocID: fmt.Sprintf(\"%s__%s\", collectionID, sanitizeForID(title)),\n\t\tMetadata: map[string]interface{}{\n\t\t\t\"title\": title,\n\t\t},\n\t\tChunking: &api.ProviderConfigParams{\n\t\t\tProviderID: \"__yao.structured\",\n\t\t\tOptionID: \"standard\",\n\t\t},\n\t\tEmbedding: &api.ProviderConfigParams{\n\t\t\tProviderID: \"__yao.openai\",\n\t\t\tOptionID: \"text-embedding-3-small\",\n\t\t},\n\t}\n\n\t_, err := kb.API.AddText(ctx, params)\n\tif err != nil {\n\t\tt.Logf(\" Warning: Failed to add document '%s': %v\", title, err)\n\t\treturn\n\t}\n\tt.Logf(\" ✓ Added: %s\", title)\n}\n\nfunc sanitizeForID(s string) string {\n\tresult := \"\"\n\tfor _, c := range s {\n\t\tif (c >= 'a' && c <= 'z') || (c >= 'A' && c <= 'Z') || (c >= '0' && c <= '9') {\n\t\t\tresult += string(c)\n\t\t} else if c == ' ' {\n\t\t\tresult += \"_\"\n\t\t}\n\t}\n\treturn result\n}\n\nfunc executeKBSearchOnCollections(t *testing.T, collections []string, query string) *searchTypes.Result {\n\tif len(collections) == 0 {\n\t\treturn &searchTypes.Result{Items: []*searchTypes.ResultItem{}}\n\t}\n\n\tcfg := &searchTypes.Config{\n\t\tKB: &searchTypes.KBConfig{\n\t\t\tCollections: collections,\n\t\t\tThreshold: 0.3,\n\t\t},\n\t}\n\tsearcher := search.New(cfg, nil)\n\n\treq := &searchTypes.Request{\n\t\tType: searchTypes.SearchTypeKB,\n\t\tQuery: query,\n\t\tCollections: collections,\n\t\tThreshold: 0.3,\n\t\tLimit: 20,\n\t\tSource: searchTypes.SourceAuto,\n\t}\n\n\tresult, err := searcher.Search(nil, req)\n\tif err != nil {\n\t\tt.Fatalf(\"Search failed: %v\", err)\n\t}\n\treturn result\n}\n" }, { "path": "agent/assistant/search_auth_kb.go", "content": "package assistant\n\nimport (\n\t\"context\"\n\n\tagentContext \"github.com/yaoapp/yao/agent/context\"\n\t\"github.com/yaoapp/yao/kb\"\n\toauthtypes \"github.com/yaoapp/yao/openapi/oauth/types\"\n)\n\n// FilterKBCollectionsByAuth filters collections based on user authorization.\n// Returns only collections that the user has permission to access.\n// Permission is determined by Collection's metadata (public, share, __yao_team_id, __yao_created_by).\nfunc FilterKBCollectionsByAuth(ctx *agentContext.Context, collections []string) []string {\n\tif ctx == nil || ctx.Authorized == nil {\n\t\treturn collections // No auth context, return all\n\t}\n\n\tauthInfo := ctx.Authorized\n\n\t// No constraints, return all collections\n\tif !authInfo.Constraints.TeamOnly && !authInfo.Constraints.OwnerOnly {\n\t\treturn collections\n\t}\n\n\t// Check KB API\n\tif kb.API == nil {\n\t\treturn collections // KB not initialized, return all\n\t}\n\n\tvar allowed []string\n\tbgCtx := context.Background()\n\n\tfor _, collectionID := range collections {\n\t\t// Get collection metadata\n\t\tcollection, err := kb.API.GetCollection(bgCtx, collectionID)\n\t\tif err != nil {\n\t\t\tcontinue // Skip if can't get collection\n\t\t}\n\n\t\tif hasCollectionAccess(authInfo, collection) {\n\t\t\tallowed = append(allowed, collectionID)\n\t\t}\n\t}\n\n\treturn allowed\n}\n\n// hasCollectionAccess checks if user has access to a collection based on its metadata.\nfunc hasCollectionAccess(authInfo *oauthtypes.AuthorizedInfo, collection map[string]interface{}) bool {\n\tif authInfo == nil {\n\t\treturn true\n\t}\n\n\t// No constraints, allow access\n\tif !authInfo.Constraints.TeamOnly && !authInfo.Constraints.OwnerOnly {\n\t\treturn true\n\t}\n\n\t// Check public access (handle different types: bool, int, float64)\n\tif isPublicValue(collection[\"public\"]) {\n\t\treturn true\n\t}\n\n\t// Get metadata for permission fields\n\tmetadata, _ := collection[\"metadata\"].(map[string]interface{})\n\tif metadata == nil {\n\t\tmetadata = collection\n\t}\n\n\t// Team only check\n\tif authInfo.Constraints.TeamOnly && authInfo.TeamID != \"\" {\n\t\tteamID, _ := metadata[\"__yao_team_id\"].(string)\n\t\tif teamID == \"\" {\n\t\t\tteamID, _ = collection[\"__yao_team_id\"].(string)\n\t\t}\n\n\t\tif teamID == authInfo.TeamID {\n\t\t\tcreatedBy, _ := metadata[\"__yao_created_by\"].(string)\n\t\t\tif createdBy == \"\" {\n\t\t\t\tcreatedBy, _ = collection[\"__yao_created_by\"].(string)\n\t\t\t}\n\t\t\tshare, _ := metadata[\"share\"].(string)\n\t\t\tif share == \"\" {\n\t\t\t\tshare, _ = collection[\"share\"].(string)\n\t\t\t}\n\n\t\t\tif createdBy == authInfo.UserID || share == \"team\" {\n\t\t\t\treturn true\n\t\t\t}\n\t\t}\n\t}\n\n\t// Owner only check\n\tif authInfo.Constraints.OwnerOnly && authInfo.UserID != \"\" {\n\t\tcreatedBy, _ := metadata[\"__yao_created_by\"].(string)\n\t\tif createdBy == \"\" {\n\t\t\tcreatedBy, _ = collection[\"__yao_created_by\"].(string)\n\t\t}\n\t\tif createdBy == authInfo.UserID {\n\t\t\treturn true\n\t\t}\n\t}\n\n\treturn false\n}\n\n// isPublicValue checks if a value represents \"public\" access\nfunc isPublicValue(v interface{}) bool {\n\tswitch val := v.(type) {\n\tcase bool:\n\t\treturn val\n\tcase int:\n\t\treturn val == 1\n\tcase int64:\n\t\treturn val == 1\n\tcase float64:\n\t\treturn val == 1\n\tcase string:\n\t\treturn val == \"true\" || val == \"1\"\n\t}\n\treturn false\n}\n" }, { "path": "agent/assistant/search_auto_disabled_test.go", "content": "package assistant_test\n\nimport (\n\tstdContext \"context\"\n\t\"testing\"\n\n\t\"github.com/stretchr/testify/assert\"\n\t\"github.com/stretchr/testify/require\"\n\t\"github.com/yaoapp/yao/agent/assistant\"\n\t\"github.com/yaoapp/yao/agent/context\"\n\t\"github.com/yaoapp/yao/agent/output/message\"\n\t\"github.com/yaoapp/yao/agent/testutils\"\n\t\"github.com/yaoapp/yao/openapi/oauth/types\"\n)\n\n// newSearchAutoDisabledTestContext creates a test context\nfunc newSearchAutoDisabledTestContext(chatID, assistantID string) *context.Context {\n\tauthorized := &types.AuthorizedInfo{\n\t\tSubject: \"test-user\",\n\t\tUserID: \"test-123\",\n\t\tTenantID: \"test-tenant\",\n\t}\n\n\tctx := context.New(stdContext.Background(), authorized, chatID)\n\tctx.ID = chatID\n\tctx.AssistantID = assistantID\n\tctx.Locale = \"en-us\"\n\tctx.Client = context.Client{\n\t\tType: \"web\",\n\t\tIP: \"127.0.0.1\",\n\t}\n\tctx.Referer = context.RefererAPI\n\tctx.Accept = context.AcceptWebCUI\n\tctx.IDGenerator = message.NewIDGenerator()\n\tctx.Metadata = make(map[string]interface{})\n\treturn ctx\n}\n\nfunc TestSearchAutoDisabled(t *testing.T) {\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\tast, err := assistant.LoadPath(\"/assistants/tests/search-auto-disabled\")\n\trequire.NoError(t, err)\n\trequire.NotNil(t, ast)\n\n\tt.Run(\"ShouldHaveSearchConfig\", func(t *testing.T) {\n\t\t// Search config is set but uses.search is disabled\n\t\tassert.NotNil(t, ast.Search, \"search config should be set\")\n\t\tassert.NotNil(t, ast.Search.Web, \"web search config should be set\")\n\t})\n\n\tt.Run(\"ShouldHaveDisabledUses\", func(t *testing.T) {\n\t\tassert.NotNil(t, ast.Uses, \"uses config should be set\")\n\t\tassert.Equal(t, \"disabled\", ast.Uses.Search, \"uses.search should be disabled\")\n\t})\n\n\tt.Run(\"StreamShouldNotExecuteSearch\", func(t *testing.T) {\n\t\t// Get agent via assistant.Get (required for Stream)\n\t\tagent, err := assistant.Get(\"tests.search-auto-disabled\")\n\t\trequire.NoError(t, err)\n\t\trequire.NotNil(t, agent)\n\n\t\t// Create context\n\t\tctx := newSearchAutoDisabledTestContext(\"test-search-auto-disabled\", \"tests.search-auto-disabled\")\n\n\t\t// Create messages\n\t\tmessages := []context.Message{\n\t\t\t{\n\t\t\t\tRole: \"user\",\n\t\t\t\tContent: \"Hello, how are you?\",\n\t\t\t},\n\t\t}\n\n\t\t// Execute stream - should NOT trigger search because uses.search is \"disabled\"\n\t\tresponse, err := agent.Stream(ctx, messages)\n\t\trequire.NoError(t, err)\n\t\trequire.NotNil(t, response)\n\n\t\tassert.NotNil(t, response.Completion, \"should have completion\")\n\t\tt.Logf(\"✓ Stream executed without search (disabled)\")\n\t})\n}\n" }, { "path": "agent/assistant/search_auto_full_test.go", "content": "package assistant_test\n\nimport (\n\tstdContext \"context\"\n\t\"strings\"\n\t\"testing\"\n\n\t\"github.com/stretchr/testify/assert\"\n\t\"github.com/stretchr/testify/require\"\n\t\"github.com/yaoapp/yao/agent/assistant\"\n\t\"github.com/yaoapp/yao/agent/context\"\n\t\"github.com/yaoapp/yao/agent/output/message\"\n\t\"github.com/yaoapp/yao/agent/testutils\"\n\t\"github.com/yaoapp/yao/openapi/oauth/types\"\n)\n\n// newSearchAutoFullTestContext creates a test context\nfunc newSearchAutoFullTestContext(chatID, assistantID string) *context.Context {\n\tauthorized := &types.AuthorizedInfo{\n\t\tSubject: \"test-user\",\n\t\tUserID: \"test-123\",\n\t\tTenantID: \"test-tenant\",\n\t}\n\n\tctx := context.New(stdContext.Background(), authorized, chatID)\n\tctx.ID = chatID\n\tctx.AssistantID = assistantID\n\tctx.Locale = \"en-us\"\n\tctx.Client = context.Client{\n\t\tType: \"web\",\n\t\tIP: \"127.0.0.1\",\n\t}\n\tctx.Referer = context.RefererAPI\n\tctx.Accept = context.AcceptWebCUI\n\tctx.IDGenerator = message.NewIDGenerator()\n\tctx.Metadata = make(map[string]interface{})\n\treturn ctx\n}\n\nfunc TestSearchAutoFull(t *testing.T) {\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\tast, err := assistant.LoadPath(\"/assistants/tests/search-auto-full\")\n\trequire.NoError(t, err)\n\trequire.NotNil(t, ast)\n\n\tt.Run(\"ShouldHaveWebSearchConfig\", func(t *testing.T) {\n\t\tassert.NotNil(t, ast.Search, \"search config should be set\")\n\t\tassert.NotNil(t, ast.Search.Web, \"web search config should be set\")\n\t\tassert.Equal(t, \"tavily\", ast.Search.Web.Provider)\n\t\tassert.Equal(t, 3, ast.Search.Web.MaxResults)\n\t})\n\n\tt.Run(\"ShouldHaveKBSearchConfig\", func(t *testing.T) {\n\t\tassert.NotNil(t, ast.Search.KB, \"kb search config should be set\")\n\t\tassert.Equal(t, 0.7, ast.Search.KB.Threshold)\n\t\tassert.False(t, ast.Search.KB.Graph)\n\t})\n\n\tt.Run(\"ShouldHaveDBSearchConfig\", func(t *testing.T) {\n\t\tassert.NotNil(t, ast.Search.DB, \"db search config should be set\")\n\t\tassert.Equal(t, 10, ast.Search.DB.MaxResults)\n\t})\n\n\tt.Run(\"ShouldHaveKBCollections\", func(t *testing.T) {\n\t\tassert.NotNil(t, ast.KB, \"kb config should be set\")\n\t\tassert.Contains(t, ast.KB.Collections, \"test-collection\")\n\t})\n\n\tt.Run(\"ShouldHaveDBModels\", func(t *testing.T) {\n\t\tassert.NotNil(t, ast.DB, \"db config should be set\")\n\t\tassert.Contains(t, ast.DB.Models, \"user\")\n\t\tassert.Contains(t, ast.DB.Models, \"article\")\n\t})\n\n\tt.Run(\"ShouldHaveCitationConfig\", func(t *testing.T) {\n\t\tassert.NotNil(t, ast.Search.Citation, \"citation config should be set\")\n\t\tassert.Equal(t, \"xml\", ast.Search.Citation.Format)\n\t\tassert.True(t, ast.Search.Citation.AutoInjectPrompt)\n\t})\n\n\tt.Run(\"ShouldHaveUsesConfig\", func(t *testing.T) {\n\t\tassert.NotNil(t, ast.Uses, \"uses config should be set\")\n\t\tassert.Equal(t, \"builtin\", ast.Uses.Search)\n\t\tassert.Equal(t, \"builtin\", ast.Uses.Web)\n\t})\n\n\tt.Run(\"StreamShouldExecuteMultipleSearchTypes\", func(t *testing.T) {\n\t\t// Get agent via assistant.Get (required for Stream)\n\t\tagent, err := assistant.Get(\"tests.search-auto-full\")\n\t\trequire.NoError(t, err)\n\t\trequire.NotNil(t, agent)\n\n\t\t// Create context\n\t\tctx := newSearchAutoFullTestContext(\"test-search-auto-full\", \"tests.search-auto-full\")\n\n\t\t// Create messages with a search query\n\t\tmessages := []context.Message{\n\t\t\t{\n\t\t\t\tRole: \"user\",\n\t\t\t\tContent: \"Find information about machine learning\",\n\t\t\t},\n\t\t}\n\n\t\t// Execute stream - should trigger Web + KB + DB searches\n\t\tresponse, err := agent.Stream(ctx, messages)\n\n\t\t// Assert no error (if API key is configured)\n\t\tif err != nil {\n\t\t\t// If error contains \"API key\", it's expected in CI without keys\n\t\t\tif strings.Contains(err.Error(), \"API key\") || strings.Contains(err.Error(), \"api_key\") {\n\t\t\t\tt.Logf(\"Expected error without API key: %v\", err)\n\t\t\t\treturn\n\t\t\t}\n\t\t\t// Other errors should fail\n\t\t\trequire.NoError(t, err)\n\t\t}\n\n\t\trequire.NotNil(t, response)\n\t\tassert.NotNil(t, response.Completion, \"should have completion\")\n\t\tt.Logf(\"✓ Stream executed with full search config (Web + KB + DB)\")\n\t})\n}\n" }, { "path": "agent/assistant/search_auto_hook_disable_test.go", "content": "package assistant_test\n\nimport (\n\tstdContext \"context\"\n\t\"testing\"\n\n\t\"github.com/stretchr/testify/assert\"\n\t\"github.com/stretchr/testify/require\"\n\t\"github.com/yaoapp/yao/agent/assistant\"\n\t\"github.com/yaoapp/yao/agent/context\"\n\t\"github.com/yaoapp/yao/agent/output/message\"\n\t\"github.com/yaoapp/yao/agent/testutils\"\n\t\"github.com/yaoapp/yao/openapi/oauth/types\"\n)\n\n// newSearchAutoHookDisableTestContext creates a test context\nfunc newSearchAutoHookDisableTestContext(chatID, assistantID string) *context.Context {\n\tauthorized := &types.AuthorizedInfo{\n\t\tSubject: \"test-user\",\n\t\tUserID: \"test-123\",\n\t\tTenantID: \"test-tenant\",\n\t}\n\n\tctx := context.New(stdContext.Background(), authorized, chatID)\n\tctx.ID = chatID\n\tctx.AssistantID = assistantID\n\tctx.Locale = \"en-us\"\n\tctx.Client = context.Client{\n\t\tType: \"web\",\n\t\tIP: \"127.0.0.1\",\n\t}\n\tctx.Referer = context.RefererAPI\n\tctx.Accept = context.AcceptWebCUI\n\tctx.IDGenerator = message.NewIDGenerator()\n\tctx.Metadata = make(map[string]interface{})\n\treturn ctx\n}\n\nfunc TestSearchAutoHookDisable(t *testing.T) {\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\tast, err := assistant.LoadPath(\"/assistants/tests/search-auto-hook-disable\")\n\trequire.NoError(t, err)\n\trequire.NotNil(t, ast)\n\n\tt.Run(\"ShouldHaveSearchConfigEnabled\", func(t *testing.T) {\n\t\t// Search config is enabled in package.yao\n\t\tassert.NotNil(t, ast.Search, \"search config should be set\")\n\t\tassert.NotNil(t, ast.Uses, \"uses config should be set\")\n\t\tassert.Equal(t, \"builtin\", ast.Uses.Search, \"uses.search should be builtin in config\")\n\t})\n\n\tt.Run(\"ShouldHaveHookScript\", func(t *testing.T) {\n\t\t// Hook script should be loaded\n\t\tassert.NotNil(t, ast.HookScript, \"hook script should be loaded\")\n\t})\n\n\tt.Run(\"HookShouldDisableSearch\", func(t *testing.T) {\n\t\t// Create context\n\t\tctx := newSearchAutoHookDisableTestContext(\"test-chat-id\", \"tests.search-auto-hook-disable\")\n\n\t\t// Create messages\n\t\tmessages := []context.Message{\n\t\t\t{\n\t\t\t\tRole: \"user\",\n\t\t\t\tContent: \"Test message\",\n\t\t\t},\n\t\t}\n\n\t\t// Call Create hook directly\n\t\topts := &context.Options{}\n\t\tresponse, _, err := ast.HookScript.Create(ctx, messages, opts)\n\t\trequire.NoError(t, err)\n\t\trequire.NotNil(t, response)\n\n\t\t// Verify hook returns uses.search = \"disabled\"\n\t\tassert.NotNil(t, response.Uses, \"hook should return uses\")\n\t\tassert.Equal(t, \"disabled\", response.Uses.Search, \"hook should disable search\")\n\t})\n\n\tt.Run(\"StreamShouldRespectHookDisable\", func(t *testing.T) {\n\t\t// Get agent via assistant.Get (required for Stream)\n\t\tagent, err := assistant.Get(\"tests.search-auto-hook-disable\")\n\t\trequire.NoError(t, err)\n\t\trequire.NotNil(t, agent)\n\n\t\t// Create context\n\t\tctx := newSearchAutoHookDisableTestContext(\"test-search-hook-disable\", \"tests.search-auto-hook-disable\")\n\n\t\t// Create messages\n\t\tmessages := []context.Message{\n\t\t\t{\n\t\t\t\tRole: \"user\",\n\t\t\t\tContent: \"What is AI?\",\n\t\t\t},\n\t\t}\n\n\t\t// Execute stream - hook will disable search\n\t\tresponse, err := agent.Stream(ctx, messages)\n\t\trequire.NoError(t, err)\n\t\trequire.NotNil(t, response)\n\n\t\tassert.NotNil(t, response.Completion, \"should have completion\")\n\t\tt.Logf(\"✓ Stream executed with hook disabling search\")\n\t})\n}\n" }, { "path": "agent/assistant/search_auto_keyword_test.go", "content": "package assistant_test\n\nimport (\n\tstdContext \"context\"\n\t\"strings\"\n\t\"testing\"\n\n\t\"github.com/stretchr/testify/assert\"\n\t\"github.com/stretchr/testify/require\"\n\t\"github.com/yaoapp/yao/agent/assistant\"\n\t\"github.com/yaoapp/yao/agent/context\"\n\t\"github.com/yaoapp/yao/agent/output/message\"\n\t\"github.com/yaoapp/yao/agent/testutils\"\n\t\"github.com/yaoapp/yao/openapi/oauth/types\"\n)\n\n// newKeywordTestContext creates a test context for keyword extraction tests\nfunc newKeywordTestContext(chatID, assistantID string) *context.Context {\n\tauthorized := &types.AuthorizedInfo{\n\t\tSubject: \"test-user\",\n\t\tUserID: \"test-123\",\n\t\tTenantID: \"test-tenant\",\n\t}\n\n\tctx := context.New(stdContext.Background(), authorized, chatID)\n\tctx.ID = chatID\n\tctx.AssistantID = assistantID\n\tctx.Locale = \"en-us\"\n\tctx.Client = context.Client{\n\t\tType: \"web\",\n\t\tIP: \"127.0.0.1\",\n\t}\n\tctx.Referer = context.RefererAPI\n\tctx.Accept = context.AcceptWebCUI\n\tctx.IDGenerator = message.NewIDGenerator()\n\tctx.Metadata = make(map[string]interface{})\n\treturn ctx\n}\n\nfunc TestSearchAutoKeyword(t *testing.T) {\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\tast, err := assistant.LoadPath(\"/assistants/tests/search-auto-keyword\")\n\trequire.NoError(t, err)\n\trequire.NotNil(t, ast)\n\n\tt.Run(\"ShouldHaveKeywordConfig\", func(t *testing.T) {\n\t\tassert.NotNil(t, ast.Search, \"search config should be set\")\n\t\tassert.NotNil(t, ast.Search.Keyword, \"keyword config should be set\")\n\t\tassert.Equal(t, 5, ast.Search.Keyword.MaxKeywords)\n\t\tassert.Equal(t, \"auto\", ast.Search.Keyword.Language)\n\t})\n\n\tt.Run(\"ShouldHaveKeywordInUses\", func(t *testing.T) {\n\t\tassert.NotNil(t, ast.Uses, \"uses config should be set\")\n\t\tassert.Equal(t, \"builtin\", ast.Uses.Keyword)\n\t})\n\n\tt.Run(\"StreamWithKeywordExtraction\", func(t *testing.T) {\n\t\t// Get agent via assistant.Get (required for Stream)\n\t\tagent, err := assistant.Get(\"tests.search-auto-keyword\")\n\t\trequire.NoError(t, err)\n\t\trequire.NotNil(t, agent)\n\n\t\t// Create context\n\t\tctx := newKeywordTestContext(\"test-search-keyword\", \"tests.search-auto-keyword\")\n\n\t\t// Create messages with a verbose query that should benefit from keyword extraction\n\t\tmessages := []context.Message{\n\t\t\t{\n\t\t\t\tRole: \"user\",\n\t\t\t\tContent: \"I want to find the best wireless headphones under 100 dollars for programming and music\",\n\t\t\t},\n\t\t}\n\n\t\t// Execute stream without Skip.Keyword (keyword extraction should happen)\n\t\tresponse, err := agent.Stream(ctx, messages)\n\n\t\t// Assert no error (if API key is configured)\n\t\tif err != nil {\n\t\t\tif strings.Contains(err.Error(), \"API key\") || strings.Contains(err.Error(), \"api_key\") {\n\t\t\t\tt.Logf(\"Expected error without API key: %v\", err)\n\t\t\t\treturn\n\t\t\t}\n\t\t\trequire.NoError(t, err)\n\t\t}\n\n\t\trequire.NotNil(t, response)\n\t\tassert.NotNil(t, response.Completion, \"should have completion\")\n\t\tt.Logf(\"✓ Stream with keyword extraction executed successfully\")\n\t})\n\n\tt.Run(\"StreamWithSkipKeyword\", func(t *testing.T) {\n\t\t// Get agent via assistant.Get (required for Stream)\n\t\tagent, err := assistant.Get(\"tests.search-auto-keyword\")\n\t\trequire.NoError(t, err)\n\t\trequire.NotNil(t, agent)\n\n\t\t// Create context\n\t\tctx := newKeywordTestContext(\"test-search-skip-keyword\", \"tests.search-auto-keyword\")\n\n\t\t// Create messages\n\t\tmessages := []context.Message{\n\t\t\t{\n\t\t\t\tRole: \"user\",\n\t\t\t\tContent: \"I want to find the best wireless headphones under 100 dollars\",\n\t\t\t},\n\t\t}\n\n\t\t// Execute stream with Skip.Keyword = true (keyword extraction should be skipped)\n\t\topts := &context.Options{\n\t\t\tSkip: &context.Skip{\n\t\t\t\tKeyword: true,\n\t\t\t},\n\t\t}\n\t\tresponse, err := agent.Stream(ctx, messages, opts)\n\n\t\t// Assert no error (if API key is configured)\n\t\tif err != nil {\n\t\t\tif strings.Contains(err.Error(), \"API key\") || strings.Contains(err.Error(), \"api_key\") {\n\t\t\t\tt.Logf(\"Expected error without API key: %v\", err)\n\t\t\t\treturn\n\t\t\t}\n\t\t\trequire.NoError(t, err)\n\t\t}\n\n\t\trequire.NotNil(t, response)\n\t\tassert.NotNil(t, response.Completion, \"should have completion\")\n\t\tt.Logf(\"✓ Stream with Skip.Keyword executed successfully\")\n\t})\n}\n\nfunc TestSearchAutoKeywordNotConfigured(t *testing.T) {\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\t// Use the search-auto-web assistant which does NOT have uses.keyword configured\n\tast, err := assistant.LoadPath(\"/assistants/tests/search-auto-web\")\n\trequire.NoError(t, err)\n\trequire.NotNil(t, ast)\n\n\tt.Run(\"ShouldNotHaveKeywordInUses\", func(t *testing.T) {\n\t\t// uses.keyword should be empty (not configured)\n\t\tif ast.Uses != nil {\n\t\t\tassert.Empty(t, ast.Uses.Keyword, \"uses.keyword should be empty\")\n\t\t}\n\t})\n\n\tt.Run(\"StreamShouldSkipKeywordExtraction\", func(t *testing.T) {\n\t\t// Get agent via assistant.Get (required for Stream)\n\t\tagent, err := assistant.Get(\"tests.search-auto-web\")\n\t\trequire.NoError(t, err)\n\t\trequire.NotNil(t, agent)\n\n\t\t// Create context\n\t\tctx := newKeywordTestContext(\"test-no-keyword\", \"tests.search-auto-web\")\n\n\t\t// Create messages\n\t\tmessages := []context.Message{\n\t\t\t{\n\t\t\t\tRole: \"user\",\n\t\t\t\tContent: \"What is the latest news about AI?\",\n\t\t\t},\n\t\t}\n\n\t\t// Execute stream - keyword extraction should NOT happen because uses.keyword is not set\n\t\tresponse, err := agent.Stream(ctx, messages)\n\n\t\t// Assert no error (if API key is configured)\n\t\tif err != nil {\n\t\t\tif strings.Contains(err.Error(), \"API key\") || strings.Contains(err.Error(), \"api_key\") {\n\t\t\t\tt.Logf(\"Expected error without API key: %v\", err)\n\t\t\t\treturn\n\t\t\t}\n\t\t\trequire.NoError(t, err)\n\t\t}\n\n\t\trequire.NotNil(t, response)\n\t\tassert.NotNil(t, response.Completion, \"should have completion\")\n\t\tt.Logf(\"✓ Stream without keyword config executed successfully\")\n\t})\n}\n" }, { "path": "agent/assistant/search_auto_web_test.go", "content": "package assistant_test\n\nimport (\n\tstdContext \"context\"\n\t\"strings\"\n\t\"testing\"\n\n\t\"github.com/stretchr/testify/assert\"\n\t\"github.com/stretchr/testify/require\"\n\t\"github.com/yaoapp/yao/agent/assistant\"\n\t\"github.com/yaoapp/yao/agent/context\"\n\t\"github.com/yaoapp/yao/agent/output/message\"\n\t\"github.com/yaoapp/yao/agent/testutils\"\n\t\"github.com/yaoapp/yao/openapi/oauth/types\"\n)\n\n// newSearchAutoTestContext creates a test context for search auto tests\nfunc newSearchAutoTestContext(chatID, assistantID string) *context.Context {\n\tauthorized := &types.AuthorizedInfo{\n\t\tSubject: \"test-user\",\n\t\tUserID: \"test-123\",\n\t\tTenantID: \"test-tenant\",\n\t}\n\n\tctx := context.New(stdContext.Background(), authorized, chatID)\n\tctx.ID = chatID\n\tctx.AssistantID = assistantID\n\tctx.Locale = \"en-us\"\n\tctx.Client = context.Client{\n\t\tType: \"web\",\n\t\tIP: \"127.0.0.1\",\n\t}\n\tctx.Referer = context.RefererAPI\n\tctx.Accept = context.AcceptWebCUI\n\tctx.IDGenerator = message.NewIDGenerator()\n\tctx.Metadata = make(map[string]interface{})\n\treturn ctx\n}\n\nfunc TestSearchAutoWeb(t *testing.T) {\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\tast, err := assistant.LoadPath(\"/assistants/tests/search-auto-web\")\n\trequire.NoError(t, err)\n\trequire.NotNil(t, ast)\n\n\tt.Run(\"ShouldHaveSearchConfig\", func(t *testing.T) {\n\t\tassert.NotNil(t, ast.Search, \"search config should be set\")\n\t\tassert.NotNil(t, ast.Search.Web, \"web search config should be set\")\n\t\tassert.Equal(t, \"tavily\", ast.Search.Web.Provider)\n\t\tassert.Equal(t, 3, ast.Search.Web.MaxResults)\n\t})\n\n\tt.Run(\"ShouldHaveUsesConfig\", func(t *testing.T) {\n\t\tassert.NotNil(t, ast.Uses, \"uses config should be set\")\n\t\tassert.Equal(t, \"builtin\", ast.Uses.Search)\n\t\tassert.Equal(t, \"builtin\", ast.Uses.Web)\n\t})\n\n\tt.Run(\"ShouldHaveCitationConfig\", func(t *testing.T) {\n\t\tassert.NotNil(t, ast.Search.Citation, \"citation config should be set\")\n\t\tassert.Equal(t, \"xml\", ast.Search.Citation.Format)\n\t\tassert.True(t, ast.Search.Citation.AutoInjectPrompt)\n\t})\n\n\tt.Run(\"StreamShouldExecuteAutoSearch\", func(t *testing.T) {\n\t\t// Get agent via assistant.Get (required for Stream)\n\t\tagent, err := assistant.Get(\"tests.search-auto-web\")\n\t\trequire.NoError(t, err)\n\t\trequire.NotNil(t, agent)\n\n\t\t// Create context\n\t\tctx := newSearchAutoTestContext(\"test-search-auto-web\", \"tests.search-auto-web\")\n\n\t\t// Create messages with a search query\n\t\tmessages := []context.Message{\n\t\t\t{\n\t\t\t\tRole: \"user\",\n\t\t\t\tContent: \"What is the latest news about artificial intelligence?\",\n\t\t\t},\n\t\t}\n\n\t\t// Execute stream\n\t\tresponse, err := agent.Stream(ctx, messages)\n\n\t\t// Assert no error (if API key is configured)\n\t\tif err != nil {\n\t\t\t// If error contains \"API key\", it's expected in CI without keys\n\t\t\tif strings.Contains(err.Error(), \"API key\") || strings.Contains(err.Error(), \"api_key\") {\n\t\t\t\tt.Logf(\"Expected error without API key: %v\", err)\n\t\t\t\treturn\n\t\t\t}\n\t\t\t// Other errors should fail\n\t\t\trequire.NoError(t, err)\n\t\t}\n\n\t\trequire.NotNil(t, response)\n\t\tassert.NotNil(t, response.Completion, \"should have completion\")\n\t\tt.Logf(\"✓ Stream executed successfully with auto search\")\n\t})\n}\n" }, { "path": "agent/assistant/source.go", "content": "package assistant\n\nimport (\n\t\"fmt\"\n\t\"strings\"\n\t\"time\"\n\n\tv8 \"github.com/yaoapp/gou/runtime/v8\"\n\t\"github.com/yaoapp/yao/agent/assistant/hook\"\n)\n\n// loadSource loads hook script from source code string\n// The source field stores TypeScript code directly (but without imports)\n// Priority: script field > source field (if script exists, source is ignored)\n// Note: Uses MakeScriptInMemory which supports TypeScript syntax without file resolution.\nfunc loadSource(source string, assistantID string) (*hook.Script, error) {\n\tif source == \"\" {\n\t\treturn nil, nil\n\t}\n\n\t// Use virtual .ts path for TypeScript support\n\t// MakeScriptInMemory handles TypeScript transform without file system access\n\tvirtualFile := fmt.Sprintf(\"assistants/%s/source.ts\", strings.ReplaceAll(assistantID, \".\", \"/\"))\n\n\tscript, err := v8.MakeScriptInMemory([]byte(source), virtualFile, 5*time.Second, true)\n\tif err != nil {\n\t\treturn nil, fmt.Errorf(\"failed to compile source script: %w\", err)\n\t}\n\n\treturn &hook.Script{Script: script}, nil\n}\n\n// TODO: Future enhancement - support multiple files merged with special comment delimiter\n// Format: // file: index.ts\n// This would allow splitting large scripts into multiple logical files while storing as single source\n// func loadSourceMultiFile(source string, assistantID string) (*hook.Script, error) {\n// // Parse source by \"// file: xxx.ts\" delimiter\n// // Merge and compile\n// }\n" }, { "path": "agent/assistant/trace.go", "content": "package assistant\n\nimport (\n\t\"fmt\"\n\n\t\"github.com/yaoapp/gou/connector/openai\"\n\t\"github.com/yaoapp/kun/log\"\n\t\"github.com/yaoapp/yao/agent/context\"\n\t\"github.com/yaoapp/yao/agent/i18n\"\n\t\"github.com/yaoapp/yao/trace/types\"\n)\n\n// initAgentTraceNode creates and returns the agent trace node\nfunc (ast *Assistant) initAgentTraceNode(ctx *context.Context, inputMessages []context.Message) types.Node {\n\ttrace, _ := ctx.Trace()\n\tif trace == nil {\n\t\treturn nil\n\t}\n\n\tagentNode, _ := trace.Add(inputMessages, types.TraceNodeOption{\n\t\tLabel: i18n.Tr(ast.ID, ctx.Locale, \"assistant.agent.stream.label\"), // \"Assistant {{name}}\"\n\t\tType: \"agent\",\n\t\tIcon: \"assistant\",\n\t\tDescription: i18n.Tr(ast.ID, ctx.Locale, \"assistant.agent.stream.description\"), // \"Assistant {{name}} is processing the request\"\n\t})\n\n\treturn agentNode\n}\n\n// traceAgentHistory logs the chat history to the agent trace node\nfunc (ast *Assistant) traceAgentHistory(ctx *context.Context, agentNode types.Node, fullMessages []context.Message) {\n\tif agentNode == nil {\n\t\treturn\n\t}\n\n\tagentNode.Info(\n\t\ti18n.Tr(ast.ID, ctx.Locale, \"assistant.agent.stream.history\"), // \"Get Chat History\"\n\t\tmap[string]any{\"messages\": fullMessages},\n\t)\n}\n\n// traceCreateHook logs the create hook response to the agent trace node\nfunc (ast *Assistant) traceCreateHook(agentNode types.Node, createResponse *context.HookCreateResponse) {\n\tif agentNode == nil {\n\t\treturn\n\t}\n\n\tagentNode.Debug(\"Call Create Hook\", map[string]any{\"response\": createResponse})\n}\n\n// traceConnectorCapabilities logs the connector capabilities to the agent trace node\nfunc (ast *Assistant) traceConnectorCapabilities(agentNode types.Node, capabilities *openai.Capabilities) {\n\tif agentNode == nil {\n\t\treturn\n\t}\n\n\tagentNode.Debug(\"Get Connector Capabilities\", map[string]any{\"capabilities\": capabilities})\n}\n\n// traceLLMRequest adds a LLM trace node to the trace\nfunc (ast *Assistant) traceLLMRequest(ctx *context.Context, connID string, completionMessages []context.Message, completionOptions *context.CompletionOptions) {\n\ttrace, _ := ctx.Trace()\n\tif trace == nil {\n\t\treturn\n\t}\n\n\ttrace.Add(\n\t\tmap[string]any{\"messages\": completionMessages, \"options\": completionOptions},\n\t\ttypes.TraceNodeOption{\n\t\t\tLabel: fmt.Sprintf(i18n.Tr(ast.ID, ctx.Locale, \"llm.openai.stream.label\"), connID), // \"LLM %s\"\n\t\t\tType: \"llm\",\n\t\t\tIcon: \"psychology\",\n\t\t\tDescription: fmt.Sprintf(i18n.Tr(ast.ID, ctx.Locale, \"llm.openai.stream.description\"), connID), // \"LLM %s is processing the request\"\n\t\t},\n\t)\n}\n\n// traceLLMComplete marks the LLM request as complete in the trace\nfunc (ast *Assistant) traceLLMComplete(ctx *context.Context, completionResponse *context.CompletionResponse) {\n\ttrace, _ := ctx.Trace()\n\tif trace == nil {\n\t\treturn\n\t}\n\n\ttrace.Complete(completionResponse)\n}\n\n// traceLLMFail marks the LLM request as failed in the trace\nfunc (ast *Assistant) traceLLMFail(ctx *context.Context, err error) {\n\ttrace, _ := ctx.Trace()\n\tif trace == nil {\n\t\treturn\n\t}\n\n\ttrace.Fail(err)\n}\n\n// traceAgentCompletion creates a completion node to report the final output\nfunc (ast *Assistant) traceAgentCompletion(ctx *context.Context, createResponse *context.HookCreateResponse, nextResponse *context.NextHookResponse, completionResponse *context.CompletionResponse, finalResponse interface{}) {\n\ttrace, _ := ctx.Trace()\n\tif trace == nil {\n\t\treturn\n\t}\n\n\t// Prepare the input data (the raw responses before processing)\n\tinput := map[string]interface{}{\n\t\t\"create\": createResponse,\n\t\t\"next\": nextResponse,\n\t\t\"completion\": completionResponse,\n\t}\n\n\t// Create a dedicated completion node\n\tcompletionNode, err := trace.Add(\n\t\tinput,\n\t\ttypes.TraceNodeOption{\n\t\t\tLabel: i18n.Tr(ast.ID, ctx.Locale, \"assistant.agent.completion.label\"), // \"Agent Completion\"\n\t\t\tType: \"agent_completion\",\n\t\t\tIcon: \"check_circle\",\n\t\t\tDescription: i18n.Tr(ast.ID, ctx.Locale, \"assistant.agent.completion.description\"), // \"Final output from assistant\"\n\t\t},\n\t)\n\tif err != nil {\n\t\tlog.Trace(\"[TRACE] Failed to create completion node: %v\", err)\n\t\treturn\n\t}\n\n\t// Immediately mark it as complete with the final response\n\tif completionNode != nil {\n\t\tcompletionNode.Complete(finalResponse)\n\t}\n}\n\n// traceAgentOutput sets the output of the agent trace node\n// Deprecated: Use traceAgentCompletion instead for better trace structure\nfunc (ast *Assistant) traceAgentOutput(agentNode types.Node, createResponse *context.HookCreateResponse, nextResponse interface{}, completionResponse *context.CompletionResponse) {\n\tif agentNode == nil {\n\t\treturn\n\t}\n\n\toutput := context.Response{\n\t\tCreate: createResponse,\n\t\tNext: nextResponse,\n\t\tCompletion: completionResponse,\n\t}\n\n\tagentNode.Complete(output)\n}\n\n// traceAgentFail marks the agent trace node as failed\nfunc (ast *Assistant) traceAgentFail(agentNode types.Node, err error) {\n\tif agentNode == nil {\n\t\treturn\n\t}\n\n\tagentNode.Fail(err)\n}\n\n// traceLLMRetryRequest adds a LLM retry trace node to the trace\nfunc (ast *Assistant) traceLLMRetryRequest(ctx *context.Context, connID string, completionMessages []context.Message, completionOptions *context.CompletionOptions) {\n\ttrace, _ := ctx.Trace()\n\tif trace == nil {\n\t\treturn\n\t}\n\n\ttrace.Add(\n\t\tmap[string]any{\"messages\": completionMessages, \"options\": completionOptions},\n\t\ttypes.TraceNodeOption{\n\t\t\tLabel: fmt.Sprintf(\"LLM %s (Tool Retry)\", connID),\n\t\t\tType: \"llm_retry\",\n\t\t\tIcon: \"refresh\",\n\t\t\tDescription: fmt.Sprintf(\"LLM %s is retrying with tool call error feedback\", connID),\n\t\t},\n\t)\n}\n" }, { "path": "agent/assistant/types.go", "content": "package assistant\n\nimport (\n\tjsoniter \"github.com/json-iterator/go\"\n\tv8 \"github.com/yaoapp/gou/runtime/v8\"\n\t\"github.com/yaoapp/yao/agent/assistant/hook\"\n\tchatctx \"github.com/yaoapp/yao/agent/context\"\n\n\toutputMessage \"github.com/yaoapp/yao/agent/output/message\"\n\tstore \"github.com/yaoapp/yao/agent/store/types\"\n)\n\nconst (\n\t// HookErrorMethodNotFound is the error message for method not found\n\tHookErrorMethodNotFound = \"method not found\"\n)\n\n// API the assistant API interface\ntype API interface {\n\tGetPlaceholder(locale string) *store.Placeholder\n}\n\n// Script the script scripts except hook script\ntype Script struct {\n\t*v8.Script\n}\n\n// Assistant the assistant\ntype Assistant struct {\n\tstore.AssistantModel\n\tHookScript *hook.Script `json:\"-\" yaml:\"-\"` // Hook Script (index.ts)\n\tScripts map[string]*Script `json:\"-\" yaml:\"-\"` // Other scripts\n\n\t// Internal\n\t// ===============================\n\tvision bool // Whether this assistant supports vision\n}\n\n// MCPTool represents a simplified MCP tool for building LLM requests\n// This is an internal representation used when collecting tools from MCP servers\n// and preparing them for the LLM's tool calling interface\ntype MCPTool struct {\n\tName string // Formatted tool name with server prefix (e.g., \"server_id__tool_name\")\n\tDescription string // Tool description from MCP server\n\tParameters interface{} // JSON Schema for tool parameters (from MCP InputSchema)\n}\n\n// ToolCallResult represents the result of a tool call execution\n// Used to track the outcome of MCP tool invocations during agent execution\ntype ToolCallResult struct {\n\tToolCallID string // Tool call ID from the LLM (matches the ID in the LLM's tool_calls response)\n\tName string // Tool name (formatted with server prefix, e.g., \"server_id__tool_name\")\n\tContent string // Result content (JSON string of the tool's output or error message)\n\tError error // Error if the call failed (nil if successful)\n\tIsRetryableError bool // Whether the error should be sent to LLM for retry\n\t// true: parameter/validation errors that LLM can fix (e.g., \"missing required field\")\n\t// false: MCP internal errors that LLM cannot fix (e.g., \"network error\", \"service unavailable\")\n}\n\n// Server extracts the MCP server ID from the formatted tool name\n// Example: \"echo__ping\" -> \"echo\"\nfunc (r *ToolCallResult) Server() string {\n\tserverID, _, _ := ParseMCPToolName(r.Name)\n\treturn serverID\n}\n\n// Tool extracts the original tool name without server prefix\n// Example: \"echo__ping\" -> \"ping\"\nfunc (r *ToolCallResult) Tool() string {\n\t_, toolName, _ := ParseMCPToolName(r.Name)\n\treturn toolName\n}\n\n// NextProcessContext encapsulates all the context needed to process Next hook responses\n// This simplifies function signatures and makes it easier to add new fields in the future\ntype NextProcessContext struct {\n\tContext *chatctx.Context // Agent context\n\tNextResponse *chatctx.NextHookResponse // Response from Next hook (already converted from JS)\n\tCompletionResponse *chatctx.CompletionResponse // LLM completion response\n\tFullMessages []chatctx.Message // Full conversation history\n\tToolCallResponses []chatctx.ToolCallResponse // Tool call results (if any)\n\tStreamHandler outputMessage.StreamFunc // Stream handler for output\n\tCreateResponse *chatctx.HookCreateResponse // Create hook response\n}\n\n// SearchIntent is an alias for context.SearchIntent\n// Used for search intent detection from __yao.needsearch agent\ntype SearchIntent = chatctx.SearchIntent\n\n// ParsedContent extracts the actual tool return value from MCP ToolContent array\n// According to MCP protocol:\n// - Content is []ToolContent array\n// - For \"text\" type, the actual value is in Text field (usually JSON string)\n// - For \"image\" type, returns the Data field\n// - For \"resource\" type, returns the Resource object\n// If there are multiple content items, returns an array of parsed values\nfunc (r *ToolCallResult) ParsedContent() (interface{}, error) {\n\tif r.Content == \"\" {\n\t\treturn nil, nil\n\t}\n\n\t// Parse Content as []ToolContent\n\tvar toolContents []map[string]interface{}\n\tif err := jsoniter.UnmarshalFromString(r.Content, &toolContents); err != nil {\n\t\t// If parsing fails, return the string content directly (error message)\n\t\treturn r.Content, nil\n\t}\n\n\t// Extract actual values from ToolContent items\n\tvar results []interface{}\n\tfor _, tc := range toolContents {\n\t\tcontentType, _ := tc[\"type\"].(string)\n\n\t\tswitch contentType {\n\t\tcase \"text\":\n\t\t\t// For text type, parse the Text field (usually JSON)\n\t\t\tif textStr, ok := tc[\"text\"].(string); ok {\n\t\t\t\t// Try to parse as JSON\n\t\t\t\tvar parsed interface{}\n\t\t\t\tif err := jsoniter.UnmarshalFromString(textStr, &parsed); err == nil {\n\t\t\t\t\tresults = append(results, parsed)\n\t\t\t\t} else {\n\t\t\t\t\t// If not JSON, return as plain string\n\t\t\t\t\tresults = append(results, textStr)\n\t\t\t\t}\n\t\t\t}\n\t\tcase \"image\":\n\t\t\t// For image type, return the data and mimeType\n\t\t\tresults = append(results, map[string]interface{}{\n\t\t\t\t\"type\": \"image\",\n\t\t\t\t\"data\": tc[\"data\"],\n\t\t\t\t\"mimeType\": tc[\"mimeType\"],\n\t\t\t})\n\t\tcase \"resource\":\n\t\t\t// For resource type, return the resource object\n\t\t\tresults = append(results, tc[\"resource\"])\n\t\tdefault:\n\t\t\t// Unknown type, return as-is\n\t\t\tresults = append(results, tc)\n\t\t}\n\t}\n\n\t// If only one result, return it directly (not as array)\n\tif len(results) == 1 {\n\t\treturn results[0], nil\n\t}\n\n\treturn results, nil\n}\n" }, { "path": "agent/assistant/utils.go", "content": "package assistant\n\nimport (\n\t\"crypto/sha256\"\n\t\"encoding/hex\"\n\t\"fmt\"\n\t\"strconv\"\n\t\"time\"\n\n\tjsoniter \"github.com/json-iterator/go\"\n\t\"github.com/kaptinlin/jsonrepair\"\n)\n\nfunc getTimestamp(v interface{}) (int64, error) {\n\tswitch v := v.(type) {\n\tcase int64:\n\t\treturn v, nil\n\tcase int:\n\t\treturn int64(v), nil\n\n\tcase string:\n\t\tif ts, err := time.Parse(time.RFC3339, v); err == nil {\n\t\t\treturn ts.UnixNano(), nil\n\t\t}\n\n\t\t// MySQL format\n\t\tif ts, err := time.Parse(\"2006-01-02 15:04:05\", v); err == nil {\n\t\t\treturn ts.UnixNano(), nil\n\t\t}\n\n\t\t// UnixNano format\n\t\tif ts, err := strconv.ParseInt(v, 10, 64); err == nil {\n\t\t\treturn ts, nil\n\t\t}\n\n\tcase time.Time:\n\t\treturn v.UnixNano(), nil\n\n\tcase nil:\n\t\treturn 0, nil\n\t}\n\n\treturn 0, fmt.Errorf(\"invalid timestamp type %T\", v)\n}\n\n// getBool gets bool from data map[string]interface{}, key string\nfunc getBool(data map[string]interface{}, key string) bool {\n\tswitch v := data[key].(type) {\n\tcase bool:\n\t\treturn v\n\tcase int64:\n\t\treturn v != 0\n\tcase int:\n\t\treturn v != 0\n\tcase float64:\n\t\treturn v != 0\n\tcase string:\n\t\treturn v == \"true\" || v == \"1\" || v == \"enabled\" || v == \"yes\" || v == \"on\"\n\tcase nil:\n\t\treturn false\n\t}\n\treturn false\n}\n\n// stringHash returns the sha256 hash of the string\nfunc stringHash(v string) string {\n\th := sha256.New()\n\th.Write([]byte(v))\n\treturn hex.EncodeToString(h.Sum(nil))\n}\n\n// ParseJSON attempts to parse a potentially malformed JSON string\nfunc ParseJSON(jsonStr string, v interface{}) error {\n\t// Try parsing as-is first\n\terr := jsoniter.UnmarshalFromString(jsonStr, v)\n\tif err == nil {\n\t\treturn nil\n\t}\n\toriginalErr := err\n\n\t// Try adding a closing brace\n\tif err := jsoniter.UnmarshalFromString(jsonStr+\"}\", v); err == nil {\n\t\treturn nil\n\t}\n\n\t// Try repairing the JSON\n\trepaired, err := jsonrepair.JSONRepair(jsonStr)\n\tif err != nil {\n\t\treturn originalErr\n\t}\n\n\t// Try parsing the repaired JSON\n\tif err := jsoniter.UnmarshalFromString(repaired, v); err == nil {\n\t\treturn nil\n\t}\n\n\t// If all attempts fail, return the original error\n\treturn originalErr\n}\n" }, { "path": "agent/caller/caller.go", "content": "// Package caller provides a shared interface for calling agents\n// This package is used by both content and search packages to avoid circular dependencies\npackage caller\n\nimport (\n\tagentContext \"github.com/yaoapp/yao/agent/context\"\n)\n\n// AgentCaller interface for calling agents (to avoid circular dependency)\n// Used by content handlers (vision, audio, etc.) and search handlers (agent mode)\ntype AgentCaller interface {\n\tStream(ctx *agentContext.Context, messages []agentContext.Message, options ...*agentContext.Options) (*agentContext.Response, error)\n}\n\n// AgentGetterFunc is a function type that gets an agent by ID\n// This should be set by the assistant package during initialization\nvar AgentGetterFunc func(agentID string) (AgentCaller, error)\n" }, { "path": "agent/caller/context.go", "content": "package caller\n\nimport (\n\t\"context\"\n\n\tagentContext \"github.com/yaoapp/yao/agent/context\"\n\t\"github.com/yaoapp/yao/openapi/oauth/types\"\n)\n\n// NewHeadlessContext creates a headless agent context from a ProcessCallRequest.\n// This is the Process equivalent of openapi.GetCompletionRequest — constructs\n// a Context + Options without HTTP dependencies (no Writer, no Interrupt).\n//\n// Key behaviors:\n// - parent context controls timeout/cancellation (caller is responsible)\n// - skip.output = true (forced): no Writer available, must skip output\n// - skip.history = true (forced): Process calls don't save chat history\n// - authorized info is passed in (from authorized.ProcessAuthInfo by caller)\n// - chatID is auto-generated if not provided\nfunc NewHeadlessContext(parent context.Context, authInfo *types.AuthorizedInfo, req *ProcessCallRequest) (*agentContext.Context, *agentContext.Options) {\n\tchatID := req.ChatID\n\tif chatID == \"\" {\n\t\tchatID = agentContext.GenChatID()\n\t}\n\n\tctx := agentContext.New(parent, authInfo, chatID)\n\tctx.AssistantID = req.AssistantID\n\tctx.Referer = agentContext.RefererProcess\n\tctx.Locale = req.Locale\n\tctx.Route = req.Route\n\tctx.Metadata = req.Metadata\n\n\t// Force skip for headless context — no Writer, no chat history\n\tskip := req.Skip\n\tif skip == nil {\n\t\tskip = &agentContext.Skip{}\n\t}\n\tskip.Output = true // no Writer available\n\tskip.History = true // Process calls don't save chat history\n\n\topts := &agentContext.Options{Skip: skip}\n\tif req.Model != \"\" {\n\t\topts.Connector = req.Model\n\t}\n\n\treturn ctx, opts\n}\n" }, { "path": "agent/caller/integration_test.go", "content": "package caller_test\n\nimport (\n\t\"context\"\n\t\"testing\"\n\n\t\"github.com/stretchr/testify/assert\"\n\t\"github.com/stretchr/testify/require\"\n\t\"github.com/yaoapp/yao/agent/assistant\"\n\t\"github.com/yaoapp/yao/agent/caller\"\n\tagentContext \"github.com/yaoapp/yao/agent/context\"\n\t\"github.com/yaoapp/yao/agent/testutils\"\n\t\"github.com/yaoapp/yao/openapi/oauth/types\"\n)\n\nfunc TestIntegration_Call_RealAgent(t *testing.T) {\n\tif testing.Short() {\n\t\tt.Skip(\"Skipping integration test in short mode\")\n\t}\n\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\t// Load the simple-greeting agent\n\tast, err := assistant.Get(\"tests.simple-greeting\")\n\trequire.NoError(t, err)\n\trequire.NotNil(t, ast)\n\n\t// Create authorized info for the context\n\tauthorized := &types.AuthorizedInfo{\n\t\tSubject: \"test-user\",\n\t\tUserID: \"test-123\",\n\t\tTenantID: \"test-tenant\",\n\t}\n\n\t// Create a context with authorization\n\tctx := agentContext.New(context.Background(), authorized, \"test-chat-integration\")\n\tctx.AssistantID = \"tests.agent-caller\"\n\n\t// Create JSAPI\n\tapi := caller.NewJSAPI(ctx)\n\n\t// Call the simple-greeting agent\n\tmessages := []interface{}{\n\t\tmap[string]interface{}{\n\t\t\t\"role\": \"user\",\n\t\t\t\"content\": \"Hello!\",\n\t\t},\n\t}\n\n\topts := map[string]interface{}{\n\t\t\"skip\": map[string]interface{}{\n\t\t\t\"history\": true,\n\t\t},\n\t}\n\n\tresult := api.Call(\"tests.simple-greeting\", messages, opts)\n\trequire.NotNil(t, result)\n\n\tr, ok := result.(*caller.Result)\n\trequire.True(t, ok)\n\tassert.Equal(t, \"tests.simple-greeting\", r.AgentID)\n\n\t// Should either have content or error\n\tif r.Error != \"\" {\n\t\tt.Logf(\"Agent call error: %s\", r.Error)\n\t} else {\n\t\tt.Logf(\"Agent response content: %s\", r.Content)\n\t\tassert.NotEmpty(t, r.Content)\n\t}\n}\n\nfunc TestIntegration_All_RealAgents(t *testing.T) {\n\tif testing.Short() {\n\t\tt.Skip(\"Skipping integration test in short mode\")\n\t}\n\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\t// Create authorized info for the context\n\tauthorized := &types.AuthorizedInfo{\n\t\tSubject: \"test-user\",\n\t\tUserID: \"test-123\",\n\t\tTenantID: \"test-tenant\",\n\t}\n\n\t// Create a context with authorization\n\tctx := agentContext.New(context.Background(), authorized, \"test-chat-all\")\n\tctx.AssistantID = \"tests.agent-caller\"\n\n\t// Create JSAPI\n\tapi := caller.NewJSAPI(ctx)\n\n\t// Call multiple agents in parallel\n\trequests := []interface{}{\n\t\tmap[string]interface{}{\n\t\t\t\"agent\": \"tests.simple-greeting\",\n\t\t\t\"messages\": []interface{}{\n\t\t\t\tmap[string]interface{}{\n\t\t\t\t\t\"role\": \"user\",\n\t\t\t\t\t\"content\": \"Hello from test 1!\",\n\t\t\t\t},\n\t\t\t},\n\t\t\t\"options\": map[string]interface{}{\n\t\t\t\t\"skip\": map[string]interface{}{\n\t\t\t\t\t\"history\": true,\n\t\t\t\t},\n\t\t\t},\n\t\t},\n\t\tmap[string]interface{}{\n\t\t\t\"agent\": \"tests.simple-greeting\",\n\t\t\t\"messages\": []interface{}{\n\t\t\t\tmap[string]interface{}{\n\t\t\t\t\t\"role\": \"user\",\n\t\t\t\t\t\"content\": \"Hello from test 2!\",\n\t\t\t\t},\n\t\t\t},\n\t\t\t\"options\": map[string]interface{}{\n\t\t\t\t\"skip\": map[string]interface{}{\n\t\t\t\t\t\"history\": true,\n\t\t\t\t},\n\t\t\t},\n\t\t},\n\t}\n\n\tresults := api.All(requests)\n\trequire.Len(t, results, 2)\n\n\tfor i, result := range results {\n\t\tr, ok := result.(*caller.Result)\n\t\trequire.True(t, ok, \"result %d should be *caller.Result\", i)\n\t\tassert.Equal(t, \"tests.simple-greeting\", r.AgentID)\n\t\tt.Logf(\"Result[%d]: content=%s, error=%s\", i, r.Content, r.Error)\n\t}\n}\n\nfunc TestIntegration_Any_RealAgents(t *testing.T) {\n\tif testing.Short() {\n\t\tt.Skip(\"Skipping integration test in short mode\")\n\t}\n\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\t// Create authorized info for the context\n\tauthorized := &types.AuthorizedInfo{\n\t\tSubject: \"test-user\",\n\t\tUserID: \"test-123\",\n\t\tTenantID: \"test-tenant\",\n\t}\n\n\t// Create a context with authorization\n\tctx := agentContext.New(context.Background(), authorized, \"test-chat-any\")\n\tctx.AssistantID = \"tests.agent-caller\"\n\n\t// Create JSAPI\n\tapi := caller.NewJSAPI(ctx)\n\n\t// Call multiple agents - return when any succeeds\n\trequests := []interface{}{\n\t\tmap[string]interface{}{\n\t\t\t\"agent\": \"tests.simple-greeting\",\n\t\t\t\"messages\": []interface{}{\n\t\t\t\tmap[string]interface{}{\n\t\t\t\t\t\"role\": \"user\",\n\t\t\t\t\t\"content\": \"Hello from any test 1!\",\n\t\t\t\t},\n\t\t\t},\n\t\t\t\"options\": map[string]interface{}{\n\t\t\t\t\"skip\": map[string]interface{}{\n\t\t\t\t\t\"history\": true,\n\t\t\t\t},\n\t\t\t},\n\t\t},\n\t\tmap[string]interface{}{\n\t\t\t\"agent\": \"tests.simple-greeting\",\n\t\t\t\"messages\": []interface{}{\n\t\t\t\tmap[string]interface{}{\n\t\t\t\t\t\"role\": \"user\",\n\t\t\t\t\t\"content\": \"Hello from any test 2!\",\n\t\t\t\t},\n\t\t\t},\n\t\t\t\"options\": map[string]interface{}{\n\t\t\t\t\"skip\": map[string]interface{}{\n\t\t\t\t\t\"history\": true,\n\t\t\t\t},\n\t\t\t},\n\t\t},\n\t}\n\n\tresults := api.Any(requests)\n\trequire.Len(t, results, 2)\n\n\t// At least one should have a result\n\thasResult := false\n\tfor i, result := range results {\n\t\tif result != nil {\n\t\t\tr, ok := result.(*caller.Result)\n\t\t\tif ok && r != nil && r.Error == \"\" {\n\t\t\t\thasResult = true\n\t\t\t\tt.Logf(\"Any Result[%d]: content=%s\", i, r.Content)\n\t\t\t}\n\t\t}\n\t}\n\tassert.True(t, hasResult, \"At least one result should succeed\")\n}\n\nfunc TestIntegration_Race_RealAgents(t *testing.T) {\n\tif testing.Short() {\n\t\tt.Skip(\"Skipping integration test in short mode\")\n\t}\n\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\t// Create authorized info for the context\n\tauthorized := &types.AuthorizedInfo{\n\t\tSubject: \"test-user\",\n\t\tUserID: \"test-123\",\n\t\tTenantID: \"test-tenant\",\n\t}\n\n\t// Create a context with authorization\n\tctx := agentContext.New(context.Background(), authorized, \"test-chat-race\")\n\tctx.AssistantID = \"tests.agent-caller\"\n\n\t// Create JSAPI\n\tapi := caller.NewJSAPI(ctx)\n\n\t// Call multiple agents - return when any completes\n\trequests := []interface{}{\n\t\tmap[string]interface{}{\n\t\t\t\"agent\": \"tests.simple-greeting\",\n\t\t\t\"messages\": []interface{}{\n\t\t\t\tmap[string]interface{}{\n\t\t\t\t\t\"role\": \"user\",\n\t\t\t\t\t\"content\": \"Hello from race test 1!\",\n\t\t\t\t},\n\t\t\t},\n\t\t\t\"options\": map[string]interface{}{\n\t\t\t\t\"skip\": map[string]interface{}{\n\t\t\t\t\t\"history\": true,\n\t\t\t\t},\n\t\t\t},\n\t\t},\n\t\tmap[string]interface{}{\n\t\t\t\"agent\": \"tests.simple-greeting\",\n\t\t\t\"messages\": []interface{}{\n\t\t\t\tmap[string]interface{}{\n\t\t\t\t\t\"role\": \"user\",\n\t\t\t\t\t\"content\": \"Hello from race test 2!\",\n\t\t\t\t},\n\t\t\t},\n\t\t\t\"options\": map[string]interface{}{\n\t\t\t\t\"skip\": map[string]interface{}{\n\t\t\t\t\t\"history\": true,\n\t\t\t\t},\n\t\t\t},\n\t\t},\n\t}\n\n\tresults := api.Race(requests)\n\trequire.Len(t, results, 2)\n\n\t// At least one should have completed\n\thasResult := false\n\tfor i, result := range results {\n\t\tif result != nil {\n\t\t\tr, ok := result.(*caller.Result)\n\t\t\tif ok && r != nil {\n\t\t\t\thasResult = true\n\t\t\t\tt.Logf(\"Race Result[%d]: content=%s, error=%s\", i, r.Content, r.Error)\n\t\t\t}\n\t\t}\n\t}\n\tassert.True(t, hasResult, \"At least one result should complete\")\n}\n" }, { "path": "agent/caller/jsapi.go", "content": "package caller\n\nimport (\n\tagentContext \"github.com/yaoapp/yao/agent/context\"\n\t\"github.com/yaoapp/yao/agent/output/message\"\n)\n\n// JSAPI implements context.AgentAPI and context.AgentAPIWithCallback interfaces\n// Provides ctx.agent.Call(), ctx.agent.All(), ctx.agent.Any(), ctx.agent.Race()\n// and their *WithHandler variants for streaming callback support\ntype JSAPI struct {\n\tctx *agentContext.Context\n\torchestrator *Orchestrator\n}\n\n// Ensure JSAPI implements AgentAPIWithCallback\nvar _ agentContext.AgentAPIWithCallback = (*JSAPI)(nil)\n\n// NewJSAPI creates a new agent JSAPI instance\nfunc NewJSAPI(ctx *agentContext.Context) *JSAPI {\n\treturn &JSAPI{\n\t\tctx: ctx,\n\t\torchestrator: NewOrchestrator(ctx),\n\t}\n}\n\n// Call executes a single agent call\n// Usage: ctx.agent.Call(\"assistant-id\", messages, options?)\n// Returns: { agent_id, response, content, error }\n// Note: For sub-agent calls, skip.history = true is automatically set\n// to prevent A2A messages from being saved to chat history.\n// Sub-agents output normally with ThreadID for SSE stream isolation.\nfunc (api *JSAPI) Call(agentID string, messages []interface{}, opts map[string]interface{}) interface{} {\n\treq := api.buildRequest(agentID, messages, opts)\n\t// Force skip options for sub-agent calls\n\tapi.forceSkipForSubAgent(req)\n\tresult := api.orchestrator.callAgent(req)\n\treturn result\n}\n\n// All executes all agent calls and waits for all to complete (like Promise.all)\n// Each request should have:\n// - agent: string - target agent ID\n// - messages: array - messages to send\n// - options?: object - call options\nfunc (api *JSAPI) All(requests []interface{}) []interface{} {\n\treqs := api.parseRequests(requests)\n\tresults := api.orchestrator.All(reqs)\n\treturn api.convertResults(results)\n}\n\n// Any returns as soon as any agent call succeeds (like Promise.any)\n// Each request should have:\n// - agent: string - target agent ID\n// - messages: array - messages to send\n// - options?: object - call options\nfunc (api *JSAPI) Any(requests []interface{}) []interface{} {\n\treqs := api.parseRequests(requests)\n\tresults := api.orchestrator.Any(reqs)\n\treturn api.convertResults(results)\n}\n\n// Race returns as soon as any agent call completes (like Promise.race)\n// Each request should have:\n// - agent: string - target agent ID\n// - messages: array - messages to send\n// - options?: object - call options\nfunc (api *JSAPI) Race(requests []interface{}) []interface{} {\n\treqs := api.parseRequests(requests)\n\tresults := api.orchestrator.Race(reqs)\n\treturn api.convertResults(results)\n}\n\n// ============================================================================\n// AgentAPIWithCallback Implementation\n// ============================================================================\n\n// CallWithHandler executes a single agent call with an OnMessage handler\n// Note: For sub-agent calls, skip.history = true is automatically set.\n// Sub-agents output normally with ThreadID. Use the handler callback\n// to receive streaming messages.\nfunc (api *JSAPI) CallWithHandler(agentID string, messages []interface{}, opts map[string]interface{}, handler agentContext.OnMessageFunc) interface{} {\n\treq := api.buildRequest(agentID, messages, opts)\n\treq.Handler = handler\n\t// Force skip options for sub-agent calls\n\tapi.forceSkipForSubAgent(req)\n\tresult := api.orchestrator.callAgent(req)\n\treturn result\n}\n\n// AllWithHandler executes all agent calls with handlers\nfunc (api *JSAPI) AllWithHandler(requests []interface{}, globalHandler agentContext.BatchOnMessageFunc) []interface{} {\n\treqs := api.parseRequestsWithHandlers(requests, globalHandler)\n\tresults := api.orchestrator.All(reqs)\n\treturn api.convertResults(results)\n}\n\n// AnyWithHandler executes agent calls and returns on first success, with handlers\nfunc (api *JSAPI) AnyWithHandler(requests []interface{}, globalHandler agentContext.BatchOnMessageFunc) []interface{} {\n\treqs := api.parseRequestsWithHandlers(requests, globalHandler)\n\tresults := api.orchestrator.Any(reqs)\n\treturn api.convertResults(results)\n}\n\n// RaceWithHandler executes agent calls and returns on first completion, with handlers\nfunc (api *JSAPI) RaceWithHandler(requests []interface{}, globalHandler agentContext.BatchOnMessageFunc) []interface{} {\n\treqs := api.parseRequestsWithHandlers(requests, globalHandler)\n\tresults := api.orchestrator.Race(reqs)\n\treturn api.convertResults(results)\n}\n\n// forceSkipForSubAgent ensures proper A2A call behavior:\n// - skip.history = true: always set — A2A messages are not saved to chat history\n// - skip.output: defaults to false (sub-agents output with ThreadID for SSE stream isolation),\n// but if the caller explicitly sets skip.output = true, it is respected.\n// This allows internal worker agents (e.g. classifiers) to run silently.\nfunc (api *JSAPI) forceSkipForSubAgent(req *Request) {\n\tif req.Options == nil {\n\t\treq.Options = &CallOptions{}\n\t}\n\n\t// Preserve caller's explicit skip.output = true before overwriting Skip struct\n\tcallerSkipOutput := req.Options.Skip != nil && req.Options.Skip.Output\n\n\tif req.Options.Skip == nil {\n\t\treq.Options.Skip = &agentContext.Skip{}\n\t}\n\treq.Options.Skip.History = true\n\n\tif callerSkipOutput {\n\t\treq.Options.Skip.Output = true\n\t}\n\t// else: skip.output remains false (default zero value) — sub-agent outputs normally\n}\n\n// parseRequestsWithHandlers parses requests and attaches handlers\n// It checks for per-request _handler fields and wraps globalHandler with agentID/index\n// For all calls, this automatically sets:\n// - skip.history = true: prevents A2A messages from being saved to chat history\n// - skip.output = false: ensures sub-agents output with ThreadID (overrides user settings)\nfunc (api *JSAPI) parseRequestsWithHandlers(requests []interface{}, globalHandler agentContext.BatchOnMessageFunc) []*Request {\n\treqs := make([]*Request, 0, len(requests))\n\n\tfor i, r := range requests {\n\t\treqMap, ok := r.(map[string]interface{})\n\t\tif !ok {\n\t\t\tcontinue\n\t\t}\n\n\t\t// Get agent ID\n\t\tagentID, ok := reqMap[\"agent\"].(string)\n\t\tif !ok {\n\t\t\tcontinue\n\t\t}\n\n\t\t// Get messages\n\t\tmessages, ok := reqMap[\"messages\"].([]interface{})\n\t\tif !ok {\n\t\t\tcontinue\n\t\t}\n\n\t\t// Get options (optional)\n\t\tvar opts map[string]interface{}\n\t\tif o, ok := reqMap[\"options\"].(map[string]interface{}); ok {\n\t\t\topts = o\n\t\t}\n\n\t\treq := api.buildRequest(agentID, messages, opts)\n\n\t\t// Force skip.output = true for all sub-agent calls\n\t\tapi.forceSkipForSubAgent(req)\n\n\t\t// Check for per-request handler first (takes precedence)\n\t\tif handler, ok := reqMap[\"_handler\"].(agentContext.OnMessageFunc); ok && handler != nil {\n\t\t\treq.Handler = handler\n\t\t} else if globalHandler != nil {\n\t\t\t// Wrap global handler with agentID and index\n\t\t\tidx := i // Capture index for closure\n\t\t\taid := agentID\n\t\t\treq.Handler = func(msg *message.Message) int {\n\t\t\t\treturn globalHandler(aid, idx, msg)\n\t\t\t}\n\t\t}\n\n\t\treqs = append(reqs, req)\n\t}\n\n\treturn reqs\n}\n\n// buildRequest builds a Request from agentID, messages, and options\nfunc (api *JSAPI) buildRequest(agentID string, messages []interface{}, opts map[string]interface{}) *Request {\n\treq := &Request{\n\t\tAgentID: agentID,\n\t\tMessages: api.parseMessages(messages),\n\t}\n\n\tif opts != nil {\n\t\treq.Options = api.parseCallOptions(opts)\n\t}\n\n\treturn req\n}\n\n// parseMessages converts []interface{} to []agentContext.Message\nfunc (api *JSAPI) parseMessages(messages []interface{}) []agentContext.Message {\n\tresult := make([]agentContext.Message, 0, len(messages))\n\tfor _, m := range messages {\n\t\tmsg, ok := m.(map[string]interface{})\n\t\tif !ok {\n\t\t\tcontinue\n\t\t}\n\n\t\tctxMsg := agentContext.Message{}\n\n\t\t// Parse role\n\t\tif role, ok := msg[\"role\"].(string); ok {\n\t\t\tctxMsg.Role = agentContext.MessageRole(role)\n\t\t}\n\n\t\t// Parse content (can be string or array)\n\t\tctxMsg.Content = msg[\"content\"]\n\n\t\t// Parse name\n\t\tif name, ok := msg[\"name\"].(string); ok {\n\t\t\tctxMsg.Name = &name\n\t\t}\n\n\t\t// Parse tool_call_id\n\t\tif toolCallID, ok := msg[\"tool_call_id\"].(string); ok {\n\t\t\tctxMsg.ToolCallID = &toolCallID\n\t\t}\n\n\t\t// Parse tool_calls\n\t\tif toolCalls, ok := msg[\"tool_calls\"].([]interface{}); ok {\n\t\t\tctxMsg.ToolCalls = api.parseToolCalls(toolCalls)\n\t\t}\n\n\t\t// Parse refusal\n\t\tif refusal, ok := msg[\"refusal\"].(string); ok {\n\t\t\tctxMsg.Refusal = &refusal\n\t\t}\n\n\t\tresult = append(result, ctxMsg)\n\t}\n\treturn result\n}\n\n// parseToolCalls converts []interface{} to []agentContext.ToolCall\nfunc (api *JSAPI) parseToolCalls(toolCalls []interface{}) []agentContext.ToolCall {\n\tresult := make([]agentContext.ToolCall, 0, len(toolCalls))\n\tfor _, tc := range toolCalls {\n\t\ttcMap, ok := tc.(map[string]interface{})\n\t\tif !ok {\n\t\t\tcontinue\n\t\t}\n\n\t\ttoolCall := agentContext.ToolCall{}\n\n\t\tif id, ok := tcMap[\"id\"].(string); ok {\n\t\t\ttoolCall.ID = id\n\t\t}\n\t\tif tcType, ok := tcMap[\"type\"].(string); ok {\n\t\t\ttoolCall.Type = agentContext.ToolCallType(tcType)\n\t\t}\n\t\tif fn, ok := tcMap[\"function\"].(map[string]interface{}); ok {\n\t\t\tif name, ok := fn[\"name\"].(string); ok {\n\t\t\t\ttoolCall.Function.Name = name\n\t\t\t}\n\t\t\tif args, ok := fn[\"arguments\"].(string); ok {\n\t\t\t\ttoolCall.Function.Arguments = args\n\t\t\t}\n\t\t}\n\n\t\tresult = append(result, toolCall)\n\t}\n\treturn result\n}\n\n// parseCallOptions converts map to CallOptions\nfunc (api *JSAPI) parseCallOptions(opts map[string]interface{}) *CallOptions {\n\tcallOpts := &CallOptions{}\n\n\tif connector, ok := opts[\"connector\"].(string); ok {\n\t\tcallOpts.Connector = connector\n\t}\n\tif mode, ok := opts[\"mode\"].(string); ok {\n\t\tcallOpts.Mode = mode\n\t}\n\tif metadata, ok := opts[\"metadata\"].(map[string]interface{}); ok {\n\t\tcallOpts.Metadata = metadata\n\t}\n\n\t// Parse skip configuration\n\tif skip, ok := opts[\"skip\"].(map[string]interface{}); ok {\n\t\tcallOpts.Skip = &agentContext.Skip{}\n\t\tif history, ok := skip[\"history\"].(bool); ok {\n\t\t\tcallOpts.Skip.History = history\n\t\t}\n\t\tif trace, ok := skip[\"trace\"].(bool); ok {\n\t\t\tcallOpts.Skip.Trace = trace\n\t\t}\n\t\tif output, ok := skip[\"output\"].(bool); ok {\n\t\t\tcallOpts.Skip.Output = output\n\t\t}\n\t\tif keyword, ok := skip[\"keyword\"].(bool); ok {\n\t\t\tcallOpts.Skip.Keyword = keyword\n\t\t}\n\t\tif search, ok := skip[\"search\"].(bool); ok {\n\t\t\tcallOpts.Skip.Search = search\n\t\t}\n\t\tif contentParsing, ok := skip[\"content_parsing\"].(bool); ok {\n\t\t\tcallOpts.Skip.ContentParsing = contentParsing\n\t\t}\n\t}\n\n\treturn callOpts\n}\n\n// parseRequests parses an array of request objects into typed Requests\nfunc (api *JSAPI) parseRequests(requests []interface{}) []*Request {\n\treturn api.parseRequestsWithHandlers(requests, nil)\n}\n\n// convertResults converts typed Results to interface slice for JS\nfunc (api *JSAPI) convertResults(results []*Result) []interface{} {\n\tout := make([]interface{}, len(results))\n\tfor i, r := range results {\n\t\tout[i] = r\n\t}\n\treturn out\n}\n\n// SetJSAPIFactory sets the factory function for creating AgentAPI instances\n// Called by assistant package during initialization\nfunc SetJSAPIFactory() {\n\tagentContext.AgentAPIFactory = func(ctx *agentContext.Context) agentContext.AgentAPI {\n\t\treturn NewJSAPI(ctx)\n\t}\n}\n" }, { "path": "agent/caller/jsapi_test.go", "content": "package caller_test\n\nimport (\n\tstdContext \"context\"\n\t\"testing\"\n\n\t\"github.com/stretchr/testify/assert\"\n\t\"github.com/stretchr/testify/require\"\n\t\"github.com/yaoapp/yao/agent/caller\"\n\t\"github.com/yaoapp/yao/agent/context\"\n)\n\nfunc TestNewJSAPI(t *testing.T) {\n\tctx := context.New(stdContext.Background(), nil, \"test-chat\")\n\tapi := caller.NewJSAPI(ctx)\n\trequire.NotNil(t, api)\n}\n\nfunc TestJSAPI_Call_NoAgentGetter(t *testing.T) {\n\t// Reset AgentGetterFunc\n\toriginalGetter := caller.AgentGetterFunc\n\tcaller.AgentGetterFunc = nil\n\tdefer func() { caller.AgentGetterFunc = originalGetter }()\n\n\tctx := context.New(stdContext.Background(), nil, \"test-chat\")\n\tapi := caller.NewJSAPI(ctx)\n\n\tmessages := []interface{}{\n\t\tmap[string]interface{}{\n\t\t\t\"role\": \"user\",\n\t\t\t\"content\": \"Hello\",\n\t\t},\n\t}\n\n\tresult := api.Call(\"test-agent\", messages, nil)\n\trequire.NotNil(t, result)\n\n\tr, ok := result.(*caller.Result)\n\trequire.True(t, ok)\n\tassert.Equal(t, \"test-agent\", r.AgentID)\n\tassert.Contains(t, r.Error, \"agent getter not initialized\")\n}\n\nfunc TestJSAPI_All_Empty(t *testing.T) {\n\tctx := context.New(stdContext.Background(), nil, \"test-chat\")\n\tapi := caller.NewJSAPI(ctx)\n\tresults := api.All([]interface{}{})\n\tassert.Len(t, results, 0)\n}\n\nfunc TestJSAPI_Any_Empty(t *testing.T) {\n\tctx := context.New(stdContext.Background(), nil, \"test-chat\")\n\tapi := caller.NewJSAPI(ctx)\n\tresults := api.Any([]interface{}{})\n\tassert.Len(t, results, 0)\n}\n\nfunc TestJSAPI_Race_Empty(t *testing.T) {\n\tctx := context.New(stdContext.Background(), nil, \"test-chat\")\n\tapi := caller.NewJSAPI(ctx)\n\tresults := api.Race([]interface{}{})\n\tassert.Len(t, results, 0)\n}\n\nfunc TestJSAPI_All_InvalidRequests(t *testing.T) {\n\tctx := context.New(stdContext.Background(), nil, \"test-chat\")\n\tapi := caller.NewJSAPI(ctx)\n\n\t// Mix of invalid and valid requests\n\trequests := []interface{}{\n\t\t\"invalid\", // Not a map\n\t\tmap[string]interface{}{\n\t\t\t\"messages\": []interface{}{}, // Missing agent\n\t\t},\n\t\tmap[string]interface{}{\n\t\t\t\"agent\": \"test-agent\", // Missing messages\n\t\t},\n\t}\n\n\tresults := api.All(requests)\n\t// None should produce a result (all invalid)\n\tassert.Len(t, results, 0)\n}\n\nfunc TestJSAPI_Call_WithOptions(t *testing.T) {\n\t// Reset AgentGetterFunc\n\toriginalGetter := caller.AgentGetterFunc\n\tcaller.AgentGetterFunc = nil\n\tdefer func() { caller.AgentGetterFunc = originalGetter }()\n\n\tctx := context.New(stdContext.Background(), nil, \"test-chat\")\n\tapi := caller.NewJSAPI(ctx)\n\n\tmessages := []interface{}{\n\t\tmap[string]interface{}{\n\t\t\t\"role\": \"user\",\n\t\t\t\"content\": \"Hello\",\n\t\t},\n\t}\n\n\topts := map[string]interface{}{\n\t\t\"connector\": \"gpt4\",\n\t\t\"mode\": \"chat\",\n\t\t\"metadata\": map[string]interface{}{\n\t\t\t\"key\": \"value\",\n\t\t},\n\t\t\"skip\": map[string]interface{}{\n\t\t\t\"history\": true,\n\t\t\t\"trace\": true,\n\t\t},\n\t}\n\n\tresult := api.Call(\"test-agent\", messages, opts)\n\trequire.NotNil(t, result)\n\n\tr, ok := result.(*caller.Result)\n\trequire.True(t, ok)\n\tassert.Equal(t, \"test-agent\", r.AgentID)\n\t// Still errors because AgentGetterFunc is nil\n\tassert.Contains(t, r.Error, \"agent getter not initialized\")\n}\n\nfunc TestSetJSAPIFactory(t *testing.T) {\n\t// Reset factory\n\tcontext.AgentAPIFactory = nil\n\n\t// Set factory\n\tcaller.SetJSAPIFactory()\n\n\t// Verify factory is set\n\trequire.NotNil(t, context.AgentAPIFactory)\n\n\t// Create a mock context\n\tctx := context.New(stdContext.Background(), nil, \"test-chat\")\n\n\t// Get agent API\n\tagentAPI := context.AgentAPIFactory(ctx)\n\trequire.NotNil(t, agentAPI)\n}\n\nfunc TestJSAPI_ImplementsAgentAPI(t *testing.T) {\n\t// Verify JSAPI implements context.AgentAPI interface\n\tctx := context.New(stdContext.Background(), nil, \"test-chat\")\n\tvar _ context.AgentAPI = caller.NewJSAPI(ctx)\n}\n" }, { "path": "agent/caller/orchestrator.go", "content": "package caller\n\nimport (\n\t\"fmt\"\n\t\"sync\"\n\n\tagentContext \"github.com/yaoapp/yao/agent/context\"\n\t\"github.com/yaoapp/yao/trace/types\"\n)\n\n// Orchestrator handles parallel agent calls with different concurrency patterns\n// Modeled after JavaScript Promise patterns (all, any, race)\ntype Orchestrator struct {\n\tctx *agentContext.Context\n}\n\n// NewOrchestrator creates a new Orchestrator for parallel agent calls\nfunc NewOrchestrator(ctx *agentContext.Context) *Orchestrator {\n\treturn &Orchestrator{ctx: ctx}\n}\n\n// callResult is used internally to pass results through channels\ntype callResult struct {\n\tidx int\n\tresult *Result\n}\n\n// All executes all agent calls and waits for all to complete (like Promise.all)\n// Returns results in the same order as requests, regardless of completion order\n// Each call uses a forked context to avoid race conditions on shared state\nfunc (o *Orchestrator) All(reqs []*Request) []*Result {\n\tif len(reqs) == 0 {\n\t\treturn []*Result{}\n\t}\n\n\tresults := make([]*Result, len(reqs))\n\tvar wg sync.WaitGroup\n\tvar mu sync.Mutex\n\n\tfor i, req := range reqs {\n\t\twg.Add(1)\n\t\tgo func(idx int, r *Request) {\n\t\t\tdefer wg.Done()\n\t\t\tdefer func() {\n\t\t\t\tif err := recover(); err != nil {\n\t\t\t\t\tmu.Lock()\n\t\t\t\t\tresults[idx] = &Result{\n\t\t\t\t\t\tAgentID: r.AgentID,\n\t\t\t\t\t\tError: \"agent call panic recovered\",\n\t\t\t\t\t}\n\t\t\t\t\tmu.Unlock()\n\t\t\t\t}\n\t\t\t}()\n\n\t\t\t// Use forked context to avoid race conditions\n\t\t\tresult := o.callAgentWithForkedContext(r)\n\t\t\tmu.Lock()\n\t\t\tresults[idx] = result\n\t\t\tmu.Unlock()\n\t\t}(i, req)\n\t}\n\n\twg.Wait()\n\treturn results\n}\n\n// Any returns as soon as any agent call succeeds (has non-error result) (like Promise.any)\n// Other calls continue in background but results are discarded after first success\n// Returns all results received so far when first success is found\n// Each call uses a forked context to avoid race conditions on shared state\nfunc (o *Orchestrator) Any(reqs []*Request) []*Result {\n\tif len(reqs) == 0 {\n\t\treturn []*Result{}\n\t}\n\n\tresults := make([]*Result, len(reqs))\n\tresultChan := make(chan callResult, len(reqs))\n\n\tvar wg sync.WaitGroup\n\tdone := make(chan struct{})\n\n\tfor i, req := range reqs {\n\t\twg.Add(1)\n\t\tgo func(idx int, r *Request) {\n\t\t\tdefer wg.Done()\n\t\t\tdefer func() {\n\t\t\t\tif err := recover(); err != nil {\n\t\t\t\t\t// Send panic result through channel\n\t\t\t\t\tselect {\n\t\t\t\t\tcase <-done:\n\t\t\t\t\tcase resultChan <- callResult{idx: idx, result: &Result{\n\t\t\t\t\t\tAgentID: r.AgentID,\n\t\t\t\t\t\tError: \"agent call panic recovered\",\n\t\t\t\t\t}}:\n\t\t\t\t\t}\n\t\t\t\t}\n\t\t\t}()\n\n\t\t\t// Check if done before starting\n\t\t\tselect {\n\t\t\tcase <-done:\n\t\t\t\treturn\n\t\t\tdefault:\n\t\t\t}\n\n\t\t\t// Use forked context to avoid race conditions\n\t\t\tresult := o.callAgentWithForkedContext(r)\n\n\t\t\t// Try to send result\n\t\t\tselect {\n\t\t\tcase <-done:\n\t\t\t\t// Already found a successful result\n\t\t\tcase resultChan <- callResult{idx: idx, result: result}:\n\t\t\t}\n\t\t}(i, req)\n\t}\n\n\t// Close channel when all goroutines complete\n\tgo func() {\n\t\twg.Wait()\n\t\tclose(resultChan)\n\t}()\n\n\t// Collect results until we find one with success (no error and has content)\n\tvar foundSuccess bool\n\tfor res := range resultChan {\n\t\tresults[res.idx] = res.result\n\t\t// Check if this result is successful (no error)\n\t\tif !foundSuccess && res.result != nil && res.result.Error == \"\" {\n\t\t\tfoundSuccess = true\n\t\t\tclose(done) // Signal other goroutines to stop\n\t\t}\n\t}\n\n\treturn results\n}\n\n// Race returns as soon as any agent call completes (like Promise.race)\n// Returns immediately when first result arrives, regardless of success/failure\n// Note: Still waits for all goroutines to complete before returning to avoid resource leaks\n// Each call uses a forked context to avoid race conditions on shared state\nfunc (o *Orchestrator) Race(reqs []*Request) []*Result {\n\tif len(reqs) == 0 {\n\t\treturn []*Result{}\n\t}\n\n\tresults := make([]*Result, len(reqs))\n\tresultChan := make(chan callResult, len(reqs))\n\n\tvar wg sync.WaitGroup\n\tdone := make(chan struct{})\n\n\tfor i, req := range reqs {\n\t\twg.Add(1)\n\t\tgo func(idx int, r *Request) {\n\t\t\tdefer wg.Done()\n\t\t\tdefer func() {\n\t\t\t\tif err := recover(); err != nil {\n\t\t\t\t\t// Send panic result through channel\n\t\t\t\t\tselect {\n\t\t\t\t\tcase <-done:\n\t\t\t\t\tcase resultChan <- callResult{idx: idx, result: &Result{\n\t\t\t\t\t\tAgentID: r.AgentID,\n\t\t\t\t\t\tError: \"agent call panic recovered\",\n\t\t\t\t\t}}:\n\t\t\t\t\t}\n\t\t\t\t}\n\t\t\t}()\n\n\t\t\t// Check if done before starting\n\t\t\tselect {\n\t\t\tcase <-done:\n\t\t\t\treturn\n\t\t\tdefault:\n\t\t\t}\n\n\t\t\t// Use forked context to avoid race conditions\n\t\t\tresult := o.callAgentWithForkedContext(r)\n\n\t\t\t// Try to send result\n\t\t\tselect {\n\t\t\tcase <-done:\n\t\t\t\t// Already got first result\n\t\t\tcase resultChan <- callResult{idx: idx, result: result}:\n\t\t\t}\n\t\t}(i, req)\n\t}\n\n\t// Close channel when all goroutines complete\n\tgo func() {\n\t\twg.Wait()\n\t\tclose(resultChan)\n\t}()\n\n\t// Get first result and signal others to stop\n\tvar gotFirst bool\n\tfor res := range resultChan {\n\t\tresults[res.idx] = res.result\n\t\tif !gotFirst {\n\t\t\tgotFirst = true\n\t\t\tclose(done) // Signal other goroutines to stop\n\t\t}\n\t}\n\n\treturn results\n}\n\n// callAgent executes a single agent call using the AgentGetterFunc\n// This method handles context sharing and result extraction\nfunc (o *Orchestrator) callAgent(req *Request) *Result {\n\treturn o.callAgentWithContext(o.ctx, req)\n}\n\n// callAgentWithForkedContext executes a single agent call with a forked context\n// This is used by batch operations (All/Any/Race) to avoid race conditions\n// when multiple goroutines modify shared context state (Stack, Logger, etc.)\nfunc (o *Orchestrator) callAgentWithForkedContext(req *Request) *Result {\n\t// Fork the context to get independent Stack and Logger\n\tforkedCtx := o.ctx.Fork()\n\treturn o.callAgentWithContext(forkedCtx, req)\n}\n\n// callAgentWithContext executes a single agent call with the given context\n// This is the core implementation used by both callAgent and callAgentWithForkedContext\nfunc (o *Orchestrator) callAgentWithContext(ctx *agentContext.Context, req *Request) *Result {\n\tif req == nil {\n\t\treturn &Result{Error: \"nil request\"}\n\t}\n\n\t// Get the agent using the getter function\n\tif AgentGetterFunc == nil {\n\t\treturn NewResult(req.AgentID, nil, fmt.Errorf(\"agent getter not initialized\"))\n\t}\n\n\tagent, err := AgentGetterFunc(req.AgentID)\n\tif err != nil {\n\t\treturn NewResult(req.AgentID, nil, fmt.Errorf(\"failed to get agent: %w\", err))\n\t}\n\n\t// Mark this as an agent-to-agent fork call for proper source tracking\n\t// RefererAgentFork distinguishes ctx.agent.Call from delegate calls\n\tctx.Referer = agentContext.RefererAgentFork\n\n\t// Build context options for the call\n\tvar ctxOpts *agentContext.Options\n\tif req.Options != nil {\n\t\tctxOpts = req.Options.ToContextOptions()\n\t} else {\n\t\tctxOpts = &agentContext.Options{}\n\t}\n\n\t// If request has a handler, set OnMessage callback\n\tif req.Handler != nil {\n\t\tif ctxOpts == nil {\n\t\t\tctxOpts = &agentContext.Options{}\n\t\t}\n\t\t// Set OnMessage to receive SSE messages\n\t\tctxOpts.OnMessage = req.Handler\n\t}\n\n\t// Add trace node for A2A call using the ORIGINAL parent context's trace\n\t// (forked contexts have nil trace and nil Stack, so ctx.Trace() would create a new orphan trace)\n\tparentTrace, _ := o.ctx.Trace()\n\tvar a2aNode types.Node\n\tif parentTrace != nil {\n\t\ta2aNode, _ = parentTrace.Add(\n\t\t\tmap[string]any{\n\t\t\t\t\"agent_id\": req.AgentID,\n\t\t\t\t\"referer\": string(ctx.Referer),\n\t\t\t},\n\t\t\ttypes.TraceNodeOption{\n\t\t\t\tLabel: fmt.Sprintf(\"Agent: %s\", req.AgentID),\n\t\t\t\tType: \"agent_call\",\n\t\t\t\tIcon: \"smart_toy\",\n\t\t\t\tDescription: fmt.Sprintf(\"A2A call to '%s'\", req.AgentID),\n\t\t\t},\n\t\t)\n\t}\n\n\tresp, err := agent.Stream(ctx, req.Messages, ctxOpts)\n\tif err != nil {\n\t\tif a2aNode != nil {\n\t\t\ta2aNode.Fail(err)\n\t\t}\n\t\treturn NewResult(req.AgentID, nil, fmt.Errorf(\"agent call failed: %w\", err))\n\t}\n\n\tif a2aNode != nil {\n\t\ta2aNode.Complete(map[string]any{\n\t\t\t\"agent_id\": req.AgentID,\n\t\t\t\"status\": \"completed\",\n\t\t})\n\t}\n\n\treturn NewResult(req.AgentID, resp, nil)\n}\n\n// extractContentFromCompletion extracts the text content from a completion response\nfunc extractContentFromCompletion(completion *agentContext.CompletionResponse) string {\n\tif completion == nil {\n\t\treturn \"\"\n\t}\n\n\t// Content can be string or []ContentPart\n\tswitch content := completion.Content.(type) {\n\tcase string:\n\t\treturn content\n\tcase []interface{}:\n\t\t// Handle array of content parts - extract text parts\n\t\tvar texts []string\n\t\tfor _, part := range content {\n\t\t\tif partMap, ok := part.(map[string]interface{}); ok {\n\t\t\t\tif partType, ok := partMap[\"type\"].(string); ok && partType == \"text\" {\n\t\t\t\t\tif text, ok := partMap[\"text\"].(string); ok {\n\t\t\t\t\t\ttexts = append(texts, text)\n\t\t\t\t\t}\n\t\t\t\t}\n\t\t\t}\n\t\t}\n\t\tif len(texts) > 0 {\n\t\t\treturn texts[0] // Return first text content\n\t\t}\n\t}\n\n\treturn \"\"\n}\n" }, { "path": "agent/caller/orchestrator_test.go", "content": "package caller_test\n\nimport (\n\tstdContext \"context\"\n\t\"testing\"\n\n\t\"github.com/stretchr/testify/assert\"\n\t\"github.com/stretchr/testify/require\"\n\t\"github.com/yaoapp/yao/agent/caller\"\n\t\"github.com/yaoapp/yao/agent/context\"\n)\n\nfunc TestNewOrchestrator(t *testing.T) {\n\tctx := context.New(stdContext.Background(), nil, \"test-chat\")\n\torch := caller.NewOrchestrator(ctx)\n\trequire.NotNil(t, orch)\n}\n\nfunc TestOrchestrator_All_Empty(t *testing.T) {\n\tctx := context.New(stdContext.Background(), nil, \"test-chat\")\n\torch := caller.NewOrchestrator(ctx)\n\n\tresults := orch.All([]*caller.Request{})\n\tassert.Len(t, results, 0)\n}\n\nfunc TestOrchestrator_Any_Empty(t *testing.T) {\n\tctx := context.New(stdContext.Background(), nil, \"test-chat\")\n\torch := caller.NewOrchestrator(ctx)\n\n\tresults := orch.Any([]*caller.Request{})\n\tassert.Len(t, results, 0)\n}\n\nfunc TestOrchestrator_Race_Empty(t *testing.T) {\n\tctx := context.New(stdContext.Background(), nil, \"test-chat\")\n\torch := caller.NewOrchestrator(ctx)\n\n\tresults := orch.Race([]*caller.Request{})\n\tassert.Len(t, results, 0)\n}\n\nfunc TestOrchestrator_All_NoGetter(t *testing.T) {\n\t// Reset AgentGetterFunc\n\toriginalGetter := caller.AgentGetterFunc\n\tcaller.AgentGetterFunc = nil\n\tdefer func() { caller.AgentGetterFunc = originalGetter }()\n\n\tctx := context.New(stdContext.Background(), nil, \"test-chat\")\n\torch := caller.NewOrchestrator(ctx)\n\n\treqs := []*caller.Request{\n\t\t{\n\t\t\tAgentID: \"agent1\",\n\t\t\tMessages: []context.Message{{Role: \"user\", Content: \"Hello\"}},\n\t\t},\n\t\t{\n\t\t\tAgentID: \"agent2\",\n\t\t\tMessages: []context.Message{{Role: \"user\", Content: \"World\"}},\n\t\t},\n\t}\n\n\tresults := orch.All(reqs)\n\trequire.Len(t, results, 2)\n\n\t// All should have errors because no getter\n\tfor i, r := range results {\n\t\trequire.NotNil(t, r, \"result %d should not be nil\", i)\n\t\tassert.Contains(t, r.Error, \"agent getter not initialized\")\n\t}\n}\n\nfunc TestOrchestrator_Any_NoGetter(t *testing.T) {\n\t// Reset AgentGetterFunc\n\toriginalGetter := caller.AgentGetterFunc\n\tcaller.AgentGetterFunc = nil\n\tdefer func() { caller.AgentGetterFunc = originalGetter }()\n\n\tctx := context.New(stdContext.Background(), nil, \"test-chat\")\n\torch := caller.NewOrchestrator(ctx)\n\n\treqs := []*caller.Request{\n\t\t{\n\t\t\tAgentID: \"agent1\",\n\t\t\tMessages: []context.Message{{Role: \"user\", Content: \"Hello\"}},\n\t\t},\n\t\t{\n\t\t\tAgentID: \"agent2\",\n\t\t\tMessages: []context.Message{{Role: \"user\", Content: \"World\"}},\n\t\t},\n\t}\n\n\tresults := orch.Any(reqs)\n\trequire.Len(t, results, 2)\n\n\t// At least one result should exist\n\thasResult := false\n\tfor _, r := range results {\n\t\tif r != nil {\n\t\t\thasResult = true\n\t\t\tassert.Contains(t, r.Error, \"agent getter not initialized\")\n\t\t}\n\t}\n\tassert.True(t, hasResult)\n}\n\nfunc TestOrchestrator_Race_NoGetter(t *testing.T) {\n\t// Reset AgentGetterFunc\n\toriginalGetter := caller.AgentGetterFunc\n\tcaller.AgentGetterFunc = nil\n\tdefer func() { caller.AgentGetterFunc = originalGetter }()\n\n\tctx := context.New(stdContext.Background(), nil, \"test-chat\")\n\torch := caller.NewOrchestrator(ctx)\n\n\treqs := []*caller.Request{\n\t\t{\n\t\t\tAgentID: \"agent1\",\n\t\t\tMessages: []context.Message{{Role: \"user\", Content: \"Hello\"}},\n\t\t},\n\t\t{\n\t\t\tAgentID: \"agent2\",\n\t\t\tMessages: []context.Message{{Role: \"user\", Content: \"World\"}},\n\t\t},\n\t}\n\n\tresults := orch.Race(reqs)\n\trequire.Len(t, results, 2)\n\n\t// At least one result should exist (first to complete)\n\thasResult := false\n\tfor _, r := range results {\n\t\tif r != nil {\n\t\t\thasResult = true\n\t\t}\n\t}\n\tassert.True(t, hasResult)\n}\n\nfunc TestOrchestrator_All_NilRequest(t *testing.T) {\n\tctx := context.New(stdContext.Background(), nil, \"test-chat\")\n\torch := caller.NewOrchestrator(ctx)\n\n\treqs := []*caller.Request{\n\t\tnil,\n\t\t{\n\t\t\tAgentID: \"agent1\",\n\t\t\tMessages: []context.Message{{Role: \"user\", Content: \"Hello\"}},\n\t\t},\n\t}\n\n\t// Reset AgentGetterFunc\n\toriginalGetter := caller.AgentGetterFunc\n\tcaller.AgentGetterFunc = nil\n\tdefer func() { caller.AgentGetterFunc = originalGetter }()\n\n\tresults := orch.All(reqs)\n\trequire.Len(t, results, 2)\n\n\t// First result should have \"nil request\" error\n\tassert.Contains(t, results[0].Error, \"nil request\")\n}\n" }, { "path": "agent/caller/process.go", "content": "package caller\n\nimport (\n\t\"context\"\n\t\"encoding/json\"\n\t\"fmt\"\n\t\"time\"\n\n\t\"github.com/yaoapp/gou/process\"\n\t\"github.com/yaoapp/kun/exception\"\n\tagentContext \"github.com/yaoapp/yao/agent/context\"\n\t\"github.com/yaoapp/yao/openapi/oauth/authorized\"\n)\n\nfunc init() {\n\tprocess.Register(\"agent.Call\", processAgentCall)\n}\n\n// processAgentCall implements the agent.Call Process handler.\n// Enables agent-to-agent calls from contexts without agent.Context (e.g., YaoJob).\n//\n// Usage: Process(\"agent.Call\", { assistant_id, messages, model?, ... })\n// Returns: *Result (same structure as ctx.agent.Call in JSAPI)\nfunc processAgentCall(p *process.Process) interface{} {\n\n\t// 1. Parse parameters via struct — fail fast on invalid input\n\tif len(p.Args) == 0 {\n\t\texception.New(\"agent.Call: argument is required\", 400).Throw()\n\t}\n\n\tvar req ProcessCallRequest\n\traw, err := json.Marshal(p.Args[0])\n\tif err != nil {\n\t\texception.New(\"agent.Call: invalid argument: %s\", 400, err.Error()).Throw()\n\t}\n\tif err := json.Unmarshal(raw, &req); err != nil {\n\t\texception.New(\"agent.Call: failed to parse request: %s\", 400, err.Error()).Throw()\n\t}\n\n\tif req.AssistantID == \"\" {\n\t\texception.New(\"agent.Call: assistant_id is required\", 400).Throw()\n\t}\n\tif len(req.Messages) == 0 {\n\t\texception.New(\"agent.Call: messages is required\", 400).Throw()\n\t}\n\n\t// 2. Auto-inject authorization info from process context\n\tauthInfo := authorized.ProcessAuthInfo(p)\n\n\t// 3. Build timeout context — LLM calls can take minutes (tool use, multi-turn)\n\t// Default: 10 minutes (DefaultProcessTimeout). Caller can override via `timeout` field.\n\ttimeoutSec := req.Timeout\n\tif timeoutSec <= 0 {\n\t\ttimeoutSec = DefaultProcessTimeout\n\t}\n\tparent := p.Context\n\tif parent == nil {\n\t\tparent = context.Background()\n\t}\n\ttimeoutCtx, cancel := context.WithTimeout(parent, time.Duration(timeoutSec)*time.Second)\n\tdefer cancel()\n\n\t// 4. Build headless context + options (encapsulated in context.go)\n\tctx, opts := NewHeadlessContext(timeoutCtx, authInfo, &req)\n\tdefer ctx.Release()\n\n\t// 5. Parse messages from []map[string]interface{} to []agentContext.Message\n\tmessages := ParseMessages(req.Messages)\n\n\t// 6. Get agent and execute\n\tif AgentGetterFunc == nil {\n\t\treturn NewResult(req.AssistantID, nil, fmt.Errorf(\"agent getter not initialized\"))\n\t}\n\n\tagent, err := AgentGetterFunc(req.AssistantID)\n\tif err != nil {\n\t\treturn NewResult(req.AssistantID, nil, fmt.Errorf(\"failed to get agent: %w\", err))\n\t}\n\n\tresp, err := agent.Stream(ctx, messages, opts)\n\tif err != nil {\n\t\treturn NewResult(req.AssistantID, nil, fmt.Errorf(\"agent call failed: %w\", err))\n\t}\n\n\t// 7. Return *Result — shared with ctx.agent.Call() via NewResult()\n\treturn NewResult(req.AssistantID, resp, nil)\n}\n\n// ParseMessages converts []map[string]interface{} to []agentContext.Message.\n// Extracted as a package-level function so it can be reused by both\n// processAgentCall and JSAPI.parseMessages.\nfunc ParseMessages(raw []map[string]interface{}) []agentContext.Message {\n\tresult := make([]agentContext.Message, 0, len(raw))\n\tfor _, msg := range raw {\n\t\tctxMsg := agentContext.Message{}\n\n\t\t// Parse role\n\t\tif role, ok := msg[\"role\"].(string); ok {\n\t\t\tctxMsg.Role = agentContext.MessageRole(role)\n\t\t}\n\n\t\t// Parse content (can be string or array of content parts)\n\t\tctxMsg.Content = msg[\"content\"]\n\n\t\t// Parse name\n\t\tif name, ok := msg[\"name\"].(string); ok {\n\t\t\tctxMsg.Name = &name\n\t\t}\n\n\t\t// Parse tool_call_id\n\t\tif toolCallID, ok := msg[\"tool_call_id\"].(string); ok {\n\t\t\tctxMsg.ToolCallID = &toolCallID\n\t\t}\n\n\t\t// Parse tool_calls\n\t\tif toolCalls, ok := msg[\"tool_calls\"].([]interface{}); ok {\n\t\t\tctxMsg.ToolCalls = parseToolCalls(toolCalls)\n\t\t}\n\n\t\t// Parse refusal\n\t\tif refusal, ok := msg[\"refusal\"].(string); ok {\n\t\t\tctxMsg.Refusal = &refusal\n\t\t}\n\n\t\tresult = append(result, ctxMsg)\n\t}\n\treturn result\n}\n\n// parseToolCalls converts []interface{} to []agentContext.ToolCall\nfunc parseToolCalls(toolCalls []interface{}) []agentContext.ToolCall {\n\tresult := make([]agentContext.ToolCall, 0, len(toolCalls))\n\tfor _, tc := range toolCalls {\n\t\ttcMap, ok := tc.(map[string]interface{})\n\t\tif !ok {\n\t\t\tcontinue\n\t\t}\n\n\t\ttoolCall := agentContext.ToolCall{}\n\n\t\tif id, ok := tcMap[\"id\"].(string); ok {\n\t\t\ttoolCall.ID = id\n\t\t}\n\t\tif tcType, ok := tcMap[\"type\"].(string); ok {\n\t\t\ttoolCall.Type = agentContext.ToolCallType(tcType)\n\t\t}\n\t\tif fn, ok := tcMap[\"function\"].(map[string]interface{}); ok {\n\t\t\tif name, ok := fn[\"name\"].(string); ok {\n\t\t\t\ttoolCall.Function.Name = name\n\t\t\t}\n\t\t\tif args, ok := fn[\"arguments\"].(string); ok {\n\t\t\t\ttoolCall.Function.Arguments = args\n\t\t\t}\n\t\t}\n\n\t\tresult = append(result, toolCall)\n\t}\n\treturn result\n}\n" }, { "path": "agent/caller/process_e2e_test.go", "content": "package caller_test\n\nimport (\n\t\"context\"\n\t\"testing\"\n\t\"time\"\n\n\t\"github.com/stretchr/testify/assert\"\n\t\"github.com/stretchr/testify/require\"\n\t\"github.com/yaoapp/gou/process\"\n\t\"github.com/yaoapp/yao/agent/caller\"\n\t\"github.com/yaoapp/yao/agent/testutils\"\n)\n\n// newLLMProcess creates a process.Process with a 120s outer timeout for LLM calls.\n// agent.Call has its own internal default timeout (DefaultProcessTimeout = 600s),\n// but the outer context (120s) takes precedence via context.WithTimeout chaining.\nfunc newLLMProcess(t *testing.T, name string, args ...interface{}) *process.Process {\n\tctx, cancel := context.WithTimeout(context.Background(), 120*time.Second)\n\tt.Cleanup(cancel)\n\treturn process.NewWithContext(ctx, name, args...)\n}\n\n// ============================================================================\n// A. Pure LLM scenarios (tests.simple-greeting — no hooks)\n// ============================================================================\n\nfunc TestProcessCall_LLM_Basic(t *testing.T) {\n\tif testing.Short() {\n\t\tt.Skip(\"Skipping: requires real LLM (source env.local.sh)\")\n\t}\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\tproc := newLLMProcess(t, \"agent.call\", map[string]interface{}{\n\t\t\"assistant_id\": \"tests.simple-greeting\",\n\t\t\"messages\": []interface{}{\n\t\t\tmap[string]interface{}{\"role\": \"user\", \"content\": \"Hello!\"},\n\t\t},\n\t})\n\n\terr := proc.Execute()\n\trequire.NoError(t, err)\n\n\tval := proc.Value()\n\trequire.NotNil(t, val, \"process should return a value\")\n\n\tresult, ok := val.(*caller.Result)\n\trequire.True(t, ok, \"value should be *caller.Result, got %T\", val)\n\n\tassert.Equal(t, \"tests.simple-greeting\", result.AgentID)\n\tassert.Empty(t, result.Error, \"should not have error\")\n\tassert.NotEmpty(t, result.Content, \"should have LLM content\")\n\tassert.NotNil(t, result.Response, \"should have full response\")\n\tt.Logf(\"LLM response: %s\", result.Content)\n}\n\nfunc TestProcessCall_LLM_MultipleMessages(t *testing.T) {\n\tif testing.Short() {\n\t\tt.Skip(\"Skipping: requires real LLM\")\n\t}\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\tproc := newLLMProcess(t, \"agent.call\", map[string]interface{}{\n\t\t\"assistant_id\": \"tests.simple-greeting\",\n\t\t\"messages\": []interface{}{\n\t\t\tmap[string]interface{}{\"role\": \"system\", \"content\": \"Always reply in JSON format.\"},\n\t\t\tmap[string]interface{}{\"role\": \"user\", \"content\": \"Say hello\"},\n\t\t},\n\t})\n\n\terr := proc.Execute()\n\trequire.NoError(t, err)\n\n\tresult, ok := proc.Value().(*caller.Result)\n\trequire.True(t, ok)\n\tassert.Empty(t, result.Error)\n\tassert.NotEmpty(t, result.Content)\n\tt.Logf(\"Multi-message response: %s\", result.Content)\n}\n\nfunc TestProcessCall_LLM_WithMetadata(t *testing.T) {\n\tif testing.Short() {\n\t\tt.Skip(\"Skipping: requires real LLM\")\n\t}\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\tproc := newLLMProcess(t, \"agent.call\", map[string]interface{}{\n\t\t\"assistant_id\": \"tests.simple-greeting\",\n\t\t\"messages\": []interface{}{\n\t\t\tmap[string]interface{}{\"role\": \"user\", \"content\": \"Hi there!\"},\n\t\t},\n\t\t\"metadata\": map[string]interface{}{\"source\": \"e2e-test\", \"mode\": \"task\"},\n\t\t\"locale\": \"zh-CN\",\n\t\t\"route\": \"/test/e2e\",\n\t\t\"chat_id\": \"e2e-test-chat-001\",\n\t})\n\n\terr := proc.Execute()\n\trequire.NoError(t, err)\n\n\tresult, ok := proc.Value().(*caller.Result)\n\trequire.True(t, ok)\n\tassert.Empty(t, result.Error)\n\tassert.NotEmpty(t, result.Content)\n\tt.Logf(\"With-metadata response: %s\", result.Content)\n}\n\nfunc TestProcessCall_LLM_SkipOutputForced(t *testing.T) {\n\tif testing.Short() {\n\t\tt.Skip(\"Skipping: requires real LLM\")\n\t}\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\t// Explicitly pass skip.output=false — headless context MUST force it to true\n\t// If the force logic fails, this would panic (nil Writer).\n\tproc := newLLMProcess(t, \"agent.call\", map[string]interface{}{\n\t\t\"assistant_id\": \"tests.simple-greeting\",\n\t\t\"messages\": []interface{}{\n\t\t\tmap[string]interface{}{\"role\": \"user\", \"content\": \"Hello!\"},\n\t\t},\n\t\t\"skip\": map[string]interface{}{\n\t\t\t\"output\": false,\n\t\t\t\"history\": false,\n\t\t},\n\t})\n\n\terr := proc.Execute()\n\trequire.NoError(t, err, \"should NOT panic even with skip.output=false — headless forces true\")\n\n\tresult, ok := proc.Value().(*caller.Result)\n\trequire.True(t, ok)\n\tassert.Empty(t, result.Error)\n\tassert.NotEmpty(t, result.Content)\n}\n\n// ============================================================================\n// B. Create Hook scenarios (tests.create)\n// ============================================================================\n\nfunc TestProcessCall_CreateHook_Default(t *testing.T) {\n\tif testing.Short() {\n\t\tt.Skip(\"Skipping: requires real LLM\")\n\t}\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\t// Send a generic message — Create Hook routes to scenarioDefault\n\tproc := newLLMProcess(t, \"agent.call\", map[string]interface{}{\n\t\t\"assistant_id\": \"tests.create\",\n\t\t\"messages\": []interface{}{\n\t\t\tmap[string]interface{}{\"role\": \"user\", \"content\": \"hello world\"},\n\t\t},\n\t})\n\n\terr := proc.Execute()\n\trequire.NoError(t, err)\n\n\tresult, ok := proc.Value().(*caller.Result)\n\trequire.True(t, ok)\n\tassert.Equal(t, \"tests.create\", result.AgentID)\n\tassert.Empty(t, result.Error)\n\tassert.NotEmpty(t, result.Content, \"Create Hook should still produce LLM response\")\n\tt.Logf(\"CreateHook default response: %s\", result.Content)\n}\n\nfunc TestProcessCall_CreateHook_ReturnFull(t *testing.T) {\n\tif testing.Short() {\n\t\tt.Skip(\"Skipping: requires real LLM\")\n\t}\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\t// Send \"return_full\" — Create Hook returns full HookCreateResponse with custom messages\n\tproc := newLLMProcess(t, \"agent.call\", map[string]interface{}{\n\t\t\"assistant_id\": \"tests.create\",\n\t\t\"messages\": []interface{}{\n\t\t\tmap[string]interface{}{\"role\": \"user\", \"content\": \"return_full\"},\n\t\t},\n\t})\n\n\terr := proc.Execute()\n\trequire.NoError(t, err)\n\n\tresult, ok := proc.Value().(*caller.Result)\n\trequire.True(t, ok)\n\tassert.Equal(t, \"tests.create\", result.AgentID)\n\tassert.Empty(t, result.Error)\n\t// The Create Hook overrides messages with system + user, then LLM responds\n\tassert.NotEmpty(t, result.Content)\n\tt.Logf(\"CreateHook return_full response: %s\", result.Content)\n}\n\n// ============================================================================\n// C. Next Hook scenarios (tests.next)\n// ============================================================================\n\nfunc TestProcessCall_NextHook_Standard(t *testing.T) {\n\tif testing.Short() {\n\t\tt.Skip(\"Skipping: requires real LLM\")\n\t}\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\t// Send \"standard\" — Next Hook returns null, standard LLM response is used\n\tproc := newLLMProcess(t, \"agent.call\", map[string]interface{}{\n\t\t\"assistant_id\": \"tests.next\",\n\t\t\"messages\": []interface{}{\n\t\t\tmap[string]interface{}{\"role\": \"user\", \"content\": \"standard\"},\n\t\t},\n\t})\n\n\terr := proc.Execute()\n\trequire.NoError(t, err)\n\n\tresult, ok := proc.Value().(*caller.Result)\n\trequire.True(t, ok)\n\tassert.Equal(t, \"tests.next\", result.AgentID)\n\tassert.Empty(t, result.Error)\n\tassert.NotEmpty(t, result.Content, \"standard scenario should return LLM content\")\n\tt.Logf(\"NextHook standard response: %s\", result.Content)\n}\n\nfunc TestProcessCall_NextHook_CustomData(t *testing.T) {\n\tif testing.Short() {\n\t\tt.Skip(\"Skipping: requires real LLM\")\n\t}\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\t// Send \"return_custom_data\" — Next Hook returns custom data\n\tproc := newLLMProcess(t, \"agent.call\", map[string]interface{}{\n\t\t\"assistant_id\": \"tests.next\",\n\t\t\"messages\": []interface{}{\n\t\t\tmap[string]interface{}{\"role\": \"user\", \"content\": \"return_custom_data\"},\n\t\t},\n\t})\n\n\terr := proc.Execute()\n\trequire.NoError(t, err)\n\n\tresult, ok := proc.Value().(*caller.Result)\n\trequire.True(t, ok)\n\tassert.Equal(t, \"tests.next\", result.AgentID)\n\tassert.Empty(t, result.Error)\n\tassert.NotNil(t, result.Response, \"should have response\")\n\n\t// Next Hook custom data is available in response.Next\n\tif result.Response != nil && result.Response.Next != nil {\n\t\tt.Logf(\"NextHook custom data: %+v\", result.Response.Next)\n\t\tnextMap, ok := result.Response.Next.(map[string]interface{})\n\t\tif ok {\n\t\t\t// The Next Hook returns { data: { message, test, timestamp } }\n\t\t\tif dataMap, ok := nextMap[\"data\"].(map[string]interface{}); ok {\n\t\t\t\tassert.Equal(t, \"Custom response from Next Hook\", dataMap[\"message\"])\n\t\t\t\tassert.Equal(t, true, dataMap[\"test\"])\n\t\t\t}\n\t\t}\n\t}\n}\n\n// ============================================================================\n// D. Timeout scenarios\n// ============================================================================\n\nfunc TestProcessCall_Timeout_Short(t *testing.T) {\n\tif testing.Short() {\n\t\tt.Skip(\"Skipping: requires real LLM (source env.local.sh)\")\n\t}\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\t// Set timeout=2 seconds — LLM round-trip will certainly exceed this.\n\t// Verifies that the timeout parameter is respected and produces an error.\n\tproc := newLLMProcess(t, \"agent.call\", map[string]interface{}{\n\t\t\"assistant_id\": \"tests.simple-greeting\",\n\t\t\"messages\": []interface{}{\n\t\t\tmap[string]interface{}{\"role\": \"user\", \"content\": \"Tell me a very long story about the history of computing.\"},\n\t\t},\n\t\t\"timeout\": 2,\n\t})\n\n\terr := proc.Execute()\n\tif err != nil {\n\t\t// Timeout may surface as a process-level error (context deadline exceeded)\n\t\tt.Logf(\"Process error (expected timeout): %s\", err.Error())\n\t\tassert.Contains(t, err.Error(), \"deadline exceeded\",\n\t\t\t\"error should indicate context deadline exceeded\")\n\t\treturn\n\t}\n\n\t// Or the agent.Stream may catch the timeout and return it in Result.Error\n\tval := proc.Value()\n\trequire.NotNil(t, val, \"process should return a value\")\n\tresult, ok := val.(*caller.Result)\n\trequire.True(t, ok, \"value should be *caller.Result, got %T\", val)\n\tassert.NotEmpty(t, result.Error, \"should have timeout error in result\")\n\tt.Logf(\"Timeout error in result: %s\", result.Error)\n}\n\n// ============================================================================\n// E. Error / validation scenarios\n// ============================================================================\n\nfunc TestProcessCall_Error_MissingAssistantID(t *testing.T) {\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\tproc := process.New(\"agent.call\", map[string]interface{}{\n\t\t\"messages\": []interface{}{\n\t\t\tmap[string]interface{}{\"role\": \"user\", \"content\": \"Hello\"},\n\t\t},\n\t})\n\n\terr := proc.Execute()\n\trequire.Error(t, err, \"should fail: assistant_id is required\")\n\tt.Logf(\"Expected error: %s\", err.Error())\n\tassert.Contains(t, err.Error(), \"assistant_id\")\n}\n\nfunc TestProcessCall_Error_EmptyMessages(t *testing.T) {\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\tproc := process.New(\"agent.call\", map[string]interface{}{\n\t\t\"assistant_id\": \"tests.simple-greeting\",\n\t\t\"messages\": []interface{}{},\n\t})\n\n\terr := proc.Execute()\n\trequire.Error(t, err, \"should fail: messages is required\")\n\tt.Logf(\"Expected error: %s\", err.Error())\n\tassert.Contains(t, err.Error(), \"messages\")\n}\n\nfunc TestProcessCall_Error_InvalidArgument(t *testing.T) {\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\t// Pass a string instead of map — json.Marshal will succeed but Unmarshal will fail\n\tproc := process.New(\"agent.call\", \"not-a-map\")\n\n\terr := proc.Execute()\n\trequire.Error(t, err, \"should fail: argument must be a map\")\n\tt.Logf(\"Expected error: %s\", err.Error())\n}\n\nfunc TestProcessCall_Error_NoArgument(t *testing.T) {\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\tproc := process.New(\"agent.call\")\n\n\terr := proc.Execute()\n\trequire.Error(t, err, \"should fail: argument is required\")\n\tt.Logf(\"Expected error: %s\", err.Error())\n}\n\nfunc TestProcessCall_Error_NonexistentAgent(t *testing.T) {\n\tif testing.Short() {\n\t\tt.Skip(\"Skipping: requires environment\")\n\t}\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\tproc := process.New(\"agent.call\", map[string]interface{}{\n\t\t\"assistant_id\": \"does.not.exist.agent\",\n\t\t\"messages\": []interface{}{\n\t\t\tmap[string]interface{}{\"role\": \"user\", \"content\": \"Hello\"},\n\t\t},\n\t})\n\n\terr := proc.Execute()\n\trequire.NoError(t, err, \"process should not error — error is in Result\")\n\n\tresult, ok := proc.Value().(*caller.Result)\n\trequire.True(t, ok)\n\tassert.Equal(t, \"does.not.exist.agent\", result.AgentID)\n\tassert.NotEmpty(t, result.Error, \"should have error for nonexistent agent\")\n\tassert.Contains(t, result.Error, \"failed to get agent\")\n\tt.Logf(\"Expected error in result: %s\", result.Error)\n}\n" }, { "path": "agent/caller/process_test.go", "content": "package caller_test\n\nimport (\n\t\"context\"\n\t\"errors\"\n\t\"testing\"\n\n\t\"github.com/stretchr/testify/assert\"\n\t\"github.com/stretchr/testify/require\"\n\t\"github.com/yaoapp/yao/agent/caller\"\n\tagentContext \"github.com/yaoapp/yao/agent/context\"\n\t\"github.com/yaoapp/yao/openapi/oauth/types\"\n)\n\n// --- NewHeadlessContext tests ---\n\nfunc TestNewHeadlessContext_Basic(t *testing.T) {\n\tauthInfo := &types.AuthorizedInfo{\n\t\tTeamID: \"team-123\",\n\t\tUserID: \"user-456\",\n\t}\n\treq := &caller.ProcessCallRequest{\n\t\tAssistantID: \"yao.keeper.classify\",\n\t\tMessages: []map[string]interface{}{\n\t\t\t{\"role\": \"user\", \"content\": \"hello\"},\n\t\t},\n\t\tLocale: \"zh-CN\",\n\t}\n\n\tctx, opts := caller.NewHeadlessContext(context.Background(), authInfo, req)\n\tdefer ctx.Release()\n\n\tassert.Equal(t, \"yao.keeper.classify\", ctx.AssistantID)\n\tassert.Equal(t, agentContext.RefererProcess, ctx.Referer)\n\tassert.Equal(t, \"zh-CN\", ctx.Locale)\n\tassert.NotEmpty(t, ctx.ChatID) // auto-generated\n\n\trequire.NotNil(t, opts)\n\trequire.NotNil(t, opts.Skip)\n\tassert.True(t, opts.Skip.Output, \"skip.output must be forced true for headless context\")\n\tassert.True(t, opts.Skip.History, \"skip.history must be forced true for headless context\")\n\tassert.Empty(t, opts.Connector)\n}\n\nfunc TestNewHeadlessContext_WithModel(t *testing.T) {\n\treq := &caller.ProcessCallRequest{\n\t\tAssistantID: \"test.agent\",\n\t\tMessages: []map[string]interface{}{{\"role\": \"user\", \"content\": \"hi\"}},\n\t\tModel: \"deepseek.v3\",\n\t}\n\n\tctx, opts := caller.NewHeadlessContext(context.Background(), nil, req)\n\tdefer ctx.Release()\n\n\tassert.Equal(t, \"deepseek.v3\", opts.Connector)\n}\n\nfunc TestNewHeadlessContext_WithChatID(t *testing.T) {\n\treq := &caller.ProcessCallRequest{\n\t\tAssistantID: \"test.agent\",\n\t\tMessages: []map[string]interface{}{{\"role\": \"user\", \"content\": \"hi\"}},\n\t\tChatID: \"custom-chat-id\",\n\t}\n\n\tctx, _ := caller.NewHeadlessContext(context.Background(), nil, req)\n\tdefer ctx.Release()\n\n\tassert.Equal(t, \"custom-chat-id\", ctx.ChatID)\n}\n\nfunc TestNewHeadlessContext_ForceSkipOverridesUserSkip(t *testing.T) {\n\treq := &caller.ProcessCallRequest{\n\t\tAssistantID: \"test.agent\",\n\t\tMessages: []map[string]interface{}{{\"role\": \"user\", \"content\": \"hi\"}},\n\t\tSkip: &agentContext.Skip{Output: false, History: false, Trace: true},\n\t}\n\n\t_, opts := caller.NewHeadlessContext(context.Background(), nil, req)\n\n\t// Output and History must be forced true regardless of user input\n\tassert.True(t, opts.Skip.Output, \"skip.output must be forced true\")\n\tassert.True(t, opts.Skip.History, \"skip.history must be forced true\")\n\t// User-specified skip.trace should be preserved\n\tassert.True(t, opts.Skip.Trace, \"skip.trace should be preserved from user input\")\n}\n\nfunc TestNewHeadlessContext_WithMetadata(t *testing.T) {\n\treq := &caller.ProcessCallRequest{\n\t\tAssistantID: \"test.agent\",\n\t\tMessages: []map[string]interface{}{{\"role\": \"user\", \"content\": \"hi\"}},\n\t\tMetadata: map[string]interface{}{\"key\": \"value\"},\n\t\tRoute: \"/test\",\n\t}\n\n\tctx, _ := caller.NewHeadlessContext(context.Background(), nil, req)\n\tdefer ctx.Release()\n\n\tassert.Equal(t, \"value\", ctx.Metadata[\"key\"])\n\tassert.Equal(t, \"/test\", ctx.Route)\n}\n\nfunc TestNewHeadlessContext_WithTimeout(t *testing.T) {\n\treq := &caller.ProcessCallRequest{\n\t\tAssistantID: \"test.agent\",\n\t\tMessages: []map[string]interface{}{{\"role\": \"user\", \"content\": \"hi\"}},\n\t\tTimeout: 30,\n\t}\n\n\t// Pass a context with timeout to verify it propagates\n\tctx, _ := caller.NewHeadlessContext(context.Background(), nil, req)\n\tdefer ctx.Release()\n\n\t// Timeout field is consumed by processAgentCall, not NewHeadlessContext.\n\t// Here we just verify the field is correctly set in the struct.\n\tassert.Equal(t, 30, req.Timeout)\n}\n\nfunc TestProcessCallRequest_DefaultTimeout(t *testing.T) {\n\treq := &caller.ProcessCallRequest{\n\t\tAssistantID: \"test.agent\",\n\t\tMessages: []map[string]interface{}{{\"role\": \"user\", \"content\": \"hi\"}},\n\t}\n\n\t// When Timeout is 0 (zero value), the default should be used\n\tassert.Equal(t, 0, req.Timeout, \"zero value means use default\")\n\tassert.Equal(t, 600, caller.DefaultProcessTimeout, \"default timeout should be 600 seconds\")\n}\n\n// --- ParseMessages tests ---\n\nfunc TestParseMessages_Basic(t *testing.T) {\n\traw := []map[string]interface{}{\n\t\t{\"role\": \"user\", \"content\": \"hello\"},\n\t\t{\"role\": \"assistant\", \"content\": \"hi there\"},\n\t}\n\n\tmessages := caller.ParseMessages(raw)\n\trequire.Len(t, messages, 2)\n\n\tassert.Equal(t, agentContext.MessageRole(\"user\"), messages[0].Role)\n\tassert.Equal(t, \"hello\", messages[0].Content)\n\tassert.Equal(t, agentContext.MessageRole(\"assistant\"), messages[1].Role)\n\tassert.Equal(t, \"hi there\", messages[1].Content)\n}\n\nfunc TestParseMessages_WithOptionalFields(t *testing.T) {\n\tname := \"test-name\"\n\traw := []map[string]interface{}{\n\t\t{\n\t\t\t\"role\": \"tool\",\n\t\t\t\"content\": \"result\",\n\t\t\t\"name\": name,\n\t\t\t\"tool_call_id\": \"tc-1\",\n\t\t},\n\t}\n\n\tmessages := caller.ParseMessages(raw)\n\trequire.Len(t, messages, 1)\n\n\tmsg := messages[0]\n\tassert.Equal(t, agentContext.MessageRole(\"tool\"), msg.Role)\n\trequire.NotNil(t, msg.Name)\n\tassert.Equal(t, name, *msg.Name)\n\trequire.NotNil(t, msg.ToolCallID)\n\tassert.Equal(t, \"tc-1\", *msg.ToolCallID)\n}\n\nfunc TestParseMessages_Empty(t *testing.T) {\n\tmessages := caller.ParseMessages(nil)\n\tassert.Empty(t, messages)\n}\n\n// --- NewResult tests ---\n\nfunc TestNewResult_Success(t *testing.T) {\n\tresp := &agentContext.Response{\n\t\tCompletion: &agentContext.CompletionResponse{\n\t\t\tContent: \"answer text\",\n\t\t},\n\t}\n\n\tresult := caller.NewResult(\"test.agent\", resp, nil)\n\n\tassert.Equal(t, \"test.agent\", result.AgentID)\n\tassert.Equal(t, \"answer text\", result.Content)\n\tassert.Empty(t, result.Error)\n\tassert.NotNil(t, result.Response)\n}\n\nfunc TestNewResult_WithError(t *testing.T) {\n\tresult := caller.NewResult(\"test.agent\", nil, errors.New(\"something failed\"))\n\n\tassert.Equal(t, \"test.agent\", result.AgentID)\n\tassert.Equal(t, \"something failed\", result.Error)\n\tassert.Empty(t, result.Content)\n\tassert.Nil(t, result.Response)\n}\n\nfunc TestNewResult_NilResponse(t *testing.T) {\n\tresult := caller.NewResult(\"test.agent\", nil, nil)\n\n\tassert.Equal(t, \"test.agent\", result.AgentID)\n\tassert.Empty(t, result.Content)\n\tassert.Empty(t, result.Error)\n\tassert.Nil(t, result.Response)\n}\n\nfunc TestNewResult_NilCompletion(t *testing.T) {\n\tresp := &agentContext.Response{Completion: nil}\n\tresult := caller.NewResult(\"test.agent\", resp, nil)\n\n\tassert.Empty(t, result.Content, \"content should be empty when completion is nil\")\n\tassert.NotNil(t, result.Response)\n}\n" }, { "path": "agent/caller/sandbox_integration_test.go", "content": "package caller_test\n\nimport (\n\t\"context\"\n\t\"strings\"\n\t\"testing\"\n\t\"time\"\n\n\t\"github.com/stretchr/testify/assert\"\n\t\"github.com/stretchr/testify/require\"\n\t\"github.com/yaoapp/yao/agent/assistant\"\n\t\"github.com/yaoapp/yao/agent/caller\"\n\tagentContext \"github.com/yaoapp/yao/agent/context\"\n\t\"github.com/yaoapp/yao/agent/testutils\"\n\t\"github.com/yaoapp/yao/openapi/oauth/types\"\n)\n\n// TestSandboxE2E_ClaudeCLIExecution tests the full sandbox + claude-proxy integration\n// This test verifies:\n// 1. Assistant loads with sandbox and prompts configured\n// 2. Claude CLI is invoked (not skipped) because prompts exist\n// 3. claude-proxy correctly translates requests to OpenAI backend\n// 4. Response is received with actual content\nfunc TestSandboxE2E_ClaudeCLIExecution(t *testing.T) {\n\tif testing.Short() {\n\t\tt.Skip(\"Skipping sandbox E2E test in short mode\")\n\t}\n\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\t// Load the e2e-test assistant\n\tast, err := assistant.Get(\"tests.sandbox.e2e-test\")\n\tif err != nil {\n\t\tt.Skipf(\"Skipping test: e2e-test assistant not available: %v\", err)\n\t}\n\n\t// Verify configuration\n\trequire.NotNil(t, ast.Sandbox, \"Sandbox should be configured\")\n\trequire.NotEmpty(t, ast.Prompts, \"Prompts should be configured (required for Claude CLI)\")\n\tt.Logf(\"✓ Assistant loaded: sandbox=%s, prompts=%d\", ast.Sandbox.Command, len(ast.Prompts))\n\n\t// Create authorized info\n\tauthorized := &types.AuthorizedInfo{\n\t\tSubject: \"sandbox-e2e-test\",\n\t\tUserID: \"e2e-user-123\",\n\t\tTenantID: \"e2e-tenant\",\n\t}\n\n\t// Create context with unique chat ID\n\tchatID := \"sandbox-e2e-\" + time.Now().Format(\"20060102-150405\")\n\tctx := agentContext.New(context.Background(), authorized, chatID)\n\tctx.AssistantID = \"tests.sandbox.e2e-test\"\n\n\t// Create JSAPI\n\tapi := caller.NewJSAPI(ctx)\n\n\t// Test 1: Simple echo command\n\tt.Run(\"EchoCommand\", func(t *testing.T) {\n\t\tmessages := []interface{}{\n\t\t\tmap[string]interface{}{\n\t\t\t\t\"role\": \"user\",\n\t\t\t\t\"content\": \"Run this command: echo 'SANDBOX_E2E_SUCCESS_12345'\",\n\t\t\t},\n\t\t}\n\n\t\topts := map[string]interface{}{\n\t\t\t\"skip\": map[string]interface{}{\n\t\t\t\t\"history\": true,\n\t\t\t},\n\t\t}\n\n\t\tstartTime := time.Now()\n\t\tresult := api.Call(\"tests.sandbox.e2e-test\", messages, opts)\n\t\tduration := time.Since(startTime)\n\t\tt.Logf(\"Execution time: %v\", duration)\n\n\t\trequire.NotNil(t, result, \"Result should not be nil\")\n\n\t\tr, ok := result.(*caller.Result)\n\t\trequire.True(t, ok, \"Result should be *caller.Result\")\n\n\t\t// Check for errors\n\t\tif r.Error != \"\" {\n\t\t\t// Check if it's a Docker/sandbox availability issue\n\t\t\tif strings.Contains(r.Error, \"Docker\") ||\n\t\t\t\tstrings.Contains(r.Error, \"sandbox\") ||\n\t\t\t\tstrings.Contains(r.Error, \"container\") {\n\t\t\t\tt.Skipf(\"Skipping: Docker/sandbox not available: %s\", r.Error)\n\t\t\t}\n\t\t\tt.Fatalf(\"Agent call failed: %s\", r.Error)\n\t\t}\n\n\t\t// Verify response\n\t\tt.Logf(\"Response content: %s\", truncateStr(r.Content, 500))\n\t\tassert.NotEmpty(t, r.Content, \"Response content should not be empty\")\n\n\t\t// Check if Claude executed the command\n\t\tif strings.Contains(r.Content, \"SANDBOX_E2E_SUCCESS_12345\") {\n\t\t\tt.Log(\"✓ Echo command executed successfully - found verification string\")\n\t\t} else if strings.Contains(strings.ToLower(r.Content), \"echo\") ||\n\t\t\tstrings.Contains(r.Content, \"SANDBOX\") {\n\t\t\tt.Log(\"✓ Response mentions the command or partial output\")\n\t\t} else {\n\t\t\tt.Log(\"⚠ Response does not contain expected output\")\n\t\t}\n\t})\n}\n\n// TestSandboxE2E_FileCreation tests that Claude can create files in the sandbox\nfunc TestSandboxE2E_FileCreation(t *testing.T) {\n\tif testing.Short() {\n\t\tt.Skip(\"Skipping sandbox E2E test in short mode\")\n\t}\n\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\t// Load the e2e-test assistant\n\tast, err := assistant.Get(\"tests.sandbox.e2e-test\")\n\tif err != nil {\n\t\tt.Skipf(\"Skipping test: e2e-test assistant not available: %v\", err)\n\t}\n\n\trequire.NotNil(t, ast.Sandbox)\n\trequire.NotEmpty(t, ast.Prompts)\n\n\t// Create context\n\tauthorized := &types.AuthorizedInfo{\n\t\tSubject: \"sandbox-e2e-test\",\n\t\tUserID: \"e2e-user-456\",\n\t\tTenantID: \"e2e-tenant\",\n\t}\n\n\tchatID := \"sandbox-file-\" + time.Now().Format(\"20060102-150405\")\n\tctx := agentContext.New(context.Background(), authorized, chatID)\n\tctx.AssistantID = \"tests.sandbox.e2e-test\"\n\n\tapi := caller.NewJSAPI(ctx)\n\n\tmessages := []interface{}{\n\t\tmap[string]interface{}{\n\t\t\t\"role\": \"user\",\n\t\t\t\"content\": \"Create a file named 'test-output.txt' with the content 'FILE_CREATION_VERIFIED_67890', then read it back and show me the content.\",\n\t\t},\n\t}\n\n\topts := map[string]interface{}{\n\t\t\"skip\": map[string]interface{}{\n\t\t\t\"history\": true,\n\t\t},\n\t}\n\n\tstartTime := time.Now()\n\tresult := api.Call(\"tests.sandbox.e2e-test\", messages, opts)\n\tduration := time.Since(startTime)\n\tt.Logf(\"Execution time: %v\", duration)\n\n\trequire.NotNil(t, result)\n\n\tr, ok := result.(*caller.Result)\n\trequire.True(t, ok)\n\n\tif r.Error != \"\" {\n\t\tif strings.Contains(r.Error, \"Docker\") ||\n\t\t\tstrings.Contains(r.Error, \"sandbox\") {\n\t\t\tt.Skipf(\"Skipping: Docker/sandbox not available: %s\", r.Error)\n\t\t}\n\t\tt.Fatalf(\"Agent call failed: %s\", r.Error)\n\t}\n\n\tt.Logf(\"Response: %s\", truncateStr(r.Content, 800))\n\n\t// Verify file was created and read back\n\tif strings.Contains(r.Content, \"FILE_CREATION_VERIFIED_67890\") {\n\t\tt.Log(\"✓ File creation and read verified\")\n\t} else if strings.Contains(strings.ToLower(r.Content), \"created\") ||\n\t\tstrings.Contains(strings.ToLower(r.Content), \"wrote\") ||\n\t\tstrings.Contains(r.Content, \"test-output.txt\") {\n\t\tt.Log(\"✓ File operation appears successful\")\n\t} else {\n\t\tt.Log(\"⚠ Could not verify file creation\")\n\t}\n}\n\n// TestSandboxE2E_HookOnlyMode tests that hooks can work without Claude CLI\nfunc TestSandboxE2E_HookOnlyMode(t *testing.T) {\n\tif testing.Short() {\n\t\tt.Skip(\"Skipping sandbox E2E test in short mode\")\n\t}\n\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\t// Load the hook-only assistant (no prompts)\n\tast, err := assistant.Get(\"tests.sandbox.hook-only\")\n\tif err != nil {\n\t\tt.Skipf(\"Skipping test: hook-only assistant not available: %v\", err)\n\t}\n\n\t// Verify configuration - no prompts means Claude CLI should be skipped\n\trequire.NotNil(t, ast.Sandbox)\n\trequire.Empty(t, ast.Prompts, \"Hook-only mode should have no prompts\")\n\tt.Logf(\"✓ Hook-only assistant loaded: sandbox=%s, prompts=%d (should be 0)\", ast.Sandbox.Command, len(ast.Prompts))\n\n\t// Create context\n\tauthorized := &types.AuthorizedInfo{\n\t\tSubject: \"sandbox-hook-test\",\n\t\tUserID: \"hook-user-789\",\n\t\tTenantID: \"hook-tenant\",\n\t}\n\n\tchatID := \"sandbox-hook-\" + time.Now().Format(\"20060102-150405\")\n\tctx := agentContext.New(context.Background(), authorized, chatID)\n\tctx.AssistantID = \"tests.sandbox.hook-only\"\n\n\tapi := caller.NewJSAPI(ctx)\n\n\tmessages := []interface{}{\n\t\tmap[string]interface{}{\n\t\t\t\"role\": \"user\",\n\t\t\t\"content\": \"test hook-only mode\",\n\t\t},\n\t}\n\n\topts := map[string]interface{}{\n\t\t\"skip\": map[string]interface{}{\n\t\t\t\"history\": true,\n\t\t},\n\t}\n\n\tstartTime := time.Now()\n\tresult := api.Call(\"tests.sandbox.hook-only\", messages, opts)\n\tduration := time.Since(startTime)\n\tt.Logf(\"Execution time: %v\", duration)\n\n\trequire.NotNil(t, result)\n\n\tr, ok := result.(*caller.Result)\n\trequire.True(t, ok)\n\n\tif r.Error != \"\" {\n\t\tif strings.Contains(r.Error, \"Docker\") ||\n\t\t\tstrings.Contains(r.Error, \"sandbox\") {\n\t\t\tt.Skipf(\"Skipping: Docker/sandbox not available: %s\", r.Error)\n\t\t}\n\t\tt.Fatalf(\"Agent call failed: %s\", r.Error)\n\t}\n\n\tt.Logf(\"Response: %s\", r.Content)\n\tt.Log(\"✓ Hook-only mode executed successfully\")\n}\n\n// TestSandboxE2E_StreamingResponse verifies streaming works correctly\nfunc TestSandboxE2E_StreamingResponse(t *testing.T) {\n\tif testing.Short() {\n\t\tt.Skip(\"Skipping sandbox E2E test in short mode\")\n\t}\n\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\t// Load the e2e-test assistant\n\tast, err := assistant.Get(\"tests.sandbox.e2e-test\")\n\tif err != nil {\n\t\tt.Skipf(\"Skipping test: e2e-test assistant not available: %v\", err)\n\t}\n\n\trequire.NotNil(t, ast.Sandbox)\n\trequire.NotEmpty(t, ast.Prompts)\n\n\t// Create context\n\tauthorized := &types.AuthorizedInfo{\n\t\tSubject: \"sandbox-stream-test\",\n\t\tUserID: \"stream-user\",\n\t\tTenantID: \"stream-tenant\",\n\t}\n\n\tchatID := \"sandbox-stream-\" + time.Now().Format(\"20060102-150405\")\n\tctx := agentContext.New(context.Background(), authorized, chatID)\n\tctx.AssistantID = \"tests.sandbox.e2e-test\"\n\n\tapi := caller.NewJSAPI(ctx)\n\n\t// Ask for a slightly longer response to verify streaming\n\tmessages := []interface{}{\n\t\tmap[string]interface{}{\n\t\t\t\"role\": \"user\",\n\t\t\t\"content\": \"Say 'Hello World' and nothing else.\",\n\t\t},\n\t}\n\n\topts := map[string]interface{}{\n\t\t\"skip\": map[string]interface{}{\n\t\t\t\"history\": true,\n\t\t},\n\t}\n\n\tstartTime := time.Now()\n\tresult := api.Call(\"tests.sandbox.e2e-test\", messages, opts)\n\tduration := time.Since(startTime)\n\tt.Logf(\"Execution time: %v\", duration)\n\n\trequire.NotNil(t, result)\n\n\tr, ok := result.(*caller.Result)\n\trequire.True(t, ok)\n\n\tif r.Error != \"\" {\n\t\tif strings.Contains(r.Error, \"Docker\") ||\n\t\t\tstrings.Contains(r.Error, \"sandbox\") {\n\t\t\tt.Skipf(\"Skipping: Docker/sandbox not available: %s\", r.Error)\n\t\t}\n\t\tt.Fatalf(\"Agent call failed: %s\", r.Error)\n\t}\n\n\tt.Logf(\"Response: %s\", r.Content)\n\n\t// Verify we got a response\n\tassert.NotEmpty(t, r.Content, \"Should have response content\")\n\n\tif strings.Contains(strings.ToLower(r.Content), \"hello\") {\n\t\tt.Log(\"✓ Streaming response received with expected content\")\n\t} else {\n\t\tt.Log(\"✓ Streaming response received\")\n\t}\n}\n\nfunc truncateStr(s string, maxLen int) string {\n\ts = strings.ReplaceAll(s, \"\\n\", \" \")\n\tif len(s) <= maxLen {\n\t\treturn s\n\t}\n\treturn s[:maxLen] + \"...\"\n}\n" }, { "path": "agent/caller/types.go", "content": "// Package caller provides types and utilities for agent-to-agent calls\npackage caller\n\nimport (\n\tagentContext \"github.com/yaoapp/yao/agent/context\"\n)\n\n// DefaultProcessTimeout is the default timeout (in seconds) for agent.Call Process.\n// LLM calls with tool use can take minutes; 10 minutes provides safe headroom.\nconst DefaultProcessTimeout = 600\n\n// Request represents a request to call an agent\ntype Request struct {\n\tAgentID string `json:\"agent\"` // Target agent ID\n\tMessages []agentContext.Message `json:\"messages\"` // Messages to send\n\tOptions *CallOptions `json:\"options,omitempty\"` // Call options\n\tHandler agentContext.OnMessageFunc `json:\"-\"` // OnMessage handler for this request (not serialized)\n}\n\n// CallOptions represents options for an agent call\ntype CallOptions struct {\n\tConnector string `json:\"connector,omitempty\"` // Override connector\n\tMode string `json:\"mode,omitempty\"` // Agent mode (chat, etc.)\n\tMetadata map[string]interface{} `json:\"metadata,omitempty\"` // Custom metadata passed to hooks\n\tSkip *agentContext.Skip `json:\"skip,omitempty\"` // Skip configuration (history, trace, output, etc.)\n}\n\n// Result represents the result of an agent call\ntype Result struct {\n\tAgentID string `json:\"agent_id\"` // Agent ID that was called\n\tResponse *agentContext.Response `json:\"response,omitempty\"` // Full response from agent\n\tContent string `json:\"content,omitempty\"` // Final text content (extracted from completion)\n\tError string `json:\"error,omitempty\"` // Error message if call failed\n}\n\n// ProcessCallRequest is the parameter structure for the agent.Call Process.\n// Fields mirror CompletionRequest + HTTP header semantics, enabling headless\n// agent calls from contexts without agent.Context (e.g., YaoJob async tasks).\ntype ProcessCallRequest struct {\n\tAssistantID string `json:\"assistant_id\"` // Required: target assistant ID (maps to X-Yao-Assistant header)\n\tMessages []map[string]interface{} `json:\"messages\"` // Required: message list (maps to CompletionRequest.Messages)\n\tModel string `json:\"model,omitempty\"` // Optional: connector ID override (maps to CompletionRequest.Model)\n\tSkip *agentContext.Skip `json:\"skip,omitempty\"` // Optional: skip config (maps to CompletionRequest.Skip)\n\tMetadata map[string]interface{} `json:\"metadata,omitempty\"` // Optional: passed to hooks (maps to CompletionRequest.Metadata)\n\tLocale string `json:\"locale,omitempty\"` // Optional (maps to locale query param)\n\tRoute string `json:\"route,omitempty\"` // Optional (maps to CompletionRequest.Route)\n\tChatID string `json:\"chat_id,omitempty\"` // Optional: auto-generated if empty (maps to chat_id query/header)\n\tTimeout int `json:\"timeout,omitempty\"` // Optional: timeout in seconds (default: DefaultProcessTimeout = 600)\n}\n\n// NewResult builds a Result from an agent call response.\n// Used by both ctx.agent.Call (orchestrator) and Process(\"agent.Call\") to\n// ensure consistent result construction.\nfunc NewResult(agentID string, resp *agentContext.Response, err error) *Result {\n\tresult := &Result{AgentID: agentID}\n\tif err != nil {\n\t\tresult.Error = err.Error()\n\t\treturn result\n\t}\n\tresult.Response = resp\n\tif resp != nil && resp.Completion != nil {\n\t\tresult.Content = extractContentFromCompletion(resp.Completion)\n\t}\n\treturn result\n}\n\n// ToContextOptions converts CallOptions to context.Options for the agent call\nfunc (o *CallOptions) ToContextOptions() *agentContext.Options {\n\tif o == nil {\n\t\treturn nil\n\t}\n\n\treturn &agentContext.Options{\n\t\tConnector: o.Connector,\n\t\tMode: o.Mode,\n\t\tMetadata: o.Metadata,\n\t\tSkip: o.Skip,\n\t}\n}\n" }, { "path": "agent/caller/types_test.go", "content": "package caller_test\n\nimport (\n\t\"testing\"\n\n\t\"github.com/stretchr/testify/assert\"\n\t\"github.com/stretchr/testify/require\"\n\t\"github.com/yaoapp/yao/agent/caller\"\n\t\"github.com/yaoapp/yao/agent/context\"\n)\n\nfunc TestCallOptions_ToContextOptions_Nil(t *testing.T) {\n\tvar opts *caller.CallOptions\n\tctxOpts := opts.ToContextOptions()\n\tassert.Nil(t, ctxOpts)\n}\n\nfunc TestCallOptions_ToContextOptions_Empty(t *testing.T) {\n\topts := &caller.CallOptions{}\n\tctxOpts := opts.ToContextOptions()\n\trequire.NotNil(t, ctxOpts)\n\tassert.Empty(t, ctxOpts.Connector)\n\tassert.Empty(t, ctxOpts.Mode)\n\tassert.Nil(t, ctxOpts.Metadata)\n\tassert.Nil(t, ctxOpts.Skip)\n}\n\nfunc TestCallOptions_ToContextOptions_Full(t *testing.T) {\n\topts := &caller.CallOptions{\n\t\tConnector: \"gpt4\",\n\t\tMode: \"chat\",\n\t\tMetadata: map[string]interface{}{\n\t\t\t\"key\": \"value\",\n\t\t},\n\t\tSkip: &context.Skip{\n\t\t\tHistory: true,\n\t\t\tTrace: true,\n\t\t\tOutput: false,\n\t\t},\n\t}\n\n\tctxOpts := opts.ToContextOptions()\n\trequire.NotNil(t, ctxOpts)\n\tassert.Equal(t, \"gpt4\", ctxOpts.Connector)\n\tassert.Equal(t, \"chat\", ctxOpts.Mode)\n\tassert.Equal(t, \"value\", ctxOpts.Metadata[\"key\"])\n\trequire.NotNil(t, ctxOpts.Skip)\n\tassert.True(t, ctxOpts.Skip.History)\n\tassert.True(t, ctxOpts.Skip.Trace)\n\tassert.False(t, ctxOpts.Skip.Output)\n}\n\nfunc TestRequest_Basic(t *testing.T) {\n\treq := &caller.Request{\n\t\tAgentID: \"test-agent\",\n\t\tMessages: []context.Message{\n\t\t\t{Role: \"user\", Content: \"Hello\"},\n\t\t},\n\t}\n\n\tassert.Equal(t, \"test-agent\", req.AgentID)\n\tassert.Len(t, req.Messages, 1)\n\tassert.Equal(t, context.MessageRole(\"user\"), req.Messages[0].Role)\n}\n\nfunc TestResult_Basic(t *testing.T) {\n\tresult := &caller.Result{\n\t\tAgentID: \"test-agent\",\n\t\tContent: \"Hello response\",\n\t}\n\n\tassert.Equal(t, \"test-agent\", result.AgentID)\n\tassert.Equal(t, \"Hello response\", result.Content)\n\tassert.Empty(t, result.Error)\n}\n\nfunc TestResult_WithError(t *testing.T) {\n\tresult := &caller.Result{\n\t\tAgentID: \"test-agent\",\n\t\tError: \"something went wrong\",\n\t}\n\n\tassert.Equal(t, \"test-agent\", result.AgentID)\n\tassert.Equal(t, \"something went wrong\", result.Error)\n\tassert.Empty(t, result.Content)\n}\n" }, { "path": "agent/content/content.go", "content": "package content\n\nimport (\n\t\"fmt\"\n\t\"strings\"\n\n\t\"github.com/yaoapp/kun/log\"\n\t\"github.com/yaoapp/yao/agent/content/docx\"\n\t\"github.com/yaoapp/yao/agent/content/image\"\n\t\"github.com/yaoapp/yao/agent/content/pdf\"\n\t\"github.com/yaoapp/yao/agent/content/pptx\"\n\t\"github.com/yaoapp/yao/agent/content/text\"\n\t\"github.com/yaoapp/yao/agent/content/types\"\n\tagentContext \"github.com/yaoapp/yao/agent/context\"\n\tsearchTypes \"github.com/yaoapp/yao/agent/search/types\"\n)\n\n// ParseUserInput ParseUserInput\nfunc ParseUserInput(ctx *agentContext.Context, messages []agentContext.Message, options *types.Options) ([]agentContext.Message, *searchTypes.ReferenceContext, error) {\n\tvar referenceContext *searchTypes.ReferenceContext = nil\n\tvar parsedMessages []agentContext.Message = make([]agentContext.Message, 0)\n\tfor _, message := range messages {\n\t\t// Only process user messages (current or from history)\n\t\tif message.Role != agentContext.RoleUser {\n\t\t\tparsedMessages = append(parsedMessages, message)\n\t\t\tcontinue\n\t\t}\n\n\t\t// Parse user input message (Ignore errors)\n\t\tparsedMessage, refs, err := parseUserInputMessage(ctx, message, options)\n\t\tif err != nil {\n\t\t\tparsedMessages = append(parsedMessages, message)\n\t\t\tlog.Error(\"Failed to parse user input message: %v, %v\", message.Content, err)\n\t\t\tcontinue\n\t\t}\n\t\tparsedMessages = append(parsedMessages, parsedMessage)\n\n\t\t// Add reference to reference context\n\t\tif refs != nil {\n\t\t\tif referenceContext == nil {\n\t\t\t\treferenceContext = &searchTypes.ReferenceContext{}\n\t\t\t}\n\t\t\treferenceContext.References = append(referenceContext.References, refs...)\n\t\t}\n\t}\n\n\treturn parsedMessages, referenceContext, nil\n}\n\n// parseUserInputMessage parse a user input message\nfunc parseUserInputMessage(ctx *agentContext.Context, message agentContext.Message, options *types.Options) (agentContext.Message, []*searchTypes.Reference, error) {\n\n\t// Context content type\n\tswitch content := message.Content.(type) {\n\tcase string:\n\t\treturn message, nil, nil\n\n\tcase []agentContext.ContentPart:\n\t\treturn parseContentParts(ctx, message, content, options)\n\n\tcase []interface{}:\n\t\t// Handle content loaded from history/JSON ([]interface{} instead of []ContentPart)\n\t\tparts, ok := convertToContentParts(content)\n\t\tif !ok {\n\t\t\treturn message, nil, nil\n\t\t}\n\t\treturn parseContentParts(ctx, message, parts, options)\n\t}\n\n\treturn message, nil, fmt.Errorf(\"unsupported content type: %T\", message.Content)\n}\n\n// parseContentParts parses content parts and returns the parsed message\nfunc parseContentParts(ctx *agentContext.Context, message agentContext.Message, content []agentContext.ContentPart, options *types.Options) (agentContext.Message, []*searchTypes.Reference, error) {\n\tallRefs := []*searchTypes.Reference{}\n\tparts := make([]agentContext.ContentPart, 0, len(content))\n\tfor _, part := range content {\n\t\tparsedPart, refs, err := parseContentPart(ctx, part, options)\n\t\tif err != nil {\n\t\t\tparts = append(parts, part)\n\t\t\tcontinue\n\t\t}\n\t\tparts = append(parts, parsedPart)\n\t\tif refs != nil {\n\t\t\tallRefs = append(allRefs, refs...)\n\t\t}\n\t}\n\n\tparsedMessage := message\n\tparsedMessage.Content = parts\n\treturn parsedMessage, allRefs, nil\n}\n\n// parseContentPart parse a content part\nfunc parseContentPart(ctx *agentContext.Context, content agentContext.ContentPart, options *types.Options) (agentContext.ContentPart, []*searchTypes.Reference, error) {\n\tswitch content.Type {\n\tcase agentContext.ContentText:\n\t\treturn content, nil, nil\n\n\tcase agentContext.ContentImageURL:\n\t\treturn image.New(options).Parse(ctx, content)\n\n\tcase agentContext.ContentInputAudio:\n\t\treturn content, nil, nil\n\n\tcase agentContext.ContentFile:\n\t\treturn parseFileContent(ctx, content, options)\n\n\tcase agentContext.ContentData:\n\t\treturn content, nil, nil\n\n\tdefault:\n\t\treturn content, nil, fmt.Errorf(\"unsupported content part type: %s\", content.Type)\n\t}\n}\n\n// parseFileContent parses file content based on file type\nfunc parseFileContent(ctx *agentContext.Context, content agentContext.ContentPart, options *types.Options) (agentContext.ContentPart, []*searchTypes.Reference, error) {\n\tif content.File == nil || content.File.URL == \"\" {\n\t\treturn content, nil, nil\n\t}\n\n\t// Determine file type from filename\n\tfilename := strings.ToLower(content.File.Filename)\n\n\t// Check file type and route to appropriate handler\n\tswitch {\n\tcase strings.HasSuffix(filename, \".pdf\"):\n\t\treturn pdf.New(options).Parse(ctx, content)\n\n\tcase strings.HasSuffix(filename, \".docx\"):\n\t\treturn docx.New(options).Parse(ctx, content)\n\n\tcase strings.HasSuffix(filename, \".pptx\"):\n\t\treturn pptx.New(options).Parse(ctx, content)\n\n\tcase text.IsSupportedExtension(filename):\n\t\treturn text.New(options).Parse(ctx, content)\n\t}\n\n\t// For unsupported file types, try to read as text\n\t// This allows any file to be converted to text content\n\treturn text.New(options).ParseRaw(ctx, content)\n}\n\n// convertToContentParts converts []interface{} to []ContentPart\n// This is needed when content is loaded from JSON/history and is []interface{} instead of []ContentPart\nfunc convertToContentParts(content []interface{}) ([]agentContext.ContentPart, bool) {\n\tparts := make([]agentContext.ContentPart, 0, len(content))\n\tfor _, item := range content {\n\t\t// Each item should be a map\n\t\tm, ok := item.(map[string]interface{})\n\t\tif !ok {\n\t\t\tcontinue\n\t\t}\n\n\t\t// Get type field\n\t\ttypeStr, _ := m[\"type\"].(string)\n\t\tif typeStr == \"\" {\n\t\t\tcontinue\n\t\t}\n\n\t\tpart := agentContext.ContentPart{\n\t\t\tType: agentContext.ContentPartType(typeStr),\n\t\t}\n\n\t\tswitch typeStr {\n\t\tcase \"text\":\n\t\t\tif text, ok := m[\"text\"].(string); ok {\n\t\t\t\tpart.Text = text\n\t\t\t}\n\n\t\tcase \"image_url\":\n\t\t\tif imgData, ok := m[\"image_url\"].(map[string]interface{}); ok {\n\t\t\t\tpart.ImageURL = &agentContext.ImageURL{}\n\t\t\t\tif url, ok := imgData[\"url\"].(string); ok {\n\t\t\t\t\tpart.ImageURL.URL = url\n\t\t\t\t}\n\t\t\t\tif detail, ok := imgData[\"detail\"].(string); ok {\n\t\t\t\t\tpart.ImageURL.Detail = agentContext.ImageDetailLevel(detail)\n\t\t\t\t}\n\t\t\t}\n\n\t\tcase \"file\":\n\t\t\tif fileData, ok := m[\"file\"].(map[string]interface{}); ok {\n\t\t\t\tpart.File = &agentContext.FileAttachment{}\n\t\t\t\tif url, ok := fileData[\"url\"].(string); ok {\n\t\t\t\t\tpart.File.URL = url\n\t\t\t\t}\n\t\t\t\tif filename, ok := fileData[\"filename\"].(string); ok {\n\t\t\t\t\tpart.File.Filename = filename\n\t\t\t\t}\n\t\t\t}\n\n\t\tcase \"input_audio\":\n\t\t\tif audioData, ok := m[\"input_audio\"].(map[string]interface{}); ok {\n\t\t\t\tpart.InputAudio = &agentContext.InputAudio{}\n\t\t\t\tif data, ok := audioData[\"data\"].(string); ok {\n\t\t\t\t\tpart.InputAudio.Data = data\n\t\t\t\t}\n\t\t\t\tif format, ok := audioData[\"format\"].(string); ok {\n\t\t\t\t\tpart.InputAudio.Format = format\n\t\t\t\t}\n\t\t\t}\n\n\t\tcase \"data\":\n\t\t\tif dataContent, ok := m[\"data\"].(map[string]interface{}); ok {\n\t\t\t\tpart.Data = &agentContext.DataContent{}\n\t\t\t\tif sources, ok := dataContent[\"sources\"].([]interface{}); ok {\n\t\t\t\t\tpart.Data.Sources = make([]agentContext.DataSource, 0, len(sources))\n\t\t\t\t\tfor _, src := range sources {\n\t\t\t\t\t\tif srcMap, ok := src.(map[string]interface{}); ok {\n\t\t\t\t\t\t\tsource := agentContext.DataSource{}\n\t\t\t\t\t\t\tif t, ok := srcMap[\"type\"].(string); ok {\n\t\t\t\t\t\t\t\tsource.Type = agentContext.DataSourceType(t)\n\t\t\t\t\t\t\t}\n\t\t\t\t\t\t\tif name, ok := srcMap[\"name\"].(string); ok {\n\t\t\t\t\t\t\t\tsource.Name = name\n\t\t\t\t\t\t\t}\n\t\t\t\t\t\t\tif id, ok := srcMap[\"id\"].(string); ok {\n\t\t\t\t\t\t\t\tsource.ID = id\n\t\t\t\t\t\t\t}\n\t\t\t\t\t\t\tif filters, ok := srcMap[\"filters\"].(map[string]interface{}); ok {\n\t\t\t\t\t\t\t\tsource.Filters = filters\n\t\t\t\t\t\t\t}\n\t\t\t\t\t\t\tif metadata, ok := srcMap[\"metadata\"].(map[string]interface{}); ok {\n\t\t\t\t\t\t\t\tsource.Metadata = metadata\n\t\t\t\t\t\t\t}\n\t\t\t\t\t\t\tpart.Data.Sources = append(part.Data.Sources, source)\n\t\t\t\t\t\t}\n\t\t\t\t\t}\n\t\t\t\t}\n\t\t\t}\n\t\t}\n\n\t\tparts = append(parts, part)\n\t}\n\n\tif len(parts) == 0 {\n\t\treturn nil, false\n\t}\n\n\treturn parts, true\n}\n" }, { "path": "agent/content/docx/docx.go", "content": "package docx\n\nimport (\n\t\"fmt\"\n\t\"os\"\n\t\"strings\"\n\n\t\"github.com/yaoapp/gou/office\"\n\t\"github.com/yaoapp/yao/agent/content/types\"\n\tagentContext \"github.com/yaoapp/yao/agent/context\"\n\tsearchTypes \"github.com/yaoapp/yao/agent/search/types\"\n\t\"github.com/yaoapp/yao/attachment\"\n)\n\n// Docx handles DOCX content\ntype Docx struct {\n\toptions *types.Options\n}\n\n// New creates a new DOCX handler\nfunc New(options *types.Options) *Docx {\n\treturn &Docx{options: options}\n}\n\n// Parse parses DOCX content and returns text\nfunc (h *Docx) Parse(ctx *agentContext.Context, content agentContext.ContentPart) (agentContext.ContentPart, []*searchTypes.Reference, error) {\n\tif content.File == nil || content.File.URL == \"\" {\n\t\treturn content, nil, fmt.Errorf(\"file content missing URL\")\n\t}\n\n\turl := content.File.URL\n\n\t// Check cache first\n\tcachedText, found, err := h.readFromCache(ctx, url)\n\tif err == nil && found {\n\t\treturn agentContext.ContentPart{\n\t\t\tType: agentContext.ContentText,\n\t\t\tText: cachedText,\n\t\t}, nil, nil\n\t}\n\n\t// Read DOCX file\n\tdata, err := h.readFile(ctx, url)\n\tif err != nil {\n\t\treturn content, nil, fmt.Errorf(\"failed to read DOCX: %w\", err)\n\t}\n\n\t// Parse DOCX using gou/office\n\tparser := office.NewParser()\n\tresult, err := parser.Parse(data)\n\tif err != nil {\n\t\treturn content, nil, fmt.Errorf(\"failed to parse DOCX: %w\", err)\n\t}\n\n\ttext := result.Markdown\n\tif text == \"\" {\n\t\treturn content, nil, fmt.Errorf(\"no text content extracted from DOCX\")\n\t}\n\n\t// Cache the result\n\tif err := h.saveToCache(ctx, url, text); err != nil {\n\t\t// Log warning but don't fail\n\t\tfmt.Printf(\"Warning: failed to cache DOCX text: %v\\n\", err)\n\t}\n\n\treturn agentContext.ContentPart{\n\t\tType: agentContext.ContentText,\n\t\tText: text,\n\t}, nil, nil\n}\n\n// readFile reads DOCX content from various sources\nfunc (h *Docx) readFile(ctx *agentContext.Context, url string) ([]byte, error) {\n\tif strings.HasPrefix(url, \"__\") {\n\t\treturn h.readFromUploader(ctx, url)\n\t}\n\n\tif strings.HasPrefix(url, \"http://\") || strings.HasPrefix(url, \"https://\") {\n\t\treturn nil, fmt.Errorf(\"HTTP URL fetch not implemented yet: %s\", url)\n\t}\n\n\t// Try to read as local file path\n\tif _, err := os.Stat(url); err == nil {\n\t\treturn os.ReadFile(url)\n\t}\n\n\treturn nil, fmt.Errorf(\"unsupported DOCX source: %s\", url)\n}\n\n// readFromUploader reads DOCX content from file uploader\nfunc (h *Docx) readFromUploader(ctx *agentContext.Context, wrapper string) ([]byte, error) {\n\tuploaderName, fileID, ok := attachment.Parse(wrapper)\n\tif !ok {\n\t\treturn nil, fmt.Errorf(\"invalid uploader wrapper format: %s\", wrapper)\n\t}\n\n\tmanager, exists := attachment.Managers[uploaderName]\n\tif !exists {\n\t\treturn nil, fmt.Errorf(\"uploader '%s' not found\", uploaderName)\n\t}\n\n\tdata, err := manager.Read(ctx.Context, fileID)\n\tif err != nil {\n\t\treturn nil, fmt.Errorf(\"failed to read file: %w\", err)\n\t}\n\n\treturn data, nil\n}\n\n// readFromCache reads cached text content for a DOCX\nfunc (h *Docx) readFromCache(ctx *agentContext.Context, url string) (string, bool, error) {\n\tuploaderName, fileID, isWrapper := attachment.Parse(url)\n\tif !isWrapper {\n\t\treturn \"\", false, nil\n\t}\n\n\tmanager, exists := attachment.Managers[uploaderName]\n\tif !exists {\n\t\treturn \"\", false, nil\n\t}\n\n\ttext, err := manager.GetText(ctx.Context, fileID, false)\n\tif err == nil && text != \"\" {\n\t\treturn text, true, nil\n\t}\n\n\treturn \"\", false, nil\n}\n\n// saveToCache saves processed text to cache\nfunc (h *Docx) saveToCache(ctx *agentContext.Context, url string, text string) error {\n\tuploaderName, fileID, isWrapper := attachment.Parse(url)\n\tif !isWrapper {\n\t\treturn nil\n\t}\n\n\tmanager, exists := attachment.Managers[uploaderName]\n\tif !exists {\n\t\treturn nil\n\t}\n\n\treturn manager.SaveText(ctx.Context, fileID, text)\n}\n" }, { "path": "agent/content/docx/docx_test.go", "content": "package docx_test\n\nimport (\n\tstdContext \"context\"\n\t\"os\"\n\t\"path/filepath\"\n\t\"testing\"\n\n\t\"github.com/stretchr/testify/assert\"\n\t\"github.com/yaoapp/gou/connector/openai\"\n\t\"github.com/yaoapp/yao/agent/content/docx\"\n\tcontentTypes \"github.com/yaoapp/yao/agent/content/types\"\n\tagentContext \"github.com/yaoapp/yao/agent/context\"\n\t\"github.com/yaoapp/yao/agent/output/message\"\n\t\"github.com/yaoapp/yao/agent/testutils\"\n\toauthTypes \"github.com/yaoapp/yao/openapi/oauth/types\"\n)\n\nconst testFilesDir = \"assistants/tests/vision-helper/tests\"\n\nfunc newTestContext() *agentContext.Context {\n\tauthorized := &oauthTypes.AuthorizedInfo{\n\t\tSubject: \"test-user\",\n\t\tClientID: \"test-client-id\",\n\t\tUserID: \"test-user-123\",\n\t}\n\tctx := agentContext.New(stdContext.Background(), authorized, \"test-chat\")\n\tctx.AssistantID = \"test-assistant\"\n\tctx.Locale = \"en-us\"\n\tctx.IDGenerator = message.NewIDGenerator()\n\treturn ctx\n}\n\nfunc newTestOptions() *contentTypes.Options {\n\treturn &contentTypes.Options{\n\t\tCapabilities: &openai.Capabilities{},\n\t}\n}\n\nfunc getTestFilePath(filename string) string {\n\tyaoRoot := os.Getenv(\"YAO_TEST_APPLICATION\")\n\tif yaoRoot == \"\" {\n\t\tyaoRoot = os.Getenv(\"YAO_ROOT\")\n\t}\n\treturn filepath.Join(yaoRoot, testFilesDir, filename)\n}\n\n// TestParseWithMissingURL tests parsing DOCX with missing URL\nfunc TestParseWithMissingURL(t *testing.T) {\n\tif testing.Short() {\n\t\tt.Skip(\"Skipping integration test\")\n\t}\n\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\toptions := newTestOptions()\n\tctx := newTestContext()\n\n\tcontent := agentContext.ContentPart{\n\t\tType: agentContext.ContentFile,\n\t\tFile: nil,\n\t}\n\n\thandler := docx.New(options)\n\t_, _, err := handler.Parse(ctx, content)\n\n\tassert.Error(t, err)\n\tassert.Contains(t, err.Error(), \"missing URL\")\n}\n\n// TestParseWithLocalDocx tests parsing a local DOCX file\nfunc TestParseWithLocalDocx(t *testing.T) {\n\tif testing.Short() {\n\t\tt.Skip(\"Skipping integration test\")\n\t}\n\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\tdocxPath := getTestFilePath(\"docx.docx\")\n\tif _, err := os.Stat(docxPath); os.IsNotExist(err) {\n\t\tt.Skipf(\"Test DOCX file not found: %s\", docxPath)\n\t}\n\n\toptions := newTestOptions()\n\tctx := newTestContext()\n\n\tcontent := agentContext.ContentPart{\n\t\tType: agentContext.ContentFile,\n\t\tFile: &agentContext.FileAttachment{\n\t\t\tURL: docxPath,\n\t\t\tFilename: \"docx.docx\",\n\t\t},\n\t}\n\n\thandler := docx.New(options)\n\tresult, refs, err := handler.Parse(ctx, content)\n\n\tassert.NoError(t, err)\n\tassert.Nil(t, refs)\n\tassert.Equal(t, agentContext.ContentText, result.Type)\n\tassert.NotEmpty(t, result.Text)\n\tt.Logf(\"DOCX parse result (first 500 chars): %.500s...\", result.Text)\n}\n\n// TestParseWithNonExistentFile tests parsing DOCX with non-existent file\nfunc TestParseWithNonExistentFile(t *testing.T) {\n\tif testing.Short() {\n\t\tt.Skip(\"Skipping integration test\")\n\t}\n\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\toptions := newTestOptions()\n\tctx := newTestContext()\n\n\tcontent := agentContext.ContentPart{\n\t\tType: agentContext.ContentFile,\n\t\tFile: &agentContext.FileAttachment{\n\t\t\tURL: \"/non/existent/path/test.docx\",\n\t\t\tFilename: \"test.docx\",\n\t\t},\n\t}\n\n\thandler := docx.New(options)\n\t_, _, err := handler.Parse(ctx, content)\n\n\tassert.Error(t, err)\n\tassert.Contains(t, err.Error(), \"unsupported DOCX source\")\n}\n" }, { "path": "agent/content/image/image.go", "content": "package image\n\nimport (\n\t\"encoding/base64\"\n\t\"fmt\"\n\t\"strings\"\n\n\t\"github.com/yaoapp/yao/agent/content/tools\"\n\t\"github.com/yaoapp/yao/agent/content/types\"\n\tagentContext \"github.com/yaoapp/yao/agent/context\"\n\t\"github.com/yaoapp/yao/agent/i18n\"\n\t\"github.com/yaoapp/yao/agent/output/message\"\n\tsearchTypes \"github.com/yaoapp/yao/agent/search/types\"\n\t\"github.com/yaoapp/yao/attachment\"\n)\n\n// Image handles image content\ntype Image struct {\n\toptions *types.Options\n}\n\n// New creates a new image handler\nfunc New(options *types.Options) *Image {\n\treturn &Image{options: options}\n}\n\n// Parse parses image content\n// Logic:\n// 1. Check model capabilities first\n// 2. If forceUses is true and uses.Vision is specified -> use vision tool regardless of model capability\n// 3. If model supports vision -> pass through or convert to base64 format\n// 4. If model doesn't support vision -> use vision agent/MCP to extract text\nfunc (h *Image) Parse(ctx *agentContext.Context, content agentContext.ContentPart) (agentContext.ContentPart, []*searchTypes.Reference, error) {\n\tif content.ImageURL == nil || content.ImageURL.URL == \"\" {\n\t\treturn content, nil, fmt.Errorf(\"image_url content missing URL\")\n\t}\n\n\t// Check model capabilities first\n\tsupportsVision, visionFormat := agentContext.GetVisionSupport(h.options.Capabilities)\n\n\t// Check if we should force using Uses tools\n\tforceUses := h.options.CompletionOptions != nil && h.options.CompletionOptions.ForceUses\n\n\t// If forceUses is true and uses.Vision is specified, use vision tool regardless of model capability\n\tif forceUses && h.options.CompletionOptions != nil && h.options.CompletionOptions.Uses != nil && h.options.CompletionOptions.Uses.Vision != \"\" {\n\t\t// Check cache first before calling agent\n\t\tcachedText, found, err := h.readFromCache(ctx, content.ImageURL.URL)\n\t\tif err == nil && found {\n\t\t\treturn agentContext.ContentPart{\n\t\t\t\tType: agentContext.ContentText,\n\t\t\t\tText: cachedText,\n\t\t\t}, nil, nil\n\t\t}\n\t\treturn h.agent(ctx, content)\n\t}\n\n\t// If model supports vision\n\tif supportsVision {\n\t\turl := content.ImageURL.URL\n\t\t// If it's already a data URI (base64), pass through directly\n\t\tif strings.HasPrefix(url, \"data:\") {\n\t\t\treturn content, nil, nil\n\t\t}\n\t\t// Convert to base64 format\n\t\treturn h.base64(ctx, content, visionFormat)\n\t}\n\n\t// Model doesn't support vision - check cache first, then use vision agent/MCP\n\t// Try to get cached text (from attachment's content_preview)\n\tcachedText, found, err := h.readFromCache(ctx, content.ImageURL.URL)\n\tif err == nil && found {\n\t\t// Cache hit! Return as text content\n\t\treturn agentContext.ContentPart{\n\t\t\tType: agentContext.ContentText,\n\t\t\tText: cachedText,\n\t\t}, nil, nil\n\t}\n\n\t// No cache, try to use vision agent/MCP\n\tif h.options.CompletionOptions != nil && h.options.CompletionOptions.Uses != nil && h.options.CompletionOptions.Uses.Vision != \"\" {\n\t\treturn h.agent(ctx, content)\n\t}\n\n\t// No vision support and no vision tool specified, return error\n\treturn content, nil, fmt.Errorf(\"model doesn't support vision and no vision tool specified in uses.Vision\")\n}\n\n// base64 encodes image content to base64 (for vision support)\nfunc (h *Image) base64(ctx *agentContext.Context, content agentContext.ContentPart, format agentContext.VisionFormat) (agentContext.ContentPart, []*searchTypes.Reference, error) {\n\tif content.ImageURL == nil || content.ImageURL.URL == \"\" {\n\t\treturn content, nil, fmt.Errorf(\"image_url content missing URL\")\n\t}\n\n\turl := content.ImageURL.URL\n\n\t// Read image data from source\n\tdata, contentType, err := h.read(ctx, url)\n\tif err != nil {\n\t\treturn content, nil, fmt.Errorf(\"failed to read image: %w\", err)\n\t}\n\n\t// Encode to base64 data URI\n\tbase64Data := EncodeToBase64DataURI(data, contentType)\n\n\t// Return as image_url ContentPart\n\treturn agentContext.ContentPart{\n\t\tType: agentContext.ContentImageURL,\n\t\tImageURL: &agentContext.ImageURL{\n\t\t\tURL: base64Data,\n\t\t\tDetail: content.ImageURL.Detail,\n\t\t},\n\t}, nil, nil\n}\n\n// read reads image content from various sources\nfunc (h *Image) read(ctx *agentContext.Context, url string) ([]byte, string, error) {\n\t// Determine source type and read accordingly\n\tif strings.HasPrefix(url, \"data:\") {\n\t\t// Data URI format: data:image/png;base64,xxxxx\n\t\treturn h.readFromDataURI(url)\n\t}\n\n\tif strings.HasPrefix(url, \"__\") {\n\t\t// Uploader wrapper format: __uploader://fileid\n\t\treturn h.readFromUploader(ctx, url)\n\t}\n\n\tif strings.HasPrefix(url, \"http://\") || strings.HasPrefix(url, \"https://\") {\n\t\t// HTTP URL - for now return error, can be implemented later\n\t\treturn nil, \"\", fmt.Errorf(\"HTTP URL fetch not implemented yet: %s\", url)\n\t}\n\n\t// Unknown source\n\treturn nil, \"\", fmt.Errorf(\"unsupported image source: %s\", url)\n}\n\n// readFromDataURI reads image content from a data URI\nfunc (h *Image) readFromDataURI(dataURI string) ([]byte, string, error) {\n\t// Parse data URI: data:image/png;base64,xxxxx\n\tif !strings.HasPrefix(dataURI, \"data:\") {\n\t\treturn nil, \"\", fmt.Errorf(\"invalid data URI format\")\n\t}\n\n\t// Find the comma separator\n\tcommaIndex := strings.Index(dataURI, \",\")\n\tif commaIndex == -1 {\n\t\treturn nil, \"\", fmt.Errorf(\"invalid data URI: missing comma separator\")\n\t}\n\n\t// Extract metadata part (e.g., \"image/png;base64\")\n\tmetadata := dataURI[5:commaIndex] // Skip \"data:\"\n\tbase64Data := dataURI[commaIndex+1:]\n\n\t// Parse content type\n\tcontentType := \"image/png\" // default\n\tif strings.Contains(metadata, \";\") {\n\t\tparts := strings.Split(metadata, \";\")\n\t\tif len(parts) > 0 && parts[0] != \"\" {\n\t\t\tcontentType = parts[0]\n\t\t}\n\t} else if metadata != \"\" && metadata != \"base64\" {\n\t\tcontentType = metadata\n\t}\n\n\t// Decode base64 data\n\tdata, err := base64.StdEncoding.DecodeString(base64Data)\n\tif err != nil {\n\t\treturn nil, \"\", fmt.Errorf(\"failed to decode base64 data: %w\", err)\n\t}\n\n\treturn data, contentType, nil\n}\n\n// readFromUploader reads image content from file uploader __uploader://fileid\nfunc (h *Image) readFromUploader(ctx *agentContext.Context, wrapper string) ([]byte, string, error) {\n\t// Parse wrapper to get uploader name and file ID\n\tuploaderName, fileID, ok := attachment.Parse(wrapper)\n\tif !ok {\n\t\treturn nil, \"\", fmt.Errorf(\"invalid uploader wrapper format: %s\", wrapper)\n\t}\n\n\t// Get attachment manager\n\tmanager, exists := attachment.Managers[uploaderName]\n\tif !exists {\n\t\treturn nil, \"\", fmt.Errorf(\"uploader '%s' not found\", uploaderName)\n\t}\n\n\t// Get file info\n\tfile, err := manager.Info(ctx.Context, fileID)\n\tif err != nil {\n\t\treturn nil, \"\", fmt.Errorf(\"failed to get file info: %w\", err)\n\t}\n\n\t// Read file content\n\tdata, err := manager.Read(ctx.Context, fileID)\n\tif err != nil {\n\t\treturn nil, \"\", fmt.Errorf(\"failed to read file: %w\", err)\n\t}\n\n\treturn data, file.ContentType, nil\n}\n\n// readFromCache reads cached text content for an image\nfunc (h *Image) readFromCache(ctx *agentContext.Context, url string) (string, bool, error) {\n\t// Parse URL to check if it's an uploader wrapper\n\tuploaderName, fileID, isWrapper := attachment.Parse(url)\n\tif !isWrapper {\n\t\treturn \"\", false, nil // Not an uploader wrapper, no cache\n\t}\n\n\t// Try attachment manager's content_preview (cross-call cache)\n\tmanager, exists := attachment.Managers[uploaderName]\n\tif !exists {\n\t\treturn \"\", false, nil\n\t}\n\n\t// GetText with fullContent=false to get preview (default)\n\ttext, err := manager.GetText(ctx.Context, fileID, false)\n\tif err == nil && text != \"\" {\n\t\treturn text, true, nil\n\t}\n\n\t// No cache found\n\treturn \"\", false, nil\n}\n\n// saveToCache saves processed text to cache\nfunc (h *Image) saveToCache(ctx *agentContext.Context, url string, text string) error {\n\t// Parse URL to get uploader name and file ID\n\tuploaderName, fileID, isWrapper := attachment.Parse(url)\n\tif !isWrapper {\n\t\treturn nil // Not an uploader wrapper, nothing to cache\n\t}\n\n\t// Save to attachment manager for future calls\n\tmanager, exists := attachment.Managers[uploaderName]\n\tif !exists {\n\t\treturn nil\n\t}\n\n\treturn manager.SaveText(ctx.Context, fileID, text)\n}\n\n// agent calls image agent to parse image content\n// Note: Cache check is done in Parse() before calling this method\nfunc (h *Image) agent(ctx *agentContext.Context, content agentContext.ContentPart) (agentContext.ContentPart, []*searchTypes.Reference, error) {\n\tif content.ImageURL == nil || content.ImageURL.URL == \"\" {\n\t\treturn content, nil, fmt.Errorf(\"image_url content missing URL\")\n\t}\n\n\turl := content.ImageURL.URL\n\n\t// Get vision tool from options\n\tvisionTool := \"\"\n\tif h.options.CompletionOptions != nil && h.options.CompletionOptions.Uses != nil {\n\t\tvisionTool = h.options.CompletionOptions.Uses.Vision\n\t}\n\n\tif visionTool == \"\" {\n\t\treturn content, nil, fmt.Errorf(\"no vision tool specified in uses.Vision\")\n\t}\n\n\t// Parse vision tool format\n\t// Format can be:\n\t// - \"agent_id\" (call agent)\n\t// - \"mcp:server_id\" (call MCP tool)\n\tvar text string\n\tvar err error\n\tif strings.HasPrefix(visionTool, \"mcp:\") {\n\t\t// MCP tool\n\t\tserverID := strings.TrimPrefix(visionTool, \"mcp:\")\n\t\ttext, err = h.callMCPVisionTool(ctx, serverID, content)\n\t} else {\n\t\t// Agent call\n\t\ttext, err = h.callVisionAgent(ctx, visionTool, content)\n\t}\n\n\tif err != nil {\n\t\treturn content, nil, fmt.Errorf(\"failed to process image with vision tool: %w\", err)\n\t}\n\n\t// Cache the result\n\tif cacheErr := h.saveToCache(ctx, url, text); cacheErr != nil {\n\t\t// Log error but don't fail the request\n\t\tfmt.Printf(\"Warning: failed to cache processed text: %v\\n\", cacheErr)\n\t}\n\n\t// Return as text content\n\treturn agentContext.ContentPart{\n\t\tType: agentContext.ContentText,\n\t\tText: text,\n\t}, nil, nil\n}\n\n// callVisionAgent calls a vision agent to describe the image\nfunc (h *Image) callVisionAgent(ctx *agentContext.Context, agentID string, content agentContext.ContentPart) (string, error) {\n\t// Read image data and convert to base64\n\tdata, contentType, err := h.read(ctx, content.ImageURL.URL)\n\tif err != nil {\n\t\treturn \"\", fmt.Errorf(\"failed to read image: %w\", err)\n\t}\n\n\tbase64Data := EncodeToBase64DataURI(data, contentType)\n\n\t// Prepare message with image\n\tmessage := agentContext.Message{\n\t\tRole: agentContext.RoleUser,\n\t\tContent: []agentContext.ContentPart{\n\t\t\t{\n\t\t\t\tType: agentContext.ContentText,\n\t\t\t\tText: \"Please analyze this image.\",\n\t\t\t},\n\t\t\t{\n\t\t\t\tType: agentContext.ContentImageURL,\n\t\t\t\tImageURL: &agentContext.ImageURL{\n\t\t\t\t\tURL: base64Data,\n\t\t\t\t\tDetail: agentContext.DetailAuto,\n\t\t\t\t},\n\t\t\t},\n\t\t},\n\t}\n\n\t// Send loading message\n\tloadingID := h.sendLoading(ctx, i18n.T(ctx.Locale, \"content.image.analyzing\"))\n\n\t// Call agent using the tools package\n\tresult, err := tools.CallAgent(ctx, agentID, message)\n\n\t// Send done message\n\th.sendLoadingDone(ctx, loadingID)\n\n\treturn result, err\n}\n\n// callMCPVisionTool calls an MCP vision tool to describe the image\nfunc (h *Image) callMCPVisionTool(ctx *agentContext.Context, serverID string, content agentContext.ContentPart) (string, error) {\n\t// Read image data and convert to base64\n\tdata, contentType, err := h.read(ctx, content.ImageURL.URL)\n\tif err != nil {\n\t\treturn \"\", fmt.Errorf(\"failed to read image: %w\", err)\n\t}\n\n\tbase64Data := EncodeToBase64DataURI(data, contentType)\n\n\t// Prepare arguments for MCP tool\n\targuments := map[string]interface{}{\n\t\t\"image\": base64Data,\n\t\t\"content_type\": contentType,\n\t}\n\n\t// Send loading message\n\tloadingID := h.sendLoading(ctx, i18n.T(ctx.Locale, \"content.image.analyzing\"))\n\n\t// Call MCP tool (typically \"describe_image\" or similar)\n\tresult, err := tools.CallMCPTool(ctx, serverID, \"describe_image\", arguments)\n\n\t// Send done message\n\th.sendLoadingDone(ctx, loadingID)\n\n\treturn result, err\n}\n\n// sendLoading sends a loading message and returns the message ID\n// Returns empty string if SilentLoading is enabled\nfunc (h *Image) sendLoading(ctx *agentContext.Context, msg string) string {\n\t// Skip loading message if SilentLoading is enabled (called from parent handler like PDF)\n\tif h.options != nil && h.options.SilentLoading {\n\t\treturn \"\"\n\t}\n\n\tloadingMsg := &message.Message{\n\t\tType: message.TypeLoading,\n\t\tProps: map[string]interface{}{\n\t\t\t\"message\": msg,\n\t\t},\n\t}\n\n\tmsgID, err := ctx.SendStream(loadingMsg)\n\tif err != nil {\n\t\treturn \"\"\n\t}\n\treturn msgID\n}\n\n// sendLoadingDone marks the loading message as done\nfunc (h *Image) sendLoadingDone(ctx *agentContext.Context, loadingID string) {\n\tif loadingID == \"\" {\n\t\treturn\n\t}\n\n\tdoneMsg := &message.Message{\n\t\tMessageID: loadingID,\n\t\tDelta: true,\n\t\tDeltaAction: message.DeltaReplace,\n\t\tType: message.TypeLoading,\n\t\tProps: map[string]interface{}{\n\t\t\t\"done\": true,\n\t\t},\n\t}\n\n\tctx.Send(doneMsg)\n}\n\n// EncodeToBase64DataURI encodes data to base64 with data URI prefix\nfunc EncodeToBase64DataURI(data []byte, contentType string) string {\n\tif contentType == \"\" {\n\t\tcontentType = \"image/png\" // default for images\n\t}\n\n\tencoded := base64.StdEncoding.EncodeToString(data)\n\treturn fmt.Sprintf(\"data:%s;base64,%s\", contentType, encoded)\n}\n" }, { "path": "agent/content/image/image_test.go", "content": "package image_test\n\nimport (\n\tstdContext \"context\"\n\t\"encoding/base64\"\n\t\"strings\"\n\t\"testing\"\n\n\t\"github.com/stretchr/testify/assert\"\n\t\"github.com/yaoapp/gou/connector/openai\"\n\t\"github.com/yaoapp/yao/agent/content/image\"\n\tcontentTypes \"github.com/yaoapp/yao/agent/content/types\"\n\tagentContext \"github.com/yaoapp/yao/agent/context\"\n\t\"github.com/yaoapp/yao/agent/output/message\"\n\t\"github.com/yaoapp/yao/agent/testutils\"\n\toauthTypes \"github.com/yaoapp/yao/openapi/oauth/types\"\n)\n\n// newTestContext creates a Context for testing with commonly used fields pre-populated\nfunc newTestContext(capabilities *openai.Capabilities) *agentContext.Context {\n\tauthorized := &oauthTypes.AuthorizedInfo{\n\t\tSubject: \"test-user\",\n\t\tClientID: \"test-client-id\",\n\t\tUserID: \"test-user-123\",\n\t\tTeamID: \"test-team-456\",\n\t\tTenantID: \"test-tenant-789\",\n\t}\n\n\tctx := agentContext.New(stdContext.Background(), authorized, \"test-chat\")\n\tctx.AssistantID = \"test-assistant\"\n\tctx.Locale = \"en-us\"\n\tctx.Theme = \"light\"\n\tctx.Client = agentContext.Client{\n\t\tType: \"web\",\n\t\tUserAgent: \"TestAgent/1.0\",\n\t\tIP: \"127.0.0.1\",\n\t}\n\tctx.Referer = agentContext.RefererAPI\n\tctx.Accept = agentContext.AcceptWebCUI\n\tctx.Route = \"\"\n\tctx.Metadata = make(map[string]interface{})\n\tctx.Capabilities = capabilities\n\tctx.IDGenerator = message.NewIDGenerator()\n\treturn ctx\n}\n\n// newTestOptions creates test options with the given capabilities\nfunc newTestOptions(capabilities *openai.Capabilities, completionOptions *agentContext.CompletionOptions) *contentTypes.Options {\n\treturn &contentTypes.Options{\n\t\tCapabilities: capabilities,\n\t\tCompletionOptions: completionOptions,\n\t}\n}\n\n// TestParseWithVisionSupport tests parsing image when model supports vision\nfunc TestParseWithVisionSupport(t *testing.T) {\n\tif testing.Short() {\n\t\tt.Skip(\"Skipping integration test\")\n\t}\n\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\t// Create capabilities with vision support\n\tcapabilities := &openai.Capabilities{\n\t\tVision: \"openai\",\n\t}\n\n\toptions := newTestOptions(capabilities, nil)\n\tctx := newTestContext(capabilities)\n\n\t// Create test image content with data URI\n\tbase64Data := \"data:image/png;base64,\" + base64.StdEncoding.EncodeToString(createTestPNG())\n\tcontent := agentContext.ContentPart{\n\t\tType: agentContext.ContentImageURL,\n\t\tImageURL: &agentContext.ImageURL{\n\t\t\tURL: base64Data,\n\t\t\tDetail: agentContext.DetailAuto,\n\t\t},\n\t}\n\n\thandler := image.New(options)\n\tresult, refs, err := handler.Parse(ctx, content)\n\n\tassert.NoError(t, err)\n\tassert.Nil(t, refs)\n\tassert.Equal(t, agentContext.ContentImageURL, result.Type)\n\tassert.NotNil(t, result.ImageURL)\n\tassert.Equal(t, base64Data, result.ImageURL.URL) // Should pass through unchanged\n}\n\n// TestParseWithoutVisionSupport tests parsing image when model doesn't support vision\nfunc TestParseWithoutVisionSupport(t *testing.T) {\n\tif testing.Short() {\n\t\tt.Skip(\"Skipping integration test\")\n\t}\n\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\t// Create capabilities WITHOUT vision support\n\tcapabilities := &openai.Capabilities{\n\t\tVision: nil,\n\t}\n\n\toptions := newTestOptions(capabilities, nil)\n\tctx := newTestContext(capabilities)\n\n\t// Create test image content\n\tbase64Data := \"data:image/png;base64,\" + base64.StdEncoding.EncodeToString(createTestPNG())\n\tcontent := agentContext.ContentPart{\n\t\tType: agentContext.ContentImageURL,\n\t\tImageURL: &agentContext.ImageURL{\n\t\t\tURL: base64Data,\n\t\t\tDetail: agentContext.DetailAuto,\n\t\t},\n\t}\n\n\thandler := image.New(options)\n\t_, _, err := handler.Parse(ctx, content)\n\n\t// Should return error because no vision support and no vision tool specified\n\tassert.Error(t, err)\n\tassert.Contains(t, err.Error(), \"no vision tool specified\")\n}\n\n// TestParseWithEmptyURL tests parsing image with empty URL\nfunc TestParseWithEmptyURL(t *testing.T) {\n\tif testing.Short() {\n\t\tt.Skip(\"Skipping integration test\")\n\t}\n\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\tcapabilities := &openai.Capabilities{\n\t\tVision: \"openai\",\n\t}\n\n\toptions := newTestOptions(capabilities, nil)\n\tctx := newTestContext(capabilities)\n\n\t// Create content with empty URL\n\tcontent := agentContext.ContentPart{\n\t\tType: agentContext.ContentImageURL,\n\t\tImageURL: &agentContext.ImageURL{\n\t\t\tURL: \"\",\n\t\t},\n\t}\n\n\thandler := image.New(options)\n\t_, _, err := handler.Parse(ctx, content)\n\n\tassert.Error(t, err)\n\tassert.Contains(t, err.Error(), \"missing URL\")\n}\n\n// TestParseWithNilImageURL tests parsing image with nil ImageURL\nfunc TestParseWithNilImageURL(t *testing.T) {\n\tif testing.Short() {\n\t\tt.Skip(\"Skipping integration test\")\n\t}\n\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\tcapabilities := &openai.Capabilities{\n\t\tVision: \"openai\",\n\t}\n\n\toptions := newTestOptions(capabilities, nil)\n\tctx := newTestContext(capabilities)\n\n\t// Create content with nil ImageURL\n\tcontent := agentContext.ContentPart{\n\t\tType: agentContext.ContentImageURL,\n\t\tImageURL: nil,\n\t}\n\n\thandler := image.New(options)\n\t_, _, err := handler.Parse(ctx, content)\n\n\tassert.Error(t, err)\n\tassert.Contains(t, err.Error(), \"missing URL\")\n}\n\n// TestEncodeToBase64DataURI tests base64 encoding\nfunc TestEncodeToBase64DataURI(t *testing.T) {\n\ttests := []struct {\n\t\tname string\n\t\tdata []byte\n\t\tcontentType string\n\t\twantPrefix string\n\t}{\n\t\t{\n\t\t\tname: \"PNG image\",\n\t\t\tdata: []byte{0x89, 0x50, 0x4E, 0x47},\n\t\t\tcontentType: \"image/png\",\n\t\t\twantPrefix: \"data:image/png;base64,\",\n\t\t},\n\t\t{\n\t\t\tname: \"JPEG image\",\n\t\t\tdata: []byte{0xFF, 0xD8, 0xFF},\n\t\t\tcontentType: \"image/jpeg\",\n\t\t\twantPrefix: \"data:image/jpeg;base64,\",\n\t\t},\n\t\t{\n\t\t\tname: \"Empty content type defaults to PNG\",\n\t\t\tdata: []byte{0x01, 0x02, 0x03},\n\t\t\tcontentType: \"\",\n\t\t\twantPrefix: \"data:image/png;base64,\",\n\t\t},\n\t}\n\n\tfor _, tt := range tests {\n\t\tt.Run(tt.name, func(t *testing.T) {\n\t\t\tresult := image.EncodeToBase64DataURI(tt.data, tt.contentType)\n\n\t\t\t// Check prefix\n\t\t\tassert.True(t, strings.HasPrefix(result, tt.wantPrefix))\n\n\t\t\t// Verify base64 encoding by decoding\n\t\t\tbase64Part := result[len(tt.wantPrefix):]\n\t\t\tdecoded, err := base64.StdEncoding.DecodeString(base64Part)\n\t\t\tassert.NoError(t, err)\n\n\t\t\t// Verify decoded data matches original\n\t\t\tassert.Equal(t, tt.data, decoded)\n\t\t})\n\t}\n}\n\n// TestParseDataURIPassthrough tests that data URI images pass through unchanged when vision is supported\nfunc TestParseDataURIPassthrough(t *testing.T) {\n\tif testing.Short() {\n\t\tt.Skip(\"Skipping integration test\")\n\t}\n\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\tcapabilities := &openai.Capabilities{\n\t\tVision: \"openai\",\n\t}\n\n\toptions := newTestOptions(capabilities, nil)\n\tctx := newTestContext(capabilities)\n\n\t// Create test image content with data URI\n\toriginalURL := \"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAAEAAAABCAYAAAAfFcSJAAAADUlEQVR42mNk+M9QDwADhgGAWjR9awAAAABJRU5ErkJggg==\"\n\tcontent := agentContext.ContentPart{\n\t\tType: agentContext.ContentImageURL,\n\t\tImageURL: &agentContext.ImageURL{\n\t\t\tURL: originalURL,\n\t\t\tDetail: agentContext.DetailHigh,\n\t\t},\n\t}\n\n\thandler := image.New(options)\n\tresult, refs, err := handler.Parse(ctx, content)\n\n\tassert.NoError(t, err)\n\tassert.Nil(t, refs)\n\tassert.Equal(t, agentContext.ContentImageURL, result.Type)\n\tassert.NotNil(t, result.ImageURL)\n\tassert.Equal(t, originalURL, result.ImageURL.URL)\n\tassert.Equal(t, agentContext.DetailHigh, result.ImageURL.Detail)\n}\n\n// TestParseWithVisionAgent tests parsing image using a vision agent when model doesn't support vision\nfunc TestParseWithVisionAgent(t *testing.T) {\n\tif testing.Short() {\n\t\tt.Skip(\"Skipping integration test\")\n\t}\n\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\t// Create capabilities WITHOUT vision support\n\tcapabilities := &openai.Capabilities{\n\t\tVision: nil, // Model doesn't support vision\n\t}\n\n\t// Configure to use vision agent\n\tcompletionOptions := &agentContext.CompletionOptions{\n\t\tUses: &agentContext.Uses{\n\t\t\tVision: \"tests.vision-test\", // Use our test vision agent\n\t\t},\n\t}\n\n\toptions := newTestOptions(capabilities, completionOptions)\n\tctx := newTestContext(capabilities)\n\n\t// Create test image content with data URI (1x1 red PNG)\n\tbase64Data := \"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAAEAAAABCAYAAAAfFcSJAAAADUlEQVR42mP8z8DwHwAFBQIAX8jx0gAAAABJRU5ErkJggg==\"\n\tcontent := agentContext.ContentPart{\n\t\tType: agentContext.ContentImageURL,\n\t\tImageURL: &agentContext.ImageURL{\n\t\t\tURL: base64Data,\n\t\t\tDetail: agentContext.DetailAuto,\n\t\t},\n\t}\n\n\thandler := image.New(options)\n\tresult, refs, err := handler.Parse(ctx, content)\n\n\t// Should succeed and return text content (image description from agent)\n\tassert.NoError(t, err)\n\tassert.Nil(t, refs)\n\tassert.Equal(t, agentContext.ContentText, result.Type)\n\tassert.NotEmpty(t, result.Text) // Agent should return some description\n\tt.Logf(\"Vision agent response: %s\", result.Text)\n}\n\n// TestParseWithForceUsesVisionAgent tests forceUses flag with vision agent\nfunc TestParseWithForceUsesVisionAgent(t *testing.T) {\n\tif testing.Short() {\n\t\tt.Skip(\"Skipping integration test\")\n\t}\n\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\t// Create capabilities WITH vision support\n\tcapabilities := &openai.Capabilities{\n\t\tVision: \"openai\", // Model supports vision\n\t}\n\n\t// Configure to FORCE use vision agent (even though model supports vision)\n\tcompletionOptions := &agentContext.CompletionOptions{\n\t\tForceUses: true, // Force using the vision tool\n\t\tUses: &agentContext.Uses{\n\t\t\tVision: \"tests.vision-test\", // Use our test vision agent\n\t\t},\n\t}\n\n\toptions := newTestOptions(capabilities, completionOptions)\n\tctx := newTestContext(capabilities)\n\n\t// Create test image content with data URI\n\tbase64Data := \"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAAEAAAABCAYAAAAfFcSJAAAADUlEQVR42mP8z8DwHwAFBQIAX8jx0gAAAABJRU5ErkJggg==\"\n\tcontent := agentContext.ContentPart{\n\t\tType: agentContext.ContentImageURL,\n\t\tImageURL: &agentContext.ImageURL{\n\t\t\tURL: base64Data,\n\t\t\tDetail: agentContext.DetailAuto,\n\t\t},\n\t}\n\n\thandler := image.New(options)\n\tresult, refs, err := handler.Parse(ctx, content)\n\n\t// Should succeed and return text content (forced to use agent even though model supports vision)\n\tassert.NoError(t, err)\n\tassert.Nil(t, refs)\n\tassert.Equal(t, agentContext.ContentText, result.Type)\n\tassert.NotEmpty(t, result.Text) // Agent should return some description\n\tt.Logf(\"Vision agent (forced) response: %s\", result.Text)\n}\n\n// createTestPNG creates a minimal valid PNG image (1x1 red pixel)\nfunc createTestPNG() []byte {\n\treturn []byte{\n\t\t0x89, 0x50, 0x4E, 0x47, 0x0D, 0x0A, 0x1A, 0x0A, // PNG signature\n\t\t0x00, 0x00, 0x00, 0x0D, 0x49, 0x48, 0x44, 0x52, // IHDR chunk\n\t\t0x00, 0x00, 0x00, 0x01, 0x00, 0x00, 0x00, 0x01, // 1x1 dimensions\n\t\t0x08, 0x02, 0x00, 0x00, 0x00, 0x90, 0x77, 0x53,\n\t\t0xDE, 0x00, 0x00, 0x00, 0x0C, 0x49, 0x44, 0x41, // IDAT chunk\n\t\t0x54, 0x08, 0xD7, 0x63, 0xF8, 0xCF, 0xC0, 0x00,\n\t\t0x00, 0x03, 0x01, 0x01, 0x00, 0x18, 0xDD, 0x8D,\n\t\t0xB4, 0x00, 0x00, 0x00, 0x00, 0x49, 0x45, 0x4E, // IEND chunk\n\t\t0x44, 0xAE, 0x42, 0x60, 0x82,\n\t}\n}\n" }, { "path": "agent/content/link/link.go", "content": "package link\n" }, { "path": "agent/content/pdf/pdf.go", "content": "package pdf\n\nimport (\n\t\"fmt\"\n\t\"os\"\n\t\"path/filepath\"\n\t\"strings\"\n\t\"time\"\n\n\tgoupdf \"github.com/yaoapp/gou/pdf\"\n\t\"github.com/yaoapp/yao/agent/content/image\"\n\t\"github.com/yaoapp/yao/agent/content/types\"\n\tagentContext \"github.com/yaoapp/yao/agent/context\"\n\t\"github.com/yaoapp/yao/agent/i18n\"\n\t\"github.com/yaoapp/yao/agent/output/message\"\n\tsearchTypes \"github.com/yaoapp/yao/agent/search/types\"\n\t\"github.com/yaoapp/yao/attachment\"\n\tkbTypes \"github.com/yaoapp/yao/kb/types\"\n)\n\n// PDF handles PDF content\ntype PDF struct {\n\toptions *types.Options\n}\n\n// New creates a new PDF handler\nfunc New(options *types.Options) *PDF {\n\treturn &PDF{options: options}\n}\n\n// Parse parses PDF content by converting to images and processing each page\n// Returns multiple ContentPart (one text part per page) combined into a single text part\nfunc (h *PDF) Parse(ctx *agentContext.Context, content agentContext.ContentPart) (agentContext.ContentPart, []*searchTypes.Reference, error) {\n\tif content.File == nil || content.File.URL == \"\" {\n\t\treturn content, nil, fmt.Errorf(\"file content missing URL\")\n\t}\n\n\turl := content.File.URL\n\n\t// Check cache first\n\tcachedText, found, err := h.readFromCache(ctx, url)\n\tif err == nil && found {\n\t\treturn agentContext.ContentPart{\n\t\t\tType: agentContext.ContentText,\n\t\t\tText: cachedText,\n\t\t}, nil, nil\n\t}\n\n\t// Convert PDF to images and process each page\n\treturn h.asImages(ctx, content)\n}\n\n// ParseMulti parses PDF content and returns multiple ContentParts (one per page)\n// This is useful when you need separate parts for each page\nfunc (h *PDF) ParseMulti(ctx *agentContext.Context, content agentContext.ContentPart) ([]agentContext.ContentPart, []*searchTypes.Reference, error) {\n\tif content.File == nil || content.File.URL == \"\" {\n\t\treturn nil, nil, fmt.Errorf(\"file content missing URL\")\n\t}\n\n\turl := content.File.URL\n\n\t// Check cache first - if cached, return as single text part\n\tcachedText, found, err := h.readFromCache(ctx, url)\n\tif err == nil && found {\n\t\treturn []agentContext.ContentPart{\n\t\t\t{\n\t\t\t\tType: agentContext.ContentText,\n\t\t\t\tText: cachedText,\n\t\t\t},\n\t\t}, nil, nil\n\t}\n\n\t// Convert PDF to images and process each page\n\treturn h.asImagesMulti(ctx, content)\n}\n\n// asImages converts PDF to images and processes each page, returning combined result\nfunc (h *PDF) asImages(ctx *agentContext.Context, content agentContext.ContentPart) (agentContext.ContentPart, []*searchTypes.Reference, error) {\n\tparts, refs, err := h.asImagesMulti(ctx, content)\n\tif err != nil {\n\t\treturn content, nil, err\n\t}\n\n\tif len(parts) == 0 {\n\t\treturn content, nil, fmt.Errorf(\"no pages extracted from PDF\")\n\t}\n\n\t// Check if any parts are text (vision agent was used) or image_url (model supports vision)\n\thasTextParts := false\n\thasImageParts := false\n\tfor _, part := range parts {\n\t\tif part.Type == agentContext.ContentText {\n\t\t\thasTextParts = true\n\t\t} else if part.Type == agentContext.ContentImageURL {\n\t\t\thasImageParts = true\n\t\t}\n\t}\n\n\t// If all parts are image_url (model supports vision), return the first image\n\t// The caller should use ParseMulti to get all images\n\tif hasImageParts && !hasTextParts {\n\t\treturn parts[0], refs, nil\n\t}\n\n\t// Combine all text parts into one\n\tvar combinedText strings.Builder\n\tpageNum := 0\n\tfor _, part := range parts {\n\t\tif part.Type == agentContext.ContentText && part.Text != \"\" {\n\t\t\tpageNum++\n\t\t\tif pageNum > 1 {\n\t\t\t\tcombinedText.WriteString(\"\\n\\n---\\n\\n\") // Page separator\n\t\t\t}\n\t\t\tcombinedText.WriteString(fmt.Sprintf(\"## Page %d\\n\\n\", pageNum))\n\t\t\tcombinedText.WriteString(part.Text)\n\t\t}\n\t}\n\n\tresult := agentContext.ContentPart{\n\t\tType: agentContext.ContentText,\n\t\tText: combinedText.String(),\n\t}\n\n\t// Cache the combined result\n\tif content.File != nil && content.File.URL != \"\" && combinedText.Len() > 0 {\n\t\th.saveToCache(ctx, content.File.URL, combinedText.String())\n\t}\n\n\treturn result, refs, nil\n}\n\n// asImagesMulti converts PDF to images and processes each page separately\nfunc (h *PDF) asImagesMulti(ctx *agentContext.Context, content agentContext.ContentPart) ([]agentContext.ContentPart, []*searchTypes.Reference, error) {\n\tif content.File == nil || content.File.URL == \"\" {\n\t\treturn nil, nil, fmt.Errorf(\"file content missing URL\")\n\t}\n\n\turl := content.File.URL\n\n\t// Read PDF file\n\tpdfData, err := h.readPDF(ctx, url)\n\tif err != nil {\n\t\treturn nil, nil, fmt.Errorf(\"failed to read PDF: %w\", err)\n\t}\n\n\t// Create temporary file for PDF\n\ttempDir := os.TempDir()\n\tpdfPath := filepath.Join(tempDir, fmt.Sprintf(\"pdf_%d.pdf\", time.Now().UnixNano()))\n\tif err := os.WriteFile(pdfPath, pdfData, 0644); err != nil {\n\t\treturn nil, nil, fmt.Errorf(\"failed to write temp PDF: %w\", err)\n\t}\n\tdefer os.Remove(pdfPath)\n\n\t// Get PDF processor with global config\n\tprocessor, err := h.getPDFProcessor()\n\tif err != nil {\n\t\treturn nil, nil, fmt.Errorf(\"failed to create PDF processor: %w\", err)\n\t}\n\n\t// Create output directory for images\n\timagesDir := filepath.Join(tempDir, fmt.Sprintf(\"pdf_images_%d\", time.Now().UnixNano()))\n\tif err := os.MkdirAll(imagesDir, 0755); err != nil {\n\t\treturn nil, nil, fmt.Errorf(\"failed to create images directory: %w\", err)\n\t}\n\tdefer os.RemoveAll(imagesDir)\n\n\t// Convert PDF to images\n\tconvertConfig := goupdf.ConvertConfig{\n\t\tOutputDir: imagesDir,\n\t\tOutputPrefix: \"page\",\n\t\tFormat: \"png\",\n\t\tDPI: 150,\n\t\tQuality: 90,\n\t\tPageRange: \"all\",\n\t}\n\n\timageFiles, err := processor.Convert(ctx.Context, pdfPath, convertConfig)\n\tif err != nil {\n\t\treturn nil, nil, fmt.Errorf(\"failed to convert PDF to images: %w\", err)\n\t}\n\n\tif len(imageFiles) == 0 {\n\t\treturn nil, nil, fmt.Errorf(\"no pages extracted from PDF\")\n\t}\n\n\t// Process each image using the image handler (with SilentLoading to suppress image loading messages)\n\timageOptions := *h.options // Copy options\n\timageOptions.SilentLoading = true\n\timageHandler := image.New(&imageOptions)\n\tvar parts []agentContext.ContentPart\n\tvar allRefs []*searchTypes.Reference\n\n\tfor i, imageFile := range imageFiles {\n\t\t// Send loading message for this page\n\t\tloadingMsg := fmt.Sprintf(i18n.T(ctx.Locale, \"content.pdf.analyzing_page\"), i+1, len(imageFiles))\n\t\tloadingID := h.sendLoading(ctx, loadingMsg)\n\n\t\t// Read image file\n\t\timageData, err := os.ReadFile(imageFile)\n\t\tif err != nil {\n\t\t\th.sendLoadingDone(ctx, loadingID)\n\t\t\tcontinue\n\t\t}\n\n\t\t// Convert to base64 data URI\n\t\tbase64Data := image.EncodeToBase64DataURI(imageData, \"image/png\")\n\n\t\t// Create image content part\n\t\timagePart := agentContext.ContentPart{\n\t\t\tType: agentContext.ContentImageURL,\n\t\t\tImageURL: &agentContext.ImageURL{\n\t\t\t\tURL: base64Data,\n\t\t\t\tDetail: agentContext.DetailAuto,\n\t\t\t},\n\t\t}\n\n\t\t// Parse image using image handler\n\t\tparsedPart, refs, err := imageHandler.Parse(ctx, imagePart)\n\n\t\t// Mark loading as done\n\t\th.sendLoadingDone(ctx, loadingID)\n\n\t\tif err != nil {\n\t\t\t// If parsing fails, skip this page\n\t\t\tcontinue\n\t\t}\n\n\t\tparts = append(parts, parsedPart)\n\t\tif refs != nil {\n\t\t\tallRefs = append(allRefs, refs...)\n\t\t}\n\t}\n\n\tif len(parts) == 0 {\n\t\treturn nil, nil, fmt.Errorf(\"failed to process any PDF pages\")\n\t}\n\n\treturn parts, allRefs, nil\n}\n\n// readPDF reads PDF content from various sources\nfunc (h *PDF) readPDF(ctx *agentContext.Context, url string) ([]byte, error) {\n\tif strings.HasPrefix(url, \"__\") {\n\t\t// Uploader wrapper format: __uploader://fileid\n\t\treturn h.readFromUploader(ctx, url)\n\t}\n\n\tif strings.HasPrefix(url, \"http://\") || strings.HasPrefix(url, \"https://\") {\n\t\treturn nil, fmt.Errorf(\"HTTP URL fetch not implemented yet: %s\", url)\n\t}\n\n\t// Try to read as local file path\n\tif _, err := os.Stat(url); err == nil {\n\t\treturn os.ReadFile(url)\n\t}\n\n\treturn nil, fmt.Errorf(\"unsupported PDF source: %s\", url)\n}\n\n// readFromUploader reads PDF content from file uploader\nfunc (h *PDF) readFromUploader(ctx *agentContext.Context, wrapper string) ([]byte, error) {\n\tuploaderName, fileID, ok := attachment.Parse(wrapper)\n\tif !ok {\n\t\treturn nil, fmt.Errorf(\"invalid uploader wrapper format: %s\", wrapper)\n\t}\n\n\tmanager, exists := attachment.Managers[uploaderName]\n\tif !exists {\n\t\treturn nil, fmt.Errorf(\"uploader '%s' not found\", uploaderName)\n\t}\n\n\tdata, err := manager.Read(ctx.Context, fileID)\n\tif err != nil {\n\t\treturn nil, fmt.Errorf(\"failed to read file: %w\", err)\n\t}\n\n\treturn data, nil\n}\n\n// readFromCache reads cached text content for a PDF\nfunc (h *PDF) readFromCache(ctx *agentContext.Context, url string) (string, bool, error) {\n\tuploaderName, fileID, isWrapper := attachment.Parse(url)\n\tif !isWrapper {\n\t\treturn \"\", false, nil\n\t}\n\n\tmanager, exists := attachment.Managers[uploaderName]\n\tif !exists {\n\t\treturn \"\", false, nil\n\t}\n\n\ttext, err := manager.GetText(ctx.Context, fileID, false)\n\tif err == nil && text != \"\" {\n\t\treturn text, true, nil\n\t}\n\n\treturn \"\", false, nil\n}\n\n// saveToCache saves processed text to cache\nfunc (h *PDF) saveToCache(ctx *agentContext.Context, url string, text string) error {\n\tuploaderName, fileID, isWrapper := attachment.Parse(url)\n\tif !isWrapper {\n\t\treturn nil\n\t}\n\n\tmanager, exists := attachment.Managers[uploaderName]\n\tif !exists {\n\t\treturn nil\n\t}\n\n\treturn manager.SaveText(ctx.Context, fileID, text)\n}\n\n// getPDFProcessor creates a PDF processor using global KB config\nfunc (h *PDF) getPDFProcessor() (*goupdf.PDF, error) {\n\tglobalPDF := kbTypes.GetGlobalPDF()\n\n\topts := goupdf.Options{\n\t\tConvertTool: goupdf.ToolPdftoppm, // default\n\t\tToolPath: \"\",\n\t}\n\n\tif globalPDF != nil {\n\t\tif globalPDF.ConvertTool != \"\" {\n\t\t\tswitch globalPDF.ConvertTool {\n\t\t\tcase \"pdftoppm\":\n\t\t\t\topts.ConvertTool = goupdf.ToolPdftoppm\n\t\t\tcase \"mutool\":\n\t\t\t\topts.ConvertTool = goupdf.ToolMutool\n\t\t\tcase \"imagemagick\", \"convert\":\n\t\t\t\topts.ConvertTool = goupdf.ToolImageMagick\n\t\t\t}\n\t\t}\n\t\tif globalPDF.ToolPath != \"\" {\n\t\t\topts.ToolPath = globalPDF.ToolPath\n\t\t}\n\t}\n\n\treturn goupdf.New(opts), nil\n}\n\n// sendLoading sends a loading message and returns the message ID\nfunc (h *PDF) sendLoading(ctx *agentContext.Context, msg string) string {\n\tloadingMsg := &message.Message{\n\t\tType: message.TypeLoading,\n\t\tProps: map[string]interface{}{\n\t\t\t\"message\": msg,\n\t\t},\n\t}\n\n\tmsgID, err := ctx.SendStream(loadingMsg)\n\tif err != nil {\n\t\treturn \"\"\n\t}\n\treturn msgID\n}\n\n// sendLoadingDone marks the loading message as done\nfunc (h *PDF) sendLoadingDone(ctx *agentContext.Context, loadingID string) {\n\tif loadingID == \"\" {\n\t\treturn\n\t}\n\n\tdoneMsg := &message.Message{\n\t\tMessageID: loadingID,\n\t\tDelta: true,\n\t\tDeltaAction: message.DeltaReplace,\n\t\tType: message.TypeLoading,\n\t\tProps: map[string]interface{}{\n\t\t\t\"done\": true,\n\t\t},\n\t}\n\n\tctx.Send(doneMsg)\n}\n" }, { "path": "agent/content/pdf/pdf_test.go", "content": "package pdf_test\n\nimport (\n\tstdContext \"context\"\n\t\"os\"\n\t\"path/filepath\"\n\t\"testing\"\n\n\t\"github.com/stretchr/testify/assert\"\n\t\"github.com/yaoapp/gou/connector/openai\"\n\t\"github.com/yaoapp/yao/agent/content/pdf\"\n\tcontentTypes \"github.com/yaoapp/yao/agent/content/types\"\n\tagentContext \"github.com/yaoapp/yao/agent/context\"\n\t\"github.com/yaoapp/yao/agent/output/message\"\n\t\"github.com/yaoapp/yao/agent/testutils\"\n\toauthTypes \"github.com/yaoapp/yao/openapi/oauth/types\"\n)\n\n// Test files directory (relative to yao-dev-app)\nconst testFilesDir = \"assistants/tests/vision-helper/tests\"\n\n// newTestContext creates a Context for testing with commonly used fields pre-populated\nfunc newTestContext(capabilities *openai.Capabilities) *agentContext.Context {\n\tauthorized := &oauthTypes.AuthorizedInfo{\n\t\tSubject: \"test-user\",\n\t\tClientID: \"test-client-id\",\n\t\tUserID: \"test-user-123\",\n\t\tTeamID: \"test-team-456\",\n\t\tTenantID: \"test-tenant-789\",\n\t}\n\n\tctx := agentContext.New(stdContext.Background(), authorized, \"test-chat\")\n\tctx.AssistantID = \"test-assistant\"\n\tctx.Locale = \"en-us\"\n\tctx.Theme = \"light\"\n\tctx.Client = agentContext.Client{\n\t\tType: \"web\",\n\t\tUserAgent: \"TestAgent/1.0\",\n\t\tIP: \"127.0.0.1\",\n\t}\n\tctx.Referer = agentContext.RefererAPI\n\tctx.Accept = agentContext.AcceptWebCUI\n\tctx.Route = \"\"\n\tctx.Metadata = make(map[string]interface{})\n\tctx.Capabilities = capabilities\n\tctx.IDGenerator = message.NewIDGenerator()\n\treturn ctx\n}\n\n// newTestOptions creates test options with the given capabilities\nfunc newTestOptions(capabilities *openai.Capabilities, completionOptions *agentContext.CompletionOptions) *contentTypes.Options {\n\treturn &contentTypes.Options{\n\t\tCapabilities: capabilities,\n\t\tCompletionOptions: completionOptions,\n\t}\n}\n\n// getTestFilePath returns the full path to a test file\nfunc getTestFilePath(filename string) string {\n\tyaoRoot := os.Getenv(\"YAO_TEST_APPLICATION\")\n\tif yaoRoot == \"\" {\n\t\tyaoRoot = os.Getenv(\"YAO_ROOT\")\n\t}\n\treturn filepath.Join(yaoRoot, testFilesDir, filename)\n}\n\n// TestParseWithMissingURL tests parsing PDF with missing URL\nfunc TestParseWithMissingURL(t *testing.T) {\n\tif testing.Short() {\n\t\tt.Skip(\"Skipping integration test\")\n\t}\n\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\tcapabilities := &openai.Capabilities{\n\t\tVision: \"openai\",\n\t}\n\n\toptions := newTestOptions(capabilities, nil)\n\tctx := newTestContext(capabilities)\n\n\t// Create content with nil File\n\tcontent := agentContext.ContentPart{\n\t\tType: agentContext.ContentFile,\n\t\tFile: nil,\n\t}\n\n\thandler := pdf.New(options)\n\t_, _, err := handler.Parse(ctx, content)\n\n\tassert.Error(t, err)\n\tassert.Contains(t, err.Error(), \"missing URL\")\n}\n\n// TestParseWithEmptyURL tests parsing PDF with empty URL\nfunc TestParseWithEmptyURL(t *testing.T) {\n\tif testing.Short() {\n\t\tt.Skip(\"Skipping integration test\")\n\t}\n\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\tcapabilities := &openai.Capabilities{\n\t\tVision: \"openai\",\n\t}\n\n\toptions := newTestOptions(capabilities, nil)\n\tctx := newTestContext(capabilities)\n\n\t// Create content with empty URL\n\tcontent := agentContext.ContentPart{\n\t\tType: agentContext.ContentFile,\n\t\tFile: &agentContext.FileAttachment{\n\t\t\tURL: \"\",\n\t\t\tFilename: \"test.pdf\",\n\t\t},\n\t}\n\n\thandler := pdf.New(options)\n\t_, _, err := handler.Parse(ctx, content)\n\n\tassert.Error(t, err)\n\tassert.Contains(t, err.Error(), \"missing URL\")\n}\n\n// TestParseWithLocalPDFAndVisionSupport tests parsing a local PDF file when model supports vision\nfunc TestParseWithLocalPDFAndVisionSupport(t *testing.T) {\n\tif testing.Short() {\n\t\tt.Skip(\"Skipping integration test\")\n\t}\n\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\t// Check if test file exists\n\tpdfPath := getTestFilePath(\"test.pdf\")\n\tif _, err := os.Stat(pdfPath); os.IsNotExist(err) {\n\t\tt.Skipf(\"Test PDF file not found: %s\", pdfPath)\n\t}\n\n\t// Create capabilities with vision support\n\tcapabilities := &openai.Capabilities{\n\t\tVision: \"openai\",\n\t}\n\n\toptions := newTestOptions(capabilities, nil)\n\tctx := newTestContext(capabilities)\n\n\t// Create content with local file path\n\tcontent := agentContext.ContentPart{\n\t\tType: agentContext.ContentFile,\n\t\tFile: &agentContext.FileAttachment{\n\t\t\tURL: pdfPath,\n\t\t\tFilename: \"test.pdf\",\n\t\t},\n\t}\n\n\thandler := pdf.New(options)\n\tresult, refs, err := handler.Parse(ctx, content)\n\n\t// Should succeed - PDF converted to images\n\tassert.NoError(t, err)\n\tassert.Nil(t, refs)\n\n\t// When model supports vision, Parse returns the first image_url part\n\t// Use ParseMulti to get all pages as separate image_url parts\n\tassert.Equal(t, agentContext.ContentImageURL, result.Type)\n\tassert.NotNil(t, result.ImageURL)\n\tassert.NotEmpty(t, result.ImageURL.URL)\n\tt.Logf(\"PDF parse result type: %s, URL prefix: %s...\", result.Type, result.ImageURL.URL[:50])\n}\n\n// TestParseWithLocalPDFAndVisionAgent tests parsing a local PDF file using vision agent\nfunc TestParseWithLocalPDFAndVisionAgent(t *testing.T) {\n\tif testing.Short() {\n\t\tt.Skip(\"Skipping integration test\")\n\t}\n\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\t// Check if test file exists\n\tpdfPath := getTestFilePath(\"test.pdf\")\n\tif _, err := os.Stat(pdfPath); os.IsNotExist(err) {\n\t\tt.Skipf(\"Test PDF file not found: %s\", pdfPath)\n\t}\n\n\t// Create capabilities WITHOUT vision support\n\tcapabilities := &openai.Capabilities{\n\t\tVision: nil,\n\t}\n\n\t// Configure to use vision agent\n\tcompletionOptions := &agentContext.CompletionOptions{\n\t\tUses: &agentContext.Uses{\n\t\t\tVision: \"tests.vision-test\", // Use our test vision agent\n\t\t},\n\t}\n\n\toptions := newTestOptions(capabilities, completionOptions)\n\tctx := newTestContext(capabilities)\n\n\t// Create content with local file path\n\tcontent := agentContext.ContentPart{\n\t\tType: agentContext.ContentFile,\n\t\tFile: &agentContext.FileAttachment{\n\t\t\tURL: pdfPath,\n\t\t\tFilename: \"test.pdf\",\n\t\t},\n\t}\n\n\thandler := pdf.New(options)\n\tresult, refs, err := handler.Parse(ctx, content)\n\n\t// Should succeed - PDF converted to images and processed by vision agent\n\tassert.NoError(t, err)\n\tassert.Nil(t, refs)\n\tassert.Equal(t, agentContext.ContentText, result.Type)\n\tassert.NotEmpty(t, result.Text)\n\tt.Logf(\"PDF parse result (via vision agent): %s\", result.Text)\n}\n\n// TestParseMultiWithLocalPDF tests ParseMulti which returns separate parts for each page\nfunc TestParseMultiWithLocalPDF(t *testing.T) {\n\tif testing.Short() {\n\t\tt.Skip(\"Skipping integration test\")\n\t}\n\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\t// Check if test file exists\n\tpdfPath := getTestFilePath(\"test.pdf\")\n\tif _, err := os.Stat(pdfPath); os.IsNotExist(err) {\n\t\tt.Skipf(\"Test PDF file not found: %s\", pdfPath)\n\t}\n\n\t// Create capabilities with vision support\n\tcapabilities := &openai.Capabilities{\n\t\tVision: \"openai\",\n\t}\n\n\toptions := newTestOptions(capabilities, nil)\n\tctx := newTestContext(capabilities)\n\n\t// Create content with local file path\n\tcontent := agentContext.ContentPart{\n\t\tType: agentContext.ContentFile,\n\t\tFile: &agentContext.FileAttachment{\n\t\t\tURL: pdfPath,\n\t\t\tFilename: \"test.pdf\",\n\t\t},\n\t}\n\n\thandler := pdf.New(options)\n\tparts, refs, err := handler.ParseMulti(ctx, content)\n\n\t// Should succeed and return at least one part (one per page)\n\tassert.NoError(t, err)\n\tassert.Nil(t, refs)\n\tassert.NotEmpty(t, parts)\n\tt.Logf(\"PDF ParseMulti returned %d parts\", len(parts))\n\n\t// When model supports vision, each part should be image_url type\n\tfor i, part := range parts {\n\t\tassert.Equal(t, agentContext.ContentImageURL, part.Type)\n\t\tassert.NotNil(t, part.ImageURL)\n\t\tt.Logf(\" Part %d: type=%s, has URL=%v\", i+1, part.Type, part.ImageURL != nil && part.ImageURL.URL != \"\")\n\t}\n}\n\n// TestParseWithUnsupportedSource tests parsing PDF with unsupported source\nfunc TestParseWithUnsupportedSource(t *testing.T) {\n\tif testing.Short() {\n\t\tt.Skip(\"Skipping integration test\")\n\t}\n\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\tcapabilities := &openai.Capabilities{\n\t\tVision: \"openai\",\n\t}\n\n\toptions := newTestOptions(capabilities, nil)\n\tctx := newTestContext(capabilities)\n\n\t// Create content with HTTP URL (not implemented)\n\tcontent := agentContext.ContentPart{\n\t\tType: agentContext.ContentFile,\n\t\tFile: &agentContext.FileAttachment{\n\t\t\tURL: \"https://example.com/test.pdf\",\n\t\t\tFilename: \"test.pdf\",\n\t\t},\n\t}\n\n\thandler := pdf.New(options)\n\t_, _, err := handler.Parse(ctx, content)\n\n\tassert.Error(t, err)\n\tassert.Contains(t, err.Error(), \"HTTP URL fetch not implemented\")\n}\n\n// TestParseWithNonExistentFile tests parsing PDF with non-existent file\nfunc TestParseWithNonExistentFile(t *testing.T) {\n\tif testing.Short() {\n\t\tt.Skip(\"Skipping integration test\")\n\t}\n\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\tcapabilities := &openai.Capabilities{\n\t\tVision: \"openai\",\n\t}\n\n\toptions := newTestOptions(capabilities, nil)\n\tctx := newTestContext(capabilities)\n\n\t// Create content with non-existent file\n\tcontent := agentContext.ContentPart{\n\t\tType: agentContext.ContentFile,\n\t\tFile: &agentContext.FileAttachment{\n\t\t\tURL: \"/non/existent/path/test.pdf\",\n\t\t\tFilename: \"test.pdf\",\n\t\t},\n\t}\n\n\thandler := pdf.New(options)\n\t_, _, err := handler.Parse(ctx, content)\n\n\tassert.Error(t, err)\n\tassert.Contains(t, err.Error(), \"unsupported PDF source\")\n}\n\n// TestSilentLoadingOption tests that SilentLoading option is respected\nfunc TestSilentLoadingOption(t *testing.T) {\n\tif testing.Short() {\n\t\tt.Skip(\"Skipping integration test\")\n\t}\n\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\tcapabilities := &openai.Capabilities{\n\t\tVision: \"openai\",\n\t}\n\n\t// Create options with SilentLoading enabled\n\toptions := &contentTypes.Options{\n\t\tCapabilities: capabilities,\n\t\tSilentLoading: true,\n\t}\n\n\t// This test just verifies the option can be set\n\t// The actual behavior is tested in the image handler tests\n\thandler := pdf.New(options)\n\tassert.NotNil(t, handler)\n\tassert.True(t, options.SilentLoading)\n}\n" }, { "path": "agent/content/pptx/pptx.go", "content": "package pptx\n\nimport (\n\t\"fmt\"\n\t\"os\"\n\t\"strings\"\n\n\t\"github.com/yaoapp/gou/office\"\n\t\"github.com/yaoapp/yao/agent/content/types\"\n\tagentContext \"github.com/yaoapp/yao/agent/context\"\n\tsearchTypes \"github.com/yaoapp/yao/agent/search/types\"\n\t\"github.com/yaoapp/yao/attachment\"\n)\n\n// Pptx handles PPTX content\ntype Pptx struct {\n\toptions *types.Options\n}\n\n// New creates a new PPTX handler\nfunc New(options *types.Options) *Pptx {\n\treturn &Pptx{options: options}\n}\n\n// Parse parses PPTX content and returns text\nfunc (h *Pptx) Parse(ctx *agentContext.Context, content agentContext.ContentPart) (agentContext.ContentPart, []*searchTypes.Reference, error) {\n\tif content.File == nil || content.File.URL == \"\" {\n\t\treturn content, nil, fmt.Errorf(\"file content missing URL\")\n\t}\n\n\turl := content.File.URL\n\n\t// Check cache first\n\tcachedText, found, err := h.readFromCache(ctx, url)\n\tif err == nil && found {\n\t\treturn agentContext.ContentPart{\n\t\t\tType: agentContext.ContentText,\n\t\t\tText: cachedText,\n\t\t}, nil, nil\n\t}\n\n\t// Read PPTX file\n\tdata, err := h.readFile(ctx, url)\n\tif err != nil {\n\t\treturn content, nil, fmt.Errorf(\"failed to read PPTX: %w\", err)\n\t}\n\n\t// Parse PPTX using gou/office\n\tparser := office.NewParser()\n\tresult, err := parser.Parse(data)\n\tif err != nil {\n\t\treturn content, nil, fmt.Errorf(\"failed to parse PPTX: %w\", err)\n\t}\n\n\ttext := result.Markdown\n\tif text == \"\" {\n\t\treturn content, nil, fmt.Errorf(\"no text content extracted from PPTX\")\n\t}\n\n\t// Cache the result\n\tif err := h.saveToCache(ctx, url, text); err != nil {\n\t\t// Log warning but don't fail\n\t\tfmt.Printf(\"Warning: failed to cache PPTX text: %v\\n\", err)\n\t}\n\n\treturn agentContext.ContentPart{\n\t\tType: agentContext.ContentText,\n\t\tText: text,\n\t}, nil, nil\n}\n\n// readFile reads PPTX content from various sources\nfunc (h *Pptx) readFile(ctx *agentContext.Context, url string) ([]byte, error) {\n\tif strings.HasPrefix(url, \"__\") {\n\t\treturn h.readFromUploader(ctx, url)\n\t}\n\n\tif strings.HasPrefix(url, \"http://\") || strings.HasPrefix(url, \"https://\") {\n\t\treturn nil, fmt.Errorf(\"HTTP URL fetch not implemented yet: %s\", url)\n\t}\n\n\t// Try to read as local file path\n\tif _, err := os.Stat(url); err == nil {\n\t\treturn os.ReadFile(url)\n\t}\n\n\treturn nil, fmt.Errorf(\"unsupported PPTX source: %s\", url)\n}\n\n// readFromUploader reads PPTX content from file uploader\nfunc (h *Pptx) readFromUploader(ctx *agentContext.Context, wrapper string) ([]byte, error) {\n\tuploaderName, fileID, ok := attachment.Parse(wrapper)\n\tif !ok {\n\t\treturn nil, fmt.Errorf(\"invalid uploader wrapper format: %s\", wrapper)\n\t}\n\n\tmanager, exists := attachment.Managers[uploaderName]\n\tif !exists {\n\t\treturn nil, fmt.Errorf(\"uploader '%s' not found\", uploaderName)\n\t}\n\n\tdata, err := manager.Read(ctx.Context, fileID)\n\tif err != nil {\n\t\treturn nil, fmt.Errorf(\"failed to read file: %w\", err)\n\t}\n\n\treturn data, nil\n}\n\n// readFromCache reads cached text content for a PPTX\nfunc (h *Pptx) readFromCache(ctx *agentContext.Context, url string) (string, bool, error) {\n\tuploaderName, fileID, isWrapper := attachment.Parse(url)\n\tif !isWrapper {\n\t\treturn \"\", false, nil\n\t}\n\n\tmanager, exists := attachment.Managers[uploaderName]\n\tif !exists {\n\t\treturn \"\", false, nil\n\t}\n\n\ttext, err := manager.GetText(ctx.Context, fileID, false)\n\tif err == nil && text != \"\" {\n\t\treturn text, true, nil\n\t}\n\n\treturn \"\", false, nil\n}\n\n// saveToCache saves processed text to cache\nfunc (h *Pptx) saveToCache(ctx *agentContext.Context, url string, text string) error {\n\tuploaderName, fileID, isWrapper := attachment.Parse(url)\n\tif !isWrapper {\n\t\treturn nil\n\t}\n\n\tmanager, exists := attachment.Managers[uploaderName]\n\tif !exists {\n\t\treturn nil\n\t}\n\n\treturn manager.SaveText(ctx.Context, fileID, text)\n}\n" }, { "path": "agent/content/pptx/pptx_test.go", "content": "package pptx_test\n\nimport (\n\tstdContext \"context\"\n\t\"os\"\n\t\"path/filepath\"\n\t\"testing\"\n\n\t\"github.com/stretchr/testify/assert\"\n\t\"github.com/yaoapp/gou/connector/openai\"\n\t\"github.com/yaoapp/yao/agent/content/pptx\"\n\tcontentTypes \"github.com/yaoapp/yao/agent/content/types\"\n\tagentContext \"github.com/yaoapp/yao/agent/context\"\n\t\"github.com/yaoapp/yao/agent/output/message\"\n\t\"github.com/yaoapp/yao/agent/testutils\"\n\toauthTypes \"github.com/yaoapp/yao/openapi/oauth/types\"\n)\n\nconst testFilesDir = \"assistants/tests/vision-helper/tests\"\n\nfunc newTestContext() *agentContext.Context {\n\tauthorized := &oauthTypes.AuthorizedInfo{\n\t\tSubject: \"test-user\",\n\t\tClientID: \"test-client-id\",\n\t\tUserID: \"test-user-123\",\n\t}\n\tctx := agentContext.New(stdContext.Background(), authorized, \"test-chat\")\n\tctx.AssistantID = \"test-assistant\"\n\tctx.Locale = \"en-us\"\n\tctx.IDGenerator = message.NewIDGenerator()\n\treturn ctx\n}\n\nfunc newTestOptions() *contentTypes.Options {\n\treturn &contentTypes.Options{\n\t\tCapabilities: &openai.Capabilities{},\n\t}\n}\n\nfunc getTestFilePath(filename string) string {\n\tyaoRoot := os.Getenv(\"YAO_TEST_APPLICATION\")\n\tif yaoRoot == \"\" {\n\t\tyaoRoot = os.Getenv(\"YAO_ROOT\")\n\t}\n\treturn filepath.Join(yaoRoot, testFilesDir, filename)\n}\n\n// TestParseWithMissingURL tests parsing PPTX with missing URL\nfunc TestParseWithMissingURL(t *testing.T) {\n\tif testing.Short() {\n\t\tt.Skip(\"Skipping integration test\")\n\t}\n\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\toptions := newTestOptions()\n\tctx := newTestContext()\n\n\tcontent := agentContext.ContentPart{\n\t\tType: agentContext.ContentFile,\n\t\tFile: nil,\n\t}\n\n\thandler := pptx.New(options)\n\t_, _, err := handler.Parse(ctx, content)\n\n\tassert.Error(t, err)\n\tassert.Contains(t, err.Error(), \"missing URL\")\n}\n\n// TestParseWithLocalPptx tests parsing a local PPTX file\nfunc TestParseWithLocalPptx(t *testing.T) {\n\tif testing.Short() {\n\t\tt.Skip(\"Skipping integration test\")\n\t}\n\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\tpptxPath := getTestFilePath(\"pptx.pptx\")\n\tif _, err := os.Stat(pptxPath); os.IsNotExist(err) {\n\t\tt.Skipf(\"Test PPTX file not found: %s\", pptxPath)\n\t}\n\n\toptions := newTestOptions()\n\tctx := newTestContext()\n\n\tcontent := agentContext.ContentPart{\n\t\tType: agentContext.ContentFile,\n\t\tFile: &agentContext.FileAttachment{\n\t\t\tURL: pptxPath,\n\t\t\tFilename: \"pptx.pptx\",\n\t\t},\n\t}\n\n\thandler := pptx.New(options)\n\tresult, refs, err := handler.Parse(ctx, content)\n\n\tassert.NoError(t, err)\n\tassert.Nil(t, refs)\n\tassert.Equal(t, agentContext.ContentText, result.Type)\n\tassert.NotEmpty(t, result.Text)\n\tt.Logf(\"PPTX parse result (first 500 chars): %.500s...\", result.Text)\n}\n\n// TestParseWithNonExistentFile tests parsing PPTX with non-existent file\nfunc TestParseWithNonExistentFile(t *testing.T) {\n\tif testing.Short() {\n\t\tt.Skip(\"Skipping integration test\")\n\t}\n\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\toptions := newTestOptions()\n\tctx := newTestContext()\n\n\tcontent := agentContext.ContentPart{\n\t\tType: agentContext.ContentFile,\n\t\tFile: &agentContext.FileAttachment{\n\t\t\tURL: \"/non/existent/path/test.pptx\",\n\t\t\tFilename: \"test.pptx\",\n\t\t},\n\t}\n\n\thandler := pptx.New(options)\n\t_, _, err := handler.Parse(ctx, content)\n\n\tassert.Error(t, err)\n\tassert.Contains(t, err.Error(), \"unsupported PPTX source\")\n}\n" }, { "path": "agent/content/text/text.go", "content": "package text\n\nimport (\n\t\"fmt\"\n\t\"os\"\n\t\"path/filepath\"\n\t\"strings\"\n\n\t\"github.com/yaoapp/yao/agent/content/types\"\n\tagentContext \"github.com/yaoapp/yao/agent/context\"\n\tsearchTypes \"github.com/yaoapp/yao/agent/search/types\"\n\t\"github.com/yaoapp/yao/attachment\"\n)\n\n// SupportedExtensions text file extensions\nvar SupportedExtensions = map[string]bool{\n\t// Markdown\n\t\".md\": true,\n\t\".markdown\": true,\n\t// Plain text\n\t\".txt\": true,\n\t// Code files\n\t\".go\": true,\n\t\".ts\": true,\n\t\".tsx\": true,\n\t\".js\": true,\n\t\".jsx\": true,\n\t\".py\": true,\n\t\".java\": true,\n\t\".c\": true,\n\t\".cpp\": true,\n\t\".h\": true,\n\t\".hpp\": true,\n\t\".rs\": true,\n\t\".rb\": true,\n\t\".php\": true,\n\t\".swift\": true,\n\t\".kt\": true,\n\t\".scala\": true,\n\t\".sh\": true,\n\t\".bash\": true,\n\t\".zsh\": true,\n\t\".fish\": true,\n\t\".ps1\": true,\n\t\".bat\": true,\n\t\".cmd\": true,\n\t\".sql\": true,\n\t\".r\": true,\n\t\".lua\": true,\n\t\".perl\": true,\n\t\".pl\": true,\n\t\".groovy\": true,\n\t\".dart\": true,\n\t\".elm\": true,\n\t\".ex\": true,\n\t\".exs\": true,\n\t\".erl\": true,\n\t\".hs\": true,\n\t\".clj\": true,\n\t\".lisp\": true,\n\t\".vim\": true,\n\t// Config files\n\t\".json\": true,\n\t\".jsonc\": true,\n\t\".yaml\": true,\n\t\".yml\": true,\n\t\".toml\": true,\n\t\".ini\": true,\n\t\".conf\": true,\n\t\".cfg\": true,\n\t\".env\": true,\n\t\".yao\": true,\n\t// Web files\n\t\".html\": true,\n\t\".htm\": true,\n\t\".css\": true,\n\t\".scss\": true,\n\t\".sass\": true,\n\t\".less\": true,\n\t\".xml\": true,\n\t\".svg\": true,\n\t// Documentation\n\t\".rst\": true,\n\t\".tex\": true,\n\t\".latex\": true,\n\t\".org\": true,\n\t\".adoc\": true,\n\t// Data files\n\t\".csv\": true,\n\t\".tsv\": true,\n\t// Log files\n\t\".log\": true,\n}\n\n// Text handles text file content\ntype Text struct {\n\toptions *types.Options\n}\n\n// New creates a new text handler\nfunc New(options *types.Options) *Text {\n\treturn &Text{options: options}\n}\n\n// IsSupportedExtension checks if a file extension is supported\nfunc IsSupportedExtension(filename string) bool {\n\text := strings.ToLower(filepath.Ext(filename))\n\treturn SupportedExtensions[ext]\n}\n\n// Parse parses text file content and returns text\nfunc (h *Text) Parse(ctx *agentContext.Context, content agentContext.ContentPart) (agentContext.ContentPart, []*searchTypes.Reference, error) {\n\tif content.File == nil || content.File.URL == \"\" {\n\t\treturn content, nil, fmt.Errorf(\"file content missing URL\")\n\t}\n\n\turl := content.File.URL\n\tfilename := content.File.Filename\n\n\t// Check cache first\n\tcachedText, found, err := h.readFromCache(ctx, url)\n\tif err == nil && found {\n\t\treturn agentContext.ContentPart{\n\t\t\tType: agentContext.ContentText,\n\t\t\tText: cachedText,\n\t\t}, nil, nil\n\t}\n\n\t// Read text file\n\tdata, err := h.readFile(ctx, url)\n\tif err != nil {\n\t\treturn content, nil, fmt.Errorf(\"failed to read text file: %w\", err)\n\t}\n\n\t// Convert to string\n\ttext := string(data)\n\n\t// Add file type context if it's a code file\n\text := strings.ToLower(filepath.Ext(filename))\n\tif isCodeFile(ext) {\n\t\t// Wrap in markdown code block with language hint\n\t\tlang := getLanguageFromExt(ext)\n\t\ttext = fmt.Sprintf(\"```%s\\n%s\\n```\", lang, text)\n\t}\n\n\t// Cache the result\n\tif err := h.saveToCache(ctx, url, text); err != nil {\n\t\t// Log warning but don't fail\n\t\tfmt.Printf(\"Warning: failed to cache text: %v\\n\", err)\n\t}\n\n\treturn agentContext.ContentPart{\n\t\tType: agentContext.ContentText,\n\t\tText: text,\n\t}, nil, nil\n}\n\n// ParseRaw parses any file as raw text content without code block wrapping\n// This is used as a fallback for unsupported file types\nfunc (h *Text) ParseRaw(ctx *agentContext.Context, content agentContext.ContentPart) (agentContext.ContentPart, []*searchTypes.Reference, error) {\n\tif content.File == nil || content.File.URL == \"\" {\n\t\treturn content, nil, fmt.Errorf(\"file content missing URL\")\n\t}\n\n\turl := content.File.URL\n\tfilename := content.File.Filename\n\n\t// Check cache first\n\tcachedText, found, err := h.readFromCache(ctx, url)\n\tif err == nil && found {\n\t\treturn agentContext.ContentPart{\n\t\t\tType: agentContext.ContentText,\n\t\t\tText: cachedText,\n\t\t}, nil, nil\n\t}\n\n\t// Read file\n\tdata, err := h.readFile(ctx, url)\n\tif err != nil {\n\t\treturn content, nil, fmt.Errorf(\"failed to read file: %w\", err)\n\t}\n\n\t// Convert to string directly (no code block wrapping)\n\ttext := string(data)\n\n\t// Add filename as context\n\tif filename != \"\" {\n\t\ttext = fmt.Sprintf(\"File: %s\\n\\n%s\", filename, text)\n\t}\n\n\t// Cache the result\n\tif err := h.saveToCache(ctx, url, text); err != nil {\n\t\t// Log warning but don't fail\n\t\tfmt.Printf(\"Warning: failed to cache text: %v\\n\", err)\n\t}\n\n\treturn agentContext.ContentPart{\n\t\tType: agentContext.ContentText,\n\t\tText: text,\n\t}, nil, nil\n}\n\n// readFile reads text content from various sources\nfunc (h *Text) readFile(ctx *agentContext.Context, url string) ([]byte, error) {\n\tif strings.HasPrefix(url, \"__\") {\n\t\treturn h.readFromUploader(ctx, url)\n\t}\n\n\tif strings.HasPrefix(url, \"http://\") || strings.HasPrefix(url, \"https://\") {\n\t\treturn nil, fmt.Errorf(\"HTTP URL fetch not implemented yet: %s\", url)\n\t}\n\n\t// Try to read as local file path\n\tif _, err := os.Stat(url); err == nil {\n\t\treturn os.ReadFile(url)\n\t}\n\n\treturn nil, fmt.Errorf(\"unsupported text file source: %s\", url)\n}\n\n// readFromUploader reads text content from file uploader\nfunc (h *Text) readFromUploader(ctx *agentContext.Context, wrapper string) ([]byte, error) {\n\tuploaderName, fileID, ok := attachment.Parse(wrapper)\n\tif !ok {\n\t\treturn nil, fmt.Errorf(\"invalid uploader wrapper format: %s\", wrapper)\n\t}\n\n\tmanager, exists := attachment.Managers[uploaderName]\n\tif !exists {\n\t\treturn nil, fmt.Errorf(\"uploader '%s' not found\", uploaderName)\n\t}\n\n\tdata, err := manager.Read(ctx.Context, fileID)\n\tif err != nil {\n\t\treturn nil, fmt.Errorf(\"failed to read file: %w\", err)\n\t}\n\n\treturn data, nil\n}\n\n// readFromCache reads cached text content\nfunc (h *Text) readFromCache(ctx *agentContext.Context, url string) (string, bool, error) {\n\tuploaderName, fileID, isWrapper := attachment.Parse(url)\n\tif !isWrapper {\n\t\treturn \"\", false, nil\n\t}\n\n\tmanager, exists := attachment.Managers[uploaderName]\n\tif !exists {\n\t\treturn \"\", false, nil\n\t}\n\n\ttext, err := manager.GetText(ctx.Context, fileID, false)\n\tif err == nil && text != \"\" {\n\t\treturn text, true, nil\n\t}\n\n\treturn \"\", false, nil\n}\n\n// saveToCache saves processed text to cache\nfunc (h *Text) saveToCache(ctx *agentContext.Context, url string, text string) error {\n\tuploaderName, fileID, isWrapper := attachment.Parse(url)\n\tif !isWrapper {\n\t\treturn nil\n\t}\n\n\tmanager, exists := attachment.Managers[uploaderName]\n\tif !exists {\n\t\treturn nil\n\t}\n\n\treturn manager.SaveText(ctx.Context, fileID, text)\n}\n\n// isCodeFile checks if the extension represents a code file\nfunc isCodeFile(ext string) bool {\n\tcodeExts := map[string]bool{\n\t\t\".go\": true, \".ts\": true, \".tsx\": true, \".js\": true, \".jsx\": true,\n\t\t\".py\": true, \".java\": true, \".c\": true, \".cpp\": true, \".h\": true,\n\t\t\".hpp\": true, \".rs\": true, \".rb\": true, \".php\": true, \".swift\": true,\n\t\t\".kt\": true, \".scala\": true, \".sh\": true, \".bash\": true, \".zsh\": true,\n\t\t\".sql\": true, \".r\": true, \".lua\": true, \".perl\": true, \".pl\": true,\n\t\t\".groovy\": true, \".dart\": true, \".elm\": true, \".ex\": true, \".exs\": true,\n\t\t\".erl\": true, \".hs\": true, \".clj\": true, \".lisp\": true, \".vim\": true,\n\t}\n\treturn codeExts[ext]\n}\n\n// getLanguageFromExt returns the language name for markdown code block\nfunc getLanguageFromExt(ext string) string {\n\tlangMap := map[string]string{\n\t\t\".go\": \"go\",\n\t\t\".ts\": \"typescript\",\n\t\t\".tsx\": \"tsx\",\n\t\t\".js\": \"javascript\",\n\t\t\".jsx\": \"jsx\",\n\t\t\".py\": \"python\",\n\t\t\".java\": \"java\",\n\t\t\".c\": \"c\",\n\t\t\".cpp\": \"cpp\",\n\t\t\".h\": \"c\",\n\t\t\".hpp\": \"cpp\",\n\t\t\".rs\": \"rust\",\n\t\t\".rb\": \"ruby\",\n\t\t\".php\": \"php\",\n\t\t\".swift\": \"swift\",\n\t\t\".kt\": \"kotlin\",\n\t\t\".scala\": \"scala\",\n\t\t\".sh\": \"bash\",\n\t\t\".bash\": \"bash\",\n\t\t\".zsh\": \"zsh\",\n\t\t\".fish\": \"fish\",\n\t\t\".ps1\": \"powershell\",\n\t\t\".bat\": \"batch\",\n\t\t\".cmd\": \"batch\",\n\t\t\".sql\": \"sql\",\n\t\t\".r\": \"r\",\n\t\t\".lua\": \"lua\",\n\t\t\".perl\": \"perl\",\n\t\t\".pl\": \"perl\",\n\t\t\".groovy\": \"groovy\",\n\t\t\".dart\": \"dart\",\n\t\t\".elm\": \"elm\",\n\t\t\".ex\": \"elixir\",\n\t\t\".exs\": \"elixir\",\n\t\t\".erl\": \"erlang\",\n\t\t\".hs\": \"haskell\",\n\t\t\".clj\": \"clojure\",\n\t\t\".lisp\": \"lisp\",\n\t\t\".vim\": \"vim\",\n\t\t\".json\": \"json\",\n\t\t\".jsonc\": \"jsonc\",\n\t\t\".yaml\": \"yaml\",\n\t\t\".yml\": \"yaml\",\n\t\t\".toml\": \"toml\",\n\t\t\".xml\": \"xml\",\n\t\t\".html\": \"html\",\n\t\t\".htm\": \"html\",\n\t\t\".css\": \"css\",\n\t\t\".scss\": \"scss\",\n\t\t\".sass\": \"sass\",\n\t\t\".less\": \"less\",\n\t\t\".svg\": \"svg\",\n\t\t\".yao\": \"json\",\n\t}\n\n\tif lang, ok := langMap[ext]; ok {\n\t\treturn lang\n\t}\n\treturn \"\"\n}\n" }, { "path": "agent/content/text/text_test.go", "content": "package text_test\n\nimport (\n\tstdContext \"context\"\n\t\"os\"\n\t\"path/filepath\"\n\t\"strings\"\n\t\"testing\"\n\n\t\"github.com/stretchr/testify/assert\"\n\t\"github.com/yaoapp/gou/connector/openai\"\n\t\"github.com/yaoapp/yao/agent/content/text\"\n\tcontentTypes \"github.com/yaoapp/yao/agent/content/types\"\n\tagentContext \"github.com/yaoapp/yao/agent/context\"\n\t\"github.com/yaoapp/yao/agent/output/message\"\n\t\"github.com/yaoapp/yao/agent/testutils\"\n\toauthTypes \"github.com/yaoapp/yao/openapi/oauth/types\"\n)\n\nconst testFilesDir = \"assistants/tests/vision-helper/tests\"\n\nfunc newTestContext() *agentContext.Context {\n\tauthorized := &oauthTypes.AuthorizedInfo{\n\t\tSubject: \"test-user\",\n\t\tClientID: \"test-client-id\",\n\t\tUserID: \"test-user-123\",\n\t}\n\tctx := agentContext.New(stdContext.Background(), authorized, \"test-chat\")\n\tctx.AssistantID = \"test-assistant\"\n\tctx.Locale = \"en-us\"\n\tctx.IDGenerator = message.NewIDGenerator()\n\treturn ctx\n}\n\nfunc newTestOptions() *contentTypes.Options {\n\treturn &contentTypes.Options{\n\t\tCapabilities: &openai.Capabilities{},\n\t}\n}\n\nfunc getTestFilePath(filename string) string {\n\tyaoRoot := os.Getenv(\"YAO_TEST_APPLICATION\")\n\tif yaoRoot == \"\" {\n\t\tyaoRoot = os.Getenv(\"YAO_ROOT\")\n\t}\n\treturn filepath.Join(yaoRoot, testFilesDir, filename)\n}\n\n// TestIsSupportedExtension tests the IsSupportedExtension function\nfunc TestIsSupportedExtension(t *testing.T) {\n\t// Supported extensions\n\tassert.True(t, text.IsSupportedExtension(\"test.md\"))\n\tassert.True(t, text.IsSupportedExtension(\"test.txt\"))\n\tassert.True(t, text.IsSupportedExtension(\"test.go\"))\n\tassert.True(t, text.IsSupportedExtension(\"test.ts\"))\n\tassert.True(t, text.IsSupportedExtension(\"test.json\"))\n\tassert.True(t, text.IsSupportedExtension(\"test.jsonc\"))\n\tassert.True(t, text.IsSupportedExtension(\"test.yao\"))\n\tassert.True(t, text.IsSupportedExtension(\"test.yaml\"))\n\tassert.True(t, text.IsSupportedExtension(\"test.yml\"))\n\tassert.True(t, text.IsSupportedExtension(\"test.py\"))\n\tassert.True(t, text.IsSupportedExtension(\"test.js\"))\n\tassert.True(t, text.IsSupportedExtension(\"test.css\"))\n\tassert.True(t, text.IsSupportedExtension(\"test.html\"))\n\n\t// Unsupported extensions\n\tassert.False(t, text.IsSupportedExtension(\"test.docx\"))\n\tassert.False(t, text.IsSupportedExtension(\"test.pptx\"))\n\tassert.False(t, text.IsSupportedExtension(\"test.pdf\"))\n\tassert.False(t, text.IsSupportedExtension(\"test.png\"))\n\tassert.False(t, text.IsSupportedExtension(\"test.jpg\"))\n\tassert.False(t, text.IsSupportedExtension(\"test.exe\"))\n\tassert.False(t, text.IsSupportedExtension(\"test.zip\"))\n}\n\n// TestParseWithMissingURL tests parsing text with missing URL\nfunc TestParseWithMissingURL(t *testing.T) {\n\tif testing.Short() {\n\t\tt.Skip(\"Skipping integration test\")\n\t}\n\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\toptions := newTestOptions()\n\tctx := newTestContext()\n\n\tcontent := agentContext.ContentPart{\n\t\tType: agentContext.ContentFile,\n\t\tFile: nil,\n\t}\n\n\thandler := text.New(options)\n\t_, _, err := handler.Parse(ctx, content)\n\n\tassert.Error(t, err)\n\tassert.Contains(t, err.Error(), \"missing URL\")\n}\n\n// TestParseWithLocalTextFile tests parsing a local text file\nfunc TestParseWithLocalTextFile(t *testing.T) {\n\tif testing.Short() {\n\t\tt.Skip(\"Skipping integration test\")\n\t}\n\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\ttxtPath := getTestFilePath(\"text.txt\")\n\tif _, err := os.Stat(txtPath); os.IsNotExist(err) {\n\t\tt.Skipf(\"Test text file not found: %s\", txtPath)\n\t}\n\n\toptions := newTestOptions()\n\tctx := newTestContext()\n\n\tcontent := agentContext.ContentPart{\n\t\tType: agentContext.ContentFile,\n\t\tFile: &agentContext.FileAttachment{\n\t\t\tURL: txtPath,\n\t\t\tFilename: \"text.txt\",\n\t\t},\n\t}\n\n\thandler := text.New(options)\n\tresult, refs, err := handler.Parse(ctx, content)\n\n\tassert.NoError(t, err)\n\tassert.Nil(t, refs)\n\tassert.Equal(t, agentContext.ContentText, result.Type)\n\tassert.NotEmpty(t, result.Text)\n\tt.Logf(\"Text parse result: %s\", result.Text)\n}\n\n// TestParseWithLocalMarkdownFile tests parsing a local markdown file\nfunc TestParseWithLocalMarkdownFile(t *testing.T) {\n\tif testing.Short() {\n\t\tt.Skip(\"Skipping integration test\")\n\t}\n\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\tmdPath := getTestFilePath(\"test.md\")\n\tif _, err := os.Stat(mdPath); os.IsNotExist(err) {\n\t\tt.Skipf(\"Test markdown file not found: %s\", mdPath)\n\t}\n\n\toptions := newTestOptions()\n\tctx := newTestContext()\n\n\tcontent := agentContext.ContentPart{\n\t\tType: agentContext.ContentFile,\n\t\tFile: &agentContext.FileAttachment{\n\t\t\tURL: mdPath,\n\t\t\tFilename: \"test.md\",\n\t\t},\n\t}\n\n\thandler := text.New(options)\n\tresult, refs, err := handler.Parse(ctx, content)\n\n\tassert.NoError(t, err)\n\tassert.Nil(t, refs)\n\tassert.Equal(t, agentContext.ContentText, result.Type)\n\tassert.NotEmpty(t, result.Text)\n\tt.Logf(\"Markdown parse result: %s\", result.Text)\n}\n\n// TestParseWithLocalCodeFile tests parsing a local code file (TypeScript)\nfunc TestParseWithLocalCodeFile(t *testing.T) {\n\tif testing.Short() {\n\t\tt.Skip(\"Skipping integration test\")\n\t}\n\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\ttsPath := getTestFilePath(\"code.ts\")\n\tif _, err := os.Stat(tsPath); os.IsNotExist(err) {\n\t\tt.Skipf(\"Test TypeScript file not found: %s\", tsPath)\n\t}\n\n\toptions := newTestOptions()\n\tctx := newTestContext()\n\n\tcontent := agentContext.ContentPart{\n\t\tType: agentContext.ContentFile,\n\t\tFile: &agentContext.FileAttachment{\n\t\t\tURL: tsPath,\n\t\t\tFilename: \"code.ts\",\n\t\t},\n\t}\n\n\thandler := text.New(options)\n\tresult, refs, err := handler.Parse(ctx, content)\n\n\tassert.NoError(t, err)\n\tassert.Nil(t, refs)\n\tassert.Equal(t, agentContext.ContentText, result.Type)\n\tassert.NotEmpty(t, result.Text)\n\t// Code files should be wrapped in markdown code blocks\n\tassert.True(t, strings.HasPrefix(result.Text, \"```typescript\"))\n\tassert.True(t, strings.HasSuffix(strings.TrimSpace(result.Text), \"```\"))\n\tt.Logf(\"Code parse result (first 500 chars): %.500s...\", result.Text)\n}\n\n// TestParseWithLocalYaoFile tests parsing a local .yao file\nfunc TestParseWithLocalYaoFile(t *testing.T) {\n\tif testing.Short() {\n\t\tt.Skip(\"Skipping integration test\")\n\t}\n\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\tyaoPath := getTestFilePath(\"hero.mod.yao\")\n\tif _, err := os.Stat(yaoPath); os.IsNotExist(err) {\n\t\tt.Skipf(\"Test .yao file not found: %s\", yaoPath)\n\t}\n\n\toptions := newTestOptions()\n\tctx := newTestContext()\n\n\tcontent := agentContext.ContentPart{\n\t\tType: agentContext.ContentFile,\n\t\tFile: &agentContext.FileAttachment{\n\t\t\tURL: yaoPath,\n\t\t\tFilename: \"hero.mod.yao\",\n\t\t},\n\t}\n\n\thandler := text.New(options)\n\tresult, refs, err := handler.Parse(ctx, content)\n\n\tassert.NoError(t, err)\n\tassert.Nil(t, refs)\n\tassert.Equal(t, agentContext.ContentText, result.Type)\n\tassert.NotEmpty(t, result.Text)\n\tt.Logf(\"Yao file parse result: %s\", result.Text)\n}\n\n// TestParseWithLocalJsonFile tests parsing a local JSON file\nfunc TestParseWithLocalJsonFile(t *testing.T) {\n\tif testing.Short() {\n\t\tt.Skip(\"Skipping integration test\")\n\t}\n\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\tjsonPath := getTestFilePath(\"test.json\")\n\tif _, err := os.Stat(jsonPath); os.IsNotExist(err) {\n\t\tt.Skipf(\"Test JSON file not found: %s\", jsonPath)\n\t}\n\n\toptions := newTestOptions()\n\tctx := newTestContext()\n\n\tcontent := agentContext.ContentPart{\n\t\tType: agentContext.ContentFile,\n\t\tFile: &agentContext.FileAttachment{\n\t\t\tURL: jsonPath,\n\t\t\tFilename: \"test.json\",\n\t\t},\n\t}\n\n\thandler := text.New(options)\n\tresult, refs, err := handler.Parse(ctx, content)\n\n\tassert.NoError(t, err)\n\tassert.Nil(t, refs)\n\tassert.Equal(t, agentContext.ContentText, result.Type)\n\tassert.NotEmpty(t, result.Text)\n\tt.Logf(\"JSON parse result: %s\", result.Text)\n}\n\n// TestParseWithNonExistentFile tests parsing text with non-existent file\nfunc TestParseWithNonExistentFile(t *testing.T) {\n\tif testing.Short() {\n\t\tt.Skip(\"Skipping integration test\")\n\t}\n\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\toptions := newTestOptions()\n\tctx := newTestContext()\n\n\tcontent := agentContext.ContentPart{\n\t\tType: agentContext.ContentFile,\n\t\tFile: &agentContext.FileAttachment{\n\t\t\tURL: \"/non/existent/path/test.txt\",\n\t\t\tFilename: \"test.txt\",\n\t\t},\n\t}\n\n\thandler := text.New(options)\n\t_, _, err := handler.Parse(ctx, content)\n\n\tassert.Error(t, err)\n\tassert.Contains(t, err.Error(), \"unsupported text file source\")\n}\n\n// TestParseRawWithLocalFile tests ParseRaw with a local file\nfunc TestParseRawWithLocalFile(t *testing.T) {\n\tif testing.Short() {\n\t\tt.Skip(\"Skipping integration test\")\n\t}\n\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\ttxtPath := getTestFilePath(\"text.txt\")\n\tif _, err := os.Stat(txtPath); os.IsNotExist(err) {\n\t\tt.Skipf(\"Test text file not found: %s\", txtPath)\n\t}\n\n\toptions := newTestOptions()\n\tctx := newTestContext()\n\n\tcontent := agentContext.ContentPart{\n\t\tType: agentContext.ContentFile,\n\t\tFile: &agentContext.FileAttachment{\n\t\t\tURL: txtPath,\n\t\t\tFilename: \"text.txt\",\n\t\t},\n\t}\n\n\thandler := text.New(options)\n\tresult, refs, err := handler.ParseRaw(ctx, content)\n\n\tassert.NoError(t, err)\n\tassert.Nil(t, refs)\n\tassert.Equal(t, agentContext.ContentText, result.Type)\n\tassert.NotEmpty(t, result.Text)\n\t// ParseRaw should include filename as context\n\tassert.True(t, strings.HasPrefix(result.Text, \"File: text.txt\"))\n\tt.Logf(\"ParseRaw result: %s\", result.Text)\n}\n" }, { "path": "agent/content/tools/tools.go", "content": "package tools\n\nimport (\n\t\"context\"\n\t\"fmt\"\n\n\tjsoniter \"github.com/json-iterator/go\"\n\t\"github.com/yaoapp/gou/mcp\"\n\t\"github.com/yaoapp/kun/log\"\n\t\"github.com/yaoapp/yao/agent/caller\"\n\tagentContext \"github.com/yaoapp/yao/agent/context\"\n)\n\n// CallAgent calls an agent to process content (vision, audio, etc.)\nfunc CallAgent(ctx *agentContext.Context, agentID string, message agentContext.Message) (string, error) {\n\tif caller.AgentGetterFunc == nil {\n\t\treturn \"\", fmt.Errorf(\"AgentGetterFunc not initialized\")\n\t}\n\n\t// Load the agent by ID using the injected function\n\tagent, err := caller.AgentGetterFunc(agentID)\n\tif err != nil {\n\t\treturn \"\", fmt.Errorf(\"failed to load agent %s: %w\", agentID, err)\n\t}\n\n\t// Call the agent with the message\n\tmessages := []agentContext.Message{message}\n\n\t// For A2A calls, skip history and output (we only need the response data)\n\topts := &agentContext.Options{Skip: &agentContext.Skip{History: true, Output: true}}\n\tresponse, err := agent.Stream(ctx, messages, opts)\n\tif err != nil {\n\t\treturn \"\", fmt.Errorf(\"failed to call agent %s: %w\", agentID, err)\n\t}\n\n\t// Extract text from agent response\n\treturn ExtractTextFromResponse(response)\n}\n\n// ExtractTextFromResponse extracts text from agent response\nfunc ExtractTextFromResponse(response *agentContext.Response) (string, error) {\n\tif response == nil {\n\t\treturn \"\", fmt.Errorf(\"agent returned nil response\")\n\t}\n\n\t// Priority 1: Check Next field (custom hook data)\n\tif response.Next != nil {\n\t\tif nextStr, ok := response.Next.(string); ok {\n\t\t\treturn nextStr, nil\n\t\t}\n\t\t// Otherwise, JSON stringify to preserve complete structure\n\t\tjsonBytes, err := jsoniter.Marshal(response.Next)\n\t\tif err != nil {\n\t\t\treturn \"\", fmt.Errorf(\"failed to serialize next hook data: %w\", err)\n\t\t}\n\t\treturn string(jsonBytes), nil\n\t}\n\n\t// Priority 2: Check Completion field (standard LLM response)\n\tif response.Completion != nil {\n\t\tswitch v := response.Completion.Content.(type) {\n\t\tcase string:\n\t\t\treturn v, nil\n\t\tcase []interface{}:\n\t\t\t// Multimodal content array - extract all text parts\n\t\t\tvar text string\n\t\t\tfor _, part := range v {\n\t\t\t\tif partMap, ok := part.(map[string]interface{}); ok {\n\t\t\t\t\tif partType, _ := partMap[\"type\"].(string); partType == \"text\" {\n\t\t\t\t\t\tif textContent, ok := partMap[\"text\"].(string); ok {\n\t\t\t\t\t\t\ttext += textContent\n\t\t\t\t\t\t}\n\t\t\t\t\t}\n\t\t\t\t}\n\t\t\t}\n\t\t\tif text != \"\" {\n\t\t\t\treturn text, nil\n\t\t\t}\n\t\t\treturn \"\", fmt.Errorf(\"no text content found in completion content parts\")\n\t\t}\n\t}\n\n\treturn \"\", fmt.Errorf(\"no content found in agent response\")\n}\n\n// CallMCPTool calls an MCP tool to process content\nfunc CallMCPTool(ctx *agentContext.Context, serverID string, toolName string, arguments map[string]interface{}) (string, error) {\n\t// Get MCP context for cancellation/timeout control\n\tmcpCtx := ctx.Context\n\tif mcpCtx == nil {\n\t\tmcpCtx = context.Background()\n\t}\n\n\t// Get MCP client\n\tclient, err := mcp.Select(serverID)\n\tif err != nil {\n\t\treturn \"\", fmt.Errorf(\"failed to select MCP client '%s': %w\", serverID, err)\n\t}\n\n\t// Call the tool\n\tlog.Trace(\"[Content] Calling MCP tool: %s (server: %s)\", toolName, serverID)\n\tcallResult, err := client.CallTool(mcpCtx, toolName, arguments)\n\tif err != nil {\n\t\treturn \"\", fmt.Errorf(\"MCP tool call failed: %w\", err)\n\t}\n\n\t// Check if result is an error\n\tif callResult.IsError {\n\t\treturn \"\", fmt.Errorf(\"MCP tool returned error: %v\", callResult.Content)\n\t}\n\n\t// Extract text content from result\n\tvar text string\n\tfor _, content := range callResult.Content {\n\t\tif content.Type == \"text\" {\n\t\t\ttext += content.Text\n\t\t}\n\t}\n\n\tif text == \"\" {\n\t\treturn \"\", fmt.Errorf(\"MCP tool returned no text content\")\n\t}\n\n\treturn text, nil\n}\n" }, { "path": "agent/content/types/types.go", "content": "package types\n\nimport (\n\t\"github.com/yaoapp/gou/connector\"\n\t\"github.com/yaoapp/gou/connector/openai\"\n\tagentContext \"github.com/yaoapp/yao/agent/context\"\n)\n\n// Options represents the options for the content\ntype Options struct {\n\n\t// Connector, Current connector instance\n\tConnector connector.Connector\n\n\t// Capabilities, Current capabilities instance\n\tCapabilities *openai.Capabilities\n\n\t// CompletionOptions, Current completion options instance\n\tCompletionOptions *agentContext.CompletionOptions\n\n\t// StreamOptions, Current stream options instance\n\tStreamOptions *agentContext.StreamOptions\n\n\t// SilentLoading, if true, suppress loading messages (used when called from parent handler)\n\tSilentLoading bool\n}\n" }, { "path": "agent/context/JSAPI.md", "content": "# Context JavaScript API Documentation\n\n## Overview\n\nThe Context JavaScript API provides a comprehensive interface for interacting with the Yao Agent system from JavaScript/TypeScript hooks (Create, Next). The Context object exposes agent state, configuration, messaging capabilities, trace operations, and MCP (Model Context Protocol) integrations.\n\n## Context Object\n\nThe Context object is automatically passed to hook functions and provides access to the agent's execution environment.\n\n### Basic Properties\n\n```typescript\ninterface Context {\n // Identifiers\n chat_id: string; // Current chat session ID\n assistant_id: string; // Assistant identifier\n\n // Configuration\n locale: string; // User locale (e.g., \"en\", \"zh-cn\")\n theme: string; // UI theme preference\n accept: string; // Output format (\"standard\", \"cui-web\", \"cui-native\", etc.)\n route: string; // Request route path\n referer: string; // Request referer\n\n // Client Information\n client: {\n type: string; // Client type\n user_agent: string; // User agent string\n ip: string; // Client IP address\n };\n\n // Dynamic Data\n metadata: Record; // Custom metadata (empty object if not set)\n authorized: Record; // Authorization data (empty object if not set)\n\n // Objects\n memory: Memory; // Agent memory with four namespaces: user, team, chat, context\n trace: Trace; // Trace object for debugging and monitoring\n mcp: MCP; // MCP object for external tool/resource access\n agent: Agent; // Agent-to-Agent calls (A2A)\n llm: LLM; // Direct LLM connector calls\n sandbox?: Sandbox; // Sandbox operations (only when sandbox configured)\n}\n```\n\n## Methods\n\n### Send Messages\n\nThe Context provides several methods for sending messages to the client:\n\n| Method | Description | Auto `message_end` | Updatable |\n| ------------------------------------ | --------------------------- | ------------------ | --------- |\n| `Send(message, block_id?)` | Send a complete message | ✅ Yes | ❌ No |\n| `SendStream(message, block_id?)` | Start a streaming message | ❌ No | ✅ Yes |\n| `Append(message_id, content, path?)` | Append content to a message | - | - |\n| `Replace(message_id, message)` | Replace message content | - | - |\n| `Merge(message_id, data, path?)` | Merge data into message | - | - |\n| `Set(message_id, data, path)` | Set a field in message | - | - |\n| `End(message_id, final_content?)` | Finalize streaming message | ✅ Yes | - |\n| `EndBlock(block_id)` | End a message block | - | - |\n| `MessageID()` | Generate unique message ID | - | - |\n| `BlockID()` | Generate unique block ID | - | - |\n| `ThreadID()` | Generate unique thread ID | - | - |\n\n> **Note:** `Append`, `Replace`, `Merge`, and `Set` only work with messages started via `SendStream()`. Messages sent via `Send()` are immediately finalized and cannot be updated.\n\n#### `ctx.Send(message, block_id?): string`\n\nSends a message to the client and automatically flushes the output.\n\n**Parameters:**\n\n- `message`: Message object or string\n- `block_id`: String (optional) - Block ID to send this message in. If omitted, no block ID is assigned.\n\n**Returns:**\n\n- `string`: The message ID (auto-generated if not provided in the message object)\n\n**Message Object Structure:**\n\n```typescript\ninterface Message {\n // Required\n type: string; // Message type: \"text\", \"tool\", \"image\", etc.\n\n // Common fields\n props?: Record; // Message properties (passed to frontend component)\n message_id?: string; // Message ID (auto-generated if omitted)\n block_id?: string; // Block ID (NOT auto-generated, has priority over block_id parameter)\n thread_id?: string; // Thread ID (auto-set from Stack for nested agents)\n\n // Metadata (optional)\n metadata?: Record; // Custom metadata\n}\n```\n\n**Examples:**\n\n```javascript\n// Send text message (object format) and capture message ID\nconst message_id = ctx.Send({\n type: \"text\",\n props: { content: \"Hello, World!\" },\n});\nconsole.log(\"Sent message:\", message_id);\n\n// Send text message (shorthand) - no block ID by default\nconst text_id = ctx.Send(\"Hello, World!\");\n\n// Send multiple messages in the same block (same bubble/card in UI)\nconst block_id = ctx.BlockID(); // Generate block ID first\nconst msg1 = ctx.Send(\"Step 1: Analyzing...\", block_id);\nconst msg2 = ctx.Send(\"Step 2: Processing...\", block_id);\nconst msg3 = ctx.Send(\"Step 3: Complete!\", block_id);\n\n// Specify block_id in message object (highest priority)\nconst msg4 = ctx.Send({\n type: \"text\",\n props: { content: \"In specific block\" },\n block_id: \"B2\", // This takes priority over second parameter\n});\n\n// Send tool message with custom IDs\nconst tool_id = ctx.Send({\n type: \"tool\",\n message_id: \"custom-tool-msg-1\",\n block_id: \"B_tools\",\n props: {\n name: \"calculator\",\n result: { sum: 42 },\n },\n});\n\n// Send image message\nconst image_id = ctx.Send({\n type: \"image\",\n props: {\n url: \"https://example.com/image.png\",\n alt: \"Example Image\",\n },\n});\n```\n\n**Block Management:**\n\n```javascript\n// Scenario 1: Simple message (most common)\nfunction Next(ctx, payload) {\n const { completion } = payload;\n\n // Send a complete message\n ctx.Send({\n type: \"text\",\n props: { content: completion.content },\n });\n}\n\n// Scenario 2: Loading indicator before slow operation\nfunction Next(ctx, payload) {\n // Start a streaming message for loading\n const loading_id = ctx.SendStream({\n type: \"loading\",\n props: { message: \"Fetching data...\" },\n });\n\n // Do slow operation (e.g., external API call)\n const result = fetchExternalData();\n\n // Replace loading with result\n ctx.Replace(loading_id, {\n type: \"text\",\n props: { content: result },\n });\n ctx.End(loading_id);\n}\n\n// Scenario 3: Grouping messages in one block (special case)\nfunction Create(ctx, messages) {\n // Generate a block ID for grouping\n const block_id = ctx.BlockID(); // \"B1\"\n\n ctx.Send(\"# Analysis Results\", block_id);\n ctx.Send(\"- Finding 1: ...\", block_id);\n ctx.Send(\"- Finding 2: ...\", block_id);\n ctx.Send(\"- Finding 3: ...\", block_id);\n\n // All messages appear in the same card/bubble in the UI\n}\n\n// Scenario 4: LLM response + follow-up card in same block\nfunction Next(ctx, payload) {\n const { completion } = payload;\n const block_id = ctx.BlockID();\n\n // LLM response\n ctx.Send({\n type: \"text\",\n props: { content: completion.content },\n block_id: block_id,\n });\n\n // Action card (grouped with LLM response)\n ctx.Send({\n type: \"card\",\n props: {\n title: \"Related Actions\",\n actions: [\"action1\", \"action2\"],\n },\n block_id: block_id,\n });\n}\n```\n\n**Notes:**\n\n- **Message ID** is automatically generated if not provided\n- **Block ID** is NOT auto-generated by default (remains empty unless manually specified)\n - Most messages don't need a Block ID (each message is independent)\n - Only specify Block ID in special cases (e.g., grouping LLM output with a follow-up card)\n - **Block ID priority**: message.block_id > block_id parameter > empty\n- **Thread ID** is automatically set from Stack for non-root calls (nested agents)\n- Returns the message ID for reference in subsequent operations\n- Output is automatically flushed after sending\n- Throws exception on failure\n- `Send()` automatically sends `message_end` event - the message is complete and cannot be updated\n- **For updatable messages**, use `ctx.SendStream()` instead (see below)\n\n#### `ctx.SendStream(message, block_id?): string`\n\nSends a streaming message that can be appended to later. Unlike `Send()`, this does NOT automatically send `message_end` event. Use `ctx.Append()` to add content, then `ctx.End()` to finalize.\n\n**Parameters:**\n\n- `message`: Message object or string\n- `block_id`: String (optional) - Block ID to send this message in\n\n**Returns:**\n\n- `string`: The message ID (for use with `Append` and `End`)\n\n**Examples:**\n\n```javascript\n// Start a streaming message\nconst msg_id = ctx.SendStream({\n type: \"text\",\n props: { content: \"# Title\\n\\n\" },\n});\n\n// Append content in chunks (simulating streaming)\nctx.Append(msg_id, \"First paragraph. \");\nctx.Append(msg_id, \"Second sentence. \");\nctx.Append(msg_id, \"Third sentence.\\n\\n\");\n\n// Finalize the message (sends message_end event)\nctx.End(msg_id);\n```\n\n**String Shorthand:**\n\n```javascript\n// SendStream with string shorthand\nconst msg_id = ctx.SendStream(\"Starting analysis...\");\nctx.Append(msg_id, \" processing...\");\nctx.Append(msg_id, \" done!\");\nctx.End(msg_id);\n// Final content: \"Starting analysis... processing... done!\"\n```\n\n**With Block ID:**\n\n```javascript\nconst block_id = ctx.BlockID();\nconst msg_id = ctx.SendStream(\"Step 1: \", block_id);\nctx.Append(msg_id, \"Analyzing data...\");\nctx.End(msg_id);\n```\n\n**Notes:**\n\n- Returns the message ID immediately for use with `Append` and `End`\n- Sends `message_start` event but NOT `message_end` (unlike `Send`)\n- Must call `ctx.End(msg_id)` to finalize the message\n- Content appended via `ctx.Append()` is accumulated for storage\n- Ideal for streaming text output where you control the timing\n\n#### `ctx.End(message_id, final_content?): string`\n\nFinalizes a streaming message started with `SendStream()`. Sends `message_end` event with the complete accumulated content.\n\n**Parameters:**\n\n- `message_id`: String - The message ID returned by `SendStream()`\n- `final_content`: String (optional) - Final content to append before ending\n\n**Returns:**\n\n- `string`: The message ID\n\n**Examples:**\n\n```javascript\n// Basic usage\nconst msg_id = ctx.SendStream(\"Hello\");\nctx.Append(msg_id, \" World\");\nctx.End(msg_id);\n// Final: \"Hello World\"\n\n// End with final content\nconst msg_id2 = ctx.SendStream(\"Processing\");\nctx.Append(msg_id2, \"...\");\nctx.End(msg_id2, \" Complete!\");\n// Final: \"Processing... Complete!\"\n```\n\n**Notes:**\n\n- Must be called after `SendStream()` to send `message_end` event\n- Optional `final_content` is appended before sending `message_end`\n- The complete accumulated content is included in `message_end.extra.content`\n- Throws exception if `message_id` is not a string\n\n**Send vs SendStream Comparison:**\n\n| Feature | `Send()` | `SendStream()` |\n| --------------------- | ----------------- | ------------------- |\n| `message_start` event | ✅ Auto | ✅ Auto |\n| `message_end` event | ✅ Auto | ❌ Manual (`End()`) |\n| Use case | Complete messages | Streaming output |\n| Content accumulation | N/A | Via `Append()` |\n| Storage | Immediate | On `End()` |\n\n**Streaming Workflow Example:**\n\n```javascript\nfunction Create(ctx, messages) {\n // Start streaming output\n const msg_id = ctx.SendStream({\n type: \"text\",\n props: { content: \"# Analysis Report\\n\\n\" },\n });\n\n // Simulate streaming chunks\n ctx.Append(msg_id, \"## Section 1\\n\");\n ctx.Append(msg_id, \"Processing data...\\n\\n\");\n\n // Do some work\n const result = analyzeData();\n\n ctx.Append(msg_id, \"## Section 2\\n\");\n ctx.Append(msg_id, `Found ${result.count} items.\\n\\n`);\n\n // Finalize with conclusion\n ctx.End(msg_id, \"## Conclusion\\nAnalysis complete.\");\n\n return { messages };\n}\n```\n\n#### `ctx.Replace(message_id, message): string`\n\nReplaces the content of a streaming message. **Only works with messages started via `SendStream()`**.\n\n**Parameters:**\n\n- `message_id`: String - The ID of the streaming message (returned by `SendStream()`)\n- `message`: Message object or string - The new message content\n\n**Returns:**\n\n- `string`: The message ID (same as the provided message_id)\n\n**Examples:**\n\n```javascript\n// Start a streaming message\nconst msg_id = ctx.SendStream({\n type: \"loading\",\n props: { message: \"Loading...\" },\n});\n\n// Replace with new content\nctx.Replace(msg_id, {\n type: \"text\",\n props: { content: \"Data loaded successfully!\" },\n});\n\n// Finalize the message\nctx.End(msg_id);\n```\n\n**Use Cases:**\n\n```javascript\n// Progress updates with replacement\nfunction Next(ctx, payload) {\n const msg_id = ctx.SendStream(\"Step 1/3: Starting...\");\n\n // ... do work ...\n ctx.Replace(msg_id, \"Step 2/3: Processing...\");\n\n // ... do more work ...\n ctx.Replace(msg_id, \"Step 3/3: Finalizing...\");\n\n // ... finish ...\n ctx.Replace(msg_id, \"Complete! ✓\");\n ctx.End(msg_id);\n}\n\n// Loading to result transition\nfunction Next(ctx, payload) {\n const msg_id = ctx.SendStream({\n type: \"loading\",\n props: { message: \"Fetching results...\" },\n });\n\n const results = fetchData();\n\n ctx.Replace(msg_id, {\n type: \"text\",\n props: { content: `Found ${results.length} results` },\n });\n ctx.End(msg_id);\n}\n```\n\n**Notes:**\n\n- **Only works with `SendStream()` messages** - `Send()` messages cannot be replaced\n- Replaces the entire message content, not just specific fields\n- Must call `ctx.End(msg_id)` after all updates to finalize the message\n- Output is automatically flushed after replacing\n- Throws exception on failure\n\n#### `ctx.Append(message_id, content, path?): string`\n\nAppends content to a streaming message. **Only works with messages started via `SendStream()`**.\n\n**Parameters:**\n\n- `message_id`: String - The ID of the streaming message (returned by `SendStream()`)\n- `content`: Message object or string - The content to append\n- `path`: String (optional) - The delta path to append to (e.g., \"props.content\", \"props.data\")\n\n**Returns:**\n\n- `string`: The message ID (same as the provided message_id)\n\n**Examples:**\n\n```javascript\n// Start a streaming message\nconst msg_id = ctx.SendStream(\"Starting\");\n\n// Append more text (default path)\nctx.Append(msg_id, \"... processing\");\nctx.Append(msg_id, \"... done!\");\n\n// Finalize the message\nctx.End(msg_id);\n// Final content: \"Starting... processing... done!\"\n\n// Append to specific path\nconst data_id = ctx.SendStream({\n type: \"data\",\n props: {\n content: \"Item 1\\n\",\n status: \"loading\",\n },\n});\n\nctx.Append(data_id, \"Item 2\\n\", \"props.content\");\nctx.Append(data_id, \"Item 3\\n\", \"props.content\");\nctx.End(data_id);\n// Final: props.content = \"Item 1\\nItem 2\\nItem 3\\n\"\n```\n\n**Use Cases:**\n\n```javascript\n// Streaming text output (simulating LLM-like output)\nfunction Create(ctx, messages) {\n const msg_id = ctx.SendStream(\"\");\n\n ctx.Append(msg_id, \"The\");\n ctx.Append(msg_id, \" quick\");\n ctx.Append(msg_id, \" brown\");\n ctx.Append(msg_id, \" fox\");\n\n ctx.End(msg_id);\n // Final: \"The quick brown fox\"\n\n return { messages };\n}\n\n// Progress logs\nfunction Next(ctx, payload) {\n const log_id = ctx.SendStream({\n type: \"log\",\n props: { content: \"Starting process\\n\" },\n });\n\n // Step 1\n doStep1();\n ctx.Append(log_id, \"Step 1 complete\\n\", \"props.content\");\n\n // Step 2\n doStep2();\n ctx.Append(log_id, \"Step 2 complete\\n\", \"props.content\");\n\n // Finish\n ctx.Append(log_id, \"All done!\\n\", \"props.content\");\n ctx.End(log_id);\n}\n```\n\n**Notes:**\n\n- **Only works with `SendStream()` messages** - `Send()` messages cannot be appended to\n- Uses delta append operation (adds to existing content, doesn't replace)\n- If `path` is omitted, appends to the default content location (`props.content`)\n- Must call `ctx.End(msg_id)` after all appends to finalize the message\n- Output is automatically flushed after appending\n- Throws exception on failure\n- block_id and ThreadID are inherited from the original message\n\n#### `ctx.Merge(message_id, data, path?): string`\n\nMerges data into a streaming message object. **Only works with messages started via `SendStream()`**.\n\n**Parameters:**\n\n- `message_id`: String - The ID of the streaming message (returned by `SendStream()`)\n- `data`: Object - The data to merge (should be an object)\n- `path`: String (optional) - The delta path to merge into (e.g., \"props\", \"props.metadata\")\n\n**Returns:**\n\n- `string`: The message ID (same as the provided message_id)\n\n**Examples:**\n\n```javascript\n// Start a streaming message with object data\nconst msg_id = ctx.SendStream({\n type: \"status\",\n props: {\n status: \"running\",\n progress: 0,\n started: true,\n },\n});\n\n// Merge updates into props (adds/updates fields, keeps others unchanged)\nctx.Merge(msg_id, { progress: 50 }, \"props\");\n// Result: props = { status: \"running\", progress: 50, started: true }\n\nctx.Merge(msg_id, { progress: 100, status: \"completed\" }, \"props\");\n// Result: props = { status: \"completed\", progress: 100, started: true }\n\n// Finalize the message\nctx.End(msg_id);\n```\n\n**Use Cases:**\n\n```javascript\n// Updating task progress\nfunction Next(ctx, payload) {\n const task_id = ctx.SendStream({\n type: \"task\",\n props: {\n name: \"Data Processing\",\n status: \"pending\",\n progress: 0,\n },\n });\n\n ctx.Merge(task_id, { status: \"running\" }, \"props\");\n doStep1();\n ctx.Merge(task_id, { progress: 25 }, \"props\");\n doStep2();\n ctx.Merge(task_id, { progress: 50 }, \"props\");\n doStep3();\n ctx.Merge(task_id, { progress: 100, status: \"completed\" }, \"props\");\n\n ctx.End(task_id);\n}\n\n// Building metadata incrementally\nfunction Create(ctx, messages) {\n const data_id = ctx.SendStream({\n type: \"data\",\n props: { content: \"Result data\" },\n });\n\n ctx.Merge(data_id, { metadata: { source: \"api\" } }, \"props\");\n ctx.Merge(data_id, { metadata: { timestamp: Date.now() } }, \"props\");\n // metadata fields are merged together\n\n ctx.End(data_id);\n return { messages };\n}\n```\n\n**Notes:**\n\n- **Only works with `SendStream()` messages** - `Send()` messages cannot be merged into\n- Uses delta merge operation (merges objects, doesn't replace)\n- Only works with object data (for merging key-value pairs)\n- Existing fields not in the merge data remain unchanged\n- If `path` is omitted, merges into the default object location\n- Must call `ctx.End(msg_id)` after all merges to finalize the message\n- Output is automatically flushed after merging\n- Throws exception on failure\n- block_id and ThreadID are inherited from the original message\n\n#### `ctx.Set(message_id, data, path): string`\n\nSets a new field or value in a streaming message. **Only works with messages started via `SendStream()`**.\n\n**Parameters:**\n\n- `message_id`: String - The ID of the streaming message (returned by `SendStream()`)\n- `data`: Any - The value to set\n- `path`: String (required) - The delta path where to set the value (e.g., \"props.newField\", \"props.metadata.key\")\n\n**Returns:**\n\n- `string`: The message ID (same as the provided message_id)\n\n**Examples:**\n\n```javascript\n// Start a streaming message\nconst msg_id = ctx.SendStream({\n type: \"result\",\n props: {\n content: \"Initial content\",\n },\n});\n\n// Set a new field\nctx.Set(msg_id, \"success\", \"props.status\");\n// Result: props.status = \"success\"\n\n// Set a nested object\nctx.Set(msg_id, { duration: 1500, cached: true }, \"props.metadata\");\n// Result: props.metadata = { duration: 1500, cached: true }\n\n// Finalize the message\nctx.End(msg_id);\n```\n\n**Use Cases:**\n\n```javascript\n// Adding computed metadata after initial send\nfunction Next(ctx, payload) {\n const result_id = ctx.SendStream({\n type: \"search_result\",\n props: { results: search_results },\n });\n\n ctx.Set(result_id, search_results.length, \"props.count\");\n ctx.Set(result_id, Date.now(), \"props.timestamp\");\n ctx.Set(result_id, \"relevance\", \"props.sort_by\");\n\n ctx.End(result_id);\n}\n\n// Conditionally adding fields\nfunction Create(ctx, messages) {\n const msg_id = ctx.SendStream({\n type: \"operation\",\n props: { name: \"Process Data\" },\n });\n\n try {\n const result = processData();\n ctx.Set(msg_id, \"success\", \"props.status\");\n ctx.Set(msg_id, result, \"props.data\");\n } catch (e) {\n ctx.Set(msg_id, e.message, \"props.error\");\n ctx.Set(msg_id, \"error\", \"props.status\");\n }\n\n ctx.End(msg_id);\n return { messages };\n}\n```\n\n**Notes:**\n\n- **Only works with `SendStream()` messages** - `Send()` messages cannot be modified\n- Uses delta set operation (creates/sets new fields)\n- The `path` parameter is **required** (must specify where to set the value)\n- Creates the path if it doesn't exist\n- Use for adding new fields or completely replacing a field's value\n- For updating existing object fields, consider using `Merge` instead\n- Must call `ctx.End(msg_id)` after all sets to finalize the message\n- Output is automatically flushed after setting\n- Throws exception on failure\n- block_id and ThreadID are inherited from the original message\n\n### ID Generators\n\nThese methods generate unique IDs for manual message management. Useful when you need to specify IDs before sending messages or for advanced Block/Thread management.\n\n#### `ctx.MessageID(): string`\n\nGenerates a unique message ID.\n\n**Returns:**\n\n- `string`: Message ID in format \"M1\", \"M2\", \"M3\"...\n\n**Example:**\n\n```javascript\n// Generate IDs manually\nconst id_1 = ctx.MessageID(); // \"M1\"\nconst id_2 = ctx.MessageID(); // \"M2\"\n\n// Use custom ID\nctx.Send({\n type: \"text\",\n message_id: id_1,\n props: { content: \"Hello\" },\n});\n```\n\n#### `ctx.BlockID(): string`\n\nGenerates a unique block ID for grouping messages.\n\n**Returns:**\n\n- `string`: Block ID in format \"B1\", \"B2\", \"B3\"...\n\n**Example:**\n\n```javascript\n// Generate block ID for grouping messages\nconst block_id = ctx.BlockID(); // \"B1\"\n\n// Send multiple messages in the same block\nctx.Send(\"Step 1: Analyzing...\", block_id);\nctx.Send(\"Step 2: Processing...\", block_id);\nctx.Send(\"Step 3: Complete!\", block_id);\n\n// All three messages appear in the same card/bubble in UI\n```\n\n**Use Cases:**\n\n```javascript\n// Scenario: LLM output + follow-up card in same block\nconst block_id = ctx.BlockID();\n\n// LLM response\nconst llm_result = Process(\"llms.chat\", {...});\nctx.Send({\n type: \"text\",\n props: { content: llm_result.content },\n block_id: block_id,\n});\n\n// Follow-up action card (grouped with LLM output)\nctx.Send({\n type: \"card\",\n props: {\n title: \"Related Actions\",\n actions: [...]\n },\n block_id: block_id,\n});\n```\n\n#### `ctx.ThreadID(): string`\n\nGenerates a unique thread ID for concurrent operations.\n\n**Returns:**\n\n- `string`: Thread ID in format \"T1\", \"T2\", \"T3\"...\n\n**Example:**\n\n```javascript\n// For advanced parallel processing scenarios\nconst thread_id = ctx.ThreadID(); // \"T1\"\n\n// Send messages in a specific thread\nctx.Send({\n type: \"text\",\n props: { content: \"Parallel task 1\" },\n thread_id: thread_id,\n});\n```\n\n**Notes:**\n\n- IDs are generated sequentially within each context\n- Each context has its own ID counter (starts from 1)\n- IDs are guaranteed to be unique within the same request/stream\n- ThreadID is usually auto-managed by Stack, manual generation is for advanced use cases\n\n### Lifecycle Management\n\n#### `ctx.EndBlock(block_id): void`\n\nManually sends a `block_end` event for the specified block. Use this to explicitly mark the end of a block.\n\n**Parameters:**\n\n- `block_id`: String - The block ID to end\n\n**Returns:**\n\n- `void`\n\n**Example:**\n\n```javascript\n// Create a block for grouped messages\nconst block_id = ctx.BlockID(); // \"B1\"\n\n// Send messages in the block\nctx.Send(\"Analyzing data...\", block_id);\nctx.Send(\"Processing results...\", block_id);\nctx.Send(\"Complete!\", block_id);\n\n// Manually end the block\nctx.EndBlock(block_id);\n```\n\n**Block Lifecycle Events:**\n\nWhen you send messages with a `block_id`:\n\n1. **First message**: Automatically sends `block_start` event\n2. **Subsequent messages**: No additional block events\n3. **Manual end**: Call `ctx.EndBlock(block_id)` to send `block_end` event\n\n**block_end Event Format:**\n\n```json\n{\n \"type\": \"event\",\n \"props\": {\n \"event\": \"block_end\",\n \"message\": \"Block ended\",\n \"data\": {\n \"block_id\": \"B1\",\n \"timestamp\": 1764483531624,\n \"duration_ms\": 1523,\n \"message_count\": 3,\n \"status\": \"completed\"\n }\n }\n}\n```\n\n**Notes:**\n\n- `block_start` is sent automatically when the first message with a new `block_id` is sent\n- `block_end` must be called manually via `ctx.EndBlock()`\n- You can track multiple blocks simultaneously (each has independent lifecycle)\n- Automatically flushes output after sending the event\n\n**Use Cases:**\n\n```javascript\n// Use case 1: Progress reporting in a block\nfunction Create(ctx, messages) {\n const block_id = ctx.BlockID();\n\n ctx.Send(\"Step 1: Analyzing data...\", block_id);\n // ... analysis logic ...\n\n ctx.Send(\"Step 2: Processing results...\", block_id);\n // ... processing logic ...\n\n ctx.Send(\"Step 3: Complete!\", block_id);\n\n // Mark the block as complete\n ctx.EndBlock(block_id);\n\n return { messages };\n}\n\n// Use case 2: Multiple parallel blocks\nfunction Create(ctx, messages) {\n const llm_block = ctx.BlockID(); // \"B1\"\n const mcp_block = ctx.BlockID(); // \"B2\"\n\n // LLM output block\n ctx.Send(\"Thinking...\", llm_block);\n const response = callLLM();\n ctx.Send(response, llm_block);\n ctx.EndBlock(llm_block);\n\n // MCP tool call block\n ctx.Send(\"Fetching data...\", mcp_block);\n const data = ctx.mcp.CallTool(\"tool\", \"method\", {});\n ctx.Send(`Found ${data.length} results`, mcp_block);\n ctx.EndBlock(mcp_block);\n\n return { messages };\n}\n```\n\n### Resource Cleanup\n\n#### `ctx.Release()`\n\nManually releases Context resources.\n\n> **Note:** In Hook functions (`Create`, `Next`), you do **NOT** need to call `Release()` - the system handles cleanup automatically. Only call `Release()` when you create a new Context manually (e.g., via `new Context()`).\n\n**Example (only for manually created Context):**\n\n```javascript\n// Only needed when creating Context manually, NOT in hooks\nconst ctx = new Context(options);\ntry {\n ctx.Send(\"Processing...\");\n} finally {\n ctx.Release(); // Required for manually created Context\n}\n```\n\n## Trace API\n\nThe `ctx.trace` object provides tracing capabilities for:\n\n1. **User Transparency** - Expose the agent's working and thinking process to users. The frontend will render these trace nodes to show users what the agent is doing.\n2. **Developer Debugging** - Help developers debug agent execution by recording detailed steps and data.\n\n> **Note:** Trace is primarily designed for developers to expose the agent's process to users. The frontend has corresponding UI components to render these trace nodes.\n\n### Properties\n\n- `ctx.trace.id`: String - The unique identifier of the trace\n\n### Methods Summary\n\n| Method | Description |\n| ------------------------- | ------------------------------- |\n| `Add(input, option)` | Create a sequential trace node |\n| `Parallel(inputs)` | Create parallel trace nodes |\n| `Info(message)` | Add info log to current node |\n| `Debug(message)` | Add debug log to current node |\n| `Warn(message)` | Add warning log to current node |\n| `Error(message)` | Add error log to current node |\n| `SetOutput(output)` | Set output for current node |\n| `SetMetadata(key, value)` | Set metadata for current node |\n| `Complete(output?)` | Mark current node as completed |\n| `Fail(error)` | Mark current node as failed |\n| `MarkComplete()` | Mark entire trace as complete |\n| `IsComplete()` | Check if trace is complete |\n| `CreateSpace(option)` | Create a visual space container |\n| `GetSpace(id)` | Get a trace space by ID |\n| `Release()` | Release trace resources |\n\n### Node Operations\n\n#### `ctx.trace.Add(input, options)`\n\nCreates a new trace node (sequential step).\n\n**Parameters:**\n\n- `input`: Input data for the node\n- `options`: Node configuration object\n\n**Options Structure:**\n\n```typescript\ninterface TraceNodeOption {\n label: string; // Display label in UI\n type?: string; // Node type identifier\n icon?: string; // Icon identifier\n description?: string; // Node description\n metadata?: Record; // Additional metadata\n autoCompleteParent?: boolean; // Auto-complete parent node(s) when this node is created (default: true)\n}\n```\n\n**Example:**\n\n```javascript\nconst search_node = ctx.trace.Add(\n { query: \"What is AI?\" },\n {\n label: \"Search Query\",\n type: \"search\",\n icon: \"search\",\n description: \"Searching for AI information\",\n }\n);\n```\n\n#### `ctx.trace.Parallel(inputs)`\n\nCreates multiple parallel trace nodes for concurrent operations.\n\n**Parameters:**\n\n- `inputs`: Array of parallel input objects\n\n**Input Structure:**\n\n```typescript\ninterface ParallelInput {\n input: any; // Input data\n option: TraceNodeOption; // Node configuration\n}\n```\n\n**Example:**\n\n```javascript\nconst parallel_nodes = ctx.trace.Parallel([\n {\n input: { url: \"https://api1.com\" },\n option: {\n label: \"API Call 1\",\n type: \"api\",\n icon: \"cloud\",\n description: \"Fetching from API 1\",\n },\n },\n {\n input: { url: \"https://api2.com\" },\n option: {\n label: \"API Call 2\",\n type: \"api\",\n icon: \"cloud\",\n description: \"Fetching from API 2\",\n },\n },\n]);\n```\n\n### Logging Methods\n\nAdd log entries to the current trace node. Each method takes a single string message and returns the trace object for chaining.\n\n```javascript\n// Information logs\nctx.trace.Info(\"Processing started\");\n\n// Debug logs\nctx.trace.Debug(\"Variable value: 42\");\n\n// Warning logs\nctx.trace.Warn(\"Deprecated feature used\");\n\n// Error logs\nctx.trace.Error(\"Operation failed: timeout\");\n```\n\n### Trace-Level Operations\n\nThese methods operate on the current trace node (managed by the trace manager).\n\n#### `ctx.trace.SetOutput(output)`\n\nSets the output data for the current trace node.\n\n```javascript\nctx.trace.SetOutput({ result: \"success\", data: [...] });\n```\n\n#### `ctx.trace.SetMetadata(key, value)`\n\nSets metadata for the current trace node.\n\n```javascript\nctx.trace.SetMetadata(\"duration\", 1500);\nctx.trace.SetMetadata(\"source\", \"cache\");\n```\n\n#### `ctx.trace.Complete(output?)`\n\nMarks the current trace node as completed (optionally with output).\n\n```javascript\nctx.trace.Complete({ status: \"done\" });\n```\n\n#### `ctx.trace.Fail(error)`\n\nMarks the current trace node as failed with an error message.\n\n```javascript\nctx.trace.Fail(\"Connection timeout\");\n```\n\n### Node Object\n\nThe `ctx.trace.Add()` and `ctx.trace.Parallel()` methods return Node objects. Each node has the following properties and methods:\n\n#### Properties\n\n- `id`: String - The unique identifier of the node\n\n#### `node.Add(input, option)`\n\nCreates a child node under this node.\n\n```javascript\nconst parent_node = ctx.trace.Add({ step: \"process\" }, { label: \"Process\" });\nconst child_node = parent_node.Add(\n { action: \"validate\" },\n { label: \"Validate Input\", type: \"validation\" }\n);\n```\n\n#### `node.Parallel(inputs)`\n\nCreates multiple parallel child nodes under this node.\n\n```javascript\nconst parent_node = ctx.trace.Add({ step: \"fetch\" }, { label: \"Fetch Data\" });\nconst child_nodes = parent_node.Parallel([\n { input: { source: \"db\" }, option: { label: \"Database Query\" } },\n { input: { source: \"api\" }, option: { label: \"API Call\" } },\n]);\n```\n\n#### `node.Info(message)`, `node.Debug(message)`, `node.Warn(message)`, `node.Error(message)`\n\nAdd log entries to the node. All methods return the node for chaining.\n\n```javascript\nconst search_node = ctx.trace.Add({ query: \"search\" }, { label: \"Search\" });\nsearch_node\n .Info(\"Starting search\")\n .Debug(\"Query parameters validated\")\n .Warn(\"Cache miss, fetching from source\");\n```\n\n#### `node.SetOutput(output)`\n\nSets the output data for a node. Returns the node for chaining.\n\n```javascript\nconst search_node = ctx.trace.Add({ query: \"search\" }, { label: \"Search\" });\nsearch_node.SetOutput({ results: [...], count: 10 });\n```\n\n#### `node.SetMetadata(key, value)`\n\nSets metadata for a node. Returns the node for chaining.\n\n```javascript\nsearch_node.SetMetadata(\"duration\", 1500).SetMetadata(\"cache_hit\", true);\n```\n\n#### `node.Complete(output?)`\n\nMarks a node as completed (optionally with output). Returns the node for chaining.\n\n```javascript\nsearch_node.Complete({ status: \"success\", data: [...] });\n```\n\n#### `node.Fail(error)`\n\nMarks a node as failed with an error message. Returns the node for chaining.\n\n```javascript\ntry {\n // Operation\n} catch (error) {\n search_node.Fail(error.message);\n}\n```\n\n### Trace Lifecycle\n\n#### `ctx.trace.IsComplete()`\n\nChecks if the trace is complete.\n\n```javascript\nif (ctx.trace.IsComplete()) {\n console.log(\"Trace completed\");\n}\n```\n\n#### `ctx.trace.MarkComplete()`\n\nMarks the entire trace as complete.\n\n```javascript\nctx.trace.MarkComplete();\n```\n\n#### `ctx.trace.Release()`\n\nReleases trace resources.\n\n> **Note:** In Hook functions, you do **NOT** need to call `Release()` - the system handles cleanup automatically. Only call this when you create a Trace manually (e.g., via `new Trace()`).\n\n### Trace Space Operations\n\nTrace spaces are visual containers for organizing trace nodes in the frontend UI. They help group related operations together for better presentation to users.\n\n> **Note:** Trace spaces are purely for visual organization and presentation. They do not store data - use `ctx.memory` for data storage between hooks.\n\n#### `ctx.trace.CreateSpace(option)`\n\nCreates a visual space container for grouping trace nodes.\n\n**Option Structure:**\n\n```typescript\ninterface TraceSpaceOption {\n label: string; // Display label in UI\n type?: string; // Space type identifier\n icon?: string; // Icon identifier\n description?: string; // Space description\n ttl?: number; // Time to live in seconds (for display only)\n metadata?: Record; // Additional metadata\n}\n```\n\n**Example:**\n\n```javascript\nconst visual_space = ctx.trace.CreateSpace({\n label: \"Search Results\",\n type: \"search\",\n icon: \"search\",\n description: \"Knowledge base search operations\",\n});\n```\n\n#### `ctx.trace.GetSpace(id)`\n\nRetrieves a trace space by ID.\n\n```javascript\nconst search_space = ctx.trace.GetSpace(\"search-space-id\");\n```\n\n## Memory API\n\nThe `ctx.memory` object provides a four-level hierarchical memory system for agent state management. Each level has different persistence and scope characteristics.\n\n### Memory Namespaces\n\n| Namespace | Scope | Persistence | Use Case |\n| -------------------- | ------------------- | ----------- | ------------------------------------------- |\n| `ctx.memory.user` | Per user | Persistent | User preferences, settings, long-term state |\n| `ctx.memory.team` | Per team | Persistent | Team-wide settings, shared configurations |\n| `ctx.memory.chat` | Per chat session | Persistent | Chat-specific context, conversation state |\n| `ctx.memory.context` | Per request context | Temporary | Request-scoped data, cleared on release |\n\n### Namespace Interface\n\nEach namespace (`user`, `team`, `chat`, `context`) provides the same interface:\n\n```typescript\ninterface MemoryNamespace {\n // Basic KV operations\n Get(key: string): any; // Get a value\n Set(key: string, value: any, ttl?: number): void; // Set a value with optional TTL (seconds)\n Del(key: string): void; // Delete a key (supports wildcards: \"prefix:*\")\n Has(key: string): boolean; // Check if key exists\n GetDel(key: string): any; // Get and delete atomically\n\n // Collection operations\n Keys(): string[]; // Get all keys\n Len(): number; // Get number of keys\n Clear(): void; // Delete all keys\n\n // Atomic counter operations\n Incr(key: string, delta?: number): number; // Increment (default delta=1)\n Decr(key: string, delta?: number): number; // Decrement (default delta=1)\n\n // List operations\n Push(key: string, values: any[]): number; // Append to list, returns new length\n Pop(key: string): any; // Remove and return last element\n Pull(key: string, count: number): any[]; // Remove and return last N elements\n PullAll(key: string): any[]; // Remove and return all elements\n AddToSet(key: string, values: any[]): number; // Add unique values to set\n\n // Array access operations\n ArrayLen(key: string): number; // Get array length\n ArrayGet(key: string, index: number): any; // Get element at index\n ArraySet(key: string, index: number, value: any): void; // Set element at index\n ArraySlice(key: string, start: number, end: number): any[]; // Get slice\n ArrayPage(key: string, page: number, size: number): any[]; // Paginated access\n ArrayAll(key: string): any[]; // Get all elements\n\n // Metadata\n id: string; // Namespace ID\n space: string; // Space type: \"user\", \"team\", \"chat\", or \"context\"\n}\n```\n\n### Basic KV Operations\n\n#### `Get(key): any`\n\nGets a value from the namespace.\n\n```javascript\n// User preferences\nconst theme = ctx.memory.user.Get(\"theme\");\nif (theme) {\n console.log(\"User prefers:\", theme);\n}\n\n// Chat context\nconst topic = ctx.memory.chat.Get(\"current_topic\");\n```\n\n#### `Set(key, value, ttl?): void`\n\nSets a value with optional TTL (time-to-live in seconds).\n\n```javascript\n// Persistent user setting\nctx.memory.user.Set(\"language\", \"en\");\n\n// Team configuration\nctx.memory.team.Set(\"api_key\", \"sk-xxx\");\n\n// Chat state\nctx.memory.chat.Set(\"last_query\", \"What is AI?\");\n\n// Temporary context data with 5 minute TTL\nctx.memory.context.Set(\"temp_result\", { data: \"...\" }, 300);\n```\n\n#### `Del(key): void`\n\nDeletes a key. Supports wildcard patterns with `*`.\n\n```javascript\n// Delete single key\nctx.memory.user.Del(\"old_setting\");\n\n// Delete with wildcard pattern\nctx.memory.chat.Del(\"cache:*\"); // Deletes all keys starting with \"cache:\"\n```\n\n#### `Has(key): boolean`\n\nChecks if a key exists.\n\n```javascript\nif (ctx.memory.user.Has(\"onboarding_complete\")) {\n // Skip onboarding\n}\n```\n\n#### `GetDel(key): any`\n\nAtomically gets and deletes a value. Useful for one-time tokens.\n\n```javascript\nconst token = ctx.memory.context.GetDel(\"one_time_token\");\nif (token) {\n // Use token (it's now deleted)\n}\n```\n\n### Collection Operations\n\n#### `Keys(): string[]`\n\nReturns all keys in the namespace.\n\n```javascript\nconst userKeys = ctx.memory.user.Keys();\nconsole.log(\"User has\", userKeys.length, \"stored values\");\n```\n\n#### `Len(): number`\n\nReturns the number of keys.\n\n```javascript\nconst count = ctx.memory.chat.Len();\nconsole.log(\"Chat has\", count, \"stored values\");\n```\n\n#### `Clear(): void`\n\nDeletes all keys in the namespace.\n\n```javascript\n// Clear temporary context data\nctx.memory.context.Clear();\n```\n\n### Atomic Counter Operations\n\n#### `Incr(key, delta?): number`\n\nAtomically increments a counter. Returns the new value.\n\n```javascript\n// Simple counter\nconst views = ctx.memory.user.Incr(\"page_views\");\nconsole.log(\"Total views:\", views);\n\n// Increment by custom amount\nconst points = ctx.memory.user.Incr(\"points\", 10);\n```\n\n#### `Decr(key, delta?): number`\n\nAtomically decrements a counter. Returns the new value.\n\n```javascript\nconst remaining = ctx.memory.user.Decr(\"credits\");\nif (remaining < 0) {\n throw new Error(\"Insufficient credits\");\n}\n```\n\n### List Operations\n\n#### `Push(key, values): number`\n\nAppends values to a list. Returns new length.\n\n```javascript\nconst len = ctx.memory.chat.Push(\"history\", [\n { role: \"user\", content: \"Hello\" },\n { role: \"assistant\", content: \"Hi there!\" },\n]);\n```\n\n#### `Pop(key): any`\n\nRemoves and returns the last element.\n\n```javascript\nconst lastItem = ctx.memory.chat.Pop(\"pending_tasks\");\n```\n\n#### `Pull(key, count): any[]`\n\nRemoves and returns the last N elements.\n\n```javascript\nconst recentItems = ctx.memory.chat.Pull(\"notifications\", 5);\n```\n\n#### `PullAll(key): any[]`\n\nRemoves and returns all elements.\n\n```javascript\nconst allTasks = ctx.memory.context.PullAll(\"batch_queue\");\n// Process all tasks, queue is now empty\n```\n\n#### `AddToSet(key, values): number`\n\nAdds unique values to a set (no duplicates). Returns new size.\n\n```javascript\nctx.memory.user.AddToSet(\"visited_pages\", [\"/home\", \"/about\"]);\nctx.memory.user.AddToSet(\"visited_pages\", [\"/home\", \"/contact\"]); // \"/home\" not added again\n```\n\n### Array Access Operations\n\n#### `ArrayLen(key): number`\n\nGets the length of an array.\n\n```javascript\nconst historyLen = ctx.memory.chat.ArrayLen(\"messages\");\n```\n\n#### `ArrayGet(key, index): any`\n\nGets an element at a specific index.\n\n```javascript\nconst firstMessage = ctx.memory.chat.ArrayGet(\"messages\", 0);\nconst lastMessage = ctx.memory.chat.ArrayGet(\"messages\", -1); // Negative index\n```\n\n#### `ArraySet(key, index, value): void`\n\nSets an element at a specific index.\n\n```javascript\nctx.memory.chat.ArraySet(\"messages\", 0, { role: \"system\", content: \"Updated\" });\n```\n\n#### `ArraySlice(key, start, end): any[]`\n\nGets a slice of the array.\n\n```javascript\nconst recent = ctx.memory.chat.ArraySlice(\"messages\", -10, -1); // Last 10 messages\n```\n\n#### `ArrayPage(key, page, size): any[]`\n\nGets a page of elements (1-indexed pages).\n\n```javascript\nconst page1 = ctx.memory.chat.ArrayPage(\"messages\", 1, 20); // First 20 messages\nconst page2 = ctx.memory.chat.ArrayPage(\"messages\", 2, 20); // Next 20 messages\n```\n\n#### `ArrayAll(key): any[]`\n\nGets all elements of the array.\n\n```javascript\nconst allMessages = ctx.memory.chat.ArrayAll(\"messages\");\n```\n\n### Use Cases\n\n```javascript\n// Use case 1: User preferences (persistent across sessions)\nfunction Create(ctx, messages) {\n // Load user preferences\n const locale = ctx.memory.user.Get(\"preferred_locale\") || \"en\";\n const style = ctx.memory.user.Get(\"response_style\") || \"concise\";\n\n return {\n messages,\n locale: locale,\n metadata: { style: style },\n };\n}\n\n// Use case 2: Chat context (persistent within chat session)\nfunction Next(ctx, payload) {\n // Track conversation topics\n const topics = ctx.memory.chat.Get(\"discussed_topics\") || [];\n const newTopic = extractTopic(payload.completion.content);\n\n if (newTopic && !topics.includes(newTopic)) {\n topics.push(newTopic);\n ctx.memory.chat.Set(\"discussed_topics\", topics);\n }\n}\n\n// Use case 3: Request-scoped data (cleared on context release)\nfunction Create(ctx, messages) {\n // Store temporary processing data\n ctx.memory.context.Set(\"request_start\", Date.now());\n ctx.memory.context.Set(\"original_query\", messages[0]?.content);\n\n return { messages };\n}\n\nfunction Next(ctx, payload) {\n // Retrieve temporary data\n const startTime = ctx.memory.context.Get(\"request_start\");\n const duration = Date.now() - startTime;\n console.log(\"Request took\", duration, \"ms\");\n\n // context memory is automatically cleared when ctx.Release() is called\n}\n\n// Use case 4: Team-wide settings\nfunction Create(ctx, messages) {\n // Check team quota\n const used = ctx.memory.team.Incr(\"monthly_requests\");\n const limit = ctx.memory.team.Get(\"monthly_limit\") || 10000;\n\n if (used > limit) {\n throw new Error(\"Team quota exceeded\");\n }\n\n return { messages };\n}\n\n// Use case 5: Rate limiting with counters\nfunction Create(ctx, messages) {\n const key = `rate:${new Date().toISOString().slice(0, 13)}`; // Hourly bucket\n const count = ctx.memory.user.Incr(key);\n\n if (count > 100) {\n throw new Error(\"Rate limit exceeded\");\n }\n\n return { messages };\n}\n```\n\n### Memory Lifecycle\n\n| Namespace | Created When | Cleared When |\n| --------- | ---------------- | --------------- |\n| `user` | First access | Manual only |\n| `team` | First access | Manual only |\n| `chat` | First access | Manual only |\n| `context` | Context creation | `ctx.Release()` |\n\n**Notes:**\n\n- `user`, `team`, `chat` namespaces are persistent (backed by database)\n- `context` namespace is temporary and cleared when the request context is released\n- All namespaces support TTL for automatic expiration\n- Wildcard deletion (`Del(\"prefix:*\")`) works on all namespaces\n- Counter operations (`Incr`, `Decr`) are atomic\n\n## MCP API\n\nThe `ctx.mcp` object provides access to Model Context Protocol operations for interacting with external tools, resources, and prompts.\n\n### Methods Summary\n\n| Method | Description |\n| ------------------------------------ | -------------------------------- |\n| `ListResources(client, cursor?)` | List available resources |\n| `ReadResource(client, uri)` | Read a specific resource |\n| `ListTools(client, cursor?)` | List available tools |\n| `CallTool(client, name, args?)` | Call a single tool |\n| `CallTools(client, tools)` | Call multiple tools sequentially |\n| `CallToolsParallel(client, tools)` | Call multiple tools in parallel |\n| `All(requests)` | Call tools across servers, wait for all |\n| `Any(requests)` | Call tools across servers, first success wins |\n| `Race(requests)` | Call tools across servers, first complete wins |\n| `ListPrompts(client, cursor?)` | List available prompts |\n| `GetPrompt(client, name, args?)` | Get a specific prompt |\n| `ListSamples(client, type, name)` | List samples for a tool/resource |\n| `GetSample(client, type, name, idx)` | Get a specific sample by index |\n\n### Resource Operations\n\n#### `ctx.mcp.ListResources(client, cursor?)`\n\nLists available resources from an MCP client.\n\n**Parameters:**\n\n- `client`: String - MCP client ID\n- `cursor`: String (optional) - Pagination cursor\n\n```javascript\nconst resources = ctx.mcp.ListResources(\"echo\", \"\");\nconsole.log(resources.resources); // Array of resources\n```\n\n#### `ctx.mcp.ReadResource(client, uri)`\n\nReads a specific resource.\n\n**Parameters:**\n\n- `client`: String - MCP client ID\n- `uri`: String - Resource URI\n\n```javascript\nconst info = ctx.mcp.ReadResource(\"echo\", \"echo://info\");\nconsole.log(info.contents); // Array of content items\n```\n\n### Tool Operations\n\n#### `ctx.mcp.ListTools(client, cursor?)`\n\nLists available tools from an MCP client.\n\n**Parameters:**\n\n- `client`: String - MCP client ID\n- `cursor`: String (optional) - Pagination cursor\n\n```javascript\nconst tools = ctx.mcp.ListTools(\"echo\", \"\");\nconsole.log(tools.tools); // Array of tools\n```\n\n#### `ctx.mcp.CallTool(client, name, arguments?)`\n\nCalls a single tool and returns the parsed result directly.\n\n**Parameters:**\n\n- `client`: String - MCP client ID\n- `name`: String - Tool name\n- `arguments`: Object (optional) - Tool arguments\n\n**Returns:** Parsed result directly (automatically extracts and parses JSON from tool response)\n\n```javascript\n// Result is returned directly - no wrapper object needed\nconst result = ctx.mcp.CallTool(\"echo\", \"echo\", { message: \"hello\" });\nconsole.log(result.echo); // \"hello\" - directly access parsed data!\n\n// Another example\nconst status = ctx.mcp.CallTool(\"echo\", \"status\", { verbose: true });\nconsole.log(status.status); // \"online\"\nconsole.log(status.uptime); // 3600\n```\n\n#### `ctx.mcp.CallTools(client, tools)`\n\nCalls multiple tools sequentially and returns array of parsed results.\n\n**Parameters:**\n\n- `client`: String - MCP client ID\n- `tools`: Array - Array of tool call objects\n\n**Returns:** Array of parsed results (same order as input tools)\n\n```javascript\nconst results = ctx.mcp.CallTools(\"echo\", [\n { name: \"ping\", arguments: { count: 1 } },\n { name: \"echo\", arguments: { message: \"hello\" } },\n]);\n\n// Results are directly accessible\nconsole.log(results[0].message); // \"pong\"\nconsole.log(results[1].echo); // \"hello\"\n```\n\n#### `ctx.mcp.CallToolsParallel(client, tools)`\n\nCalls multiple tools in parallel and returns array of parsed results.\n\n**Parameters:**\n\n- `client`: String - MCP client ID\n- `tools`: Array - Array of tool call objects\n\n**Returns:** Array of parsed results (same order as input tools)\n\n```javascript\nconst results = ctx.mcp.CallToolsParallel(\"echo\", [\n { name: \"ping\", arguments: { count: 1 } },\n { name: \"echo\", arguments: { message: \"hello\" } },\n]);\n\n// Results are directly accessible (order matches input order)\nconsole.log(results[0].message); // \"pong\" (ping result)\nconsole.log(results[1].echo); // \"hello\" (echo result)\n```\n\n### Prompt Operations\n\n#### `ctx.mcp.ListPrompts(client, cursor?)`\n\nLists available prompts from an MCP client.\n\n**Parameters:**\n\n- `client`: String - MCP client ID\n- `cursor`: String (optional) - Pagination cursor\n\n```javascript\nconst prompts = ctx.mcp.ListPrompts(\"echo\", \"\");\nconsole.log(prompts.prompts); // Array of prompts\n```\n\n#### `ctx.mcp.GetPrompt(client, name, arguments?)`\n\nRetrieves a specific prompt with optional arguments.\n\n**Parameters:**\n\n- `client`: String - MCP client ID\n- `name`: String - Prompt name\n- `arguments`: Object (optional) - Prompt arguments\n\n```javascript\nconst prompt = ctx.mcp.GetPrompt(\"echo\", \"test_connection\", {\n detailed: \"true\",\n});\nconsole.log(prompt.messages); // Array of prompt messages\n```\n\n### Sample Operations\n\n#### `ctx.mcp.ListSamples(client, type, name)`\n\nLists available samples for a tool or resource.\n\n**Parameters:**\n\n- `client`: String - MCP client ID\n- `type`: String - Sample type (\"tool\" or \"resource\")\n- `name`: String - Tool or resource name\n\n```javascript\nconst samples = ctx.mcp.ListSamples(\"echo\", \"tool\", \"ping\");\nconsole.log(samples.samples); // Array of samples\n```\n\n#### `ctx.mcp.GetSample(client, type, name, index)`\n\nGets a specific sample by index.\n\n**Parameters:**\n\n- `client`: String - MCP client ID\n- `type`: String - Sample type (\"tool\" or \"resource\")\n- `name`: String - Tool or resource name\n- `index`: Number - Sample index (0-based)\n\n```javascript\nconst sample = ctx.mcp.GetSample(\"echo\", \"tool\", \"ping\", 0);\nconsole.log(sample.name, sample.input); // Sample name and input data\n```\n\n### Cross-Server Tool Operations\n\nThese methods enable calling tools across multiple MCP servers concurrently, similar to JavaScript Promise patterns. This is useful for:\n\n- **Parallel data fetching**: Query multiple data sources simultaneously\n- **Redundancy/Fallback**: Try multiple servers, use first successful result\n- **Load balancing**: Distribute load across servers\n\n#### `ctx.mcp.All(requests)`\n\nCalls tools on multiple MCP servers concurrently and waits for all to complete (like `Promise.all`).\n\n**Parameters:**\n\n- `requests`: Array of request objects with `mcp`, `tool`, and optional `arguments`\n\n**Returns:** Array of `MCPToolResult` objects in the same order as requests\n\n```javascript\nconst results = ctx.mcp.All([\n { mcp: \"server1\", tool: \"search\", arguments: { query: \"topic\" } },\n { mcp: \"server2\", tool: \"fetch\", arguments: { id: 123 } },\n { mcp: \"server3\", tool: \"analyze\", arguments: { data: \"input\" } }\n]);\n\n// Process all results\nresults.forEach((r, i) => {\n if (r.error) {\n console.log(`Request ${i} failed: ${r.error}`);\n } else {\n console.log(`Request ${i} result:`, r.result);\n }\n});\n```\n\n#### `ctx.mcp.Any(requests)`\n\nCalls tools on multiple MCP servers concurrently and returns when any succeeds (like `Promise.any`). Useful for redundancy/fallback scenarios.\n\n**Parameters:**\n\n- `requests`: Array of request objects\n\n**Returns:** Array of `MCPToolResult` objects (only contains results received before first success)\n\n```javascript\n// Try multiple search providers, use first successful result\nconst results = ctx.mcp.Any([\n { mcp: \"search-primary\", tool: \"search\", arguments: { q: \"query\" } },\n { mcp: \"search-backup\", tool: \"search\", arguments: { q: \"query\" } }\n]);\n\nconst success = results.find(r => r && !r.error);\nif (success) {\n console.log(\"Search result:\", success.result);\n}\n```\n\n#### `ctx.mcp.Race(requests)`\n\nCalls tools on multiple MCP servers concurrently and returns when any completes (like `Promise.race`). Returns immediately with first completion, regardless of success or failure.\n\n**Parameters:**\n\n- `requests`: Array of request objects\n\n**Returns:** Array of `MCPToolResult` objects (only first completed result is populated)\n\n```javascript\n// Get fastest response\nconst results = ctx.mcp.Race([\n { mcp: \"region-us\", tool: \"ping\", arguments: {} },\n { mcp: \"region-eu\", tool: \"ping\", arguments: {} },\n { mcp: \"region-asia\", tool: \"ping\", arguments: {} }\n]);\n\nconst first = results.find(r => r !== undefined && r !== null);\nconsole.log(`Fastest server: ${first.mcp}`);\n```\n\n#### MCPToolRequest Structure\n\n```typescript\ninterface MCPToolRequest {\n mcp: string; // MCP server ID (required)\n tool: string; // Tool name (required)\n arguments?: any; // Tool arguments (optional)\n}\n```\n\n#### MCPToolResult Structure\n\n```typescript\ninterface MCPToolResult {\n mcp: string; // MCP server ID\n tool: string; // Tool name\n result?: any; // Parsed result content (directly usable)\n error?: string; // Error message (on failure)\n}\n```\n\nThe `result` field contains the automatically parsed content from the MCP response:\n- For text content: JSON parsed if valid JSON, otherwise plain string\n- For image content: `{ type: \"image\", data: \"...\", mimeType: \"...\" }`\n- For resource content: The resource object directly\n- If only one content item exists, returns it directly (not as array)\n\n**Example using parsed result:**\n\n```javascript\n// Single server - direct result\nconst result = ctx.mcp.CallTool(\"echo\", \"echo\", { message: \"hello\" });\nconsole.log(result.echo); // Directly access parsed data\n\n// Cross-server - results array with MCPToolResult objects\nconst results = ctx.mcp.All([\n { mcp: \"echo\", tool: \"echo\", arguments: { message: \"hello\" } }\n]);\nconsole.log(results[0].result.echo); // Access via .result field\n```\n\n## Agent API\n\nThe `ctx.agent` object provides methods to call other agents from within hooks, enabling agent-to-agent communication (A2A). This allows building complex multi-agent workflows where agents can delegate tasks, consult specialists, or orchestrate parallel operations.\n\n### Methods Summary\n\n| Method | Description |\n| ------------------------------- | ---------------------------------------- |\n| `Call(agentID, messages, opts)` | Call a single agent |\n| `All(requests, opts?)` | Call multiple agents, wait for all |\n| `Any(requests, opts?)` | Call multiple agents, first success wins |\n| `Race(requests, opts?)` | Call multiple agents, first complete wins|\n\n### Single Agent Call\n\n#### `ctx.agent.Call(agentID, messages, options?)`\n\nCalls a single agent and streams the response to the current context's output.\n\n**Parameters:**\n\n- `agentID`: String - The target agent/assistant ID\n- `messages`: Array - Messages to send to the agent\n- `options`: Object (optional) - Call options including callback\n\n**Options:**\n\n```typescript\ninterface AgentCallOptions {\n connector?: string; // Override LLM connector\n mode?: string; // Agent mode (\"chat\", \"task\", etc.)\n metadata?: Record; // Custom metadata passed to hooks\n skip?: {\n history?: boolean; // Skip loading chat history\n trace?: boolean; // Skip trace recording\n output?: boolean; // Skip output to client\n keyword?: boolean; // Skip keyword extraction\n search?: boolean; // Skip search\n content_parsing?: boolean; // Skip content parsing\n };\n onChunk?: (msg: Message) => number; // Callback for each message chunk\n}\n```\n\n**Example:**\n\n```javascript\n// Basic call\nconst result = ctx.agent.Call(\"specialist.agent\", [\n { role: \"user\", content: \"Analyze this data\" }\n]);\n\n// With callback\nconst result = ctx.agent.Call(\"specialist.agent\", messages, {\n connector: \"gpt-4o\",\n onChunk: (msg) => {\n console.log(\"Received:\", msg.type, msg.props?.content);\n return 0; // 0 = continue, non-zero = stop\n }\n});\n```\n\n**Returns:**\n\n```typescript\ninterface AgentResult {\n agent_id: string; // Agent ID that was called\n response?: Response; // Full agent response\n content?: string; // Extracted text content\n error?: string; // Error message if failed\n}\n```\n\n**Message Object (received in onChunk callback):**\n\nThe `onChunk` callback receives a `Message` object with the following structure:\n\n```typescript\ninterface Message {\n type: string; // Message type: \"text\", \"thinking\", \"tool_call\", \"error\", etc.\n props?: Record; // Message properties (e.g., { content: \"Hello\" })\n \n // Streaming identifiers\n chunk_id?: string; // Unique chunk ID (C1, C2, ...)\n message_id?: string; // Logical message ID (M1, M2, ...)\n block_id?: string; // Output block ID (B1, B2, ...)\n thread_id?: string; // Thread ID for concurrent calls (T1, T2, ...)\n \n // Delta control\n delta?: boolean; // Whether this is an incremental update\n delta_path?: string; // Update path (e.g., \"content\")\n delta_action?: string; // Update action: \"append\", \"replace\", \"merge\", \"set\"\n}\n```\n\nCommon message types:\n- `\"text\"` - Text content (`props.content` contains the text)\n- `\"thinking\"` - Reasoning/thinking content (o1, DeepSeek R1 models)\n- `\"tool_call\"` - Tool/function call\n- `\"error\"` - Error message (`props.error` contains error details)\n\n### Parallel Agent Calls\n\nThe parallel methods allow calling multiple agents concurrently, similar to JavaScript Promise patterns.\n\n> **Important: SSE Output is Automatically Disabled**\n>\n> For all batch calls (`All`, `Any`, `Race`), SSE output is **automatically disabled** (`skip.output = true`).\n> This prevents multiple agents from writing to the same SSE stream simultaneously, which would cause\n> client disconnection and message corruption. Use the `onChunk` callback to receive streaming messages\n> if needed.\n\n#### `ctx.agent.All(requests, options?)`\n\nExecutes all agent calls and waits for all to complete (like `Promise.all`).\n\n**Parameters:**\n\n- `requests`: Array of request objects\n- `options`: Object (optional) - Global options including callback\n\n**Request Structure:**\n\n```typescript\ninterface AgentRequest {\n agent: string; // Target agent ID\n messages: Message[]; // Messages to send\n options?: AgentCallOptions; // Per-request options (excluding onChunk)\n}\n\n// Note: Per-request onChunk is NOT supported in batch calls.\n// Use the global onChunk callback in the second argument instead.\n// Note: skip.output is automatically set to true for all batch calls.\n```\n\n**Example:**\n\n```javascript\n// Call multiple agents in parallel\nconst results = ctx.agent.All([\n { agent: \"analyzer\", messages: [{ role: \"user\", content: \"Analyze X\" }] },\n { agent: \"summarizer\", messages: [{ role: \"user\", content: \"Summarize Y\" }] }\n]);\n\n// Results array matches request order\nresults.forEach((r, i) => {\n if (r.error) {\n console.log(`Agent ${r.agent_id} failed:`, r.error);\n } else {\n console.log(`Agent ${r.agent_id} response:`, r.content);\n }\n});\n\n// With global callback for all responses\nconst results = ctx.agent.All([\n { agent: \"agent-1\", messages: [...] },\n { agent: \"agent-2\", messages: [...] }\n], {\n onChunk: (agentId, index, msg) => {\n console.log(`Agent ${agentId} [${index}]:`, msg.type, msg.props?.content);\n return 0;\n }\n});\n```\n\n#### `ctx.agent.Any(requests, options?)`\n\nReturns as soon as any agent call succeeds (like `Promise.any`). Other calls continue in background.\n\n**Example:**\n\n```javascript\n// Try multiple agents, use first successful response\nconst results = ctx.agent.Any([\n { agent: \"primary.agent\", messages: [...] },\n { agent: \"fallback.agent\", messages: [...] }\n]);\n\n// First successful result is returned\nconst success = results.find(r => !r.error);\nif (success) {\n console.log(\"Got response from:\", success.agent_id);\n}\n```\n\n#### `ctx.agent.Race(requests, options?)`\n\nReturns as soon as any agent call completes, regardless of success/failure (like `Promise.race`).\n\n**Example:**\n\n```javascript\n// Race multiple agents for fastest response\nconst results = ctx.agent.Race([\n { agent: \"fast.agent\", messages: [...] },\n { agent: \"slow.agent\", messages: [...] }\n]);\n\n// First completed result (may be error or success)\nconst first = results.find(r => r !== null);\nconsole.log(\"Fastest agent:\", first.agent_id);\n```\n\n### Use Cases\n\n```javascript\n// Use case 1: Specialist consultation\nfunction Next(ctx, payload) {\n const { completion } = payload;\n \n if (completion?.content?.includes(\"complex analysis\")) {\n // Delegate to specialist\n const result = ctx.agent.Call(\"specialist.analyzer\", [\n { role: \"user\", content: completion.content }\n ]);\n \n return {\n data: {\n status: \"delegated\",\n specialist_response: result.content\n }\n };\n }\n \n return null;\n}\n\n// Use case 2: Parallel processing\nfunction Create(ctx, messages) {\n const userQuery = messages[messages.length - 1]?.content;\n \n // Query multiple knowledge sources in parallel\n const results = ctx.agent.All([\n { agent: \"kb.technical\", messages: [{ role: \"user\", content: userQuery }] },\n { agent: \"kb.business\", messages: [{ role: \"user\", content: userQuery }] },\n { agent: \"kb.legal\", messages: [{ role: \"user\", content: userQuery }] }\n ]);\n \n // Combine results\n const combinedKnowledge = results\n .filter(r => !r.error)\n .map(r => r.content)\n .join(\"\\n\\n\");\n \n // Add to messages\n return {\n messages: [\n ...messages,\n { role: \"system\", content: `Relevant knowledge:\\n${combinedKnowledge}` }\n ]\n };\n}\n\n// Use case 3: Fallback strategy\nfunction Next(ctx, payload) {\n if (payload.error) {\n // Try backup agents\n const results = ctx.agent.Any([\n { agent: \"backup.gpt4\", messages: payload.messages },\n { agent: \"backup.claude\", messages: payload.messages }\n ]);\n \n const success = results.find(r => !r.error);\n if (success) {\n return { data: { recovered: true, content: success.content } };\n }\n }\n \n return null;\n}\n```\n\n## Sandbox API\n\nThe `ctx.sandbox` object provides access to sandbox operations when the assistant is configured with a sandbox executor (e.g., Claude CLI, Cursor CLI). The sandbox allows hooks to interact with an isolated Docker container environment for file operations and command execution.\n\n> **Note:** `ctx.sandbox` is only available when the assistant has `sandbox` configuration in `package.yao`. If no sandbox is configured, `ctx.sandbox` will be `null`.\n\n### Properties\n\n- `ctx.sandbox.workdir`: String - The workspace directory path inside the container (e.g., `/workspace`)\n\n### Methods Summary\n\n| Method | Description |\n| ----------------------------- | ---------------------------------------- |\n| `ReadFile(path)` | Read a file from the container |\n| `WriteFile(path, content)` | Write content to a file in the container |\n| `ListDir(path)` | List directory contents |\n| `Exec(command)` | Execute a command in the container |\n\n### File Operations\n\n#### `ctx.sandbox.ReadFile(path): string`\n\nReads a file from the sandbox container.\n\n**Parameters:**\n\n- `path`: String - File path (relative to workdir or absolute)\n\n**Returns:**\n\n- `string`: File contents as string\n\n**Example:**\n\n```javascript\n// Read a file from workspace\nconst content = ctx.sandbox.ReadFile(\"config.json\");\nconsole.log(content);\n\n// Read with absolute path\nconst readme = ctx.sandbox.ReadFile(\"/workspace/README.md\");\n```\n\n#### `ctx.sandbox.WriteFile(path, content): void`\n\nWrites content to a file in the sandbox container.\n\n**Parameters:**\n\n- `path`: String - File path (relative to workdir or absolute)\n- `content`: String - Content to write\n\n**Example:**\n\n```javascript\n// Write a configuration file\nctx.sandbox.WriteFile(\"config.json\", JSON.stringify({ debug: true }));\n\n// Write a script\nctx.sandbox.WriteFile(\"script.sh\", \"#!/bin/bash\\necho 'Hello'\");\n```\n\n#### `ctx.sandbox.ListDir(path): FileInfo[]`\n\nLists the contents of a directory in the sandbox container.\n\n**Parameters:**\n\n- `path`: String - Directory path (relative to workdir or absolute)\n\n**Returns:**\n\n- `FileInfo[]`: Array of file information objects\n\n**FileInfo Structure:**\n\n```typescript\ninterface FileInfo {\n name: string; // File or directory name\n size: number; // Size in bytes\n is_dir: boolean; // True if directory\n}\n```\n\n**Example:**\n\n```javascript\n// List workspace contents\nconst files = ctx.sandbox.ListDir(\".\");\nfiles.forEach(f => {\n console.log(`${f.is_dir ? \"DIR\" : \"FILE\"} ${f.name} (${f.size} bytes)`);\n});\n\n// List specific directory\nconst srcFiles = ctx.sandbox.ListDir(\"src\");\n```\n\n### Command Execution\n\n#### `ctx.sandbox.Exec(command): string`\n\nExecutes a command in the sandbox container and returns the output.\n\n**Parameters:**\n\n- `command`: String[] - Command and arguments as an array\n\n**Returns:**\n\n- `string`: Command stdout output\n\n**Throws:**\n\n- Error if command exits with non-zero code (includes stderr in error message)\n\n**Example:**\n\n```javascript\n// Run a simple command\nconst output = ctx.sandbox.Exec([\"echo\", \"Hello, World!\"]);\nconsole.log(output); // \"Hello, World!\\n\"\n\n// Run git commands\nconst status = ctx.sandbox.Exec([\"git\", \"status\"]);\nconsole.log(status);\n\n// Run npm install\ntry {\n const result = ctx.sandbox.Exec([\"npm\", \"install\"]);\n console.log(\"Install complete:\", result);\n} catch (e) {\n console.error(\"Install failed:\", e.message);\n}\n\n// Run shell script\nctx.sandbox.WriteFile(\"test.sh\", \"#!/bin/bash\\necho 'Running script'\\nls -la\");\nctx.sandbox.Exec([\"chmod\", \"+x\", \"test.sh\"]);\nconst scriptOutput = ctx.sandbox.Exec([\"./test.sh\"]);\n```\n\n### Use Cases\n\n```javascript\n// Use case 1: Prepare workspace before Claude CLI execution\nfunction Create(ctx, messages) {\n if (ctx.sandbox) {\n // Create project structure\n ctx.sandbox.WriteFile(\"package.json\", JSON.stringify({\n name: \"project\",\n version: \"1.0.0\"\n }, null, 2));\n \n // Write initial code\n ctx.sandbox.WriteFile(\"src/index.ts\", \"console.log('Hello');\");\n \n ctx.trace.Info(\"Workspace prepared\");\n }\n return { messages };\n}\n\n// Use case 2: Post-process sandbox results\nfunction Next(ctx, payload) {\n if (ctx.sandbox && !payload.error) {\n // Read generated files\n try {\n const files = ctx.sandbox.ListDir(\"output\");\n const results = files.map(f => ({\n name: f.name,\n content: ctx.sandbox.ReadFile(`output/${f.name}`)\n }));\n \n return {\n data: {\n status: \"success\",\n generated_files: results\n }\n };\n } catch (e) {\n ctx.trace.Warn(\"No output directory found\");\n }\n }\n return null;\n}\n\n// Use case 3: Run tests after code generation\nfunction Next(ctx, payload) {\n if (ctx.sandbox && payload.completion) {\n try {\n // Run tests\n const testOutput = ctx.sandbox.Exec([\"npm\", \"test\"]);\n ctx.trace.Info(\"Tests passed\");\n \n return {\n data: {\n status: \"success\",\n test_output: testOutput\n }\n };\n } catch (e) {\n ctx.trace.Error(\"Tests failed: \" + e.message);\n return {\n data: {\n status: \"test_failed\",\n error: e.message\n }\n };\n }\n }\n return null;\n}\n```\n\n### Sandbox Configuration\n\nThe sandbox is configured in the assistant's `package.yao`:\n\n```jsonc\n{\n \"name\": \"Coder Assistant\",\n \"connector\": \"deepseek.v3\",\n \"sandbox\": {\n \"command\": \"claude\", // claude | cursor (future)\n \"image\": \"yaoapp/sandbox-claude:latest\", // Optional, auto-selected by command\n \"max_memory\": \"4g\", // Memory limit (optional)\n \"max_cpu\": 2.0, // CPU limit (optional)\n \"timeout\": \"10m\", // Execution timeout\n \"arguments\": { // Command-specific arguments\n \"max_turns\": 20,\n \"permission_mode\": \"acceptEdits\"\n }\n }\n}\n```\n\n### Notes\n\n- Sandbox operations are **synchronous** - they block until complete\n- File paths can be relative (to workdir) or absolute\n- Relative paths are resolved against the `workdir` directory\n- The sandbox container is created at the start of the request and removed when the request completes\n- Commands are executed with the sandbox user's permissions\n- Errors throw JavaScript exceptions - use try/catch for error handling\n- Large file operations may timeout - use appropriate timeout settings\n\n## LLM API\n\nThe `ctx.llm` object provides direct access to LLM connectors for streaming completions. This allows calling LLM models directly without going through the full agent pipeline, useful for quick completions, model comparisons, or building custom workflows.\n\n### Methods Summary\n\n| Method | Description |\n| --------------------------------- | -------------------------------------- |\n| `Stream(connector, messages, opts)` | Stream LLM completion |\n| `All(requests, opts?)` | Call multiple LLMs, wait for all |\n| `Any(requests, opts?)` | Call multiple LLMs, first success wins |\n| `Race(requests, opts?)` | Call multiple LLMs, first complete wins|\n\n### Single LLM Call\n\n#### `ctx.llm.Stream(connector, messages, options?)`\n\nCalls an LLM connector with streaming output to the current context's writer.\n\n**Parameters:**\n\n- `connector`: String - The LLM connector ID (e.g., \"gpt-4o\", \"claude-3\")\n- `messages`: Array - Messages to send to the LLM\n- `options`: Object (optional) - LLM options including callback\n\n**Options:**\n\n```typescript\ninterface LlmOptions {\n temperature?: number; // Sampling temperature (0-2)\n max_tokens?: number; // Max tokens (legacy, use max_completion_tokens)\n max_completion_tokens?: number; // Max completion tokens\n top_p?: number; // Nucleus sampling\n presence_penalty?: number; // Presence penalty (-2 to 2)\n frequency_penalty?: number; // Frequency penalty (-2 to 2)\n stop?: string | string[]; // Stop sequences\n user?: string; // User identifier for tracking\n seed?: number; // Random seed for reproducibility\n tools?: object[]; // Function/tool definitions\n tool_choice?: string | object; // Tool choice strategy\n response_format?: { // Response format\n type: string; // \"text\" | \"json_object\" | \"json_schema\"\n json_schema?: {\n name: string;\n description?: string;\n schema: object;\n strict?: boolean;\n };\n };\n reasoning_effort?: string; // For reasoning models (e.g., \"low\", \"medium\", \"high\")\n onChunk?: (msg: Message) => number; // Callback for each chunk\n}\n```\n\n**Example:**\n\n```javascript\n// Basic streaming call\nconst result = ctx.llm.Stream(\"gpt-4o\", [\n { role: \"system\", content: \"You are a helpful assistant.\" },\n { role: \"user\", content: \"Explain quantum computing\" }\n]);\n\n// With options and callback\nconst result = ctx.llm.Stream(\"gpt-4o\", messages, {\n temperature: 0.7,\n max_tokens: 2000,\n onChunk: (msg) => {\n console.log(\"Chunk:\", msg.type, msg.props?.content);\n return 0; // 0 = continue, non-zero = stop\n }\n});\n\nconsole.log(\"Full response:\", result.content);\n```\n\n**Returns:**\n\n```typescript\ninterface LlmResult {\n connector: string; // Connector ID used\n response?: CompletionResponse; // Full completion response\n content?: string; // Extracted text content\n error?: string; // Error message if failed\n}\n```\n\n### Parallel LLM Calls\n\nThe parallel methods allow calling multiple LLM connectors concurrently, useful for model comparison, ensemble methods, or fallback strategies.\n\n#### `ctx.llm.All(requests, options?)`\n\nExecutes all LLM calls and waits for all to complete (like `Promise.all`).\n\n**Request Structure:**\n\n```typescript\ninterface LlmRequest {\n connector: string; // LLM connector ID\n messages: Message[]; // Messages to send\n options?: LlmOptions; // Per-request options (excluding onChunk)\n}\n```\n\n**Example:**\n\n```javascript\n// Compare responses from multiple models\nconst results = ctx.llm.All([\n { connector: \"gpt-4o\", messages: [...], options: { temperature: 0.7 } },\n { connector: \"claude-3\", messages: [...], options: { temperature: 0.7 } },\n { connector: \"gemini-pro\", messages: [...] }\n]);\n\nresults.forEach((r) => {\n console.log(`${r.connector}: ${r.content?.substring(0, 100)}...`);\n});\n\n// With global callback\nconst results = ctx.llm.All([\n { connector: \"gpt-4o\", messages: [...] },\n { connector: \"claude-3\", messages: [...] }\n], {\n onChunk: (connectorId, index, msg) => {\n console.log(`LLM ${connectorId} [${index}]:`, msg.props?.content);\n return 0;\n }\n});\n```\n\n#### `ctx.llm.Any(requests, options?)`\n\nReturns as soon as any LLM call succeeds (like `Promise.any`).\n\n**Example:**\n\n```javascript\n// Use first successful response from any model\nconst results = ctx.llm.Any([\n { connector: \"gpt-4o\", messages: [...] },\n { connector: \"gpt-4o-mini\", messages: [...] }\n]);\n\nconst success = results.find(r => !r.error);\nif (success) {\n ctx.Send(success.content);\n}\n```\n\n#### `ctx.llm.Race(requests, options?)`\n\nReturns as soon as any LLM call completes (like `Promise.race`).\n\n**Example:**\n\n```javascript\n// Get fastest response\nconst results = ctx.llm.Race([\n { connector: \"gpt-4o-mini\", messages: [...] }, // Usually faster\n { connector: \"gpt-4o\", messages: [...] } // Usually slower\n]);\n\nconst first = results.find(r => r !== null);\nconsole.log(\"Fastest model:\", first.connector);\n```\n\n### Use Cases\n\n```javascript\n// Use case 1: Quick classification without full agent pipeline\nfunction Create(ctx, messages) {\n const userMessage = messages[messages.length - 1]?.content;\n \n // Quick intent classification\n const result = ctx.llm.Stream(\"gpt-4o-mini\", [\n { role: \"system\", content: \"Classify intent as: question, command, or chat\" },\n { role: \"user\", content: userMessage }\n ], { temperature: 0, max_tokens: 10 });\n \n const intent = result.content?.toLowerCase();\n ctx.memory.context.Set(\"intent\", intent);\n \n return { messages };\n}\n\n// Use case 2: Model comparison for quality assurance\nfunction Next(ctx, payload) {\n const { completion } = payload;\n \n // Get second opinion from different model\n const results = ctx.llm.All([\n { connector: \"gpt-4o\", messages: payload.messages },\n { connector: \"claude-3-opus\", messages: payload.messages }\n ]);\n \n // Compare responses\n const gptResponse = results[0].content;\n const claudeResponse = results[1].content;\n \n return {\n data: {\n primary: completion.content,\n comparisons: {\n gpt4o: gptResponse,\n claude: claudeResponse\n }\n }\n };\n}\n\n// Use case 3: Ensemble with voting\nfunction Create(ctx, messages) {\n // Get multiple model opinions for important decisions\n const results = ctx.llm.All([\n { connector: \"gpt-4o\", messages: [...] },\n { connector: \"claude-3\", messages: [...] },\n { connector: \"gemini-pro\", messages: [...] }\n ]);\n \n // Simple majority voting (in real use, implement proper consensus)\n const responses = results.filter(r => !r.error).map(r => r.content);\n \n return {\n messages: [\n ...messages,\n { \n role: \"system\", \n content: `Multiple model opinions:\\n${responses.map((r, i) => `Model ${i+1}: ${r}`).join('\\n')}`\n }\n ]\n };\n}\n\n// Use case 4: Fallback with latency optimization\nfunction Next(ctx, payload) {\n if (payload.error) {\n // Race multiple fallback models\n const results = ctx.llm.Race([\n { connector: \"gpt-4o-mini\", messages: payload.messages },\n { connector: \"claude-3-haiku\", messages: payload.messages }\n ]);\n \n const fastest = results.find(r => r !== null);\n if (fastest && !fastest.error) {\n ctx.Send(fastest.content);\n return { data: { recovered: true, model: fastest.connector } };\n }\n }\n \n return null;\n}\n```\n\n## Hooks\n\nThe Agent system supports two hooks that can be defined in the assistant's `index.ts` file: `Create` and `Next`.\n\n### Agent Execution Lifecycle\n\n```mermaid\nflowchart TD\n A[User Input] --> B[Load History]\n B --> C{Create Hook?}\n C -->|Yes| D[Execute Create Hook]\n C -->|No| E{Has Prompts/MCP?}\n D --> E\n E -->|Yes| F[Build LLM Request]\n E -->|No| K\n F --> G[LLM Stream Call]\n G --> H{Tool Calls?}\n H -->|Yes| I[Execute Tools]\n I --> J{Tool Errors?}\n J -->|Yes, Retry| G\n J -->|No| K\n H -->|No| K\n K{Next Hook?}\n K -->|Yes| L[Execute Next Hook]\n K -->|No| M[Return Response]\n L --> N{Delegate?}\n N -->|Yes| O[Call Target Agent]\n O --> M\n N -->|No| M\n M --> P[End]\n\n style D fill:#e1f5fe\n style L fill:#e1f5fe\n style G fill:#fff3e0\n style I fill:#f3e5f5\n```\n\n> **Note:** LLM call is optional. If the assistant has no prompts and no MCP servers configured, the LLM call is skipped. Hooks can be used independently to implement custom logic without LLM involvement.\n\n### Create Hook\n\nCalled at the beginning of agent execution, before any LLM call. Use this to preprocess messages, add context, configure the request, or implement custom logic.\n\n**Signature:**\n\n```typescript\nfunction Create(\n ctx: Context,\n messages: Message[],\n options?: Record\n): HookCreateResponse | null;\n```\n\n**Parameters:**\n\n- `ctx`: Context object\n- `messages`: Array of input messages (including chat history if enabled)\n- `options`: Optional call-level options (see below)\n\n**Options Structure:**\n\n```typescript\ninterface Options {\n skip?: {\n history?: boolean; // Skip loading/saving chat history\n trace?: boolean; // Skip trace recording\n output?: boolean; // Skip output to client (for internal A2A calls that only need response data)\n };\n connector?: string; // Override LLM connector ID\n disable_global_prompts?: boolean; // Disable global prompts for this request\n search?: boolean; // Enable/disable search mode\n mode?: string; // Agent mode (default: \"chat\")\n}\n```\n\n**Return Value (`HookCreateResponse`):**\n\n```typescript\ninterface HookCreateResponse {\n // Messages to be sent to the assistant (can modify/replace input messages)\n messages?: Message[];\n\n // Audio configuration (for models that support audio output)\n audio?: AudioConfig;\n\n // Generation parameters (override assistant defaults)\n temperature?: number;\n max_tokens?: number;\n max_completion_tokens?: number;\n\n // MCP configuration - add/override MCP servers for this request\n mcp_servers?: MCPServerConfig[];\n\n // Prompt configuration\n prompt_preset?: string; // Select prompt preset (e.g., \"chat.friendly\", \"task.analysis\")\n disable_global_prompts?: boolean; // Temporarily disable global prompts for this request\n\n // Context adjustments - allow hook to modify context fields\n connector?: string; // Override connector (call-level)\n locale?: string; // Override locale (session-level)\n theme?: string; // Override theme (session-level)\n route?: string; // Override route (session-level)\n metadata?: Record; // Override or merge metadata (session-level)\n\n // Uses configuration - allow hook to override wrapper configurations\n uses?: UsesConfig; // Override wrapper configurations for vision, audio, search, and fetch\n force_uses?: boolean; // Force using Uses tools regardless of model capabilities\n}\n\n// Audio output configuration\ninterface AudioConfig {\n voice: string; // Voice to use (e.g., \"alloy\", \"echo\", \"fable\", \"onyx\", \"nova\", \"shimmer\")\n format: string; // Audio format (e.g., \"wav\", \"mp3\", \"flac\", \"opus\", \"pcm16\")\n}\n\n// MCP server configuration\ninterface MCPServerConfig {\n server_id: string; // MCP server ID (required)\n tools?: string[]; // Tool name filter (empty = all tools)\n resources?: string[]; // Resource URI filter (empty = all resources)\n}\n\n// Uses wrapper configuration\ninterface UsesConfig {\n vision?: string; // Vision processing tool. Format: \"agent\" or \"mcp:server_id\"\n audio?: string; // Audio processing tool. Format: \"agent\" or \"mcp:server_id\"\n search?: string; // Search tool. Format: \"agent\" or \"mcp:server_id\"\n fetch?: string; // Fetch/retrieval tool. Format: \"agent\" or \"mcp:server_id\"\n}\n```\n\n**Example:**\n\n```javascript\nfunction Create(ctx, messages) {\n // Store data for Next hook\n ctx.memory.context.Set(\"user_query\", messages[0]?.content);\n\n // Modify messages\n const enhanced_messages = messages.map((msg) => ({\n ...msg,\n content: msg.content + \"\\n\\nPlease be concise.\",\n }));\n\n // Return configuration\n return {\n messages: enhanced_messages,\n temperature: 0.7,\n max_tokens: 2000,\n };\n}\n```\n\n### Next Hook\n\nCalled after the LLM response and tool calls (if any), or directly after Create Hook if no LLM call is configured. Use this to post-process the response, send custom messages, delegate to another agent, or implement custom response logic.\n\n**Signature:**\n\n```typescript\nfunction Next(\n ctx: Context,\n payload: NextHookPayload,\n options?: Record\n): NextHookResponse | null;\n```\n\n**Parameters:**\n\n- `ctx`: Context object\n- `payload`: Object containing:\n- `options`: Optional call-level options (same structure as Create Hook options)\n\n```typescript\ninterface NextHookPayload {\n messages: Message[]; // Messages sent to the assistant\n completion?: CompletionResponse; // LLM response\n tools?: ToolCallResponse[]; // Tool call results (if any)\n error?: string; // Error message if LLM call failed\n}\n\ninterface CompletionResponse {\n content: string; // LLM text response\n tool_calls?: ToolCall[]; // Tool calls requested by LLM\n usage?: UsageInfo; // Token usage statistics\n}\n\ninterface ToolCallResponse {\n toolcall_id: string;\n server: string; // MCP server name\n tool: string; // Tool name\n arguments?: any; // Arguments passed to tool\n result?: any; // Tool execution result\n error?: string; // Error if tool failed\n}\n```\n\n**Return Value (`NextHookResponse`):**\n\n```typescript\ninterface NextHookResponse {\n // Delegate to another agent (recursive call)\n // If provided, the current agent will call the target agent\n delegate?: {\n agent_id: string; // Required: target agent ID\n messages: Message[]; // Messages to send to target agent\n options?: Record; // Optional: call-level options for delegation\n };\n\n // Custom response data\n // Will be placed in Response.next field and returned to user\n // If both delegate and data are null/undefined, standard Response is returned\n data?: any;\n\n // Metadata for debugging and logging\n metadata?: Record;\n}\n```\n\n**Agent Response Structure:**\n\nThe agent's `Stream()` method returns a `Response` object:\n\n```typescript\ninterface Response {\n request_id: string; // Request ID\n context_id: string; // Context ID\n trace_id: string; // Trace ID\n chat_id: string; // Chat ID\n assistant_id: string; // Assistant ID\n create?: HookCreateResponse; // Create hook response\n next?: any; // See below for what this contains\n completion?: CompletionResponse; // LLM completion response\n}\n```\n\n**Response.next field logic:**\n\n- If `NextHookResponse.data` is provided → `Response.next` = custom data\n- If `NextHookResponse.data` is null/undefined → `Response.next` = entire `NextHookResponse` object\n- If no Next hook defined → `Response.next` = null\n\n**Example:**\n\n```javascript\n/**\n * Next Hook - Process LLM response\n * @param {Context} ctx - Agent context\n * @param {NextHookPayload} payload - Contains messages, completion, tools, error\n * @returns {NextHookResponse | null} - Return null for standard response\n */\nfunction Next(ctx, payload) {\n const { messages, completion, tools, error } = payload;\n\n // Handle errors gracefully\n if (error) {\n return {\n data: {\n status: \"error\",\n message: error,\n recovery: \"Please try again\",\n },\n metadata: { error_handled: true },\n };\n }\n\n // Process tool results if any\n if (tools && tools.length > 0) {\n const successful = tools.filter((t) => !t.error);\n const failed = tools.filter((t) => t.error);\n\n return {\n data: {\n status: \"tools_processed\",\n total: tools.length,\n successful: successful.length,\n failed: failed.length,\n results: successful.map((t) => t.result),\n },\n metadata: { has_failures: failed.length > 0 },\n };\n }\n\n // Return custom data based on completion\n if (completion && completion.content) {\n return {\n data: {\n status: \"success\",\n response: completion.content,\n processed: true,\n },\n metadata: { source: \"next_hook\" },\n };\n }\n\n // Return null to use standard response\n return null;\n}\n```\n\n### Hook Execution Flow\n\nSee the [Agent Execution Lifecycle](#agent-execution-lifecycle) diagram above for a visual representation.\n\n**Key Points:**\n\n- **Hooks are optional** - if not defined, the agent uses default behavior\n- **LLM call is optional** - only executed if the assistant has prompts or MCP servers configured\n- **Return `null` or `undefined`** from hooks to use default behavior\n- **Hooks can send messages directly** via `ctx.Send()`, `ctx.SendStream()`, etc.\n- **Create Hook** runs before LLM call (if any), can modify messages and configure the request\n- **Next Hook** runs after LLM call and tool execution (if any), can post-process or delegate\n- Use `ctx.memory.context` to pass data between Create and Next hooks within a request\n\n## Complete Example\n\nHere's a comprehensive example demonstrating Create and Next hooks with various Context API features:\n\n```javascript\n/**\n * Create Hook - Preprocess messages and configure the request\n *\n * @param {Context} ctx - Agent context object\n * @param {Message[]} messages - Input messages (including history if enabled)\n * @returns {HookCreateResponse | null} - Configuration for LLM call, or null for defaults\n */\nfunction Create(ctx, messages) {\n // Extract user query from the last message\n const user_query = messages[messages.length - 1]?.content || \"\";\n\n // Store data in context memory for use in Next hook\n ctx.memory.context.Set(\"original_query\", user_query);\n ctx.memory.context.Set(\"request_time\", Date.now());\n\n // Add trace node to show processing in UI\n const create_node = ctx.trace.Add(\n { query: user_query },\n {\n label: \"Create Hook\",\n type: \"preprocessing\",\n icon: \"play\",\n description: \"Analyzing user request\",\n }\n );\n\n // Check if user needs search functionality\n const needs_search =\n user_query.toLowerCase().includes(\"search\") ||\n user_query.toLowerCase().includes(\"find\");\n\n if (needs_search) {\n create_node.Info(\"Search mode enabled\");\n\n // Configure MCP servers for search\n return {\n messages: messages,\n mcp_servers: [{ server_id: \"search_engine\" }],\n prompt_preset: \"search.assistant\",\n metadata: { mode: \"search\" },\n };\n }\n\n create_node.Complete({ mode: \"standard\" });\n\n // Return modified messages or configuration\n return {\n messages: messages,\n temperature: 0.7,\n max_tokens: 2000,\n };\n}\n\n/**\n * Next Hook - Process LLM response and optionally customize output\n *\n * @param {Context} ctx - Agent context object\n * @param {NextHookPayload} payload - Contains messages, completion, tools, error\n * @returns {NextHookResponse | null} - Custom response, delegation, or null for standard\n */\nfunction Next(ctx, payload) {\n const { messages, completion, tools, error } = payload;\n\n // Retrieve data from Create hook via context memory\n const original_query = ctx.memory.context.Get(\"original_query\");\n const request_time = ctx.memory.context.Get(\"request_time\");\n const duration = Date.now() - request_time;\n\n // Create trace node for Next hook processing\n const next_node = ctx.trace.Add(\n { completion_length: completion?.content?.length || 0 },\n {\n label: \"Next Hook\",\n type: \"postprocessing\",\n icon: \"check\",\n description: \"Processing LLM response\",\n }\n );\n\n // Handle errors\n if (error) {\n next_node.Fail(error);\n return {\n data: {\n status: \"error\",\n message: \"An error occurred while processing your request\",\n error: error,\n },\n };\n }\n\n // Process tool call results\n if (tools && tools.length > 0) {\n next_node.Info(`Processing ${tools.length} tool results`);\n\n const successful = tools.filter((t) => !t.error);\n const results = successful.map((t) => ({\n tool: t.tool,\n server: t.server,\n result: t.result,\n }));\n\n // Send streaming message with results\n const msg_id = ctx.SendStream(\"## Tool Results\\n\\n\");\n results.forEach((r, i) => {\n ctx.Append(msg_id, `**${i + 1}. ${r.tool}**\\n`);\n ctx.Append(msg_id, `${JSON.stringify(r.result, null, 2)}\\n\\n`);\n });\n ctx.End(msg_id);\n\n next_node.SetMetadata(\"tools_processed\", tools.length);\n next_node.Complete({ status: \"tools_processed\" });\n\n return {\n data: {\n status: \"success\",\n tool_results: results,\n duration_ms: duration,\n },\n metadata: { processed_by: \"next_hook\" },\n };\n }\n\n // Check if delegation is needed based on completion content\n if (completion?.content?.toLowerCase().includes(\"delegate to specialist\")) {\n next_node.Info(\"Delegating to specialist agent\");\n\n return {\n delegate: {\n agent_id: \"specialist.agent\",\n messages: [\n { role: \"system\", content: \"Handle this specialized request\" },\n { role: \"user\", content: original_query },\n ],\n options: { priority: \"high\" },\n },\n metadata: { reason: \"specialist_needed\" },\n };\n }\n\n // Standard processing - add metadata and return\n next_node.SetMetadata(\"duration_ms\", duration);\n next_node.Complete({ status: \"success\" });\n\n // Return null to use standard LLM response\n // Or return custom data to override\n return null;\n}\n```\n\n## Best Practices\n\n1. **Error Handling**: Always wrap Context operations in try-catch blocks\n2. **Resource Cleanup**: Only call `ctx.Release()` for manually created Context, not in hooks\n3. **Trace Organization**: Create meaningful trace nodes with descriptive labels\n4. **Logging Levels**: Use appropriate log levels (Debug for development, Info for progress, Error for failures)\n5. **Message IDs**: Let the system auto-generate message IDs unless you need specific tracking\n6. **Parallel Operations**: Use `Trace.Parallel()` for concurrent operations to maintain trace clarity\n7. **Memory Usage**: Use `ctx.memory.context` for request-scoped data, `ctx.memory.chat` for chat state, `ctx.memory.user` for user preferences\n8. **Streaming Messages**: Use `SendStream()` + `Append()` + `End()` for streaming output; use `Send()` for complete messages\n9. **Block Grouping**: Only use Block IDs when you need to group multiple messages together (e.g., LLM output + follow-up card)\n\n## Error Handling\n\nAll Context methods throw exceptions on failure. Always handle errors appropriately:\n\n```javascript\ntry {\n ctx.Send(message);\n} catch (error) {\n ctx.trace.Error(\"Failed to send message\", { error: error.message });\n throw error;\n}\n```\n\n## TypeScript Support\n\nFor TypeScript projects, the Context types are automatically inferred. You can also import explicit types:\n\n```typescript\nimport { Context, Message, TraceNodeOption } from \"@yao/runtime\";\n\ninterface NextPayload {\n messages: Message[];\n completion: any;\n tools: any[];\n error?: string;\n}\n\nfunction Next(\n ctx: Context,\n payload: NextHookPayload,\n options?: Record\n): NextHookResponse | null {\n // Your code with full type checking\n const { messages, completion, tools, error } = payload;\n // ...\n}\n```\n\n## See Also\n\n- [Agent Hooks Documentation](../hooks/README.md)\n- [MCP Protocol Specification](../mcp/README.md)\n- [Trace System Documentation](../../trace/README.md)\n- [Message Format Specification](../message/README.md)\n" }, { "path": "agent/context/RESOURCE_MANAGEMENT.md", "content": "# Context Resource Management\n\nThis document explains the resource management strategy for Context and Trace objects in JavaScript.\n\n## Overview\n\nBoth `Context` and `Trace` objects provide two cleanup methods:\n\n- **`__release()`** - Internal method called automatically by:\n - V8 garbage collector (when object is collected)\n - `Use()` function (immediate cleanup after callback)\n\n- **`Release()`** - Public method for explicit manual cleanup:\n - Called in `try-finally` blocks\n - Provides immediate resource cleanup\n - Same implementation as `__release()` - they do the same thing\n\n## Resource Hierarchy\n\nWhen `Context.Release()` is called, it automatically releases:\n\n1. **Trace object** - If present, calls `Trace.__release()` to cleanup:\n - Go bridge registry entries\n - Trace manager resources\n - Background goroutines\n\n2. **Context object** - Releases:\n - Go bridge registry entry for the Context itself\n\nThis ensures proper cleanup of the entire resource tree.\n\n## Usage Patterns\n\n### Pattern 1: Automatic Cleanup with `Use()` (Recommended)\n\n**Best for**: Most cases, clean code, automatic resource management\n\n```javascript\n// Context is released automatically after callback\nUse(Context, contextData, (ctx) => {\n // Access Trace (released automatically with context)\n const trace = ctx.Trace\n const node = trace.Add({ type: \"step\" }, { label: \"Processing\" })\n \n trace.Info(\"Doing work\")\n node.Complete({ result: \"done\" })\n \n return result\n})\n// ctx.Release() called automatically, which also releases Trace\n```\n\n### Pattern 2: Manual Cleanup with `try-finally`\n\n**Best for**: Explicit control, critical memory scenarios\n\n```javascript\nconst ctx = getContext() // or passed as parameter\nconst trace = ctx.Trace\n\ntry {\n const node = trace.Add({ type: \"step\" }, { label: \"Processing\" })\n \n trace.Info(\"Doing work\")\n node.Complete({ result: \"done\" })\n \n return result\n} finally {\n // Explicit cleanup (also releases Trace)\n ctx.Release()\n}\n```\n\n### Pattern 3: Separate Trace Cleanup\n\n**Best for**: When you want to release Trace independently\n\n```javascript\nconst ctx = getContext()\nconst trace = ctx.Trace\n\ntry {\n const node = trace.Add({ type: \"step\" }, { label: \"Processing\" })\n trace.Info(\"Doing work\")\n node.Complete({ result: \"done\" })\n \n // Release trace early if needed\n trace.Release()\n \n // Continue using ctx...\n return result\n} finally {\n // Release context (Trace already released, safe to call again)\n ctx.Release()\n}\n```\n\n### Pattern 4: No Explicit Cleanup (Not Recommended)\n\n**Avoid in production**: Relies on GC, unpredictable timing\n\n```javascript\nfunction processData(ctx) {\n const trace = ctx.Trace\n const node = trace.Add({ type: \"step\" }, { label: \"Processing\" })\n \n trace.Info(\"Doing work\")\n node.Complete({ result: \"done\" })\n \n return result\n // Waits for V8 GC to call __release() - SLOW!\n}\n```\n\n## No-op Trace Handling\n\nWhen Trace is not initialized, `ctx.Trace` returns a no-op object:\n\n- All methods are no-ops (do nothing)\n- `Release()` is safe to call (no-op)\n- No errors are thrown\n- Provides consistent API regardless of trace initialization\n\n```javascript\n// Works even if Trace is not initialized\nconst ctx = getContext()\nconst trace = ctx.Trace // might be no-op\n\ntrace.Info(\"Message\") // safe even if no-op\ntrace.Release() // safe even if no-op\nctx.Release() // always safe\n```\n\n## Error Handling\n\nCleanup happens even when errors occur:\n\n```javascript\nconst ctx = getContext()\ntry {\n const trace = ctx.Trace\n const node = trace.Add({ type: \"step\" }, { label: \"Processing\" })\n \n throw new Error(\"Something went wrong\")\n \n} finally {\n // Cleanup still happens\n ctx.Release() // also releases Trace\n}\n```\n\nWith `Use()`:\n\n```javascript\ntry {\n Use(Context, contextData, (ctx) => {\n throw new Error(\"Something went wrong\")\n })\n} catch (error) {\n // Error is caught\n // ctx.Release() was already called automatically\n}\n```\n\n## Memory Management\n\n### ✅ Good: Immediate Cleanup\n\n```javascript\n// Loop with immediate cleanup\nfor (let i = 0; i < 10000; i++) {\n Use(Context, data, (ctx) => {\n const trace = ctx.Trace\n trace.Info(`Processing item ${i}`)\n // Released immediately after each iteration\n })\n}\n```\n\n### ❌ Bad: Waiting for GC\n\n```javascript\n// Memory accumulates until GC runs\nfor (let i = 0; i < 10000; i++) {\n const ctx = getContext()\n const trace = ctx.Trace\n trace.Info(`Processing item ${i}`)\n // No cleanup - may run out of memory!\n}\n```\n\n## Implementation Details\n\n### Context.Release() / Context.__release()\n\n1. Checks if `ctx.Trace` exists\n2. If yes, calls `trace.__release()` to cleanup Trace resources\n3. Releases Context from bridge registry\n4. Safe to call multiple times (idempotent)\n5. Errors in cleanup are silently ignored\n\n### Trace.Release() / Trace.__release()\n\n1. Releases Go manager object from bridge registry\n2. Calls `trace.Release(traceID)` to cleanup:\n - Remove from global trace registry\n - Stop background goroutines\n - Free associated resources\n3. Safe to call multiple times (idempotent)\n\n### No-op Objects\n\nBoth no-op Trace and no-op Node provide:\n- All methods as no-ops\n- `Release()` and `__release()` methods\n- Consistent API for error-free operation\n- Zero memory overhead\n\n## Best Practices\n\n1. **✅ Use `Use()` for automatic cleanup** in most cases\n2. **✅ Use `try-finally` with `Release()`** when you need explicit control\n3. **✅ Release Context** (which also releases Trace) rather than releasing each separately\n4. **✅ Release resources in loops** to prevent memory accumulation\n5. **❌ Don't rely on GC** for resource cleanup in production code\n6. **❌ Don't worry about calling `Release()` twice** - it's idempotent\n\n## Testing\n\nSee `jsapi_release_test.go` for comprehensive tests of:\n- Context Release\n- Trace Release \n- Cascading cleanup (Context → Trace)\n- try-finally pattern\n- No-op object Release\n- Error handling with cleanup\n\nRun tests:\n```bash\ncd yao\ngo test -v ./agent/context -run Release\n```\n\n" }, { "path": "agent/context/authorized_test.go", "content": "package context_test\n\nimport (\n\tstdContext \"context\"\n\t\"testing\"\n\n\t\"github.com/stretchr/testify/assert\"\n\t\"github.com/yaoapp/yao/agent/context\"\n\t\"github.com/yaoapp/yao/config\"\n\t\"github.com/yaoapp/yao/openapi/oauth/types\"\n\t\"github.com/yaoapp/yao/test\"\n\t\"github.com/yaoapp/yao/trace\"\n)\n\nfunc TestContextNew_PreservesAuthorizedInfo(t *testing.T) {\n\ttest.Prepare(t, config.Conf)\n\tdefer test.Clean()\n\n\t// Create authorized info\n\tauthInfo := &types.AuthorizedInfo{\n\t\tUserID: \"716942074991\",\n\t\tTeamID: \"565955042879\",\n\t\tTenantID: \"tenant-001\",\n\t}\n\n\t// Create context using New()\n\tctx := context.New(stdContext.Background(), authInfo, \"test-chat-123\")\n\tdefer ctx.Release()\n\n\t// Verify authorized info is preserved\n\tassert.NotNil(t, ctx)\n\tassert.NotNil(t, ctx.Authorized)\n\tassert.Equal(t, \"716942074991\", ctx.Authorized.UserID)\n\tassert.Equal(t, \"565955042879\", ctx.Authorized.TeamID)\n\tassert.Equal(t, \"tenant-001\", ctx.Authorized.TenantID)\n\tassert.Equal(t, \"test-chat-123\", ctx.ChatID)\n}\n\nfunc TestContextTrace_SavesAuthorizedInfo(t *testing.T) {\n\ttest.Prepare(t, config.Conf)\n\tdefer test.Clean()\n\n\t// Create authorized info\n\tauthInfo := &types.AuthorizedInfo{\n\t\tUserID: \"716942074991\",\n\t\tTeamID: \"565955042879\",\n\t\tTenantID: \"tenant-001\",\n\t}\n\n\t// Create context using New\n\tctx := context.New(stdContext.Background(), authInfo, \"test-chat-456\")\n\tctx.AssistantID = \"test-assistant\"\n\tctx.Referer = context.RefererAPI\n\n\t// Initialize stack (required for trace)\n\tstack, _, done := context.EnterStack(ctx, \"test-assistant\", &context.Options{})\n\tctx.Stack = stack\n\tdefer done()\n\n\t// Initialize trace\n\tmanager, err := ctx.Trace()\n\tassert.NoError(t, err)\n\tassert.NotNil(t, manager)\n\n\t// Get trace info\n\tinfo, err := manager.GetTraceInfo()\n\tassert.NoError(t, err)\n\tassert.NotNil(t, info)\n\n\t// Verify auth info is saved in trace\n\tassert.Equal(t, \"716942074991\", info.CreatedBy)\n\tassert.Equal(t, \"565955042879\", info.TeamID)\n\tassert.Equal(t, \"tenant-001\", info.TenantID)\n\n\t// Clean up\n\tif ctx.Stack != nil && ctx.Stack.TraceID != \"\" {\n\t\ttrace.Release(ctx.Stack.TraceID)\n\t\ttrace.Remove(stdContext.Background(), trace.Local, ctx.Stack.TraceID)\n\t}\n}\n\nfunc TestContextNew_NilAuthorized(t *testing.T) {\n\ttest.Prepare(t, config.Conf)\n\tdefer test.Clean()\n\n\t// Create context with nil authorized info (should not panic)\n\tctx := context.New(stdContext.Background(), nil, \"test-chat-789\")\n\tdefer ctx.Release()\n\n\tassert.NotNil(t, ctx)\n\tassert.Nil(t, ctx.Authorized)\n\tassert.Equal(t, \"test-chat-789\", ctx.ChatID)\n}\n" }, { "path": "agent/context/buffer.go", "content": "package context\n\nimport (\n\t\"sync\"\n\t\"time\"\n\n\t\"github.com/google/uuid\"\n)\n\n// =============================================================================\n// Chat Buffer - Buffers messages and steps during execution for batch saving\n// =============================================================================\n\n// ChatBuffer buffers messages and resume steps during agent execution\n// All data is held in memory and batch-written at the end of Stream()\ntype ChatBuffer struct {\n\t// Identity\n\tchatID string\n\trequestID string\n\tassistantID string\n\tconnector string // Current connector ID (for data analysis)\n\tmode string // Current chat mode (chat or task)\n\n\t// Message buffer\n\tmessages []*BufferedMessage\n\tmsgSequence int\n\n\t// Step buffer (for Resume)\n\tsteps []*BufferedStep\n\tcurrentStep *BufferedStep\n\tstepSequence int\n\n\t// Space snapshot (captured when step starts, for recovery)\n\tspaceSnapshot map[string]interface{}\n\n\tmu sync.Mutex\n}\n\n// BufferedMessage represents a message waiting to be saved\ntype BufferedMessage struct {\n\tMessageID string `json:\"message_id\"`\n\tChatID string `json:\"chat_id\"`\n\tRequestID string `json:\"request_id,omitempty\"`\n\tRole string `json:\"role\"` // \"user\" or \"assistant\"\n\tType string `json:\"type\"` // \"text\", \"image\", \"loading\", \"tool_call\", \"retrieval\", etc.\n\tProps map[string]interface{} `json:\"props\"`\n\tBlockID string `json:\"block_id,omitempty\"`\n\tThreadID string `json:\"thread_id,omitempty\"`\n\tAssistantID string `json:\"assistant_id,omitempty\"`\n\tConnector string `json:\"connector,omitempty\"` // Connector ID used for this message\n\tMode string `json:\"mode,omitempty\"` // Chat mode used for this message (chat or task)\n\tSequence int `json:\"sequence\"`\n\tMetadata map[string]interface{} `json:\"metadata,omitempty\"`\n\tCreatedAt time.Time `json:\"created_at\"`\n\tIsStreaming bool `json:\"-\"` // Internal flag: true if message is still streaming (not saved until End)\n}\n\n// BufferedStep represents an execution step waiting to be saved (for Resume)\n// Only saved when request is interrupted or failed\ntype BufferedStep struct {\n\tResumeID string `json:\"resume_id\"`\n\tChatID string `json:\"chat_id\"`\n\tRequestID string `json:\"request_id\"`\n\tAssistantID string `json:\"assistant_id\"`\n\tStackID string `json:\"stack_id\"`\n\tStackParentID string `json:\"stack_parent_id,omitempty\"`\n\tStackDepth int `json:\"stack_depth\"`\n\tType string `json:\"type\"` // \"input\", \"hook_create\", \"llm\", \"tool\", \"hook_next\", \"delegate\"\n\tStatus string `json:\"status\"` // \"running\", \"completed\", \"failed\", \"interrupted\"\n\tInput map[string]interface{} `json:\"input,omitempty\"`\n\tOutput map[string]interface{} `json:\"output,omitempty\"`\n\tSpaceSnapshot map[string]interface{} `json:\"space_snapshot,omitempty\"`\n\tError string `json:\"error,omitempty\"`\n\tSequence int `json:\"sequence\"`\n\tMetadata map[string]interface{} `json:\"metadata,omitempty\"`\n\tCreatedAt time.Time `json:\"created_at\"`\n}\n\n// Step status constants (internal use only, not stored in database)\nconst (\n\tStepStatusRunning = \"running\"\n\tStepStatusCompleted = \"completed\"\n)\n\n// Step type constants\nconst (\n\tStepTypeInput = \"input\"\n\tStepTypeHookCreate = \"hook_create\"\n\tStepTypeLLM = \"llm\"\n\tStepTypeTool = \"tool\"\n\tStepTypeHookNext = \"hook_next\"\n\tStepTypeDelegate = \"delegate\"\n)\n\n// Resume status constants (for database storage)\nconst (\n\tResumeStatusFailed = \"failed\"\n\tResumeStatusInterrupted = \"interrupted\"\n)\n\n// NewChatBuffer creates a new chat buffer\nfunc NewChatBuffer(chatID, requestID, assistantID, connector, mode string) *ChatBuffer {\n\treturn &ChatBuffer{\n\t\tchatID: chatID,\n\t\trequestID: requestID,\n\t\tassistantID: assistantID,\n\t\tconnector: connector,\n\t\tmode: mode,\n\t\tmessages: make([]*BufferedMessage, 0),\n\t\tsteps: make([]*BufferedStep, 0),\n\t}\n}\n\n// =============================================================================\n// Message Buffer Methods\n// =============================================================================\n\n// AddMessage adds a message to the buffer\nfunc (b *ChatBuffer) AddMessage(msg *BufferedMessage) {\n\tif msg == nil {\n\t\treturn\n\t}\n\n\tb.mu.Lock()\n\tdefer b.mu.Unlock()\n\n\t// Auto-generate IDs if not provided\n\tif msg.MessageID == \"\" {\n\t\tmsg.MessageID = uuid.New().String()\n\t}\n\tif msg.ChatID == \"\" {\n\t\tmsg.ChatID = b.chatID\n\t}\n\tif msg.RequestID == \"\" {\n\t\tmsg.RequestID = b.requestID\n\t}\n\tif msg.CreatedAt.IsZero() {\n\t\tmsg.CreatedAt = time.Now()\n\t}\n\n\t// Set mode from buffer if not provided\n\tif msg.Mode == \"\" && b.mode != \"\" {\n\t\tmsg.Mode = b.mode\n\t}\n\n\t// Auto-increment sequence\n\tb.msgSequence++\n\tmsg.Sequence = b.msgSequence\n\n\tb.messages = append(b.messages, msg)\n}\n\n// AddUserInput adds user input message to the buffer\nfunc (b *ChatBuffer) AddUserInput(content interface{}, name string) {\n\tprops := map[string]interface{}{\n\t\t\"content\": content,\n\t\t\"role\": \"user\",\n\t}\n\tif name != \"\" {\n\t\tprops[\"name\"] = name\n\t}\n\n\tb.AddMessage(&BufferedMessage{\n\t\tRole: \"user\",\n\t\tType: \"user_input\",\n\t\tProps: props,\n\t})\n}\n\n// AddAssistantMessage adds an assistant message to the buffer\n// This is called by ctx.Send() to buffer messages for batch saving\nfunc (b *ChatBuffer) AddAssistantMessage(messageID, msgType string, props map[string]interface{}, blockID, threadID, assistantID string, metadata map[string]interface{}) {\n\t// Skip event type messages (transient, not stored)\n\tif msgType == \"event\" {\n\t\treturn\n\t}\n\n\tb.AddMessage(&BufferedMessage{\n\t\tMessageID: messageID, // Use the same MessageID as sent to client\n\t\tRole: \"assistant\",\n\t\tType: msgType,\n\t\tProps: props,\n\t\tBlockID: blockID,\n\t\tThreadID: threadID,\n\t\tAssistantID: assistantID,\n\t\tConnector: b.connector, // Use current connector\n\t\tMetadata: metadata,\n\t})\n}\n\n// AddStreamingMessage adds a streaming message to the buffer\n// Streaming messages are not saved until CompleteStreamingMessage is called\n// This is called by ctx.SendStream() to start a streaming message\nfunc (b *ChatBuffer) AddStreamingMessage(messageID, msgType string, props map[string]interface{}, blockID, threadID, assistantID string, metadata map[string]interface{}) {\n\t// Skip event type messages (transient, not stored)\n\tif msgType == \"event\" {\n\t\treturn\n\t}\n\n\t// Deep copy props to avoid mutation issues\n\tpropsCopy := make(map[string]interface{})\n\tfor k, v := range props {\n\t\tpropsCopy[k] = v\n\t}\n\n\tb.AddMessage(&BufferedMessage{\n\t\tMessageID: messageID, // Use provided message ID\n\t\tRole: \"assistant\",\n\t\tType: msgType,\n\t\tProps: propsCopy,\n\t\tBlockID: blockID,\n\t\tThreadID: threadID,\n\t\tAssistantID: assistantID,\n\t\tConnector: b.connector,\n\t\tMetadata: metadata,\n\t\tIsStreaming: true, // Mark as streaming\n\t})\n}\n\n// AppendMessageContent appends content to a streaming message\n// This is called by ctx.Append() to accumulate content\nfunc (b *ChatBuffer) AppendMessageContent(messageID string, content string) bool {\n\tb.mu.Lock()\n\tdefer b.mu.Unlock()\n\n\t// Find the message by ID\n\tfor _, msg := range b.messages {\n\t\tif msg.MessageID == messageID && msg.IsStreaming {\n\t\t\t// Append to existing content\n\t\t\tif msg.Props == nil {\n\t\t\t\tmsg.Props = make(map[string]interface{})\n\t\t\t}\n\t\t\tif existing, ok := msg.Props[\"content\"].(string); ok {\n\t\t\t\tmsg.Props[\"content\"] = existing + content\n\t\t\t} else {\n\t\t\t\tmsg.Props[\"content\"] = content\n\t\t\t}\n\t\t\treturn true\n\t\t}\n\t}\n\treturn false\n}\n\n// CompleteStreamingMessage marks a streaming message as complete\n// This is called by ctx.End() to finalize the message\n// Returns the complete content for the message_end event\nfunc (b *ChatBuffer) CompleteStreamingMessage(messageID string) (string, bool) {\n\tb.mu.Lock()\n\tdefer b.mu.Unlock()\n\n\t// Find the message by ID\n\tfor _, msg := range b.messages {\n\t\tif msg.MessageID == messageID && msg.IsStreaming {\n\t\t\tmsg.IsStreaming = false\n\t\t\t// Return the accumulated content\n\t\t\tif content, ok := msg.Props[\"content\"].(string); ok {\n\t\t\t\treturn content, true\n\t\t\t}\n\t\t\treturn \"\", true\n\t\t}\n\t}\n\treturn \"\", false\n}\n\n// GetStreamingMessage returns a streaming message by ID\nfunc (b *ChatBuffer) GetStreamingMessage(messageID string) *BufferedMessage {\n\tb.mu.Lock()\n\tdefer b.mu.Unlock()\n\n\tfor _, msg := range b.messages {\n\t\tif msg.MessageID == messageID && msg.IsStreaming {\n\t\t\treturn msg\n\t\t}\n\t}\n\treturn nil\n}\n\n// GetMessages returns all buffered messages\nfunc (b *ChatBuffer) GetMessages() []*BufferedMessage {\n\tb.mu.Lock()\n\tdefer b.mu.Unlock()\n\n\tresult := make([]*BufferedMessage, len(b.messages))\n\tcopy(result, b.messages)\n\treturn result\n}\n\n// GetMessageCount returns the number of buffered messages\nfunc (b *ChatBuffer) GetMessageCount() int {\n\tb.mu.Lock()\n\tdefer b.mu.Unlock()\n\treturn len(b.messages)\n}\n\n// =============================================================================\n// Step Buffer Methods (for Resume)\n// =============================================================================\n\n// BeginStep starts tracking a new execution step\n// Returns the step for further updates\nfunc (b *ChatBuffer) BeginStep(stepType string, input map[string]interface{}, stack *Stack) *BufferedStep {\n\tb.mu.Lock()\n\tdefer b.mu.Unlock()\n\n\tb.stepSequence++\n\n\tstep := &BufferedStep{\n\t\tResumeID: uuid.New().String(),\n\t\tChatID: b.chatID,\n\t\tRequestID: b.requestID,\n\t\tAssistantID: b.assistantID,\n\t\tType: stepType,\n\t\tStatus: StepStatusRunning,\n\t\tInput: input,\n\t\tSequence: b.stepSequence,\n\t\tCreatedAt: time.Now(),\n\t}\n\n\t// Set stack information if available\n\tif stack != nil {\n\t\tstep.StackID = stack.ID\n\t\tstep.StackParentID = stack.ParentID\n\t\tstep.StackDepth = stack.Depth\n\t}\n\n\t// Capture current space snapshot\n\tif b.spaceSnapshot != nil {\n\t\tstep.SpaceSnapshot = copyMap(b.spaceSnapshot)\n\t}\n\n\tb.steps = append(b.steps, step)\n\tb.currentStep = step\n\n\treturn step\n}\n\n// CompleteStep marks the current step as completed\nfunc (b *ChatBuffer) CompleteStep(output map[string]interface{}) {\n\tb.mu.Lock()\n\tdefer b.mu.Unlock()\n\n\tif b.currentStep != nil {\n\t\tb.currentStep.Output = output\n\t\tb.currentStep.Status = StepStatusCompleted\n\t\tb.currentStep = nil\n\t}\n}\n\n// FailCurrentStep marks the current step as failed or interrupted\nfunc (b *ChatBuffer) FailCurrentStep(status string, err error) {\n\tb.mu.Lock()\n\tdefer b.mu.Unlock()\n\n\tif b.currentStep != nil && b.currentStep.Status == StepStatusRunning {\n\t\tb.currentStep.Status = status\n\t\tif err != nil {\n\t\t\tb.currentStep.Error = err.Error()\n\t\t}\n\t}\n}\n\n// GetCurrentStep returns the current running step\nfunc (b *ChatBuffer) GetCurrentStep() *BufferedStep {\n\tb.mu.Lock()\n\tdefer b.mu.Unlock()\n\treturn b.currentStep\n}\n\n// GetStepsForResume returns steps that need to be saved for resume\n// Only returns steps with failed or interrupted status\nfunc (b *ChatBuffer) GetStepsForResume(finalStatus string) []*BufferedStep {\n\tb.mu.Lock()\n\tdefer b.mu.Unlock()\n\n\t// If completed successfully, no steps need to be saved\n\tif finalStatus == StepStatusCompleted {\n\t\treturn nil\n\t}\n\n\t// Mark current running step with final status\n\tif b.currentStep != nil && b.currentStep.Status == StepStatusRunning {\n\t\tb.currentStep.Status = finalStatus\n\t}\n\n\t// Return all steps (they will all have the context for recovery)\n\tresult := make([]*BufferedStep, len(b.steps))\n\tcopy(result, b.steps)\n\treturn result\n}\n\n// GetAllSteps returns all buffered steps (for debugging/testing)\nfunc (b *ChatBuffer) GetAllSteps() []*BufferedStep {\n\tb.mu.Lock()\n\tdefer b.mu.Unlock()\n\n\tresult := make([]*BufferedStep, len(b.steps))\n\tcopy(result, b.steps)\n\treturn result\n}\n\n// =============================================================================\n// Space Snapshot Methods\n// =============================================================================\n\n// SetSpaceSnapshot sets the space snapshot for recovery\n// Should be called when space data changes\nfunc (b *ChatBuffer) SetSpaceSnapshot(snapshot map[string]interface{}) {\n\tb.mu.Lock()\n\tdefer b.mu.Unlock()\n\tb.spaceSnapshot = copyMap(snapshot)\n}\n\n// GetSpaceSnapshot returns the current space snapshot\nfunc (b *ChatBuffer) GetSpaceSnapshot() map[string]interface{} {\n\tb.mu.Lock()\n\tdefer b.mu.Unlock()\n\treturn copyMap(b.spaceSnapshot)\n}\n\n// =============================================================================\n// Identity Methods\n// =============================================================================\n\n// ChatID returns the chat ID\nfunc (b *ChatBuffer) ChatID() string {\n\treturn b.chatID\n}\n\n// RequestID returns the request ID\nfunc (b *ChatBuffer) RequestID() string {\n\treturn b.requestID\n}\n\n// AssistantID returns the assistant ID\nfunc (b *ChatBuffer) AssistantID() string {\n\treturn b.assistantID\n}\n\n// SetAssistantID updates the assistant ID (for A2A calls)\nfunc (b *ChatBuffer) SetAssistantID(assistantID string) {\n\tb.mu.Lock()\n\tdefer b.mu.Unlock()\n\tb.assistantID = assistantID\n}\n\n// Connector returns the current connector ID\nfunc (b *ChatBuffer) Connector() string {\n\treturn b.connector\n}\n\n// SetConnector updates the connector ID (when user switches connector)\nfunc (b *ChatBuffer) SetConnector(connector string) {\n\tb.mu.Lock()\n\tdefer b.mu.Unlock()\n\tb.connector = connector\n}\n\n// Mode returns the current chat mode\nfunc (b *ChatBuffer) Mode() string {\n\treturn b.mode\n}\n\n// SetMode updates the chat mode (when user switches mode)\nfunc (b *ChatBuffer) SetMode(mode string) {\n\tb.mu.Lock()\n\tdefer b.mu.Unlock()\n\tb.mode = mode\n}\n\n// =============================================================================\n// Helper Functions\n// =============================================================================\n\n// copyMap creates a shallow copy of a map\nfunc copyMap(src map[string]interface{}) map[string]interface{} {\n\tif src == nil {\n\t\treturn nil\n\t}\n\tdst := make(map[string]interface{}, len(src))\n\tfor k, v := range src {\n\t\tdst[k] = v\n\t}\n\treturn dst\n}\n" }, { "path": "agent/context/buffer_test.go", "content": "package context_test\n\nimport (\n\t\"fmt\"\n\t\"sync\"\n\t\"testing\"\n\t\"time\"\n\n\t\"github.com/stretchr/testify/assert\"\n\t\"github.com/stretchr/testify/require\"\n\t\"github.com/yaoapp/yao/agent/context\"\n)\n\n// =============================================================================\n// ChatBuffer Creation Tests\n// =============================================================================\n\nfunc TestBufferNewChatBuffer(t *testing.T) {\n\tt.Run(\"CreateWithAllFields\", func(t *testing.T) {\n\t\tbuffer := context.NewChatBuffer(\"chat-123\", \"req-456\", \"assistant-789\", \"\", \"\")\n\n\t\tassert.NotNil(t, buffer)\n\t\tassert.Equal(t, \"chat-123\", buffer.ChatID())\n\t\tassert.Equal(t, \"req-456\", buffer.RequestID())\n\t\tassert.Equal(t, \"assistant-789\", buffer.AssistantID())\n\t\tassert.Empty(t, buffer.GetMessages())\n\t\tassert.Empty(t, buffer.GetAllSteps())\n\t\tassert.Equal(t, 0, buffer.GetMessageCount())\n\t})\n\n\tt.Run(\"CreateWithEmptyFields\", func(t *testing.T) {\n\t\tbuffer := context.NewChatBuffer(\"\", \"\", \"\", \"\", \"\")\n\n\t\tassert.NotNil(t, buffer)\n\t\tassert.Empty(t, buffer.ChatID())\n\t\tassert.Empty(t, buffer.RequestID())\n\t\tassert.Empty(t, buffer.AssistantID())\n\t})\n}\n\n// =============================================================================\n// Message Buffer Tests\n// =============================================================================\n\nfunc TestBufferAddMessage(t *testing.T) {\n\tt.Run(\"AddSingleMessage\", func(t *testing.T) {\n\t\tbuffer := context.NewChatBuffer(\"chat-1\", \"req-1\", \"assistant-1\", \"\", \"\")\n\n\t\tmsg := &context.BufferedMessage{\n\t\t\tRole: \"assistant\",\n\t\t\tType: \"text\",\n\t\t\tProps: map[string]interface{}{\"content\": \"Hello\"},\n\t\t}\n\t\tbuffer.AddMessage(msg)\n\n\t\tmessages := buffer.GetMessages()\n\t\trequire.Len(t, messages, 1)\n\t\tassert.Equal(t, \"assistant\", messages[0].Role)\n\t\tassert.Equal(t, \"text\", messages[0].Type)\n\t\tassert.Equal(t, 1, messages[0].Sequence)\n\t\tassert.NotEmpty(t, messages[0].MessageID) // Auto-generated\n\t\tassert.Equal(t, \"chat-1\", messages[0].ChatID)\n\t\tassert.Equal(t, \"req-1\", messages[0].RequestID)\n\t\tassert.False(t, messages[0].CreatedAt.IsZero())\n\t})\n\n\tt.Run(\"AddMultipleMessages\", func(t *testing.T) {\n\t\tbuffer := context.NewChatBuffer(\"chat-2\", \"req-2\", \"assistant-2\", \"\", \"\")\n\n\t\tfor i := 0; i < 5; i++ {\n\t\t\tbuffer.AddMessage(&context.BufferedMessage{\n\t\t\t\tRole: \"assistant\",\n\t\t\t\tType: \"text\",\n\t\t\t\tProps: map[string]interface{}{\"content\": fmt.Sprintf(\"Message %d\", i+1)},\n\t\t\t})\n\t\t}\n\n\t\tmessages := buffer.GetMessages()\n\t\trequire.Len(t, messages, 5)\n\n\t\t// Verify sequence numbers\n\t\tfor i, msg := range messages {\n\t\t\tassert.Equal(t, i+1, msg.Sequence)\n\t\t}\n\t})\n\n\tt.Run(\"AddNilMessage\", func(t *testing.T) {\n\t\tbuffer := context.NewChatBuffer(\"chat-3\", \"req-3\", \"assistant-3\", \"\", \"\")\n\t\tbuffer.AddMessage(nil)\n\n\t\tassert.Equal(t, 0, buffer.GetMessageCount())\n\t})\n\n\tt.Run(\"AddMessageWithExistingID\", func(t *testing.T) {\n\t\tbuffer := context.NewChatBuffer(\"chat-4\", \"req-4\", \"assistant-4\", \"\", \"\")\n\n\t\tmsg := &context.BufferedMessage{\n\t\t\tMessageID: \"custom-id-123\",\n\t\t\tRole: \"assistant\",\n\t\t\tType: \"text\",\n\t\t}\n\t\tbuffer.AddMessage(msg)\n\n\t\tmessages := buffer.GetMessages()\n\t\trequire.Len(t, messages, 1)\n\t\tassert.Equal(t, \"custom-id-123\", messages[0].MessageID) // Preserved\n\t})\n\n\tt.Run(\"AddMessageWithExistingTimestamp\", func(t *testing.T) {\n\t\tbuffer := context.NewChatBuffer(\"chat-5\", \"req-5\", \"assistant-5\", \"\", \"\")\n\n\t\tcustomTime := time.Date(2024, 1, 1, 12, 0, 0, 0, time.UTC)\n\t\tmsg := &context.BufferedMessage{\n\t\t\tRole: \"assistant\",\n\t\t\tType: \"text\",\n\t\t\tCreatedAt: customTime,\n\t\t}\n\t\tbuffer.AddMessage(msg)\n\n\t\tmessages := buffer.GetMessages()\n\t\trequire.Len(t, messages, 1)\n\t\tassert.Equal(t, customTime, messages[0].CreatedAt) // Preserved\n\t})\n}\n\nfunc TestBufferAddUserInput(t *testing.T) {\n\tt.Run(\"AddStringContent\", func(t *testing.T) {\n\t\tbuffer := context.NewChatBuffer(\"chat-1\", \"req-1\", \"assistant-1\", \"\", \"\")\n\t\tbuffer.AddUserInput(\"What is the weather?\", \"\")\n\n\t\tmessages := buffer.GetMessages()\n\t\trequire.Len(t, messages, 1)\n\t\tassert.Equal(t, \"user\", messages[0].Role)\n\t\tassert.Equal(t, \"user_input\", messages[0].Type)\n\t\tassert.Equal(t, \"What is the weather?\", messages[0].Props[\"content\"])\n\t\tassert.Equal(t, \"user\", messages[0].Props[\"role\"])\n\t})\n\n\tt.Run(\"AddUserInputWithName\", func(t *testing.T) {\n\t\tbuffer := context.NewChatBuffer(\"chat-2\", \"req-2\", \"assistant-2\", \"\", \"\")\n\t\tbuffer.AddUserInput(\"Hello\", \"John\")\n\n\t\tmessages := buffer.GetMessages()\n\t\trequire.Len(t, messages, 1)\n\t\tassert.Equal(t, \"John\", messages[0].Props[\"name\"])\n\t})\n\n\tt.Run(\"AddComplexContent\", func(t *testing.T) {\n\t\tbuffer := context.NewChatBuffer(\"chat-3\", \"req-3\", \"assistant-3\", \"\", \"\")\n\t\tcomplexContent := []map[string]interface{}{\n\t\t\t{\"type\": \"text\", \"text\": \"Look at this image\"},\n\t\t\t{\"type\": \"image_url\", \"image_url\": map[string]string{\"url\": \"https://example.com/image.jpg\"}},\n\t\t}\n\t\tbuffer.AddUserInput(complexContent, \"\")\n\n\t\tmessages := buffer.GetMessages()\n\t\trequire.Len(t, messages, 1)\n\t\tcontent, ok := messages[0].Props[\"content\"].([]map[string]interface{})\n\t\trequire.True(t, ok)\n\t\tassert.Len(t, content, 2)\n\t})\n}\n\nfunc TestBufferAddAssistantMessage(t *testing.T) {\n\tt.Run(\"AddTextMessage\", func(t *testing.T) {\n\t\tbuffer := context.NewChatBuffer(\"chat-1\", \"req-1\", \"assistant-1\", \"\", \"\")\n\t\tbuffer.AddAssistantMessage(\n\t\t\t\"M1\",\n\t\t\t\"text\",\n\t\t\tmap[string]interface{}{\"content\": \"Hello, how can I help?\"},\n\t\t\t\"block-1\",\n\t\t\t\"thread-1\",\n\t\t\t\"assistant-1\",\n\t\t\tmap[string]interface{}{\"model\": \"gpt-4\"},\n\t\t)\n\n\t\tmessages := buffer.GetMessages()\n\t\trequire.Len(t, messages, 1)\n\t\tassert.Equal(t, \"M1\", messages[0].MessageID)\n\t\tassert.Equal(t, \"assistant\", messages[0].Role)\n\t\tassert.Equal(t, \"text\", messages[0].Type)\n\t\tassert.Equal(t, \"block-1\", messages[0].BlockID)\n\t\tassert.Equal(t, \"thread-1\", messages[0].ThreadID)\n\t\tassert.Equal(t, \"assistant-1\", messages[0].AssistantID)\n\t\tassert.Equal(t, \"gpt-4\", messages[0].Metadata[\"model\"])\n\t})\n\n\tt.Run(\"SkipEventMessage\", func(t *testing.T) {\n\t\tbuffer := context.NewChatBuffer(\"chat-2\", \"req-2\", \"assistant-2\", \"\", \"\")\n\t\tbuffer.AddAssistantMessage(\n\t\t\t\"E1\",\n\t\t\t\"event\",\n\t\t\tmap[string]interface{}{\"event\": \"message_start\"},\n\t\t\t\"\", \"\", \"\", nil,\n\t\t)\n\n\t\t// Event messages should be skipped\n\t\tassert.Equal(t, 0, buffer.GetMessageCount())\n\t})\n\n\tt.Run(\"AddRetrievalMessage\", func(t *testing.T) {\n\t\tbuffer := context.NewChatBuffer(\"chat-3\", \"req-3\", \"assistant-3\", \"\", \"\")\n\t\tbuffer.AddAssistantMessage(\n\t\t\t\"M2\",\n\t\t\t\"retrieval\",\n\t\t\tmap[string]interface{}{\n\t\t\t\t\"sources\": []map[string]interface{}{\n\t\t\t\t\t{\"title\": \"Doc 1\", \"score\": 0.95},\n\t\t\t\t\t{\"title\": \"Doc 2\", \"score\": 0.87},\n\t\t\t\t},\n\t\t\t},\n\t\t\t\"block-1\", \"\", \"assistant-3\", nil,\n\t\t)\n\n\t\tmessages := buffer.GetMessages()\n\t\trequire.Len(t, messages, 1)\n\t\tassert.Equal(t, \"retrieval\", messages[0].Type)\n\t})\n\n\tt.Run(\"AddToolCallMessage\", func(t *testing.T) {\n\t\tbuffer := context.NewChatBuffer(\"chat-4\", \"req-4\", \"assistant-4\", \"\", \"\")\n\t\tbuffer.AddAssistantMessage(\n\t\t\t\"M3\",\n\t\t\t\"tool_call\",\n\t\t\tmap[string]interface{}{\n\t\t\t\t\"name\": \"get_weather\",\n\t\t\t\t\"arguments\": `{\"location\": \"San Francisco\"}`,\n\t\t\t},\n\t\t\t\"block-1\", \"\", \"assistant-4\", nil,\n\t\t)\n\n\t\tmessages := buffer.GetMessages()\n\t\trequire.Len(t, messages, 1)\n\t\tassert.Equal(t, \"tool_call\", messages[0].Type)\n\t\tassert.Equal(t, \"get_weather\", messages[0].Props[\"name\"])\n\t})\n\n\tt.Run(\"AddCustomTypeMessage\", func(t *testing.T) {\n\t\tbuffer := context.NewChatBuffer(\"chat-5\", \"req-5\", \"assistant-5\", \"\", \"\")\n\t\tbuffer.AddAssistantMessage(\n\t\t\t\"M4\",\n\t\t\t\"custom_chart\",\n\t\t\tmap[string]interface{}{\n\t\t\t\t\"chart_type\": \"bar\",\n\t\t\t\t\"data\": []int{1, 2, 3, 4, 5},\n\t\t\t},\n\t\t\t\"block-1\", \"\", \"assistant-5\", nil,\n\t\t)\n\n\t\tmessages := buffer.GetMessages()\n\t\trequire.Len(t, messages, 1)\n\t\tassert.Equal(t, \"custom_chart\", messages[0].Type)\n\t})\n}\n\nfunc TestBufferGetMessages(t *testing.T) {\n\tt.Run(\"GetMessagesReturnsSliceCopy\", func(t *testing.T) {\n\t\tbuffer := context.NewChatBuffer(\"chat-1\", \"req-1\", \"assistant-1\", \"\", \"\")\n\t\tbuffer.AddUserInput(\"Hello\", \"\")\n\n\t\tmessages1 := buffer.GetMessages()\n\t\tmessages2 := buffer.GetMessages()\n\n\t\t// Slices should be different (copy of slice)\n\t\t// But pointers point to same underlying objects (shallow copy)\n\t\tassert.Len(t, messages1, 1)\n\t\tassert.Len(t, messages2, 1)\n\t})\n\n\tt.Run(\"GetEmptyMessages\", func(t *testing.T) {\n\t\tbuffer := context.NewChatBuffer(\"chat-2\", \"req-2\", \"assistant-2\", \"\", \"\")\n\t\tmessages := buffer.GetMessages()\n\n\t\tassert.NotNil(t, messages)\n\t\tassert.Empty(t, messages)\n\t})\n}\n\nfunc TestBufferGetMessageCount(t *testing.T) {\n\tbuffer := context.NewChatBuffer(\"chat-1\", \"req-1\", \"assistant-1\", \"\", \"\")\n\tassert.Equal(t, 0, buffer.GetMessageCount())\n\n\tbuffer.AddUserInput(\"Message 1\", \"\")\n\tassert.Equal(t, 1, buffer.GetMessageCount())\n\n\tbuffer.AddAssistantMessage(\"M1\", \"text\", map[string]interface{}{\"content\": \"Reply\"}, \"\", \"\", \"\", nil)\n\tassert.Equal(t, 2, buffer.GetMessageCount())\n}\n\n// =============================================================================\n// Step Buffer Tests (for Resume)\n// =============================================================================\n\nfunc TestBufferBeginStep(t *testing.T) {\n\tt.Run(\"BeginStepWithStack\", func(t *testing.T) {\n\t\tbuffer := context.NewChatBuffer(\"chat-1\", \"req-1\", \"assistant-1\", \"\", \"\")\n\n\t\tstack := &context.Stack{\n\t\t\tID: \"stack-123\",\n\t\t\tParentID: \"stack-parent-456\",\n\t\t\tDepth: 2,\n\t\t}\n\n\t\tstep := buffer.BeginStep(context.StepTypeLLM, map[string]interface{}{\"prompt\": \"Hello\"}, stack)\n\n\t\trequire.NotNil(t, step)\n\t\tassert.NotEmpty(t, step.ResumeID)\n\t\tassert.Equal(t, \"chat-1\", step.ChatID)\n\t\tassert.Equal(t, \"req-1\", step.RequestID)\n\t\tassert.Equal(t, \"assistant-1\", step.AssistantID)\n\t\tassert.Equal(t, \"stack-123\", step.StackID)\n\t\tassert.Equal(t, \"stack-parent-456\", step.StackParentID)\n\t\tassert.Equal(t, 2, step.StackDepth)\n\t\tassert.Equal(t, context.StepTypeLLM, step.Type)\n\t\tassert.Equal(t, context.StepStatusRunning, step.Status)\n\t\tassert.Equal(t, 1, step.Sequence)\n\t\tassert.Equal(t, \"Hello\", step.Input[\"prompt\"])\n\t\tassert.False(t, step.CreatedAt.IsZero())\n\t})\n\n\tt.Run(\"BeginStepWithNilStack\", func(t *testing.T) {\n\t\tbuffer := context.NewChatBuffer(\"chat-2\", \"req-2\", \"assistant-2\", \"\", \"\")\n\n\t\tstep := buffer.BeginStep(context.StepTypeInput, nil, nil)\n\n\t\trequire.NotNil(t, step)\n\t\tassert.Empty(t, step.StackID)\n\t\tassert.Empty(t, step.StackParentID)\n\t\tassert.Equal(t, 0, step.StackDepth)\n\t})\n\n\tt.Run(\"BeginMultipleSteps\", func(t *testing.T) {\n\t\tbuffer := context.NewChatBuffer(\"chat-3\", \"req-3\", \"assistant-3\", \"\", \"\")\n\n\t\tstep1 := buffer.BeginStep(context.StepTypeInput, nil, nil)\n\t\tstep2 := buffer.BeginStep(context.StepTypeHookCreate, nil, nil)\n\t\tstep3 := buffer.BeginStep(context.StepTypeLLM, nil, nil)\n\n\t\tassert.Equal(t, 1, step1.Sequence)\n\t\tassert.Equal(t, 2, step2.Sequence)\n\t\tassert.Equal(t, 3, step3.Sequence)\n\n\t\tsteps := buffer.GetAllSteps()\n\t\trequire.Len(t, steps, 3)\n\t})\n\n\tt.Run(\"BeginStepWithSpaceSnapshot\", func(t *testing.T) {\n\t\tbuffer := context.NewChatBuffer(\"chat-4\", \"req-4\", \"assistant-4\", \"\", \"\")\n\n\t\t// Set space snapshot before beginning step\n\t\tbuffer.SetSpaceSnapshot(map[string]interface{}{\n\t\t\t\"key1\": \"value1\",\n\t\t\t\"key2\": 42,\n\t\t})\n\n\t\tstep := buffer.BeginStep(context.StepTypeLLM, nil, nil)\n\n\t\trequire.NotNil(t, step.SpaceSnapshot)\n\t\tassert.Equal(t, \"value1\", step.SpaceSnapshot[\"key1\"])\n\t\tassert.Equal(t, 42, step.SpaceSnapshot[\"key2\"])\n\t})\n}\n\nfunc TestBufferCompleteStep(t *testing.T) {\n\tt.Run(\"CompleteCurrentStep\", func(t *testing.T) {\n\t\tbuffer := context.NewChatBuffer(\"chat-1\", \"req-1\", \"assistant-1\", \"\", \"\")\n\n\t\tbuffer.BeginStep(context.StepTypeLLM, map[string]interface{}{\"prompt\": \"Hello\"}, nil)\n\t\tbuffer.CompleteStep(map[string]interface{}{\"response\": \"Hi there!\"})\n\n\t\tsteps := buffer.GetAllSteps()\n\t\trequire.Len(t, steps, 1)\n\t\tassert.Equal(t, context.StepStatusCompleted, steps[0].Status)\n\t\tassert.Equal(t, \"Hi there!\", steps[0].Output[\"response\"])\n\t\tassert.Nil(t, buffer.GetCurrentStep()) // Current step cleared\n\t})\n\n\tt.Run(\"CompleteWithNoCurrentStep\", func(t *testing.T) {\n\t\tbuffer := context.NewChatBuffer(\"chat-2\", \"req-2\", \"assistant-2\", \"\", \"\")\n\n\t\t// Should not panic\n\t\tbuffer.CompleteStep(map[string]interface{}{\"response\": \"test\"})\n\t\tassert.Nil(t, buffer.GetCurrentStep())\n\t})\n\n\tt.Run(\"CompleteMultipleStepsSequentially\", func(t *testing.T) {\n\t\tbuffer := context.NewChatBuffer(\"chat-3\", \"req-3\", \"assistant-3\", \"\", \"\")\n\n\t\tbuffer.BeginStep(context.StepTypeInput, nil, nil)\n\t\tbuffer.CompleteStep(map[string]interface{}{\"done\": true})\n\n\t\tbuffer.BeginStep(context.StepTypeHookCreate, nil, nil)\n\t\tbuffer.CompleteStep(map[string]interface{}{\"hook_result\": \"ok\"})\n\n\t\tbuffer.BeginStep(context.StepTypeLLM, nil, nil)\n\t\tbuffer.CompleteStep(map[string]interface{}{\"llm_response\": \"hello\"})\n\n\t\tsteps := buffer.GetAllSteps()\n\t\trequire.Len(t, steps, 3)\n\t\tfor _, step := range steps {\n\t\t\tassert.Equal(t, context.StepStatusCompleted, step.Status)\n\t\t}\n\t})\n}\n\nfunc TestBufferFailCurrentStep(t *testing.T) {\n\tt.Run(\"FailWithError\", func(t *testing.T) {\n\t\tbuffer := context.NewChatBuffer(\"chat-1\", \"req-1\", \"assistant-1\", \"\", \"\")\n\n\t\tbuffer.BeginStep(context.StepTypeLLM, nil, nil)\n\t\tbuffer.FailCurrentStep(context.ResumeStatusFailed, fmt.Errorf(\"API error: rate limit exceeded\"))\n\n\t\tsteps := buffer.GetAllSteps()\n\t\trequire.Len(t, steps, 1)\n\t\tassert.Equal(t, context.ResumeStatusFailed, steps[0].Status)\n\t\tassert.Equal(t, \"API error: rate limit exceeded\", steps[0].Error)\n\t})\n\n\tt.Run(\"FailWithInterrupted\", func(t *testing.T) {\n\t\tbuffer := context.NewChatBuffer(\"chat-2\", \"req-2\", \"assistant-2\", \"\", \"\")\n\n\t\tbuffer.BeginStep(context.StepTypeLLM, nil, nil)\n\t\tbuffer.FailCurrentStep(context.ResumeStatusInterrupted, nil)\n\n\t\tsteps := buffer.GetAllSteps()\n\t\trequire.Len(t, steps, 1)\n\t\tassert.Equal(t, context.ResumeStatusInterrupted, steps[0].Status)\n\t\tassert.Empty(t, steps[0].Error)\n\t})\n\n\tt.Run(\"FailAlreadyCompletedStep\", func(t *testing.T) {\n\t\tbuffer := context.NewChatBuffer(\"chat-3\", \"req-3\", \"assistant-3\", \"\", \"\")\n\n\t\tbuffer.BeginStep(context.StepTypeLLM, nil, nil)\n\t\tbuffer.CompleteStep(map[string]interface{}{\"done\": true})\n\n\t\t// Try to fail completed step (should be no-op since currentStep is nil)\n\t\tbuffer.FailCurrentStep(context.ResumeStatusFailed, fmt.Errorf(\"late error\"))\n\n\t\tsteps := buffer.GetAllSteps()\n\t\trequire.Len(t, steps, 1)\n\t\tassert.Equal(t, context.StepStatusCompleted, steps[0].Status) // Still completed\n\t})\n\n\tt.Run(\"FailWithNoCurrentStep\", func(t *testing.T) {\n\t\tbuffer := context.NewChatBuffer(\"chat-4\", \"req-4\", \"assistant-4\", \"\", \"\")\n\n\t\t// Should not panic\n\t\tbuffer.FailCurrentStep(context.ResumeStatusFailed, fmt.Errorf(\"error\"))\n\t})\n}\n\nfunc TestBufferGetCurrentStep(t *testing.T) {\n\tt.Run(\"NoCurrentStep\", func(t *testing.T) {\n\t\tbuffer := context.NewChatBuffer(\"chat-1\", \"req-1\", \"assistant-1\", \"\", \"\")\n\t\tassert.Nil(t, buffer.GetCurrentStep())\n\t})\n\n\tt.Run(\"HasCurrentStep\", func(t *testing.T) {\n\t\tbuffer := context.NewChatBuffer(\"chat-2\", \"req-2\", \"assistant-2\", \"\", \"\")\n\t\tbuffer.BeginStep(context.StepTypeLLM, nil, nil)\n\n\t\tcurrent := buffer.GetCurrentStep()\n\t\trequire.NotNil(t, current)\n\t\tassert.Equal(t, context.StepTypeLLM, current.Type)\n\t})\n\n\tt.Run(\"CurrentStepClearedAfterComplete\", func(t *testing.T) {\n\t\tbuffer := context.NewChatBuffer(\"chat-3\", \"req-3\", \"assistant-3\", \"\", \"\")\n\t\tbuffer.BeginStep(context.StepTypeLLM, nil, nil)\n\t\tbuffer.CompleteStep(nil)\n\n\t\tassert.Nil(t, buffer.GetCurrentStep())\n\t})\n}\n\nfunc TestBufferGetStepsForResume(t *testing.T) {\n\tt.Run(\"CompletedSuccessfully\", func(t *testing.T) {\n\t\tbuffer := context.NewChatBuffer(\"chat-1\", \"req-1\", \"assistant-1\", \"\", \"\")\n\n\t\tbuffer.BeginStep(context.StepTypeInput, nil, nil)\n\t\tbuffer.CompleteStep(nil)\n\t\tbuffer.BeginStep(context.StepTypeLLM, nil, nil)\n\t\tbuffer.CompleteStep(nil)\n\n\t\t// Completed successfully - no steps need to be saved\n\t\tsteps := buffer.GetStepsForResume(context.StepStatusCompleted)\n\t\tassert.Nil(t, steps)\n\t})\n\n\tt.Run(\"FailedRequest\", func(t *testing.T) {\n\t\tbuffer := context.NewChatBuffer(\"chat-2\", \"req-2\", \"assistant-2\", \"\", \"\")\n\n\t\tbuffer.BeginStep(context.StepTypeInput, nil, nil)\n\t\tbuffer.CompleteStep(nil)\n\t\tbuffer.BeginStep(context.StepTypeLLM, nil, nil)\n\t\t// Step still running when failure occurs\n\n\t\tsteps := buffer.GetStepsForResume(context.ResumeStatusFailed)\n\t\trequire.NotNil(t, steps)\n\t\tassert.Len(t, steps, 2)\n\n\t\t// Current step should be marked as failed\n\t\tassert.Equal(t, context.ResumeStatusFailed, steps[1].Status)\n\t})\n\n\tt.Run(\"InterruptedRequest\", func(t *testing.T) {\n\t\tbuffer := context.NewChatBuffer(\"chat-3\", \"req-3\", \"assistant-3\", \"\", \"\")\n\n\t\tbuffer.BeginStep(context.StepTypeInput, nil, nil)\n\t\tbuffer.CompleteStep(nil)\n\t\tbuffer.BeginStep(context.StepTypeHookCreate, nil, nil)\n\t\tbuffer.CompleteStep(nil)\n\t\tbuffer.BeginStep(context.StepTypeLLM, nil, nil)\n\t\t// Interrupted during LLM\n\n\t\tsteps := buffer.GetStepsForResume(context.ResumeStatusInterrupted)\n\t\trequire.NotNil(t, steps)\n\t\tassert.Len(t, steps, 3)\n\t\tassert.Equal(t, context.ResumeStatusInterrupted, steps[2].Status)\n\t})\n}\n\nfunc TestBufferGetAllSteps(t *testing.T) {\n\tt.Run(\"GetStepsReturnsSliceCopy\", func(t *testing.T) {\n\t\tbuffer := context.NewChatBuffer(\"chat-1\", \"req-1\", \"assistant-1\", \"\", \"\")\n\t\tbuffer.BeginStep(context.StepTypeLLM, nil, nil)\n\n\t\tsteps1 := buffer.GetAllSteps()\n\t\tsteps2 := buffer.GetAllSteps()\n\n\t\t// Slices should be different (copy of slice)\n\t\tassert.Len(t, steps1, 1)\n\t\tassert.Len(t, steps2, 1)\n\t})\n\n\tt.Run(\"GetEmptySteps\", func(t *testing.T) {\n\t\tbuffer := context.NewChatBuffer(\"chat-2\", \"req-2\", \"assistant-2\", \"\", \"\")\n\t\tsteps := buffer.GetAllSteps()\n\n\t\tassert.NotNil(t, steps)\n\t\tassert.Empty(t, steps)\n\t})\n}\n\n// =============================================================================\n// Space Snapshot Tests\n// =============================================================================\n\nfunc TestBufferSpaceSnapshot(t *testing.T) {\n\tt.Run(\"SetAndGetSnapshot\", func(t *testing.T) {\n\t\tbuffer := context.NewChatBuffer(\"chat-1\", \"req-1\", \"assistant-1\", \"\", \"\")\n\n\t\tsnapshot := map[string]interface{}{\n\t\t\t\"user_id\": \"user-123\",\n\t\t\t\"session\": map[string]interface{}{\"token\": \"abc\"},\n\t\t\t\"counter\": 42,\n\t\t\t\"is_active\": true,\n\t\t}\n\t\tbuffer.SetSpaceSnapshot(snapshot)\n\n\t\tretrieved := buffer.GetSpaceSnapshot()\n\t\tassert.Equal(t, \"user-123\", retrieved[\"user_id\"])\n\t\tassert.Equal(t, 42, retrieved[\"counter\"])\n\t\tassert.Equal(t, true, retrieved[\"is_active\"])\n\t})\n\n\tt.Run(\"SnapshotIsCopy\", func(t *testing.T) {\n\t\tbuffer := context.NewChatBuffer(\"chat-2\", \"req-2\", \"assistant-2\", \"\", \"\")\n\n\t\toriginal := map[string]interface{}{\"key\": \"original\"}\n\t\tbuffer.SetSpaceSnapshot(original)\n\n\t\t// Modify original\n\t\toriginal[\"key\"] = \"modified\"\n\n\t\t// Buffer should have original value\n\t\tretrieved := buffer.GetSpaceSnapshot()\n\t\tassert.Equal(t, \"original\", retrieved[\"key\"])\n\t})\n\n\tt.Run(\"GetSnapshotReturnsCopy\", func(t *testing.T) {\n\t\tbuffer := context.NewChatBuffer(\"chat-3\", \"req-3\", \"assistant-3\", \"\", \"\")\n\t\tbuffer.SetSpaceSnapshot(map[string]interface{}{\"key\": \"value\"})\n\n\t\tretrieved1 := buffer.GetSpaceSnapshot()\n\t\tretrieved1[\"key\"] = \"modified\"\n\n\t\tretrieved2 := buffer.GetSpaceSnapshot()\n\t\tassert.Equal(t, \"value\", retrieved2[\"key\"]) // Original unchanged\n\t})\n\n\tt.Run(\"GetNilSnapshot\", func(t *testing.T) {\n\t\tbuffer := context.NewChatBuffer(\"chat-4\", \"req-4\", \"assistant-4\", \"\", \"\")\n\t\tsnapshot := buffer.GetSpaceSnapshot()\n\t\tassert.Nil(t, snapshot)\n\t})\n\n\tt.Run(\"SetNilSnapshot\", func(t *testing.T) {\n\t\tbuffer := context.NewChatBuffer(\"chat-5\", \"req-5\", \"assistant-5\", \"\", \"\")\n\t\tbuffer.SetSpaceSnapshot(map[string]interface{}{\"key\": \"value\"})\n\t\tbuffer.SetSpaceSnapshot(nil)\n\n\t\tsnapshot := buffer.GetSpaceSnapshot()\n\t\tassert.Nil(t, snapshot)\n\t})\n}\n\n// =============================================================================\n// Identity Methods Tests\n// =============================================================================\n\nfunc TestBufferIdentityMethods(t *testing.T) {\n\tt.Run(\"SetAssistantID\", func(t *testing.T) {\n\t\tbuffer := context.NewChatBuffer(\"chat-1\", \"req-1\", \"assistant-original\", \"\", \"\")\n\n\t\tassert.Equal(t, \"assistant-original\", buffer.AssistantID())\n\n\t\tbuffer.SetAssistantID(\"assistant-new\")\n\t\tassert.Equal(t, \"assistant-new\", buffer.AssistantID())\n\t})\n\n\tt.Run(\"ChatID\", func(t *testing.T) {\n\t\tbuffer := context.NewChatBuffer(\"chat-test\", \"req-test\", \"assistant-test\", \"\", \"\")\n\t\tassert.Equal(t, \"chat-test\", buffer.ChatID())\n\t})\n\n\tt.Run(\"RequestID\", func(t *testing.T) {\n\t\tbuffer := context.NewChatBuffer(\"chat-test\", \"req-test\", \"assistant-test\", \"\", \"\")\n\t\tassert.Equal(t, \"req-test\", buffer.RequestID())\n\t})\n\n\tt.Run(\"Connector\", func(t *testing.T) {\n\t\tbuffer := context.NewChatBuffer(\"chat-test\", \"req-test\", \"assistant-test\", \"openai\", \"\")\n\t\tassert.Equal(t, \"openai\", buffer.Connector())\n\t})\n\n\tt.Run(\"SetConnector\", func(t *testing.T) {\n\t\tbuffer := context.NewChatBuffer(\"chat-1\", \"req-1\", \"assistant-1\", \"openai\", \"\")\n\t\tassert.Equal(t, \"openai\", buffer.Connector())\n\n\t\t// Simulate user switching connector mid-conversation\n\t\tbuffer.SetConnector(\"anthropic\")\n\t\tassert.Equal(t, \"anthropic\", buffer.Connector())\n\t})\n\n\tt.Run(\"EmptyConnector\", func(t *testing.T) {\n\t\tbuffer := context.NewChatBuffer(\"chat-test\", \"req-test\", \"assistant-test\", \"\", \"\")\n\t\tassert.Equal(t, \"\", buffer.Connector())\n\t})\n}\n\nfunc TestBufferConnectorInMessages(t *testing.T) {\n\tt.Run(\"MessageInheritsConnector\", func(t *testing.T) {\n\t\tbuffer := context.NewChatBuffer(\"chat-1\", \"req-1\", \"assistant-1\", \"openai\", \"\")\n\n\t\t// Add assistant message - should inherit connector from buffer\n\t\tbuffer.AddAssistantMessage(\n\t\t\t\"M1\",\n\t\t\t\"text\",\n\t\t\tmap[string]interface{}{\"content\": \"Hello\"},\n\t\t\t\"block-1\", \"thread-1\", \"assistant-1\", nil,\n\t\t)\n\n\t\tmessages := buffer.GetMessages()\n\t\trequire.Len(t, messages, 1)\n\t\tassert.Equal(t, \"openai\", messages[0].Connector)\n\t})\n\n\tt.Run(\"MessageConnectorUpdatesWithBuffer\", func(t *testing.T) {\n\t\tbuffer := context.NewChatBuffer(\"chat-1\", \"req-1\", \"assistant-1\", \"openai\", \"\")\n\n\t\t// First message with openai\n\t\tbuffer.AddAssistantMessage(\n\t\t\t\"M1\",\n\t\t\t\"text\",\n\t\t\tmap[string]interface{}{\"content\": \"Using OpenAI\"},\n\t\t\t\"\", \"\", \"assistant-1\", nil,\n\t\t)\n\n\t\t// User switches connector\n\t\tbuffer.SetConnector(\"anthropic\")\n\n\t\t// Second message with anthropic\n\t\tbuffer.AddAssistantMessage(\n\t\t\t\"M2\",\n\t\t\t\"text\",\n\t\t\tmap[string]interface{}{\"content\": \"Now using Claude\"},\n\t\t\t\"\", \"\", \"assistant-1\", nil,\n\t\t)\n\n\t\tmessages := buffer.GetMessages()\n\t\trequire.Len(t, messages, 2)\n\t\tassert.Equal(t, \"openai\", messages[0].Connector, \"First message should use openai\")\n\t\tassert.Equal(t, \"anthropic\", messages[1].Connector, \"Second message should use anthropic\")\n\t})\n\n\tt.Run(\"UserInputDoesNotSetConnector\", func(t *testing.T) {\n\t\tbuffer := context.NewChatBuffer(\"chat-1\", \"req-1\", \"assistant-1\", \"openai\", \"\")\n\n\t\t// User input doesn't have connector (it's set by the system based on which model processes it)\n\t\tbuffer.AddUserInput(\"Hello\", \"\")\n\n\t\tmessages := buffer.GetMessages()\n\t\trequire.Len(t, messages, 1)\n\t\t// User input messages don't have connector field set by AddUserInput\n\t\t// Connector is only set for assistant messages\n\t\tassert.Equal(t, \"\", messages[0].Connector)\n\t})\n\n\tt.Run(\"MultipleConnectorSwitches\", func(t *testing.T) {\n\t\tbuffer := context.NewChatBuffer(\"chat-1\", \"req-1\", \"assistant-1\", \"openai\", \"\")\n\n\t\t// Simulate a conversation with multiple connector switches\n\t\tconnectors := []string{\"openai\", \"anthropic\", \"openai\", \"google\"}\n\t\tfor i, conn := range connectors {\n\t\t\tbuffer.SetConnector(conn)\n\t\t\tbuffer.AddAssistantMessage(\n\t\t\t\tfmt.Sprintf(\"M%d\", i+1),\n\t\t\t\t\"text\",\n\t\t\t\tmap[string]interface{}{\"content\": fmt.Sprintf(\"Message %d\", i+1)},\n\t\t\t\t\"\", \"\", \"assistant-1\", nil,\n\t\t\t)\n\t\t}\n\n\t\tmessages := buffer.GetMessages()\n\t\trequire.Len(t, messages, 4)\n\n\t\tfor i, msg := range messages {\n\t\t\tassert.Equal(t, connectors[i], msg.Connector, \"Message %d should have connector %s\", i+1, connectors[i])\n\t\t}\n\t})\n}\n\n// =============================================================================\n// Concurrency Tests\n// =============================================================================\n\nfunc TestBufferConcurrentMessageOperations(t *testing.T) {\n\tbuffer := context.NewChatBuffer(\"chat-concurrent\", \"req-concurrent\", \"assistant-concurrent\", \"\", \"\")\n\n\tvar wg sync.WaitGroup\n\tnumGoroutines := 100\n\n\t// Concurrent writes\n\tfor i := 0; i < numGoroutines; i++ {\n\t\twg.Add(1)\n\t\tgo func(idx int) {\n\t\t\tdefer wg.Done()\n\t\t\tbuffer.AddMessage(&context.BufferedMessage{\n\t\t\t\tRole: \"assistant\",\n\t\t\t\tType: \"text\",\n\t\t\t\tProps: map[string]interface{}{\"content\": fmt.Sprintf(\"Message %d\", idx)},\n\t\t\t})\n\t\t}(i)\n\t}\n\n\twg.Wait()\n\n\t// Verify all messages were added\n\tmessages := buffer.GetMessages()\n\tassert.Len(t, messages, numGoroutines)\n\n\t// Verify sequences are unique\n\tsequences := make(map[int]bool)\n\tfor _, msg := range messages {\n\t\tassert.False(t, sequences[msg.Sequence], \"Duplicate sequence found: %d\", msg.Sequence)\n\t\tsequences[msg.Sequence] = true\n\t}\n}\n\nfunc TestBufferConcurrentStepOperations(t *testing.T) {\n\tbuffer := context.NewChatBuffer(\"chat-concurrent\", \"req-concurrent\", \"assistant-concurrent\", \"\", \"\")\n\n\tvar wg sync.WaitGroup\n\tnumGoroutines := 50\n\n\t// Concurrent step operations\n\tfor i := 0; i < numGoroutines; i++ {\n\t\twg.Add(1)\n\t\tgo func(idx int) {\n\t\t\tdefer wg.Done()\n\t\t\tbuffer.BeginStep(context.StepTypeLLM, map[string]interface{}{\"idx\": idx}, nil)\n\t\t\ttime.Sleep(time.Millisecond) // Simulate some work\n\t\t\tbuffer.CompleteStep(map[string]interface{}{\"result\": idx})\n\t\t}(i)\n\t}\n\n\twg.Wait()\n\n\t// Verify all steps were recorded\n\tsteps := buffer.GetAllSteps()\n\tassert.Len(t, steps, numGoroutines)\n}\n\nfunc TestBufferConcurrentReadWrite(t *testing.T) {\n\tbuffer := context.NewChatBuffer(\"chat-rw\", \"req-rw\", \"assistant-rw\", \"\", \"\")\n\n\tvar wg sync.WaitGroup\n\tdone := make(chan bool)\n\n\t// Writer goroutine\n\twg.Add(1)\n\tgo func() {\n\t\tdefer wg.Done()\n\t\tfor i := 0; i < 100; i++ {\n\t\t\tbuffer.AddMessage(&context.BufferedMessage{\n\t\t\t\tRole: \"assistant\",\n\t\t\t\tType: \"text\",\n\t\t\t\tProps: map[string]interface{}{\"content\": fmt.Sprintf(\"Message %d\", i)},\n\t\t\t})\n\t\t\ttime.Sleep(time.Microsecond)\n\t\t}\n\t}()\n\n\t// Reader goroutine\n\twg.Add(1)\n\tgo func() {\n\t\tdefer wg.Done()\n\t\tfor {\n\t\t\tselect {\n\t\t\tcase <-done:\n\t\t\t\treturn\n\t\t\tdefault:\n\t\t\t\t_ = buffer.GetMessages()\n\t\t\t\t_ = buffer.GetMessageCount()\n\t\t\t\ttime.Sleep(time.Microsecond)\n\t\t\t}\n\t\t}\n\t}()\n\n\t// Let it run for a bit\n\ttime.Sleep(50 * time.Millisecond)\n\tclose(done)\n\twg.Wait()\n\n\t// Should complete without race conditions\n\tassert.Equal(t, 100, buffer.GetMessageCount())\n}\n\n// =============================================================================\n// Step Type Constants Tests\n// =============================================================================\n\nfunc TestBufferStepTypeConstants(t *testing.T) {\n\t// Verify all step types are defined\n\tassert.Equal(t, \"input\", context.StepTypeInput)\n\tassert.Equal(t, \"hook_create\", context.StepTypeHookCreate)\n\tassert.Equal(t, \"llm\", context.StepTypeLLM)\n\tassert.Equal(t, \"tool\", context.StepTypeTool)\n\tassert.Equal(t, \"hook_next\", context.StepTypeHookNext)\n\tassert.Equal(t, \"delegate\", context.StepTypeDelegate)\n}\n\nfunc TestBufferResumeStatusConstants(t *testing.T) {\n\tassert.Equal(t, \"failed\", context.ResumeStatusFailed)\n\tassert.Equal(t, \"interrupted\", context.ResumeStatusInterrupted)\n}\n\nfunc TestBufferStepStatusConstants(t *testing.T) {\n\tassert.Equal(t, \"running\", context.StepStatusRunning)\n\tassert.Equal(t, \"completed\", context.StepStatusCompleted)\n}\n\n// =============================================================================\n// Edge Cases and Error Handling Tests\n// =============================================================================\n\nfunc TestBufferEdgeCases(t *testing.T) {\n\tt.Run(\"LargeNumberOfMessages\", func(t *testing.T) {\n\t\tbuffer := context.NewChatBuffer(\"chat-large\", \"req-large\", \"assistant-large\", \"\", \"\")\n\n\t\t// Add 10000 messages\n\t\tfor i := 0; i < 10000; i++ {\n\t\t\tbuffer.AddMessage(&context.BufferedMessage{\n\t\t\t\tRole: \"assistant\",\n\t\t\t\tType: \"text\",\n\t\t\t\tProps: map[string]interface{}{\"content\": fmt.Sprintf(\"Message %d\", i)},\n\t\t\t})\n\t\t}\n\n\t\tassert.Equal(t, 10000, buffer.GetMessageCount())\n\t\tmessages := buffer.GetMessages()\n\t\tassert.Len(t, messages, 10000)\n\t})\n\n\tt.Run(\"MessageWithEmptyProps\", func(t *testing.T) {\n\t\tbuffer := context.NewChatBuffer(\"chat-empty\", \"req-empty\", \"assistant-empty\", \"\", \"\")\n\n\t\tbuffer.AddMessage(&context.BufferedMessage{\n\t\t\tRole: \"assistant\",\n\t\t\tType: \"text\",\n\t\t\tProps: nil,\n\t\t})\n\n\t\tmessages := buffer.GetMessages()\n\t\trequire.Len(t, messages, 1)\n\t\tassert.Nil(t, messages[0].Props)\n\t})\n\n\tt.Run(\"StepWithEmptyInput\", func(t *testing.T) {\n\t\tbuffer := context.NewChatBuffer(\"chat-step\", \"req-step\", \"assistant-step\", \"\", \"\")\n\n\t\tstep := buffer.BeginStep(context.StepTypeLLM, nil, nil)\n\t\tassert.Nil(t, step.Input)\n\n\t\tbuffer.CompleteStep(nil)\n\t\tsteps := buffer.GetAllSteps()\n\t\tassert.Nil(t, steps[0].Output)\n\t})\n\n\tt.Run(\"AllMessageTypes\", func(t *testing.T) {\n\t\tbuffer := context.NewChatBuffer(\"chat-types\", \"req-types\", \"assistant-types\", \"\", \"\")\n\n\t\tmessageTypes := []string{\n\t\t\t\"text\", \"image\", \"loading\", \"tool_call\", \"tool_result\",\n\t\t\t\"retrieval\", \"thinking\", \"action\", \"chart\", \"table\",\n\t\t\t\"custom_type_1\", \"custom_type_2\",\n\t\t}\n\n\t\tfor i, msgType := range messageTypes {\n\t\t\tbuffer.AddAssistantMessage(fmt.Sprintf(\"M%d\", i+1), msgType, map[string]interface{}{\"type\": msgType}, \"\", \"\", \"\", nil)\n\t\t}\n\n\t\tassert.Equal(t, len(messageTypes), buffer.GetMessageCount())\n\t})\n\n\tt.Run(\"AllStepTypes\", func(t *testing.T) {\n\t\tbuffer := context.NewChatBuffer(\"chat-step-types\", \"req-step-types\", \"assistant-step-types\", \"\", \"\")\n\n\t\tstepTypes := []string{\n\t\t\tcontext.StepTypeInput, context.StepTypeHookCreate, context.StepTypeLLM,\n\t\t\tcontext.StepTypeTool, context.StepTypeHookNext, context.StepTypeDelegate,\n\t\t}\n\n\t\tfor _, stepType := range stepTypes {\n\t\t\tbuffer.BeginStep(stepType, nil, nil)\n\t\t\tbuffer.CompleteStep(nil)\n\t\t}\n\n\t\tsteps := buffer.GetAllSteps()\n\t\tassert.Len(t, steps, len(stepTypes))\n\t})\n}\n\n// =============================================================================\n// Integration-like Tests (Simulating Real Workflow)\n// =============================================================================\n\nfunc TestBufferCompleteWorkflow(t *testing.T) {\n\tt.Run(\"SuccessfulChatFlow\", func(t *testing.T) {\n\t\tbuffer := context.NewChatBuffer(\"chat-workflow\", \"req-workflow\", \"assistant-main\", \"\", \"\")\n\n\t\t// 1. User input\n\t\tbuffer.AddUserInput(\"What's the weather in San Francisco?\", \"John\")\n\t\tbuffer.BeginStep(context.StepTypeInput, map[string]interface{}{\"content\": \"What's the weather in San Francisco?\"}, nil)\n\t\tbuffer.CompleteStep(nil)\n\n\t\t// 2. Create hook\n\t\tbuffer.BeginStep(context.StepTypeHookCreate, nil, nil)\n\t\tbuffer.AddAssistantMessage(\"M1\", \"thinking\", map[string]interface{}{\"content\": \"Processing your request...\"}, \"block-1\", \"\", \"assistant-main\", nil)\n\t\tbuffer.CompleteStep(nil)\n\n\t\t// 3. LLM call with tool\n\t\tbuffer.BeginStep(context.StepTypeLLM, map[string]interface{}{\"model\": \"gpt-4\"}, nil)\n\t\tbuffer.AddAssistantMessage(\"M2\", \"tool_call\", map[string]interface{}{\n\t\t\t\"name\": \"get_weather\",\n\t\t\t\"arguments\": `{\"location\":\"San Francisco\"}`,\n\t\t}, \"block-2\", \"\", \"assistant-main\", nil)\n\t\tbuffer.CompleteStep(map[string]interface{}{\"tool_calls\": 1})\n\n\t\t// 4. Tool execution\n\t\tbuffer.BeginStep(context.StepTypeTool, map[string]interface{}{\"tool\": \"get_weather\"}, nil)\n\t\tbuffer.AddAssistantMessage(\"M3\", \"tool_result\", map[string]interface{}{\n\t\t\t\"result\": \"72°F, Sunny\",\n\t\t}, \"block-2\", \"\", \"assistant-main\", nil)\n\t\tbuffer.CompleteStep(map[string]interface{}{\"result\": \"72°F, Sunny\"})\n\n\t\t// 5. Final LLM response\n\t\tbuffer.BeginStep(context.StepTypeLLM, nil, nil)\n\t\tbuffer.AddAssistantMessage(\"M4\", \"text\", map[string]interface{}{\n\t\t\t\"content\": \"The weather in San Francisco is currently 72°F and sunny.\",\n\t\t}, \"block-3\", \"\", \"assistant-main\", nil)\n\t\tbuffer.CompleteStep(nil)\n\n\t\t// Verify: 1 user_input + 4 assistant messages (thinking, tool_call, tool_result, text)\n\t\tassert.Equal(t, 5, buffer.GetMessageCount())\n\t\tassert.Len(t, buffer.GetAllSteps(), 5) // 5 steps (no hook_next in this flow)\n\n\t\t// All steps should be completed\n\t\tsteps := buffer.GetStepsForResume(context.StepStatusCompleted)\n\t\tassert.Nil(t, steps)\n\t})\n\n\tt.Run(\"InterruptedChatFlow\", func(t *testing.T) {\n\t\tbuffer := context.NewChatBuffer(\"chat-interrupted\", \"req-interrupted\", \"assistant-main\", \"\", \"\")\n\n\t\t// Set space snapshot\n\t\tbuffer.SetSpaceSnapshot(map[string]interface{}{\n\t\t\t\"user_context\": \"previous conversation\",\n\t\t\t\"preferences\": map[string]interface{}{\"language\": \"en\"},\n\t\t})\n\n\t\t// 1. User input\n\t\tbuffer.AddUserInput(\"Generate a long story\", \"\")\n\t\tbuffer.BeginStep(context.StepTypeInput, nil, nil)\n\t\tbuffer.CompleteStep(nil)\n\n\t\t// 2. LLM starts generating\n\t\tbuffer.BeginStep(context.StepTypeLLM, map[string]interface{}{\"model\": \"gpt-4\"}, nil)\n\t\tbuffer.AddAssistantMessage(\"M1\", \"text\", map[string]interface{}{\"content\": \"Once upon a time...\"}, \"block-1\", \"\", \"assistant-main\", nil)\n\t\t// User interrupts here!\n\n\t\t// Get steps for resume\n\t\tsteps := buffer.GetStepsForResume(context.ResumeStatusInterrupted)\n\t\trequire.NotNil(t, steps)\n\t\tassert.Len(t, steps, 2)\n\n\t\t// Last step should be interrupted with space snapshot\n\t\tlastStep := steps[len(steps)-1]\n\t\tassert.Equal(t, context.ResumeStatusInterrupted, lastStep.Status)\n\t\tassert.NotNil(t, lastStep.SpaceSnapshot)\n\t\tassert.Equal(t, \"previous conversation\", lastStep.SpaceSnapshot[\"user_context\"])\n\t})\n\n\tt.Run(\"A2ACallWithDelegation\", func(t *testing.T) {\n\t\tbuffer := context.NewChatBuffer(\"chat-a2a\", \"req-a2a\", \"assistant-main\", \"\", \"\")\n\n\t\tmainStack := &context.Stack{ID: \"stack-main\", Depth: 0}\n\t\tchildStack := &context.Stack{ID: \"stack-child\", ParentID: \"stack-main\", Depth: 1}\n\n\t\t// Main assistant starts\n\t\tbuffer.BeginStep(context.StepTypeInput, nil, mainStack)\n\t\tbuffer.CompleteStep(nil)\n\n\t\t// Delegate to child assistant\n\t\tbuffer.SetAssistantID(\"assistant-child\")\n\t\tbuffer.BeginStep(context.StepTypeDelegate, map[string]interface{}{\"delegate_to\": \"assistant-child\"}, childStack)\n\n\t\t// Child assistant messages\n\t\tbuffer.AddAssistantMessage(\"M1\", \"text\", map[string]interface{}{\"content\": \"Child assistant responding\"}, \"block-child\", \"\", \"assistant-child\", nil)\n\t\tbuffer.CompleteStep(map[string]interface{}{\"delegate_result\": \"success\"})\n\n\t\t// Return to main assistant\n\t\tbuffer.SetAssistantID(\"assistant-main\")\n\t\tbuffer.BeginStep(context.StepTypeLLM, nil, mainStack)\n\t\tbuffer.AddAssistantMessage(\"M2\", \"text\", map[string]interface{}{\"content\": \"Main assistant continuing\"}, \"block-main\", \"\", \"assistant-main\", nil)\n\t\tbuffer.CompleteStep(nil)\n\n\t\t// Verify\n\t\tmessages := buffer.GetMessages()\n\t\tassert.Len(t, messages, 2)\n\t\tassert.Equal(t, \"assistant-child\", messages[0].AssistantID)\n\t\tassert.Equal(t, \"assistant-main\", messages[1].AssistantID)\n\n\t\tsteps := buffer.GetAllSteps()\n\t\tassert.Len(t, steps, 3)\n\t\tassert.Equal(t, \"stack-child\", steps[1].StackID)\n\t\tassert.Equal(t, \"stack-main\", steps[1].StackParentID)\n\t})\n\n\tt.Run(\"ConcurrentAgentCalls\", func(t *testing.T) {\n\t\tbuffer := context.NewChatBuffer(\"chat-concurrent-a2a\", \"req-concurrent-a2a\", \"assistant-main\", \"\", \"\")\n\n\t\t// Main assistant spawns multiple concurrent calls\n\t\tbuffer.BeginStep(context.StepTypeInput, nil, nil)\n\t\tbuffer.CompleteStep(nil)\n\n\t\t// Simulate concurrent responses with thread IDs\n\t\tvar wg sync.WaitGroup\n\t\tfor i := 0; i < 3; i++ {\n\t\t\twg.Add(1)\n\t\t\tgo func(idx int) {\n\t\t\t\tdefer wg.Done()\n\t\t\t\tthreadID := fmt.Sprintf(\"thread-%d\", idx)\n\t\t\t\tbuffer.AddAssistantMessage(\n\t\t\t\t\tfmt.Sprintf(\"M%d\", idx),\n\t\t\t\t\t\"text\",\n\t\t\t\t\tmap[string]interface{}{\"content\": fmt.Sprintf(\"Response from thread %d\", idx)},\n\t\t\t\t\t\"block-concurrent\",\n\t\t\t\t\tthreadID,\n\t\t\t\t\tfmt.Sprintf(\"assistant-%d\", idx),\n\t\t\t\t\tnil,\n\t\t\t\t)\n\t\t\t}(i)\n\t\t}\n\t\twg.Wait()\n\n\t\tmessages := buffer.GetMessages()\n\t\tassert.Len(t, messages, 3)\n\n\t\t// Verify all have same block ID but different thread IDs\n\t\tthreadIDs := make(map[string]bool)\n\t\tfor _, msg := range messages {\n\t\t\tassert.Equal(t, \"block-concurrent\", msg.BlockID)\n\t\t\tassert.False(t, threadIDs[msg.ThreadID], \"Duplicate thread ID\")\n\t\t\tthreadIDs[msg.ThreadID] = true\n\t\t}\n\t})\n}\n\n// =============================================================================\n// Message Sequence Tests\n// =============================================================================\n\nfunc TestBufferMessageSequence(t *testing.T) {\n\tt.Run(\"SequenceAutoIncrement\", func(t *testing.T) {\n\t\tbuffer := context.NewChatBuffer(\"chat-seq\", \"req-seq\", \"assistant-seq\", \"\", \"\")\n\n\t\tfor i := 0; i < 10; i++ {\n\t\t\tbuffer.AddMessage(&context.BufferedMessage{\n\t\t\t\tRole: \"assistant\",\n\t\t\t\tType: \"text\",\n\t\t\t})\n\t\t}\n\n\t\tmessages := buffer.GetMessages()\n\t\tfor i, msg := range messages {\n\t\t\tassert.Equal(t, i+1, msg.Sequence)\n\t\t}\n\t})\n\n\tt.Run(\"MixedMessageTypes\", func(t *testing.T) {\n\t\tbuffer := context.NewChatBuffer(\"chat-mixed\", \"req-mixed\", \"assistant-mixed\", \"\", \"\")\n\n\t\tbuffer.AddUserInput(\"Hello\", \"\")\n\t\tbuffer.AddAssistantMessage(\"M1\", \"text\", nil, \"\", \"\", \"\", nil)\n\t\tbuffer.AddUserInput(\"Follow up\", \"\")\n\t\tbuffer.AddAssistantMessage(\"M2\", \"tool_call\", nil, \"\", \"\", \"\", nil)\n\n\t\tmessages := buffer.GetMessages()\n\t\tassert.Len(t, messages, 4)\n\t\tfor i, msg := range messages {\n\t\t\tassert.Equal(t, i+1, msg.Sequence)\n\t\t}\n\t})\n}\n\n// =============================================================================\n// Step Sequence Tests\n// =============================================================================\n\nfunc TestBufferStepSequence(t *testing.T) {\n\tt.Run(\"SequenceAutoIncrement\", func(t *testing.T) {\n\t\tbuffer := context.NewChatBuffer(\"chat-step-seq\", \"req-step-seq\", \"assistant-step-seq\", \"\", \"\")\n\n\t\tfor i := 0; i < 5; i++ {\n\t\t\tbuffer.BeginStep(context.StepTypeLLM, nil, nil)\n\t\t\tbuffer.CompleteStep(nil)\n\t\t}\n\n\t\tsteps := buffer.GetAllSteps()\n\t\tfor i, step := range steps {\n\t\t\tassert.Equal(t, i+1, step.Sequence)\n\t\t}\n\t})\n}\n\n// =============================================================================\n// Buffer Reset/Clear Tests (if needed in future)\n// =============================================================================\n\nfunc TestBufferMultipleRequests(t *testing.T) {\n\tt.Run(\"NewBufferPerRequest\", func(t *testing.T) {\n\t\t// Simulate multiple requests with separate buffers\n\t\tbuffer1 := context.NewChatBuffer(\"chat-1\", \"req-1\", \"assistant-1\", \"\", \"\")\n\t\tbuffer1.AddUserInput(\"Request 1\", \"\")\n\n\t\tbuffer2 := context.NewChatBuffer(\"chat-1\", \"req-2\", \"assistant-1\", \"\", \"\")\n\t\tbuffer2.AddUserInput(\"Request 2\", \"\")\n\n\t\t// Buffers should be independent\n\t\tassert.Equal(t, 1, buffer1.GetMessageCount())\n\t\tassert.Equal(t, 1, buffer2.GetMessageCount())\n\n\t\tmsg1 := buffer1.GetMessages()[0]\n\t\tmsg2 := buffer2.GetMessages()[0]\n\n\t\tassert.Equal(t, \"req-1\", msg1.RequestID)\n\t\tassert.Equal(t, \"req-2\", msg2.RequestID)\n\t})\n}\n\n// =============================================================================\n// Streaming Message Tests\n// =============================================================================\n\nfunc TestBufferStreamingMessage(t *testing.T) {\n\tt.Run(\"AddStreamingMessage\", func(t *testing.T) {\n\t\tbuffer := context.NewChatBuffer(\"chat-1\", \"req-1\", \"assistant-1\", \"openai\", \"\")\n\n\t\tbuffer.AddStreamingMessage(\n\t\t\t\"msg-stream-1\",\n\t\t\t\"text\",\n\t\t\tmap[string]interface{}{\"content\": \"# Title\\n\\n\"},\n\t\t\t\"block-1\",\n\t\t\t\"thread-1\",\n\t\t\t\"assistant-1\",\n\t\t\tnil,\n\t\t)\n\n\t\tassert.Equal(t, 1, buffer.GetMessageCount())\n\n\t\t// Verify streaming message is added\n\t\tmsg := buffer.GetStreamingMessage(\"msg-stream-1\")\n\t\tassert.NotNil(t, msg)\n\t\tassert.Equal(t, \"msg-stream-1\", msg.MessageID)\n\t\tassert.Equal(t, \"text\", msg.Type)\n\t\tassert.Equal(t, \"# Title\\n\\n\", msg.Props[\"content\"])\n\t\tassert.True(t, msg.IsStreaming)\n\t})\n\n\tt.Run(\"AppendMessageContent\", func(t *testing.T) {\n\t\tbuffer := context.NewChatBuffer(\"chat-1\", \"req-1\", \"assistant-1\", \"openai\", \"\")\n\n\t\t// Add streaming message\n\t\tbuffer.AddStreamingMessage(\n\t\t\t\"msg-stream-2\",\n\t\t\t\"text\",\n\t\t\tmap[string]interface{}{\"content\": \"Initial \"},\n\t\t\t\"\", \"\", \"\", nil,\n\t\t)\n\n\t\t// Append content\n\t\tok := buffer.AppendMessageContent(\"msg-stream-2\", \"Line 1\\n\")\n\t\tassert.True(t, ok)\n\n\t\tok = buffer.AppendMessageContent(\"msg-stream-2\", \"Line 2\\n\")\n\t\tassert.True(t, ok)\n\n\t\t// Verify accumulated content\n\t\tmsg := buffer.GetStreamingMessage(\"msg-stream-2\")\n\t\tassert.NotNil(t, msg)\n\t\tassert.Equal(t, \"Initial Line 1\\nLine 2\\n\", msg.Props[\"content\"])\n\t})\n\n\tt.Run(\"AppendToNonExistentMessage\", func(t *testing.T) {\n\t\tbuffer := context.NewChatBuffer(\"chat-1\", \"req-1\", \"assistant-1\", \"openai\", \"\")\n\n\t\t// Try to append to non-existent message\n\t\tok := buffer.AppendMessageContent(\"non-existent\", \"content\")\n\t\tassert.False(t, ok)\n\t})\n\n\tt.Run(\"AppendToCompletedMessage\", func(t *testing.T) {\n\t\tbuffer := context.NewChatBuffer(\"chat-1\", \"req-1\", \"assistant-1\", \"openai\", \"\")\n\n\t\t// Add and complete streaming message\n\t\tbuffer.AddStreamingMessage(\n\t\t\t\"msg-stream-3\",\n\t\t\t\"text\",\n\t\t\tmap[string]interface{}{\"content\": \"Initial\"},\n\t\t\t\"\", \"\", \"\", nil,\n\t\t)\n\t\tbuffer.CompleteStreamingMessage(\"msg-stream-3\")\n\n\t\t// Try to append to completed message (should fail)\n\t\tok := buffer.AppendMessageContent(\"msg-stream-3\", \" more\")\n\t\tassert.False(t, ok)\n\t})\n\n\tt.Run(\"CompleteStreamingMessage\", func(t *testing.T) {\n\t\tbuffer := context.NewChatBuffer(\"chat-1\", \"req-1\", \"assistant-1\", \"openai\", \"\")\n\n\t\t// Add streaming message\n\t\tbuffer.AddStreamingMessage(\n\t\t\t\"msg-stream-4\",\n\t\t\t\"text\",\n\t\t\tmap[string]interface{}{\"content\": \"Hello \"},\n\t\t\t\"\", \"\", \"\", nil,\n\t\t)\n\n\t\t// Append content\n\t\tbuffer.AppendMessageContent(\"msg-stream-4\", \"World!\")\n\n\t\t// Complete the message\n\t\tcontent, ok := buffer.CompleteStreamingMessage(\"msg-stream-4\")\n\t\tassert.True(t, ok)\n\t\tassert.Equal(t, \"Hello World!\", content)\n\n\t\t// Message should no longer be streaming\n\t\tmsg := buffer.GetStreamingMessage(\"msg-stream-4\")\n\t\tassert.Nil(t, msg)\n\n\t\t// But should still exist in messages\n\t\tmessages := buffer.GetMessages()\n\t\tassert.Equal(t, 1, len(messages))\n\t\tassert.False(t, messages[0].IsStreaming)\n\t})\n\n\tt.Run(\"CompleteNonExistentMessage\", func(t *testing.T) {\n\t\tbuffer := context.NewChatBuffer(\"chat-1\", \"req-1\", \"assistant-1\", \"openai\", \"\")\n\n\t\tcontent, ok := buffer.CompleteStreamingMessage(\"non-existent\")\n\t\tassert.False(t, ok)\n\t\tassert.Empty(t, content)\n\t})\n\n\tt.Run(\"StreamingMessageWorkflow\", func(t *testing.T) {\n\t\tbuffer := context.NewChatBuffer(\"chat-1\", \"req-1\", \"assistant-1\", \"deepseek\", \"\")\n\n\t\t// Simulate a typical streaming workflow:\n\t\t// 1. SendStream sends initial content\n\t\tbuffer.AddStreamingMessage(\n\t\t\t\"msg-workflow\",\n\t\t\t\"text\",\n\t\t\tmap[string]interface{}{\"content\": \"# Available Tests\\n\\n\"},\n\t\t\t\"block-main\",\n\t\t\t\"\",\n\t\t\t\"assistant-1\",\n\t\t\tnil,\n\t\t)\n\n\t\t// 2. Multiple Append calls add content\n\t\tbuffer.AppendMessageContent(\"msg-workflow\", \"Send one of these keywords:\\n\\n\")\n\t\tbuffer.AppendMessageContent(\"msg-workflow\", \"- **basic** - Basic tests\\n\")\n\t\tbuffer.AppendMessageContent(\"msg-workflow\", \"- **advanced** - Advanced tests\\n\")\n\n\t\t// 3. End completes the message\n\t\tfinalContent, ok := buffer.CompleteStreamingMessage(\"msg-workflow\")\n\t\tassert.True(t, ok)\n\n\t\texpectedContent := \"# Available Tests\\n\\nSend one of these keywords:\\n\\n- **basic** - Basic tests\\n- **advanced** - Advanced tests\\n\"\n\t\tassert.Equal(t, expectedContent, finalContent)\n\n\t\t// Verify final message state\n\t\tmessages := buffer.GetMessages()\n\t\tassert.Equal(t, 1, len(messages))\n\t\tassert.Equal(t, \"msg-workflow\", messages[0].MessageID)\n\t\tassert.Equal(t, \"deepseek\", messages[0].Connector) // Connector should be set\n\t\tassert.False(t, messages[0].IsStreaming)\n\t})\n\n\tt.Run(\"MixedStreamingAndRegularMessages\", func(t *testing.T) {\n\t\tbuffer := context.NewChatBuffer(\"chat-1\", \"req-1\", \"assistant-1\", \"openai\", \"\")\n\n\t\t// Add user input (regular)\n\t\tbuffer.AddUserInput(\"Hello\", \"user1\")\n\n\t\t// Add streaming assistant message\n\t\tbuffer.AddStreamingMessage(\n\t\t\t\"msg-stream\",\n\t\t\t\"text\",\n\t\t\tmap[string]interface{}{\"content\": \"Hi \"},\n\t\t\t\"\", \"\", \"\", nil,\n\t\t)\n\t\tbuffer.AppendMessageContent(\"msg-stream\", \"there!\")\n\t\tbuffer.CompleteStreamingMessage(\"msg-stream\")\n\n\t\t// Add regular assistant message\n\t\tbuffer.AddAssistantMessage(\"M3\", \"text\", map[string]interface{}{\"content\": \"How can I help?\"}, \"\", \"\", \"\", nil)\n\n\t\t// Verify all messages\n\t\tmessages := buffer.GetMessages()\n\t\tassert.Equal(t, 3, len(messages))\n\n\t\t// Check sequence\n\t\tassert.Equal(t, 1, messages[0].Sequence)\n\t\tassert.Equal(t, 2, messages[1].Sequence)\n\t\tassert.Equal(t, 3, messages[2].Sequence)\n\n\t\t// Check content\n\t\tassert.Equal(t, \"user\", messages[0].Role)\n\t\tassert.Equal(t, \"Hi there!\", messages[1].Props[\"content\"])\n\t\tassert.Equal(t, \"How can I help?\", messages[2].Props[\"content\"])\n\t})\n\n\tt.Run(\"StreamingMessageWithEmptyInitialContent\", func(t *testing.T) {\n\t\tbuffer := context.NewChatBuffer(\"chat-1\", \"req-1\", \"assistant-1\", \"openai\", \"\")\n\n\t\t// Add streaming message with nil props\n\t\tbuffer.AddStreamingMessage(\n\t\t\t\"msg-empty\",\n\t\t\t\"text\",\n\t\t\tnil,\n\t\t\t\"\", \"\", \"\", nil,\n\t\t)\n\n\t\t// Append content\n\t\tbuffer.AppendMessageContent(\"msg-empty\", \"Content\")\n\n\t\t// Complete\n\t\tcontent, ok := buffer.CompleteStreamingMessage(\"msg-empty\")\n\t\tassert.True(t, ok)\n\t\tassert.Equal(t, \"Content\", content)\n\t})\n\n\tt.Run(\"ConcurrentStreamingOperations\", func(t *testing.T) {\n\t\tbuffer := context.NewChatBuffer(\"chat-1\", \"req-1\", \"assistant-1\", \"openai\", \"\")\n\n\t\t// Add streaming message\n\t\tbuffer.AddStreamingMessage(\n\t\t\t\"msg-concurrent\",\n\t\t\t\"text\",\n\t\t\tmap[string]interface{}{\"content\": \"\"},\n\t\t\t\"\", \"\", \"\", nil,\n\t\t)\n\n\t\t// Concurrent appends with fixed-length content\n\t\tvar wg sync.WaitGroup\n\t\tfor i := 0; i < 100; i++ {\n\t\t\twg.Add(1)\n\t\t\tgo func() {\n\t\t\t\tdefer wg.Done()\n\t\t\t\tbuffer.AppendMessageContent(\"msg-concurrent\", \"x\")\n\t\t\t}()\n\t\t}\n\t\twg.Wait()\n\n\t\t// Complete\n\t\tcontent, ok := buffer.CompleteStreamingMessage(\"msg-concurrent\")\n\t\tassert.True(t, ok)\n\n\t\t// Content should have 100 'x' characters\n\t\tassert.Equal(t, 100, len(content))\n\t})\n}\n" }, { "path": "agent/context/chat.go", "content": "package context\n\nimport (\n\t\"crypto/sha256\"\n\t\"encoding/hex\"\n\t\"encoding/json\"\n\t\"fmt\"\n\t\"time\"\n\n\tgonanoid \"github.com/matoous/go-nanoid/v2\"\n\t\"github.com/yaoapp/gou/store\"\n)\n\nconst (\n\tchatCachePrefix = \"chat:messages:\"\n\tchatCacheTTL = time.Hour * 24 * 7 // 7 days\n)\n\n// filterNonAssistantMessages returns messages excluding assistant messages\nfunc filterNonAssistantMessages(messages []Message) []Message {\n\tvar filtered []Message\n\tfor _, msg := range messages {\n\t\tif msg.Role != RoleAssistant {\n\t\t\tfiltered = append(filtered, msg)\n\t\t}\n\t}\n\treturn filtered\n}\n\n// countUserMessages returns the number of user role messages\nfunc countUserMessages(messages []Message) int {\n\tcount := 0\n\tfor _, msg := range messages {\n\t\tif msg.Role == RoleUser {\n\t\t\tcount++\n\t\t}\n\t}\n\treturn count\n}\n\n// GetChatIDByMessages gets or generates a chat ID based on message content\n// Matching strategy:\n// - Only non-assistant messages (system, developer, user, tool) are used for matching\n// - User adds a new message at the end each time\n// - To detect continuation, we match messages BEFORE the last non-assistant message\n// - If previous non-assistant messages match cached conversation → same chat\n// - For first user message (even with system/developer messages): always generate new chat ID\nfunc GetChatIDByMessages(cache store.Store, messages []Message) (string, error) {\n\tif len(messages) == 0 {\n\t\treturn \"\", fmt.Errorf(\"messages cannot be empty\")\n\t}\n\n\t// Filter out assistant messages for matching\n\tnonAssistantMessages := filterNonAssistantMessages(messages)\n\n\t// Count user messages to determine matching strategy\n\tuserMessageCount := countUserMessages(nonAssistantMessages)\n\n\tvar chatID string\n\tvar matched bool\n\n\t// Matching strategy based on user message count:\n\t// - 1 user message: generate new chat ID (cannot determine continuation)\n\t// - 2+ user messages: match all except last (which is the new user input)\n\tif userMessageCount >= 2 {\n\t\t// Match previous messages (all except last non-assistant message)\n\t\tmatchMessages := nonAssistantMessages[:len(nonAssistantMessages)-1]\n\t\thash, err := hashMessages(matchMessages)\n\t\tif err == nil {\n\t\t\tkey := getKey(hash)\n\t\t\tif cachedID, ok := cache.Get(key); ok {\n\t\t\t\tif chatIDStr, ok := cachedID.(string); ok && chatIDStr != \"\" {\n\t\t\t\t\tchatID = chatIDStr\n\t\t\t\t\tmatched = true\n\t\t\t\t}\n\t\t\t}\n\t\t}\n\t}\n\n\t// If no match, generate new chat ID\n\tif !matched {\n\t\tchatID = GenChatID()\n\t}\n\n\t// Cache the current messages for future matching\n\t// CacheChatID will handle filtering assistant messages\n\t// Next request will have one more message and will try to match current messages\n\t_ = CacheChatID(cache, messages, chatID)\n\n\treturn chatID, nil\n}\n\n// CacheChatID cache the chat ID with all message prefixes for future matching\n// It caches ALL prefixes of the message array to enable conversation continuation detection\n// Assistant messages are automatically filtered out before caching\n// Example: For messages [A,B,C], it caches hashes for [A], [A,B], and [A,B,C]\nfunc CacheChatID(cache store.Store, messages []Message, chatID string) error {\n\tif len(messages) == 0 {\n\t\treturn fmt.Errorf(\"messages cannot be empty\")\n\t}\n\n\tif chatID == \"\" {\n\t\treturn fmt.Errorf(\"chatID cannot be empty\")\n\t}\n\n\t// Filter out assistant messages\n\tnonAssistantMessages := filterNonAssistantMessages(messages)\n\tif len(nonAssistantMessages) == 0 {\n\t\treturn fmt.Errorf(\"no non-assistant messages to cache\")\n\t}\n\n\t// Cache all prefixes of the non-assistant messages array\n\t// This allows detecting conversation continuation when new messages are added\n\tfor length := 1; length <= len(nonAssistantMessages); length++ {\n\t\tprefix := nonAssistantMessages[:length]\n\t\thash, err := hashMessages(prefix)\n\t\tif err != nil {\n\t\t\tcontinue // Skip this prefix if hashing fails\n\t\t}\n\n\t\tkey := getKey(hash)\n\t\t// Ignore errors for individual cache sets\n\t\t_ = cache.Set(key, chatID, chatCacheTTL)\n\t}\n\n\treturn nil\n}\n\n// GenChatID generate a new chat ID using NanoID algorithm\n// safe: optional parameter, reserved for future safe mode implementation (collision detection)\nfunc GenChatID(safe ...bool) string {\n\t// TODO: Implement safe mode with collision detection when needed\n\t// For now, NanoID provides sufficient uniqueness without collision checking\n\n\t// URL-safe alphabet (no ambiguous characters like 0/O, 1/l/I)\n\tconst alphabet = \"23456789ABCDEFGHJKMNPQRSTUVWXYZabcdefghijkmnpqrstuvwxyz\"\n\tconst length = 16 // 16 characters provides good balance of uniqueness and readability\n\n\tid, err := gonanoid.Generate(alphabet, length)\n\tif err != nil {\n\t\t// Fallback to timestamp-based ID if NanoID generation fails\n\t\treturn fmt.Sprintf(\"%d\", time.Now().UnixNano())\n\t}\n\n\treturn id\n}\n\n// getKey generates a cache key for messages\nfunc getKey(messageHash string) string {\n\treturn chatCachePrefix + messageHash\n}\n\n// hashMessage generates a hash for a single message\nfunc hashMessage(msg Message) (string, error) {\n\tdata, err := json.Marshal(msg)\n\tif err != nil {\n\t\treturn \"\", err\n\t}\n\thash := sha256.Sum256(data)\n\treturn hex.EncodeToString(hash[:]), nil\n}\n\n// hashMessages generates a combined hash for a slice of messages\n// Note: Caller is responsible for filtering messages (e.g., removing assistant messages)\nfunc hashMessages(messages []Message) (string, error) {\n\tif len(messages) == 0 {\n\t\treturn \"\", fmt.Errorf(\"messages cannot be empty\")\n\t}\n\n\tvar hashes string\n\tfor _, msg := range messages {\n\t\thash, err := hashMessage(msg)\n\t\tif err != nil {\n\t\t\treturn \"\", err\n\t\t}\n\t\thashes += hash\n\t}\n\n\t// Generate final hash from combined hashes\n\tfinalHash := sha256.Sum256([]byte(hashes))\n\treturn hex.EncodeToString(finalHash[:]), nil\n}\n" }, { "path": "agent/context/chat_test.go", "content": "package context_test\n\nimport (\n\t\"testing\"\n\n\t\"github.com/yaoapp/gou/store\"\n\t\"github.com/yaoapp/yao/agent/context\"\n\t\"github.com/yaoapp/yao/config\"\n\t\"github.com/yaoapp/yao/test\"\n)\n\nfunc getTestCache(t *testing.T) store.Store {\n\tcache, err := store.Get(\"__yao.agent.cache\")\n\tif err != nil {\n\t\tt.Fatalf(\"Failed to get cache store: %v\", err)\n\t}\n\tcache.Clear() // Clean before test\n\treturn cache\n}\n\nfunc TestGetChatIDByMessages_NewConversation(t *testing.T) {\n\ttest.Prepare(t, config.Conf)\n\tdefer test.Clean()\n\n\tcache := getTestCache(t)\n\n\tmessages := []context.Message{\n\t\t{\n\t\t\tRole: context.RoleUser,\n\t\t\tContent: \"Hello, how are you?\",\n\t\t},\n\t}\n\n\t// First request - should generate new chat ID\n\tchatID1, err := context.GetChatIDByMessages(cache, messages)\n\tif err != nil {\n\t\tt.Fatalf(\"Failed to get chat ID: %v\", err)\n\t}\n\n\tif chatID1 == \"\" {\n\t\tt.Fatal(\"Expected non-empty chat ID\")\n\t}\n\n\t// Second request with same single user message - should generate DIFFERENT chat ID\n\t// (single user message always generates new chat ID to avoid false matches)\n\tchatID2, err := context.GetChatIDByMessages(cache, messages)\n\tif err != nil {\n\t\tt.Fatalf(\"Failed to get chat ID: %v\", err)\n\t}\n\n\tif chatID2 == \"\" {\n\t\tt.Fatal(\"Expected non-empty chat ID\")\n\t}\n\n\t// Both should be valid but different (single user message = new conversation each time)\n\tif chatID1 == chatID2 {\n\t\tt.Errorf(\"Expected different chat IDs for single user message, got same ID: %s\", chatID1)\n\t}\n}\n\nfunc TestGetChatIDByMessages_ContinuousConversation(t *testing.T) {\n\ttest.Prepare(t, config.Conf)\n\tdefer test.Clean()\n\n\tcache := getTestCache(t)\n\n\t// Scenario: User conversation with incrementally added messages\n\t// Request 1: [user1]\n\tmessages1 := []context.Message{\n\t\t{Role: context.RoleUser, Content: \"First message\"},\n\t}\n\tchatID1, err := context.GetChatIDByMessages(cache, messages1)\n\tif err != nil {\n\t\tt.Fatalf(\"Failed to get chat ID: %v\", err)\n\t}\n\n\t// Request 2: [user1, user2]\n\t// For 2 messages, matches last 1 message\n\t// Should match chatID1 because last message is cached\n\tmessages2 := []context.Message{\n\t\t{Role: context.RoleUser, Content: \"First message\"},\n\t\t{Role: context.RoleUser, Content: \"Second message\"},\n\t}\n\tchatID2, err := context.GetChatIDByMessages(cache, messages2)\n\tif err != nil {\n\t\tt.Fatalf(\"Failed to get chat ID: %v\", err)\n\t}\n\n\tif chatID1 != chatID2 {\n\t\tt.Errorf(\"Expected chatID2 to match chatID1, got %s and %s\", chatID2, chatID1)\n\t}\n\n\t// Request 3: [user1, user2, user3]\n\t// For 3+ messages, matches last 2 messages\n\t// Should match chatID2 because last 2 messages are cached\n\tmessages3 := []context.Message{\n\t\t{Role: context.RoleUser, Content: \"First message\"},\n\t\t{Role: context.RoleUser, Content: \"Second message\"},\n\t\t{Role: context.RoleUser, Content: \"Third message\"},\n\t}\n\tchatID3, err := context.GetChatIDByMessages(cache, messages3)\n\tif err != nil {\n\t\tt.Fatalf(\"Failed to get chat ID: %v\", err)\n\t}\n\n\tif chatID2 != chatID3 {\n\t\tt.Errorf(\"Expected chatID3 to match chatID2, got %s and %s\", chatID3, chatID2)\n\t}\n\n\t// Request 4: [user1, user2, user3, user4]\n\t// Should match chatID3 because last 2 messages are cached\n\tmessages4 := []context.Message{\n\t\t{Role: context.RoleUser, Content: \"First message\"},\n\t\t{Role: context.RoleUser, Content: \"Second message\"},\n\t\t{Role: context.RoleUser, Content: \"Third message\"},\n\t\t{Role: context.RoleUser, Content: \"Fourth message\"},\n\t}\n\tchatID4, err := context.GetChatIDByMessages(cache, messages4)\n\tif err != nil {\n\t\tt.Fatalf(\"Failed to get chat ID: %v\", err)\n\t}\n\n\tif chatID3 != chatID4 {\n\t\tt.Errorf(\"Expected chatID4 to match chatID3, got %s and %s\", chatID4, chatID3)\n\t}\n\n\t// All should be the same conversation\n\tif chatID1 != chatID4 {\n\t\tt.Errorf(\"Expected all chat IDs to be the same, got %s and %s\", chatID1, chatID4)\n\t}\n}\n\nfunc TestGetChatIDByMessages_DifferentConversations(t *testing.T) {\n\ttest.Prepare(t, config.Conf)\n\tdefer test.Clean()\n\n\tcache := getTestCache(t)\n\n\t// First conversation\n\tmessages1 := []context.Message{\n\t\t{\n\t\t\tRole: context.RoleUser,\n\t\t\tContent: \"Hello\",\n\t\t},\n\t}\n\n\tchatID1, err := context.GetChatIDByMessages(cache, messages1)\n\tif err != nil {\n\t\tt.Fatalf(\"Failed to get chat ID: %v\", err)\n\t}\n\n\terr = context.CacheChatID(cache, messages1, chatID1)\n\tif err != nil {\n\t\tt.Fatalf(\"Failed to cache chat ID: %v\", err)\n\t}\n\n\t// Different conversation\n\tmessages2 := []context.Message{\n\t\t{\n\t\t\tRole: context.RoleUser,\n\t\t\tContent: \"Goodbye\",\n\t\t},\n\t}\n\n\tchatID2, err := context.GetChatIDByMessages(cache, messages2)\n\tif err != nil {\n\t\tt.Fatalf(\"Failed to get chat ID: %v\", err)\n\t}\n\n\tif chatID1 == chatID2 {\n\t\tt.Errorf(\"Expected different chat IDs for different conversations, got %s\", chatID1)\n\t}\n}\n\nfunc TestGetChatIDByMessages_MultiModalContent(t *testing.T) {\n\ttest.Prepare(t, config.Conf)\n\tdefer test.Clean()\n\n\tcache := getTestCache(t)\n\n\t// First request with multimodal content\n\tmessages1 := []context.Message{\n\t\t{\n\t\t\tRole: context.RoleUser,\n\t\t\tContent: []context.ContentPart{\n\t\t\t\t{\n\t\t\t\t\tType: context.ContentText,\n\t\t\t\t\tText: \"What's in this image?\",\n\t\t\t\t},\n\t\t\t\t{\n\t\t\t\t\tType: context.ContentImageURL,\n\t\t\t\t\tImageURL: &context.ImageURL{\n\t\t\t\t\t\tURL: \"https://example.com/image.jpg\",\n\t\t\t\t\t\tDetail: context.DetailHigh,\n\t\t\t\t\t},\n\t\t\t\t},\n\t\t\t},\n\t\t},\n\t}\n\n\tchatID1, err := context.GetChatIDByMessages(cache, messages1)\n\tif err != nil {\n\t\tt.Fatalf(\"Failed to get chat ID: %v\", err)\n\t}\n\n\t// Second request - add another message to continue conversation\n\tmessages2 := append(messages1, context.Message{\n\t\tRole: context.RoleUser,\n\t\tContent: \"Tell me more details\",\n\t})\n\n\tchatID2, err := context.GetChatIDByMessages(cache, messages2)\n\tif err != nil {\n\t\tt.Fatalf(\"Failed to get chat ID: %v\", err)\n\t}\n\n\t// Should get same chat ID (continuation)\n\tif chatID1 != chatID2 {\n\t\tt.Errorf(\"Expected same chat ID for multimodal continuation, got %s and %s\", chatID1, chatID2)\n\t}\n}\n\nfunc TestGetChatIDByMessages_WithToolCalls(t *testing.T) {\n\ttest.Prepare(t, config.Conf)\n\tdefer test.Clean()\n\n\tcache := getTestCache(t)\n\n\t// First request with user message\n\tmessages1 := []context.Message{\n\t\t{\n\t\t\tRole: context.RoleUser,\n\t\t\tContent: \"What's the weather in Tokyo?\",\n\t\t},\n\t}\n\n\tchatID1, err := context.GetChatIDByMessages(cache, messages1)\n\tif err != nil {\n\t\tt.Fatalf(\"Failed to get chat ID: %v\", err)\n\t}\n\n\t// Second request - add assistant response and another user message\n\tmessages2 := []context.Message{\n\t\t{\n\t\t\tRole: context.RoleUser,\n\t\t\tContent: \"What's the weather in Tokyo?\",\n\t\t},\n\t\t{\n\t\t\tRole: context.RoleAssistant,\n\t\t\tContent: nil,\n\t\t\tToolCalls: []context.ToolCall{\n\t\t\t\t{\n\t\t\t\t\tID: \"call_123\",\n\t\t\t\t\tType: context.ToolTypeFunction,\n\t\t\t\t\tFunction: context.Function{\n\t\t\t\t\t\tName: \"get_weather\",\n\t\t\t\t\t\tArguments: `{\"location\":\"Tokyo\"}`,\n\t\t\t\t\t},\n\t\t\t\t},\n\t\t\t},\n\t\t},\n\t\t{\n\t\t\tRole: context.RoleUser,\n\t\t\tContent: \"How about tomorrow?\",\n\t\t},\n\t}\n\n\tchatID2, err := context.GetChatIDByMessages(cache, messages2)\n\tif err != nil {\n\t\tt.Fatalf(\"Failed to get chat ID: %v\", err)\n\t}\n\n\t// Should get same chat ID (assistant messages are ignored, so it matches the first user message)\n\tif chatID1 != chatID2 {\n\t\tt.Errorf(\"Expected same chat ID for messages with tool calls, got %s and %s\", chatID1, chatID2)\n\t}\n}\n\nfunc TestCacheChatID_EmptyMessages(t *testing.T) {\n\ttest.Prepare(t, config.Conf)\n\tdefer test.Clean()\n\n\tcache := getTestCache(t)\n\n\terr := context.CacheChatID(cache, []context.Message{}, \"chat_123\")\n\tif err == nil {\n\t\tt.Error(\"Expected error for empty messages\")\n\t}\n}\n\nfunc TestCacheChatID_EmptyChatID(t *testing.T) {\n\ttest.Prepare(t, config.Conf)\n\tdefer test.Clean()\n\n\tcache := getTestCache(t)\n\n\tmessages := []context.Message{\n\t\t{\n\t\t\tRole: context.RoleUser,\n\t\t\tContent: \"Hello\",\n\t\t},\n\t}\n\n\terr := context.CacheChatID(cache, messages, \"\")\n\tif err == nil {\n\t\tt.Error(\"Expected error for empty chat ID\")\n\t}\n}\n\nfunc TestGetChatIDByMessages_EmptyMessages(t *testing.T) {\n\ttest.Prepare(t, config.Conf)\n\tdefer test.Clean()\n\n\tcache := getTestCache(t)\n\n\t_, err := context.GetChatIDByMessages(cache, []context.Message{})\n\tif err == nil {\n\t\tt.Error(\"Expected error for empty messages\")\n\t}\n}\n\nfunc TestGenChatID(t *testing.T) {\n\tid1 := context.GenChatID()\n\n\tif id1 == \"\" {\n\t\tt.Error(\"Expected non-empty chat ID\")\n\t}\n\n\t// Check length - NanoID with length 16 should produce 16 character strings\n\tif len(id1) < 10 {\n\t\tt.Errorf(\"Expected chat ID to have reasonable length, got %d characters: %s\", len(id1), id1)\n\t}\n\n\t// Note: We don't test uniqueness here because nano timestamp-based IDs\n\t// can occasionally be the same when generated in rapid succession.\n\t// The uniqueness is good enough for production use.\n}\n" }, { "path": "agent/context/context.go", "content": "package context\n\nimport (\n\t\"context\"\n\t\"fmt\"\n\t\"sync\"\n\n\t\"github.com/google/uuid\"\n\t\"github.com/yaoapp/yao/agent/memory\"\n\t\"github.com/yaoapp/yao/agent/output/message\"\n\t\"github.com/yaoapp/yao/config\"\n\t\"github.com/yaoapp/yao/openapi/oauth/types\"\n\t\"github.com/yaoapp/yao/trace\"\n\ttraceTypes \"github.com/yaoapp/yao/trace/types\"\n)\n\n// Global context registry for interrupt management\nvar (\n\tcontextRegistry = &sync.Map{} // map[contextID]*Context\n)\n\n// New create a new context with basic initialization\nfunc New(parent context.Context, authorized *types.AuthorizedInfo, chatID string) *Context {\n\tif parent == nil {\n\t\tparent = context.Background()\n\t}\n\n\tcontextID := generateContextID()\n\n\t// Extract user and team IDs from authorized info\n\tvar userID, teamID string\n\tif authorized != nil {\n\t\tuserID = authorized.UserID\n\t\tteamID = authorized.TeamID\n\t}\n\n\t// Create memory instance using global manager\n\tmem, _ := memory.GetMemory(userID, teamID, chatID, contextID)\n\n\tctx := &Context{\n\t\tContext: parent,\n\t\tID: contextID, // Generate unique ID for the context\n\t\tAuthorized: authorized, // Set authorized info\n\t\tMemory: mem,\n\t\tChatID: chatID,\n\t\tIDGenerator: message.NewIDGenerator(), // Initialize ID generator for this context\n\t\tmessageMetadata: newMessageMetadataStore(), // Initialize message metadata store\n\t\tLogger: NewRequestLogger(\"\", chatID, contextID), // Initialize logger (assistantID set later)\n\t}\n\n\treturn ctx\n}\n\n// Release the context and clean up all resources including stacks and trace\nfunc (ctx *Context) Release() {\n\tif ctx.Logger != nil {\n\t\tctx.Logger.Release()\n\t}\n\n\t// Unregister from global registry\n\tif ctx.ID != \"\" {\n\t\tUnregister(ctx.ID)\n\t}\n\n\t// Stop interrupt controller\n\tif ctx.Interrupt != nil {\n\t\tif ctx.Logger != nil {\n\t\t\tctx.Logger.Cleanup(\"Interrupt controller\")\n\t\t}\n\t\tctx.Interrupt.Stop()\n\t\tctx.Interrupt = nil\n\t}\n\n\t// Complete and release trace if exists.\n\t// Only the root context (non-forked) owns the trace lifecycle.\n\t// Forked contexts share the same trace manager but must not release it.\n\tif ctx.trace != nil && ctx.Stack != nil && ctx.Stack.TraceID != \"\" {\n\t\tif ctx.ForkParent == nil {\n\t\t\tif ctx.Logger != nil {\n\t\t\t\tctx.Logger.Cleanup(\"Trace: \" + ctx.Stack.TraceID)\n\t\t\t}\n\t\t\tif ctx.Context != nil && ctx.Context.Err() != nil {\n\t\t\t\ttrace.MarkCancelled(ctx.Stack.TraceID, ctx.Context.Err().Error())\n\t\t\t\ttrace.Release(ctx.Stack.TraceID)\n\t\t\t} else {\n\t\t\t\tctx.trace.MarkComplete()\n\t\t\t\ttrace.Release(ctx.Stack.TraceID)\n\t\t\t}\n\t\t}\n\t\tctx.trace = nil\n\t}\n\n\t// Clear context-level memory only (request-scoped temporary data)\n\t// User, Team, Chat level memory is persistent and should NOT be cleared\n\tif ctx.Memory != nil && ctx.Memory.Context != nil {\n\t\tif ctx.Logger != nil {\n\t\t\tctx.Logger.Cleanup(\"Memory.Context\")\n\t\t}\n\t\tctx.Memory.Context.Clear()\n\t}\n\tctx.Memory = nil\n\n\t// Clear stacks\n\tif ctx.Stacks != nil {\n\t\tif ctx.Logger != nil {\n\t\t\tctx.Logger.Cleanup(fmt.Sprintf(\"Stacks (%d)\", len(ctx.Stacks)))\n\t\t}\n\t\tfor k := range ctx.Stacks {\n\t\t\tdelete(ctx.Stacks, k)\n\t\t}\n\t\tctx.Stacks = nil\n\t}\n\n\t// Clear current stack reference\n\tctx.Stack = nil\n\n\t// Close SafeWriter if exists (must be before setting Writer to nil)\n\t// This ensures the background goroutine is properly stopped\n\tctx.CloseSafeWriter()\n\n\t// Clear writer reference\n\tctx.Writer = nil\n\n\t// Close logger (MUST be last)\n\tif ctx.Logger != nil {\n\t\tctx.Logger.Close()\n\t\tctx.Logger = nil\n\t}\n}\n\n// GetAuthorizedMap returns the authorized information as a map\n// This implements the AuthorizedProvider interface for MCP process calls\n// Allows MCP tools to receive authorization context when called via Process transport\nfunc (ctx *Context) GetAuthorizedMap() map[string]interface{} {\n\tif ctx.Authorized == nil {\n\t\treturn nil\n\t}\n\treturn ctx.Authorized.AuthorizedToMap()\n}\n\n// Fork creates a child context for concurrent agent/LLM calls\n// The forked context shares read-only resources (Authorized, Cache, Writer)\n// but has its own independent Stack, Logger, and Memory.Context namespace\n// to avoid race conditions and state sharing issues.\n//\n// This is essential for batch operations (All/Any/Race) where multiple goroutines\n// need to execute concurrently without interfering with each other's state.\n//\n// Key behavior:\n// - Memory.User, Memory.Team, Memory.Chat are shared (cross-request state)\n// - Memory.Context is INDEPENDENT (request-scoped state, isolated per fork)\n//\n// The forked context does NOT need to be released separately - the parent context\n// manages shared resources. However, the child's Stack will be collected in parent's Stacks map.\nfunc (ctx *Context) Fork() *Context {\n\tchildID := generateContextID()\n\n\t// Fork memory with independent Context namespace\n\t// This prevents parallel sub-agents from sharing ctx.memory.context state\n\tvar forkedMemory *memory.Memory\n\tif ctx.Memory != nil {\n\t\tvar err error\n\t\tforkedMemory, err = ctx.Memory.Fork(childID)\n\t\tif err != nil {\n\t\t\t// Fallback to shared memory if fork fails (log warning)\n\t\t\tforkedMemory = ctx.Memory\n\t\t}\n\t}\n\n\tchild := &Context{\n\t\t// Inherit parent's standard context\n\t\tContext: ctx.Context,\n\n\t\t// New unique ID for this forked context\n\t\tID: childID,\n\n\t\t// Memory with independent Context namespace (see above)\n\t\tMemory: forkedMemory,\n\n\t\t// Share read-only/thread-safe resources with parent\n\t\tCache: ctx.Cache, // Cache store is thread-safe\n\t\tWriter: ctx.Writer, // Output writer is thread-safe (output module handles concurrency)\n\t\tAuthorized: ctx.Authorized, // Read-only auth info\n\t\tCapabilities: ctx.Capabilities, // Read-only model capabilities\n\n\t\t// Share reference to parent's Stacks map for trace collection\n\t\t// Child stacks will be added here by EnterStack\n\t\tStacks: ctx.Stacks,\n\n\t\t// Stack is nil for forked contexts - will be set by EnterStack\n\t\t// ForkParent stores parent stack info so EnterStack can create child stack\n\t\tStack: nil,\n\n\t\t// Create independent resources to avoid race conditions\n\t\tIDGenerator: message.NewIDGenerator(),\n\t\tLogger: NewRequestLogger(ctx.AssistantID, ctx.ChatID, childID, WithParentID(ctx.ID)),\n\t\tmessageMetadata: newMessageMetadataStore(),\n\n\t\t// Inherit context metadata\n\t\tChatID: ctx.ChatID,\n\t\tAssistantID: ctx.AssistantID,\n\t\tLocale: ctx.Locale,\n\t\tTheme: ctx.Theme,\n\t\tClient: ctx.Client,\n\t\tReferer: ctx.Referer,\n\t\tAccept: ctx.Accept,\n\t\tRoute: ctx.Route,\n\t\tMetadata: ctx.Metadata,\n\n\t\t// Don't inherit these - they are request-specific\n\t\tBuffer: nil, // Buffer belongs to root context\n\t\tInterrupt: nil, // Interrupt controller belongs to root context\n\t\ttrace: nil, // Trace will be inherited via TraceID in Stack\n\t}\n\n\t// Set ForkParent info if parent has a Stack\n\t// This allows EnterStack to create a child stack instead of root stack\n\tif ctx.Stack != nil {\n\t\tchild.ForkParent = &ForkParentInfo{\n\t\t\tStackID: ctx.Stack.ID,\n\t\t\tTraceID: ctx.Stack.TraceID,\n\t\t\tDepth: ctx.Stack.Depth,\n\t\t\tPath: append([]string{}, ctx.Stack.Path...), // Copy path slice\n\t\t}\n\t}\n\n\treturn child\n}\n\n// Send sends data to the context's writer\n// This is used by the output module to send messages to the client\n// func (ctx *Context) Send(data []byte) error {\n// \tif ctx.Writer == nil {\n// \t\treturn nil // No writer, silently ignore\n// \t}\n\n// \t_, err := ctx.Writer.Write(data)\n// \treturn err\n// }\n\n// Trace returns the trace manager for this context, lazily initialized on first call\n// Uses the TraceID from ctx.Stack if available, or generates a new one\nfunc (ctx *Context) Trace() (traceTypes.Manager, error) {\n\t// Return trace if already initialized\n\tif ctx.trace != nil {\n\t\treturn ctx.trace, nil\n\t}\n\n\t// Get TraceID from Stack or generate new one\n\tvar traceID string\n\tif ctx.Stack != nil && ctx.Stack.TraceID != \"\" {\n\t\ttraceID = ctx.Stack.TraceID\n\n\t\t// Try to load existing trace first\n\t\tmanager, err := trace.Load(traceID)\n\t\tif err == nil {\n\t\t\t// Found in registry, reuse it\n\t\t\tctx.trace = manager\n\t\t\treturn manager, nil\n\t\t}\n\t}\n\n\t// Get trace configuration from global config\n\tcfg := config.Conf\n\n\t// Prepare driver options\n\tvar driverOptions []any\n\tvar driverType string\n\n\tswitch cfg.Trace.Driver {\n\tcase \"store\":\n\t\tdriverType = trace.Store\n\t\tif cfg.Trace.Store == \"\" {\n\t\t\treturn nil, fmt.Errorf(\"trace store ID not configured\")\n\t\t}\n\t\tdriverOptions = []any{cfg.Trace.Store, cfg.Trace.Prefix}\n\n\tcase \"local\", \"\":\n\t\tdriverType = trace.Local\n\t\tdriverOptions = []any{cfg.Trace.Path}\n\n\tdefault:\n\t\treturn nil, fmt.Errorf(\"unsupported trace driver: %s\", cfg.Trace.Driver)\n\t}\n\n\t// Prepare trace options\n\ttraceOption := &traceTypes.TraceOption{ID: traceID, AutoArchive: config.Conf.Mode == \"production\"}\n\n\t// Set trace options from authorized information\n\tif ctx.Authorized != nil {\n\t\ttraceOption.CreatedBy = ctx.Authorized.UserID\n\t\ttraceOption.TeamID = ctx.Authorized.TeamID\n\t\ttraceOption.TenantID = ctx.Authorized.TenantID\n\t}\n\n\t// Create trace using trace.New (handles registry)\n\tcreatedTraceID, manager, err := trace.New(ctx.Context, driverType, traceOption, driverOptions...)\n\tif err != nil {\n\t\treturn nil, fmt.Errorf(\"failed to create trace: %w\", err)\n\t}\n\n\t// Update Stack with the created TraceID if needed\n\tif ctx.Stack != nil && ctx.Stack.TraceID == \"\" {\n\t\tctx.Stack.TraceID = createdTraceID\n\t}\n\n\t// Store for future calls\n\tctx.trace = manager\n\n\treturn manager, nil\n}\n\n// Map the context to a map\nfunc (ctx *Context) Map() map[string]interface{} {\n\tdata := map[string]interface{}{}\n\n\t// Authorized information\n\tif ctx.Authorized != nil {\n\t\tdata[\"authorized\"] = ctx.Authorized\n\t}\n\tif ctx.ChatID != \"\" {\n\t\tdata[\"chat_id\"] = ctx.ChatID\n\t}\n\tif ctx.AssistantID != \"\" {\n\t\tdata[\"assistant_id\"] = ctx.AssistantID\n\t}\n\n\t// Locale information\n\tif ctx.Locale != \"\" {\n\t\tdata[\"locale\"] = ctx.Locale\n\t}\n\tif ctx.Theme != \"\" {\n\t\tdata[\"theme\"] = ctx.Theme\n\t}\n\n\t// Request information\n\tif ctx.Client.Type != \"\" || ctx.Client.UserAgent != \"\" || ctx.Client.IP != \"\" {\n\t\tdata[\"client\"] = map[string]interface{}{\n\t\t\t\"type\": ctx.Client.Type,\n\t\t\t\"user_agent\": ctx.Client.UserAgent,\n\t\t\t\"ip\": ctx.Client.IP,\n\t\t}\n\t}\n\tif ctx.Referer != \"\" {\n\t\tdata[\"referer\"] = ctx.Referer\n\t}\n\tif ctx.Accept != \"\" {\n\t\tdata[\"accept\"] = ctx.Accept\n\t}\n\n\t// CUI Context information\n\tif ctx.Route != \"\" {\n\t\tdata[\"route\"] = ctx.Route\n\t}\n\tif len(ctx.Metadata) > 0 {\n\t\tdata[\"metadata\"] = ctx.Metadata\n\t}\n\n\treturn data\n}\n\n// Global Registry Functions\n// ===================================\n\n// Register registers a context to the global registry\nfunc Register(ctx *Context) error {\n\tif ctx == nil {\n\t\treturn fmt.Errorf(\"context is nil\")\n\t}\n\n\tif ctx.ID == \"\" {\n\t\treturn fmt.Errorf(\"context ID is empty\")\n\t}\n\n\tcontextRegistry.Store(ctx.ID, ctx)\n\treturn nil\n}\n\n// Unregister removes a context from the global registry\nfunc Unregister(contextID string) {\n\tcontextRegistry.Delete(contextID)\n}\n\n// Get retrieves a context from the global registry by ID\nfunc Get(contextID string) (*Context, error) {\n\tvalue, ok := contextRegistry.Load(contextID)\n\tif !ok {\n\t\treturn nil, fmt.Errorf(\"context not found: %s\", contextID)\n\t}\n\n\tctx, ok := value.(*Context)\n\tif !ok {\n\t\treturn nil, fmt.Errorf(\"invalid context type\")\n\t}\n\n\treturn ctx, nil\n}\n\n// SendInterrupt sends an interrupt signal to a context by ID\n// This is the main entry point for external interrupt requests\nfunc SendInterrupt(contextID string, signal *InterruptSignal) error {\n\tctx, err := Get(contextID)\n\tif err != nil {\n\t\treturn err\n\t}\n\n\tif ctx.Interrupt == nil {\n\t\treturn fmt.Errorf(\"interrupt controller not initialized for context: %s\", contextID)\n\t}\n\n\treturn ctx.Interrupt.SendSignal(signal)\n}\n\n// generateContextID generates a unique context ID\nfunc generateContextID() string {\n\treturn uuid.New().String()\n}\n\n// RequestID returns a unique request ID using NanoID\nfunc (ctx *Context) RequestID() string {\n\treturn ctx.ID\n}\n\n// TraceID returns the trace ID for the context\nfunc (ctx *Context) TraceID() string {\n\tif ctx.Stack != nil {\n\t\treturn ctx.Stack.TraceID\n\t}\n\treturn \"\"\n}\n\n// getMessageMetadata retrieves metadata for a message by ID\n// Returns nil if message metadata is not found\nfunc (ctx *Context) getMessageMetadata(messageID string) *MessageMetadata {\n\tif ctx.messageMetadata == nil {\n\t\treturn nil\n\t}\n\treturn ctx.messageMetadata.getMessage(messageID)\n}\n\n// GetMessageMetadata returns metadata for a message (public version)\nfunc (ctx *Context) GetMessageMetadata(messageID string) *MessageMetadata {\n\treturn ctx.getMessageMetadata(messageID)\n}\n\n// =============================================================================\n// Chat Buffer Methods\n// =============================================================================\n\n// InitBuffer initializes the chat buffer for this context\n// Should be called at the start of Stream() to begin buffering messages and steps\nfunc (ctx *Context) InitBuffer(assistantID, connector, mode string) *ChatBuffer {\n\tctx.Buffer = NewChatBuffer(ctx.ChatID, ctx.RequestID(), assistantID, connector, mode)\n\treturn ctx.Buffer\n}\n\n// HasBuffer returns true if the buffer is initialized\nfunc (ctx *Context) HasBuffer() bool {\n\treturn ctx.Buffer != nil\n}\n\n// BufferUserInput adds user input to the buffer\n// Should be called at the start of Stream() to buffer the user's input message\nfunc (ctx *Context) BufferUserInput(messages []Message) {\n\tif ctx.Buffer == nil {\n\t\treturn\n\t}\n\n\tfor _, msg := range messages {\n\t\tif msg.Role == RoleUser {\n\t\t\t// Get name if available\n\t\t\tvar name string\n\t\t\tif msg.Name != nil {\n\t\t\t\tname = *msg.Name\n\t\t\t}\n\t\t\tctx.Buffer.AddUserInput(msg.Content, name)\n\t\t}\n\t}\n}\n\n// BufferAssistantMessage adds an assistant message to the buffer\n// Called by ctx.Send() to buffer messages for batch saving\nfunc (ctx *Context) BufferAssistantMessage(messageID, msgType string, props map[string]interface{}, blockID, threadID string, metadata map[string]interface{}) {\n\tif ctx.Buffer == nil {\n\t\treturn\n\t}\n\n\tctx.Buffer.AddAssistantMessage(messageID, msgType, props, blockID, threadID, ctx.AssistantID, metadata)\n}\n\n// BeginStep starts tracking a new execution step\n// Returns the step for further updates\nfunc (ctx *Context) BeginStep(stepType string, input map[string]interface{}) *BufferedStep {\n\tif ctx.Buffer == nil {\n\t\treturn nil\n\t}\n\n\t// Update context memory snapshot before starting step (for recovery)\n\tif ctx.Memory != nil && ctx.Memory.Context != nil {\n\t\tctx.Buffer.SetSpaceSnapshot(ctx.Memory.Context.Snapshot())\n\t}\n\n\treturn ctx.Buffer.BeginStep(stepType, input, ctx.Stack)\n}\n\n// CompleteStep marks the current step as completed\nfunc (ctx *Context) CompleteStep(output map[string]interface{}) {\n\tif ctx.Buffer == nil {\n\t\treturn\n\t}\n\tctx.Buffer.CompleteStep(output)\n}\n\n// FailCurrentStep marks the current step as failed or interrupted\nfunc (ctx *Context) FailCurrentStep(status string, err error) {\n\tif ctx.Buffer == nil {\n\t\treturn\n\t}\n\tctx.Buffer.FailCurrentStep(status, err)\n}\n\n// shouldSkipHistory checks if history saving should be skipped\n// Returns true if Skip.History is set in the current stack options\nfunc (ctx *Context) shouldSkipHistory() bool {\n\tif ctx.Stack == nil || ctx.Stack.Options == nil || ctx.Stack.Options.Skip == nil {\n\t\treturn false\n\t}\n\treturn ctx.Stack.Options.Skip.History\n}\n\n// IsA2ACall returns true if this is any Agent-to-Agent call (delegate or fork)\n// A2A calls are identified by Referer being \"agent\" or \"agent_fork\":\n// - ctx.agent.Call/All/Any/Race uses RefererAgentFork (forked context, skips history)\n// - delegate uses RefererAgent (same context flow, saves history)\nfunc (ctx *Context) IsA2ACall() bool {\n\treturn ctx.Referer == RefererAgent || ctx.Referer == RefererAgentFork\n}\n\n// IsForkedA2ACall returns true if this is a forked A2A call (ctx.agent.Call/All/Any/Race)\n// Forked calls use RefererAgentFork, while delegate calls use RefererAgent.\n// This is used to skip history saving for forked sub-agent calls,\n// while allowing delegate calls to save history normally.\nfunc (ctx *Context) IsForkedA2ACall() bool {\n\treturn ctx.Referer == RefererAgentFork\n}\n\n// MergeMetadata merges the given metadata into ctx.Metadata.\n// Existing keys are overwritten by incoming values.\n// This enables A2A callers to pass custom metadata (e.g. oneshot, async)\n// to sub-agent hooks via ctx.metadata in JavaScript.\nfunc (ctx *Context) MergeMetadata(metadata map[string]interface{}) {\n\tif len(metadata) == 0 {\n\t\treturn\n\t}\n\tif ctx.Metadata == nil {\n\t\tctx.Metadata = make(map[string]interface{}, len(metadata))\n\t}\n\tfor k, v := range metadata {\n\t\tctx.Metadata[k] = v\n\t}\n}\n" }, { "path": "agent/context/context_test.go", "content": "package context_test\n\nimport (\n\t\"bytes\"\n\tstdContext \"context\"\n\t\"encoding/json\"\n\t\"net/http\"\n\t\"net/http/httptest\"\n\t\"testing\"\n\n\t\"github.com/gin-gonic/gin\"\n\t\"github.com/stretchr/testify/assert\"\n\t\"github.com/yaoapp/gou/store\"\n\t\"github.com/yaoapp/yao/agent/context\"\n\t\"github.com/yaoapp/yao/config\"\n\t\"github.com/yaoapp/yao/test\"\n)\n\nfunc TestGetCompletionRequest(t *testing.T) {\n\ttest.Prepare(t, config.Conf)\n\tdefer test.Clean()\n\n\tgin.SetMode(gin.TestMode)\n\n\tcache, err := store.Get(\"__yao.agent.cache\")\n\tif err != nil {\n\t\tt.Fatalf(\"Failed to get cache: %v\", err)\n\t}\n\n\ttests := []struct {\n\t\tname string\n\t\trequestBody map[string]interface{}\n\t\tqueryParams map[string]string\n\t\theaders map[string]string\n\t\texpectedModel string\n\t\texpectedMsgCount int\n\t\texpectedTemp *float64\n\t\texpectedStream *bool\n\t\texpectedLocale string\n\t\texpectedTheme string\n\t\texpectedReferer string\n\t\texpectedAccept context.Accept\n\t\texpectedAssistantID string\n\t\texpectError bool\n\t}{\n\t\t{\n\t\t\tname: \"Complete request from body with metadata\",\n\t\t\trequestBody: map[string]interface{}{\n\t\t\t\t\"model\": \"gpt-4-yao_assistant123\",\n\t\t\t\t\"messages\": []map[string]interface{}{\n\t\t\t\t\t{\"role\": \"user\", \"content\": \"Hello\"},\n\t\t\t\t},\n\t\t\t\t\"temperature\": 0.7,\n\t\t\t\t\"stream\": true,\n\t\t\t\t\"metadata\": map[string]string{\n\t\t\t\t\t\"locale\": \"zh-cn\",\n\t\t\t\t\t\"theme\": \"dark\",\n\t\t\t\t\t\"referer\": \"process\",\n\t\t\t\t\t\"accept\": \"cui-web\",\n\t\t\t\t\t\"chat_id\": \"chat-from-metadata\",\n\t\t\t\t},\n\t\t\t},\n\t\t\texpectedModel: \"gpt-4-yao_assistant123\",\n\t\t\texpectedMsgCount: 1,\n\t\t\texpectedTemp: floatPtr(0.7),\n\t\t\texpectedStream: boolPtr(true),\n\t\t\texpectedLocale: \"zh-cn\",\n\t\t\texpectedTheme: \"dark\",\n\t\t\texpectedReferer: context.RefererProcess,\n\t\t\texpectedAccept: context.AcceptWebCUI,\n\t\t\texpectedAssistantID: \"assistant123\",\n\t\t\texpectError: false,\n\t\t},\n\t\t{\n\t\t\tname: \"Query params override payload metadata\",\n\t\t\trequestBody: map[string]interface{}{\n\t\t\t\t\"model\": \"gpt-4-yao_test456\",\n\t\t\t\t\"messages\": []map[string]interface{}{\n\t\t\t\t\t{\"role\": \"user\", \"content\": \"Test\"},\n\t\t\t\t},\n\t\t\t\t\"metadata\": map[string]string{\n\t\t\t\t\t\"locale\": \"en-us\",\n\t\t\t\t\t\"theme\": \"light\",\n\t\t\t\t},\n\t\t\t},\n\t\t\tqueryParams: map[string]string{\n\t\t\t\t\"locale\": \"fr-FR\",\n\t\t\t\t\"theme\": \"auto\",\n\t\t\t},\n\t\t\texpectedModel: \"gpt-4-yao_test456\",\n\t\t\texpectedMsgCount: 1,\n\t\t\texpectedLocale: \"fr-fr\",\n\t\t\texpectedTheme: \"auto\",\n\t\t\texpectedReferer: context.RefererAPI,\n\t\t\texpectedAccept: context.AcceptStandard,\n\t\t\texpectedAssistantID: \"test456\",\n\t\t\texpectError: false,\n\t\t},\n\t\t{\n\t\t\tname: \"Headers override payload metadata\",\n\t\t\trequestBody: map[string]interface{}{\n\t\t\t\t\"model\": \"gpt-3.5-turbo-yao_header789\",\n\t\t\t\t\"messages\": []map[string]interface{}{\n\t\t\t\t\t{\"role\": \"user\", \"content\": \"Test\"},\n\t\t\t\t},\n\t\t\t\t\"metadata\": map[string]string{\n\t\t\t\t\t\"referer\": \"process\",\n\t\t\t\t\t\"accept\": \"cui-web\",\n\t\t\t\t},\n\t\t\t},\n\t\t\theaders: map[string]string{\n\t\t\t\t\"X-Yao-Referer\": \"mcp\",\n\t\t\t\t\"X-Yao-Accept\": \"cui-desktop\",\n\t\t\t},\n\t\t\texpectedModel: \"gpt-3.5-turbo-yao_header789\",\n\t\t\texpectedMsgCount: 1,\n\t\t\texpectedLocale: \"\",\n\t\t\texpectedTheme: \"\",\n\t\t\texpectedReferer: context.RefererMCP,\n\t\t\texpectedAccept: context.AcceptDesktopCUI,\n\t\t\texpectedAssistantID: \"header789\",\n\t\t\texpectError: false,\n\t\t},\n\t\t{\n\t\t\tname: \"Minimal request without metadata\",\n\t\t\trequestBody: map[string]interface{}{\n\t\t\t\t\"model\": \"gpt-4o-yao_minimal\",\n\t\t\t\t\"messages\": []map[string]interface{}{\n\t\t\t\t\t{\"role\": \"user\", \"content\": \"Hello\"},\n\t\t\t\t},\n\t\t\t},\n\t\t\texpectedModel: \"gpt-4o-yao_minimal\",\n\t\t\texpectedMsgCount: 1,\n\t\t\texpectedLocale: \"\",\n\t\t\texpectedTheme: \"\",\n\t\t\texpectedReferer: context.RefererAPI,\n\t\t\texpectedAccept: context.AcceptStandard,\n\t\t\texpectedAssistantID: \"minimal\",\n\t\t\texpectError: false,\n\t\t},\n\t\t{\n\t\t\tname: \"Missing model\",\n\t\t\trequestBody: map[string]interface{}{\n\t\t\t\t\"messages\": []map[string]interface{}{\n\t\t\t\t\t{\"role\": \"user\", \"content\": \"Hello\"},\n\t\t\t\t},\n\t\t\t},\n\t\t\texpectError: true,\n\t\t},\n\t\t{\n\t\t\tname: \"Missing messages\",\n\t\t\trequestBody: map[string]interface{}{\n\t\t\t\t\"model\": \"gpt-4\",\n\t\t\t},\n\t\t\texpectError: true,\n\t\t},\n\t}\n\n\tfor _, tt := range tests {\n\t\tt.Run(tt.name, func(t *testing.T) {\n\t\t\tw := httptest.NewRecorder()\n\t\t\tc, _ := gin.CreateTestContext(w)\n\n\t\t\t// Build request\n\t\t\tbodyBytes, _ := json.Marshal(tt.requestBody)\n\t\t\treq, _ := http.NewRequest(\"POST\", \"http://example.com/chat/completions\", bytes.NewBuffer(bodyBytes))\n\t\t\treq.Header.Set(\"Content-Type\", \"application/json\")\n\n\t\t\t// Add query params\n\t\t\tq := req.URL.Query()\n\t\t\tfor key, value := range tt.queryParams {\n\t\t\t\tq.Add(key, value)\n\t\t\t}\n\t\t\treq.URL.RawQuery = q.Encode()\n\n\t\t\t// Add headers\n\t\t\tfor key, value := range tt.headers {\n\t\t\t\treq.Header.Set(key, value)\n\t\t\t}\n\n\t\t\tc.Request = req\n\n\t\t\t// Call GetCompletionRequest\n\t\t\tcompletionReq, ctx, opts, err := context.GetCompletionRequest(c, cache)\n\n\t\t\tif tt.expectError {\n\t\t\t\tassert.Error(t, err)\n\t\t\t\treturn\n\t\t\t}\n\n\t\t\tassert.NoError(t, err)\n\t\t\tassert.NotNil(t, completionReq)\n\t\t\tassert.NotNil(t, ctx)\n\t\t\tassert.NotNil(t, opts)\n\n\t\t\t// Verify CompletionRequest\n\t\t\tassert.Equal(t, tt.expectedModel, completionReq.Model)\n\t\t\tassert.Equal(t, tt.expectedMsgCount, len(completionReq.Messages))\n\t\t\tif tt.expectedTemp != nil {\n\t\t\t\tassert.NotNil(t, completionReq.Temperature)\n\t\t\t\tassert.Equal(t, *tt.expectedTemp, *completionReq.Temperature)\n\t\t\t}\n\t\t\tif tt.expectedStream != nil {\n\t\t\t\tassert.NotNil(t, completionReq.Stream)\n\t\t\t\tassert.Equal(t, *tt.expectedStream, *completionReq.Stream)\n\t\t\t}\n\n\t\t\t// Verify Context\n\t\t\tassert.Equal(t, tt.expectedLocale, ctx.Locale)\n\t\t\tassert.Equal(t, tt.expectedTheme, ctx.Theme)\n\t\t\tassert.Equal(t, tt.expectedReferer, ctx.Referer)\n\t\t\tassert.Equal(t, tt.expectedAccept, ctx.Accept)\n\t\t\tassert.Equal(t, tt.expectedAssistantID, ctx.AssistantID)\n\t\t\tassert.NotNil(t, ctx.Memory)\n\t\t\tassert.NotNil(t, ctx.Cache)\n\t\t})\n\t}\n}\n\nfunc TestContextNew_WithAuthorized(t *testing.T) {\n\ttest.Prepare(t, config.Conf)\n\tdefer test.Clean()\n\n\t// Create context using New()\n\tctx := context.New(stdContext.Background(), nil, \"test-chat-id\")\n\tdefer ctx.Release()\n\n\tassert.NotNil(t, ctx)\n\tassert.Equal(t, \"test-chat-id\", ctx.ChatID)\n\tassert.NotNil(t, ctx.Memory)\n\tassert.NotNil(t, ctx.IDGenerator)\n}\n\n// Helper functions for context_test package\nfunc floatPtr(f float64) *float64 {\n\treturn &f\n}\n\nfunc boolPtr(b bool) *bool {\n\treturn &b\n}\n" }, { "path": "agent/context/grpc.go", "content": "package context\n\nimport (\n\t\"context\"\n\t\"encoding/json\"\n\t\"fmt\"\n\t\"net/http\"\n\n\t\"github.com/yaoapp/gou/connector\"\n\t\"github.com/yaoapp/gou/store\"\n\t\"github.com/yaoapp/yao/openapi/oauth/types\"\n)\n\n// GRPCAgentInput holds the raw inputs from a gRPC AgentStream request.\ntype GRPCAgentInput struct {\n\tAssistantID string\n\tMessages []byte\n\tOptions []byte\n\tAuthInfo *types.AuthorizedInfo\n\tCache store.Store\n\tWriter http.ResponseWriter\n}\n\n// GetGRPCAgentRequest parses a gRPC agent request and creates a Context + Options,\n// mirroring openapi.go GetCompletionRequest.\n//\n// Flow: validate → parse messages → parse options → build Context → build Options → register interrupt\nfunc GetGRPCAgentRequest(parent context.Context, input GRPCAgentInput) ([]Message, *Context, *Options, error) {\n\tif input.AssistantID == \"\" {\n\t\treturn nil, nil, nil, fmt.Errorf(\"assistant_id is required\")\n\t}\n\n\tmessages, err := parseGRPCMessages(input.Messages)\n\tif err != nil {\n\t\treturn nil, nil, nil, err\n\t}\n\n\tvar rawOpts map[string]interface{}\n\tif len(input.Options) > 0 {\n\t\tif err := json.Unmarshal(input.Options, &rawOpts); err != nil {\n\t\t\treturn nil, nil, nil, fmt.Errorf(\"invalid options JSON: %w\", err)\n\t\t}\n\t}\n\n\tchatID := getChatIDFromOpts(rawOpts)\n\tctx := New(parent, input.AuthInfo, chatID)\n\n\tctx.Cache = input.Cache\n\tctx.Writer = input.Writer\n\tctx.AssistantID = input.AssistantID\n\tctx.Locale = getStringOpt(rawOpts, \"locale\")\n\tctx.Theme = getStringOpt(rawOpts, \"theme\")\n\tctx.Referer = getRefererOpt(rawOpts)\n\tctx.Accept = getAcceptOpt(rawOpts)\n\tctx.Route = getStringOpt(rawOpts, \"route\")\n\tctx.Metadata = getMapOpt(rawOpts, \"metadata\")\n\tctx.Client = Client{Type: \"grpc\"}\n\n\topts := &Options{\n\t\tContext: parent,\n\t\tSkip: getSkipOpt(rawOpts),\n\t\tMode: getStringOpt(rawOpts, \"mode\"),\n\t}\n\n\tif connectorID := getStringOpt(rawOpts, \"connector\"); connectorID != \"\" {\n\t\tif _, err := connector.Select(connectorID); err == nil {\n\t\t\topts.Connector = connectorID\n\t\t}\n\t}\n\n\tctx.Interrupt = NewInterruptController()\n\tif err := Register(ctx); err != nil {\n\t\treturn nil, nil, nil, fmt.Errorf(\"failed to register context: %w\", err)\n\t}\n\tctx.Interrupt.Start(ctx.ID)\n\n\treturn messages, ctx, opts, nil\n}\n\nfunc parseGRPCMessages(raw []byte) ([]Message, error) {\n\tif len(raw) == 0 {\n\t\treturn nil, fmt.Errorf(\"messages are required\")\n\t}\n\tvar messages []Message\n\tif err := json.Unmarshal(raw, &messages); err != nil {\n\t\treturn nil, fmt.Errorf(\"invalid messages JSON: %w\", err)\n\t}\n\tif len(messages) == 0 {\n\t\treturn nil, fmt.Errorf(\"messages must not be empty\")\n\t}\n\treturn messages, nil\n}\n\nfunc getChatIDFromOpts(opts map[string]interface{}) string {\n\tif opts != nil {\n\t\tif v, ok := opts[\"chat_id\"].(string); ok && v != \"\" {\n\t\t\treturn v\n\t\t}\n\t}\n\treturn GenChatID()\n}\n\nfunc getStringOpt(opts map[string]interface{}, key string) string {\n\tif opts == nil {\n\t\treturn \"\"\n\t}\n\tv, _ := opts[key].(string)\n\treturn v\n}\n\nfunc getRefererOpt(opts map[string]interface{}) string {\n\tr := getStringOpt(opts, \"referer\")\n\tif r != \"\" {\n\t\treturn validateReferer(r)\n\t}\n\treturn RefererAPI\n}\n\nfunc getAcceptOpt(opts map[string]interface{}) Accept {\n\ta := getStringOpt(opts, \"accept\")\n\tif a != \"\" {\n\t\treturn validateAccept(a)\n\t}\n\treturn AcceptStandard\n}\n\nfunc getMapOpt(opts map[string]interface{}, key string) map[string]interface{} {\n\tif opts == nil {\n\t\treturn nil\n\t}\n\tv, _ := opts[key].(map[string]interface{})\n\treturn v\n}\n\nfunc getSkipOpt(opts map[string]interface{}) *Skip {\n\tif opts == nil {\n\t\treturn nil\n\t}\n\traw, ok := opts[\"skip\"]\n\tif !ok {\n\t\treturn nil\n\t}\n\n\tdata, err := json.Marshal(raw)\n\tif err != nil {\n\t\treturn nil\n\t}\n\tvar skip Skip\n\tif err := json.Unmarshal(data, &skip); err != nil {\n\t\treturn nil\n\t}\n\treturn &skip\n}\n" }, { "path": "agent/context/interfaces.go", "content": "package context\n\nimport (\n\t\"net/http\"\n\n\t\"github.com/yaoapp/yao/agent/output/message\"\n)\n\n// StreamChunkType represents the type of content in a streaming chunk\ntype StreamChunkType string\n\n// Stream chunk type constants - indicates what type of content is in the current chunk\nconst (\n\t// Content chunk types - actual data from the LLM\n\tChunkText StreamChunkType = \"text\" // Regular text content\n\tChunkThinking StreamChunkType = \"thinking\" // Reasoning/thinking content (o1, DeepSeek R1)\n\tChunkToolCall StreamChunkType = \"tool_call\" // Tool/function call\n\tChunkRefusal StreamChunkType = \"refusal\" // Model refusal\n\tChunkMetadata StreamChunkType = \"metadata\" // Metadata (usage, finish_reason, etc.)\n\tChunkError StreamChunkType = \"error\" // Error chunk\n\tChunkUnknown StreamChunkType = \"unknown\" // Unknown/unrecognized chunk type\n\n\t// Lifecycle event types - stream and message boundaries\n\tChunkStreamStart StreamChunkType = \"stream_start\" // Stream begins (entire request starts)\n\tChunkStreamEnd StreamChunkType = \"stream_end\" // Stream ends (entire request completes)\n\tChunkMessageStart StreamChunkType = \"message_start\" // Message begins (text/tool_call/thinking message starts)\n\tChunkMessageEnd StreamChunkType = \"message_end\" // Message ends (text/tool_call/thinking message completes)\n)\n\n// Writer is an alias for http.ResponseWriter interface used by an agent to construct a response.\n// A Writer may not be used after the agent execution has completed.\ntype Writer = http.ResponseWriter\n\n// Agent the agent interface\ntype Agent interface {\n\n\t// Stream stream the agent\n\tStream(ctx *Context, messages []Message, handler message.StreamFunc) error\n\n\t// Run run the agent\n\tRun(ctx *Context, messages []Message) (*Response, error)\n}\n" }, { "path": "agent/context/interrupt.go", "content": "package context\n\nimport (\n\t\"context\"\n\t\"fmt\"\n\t\"time\"\n\n\t\"github.com/yaoapp/kun/log\"\n)\n\n// NewInterruptController creates a new interrupt controller\nfunc NewInterruptController() *InterruptController {\n\tctrl := &InterruptController{\n\t\tqueue: make(chan *InterruptSignal, 10), // Buffer for 10 interrupts\n\t\tpending: make([]*InterruptSignal, 0),\n\t}\n\tctrl.ctx, ctrl.cancel = context.WithCancel(context.Background())\n\treturn ctrl\n}\n\n// Start starts the interrupt listener goroutine\nfunc (ic *InterruptController) Start(contextID string) {\n\tif ic.listenerStarted {\n\t\treturn\n\t}\n\n\tic.mutex.Lock()\n\tic.listenerStarted = true\n\tic.contextID = contextID\n\tic.mutex.Unlock()\n\n\tgo ic.listen()\n}\n\n// SetHandler sets the handler for interrupt signals\nfunc (ic *InterruptController) SetHandler(handler InterruptHandler) {\n\tif ic == nil {\n\t\treturn\n\t}\n\tic.handler = handler\n}\n\n// listen is the main listener goroutine that processes interrupt signals\nfunc (ic *InterruptController) listen() {\n\tfor {\n\t\tselect {\n\t\tcase signal := <-ic.queue:\n\t\t\t// Handle user interrupt signal (stop button, for appending messages)\n\t\t\tic.handleSignal(signal)\n\n\t\tcase <-ic.ctx.Done():\n\t\t\t// Internal context cancelled, stop listening\n\t\t\treturn\n\t\t}\n\t}\n}\n\n// handleSignal processes an interrupt signal\nfunc (ic *InterruptController) handleSignal(signal *InterruptSignal) {\n\tif signal == nil {\n\t\treturn\n\t}\n\n\tlog.Trace(\"[INTERRUPT] Signal received: type=%s, messages=%d, timestamp=%d\", signal.Type, len(signal.Messages), signal.Timestamp)\n\n\tic.mutex.Lock()\n\n\t// If no current interrupt, set it as current\n\tif ic.current == nil {\n\t\tic.current = signal\n\t} else {\n\t\t// If there's already a current interrupt, add to pending queue\n\t\tic.pending = append(ic.pending, signal)\n\t}\n\n\t// For force interrupt with no messages (pure cancellation), cancel the interrupt context\n\t// This allows LLM streaming and other operations to check and stop\n\tif signal.Type == InterruptForce && len(signal.Messages) == 0 {\n\t\tif ic.cancel != nil {\n\t\t\tic.cancel()\n\t\t\t// Create a new context for potential future operations\n\t\t\tic.ctx, ic.cancel = context.WithCancel(context.Background())\n\t\t}\n\t}\n\n\tic.mutex.Unlock()\n\n\t// Call the registered handler if available (outside lock to avoid deadlock)\n\tif ic.handler != nil && ic.contextID != \"\" {\n\t\tgo func() {\n\t\t\t// Retrieve the parent context from global registry\n\t\t\tctx, err := Get(ic.contextID)\n\t\t\tif err != nil {\n\t\t\t\tfmt.Printf(\"Failed to get context for interrupt handler: %v\\n\", err)\n\t\t\t\treturn\n\t\t\t}\n\n\t\t\t// Call the handler\n\t\t\tif err := ic.handler(ctx, signal); err != nil {\n\t\t\t\tfmt.Printf(\"Interrupt handler error: %v\\n\", err)\n\t\t\t}\n\t\t}()\n\t}\n}\n\n// Check checks for current interrupt signal (non-blocking)\n// Returns the current interrupt and moves to next one if available\nfunc (ic *InterruptController) Check() *InterruptSignal {\n\tif ic == nil {\n\t\treturn nil\n\t}\n\n\tic.mutex.Lock()\n\tdefer ic.mutex.Unlock()\n\n\tif ic.current == nil {\n\t\treturn nil\n\t}\n\n\t// Get current interrupt\n\tsignal := ic.current\n\n\t// Move to next interrupt in queue\n\tif len(ic.pending) > 0 {\n\t\tic.current = ic.pending[0]\n\t\tic.pending = ic.pending[1:]\n\t} else {\n\t\tic.current = nil\n\t}\n\n\treturn signal\n}\n\n// CheckWithMerge checks for interrupts and merges all pending messages\n// This is useful when multiple interrupts should be handled together\nfunc (ic *InterruptController) CheckWithMerge() *InterruptSignal {\n\tif ic == nil {\n\t\treturn nil\n\t}\n\n\tic.mutex.Lock()\n\tdefer ic.mutex.Unlock()\n\n\tif ic.current == nil {\n\t\treturn nil\n\t}\n\n\t// If there are pending interrupts, merge all messages\n\tif len(ic.pending) > 0 {\n\t\t// Collect all messages\n\t\tallMessages := append([]Message{}, ic.current.Messages...)\n\t\tfor _, pending := range ic.pending {\n\t\t\tallMessages = append(allMessages, pending.Messages...)\n\t\t}\n\n\t\t// Create merged signal\n\t\tmergedSignal := &InterruptSignal{\n\t\t\tType: ic.current.Type, // Use first signal's type\n\t\t\tMessages: allMessages,\n\t\t\tTimestamp: time.Now().UnixMilli(),\n\t\t\tMetadata: map[string]interface{}{\n\t\t\t\t\"merged\": true,\n\t\t\t\t\"merged_count\": len(ic.pending) + 1,\n\t\t\t\t\"original_time\": ic.current.Timestamp,\n\t\t\t},\n\t\t}\n\n\t\t// Clear all interrupts\n\t\tic.current = nil\n\t\tic.pending = make([]*InterruptSignal, 0)\n\n\t\treturn mergedSignal\n\t}\n\n\t// No pending interrupts, return current\n\tsignal := ic.current\n\tic.current = nil\n\treturn signal\n}\n\n// Peek returns the current interrupt without removing it\nfunc (ic *InterruptController) Peek() *InterruptSignal {\n\tif ic == nil {\n\t\treturn nil\n\t}\n\n\tic.mutex.RLock()\n\tdefer ic.mutex.RUnlock()\n\n\treturn ic.current\n}\n\n// IsInterrupted checks if interrupt context is cancelled (force interrupt)\nfunc (ic *InterruptController) IsInterrupted() bool {\n\tif ic == nil || ic.ctx == nil {\n\t\treturn false\n\t}\n\n\tselect {\n\tcase <-ic.ctx.Done():\n\t\treturn true\n\tdefault:\n\t\treturn false\n\t}\n}\n\n// Context returns the interrupt control context\n// This can be used in select statements to check for force interrupts\nfunc (ic *InterruptController) Context() context.Context {\n\tif ic == nil {\n\t\treturn context.Background()\n\t}\n\treturn ic.ctx\n}\n\n// GetPendingCount returns the number of pending interrupts\nfunc (ic *InterruptController) GetPendingCount() int {\n\tif ic == nil {\n\t\treturn 0\n\t}\n\n\tic.mutex.RLock()\n\tdefer ic.mutex.RUnlock()\n\n\tcount := len(ic.pending)\n\tif ic.current != nil {\n\t\tcount++\n\t}\n\treturn count\n}\n\n// Clear clears all interrupts (current and pending)\nfunc (ic *InterruptController) Clear() {\n\tif ic == nil {\n\t\treturn\n\t}\n\n\tic.mutex.Lock()\n\tdefer ic.mutex.Unlock()\n\n\tic.current = nil\n\tic.pending = make([]*InterruptSignal, 0)\n}\n\n// Stop stops the interrupt controller and cleans up resources\nfunc (ic *InterruptController) Stop() {\n\tif ic == nil {\n\t\treturn\n\t}\n\n\t// Cancel context to stop listener\n\tif ic.cancel != nil {\n\t\tic.cancel()\n\t}\n\n\t// Close channel\n\tif ic.queue != nil {\n\t\tclose(ic.queue)\n\t}\n\n\t// Clear interrupts\n\tic.Clear()\n}\n\n// SendSignal sends an interrupt signal to the controller\n// This is called from external sources (e.g., another HTTP request)\nfunc (ic *InterruptController) SendSignal(signal *InterruptSignal) error {\n\tif ic == nil {\n\t\treturn fmt.Errorf(\"interrupt controller is nil\")\n\t}\n\n\tif ic.queue == nil {\n\t\treturn fmt.Errorf(\"interrupt queue is not initialized\")\n\t}\n\n\t// Non-blocking send\n\tselect {\n\tcase ic.queue <- signal:\n\t\treturn nil\n\tcase <-time.After(500 * time.Millisecond):\n\t\treturn fmt.Errorf(\"failed to send interrupt: timeout\")\n\t}\n}\n" }, { "path": "agent/context/interrupt_test.go", "content": "package context_test\n\nimport (\n\tstdContext \"context\"\n\t\"fmt\"\n\t\"testing\"\n\t\"time\"\n\n\t\"github.com/yaoapp/yao/agent/context\"\n\t\"github.com/yaoapp/yao/openapi/oauth/types\"\n)\n\n// newTestContextWithInterrupt creates a Context with interrupt controller for testing\nfunc newTestContextWithInterrupt(chatID, assistantID string) *context.Context {\n\tctx := context.New(stdContext.Background(), &types.AuthorizedInfo{\n\t\tSubject: \"test-user\",\n\t\tClientID: \"test-client-id\",\n\t\tUserID: \"test-user-123\",\n\t\tTeamID: \"test-team-456\",\n\t\tTenantID: \"test-tenant-789\",\n\t\tSessionID: \"test-session-id\",\n\t}, chatID)\n\n\tctx.AssistantID = assistantID\n\tctx.Locale = \"en-us\"\n\tctx.Theme = \"light\"\n\tctx.Client = context.Client{\n\t\tType: \"web\",\n\t\tUserAgent: \"TestAgent/1.0\",\n\t\tIP: \"127.0.0.1\",\n\t}\n\tctx.Referer = context.RefererAPI\n\tctx.Accept = context.AcceptWebCUI\n\tctx.Route = \"/test/route\"\n\tctx.Metadata = map[string]interface{}{\n\t\t\"test\": \"context_metadata\",\n\t}\n\n\t// Initialize interrupt controller\n\tctx.Interrupt = context.NewInterruptController()\n\n\t// Register context globally\n\tif err := context.Register(ctx); err != nil {\n\t\tpanic(fmt.Sprintf(\"Failed to register context: %v\", err))\n\t}\n\n\t// Start interrupt listener\n\tctx.Interrupt.Start(ctx.ID)\n\n\treturn ctx\n}\n\n// TestInterruptBasic tests basic interrupt signal sending and receiving\nfunc TestInterruptBasic(t *testing.T) {\n\t// Create context with interrupt support\n\tctx := newTestContextWithInterrupt(\"chat-test-interrupt\", \"test-assistant\")\n\tdefer ctx.Release()\n\n\tt.Run(\"SendGracefulInterrupt\", func(t *testing.T) {\n\t\t// Create a graceful interrupt signal\n\t\tsignal := &context.InterruptSignal{\n\t\t\tType: context.InterruptGraceful,\n\t\t\tMessages: []context.Message{\n\t\t\t\t{Role: context.RoleUser, Content: \"This is a graceful interrupt\"},\n\t\t\t},\n\t\t\tTimestamp: time.Now().UnixMilli(),\n\t\t}\n\n\t\t// Send interrupt signal\n\t\terr := context.SendInterrupt(ctx.ID, signal)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to send interrupt signal: %v\", err)\n\t\t}\n\n\t\t// Wait a bit for the signal to be processed\n\t\ttime.Sleep(100 * time.Millisecond)\n\n\t\t// Check if signal was received\n\t\treceivedSignal := ctx.Interrupt.Peek()\n\t\tif receivedSignal == nil {\n\t\t\tt.Fatal(\"Expected interrupt signal to be received, got nil\")\n\t\t}\n\n\t\tif receivedSignal.Type != context.InterruptGraceful {\n\t\t\tt.Errorf(\"Expected interrupt type 'graceful', got: %s\", receivedSignal.Type)\n\t\t}\n\n\t\tif len(receivedSignal.Messages) != 1 {\n\t\t\tt.Errorf(\"Expected 1 message, got: %d\", len(receivedSignal.Messages))\n\t\t}\n\n\t\tif receivedSignal.Messages[0].Content != \"This is a graceful interrupt\" {\n\t\t\tt.Errorf(\"Expected message content 'This is a graceful interrupt', got: %s\", receivedSignal.Messages[0].Content)\n\t\t}\n\n\t\tt.Log(\"✓ Graceful interrupt signal sent and received successfully\")\n\t})\n\n\tt.Run(\"SendForceInterrupt\", func(t *testing.T) {\n\t\t// Clear previous signals\n\t\tctx.Interrupt.Clear()\n\n\t\t// Create a force interrupt signal\n\t\tsignal := &context.InterruptSignal{\n\t\t\tType: context.InterruptForce,\n\t\t\tMessages: []context.Message{\n\t\t\t\t{Role: context.RoleUser, Content: \"This is a force interrupt\"},\n\t\t\t},\n\t\t\tTimestamp: time.Now().UnixMilli(),\n\t\t}\n\n\t\t// Send interrupt signal\n\t\terr := context.SendInterrupt(ctx.ID, signal)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to send interrupt signal: %v\", err)\n\t\t}\n\n\t\t// Wait a bit for the signal to be processed\n\t\ttime.Sleep(100 * time.Millisecond)\n\n\t\t// Check if signal was received\n\t\treceivedSignal := ctx.Interrupt.Peek()\n\t\tif receivedSignal == nil {\n\t\t\tt.Fatal(\"Expected interrupt signal to be received, got nil\")\n\t\t}\n\n\t\tif receivedSignal.Type != context.InterruptForce {\n\t\t\tt.Errorf(\"Expected interrupt type 'force', got: %s\", receivedSignal.Type)\n\t\t}\n\n\t\tt.Log(\"✓ Force interrupt signal sent and received successfully\")\n\t})\n\n\tt.Run(\"MultipleInterrupts\", func(t *testing.T) {\n\t\t// Clear previous signals\n\t\tctx.Interrupt.Clear()\n\n\t\t// Send multiple interrupt signals\n\t\tfor i := 0; i < 3; i++ {\n\t\t\tsignal := &context.InterruptSignal{\n\t\t\t\tType: context.InterruptGraceful,\n\t\t\t\tMessages: []context.Message{\n\t\t\t\t\t{Role: context.RoleUser, Content: fmt.Sprintf(\"Message %d\", i+1)},\n\t\t\t\t},\n\t\t\t\tTimestamp: time.Now().UnixMilli(),\n\t\t\t}\n\n\t\t\terr := context.SendInterrupt(ctx.ID, signal)\n\t\t\tif err != nil {\n\t\t\t\tt.Fatalf(\"Failed to send interrupt signal %d: %v\", i+1, err)\n\t\t\t}\n\t\t}\n\n\t\t// Wait a bit for signals to be processed\n\t\ttime.Sleep(100 * time.Millisecond)\n\n\t\t// Check pending count\n\t\tpendingCount := ctx.Interrupt.GetPendingCount()\n\t\tif pendingCount != 3 {\n\t\t\tt.Errorf(\"Expected 3 pending interrupts, got: %d\", pendingCount)\n\t\t}\n\n\t\t// Check merged signal\n\t\tmergedSignal := ctx.Interrupt.CheckWithMerge()\n\t\tif mergedSignal == nil {\n\t\t\tt.Fatal(\"Expected merged signal, got nil\")\n\t\t}\n\n\t\tif len(mergedSignal.Messages) != 3 {\n\t\t\tt.Errorf(\"Expected 3 merged messages, got: %d\", len(mergedSignal.Messages))\n\t\t}\n\n\t\t// Verify all messages are present\n\t\tfor i := 0; i < 3; i++ {\n\t\t\texpectedContent := fmt.Sprintf(\"Message %d\", i+1)\n\t\t\tif mergedSignal.Messages[i].Content != expectedContent {\n\t\t\t\tt.Errorf(\"Expected message %d content '%s', got: %s\", i+1, expectedContent, mergedSignal.Messages[i].Content)\n\t\t\t}\n\t\t}\n\n\t\tt.Log(\"✓ Multiple interrupt signals merged successfully\")\n\t})\n}\n\n// TestInterruptHandler tests interrupt handler invocation\nfunc TestInterruptHandler(t *testing.T) {\n\t// Create context with interrupt support\n\tctx := newTestContextWithInterrupt(\"chat-test-interrupt-handler\", \"test-assistant\")\n\tdefer ctx.Release()\n\n\tt.Run(\"HandlerInvocation\", func(t *testing.T) {\n\t\t// Track if handler was called\n\t\thandlerCalled := false\n\t\tvar receivedSignal *context.InterruptSignal\n\n\t\t// Set up handler\n\t\tctx.Interrupt.SetHandler(func(c *context.Context, signal *context.InterruptSignal) error {\n\t\t\thandlerCalled = true\n\t\t\treceivedSignal = signal\n\t\t\tt.Logf(\"Handler called with signal type: %s, messages: %d\", signal.Type, len(signal.Messages))\n\t\t\treturn nil\n\t\t})\n\n\t\t// Send interrupt signal\n\t\tsignal := &context.InterruptSignal{\n\t\t\tType: context.InterruptGraceful,\n\t\t\tMessages: []context.Message{\n\t\t\t\t{Role: context.RoleUser, Content: \"Test handler invocation\"},\n\t\t\t},\n\t\t\tTimestamp: time.Now().UnixMilli(),\n\t\t}\n\n\t\terr := context.SendInterrupt(ctx.ID, signal)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to send interrupt signal: %v\", err)\n\t\t}\n\n\t\t// Wait for handler to be called\n\t\ttime.Sleep(200 * time.Millisecond)\n\n\t\t// Verify handler was called\n\t\tif !handlerCalled {\n\t\t\tt.Error(\"Expected handler to be called, but it wasn't\")\n\t\t}\n\n\t\tif receivedSignal == nil {\n\t\t\tt.Fatal(\"Expected signal in handler, got nil\")\n\t\t}\n\n\t\tif receivedSignal.Type != context.InterruptGraceful {\n\t\t\tt.Errorf(\"Expected graceful interrupt in handler, got: %s\", receivedSignal.Type)\n\t\t}\n\n\t\tif len(receivedSignal.Messages) != 1 {\n\t\t\tt.Errorf(\"Expected 1 message in handler, got: %d\", len(receivedSignal.Messages))\n\t\t}\n\n\t\tt.Log(\"✓ Interrupt handler invoked successfully\")\n\t})\n\n\tt.Run(\"HandlerWithError\", func(t *testing.T) {\n\t\t// Create new context\n\t\tctx2 := newTestContextWithInterrupt(\"chat-test-handler-error\", \"test-assistant\")\n\t\tdefer ctx2.Release()\n\n\t\t// Set up handler that returns error\n\t\thandlerCalled := false\n\t\tctx2.Interrupt.SetHandler(func(c *context.Context, signal *context.InterruptSignal) error {\n\t\t\thandlerCalled = true\n\t\t\treturn fmt.Errorf(\"test error from handler\")\n\t\t})\n\n\t\t// Send interrupt signal\n\t\tsignal := &context.InterruptSignal{\n\t\t\tType: context.InterruptForce,\n\t\t\tMessages: []context.Message{\n\t\t\t\t{Role: context.RoleUser, Content: \"Test error handling\"},\n\t\t\t},\n\t\t\tTimestamp: time.Now().UnixMilli(),\n\t\t}\n\n\t\terr := context.SendInterrupt(ctx2.ID, signal)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to send interrupt signal: %v\", err)\n\t\t}\n\n\t\t// Wait for handler to be called\n\t\ttime.Sleep(200 * time.Millisecond)\n\n\t\t// Handler should still be called even if it returns error\n\t\tif !handlerCalled {\n\t\t\tt.Error(\"Expected handler to be called even with error\")\n\t\t}\n\n\t\tt.Log(\"✓ Handler error handling works correctly\")\n\t})\n}\n\n// TestInterruptContextLifecycle tests context registration and cleanup\nfunc TestInterruptContextLifecycle(t *testing.T) {\n\tt.Run(\"RegisterAndRetrieve\", func(t *testing.T) {\n\t\tctx := newTestContextWithInterrupt(\"chat-test-lifecycle\", \"test-assistant\")\n\n\t\t// Verify context can be retrieved\n\t\tretrievedCtx, err := context.Get(ctx.ID)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to retrieve context: %v\", err)\n\t\t}\n\n\t\tif retrievedCtx.ID != ctx.ID {\n\t\t\tt.Errorf(\"Expected context ID %s, got: %s\", ctx.ID, retrievedCtx.ID)\n\t\t}\n\n\t\tctx.Release()\n\n\t\t// After release, context should be removed\n\t\t_, err = context.Get(ctx.ID)\n\t\tif err == nil {\n\t\t\tt.Error(\"Expected error when retrieving released context\")\n\t\t}\n\n\t\tt.Log(\"✓ Context registration and cleanup works correctly\")\n\t})\n\n\tt.Run(\"SendToNonExistentContext\", func(t *testing.T) {\n\t\tsignal := &context.InterruptSignal{\n\t\t\tType: context.InterruptGraceful,\n\t\t\tMessages: []context.Message{{Role: context.RoleUser, Content: \"test\"}},\n\t\t\tTimestamp: time.Now().UnixMilli(),\n\t\t}\n\n\t\terr := context.SendInterrupt(\"non-existent-id\", signal)\n\t\tif err == nil {\n\t\t\tt.Error(\"Expected error when sending to non-existent context\")\n\t\t}\n\n\t\tt.Log(\"✓ Sending to non-existent context returns error\")\n\t})\n}\n\n// TestInterruptCheckMethods tests different check methods\nfunc TestInterruptCheckMethods(t *testing.T) {\n\tctx := newTestContextWithInterrupt(\"chat-test-check-methods\", \"test-assistant\")\n\tdefer ctx.Release()\n\n\tt.Run(\"PeekDoesNotRemove\", func(t *testing.T) {\n\t\t// Send signal\n\t\tsignal := &context.InterruptSignal{\n\t\t\tType: context.InterruptGraceful,\n\t\t\tMessages: []context.Message{{Role: context.RoleUser, Content: \"peek test\"}},\n\t\t\tTimestamp: time.Now().UnixMilli(),\n\t\t}\n\t\tcontext.SendInterrupt(ctx.ID, signal)\n\t\ttime.Sleep(100 * time.Millisecond)\n\n\t\t// Peek should return signal but not remove it\n\t\tpeeked1 := ctx.Interrupt.Peek()\n\t\tif peeked1 == nil {\n\t\t\tt.Fatal(\"Expected signal from first peek\")\n\t\t}\n\n\t\tpeeked2 := ctx.Interrupt.Peek()\n\t\tif peeked2 == nil {\n\t\t\tt.Fatal(\"Expected signal from second peek\")\n\t\t}\n\n\t\tif peeked1.Messages[0].Content != peeked2.Messages[0].Content {\n\t\t\tt.Error(\"Peek should return the same signal\")\n\t\t}\n\n\t\tt.Log(\"✓ Peek does not remove signal\")\n\t})\n\n\tt.Run(\"CheckRemovesSignal\", func(t *testing.T) {\n\t\tctx.Interrupt.Clear()\n\n\t\t// Send signal\n\t\tsignal := &context.InterruptSignal{\n\t\t\tType: context.InterruptGraceful,\n\t\t\tMessages: []context.Message{{Role: context.RoleUser, Content: \"check test\"}},\n\t\t\tTimestamp: time.Now().UnixMilli(),\n\t\t}\n\t\tcontext.SendInterrupt(ctx.ID, signal)\n\t\ttime.Sleep(100 * time.Millisecond)\n\n\t\t// Check should return and remove signal\n\t\tchecked := ctx.Interrupt.Check()\n\t\tif checked == nil {\n\t\t\tt.Fatal(\"Expected signal from check\")\n\t\t}\n\n\t\t// Second check should return nil\n\t\tchecked2 := ctx.Interrupt.Check()\n\t\tif checked2 != nil {\n\t\t\tt.Error(\"Expected nil from second check after removal\")\n\t\t}\n\n\t\tt.Log(\"✓ Check removes signal after retrieval\")\n\t})\n\n\tt.Run(\"CheckWithMergeMultipleSignals\", func(t *testing.T) {\n\t\tctx.Interrupt.Clear()\n\n\t\t// Send 5 signals with different messages\n\t\tmessages := []string{\n\t\t\t\"First message\",\n\t\t\t\"Second message\",\n\t\t\t\"Third message\",\n\t\t\t\"Fourth message\",\n\t\t\t\"Fifth message\",\n\t\t}\n\n\t\tfor i, msg := range messages {\n\t\t\tsignal := &context.InterruptSignal{\n\t\t\t\tType: context.InterruptGraceful,\n\t\t\t\tMessages: []context.Message{\n\t\t\t\t\t{Role: context.RoleUser, Content: msg},\n\t\t\t\t},\n\t\t\t\tTimestamp: time.Now().UnixMilli(),\n\t\t\t\tMetadata: map[string]interface{}{\n\t\t\t\t\t\"sequence\": i + 1,\n\t\t\t\t},\n\t\t\t}\n\t\t\terr := context.SendInterrupt(ctx.ID, signal)\n\t\t\tif err != nil {\n\t\t\t\tt.Fatalf(\"Failed to send signal %d: %v\", i+1, err)\n\t\t\t}\n\t\t\ttime.Sleep(10 * time.Millisecond) // Small delay between signals\n\t\t}\n\n\t\ttime.Sleep(100 * time.Millisecond)\n\n\t\t// Verify all signals are queued\n\t\tpendingCount := ctx.Interrupt.GetPendingCount()\n\t\tif pendingCount != 5 {\n\t\t\tt.Errorf(\"Expected 5 pending signals, got: %d\", pendingCount)\n\t\t}\n\n\t\t// CheckWithMerge should merge all messages into one signal\n\t\tmerged := ctx.Interrupt.CheckWithMerge()\n\t\tif merged == nil {\n\t\t\tt.Fatal(\"Expected merged signal, got nil\")\n\t\t}\n\n\t\t// Verify all messages are merged\n\t\tif len(merged.Messages) != 5 {\n\t\t\tt.Errorf(\"Expected 5 merged messages, got: %d\", len(merged.Messages))\n\t\t}\n\n\t\t// Verify message order\n\t\tfor i, msg := range messages {\n\t\t\tif merged.Messages[i].Content != msg {\n\t\t\t\tt.Errorf(\"Message %d mismatch: expected '%s', got '%s'\", i+1, msg, merged.Messages[i].Content)\n\t\t\t}\n\t\t}\n\n\t\t// Verify metadata indicates merge\n\t\tif merged.Metadata[\"merged\"] != true {\n\t\t\tt.Error(\"Expected merged metadata to be true\")\n\t\t}\n\t\tif merged.Metadata[\"merged_count\"] != 5 {\n\t\t\tt.Errorf(\"Expected merged_count 5, got: %v\", merged.Metadata[\"merged_count\"])\n\t\t}\n\n\t\t// After merge, queue should be empty\n\t\tif ctx.Interrupt.GetPendingCount() != 0 {\n\t\t\tt.Errorf(\"Expected empty queue after merge, got: %d\", ctx.Interrupt.GetPendingCount())\n\t\t}\n\n\t\tt.Log(\"✓ CheckWithMerge correctly merged 5 signals into one\")\n\t})\n\n\tt.Run(\"CheckWithMergeSingleSignal\", func(t *testing.T) {\n\t\tctx.Interrupt.Clear()\n\n\t\t// Send single signal\n\t\tsignal := &context.InterruptSignal{\n\t\t\tType: context.InterruptGraceful,\n\t\t\tMessages: []context.Message{{Role: context.RoleUser, Content: \"single signal\"}},\n\t\t\tTimestamp: time.Now().UnixMilli(),\n\t\t}\n\t\tcontext.SendInterrupt(ctx.ID, signal)\n\t\ttime.Sleep(100 * time.Millisecond)\n\n\t\t// CheckWithMerge with single signal should return it without merge metadata\n\t\tmerged := ctx.Interrupt.CheckWithMerge()\n\t\tif merged == nil {\n\t\t\tt.Fatal(\"Expected signal, got nil\")\n\t\t}\n\n\t\tif len(merged.Messages) != 1 {\n\t\t\tt.Errorf(\"Expected 1 message, got: %d\", len(merged.Messages))\n\t\t}\n\n\t\t// Single signal should not have merge metadata\n\t\tif merged.Metadata != nil && merged.Metadata[\"merged\"] == true {\n\t\t\tt.Error(\"Single signal should not have merge metadata\")\n\t\t}\n\n\t\tt.Log(\"✓ CheckWithMerge handles single signal correctly\")\n\t})\n}\n\n// TestInterruptContext tests interrupt context methods\nfunc TestInterruptContext(t *testing.T) {\n\tctx := newTestContextWithInterrupt(\"chat-test-interrupt-context\", \"test-assistant\")\n\tdefer ctx.Release()\n\n\tt.Run(\"InterruptContextMethod\", func(t *testing.T) {\n\t\t// Get interrupt context\n\t\tinterruptCtx := ctx.Interrupt.Context()\n\t\tif interruptCtx == nil {\n\t\t\tt.Fatal(\"Expected interrupt context, got nil\")\n\t\t}\n\n\t\t// Context should not be done initially\n\t\tselect {\n\t\tcase <-interruptCtx.Done():\n\t\t\tt.Error(\"Interrupt context should not be done initially\")\n\t\tdefault:\n\t\t\tt.Log(\"✓ Interrupt context is not done initially\")\n\t\t}\n\t})\n\n\tt.Run(\"IsInterruptedFalseInitially\", func(t *testing.T) {\n\t\t// Should not be interrupted initially\n\t\tif ctx.Interrupt.IsInterrupted() {\n\t\t\tt.Error(\"Should not be interrupted initially\")\n\t\t}\n\t\tt.Log(\"✓ IsInterrupted returns false initially\")\n\t})\n\n\tt.Run(\"ForceInterruptCancelsContext\", func(t *testing.T) {\n\t\t// Get context before interrupt\n\t\tinterruptCtx := ctx.Interrupt.Context()\n\n\t\t// Send force interrupt with empty messages (pure cancellation)\n\t\t// This is the pattern for stopping streaming without appending messages\n\t\tsignal := &context.InterruptSignal{\n\t\t\tType: context.InterruptForce,\n\t\t\tMessages: []context.Message{}, // Empty messages = pure cancellation\n\t\t\tTimestamp: time.Now().UnixMilli(),\n\t\t}\n\t\terr := context.SendInterrupt(ctx.ID, signal)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to send interrupt: %v\", err)\n\t\t}\n\n\t\ttime.Sleep(100 * time.Millisecond)\n\n\t\t// The OLD context should be cancelled\n\t\tselect {\n\t\tcase <-interruptCtx.Done():\n\t\t\tt.Log(\"✓ Force interrupt with empty messages cancelled the old context\")\n\t\tcase <-time.After(200 * time.Millisecond):\n\t\t\tt.Error(\"Old context was not cancelled after force interrupt with empty messages\")\n\t\t}\n\n\t\t// Note: IsInterrupted() checks the NEW context (which was recreated)\n\t\t// So it will return false. This is expected behavior.\n\t\t// The key is that the old context was cancelled (checked above)\n\t\tt.Log(\"✓ Context was recreated after force interrupt (expected behavior)\")\n\t})\n\n\tt.Run(\"GracefulInterruptDoesNotCancelContext\", func(t *testing.T) {\n\t\t// Create new context for this test\n\t\tctx2 := newTestContextWithInterrupt(\"chat-test-graceful-no-cancel\", \"test-assistant\")\n\t\tdefer ctx2.Release()\n\n\t\tinterruptCtx := ctx2.Interrupt.Context()\n\n\t\t// Send graceful interrupt\n\t\tsignal := &context.InterruptSignal{\n\t\t\tType: context.InterruptGraceful,\n\t\t\tMessages: []context.Message{{Role: context.RoleUser, Content: \"graceful\"}},\n\t\t\tTimestamp: time.Now().UnixMilli(),\n\t\t}\n\t\tcontext.SendInterrupt(ctx2.ID, signal)\n\t\ttime.Sleep(100 * time.Millisecond)\n\n\t\t// Context should NOT be cancelled for graceful interrupt\n\t\tselect {\n\t\tcase <-interruptCtx.Done():\n\t\t\tt.Error(\"Graceful interrupt should not cancel context\")\n\t\tdefault:\n\t\t\tt.Log(\"✓ Graceful interrupt does not cancel context\")\n\t\t}\n\n\t\t// IsInterrupted should still return false for graceful\n\t\tif ctx2.Interrupt.IsInterrupted() {\n\t\t\tt.Error(\"IsInterrupted should return false for graceful interrupt\")\n\t\t} else {\n\t\t\tt.Log(\"✓ IsInterrupted returns false for graceful interrupt\")\n\t\t}\n\t})\n}\n\n// TestInterruptSendSignalDirectly tests SendSignal method directly\nfunc TestInterruptSendSignalDirectly(t *testing.T) {\n\tctx := newTestContextWithInterrupt(\"chat-test-send-signal\", \"test-assistant\")\n\tdefer ctx.Release()\n\n\tt.Run(\"SendSignalSuccess\", func(t *testing.T) {\n\t\tsignal := &context.InterruptSignal{\n\t\t\tType: context.InterruptGraceful,\n\t\t\tMessages: []context.Message{{Role: context.RoleUser, Content: \"direct send\"}},\n\t\t\tTimestamp: time.Now().UnixMilli(),\n\t\t}\n\n\t\terr := ctx.Interrupt.SendSignal(signal)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"SendSignal failed: %v\", err)\n\t\t}\n\n\t\ttime.Sleep(100 * time.Millisecond)\n\n\t\t// Verify signal was received\n\t\treceived := ctx.Interrupt.Peek()\n\t\tif received == nil {\n\t\t\tt.Fatal(\"Signal not received\")\n\t\t}\n\n\t\tif received.Messages[0].Content != \"direct send\" {\n\t\t\tt.Errorf(\"Expected 'direct send', got: %s\", received.Messages[0].Content)\n\t\t}\n\n\t\tt.Log(\"✓ SendSignal directly works\")\n\t})\n\n\tt.Run(\"SendSignalToNilController\", func(t *testing.T) {\n\t\tvar nilController *context.InterruptController\n\t\tsignal := &context.InterruptSignal{\n\t\t\tType: context.InterruptGraceful,\n\t\t\tMessages: []context.Message{{Role: context.RoleUser, Content: \"test\"}},\n\t\t\tTimestamp: time.Now().UnixMilli(),\n\t\t}\n\n\t\terr := nilController.SendSignal(signal)\n\t\tif err == nil {\n\t\t\tt.Error(\"Expected error when sending to nil controller\")\n\t\t} else {\n\t\t\tt.Logf(\"✓ Correctly returned error for nil controller: %v\", err)\n\t\t}\n\t})\n\n\tt.Run(\"SendSignalTimeout\", func(t *testing.T) {\n\t\t// Create controller but don't start listener\n\t\ttestCtrl := context.NewInterruptController()\n\t\t// Don't call Start(), so channel won't be read\n\n\t\t// Fill the buffer (capacity is 10)\n\t\tfor i := 0; i < 10; i++ {\n\t\t\tsignal := &context.InterruptSignal{\n\t\t\t\tType: context.InterruptGraceful,\n\t\t\t\tMessages: []context.Message{{Role: context.RoleUser, Content: fmt.Sprintf(\"msg %d\", i)}},\n\t\t\t\tTimestamp: time.Now().UnixMilli(),\n\t\t\t}\n\t\t\ttestCtrl.SendSignal(signal)\n\t\t}\n\n\t\t// This should timeout since buffer is full and no listener\n\t\tsignal := &context.InterruptSignal{\n\t\t\tType: context.InterruptGraceful,\n\t\t\tMessages: []context.Message{{Role: context.RoleUser, Content: \"overflow\"}},\n\t\t\tTimestamp: time.Now().UnixMilli(),\n\t\t}\n\n\t\terr := testCtrl.SendSignal(signal)\n\t\tif err == nil {\n\t\t\tt.Error(\"Expected timeout error when buffer is full\")\n\t\t} else {\n\t\t\tt.Logf(\"✓ SendSignal correctly times out when buffer full: %v\", err)\n\t\t}\n\t})\n}\n" }, { "path": "agent/context/jsapi.go", "content": "package context\n\nimport (\n\t\"time\"\n\n\t\"github.com/yaoapp/gou/runtime/v8/bridge\"\n\t\"github.com/yaoapp/yao/agent/memory\"\n\t\"github.com/yaoapp/yao/agent/output\"\n\t\"github.com/yaoapp/yao/agent/output/message\"\n\ttraceJsapi \"github.com/yaoapp/yao/trace/jsapi\"\n\t\"rogchap.com/v8go\"\n)\n\n// JsValue return the JavaScript value of the context\nfunc (ctx *Context) JsValue(v8ctx *v8go.Context) (*v8go.Value, error) {\n\treturn ctx.NewObject(v8ctx)\n}\n\n// NewObject Create a new JavaScript object from the context\nfunc (ctx *Context) NewObject(v8ctx *v8go.Context) (*v8go.Value, error) {\n\n\tjsObject := v8go.NewObjectTemplate(v8ctx.Isolate())\n\n\t// Set internal field count to 1 to store the goValueID\n\t// Internal fields are not accessible from JavaScript, providing better security\n\tjsObject.SetInternalFieldCount(1)\n\n\t// Register context in global bridge registry for efficient Go object retrieval\n\t// The goValueID will be stored in internal field (index 0) after instance creation\n\tgoValueID := bridge.RegisterGoObject(ctx)\n\n\t// Set release function (both __release and Release do the same thing)\n\t// __release: Internal cleanup (called by GC or Use())\n\t// Release: Public method for manual cleanup (try-finally pattern)\n\treleaseFunc := ctx.objectRelease(v8ctx.Isolate(), goValueID)\n\tjsObject.Set(\"__release\", releaseFunc)\n\tjsObject.Set(\"Release\", releaseFunc)\n\n\t// Set primitive fields in template\n\tjsObject.Set(\"chat_id\", ctx.ChatID)\n\tjsObject.Set(\"assistant_id\", ctx.AssistantID)\n\tjsObject.Set(\"locale\", ctx.Locale)\n\tjsObject.Set(\"theme\", ctx.Theme)\n\tjsObject.Set(\"referer\", ctx.Referer)\n\tjsObject.Set(\"accept\", string(ctx.Accept))\n\tjsObject.Set(\"route\", ctx.Route)\n\n\t// Set methods\n\tjsObject.Set(\"Send\", ctx.sendMethod(v8ctx.Isolate()))\n\tjsObject.Set(\"SendStream\", ctx.sendStreamMethod(v8ctx.Isolate()))\n\tjsObject.Set(\"Replace\", ctx.replaceMethod(v8ctx.Isolate()))\n\tjsObject.Set(\"Append\", ctx.appendMethod(v8ctx.Isolate()))\n\tjsObject.Set(\"Merge\", ctx.mergeMethod(v8ctx.Isolate()))\n\tjsObject.Set(\"Set\", ctx.setMethod(v8ctx.Isolate()))\n\tjsObject.Set(\"End\", ctx.endMethod(v8ctx.Isolate()))\n\n\t// Set ID generator methods\n\tjsObject.Set(\"MessageID\", ctx.messageIDMethod(v8ctx.Isolate()))\n\tjsObject.Set(\"BlockID\", ctx.blockIDMethod(v8ctx.Isolate()))\n\tjsObject.Set(\"ThreadID\", ctx.threadIDMethod(v8ctx.Isolate()))\n\n\t// Lifecycle methods\n\tjsObject.Set(\"EndBlock\", ctx.endBlockMethod(v8ctx.Isolate()))\n\n\t// Set mcp object\n\tjsObject.Set(\"mcp\", ctx.newMCPObject(v8ctx.Isolate()))\n\n\t// Set search object\n\tjsObject.Set(\"search\", ctx.newSearchObject(v8ctx.Isolate()))\n\n\t// Set agent object for calling other agents\n\tjsObject.Set(\"agent\", ctx.newAgentObject(v8ctx.Isolate()))\n\n\t// Set llm object for direct LLM calls\n\tjsObject.Set(\"llm\", ctx.newLlmObject(v8ctx.Isolate()))\n\n\t// Note: Space object will be set after instance creation (requires v8ctx)\n\n\t// Create instance\n\tinstance, err := jsObject.NewInstance(v8ctx)\n\tif err != nil {\n\t\t// Clean up: release from global registry if instance creation failed\n\t\tbridge.ReleaseGoObject(goValueID)\n\t\treturn nil, err\n\t}\n\n\t// Store the goValueID in internal field (index 0)\n\t// This is not accessible from JavaScript, providing better security\n\tobj, err := instance.Value.AsObject()\n\tif err != nil {\n\t\tbridge.ReleaseGoObject(goValueID)\n\t\treturn nil, err\n\t}\n\n\terr = obj.SetInternalField(0, goValueID)\n\tif err != nil {\n\t\tbridge.ReleaseGoObject(goValueID)\n\t\treturn nil, err\n\t}\n\n\t// Set trace object (property, not method)\n\t// If trace is not initialized, use no-op object\n\ttraceObj := ctx.createTraceObject(v8ctx)\n\tif traceObj != nil {\n\t\tobj.Set(\"trace\", traceObj)\n\t}\n\n\t// Set complex objects (maps, arrays) after instance creation using bridge\n\t// Client object\n\tclientData := map[string]interface{}{\n\t\t\"type\": ctx.Client.Type,\n\t\t\"user_agent\": ctx.Client.UserAgent,\n\t\t\"ip\": ctx.Client.IP,\n\t}\n\tclientVal, err := bridge.JsValue(v8ctx, clientData)\n\tif err == nil {\n\t\tobj.Set(\"client\", clientVal)\n\t\tclientVal.Release() // Release Go-side Persistent handle, V8 internal reference remains\n\t}\n\n\t// Metadata object - always set to empty map if nil\n\tmetadataData := ctx.Metadata\n\tif metadataData == nil {\n\t\tmetadataData = map[string]interface{}{}\n\t}\n\tmetadataVal, err := bridge.JsValue(v8ctx, metadataData)\n\tif err == nil {\n\t\tobj.Set(\"metadata\", metadataVal)\n\t\tmetadataVal.Release() // Release Go-side Persistent handle, V8 internal reference remains\n\t}\n\n\t// Authorized object - pass the complete structure\n\tif ctx.Authorized != nil {\n\t\tauthorizedVal, err := bridge.JsValue(v8ctx, ctx.Authorized)\n\t\tif err == nil {\n\t\t\tobj.Set(\"authorized\", authorizedVal)\n\t\t\tauthorizedVal.Release() // Release Go-side Persistent handle, V8 internal reference remains\n\t\t}\n\t} else {\n\t\t// Set to empty object when nil\n\t\temptyObj, err := bridge.JsValue(v8ctx, map[string]interface{}{})\n\t\tif err == nil {\n\t\t\tobj.Set(\"authorized\", emptyObj)\n\t\t\temptyObj.Release()\n\t\t}\n\t}\n\n\t// Memory object - create a JavaScript object with User/Team/Chat/Context namespaces\n\tif ctx.Memory != nil {\n\t\tmemoryObj := ctx.createMemoryObject(v8ctx)\n\t\tobj.Set(\"memory\", memoryObj)\n\t\tmemoryObj.Release()\n\t}\n\n\t// Sandbox object - only set if sandbox executor is available (V1)\n\tif ctx.sandboxExecutor != nil {\n\t\tsandboxObj := ctx.createSandboxInstance(v8ctx)\n\t\tif sandboxObj != nil {\n\t\t\tobj.Set(\"sandbox\", sandboxObj)\n\t\t\tsandboxObj.Release()\n\t\t}\n\t}\n\n\t// Computer object - only set if V2 computer is available\n\tif ctx.computer != nil {\n\t\tcomputerObj := ctx.createComputerInstance(v8ctx)\n\t\tif computerObj != nil {\n\t\t\tobj.Set(\"computer\", computerObj)\n\t\t\tcomputerObj.Release()\n\t\t}\n\t}\n\n\t// Workspace object - only set if V2 workspace is available\n\tif ctx.workspace != nil {\n\t\twsObj := ctx.createWorkspaceInstance(v8ctx)\n\t\tif wsObj != nil {\n\t\t\tobj.Set(\"workspace\", wsObj)\n\t\t\twsObj.Release()\n\t\t}\n\t}\n\n\treturn instance.Value, nil\n}\n\n// objectRelease releases the Go object from the global bridge registry\n// It retrieves the goValueID from internal field (index 0) and releases the Go object\n// Also releases associated Trace object if present\nfunc (ctx *Context) objectRelease(iso *v8go.Isolate, goValueID string) *v8go.FunctionTemplate {\n\treturn v8go.NewFunctionTemplate(iso, func(info *v8go.FunctionCallbackInfo) *v8go.Value {\n\t\t// Get the context object (this)\n\t\tthisObj, err := info.This().AsObject()\n\t\tif err == nil {\n\t\t\t// NOTE: We do NOT automatically release Trace object here\n\t\t\t//\n\t\t\t// Rationale:\n\t\t\t// 1. Each Hook execution creates a new V8 script context (scriptCtx)\n\t\t\t// 2. The agent Context (ctx) is passed to the Hook as a parameter\n\t\t\t// 3. When scriptCtx.Close() is called (via defer), V8 cleanup triggers ctx.__release()\n\t\t\t// 4. If we release Trace here, it gets released after EVERY Hook execution\n\t\t\t// 5. This causes \"context canceled\" errors in subsequent operations\n\t\t\t//\n\t\t\t// Trace lifecycle:\n\t\t\t// - Trace is created when agent.Stream() starts (in Context.Trace())\n\t\t\t// - Trace should persist across ALL Hook executions (Create, Next, Done)\n\t\t\t// - Trace is released when agent Context.Release() is called (after agent.Stream() completes)\n\t\t\t//\n\t\t\t// Memory management:\n\t\t\t// - If JS code explicitly calls trace.Release(), it will work (trace/jsapi/trace.go:traceGoRelease)\n\t\t\t// - If not explicitly called, Context.Release() will clean it up (context/context.go:Release)\n\t\t\t// - This is the correct lifecycle: one Context -> one Trace -> multiple Hook executions\n\n\t\t\t// Release Context Go object from bridge registry\n\t\t\tif thisObj.InternalFieldCount() > 0 {\n\t\t\t\t// Get goValueID from internal field (index 0)\n\t\t\t\tgoValueID := thisObj.GetInternalField(0)\n\t\t\t\tif goValueID != nil && goValueID.IsString() {\n\t\t\t\t\t// Release from global bridge registry\n\t\t\t\t\tbridge.ReleaseGoObject(goValueID.String())\n\t\t\t\t}\n\t\t\t}\n\t\t}\n\n\t\treturn v8go.Undefined(info.Context().Isolate())\n\t})\n}\n\n// createTraceObject creates a Trace object instance\n// Returns a no-op Trace object if trace is not initialized\nfunc (ctx *Context) createTraceObject(v8ctx *v8go.Context) *v8go.Value {\n\t// Try to get trace manager\n\tmanager, err := ctx.Trace()\n\tif err != nil || manager == nil {\n\t\t// Return no-op trace object if initialization fails\n\t\tnoOpTrace, _ := traceJsapi.NewNoOpTraceObject(v8ctx)\n\t\treturn noOpTrace\n\t}\n\n\t// Get trace ID\n\ttraceID := \"\"\n\tif ctx.Stack != nil {\n\t\ttraceID = ctx.Stack.TraceID\n\t}\n\n\t// Create JavaScript Trace object\n\ttraceObj, err := traceJsapi.NewTraceObject(v8ctx, traceID, manager)\n\tif err != nil {\n\t\t// Return no-op trace object if creation fails\n\t\tnoOpTrace, _ := traceJsapi.NewNoOpTraceObject(v8ctx)\n\t\treturn noOpTrace\n\t}\n\n\treturn traceObj\n}\n\n// sendMethod implements ctx.Send(message, blockId?)\n// Usage: const messageId = ctx.Send({ type: \"text\", props: { content: \"Hello\" } })\n// Usage: const messageId = ctx.Send(\"Hello\") // shorthand for text message\n// Usage: const messageId = ctx.Send(\"Hello\", \"B1\") // specify block ID\n// Automatically generates MessageID and BlockID (if not specified), flushes output\n// Returns: message_id (string)\nfunc (ctx *Context) sendMethod(iso *v8go.Isolate) *v8go.FunctionTemplate {\n\treturn v8go.NewFunctionTemplate(iso, func(info *v8go.FunctionCallbackInfo) *v8go.Value {\n\t\tv8ctx := info.Context()\n\t\targs := info.Args()\n\n\t\tif len(args) < 1 {\n\t\t\treturn bridge.JsException(v8ctx, \"Send requires a message argument\")\n\t\t}\n\n\t\t// Parse message argument\n\t\tmsg, err := parseMessage(v8ctx, args[0])\n\t\tif err != nil {\n\t\t\treturn bridge.JsException(v8ctx, \"invalid message: \"+err.Error())\n\t\t}\n\n\t\t// Get optional blockId argument (second argument)\n\t\t// Note: message object's block_id has higher priority\n\t\tif len(args) >= 2 && args[1].IsString() && msg.BlockID == \"\" {\n\t\t\tmsg.BlockID = args[1].String()\n\t\t}\n\n\t\t// Generate unique MessageID if not provided\n\t\tif msg.MessageID == \"\" {\n\t\t\tif ctx.IDGenerator != nil {\n\t\t\t\tmsg.MessageID = ctx.IDGenerator.GenerateMessageID()\n\t\t\t} else {\n\t\t\t\tmsg.MessageID = output.GenerateID()\n\t\t\t}\n\t\t}\n\n\t\t// Call ctx.Send (will auto-generate BlockID if still empty)\n\t\tif err := ctx.Send(msg); err != nil {\n\t\t\treturn bridge.JsException(v8ctx, \"Send failed: \"+err.Error())\n\t\t}\n\n\t\t// Automatically flush after sending\n\t\tif err := ctx.Flush(); err != nil {\n\t\t\treturn bridge.JsException(v8ctx, \"Flush failed: \"+err.Error())\n\t\t}\n\n\t\t// Return the message ID\n\t\tmessageID, err := v8go.NewValue(iso, msg.MessageID)\n\t\tif err != nil {\n\t\t\treturn bridge.JsException(v8ctx, \"Failed to create return value: \"+err.Error())\n\t\t}\n\t\treturn messageID\n\t})\n}\n\n// sendStreamMethod implements ctx.SendStream(message)\n// Usage: const msgId = ctx.SendStream({ type: \"text\", props: { content: \"Initial content\" } })\n// Starts a streaming message that can be appended to with ctx.Append()\n// Must be finalized with ctx.End(msgId) or ctx.End(msgId, \"final content\")\n// Unlike Send(), this does NOT automatically send message_end event\n// Returns: message_id (string)\nfunc (ctx *Context) sendStreamMethod(iso *v8go.Isolate) *v8go.FunctionTemplate {\n\treturn v8go.NewFunctionTemplate(iso, func(info *v8go.FunctionCallbackInfo) *v8go.Value {\n\t\tv8ctx := info.Context()\n\t\targs := info.Args()\n\n\t\tif len(args) < 1 {\n\t\t\treturn bridge.JsException(v8ctx, \"SendStream requires a message argument\")\n\t\t}\n\n\t\t// Parse message argument\n\t\tmsg, err := parseMessage(v8ctx, args[0])\n\t\tif err != nil {\n\t\t\treturn bridge.JsException(v8ctx, \"invalid message: \"+err.Error())\n\t\t}\n\n\t\t// Get optional blockId argument (second argument)\n\t\tif len(args) >= 2 && args[1].IsString() && msg.BlockID == \"\" {\n\t\t\tmsg.BlockID = args[1].String()\n\t\t}\n\n\t\t// Call ctx.SendStream\n\t\tmessageID, err := ctx.SendStream(msg)\n\t\tif err != nil {\n\t\t\treturn bridge.JsException(v8ctx, \"SendStream failed: \"+err.Error())\n\t\t}\n\n\t\t// Automatically flush after sending\n\t\tif err := ctx.Flush(); err != nil {\n\t\t\treturn bridge.JsException(v8ctx, \"Flush failed: \"+err.Error())\n\t\t}\n\n\t\t// Return the message ID\n\t\treturnID, err := v8go.NewValue(iso, messageID)\n\t\tif err != nil {\n\t\t\treturn bridge.JsException(v8ctx, \"Failed to create return value: \"+err.Error())\n\t\t}\n\t\treturn returnID\n\t})\n}\n\n// endMethod implements ctx.End(messageId, finalContent?)\n// Usage: ctx.End(msgId) or ctx.End(msgId, \"final content to append\")\n// Finalizes a streaming message started with SendStream()\n// Sends message_end event with the complete accumulated content\n// Returns: message_id (string)\nfunc (ctx *Context) endMethod(iso *v8go.Isolate) *v8go.FunctionTemplate {\n\treturn v8go.NewFunctionTemplate(iso, func(info *v8go.FunctionCallbackInfo) *v8go.Value {\n\t\tv8ctx := info.Context()\n\t\targs := info.Args()\n\n\t\tif len(args) < 1 {\n\t\t\treturn bridge.JsException(v8ctx, \"End requires a messageId argument\")\n\t\t}\n\n\t\t// Get message ID (first argument)\n\t\tif !args[0].IsString() {\n\t\t\treturn bridge.JsException(v8ctx, \"messageId must be a string\")\n\t\t}\n\t\tmessageID := args[0].String()\n\n\t\t// Get optional final content (second argument)\n\t\tvar finalContent string\n\t\tif len(args) >= 2 && args[1].IsString() {\n\t\t\tfinalContent = args[1].String()\n\t\t}\n\n\t\t// Call ctx.End\n\t\tvar err error\n\t\tif finalContent != \"\" {\n\t\t\terr = ctx.End(messageID, finalContent)\n\t\t} else {\n\t\t\terr = ctx.End(messageID)\n\t\t}\n\t\tif err != nil {\n\t\t\treturn bridge.JsException(v8ctx, \"End failed: \"+err.Error())\n\t\t}\n\n\t\t// Automatically flush after sending\n\t\tif err := ctx.Flush(); err != nil {\n\t\t\treturn bridge.JsException(v8ctx, \"Flush failed: \"+err.Error())\n\t\t}\n\n\t\t// Return the message ID\n\t\treturnID, err := v8go.NewValue(iso, messageID)\n\t\tif err != nil {\n\t\t\treturn bridge.JsException(v8ctx, \"Failed to create return value: \"+err.Error())\n\t\t}\n\t\treturn returnID\n\t})\n}\n\n// replaceMethod implements ctx.Replace(messageId, message)\n// Usage: ctx.Replace(messageId, { type: \"text\", props: { content: \"Updated content\" } })\n// Replaces the entire message content with the specified message_id\n// Automatically flushes output\n// Returns: message_id (string)\nfunc (ctx *Context) replaceMethod(iso *v8go.Isolate) *v8go.FunctionTemplate {\n\treturn v8go.NewFunctionTemplate(iso, func(info *v8go.FunctionCallbackInfo) *v8go.Value {\n\t\tv8ctx := info.Context()\n\t\targs := info.Args()\n\n\t\t// Validate arguments\n\t\tif len(args) < 2 {\n\t\t\treturn bridge.JsException(v8ctx, \"Replace requires messageId and message arguments\")\n\t\t}\n\n\t\t// Get message ID (first argument)\n\t\tif !args[0].IsString() {\n\t\t\treturn bridge.JsException(v8ctx, \"messageId must be a string\")\n\t\t}\n\t\tmessageID := args[0].String()\n\n\t\t// Parse message argument (second argument)\n\t\tmsg, err := parseMessage(v8ctx, args[1])\n\t\tif err != nil {\n\t\t\treturn bridge.JsException(v8ctx, \"invalid message: \"+err.Error())\n\t\t}\n\n\t\t// Set message ID to the provided ID\n\t\tmsg.MessageID = messageID\n\n\t\t// Set delta mode for replacement\n\t\tmsg.Delta = true\n\t\tmsg.DeltaAction = message.DeltaReplace\n\t\tmsg.DeltaPath = \"\" // Empty path means replace entire message\n\n\t\t// Call ctx.Send\n\t\tif err := ctx.Send(msg); err != nil {\n\t\t\treturn bridge.JsException(v8ctx, \"Replace failed: \"+err.Error())\n\t\t}\n\n\t\t// Automatically flush after sending\n\t\tif err := ctx.Flush(); err != nil {\n\t\t\treturn bridge.JsException(v8ctx, \"Flush failed: \"+err.Error())\n\t\t}\n\n\t\t// Return the message ID\n\t\treturnID, err := v8go.NewValue(iso, messageID)\n\t\tif err != nil {\n\t\t\treturn bridge.JsException(v8ctx, \"Failed to create return value: \"+err.Error())\n\t\t}\n\t\treturn returnID\n\t})\n}\n\n// appendMethod implements ctx.Append(messageId, content, path?)\n// Usage: ctx.Append(messageId, \"more text\") // append to default content path\n// Usage: ctx.Append(messageId, \"more text\", \"props.content\") // append to specific path\n// Usage: ctx.Append(messageId, { type: \"text\", props: { content: \"more text\" } })\n// Usage: ctx.Append(messageId, { props: { content: \"more text\" } }, \"props.data\") // append to custom path\n// Appends content to an existing message (delta append operation)\n// Automatically flushes output\n// Returns: message_id (string)\nfunc (ctx *Context) appendMethod(iso *v8go.Isolate) *v8go.FunctionTemplate {\n\treturn v8go.NewFunctionTemplate(iso, func(info *v8go.FunctionCallbackInfo) *v8go.Value {\n\t\tv8ctx := info.Context()\n\t\targs := info.Args()\n\n\t\t// Validate arguments\n\t\tif len(args) < 2 {\n\t\t\treturn bridge.JsException(v8ctx, \"Append requires messageId and content arguments\")\n\t\t}\n\n\t\t// Get message ID (first argument)\n\t\tif !args[0].IsString() {\n\t\t\treturn bridge.JsException(v8ctx, \"messageId must be a string\")\n\t\t}\n\t\tmessageID := args[0].String()\n\n\t\t// Parse content argument (second argument)\n\t\tmsg, err := parseMessage(v8ctx, args[1])\n\t\tif err != nil {\n\t\t\treturn bridge.JsException(v8ctx, \"invalid content: \"+err.Error())\n\t\t}\n\n\t\t// Get optional path argument (third argument)\n\t\tdeltaPath := \"\"\n\t\tif len(args) >= 3 && args[2].IsString() {\n\t\t\tdeltaPath = args[2].String()\n\t\t}\n\n\t\t// Set message ID to the provided ID\n\t\tmsg.MessageID = messageID\n\n\t\t// Set delta mode for append\n\t\tmsg.Delta = true\n\t\tmsg.DeltaAction = message.DeltaAppend\n\t\tmsg.DeltaPath = deltaPath // Empty path means append to default content, or specify custom path\n\n\t\t// Call ctx.Send\n\t\tif err := ctx.Send(msg); err != nil {\n\t\t\treturn bridge.JsException(v8ctx, \"Append failed: \"+err.Error())\n\t\t}\n\n\t\t// Automatically flush after sending\n\t\tif err := ctx.Flush(); err != nil {\n\t\t\treturn bridge.JsException(v8ctx, \"Flush failed: \"+err.Error())\n\t\t}\n\n\t\t// Return the message ID\n\t\treturnID, err := v8go.NewValue(iso, messageID)\n\t\tif err != nil {\n\t\t\treturn bridge.JsException(v8ctx, \"Failed to create return value: \"+err.Error())\n\t\t}\n\t\treturn returnID\n\t})\n}\n\n// mergeMethod implements ctx.Merge(messageId, data, path?)\n// Usage: ctx.Merge(messageId, { key: \"value\" }) // merge to default object path\n// Usage: ctx.Merge(messageId, { status: \"done\" }, \"props\") // merge to specific path\n// Usage: ctx.Merge(messageId, { props: { status: \"done\", progress: 100 } })\n// Merges data into an existing message object (delta merge operation)\n// Automatically flushes output\n// Returns: message_id (string)\nfunc (ctx *Context) mergeMethod(iso *v8go.Isolate) *v8go.FunctionTemplate {\n\treturn v8go.NewFunctionTemplate(iso, func(info *v8go.FunctionCallbackInfo) *v8go.Value {\n\t\tv8ctx := info.Context()\n\t\targs := info.Args()\n\n\t\t// Validate arguments\n\t\tif len(args) < 2 {\n\t\t\treturn bridge.JsException(v8ctx, \"Merge requires messageId and data arguments\")\n\t\t}\n\n\t\t// Get message ID (first argument)\n\t\tif !args[0].IsString() {\n\t\t\treturn bridge.JsException(v8ctx, \"messageId must be a string\")\n\t\t}\n\t\tmessageID := args[0].String()\n\n\t\t// Parse data argument (second argument)\n\t\tmsg, err := parseMessage(v8ctx, args[1])\n\t\tif err != nil {\n\t\t\treturn bridge.JsException(v8ctx, \"invalid data: \"+err.Error())\n\t\t}\n\n\t\t// Get optional path argument (third argument)\n\t\tdeltaPath := \"\"\n\t\tif len(args) >= 3 && args[2].IsString() {\n\t\t\tdeltaPath = args[2].String()\n\t\t}\n\n\t\t// Set message ID to the provided ID\n\t\tmsg.MessageID = messageID\n\n\t\t// Set delta mode for merge\n\t\tmsg.Delta = true\n\t\tmsg.DeltaAction = message.DeltaMerge\n\t\tmsg.DeltaPath = deltaPath // Empty path means merge to default object, or specify custom path\n\n\t\t// Call ctx.Send\n\t\tif err := ctx.Send(msg); err != nil {\n\t\t\treturn bridge.JsException(v8ctx, \"Merge failed: \"+err.Error())\n\t\t}\n\n\t\t// Automatically flush after sending\n\t\tif err := ctx.Flush(); err != nil {\n\t\t\treturn bridge.JsException(v8ctx, \"Flush failed: \"+err.Error())\n\t\t}\n\n\t\t// Return the message ID\n\t\treturnID, err := v8go.NewValue(iso, messageID)\n\t\tif err != nil {\n\t\t\treturn bridge.JsException(v8ctx, \"Failed to create return value: \"+err.Error())\n\t\t}\n\t\treturn returnID\n\t})\n}\n\n// setMethod implements ctx.Set(messageId, data, path)\n// Usage: ctx.Set(messageId, \"value\", \"props.newField\") // set new field at specific path\n// Usage: ctx.Set(messageId, { newKey: \"value\" }, \"props\") // set new fields in props\n// Sets a new field or value in an existing message (delta set operation)\n// Automatically flushes output\n// Returns: message_id (string)\nfunc (ctx *Context) setMethod(iso *v8go.Isolate) *v8go.FunctionTemplate {\n\treturn v8go.NewFunctionTemplate(iso, func(info *v8go.FunctionCallbackInfo) *v8go.Value {\n\t\tv8ctx := info.Context()\n\t\targs := info.Args()\n\n\t\t// Validate arguments (path is required for Set operation)\n\t\tif len(args) < 3 {\n\t\t\treturn bridge.JsException(v8ctx, \"Set requires messageId, data, and path arguments\")\n\t\t}\n\n\t\t// Get message ID (first argument)\n\t\tif !args[0].IsString() {\n\t\t\treturn bridge.JsException(v8ctx, \"messageId must be a string\")\n\t\t}\n\t\tmessageID := args[0].String()\n\n\t\t// Parse data argument (second argument)\n\t\tmsg, err := parseMessage(v8ctx, args[1])\n\t\tif err != nil {\n\t\t\treturn bridge.JsException(v8ctx, \"invalid data: \"+err.Error())\n\t\t}\n\n\t\t// Get path argument (third argument - required)\n\t\tif !args[2].IsString() {\n\t\t\treturn bridge.JsException(v8ctx, \"path must be a string\")\n\t\t}\n\t\tdeltaPath := args[2].String()\n\n\t\tif deltaPath == \"\" {\n\t\t\treturn bridge.JsException(v8ctx, \"path cannot be empty for Set operation\")\n\t\t}\n\n\t\t// Set message ID to the provided ID\n\t\tmsg.MessageID = messageID\n\n\t\t// Set delta mode for set\n\t\tmsg.Delta = true\n\t\tmsg.DeltaAction = message.DeltaSet\n\t\tmsg.DeltaPath = deltaPath // Path is required for Set operation\n\n\t\t// Call ctx.Send\n\t\tif err := ctx.Send(msg); err != nil {\n\t\t\treturn bridge.JsException(v8ctx, \"Set failed: \"+err.Error())\n\t\t}\n\n\t\t// Automatically flush after sending\n\t\tif err := ctx.Flush(); err != nil {\n\t\t\treturn bridge.JsException(v8ctx, \"Flush failed: \"+err.Error())\n\t\t}\n\n\t\t// Return the message ID\n\t\treturnID, err := v8go.NewValue(iso, messageID)\n\t\tif err != nil {\n\t\t\treturn bridge.JsException(v8ctx, \"Failed to create return value: \"+err.Error())\n\t\t}\n\t\treturn returnID\n\t})\n}\n\n// messageIDMethod implements ctx.MessageID()\n// Usage: const msgId = ctx.MessageID() // Returns: \"M1\", \"M2\", \"M3\"...\n// Generates a unique message ID for manual message management\n// Returns: message_id (string)\nfunc (ctx *Context) messageIDMethod(iso *v8go.Isolate) *v8go.FunctionTemplate {\n\treturn v8go.NewFunctionTemplate(iso, func(info *v8go.FunctionCallbackInfo) *v8go.Value {\n\t\tv8ctx := info.Context()\n\n\t\tvar messageID string\n\t\tif ctx.IDGenerator != nil {\n\t\t\tmessageID = ctx.IDGenerator.GenerateMessageID()\n\t\t} else {\n\t\t\tmessageID = output.GenerateID()\n\t\t}\n\n\t\t// Return the generated ID\n\t\tid, err := v8go.NewValue(iso, messageID)\n\t\tif err != nil {\n\t\t\treturn bridge.JsException(v8ctx, \"Failed to generate message ID: \"+err.Error())\n\t\t}\n\t\treturn id\n\t})\n}\n\n// blockIDMethod implements ctx.BlockID()\n// Usage: const blockId = ctx.BlockID() // Returns: \"B1\", \"B2\", \"B3\"...\n// Generates a unique block ID for grouping messages\n// Returns: block_id (string)\nfunc (ctx *Context) blockIDMethod(iso *v8go.Isolate) *v8go.FunctionTemplate {\n\treturn v8go.NewFunctionTemplate(iso, func(info *v8go.FunctionCallbackInfo) *v8go.Value {\n\t\tv8ctx := info.Context()\n\n\t\tvar blockID string\n\t\tif ctx.IDGenerator != nil {\n\t\t\tblockID = ctx.IDGenerator.GenerateBlockID()\n\t\t} else {\n\t\t\tblockID = output.GenerateID()\n\t\t}\n\n\t\t// Return the generated ID\n\t\tid, err := v8go.NewValue(iso, blockID)\n\t\tif err != nil {\n\t\t\treturn bridge.JsException(v8ctx, \"Failed to generate block ID: \"+err.Error())\n\t\t}\n\t\treturn id\n\t})\n}\n\n// threadIDMethod implements ctx.ThreadID()\n// Usage: const threadId = ctx.ThreadID() // Returns: \"T1\", \"T2\", \"T3\"...\n// Generates a unique thread ID for concurrent operations\n// Returns: thread_id (string)\nfunc (ctx *Context) threadIDMethod(iso *v8go.Isolate) *v8go.FunctionTemplate {\n\treturn v8go.NewFunctionTemplate(iso, func(info *v8go.FunctionCallbackInfo) *v8go.Value {\n\t\tv8ctx := info.Context()\n\n\t\tvar threadID string\n\t\tif ctx.IDGenerator != nil {\n\t\t\tthreadID = ctx.IDGenerator.GenerateThreadID()\n\t\t} else {\n\t\t\tthreadID = output.GenerateID()\n\t\t}\n\n\t\t// Return the generated ID\n\t\tid, err := v8go.NewValue(iso, threadID)\n\t\tif err != nil {\n\t\t\treturn bridge.JsException(v8ctx, \"Failed to generate thread ID: \"+err.Error())\n\t\t}\n\t\treturn id\n\t})\n}\n\n// endBlockMethod implements ctx.EndBlock(block_id)\n// Usage: ctx.EndBlock(\"B1\")\n// Sends a block_end event for the specified block\n// Returns: undefined\nfunc (ctx *Context) endBlockMethod(iso *v8go.Isolate) *v8go.FunctionTemplate {\n\treturn v8go.NewFunctionTemplate(iso, func(info *v8go.FunctionCallbackInfo) *v8go.Value {\n\t\tv8ctx := info.Context()\n\t\targs := info.Args()\n\n\t\t// Validate arguments\n\t\tif len(args) < 1 {\n\t\t\treturn bridge.JsException(v8ctx, \"EndBlock requires block_id argument\")\n\t\t}\n\n\t\tif !args[0].IsString() {\n\t\t\treturn bridge.JsException(v8ctx, \"block_id must be a string\")\n\t\t}\n\n\t\tblockID := args[0].String()\n\n\t\t// Call ctx.EndBlock\n\t\tif err := ctx.EndBlock(blockID); err != nil {\n\t\t\treturn bridge.JsException(v8ctx, \"EndBlock failed: \"+err.Error())\n\t\t}\n\n\t\t// Automatically flush after ending block\n\t\tif err := ctx.Flush(); err != nil {\n\t\t\treturn bridge.JsException(v8ctx, \"Flush failed: \"+err.Error())\n\t\t}\n\n\t\treturn v8go.Undefined(iso)\n\t})\n}\n\n// createMemoryObject creates a Memory object for JavaScript access\n// Memory provides four namespaces: User, Team, Chat, Context\n// Each namespace supports: Get, Set, Del, Has, Keys, Len, Clear, Incr, Decr\nfunc (ctx *Context) createMemoryObject(v8ctx *v8go.Context) *v8go.Value {\n\tiso := v8ctx.Isolate()\n\tobjTpl := v8go.NewObjectTemplate(iso)\n\tobj, _ := objTpl.NewInstance(v8ctx)\n\n\t// Create namespace accessors\n\tif ctx.Memory.User != nil {\n\t\tuserObj := ctx.createNamespaceObject(v8ctx, ctx.Memory.User)\n\t\tobj.Set(\"user\", userObj)\n\t\tuserObj.Release()\n\t}\n\n\tif ctx.Memory.Team != nil {\n\t\tteamObj := ctx.createNamespaceObject(v8ctx, ctx.Memory.Team)\n\t\tobj.Set(\"team\", teamObj)\n\t\tteamObj.Release()\n\t}\n\n\tif ctx.Memory.Chat != nil {\n\t\tchatObj := ctx.createNamespaceObject(v8ctx, ctx.Memory.Chat)\n\t\tobj.Set(\"chat\", chatObj)\n\t\tchatObj.Release()\n\t}\n\n\tif ctx.Memory.Context != nil {\n\t\tcontextObj := ctx.createNamespaceObject(v8ctx, ctx.Memory.Context)\n\t\tobj.Set(\"context\", contextObj)\n\t\tcontextObj.Release()\n\t}\n\n\treturn obj.Value\n}\n\n// createNamespaceObject creates a namespace object with KV store methods\nfunc (ctx *Context) createNamespaceObject(v8ctx *v8go.Context, ns *memory.Namespace) *v8go.Value {\n\tiso := v8ctx.Isolate()\n\tobjTpl := v8go.NewObjectTemplate(iso)\n\tobj, _ := objTpl.NewInstance(v8ctx)\n\n\t// Get method: ns.Get(key)\n\tgetFunc := v8go.NewFunctionTemplate(iso, func(info *v8go.FunctionCallbackInfo) *v8go.Value {\n\t\tif len(info.Args()) < 1 {\n\t\t\treturn bridge.JsException(info.Context(), \"Get requires a key argument\")\n\t\t}\n\n\t\tkey := info.Args()[0].String()\n\t\tvalue, ok := ns.Get(key)\n\t\tif !ok {\n\t\t\treturn v8go.Null(iso)\n\t\t}\n\n\t\tjsValue, err := bridge.JsValue(info.Context(), value)\n\t\tif err != nil {\n\t\t\treturn v8go.Null(iso)\n\t\t}\n\n\t\treturn jsValue\n\t})\n\tgetFuncVal := getFunc.GetFunction(v8ctx)\n\tobj.Set(\"Get\", getFuncVal.Value)\n\n\t// Set method: ns.Set(key, value, ttl?)\n\tsetFunc := v8go.NewFunctionTemplate(iso, func(info *v8go.FunctionCallbackInfo) *v8go.Value {\n\t\tif len(info.Args()) < 2 {\n\t\t\treturn bridge.JsException(info.Context(), \"Set requires key and value arguments\")\n\t\t}\n\n\t\tkey := info.Args()[0].String()\n\t\tvalue, err := bridge.GoValue(info.Args()[1], info.Context())\n\t\tif err != nil {\n\t\t\treturn bridge.JsException(info.Context(), \"Failed to convert value: \"+err.Error())\n\t\t}\n\n\t\t// Optional TTL in milliseconds (third argument)\n\t\tvar ttl time.Duration\n\t\tif len(info.Args()) >= 3 && info.Args()[2].IsNumber() {\n\t\t\tttlMs := info.Args()[2].Integer()\n\t\t\tttl = time.Duration(ttlMs) * time.Millisecond\n\t\t}\n\n\t\tif err := ns.Set(key, value, ttl); err != nil {\n\t\t\treturn bridge.JsException(info.Context(), \"Failed to set value: \"+err.Error())\n\t\t}\n\n\t\treturn v8go.Undefined(iso)\n\t})\n\tsetFuncVal := setFunc.GetFunction(v8ctx)\n\tobj.Set(\"Set\", setFuncVal.Value)\n\n\t// Del method: ns.Del(key)\n\tdelFunc := v8go.NewFunctionTemplate(iso, func(info *v8go.FunctionCallbackInfo) *v8go.Value {\n\t\tif len(info.Args()) < 1 {\n\t\t\treturn bridge.JsException(info.Context(), \"Del requires a key argument\")\n\t\t}\n\n\t\tkey := info.Args()[0].String()\n\t\tif err := ns.Del(key); err != nil {\n\t\t\treturn bridge.JsException(info.Context(), \"Failed to delete key: \"+err.Error())\n\t\t}\n\n\t\treturn v8go.Undefined(iso)\n\t})\n\tdelFuncVal := delFunc.GetFunction(v8ctx)\n\tobj.Set(\"Del\", delFuncVal.Value)\n\n\t// Has method: ns.Has(key)\n\thasFunc := v8go.NewFunctionTemplate(iso, func(info *v8go.FunctionCallbackInfo) *v8go.Value {\n\t\tif len(info.Args()) < 1 {\n\t\t\treturn bridge.JsException(info.Context(), \"Has requires a key argument\")\n\t\t}\n\n\t\tkey := info.Args()[0].String()\n\t\texists := ns.Has(key)\n\n\t\tjsValue, _ := v8go.NewValue(iso, exists)\n\t\treturn jsValue\n\t})\n\thasFuncVal := hasFunc.GetFunction(v8ctx)\n\tobj.Set(\"Has\", hasFuncVal.Value)\n\n\t// Keys method: ns.Keys()\n\tkeysFunc := v8go.NewFunctionTemplate(iso, func(info *v8go.FunctionCallbackInfo) *v8go.Value {\n\t\tkeys := ns.Keys()\n\t\tjsValue, err := bridge.JsValue(info.Context(), keys)\n\t\tif err != nil {\n\t\t\treturn bridge.JsException(info.Context(), \"Failed to get keys: \"+err.Error())\n\t\t}\n\t\treturn jsValue\n\t})\n\tkeysFuncVal := keysFunc.GetFunction(v8ctx)\n\tobj.Set(\"Keys\", keysFuncVal.Value)\n\n\t// Len method: ns.Len()\n\tlenFunc := v8go.NewFunctionTemplate(iso, func(info *v8go.FunctionCallbackInfo) *v8go.Value {\n\t\tlength := ns.Len()\n\t\t// Use int32 for JavaScript Number (int64 becomes BigInt which is incompatible)\n\t\tjsValue, _ := v8go.NewValue(iso, int32(length))\n\t\treturn jsValue\n\t})\n\tlenFuncVal := lenFunc.GetFunction(v8ctx)\n\tobj.Set(\"Len\", lenFuncVal.Value)\n\n\t// Clear method: ns.Clear()\n\tclearFunc := v8go.NewFunctionTemplate(iso, func(info *v8go.FunctionCallbackInfo) *v8go.Value {\n\t\tns.Clear()\n\t\treturn v8go.Undefined(iso)\n\t})\n\tclearFuncVal := clearFunc.GetFunction(v8ctx)\n\tobj.Set(\"Clear\", clearFuncVal.Value)\n\n\t// Incr method: ns.Incr(key, delta?)\n\tincrFunc := v8go.NewFunctionTemplate(iso, func(info *v8go.FunctionCallbackInfo) *v8go.Value {\n\t\tif len(info.Args()) < 1 {\n\t\t\treturn bridge.JsException(info.Context(), \"Incr requires a key argument\")\n\t\t}\n\n\t\tkey := info.Args()[0].String()\n\t\tdelta := int64(1)\n\t\tif len(info.Args()) >= 2 && info.Args()[1].IsNumber() {\n\t\t\tdelta = info.Args()[1].Integer()\n\t\t}\n\n\t\tnewValue, err := ns.Incr(key, delta)\n\t\tif err != nil {\n\t\t\treturn bridge.JsException(info.Context(), \"Failed to increment: \"+err.Error())\n\t\t}\n\n\t\t// Use int32 for JavaScript Number (int64 becomes BigInt which is incompatible with ===)\n\t\t// For counters, int32 range (-2^31 to 2^31-1) is sufficient\n\t\tjsValue, _ := v8go.NewValue(iso, int32(newValue))\n\t\treturn jsValue\n\t})\n\tincrFuncVal := incrFunc.GetFunction(v8ctx)\n\tobj.Set(\"Incr\", incrFuncVal.Value)\n\n\t// Decr method: ns.Decr(key, delta?)\n\tdecrFunc := v8go.NewFunctionTemplate(iso, func(info *v8go.FunctionCallbackInfo) *v8go.Value {\n\t\tif len(info.Args()) < 1 {\n\t\t\treturn bridge.JsException(info.Context(), \"Decr requires a key argument\")\n\t\t}\n\n\t\tkey := info.Args()[0].String()\n\t\tdelta := int64(1)\n\t\tif len(info.Args()) >= 2 && info.Args()[1].IsNumber() {\n\t\t\tdelta = info.Args()[1].Integer()\n\t\t}\n\n\t\tnewValue, err := ns.Decr(key, delta)\n\t\tif err != nil {\n\t\t\treturn bridge.JsException(info.Context(), \"Failed to decrement: \"+err.Error())\n\t\t}\n\n\t\t// Use int32 for JavaScript Number (int64 becomes BigInt which is incompatible with ===)\n\t\tjsValue, _ := v8go.NewValue(iso, int32(newValue))\n\t\treturn jsValue\n\t})\n\tdecrFuncVal := decrFunc.GetFunction(v8ctx)\n\tobj.Set(\"Decr\", decrFuncVal.Value)\n\n\t// GetDel method: ns.GetDel(key) - Get value and delete immediately\n\tgetDelFunc := v8go.NewFunctionTemplate(iso, func(info *v8go.FunctionCallbackInfo) *v8go.Value {\n\t\tif len(info.Args()) < 1 {\n\t\t\treturn bridge.JsException(info.Context(), \"GetDel requires a key argument\")\n\t\t}\n\n\t\tkey := info.Args()[0].String()\n\t\tvalue, ok := ns.GetDel(key)\n\t\tif !ok {\n\t\t\treturn v8go.Null(iso)\n\t\t}\n\n\t\tjsValue, err := bridge.JsValue(info.Context(), value)\n\t\tif err != nil {\n\t\t\treturn v8go.Null(iso)\n\t\t}\n\n\t\treturn jsValue\n\t})\n\tgetDelFuncVal := getDelFunc.GetFunction(v8ctx)\n\tobj.Set(\"GetDel\", getDelFuncVal.Value)\n\n\treturn obj.Value\n}\n\n// sendGroupMethod implements ctx.SendGroup(group)\n// Usage: ctx.SendGroup({ id: \"group1\", messages: [...] })\n// Automatically generates IDs, sends group_start/group_end events, and flushes output\n" }, { "path": "agent/context/jsapi_agent.go", "content": "package context\n\nimport (\n\t\"github.com/yaoapp/gou/runtime/v8/bridge\"\n\t\"github.com/yaoapp/yao/agent/output/message\"\n\t\"rogchap.com/v8go\"\n)\n\n// AgentAPI defines the agent JSAPI interface for ctx.agent.*\n// This interface is defined here to avoid circular dependency between context and caller packages.\n// The actual implementation is in agent/caller/jsapi.go\ntype AgentAPI interface {\n\t// Call executes a single agent call\n\t// Returns *caller.Result or error information\n\tCall(agentID string, messages []interface{}, opts map[string]interface{}) interface{}\n\n\t// Parallel agent call methods - inspired by JavaScript Promise\n\t// All waits for all agent calls to complete (like Promise.all)\n\tAll(requests []interface{}) []interface{}\n\t// Any returns when any agent call succeeds (like Promise.any)\n\tAny(requests []interface{}) []interface{}\n\t// Race returns when any agent call completes (like Promise.race)\n\tRace(requests []interface{}) []interface{}\n}\n\n// AgentAPIWithCallback extends AgentAPI with callback support\n// This interface provides methods that accept OnMessage handlers for real-time message processing\ntype AgentAPIWithCallback interface {\n\tAgentAPI\n\n\t// CallWithHandler executes a single agent call with an OnMessage handler\n\t// handler receives SSE messages: func(msg *message.Message) int\n\tCallWithHandler(agentID string, messages []interface{}, opts map[string]interface{}, handler OnMessageFunc) interface{}\n\n\t// AllWithHandler executes all agent calls with handlers\n\t// globalHandler receives messages with agentID and index: func(agentID, index, msg) int\n\t// Individual request handlers (if set) take precedence over globalHandler\n\tAllWithHandler(requests []interface{}, globalHandler BatchOnMessageFunc) []interface{}\n\n\t// AnyWithHandler executes agent calls and returns on first success, with handlers\n\tAnyWithHandler(requests []interface{}, globalHandler BatchOnMessageFunc) []interface{}\n\n\t// RaceWithHandler executes agent calls and returns on first completion, with handlers\n\tRaceWithHandler(requests []interface{}, globalHandler BatchOnMessageFunc) []interface{}\n}\n\n// BatchOnMessageFunc is the OnMessage function for batch calls\n// It includes agentID and index to identify the source of each message\ntype BatchOnMessageFunc func(agentID string, index int, msg *message.Message) int\n\n// AgentAPIFactory is a function type that creates an AgentAPI for a context\n// This is set by the caller package during initialization\nvar AgentAPIFactory func(ctx *Context) AgentAPI\n\n// Agent returns the agent API for this context\n// Returns nil if AgentAPIFactory is not set\nfunc (ctx *Context) Agent() AgentAPI {\n\tif AgentAPIFactory == nil {\n\t\treturn nil\n\t}\n\treturn AgentAPIFactory(ctx)\n}\n\n// newAgentObject creates a new agent object with all agent methods\n// This is called from jsapi.go NewObject() to mount ctx.agent\nfunc (ctx *Context) newAgentObject(iso *v8go.Isolate) *v8go.ObjectTemplate {\n\tagentObj := v8go.NewObjectTemplate(iso)\n\n\t// Single agent call method\n\tagentObj.Set(\"Call\", ctx.agentCallMethod(iso))\n\n\t// Parallel agent call methods - inspired by JavaScript Promise\n\tagentObj.Set(\"All\", ctx.agentAllMethod(iso))\n\tagentObj.Set(\"Any\", ctx.agentAnyMethod(iso))\n\tagentObj.Set(\"Race\", ctx.agentRaceMethod(iso))\n\n\treturn agentObj\n}\n\n// agentCallMethod implements ctx.agent.Call(agentID, messages, options?)\n// Usage: const result = ctx.agent.Call(\"assistant-id\", [{ role: \"user\", content: \"Hello\" }], { connector: \"gpt4\", onChunk: (type, data) => 0 })\n// Returns: { agent_id, response, content, error }\nfunc (ctx *Context) agentCallMethod(iso *v8go.Isolate) *v8go.FunctionTemplate {\n\treturn v8go.NewFunctionTemplate(iso, func(info *v8go.FunctionCallbackInfo) *v8go.Value {\n\t\tv8ctx := info.Context()\n\t\targs := info.Args()\n\n\t\t// Validate arguments\n\t\tif len(args) < 2 {\n\t\t\treturn bridge.JsException(v8ctx, \"Call requires agentID and messages parameters\")\n\t\t}\n\n\t\t// Get agent ID (first argument)\n\t\tif !args[0].IsString() {\n\t\t\treturn bridge.JsException(v8ctx, \"agentID must be a string\")\n\t\t}\n\t\tagentID := args[0].String()\n\n\t\t// Parse messages (second argument)\n\t\tmessagesVal, err := bridge.GoValue(args[1], v8ctx)\n\t\tif err != nil {\n\t\t\treturn bridge.JsException(v8ctx, \"invalid messages: \"+err.Error())\n\t\t}\n\t\tmessages, ok := messagesVal.([]interface{})\n\t\tif !ok {\n\t\t\treturn bridge.JsException(v8ctx, \"messages must be an array\")\n\t\t}\n\n\t\t// Parse options (optional third argument) - extract onChunk separately\n\t\tvar opts map[string]interface{}\n\t\tvar onChunkFn *v8go.Function\n\n\t\tif len(args) >= 3 && !args[2].IsUndefined() && !args[2].IsNull() {\n\t\t\toptsObj, err := args[2].AsObject()\n\t\t\tif err == nil && optsObj != nil {\n\t\t\t\t// Extract onChunk callback before converting to Go value\n\t\t\t\tonChunkVal, _ := optsObj.Get(\"onChunk\")\n\t\t\t\tif onChunkVal != nil && onChunkVal.IsFunction() {\n\t\t\t\t\tonChunkFn, _ = onChunkVal.AsFunction()\n\t\t\t\t}\n\n\t\t\t\t// Convert the rest of options to Go map\n\t\t\t\tgoVal, err := bridge.GoValue(args[2], v8ctx)\n\t\t\t\tif err == nil {\n\t\t\t\t\tif optsMap, ok := goVal.(map[string]interface{}); ok {\n\t\t\t\t\t\t// Remove onChunk from the map (it's handled separately)\n\t\t\t\t\t\tdelete(optsMap, \"onChunk\")\n\t\t\t\t\t\topts = optsMap\n\t\t\t\t\t}\n\t\t\t\t}\n\t\t\t}\n\t\t}\n\n\t\t// Get agent API\n\t\tagentAPI := ctx.Agent()\n\t\tif agentAPI == nil {\n\t\t\treturn bridge.JsException(v8ctx, \"agent API not available\")\n\t\t}\n\n\t\tvar result interface{}\n\n\t\t// If onChunk callback is provided and API supports it, use CallWithHandler\n\t\tif onChunkFn != nil {\n\t\t\tif apiWithCb, ok := agentAPI.(AgentAPIWithCallback); ok {\n\t\t\t\t// Create Go StreamFunc that calls JS callback\n\t\t\t\thandler := createJSStreamHandler(v8ctx, onChunkFn)\n\t\t\t\tresult = apiWithCb.CallWithHandler(agentID, messages, opts, handler)\n\t\t\t} else {\n\t\t\t\t// Fallback: ignore callback if API doesn't support it\n\t\t\t\tresult = agentAPI.Call(agentID, messages, opts)\n\t\t\t}\n\t\t} else {\n\t\t\t// No callback, use regular Call\n\t\t\tresult = agentAPI.Call(agentID, messages, opts)\n\t\t}\n\n\t\t// Convert result to JS value\n\t\tjsVal, err := bridge.JsValue(v8ctx, result)\n\t\tif err != nil {\n\t\t\treturn bridge.JsException(v8ctx, \"failed to convert result: \"+err.Error())\n\t\t}\n\n\t\treturn jsVal\n\t})\n}\n\n// createJSOnMessageHandler creates a Go OnMessageFunc that calls a JS callback\n// JS callback signature: (msg: object) => number\n// msg contains: type, props, delta, message_id, chunk_id, etc.\nfunc createJSStreamHandler(v8ctx *v8go.Context, callback *v8go.Function) OnMessageFunc {\n\treturn func(msg *message.Message) int {\n\t\tif callback == nil || v8ctx == nil || msg == nil {\n\t\t\treturn 0 // Continue if no callback\n\t\t}\n\n\t\t// Convert message to JS value\n\t\tjsMsg, err := bridge.JsValue(v8ctx, msg)\n\t\tif err != nil {\n\t\t\treturn 1 // Stop on error\n\t\t}\n\n\t\t// Call the JS callback with the message object\n\t\tresult, err := callback.Call(v8ctx.Global(), jsMsg)\n\t\tif err != nil {\n\t\t\treturn 1 // Stop on error\n\t\t}\n\n\t\t// Check return value (0 = continue, non-zero = stop)\n\t\tif result != nil && result.IsNumber() {\n\t\t\tret := result.Integer()\n\t\t\tif ret != 0 {\n\t\t\t\treturn int(ret)\n\t\t\t}\n\t\t}\n\n\t\treturn 0 // Continue\n\t}\n}\n\n// agentAllMethod implements ctx.agent.All(requests, options?)\n// Waits for all agent calls to complete (like Promise.all)\n// Each request should have:\n// - agent: string - target agent ID\n// - messages: array - messages to send\n// - options?: object - call options\n//\n// Global options (second argument):\n// - onChunk?: (agentID, index, msg) => number - callback for all messages (uses channel for V8 safety)\nfunc (ctx *Context) agentAllMethod(iso *v8go.Isolate) *v8go.FunctionTemplate {\n\treturn v8go.NewFunctionTemplate(iso, func(info *v8go.FunctionCallbackInfo) *v8go.Value {\n\t\tv8ctx := info.Context()\n\t\targs := info.Args()\n\n\t\t// Validate arguments\n\t\tif len(args) < 1 {\n\t\t\treturn bridge.JsException(v8ctx, \"All requires requests parameter\")\n\t\t}\n\n\t\t// Parse requests and extract global callback\n\t\trequests, globalCallback := ctx.parseRequestsForBatch(args, v8ctx)\n\n\t\t// Get agent API\n\t\tagentAPI := ctx.Agent()\n\t\tif agentAPI == nil {\n\t\t\treturn bridge.JsException(v8ctx, \"agent API not available\")\n\t\t}\n\n\t\t// Execute with channel-based callback handling\n\t\tresults := ctx.executeBatchWithCallback(BatchMethodAll, requests, globalCallback, v8ctx)\n\n\t\t// Convert results to JS value\n\t\tjsVal, err := bridge.JsValue(v8ctx, results)\n\t\tif err != nil {\n\t\t\treturn bridge.JsException(v8ctx, \"failed to convert results: \"+err.Error())\n\t\t}\n\n\t\treturn jsVal\n\t})\n}\n\n// agentAnyMethod implements ctx.agent.Any(requests, options?)\n// Returns when any agent call succeeds (like Promise.any)\n// Each request should have:\n// - agent: string - target agent ID\n// - messages: array - messages to send\n// - options?: object - call options\n//\n// Global options (second argument):\n// - onChunk?: (agentID, index, msg) => number - callback for all messages (uses channel for V8 safety)\nfunc (ctx *Context) agentAnyMethod(iso *v8go.Isolate) *v8go.FunctionTemplate {\n\treturn v8go.NewFunctionTemplate(iso, func(info *v8go.FunctionCallbackInfo) *v8go.Value {\n\t\tv8ctx := info.Context()\n\t\targs := info.Args()\n\n\t\t// Validate arguments\n\t\tif len(args) < 1 {\n\t\t\treturn bridge.JsException(v8ctx, \"Any requires requests parameter\")\n\t\t}\n\n\t\t// Parse requests and extract global callback\n\t\trequests, globalCallback := ctx.parseRequestsForBatch(args, v8ctx)\n\n\t\t// Get agent API\n\t\tagentAPI := ctx.Agent()\n\t\tif agentAPI == nil {\n\t\t\treturn bridge.JsException(v8ctx, \"agent API not available\")\n\t\t}\n\n\t\t// Execute with channel-based callback handling\n\t\tresults := ctx.executeBatchWithCallback(BatchMethodAny, requests, globalCallback, v8ctx)\n\n\t\t// Convert results to JS value\n\t\tjsVal, err := bridge.JsValue(v8ctx, results)\n\t\tif err != nil {\n\t\t\treturn bridge.JsException(v8ctx, \"failed to convert results: \"+err.Error())\n\t\t}\n\n\t\treturn jsVal\n\t})\n}\n\n// agentRaceMethod implements ctx.agent.Race(requests, options?)\n// Returns when any agent call completes (like Promise.race)\n// Each request should have:\n// - agent: string - target agent ID\n// - messages: array - messages to send\n// - options?: object - call options\n//\n// Global options (second argument):\n// - onChunk?: (agentID, index, msg) => number - callback for all messages (uses channel for V8 safety)\nfunc (ctx *Context) agentRaceMethod(iso *v8go.Isolate) *v8go.FunctionTemplate {\n\treturn v8go.NewFunctionTemplate(iso, func(info *v8go.FunctionCallbackInfo) *v8go.Value {\n\t\tv8ctx := info.Context()\n\t\targs := info.Args()\n\n\t\t// Validate arguments\n\t\tif len(args) < 1 {\n\t\t\treturn bridge.JsException(v8ctx, \"Race requires requests parameter\")\n\t\t}\n\n\t\t// Parse requests and extract global callback\n\t\trequests, globalCallback := ctx.parseRequestsForBatch(args, v8ctx)\n\n\t\t// Get agent API\n\t\tagentAPI := ctx.Agent()\n\t\tif agentAPI == nil {\n\t\t\treturn bridge.JsException(v8ctx, \"agent API not available\")\n\t\t}\n\n\t\t// Execute with channel-based callback handling\n\t\tresults := ctx.executeBatchWithCallback(BatchMethodRace, requests, globalCallback, v8ctx)\n\n\t\t// Convert results to JS value\n\t\tjsVal, err := bridge.JsValue(v8ctx, results)\n\t\tif err != nil {\n\t\t\treturn bridge.JsException(v8ctx, \"failed to convert results: \"+err.Error())\n\t\t}\n\n\t\treturn jsVal\n\t})\n}\n\n// batchMessage represents a message from a batch call for channel-based callback handling\ntype batchMessage struct {\n\tAgentID string // Agent ID that generated this message\n\tIndex int // Index of the request in the batch\n\tMessage *message.Message // The message object\n}\n\n// parseRequestsForBatch parses the requests array and extracts global callback for batch calls\n// Returns the requests array and the global JS callback function (if any)\nfunc (ctx *Context) parseRequestsForBatch(args []*v8go.Value, v8ctx *v8go.Context) ([]interface{}, *v8go.Function) {\n\tvar globalCallback *v8go.Function\n\n\t// Parse global options (second argument) for global onChunk\n\tif len(args) >= 2 && !args[1].IsUndefined() && !args[1].IsNull() {\n\t\tglobalOptsObj, err := args[1].AsObject()\n\t\tif err == nil && globalOptsObj != nil {\n\t\t\tonChunkVal, _ := globalOptsObj.Get(\"onChunk\")\n\t\t\tif onChunkVal != nil && onChunkVal.IsFunction() {\n\t\t\t\tglobalCallback, _ = onChunkVal.AsFunction()\n\t\t\t}\n\t\t}\n\t}\n\n\t// Parse requests array\n\tif len(args) < 1 || args[0].IsUndefined() || args[0].IsNull() {\n\t\treturn []interface{}{}, globalCallback\n\t}\n\n\trequestsObj, err := args[0].AsObject()\n\tif err != nil {\n\t\treturn []interface{}{}, globalCallback\n\t}\n\n\t// Get array length\n\tlengthVal, err := requestsObj.Get(\"length\")\n\tif err != nil {\n\t\treturn []interface{}{}, globalCallback\n\t}\n\n\tlength := int(lengthVal.Integer())\n\trequests := make([]interface{}, 0, length)\n\n\tfor i := 0; i < length; i++ {\n\t\titemVal, err := requestsObj.GetIdx(uint32(i))\n\t\tif err != nil || itemVal.IsUndefined() || itemVal.IsNull() {\n\t\t\tcontinue\n\t\t}\n\n\t\t// Convert to Go map\n\t\tgoVal, err := bridge.GoValue(itemVal, v8ctx)\n\t\tif err != nil {\n\t\t\tcontinue\n\t\t}\n\n\t\treqMap, ok := goVal.(map[string]interface{})\n\t\tif !ok {\n\t\t\tcontinue\n\t\t}\n\n\t\t// Remove onChunk from per-request options (only global callback is supported)\n\t\tif opts, ok := reqMap[\"options\"].(map[string]interface{}); ok {\n\t\t\tdelete(opts, \"onChunk\")\n\t\t}\n\n\t\trequests = append(requests, reqMap)\n\t}\n\n\treturn requests, globalCallback\n}\n\n// BatchMethod represents the type of batch operation\ntype BatchMethod int\n\nconst (\n\tBatchMethodAll BatchMethod = iota\n\tBatchMethodAny\n\tBatchMethodRace\n)\n\n// executeBatchWithCallback executes a batch operation with channel-based callback handling\n// This ensures V8 thread safety by processing all callbacks in the main goroutine\nfunc (ctx *Context) executeBatchWithCallback(\n\tmethod BatchMethod,\n\trequests []interface{},\n\tcallback *v8go.Function,\n\tv8ctx *v8go.Context,\n) []interface{} {\n\t// Get agent API\n\tagentAPI := ctx.Agent()\n\tif agentAPI == nil {\n\t\treturn []interface{}{}\n\t}\n\n\t// If no callback, just execute directly\n\tif callback == nil {\n\t\tswitch method {\n\t\tcase BatchMethodAll:\n\t\t\treturn agentAPI.All(requests)\n\t\tcase BatchMethodAny:\n\t\t\treturn agentAPI.Any(requests)\n\t\tcase BatchMethodRace:\n\t\t\treturn agentAPI.Race(requests)\n\t\t}\n\t\treturn []interface{}{}\n\t}\n\n\t// Check if API supports callbacks\n\tapiWithCb, ok := agentAPI.(AgentAPIWithCallback)\n\tif !ok {\n\t\tswitch method {\n\t\tcase BatchMethodAll:\n\t\t\treturn agentAPI.All(requests)\n\t\tcase BatchMethodAny:\n\t\t\treturn agentAPI.Any(requests)\n\t\tcase BatchMethodRace:\n\t\t\treturn agentAPI.Race(requests)\n\t\t}\n\t\treturn []interface{}{}\n\t}\n\n\t// Create message channel for callback handling\n\t// Use a large buffer (1000) to reduce blocking, with blocking send to guarantee no message loss\n\tmsgChan := make(chan batchMessage, 1000)\n\tdoneChan := make(chan []interface{}, 1)\n\n\t// Create Go handler that sends messages to channel\n\t// Blocking send ensures no message is lost (natural backpressure)\n\tgoHandler := func(agentID string, index int, msg *message.Message) int {\n\t\tmsgChan <- batchMessage{AgentID: agentID, Index: index, Message: msg}\n\t\treturn 0\n\t}\n\n\t// Start batch execution in background goroutine\n\tgo func() {\n\t\tdefer close(msgChan)\n\t\tvar results []interface{}\n\n\t\tswitch method {\n\t\tcase BatchMethodAll:\n\t\t\tresults = apiWithCb.AllWithHandler(requests, goHandler)\n\t\tcase BatchMethodAny:\n\t\t\tresults = apiWithCb.AnyWithHandler(requests, goHandler)\n\t\tcase BatchMethodRace:\n\t\t\tresults = apiWithCb.RaceWithHandler(requests, goHandler)\n\t\t}\n\n\t\tdoneChan <- results\n\t}()\n\n\t// Process messages in main goroutine (V8 thread-safe)\n\tfor msg := range msgChan {\n\t\tcallJSBatchCallback(v8ctx, callback, msg.AgentID, msg.Index, msg.Message)\n\t}\n\n\t// Wait for results\n\treturn <-doneChan\n}\n\n// callJSBatchCallback calls the JS callback with batch message parameters\n// Must be called from the main V8 goroutine\nfunc callJSBatchCallback(v8ctx *v8go.Context, callback *v8go.Function, agentID string, index int, msg *message.Message) {\n\tif callback == nil || v8ctx == nil || msg == nil {\n\t\treturn\n\t}\n\n\tiso := v8ctx.Isolate()\n\n\tagentIDVal, err := v8go.NewValue(iso, agentID)\n\tif err != nil {\n\t\treturn\n\t}\n\n\tindexVal, err := v8go.NewValue(iso, int32(index))\n\tif err != nil {\n\t\treturn\n\t}\n\n\t// Convert message to JS value\n\tjsMsg, err := bridge.JsValue(v8ctx, msg)\n\tif err != nil {\n\t\treturn\n\t}\n\n\tcallback.Call(v8ctx.Global(), agentIDVal, indexVal, jsMsg)\n}\n" }, { "path": "agent/context/jsapi_agent_test.go", "content": "package context_test\n\nimport (\n\tstdContext \"context\"\n\t\"testing\"\n\n\t\"github.com/stretchr/testify/assert\"\n\t\"github.com/stretchr/testify/require\"\n\t\"github.com/yaoapp/yao/agent/context\"\n)\n\nfunc TestContext_Agent_NilFactory(t *testing.T) {\n\t// Reset factory\n\tcontext.AgentAPIFactory = nil\n\n\tctx := context.New(stdContext.Background(), nil, \"test-chat\")\n\tagentAPI := ctx.Agent()\n\tassert.Nil(t, agentAPI)\n}\n\nfunc TestContext_Agent_WithFactory(t *testing.T) {\n\t// Set up a mock factory\n\tvar capturedCtx *context.Context\n\tcontext.AgentAPIFactory = func(ctx *context.Context) context.AgentAPI {\n\t\tcapturedCtx = ctx\n\t\treturn &mockAgentAPI{}\n\t}\n\tdefer func() { context.AgentAPIFactory = nil }()\n\n\tctx := context.New(stdContext.Background(), nil, \"test-chat\")\n\tagentAPI := ctx.Agent()\n\n\trequire.NotNil(t, agentAPI)\n\tassert.Equal(t, ctx, capturedCtx)\n}\n\n// mockAgentAPI implements context.AgentAPI for testing\ntype mockAgentAPI struct{}\n\nfunc (m *mockAgentAPI) Call(agentID string, messages []interface{}, opts map[string]interface{}) interface{} {\n\treturn map[string]interface{}{\n\t\t\"agent_id\": agentID,\n\t\t\"content\": \"mock response\",\n\t}\n}\n\nfunc (m *mockAgentAPI) All(requests []interface{}) []interface{} {\n\treturn []interface{}{}\n}\n\nfunc (m *mockAgentAPI) Any(requests []interface{}) []interface{} {\n\treturn []interface{}{}\n}\n\nfunc (m *mockAgentAPI) Race(requests []interface{}) []interface{} {\n\treturn []interface{}{}\n}\n" }, { "path": "agent/context/jsapi_agent_v8_test.go", "content": "package context_test\n\nimport (\n\tstdContext \"context\"\n\t\"strings\"\n\t\"testing\"\n\n\t\"github.com/stretchr/testify/assert\"\n\t\"github.com/stretchr/testify/require\"\n\tv8 \"github.com/yaoapp/gou/runtime/v8\"\n\t\"github.com/yaoapp/yao/agent/context\"\n\t\"github.com/yaoapp/yao/agent/testutils\"\n\t\"github.com/yaoapp/yao/openapi/oauth/types\"\n\n\t// Import assistant package to register AgentAPIFactory\n\t_ \"github.com/yaoapp/yao/agent/assistant\"\n)\n\n// TestAgent_Call_V8 tests basic ctx.agent.Call() functionality with real V8 execution\nfunc TestAgent_Call_V8(t *testing.T) {\n\tif testing.Short() {\n\t\tt.Skip(\"Skipping integration test in short mode\")\n\t}\n\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\t// Create authorized info for the context\n\tauthorized := &types.AuthorizedInfo{\n\t\tSubject: \"test-user\",\n\t\tUserID: \"test-123\",\n\t\tTenantID: \"test-tenant\",\n\t}\n\n\tctx := context.New(stdContext.Background(), authorized, \"test-chat-v8-call\")\n\tctx.AssistantID = \"tests.agent-caller\"\n\tdefer ctx.Release()\n\n\tres, err := v8.Call(v8.CallOptions{}, `\n\t\tfunction test(ctx) {\n\t\t\ttry {\n\t\t\t\tconst result = ctx.agent.Call(\n\t\t\t\t\t\"tests.simple-greeting\",\n\t\t\t\t\t[{ role: \"user\", content: \"Hello\" }]\n\t\t\t\t);\n\t\t\t\t\n\t\t\t\treturn { \n\t\t\t\t\tsuccess: true,\n\t\t\t\t\tagent_id: result.agent_id,\n\t\t\t\t\thas_content: result.content && result.content.length > 0,\n\t\t\t\t\thas_response: result.response !== undefined,\n\t\t\t\t\terror: result.error || \"\"\n\t\t\t\t};\n\t\t\t} catch (error) {\n\t\t\t\treturn { success: false, error: error.message };\n\t\t\t}\n\t\t}`, ctx)\n\n\trequire.NoError(t, err)\n\tresult, ok := res.(map[string]interface{})\n\trequire.True(t, ok, \"Result should be a map\")\n\n\tsuccess, _ := result[\"success\"].(bool)\n\tif !success {\n\t\tt.Fatalf(\"Test failed: %v\", result[\"error\"])\n\t}\n\n\tassert.Equal(t, \"tests.simple-greeting\", result[\"agent_id\"])\n\n\thasContent, _ := result[\"has_content\"].(bool)\n\tassert.True(t, hasContent, \"Should have content in response\")\n\n\thasResponse, _ := result[\"has_response\"].(bool)\n\tassert.True(t, hasResponse, \"Should have response object\")\n\n\terrorStr, _ := result[\"error\"].(string)\n\tassert.Empty(t, errorStr, \"Should not have error\")\n}\n\n// TestAgent_Call_WithOptions_V8 tests ctx.agent.Call() with options\nfunc TestAgent_Call_WithOptions_V8(t *testing.T) {\n\tif testing.Short() {\n\t\tt.Skip(\"Skipping integration test in short mode\")\n\t}\n\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\tauthorized := &types.AuthorizedInfo{\n\t\tSubject: \"test-user\",\n\t\tUserID: \"test-123\",\n\t\tTenantID: \"test-tenant\",\n\t}\n\n\tctx := context.New(stdContext.Background(), authorized, \"test-chat-v8-options\")\n\tctx.AssistantID = \"tests.agent-caller\"\n\tdefer ctx.Release()\n\n\tres, err := v8.Call(v8.CallOptions{}, `\n\t\tfunction test(ctx) {\n\t\t\ttry {\n\t\t\t\tconst result = ctx.agent.Call(\n\t\t\t\t\t\"tests.simple-greeting\",\n\t\t\t\t\t[{ role: \"user\", content: \"Hi there!\" }],\n\t\t\t\t\t{\n\t\t\t\t\t\tskip: {\n\t\t\t\t\t\t\thistory: true,\n\t\t\t\t\t\t\ttrace: true\n\t\t\t\t\t\t}\n\t\t\t\t\t}\n\t\t\t\t);\n\t\t\t\t\n\t\t\t\treturn { \n\t\t\t\t\tsuccess: true,\n\t\t\t\t\tagent_id: result.agent_id,\n\t\t\t\t\tcontent: result.content || \"\",\n\t\t\t\t\terror: result.error || \"\"\n\t\t\t\t};\n\t\t\t} catch (error) {\n\t\t\t\treturn { success: false, error: error.message };\n\t\t\t}\n\t\t}`, ctx)\n\n\trequire.NoError(t, err)\n\tresult := res.(map[string]interface{})\n\n\tif !result[\"success\"].(bool) {\n\t\tt.Fatalf(\"Test failed: %v\", result[\"error\"])\n\t}\n\n\tassert.Equal(t, \"tests.simple-greeting\", result[\"agent_id\"])\n\tassert.NotEmpty(t, result[\"content\"], \"Should have content\")\n}\n\n// TestAgent_All_V8 tests ctx.agent.All() for parallel execution\nfunc TestAgent_All_V8(t *testing.T) {\n\tif testing.Short() {\n\t\tt.Skip(\"Skipping integration test in short mode\")\n\t}\n\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\tauthorized := &types.AuthorizedInfo{\n\t\tSubject: \"test-user\",\n\t\tUserID: \"test-123\",\n\t\tTenantID: \"test-tenant\",\n\t}\n\n\tctx := context.New(stdContext.Background(), authorized, \"test-chat-v8-all\")\n\tctx.AssistantID = \"tests.agent-caller\"\n\tdefer ctx.Release()\n\n\tres, err := v8.Call(v8.CallOptions{}, `\n\t\tfunction test(ctx) {\n\t\t\ttry {\n\t\t\t\tconst results = ctx.agent.All([\n\t\t\t\t\t{\n\t\t\t\t\t\tagent: \"tests.simple-greeting\",\n\t\t\t\t\t\tmessages: [{ role: \"user\", content: \"Hello from request 1\" }]\n\t\t\t\t\t},\n\t\t\t\t\t{\n\t\t\t\t\t\tagent: \"tests.simple-greeting\",\n\t\t\t\t\t\tmessages: [{ role: \"user\", content: \"Hello from request 2\" }]\n\t\t\t\t\t}\n\t\t\t\t]);\n\t\t\t\t\n\t\t\t\treturn { \n\t\t\t\t\tsuccess: true,\n\t\t\t\t\tcount: results.length,\n\t\t\t\t\tfirst_agent: results[0] ? results[0].agent_id : \"\",\n\t\t\t\t\tsecond_agent: results[1] ? results[1].agent_id : \"\",\n\t\t\t\t\tfirst_has_content: results[0] && results[0].content && results[0].content.length > 0,\n\t\t\t\t\tsecond_has_content: results[1] && results[1].content && results[1].content.length > 0,\n\t\t\t\t\tfirst_error: results[0] ? (results[0].error || \"\") : \"no result\",\n\t\t\t\t\tsecond_error: results[1] ? (results[1].error || \"\") : \"no result\"\n\t\t\t\t};\n\t\t\t} catch (error) {\n\t\t\t\treturn { success: false, error: error.message };\n\t\t\t}\n\t\t}`, ctx)\n\n\trequire.NoError(t, err)\n\tresult := res.(map[string]interface{})\n\n\tif !result[\"success\"].(bool) {\n\t\tt.Fatalf(\"Test failed: %v\", result[\"error\"])\n\t}\n\n\tassert.Equal(t, float64(2), result[\"count\"])\n\tassert.Equal(t, \"tests.simple-greeting\", result[\"first_agent\"])\n\tassert.Equal(t, \"tests.simple-greeting\", result[\"second_agent\"])\n\tassert.True(t, result[\"first_has_content\"].(bool), \"First result should have content\")\n\tassert.True(t, result[\"second_has_content\"].(bool), \"Second result should have content\")\n\tassert.Empty(t, result[\"first_error\"], \"First result should not have error\")\n\tassert.Empty(t, result[\"second_error\"], \"Second result should not have error\")\n}\n\n// TestAgent_Any_V8 tests ctx.agent.Any() returns on first success\nfunc TestAgent_Any_V8(t *testing.T) {\n\tif testing.Short() {\n\t\tt.Skip(\"Skipping integration test in short mode\")\n\t}\n\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\tauthorized := &types.AuthorizedInfo{\n\t\tSubject: \"test-user\",\n\t\tUserID: \"test-123\",\n\t\tTenantID: \"test-tenant\",\n\t}\n\n\tctx := context.New(stdContext.Background(), authorized, \"test-chat-v8-any\")\n\tctx.AssistantID = \"tests.agent-caller\"\n\tdefer ctx.Release()\n\n\tres, err := v8.Call(v8.CallOptions{}, `\n\t\tfunction test(ctx) {\n\t\t\ttry {\n\t\t\t\tconst results = ctx.agent.Any([\n\t\t\t\t\t{\n\t\t\t\t\t\tagent: \"tests.simple-greeting\",\n\t\t\t\t\t\tmessages: [{ role: \"user\", content: \"Hello\" }]\n\t\t\t\t\t},\n\t\t\t\t\t{\n\t\t\t\t\t\tagent: \"tests.simple-greeting\",\n\t\t\t\t\t\tmessages: [{ role: \"user\", content: \"Hi\" }]\n\t\t\t\t\t}\n\t\t\t\t]);\n\t\t\t\t\n\t\t\t\t// At least one result should be successful\n\t\t\t\tlet hasSuccess = false;\n\t\t\t\tfor (const r of results) {\n\t\t\t\t\tif (r && r.content && !r.error) {\n\t\t\t\t\t\thasSuccess = true;\n\t\t\t\t\t\tbreak;\n\t\t\t\t\t}\n\t\t\t\t}\n\t\t\t\t\n\t\t\t\treturn { \n\t\t\t\t\tsuccess: true,\n\t\t\t\t\tcount: results.length,\n\t\t\t\t\thas_successful_result: hasSuccess\n\t\t\t\t};\n\t\t\t} catch (error) {\n\t\t\t\treturn { success: false, error: error.message };\n\t\t\t}\n\t\t}`, ctx)\n\n\trequire.NoError(t, err)\n\tresult := res.(map[string]interface{})\n\n\tif !result[\"success\"].(bool) {\n\t\tt.Fatalf(\"Test failed: %v\", result[\"error\"])\n\t}\n\n\tassert.Equal(t, float64(2), result[\"count\"])\n\tassert.True(t, result[\"has_successful_result\"].(bool), \"Should have at least one successful result\")\n}\n\n// TestAgent_Race_V8 tests ctx.agent.Race() returns on first completion\nfunc TestAgent_Race_V8(t *testing.T) {\n\tif testing.Short() {\n\t\tt.Skip(\"Skipping integration test in short mode\")\n\t}\n\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\tauthorized := &types.AuthorizedInfo{\n\t\tSubject: \"test-user\",\n\t\tUserID: \"test-123\",\n\t\tTenantID: \"test-tenant\",\n\t}\n\n\tctx := context.New(stdContext.Background(), authorized, \"test-chat-v8-race\")\n\tctx.AssistantID = \"tests.agent-caller\"\n\tdefer ctx.Release()\n\n\tres, err := v8.Call(v8.CallOptions{}, `\n\t\tfunction test(ctx) {\n\t\t\ttry {\n\t\t\t\tconst results = ctx.agent.Race([\n\t\t\t\t\t{\n\t\t\t\t\t\tagent: \"tests.simple-greeting\",\n\t\t\t\t\t\tmessages: [{ role: \"user\", content: \"Hello\" }]\n\t\t\t\t\t},\n\t\t\t\t\t{\n\t\t\t\t\t\tagent: \"tests.simple-greeting\",\n\t\t\t\t\t\tmessages: [{ role: \"user\", content: \"Hi\" }]\n\t\t\t\t\t}\n\t\t\t\t]);\n\t\t\t\t\n\t\t\t\t// At least one result should exist (first to complete)\n\t\t\t\tlet hasResult = false;\n\t\t\t\tfor (const r of results) {\n\t\t\t\t\tif (r && (r.content || r.error)) {\n\t\t\t\t\t\thasResult = true;\n\t\t\t\t\t\tbreak;\n\t\t\t\t\t}\n\t\t\t\t}\n\t\t\t\t\n\t\t\t\treturn { \n\t\t\t\t\tsuccess: true,\n\t\t\t\t\tcount: results.length,\n\t\t\t\t\thas_result: hasResult\n\t\t\t\t};\n\t\t\t} catch (error) {\n\t\t\t\treturn { success: false, error: error.message };\n\t\t\t}\n\t\t}`, ctx)\n\n\trequire.NoError(t, err)\n\tresult := res.(map[string]interface{})\n\n\tif !result[\"success\"].(bool) {\n\t\tt.Fatalf(\"Test failed: %v\", result[\"error\"])\n\t}\n\n\tassert.Equal(t, float64(2), result[\"count\"])\n\tassert.True(t, result[\"has_result\"].(bool), \"Should have at least one result\")\n}\n\n// TestAgent_ErrorHandling_V8 tests error handling when calling non-existent agent\nfunc TestAgent_ErrorHandling_V8(t *testing.T) {\n\tif testing.Short() {\n\t\tt.Skip(\"Skipping integration test in short mode\")\n\t}\n\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\tauthorized := &types.AuthorizedInfo{\n\t\tSubject: \"test-user\",\n\t\tUserID: \"test-123\",\n\t\tTenantID: \"test-tenant\",\n\t}\n\n\tctx := context.New(stdContext.Background(), authorized, \"test-chat-v8-error\")\n\tctx.AssistantID = \"tests.agent-caller\"\n\tdefer ctx.Release()\n\n\tres, err := v8.Call(v8.CallOptions{}, `\n\t\tfunction test(ctx) {\n\t\t\ttry {\n\t\t\t\tconst result = ctx.agent.Call(\n\t\t\t\t\t\"non-existent-agent\",\n\t\t\t\t\t[{ role: \"user\", content: \"Hello\" }]\n\t\t\t\t);\n\t\t\t\t\n\t\t\t\treturn { \n\t\t\t\t\tsuccess: true,\n\t\t\t\t\thas_error: result.error && result.error.length > 0,\n\t\t\t\t\terror_message: result.error || \"\"\n\t\t\t\t};\n\t\t\t} catch (error) {\n\t\t\t\treturn { success: false, error: error.message };\n\t\t\t}\n\t\t}`, ctx)\n\n\trequire.NoError(t, err)\n\tresult := res.(map[string]interface{})\n\n\t// The call should succeed (no JS exception), but result should contain error\n\tassert.True(t, result[\"success\"].(bool), \"JS execution should succeed\")\n\tassert.True(t, result[\"has_error\"].(bool), \"Result should have error for non-existent agent\")\n\tassert.True(t, strings.Contains(result[\"error_message\"].(string), \"failed to get agent\"), \"Error should mention failed to get agent\")\n}\n\n// TestAgent_EmptyRequests_V8 tests handling of empty requests array\nfunc TestAgent_EmptyRequests_V8(t *testing.T) {\n\tif testing.Short() {\n\t\tt.Skip(\"Skipping integration test in short mode\")\n\t}\n\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\tauthorized := &types.AuthorizedInfo{\n\t\tSubject: \"test-user\",\n\t\tUserID: \"test-123\",\n\t\tTenantID: \"test-tenant\",\n\t}\n\n\tctx := context.New(stdContext.Background(), authorized, \"test-chat-v8-empty\")\n\tctx.AssistantID = \"tests.agent-caller\"\n\tdefer ctx.Release()\n\n\tres, err := v8.Call(v8.CallOptions{}, `\n\t\tfunction test(ctx) {\n\t\t\ttry {\n\t\t\t\tconst results = ctx.agent.All([]);\n\t\t\t\treturn { \n\t\t\t\t\tsuccess: true,\n\t\t\t\t\tcount: results.length\n\t\t\t\t};\n\t\t\t} catch (error) {\n\t\t\t\treturn { success: false, error: error.message };\n\t\t\t}\n\t\t}`, ctx)\n\n\trequire.NoError(t, err)\n\tresult := res.(map[string]interface{})\n\n\tassert.True(t, result[\"success\"].(bool))\n\tassert.Equal(t, float64(0), result[\"count\"])\n}\n\n// TestAgent_InvalidArguments_V8 tests error handling for invalid arguments\nfunc TestAgent_InvalidArguments_V8(t *testing.T) {\n\tif testing.Short() {\n\t\tt.Skip(\"Skipping integration test in short mode\")\n\t}\n\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\tauthorized := &types.AuthorizedInfo{\n\t\tSubject: \"test-user\",\n\t\tUserID: \"test-123\",\n\t\tTenantID: \"test-tenant\",\n\t}\n\n\tctx := context.New(stdContext.Background(), authorized, \"test-chat-v8-invalid\")\n\tctx.AssistantID = \"tests.agent-caller\"\n\tdefer ctx.Release()\n\n\t// Test missing arguments\n\tres, err := v8.Call(v8.CallOptions{}, `\n\t\tfunction test(ctx) {\n\t\t\ttry {\n\t\t\t\t// Call with no arguments should throw\n\t\t\t\tctx.agent.Call();\n\t\t\t\treturn { success: false, error: \"Should have thrown\" };\n\t\t\t} catch (error) {\n\t\t\t\treturn { success: true, error: error.message };\n\t\t\t}\n\t\t}`, ctx)\n\n\trequire.NoError(t, err)\n\tresult := res.(map[string]interface{})\n\tassert.True(t, result[\"success\"].(bool), \"Should catch the error\")\n\tassert.Contains(t, result[\"error\"].(string), \"requires\")\n}\n\n// ============================================================================\n// Callback Tests\n// ============================================================================\n\n// TestAgent_Call_WithCallback_V8 tests ctx.agent.Call() with onChunk callback\nfunc TestAgent_Call_WithCallback_V8(t *testing.T) {\n\tif testing.Short() {\n\t\tt.Skip(\"Skipping integration test in short mode\")\n\t}\n\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\tauthorized := &types.AuthorizedInfo{\n\t\tSubject: \"test-user\",\n\t\tUserID: \"test-123\",\n\t\tTenantID: \"test-tenant\",\n\t}\n\n\tctx := context.New(stdContext.Background(), authorized, \"test-chat-v8-callback\")\n\tctx.AssistantID = \"tests.agent-caller\"\n\tdefer ctx.Release()\n\n\tres, err := v8.Call(v8.CallOptions{}, `\n\t\tfunction test(ctx) {\n\t\t\ttry {\n\t\t\t\tconst messages = [];\n\t\t\t\tlet messageCount = 0;\n\t\t\t\t\n\t\t\t\tconst result = ctx.agent.Call(\n\t\t\t\t\t\"tests.simple-greeting\",\n\t\t\t\t\t[{ role: \"user\", content: \"Hello\" }],\n\t\t\t\t\t{\n\t\t\t\t\t\tonChunk: (msg) => {\n\t\t\t\t\t\t\t// msg is the SSE message object\n\t\t\t\t\t\t\tmessageCount++;\n\t\t\t\t\t\t\tmessages.push({\n\t\t\t\t\t\t\t\ttype: msg.type,\n\t\t\t\t\t\t\t\thas_props: msg.props !== undefined\n\t\t\t\t\t\t\t});\n\t\t\t\t\t\t\treturn 0; // Continue\n\t\t\t\t\t\t}\n\t\t\t\t\t}\n\t\t\t\t);\n\t\t\t\t\n\t\t\t\treturn { \n\t\t\t\t\tsuccess: true,\n\t\t\t\t\tagent_id: result.agent_id,\n\t\t\t\t\thas_content: result.content && result.content.length > 0,\n\t\t\t\t\tmessage_count: messageCount,\n\t\t\t\t\treceived_messages: messages.slice(0, 5), // First 5 messages\n\t\t\t\t\terror: result.error || \"\"\n\t\t\t\t};\n\t\t\t} catch (error) {\n\t\t\t\treturn { success: false, error: error.message };\n\t\t\t}\n\t\t}`, ctx)\n\n\trequire.NoError(t, err)\n\tresult, ok := res.(map[string]interface{})\n\trequire.True(t, ok, \"Result should be a map\")\n\n\tsuccess, _ := result[\"success\"].(bool)\n\tif !success {\n\t\tt.Fatalf(\"Test failed: %v\", result[\"error\"])\n\t}\n\n\tassert.Equal(t, \"tests.simple-greeting\", result[\"agent_id\"])\n\n\t// Should have received some messages via callback\n\tmessageCount, _ := result[\"message_count\"].(float64)\n\tt.Logf(\"Received %v messages via callback\", messageCount)\n\tassert.Greater(t, messageCount, float64(0), \"Should have received messages via callback\")\n\n\t// Check that we received message objects with type and props\n\treceivedMsgs, _ := result[\"received_messages\"].([]interface{})\n\tif len(receivedMsgs) > 0 {\n\t\tfirstMsg := receivedMsgs[0].(map[string]interface{})\n\t\tt.Logf(\"First message type: %v\", firstMsg[\"type\"])\n\t\tassert.NotEmpty(t, firstMsg[\"type\"], \"Message should have type\")\n\t}\n}\n\n// TestAgent_Call_WithCallback_Stop_V8 tests that callback can stop streaming\nfunc TestAgent_Call_WithCallback_Stop_V8(t *testing.T) {\n\tif testing.Short() {\n\t\tt.Skip(\"Skipping integration test in short mode\")\n\t}\n\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\tauthorized := &types.AuthorizedInfo{\n\t\tSubject: \"test-user\",\n\t\tUserID: \"test-123\",\n\t\tTenantID: \"test-tenant\",\n\t}\n\n\tctx := context.New(stdContext.Background(), authorized, \"test-chat-v8-callback-stop\")\n\tctx.AssistantID = \"tests.agent-caller\"\n\tdefer ctx.Release()\n\n\tres, err := v8.Call(v8.CallOptions{}, `\n\t\tfunction test(ctx) {\n\t\t\ttry {\n\t\t\t\tlet messageCount = 0;\n\t\t\t\t\n\t\t\t\tconst result = ctx.agent.Call(\n\t\t\t\t\t\"tests.simple-greeting\",\n\t\t\t\t\t[{ role: \"user\", content: \"Hello\" }],\n\t\t\t\t\t{\n\t\t\t\t\t\tonChunk: (msg) => {\n\t\t\t\t\t\t\tmessageCount++;\n\t\t\t\t\t\t\t// Stop after receiving 3 messages\n\t\t\t\t\t\t\tif (messageCount >= 3) {\n\t\t\t\t\t\t\t\treturn 1; // Stop\n\t\t\t\t\t\t\t}\n\t\t\t\t\t\t\treturn 0; // Continue\n\t\t\t\t\t\t}\n\t\t\t\t\t}\n\t\t\t\t);\n\t\t\t\t\n\t\t\t\treturn { \n\t\t\t\t\tsuccess: true,\n\t\t\t\t\tmessage_count: messageCount,\n\t\t\t\t\tstopped_early: messageCount <= 5 // Should have stopped early\n\t\t\t\t};\n\t\t\t} catch (error) {\n\t\t\t\treturn { success: false, error: error.message };\n\t\t\t}\n\t\t}`, ctx)\n\n\trequire.NoError(t, err)\n\tresult, ok := res.(map[string]interface{})\n\trequire.True(t, ok, \"Result should be a map\")\n\n\tsuccess, _ := result[\"success\"].(bool)\n\tif !success {\n\t\tt.Fatalf(\"Test failed: %v\", result[\"error\"])\n\t}\n\n\tmessageCount, _ := result[\"message_count\"].(float64)\n\tt.Logf(\"Received %v messages before stopping\", messageCount)\n\t// Note: The exact count may vary based on when the stop is processed\n}\n\n// TestAgent_All_WithGlobalCallback_V8 tests ctx.agent.All() with global onChunk callback\n// Uses channel-based callback handling for V8 thread safety\nfunc TestAgent_All_WithGlobalCallback_V8(t *testing.T) {\n\tif testing.Short() {\n\t\tt.Skip(\"Skipping integration test in short mode\")\n\t}\n\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\tauthorized := &types.AuthorizedInfo{\n\t\tSubject: \"test-user\",\n\t\tUserID: \"test-123\",\n\t\tTenantID: \"test-tenant\",\n\t}\n\n\tctx := context.New(stdContext.Background(), authorized, \"test-chat-v8-all-callback\")\n\tctx.AssistantID = \"tests.agent-caller\"\n\tdefer ctx.Release()\n\n\tres, err := v8.Call(v8.CallOptions{}, `\n\t\tfunction test(ctx) {\n\t\t\ttry {\n\t\t\t\tconst messagesByAgent = {};\n\t\t\t\t\n\t\t\t\tconst results = ctx.agent.All(\n\t\t\t\t\t[\n\t\t\t\t\t\t{\n\t\t\t\t\t\t\tagent: \"tests.simple-greeting\",\n\t\t\t\t\t\t\tmessages: [{ role: \"user\", content: \"Hello from 1\" }]\n\t\t\t\t\t\t},\n\t\t\t\t\t\t{\n\t\t\t\t\t\t\tagent: \"tests.simple-greeting\",\n\t\t\t\t\t\t\tmessages: [{ role: \"user\", content: \"Hello from 2\" }]\n\t\t\t\t\t\t}\n\t\t\t\t\t],\n\t\t\t\t\t{\n\t\t\t\t\t\t// Global callback receives agentID, index, and message\n\t\t\t\t\t\tonChunk: (agentID, index, msg) => {\n\t\t\t\t\t\t\tconst key = agentID + \"_\" + index;\n\t\t\t\t\t\t\tif (!messagesByAgent[key]) {\n\t\t\t\t\t\t\t\tmessagesByAgent[key] = 0;\n\t\t\t\t\t\t\t}\n\t\t\t\t\t\t\tmessagesByAgent[key]++;\n\t\t\t\t\t\t\treturn 0;\n\t\t\t\t\t\t}\n\t\t\t\t\t}\n\t\t\t\t);\n\t\t\t\t\n\t\t\t\treturn { \n\t\t\t\t\tsuccess: true,\n\t\t\t\t\tresult_count: results.length,\n\t\t\t\t\tmessages_by_agent: messagesByAgent\n\t\t\t\t};\n\t\t\t} catch (error) {\n\t\t\t\treturn { success: false, error: error.message };\n\t\t\t}\n\t\t}`, ctx)\n\n\trequire.NoError(t, err)\n\tresult, ok := res.(map[string]interface{})\n\trequire.True(t, ok, \"Result should be a map\")\n\n\tsuccess, _ := result[\"success\"].(bool)\n\tif !success {\n\t\tt.Fatalf(\"Test failed: %v\", result[\"error\"])\n\t}\n\n\tassert.Equal(t, float64(2), result[\"result_count\"])\n\n\t// Should have received messages from both agents\n\tmessagesByAgent, _ := result[\"messages_by_agent\"].(map[string]interface{})\n\tt.Logf(\"Messages by agent: %v\", messagesByAgent)\n\n\t// At least one agent should have sent messages\n\tassert.Greater(t, len(messagesByAgent), 0, \"Should have received messages from agents\")\n}\n" }, { "path": "agent/context/jsapi_computer.go", "content": "package context\n\nimport (\n\t\"context\"\n\t\"fmt\"\n\t\"strings\"\n\n\t\"github.com/yaoapp/gou/runtime/v8/bridge\"\n\tinfraV2 \"github.com/yaoapp/yao/sandbox/v2\"\n\t\"github.com/yaoapp/yao/tai/workspace\"\n\t\"rogchap.com/v8go\"\n)\n\n// SetComputer sets the V2 computer and its workspace for this context.\n// Should be called after Runner.Prepare succeeds in initSandboxV2.\nfunc (ctx *Context) SetComputer(computer infraV2.Computer) {\n\tctx.computer = computer\n\tif computer != nil {\n\t\tctx.workspace = computer.Workplace()\n\t}\n}\n\n// SetWorkspace sets the workspace FS directly without requiring a Computer.\n// Use this when the user selected a workspace but no sandbox is configured.\nfunc (ctx *Context) SetWorkspace(ws workspace.FS) {\n\tctx.workspace = ws\n}\n\n// GetComputer returns the V2 computer if available.\nfunc (ctx *Context) GetComputer() infraV2.Computer {\n\treturn ctx.computer\n}\n\n// GetWorkspace returns the V2 workspace FS if available.\nfunc (ctx *Context) GetWorkspace() workspace.FS {\n\treturn ctx.workspace\n}\n\n// HasComputer returns true if V2 computer is available.\nfunc (ctx *Context) HasComputer() bool {\n\treturn ctx.computer != nil\n}\n\n// HasWorkspace returns true if workspace FS is available.\nfunc (ctx *Context) HasWorkspace() bool {\n\treturn ctx.workspace != nil\n}\n\n// createComputerInstance creates the ctx.computer JavaScript object.\nfunc (ctx *Context) createComputerInstance(v8ctx *v8go.Context) *v8go.Value {\n\tif ctx.computer == nil {\n\t\treturn nil\n\t}\n\n\tiso := v8ctx.Isolate()\n\tobjTpl := v8go.NewObjectTemplate(iso)\n\n\tinfo := ctx.computer.ComputerInfo()\n\tid := info.BoxID\n\tif id == \"\" {\n\t\tid = info.NodeID\n\t}\n\tobjTpl.Set(\"id\", id)\n\n\tobjTpl.Set(\"Exec\", ctx.computerExecMethod(iso))\n\tobjTpl.Set(\"VNC\", ctx.computerVNCMethod(iso))\n\tobjTpl.Set(\"Proxy\", ctx.computerProxyMethod(iso))\n\tobjTpl.Set(\"Info\", ctx.computerInfoMethod(iso))\n\n\tinstance, err := objTpl.NewInstance(v8ctx)\n\tif err != nil {\n\t\treturn nil\n\t}\n\treturn instance.Value\n}\n\n// computerExecMethod implements ctx.computer.Exec(cmd)\n// cmd can be a string or an array of strings.\n// Returns: { stdout, stderr, exit_code }\nfunc (ctx *Context) computerExecMethod(iso *v8go.Isolate) *v8go.FunctionTemplate {\n\treturn v8go.NewFunctionTemplate(iso, func(info *v8go.FunctionCallbackInfo) *v8go.Value {\n\t\tv8ctx := info.Context()\n\t\targs := info.Args()\n\n\t\tif ctx.computer == nil {\n\t\t\treturn bridge.JsException(v8ctx, \"computer not available\")\n\t\t}\n\t\tif len(args) < 1 {\n\t\t\treturn bridge.JsException(v8ctx, \"Exec requires a command argument\")\n\t\t}\n\n\t\tcmd, err := parseCommandArg(v8ctx, args[0])\n\t\tif err != nil {\n\t\t\treturn bridge.JsException(v8ctx, err.Error())\n\t\t}\n\n\t\tresult, err := ctx.computer.Exec(context.Background(), cmd)\n\t\tif err != nil {\n\t\t\treturn bridge.JsException(v8ctx, \"Exec failed: \"+err.Error())\n\t\t}\n\n\t\tres := map[string]interface{}{\n\t\t\t\"stdout\": result.Stdout,\n\t\t\t\"stderr\": result.Stderr,\n\t\t\t\"exit_code\": int32(result.ExitCode),\n\t\t}\n\t\tjsVal, err := bridge.JsValue(v8ctx, res)\n\t\tif err != nil {\n\t\t\treturn bridge.JsException(v8ctx, err.Error())\n\t\t}\n\t\treturn jsVal\n\t})\n}\n\n// computerVNCMethod implements ctx.computer.VNC()\n// Returns the VNC URL string.\nfunc (ctx *Context) computerVNCMethod(iso *v8go.Isolate) *v8go.FunctionTemplate {\n\treturn v8go.NewFunctionTemplate(iso, func(info *v8go.FunctionCallbackInfo) *v8go.Value {\n\t\tv8ctx := info.Context()\n\n\t\tif ctx.computer == nil {\n\t\t\treturn bridge.JsException(v8ctx, \"computer not available\")\n\t\t}\n\n\t\turl, err := ctx.computer.VNC(context.Background())\n\t\tif err != nil {\n\t\t\treturn bridge.JsException(v8ctx, \"VNC failed: \"+err.Error())\n\t\t}\n\n\t\tjsVal, err := v8go.NewValue(iso, url)\n\t\tif err != nil {\n\t\t\treturn bridge.JsException(v8ctx, err.Error())\n\t\t}\n\t\treturn jsVal\n\t})\n}\n\n// computerProxyMethod implements ctx.computer.Proxy(port, path?)\n// Returns the proxy URL string.\nfunc (ctx *Context) computerProxyMethod(iso *v8go.Isolate) *v8go.FunctionTemplate {\n\treturn v8go.NewFunctionTemplate(iso, func(info *v8go.FunctionCallbackInfo) *v8go.Value {\n\t\tv8ctx := info.Context()\n\t\targs := info.Args()\n\n\t\tif ctx.computer == nil {\n\t\t\treturn bridge.JsException(v8ctx, \"computer not available\")\n\t\t}\n\t\tif len(args) < 1 || !args[0].IsNumber() {\n\t\t\treturn bridge.JsException(v8ctx, \"Proxy requires a port number\")\n\t\t}\n\n\t\tport := int(args[0].Integer())\n\t\tpath := \"\"\n\t\tif len(args) >= 2 && args[1].IsString() {\n\t\t\tpath = args[1].String()\n\t\t}\n\n\t\turl, err := ctx.computer.Proxy(context.Background(), port, path)\n\t\tif err != nil {\n\t\t\treturn bridge.JsException(v8ctx, \"Proxy failed: \"+err.Error())\n\t\t}\n\n\t\tjsVal, err := v8go.NewValue(iso, url)\n\t\tif err != nil {\n\t\t\treturn bridge.JsException(v8ctx, err.Error())\n\t\t}\n\t\treturn jsVal\n\t})\n}\n\n// computerInfoMethod implements ctx.computer.Info()\n// Returns a JS object with computer identity and system information.\nfunc (ctx *Context) computerInfoMethod(iso *v8go.Isolate) *v8go.FunctionTemplate {\n\treturn v8go.NewFunctionTemplate(iso, func(info *v8go.FunctionCallbackInfo) *v8go.Value {\n\t\tv8ctx := info.Context()\n\n\t\tif ctx.computer == nil {\n\t\t\treturn bridge.JsException(v8ctx, \"computer not available\")\n\t\t}\n\n\t\tci := ctx.computer.ComputerInfo()\n\t\tresult := map[string]interface{}{\n\t\t\t\"kind\": ci.Kind,\n\t\t\t\"node_id\": ci.NodeID,\n\t\t\t\"tai_id\": ci.TaiID,\n\t\t\t\"status\": ci.Status,\n\t\t\t\"system\": map[string]interface{}{\n\t\t\t\t\"os\": ci.System.OS,\n\t\t\t\t\"arch\": ci.System.Arch,\n\t\t\t\t\"hostname\": ci.System.Hostname,\n\t\t\t\t\"num_cpu\": int32(ci.System.NumCPU),\n\t\t\t\t\"shell\": ci.System.Shell,\n\t\t\t},\n\t\t}\n\t\tif ci.BoxID != \"\" {\n\t\t\tresult[\"box_id\"] = ci.BoxID\n\t\t\tresult[\"container_id\"] = ci.ContainerID\n\t\t\tresult[\"image\"] = ci.Image\n\t\t\tresult[\"policy\"] = string(ci.Policy)\n\t\t}\n\n\t\tjsVal, err := bridge.JsValue(v8ctx, result)\n\t\tif err != nil {\n\t\t\treturn bridge.JsException(v8ctx, err.Error())\n\t\t}\n\t\treturn jsVal\n\t})\n}\n\n// parseCommandArg converts a JS value (string or string array) to []string.\nfunc parseCommandArg(v8ctx *v8go.Context, val *v8go.Value) ([]string, error) {\n\tif val.IsString() {\n\t\traw := val.String()\n\t\treturn strings.Fields(raw), nil\n\t}\n\n\tif val.IsArray() {\n\t\tobj, err := val.AsObject()\n\t\tif err != nil {\n\t\t\treturn nil, err\n\t\t}\n\t\tlengthVal, err := obj.Get(\"length\")\n\t\tif err != nil {\n\t\t\treturn nil, err\n\t\t}\n\t\tlength := int(lengthVal.Integer())\n\t\tcmd := make([]string, length)\n\t\tfor i := 0; i < length; i++ {\n\t\t\titem, err := obj.GetIdx(uint32(i))\n\t\t\tif err != nil {\n\t\t\t\treturn nil, err\n\t\t\t}\n\t\t\tcmd[i] = item.String()\n\t\t}\n\t\treturn cmd, nil\n\t}\n\n\treturn nil, fmt.Errorf(\"command must be a string or array of strings\")\n}\n" }, { "path": "agent/context/jsapi_helpers.go", "content": "package context\n\nimport (\n\t\"fmt\"\n\n\t\"github.com/yaoapp/gou/runtime/v8/bridge\"\n\t\"github.com/yaoapp/yao/agent/output/message\"\n\t\"rogchap.com/v8go\"\n)\n\n// parseMessage parses a JavaScript value into a message.Message\nfunc parseMessage(v8ctx *v8go.Context, jsValue *v8go.Value) (*message.Message, error) {\n\t// Handle string shorthand: convert to text message\n\tif jsValue.IsString() {\n\t\treturn &message.Message{\n\t\t\tType: message.TypeText,\n\t\t\tProps: map[string]interface{}{\n\t\t\t\t\"content\": jsValue.String(),\n\t\t\t},\n\t\t}, nil\n\t}\n\n\t// Handle object\n\tif !jsValue.IsObject() {\n\t\treturn nil, fmt.Errorf(\"message must be a string or object\")\n\t}\n\n\t// Convert to Go map\n\tgoValue, err := bridge.GoValue(jsValue, v8ctx)\n\tif err != nil {\n\t\treturn nil, fmt.Errorf(\"failed to convert message: %w\", err)\n\t}\n\n\tmsgMap, ok := goValue.(map[string]interface{})\n\tif !ok {\n\t\treturn nil, fmt.Errorf(\"message must be an object\")\n\t}\n\n\t// Build message\n\tmsg := &message.Message{}\n\n\t// Type field (required)\n\tif msgType, ok := msgMap[\"type\"].(string); ok {\n\t\tmsg.Type = msgType\n\t} else {\n\t\treturn nil, fmt.Errorf(\"message.type is required and must be a string\")\n\t}\n\n\t// Props field (optional)\n\tif props, ok := msgMap[\"props\"].(map[string]interface{}); ok {\n\t\tmsg.Props = props\n\t}\n\n\t// Optional fields - Streaming control\n\tif chunkID, ok := msgMap[\"chunk_id\"].(string); ok {\n\t\tmsg.ChunkID = chunkID\n\t}\n\tif messageID, ok := msgMap[\"message_id\"].(string); ok {\n\t\tmsg.MessageID = messageID\n\t}\n\tif blockID, ok := msgMap[\"block_id\"].(string); ok {\n\t\tmsg.BlockID = blockID\n\t}\n\tif threadID, ok := msgMap[\"thread_id\"].(string); ok {\n\t\tmsg.ThreadID = threadID\n\t}\n\n\t// Delta control\n\tif delta, ok := msgMap[\"delta\"].(bool); ok {\n\t\tmsg.Delta = delta\n\t}\n\tif deltaPath, ok := msgMap[\"delta_path\"].(string); ok {\n\t\tmsg.DeltaPath = deltaPath\n\t}\n\tif deltaAction, ok := msgMap[\"delta_action\"].(string); ok {\n\t\tmsg.DeltaAction = deltaAction\n\t}\n\tif typeChange, ok := msgMap[\"type_change\"].(bool); ok {\n\t\tmsg.TypeChange = typeChange\n\t}\n\n\t// Metadata (optional)\n\tif metadataMap, ok := msgMap[\"metadata\"].(map[string]interface{}); ok {\n\t\tmetadata := &message.Metadata{}\n\t\tif timestamp, ok := metadataMap[\"timestamp\"].(float64); ok {\n\t\t\tmetadata.Timestamp = int64(timestamp)\n\t\t}\n\t\tif sequence, ok := metadataMap[\"sequence\"].(float64); ok {\n\t\t\tmetadata.Sequence = int(sequence)\n\t\t}\n\t\tif traceID, ok := metadataMap[\"trace_id\"].(string); ok {\n\t\t\tmetadata.TraceID = traceID\n\t\t}\n\t\tmsg.Metadata = metadata\n\t}\n\n\treturn msg, nil\n}\n\n// parseGroup parses a JavaScript value into a message.Group\nfunc parseGroup(v8ctx *v8go.Context, jsValue *v8go.Value) (*message.Group, error) {\n\t// Must be an object\n\tif !jsValue.IsObject() {\n\t\treturn nil, fmt.Errorf(\"group must be an object\")\n\t}\n\n\t// Convert to Go map\n\tgoValue, err := bridge.GoValue(jsValue, v8ctx)\n\tif err != nil {\n\t\treturn nil, fmt.Errorf(\"failed to convert group: %w\", err)\n\t}\n\n\tgroupMap, ok := goValue.(map[string]interface{})\n\tif !ok {\n\t\treturn nil, fmt.Errorf(\"group must be an object\")\n\t}\n\n\t// Build group\n\tgroup := &message.Group{}\n\n\t// ID field (required)\n\tif id, ok := groupMap[\"id\"].(string); ok {\n\t\tgroup.ID = id\n\t} else {\n\t\treturn nil, fmt.Errorf(\"group.id is required and must be a string\")\n\t}\n\n\t// Messages field (required)\n\tif messagesArray, ok := groupMap[\"messages\"].([]interface{}); ok {\n\t\tgroup.Messages = make([]*message.Message, 0, len(messagesArray))\n\t\tfor i, msgInterface := range messagesArray {\n\t\t\t// Convert to map\n\t\t\tmsgMap, ok := msgInterface.(map[string]interface{})\n\t\t\tif !ok {\n\t\t\t\treturn nil, fmt.Errorf(\"group.messages[%d] must be an object\", i)\n\t\t\t}\n\n\t\t\t// Convert map to Message\n\t\t\tmsg := &message.Message{}\n\n\t\t\t// Type field (required)\n\t\t\tif msgType, ok := msgMap[\"type\"].(string); ok {\n\t\t\t\tmsg.Type = msgType\n\t\t\t} else {\n\t\t\t\treturn nil, fmt.Errorf(\"group.messages[%d].type is required\", i)\n\t\t\t}\n\n\t\t\t// Props field (optional)\n\t\t\tif props, ok := msgMap[\"props\"].(map[string]interface{}); ok {\n\t\t\t\tmsg.Props = props\n\t\t\t}\n\n\t\t\t// Optional fields - Streaming control\n\t\t\tif chunkID, ok := msgMap[\"chunk_id\"].(string); ok {\n\t\t\t\tmsg.ChunkID = chunkID\n\t\t\t}\n\t\t\tif messageID, ok := msgMap[\"message_id\"].(string); ok {\n\t\t\t\tmsg.MessageID = messageID\n\t\t\t}\n\t\t\tif blockID, ok := msgMap[\"block_id\"].(string); ok {\n\t\t\t\tmsg.BlockID = blockID\n\t\t\t}\n\t\t\tif threadID, ok := msgMap[\"thread_id\"].(string); ok {\n\t\t\t\tmsg.ThreadID = threadID\n\t\t\t}\n\n\t\t\t// Delta control\n\t\t\tif delta, ok := msgMap[\"delta\"].(bool); ok {\n\t\t\t\tmsg.Delta = delta\n\t\t\t}\n\t\t\tif deltaPath, ok := msgMap[\"delta_path\"].(string); ok {\n\t\t\t\tmsg.DeltaPath = deltaPath\n\t\t\t}\n\t\t\tif deltaAction, ok := msgMap[\"delta_action\"].(string); ok {\n\t\t\t\tmsg.DeltaAction = deltaAction\n\t\t\t}\n\t\t\tif typeChange, ok := msgMap[\"type_change\"].(bool); ok {\n\t\t\t\tmsg.TypeChange = typeChange\n\t\t\t}\n\n\t\t\t// Metadata (optional)\n\t\t\tif metadataMap, ok := msgMap[\"metadata\"].(map[string]interface{}); ok {\n\t\t\t\tmetadata := &message.Metadata{}\n\t\t\t\tif timestamp, ok := metadataMap[\"timestamp\"].(float64); ok {\n\t\t\t\t\tmetadata.Timestamp = int64(timestamp)\n\t\t\t\t}\n\t\t\t\tif sequence, ok := metadataMap[\"sequence\"].(float64); ok {\n\t\t\t\t\tmetadata.Sequence = int(sequence)\n\t\t\t\t}\n\t\t\t\tif traceID, ok := metadataMap[\"trace_id\"].(string); ok {\n\t\t\t\t\tmetadata.TraceID = traceID\n\t\t\t\t}\n\t\t\t\tmsg.Metadata = metadata\n\t\t\t}\n\n\t\t\tgroup.Messages = append(group.Messages, msg)\n\t\t}\n\t} else {\n\t\treturn nil, fmt.Errorf(\"group.messages is required and must be an array\")\n\t}\n\n\t// Metadata (optional)\n\tif metadataMap, ok := groupMap[\"metadata\"].(map[string]interface{}); ok {\n\t\tmetadata := &message.Metadata{}\n\t\tif timestamp, ok := metadataMap[\"timestamp\"].(float64); ok {\n\t\t\tmetadata.Timestamp = int64(timestamp)\n\t\t}\n\t\tif sequence, ok := metadataMap[\"sequence\"].(float64); ok {\n\t\t\tmetadata.Sequence = int(sequence)\n\t\t}\n\t\tif traceID, ok := metadataMap[\"trace_id\"].(string); ok {\n\t\t\tmetadata.TraceID = traceID\n\t\t}\n\t\tgroup.Metadata = metadata\n\t}\n\n\treturn group, nil\n}\n" }, { "path": "agent/context/jsapi_llm.go", "content": "package context\n\nimport (\n\t\"github.com/yaoapp/gou/runtime/v8/bridge\"\n\t\"github.com/yaoapp/yao/agent/output/message\"\n\t\"rogchap.com/v8go\"\n)\n\n// LlmAPI defines the LLM JSAPI interface for ctx.llm.*\n// This interface is defined here to avoid circular dependency between context and llm packages.\n// The actual implementation is in agent/llm/jsapi.go\ntype LlmAPI interface {\n\t// Stream calls LLM with streaming output to ctx.Writer\n\t// Returns *llm.Result or error information\n\tStream(connector string, messages []interface{}, opts map[string]interface{}) interface{}\n\n\t// Parallel LLM call methods - inspired by JavaScript Promise\n\t// All waits for all LLM calls to complete (like Promise.all)\n\tAll(requests []interface{}) []interface{}\n\t// Any returns when any LLM call succeeds (like Promise.any)\n\tAny(requests []interface{}) []interface{}\n\t// Race returns when any LLM call completes (like Promise.race)\n\tRace(requests []interface{}) []interface{}\n}\n\n// LlmAPIWithCallback extends LlmAPI with callback support\n// This interface provides methods that accept OnMessage handlers for real-time message processing\ntype LlmAPIWithCallback interface {\n\tLlmAPI\n\n\t// StreamWithHandler calls LLM with an OnMessage handler\n\t// handler receives SSE messages: func(msg *message.Message) int\n\tStreamWithHandler(connector string, messages []interface{}, opts map[string]interface{}, handler OnMessageFunc) interface{}\n\n\t// AllWithHandler executes all LLM calls with handlers\n\t// globalHandler receives messages with connectorID and index: func(connectorID, index, msg) int\n\tAllWithHandler(requests []interface{}, globalHandler LlmBatchOnMessageFunc) []interface{}\n\n\t// AnyWithHandler executes LLM calls and returns on first success, with handlers\n\tAnyWithHandler(requests []interface{}, globalHandler LlmBatchOnMessageFunc) []interface{}\n\n\t// RaceWithHandler executes LLM calls and returns on first completion, with handlers\n\tRaceWithHandler(requests []interface{}, globalHandler LlmBatchOnMessageFunc) []interface{}\n}\n\n// LlmBatchOnMessageFunc is the OnMessage function for batch LLM calls\n// It includes connectorID and index to identify the source of each message\ntype LlmBatchOnMessageFunc func(connectorID string, index int, msg *message.Message) int\n\n// LlmAPIFactory is a function type that creates a LlmAPI for a context\n// This is set by the llm package during initialization\nvar LlmAPIFactory func(ctx *Context) LlmAPI\n\n// Llm returns the LLM API for this context\n// Returns nil if LlmAPIFactory is not set\nfunc (ctx *Context) Llm() LlmAPI {\n\tif LlmAPIFactory == nil {\n\t\treturn nil\n\t}\n\treturn LlmAPIFactory(ctx)\n}\n\n// newLlmObject creates a new llm object with all llm methods\n// This is called from jsapi.go NewObject() to mount ctx.llm\nfunc (ctx *Context) newLlmObject(iso *v8go.Isolate) *v8go.ObjectTemplate {\n\tllmObj := v8go.NewObjectTemplate(iso)\n\n\t// Single LLM call method\n\tllmObj.Set(\"Stream\", ctx.llmStreamMethod(iso))\n\n\t// Parallel LLM call methods - inspired by JavaScript Promise\n\tllmObj.Set(\"All\", ctx.llmAllMethod(iso))\n\tllmObj.Set(\"Any\", ctx.llmAnyMethod(iso))\n\tllmObj.Set(\"Race\", ctx.llmRaceMethod(iso))\n\n\treturn llmObj\n}\n\n// llmStreamMethod implements ctx.llm.Stream(connector, messages, options?)\n// Usage: const result = ctx.llm.Stream(\"gpt-4o\", [{ role: \"user\", content: \"Hello\" }], { temperature: 0.7, onChunk: (msg) => 0 })\n// Returns: { connector, response, content, error }\nfunc (ctx *Context) llmStreamMethod(iso *v8go.Isolate) *v8go.FunctionTemplate {\n\treturn v8go.NewFunctionTemplate(iso, func(info *v8go.FunctionCallbackInfo) *v8go.Value {\n\t\tv8ctx := info.Context()\n\t\targs := info.Args()\n\n\t\t// Validate arguments\n\t\tif len(args) < 2 {\n\t\t\treturn bridge.JsException(v8ctx, \"Stream requires connector and messages parameters\")\n\t\t}\n\n\t\t// Get connector ID (first argument)\n\t\tif !args[0].IsString() {\n\t\t\treturn bridge.JsException(v8ctx, \"connector must be a string\")\n\t\t}\n\t\tconnector := args[0].String()\n\n\t\t// Parse messages (second argument)\n\t\tmessagesVal, err := bridge.GoValue(args[1], v8ctx)\n\t\tif err != nil {\n\t\t\treturn bridge.JsException(v8ctx, \"invalid messages: \"+err.Error())\n\t\t}\n\t\tmessages, ok := messagesVal.([]interface{})\n\t\tif !ok {\n\t\t\treturn bridge.JsException(v8ctx, \"messages must be an array\")\n\t\t}\n\n\t\t// Parse options (optional third argument) - extract onChunk separately\n\t\tvar opts map[string]interface{}\n\t\tvar onChunkFn *v8go.Function\n\n\t\tif len(args) >= 3 && !args[2].IsUndefined() && !args[2].IsNull() {\n\t\t\toptsObj, err := args[2].AsObject()\n\t\t\tif err == nil && optsObj != nil {\n\t\t\t\t// Extract onChunk callback before converting to Go value\n\t\t\t\tonChunkVal, _ := optsObj.Get(\"onChunk\")\n\t\t\t\tif onChunkVal != nil && onChunkVal.IsFunction() {\n\t\t\t\t\tonChunkFn, _ = onChunkVal.AsFunction()\n\t\t\t\t}\n\n\t\t\t\t// Convert the rest of options to Go map\n\t\t\t\tgoVal, err := bridge.GoValue(args[2], v8ctx)\n\t\t\t\tif err == nil {\n\t\t\t\t\tif optsMap, ok := goVal.(map[string]interface{}); ok {\n\t\t\t\t\t\t// Remove onChunk from the map (it's handled separately)\n\t\t\t\t\t\tdelete(optsMap, \"onChunk\")\n\t\t\t\t\t\topts = optsMap\n\t\t\t\t\t}\n\t\t\t\t}\n\t\t\t}\n\t\t}\n\n\t\t// Get LLM API\n\t\tllmAPI := ctx.Llm()\n\t\tif llmAPI == nil {\n\t\t\treturn bridge.JsException(v8ctx, \"LLM API not available\")\n\t\t}\n\n\t\tvar result interface{}\n\n\t\t// If onChunk callback is provided and API supports it, use StreamWithHandler\n\t\tif onChunkFn != nil {\n\t\t\tif apiWithCb, ok := llmAPI.(LlmAPIWithCallback); ok {\n\t\t\t\t// Create Go OnMessageFunc that calls JS callback\n\t\t\t\thandler := createJSStreamHandler(v8ctx, onChunkFn)\n\t\t\t\tresult = apiWithCb.StreamWithHandler(connector, messages, opts, handler)\n\t\t\t} else {\n\t\t\t\t// Fallback: ignore callback if API doesn't support it\n\t\t\t\tresult = llmAPI.Stream(connector, messages, opts)\n\t\t\t}\n\t\t} else {\n\t\t\t// No callback, use regular Stream\n\t\t\tresult = llmAPI.Stream(connector, messages, opts)\n\t\t}\n\n\t\t// Convert result to JS value\n\t\tjsVal, err := bridge.JsValue(v8ctx, result)\n\t\tif err != nil {\n\t\t\treturn bridge.JsException(v8ctx, \"failed to convert result: \"+err.Error())\n\t\t}\n\n\t\treturn jsVal\n\t})\n}\n\n// llmAllMethod implements ctx.llm.All(requests, options?)\n// Usage: const results = ctx.llm.All([\n//\n//\t{ connector: \"gpt-4o\", messages: [...], options: {...} },\n//\t{ connector: \"claude-3\", messages: [...] }\n//\n// ], { onChunk: (connectorID, index, msg) => 0 })\n// Returns: [{ connector, response, content, error }, ...]\nfunc (ctx *Context) llmAllMethod(iso *v8go.Isolate) *v8go.FunctionTemplate {\n\treturn v8go.NewFunctionTemplate(iso, func(info *v8go.FunctionCallbackInfo) *v8go.Value {\n\t\treturn ctx.executeLlmBatchMethod(info, LlmBatchMethodAll)\n\t})\n}\n\n// llmAnyMethod implements ctx.llm.Any(requests, options?)\n// Returns first successful result\nfunc (ctx *Context) llmAnyMethod(iso *v8go.Isolate) *v8go.FunctionTemplate {\n\treturn v8go.NewFunctionTemplate(iso, func(info *v8go.FunctionCallbackInfo) *v8go.Value {\n\t\treturn ctx.executeLlmBatchMethod(info, LlmBatchMethodAny)\n\t})\n}\n\n// llmRaceMethod implements ctx.llm.Race(requests, options?)\n// Returns first completed result (success or failure)\nfunc (ctx *Context) llmRaceMethod(iso *v8go.Isolate) *v8go.FunctionTemplate {\n\treturn v8go.NewFunctionTemplate(iso, func(info *v8go.FunctionCallbackInfo) *v8go.Value {\n\t\treturn ctx.executeLlmBatchMethod(info, LlmBatchMethodRace)\n\t})\n}\n\n// LlmBatchMethod represents the type of batch LLM operation\ntype LlmBatchMethod int\n\nconst (\n\tLlmBatchMethodAll LlmBatchMethod = iota\n\tLlmBatchMethodAny\n\tLlmBatchMethodRace\n)\n\n// executeLlmBatchMethod handles All/Any/Race batch LLM calls\nfunc (ctx *Context) executeLlmBatchMethod(info *v8go.FunctionCallbackInfo, method LlmBatchMethod) *v8go.Value {\n\tv8ctx := info.Context()\n\targs := info.Args()\n\n\t// Validate arguments\n\tif len(args) < 1 {\n\t\treturn bridge.JsException(v8ctx, \"requires requests array parameter\")\n\t}\n\n\t// Parse requests array (first argument)\n\trequestsVal, err := bridge.GoValue(args[0], v8ctx)\n\tif err != nil {\n\t\treturn bridge.JsException(v8ctx, \"invalid requests: \"+err.Error())\n\t}\n\trequests, ok := requestsVal.([]interface{})\n\tif !ok {\n\t\treturn bridge.JsException(v8ctx, \"requests must be an array\")\n\t}\n\n\t// Get LLM API\n\tllmAPI := ctx.Llm()\n\tif llmAPI == nil {\n\t\treturn bridge.JsException(v8ctx, \"LLM API not available\")\n\t}\n\n\t// Parse optional global callback from second argument (options object)\n\tvar globalCallback *v8go.Function\n\tif len(args) >= 2 && !args[1].IsUndefined() && !args[1].IsNull() {\n\t\toptsObj, err := args[1].AsObject()\n\t\tif err == nil && optsObj != nil {\n\t\t\tonChunkVal, _ := optsObj.Get(\"onChunk\")\n\t\t\tif onChunkVal != nil && onChunkVal.IsFunction() {\n\t\t\t\tglobalCallback, _ = onChunkVal.AsFunction()\n\t\t\t}\n\t\t}\n\t}\n\n\tvar results []interface{}\n\n\t// If callback is provided and API supports it, use channel-based execution\n\tif globalCallback != nil {\n\t\tif apiWithCb, ok := llmAPI.(LlmAPIWithCallback); ok {\n\t\t\tresults = ctx.executeLlmBatchWithCallback(method, requests, globalCallback, v8ctx, apiWithCb)\n\t\t} else {\n\t\t\t// Fallback: execute without callback\n\t\t\tresults = ctx.executeLlmBatchWithoutCallback(method, requests, llmAPI)\n\t\t}\n\t} else {\n\t\t// No callback, use regular batch methods\n\t\tresults = ctx.executeLlmBatchWithoutCallback(method, requests, llmAPI)\n\t}\n\n\t// Convert results to JS value\n\tjsVal, err := bridge.JsValue(v8ctx, results)\n\tif err != nil {\n\t\treturn bridge.JsException(v8ctx, \"failed to convert results: \"+err.Error())\n\t}\n\n\treturn jsVal\n}\n\n// executeLlmBatchWithoutCallback executes batch LLM calls without callback\nfunc (ctx *Context) executeLlmBatchWithoutCallback(method LlmBatchMethod, requests []interface{}, llmAPI LlmAPI) []interface{} {\n\tswitch method {\n\tcase LlmBatchMethodAll:\n\t\treturn llmAPI.All(requests)\n\tcase LlmBatchMethodAny:\n\t\treturn llmAPI.Any(requests)\n\tcase LlmBatchMethodRace:\n\t\treturn llmAPI.Race(requests)\n\tdefault:\n\t\treturn llmAPI.All(requests)\n\t}\n}\n\n// llmBatchMessage is used for channel communication in batch LLM calls\ntype llmBatchMessage struct {\n\tConnectorID string\n\tIndex int\n\tMessage *message.Message\n}\n\n// executeLlmBatchWithCallback executes batch LLM calls with callback using channel\n// This ensures V8 thread safety by serializing callback invocations\nfunc (ctx *Context) executeLlmBatchWithCallback(method LlmBatchMethod, requests []interface{}, callback *v8go.Function, v8ctx *v8go.Context, apiWithCb LlmAPIWithCallback) []interface{} {\n\t// Create a buffered channel for messages\n\t// Using blocking send to ensure all messages are delivered\n\tmsgChan := make(chan llmBatchMessage, 1000)\n\tdoneChan := make(chan []interface{}, 1)\n\n\t// Create Go handler that sends to channel\n\tgoHandler := func(connectorID string, index int, msg *message.Message) int {\n\t\tmsgChan <- llmBatchMessage{\n\t\t\tConnectorID: connectorID,\n\t\t\tIndex: index,\n\t\t\tMessage: msg,\n\t\t}\n\t\treturn 0\n\t}\n\n\t// Execute batch calls in background goroutine\n\tgo func() {\n\t\tdefer close(msgChan)\n\n\t\tvar results []interface{}\n\t\tswitch method {\n\t\tcase LlmBatchMethodAll:\n\t\t\tresults = apiWithCb.AllWithHandler(requests, goHandler)\n\t\tcase LlmBatchMethodAny:\n\t\t\tresults = apiWithCb.AnyWithHandler(requests, goHandler)\n\t\tcase LlmBatchMethodRace:\n\t\t\tresults = apiWithCb.RaceWithHandler(requests, goHandler)\n\t\tdefault:\n\t\t\tresults = apiWithCb.AllWithHandler(requests, goHandler)\n\t\t}\n\t\tdoneChan <- results\n\t}()\n\n\t// Process messages in main goroutine (V8 thread)\n\tfor msg := range msgChan {\n\t\tcallJSLlmBatchCallback(v8ctx, callback, msg.ConnectorID, msg.Index, msg.Message)\n\t}\n\n\t// Wait for results\n\treturn <-doneChan\n}\n\n// callJSLlmBatchCallback calls the JS callback function for batch LLM calls\nfunc callJSLlmBatchCallback(v8ctx *v8go.Context, callback *v8go.Function, connectorID string, index int, msg *message.Message) {\n\tif callback == nil || v8ctx == nil {\n\t\treturn\n\t}\n\n\tiso := v8ctx.Isolate()\n\n\t// Create arguments: connectorID, index, message\n\tconnectorVal, err := v8go.NewValue(iso, connectorID)\n\tif err != nil {\n\t\treturn\n\t}\n\n\tindexVal, err := v8go.NewValue(iso, int32(index))\n\tif err != nil {\n\t\treturn\n\t}\n\n\t// Convert message to JS object\n\tmsgVal, err := bridge.JsValue(v8ctx, msg)\n\tif err != nil {\n\t\treturn\n\t}\n\n\t// Call the callback\n\t_, _ = callback.Call(v8go.Undefined(iso), connectorVal, indexVal, msgVal)\n}\n" }, { "path": "agent/context/jsapi_llm_v8_test.go", "content": "package context_test\n\nimport (\n\tstdContext \"context\"\n\t\"testing\"\n\n\t\"github.com/stretchr/testify/assert\"\n\t\"github.com/stretchr/testify/require\"\n\tv8 \"github.com/yaoapp/gou/runtime/v8\"\n\t\"github.com/yaoapp/yao/agent/context\"\n\t\"github.com/yaoapp/yao/agent/testutils\"\n\t\"github.com/yaoapp/yao/openapi/oauth/types\"\n\n\t// Import assistant package to register LlmAPIFactory\n\t_ \"github.com/yaoapp/yao/agent/assistant\"\n)\n\n// TestLlm_Stream_V8 tests basic ctx.llm.Stream functionality with real V8 execution\nfunc TestLlm_Stream_V8(t *testing.T) {\n\tif testing.Short() {\n\t\tt.Skip(\"Skipping integration test in short mode\")\n\t}\n\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\t// Create authorized info for the context\n\tauthorized := &types.AuthorizedInfo{\n\t\tSubject: \"test-user\",\n\t\tUserID: \"test-123\",\n\t\tTenantID: \"test-tenant\",\n\t}\n\n\t// Create a context\n\tctx := context.New(stdContext.Background(), authorized, \"test-chat-llm-stream\")\n\tctx.AssistantID = \"tests.simple-greeting\"\n\tdefer ctx.Release()\n\n\t// Test basic Stream call with real connector\n\tres, err := v8.Call(v8.CallOptions{}, `\n\t\tfunction test(ctx) {\n\t\t\ttry {\n\t\t\t\tconst result = ctx.llm.Stream(\"gpt-4o-mini\", [\n\t\t\t\t\t{ role: \"user\", content: \"Say hello in one word\" }\n\t\t\t\t], {\n\t\t\t\t\ttemperature: 0.1,\n\t\t\t\t\tmax_tokens: 10\n\t\t\t\t});\n\t\t\t\t\n\t\t\t\treturn {\n\t\t\t\t\tsuccess: true,\n\t\t\t\t\tconnector: result.connector,\n\t\t\t\t\thas_content: result.content && result.content.length > 0,\n\t\t\t\t\thas_response: result.response !== undefined,\n\t\t\t\t\terror: result.error || \"\"\n\t\t\t\t};\n\t\t\t} catch (error) {\n\t\t\t\treturn { success: false, error: error.message };\n\t\t\t}\n\t\t}`, ctx)\n\n\trequire.NoError(t, err)\n\trequire.NotNil(t, res)\n\n\tresult, ok := res.(map[string]interface{})\n\trequire.True(t, ok, \"result should be a map\")\n\n\tsuccess, _ := result[\"success\"].(bool)\n\tif !success {\n\t\tt.Logf(\"Test result: %v\", result)\n\t}\n\trequire.True(t, success, \"Test should succeed, error: %v\", result[\"error\"])\n\n\tassert.Equal(t, \"gpt-4o-mini\", result[\"connector\"])\n\n\thasContent, _ := result[\"has_content\"].(bool)\n\tassert.True(t, hasContent, \"Should have content in response\")\n\n\thasResponse, _ := result[\"has_response\"].(bool)\n\tassert.True(t, hasResponse, \"Should have response object\")\n}\n\n// TestLlm_Stream_WithCallback_V8 tests ctx.llm.Stream with onChunk callback\nfunc TestLlm_Stream_WithCallback_V8(t *testing.T) {\n\tif testing.Short() {\n\t\tt.Skip(\"Skipping integration test in short mode\")\n\t}\n\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\tauthorized := &types.AuthorizedInfo{\n\t\tSubject: \"test-user\",\n\t\tUserID: \"test-123\",\n\t\tTenantID: \"test-tenant\",\n\t}\n\n\tctx := context.New(stdContext.Background(), authorized, \"test-chat-llm-callback\")\n\tctx.AssistantID = \"tests.simple-greeting\"\n\tdefer ctx.Release()\n\n\t// Test Stream call with callback\n\tres, err := v8.Call(v8.CallOptions{}, `\n\t\tfunction test(ctx) {\n\t\t\ttry {\n\t\t\t\tlet callbackCount = 0;\n\t\t\t\tlet receivedTypes = [];\n\t\t\t\t\n\t\t\t\tconst result = ctx.llm.Stream(\"gpt-4o-mini\", [\n\t\t\t\t\t{ role: \"user\", content: \"Say hi\" }\n\t\t\t\t], {\n\t\t\t\t\ttemperature: 0.1,\n\t\t\t\t\tmax_tokens: 10,\n\t\t\t\t\tonChunk: function(msg) {\n\t\t\t\t\t\tcallbackCount++;\n\t\t\t\t\t\tif (msg && msg.type) {\n\t\t\t\t\t\t\treceivedTypes.push(msg.type);\n\t\t\t\t\t\t}\n\t\t\t\t\t\treturn 0; // Continue\n\t\t\t\t\t}\n\t\t\t\t});\n\t\t\t\t\n\t\t\t\treturn {\n\t\t\t\t\tsuccess: true,\n\t\t\t\t\tconnector: result.connector,\n\t\t\t\t\tcallbackCount: callbackCount,\n\t\t\t\t\treceivedTypes: receivedTypes,\n\t\t\t\t\thas_content: result.content && result.content.length > 0,\n\t\t\t\t\terror: result.error || \"\"\n\t\t\t\t};\n\t\t\t} catch (error) {\n\t\t\t\treturn { success: false, error: error.message };\n\t\t\t}\n\t\t}`, ctx)\n\n\trequire.NoError(t, err)\n\trequire.NotNil(t, res)\n\n\tresult, ok := res.(map[string]interface{})\n\trequire.True(t, ok, \"result should be a map\")\n\n\tsuccess, _ := result[\"success\"].(bool)\n\tif !success {\n\t\tt.Logf(\"Test result: %v\", result)\n\t}\n\trequire.True(t, success, \"Test should succeed, error: %v\", result[\"error\"])\n\n\t// Callback should have been called at least once\n\tcallbackCount, _ := result[\"callbackCount\"].(float64)\n\tassert.Greater(t, callbackCount, float64(0), \"Callback should be called at least once\")\n}\n\n// TestLlm_All_V8 tests ctx.llm.All with multiple connectors\nfunc TestLlm_All_V8(t *testing.T) {\n\tif testing.Short() {\n\t\tt.Skip(\"Skipping integration test in short mode\")\n\t}\n\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\tauthorized := &types.AuthorizedInfo{\n\t\tSubject: \"test-user\",\n\t\tUserID: \"test-123\",\n\t\tTenantID: \"test-tenant\",\n\t}\n\n\tctx := context.New(stdContext.Background(), authorized, \"test-chat-llm-all\")\n\tctx.AssistantID = \"tests.simple-greeting\"\n\tdefer ctx.Release()\n\n\t// Test All with multiple requests to same connector (different prompts)\n\tres, err := v8.Call(v8.CallOptions{}, `\n\t\tfunction test(ctx) {\n\t\t\ttry {\n\t\t\t\tconst results = ctx.llm.All([\n\t\t\t\t\t{\n\t\t\t\t\t\tconnector: \"gpt-4o-mini\",\n\t\t\t\t\t\tmessages: [{ role: \"user\", content: \"Say 'one'\" }],\n\t\t\t\t\t\toptions: { temperature: 0.1, max_tokens: 5 }\n\t\t\t\t\t},\n\t\t\t\t\t{\n\t\t\t\t\t\tconnector: \"gpt-4o-mini\",\n\t\t\t\t\t\tmessages: [{ role: \"user\", content: \"Say 'two'\" }],\n\t\t\t\t\t\toptions: { temperature: 0.1, max_tokens: 5 }\n\t\t\t\t\t}\n\t\t\t\t]);\n\t\t\t\t\n\t\t\t\treturn {\n\t\t\t\t\tsuccess: true,\n\t\t\t\t\tcount: results.length,\n\t\t\t\t\tresults: results.map(r => ({\n\t\t\t\t\t\tconnector: r.connector,\n\t\t\t\t\t\thas_content: r.content && r.content.length > 0,\n\t\t\t\t\t\terror: r.error || \"\"\n\t\t\t\t\t}))\n\t\t\t\t};\n\t\t\t} catch (error) {\n\t\t\t\treturn { success: false, error: error.message };\n\t\t\t}\n\t\t}`, ctx)\n\n\trequire.NoError(t, err)\n\trequire.NotNil(t, res)\n\n\tresult, ok := res.(map[string]interface{})\n\trequire.True(t, ok, \"result should be a map\")\n\n\tsuccess, _ := result[\"success\"].(bool)\n\tif !success {\n\t\tt.Logf(\"Test result: %v\", result)\n\t}\n\trequire.True(t, success, \"Test should succeed, error: %v\", result[\"error\"])\n\n\t// Should have 2 results\n\tcount, _ := result[\"count\"].(float64)\n\tassert.Equal(t, float64(2), count, \"Should have 2 results\")\n\n\t// Check individual results\n\tresults, _ := result[\"results\"].([]interface{})\n\trequire.Len(t, results, 2)\n\n\tfor i, r := range results {\n\t\trMap, _ := r.(map[string]interface{})\n\t\thasContent, _ := rMap[\"has_content\"].(bool)\n\t\tassert.True(t, hasContent, \"Result %d should have content\", i)\n\t\terrorStr, _ := rMap[\"error\"].(string)\n\t\tassert.Empty(t, errorStr, \"Result %d should not have error\", i)\n\t}\n}\n\n// TestLlm_All_WithCallback_V8 tests ctx.llm.All with global callback\nfunc TestLlm_All_WithCallback_V8(t *testing.T) {\n\tif testing.Short() {\n\t\tt.Skip(\"Skipping integration test in short mode\")\n\t}\n\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\tauthorized := &types.AuthorizedInfo{\n\t\tSubject: \"test-user\",\n\t\tUserID: \"test-123\",\n\t\tTenantID: \"test-tenant\",\n\t}\n\n\tctx := context.New(stdContext.Background(), authorized, \"test-chat-llm-all-callback\")\n\tctx.AssistantID = \"tests.simple-greeting\"\n\tdefer ctx.Release()\n\n\t// Test All with global callback\n\tres, err := v8.Call(v8.CallOptions{}, `\n\t\tfunction test(ctx) {\n\t\t\ttry {\n\t\t\t\tlet callbackCount = 0;\n\t\t\t\tlet indexesSeen = new Set();\n\t\t\t\t\n\t\t\t\tconst results = ctx.llm.All([\n\t\t\t\t\t{\n\t\t\t\t\t\tconnector: \"gpt-4o-mini\",\n\t\t\t\t\t\tmessages: [{ role: \"user\", content: \"Say 'A'\" }],\n\t\t\t\t\t\toptions: { temperature: 0.1, max_tokens: 5 }\n\t\t\t\t\t},\n\t\t\t\t\t{\n\t\t\t\t\t\tconnector: \"gpt-4o-mini\",\n\t\t\t\t\t\tmessages: [{ role: \"user\", content: \"Say 'B'\" }],\n\t\t\t\t\t\toptions: { temperature: 0.1, max_tokens: 5 }\n\t\t\t\t\t}\n\t\t\t\t], {\n\t\t\t\t\tonChunk: function(connectorID, index, msg) {\n\t\t\t\t\t\tcallbackCount++;\n\t\t\t\t\t\tindexesSeen.add(index);\n\t\t\t\t\t\treturn 0;\n\t\t\t\t\t}\n\t\t\t\t});\n\t\t\t\t\n\t\t\t\treturn {\n\t\t\t\t\tsuccess: true,\n\t\t\t\t\tcount: results.length,\n\t\t\t\t\tcallbackCount: callbackCount,\n\t\t\t\t\tindexesSeen: indexesSeen.size\n\t\t\t\t};\n\t\t\t} catch (error) {\n\t\t\t\treturn { success: false, error: error.message };\n\t\t\t}\n\t\t}`, ctx)\n\n\trequire.NoError(t, err)\n\trequire.NotNil(t, res)\n\n\tresult, ok := res.(map[string]interface{})\n\trequire.True(t, ok, \"result should be a map\")\n\n\tsuccess, _ := result[\"success\"].(bool)\n\tif !success {\n\t\tt.Logf(\"Test result: %v\", result)\n\t}\n\trequire.True(t, success, \"Test should succeed, error: %v\", result[\"error\"])\n\n\t// Callback should have been called\n\tcallbackCount, _ := result[\"callbackCount\"].(float64)\n\tassert.Greater(t, callbackCount, float64(0), \"Callback should be called\")\n\n\t// Should have seen at least one index (both requests may complete so fast that only one is tracked)\n\t// Note: Due to V8 thread safety with channel-based approach, callbacks are serialized\n\tindexesSeen, _ := result[\"indexesSeen\"].(float64)\n\tassert.GreaterOrEqual(t, indexesSeen, float64(1), \"Should have seen callbacks from at least one request\")\n}\n\n// TestLlm_Any_V8 tests ctx.llm.Any - returns first success\nfunc TestLlm_Any_V8(t *testing.T) {\n\tif testing.Short() {\n\t\tt.Skip(\"Skipping integration test in short mode\")\n\t}\n\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\tauthorized := &types.AuthorizedInfo{\n\t\tSubject: \"test-user\",\n\t\tUserID: \"test-123\",\n\t\tTenantID: \"test-tenant\",\n\t}\n\n\tctx := context.New(stdContext.Background(), authorized, \"test-chat-llm-any\")\n\tctx.AssistantID = \"tests.simple-greeting\"\n\tdefer ctx.Release()\n\n\t// Test Any - should return first successful result\n\tres, err := v8.Call(v8.CallOptions{}, `\n\t\tfunction test(ctx) {\n\t\t\ttry {\n\t\t\t\tconst results = ctx.llm.Any([\n\t\t\t\t\t{\n\t\t\t\t\t\tconnector: \"gpt-4o-mini\",\n\t\t\t\t\t\tmessages: [{ role: \"user\", content: \"Say 'hello'\" }],\n\t\t\t\t\t\toptions: { temperature: 0.1, max_tokens: 5 }\n\t\t\t\t\t},\n\t\t\t\t\t{\n\t\t\t\t\t\tconnector: \"gpt-4o-mini\",\n\t\t\t\t\t\tmessages: [{ role: \"user\", content: \"Say 'world'\" }],\n\t\t\t\t\t\toptions: { temperature: 0.1, max_tokens: 5 }\n\t\t\t\t\t}\n\t\t\t\t]);\n\t\t\t\t\n\t\t\t\t// Any returns array with single successful result\n\t\t\t\treturn {\n\t\t\t\t\tsuccess: true,\n\t\t\t\t\tcount: results.length,\n\t\t\t\t\tfirst_has_content: results[0] && results[0].content && results[0].content.length > 0,\n\t\t\t\t\tfirst_error: results[0] ? (results[0].error || \"\") : \"no result\"\n\t\t\t\t};\n\t\t\t} catch (error) {\n\t\t\t\treturn { success: false, error: error.message };\n\t\t\t}\n\t\t}`, ctx)\n\n\trequire.NoError(t, err)\n\trequire.NotNil(t, res)\n\n\tresult, ok := res.(map[string]interface{})\n\trequire.True(t, ok, \"result should be a map\")\n\n\tsuccess, _ := result[\"success\"].(bool)\n\tif !success {\n\t\tt.Logf(\"Test result: %v\", result)\n\t}\n\trequire.True(t, success, \"Test should succeed, error: %v\", result[\"error\"])\n\n\t// Any returns single result on success\n\tcount, _ := result[\"count\"].(float64)\n\tassert.Equal(t, float64(1), count, \"Should have 1 result (first success)\")\n\n\tfirstHasContent, _ := result[\"first_has_content\"].(bool)\n\tassert.True(t, firstHasContent, \"First result should have content\")\n\n\tfirstError, _ := result[\"first_error\"].(string)\n\tassert.Empty(t, firstError, \"First result should not have error\")\n}\n\n// TestLlm_Race_V8 tests ctx.llm.Race - returns first completion\nfunc TestLlm_Race_V8(t *testing.T) {\n\tif testing.Short() {\n\t\tt.Skip(\"Skipping integration test in short mode\")\n\t}\n\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\tauthorized := &types.AuthorizedInfo{\n\t\tSubject: \"test-user\",\n\t\tUserID: \"test-123\",\n\t\tTenantID: \"test-tenant\",\n\t}\n\n\tctx := context.New(stdContext.Background(), authorized, \"test-chat-llm-race\")\n\tctx.AssistantID = \"tests.simple-greeting\"\n\tdefer ctx.Release()\n\n\t// Test Race - should return first completed result\n\tres, err := v8.Call(v8.CallOptions{}, `\n\t\tfunction test(ctx) {\n\t\t\ttry {\n\t\t\t\tconst results = ctx.llm.Race([\n\t\t\t\t\t{\n\t\t\t\t\t\tconnector: \"gpt-4o-mini\",\n\t\t\t\t\t\tmessages: [{ role: \"user\", content: \"Say 'fast'\" }],\n\t\t\t\t\t\toptions: { temperature: 0.1, max_tokens: 5 }\n\t\t\t\t\t},\n\t\t\t\t\t{\n\t\t\t\t\t\tconnector: \"gpt-4o-mini\",\n\t\t\t\t\t\tmessages: [{ role: \"user\", content: \"Say 'slow'\" }],\n\t\t\t\t\t\toptions: { temperature: 0.1, max_tokens: 5 }\n\t\t\t\t\t}\n\t\t\t\t]);\n\t\t\t\t\n\t\t\t\t// Race returns array with single result (first to complete)\n\t\t\t\treturn {\n\t\t\t\t\tsuccess: true,\n\t\t\t\t\tcount: results.length,\n\t\t\t\t\thas_result: results[0] !== undefined,\n\t\t\t\t\tfirst_connector: results[0] ? results[0].connector : \"\",\n\t\t\t\t\tfirst_has_content: results[0] && results[0].content && results[0].content.length > 0\n\t\t\t\t};\n\t\t\t} catch (error) {\n\t\t\t\treturn { success: false, error: error.message };\n\t\t\t}\n\t\t}`, ctx)\n\n\trequire.NoError(t, err)\n\trequire.NotNil(t, res)\n\n\tresult, ok := res.(map[string]interface{})\n\trequire.True(t, ok, \"result should be a map\")\n\n\tsuccess, _ := result[\"success\"].(bool)\n\tif !success {\n\t\tt.Logf(\"Test result: %v\", result)\n\t}\n\trequire.True(t, success, \"Test should succeed, error: %v\", result[\"error\"])\n\n\t// Race returns single result\n\tcount, _ := result[\"count\"].(float64)\n\tassert.Equal(t, float64(1), count, \"Should have 1 result (first to complete)\")\n\n\thasResult, _ := result[\"has_result\"].(bool)\n\tassert.True(t, hasResult, \"Should have a result\")\n\n\tfirstConnector, _ := result[\"first_connector\"].(string)\n\tassert.Equal(t, \"gpt-4o-mini\", firstConnector, \"First result should have connector\")\n}\n\n// TestLlm_Stream_InvalidConnector_V8 tests error handling for invalid connector\nfunc TestLlm_Stream_InvalidConnector_V8(t *testing.T) {\n\tif testing.Short() {\n\t\tt.Skip(\"Skipping integration test in short mode\")\n\t}\n\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\tauthorized := &types.AuthorizedInfo{\n\t\tSubject: \"test-user\",\n\t\tUserID: \"test-123\",\n\t\tTenantID: \"test-tenant\",\n\t}\n\n\tctx := context.New(stdContext.Background(), authorized, \"test-chat-llm-invalid\")\n\tctx.AssistantID = \"tests.simple-greeting\"\n\tdefer ctx.Release()\n\n\t// Test Stream with invalid connector\n\tres, err := v8.Call(v8.CallOptions{}, `\n\t\tfunction test(ctx) {\n\t\t\ttry {\n\t\t\t\tconst result = ctx.llm.Stream(\"invalid-connector-that-does-not-exist\", [\n\t\t\t\t\t{ role: \"user\", content: \"Hello\" }\n\t\t\t\t]);\n\t\t\t\t\n\t\t\t\treturn {\n\t\t\t\t\thas_error: result.error && result.error.length > 0,\n\t\t\t\t\terror: result.error || \"\"\n\t\t\t\t};\n\t\t\t} catch (error) {\n\t\t\t\treturn { has_error: true, error: error.message };\n\t\t\t}\n\t\t}`, ctx)\n\n\trequire.NoError(t, err)\n\trequire.NotNil(t, res)\n\n\tresult, ok := res.(map[string]interface{})\n\trequire.True(t, ok, \"result should be a map\")\n\n\thasError, _ := result[\"has_error\"].(bool)\n\tassert.True(t, hasError, \"Should have error for invalid connector\")\n}\n" }, { "path": "agent/context/jsapi_mcp.go", "content": "package context\n\nimport (\n\t\"fmt\"\n\n\t\"github.com/yaoapp/gou/mcp/types\"\n\t\"github.com/yaoapp/gou/runtime/v8/bridge\"\n\t\"rogchap.com/v8go\"\n)\n\n// Suppress unused import warning - types is used in other functions\nvar _ = types.ToolCall{}\n\n// MCP JavaScript API methods\n// These methods expose MCP functionality to JavaScript runtime\n\n// mcpListResourcesMethod implements ctx.MCP.ListResources(mcp, cursor)\n// Lists all available resources from an MCP client\nfunc (ctx *Context) mcpListResourcesMethod(iso *v8go.Isolate) *v8go.FunctionTemplate {\n\treturn v8go.NewFunctionTemplate(iso, func(info *v8go.FunctionCallbackInfo) *v8go.Value {\n\t\tv8ctx := info.Context()\n\t\targs := info.Args()\n\n\t\tif len(args) < 1 {\n\t\t\treturn bridge.JsException(v8ctx, \"ListResources requires mcp parameter\")\n\t\t}\n\n\t\tmcpID := args[0].String()\n\t\tcursor := \"\"\n\t\tif len(args) >= 2 && !args[1].IsUndefined() {\n\t\t\tcursor = args[1].String()\n\t\t}\n\n\t\tresult, err := ctx.ListResources(mcpID, cursor)\n\t\tif err != nil {\n\t\t\treturn bridge.JsException(v8ctx, err.Error())\n\t\t}\n\n\t\tjsVal, err := bridge.JsValue(v8ctx, result)\n\t\tif err != nil {\n\t\t\treturn bridge.JsException(v8ctx, err.Error())\n\t\t}\n\n\t\treturn jsVal\n\t})\n}\n\n// mcpReadResourceMethod implements ctx.MCP.ReadResource(mcp, uri)\n// Reads a specific resource from an MCP client\nfunc (ctx *Context) mcpReadResourceMethod(iso *v8go.Isolate) *v8go.FunctionTemplate {\n\treturn v8go.NewFunctionTemplate(iso, func(info *v8go.FunctionCallbackInfo) *v8go.Value {\n\t\tv8ctx := info.Context()\n\t\targs := info.Args()\n\n\t\tif len(args) < 2 {\n\t\t\treturn bridge.JsException(v8ctx, \"ReadResource requires mcp and uri parameters\")\n\t\t}\n\n\t\tmcpID := args[0].String()\n\t\turi := args[1].String()\n\n\t\tresult, err := ctx.ReadResource(mcpID, uri)\n\t\tif err != nil {\n\t\t\treturn bridge.JsException(v8ctx, err.Error())\n\t\t}\n\n\t\tjsVal, err := bridge.JsValue(v8ctx, result)\n\t\tif err != nil {\n\t\t\treturn bridge.JsException(v8ctx, err.Error())\n\t\t}\n\n\t\treturn jsVal\n\t})\n}\n\n// mcpListToolsMethod implements ctx.MCP.ListTools(mcp, cursor)\n// Lists all available tools from an MCP client\nfunc (ctx *Context) mcpListToolsMethod(iso *v8go.Isolate) *v8go.FunctionTemplate {\n\treturn v8go.NewFunctionTemplate(iso, func(info *v8go.FunctionCallbackInfo) *v8go.Value {\n\t\tv8ctx := info.Context()\n\t\targs := info.Args()\n\n\t\tif len(args) < 1 {\n\t\t\treturn bridge.JsException(v8ctx, \"ListTools requires mcp parameter\")\n\t\t}\n\n\t\tmcpID := args[0].String()\n\t\tcursor := \"\"\n\t\tif len(args) >= 2 && !args[1].IsUndefined() {\n\t\t\tcursor = args[1].String()\n\t\t}\n\n\t\tresult, err := ctx.ListTools(mcpID, cursor)\n\t\tif err != nil {\n\t\t\treturn bridge.JsException(v8ctx, err.Error())\n\t\t}\n\n\t\tjsVal, err := bridge.JsValue(v8ctx, result)\n\t\tif err != nil {\n\t\t\treturn bridge.JsException(v8ctx, err.Error())\n\t\t}\n\n\t\treturn jsVal\n\t})\n}\n\n// mcpCallToolMethod implements ctx.MCP.CallTool(mcp, name, args)\n// Calls a specific tool from an MCP client\n// Returns CallToolResult with parsed 'result' field for convenience\nfunc (ctx *Context) mcpCallToolMethod(iso *v8go.Isolate) *v8go.FunctionTemplate {\n\treturn v8go.NewFunctionTemplate(iso, func(info *v8go.FunctionCallbackInfo) *v8go.Value {\n\t\tv8ctx := info.Context()\n\t\targs := info.Args()\n\n\t\tif len(args) < 2 {\n\t\t\treturn bridge.JsException(v8ctx, \"CallTool requires mcp and name parameters\")\n\t\t}\n\n\t\tmcpID := args[0].String()\n\t\ttoolName := args[1].String()\n\n\t\t// Parse arguments (optional)\n\t\tvar toolArgs map[string]interface{}\n\t\tif len(args) >= 3 && !args[2].IsUndefined() {\n\t\t\tgoVal, err := bridge.GoValue(args[2], v8ctx)\n\t\t\tif err != nil {\n\t\t\t\treturn bridge.JsException(v8ctx, \"invalid tool arguments: \"+err.Error())\n\t\t\t}\n\t\t\tif argsMap, ok := goVal.(map[string]interface{}); ok {\n\t\t\t\ttoolArgs = argsMap\n\t\t\t}\n\t\t}\n\n\t\tresponse, err := ctx.CallTool(mcpID, toolName, toolArgs)\n\t\tif err != nil {\n\t\t\treturn bridge.JsException(v8ctx, err.Error())\n\t\t}\n\n\t\t// Return parsed result directly\n\t\tresult := parseCallToolResponse(response)\n\n\t\tjsVal, err := bridge.JsValue(v8ctx, result)\n\t\tif err != nil {\n\t\t\treturn bridge.JsException(v8ctx, err.Error())\n\t\t}\n\n\t\treturn jsVal\n\t})\n}\n\n// mcpCallToolsMethod implements ctx.MCP.CallTools(mcp, tools)\n// Calls multiple tools sequentially from an MCP client\n// Returns CallToolsResult with parsed 'result' field in each item\nfunc (ctx *Context) mcpCallToolsMethod(iso *v8go.Isolate) *v8go.FunctionTemplate {\n\treturn v8go.NewFunctionTemplate(iso, func(info *v8go.FunctionCallbackInfo) *v8go.Value {\n\t\tv8ctx := info.Context()\n\t\targs := info.Args()\n\n\t\tif len(args) < 2 {\n\t\t\treturn bridge.JsException(v8ctx, \"CallTools requires mcp and tools parameters\")\n\t\t}\n\n\t\tmcpID := args[0].String()\n\n\t\t// Parse tools array\n\t\tgoVal, err := bridge.GoValue(args[1], v8ctx)\n\t\tif err != nil {\n\t\t\treturn bridge.JsException(v8ctx, \"invalid tools parameter: \"+err.Error())\n\t\t}\n\n\t\ttoolsArray, ok := goVal.([]interface{})\n\t\tif !ok {\n\t\t\treturn bridge.JsException(v8ctx, \"tools parameter must be an array\")\n\t\t}\n\n\t\t// Convert to ToolCall array\n\t\ttools := make([]types.ToolCall, 0, len(toolsArray))\n\t\tfor i, item := range toolsArray {\n\t\t\ttoolMap, ok := item.(map[string]interface{})\n\t\t\tif !ok {\n\t\t\t\treturn bridge.JsException(v8ctx, \"each tool must be an object\")\n\t\t\t}\n\n\t\t\tname, ok := toolMap[\"name\"].(string)\n\t\t\tif !ok {\n\t\t\t\treturn bridge.JsException(v8ctx, \"tool name is required\")\n\t\t\t}\n\n\t\t\ttoolCall := types.ToolCall{\n\t\t\t\tName: name,\n\t\t\t}\n\n\t\t\tif argsVal, exists := toolMap[\"arguments\"]; exists && argsVal != nil {\n\t\t\t\tif argsMap, ok := argsVal.(map[string]interface{}); ok {\n\t\t\t\t\ttoolCall.Arguments = argsMap\n\t\t\t\t}\n\t\t\t}\n\n\t\t\ttools = append(tools, toolCall)\n\n\t\t\t// Suppress unused variable warning\n\t\t\t_ = i\n\t\t}\n\n\t\tresponse, err := ctx.CallTools(mcpID, tools)\n\t\tif err != nil {\n\t\t\treturn bridge.JsException(v8ctx, err.Error())\n\t\t}\n\n\t\t// Return parsed results directly as array\n\t\tresult := parseCallToolsResponse(response)\n\n\t\tjsVal, err := bridge.JsValue(v8ctx, result)\n\t\tif err != nil {\n\t\t\treturn bridge.JsException(v8ctx, err.Error())\n\t\t}\n\n\t\treturn jsVal\n\t})\n}\n\n// mcpCallToolsParallelMethod implements ctx.MCP.CallToolsParallel(mcp, tools)\n// Calls multiple tools in parallel from an MCP client\n// Returns CallToolsResult with parsed 'result' field in each item\nfunc (ctx *Context) mcpCallToolsParallelMethod(iso *v8go.Isolate) *v8go.FunctionTemplate {\n\treturn v8go.NewFunctionTemplate(iso, func(info *v8go.FunctionCallbackInfo) *v8go.Value {\n\t\tv8ctx := info.Context()\n\t\targs := info.Args()\n\n\t\tif len(args) < 2 {\n\t\t\treturn bridge.JsException(v8ctx, \"CallToolsParallel requires mcp and tools parameters\")\n\t\t}\n\n\t\tmcpID := args[0].String()\n\n\t\t// Parse tools array\n\t\tgoVal, err := bridge.GoValue(args[1], v8ctx)\n\t\tif err != nil {\n\t\t\treturn bridge.JsException(v8ctx, \"invalid tools parameter: \"+err.Error())\n\t\t}\n\n\t\ttoolsArray, ok := goVal.([]interface{})\n\t\tif !ok {\n\t\t\treturn bridge.JsException(v8ctx, \"tools parameter must be an array\")\n\t\t}\n\n\t\t// Convert to ToolCall array\n\t\ttools := make([]types.ToolCall, 0, len(toolsArray))\n\t\tfor i, item := range toolsArray {\n\t\t\ttoolMap, ok := item.(map[string]interface{})\n\t\t\tif !ok {\n\t\t\t\treturn bridge.JsException(v8ctx, \"each tool must be an object\")\n\t\t\t}\n\n\t\t\tname, ok := toolMap[\"name\"].(string)\n\t\t\tif !ok {\n\t\t\t\treturn bridge.JsException(v8ctx, \"tool name is required\")\n\t\t\t}\n\n\t\t\ttoolCall := types.ToolCall{\n\t\t\t\tName: name,\n\t\t\t}\n\n\t\t\tif argsVal, exists := toolMap[\"arguments\"]; exists && argsVal != nil {\n\t\t\t\tif argsMap, ok := argsVal.(map[string]interface{}); ok {\n\t\t\t\t\ttoolCall.Arguments = argsMap\n\t\t\t\t}\n\t\t\t}\n\n\t\t\ttools = append(tools, toolCall)\n\n\t\t\t// Suppress unused variable warning\n\t\t\t_ = i\n\t\t}\n\n\t\tresponse, err := ctx.CallToolsParallel(mcpID, tools)\n\t\tif err != nil {\n\t\t\treturn bridge.JsException(v8ctx, err.Error())\n\t\t}\n\n\t\t// Return parsed results directly as array\n\t\tresult := parseCallToolsResponse(response)\n\n\t\tjsVal, err := bridge.JsValue(v8ctx, result)\n\t\tif err != nil {\n\t\t\treturn bridge.JsException(v8ctx, err.Error())\n\t\t}\n\n\t\treturn jsVal\n\t})\n}\n\n// mcpListPromptsMethod implements ctx.MCP.ListPrompts(mcp, cursor)\n// Lists all available prompts from an MCP client\nfunc (ctx *Context) mcpListPromptsMethod(iso *v8go.Isolate) *v8go.FunctionTemplate {\n\treturn v8go.NewFunctionTemplate(iso, func(info *v8go.FunctionCallbackInfo) *v8go.Value {\n\t\tv8ctx := info.Context()\n\t\targs := info.Args()\n\n\t\tif len(args) < 1 {\n\t\t\treturn bridge.JsException(v8ctx, \"ListPrompts requires mcp parameter\")\n\t\t}\n\n\t\tmcpID := args[0].String()\n\t\tcursor := \"\"\n\t\tif len(args) >= 2 && !args[1].IsUndefined() {\n\t\t\tcursor = args[1].String()\n\t\t}\n\n\t\tresult, err := ctx.ListPrompts(mcpID, cursor)\n\t\tif err != nil {\n\t\t\treturn bridge.JsException(v8ctx, err.Error())\n\t\t}\n\n\t\tjsVal, err := bridge.JsValue(v8ctx, result)\n\t\tif err != nil {\n\t\t\treturn bridge.JsException(v8ctx, err.Error())\n\t\t}\n\n\t\treturn jsVal\n\t})\n}\n\n// mcpGetPromptMethod implements ctx.MCP.GetPrompt(mcp, name, args)\n// Gets a specific prompt from an MCP client\nfunc (ctx *Context) mcpGetPromptMethod(iso *v8go.Isolate) *v8go.FunctionTemplate {\n\treturn v8go.NewFunctionTemplate(iso, func(info *v8go.FunctionCallbackInfo) *v8go.Value {\n\t\tv8ctx := info.Context()\n\t\targs := info.Args()\n\n\t\tif len(args) < 2 {\n\t\t\treturn bridge.JsException(v8ctx, \"GetPrompt requires mcp and name parameters\")\n\t\t}\n\n\t\tmcpID := args[0].String()\n\t\tpromptName := args[1].String()\n\n\t\t// Parse arguments (optional)\n\t\tvar promptArgs map[string]interface{}\n\t\tif len(args) >= 3 && !args[2].IsUndefined() {\n\t\t\tgoVal, err := bridge.GoValue(args[2], v8ctx)\n\t\t\tif err != nil {\n\t\t\t\treturn bridge.JsException(v8ctx, \"invalid prompt arguments: \"+err.Error())\n\t\t\t}\n\t\t\tif argsMap, ok := goVal.(map[string]interface{}); ok {\n\t\t\t\tpromptArgs = argsMap\n\t\t\t}\n\t\t}\n\n\t\tresult, err := ctx.GetPrompt(mcpID, promptName, promptArgs)\n\t\tif err != nil {\n\t\t\treturn bridge.JsException(v8ctx, err.Error())\n\t\t}\n\n\t\tjsVal, err := bridge.JsValue(v8ctx, result)\n\t\tif err != nil {\n\t\t\treturn bridge.JsException(v8ctx, err.Error())\n\t\t}\n\n\t\treturn jsVal\n\t})\n}\n\n// mcpListSamplesMethod implements ctx.MCP.ListSamples(mcp, type, name)\n// Lists all available samples from an MCP client\nfunc (ctx *Context) mcpListSamplesMethod(iso *v8go.Isolate) *v8go.FunctionTemplate {\n\treturn v8go.NewFunctionTemplate(iso, func(info *v8go.FunctionCallbackInfo) *v8go.Value {\n\t\tv8ctx := info.Context()\n\t\targs := info.Args()\n\n\t\tif len(args) < 3 {\n\t\t\treturn bridge.JsException(v8ctx, \"ListSamples requires mcp, type, and name parameters\")\n\t\t}\n\n\t\tmcpID := args[0].String()\n\t\tsampleType := types.SampleItemType(args[1].String())\n\t\tname := args[2].String()\n\n\t\tresult, err := ctx.ListSamples(mcpID, sampleType, name)\n\t\tif err != nil {\n\t\t\treturn bridge.JsException(v8ctx, err.Error())\n\t\t}\n\n\t\tjsVal, err := bridge.JsValue(v8ctx, result)\n\t\tif err != nil {\n\t\t\treturn bridge.JsException(v8ctx, err.Error())\n\t\t}\n\n\t\treturn jsVal\n\t})\n}\n\n// mcpGetSampleMethod implements ctx.MCP.GetSample(mcp, type, name, index)\n// Gets a specific sample from an MCP client\nfunc (ctx *Context) mcpGetSampleMethod(iso *v8go.Isolate) *v8go.FunctionTemplate {\n\treturn v8go.NewFunctionTemplate(iso, func(info *v8go.FunctionCallbackInfo) *v8go.Value {\n\t\tv8ctx := info.Context()\n\t\targs := info.Args()\n\n\t\tif len(args) < 4 {\n\t\t\treturn bridge.JsException(v8ctx, \"GetSample requires mcp, type, name, and index parameters\")\n\t\t}\n\n\t\tmcpID := args[0].String()\n\t\tsampleType := types.SampleItemType(args[1].String())\n\t\tname := args[2].String()\n\n\t\t// Parse index\n\t\tindexVal, err := bridge.GoValue(args[3], v8ctx)\n\t\tif err != nil {\n\t\t\treturn bridge.JsException(v8ctx, \"invalid index parameter: \"+err.Error())\n\t\t}\n\n\t\tvar index int\n\t\tswitch v := indexVal.(type) {\n\t\tcase int:\n\t\t\tindex = v\n\t\tcase int32:\n\t\t\tindex = int(v)\n\t\tcase int64:\n\t\t\tindex = int(v)\n\t\tcase float64:\n\t\t\tindex = int(v)\n\t\tdefault:\n\t\t\treturn bridge.JsException(v8ctx, \"index must be a number\")\n\t\t}\n\n\t\tresult, err := ctx.GetSample(mcpID, sampleType, name, index)\n\t\tif err != nil {\n\t\t\treturn bridge.JsException(v8ctx, err.Error())\n\t\t}\n\n\t\tjsVal, err := bridge.JsValue(v8ctx, result)\n\t\tif err != nil {\n\t\t\treturn bridge.JsException(v8ctx, err.Error())\n\t\t}\n\n\t\treturn jsVal\n\t})\n}\n\n// mcpAllMethod implements ctx.mcp.All(requests)\n// Calls tools on multiple MCP servers concurrently and waits for all to complete (like Promise.all)\n// Each request should have: { mcp: string, tool: string, arguments?: object }\nfunc (ctx *Context) mcpAllMethod(iso *v8go.Isolate) *v8go.FunctionTemplate {\n\treturn v8go.NewFunctionTemplate(iso, func(info *v8go.FunctionCallbackInfo) *v8go.Value {\n\t\tv8ctx := info.Context()\n\t\targs := info.Args()\n\n\t\tif len(args) < 1 {\n\t\t\treturn bridge.JsException(v8ctx, \"All requires requests parameter\")\n\t\t}\n\n\t\t// Parse requests array\n\t\trequests, err := ctx.parseMCPToolRequests(args[0], v8ctx)\n\t\tif err != nil {\n\t\t\treturn bridge.JsException(v8ctx, err.Error())\n\t\t}\n\n\t\t// Execute all requests\n\t\tresults := ctx.CallToolAll(requests)\n\n\t\t// Convert results to JS value\n\t\tjsVal, err := bridge.JsValue(v8ctx, results)\n\t\tif err != nil {\n\t\t\treturn bridge.JsException(v8ctx, \"failed to convert results: \"+err.Error())\n\t\t}\n\n\t\treturn jsVal\n\t})\n}\n\n// mcpAnyMethod implements ctx.mcp.Any(requests)\n// Calls tools on multiple MCP servers concurrently and returns when any succeeds (like Promise.any)\n// Each request should have: { mcp: string, tool: string, arguments?: object }\nfunc (ctx *Context) mcpAnyMethod(iso *v8go.Isolate) *v8go.FunctionTemplate {\n\treturn v8go.NewFunctionTemplate(iso, func(info *v8go.FunctionCallbackInfo) *v8go.Value {\n\t\tv8ctx := info.Context()\n\t\targs := info.Args()\n\n\t\tif len(args) < 1 {\n\t\t\treturn bridge.JsException(v8ctx, \"Any requires requests parameter\")\n\t\t}\n\n\t\t// Parse requests array\n\t\trequests, err := ctx.parseMCPToolRequests(args[0], v8ctx)\n\t\tif err != nil {\n\t\t\treturn bridge.JsException(v8ctx, err.Error())\n\t\t}\n\n\t\t// Execute requests until any succeeds\n\t\tresults := ctx.CallToolAny(requests)\n\n\t\t// Convert results to JS value\n\t\tjsVal, err := bridge.JsValue(v8ctx, results)\n\t\tif err != nil {\n\t\t\treturn bridge.JsException(v8ctx, \"failed to convert results: \"+err.Error())\n\t\t}\n\n\t\treturn jsVal\n\t})\n}\n\n// mcpRaceMethod implements ctx.mcp.Race(requests)\n// Calls tools on multiple MCP servers concurrently and returns when any completes (like Promise.race)\n// Each request should have: { mcp: string, tool: string, arguments?: object }\nfunc (ctx *Context) mcpRaceMethod(iso *v8go.Isolate) *v8go.FunctionTemplate {\n\treturn v8go.NewFunctionTemplate(iso, func(info *v8go.FunctionCallbackInfo) *v8go.Value {\n\t\tv8ctx := info.Context()\n\t\targs := info.Args()\n\n\t\tif len(args) < 1 {\n\t\t\treturn bridge.JsException(v8ctx, \"Race requires requests parameter\")\n\t\t}\n\n\t\t// Parse requests array\n\t\trequests, err := ctx.parseMCPToolRequests(args[0], v8ctx)\n\t\tif err != nil {\n\t\t\treturn bridge.JsException(v8ctx, err.Error())\n\t\t}\n\n\t\t// Execute requests and return first completion\n\t\tresults := ctx.CallToolRace(requests)\n\n\t\t// Convert results to JS value\n\t\tjsVal, err := bridge.JsValue(v8ctx, results)\n\t\tif err != nil {\n\t\t\treturn bridge.JsException(v8ctx, \"failed to convert results: \"+err.Error())\n\t\t}\n\n\t\treturn jsVal\n\t})\n}\n\n// parseMCPToolRequests parses JS requests array into MCPToolRequest slice\nfunc (ctx *Context) parseMCPToolRequests(arg *v8go.Value, v8ctx *v8go.Context) ([]*MCPToolRequest, error) {\n\tgoVal, err := bridge.GoValue(arg, v8ctx)\n\tif err != nil {\n\t\treturn nil, fmt.Errorf(\"invalid requests: %w\", err)\n\t}\n\n\trequestsArray, ok := goVal.([]interface{})\n\tif !ok {\n\t\treturn nil, fmt.Errorf(\"requests must be an array\")\n\t}\n\n\trequests := make([]*MCPToolRequest, 0, len(requestsArray))\n\tfor _, item := range requestsArray {\n\t\treqMap, ok := item.(map[string]interface{})\n\t\tif !ok {\n\t\t\treturn nil, fmt.Errorf(\"each request must be an object\")\n\t\t}\n\n\t\t// Required: mcp\n\t\tmcpID, ok := reqMap[\"mcp\"].(string)\n\t\tif !ok || mcpID == \"\" {\n\t\t\treturn nil, fmt.Errorf(\"request.mcp is required and must be a string\")\n\t\t}\n\n\t\t// Required: tool\n\t\ttool, ok := reqMap[\"tool\"].(string)\n\t\tif !ok || tool == \"\" {\n\t\t\treturn nil, fmt.Errorf(\"request.tool is required and must be a string\")\n\t\t}\n\n\t\treq := &MCPToolRequest{\n\t\t\tMCP: mcpID,\n\t\t\tTool: tool,\n\t\t}\n\n\t\t// Optional: arguments\n\t\tif args, exists := reqMap[\"arguments\"]; exists && args != nil {\n\t\t\treq.Arguments = args\n\t\t}\n\n\t\trequests = append(requests, req)\n\t}\n\n\treturn requests, nil\n}\n\n// newMCPObject creates a new MCP object with all MCP methods\nfunc (ctx *Context) newMCPObject(iso *v8go.Isolate) *v8go.ObjectTemplate {\n\tmcpObj := v8go.NewObjectTemplate(iso)\n\n\t// Resource operations\n\tmcpObj.Set(\"ListResources\", ctx.mcpListResourcesMethod(iso))\n\tmcpObj.Set(\"ReadResource\", ctx.mcpReadResourceMethod(iso))\n\n\t// Tool operations\n\tmcpObj.Set(\"ListTools\", ctx.mcpListToolsMethod(iso))\n\tmcpObj.Set(\"CallTool\", ctx.mcpCallToolMethod(iso))\n\tmcpObj.Set(\"CallTools\", ctx.mcpCallToolsMethod(iso))\n\tmcpObj.Set(\"CallToolsParallel\", ctx.mcpCallToolsParallelMethod(iso))\n\n\t// Cross-server parallel tool operations (Promise-like patterns)\n\tmcpObj.Set(\"All\", ctx.mcpAllMethod(iso))\n\tmcpObj.Set(\"Any\", ctx.mcpAnyMethod(iso))\n\tmcpObj.Set(\"Race\", ctx.mcpRaceMethod(iso))\n\n\t// Prompt operations\n\tmcpObj.Set(\"ListPrompts\", ctx.mcpListPromptsMethod(iso))\n\tmcpObj.Set(\"GetPrompt\", ctx.mcpGetPromptMethod(iso))\n\n\t// Sample operations\n\tmcpObj.Set(\"ListSamples\", ctx.mcpListSamplesMethod(iso))\n\tmcpObj.Set(\"GetSample\", ctx.mcpGetSampleMethod(iso))\n\n\treturn mcpObj\n}\n" }, { "path": "agent/context/jsapi_mcp_test.go", "content": "package context_test\n\nimport (\n\tstdContext \"context\"\n\t\"testing\"\n\n\t\"github.com/stretchr/testify/assert\"\n\tv8 \"github.com/yaoapp/gou/runtime/v8\"\n\t\"github.com/yaoapp/yao/agent/context\"\n\t\"github.com/yaoapp/yao/config\"\n\t\"github.com/yaoapp/yao/test\"\n)\n\n// newMCPTestContext creates a test context for MCP testing\nfunc newMCPTestContext() *context.Context {\n\tctx := context.New(stdContext.Background(), nil, \"test-chat-id\")\n\tctx.AssistantID = \"test-assistant-id\"\n\tctx.Locale = \"en\"\n\tctx.Referer = context.RefererAPI\n\tstack, _, _ := context.EnterStack(ctx, \"test-assistant\", &context.Options{})\n\tctx.Stack = stack\n\treturn ctx\n}\n\n// TestMCPListResources tests MCP.ListResources from JavaScript\nfunc TestMCPListResources(t *testing.T) {\n\ttest.Prepare(t, config.Conf)\n\tdefer test.Clean()\n\n\tctx := newMCPTestContext()\n\n\tres, err := v8.Call(v8.CallOptions{}, `\n\t\tfunction test(ctx) {\n\t\t\t// List resources from echo MCP\n\t\t\tconst result = ctx.mcp.ListResources(\"echo\", \"\")\n\t\t\t\n\t\t\tif (!result || !result.resources) {\n\t\t\t\tthrow new Error(\"Expected resources\")\n\t\t\t}\n\t\t\t\n\t\t\treturn {\n\t\t\t\tcount: result.resources.length,\n\t\t\t\thas_info: result.resources.some(r => r.name === \"info\"),\n\t\t\t\thas_health: result.resources.some(r => r.name === \"health\")\n\t\t\t}\n\t\t}`, ctx)\n\n\tif err != nil {\n\t\tt.Fatalf(\"Call failed: %v\", err)\n\t}\n\n\tresult, ok := res.(map[string]interface{})\n\tif !ok {\n\t\tt.Fatalf(\"Expected map result, got %T\", res)\n\t}\n\n\tassert.Equal(t, float64(2), result[\"count\"], \"should have 2 resources\")\n\tassert.Equal(t, true, result[\"has_info\"], \"should have info resource\")\n\tassert.Equal(t, true, result[\"has_health\"], \"should have health resource\")\n}\n\n// TestMCPReadResource tests MCP.ReadResource from JavaScript\nfunc TestMCPReadResource(t *testing.T) {\n\ttest.Prepare(t, config.Conf)\n\tdefer test.Clean()\n\n\tctx := newMCPTestContext()\n\n\tres, err := v8.Call(v8.CallOptions{}, `\n\t\tfunction test(ctx) {\n\t\t\t// Read info resource\n\t\t\tconst result = ctx.mcp.ReadResource(\"echo\", \"echo://info\")\n\t\t\t\n\t\t\tif (!result || !result.contents) {\n\t\t\t\tthrow new Error(\"Expected contents\")\n\t\t\t}\n\t\t\t\n\t\t\treturn {\n\t\t\t\tcount: result.contents.length,\n\t\t\t\thas_content: result.contents.length > 0\n\t\t\t}\n\t\t}`, ctx)\n\n\tif err != nil {\n\t\tt.Fatalf(\"Call failed: %v\", err)\n\t}\n\n\tresult, ok := res.(map[string]interface{})\n\tif !ok {\n\t\tt.Fatalf(\"Expected map result, got %T\", res)\n\t}\n\n\tassert.Equal(t, float64(1), result[\"count\"], \"should have 1 content\")\n\tassert.Equal(t, true, result[\"has_content\"], \"should have content\")\n}\n\n// TestMCPListTools tests MCP.ListTools from JavaScript\nfunc TestMCPListTools(t *testing.T) {\n\ttest.Prepare(t, config.Conf)\n\tdefer test.Clean()\n\n\tctx := newMCPTestContext()\n\n\tres, err := v8.Call(v8.CallOptions{}, `\n\t\tfunction test(ctx) {\n\t\t\t// List tools from echo MCP\n\t\t\tconst result = ctx.mcp.ListTools(\"echo\", \"\")\n\t\t\t\n\t\t\tif (!result || !result.tools) {\n\t\t\t\tthrow new Error(\"Expected tools\")\n\t\t\t}\n\t\t\t\n\t\t\treturn {\n\t\t\t\tcount: result.tools.length,\n\t\t\t\thas_ping: result.tools.some(t => t.name === \"ping\"),\n\t\t\t\thas_status: result.tools.some(t => t.name === \"status\"),\n\t\t\t\thas_echo: result.tools.some(t => t.name === \"echo\")\n\t\t\t}\n\t\t}`, ctx)\n\n\tif err != nil {\n\t\tt.Fatalf(\"Call failed: %v\", err)\n\t}\n\n\tresult, ok := res.(map[string]interface{})\n\tif !ok {\n\t\tt.Fatalf(\"Expected map result, got %T\", res)\n\t}\n\n\tassert.Equal(t, float64(3), result[\"count\"], \"should have 3 tools\")\n\tassert.Equal(t, true, result[\"has_ping\"], \"should have ping tool\")\n\tassert.Equal(t, true, result[\"has_status\"], \"should have status tool\")\n\tassert.Equal(t, true, result[\"has_echo\"], \"should have echo tool\")\n}\n\n// TestMCPCallTool tests MCP.CallTool from JavaScript\nfunc TestMCPCallTool(t *testing.T) {\n\ttest.Prepare(t, config.Conf)\n\tdefer test.Clean()\n\n\tctx := newMCPTestContext()\n\n\tres, err := v8.Call(v8.CallOptions{}, `\n\t\tfunction test(ctx) {\n\t\t\t// Call ping tool - returns parsed result directly\n\t\t\tconst result = ctx.mcp.CallTool(\"echo\", \"ping\", { count: 3, message: \"test\" })\n\t\t\t\n\t\t\tif (result === undefined || result === null) {\n\t\t\t\tthrow new Error(\"Expected result\")\n\t\t\t}\n\t\t\t\n\t\t\treturn {\n\t\t\t\thas_result: true,\n\t\t\t\tmessage: result.message\n\t\t\t}\n\t\t}`, ctx)\n\n\tif err != nil {\n\t\tt.Fatalf(\"Call failed: %v\", err)\n\t}\n\n\tresult, ok := res.(map[string]interface{})\n\tif !ok {\n\t\tt.Fatalf(\"Expected map result, got %T\", res)\n\t}\n\n\tassert.Equal(t, true, result[\"has_result\"], \"should have result\")\n\tassert.Equal(t, \"test\", result[\"message\"], \"should have message\")\n}\n\n// TestMCPCallTools tests MCP.CallTools from JavaScript\nfunc TestMCPCallTools(t *testing.T) {\n\ttest.Prepare(t, config.Conf)\n\tdefer test.Clean()\n\n\tctx := newMCPTestContext()\n\n\tres, err := v8.Call(v8.CallOptions{}, `\n\t\tfunction test(ctx) {\n\t\t\t// Call multiple tools sequentially - returns array of parsed results\n\t\t\tconst tools = [\n\t\t\t\t{ name: \"ping\", arguments: { count: 1 } },\n\t\t\t\t{ name: \"status\", arguments: { verbose: false } }\n\t\t\t]\n\t\t\t\n\t\t\tconst results = ctx.mcp.CallTools(\"echo\", tools)\n\t\t\t\n\t\t\tif (!Array.isArray(results)) {\n\t\t\t\tthrow new Error(\"Expected array of results\")\n\t\t\t}\n\t\t\t\n\t\t\treturn {\n\t\t\t\tcount: results.length,\n\t\t\t\tping_message: results[0]?.message,\n\t\t\t\tstatus_online: results[1]?.status === \"online\"\n\t\t\t}\n\t\t}`, ctx)\n\n\tif err != nil {\n\t\tt.Fatalf(\"Call failed: %v\", err)\n\t}\n\n\tresult, ok := res.(map[string]interface{})\n\tif !ok {\n\t\tt.Fatalf(\"Expected map result, got %T\", res)\n\t}\n\n\tassert.Equal(t, float64(2), result[\"count\"], \"should have 2 results\")\n\tassert.Equal(t, \"pong\", result[\"ping_message\"], \"ping should return pong\")\n\tassert.Equal(t, true, result[\"status_online\"], \"status should be online\")\n}\n\n// TestMCPCallToolsParallel tests MCP.CallToolsParallel from JavaScript\nfunc TestMCPCallToolsParallel(t *testing.T) {\n\ttest.Prepare(t, config.Conf)\n\tdefer test.Clean()\n\n\tctx := newMCPTestContext()\n\n\tres, err := v8.Call(v8.CallOptions{}, `\n\t\tfunction test(ctx) {\n\t\t\t// Call multiple tools in parallel - returns array of parsed results\n\t\t\tconst tools = [\n\t\t\t\t{ name: \"ping\", arguments: { count: 1 } },\n\t\t\t\t{ name: \"status\", arguments: { verbose: true } }\n\t\t\t]\n\t\t\t\n\t\t\tconst results = ctx.mcp.CallToolsParallel(\"echo\", tools)\n\t\t\t\n\t\t\tif (!Array.isArray(results)) {\n\t\t\t\tthrow new Error(\"Expected array of results\")\n\t\t\t}\n\t\t\t\n\t\t\treturn {\n\t\t\t\tcount: results.length,\n\t\t\t\tping_message: results[0]?.message,\n\t\t\t\tstatus_online: results[1]?.status === \"online\"\n\t\t\t}\n\t\t}`, ctx)\n\n\tif err != nil {\n\t\tt.Fatalf(\"Call failed: %v\", err)\n\t}\n\n\tresult, ok := res.(map[string]interface{})\n\tif !ok {\n\t\tt.Fatalf(\"Expected map result, got %T\", res)\n\t}\n\n\tassert.Equal(t, float64(2), result[\"count\"], \"should have 2 results\")\n\tassert.Equal(t, \"pong\", result[\"ping_message\"], \"ping should return pong\")\n\tassert.Equal(t, true, result[\"status_online\"], \"status should be online\")\n}\n\n// TestMCPListPrompts tests MCP.ListPrompts from JavaScript\nfunc TestMCPListPrompts(t *testing.T) {\n\ttest.Prepare(t, config.Conf)\n\tdefer test.Clean()\n\n\tctx := newMCPTestContext()\n\n\tres, err := v8.Call(v8.CallOptions{}, `\n\t\tfunction test(ctx) {\n\t\t\t// List prompts from echo MCP\n\t\t\tconst result = ctx.mcp.ListPrompts(\"echo\", \"\")\n\t\t\t\n\t\t\tif (!result || !result.prompts) {\n\t\t\t\tthrow new Error(\"Expected prompts\")\n\t\t\t}\n\t\t\t\n\t\t\treturn {\n\t\t\t\tcount: result.prompts.length,\n\t\t\t\thas_test_connection: result.prompts.some(p => p.name === \"test_connection\"),\n\t\t\t\thas_test_echo: result.prompts.some(p => p.name === \"test_echo\")\n\t\t\t}\n\t\t}`, ctx)\n\n\tif err != nil {\n\t\tt.Fatalf(\"Call failed: %v\", err)\n\t}\n\n\tresult, ok := res.(map[string]interface{})\n\tif !ok {\n\t\tt.Fatalf(\"Expected map result, got %T\", res)\n\t}\n\n\tassert.Equal(t, float64(2), result[\"count\"], \"should have 2 prompts\")\n\tassert.Equal(t, true, result[\"has_test_connection\"], \"should have test_connection prompt\")\n\tassert.Equal(t, true, result[\"has_test_echo\"], \"should have test_echo prompt\")\n}\n\n// TestMCPGetPrompt tests MCP.GetPrompt from JavaScript\nfunc TestMCPGetPrompt(t *testing.T) {\n\ttest.Prepare(t, config.Conf)\n\tdefer test.Clean()\n\n\tctx := newMCPTestContext()\n\n\tres, err := v8.Call(v8.CallOptions{}, `\n\t\tfunction test(ctx) {\n\t\t\t// Get test_connection prompt\n\t\t\tconst result = ctx.mcp.GetPrompt(\"echo\", \"test_connection\", { detailed: \"true\" })\n\t\t\t\n\t\t\tif (!result || !result.messages) {\n\t\t\t\tthrow new Error(\"Expected messages\")\n\t\t\t}\n\t\t\t\n\t\t\treturn {\n\t\t\t\tcount: result.messages.length,\n\t\t\t\thas_messages: result.messages.length > 0\n\t\t\t}\n\t\t}`, ctx)\n\n\tif err != nil {\n\t\tt.Fatalf(\"Call failed: %v\", err)\n\t}\n\n\tresult, ok := res.(map[string]interface{})\n\tif !ok {\n\t\tt.Fatalf(\"Expected map result, got %T\", res)\n\t}\n\n\tassert.Equal(t, float64(1), result[\"count\"], \"should have 1 message\")\n\tassert.Equal(t, true, result[\"has_messages\"], \"should have messages\")\n}\n\n// TestMCPListSamples tests MCP.ListSamples from JavaScript\nfunc TestMCPListSamples(t *testing.T) {\n\ttest.Prepare(t, config.Conf)\n\tdefer test.Clean()\n\n\tctx := newMCPTestContext()\n\n\tres, err := v8.Call(v8.CallOptions{}, `\n\t\tfunction test(ctx) {\n\t\t\t// List samples for ping tool\n\t\t\tconst result = ctx.mcp.ListSamples(\"echo\", \"tool\", \"ping\")\n\t\t\t\n\t\t\tif (!result || !result.samples) {\n\t\t\t\tthrow new Error(\"Expected samples\")\n\t\t\t}\n\t\t\t\n\t\t\treturn {\n\t\t\t\tcount: result.samples.length,\n\t\t\t\thas_samples: result.samples.length > 0\n\t\t\t}\n\t\t}`, ctx)\n\n\tif err != nil {\n\t\tt.Fatalf(\"Call failed: %v\", err)\n\t}\n\n\tresult, ok := res.(map[string]interface{})\n\tif !ok {\n\t\tt.Fatalf(\"Expected map result, got %T\", res)\n\t}\n\n\tassert.Equal(t, float64(3), result[\"count\"], \"should have 3 samples\")\n\tassert.Equal(t, true, result[\"has_samples\"], \"should have samples\")\n}\n\n// TestMCPGetSample tests MCP.GetSample from JavaScript\nfunc TestMCPGetSample(t *testing.T) {\n\ttest.Prepare(t, config.Conf)\n\tdefer test.Clean()\n\n\tctx := newMCPTestContext()\n\n\tres, err := v8.Call(v8.CallOptions{}, `\n\t\tfunction test(ctx) {\n\t\t\t// Get first sample for ping tool\n\t\t\tconst result = ctx.mcp.GetSample(\"echo\", \"tool\", \"ping\", 0)\n\t\t\t\n\t\t\tif (!result) {\n\t\t\t\tthrow new Error(\"Expected sample\")\n\t\t\t}\n\t\t\t\n\t\t\treturn {\n\t\t\t\thas_name: !!result.name,\n\t\t\t\thas_input: !!result.input,\n\t\t\t\tname: result.name\n\t\t\t}\n\t\t}`, ctx)\n\n\tif err != nil {\n\t\tt.Fatalf(\"Call failed: %v\", err)\n\t}\n\n\tresult, ok := res.(map[string]interface{})\n\tif !ok {\n\t\tt.Fatalf(\"Expected map result, got %T\", res)\n\t}\n\n\tassert.Equal(t, true, result[\"has_name\"], \"should have name\")\n\tassert.Equal(t, true, result[\"has_input\"], \"should have input\")\n\tassert.Equal(t, \"single_ping\", result[\"name\"], \"name should be single_ping\")\n}\n\n// TestMCPJsApiWithTrace tests MCP operations with trace from JavaScript\nfunc TestMCPJsApiWithTrace(t *testing.T) {\n\ttest.Prepare(t, config.Conf)\n\tdefer test.Clean()\n\n\tctx := newMCPTestContext()\n\n\tres, err := v8.Call(v8.CallOptions{}, `\n\t\tfunction test(ctx) {\n\t\t\t// Get trace (property, not method call)\n\t\t\tconst trace = ctx.trace\n\t\t\t\n\t\t\t// Call MCP tool - returns parsed result directly\n\t\t\tconst result = ctx.mcp.CallTool(\"echo\", \"ping\", { count: 5 })\n\t\t\t\n\t\t\t// Verify trace and result exist\n\t\t\treturn {\n\t\t\t\thas_trace: !!trace,\n\t\t\t\thas_result: result !== undefined && result !== null,\n\t\t\t\tping_message: result?.message\n\t\t\t}\n\t\t}`, ctx)\n\n\tif err != nil {\n\t\tt.Fatalf(\"Call failed: %v\", err)\n\t}\n\n\tresult, ok := res.(map[string]interface{})\n\tif !ok {\n\t\tt.Fatalf(\"Expected map result, got %T\", res)\n\t}\n\n\tassert.Equal(t, true, result[\"has_trace\"], \"should have trace\")\n\tassert.Equal(t, true, result[\"has_result\"], \"should have result\")\n\tassert.Equal(t, \"pong\", result[\"ping_message\"], \"should have ping response\")\n}\n" }, { "path": "agent/context/jsapi_mcp_v8_test.go", "content": "package context_test\n\nimport (\n\tstdContext \"context\"\n\t\"testing\"\n\n\t\"github.com/stretchr/testify/assert\"\n\t\"github.com/stretchr/testify/require\"\n\tv8 \"github.com/yaoapp/gou/runtime/v8\"\n\t\"github.com/yaoapp/yao/agent/context\"\n\t\"github.com/yaoapp/yao/agent/testutils\"\n\t\"github.com/yaoapp/yao/openapi/oauth/types\"\n)\n\n// TestMCP_All_V8 tests ctx.mcp.All() with real V8 execution\nfunc TestMCP_All_V8(t *testing.T) {\n\tif testing.Short() {\n\t\tt.Skip(\"Skipping integration test in short mode\")\n\t}\n\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\tauthorized := &types.AuthorizedInfo{\n\t\tSubject: \"test-user\",\n\t\tUserID: \"test-123\",\n\t\tTenantID: \"test-tenant\",\n\t}\n\n\tctx := context.New(stdContext.Background(), authorized, \"test-chat-mcp-all\")\n\tctx.AssistantID = \"tests.agent-caller\"\n\tdefer ctx.Release()\n\n\tres, err := v8.Call(v8.CallOptions{}, `\n\t\tfunction test(ctx) {\n\t\t\ttry {\n\t\t\t\tconst results = ctx.mcp.All([\n\t\t\t\t\t{ mcp: \"echo\", tool: \"ping\", arguments: { count: 1 } },\n\t\t\t\t\t{ mcp: \"echo\", tool: \"status\", arguments: { verbose: false } },\n\t\t\t\t\t{ mcp: \"echo\", tool: \"echo\", arguments: { message: \"hello\" } }\n\t\t\t\t]);\n\t\t\t\t\n\t\t\t\treturn {\n\t\t\t\t\tsuccess: true,\n\t\t\t\t\tcount: results.length,\n\t\t\t\t\t// Each result has mcp, tool, result (parsed), error\n\t\t\t\t\tresults: results.map(r => ({\n\t\t\t\t\t\tmcp: r.mcp,\n\t\t\t\t\t\ttool: r.tool,\n\t\t\t\t\t\thas_result: r.result !== undefined && r.result !== null,\n\t\t\t\t\t\terror: r.error || \"\"\n\t\t\t\t\t}))\n\t\t\t\t};\n\t\t\t} catch (error) {\n\t\t\t\treturn { success: false, error: error.message };\n\t\t\t}\n\t\t}`, ctx)\n\n\trequire.NoError(t, err)\n\tresult, ok := res.(map[string]interface{})\n\trequire.True(t, ok, \"Result should be a map\")\n\n\tsuccess, _ := result[\"success\"].(bool)\n\tif !success {\n\t\tt.Fatalf(\"Test failed: %v\", result[\"error\"])\n\t}\n\n\t// Should have 3 results - handle different integer types\n\tvar count int\n\tswitch v := result[\"count\"].(type) {\n\tcase int:\n\t\tcount = v\n\tcase int32:\n\t\tcount = int(v)\n\tcase int64:\n\t\tcount = int(v)\n\tcase float64:\n\t\tcount = int(v)\n\tdefault:\n\t\tt.Logf(\"Unexpected count type: %T, value: %v\", result[\"count\"], result[\"count\"])\n\t}\n\tassert.Equal(t, 3, count, \"Should have 3 results\")\n\n\t// Check each result\n\tresults, ok := result[\"results\"].([]interface{})\n\trequire.True(t, ok, \"Results should be an array\")\n\trequire.Len(t, results, 3)\n\n\tfor i, r := range results {\n\t\tresMap, ok := r.(map[string]interface{})\n\t\trequire.True(t, ok, \"Result %d should be a map\", i)\n\n\t\thasResult, _ := resMap[\"has_result\"].(bool)\n\t\tassert.True(t, hasResult, \"Result %d should have parsed result\", i)\n\n\t\terrorStr, _ := resMap[\"error\"].(string)\n\t\tassert.Empty(t, errorStr, \"Result %d should not have error\", i)\n\t}\n}\n\n// TestMCP_All_WithError_V8 tests ctx.mcp.All() with some failing requests\nfunc TestMCP_All_WithError_V8(t *testing.T) {\n\tif testing.Short() {\n\t\tt.Skip(\"Skipping integration test in short mode\")\n\t}\n\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\tauthorized := &types.AuthorizedInfo{\n\t\tSubject: \"test-user\",\n\t\tUserID: \"test-123\",\n\t\tTenantID: \"test-tenant\",\n\t}\n\n\tctx := context.New(stdContext.Background(), authorized, \"test-chat-mcp-all-error\")\n\tctx.AssistantID = \"tests.agent-caller\"\n\tdefer ctx.Release()\n\n\tres, err := v8.Call(v8.CallOptions{}, `\n\t\tfunction test(ctx) {\n\t\t\ttry {\n\t\t\t\tconst results = ctx.mcp.All([\n\t\t\t\t\t{ mcp: \"echo\", tool: \"ping\", arguments: { count: 1 } },\n\t\t\t\t\t{ mcp: \"nonexistent-mcp\", tool: \"some-tool\", arguments: {} },\n\t\t\t\t\t{ mcp: \"echo\", tool: \"status\", arguments: {} }\n\t\t\t\t]);\n\t\t\t\t\n\t\t\t\treturn {\n\t\t\t\t\tsuccess: true,\n\t\t\t\t\tcount: results.length,\n\t\t\t\t\tresults: results.map(r => ({\n\t\t\t\t\t\tmcp: r.mcp,\n\t\t\t\t\t\ttool: r.tool,\n\t\t\t\t\t\thas_result: r.result !== undefined && r.result !== null,\n\t\t\t\t\t\thas_error: r.error !== undefined && r.error !== \"\" && r.error !== null\n\t\t\t\t\t}))\n\t\t\t\t};\n\t\t\t} catch (error) {\n\t\t\t\treturn { success: false, error: error.message };\n\t\t\t}\n\t\t}`, ctx)\n\n\trequire.NoError(t, err)\n\tresult, ok := res.(map[string]interface{})\n\trequire.True(t, ok, \"Result should be a map\")\n\n\tsuccess, _ := result[\"success\"].(bool)\n\tif !success {\n\t\tt.Fatalf(\"Test failed: %v\", result[\"error\"])\n\t}\n\n\t// Should have 3 results - handle different integer types\n\tvar count int\n\tswitch v := result[\"count\"].(type) {\n\tcase int:\n\t\tcount = v\n\tcase int32:\n\t\tcount = int(v)\n\tcase int64:\n\t\tcount = int(v)\n\tcase float64:\n\t\tcount = int(v)\n\t}\n\tassert.Equal(t, 3, count, \"Should have 3 results\")\n\n\t// Check results\n\tresults, ok := result[\"results\"].([]interface{})\n\trequire.True(t, ok, \"Results should be an array\")\n\n\t// First result (ping) should succeed\n\tr0, _ := results[0].(map[string]interface{})\n\tassert.True(t, r0[\"has_result\"].(bool), \"Ping should have result\")\n\tassert.False(t, r0[\"has_error\"].(bool), \"Ping should not have error\")\n\n\t// Second result (nonexistent) should fail\n\tr1, _ := results[1].(map[string]interface{})\n\tassert.False(t, r1[\"has_result\"].(bool), \"Nonexistent should not have result\")\n\tassert.True(t, r1[\"has_error\"].(bool), \"Nonexistent should have error\")\n\n\t// Third result (status) should succeed\n\tr2, _ := results[2].(map[string]interface{})\n\tassert.True(t, r2[\"has_result\"].(bool), \"Status should have result\")\n\tassert.False(t, r2[\"has_error\"].(bool), \"Status should not have error\")\n}\n\n// TestMCP_Any_V8 tests ctx.mcp.Any() with real V8 execution\nfunc TestMCP_Any_V8(t *testing.T) {\n\tif testing.Short() {\n\t\tt.Skip(\"Skipping integration test in short mode\")\n\t}\n\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\tauthorized := &types.AuthorizedInfo{\n\t\tSubject: \"test-user\",\n\t\tUserID: \"test-123\",\n\t\tTenantID: \"test-tenant\",\n\t}\n\n\tctx := context.New(stdContext.Background(), authorized, \"test-chat-mcp-any\")\n\tctx.AssistantID = \"tests.agent-caller\"\n\tdefer ctx.Release()\n\n\tres, err := v8.Call(v8.CallOptions{}, `\n\t\tfunction test(ctx) {\n\t\t\ttry {\n\t\t\t\tconst results = ctx.mcp.Any([\n\t\t\t\t\t{ mcp: \"echo\", tool: \"ping\", arguments: { count: 1 } },\n\t\t\t\t\t{ mcp: \"echo\", tool: \"status\", arguments: {} }\n\t\t\t\t]);\n\t\t\t\t\n\t\t\t\t// Find success results (has result field, no error)\n\t\t\t\tconst successResults = results.filter(r => r && r.result && !r.error);\n\t\t\t\t\n\t\t\t\treturn {\n\t\t\t\t\tsuccess: true,\n\t\t\t\t\ttotal_count: results.length,\n\t\t\t\t\tsuccess_count: successResults.length,\n\t\t\t\t\thas_at_least_one_success: successResults.length >= 1\n\t\t\t\t};\n\t\t\t} catch (error) {\n\t\t\t\treturn { success: false, error: error.message };\n\t\t\t}\n\t\t}`, ctx)\n\n\trequire.NoError(t, err)\n\tresult, ok := res.(map[string]interface{})\n\trequire.True(t, ok, \"Result should be a map\")\n\n\tsuccess, _ := result[\"success\"].(bool)\n\tif !success {\n\t\tt.Fatalf(\"Test failed: %v\", result[\"error\"])\n\t}\n\n\thasAtLeastOne, _ := result[\"has_at_least_one_success\"].(bool)\n\tassert.True(t, hasAtLeastOne, \"Should have at least one successful result\")\n}\n\n// TestMCP_Any_AllFail_V8 tests ctx.mcp.Any() when all requests fail\nfunc TestMCP_Any_AllFail_V8(t *testing.T) {\n\tif testing.Short() {\n\t\tt.Skip(\"Skipping integration test in short mode\")\n\t}\n\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\tauthorized := &types.AuthorizedInfo{\n\t\tSubject: \"test-user\",\n\t\tUserID: \"test-123\",\n\t\tTenantID: \"test-tenant\",\n\t}\n\n\tctx := context.New(stdContext.Background(), authorized, \"test-chat-mcp-any-fail\")\n\tctx.AssistantID = \"tests.agent-caller\"\n\tdefer ctx.Release()\n\n\tres, err := v8.Call(v8.CallOptions{}, `\n\t\tfunction test(ctx) {\n\t\t\ttry {\n\t\t\t\tconst results = ctx.mcp.Any([\n\t\t\t\t\t{ mcp: \"nonexistent-1\", tool: \"tool1\", arguments: {} },\n\t\t\t\t\t{ mcp: \"nonexistent-2\", tool: \"tool2\", arguments: {} }\n\t\t\t\t]);\n\t\t\t\t\n\t\t\t\t// All should fail\n\t\t\t\tconst failedResults = results.filter(r => r && r.error);\n\t\t\t\t\n\t\t\t\treturn {\n\t\t\t\t\tsuccess: true,\n\t\t\t\t\ttotal_count: results.length,\n\t\t\t\t\tfailed_count: failedResults.length,\n\t\t\t\t\tall_failed: failedResults.length === results.length\n\t\t\t\t};\n\t\t\t} catch (error) {\n\t\t\t\treturn { success: false, error: error.message };\n\t\t\t}\n\t\t}`, ctx)\n\n\trequire.NoError(t, err)\n\tresult, ok := res.(map[string]interface{})\n\trequire.True(t, ok, \"Result should be a map\")\n\n\tsuccess, _ := result[\"success\"].(bool)\n\tif !success {\n\t\tt.Fatalf(\"Test failed: %v\", result[\"error\"])\n\t}\n\n\tallFailed, _ := result[\"all_failed\"].(bool)\n\tassert.True(t, allFailed, \"All requests should fail\")\n}\n\n// TestMCP_Race_V8 tests ctx.mcp.Race() with real V8 execution\nfunc TestMCP_Race_V8(t *testing.T) {\n\tif testing.Short() {\n\t\tt.Skip(\"Skipping integration test in short mode\")\n\t}\n\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\tauthorized := &types.AuthorizedInfo{\n\t\tSubject: \"test-user\",\n\t\tUserID: \"test-123\",\n\t\tTenantID: \"test-tenant\",\n\t}\n\n\tctx := context.New(stdContext.Background(), authorized, \"test-chat-mcp-race\")\n\tctx.AssistantID = \"tests.agent-caller\"\n\tdefer ctx.Release()\n\n\tres, err := v8.Call(v8.CallOptions{}, `\n\t\tfunction test(ctx) {\n\t\t\ttry {\n\t\t\t\tconst results = ctx.mcp.Race([\n\t\t\t\t\t{ mcp: \"echo\", tool: \"ping\", arguments: { count: 1 } },\n\t\t\t\t\t{ mcp: \"echo\", tool: \"status\", arguments: {} }\n\t\t\t\t]);\n\t\t\t\t\n\t\t\t\t// Find completed results (could be success or error)\n\t\t\t\tconst completedResults = results.filter(r => r !== undefined && r !== null);\n\t\t\t\t\n\t\t\t\treturn {\n\t\t\t\t\tsuccess: true,\n\t\t\t\t\ttotal_count: results.length,\n\t\t\t\t\tcompleted_count: completedResults.length,\n\t\t\t\t\thas_at_least_one_completed: completedResults.length >= 1\n\t\t\t\t};\n\t\t\t} catch (error) {\n\t\t\t\treturn { success: false, error: error.message };\n\t\t\t}\n\t\t}`, ctx)\n\n\trequire.NoError(t, err)\n\tresult, ok := res.(map[string]interface{})\n\trequire.True(t, ok, \"Result should be a map\")\n\n\tsuccess, _ := result[\"success\"].(bool)\n\tif !success {\n\t\tt.Fatalf(\"Test failed: %v\", result[\"error\"])\n\t}\n\n\thasAtLeastOne, _ := result[\"has_at_least_one_completed\"].(bool)\n\tassert.True(t, hasAtLeastOne, \"Should have at least one completed result\")\n}\n\n// TestMCP_All_ResultContent_V8 tests that the result contains parsed content directly\nfunc TestMCP_All_ResultContent_V8(t *testing.T) {\n\tif testing.Short() {\n\t\tt.Skip(\"Skipping integration test in short mode\")\n\t}\n\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\tauthorized := &types.AuthorizedInfo{\n\t\tSubject: \"test-user\",\n\t\tUserID: \"test-123\",\n\t\tTenantID: \"test-tenant\",\n\t}\n\n\tctx := context.New(stdContext.Background(), authorized, \"test-chat-mcp-content\")\n\tctx.AssistantID = \"tests.agent-caller\"\n\tdefer ctx.Release()\n\n\tres, err := v8.Call(v8.CallOptions{}, `\n\t\tfunction test(ctx) {\n\t\t\ttry {\n\t\t\t\tconst results = ctx.mcp.All([\n\t\t\t\t\t{ mcp: \"echo\", tool: \"echo\", arguments: { message: \"hello world\", uppercase: true } }\n\t\t\t\t]);\n\t\t\t\t\n\t\t\t\tif (results.length !== 1) {\n\t\t\t\t\treturn { success: false, error: \"Expected 1 result\" };\n\t\t\t\t}\n\t\t\t\t\n\t\t\t\tconst r = results[0];\n\t\t\t\tif (r.error) {\n\t\t\t\t\treturn { success: false, error: \"Tool call failed: \" + r.error };\n\t\t\t\t}\n\t\t\t\t\n\t\t\t\t// Result should contain parsed data directly\n\t\t\t\tconst data = r.result;\n\t\t\t\tif (!data) {\n\t\t\t\t\treturn { success: false, error: \"Result should have parsed data\" };\n\t\t\t\t}\n\t\t\t\t\n\t\t\t\treturn {\n\t\t\t\t\tsuccess: true,\n\t\t\t\t\techo_message: data.echo,\n\t\t\t\t\tuppercase_flag: data.uppercase,\n\t\t\t\t\toriginal_length: data.length\n\t\t\t\t};\n\t\t\t} catch (error) {\n\t\t\t\treturn { success: false, error: error.message };\n\t\t\t}\n\t\t}`, ctx)\n\n\trequire.NoError(t, err)\n\tresult, ok := res.(map[string]interface{})\n\trequire.True(t, ok, \"Result should be a map\")\n\n\tsuccess, _ := result[\"success\"].(bool)\n\tif !success {\n\t\tt.Fatalf(\"Test failed: %v\", result[\"error\"])\n\t}\n\n\techoMessage, _ := result[\"echo_message\"].(string)\n\tassert.Equal(t, \"HELLO WORLD\", echoMessage, \"Echo message should be uppercase\")\n\n\tuppercaseFlag, _ := result[\"uppercase_flag\"].(bool)\n\tassert.True(t, uppercaseFlag, \"Uppercase flag should be true\")\n\n\t// Handle different integer types from V8\n\tvar originalLength int\n\tswitch v := result[\"original_length\"].(type) {\n\tcase int:\n\t\toriginalLength = v\n\tcase int32:\n\t\toriginalLength = int(v)\n\tcase int64:\n\t\toriginalLength = int(v)\n\tcase float64:\n\t\toriginalLength = int(v)\n\t}\n\tassert.Equal(t, 11, originalLength, \"Original message length should be 11\")\n}\n\n// TestMCP_All_MultipleTools_V8 tests All with multiple tools and verifies parsed results\nfunc TestMCP_All_MultipleTools_V8(t *testing.T) {\n\tif testing.Short() {\n\t\tt.Skip(\"Skipping integration test in short mode\")\n\t}\n\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\tauthorized := &types.AuthorizedInfo{\n\t\tSubject: \"test-user\",\n\t\tUserID: \"test-123\",\n\t\tTenantID: \"test-tenant\",\n\t}\n\n\tctx := context.New(stdContext.Background(), authorized, \"test-chat-mcp-multi\")\n\tctx.AssistantID = \"tests.agent-caller\"\n\tdefer ctx.Release()\n\n\tres, err := v8.Call(v8.CallOptions{}, `\n\t\tfunction test(ctx) {\n\t\t\ttry {\n\t\t\t\tconst results = ctx.mcp.All([\n\t\t\t\t\t{ mcp: \"echo\", tool: \"ping\", arguments: { count: 5 } },\n\t\t\t\t\t{ mcp: \"echo\", tool: \"echo\", arguments: { message: \"test\", uppercase: false } }\n\t\t\t\t]);\n\t\t\t\t\n\t\t\t\tif (results.length !== 2) {\n\t\t\t\t\treturn { success: false, error: \"Expected 2 results\" };\n\t\t\t\t}\n\t\t\t\t\n\t\t\t\t// Access parsed results directly\n\t\t\t\tconst ping = results[0];\n\t\t\t\tconst echo = results[1];\n\t\t\t\t\n\t\t\t\treturn {\n\t\t\t\t\tsuccess: true,\n\t\t\t\t\tping_message: ping.result?.message,\n\t\t\t\t\techo_message: echo.result?.echo\n\t\t\t\t};\n\t\t\t} catch (error) {\n\t\t\t\treturn { success: false, error: error.message };\n\t\t\t}\n\t\t}`, ctx)\n\n\trequire.NoError(t, err)\n\tresult, ok := res.(map[string]interface{})\n\trequire.True(t, ok, \"Result should be a map\")\n\n\tsuccess, _ := result[\"success\"].(bool)\n\tif !success {\n\t\tt.Fatalf(\"Test failed: %v\", result[\"error\"])\n\t}\n\n\tpingMessage, _ := result[\"ping_message\"].(string)\n\tassert.Equal(t, \"pong\", pingMessage, \"Ping should return pong\")\n\n\techoMessage, _ := result[\"echo_message\"].(string)\n\tassert.Equal(t, \"test\", echoMessage, \"Echo should return the message\")\n}\n\n// TestMCP_CallTool_ParsedResult_V8 tests that ctx.mcp.CallTool() returns parsed result directly\nfunc TestMCP_CallTool_ParsedResult_V8(t *testing.T) {\n\tif testing.Short() {\n\t\tt.Skip(\"Skipping integration test in short mode\")\n\t}\n\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\tauthorized := &types.AuthorizedInfo{\n\t\tSubject: \"test-user\",\n\t\tUserID: \"test-123\",\n\t\tTenantID: \"test-tenant\",\n\t}\n\n\tctx := context.New(stdContext.Background(), authorized, \"test-chat-mcp-calltool-parsed\")\n\tctx.AssistantID = \"tests.agent-caller\"\n\tdefer ctx.Release()\n\n\tres, err := v8.Call(v8.CallOptions{}, `\n\t\tfunction test(ctx) {\n\t\t\ttry {\n\t\t\t\t// Test CallTool returns parsed result directly\n\t\t\t\tconst result = ctx.mcp.CallTool(\"echo\", \"echo\", { \n\t\t\t\t\tmessage: \"test message\", \n\t\t\t\t\tuppercase: true \n\t\t\t\t});\n\t\t\t\t\n\t\t\t\t// Result should be the parsed data directly\n\t\t\t\tif (result === undefined || result === null) {\n\t\t\t\t\treturn { success: false, error: \"Result should not be null\" };\n\t\t\t\t}\n\t\t\t\t\n\t\t\t\treturn {\n\t\t\t\t\tsuccess: true,\n\t\t\t\t\techo_message: result.echo,\n\t\t\t\t\toriginal_length: result.length\n\t\t\t\t};\n\t\t\t} catch (error) {\n\t\t\t\treturn { success: false, error: error.message };\n\t\t\t}\n\t\t}`, ctx)\n\n\trequire.NoError(t, err)\n\tresult, ok := res.(map[string]interface{})\n\trequire.True(t, ok, \"Result should be a map\")\n\n\tsuccess, _ := result[\"success\"].(bool)\n\tif !success {\n\t\tt.Fatalf(\"Test failed: %v\", result[\"error\"])\n\t}\n\n\techoMessage, _ := result[\"echo_message\"].(string)\n\tassert.Equal(t, \"TEST MESSAGE\", echoMessage, \"Echo message should be uppercase\")\n\n\t// Handle different integer types from V8\n\tvar length int\n\tswitch v := result[\"original_length\"].(type) {\n\tcase int:\n\t\tlength = v\n\tcase int32:\n\t\tlength = int(v)\n\tcase int64:\n\t\tlength = int(v)\n\tcase float64:\n\t\tlength = int(v)\n\t}\n\tassert.Equal(t, 12, length, \"Original message length should be 12\")\n}\n\n// TestMCP_CallToolsParallel_ParsedResult_V8 tests that ctx.mcp.CallToolsParallel() returns parsed results directly\nfunc TestMCP_CallToolsParallel_ParsedResult_V8(t *testing.T) {\n\tif testing.Short() {\n\t\tt.Skip(\"Skipping integration test in short mode\")\n\t}\n\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\tauthorized := &types.AuthorizedInfo{\n\t\tSubject: \"test-user\",\n\t\tUserID: \"test-123\",\n\t\tTenantID: \"test-tenant\",\n\t}\n\n\tctx := context.New(stdContext.Background(), authorized, \"test-chat-mcp-calltools-parsed\")\n\tctx.AssistantID = \"tests.agent-caller\"\n\tdefer ctx.Release()\n\n\tres, err := v8.Call(v8.CallOptions{}, `\n\t\tfunction test(ctx) {\n\t\t\ttry {\n\t\t\t\t// Test CallToolsParallel returns parsed results directly as array\n\t\t\t\tconst results = ctx.mcp.CallToolsParallel(\"echo\", [\n\t\t\t\t\t{ name: \"ping\", arguments: { count: 2 } },\n\t\t\t\t\t{ name: \"echo\", arguments: { message: \"hello\", uppercase: false } }\n\t\t\t\t]);\n\t\t\t\t\n\t\t\t\tif (!Array.isArray(results)) {\n\t\t\t\t\treturn { success: false, error: \"Results should be an array\" };\n\t\t\t\t}\n\t\t\t\tif (results.length !== 2) {\n\t\t\t\t\treturn { success: false, error: \"Expected 2 results, got \" + results.length };\n\t\t\t\t}\n\t\t\t\t\n\t\t\t\t// Each result is the parsed data directly\n\t\t\t\tconst pingResult = results[0];\n\t\t\t\tconst echoResult = results[1];\n\t\t\t\t\n\t\t\t\treturn {\n\t\t\t\t\tsuccess: true,\n\t\t\t\t\tping_message: pingResult?.message,\n\t\t\t\t\techo_message: echoResult?.echo\n\t\t\t\t};\n\t\t\t} catch (error) {\n\t\t\t\treturn { success: false, error: error.message };\n\t\t\t}\n\t\t}`, ctx)\n\n\trequire.NoError(t, err)\n\tresult, ok := res.(map[string]interface{})\n\trequire.True(t, ok, \"Result should be a map\")\n\n\tsuccess, _ := result[\"success\"].(bool)\n\tif !success {\n\t\tt.Fatalf(\"Test failed: %v\", result[\"error\"])\n\t}\n\n\tpingMessage, _ := result[\"ping_message\"].(string)\n\tassert.Equal(t, \"pong\", pingMessage, \"Ping result should have message='pong'\")\n\n\techoMessage, _ := result[\"echo_message\"].(string)\n\tassert.Equal(t, \"hello\", echoMessage, \"Echo message should be preserved (no uppercase)\")\n}\n" }, { "path": "agent/context/jsapi_memory_test.go", "content": "package context_test\n\nimport (\n\tstdContext \"context\"\n\t\"testing\"\n\n\t\"github.com/stretchr/testify/assert\"\n\t\"github.com/stretchr/testify/require\"\n\tv8 \"github.com/yaoapp/gou/runtime/v8\"\n\t\"github.com/yaoapp/yao/agent/context\"\n\t\"github.com/yaoapp/yao/agent/memory\"\n\t\"github.com/yaoapp/yao/config\"\n\t\"github.com/yaoapp/yao/openapi/oauth/types\"\n\t\"github.com/yaoapp/yao/test\"\n)\n\nfunc TestMemoryUserNamespace(t *testing.T) {\n\ttest.Prepare(t, config.Conf)\n\tdefer test.Clean()\n\n\tmem, err := memory.New(nil, \"user1\", \"team1\", \"chat1\", \"ctx1\")\n\trequire.NoError(t, err)\n\n\tctx := &context.Context{\n\t\tChatID: \"test-chat-id\",\n\t\tAssistantID: \"test-assistant-id\",\n\t\tLocale: \"en\",\n\t\tContext: stdContext.Background(),\n\t\tMemory: mem,\n\t}\n\n\tres, err := v8.Call(v8.CallOptions{}, `\n\t\tfunction test(ctx) {\n\t\t\ttry {\n\t\t\t\t// Set values in user namespace\n\t\t\t\tctx.memory.user.Set(\"name\", \"John\");\n\t\t\t\tctx.memory.user.Set(\"age\", 30);\n\t\t\t\tctx.memory.user.Set(\"active\", true);\n\t\t\t\t\n\t\t\t\t// Get values back\n\t\t\t\tconst name = ctx.memory.user.Get(\"name\");\n\t\t\t\tconst age = ctx.memory.user.Get(\"age\");\n\t\t\t\tconst active = ctx.memory.user.Get(\"active\");\n\t\t\t\t\n\t\t\t\t// Verify\n\t\t\t\tif (name !== \"John\") throw new Error(\"Name mismatch\");\n\t\t\t\tif (age !== 30) throw new Error(\"Age mismatch\");\n\t\t\t\tif (active !== true) throw new Error(\"Active mismatch\");\n\t\t\t\t\n\t\t\t\treturn { \n\t\t\t\t\tsuccess: true,\n\t\t\t\t\tname: name,\n\t\t\t\t\tage: age,\n\t\t\t\t\tactive: active\n\t\t\t\t};\n\t\t\t} catch (error) {\n\t\t\t\treturn { success: false, error: error.message };\n\t\t\t}\n\t\t}`, ctx)\n\n\trequire.NoError(t, err)\n\tresult := res.(map[string]interface{})\n\tassert.True(t, result[\"success\"].(bool))\n\tassert.Equal(t, \"John\", result[\"name\"])\n\tassert.Equal(t, float64(30), result[\"age\"])\n\tassert.Equal(t, true, result[\"active\"])\n}\n\nfunc TestMemoryTeamNamespace(t *testing.T) {\n\ttest.Prepare(t, config.Conf)\n\tdefer test.Clean()\n\n\tmem, err := memory.New(nil, \"user1\", \"team1\", \"chat1\", \"ctx1\")\n\trequire.NoError(t, err)\n\n\tctx := &context.Context{\n\t\tChatID: \"test-chat-id\",\n\t\tAssistantID: \"test-assistant-id\",\n\t\tLocale: \"en\",\n\t\tContext: stdContext.Background(),\n\t\tMemory: mem,\n\t}\n\n\tres, err := v8.Call(v8.CallOptions{}, `\n\t\tfunction test(ctx) {\n\t\t\ttry {\n\t\t\t\t// Set team-wide settings\n\t\t\t\tctx.memory.team.Set(\"settings\", { theme: \"dark\", language: \"en\" });\n\t\t\t\t\n\t\t\t\t// Get back\n\t\t\t\tconst settings = ctx.memory.team.Get(\"settings\");\n\t\t\t\t\n\t\t\t\tif (settings.theme !== \"dark\") throw new Error(\"Theme mismatch\");\n\t\t\t\tif (settings.language !== \"en\") throw new Error(\"Language mismatch\");\n\t\t\t\t\n\t\t\t\treturn { \n\t\t\t\t\tsuccess: true,\n\t\t\t\t\tsettings: settings\n\t\t\t\t};\n\t\t\t} catch (error) {\n\t\t\t\treturn { success: false, error: error.message };\n\t\t\t}\n\t\t}`, ctx)\n\n\trequire.NoError(t, err)\n\tresult := res.(map[string]interface{})\n\tassert.True(t, result[\"success\"].(bool))\n}\n\nfunc TestMemoryChatNamespace(t *testing.T) {\n\ttest.Prepare(t, config.Conf)\n\tdefer test.Clean()\n\n\tmem, err := memory.New(nil, \"user1\", \"team1\", \"chat1\", \"ctx1\")\n\trequire.NoError(t, err)\n\n\tctx := &context.Context{\n\t\tChatID: \"test-chat-id\",\n\t\tAssistantID: \"test-assistant-id\",\n\t\tLocale: \"en\",\n\t\tContext: stdContext.Background(),\n\t\tMemory: mem,\n\t}\n\n\tres, err := v8.Call(v8.CallOptions{}, `\n\t\tfunction test(ctx) {\n\t\t\ttry {\n\t\t\t\t// Set chat context\n\t\t\t\tctx.memory.chat.Set(\"topic\", \"AI Discussion\");\n\t\t\t\tctx.memory.chat.Set(\"participants\", [\"Alice\", \"Bob\"]);\n\t\t\t\t\n\t\t\t\t// Get back\n\t\t\t\tconst topic = ctx.memory.chat.Get(\"topic\");\n\t\t\t\tconst participants = ctx.memory.chat.Get(\"participants\");\n\t\t\t\t\n\t\t\t\tif (topic !== \"AI Discussion\") throw new Error(\"Topic mismatch\");\n\t\t\t\tif (participants.length !== 2) throw new Error(\"Participants mismatch\");\n\t\t\t\t\n\t\t\t\treturn { \n\t\t\t\t\tsuccess: true,\n\t\t\t\t\ttopic: topic,\n\t\t\t\t\tparticipants: participants\n\t\t\t\t};\n\t\t\t} catch (error) {\n\t\t\t\treturn { success: false, error: error.message };\n\t\t\t}\n\t\t}`, ctx)\n\n\trequire.NoError(t, err)\n\tresult := res.(map[string]interface{})\n\tassert.True(t, result[\"success\"].(bool))\n\tassert.Equal(t, \"AI Discussion\", result[\"topic\"])\n}\n\nfunc TestMemoryContextNamespace(t *testing.T) {\n\ttest.Prepare(t, config.Conf)\n\tdefer test.Clean()\n\n\tmem, err := memory.New(nil, \"user1\", \"team1\", \"chat1\", \"ctx1\")\n\trequire.NoError(t, err)\n\n\tctx := &context.Context{\n\t\tChatID: \"test-chat-id\",\n\t\tAssistantID: \"test-assistant-id\",\n\t\tLocale: \"en\",\n\t\tContext: stdContext.Background(),\n\t\tMemory: mem,\n\t}\n\n\tres, err := v8.Call(v8.CallOptions{}, `\n\t\tfunction test(ctx) {\n\t\t\ttry {\n\t\t\t\t// Set temporary context data\n\t\t\t\tctx.memory.context.Set(\"temp_result\", { step: 1, data: \"processing\" });\n\t\t\t\t\n\t\t\t\t// Get back\n\t\t\t\tconst result = ctx.memory.context.Get(\"temp_result\");\n\t\t\t\t\n\t\t\t\tif (result.step !== 1) throw new Error(\"Step mismatch\");\n\t\t\t\tif (result.data !== \"processing\") throw new Error(\"Data mismatch\");\n\t\t\t\t\n\t\t\t\treturn { \n\t\t\t\t\tsuccess: true,\n\t\t\t\t\tresult: result\n\t\t\t\t};\n\t\t\t} catch (error) {\n\t\t\t\treturn { success: false, error: error.message };\n\t\t\t}\n\t\t}`, ctx)\n\n\trequire.NoError(t, err)\n\tresult := res.(map[string]interface{})\n\tassert.True(t, result[\"success\"].(bool))\n}\n\nfunc TestMemoryHasAndDel(t *testing.T) {\n\ttest.Prepare(t, config.Conf)\n\tdefer test.Clean()\n\n\tmem, err := memory.New(nil, \"user1\", \"team1\", \"chat1\", \"ctx1\")\n\trequire.NoError(t, err)\n\n\tctx := &context.Context{\n\t\tChatID: \"test-chat-id\",\n\t\tAssistantID: \"test-assistant-id\",\n\t\tLocale: \"en\",\n\t\tContext: stdContext.Background(),\n\t\tMemory: mem,\n\t}\n\n\tres, err := v8.Call(v8.CallOptions{}, `\n\t\tfunction test(ctx) {\n\t\t\ttry {\n\t\t\t\t// Set a value\n\t\t\t\tctx.memory.user.Set(\"key\", \"value\");\n\t\t\t\t\n\t\t\t\t// Check Has\n\t\t\t\tconst hasBefore = ctx.memory.user.Has(\"key\");\n\t\t\t\tif (!hasBefore) throw new Error(\"Should have key before delete\");\n\t\t\t\t\n\t\t\t\t// Delete\n\t\t\t\tctx.memory.user.Del(\"key\");\n\t\t\t\t\n\t\t\t\t// Check Has again\n\t\t\t\tconst hasAfter = ctx.memory.user.Has(\"key\");\n\t\t\t\tif (hasAfter) throw new Error(\"Should not have key after delete\");\n\t\t\t\t\n\t\t\t\treturn { \n\t\t\t\t\tsuccess: true,\n\t\t\t\t\thasBefore: hasBefore,\n\t\t\t\t\thasAfter: hasAfter\n\t\t\t\t};\n\t\t\t} catch (error) {\n\t\t\t\treturn { success: false, error: error.message };\n\t\t\t}\n\t\t}`, ctx)\n\n\trequire.NoError(t, err)\n\tresult := res.(map[string]interface{})\n\tassert.True(t, result[\"success\"].(bool))\n\tassert.True(t, result[\"hasBefore\"].(bool))\n\tassert.False(t, result[\"hasAfter\"].(bool))\n}\n\nfunc TestMemoryIncrDecr(t *testing.T) {\n\ttest.Prepare(t, config.Conf)\n\tdefer test.Clean()\n\n\tmem, err := memory.New(nil, \"user1\", \"team1\", \"chat1\", \"ctx1\")\n\trequire.NoError(t, err)\n\n\tctx := &context.Context{\n\t\tChatID: \"test-chat-id\",\n\t\tAssistantID: \"test-assistant-id\",\n\t\tLocale: \"en\",\n\t\tContext: stdContext.Background(),\n\t\tMemory: mem,\n\t}\n\n\tres, err := v8.Call(v8.CallOptions{}, `\n\t\tfunction test(ctx) {\n\t\t\ttry {\n\t\t\t\t// Incr on non-existent key\n\t\t\t\tconst v1 = ctx.memory.user.Incr(\"counter\");\n\t\t\t\tif (v1 !== 1) throw new Error(\"First incr should be 1, got \" + v1);\n\t\t\t\t\n\t\t\t\t// Incr with delta\n\t\t\t\tconst v2 = ctx.memory.user.Incr(\"counter\", 5);\n\t\t\t\tif (v2 !== 6) throw new Error(\"Second incr should be 6, got \" + v2);\n\t\t\t\t\n\t\t\t\t// Decr\n\t\t\t\tconst v3 = ctx.memory.user.Decr(\"counter\", 2);\n\t\t\t\tif (v3 !== 4) throw new Error(\"Decr should be 4, got \" + v3);\n\t\t\t\t\n\t\t\t\treturn { \n\t\t\t\t\tsuccess: true,\n\t\t\t\t\tv1: v1,\n\t\t\t\t\tv2: v2,\n\t\t\t\t\tv3: v3\n\t\t\t\t};\n\t\t\t} catch (error) {\n\t\t\t\treturn { success: false, error: error.message };\n\t\t\t}\n\t\t}`, ctx)\n\n\trequire.NoError(t, err)\n\tresult := res.(map[string]interface{})\n\tassert.True(t, result[\"success\"].(bool))\n\tassert.Equal(t, float64(1), result[\"v1\"])\n\tassert.Equal(t, float64(6), result[\"v2\"])\n\tassert.Equal(t, float64(4), result[\"v3\"])\n}\n\nfunc TestMemoryKeysAndLen(t *testing.T) {\n\ttest.Prepare(t, config.Conf)\n\tdefer test.Clean()\n\n\t// Use unique IDs to avoid data pollution from other tests\n\tmem, err := memory.New(nil, \"user-keys-len\", \"team-keys-len\", \"chat-keys-len\", \"ctx-keys-len\")\n\trequire.NoError(t, err)\n\n\tctx := &context.Context{\n\t\tChatID: \"test-chat-id\",\n\t\tAssistantID: \"test-assistant-id\",\n\t\tLocale: \"en\",\n\t\tContext: stdContext.Background(),\n\t\tMemory: mem,\n\t}\n\n\tres, err := v8.Call(v8.CallOptions{}, `\n\t\tfunction test(ctx) {\n\t\t\ttry {\n\t\t\t\t// Set multiple values\n\t\t\t\tctx.memory.user.Set(\"a\", 1);\n\t\t\t\tctx.memory.user.Set(\"b\", 2);\n\t\t\t\tctx.memory.user.Set(\"c\", 3);\n\t\t\t\t\n\t\t\t\t// Get keys\n\t\t\t\tconst keys = ctx.memory.user.Keys();\n\t\t\t\tif (keys.length !== 3) throw new Error(\"Should have 3 keys, got \" + keys.length);\n\t\t\t\t\n\t\t\t\t// Get len\n\t\t\t\tconst len = ctx.memory.user.Len();\n\t\t\t\tif (len !== 3) throw new Error(\"Len should be 3, got \" + len);\n\t\t\t\t\n\t\t\t\treturn { \n\t\t\t\t\tsuccess: true,\n\t\t\t\t\tkeys: keys,\n\t\t\t\t\tlen: len\n\t\t\t\t};\n\t\t\t} catch (error) {\n\t\t\t\treturn { success: false, error: error.message };\n\t\t\t}\n\t\t}`, ctx)\n\n\trequire.NoError(t, err)\n\tresult := res.(map[string]interface{})\n\tif !result[\"success\"].(bool) {\n\t\tt.Fatalf(\"Test failed: %v\", result[\"error\"])\n\t}\n\tassert.Equal(t, float64(3), result[\"len\"])\n}\n\nfunc TestMemoryClear(t *testing.T) {\n\ttest.Prepare(t, config.Conf)\n\tdefer test.Clean()\n\n\tmem, err := memory.New(nil, \"user1\", \"team1\", \"chat1\", \"ctx1\")\n\trequire.NoError(t, err)\n\n\tctx := &context.Context{\n\t\tChatID: \"test-chat-id\",\n\t\tAssistantID: \"test-assistant-id\",\n\t\tLocale: \"en\",\n\t\tContext: stdContext.Background(),\n\t\tMemory: mem,\n\t}\n\n\tres, err := v8.Call(v8.CallOptions{}, `\n\t\tfunction test(ctx) {\n\t\t\ttry {\n\t\t\t\t// Set values\n\t\t\t\tctx.memory.user.Set(\"a\", 1);\n\t\t\t\tctx.memory.user.Set(\"b\", 2);\n\t\t\t\t\n\t\t\t\t// Clear\n\t\t\t\tctx.memory.user.Clear();\n\t\t\t\t\n\t\t\t\t// Check len\n\t\t\t\tconst len = ctx.memory.user.Len();\n\t\t\t\tif (len !== 0) throw new Error(\"Len should be 0 after clear, got \" + len);\n\t\t\t\t\n\t\t\t\treturn { \n\t\t\t\t\tsuccess: true,\n\t\t\t\t\tlen: len\n\t\t\t\t};\n\t\t\t} catch (error) {\n\t\t\t\treturn { success: false, error: error.message };\n\t\t\t}\n\t\t}`, ctx)\n\n\trequire.NoError(t, err)\n\tresult := res.(map[string]interface{})\n\tassert.True(t, result[\"success\"].(bool))\n\tassert.Equal(t, float64(0), result[\"len\"])\n}\n\nfunc TestMemoryGetDel(t *testing.T) {\n\ttest.Prepare(t, config.Conf)\n\tdefer test.Clean()\n\n\tmem, err := memory.New(nil, \"user1\", \"team1\", \"chat1\", \"ctx1\")\n\trequire.NoError(t, err)\n\n\tctx := &context.Context{\n\t\tChatID: \"test-chat-id\",\n\t\tAssistantID: \"test-assistant-id\",\n\t\tLocale: \"en\",\n\t\tContext: stdContext.Background(),\n\t\tMemory: mem,\n\t}\n\n\tres, err := v8.Call(v8.CallOptions{}, `\n\t\tfunction test(ctx) {\n\t\t\ttry {\n\t\t\t\t// Set a one-time value\n\t\t\t\tctx.memory.user.Set(\"token\", \"secret123\");\n\t\t\t\t\n\t\t\t\t// GetDel\n\t\t\t\tconst value = ctx.memory.user.GetDel(\"token\");\n\t\t\t\tif (value !== \"secret123\") throw new Error(\"Value mismatch\");\n\t\t\t\t\n\t\t\t\t// Should be deleted\n\t\t\t\tconst after = ctx.memory.user.Get(\"token\");\n\t\t\t\tif (after !== null) throw new Error(\"Should be null after GetDel\");\n\t\t\t\t\n\t\t\t\treturn { \n\t\t\t\t\tsuccess: true,\n\t\t\t\t\tvalue: value,\n\t\t\t\t\tafter: after\n\t\t\t\t};\n\t\t\t} catch (error) {\n\t\t\t\treturn { success: false, error: error.message };\n\t\t\t}\n\t\t}`, ctx)\n\n\trequire.NoError(t, err)\n\tresult := res.(map[string]interface{})\n\tassert.True(t, result[\"success\"].(bool))\n\tassert.Equal(t, \"secret123\", result[\"value\"])\n\tassert.Nil(t, result[\"after\"])\n}\n\nfunc TestMemoryIsolation(t *testing.T) {\n\ttest.Prepare(t, config.Conf)\n\tdefer test.Clean()\n\n\t// Create two different memory instances\n\tmem1, err := memory.New(nil, \"user1\", \"\", \"\", \"\")\n\trequire.NoError(t, err)\n\n\tmem2, err := memory.New(nil, \"user2\", \"\", \"\", \"\")\n\trequire.NoError(t, err)\n\n\tctx1 := &context.Context{\n\t\tChatID: \"chat1\",\n\t\tContext: stdContext.Background(),\n\t\tMemory: mem1,\n\t}\n\n\tctx2 := &context.Context{\n\t\tChatID: \"chat2\",\n\t\tContext: stdContext.Background(),\n\t\tMemory: mem2,\n\t}\n\n\t// Set value in user1\n\tres1, err := v8.Call(v8.CallOptions{}, `\n\t\tfunction test(ctx) {\n\t\t\tctx.memory.user.Set(\"key\", \"user1_value\");\n\t\t\treturn ctx.memory.user.Get(\"key\");\n\t\t}`, ctx1)\n\trequire.NoError(t, err)\n\tassert.Equal(t, \"user1_value\", res1)\n\n\t// Set value in user2\n\tres2, err := v8.Call(v8.CallOptions{}, `\n\t\tfunction test(ctx) {\n\t\t\tctx.memory.user.Set(\"key\", \"user2_value\");\n\t\t\treturn ctx.memory.user.Get(\"key\");\n\t\t}`, ctx2)\n\trequire.NoError(t, err)\n\tassert.Equal(t, \"user2_value\", res2)\n\n\t// Verify user1 still has its own value\n\tres3, err := v8.Call(v8.CallOptions{}, `\n\t\tfunction test(ctx) {\n\t\t\treturn ctx.memory.user.Get(\"key\");\n\t\t}`, ctx1)\n\trequire.NoError(t, err)\n\tassert.Equal(t, \"user1_value\", res3)\n}\n\nfunc TestMemoryNoMemory(t *testing.T) {\n\ttest.Prepare(t, config.Conf)\n\tdefer test.Clean()\n\n\tctx := &context.Context{\n\t\tChatID: \"test-chat-id\",\n\t\tContext: stdContext.Background(),\n\t\tMemory: nil, // No memory\n\t}\n\n\tres, err := v8.Call(v8.CallOptions{}, `\n\t\tfunction test(ctx) {\n\t\t\ttry {\n\t\t\t\tconst hasMemory = ctx.memory !== undefined && ctx.memory !== null;\n\t\t\t\treturn { \n\t\t\t\t\tsuccess: true,\n\t\t\t\t\thasMemory: hasMemory\n\t\t\t\t};\n\t\t\t} catch (error) {\n\t\t\t\treturn { success: false, error: error.message };\n\t\t\t}\n\t\t}`, ctx)\n\n\trequire.NoError(t, err)\n\tresult := res.(map[string]interface{})\n\tassert.True(t, result[\"success\"].(bool))\n\tassert.False(t, result[\"hasMemory\"].(bool))\n}\n\nfunc TestMemoryWithAuthorizedInfo(t *testing.T) {\n\ttest.Prepare(t, config.Conf)\n\tdefer test.Clean()\n\n\t// Use context.New to create context with authorized info\n\tauthorized := &types.AuthorizedInfo{\n\t\tUserID: \"user123\",\n\t\tTeamID: \"team456\",\n\t}\n\n\tctx := context.New(stdContext.Background(), authorized, \"chat789\")\n\tdefer ctx.Release()\n\n\t// Verify memory was created with correct IDs\n\trequire.NotNil(t, ctx.Memory)\n\trequire.NotNil(t, ctx.Memory.User)\n\trequire.NotNil(t, ctx.Memory.Team)\n\trequire.NotNil(t, ctx.Memory.Chat)\n\trequire.NotNil(t, ctx.Memory.Context)\n\n\tres, err := v8.Call(v8.CallOptions{}, `\n\t\tfunction test(ctx) {\n\t\t\ttry {\n\t\t\t\t// Set values in different namespaces\n\t\t\t\tctx.memory.user.Set(\"pref\", \"dark\");\n\t\t\t\tctx.memory.team.Set(\"setting\", \"shared\");\n\t\t\t\tctx.memory.chat.Set(\"topic\", \"test\");\n\t\t\t\tctx.memory.context.Set(\"temp\", \"data\");\n\t\t\t\t\n\t\t\t\treturn { \n\t\t\t\t\tsuccess: true,\n\t\t\t\t\tuser: ctx.memory.user.Get(\"pref\"),\n\t\t\t\t\tteam: ctx.memory.team.Get(\"setting\"),\n\t\t\t\t\tchat: ctx.memory.chat.Get(\"topic\"),\n\t\t\t\t\tcontext: ctx.memory.context.Get(\"temp\")\n\t\t\t\t};\n\t\t\t} catch (error) {\n\t\t\t\treturn { success: false, error: error.message };\n\t\t\t}\n\t\t}`, ctx)\n\n\trequire.NoError(t, err)\n\tresult := res.(map[string]interface{})\n\tassert.True(t, result[\"success\"].(bool))\n\tassert.Equal(t, \"dark\", result[\"user\"])\n\tassert.Equal(t, \"shared\", result[\"team\"])\n\tassert.Equal(t, \"test\", result[\"chat\"])\n\tassert.Equal(t, \"data\", result[\"context\"])\n}\n" }, { "path": "agent/context/jsapi_output_test.go", "content": "package context_test\n\nimport (\n\t\"bytes\"\n\tstdContext \"context\"\n\t\"net/http\"\n\t\"testing\"\n\n\t\"github.com/stretchr/testify/assert\"\n\tv8 \"github.com/yaoapp/gou/runtime/v8\"\n\t\"github.com/yaoapp/yao/agent/context\"\n\t\"github.com/yaoapp/yao/config\"\n\t\"github.com/yaoapp/yao/test\"\n)\n\n// testMockResponseWriter is a mock implementation of http.ResponseWriter for testing\ntype testMockResponseWriter struct {\n\theaders http.Header\n\tbuffer *bytes.Buffer\n\tstatus int\n}\n\nfunc newTestMockResponseWriter() *testMockResponseWriter {\n\treturn &testMockResponseWriter{\n\t\theaders: make(http.Header),\n\t\tbuffer: &bytes.Buffer{},\n\t\tstatus: http.StatusOK,\n\t}\n}\n\nfunc (m *testMockResponseWriter) Header() http.Header {\n\treturn m.headers\n}\n\nfunc (m *testMockResponseWriter) Write(b []byte) (int, error) {\n\treturn m.buffer.Write(b)\n}\n\nfunc (m *testMockResponseWriter) WriteHeader(statusCode int) {\n\tm.status = statusCode\n}\n\n// TestJsValueSend test the Send method on Context\nfunc TestJsValueSend(t *testing.T) {\n\n\ttest.Prepare(t, config.Conf)\n\tdefer test.Clean()\n\n\tcxt := context.New(stdContext.Background(), nil, \"test-chat-id\")\n\tcxt.AssistantID = \"test-assistant-id\"\n\tcxt.Accept = context.AcceptStandard\n\tcxt.Locale = \"en\"\n\tcxt.Writer = newTestMockResponseWriter()\n\n\t// Test sending string shorthand\n\tres, err := v8.Call(v8.CallOptions{}, `\n\t\tfunction test(ctx) {\n\t\t\ttry {\n\t\t\t\t// Send simple string\n\t\t\t\tctx.Send(\"Hello World\");\n\t\t\t\treturn { success: true };\n\t\t\t} catch (error) {\n\t\t\t\treturn { success: false, error: error.message };\n\t\t\t}\n\t\t}`, cxt)\n\tif err != nil {\n\t\tt.Fatalf(\"Call failed: %v\", err)\n\t}\n\n\tresult, ok := res.(map[string]interface{})\n\tif !ok {\n\t\tt.Fatalf(\"Expected map result, got %T\", res)\n\t}\n\tassert.Equal(t, true, result[\"success\"], \"Send string should succeed\")\n\n\t// Test sending message object\n\tres, err = v8.Call(v8.CallOptions{}, `\n\t\tfunction test(ctx) {\n\t\t\ttry {\n\t\t\t\t// Send message object\n\t\t\t\tctx.Send({\n\t\t\t\t\ttype: \"text\",\n\t\t\t\t\tprops: {\n\t\t\t\t\t\tcontent: \"Hello from JavaScript\"\n\t\t\t\t\t},\n\t\t\t\t\tid: \"msg_123\",\n\t\t\t\t\tmetadata: {\n\t\t\t\t\t\ttimestamp: Date.now(),\n\t\t\t\t\t\tsequence: 1\n\t\t\t\t\t}\n\t\t\t\t});\n\t\t\t\treturn { success: true };\n\t\t\t} catch (error) {\n\t\t\t\treturn { success: false, error: error.message };\n\t\t\t}\n\t\t}`, cxt)\n\tif err != nil {\n\t\tt.Fatalf(\"Call failed: %v\", err)\n\t}\n\n\tresult, ok = res.(map[string]interface{})\n\tif !ok {\n\t\tt.Fatalf(\"Expected map result, got %T\", res)\n\t}\n\tassert.Equal(t, true, result[\"success\"], \"Send message object should succeed\")\n}\n\n// TestJsValueSendGroup test the SendGroup method on Context\n// TestJsValueSendDeltaUpdates test delta updates in Send\nfunc TestJsValueSendDeltaUpdates(t *testing.T) {\n\n\ttest.Prepare(t, config.Conf)\n\tdefer test.Clean()\n\n\tcxt := context.New(stdContext.Background(), nil, \"test-chat-id\")\n\tcxt.AssistantID = \"test-assistant-id\"\n\tcxt.Accept = context.AcceptStandard\n\tcxt.Locale = \"en\"\n\tcxt.Writer = newTestMockResponseWriter()\n\n\tres, err := v8.Call(v8.CallOptions{}, `\n\t\tfunction test(ctx) {\n\t\t\ttry {\n\t\t\t\t// Send initial message\n\t\t\t\tctx.Send({\n\t\t\t\t\ttype: \"text\",\n\t\t\t\t\tprops: { content: \"Hello\" },\n\t\t\t\t\tid: \"msg_1\",\n\t\t\t\t\tdelta: false\n\t\t\t\t});\n\t\t\t\t\n\t\t\t\t// Send delta update (append)\n\t\t\t\tctx.Send({\n\t\t\t\t\ttype: \"text\",\n\t\t\t\t\tprops: { content: \" World\" },\n\t\t\t\t\tid: \"msg_1\",\n\t\t\t\t\tdelta: true,\n\t\t\t\t\tdelta_path: \"content\",\n\t\t\t\t\tdelta_action: \"append\"\n\t\t\t\t});\n\t\t\t\t\n\t\t\t\t// Send completion (no done field needed)\n\t\t\t\t\n\t\t\t\treturn { success: true };\n\t\t\t} catch (error) {\n\t\t\t\treturn { success: false, error: error.message };\n\t\t\t}\n\t\t}`, cxt)\n\tif err != nil {\n\t\tt.Fatalf(\"Call failed: %v\", err)\n\t}\n\n\tresult, ok := res.(map[string]interface{})\n\tif !ok {\n\t\tt.Fatalf(\"Expected map result, got %T\", res)\n\t}\n\tassert.Equal(t, true, result[\"success\"], \"Delta updates should succeed\")\n}\n\n// TestJsValueSendMultipleTypes test sending different message types\nfunc TestJsValueSendMultipleTypes(t *testing.T) {\n\n\ttest.Prepare(t, config.Conf)\n\tdefer test.Clean()\n\n\tcxt := context.New(stdContext.Background(), nil, \"test-chat-id\")\n\tcxt.AssistantID = \"test-assistant-id\"\n\tcxt.Accept = context.AcceptStandard\n\tcxt.Locale = \"en\"\n\tcxt.Writer = newTestMockResponseWriter()\n\n\tres, err := v8.Call(v8.CallOptions{}, `\n\t\tfunction test(ctx) {\n\t\t\ttry {\n\t\t\t\t// Text message\n\t\t\t\tctx.Send({\n\t\t\t\t\ttype: \"text\",\n\t\t\t\t\tprops: { content: \"Hello\" }\n\t\t\t\t});\n\t\t\t\t\n\t\t\t\t// Thinking message\n\t\t\t\tctx.Send({\n\t\t\t\t\ttype: \"thinking\",\n\t\t\t\t\tprops: { content: \"Let me think...\" }\n\t\t\t\t});\n\t\t\t\t\n\t\t\t\t// Loading message\n\t\t\t\tctx.Send({\n\t\t\t\t\ttype: \"loading\",\n\t\t\t\t\tprops: { message: \"Processing...\" }\n\t\t\t\t});\n\t\t\t\t\n\t\t\t\t// Tool call message\n\t\t\t\tctx.Send({\n\t\t\t\t\ttype: \"tool_call\",\n\t\t\t\t\tprops: {\n\t\t\t\t\t\tid: \"call_123\",\n\t\t\t\t\t\tname: \"get_weather\",\n\t\t\t\t\t\targuments: '{\"location\": \"San Francisco\"}'\n\t\t\t\t\t}\n\t\t\t\t});\n\t\t\t\t\n\t\t\t\t// Error message\n\t\t\t\tctx.Send({\n\t\t\t\t\ttype: \"error\",\n\t\t\t\t\tprops: {\n\t\t\t\t\t\tmessage: \"Something went wrong\",\n\t\t\t\t\t\tcode: \"ERR_500\"\n\t\t\t\t\t}\n\t\t\t\t});\n\t\t\t\t\n\t\t\t\t// Image message\n\t\t\t\tctx.Send({\n\t\t\t\t\ttype: \"image\",\n\t\t\t\t\tprops: {\n\t\t\t\t\t\turl: \"https://example.com/image.jpg\",\n\t\t\t\t\t\talt: \"Example image\",\n\t\t\t\t\t\twidth: 800,\n\t\t\t\t\t\theight: 600\n\t\t\t\t\t}\n\t\t\t\t});\n\t\t\t\t\n\t\t\t\treturn { success: true };\n\t\t\t} catch (error) {\n\t\t\t\treturn { success: false, error: error.message };\n\t\t\t}\n\t\t}`, cxt)\n\tif err != nil {\n\t\tt.Fatalf(\"Call failed: %v\", err)\n\t}\n\n\tresult, ok := res.(map[string]interface{})\n\tif !ok {\n\t\tt.Fatalf(\"Expected map result, got %T\", res)\n\t}\n\tassert.Equal(t, true, result[\"success\"], \"Multiple message types should succeed\")\n}\n\n// TestJsValueSendErrorHandling test error handling in Send\nfunc TestJsValueSendErrorHandling(t *testing.T) {\n\n\ttest.Prepare(t, config.Conf)\n\tdefer test.Clean()\n\n\tcxt := context.New(stdContext.Background(), nil, \"test-chat-id\")\n\tcxt.AssistantID = \"test-assistant-id\"\n\tcxt.Accept = context.AcceptStandard\n\tcxt.Locale = \"en\"\n\tcxt.Writer = newTestMockResponseWriter()\n\n\t// Test invalid argument - no arguments\n\tres, err := v8.Call(v8.CallOptions{}, `\n\t\tfunction test(ctx) {\n\t\t\ttry {\n\t\t\t\tctx.Send();\n\t\t\t\treturn { success: true };\n\t\t\t} catch (error) {\n\t\t\t\treturn { success: false, error: error.message };\n\t\t\t}\n\t\t}`, cxt)\n\tif err != nil {\n\t\tt.Fatalf(\"Call failed: %v\", err)\n\t}\n\n\tresult, ok := res.(map[string]interface{})\n\tif !ok {\n\t\tt.Fatalf(\"Expected map result, got %T\", res)\n\t}\n\tassert.Equal(t, false, result[\"success\"], \"Send without arguments should fail\")\n\tassert.Contains(t, result[\"error\"], \"Send requires a message argument\", \"Error should mention missing message\")\n}\n\n// TestJsValueSendGroupErrorHandling test error handling in SendGroup\n// TestJsValueSendWithCUIAccept test Send with CUI accept types\nfunc TestJsValueSendWithCUIAccept(t *testing.T) {\n\n\ttest.Prepare(t, config.Conf)\n\tdefer test.Clean()\n\n\tacceptTypes := []context.Accept{context.AcceptWebCUI, context.AccepNativeCUI, context.AcceptDesktopCUI}\n\n\tfor _, acceptType := range acceptTypes {\n\t\tt.Run(string(acceptType), func(t *testing.T) {\n\t\t\tcxt := context.New(stdContext.Background(), nil, \"test-chat-id\")\n\t\t\tcxt.AssistantID = \"test-assistant-id\"\n\t\t\tcxt.Accept = acceptType\n\t\t\tcxt.Locale = \"en\"\n\t\t\tcxt.Writer = newTestMockResponseWriter()\n\n\t\t\tres, err := v8.Call(v8.CallOptions{}, `\n\t\t\t\tfunction test(ctx) {\n\t\t\t\t\ttry {\n\t\t\t\t\t\tctx.Send({\n\t\t\t\t\t\t\ttype: \"text\",\n\t\t\t\t\t\t\tprops: { content: \"Hello CUI\" }\n\t\t\t\t\t\t});\n\t\t\t\t\t\treturn { success: true };\n\t\t\t\t\t} catch (error) {\n\t\t\t\t\t\treturn { success: false, error: error.message };\n\t\t\t\t\t}\n\t\t\t\t}`, cxt)\n\t\t\tif err != nil {\n\t\t\t\tt.Fatalf(\"Call failed: %v\", err)\n\t\t\t}\n\n\t\t\tresult, ok := res.(map[string]interface{})\n\t\t\tif !ok {\n\t\t\t\tt.Fatalf(\"Expected map result, got %T\", res)\n\t\t\t}\n\t\t\tassert.Equal(t, true, result[\"success\"], \"Send with \"+string(acceptType)+\" should succeed\")\n\t\t})\n\t}\n}\n\n// TestJsValueSendGroupWithMetadata test SendGroup with various metadata\n// TestJsValueSendChainedCalls test chained Send calls\nfunc TestJsValueSendChainedCalls(t *testing.T) {\n\n\ttest.Prepare(t, config.Conf)\n\tdefer test.Clean()\n\n\tcxt := context.New(stdContext.Background(), nil, \"test-chat-id\")\n\tcxt.AssistantID = \"test-assistant-id\"\n\tcxt.Accept = context.AcceptStandard\n\tcxt.Locale = \"en\"\n\tcxt.Writer = newTestMockResponseWriter()\n\n\tres, err := v8.Call(v8.CallOptions{}, `\n\t\tfunction test(ctx) {\n\t\t\ttry {\n\t\t\t\t// Multiple sequential sends (each auto-flushes)\n\t\t\t\tctx.Send(\"Step 1\");\n\t\t\t\tctx.Send(\"Step 2\");\n\t\t\t\tctx.Send(\"Step 3\");\n\t\t\t\tctx.Send(\"Step 4\");\n\t\t\t\t\n\t\t\t\treturn { success: true };\n\t\t\t} catch (error) {\n\t\t\t\treturn { success: false, error: error.message };\n\t\t\t}\n\t\t}`, cxt)\n\tif err != nil {\n\t\tt.Fatalf(\"Call failed: %v\", err)\n\t}\n\n\tresult, ok := res.(map[string]interface{})\n\tif !ok {\n\t\tt.Fatalf(\"Expected map result, got %T\", res)\n\t}\n\tassert.Equal(t, true, result[\"success\"], \"Chained Send calls should succeed\")\n}\n\n// TestJsValueIDGenerators test ID generator methods\nfunc TestJsValueIDGenerators(t *testing.T) {\n\n\ttest.Prepare(t, config.Conf)\n\tdefer test.Clean()\n\n\tcxt := context.New(stdContext.Background(), nil, \"test-chat-id\")\n\tcxt.AssistantID = \"test-assistant-id\"\n\tcxt.Accept = context.AcceptStandard\n\tcxt.Locale = \"en\"\n\tcxt.Writer = newTestMockResponseWriter()\n\n\tres, err := v8.Call(v8.CallOptions{}, `\n\t\tfunction test(ctx) {\n\t\t\ttry {\n\t\t\t\t// Test MessageID generator\n\t\t\t\tconst msgId1 = ctx.MessageID();\n\t\t\t\tconst msgId2 = ctx.MessageID();\n\t\t\t\t\n\t\t\t\t// Test BlockID generator\n\t\t\t\tconst blockId1 = ctx.BlockID();\n\t\t\t\tconst blockId2 = ctx.BlockID();\n\t\t\t\t\n\t\t\t\t// Test ThreadID generator\n\t\t\t\tconst threadId1 = ctx.ThreadID();\n\t\t\t\tconst threadId2 = ctx.ThreadID();\n\t\t\t\t\n\t\t\t\t// Verify IDs are strings and sequential\n\t\t\t\tif (typeof msgId1 !== 'string' || typeof msgId2 !== 'string') {\n\t\t\t\t\tthrow new Error('MessageID should return string');\n\t\t\t\t}\n\t\t\t\tif (typeof blockId1 !== 'string' || typeof blockId2 !== 'string') {\n\t\t\t\t\tthrow new Error('BlockID should return string');\n\t\t\t\t}\n\t\t\t\tif (typeof threadId1 !== 'string' || typeof threadId2 !== 'string') {\n\t\t\t\t\tthrow new Error('ThreadID should return string');\n\t\t\t\t}\n\t\t\t\t\n\t\t\t\t// Verify they follow the pattern (M1, M2, B1, B2, T1, T2)\n\t\t\t\tif (!msgId1.startsWith('M') || !msgId2.startsWith('M')) {\n\t\t\t\t\tthrow new Error('MessageID should start with M');\n\t\t\t\t}\n\t\t\t\tif (!blockId1.startsWith('B') || !blockId2.startsWith('B')) {\n\t\t\t\t\tthrow new Error('BlockID should start with B');\n\t\t\t\t}\n\t\t\t\tif (!threadId1.startsWith('T') || !threadId2.startsWith('T')) {\n\t\t\t\t\tthrow new Error('ThreadID should start with T');\n\t\t\t\t}\n\t\t\t\t\n\t\t\t\treturn { \n\t\t\t\t\tsuccess: true,\n\t\t\t\t\tmsgId1: msgId1,\n\t\t\t\t\tmsgId2: msgId2,\n\t\t\t\t\tblockId1: blockId1,\n\t\t\t\t\tblockId2: blockId2,\n\t\t\t\t\tthreadId1: threadId1,\n\t\t\t\t\tthreadId2: threadId2\n\t\t\t\t};\n\t\t\t} catch (error) {\n\t\t\t\treturn { success: false, error: error.message };\n\t\t\t}\n\t\t}`, cxt)\n\tif err != nil {\n\t\tt.Fatalf(\"Call failed: %v\", err)\n\t}\n\n\tresult, ok := res.(map[string]interface{})\n\tif !ok {\n\t\tt.Fatalf(\"Expected map result, got %T\", res)\n\t}\n\tassert.Equal(t, true, result[\"success\"], \"ID generators should succeed\")\n}\n\n// TestJsValueSendWithBlockID test Send with block_id parameter\nfunc TestJsValueSendWithBlockID(t *testing.T) {\n\n\ttest.Prepare(t, config.Conf)\n\tdefer test.Clean()\n\n\tcxt := context.New(stdContext.Background(), nil, \"test-chat-id\")\n\tcxt.AssistantID = \"test-assistant-id\"\n\tcxt.Accept = context.AcceptStandard\n\tcxt.Locale = \"en\"\n\tcxt.Writer = newTestMockResponseWriter()\n\n\tres, err := v8.Call(v8.CallOptions{}, `\n\t\tfunction test(ctx) {\n\t\t\ttry {\n\t\t\t\t// Generate block ID manually\n\t\t\t\tconst blockId = ctx.BlockID();\n\t\t\t\t\n\t\t\t\t// Send multiple messages with same block ID\n\t\t\t\tconst msg1 = ctx.Send(\"Message 1\", blockId);\n\t\t\t\tconst msg2 = ctx.Send(\"Message 2\", blockId);\n\t\t\t\tconst msg3 = ctx.Send(\"Message 3\", blockId);\n\t\t\t\t\n\t\t\t\t// Send message with block_id in object (higher priority)\n\t\t\t\tconst msg4 = ctx.Send({\n\t\t\t\t\ttype: \"text\",\n\t\t\t\t\tprops: { content: \"Message 4\" },\n\t\t\t\t\tblock_id: \"B_custom\"\n\t\t\t\t}, blockId); // blockId parameter should be ignored\n\t\t\t\t\n\t\t\t\treturn { \n\t\t\t\t\tsuccess: true,\n\t\t\t\t\tmsg1: msg1,\n\t\t\t\t\tmsg2: msg2,\n\t\t\t\t\tmsg3: msg3,\n\t\t\t\t\tmsg4: msg4,\n\t\t\t\t\tblockId: blockId\n\t\t\t\t};\n\t\t\t} catch (error) {\n\t\t\t\treturn { success: false, error: error.message };\n\t\t\t}\n\t\t}`, cxt)\n\tif err != nil {\n\t\tt.Fatalf(\"Call failed: %v\", err)\n\t}\n\n\tresult, ok := res.(map[string]interface{})\n\tif !ok {\n\t\tt.Fatalf(\"Expected map result, got %T\", res)\n\t}\n\tassert.Equal(t, true, result[\"success\"], \"Send with blockId should succeed\")\n}\n\n// TestJsValueReplace test ctx.Replace method\nfunc TestJsValueReplace(t *testing.T) {\n\n\ttest.Prepare(t, config.Conf)\n\tdefer test.Clean()\n\n\tcxt := context.New(stdContext.Background(), nil, \"test-chat-id\")\n\tcxt.AssistantID = \"test-assistant-id\"\n\tcxt.Accept = context.AcceptStandard\n\tcxt.Locale = \"en\"\n\tcxt.Writer = newTestMockResponseWriter()\n\n\tres, err := v8.Call(v8.CallOptions{}, `\n\t\tfunction test(ctx) {\n\t\t\ttry {\n\t\t\t\t// Send initial message\n\t\t\t\tconst msgId = ctx.Send(\"Initial content\");\n\t\t\t\t\n\t\t\t\t// Replace with new content\n\t\t\t\tctx.Replace(msgId, \"Updated content\");\n\t\t\t\t\n\t\t\t\t// Replace with object\n\t\t\t\tctx.Replace(msgId, {\n\t\t\t\t\ttype: \"text\",\n\t\t\t\t\tprops: { content: \"Final content\" }\n\t\t\t\t});\n\t\t\t\t\n\t\t\t\treturn { success: true, msgId: msgId };\n\t\t\t} catch (error) {\n\t\t\t\treturn { success: false, error: error.message };\n\t\t\t}\n\t\t}`, cxt)\n\tif err != nil {\n\t\tt.Fatalf(\"Call failed: %v\", err)\n\t}\n\n\tresult, ok := res.(map[string]interface{})\n\tif !ok {\n\t\tt.Fatalf(\"Expected map result, got %T\", res)\n\t}\n\tassert.Equal(t, true, result[\"success\"], \"Replace should succeed\")\n}\n\n// TestJsValueAppend test ctx.Append method\nfunc TestJsValueAppend(t *testing.T) {\n\n\ttest.Prepare(t, config.Conf)\n\tdefer test.Clean()\n\n\tcxt := context.New(stdContext.Background(), nil, \"test-chat-id\")\n\tcxt.AssistantID = \"test-assistant-id\"\n\tcxt.Accept = context.AcceptStandard\n\tcxt.Locale = \"en\"\n\tcxt.Writer = newTestMockResponseWriter()\n\n\tres, err := v8.Call(v8.CallOptions{}, `\n\t\tfunction test(ctx) {\n\t\t\ttry {\n\t\t\t\t// Send initial message\n\t\t\t\tconst msgId = ctx.Send(\"Hello\");\n\t\t\t\t\n\t\t\t\t// Append to default path\n\t\t\t\tctx.Append(msgId, \" World\");\n\t\t\t\tctx.Append(msgId, \"!\");\n\t\t\t\t\n\t\t\t\t// Append to specific path\n\t\t\t\tconst msgId2 = ctx.Send({\n\t\t\t\t\ttype: \"data\",\n\t\t\t\t\tprops: { content: \"Line 1\\n\" }\n\t\t\t\t});\n\t\t\t\tctx.Append(msgId2, \"Line 2\\n\", \"props.content\");\n\t\t\t\tctx.Append(msgId2, \"Line 3\\n\", \"props.content\");\n\t\t\t\t\n\t\t\t\treturn { success: true, msgId: msgId, msgId2: msgId2 };\n\t\t\t} catch (error) {\n\t\t\t\treturn { success: false, error: error.message };\n\t\t\t}\n\t\t}`, cxt)\n\tif err != nil {\n\t\tt.Fatalf(\"Call failed: %v\", err)\n\t}\n\n\tresult, ok := res.(map[string]interface{})\n\tif !ok {\n\t\tt.Fatalf(\"Expected map result, got %T\", res)\n\t}\n\tassert.Equal(t, true, result[\"success\"], \"Append should succeed\")\n}\n\n// TestJsValueMerge test ctx.Merge method\nfunc TestJsValueMerge(t *testing.T) {\n\n\ttest.Prepare(t, config.Conf)\n\tdefer test.Clean()\n\n\tcxt := context.New(stdContext.Background(), nil, \"test-chat-id\")\n\tcxt.AssistantID = \"test-assistant-id\"\n\tcxt.Accept = context.AcceptStandard\n\tcxt.Locale = \"en\"\n\tcxt.Writer = newTestMockResponseWriter()\n\n\tres, err := v8.Call(v8.CallOptions{}, `\n\t\tfunction test(ctx) {\n\t\t\ttry {\n\t\t\t\t// Send initial message with object\n\t\t\t\tconst msgId = ctx.Send({\n\t\t\t\t\ttype: \"status\",\n\t\t\t\t\tprops: {\n\t\t\t\t\t\tstatus: \"running\",\n\t\t\t\t\t\tprogress: 0,\n\t\t\t\t\t\tstarted: true\n\t\t\t\t\t}\n\t\t\t\t});\n\t\t\t\t\n\t\t\t\t// Merge updates (keeps other fields)\n\t\t\t\tctx.Merge(msgId, {\n\t\t\t\t\ttype: \"status\",\n\t\t\t\t\tprops: { progress: 50 }\n\t\t\t\t}, \"props\");\n\t\t\t\tctx.Merge(msgId, {\n\t\t\t\t\ttype: \"status\",\n\t\t\t\t\tprops: { progress: 100, status: \"completed\" }\n\t\t\t\t}, \"props\");\n\t\t\t\t\n\t\t\t\treturn { success: true, msgId: msgId };\n\t\t\t} catch (error) {\n\t\t\t\treturn { success: false, error: error.message };\n\t\t\t}\n\t\t}`, cxt)\n\tif err != nil {\n\t\tt.Fatalf(\"Call failed: %v\", err)\n\t}\n\n\tresult, ok := res.(map[string]interface{})\n\tif !ok {\n\t\tt.Fatalf(\"Expected map result, got %T\", res)\n\t}\n\tif !result[\"success\"].(bool) {\n\t\tt.Logf(\"Error: %v\", result[\"error\"])\n\t}\n\tassert.Equal(t, true, result[\"success\"], \"Merge should succeed\")\n}\n\n// TestJsValueSet test ctx.Set method\nfunc TestJsValueSet(t *testing.T) {\n\n\ttest.Prepare(t, config.Conf)\n\tdefer test.Clean()\n\n\tcxt := context.New(stdContext.Background(), nil, \"test-chat-id\")\n\tcxt.AssistantID = \"test-assistant-id\"\n\tcxt.Accept = context.AcceptStandard\n\tcxt.Locale = \"en\"\n\tcxt.Writer = newTestMockResponseWriter()\n\n\tres, err := v8.Call(v8.CallOptions{}, `\n\t\tfunction test(ctx) {\n\t\t\ttry {\n\t\t\t\t// Send initial message\n\t\t\t\tconst msgId = ctx.Send({\n\t\t\t\t\ttype: \"result\",\n\t\t\t\t\tprops: { content: \"Initial\" }\n\t\t\t\t});\n\t\t\t\t\n\t\t\t\t// Set new fields\n\t\t\t\tctx.Set(msgId, {\n\t\t\t\t\ttype: \"result\",\n\t\t\t\t\tprops: { status: \"success\" }\n\t\t\t\t}, \"props.status\");\n\t\t\t\tctx.Set(msgId, {\n\t\t\t\t\ttype: \"result\",\n\t\t\t\t\tprops: { timestamp: Date.now() }\n\t\t\t\t}, \"props.timestamp\");\n\t\t\t\tctx.Set(msgId, {\n\t\t\t\t\ttype: \"result\",\n\t\t\t\t\tprops: { metadata: { duration: 1500 } }\n\t\t\t\t}, \"props.metadata\");\n\t\t\t\t\n\t\t\t\treturn { success: true, msgId: msgId };\n\t\t\t} catch (error) {\n\t\t\t\treturn { success: false, error: error.message };\n\t\t\t}\n\t\t}`, cxt)\n\tif err != nil {\n\t\tt.Fatalf(\"Call failed: %v\", err)\n\t}\n\n\tresult, ok := res.(map[string]interface{})\n\tif !ok {\n\t\tt.Fatalf(\"Expected map result, got %T\", res)\n\t}\n\tif !result[\"success\"].(bool) {\n\t\tt.Logf(\"Error: %v\", result[\"error\"])\n\t}\n\tassert.Equal(t, true, result[\"success\"], \"Set should succeed\")\n}\n\n// TestJsValueBlockIDInheritance test that delta operations inherit block_id\nfunc TestJsValueBlockIDInheritance(t *testing.T) {\n\n\ttest.Prepare(t, config.Conf)\n\tdefer test.Clean()\n\n\tcxt := context.New(stdContext.Background(), nil, \"test-chat-id\")\n\tcxt.AssistantID = \"test-assistant-id\"\n\tcxt.Accept = context.AcceptStandard\n\tcxt.Locale = \"en\"\n\tcxt.Writer = newTestMockResponseWriter()\n\n\tres, err := v8.Call(v8.CallOptions{}, `\n\t\tfunction test(ctx) {\n\t\t\ttry {\n\t\t\t\t// Send message with block_id\n\t\t\t\tconst blockId = ctx.BlockID();\n\t\t\t\tconst msgId = ctx.Send(\"Initial message\", blockId);\n\t\t\t\t\n\t\t\t\t// Delta operations should inherit block_id automatically\n\t\t\t\tctx.Append(msgId, \" appended\");\n\t\t\t\tctx.Replace(msgId, \"Replaced message\");\n\t\t\t\tctx.Merge(msgId, {\n\t\t\t\t\ttype: \"text\",\n\t\t\t\t\tprops: { status: \"done\" }\n\t\t\t\t}, \"props\");\n\t\t\t\tctx.Set(msgId, {\n\t\t\t\t\ttype: \"text\",\n\t\t\t\t\tprops: { state: \"final\" }\n\t\t\t\t}, \"props.state\");\n\t\t\t\t\n\t\t\t\treturn { success: true, msgId: msgId, blockId: blockId };\n\t\t\t} catch (error) {\n\t\t\t\treturn { success: false, error: error.message };\n\t\t\t}\n\t\t}`, cxt)\n\tif err != nil {\n\t\tt.Fatalf(\"Call failed: %v\", err)\n\t}\n\n\tresult, ok := res.(map[string]interface{})\n\tif !ok {\n\t\tt.Fatalf(\"Expected map result, got %T\", res)\n\t}\n\tif !result[\"success\"].(bool) {\n\t\tt.Logf(\"Error: %v\", result[\"error\"])\n\t}\n\tassert.Equal(t, true, result[\"success\"], \"Delta operations should inherit block_id\")\n}\n\n// TestJsValueEndBlock tests the EndBlock method on Context\nfunc TestJsValueEndBlock(t *testing.T) {\n\n\ttest.Prepare(t, config.Conf)\n\tdefer test.Clean()\n\n\t// Setup mock writer\n\tmockWriter := newTestMockResponseWriter()\n\n\t// Use New() to properly initialize messageMetadata\n\tcxt := context.New(stdContext.Background(), nil, \"test-chat-id\")\n\tcxt.AssistantID = \"test-assistant-id\"\n\tcxt.Accept = context.AcceptWebCUI\n\tcxt.Locale = \"en\"\n\tcxt.Writer = mockWriter\n\n\t// Test EndBlock method\n\tres, err := v8.Call(v8.CallOptions{}, `\n\t\tfunction test(ctx) {\n\t\t\ttry {\n\t\t\t\t// Create a block and send messages\n\t\t\t\tconst block_id = ctx.BlockID(); // \"B1\"\n\t\t\t\t\n\t\t\t\tctx.Send(\"Message 1\", block_id);\n\t\t\t\tctx.Send(\"Message 2\", block_id);\n\t\t\t\tctx.Send(\"Message 3\", block_id);\n\t\t\t\t\n\t\t\t\t// End the block manually\n\t\t\t\tctx.EndBlock(block_id);\n\t\t\t\t\n\t\t\t\treturn { success: true, block_id: block_id };\n\t\t\t} catch (error) {\n\t\t\t\treturn { success: false, error: error.message };\n\t\t\t}\n\t\t}`, cxt)\n\tif err != nil {\n\t\tt.Fatalf(\"Call failed: %v\", err)\n\t}\n\n\tresult, ok := res.(map[string]interface{})\n\tif !ok {\n\t\tt.Fatalf(\"Expected map result, got %T\", res)\n\t}\n\tif !result[\"success\"].(bool) {\n\t\tt.Logf(\"Error: %v\", result[\"error\"])\n\t}\n\tassert.Equal(t, true, result[\"success\"], \"EndBlock should work correctly\")\n\n\t// Close SafeWriter to wait for all async writes to complete\n\tcxt.CloseSafeWriter()\n\n\t// Verify that block_end event was sent\n\toutput := mockWriter.buffer.String()\n\tassert.Contains(t, output, \"block_end\", \"Output should contain block_end event\")\n}\n\n// TestJsValueSendStream tests the SendStream method on Context\nfunc TestJsValueSendStream(t *testing.T) {\n\n\ttest.Prepare(t, config.Conf)\n\tdefer test.Clean()\n\n\t// Setup mock writer\n\tmockWriter := newTestMockResponseWriter()\n\n\t// Use New() to properly initialize messageMetadata\n\tcxt := context.New(stdContext.Background(), nil, \"test-chat-id\")\n\tcxt.AssistantID = \"test-assistant-id\"\n\tcxt.Accept = context.AcceptWebCUI\n\tcxt.Locale = \"en\"\n\tcxt.Writer = mockWriter\n\n\t// Test SendStream method\n\tres, err := v8.Call(v8.CallOptions{}, `\n\t\tfunction test(ctx) {\n\t\t\ttry {\n\t\t\t\t// Start a streaming message\n\t\t\t\tconst msgId = ctx.SendStream({\n\t\t\t\t\ttype: \"text\",\n\t\t\t\t\tprops: { content: \"Initial content\" }\n\t\t\t\t});\n\t\t\t\t\n\t\t\t\t// Verify msgId is returned\n\t\t\t\tif (typeof msgId !== 'string' || msgId === '') {\n\t\t\t\t\tthrow new Error('SendStream should return a message ID');\n\t\t\t\t}\n\t\t\t\t\n\t\t\t\treturn { success: true, msgId: msgId };\n\t\t\t} catch (error) {\n\t\t\t\treturn { success: false, error: error.message };\n\t\t\t}\n\t\t}`, cxt)\n\tif err != nil {\n\t\tt.Fatalf(\"Call failed: %v\", err)\n\t}\n\n\tresult, ok := res.(map[string]interface{})\n\tif !ok {\n\t\tt.Fatalf(\"Expected map result, got %T\", res)\n\t}\n\tif !result[\"success\"].(bool) {\n\t\tt.Logf(\"Error: %v\", result[\"error\"])\n\t}\n\tassert.Equal(t, true, result[\"success\"], \"SendStream should work correctly\")\n\n\t// Close SafeWriter to wait for all async writes to complete\n\tcxt.CloseSafeWriter()\n\n\t// Verify message_start was sent but NOT message_end\n\toutput := mockWriter.buffer.String()\n\tassert.Contains(t, output, \"message_start\", \"Output should contain message_start event\")\n\tassert.NotContains(t, output, \"message_end\", \"Output should NOT contain message_end event (streaming)\")\n}\n\n// TestJsValueSendStreamWithBlockID tests SendStream with block_id parameter\nfunc TestJsValueSendStreamWithBlockID(t *testing.T) {\n\n\ttest.Prepare(t, config.Conf)\n\tdefer test.Clean()\n\n\tmockWriter := newTestMockResponseWriter()\n\n\tcxt := context.New(stdContext.Background(), nil, \"test-chat-id\")\n\tcxt.AssistantID = \"test-assistant-id\"\n\tcxt.Accept = context.AcceptWebCUI\n\tcxt.Locale = \"en\"\n\tcxt.Writer = mockWriter\n\n\tres, err := v8.Call(v8.CallOptions{}, `\n\t\tfunction test(ctx) {\n\t\t\ttry {\n\t\t\t\t// Generate block ID\n\t\t\t\tconst blockId = ctx.BlockID();\n\t\t\t\t\n\t\t\t\t// Start streaming with block_id\n\t\t\t\tconst msgId = ctx.SendStream({\n\t\t\t\t\ttype: \"text\",\n\t\t\t\t\tprops: { content: \"Streaming with block\" },\n\t\t\t\t\tblock_id: blockId\n\t\t\t\t});\n\t\t\t\t\n\t\t\t\treturn { success: true, msgId: msgId, blockId: blockId };\n\t\t\t} catch (error) {\n\t\t\t\treturn { success: false, error: error.message };\n\t\t\t}\n\t\t}`, cxt)\n\tif err != nil {\n\t\tt.Fatalf(\"Call failed: %v\", err)\n\t}\n\n\tresult, ok := res.(map[string]interface{})\n\tif !ok {\n\t\tt.Fatalf(\"Expected map result, got %T\", res)\n\t}\n\tassert.Equal(t, true, result[\"success\"], \"SendStream with blockId should succeed\")\n\n\t// Close SafeWriter to wait for all async writes to complete\n\tcxt.CloseSafeWriter()\n\n\t// Verify block_start was also sent\n\toutput := mockWriter.buffer.String()\n\tassert.Contains(t, output, \"block_start\", \"Output should contain block_start event\")\n}\n\n// TestJsValueEnd tests the End method on Context\nfunc TestJsValueEnd(t *testing.T) {\n\n\ttest.Prepare(t, config.Conf)\n\tdefer test.Clean()\n\n\tmockWriter := newTestMockResponseWriter()\n\n\tcxt := context.New(stdContext.Background(), nil, \"test-chat-id\")\n\tcxt.AssistantID = \"test-assistant-id\"\n\tcxt.Accept = context.AcceptWebCUI\n\tcxt.Locale = \"en\"\n\tcxt.Writer = mockWriter\n\n\tres, err := v8.Call(v8.CallOptions{}, `\n\t\tfunction test(ctx) {\n\t\t\ttry {\n\t\t\t\t// Start a streaming message\n\t\t\t\tconst msgId = ctx.SendStream({\n\t\t\t\t\ttype: \"text\",\n\t\t\t\t\tprops: { content: \"Hello\" }\n\t\t\t\t});\n\t\t\t\t\n\t\t\t\t// End the message\n\t\t\t\tctx.End(msgId);\n\t\t\t\t\n\t\t\t\treturn { success: true, msgId: msgId };\n\t\t\t} catch (error) {\n\t\t\t\treturn { success: false, error: error.message };\n\t\t\t}\n\t\t}`, cxt)\n\tif err != nil {\n\t\tt.Fatalf(\"Call failed: %v\", err)\n\t}\n\n\tresult, ok := res.(map[string]interface{})\n\tif !ok {\n\t\tt.Fatalf(\"Expected map result, got %T\", res)\n\t}\n\tif !result[\"success\"].(bool) {\n\t\tt.Logf(\"Error: %v\", result[\"error\"])\n\t}\n\tassert.Equal(t, true, result[\"success\"], \"End should work correctly\")\n\n\t// Close SafeWriter to wait for all async writes to complete\n\tcxt.CloseSafeWriter()\n\n\t// Verify message_end was sent\n\toutput := mockWriter.buffer.String()\n\tassert.Contains(t, output, \"message_end\", \"Output should contain message_end event after End()\")\n}\n\n// TestJsValueEndWithFinalContent tests End with final content parameter\nfunc TestJsValueEndWithFinalContent(t *testing.T) {\n\n\ttest.Prepare(t, config.Conf)\n\tdefer test.Clean()\n\n\tmockWriter := newTestMockResponseWriter()\n\n\tcxt := context.New(stdContext.Background(), nil, \"test-chat-id\")\n\tcxt.AssistantID = \"test-assistant-id\"\n\tcxt.Accept = context.AcceptWebCUI\n\tcxt.Locale = \"en\"\n\tcxt.Writer = mockWriter\n\n\tres, err := v8.Call(v8.CallOptions{}, `\n\t\tfunction test(ctx) {\n\t\t\ttry {\n\t\t\t\t// Start a streaming message\n\t\t\t\tconst msgId = ctx.SendStream({\n\t\t\t\t\ttype: \"text\",\n\t\t\t\t\tprops: { content: \"Start\" }\n\t\t\t\t});\n\t\t\t\t\n\t\t\t\t// End with final content\n\t\t\t\tctx.End(msgId, \" End\");\n\t\t\t\t\n\t\t\t\treturn { success: true, msgId: msgId };\n\t\t\t} catch (error) {\n\t\t\t\treturn { success: false, error: error.message };\n\t\t\t}\n\t\t}`, cxt)\n\tif err != nil {\n\t\tt.Fatalf(\"Call failed: %v\", err)\n\t}\n\n\tresult, ok := res.(map[string]interface{})\n\tif !ok {\n\t\tt.Fatalf(\"Expected map result, got %T\", res)\n\t}\n\tif !result[\"success\"].(bool) {\n\t\tt.Logf(\"Error: %v\", result[\"error\"])\n\t}\n\tassert.Equal(t, true, result[\"success\"], \"End with final content should work correctly\")\n\n\t// Close SafeWriter to wait for all async writes to complete\n\tcxt.CloseSafeWriter()\n\n\t// Verify message_end was sent\n\toutput := mockWriter.buffer.String()\n\tassert.Contains(t, output, \"message_end\", \"Output should contain message_end event\")\n}\n\n// TestJsValueStreamingWorkflow tests the complete streaming workflow: SendStream -> Append -> End\nfunc TestJsValueStreamingWorkflow(t *testing.T) {\n\n\ttest.Prepare(t, config.Conf)\n\tdefer test.Clean()\n\n\tmockWriter := newTestMockResponseWriter()\n\n\tcxt := context.New(stdContext.Background(), nil, \"test-chat-id\")\n\tcxt.AssistantID = \"test-assistant-id\"\n\tcxt.Accept = context.AcceptWebCUI\n\tcxt.Locale = \"en\"\n\tcxt.Writer = mockWriter\n\n\tres, err := v8.Call(v8.CallOptions{}, `\n\t\tfunction test(ctx) {\n\t\t\ttry {\n\t\t\t\t// Start a streaming message\n\t\t\t\tconst msgId = ctx.SendStream({\n\t\t\t\t\ttype: \"text\",\n\t\t\t\t\tprops: { content: \"# Title\\n\\n\" }\n\t\t\t\t});\n\t\t\t\t\n\t\t\t\t// Append content in chunks (simulating streaming)\n\t\t\t\tctx.Append(msgId, \"First paragraph. \");\n\t\t\t\tctx.Append(msgId, \"Second sentence. \");\n\t\t\t\tctx.Append(msgId, \"Third sentence.\\n\\n\");\n\t\t\t\tctx.Append(msgId, \"Second paragraph.\");\n\t\t\t\t\n\t\t\t\t// Finalize the message\n\t\t\t\tctx.End(msgId);\n\t\t\t\t\n\t\t\t\treturn { success: true, msgId: msgId };\n\t\t\t} catch (error) {\n\t\t\t\treturn { success: false, error: error.message };\n\t\t\t}\n\t\t}`, cxt)\n\tif err != nil {\n\t\tt.Fatalf(\"Call failed: %v\", err)\n\t}\n\n\tresult, ok := res.(map[string]interface{})\n\tif !ok {\n\t\tt.Fatalf(\"Expected map result, got %T\", res)\n\t}\n\tif !result[\"success\"].(bool) {\n\t\tt.Logf(\"Error: %v\", result[\"error\"])\n\t}\n\tassert.Equal(t, true, result[\"success\"], \"Streaming workflow should work correctly\")\n\n\t// Close SafeWriter to flush all pending async writes before reading buffer.\n\t// SafeWriter processes writes in a background goroutine via channel;\n\t// without this, the buffer may still be empty on slow CI runners.\n\tcxt.CloseSafeWriter()\n\n\t// Verify the complete workflow events\n\toutput := mockWriter.buffer.String()\n\tassert.Contains(t, output, \"message_start\", \"Output should contain message_start\")\n\tassert.Contains(t, output, \"message_end\", \"Output should contain message_end\")\n\tassert.Contains(t, output, \"# Title\", \"Output should contain initial content\")\n\tassert.Contains(t, output, \"First paragraph\", \"Output should contain appended content\")\n}\n\n// TestJsValueSendStreamStringShorthand tests SendStream with string shorthand\nfunc TestJsValueSendStreamStringShorthand(t *testing.T) {\n\n\ttest.Prepare(t, config.Conf)\n\tdefer test.Clean()\n\n\tmockWriter := newTestMockResponseWriter()\n\n\tcxt := context.New(stdContext.Background(), nil, \"test-chat-id\")\n\tcxt.AssistantID = \"test-assistant-id\"\n\tcxt.Accept = context.AcceptWebCUI\n\tcxt.Locale = \"en\"\n\tcxt.Writer = mockWriter\n\n\tres, err := v8.Call(v8.CallOptions{}, `\n\t\tfunction test(ctx) {\n\t\t\ttry {\n\t\t\t\t// SendStream with string shorthand\n\t\t\t\tconst msgId = ctx.SendStream(\"Hello streaming\");\n\t\t\t\t\n\t\t\t\tif (typeof msgId !== 'string' || msgId === '') {\n\t\t\t\t\tthrow new Error('SendStream should return a message ID');\n\t\t\t\t}\n\t\t\t\t\n\t\t\t\tctx.End(msgId);\n\t\t\t\t\n\t\t\t\treturn { success: true, msgId: msgId };\n\t\t\t} catch (error) {\n\t\t\t\treturn { success: false, error: error.message };\n\t\t\t}\n\t\t}`, cxt)\n\tif err != nil {\n\t\tt.Fatalf(\"Call failed: %v\", err)\n\t}\n\n\tresult, ok := res.(map[string]interface{})\n\tif !ok {\n\t\tt.Fatalf(\"Expected map result, got %T\", res)\n\t}\n\tassert.Equal(t, true, result[\"success\"], \"SendStream with string shorthand should succeed\")\n}\n\n// TestJsValueEndErrorHandling tests error handling in End method\nfunc TestJsValueEndErrorHandling(t *testing.T) {\n\n\ttest.Prepare(t, config.Conf)\n\tdefer test.Clean()\n\n\tmockWriter := newTestMockResponseWriter()\n\n\tcxt := context.New(stdContext.Background(), nil, \"test-chat-id\")\n\tcxt.AssistantID = \"test-assistant-id\"\n\tcxt.Accept = context.AcceptWebCUI\n\tcxt.Locale = \"en\"\n\tcxt.Writer = mockWriter\n\n\t// Test End without arguments\n\tres, err := v8.Call(v8.CallOptions{}, `\n\t\tfunction test(ctx) {\n\t\t\ttry {\n\t\t\t\tctx.End();\n\t\t\t\treturn { success: true };\n\t\t\t} catch (error) {\n\t\t\t\treturn { success: false, error: error.message };\n\t\t\t}\n\t\t}`, cxt)\n\tif err != nil {\n\t\tt.Fatalf(\"Call failed: %v\", err)\n\t}\n\n\tresult, ok := res.(map[string]interface{})\n\tif !ok {\n\t\tt.Fatalf(\"Expected map result, got %T\", res)\n\t}\n\tassert.Equal(t, false, result[\"success\"], \"End without arguments should fail\")\n\tassert.Contains(t, result[\"error\"], \"messageId\", \"Error should mention missing messageId\")\n}\n\n// TestJsValueEndWithInvalidMessageID tests End with invalid messageId type\nfunc TestJsValueEndWithInvalidMessageID(t *testing.T) {\n\n\ttest.Prepare(t, config.Conf)\n\tdefer test.Clean()\n\n\tmockWriter := newTestMockResponseWriter()\n\n\tcxt := context.New(stdContext.Background(), nil, \"test-chat-id\")\n\tcxt.AssistantID = \"test-assistant-id\"\n\tcxt.Accept = context.AcceptWebCUI\n\tcxt.Locale = \"en\"\n\tcxt.Writer = mockWriter\n\n\t// Test End with non-string messageId\n\tres, err := v8.Call(v8.CallOptions{}, `\n\t\tfunction test(ctx) {\n\t\t\ttry {\n\t\t\t\tctx.End(123);\n\t\t\t\treturn { success: true };\n\t\t\t} catch (error) {\n\t\t\t\treturn { success: false, error: error.message };\n\t\t\t}\n\t\t}`, cxt)\n\tif err != nil {\n\t\tt.Fatalf(\"Call failed: %v\", err)\n\t}\n\n\tresult, ok := res.(map[string]interface{})\n\tif !ok {\n\t\tt.Fatalf(\"Expected map result, got %T\", res)\n\t}\n\tassert.Equal(t, false, result[\"success\"], \"End with non-string messageId should fail\")\n\tassert.Contains(t, result[\"error\"], \"string\", \"Error should mention messageId must be string\")\n}\n\n// TestJsValueSendStreamErrorHandling tests error handling in SendStream method\nfunc TestJsValueSendStreamErrorHandling(t *testing.T) {\n\n\ttest.Prepare(t, config.Conf)\n\tdefer test.Clean()\n\n\tmockWriter := newTestMockResponseWriter()\n\n\tcxt := context.New(stdContext.Background(), nil, \"test-chat-id\")\n\tcxt.AssistantID = \"test-assistant-id\"\n\tcxt.Accept = context.AcceptWebCUI\n\tcxt.Locale = \"en\"\n\tcxt.Writer = mockWriter\n\n\t// Test SendStream without arguments\n\tres, err := v8.Call(v8.CallOptions{}, `\n\t\tfunction test(ctx) {\n\t\t\ttry {\n\t\t\t\tctx.SendStream();\n\t\t\t\treturn { success: true };\n\t\t\t} catch (error) {\n\t\t\t\treturn { success: false, error: error.message };\n\t\t\t}\n\t\t}`, cxt)\n\tif err != nil {\n\t\tt.Fatalf(\"Call failed: %v\", err)\n\t}\n\n\tresult, ok := res.(map[string]interface{})\n\tif !ok {\n\t\tt.Fatalf(\"Expected map result, got %T\", res)\n\t}\n\tassert.Equal(t, false, result[\"success\"], \"SendStream without arguments should fail\")\n\tassert.Contains(t, result[\"error\"], \"SendStream requires a message argument\", \"Error should mention missing message\")\n}\n\n// TestJsValueMultipleStreams tests handling multiple concurrent streaming messages\nfunc TestJsValueMultipleStreams(t *testing.T) {\n\n\ttest.Prepare(t, config.Conf)\n\tdefer test.Clean()\n\n\tmockWriter := newTestMockResponseWriter()\n\n\tcxt := context.New(stdContext.Background(), nil, \"test-chat-id\")\n\tcxt.AssistantID = \"test-assistant-id\"\n\tcxt.Accept = context.AcceptWebCUI\n\tcxt.Locale = \"en\"\n\tcxt.Writer = mockWriter\n\n\tres, err := v8.Call(v8.CallOptions{}, `\n\t\tfunction test(ctx) {\n\t\t\ttry {\n\t\t\t\t// Start multiple streaming messages\n\t\t\t\tconst msg1 = ctx.SendStream({ type: \"text\", props: { content: \"Stream 1: \" } });\n\t\t\t\tconst msg2 = ctx.SendStream({ type: \"text\", props: { content: \"Stream 2: \" } });\n\t\t\t\t\n\t\t\t\t// Interleave appends\n\t\t\t\tctx.Append(msg1, \"A\");\n\t\t\t\tctx.Append(msg2, \"X\");\n\t\t\t\tctx.Append(msg1, \"B\");\n\t\t\t\tctx.Append(msg2, \"Y\");\n\t\t\t\tctx.Append(msg1, \"C\");\n\t\t\t\tctx.Append(msg2, \"Z\");\n\t\t\t\t\n\t\t\t\t// End both streams\n\t\t\t\tctx.End(msg1);\n\t\t\t\tctx.End(msg2);\n\t\t\t\t\n\t\t\t\treturn { success: true, msg1: msg1, msg2: msg2 };\n\t\t\t} catch (error) {\n\t\t\t\treturn { success: false, error: error.message };\n\t\t\t}\n\t\t}`, cxt)\n\tif err != nil {\n\t\tt.Fatalf(\"Call failed: %v\", err)\n\t}\n\n\tresult, ok := res.(map[string]interface{})\n\tif !ok {\n\t\tt.Fatalf(\"Expected map result, got %T\", res)\n\t}\n\tif !result[\"success\"].(bool) {\n\t\tt.Logf(\"Error: %v\", result[\"error\"])\n\t}\n\tassert.Equal(t, true, result[\"success\"], \"Multiple streams should work correctly\")\n\tassert.NotEqual(t, result[\"msg1\"], result[\"msg2\"], \"Message IDs should be different\")\n}\n\n// TestJsValueSendVsSendStream tests the difference between Send and SendStream\nfunc TestJsValueSendVsSendStream(t *testing.T) {\n\n\ttest.Prepare(t, config.Conf)\n\tdefer test.Clean()\n\n\t// Test Send - should auto-send message_end\n\tt.Run(\"Send auto-ends\", func(t *testing.T) {\n\t\tmockWriter := newTestMockResponseWriter()\n\t\tcxt := context.New(stdContext.Background(), nil, \"test-chat-id\")\n\t\tcxt.AssistantID = \"test-assistant-id\"\n\t\tcxt.Accept = context.AcceptWebCUI\n\t\tcxt.Locale = \"en\"\n\t\tcxt.Writer = mockWriter\n\n\t\t_, err := v8.Call(v8.CallOptions{}, `\n\t\t\tfunction test(ctx) {\n\t\t\t\tctx.Send(\"Complete message\");\n\t\t\t\treturn true;\n\t\t\t}`, cxt)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Call failed: %v\", err)\n\t\t}\n\n\t\t// Close SafeWriter to flush all pending async writes before reading buffer.\n\t\t// SafeWriter processes writes in a background goroutine via channel;\n\t\t// without this, the buffer may still be empty on slow CI runners.\n\t\tcxt.CloseSafeWriter()\n\n\t\toutput := mockWriter.buffer.String()\n\t\tassert.Contains(t, output, \"message_start\", \"Send should emit message_start\")\n\t\tassert.Contains(t, output, \"message_end\", \"Send should auto-emit message_end\")\n\t})\n\n\t// Test SendStream - should NOT auto-send message_end\n\tt.Run(\"SendStream requires explicit End\", func(t *testing.T) {\n\t\tmockWriter := newTestMockResponseWriter()\n\t\tcxt := context.New(stdContext.Background(), nil, \"test-chat-id\")\n\t\tcxt.AssistantID = \"test-assistant-id\"\n\t\tcxt.Accept = context.AcceptWebCUI\n\t\tcxt.Locale = \"en\"\n\t\tcxt.Writer = mockWriter\n\n\t\t_, err := v8.Call(v8.CallOptions{}, `\n\t\t\tfunction test(ctx) {\n\t\t\t\tconst msgId = ctx.SendStream(\"Streaming message\");\n\t\t\t\t// Intentionally NOT calling ctx.End(msgId)\n\t\t\t\treturn msgId;\n\t\t\t}`, cxt)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Call failed: %v\", err)\n\t\t}\n\n\t\t// Close SafeWriter to flush all pending async writes before reading buffer.\n\t\tcxt.CloseSafeWriter()\n\n\t\toutput := mockWriter.buffer.String()\n\t\tassert.Contains(t, output, \"message_start\", \"SendStream should emit message_start\")\n\t\tassert.NotContains(t, output, \"message_end\", \"SendStream should NOT auto-emit message_end\")\n\t})\n}\n" }, { "path": "agent/context/jsapi_release_test.go", "content": "package context_test\n\nimport (\n\tstdContext \"context\"\n\t\"testing\"\n\n\t\"github.com/stretchr/testify/assert\"\n\tv8 \"github.com/yaoapp/gou/runtime/v8\"\n\t\"github.com/yaoapp/yao/agent/context\"\n\t\"github.com/yaoapp/yao/config\"\n\t\"github.com/yaoapp/yao/test\"\n)\n\n// newReleaseTestContext creates a test context for release testing\nfunc newReleaseTestContext() *context.Context {\n\tctx := context.New(stdContext.Background(), nil, \"test-chat-id\")\n\tctx.AssistantID = \"test-assistant-id\"\n\tctx.Referer = context.RefererAPI\n\tstack, _, _ := context.EnterStack(ctx, \"test-assistant\", &context.Options{})\n\tctx.Stack = stack\n\treturn ctx\n}\n\n// TestContextRelease tests explicit Release() method on Context\nfunc TestContextRelease(t *testing.T) {\n\ttest.Prepare(t, config.Conf)\n\tdefer test.Clean()\n\n\tcxt := newReleaseTestContext()\n\n\tres, err := v8.Call(v8.CallOptions{}, `\n\t\tfunction test(ctx) {\n\t\t\t// Verify context has Release method\n\t\t\tif (typeof ctx.Release !== 'function') {\n\t\t\t\tthrow new Error(\"ctx.Release is not a function\")\n\t\t\t}\n\t\t\t\n\t\t\t// Verify context has __release method\n\t\t\tif (typeof ctx.__release !== 'function') {\n\t\t\t\tthrow new Error(\"ctx.__release is not a function\")\n\t\t\t}\n\t\t\t\n\t\t\t// Call Release explicitly\n\t\t\tctx.Release()\n\t\t\t\n\t\t\t// Can call Release multiple times safely (idempotent)\n\t\t\tctx.Release()\n\t\t\t\n\t\t\treturn {\n\t\t\t\thas_release: true,\n\t\t\t\tsuccess: true\n\t\t\t}\n\t\t}`, cxt)\n\n\tif err != nil {\n\t\tt.Fatalf(\"Call failed: %v\", err)\n\t}\n\n\tresult, ok := res.(map[string]interface{})\n\tif !ok {\n\t\tt.Fatalf(\"Expected map result, got %T\", res)\n\t}\n\n\tassert.Equal(t, true, result[\"has_release\"], \"should have Release method\")\n\tassert.Equal(t, true, result[\"success\"], \"release should succeed\")\n}\n\n// TestTraceRelease tests explicit Release() method on Trace\nfunc TestTraceRelease(t *testing.T) {\n\ttest.Prepare(t, config.Conf)\n\tdefer test.Clean()\n\n\tcxt := newReleaseTestContext()\n\n\tres, err := v8.Call(v8.CallOptions{}, `\n\t\tfunction test(ctx) {\n\t\t\t// Get trace\n\t\t\tconst trace = ctx.trace\n\t\t\t\n\t\t\t// Verify trace has Release method\n\t\t\tif (typeof trace.Release !== 'function') {\n\t\t\t\tthrow new Error(\"trace.Release is not a function\")\n\t\t\t}\n\t\t\t\n\t\t\t// Verify trace has __release method\n\t\t\tif (typeof trace.__release !== 'function') {\n\t\t\t\tthrow new Error(\"trace.__release is not a function\")\n\t\t\t}\n\t\t\t\n\t\t\t// Use trace\n\t\t\tconst node = trace.Add({ type: \"test\" }, { label: \"Test Node\" })\n\t\t\ttrace.Info(\"Test message\")\n\t\t\t\n\t\t\t// Release trace explicitly\n\t\t\ttrace.Release()\n\t\t\t\n\t\t\t// Can call Release multiple times safely (idempotent)\n\t\t\ttrace.Release()\n\t\t\t\n\t\t\treturn {\n\t\t\t\thas_release: true,\n\t\t\t\thas_node: !!node,\n\t\t\t\tsuccess: true\n\t\t\t}\n\t\t}`, cxt)\n\n\tif err != nil {\n\t\tt.Fatalf(\"Call failed: %v\", err)\n\t}\n\n\tresult, ok := res.(map[string]interface{})\n\tif !ok {\n\t\tt.Fatalf(\"Expected map result, got %T\", res)\n\t}\n\n\tassert.Equal(t, true, result[\"has_release\"], \"should have Release method\")\n\tassert.Equal(t, true, result[\"has_node\"], \"should create node\")\n\tassert.Equal(t, true, result[\"success\"], \"release should succeed\")\n}\n\n// TestContextReleaseWithTrace tests that releasing Context also releases Trace\nfunc TestContextReleaseWithTrace(t *testing.T) {\n\ttest.Prepare(t, config.Conf)\n\tdefer test.Clean()\n\n\tcxt := newReleaseTestContext()\n\n\tres, err := v8.Call(v8.CallOptions{}, `\n\t\tfunction test(ctx) {\n\t\t\t// Get trace\n\t\t\tconst trace = ctx.trace\n\t\t\t\n\t\t\t// Use trace\n\t\t\tconst node = trace.Add({ type: \"test\" }, { label: \"Test Node\" })\n\t\t\ttrace.Info(\"Test message\")\n\t\t\tnode.Complete({ result: \"done\" })\n\t\t\t\n\t\t\t// Release context (should also release trace)\n\t\t\tctx.Release()\n\t\t\t\n\t\t\treturn {\n\t\t\t\ttrace_released_via_context: true,\n\t\t\t\tsuccess: true\n\t\t\t}\n\t\t}`, cxt)\n\n\tif err != nil {\n\t\tt.Fatalf(\"Call failed: %v\", err)\n\t}\n\n\tresult, ok := res.(map[string]interface{})\n\tif !ok {\n\t\tt.Fatalf(\"Expected map result, got %T\", res)\n\t}\n\n\tassert.Equal(t, true, result[\"trace_released_via_context\"], \"trace should be released via context\")\n\tassert.Equal(t, true, result[\"success\"], \"release should succeed\")\n}\n\n// TestTryFinallyPattern tests the try-finally pattern with Release()\nfunc TestTryFinallyPattern(t *testing.T) {\n\ttest.Prepare(t, config.Conf)\n\tdefer test.Clean()\n\n\tcxt := newReleaseTestContext()\n\n\tres, err := v8.Call(v8.CallOptions{}, `\n\t\tfunction test(ctx) {\n\t\t\tconst trace = ctx.trace\n\t\t\t\n\t\t\t// Try-finally pattern for explicit resource management\n\t\t\ttry {\n\t\t\t\tconst node = trace.Add({ type: \"step\" }, { label: \"Processing\" })\n\t\t\t\t\n\t\t\t\t// Simulate some work\n\t\t\t\ttrace.Info(\"Step 1: Initialize\")\n\t\t\t\ttrace.Info(\"Step 2: Process\")\n\t\t\t\t\n\t\t\t\tnode.Complete({ result: \"success\" })\n\t\t\t\t\n\t\t\t\treturn {\n\t\t\t\t\tcompleted: true\n\t\t\t\t}\n\t\t\t} finally {\n\t\t\t\t// Explicit cleanup\n\t\t\t\ttrace.Release()\n\t\t\t\tctx.Release()\n\t\t\t}\n\t\t}`, cxt)\n\n\tif err != nil {\n\t\tt.Fatalf(\"Call failed: %v\", err)\n\t}\n\n\tresult, ok := res.(map[string]interface{})\n\tif !ok {\n\t\tt.Fatalf(\"Expected map result, got %T\", res)\n\t}\n\n\tassert.Equal(t, true, result[\"completed\"], \"should complete successfully\")\n}\n\n// TestNoOpTraceRelease tests that no-op Trace also has Release method\nfunc TestNoOpTraceRelease(t *testing.T) {\n\ttest.Prepare(t, config.Conf)\n\tdefer test.Clean()\n\n\t// Context without trace initialization\n\tcxt := context.New(stdContext.Background(), nil, \"test-chat-id\")\n\tcxt.AssistantID = \"test-assistant-id\"\n\n\tres, err := v8.Call(v8.CallOptions{}, `\n\t\tfunction test(ctx) {\n\t\t\t// Get trace (should be no-op)\n\t\t\tconst trace = ctx.trace\n\t\t\t\n\t\t\t// Verify trace has Release method even when it's no-op\n\t\t\tif (typeof trace.Release !== 'function') {\n\t\t\t\tthrow new Error(\"no-op trace.Release is not a function\")\n\t\t\t}\n\t\t\t\n\t\t\t// Call methods on no-op trace (should not error)\n\t\t\ttrace.Info(\"This is a no-op\")\n\t\t\tconst node = trace.Add({ type: \"test\" }, { label: \"No-op\" })\n\t\t\tnode.Complete({ result: \"done\" })\n\t\t\t\n\t\t\t// Release no-op trace (should not error)\n\t\t\ttrace.Release()\n\t\t\t\n\t\t\treturn {\n\t\t\t\tnoop_trace_works: true,\n\t\t\t\tsuccess: true\n\t\t\t}\n\t\t}`, cxt)\n\n\tif err != nil {\n\t\tt.Fatalf(\"Call failed: %v\", err)\n\t}\n\n\tresult, ok := res.(map[string]interface{})\n\tif !ok {\n\t\tt.Fatalf(\"Expected map result, got %T\", res)\n\t}\n\n\tassert.Equal(t, true, result[\"noop_trace_works\"], \"no-op trace should work\")\n\tassert.Equal(t, true, result[\"success\"], \"release should succeed\")\n}\n\n// TestTryFinallyPatternWithError tests try-finally with error handling\nfunc TestTryFinallyPatternWithError(t *testing.T) {\n\ttest.Prepare(t, config.Conf)\n\tdefer test.Clean()\n\n\tcxt := newReleaseTestContext()\n\n\t_, err := v8.Call(v8.CallOptions{}, `\n\t\tfunction test(ctx) {\n\t\t\tconst trace = ctx.trace\n\t\t\t\n\t\t\t// Try-finally pattern ensures cleanup even when error occurs\n\t\t\ttry {\n\t\t\t\tconst node = trace.Add({ type: \"step\" }, { label: \"Processing\" })\n\t\t\t\ttrace.Info(\"Starting work\")\n\t\t\t\t\n\t\t\t\t// Simulate an error\n\t\t\t\tthrow new Error(\"Simulated error\")\n\t\t\t\t\n\t\t\t} finally {\n\t\t\t\t// Cleanup happens even after error\n\t\t\t\ttrace.Release()\n\t\t\t\tctx.Release()\n\t\t\t}\n\t\t}`, cxt)\n\n\t// Error should be propagated\n\tif err == nil {\n\t\tt.Fatal(\"Expected error to be propagated\")\n\t}\n\n\t// But cleanup should have happened (no way to verify directly, but test should not crash)\n\tassert.Contains(t, err.Error(), \"Simulated error\", \"error should be propagated\")\n}\n" }, { "path": "agent/context/jsapi_sandbox.go", "content": "package context\n\nimport (\n\t\"context\"\n\n\t\"github.com/yaoapp/gou/runtime/v8/bridge\"\n\topenapiSandbox \"github.com/yaoapp/yao/openapi/sandbox\"\n\tinfraSandbox \"github.com/yaoapp/yao/sandbox\"\n\t\"rogchap.com/v8go\"\n)\n\n// SandboxExecutor defines the interface for sandbox operations\n// This interface is implemented by agent/sandbox.Executor\n// It's defined here to avoid import cycles\ntype SandboxExecutor interface {\n\t// Filesystem operations\n\tReadFile(ctx context.Context, path string) ([]byte, error)\n\tWriteFile(ctx context.Context, path string, content []byte) error\n\tListDir(ctx context.Context, path string) ([]infraSandbox.FileInfo, error)\n\n\t// Command execution\n\tExec(ctx context.Context, cmd []string) (string, error)\n\n\t// Workspace info\n\tGetWorkDir() string\n\n\t// Sandbox identification\n\tGetSandboxID() string\n\n\t// VNC access (returns empty string if not available)\n\tGetVNCUrl() string\n}\n\n// SetSandboxExecutor sets the sandbox executor for this context\n// This should be called before hooks are executed\nfunc (ctx *Context) SetSandboxExecutor(executor SandboxExecutor) {\n\tctx.sandboxExecutor = executor\n}\n\n// GetSandboxExecutor returns the sandbox executor if available\nfunc (ctx *Context) GetSandboxExecutor() SandboxExecutor {\n\treturn ctx.sandboxExecutor\n}\n\n// HasSandbox returns true if sandbox executor is available\nfunc (ctx *Context) HasSandbox() bool {\n\treturn ctx.sandboxExecutor != nil\n}\n\n// newSandboxObject creates the ctx.sandbox JavaScript object\n// Returns nil if sandbox executor is not available\nfunc (ctx *Context) newSandboxObject(iso *v8go.Isolate) *v8go.ObjectTemplate {\n\tif ctx.sandboxExecutor == nil {\n\t\treturn nil\n\t}\n\n\tsandboxObj := v8go.NewObjectTemplate(iso)\n\n\t// Set methods\n\tsandboxObj.Set(\"ReadFile\", ctx.sandboxReadFileMethod(iso))\n\tsandboxObj.Set(\"WriteFile\", ctx.sandboxWriteFileMethod(iso))\n\tsandboxObj.Set(\"ListDir\", ctx.sandboxListDirMethod(iso))\n\tsandboxObj.Set(\"Exec\", ctx.sandboxExecMethod(iso))\n\tsandboxObj.Set(\"GetVNCUrl\", ctx.sandboxGetVNCUrlMethod(iso))\n\tsandboxObj.Set(\"GetSandboxID\", ctx.sandboxGetSandboxIDMethod(iso))\n\n\treturn sandboxObj\n}\n\n// createSandboxInstance creates the sandbox object instance with workdir property\nfunc (ctx *Context) createSandboxInstance(v8ctx *v8go.Context) *v8go.Value {\n\tif ctx.sandboxExecutor == nil {\n\t\treturn nil\n\t}\n\n\tsandboxTemplate := ctx.newSandboxObject(v8ctx.Isolate())\n\tif sandboxTemplate == nil {\n\t\treturn nil\n\t}\n\n\t// Set workdir as a property\n\tsandboxTemplate.Set(\"workdir\", ctx.sandboxExecutor.GetWorkDir())\n\n\t// Set sandbox_id as a property\n\tsandboxID := ctx.sandboxExecutor.GetSandboxID()\n\tsandboxTemplate.Set(\"sandbox_id\", sandboxID)\n\n\t// Set vnc_url as a property (empty string if not available)\n\t// GetVNCUrl returns sandbox ID if VNC is supported, empty otherwise\n\tvncSandboxID := ctx.sandboxExecutor.GetVNCUrl()\n\tif vncSandboxID != \"\" {\n\t\tsandboxTemplate.Set(\"vnc_url\", openapiSandbox.GetVNCClientURL(vncSandboxID))\n\t} else {\n\t\tsandboxTemplate.Set(\"vnc_url\", \"\")\n\t}\n\n\tinstance, err := sandboxTemplate.NewInstance(v8ctx)\n\tif err != nil {\n\t\treturn nil\n\t}\n\n\treturn instance.Value\n}\n\n// sandboxReadFileMethod implements ctx.sandbox.ReadFile(path)\nfunc (ctx *Context) sandboxReadFileMethod(iso *v8go.Isolate) *v8go.FunctionTemplate {\n\treturn v8go.NewFunctionTemplate(iso, func(info *v8go.FunctionCallbackInfo) *v8go.Value {\n\t\tv8ctx := info.Context()\n\t\targs := info.Args()\n\n\t\tif ctx.sandboxExecutor == nil {\n\t\t\treturn bridge.JsException(v8ctx, \"sandbox executor not available\")\n\t\t}\n\n\t\tif len(args) < 1 {\n\t\t\treturn bridge.JsException(v8ctx, \"ReadFile requires path parameter\")\n\t\t}\n\n\t\tpath := args[0].String()\n\n\t\tcontent, err := ctx.sandboxExecutor.ReadFile(context.Background(), path)\n\t\tif err != nil {\n\t\t\treturn bridge.JsException(v8ctx, err.Error())\n\t\t}\n\n\t\t// Return as string\n\t\tjsVal, err := v8go.NewValue(iso, string(content))\n\t\tif err != nil {\n\t\t\treturn bridge.JsException(v8ctx, err.Error())\n\t\t}\n\n\t\treturn jsVal\n\t})\n}\n\n// sandboxWriteFileMethod implements ctx.sandbox.WriteFile(path, content)\nfunc (ctx *Context) sandboxWriteFileMethod(iso *v8go.Isolate) *v8go.FunctionTemplate {\n\treturn v8go.NewFunctionTemplate(iso, func(info *v8go.FunctionCallbackInfo) *v8go.Value {\n\t\tv8ctx := info.Context()\n\t\targs := info.Args()\n\n\t\tif ctx.sandboxExecutor == nil {\n\t\t\treturn bridge.JsException(v8ctx, \"sandbox executor not available\")\n\t\t}\n\n\t\tif len(args) < 2 {\n\t\t\treturn bridge.JsException(v8ctx, \"WriteFile requires path and content parameters\")\n\t\t}\n\n\t\tpath := args[0].String()\n\t\tcontent := args[1].String()\n\n\t\terr := ctx.sandboxExecutor.WriteFile(context.Background(), path, []byte(content))\n\t\tif err != nil {\n\t\t\treturn bridge.JsException(v8ctx, err.Error())\n\t\t}\n\n\t\t// Return undefined on success\n\t\treturn v8go.Undefined(iso)\n\t})\n}\n\n// sandboxListDirMethod implements ctx.sandbox.ListDir(path)\nfunc (ctx *Context) sandboxListDirMethod(iso *v8go.Isolate) *v8go.FunctionTemplate {\n\treturn v8go.NewFunctionTemplate(iso, func(info *v8go.FunctionCallbackInfo) *v8go.Value {\n\t\tv8ctx := info.Context()\n\t\targs := info.Args()\n\n\t\tif ctx.sandboxExecutor == nil {\n\t\t\treturn bridge.JsException(v8ctx, \"sandbox executor not available\")\n\t\t}\n\n\t\tif len(args) < 1 {\n\t\t\treturn bridge.JsException(v8ctx, \"ListDir requires path parameter\")\n\t\t}\n\n\t\tpath := args[0].String()\n\n\t\tfiles, err := ctx.sandboxExecutor.ListDir(context.Background(), path)\n\t\tif err != nil {\n\t\t\treturn bridge.JsException(v8ctx, err.Error())\n\t\t}\n\n\t\t// Convert to JavaScript array of objects\n\t\tresult := make([]map[string]interface{}, len(files))\n\t\tfor i, f := range files {\n\t\t\tresult[i] = map[string]interface{}{\n\t\t\t\t\"name\": f.Name,\n\t\t\t\t\"size\": f.Size,\n\t\t\t\t\"is_dir\": f.IsDir,\n\t\t\t}\n\t\t}\n\n\t\tjsVal, err := bridge.JsValue(v8ctx, result)\n\t\tif err != nil {\n\t\t\treturn bridge.JsException(v8ctx, err.Error())\n\t\t}\n\n\t\treturn jsVal\n\t})\n}\n\n// sandboxExecMethod implements ctx.sandbox.Exec(cmd)\nfunc (ctx *Context) sandboxExecMethod(iso *v8go.Isolate) *v8go.FunctionTemplate {\n\treturn v8go.NewFunctionTemplate(iso, func(info *v8go.FunctionCallbackInfo) *v8go.Value {\n\t\tv8ctx := info.Context()\n\t\targs := info.Args()\n\n\t\tif ctx.sandboxExecutor == nil {\n\t\t\treturn bridge.JsException(v8ctx, \"sandbox executor not available\")\n\t\t}\n\n\t\tif len(args) < 1 {\n\t\t\treturn bridge.JsException(v8ctx, \"Exec requires cmd parameter (array of strings)\")\n\t\t}\n\n\t\t// Parse command array\n\t\tcmdArg := args[0]\n\t\tif !cmdArg.IsArray() {\n\t\t\treturn bridge.JsException(v8ctx, \"Exec requires cmd to be an array of strings\")\n\t\t}\n\n\t\tcmdObj, err := cmdArg.AsObject()\n\t\tif err != nil {\n\t\t\treturn bridge.JsException(v8ctx, \"failed to parse cmd array: \"+err.Error())\n\t\t}\n\n\t\t// Get array length\n\t\tlengthVal, err := cmdObj.Get(\"length\")\n\t\tif err != nil {\n\t\t\treturn bridge.JsException(v8ctx, \"failed to get cmd array length: \"+err.Error())\n\t\t}\n\t\tlength := int(lengthVal.Integer())\n\n\t\t// Build command slice\n\t\tcmd := make([]string, length)\n\t\tfor i := 0; i < length; i++ {\n\t\t\titemVal, err := cmdObj.GetIdx(uint32(i))\n\t\t\tif err != nil {\n\t\t\t\treturn bridge.JsException(v8ctx, \"failed to get cmd array element: \"+err.Error())\n\t\t\t}\n\t\t\tcmd[i] = itemVal.String()\n\t\t}\n\n\t\toutput, err := ctx.sandboxExecutor.Exec(context.Background(), cmd)\n\t\tif err != nil {\n\t\t\treturn bridge.JsException(v8ctx, err.Error())\n\t\t}\n\n\t\tjsVal, err := v8go.NewValue(v8ctx.Isolate(), output)\n\t\tif err != nil {\n\t\t\treturn bridge.JsException(v8ctx, err.Error())\n\t\t}\n\n\t\treturn jsVal\n\t})\n}\n\n// sandboxGetVNCUrlMethod implements ctx.sandbox.GetVNCUrl()\nfunc (ctx *Context) sandboxGetVNCUrlMethod(iso *v8go.Isolate) *v8go.FunctionTemplate {\n\treturn v8go.NewFunctionTemplate(iso, func(info *v8go.FunctionCallbackInfo) *v8go.Value {\n\t\tv8ctx := info.Context()\n\n\t\tif ctx.sandboxExecutor == nil {\n\t\t\treturn bridge.JsException(v8ctx, \"sandbox executor not available\")\n\t\t}\n\n\t\t// GetVNCUrl returns sandbox ID if VNC is supported, empty otherwise\n\t\tvncSandboxID := ctx.sandboxExecutor.GetVNCUrl()\n\t\tvncUrl := \"\"\n\t\tif vncSandboxID != \"\" {\n\t\t\tvncUrl = openapiSandbox.GetVNCClientURL(vncSandboxID)\n\t\t}\n\n\t\tjsVal, err := v8go.NewValue(iso, vncUrl)\n\t\tif err != nil {\n\t\t\treturn bridge.JsException(v8ctx, err.Error())\n\t\t}\n\n\t\treturn jsVal\n\t})\n}\n\n// sandboxGetSandboxIDMethod implements ctx.sandbox.GetSandboxID()\nfunc (ctx *Context) sandboxGetSandboxIDMethod(iso *v8go.Isolate) *v8go.FunctionTemplate {\n\treturn v8go.NewFunctionTemplate(iso, func(info *v8go.FunctionCallbackInfo) *v8go.Value {\n\t\tv8ctx := info.Context()\n\n\t\tif ctx.sandboxExecutor == nil {\n\t\t\treturn bridge.JsException(v8ctx, \"sandbox executor not available\")\n\t\t}\n\n\t\tsandboxID := ctx.sandboxExecutor.GetSandboxID()\n\n\t\tjsVal, err := v8go.NewValue(iso, sandboxID)\n\t\tif err != nil {\n\t\t\treturn bridge.JsException(v8ctx, err.Error())\n\t\t}\n\n\t\treturn jsVal\n\t})\n}\n" }, { "path": "agent/context/jsapi_sandbox_test.go", "content": "package context_test\n\nimport (\n\tstdContext \"context\"\n\t\"os\"\n\t\"testing\"\n\n\t\"github.com/stretchr/testify/assert\"\n\t\"github.com/stretchr/testify/require\"\n\tv8 \"github.com/yaoapp/gou/runtime/v8\"\n\t\"github.com/yaoapp/yao/agent/context\"\n\t\"github.com/yaoapp/yao/config\"\n\tinfraSandbox \"github.com/yaoapp/yao/sandbox\"\n\t\"github.com/yaoapp/yao/test\"\n)\n\n// createTestSandboxManager creates a real sandbox manager for testing\nfunc createTestSandboxManager(t *testing.T) *infraSandbox.Manager {\n\t// Get data root from environment or use temp directory\n\tdataRoot := os.Getenv(\"YAO_ROOT\")\n\tif dataRoot == \"\" {\n\t\tdataRoot = t.TempDir()\n\t}\n\n\t// Create config with proper paths\n\tcfg := infraSandbox.DefaultConfig()\n\tcfg.Init(dataRoot)\n\n\tmanager, err := infraSandbox.NewManager(cfg)\n\tif err != nil {\n\t\tt.Skipf(\"Skipping test: Docker not available: %v\", err)\n\t\treturn nil\n\t}\n\n\treturn manager\n}\n\n// createTestContainer creates a container and returns a cleanup function\nfunc createTestContainer(t *testing.T, manager *infraSandbox.Manager, userID, chatID string) (*infraSandbox.Container, func()) {\n\tcontainer, err := manager.GetOrCreate(stdContext.Background(), userID, chatID)\n\trequire.NoError(t, err)\n\trequire.NotNil(t, container)\n\n\t// Return cleanup function that removes the container\n\tcleanup := func() {\n\t\terr := manager.Remove(stdContext.Background(), container.Name)\n\t\tif err != nil {\n\t\t\tt.Logf(\"Warning: failed to cleanup container %s: %v\", container.Name, err)\n\t\t}\n\t}\n\n\treturn container, cleanup\n}\n\n// realSandboxExecutor wraps infraSandbox.Manager to implement context.SandboxExecutor\ntype realSandboxExecutor struct {\n\tmanager *infraSandbox.Manager\n\tcontainerName string\n\tworkDir string\n}\n\nfunc (e *realSandboxExecutor) ReadFile(ctx stdContext.Context, path string) ([]byte, error) {\n\tfullPath := e.workDir + \"/\" + path\n\treturn e.manager.ReadFile(ctx, e.containerName, fullPath)\n}\n\nfunc (e *realSandboxExecutor) WriteFile(ctx stdContext.Context, path string, content []byte) error {\n\tfullPath := e.workDir + \"/\" + path\n\treturn e.manager.WriteFile(ctx, e.containerName, fullPath, content)\n}\n\nfunc (e *realSandboxExecutor) ListDir(ctx stdContext.Context, path string) ([]infraSandbox.FileInfo, error) {\n\tfullPath := e.workDir + \"/\" + path\n\treturn e.manager.ListDir(ctx, e.containerName, fullPath)\n}\n\nfunc (e *realSandboxExecutor) Exec(ctx stdContext.Context, cmd []string) (string, error) {\n\tresult, err := e.manager.Exec(ctx, e.containerName, cmd, &infraSandbox.ExecOptions{\n\t\tWorkDir: e.workDir,\n\t})\n\tif err != nil {\n\t\treturn \"\", err\n\t}\n\treturn result.Stdout, nil\n}\n\nfunc (e *realSandboxExecutor) GetWorkDir() string {\n\treturn e.workDir\n}\n\nfunc (e *realSandboxExecutor) GetSandboxID() string {\n\t// Extract sandbox ID from container name (format: yao-sandbox-{userID}-{chatID})\n\t// For tests, just return a mock ID\n\treturn \"test-user-test-chat\"\n}\n\nfunc (e *realSandboxExecutor) GetVNCUrl() string {\n\t// Tests don't use VNC, return empty\n\treturn \"\"\n}\n\n// TestJsSandboxNotAvailable tests ctx.sandbox when not configured\nfunc TestJsSandboxNotAvailable(t *testing.T) {\n\ttest.Prepare(t, config.Conf)\n\tdefer test.Clean()\n\n\tctx := context.New(stdContext.Background(), nil, \"test-chat-no-sandbox\")\n\tctx.AssistantID = \"test-assistant\"\n\n\t// Test that ctx.sandbox is undefined when not configured\n\tres, err := v8.Call(v8.CallOptions{}, `\n\t\tfunction test(ctx) {\n\t\t\ttry {\n\t\t\t\tif (ctx.sandbox === undefined || ctx.sandbox === null) {\n\t\t\t\t\treturn { success: true, hasSandbox: false };\n\t\t\t\t}\n\t\t\t\treturn { success: true, hasSandbox: true };\n\t\t\t} catch (error) {\n\t\t\t\treturn { success: false, error: error.message };\n\t\t\t}\n\t\t}`, ctx)\n\n\trequire.NoError(t, err)\n\tresult, ok := res.(map[string]interface{})\n\trequire.True(t, ok, \"Expected map result\")\n\tassert.Equal(t, true, result[\"success\"])\n\tassert.Equal(t, false, result[\"hasSandbox\"], \"ctx.sandbox should not be available when not configured\")\n}\n\n// TestJsSandboxWriteFile tests ctx.sandbox.WriteFile via JavaScript\nfunc TestJsSandboxWriteFile(t *testing.T) {\n\ttest.Prepare(t, config.Conf)\n\tdefer test.Clean()\n\n\tmanager := createTestSandboxManager(t)\n\tif manager == nil {\n\t\treturn\n\t}\n\tdefer manager.Close()\n\n\t// Create container with auto-cleanup\n\tcontainer, cleanup := createTestContainer(t, manager, \"test-user\", \"test-js-writefile\")\n\tdefer cleanup()\n\n\texecutor := &realSandboxExecutor{\n\t\tmanager: manager,\n\t\tcontainerName: container.Name,\n\t\tworkDir: \"/workspace\",\n\t}\n\n\t// Create context with sandbox\n\tctx := context.New(stdContext.Background(), nil, \"test-chat-writefile\")\n\tctx.AssistantID = \"test-assistant\"\n\tctx.SetSandboxExecutor(executor)\n\n\t// Test WriteFile via JavaScript\n\tres, err := v8.Call(v8.CallOptions{}, `\n\t\tfunction test(ctx) {\n\t\t\ttry {\n\t\t\t\tif (!ctx.sandbox) {\n\t\t\t\t\treturn { success: false, error: \"sandbox not available\" };\n\t\t\t\t}\n\t\t\t\t\n\t\t\t\t// Write a file\n\t\t\t\tctx.sandbox.WriteFile(\"js-test.txt\", \"Hello from JavaScript!\");\n\t\t\t\t\n\t\t\t\treturn { success: true };\n\t\t\t} catch (error) {\n\t\t\t\treturn { success: false, error: error.message };\n\t\t\t}\n\t\t}`, ctx)\n\n\trequire.NoError(t, err)\n\tresult, ok := res.(map[string]interface{})\n\trequire.True(t, ok, \"Expected map result\")\n\tassert.Equal(t, true, result[\"success\"], \"WriteFile should succeed: %v\", result[\"error\"])\n\n\t// Verify file was written by reading it back directly\n\tcontent, err := executor.ReadFile(stdContext.Background(), \"js-test.txt\")\n\trequire.NoError(t, err)\n\tassert.Equal(t, \"Hello from JavaScript!\", string(content))\n}\n\n// TestJsSandboxReadFile tests ctx.sandbox.ReadFile via JavaScript\nfunc TestJsSandboxReadFile(t *testing.T) {\n\ttest.Prepare(t, config.Conf)\n\tdefer test.Clean()\n\n\tmanager := createTestSandboxManager(t)\n\tif manager == nil {\n\t\treturn\n\t}\n\tdefer manager.Close()\n\n\t// Create container with auto-cleanup\n\tcontainer, cleanup := createTestContainer(t, manager, \"test-user\", \"test-js-readfile\")\n\tdefer cleanup()\n\n\texecutor := &realSandboxExecutor{\n\t\tmanager: manager,\n\t\tcontainerName: container.Name,\n\t\tworkDir: \"/workspace\",\n\t}\n\n\t// Write a file first\n\terr := executor.WriteFile(stdContext.Background(), \"read-test.txt\", []byte(\"Content to read\"))\n\trequire.NoError(t, err)\n\n\t// Create context with sandbox\n\tctx := context.New(stdContext.Background(), nil, \"test-chat-readfile\")\n\tctx.AssistantID = \"test-assistant\"\n\tctx.SetSandboxExecutor(executor)\n\n\t// Test ReadFile via JavaScript\n\tres, err := v8.Call(v8.CallOptions{}, `\n\t\tfunction test(ctx) {\n\t\t\ttry {\n\t\t\t\tif (!ctx.sandbox) {\n\t\t\t\t\treturn { success: false, error: \"sandbox not available\" };\n\t\t\t\t}\n\t\t\t\t\n\t\t\t\t// Read the file\n\t\t\t\tconst content = ctx.sandbox.ReadFile(\"read-test.txt\");\n\t\t\t\t\n\t\t\t\treturn { success: true, content: content };\n\t\t\t} catch (error) {\n\t\t\t\treturn { success: false, error: error.message };\n\t\t\t}\n\t\t}`, ctx)\n\n\trequire.NoError(t, err)\n\tresult, ok := res.(map[string]interface{})\n\trequire.True(t, ok, \"Expected map result\")\n\tassert.Equal(t, true, result[\"success\"], \"ReadFile should succeed: %v\", result[\"error\"])\n\tassert.Equal(t, \"Content to read\", result[\"content\"])\n}\n\n// TestJsSandboxListDir tests ctx.sandbox.ListDir via JavaScript\nfunc TestJsSandboxListDir(t *testing.T) {\n\ttest.Prepare(t, config.Conf)\n\tdefer test.Clean()\n\n\tmanager := createTestSandboxManager(t)\n\tif manager == nil {\n\t\treturn\n\t}\n\tdefer manager.Close()\n\n\t// Create container with auto-cleanup\n\tcontainer, cleanup := createTestContainer(t, manager, \"test-user\", \"test-js-listdir\")\n\tdefer cleanup()\n\n\texecutor := &realSandboxExecutor{\n\t\tmanager: manager,\n\t\tcontainerName: container.Name,\n\t\tworkDir: \"/workspace\",\n\t}\n\n\t// Write some files first\n\terr := executor.WriteFile(stdContext.Background(), \"file1.txt\", []byte(\"content1\"))\n\trequire.NoError(t, err)\n\terr = executor.WriteFile(stdContext.Background(), \"file2.txt\", []byte(\"content2\"))\n\trequire.NoError(t, err)\n\n\t// Create context with sandbox\n\tctx := context.New(stdContext.Background(), nil, \"test-chat-listdir\")\n\tctx.AssistantID = \"test-assistant\"\n\tctx.SetSandboxExecutor(executor)\n\n\t// Test ListDir via JavaScript\n\tres, err := v8.Call(v8.CallOptions{}, `\n\t\tfunction test(ctx) {\n\t\t\ttry {\n\t\t\t\tif (!ctx.sandbox) {\n\t\t\t\t\treturn { success: false, error: \"sandbox not available\" };\n\t\t\t\t}\n\t\t\t\t\n\t\t\t\t// List directory\n\t\t\t\tconst files = ctx.sandbox.ListDir(\".\");\n\t\t\t\t\n\t\t\t\t// Find our test files\n\t\t\t\tconst fileNames = files.map(f => f.name);\n\t\t\t\tconst hasFile1 = fileNames.includes(\"file1.txt\");\n\t\t\t\tconst hasFile2 = fileNames.includes(\"file2.txt\");\n\t\t\t\t\n\t\t\t\treturn { \n\t\t\t\t\tsuccess: true, \n\t\t\t\t\tfileCount: files.length,\n\t\t\t\t\thasFile1: hasFile1,\n\t\t\t\t\thasFile2: hasFile2,\n\t\t\t\t\tfiles: fileNames\n\t\t\t\t};\n\t\t\t} catch (error) {\n\t\t\t\treturn { success: false, error: error.message };\n\t\t\t}\n\t\t}`, ctx)\n\n\trequire.NoError(t, err)\n\tresult, ok := res.(map[string]interface{})\n\trequire.True(t, ok, \"Expected map result\")\n\tassert.Equal(t, true, result[\"success\"], \"ListDir should succeed: %v\", result[\"error\"])\n\tassert.Equal(t, true, result[\"hasFile1\"], \"Should find file1.txt\")\n\tassert.Equal(t, true, result[\"hasFile2\"], \"Should find file2.txt\")\n}\n\n// TestJsSandboxExec tests ctx.sandbox.Exec via JavaScript\nfunc TestJsSandboxExec(t *testing.T) {\n\ttest.Prepare(t, config.Conf)\n\tdefer test.Clean()\n\n\tmanager := createTestSandboxManager(t)\n\tif manager == nil {\n\t\treturn\n\t}\n\tdefer manager.Close()\n\n\t// Create container with auto-cleanup\n\tcontainer, cleanup := createTestContainer(t, manager, \"test-user\", \"test-js-exec\")\n\tdefer cleanup()\n\n\texecutor := &realSandboxExecutor{\n\t\tmanager: manager,\n\t\tcontainerName: container.Name,\n\t\tworkDir: \"/workspace\",\n\t}\n\n\t// Create context with sandbox\n\tctx := context.New(stdContext.Background(), nil, \"test-chat-exec\")\n\tctx.AssistantID = \"test-assistant\"\n\tctx.SetSandboxExecutor(executor)\n\n\t// Test Exec via JavaScript\n\tres, err := v8.Call(v8.CallOptions{}, `\n\t\tfunction test(ctx) {\n\t\t\ttry {\n\t\t\t\tif (!ctx.sandbox) {\n\t\t\t\t\treturn { success: false, error: \"sandbox not available\" };\n\t\t\t\t}\n\t\t\t\t\n\t\t\t\t// Execute echo command\n\t\t\t\tconst output = ctx.sandbox.Exec([\"echo\", \"hello-from-js\"]);\n\t\t\t\t\n\t\t\t\treturn { success: true, output: output.trim() };\n\t\t\t} catch (error) {\n\t\t\t\treturn { success: false, error: error.message };\n\t\t\t}\n\t\t}`, ctx)\n\n\trequire.NoError(t, err)\n\tresult, ok := res.(map[string]interface{})\n\trequire.True(t, ok, \"Expected map result\")\n\tassert.Equal(t, true, result[\"success\"], \"Exec should succeed: %v\", result[\"error\"])\n\t// Output may contain Docker stream header bytes, so use Contains\n\toutput, _ := result[\"output\"].(string)\n\tassert.Contains(t, output, \"hello-from-js\", \"Exec output should contain expected text\")\n}\n\n// TestJsSandboxWorkdir tests ctx.sandbox.workdir property via JavaScript\nfunc TestJsSandboxWorkdir(t *testing.T) {\n\ttest.Prepare(t, config.Conf)\n\tdefer test.Clean()\n\n\tmanager := createTestSandboxManager(t)\n\tif manager == nil {\n\t\treturn\n\t}\n\tdefer manager.Close()\n\n\t// Create container with auto-cleanup\n\tcontainer, cleanup := createTestContainer(t, manager, \"test-user\", \"test-js-workdir\")\n\tdefer cleanup()\n\n\texecutor := &realSandboxExecutor{\n\t\tmanager: manager,\n\t\tcontainerName: container.Name,\n\t\tworkDir: \"/workspace\",\n\t}\n\n\t// Create context with sandbox\n\tctx := context.New(stdContext.Background(), nil, \"test-chat-workdir\")\n\tctx.AssistantID = \"test-assistant\"\n\tctx.SetSandboxExecutor(executor)\n\n\t// Test workdir property via JavaScript\n\tres, err := v8.Call(v8.CallOptions{}, `\n\t\tfunction test(ctx) {\n\t\t\ttry {\n\t\t\t\tif (!ctx.sandbox) {\n\t\t\t\t\treturn { success: false, error: \"sandbox not available\" };\n\t\t\t\t}\n\t\t\t\t\n\t\t\t\t// Get workdir property\n\t\t\t\tconst workdir = ctx.sandbox.workdir;\n\t\t\t\t\n\t\t\t\treturn { success: true, workdir: workdir };\n\t\t\t} catch (error) {\n\t\t\t\treturn { success: false, error: error.message };\n\t\t\t}\n\t\t}`, ctx)\n\n\trequire.NoError(t, err)\n\tresult, ok := res.(map[string]interface{})\n\trequire.True(t, ok, \"Expected map result\")\n\tassert.Equal(t, true, result[\"success\"], \"workdir access should succeed: %v\", result[\"error\"])\n\tassert.Equal(t, \"/workspace\", result[\"workdir\"])\n}\n\n// TestJsSandboxCompleteWorkflow tests a complete workflow via JavaScript\nfunc TestJsSandboxCompleteWorkflow(t *testing.T) {\n\ttest.Prepare(t, config.Conf)\n\tdefer test.Clean()\n\n\tmanager := createTestSandboxManager(t)\n\tif manager == nil {\n\t\treturn\n\t}\n\tdefer manager.Close()\n\n\t// Create container with auto-cleanup\n\tcontainer, cleanup := createTestContainer(t, manager, \"test-user\", \"test-js-workflow\")\n\tdefer cleanup()\n\n\texecutor := &realSandboxExecutor{\n\t\tmanager: manager,\n\t\tcontainerName: container.Name,\n\t\tworkDir: \"/workspace\",\n\t}\n\n\t// Create context with sandbox\n\tctx := context.New(stdContext.Background(), nil, \"test-chat-workflow\")\n\tctx.AssistantID = \"test-assistant\"\n\tctx.SetSandboxExecutor(executor)\n\n\t// Test complete workflow: write file, exec cat, verify content\n\tres, err := v8.Call(v8.CallOptions{}, `\n\t\tfunction test(ctx) {\n\t\t\ttry {\n\t\t\t\tif (!ctx.sandbox) {\n\t\t\t\t\treturn { success: false, error: \"sandbox not available\" };\n\t\t\t\t}\n\t\t\t\t\n\t\t\t\t// 1. Check workdir\n\t\t\t\tconst workdir = ctx.sandbox.workdir;\n\t\t\t\tif (workdir !== \"/workspace\") {\n\t\t\t\t\treturn { success: false, error: \"unexpected workdir: \" + workdir };\n\t\t\t\t}\n\t\t\t\t\n\t\t\t\t// 2. Write a file\n\t\t\t\tconst testContent = \"Test workflow content: \" + Date.now();\n\t\t\t\tctx.sandbox.WriteFile(\"workflow-test.txt\", testContent);\n\t\t\t\t\n\t\t\t\t// 3. Read it back\n\t\t\t\tconst readContent = ctx.sandbox.ReadFile(\"workflow-test.txt\");\n\t\t\t\tif (readContent !== testContent) {\n\t\t\t\t\treturn { success: false, error: \"content mismatch after read\" };\n\t\t\t\t}\n\t\t\t\t\n\t\t\t\t// 4. List directory and verify file exists\n\t\t\t\tconst files = ctx.sandbox.ListDir(\".\");\n\t\t\t\tconst fileNames = files.map(f => f.name);\n\t\t\t\tif (!fileNames.includes(\"workflow-test.txt\")) {\n\t\t\t\t\treturn { success: false, error: \"file not found in listing\" };\n\t\t\t\t}\n\t\t\t\t\n\t\t\t\t// 5. Execute cat command\n\t\t\t\tconst catOutput = ctx.sandbox.Exec([\"cat\", workdir + \"/workflow-test.txt\"]);\n\t\t\t\tif (!catOutput.includes(\"Test workflow content\")) {\n\t\t\t\t\treturn { success: false, error: \"cat output mismatch\" };\n\t\t\t\t}\n\t\t\t\t\n\t\t\t\t// 6. Execute pwd command\n\t\t\t\tconst pwdOutput = ctx.sandbox.Exec([\"pwd\"]);\n\t\t\t\tif (!pwdOutput.includes(\"/workspace\")) {\n\t\t\t\t\treturn { success: false, error: \"pwd output mismatch: \" + pwdOutput };\n\t\t\t\t}\n\t\t\t\t\n\t\t\t\treturn { \n\t\t\t\t\tsuccess: true, \n\t\t\t\t\tworkdir: workdir,\n\t\t\t\t\tcontent: readContent,\n\t\t\t\t\tfileCount: files.length\n\t\t\t\t};\n\t\t\t} catch (error) {\n\t\t\t\treturn { success: false, error: error.message, stack: error.stack };\n\t\t\t}\n\t\t}`, ctx)\n\n\trequire.NoError(t, err)\n\tresult, ok := res.(map[string]interface{})\n\trequire.True(t, ok, \"Expected map result\")\n\tassert.Equal(t, true, result[\"success\"], \"Complete workflow should succeed: %v\", result[\"error\"])\n\tassert.Equal(t, \"/workspace\", result[\"workdir\"])\n}\n" }, { "path": "agent/context/jsapi_search.go", "content": "package context\n\nimport (\n\t\"github.com/yaoapp/gou/runtime/v8/bridge\"\n\t\"rogchap.com/v8go\"\n)\n\n// SearchAPI defines the search JSAPI interface for ctx.search.*\n// This interface is defined here to avoid circular dependency between context and search packages.\n// The actual implementation is in agent/search/jsapi.go\ntype SearchAPI interface {\n\t// Web executes web search\n\t// Returns *types.Result or error information\n\tWeb(query string, opts map[string]interface{}) interface{}\n\n\t// KB executes knowledge base search\n\t// Returns *types.Result or error information\n\tKB(query string, opts map[string]interface{}) interface{}\n\n\t// DB executes database search\n\t// Returns *types.Result or error information\n\tDB(query string, opts map[string]interface{}) interface{}\n\n\t// Parallel search methods - inspired by JavaScript Promise\n\t// All waits for all searches to complete (like Promise.all)\n\tAll(requests []interface{}) []interface{}\n\t// Any returns when any search succeeds with results (like Promise.any)\n\tAny(requests []interface{}) []interface{}\n\t// Race returns when any search completes (like Promise.race)\n\tRace(requests []interface{}) []interface{}\n}\n\n// SearchAPIFactory is a function type that creates a SearchAPI for a context\n// This is set by the search package during initialization\nvar SearchAPIFactory func(ctx *Context) SearchAPI\n\n// Search returns the search API for this context\n// Returns nil if SearchAPIFactory is not set\nfunc (ctx *Context) Search() SearchAPI {\n\tif SearchAPIFactory == nil {\n\t\treturn nil\n\t}\n\treturn SearchAPIFactory(ctx)\n}\n\n// newSearchObject creates a new search object with all search methods\n// This is called from jsapi.go NewObject() to mount ctx.search\nfunc (ctx *Context) newSearchObject(iso *v8go.Isolate) *v8go.ObjectTemplate {\n\tsearchObj := v8go.NewObjectTemplate(iso)\n\n\t// Single search methods\n\tsearchObj.Set(\"Web\", ctx.searchWebMethod(iso))\n\tsearchObj.Set(\"KB\", ctx.searchKBMethod(iso))\n\tsearchObj.Set(\"DB\", ctx.searchDBMethod(iso))\n\n\t// Parallel search methods - inspired by JavaScript Promise\n\tsearchObj.Set(\"All\", ctx.searchAllMethod(iso))\n\tsearchObj.Set(\"Any\", ctx.searchAnyMethod(iso))\n\tsearchObj.Set(\"Race\", ctx.searchRaceMethod(iso))\n\n\treturn searchObj\n}\n\n// searchWebMethod implements ctx.search.Web(query, options?)\n// Options:\n// - limit: number - max results (default: 10)\n// - sites: string[] - restrict to specific sites\n// - time_range: string - \"day\", \"week\", \"month\", \"year\"\n// - rerank: { top_n: number } - rerank options\nfunc (ctx *Context) searchWebMethod(iso *v8go.Isolate) *v8go.FunctionTemplate {\n\treturn v8go.NewFunctionTemplate(iso, func(info *v8go.FunctionCallbackInfo) *v8go.Value {\n\t\tv8ctx := info.Context()\n\t\targs := info.Args()\n\n\t\t// Validate arguments\n\t\tif len(args) < 1 {\n\t\t\treturn bridge.JsException(v8ctx, \"Web requires query parameter\")\n\t\t}\n\n\t\t// Get query string\n\t\tif !args[0].IsString() {\n\t\t\treturn bridge.JsException(v8ctx, \"query must be a string\")\n\t\t}\n\t\tquery := args[0].String()\n\n\t\t// Parse options (optional)\n\t\tvar opts map[string]interface{}\n\t\tif len(args) >= 2 && !args[1].IsUndefined() && !args[1].IsNull() {\n\t\t\tgoVal, err := bridge.GoValue(args[1], v8ctx)\n\t\t\tif err != nil {\n\t\t\t\treturn bridge.JsException(v8ctx, \"invalid options: \"+err.Error())\n\t\t\t}\n\t\t\tif optsMap, ok := goVal.(map[string]interface{}); ok {\n\t\t\t\topts = optsMap\n\t\t\t}\n\t\t}\n\n\t\t// Get search API\n\t\tsearchAPI := ctx.Search()\n\t\tif searchAPI == nil {\n\t\t\treturn bridge.JsException(v8ctx, \"search API not available\")\n\t\t}\n\n\t\t// Execute search\n\t\tresult := searchAPI.Web(query, opts)\n\n\t\t// Convert result to JS value\n\t\tjsVal, err := bridge.JsValue(v8ctx, result)\n\t\tif err != nil {\n\t\t\treturn bridge.JsException(v8ctx, \"failed to convert result: \"+err.Error())\n\t\t}\n\n\t\treturn jsVal\n\t})\n}\n\n// searchKBMethod implements ctx.search.KB(query, options?)\n// Options:\n// - collections: string[] - collection IDs\n// - threshold: number - similarity threshold (0-1)\n// - limit: number - max results\n// - graph: boolean - enable graph association\n// - rerank: { top_n: number } - rerank options\nfunc (ctx *Context) searchKBMethod(iso *v8go.Isolate) *v8go.FunctionTemplate {\n\treturn v8go.NewFunctionTemplate(iso, func(info *v8go.FunctionCallbackInfo) *v8go.Value {\n\t\tv8ctx := info.Context()\n\t\targs := info.Args()\n\n\t\t// Validate arguments\n\t\tif len(args) < 1 {\n\t\t\treturn bridge.JsException(v8ctx, \"KB requires query parameter\")\n\t\t}\n\n\t\t// Get query string\n\t\tif !args[0].IsString() {\n\t\t\treturn bridge.JsException(v8ctx, \"query must be a string\")\n\t\t}\n\t\tquery := args[0].String()\n\n\t\t// Parse options (optional)\n\t\tvar opts map[string]interface{}\n\t\tif len(args) >= 2 && !args[1].IsUndefined() && !args[1].IsNull() {\n\t\t\tgoVal, err := bridge.GoValue(args[1], v8ctx)\n\t\t\tif err != nil {\n\t\t\t\treturn bridge.JsException(v8ctx, \"invalid options: \"+err.Error())\n\t\t\t}\n\t\t\tif optsMap, ok := goVal.(map[string]interface{}); ok {\n\t\t\t\topts = optsMap\n\t\t\t}\n\t\t}\n\n\t\t// Get search API\n\t\tsearchAPI := ctx.Search()\n\t\tif searchAPI == nil {\n\t\t\treturn bridge.JsException(v8ctx, \"search API not available\")\n\t\t}\n\n\t\t// Execute search\n\t\tresult := searchAPI.KB(query, opts)\n\n\t\t// Convert result to JS value\n\t\tjsVal, err := bridge.JsValue(v8ctx, result)\n\t\tif err != nil {\n\t\t\treturn bridge.JsException(v8ctx, \"failed to convert result: \"+err.Error())\n\t\t}\n\n\t\treturn jsVal\n\t})\n}\n\n// searchDBMethod implements ctx.search.DB(query, options?)\n// Options:\n// - models: string[] - model IDs\n// - wheres: Where[] - pre-defined filters (GOU QueryDSL Where format)\n// - orders: Order[] - sort orders (GOU QueryDSL Order format)\n// - select: string[] - fields to return\n// - limit: number - max results\n// - rerank: { top_n: number } - rerank options\nfunc (ctx *Context) searchDBMethod(iso *v8go.Isolate) *v8go.FunctionTemplate {\n\treturn v8go.NewFunctionTemplate(iso, func(info *v8go.FunctionCallbackInfo) *v8go.Value {\n\t\tv8ctx := info.Context()\n\t\targs := info.Args()\n\n\t\t// Validate arguments\n\t\tif len(args) < 1 {\n\t\t\treturn bridge.JsException(v8ctx, \"DB requires query parameter\")\n\t\t}\n\n\t\t// Get query string\n\t\tif !args[0].IsString() {\n\t\t\treturn bridge.JsException(v8ctx, \"query must be a string\")\n\t\t}\n\t\tquery := args[0].String()\n\n\t\t// Parse options (optional)\n\t\tvar opts map[string]interface{}\n\t\tif len(args) >= 2 && !args[1].IsUndefined() && !args[1].IsNull() {\n\t\t\tgoVal, err := bridge.GoValue(args[1], v8ctx)\n\t\t\tif err != nil {\n\t\t\t\treturn bridge.JsException(v8ctx, \"invalid options: \"+err.Error())\n\t\t\t}\n\t\t\tif optsMap, ok := goVal.(map[string]interface{}); ok {\n\t\t\t\topts = optsMap\n\t\t\t}\n\t\t}\n\n\t\t// Get search API\n\t\tsearchAPI := ctx.Search()\n\t\tif searchAPI == nil {\n\t\t\treturn bridge.JsException(v8ctx, \"search API not available\")\n\t\t}\n\n\t\t// Execute search\n\t\tresult := searchAPI.DB(query, opts)\n\n\t\t// Convert result to JS value\n\t\tjsVal, err := bridge.JsValue(v8ctx, result)\n\t\tif err != nil {\n\t\t\treturn bridge.JsException(v8ctx, \"failed to convert result: \"+err.Error())\n\t\t}\n\n\t\treturn jsVal\n\t})\n}\n\n// searchAllMethod implements ctx.search.All(requests)\n// Waits for all searches to complete (like Promise.all)\n// Each request should have:\n// - type: string - \"web\", \"kb\", or \"db\"\n// - query: string - search query\n// - ... other type-specific options\nfunc (ctx *Context) searchAllMethod(iso *v8go.Isolate) *v8go.FunctionTemplate {\n\treturn v8go.NewFunctionTemplate(iso, func(info *v8go.FunctionCallbackInfo) *v8go.Value {\n\t\tv8ctx := info.Context()\n\t\targs := info.Args()\n\n\t\t// Validate arguments\n\t\tif len(args) < 1 {\n\t\t\treturn bridge.JsException(v8ctx, \"All requires requests parameter\")\n\t\t}\n\n\t\t// Parse requests array\n\t\tgoVal, err := bridge.GoValue(args[0], v8ctx)\n\t\tif err != nil {\n\t\t\treturn bridge.JsException(v8ctx, \"invalid requests: \"+err.Error())\n\t\t}\n\n\t\trequestsArray, ok := goVal.([]interface{})\n\t\tif !ok {\n\t\t\treturn bridge.JsException(v8ctx, \"requests must be an array\")\n\t\t}\n\n\t\t// Get search API\n\t\tsearchAPI := ctx.Search()\n\t\tif searchAPI == nil {\n\t\t\treturn bridge.JsException(v8ctx, \"search API not available\")\n\t\t}\n\n\t\t// Execute parallel search\n\t\tresults := searchAPI.All(requestsArray)\n\n\t\t// Convert results to JS value\n\t\tjsVal, err := bridge.JsValue(v8ctx, results)\n\t\tif err != nil {\n\t\t\treturn bridge.JsException(v8ctx, \"failed to convert results: \"+err.Error())\n\t\t}\n\n\t\treturn jsVal\n\t})\n}\n\n// searchAnyMethod implements ctx.search.Any(requests)\n// Returns when any search succeeds with results (like Promise.any)\n// Each request should have:\n// - type: string - \"web\", \"kb\", or \"db\"\n// - query: string - search query\n// - ... other type-specific options\nfunc (ctx *Context) searchAnyMethod(iso *v8go.Isolate) *v8go.FunctionTemplate {\n\treturn v8go.NewFunctionTemplate(iso, func(info *v8go.FunctionCallbackInfo) *v8go.Value {\n\t\tv8ctx := info.Context()\n\t\targs := info.Args()\n\n\t\t// Validate arguments\n\t\tif len(args) < 1 {\n\t\t\treturn bridge.JsException(v8ctx, \"Any requires requests parameter\")\n\t\t}\n\n\t\t// Parse requests array\n\t\tgoVal, err := bridge.GoValue(args[0], v8ctx)\n\t\tif err != nil {\n\t\t\treturn bridge.JsException(v8ctx, \"invalid requests: \"+err.Error())\n\t\t}\n\n\t\trequestsArray, ok := goVal.([]interface{})\n\t\tif !ok {\n\t\t\treturn bridge.JsException(v8ctx, \"requests must be an array\")\n\t\t}\n\n\t\t// Get search API\n\t\tsearchAPI := ctx.Search()\n\t\tif searchAPI == nil {\n\t\t\treturn bridge.JsException(v8ctx, \"search API not available\")\n\t\t}\n\n\t\t// Execute parallel search\n\t\tresults := searchAPI.Any(requestsArray)\n\n\t\t// Convert results to JS value\n\t\tjsVal, err := bridge.JsValue(v8ctx, results)\n\t\tif err != nil {\n\t\t\treturn bridge.JsException(v8ctx, \"failed to convert results: \"+err.Error())\n\t\t}\n\n\t\treturn jsVal\n\t})\n}\n\n// searchRaceMethod implements ctx.search.Race(requests)\n// Returns when any search completes (like Promise.race)\n// Each request should have:\n// - type: string - \"web\", \"kb\", or \"db\"\n// - query: string - search query\n// - ... other type-specific options\nfunc (ctx *Context) searchRaceMethod(iso *v8go.Isolate) *v8go.FunctionTemplate {\n\treturn v8go.NewFunctionTemplate(iso, func(info *v8go.FunctionCallbackInfo) *v8go.Value {\n\t\tv8ctx := info.Context()\n\t\targs := info.Args()\n\n\t\t// Validate arguments\n\t\tif len(args) < 1 {\n\t\t\treturn bridge.JsException(v8ctx, \"Race requires requests parameter\")\n\t\t}\n\n\t\t// Parse requests array\n\t\tgoVal, err := bridge.GoValue(args[0], v8ctx)\n\t\tif err != nil {\n\t\t\treturn bridge.JsException(v8ctx, \"invalid requests: \"+err.Error())\n\t\t}\n\n\t\trequestsArray, ok := goVal.([]interface{})\n\t\tif !ok {\n\t\t\treturn bridge.JsException(v8ctx, \"requests must be an array\")\n\t\t}\n\n\t\t// Get search API\n\t\tsearchAPI := ctx.Search()\n\t\tif searchAPI == nil {\n\t\t\treturn bridge.JsException(v8ctx, \"search API not available\")\n\t\t}\n\n\t\t// Execute parallel search\n\t\tresults := searchAPI.Race(requestsArray)\n\n\t\t// Convert results to JS value\n\t\tjsVal, err := bridge.JsValue(v8ctx, results)\n\t\tif err != nil {\n\t\t\treturn bridge.JsException(v8ctx, \"failed to convert results: \"+err.Error())\n\t\t}\n\n\t\treturn jsVal\n\t})\n}\n" }, { "path": "agent/context/jsapi_search_test.go", "content": "package context_test\n\nimport (\n\tstdContext \"context\"\n\t\"encoding/json\"\n\t\"strings\"\n\t\"testing\"\n\n\t\"github.com/stretchr/testify/assert\"\n\t\"github.com/stretchr/testify/require\"\n\t\"github.com/yaoapp/yao/agent/assistant\"\n\t\"github.com/yaoapp/yao/agent/context\"\n\t\"github.com/yaoapp/yao/agent/search/types\"\n\t\"github.com/yaoapp/yao/agent/testutils\"\n\toauthTypes \"github.com/yaoapp/yao/openapi/oauth/types\"\n)\n\n// Note: SearchAPIFactory is set by assistant.init() with proper config getter\n// We import assistant package to ensure init() runs before tests\n\n// newSearchTestContext creates a Context for search JSAPI testing\nfunc newSearchTestContext(chatID, assistantID string) *context.Context {\n\tauthorized := &oauthTypes.AuthorizedInfo{\n\t\tSubject: \"test-user\",\n\t\tClientID: \"test-client-id\",\n\t\tScope: \"openid profile email\",\n\t\tSessionID: \"test-session-id\",\n\t\tUserID: \"test-user-123\",\n\t}\n\n\tctx := context.New(stdContext.Background(), authorized, chatID)\n\tctx.AssistantID = assistantID\n\tctx.Locale = \"en-us\"\n\tctx.Referer = context.RefererAPI\n\tctx.Accept = context.AcceptWebCUI\n\tctx.Metadata = make(map[string]interface{})\n\treturn ctx\n}\n\n// getResponseContent extracts the content from the first assistant message\nfunc getResponseContent(res *context.HookCreateResponse) string {\n\tif res == nil || len(res.Messages) == 0 {\n\t\treturn \"\"\n\t}\n\tfor _, msg := range res.Messages {\n\t\tif msg.Role == \"assistant\" {\n\t\t\tif content, ok := msg.Content.(string); ok {\n\t\t\t\treturn content\n\t\t\t}\n\t\t}\n\t}\n\treturn \"\"\n}\n\n// TestSearchJSAPI_Web tests ctx.search.Web() via Create Hook\n// Skip: requires external API key (Tavily/Serper)\nfunc TestSearchJSAPI_Web(t *testing.T) {\n\tt.Skip(\"Skipping: requires external API key (Tavily/Serper)\")\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\t// Load the search-jsapi test assistant\n\tagent, err := assistant.Get(\"tests.search-jsapi\")\n\trequire.NoError(t, err, \"Failed to get tests.search-jsapi assistant\")\n\trequire.NotNil(t, agent.HookScript, \"The tests.search-jsapi assistant has no script\")\n\n\tctx := newSearchTestContext(\"chat-search-web\", \"tests.search-jsapi\")\n\n\t// Call Create hook with test:web command\n\tres, _, err := agent.HookScript.Create(ctx, []context.Message{{Role: \"user\", Content: \"test:web Yao App Engine\"}})\n\trequire.NoError(t, err, \"Create hook failed\")\n\trequire.NotNil(t, res, \"Expected non-nil response\")\n\n\t// Get response content from messages\n\tcontent := getResponseContent(res)\n\trequire.NotEmpty(t, content, \"Expected response content\")\n\n\t// Parse the JSON response\n\tvar result types.Result\n\terr = json.Unmarshal([]byte(content), &result)\n\trequire.NoError(t, err, \"Response should be valid JSON: %s\", content)\n\n\t// Verify result\n\tassert.Equal(t, types.SearchTypeWeb, result.Type, \"type should be web\")\n\tassert.Equal(t, \"Yao App Engine\", result.Query, \"query should match\")\n\tassert.Empty(t, result.Error, \"should not have error: %s\", result.Error)\n\tassert.Greater(t, len(result.Items), 0, \"should have items\")\n\n\tt.Logf(\"Web search returned %d items\", len(result.Items))\n\tfor i, item := range result.Items {\n\t\tif i < 3 {\n\t\t\tt.Logf(\" [%s] %s - %s\", item.CitationID, item.Title, item.URL)\n\t\t}\n\t}\n}\n\n// TestSearchJSAPI_WebWithSites tests ctx.search.Web() with site restriction\n// Skip: requires external API key (Tavily/Serper)\nfunc TestSearchJSAPI_WebWithSites(t *testing.T) {\n\tt.Skip(\"Skipping: requires external API key (Tavily/Serper)\")\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\tagent, err := assistant.Get(\"tests.search-jsapi\")\n\trequire.NoError(t, err)\n\trequire.NotNil(t, agent.HookScript)\n\n\tctx := newSearchTestContext(\"chat-search-web-sites\", \"tests.search-jsapi\")\n\n\tres, _, err := agent.HookScript.Create(ctx, []context.Message{{Role: \"user\", Content: \"test:web_sites Yao App Engine\"}})\n\trequire.NoError(t, err)\n\trequire.NotNil(t, res)\n\n\tcontent := getResponseContent(res)\n\trequire.NotEmpty(t, content, \"Expected response content\")\n\n\tvar result types.Result\n\terr = json.Unmarshal([]byte(content), &result)\n\trequire.NoError(t, err, \"Response should be valid JSON: %s\", content)\n\n\tassert.Equal(t, types.SearchTypeWeb, result.Type)\n\tassert.Empty(t, result.Error, \"should not have error: %s\", result.Error)\n\tassert.Greater(t, len(result.Items), 0, \"should have items\")\n\n\t// Verify all results are from allowed sites\n\tallowedSites := []string{\"github.com\", \"yaoapps.com\"}\n\tfor _, item := range result.Items {\n\t\tisAllowed := false\n\t\tfor _, site := range allowedSites {\n\t\t\tif strings.Contains(item.URL, site) {\n\t\t\t\tisAllowed = true\n\t\t\t\tbreak\n\t\t\t}\n\t\t}\n\t\tassert.True(t, isAllowed, \"URL %s should be from allowed sites\", item.URL)\n\t}\n\n\tt.Logf(\"Site-restricted search returned %d items\", len(result.Items))\n}\n\n// TestSearchJSAPI_KB tests ctx.search.KB() via Create Hook (skeleton)\nfunc TestSearchJSAPI_KB(t *testing.T) {\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\tagent, err := assistant.Get(\"tests.search-jsapi\")\n\trequire.NoError(t, err)\n\trequire.NotNil(t, agent.HookScript)\n\n\tctx := newSearchTestContext(\"chat-search-kb\", \"tests.search-jsapi\")\n\n\tres, _, err := agent.HookScript.Create(ctx, []context.Message{{Role: \"user\", Content: \"test:kb test query\"}})\n\trequire.NoError(t, err)\n\trequire.NotNil(t, res)\n\n\tcontent := getResponseContent(res)\n\trequire.NotEmpty(t, content, \"Expected response content\")\n\n\tvar result types.Result\n\terr = json.Unmarshal([]byte(content), &result)\n\trequire.NoError(t, err, \"Response should be valid JSON: %s\", content)\n\n\tassert.Equal(t, types.SearchTypeKB, result.Type, \"type should be kb\")\n\tassert.Equal(t, \"test query\", result.Query, \"query should match\")\n\tassert.Equal(t, types.SourceHook, result.Source, \"source should be hook\")\n}\n\n// TestSearchJSAPI_DB tests ctx.search.DB() via Create Hook (skeleton)\nfunc TestSearchJSAPI_DB(t *testing.T) {\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\tagent, err := assistant.Get(\"tests.search-jsapi\")\n\trequire.NoError(t, err)\n\trequire.NotNil(t, agent.HookScript)\n\n\tctx := newSearchTestContext(\"chat-search-db\", \"tests.search-jsapi\")\n\n\tres, _, err := agent.HookScript.Create(ctx, []context.Message{{Role: \"user\", Content: \"test:db test query\"}})\n\trequire.NoError(t, err)\n\trequire.NotNil(t, res)\n\n\tcontent := getResponseContent(res)\n\trequire.NotEmpty(t, content, \"Expected response content\")\n\n\tvar result types.Result\n\terr = json.Unmarshal([]byte(content), &result)\n\trequire.NoError(t, err, \"Response should be valid JSON: %s\", content)\n\n\tassert.Equal(t, types.SearchTypeDB, result.Type, \"type should be db\")\n\tassert.Equal(t, \"test query\", result.Query, \"query should match\")\n\tassert.Equal(t, types.SourceHook, result.Source, \"source should be hook\")\n}\n\n// TestSearchJSAPI_All tests ctx.search.All() via Create Hook\n// Skip: requires external API key (Tavily/Serper)\nfunc TestSearchJSAPI_All(t *testing.T) {\n\tt.Skip(\"Skipping: requires external API key (Tavily/Serper)\")\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\tagent, err := assistant.Get(\"tests.search-jsapi\")\n\trequire.NoError(t, err)\n\trequire.NotNil(t, agent.HookScript)\n\n\tctx := newSearchTestContext(\"chat-search-all\", \"tests.search-jsapi\")\n\n\tres, _, err := agent.HookScript.Create(ctx, []context.Message{{Role: \"user\", Content: \"test:all\"}})\n\trequire.NoError(t, err)\n\trequire.NotNil(t, res)\n\n\tcontent := getResponseContent(res)\n\trequire.NotEmpty(t, content, \"Expected response content\")\n\n\t// Parse as array of results\n\tvar results []*types.Result\n\terr = json.Unmarshal([]byte(content), &results)\n\trequire.NoError(t, err, \"Response should be valid JSON array: %s\", content)\n\n\tassert.Len(t, results, 2, \"should have 2 results\")\n\n\t// Both should succeed\n\tsuccessCount := 0\n\ttotalItems := 0\n\tfor _, r := range results {\n\t\tif r != nil && r.Error == \"\" {\n\t\t\tsuccessCount++\n\t\t\ttotalItems += len(r.Items)\n\t\t}\n\t}\n\n\tassert.Equal(t, 2, successCount, \"both searches should succeed\")\n\tassert.Greater(t, totalItems, 0, \"should have items\")\n\n\tt.Logf(\"All search: %d results, %d total items\", len(results), totalItems)\n}\n\n// TestSearchJSAPI_Any tests ctx.search.Any() via Create Hook\n// Skip: requires external API key (Tavily/Serper)\nfunc TestSearchJSAPI_Any(t *testing.T) {\n\tt.Skip(\"Skipping: requires external API key (Tavily/Serper)\")\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\tagent, err := assistant.Get(\"tests.search-jsapi\")\n\trequire.NoError(t, err)\n\trequire.NotNil(t, agent.HookScript)\n\n\tctx := newSearchTestContext(\"chat-search-any\", \"tests.search-jsapi\")\n\n\tres, _, err := agent.HookScript.Create(ctx, []context.Message{{Role: \"user\", Content: \"test:any\"}})\n\trequire.NoError(t, err)\n\trequire.NotNil(t, res)\n\n\tcontent := getResponseContent(res)\n\trequire.NotEmpty(t, content, \"Expected response content\")\n\n\tvar results []*types.Result\n\terr = json.Unmarshal([]byte(content), &results)\n\trequire.NoError(t, err, \"Response should be valid JSON array: %s\", content)\n\n\tassert.Len(t, results, 2, \"should have 2 result slots\")\n\n\t// At least one should have results\n\thasSuccess := false\n\tfor _, r := range results {\n\t\tif r != nil && len(r.Items) > 0 && r.Error == \"\" {\n\t\t\thasSuccess = true\n\t\t\tbreak\n\t\t}\n\t}\n\tassert.True(t, hasSuccess, \"at least one search should succeed\")\n\n\tt.Logf(\"Any search completed\")\n}\n\n// TestSearchJSAPI_Race tests ctx.search.Race() via Create Hook\n// Skip: requires external API key (Tavily/Serper)\nfunc TestSearchJSAPI_Race(t *testing.T) {\n\tt.Skip(\"Skipping: requires external API key (Tavily/Serper)\")\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\tagent, err := assistant.Get(\"tests.search-jsapi\")\n\trequire.NoError(t, err)\n\trequire.NotNil(t, agent.HookScript)\n\n\tctx := newSearchTestContext(\"chat-search-race\", \"tests.search-jsapi\")\n\n\tres, _, err := agent.HookScript.Create(ctx, []context.Message{{Role: \"user\", Content: \"test:race\"}})\n\trequire.NoError(t, err)\n\trequire.NotNil(t, res)\n\n\tcontent := getResponseContent(res)\n\trequire.NotEmpty(t, content, \"Expected response content\")\n\n\tvar results []*types.Result\n\terr = json.Unmarshal([]byte(content), &results)\n\trequire.NoError(t, err, \"Response should be valid JSON array: %s\", content)\n\n\tassert.Len(t, results, 2, \"should have 2 result slots\")\n\n\t// At least one should have completed\n\thasResult := false\n\tfor _, r := range results {\n\t\tif r != nil {\n\t\t\thasResult = true\n\t\t\tbreak\n\t\t}\n\t}\n\tassert.True(t, hasResult, \"at least one search should complete\")\n\n\tt.Logf(\"Race search completed\")\n}\n\n// TestSearchJSAPI_InvalidCommand tests invalid test command\nfunc TestSearchJSAPI_InvalidCommand(t *testing.T) {\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\tagent, err := assistant.Get(\"tests.search-jsapi\")\n\trequire.NoError(t, err)\n\trequire.NotNil(t, agent.HookScript)\n\n\tctx := newSearchTestContext(\"chat-search-invalid\", \"tests.search-jsapi\")\n\n\tres, _, err := agent.HookScript.Create(ctx, []context.Message{{Role: \"user\", Content: \"invalid command\"}})\n\trequire.NoError(t, err)\n\trequire.NotNil(t, res)\n\n\tcontent := getResponseContent(res)\n\tassert.Contains(t, content, \"Invalid test command\", \"should return error message\")\n}\n\n// TestSearchJSAPI_UnknownMethod tests unknown test method\nfunc TestSearchJSAPI_UnknownMethod(t *testing.T) {\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\tagent, err := assistant.Get(\"tests.search-jsapi\")\n\trequire.NoError(t, err)\n\trequire.NotNil(t, agent.HookScript)\n\n\tctx := newSearchTestContext(\"chat-search-unknown\", \"tests.search-jsapi\")\n\n\tres, _, err := agent.HookScript.Create(ctx, []context.Message{{Role: \"user\", Content: \"test:unknown\"}})\n\trequire.NoError(t, err)\n\trequire.NotNil(t, res)\n\n\tcontent := getResponseContent(res)\n\tassert.Contains(t, content, \"Unknown test method\", \"should return error message\")\n}\n" }, { "path": "agent/context/jsapi_stress_test.go", "content": "package context_test\n\nimport (\n\tstdContext \"context\"\n\t\"fmt\"\n\t\"os\"\n\t\"runtime\"\n\t\"sync\"\n\t\"testing\"\n\t\"time\"\n\n\t\"github.com/stretchr/testify/assert\"\n\tv8 \"github.com/yaoapp/gou/runtime/v8\"\n\t\"github.com/yaoapp/yao/agent/context\"\n\t\"github.com/yaoapp/yao/config\"\n\t\"github.com/yaoapp/yao/test\"\n)\n\n// newStressTestContext creates a test context for stress testing\nfunc newStressTestContext(chatID string) *context.Context {\n\tctx := context.New(stdContext.Background(), nil, chatID)\n\tctx.AssistantID = \"test-assistant\"\n\tctx.Referer = context.RefererAPI\n\tstack, _, _ := context.EnterStack(ctx, \"test-assistant\", &context.Options{})\n\tctx.Stack = stack\n\treturn ctx\n}\n\n// TestStressContextCreationAndRelease tests massive context creation and cleanup\nfunc TestStressContextCreationAndRelease(t *testing.T) {\n\tif testing.Short() {\n\t\tt.Skip(\"Skipping stress test in short mode\")\n\t}\n\n\ttest.Prepare(t, config.Conf)\n\tdefer test.Clean()\n\n\titerations := 1000\n\tstartMemory := getMemStats()\n\n\tfor i := 0; i < iterations; i++ {\n\t\tcxt := newStressTestContext(fmt.Sprintf(\"chat-%d\", i))\n\n\t\t_, err := v8.Call(v8.CallOptions{}, `\n\t\t\tfunction test(ctx) {\n\t\t\t\t// Use trace\n\t\t\t\tctx.trace.Add({ type: \"test\" }, { label: \"Test\" })\n\t\t\t\tctx.trace.Info(\"Processing\")\n\t\t\t\t\n\t\t\t\t// Explicit release\n\t\t\t\tctx.Release()\n\t\t\t\t\n\t\t\t\treturn { iteration: true }\n\t\t\t}`, cxt)\n\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Iteration %d failed: %v\", i, err)\n\t\t}\n\n\t\t// Force GC every 100 iterations to check for leaks\n\t\tif i%100 == 0 {\n\t\t\truntime.GC()\n\t\t\tcurrentMemory := getMemStats()\n\t\t\tt.Logf(\"Iteration %d: Memory usage: %d MB\", i, currentMemory/1024/1024)\n\t\t}\n\t}\n\n\t// Final GC and memory check\n\truntime.GC()\n\ttime.Sleep(100 * time.Millisecond)\n\tendMemory := getMemStats()\n\n\tt.Logf(\"Start memory: %d MB\", startMemory/1024/1024)\n\tt.Logf(\"End memory: %d MB\", endMemory/1024/1024)\n\n\t// Calculate memory growth (handle case where end < start)\n\tvar memoryGrowth int64\n\tif endMemory > startMemory {\n\t\tmemoryGrowth = int64(endMemory - startMemory)\n\t\tt.Logf(\"Memory growth: %d MB\", memoryGrowth/1024/1024)\n\t} else {\n\t\tmemoryGrowth = -int64(startMemory - endMemory)\n\t\tt.Logf(\"Memory decreased: %d MB\", -memoryGrowth/1024/1024)\n\t}\n\n\t// Allow reasonable memory growth (not more than 50MB for 1000 iterations)\n\t// Memory can decrease due to GC, which is fine\n\tif memoryGrowth > 0 {\n\t\tassert.Less(t, memoryGrowth, int64(50*1024*1024), \"Memory leak detected\")\n\t}\n}\n\n// TestStressTraceOperations tests intensive trace operations\nfunc TestStressTraceOperations(t *testing.T) {\n\tif testing.Short() {\n\t\tt.Skip(\"Skipping stress test in short mode\")\n\t}\n\n\ttest.Prepare(t, config.Conf)\n\tdefer test.Clean()\n\n\titerations := 500\n\tnodesPerIteration := 10\n\n\tstartMemory := getMemStats()\n\n\tfor i := 0; i < iterations; i++ {\n\t\t// Create new context for each iteration to avoid context cancellation issues\n\t\tcxt := newStressTestContext(fmt.Sprintf(\"stress-test-chat-%d\", i))\n\t\t_, err := v8.Call(v8.CallOptions{}, fmt.Sprintf(`\n\t\t\tfunction test(ctx) {\n\t\t\t\tconst trace = ctx.trace\n\t\t\t\tconst nodes = []\n\t\t\t\t\n\t\t\t\t// Create multiple nodes\n\t\t\t\tfor (let j = 0; j < %d; j++) {\n\t\t\t\t\tconst node = trace.Add(\n\t\t\t\t\t\t{ type: \"step\", data: \"data-\" + j },\n\t\t\t\t\t\t{ label: \"Step \" + j }\n\t\t\t\t\t)\n\t\t\t\t\tnodes.push(node)\n\t\t\t\t\t\n\t\t\t\t\t// Add logs\n\t\t\t\t\tnode.Info(\"Processing step \" + j)\n\t\t\t\t\tnode.Debug(\"Debug info \" + j)\n\t\t\t\t}\n\t\t\t\t\n\t\t\t\t// Complete all nodes\n\t\t\t\tfor (const node of nodes) {\n\t\t\t\t\tnode.Complete({ result: \"success\" })\n\t\t\t\t}\n\t\t\t\t\n\t\t\t\t// Release resources\n\t\t\t\tctx.Release()\n\t\t\t\t\n\t\t\t\treturn { nodes: nodes.length }\n\t\t\t}`, nodesPerIteration), cxt)\n\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Iteration %d failed: %v\", i, err)\n\t\t}\n\n\t\tif i%50 == 0 {\n\t\t\truntime.GC()\n\t\t\tcurrentMemory := getMemStats()\n\t\t\tt.Logf(\"Iteration %d: Created %d nodes, Memory: %d MB\",\n\t\t\t\ti, i*nodesPerIteration, currentMemory/1024/1024)\n\t\t}\n\t}\n\n\truntime.GC()\n\ttime.Sleep(100 * time.Millisecond)\n\tendMemory := getMemStats()\n\n\tt.Logf(\"Total nodes created: %d\", iterations*nodesPerIteration)\n\tt.Logf(\"Start memory: %d MB\", startMemory/1024/1024)\n\tt.Logf(\"End memory: %d MB\", endMemory/1024/1024)\n\n\tif endMemory > startMemory {\n\t\tt.Logf(\"Memory growth: %d MB\", (endMemory-startMemory)/1024/1024)\n\t} else {\n\t\tt.Logf(\"Memory decreased: %d MB\", (startMemory-endMemory)/1024/1024)\n\t}\n}\n\n// TestStressMCPOperations tests intensive MCP operations\nfunc TestStressMCPOperations(t *testing.T) {\n\tif testing.Short() {\n\t\tt.Skip(\"Skipping stress test in short mode\")\n\t}\n\n\ttest.Prepare(t, config.Conf)\n\tdefer test.Clean()\n\n\titerations := 500\n\n\tcxt := newStressTestContext(\"mcp-stress-test\")\n\n\tstartMemory := getMemStats()\n\n\tfor i := 0; i < iterations; i++ {\n\t\t_, err := v8.Call(v8.CallOptions{}, `\n\t\t\tfunction test(ctx) {\n\t\t\t\t// List operations\n\t\t\t\tconst tools = ctx.mcp.ListTools(\"echo\", \"\")\n\t\t\t\tconst resources = ctx.mcp.ListResources(\"echo\", \"\")\n\t\t\t\tconst prompts = ctx.mcp.ListPrompts(\"echo\", \"\")\n\t\t\t\t\n\t\t\t\t// Call operations\n\t\t\t\tconst result1 = ctx.mcp.CallTool(\"echo\", \"ping\", { count: 1 })\n\t\t\t\tconst result2 = ctx.mcp.CallTool(\"echo\", \"status\", { verbose: false })\n\t\t\t\t\n\t\t\t\t// Read operations\n\t\t\t\tconst info = ctx.mcp.ReadResource(\"echo\", \"echo://info\")\n\t\t\t\t\n\t\t\t\treturn {\n\t\t\t\t\ttools: tools.tools.length,\n\t\t\t\t\tresources: resources.resources.length,\n\t\t\t\t\tprompts: prompts.prompts.length\n\t\t\t\t}\n\t\t\t}`, cxt)\n\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Iteration %d failed: %v\", i, err)\n\t\t}\n\n\t\tif i%50 == 0 {\n\t\t\truntime.GC()\n\t\t\tcurrentMemory := getMemStats()\n\t\t\tt.Logf(\"Iteration %d: Memory: %d MB\", i, currentMemory/1024/1024)\n\t\t}\n\t}\n\n\truntime.GC()\n\ttime.Sleep(100 * time.Millisecond)\n\tendMemory := getMemStats()\n\n\tt.Logf(\"MCP operations: %d iterations\", iterations)\n\tt.Logf(\"Start memory: %d MB\", startMemory/1024/1024)\n\tt.Logf(\"End memory: %d MB\", endMemory/1024/1024)\n\n\tif endMemory > startMemory {\n\t\tt.Logf(\"Memory growth: %d MB\", (endMemory-startMemory)/1024/1024)\n\t} else {\n\t\tt.Logf(\"Memory decreased: %d MB\", (startMemory-endMemory)/1024/1024)\n\t}\n}\n\n// TestStressConcurrentContexts tests concurrent context creation and usage\nfunc TestStressConcurrentContexts(t *testing.T) {\n\tif testing.Short() {\n\t\tt.Skip(\"Skipping stress test in short mode\")\n\t}\n\n\ttest.Prepare(t, config.Conf)\n\tdefer test.Clean()\n\n\tgoroutines := 50\n\titerationsPerGoroutine := 20\n\n\tstartMemory := getMemStats()\n\n\tvar wg sync.WaitGroup\n\terrors := make(chan error, goroutines*iterationsPerGoroutine)\n\n\tfor g := 0; g < goroutines; g++ {\n\t\twg.Add(1)\n\t\tgo func(goroutineID int) {\n\t\t\tdefer wg.Done()\n\n\t\t\tfor i := 0; i < iterationsPerGoroutine; i++ {\n\t\t\t\tcxt := newStressTestContext(fmt.Sprintf(\"chat-%d-%d\", goroutineID, i))\n\n\t\t\t\t_, err := v8.Call(v8.CallOptions{}, `\n\t\t\t\t\tfunction test(ctx) {\n\t\t\t\t\t\t// Use trace\n\t\t\t\t\t\tconst node = ctx.trace.Add({ type: \"test\" }, { label: \"Concurrent Test\" })\n\t\t\t\t\t\tctx.trace.Info(\"Processing concurrent request\")\n\t\t\t\t\t\tnode.Complete({ result: \"success\" })\n\t\t\t\t\t\t\n\t\t\t\t\t\t// Use MCP\n\t\t\t\t\t\tconst tools = ctx.mcp.ListTools(\"echo\", \"\")\n\t\t\t\t\t\t\n\t\t\t\t\t\t// Release resources\n\t\t\t\t\t\tctx.Release()\n\t\t\t\t\t\t\n\t\t\t\t\t\treturn { success: true }\n\t\t\t\t\t}`, cxt)\n\n\t\t\t\tif err != nil {\n\t\t\t\t\terrors <- fmt.Errorf(\"goroutine %d iteration %d: %v\", goroutineID, i, err)\n\t\t\t\t\treturn\n\t\t\t\t}\n\t\t\t}\n\t\t}(g)\n\t}\n\n\twg.Wait()\n\tclose(errors)\n\n\t// Check for errors\n\terrorCount := 0\n\tfor err := range errors {\n\t\tt.Error(err)\n\t\terrorCount++\n\t}\n\n\tassert.Equal(t, 0, errorCount, \"No errors should occur in concurrent operations\")\n\n\truntime.GC()\n\ttime.Sleep(100 * time.Millisecond)\n\tendMemory := getMemStats()\n\n\ttotalOperations := goroutines * iterationsPerGoroutine\n\tt.Logf(\"Total operations: %d (goroutines: %d, iterations: %d)\",\n\t\ttotalOperations, goroutines, iterationsPerGoroutine)\n\tt.Logf(\"Start memory: %d MB\", startMemory/1024/1024)\n\tt.Logf(\"End memory: %d MB\", endMemory/1024/1024)\n\n\tif endMemory > startMemory {\n\t\tt.Logf(\"Memory growth: %d MB\", (endMemory-startMemory)/1024/1024)\n\t} else {\n\t\tt.Logf(\"Memory decreased: %d MB\", (startMemory-endMemory)/1024/1024)\n\t}\n}\n\n// TestStressNoOpTracePerformance tests no-op trace performance\nfunc TestStressNoOpTracePerformance(t *testing.T) {\n\tif testing.Short() {\n\t\tt.Skip(\"Skipping stress test in short mode\")\n\t}\n\n\ttest.Prepare(t, config.Conf)\n\tdefer test.Clean()\n\n\titerations := 1000\n\n\t// Context without trace initialization (no-op trace)\n\tcxt := context.New(stdContext.Background(), nil, \"noop-stress-test\")\n\tcxt.AssistantID = \"test-assistant\"\n\n\tstartMemory := getMemStats()\n\tstartTime := time.Now()\n\n\tfor i := 0; i < iterations; i++ {\n\t\t_, err := v8.Call(v8.CallOptions{}, `\n\t\t\tfunction test(ctx) {\n\t\t\t\tconst trace = ctx.trace // no-op trace\n\t\t\t\t\n\t\t\t\t// All operations should be no-ops and fast\n\t\t\t\ttrace.Info(\"No-op info\")\n\t\t\t\tconst node = trace.Add({ type: \"test\" }, { label: \"No-op\" })\n\t\t\t\tnode.Info(\"No-op node info\")\n\t\t\t\tnode.Complete({ result: \"done\" })\n\t\t\t\ttrace.Release()\n\t\t\t\t\n\t\t\t\treturn { noop: true }\n\t\t\t}`, cxt)\n\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Iteration %d failed: %v\", i, err)\n\t\t}\n\t}\n\n\tduration := time.Since(startTime)\n\truntime.GC()\n\tendMemory := getMemStats()\n\n\tavgTimePerOp := duration / time.Duration(iterations)\n\tt.Logf(\"No-op trace operations: %d iterations\", iterations)\n\tt.Logf(\"Total time: %v\", duration)\n\tt.Logf(\"Average time per operation: %v\", avgTimePerOp)\n\tt.Logf(\"Start memory: %d MB\", startMemory/1024/1024)\n\tt.Logf(\"End memory: %d MB\", endMemory/1024/1024)\n\n\t// No-op operations should be reasonably fast\n\t// Note: CI environments may be slower due to resource limits\n\t// Local: ~2ms, CI: ~10ms\n\tmaxTimePerOp := 5 * time.Millisecond\n\tif os.Getenv(\"CI\") != \"\" || os.Getenv(\"GITHUB_ACTIONS\") != \"\" {\n\t\tmaxTimePerOp = 15 * time.Millisecond // More lenient for CI\n\t}\n\tassert.Less(t, avgTimePerOp, maxTimePerOp, \"No-op operations should be fast\")\n\n\t// No-op operations should not leak memory (< 5MB growth)\n\tif endMemory > startMemory {\n\t\tmemoryGrowth := int64(endMemory - startMemory)\n\t\tassert.Less(t, memoryGrowth, int64(5*1024*1024), \"No-op operations should not leak memory\")\n\t\tt.Logf(\"Memory growth: %d MB\", memoryGrowth/1024/1024)\n\t} else {\n\t\tt.Logf(\"Memory decreased: %d MB\", (startMemory-endMemory)/1024/1024)\n\t}\n}\n\n// TestStressReleasePatterns tests different release patterns\nfunc TestStressReleasePatterns(t *testing.T) {\n\tif testing.Short() {\n\t\tt.Skip(\"Skipping stress test in short mode\")\n\t}\n\n\ttest.Prepare(t, config.Conf)\n\tdefer test.Clean()\n\n\titerations := 200\n\n\tt.Run(\"ManualRelease\", func(t *testing.T) {\n\t\tstartMemory := getMemStats()\n\n\t\tfor i := 0; i < iterations; i++ {\n\t\t\tcxt := newStressTestContext(fmt.Sprintf(\"manual-%d\", i))\n\n\t\t\t_, err := v8.Call(v8.CallOptions{}, `\n\t\t\t\tfunction test(ctx) {\n\t\t\t\t\ttry {\n\t\t\t\t\t\tctx.trace.Add({ type: \"test\" }, { label: \"Manual Release\" })\n\t\t\t\t\t\treturn { success: true }\n\t\t\t\t\t} finally {\n\t\t\t\t\t\tctx.Release() // Manual release\n\t\t\t\t\t}\n\t\t\t\t}`, cxt)\n\n\t\t\tif err != nil {\n\t\t\t\tt.Fatalf(\"Manual release iteration %d failed: %v\", i, err)\n\t\t\t}\n\t\t}\n\n\t\truntime.GC()\n\t\tendMemory := getMemStats()\n\t\tif endMemory > startMemory {\n\t\t\tt.Logf(\"Manual release: Memory growth: %d MB\", (endMemory-startMemory)/1024/1024)\n\t\t} else {\n\t\t\tt.Logf(\"Manual release: Memory decreased: %d MB\", (startMemory-endMemory)/1024/1024)\n\t\t}\n\t})\n\n\tt.Run(\"NoRelease_RelyOnGC\", func(t *testing.T) {\n\t\tstartMemory := getMemStats()\n\n\t\tfor i := 0; i < iterations; i++ {\n\t\t\tcxt := newStressTestContext(fmt.Sprintf(\"gc-%d\", i))\n\n\t\t\t_, err := v8.Call(v8.CallOptions{}, `\n\t\t\t\tfunction test(ctx) {\n\t\t\t\t\tctx.trace.Add({ type: \"test\" }, { label: \"GC Release\" })\n\t\t\t\t\treturn { success: true }\n\t\t\t\t\t// No manual release - rely on GC\n\t\t\t\t}`, cxt)\n\n\t\t\tif err != nil {\n\t\t\t\tt.Fatalf(\"GC release iteration %d failed: %v\", i, err)\n\t\t\t}\n\t\t}\n\n\t\t// Force GC multiple times\n\t\tfor i := 0; i < 3; i++ {\n\t\t\truntime.GC()\n\t\t\ttime.Sleep(50 * time.Millisecond)\n\t\t}\n\n\t\tendMemory := getMemStats()\n\t\tif endMemory > startMemory {\n\t\t\tt.Logf(\"GC release: Memory growth: %d MB\", (endMemory-startMemory)/1024/1024)\n\t\t} else {\n\t\t\tt.Logf(\"GC release: Memory decreased: %d MB\", (startMemory-endMemory)/1024/1024)\n\t\t}\n\t})\n\n\tt.Run(\"SeparateTraceRelease\", func(t *testing.T) {\n\t\tstartMemory := getMemStats()\n\n\t\tfor i := 0; i < iterations; i++ {\n\t\t\tcxt := newStressTestContext(fmt.Sprintf(\"separate-%d\", i))\n\n\t\t\t_, err := v8.Call(v8.CallOptions{}, `\n\t\t\t\tfunction test(ctx) {\n\t\t\t\t\ttry {\n\t\t\t\t\t\tctx.trace.Add({ type: \"test\" }, { label: \"Separate Release\" })\n\t\t\t\t\t\tctx.trace.Release() // Release trace separately\n\t\t\t\t\t\treturn { success: true }\n\t\t\t\t\t} finally {\n\t\t\t\t\t\tctx.Release() // Release context\n\t\t\t\t\t}\n\t\t\t\t}`, cxt)\n\n\t\t\tif err != nil {\n\t\t\t\tt.Fatalf(\"Separate release iteration %d failed: %v\", i, err)\n\t\t\t}\n\t\t}\n\n\t\truntime.GC()\n\t\tendMemory := getMemStats()\n\t\tif endMemory > startMemory {\n\t\t\tt.Logf(\"Separate release: Memory growth: %d MB\", (endMemory-startMemory)/1024/1024)\n\t\t} else {\n\t\t\tt.Logf(\"Separate release: Memory decreased: %d MB\", (startMemory-endMemory)/1024/1024)\n\t\t}\n\t})\n}\n\n// TestStressLongRunningTrace tests long-running trace with many operations\nfunc TestStressLongRunningTrace(t *testing.T) {\n\tif testing.Short() {\n\t\tt.Skip(\"Skipping stress test in short mode\")\n\t}\n\n\ttest.Prepare(t, config.Conf)\n\tdefer test.Clean()\n\n\tcxt := newStressTestContext(\"long-running-test\")\n\n\tstartMemory := getMemStats()\n\toperations := 100\n\n\t_, err := v8.Call(v8.CallOptions{}, fmt.Sprintf(`\n\t\tfunction test(ctx) {\n\t\t\tconst trace = ctx.trace\n\t\t\tconst allNodes = []\n\t\t\t\n\t\t\t// Create many nested nodes\n\t\t\tfor (let i = 0; i < %d; i++) {\n\t\t\t\tconst parentNode = trace.Add(\n\t\t\t\t\t{ type: \"parent\", index: i },\n\t\t\t\t\t{ label: \"Parent \" + i }\n\t\t\t\t)\n\t\t\t\tallNodes.push(parentNode)\n\t\t\t\t\n\t\t\t\t// Create child nodes\n\t\t\t\tfor (let j = 0; j < 5; j++) {\n\t\t\t\t\tconst childNode = parentNode.Add(\n\t\t\t\t\t\t{ type: \"child\", parent: i, index: j },\n\t\t\t\t\t\t{ label: \"Child \" + i + \"-\" + j }\n\t\t\t\t\t)\n\t\t\t\t\tallNodes.push(childNode)\n\t\t\t\t\t\n\t\t\t\t\t// Add logs\n\t\t\t\t\tchildNode.Info(\"Processing child \" + i + \"-\" + j)\n\t\t\t\t\tchildNode.Complete({ result: \"success\" })\n\t\t\t\t}\n\t\t\t\t\n\t\t\t\tparentNode.Complete({ result: \"all children completed\" })\n\t\t\t}\n\t\t\t\n\t\t\t// Release at the end\n\t\t\ttrace.Release()\n\t\t\tctx.Release()\n\t\t\t\n\t\t\treturn { \n\t\t\t\ttotalNodes: allNodes.length,\n\t\t\t\toperations: %d\n\t\t\t}\n\t\t}`, operations, operations), cxt)\n\n\tif err != nil {\n\t\tt.Fatalf(\"Long running trace failed: %v\", err)\n\t}\n\n\truntime.GC()\n\tendMemory := getMemStats()\n\n\texpectedNodes := operations * 6 // parent + 5 children\n\tt.Logf(\"Long-running trace: %d operations, %d nodes\", operations, expectedNodes)\n\tt.Logf(\"Start memory: %d MB\", startMemory/1024/1024)\n\tt.Logf(\"End memory: %d MB\", endMemory/1024/1024)\n\n\tif endMemory > startMemory {\n\t\tt.Logf(\"Memory growth: %d MB\", (endMemory-startMemory)/1024/1024)\n\t} else {\n\t\tt.Logf(\"Memory decreased: %d MB\", (startMemory-endMemory)/1024/1024)\n\t}\n}\n\n// Helper function to get current memory usage\nfunc getMemStats() uint64 {\n\truntime.GC()\n\tvar m runtime.MemStats\n\truntime.ReadMemStats(&m)\n\treturn m.Alloc\n}\n" }, { "path": "agent/context/jsapi_test.go", "content": "package context_test\n\nimport (\n\tstdContext \"context\"\n\t\"fmt\"\n\t\"sync\"\n\t\"testing\"\n\n\t\"github.com/stretchr/testify/assert\"\n\tv8 \"github.com/yaoapp/gou/runtime/v8\"\n\t\"github.com/yaoapp/gou/runtime/v8/bridge\"\n\t\"github.com/yaoapp/yao/agent/context\"\n\t\"github.com/yaoapp/yao/config\"\n\t\"github.com/yaoapp/yao/openapi/oauth/types\"\n\t\"github.com/yaoapp/yao/test\"\n\t\"rogchap.com/v8go\"\n)\n\n// TestJsValue test the JsValue function\nfunc TestJsValue(t *testing.T) {\n\n\ttest.Prepare(t, config.Conf)\n\tdefer test.Clean()\n\n\tcxt := context.New(stdContext.Background(), nil, \"ChatID-123456\")\n\tcxt.AssistantID = \"AssistantID-1234\"\n\n\tv8.RegisterFunction(\"testContextJsvalue\", testContextJsvalueEmbed)\n\tres, err := v8.Call(v8.CallOptions{}, `\n\t\tfunction test(cxt) {\n\t\t\treturn testContextJsvalue(cxt)\n\t\t}`, cxt)\n\tif err != nil {\n\t\tt.Fatalf(\"Call failed: %v\", err)\n\t}\n\tassert.Equal(t, \"ChatID-123456\", res)\n\t// Note: We can't directly check goMaps cleanup as it's in the bridge package\n}\n\nfunc testContextJsvalueEmbed(iso *v8go.Isolate) *v8go.FunctionTemplate {\n\treturn v8go.NewFunctionTemplate(iso, testContextJsvalueFunction)\n}\n\nfunc testContextJsvalueFunction(info *v8go.FunctionCallbackInfo) *v8go.Value {\n\tvar args = info.Args()\n\tif len(args) < 1 {\n\t\treturn bridge.JsException(info.Context(), \"Missing parameters\")\n\t}\n\n\tctx, err := args[0].AsObject()\n\tif err != nil {\n\t\treturn bridge.JsException(info.Context(), err)\n\t}\n\n\tchatID, err := ctx.Get(\"chat_id\")\n\tif err != nil {\n\t\treturn bridge.JsException(info.Context(), err)\n\t}\n\n\treturn chatID\n}\n\n// TestJsValueConcurrent test the JsValue function with concurrent requests\nfunc TestJsValueConcurrent(t *testing.T) {\n\n\ttest.Prepare(t, config.Conf)\n\tdefer test.Clean()\n\n\tv8.RegisterFunction(\"testContextJsvalue\", testContextJsvalueEmbed)\n\n\t// Number of concurrent goroutines\n\tconcurrency := 10\n\titerationsPerGoroutine := 5\n\n\tvar wg sync.WaitGroup\n\terrors := make(chan error, concurrency*iterationsPerGoroutine)\n\tresults := make(chan string, concurrency*iterationsPerGoroutine)\n\n\t// Launch concurrent goroutines\n\tfor i := 0; i < concurrency; i++ {\n\t\twg.Add(1)\n\t\tgo func(routineID int) {\n\t\t\tdefer wg.Done()\n\n\t\t\tfor j := 0; j < iterationsPerGoroutine; j++ {\n\t\t\t\tchatID := fmt.Sprintf(\"ChatID-%d-%d\", routineID, j)\n\t\t\t\tassistantID := fmt.Sprintf(\"AssistantID-%d-%d\", routineID, j)\n\n\t\t\t\tcxt := context.New(stdContext.Background(), nil, chatID)\n\t\t\t\tcxt.AssistantID = assistantID\n\n\t\t\t\tres, err := v8.Call(v8.CallOptions{}, `\n\t\t\t\t\tfunction test(cxt) {\n\t\t\t\t\t\treturn testContextJsvalue(cxt)\n\t\t\t\t\t}`, cxt)\n\n\t\t\t\tif err != nil {\n\t\t\t\t\terrors <- fmt.Errorf(\"routine %d iteration %d failed: %v\", routineID, j, err)\n\t\t\t\t\treturn\n\t\t\t\t}\n\n\t\t\t\tresults <- res.(string)\n\t\t\t}\n\t\t}(i)\n\t}\n\n\t// Wait for all goroutines to complete\n\twg.Wait()\n\tclose(errors)\n\tclose(results)\n\n\t// Check for errors\n\tfor err := range errors {\n\t\tt.Error(err)\n\t}\n\n\t// Verify all results\n\tresultCount := 0\n\tfor res := range results {\n\t\tassert.Contains(t, res, \"ChatID-\")\n\t\tresultCount++\n\t}\n\n\t// Verify the correct number of results\n\texpectedResults := concurrency * iterationsPerGoroutine\n\tassert.Equal(t, expectedResults, resultCount, \"Should have %d results\", expectedResults)\n\n\t// Verify all objects are cleaned up after GC\n\t// Note: objects should be released when v8 values are garbage collected\n\t// Note: We can't directly check goMaps cleanup as it's in the bridge package\n}\n\n// TestJsValueRegistrationAndCleanup test the object registration and cleanup mechanism\nfunc TestJsValueRegistrationAndCleanup(t *testing.T) {\n\n\ttest.Prepare(t, config.Conf)\n\tdefer test.Clean()\n\n\tv8.RegisterFunction(\"testContextRegistration\", testContextRegistrationEmbed)\n\n\t// Create multiple contexts and verify registration\n\tcontextCount := 5\n\tfor i := 0; i < contextCount; i++ {\n\t\tcxt := context.New(stdContext.Background(), nil, fmt.Sprintf(\"ChatID-%d\", i))\n\t\tcxt.AssistantID = fmt.Sprintf(\"AssistantID-%d\", i)\n\n\t\t_, err := v8.Call(v8.CallOptions{}, `\n\t\t\tfunction test(cxt) {\n\t\t\t\treturn testContextRegistration(cxt)\n\t\t\t}`, cxt)\n\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Call %d failed: %v\", i, err)\n\t\t}\n\t}\n\n\t// All objects should be cleaned up after v8.Call completes\n\t// Note: We can't directly check goMaps cleanup as it's in the bridge package\n}\n\nfunc testContextRegistrationEmbed(iso *v8go.Isolate) *v8go.FunctionTemplate {\n\treturn v8go.NewFunctionTemplate(iso, testContextRegistrationFunction)\n}\n\nfunc testContextRegistrationFunction(info *v8go.FunctionCallbackInfo) *v8go.Value {\n\tvar args = info.Args()\n\tif len(args) < 1 {\n\t\treturn bridge.JsException(info.Context(), \"Missing parameters\")\n\t}\n\n\tctx, err := args[0].AsObject()\n\tif err != nil {\n\t\treturn bridge.JsException(info.Context(), err)\n\t}\n\n\t// Verify the object has __release function\n\trelease, err := ctx.Get(\"__release\")\n\tif err != nil {\n\t\treturn bridge.JsException(info.Context(), err)\n\t}\n\n\tif !release.IsFunction() {\n\t\treturn bridge.JsException(info.Context(), fmt.Errorf(\"__release should be a function\"))\n\t}\n\n\t// Verify the object has internal field (goValueID is stored in internal field, not accessible from JS)\n\tif ctx.InternalFieldCount() == 0 {\n\t\treturn bridge.JsException(info.Context(), fmt.Errorf(\"object should have internal field\"))\n\t}\n\n\tgoValueID := ctx.GetInternalField(0)\n\tif goValueID == nil || !goValueID.IsString() {\n\t\treturn bridge.JsException(info.Context(), fmt.Errorf(\"internal field should contain goValueID string\"))\n\t}\n\n\tval, err := v8go.NewValue(info.Context().Isolate(), true)\n\tif err != nil {\n\t\treturn bridge.JsException(info.Context(), err)\n\t}\n\treturn val\n}\n\n// TestJsValueAllFields test that all Context fields are properly exported to JavaScript\nfunc TestJsValueAllFields(t *testing.T) {\n\n\ttest.Prepare(t, config.Conf)\n\tdefer test.Clean()\n\n\tauthInfo := &types.AuthorizedInfo{\n\t\tSubject: \"test-user\",\n\t\tClientID: \"test-client\",\n\t\tUserID: \"user-123\",\n\t\tTeamID: \"team-456\",\n\t\tTenantID: \"tenant-789\",\n\t\tConstraints: types.DataConstraints{\n\t\t\tOwnerOnly: true,\n\t\t\tCreatorOnly: false,\n\t\t\tTeamOnly: true,\n\t\t\tExtra: map[string]interface{}{\n\t\t\t\t\"department\": \"engineering\",\n\t\t\t\t\"region\": \"us-west\",\n\t\t\t},\n\t\t},\n\t}\n\n\tcxt := context.New(stdContext.Background(), authInfo, \"test-chat-id\")\n\tcxt.AssistantID = \"test-assistant-id\"\n\tcxt.Locale = \"zh-cn\"\n\tcxt.Theme = \"dark\"\n\tcxt.Client = context.Client{\n\t\tType: \"web\",\n\t\tUserAgent: \"Mozilla/5.0\",\n\t\tIP: \"127.0.0.1\",\n\t}\n\tcxt.Referer = \"api\"\n\tcxt.Accept = \"cui-web\"\n\tcxt.Route = \"/dashboard/home\"\n\tcxt.Metadata = map[string]interface{}{\n\t\t\"key1\": \"value1\",\n\t\t\"key2\": 123,\n\t\t\"key3\": true,\n\t}\n\n\tv8.RegisterFunction(\"testAllFields\", testAllFieldsEmbed)\n\tres, err := v8.Call(v8.CallOptions{}, `\n\t\tfunction test(cxt) {\n\t\t\treturn testAllFields(cxt)\n\t\t}`, cxt)\n\tif err != nil {\n\t\tt.Fatalf(\"Call failed: %v\", err)\n\t}\n\n\tresult, ok := res.(map[string]interface{})\n\tif !ok {\n\t\tt.Fatalf(\"Expected map result, got %T\", res)\n\t}\n\n\t// Verify all fields\n\tassert.Equal(t, \"test-chat-id\", result[\"chat_id\"], \"chat_id mismatch\")\n\tassert.Equal(t, \"test-assistant-id\", result[\"assistant_id\"], \"assistant_id mismatch\")\n\tassert.Equal(t, \"zh-cn\", result[\"locale\"], \"locale mismatch\")\n\tassert.Equal(t, \"dark\", result[\"theme\"], \"theme mismatch\")\n\tassert.Equal(t, \"api\", result[\"referer\"], \"referer mismatch\")\n\tassert.Equal(t, \"cui-web\", result[\"accept\"], \"accept mismatch\")\n\tassert.Equal(t, \"/dashboard/home\", result[\"route\"], \"route mismatch\")\n\n\t// Verify client object\n\tclient, ok := result[\"client\"].(map[string]interface{})\n\tassert.True(t, ok, \"client should be an object\")\n\tassert.Equal(t, \"web\", client[\"type\"], \"client.type mismatch\")\n\tassert.Equal(t, \"Mozilla/5.0\", client[\"user_agent\"], \"client.user_agent mismatch\")\n\tassert.Equal(t, \"127.0.0.1\", client[\"ip\"], \"client.ip mismatch\")\n\n\t// Verify metadata object\n\tmetadata, ok := result[\"metadata\"].(map[string]interface{})\n\tassert.True(t, ok, \"metadata should be an object\")\n\tassert.Equal(t, \"value1\", metadata[\"key1\"], \"metadata.key1 mismatch\")\n\tassert.Equal(t, float64(123), metadata[\"key2\"], \"metadata.key2 mismatch\")\n\tassert.Equal(t, true, metadata[\"key3\"], \"metadata.key3 mismatch\")\n\n\t// Verify authorized object\n\tauthorized, ok := result[\"authorized\"].(map[string]interface{})\n\tassert.True(t, ok, \"authorized should be an object\")\n\tassert.Equal(t, \"test-user\", authorized[\"sub\"], \"authorized.sub mismatch\")\n\tassert.Equal(t, \"test-client\", authorized[\"client_id\"], \"authorized.client_id mismatch\")\n\tassert.Equal(t, \"user-123\", authorized[\"user_id\"], \"authorized.user_id mismatch\")\n\tassert.Equal(t, \"team-456\", authorized[\"team_id\"], \"authorized.team_id mismatch\")\n\tassert.Equal(t, \"tenant-789\", authorized[\"tenant_id\"], \"authorized.tenant_id mismatch\")\n\n\t// Verify authorized.constraints object\n\tconstraints, ok := authorized[\"constraints\"].(map[string]interface{})\n\tassert.True(t, ok, \"authorized.constraints should be an object\")\n\tassert.Equal(t, true, constraints[\"owner_only\"], \"constraints.owner_only mismatch\")\n\t// creator_only is false, and with omitempty it may not be present\n\tif creatorOnly, exists := constraints[\"creator_only\"]; exists {\n\t\tassert.Equal(t, false, creatorOnly, \"constraints.creator_only mismatch\")\n\t}\n\tassert.Equal(t, true, constraints[\"team_only\"], \"constraints.team_only mismatch\")\n\n\t// Verify constraints.extra object\n\textra, ok := constraints[\"extra\"].(map[string]interface{})\n\tassert.True(t, ok, \"constraints.extra should be an object\")\n\tassert.Equal(t, \"engineering\", extra[\"department\"], \"constraints.extra.department mismatch\")\n\tassert.Equal(t, \"us-west\", extra[\"region\"], \"constraints.extra.region mismatch\")\n\n\t// Note: We can't directly check goMaps cleanup as it's in the bridge package\n}\n\nfunc testAllFieldsEmbed(iso *v8go.Isolate) *v8go.FunctionTemplate {\n\treturn v8go.NewFunctionTemplate(iso, testAllFieldsFunction)\n}\n\nfunc testAllFieldsFunction(info *v8go.FunctionCallbackInfo) *v8go.Value {\n\tvar args = info.Args()\n\tif len(args) < 1 {\n\t\treturn bridge.JsException(info.Context(), \"Missing parameters\")\n\t}\n\n\tctx, err := args[0].AsObject()\n\tif err != nil {\n\t\treturn bridge.JsException(info.Context(), err)\n\t}\n\n\t// Extract all fields and return as a map\n\tresult := map[string]interface{}{}\n\n\t// Helper function to get field value\n\tgetField := func(name string) (interface{}, bool) {\n\t\tval, err := ctx.Get(name)\n\t\tif err != nil || val.IsUndefined() {\n\t\t\treturn nil, false\n\t\t}\n\t\tgoVal, err := bridge.GoValue(val, info.Context())\n\t\tif err != nil {\n\t\t\treturn nil, false\n\t\t}\n\t\treturn goVal, true\n\t}\n\n\tif val, ok := getField(\"chat_id\"); ok {\n\t\tresult[\"chat_id\"] = val\n\t}\n\tif val, ok := getField(\"assistant_id\"); ok {\n\t\tresult[\"assistant_id\"] = val\n\t}\n\tif val, ok := getField(\"locale\"); ok {\n\t\tresult[\"locale\"] = val\n\t}\n\tif val, ok := getField(\"theme\"); ok {\n\t\tresult[\"theme\"] = val\n\t}\n\tif val, ok := getField(\"client\"); ok {\n\t\tresult[\"client\"] = val\n\t}\n\tif val, ok := getField(\"referer\"); ok {\n\t\tresult[\"referer\"] = val\n\t}\n\tif val, ok := getField(\"accept\"); ok {\n\t\tresult[\"accept\"] = val\n\t}\n\tif val, ok := getField(\"route\"); ok {\n\t\tresult[\"route\"] = val\n\t}\n\tif val, ok := getField(\"metadata\"); ok {\n\t\tresult[\"metadata\"] = val\n\t}\n\tif val, ok := getField(\"authorized\"); ok {\n\t\tresult[\"authorized\"] = val\n\t}\n\n\t// Check for deprecated fields - they should NOT exist\n\tif val, ok := getField(\"sid\"); ok {\n\t\tresult[\"sid\"] = val\n\t}\n\tif val, ok := getField(\"silent\"); ok {\n\t\tresult[\"silent\"] = val\n\t}\n\n\tjsVal, err := bridge.JsValue(info.Context(), result)\n\tif err != nil {\n\t\treturn bridge.JsException(info.Context(), err)\n\t}\n\treturn jsVal\n}\n\n// TestJsValueTrace test the Trace method on Context\nfunc TestJsValueTrace(t *testing.T) {\n\n\ttest.Prepare(t, config.Conf)\n\tdefer test.Clean()\n\n\tcxt := context.New(stdContext.Background(), nil, \"test-chat-id\")\n\tcxt.AssistantID = \"test-assistant-id\"\n\tcxt.Stack = &context.Stack{\n\t\tTraceID: \"test-trace-id\",\n\t}\n\n\tres, err := v8.Call(v8.CallOptions{}, `\n\t\tfunction test(cxt) {\n\t\t\t// Get trace from context (property, not method call)\n\t\t\tconst trace = cxt.trace\n\t\t\t\n\t\t\t// Verify trace object exists\n\t\t\tif (!trace) {\n\t\t\t\tthrow new Error(\"Trace returned null or undefined\")\n\t\t\t}\n\t\t\t\n\t\t\t// Verify trace has expected methods\n\t\t\tif (typeof trace.Add !== 'function') {\n\t\t\t\tthrow new Error(\"trace.Add is not a function\")\n\t\t\t}\n\t\t\tif (typeof trace.Info !== 'function') {\n\t\t\t\tthrow new Error(\"trace.Info is not a function\")\n\t\t\t}\n\t\t\t\n\t\t\t// Actually use the trace - add a node\n\t\t\tconst node = trace.Add({ type: \"test\", content: \"Test from context\" }, { label: \"Test Node\" })\n\t\t\t\n\t\t\t// Log some info\n\t\t\ttrace.Info(\"Testing trace from context\")\n\t\t\tnode.Info(\"Node info message\")\n\t\t\t\n\t\t\t// Complete the node\n\t\t\tnode.Complete({ result: \"success\" })\n\t\t\t\n\t\t\t// Return verification info\n\t\t\treturn {\n\t\t\t\ttrace_id: trace.id,\n\t\t\t\tnode_id: node.id,\n\t\t\t\tsuccess: true\n\t\t\t}\n\t\t}`, cxt)\n\tif err != nil {\n\t\tt.Fatalf(\"Call failed: %v\", err)\n\t}\n\n\tresult, ok := res.(map[string]interface{})\n\tif !ok {\n\t\tt.Fatalf(\"Expected map result, got %T\", res)\n\t}\n\n\t// Verify trace was accessible and operations succeeded\n\tassert.Equal(t, \"test-trace-id\", result[\"trace_id\"], \"trace_id should match\")\n\tassert.NotEmpty(t, result[\"node_id\"], \"node_id should not be empty\")\n\tassert.Equal(t, true, result[\"success\"], \"operation should succeed\")\n}\n\n// TestJsValueAuthorizedAndMetadata test the authorized and metadata fields\nfunc TestJsValueAuthorizedAndMetadata(t *testing.T) {\n\n\ttest.Prepare(t, config.Conf)\n\tdefer test.Clean()\n\n\tauthInfo := &types.AuthorizedInfo{\n\t\tUserID: \"user-123\",\n\t\tTenantID: \"tenant-456\",\n\t\tClientID: \"client-789\",\n\t}\n\tcxt := context.New(stdContext.Background(), authInfo, \"test-chat-id\")\n\tcxt.AssistantID = \"test-assistant-id\"\n\tcxt.Metadata = map[string]interface{}{\n\t\t\"request_id\": \"req-001\",\n\t\t\"source\": \"api\",\n\t\t\"version\": \"1.0.0\",\n\t}\n\n\tv8.RegisterFunction(\"testAuthorizedMetadata\", testAuthorizedMetadataEmbed)\n\tres, err := v8.Call(v8.CallOptions{}, `\n\t\tfunction test(cxt) {\n\t\t\treturn testAuthorizedMetadata(cxt)\n\t\t}`, cxt)\n\tif err != nil {\n\t\tt.Fatalf(\"Call failed: %v\", err)\n\t}\n\n\tresult, ok := res.(map[string]interface{})\n\tif !ok {\n\t\tt.Fatalf(\"Expected map result, got %T\", res)\n\t}\n\n\t// Verify authorized object\n\tauthorized, ok := result[\"authorized\"].(map[string]interface{})\n\tassert.True(t, ok, \"authorized should be an object\")\n\tassert.Equal(t, \"user-123\", authorized[\"user_id\"], \"authorized.user_id mismatch\")\n\tassert.Equal(t, \"tenant-456\", authorized[\"tenant_id\"], \"authorized.tenant_id mismatch\")\n\tassert.Equal(t, \"client-789\", authorized[\"client_id\"], \"authorized.client_id mismatch\")\n\n\t// Verify metadata object\n\tmetadata, ok := result[\"metadata\"].(map[string]interface{})\n\tassert.True(t, ok, \"metadata should be an object\")\n\tassert.Equal(t, \"req-001\", metadata[\"request_id\"], \"metadata.request_id mismatch\")\n\tassert.Equal(t, \"api\", metadata[\"source\"], \"metadata.source mismatch\")\n\tassert.Equal(t, \"1.0.0\", metadata[\"version\"], \"metadata.version mismatch\")\n}\n\nfunc testAuthorizedMetadataEmbed(iso *v8go.Isolate) *v8go.FunctionTemplate {\n\treturn v8go.NewFunctionTemplate(iso, testAuthorizedMetadataFunction)\n}\n\nfunc testAuthorizedMetadataFunction(info *v8go.FunctionCallbackInfo) *v8go.Value {\n\tvar args = info.Args()\n\tif len(args) < 1 {\n\t\treturn bridge.JsException(info.Context(), \"Missing parameters\")\n\t}\n\n\tctx, err := args[0].AsObject()\n\tif err != nil {\n\t\treturn bridge.JsException(info.Context(), err)\n\t}\n\n\t// Extract authorized and metadata fields\n\tresult := map[string]interface{}{}\n\n\t// Get authorized\n\tauthorizedVal, err := ctx.Get(\"authorized\")\n\tif err != nil {\n\t\treturn bridge.JsException(info.Context(), err)\n\t}\n\tif !authorizedVal.IsUndefined() && !authorizedVal.IsNull() {\n\t\tauthorized, err := bridge.GoValue(authorizedVal, info.Context())\n\t\tif err != nil {\n\t\t\treturn bridge.JsException(info.Context(), err)\n\t\t}\n\t\tresult[\"authorized\"] = authorized\n\t}\n\n\t// Get metadata\n\tmetadataVal, err := ctx.Get(\"metadata\")\n\tif err != nil {\n\t\treturn bridge.JsException(info.Context(), err)\n\t}\n\tif !metadataVal.IsUndefined() && !metadataVal.IsNull() {\n\t\tmetadata, err := bridge.GoValue(metadataVal, info.Context())\n\t\tif err != nil {\n\t\t\treturn bridge.JsException(info.Context(), err)\n\t\t}\n\t\tresult[\"metadata\"] = metadata\n\t}\n\n\tjsVal, err := bridge.JsValue(info.Context(), result)\n\tif err != nil {\n\t\treturn bridge.JsException(info.Context(), err)\n\t}\n\treturn jsVal\n}\n\n// TestJsValueAuthorizedNil test when authorized is nil\nfunc TestJsValueAuthorizedNil(t *testing.T) {\n\n\ttest.Prepare(t, config.Conf)\n\tdefer test.Clean()\n\n\tcxt := context.New(stdContext.Background(), nil, \"test-chat-id\")\n\tcxt.AssistantID = \"test-assistant-id\"\n\tcxt.Metadata = nil // Explicitly nil (should be empty object)\n\n\tres, err := v8.Call(v8.CallOptions{}, `\n\t\tfunction test(cxt) {\n\t\t\t// Debug: check the actual values\n\t\t\tconst authorized = cxt.authorized;\n\t\t\tconst metadata = cxt.metadata;\n\t\t\t\n\t\t\treturn {\n\t\t\t\tauthorized_type: typeof authorized,\n\t\t\t\tauthorized_is_null: authorized === null,\n\t\t\t\tauthorized_is_undefined: authorized === undefined,\n\t\t\t\tmetadata_type: typeof metadata,\n\t\t\t\tmetadata_is_object: typeof metadata === 'object' && metadata !== null,\n\t\t\t\tmetadata_is_empty: metadata && Object.keys(metadata).length === 0,\n\t\t\t\thas_authorized: 'authorized' in cxt,\n\t\t\t\thas_metadata: 'metadata' in cxt\n\t\t\t}\n\t\t}`, cxt)\n\tif err != nil {\n\t\tt.Fatalf(\"Call failed: %v\", err)\n\t}\n\n\tresult, ok := res.(map[string]interface{})\n\tif !ok {\n\t\tt.Fatalf(\"Expected map result, got %T\", res)\n\t}\n\n\t// Verify authorized exists and is an empty object when nil\n\tassert.Equal(t, true, result[\"has_authorized\"], \"authorized property should exist\")\n\tassert.Equal(t, \"object\", result[\"authorized_type\"], \"authorized should be an object\")\n\tassert.Equal(t, true, result[\"metadata_is_object\"], \"authorized should be an object (not null)\")\n\n\t// Verify metadata is an empty object when not set\n\tassert.Equal(t, true, result[\"has_metadata\"], \"metadata property should exist\")\n\tassert.Equal(t, \"object\", result[\"metadata_type\"], \"metadata should be an object\")\n\tassert.Equal(t, true, result[\"metadata_is_object\"], \"metadata should be an object\")\n\tassert.Equal(t, true, result[\"metadata_is_empty\"], \"metadata should be empty object when not set\")\n}\n" }, { "path": "agent/context/jsapi_workspace.go", "content": "package context\n\nimport (\n\t\"io/fs\"\n\t\"os\"\n\n\t\"github.com/yaoapp/gou/runtime/v8/bridge\"\n\t\"rogchap.com/v8go\"\n)\n\n// createWorkspaceInstance creates the ctx.workspace JavaScript object.\nfunc (ctx *Context) createWorkspaceInstance(v8ctx *v8go.Context) *v8go.Value {\n\tif ctx.workspace == nil {\n\t\treturn nil\n\t}\n\n\tiso := v8ctx.Isolate()\n\tobjTpl := v8go.NewObjectTemplate(iso)\n\n\tobjTpl.Set(\"ReadFile\", ctx.wsReadFileMethod(iso))\n\tobjTpl.Set(\"WriteFile\", ctx.wsWriteFileMethod(iso))\n\tobjTpl.Set(\"ReadDir\", ctx.wsReadDirMethod(iso))\n\tobjTpl.Set(\"MkdirAll\", ctx.wsMkdirAllMethod(iso))\n\tobjTpl.Set(\"Remove\", ctx.wsRemoveMethod(iso))\n\tobjTpl.Set(\"RemoveAll\", ctx.wsRemoveAllMethod(iso))\n\tobjTpl.Set(\"Rename\", ctx.wsRenameMethod(iso))\n\tobjTpl.Set(\"Copy\", ctx.wsCopyMethod(iso))\n\tobjTpl.Set(\"Stat\", ctx.wsStatMethod(iso))\n\tobjTpl.Set(\"Exists\", ctx.wsExistsMethod(iso))\n\n\tinstance, err := objTpl.NewInstance(v8ctx)\n\tif err != nil {\n\t\treturn nil\n\t}\n\treturn instance.Value\n}\n\n// wsReadFileMethod implements ctx.workspace.ReadFile(path)\nfunc (ctx *Context) wsReadFileMethod(iso *v8go.Isolate) *v8go.FunctionTemplate {\n\treturn v8go.NewFunctionTemplate(iso, func(info *v8go.FunctionCallbackInfo) *v8go.Value {\n\t\tv8ctx := info.Context()\n\t\targs := info.Args()\n\n\t\tif ctx.workspace == nil {\n\t\t\treturn bridge.JsException(v8ctx, \"workspace not available\")\n\t\t}\n\t\tif len(args) < 1 {\n\t\t\treturn bridge.JsException(v8ctx, \"ReadFile requires a path argument\")\n\t\t}\n\n\t\tdata, err := ctx.workspace.ReadFile(args[0].String())\n\t\tif err != nil {\n\t\t\treturn bridge.JsException(v8ctx, \"ReadFile failed: \"+err.Error())\n\t\t}\n\n\t\tjsVal, err := v8go.NewValue(iso, string(data))\n\t\tif err != nil {\n\t\t\treturn bridge.JsException(v8ctx, err.Error())\n\t\t}\n\t\treturn jsVal\n\t})\n}\n\n// wsWriteFileMethod implements ctx.workspace.WriteFile(path, content)\nfunc (ctx *Context) wsWriteFileMethod(iso *v8go.Isolate) *v8go.FunctionTemplate {\n\treturn v8go.NewFunctionTemplate(iso, func(info *v8go.FunctionCallbackInfo) *v8go.Value {\n\t\tv8ctx := info.Context()\n\t\targs := info.Args()\n\n\t\tif ctx.workspace == nil {\n\t\t\treturn bridge.JsException(v8ctx, \"workspace not available\")\n\t\t}\n\t\tif len(args) < 2 {\n\t\t\treturn bridge.JsException(v8ctx, \"WriteFile requires path and content arguments\")\n\t\t}\n\n\t\tpath := args[0].String()\n\t\tcontent := args[1].String()\n\n\t\tif err := ctx.workspace.WriteFile(path, []byte(content), 0o644); err != nil {\n\t\t\treturn bridge.JsException(v8ctx, \"WriteFile failed: \"+err.Error())\n\t\t}\n\t\treturn v8go.Undefined(iso)\n\t})\n}\n\n// wsReadDirMethod implements ctx.workspace.ReadDir(path)\nfunc (ctx *Context) wsReadDirMethod(iso *v8go.Isolate) *v8go.FunctionTemplate {\n\treturn v8go.NewFunctionTemplate(iso, func(info *v8go.FunctionCallbackInfo) *v8go.Value {\n\t\tv8ctx := info.Context()\n\t\targs := info.Args()\n\n\t\tif ctx.workspace == nil {\n\t\t\treturn bridge.JsException(v8ctx, \"workspace not available\")\n\t\t}\n\n\t\tpath := \".\"\n\t\tif len(args) >= 1 && args[0].IsString() {\n\t\t\tpath = args[0].String()\n\t\t}\n\n\t\tentries, err := ctx.workspace.ReadDir(path)\n\t\tif err != nil {\n\t\t\treturn bridge.JsException(v8ctx, \"ReadDir failed: \"+err.Error())\n\t\t}\n\n\t\tresult := make([]map[string]interface{}, 0, len(entries))\n\t\tfor _, e := range entries {\n\t\t\tfi, _ := e.Info()\n\t\t\titem := map[string]interface{}{\n\t\t\t\t\"name\": e.Name(),\n\t\t\t\t\"is_dir\": e.IsDir(),\n\t\t\t}\n\t\t\tif fi != nil {\n\t\t\t\titem[\"size\"] = int32(fi.Size())\n\t\t\t}\n\t\t\tresult = append(result, item)\n\t\t}\n\n\t\tjsVal, err := bridge.JsValue(v8ctx, result)\n\t\tif err != nil {\n\t\t\treturn bridge.JsException(v8ctx, err.Error())\n\t\t}\n\t\treturn jsVal\n\t})\n}\n\n// wsMkdirAllMethod implements ctx.workspace.MkdirAll(path)\nfunc (ctx *Context) wsMkdirAllMethod(iso *v8go.Isolate) *v8go.FunctionTemplate {\n\treturn v8go.NewFunctionTemplate(iso, func(info *v8go.FunctionCallbackInfo) *v8go.Value {\n\t\tv8ctx := info.Context()\n\t\targs := info.Args()\n\n\t\tif ctx.workspace == nil {\n\t\t\treturn bridge.JsException(v8ctx, \"workspace not available\")\n\t\t}\n\t\tif len(args) < 1 {\n\t\t\treturn bridge.JsException(v8ctx, \"MkdirAll requires a path argument\")\n\t\t}\n\n\t\tif err := ctx.workspace.MkdirAll(args[0].String(), 0o755); err != nil {\n\t\t\treturn bridge.JsException(v8ctx, \"MkdirAll failed: \"+err.Error())\n\t\t}\n\t\treturn v8go.Undefined(iso)\n\t})\n}\n\n// wsRemoveMethod implements ctx.workspace.Remove(path)\nfunc (ctx *Context) wsRemoveMethod(iso *v8go.Isolate) *v8go.FunctionTemplate {\n\treturn v8go.NewFunctionTemplate(iso, func(info *v8go.FunctionCallbackInfo) *v8go.Value {\n\t\tv8ctx := info.Context()\n\t\targs := info.Args()\n\n\t\tif ctx.workspace == nil {\n\t\t\treturn bridge.JsException(v8ctx, \"workspace not available\")\n\t\t}\n\t\tif len(args) < 1 {\n\t\t\treturn bridge.JsException(v8ctx, \"Remove requires a path argument\")\n\t\t}\n\n\t\tif err := ctx.workspace.Remove(args[0].String()); err != nil {\n\t\t\treturn bridge.JsException(v8ctx, \"Remove failed: \"+err.Error())\n\t\t}\n\t\treturn v8go.Undefined(iso)\n\t})\n}\n\n// wsRemoveAllMethod implements ctx.workspace.RemoveAll(path)\nfunc (ctx *Context) wsRemoveAllMethod(iso *v8go.Isolate) *v8go.FunctionTemplate {\n\treturn v8go.NewFunctionTemplate(iso, func(info *v8go.FunctionCallbackInfo) *v8go.Value {\n\t\tv8ctx := info.Context()\n\t\targs := info.Args()\n\n\t\tif ctx.workspace == nil {\n\t\t\treturn bridge.JsException(v8ctx, \"workspace not available\")\n\t\t}\n\t\tif len(args) < 1 {\n\t\t\treturn bridge.JsException(v8ctx, \"RemoveAll requires a path argument\")\n\t\t}\n\n\t\tif err := ctx.workspace.RemoveAll(args[0].String()); err != nil {\n\t\t\treturn bridge.JsException(v8ctx, \"RemoveAll failed: \"+err.Error())\n\t\t}\n\t\treturn v8go.Undefined(iso)\n\t})\n}\n\n// wsRenameMethod implements ctx.workspace.Rename(oldName, newName)\nfunc (ctx *Context) wsRenameMethod(iso *v8go.Isolate) *v8go.FunctionTemplate {\n\treturn v8go.NewFunctionTemplate(iso, func(info *v8go.FunctionCallbackInfo) *v8go.Value {\n\t\tv8ctx := info.Context()\n\t\targs := info.Args()\n\n\t\tif ctx.workspace == nil {\n\t\t\treturn bridge.JsException(v8ctx, \"workspace not available\")\n\t\t}\n\t\tif len(args) < 2 {\n\t\t\treturn bridge.JsException(v8ctx, \"Rename requires oldName and newName arguments\")\n\t\t}\n\n\t\tif err := ctx.workspace.Rename(args[0].String(), args[1].String()); err != nil {\n\t\t\treturn bridge.JsException(v8ctx, \"Rename failed: \"+err.Error())\n\t\t}\n\t\treturn v8go.Undefined(iso)\n\t})\n}\n\n// wsCopyMethod implements ctx.workspace.Copy(src, dst)\nfunc (ctx *Context) wsCopyMethod(iso *v8go.Isolate) *v8go.FunctionTemplate {\n\treturn v8go.NewFunctionTemplate(iso, func(info *v8go.FunctionCallbackInfo) *v8go.Value {\n\t\tv8ctx := info.Context()\n\t\targs := info.Args()\n\n\t\tif ctx.workspace == nil {\n\t\t\treturn bridge.JsException(v8ctx, \"workspace not available\")\n\t\t}\n\t\tif len(args) < 2 {\n\t\t\treturn bridge.JsException(v8ctx, \"Copy requires src and dst arguments\")\n\t\t}\n\n\t\tif _, err := ctx.workspace.Copy(args[0].String(), args[1].String()); err != nil {\n\t\t\treturn bridge.JsException(v8ctx, \"Copy failed: \"+err.Error())\n\t\t}\n\t\treturn v8go.Undefined(iso)\n\t})\n}\n\n// wsStatMethod implements ctx.workspace.Stat(path)\nfunc (ctx *Context) wsStatMethod(iso *v8go.Isolate) *v8go.FunctionTemplate {\n\treturn v8go.NewFunctionTemplate(iso, func(info *v8go.FunctionCallbackInfo) *v8go.Value {\n\t\tv8ctx := info.Context()\n\t\targs := info.Args()\n\n\t\tif ctx.workspace == nil {\n\t\t\treturn bridge.JsException(v8ctx, \"workspace not available\")\n\t\t}\n\t\tif len(args) < 1 {\n\t\t\treturn bridge.JsException(v8ctx, \"Stat requires a path argument\")\n\t\t}\n\n\t\tfi, err := ctx.workspace.Stat(args[0].String())\n\t\tif err != nil {\n\t\t\treturn bridge.JsException(v8ctx, \"Stat failed: \"+err.Error())\n\t\t}\n\n\t\tresult := map[string]interface{}{\n\t\t\t\"name\": fi.Name(),\n\t\t\t\"size\": int32(fi.Size()),\n\t\t\t\"is_dir\": fi.IsDir(),\n\t\t\t\"mode\": int32(fi.Mode()),\n\t\t\t\"mtime\": fi.ModTime().UnixMilli(),\n\t\t}\n\t\tjsVal, err := bridge.JsValue(v8ctx, result)\n\t\tif err != nil {\n\t\t\treturn bridge.JsException(v8ctx, err.Error())\n\t\t}\n\t\treturn jsVal\n\t})\n}\n\n// wsExistsMethod implements ctx.workspace.Exists(path)\nfunc (ctx *Context) wsExistsMethod(iso *v8go.Isolate) *v8go.FunctionTemplate {\n\treturn v8go.NewFunctionTemplate(iso, func(info *v8go.FunctionCallbackInfo) *v8go.Value {\n\t\tv8ctx := info.Context()\n\t\targs := info.Args()\n\n\t\tif ctx.workspace == nil {\n\t\t\treturn bridge.JsException(v8ctx, \"workspace not available\")\n\t\t}\n\t\tif len(args) < 1 {\n\t\t\treturn bridge.JsException(v8ctx, \"Exists requires a path argument\")\n\t\t}\n\n\t\t_, err := ctx.workspace.Stat(args[0].String())\n\t\texists := err == nil || !isNotExist(err)\n\n\t\tjsVal, _ := v8go.NewValue(iso, exists)\n\t\treturn jsVal\n\t})\n}\n\nfunc isNotExist(err error) bool {\n\tif os.IsNotExist(err) {\n\t\treturn true\n\t}\n\tpathErr, ok := err.(*fs.PathError)\n\tif ok && os.IsNotExist(pathErr.Err) {\n\t\treturn true\n\t}\n\treturn false\n}\n" }, { "path": "agent/context/log.go", "content": "package context\n\nimport (\n\t\"fmt\"\n\t\"strings\"\n\t\"sync\"\n\t\"time\"\n\n\tkunlog \"github.com/yaoapp/kun/log\"\n\t\"github.com/yaoapp/yao/config\"\n)\n\n// =============================================================================\n// ANSI Color Codes\n// =============================================================================\n\nconst (\n\tcolorReset = \"\\033[0m\"\n\tcolorRed = \"\\033[31m\"\n\tcolorGreen = \"\\033[32m\"\n\tcolorYellow = \"\\033[33m\"\n\tcolorBlue = \"\\033[34m\"\n\tcolorMagenta = \"\\033[35m\"\n\tcolorCyan = \"\\033[36m\"\n\tcolorWhite = \"\\033[37m\"\n\tcolorGray = \"\\033[90m\"\n\n\tcolorBoldRed = \"\\033[1;31m\"\n\tcolorBoldGreen = \"\\033[1;32m\"\n\tcolorBoldYellow = \"\\033[1;33m\"\n\tcolorBoldBlue = \"\\033[1;34m\"\n\tcolorBoldMagenta = \"\\033[1;35m\"\n\tcolorBoldCyan = \"\\033[1;36m\"\n)\n\n// =============================================================================\n// Log Level\n// =============================================================================\n\n// LogLevel represents log severity\ntype LogLevel int\n\nconst (\n\t// LogLevelTrace represents the most verbose logging level for detailed tracing\n\tLogLevelTrace LogLevel = iota\n\t// LogLevelDebug represents debug level logging for development diagnostics\n\tLogLevelDebug\n\t// LogLevelInfo represents informational messages for normal operation\n\tLogLevelInfo\n\t// LogLevelWarn represents warning messages for potentially harmful situations\n\tLogLevelWarn\n\t// LogLevelError represents error messages for serious problems\n\tLogLevelError\n)\n\n// =============================================================================\n// Log Entry\n// =============================================================================\n\n// LogEntry represents a single log entry\ntype LogEntry struct {\n\tLevel LogLevel\n\tMessage string\n\tTimestamp time.Time\n\tPhase string // For phase logging\n\tElapsed time.Duration\n}\n\n// =============================================================================\n// Request Logger\n// =============================================================================\n\n// RequestLogger provides request-scoped async logging\ntype RequestLogger struct {\n\tassistantIDStack []string // Stack-based: delegate calls push, pop on exit; top = current\n\tchatID string\n\trequestID string\n\tshortID string // Short version of requestID for display\n\tparentID string // Parent request ID for A2A tree structure\n\tstartTime time.Time\n\n\tch chan LogEntry\n\tdone chan struct{}\n\tonce sync.Once\n\tclosed bool\n\tnoop bool // noop logger does nothing (for nil safety)\n\tmu sync.RWMutex\n}\n\n// LoggerOption configures a RequestLogger\ntype LoggerOption func(*RequestLogger)\n\n// WithParentID sets the parent request ID for A2A tree structure\nfunc WithParentID(parentID string) LoggerOption {\n\treturn func(l *RequestLogger) {\n\t\tl.parentID = parentID\n\t}\n}\n\n// noopLogger is a shared no-op logger instance\nvar noopLogger = &RequestLogger{noop: true}\n\n// NewRequestLogger creates a new request-scoped logger with async processing\nfunc NewRequestLogger(assistantID, chatID, requestID string, opts ...LoggerOption) *RequestLogger {\n\tl := &RequestLogger{\n\t\tassistantIDStack: []string{assistantID},\n\t\tchatID: chatID,\n\t\trequestID: requestID,\n\t\tshortID: shortID(requestID),\n\t\tstartTime: time.Now(),\n\t\tch: make(chan LogEntry, 100), // Buffered channel\n\t\tdone: make(chan struct{}),\n\t}\n\n\tfor _, opt := range opts {\n\t\topt(l)\n\t}\n\n\t// Start consumer goroutine\n\tgo l.consume()\n\n\treturn l\n}\n\n// Noop returns a no-op logger that does nothing (nil-safe)\nfunc Noop() *RequestLogger {\n\treturn noopLogger\n}\n\n// SetAssistantID pushes a new assistant ID onto the stack (called when entering Stream).\n// Each SetAssistantID must be paired with a RestoreAssistantID on exit.\nfunc (l *RequestLogger) SetAssistantID(id string) {\n\tif l.noop {\n\t\treturn\n\t}\n\tl.mu.Lock()\n\tl.assistantIDStack = append(l.assistantIDStack, id)\n\tl.mu.Unlock()\n}\n\n// RestoreAssistantID pops the current assistant ID, reverting to the previous one.\n// Safe to call even if the stack has only one entry (the initial ID is never removed).\nfunc (l *RequestLogger) RestoreAssistantID() {\n\tif l.noop {\n\t\treturn\n\t}\n\tl.mu.Lock()\n\tif len(l.assistantIDStack) > 1 {\n\t\tl.assistantIDStack = l.assistantIDStack[:len(l.assistantIDStack)-1]\n\t}\n\tl.mu.Unlock()\n}\n\nfunc (l *RequestLogger) currentAssistantID() string {\n\tif len(l.assistantIDStack) == 0 {\n\t\treturn \"\"\n\t}\n\treturn l.assistantIDStack[len(l.assistantIDStack)-1]\n}\n\n// Close closes the logger and waits for all entries to be processed\nfunc (l *RequestLogger) Close() {\n\tif l.noop {\n\t\treturn\n\t}\n\tl.once.Do(func() {\n\t\tl.mu.Lock()\n\t\tl.closed = true\n\t\tl.mu.Unlock()\n\n\t\tclose(l.ch)\n\t\t<-l.done // Wait for consumer to finish\n\t})\n}\n\n// consume processes log entries from the channel\nfunc (l *RequestLogger) consume() {\n\tdefer close(l.done)\n\n\tfor entry := range l.ch {\n\t\tl.processEntry(entry)\n\t}\n}\n\n// processEntry handles a single log entry based on mode\nfunc (l *RequestLogger) processEntry(entry LogEntry) {\n\tif config.IsDevelopment() {\n\t\tl.printDev(entry)\n\t\tl.writeLog(entry, true)\n\t} else {\n\t\tl.writeLog(entry, false)\n\t}\n}\n\n// printDev prints colored output to stdout in development mode\nfunc (l *RequestLogger) printDev(entry LogEntry) {\n\tswitch entry.Level {\n\tcase LogLevelTrace:\n\t\tfmt.Printf(\"%s → %s%s\\n\", colorGray, entry.Message, colorReset)\n\tcase LogLevelDebug:\n\t\tfmt.Printf(\"%s • %s%s\\n\", colorGray, entry.Message, colorReset)\n\tcase LogLevelInfo:\n\t\tfmt.Printf(\"%s ℹ %s%s\\n\", colorCyan, entry.Message, colorReset)\n\tcase LogLevelWarn:\n\t\tfmt.Printf(\"%s ⚠ %s%s\\n\", colorYellow, entry.Message, colorReset)\n\tcase LogLevelError:\n\t\tfmt.Printf(\"%s ✗ %s%s\\n\", colorRed, entry.Message, colorReset)\n\t}\n}\n\n// writeLog writes structured events to kun/log\nfunc (l *RequestLogger) writeLog(entry LogEntry, devMode bool) {\n\tprefix := fmt.Sprintf(\"[AGENT] %s \", l.shortID)\n\tif devMode {\n\t\tkunlog.Trace(\"%s%s\", prefix, entry.Message)\n\t\treturn\n\t}\n\tswitch entry.Level {\n\tcase LogLevelTrace:\n\t\tkunlog.Trace(\"%s%s\", prefix, entry.Message)\n\tcase LogLevelDebug:\n\t\t// Skip debug in production\n\tcase LogLevelInfo:\n\t\tkunlog.Info(\"%s%s\", prefix, entry.Message)\n\tcase LogLevelWarn:\n\t\tkunlog.Warn(\"%s%s\", prefix, entry.Message)\n\tcase LogLevelError:\n\t\tkunlog.Error(\"%s%s\", prefix, entry.Message)\n\t}\n}\n\n// send sends an entry to the channel (non-blocking if closed)\nfunc (l *RequestLogger) send(entry LogEntry) {\n\tif l.noop {\n\t\treturn\n\t}\n\n\tl.mu.RLock()\n\tclosed := l.closed\n\tl.mu.RUnlock()\n\n\tif closed {\n\t\treturn\n\t}\n\n\tentry.Timestamp = time.Now()\n\tselect {\n\tcase l.ch <- entry:\n\tdefault:\n\t\t// Channel full, drop the log (shouldn't happen with buffered channel)\n\t}\n}\n\n// =============================================================================\n// Standard Log Interface\n// =============================================================================\n\n// Trace logs a trace level message\nfunc (l *RequestLogger) Trace(format string, args ...interface{}) {\n\tl.send(LogEntry{\n\t\tLevel: LogLevelTrace,\n\t\tMessage: fmt.Sprintf(format, args...),\n\t})\n}\n\n// Debug logs a debug level message\nfunc (l *RequestLogger) Debug(format string, args ...interface{}) {\n\tl.send(LogEntry{\n\t\tLevel: LogLevelDebug,\n\t\tMessage: fmt.Sprintf(format, args...),\n\t})\n}\n\n// Info logs an info level message\nfunc (l *RequestLogger) Info(format string, args ...interface{}) {\n\tl.send(LogEntry{\n\t\tLevel: LogLevelInfo,\n\t\tMessage: fmt.Sprintf(format, args...),\n\t})\n}\n\n// Warn logs a warning level message\nfunc (l *RequestLogger) Warn(format string, args ...interface{}) {\n\tl.send(LogEntry{\n\t\tLevel: LogLevelWarn,\n\t\tMessage: fmt.Sprintf(format, args...),\n\t})\n}\n\n// Error logs an error level message\nfunc (l *RequestLogger) Error(format string, args ...interface{}) {\n\tl.send(LogEntry{\n\t\tLevel: LogLevelError,\n\t\tMessage: fmt.Sprintf(format, args...),\n\t})\n}\n\n// =============================================================================\n// Business Quick Functions\n// =============================================================================\n\n// Start logs the start of a request with visual separator\nfunc (l *RequestLogger) Start() {\n\tif l.noop {\n\t\treturn\n\t}\n\n\tkunlog.Trace(\"[AGENT] Request %s started: assistant=%s, chat=%s, request=%s\",\n\t\tl.shortID, l.currentAssistantID(), shortID(l.chatID), shortID(l.requestID))\n\n\tif !config.IsDevelopment() {\n\t\treturn\n\t}\n\n\tfmt.Println()\n\tfmt.Printf(\"%s%s%s\\n\", colorBoldCyan, strings.Repeat(\"═\", 60), colorReset)\n\tfmt.Printf(\"%s AGENT REQUEST %s%s\\n\", colorBoldCyan, l.shortID, colorReset)\n\tfmt.Printf(\"%s%s%s\\n\", colorBoldCyan, strings.Repeat(\"─\", 60), colorReset)\n\tfmt.Printf(\"%s Assistant: %s%s%s\\n\", colorGray, colorWhite, l.currentAssistantID(), colorReset)\n\tfmt.Printf(\"%s Chat ID: %s%s%s\\n\", colorGray, colorWhite, l.chatID, colorReset)\n\tfmt.Printf(\"%s Request: %s%s%s\\n\", colorGray, colorWhite, l.requestID, colorReset)\n\tfmt.Printf(\"%s Time: %s%s%s\\n\", colorGray, colorWhite, l.startTime.Format(\"15:04:05.000\"), colorReset)\n\tfmt.Printf(\"%s%s%s\\n\", colorCyan, strings.Repeat(\"─\", 60), colorReset)\n}\n\n// End logs the end of a request with summary\nfunc (l *RequestLogger) End(success bool, err error) {\n\tif l.noop {\n\t\treturn\n\t}\n\n\tduration := time.Since(l.startTime)\n\n\tif success {\n\t\tkunlog.Trace(\"[AGENT] Request %s completed: assistant=%s, duration=%v\",\n\t\t\tl.shortID, l.currentAssistantID(), duration.Round(time.Millisecond))\n\t} else {\n\t\tkunlog.Error(\"[AGENT] Request %s failed: assistant=%s, duration=%v, error=%v\",\n\t\t\tl.shortID, l.currentAssistantID(), duration.Round(time.Millisecond), err)\n\t}\n\n\tif !config.IsDevelopment() {\n\t\treturn\n\t}\n\n\tfmt.Printf(\"%s%s%s\\n\", colorCyan, strings.Repeat(\"─\", 60), colorReset)\n\tif success {\n\t\tfmt.Printf(\"%s REQUEST %s COMPLETED%s\\n\", colorBoldGreen, l.shortID, colorReset)\n\t} else {\n\t\tfmt.Printf(\"%s REQUEST %s FAILED%s\\n\", colorBoldRed, l.shortID, colorReset)\n\t\tif err != nil {\n\t\t\tfmt.Printf(\"%s Error: %s%v%s\\n\", colorGray, colorRed, err, colorReset)\n\t\t}\n\t}\n\tfmt.Printf(\"%s Assistant: %s%s%s\\n\", colorGray, colorWhite, l.currentAssistantID(), colorReset)\n\tfmt.Printf(\"%s Duration: %s%v%s\\n\", colorGray, colorWhite, duration.Round(time.Millisecond), colorReset)\n\tfmt.Printf(\"%s%s%s\\n\", colorCyan, strings.Repeat(\"─\", 60), colorReset)\n\tfmt.Println()\n}\n\n// Phase logs a major phase in the request lifecycle\nfunc (l *RequestLogger) Phase(name string) {\n\tif l.noop {\n\t\treturn\n\t}\n\n\telapsed := time.Since(l.startTime).Round(time.Millisecond)\n\tkunlog.Trace(\"[AGENT] %s Phase: %s (+%v)\", l.shortID, name, elapsed)\n\n\tif !config.IsDevelopment() {\n\t\treturn\n\t}\n\n\tfmt.Printf(\"%s > %s%s %s[+%v]%s\\n\", colorBoldBlue, name, colorReset, colorGray, elapsed, colorReset)\n}\n\n// PhaseComplete logs the completion of a phase\nfunc (l *RequestLogger) PhaseComplete(name string) {\n\tif l.noop {\n\t\treturn\n\t}\n\n\telapsed := time.Since(l.startTime).Round(time.Millisecond)\n\tkunlog.Trace(\"[AGENT] %s Phase completed: %s (+%v)\", l.shortID, name, elapsed)\n\n\tif !config.IsDevelopment() {\n\t\treturn\n\t}\n\n\tfmt.Printf(\"%s + %s%s %s[+%v]%s\\n\", colorGreen, name, colorReset, colorGray, elapsed, colorReset)\n}\n\n// PhaseSkip logs a skipped phase (development only)\nfunc (l *RequestLogger) PhaseSkip(name, reason string) {\n\tif l.noop {\n\t\treturn\n\t}\n\n\tif !config.IsDevelopment() {\n\t\treturn\n\t}\n\n\tfmt.Printf(\"%s - %s (%s)%s\\n\", colorGray, name, reason, colorReset)\n}\n\n// LLMStart logs the start of an LLM call\nfunc (l *RequestLogger) LLMStart(connector, model string, messageCount int) {\n\tif l.noop {\n\t\treturn\n\t}\n\n\telapsed := time.Since(l.startTime).Round(time.Millisecond)\n\tkunlog.Trace(\"[AGENT] %s LLM call: connector=%s, model=%s, messages=%d (+%v)\", l.shortID, connector, model, messageCount, elapsed)\n\n\tif !config.IsDevelopment() {\n\t\treturn\n\t}\n\n\tfmt.Printf(\"%s LLM Call%s %s[+%v]%s\\n\", colorBoldMagenta, colorReset, colorGray, elapsed, colorReset)\n\tfmt.Printf(\"%s Connector: %s%s%s\\n\", colorGray, colorWhite, connector, colorReset)\n\tif model != \"\" {\n\t\tfmt.Printf(\"%s Model: %s%s%s\\n\", colorGray, colorWhite, model, colorReset)\n\t}\n\tfmt.Printf(\"%s Messages: %s%d%s\\n\", colorGray, colorWhite, messageCount, colorReset)\n}\n\n// LLMComplete logs the completion of an LLM call\nfunc (l *RequestLogger) LLMComplete(tokens int, hasToolCalls bool) {\n\tif l.noop {\n\t\treturn\n\t}\n\n\telapsed := time.Since(l.startTime).Round(time.Millisecond)\n\tstatus := \"streaming\"\n\tif hasToolCalls {\n\t\tstatus = \"tool_calls\"\n\t}\n\n\tkunlog.Trace(\"[AGENT] %s LLM response: status=%s, tokens=%d (+%v)\", l.shortID, status, tokens, elapsed)\n\n\tif !config.IsDevelopment() {\n\t\treturn\n\t}\n\n\tfmt.Printf(\"%s + LLM Response (%s)%s\", colorGreen, status, colorReset)\n\tif tokens > 0 {\n\t\tfmt.Printf(\" %s[tokens: %d]%s\", colorGray, tokens, colorReset)\n\t}\n\tfmt.Printf(\" %s[+%v]%s\\n\", colorGray, elapsed, colorReset)\n}\n\n// ToolStart logs the start of tool execution\nfunc (l *RequestLogger) ToolStart(toolName string) {\n\tif l.noop {\n\t\treturn\n\t}\n\n\tkunlog.Trace(\"[AGENT] %s Tool call: %s\", l.shortID, toolName)\n\n\tif !config.IsDevelopment() {\n\t\treturn\n\t}\n\n\tfmt.Printf(\"%s Tool: %s%s\\n\", colorYellow, toolName, colorReset)\n}\n\n// ToolComplete logs the completion of tool execution\nfunc (l *RequestLogger) ToolComplete(toolName string, success bool) {\n\tif l.noop {\n\t\treturn\n\t}\n\n\tif success {\n\t\tkunlog.Trace(\"[AGENT] %s Tool completed: %s\", l.shortID, toolName)\n\t} else {\n\t\tkunlog.Error(\"[AGENT] %s Tool failed: %s\", l.shortID, toolName)\n\t}\n\n\tif !config.IsDevelopment() {\n\t\treturn\n\t}\n\n\tif success {\n\t\tfmt.Printf(\"%s + %s completed%s\\n\", colorGreen, toolName, colorReset)\n\t} else {\n\t\tfmt.Printf(\"%s x %s failed%s\\n\", colorRed, toolName, colorReset)\n\t}\n}\n\n// HookStart logs the start of a hook execution\nfunc (l *RequestLogger) HookStart(hookName string) {\n\tif l.noop {\n\t\treturn\n\t}\n\n\telapsed := time.Since(l.startTime).Round(time.Millisecond)\n\tkunlog.Trace(\"[AGENT] %s Hook: %s (+%v)\", l.shortID, hookName, elapsed)\n\n\tif !config.IsDevelopment() {\n\t\treturn\n\t}\n\n\tfmt.Printf(\"%s Hook: %s%s %s[+%v]%s\\n\", colorMagenta, hookName, colorReset, colorGray, elapsed, colorReset)\n}\n\n// HookComplete logs the completion of a hook\nfunc (l *RequestLogger) HookComplete(hookName string) {\n\tif l.noop {\n\t\treturn\n\t}\n\n\tkunlog.Trace(\"[AGENT] %s Hook completed: %s\", l.shortID, hookName)\n\n\tif !config.IsDevelopment() {\n\t\treturn\n\t}\n\n\tfmt.Printf(\"%s + %s done%s\\n\", colorGreen, hookName, colorReset)\n}\n\n// Cleanup logs resource cleanup\nfunc (l *RequestLogger) Cleanup(resource string) {\n\tif l.noop {\n\t\treturn\n\t}\n\n\tkunlog.Trace(\"[AGENT] %s Cleanup: %s\", l.shortID, resource)\n\n\tif !config.IsDevelopment() {\n\t\treturn\n\t}\n\tfmt.Printf(\"%s + %s%s\\n\", colorGray, resource, colorReset)\n}\n\n// HistoryLoad logs history loading\nfunc (l *RequestLogger) HistoryLoad(count, maxSize int) {\n\tif l.noop {\n\t\treturn\n\t}\n\n\tkunlog.Trace(\"[AGENT] %s History loaded: %d/%d messages\", l.shortID, count, maxSize)\n\n\tif !config.IsDevelopment() {\n\t\treturn\n\t}\n\tfmt.Printf(\"%s Loaded %d/%d history messages%s\\n\", colorGray, count, maxSize, colorReset)\n}\n\n// HistoryOverlap logs overlap detection\nfunc (l *RequestLogger) HistoryOverlap(overlapCount int) {\n\tif l.noop {\n\t\treturn\n\t}\n\n\tif overlapCount > 0 {\n\t\tkunlog.Trace(\"[AGENT] %s History overlap removed: %d messages\", l.shortID, overlapCount)\n\n\t\tif !config.IsDevelopment() {\n\t\t\treturn\n\t\t}\n\t\tfmt.Printf(\"%s Removed %d overlapping messages%s\\n\", colorYellow, overlapCount, colorReset)\n\t}\n}\n\n// Release logs the start of resource release phase\nfunc (l *RequestLogger) Release() {\n\tif l.noop {\n\t\treturn\n\t}\n\n\tkunlog.Trace(\"[AGENT] %s Release started\", l.shortID)\n\n\tif !config.IsDevelopment() {\n\t\treturn\n\t}\n\n\tfmt.Printf(\"%s RELEASE %s%s %s(%s)%s\\n\", colorBoldYellow, l.shortID, colorReset, colorGray, l.currentAssistantID(), colorReset)\n}\n\n// =============================================================================\n// Helper\n// =============================================================================\n\n// shortID returns first 8 characters of an ID\nfunc shortID(id string) string {\n\tif len(id) > 8 {\n\t\treturn id[:8]\n\t}\n\treturn id\n}\n" }, { "path": "agent/context/mcp.go", "content": "package context\n\nimport (\n\t\"fmt\"\n\n\tjsoniter \"github.com/json-iterator/go\"\n\t\"github.com/yaoapp/gou/mcp\"\n\t\"github.com/yaoapp/gou/mcp/types\"\n\t\"github.com/yaoapp/yao/agent/i18n\"\n\ttraceTypes \"github.com/yaoapp/yao/trace/types\"\n)\n\n// MCP Client Operations with automatic trace logging and resource management\n\n// Resource Operations\n// ==================\n\n// ListResources lists all available resources from an MCP client\n// Automatically creates trace node and handles client lifecycle\nfunc (ctx *Context) ListResources(mcpID string, cursor string) (*types.ListResourcesResponse, error) {\n\t// Get MCP client\n\tclient, err := mcp.Select(mcpID)\n\tif err != nil {\n\t\treturn nil, fmt.Errorf(\"failed to select MCP client '%s': %w\", mcpID, err)\n\t}\n\n\t// Get client label for display\n\tclientLabel := client.GetMetaInfo().Label\n\tif clientLabel == \"\" {\n\t\tclientLabel = mcpID\n\t}\n\n\t// Get trace manager\n\ttrace, _ := ctx.Trace()\n\n\t// Create trace node\n\tvar node traceTypes.Node\n\tif trace != nil {\n\t\tnode, _ = trace.Add(\n\t\t\tmap[string]any{\n\t\t\t\t\"mcp\": mcpID,\n\t\t\t\t\"cursor\": cursor,\n\t\t\t},\n\t\t\ttraceTypes.TraceNodeOption{\n\t\t\t\tLabel: i18n.T(ctx.Locale, \"mcp.list_resources.label\"), // \"MCP: List Resources\"\n\t\t\t\tType: \"mcp\",\n\t\t\t\tIcon: \"list\",\n\t\t\t\tDescription: fmt.Sprintf(i18n.T(ctx.Locale, \"mcp.list_resources.description\"), clientLabel), // \"List resources from MCP client '%s'\"\n\t\t\t},\n\t\t)\n\t}\n\n\t// Call ListResources\n\tresult, err := client.ListResources(ctx.Context, cursor)\n\tif err != nil {\n\t\tif node != nil {\n\t\t\tnode.Fail(err)\n\t\t}\n\t\treturn nil, err\n\t}\n\n\t// Complete trace node with result\n\tif node != nil {\n\t\tnode.Complete(map[string]any{\n\t\t\t\"resources\": len(result.Resources),\n\t\t\t\"nextCursor\": result.NextCursor,\n\t\t})\n\t}\n\n\treturn result, nil\n}\n\n// ReadResource reads a specific resource from an MCP client\n// Automatically creates trace node and handles client lifecycle\nfunc (ctx *Context) ReadResource(mcpID string, uri string) (*types.ReadResourceResponse, error) {\n\t// Get MCP client\n\tclient, err := mcp.Select(mcpID)\n\tif err != nil {\n\t\treturn nil, fmt.Errorf(\"failed to select MCP client '%s': %w\", mcpID, err)\n\t}\n\n\t// Get client label for display\n\tclientLabel := client.GetMetaInfo().Label\n\tif clientLabel == \"\" {\n\t\tclientLabel = mcpID\n\t}\n\n\t// Get trace manager\n\ttrace, _ := ctx.Trace()\n\n\t// Create trace node\n\tvar node traceTypes.Node\n\tif trace != nil {\n\t\tnode, _ = trace.Add(\n\t\t\tmap[string]any{\n\t\t\t\t\"mcp\": mcpID,\n\t\t\t\t\"uri\": uri,\n\t\t\t},\n\t\t\ttraceTypes.TraceNodeOption{\n\t\t\t\tLabel: i18n.T(ctx.Locale, \"mcp.read_resource.label\"), // \"MCP: Read Resource\"\n\t\t\t\tType: \"mcp\",\n\t\t\t\tIcon: \"description\",\n\t\t\t\tDescription: fmt.Sprintf(i18n.T(ctx.Locale, \"mcp.read_resource.description\"), uri, clientLabel), // \"Read resource '%s' from MCP client '%s'\"\n\t\t\t},\n\t\t)\n\t}\n\n\t// Call ReadResource\n\tresult, err := client.ReadResource(ctx.Context, uri)\n\tif err != nil {\n\t\tif node != nil {\n\t\t\tnode.Fail(err)\n\t\t}\n\t\treturn nil, err\n\t}\n\n\t// Complete trace node with result\n\tif node != nil {\n\t\tnode.Complete(map[string]any{\n\t\t\t\"contents\": len(result.Contents),\n\t\t})\n\t}\n\n\treturn result, nil\n}\n\n// Tool Operations\n// ===============\n\n// ListTools lists all available tools from an MCP client\n// Automatically creates trace node and handles client lifecycle\nfunc (ctx *Context) ListTools(mcpID string, cursor string) (*types.ListToolsResponse, error) {\n\t// Get MCP client\n\tclient, err := mcp.Select(mcpID)\n\tif err != nil {\n\t\treturn nil, fmt.Errorf(\"failed to select MCP client '%s': %w\", mcpID, err)\n\t}\n\n\t// Get client label for display\n\tclientLabel := client.GetMetaInfo().Label\n\tif clientLabel == \"\" {\n\t\tclientLabel = mcpID\n\t}\n\n\t// Get trace manager\n\ttrace, _ := ctx.Trace()\n\n\t// Create trace node\n\tvar node traceTypes.Node\n\tif trace != nil {\n\t\tnode, _ = trace.Add(\n\t\t\tmap[string]any{\n\t\t\t\t\"mcp\": mcpID,\n\t\t\t\t\"cursor\": cursor,\n\t\t\t},\n\t\t\ttraceTypes.TraceNodeOption{\n\t\t\t\tLabel: i18n.T(ctx.Locale, \"mcp.list_tools.label\"), // \"MCP: List Tools\"\n\t\t\t\tType: \"mcp\",\n\t\t\t\tIcon: \"build\",\n\t\t\t\tDescription: fmt.Sprintf(i18n.T(ctx.Locale, \"mcp.list_tools.description\"), clientLabel), // \"List tools from MCP client '%s'\"\n\t\t\t},\n\t\t)\n\t}\n\n\t// Call ListTools\n\tresult, err := client.ListTools(ctx.Context, cursor)\n\tif err != nil {\n\t\tif node != nil {\n\t\t\tnode.Fail(err)\n\t\t}\n\t\treturn nil, err\n\t}\n\n\t// Complete trace node with result\n\tif node != nil {\n\t\tnode.Complete(map[string]any{\n\t\t\t\"tools\": len(result.Tools),\n\t\t\t\"nextCursor\": result.NextCursor,\n\t\t})\n\t}\n\n\treturn result, nil\n}\n\n// CallTool calls a single tool from an MCP client\n// Automatically creates trace node and handles client lifecycle\nfunc (ctx *Context) CallTool(mcpID string, name string, arguments interface{}) (*types.CallToolResponse, error) {\n\t// Get MCP client\n\tclient, err := mcp.Select(mcpID)\n\tif err != nil {\n\t\treturn nil, fmt.Errorf(\"failed to select MCP client '%s': %w\", mcpID, err)\n\t}\n\n\t// Get client label for display\n\tclientLabel := client.GetMetaInfo().Label\n\tif clientLabel == \"\" {\n\t\tclientLabel = mcpID\n\t}\n\n\t// Get trace manager\n\ttrace, _ := ctx.Trace()\n\n\t// Create trace node\n\tvar node traceTypes.Node\n\tif trace != nil {\n\t\tnode, _ = trace.Add(\n\t\t\tmap[string]any{\n\t\t\t\t\"mcp\": mcpID,\n\t\t\t\t\"tool\": name,\n\t\t\t\t\"arguments\": arguments,\n\t\t\t},\n\t\t\ttraceTypes.TraceNodeOption{\n\t\t\t\tLabel: i18n.T(ctx.Locale, \"mcp.call_tool.label\"), // \"MCP: Call Tool\"\n\t\t\t\tType: \"mcp\",\n\t\t\t\tIcon: \"settings\",\n\t\t\t\tDescription: fmt.Sprintf(i18n.T(ctx.Locale, \"mcp.call_tool.description\"), name, clientLabel), // \"Call tool '%s' from MCP client '%s'\"\n\t\t\t},\n\t\t)\n\t}\n\n\t// Call tool (pass ctx as extraArgs for Process transport to propagate Authorized())\n\tresult, err := client.CallTool(ctx.Context, name, arguments, ctx)\n\tif err != nil {\n\t\tif node != nil {\n\t\t\tnode.Fail(err)\n\t\t}\n\t\treturn nil, err\n\t}\n\n\t// Complete trace node with result\n\tif node != nil {\n\t\tnode.Complete(map[string]any{\n\t\t\t\"contents\": len(result.Content),\n\t\t})\n\t}\n\n\treturn result, nil\n}\n\n// CallTools calls multiple tools sequentially from an MCP client\n// Automatically creates trace node and handles client lifecycle\nfunc (ctx *Context) CallTools(mcpID string, tools []types.ToolCall) (*types.CallToolsResponse, error) {\n\t// Get MCP client\n\tclient, err := mcp.Select(mcpID)\n\tif err != nil {\n\t\treturn nil, fmt.Errorf(\"failed to select MCP client '%s': %w\", mcpID, err)\n\t}\n\n\t// Get client label for display\n\tclientLabel := client.GetMetaInfo().Label\n\tif clientLabel == \"\" {\n\t\tclientLabel = mcpID\n\t}\n\n\t// Get trace manager\n\ttrace, _ := ctx.Trace()\n\n\t// Create trace node\n\tvar node traceTypes.Node\n\tif trace != nil {\n\t\tnode, _ = trace.Add(\n\t\t\tmap[string]any{\n\t\t\t\t\"mcp\": mcpID,\n\t\t\t\t\"tools\": tools,\n\t\t\t\t\"count\": len(tools),\n\t\t\t},\n\t\t\ttraceTypes.TraceNodeOption{\n\t\t\t\tLabel: i18n.T(ctx.Locale, \"mcp.call_tools.label\"), // \"MCP: Call Tools\"\n\t\t\t\tType: \"mcp\",\n\t\t\t\tIcon: \"settings\",\n\t\t\t\tDescription: fmt.Sprintf(i18n.T(ctx.Locale, \"mcp.call_tools.description\"), len(tools), clientLabel), // \"Call %d tools sequentially from MCP client '%s'\"\n\t\t\t},\n\t\t)\n\t}\n\n\t// Call tools sequentially (pass ctx as extraArgs for Process transport to propagate Authorized())\n\tresult, err := client.CallTools(ctx.Context, tools, ctx)\n\tif err != nil {\n\t\tif node != nil {\n\t\t\tnode.Fail(err)\n\t\t}\n\t\treturn nil, err\n\t}\n\n\t// Complete trace node with result\n\tif node != nil {\n\t\tnode.Complete(map[string]any{\n\t\t\t\"results\": len(result.Results),\n\t\t})\n\t}\n\n\treturn result, nil\n}\n\n// CallToolsParallel calls multiple tools in parallel from an MCP client\n// Automatically creates trace node and handles client lifecycle\nfunc (ctx *Context) CallToolsParallel(mcpID string, tools []types.ToolCall) (*types.CallToolsResponse, error) {\n\t// Get MCP client\n\tclient, err := mcp.Select(mcpID)\n\tif err != nil {\n\t\treturn nil, fmt.Errorf(\"failed to select MCP client '%s': %w\", mcpID, err)\n\t}\n\n\t// Get client label for display\n\tclientLabel := client.GetMetaInfo().Label\n\tif clientLabel == \"\" {\n\t\tclientLabel = mcpID\n\t}\n\n\t// Get trace manager\n\ttrace, _ := ctx.Trace()\n\n\t// Create trace node\n\tvar node traceTypes.Node\n\tif trace != nil {\n\t\tnode, _ = trace.Add(\n\t\t\tmap[string]any{\n\t\t\t\t\"mcp\": mcpID,\n\t\t\t\t\"tools\": tools,\n\t\t\t\t\"count\": len(tools),\n\t\t\t},\n\t\t\ttraceTypes.TraceNodeOption{\n\t\t\t\tLabel: i18n.T(ctx.Locale, \"mcp.call_tools_parallel.label\"), // \"MCP: Call Tools (Parallel)\"\n\t\t\t\tType: \"mcp\",\n\t\t\t\tIcon: \"settings\",\n\t\t\t\tDescription: fmt.Sprintf(i18n.T(ctx.Locale, \"mcp.call_tools_parallel.description\"), len(tools), clientLabel), // \"Call %d tools in parallel from MCP client '%s'\"\n\t\t\t},\n\t\t)\n\t}\n\n\t// Call tools in parallel (pass ctx as extraArgs for Process transport to propagate Authorized())\n\tresult, err := client.CallToolsParallel(ctx.Context, tools, ctx)\n\tif err != nil {\n\t\tif node != nil {\n\t\t\tnode.Fail(err)\n\t\t}\n\t\treturn nil, err\n\t}\n\n\t// Complete trace node with result\n\tif node != nil {\n\t\tnode.Complete(map[string]any{\n\t\t\t\"results\": len(result.Results),\n\t\t})\n\t}\n\n\treturn result, nil\n}\n\n// Prompt Operations\n// =================\n\n// ListPrompts lists all available prompts from an MCP client\n// Automatically creates trace node and handles client lifecycle\nfunc (ctx *Context) ListPrompts(mcpID string, cursor string) (*types.ListPromptsResponse, error) {\n\t// Get MCP client\n\tclient, err := mcp.Select(mcpID)\n\tif err != nil {\n\t\treturn nil, fmt.Errorf(\"failed to select MCP client '%s': %w\", mcpID, err)\n\t}\n\n\t// Get client label for display\n\tclientLabel := client.GetMetaInfo().Label\n\tif clientLabel == \"\" {\n\t\tclientLabel = mcpID\n\t}\n\n\t// Get trace manager\n\ttrace, _ := ctx.Trace()\n\n\t// Create trace node\n\tvar node traceTypes.Node\n\tif trace != nil {\n\t\tnode, _ = trace.Add(\n\t\t\tmap[string]any{\n\t\t\t\t\"mcp\": mcpID,\n\t\t\t\t\"cursor\": cursor,\n\t\t\t},\n\t\t\ttraceTypes.TraceNodeOption{\n\t\t\t\tLabel: i18n.T(ctx.Locale, \"mcp.list_prompts.label\"), // \"MCP: List Prompts\"\n\t\t\t\tType: \"mcp\",\n\t\t\t\tIcon: \"chat\",\n\t\t\t\tDescription: fmt.Sprintf(i18n.T(ctx.Locale, \"mcp.list_prompts.description\"), clientLabel), // \"List prompts from MCP client '%s'\"\n\t\t\t},\n\t\t)\n\t}\n\n\t// Call ListPrompts\n\tresult, err := client.ListPrompts(ctx.Context, cursor)\n\tif err != nil {\n\t\tif node != nil {\n\t\t\tnode.Fail(err)\n\t\t}\n\t\treturn nil, err\n\t}\n\n\t// Complete trace node with result\n\tif node != nil {\n\t\tnode.Complete(map[string]any{\n\t\t\t\"prompts\": len(result.Prompts),\n\t\t\t\"nextCursor\": result.NextCursor,\n\t\t})\n\t}\n\n\treturn result, nil\n}\n\n// GetPrompt gets a prompt with arguments from an MCP client\n// Automatically creates trace node and handles client lifecycle\nfunc (ctx *Context) GetPrompt(mcpID string, name string, arguments map[string]interface{}) (*types.GetPromptResponse, error) {\n\t// Get MCP client\n\tclient, err := mcp.Select(mcpID)\n\tif err != nil {\n\t\treturn nil, fmt.Errorf(\"failed to select MCP client '%s': %w\", mcpID, err)\n\t}\n\n\t// Get client label for display\n\tclientLabel := client.GetMetaInfo().Label\n\tif clientLabel == \"\" {\n\t\tclientLabel = mcpID\n\t}\n\n\t// Get trace manager\n\ttrace, _ := ctx.Trace()\n\n\t// Create trace node\n\tvar node traceTypes.Node\n\tif trace != nil {\n\t\tnode, _ = trace.Add(\n\t\t\tmap[string]any{\n\t\t\t\t\"mcp\": mcpID,\n\t\t\t\t\"prompt\": name,\n\t\t\t\t\"arguments\": arguments,\n\t\t\t},\n\t\t\ttraceTypes.TraceNodeOption{\n\t\t\t\tLabel: i18n.T(ctx.Locale, \"mcp.get_prompt.label\"), // \"MCP: Get Prompt\"\n\t\t\t\tType: \"mcp\",\n\t\t\t\tIcon: \"chat\",\n\t\t\t\tDescription: fmt.Sprintf(i18n.T(ctx.Locale, \"mcp.get_prompt.description\"), name, clientLabel), // \"Get prompt '%s' from MCP client '%s'\"\n\t\t\t},\n\t\t)\n\t}\n\n\t// Get prompt\n\tresult, err := client.GetPrompt(ctx.Context, name, arguments)\n\tif err != nil {\n\t\tif node != nil {\n\t\t\tnode.Fail(err)\n\t\t}\n\t\treturn nil, err\n\t}\n\n\t// Complete trace node with result\n\tif node != nil {\n\t\tnode.Complete(map[string]any{\n\t\t\t\"messages\": len(result.Messages),\n\t\t})\n\t}\n\n\treturn result, nil\n}\n\n// Sample Operations\n// =================\n\n// ListSamples lists samples for a tool or resource from an MCP client\n// Automatically creates trace node and handles client lifecycle\nfunc (ctx *Context) ListSamples(mcpID string, itemType types.SampleItemType, itemName string) (*types.ListSamplesResponse, error) {\n\t// Get MCP client\n\tclient, err := mcp.Select(mcpID)\n\tif err != nil {\n\t\treturn nil, fmt.Errorf(\"failed to select MCP client '%s': %w\", mcpID, err)\n\t}\n\n\t// Get client label for display\n\tclientLabel := client.GetMetaInfo().Label\n\tif clientLabel == \"\" {\n\t\tclientLabel = mcpID\n\t}\n\n\t// Get trace manager\n\ttrace, _ := ctx.Trace()\n\n\t// Create trace node\n\tvar node traceTypes.Node\n\tif trace != nil {\n\t\tnode, _ = trace.Add(\n\t\t\tmap[string]any{\n\t\t\t\t\"mcp\": mcpID,\n\t\t\t\t\"itemType\": itemType,\n\t\t\t\t\"itemName\": itemName,\n\t\t\t},\n\t\t\ttraceTypes.TraceNodeOption{\n\t\t\t\tLabel: i18n.T(ctx.Locale, \"mcp.list_samples.label\"), // \"MCP: List Samples\"\n\t\t\t\tType: \"mcp\",\n\t\t\t\tIcon: \"library_books\",\n\t\t\t\tDescription: fmt.Sprintf(i18n.T(ctx.Locale, \"mcp.list_samples.description\"), itemName, clientLabel), // \"List samples for '%s' from MCP client '%s'\"\n\t\t\t},\n\t\t)\n\t}\n\n\t// Call ListSamples\n\tresult, err := client.ListSamples(ctx.Context, itemType, itemName)\n\tif err != nil {\n\t\tif node != nil {\n\t\t\tnode.Fail(err)\n\t\t}\n\t\treturn nil, err\n\t}\n\n\t// Complete trace node with result\n\tif node != nil {\n\t\tnode.Complete(map[string]any{\n\t\t\t\"samples\": len(result.Samples),\n\t\t})\n\t}\n\n\treturn result, nil\n}\n\n// GetSample gets a specific sample by index from an MCP client\n// Automatically creates trace node and handles client lifecycle\nfunc (ctx *Context) GetSample(mcpID string, itemType types.SampleItemType, itemName string, index int) (*types.SampleData, error) {\n\t// Get MCP client\n\tclient, err := mcp.Select(mcpID)\n\tif err != nil {\n\t\treturn nil, fmt.Errorf(\"failed to select MCP client '%s': %w\", mcpID, err)\n\t}\n\n\t// Get client label for display\n\tclientLabel := client.GetMetaInfo().Label\n\tif clientLabel == \"\" {\n\t\tclientLabel = mcpID\n\t}\n\n\t// Get trace manager\n\ttrace, _ := ctx.Trace()\n\n\t// Create trace node\n\tvar node traceTypes.Node\n\tif trace != nil {\n\t\tnode, _ = trace.Add(\n\t\t\tmap[string]any{\n\t\t\t\t\"mcp\": mcpID,\n\t\t\t\t\"itemType\": itemType,\n\t\t\t\t\"itemName\": itemName,\n\t\t\t\t\"index\": index,\n\t\t\t},\n\t\t\ttraceTypes.TraceNodeOption{\n\t\t\t\tLabel: i18n.T(ctx.Locale, \"mcp.get_sample.label\"), // \"MCP: Get Sample\"\n\t\t\t\tType: \"mcp\",\n\t\t\t\tIcon: \"library_books\",\n\t\t\t\tDescription: fmt.Sprintf(i18n.T(ctx.Locale, \"mcp.get_sample.description\"), index, itemName, clientLabel), // \"Get sample #%d for '%s' from MCP client '%s'\"\n\t\t\t},\n\t\t)\n\t}\n\n\t// Get sample\n\tresult, err := client.GetSample(ctx.Context, itemType, itemName, index)\n\tif err != nil {\n\t\tif node != nil {\n\t\t\tnode.Fail(err)\n\t\t}\n\t\treturn nil, err\n\t}\n\n\t// Complete trace node with result\n\tif node != nil {\n\t\tnode.Complete(result)\n\t}\n\n\treturn result, nil\n}\n\n// Single-Server Tool Response Helpers\n// ====================================\n\n// parseCallToolResponse parses a CallToolResponse and returns the parsed content directly\nfunc parseCallToolResponse(response *types.CallToolResponse) interface{} {\n\tif response == nil {\n\t\treturn nil\n\t}\n\treturn parseToolResponseContent(response)\n}\n\n// parseCallToolsResponse parses a CallToolsResponse and returns an array of parsed results\nfunc parseCallToolsResponse(response *types.CallToolsResponse) []interface{} {\n\tif response == nil {\n\t\treturn nil\n\t}\n\tresults := make([]interface{}, len(response.Results))\n\tfor i, r := range response.Results {\n\t\tresults[i] = parseToolResponseContent(&r)\n\t}\n\treturn results\n}\n\n// Cross-Server Tool Operations\n// ============================\n\n// MCPToolRequest represents a request to call a tool on a specific MCP server\ntype MCPToolRequest struct {\n\tMCP string `json:\"mcp\"` // MCP server ID\n\tTool string `json:\"tool\"` // Tool name\n\tArguments interface{} `json:\"arguments\"` // Tool arguments\n}\n\n// MCPToolResult represents the result of a cross-server tool call\n// Returns parsed result directly, with error field for failures\ntype MCPToolResult struct {\n\tMCP string `json:\"mcp\"` // MCP server ID\n\tTool string `json:\"tool\"` // Tool name\n\tResult interface{} `json:\"result,omitempty\"` // Parsed result content (directly usable)\n\tError string `json:\"error,omitempty\"` // Error message (on failure)\n}\n\n// callToolResult is used internally to pass results through channels\ntype callToolResult struct {\n\tidx int\n\tresult *MCPToolResult\n}\n\n// CallToolAll calls tools on multiple MCP servers concurrently and waits for all to complete\n// Returns results in the same order as requests, regardless of completion order (like Promise.all)\nfunc (ctx *Context) CallToolAll(requests []*MCPToolRequest) []*MCPToolResult {\n\tif len(requests) == 0 {\n\t\treturn []*MCPToolResult{}\n\t}\n\n\tresults := make([]*MCPToolResult, len(requests))\n\tdone := make(chan struct{})\n\tremaining := len(requests)\n\n\tfor i, req := range requests {\n\t\tgo func(idx int, r *MCPToolRequest) {\n\t\t\tdefer func() {\n\t\t\t\tif err := recover(); err != nil {\n\t\t\t\t\tresults[idx] = &MCPToolResult{\n\t\t\t\t\t\tMCP: r.MCP,\n\t\t\t\t\t\tTool: r.Tool,\n\t\t\t\t\t\tError: fmt.Sprintf(\"panic: %v\", err),\n\t\t\t\t\t}\n\t\t\t\t}\n\t\t\t\tdone <- struct{}{}\n\t\t\t}()\n\n\t\t\tresults[idx] = ctx.callToolSingle(r)\n\t\t}(i, req)\n\t}\n\n\t// Wait for all to complete\n\tfor remaining > 0 {\n\t\t<-done\n\t\tremaining--\n\t}\n\n\treturn results\n}\n\n// CallToolAny calls tools on multiple MCP servers concurrently and returns when any succeeds\n// Returns all results received so far when first success is found (like Promise.any)\nfunc (ctx *Context) CallToolAny(requests []*MCPToolRequest) []*MCPToolResult {\n\tif len(requests) == 0 {\n\t\treturn []*MCPToolResult{}\n\t}\n\n\tresultChan := make(chan callToolResult, len(requests))\n\tremaining := len(requests)\n\n\tfor i, req := range requests {\n\t\tgo func(idx int, r *MCPToolRequest) {\n\t\t\tdefer func() {\n\t\t\t\tif err := recover(); err != nil {\n\t\t\t\t\tresultChan <- callToolResult{\n\t\t\t\t\t\tidx: idx,\n\t\t\t\t\t\tresult: &MCPToolResult{\n\t\t\t\t\t\t\tMCP: r.MCP,\n\t\t\t\t\t\t\tTool: r.Tool,\n\t\t\t\t\t\t\tError: fmt.Sprintf(\"panic: %v\", err),\n\t\t\t\t\t\t},\n\t\t\t\t\t}\n\t\t\t\t}\n\t\t\t}()\n\n\t\t\tresultChan <- callToolResult{idx: idx, result: ctx.callToolSingle(r)}\n\t\t}(i, req)\n\t}\n\n\t// Collect results until we find a success or all fail\n\tresults := make([]*MCPToolResult, len(requests))\n\n\tfor remaining > 0 {\n\t\tcr := <-resultChan\n\t\tremaining--\n\t\tresults[cr.idx] = cr.result\n\n\t\t// Check if this is a success (no error)\n\t\tif cr.result.Error == \"\" {\n\t\t\tbreak // Stop waiting, we have a success\n\t\t}\n\t}\n\n\t// Drain remaining results in background (don't block)\n\tif remaining > 0 {\n\t\tgo func(count int) {\n\t\t\tfor i := 0; i < count; i++ {\n\t\t\t\t<-resultChan\n\t\t\t}\n\t\t}(remaining)\n\t}\n\n\treturn results\n}\n\n// CallToolRace calls tools on multiple MCP servers concurrently and returns when any completes\n// Returns all results received so far when first completion (like Promise.race)\nfunc (ctx *Context) CallToolRace(requests []*MCPToolRequest) []*MCPToolResult {\n\tif len(requests) == 0 {\n\t\treturn []*MCPToolResult{}\n\t}\n\n\tresultChan := make(chan callToolResult, len(requests))\n\tremaining := len(requests)\n\n\tfor i, req := range requests {\n\t\tgo func(idx int, r *MCPToolRequest) {\n\t\t\tdefer func() {\n\t\t\t\tif err := recover(); err != nil {\n\t\t\t\t\tresultChan <- callToolResult{\n\t\t\t\t\t\tidx: idx,\n\t\t\t\t\t\tresult: &MCPToolResult{\n\t\t\t\t\t\t\tMCP: r.MCP,\n\t\t\t\t\t\t\tTool: r.Tool,\n\t\t\t\t\t\t\tError: fmt.Sprintf(\"panic: %v\", err),\n\t\t\t\t\t\t},\n\t\t\t\t\t}\n\t\t\t\t}\n\t\t\t}()\n\n\t\t\tresultChan <- callToolResult{idx: idx, result: ctx.callToolSingle(r)}\n\t\t}(i, req)\n\t}\n\n\t// Get first result (success or failure)\n\tresults := make([]*MCPToolResult, len(requests))\n\tcr := <-resultChan\n\tremaining--\n\tresults[cr.idx] = cr.result\n\n\t// Drain remaining results in background (don't block)\n\tif remaining > 0 {\n\t\tgo func(count int) {\n\t\t\tfor i := 0; i < count; i++ {\n\t\t\t\t<-resultChan\n\t\t\t}\n\t\t}(remaining)\n\t}\n\n\treturn results\n}\n\n// callToolSingle executes a single tool call on an MCP server\n// This is a helper method for the parallel call methods\nfunc (ctx *Context) callToolSingle(req *MCPToolRequest) *MCPToolResult {\n\tresult := &MCPToolResult{\n\t\tMCP: req.MCP,\n\t\tTool: req.Tool,\n\t}\n\n\t// Call the tool using existing CallTool method\n\tresponse, err := ctx.CallTool(req.MCP, req.Tool, req.Arguments)\n\tif err != nil {\n\t\tresult.Error = err.Error()\n\t\treturn result\n\t}\n\n\t// Parse and return result directly\n\tresult.Result = parseToolResponseContent(response)\n\treturn result\n}\n\n// parseToolResponseContent extracts and parses the actual content from a CallToolResponse\n// Similar to ToolCallResult.ParsedContent() in assistant/types.go\n// - For \"text\" type, parses the Text field as JSON (or returns as string if not JSON)\n// - For \"image\" type, returns the Data and MimeType\n// - For \"resource\" type, returns the Resource object\n// - If only one content item, returns it directly (not as array)\nfunc parseToolResponseContent(response *types.CallToolResponse) interface{} {\n\tif response == nil || len(response.Content) == 0 {\n\t\treturn nil\n\t}\n\n\tvar results []interface{}\n\tfor _, tc := range response.Content {\n\t\tswitch tc.Type {\n\t\tcase types.ToolContentTypeText:\n\t\t\t// For text type, try to parse as JSON\n\t\t\tif tc.Text != \"\" {\n\t\t\t\tvar parsed interface{}\n\t\t\t\tif err := jsoniter.UnmarshalFromString(tc.Text, &parsed); err == nil {\n\t\t\t\t\tresults = append(results, parsed)\n\t\t\t\t} else {\n\t\t\t\t\t// If not JSON, return as plain string\n\t\t\t\t\tresults = append(results, tc.Text)\n\t\t\t\t}\n\t\t\t}\n\t\tcase types.ToolContentTypeImage:\n\t\t\t// For image type, return data and mimeType\n\t\t\tresults = append(results, map[string]interface{}{\n\t\t\t\t\"type\": \"image\",\n\t\t\t\t\"data\": tc.Data,\n\t\t\t\t\"mimeType\": tc.MimeType,\n\t\t\t})\n\t\tcase types.ToolContentTypeResource:\n\t\t\t// For resource type, return the resource object\n\t\t\tif tc.Resource != nil {\n\t\t\t\tresults = append(results, tc.Resource)\n\t\t\t}\n\t\tdefault:\n\t\t\t// Unknown type, include as-is with type info\n\t\t\tresults = append(results, map[string]interface{}{\n\t\t\t\t\"type\": tc.Type,\n\t\t\t\t\"text\": tc.Text,\n\t\t\t})\n\t\t}\n\t}\n\n\t// If only one result, return it directly (not as array)\n\tif len(results) == 1 {\n\t\treturn results[0]\n\t}\n\n\treturn results\n}\n" }, { "path": "agent/context/mcp_test.go", "content": "package context_test\n\nimport (\n\tstdContext \"context\"\n\t\"testing\"\n\n\t\"github.com/yaoapp/gou/mcp/types\"\n\t\"github.com/yaoapp/yao/agent/context\"\n\t\"github.com/yaoapp/yao/config\"\n\t\"github.com/yaoapp/yao/test\"\n)\n\n// newTestMCPContext creates a test context\nfunc newTestMCPContext() *context.Context {\n\tctx := context.New(stdContext.Background(), nil, \"test-chat\")\n\tctx.AssistantID = \"test-assistant\"\n\tctx.Locale = \"en\"\n\tctx.Referer = context.RefererAPI\n\n\t// Initialize stack and trace\n\tstack, traceID, _ := context.EnterStack(ctx, \"test-assistant\", &context.Options{})\n\tctx.Stack = stack\n\t_ = traceID // traceID is set in stack\n\n\treturn ctx\n}\n\n// TestListResources tests the ListResources function\nfunc TestListResources(t *testing.T) {\n\ttest.Prepare(t, config.Conf)\n\tdefer test.Clean()\n\n\tctx := newTestMCPContext()\n\n\tresult, err := ctx.ListResources(\"echo\", \"\")\n\tif err != nil {\n\t\tt.Fatalf(\"ListResources failed: %v\", err)\n\t}\n\n\tif result == nil {\n\t\tt.Fatal(\"Expected result, got nil\")\n\t}\n\n\tif len(result.Resources) == 0 {\n\t\tt.Error(\"Expected resources, got empty list\")\n\t}\n\n\tt.Logf(\"✓ ListResources returned %d resources\", len(result.Resources))\n\n\t// Check if specific resources exist\n\tresourceNames := make(map[string]bool)\n\tfor _, resource := range result.Resources {\n\t\tresourceNames[resource.Name] = true\n\t\tt.Logf(\" - Resource: %s (URI: %s)\", resource.Name, resource.URI)\n\t}\n\n\tif !resourceNames[\"info\"] {\n\t\tt.Error(\"Expected 'info' resource not found\")\n\t}\n\tif !resourceNames[\"health\"] {\n\t\tt.Error(\"Expected 'health' resource not found\")\n\t}\n}\n\n// TestReadResource tests the ReadResource function\nfunc TestReadResource(t *testing.T) {\n\ttest.Prepare(t, config.Conf)\n\tdefer test.Clean()\n\n\tctx := newTestMCPContext()\n\n\tt.Run(\"ReadServerInfo\", func(t *testing.T) {\n\t\tresult, err := ctx.ReadResource(\"echo\", \"echo://info\")\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"ReadResource failed: %v\", err)\n\t\t}\n\n\t\tif result == nil {\n\t\t\tt.Fatal(\"Expected result, got nil\")\n\t\t}\n\n\t\tif len(result.Contents) == 0 {\n\t\t\tt.Error(\"Expected contents, got empty list\")\n\t\t}\n\n\t\tt.Logf(\"✓ ReadResource returned %d contents\", len(result.Contents))\n\t})\n\n\tt.Run(\"ReadHealthCheck\", func(t *testing.T) {\n\t\tresult, err := ctx.ReadResource(\"echo\", \"echo://health?check=all\")\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"ReadResource failed: %v\", err)\n\t\t}\n\n\t\tif result == nil {\n\t\t\tt.Fatal(\"Expected result, got nil\")\n\t\t}\n\n\t\tif len(result.Contents) == 0 {\n\t\t\tt.Error(\"Expected contents, got empty list\")\n\t\t}\n\n\t\tt.Logf(\"✓ ReadResource for health check returned %d contents\", len(result.Contents))\n\t})\n}\n\n// TestListTools tests the ListTools function\nfunc TestListTools(t *testing.T) {\n\ttest.Prepare(t, config.Conf)\n\tdefer test.Clean()\n\n\tctx := newTestMCPContext()\n\n\tresult, err := ctx.ListTools(\"echo\", \"\")\n\tif err != nil {\n\t\tt.Fatalf(\"ListTools failed: %v\", err)\n\t}\n\n\tif result == nil {\n\t\tt.Fatal(\"Expected result, got nil\")\n\t}\n\n\tif len(result.Tools) == 0 {\n\t\tt.Error(\"Expected tools, got empty list\")\n\t}\n\n\tt.Logf(\"✓ ListTools returned %d tools\", len(result.Tools))\n\n\t// Check if specific tools exist\n\ttoolNames := make(map[string]bool)\n\tfor _, tool := range result.Tools {\n\t\ttoolNames[tool.Name] = true\n\t}\n\n\tif !toolNames[\"ping\"] {\n\t\tt.Error(\"Expected 'ping' tool not found\")\n\t}\n\tif !toolNames[\"status\"] {\n\t\tt.Error(\"Expected 'status' tool not found\")\n\t}\n\tif !toolNames[\"echo\"] {\n\t\tt.Error(\"Expected 'echo' tool not found\")\n\t}\n}\n\n// TestCallTool tests the CallTool function\nfunc TestCallTool(t *testing.T) {\n\ttest.Prepare(t, config.Conf)\n\tdefer test.Clean()\n\n\tctx := newTestMCPContext()\n\n\tt.Run(\"CallPing\", func(t *testing.T) {\n\t\tresult, err := ctx.CallTool(\"echo\", \"ping\", map[string]interface{}{\n\t\t\t\"count\": 3,\n\t\t\t\"message\": \"test\",\n\t\t})\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"CallTool failed: %v\", err)\n\t\t}\n\n\t\tif result == nil {\n\t\t\tt.Fatal(\"Expected result, got nil\")\n\t\t}\n\n\t\tif len(result.Content) == 0 {\n\t\t\tt.Error(\"Expected content, got empty list\")\n\t\t}\n\n\t\tt.Logf(\"✓ CallTool (ping) returned %d contents\", len(result.Content))\n\t})\n\n\tt.Run(\"CallStatus\", func(t *testing.T) {\n\t\tresult, err := ctx.CallTool(\"echo\", \"status\", map[string]interface{}{\n\t\t\t\"verbose\": true,\n\t\t})\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"CallTool failed: %v\", err)\n\t\t}\n\n\t\tif result == nil {\n\t\t\tt.Fatal(\"Expected result, got nil\")\n\t\t}\n\n\t\tif len(result.Content) == 0 {\n\t\t\tt.Error(\"Expected content, got empty list\")\n\t\t}\n\n\t\tt.Logf(\"✓ CallTool (status) returned %d contents\", len(result.Content))\n\t})\n\n\tt.Run(\"CallEcho\", func(t *testing.T) {\n\t\tresult, err := ctx.CallTool(\"echo\", \"echo\", map[string]interface{}{\n\t\t\t\"message\": \"Hello World\",\n\t\t\t\"uppercase\": true,\n\t\t})\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"CallTool failed: %v\", err)\n\t\t}\n\n\t\tif result == nil {\n\t\t\tt.Fatal(\"Expected result, got nil\")\n\t\t}\n\n\t\tif len(result.Content) == 0 {\n\t\t\tt.Error(\"Expected content, got empty list\")\n\t\t}\n\n\t\tt.Logf(\"✓ CallTool (echo) returned %d contents\", len(result.Content))\n\t})\n}\n\n// TestCallTools tests the CallTools function (sequential)\nfunc TestCallTools(t *testing.T) {\n\ttest.Prepare(t, config.Conf)\n\tdefer test.Clean()\n\n\tctx := newTestMCPContext()\n\n\ttools := []types.ToolCall{\n\t\t{\n\t\t\tName: \"ping\",\n\t\t\tArguments: map[string]interface{}{\n\t\t\t\t\"count\": 1,\n\t\t\t},\n\t\t},\n\t\t{\n\t\t\tName: \"status\",\n\t\t\tArguments: map[string]interface{}{\n\t\t\t\t\"verbose\": false,\n\t\t\t},\n\t\t},\n\t\t{\n\t\t\tName: \"echo\",\n\t\t\tArguments: map[string]interface{}{\n\t\t\t\t\"message\": \"test\",\n\t\t\t},\n\t\t},\n\t}\n\n\tresult, err := ctx.CallTools(\"echo\", tools)\n\tif err != nil {\n\t\tt.Fatalf(\"CallTools failed: %v\", err)\n\t}\n\n\tif result == nil {\n\t\tt.Fatal(\"Expected result, got nil\")\n\t}\n\n\tif len(result.Results) != 3 {\n\t\tt.Errorf(\"Expected 3 results, got %d\", len(result.Results))\n\t}\n\n\tt.Logf(\"✓ CallTools returned %d results\", len(result.Results))\n}\n\n// TestCallToolsParallel tests the CallToolsParallel function\nfunc TestCallToolsParallel(t *testing.T) {\n\ttest.Prepare(t, config.Conf)\n\tdefer test.Clean()\n\n\tctx := newTestMCPContext()\n\n\ttools := []types.ToolCall{\n\t\t{\n\t\t\tName: \"ping\",\n\t\t\tArguments: map[string]interface{}{\n\t\t\t\t\"count\": 1,\n\t\t\t},\n\t\t},\n\t\t{\n\t\t\tName: \"status\",\n\t\t\tArguments: map[string]interface{}{\n\t\t\t\t\"verbose\": true,\n\t\t\t},\n\t\t},\n\t}\n\n\tresult, err := ctx.CallToolsParallel(\"echo\", tools)\n\tif err != nil {\n\t\tt.Fatalf(\"CallToolsParallel failed: %v\", err)\n\t}\n\n\tif result == nil {\n\t\tt.Fatal(\"Expected result, got nil\")\n\t}\n\n\tif len(result.Results) != 2 {\n\t\tt.Errorf(\"Expected 2 results, got %d\", len(result.Results))\n\t}\n\n\tt.Logf(\"✓ CallToolsParallel returned %d results\", len(result.Results))\n}\n\n// TestListPrompts tests the ListPrompts function\nfunc TestListPrompts(t *testing.T) {\n\ttest.Prepare(t, config.Conf)\n\tdefer test.Clean()\n\n\tctx := newTestMCPContext()\n\n\tresult, err := ctx.ListPrompts(\"echo\", \"\")\n\tif err != nil {\n\t\tt.Fatalf(\"ListPrompts failed: %v\", err)\n\t}\n\n\tif result == nil {\n\t\tt.Fatal(\"Expected result, got nil\")\n\t}\n\n\tif len(result.Prompts) == 0 {\n\t\tt.Error(\"Expected prompts, got empty list\")\n\t}\n\n\tt.Logf(\"✓ ListPrompts returned %d prompts\", len(result.Prompts))\n\n\t// Check if specific prompts exist\n\tpromptNames := make(map[string]bool)\n\tfor _, prompt := range result.Prompts {\n\t\tpromptNames[prompt.Name] = true\n\t}\n\n\tif !promptNames[\"test_connection\"] {\n\t\tt.Error(\"Expected 'test_connection' prompt not found\")\n\t}\n\tif !promptNames[\"test_echo\"] {\n\t\tt.Error(\"Expected 'test_echo' prompt not found\")\n\t}\n}\n\n// TestGetPrompt tests the GetPrompt function\nfunc TestGetPrompt(t *testing.T) {\n\ttest.Prepare(t, config.Conf)\n\tdefer test.Clean()\n\n\tctx := newTestMCPContext()\n\n\tt.Run(\"GetTestConnectionPrompt\", func(t *testing.T) {\n\t\tresult, err := ctx.GetPrompt(\"echo\", \"test_connection\", map[string]interface{}{\n\t\t\t\"detailed\": \"true\",\n\t\t})\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"GetPrompt failed: %v\", err)\n\t\t}\n\n\t\tif result == nil {\n\t\t\tt.Fatal(\"Expected result, got nil\")\n\t\t}\n\n\t\tif len(result.Messages) == 0 {\n\t\t\tt.Error(\"Expected messages, got empty list\")\n\t\t}\n\n\t\tt.Logf(\"✓ GetPrompt returned %d messages\", len(result.Messages))\n\t})\n\n\tt.Run(\"GetTestEchoPrompt\", func(t *testing.T) {\n\t\tresult, err := ctx.GetPrompt(\"echo\", \"test_echo\", map[string]interface{}{\n\t\t\t\"message\": \"Hello\",\n\t\t\t\"format\": \"uppercase\",\n\t\t})\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"GetPrompt failed: %v\", err)\n\t\t}\n\n\t\tif result == nil {\n\t\t\tt.Fatal(\"Expected result, got nil\")\n\t\t}\n\n\t\tif len(result.Messages) == 0 {\n\t\t\tt.Error(\"Expected messages, got empty list\")\n\t\t}\n\n\t\tt.Logf(\"✓ GetPrompt returned %d messages\", len(result.Messages))\n\t})\n}\n\n// TestListSamples tests the ListSamples function\nfunc TestListSamples(t *testing.T) {\n\ttest.Prepare(t, config.Conf)\n\tdefer test.Clean()\n\n\tctx := newTestMCPContext()\n\n\tt.Run(\"ListToolSamples\", func(t *testing.T) {\n\t\tresult, err := ctx.ListSamples(\"echo\", types.SampleTool, \"ping\")\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"ListSamples failed: %v\", err)\n\t\t}\n\n\t\tif result == nil {\n\t\t\tt.Fatal(\"Expected result, got nil\")\n\t\t}\n\n\t\tif len(result.Samples) == 0 {\n\t\t\tt.Error(\"Expected samples, got empty list\")\n\t\t}\n\n\t\tt.Logf(\"✓ ListSamples for tool 'ping' returned %d samples\", len(result.Samples))\n\t})\n\n\tt.Run(\"ListResourceSamples\", func(t *testing.T) {\n\t\tresult, err := ctx.ListSamples(\"echo\", types.SampleResource, \"info\")\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"ListSamples failed: %v\", err)\n\t\t}\n\n\t\tif result == nil {\n\t\t\tt.Fatal(\"Expected result, got nil\")\n\t\t}\n\n\t\tif len(result.Samples) == 0 {\n\t\t\tt.Error(\"Expected samples, got empty list\")\n\t\t}\n\n\t\tt.Logf(\"✓ ListSamples for resource 'info' returned %d samples\", len(result.Samples))\n\t})\n}\n\n// TestGetSample tests the GetSample function\nfunc TestGetSample(t *testing.T) {\n\ttest.Prepare(t, config.Conf)\n\tdefer test.Clean()\n\n\tctx := newTestMCPContext()\n\n\tt.Run(\"GetToolSample\", func(t *testing.T) {\n\t\tresult, err := ctx.GetSample(\"echo\", types.SampleTool, \"ping\", 0)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"GetSample failed: %v\", err)\n\t\t}\n\n\t\tif result == nil {\n\t\t\tt.Fatal(\"Expected result, got nil\")\n\t\t}\n\n\t\tif result.Name == \"\" {\n\t\t\tt.Error(\"Expected sample name, got empty string\")\n\t\t}\n\n\t\tt.Logf(\"✓ GetSample for tool 'ping' returned sample '%s'\", result.Name)\n\t})\n\n\tt.Run(\"GetResourceSample\", func(t *testing.T) {\n\t\tresult, err := ctx.GetSample(\"echo\", types.SampleResource, \"info\", 0)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"GetSample failed: %v\", err)\n\t\t}\n\n\t\tif result == nil {\n\t\t\tt.Fatal(\"Expected result, got nil\")\n\t\t}\n\n\t\tif result.Name == \"\" {\n\t\t\tt.Error(\"Expected sample name, got empty string\")\n\t\t}\n\n\t\tt.Logf(\"✓ GetSample for resource 'info' returned sample '%s'\", result.Name)\n\t})\n}\n\n// TestMCPWithTrace tests MCP operations with trace\nfunc TestMCPWithTrace(t *testing.T) {\n\ttest.Prepare(t, config.Conf)\n\tdefer test.Clean()\n\n\tctx := newTestMCPContext()\n\n\t// Initialize trace\n\ttrace, err := ctx.Trace()\n\tif err != nil {\n\t\tt.Fatalf(\"Failed to initialize trace: %v\", err)\n\t}\n\n\tif trace == nil {\n\t\tt.Fatal(\"Expected trace, got nil\")\n\t}\n\n\t// Call tool with trace\n\tresult, err := ctx.CallTool(\"echo\", \"ping\", map[string]interface{}{\n\t\t\"count\": 5,\n\t})\n\tif err != nil {\n\t\tt.Fatalf(\"CallTool with trace failed: %v\", err)\n\t}\n\n\tif result == nil {\n\t\tt.Fatal(\"Expected result, got nil\")\n\t}\n\n\t// Get trace nodes to verify trace was created\n\tnodes, err := trace.GetAllNodes()\n\tif err != nil {\n\t\tt.Fatalf(\"Failed to get trace nodes: %v\", err)\n\t}\n\n\tif len(nodes) == 0 {\n\t\tt.Error(\"Expected trace nodes, got empty list\")\n\t}\n\n\tt.Logf(\"✓ MCP operation created %d trace nodes\", len(nodes))\n}\n" }, { "path": "agent/context/message.go", "content": "package context\n\nimport (\n\t\"encoding/json\"\n\t\"fmt\"\n\t\"sync\"\n)\n\n// messageMetadataStore provides thread-safe storage for message and block metadata\ntype messageMetadataStore struct {\n\tmessages map[string]*MessageMetadata // Message metadata by MessageID\n\tblocks map[string]*BlockMetadata // Block metadata by BlockID\n\tmu sync.RWMutex\n}\n\n// newMessageMetadataStore creates a new message metadata store\nfunc newMessageMetadataStore() *messageMetadataStore {\n\treturn &messageMetadataStore{\n\t\tmessages: make(map[string]*MessageMetadata),\n\t\tblocks: make(map[string]*BlockMetadata),\n\t}\n}\n\n// setMessage stores metadata for a message (thread-safe)\nfunc (s *messageMetadataStore) setMessage(messageID string, metadata *MessageMetadata) {\n\ts.mu.Lock()\n\tdefer s.mu.Unlock()\n\ts.messages[messageID] = metadata\n}\n\n// getMessage retrieves metadata for a message (thread-safe)\nfunc (s *messageMetadataStore) getMessage(messageID string) *MessageMetadata {\n\ts.mu.RLock()\n\tdefer s.mu.RUnlock()\n\treturn s.messages[messageID]\n}\n\n// setBlock stores metadata for a block (thread-safe)\nfunc (s *messageMetadataStore) setBlock(blockID string, metadata *BlockMetadata) {\n\ts.mu.Lock()\n\tdefer s.mu.Unlock()\n\ts.blocks[blockID] = metadata\n}\n\n// getBlock retrieves metadata for a block (thread-safe)\nfunc (s *messageMetadataStore) getBlock(blockID string) *BlockMetadata {\n\ts.mu.RLock()\n\tdefer s.mu.RUnlock()\n\treturn s.blocks[blockID]\n}\n\n// updateBlock updates block metadata (thread-safe)\nfunc (s *messageMetadataStore) updateBlock(blockID string, update func(*BlockMetadata)) {\n\ts.mu.Lock()\n\tdefer s.mu.Unlock()\n\tif block, exists := s.blocks[blockID]; exists {\n\t\tupdate(block)\n\t}\n}\n\n// UnmarshalJSON custom unmarshaler for Message to handle Content field\nfunc (m *Message) UnmarshalJSON(data []byte) error {\n\t// Define a temporary struct to avoid infinite recursion\n\ttype Alias Message\n\taux := &struct {\n\t\tContent json.RawMessage `json:\"content,omitempty\"`\n\t\t*Alias\n\t}{\n\t\tAlias: (*Alias)(m),\n\t}\n\n\tif err := json.Unmarshal(data, &aux); err != nil {\n\t\treturn err\n\t}\n\n\t// If content is empty, return early\n\tif len(aux.Content) == 0 || string(aux.Content) == \"null\" {\n\t\tm.Content = nil\n\t\treturn nil\n\t}\n\n\t// Try to unmarshal as string first\n\tvar contentStr string\n\tif err := json.Unmarshal(aux.Content, &contentStr); err == nil {\n\t\tm.Content = contentStr\n\t\treturn nil\n\t}\n\n\t// Try to unmarshal as array of ContentPart\n\tvar contentParts []ContentPart\n\tif err := json.Unmarshal(aux.Content, &contentParts); err == nil {\n\t\tm.Content = contentParts\n\t\treturn nil\n\t}\n\n\treturn fmt.Errorf(\"content must be either a string or an array of ContentPart\")\n}\n\n// MarshalJSON custom marshaler for Message\nfunc (m *Message) MarshalJSON() ([]byte, error) {\n\ttype Alias Message\n\treturn json.Marshal(&struct {\n\t\t*Alias\n\t}{\n\t\tAlias: (*Alias)(m),\n\t})\n}\n\n// NewTextMessage creates a new message with text content\nfunc NewTextMessage(role MessageRole, text string) *Message {\n\treturn &Message{\n\t\tRole: role,\n\t\tContent: text,\n\t}\n}\n\n// NewMultipartMessage creates a new message with multipart content\nfunc NewMultipartMessage(role MessageRole, parts []ContentPart) *Message {\n\treturn &Message{\n\t\tRole: role,\n\t\tContent: parts,\n\t}\n}\n\n// GetContentAsString returns content as string if possible\nfunc (m *Message) GetContentAsString() (string, bool) {\n\tif str, ok := m.Content.(string); ok {\n\t\treturn str, true\n\t}\n\treturn \"\", false\n}\n\n// GetContentAsParts returns content as ContentPart array if possible\nfunc (m *Message) GetContentAsParts() ([]ContentPart, bool) {\n\tif parts, ok := m.Content.([]ContentPart); ok {\n\t\treturn parts, true\n\t}\n\treturn nil, false\n}\n\n// HasToolCalls checks if the message has tool calls\nfunc (m *Message) HasToolCalls() bool {\n\treturn len(m.ToolCalls) > 0\n}\n\n// IsRefusal checks if the message is a refusal\nfunc (m *Message) IsRefusal() bool {\n\treturn m.Refusal != nil && *m.Refusal != \"\"\n}\n" }, { "path": "agent/context/message_events_test.go", "content": "package context_test\n\nimport (\n\t\"bytes\"\n\tstdContext \"context\"\n\t\"encoding/json\"\n\t\"net/http\"\n\t\"testing\"\n\n\t\"github.com/stretchr/testify/assert\"\n\t\"github.com/yaoapp/yao/agent/context\"\n\t\"github.com/yaoapp/yao/agent/output/message\"\n)\n\nfunc TestMessageLifecycleEvents(t *testing.T) {\n\t// Create a mock response writer\n\tvar buf bytes.Buffer\n\tmockWriter := &mockResponseWriter{\n\t\tbuffer: &buf,\n\t\theaders: make(http.Header),\n\t}\n\n\t// Create context using New() to ensure proper initialization\n\tctx := context.New(stdContext.Background(), nil, \"test-chat\")\n\tctx.Accept = context.AcceptWebCUI\n\tctx.Writer = mockWriter\n\tctx.AssistantID = \"test-assistant\"\n\tctx.Locale = \"en\"\n\n\t// Send a simple text message\n\terr := ctx.Send(&message.Message{\n\t\tType: message.TypeText,\n\t\tProps: map[string]interface{}{\n\t\t\t\"content\": \"Hello World\",\n\t\t},\n\t})\n\tassert.NoError(t, err)\n\n\t// Flush to ensure all messages are written\n\tctx.Flush()\n\n\t// Close SafeWriter to wait for all async writes to complete\n\t// SafeWriter uses a channel-based queue, so we must close it before reading buffer\n\tctx.CloseSafeWriter()\n\n\t// Parse output to find events\n\toutput := buf.String()\n\tt.Logf(\"Output:\\n%s\", output)\n\n\tlines := bytes.Split([]byte(output), []byte(\"\\n\"))\n\n\tvar messages []map[string]interface{}\n\tfor _, line := range lines {\n\t\tif len(line) == 0 {\n\t\t\tcontinue\n\t\t}\n\t\t// CUI format: data: {...}\n\t\tif bytes.HasPrefix(line, []byte(\"data: \")) {\n\t\t\tline = bytes.TrimPrefix(line, []byte(\"data: \"))\n\t\t}\n\n\t\tvar msg map[string]interface{}\n\t\tif err := json.Unmarshal(line, &msg); err == nil {\n\t\t\tmessages = append(messages, msg)\n\t\t\tt.Logf(\"Message: type=%s\", msg[\"type\"])\n\t\t}\n\t}\n\n\t// Check for events\n\thasMessageStart := false\n\thasMessageEnd := false\n\thasTextMessage := false\n\n\tfor _, msg := range messages {\n\t\tmsgType, _ := msg[\"type\"].(string)\n\n\t\tif msgType == \"event\" {\n\t\t\tif props, ok := msg[\"props\"].(map[string]interface{}); ok {\n\t\t\t\tif eventType, ok := props[\"event\"].(string); ok {\n\t\t\t\t\tt.Logf(\"Event type: %s\", eventType)\n\t\t\t\t\tif eventType == \"message_start\" {\n\t\t\t\t\t\thasMessageStart = true\n\t\t\t\t\t\tt.Logf(\"✓ Found message_start event\")\n\t\t\t\t\t}\n\t\t\t\t\tif eventType == \"message_end\" {\n\t\t\t\t\t\thasMessageEnd = true\n\t\t\t\t\t\tt.Logf(\"✓ Found message_end event: %+v\", props)\n\t\t\t\t\t}\n\t\t\t\t}\n\t\t\t}\n\t\t} else if msgType == \"text\" {\n\t\t\thasTextMessage = true\n\t\t\tt.Logf(\"✓ Found text message\")\n\t\t}\n\t}\n\n\tt.Logf(\"Summary: start=%v, text=%v, end=%v\", hasMessageStart, hasTextMessage, hasMessageEnd)\n\n\tassert.True(t, hasMessageStart, \"Should have message_start event\")\n\tassert.True(t, hasTextMessage, \"Should have text message\")\n\tassert.True(t, hasMessageEnd, \"Should have message_end event\")\n}\n\n// mockResponseWriter implements http.ResponseWriter for testing\ntype mockResponseWriter struct {\n\tbuffer *bytes.Buffer\n\tstatusCode int\n\theaders http.Header\n}\n\nfunc (m *mockResponseWriter) Header() http.Header {\n\treturn m.headers\n}\n\nfunc (m *mockResponseWriter) Write(data []byte) (int, error) {\n\treturn m.buffer.Write(data)\n}\n\nfunc (m *mockResponseWriter) WriteHeader(statusCode int) {\n\tm.statusCode = statusCode\n}\n" }, { "path": "agent/context/message_test.go", "content": "package context_test\n\nimport (\n\t\"encoding/json\"\n\t\"testing\"\n\n\t\"github.com/yaoapp/yao/agent/context\"\n)\n\nfunc TestMessage_UnmarshalJSON_StringContent(t *testing.T) {\n\tjsonData := `{\n\t\t\"role\": \"user\",\n\t\t\"content\": \"Hello, world!\"\n\t}`\n\n\tvar msg context.Message\n\terr := json.Unmarshal([]byte(jsonData), &msg)\n\tif err != nil {\n\t\tt.Fatalf(\"Failed to unmarshal: %v\", err)\n\t}\n\n\tif msg.Role != context.RoleUser {\n\t\tt.Errorf(\"Expected role %s, got %s\", context.RoleUser, msg.Role)\n\t}\n\n\tcontent, ok := msg.GetContentAsString()\n\tif !ok {\n\t\tt.Fatal(\"Expected content to be string\")\n\t}\n\n\tif content != \"Hello, world!\" {\n\t\tt.Errorf(\"Expected content 'Hello, world!', got '%s'\", content)\n\t}\n}\n\nfunc TestMessage_UnmarshalJSON_ArrayContent(t *testing.T) {\n\tjsonData := `{\n\t\t\"role\": \"user\",\n\t\t\"content\": [\n\t\t\t{\n\t\t\t\t\"type\": \"text\",\n\t\t\t\t\"text\": \"What's in this image?\"\n\t\t\t},\n\t\t\t{\n\t\t\t\t\"type\": \"image_url\",\n\t\t\t\t\"image_url\": {\n\t\t\t\t\t\"url\": \"https://example.com/image.jpg\",\n\t\t\t\t\t\"detail\": \"high\"\n\t\t\t\t}\n\t\t\t}\n\t\t]\n\t}`\n\n\tvar msg context.Message\n\terr := json.Unmarshal([]byte(jsonData), &msg)\n\tif err != nil {\n\t\tt.Fatalf(\"Failed to unmarshal: %v\", err)\n\t}\n\n\tif msg.Role != context.RoleUser {\n\t\tt.Errorf(\"Expected role %s, got %s\", context.RoleUser, msg.Role)\n\t}\n\n\tparts, ok := msg.GetContentAsParts()\n\tif !ok {\n\t\tt.Fatal(\"Expected content to be array of ContentPart\")\n\t}\n\n\tif len(parts) != 2 {\n\t\tt.Fatalf(\"Expected 2 content parts, got %d\", len(parts))\n\t}\n\n\t// Check first part (text)\n\tif parts[0].Type != context.ContentText {\n\t\tt.Errorf(\"Expected type %s, got %s\", context.ContentText, parts[0].Type)\n\t}\n\tif parts[0].Text != \"What's in this image?\" {\n\t\tt.Errorf(\"Expected text 'What's in this image?', got '%s'\", parts[0].Text)\n\t}\n\n\t// Check second part (image)\n\tif parts[1].Type != context.ContentImageURL {\n\t\tt.Errorf(\"Expected type %s, got %s\", context.ContentImageURL, parts[1].Type)\n\t}\n\tif parts[1].ImageURL == nil {\n\t\tt.Fatal(\"Expected ImageURL to be non-nil\")\n\t}\n\tif parts[1].ImageURL.URL != \"https://example.com/image.jpg\" {\n\t\tt.Errorf(\"Expected URL 'https://example.com/image.jpg', got '%s'\", parts[1].ImageURL.URL)\n\t}\n\tif parts[1].ImageURL.Detail != context.DetailHigh {\n\t\tt.Errorf(\"Expected detail %s, got %s\", context.DetailHigh, parts[1].ImageURL.Detail)\n\t}\n}\n\nfunc TestMessage_UnmarshalJSON_NullContent(t *testing.T) {\n\tjsonData := `{\n\t\t\"role\": \"assistant\",\n\t\t\"content\": null,\n\t\t\"tool_calls\": [\n\t\t\t{\n\t\t\t\t\"id\": \"call_123\",\n\t\t\t\t\"type\": \"function\",\n\t\t\t\t\"function\": {\n\t\t\t\t\t\"name\": \"get_weather\",\n\t\t\t\t\t\"arguments\": \"{\\\"location\\\":\\\"Tokyo\\\"}\"\n\t\t\t\t}\n\t\t\t}\n\t\t]\n\t}`\n\n\tvar msg context.Message\n\terr := json.Unmarshal([]byte(jsonData), &msg)\n\tif err != nil {\n\t\tt.Fatalf(\"Failed to unmarshal: %v\", err)\n\t}\n\n\tif msg.Role != context.RoleAssistant {\n\t\tt.Errorf(\"Expected role %s, got %s\", context.RoleAssistant, msg.Role)\n\t}\n\n\tif msg.Content != nil {\n\t\tt.Errorf(\"Expected content to be nil, got %v\", msg.Content)\n\t}\n\n\tif !msg.HasToolCalls() {\n\t\tt.Fatal(\"Expected message to have tool calls\")\n\t}\n\n\tif len(msg.ToolCalls) != 1 {\n\t\tt.Fatalf(\"Expected 1 tool call, got %d\", len(msg.ToolCalls))\n\t}\n\n\tif msg.ToolCalls[0].ID != \"call_123\" {\n\t\tt.Errorf(\"Expected tool call ID 'call_123', got '%s'\", msg.ToolCalls[0].ID)\n\t}\n}\n\nfunc TestMessage_UnmarshalJSON_WithRefusal(t *testing.T) {\n\trefusalText := \"I cannot help with that request.\"\n\tjsonData := `{\n\t\t\"role\": \"assistant\",\n\t\t\"content\": \"I'm sorry, but I can't assist with that.\",\n\t\t\"refusal\": \"I cannot help with that request.\"\n\t}`\n\n\tvar msg context.Message\n\terr := json.Unmarshal([]byte(jsonData), &msg)\n\tif err != nil {\n\t\tt.Fatalf(\"Failed to unmarshal: %v\", err)\n\t}\n\n\tif !msg.IsRefusal() {\n\t\tt.Error(\"Expected message to be a refusal\")\n\t}\n\n\tif msg.Refusal == nil {\n\t\tt.Fatal(\"Expected refusal to be non-nil\")\n\t}\n\n\tif *msg.Refusal != refusalText {\n\t\tt.Errorf(\"Expected refusal '%s', got '%s'\", refusalText, *msg.Refusal)\n\t}\n}\n\nfunc TestMessage_UnmarshalJSON_AudioContent(t *testing.T) {\n\tjsonData := `{\n\t\t\"role\": \"user\",\n\t\t\"content\": [\n\t\t\t{\n\t\t\t\t\"type\": \"text\",\n\t\t\t\t\"text\": \"Transcribe this audio\"\n\t\t\t},\n\t\t\t{\n\t\t\t\t\"type\": \"input_audio\",\n\t\t\t\t\"input_audio\": {\n\t\t\t\t\t\"data\": \"base64encodedaudiodata\",\n\t\t\t\t\t\"format\": \"wav\"\n\t\t\t\t}\n\t\t\t}\n\t\t]\n\t}`\n\n\tvar msg context.Message\n\terr := json.Unmarshal([]byte(jsonData), &msg)\n\tif err != nil {\n\t\tt.Fatalf(\"Failed to unmarshal: %v\", err)\n\t}\n\n\tparts, ok := msg.GetContentAsParts()\n\tif !ok {\n\t\tt.Fatal(\"Expected content to be array of ContentPart\")\n\t}\n\n\tif len(parts) != 2 {\n\t\tt.Fatalf(\"Expected 2 content parts, got %d\", len(parts))\n\t}\n\n\t// Check audio part\n\tif parts[1].Type != context.ContentInputAudio {\n\t\tt.Errorf(\"Expected type %s, got %s\", context.ContentInputAudio, parts[1].Type)\n\t}\n\tif parts[1].InputAudio == nil {\n\t\tt.Fatal(\"Expected InputAudio to be non-nil\")\n\t}\n\tif parts[1].InputAudio.Data != \"base64encodedaudiodata\" {\n\t\tt.Errorf(\"Expected audio data 'base64encodedaudiodata', got '%s'\", parts[1].InputAudio.Data)\n\t}\n\tif parts[1].InputAudio.Format != \"wav\" {\n\t\tt.Errorf(\"Expected format 'wav', got '%s'\", parts[1].InputAudio.Format)\n\t}\n}\n\nfunc TestMessage_MarshalJSON_StringContent(t *testing.T) {\n\tmsg := context.NewTextMessage(context.RoleUser, \"Hello, AI!\")\n\n\tdata, err := json.Marshal(msg)\n\tif err != nil {\n\t\tt.Fatalf(\"Failed to marshal: %v\", err)\n\t}\n\n\tvar result map[string]interface{}\n\terr = json.Unmarshal(data, &result)\n\tif err != nil {\n\t\tt.Fatalf(\"Failed to unmarshal result: %v\", err)\n\t}\n\n\tif result[\"role\"] != string(context.RoleUser) {\n\t\tt.Errorf(\"Expected role %s, got %v\", context.RoleUser, result[\"role\"])\n\t}\n\n\tif result[\"content\"] != \"Hello, AI!\" {\n\t\tt.Errorf(\"Expected content 'Hello, AI!', got %v\", result[\"content\"])\n\t}\n}\n\nfunc TestMessage_MarshalJSON_ArrayContent(t *testing.T) {\n\tparts := []context.ContentPart{\n\t\t{\n\t\t\tType: context.ContentText,\n\t\t\tText: \"Describe this image\",\n\t\t},\n\t\t{\n\t\t\tType: context.ContentImageURL,\n\t\t\tImageURL: &context.ImageURL{\n\t\t\t\tURL: \"https://example.com/test.jpg\",\n\t\t\t\tDetail: context.DetailLow,\n\t\t\t},\n\t\t},\n\t}\n\n\tmsg := context.NewMultipartMessage(context.RoleUser, parts)\n\n\tdata, err := json.Marshal(msg)\n\tif err != nil {\n\t\tt.Fatalf(\"Failed to marshal: %v\", err)\n\t}\n\n\t// Unmarshal back to verify\n\tvar result context.Message\n\terr = json.Unmarshal(data, &result)\n\tif err != nil {\n\t\tt.Fatalf(\"Failed to unmarshal result: %v\", err)\n\t}\n\n\tresultParts, ok := result.GetContentAsParts()\n\tif !ok {\n\t\tt.Fatal(\"Expected content to be array of ContentPart\")\n\t}\n\n\tif len(resultParts) != 2 {\n\t\tt.Fatalf(\"Expected 2 content parts, got %d\", len(resultParts))\n\t}\n}\n\nfunc TestMessage_MarshalJSON_WithToolCalls(t *testing.T) {\n\tmsg := &context.Message{\n\t\tRole: context.RoleAssistant,\n\t\tContent: nil,\n\t\tToolCalls: []context.ToolCall{\n\t\t\t{\n\t\t\t\tID: \"call_abc123\",\n\t\t\t\tType: context.ToolTypeFunction,\n\t\t\t\tFunction: context.Function{\n\t\t\t\t\tName: \"get_weather\",\n\t\t\t\t\tArguments: `{\"location\":\"San Francisco\"}`,\n\t\t\t\t},\n\t\t\t},\n\t\t},\n\t}\n\n\tdata, err := json.Marshal(msg)\n\tif err != nil {\n\t\tt.Fatalf(\"Failed to marshal: %v\", err)\n\t}\n\n\t// Unmarshal back to verify\n\tvar result context.Message\n\terr = json.Unmarshal(data, &result)\n\tif err != nil {\n\t\tt.Fatalf(\"Failed to unmarshal result: %v\", err)\n\t}\n\n\tif !result.HasToolCalls() {\n\t\tt.Error(\"Expected message to have tool calls\")\n\t}\n\n\tif len(result.ToolCalls) != 1 {\n\t\tt.Fatalf(\"Expected 1 tool call, got %d\", len(result.ToolCalls))\n\t}\n\n\tif result.ToolCalls[0].Function.Name != \"get_weather\" {\n\t\tt.Errorf(\"Expected function name 'get_weather', got '%s'\", result.ToolCalls[0].Function.Name)\n\t}\n}\n\nfunc TestMessage_ToolMessage(t *testing.T) {\n\ttoolCallID := \"call_abc123\"\n\tjsonData := `{\n\t\t\"role\": \"tool\",\n\t\t\"tool_call_id\": \"call_abc123\",\n\t\t\"content\": \"The weather in San Francisco is sunny, 72°F\"\n\t}`\n\n\tvar msg context.Message\n\terr := json.Unmarshal([]byte(jsonData), &msg)\n\tif err != nil {\n\t\tt.Fatalf(\"Failed to unmarshal: %v\", err)\n\t}\n\n\tif msg.Role != context.RoleTool {\n\t\tt.Errorf(\"Expected role %s, got %s\", context.RoleTool, msg.Role)\n\t}\n\n\tif msg.ToolCallID == nil {\n\t\tt.Fatal(\"Expected tool_call_id to be non-nil\")\n\t}\n\n\tif *msg.ToolCallID != toolCallID {\n\t\tt.Errorf(\"Expected tool_call_id '%s', got '%s'\", toolCallID, *msg.ToolCallID)\n\t}\n\n\tcontent, ok := msg.GetContentAsString()\n\tif !ok {\n\t\tt.Fatal(\"Expected content to be string\")\n\t}\n\n\tif content != \"The weather in San Francisco is sunny, 72°F\" {\n\t\tt.Errorf(\"Unexpected content: %s\", content)\n\t}\n}\n\nfunc TestNewTextMessage(t *testing.T) {\n\tmsg := context.NewTextMessage(context.RoleSystem, \"You are a helpful assistant.\")\n\n\tif msg.Role != context.RoleSystem {\n\t\tt.Errorf(\"Expected role %s, got %s\", context.RoleSystem, msg.Role)\n\t}\n\n\tcontent, ok := msg.GetContentAsString()\n\tif !ok {\n\t\tt.Fatal(\"Expected content to be string\")\n\t}\n\n\tif content != \"You are a helpful assistant.\" {\n\t\tt.Errorf(\"Expected content 'You are a helpful assistant.', got '%s'\", content)\n\t}\n}\n\nfunc TestNewMultipartMessage(t *testing.T) {\n\tparts := []context.ContentPart{\n\t\t{Type: context.ContentText, Text: \"Hello\"},\n\t}\n\n\tmsg := context.NewMultipartMessage(context.RoleUser, parts)\n\n\tif msg.Role != context.RoleUser {\n\t\tt.Errorf(\"Expected role %s, got %s\", context.RoleUser, msg.Role)\n\t}\n\n\tresultParts, ok := msg.GetContentAsParts()\n\tif !ok {\n\t\tt.Fatal(\"Expected content to be array of ContentPart\")\n\t}\n\n\tif len(resultParts) != 1 {\n\t\tt.Fatalf(\"Expected 1 content part, got %d\", len(resultParts))\n\t}\n}\n" }, { "path": "agent/context/openapi.go", "content": "package context\n\nimport (\n\t\"bytes\"\n\t\"encoding/base64\"\n\t\"encoding/json\"\n\t\"fmt\"\n\t\"io\"\n\t\"strings\"\n\n\t\"github.com/gin-gonic/gin\"\n\t\"github.com/yaoapp/gou/connector\"\n\t\"github.com/yaoapp/gou/store\"\n\t\"github.com/yaoapp/yao/openapi/oauth/authorized\"\n)\n\n// GetCompletionRequest parse completion request and create context from openapi request\n// Returns: *CompletionRequest, *Context, *Options, error\nfunc GetCompletionRequest(c *gin.Context, cache store.Store) (*CompletionRequest, *Context, *Options, error) {\n\t// Get authorized information\n\tauthInfo := authorized.GetInfo(c)\n\n\t// Parse completion request from payload or query first\n\tcompletionReq, err := parseCompletionRequestData(c)\n\tif err != nil {\n\t\treturn nil, nil, nil, fmt.Errorf(\"failed to parse completion request: %w\", err)\n\t}\n\n\t// Extract assistant ID using completionReq (can extract from model field)\n\tassistantID, err := GetAssistantID(c, completionReq)\n\tif err != nil {\n\t\treturn nil, nil, nil, fmt.Errorf(\"failed to get assistant ID: %w\", err)\n\t}\n\n\t// Extract chat ID (may generate from messages if not provided)\n\tchatID, err := GetChatID(c, cache, completionReq)\n\tif err != nil {\n\t\t// Fallback: Generate a new chat ID if extraction fails\n\t\tchatID = GenChatID()\n\t}\n\n\t// Parse client information from User-Agent header\n\tuserAgent := c.GetHeader(\"User-Agent\")\n\tclientType := getClientType(userAgent)\n\tclientIP := c.ClientIP()\n\n\t// Create context with unique ID using New() to ensure proper initialization\n\tctx := New(c.Request.Context(), authInfo, chatID)\n\n\t// Set context fields (session-level state)\n\tctx.Cache = cache\n\tctx.Writer = c.Writer\n\tctx.AssistantID = assistantID\n\tctx.Locale = GetLocale(c, completionReq)\n\tctx.Theme = GetTheme(c, completionReq)\n\tctx.Referer = GetReferer(c, completionReq)\n\tctx.Accept = GetAccept(c, completionReq)\n\tctx.Client = Client{\n\t\tType: clientType,\n\t\tUserAgent: userAgent,\n\t\tIP: clientIP,\n\t}\n\tctx.Route = GetRoute(c, completionReq)\n\tctx.Metadata = GetMetadata(c, completionReq)\n\n\t// Create Options (call-level parameters)\n\topts := &Options{\n\t\tContext: c.Request.Context(),\n\t\tSkip: GetSkip(c, completionReq),\n\t\tMode: GetMode(c, completionReq),\n\t}\n\n\t// Try to extract custom connector from model field\n\t// If model is a valid connector ID, set it to opts.Connector\n\t// Otherwise, keep the standard OpenAI-compatible behavior (model as assistant ID)\n\tif completionReq != nil && completionReq.Model != \"\" {\n\t\t// Check if model is a valid connector (not containing \"-yao_\" which indicates assistant ID format)\n\t\tif !strings.Contains(completionReq.Model, \"-yao_\") {\n\t\t\t// Try to validate if it's a real connector\n\t\t\tif _, err := connector.Select(completionReq.Model); err == nil {\n\t\t\t\t// It's a valid connector, use it\n\t\t\t\topts.Connector = completionReq.Model\n\t\t\t}\n\t\t\t// If not a valid connector, ignore it (keep opts.Connector empty to use assistant's default)\n\t\t}\n\t}\n\n\t// Initialize interrupt controller\n\tctx.Interrupt = NewInterruptController()\n\n\t// Register context to global registry first (required for interrupt handler callback)\n\tif err := Register(ctx); err != nil {\n\t\treturn nil, nil, nil, fmt.Errorf(\"failed to register context: %w\", err)\n\t}\n\n\t// Start interrupt listener after registration\n\t// Only monitors interrupt signals (user stop button for appending messages)\n\t// HTTP context cancellation is handled by LLM/Agent layers naturally\n\tctx.Interrupt.Start(ctx.ID)\n\n\treturn completionReq, ctx, opts, nil\n}\n\n// getClientType parses the client type from User-Agent header\nfunc getClientType(userAgent string) string {\n\tif userAgent == \"\" {\n\t\treturn \"web\" // Default to web\n\t}\n\n\tua := strings.ToLower(userAgent)\n\n\t// Check for specific client types\n\tswitch {\n\tcase strings.Contains(ua, \"yao-agent\") || strings.Contains(ua, \"agent\"):\n\t\treturn \"agent\"\n\tcase strings.Contains(ua, \"yao-jssdk\") || strings.Contains(ua, \"jssdk\"):\n\t\treturn \"jssdk\"\n\tcase strings.Contains(ua, \"android\"):\n\t\treturn \"android\"\n\tcase strings.Contains(ua, \"iphone\") || strings.Contains(ua, \"ipad\") || strings.Contains(ua, \"ipod\"):\n\t\treturn \"ios\"\n\tcase strings.Contains(ua, \"windows\"):\n\t\treturn \"windows\"\n\tcase strings.Contains(ua, \"mac os x\") || strings.Contains(ua, \"macintosh\"):\n\t\treturn \"macos\"\n\tcase strings.Contains(ua, \"linux\"):\n\t\treturn \"linux\"\n\tdefault:\n\t\treturn \"web\"\n\t}\n}\n\n// GetAssistantID extracts assistant ID from request with priority:\n// 1. Query parameter \"assistant_id\"\n// 2. Header \"X-Yao-Assistant\"\n// 3. Extract from model field (from CompletionRequest or Query) - splits by \"-\" takes last field, extracts ID from \"yao_xxx\" prefix\nfunc GetAssistantID(c *gin.Context, req *CompletionRequest) (string, error) {\n\t// Priority 1: Query parameter assistant_id\n\tif assistantID := c.Query(\"assistant_id\"); assistantID != \"\" {\n\t\treturn assistantID, nil\n\t}\n\n\t// Priority 2: Header X-Yao-Assistant\n\tif assistantID := c.GetHeader(\"X-Yao-Assistant\"); assistantID != \"\" {\n\t\treturn assistantID, nil\n\t}\n\n\t// Priority 3: Extract from model field (from CompletionRequest or Query)\n\tmodel := \"\"\n\tif req != nil && req.Model != \"\" {\n\t\tmodel = req.Model\n\t} else {\n\t\tmodel = c.Query(\"model\")\n\t}\n\n\tif model != \"\" {\n\t\t// Parse model ID using the same logic as ParseModelID\n\t\t// Expected format: [prefix-]assistantName-model-yao_assistantID\n\t\t// Find the last occurrence of \"-yao_\"\n\t\tparts := strings.Split(model, \"-yao_\")\n\t\tif len(parts) >= 2 {\n\t\t\tassistantID := parts[len(parts)-1]\n\t\t\tif assistantID != \"\" {\n\t\t\t\treturn assistantID, nil\n\t\t\t}\n\t\t}\n\t}\n\n\t// If no assistant ID found, return error\n\treturn \"\", fmt.Errorf(\"assistant_id is required\")\n}\n\n// GetMessages extracts messages from the request\n// Priority:\n// 1. Query parameter \"messages\" (JSON string)\n// 2. CompletionRequest.Messages (from payload)\nfunc GetMessages(c *gin.Context, req *CompletionRequest) ([]Message, error) {\n\t// Priority 1: Query parameter messages\n\tif messagesJSON := c.Query(\"messages\"); messagesJSON != \"\" {\n\t\tvar messages []Message\n\t\tif err := json.Unmarshal([]byte(messagesJSON), &messages); err == nil && len(messages) > 0 {\n\t\t\treturn messages, nil\n\t\t}\n\t}\n\n\t// Priority 2: From CompletionRequest (payload)\n\tif req != nil && len(req.Messages) > 0 {\n\t\treturn req.Messages, nil\n\t}\n\n\treturn nil, fmt.Errorf(\"messages field is required\")\n}\n\n// GetLocale extracts locale from request with priority:\n// 1. Query parameter \"locale\"\n// 2. Header \"Accept-Language\"\n// 3. CompletionRequest metadata \"locale\" (from payload)\nfunc GetLocale(c *gin.Context, req *CompletionRequest) string {\n\t// Priority 1: Query parameter\n\tif locale := c.Query(\"locale\"); locale != \"\" {\n\t\treturn strings.ToLower(locale)\n\t}\n\n\t// Priority 2: Header Accept-Language\n\tif acceptLang := c.GetHeader(\"Accept-Language\"); acceptLang != \"\" {\n\t\t// Parse Accept-Language header (e.g., \"en-US,en;q=0.9,zh;q=0.8\")\n\t\t// Take the first language\n\t\tparts := strings.Split(acceptLang, \",\")\n\t\tif len(parts) > 0 {\n\t\t\t// Remove quality value if present\n\t\t\tlang := strings.Split(parts[0], \";\")[0]\n\t\t\treturn strings.ToLower(strings.TrimSpace(lang))\n\t\t}\n\t}\n\n\t// Priority 3: From CompletionRequest metadata\n\tif req != nil && req.Metadata != nil {\n\t\tif locale, ok := req.Metadata[\"locale\"]; ok {\n\t\t\tif localeStr, ok := locale.(string); ok && localeStr != \"\" {\n\t\t\t\treturn strings.ToLower(localeStr)\n\t\t\t}\n\t\t}\n\t}\n\n\treturn \"\"\n}\n\n// GetTheme extracts theme from request with priority:\n// 1. Query parameter \"theme\"\n// 2. Header \"X-Yao-Theme\"\n// 3. CompletionRequest metadata \"theme\" (from payload)\nfunc GetTheme(c *gin.Context, req *CompletionRequest) string {\n\t// Priority 1: Query parameter\n\tif theme := c.Query(\"theme\"); theme != \"\" {\n\t\treturn strings.ToLower(theme)\n\t}\n\n\t// Priority 2: Header\n\tif theme := c.GetHeader(\"X-Yao-Theme\"); theme != \"\" {\n\t\treturn strings.ToLower(theme)\n\t}\n\n\t// Priority 3: From CompletionRequest metadata\n\tif req != nil && req.Metadata != nil {\n\t\tif theme, ok := req.Metadata[\"theme\"]; ok {\n\t\t\tif themeStr, ok := theme.(string); ok && themeStr != \"\" {\n\t\t\t\treturn strings.ToLower(themeStr)\n\t\t\t}\n\t\t}\n\t}\n\n\treturn \"\"\n}\n\n// GetReferer extracts referer from request with priority:\n// 1. Query parameter \"referer\"\n// 2. Header \"X-Yao-Referer\"\n// 3. CompletionRequest metadata \"referer\" (from payload)\n// 4. Default to \"api\"\nfunc GetReferer(c *gin.Context, req *CompletionRequest) string {\n\t// Priority 1: Query parameter\n\tif referer := c.Query(\"referer\"); referer != \"\" {\n\t\treturn validateReferer(referer)\n\t}\n\n\t// Priority 2: Header\n\tif referer := c.GetHeader(\"X-Yao-Referer\"); referer != \"\" {\n\t\treturn validateReferer(referer)\n\t}\n\n\t// Priority 3: From CompletionRequest metadata\n\tif req != nil && req.Metadata != nil {\n\t\tif referer, ok := req.Metadata[\"referer\"]; ok {\n\t\t\tif refererStr, ok := referer.(string); ok && refererStr != \"\" {\n\t\t\t\treturn validateReferer(refererStr)\n\t\t\t}\n\t\t}\n\t}\n\n\t// Priority 4: Default\n\treturn RefererAPI\n}\n\n// GetAccept extracts accept type from request with priority:\n// 1. Query parameter \"accept\"\n// 2. Header \"X-Yao-Accept\"\n// 3. CompletionRequest metadata \"accept\" (from payload)\n// 4. Default to \"standard\" (OpenAI-compatible format)\nfunc GetAccept(c *gin.Context, req *CompletionRequest) Accept {\n\t// Priority 1: Query parameter\n\tif accept := c.Query(\"accept\"); accept != \"\" {\n\t\treturn validateAccept(accept)\n\t}\n\n\t// Priority 2: Header\n\tif accept := c.GetHeader(\"X-Yao-Accept\"); accept != \"\" {\n\t\treturn validateAccept(accept)\n\t}\n\n\t// Priority 3: From CompletionRequest metadata\n\tif req != nil && req.Metadata != nil {\n\t\tif accept, ok := req.Metadata[\"accept\"]; ok {\n\t\t\tif acceptStr, ok := accept.(string); ok && acceptStr != \"\" {\n\t\t\t\treturn validateAccept(acceptStr)\n\t\t\t}\n\t\t}\n\t}\n\n\t// Priority 4: Default to \"standard\" (OpenAI-compatible format)\n\treturn AcceptStandard\n\n\t// // Future: Parse from User-Agent if needed\n\t// userAgent := c.GetHeader(\"User-Agent\")\n\t// clientType := getClientType(userAgent)\n\t// return parseAccept(clientType)\n}\n\n// GetChatID get the chat ID from the request\n// Priority:\n// 1. Query parameter \"chat_id\"\n// 2. Header \"X-Yao-Chat\"\n// 3. CompletionRequest metadata \"chat_id\" (from payload)\n// 4. Generate from messages using GetChatIDByMessages\nfunc GetChatID(c *gin.Context, cache store.Store, req *CompletionRequest) (string, error) {\n\t// Priority 1: Query parameter chat_id\n\tif chatID := c.Query(\"chat_id\"); chatID != \"\" {\n\t\treturn chatID, nil\n\t}\n\n\t// Priority 2: Header X-Yao-Chat\n\tif chatID := c.GetHeader(\"X-Yao-Chat\"); chatID != \"\" {\n\t\treturn chatID, nil\n\t}\n\n\t// Priority 3: From CompletionRequest metadata\n\tif req != nil && req.Metadata != nil {\n\t\tif chatID, ok := req.Metadata[\"chat_id\"]; ok {\n\t\t\tif chatIDStr, ok := chatID.(string); ok && chatIDStr != \"\" {\n\t\t\t\treturn chatIDStr, nil\n\t\t\t}\n\t\t}\n\t}\n\n\t// Priority 4: Generate from messages\n\tmessages, err := GetMessages(c, req)\n\tif err != nil {\n\t\treturn \"\", fmt.Errorf(\"failed to get messages for chat ID generation: %w\", err)\n\t}\n\n\tchatID, err := GetChatIDByMessages(cache, messages)\n\tif err != nil {\n\t\treturn \"\", fmt.Errorf(\"failed to generate chat ID from messages: %w\", err)\n\t}\n\n\treturn chatID, nil\n}\n\n// GetRoute extracts route from request with priority:\n// 1. Query parameter \"route\"\n// 2. Header \"X-Yao-Route\"\n// 3. CompletionRequest.Route (from payload)\nfunc GetRoute(c *gin.Context, req *CompletionRequest) string {\n\t// Priority 1: Query parameter\n\tif route := c.Query(\"route\"); route != \"\" {\n\t\treturn route\n\t}\n\n\t// Priority 2: Header\n\tif route := c.GetHeader(\"X-Yao-Route\"); route != \"\" {\n\t\treturn route\n\t}\n\n\t// Priority 3: From CompletionRequest\n\tif req != nil && req.Route != \"\" {\n\t\treturn req.Route\n\t}\n\n\treturn \"\"\n}\n\n// GetMode extracts mode from request with priority:\n// 1. Query parameter \"mode\"\n// 2. Header \"X-Yao-Mode\"\n// 3. CompletionRequest metadata \"mode\" (from payload)\nfunc GetMode(c *gin.Context, req *CompletionRequest) string {\n\t// Priority 1: Query parameter\n\tif mode := c.Query(\"mode\"); mode != \"\" {\n\t\treturn mode\n\t}\n\n\t// Priority 2: Header\n\tif mode := c.GetHeader(\"X-Yao-Mode\"); mode != \"\" {\n\t\treturn mode\n\t}\n\n\t// Priority 3: From CompletionRequest metadata\n\tif req != nil && req.Metadata != nil {\n\t\tif mode, ok := req.Metadata[\"mode\"]; ok {\n\t\t\tif modeStr, ok := mode.(string); ok && modeStr != \"\" {\n\t\t\t\treturn modeStr\n\t\t\t}\n\t\t}\n\t}\n\n\treturn \"\"\n}\n\n// GetSkip extracts skip configuration from request with priority:\n// 1. CompletionRequest.Skip (from payload body) - Priority\n// 2. Individual query parameters: \"skip_history\", \"skip_trace\"\nfunc GetSkip(c *gin.Context, req *CompletionRequest) *Skip {\n\t// Priority 1: From CompletionRequest body (most direct)\n\tif req != nil && req.Skip != nil {\n\t\treturn req.Skip\n\t}\n\n\t// Priority 2: Individual query parameters (recommended for query usage)\n\tskipHistory := c.Query(\"skip_history\") == \"true\" || c.Query(\"skip_history\") == \"1\"\n\tskipTrace := c.Query(\"skip_trace\") == \"true\" || c.Query(\"skip_trace\") == \"1\"\n\n\t// Check if any skip parameter is set\n\tif c.Query(\"skip_history\") != \"\" || c.Query(\"skip_trace\") != \"\" {\n\t\treturn &Skip{\n\t\t\tHistory: skipHistory,\n\t\t\tTrace: skipTrace,\n\t\t}\n\t}\n\n\treturn nil\n}\n\n// GetMetadata extracts metadata from request with priority:\n// 1. Query parameter \"metadata\" (JSON string)\n// 2. Header \"X-Yao-Metadata\" (Base64 encoded JSON string)\n// 3. CompletionRequest.Metadata (from payload)\nfunc GetMetadata(c *gin.Context, req *CompletionRequest) map[string]interface{} {\n\t// Priority 1: Query parameter (JSON string)\n\tif metadataJSON := c.Query(\"metadata\"); metadataJSON != \"\" {\n\t\tvar metadata map[string]interface{}\n\t\tif err := json.Unmarshal([]byte(metadataJSON), &metadata); err == nil {\n\t\t\treturn metadata\n\t\t}\n\t}\n\n\t// Priority 2: Header (Base64 encoded JSON string)\n\tif metadataBase64 := c.GetHeader(\"X-Yao-Metadata\"); metadataBase64 != \"\" {\n\t\t// Try to decode Base64\n\t\tif decoded, err := base64.StdEncoding.DecodeString(metadataBase64); err == nil {\n\t\t\tvar metadata map[string]interface{}\n\t\t\tif err := json.Unmarshal(decoded, &metadata); err == nil {\n\t\t\t\treturn metadata\n\t\t\t}\n\t\t}\n\t\t// Fallback: try to parse as plain JSON\n\t\tvar metadata map[string]interface{}\n\t\tif err := json.Unmarshal([]byte(metadataBase64), &metadata); err == nil {\n\t\t\treturn metadata\n\t\t}\n\t}\n\n\t// Priority 3: From CompletionRequest\n\tif req != nil && req.Metadata != nil {\n\t\treturn req.Metadata\n\t}\n\n\treturn nil\n}\n\n// parseCompletionRequestData extracts CompletionRequest from the request\n// Data can be passed via:\n// 1. Request body (JSON payload) - Priority\n// 2. Query parameters\nfunc parseCompletionRequestData(c *gin.Context) (*CompletionRequest, error) {\n\tvar req CompletionRequest\n\n\t// Try to parse from request body first\n\tif c.Request.Body != nil {\n\t\tbody, err := io.ReadAll(c.Request.Body)\n\t\tif err != nil {\n\t\t\treturn nil, fmt.Errorf(\"failed to read request body: %w\", err)\n\t\t}\n\n\t\t// Restore body for further processing\n\t\tc.Request.Body = io.NopCloser(bytes.NewBuffer(body))\n\n\t\t// If body is not empty, try to parse it\n\t\tif len(body) > 0 {\n\t\t\tif err := json.Unmarshal(body, &req); err != nil {\n\t\t\t\treturn nil, fmt.Errorf(\"failed to parse completion request from body: %w\", err)\n\t\t\t}\n\n\t\t\t// If we got valid data from body, validate and return\n\t\t\t// Model is optional if assistant_id can be extracted later\n\t\t\tif len(req.Messages) > 0 {\n\t\t\t\treturn &req, nil\n\t\t\t}\n\t\t}\n\t}\n\n\t// Fallback: Try to parse from query parameters\n\t// Model is optional (can be extracted from assistant_id)\n\tmodel := c.Query(\"model\")\n\treq.Model = model\n\n\t// Messages (required, must be JSON string in query)\n\tmessagesJSON := c.Query(\"messages\")\n\tif messagesJSON == \"\" {\n\t\treturn nil, fmt.Errorf(\"messages field is required\")\n\t}\n\n\tvar messages []Message\n\tif err := json.Unmarshal([]byte(messagesJSON), &messages); err != nil {\n\t\treturn nil, fmt.Errorf(\"failed to parse messages from query: %w\", err)\n\t}\n\tif len(messages) == 0 {\n\t\treturn nil, fmt.Errorf(\"messages field must not be empty\")\n\t}\n\treq.Messages = messages\n\n\t// Optional fields from query\n\tif tempStr := c.Query(\"temperature\"); tempStr != \"\" {\n\t\tvar temp float64\n\t\tif _, err := fmt.Sscanf(tempStr, \"%f\", &temp); err == nil {\n\t\t\treq.Temperature = &temp\n\t\t}\n\t}\n\n\tif maxTokensStr := c.Query(\"max_tokens\"); maxTokensStr != \"\" {\n\t\tvar maxTokens int\n\t\tif _, err := fmt.Sscanf(maxTokensStr, \"%d\", &maxTokens); err == nil {\n\t\t\treq.MaxTokens = &maxTokens\n\t\t}\n\t}\n\n\tif maxCompletionTokensStr := c.Query(\"max_completion_tokens\"); maxCompletionTokensStr != \"\" {\n\t\tvar maxCompletionTokens int\n\t\tif _, err := fmt.Sscanf(maxCompletionTokensStr, \"%d\", &maxCompletionTokens); err == nil {\n\t\t\treq.MaxCompletionTokens = &maxCompletionTokens\n\t\t}\n\t}\n\n\tif streamStr := c.Query(\"stream\"); streamStr != \"\" {\n\t\tstream := streamStr == \"true\" || streamStr == \"1\"\n\t\treq.Stream = &stream\n\t}\n\n\t// Audio config from query (JSON string)\n\tif audioJSON := c.Query(\"audio\"); audioJSON != \"\" {\n\t\tvar audio AudioConfig\n\t\tif err := json.Unmarshal([]byte(audioJSON), &audio); err == nil {\n\t\t\treq.Audio = &audio\n\t\t}\n\t}\n\n\t// Stream options from query (JSON string)\n\tif streamOptionsJSON := c.Query(\"stream_options\"); streamOptionsJSON != \"\" {\n\t\tvar streamOptions StreamOptions\n\t\tif err := json.Unmarshal([]byte(streamOptionsJSON), &streamOptions); err == nil {\n\t\t\treq.StreamOptions = &streamOptions\n\t\t}\n\t}\n\n\t// Metadata from query (JSON string)\n\tif metadataJSON := c.Query(\"metadata\"); metadataJSON != \"\" {\n\t\tvar metadata map[string]interface{}\n\t\tif err := json.Unmarshal([]byte(metadataJSON), &metadata); err == nil {\n\t\t\treq.Metadata = metadata\n\t\t}\n\t}\n\n\treturn &req, nil\n}\n" }, { "path": "agent/context/openapi_test.go", "content": "package context_test\n\nimport (\n\t\"bytes\"\n\t\"encoding/json\"\n\t\"io\"\n\t\"net/http/httptest\"\n\t\"testing\"\n\n\t\"github.com/gin-gonic/gin\"\n\t\"github.com/yaoapp/gou/store\"\n\t\"github.com/yaoapp/yao/agent/context\"\n\t\"github.com/yaoapp/yao/config\"\n\t\"github.com/yaoapp/yao/test\"\n)\n\n// parseCompletionRequestData is a helper function for tests to parse completion request data\nfunc parseCompletionRequestData(c *gin.Context) (*context.CompletionRequest, error) {\n\tvar req context.CompletionRequest\n\n\tif c.Request.Body != nil {\n\t\tbody, err := io.ReadAll(c.Request.Body)\n\t\tif err != nil {\n\t\t\treturn nil, err\n\t\t}\n\t\tc.Request.Body = io.NopCloser(bytes.NewBuffer(body))\n\t\tif len(body) > 0 {\n\t\t\tif err := json.Unmarshal(body, &req); err != nil {\n\t\t\t\treturn nil, err\n\t\t\t}\n\t\t\tif len(req.Messages) > 0 {\n\t\t\t\treturn &req, nil\n\t\t\t}\n\t\t}\n\t}\n\treturn &req, nil\n}\n\nfunc TestGetMessages_FromBody(t *testing.T) {\n\ttest.Prepare(t, config.Conf)\n\tdefer test.Clean()\n\n\tgin.SetMode(gin.TestMode)\n\n\tmessages := []context.Message{\n\t\t{\n\t\t\tRole: context.RoleUser,\n\t\t\tContent: \"Hello, world!\",\n\t\t},\n\t\t{\n\t\t\tRole: context.RoleAssistant,\n\t\t\tContent: \"Hi there!\",\n\t\t},\n\t}\n\n\trequestBody := map[string]interface{}{\n\t\t\"messages\": messages,\n\t\t\"model\": \"gpt-4\",\n\t}\n\n\tbodyBytes, _ := json.Marshal(requestBody)\n\n\treq := httptest.NewRequest(\"POST\", \"/chat/completions\", bytes.NewBuffer(bodyBytes))\n\treq.Header.Set(\"Content-Type\", \"application/json\")\n\tw := httptest.NewRecorder()\n\tc, _ := gin.CreateTestContext(w)\n\tc.Request = req\n\n\t// Parse request first\n\tcompletionReq, _ := parseCompletionRequestData(c)\n\n\tresult, err := context.GetMessages(c, completionReq)\n\tif err != nil {\n\t\tt.Fatalf(\"Failed to get messages: %v\", err)\n\t}\n\n\tif len(result) != 2 {\n\t\tt.Errorf(\"Expected 2 messages, got %d\", len(result))\n\t}\n\n\tif result[0].Role != context.RoleUser {\n\t\tt.Errorf(\"Expected first message role to be %s, got %s\", context.RoleUser, result[0].Role)\n\t}\n}\n\nfunc TestGetMessages_FromQuery(t *testing.T) {\n\ttest.Prepare(t, config.Conf)\n\tdefer test.Clean()\n\n\tgin.SetMode(gin.TestMode)\n\n\tmessages := []context.Message{\n\t\t{\n\t\t\tRole: context.RoleUser,\n\t\t\tContent: \"Test message\",\n\t\t},\n\t}\n\n\tmessagesJSON, _ := json.Marshal(messages)\n\n\treq := httptest.NewRequest(\"GET\", \"/chat/completions\", nil)\n\tq := req.URL.Query()\n\tq.Add(\"messages\", string(messagesJSON))\n\treq.URL.RawQuery = q.Encode()\n\n\tw := httptest.NewRecorder()\n\tc, _ := gin.CreateTestContext(w)\n\tc.Request = req\n\n\tresult, err := context.GetMessages(c, nil)\n\tif err != nil {\n\t\tt.Fatalf(\"Failed to get messages: %v\", err)\n\t}\n\n\tif len(result) != 1 {\n\t\tt.Errorf(\"Expected 1 message, got %d\", len(result))\n\t}\n}\n\nfunc TestGetMessages_EmptyMessages(t *testing.T) {\n\ttest.Prepare(t, config.Conf)\n\tdefer test.Clean()\n\n\tgin.SetMode(gin.TestMode)\n\n\trequestBody := map[string]interface{}{\n\t\t\"messages\": []context.Message{},\n\t\t\"model\": \"gpt-4\",\n\t}\n\n\tbodyBytes, _ := json.Marshal(requestBody)\n\n\treq := httptest.NewRequest(\"POST\", \"/chat/completions\", bytes.NewBuffer(bodyBytes))\n\treq.Header.Set(\"Content-Type\", \"application/json\")\n\tw := httptest.NewRecorder()\n\tc, _ := gin.CreateTestContext(w)\n\tc.Request = req\n\n\tcompletionReq, _ := parseCompletionRequestData(c)\n\n\t_, err := context.GetMessages(c, completionReq)\n\tif err == nil {\n\t\tt.Error(\"Expected error for empty messages\")\n\t}\n}\n\nfunc TestGetChatID_FromQuery(t *testing.T) {\n\ttest.Prepare(t, config.Conf)\n\tdefer test.Clean()\n\n\tgin.SetMode(gin.TestMode)\n\n\tcache, err := store.Get(\"__yao.agent.cache\")\n\tif err != nil {\n\t\tt.Fatalf(\"Failed to get cache: %v\", err)\n\t}\n\n\texpectedChatID := \"test-chat-123\"\n\n\treq := httptest.NewRequest(\"GET\", \"/chat/completions?chat_id=\"+expectedChatID, nil)\n\tw := httptest.NewRecorder()\n\tc, _ := gin.CreateTestContext(w)\n\tc.Request = req\n\n\tchatID, err := context.GetChatID(c, cache, nil)\n\tif err != nil {\n\t\tt.Fatalf(\"Failed to get chat ID: %v\", err)\n\t}\n\n\tif chatID != expectedChatID {\n\t\tt.Errorf(\"Expected chat ID %s, got %s\", expectedChatID, chatID)\n\t}\n}\n\nfunc TestGetChatID_FromHeader(t *testing.T) {\n\ttest.Prepare(t, config.Conf)\n\tdefer test.Clean()\n\n\tgin.SetMode(gin.TestMode)\n\n\tcache, err := store.Get(\"__yao.agent.cache\")\n\tif err != nil {\n\t\tt.Fatalf(\"Failed to get cache: %v\", err)\n\t}\n\n\texpectedChatID := \"header-chat-456\"\n\n\treq := httptest.NewRequest(\"GET\", \"/chat/completions\", nil)\n\treq.Header.Set(\"X-Yao-Chat\", expectedChatID)\n\tw := httptest.NewRecorder()\n\tc, _ := gin.CreateTestContext(w)\n\tc.Request = req\n\n\tchatID, err := context.GetChatID(c, cache, nil)\n\tif err != nil {\n\t\tt.Fatalf(\"Failed to get chat ID: %v\", err)\n\t}\n\n\tif chatID != expectedChatID {\n\t\tt.Errorf(\"Expected chat ID %s, got %s\", expectedChatID, chatID)\n\t}\n}\n\nfunc TestGetChatID_FromMetadata(t *testing.T) {\n\ttest.Prepare(t, config.Conf)\n\tdefer test.Clean()\n\n\tgin.SetMode(gin.TestMode)\n\n\tcache, err := store.Get(\"__yao.agent.cache\")\n\tif err != nil {\n\t\tt.Fatalf(\"Failed to get cache: %v\", err)\n\t}\n\n\texpectedChatID := \"metadata-chat-789\"\n\n\trequestBody := map[string]interface{}{\n\t\t\"model\": \"gpt-4\",\n\t\t\"messages\": []map[string]interface{}{\n\t\t\t{\"role\": \"user\", \"content\": \"Test\"},\n\t\t},\n\t\t\"metadata\": map[string]interface{}{\n\t\t\t\"chat_id\": expectedChatID,\n\t\t},\n\t}\n\n\tbodyBytes, _ := json.Marshal(requestBody)\n\n\treq := httptest.NewRequest(\"POST\", \"/chat/completions\", bytes.NewBuffer(bodyBytes))\n\treq.Header.Set(\"Content-Type\", \"application/json\")\n\tw := httptest.NewRecorder()\n\tc, _ := gin.CreateTestContext(w)\n\tc.Request = req\n\n\tcompletionReq, _ := parseCompletionRequestData(c)\n\n\tchatID, err := context.GetChatID(c, cache, completionReq)\n\tif err != nil {\n\t\tt.Fatalf(\"Failed to get chat ID: %v\", err)\n\t}\n\n\tif chatID != expectedChatID {\n\t\tt.Errorf(\"Expected chat ID %s, got %s\", expectedChatID, chatID)\n\t}\n}\n\nfunc TestGetChatID_FromMessages(t *testing.T) {\n\ttest.Prepare(t, config.Conf)\n\tdefer test.Clean()\n\n\tgin.SetMode(gin.TestMode)\n\n\tcache, err := store.Get(\"__yao.agent.cache\")\n\tif err != nil {\n\t\tt.Fatalf(\"Failed to get cache: %v\", err)\n\t}\n\tcache.Clear()\n\n\t// First request with one user message\n\tmessages1 := []context.Message{\n\t\t{\n\t\t\tRole: context.RoleUser,\n\t\t\tContent: \"First message\",\n\t\t},\n\t}\n\n\trequestBody1 := map[string]interface{}{\n\t\t\"model\": \"gpt-4\",\n\t\t\"messages\": messages1,\n\t}\n\n\tbodyBytes1, _ := json.Marshal(requestBody1)\n\n\treq := httptest.NewRequest(\"POST\", \"/chat/completions\", bytes.NewBuffer(bodyBytes1))\n\treq.Header.Set(\"Content-Type\", \"application/json\")\n\tw := httptest.NewRecorder()\n\tc, _ := gin.CreateTestContext(w)\n\tc.Request = req\n\n\tcompletionReq1, _ := parseCompletionRequestData(c)\n\n\tchatID1, err := context.GetChatID(c, cache, completionReq1)\n\tif err != nil {\n\t\tt.Fatalf(\"Failed to get chat ID: %v\", err)\n\t}\n\n\tif chatID1 == \"\" {\n\t\tt.Error(\"Expected non-empty chat ID\")\n\t}\n\n\t// Second request with two user messages (continuation)\n\tmessages2 := []context.Message{\n\t\t{\n\t\t\tRole: context.RoleUser,\n\t\t\tContent: \"First message\",\n\t\t},\n\t\t{\n\t\t\tRole: context.RoleUser,\n\t\t\tContent: \"Second message\",\n\t\t},\n\t}\n\n\trequestBody2 := map[string]interface{}{\n\t\t\"model\": \"gpt-4\",\n\t\t\"messages\": messages2,\n\t}\n\n\tbodyBytes2, _ := json.Marshal(requestBody2)\n\n\treq2 := httptest.NewRequest(\"POST\", \"/chat/completions\", bytes.NewBuffer(bodyBytes2))\n\treq2.Header.Set(\"Content-Type\", \"application/json\")\n\tw2 := httptest.NewRecorder()\n\tc2, _ := gin.CreateTestContext(w2)\n\tc2.Request = req2\n\n\tcompletionReq2, _ := parseCompletionRequestData(c2)\n\n\tchatID2, err := context.GetChatID(c2, cache, completionReq2)\n\tif err != nil {\n\t\tt.Fatalf(\"Failed to get chat ID second time: %v\", err)\n\t}\n\n\t// Should get same chat ID (continuation of conversation)\n\tif chatID1 != chatID2 {\n\t\tt.Errorf(\"Expected same chat ID for continuation, got %s and %s\", chatID1, chatID2)\n\t}\n}\n\nfunc TestGetChatID_Priority(t *testing.T) {\n\ttest.Prepare(t, config.Conf)\n\tdefer test.Clean()\n\n\tgin.SetMode(gin.TestMode)\n\n\tcache, err := store.Get(\"__yao.agent.cache\")\n\tif err != nil {\n\t\tt.Fatalf(\"Failed to get cache: %v\", err)\n\t}\n\n\tqueryChatID := \"query-chat-id\"\n\theaderChatID := \"header-chat-id\"\n\tmetadataChatID := \"metadata-chat-id\"\n\n\tmessages := []context.Message{\n\t\t{\n\t\t\tRole: context.RoleUser,\n\t\t\tContent: \"This should not be used\",\n\t\t},\n\t}\n\n\trequestBody := map[string]interface{}{\n\t\t\"model\": \"gpt-4\",\n\t\t\"messages\": messages,\n\t\t\"metadata\": map[string]interface{}{\n\t\t\t\"chat_id\": metadataChatID,\n\t\t},\n\t}\n\n\tbodyBytes, _ := json.Marshal(requestBody)\n\n\t// Test priority: query > header > metadata > messages\n\treq := httptest.NewRequest(\"POST\", \"/chat/completions?chat_id=\"+queryChatID, bytes.NewBuffer(bodyBytes))\n\treq.Header.Set(\"Content-Type\", \"application/json\")\n\treq.Header.Set(\"X-Yao-Chat\", headerChatID)\n\tw := httptest.NewRecorder()\n\tc, _ := gin.CreateTestContext(w)\n\tc.Request = req\n\n\tcompletionReq, _ := parseCompletionRequestData(c)\n\n\tchatID, err := context.GetChatID(c, cache, completionReq)\n\tif err != nil {\n\t\tt.Fatalf(\"Failed to get chat ID: %v\", err)\n\t}\n\n\tif chatID != queryChatID {\n\t\tt.Errorf(\"Expected query parameter to take priority, got %s instead of %s\", chatID, queryChatID)\n\t}\n}\n\nfunc TestGetLocale_FromQuery(t *testing.T) {\n\tgin.SetMode(gin.TestMode)\n\n\treq := httptest.NewRequest(\"GET\", \"/chat/completions?locale=zh-CN\", nil)\n\tw := httptest.NewRecorder()\n\tc, _ := gin.CreateTestContext(w)\n\tc.Request = req\n\n\tlocale := context.GetLocale(c, nil)\n\tif locale != \"zh-cn\" {\n\t\tt.Errorf(\"Expected locale 'zh-cn', got '%s'\", locale)\n\t}\n}\n\nfunc TestGetLocale_FromHeader(t *testing.T) {\n\tgin.SetMode(gin.TestMode)\n\n\treq := httptest.NewRequest(\"GET\", \"/chat/completions\", nil)\n\treq.Header.Set(\"Accept-Language\", \"en-US,en;q=0.9,zh;q=0.8\")\n\tw := httptest.NewRecorder()\n\tc, _ := gin.CreateTestContext(w)\n\tc.Request = req\n\n\tlocale := context.GetLocale(c, nil)\n\tif locale != \"en-us\" {\n\t\tt.Errorf(\"Expected locale 'en-us', got '%s'\", locale)\n\t}\n}\n\nfunc TestGetLocale_FromMetadata(t *testing.T) {\n\tgin.SetMode(gin.TestMode)\n\n\treq := httptest.NewRequest(\"POST\", \"/chat/completions\", nil)\n\tw := httptest.NewRecorder()\n\tc, _ := gin.CreateTestContext(w)\n\tc.Request = req\n\n\tcompletionReq := &context.CompletionRequest{\n\t\tMetadata: map[string]interface{}{\n\t\t\t\"locale\": \"ja-JP\",\n\t\t},\n\t}\n\n\tlocale := context.GetLocale(c, completionReq)\n\tif locale != \"ja-jp\" {\n\t\tt.Errorf(\"Expected locale 'ja-jp' from metadata, got '%s'\", locale)\n\t}\n}\n\nfunc TestGetLocale_Priority(t *testing.T) {\n\tgin.SetMode(gin.TestMode)\n\n\treq := httptest.NewRequest(\"GET\", \"/chat/completions?locale=fr-FR\", nil)\n\treq.Header.Set(\"Accept-Language\", \"en-US\")\n\tw := httptest.NewRecorder()\n\tc, _ := gin.CreateTestContext(w)\n\tc.Request = req\n\n\tcompletionReq := &context.CompletionRequest{\n\t\tMetadata: map[string]interface{}{\n\t\t\t\"locale\": \"de-DE\",\n\t\t},\n\t}\n\n\tlocale := context.GetLocale(c, completionReq)\n\tif locale != \"fr-fr\" {\n\t\tt.Errorf(\"Expected query parameter to take priority, got '%s'\", locale)\n\t}\n}\n\nfunc TestGetTheme_FromQuery(t *testing.T) {\n\tgin.SetMode(gin.TestMode)\n\n\treq := httptest.NewRequest(\"GET\", \"/chat/completions?theme=dark\", nil)\n\tw := httptest.NewRecorder()\n\tc, _ := gin.CreateTestContext(w)\n\tc.Request = req\n\n\ttheme := context.GetTheme(c, nil)\n\tif theme != \"dark\" {\n\t\tt.Errorf(\"Expected theme 'dark', got '%s'\", theme)\n\t}\n}\n\nfunc TestGetTheme_FromHeader(t *testing.T) {\n\tgin.SetMode(gin.TestMode)\n\n\treq := httptest.NewRequest(\"GET\", \"/chat/completions\", nil)\n\treq.Header.Set(\"X-Yao-Theme\", \"light\")\n\tw := httptest.NewRecorder()\n\tc, _ := gin.CreateTestContext(w)\n\tc.Request = req\n\n\ttheme := context.GetTheme(c, nil)\n\tif theme != \"light\" {\n\t\tt.Errorf(\"Expected theme 'light', got '%s'\", theme)\n\t}\n}\n\nfunc TestGetTheme_FromMetadata(t *testing.T) {\n\tgin.SetMode(gin.TestMode)\n\n\treq := httptest.NewRequest(\"POST\", \"/chat/completions\", nil)\n\tw := httptest.NewRecorder()\n\tc, _ := gin.CreateTestContext(w)\n\tc.Request = req\n\n\tcompletionReq := &context.CompletionRequest{\n\t\tMetadata: map[string]interface{}{\n\t\t\t\"theme\": \"auto\",\n\t\t},\n\t}\n\n\ttheme := context.GetTheme(c, completionReq)\n\tif theme != \"auto\" {\n\t\tt.Errorf(\"Expected theme 'auto' from metadata, got '%s'\", theme)\n\t}\n}\n\nfunc TestGetReferer_FromMetadata(t *testing.T) {\n\tgin.SetMode(gin.TestMode)\n\n\treq := httptest.NewRequest(\"POST\", \"/chat/completions\", nil)\n\tw := httptest.NewRecorder()\n\tc, _ := gin.CreateTestContext(w)\n\tc.Request = req\n\n\tcompletionReq := &context.CompletionRequest{\n\t\tMetadata: map[string]interface{}{\n\t\t\t\"referer\": \"tool\",\n\t\t},\n\t}\n\n\treferer := context.GetReferer(c, completionReq)\n\tif referer != context.RefererTool {\n\t\tt.Errorf(\"Expected referer 'tool' from metadata, got '%s'\", referer)\n\t}\n}\n\nfunc TestGetAccept_FromQuery(t *testing.T) {\n\tgin.SetMode(gin.TestMode)\n\n\treq := httptest.NewRequest(\"GET\", \"/chat/completions?accept=cui-web\", nil)\n\tw := httptest.NewRecorder()\n\tc, _ := gin.CreateTestContext(w)\n\tc.Request = req\n\n\taccept := context.GetAccept(c, nil)\n\tif accept != context.AcceptWebCUI {\n\t\tt.Errorf(\"Expected accept 'cui-web' from query, got '%s'\", accept)\n\t}\n}\n\nfunc TestGetAccept_FromHeader(t *testing.T) {\n\tgin.SetMode(gin.TestMode)\n\n\treq := httptest.NewRequest(\"GET\", \"/chat/completions\", nil)\n\treq.Header.Set(\"X-Yao-Accept\", \"cui-desktop\")\n\tw := httptest.NewRecorder()\n\tc, _ := gin.CreateTestContext(w)\n\tc.Request = req\n\n\taccept := context.GetAccept(c, nil)\n\tif accept != context.AcceptDesktopCUI {\n\t\tt.Errorf(\"Expected accept 'cui-desktop' from header, got '%s'\", accept)\n\t}\n}\n\nfunc TestGetAccept_FromMetadata(t *testing.T) {\n\tgin.SetMode(gin.TestMode)\n\n\treq := httptest.NewRequest(\"POST\", \"/chat/completions\", nil)\n\tw := httptest.NewRecorder()\n\tc, _ := gin.CreateTestContext(w)\n\tc.Request = req\n\n\tcompletionReq := &context.CompletionRequest{\n\t\tMetadata: map[string]interface{}{\n\t\t\t\"accept\": \"cui-native\",\n\t\t},\n\t}\n\n\taccept := context.GetAccept(c, completionReq)\n\tif accept != context.AccepNativeCUI {\n\t\tt.Errorf(\"Expected accept 'cui-native' from metadata, got '%s'\", accept)\n\t}\n}\n\nfunc TestGetAccept_Default(t *testing.T) {\n\tgin.SetMode(gin.TestMode)\n\n\treq := httptest.NewRequest(\"GET\", \"/chat/completions\", nil)\n\tw := httptest.NewRecorder()\n\tc, _ := gin.CreateTestContext(w)\n\tc.Request = req\n\n\taccept := context.GetAccept(c, nil)\n\tif accept != context.AcceptStandard {\n\t\tt.Errorf(\"Expected default accept 'standard', got '%s'\", accept)\n\t}\n}\n\nfunc TestGetAccept_Priority(t *testing.T) {\n\tgin.SetMode(gin.TestMode)\n\n\treq := httptest.NewRequest(\"GET\", \"/chat/completions?accept=cui-web\", nil)\n\treq.Header.Set(\"X-Yao-Accept\", \"cui-desktop\")\n\tw := httptest.NewRecorder()\n\tc, _ := gin.CreateTestContext(w)\n\tc.Request = req\n\n\tcompletionReq := &context.CompletionRequest{\n\t\tMetadata: map[string]interface{}{\n\t\t\t\"accept\": \"cui-native\",\n\t\t},\n\t}\n\n\taccept := context.GetAccept(c, completionReq)\n\tif accept != context.AcceptWebCUI {\n\t\tt.Errorf(\"Expected query parameter to take priority, got '%s'\", accept)\n\t}\n}\n\nfunc TestGetAssistantID_FromModel(t *testing.T) {\n\tgin.SetMode(gin.TestMode)\n\n\treq := httptest.NewRequest(\"POST\", \"/chat/completions\", nil)\n\tw := httptest.NewRecorder()\n\tc, _ := gin.CreateTestContext(w)\n\tc.Request = req\n\n\tcompletionReq := &context.CompletionRequest{\n\t\tModel: \"gpt-4-turbo-yao_myassistant\",\n\t}\n\n\tassistantID, err := context.GetAssistantID(c, completionReq)\n\tif err != nil {\n\t\tt.Fatalf(\"Failed to get assistant ID: %v\", err)\n\t}\n\n\tif assistantID != \"myassistant\" {\n\t\tt.Errorf(\"Expected assistant ID 'myassistant', got '%s'\", assistantID)\n\t}\n}\n\nfunc TestGetAssistantID_Priority(t *testing.T) {\n\tgin.SetMode(gin.TestMode)\n\n\treq := httptest.NewRequest(\"GET\", \"/chat/completions?assistant_id=from_query\", nil)\n\treq.Header.Set(\"X-Yao-Assistant\", \"from_header\")\n\tw := httptest.NewRecorder()\n\tc, _ := gin.CreateTestContext(w)\n\tc.Request = req\n\n\tcompletionReq := &context.CompletionRequest{\n\t\tModel: \"gpt-4-yao_from_model\",\n\t}\n\n\tassistantID, err := context.GetAssistantID(c, completionReq)\n\tif err != nil {\n\t\tt.Fatalf(\"Failed to get assistant ID: %v\", err)\n\t}\n\n\tif assistantID != \"from_query\" {\n\t\tt.Errorf(\"Expected query parameter to take priority, got '%s'\", assistantID)\n\t}\n}\n\nfunc TestGetRoute_FromQuery(t *testing.T) {\n\tgin.SetMode(gin.TestMode)\n\n\treq := httptest.NewRequest(\"GET\", \"/chat/completions?route=/dashboard/home\", nil)\n\tw := httptest.NewRecorder()\n\tc, _ := gin.CreateTestContext(w)\n\tc.Request = req\n\n\troute := context.GetRoute(c, nil)\n\tif route != \"/dashboard/home\" {\n\t\tt.Errorf(\"Expected route '/dashboard/home', got '%s'\", route)\n\t}\n}\n\nfunc TestGetRoute_FromHeader(t *testing.T) {\n\tgin.SetMode(gin.TestMode)\n\n\treq := httptest.NewRequest(\"GET\", \"/chat/completions\", nil)\n\treq.Header.Set(\"X-Yao-Route\", \"/settings/profile\")\n\tw := httptest.NewRecorder()\n\tc, _ := gin.CreateTestContext(w)\n\tc.Request = req\n\n\troute := context.GetRoute(c, nil)\n\tif route != \"/settings/profile\" {\n\t\tt.Errorf(\"Expected route '/settings/profile', got '%s'\", route)\n\t}\n}\n\nfunc TestGetRoute_FromPayload(t *testing.T) {\n\tgin.SetMode(gin.TestMode)\n\n\treq := httptest.NewRequest(\"POST\", \"/chat/completions\", nil)\n\tw := httptest.NewRecorder()\n\tc, _ := gin.CreateTestContext(w)\n\tc.Request = req\n\n\tcompletionReq := &context.CompletionRequest{\n\t\tRoute: \"/admin/users\",\n\t}\n\n\troute := context.GetRoute(c, completionReq)\n\tif route != \"/admin/users\" {\n\t\tt.Errorf(\"Expected route '/admin/users' from payload, got '%s'\", route)\n\t}\n}\n\nfunc TestGetRoute_Priority(t *testing.T) {\n\tgin.SetMode(gin.TestMode)\n\n\treq := httptest.NewRequest(\"GET\", \"/chat/completions?route=/from/query\", nil)\n\treq.Header.Set(\"X-Yao-Route\", \"/from/header\")\n\tw := httptest.NewRecorder()\n\tc, _ := gin.CreateTestContext(w)\n\tc.Request = req\n\n\tcompletionReq := &context.CompletionRequest{\n\t\tRoute: \"/from/payload\",\n\t}\n\n\troute := context.GetRoute(c, completionReq)\n\tif route != \"/from/query\" {\n\t\tt.Errorf(\"Expected query parameter to take priority, got '%s'\", route)\n\t}\n}\n\nfunc TestGetMetadata_FromQuery(t *testing.T) {\n\tgin.SetMode(gin.TestMode)\n\n\tdata := map[string]interface{}{\n\t\t\"key1\": \"value1\",\n\t\t\"key2\": float64(123),\n\t}\n\tdataJSON, _ := json.Marshal(data)\n\n\treq := httptest.NewRequest(\"GET\", \"/chat/completions?metadata=\"+string(dataJSON), nil)\n\tw := httptest.NewRecorder()\n\tc, _ := gin.CreateTestContext(w)\n\tc.Request = req\n\n\tresult := context.GetMetadata(c, nil)\n\tif result == nil {\n\t\tt.Fatal(\"Expected data to be returned\")\n\t}\n\n\tif result[\"key1\"] != \"value1\" {\n\t\tt.Errorf(\"Expected key1='value1', got '%v'\", result[\"key1\"])\n\t}\n\n\tif result[\"key2\"] != float64(123) {\n\t\tt.Errorf(\"Expected key2=123, got '%v'\", result[\"key2\"])\n\t}\n}\n\nfunc TestGetMetadata_FromHeader_Base64(t *testing.T) {\n\tgin.SetMode(gin.TestMode)\n\n\tdataBase64 := \"eyJ1c2VyX2lkIjo0NTYsImFjdGlvbiI6ImNyZWF0ZSJ9\" // base64 of {\"user_id\":456,\"action\":\"create\"}\n\n\treq := httptest.NewRequest(\"GET\", \"/chat/completions\", nil)\n\treq.Header.Set(\"X-Yao-Metadata\", dataBase64)\n\tw := httptest.NewRecorder()\n\tc, _ := gin.CreateTestContext(w)\n\tc.Request = req\n\n\tresult := context.GetMetadata(c, nil)\n\tif result == nil {\n\t\tt.Fatal(\"Expected data to be returned\")\n\t}\n\n\tif result[\"action\"] != \"create\" {\n\t\tt.Errorf(\"Expected action='create', got '%v'\", result[\"action\"])\n\t}\n\n\tif result[\"user_id\"] != float64(456) {\n\t\tt.Errorf(\"Expected user_id=456, got '%v'\", result[\"user_id\"])\n\t}\n}\n\nfunc TestGetMetadata_FromPayload(t *testing.T) {\n\tgin.SetMode(gin.TestMode)\n\n\treq := httptest.NewRequest(\"POST\", \"/chat/completions\", nil)\n\tw := httptest.NewRecorder()\n\tc, _ := gin.CreateTestContext(w)\n\tc.Request = req\n\n\tdata := map[string]interface{}{\n\t\t\"page\": float64(1),\n\t\t\"limit\": float64(10),\n\t}\n\n\tcompletionReq := &context.CompletionRequest{\n\t\tMetadata: data,\n\t}\n\n\tresult := context.GetMetadata(c, completionReq)\n\tif result == nil {\n\t\tt.Fatal(\"Expected data to be returned\")\n\t}\n\n\tif result[\"page\"] != float64(1) {\n\t\tt.Errorf(\"Expected page=1, got '%v'\", result[\"page\"])\n\t}\n\n\tif result[\"limit\"] != float64(10) {\n\t\tt.Errorf(\"Expected limit=10, got '%v'\", result[\"limit\"])\n\t}\n}\n\nfunc TestGetMetadata_Priority(t *testing.T) {\n\tgin.SetMode(gin.TestMode)\n\n\tqueryData := map[string]interface{}{\n\t\t\"source\": \"query\",\n\t}\n\tqueryDataJSON, _ := json.Marshal(queryData)\n\n\theaderDataBase64 := \"eyJzb3VyY2UiOiJoZWFkZXIifQ==\" // base64 of {\"source\":\"header\"}\n\n\treq := httptest.NewRequest(\"GET\", \"/chat/completions?metadata=\"+string(queryDataJSON), nil)\n\treq.Header.Set(\"X-Yao-Metadata\", headerDataBase64)\n\tw := httptest.NewRecorder()\n\tc, _ := gin.CreateTestContext(w)\n\tc.Request = req\n\n\tpayloadData := map[string]interface{}{\n\t\t\"source\": \"payload\",\n\t}\n\n\tcompletionReq := &context.CompletionRequest{\n\t\tMetadata: payloadData,\n\t}\n\n\tresult := context.GetMetadata(c, completionReq)\n\tif result == nil {\n\t\tt.Fatal(\"Expected data to be returned\")\n\t}\n\n\tif result[\"source\"] != \"query\" {\n\t\tt.Errorf(\"Expected query parameter to take priority, got '%v'\", result[\"source\"])\n\t}\n}\n\nfunc TestGetMetadata_EmptyData(t *testing.T) {\n\tgin.SetMode(gin.TestMode)\n\n\treq := httptest.NewRequest(\"GET\", \"/chat/completions\", nil)\n\tw := httptest.NewRecorder()\n\tc, _ := gin.CreateTestContext(w)\n\tc.Request = req\n\n\tresult := context.GetMetadata(c, nil)\n\tif result != nil {\n\t\tt.Errorf(\"Expected nil data, got '%v'\", result)\n\t}\n}\n\nfunc TestGetCompletionRequest_WriterInitialized(t *testing.T) {\n\ttest.Prepare(t, config.Conf)\n\tdefer test.Clean()\n\n\tgin.SetMode(gin.TestMode)\n\n\tcache, err := store.Get(\"__yao.agent.cache\")\n\tif err != nil {\n\t\tt.Fatalf(\"Failed to get cache: %v\", err)\n\t}\n\n\tmessages := []context.Message{\n\t\t{\n\t\t\tRole: context.RoleUser,\n\t\t\tContent: \"Test message\",\n\t\t},\n\t}\n\n\trequestBody := map[string]interface{}{\n\t\t\"model\": \"gpt-4-yao_test\",\n\t\t\"messages\": messages,\n\t}\n\n\tbodyBytes, _ := json.Marshal(requestBody)\n\n\treq := httptest.NewRequest(\"POST\", \"/chat/completions\", bytes.NewBuffer(bodyBytes))\n\treq.Header.Set(\"Content-Type\", \"application/json\")\n\tw := httptest.NewRecorder()\n\tc, _ := gin.CreateTestContext(w)\n\tc.Request = req\n\n\tcompletionReq, ctx, opts, err := context.GetCompletionRequest(c, cache)\n\tif err != nil {\n\t\tt.Fatalf(\"Failed to get completion request: %v\", err)\n\t}\n\tdefer ctx.Release()\n\n\t// Check that Writer is initialized\n\tif ctx.Writer == nil {\n\t\tt.Error(\"Expected ctx.Writer to be initialized, got nil\")\n\t}\n\n\t// Check that Writer is the same as gin context writer\n\tif ctx.Writer != c.Writer {\n\t\tt.Error(\"Expected ctx.Writer to be the same as gin context writer\")\n\t}\n\n\t// Check that Options is initialized\n\tif opts == nil {\n\t\tt.Error(\"Expected opts to be initialized, got nil\")\n\t}\n\n\t// Check other fields\n\tif completionReq.Model != \"gpt-4-yao_test\" {\n\t\tt.Errorf(\"Expected model 'gpt-4-yao_test', got '%s'\", completionReq.Model)\n\t}\n\n\tif ctx.AssistantID != \"test\" {\n\t\tt.Errorf(\"Expected assistant ID 'test', got '%s'\", ctx.AssistantID)\n\t}\n\n\t// Check that ChatID was generated (fallback)\n\tif ctx.ChatID == \"\" {\n\t\tt.Error(\"Expected ChatID to be generated, got empty string\")\n\t}\n}\n\nfunc TestGetCompletionRequest_ChatIDFallback(t *testing.T) {\n\ttest.Prepare(t, config.Conf)\n\tdefer test.Clean()\n\n\tgin.SetMode(gin.TestMode)\n\n\tcache, err := store.Get(\"__yao.agent.cache\")\n\tif err != nil {\n\t\tt.Fatalf(\"Failed to get cache: %v\", err)\n\t}\n\n\t// Request without explicit chat_id should generate one\n\tmessages := []context.Message{\n\t\t{\n\t\t\tRole: context.RoleUser,\n\t\t\tContent: \"Test message\",\n\t\t},\n\t}\n\n\trequestBody := map[string]interface{}{\n\t\t\"model\": \"gpt-4-yao_assistant1\",\n\t\t\"messages\": messages,\n\t}\n\n\tbodyBytes, _ := json.Marshal(requestBody)\n\n\treq := httptest.NewRequest(\"POST\", \"/chat/completions\", bytes.NewBuffer(bodyBytes))\n\treq.Header.Set(\"Content-Type\", \"application/json\")\n\tw := httptest.NewRecorder()\n\tc, _ := gin.CreateTestContext(w)\n\tc.Request = req\n\n\t_, ctx, opts, err := context.GetCompletionRequest(c, cache)\n\tif err != nil {\n\t\tt.Fatalf(\"Failed to get completion request: %v\", err)\n\t}\n\tdefer ctx.Release()\n\n\t// Check that Options is initialized\n\tif opts == nil {\n\t\tt.Error(\"Expected opts to be initialized, got nil\")\n\t}\n\n\t// ChatID should be generated (not empty)\n\tif ctx.ChatID == \"\" {\n\t\tt.Error(\"Expected ChatID to be generated via fallback, got empty string\")\n\t}\n\n\t// ChatID should be a valid NanoID format (16 characters)\n\tif len(ctx.ChatID) < 8 {\n\t\tt.Errorf(\"Expected ChatID to be at least 8 characters, got %d\", len(ctx.ChatID))\n\t}\n}\n\nfunc TestGetSkip_FromBody(t *testing.T) {\n\tgin.SetMode(gin.TestMode)\n\n\treq := httptest.NewRequest(\"POST\", \"/chat/completions\", nil)\n\tw := httptest.NewRecorder()\n\tc, _ := gin.CreateTestContext(w)\n\tc.Request = req\n\n\tcompletionReq := &context.CompletionRequest{\n\t\tSkip: &context.Skip{\n\t\t\tHistory: true,\n\t\t\tTrace: false,\n\t\t},\n\t}\n\n\tskip := context.GetSkip(c, completionReq)\n\tif skip == nil {\n\t\tt.Fatal(\"Expected skip to be returned\")\n\t}\n\n\tif !skip.History {\n\t\tt.Error(\"Expected skip.History to be true\")\n\t}\n\n\tif skip.Trace {\n\t\tt.Error(\"Expected skip.Trace to be false\")\n\t}\n}\n\nfunc TestGetSkip_FromQueryParams(t *testing.T) {\n\tgin.SetMode(gin.TestMode)\n\n\treq := httptest.NewRequest(\"GET\", \"/chat/completions?skip_history=true&skip_trace=false\", nil)\n\tw := httptest.NewRecorder()\n\tc, _ := gin.CreateTestContext(w)\n\tc.Request = req\n\n\tskip := context.GetSkip(c, nil)\n\tif skip == nil {\n\t\tt.Fatal(\"Expected skip to be returned\")\n\t}\n\n\tif !skip.History {\n\t\tt.Error(\"Expected skip.History to be true from query param\")\n\t}\n\n\tif skip.Trace {\n\t\tt.Error(\"Expected skip.Trace to be false\")\n\t}\n}\n\nfunc TestGetSkip_FromQueryParams_ShortForm(t *testing.T) {\n\tgin.SetMode(gin.TestMode)\n\n\treq := httptest.NewRequest(\"GET\", \"/chat/completions?skip_history=1&skip_trace=1\", nil)\n\tw := httptest.NewRecorder()\n\tc, _ := gin.CreateTestContext(w)\n\tc.Request = req\n\n\tskip := context.GetSkip(c, nil)\n\tif skip == nil {\n\t\tt.Fatal(\"Expected skip to be returned\")\n\t}\n\n\tif !skip.History {\n\t\tt.Error(\"Expected skip.History to be true from query param (1)\")\n\t}\n\n\tif !skip.Trace {\n\t\tt.Error(\"Expected skip.Trace to be true from query param (1)\")\n\t}\n}\n\nfunc TestGetSkip_Priority(t *testing.T) {\n\tgin.SetMode(gin.TestMode)\n\n\t// Body should take priority over query\n\treq := httptest.NewRequest(\"POST\", \"/chat/completions?skip_history=false&skip_trace=false\", nil)\n\tw := httptest.NewRecorder()\n\tc, _ := gin.CreateTestContext(w)\n\tc.Request = req\n\n\tcompletionReq := &context.CompletionRequest{\n\t\tSkip: &context.Skip{\n\t\t\tHistory: true,\n\t\t\tTrace: true,\n\t\t},\n\t}\n\n\tskip := context.GetSkip(c, completionReq)\n\tif skip == nil {\n\t\tt.Fatal(\"Expected skip to be returned\")\n\t}\n\n\t// Body should take priority\n\tif !skip.History {\n\t\tt.Error(\"Expected body parameter to take priority, skip.History should be true\")\n\t}\n\n\tif !skip.Trace {\n\t\tt.Error(\"Expected body parameter to take priority, skip.Trace should be true\")\n\t}\n}\n\nfunc TestGetSkip_Nil(t *testing.T) {\n\tgin.SetMode(gin.TestMode)\n\n\treq := httptest.NewRequest(\"GET\", \"/chat/completions\", nil)\n\tw := httptest.NewRecorder()\n\tc, _ := gin.CreateTestContext(w)\n\tc.Request = req\n\n\tskip := context.GetSkip(c, nil)\n\tif skip != nil {\n\t\tt.Errorf(\"Expected skip to be nil, got %v\", skip)\n\t}\n}\n\nfunc TestGetSkip_OnlyHistorySet(t *testing.T) {\n\tgin.SetMode(gin.TestMode)\n\n\treq := httptest.NewRequest(\"GET\", \"/chat/completions?skip_history=true\", nil)\n\tw := httptest.NewRecorder()\n\tc, _ := gin.CreateTestContext(w)\n\tc.Request = req\n\n\tskip := context.GetSkip(c, nil)\n\tif skip == nil {\n\t\tt.Fatal(\"Expected skip to be returned\")\n\t}\n\n\tif !skip.History {\n\t\tt.Error(\"Expected skip.History to be true\")\n\t}\n\n\tif skip.Trace {\n\t\tt.Error(\"Expected skip.Trace to be false (default)\")\n\t}\n}\n\nfunc TestGetSkip_FromBodyViaParseRequest(t *testing.T) {\n\ttest.Prepare(t, config.Conf)\n\tdefer test.Clean()\n\n\tgin.SetMode(gin.TestMode)\n\n\t// Test parsing Skip from full request body\n\tmessages := []context.Message{\n\t\t{\n\t\t\tRole: context.RoleUser,\n\t\t\tContent: \"Generate a title for this chat\",\n\t\t},\n\t}\n\n\trequestBody := map[string]interface{}{\n\t\t\"model\": \"workers.system.title-yao_test\",\n\t\t\"messages\": messages,\n\t\t\"skip\": map[string]interface{}{\n\t\t\t\"history\": true,\n\t\t\t\"trace\": false,\n\t\t},\n\t}\n\n\tbodyBytes, _ := json.Marshal(requestBody)\n\n\treq := httptest.NewRequest(\"POST\", \"/chat/completions\", bytes.NewBuffer(bodyBytes))\n\treq.Header.Set(\"Content-Type\", \"application/json\")\n\tw := httptest.NewRecorder()\n\tc, _ := gin.CreateTestContext(w)\n\tc.Request = req\n\n\t// Parse the request\n\tcompletionReq, err := parseCompletionRequestData(c)\n\tif err != nil {\n\t\tt.Fatalf(\"Failed to parse completion request: %v\", err)\n\t}\n\n\t// Verify Skip was parsed correctly\n\tif completionReq.Skip == nil {\n\t\tt.Fatal(\"Expected Skip to be parsed from body, got nil\")\n\t}\n\n\tif !completionReq.Skip.History {\n\t\tt.Error(\"Expected Skip.History to be true from body\")\n\t}\n\n\tif completionReq.Skip.Trace {\n\t\tt.Error(\"Expected Skip.Trace to be false from body\")\n\t}\n\n\t// Now test GetSkip function with the parsed request\n\tskip := context.GetSkip(c, completionReq)\n\tif skip == nil {\n\t\tt.Fatal(\"Expected GetSkip to return skip configuration\")\n\t}\n\n\tif !skip.History {\n\t\tt.Error(\"Expected GetSkip to return History=true\")\n\t}\n\n\tif skip.Trace {\n\t\tt.Error(\"Expected GetSkip to return Trace=false\")\n\t}\n}\n\nfunc TestGetMode_FromQuery(t *testing.T) {\n\ttest.Prepare(t, config.Conf)\n\tdefer test.Clean()\n\n\tgin.SetMode(gin.TestMode)\n\n\treq := httptest.NewRequest(\"GET\", \"/chat/completions?mode=task\", nil)\n\tw := httptest.NewRecorder()\n\tc, _ := gin.CreateTestContext(w)\n\tc.Request = req\n\n\tmode := context.GetMode(c, nil)\n\tif mode != \"task\" {\n\t\tt.Errorf(\"Expected mode 'task' from query, got '%s'\", mode)\n\t}\n}\n\nfunc TestGetMode_FromHeader(t *testing.T) {\n\ttest.Prepare(t, config.Conf)\n\tdefer test.Clean()\n\n\tgin.SetMode(gin.TestMode)\n\n\treq := httptest.NewRequest(\"GET\", \"/chat/completions\", nil)\n\treq.Header.Set(\"X-Yao-Mode\", \"chat\")\n\tw := httptest.NewRecorder()\n\tc, _ := gin.CreateTestContext(w)\n\tc.Request = req\n\n\tmode := context.GetMode(c, nil)\n\tif mode != \"chat\" {\n\t\tt.Errorf(\"Expected mode 'chat' from header, got '%s'\", mode)\n\t}\n}\n\nfunc TestGetMode_FromMetadata(t *testing.T) {\n\ttest.Prepare(t, config.Conf)\n\tdefer test.Clean()\n\n\tgin.SetMode(gin.TestMode)\n\n\treq := httptest.NewRequest(\"GET\", \"/chat/completions\", nil)\n\tw := httptest.NewRecorder()\n\tc, _ := gin.CreateTestContext(w)\n\tc.Request = req\n\n\tcompletionReq := &context.CompletionRequest{\n\t\tMetadata: map[string]interface{}{\n\t\t\t\"mode\": \"task\",\n\t\t},\n\t}\n\n\tmode := context.GetMode(c, completionReq)\n\tif mode != \"task\" {\n\t\tt.Errorf(\"Expected mode 'task' from metadata, got '%s'\", mode)\n\t}\n}\n\nfunc TestGetMode_Priority(t *testing.T) {\n\ttest.Prepare(t, config.Conf)\n\tdefer test.Clean()\n\n\tgin.SetMode(gin.TestMode)\n\n\t// Query has highest priority\n\treq := httptest.NewRequest(\"GET\", \"/chat/completions?mode=query_mode\", nil)\n\treq.Header.Set(\"X-Yao-Mode\", \"header_mode\")\n\tw := httptest.NewRecorder()\n\tc, _ := gin.CreateTestContext(w)\n\tc.Request = req\n\n\tcompletionReq := &context.CompletionRequest{\n\t\tMetadata: map[string]interface{}{\n\t\t\t\"mode\": \"metadata_mode\",\n\t\t},\n\t}\n\n\tmode := context.GetMode(c, completionReq)\n\tif mode != \"query_mode\" {\n\t\tt.Errorf(\"Expected mode 'query_mode' (query has priority), got '%s'\", mode)\n\t}\n}\n\nfunc TestGetMode_Empty(t *testing.T) {\n\ttest.Prepare(t, config.Conf)\n\tdefer test.Clean()\n\n\tgin.SetMode(gin.TestMode)\n\n\treq := httptest.NewRequest(\"GET\", \"/chat/completions\", nil)\n\tw := httptest.NewRecorder()\n\tc, _ := gin.CreateTestContext(w)\n\tc.Request = req\n\n\tmode := context.GetMode(c, nil)\n\tif mode != \"\" {\n\t\tt.Errorf(\"Expected empty mode, got '%s'\", mode)\n\t}\n}\n" }, { "path": "agent/context/options.go", "content": "package context\n\n// ToMap converts Options struct to map for JSON serialization\nfunc (opts *Options) ToMap() map[string]interface{} {\n\tif opts == nil {\n\t\treturn nil\n\t}\n\n\tresult := make(map[string]interface{})\n\n\t// Add configurable fields (with json tags)\n\tif opts.Connector != \"\" {\n\t\tresult[\"connector\"] = opts.Connector\n\t}\n\tif opts.Mode != \"\" {\n\t\tresult[\"mode\"] = opts.Mode\n\t}\n\tif opts.Search != nil {\n\t\tresult[\"search\"] = opts.Search\n\t}\n\tif opts.Skip != nil {\n\t\tresult[\"skip\"] = opts.Skip\n\t}\n\t// Only add DisableGlobalPrompts if true (avoid false values in map)\n\tif opts.DisableGlobalPrompts {\n\t\tresult[\"disable_global_prompts\"] = opts.DisableGlobalPrompts\n\t}\n\tif opts.Metadata != nil {\n\t\tresult[\"metadata\"] = opts.Metadata\n\t}\n\n\t// Note: Runtime fields (Context, Writer) are not serialized (json:\"-\")\n\t// They should not be included in the map\n\n\treturn result\n}\n\n// OptionsFromMap creates Options struct from map (e.g., from JS Hook)\nfunc OptionsFromMap(m map[string]interface{}) *Options {\n\tif m == nil {\n\t\treturn &Options{}\n\t}\n\n\topts := &Options{}\n\n\t// Extract configurable fields\n\tif connector, ok := m[\"connector\"].(string); ok {\n\t\topts.Connector = connector\n\t}\n\tif mode, ok := m[\"mode\"].(string); ok {\n\t\topts.Mode = mode\n\t}\n\t// Search supports: bool | SearchIntent | map[string]any | nil\n\tif search := m[\"search\"]; search != nil {\n\t\topts.Search = search\n\t}\n\tif skipMap, ok := m[\"skip\"].(map[string]interface{}); ok {\n\t\tskip := &Skip{}\n\t\tif history, ok := skipMap[\"history\"].(bool); ok {\n\t\t\tskip.History = history\n\t\t}\n\t\tif trace, ok := skipMap[\"trace\"].(bool); ok {\n\t\t\tskip.Trace = trace\n\t\t}\n\t\tif output, ok := skipMap[\"output\"].(bool); ok {\n\t\t\tskip.Output = output\n\t\t}\n\t\topts.Skip = skip\n\t}\n\tif disableGlobalPrompts, ok := m[\"disable_global_prompts\"].(bool); ok {\n\t\topts.DisableGlobalPrompts = disableGlobalPrompts\n\t}\n\tif metadata, ok := m[\"metadata\"].(map[string]interface{}); ok {\n\t\topts.Metadata = metadata\n\t}\n\n\t// Note: Context and Writer are runtime fields, not restored from map\n\t// They should be set by the caller if needed\n\n\treturn opts\n}\n" }, { "path": "agent/context/output.go", "content": "package context\n\nimport (\n\t\"time\"\n\n\t\"github.com/yaoapp/gou/llm\"\n\t\"github.com/yaoapp/yao/agent/output\"\n\t\"github.com/yaoapp/yao/agent/output/message\"\n)\n\n// Send sends a message via the output module\n// Automatically manages BlockID, ThreadID, lifecycle events, and metadata for delta operations\n// - For delta operations: inherits BlockID and ThreadID from original message, increments chunk count\n// - For new messages: auto-sets ThreadID from Stack, sends message_start event\n// - Sends block_start event when a new BlockID is first encountered\n// - Records metadata for all sent messages to enable delta inheritance\nfunc (ctx *Context) Send(msg *message.Message) error {\n\t// Call OnMessage callback if provided (for ctx.agent.Call with onChunk)\n\tif ctx.Stack != nil && ctx.Stack.Options != nil && ctx.Stack.Options.OnMessage != nil {\n\t\tif ret := ctx.Stack.Options.OnMessage(msg); ret != 0 {\n\t\t\treturn nil // Callback requested stop\n\t\t}\n\t}\n\n\tout, err := ctx.getOutput()\n\tif err != nil {\n\t\treturn err\n\t}\n\n\t// Skip lifecycle events for event-type messages (prevent recursion)\n\tisEventMessage := msg.Type == message.TypeEvent\n\n\t// === Handle message_start event: record metadata for future delta chunks ===\n\tif isEventMessage && msg.Props != nil {\n\t\tif event, ok := msg.Props[\"event\"].(string); ok && event == message.EventMessageStart {\n\t\t\tif data, ok := msg.Props[\"data\"].(message.EventMessageStartData); ok {\n\t\t\t\t// Record metadata from message_start event\n\t\t\t\tif data.MessageID != \"\" && ctx.messageMetadata != nil {\n\t\t\t\t\tctx.messageMetadata.setMessage(data.MessageID, &MessageMetadata{\n\t\t\t\t\t\tMessageID: data.MessageID,\n\t\t\t\t\t\tThreadID: data.ThreadID,\n\t\t\t\t\t\tType: data.Type,\n\t\t\t\t\t\tStartTime: time.Now(),\n\t\t\t\t\t\tChunkCount: 0, // Will be incremented by delta chunks\n\t\t\t\t\t})\n\t\t\t\t}\n\t\t\t}\n\t\t}\n\t}\n\n\t// === Delta operations: Auto-inherit and update metadata ===\n\tif msg.Delta && msg.MessageID != \"\" && ctx.messageMetadata != nil {\n\t\tif metadata := ctx.getMessageMetadata(msg.MessageID); metadata != nil {\n\t\t\t// Inherit BlockID if not specified\n\t\t\tif msg.BlockID == \"\" {\n\t\t\t\tmsg.BlockID = metadata.BlockID\n\t\t\t}\n\t\t\t// Inherit ThreadID if not specified\n\t\t\tif msg.ThreadID == \"\" {\n\t\t\t\tmsg.ThreadID = metadata.ThreadID\n\t\t\t}\n\n\t\t\t// Increment chunk count for this message\n\t\t\tmetadata.ChunkCount++\n\n\t\t\t// Update Buffer content for streaming messages (for storage)\n\t\t\tif ctx.Buffer != nil && msg.Props != nil {\n\t\t\t\tif content, ok := msg.Props[\"content\"].(string); ok {\n\t\t\t\t\tctx.Buffer.AppendMessageContent(msg.MessageID, content)\n\t\t\t\t}\n\t\t\t}\n\t\t}\n\t}\n\n\t// === Auto-generate ChunkID (always) ===\n\tif msg.ChunkID == \"\" && !isEventMessage {\n\t\tif ctx.IDGenerator != nil {\n\t\t\tmsg.ChunkID = ctx.IDGenerator.GenerateChunkID()\n\t\t} else {\n\t\t\tmsg.ChunkID = message.GenerateNanoID()\n\t\t}\n\t}\n\n\t// === Non-delta operations: New message logic ===\n\tif !msg.Delta && !isEventMessage {\n\t\t// Auto-set ThreadID for non-root Stack (nested agent calls)\n\t\tif msg.ThreadID == \"\" && ctx.Stack != nil && !ctx.Stack.IsRoot() {\n\t\t\tmsg.ThreadID = ctx.Stack.ID\n\t\t}\n\n\t\t// BlockID is NOT auto-generated by default (only manually specified in special cases)\n\t\t// Example: Send a web card after LLM output, group them in the same Block\n\t\t// Developers can specify via ctx.Send(message, blockId) or message.block_id\n\n\t\t// === Send block_start event if this is a new block ===\n\t\tif msg.BlockID != \"\" && ctx.messageMetadata != nil {\n\t\t\tif ctx.messageMetadata.getBlock(msg.BlockID) == nil {\n\t\t\t\t// New block, send block_start event\n\t\t\t\tblockStartData := message.EventBlockStartData{\n\t\t\t\t\tBlockID: msg.BlockID,\n\t\t\t\t\tType: \"mixed\", // Default type, can be enhanced later\n\t\t\t\t\tTimestamp: time.Now().UnixMilli(),\n\t\t\t\t}\n\t\t\t\tblockStartEvent := output.NewEventMessage(message.EventBlockStart, \"Block started\", blockStartData)\n\t\t\t\tif err := ctx.sendRaw(blockStartEvent); err != nil {\n\t\t\t\t\treturn err\n\t\t\t\t}\n\n\t\t\t\t// Record block metadata\n\t\t\t\tctx.messageMetadata.setBlock(msg.BlockID, &BlockMetadata{\n\t\t\t\t\tBlockID: msg.BlockID,\n\t\t\t\t\tType: \"mixed\",\n\t\t\t\t\tStartTime: time.Now(),\n\t\t\t\t\tMessageCount: 0,\n\t\t\t\t})\n\t\t\t}\n\n\t\t\t// Increment message count for this block\n\t\t\tctx.messageMetadata.updateBlock(msg.BlockID, func(block *BlockMetadata) {\n\t\t\t\tblock.MessageCount++\n\t\t\t})\n\t\t}\n\n\t\t// === Generate MessageID if not provided ===\n\t\tif msg.MessageID == \"\" {\n\t\t\tif ctx.IDGenerator != nil {\n\t\t\t\tmsg.MessageID = ctx.IDGenerator.GenerateMessageID()\n\t\t\t} else {\n\t\t\t\tmsg.MessageID = message.GenerateNanoID() // Use NanoID generator\n\t\t\t}\n\t\t}\n\n\t\t// === Send message_start event ===\n\t\tmessageStartData := message.EventMessageStartData{\n\t\t\tMessageID: msg.MessageID,\n\t\t\tType: msg.Type,\n\t\t\tTimestamp: time.Now().UnixMilli(),\n\t\t\tThreadID: msg.ThreadID, // Include ThreadID for concurrent stream identification\n\t\t}\n\t\tmessageStartEvent := output.NewEventMessage(message.EventMessageStart, \"Message started\", messageStartData)\n\t\tif err := ctx.sendRaw(messageStartEvent); err != nil {\n\t\t\treturn err\n\t\t}\n\n\t\t// === Record message metadata with start time ===\n\t\tif ctx.messageMetadata != nil {\n\t\t\tctx.messageMetadata.setMessage(msg.MessageID, &MessageMetadata{\n\t\t\t\tMessageID: msg.MessageID,\n\t\t\t\tBlockID: msg.BlockID,\n\t\t\t\tThreadID: msg.ThreadID,\n\t\t\t\tType: msg.Type,\n\t\t\t\tStartTime: time.Now(),\n\t\t\t\tChunkCount: 1, // Initial chunk\n\t\t\t})\n\t\t}\n\t}\n\n\t// === Actually send the message ===\n\tif err := out.Send(msg); err != nil {\n\t\treturn err\n\t}\n\n\t// === Buffer message for batch saving (non-delta, non-event messages only) ===\n\t// Delta messages are streaming chunks; only final content should be saved\n\t// Event messages are transient lifecycle signals, not stored\n\t// Skip if History is disabled in options\n\tif !msg.Delta && !isEventMessage && ctx.Buffer != nil && !ctx.shouldSkipHistory() {\n\t\tassistantID := \"\"\n\t\tif ctx.Stack != nil {\n\t\t\tassistantID = ctx.Stack.AssistantID\n\t\t}\n\t\tctx.Buffer.AddAssistantMessage(\n\t\t\tmsg.MessageID, // Use the same MessageID as sent to client\n\t\t\tmsg.Type,\n\t\t\tmsg.Props,\n\t\t\tmsg.BlockID,\n\t\t\tmsg.ThreadID,\n\t\t\tassistantID,\n\t\t\tnil, // metadata can be added if needed\n\t\t)\n\t}\n\n\t// === Auto-send message_end for non-delta messages (complete messages) ===\n\tif !msg.Delta && !isEventMessage && msg.MessageID != \"\" && ctx.messageMetadata != nil {\n\t\tmetadata := ctx.messageMetadata.getMessage(msg.MessageID)\n\t\tif metadata != nil {\n\t\t\t// Calculate duration\n\t\t\tdurationMs := time.Since(metadata.StartTime).Milliseconds()\n\n\t\t\t// Extract content for the extra field\n\t\t\tvar content interface{}\n\t\t\tif msg.Props != nil {\n\t\t\t\tif c, ok := msg.Props[\"content\"]; ok {\n\t\t\t\t\tcontent = c\n\t\t\t\t}\n\t\t\t}\n\n\t\t\t// Build message_end event data\n\t\t\tendData := message.EventMessageEndData{\n\t\t\t\tMessageID: msg.MessageID,\n\t\t\t\tType: msg.Type,\n\t\t\t\tTimestamp: time.Now().UnixMilli(),\n\t\t\t\tThreadID: metadata.ThreadID, // Include ThreadID for concurrent stream identification\n\t\t\t\tDurationMs: durationMs,\n\t\t\t\tChunkCount: metadata.ChunkCount,\n\t\t\t\tStatus: \"completed\",\n\t\t\t}\n\n\t\t\t// Add content to extra if available\n\t\t\tif content != nil {\n\t\t\t\tendData.Extra = map[string]interface{}{\n\t\t\t\t\t\"content\": content,\n\t\t\t\t}\n\t\t\t}\n\n\t\t\t// Send message_end event\n\t\t\tmessageEndEvent := output.NewEventMessage(message.EventMessageEnd, \"Message completed\", endData)\n\t\t\tctx.sendRaw(messageEndEvent)\n\t\t}\n\t}\n\n\treturn nil\n}\n\n// SendStream sends a streaming message that can be appended to later\n// Unlike Send(), this does NOT automatically send message_end event\n// Use ctx.Append() to add content, then ctx.End() to finalize\n// Returns the message ID for use with Append/End\nfunc (ctx *Context) SendStream(msg *message.Message) (string, error) {\n\tout, err := ctx.getOutput()\n\tif err != nil {\n\t\treturn \"\", err\n\t}\n\n\t// Skip lifecycle events for event-type messages\n\tisEventMessage := msg.Type == message.TypeEvent\n\tif isEventMessage {\n\t\t// Event messages should use Send(), not SendStream()\n\t\treturn \"\", ctx.Send(msg)\n\t}\n\n\t// === Auto-generate ChunkID ===\n\tif msg.ChunkID == \"\" {\n\t\tif ctx.IDGenerator != nil {\n\t\t\tmsg.ChunkID = ctx.IDGenerator.GenerateChunkID()\n\t\t} else {\n\t\t\tmsg.ChunkID = message.GenerateNanoID()\n\t\t}\n\t}\n\n\t// === Auto-set ThreadID for non-root Stack ===\n\tif msg.ThreadID == \"\" && ctx.Stack != nil && !ctx.Stack.IsRoot() {\n\t\tmsg.ThreadID = ctx.Stack.ID\n\t}\n\n\t// === Handle BlockID and block_start event ===\n\tif msg.BlockID != \"\" && ctx.messageMetadata != nil {\n\t\tif ctx.messageMetadata.getBlock(msg.BlockID) == nil {\n\t\t\tblockStartData := message.EventBlockStartData{\n\t\t\t\tBlockID: msg.BlockID,\n\t\t\t\tType: \"mixed\",\n\t\t\t\tTimestamp: time.Now().UnixMilli(),\n\t\t\t}\n\t\t\tblockStartEvent := output.NewEventMessage(message.EventBlockStart, \"Block started\", blockStartData)\n\t\t\tif err := ctx.sendRaw(blockStartEvent); err != nil {\n\t\t\t\treturn \"\", err\n\t\t\t}\n\t\t\tctx.messageMetadata.setBlock(msg.BlockID, &BlockMetadata{\n\t\t\t\tBlockID: msg.BlockID,\n\t\t\t\tType: \"mixed\",\n\t\t\t\tStartTime: time.Now(),\n\t\t\t\tMessageCount: 0,\n\t\t\t})\n\t\t}\n\t\tctx.messageMetadata.updateBlock(msg.BlockID, func(block *BlockMetadata) {\n\t\t\tblock.MessageCount++\n\t\t})\n\t}\n\n\t// === Generate MessageID if not provided ===\n\tif msg.MessageID == \"\" {\n\t\tif ctx.IDGenerator != nil {\n\t\t\tmsg.MessageID = ctx.IDGenerator.GenerateMessageID()\n\t\t} else {\n\t\t\tmsg.MessageID = message.GenerateNanoID()\n\t\t}\n\t}\n\n\t// === Send message_start event ===\n\tmessageStartData := message.EventMessageStartData{\n\t\tMessageID: msg.MessageID,\n\t\tType: msg.Type,\n\t\tTimestamp: time.Now().UnixMilli(),\n\t\tThreadID: msg.ThreadID,\n\t}\n\tmessageStartEvent := output.NewEventMessage(message.EventMessageStart, \"Message started\", messageStartData)\n\tif err := ctx.sendRaw(messageStartEvent); err != nil {\n\t\treturn \"\", err\n\t}\n\n\t// === Record message metadata ===\n\tif ctx.messageMetadata != nil {\n\t\tctx.messageMetadata.setMessage(msg.MessageID, &MessageMetadata{\n\t\t\tMessageID: msg.MessageID,\n\t\t\tBlockID: msg.BlockID,\n\t\t\tThreadID: msg.ThreadID,\n\t\t\tType: msg.Type,\n\t\t\tStartTime: time.Now(),\n\t\t\tChunkCount: 1,\n\t\t})\n\t}\n\n\t// === Actually send the message ===\n\tif err := out.Send(msg); err != nil {\n\t\treturn \"\", err\n\t}\n\n\t// === Buffer streaming message (will be completed by End()) ===\n\tif ctx.Buffer != nil && !ctx.shouldSkipHistory() {\n\t\tassistantID := \"\"\n\t\tif ctx.Stack != nil {\n\t\t\tassistantID = ctx.Stack.AssistantID\n\t\t}\n\t\tctx.Buffer.AddStreamingMessage(\n\t\t\tmsg.MessageID,\n\t\t\tmsg.Type,\n\t\t\tmsg.Props,\n\t\t\tmsg.BlockID,\n\t\t\tmsg.ThreadID,\n\t\t\tassistantID,\n\t\t\tnil,\n\t\t)\n\t}\n\n\t// NOTE: No message_end event here - will be sent by End()\n\treturn msg.MessageID, nil\n}\n\n// End finalizes a streaming message started with SendStream\n// Optionally appends final content before sending message_end event\n// This also saves the complete message to the buffer for storage\nfunc (ctx *Context) End(messageID string, finalContent ...string) error {\n\tif messageID == \"\" {\n\t\treturn nil\n\t}\n\n\t// Append final content if provided\n\tif len(finalContent) > 0 && finalContent[0] != \"\" {\n\t\t// Create a delta message for the final content\n\t\tdeltaMsg := &message.Message{\n\t\t\tMessageID: messageID,\n\t\t\tType: message.TypeText,\n\t\t\tDelta: true,\n\t\t\tDeltaAction: message.DeltaAppend,\n\t\t\tProps: map[string]interface{}{\n\t\t\t\t\"content\": finalContent[0],\n\t\t\t},\n\t\t}\n\t\tif err := ctx.Send(deltaMsg); err != nil {\n\t\t\treturn err\n\t\t}\n\t}\n\n\t// Get complete content from buffer\n\tvar completeContent string\n\tif ctx.Buffer != nil {\n\t\tcompleteContent, _ = ctx.Buffer.CompleteStreamingMessage(messageID)\n\t}\n\n\t// Get metadata for duration calculation\n\tvar durationMs int64\n\tvar threadID string\n\tvar chunkCount int\n\tvar msgType string = message.TypeText\n\n\tif ctx.messageMetadata != nil {\n\t\tif metadata := ctx.messageMetadata.getMessage(messageID); metadata != nil {\n\t\t\tdurationMs = time.Since(metadata.StartTime).Milliseconds()\n\t\t\tthreadID = metadata.ThreadID\n\t\t\tchunkCount = metadata.ChunkCount\n\t\t\tmsgType = metadata.Type\n\t\t}\n\t}\n\n\t// Build message_end event data\n\tendData := message.EventMessageEndData{\n\t\tMessageID: messageID,\n\t\tType: msgType,\n\t\tTimestamp: time.Now().UnixMilli(),\n\t\tThreadID: threadID,\n\t\tDurationMs: durationMs,\n\t\tChunkCount: chunkCount,\n\t\tStatus: \"completed\",\n\t}\n\n\t// Add complete content to extra\n\tif completeContent != \"\" {\n\t\tendData.Extra = map[string]interface{}{\n\t\t\t\"content\": completeContent,\n\t\t}\n\t}\n\n\t// Send message_end event\n\tmessageEndEvent := output.NewEventMessage(message.EventMessageEnd, \"Message completed\", endData)\n\treturn ctx.sendRaw(messageEndEvent)\n}\n\n// EndMessage sends a message_end event for a completed message\n// Note: For non-delta messages, message_end is automatically sent by Send()\n// This method is primarily for delta streaming scenarios:\n// - After all delta chunks are sent for a message, call EndMessage() to finalize it\n// - For LLM streaming, this is typically called after receiving ChunkMessageEnd\nfunc (ctx *Context) EndMessage(messageID string, content interface{}) error {\n\tif messageID == \"\" || ctx.messageMetadata == nil {\n\t\treturn nil\n\t}\n\n\tmetadata := ctx.messageMetadata.getMessage(messageID)\n\tif metadata == nil {\n\t\treturn nil // Message not found, skip\n\t}\n\n\t// Calculate duration\n\tdurationMs := time.Since(metadata.StartTime).Milliseconds()\n\n\t// Build message_end event data\n\tendData := message.EventMessageEndData{\n\t\tMessageID: messageID,\n\t\tType: metadata.Type,\n\t\tTimestamp: time.Now().UnixMilli(),\n\t\tThreadID: metadata.ThreadID, // Include ThreadID for concurrent stream identification\n\t\tDurationMs: durationMs,\n\t\tChunkCount: metadata.ChunkCount,\n\t\tStatus: \"completed\",\n\t}\n\n\t// Add content to extra if provided\n\tif content != nil {\n\t\tendData.Extra = map[string]interface{}{\n\t\t\t\"content\": content,\n\t\t}\n\t}\n\n\t// Send message_end event\n\tmessageEndEvent := output.NewEventMessage(message.EventMessageEnd, \"Message completed\", endData)\n\treturn ctx.sendRaw(messageEndEvent)\n}\n\n// EndBlock sends a block_end event for a completed block\n// This should be called explicitly when all messages in a block are complete\nfunc (ctx *Context) EndBlock(blockID string) error {\n\tif blockID == \"\" || ctx.messageMetadata == nil {\n\t\treturn nil\n\t}\n\n\tblockMetadata := ctx.messageMetadata.getBlock(blockID)\n\tif blockMetadata == nil {\n\t\treturn nil // Block not found, skip\n\t}\n\n\t// Calculate duration\n\tdurationMs := time.Since(blockMetadata.StartTime).Milliseconds()\n\n\t// Build block_end event data\n\tendData := message.EventBlockEndData{\n\t\tBlockID: blockID,\n\t\tType: blockMetadata.Type,\n\t\tTimestamp: time.Now().UnixMilli(),\n\t\tDurationMs: durationMs,\n\t\tMessageCount: blockMetadata.MessageCount,\n\t\tStatus: \"completed\",\n\t}\n\n\t// Send block_end event\n\tblockEndEvent := output.NewEventMessage(message.EventBlockEnd, \"Block completed\", endData)\n\treturn ctx.sendRaw(blockEndEvent)\n}\n\n// SendGroup sends a group of messages via the output module\n// Deprecated: This method is deprecated and will be removed in future versions\nfunc (ctx *Context) SendGroup(group *message.Group) error {\n\toutput, err := ctx.getOutput()\n\tif err != nil {\n\t\treturn err\n\t}\n\treturn output.SendGroup(group)\n}\n\n// Flush flushes the output writer\nfunc (ctx *Context) Flush() error {\n\toutput, err := ctx.getOutput()\n\tif err != nil {\n\t\treturn err\n\t}\n\treturn output.Flush()\n}\n\n// CloseOutput closes the output writer\nfunc (ctx *Context) CloseOutput() error {\n\toutput, err := ctx.getOutput()\n\tif err != nil {\n\t\treturn err\n\t}\n\treturn output.Close()\n}\n\n// sendRaw sends a message directly without triggering lifecycle events\n// Used internally to send event messages without recursion\nfunc (ctx *Context) sendRaw(msg *message.Message) error {\n\tout, err := ctx.getOutput()\n\tif err != nil {\n\t\treturn err\n\t}\n\treturn out.Send(msg)\n}\n\n// getWriter gets the effective Writer for the current context\n// Priority: Skip.Output > Stack.Options.Writer > ctx.Writer\n// Note: The Writer returned is always a SafeWriter (wrapped at context creation)\n// to ensure thread-safe concurrent writes for SSE streaming.\nfunc (ctx *Context) getWriter() Writer {\n\t// Check if output is explicitly skipped (for internal A2A calls)\n\tif ctx.Stack != nil && ctx.Stack.Options != nil && ctx.Stack.Options.Skip != nil && ctx.Stack.Options.Skip.Output {\n\t\treturn nil // Explicitly disable output\n\t}\n\n\t// Check if current Stack has a Writer override\n\tif ctx.Stack != nil && ctx.Stack.Options != nil && ctx.Stack.Options.Writer != nil {\n\t\treturn ctx.Stack.Options.Writer\n\t}\n\n\treturn ctx.Writer\n}\n\n// getOutput gets the output writer for the context\nfunc (ctx *Context) getOutput() (*output.Output, error) {\n\t// Check if current Stack has cached output\n\tif ctx.Stack != nil && ctx.Stack.output != nil {\n\t\treturn ctx.Stack.output, nil\n\t}\n\n\t// Ensure Writer is wrapped in SafeWriter for concurrent-safe SSE writes\n\t// This is essential for ctx.agent.All where multiple sub-agents\n\t// write to the same SSE stream concurrently.\n\t// We wrap once at the context level so all forked contexts share the same SafeWriter.\n\twriter := ctx.getWriter()\n\tif writer != nil {\n\t\t// Check if it's already a SafeWriter\n\t\tif _, ok := writer.(*output.SafeWriter); !ok {\n\t\t\t// Wrap in SafeWriter with context for automatic cleanup on client disconnect\n\t\t\t// This prevents goroutine leaks in enterprise applications\n\t\t\tvar safeWriter *output.SafeWriter\n\t\t\tif ctx.Context != nil {\n\t\t\t\t// Use request context to detect client disconnection\n\t\t\t\tsafeWriter = output.NewSafeWriterWithContext(ctx.Context, writer)\n\t\t\t} else {\n\t\t\t\t// Fallback to basic SafeWriter if no context available\n\t\t\t\tsafeWriter = output.NewSafeWriter(writer)\n\t\t\t}\n\t\t\tctx.Writer = safeWriter\n\t\t\twriter = safeWriter\n\t\t}\n\t}\n\n\ttrace, _ := ctx.Trace()\n\tvar options message.Options = message.Options{\n\t\tBaseURL: \"/\",\n\t\tWriter: writer,\n\t\tTrace: trace,\n\t\tLocale: ctx.Locale,\n\t\tAccept: string(ctx.Accept),\n\t}\n\n\tif ctx.Capabilities != nil {\n\t\tcaps := llm.Capabilities(*ctx.Capabilities)\n\t\toptions.Capabilities = &caps\n\t}\n\n\tout, err := output.NewOutput(options)\n\tif err != nil {\n\t\treturn nil, err\n\t}\n\n\t// Cache to current Stack (each Stack has its own output with its own Writer)\n\tif ctx.Stack != nil {\n\t\tctx.Stack.output = out\n\t}\n\n\treturn out, nil\n}\n\n// CloseSafeWriter closes the SafeWriter if one was created\n// This should be called at the end of the root request to flush any pending writes\n// and stop the background goroutine.\nfunc (ctx *Context) CloseSafeWriter() {\n\tif ctx.Writer == nil {\n\t\treturn\n\t}\n\tif sw, ok := ctx.Writer.(*output.SafeWriter); ok {\n\t\tsw.Close()\n\t}\n}\n" }, { "path": "agent/context/stack.go", "content": "package context\n\nimport (\n\t\"fmt\"\n\t\"time\"\n\n\t\"github.com/google/uuid\"\n\t\"github.com/yaoapp/yao/trace\"\n)\n\n// NewStack creates a new root stack with the given trace ID and assistant ID\nfunc NewStack(traceID, assistantID, referer string, opts *Options) *Stack {\n\tif traceID == \"\" {\n\t\ttraceID = uuid.New().String()\n\t}\n\n\tstackID := uuid.New().String()\n\tnow := time.Now().UnixMilli()\n\n\treturn &Stack{\n\t\tID: stackID,\n\t\tTraceID: traceID,\n\t\tAssistantID: assistantID,\n\t\tReferer: referer,\n\t\tDepth: 0,\n\t\tParentID: \"\",\n\t\tPath: []string{stackID},\n\t\tOptions: opts,\n\t\tCreatedAt: now,\n\t\tStatus: StackStatusRunning,\n\t}\n}\n\n// NewChildStack creates a child stack from the current stack\nfunc (s *Stack) NewChildStack(assistantID, referer string, opts *Options) *Stack {\n\tstackID := uuid.New().String()\n\tnow := time.Now().UnixMilli()\n\n\t// Build path by appending current stack's path with new ID\n\tpath := make([]string, len(s.Path)+1)\n\tcopy(path, s.Path)\n\tpath[len(s.Path)] = stackID\n\n\treturn &Stack{\n\t\tID: stackID,\n\t\tTraceID: s.TraceID, // Inherit trace ID\n\t\tAssistantID: assistantID,\n\t\tReferer: referer,\n\t\tDepth: s.Depth + 1,\n\t\tParentID: s.ID,\n\t\tPath: path,\n\t\tOptions: opts,\n\t\tCreatedAt: now,\n\t\tStatus: StackStatusRunning,\n\t}\n}\n\n// NewChildStackFromForkParent creates a child stack from ForkParentInfo\n// This is used by forked contexts (ctx.agent.Call) to create a child stack\n// without sharing the actual Stack reference (which would cause race conditions)\nfunc NewChildStackFromForkParent(parent *ForkParentInfo, assistantID, referer string, opts *Options) *Stack {\n\tstackID := uuid.New().String()\n\tnow := time.Now().UnixMilli()\n\n\t// Build path by appending parent's path with new ID\n\tpath := make([]string, len(parent.Path)+1)\n\tcopy(path, parent.Path)\n\tpath[len(parent.Path)] = stackID\n\n\treturn &Stack{\n\t\tID: stackID,\n\t\tTraceID: parent.TraceID, // Inherit trace ID from parent\n\t\tAssistantID: assistantID,\n\t\tReferer: referer,\n\t\tDepth: parent.Depth + 1,\n\t\tParentID: parent.StackID, // Use parent's stack ID\n\t\tPath: path,\n\t\tOptions: opts,\n\t\tCreatedAt: now,\n\t\tStatus: StackStatusRunning,\n\t}\n}\n\n// Complete marks the stack as completed and calculates duration\nfunc (s *Stack) Complete() {\n\tnow := time.Now().UnixMilli()\n\ts.CompletedAt = &now\n\ts.Status = StackStatusCompleted\n\tduration := now - s.CreatedAt\n\ts.DurationMs = &duration\n}\n\n// Fail marks the stack as failed with an error message\nfunc (s *Stack) Fail(err error) {\n\tnow := time.Now().UnixMilli()\n\ts.CompletedAt = &now\n\ts.Status = StackStatusFailed\n\tif err != nil {\n\t\ts.Error = err.Error()\n\t}\n\tduration := now - s.CreatedAt\n\ts.DurationMs = &duration\n}\n\n// Timeout marks the stack as timeout\nfunc (s *Stack) Timeout() {\n\tnow := time.Now().UnixMilli()\n\ts.CompletedAt = &now\n\ts.Status = StackStatusTimeout\n\tduration := now - s.CreatedAt\n\ts.DurationMs = &duration\n}\n\n// IsRoot returns true if this is a root stack (no parent)\nfunc (s *Stack) IsRoot() bool {\n\treturn s.ParentID == \"\"\n}\n\n// IsCompleted returns true if the stack has completed (success, failed, or timeout)\nfunc (s *Stack) IsCompleted() bool {\n\treturn s.Status == StackStatusCompleted ||\n\t\ts.Status == StackStatusFailed ||\n\t\ts.Status == StackStatusTimeout\n}\n\n// IsRunning returns true if the stack is currently running\nfunc (s *Stack) IsRunning() bool {\n\treturn s.Status == StackStatusRunning\n}\n\n// GetPathString returns the path as a string (e.g., \"root_id -> parent_id -> current_id\")\nfunc (s *Stack) GetPathString() string {\n\tif len(s.Path) == 0 {\n\t\treturn s.ID\n\t}\n\n\tresult := s.Path[0]\n\tfor i := 1; i < len(s.Path); i++ {\n\t\tresult += \" -> \" + s.Path[i]\n\t}\n\treturn result\n}\n\n// String returns a string representation of the stack for debugging\nfunc (s *Stack) String() string {\n\tstatus := s.Status\n\tif s.IsCompleted() && s.DurationMs != nil {\n\t\tstatus = fmt.Sprintf(\"%s (%dms)\", s.Status, *s.DurationMs)\n\t}\n\n\treturn fmt.Sprintf(\"Stack[ID=%s, TraceID=%s, Assistant=%s, Depth=%d, Status=%s]\",\n\t\ts.ID[:8], s.TraceID[:8], s.AssistantID, s.Depth, status)\n}\n\n// Clone creates a deep copy of the stack\nfunc (s *Stack) Clone() *Stack {\n\tclone := &Stack{\n\t\tID: s.ID,\n\t\tTraceID: s.TraceID,\n\t\tAssistantID: s.AssistantID,\n\t\tReferer: s.Referer,\n\t\tDepth: s.Depth,\n\t\tParentID: s.ParentID,\n\t\tPath: make([]string, len(s.Path)),\n\t\tOptions: s.Options, // Shallow copy of Options pointer\n\t\tCreatedAt: s.CreatedAt,\n\t\tStatus: s.Status,\n\t\tError: s.Error,\n\t}\n\n\tcopy(clone.Path, s.Path)\n\n\tif s.CompletedAt != nil {\n\t\tcompletedAt := *s.CompletedAt\n\t\tclone.CompletedAt = &completedAt\n\t}\n\n\tif s.DurationMs != nil {\n\t\tdurationMs := *s.DurationMs\n\t\tclone.DurationMs = &durationMs\n\t}\n\n\treturn clone\n}\n\n// EnterStack initializes or creates a child stack and returns it along with trace ID and completion function\n// This is a helper function to manage stack context for nested calls\n// The stack will be automatically saved to ctx.Stacks for trace logging\n//\n// Returns:\n// - *Stack: current stack\n// - string: trace ID (generated for root, inherited for children)\n// - func(): completion function to be deferred\n//\n// Usage:\n//\n//\tstack, traceID, done := context.EnterStack(ctx, assistantID, opts)\n//\tdefer done()\n//\t// ... your code here ...\nfunc EnterStack(ctx *Context, assistantID string, opts *Options) (*Stack, string, func()) {\n\tvar stack *Stack\n\tvar parentStack *Stack\n\tvar traceID string\n\n\t// Get referer from ctx (request source)\n\treferer := ctx.Referer\n\n\t// Initialize Stacks map if not exists\n\tif ctx.Stacks == nil {\n\t\tctx.Stacks = make(map[string]*Stack)\n\t}\n\n\tif ctx.Stack == nil {\n\t\t// Check if this is a forked context with parent stack info\n\t\tif ctx.ForkParent != nil {\n\t\t\t// Create child stack using ForkParent info\n\t\t\t// This is for forked contexts (ctx.agent.Call) to have proper ThreadID\n\t\t\ttraceID = ctx.ForkParent.TraceID\n\t\t\tstack = NewChildStackFromForkParent(ctx.ForkParent, assistantID, referer, opts)\n\t\t\tctx.Stack = stack\n\t\t} else {\n\t\t\t// Create root stack for this assistant call (entry point)\n\t\t\t// Generate a new trace ID for root\n\t\t\ttraceID = trace.GenTraceID()\n\t\t\tstack = NewStack(traceID, assistantID, referer, opts)\n\t\t\tctx.Stack = stack\n\t\t}\n\t} else {\n\t\t// Create child stack for nested agent call (delegate)\n\t\t// Inherit trace ID from parent\n\t\tparentStack = ctx.Stack\n\t\ttraceID = parentStack.TraceID\n\t\tstack = ctx.Stack.NewChildStack(assistantID, referer, opts)\n\t\tctx.Stack = stack\n\t}\n\n\t// Mark stack as running (in case it was pending)\n\tif stack.Status == StackStatusPending {\n\t\tstack.Status = StackStatusRunning\n\t}\n\n\t// Save stack to collection for trace logging\n\tctx.Stacks[stack.ID] = stack\n\n\t// Return completion function\n\tdone := func() {\n\t\t// Mark as completed if no panic occurred\n\t\tif !stack.IsCompleted() {\n\t\t\tstack.Complete()\n\t\t}\n\n\t\t// Restore parent stack\n\t\tif parentStack != nil {\n\t\t\tctx.Stack = parentStack\n\t\t}\n\t}\n\n\treturn stack, traceID, done\n}\n\n// GetAllStacks returns all stacks collected during the request\n// This is useful for trace logging after the request completes\nfunc (ctx *Context) GetAllStacks() []*Stack {\n\tif ctx.Stacks == nil {\n\t\treturn nil\n\t}\n\n\tstacks := make([]*Stack, 0, len(ctx.Stacks))\n\tfor _, s := range ctx.Stacks {\n\t\tstacks = append(stacks, s)\n\t}\n\treturn stacks\n}\n\n// GetStackByID returns a specific stack by its ID\n// This is useful for querying stack information during request processing\nfunc (ctx *Context) GetStackByID(id string) *Stack {\n\tif ctx.Stacks == nil {\n\t\treturn nil\n\t}\n\treturn ctx.Stacks[id]\n}\n\n// GetStacksByTraceID returns all stacks with the given trace ID\n// This is useful for getting the complete call tree for a trace\nfunc (ctx *Context) GetStacksByTraceID(traceID string) []*Stack {\n\tif ctx.Stacks == nil {\n\t\treturn nil\n\t}\n\n\tstacks := make([]*Stack, 0)\n\tfor _, s := range ctx.Stacks {\n\t\tif s.TraceID == traceID {\n\t\t\tstacks = append(stacks, s)\n\t\t}\n\t}\n\treturn stacks\n}\n\n// GetRootStack returns the root stack (depth = 0) of current trace\nfunc (ctx *Context) GetRootStack() *Stack {\n\tif ctx.Stacks == nil {\n\t\treturn nil\n\t}\n\n\tfor _, s := range ctx.Stacks {\n\t\tif s.IsRoot() {\n\t\t\treturn s\n\t\t}\n\t}\n\treturn nil\n}\n" }, { "path": "agent/context/stack_test.go", "content": "package context_test\n\nimport (\n\tstdContext \"context\"\n\t\"testing\"\n\t\"time\"\n\n\t\"github.com/yaoapp/yao/agent/context\"\n\t\"github.com/yaoapp/yao/config\"\n\t\"github.com/yaoapp/yao/test\"\n)\n\nfunc TestNewStack(t *testing.T) {\n\ttest.Prepare(t, config.Conf)\n\tdefer test.Clean()\n\n\ttraceID := \"12345678\"\n\tassistantID := \"test-assistant\"\n\treferer := context.RefererAPI\n\topts := &context.Options{}\n\n\tstack := context.NewStack(traceID, assistantID, referer, opts)\n\n\tif stack == nil {\n\t\tt.Fatal(\"Expected stack to be created, got nil\")\n\t}\n\n\tif stack.TraceID != traceID {\n\t\tt.Errorf(\"Expected TraceID '%s', got '%s'\", traceID, stack.TraceID)\n\t}\n\n\tif stack.AssistantID != assistantID {\n\t\tt.Errorf(\"Expected AssistantID '%s', got '%s'\", assistantID, stack.AssistantID)\n\t}\n\n\tif stack.Referer != referer {\n\t\tt.Errorf(\"Expected Referer '%s', got '%s'\", referer, stack.Referer)\n\t}\n\n\tif stack.Depth != 0 {\n\t\tt.Errorf(\"Expected Depth 0, got %d\", stack.Depth)\n\t}\n\n\tif stack.ParentID != \"\" {\n\t\tt.Errorf(\"Expected empty ParentID, got '%s'\", stack.ParentID)\n\t}\n\n\tif !stack.IsRoot() {\n\t\tt.Error(\"Expected stack to be root\")\n\t}\n\n\tif stack.Status != context.StackStatusRunning {\n\t\tt.Errorf(\"Expected Status '%s', got '%s'\", context.StackStatusRunning, stack.Status)\n\t}\n}\n\nfunc TestNewStack_GenerateTraceID(t *testing.T) {\n\ttest.Prepare(t, config.Conf)\n\tdefer test.Clean()\n\n\t// Empty traceID should generate a UUID\n\tstack := context.NewStack(\"\", \"test-assistant\", context.RefererAPI, &context.Options{})\n\n\tif stack.TraceID == \"\" {\n\t\tt.Error(\"Expected TraceID to be generated, got empty string\")\n\t}\n\n\t// Should be a valid UUID (36 characters with dashes)\n\tif len(stack.TraceID) < 8 {\n\t\tt.Errorf(\"Expected TraceID to be at least 8 characters, got %d\", len(stack.TraceID))\n\t}\n}\n\nfunc TestNewChildStack(t *testing.T) {\n\ttest.Prepare(t, config.Conf)\n\tdefer test.Clean()\n\n\t// Create parent stack\n\tparentStack := context.NewStack(\"12345678\", \"parent-assistant\", context.RefererAPI, &context.Options{})\n\n\t// Create child stack\n\tchildStack := parentStack.NewChildStack(\"child-assistant\", context.RefererAgent, &context.Options{})\n\n\tif childStack == nil {\n\t\tt.Fatal(\"Expected child stack to be created, got nil\")\n\t}\n\n\t// Child should inherit TraceID\n\tif childStack.TraceID != parentStack.TraceID {\n\t\tt.Errorf(\"Expected child TraceID '%s', got '%s'\", parentStack.TraceID, childStack.TraceID)\n\t}\n\n\t// Child should have parent ID\n\tif childStack.ParentID != parentStack.ID {\n\t\tt.Errorf(\"Expected ParentID '%s', got '%s'\", parentStack.ID, childStack.ParentID)\n\t}\n\n\t// Child should have incremented depth\n\tif childStack.Depth != parentStack.Depth+1 {\n\t\tt.Errorf(\"Expected Depth %d, got %d\", parentStack.Depth+1, childStack.Depth)\n\t}\n\n\t// Child should not be root\n\tif childStack.IsRoot() {\n\t\tt.Error(\"Expected child stack not to be root\")\n\t}\n\n\t// Path should include both parent and child\n\tif len(childStack.Path) != 2 {\n\t\tt.Errorf(\"Expected Path length 2, got %d\", len(childStack.Path))\n\t}\n\n\tif childStack.Path[0] != parentStack.ID {\n\t\tt.Errorf(\"Expected first path element '%s', got '%s'\", parentStack.ID, childStack.Path[0])\n\t}\n\n\tif childStack.Path[1] != childStack.ID {\n\t\tt.Errorf(\"Expected second path element '%s', got '%s'\", childStack.ID, childStack.Path[1])\n\t}\n}\n\nfunc TestStackComplete(t *testing.T) {\n\ttest.Prepare(t, config.Conf)\n\tdefer test.Clean()\n\n\tstack := context.NewStack(\"12345678\", \"test-assistant\", context.RefererAPI, &context.Options{})\n\n\t// Wait a bit to have measurable duration\n\ttime.Sleep(10 * time.Millisecond)\n\n\tstack.Complete()\n\n\tif stack.Status != context.StackStatusCompleted {\n\t\tt.Errorf(\"Expected Status '%s', got '%s'\", context.StackStatusCompleted, stack.Status)\n\t}\n\n\tif stack.CompletedAt == nil {\n\t\tt.Error(\"Expected CompletedAt to be set, got nil\")\n\t}\n\n\tif stack.DurationMs == nil {\n\t\tt.Error(\"Expected DurationMs to be set, got nil\")\n\t}\n\n\tif *stack.DurationMs < 10 {\n\t\tt.Errorf(\"Expected DurationMs to be at least 10ms, got %d\", *stack.DurationMs)\n\t}\n\n\tif !stack.IsCompleted() {\n\t\tt.Error(\"Expected stack to be completed\")\n\t}\n\n\tif stack.IsRunning() {\n\t\tt.Error(\"Expected stack not to be running\")\n\t}\n}\n\nfunc TestStackFail(t *testing.T) {\n\ttest.Prepare(t, config.Conf)\n\tdefer test.Clean()\n\n\tstack := context.NewStack(\"12345678\", \"test-assistant\", context.RefererAPI, &context.Options{})\n\n\ttestError := \"test error message\"\n\tstack.Fail(nil)\n\tstack.Error = testError\n\n\tif stack.Status != context.StackStatusFailed {\n\t\tt.Errorf(\"Expected Status '%s', got '%s'\", context.StackStatusFailed, stack.Status)\n\t}\n\n\tif stack.Error != testError {\n\t\tt.Errorf(\"Expected Error '%s', got '%s'\", testError, stack.Error)\n\t}\n\n\tif !stack.IsCompleted() {\n\t\tt.Error(\"Expected failed stack to be completed\")\n\t}\n}\n\nfunc TestStackTimeout(t *testing.T) {\n\ttest.Prepare(t, config.Conf)\n\tdefer test.Clean()\n\n\tstack := context.NewStack(\"12345678\", \"test-assistant\", context.RefererAPI, &context.Options{})\n\n\tstack.Timeout()\n\n\tif stack.Status != context.StackStatusTimeout {\n\t\tt.Errorf(\"Expected Status '%s', got '%s'\", context.StackStatusTimeout, stack.Status)\n\t}\n\n\tif !stack.IsCompleted() {\n\t\tt.Error(\"Expected timeout stack to be completed\")\n\t}\n}\n\nfunc TestEnterStack_RootCreation(t *testing.T) {\n\ttest.Prepare(t, config.Conf)\n\tdefer test.Clean()\n\n\tctx := context.New(stdContext.Background(), nil, \"test-chat-id\")\n\tctx.Referer = context.RefererAPI\n\n\tstack, traceID, done := context.EnterStack(ctx, \"test-assistant\", &context.Options{})\n\tdefer done()\n\n\tif stack == nil {\n\t\tt.Fatal(\"Expected stack to be created, got nil\")\n\t}\n\n\tif traceID == \"\" {\n\t\tt.Error(\"Expected traceID to be generated, got empty string\")\n\t}\n\n\t// TraceID should be at least 8 digits (from trace.GenTraceID)\n\tif len(traceID) < 8 {\n\t\tt.Errorf(\"Expected traceID length at least 8, got %d\", len(traceID))\n\t}\n\n\tif stack.TraceID != traceID {\n\t\tt.Errorf(\"Expected stack TraceID '%s', got '%s'\", traceID, stack.TraceID)\n\t}\n\n\tif ctx.Stack != stack {\n\t\tt.Error(\"Expected ctx.Stack to be set to created stack\")\n\t}\n\n\tif ctx.Stacks == nil {\n\t\tt.Fatal(\"Expected ctx.Stacks to be initialized, got nil\")\n\t}\n\n\tif ctx.Stacks[stack.ID] != stack {\n\t\tt.Error(\"Expected stack to be saved in ctx.Stacks\")\n\t}\n\n\tif !stack.IsRoot() {\n\t\tt.Error(\"Expected stack to be root\")\n\t}\n}\n\nfunc TestEnterStack_ChildCreation(t *testing.T) {\n\ttest.Prepare(t, config.Conf)\n\tdefer test.Clean()\n\n\tctx := context.New(stdContext.Background(), nil, \"test-chat-id\")\n\tctx.Referer = context.RefererAPI\n\n\t// Create parent\n\tparentStack, parentTraceID, parentDone := context.EnterStack(ctx, \"parent-assistant\", &context.Options{})\n\tdefer parentDone()\n\n\tif parentStack == nil {\n\t\tt.Fatal(\"Expected parent stack to be created, got nil\")\n\t}\n\n\t// Create child\n\tchildStack, childTraceID, childDone := context.EnterStack(ctx, \"child-assistant\", &context.Options{})\n\tdefer childDone()\n\n\tif childStack == nil {\n\t\tt.Fatal(\"Expected child stack to be created, got nil\")\n\t}\n\n\t// Child should inherit trace ID\n\tif childTraceID != parentTraceID {\n\t\tt.Errorf(\"Expected child traceID '%s', got '%s'\", parentTraceID, childTraceID)\n\t}\n\n\t// Child should have parent ID\n\tif childStack.ParentID != parentStack.ID {\n\t\tt.Errorf(\"Expected child ParentID '%s', got '%s'\", parentStack.ID, childStack.ParentID)\n\t}\n\n\t// Both should be saved in ctx.Stacks\n\tif len(ctx.Stacks) != 2 {\n\t\tt.Errorf(\"Expected 2 stacks in ctx.Stacks, got %d\", len(ctx.Stacks))\n\t}\n\n\t// Current stack should be child\n\tif ctx.Stack != childStack {\n\t\tt.Error(\"Expected ctx.Stack to be child stack\")\n\t}\n}\n\nfunc TestEnterStack_DoneCallback(t *testing.T) {\n\ttest.Prepare(t, config.Conf)\n\tdefer test.Clean()\n\n\tctx := context.New(stdContext.Background(), nil, \"test-chat-id\")\n\tctx.Referer = context.RefererAPI\n\n\t// Create parent\n\tparentStack, _, parentDone := context.EnterStack(ctx, \"parent-assistant\", &context.Options{})\n\n\t// Create child\n\tchildStack, _, childDone := context.EnterStack(ctx, \"child-assistant\", &context.Options{})\n\n\t// Child should be current\n\tif ctx.Stack != childStack {\n\t\tt.Error(\"Expected ctx.Stack to be child stack before done\")\n\t}\n\n\t// Call child done\n\tchildDone()\n\n\t// Parent should be restored\n\tif ctx.Stack != parentStack {\n\t\tt.Error(\"Expected ctx.Stack to be restored to parent stack after child done\")\n\t}\n\n\t// Child should be completed\n\tif !childStack.IsCompleted() {\n\t\tt.Error(\"Expected child stack to be completed after done\")\n\t}\n\n\t// Call parent done\n\tparentDone()\n\n\t// Parent should be completed\n\tif !parentStack.IsCompleted() {\n\t\tt.Error(\"Expected parent stack to be completed after done\")\n\t}\n}\n\nfunc TestContextGetAllStacks(t *testing.T) {\n\ttest.Prepare(t, config.Conf)\n\tdefer test.Clean()\n\n\tctx := context.New(stdContext.Background(), nil, \"test-chat-id\")\n\tctx.Referer = context.RefererAPI\n\n\t// Create multiple stacks\n\t_, _, done1 := context.EnterStack(ctx, \"assistant1\", &context.Options{})\n\tdefer done1()\n\n\t_, _, done2 := context.EnterStack(ctx, \"assistant2\", &context.Options{})\n\tdefer done2()\n\n\t_, _, done3 := context.EnterStack(ctx, \"assistant3\", &context.Options{})\n\tdefer done3()\n\n\t// Get all stacks\n\tallStacks := ctx.GetAllStacks()\n\n\tif len(allStacks) != 3 {\n\t\tt.Errorf(\"Expected 3 stacks, got %d\", len(allStacks))\n\t}\n}\n\nfunc TestContextGetStackByID(t *testing.T) {\n\ttest.Prepare(t, config.Conf)\n\tdefer test.Clean()\n\n\tctx := context.New(stdContext.Background(), nil, \"test-chat-id\")\n\tctx.Referer = context.RefererAPI\n\n\tstack, _, done := context.EnterStack(ctx, \"test-assistant\", &context.Options{})\n\tdefer done()\n\n\t// Get stack by ID\n\tfound := ctx.GetStackByID(stack.ID)\n\n\tif found == nil {\n\t\tt.Fatal(\"Expected to find stack, got nil\")\n\t}\n\n\tif found.ID != stack.ID {\n\t\tt.Errorf(\"Expected stack ID '%s', got '%s'\", stack.ID, found.ID)\n\t}\n\n\t// Try to get non-existent stack\n\tnotFound := ctx.GetStackByID(\"non-existent-id\")\n\tif notFound != nil {\n\t\tt.Error(\"Expected nil for non-existent stack ID\")\n\t}\n}\n\nfunc TestContextGetStacksByTraceID(t *testing.T) {\n\ttest.Prepare(t, config.Conf)\n\tdefer test.Clean()\n\n\tctx := context.New(stdContext.Background(), nil, \"test-chat-id\")\n\tctx.Referer = context.RefererAPI\n\n\t// Create parent and child (same trace ID)\n\t_, traceID, done1 := context.EnterStack(ctx, \"parent-assistant\", &context.Options{})\n\tdefer done1()\n\n\t_, _, done2 := context.EnterStack(ctx, \"child-assistant\", &context.Options{})\n\tdefer done2()\n\n\t// Get stacks by trace ID\n\tstacks := ctx.GetStacksByTraceID(traceID)\n\n\tif len(stacks) != 2 {\n\t\tt.Errorf(\"Expected 2 stacks with trace ID '%s', got %d\", traceID, len(stacks))\n\t}\n\n\t// All should have same trace ID\n\tfor _, s := range stacks {\n\t\tif s.TraceID != traceID {\n\t\t\tt.Errorf(\"Expected TraceID '%s', got '%s'\", traceID, s.TraceID)\n\t\t}\n\t}\n}\n\nfunc TestContextGetRootStack(t *testing.T) {\n\ttest.Prepare(t, config.Conf)\n\tdefer test.Clean()\n\n\tctx := context.New(stdContext.Background(), nil, \"test-chat-id\")\n\tctx.Referer = context.RefererAPI\n\n\t// Create parent\n\tparentStack, _, done1 := context.EnterStack(ctx, \"parent-assistant\", &context.Options{})\n\tdefer done1()\n\n\t// Create child\n\t_, _, done2 := context.EnterStack(ctx, \"child-assistant\", &context.Options{})\n\tdefer done2()\n\n\t// Get root stack\n\trootStack := ctx.GetRootStack()\n\n\tif rootStack == nil {\n\t\tt.Fatal(\"Expected to find root stack, got nil\")\n\t}\n\n\tif rootStack.ID != parentStack.ID {\n\t\tt.Errorf(\"Expected root stack ID '%s', got '%s'\", parentStack.ID, rootStack.ID)\n\t}\n\n\tif !rootStack.IsRoot() {\n\t\tt.Error(\"Expected returned stack to be root\")\n\t}\n}\n\nfunc TestStackClone(t *testing.T) {\n\ttest.Prepare(t, config.Conf)\n\tdefer test.Clean()\n\n\toriginal := context.NewStack(\"12345678\", \"test-assistant\", context.RefererAPI, &context.Options{})\n\toriginal.Complete()\n\n\tclone := original.Clone()\n\n\tif clone == nil {\n\t\tt.Fatal(\"Expected clone to be created, got nil\")\n\t}\n\n\t// Check all fields are copied\n\tif clone.ID != original.ID {\n\t\tt.Error(\"ID not cloned correctly\")\n\t}\n\n\tif clone.TraceID != original.TraceID {\n\t\tt.Error(\"TraceID not cloned correctly\")\n\t}\n\n\tif clone.AssistantID != original.AssistantID {\n\t\tt.Error(\"AssistantID not cloned correctly\")\n\t}\n\n\tif clone.Status != original.Status {\n\t\tt.Error(\"Status not cloned correctly\")\n\t}\n\n\t// Check deep copy of Path\n\tif len(clone.Path) != len(original.Path) {\n\t\tt.Error(\"Path length not cloned correctly\")\n\t}\n\n\t// Modify clone's path shouldn't affect original\n\tif len(clone.Path) > 0 {\n\t\tclone.Path[0] = \"modified\"\n\t\tif original.Path[0] == \"modified\" {\n\t\t\tt.Error(\"Path is not deeply copied\")\n\t\t}\n\t}\n}\n" }, { "path": "agent/context/types.go", "content": "package context\n\nimport (\n\t\"context\"\n\t\"sync\"\n\t\"time\"\n\n\t\"github.com/yaoapp/gou/llm\"\n\t\"github.com/yaoapp/gou/store\"\n\t\"github.com/yaoapp/yao/agent/memory\"\n\t\"github.com/yaoapp/yao/agent/output\"\n\t\"github.com/yaoapp/yao/agent/output/message\"\n\t\"github.com/yaoapp/yao/openapi/oauth/types\"\n\tinfraV2 \"github.com/yaoapp/yao/sandbox/v2\"\n\t\"github.com/yaoapp/yao/tai/workspace\"\n\ttraceTypes \"github.com/yaoapp/yao/trace/types\"\n)\n\n// Accept the accept of the request, it will be used to identify the accept of the request.\ntype Accept string\n\n// Referer the referer of the request, it will be used to identify the referer of the request.\ntype Referer string\n\n// Client represents the client information from HTTP request\ntype Client struct {\n\tType string `json:\"type,omitempty\"` // Client type: web, android, ios, windows, macos, linux, agent, jssdk\n\tUserAgent string `json:\"user_agent,omitempty\"` // Original User-Agent header\n\tIP string `json:\"ip,omitempty\"` // Client IP address\n}\n\nconst (\n\t// AcceptStandard standard response format compatible with OpenAI API and general chat UIs (default)\n\tAcceptStandard = \"standard\"\n\n\t// AcceptWebCUI web-based CUI format with action request support for Yao Chat User Interface\n\tAcceptWebCUI = \"cui-web\"\n\n\t// AccepNativeCUI native mobile/tablet CUI format with action request support\n\tAccepNativeCUI = \"cui-native\"\n\n\t// AcceptDesktopCUI desktop CUI format with action request support\n\tAcceptDesktopCUI = \"cui-desktop\"\n)\n\n// ValidAccepts is the map of valid accept types\nvar ValidAccepts = map[string]bool{\n\tAcceptStandard: true,\n\tAcceptWebCUI: true,\n\tAccepNativeCUI: true,\n\tAcceptDesktopCUI: true,\n}\n\nconst (\n\t// RefererAPI request from HTTP API endpoint\n\tRefererAPI = \"api\"\n\n\t// RefererProcess request from Yao Process call\n\tRefererProcess = \"process\"\n\n\t// RefererMCP request from MCP (Model Context Protocol) server\n\tRefererMCP = \"mcp\"\n\n\t// RefererJSSDK request from JavaScript SDK\n\tRefererJSSDK = \"jssdk\"\n\n\t// RefererAgent request from agent-to-agent delegate call (same context, saves history)\n\tRefererAgent = \"agent\"\n\n\t// RefererAgentFork request from agent-to-agent fork call (ctx.agent.Call/All/Any/Race, skips history)\n\tRefererAgentFork = \"agent_fork\"\n\n\t// RefererTool request from tool/function execution\n\tRefererTool = \"tool\"\n\n\t// RefererHook request from hook trigger (on_message, on_error, etc.)\n\tRefererHook = \"hook\"\n\n\t// RefererSchedule request from scheduled task or cron job\n\tRefererSchedule = \"schedule\"\n\n\t// RefererScript request from custom script execution\n\tRefererScript = \"script\"\n\n\t// RefererInternal request from internal system call\n\tRefererInternal = \"internal\"\n)\n\n// ValidReferers is the map of valid referer types\nvar ValidReferers = map[string]bool{\n\tRefererAPI: true,\n\tRefererProcess: true,\n\tRefererMCP: true,\n\tRefererJSSDK: true,\n\tRefererAgent: true,\n\tRefererAgentFork: true,\n\tRefererTool: true,\n\tRefererHook: true,\n\tRefererSchedule: true,\n\tRefererScript: true,\n\tRefererInternal: true,\n}\n\nconst (\n\t// StackStatusPending stack is created but not started yet\n\tStackStatusPending = \"pending\"\n\n\t// StackStatusRunning stack is currently executing\n\tStackStatusRunning = \"running\"\n\n\t// StackStatusCompleted stack completed successfully\n\tStackStatusCompleted = \"completed\"\n\n\t// StackStatusFailed stack failed with error\n\tStackStatusFailed = \"failed\"\n\n\t// StackStatusTimeout stack execution timeout\n\tStackStatusTimeout = \"timeout\"\n)\n\n// ValidStackStatus is the map of valid stack status types\nvar ValidStackStatus = map[string]bool{\n\tStackStatusPending: true,\n\tStackStatusRunning: true,\n\tStackStatusCompleted: true,\n\tStackStatusFailed: true,\n\tStackStatusTimeout: true,\n}\n\n// Interrupt Types and Constants\n// ===============================\n\n// InterruptType represents the type of interrupt\ntype InterruptType string\n\nconst (\n\t// InterruptGraceful waits for current step to complete before handling interrupt\n\tInterruptGraceful InterruptType = \"graceful\"\n\n\t// InterruptForce immediately cancels current operation and handles interrupt\n\tInterruptForce InterruptType = \"force\"\n)\n\n// InterruptAction represents the action to take after interrupt is handled\ntype InterruptAction string\n\nconst (\n\t// InterruptActionContinue appends new messages and continues execution\n\tInterruptActionContinue InterruptAction = \"continue\"\n\n\t// InterruptActionRestart restarts execution with only new messages\n\tInterruptActionRestart InterruptAction = \"restart\"\n\n\t// InterruptActionAbort terminates the request\n\tInterruptActionAbort InterruptAction = \"abort\"\n)\n\n// InterruptSignal represents an interrupt signal with new messages from user\ntype InterruptSignal struct {\n\tType InterruptType `json:\"type\"` // Interrupt type: graceful or force\n\tMessages []Message `json:\"messages\"` // User's new messages (can be multiple)\n\tTimestamp int64 `json:\"timestamp\"` // Interrupt timestamp in milliseconds\n\tMetadata map[string]interface{} `json:\"metadata,omitempty\"` // Additional metadata\n}\n\n// InterruptHandler is the function signature for handling interrupts\n// This handler is registered in the InterruptController and called when interrupt signal is received\n// Parameters:\n// - ctx: The context being interrupted\n// - signal: The interrupt signal (contains Type and Messages)\n//\n// Returns:\n// - error: Error if interrupt handling failed\ntype InterruptHandler func(ctx *Context, signal *InterruptSignal) error\n\n// InterruptController manages interrupt handling for a context\n// All interrupt-related fields are encapsulated in this type\ntype InterruptController struct {\n\tqueue chan *InterruptSignal `json:\"-\"` // Queue to receive interrupt signals\n\tcurrent *InterruptSignal `json:\"-\"` // Current interrupt being processed\n\tpending []*InterruptSignal `json:\"-\"` // Pending interrupts in queue\n\tmutex sync.RWMutex `json:\"-\"` // Protects current and pending\n\tctx context.Context `json:\"-\"` // Interrupt control context (independent from HTTP context)\n\tcancel context.CancelFunc `json:\"-\"` // Cancel function for force interrupt\n\tlistenerStarted bool `json:\"-\"` // Whether listener goroutine is started\n\thandler InterruptHandler `json:\"-\"` // Handler to process interrupt signals\n\tcontextID string `json:\"-\"` // Context ID to retrieve the parent context\n}\n\n// AssistantInfo represents the assistant information structure\ntype AssistantInfo struct {\n\tID string `json:\"assistant_id\"` // Assistant ID\n\tType string `json:\"type,omitempty\"` // Assistant Type, default is assistant\n\tName string `json:\"name,omitempty\"` // Assistant Name\n\tAvatar string `json:\"avatar,omitempty\"` // Assistant Avatar\n\tDescription string `json:\"description,omitempty\"` // Assistant Description\n}\n\n// Skip configuration for what to skip in this request\ntype Skip struct {\n\tHistory bool `json:\"history\"` // Skip saving chat history (for internal calls like title/prompt generation)\n\tTrace bool `json:\"trace\"` // Skip trace logging\n\tOutput bool `json:\"output\"` // Skip output to client (for internal A2A calls that only need response data)\n\tKeyword bool `json:\"keyword\"` // Skip keyword extraction for web search (use raw query directly)\n\tSearch bool `json:\"search\"` // Skip auto search (for internal calls like needsearch intent detection)\n\tContentParsing bool `json:\"content_parsing\"` // Skip content parsing (vision, PDF, docx, etc.), convert files to raw text directly\n}\n\n// MessageMetadata stores metadata for sent messages\n// Used to inherit BlockID and ThreadID in delta operations\ntype MessageMetadata struct {\n\tMessageID string // Message ID\n\tBlockID string // Block ID\n\tThreadID string // Thread ID\n\tType string // Message type (text, thinking, etc.)\n\tStartTime time.Time // Message start time (for calculating duration)\n\tChunkCount int // Number of chunks sent for this message\n}\n\n// BlockMetadata stores metadata for output blocks\ntype BlockMetadata struct {\n\tBlockID string // Block ID\n\tType string // Block type (llm, mcp, agent, etc.)\n\tStartTime time.Time // Block start time\n\tMessageCount int // Number of messages in this block\n}\n\n// Context the context\ntype Context struct {\n\n\t// Context\n\tcontext.Context\n\n\t// External\n\tID string `json:\"id\"` // Context ID for external interrupt identification\n\tMemory *memory.Memory `json:\"-\"` // Agent memory with four spaces: User, Team, Chat, Context\n\tCache store.Store `json:\"-\"` // Cache store, it will be used to store the message cache, default is \"__yao.agent.cache\"\n\tStack *Stack `json:\"-\"` // Stack, current active stack of the request\n\tStacks map[string]*Stack `json:\"-\"` // Stacks, all stacks in this request (for trace logging)\n\tWriter Writer `json:\"-\"` // Writer, it will be used to write response data to the client\n\tIDGenerator *message.IDGenerator `json:\"-\"` // ID generator for this context (chunk, message, block, thread IDs)\n\tLogger *RequestLogger `json:\"-\"` // Request-scoped async logger\n\n\t// ForkParent stores parent stack info for forked contexts (set by Fork())\n\t// This allows EnterStack to create a child stack instead of root stack\n\t// without sharing the actual Stack reference (which would cause race conditions)\n\tForkParent *ForkParentInfo `json:\"-\"`\n\n\t// Chat buffer for batch saving messages and resume steps\n\tBuffer *ChatBuffer `json:\"-\"` // Chat buffer for batch saving at end of Stream()\n\n\t// Internal\n\ttrace traceTypes.Manager `json:\"-\"` // Trace manager, lazy initialized on first access\n\tmessageMetadata *messageMetadataStore `json:\"-\"` // Thread-safe message metadata store for delta operations\n\tsandboxExecutor SandboxExecutor `json:\"-\"` // Sandbox executor for hooks (set by assistant when sandbox is configured)\n\tcomputer infraV2.Computer `json:\"-\"` // V2 sandbox computer (set by assistant when V2 sandbox is configured)\n\tworkspace workspace.FS `json:\"-\"` // V2 workspace FS (derived from computer.Workplace())\n\n\t// Model capabilities (set by assistant, used by output adapters)\n\tCapabilities *llm.Capabilities `json:\"-\"` // Model capabilities for the current connector\n\n\t// Interrupt control (all interrupt-related logic is encapsulated in InterruptController)\n\tInterrupt *InterruptController `json:\"-\"` // Interrupt controller for handling user interrupts during streaming\n\n\t// Authorized information\n\tAuthorized *types.AuthorizedInfo `json:\"authorized,omitempty\"` // Authorized information\n\tChatID string `json:\"chat_id,omitempty\"` // Chat ID, use to select chat\n\tAssistantID string `json:\"assistant_id,omitempty\"` // Assistant ID, use to select assistant\n\n\t// Locale information\n\tLocale string `json:\"locale,omitempty\"` // Locale\n\tTheme string `json:\"theme,omitempty\"` // Theme\n\n\t// Request information\n\tClient Client `json:\"client,omitempty\"` // Client information from HTTP request\n\tReferer string `json:\"referer,omitempty\"` // Request source: api, process, mcp, jssdk, agent, tool, hook, schedule, script, internal\n\tAccept Accept `json:\"accept,omitempty\"` // Response format: standard, cui-web, cui-native, cui-desktop\n\n\t// CUI Context information\n\tRoute string `json:\"route,omitempty\"` // The route of the request, it will be used to identify the route of the request\n\tMetadata map[string]interface{} `json:\"metadata,omitempty\"` // The metadata of the request, it will be used to pass data to the page\n}\n\n// SearchIntent represents the result of search intent detection\n// Used by Create hook to specify fine-grained search behavior\ntype SearchIntent struct {\n\tNeedSearch bool `json:\"need_search\"` // Whether search is needed\n\tSearchTypes []string `json:\"search_types,omitempty\"` // Types of search to perform: \"web\", \"kb\", \"db\"\n\tConfidence float64 `json:\"confidence,omitempty\"` // Confidence level (0-1)\n\tReason string `json:\"reason,omitempty\"` // Reason for the decision\n}\n\n// Options represents the options for the context\ntype Options struct {\n\n\t// Original context, override the default context\n\tContext context.Context `json:\"-\"` // Context, it will be used to pass the context to the call\n\n\t// Writer, use to write response data to the client (override the default writer)\n\tWriter Writer `json:\"writer,omitempty\"` // Writer, use to write response data to the client\n\n\t// Skip configuration (history, trace, etc.), nil means don't skip anything\n\tSkip *Skip `json:\"skip,omitempty\"` // Skip configuration (history, trace, etc.), nil means don't skip anything\n\n\t// Connector, use to select the connector of the LLM Model, Default is Assistant.Connector\n\tConnector string `json:\"connector,omitempty\"` // Connector, use to select the connector of the LLM Model, Default is Assistant.Connector\n\n\t// Disable global prompts, default is false\n\tDisableGlobalPrompts bool `json:\"disable_global_prompts,omitempty\"` // Temporarily disable global prompts for this request\n\n\t// Search controls search behavior, supports multiple types:\n\t// - bool: true = enable all search types, false = disable all search\n\t// - SearchIntent: fine-grained control with specific types, confidence, etc.\n\t// - nil: use default behavior (determined by __yao.needsearch agent)\n\tSearch any `json:\"search,omitempty\"` // Search mode: bool | SearchIntent | nil\n\n\t// Agent mode, use to select the mode of the request, default is \"chat\"\n\tMode string `json:\"mode,omitempty\"` // Agent mode, use to select the mode of the request, default is \"chat\"\n\n\t// Uses configuration, allow hook to override wrapper configurations for vision, audio, search, and fetch\n\tUses *Uses `json:\"uses,omitempty\"` // Uses configuration, allow hook to override wrapper configurations for vision, audio, search, and fetch\n\n\t// Metadata for passing custom data to hooks (e.g., scenario selection)\n\tMetadata map[string]any `json:\"metadata,omitempty\"` // Custom metadata passed to Create/Next hooks\n\n\t// HistorySize controls the max number of history messages loaded for LLM context.\n\t// Priority: HistorySize > StoreSetting.MaxSize > default (20)\n\t// 0 means use StoreSetting or default.\n\tHistorySize int `json:\"history_size,omitempty\"`\n\n\t// OnMessage is called for each message sent via ctx.Send()\n\t// Used by ctx.agent.Call with onChunk callback to receive SSE messages\n\t// Returns: 0 = continue, non-zero = stop\n\tOnMessage OnMessageFunc `json:\"-\"`\n}\n\n// ForceA2A sets the options for Agent-to-Agent (A2A) calls.\n// For A2A calls:\n// - Output is NOT skipped - sub-agents output normally with ThreadID\n// - History IS skipped - A2A messages should not be saved to chat history\n// If Skip is nil, it creates a new Skip instance.\nfunc (opts *Options) ForceA2A() {\n\tif opts.Skip == nil {\n\t\topts.Skip = &Skip{}\n\t}\n\topts.Skip.History = true\n\t// Note: skip.output is NOT set - sub-agents output normally with ThreadID\n}\n\n// OnMessageFunc is a callback function for receiving output messages\n// Called for each message sent via ctx.Send() - same as SSE messages to client\n// Returns: 0 = continue, non-zero = stop sending\ntype OnMessageFunc func(msg *message.Message) int\n\n// ForkParentInfo stores parent stack information for forked contexts\n// This is used by EnterStack to create a child stack with proper inheritance\n// without sharing the actual Stack reference (which would cause race conditions in parallel calls)\ntype ForkParentInfo struct {\n\tStackID string // Parent stack ID (used as ParentID for child stack)\n\tTraceID string // Parent trace ID (inherited by child stack)\n\tDepth int // Parent depth (child depth = parent depth + 1)\n\tPath []string // Parent path (child path = parent path + child ID)\n}\n\n// Stack represents the call stack node for tracing agent-to-agent calls\n// Uses a flat structure to avoid circular references and memory overhead\ntype Stack struct {\n\t// Identity\n\tID string `json:\"id\"` // Unique stack node ID, used to identify this specific call\n\tTraceID string `json:\"trace_id\"` // Shared trace ID for entire call tree, inherited from root\n\n\t// Options\n\tOptions *Options `json:\"options,omitempty\"` // Options for the call\n\n\t// Call context\n\tAssistantID string `json:\"assistant_id\"` // Assistant handling this call\n\tReferer string `json:\"referer,omitempty\"` // Call source: api, agent, tool, process, etc.\n\tDepth int `json:\"depth\"` // Call depth in the tree (0=root)\n\n\t// Relationships\n\tParentID string `json:\"parent_id,omitempty\"` // Parent stack ID (empty for root call)\n\tPath []string `json:\"path\"` // Full path from root: [root_id, parent_id, ..., this_id]\n\n\t// Tracking\n\tCreatedAt int64 `json:\"created_at\"` // Unix timestamp in milliseconds\n\tCompletedAt *int64 `json:\"completed_at,omitempty\"` // Unix timestamp when completed (nil if ongoing)\n\tStatus string `json:\"status\"` // Status: pending, running, completed, failed, timeout\n\tError string `json:\"error,omitempty\"` // Error message if failed\n\n\t// Metrics\n\tDurationMs *int64 `json:\"duration_ms,omitempty\"` // Duration in milliseconds (calculated when completed)\n\n\t// Runtime cache (not serialized)\n\toutput *output.Output `json:\"-\"` // Cached output instance for this stack\n}\n\n// Response the response\n// 100% compatible with the OpenAI API\ntype Response struct {\n\tRequestID string `json:\"request_id\"` // Request ID for the response\n\tContextID string `json:\"context_id\"` // Context ID for the response\n\tTraceID string `json:\"trace_id\"` // Trace ID for the response\n\tChatID string `json:\"chat_id\"` // Chat ID for the response\n\tAssistantID string `json:\"assistant_id\"` // Assistant ID for the response\n\tCreate *HookCreateResponse `json:\"create,omitempty\"` // Create response from the create hook\n\tNext interface{} `json:\"next,omitempty\"` // Next response from the next hook\n\tCompletion *CompletionResponse `json:\"completion,omitempty\"` // Completion response from the completion hook\n\tTools []ToolCallResponse `json:\"tools,omitempty\"` // Tool call results (if any tools were executed)\n}\n\n// HookCreateResponse the response of the create hook\ntype HookCreateResponse struct {\n\n\t// Messages to be sent to the assistant\n\tMessages []Message `json:\"messages,omitempty\"`\n\n\t// Audio configuration (for models that support audio output)\n\tAudio *AudioConfig `json:\"audio,omitempty\"`\n\n\t// Generation parameters\n\tTemperature *float64 `json:\"temperature,omitempty\"`\n\tMaxTokens *int `json:\"max_tokens,omitempty\"`\n\tMaxCompletionTokens *int `json:\"max_completion_tokens,omitempty\"`\n\n\t// MCP configuration - allow hook to add/override MCP servers for this request\n\tMCPServers []MCPServerConfig `json:\"mcp_servers,omitempty\"`\n\n\t// Prompt configuration\n\tPromptPreset string `json:\"prompt_preset,omitempty\"` // Select prompt preset (e.g., \"chat.friendly\", \"task.analysis\")\n\tDisableGlobalPrompts *bool `json:\"disable_global_prompts,omitempty\"` // Temporarily disable global prompts for this request\n\n\t// Context adjustments - allow hook to modify context fields\n\tConnector string `json:\"connector,omitempty\"` // Override connector (call-level)\n\tLocale string `json:\"locale,omitempty\"` // Override locale (session-level)\n\tTheme string `json:\"theme,omitempty\"` // Override theme (session-level)\n\tRoute string `json:\"route,omitempty\"` // Override route (session-level)\n\tMetadata map[string]interface{} `json:\"metadata,omitempty\"` // Override or merge metadata (session-level)\n\n\t// Uses configuration - allow hook to override wrapper configurations\n\tUses *Uses `json:\"uses,omitempty\"` // Override wrapper configurations for vision, audio, search, and fetch\n\n\t// ForceUses controls whether to force using Uses tools even when model has native capabilities\n\tForceUses *bool `json:\"force_uses,omitempty\"` // Force using Uses tools regardless of model capabilities\n\n\t// Search controls search behavior, supports multiple types:\n\t// - bool: true = enable all search types, false = disable all search\n\t// - SearchIntent: fine-grained control with specific types, confidence, etc.\n\t// - nil: use default behavior (determined by __yao.needsearch agent)\n\tSearch any `json:\"search,omitempty\"` // Search mode: bool | SearchIntent | nil\n\n\t// Delegate: if provided, delegate to another agent immediately (skip LLM call)\n\t// This allows Create hook to route to sub-agents before any LLM processing\n\tDelegate *DelegateConfig `json:\"delegate,omitempty\"`\n}\n\n// NextHookPayload payload for the next hook\ntype NextHookPayload struct {\n\tMessages []Message `json:\"messages,omitempty\"` // Messages to be sent to the assistant\n\tCompletion *CompletionResponse `json:\"completion,omitempty\"` // Completion response from the completion hook\n\tTools []ToolCallResponse `json:\"tools,omitempty\"` // Tools results from the assistant\n\tError string `json:\"error,omitempty\"` // Error message if failed\n}\n\n// ToolCallResponse the response of a tool call\ntype ToolCallResponse struct {\n\tToolCallID string `json:\"toolcall_id\"`\n\tServer string `json:\"server\"`\n\tTool string `json:\"tool\"`\n\tArguments interface{} `json:\"arguments,omitempty\"`\n\tResult interface{} `json:\"result,omitempty\"`\n\tError string `json:\"error,omitempty\"`\n}\n\n// NextHookResponse represents the response from Next hook\ntype NextHookResponse struct {\n\t// Delegate: if provided, delegate to another agent (recursive call)\n\tDelegate *DelegateConfig `json:\"delegate,omitempty\"`\n\n\t// Data: custom response data to return to user\n\t// If both Delegate and Data are nil, use standard CompletionResponse\n\tData interface{} `json:\"data,omitempty\"`\n\n\t// Metadata: for debugging and logging\n\tMetadata map[string]interface{} `json:\"metadata,omitempty\"`\n}\n\n// DelegateConfig configuration for delegating to another agent\ntype DelegateConfig struct {\n\tAgentID string `json:\"agent_id\"` // Required: target agent ID\n\tMessages []Message `json:\"messages\"` // Messages to send to target agent\n\tOptions map[string]interface{} `json:\"options,omitempty\"` // Optional: call-level options for delegation\n}\n\n// NextAction defines the action determined by Next hook response\ntype NextAction string\n\nconst (\n\t// NextActionReturn returns data to user (standard or custom)\n\tNextActionReturn NextAction = \"return\"\n\n\t// NextActionDelegate delegates to another agent\n\tNextActionDelegate NextAction = \"delegate\"\n)\n\n// Action returns the determined action based on NextHookResponse fields\nfunc (n *NextHookResponse) Action() NextAction {\n\tif n.Delegate != nil {\n\t\treturn NextActionDelegate\n\t}\n\treturn NextActionReturn\n}\n\n// ResponseHookNext the response of the next hook\ntype ResponseHookNext interface{}\n\n// ResponseHookMCP the response of the mcp hook\ntype ResponseHookMCP struct{}\n\n// ResponseHookFailback the response of the failback hook\ntype ResponseHookFailback struct{}\n\n// HookInterruptedResponse the response of the interrupted hook\ntype HookInterruptedResponse struct {\n\t// Action to take after interrupt is handled\n\tAction InterruptAction `json:\"action\"` // continue, restart, or abort\n\n\t// Messages to use for next execution (if action is continue or restart)\n\tMessages []Message `json:\"messages,omitempty\"`\n\n\t// Context adjustments - allow hook to modify context fields\n\tAssistantID string `json:\"assistant_id,omitempty\"` // Override assistant ID\n\tConnector string `json:\"connector,omitempty\"` // Override connector\n\tLocale string `json:\"locale,omitempty\"` // Override locale\n\tTheme string `json:\"theme,omitempty\"` // Override theme\n\tRoute string `json:\"route,omitempty\"` // Override route\n\tMetadata map[string]interface{} `json:\"metadata,omitempty\"` // Override or merge metadata\n\n\t// Notice to send to client\n\tNotice string `json:\"notice,omitempty\"` // Message to display to user (e.g., \"Processing your new question...\")\n}\n\n// Message Structure ( OpenAI Chat Completion Input Message Structure, https://platform.openai.com/docs/api-reference/chat/create#chat/create-messages )\n// ===============================\n\n// MessageRole represents the role of a message author\ntype MessageRole string\n\n// Message role constants\nconst (\n\tRoleDeveloper MessageRole = \"developer\" // Developer-provided instructions (o1 models and newer)\n\tRoleSystem MessageRole = \"system\" // System instructions\n\tRoleUser MessageRole = \"user\" // User messages\n\tRoleAssistant MessageRole = \"assistant\" // Assistant responses\n\tRoleTool MessageRole = \"tool\" // Tool responses\n)\n\n// Message represents a message in the conversation, compatible with OpenAI's chat completion API\n// Supports message types: developer, system, user, assistant, and tool\ntype Message struct {\n\t// Common fields for all message types\n\tRole MessageRole `json:\"role\"` // Required: message author role\n\tContent interface{} `json:\"content,omitempty\"` // string or array of ContentPart; Required for most types, optional for assistant with tool_calls\n\tName *string `json:\"name,omitempty\"` // Optional: participant name to differentiate between participants of the same role\n\n\t// Tool message specific fields\n\tToolCallID *string `json:\"tool_call_id,omitempty\"` // Required for tool messages: tool call that this message is responding to\n\n\t// Assistant message specific fields\n\tToolCalls []ToolCall `json:\"tool_calls,omitempty\"` // Optional for assistant: tool calls generated by the model\n\tRefusal *string `json:\"refusal,omitempty\"` // Optional for assistant: refusal message (null when not refusing)\n}\n\n// ContentPartType represents the type of content part\ntype ContentPartType string\n\n// Content part type constants\nconst (\n\tContentText ContentPartType = \"text\" // Text content\n\tContentImageURL ContentPartType = \"image_url\" // Image URL content (Vision)\n\tContentInputAudio ContentPartType = \"input_audio\" // Input audio content (Audio)\n\tContentFile ContentPartType = \"file\" // File attachment (documents, etc.)\n\tContentData ContentPartType = \"data\" // Generic data content (base64, binary, etc.)\n)\n\n// ContentPart represents a part of the message content (for multimodal messages)\n// Used when Content is an array instead of a simple string\ntype ContentPart struct {\n\tType ContentPartType `json:\"type\"` // Required: content part type\n\tText string `json:\"text,omitempty\"` // For type=\"text\": the text content\n\tImageURL *ImageURL `json:\"image_url,omitempty\"` // For type=\"image_url\": the image URL\n\tInputAudio *InputAudio `json:\"input_audio,omitempty\"` // For type=\"input_audio\": the input audio data\n\tFile *FileAttachment `json:\"file,omitempty\"` // For type=\"file\": file attachment\n\tData *DataContent `json:\"data,omitempty\"` // For type=\"data\": generic data content\n}\n\n// ImageDetailLevel represents the detail level for image processing\ntype ImageDetailLevel string\n\n// Image detail level constants\nconst (\n\tDetailAuto ImageDetailLevel = \"auto\" // Let the model decide\n\tDetailLow ImageDetailLevel = \"low\" // Low detail (faster, cheaper)\n\tDetailHigh ImageDetailLevel = \"high\" // High detail (slower, more expensive)\n)\n\n// ImageURL represents an image URL in the message content\ntype ImageURL struct {\n\tURL string `json:\"url\"` // Required: URL of the image or base64 encoded image data\n\tDetail ImageDetailLevel `json:\"detail,omitempty\"` // Optional: how the model processes the image\n}\n\n// InputAudio represents input audio data in the message content\ntype InputAudio struct {\n\tData string `json:\"data\"` // Required: Base64 encoded audio data\n\tFormat string `json:\"format\"` // Required: Audio format (e.g., \"wav\", \"mp3\")\n}\n\n// FileAttachment represents a file attachment in the message content\n// Compatible with frontend InputArea format: { type: 'file', file: { url, filename } }\ntype FileAttachment struct {\n\tURL string `json:\"url\"` // Required: URL of the file (http:// or __uploader://fileid wrapper)\n\tFilename string `json:\"filename,omitempty\"` // Optional: original filename\n}\n\n// DataSourceType represents the type of data source\ntype DataSourceType string\n\n// Data source type constants\nconst (\n\tDataSourceModel DataSourceType = \"model\" // Data model\n\tDataSourceKBCollection DataSourceType = \"kb_collection\" // Knowledge base collection\n\tDataSourceKBDocument DataSourceType = \"kb_document\" // Knowledge base document/file\n\tDataSourceTable DataSourceType = \"table\" // Database table\n\tDataSourceAPI DataSourceType = \"api\" // API endpoint\n\tDataSourceMCPResource DataSourceType = \"mcp_resource\" // MCP (Model Context Protocol) resource\n)\n\n// DataSource represents a single data source reference\ntype DataSource struct {\n\tType DataSourceType `json:\"type\"` // Required: type of data source\n\tName string `json:\"name\"` // Required: name/identifier of the data source\n\tID string `json:\"id,omitempty\"` // Optional: specific ID (e.g., document ID, record ID)\n\tFilters map[string]interface{} `json:\"filters,omitempty\"` // Optional: filters to apply\n\tMetadata map[string]interface{} `json:\"metadata,omitempty\"` // Optional: additional metadata\n}\n\n// DataContent represents data source references in the message\n// Used to reference data models, knowledge base collections, KB documents, etc.\ntype DataContent struct {\n\tSources []DataSource `json:\"sources\"` // Required: array of data source references\n}\n\n// ToolCallType represents the type of tool call\ntype ToolCallType string\n\n// Tool call type constants\nconst (\n\tToolTypeFunction ToolCallType = \"function\" // Function call\n)\n\n// ToolCall represents a tool call generated by the model (for assistant messages)\ntype ToolCall struct {\n\tID string `json:\"id\"` // Required: unique identifier for the tool call\n\tType ToolCallType `json:\"type\"` // Required: type of tool call, currently only \"function\"\n\tFunction Function `json:\"function\"` // Required: function call details\n}\n\n// Function represents a function call with name and arguments\ntype Function struct {\n\tName string `json:\"name\"` // Required: name of the function to call\n\tArguments string `json:\"arguments,omitempty\"` // Optional: arguments to pass to the function, as a JSON string\n}\n\n// Completion Request Structure ( OpenAI Chat Completion Request, https://platform.openai.com/docs/api-reference/chat/create )\n// ===============================\n\n// CompletionRequest represents a chat completion request compatible with OpenAI's API\ntype CompletionRequest struct {\n\t// Required fields\n\tModel string `json:\"model\"` // Required: ID of the model to use\n\tMessages []Message `json:\"messages\"` // Required: list of messages comprising the conversation so far\n\n\t// Audio configuration (for models that support audio output)\n\tAudio *AudioConfig `json:\"audio,omitempty\"` // Optional: audio output configuration\n\n\t// Generation parameters\n\tTemperature *float64 `json:\"temperature,omitempty\"` // Optional: sampling temperature (0-2), defaults to 1\n\tMaxTokens *int `json:\"max_tokens,omitempty\"` // Optional: maximum number of tokens to generate (deprecated, use max_completion_tokens)\n\tMaxCompletionTokens *int `json:\"max_completion_tokens,omitempty\"` // Optional: maximum number of tokens that can be generated in the completion\n\n\t// Streaming configuration\n\tStream *bool `json:\"stream,omitempty\"` // Optional: if true, stream partial message deltas\n\tStreamOptions *StreamOptions `json:\"stream_options,omitempty\"` // Optional: options for streaming response\n\n\t// CUI Context information\n\tRoute string `json:\"route,omitempty\"` // Optional: route of the request for CUI context\n\tMetadata map[string]interface{} `json:\"metadata,omitempty\"` // Optional: metadata to pass to the page for CUI context\n\tSkip *Skip `json:\"skip,omitempty\"` // Optional: skip configuration (history, trace, etc.)\n}\n\n// AudioConfig represents the audio output configuration for models that support audio\ntype AudioConfig struct {\n\tVoice string `json:\"voice\"` // Required: voice to use for audio output (e.g., \"alloy\", \"echo\", \"fable\", \"onyx\", \"nova\", \"shimmer\")\n\tFormat string `json:\"format\"` // Required: audio output format (e.g., \"wav\", \"mp3\", \"flac\", \"opus\", \"pcm16\")\n}\n\n// StreamOptions represents options for streaming responses\ntype StreamOptions struct {\n\tIncludeUsage bool `json:\"include_usage,omitempty\"` // If true, include usage statistics in the final chunk\n}\n\n// MCPServerConfig represents an MCP server configuration\n// This mirrors agent/store/types.MCPServerConfig to avoid import cycles\ntype MCPServerConfig struct {\n\tServerID string `json:\"server_id\"` // MCP server ID (required)\n\tTools []string `json:\"tools,omitempty\"` // Tool name filter (empty = all tools)\n\tResources []string `json:\"resources,omitempty\"` // Resource URI filter (empty = all resources)\n}\n" }, { "path": "agent/context/types_llm.go", "content": "package context\n\nimport (\n\t\"github.com/yaoapp/gou/llm\"\n\t\"github.com/yaoapp/yao/agent/output/message\"\n)\n\n// Uses represents the wrapper configurations for assistant\n// Used to specify which assistant or MCP server to use for vision, audio, search, and fetch operations\ntype Uses struct {\n\tVision string `json:\"vision,omitempty\"` // Vision processing tool. Format: \"agent\" or \"mcp:server_id\"\n\tAudio string `json:\"audio,omitempty\"` // Audio processing tool. Format: \"agent\" or \"mcp:server_id\"\n\tSearch string `json:\"search,omitempty\"` // Search tool. Format: \"builtin\", \"disabled\", \"\", \"mcp:.\"\n\tFetch string `json:\"fetch,omitempty\"` // Fetch/retrieval tool. Format: \"agent\" or \"mcp:server_id\"\n\n\t// Search-related processing tools (NLP)\n\tWeb string `json:\"web,omitempty\"` // Web search handler: \"builtin\", \"\", \"mcp:.\"\n\tKeyword string `json:\"keyword,omitempty\"` // Keyword extraction: \"builtin\", \"\", \"mcp:.\"\n\tQueryDSL string `json:\"querydsl,omitempty\"` // QueryDSL generation: \"builtin\", \"\", \"mcp:.\"\n\tRerank string `json:\"rerank,omitempty\"` // Result reranking: \"builtin\", \"\", \"mcp:.\"\n}\n\n// VisionFormat specifies the vision input format\ntype VisionFormat string\n\n// Vision format constants define how image inputs are processed\nconst (\n\t// VisionFormatNone indicates no vision support\n\tVisionFormatNone VisionFormat = \"\"\n\t// VisionFormatOpenAI indicates OpenAI format (image_url with URL)\n\tVisionFormatOpenAI VisionFormat = \"openai\"\n\t// VisionFormatClaude indicates Claude/Anthropic format (image with base64)\n\tVisionFormatClaude VisionFormat = \"claude\"\n\t// VisionFormatBase64 forces base64 conversion (alias for claude)\n\tVisionFormatBase64 VisionFormat = \"base64\"\n\t// VisionFormatDefault enables auto-detection of format\n\tVisionFormatDefault VisionFormat = \"default\"\n)\n\n// GetVisionSupport returns whether vision is supported and the format\nfunc GetVisionSupport(cap *llm.Capabilities) (bool, VisionFormat) {\n\tif cap == nil || cap.Vision == nil {\n\t\treturn false, VisionFormatNone\n\t}\n\n\tswitch v := cap.Vision.(type) {\n\tcase bool:\n\t\t// Legacy bool format\n\t\treturn v, VisionFormatDefault\n\tcase string:\n\t\t// String format\n\t\tif v == \"\" || v == string(VisionFormatNone) {\n\t\t\treturn false, VisionFormatNone\n\t\t}\n\t\treturn true, VisionFormat(v)\n\tcase VisionFormat:\n\t\t// Direct VisionFormat type\n\t\tif v == VisionFormatNone || v == \"\" {\n\t\t\treturn false, VisionFormatNone\n\t\t}\n\t\treturn true, v\n\tdefault:\n\t\treturn false, VisionFormatNone\n\t}\n}\n\n// CompletionOptions the completion request options\n// These options are extracted from HookCreateResponse and Context, then passed to the LLM connector\n// Compatible with OpenAI Chat Completion API: https://platform.openai.com/docs/api-reference/chat/create\ntype CompletionOptions struct {\n\t// Model capabilities (used by LLM to select appropriate provider)\n\t// nil means capabilities are not specified/checked\n\tCapabilities *llm.Capabilities `json:\"capabilities,omitempty\"`\n\n\t// User-specified tools for vision, audio, search, and fetch processing\n\tUses *Uses `json:\"uses,omitempty\"`\n\n\t// ForceUses controls whether to force using Uses tools even when model has native capabilities\n\t// When true: Always use tools specified in Uses, ignore model's native multimodal capabilities\n\t// When false (default): Use model's native capabilities if available, fallback to Uses tools\n\t// This is useful when you want consistent behavior across different models or prefer specific tools\n\tForceUses bool `json:\"force_uses,omitempty\"`\n\n\t// Audio configuration (for models that support audio output)\n\tAudio *AudioConfig `json:\"audio,omitempty\"`\n\n\t// Generation parameters\n\tTemperature *float64 `json:\"temperature,omitempty\"` // Sampling temperature (0-2), defaults to 1\n\tMaxTokens *int `json:\"max_tokens,omitempty\"` // Maximum tokens to generate (deprecated, use MaxCompletionTokens)\n\tMaxCompletionTokens *int `json:\"max_completion_tokens,omitempty\"` // Maximum tokens in completion\n\tTopP *float64 `json:\"top_p,omitempty\"` // Nucleus sampling parameter (0-1), alternative to temperature\n\tN *int `json:\"n,omitempty\"` // Number of chat completion choices to generate\n\n\t// Control parameters\n\tStop interface{} `json:\"stop,omitempty\"` // Up to 4 sequences where the API will stop generating (string or []string)\n\tPresencePenalty *float64 `json:\"presence_penalty,omitempty\"` // Presence penalty (-2.0 to 2.0)\n\tFrequencyPenalty *float64 `json:\"frequency_penalty,omitempty\"` // Frequency penalty (-2.0 to 2.0)\n\tLogitBias map[string]float64 `json:\"logit_bias,omitempty\"` // Modify likelihood of specified tokens appearing\n\n\t// User and response format\n\tUser string `json:\"user,omitempty\"` // Unique identifier representing end-user\n\tResponseFormat *ResponseFormat `json:\"response_format,omitempty\"` // Format of the model's output\n\tSeed *int `json:\"seed,omitempty\"` // Seed for deterministic sampling\n\n\t// Tool calling\n\tTools []map[string]interface{} `json:\"tools,omitempty\"` // List of tools the model may call\n\tToolChoice interface{} `json:\"tool_choice,omitempty\"` // Controls which tool is called (\"none\", \"auto\", \"required\", or specific tool)\n\n\t// Streaming configuration\n\tStream *bool `json:\"stream,omitempty\"` // If true, stream partial message deltas\n\tStreamOptions *StreamOptions `json:\"stream_options,omitempty\"` // Options for streaming response\n\n\t// Reasoning configuration (for reasoning models like o1, GPT-5)\n\tReasoningEffort *string `json:\"reasoning_effort,omitempty\"` // Reasoning effort level: \"low\", \"medium\", \"high\" (o1 and GPT-5 only)\n\n\t// CUI Context information (from Context)\n\tRoute string `json:\"route,omitempty\"` // Route of the request for CUI context\n\tMetadata map[string]interface{} `json:\"metadata,omitempty\"` // Metadata to pass to the page for CUI context\n}\n\n// CompletionResponse represents the unified LLM completion response\n// This is Yao's internal representation that works with multiple LLM providers (OpenAI, Claude, DeepSeek, etc.)\ntype CompletionResponse struct {\n\t// Response metadata\n\tID string `json:\"id\"` // Unique identifier for the completion\n\tObject string `json:\"object\"` // Object type (e.g., \"chat.completion\")\n\tCreated int64 `json:\"created\"` // Unix timestamp of creation\n\tModel string `json:\"model\"` // Model used for completion\n\n\t// Response message (similar to OpenAI's message structure)\n\tRole string `json:\"role\"` // Role of the response, typically \"assistant\"\n\tContent interface{} `json:\"content,omitempty\"` // string (text) or []ContentPart (multimodal: text, image, audio)\n\n\t// Tool calls (when model calls functions/tools)\n\tToolCalls []ToolCall `json:\"tool_calls,omitempty\"` // Tool calls made by the model\n\n\t// Refusal (when model refuses to respond due to policy)\n\tRefusal string `json:\"refusal,omitempty\"` // Refusal message if model refused to answer\n\n\t// Reasoning content (for reasoning models like o1, DeepSeek R1)\n\tReasoningContent string `json:\"reasoning_content,omitempty\"` // Thinking/reasoning process\n\n\t// Completion metadata\n\tFinishReason string `json:\"finish_reason\"` // Why generation stopped (stop, length, tool_calls, content_filter, etc.)\n\n\t// Usage statistics\n\tUsage *message.UsageInfo `json:\"usage,omitempty\"` // Token usage statistics\n\n\t// Additional metadata\n\tSystemFingerprint string `json:\"system_fingerprint,omitempty\"` // System fingerprint for reproducibility\n\tMetadata map[string]interface{} `json:\"metadata,omitempty\"` // Additional provider-specific metadata\n\n\t// Raw response data (for debugging and special cases)\n\tRaw interface{} `json:\"raw,omitempty\"` // Original raw response from the LLM provider\n}\n\n// FinishReason constants - why the model stopped generating tokens\nconst (\n\tFinishReasonStop = \"stop\" // Natural stop point or provided stop sequence reached\n\tFinishReasonLength = \"length\" // Max tokens limit reached\n\tFinishReasonToolCalls = \"tool_calls\" // Model called a tool\n\tFinishReasonContentFilter = \"content_filter\" // Content filtered due to safety\n\tFinishReasonFunctionCall = \"function_call\" // Model called a function (deprecated, use tool_calls)\n)\n\n// ResponseFormat specifies the format of the model's output\n// Reference: https://platform.openai.com/docs/api-reference/chat/create#chat_create-response_format\ntype ResponseFormat struct {\n\tType ResponseFormatType `json:\"type\"` // Required: type of response format\n\tJSONSchema *JSONSchema `json:\"json_schema,omitempty\"` // Optional: for type=\"json_schema\", defines the schema\n}\n\n// ResponseFormatType represents the type of response format\ntype ResponseFormatType string\n\n// Response format type constants\nconst (\n\tResponseFormatText ResponseFormatType = \"text\" // Default text format\n\tResponseFormatJSON ResponseFormatType = \"json_object\" // JSON object format (no schema)\n\tResponseFormatJSONSchema ResponseFormatType = \"json_schema\" // JSON with strict schema validation\n)\n\n// JSONSchema defines a JSON schema for structured output\n// Used when ResponseFormat.Type is \"json_schema\"\ntype JSONSchema struct {\n\tName string `json:\"name\"` // Required: name of the schema\n\tDescription string `json:\"description,omitempty\"` // Optional: description of the schema\n\tSchema interface{} `json:\"schema\"` // Required: JSON schema (*jsonschema.Schema or map[string]interface{})\n\tStrict *bool `json:\"strict,omitempty\"` // Optional: whether to enforce strict schema validation (default: true)\n}\n" }, { "path": "agent/context/utils.go", "content": "package context\n\n// getValidatedValue gets value from query, header, or default, and validates it\nfunc getValidatedValue(queryValue, headerValue, defaultValue string, validator func(string) string) string {\n\tif queryValue != \"\" {\n\t\treturn validator(queryValue)\n\t}\n\tif headerValue != \"\" {\n\t\treturn validator(headerValue)\n\t}\n\treturn defaultValue\n}\n\n// getValidatedAccept gets Accept from query, header, or parse from client type\nfunc getValidatedAccept(queryValue, headerValue, clientType string) Accept {\n\tif queryValue != \"\" {\n\t\treturn validateAccept(queryValue)\n\t}\n\tif headerValue != \"\" {\n\t\treturn validateAccept(headerValue)\n\t}\n\treturn parseAccept(clientType)\n}\n\n// validateReferer validates and returns a valid Referer, returns RefererAPI if invalid\nfunc validateReferer(referer string) string {\n\tif ValidReferers[referer] {\n\t\treturn referer\n\t}\n\treturn RefererAPI\n}\n\n// validateAccept validates and returns a valid Accept type, returns AcceptStandard if invalid\nfunc validateAccept(accept string) Accept {\n\tif ValidAccepts[accept] {\n\t\treturn Accept(accept)\n\t}\n\treturn AcceptStandard\n}\n\n// parseAccept determines the accept type based on client type\nfunc parseAccept(clientType string) Accept {\n\tswitch clientType {\n\tcase \"web\":\n\t\treturn AcceptWebCUI\n\tcase \"android\", \"ios\":\n\t\treturn AccepNativeCUI\n\tcase \"windows\", \"macos\", \"linux\":\n\t\treturn AcceptDesktopCUI\n\tdefault:\n\t\treturn AcceptStandard\n\t}\n}\n" }, { "path": "agent/docs/configuration.md", "content": "# Assistant Configuration\n\n## Directory Structure\n\n```\nassistants/\n└── /\n ├── package.yao # Required: Configuration\n ├── prompts.yml # Optional: Default prompts\n ├── prompts/ # Optional: Prompt presets\n │ ├── chat.yml\n │ └── task.yml\n ├── locales/ # Optional: Translations\n │ ├── en-us.yml\n │ └── zh-cn.yml\n ├── src/ # Optional: Hook scripts\n │ └── index.ts\n └── mcps/ # Optional: MCP servers\n └── tools.mcp.yao\n```\n\n## package.yao\n\n### Basic Fields\n\n```json\n{\n \"name\": \"{{ name }}\",\n \"type\": \"assistant\",\n \"avatar\": \"/assets/avatar.png\",\n \"description\": \"{{ description }}\",\n \"connector\": \"gpt-4o\",\n \"tags\": [\"Category1\", \"Category2\"],\n \"sort\": 1\n}\n```\n\n| Field | Type | Description |\n| ------------- | -------- | ------------------------------------ |\n| `name` | string | Display name (supports i18n `{{ }}`) |\n| `type` | string | Type: `assistant` (default) |\n| `avatar` | string | Avatar image path |\n| `description` | string | Description (supports i18n) |\n| `connector` | string | LLM connector ID |\n| `tags` | string[] | Categorization tags |\n| `sort` | number | Display order |\n\n### Connector Options\n\n```json\n{\n \"connector\": \"gpt-4o\",\n \"connector_options\": {\n \"optional\": true,\n \"connectors\": [\"gpt-4o\", \"gpt-4o-mini\", \"claude-3\"],\n \"filters\": [\"tool_calls\", \"vision\"]\n }\n}\n```\n\n| Field | Type | Description |\n| ------------ | -------- | -------------------------------------------- |\n| `optional` | boolean | Allow user to select connector |\n| `connectors` | string[] | Available connectors (empty = all) |\n| `filters` | string[] | Required capabilities: `vision`, `audio`, `tool_calls`, `reasoning` |\n\n### Generation Options\n\n```json\n{\n \"options\": {\n \"temperature\": 0.7,\n \"max_tokens\": 4096\n }\n}\n```\n\n### Placeholder (UI Hints)\n\n```json\n{\n \"placeholder\": {\n \"title\": \"{{ chat.title }}\",\n \"description\": \"{{ chat.description }}\",\n \"prompts\": [\n \"{{ chat.prompts.0 }}\",\n \"{{ chat.prompts.1 }}\"\n ]\n }\n}\n```\n\n### Visibility & Access\n\n```json\n{\n \"public\": true,\n \"share\": \"team\",\n \"readonly\": true,\n \"built_in\": true,\n \"mentionable\": true,\n \"automated\": false\n}\n```\n\n| Field | Type | Description |\n| ------------- | ------- | --------------------------------- |\n| `public` | boolean | Visible to all users |\n| `share` | string | Sharing scope: `private`, `team` |\n| `readonly` | boolean | Prevent user modifications |\n| `built_in` | boolean | System-managed assistant |\n| `mentionable` | boolean | Can be @mentioned in chat |\n| `automated` | boolean | Can be triggered automatically |\n\n### Modes\n\n```json\n{\n \"modes\": [\"chat\", \"task\"],\n \"default_mode\": \"task\"\n}\n```\n\n### MCP Servers\n\n```json\n{\n \"mcp\": {\n \"servers\": [\n \"server-id\",\n { \"server_id\": \"tools\", \"tools\": [\"tool1\", \"tool2\"] },\n { \"server_id\": \"resources\", \"resources\": [\"uri://pattern\"] }\n ]\n }\n}\n```\n\n### Knowledge Base\n\n```json\n{\n \"kb\": {\n \"collections\": [\"collection-id-1\", \"collection-id-2\"]\n }\n}\n```\n\n### Database Models\n\n```json\n{\n \"db\": {\n \"models\": [\"model.name\", \"another.model\"]\n }\n}\n```\n\n### Uses (Wrapper Tools)\n\n```json\n{\n \"uses\": {\n \"vision\": \"vision-agent\",\n \"audio\": \"audio-agent\",\n \"search\": \"disabled\",\n \"fetch\": \"mcp:fetcher\"\n }\n}\n```\n\n| Field | Description |\n| -------- | -------------------------------------------------- |\n| `vision` | Vision processing: `` or `mcp:` |\n| `audio` | Audio processing: `` or `mcp:` |\n| `search` | Search: `disabled`, ``, or `mcp:`|\n| `fetch` | HTTP fetching: `` or `mcp:` |\n\n### Search Configuration\n\n```json\n{\n \"search\": {\n \"web\": {\n \"provider\": \"tavily\",\n \"max_results\": 10\n },\n \"kb\": {\n \"threshold\": 0.7,\n \"graph\": true\n },\n \"db\": {\n \"max_results\": 20\n },\n \"citation\": {\n \"format\": \"[{index}]\",\n \"auto_inject_prompt\": true\n }\n }\n}\n```\n\n## Environment Variables\n\nUse `$ENV.VAR_NAME` for sensitive values:\n\n```json\n{\n \"connector\": \"$ENV.LLM_CONNECTOR\"\n}\n```\n\n## Complete Example\n\n```json\n{\n \"name\": \"{{ name }}\",\n \"type\": \"assistant\",\n \"avatar\": \"/assets/assistant.png\",\n \"connector\": \"gpt-4o\",\n \"connector_options\": {\n \"optional\": true,\n \"connectors\": [\"gpt-4o\", \"gpt-4o-mini\"],\n \"filters\": [\"tool_calls\"]\n },\n \"mcp\": {\n \"servers\": [{ \"server_id\": \"tools\", \"tools\": [\"search\", \"calculate\"] }]\n },\n \"description\": \"{{ description }}\",\n \"options\": { \"temperature\": 0.7 },\n \"public\": true,\n \"placeholder\": {\n \"title\": \"{{ chat.title }}\",\n \"description\": \"{{ chat.description }}\",\n \"prompts\": [\"{{ chat.prompts.0 }}\", \"{{ chat.prompts.1 }}\"]\n },\n \"tags\": [\"Productivity\"],\n \"modes\": [\"chat\", \"task\"],\n \"default_mode\": \"chat\",\n \"sort\": 1,\n \"readonly\": true,\n \"mentionable\": true\n}\n```\n" }, { "path": "agent/docs/context-api.md", "content": "# Context API\n\nThe `ctx` object provides access to messaging, memory, tracing, and MCP operations.\n\n## Properties\n\n```typescript\ninterface Context {\n chat_id: string; // Chat session ID\n assistant_id: string; // Assistant ID\n locale: string; // User locale (e.g., \"en-us\")\n theme: string; // UI theme\n route: string; // Request route\n referer: string; // Request source\n metadata: Record; // Custom metadata\n authorized: Record; // Auth info\n\n memory: Memory; // Memory namespaces\n trace: Trace; // Tracing API\n mcp: MCP; // MCP operations\n search: Search; // Search API\n agent: Agent; // Agent-to-Agent calls (A2A)\n llm: LLM; // Direct LLM calls\n sandbox?: Sandbox; // Sandbox operations (optional)\n}\n```\n\n## Messaging\n\n### Send Complete Message\n\n```typescript\nctx.Send({ type: \"text\", props: { content: \"Hello!\" } });\nctx.Send(\"Hello!\"); // Shorthand for text\n```\n\n### Streaming Messages\n\n```typescript\nconst msgId = ctx.SendStream(\"Starting...\");\nctx.Append(msgId, \" processing...\");\nctx.Append(msgId, \" done!\");\nctx.End(msgId);\n```\n\n### Update Streaming Message\n\n```typescript\nconst msgId = ctx.SendStream({ type: \"loading\", props: { message: \"Loading...\" } });\n// ... do work ...\nctx.Replace(msgId, { type: \"text\", props: { content: \"Complete!\" } });\nctx.End(msgId);\n```\n\n### Merge Data\n\n```typescript\nconst msgId = ctx.SendStream({ type: \"status\", props: { progress: 0 } });\nctx.Merge(msgId, { progress: 50 }, \"props\");\nctx.Merge(msgId, { progress: 100, status: \"done\" }, \"props\");\nctx.End(msgId);\n```\n\n### Set Field\n\n```typescript\nconst msgId = ctx.SendStream({ type: \"result\", props: {} });\nctx.Set(msgId, \"success\", \"props.status\");\nctx.Set(msgId, { count: 10 }, \"props.data\");\nctx.End(msgId);\n```\n\n### Block Grouping\n\n```typescript\nconst blockId = ctx.BlockID();\nctx.Send(\"Step 1\", blockId);\nctx.Send(\"Step 2\", blockId);\nctx.Send(\"Step 3\", blockId);\nctx.EndBlock(blockId);\n```\n\n### ID Generators\n\n```typescript\nconst msgId = ctx.MessageID(); // \"M1\", \"M2\", ...\nconst blockId = ctx.BlockID(); // \"B1\", \"B2\", ...\nconst threadId = ctx.ThreadID(); // \"T1\", \"T2\", ...\n```\n\n## Memory\n\nFour-level hierarchical memory system:\n\n| Namespace | Scope | Persistence |\n| -------------------- | ------------ | ----------- |\n| `ctx.memory.user` | Per user | Persistent |\n| `ctx.memory.team` | Per team | Persistent |\n| `ctx.memory.chat` | Per chat | Persistent |\n| `ctx.memory.context` | Per request | Temporary |\n\n### Basic Operations\n\n```typescript\n// Get/Set\nctx.memory.user.Set(\"theme\", \"dark\");\nconst theme = ctx.memory.user.Get(\"theme\");\n\n// With TTL (seconds)\nctx.memory.context.Set(\"temp\", data, 300);\n\n// Check/Delete\nif (ctx.memory.chat.Has(\"topic\")) {\n ctx.memory.chat.Del(\"topic\");\n}\n\n// Get and delete atomically\nconst token = ctx.memory.context.GetDel(\"one_time_token\");\n\n// Collection operations\nconst keys = ctx.memory.user.Keys();\nconst count = ctx.memory.chat.Len();\nctx.memory.context.Clear();\n```\n\n### Counters\n\n```typescript\nconst views = ctx.memory.user.Incr(\"page_views\");\nconst credits = ctx.memory.user.Decr(\"credits\", 5);\n```\n\n### Lists\n\n```typescript\nctx.memory.chat.Push(\"history\", [msg1, msg2]);\nconst last = ctx.memory.chat.Pop(\"queue\");\nconst items = ctx.memory.chat.Pull(\"queue\", 5);\nconst all = ctx.memory.chat.PullAll(\"queue\");\n```\n\n### Sets\n\n```typescript\nctx.memory.user.AddToSet(\"visited\", [\"/home\", \"/about\"]);\n```\n\n### Array Access\n\n```typescript\nconst len = ctx.memory.chat.ArrayLen(\"messages\");\nconst first = ctx.memory.chat.ArrayGet(\"messages\", 0);\nconst last = ctx.memory.chat.ArrayGet(\"messages\", -1);\nctx.memory.chat.ArraySet(\"messages\", 0, newMsg);\nconst slice = ctx.memory.chat.ArraySlice(\"messages\", -10, -1);\nconst page = ctx.memory.chat.ArrayPage(\"messages\", 1, 20);\nconst all = ctx.memory.chat.ArrayAll(\"messages\");\n```\n\n## Trace\n\n### Create Nodes\n\n```typescript\nconst node = ctx.trace.Add(\n { query: \"input data\" },\n {\n label: \"Processing\",\n type: \"process\",\n icon: \"play\",\n description: \"Processing user request\"\n }\n);\n```\n\n### Logging\n\n```typescript\nctx.trace.Info(\"Starting process\");\nctx.trace.Debug(\"Variable: \" + value);\nctx.trace.Warn(\"Deprecated feature\");\nctx.trace.Error(\"Operation failed\");\n\n// Or on node\nnode.Info(\"Step completed\");\n```\n\n### Node Lifecycle\n\n```typescript\nnode.SetOutput({ result: data });\nnode.SetMetadata(\"duration\", 1500);\nnode.Complete({ status: \"done\" });\n// or\nnode.Fail(\"Error message\");\n```\n\n### Parallel Nodes\n\n```typescript\nconst nodes = ctx.trace.Parallel([\n { input: { url: \"api1\" }, option: { label: \"API 1\" } },\n { input: { url: \"api2\" }, option: { label: \"API 2\" } }\n]);\n```\n\n### Child Nodes\n\n```typescript\nconst parent = ctx.trace.Add({}, { label: \"Parent\" });\nconst child = parent.Add({}, { label: \"Child\" });\n```\n\n## MCP\n\n### Tools\n\n```typescript\n// List tools\nconst tools = ctx.mcp.ListTools(\"server-id\");\n\n// Call single tool - returns parsed result directly\nconst result = ctx.mcp.CallTool(\"server-id\", \"tool-name\", { arg: \"value\" });\nconsole.log(result.field); // Direct access to parsed data\n\n// Call multiple sequentially - returns array of parsed results\nconst results = ctx.mcp.CallTools(\"server-id\", [\n { name: \"tool1\", arguments: { a: 1 } },\n { name: \"tool2\", arguments: { b: 2 } }\n]);\nresults.forEach(r => console.log(r));\n\n// Call multiple in parallel - returns array of parsed results\nconst results = ctx.mcp.CallToolsParallel(\"server-id\", [\n { name: \"tool1\", arguments: {} },\n { name: \"tool2\", arguments: {} }\n]);\nresults.forEach(r => console.log(r));\n```\n\n### Cross-Server Tool Calls\n\n```typescript\n// Call tools across multiple MCP servers (like Promise.all)\nconst results = ctx.mcp.All([\n { mcp: \"server1\", tool: \"search\", arguments: { q: \"query\" } },\n { mcp: \"server2\", tool: \"fetch\", arguments: { id: 123 } }\n]);\n\n// First success wins (like Promise.any)\nconst results = ctx.mcp.Any([\n { mcp: \"primary\", tool: \"search\", arguments: { q: \"query\" } },\n { mcp: \"backup\", tool: \"search\", arguments: { q: \"query\" } }\n]);\n\n// First complete wins (like Promise.race)\nconst results = ctx.mcp.Race([\n { mcp: \"region-us\", tool: \"ping\", arguments: {} },\n { mcp: \"region-eu\", tool: \"ping\", arguments: {} }\n]);\n\n// Result structure\ninterface MCPToolResult {\n mcp: string; // Server ID\n tool: string; // Tool name\n result?: any; // Parsed result content\n error?: string; // Error if failed\n}\n```\n\n### Resources\n\n```typescript\nconst resources = ctx.mcp.ListResources(\"server-id\");\nconst data = ctx.mcp.ReadResource(\"server-id\", \"resource://uri\");\n```\n\n### Prompts\n\n```typescript\nconst prompts = ctx.mcp.ListPrompts(\"server-id\");\nconst prompt = ctx.mcp.GetPrompt(\"server-id\", \"prompt-name\", { arg: \"value\" });\n```\n\n## Search\n\n### Single Search\n\n```typescript\n// Web search\nconst webResult = ctx.search.Web(\"query\", {\n limit: 10,\n sites: [\"example.com\"],\n time_range: \"week\"\n});\n\n// Knowledge base\nconst kbResult = ctx.search.KB(\"query\", {\n collections: [\"docs\"],\n threshold: 0.7,\n graph: true\n});\n\n// Database\nconst dbResult = ctx.search.DB(\"query\", {\n models: [\"model.name\"],\n wheres: [{ column: \"status\", value: \"active\" }],\n limit: 20\n});\n```\n\n### Parallel Search\n\n```typescript\n// Wait for all\nconst results = ctx.search.All([\n { type: \"web\", query: \"topic\" },\n { type: \"kb\", query: \"topic\", collections: [\"docs\"] }\n]);\n\n// First success\nconst results = ctx.search.Any([\n { type: \"web\", query: \"topic\" },\n { type: \"kb\", query: \"topic\" }\n]);\n\n// First complete\nconst results = ctx.search.Race([\n { type: \"web\", query: \"topic\" },\n { type: \"kb\", query: \"topic\" }\n]);\n```\n\n### Result Structure\n\n```typescript\ninterface SearchResult {\n type: \"web\" | \"kb\" | \"db\";\n query: string;\n source: \"hook\" | \"auto\" | \"user\";\n items: {\n citation_id: string;\n title: string;\n url: string;\n content: string;\n score: number;\n }[];\n error?: string;\n}\n```\n\n## Agent API\n\nThe `ctx.agent` object provides methods to call other agents from within hooks, enabling agent-to-agent communication (A2A).\n\n### Single Agent Call\n\n```typescript\n// Basic call\nconst result = ctx.agent.Call(\"assistant-id\", messages);\n\n// With options and callback\nconst result = ctx.agent.Call(\"assistant-id\", messages, {\n connector: \"gpt-4o\",\n mode: \"chat\",\n metadata: { source: \"hook\" },\n skip: { history: false, trace: false, output: false },\n onChunk: (msg) => {\n console.log(\"Received:\", msg.type, msg.props);\n return 0; // 0 = continue, non-zero = stop\n }\n});\n```\n\n### Agent Options\n\n```typescript\ninterface AgentCallOptions {\n connector?: string; // Override LLM connector\n mode?: string; // Agent mode (\"chat\", \"task\")\n metadata?: Record; // Custom metadata passed to hooks\n skip?: {\n history?: boolean; // Skip loading chat history\n trace?: boolean; // Skip trace recording\n output?: boolean; // Skip output to client\n keyword?: boolean; // Skip keyword extraction\n search?: boolean; // Skip search\n content_parsing?: boolean; // Skip content parsing\n };\n onChunk?: (msg: Message) => number; // Callback (0=continue, non-zero=stop)\n}\n```\n\n### Parallel Agent Calls\n\n```typescript\n// Wait for all agents to complete (like Promise.all)\nconst results = ctx.agent.All([\n { agent: \"agent-1\", messages: [...] },\n { agent: \"agent-2\", messages: [...] }\n]);\n\n// Return first successful result (like Promise.any)\nconst results = ctx.agent.Any([\n { agent: \"agent-1\", messages: [...] },\n { agent: \"agent-2\", messages: [...] }\n]);\n\n// Return first completed result (like Promise.race)\nconst results = ctx.agent.Race([\n { agent: \"agent-1\", messages: [...] },\n { agent: \"agent-2\", messages: [...] }\n]);\n\n// With global callback for all responses\nconst results = ctx.agent.All([\n { agent: \"agent-1\", messages: [...] },\n { agent: \"agent-2\", messages: [...] }\n], {\n onChunk: (agentId, index, msg) => {\n console.log(`Agent ${agentId} [${index}]:`, msg.type);\n return 0;\n }\n});\n```\n\n### Result Structure\n\n```typescript\ninterface AgentResult {\n agent_id: string;\n response?: Response;\n content?: string;\n error?: string;\n}\n```\n\n### Message Object (onChunk callback)\n\n```typescript\ninterface Message {\n type: string; // \"text\", \"thinking\", \"tool_call\", \"error\"\n props?: Record; // e.g., { content: \"Hello\" }\n chunk_id?: string; // C1, C2, ...\n message_id?: string; // M1, M2, ...\n delta?: boolean; // Incremental update flag\n}\n```\n\n## Sandbox API\n\nThe `ctx.sandbox` object provides access to sandbox operations when the assistant is configured with a sandbox executor (e.g., Claude CLI). Only available when `sandbox` is configured in `package.yao`.\n\n### Properties\n\n```typescript\nctx.sandbox.workdir // Workspace directory path (e.g., \"/workspace\")\n```\n\n### File Operations\n\n```typescript\n// Read file\nconst content = ctx.sandbox.ReadFile(\"config.json\");\n\n// Write file\nctx.sandbox.WriteFile(\"output.txt\", \"Hello World\");\n\n// List directory\nconst files = ctx.sandbox.ListDir(\"src\");\nfiles.forEach(f => console.log(f.name, f.is_dir, f.size));\n```\n\n### Command Execution\n\n```typescript\n// Execute command (returns stdout)\nconst output = ctx.sandbox.Exec([\"npm\", \"test\"]);\n\n// Handle errors\ntry {\n ctx.sandbox.Exec([\"git\", \"commit\", \"-m\", \"fix\"]);\n} catch (e) {\n console.error(\"Command failed:\", e.message);\n}\n```\n\n### FileInfo Structure\n\n```typescript\ninterface FileInfo {\n name: string; // File/directory name\n size: number; // Size in bytes\n is_dir: boolean; // True if directory\n}\n```\n\n### Use Cases\n\n```typescript\n// Prepare workspace before execution\nfunction Create(ctx, messages) {\n if (ctx.sandbox) {\n ctx.sandbox.WriteFile(\"config.json\", JSON.stringify({ debug: true }));\n }\n return { messages };\n}\n\n// Post-process results\nfunction Next(ctx, payload) {\n if (ctx.sandbox && !payload.error) {\n const files = ctx.sandbox.ListDir(\"output\");\n return { data: { generated: files.map(f => f.name) } };\n }\n return null;\n}\n```\n\n## LLM API\n\nThe `ctx.llm` object provides direct access to LLM connectors for streaming completions.\n\n### Single LLM Call\n\n```typescript\n// Basic streaming call\nconst result = ctx.llm.Stream(\"gpt-4o\", [\n { role: \"user\", content: \"Hello\" }\n]);\n\n// With options and callback\nconst result = ctx.llm.Stream(\"gpt-4o\", messages, {\n temperature: 0.7,\n max_tokens: 2000,\n onChunk: (msg) => {\n console.log(\"Chunk:\", msg.props?.content);\n return 0;\n }\n});\n```\n\n### Parallel LLM Calls\n\n```typescript\n// Wait for all LLM calls (like Promise.all)\nconst results = ctx.llm.All([\n { connector: \"gpt-4o\", messages: [...] },\n { connector: \"claude-3\", messages: [...] }\n]);\n\n// Return first successful result (like Promise.any)\nconst results = ctx.llm.Any([\n { connector: \"gpt-4o\", messages: [...] },\n { connector: \"claude-3\", messages: [...] }\n]);\n\n// Return first completed result (like Promise.race)\nconst results = ctx.llm.Race([\n { connector: \"gpt-4o\", messages: [...] },\n { connector: \"claude-3\", messages: [...] }\n]);\n\n// With global callback\nconst results = ctx.llm.All([\n { connector: \"gpt-4o\", messages: [...] },\n { connector: \"claude-3\", messages: [...] }\n], {\n onChunk: (connectorId, index, msg) => {\n console.log(`LLM ${connectorId} [${index}]:`, msg.type);\n return 0;\n }\n});\n```\n\n### LLM Options\n\n```typescript\ninterface LlmOptions {\n temperature?: number;\n max_tokens?: number;\n max_completion_tokens?: number;\n top_p?: number;\n presence_penalty?: number;\n frequency_penalty?: number;\n stop?: string | string[];\n user?: string;\n seed?: number;\n tools?: object[];\n tool_choice?: string | object;\n response_format?: { type: string; json_schema?: object };\n reasoning_effort?: string;\n onChunk?: (msg: Message) => number;\n}\n```\n\n### Result Structure\n\n```typescript\ninterface LlmResult {\n connector: string;\n response?: CompletionResponse;\n content?: string;\n error?: string;\n}\n" }, { "path": "agent/docs/hooks.md", "content": "# Hooks\n\nHooks allow you to customize agent behavior at key points in the execution lifecycle.\n\n## Lifecycle\n\n```\nUser Input → Create Hook → LLM Call → Tool Execution → Next Hook → Response\n```\n\n## Create Hook\n\nCalled before LLM call. Use to preprocess messages, configure request, or delegate.\n\n```typescript\nfunction Create(ctx: agent.Context, messages: agent.Message[]): agent.Create {\n // Return null for default behavior\n return null;\n\n // Or return configuration\n return {\n messages, // Modified messages\n temperature: 0.7, // Override temperature\n max_tokens: 2000, // Override max tokens\n connector: \"gpt-4o-mini\", // Override connector\n prompt_preset: \"task\", // Select prompt preset\n disable_global_prompts: true,// Skip global prompts\n mcp_servers: [ // Add MCP servers\n { server_id: \"tools\", tools: [\"search\"] }\n ],\n uses: { // Override wrapper tools\n vision: \"vision-agent\",\n search: \"disabled\"\n },\n force_uses: true, // Force use wrapper tools\n locale: \"zh-cn\", // Override locale\n metadata: { key: \"value\" }, // Pass data to context\n };\n}\n```\n\n### Delegation (Skip LLM)\n\nRoute to another agent immediately:\n\n```typescript\nfunction Create(ctx: agent.Context, messages: agent.Message[]): agent.Create {\n if (shouldDelegate(messages)) {\n return {\n delegate: {\n agent_id: \"specialist.agent\",\n messages: messages,\n options: { metadata: { source: \"main\" } }\n }\n };\n }\n return { messages };\n}\n```\n\n## Next Hook\n\nCalled after LLM response and tool execution. Use to post-process or delegate.\n\n```typescript\nfunction Next(ctx: agent.Context, payload: agent.Payload): agent.Next {\n const { messages, completion, tools, error } = payload;\n\n // Handle errors\n if (error) {\n return { data: { status: \"error\", message: error } };\n }\n\n // Process tool results\n if (tools?.length > 0) {\n const results = tools.map(t => t.result);\n return { data: { status: \"success\", results } };\n }\n\n // Delegate based on response\n if (completion?.content?.includes(\"transfer\")) {\n return {\n delegate: {\n agent_id: \"transfer.agent\",\n messages: payload.messages\n }\n };\n }\n\n // Return null for standard response\n return null;\n}\n```\n\n### Payload Structure\n\n```typescript\ninterface Payload {\n messages: Message[]; // Messages sent to LLM\n completion?: {\n content: string; // LLM text response\n tool_calls?: ToolCall[]; // Tool calls from LLM\n usage?: UsageInfo; // Token usage\n };\n tools?: ToolCallResponse[]; // Tool execution results\n error?: string; // Error message\n}\n\ninterface ToolCallResponse {\n toolcall_id: string;\n server: string; // MCP server ID\n tool: string; // Tool name\n arguments?: any; // Tool arguments\n result?: any; // Tool result\n error?: string; // Tool error\n}\n```\n\n### Return Values\n\n```typescript\ninterface NextResponse {\n delegate?: { // Route to another agent\n agent_id: string;\n messages: Message[];\n options?: Record;\n };\n data?: any; // Custom response data\n metadata?: Record;// Debug metadata\n}\n```\n\n## Sending Messages\n\nUse `ctx` to send messages to the client:\n\n```typescript\nfunction Create(ctx: agent.Context, messages: agent.Message[]): agent.Create {\n // Send complete message\n ctx.Send({ type: \"text\", props: { content: \"Processing...\" } });\n\n // Streaming message\n const msgId = ctx.SendStream(\"Starting...\");\n ctx.Append(msgId, \" step 1...\");\n ctx.Append(msgId, \" step 2...\");\n ctx.End(msgId);\n\n return { messages };\n}\n```\n\n## Memory\n\nShare data between hooks:\n\n```typescript\nfunction Create(ctx: agent.Context, messages: agent.Message[]): agent.Create {\n // Store in request-scoped memory\n ctx.memory.context.Set(\"start_time\", Date.now());\n ctx.memory.context.Set(\"query\", messages[0]?.content);\n\n return { messages };\n}\n\nfunction Next(ctx: agent.Context, payload: agent.Payload): agent.Next {\n // Retrieve data\n const startTime = ctx.memory.context.Get(\"start_time\");\n const duration = Date.now() - startTime;\n\n return { data: { duration_ms: duration } };\n}\n```\n\n## Tracing\n\nAdd trace nodes for debugging and UI:\n\n```typescript\nfunction Create(ctx: agent.Context, messages: agent.Message[]): agent.Create {\n const node = ctx.trace.Add(\n { query: messages[0]?.content },\n { label: \"Preprocessing\", type: \"process\", icon: \"play\" }\n );\n\n node.Info(\"Starting analysis\");\n // ... processing ...\n node.Complete({ status: \"done\" });\n\n return { messages };\n}\n```\n\n## Error Handling\n\n```typescript\nfunction Next(ctx: agent.Context, payload: agent.Payload): agent.Next {\n try {\n if (payload.error) {\n ctx.trace.Error(payload.error);\n return {\n data: { status: \"error\", message: \"Something went wrong\" }\n };\n }\n // ... normal processing\n } catch (e) {\n ctx.trace.Error(e.message);\n return { data: { status: \"error\", message: e.message } };\n }\n}\n```\n\n## Multi-Agent Orchestration\n\n```typescript\n// Main agent delegates based on intent\nfunction Next(ctx: agent.Context, payload: agent.Payload): agent.Next {\n const { tools } = payload;\n\n // Route based on tool result\n const intent = tools?.[0]?.result?.intent;\n const agentMap = {\n \"search\": \"search.agent\",\n \"calculate\": \"calc.agent\",\n \"translate\": \"translate.agent\"\n };\n\n if (intent && agentMap[intent]) {\n return {\n delegate: {\n agent_id: agentMap[intent],\n messages: payload.messages\n }\n };\n }\n\n return null;\n}\n```\n\n## Complete Example\n\n```typescript\nimport { agent } from \"@yao/runtime\";\n\nfunction Create(ctx: agent.Context, messages: agent.Message[]): agent.Create {\n const query = messages[messages.length - 1]?.content || \"\";\n \n // Store for Next hook\n ctx.memory.context.Set(\"query\", query);\n ctx.memory.context.Set(\"start\", Date.now());\n\n // Add trace\n ctx.trace.Add({ query }, { label: \"Create\", type: \"hook\" });\n\n // Check if needs special handling\n if (query.toLowerCase().includes(\"urgent\")) {\n return {\n messages,\n temperature: 0,\n prompt_preset: \"task\"\n };\n }\n\n return { messages };\n}\n\nfunction Next(ctx: agent.Context, payload: agent.Payload): agent.Next {\n const { completion, tools, error } = payload;\n const start = ctx.memory.context.Get(\"start\");\n const duration = Date.now() - start;\n\n ctx.trace.Add(\n { duration },\n { label: \"Next\", type: \"hook\" }\n ).Complete();\n\n if (error) {\n return { data: { status: \"error\", error } };\n }\n\n if (tools?.length > 0) {\n return {\n data: {\n status: \"success\",\n response: completion?.content,\n tools: tools.map(t => ({ name: t.tool, result: t.result })),\n duration_ms: duration\n }\n };\n }\n\n return null;\n}\n```\n" }, { "path": "agent/docs/i18n.md", "content": "# Internationalization (i18n)\n\n## Locale Files\n\nCreate `locales/` directory in the assistant:\n\n```\nassistants/my-assistant/\n└── locales/\n ├── en-us.yml\n ├── zh-cn.yml\n └── ja.yml\n```\n\n## Locale File Format\n\n```yaml\n# locales/en-us.yml\nname: My Assistant\ndescription: A helpful AI assistant\n\nchat:\n title: New Chat\n description: How can I help you today?\n prompts:\n - What can you do?\n - Help me with a task\n - Tell me about yourself\n\nmessages:\n welcome: Welcome back!\n error: Something went wrong\n processing: Processing your request...\n```\n\n```yaml\n# locales/zh-cn.yml\nname: 我的助手\ndescription: 一个有帮助的AI助手\n\nchat:\n title: 新对话\n description: 今天我能帮您什么？\n prompts:\n - 你能做什么？\n - 帮我完成一个任务\n - 介绍一下你自己\n\nmessages:\n welcome: 欢迎回来！\n error: 出了点问题\n processing: 正在处理您的请求...\n```\n\n## Using Translations\n\n### In package.yao\n\nUse `{{ key }}` syntax:\n\n```json\n{\n \"name\": \"{{ name }}\",\n \"description\": \"{{ description }}\",\n \"placeholder\": {\n \"title\": \"{{ chat.title }}\",\n \"description\": \"{{ chat.description }}\",\n \"prompts\": [\n \"{{ chat.prompts.0 }}\",\n \"{{ chat.prompts.1 }}\",\n \"{{ chat.prompts.2 }}\"\n ]\n }\n}\n```\n\n### In Prompts\n\n```yaml\n- role: system\n content: |\n You are {{ name }}.\n {{ description }}\n\n Respond in the user's language.\n```\n\n## Locale Detection\n\nThe system detects locale from:\n\n1. Request header `Accept-Language`\n2. User preference (stored in memory)\n3. Default: `en-us`\n\n### Override in Hook\n\n```typescript\nfunction Create(ctx: agent.Context, messages: agent.Message[]): agent.Create {\n // Get user preference\n const userLocale = ctx.memory.user.Get(\"preferred_locale\");\n\n return {\n messages,\n locale: userLocale || \"en-us\"\n };\n}\n```\n\n## Global Translations\n\nDefine global translations in `agent/locales/`:\n\n```\nagent/\n└── locales/\n ├── en-us.yml\n └── zh-cn.yml\n```\n\nThese are available to all assistants via the `__global__` namespace.\n\n## Accessing Translations in Hooks\n\n```typescript\nfunction Create(ctx: agent.Context, messages: agent.Message[]): agent.Create {\n const locale = ctx.locale; // e.g., \"en-us\"\n\n // Use locale for custom logic\n if (locale.startsWith(\"zh\")) {\n return {\n messages,\n prompt_preset: \"chinese\"\n };\n }\n\n return { messages };\n}\n```\n\n## Nested Keys\n\nAccess nested values with dot notation:\n\n```yaml\n# locales/en-us.yml\nerrors:\n validation:\n required: This field is required\n invalid: Invalid value\n network:\n timeout: Connection timed out\n```\n\n```json\n{\n \"placeholder\": {\n \"title\": \"{{ errors.validation.required }}\"\n }\n}\n```\n\n## Fallback Behavior\n\nIf a translation key is not found:\n\n1. Try the requested locale\n2. Fall back to `en-us`\n3. Return the key itself if not found\n\n## Dynamic Locale Content\n\n```typescript\nfunction Create(ctx: agent.Context, messages: agent.Message[]): agent.Create {\n // Add locale-specific system message\n const localeGreeting = {\n \"en-us\": \"Hello! How can I help you?\",\n \"zh-cn\": \"你好！有什么可以帮您的？\",\n \"ja\": \"こんにちは！何かお手伝いできますか？\"\n };\n\n const greeting = localeGreeting[ctx.locale] || localeGreeting[\"en-us\"];\n\n return {\n messages: [\n { role: \"system\", content: `Greeting: ${greeting}` },\n ...messages\n ]\n };\n}\n```\n\n## Best Practices\n\n1. **Keep keys consistent** - Use the same keys across all locale files\n2. **Use nested structure** - Organize related translations together\n3. **Provide fallbacks** - Always have `en-us` as the base locale\n4. **Test all locales** - Verify translations render correctly\n5. **Use context variables** - Combine with `$CTX.locale` in prompts\n" }, { "path": "agent/docs/iframe.md", "content": "# Iframe Integration\n\nAgent Pages can be embedded in CUI via `/web/` routes. This document covers the iframe communication mechanism between embedded pages and the CUI host.\n\n## Route Mapping\n\nPages are accessible via:\n\n```\n/web//\n```\n\nExample:\n\n| Page File | URL |\n| -------------------------- | --------------------------------- |\n| `pages/index/index.html` | `/web/my-assistant/index` |\n| `pages/result/index.html` | `/web/my-assistant/result` |\n| `pages/report/detail.html` | `/web/my-assistant/report/detail` |\n\n## URL Parameters\n\nCUI automatically injects context via URL parameters:\n\n| Parameter | Value | Description |\n| ---------- | ---------------------- | ------------- |\n| `__theme` | `light` / `dark` | Current theme |\n| `__locale` | `en-us`, `zh-cn`, etc. | User locale |\n\n> **Note**: Authentication uses secure HTTP-only cookies, so `__token` parameter is not needed.\n\n**Usage in page URL:**\n\n```\n/web/my-assistant/result?theme=__theme&locale=__locale\n```\n\nCUI replaces `__theme`, `__locale` with actual values before loading.\n\n## Message Communication\n\n### Receiving Setup Message\n\nWhen the iframe loads, CUI sends a `setup` message:\n\n```typescript\n// In your page script\nwindow.addEventListener(\"message\", (e) => {\n if (e.data.type === \"setup\") {\n const { theme, locale } = e.data.message;\n // Apply theme, set locale\n document.documentElement.setAttribute(\"data-theme\", theme);\n }\n});\n```\n\n### Sending Actions to CUI\n\nPages can trigger CUI actions via `postMessage` using the unified Action system:\n\n```typescript\n// Send action to parent CUI\nwindow.parent.postMessage(\n {\n type: \"action\",\n message: {\n name: \"notify.success\",\n payload: { message: \"Operation completed\" },\n },\n },\n window.location.origin\n);\n```\n\n### Action Types\n\n#### Navigate\n\n| Action | Description | Payload |\n| --------------- | ------------------------------- | ------------------------------------------- |\n| `navigate` | Open page in sidebar or new tab | `{ route, title?, icon?, query?, target? }` |\n| `navigate.back` | Navigate back in history | - |\n\n**Navigate Payload:**\n\n| Field | Type | Required | Description |\n| -------- | ------------------------ | -------- | ----------------------------------------------- |\n| `route` | `string` | ✅ | Target route (`$dashboard/xxx`, `/xxx`, or URL) |\n| `title` | `string` | - | Page title (shows title bar with back button) |\n| `icon` | `string` | - | Tab icon (e.g., `material-folder`) |\n| `query` | `Record` | - | Query parameters |\n| `target` | `'_self'` \\| `'_blank'` | - | `_self` (sidebar) or `_blank` (new window) |\n\n#### Notify\n\n| Action | Description | Payload |\n| ---------------- | ------------------------- | ------------------------------------------ |\n| `notify.success` | Show success notification | `{ message, duration?, icon?, closable? }` |\n| `notify.error` | Show error notification | `{ message, duration?, icon?, closable? }` |\n| `notify.warning` | Show warning notification | `{ message, duration?, icon?, closable? }` |\n| `notify.info` | Show info notification | `{ message, duration?, icon?, closable? }` |\n\n#### App\n\n| Action | Description |\n| ----------------- | ------------------------ |\n| `app.menu.reload` | Refresh application menu |\n\n#### Modal\n\n| Action | Description |\n| ------------- | ----------------- |\n| `modal.open` | Open modal dialog |\n| `modal.close` | Close modal |\n\n#### Table\n\n| Action | Description |\n| --------------- | -------------------- |\n| `table.search` | Trigger table search |\n| `table.refresh` | Refresh table data |\n| `table.save` | Save table row |\n| `table.delete` | Delete table row(s) |\n\n#### Form\n\n| Action | Description |\n| ----------------- | --------------------- |\n| `form.find` | Load form data by ID |\n| `form.submit` | Submit form |\n| `form.reset` | Reset form |\n| `form.setFields` | Set form field values |\n| `form.fullscreen` | Toggle fullscreen |\n\n#### MCP (Client-side)\n\n| Action | Description |\n| ------------------- | ------------------ |\n| `mcp.tool.call` | Execute MCP tool |\n| `mcp.resource.read` | Read MCP resource |\n| `mcp.resource.list` | List MCP resources |\n| `mcp.prompt.get` | Get MCP prompt |\n| `mcp.prompt.list` | List MCP prompts |\n\n#### Event\n\n| Action | Description |\n| ------------ | ----------------- |\n| `event.emit` | Emit custom event |\n\n#### Confirm\n\n| Action | Description |\n| --------- | ------------------------ |\n| `confirm` | Show confirmation dialog |\n\n### Receiving Events from CUI\n\nCUI can send messages to iframe via `web/sendMessage` event:\n\n```typescript\n// In your page script\nwindow.addEventListener(\"message\", (e) => {\n const { type, message } = e.data;\n\n switch (type) {\n case \"setup\":\n // Initial setup with theme, locale\n break;\n case \"refresh\":\n // CUI requests page refresh\n location.reload();\n break;\n case \"data\":\n // CUI sends data update\n handleDataUpdate(message);\n break;\n }\n});\n```\n\n## Complete Example\n\n### Page HTML (pages/result/index.html)\n\n```html\n\n\n \n Result Page\n \n \n \n
\n \n\n```\n\n### Page Script (pages/result/result.ts)\n\n```typescript\nimport { $Backend, Component, EventData } from \"@yao/sui\";\n\nconst self = this as Component;\n\n// Helper: Send action to CUI parent\nconst sendAction = (name: string, payload?: any) => {\n try {\n window.parent.postMessage(\n { type: \"action\", message: { name, payload } },\n window.location.origin\n );\n } catch (err) {\n console.error(\"Failed to send action to parent:\", err);\n }\n};\n\n// Initialize message listener\nfunction init() {\n window.addEventListener(\"message\", (e) => {\n if (e.origin !== window.location.origin) return;\n\n const { type, message } = e.data;\n switch (type) {\n case \"setup\":\n // Apply theme, locale from CUI\n document.documentElement.setAttribute(\"data-theme\", message.theme);\n break;\n case \"update\":\n // Handle data updates from CUI\n console.log(\"Received update:\", message);\n break;\n }\n });\n\n // Make helper available globally\n (window as any).sendAction = sendAction;\n}\n\ninit();\n\n// Event handler: Show success notification\nself.HandleSuccess = (event: Event, data: EventData) => {\n sendAction(\"notify.success\", { message: data.message || \"Success!\" });\n};\n\n// Event handler: Navigate to page\nself.HandleNavigate = (event: Event, data: EventData) => {\n sendAction(\"navigate\", {\n route: data.path,\n title: data.title,\n });\n};\n\n// Event handler: Close sidebar\nself.HandleClose = () => {\n sendAction(\"event.emit\", { key: \"app/closeSidebar\", value: {} });\n};\n\n// Event handler: Call backend and display result\nself.HandleQuery = async (event: Event, data: EventData) => {\n try {\n const result = await $Backend().Call(\"Query\", data.id);\n console.log(result);\n } catch (error: any) {\n sendAction(\"notify.error\", { message: error.message });\n }\n};\n```\n\n## Triggering from Hooks\n\nOpen page in sidebar from agent hooks:\n\n```typescript\nfunction Next(ctx: agent.Context, payload: agent.Payload): agent.Next {\n // Open result page in sidebar\n ctx.Send({\n type: \"action\",\n props: {\n name: \"navigate\",\n payload: {\n route: `/agents/my-assistant/result`,\n title: \"Results\",\n query: { id: resultId },\n },\n },\n });\n\n return null;\n}\n```\n\nSee [Pages](pages.md) for more details on triggering pages from hooks.\n\n## Security Notes\n\n1. **Same-origin only**: Messages are only processed from same-origin iframes\n2. **Secure cookies**: Authentication uses HTTP-only cookies, no token in URL\n3. **Validate messages**: Always validate message structure before processing\n" }, { "path": "agent/docs/mcp.md", "content": "# MCP Integration\n\nModel Context Protocol (MCP) enables tool integration with external services.\n\n## Directory Structure\n\nAssistants can define their own namespaced MCP servers in the `mcps/` directory:\n\n```\nassistants/\n└── my-assistant/\n ├── package.yao\n └── mcps/\n ├── tools.mcp.yao # → agents.my-assistant.tools\n ├── calculator.mcp.yao # → agents.my-assistant.calculator\n └── mapping/\n └── tools/\n └── schemes/\n ├── search.in.yao\n └── search.out.yao\n```\n\nMCP servers are automatically loaded with `agents..` prefix.\n\n## Defining MCP Servers\n\nCreate `mcps/tools.mcp.yao` in the assistant directory:\n\n```json\n{\n \"label\": \"Tools\",\n \"description\": \"Custom tools for the assistant\",\n \"transport\": \"process\",\n \"tools\": {\n \"search\": \"scripts.tools.Search\",\n \"create\": \"models.data.Create\"\n }\n}\n```\n\n### Transport Types\n\n**Process (Yao Internal)**\n\nMap Yao Processes directly to MCP tools:\n\n```json\n{\n \"transport\": \"process\",\n \"tools\": {\n \"search\": \"models.data.Paginate\",\n \"create\": \"models.data.Create\"\n },\n \"resources\": {\n \"detail\": \"models.data.Find\"\n }\n}\n```\n\n**STDIO (Local Server)**\n\n```json\n{\n \"transport\": \"stdio\",\n \"command\": \"python\",\n \"arguments\": [\"mcp_server.py\"],\n \"env\": { \"API_KEY\": \"$ENV.API_KEY\" }\n}\n```\n\n**HTTP (REST API)**\n\n```json\n{\n \"transport\": \"http\",\n \"url\": \"https://mcp.example.com/api\",\n \"authorization_token\": \"$ENV.TOKEN\"\n}\n```\n\n**SSE (Server-Sent Events)**\n\n```json\n{\n \"transport\": \"sse\",\n \"url\": \"https://mcp.example.com/events\",\n \"authorization_token\": \"$ENV.TOKEN\"\n}\n```\n\n## Configuring in package.yao\n\n### All Tools\n\n```json\n{\n \"mcp\": {\n \"servers\": [\"tools\"]\n }\n}\n```\n\n### Specific Tools\n\n```json\n{\n \"mcp\": {\n \"servers\": [{ \"server_id\": \"tools\", \"tools\": [\"search\", \"calculate\"] }]\n }\n}\n```\n\n### With Resources\n\n```json\n{\n \"mcp\": {\n \"servers\": [\n {\n \"server_id\": \"data\",\n \"tools\": [\"query\"],\n \"resources\": [\"data://users/*\"]\n }\n ]\n }\n}\n```\n\n## Dynamic Configuration in Hooks\n\n```typescript\nfunction Create(ctx: agent.Context, messages: agent.Message[]): agent.Create {\n return {\n messages,\n mcp_servers: [\n { server_id: \"tools\", tools: [\"search\"] },\n { server_id: \"data\", resources: [\"data://reports\"] },\n ],\n };\n}\n```\n\n## Using MCP in Hooks\n\n### List Available Tools\n\n```typescript\nconst tools = ctx.mcp.ListTools(\"server-id\");\n// { tools: [{ name: \"search\", description: \"...\", inputSchema: {...} }] }\n```\n\n### Call Tool\n\n```typescript\n// Returns parsed result directly - no wrapper object\nconst result = ctx.mcp.CallTool(\"server-id\", \"search\", {\n query: \"example\",\n limit: 10,\n});\nconsole.log(result.items); // Direct access to parsed data\n```\n\n### Batch Tool Calls\n\n```typescript\n// Sequential - returns array of parsed results\nconst results = ctx.mcp.CallTools(\"server-id\", [\n { name: \"step1\", arguments: { input: \"a\" } },\n { name: \"step2\", arguments: { input: \"b\" } },\n]);\nresults.forEach(r => console.log(r));\n\n// Parallel - returns array of parsed results\nconst results = ctx.mcp.CallToolsParallel(\"server-id\", [\n { name: \"api1\", arguments: {} },\n { name: \"api2\", arguments: {} },\n]);\nresults.forEach(r => console.log(r));\n```\n\n### Read Resources\n\n```typescript\nconst resources = ctx.mcp.ListResources(\"server-id\");\nconst data = ctx.mcp.ReadResource(\"server-id\", \"data://users/123\");\n```\n\n### Get Prompts\n\n```typescript\nconst prompts = ctx.mcp.ListPrompts(\"server-id\");\nconst prompt = ctx.mcp.GetPrompt(\"server-id\", \"system\", { role: \"helper\" });\n```\n\n### Cross-Server Tool Calls\n\nCall tools across multiple MCP servers concurrently:\n\n```typescript\n// Wait for all (like Promise.all)\nconst results = ctx.mcp.All([\n { mcp: \"server1\", tool: \"search\", arguments: { q: \"query\" } },\n { mcp: \"server2\", tool: \"analyze\", arguments: { data: \"input\" } }\n]);\n\n// First success (like Promise.any) - good for fallback\nconst results = ctx.mcp.Any([\n { mcp: \"primary\", tool: \"fetch\", arguments: { id: 1 } },\n { mcp: \"backup\", tool: \"fetch\", arguments: { id: 1 } }\n]);\n\n// First complete (like Promise.race) - good for latency\nconst results = ctx.mcp.Race([\n { mcp: \"region-us\", tool: \"ping\", arguments: {} },\n { mcp: \"region-eu\", tool: \"ping\", arguments: {} }\n]);\n\n// Access results\nresults.forEach(r => {\n if (r.error) {\n console.log(`${r.mcp}/${r.tool} failed: ${r.error}`);\n } else {\n console.log(`${r.mcp}/${r.tool} result:`, r.result);\n }\n});\n```\n\n## Tool Schema Mapping\n\nDefine input schemas for process transport tools:\n\n```\nmcps/\n└── mapping/\n └── /\n └── schemes/\n ├── search.in.yao # Input schema\n └── search.out.yao # Output schema (optional)\n```\n\n**mapping/tools/schemes/search.in.yao**\n\n```json\n{\n \"type\": \"object\",\n \"description\": \"Search data\",\n \"properties\": {\n \"keyword\": { \"type\": \"string\" },\n \"page\": { \"type\": \"integer\" }\n },\n \"x-process-args\": [\":arguments\"]\n}\n```\n\nThe `x-process-args` maps MCP arguments to Yao Process parameters:\n\n- `\":arguments\"` - Pass entire arguments object\n- `\"$args.field\"` - Extract specific field\n\n### Schema with Nested Objects\n\n```json\n{\n \"type\": \"object\",\n \"description\": \"Extract structured data from input\",\n \"properties\": {\n \"intent\": {\n \"type\": \"string\",\n \"enum\": [\"query\", \"create\", \"update\"],\n \"description\": \"Operation intent\"\n },\n \"items\": {\n \"type\": \"array\",\n \"items\": {\n \"type\": \"object\",\n \"properties\": {\n \"name\": { \"type\": \"string\" },\n \"value\": { \"type\": \"number\" }\n },\n \"required\": [\"name\", \"value\"]\n }\n }\n },\n \"required\": [\"intent\"],\n \"x-process-args\": [\":arguments\"]\n}\n```\n\n## Using Assistant Models in MCP\n\nMCP tools can reference assistant's own models:\n\n**mcps/data.mcp.yao**\n\n```json\n{\n \"label\": \"Data Tools\",\n \"transport\": \"process\",\n \"tools\": {\n \"list_orders\": \"models.agents.my-assistant.order.Paginate\",\n \"get_order\": \"models.agents.my-assistant.order.Find\",\n \"create_order\": \"models.agents.my-assistant.order.Create\",\n \"custom_query\": \"agents.my-assistant.orders.Query\"\n }\n}\n```\n\nSee [Models](models.md) for defining assistant models.\n\n## Error Handling\n\n```typescript\nfunction Next(ctx: agent.Context, payload: agent.Payload): agent.Next {\n const { tools } = payload;\n\n if (tools) {\n for (const tool of tools) {\n if (tool.error) {\n ctx.trace.Error(`Tool ${tool.tool} failed: ${tool.error}`);\n // Handle error\n } else {\n // Process result\n console.log(tool.result);\n }\n }\n }\n\n return null;\n}\n```\n\n## Complete Example\n\n**mcps/calculator.mcp.yao**\n\n```json\n{\n \"label\": \"Calculator\",\n \"description\": \"Math operations\",\n \"transport\": \"process\",\n \"tools\": {\n \"add\": \"scripts.math.Add\",\n \"multiply\": \"scripts.math.Multiply\"\n }\n}\n```\n\n**package.yao**\n\n```json\n{\n \"name\": \"Math Assistant\",\n \"connector\": \"gpt-4o\",\n \"mcp\": {\n \"servers\": [{ \"server_id\": \"calculator\", \"tools\": [\"add\", \"multiply\"] }]\n }\n}\n```\n\n**src/index.ts**\n\n```typescript\nfunction Create(ctx: agent.Context, messages: agent.Message[]): agent.Create {\n // Check if calculation is needed\n const query = messages[messages.length - 1]?.content || \"\";\n if (/\\d+\\s*[\\+\\-\\*\\/]\\s*\\d+/.test(query)) {\n // Enable calculator\n return {\n messages,\n mcp_servers: [{ server_id: \"calculator\" }],\n };\n }\n return { messages };\n}\n\nfunction Next(ctx: agent.Context, payload: agent.Payload): agent.Next {\n const { tools } = payload;\n\n if (tools?.length > 0) {\n const calcResult = tools.find((t) => t.server === \"calculator\");\n if (calcResult?.result) {\n return {\n data: {\n answer: calcResult.result,\n expression: calcResult.arguments,\n },\n };\n }\n }\n\n return null;\n}\n```\n" }, { "path": "agent/docs/models.md", "content": "# Assistant Models\n\nAssistants can define their own namespaced data models in the `models/` directory. These models are automatically loaded with the `agents..` prefix and use isolated database tables.\n\n## Directory Structure\n\n```\nassistants/\n└── my-assistant/\n ├── package.yao\n └── models/\n ├── order.mod.yao # → agents.my-assistant.order\n ├── item.mod.yao # → agents.my-assistant.item\n └── nested/\n └── log.mod.yao # → agents.my-assistant.nested.log\n```\n\n## Model Definition\n\nStandard Yao model definition with automatic table prefixing.\n\n**models/order.mod.yao**\n\n```json\n{\n \"name\": \"Order\",\n \"label\": \"Order Record\",\n \"description\": \"Customer orders\",\n \"table\": {\n \"name\": \"order\",\n \"comment\": \"Order records\"\n },\n \"columns\": [\n {\n \"name\": \"id\",\n \"type\": \"ID\",\n \"label\": \"ID\",\n \"primary\": true\n },\n {\n \"name\": \"order_no\",\n \"type\": \"string\",\n \"label\": \"Order Number\",\n \"length\": 100,\n \"nullable\": false,\n \"unique\": true,\n \"index\": true\n },\n {\n \"name\": \"customer_id\",\n \"type\": \"string\",\n \"label\": \"Customer ID\",\n \"length\": 255,\n \"nullable\": false,\n \"index\": true\n },\n {\n \"name\": \"total_amount\",\n \"type\": \"decimal\",\n \"label\": \"Total Amount\",\n \"precision\": 15,\n \"scale\": 2,\n \"nullable\": false\n },\n {\n \"name\": \"status\",\n \"type\": \"enum\",\n \"label\": \"Status\",\n \"option\": [\"pending\", \"confirmed\", \"shipped\", \"completed\", \"cancelled\"],\n \"default\": \"pending\",\n \"nullable\": false,\n \"index\": true\n },\n {\n \"name\": \"metadata\",\n \"type\": \"json\",\n \"label\": \"Metadata\",\n \"nullable\": true\n }\n ],\n \"relations\": {\n \"items\": {\n \"type\": \"hasMany\",\n \"model\": \"item\",\n \"key\": \"order_id\",\n \"foreign\": \"id\"\n }\n },\n \"indexes\": [\n {\n \"name\": \"idx_customer_status\",\n \"columns\": [\"customer_id\", \"status\"],\n \"type\": \"index\"\n }\n ],\n \"option\": {\n \"timestamps\": true,\n \"soft_deletes\": true\n }\n}\n```\n\n## Table Naming\n\nTables are automatically prefixed with `agents__`:\n\n| Assistant ID | Model File | Model ID | Table Name |\n| ------------ | ---------------------- | ------------------------ | ------------------------ |\n| `expense` | `models/order.mod.yao` | `agents.expense.order` | `agents_expense_order` |\n| `tests.demo` | `models/user.mod.yao` | `agents.tests.demo.user` | `agents_tests_demo_user` |\n\n## Using Models\n\n### In Hooks\n\n```typescript\nimport { Process } from \"@yao/runtime\";\n\nfunction Create(ctx: agent.Context, messages: agent.Message[]): agent.Create {\n // Query assistant's own model\n const orders = Process(\"models.agents.my-assistant.order.Paginate\", {\n wheres: [{ column: \"status\", value: \"pending\" }],\n limit: 10,\n });\n\n return { messages };\n}\n```\n\n### In MCP Tools\n\n```json\n{\n \"transport\": \"process\",\n \"tools\": {\n \"list_orders\": \"models.agents.my-assistant.order.Paginate\",\n \"get_order\": \"models.agents.my-assistant.order.Find\",\n \"create_order\": \"models.agents.my-assistant.order.Create\",\n \"update_order\": \"models.agents.my-assistant.order.Update\"\n }\n}\n```\n\n### In Scripts\n\n**src/orders.ts**\n\n```typescript\nimport { Process } from \"@yao/runtime\";\n\nexport function ListPending(): any[] {\n return Process(\"models.agents.my-assistant.order.Get\", {\n wheres: [{ column: \"status\", value: \"pending\" }],\n orders: [{ column: \"created_at\", option: \"desc\" }],\n });\n}\n\nexport function CreateOrder(data: any): any {\n return Process(\"models.agents.my-assistant.order.Create\", data);\n}\n\nexport function UpdateStatus(id: number, status: string): any {\n return Process(\"models.agents.my-assistant.order.Update\", id, { status });\n}\n```\n\n## Column Types\n\n| Type | Description | Options |\n| ------------ | -------------------------- | ----------------------- |\n| `ID` | Auto-increment primary key | `primary: true` |\n| `string` | VARCHAR | `length` (default: 255) |\n| `text` | TEXT | - |\n| `integer` | INT | - |\n| `bigInteger` | BIGINT | - |\n| `float` | FLOAT | `precision`, `scale` |\n| `decimal` | DECIMAL | `precision`, `scale` |\n| `boolean` | BOOLEAN | - |\n| `date` | DATE | - |\n| `datetime` | DATETIME | - |\n| `timestamp` | TIMESTAMP | - |\n| `json` | JSON/JSONB | - |\n| `enum` | ENUM | `option: [...]` |\n\n## Column Options\n\n| Option | Type | Description |\n| ----------- | --------- | ----------------- |\n| `nullable` | `boolean` | Allow NULL values |\n| `default` | `any` | Default value |\n| `unique` | `boolean` | Unique constraint |\n| `index` | `boolean` | Create index |\n| `primary` | `boolean` | Primary key |\n| `length` | `integer` | String length |\n| `precision` | `integer` | Decimal precision |\n| `scale` | `integer` | Decimal scale |\n| `comment` | `string` | Column comment |\n\n## Relations\n\n```json\n{\n \"relations\": {\n \"items\": {\n \"type\": \"hasMany\",\n \"model\": \"item\",\n \"key\": \"order_id\",\n \"foreign\": \"id\"\n },\n \"customer\": {\n \"type\": \"hasOne\",\n \"model\": \"customer\",\n \"key\": \"id\",\n \"foreign\": \"customer_id\"\n }\n }\n}\n```\n\n| Type | Description |\n| ---------------- | ----------------------------- |\n| `hasOne` | One-to-one relationship |\n| `hasMany` | One-to-many relationship |\n| `hasOneThrough` | Has one through intermediate |\n| `hasManyThrough` | Has many through intermediate |\n\n## Model Options\n\n```json\n{\n \"option\": {\n \"timestamps\": true,\n \"soft_deletes\": true,\n \"permission\": true\n }\n}\n```\n\n| Option | Description |\n| -------------- | -------------------------------------- |\n| `timestamps` | Add `created_at`, `updated_at` columns |\n| `soft_deletes` | Add `deleted_at` for soft delete |\n| `permission` | Enable permission checks |\n\n## Process Reference\n\nCommon model processes:\n\n| Process | Description | Arguments |\n| ------------- | ---------------- | --------------------------- |\n| `Find` | Get by ID | `id`, `query?` |\n| `Get` | Get records | `query` |\n| `Paginate` | Paginated list | `query`, `page`, `pagesize` |\n| `Create` | Create record | `data` |\n| `Update` | Update record | `id`, `data` |\n| `Save` | Create or update | `data` |\n| `Delete` | Delete record | `id` |\n| `Destroy` | Hard delete | `id` |\n| `Insert` | Batch insert | `columns`, `rows` |\n| `UpdateWhere` | Batch update | `query`, `data` |\n| `DeleteWhere` | Batch delete | `query` |\n\n## Migration\n\nModels are automatically migrated when Yao starts. The migration:\n\n1. Creates tables if not exist\n2. Adds new columns\n3. Creates indexes\n4. Does NOT drop columns (safe migration)\n\nTo force schema sync:\n\n```bash\nyao migrate --reset # Warning: drops and recreates tables\n```\n\n## Example: Complete Assistant with Models\n\n**assistants/inventory/package.yao**\n\n```json\n{\n \"name\": \"Inventory Assistant\",\n \"connector\": \"gpt-4o\",\n \"mcp\": {\n \"servers\": [{ \"server_id\": \"inventory\" }]\n }\n}\n```\n\n**assistants/inventory/models/product.mod.yao**\n\n```json\n{\n \"name\": \"Product\",\n \"table\": { \"name\": \"product\" },\n \"columns\": [\n { \"name\": \"id\", \"type\": \"ID\", \"primary\": true },\n { \"name\": \"sku\", \"type\": \"string\", \"length\": 50, \"unique\": true },\n { \"name\": \"name\", \"type\": \"string\", \"length\": 200 },\n { \"name\": \"quantity\", \"type\": \"integer\", \"default\": 0 },\n { \"name\": \"price\", \"type\": \"decimal\", \"precision\": 10, \"scale\": 2 }\n ],\n \"option\": { \"timestamps\": true }\n}\n```\n\n**assistants/inventory/mcps/inventory.mcp.yao**\n\n```json\n{\n \"label\": \"Inventory\",\n \"transport\": \"process\",\n \"tools\": {\n \"list_products\": \"models.agents.inventory.product.Paginate\",\n \"get_product\": \"models.agents.inventory.product.Find\",\n \"update_stock\": \"agents.inventory.stock.Update\"\n }\n}\n```\n\n**assistants/inventory/src/stock.ts**\n\n```typescript\nimport { Process } from \"@yao/runtime\";\n\nexport function Update(args: { sku: string; quantity: number }): any {\n const product = Process(\"models.agents.inventory.product.Get\", {\n wheres: [{ column: \"sku\", value: args.sku }],\n limit: 1,\n });\n\n if (!product || product.length === 0) {\n throw new Error(`Product not found: ${args.sku}`);\n }\n\n return Process(\"models.agents.inventory.product.Update\", product[0].id, {\n quantity: args.quantity,\n });\n}\n```\n" }, { "path": "agent/docs/pages.md", "content": "# Agent Pages\n\nAgent Pages provide a built-in SUI (Simple User Interface) framework for building web interfaces for AI agents. Pages are automatically loaded from the `/agent/template/` directory for global templates and `/assistants//pages/` for individual assistant pages.\n\n## Directory Structure\n\n```\n/\n├── agent/\n│ └── template/ # Global template directory\n│ ├── __document.html # Document template\n│ ├── __data.json # Global data\n│ ├── __assets/ # Global assets\n│ │ ├── css/\n│ │ ├── js/\n│ │ └── images/\n│ ├── pages/ # Global pages (login, error, etc.)\n│ │ └── login/\n│ │ └── login.html\n│ └── __locales/ # Internationalization\n│\n└── assistants/\n └── my-assistant/\n ├── package.yao\n └── pages/ # Assistant-specific pages\n ├── index/\n │ ├── index.html\n │ ├── index.css\n │ ├── index.ts\n │ └── index.backend.ts\n └── __assets/ # Optional assistant assets\n```\n\n## Route Mapping\n\n| File Path | Public URL |\n| ----------------------------------------- | -------------------- |\n| `/agent/template/pages/login/login.html` | `/agents/login` |\n| `/assistants/demo/pages/index/index.html` | `/agents/demo/index` |\n| `/assistants/demo/pages/chat/chat.html` | `/agents/demo/chat` |\n\n## Quick Start\n\n### 1. Create Document Template\n\n**`/agent/template/__document.html`**:\n\n```html\n\n\n \n \n {{ $global.title }}\n \n \n \n \n
{{ __page }}
\n \n\n```\n\n### 2. Create Global Data\n\n**`/agent/template/__data.json`**:\n\n```json\n{\n \"title\": \"AI Agent\",\n \"version\": \"1.0.0\"\n}\n```\n\n### 3. Create a Page\n\n**`/assistants/my-assistant/pages/index/index.html`**:\n\n```html\n
\n

{{ title }}

\n
\n
{{ msg.content }}
\n
\n \n
\n```\n\n**`/assistants/my-assistant/pages/index/index.json`**:\n\n```json\n{\n \"title\": \"Chat\",\n \"messages\": []\n}\n```\n\n**`/assistants/my-assistant/pages/index/index.css`**:\n\n```css\n.page {\n max-width: 800px;\n margin: 0 auto;\n padding: 24px;\n}\n\n.messages {\n display: flex;\n flex-direction: column;\n gap: 12px;\n}\n\n.message.user {\n align-self: flex-end;\n background: #007bff;\n color: white;\n}\n\n.message.assistant {\n align-self: flex-start;\n background: #f0f0f0;\n}\n```\n\n### 4. Add Backend Script\n\n**`/assistants/my-assistant/pages/index/index.backend.ts`**:\n\n```typescript\nfunction BeforeRender(request: Request): Record {\n const chatId = request.query.chat_id;\n return {\n messages: chatId ? Process(\"scripts.chat.GetHistory\", chatId) : [],\n user: request.authorized?.user_id,\n };\n}\n\nfunction ApiGetData(request: Request): any {\n const { id } = request.payload;\n return Process(\"models.data.Find\", id, {});\n}\n```\n\n### 5. Add Frontend Script\n\n**`/assistants/my-assistant/pages/index/index.ts`**:\n\nFrontend scripts can be written in two styles:\n\n**Style 1: Direct Code (Simple Pages)**\n\n```typescript\n// Runs immediately when script loads\ndocument.addEventListener(\"DOMContentLoaded\", () => {\n const form = document.querySelector(\"#myForm\") as HTMLFormElement;\n\n form.addEventListener(\"submit\", async (e) => {\n e.preventDefault();\n // Handle form submission\n });\n});\n\n// Smooth scrolling for navigation\ndocument.querySelectorAll('a[href^=\"#\"]').forEach((anchor) => {\n anchor.addEventListener(\"click\", function (e) {\n e.preventDefault();\n const target = document.querySelector(this.getAttribute(\"href\"));\n target?.scrollIntoView({ behavior: \"smooth\" });\n });\n});\n```\n\n**Style 2: Component Pattern (Interactive Pages)**\n\n```typescript\nimport { $Backend, Component, EventData } from \"@yao/sui\";\n\nconst self = this as Component;\n\n// Event handler bound to s:on-click=\"HandleClick\"\nself.HandleClick = async (event: Event, data: EventData) => {\n const result = await $Backend().Call(\"GetData\", data.id);\n console.log(result);\n};\n\n// Form submission handler\nself.HandleSubmit = async (event: Event) => {\n event.preventDefault();\n const form = event.target as HTMLFormElement;\n const formData = new FormData(form);\n await $Backend().Call(\"Submit\", Object.fromEntries(formData));\n};\n```\n\n**Using Backend API:**\n\n```typescript\nimport { $Backend, Yao } from \"@yao/sui\";\n\n// Call backend method\nconst data = await $Backend().Call(\"MethodName\", arg1, arg2);\n\n// Direct API calls\nconst yao = new Yao();\nconst res = await yao.Get(\"/api/endpoint\", { param: \"value\" });\nawait yao.Post(\"/api/endpoint\", { data: \"value\" });\n```\n\n### 6. Build and Run\n\n```bash\n# Build pages\nyao sui build agent\n\n# Or watch for changes\nyao sui watch agent\n\n# Start server\nyao start\n```\n\nAccess at: `http://localhost:5099/agents/my-assistant/index`\n\n## Template Syntax\n\n### Data Binding\n\n```html\n\n

{{ title }}

\n\n\n

{{ user.name }}

\n\n\n

{{ description || \"No description\" }}

\n```\n\n### Conditionals\n\n```html\n
Welcome, {{ user.name }}!
\n
Welcome, Guest!
\n
Please log in
\n```\n\n### Loops\n\n```html\n
    \n
  • \n {{ i + 1 }}. {{ item.name }}\n
    • \n
\n```\n\n### Events\n\n```html\n\n\n```\n\n### Components\n\nPages can use other pages as components:\n\n```html\n\n\n\n
\n
\n
Content
\n
\n
\n```\n\n## Built-in Variables\n\n| Variable | Description |\n| ---------- | ------------------------------------------- |\n| `$global` | Global data from `__data.json` |\n| `$query` | URL query parameters |\n| `$param` | URL path parameters |\n| `$payload` | POST request body |\n| `$cookie` | Cookie values |\n| `$url` | Current URL info |\n| `$theme` | Current theme |\n| `$locale` | Current locale |\n| `$auth` | OAuth authorization info (if authenticated) |\n\n## Page Configuration\n\nCreate `.config` for page settings:\n\n```json\n{\n \"title\": \"Page Title\",\n \"guard\": \"bearer-jwt\",\n \"cache\": 3600,\n \"data\": {\n \"key\": \"value\"\n }\n}\n```\n\n## Asset Paths\n\n- **Global assets**: `/agents/assets/...` → `/agent/template/__assets/...`\n- **Assistant assets**: `/agents//assets/...` → `/assistants//pages/__assets/...`\n\n## Build Output\n\n```\n/public/agents/\n├── assets/\n│ ├── libsui.min.js # SUI frontend SDK\n│ ├── css/ # Global CSS\n│ ├── js/ # Global JS\n│ └── images/ # Global images\n│\n├── login.sui # Global page\n├── login.cfg\n│\n└── my-assistant/\n ├── index.sui # Assistant page\n └── index.cfg\n```\n\n## Authentication\n\nPages default to public access. To require authentication:\n\n**`/assistants/my-assistant/pages/dashboard/dashboard.config`**:\n\n```json\n{\n \"guard\": \"bearer-jwt\"\n}\n```\n\nAvailable guards:\n\n| Guard | Description |\n| ------------ | --------------------------------- |\n| `-` | No authentication (default) |\n| `bearer-jwt` | JWT token in Authorization header |\n| `cookie-jwt` | JWT token in cookie |\n| `oauth` | OAuth 2.0 authentication |\n\n## Triggering Pages from Hooks\n\nUse `action` messages to open pages in the sidebar during conversation:\n\n```typescript\n// Navigate to a page in sidebar\nctx.Send({\n type: \"action\",\n props: {\n name: \"navigate\",\n payload: {\n route: \"/agents/my-assistant/result\", // Page route\n title: \"Query Results\", // Sidebar title\n query: { id: \"123\" }, // Passed as $query in page\n },\n },\n});\n\n// Open in new tab\nctx.Send({\n type: \"action\",\n props: {\n name: \"navigate\",\n payload: {\n route: \"/agents/my-assistant/detail\",\n target: \"_blank\",\n },\n },\n});\n```\n\n### Action Reference\n\n#### Navigate\n\nOpen a route in the sidebar or new window.\n\n**Payload:**\n\n| Field | Type | Required | Description |\n| -------- | ------------------------ | -------- | ---------------------------------------------------- |\n| `route` | `string` | ✅ | Target route or URL |\n| `title` | `string` | - | Page title (shows custom title bar with back button) |\n| `icon` | `string` | - | Tab icon (e.g., `material-folder`) |\n| `query` | `Record` | - | Query parameters (passed as `$query` in page) |\n| `target` | `'_self'` \\| `'_blank'` | - | `_self` (sidebar, default) or `_blank` (new window) |\n\n**Route Types:**\n\n| Prefix | Type | Description |\n| ----------------- | -------- | ----------------------------------------------- |\n| `$dashboard/` | CUI Page | Dashboard pages (e.g., `$dashboard/kb` → `/kb`) |\n| `/` | SUI Page | Custom pages (e.g., `/agents/demo/result`) |\n| `http://https://` | External | External URL (loaded in iframe) |\n\n**Examples:**\n\n```typescript\n// Open agent page in sidebar with title\nctx.Send({\n type: \"action\",\n props: {\n name: \"navigate\",\n payload: {\n route: \"/agents/my-assistant/result\",\n title: \"Query Results\",\n icon: \"material-table_chart\",\n query: { id: \"123\" },\n },\n },\n});\n\n// Open CUI dashboard page\nctx.Send({\n type: \"action\",\n props: {\n name: \"navigate\",\n payload: { route: \"$dashboard/users\" },\n },\n});\n\n// Open external URL in new tab\nctx.Send({\n type: \"action\",\n props: {\n name: \"navigate\",\n payload: {\n route: \"https://docs.example.com\",\n target: \"_blank\",\n },\n },\n});\n```\n\n#### Navigate Back\n\nNavigate back in history.\n\n```typescript\nctx.Send({\n type: \"action\",\n props: { name: \"navigate.back\" },\n});\n```\n\n#### Notify\n\nShow notification messages.\n\n**Actions:**\n\n| Action | Description |\n| ---------------- | ----------------------------- |\n| `notify.success` | Success notification (green) |\n| `notify.error` | Error notification (red) |\n| `notify.warning` | Warning notification (yellow) |\n| `notify.info` | Info notification (blue) |\n\n**Payload:**\n\n| Field | Type | Required | Description |\n| ---------- | --------- | -------- | ---------------------------------------------- |\n| `message` | `string` | ✅ | Notification message |\n| `duration` | `number` | - | Auto-close seconds (default: 3, 0 = keep open) |\n| `icon` | `string` | - | Custom icon (overrides default) |\n| `closable` | `boolean` | - | Show close button (default: false) |\n\n**Examples:**\n\n```typescript\n// Success notification\nctx.Send({\n type: \"action\",\n props: {\n name: \"notify.success\",\n payload: { message: \"Data saved successfully!\" },\n },\n});\n\n// Error with custom duration\nctx.Send({\n type: \"action\",\n props: {\n name: \"notify.error\",\n payload: {\n message: \"Operation failed\",\n duration: 5,\n closable: true,\n },\n },\n});\n```\n\n#### App Menu\n\nRefresh application menu/navigation.\n\n```typescript\nctx.Send({\n type: \"action\",\n props: { name: \"app.menu.reload\" },\n});\n```\n\n#### All Actions\n\n| Category | Action | Description |\n| -------- | ------------------- | ------------------------------- |\n| Navigate | `navigate` | Open page in sidebar or new tab |\n| | `navigate.back` | Navigate back in history |\n| Notify | `notify.success` | Show success notification |\n| | `notify.error` | Show error notification |\n| | `notify.warning` | Show warning notification |\n| | `notify.info` | Show info notification |\n| App | `app.menu.reload` | Refresh application menu |\n| Modal | `modal.open` | Open content in modal dialog |\n| | `modal.close` | Close modal |\n| Table | `table.search` | Trigger table search |\n| | `table.refresh` | Refresh table data |\n| | `table.save` | Save table row data |\n| | `table.delete` | Delete table row(s) |\n| Form | `form.find` | Load form data by ID |\n| | `form.submit` | Submit form data |\n| | `form.reset` | Reset form to initial state |\n| | `form.setFields` | Set form field values |\n| MCP | `mcp.tool.call` | Execute MCP tool (client-side) |\n| | `mcp.resource.read` | Read MCP resource |\n| Event | `event.emit` | Emit custom event |\n| Confirm | `confirm` | Show confirmation dialog |\n\n## Frontend API\n\n### Backend Calls\n\n```typescript\nimport { $Backend, Yao } from \"@yao/sui\";\n\n// Call backend method defined in .backend.ts\nconst data = await $Backend().Call(\"MethodName\", arg1, arg2);\n\n// Direct API calls\nconst yao = new Yao();\nconst res = await yao.Get(\"/api/endpoint\", { query: \"value\" });\nawait yao.Post(\"/api/endpoint\", { body: \"data\" });\n```\n\n### State Management\n\n```typescript\nimport { Component } from \"@yao/sui\";\n\nconst self = this as Component;\n\n// Store values (per component instance)\nself.store.Set(\"key\", value);\nconst value = self.store.Get(\"key\");\n```\n\n### Parent Communication (Iframe)\n\n```typescript\n// Helper: Send action to CUI parent\nconst sendAction = (name: string, payload?: any) => {\n window.parent.postMessage(\n { type: \"action\", message: { name, payload } },\n window.location.origin\n );\n};\n\n// Usage\nsendAction(\"notify.success\", { message: \"Done!\" });\nsendAction(\"navigate\", {\n route: \"/agents/my-assistant/detail\",\n title: \"Details\",\n});\n\n// Receive messages from parent\nwindow.addEventListener(\"message\", (e) => {\n if (e.origin !== window.location.origin) return;\n const { type, message } = e.data;\n if (type === \"setup\") {\n document.documentElement.setAttribute(\"data-theme\", message.theme);\n }\n});\n```\n\n## Iframe Communication\n\nWhen pages are embedded in CUI via `/web//`, they can communicate with the host:\n\n### Receiving Context\n\n```javascript\nwindow.addEventListener(\"message\", (e) => {\n if (e.origin !== window.location.origin) return;\n if (e.data.type === \"setup\") {\n const { theme, locale } = e.data.message;\n // Apply theme, set locale\n document.documentElement.setAttribute(\"data-theme\", theme);\n }\n});\n```\n\n### Sending Actions\n\n```javascript\n// Helper function\nconst sendAction = (name, payload) => {\n window.parent.postMessage(\n { type: \"action\", message: { name, payload } },\n window.location.origin\n );\n};\n\n// Show notification\nsendAction(\"notify.success\", { message: \"Done!\" });\n\n// Navigate to page\nsendAction(\"navigate\", {\n route: \"/agents/my-assistant/detail\",\n title: \"Details\",\n});\n```\n\nSee [Iframe Integration](iframe.md) for complete documentation.\n\n## Related Documentation\n\n- [Iframe Integration](iframe.md) - CUI iframe communication\n- [SUI Template Syntax](../../sui/docs/template-syntax.md)\n- [SUI Data Binding](../../sui/docs/data-binding.md)\n- [SUI Components](../../sui/docs/components.md)\n- [SUI Frontend API](../../sui/docs/frontend-api.md)\n" }, { "path": "agent/docs/prompts.md", "content": "# Prompts\n\n## Default Prompts\n\nCreate `prompts.yml` in the assistant directory:\n\n```yaml\n- role: system\n content: |\n You are a helpful assistant.\n\n ## Guidelines\n - Be concise and accurate\n - Ask clarifying questions when needed\n\n- role: system\n name: context\n content: |\n Current date: {{ $CTX.date }}\n User locale: {{ $CTX.locale }}\n```\n\n### Prompt Structure\n\n```yaml\n- role: system | user | assistant\n content: string\n name: string (optional)\n```\n\n### Context Variables\n\nUse `$CTX.*` for runtime context:\n\n| Variable | Description |\n| ---------------- | -------------------------- |\n| `$CTX.date` | Current date |\n| `$CTX.time` | Current time |\n| `$CTX.locale` | User locale (e.g., en-us) |\n| `$CTX.timezone` | User timezone |\n| `$CTX.user_id` | Current user ID |\n| `$CTX.team_id` | Current team ID |\n| `$CTX.chat_id` | Current chat session ID |\n\n## Prompt Presets\n\nCreate presets in `prompts/` directory for different scenarios:\n\n```\nprompts/\n├── chat.yml # Casual conversation\n├── task.yml # Task-oriented\n└── analysis.yml # Data analysis\n```\n\n**prompts/chat.yml**\n\n```yaml\n- role: system\n content: |\n You are a friendly conversational assistant.\n Be warm and engaging.\n```\n\n**prompts/task.yml**\n\n```yaml\n- role: system\n content: |\n You are a task-focused assistant.\n Be precise and efficient.\n```\n\n### Using Presets\n\nSelect preset in Create hook:\n\n```typescript\nfunction Create(ctx: agent.Context, messages: agent.Message[]): agent.Create {\n return {\n messages,\n prompt_preset: \"task\", // Use prompts/task.yml\n };\n}\n```\n\nOr via mode configuration in `package.yao`:\n\n```json\n{\n \"modes\": [\"chat\", \"task\"],\n \"default_mode\": \"task\"\n}\n```\n\n## Global Prompts\n\nDefine global prompts in `agent/prompts.yml` (applies to all assistants):\n\n```yaml\n- role: system\n content: |\n # Global Guidelines\n - Always be helpful and respectful\n - Follow company policies\n```\n\n### Disabling Global Prompts\n\nPer assistant:\n\n```json\n{\n \"disable_global_prompts\": true\n}\n```\n\nPer request (in Create hook):\n\n```typescript\nfunction Create(ctx: agent.Context, messages: agent.Message[]): agent.Create {\n return {\n messages,\n disable_global_prompts: true,\n };\n}\n```\n\n## Multi-line Content\n\nUse YAML block scalars for long content:\n\n```yaml\n- role: system\n content: |\n # Assistant Role\n \n You are an expert in data analysis.\n \n ## Capabilities\n - Statistical analysis\n - Data visualization\n - Report generation\n \n ## Guidelines\n 1. Always validate input data\n 2. Explain your methodology\n 3. Provide actionable insights\n```\n\n## Dynamic Prompts\n\nInject dynamic content in Create hook:\n\n```typescript\nfunction Create(ctx: agent.Context, messages: agent.Message[]): agent.Create {\n const userPrefs = ctx.memory.user.Get(\"preferences\");\n\n // Add dynamic system message\n const dynamicPrompt = {\n role: \"system\",\n content: `User preferences: ${JSON.stringify(userPrefs)}`,\n };\n\n return {\n messages: [dynamicPrompt, ...messages],\n };\n}\n```\n\n## Prompt Best Practices\n\n1. **Be specific** - Clear instructions produce better results\n2. **Use structure** - Headers, lists, and sections improve readability\n3. **Set boundaries** - Define what the assistant should and shouldn't do\n4. **Include examples** - Show expected input/output formats\n5. **Layer prompts** - Use global + assistant + dynamic prompts together\n" }, { "path": "agent/docs/search.md", "content": "# Search\n\nThe agent search system provides automatic search across web, knowledge base (KB), and database (DB).\n\n## Auto Search Flow\n\n1. **Intent Detection** - `__yao.needsearch` agent analyzes user message\n2. **Search Execution** - Executes web/kb/db searches based on intent\n3. **Context Injection** - Results injected as system message\n4. **Citation** - LLM can cite results using `[1]`, `[2]` format\n\n## Configuration\n\n### Global (agent/search.yml)\n\n```yaml\nweb:\n provider: tavily # tavily, serper, serpapi\n api_key_env: TAVILY_API_KEY\n max_results: 10\n\nkb:\n threshold: 0.7\n graph: true\n\ndb:\n max_results: 20\n\nkeyword:\n max_keywords: 5\n language: en\n\ncitation:\n format: \"[{index}]\"\n auto_inject_prompt: true\n```\n\n### Per Assistant (package.yao)\n\n```json\n{\n \"search\": {\n \"web\": { \"max_results\": 5 },\n \"kb\": { \"threshold\": 0.8 },\n \"citation\": { \"format\": \"[{index}]\" }\n },\n \"kb\": {\n \"collections\": [\"docs\", \"faq\"]\n },\n \"db\": {\n \"models\": [\"articles\", \"products\"]\n }\n}\n```\n\n## Controlling Search in Hooks\n\n### Disable Search\n\n```typescript\nfunction Create(ctx: agent.Context, messages: agent.Message[]): agent.Create {\n return {\n messages,\n search: false\n };\n}\n```\n\n### Enable Specific Types\n\n```typescript\nfunction Create(ctx: agent.Context, messages: agent.Message[]): agent.Create {\n return {\n messages,\n search: {\n need_search: true,\n search_types: [\"kb\", \"db\"], // Only KB and DB\n confidence: 1.0,\n reason: \"controlled by hook\"\n }\n };\n}\n```\n\n### Disable via Uses\n\n```typescript\nfunction Create(ctx: agent.Context, messages: agent.Message[]): agent.Create {\n return {\n messages,\n uses: { search: \"disabled\" }\n };\n}\n```\n\n## Search API (ctx.search)\n\n### Web Search\n\n```typescript\nconst result = ctx.search.Web(\"query\", {\n limit: 10,\n sites: [\"example.com\", \"docs.example.com\"],\n time_range: \"week\", // day, week, month, year\n rerank: { top_n: 5 }\n});\n```\n\n### Knowledge Base Search\n\n```typescript\nconst result = ctx.search.KB(\"query\", {\n collections: [\"docs\", \"faq\"],\n threshold: 0.7,\n limit: 10,\n graph: true,\n rerank: { top_n: 5 }\n});\n```\n\n### Database Search\n\n```typescript\nconst result = ctx.search.DB(\"query\", {\n models: [\"articles\"],\n wheres: [{ column: \"status\", value: \"published\" }],\n orders: [{ column: \"created_at\", option: \"desc\" }],\n select: [\"id\", \"title\", \"content\"],\n limit: 20,\n rerank: { top_n: 10 }\n});\n```\n\n### Parallel Search\n\n```typescript\n// Wait for all\nconst results = ctx.search.All([\n { type: \"web\", query: \"topic\" },\n { type: \"kb\", query: \"topic\", collections: [\"docs\"] },\n { type: \"db\", query: \"topic\", models: [\"articles\"] }\n]);\n\n// First success with results\nconst results = ctx.search.Any([\n { type: \"web\", query: \"topic\" },\n { type: \"kb\", query: \"topic\" }\n]);\n\n// First to complete\nconst results = ctx.search.Race([\n { type: \"web\", query: \"topic\" },\n { type: \"kb\", query: \"topic\" }\n]);\n```\n\n## Result Structure\n\n```typescript\ninterface SearchResult {\n type: \"web\" | \"kb\" | \"db\";\n query: string;\n source: \"hook\" | \"auto\" | \"user\";\n items: SearchItem[];\n error?: string;\n}\n\ninterface SearchItem {\n citation_id: string; // \"1\", \"2\", etc.\n title: string;\n url: string;\n content: string;\n score: number;\n}\n```\n\n## Custom Search in Hooks\n\n```typescript\nfunction Create(ctx: agent.Context, messages: agent.Message[]): agent.Create {\n const query = messages[messages.length - 1]?.content || \"\";\n\n // Custom KB search\n const kbResult = ctx.search.KB(query, {\n collections: [\"internal_docs\"],\n threshold: 0.8\n });\n\n if (kbResult.items?.length > 0) {\n // Format results as context\n const context = kbResult.items\n .map((item, i) => `[${i + 1}] ${item.title}\\n${item.content}`)\n .join(\"\\n\\n\");\n\n // Inject as system message\n const contextMsg = {\n role: \"system\",\n content: `Reference information:\\n${context}`\n };\n\n return {\n messages: [contextMsg, ...messages],\n search: false // Skip auto search\n };\n }\n\n return { messages };\n}\n```\n\n## Authorization\n\n### KB Collections\n\nCollections are filtered by user authorization:\n\n```typescript\n// Only collections user has access to are searched\nconst result = ctx.search.KB(\"query\", {\n collections: [\"public\", \"internal\", \"secret\"]\n // User without \"secret\" access won't search that collection\n});\n```\n\n### DB Models\n\nDatabase queries include permission filters:\n\n```typescript\n// Auth filters are automatically added\n// e.g., { column: \"__yao_created_by\", value: user_id }\nconst result = ctx.search.DB(\"query\", {\n models: [\"user_documents\"]\n});\n```\n\n## Web Search Providers\n\n### Tavily\n\n```yaml\nweb:\n provider: tavily\n api_key_env: TAVILY_API_KEY\n```\n\n### Serper\n\n```yaml\nweb:\n provider: serper\n api_key_env: SERPER_API_KEY\n```\n\n### SerpAPI\n\n```yaml\nweb:\n provider: serpapi\n api_key_env: SERPAPI_API_KEY\n```\n\n## Citation\n\nLLM responses can include citations:\n\n```\nBased on the documentation [1], the feature works by... [2]\n\nReferences:\n[1] Getting Started Guide - https://docs.example.com/start\n[2] API Reference - https://docs.example.com/api\n```\n\n### Citation Format\n\n```yaml\ncitation:\n format: \"[{index}]\" # or \"({index})\" or \"[^{index}]\"\n auto_inject_prompt: true\n custom_prompt: |\n When citing sources, use the format [N] where N is the reference number.\n```\n" }, { "path": "agent/docs/testing.md", "content": "# Agent Testing\n\nA comprehensive testing framework for Yao AI agents with support for standard testing, dynamic (simulator-driven) testing, agent-driven assertions, and CI integration.\n\n## Quick Start\n\n```bash\n# Test with direct message (auto-detect agent from current directory)\ncd assistants/my-assistant\nyao agent test -i \"Hello, how are you?\"\n\n# Test with JSONL file\nyao agent test -i tests/inputs.jsonl\n\n# Generate HTML report\nyao agent test -i tests/inputs.jsonl -o report.html\n\n# Stability analysis (run each test 5 times)\nyao agent test -i tests/inputs.jsonl --runs 5\n```\n\n## Input Modes\n\nThe `-i` flag supports multiple input modes:\n\n| Mode | Example | Description |\n|------|---------|-------------|\n| Direct message | `-i \"Hello\"` | Single message test |\n| JSONL file | `-i tests/inputs.jsonl` | Multiple test cases |\n| Agent-driven | `-i \"agents:tests.generator?count=10\"` | Generate tests with agent |\n| Script test | `-i scripts.expense.setup` | Test handler scripts |\n| Script-generated | `-i \"scripts:tests.gen.Generate\"` | Generate tests from script |\n\n## Test Case Format (JSONL)\n\n### Basic Test\n\n```jsonl\n{\"id\": \"greeting\", \"input\": \"Hello\", \"assert\": {\"type\": \"contains\", \"value\": \"Hi\"}}\n```\n\n### With Conversation History\n\n```jsonl\n{\n \"id\": \"multi-turn\",\n \"input\": [\n {\"role\": \"user\", \"content\": \"What's 2+2?\"},\n {\"role\": \"assistant\", \"content\": \"4\"},\n {\"role\": \"user\", \"content\": \"Multiply by 3\"}\n ],\n \"assert\": {\"type\": \"contains\", \"value\": \"12\"}\n}\n```\n\n### With File Attachments\n\n```jsonl\n{\n \"id\": \"image-test\",\n \"input\": {\n \"role\": \"user\",\n \"content\": [\n {\"type\": \"text\", \"text\": \"Describe this image\"},\n {\"type\": \"image\", \"source\": \"file://fixtures/test.jpg\"}\n ]\n }\n}\n```\n\n## Assertions\n\n### Static Assertions\n\n| Type | Description | Example |\n|------|-------------|---------|\n| `equals` | Exact match | `{\"type\": \"equals\", \"value\": {\"key\": \"val\"}}` |\n| `contains` | Output contains value | `{\"type\": \"contains\", \"value\": \"keyword\"}` |\n| `not_contains` | Output does not contain | `{\"type\": \"not_contains\", \"value\": \"error\"}` |\n| `regex` | Match regex pattern | `{\"type\": \"regex\", \"value\": \"\\\\d+\"}` |\n| `json_path` | Extract and compare | `{\"type\": \"json_path\", \"path\": \"$.field\", \"value\": true}` |\n| `type` | Check output type | `{\"type\": \"type\", \"value\": \"object\"}` |\n| `tool_called` | Check tool was called | `{\"type\": \"tool_called\", \"value\": \"setup\"}` |\n| `tool_result` | Check tool result | `{\"type\": \"tool_result\", \"value\": {\"tool\": \"setup\", \"result\": {\"success\": true}}}` |\n\n### Agent-Driven Assertions\n\nUse LLM to validate response semantics:\n\n```jsonl\n{\n \"id\": \"helpful-response\",\n \"input\": \"How do I reset my password?\",\n \"assert\": {\n \"type\": \"agent\",\n \"use\": \"agents:tests.validator-agent\",\n \"value\": \"Response should provide clear step-by-step instructions\"\n }\n}\n```\n\n### Multiple Assertions\n\nAll assertions must pass:\n\n```jsonl\n{\n \"id\": \"complete-check\",\n \"input\": \"Submit expense\",\n \"assert\": [\n {\"type\": \"contains\", \"value\": \"expense\"},\n {\"type\": \"not_contains\", \"value\": \"error\"},\n {\"type\": \"regex\", \"value\": \"(?i)(submitted|created)\"}\n ]\n}\n```\n\n## Dynamic Mode (Simulator)\n\nFor testing complex conversation flows with a user simulator:\n\n```jsonl\n{\n \"id\": \"order-flow\",\n \"input\": \"I want to order coffee\",\n \"simulator\": {\n \"use\": \"tests.simulator-agent\",\n \"options\": {\n \"metadata\": {\n \"persona\": \"Customer\",\n \"goal\": \"Order a medium latte\"\n }\n }\n },\n \"checkpoints\": [\n {\n \"id\": \"greeting\",\n \"assert\": {\"type\": \"regex\", \"value\": \"(?i)(hello|hi)\"}\n },\n {\n \"id\": \"ask-size\",\n \"after\": [\"greeting\"],\n \"assert\": {\"type\": \"regex\", \"value\": \"(?i)size\"}\n },\n {\n \"id\": \"confirm\",\n \"after\": [\"ask-size\"],\n \"assert\": {\"type\": \"regex\", \"value\": \"(?i)confirm\"}\n }\n ],\n \"max_turns\": 10\n}\n```\n\nRun with:\n\n```bash\nyao agent test -i tests/dynamic.jsonl --simulator tests.simulator-agent -v\n```\n\n## Script Testing\n\nTest agent handler scripts with the `t.assert` API:\n\n```typescript\n// assistants/my-assistant/src/setup_test.ts\nimport { SystemReady } from \"./setup\";\n\nexport function TestSystemReady(t: TestingT, ctx: Context) {\n const result = SystemReady(ctx);\n \n t.assert.True(result.success, \"Should succeed\");\n t.assert.Equal(result.status, \"ready\", \"Status should be ready\");\n t.assert.NotNil(result.data, \"Data should not be nil\");\n}\n\nexport function TestWithAgentAssertion(t: TestingT, ctx: Context) {\n const response = Process(\"agents.my-assistant.Stream\", ctx, messages);\n \n // Static assertion\n t.assert.Contains(response.content, \"confirm\");\n \n // Agent-driven assertion\n t.assert.Agent(response.content, \"tests.validator-agent\", {\n criteria: \"Response should ask for confirmation\"\n });\n}\n```\n\nRun with:\n\n```bash\nyao agent test -i scripts.my-assistant.setup -v\n```\n\n### Available Assertions\n\n| Method | Description |\n|--------|-------------|\n| `t.assert.True(value, msg)` | Assert value is true |\n| `t.assert.False(value, msg)` | Assert value is false |\n| `t.assert.Equal(a, b, msg)` | Assert a equals b |\n| `t.assert.NotEqual(a, b, msg)` | Assert a not equals b |\n| `t.assert.Nil(value, msg)` | Assert value is null/undefined |\n| `t.assert.NotNil(value, msg)` | Assert value is not nil |\n| `t.assert.Contains(s, sub, msg)` | Assert string contains substr |\n| `t.assert.Len(arr, n, msg)` | Assert array/string length |\n| `t.assert.Agent(resp, id, opts)` | Agent-driven assertion |\n\n## Before/After Hooks\n\n### Per-Test Hooks\n\n```jsonl\n{\n \"id\": \"with-setup\",\n \"input\": \"Show my data\",\n \"before\": \"env_test.Before\",\n \"after\": \"env_test.After\"\n}\n```\n\n### Global Hooks\n\n```bash\nyao agent test -i tests/inputs.jsonl --before env_test.BeforeAll --after env_test.AfterAll\n```\n\n### Hook Implementation\n\n```typescript\n// assistants/my-assistant/src/env_test.ts\n\nexport function Before(ctx: Context, testCase: TestCase): any {\n const userId = Process(\"models.user.Create\", { name: \"Test User\" });\n return { userId }; // Passed to After\n}\n\nexport function After(ctx: Context, testCase: TestCase, result: TestResult, beforeData: any) {\n if (beforeData?.userId) {\n Process(\"models.user.Delete\", beforeData.userId);\n }\n}\n\nexport function BeforeAll(ctx: Context, testCases: TestCase[]): any {\n Process(\"models.migrate\");\n return { initialized: true };\n}\n\nexport function AfterAll(ctx: Context, results: TestResult[], beforeData: any) {\n const passed = results.filter(r => r.status === \"passed\").length;\n console.log(`Tests completed: ${passed}/${results.length} passed`);\n}\n```\n\n## Custom Context\n\nCreate a JSON file for custom authorization:\n\n```json\n{\n \"chat_id\": \"test-chat-001\",\n \"authorized\": {\n \"user_id\": \"test-user-123\",\n \"team_id\": \"test-team-456\",\n \"constraints\": {\n \"owner_only\": true,\n \"extra\": { \"department\": \"engineering\" }\n }\n }\n}\n```\n\nUse with `--ctx`:\n\n```bash\nyao agent test -i scripts.my-assistant.setup --ctx tests/context.json -v\n```\n\n## Command Line Options\n\n| Flag | Description | Default |\n|------|-------------|---------|\n| `-i` | Input: JSONL file, message, `agents:xxx`, or `scripts:xxx` | (required) |\n| `-o` | Output file path | `output-{timestamp}.jsonl` |\n| `-n` | Agent ID (optional, auto-detected) | auto-detect |\n| `-a` | Application directory | auto-detect |\n| `-e` | Environment file | - |\n| `-c` | Override connector | agent default |\n| `-u` | Test user ID | `test-user` |\n| `-t` | Test team ID | `test-team` |\n| `-r` | Reporter agent ID | built-in |\n| `-v` | Verbose output | false |\n| `--ctx` | Path to context JSON file | - |\n| `--simulator` | Default simulator agent ID | - |\n| `--before` | Global BeforeAll hook | - |\n| `--after` | Global AfterAll hook | - |\n| `--runs` | Runs per test (stability analysis) | 1 |\n| `--run` | Regex pattern to filter tests | - |\n| `--timeout` | Timeout per test | 2m |\n| `--parallel` | Parallel test cases | 1 |\n| `--fail-fast` | Stop on first failure | false |\n| `--dry-run` | Generate tests without running | false |\n\n## Output Formats\n\nDetermined by `-o` file extension:\n\n| Extension | Format | Description |\n|-----------|--------|-------------|\n| `.jsonl` | JSONL | Streaming (default) |\n| `.json` | JSON | Complete structured |\n| `.md` | Markdown | Human-readable |\n| `.html` | HTML | Interactive web report |\n\n## Stability Analysis\n\nRun each test multiple times to measure consistency:\n\n```bash\nyao agent test -i tests/inputs.jsonl --runs 5 -o stability.json\n```\n\n| Pass Rate | Classification |\n|-----------|----------------|\n| 100% | Stable |\n| 80-99% | Mostly Stable |\n| 50-79% | Unstable |\n| < 50% | Highly Unstable |\n\n## CI Integration\n\n```bash\n# Exit code: 0 = all passed, 1 = failures\nyao agent test -i tests/inputs.jsonl --fail-fast\n\n# Run with parallel execution\nyao agent test -i tests/inputs.jsonl --parallel 4\n```\n\n### GitHub Actions Example\n\n```yaml\n- name: Run Agent Tests\n run: |\n yao agent test -i assistants/my-assistant/tests/inputs.jsonl \\\n -u ci-user -t ci-team \\\n --runs 3 \\\n -o report.json\n\n- name: Run Dynamic Tests\n run: |\n yao agent test -i assistants/my-assistant/tests/dynamic.jsonl \\\n --simulator tests.simulator-agent \\\n -v\n\n- name: Run Script Tests\n run: |\n yao agent test -i scripts.my-assistant.setup -v\n```\n\n## Exit Codes\n\n| Code | Description |\n|------|-------------|\n| 0 | All tests passed |\n| 1 | Tests failed, configuration error, or runtime error |\n" }, { "path": "agent/i18n/builtin.go", "content": "package i18n\n\n// init registers built-in global messages\nfunc init() {\n\t// Initialize __global__ if not exists\n\tif Locales[\"__global__\"] == nil {\n\t\tLocales[\"__global__\"] = make(map[string]I18n)\n\t}\n\n\t// Built-in English messages\n\tLocales[\"__global__\"][\"en\"] = I18n{\n\t\tLocale: \"en\",\n\t\tMessages: map[string]any{\n\t\t\t// Assistant: agent.go Stream() function\n\t\t\t\"assistant.agent.stream.label\": \"{{name}}\",\n\t\t\t\"assistant.agent.stream.description\": \"{{name}} is processing the request\",\n\t\t\t\"assistant.agent.stream.history\": \"Get Chat History\",\n\t\t\t\"assistant.agent.stream.capabilities\": \"Get Connector Capabilities\",\n\t\t\t\"assistant.agent.stream.create_hook\": \"Call Create Hook\",\n\t\t\t\"assistant.agent.stream.closing\": \"Closing output (root call)\",\n\t\t\t\"assistant.agent.stream.skipping\": \"Skipping output close (nested call)\",\n\t\t\t\"assistant.agent.stream.close_error\": \"Failed to close output\",\n\t\t\t\"assistant.agent.completion.label\": \"Agent Completion\",\n\t\t\t\"assistant.agent.completion.description\": \"Final output from {{name}}\",\n\n\t\t\t// LLM: providers/openai/openai.go Stream() function\n\t\t\t\"llm.openai.stream.label\": \"LLM %s\",\n\t\t\t\"llm.openai.stream.description\": \"LLM %s is processing the request\",\n\t\t\t\"llm.openai.stream.starting\": \"Starting stream request\",\n\t\t\t\"llm.openai.stream.request\": \"Stream Request\",\n\t\t\t\"llm.openai.stream.retry\": \"Stream request failed, retrying\",\n\t\t\t\"llm.openai.stream.api_error\": \"OpenAI API returned error response\",\n\t\t\t\"llm.openai.stream.error\": \"OpenAI Stream Error\",\n\t\t\t\"llm.openai.stream.no_data\": \"Request body that caused empty response\",\n\t\t\t\"llm.openai.stream.no_data_info\": \"Request details\",\n\t\t\t\"llm.openai.post.api_error\": \"OpenAI API error response\",\n\n\t\t\t// LLM: handlers/stream.go (general LLM stream handler)\n\t\t\t\"llm.handlers.stream.info\": \"LLM Stream\",\n\t\t\t\"llm.handlers.stream.raw_output\": \"LLM Raw Output\",\n\n\t\t\t// Output: adapters/openai/writer.go\n\t\t\t\"output.openai.writer.sending_chunk\": \"Sending chunk to client\",\n\t\t\t\"output.openai.writer.sending_done\": \"Sending [DONE] to client\",\n\t\t\t\"output.openai.writer.adapt_error\": \"Failed to adapt message\",\n\t\t\t\"output.openai.writer.chunk_error\": \"Failed to send chunk\",\n\t\t\t\"output.openai.writer.group_error\": \"Failed to write message in group\",\n\t\t\t\"output.openai.writer.send_error\": \"Failed to send data to client\",\n\t\t\t\"output.openai.writer.marshal_error\": \"Failed to marshal chunk\",\n\t\t\t\"output.openai.writer.done_error\": \"Failed to send [DONE] to client\",\n\n\t\t\t// Output: adapters/cui/writer.go\n\t\t\t\"output.cui.writer.sending_chunk\": \"Sending chunk to client\",\n\t\t\t\"output.cui.writer.adapt_error\": \"Failed to adapt message\",\n\t\t\t\"output.cui.writer.chunk_error\": \"Failed to send chunk\",\n\t\t\t\"output.cui.writer.group_error\": \"Failed to send message group\",\n\t\t\t\"output.cui.writer.send_error\": \"Failed to send data to client\",\n\t\t\t\"output.cui.writer.marshal_error\": \"Failed to marshal chunk\",\n\n\t\t\t// Output: Stream event messages\n\t\t\t\"output.stream_start\": \"Assistant is processing\",\n\t\t\t\"output.view_trace\": \"View process\",\n\n\t\t\t// Common status messages\n\t\t\t\"common.status.processing\": \"Processing\",\n\t\t\t\"common.status.completed\": \"Completed\",\n\t\t\t\"common.status.failed\": \"Failed\",\n\t\t\t\"common.status.retrying\": \"Retrying\",\n\n\t\t\t// MCP: context/mcp.go - Resource operations\n\t\t\t\"mcp.list_resources.label\": \"MCP: List Resources\",\n\t\t\t\"mcp.list_resources.description\": \"List resources from MCP client '%s'\",\n\t\t\t\"mcp.read_resource.label\": \"MCP: Read Resource\",\n\t\t\t\"mcp.read_resource.description\": \"Read resource '%s' from MCP client '%s'\",\n\n\t\t\t// MCP: context/mcp.go - Tool operations\n\t\t\t\"mcp.list_tools.label\": \"MCP: List Tools\",\n\t\t\t\"mcp.list_tools.description\": \"List tools from MCP client '%s'\",\n\t\t\t\"mcp.call_tool.label\": \"MCP: Call Tool\",\n\t\t\t\"mcp.call_tool.description\": \"Call tool '%s' from MCP client '%s'\",\n\t\t\t\"mcp.call_tools.label\": \"MCP: Call Tools\",\n\t\t\t\"mcp.call_tools.description\": \"Call %d tools sequentially from MCP client '%s'\",\n\t\t\t\"mcp.call_tools_parallel.label\": \"MCP: Call Tools (Parallel)\",\n\t\t\t\"mcp.call_tools_parallel.description\": \"Call %d tools in parallel from MCP client '%s'\",\n\n\t\t\t// MCP: context/mcp.go - Prompt operations\n\t\t\t\"mcp.list_prompts.label\": \"MCP: List Prompts\",\n\t\t\t\"mcp.list_prompts.description\": \"List prompts from MCP client '%s'\",\n\t\t\t\"mcp.get_prompt.label\": \"MCP: Get Prompt\",\n\t\t\t\"mcp.get_prompt.description\": \"Get prompt '%s' from MCP client '%s'\",\n\n\t\t\t// MCP: context/mcp.go - Sample operations\n\t\t\t\"mcp.list_samples.label\": \"MCP: List Samples\",\n\t\t\t\"mcp.list_samples.description\": \"List samples for '%s' from MCP client '%s'\",\n\t\t\t\"mcp.get_sample.label\": \"MCP: Get Sample\",\n\t\t\t\"mcp.get_sample.description\": \"Get sample #%d for '%s' from MCP client '%s'\",\n\n\t\t\t// KB: Chat collection\n\t\t\t\"kb.chat.name\": \"Chat Knowledge Base\",\n\t\t\t\"kb.chat.description\": \"Auto-created knowledge base collection for chat sessions\",\n\n\t\t\t// Sandbox: assistant/sandbox.go - Sandbox status messages\n\t\t\t\"sandbox.preparing\": \"Getting things ready...\",\n\t\t\t\"sandbox.ready\": \"Sandbox ready\",\n\t\t\t\"sandbox.working\": \"Working on your request\",\n\t\t\t\"sandbox.completed\": \"Completed\",\n\t\t\t\"sandbox.failed\": \"Execution failed\",\n\t\t\t\"sandbox.starting\": \"Setting up workspace...\",\n\t\t\t\"sandbox.pulling_image\": \"Preparing environment (first time may take a moment)\",\n\t\t\t\"sandbox.waiting_response\": \"Waiting for AI response...\",\n\n\t\t\t// Sandbox: claude/executor.go - Tool execution messages\n\t\t\t\"sandbox.tool.read\": \"Reading file\",\n\t\t\t\"sandbox.tool.write\": \"Writing file\",\n\t\t\t\"sandbox.tool.edit\": \"Editing file\",\n\t\t\t\"sandbox.tool.bash\": \"Running command\",\n\t\t\t\"sandbox.tool.glob\": \"Finding files\",\n\t\t\t\"sandbox.tool.grep\": \"Searching code\",\n\t\t\t\"sandbox.tool.ls\": \"Listing directory\",\n\t\t\t\"sandbox.tool.task\": \"Running subtask\",\n\t\t\t\"sandbox.tool.web_search\": \"Searching web\",\n\t\t\t\"sandbox.tool.web_fetch\": \"Fetching URL\",\n\t\t\t\"sandbox.tool.todo_write\": \"Managing tasks\",\n\t\t\t\"sandbox.tool.ask_question\": \"Asking question\",\n\t\t\t\"sandbox.tool.switch_mode\": \"Switching mode\",\n\t\t\t\"sandbox.tool.read_lints\": \"Checking lints\",\n\t\t\t\"sandbox.tool.edit_notebook\": \"Editing notebook\",\n\t\t\t\"sandbox.tool.unknown\": \"Executing {{name}}\",\n\n\t\t\t// Content: content/image/image.go - Image processing messages\n\t\t\t\"content.image.analyzing\": \"Analyzing image\",\n\n\t\t\t// Content: content/pdf/pdf.go - PDF processing messages\n\t\t\t\"content.pdf.analyzing_page\": \"Analyzing PDF page %d/%d\",\n\n\t\t\t// Search: assistant/search.go - Output messages\n\t\t\t\"search.loading\": \"Searching\",\n\t\t\t\"search.success\": \"Found %d references\",\n\t\t\t\"search.success.one\": \"Found 1 reference\",\n\t\t\t\"search.partial\": \"Found %d references (some sources failed)\",\n\t\t\t\"search.failed\": \"Search failed\",\n\t\t\t\"search.no_results\": \"No references found\",\n\n\t\t\t// Search Intent: assistant/search.go - Intent detection messages\n\t\t\t\"search.intent.loading\": \"Checking if references are needed\",\n\t\t\t\"search.intent.need_search\": \"Searching for references\",\n\t\t\t\"search.intent.no_search\": \"No references needed\",\n\n\t\t\t// Keyword Extraction: assistant/search.go - Keyword extraction messages\n\t\t\t\"search.keyword.loading\": \"Analyzing conversation\",\n\t\t\t\"search.keyword.done\": \"Analysis complete\",\n\n\t\t\t// Search: assistant/search.go - Trace labels\n\t\t\t\"search.trace.label\": \"Search\",\n\t\t\t\"search.trace.description\": \"Search the web and knowledge base for relevant information\",\n\t\t\t\"search.trace.web.label\": \"Web Search\",\n\t\t\t\"search.trace.web.description\": \"Searching the web\",\n\t\t\t\"search.trace.kb.label\": \"KB Search\",\n\t\t\t\"search.trace.kb.description\": \"Searching knowledge base\",\n\t\t\t\"search.trace.db.label\": \"DB Search\",\n\t\t\t\"search.trace.db.description\": \"Searching database\",\n\t\t},\n\t}\n\n\t// Built-in Chinese (Simplified) messages\n\tLocales[\"__global__\"][\"zh-cn\"] = I18n{\n\t\tLocale: \"zh-cn\",\n\t\tMessages: map[string]any{\n\t\t\t// Assistant: agent.go Stream() function\n\t\t\t\"assistant.agent.stream.label\": \"{{name}}\",\n\t\t\t\"assistant.agent.stream.description\": \"{{name}} 正在处理请求\",\n\t\t\t\"assistant.agent.stream.history\": \"获取聊天历史\",\n\t\t\t\"assistant.agent.stream.capabilities\": \"获取连接器能力\",\n\t\t\t\"assistant.agent.stream.create_hook\": \"调用 Create Hook\",\n\t\t\t\"assistant.agent.stream.closing\": \"关闭输出（根调用）\",\n\t\t\t\"assistant.agent.stream.skipping\": \"跳过输出关闭（嵌套调用）\",\n\t\t\t\"assistant.agent.stream.close_error\": \"关闭输出失败\",\n\t\t\t\"assistant.agent.completion.label\": \"智能体完成\",\n\t\t\t\"assistant.agent.completion.description\": \"{{name}} 最终输出\",\n\n\t\t\t// LLM: providers/openai/openai.go Stream() function\n\t\t\t\"llm.openai.stream.label\": \"LLM %s\",\n\t\t\t\"llm.openai.stream.description\": \"LLM %s 正在处理请求\",\n\t\t\t\"llm.openai.stream.starting\": \"开始流式请求\",\n\t\t\t\"llm.openai.stream.request\": \"流式请求\",\n\t\t\t\"llm.openai.stream.retry\": \"流式请求失败，正在重试\",\n\t\t\t\"llm.openai.stream.api_error\": \"OpenAI API 返回错误响应\",\n\t\t\t\"llm.openai.stream.error\": \"OpenAI 流错误\",\n\t\t\t\"llm.openai.stream.no_data\": \"导致空响应的请求体\",\n\t\t\t\"llm.openai.stream.no_data_info\": \"请求详情\",\n\t\t\t\"llm.openai.post.api_error\": \"OpenAI API 错误响应\",\n\n\t\t\t// LLM: handlers/stream.go (general LLM stream handler)\n\t\t\t\"llm.handlers.stream.info\": \"LLM 流式输出\",\n\t\t\t\"llm.handlers.stream.raw_output\": \"LLM 原始输出\",\n\n\t\t\t// Output: adapters/openai/writer.go\n\t\t\t\"output.openai.writer.sending_chunk\": \"向客户端发送数据块\",\n\t\t\t\"output.openai.writer.sending_done\": \"向客户端发送 [DONE]\",\n\t\t\t\"output.openai.writer.adapt_error\": \"适配消息失败\",\n\t\t\t\"output.openai.writer.chunk_error\": \"发送数据块失败\",\n\t\t\t\"output.openai.writer.group_error\": \"写入消息组中的消息失败\",\n\t\t\t\"output.openai.writer.send_error\": \"发送数据到客户端失败\",\n\t\t\t\"output.openai.writer.marshal_error\": \"序列化数据块失败\",\n\t\t\t\"output.openai.writer.done_error\": \"发送 [DONE] 到客户端失败\",\n\n\t\t\t// Output: adapters/cui/writer.go\n\t\t\t\"output.cui.writer.sending_chunk\": \"向客户端发送数据块\",\n\t\t\t\"output.cui.writer.adapt_error\": \"适配消息失败\",\n\t\t\t\"output.cui.writer.chunk_error\": \"发送数据块失败\",\n\t\t\t\"output.cui.writer.group_error\": \"发送消息组失败\",\n\t\t\t\"output.cui.writer.send_error\": \"发送数据到客户端失败\",\n\t\t\t\"output.cui.writer.marshal_error\": \"序列化数据块失败\",\n\n\t\t\t// Output: Stream event messages\n\t\t\t\"output.stream_start\": \"智能体正在处理\",\n\t\t\t\"output.view_trace\": \"查看处理详情\",\n\n\t\t\t// Common status messages\n\t\t\t\"common.status.processing\": \"处理中\",\n\t\t\t\"common.status.completed\": \"已完成\",\n\t\t\t\"common.status.failed\": \"失败\",\n\t\t\t\"common.status.retrying\": \"重试中\",\n\n\t\t\t// KB: Chat collection\n\t\t\t\"kb.chat.name\": \"聊天知识库\",\n\t\t\t\"kb.chat.description\": \"自动为聊天会话创建的知识库集合\",\n\n\t\t\t// Sandbox: assistant/sandbox.go - Sandbox status messages\n\t\t\t\"sandbox.preparing\": \"正在准备...\",\n\t\t\t\"sandbox.ready\": \"就绪\",\n\t\t\t\"sandbox.working\": \"正在处理您的请求\",\n\t\t\t\"sandbox.completed\": \"处理完成\",\n\t\t\t\"sandbox.failed\": \"执行失败\",\n\t\t\t\"sandbox.starting\": \"正在启动工作区...\",\n\t\t\t\"sandbox.pulling_image\": \"正在准备运行环境（首次可能需要一点时间）\",\n\t\t\t\"sandbox.waiting_response\": \"等待 AI 响应...\",\n\n\t\t\t// Sandbox: claude/executor.go - Tool execution messages\n\t\t\t\"sandbox.tool.read\": \"正在读取文件\",\n\t\t\t\"sandbox.tool.write\": \"正在写入文件\",\n\t\t\t\"sandbox.tool.edit\": \"正在编辑文件\",\n\t\t\t\"sandbox.tool.bash\": \"正在执行命令\",\n\t\t\t\"sandbox.tool.glob\": \"正在查找文件\",\n\t\t\t\"sandbox.tool.grep\": \"正在搜索代码\",\n\t\t\t\"sandbox.tool.ls\": \"正在列出目录\",\n\t\t\t\"sandbox.tool.task\": \"正在执行子任务\",\n\t\t\t\"sandbox.tool.web_search\": \"正在搜索网页\",\n\t\t\t\"sandbox.tool.web_fetch\": \"正在获取网页\",\n\t\t\t\"sandbox.tool.todo_write\": \"正在管理任务\",\n\t\t\t\"sandbox.tool.ask_question\": \"正在询问问题\",\n\t\t\t\"sandbox.tool.switch_mode\": \"正在切换模式\",\n\t\t\t\"sandbox.tool.read_lints\": \"正在检查代码\",\n\t\t\t\"sandbox.tool.edit_notebook\": \"正在编辑笔记本\",\n\t\t\t\"sandbox.tool.unknown\": \"正在执行 {{name}}\",\n\n\t\t\t// Content: content/image/image.go - Image processing messages\n\t\t\t\"content.image.analyzing\": \"正在分析图片\",\n\n\t\t\t// Content: content/pdf/pdf.go - PDF processing messages\n\t\t\t\"content.pdf.analyzing_page\": \"正在分析 PDF 第 %d/%d 页\",\n\n\t\t\t// Search: assistant/search.go - Output messages\n\t\t\t\"search.loading\": \"正在搜索\",\n\t\t\t\"search.success\": \"找到 %d 条参考资料\",\n\t\t\t\"search.success.one\": \"找到 1 条参考资料\",\n\t\t\t\"search.partial\": \"找到 %d 条参考资料（部分来源失败）\",\n\t\t\t\"search.failed\": \"搜索失败\",\n\t\t\t\"search.no_results\": \"未找到相关资料\",\n\n\t\t\t// Search Intent: assistant/search.go - Intent detection messages\n\t\t\t\"search.intent.loading\": \"检查是否需要查询资料\",\n\t\t\t\"search.intent.need_search\": \"正在查询相关资料\",\n\t\t\t\"search.intent.no_search\": \"无需查询资料\",\n\n\t\t\t// Keyword Extraction: assistant/search.go - Keyword extraction messages\n\t\t\t\"search.keyword.loading\": \"正在分析对话内容\",\n\t\t\t\"search.keyword.done\": \"分析完成\",\n\n\t\t\t// Search: assistant/search.go - Trace labels\n\t\t\t\"search.trace.label\": \"搜索\",\n\t\t\t\"search.trace.description\": \"搜索网络和知识库获取相关信息\",\n\t\t\t\"search.trace.web.label\": \"网页搜索\",\n\t\t\t\"search.trace.web.description\": \"搜索网页获取相关信息\",\n\t\t\t\"search.trace.kb.label\": \"知识库搜索\",\n\t\t\t\"search.trace.kb.description\": \"搜索知识库获取相关信息\",\n\t\t\t\"search.trace.db.label\": \"数据库搜索\",\n\t\t\t\"search.trace.db.description\": \"搜索数据库获取相关信息\",\n\t\t},\n\t}\n\n\t// Built-in Chinese (short code) - same as zh-cn\n\tLocales[\"__global__\"][\"zh\"] = I18n{\n\t\tLocale: \"zh\",\n\t\tMessages: map[string]any{\n\t\t\t// Assistant: agent.go Stream() function\n\t\t\t\"assistant.agent.stream.label\": \"{{name}}\",\n\t\t\t\"assistant.agent.stream.description\": \"{{name}} 正在处理请求\",\n\t\t\t\"assistant.agent.stream.history\": \"获取聊天历史\",\n\t\t\t\"assistant.agent.stream.capabilities\": \"获取连接器能力\",\n\t\t\t\"assistant.agent.stream.create_hook\": \"调用 Create Hook\",\n\t\t\t\"assistant.agent.stream.closing\": \"关闭输出（根调用）\",\n\t\t\t\"assistant.agent.stream.skipping\": \"跳过输出关闭（嵌套调用）\",\n\t\t\t\"assistant.agent.stream.close_error\": \"关闭输出失败\",\n\t\t\t\"assistant.agent.completion.label\": \"智能体完成\",\n\t\t\t\"assistant.agent.completion.description\": \"{{name}} 最终输出\",\n\n\t\t\t// LLM: providers/openai/openai.go Stream() function\n\t\t\t\"llm.openai.stream.label\": \"LLM %s\",\n\t\t\t\"llm.openai.stream.description\": \"LLM %s 正在处理请求\",\n\t\t\t\"llm.openai.stream.starting\": \"开始流式请求\",\n\t\t\t\"llm.openai.stream.request\": \"流式请求\",\n\t\t\t\"llm.openai.stream.retry\": \"流式请求失败，正在重试\",\n\t\t\t\"llm.openai.stream.api_error\": \"OpenAI API 返回错误响应\",\n\t\t\t\"llm.openai.stream.error\": \"OpenAI 流错误\",\n\t\t\t\"llm.openai.stream.no_data\": \"导致空响应的请求体\",\n\t\t\t\"llm.openai.stream.no_data_info\": \"请求详情\",\n\t\t\t\"llm.openai.post.api_error\": \"OpenAI API 错误响应\",\n\n\t\t\t// LLM: handlers/stream.go (general LLM stream handler)\n\t\t\t\"llm.handlers.stream.info\": \"LLM 流式输出\",\n\t\t\t\"llm.handlers.stream.raw_output\": \"LLM 原始输出\",\n\n\t\t\t// Output: adapters/openai/writer.go\n\t\t\t\"output.openai.writer.sending_chunk\": \"向客户端发送数据块\",\n\t\t\t\"output.openai.writer.sending_done\": \"向客户端发送 [DONE]\",\n\t\t\t\"output.openai.writer.adapt_error\": \"适配消息失败\",\n\t\t\t\"output.openai.writer.chunk_error\": \"发送数据块失败\",\n\t\t\t\"output.openai.writer.group_error\": \"写入消息组中的消息失败\",\n\t\t\t\"output.openai.writer.send_error\": \"发送数据到客户端失败\",\n\t\t\t\"output.openai.writer.marshal_error\": \"序列化数据块失败\",\n\t\t\t\"output.openai.writer.done_error\": \"发送 [DONE] 到客户端失败\",\n\n\t\t\t// Output: adapters/cui/writer.go\n\t\t\t\"output.cui.writer.sending_chunk\": \"向客户端发送数据块\",\n\t\t\t\"output.cui.writer.adapt_error\": \"适配消息失败\",\n\t\t\t\"output.cui.writer.chunk_error\": \"发送数据块失败\",\n\t\t\t\"output.cui.writer.group_error\": \"发送消息组失败\",\n\t\t\t\"output.cui.writer.send_error\": \"发送数据到客户端失败\",\n\t\t\t\"output.cui.writer.marshal_error\": \"序列化数据块失败\",\n\n\t\t\t// Output: Stream event messages\n\t\t\t\"output.stream_start\": \"智能体正在处理\",\n\t\t\t\"output.view_trace\": \"查看处理详情\",\n\n\t\t\t// Common status messages\n\t\t\t\"common.status.processing\": \"处理中\",\n\t\t\t\"common.status.completed\": \"已完成\",\n\t\t\t\"common.status.failed\": \"失败\",\n\t\t\t\"common.status.retrying\": \"重试中\",\n\n\t\t\t// MCP: context/mcp.go - Resource operations\n\t\t\t\"mcp.list_resources.label\": \"MCP: 列出资源\",\n\t\t\t\"mcp.list_resources.description\": \"从 MCP 客户端 '%s' 列出资源\",\n\t\t\t\"mcp.read_resource.label\": \"MCP: 读取资源\",\n\t\t\t\"mcp.read_resource.description\": \"从 MCP 客户端 '%s' 读取资源 '%s'\",\n\n\t\t\t// MCP: context/mcp.go - Tool operations\n\t\t\t\"mcp.list_tools.label\": \"MCP: 列出工具\",\n\t\t\t\"mcp.list_tools.description\": \"从 MCP 客户端 '%s' 列出工具\",\n\t\t\t\"mcp.call_tool.label\": \"MCP: 调用工具\",\n\t\t\t\"mcp.call_tool.description\": \"从 MCP 客户端 '%s' 调用工具 '%s'\",\n\t\t\t\"mcp.call_tools.label\": \"MCP: 调用工具\",\n\t\t\t\"mcp.call_tools.description\": \"从 MCP 客户端 '%s' 顺序调用 %d 个工具\",\n\t\t\t\"mcp.call_tools_parallel.label\": \"MCP: 调用工具（并行）\",\n\t\t\t\"mcp.call_tools_parallel.description\": \"从 MCP 客户端 '%s' 并行调用 %d 个工具\",\n\n\t\t\t// MCP: context/mcp.go - Prompt operations\n\t\t\t\"mcp.list_prompts.label\": \"MCP: 列出提示词\",\n\t\t\t\"mcp.list_prompts.description\": \"从 MCP 客户端 '%s' 列出提示词\",\n\t\t\t\"mcp.get_prompt.label\": \"MCP: 获取提示词\",\n\t\t\t\"mcp.get_prompt.description\": \"从 MCP 客户端 '%s' 获取提示词 '%s'\",\n\n\t\t\t// MCP: context/mcp.go - Sample operations\n\t\t\t\"mcp.list_samples.label\": \"MCP: 列出示例\",\n\t\t\t\"mcp.list_samples.description\": \"从 MCP 客户端 '%s' 列出 '%s' 的示例\",\n\t\t\t\"mcp.get_sample.label\": \"MCP: 获取示例\",\n\t\t\t\"mcp.get_sample.description\": \"从 MCP 客户端 '%s' 获取 '%s' 的第 %d 个示例\",\n\n\t\t\t// KB: Chat collection\n\t\t\t\"kb.chat.name\": \"聊天知识库\",\n\t\t\t\"kb.chat.description\": \"自动为聊天会话创建的知识库集合\",\n\n\t\t\t// Sandbox: assistant/sandbox.go - Sandbox status messages\n\t\t\t\"sandbox.preparing\": \"正在准备...\",\n\t\t\t\"sandbox.ready\": \"就绪\",\n\t\t\t\"sandbox.working\": \"正在处理您的请求\",\n\t\t\t\"sandbox.completed\": \"处理完成\",\n\t\t\t\"sandbox.failed\": \"执行失败\",\n\t\t\t\"sandbox.starting\": \"正在启动工作区...\",\n\t\t\t\"sandbox.pulling_image\": \"正在准备运行环境（首次可能需要一点时间）\",\n\t\t\t\"sandbox.waiting_response\": \"等待 AI 响应...\",\n\n\t\t\t// Sandbox: claude/executor.go - Tool execution messages\n\t\t\t\"sandbox.tool.read\": \"正在读取文件\",\n\t\t\t\"sandbox.tool.write\": \"正在写入文件\",\n\t\t\t\"sandbox.tool.edit\": \"正在编辑文件\",\n\t\t\t\"sandbox.tool.bash\": \"正在执行命令\",\n\t\t\t\"sandbox.tool.glob\": \"正在查找文件\",\n\t\t\t\"sandbox.tool.grep\": \"正在搜索代码\",\n\t\t\t\"sandbox.tool.ls\": \"正在列出目录\",\n\t\t\t\"sandbox.tool.task\": \"正在执行子任务\",\n\t\t\t\"sandbox.tool.web_search\": \"正在搜索网页\",\n\t\t\t\"sandbox.tool.web_fetch\": \"正在获取网页\",\n\t\t\t\"sandbox.tool.todo_write\": \"正在管理任务\",\n\t\t\t\"sandbox.tool.ask_question\": \"正在询问问题\",\n\t\t\t\"sandbox.tool.switch_mode\": \"正在切换模式\",\n\t\t\t\"sandbox.tool.read_lints\": \"正在检查代码\",\n\t\t\t\"sandbox.tool.edit_notebook\": \"正在编辑笔记本\",\n\t\t\t\"sandbox.tool.unknown\": \"正在执行 {{name}}\",\n\n\t\t\t// Content: content/image/image.go - Image processing messages\n\t\t\t\"content.image.analyzing\": \"正在分析图片\",\n\n\t\t\t// Content: content/pdf/pdf.go - PDF processing messages\n\t\t\t\"content.pdf.analyzing_page\": \"正在分析 PDF 第 %d/%d 页\",\n\n\t\t\t// Search: assistant/search.go - Output messages\n\t\t\t\"search.loading\": \"正在搜索\",\n\t\t\t\"search.success\": \"找到 %d 条参考资料\",\n\t\t\t\"search.success.one\": \"找到 1 条参考资料\",\n\t\t\t\"search.partial\": \"找到 %d 条参考资料（部分来源失败）\",\n\t\t\t\"search.failed\": \"搜索失败\",\n\t\t\t\"search.no_results\": \"未找到相关资料\",\n\n\t\t\t// Search Intent: assistant/search.go - Intent detection messages\n\t\t\t\"search.intent.loading\": \"检查是否需要查询资料\",\n\t\t\t\"search.intent.need_search\": \"正在查询相关资料\",\n\t\t\t\"search.intent.no_search\": \"无需查询资料\",\n\n\t\t\t// Keyword Extraction: assistant/search.go - Keyword extraction messages\n\t\t\t\"search.keyword.loading\": \"正在分析对话内容\",\n\t\t\t\"search.keyword.done\": \"分析完成\",\n\n\t\t\t// Search: assistant/search.go - Trace labels\n\t\t\t\"search.trace.label\": \"搜索\",\n\t\t\t\"search.trace.description\": \"搜索网络和知识库获取相关信息\",\n\t\t\t\"search.trace.web.label\": \"网页搜索\",\n\t\t\t\"search.trace.web.description\": \"搜索网页获取相关信息\",\n\t\t\t\"search.trace.kb.label\": \"知识库搜索\",\n\t\t\t\"search.trace.kb.description\": \"搜索知识库获取相关信息\",\n\t\t\t\"search.trace.db.label\": \"数据库搜索\",\n\t\t\t\"search.trace.db.description\": \"搜索数据库获取相关信息\",\n\t\t},\n\t}\n}\n" }, { "path": "agent/i18n/i18n.go", "content": "package i18n\n\nimport (\n\t\"path/filepath\"\n\t\"regexp\"\n\t\"strings\"\n\n\t\"github.com/yaoapp/gou/application\"\n\t\"github.com/yaoapp/gou/fs\"\n\t\"github.com/yaoapp/kun/maps\"\n)\n\n// Locales the locales\nvar Locales = map[string]map[string]I18n{}\n\n// I18n the i18n struct\ntype I18n struct {\n\tLocale string `json:\"locale,omitempty\" yaml:\"locale,omitempty\"`\n\tMessages map[string]any `json:\"messages,omitempty\" yaml:\"messages,omitempty\"`\n}\n\n// Map the i18n map\ntype Map map[string]I18n\n\n// Parse parse the input\nfunc (i18n I18n) Parse(input any) any {\n\tif input == nil {\n\t\treturn nil\n\t}\n\n\tswitch in := input.(type) {\n\tcase string:\n\t\treturn i18n.parseString(in)\n\n\tcase map[string]any:\n\t\tnew := map[string]any{}\n\t\tfor key, value := range in {\n\t\t\tnew[key] = i18n.Parse(value)\n\t\t}\n\t\treturn new\n\n\tcase []any:\n\t\tnew := []any{}\n\t\tfor _, value := range in {\n\t\t\tnew = append(new, i18n.Parse(value))\n\t\t}\n\t\treturn new\n\n\tcase []string:\n\t\tnew := []string{}\n\t\tfor _, value := range in {\n\t\t\tif parsed := i18n.Parse(value); parsed != nil {\n\t\t\t\tif s, ok := parsed.(string); ok {\n\t\t\t\t\tnew = append(new, s)\n\t\t\t\t} else {\n\t\t\t\t\tnew = append(new, value)\n\t\t\t\t}\n\t\t\t} else {\n\t\t\t\tnew = append(new, value)\n\t\t\t}\n\t\t}\n\t\treturn new\n\t}\n\n\treturn input\n}\n\n// parseString parse a string value\nfunc (i18n I18n) parseString(in string) string {\n\ttrimed := strings.TrimSpace(in)\n\n\t// Check if it's a direct message key (no template markers)\n\tif !strings.Contains(trimed, \"{{\") && !strings.Contains(trimed, \"}}\") {\n\t\tif val, ok := i18n.Messages[trimed]; ok {\n\t\t\tif s, ok := val.(string); ok {\n\t\t\t\treturn s\n\t\t\t}\n\t\t}\n\t\treturn in\n\t}\n\n\t// Check if it's a full template expression {{...}} (exact match - entire string is one template)\n\thasExp := strings.HasPrefix(trimed, \"{{\") && strings.HasSuffix(trimed, \"}}\")\n\tif hasExp {\n\t\t// Check if there's only ONE template pattern (no text before/after or multiple templates)\n\t\tre := regexp.MustCompile(`\\{\\{\\s*([^}]+?)\\s*\\}\\}`)\n\t\tmatches := re.FindAllString(trimed, -1)\n\n\t\t// Only treat as full template if there's exactly one match and it equals the trimmed string\n\t\tif len(matches) == 1 && matches[0] == trimed {\n\t\t\texp := strings.TrimSpace(strings.TrimPrefix(strings.TrimSuffix(trimed, \"}}\"), \"{{\"))\n\t\t\tif val, ok := i18n.Messages[exp]; ok {\n\t\t\t\tif s, ok := val.(string); ok {\n\t\t\t\t\treturn s\n\t\t\t\t}\n\t\t\t}\n\t\t\treturn in\n\t\t}\n\t}\n\n\t// Handle embedded template variables: \"text {{var}} more {{var2}}\"\n\tif strings.Contains(in, \"{{\") && strings.Contains(in, \"}}\") {\n\t\tresult := in\n\t\t// Use regex to find all {{...}} patterns\n\t\tre := regexp.MustCompile(`\\{\\{\\s*([^}]+?)\\s*\\}\\}`)\n\t\tmatches := re.FindAllStringSubmatch(in, -1)\n\n\t\tfor _, match := range matches {\n\t\t\tif len(match) >= 2 {\n\t\t\t\tfullMatch := match[0] // Full match including {{ }}\n\t\t\t\tvarName := strings.TrimSpace(match[1]) // Variable name without {{ }}\n\n\t\t\t\t// Try to replace with value from Messages\n\t\t\t\tif val, ok := i18n.Messages[varName]; ok {\n\t\t\t\t\tif s, ok := val.(string); ok {\n\t\t\t\t\t\tresult = strings.Replace(result, fullMatch, s, 1)\n\t\t\t\t\t}\n\t\t\t\t}\n\t\t\t}\n\t\t}\n\t\treturn result\n\t}\n\n\treturn in\n}\n\n// GetLocales get the locales from path\nfunc GetLocales(path string) (Map, error) {\n\tapp, err := fs.Get(\"app\")\n\tif err != nil {\n\t\treturn nil, err\n\t}\n\n\ti18ns := Map{}\n\tlocalesdir := filepath.Join(path, \"locales\")\n\tif has, _ := app.Exists(localesdir); has {\n\t\tlocales, err := app.ReadDir(localesdir, true)\n\t\tif err != nil {\n\t\t\treturn nil, err\n\t\t}\n\n\t\t// load locales\n\t\tfor _, locale := range locales {\n\t\t\tlocaleData, err := app.ReadFile(locale)\n\t\t\tif err != nil {\n\t\t\t\treturn nil, err\n\t\t\t}\n\t\t\tvar messages maps.Map = map[string]any{}\n\t\t\terr = application.Parse(locale, localeData, &messages)\n\t\t\tif err != nil {\n\t\t\t\treturn nil, err\n\t\t\t}\n\t\t\tname := strings.ToLower(strings.TrimSuffix(filepath.Base(locale), \".yml\"))\n\t\t\ti18ns[name] = I18n{Locale: name, Messages: messages}\n\t\t}\n\t}\n\treturn i18ns, nil\n}\n\n// Flatten flattens the map of locales by adding short language codes and region codes\n// e.g., \"en-us\" will also create \"en\" and \"us\" entries\n// If __global__ locales exist, they are merged (local/user messages override global built-in messages)\nfunc (m Map) Flatten() Map {\n\tflattened := make(Map)\n\n\t// First, process local messages with Dot() flattening\n\tfor localeCode, i18n := range m {\n\t\t// Flatten nested messages to dot notation (e.g., {\"local\": {\"key\": \"value\"}} -> {\"local.key\": \"value\"})\n\t\tflattened[localeCode] = I18n{\n\t\t\tLocale: localeCode,\n\t\t\tMessages: maps.MapOf(i18n.Messages).Dot(),\n\t\t}\n\n\t\t// Add short language codes\n\t\tparts := strings.Split(localeCode, \"-\")\n\t\tif len(parts) > 1 {\n\t\t\t// Add short language code (e.g., \"en\" from \"en-us\")\n\t\t\tif _, ok := flattened[parts[0]]; !ok {\n\t\t\t\tflattened[parts[0]] = flattened[localeCode]\n\t\t\t}\n\t\t\t// Add region code (e.g., \"us\" from \"en-us\")\n\t\t\tif _, ok := flattened[parts[1]]; !ok {\n\t\t\t\tflattened[parts[1]] = flattened[localeCode]\n\t\t\t}\n\t\t}\n\t}\n\n\t// Merge with global locales if they exist\n\t// Strategy: Start with global (built-in), then override with local (user)\n\tglobalLocales, hasGlobal := Locales[\"__global__\"]\n\tif !hasGlobal {\n\t\treturn flattened\n\t}\n\n\tfor globalLocaleCode, globalI18n := range globalLocales {\n\t\t// Ensure global messages are also flattened (though builtin.go already uses flat keys)\n\t\tglobalFlattened := maps.MapOf(globalI18n.Messages).Dot()\n\n\t\tif localI18n, ok := flattened[globalLocaleCode]; ok {\n\t\t\t// Both global and local exist: merge with local overriding global\n\t\t\tmergedMessages := make(map[string]any)\n\t\t\t// First copy all global messages\n\t\t\tfor k, v := range globalFlattened {\n\t\t\t\tmergedMessages[k] = v\n\t\t\t}\n\t\t\t// Then override with local messages\n\t\t\tfor k, v := range localI18n.Messages {\n\t\t\t\tmergedMessages[k] = v\n\t\t\t}\n\t\t\tflattened[globalLocaleCode] = I18n{\n\t\t\t\tLocale: globalLocaleCode,\n\t\t\t\tMessages: mergedMessages,\n\t\t\t}\n\t\t} else {\n\t\t\t// Only global exists, add it with flattened messages\n\t\t\tflattened[globalLocaleCode] = I18n{\n\t\t\t\tLocale: globalLocaleCode,\n\t\t\t\tMessages: globalFlattened,\n\t\t\t}\n\t\t}\n\t}\n\n\treturn flattened\n}\n\n// FlattenWithGlobal is deprecated. Use Flatten() instead, which now automatically merges with global locales.\n// Kept for backward compatibility.\nfunc (m Map) FlattenWithGlobal() Map {\n\treturn m.Flatten()\n}\n\n// Translate translate the input with recursive variable resolution\n// Fallback strategy: assistant locale -> assistant short codes -> global locale -> global short codes\nfunc Translate(assistantID string, locale string, input any) any {\n\n\tlocale = strings.ToLower(strings.TrimSpace(locale))\n\n\t// Helper function to try translation with a specific i18n object\n\ttryTranslate := func(i18n I18n, input any) (any, bool) {\n\t\tresult := i18n.Parse(input)\n\t\t// For string input, check if translation was found by comparing with input\n\t\t// For other types, Parse always returns a result (transformed or original)\n\t\tif inputStr, ok := input.(string); ok {\n\t\t\tif resultStr, ok := result.(string); ok {\n\t\t\t\t// Translation found if result is different from input\n\t\t\t\tif resultStr != inputStr {\n\t\t\t\t\treturn result, true\n\t\t\t\t}\n\t\t\t\treturn input, false\n\t\t\t}\n\t\t}\n\t\t// For non-string inputs (maps, slices), Parse always processes them\n\t\treturn result, true\n\t}\n\n\t// Helper function to process recursive templates\n\tprocessTemplates := func(result any, assistantID string, locale string) any {\n\t\tif resultStr, ok := result.(string); ok && strings.Contains(resultStr, \"{{\") && strings.Contains(resultStr, \"}}\") {\n\t\t\tre := regexp.MustCompile(`\\{\\{\\s*([^}]+?)\\s*\\}\\}`)\n\t\t\tresultStr = re.ReplaceAllStringFunc(resultStr, func(match string) string {\n\t\t\t\tvarName := strings.TrimSpace(strings.TrimPrefix(strings.TrimSuffix(match, \"}}\"), \"{{\"))\n\t\t\t\ttranslated := Translate(assistantID, locale, varName)\n\t\t\t\tif translatedStr, ok := translated.(string); ok && translatedStr != varName {\n\t\t\t\t\treturn translatedStr\n\t\t\t\t}\n\t\t\t\treturn match\n\t\t\t})\n\t\t\treturn resultStr\n\t\t}\n\t\treturn result\n\t}\n\n\t// Try assistant locale first\n\tif i18ns, has := Locales[assistantID]; has {\n\t\t// Try exact locale\n\t\tif i18n, hasLocale := i18ns[locale]; hasLocale {\n\t\t\tif result, found := tryTranslate(i18n, input); found {\n\t\t\t\treturn processTemplates(result, assistantID, locale)\n\t\t\t}\n\t\t}\n\n\t\t// Try short codes\n\t\tparts := strings.Split(locale, \"-\")\n\t\tif len(parts) > 1 {\n\t\t\tif i18n, hasLocale := i18ns[parts[1]]; hasLocale {\n\t\t\t\tif result, found := tryTranslate(i18n, input); found {\n\t\t\t\t\treturn processTemplates(result, assistantID, locale)\n\t\t\t\t}\n\t\t\t}\n\t\t\tif i18n, hasLocale := i18ns[parts[0]]; hasLocale {\n\t\t\t\tif result, found := tryTranslate(i18n, input); found {\n\t\t\t\t\treturn processTemplates(result, assistantID, locale)\n\t\t\t\t}\n\t\t\t}\n\t\t}\n\t}\n\n\t// Fallback to global locales\n\tif globalI18ns, hasGlobal := Locales[\"__global__\"]; hasGlobal {\n\t\t// Try exact locale\n\t\tif i18n, hasLocale := globalI18ns[locale]; hasLocale {\n\t\t\tif result, found := tryTranslate(i18n, input); found {\n\t\t\t\treturn processTemplates(result, assistantID, locale)\n\t\t\t}\n\t\t}\n\n\t\t// Try short codes\n\t\tparts := strings.Split(locale, \"-\")\n\t\tif len(parts) > 1 {\n\t\t\tif i18n, hasLocale := globalI18ns[parts[1]]; hasLocale {\n\t\t\t\tif result, found := tryTranslate(i18n, input); found {\n\t\t\t\t\treturn processTemplates(result, assistantID, locale)\n\t\t\t\t}\n\t\t\t}\n\t\t\tif i18n, hasLocale := globalI18ns[parts[0]]; hasLocale {\n\t\t\t\tif result, found := tryTranslate(i18n, input); found {\n\t\t\t\t\treturn processTemplates(result, assistantID, locale)\n\t\t\t\t}\n\t\t\t}\n\t\t}\n\t}\n\n\treturn input\n}\n\n// TranslateGlobal translate the input with global i18n\nfunc TranslateGlobal(locale string, input any) any {\n\tlocale = strings.ToLower(strings.TrimSpace(locale))\n\ti18ns, has := Locales[\"__global__\"]\n\tif !has {\n\t\ti18ns = map[string]I18n{}\n\t}\n\n\t// Try the exact locale first\n\ti18n, has := i18ns[locale]\n\tif has {\n\t\tresult := i18n.Parse(input)\n\t\t// If the result is the same as input (not translated), try fallback\n\t\tif result != input {\n\t\t\treturn result\n\t\t}\n\t}\n\n\t// Fallback logic: for \"en-us\", try \"en\"\n\tparts := strings.Split(locale, \"-\")\n\tif len(parts) > 1 {\n\t\t// Try the language code (e.g., \"en\" for \"en-us\")\n\t\tif fallbackI18n, hasFallback := i18ns[parts[0]]; hasFallback {\n\t\t\tresult := fallbackI18n.Parse(input)\n\t\t\tif result != input {\n\t\t\t\treturn result\n\t\t\t}\n\t\t}\n\t\t// Try the country code (e.g., \"us\" for \"en-us\")\n\t\tif fallbackI18n, hasFallback := i18ns[parts[1]]; hasFallback {\n\t\t\tresult := fallbackI18n.Parse(input)\n\t\t\tif result != input {\n\t\t\t\treturn result\n\t\t\t}\n\t\t}\n\t}\n\n\treturn input\n}\n\n// T is a short alias for TranslateGlobal that returns string\n// Usage: i18n.T(ctx.Locale, \"assistant.agent.stream.label\")\n// Variables in templates like {{variable}} will be recursively resolved from the global language pack\nfunc T(locale string, key string) string {\n\tresult := TranslateGlobal(locale, key)\n\tif str, ok := result.(string); ok {\n\t\treturn str\n\t}\n\treturn key\n}\n\n// Tr translates with assistantID and returns string\n// Supports recursive translation of {{variable}} templates\n// Usage: i18n.Tr(assistantID, locale, \"key\")\nfunc Tr(assistantID string, locale string, key string) string {\n\tresult := Translate(assistantID, locale, key)\n\tif str, ok := result.(string); ok {\n\t\treturn str\n\t}\n\treturn key\n}\n" }, { "path": "agent/i18n/i18n_test.go", "content": "package i18n\n\nimport (\n\t\"testing\"\n\n\t\"github.com/yaoapp/yao/config\"\n\t\"github.com/yaoapp/yao/test\"\n)\n\n// TestParseString tests the parseString method\nfunc TestParseString(t *testing.T) {\n\ti18n := I18n{\n\t\tLocale: \"en\",\n\t\tMessages: map[string]any{\n\t\t\t\"hello\": \"Hello\",\n\t\t\t\"world\": \"World\",\n\t\t\t\"greeting\": \"Hello, World!\",\n\t\t\t\"description\": \"This is a test\",\n\t\t},\n\t}\n\n\ttests := []struct {\n\t\tname string\n\t\tinput string\n\t\texpected string\n\t}{\n\t\t{\n\t\t\tname: \"Template expression with match\",\n\t\t\tinput: \"{{greeting}}\",\n\t\t\texpected: \"Hello, World!\",\n\t\t},\n\t\t{\n\t\t\tname: \"Template expression with spaces\",\n\t\t\tinput: \"{{ greeting }}\",\n\t\t\texpected: \"Hello, World!\",\n\t\t},\n\t\t{\n\t\t\tname: \"Template expression without match\",\n\t\t\tinput: \"{{notfound}}\",\n\t\t\texpected: \"{{notfound}}\",\n\t\t},\n\t\t{\n\t\t\tname: \"Direct message key\",\n\t\t\tinput: \"hello\",\n\t\t\texpected: \"Hello\",\n\t\t},\n\t\t{\n\t\t\tname: \"Direct message key with spaces\",\n\t\t\tinput: \" world \",\n\t\t\texpected: \"World\",\n\t\t},\n\t\t{\n\t\t\tname: \"Non-existent key\",\n\t\t\tinput: \"notfound\",\n\t\t\texpected: \"notfound\",\n\t\t},\n\t\t{\n\t\t\tname: \"Regular text\",\n\t\t\tinput: \"Just some text\",\n\t\t\texpected: \"Just some text\",\n\t\t},\n\t\t{\n\t\t\tname: \"Empty string\",\n\t\t\tinput: \"\",\n\t\t\texpected: \"\",\n\t\t},\n\t\t// Embedded template tests (new feature)\n\t\t{\n\t\t\tname: \"Embedded single template\",\n\t\t\tinput: \"Hello {{hello}}\",\n\t\t\texpected: \"Hello Hello\",\n\t\t},\n\t\t{\n\t\t\tname: \"Embedded multiple templates\",\n\t\t\tinput: \"{{hello}} {{world}}!\",\n\t\t\texpected: \"Hello World!\",\n\t\t},\n\t\t{\n\t\t\tname: \"Embedded template with spaces\",\n\t\t\tinput: \"Say {{ hello }} to the {{ world }}\",\n\t\t\texpected: \"Say Hello to the World\",\n\t\t},\n\t\t{\n\t\t\tname: \"Embedded template mixed with text\",\n\t\t\tinput: \"Message: {{greeting}} - {{description}}\",\n\t\t\texpected: \"Message: Hello, World! - This is a test\",\n\t\t},\n\t\t{\n\t\t\tname: \"Embedded template not found\",\n\t\t\tinput: \"Hello {{notfound}} World\",\n\t\t\texpected: \"Hello {{notfound}} World\",\n\t\t},\n\t\t{\n\t\t\tname: \"Embedded template partial match\",\n\t\t\tinput: \"{{hello}} {{notfound}} {{world}}\",\n\t\t\texpected: \"Hello {{notfound}} World\",\n\t\t},\n\t}\n\n\tfor _, tt := range tests {\n\t\tt.Run(tt.name, func(t *testing.T) {\n\t\t\tresult := i18n.parseString(tt.input)\n\t\t\tif result != tt.expected {\n\t\t\t\tt.Errorf(\"parseString(%q) = %q, want %q\", tt.input, result, tt.expected)\n\t\t\t}\n\t\t})\n\t}\n}\n\n// TestParseStringNonStringValue tests parseString when message value is not a string\nfunc TestParseStringNonStringValue(t *testing.T) {\n\ti18n := I18n{\n\t\tLocale: \"en\",\n\t\tMessages: map[string]any{\n\t\t\t\"number\": 123,\n\t\t\t\"object\": map[string]any{\"key\": \"value\"},\n\t\t},\n\t}\n\n\ttests := []struct {\n\t\tname string\n\t\tinput string\n\t\texpected string\n\t}{\n\t\t{\n\t\t\tname: \"Template with number value\",\n\t\t\tinput: \"{{number}}\",\n\t\t\texpected: \"{{number}}\",\n\t\t},\n\t\t{\n\t\t\tname: \"Direct key with number value\",\n\t\t\tinput: \"number\",\n\t\t\texpected: \"number\",\n\t\t},\n\t\t{\n\t\t\tname: \"Template with object value\",\n\t\t\tinput: \"{{object}}\",\n\t\t\texpected: \"{{object}}\",\n\t\t},\n\t}\n\n\tfor _, tt := range tests {\n\t\tt.Run(tt.name, func(t *testing.T) {\n\t\t\tresult := i18n.parseString(tt.input)\n\t\t\tif result != tt.expected {\n\t\t\t\tt.Errorf(\"parseString(%q) = %q, want %q\", tt.input, result, tt.expected)\n\t\t\t}\n\t\t})\n\t}\n}\n\n// TestParse tests the Parse method with various input types\nfunc TestParse(t *testing.T) {\n\ti18n := I18n{\n\t\tLocale: \"en\",\n\t\tMessages: map[string]any{\n\t\t\t\"name\": \"John\",\n\t\t\t\"description\": \"A developer\",\n\t\t\t\"title\": \"Welcome\",\n\t\t},\n\t}\n\n\tt.Run(\"Nil input\", func(t *testing.T) {\n\t\tresult := i18n.Parse(nil)\n\t\tif result != nil {\n\t\t\tt.Errorf(\"Parse(nil) = %v, want nil\", result)\n\t\t}\n\t})\n\n\tt.Run(\"String input\", func(t *testing.T) {\n\t\tresult := i18n.Parse(\"{{name}}\")\n\t\tif result != \"John\" {\n\t\t\tt.Errorf(\"Parse({{name}}) = %v, want 'John'\", result)\n\t\t}\n\t})\n\n\tt.Run(\"Map input\", func(t *testing.T) {\n\t\tinput := map[string]any{\n\t\t\t\"name\": \"{{name}}\",\n\t\t\t\"description\": \"{{description}}\",\n\t\t\t\"age\": 30,\n\t\t}\n\t\tresult := i18n.Parse(input)\n\t\tif resultMap, ok := result.(map[string]any); ok {\n\t\t\tif resultMap[\"name\"] != \"John\" {\n\t\t\t\tt.Errorf(\"Expected name 'John', got %v\", resultMap[\"name\"])\n\t\t\t}\n\t\t\tif resultMap[\"description\"] != \"A developer\" {\n\t\t\t\tt.Errorf(\"Expected description 'A developer', got %v\", resultMap[\"description\"])\n\t\t\t}\n\t\t\tif resultMap[\"age\"] != 30 {\n\t\t\t\tt.Errorf(\"Expected age 30, got %v\", resultMap[\"age\"])\n\t\t\t}\n\t\t} else {\n\t\t\tt.Errorf(\"Expected map[string]any, got %T\", result)\n\t\t}\n\t})\n\n\tt.Run(\"Slice of any\", func(t *testing.T) {\n\t\tinput := []any{\"{{name}}\", \"{{description}}\", 123}\n\t\tresult := i18n.Parse(input)\n\t\tif resultSlice, ok := result.([]any); ok {\n\t\t\tif len(resultSlice) != 3 {\n\t\t\t\tt.Errorf(\"Expected 3 elements, got %d\", len(resultSlice))\n\t\t\t}\n\t\t\tif resultSlice[0] != \"John\" {\n\t\t\t\tt.Errorf(\"Expected 'John', got %v\", resultSlice[0])\n\t\t\t}\n\t\t\tif resultSlice[1] != \"A developer\" {\n\t\t\t\tt.Errorf(\"Expected 'A developer', got %v\", resultSlice[1])\n\t\t\t}\n\t\t\tif resultSlice[2] != 123 {\n\t\t\t\tt.Errorf(\"Expected 123, got %v\", resultSlice[2])\n\t\t\t}\n\t\t} else {\n\t\t\tt.Errorf(\"Expected []any, got %T\", result)\n\t\t}\n\t})\n\n\tt.Run(\"Slice of strings\", func(t *testing.T) {\n\t\tinput := []string{\"{{name}}\", \"{{description}}\", \"plain text\"}\n\t\tresult := i18n.Parse(input)\n\t\tif resultSlice, ok := result.([]string); ok {\n\t\t\tif len(resultSlice) != 3 {\n\t\t\t\tt.Errorf(\"Expected 3 elements, got %d\", len(resultSlice))\n\t\t\t}\n\t\t\tif resultSlice[0] != \"John\" {\n\t\t\t\tt.Errorf(\"Expected 'John', got %v\", resultSlice[0])\n\t\t\t}\n\t\t\tif resultSlice[1] != \"A developer\" {\n\t\t\t\tt.Errorf(\"Expected 'A developer', got %v\", resultSlice[1])\n\t\t\t}\n\t\t\tif resultSlice[2] != \"plain text\" {\n\t\t\t\tt.Errorf(\"Expected 'plain text', got %v\", resultSlice[2])\n\t\t\t}\n\t\t} else {\n\t\t\tt.Errorf(\"Expected []string, got %T\", result)\n\t\t}\n\t})\n\n\tt.Run(\"Nested structures\", func(t *testing.T) {\n\t\tinput := map[string]any{\n\t\t\t\"user\": map[string]any{\n\t\t\t\t\"name\": \"{{name}}\",\n\t\t\t\t\"info\": []string{\"{{title}}\", \"{{description}}\"},\n\t\t\t},\n\t\t}\n\t\tresult := i18n.Parse(input)\n\t\tif resultMap, ok := result.(map[string]any); ok {\n\t\t\tif userMap, ok := resultMap[\"user\"].(map[string]any); ok {\n\t\t\t\tif userMap[\"name\"] != \"John\" {\n\t\t\t\t\tt.Errorf(\"Expected nested name 'John', got %v\", userMap[\"name\"])\n\t\t\t\t}\n\t\t\t\tif infoSlice, ok := userMap[\"info\"].([]any); ok {\n\t\t\t\t\tif infoSlice[0] != \"Welcome\" {\n\t\t\t\t\t\tt.Errorf(\"Expected 'Welcome', got %v\", infoSlice[0])\n\t\t\t\t\t}\n\t\t\t\t}\n\t\t\t}\n\t\t}\n\t})\n\n\tt.Run(\"Other types pass through\", func(t *testing.T) {\n\t\tinput := 12345\n\t\tresult := i18n.Parse(input)\n\t\tif result != input {\n\t\t\tt.Errorf(\"Expected %v, got %v\", input, result)\n\t\t}\n\t})\n}\n\n// TestParseSliceStringWithNilAndNonString tests []string parsing edge cases\nfunc TestParseSliceStringWithNilAndNonString(t *testing.T) {\n\ti18n := I18n{\n\t\tLocale: \"en\",\n\t\tMessages: map[string]any{\n\t\t\t\"key1\": \"value1\",\n\t\t\t\"key2\": 123, // Non-string value\n\t\t\t\"key3\": nil, // Nil value\n\t\t},\n\t}\n\n\tt.Run(\"String slice with fallback\", func(t *testing.T) {\n\t\tinput := []string{\"{{key1}}\", \"{{key2}}\", \"{{notfound}}\"}\n\t\tresult := i18n.Parse(input)\n\t\tif resultSlice, ok := result.([]string); ok {\n\t\t\tif resultSlice[0] != \"value1\" {\n\t\t\t\tt.Errorf(\"Expected 'value1', got %v\", resultSlice[0])\n\t\t\t}\n\t\t\t// key2 has non-string value, should fallback to original\n\t\t\tif resultSlice[1] != \"{{key2}}\" {\n\t\t\t\tt.Errorf(\"Expected '{{key2}}', got %v\", resultSlice[1])\n\t\t\t}\n\t\t\tif resultSlice[2] != \"{{notfound}}\" {\n\t\t\t\tt.Errorf(\"Expected '{{notfound}}', got %v\", resultSlice[2])\n\t\t\t}\n\t\t} else {\n\t\t\tt.Errorf(\"Expected []string, got %T\", result)\n\t\t}\n\t})\n\n\tt.Run(\"String slice with nil parsed result\", func(t *testing.T) {\n\t\t// This tests the case where Parse returns nil for a string\n\t\tinput := []string{\"{{key3}}\", \"normal\"}\n\t\tresult := i18n.Parse(input)\n\t\tif resultSlice, ok := result.([]string); ok {\n\t\t\t// When parsed is nil, should fallback to original\n\t\t\tif resultSlice[0] != \"{{key3}}\" {\n\t\t\t\tt.Errorf(\"Expected '{{key3}}' (tests nil parsed branch), got %v\", resultSlice[0])\n\t\t\t}\n\t\t\tif resultSlice[1] != \"normal\" {\n\t\t\t\tt.Errorf(\"Expected 'normal', got %v\", resultSlice[1])\n\t\t\t}\n\t\t} else {\n\t\t\tt.Errorf(\"Expected []string, got %T\", result)\n\t\t}\n\t})\n\n\tt.Run(\"String slice with non-string parsed result from map\", func(t *testing.T) {\n\t\ti18nWithMap := I18n{\n\t\t\tLocale: \"en\",\n\t\t\tMessages: map[string]any{\n\t\t\t\t\"map_key\": map[string]any{\"nested\": \"value\"},\n\t\t\t},\n\t\t}\n\t\t// When Parse returns a non-string type (like a map), should fallback\n\t\tinput := []string{\"{{map_key}}\", \"text\"}\n\t\tresult := i18nWithMap.Parse(input)\n\t\tif resultSlice, ok := result.([]string); ok {\n\t\t\t// Should fallback to original when parsed is not string\n\t\t\tif resultSlice[0] != \"{{map_key}}\" {\n\t\t\t\tt.Errorf(\"Expected '{{map_key}}' (tests non-string parsed branch), got %v\", resultSlice[0])\n\t\t\t}\n\t\t} else {\n\t\t\tt.Errorf(\"Expected []string, got %T\", result)\n\t\t}\n\t})\n}\n\n// TestMapFlatten tests the Flatten method\nfunc TestMapFlatten(t *testing.T) {\n\ti18ns := Map{\n\t\t\"en-us\": I18n{\n\t\t\tLocale: \"en-us\",\n\t\t\tMessages: map[string]any{\n\t\t\t\t\"greeting\": \"Hello\",\n\t\t\t},\n\t\t},\n\t\t\"zh-cn\": I18n{\n\t\t\tLocale: \"zh-cn\",\n\t\t\tMessages: map[string]any{\n\t\t\t\t\"greeting\": \"你好\",\n\t\t\t},\n\t\t},\n\t}\n\n\tflattened := i18ns.Flatten()\n\n\t// Should have original keys\n\tif _, ok := flattened[\"en-us\"]; !ok {\n\t\tt.Error(\"Expected 'en-us' key in flattened map\")\n\t}\n\tif _, ok := flattened[\"zh-cn\"]; !ok {\n\t\tt.Error(\"Expected 'zh-cn' key in flattened map\")\n\t}\n\n\t// Should have short lang codes\n\tif _, ok := flattened[\"en\"]; !ok {\n\t\tt.Error(\"Expected 'en' short code in flattened map\")\n\t}\n\tif _, ok := flattened[\"us\"]; !ok {\n\t\tt.Error(\"Expected 'us' region code in flattened map\")\n\t}\n\tif _, ok := flattened[\"zh\"]; !ok {\n\t\tt.Error(\"Expected 'zh' short code in flattened map\")\n\t}\n\tif _, ok := flattened[\"cn\"]; !ok {\n\t\tt.Error(\"Expected 'cn' region code in flattened map\")\n\t}\n\n\t// Verify messages are preserved\n\tif msg, ok := flattened[\"en\"].Messages[\"greeting\"].(string); !ok || msg != \"Hello\" {\n\t\tt.Errorf(\"Expected 'Hello', got %v\", flattened[\"en\"].Messages[\"greeting\"])\n\t}\n\tif msg, ok := flattened[\"zh\"].Messages[\"greeting\"].(string); !ok || msg != \"你好\" {\n\t\tt.Errorf(\"Expected '你好', got %v\", flattened[\"zh\"].Messages[\"greeting\"])\n\t}\n}\n\n// TestMapFlattenWithGlobal tests the FlattenWithGlobal method\nfunc TestMapFlattenWithGlobal(t *testing.T) {\n\t// Save and restore __global__\n\toriginalGlobal := Locales[\"__global__\"]\n\tdefer func() {\n\t\tLocales[\"__global__\"] = originalGlobal\n\t}()\n\n\t// Setup global locales\n\tLocales[\"__global__\"] = map[string]I18n{\n\t\t\"en\": {\n\t\t\tLocale: \"en\",\n\t\t\tMessages: map[string]any{\n\t\t\t\t\"global.key\": \"Global Value\",\n\t\t\t\t\"common\": \"Common\",\n\t\t\t},\n\t\t},\n\t}\n\n\ti18ns := Map{\n\t\t\"en\": I18n{\n\t\t\tLocale: \"en\",\n\t\t\tMessages: map[string]any{\n\t\t\t\t\"local.key\": \"Local Value\",\n\t\t\t\t\"common\": \"Local Common\", // Should override global\n\t\t\t},\n\t\t},\n\t}\n\n\tflattened := i18ns.FlattenWithGlobal()\n\n\tif _, ok := flattened[\"en\"]; !ok {\n\t\tt.Fatal(\"Expected 'en' key in flattened map\")\n\t}\n\n\t// Should have local key\n\tif val, ok := flattened[\"en\"].Messages[\"local.key\"].(string); !ok || val != \"Local Value\" {\n\t\tt.Errorf(\"Expected 'Local Value', got %v\", flattened[\"en\"].Messages[\"local.key\"])\n\t}\n\n\t// Should have global key\n\tif val, ok := flattened[\"en\"].Messages[\"global.key\"].(string); !ok || val != \"Global Value\" {\n\t\tt.Errorf(\"Expected 'Global Value', got %v\", flattened[\"en\"].Messages[\"global.key\"])\n\t}\n\n\t// Local should override global\n\tif val, ok := flattened[\"en\"].Messages[\"common\"].(string); !ok || val != \"Local Common\" {\n\t\tt.Errorf(\"Expected 'Local Common', got %v\", flattened[\"en\"].Messages[\"common\"])\n\t}\n}\n\n// TestMapFlattenWithGlobalNoGlobal tests FlattenWithGlobal when no global exists\nfunc TestMapFlattenWithGlobalNoGlobal(t *testing.T) {\n\t// Save and restore __global__\n\toriginalGlobal := Locales[\"__global__\"]\n\tdefer func() {\n\t\tLocales[\"__global__\"] = originalGlobal\n\t}()\n\n\t// Make sure no global exists\n\tdelete(Locales, \"__global__\")\n\n\ti18ns := Map{\n\t\t\"en\": I18n{\n\t\t\tLocale: \"en\",\n\t\t\tMessages: map[string]any{\n\t\t\t\t\"key\": \"value\",\n\t\t\t},\n\t\t},\n\t}\n\n\tflattened := i18ns.FlattenWithGlobal()\n\n\tif _, ok := flattened[\"en\"]; !ok {\n\t\tt.Fatal(\"Expected 'en' key in flattened map\")\n\t}\n\n\tif val, ok := flattened[\"en\"].Messages[\"key\"].(string); !ok || val != \"value\" {\n\t\tt.Errorf(\"Expected 'value', got %v\", flattened[\"en\"].Messages[\"key\"])\n\t}\n}\n\n// TestMapFlattenWithGlobalKeyConflict tests FlattenWithGlobal when local keys already exist\nfunc TestMapFlattenWithGlobalKeyConflict(t *testing.T) {\n\t// Save and restore __global__\n\toriginalGlobal := Locales[\"__global__\"]\n\tdefer func() {\n\t\tLocales[\"__global__\"] = originalGlobal\n\t}()\n\n\t// Setup global with keys in flat format (after Dot())\n\tLocales[\"__global__\"] = map[string]I18n{\n\t\t\"en\": {\n\t\t\tLocale: \"en\",\n\t\t\tMessages: map[string]any{\n\t\t\t\t\"shared.key\": \"Global Shared\",\n\t\t\t\t\"global.only\": \"Global Only\",\n\t\t\t\t\"local.key\": \"Global Local\", // Will be overridden\n\t\t\t},\n\t\t},\n\t}\n\n\t// Local messages in nested format (will be flattened by Dot())\n\ti18ns := Map{\n\t\t\"en\": I18n{\n\t\t\tLocale: \"en\",\n\t\t\tMessages: map[string]any{\n\t\t\t\t\"local\": map[string]any{\n\t\t\t\t\t\"key\": \"Local Value\", // After Dot() becomes \"local.key\", should override global\n\t\t\t\t},\n\t\t\t\t\"unique\": map[string]any{\n\t\t\t\t\t\"key\": \"Local Unique\",\n\t\t\t\t},\n\t\t\t},\n\t\t},\n\t}\n\n\tflattened := i18ns.FlattenWithGlobal()\n\n\tif _, ok := flattened[\"en\"]; !ok {\n\t\tt.Fatal(\"Expected 'en' key in flattened map\")\n\t}\n\n\t// Local key should exist and NOT be overridden by global\n\tif val, ok := flattened[\"en\"].Messages[\"local.key\"].(string); !ok || val != \"Local Value\" {\n\t\tt.Errorf(\"Expected 'Local Value' (local should override global), got %v\", flattened[\"en\"].Messages[\"local.key\"])\n\t}\n\n\t// Global only key should exist\n\tif val, ok := flattened[\"en\"].Messages[\"global.only\"].(string); !ok || val != \"Global Only\" {\n\t\tt.Errorf(\"Expected 'Global Only', got %v\", flattened[\"en\"].Messages[\"global.only\"])\n\t}\n\n\t// Unique local key should exist\n\tif val, ok := flattened[\"en\"].Messages[\"unique.key\"].(string); !ok || val != \"Local Unique\" {\n\t\tt.Errorf(\"Expected 'Local Unique', got %v\", flattened[\"en\"].Messages[\"unique.key\"])\n\t}\n\n\t// Shared key from global should exist\n\tif val, ok := flattened[\"en\"].Messages[\"shared.key\"].(string); !ok || val != \"Global Shared\" {\n\t\tt.Errorf(\"Expected 'Global Shared', got %v\", flattened[\"en\"].Messages[\"shared.key\"])\n\t}\n}\n\n// TestTranslate tests the Translate function\nfunc TestTranslate(t *testing.T) {\n\tassistantID := \"test-assistant\"\n\tLocales[assistantID] = map[string]I18n{\n\t\t\"en\": {\n\t\t\tLocale: \"en\",\n\t\t\tMessages: map[string]any{\n\t\t\t\t\"greeting\": \"Hello\",\n\t\t\t\t\"name\": \"John\",\n\t\t\t},\n\t\t},\n\t\t\"zh-cn\": {\n\t\t\tLocale: \"zh-cn\",\n\t\t\tMessages: map[string]any{\n\t\t\t\t\"greeting\": \"你好\",\n\t\t\t\t\"name\": \"张三\",\n\t\t\t},\n\t\t},\n\t}\n\tdefer delete(Locales, assistantID)\n\n\tt.Run(\"Translate with exact locale match\", func(t *testing.T) {\n\t\tresult := Translate(assistantID, \"en\", \"{{greeting}}\")\n\t\tif result != \"Hello\" {\n\t\t\tt.Errorf(\"Expected 'Hello', got %v\", result)\n\t\t}\n\t})\n\n\tt.Run(\"Translate with locale variant\", func(t *testing.T) {\n\t\tresult := Translate(assistantID, \"zh-CN\", \"{{greeting}}\")\n\t\tif result != \"你好\" {\n\t\t\tt.Errorf(\"Expected '你好', got %v\", result)\n\t\t}\n\t})\n\n\tt.Run(\"Translate with short locale code\", func(t *testing.T) {\n\t\tresult := Translate(assistantID, \"en-us\", \"{{name}}\")\n\t\tif result != \"John\" {\n\t\t\tt.Errorf(\"Expected 'John', got %v\", result)\n\t\t}\n\t})\n\n\tt.Run(\"Translate without locale match\", func(t *testing.T) {\n\t\tresult := Translate(assistantID, \"fr\", \"{{greeting}}\")\n\t\t// Should return original when no locale found\n\t\tif result != \"{{greeting}}\" {\n\t\t\tt.Errorf(\"Expected '{{greeting}}', got %v\", result)\n\t\t}\n\t})\n\n\tt.Run(\"Translate non-existent assistant\", func(t *testing.T) {\n\t\tresult := Translate(\"nonexistent\", \"en\", \"{{greeting}}\")\n\t\tif result != \"{{greeting}}\" {\n\t\t\tt.Errorf(\"Expected '{{greeting}}', got %v\", result)\n\t\t}\n\t})\n\n\tt.Run(\"Translate with fallback to global\", func(t *testing.T) {\n\t\t// Save and restore __global__\n\t\toriginalGlobal := Locales[\"__global__\"]\n\t\tdefer func() {\n\t\t\tLocales[\"__global__\"] = originalGlobal\n\t\t}()\n\n\t\tLocales[\"__global__\"] = map[string]I18n{\n\t\t\t\"es\": {\n\t\t\t\tLocale: \"es\",\n\t\t\t\tMessages: map[string]any{\n\t\t\t\t\t\"greeting\": \"Hola\",\n\t\t\t\t},\n\t\t\t},\n\t\t}\n\n\t\tresult := Translate(assistantID, \"es\", \"{{greeting}}\")\n\t\tif result != \"Hola\" {\n\t\t\tt.Errorf(\"Expected 'Hola', got %v\", result)\n\t\t}\n\t})\n\n\tt.Run(\"Translate complex structure\", func(t *testing.T) {\n\t\tinput := map[string]any{\n\t\t\t\"title\": \"{{greeting}}\",\n\t\t\t\"user\": \"{{name}}\",\n\t\t}\n\t\tresult := Translate(assistantID, \"zh-cn\", input)\n\t\tif resultMap, ok := result.(map[string]any); ok {\n\t\t\tif resultMap[\"title\"] != \"你好\" {\n\t\t\t\tt.Errorf(\"Expected '你好', got %v\", resultMap[\"title\"])\n\t\t\t}\n\t\t\tif resultMap[\"user\"] != \"张三\" {\n\t\t\t\tt.Errorf(\"Expected '张三', got %v\", resultMap[\"user\"])\n\t\t\t}\n\t\t} else {\n\t\t\tt.Errorf(\"Expected map[string]any, got %T\", result)\n\t\t}\n\t})\n}\n\n// TestTranslateGlobal tests the TranslateGlobal function with custom messages\nfunc TestTranslateGlobal(t *testing.T) {\n\ttest.Prepare(t, config.Conf)\n\tdefer test.Clean()\n\n\t// Save existing __global__ and restore after test\n\toriginalGlobal := make(map[string]I18n)\n\tif existing, ok := Locales[\"__global__\"]; ok {\n\t\tfor k, v := range existing {\n\t\t\toriginalGlobal[k] = v\n\t\t}\n\t}\n\tdefer func() {\n\t\tLocales[\"__global__\"] = originalGlobal\n\t}()\n\n\t// Add custom test messages to existing global (not replacing)\n\tif Locales[\"__global__\"] == nil {\n\t\tLocales[\"__global__\"] = make(map[string]I18n)\n\t}\n\n\t// Extend existing English messages\n\tenMessages := make(map[string]any)\n\tif existing, ok := Locales[\"__global__\"][\"en\"]; ok {\n\t\tfor k, v := range existing.Messages {\n\t\t\tenMessages[k] = v\n\t\t}\n\t}\n\tenMessages[\"button.ok\"] = \"OK\"\n\tenMessages[\"button.cancel\"] = \"Cancel\"\n\tLocales[\"__global__\"][\"en\"] = I18n{\n\t\tLocale: \"en\",\n\t\tMessages: enMessages,\n\t}\n\n\t// Extend existing Chinese messages\n\tzhcnMessages := make(map[string]any)\n\tif existing, ok := Locales[\"__global__\"][\"zh-cn\"]; ok {\n\t\tfor k, v := range existing.Messages {\n\t\t\tzhcnMessages[k] = v\n\t\t}\n\t}\n\tzhcnMessages[\"button.ok\"] = \"确定\"\n\tzhcnMessages[\"button.cancel\"] = \"取消\"\n\tLocales[\"__global__\"][\"zh-cn\"] = I18n{\n\t\tLocale: \"zh-cn\",\n\t\tMessages: zhcnMessages,\n\t}\n\n\t// Extend existing Chinese short code messages\n\tzhMessages := make(map[string]any)\n\tif existing, ok := Locales[\"__global__\"][\"zh\"]; ok {\n\t\tfor k, v := range existing.Messages {\n\t\t\tzhMessages[k] = v\n\t\t}\n\t}\n\tzhMessages[\"button.ok\"] = \"确定\"\n\tzhMessages[\"button.cancel\"] = \"取消\"\n\tLocales[\"__global__\"][\"zh\"] = I18n{\n\t\tLocale: \"zh\",\n\t\tMessages: zhMessages,\n\t}\n\n\tt.Run(\"TranslateGlobal with match\", func(t *testing.T) {\n\t\tresult := TranslateGlobal(\"en\", \"{{button.ok}}\")\n\t\tif result != \"OK\" {\n\t\t\tt.Errorf(\"Expected 'OK', got %v\", result)\n\t\t}\n\t})\n\n\tt.Run(\"TranslateGlobal with Chinese\", func(t *testing.T) {\n\t\tresult := TranslateGlobal(\"zh-cn\", \"{{button.cancel}}\")\n\t\tif result != \"取消\" {\n\t\t\tt.Errorf(\"Expected '取消', got %v\", result)\n\t\t}\n\t})\n\n\tt.Run(\"TranslateGlobal with short code\", func(t *testing.T) {\n\t\tresult := TranslateGlobal(\"zh-TW\", \"{{button.ok}}\")\n\t\tif result != \"确定\" {\n\t\t\tt.Errorf(\"Expected '确定', got %v\", result)\n\t\t}\n\t})\n\n\tt.Run(\"TranslateGlobal without match\", func(t *testing.T) {\n\t\tresult := TranslateGlobal(\"fr\", \"{{button.ok}}\")\n\t\tif result != \"{{button.ok}}\" {\n\t\t\tt.Errorf(\"Expected '{{button.ok}}', got %v\", result)\n\t\t}\n\t})\n\n\tt.Run(\"TranslateGlobal no global\", func(t *testing.T) {\n\t\t// Temporarily remove global\n\t\ttemp := Locales[\"__global__\"]\n\t\tdelete(Locales, \"__global__\")\n\n\t\tresult := TranslateGlobal(\"en\", \"{{button.ok}}\")\n\t\tif result != \"{{button.ok}}\" {\n\t\t\tt.Errorf(\"Expected '{{button.ok}}', got %v\", result)\n\t\t}\n\n\t\t// Restore\n\t\tLocales[\"__global__\"] = temp\n\t})\n\n\tt.Run(\"TranslateGlobal fallback from en-us to en\", func(t *testing.T) {\n\t\t// Create a scenario where en-us has limited messages, but en has more\n\t\t// This simulates the real-world case: en-us locale exists with 3 messages,\n\t\t// but en has 41 messages including llm.handlers.stream.info\n\n\t\t// Create en-us with only a few messages\n\t\tenUSMessages := map[string]any{\n\t\t\t\"button.ok\": \"OK (US)\", // en-us specific\n\t\t\t\"app.name\": \"My App\",\n\t\t\t\"app.version\": \"1.0\",\n\t\t}\n\t\tLocales[\"__global__\"][\"en-us\"] = I18n{\n\t\t\tLocale: \"en-us\",\n\t\t\tMessages: enUSMessages,\n\t\t}\n\n\t\t// Test 1: Key exists in en-us - should use en-us\n\t\tresult := TranslateGlobal(\"en-us\", \"button.ok\")\n\t\tif result != \"OK (US)\" {\n\t\t\tt.Errorf(\"Expected 'OK (US)' from en-us, got %v\", result)\n\t\t}\n\n\t\t// Test 2: Key does NOT exist in en-us but exists in en - should fallback to en\n\t\tresult = TranslateGlobal(\"en-us\", \"button.cancel\")\n\t\tif result != \"Cancel\" {\n\t\t\tt.Errorf(\"Expected 'Cancel' (fallback from en-us to en), got %v\", result)\n\t\t}\n\n\t\t// Test 3: Built-in key that exists in en but not in en-us\n\t\tresult = TranslateGlobal(\"en-us\", \"llm.handlers.stream.info\")\n\t\texpected := \"LLM Stream\"\n\t\tif result != expected {\n\t\t\tt.Errorf(\"Expected '%s' (fallback from en-us to en), got %v\", expected, result)\n\t\t}\n\n\t\t// Test 4: Direct key (not template) also should fallback\n\t\tresult = TranslateGlobal(\"en-us\", \"{{llm.handlers.stream.info}}\")\n\t\tif result != expected {\n\t\t\tt.Errorf(\"Expected '%s' (fallback from en-us to en with template), got %v\", expected, result)\n\t\t}\n\t})\n}\n\n// TestGetLocalesIntegration tests GetLocales with real assistant data\nfunc TestGetLocalesIntegration(t *testing.T) {\n\ttest.Prepare(t, config.Conf)\n\tdefer test.Clean()\n\n\t// Use the real mohe assistant path (relative to app root)\n\tassistantPath := \"/assistants/mohe\"\n\n\tt.Run(\"Load real locale files\", func(t *testing.T) {\n\t\tlocales, err := GetLocales(assistantPath)\n\t\tif err != nil {\n\t\t\tt.Skipf(\"Skipping: %v\", err)\n\t\t\treturn\n\t\t}\n\n\t\t// Should have at least 2 locales (en-us and zh-cn)\n\t\tif len(locales) < 2 {\n\t\t\tt.Errorf(\"Expected at least 2 locales, got %d\", len(locales))\n\t\t}\n\n\t\t// Check en-us locale\n\t\tif enUS, ok := locales[\"en-us\"]; ok {\n\t\t\tif enUS.Locale != \"en-us\" {\n\t\t\t\tt.Errorf(\"Expected locale 'en-us', got %s\", enUS.Locale)\n\t\t\t}\n\n\t\t\t// Check some messages\n\t\t\tif desc, ok := enUS.Messages[\"description\"].(string); ok {\n\t\t\t\tif desc == \"\" {\n\t\t\t\t\tt.Error(\"Expected non-empty description\")\n\t\t\t\t}\n\t\t\t\tt.Logf(\"English description: %s\", desc)\n\t\t\t}\n\n\t\t\tif chat, ok := enUS.Messages[\"chat\"].(map[string]interface{}); ok {\n\t\t\t\tif title, ok := chat[\"title\"].(string); ok {\n\t\t\t\t\tif title != \"New Chat\" {\n\t\t\t\t\t\tt.Errorf(\"Expected 'New Chat', got %s\", title)\n\t\t\t\t\t}\n\t\t\t\t}\n\t\t\t}\n\t\t} else {\n\t\t\tt.Error(\"Expected 'en-us' locale\")\n\t\t}\n\n\t\t// Check zh-cn locale\n\t\tif zhCN, ok := locales[\"zh-cn\"]; ok {\n\t\t\tif zhCN.Locale != \"zh-cn\" {\n\t\t\t\tt.Errorf(\"Expected locale 'zh-cn', got %s\", zhCN.Locale)\n\t\t\t}\n\n\t\t\t// Check some messages\n\t\t\tif desc, ok := zhCN.Messages[\"description\"].(string); ok {\n\t\t\t\tif desc == \"\" {\n\t\t\t\t\tt.Error(\"Expected non-empty description\")\n\t\t\t\t}\n\t\t\t\tt.Logf(\"Chinese description: %s\", desc)\n\t\t\t}\n\n\t\t\tif chat, ok := zhCN.Messages[\"chat\"].(map[string]interface{}); ok {\n\t\t\t\tif title, ok := chat[\"title\"].(string); ok {\n\t\t\t\t\tif title != \"新对话\" {\n\t\t\t\t\t\tt.Errorf(\"Expected '新对话', got %s\", title)\n\t\t\t\t\t}\n\t\t\t\t}\n\t\t\t}\n\t\t} else {\n\t\t\tt.Error(\"Expected 'zh-cn' locale\")\n\t\t}\n\n\t\tt.Logf(\"Loaded %d locales successfully\", len(locales))\n\t})\n\n\tt.Run(\"Flatten loaded locales\", func(t *testing.T) {\n\t\tlocales, err := GetLocales(assistantPath)\n\t\tif err != nil {\n\t\t\tt.Skipf(\"Skipping: %v\", err)\n\t\t\treturn\n\t\t}\n\n\t\tflattened := locales.Flatten()\n\n\t\t// Should have short codes\n\t\tif _, ok := flattened[\"en\"]; !ok {\n\t\t\tt.Error(\"Expected 'en' short code after flatten\")\n\t\t}\n\t\tif _, ok := flattened[\"zh\"]; !ok {\n\t\t\tt.Error(\"Expected 'zh' short code after flatten\")\n\t\t}\n\t\tif _, ok := flattened[\"us\"]; !ok {\n\t\t\tt.Error(\"Expected 'us' region code after flatten\")\n\t\t}\n\t\tif _, ok := flattened[\"cn\"]; !ok {\n\t\t\tt.Error(\"Expected 'cn' region code after flatten\")\n\t\t}\n\n\t\t// Verify flattened messages structure\n\t\tif en, ok := flattened[\"en\"]; ok {\n\t\t\tif _, ok := en.Messages[\"chat.title\"]; !ok {\n\t\t\t\tt.Error(\"Expected flattened 'chat.title' key\")\n\t\t\t}\n\t\t\tif _, ok := en.Messages[\"chat.description\"]; !ok {\n\t\t\t\tt.Error(\"Expected flattened 'chat.description' key\")\n\t\t\t}\n\t\t\tif _, ok := en.Messages[\"chat.prompts.0\"]; !ok {\n\t\t\t\tt.Error(\"Expected flattened 'chat.prompts.0' key\")\n\t\t\t}\n\t\t}\n\n\t\tt.Logf(\"Flattened to %d locale codes\", len(flattened))\n\t})\n}\n\n// TestEdgeCases tests various edge cases\nfunc TestEdgeCases(t *testing.T) {\n\tt.Run(\"Empty Messages map\", func(t *testing.T) {\n\t\ti18n := I18n{\n\t\t\tLocale: \"en\",\n\t\t\tMessages: map[string]any{},\n\t\t}\n\t\tresult := i18n.Parse(\"{{key}}\")\n\t\tif result != \"{{key}}\" {\n\t\t\tt.Errorf(\"Expected '{{key}}', got %v\", result)\n\t\t}\n\t})\n\n\tt.Run(\"Nil Messages map\", func(t *testing.T) {\n\t\ti18n := I18n{\n\t\t\tLocale: \"en\",\n\t\t\tMessages: nil,\n\t\t}\n\t\tresult := i18n.Parse(\"{{key}}\")\n\t\tif result != \"{{key}}\" {\n\t\t\tt.Errorf(\"Expected '{{key}}', got %v\", result)\n\t\t}\n\t})\n\n\tt.Run(\"Empty locale string\", func(t *testing.T) {\n\t\tLocales[\"test\"] = map[string]I18n{\n\t\t\t\"en\": {\n\t\t\t\tLocale: \"en\",\n\t\t\t\tMessages: map[string]any{\"key\": \"value\"},\n\t\t\t},\n\t\t}\n\t\tdefer delete(Locales, \"test\")\n\n\t\tresult := Translate(\"test\", \"\", \"{{key}}\")\n\t\t// Should still work with empty string after trim\n\t\tif result != \"{{key}}\" {\n\t\t\tt.Logf(\"Result: %v\", result)\n\t\t}\n\t})\n\n\tt.Run(\"Locale with only spaces\", func(t *testing.T) {\n\t\tLocales[\"test\"] = map[string]I18n{\n\t\t\t\"\": {\n\t\t\t\tLocale: \"\",\n\t\t\t\tMessages: map[string]any{\"key\": \"value\"},\n\t\t\t},\n\t\t}\n\t\tdefer delete(Locales, \"test\")\n\n\t\tresult := Translate(\"test\", \" \", \"{{key}}\")\n\t\tif result != \"value\" {\n\t\t\tt.Errorf(\"Expected 'value', got %v\", result)\n\t\t}\n\t})\n}\n\n// TestBuiltinMessages tests the built-in global messages\nfunc TestBuiltinMessages(t *testing.T) {\n\t// Save and restore __global__ to avoid test interference\n\toriginalGlobal := make(map[string]I18n)\n\tif existing, ok := Locales[\"__global__\"]; ok {\n\t\tfor k, v := range existing {\n\t\t\toriginalGlobal[k] = v\n\t\t}\n\t}\n\tdefer func() {\n\t\tLocales[\"__global__\"] = originalGlobal\n\t}()\n\n\tt.Run(\"English built-in messages\", func(t *testing.T) {\n\t\t// Test assistant messages\n\t\t// Updated: label now only shows {{name}} without \"Assistant\" prefix\n\t\tresult := TranslateGlobal(\"en\", \"{{assistant.agent.stream.label}}\")\n\t\texpected := \"{{name}}\"\n\t\tif result != expected {\n\t\t\tt.Errorf(\"Expected '%s', got '%v'\", expected, result)\n\t\t}\n\n\t\tresult = TranslateGlobal(\"en\", \"{{assistant.agent.stream.description}}\")\n\t\texpected = \"{{name}} is processing the request\"\n\t\tif result != expected {\n\t\t\tt.Errorf(\"Expected '%s', got '%v'\", expected, result)\n\t\t}\n\n\t\tresult = TranslateGlobal(\"en\", \"{{assistant.agent.stream.history}}\")\n\t\texpected = \"Get Chat History\"\n\t\tif result != expected {\n\t\t\tt.Errorf(\"Expected '%s', got '%v'\", expected, result)\n\t\t}\n\n\t\t// Test LLM messages (note: LLM uses %s for fmt.Sprintf, not {{name}} for recursive translation)\n\t\tresult = TranslateGlobal(\"en\", \"{{llm.openai.stream.label}}\")\n\t\texpected = \"LLM %s\"\n\t\tif result != expected {\n\t\t\tt.Errorf(\"Expected '%s', got '%v'\", expected, result)\n\t\t}\n\n\t\tresult = TranslateGlobal(\"en\", \"{{llm.handlers.stream.info}}\")\n\t\texpected = \"LLM Stream\"\n\t\tif result != expected {\n\t\t\tt.Errorf(\"Expected '%s', got '%v'\", expected, result)\n\t\t}\n\n\t\t// Test common messages\n\t\tresult = TranslateGlobal(\"en\", \"{{common.status.processing}}\")\n\t\texpected = \"Processing\"\n\t\tif result != expected {\n\t\t\tt.Errorf(\"Expected '%s', got '%v'\", expected, result)\n\t\t}\n\t})\n\n\tt.Run(\"Chinese (zh-cn) built-in messages\", func(t *testing.T) {\n\t\t// Test assistant messages\n\t\t// Updated: label now only shows {{name}} without \"助手\" prefix\n\t\tresult := TranslateGlobal(\"zh-cn\", \"{{assistant.agent.stream.label}}\")\n\t\texpected := \"{{name}}\"\n\t\tif result != expected {\n\t\t\tt.Errorf(\"Expected '%s', got '%v'\", expected, result)\n\t\t}\n\n\t\tresult = TranslateGlobal(\"zh-cn\", \"{{assistant.agent.stream.description}}\")\n\t\texpected = \"{{name}} 正在处理请求\"\n\t\tif result != expected {\n\t\t\tt.Errorf(\"Expected '%s', got '%v'\", expected, result)\n\t\t}\n\n\t\tresult = TranslateGlobal(\"zh-cn\", \"{{assistant.agent.stream.history}}\")\n\t\texpected = \"获取聊天历史\"\n\t\tif result != expected {\n\t\t\tt.Errorf(\"Expected '%s', got '%v'\", expected, result)\n\t\t}\n\n\t\t// Test LLM messages\n\t\tresult = TranslateGlobal(\"zh-cn\", \"{{llm.handlers.stream.info}}\")\n\t\texpected = \"LLM 流式输出\"\n\t\tif result != expected {\n\t\t\tt.Errorf(\"Expected '%s', got '%v'\", expected, result)\n\t\t}\n\n\t\t// Test common messages\n\t\tresult = TranslateGlobal(\"zh-cn\", \"{{common.status.processing}}\")\n\t\texpected = \"处理中\"\n\t\tif result != expected {\n\t\t\tt.Errorf(\"Expected '%s', got '%v'\", expected, result)\n\t\t}\n\t})\n\n\tt.Run(\"Chinese (zh) short code\", func(t *testing.T) {\n\t\t// Updated: label now only shows {{name}} without \"助手\" prefix\n\t\tresult := TranslateGlobal(\"zh\", \"{{assistant.agent.stream.label}}\")\n\t\texpected := \"{{name}}\"\n\t\tif result != expected {\n\t\t\tt.Errorf(\"Expected '%s', got '%v'\", expected, result)\n\t\t}\n\n\t\tresult = TranslateGlobal(\"zh\", \"{{common.status.processing}}\")\n\t\texpected = \"处理中\"\n\t\tif result != expected {\n\t\t\tt.Errorf(\"Expected '%s', got '%v'\", expected, result)\n\t\t}\n\t})\n\n\tt.Run(\"Embedded template with built-in messages\", func(t *testing.T) {\n\t\t// English\n\t\tresult := TranslateGlobal(\"en\", \"Status: {{common.status.processing}}\")\n\t\texpected := \"Status: Processing\"\n\t\tif result != expected {\n\t\t\tt.Errorf(\"Expected '%s', got '%v'\", expected, result)\n\t\t}\n\n\t\t// Chinese\n\t\tresult = TranslateGlobal(\"zh-cn\", \"状态: {{common.status.processing}}\")\n\t\texpected = \"状态: 处理中\"\n\t\tif result != expected {\n\t\t\tt.Errorf(\"Expected '%s', got '%v'\", expected, result)\n\t\t}\n\t})\n\n\tt.Run(\"Non-existent key in global\", func(t *testing.T) {\n\t\tresult := TranslateGlobal(\"en\", \"{{unknown.key}}\")\n\t\tif result != \"{{unknown.key}}\" {\n\t\t\tt.Errorf(\"Expected '{{unknown.key}}', got '%v'\", result)\n\t\t}\n\t})\n}\n\n// TestTAlias tests the T function alias\nfunc TestTr(t *testing.T) {\n\t// Save original global locales\n\toriginalGlobal := Locales[\"__global__\"]\n\tdefer func() {\n\t\tif originalGlobal != nil {\n\t\t\tLocales[\"__global__\"] = originalGlobal\n\t\t} else {\n\t\t\tdelete(Locales, \"__global__\")\n\t\t}\n\t}()\n\n\t// Setup test locales with nested templates\n\tLocales[\"__global__\"] = map[string]I18n{\n\t\t\"en\": {\n\t\t\tLocale: \"en\",\n\t\t\tMessages: map[string]any{\n\t\t\t\t\"assistant.label\": \"Assistant {{assistant.name}}\", // Use full key path\n\t\t\t\t\"assistant.name\": \"AI Helper\",\n\t\t\t\t\"assistant.description\": \"{{assistant.label}} is processing\",\n\t\t\t\t\"llm.label\": \"LLM {{model.deepseek}}\", // Use full key path\n\t\t\t\t\"model.deepseek\": \"DeepSeek\",\n\t\t\t\t\"deeply.nested\": \"Level1 {{level2}}\",\n\t\t\t\t\"level2\": \"Level2 {{level3}}\",\n\t\t\t\t\"level3\": \"Level3 End\",\n\t\t\t\t\"simple.message\": \"Hello World\",\n\t\t\t},\n\t\t},\n\t\t\"zh-cn\": {\n\t\t\tLocale: \"zh-cn\",\n\t\t\tMessages: map[string]any{\n\t\t\t\t\"assistant.label\": \"助手 {{assistant.name}}\", // Use full key path\n\t\t\t\t\"assistant.name\": \"智能助手\",\n\t\t\t\t\"assistant.description\": \"{{assistant.label}} 正在处理\",\n\t\t\t\t\"llm.label\": \"模型 {{model.deepseek}}\", // Use full key path\n\t\t\t\t\"model.deepseek\": \"深度求索\",\n\t\t\t\t\"deeply.nested\": \"第一层 {{level2}}\",\n\t\t\t\t\"level2\": \"第二层 {{level3}}\",\n\t\t\t\t\"level3\": \"第三层结束\",\n\t\t\t\t\"simple.message\": \"你好世界\",\n\t\t\t},\n\t\t},\n\t}\n\n\t// Setup assistant-specific locale (overrides assistant.name, but inherits assistant.label from global)\n\tLocales[\"test-assistant\"] = map[string]I18n{\n\t\t\"en\": {\n\t\t\tLocale: \"en\",\n\t\t\tMessages: map[string]any{\n\t\t\t\t\"assistant.name\": \"Custom Assistant\", // This will override global when assistant.label is resolved\n\t\t\t},\n\t\t},\n\t}\n\tdefer delete(Locales, \"test-assistant\")\n\n\tt.Run(\"Simple translation without variables\", func(t *testing.T) {\n\t\tresult := Tr(\"__global__\", \"en\", \"simple.message\")\n\t\tif result != \"Hello World\" {\n\t\t\tt.Errorf(\"Expected 'Hello World', got '%s'\", result)\n\t\t}\n\n\t\tresult = Tr(\"__global__\", \"zh-cn\", \"simple.message\")\n\t\tif result != \"你好世界\" {\n\t\t\tt.Errorf(\"Expected '你好世界', got '%s'\", result)\n\t\t}\n\t})\n\n\tt.Run(\"One level nested variable\", func(t *testing.T) {\n\t\t// \"Assistant {{name}}\" -> \"Assistant AI Helper\"\n\t\tresult := Tr(\"__global__\", \"en\", \"assistant.label\")\n\t\tif result != \"Assistant AI Helper\" {\n\t\t\tt.Errorf(\"Expected 'Assistant AI Helper', got '%s'\", result)\n\t\t}\n\n\t\tresult = Tr(\"__global__\", \"zh-cn\", \"assistant.label\")\n\t\tif result != \"助手 智能助手\" {\n\t\t\tt.Errorf(\"Expected '助手 智能助手', got '%s'\", result)\n\t\t}\n\t})\n\n\tt.Run(\"Two levels nested variables\", func(t *testing.T) {\n\t\t// \"{{assistant.label}} is processing\" -> \"Assistant AI Helper is processing\"\n\t\tresult := Tr(\"__global__\", \"en\", \"assistant.description\")\n\t\tif result != \"Assistant AI Helper is processing\" {\n\t\t\tt.Errorf(\"Expected 'Assistant AI Helper is processing', got '%s'\", result)\n\t\t}\n\n\t\tresult = Tr(\"__global__\", \"zh-cn\", \"assistant.description\")\n\t\tif result != \"助手 智能助手 正在处理\" {\n\t\t\tt.Errorf(\"Expected '助手 智能助手 正在处理', got '%s'\", result)\n\t\t}\n\t})\n\n\tt.Run(\"Three levels deeply nested\", func(t *testing.T) {\n\t\t// \"Level1 {{level2}}\" -> \"Level1 Level2 {{level3}}\" -> \"Level1 Level2 Level3 End\"\n\t\tresult := Tr(\"__global__\", \"en\", \"deeply.nested\")\n\t\tif result != \"Level1 Level2 Level3 End\" {\n\t\t\tt.Errorf(\"Expected 'Level1 Level2 Level3 End', got '%s'\", result)\n\t\t}\n\n\t\tresult = Tr(\"__global__\", \"zh-cn\", \"deeply.nested\")\n\t\tif result != \"第一层 第二层 第三层结束\" {\n\t\t\tt.Errorf(\"Expected '第一层 第二层 第三层结束', got '%s'\", result)\n\t\t}\n\t})\n\n\tt.Run(\"Assistant-specific override\", func(t *testing.T) {\n\t\t// When assistant locale exists but doesn't have a key, it WILL fallback to global\n\t\t// This is key-level fallback: try assistant first, then fallback to global\n\t\tresult := Tr(\"test-assistant\", \"en\", \"assistant.label\")\n\t\t// \"Assistant {{assistant.name}}\" from global, then {{assistant.name}} -> \"Custom Assistant\" from assistant\n\t\tif result != \"Assistant Custom Assistant\" {\n\t\t\tt.Errorf(\"Expected 'Assistant Custom Assistant' (fallback to global with assistant override), got '%s'\", result)\n\t\t}\n\n\t\t// assistant has 'en' locale but doesn't have this key, fallback to global\n\t\tresult = Tr(\"test-assistant\", \"en\", \"simple.message\")\n\t\tif result != \"Hello World\" {\n\t\t\tt.Errorf(\"Expected 'Hello World' (fallback to global), got '%s'\", result)\n\t\t}\n\n\t\t// If assistant locale has the key, it will use assistant's value\n\t\tresult = Tr(\"test-assistant\", \"en\", \"assistant.name\")\n\t\tif result != \"Custom Assistant\" {\n\t\t\tt.Errorf(\"Expected 'Custom Assistant' (from assistant locale), got '%s'\", result)\n\t\t}\n\t})\n\n\tt.Run(\"Non-existent key returns original\", func(t *testing.T) {\n\t\tresult := Tr(\"__global__\", \"en\", \"non.existent.key\")\n\t\tif result != \"non.existent.key\" {\n\t\t\tt.Errorf(\"Expected 'non.existent.key', got '%s'\", result)\n\t\t}\n\t})\n\n\tt.Run(\"LLM with model variable\", func(t *testing.T) {\n\t\tresult := Tr(\"__global__\", \"en\", \"llm.label\")\n\t\tif result != \"LLM DeepSeek\" {\n\t\t\tt.Errorf(\"Expected 'LLM DeepSeek', got '%s'\", result)\n\t\t}\n\n\t\tresult = Tr(\"__global__\", \"zh-cn\", \"llm.label\")\n\t\tif result != \"模型 深度求索\" {\n\t\t\tt.Errorf(\"Expected '模型 深度求索', got '%s'\", result)\n\t\t}\n\t})\n}\n\nfunc TestTAlias(t *testing.T) {\n\t// Save and restore __global__ to avoid test interference\n\toriginalGlobal := make(map[string]I18n)\n\tif existing, ok := Locales[\"__global__\"]; ok {\n\t\tfor k, v := range existing {\n\t\t\toriginalGlobal[k] = v\n\t\t}\n\t}\n\tdefer func() {\n\t\tLocales[\"__global__\"] = originalGlobal\n\t}()\n\n\tt.Run(\"T alias works like TranslateGlobal\", func(t *testing.T) {\n\t\t// Test that T and TranslateGlobal return the same results\n\t\tinput := \"{{assistant.agent.stream.label}}\"\n\n\t\tresultT := T(\"en\", input)\n\t\tresultGlobal := TranslateGlobal(\"en\", input)\n\n\t\tif resultT != resultGlobal {\n\t\t\tt.Errorf(\"T and TranslateGlobal should return same result. T: %v, TranslateGlobal: %v\", resultT, resultGlobal)\n\t\t}\n\n\t\t// Updated: label now only shows {{name}} without \"Assistant\" prefix\n\t\texpected := \"{{name}}\"\n\t\tif resultT != expected {\n\t\t\tt.Errorf(\"Expected '%s', got '%v'\", expected, resultT)\n\t\t}\n\t})\n\n\tt.Run(\"T alias with Chinese\", func(t *testing.T) {\n\t\tresult := T(\"zh-cn\", \"{{assistant.agent.stream.history}}\")\n\t\texpected := \"获取聊天历史\"\n\t\tif result != expected {\n\t\t\tt.Errorf(\"Expected '%s', got '%v'\", expected, result)\n\t\t}\n\t})\n\n\tt.Run(\"T alias with embedded template\", func(t *testing.T) {\n\t\tresult := T(\"en\", \"Status: {{common.status.completed}}\")\n\t\texpected := \"Status: Completed\"\n\t\tif result != expected {\n\t\t\tt.Errorf(\"Expected '%s', got '%v'\", expected, result)\n\t\t}\n\t})\n\n\tt.Run(\"T with nested template (template in template value)\", func(t *testing.T) {\n\t\t// assistant.agent.stream.label = \"{{name}}\" (contains {{name}} template)\n\t\t// Updated: label now only shows {{name}} without prefix\n\t\t// This tests if we can get the template string itself\n\t\tresult := T(\"en\", \"{{assistant.agent.stream.label}}\")\n\t\texpected := \"{{name}}\"\n\t\tif result != expected {\n\t\t\tt.Errorf(\"Expected '%s', got '%v'\", expected, result)\n\t\t}\n\n\t\t// Verify Chinese version too\n\t\tresultZh := T(\"zh-cn\", \"{{assistant.agent.stream.label}}\")\n\t\texpectedZh := \"{{name}}\"\n\t\tif resultZh != expectedZh {\n\t\t\tt.Errorf(\"Expected '%s', got '%v'\", expectedZh, resultZh)\n\t\t}\n\t})\n}\n" }, { "path": "agent/llm/adapters/adapter.go", "content": "package adapters\n\nimport (\n\t\"github.com/yaoapp/yao/agent/context\"\n)\n\n// CapabilityAdapter is the interface for capability-specific message and response processing\n// Each adapter handles one capability dimension (tool calls, vision, audio, reasoning, etc.)\ntype CapabilityAdapter interface {\n\t// Name returns the adapter name for debugging\n\tName() string\n\n\t// PreprocessMessages preprocesses messages before sending to LLM\n\t// Returns modified messages or error\n\tPreprocessMessages(messages []context.Message) ([]context.Message, error)\n\n\t// PreprocessOptions preprocesses completion options before sending to LLM\n\t// Returns modified options or error\n\tPreprocessOptions(options *context.CompletionOptions) (*context.CompletionOptions, error)\n\n\t// PostprocessResponse postprocesses the LLM response\n\t// Returns modified response or error\n\tPostprocessResponse(response *context.CompletionResponse) (*context.CompletionResponse, error)\n\n\t// ProcessStreamChunk processes a streaming chunk\n\t// Returns modified chunk type and data, or error\n\tProcessStreamChunk(chunkType context.StreamChunkType, data []byte) (context.StreamChunkType, []byte, error)\n}\n\n// BaseAdapter provides default implementations for CapabilityAdapter\n// Adapters can embed this and override only the methods they need\ntype BaseAdapter struct {\n\tname string\n}\n\n// NewBaseAdapter creates a new base adapter\nfunc NewBaseAdapter(name string) *BaseAdapter {\n\treturn &BaseAdapter{name: name}\n}\n\n// Name returns the adapter name\nfunc (a *BaseAdapter) Name() string {\n\treturn a.name\n}\n\n// PreprocessMessages default implementation (no-op)\nfunc (a *BaseAdapter) PreprocessMessages(messages []context.Message) ([]context.Message, error) {\n\treturn messages, nil\n}\n\n// PreprocessOptions default implementation (no-op)\nfunc (a *BaseAdapter) PreprocessOptions(options *context.CompletionOptions) (*context.CompletionOptions, error) {\n\treturn options, nil\n}\n\n// PostprocessResponse default implementation (no-op)\nfunc (a *BaseAdapter) PostprocessResponse(response *context.CompletionResponse) (*context.CompletionResponse, error) {\n\treturn response, nil\n}\n\n// ProcessStreamChunk default implementation (pass through)\nfunc (a *BaseAdapter) ProcessStreamChunk(chunkType context.StreamChunkType, data []byte) (context.StreamChunkType, []byte, error) {\n\treturn chunkType, data, nil\n}\n" }, { "path": "agent/llm/adapters/audio.go", "content": "package adapters\n\nimport (\n\t\"github.com/yaoapp/yao/agent/context\"\n)\n\n// AudioAdapter handles audio capability\n// If model doesn't support audio, it removes or converts audio content\ntype AudioAdapter struct {\n\t*BaseAdapter\n\tnativeSupport bool\n}\n\n// NewAudioAdapter creates a new audio adapter\nfunc NewAudioAdapter(nativeSupport bool) *AudioAdapter {\n\treturn &AudioAdapter{\n\t\tBaseAdapter: NewBaseAdapter(\"AudioAdapter\"),\n\t\tnativeSupport: nativeSupport,\n\t}\n}\n\n// PreprocessMessages removes or converts audio content if not supported\nfunc (a *AudioAdapter) PreprocessMessages(messages []context.Message) ([]context.Message, error) {\n\tif a.nativeSupport {\n\t\t// Native support, no preprocessing needed\n\t\treturn messages, nil\n\t}\n\n\t// Process messages to remove audio content\n\tprocessed := make([]context.Message, 0, len(messages))\n\tfor _, msg := range messages {\n\t\tprocessedMsg := msg\n\n\t\t// Handle multimodal content (array of ContentPart)\n\t\tif contentParts, ok := msg.Content.([]context.ContentPart); ok {\n\t\t\tfilteredParts := make([]context.ContentPart, 0)\n\n\t\t\tfor _, part := range contentParts {\n\t\t\t\t// Skip audio content if not supported\n\t\t\t\tif part.Type == context.ContentInputAudio {\n\t\t\t\t\t// TODO: Optionally convert to transcription text if available\n\t\t\t\t\tcontinue\n\t\t\t\t}\n\t\t\t\tfilteredParts = append(filteredParts, part)\n\t\t\t}\n\n\t\t\t// If all parts were filtered out, add placeholder text\n\t\t\tif len(filteredParts) == 0 {\n\t\t\t\tprocessedMsg.Content = \"[Audio content not supported by this model]\"\n\t\t\t} else if len(filteredParts) == 1 && filteredParts[0].Type == context.ContentText {\n\t\t\t\t// Single text part, convert to string\n\t\t\t\tprocessedMsg.Content = filteredParts[0].Text\n\t\t\t} else {\n\t\t\t\tprocessedMsg.Content = filteredParts\n\t\t\t}\n\t\t}\n\n\t\tprocessed = append(processed, processedMsg)\n\t}\n\n\treturn processed, nil\n}\n" }, { "path": "agent/llm/adapters/reasoning.go", "content": "package adapters\n\nimport (\n\t\"github.com/yaoapp/gou/connector/openai\"\n\t\"github.com/yaoapp/yao/agent/context\"\n)\n\n// ReasoningFormat represents the reasoning content format\ntype ReasoningFormat string\n\nconst (\n\tReasoningFormatNone ReasoningFormat = \"none\" // No reasoning support\n\tReasoningFormatOpenAI ReasoningFormat = \"openai-o1\" // OpenAI o1 format (hidden reasoning)\n\tReasoningFormatGPT5 ReasoningFormat = \"gpt-5\" // GPT-5 format (hidden reasoning)\n\tReasoningFormatDeepSeek ReasoningFormat = \"deepseek-r1\" // DeepSeek R1 format (visible reasoning)\n)\n\n// ReasoningAdapter handles reasoning content capability\n// - Manages reasoning_effort parameter (o1, GPT-5)\n// - Manages temperature parameter constraints (reasoning models typically require temperature=1)\n// - Extracts reasoning_tokens from usage\n// - Parses visible reasoning content (DeepSeek R1)\ntype ReasoningAdapter struct {\n\t*BaseAdapter\n\tformat ReasoningFormat\n\tsupportsEffort bool // Whether the model supports reasoning_effort parameter\n\tsupportsTemperature bool // Whether the model supports temperature adjustment\n}\n\n// NewReasoningAdapter creates a new reasoning adapter\n// If cap.TemperatureAdjustable is provided, it overrides the default behavior\nfunc NewReasoningAdapter(format ReasoningFormat, cap *openai.Capabilities) *ReasoningAdapter {\n\tsupportsEffort := false\n\tsupportsTemperature := true\n\n\t// Set defaults based on reasoning format\n\tswitch format {\n\tcase ReasoningFormatOpenAI, ReasoningFormatGPT5:\n\t\t// OpenAI o1 and GPT-5: support reasoning_effort, but NOT temperature adjustment\n\t\tsupportsEffort = true\n\t\tsupportsTemperature = false\n\tcase ReasoningFormatDeepSeek:\n\t\t// DeepSeek R1: no reasoning_effort, no temperature adjustment\n\t\tsupportsEffort = false\n\t\tsupportsTemperature = false\n\tcase ReasoningFormatNone:\n\t\t// Non-reasoning models: no reasoning_effort, but support temperature\n\t\tsupportsEffort = false\n\t\tsupportsTemperature = true\n\t}\n\n\t// Override with explicit capability if provided\n\tif cap != nil {\n\t\tsupportsTemperature = cap.TemperatureAdjustable\n\t}\n\n\treturn &ReasoningAdapter{\n\t\tBaseAdapter: NewBaseAdapter(\"ReasoningAdapter\"),\n\t\tformat: format,\n\t\tsupportsEffort: supportsEffort,\n\t\tsupportsTemperature: supportsTemperature,\n\t}\n}\n\n// PreprocessOptions handles reasoning_effort and temperature parameters\nfunc (a *ReasoningAdapter) PreprocessOptions(options *context.CompletionOptions) (*context.CompletionOptions, error) {\n\tif options == nil {\n\t\treturn options, nil\n\t}\n\n\tnewOptions := *options\n\tmodified := false\n\n\t// 1. Handle reasoning_effort parameter\n\tif !a.supportsEffort && newOptions.ReasoningEffort != nil {\n\t\t// Model doesn't support reasoning_effort, remove the parameter\n\t\tnewOptions.ReasoningEffort = nil\n\t\tmodified = true\n\t}\n\n\t// 2. Handle temperature parameter\n\tif !a.supportsTemperature && newOptions.Temperature != nil {\n\t\tcurrentTemp := *newOptions.Temperature\n\t\tif currentTemp != 1.0 {\n\t\t\t// Model doesn't support temperature adjustment, reset to default (1.0)\n\t\t\tdefaultTemp := 1.0\n\t\t\tnewOptions.Temperature = &defaultTemp\n\t\t\tmodified = true\n\t\t}\n\t}\n\n\tif modified {\n\t\treturn &newOptions, nil\n\t}\n\n\t// No modifications needed\n\treturn options, nil\n}\n\n// ProcessStreamChunk processes streaming chunks with reasoning content\nfunc (a *ReasoningAdapter) ProcessStreamChunk(chunkType context.StreamChunkType, data []byte) (context.StreamChunkType, []byte, error) {\n\tif a.format == ReasoningFormatNone {\n\t\t// No reasoning support, pass through\n\t\treturn chunkType, data, nil\n\t}\n\n\t// TODO: Parse reasoning_content based on format\n\t// - OpenAI o1: No visible reasoning in stream (reasoning happens internally)\n\t// - GPT-5: No visible reasoning in stream (reasoning happens internally)\n\t// - DeepSeek R1: May have ... tags or reasoning_content field\n\n\treturn chunkType, data, nil\n}\n\n// PostprocessResponse extracts reasoning content and tokens from the final response\nfunc (a *ReasoningAdapter) PostprocessResponse(response *context.CompletionResponse) (*context.CompletionResponse, error) {\n\tif a.format == ReasoningFormatNone {\n\t\t// No reasoning support\n\t\treturn response, nil\n\t}\n\n\t// Reasoning tokens are already extracted in Usage.CompletionTokensDetails.ReasoningTokens\n\t// by the OpenAI response parser, no additional processing needed for o1/GPT-5\n\n\t// TODO: For DeepSeek R1, extract visible reasoning content\n\t// - Parse ... tags from content\n\t// - Set response.ReasoningContent\n\t// - Remove tags from response.Content (keep only final answer)\n\n\treturn response, nil\n}\n" }, { "path": "agent/llm/adapters/toolcall.go", "content": "package adapters\n\nimport (\n\t\"github.com/yaoapp/yao/agent/context\"\n)\n\n// ToolCallAdapter handles tool calling capability\n// If model doesn't support native tool calls, it injects tool instructions into prompts\ntype ToolCallAdapter struct {\n\t*BaseAdapter\n\tnativeSupport bool\n}\n\n// NewToolCallAdapter creates a new tool call adapter\nfunc NewToolCallAdapter(nativeSupport bool) *ToolCallAdapter {\n\treturn &ToolCallAdapter{\n\t\tBaseAdapter: NewBaseAdapter(\"ToolCallAdapter\"),\n\t\tnativeSupport: nativeSupport,\n\t}\n}\n\n// PreprocessMessages injects tool calling instructions if not natively supported\nfunc (a *ToolCallAdapter) PreprocessMessages(messages []context.Message) ([]context.Message, error) {\n\tif a.nativeSupport {\n\t\t// Native support, no preprocessing needed\n\t\treturn messages, nil\n\t}\n\n\t// TODO: Inject tool calling instructions into system prompt\n\t// - Generate tool description prompt\n\t// - Add to system message or create new system message\n\t// - Include tool schemas and usage instructions\n\treturn messages, nil\n}\n\n// PreprocessOptions removes tool-related options if not natively supported\nfunc (a *ToolCallAdapter) PreprocessOptions(options *context.CompletionOptions) (*context.CompletionOptions, error) {\n\tif a.nativeSupport {\n\t\t// Native support, keep options as-is\n\t\treturn options, nil\n\t}\n\n\tif options == nil {\n\t\treturn options, nil\n\t}\n\n\t// Remove tool parameters for non-native models\n\tnewOptions := *options\n\tnewOptions.Tools = nil\n\tnewOptions.ToolChoice = nil\n\treturn &newOptions, nil\n}\n\n// PostprocessResponse extracts tool calls from text if not natively supported\nfunc (a *ToolCallAdapter) PostprocessResponse(response *context.CompletionResponse) (*context.CompletionResponse, error) {\n\tif a.nativeSupport {\n\t\t// Native support, response already has structured tool calls\n\t\treturn response, nil\n\t}\n\n\t// TODO: Extract tool calls from text response\n\t// - Look for JSON blocks or specific patterns\n\t// - Parse tool name and arguments\n\t// - Add to response.ToolCalls\n\treturn response, nil\n}\n" }, { "path": "agent/llm/adapters/vision.go", "content": "package adapters\n\nimport (\n\t\"encoding/base64\"\n\t\"fmt\"\n\t\"io\"\n\t\"net/http\"\n\t\"strings\"\n\t\"time\"\n\n\t\"github.com/yaoapp/yao/agent/context\"\n)\n\n// VisionAdapter handles vision (image) capability\n// If model doesn't support vision, it removes or converts image content\ntype VisionAdapter struct {\n\t*BaseAdapter\n\tnativeSupport bool\n\tformat context.VisionFormat\n}\n\n// NewVisionAdapter creates a new vision adapter\nfunc NewVisionAdapter(nativeSupport bool, format context.VisionFormat) *VisionAdapter {\n\treturn &VisionAdapter{\n\t\tBaseAdapter: NewBaseAdapter(\"VisionAdapter\"),\n\t\tnativeSupport: nativeSupport,\n\t\tformat: format,\n\t}\n}\n\n// PreprocessMessages removes or converts image content if not supported\nfunc (a *VisionAdapter) PreprocessMessages(messages []context.Message) ([]context.Message, error) {\n\tif !a.nativeSupport {\n\t\t// No vision support, remove image content\n\t\treturn a.removeImageContent(messages), nil\n\t}\n\n\t// Check if we need to convert format\n\tneedsConversion := a.format == context.VisionFormatClaude || a.format == context.VisionFormatBase64\n\n\tif !needsConversion {\n\t\t// Native support with OpenAI format or default, no preprocessing needed\n\t\treturn messages, nil\n\t}\n\n\t// Convert image_url format to Claude base64 format\n\treturn a.convertToBase64Format(messages)\n}\n\n// removeImageContent removes image content from messages\nfunc (a *VisionAdapter) removeImageContent(messages []context.Message) []context.Message {\n\tprocessed := make([]context.Message, 0, len(messages))\n\tfor _, msg := range messages {\n\t\tprocessedMsg := msg\n\n\t\t// Handle multimodal content (array of map)\n\t\tif contentParts, ok := msg.Content.([]map[string]interface{}); ok {\n\t\t\tfilteredParts := make([]map[string]interface{}, 0)\n\n\t\t\tfor _, part := range contentParts {\n\t\t\t\tpartType, _ := part[\"type\"].(string)\n\t\t\t\t// Skip image content\n\t\t\t\tif partType != \"image_url\" && partType != \"image\" {\n\t\t\t\t\tfilteredParts = append(filteredParts, part)\n\t\t\t\t}\n\t\t\t}\n\n\t\t\t// If all parts were filtered out, add placeholder text\n\t\t\tif len(filteredParts) == 0 {\n\t\t\t\tprocessedMsg.Content = \"[Image content not supported by this model]\"\n\t\t\t} else if len(filteredParts) == 1 {\n\t\t\t\tif textVal, ok := filteredParts[0][\"text\"].(string); ok {\n\t\t\t\t\tprocessedMsg.Content = textVal\n\t\t\t\t} else {\n\t\t\t\t\tprocessedMsg.Content = filteredParts\n\t\t\t\t}\n\t\t\t} else {\n\t\t\t\tprocessedMsg.Content = filteredParts\n\t\t\t}\n\t\t}\n\n\t\tprocessed = append(processed, processedMsg)\n\t}\n\n\treturn processed\n}\n\n// convertToBase64Format converts image_url format to Claude base64 format\nfunc (a *VisionAdapter) convertToBase64Format(messages []context.Message) ([]context.Message, error) {\n\tprocessed := make([]context.Message, 0, len(messages))\n\n\tfor _, msg := range messages {\n\t\tprocessedMsg := msg\n\n\t\t// Handle multimodal content\n\t\tif contentParts, ok := msg.Content.([]map[string]interface{}); ok {\n\t\t\tconvertedParts := make([]map[string]interface{}, 0)\n\n\t\t\tfor _, part := range contentParts {\n\t\t\t\tpartType, _ := part[\"type\"].(string)\n\n\t\t\t\tif partType == \"image_url\" {\n\t\t\t\t\t// Convert to base64 format\n\t\t\t\t\tconvertedPart, err := a.convertImageURLToBase64(part)\n\t\t\t\t\tif err != nil {\n\t\t\t\t\t\t// If conversion fails, skip this image\n\t\t\t\t\t\tcontinue\n\t\t\t\t\t}\n\t\t\t\t\tconvertedParts = append(convertedParts, convertedPart)\n\t\t\t\t} else {\n\t\t\t\t\t// Keep non-image parts as-is\n\t\t\t\t\tconvertedParts = append(convertedParts, part)\n\t\t\t\t}\n\t\t\t}\n\n\t\t\tprocessedMsg.Content = convertedParts\n\t\t}\n\n\t\tprocessed = append(processed, processedMsg)\n\t}\n\n\treturn processed, nil\n}\n\n// convertImageURLToBase64 converts OpenAI image_url format to Claude base64 format\nfunc (a *VisionAdapter) convertImageURLToBase64(part map[string]interface{}) (map[string]interface{}, error) {\n\t// Extract URL from image_url object\n\timageURLObj, ok := part[\"image_url\"].(map[string]interface{})\n\tif !ok {\n\t\treturn nil, fmt.Errorf(\"invalid image_url format\")\n\t}\n\n\turl, ok := imageURLObj[\"url\"].(string)\n\tif !ok || url == \"\" {\n\t\treturn nil, fmt.Errorf(\"missing or invalid URL in image_url\")\n\t}\n\n\t// Check if already base64 data URL\n\tif strings.HasPrefix(url, \"data:\") {\n\t\t// Extract media type and base64 data from data URL\n\t\t// Format: data:image/jpeg;base64,\n\t\tparts := strings.SplitN(url, \",\", 2)\n\t\tif len(parts) != 2 {\n\t\t\treturn nil, fmt.Errorf(\"invalid data URL format\")\n\t\t}\n\n\t\t// Extract media type from first part\n\t\tmediaParts := strings.Split(parts[0], \";\")\n\t\tmediaType := strings.TrimPrefix(mediaParts[0], \"data:\")\n\t\tbase64Data := parts[1]\n\n\t\treturn map[string]interface{}{\n\t\t\t\"type\": \"image\",\n\t\t\t\"source\": map[string]interface{}{\n\t\t\t\t\"type\": \"base64\",\n\t\t\t\t\"media_type\": mediaType,\n\t\t\t\t\"data\": base64Data,\n\t\t\t},\n\t\t}, nil\n\t}\n\n\t// Download image from URL and convert to base64\n\tbase64Data, mediaType, err := a.downloadAndEncodeImage(url)\n\tif err != nil {\n\t\treturn nil, fmt.Errorf(\"failed to download image: %w\", err)\n\t}\n\n\t// Return Claude/Anthropic format\n\treturn map[string]interface{}{\n\t\t\"type\": \"image\",\n\t\t\"source\": map[string]interface{}{\n\t\t\t\"type\": \"base64\",\n\t\t\t\"media_type\": mediaType,\n\t\t\t\"data\": base64Data,\n\t\t},\n\t}, nil\n}\n\n// downloadAndEncodeImage downloads an image from URL and returns base64 encoded data\nfunc (a *VisionAdapter) downloadAndEncodeImage(url string) (string, string, error) {\n\t// Create HTTP client with timeout\n\tclient := &http.Client{\n\t\tTimeout: 30 * time.Second,\n\t}\n\n\t// Download image\n\tresp, err := client.Get(url)\n\tif err != nil {\n\t\treturn \"\", \"\", fmt.Errorf(\"failed to download image: %w\", err)\n\t}\n\tdefer resp.Body.Close()\n\n\tif resp.StatusCode != 200 {\n\t\treturn \"\", \"\", fmt.Errorf(\"failed to download image: HTTP %d\", resp.StatusCode)\n\t}\n\n\t// Read image data\n\timageData, err := io.ReadAll(resp.Body)\n\tif err != nil {\n\t\treturn \"\", \"\", fmt.Errorf(\"failed to read image data: %w\", err)\n\t}\n\n\t// Detect media type from Content-Type header\n\tmediaType := resp.Header.Get(\"Content-Type\")\n\n\t// Normalize media type (remove charset and other parameters)\n\tif mediaType != \"\" {\n\t\t// Split by semicolon to remove parameters like \"; charset=utf-8\"\n\t\tif idx := strings.Index(mediaType, \";\"); idx != -1 {\n\t\t\tmediaType = strings.TrimSpace(mediaType[:idx])\n\t\t}\n\t}\n\n\tif mediaType == \"\" {\n\t\t// Fallback to detecting from URL extension or default to jpeg\n\t\turlLower := strings.ToLower(url)\n\t\tif strings.HasSuffix(urlLower, \".png\") {\n\t\t\tmediaType = \"image/png\"\n\t\t} else if strings.HasSuffix(urlLower, \".gif\") {\n\t\t\tmediaType = \"image/gif\"\n\t\t} else if strings.HasSuffix(urlLower, \".webp\") {\n\t\t\tmediaType = \"image/webp\"\n\t\t} else if strings.Contains(urlLower, \".jpg\") || strings.Contains(urlLower, \".jpeg\") {\n\t\t\tmediaType = \"image/jpeg\"\n\t\t} else {\n\t\t\t// Default to jpeg\n\t\t\tmediaType = \"image/jpeg\"\n\t\t}\n\t}\n\n\t// Encode to base64\n\tbase64Data := base64.StdEncoding.EncodeToString(imageData)\n\n\treturn base64Data, mediaType, nil\n}\n" }, { "path": "agent/llm/capabilities.go", "content": "package llm\n\nimport (\n\t\"github.com/yaoapp/gou/connector\"\n\tgoullm \"github.com/yaoapp/gou/llm\"\n)\n\n// GetCapabilities get the capabilities of a connector by connector ID\n// Reads capabilities from connector's Setting()[\"capabilities\"], with fallback to defaults.\nfunc GetCapabilities(connectorID string) *goullm.Capabilities {\n\tif connectorID == \"\" {\n\t\treturn getDefaultCapabilities()\n\t}\n\n\tconn, err := connector.Select(connectorID)\n\tif err != nil {\n\t\treturn getDefaultCapabilities()\n\t}\n\n\treturn GetCapabilitiesFromConn(conn)\n}\n\n// GetCapabilitiesFromConn get the capabilities from a connector instance\nfunc GetCapabilitiesFromConn(conn connector.Connector) *goullm.Capabilities {\n\tif conn == nil {\n\t\treturn getDefaultCapabilities()\n\t}\n\n\tsettings := conn.Setting()\n\tif settings != nil {\n\t\tif caps, ok := settings[\"capabilities\"]; ok {\n\t\t\tif capabilities, ok := caps.(*goullm.Capabilities); ok {\n\t\t\t\treturn capabilities\n\t\t\t}\n\t\t\tif capabilities, ok := caps.(goullm.Capabilities); ok {\n\t\t\t\treturn &capabilities\n\t\t\t}\n\t\t}\n\t}\n\n\treturn getDefaultCapabilities()\n}\n\n// getDefaultCapabilities returns minimal default capabilities\nfunc getDefaultCapabilities() *goullm.Capabilities {\n\treturn &goullm.Capabilities{\n\t\tVision: false,\n\t\tToolCalls: false,\n\t\tAudio: false,\n\t\tReasoning: false,\n\t\tStreaming: false,\n\t\tJSON: false,\n\t\tMultimodal: false,\n\t\tTemperatureAdjustable: true,\n\t}\n}\n\n// GetCapabilitiesMap get capabilities as map[string]interface{} for API responses\nfunc GetCapabilitiesMap(connectorID string) map[string]interface{} {\n\tcaps := GetCapabilities(connectorID)\n\tif caps == nil {\n\t\treturn nil\n\t}\n\n\treturn ToMap(caps)\n}\n\n// ToMap converts Capabilities to map[string]interface{}\nfunc ToMap(caps *goullm.Capabilities) map[string]interface{} {\n\tif caps == nil {\n\t\treturn nil\n\t}\n\n\tresult := make(map[string]interface{})\n\n\tif caps.Vision != nil {\n\t\tresult[\"vision\"] = caps.Vision\n\t}\n\n\tresult[\"audio\"] = caps.Audio\n\tresult[\"stt\"] = caps.STT\n\tresult[\"tool_calls\"] = caps.ToolCalls\n\tresult[\"reasoning\"] = caps.Reasoning\n\tresult[\"streaming\"] = caps.Streaming\n\tresult[\"json\"] = caps.JSON\n\tresult[\"multimodal\"] = caps.Multimodal\n\tresult[\"temperature_adjustable\"] = caps.TemperatureAdjustable\n\n\treturn result\n}\n" }, { "path": "agent/llm/interfaces.go", "content": "package llm\n\nimport (\n\t\"github.com/yaoapp/yao/agent/context\"\n\t\"github.com/yaoapp/yao/agent/output/message\"\n)\n\n// LLM the LLM interface\ntype LLM interface {\n\tStream(ctx *context.Context, messages []context.Message, options *context.CompletionOptions, handler message.StreamFunc) (*context.CompletionResponse, error)\n\tPost(ctx *context.Context, messages []context.Message, options *context.CompletionOptions) (*context.CompletionResponse, error)\n}\n" }, { "path": "agent/llm/jsapi.go", "content": "// Package llm provides the LLM JSAPI implementation\npackage llm\n\nimport (\n\t\"fmt\"\n\n\t\"github.com/yaoapp/gou/connector\"\n\tagentContext \"github.com/yaoapp/yao/agent/context\"\n\t\"github.com/yaoapp/yao/agent/output/message\"\n)\n\n// JSAPI implements LlmAPI interface for ctx.llm.* methods\ntype JSAPI struct {\n\tctx *agentContext.Context\n}\n\n// Ensure JSAPI implements both interfaces\nvar _ agentContext.LlmAPI = (*JSAPI)(nil)\nvar _ agentContext.LlmAPIWithCallback = (*JSAPI)(nil)\n\n// NewJSAPI creates a new JSAPI for the given context\nfunc NewJSAPI(ctx *agentContext.Context) *JSAPI {\n\treturn &JSAPI{ctx: ctx}\n}\n\n// SetJSAPIFactory registers the JSAPI factory with the context package\n// This should be called during initialization\nfunc SetJSAPIFactory() {\n\tagentContext.LlmAPIFactory = func(ctx *agentContext.Context) agentContext.LlmAPI {\n\t\treturn NewJSAPI(ctx)\n\t}\n}\n\n// Stream implements LlmAPI.Stream - calls LLM with streaming output to ctx.Writer\nfunc (api *JSAPI) Stream(connectorID string, messages []interface{}, opts map[string]interface{}) interface{} {\n\treturn api.StreamWithHandler(connectorID, messages, opts, nil)\n}\n\n// StreamWithHandler implements LlmAPIWithCallback.StreamWithHandler - calls LLM with OnMessage handler\nfunc (api *JSAPI) StreamWithHandler(connectorID string, messages []interface{}, opts map[string]interface{}, handler agentContext.OnMessageFunc) interface{} {\n\tresult := &Result{\n\t\tConnector: connectorID,\n\t}\n\n\t// Validate context\n\tif api.ctx == nil {\n\t\tresult.Error = \"context is nil\"\n\t\treturn result\n\t}\n\n\t// Get connector\n\tconn, err := connector.Select(connectorID)\n\tif err != nil {\n\t\tresult.Error = fmt.Sprintf(\"failed to select connector %s: %v\", connectorID, err)\n\t\treturn result\n\t}\n\n\t// Parse messages to context.Message format\n\tctxMessages, err := parseMessages(messages)\n\tif err != nil {\n\t\tresult.Error = fmt.Sprintf(\"failed to parse messages: %v\", err)\n\t\treturn result\n\t}\n\n\t// Build CompletionOptions from opts\n\tcompletionOptions := buildCompletionOptions(conn, opts)\n\n\t// Create LLM instance\n\tllmInstance, err := New(conn, completionOptions)\n\tif err != nil {\n\t\tresult.Error = fmt.Sprintf(\"failed to create LLM instance: %v\", err)\n\t\treturn result\n\t}\n\n\t// Create stream handler with the provided callback\n\t// Note: We pass handler directly to the stream handler instead of setting ctx.Stack.Options.OnMessage\n\t// This avoids race conditions in concurrent batch calls where multiple goroutines\n\t// would otherwise overwrite the same ctx.Stack.Options.OnMessage\n\tstreamHandler := createStreamHandlerWithCallback(api.ctx, handler)\n\n\t// Execute LLM stream call\n\tresponse, err := llmInstance.Stream(api.ctx, ctxMessages, completionOptions, streamHandler)\n\tif err != nil {\n\t\tresult.Error = fmt.Sprintf(\"LLM stream failed: %v\", err)\n\t\treturn result\n\t}\n\n\t// Set response\n\tresult.Response = response\n\n\t// Extract text content from response\n\tif response != nil {\n\t\tresult.Content = extractContent(response)\n\t}\n\n\treturn result\n}\n\n// createStreamHandlerWithCallback creates a stream handler that uses the provided callback directly\n// This is used instead of setting ctx.Stack.Options.OnMessage to avoid race conditions\n// in concurrent batch calls\nfunc createStreamHandlerWithCallback(ctx *agentContext.Context, handler agentContext.OnMessageFunc) message.StreamFunc {\n\t// Handle nil context\n\tif ctx == nil {\n\t\treturn func(chunkType message.StreamChunkType, data []byte) int {\n\t\t\treturn 0 // No-op handler when context is nil\n\t\t}\n\t}\n\n\t// Stream state for tracking message groups\n\tstate := &streamState{\n\t\tctx: ctx,\n\t\tbuffer: []byte{},\n\t\thandler: handler, // Store the handler directly in state\n\t}\n\n\treturn func(chunkType message.StreamChunkType, data []byte) int {\n\t\treturn state.handleChunk(chunkType, data)\n\t}\n}\n\n// parseMessages converts JS message array to context.Message slice\nfunc parseMessages(messages []interface{}) ([]agentContext.Message, error) {\n\tresult := make([]agentContext.Message, 0, len(messages))\n\n\tfor i, msg := range messages {\n\t\tmsgMap, ok := msg.(map[string]interface{})\n\t\tif !ok {\n\t\t\treturn nil, fmt.Errorf(\"message %d is not an object\", i)\n\t\t}\n\n\t\tctxMsg := agentContext.Message{}\n\n\t\t// Required: role\n\t\tif role, ok := msgMap[\"role\"].(string); ok {\n\t\t\tctxMsg.Role = agentContext.MessageRole(role)\n\t\t} else {\n\t\t\treturn nil, fmt.Errorf(\"message %d missing role\", i)\n\t\t}\n\n\t\t// Optional: content (can be string or array for multimodal)\n\t\tif content, ok := msgMap[\"content\"]; ok {\n\t\t\tctxMsg.Content = content\n\t\t}\n\n\t\t// Optional: name\n\t\tif name, ok := msgMap[\"name\"].(string); ok {\n\t\t\tctxMsg.Name = &name\n\t\t}\n\n\t\t// Optional: tool_calls\n\t\tif toolCalls, ok := msgMap[\"tool_calls\"]; ok {\n\t\t\tif tcArray, ok := toolCalls.([]interface{}); ok {\n\t\t\t\tctxMsg.ToolCalls = parseToolCalls(tcArray)\n\t\t\t}\n\t\t}\n\n\t\t// Optional: tool_call_id (for tool response messages)\n\t\tif toolCallID, ok := msgMap[\"tool_call_id\"].(string); ok {\n\t\t\tctxMsg.ToolCallID = &toolCallID\n\t\t}\n\n\t\tresult = append(result, ctxMsg)\n\t}\n\n\treturn result, nil\n}\n\n// parseToolCalls converts JS tool_calls array to context.ToolCall slice\nfunc parseToolCalls(toolCalls []interface{}) []agentContext.ToolCall {\n\tresult := make([]agentContext.ToolCall, 0, len(toolCalls))\n\n\tfor _, tc := range toolCalls {\n\t\ttcMap, ok := tc.(map[string]interface{})\n\t\tif !ok {\n\t\t\tcontinue\n\t\t}\n\n\t\ttoolCall := agentContext.ToolCall{}\n\n\t\tif id, ok := tcMap[\"id\"].(string); ok {\n\t\t\ttoolCall.ID = id\n\t\t}\n\t\tif typ, ok := tcMap[\"type\"].(string); ok {\n\t\t\ttoolCall.Type = agentContext.ToolCallType(typ)\n\t\t}\n\t\tif fn, ok := tcMap[\"function\"].(map[string]interface{}); ok {\n\t\t\ttoolCall.Function = agentContext.Function{}\n\t\t\tif name, ok := fn[\"name\"].(string); ok {\n\t\t\t\ttoolCall.Function.Name = name\n\t\t\t}\n\t\t\tif args, ok := fn[\"arguments\"].(string); ok {\n\t\t\t\ttoolCall.Function.Arguments = args\n\t\t\t}\n\t\t}\n\n\t\tresult = append(result, toolCall)\n\t}\n\n\treturn result\n}\n\n// buildCompletionOptions creates CompletionOptions from JS opts map\n// BuildCompletionOptions builds CompletionOptions from a connector and raw opts map.\n// Exported for reuse by gRPC handlers.\nfunc BuildCompletionOptions(conn connector.Connector, opts map[string]interface{}) *agentContext.CompletionOptions {\n\treturn buildCompletionOptions(conn, opts)\n}\n\nfunc buildCompletionOptions(conn connector.Connector, opts map[string]interface{}) *agentContext.CompletionOptions {\n\t// Get capabilities from connector\n\tcapabilities := GetCapabilitiesFromConn(conn)\n\n\tcompletionOptions := &agentContext.CompletionOptions{\n\t\tCapabilities: capabilities,\n\t}\n\n\tif opts == nil {\n\t\treturn completionOptions\n\t}\n\n\t// Temperature\n\tif temp, ok := opts[\"temperature\"].(float64); ok {\n\t\tcompletionOptions.Temperature = &temp\n\t}\n\n\t// Max tokens\n\tif maxTokens, ok := opts[\"max_tokens\"].(float64); ok {\n\t\tmt := int(maxTokens)\n\t\tcompletionOptions.MaxTokens = &mt\n\t}\n\tif maxCompletionTokens, ok := opts[\"max_completion_tokens\"].(float64); ok {\n\t\tmct := int(maxCompletionTokens)\n\t\tcompletionOptions.MaxCompletionTokens = &mct\n\t}\n\n\t// Top P\n\tif topP, ok := opts[\"top_p\"].(float64); ok {\n\t\tcompletionOptions.TopP = &topP\n\t}\n\n\t// Presence penalty\n\tif presencePenalty, ok := opts[\"presence_penalty\"].(float64); ok {\n\t\tcompletionOptions.PresencePenalty = &presencePenalty\n\t}\n\n\t// Frequency penalty\n\tif frequencyPenalty, ok := opts[\"frequency_penalty\"].(float64); ok {\n\t\tcompletionOptions.FrequencyPenalty = &frequencyPenalty\n\t}\n\n\t// Stop sequences\n\tif stop, ok := opts[\"stop\"]; ok {\n\t\tcompletionOptions.Stop = stop\n\t}\n\n\t// User\n\tif user, ok := opts[\"user\"].(string); ok {\n\t\tcompletionOptions.User = user\n\t}\n\n\t// Seed\n\tif seed, ok := opts[\"seed\"].(float64); ok {\n\t\ts := int(seed)\n\t\tcompletionOptions.Seed = &s\n\t}\n\n\t// Tools\n\tif tools, ok := opts[\"tools\"].([]interface{}); ok {\n\t\tcompletionOptions.Tools = make([]map[string]interface{}, 0, len(tools))\n\t\tfor _, tool := range tools {\n\t\t\tif toolMap, ok := tool.(map[string]interface{}); ok {\n\t\t\t\tcompletionOptions.Tools = append(completionOptions.Tools, toolMap)\n\t\t\t}\n\t\t}\n\t}\n\n\t// Tool choice\n\tif toolChoice, ok := opts[\"tool_choice\"]; ok {\n\t\tcompletionOptions.ToolChoice = toolChoice\n\t}\n\n\t// Response format\n\tif responseFormat, ok := opts[\"response_format\"].(map[string]interface{}); ok {\n\t\trf := &agentContext.ResponseFormat{}\n\t\tif rfType, ok := responseFormat[\"type\"].(string); ok {\n\t\t\trf.Type = agentContext.ResponseFormatType(rfType)\n\t\t}\n\t\tif jsonSchema, ok := responseFormat[\"json_schema\"].(map[string]interface{}); ok {\n\t\t\trf.JSONSchema = &agentContext.JSONSchema{}\n\t\t\tif name, ok := jsonSchema[\"name\"].(string); ok {\n\t\t\t\trf.JSONSchema.Name = name\n\t\t\t}\n\t\t\tif desc, ok := jsonSchema[\"description\"].(string); ok {\n\t\t\t\trf.JSONSchema.Description = desc\n\t\t\t}\n\t\t\tif schema, ok := jsonSchema[\"schema\"]; ok {\n\t\t\t\trf.JSONSchema.Schema = schema\n\t\t\t}\n\t\t\tif strict, ok := jsonSchema[\"strict\"].(bool); ok {\n\t\t\t\trf.JSONSchema.Strict = &strict\n\t\t\t}\n\t\t}\n\t\tcompletionOptions.ResponseFormat = rf\n\t}\n\n\t// Reasoning effort (for reasoning models)\n\tif reasoningEffort, ok := opts[\"reasoning_effort\"].(string); ok {\n\t\tcompletionOptions.ReasoningEffort = &reasoningEffort\n\t}\n\n\treturn completionOptions\n}\n\n// streamState manages stream handler state\ntype streamState struct {\n\tctx *agentContext.Context\n\tinMessage bool\n\tcurrentMsgID string\n\tcurrentMsgType string\n\tbuffer []byte\n\tmsgCounter int // Counter for generating message IDs when IDGenerator is nil\n\tchunkCounter int // Counter for generating chunk IDs when IDGenerator is nil\n\thandler agentContext.OnMessageFunc // Direct handler reference (avoids race condition via ctx.Stack.Options)\n}\n\n// generateMessageID generates a unique message ID\nfunc (s *streamState) generateMessageID() string {\n\tif s.ctx != nil && s.ctx.IDGenerator != nil {\n\t\treturn s.ctx.IDGenerator.GenerateMessageID()\n\t}\n\ts.msgCounter++\n\treturn fmt.Sprintf(\"M%d\", s.msgCounter)\n}\n\n// generateChunkID generates a unique chunk ID\nfunc (s *streamState) generateChunkID() string {\n\tif s.ctx != nil && s.ctx.IDGenerator != nil {\n\t\treturn s.ctx.IDGenerator.GenerateChunkID()\n\t}\n\ts.chunkCounter++\n\treturn fmt.Sprintf(\"C%d\", s.chunkCounter)\n}\n\n// handleChunk processes a single stream chunk\nfunc (s *streamState) handleChunk(chunkType message.StreamChunkType, data []byte) int {\n\tswitch chunkType {\n\tcase message.ChunkMessageStart:\n\t\ts.inMessage = true\n\t\ts.currentMsgID = s.generateMessageID()\n\t\ts.buffer = []byte{}\n\t\treturn 0\n\n\tcase message.ChunkText:\n\t\tif !s.inMessage {\n\t\t\ts.inMessage = true\n\t\t\ts.currentMsgID = s.generateMessageID()\n\t\t}\n\t\ts.currentMsgType = message.TypeText\n\t\ts.buffer = append(s.buffer, data...)\n\n\t\t// Create message\n\t\tmsg := &message.Message{\n\t\t\tChunkID: s.generateChunkID(),\n\t\t\tMessageID: s.currentMsgID,\n\t\t\tType: message.TypeText,\n\t\t\tDelta: true,\n\t\t\tProps: map[string]interface{}{\n\t\t\t\t\"content\": string(data),\n\t\t\t},\n\t\t}\n\n\t\t// Call handler directly if provided (for batch calls and single calls with callback)\n\t\t// We use direct handler instead of ctx.Stack.Options.OnMessage to avoid race conditions\n\t\t// in concurrent batch calls where multiple goroutines would overwrite the shared OnMessage\n\t\tif s.handler != nil {\n\t\t\tif ret := s.handler(msg); ret != 0 {\n\t\t\t\treturn ret\n\t\t\t}\n\t\t}\n\n\t\t// Send to output for actual message delivery to client\n\t\t// Note: ctx.Send may also call ctx.Stack.Options.OnMessage if set (for agent calls),\n\t\t// but for LLM calls we don't set OnMessage, so no double callback occurs\n\t\tif err := s.ctx.Send(msg); err != nil {\n\t\t\t// Log error but continue streaming\n\t\t\treturn 0\n\t\t}\n\t\treturn 0\n\n\tcase message.ChunkThinking:\n\t\tif !s.inMessage {\n\t\t\ts.inMessage = true\n\t\t\ts.currentMsgID = s.generateMessageID()\n\t\t}\n\t\ts.currentMsgType = message.TypeThinking\n\t\ts.buffer = append(s.buffer, data...)\n\n\t\tmsg := &message.Message{\n\t\t\tChunkID: s.generateChunkID(),\n\t\t\tMessageID: s.currentMsgID,\n\t\t\tType: message.TypeThinking,\n\t\t\tDelta: true,\n\t\t\tProps: map[string]interface{}{\n\t\t\t\t\"content\": string(data),\n\t\t\t},\n\t\t}\n\n\t\t// Call handler directly if provided\n\t\tif s.handler != nil {\n\t\t\tif ret := s.handler(msg); ret != 0 {\n\t\t\t\treturn ret\n\t\t\t}\n\t\t}\n\n\t\tif err := s.ctx.Send(msg); err != nil {\n\t\t\treturn 0\n\t\t}\n\t\treturn 0\n\n\tcase message.ChunkToolCall:\n\t\tif !s.inMessage {\n\t\t\ts.inMessage = true\n\t\t\ts.currentMsgID = s.generateMessageID()\n\t\t}\n\t\ts.currentMsgType = message.TypeToolCall\n\t\ts.buffer = append(s.buffer, data...)\n\n\t\t// Tool call chunks are more complex - parse and forward\n\t\tmsg := &message.Message{\n\t\t\tChunkID: s.generateChunkID(),\n\t\t\tMessageID: s.currentMsgID,\n\t\t\tType: message.TypeToolCall,\n\t\t\tDelta: true,\n\t\t\tProps: map[string]interface{}{\n\t\t\t\t\"raw\": string(data),\n\t\t\t},\n\t\t}\n\n\t\t// Call handler directly if provided\n\t\tif s.handler != nil {\n\t\t\tif ret := s.handler(msg); ret != 0 {\n\t\t\t\treturn ret\n\t\t\t}\n\t\t}\n\n\t\tif err := s.ctx.Send(msg); err != nil {\n\t\t\treturn 0\n\t\t}\n\t\treturn 0\n\n\tcase message.ChunkMessageEnd:\n\t\tif s.inMessage {\n\t\t\ts.inMessage = false\n\t\t\ts.currentMsgID = \"\"\n\t\t\ts.buffer = []byte{}\n\t\t}\n\t\treturn 0\n\n\tcase message.ChunkError:\n\t\t// Send error and stop\n\t\tmsg := &message.Message{\n\t\t\tType: message.TypeError,\n\t\t\tProps: map[string]interface{}{\n\t\t\t\t\"error\": string(data),\n\t\t\t},\n\t\t}\n\n\t\t// Call handler directly if provided\n\t\tif s.handler != nil {\n\t\t\ts.handler(msg)\n\t\t}\n\n\t\t_ = s.ctx.Send(msg) // Ignore error on error message\n\t\treturn 1 // Stop on error\n\n\tdefault:\n\t\t// Other chunk types (stream_start, stream_end, metadata) - ignore\n\t\treturn 0\n\t}\n}\n\n// extractContent extracts text content from CompletionResponse\nfunc extractContent(response *agentContext.CompletionResponse) string {\n\tif response == nil || response.Content == nil {\n\t\treturn \"\"\n\t}\n\n\tswitch content := response.Content.(type) {\n\tcase string:\n\t\treturn content\n\tcase []interface{}:\n\t\t// Multimodal response - extract text parts\n\t\tvar text string\n\t\tfor _, part := range content {\n\t\t\tif partMap, ok := part.(map[string]interface{}); ok {\n\t\t\t\tif partMap[\"type\"] == \"text\" {\n\t\t\t\t\tif t, ok := partMap[\"text\"].(string); ok {\n\t\t\t\t\t\ttext += t\n\t\t\t\t\t}\n\t\t\t\t}\n\t\t\t}\n\t\t}\n\t\treturn text\n\tdefault:\n\t\treturn \"\"\n\t}\n}\n\n// ============================================================================\n// Batch LLM Methods: All, Any, Race\n// ============================================================================\n\n// All executes all LLM requests concurrently and returns all results\nfunc (api *JSAPI) All(requests []interface{}) []interface{} {\n\treturn api.AllWithHandler(requests, nil)\n}\n\n// Any executes LLM requests concurrently and returns first successful result\nfunc (api *JSAPI) Any(requests []interface{}) []interface{} {\n\treturn api.AnyWithHandler(requests, nil)\n}\n\n// Race executes LLM requests concurrently and returns first completed result\nfunc (api *JSAPI) Race(requests []interface{}) []interface{} {\n\treturn api.RaceWithHandler(requests, nil)\n}\n\n// AllWithHandler executes all LLM requests with global handler\nfunc (api *JSAPI) AllWithHandler(requests []interface{}, globalHandler agentContext.LlmBatchOnMessageFunc) []interface{} {\n\tparsedRequests := api.parseRequests(requests, globalHandler)\n\treturn api.executeAll(parsedRequests)\n}\n\n// AnyWithHandler executes LLM requests and returns first success with handler\nfunc (api *JSAPI) AnyWithHandler(requests []interface{}, globalHandler agentContext.LlmBatchOnMessageFunc) []interface{} {\n\tparsedRequests := api.parseRequests(requests, globalHandler)\n\treturn api.executeAny(parsedRequests)\n}\n\n// RaceWithHandler executes LLM requests and returns first completion with handler\nfunc (api *JSAPI) RaceWithHandler(requests []interface{}, globalHandler agentContext.LlmBatchOnMessageFunc) []interface{} {\n\tparsedRequests := api.parseRequests(requests, globalHandler)\n\treturn api.executeRace(parsedRequests)\n}\n\n// parseRequests converts JS request array to internal Request slice\nfunc (api *JSAPI) parseRequests(requests []interface{}, globalHandler agentContext.LlmBatchOnMessageFunc) []*Request {\n\tresult := make([]*Request, 0, len(requests))\n\n\tfor i, req := range requests {\n\t\treqMap, ok := req.(map[string]interface{})\n\t\tif !ok {\n\t\t\tcontinue\n\t\t}\n\n\t\trequest := &Request{}\n\n\t\t// Required: connector\n\t\tif connector, ok := reqMap[\"connector\"].(string); ok {\n\t\t\trequest.Connector = connector\n\t\t} else {\n\t\t\tcontinue // Skip invalid request\n\t\t}\n\n\t\t// Required: messages\n\t\tif messages, ok := reqMap[\"messages\"].([]interface{}); ok {\n\t\t\trequest.Messages = messages\n\t\t} else {\n\t\t\tcontinue // Skip invalid request\n\t\t}\n\n\t\t// Optional: options\n\t\tif options, ok := reqMap[\"options\"].(map[string]interface{}); ok {\n\t\t\t// Remove onChunk from options if present (handled via globalHandler)\n\t\t\tdelete(options, \"onChunk\")\n\t\t\trequest.Options = options\n\t\t}\n\n\t\t// Set handler based on globalHandler\n\t\tif globalHandler != nil {\n\t\t\tindex := i\n\t\t\tconnectorID := request.Connector\n\t\t\trequest.Handler = func(msg *message.Message) int {\n\t\t\t\treturn globalHandler(connectorID, index, msg)\n\t\t\t}\n\t\t}\n\n\t\tresult = append(result, request)\n\t}\n\n\treturn result\n}\n\n// executeAll executes all requests concurrently and waits for all to complete\n// Each request uses a forked context to avoid race conditions on shared state\nfunc (api *JSAPI) executeAll(requests []*Request) []interface{} {\n\tif len(requests) == 0 {\n\t\treturn []interface{}{}\n\t}\n\n\tresults := make([]interface{}, len(requests))\n\tdone := make(chan struct{})\n\tremaining := len(requests)\n\n\tfor i, req := range requests {\n\t\tgo func(index int, request *Request) {\n\t\t\tdefer func() {\n\t\t\t\tif err := recover(); err != nil {\n\t\t\t\t\tresults[index] = &Result{\n\t\t\t\t\t\tConnector: request.Connector,\n\t\t\t\t\t\tError: fmt.Sprintf(\"panic: %v\", err),\n\t\t\t\t\t}\n\t\t\t\t}\n\t\t\t\tdone <- struct{}{}\n\t\t\t}()\n\n\t\t\t// Use forked context to avoid race conditions\n\t\t\tresults[index] = api.executeSingleRequestWithForkedContext(request)\n\t\t}(i, req)\n\t}\n\n\t// Wait for all to complete\n\tfor remaining > 0 {\n\t\t<-done\n\t\tremaining--\n\t}\n\n\treturn results\n}\n\n// executeAny executes requests and returns first successful result\n// Each request uses a forked context to avoid race conditions on shared state\nfunc (api *JSAPI) executeAny(requests []*Request) []interface{} {\n\tif len(requests) == 0 {\n\t\treturn []interface{}{}\n\t}\n\n\ttype indexedResult struct {\n\t\tindex int\n\t\tresult *Result\n\t}\n\n\tresultChan := make(chan indexedResult, len(requests))\n\tremaining := len(requests)\n\n\tfor i, req := range requests {\n\t\tgo func(index int, request *Request) {\n\t\t\tdefer func() {\n\t\t\t\tif err := recover(); err != nil {\n\t\t\t\t\tresultChan <- indexedResult{\n\t\t\t\t\t\tindex: index,\n\t\t\t\t\t\tresult: &Result{\n\t\t\t\t\t\t\tConnector: request.Connector,\n\t\t\t\t\t\t\tError: fmt.Sprintf(\"panic: %v\", err),\n\t\t\t\t\t\t},\n\t\t\t\t\t}\n\t\t\t\t}\n\t\t\t}()\n\n\t\t\t// Use forked context to avoid race conditions\n\t\t\tres := api.executeSingleRequestWithForkedContext(request)\n\t\t\tresultChan <- indexedResult{index: index, result: res.(*Result)}\n\t\t}(i, req)\n\t}\n\n\t// Wait for first success or all failures\n\tvar firstSuccess *indexedResult\n\terrors := make([]*indexedResult, 0)\n\n\tfor remaining > 0 {\n\t\tir := <-resultChan\n\t\tremaining--\n\n\t\tif ir.result.Error == \"\" {\n\t\t\t// Success!\n\t\t\tfirstSuccess = &ir\n\t\t\tbreak\n\t\t}\n\t\terrors = append(errors, &ir)\n\t}\n\n\t// Drain remaining results in background (don't block)\n\tif remaining > 0 {\n\t\tgo func(count int) {\n\t\t\tfor i := 0; i < count; i++ {\n\t\t\t\t<-resultChan\n\t\t\t}\n\t\t}(remaining)\n\t}\n\n\tif firstSuccess != nil {\n\t\treturn []interface{}{firstSuccess.result}\n\t}\n\n\t// All failed - return all errors\n\tresults := make([]interface{}, len(errors))\n\tfor i, e := range errors {\n\t\tresults[i] = e.result\n\t}\n\treturn results\n}\n\n// executeRace executes requests and returns first completed result (success or failure)\n// Each request uses a forked context to avoid race conditions on shared state\nfunc (api *JSAPI) executeRace(requests []*Request) []interface{} {\n\tif len(requests) == 0 {\n\t\treturn []interface{}{}\n\t}\n\n\tresultChan := make(chan *Result, len(requests))\n\n\tfor _, req := range requests {\n\t\tgo func(request *Request) {\n\t\t\tdefer func() {\n\t\t\t\tif err := recover(); err != nil {\n\t\t\t\t\tresultChan <- &Result{\n\t\t\t\t\t\tConnector: request.Connector,\n\t\t\t\t\t\tError: fmt.Sprintf(\"panic: %v\", err),\n\t\t\t\t\t}\n\t\t\t\t}\n\t\t\t}()\n\n\t\t\t// Use forked context to avoid race conditions\n\t\t\tres := api.executeSingleRequestWithForkedContext(request)\n\t\t\tresultChan <- res.(*Result)\n\t\t}(req)\n\t}\n\n\t// Return first result\n\tresult := <-resultChan\n\n\t// Drain remaining results in background (don't block)\n\tremaining := len(requests) - 1\n\tif remaining > 0 {\n\t\tgo func(count int) {\n\t\t\tfor i := 0; i < count; i++ {\n\t\t\t\t<-resultChan\n\t\t\t}\n\t\t}(remaining)\n\t}\n\n\treturn []interface{}{result}\n}\n\n// executeSingleRequestWithForkedContext executes a single LLM request with a forked context\n// This is used by batch operations (All/Any/Race) to avoid race conditions\n// when multiple goroutines access shared context state\nfunc (api *JSAPI) executeSingleRequestWithForkedContext(request *Request) interface{} {\n\t// Fork the context to get independent resources (IDGenerator, Logger, etc.)\n\tforkedCtx := api.ctx.Fork()\n\n\t// Create a temporary JSAPI with the forked context\n\tforkedAPI := &JSAPI{ctx: forkedCtx}\n\n\treturn forkedAPI.StreamWithHandler(request.Connector, request.Messages, request.Options, request.Handler)\n}\n" }, { "path": "agent/llm/jsapi_types.go", "content": "// Package llm provides types and utilities for LLM JSAPI\npackage llm\n\nimport (\n\tagentContext \"github.com/yaoapp/yao/agent/context\"\n)\n\n// Request represents a request to call an LLM connector\ntype Request struct {\n\tConnector string `json:\"connector\"` // Target connector ID\n\tMessages []interface{} `json:\"messages\"` // Messages to send\n\tOptions map[string]interface{} `json:\"options,omitempty\"` // LLM call options (temperature, max_tokens, etc.)\n\tHandler agentContext.OnMessageFunc `json:\"-\"` // OnMessage handler for this request (not serialized)\n}\n\n// Result represents the result of a LLM call via JSAPI\ntype Result struct {\n\tConnector string `json:\"connector\"` // Connector ID that was used\n\tResponse *agentContext.CompletionResponse `json:\"response,omitempty\"` // Full LLM response\n\tContent string `json:\"content,omitempty\"` // Extracted text content\n\tError string `json:\"error,omitempty\"` // Error message if call failed\n}\n\n// Note: LlmBatchOnMessageFunc is defined in agent/context/jsapi_llm.go\n// to avoid circular dependencies\n" }, { "path": "agent/llm/llm.go", "content": "package llm\n\nimport (\n\t\"github.com/yaoapp/gou/connector\"\n\t\"github.com/yaoapp/yao/agent/context\"\n\t\"github.com/yaoapp/yao/agent/llm/providers\"\n)\n\n// New create a new LLM instance\n// conn: connector object from connector.Select()\n// options: completion options containing capabilities and other settings\nfunc New(conn connector.Connector, options *context.CompletionOptions) (LLM, error) {\n\t// Select appropriate provider based on capabilities\n\treturn providers.SelectProvider(conn, options)\n}\n" }, { "path": "agent/llm/process.go", "content": "package llm\n\nimport (\n\t\"context\"\n\t\"encoding/json\"\n\t\"fmt\"\n\n\t\"github.com/yaoapp/gou/connector\"\n\tgouHTTP \"github.com/yaoapp/gou/http\"\n\t\"github.com/yaoapp/gou/process\"\n\t\"github.com/yaoapp/gou/runtime/v8/bridge\"\n\tagentContext \"github.com/yaoapp/yao/agent/context\"\n\t\"github.com/yaoapp/yao/agent/output/message\"\n\t\"github.com/yaoapp/yao/openapi/oauth/authorized\"\n)\n\nfunc init() {\n\tprocess.Register(\"llm.ChatCompletions\", ProcessChatCompletions)\n}\n\n// ProcessChatCompletions implements the llm.ChatCompletions Process.\n// A universal replacement for openai.chat.Completions that auto-detects\n// connector type (openai, anthropic, etc.) and routes accordingly.\n//\n// Usage:\n//\n//\tProcess(\"llm.ChatCompletions\", connector, messages)\n//\tProcess(\"llm.ChatCompletions\", connector, messages, opts)\n//\tProcess(\"llm.ChatCompletions\", connector, messages, opts, callback)\n//\n// Args:\n// - connector (string): Connector ID, any type (openai / anthropic / ...)\n// - messages ([]map): Message array, supports multimodal content (image_url, etc.)\n// - opts (map): Optional. temperature, max_tokens, etc.\n// - callback (func): Optional. Streaming callback func(data []byte) int\n//\n// Returns: OpenAI-compatible format { choices: [{ message: { role, content } }], ... }\nfunc ProcessChatCompletions(p *process.Process) interface{} {\n\tp.ValidateArgNums(2)\n\n\t// 1. Parse connector ID\n\tconnectorID := p.ArgsString(0)\n\tif connectorID == \"\" {\n\t\treturn newErrorResponse(\"llm.ChatCompletions: connector is required\")\n\t}\n\n\t// 2. Parse messages\n\trawMessages := p.ArgsArray(1)\n\tmessages := make([]map[string]interface{}, 0, len(rawMessages))\n\tfor i, v := range rawMessages {\n\t\tmsg, ok := v.(map[string]interface{})\n\t\tif !ok {\n\t\t\treturn newErrorResponse(fmt.Sprintf(\"llm.ChatCompletions: message %d is not an object\", i))\n\t\t}\n\t\tmessages = append(messages, msg)\n\t}\n\n\t// 3. Parse optional opts\n\tvar opts map[string]interface{}\n\tif p.NumOfArgs() > 2 && p.Args[2] != nil {\n\t\tif o, ok := p.Args[2].(map[string]interface{}); ok {\n\t\t\topts = o\n\t\t}\n\t}\n\n\t// 4. Parse optional callback (for streaming)\n\tvar callback func(data []byte) int\n\tif p.NumOfArgs() > 3 && p.Args[3] != nil {\n\t\tswitch cb := p.Args[3].(type) {\n\t\tcase func(data []byte) int:\n\t\t\tcallback = cb\n\t\tcase bridge.FunctionT:\n\t\t\tcallback = func(data []byte) int {\n\t\t\t\tv, err := cb.Call(string(data))\n\t\t\t\tif err != nil {\n\t\t\t\t\treturn gouHTTP.HandlerReturnError\n\t\t\t\t}\n\t\t\t\tret, ok := v.(int)\n\t\t\t\tif !ok {\n\t\t\t\t\treturn gouHTTP.HandlerReturnError\n\t\t\t\t}\n\t\t\t\treturn ret\n\t\t\t}\n\t\t}\n\t}\n\n\t// 5. Select connector\n\tconn, err := connector.Select(connectorID)\n\tif err != nil {\n\t\treturn newErrorResponse(fmt.Sprintf(\"llm.ChatCompletions: connector %s not found: %v\", connectorID, err))\n\t}\n\n\t// 6. Build completion options (reuse jsapi.go logic)\n\tcompletionOptions := buildCompletionOptions(conn, opts)\n\n\t// 7. Create LLM instance (auto-selects openai/anthropic provider)\n\tllmInstance, err := New(conn, completionOptions)\n\tif err != nil {\n\t\treturn newErrorResponse(fmt.Sprintf(\"llm.ChatCompletions: failed to create LLM: %v\", err))\n\t}\n\n\t// 8. Parse messages to context.Message format (reuse jsapi.go logic)\n\tinterfaceMessages := make([]interface{}, len(messages))\n\tfor i, m := range messages {\n\t\tinterfaceMessages[i] = m\n\t}\n\tctxMessages, err := parseMessages(interfaceMessages)\n\tif err != nil {\n\t\treturn newErrorResponse(fmt.Sprintf(\"llm.ChatCompletions: invalid messages: %v\", err))\n\t}\n\n\t// 8.1 Normalize multimodal content: convert []interface{} maps to []ContentPart\n\t// so that providers (especially Anthropic) can type-assert correctly.\n\tfor i := range ctxMessages {\n\t\tif parts, ok := ctxMessages[i].Content.([]interface{}); ok {\n\t\t\tctxMessages[i].Content = normalizeContentParts(parts)\n\t\t}\n\t}\n\n\t// 9. Build a minimal headless context for LLM call\n\tparent := p.Context\n\tif parent == nil {\n\t\tparent = context.Background()\n\t}\n\tauthInfo := authorized.ProcessAuthInfo(p)\n\tchatID := agentContext.GenChatID()\n\tctx := agentContext.New(parent, authInfo, chatID)\n\tdefer ctx.Release()\n\n\t// 10. Create stream handler\n\tvar streamHandler message.StreamFunc\n\tif callback != nil {\n\t\t// With callback: forward raw chunks to caller\n\t\tstreamHandler = func(chunkType message.StreamChunkType, data []byte) int {\n\t\t\tif chunkType == message.ChunkText || chunkType == message.ChunkThinking {\n\t\t\t\treturn callback(data)\n\t\t\t}\n\t\t\treturn 0\n\t\t}\n\t} else {\n\t\t// No callback: no-op handler, just collect final response\n\t\tstreamHandler = func(chunkType message.StreamChunkType, data []byte) int {\n\t\t\treturn 0\n\t\t}\n\t}\n\n\t// 11. Execute LLM stream call\n\tresponse, err := llmInstance.Stream(ctx, ctxMessages, completionOptions, streamHandler)\n\tif err != nil {\n\t\treturn newErrorResponse(fmt.Sprintf(\"llm.ChatCompletions: LLM call failed: %v\", err))\n\t}\n\n\t// 12. Convert CompletionResponse to OpenAI-compatible format\n\t// { choices: [{ message: { role, content } }], id, model, ... }\n\treturn toOpenAIFormat(response)\n}\n\n// toOpenAIFormat converts CompletionResponse to OpenAI chat.completions format\n// for backward compatibility with code that consumed openai.chat.Completions.\nfunc toOpenAIFormat(resp *agentContext.CompletionResponse) map[string]interface{} {\n\tif resp == nil {\n\t\treturn map[string]interface{}{\n\t\t\t\"choices\": []interface{}{},\n\t\t}\n\t}\n\n\tmsgMap := map[string]interface{}{\n\t\t\"role\": resp.Role,\n\t\t\"content\": resp.Content,\n\t}\n\tif len(resp.ToolCalls) > 0 {\n\t\tmsgMap[\"tool_calls\"] = resp.ToolCalls\n\t}\n\n\tchoice := map[string]interface{}{\n\t\t\"index\": 0,\n\t\t\"message\": msgMap,\n\t\t\"finish_reason\": \"stop\",\n\t}\n\n\tresult := map[string]interface{}{\n\t\t\"id\": resp.ID,\n\t\t\"object\": \"chat.completion\",\n\t\t\"created\": resp.Created,\n\t\t\"model\": resp.Model,\n\t\t\"choices\": []interface{}{choice},\n\t}\n\n\tif resp.Usage != nil {\n\t\tresult[\"usage\"] = resp.Usage\n\t}\n\n\treturn result\n}\n\n// newErrorResponse creates an error response in OpenAI-compatible format\nfunc newErrorResponse(errMsg string) map[string]interface{} {\n\treturn map[string]interface{}{\n\t\t\"error\": map[string]interface{}{\n\t\t\t\"message\": errMsg,\n\t\t\t\"type\": \"invalid_request_error\",\n\t\t},\n\t}\n}\n\n// normalizeContentParts converts []interface{} (raw maps from Process args)\n// to []agentContext.ContentPart (strongly typed) via JSON round-trip.\n// This is essential for providers (e.g. Anthropic) that type-assert on\n// []ContentPart to apply format-specific conversions (image_url → image).\nfunc normalizeContentParts(parts []interface{}) []agentContext.ContentPart {\n\traw, err := json.Marshal(parts)\n\tif err != nil {\n\t\treturn nil\n\t}\n\tvar typed []agentContext.ContentPart\n\tif err := json.Unmarshal(raw, &typed); err != nil {\n\t\treturn nil\n\t}\n\treturn typed\n}\n" }, { "path": "agent/llm/providers/ANTHROPIC_PROVIDER_PROPOSAL.md", "content": "# Anthropic Provider Implementation Proposal\n\n## Overview\n\nThis proposal outlines the implementation plan for native Anthropic Claude API support in Yao Agent. Currently, all LLM connectors use `type: \"openai\"`, and Anthropic detection relies on URL pattern matching, which is unreliable and architecturally incorrect.\n\n## Current Architecture\n\n```\ngou/connector/\n├── openai/ # type: \"openai\" - handles all OpenAI-compatible APIs\n├── moapi/ # type: \"moapi\"\n├── redis/ # type: \"redis\"\n└── ...\n\nyao/agent/llm/\n├── providers/\n│ ├── factory.go # SelectProvider() - selects provider based on connector type\n│ ├── base/ # Base provider implementation\n│ └── openai/ # OpenAI-compatible provider\n```\n\n**Current Flow:**\n1. All LLM connectors declare `\"type\": \"openai\"`\n2. `factory.go` uses `conn.Is(connector.OPENAI)` → always true for LLMs\n3. `DetectAPIFormat()` guesses API format by URL patterns (unreliable)\n\n## Problem Statement\n\n1. **No type distinction**: Cannot differentiate Anthropic from OpenAI at connector level\n2. **URL-based detection is fragile**: Relies on hardcoded patterns like `\"anthropic.com\"`\n3. **API incompatibility**: Anthropic API uses different:\n - Endpoint: `/messages` vs `/chat/completions`\n - Auth header: `x-api-key` vs `Bearer` token\n - Request format: `system` as separate field, `max_tokens` required\n - Response format: Different structure\n\n## Proposed Solution\n\n### Phase 1: gou/connector - Add Anthropic Connector Type\n\n**New files:**\n```\ngou/connector/\n├── anthropic/\n│ ├── anthropic.go # Connector implementation\n│ ├── types.go # Options, Capabilities structs\n│ └── defaults.go # Default model capabilities\n```\n\n**connector/types.go changes:**\n```go\nconst (\n // ... existing types\n ANTHROPIC = 7 // New connector type\n)\n```\n\n**connector/anthropic/anthropic.go:**\n```go\npackage anthropic\n\ntype Connector struct {\n id string\n file string\n Name string `json:\"name\"`\n Options Options `json:\"options\"`\n}\n\ntype Options struct {\n Host string `json:\"host,omitempty\"` // Default: https://api.anthropic.com\n Model string `json:\"model,omitempty\"` // e.g., claude-sonnet-4-5\n Key string `json:\"key\"` // API key\n Version string `json:\"version,omitempty\"` // API version, default: 2024-01-01\n Capabilities *Capabilities `json:\"capabilities,omitempty\"`\n}\n\ntype Capabilities struct {\n Vision interface{} `json:\"vision,omitempty\"`\n ToolCalls bool `json:\"tool_calls,omitempty\"`\n Streaming bool `json:\"streaming,omitempty\"`\n // ... same as openai.Capabilities\n}\n```\n\n**DSL Example:**\n```json\n{\n \"label\": \"Claude Sonnet 4.5\",\n \"type\": \"anthropic\",\n \"options\": {\n \"model\": \"claude-sonnet-4-5\",\n \"key\": \"$ENV.ANTHROPIC_API_KEY\",\n \"capabilities\": {\n \"vision\": \"claude\",\n \"tool_calls\": true,\n \"streaming\": true\n }\n }\n}\n```\n\n### Phase 2: yao/agent/llm - Add Anthropic Provider\n\n**New files:**\n```\nyao/agent/llm/providers/\n├── anthropic/\n│ ├── anthropic.go # Provider implementation\n│ └── types.go # Request/Response types\n```\n\n**anthropic/anthropic.go:**\n```go\npackage anthropic\n\ntype Provider struct {\n *base.Provider\n adapters []adapters.CapabilityAdapter\n}\n\nfunc New(conn connector.Connector, capabilities *Capabilities) *Provider\n\nfunc (p *Provider) Stream(ctx, messages, options, handler) (*CompletionResponse, error)\nfunc (p *Provider) Post(ctx, messages, options) (*CompletionResponse, error)\n\n// Internal methods\nfunc (p *Provider) buildRequestBody(messages, options, streaming) (map[string]interface{}, error)\nfunc (p *Provider) convertMessages(messages []context.Message) []map[string]interface{}\n```\n\n**Key Implementation Details:**\n\n1. **Message Conversion** (OpenAI format → Anthropic format):\n```go\n// OpenAI format:\n// {\"role\": \"system\", \"content\": \"...\"}\n// {\"role\": \"user\", \"content\": \"...\"}\n\n// Anthropic format:\n// system: \"...\" (separate field)\n// messages: [{\"role\": \"user\", \"content\": \"...\"}]\n```\n\n2. **Request Building:**\n```go\nbody := map[string]interface{}{\n \"model\": model,\n \"max_tokens\": maxTokens, // Required in Anthropic\n \"messages\": convertedMessages,\n}\nif systemPrompt != \"\" {\n body[\"system\"] = systemPrompt\n}\n```\n\n3. **HTTP Headers:**\n```go\nreq.SetHeader(\"Content-Type\", \"application/json\")\nreq.SetHeader(\"x-api-key\", apiKey) // Not Bearer token\nreq.SetHeader(\"anthropic-version\", \"2024-01-01\")\n```\n\n4. **SSE Parsing** (different from OpenAI):\n```go\n// Anthropic SSE events:\n// event: message_start\n// event: content_block_start\n// event: content_block_delta\n// event: content_block_stop\n// event: message_delta\n// event: message_stop\n```\n\n**factory.go changes:**\n```go\nfunc SelectProvider(conn connector.Connector, options *CompletionOptions) (LLM, error) {\n // ...\n \n // Check connector type directly\n if conn.Is(connector.ANTHROPIC) {\n return anthropic.New(conn, options.Capabilities), nil\n }\n \n if conn.Is(connector.OPENAI) {\n return openai.New(conn, options.Capabilities), nil\n }\n \n // Default fallback\n return openai.New(conn, options.Capabilities), nil\n}\n```\n\n## Implementation Effort\n\n| Component | Files | Lines (est.) | Effort |\n|-----------|-------|--------------|--------|\n| gou/connector/anthropic | 3 | ~250 | 2-3 hours |\n| yao/agent/llm/providers/anthropic | 2 | ~600 | 4-6 hours |\n| Tests | 4 | ~400 | 2-3 hours |\n| **Total** | **9** | **~1250** | **8-12 hours** |\n\n## Migration Path\n\n1. **Backward Compatible**: Existing `type: \"openai\"` connectors continue to work\n2. **New Connectors**: Use `type: \"anthropic\"` for direct Anthropic API access\n3. **Proxy Services**: OpenRouter, AWS Bedrock still use `type: \"openai\"` (they provide OpenAI-compatible endpoints)\n\n## Testing Strategy\n\n1. **Unit Tests**: Message conversion, request building\n2. **Integration Tests**: Real API calls (with test API key)\n3. **Connector Tests**: gou connector parsing and validation\n\n## Alternative Considered\n\n**URL-based detection in yao layer only** (current approach):\n- Pros: No gou changes needed\n- Cons: Fragile, architecturally incorrect, no connector-level validation\n\n**Conclusion**: Rejected. Proper connector type is the cleaner solution.\n\n## References\n\n- [Anthropic API Documentation](https://docs.anthropic.com/en/api)\n- [Anthropic Go SDK](https://github.com/anthropics/anthropic-sdk-go)\n- [OpenAI Compatibility Guide](https://platform.claude.com/docs/en/api/openai-sdk)\n\n## Next Steps\n\n1. Review and approve this proposal\n2. Implement gou/connector/anthropic (Phase 1)\n3. Implement yao/agent/llm/providers/anthropic (Phase 2)\n4. Update yao-init connectors to use `type: \"anthropic\"`\n5. Write tests and documentation\n" }, { "path": "agent/llm/providers/README.md", "content": "# LLM Providers Architecture (New)\n\n## Overview\n\nThis directory contains LLM provider implementations using the **Capability Adapters** pattern. The new architecture separates API format handling from capability handling.\n\n## Architecture Design\n\n```\n┌─────────────────────────────────────────────────┐\n│ LLM Provider (API Format) │\n│ - OpenAI-compatible │\n│ - Claude (TODO) │\n│ - Custom (TODO) │\n└──────────────┬──────────────────────────────────┘\n │\n ↓\n┌─────────────────────────────────────────────────┐\n│ Capability Adapters (Modular) │\n│ - ToolCallAdapter (native or prompt eng.) │\n│ - VisionAdapter (native or removal) │\n│ - AudioAdapter (native or removal) │\n│ - ReasoningAdapter (o1/R1/GPT-Think) │\n└─────────────────────────────────────────────────┘\n```\n\n## Key Concepts\n\n### 1. Provider = API Format\n\nProviders handle the **API communication format**:\n- OpenAI-compatible API (`/v1/chat/completions`)\n- Claude API (TODO)\n- Custom API formats (TODO)\n\n### 2. Adapters = Capabilities\n\nAdapters handle **model capabilities** independently:\n- **ToolCallAdapter**: Tool calling (native or prompt engineering)\n- **VisionAdapter**: Image input (native or removal/conversion)\n- **AudioAdapter**: Audio input (native or removal/conversion)\n- **ReasoningAdapter**: Reasoning content (o1/DeepSeek R1/GPT-4o thinking)\n\n## Provider Selection\n\n```go\n// factory.go\nfunc SelectProvider(conn connector.Connector, options *context.CompletionOptions) (LLM, error) {\n apiFormat := DetectAPIFormat(conn)\n \n switch apiFormat {\n case \"openai\":\n // Adapters automatically configured based on capabilities\n return openai.New(conn, options.Capabilities), nil\n case \"claude\":\n return claude.New(conn, options.Capabilities), nil\n default:\n return openai.New(conn, options.Capabilities), nil\n }\n}\n```\n\n## Directory Structure\n\n```\nproviders/\n├── factory.go # Provider selection based on API format\n├── base/ # Common functionality\n│ └── base.go\n├── openai/ # OpenAI-compatible API provider\n│ └── openai.go # Includes adapter integration\n└── README.md # This file\n\n../adapters/ # Capability adapters (separate package)\n├── adapter.go # Base interface\n├── toolcall.go # Tool calling adapter\n├── vision.go # Vision adapter\n├── audio.go # Audio adapter\n└── reasoning.go # Reasoning adapter\n```\n\n## OpenAI Provider\n\nThe OpenAI provider supports **all capabilities** through adapters:\n\n```go\ntype Provider struct {\n *base.Provider\n adapters []adapters.CapabilityAdapter\n}\n\nfunc New(conn connector.Connector, capabilities *context.ModelCapabilities) *Provider {\n return &Provider{\n Provider: base.NewProvider(conn, capabilities),\n adapters: buildAdapters(capabilities), // Auto-configured\n }\n}\n```\n\n### Adapter Pipeline\n\n**Preprocessing** (before API call):\n```\nMessages → ToolCallAdapter → VisionAdapter → AudioAdapter → API Request\n```\n\n**Streaming** (during API call):\n```\nAPI Chunk → ReasoningAdapter → ToolCallAdapter → Output\n```\n\n**Postprocessing** (after API call):\n```\nAPI Response → All Adapters → Final Response\n```\n\n## Model Examples\n\n### Full-Featured Model (GPT-4o)\n\n```yaml\n# connectors.yml\ngpt-4o:\n vision: true\n tool_calls: true\n audio: true\n reasoning: false\n```\n\n**Adapters created**:\n- ToolCallAdapter(native=true)\n- VisionAdapter(native=true)\n- AudioAdapter(native=true)\n\n### Reasoning Model with Tools (OpenAI o1)\n\n```yaml\no1-preview:\n reasoning: true\n tool_calls: true\n```\n\n**Adapters created**:\n- ToolCallAdapter(native=true)\n- ReasoningAdapter(format=openai-o1)\n\n### Reasoning Model without Tools (DeepSeek R1)\n\n```yaml\ndeepseek-reasoner:\n reasoning: true\n tool_calls: false\n```\n\n**Adapters created**:\n- ToolCallAdapter(native=false) → Uses prompt engineering\n- ReasoningAdapter(format=deepseek-r1)\n\n### Legacy Model (GPT-3.5-instruct)\n\n```yaml\ngpt-3.5-turbo-instruct:\n tool_calls: false\n vision: false\n audio: false\n```\n\n**Adapters created**:\n- ToolCallAdapter(native=false) → Prompt engineering\n- VisionAdapter(native=false) → Removes images\n- AudioAdapter(native=false) → Removes audio\n\n## Capability Adapters\n\n### ToolCallAdapter\n\n**When native=true**:\n- Passes tool definitions to API\n- Parses structured tool_calls from response\n\n**When native=false**:\n- Injects tool schemas into system prompt\n- Extracts tool calls from text response (JSON parsing)\n\n### VisionAdapter\n\n**When native=true**:\n- Passes image URLs/data directly to API\n\n**When native=false**:\n- Removes image content from messages\n- Optionally converts to text descriptions\n\n### AudioAdapter\n\n**When native=true**:\n- Passes audio data directly to API\n\n**When native=false**:\n- Removes audio content from messages\n- Optionally converts to text transcriptions\n\n### ReasoningAdapter\n\nHandles different reasoning formats:\n\n**OpenAI o1** (`reasoning_content` field):\n```json\n{\n \"delta\": {\n \"reasoning_content\": \"Let me think...\",\n \"content\": \"The answer is 42\"\n }\n}\n```\n\n**DeepSeek R1** (may have different format):\n```json\n{\n \"delta\": {\n \"content\": \"Let me think...The answer is 42\"\n }\n}\n```\n\n**GPT-4o thinking** (future):\n```json\n{\n \"delta\": {\n \"thinking\": \"Let me think...\",\n \"content\": \"The answer is 42\"\n }\n}\n```\n\n## Adding New Capabilities\n\n1. Create new adapter in `../adapters/`:\n ```go\n type NewCapabilityAdapter struct {\n *BaseAdapter\n nativeSupport bool\n }\n ```\n\n2. Implement CapabilityAdapter interface\n\n3. Add to `buildAdapters()` in `openai/openai.go`:\n ```go\n if cap.NewCapability != nil {\n result = append(result, adapters.NewNewCapabilityAdapter(*cap.NewCapability))\n }\n ```\n\n## Adding New API Format Provider\n\n1. Create new directory: `providers/newapi/`\n\n2. Implement LLM interface:\n ```go\n type Provider struct {\n *base.Provider\n adapters []adapters.CapabilityAdapter\n }\n \n func (p *Provider) Stream(...) (*CompletionResponse, error) {\n // Apply adapter preprocessing\n // Make API call\n // Apply adapter postprocessing\n }\n ```\n\n3. Update `factory.go`:\n ```go\n case \"newapi\":\n return newapi.New(conn, options.Capabilities), nil\n ```\n\n## Benefits of New Architecture\n\n1. **Separation of Concerns**:\n - Providers handle API format\n - Adapters handle capabilities\n\n2. **Code Reuse**:\n - Same adapters work across different providers\n - No duplication of capability logic\n\n3. **Easy Extension**:\n - Add new capability = add one adapter\n - Add new API = add one provider\n\n4. **Flexible Combinations**:\n - Any provider can use any adapter combination\n - Capabilities are composable\n\n5. **Clear Responsibility**:\n - Each adapter handles exactly one capability dimension\n - Easy to test and maintain\n\n## Testing Strategy\n\n### Unit Tests (per adapter)\n- Test preprocessing logic\n- Test postprocessing logic\n- Test stream chunk processing\n\n### Integration Tests (per provider)\n- Test with different adapter combinations\n- Test full request/response flow\n- Test error handling\n\n### End-to-End Tests\n- Test real API calls with different models\n- Verify capability detection\n- Verify adapter selection\n\n## Migration Notes\n\n### Old Architecture → New Architecture\n\n**Before**:\n```\nreasoning.Provider → Reasoning models (o1, R1)\nopenai.Provider → Full-featured models (GPT-4o)\nlegacy.Provider → Old models (GPT-3)\n```\n\n**After**:\n```\nopenai.Provider + adapters → ALL models\n```\n\nThe same OpenAI provider now handles all cases through different adapter combinations.\n\n" }, { "path": "agent/llm/providers/anthropic/anthropic.go", "content": "package anthropic\n\nimport (\n\tgocontext \"context\"\n\t\"fmt\"\n\t\"sort\"\n\t\"strings\"\n\t\"time\"\n\n\tjsoniter \"github.com/json-iterator/go\"\n\t\"github.com/yaoapp/gou/connector\"\n\t\"github.com/yaoapp/gou/http\"\n\tgoullm \"github.com/yaoapp/gou/llm\"\n\t\"github.com/yaoapp/kun/log\"\n\t\"github.com/yaoapp/yao/agent/context\"\n\t\"github.com/yaoapp/yao/agent/i18n\"\n\t\"github.com/yaoapp/yao/agent/llm/adapters\"\n\t\"github.com/yaoapp/yao/agent/llm/providers/base\"\n\t\"github.com/yaoapp/yao/agent/output/message\"\n)\n\n// Provider Anthropic Messages API provider\ntype Provider struct {\n\t*base.Provider\n\tadapters []adapters.CapabilityAdapter\n}\n\n// New create a new Anthropic provider\nfunc New(conn connector.Connector, capabilities *goullm.Capabilities) *Provider {\n\treturn &Provider{\n\t\tProvider: base.NewProvider(conn, capabilities),\n\t\tadapters: buildAdapters(capabilities),\n\t}\n}\n\n// buildAdapters builds capability adapters based on model capabilities\nfunc buildAdapters(cap *goullm.Capabilities) []adapters.CapabilityAdapter {\n\tif cap == nil {\n\t\treturn []adapters.CapabilityAdapter{}\n\t}\n\n\tresult := make([]adapters.CapabilityAdapter, 0)\n\n\t// Tool call adapter\n\tresult = append(result, adapters.NewToolCallAdapter(cap.ToolCalls))\n\n\t// Vision adapter\n\tvisionSupport, visionFormat := context.GetVisionSupport(cap)\n\tif visionSupport {\n\t\tresult = append(result, adapters.NewVisionAdapter(true, visionFormat))\n\t} else if cap.Vision != nil {\n\t\tresult = append(result, adapters.NewVisionAdapter(false, context.VisionFormatNone))\n\t}\n\n\t// Audio adapter\n\tresult = append(result, adapters.NewAudioAdapter(cap.Audio))\n\n\t// Reasoning adapter\n\tif cap.Reasoning {\n\t\tresult = append(result, adapters.NewReasoningAdapter(adapters.ReasoningFormatOpenAI, cap))\n\t} else {\n\t\tresult = append(result, adapters.NewReasoningAdapter(adapters.ReasoningFormatNone, cap))\n\t}\n\n\treturn result\n}\n\n// Stream stream completion from Anthropic API\nfunc (p *Provider) Stream(ctx *context.Context, messages []context.Message, options *context.CompletionOptions, handler message.StreamFunc) (*context.CompletionResponse, error) {\n\ttrace, _ := ctx.Trace()\n\tif trace != nil {\n\t\ttrace.Debug(\"Anthropic Stream: Starting stream request\", map[string]any{\n\t\t\t\"message_count\": len(messages),\n\t\t})\n\t}\n\n\tmaxRetries := 3\n\tvar lastErr error\n\n\tgoCtx := ctx.Context\n\tif ctx.Stack != nil && ctx.Stack.Options != nil && ctx.Stack.Options.Context != nil {\n\t\tgoCtx = ctx.Stack.Options.Context\n\t}\n\tif goCtx == nil {\n\t\tgoCtx = gocontext.Background()\n\t}\n\n\tcurrentMessages := make([]context.Message, len(messages))\n\tcopy(currentMessages, messages)\n\n\tfor attempt := 0; attempt < maxRetries; attempt++ {\n\t\tselect {\n\t\tcase <-goCtx.Done():\n\t\t\treturn nil, fmt.Errorf(\"context cancelled: %w\", goCtx.Err())\n\t\tdefault:\n\t\t}\n\n\t\tif ctx.Interrupt != nil {\n\t\t\tif signal := ctx.Interrupt.Peek(); signal != nil && signal.Type == context.InterruptForce {\n\t\t\t\treturn nil, fmt.Errorf(\"force interrupted by user\")\n\t\t\t}\n\t\t}\n\n\t\tif attempt > 0 {\n\t\t\tbackoff := time.Duration(1<\n\t// data: \n\tvar currentEventType string\n\n\tstreamHandler := func(data []byte) int {\n\t\tselect {\n\t\tcase <-goCtx.Done():\n\t\t\treturn http.HandlerReturnBreak\n\t\tdefault:\n\t\t}\n\n\t\tif ctx.Interrupt != nil {\n\t\t\tif signal := ctx.Interrupt.Peek(); signal != nil && signal.Type == context.InterruptForce {\n\t\t\t\treturn http.HandlerReturnBreak\n\t\t\t}\n\t\t}\n\n\t\tif len(data) == 0 {\n\t\t\treturn http.HandlerReturnOk\n\t\t}\n\n\t\tdataStr := string(data)\n\t\ttrimmed := strings.TrimSpace(dataStr)\n\n\t\tif trimmed == \"\" {\n\t\t\treturn http.HandlerReturnOk\n\t\t}\n\n\t\t// Parse event type line\n\t\t// Support both \"event: type\" (with space) and \"event:type\" (without space) formats\n\t\tif strings.HasPrefix(trimmed, \"event:\") {\n\t\t\tcurrentEventType = strings.TrimSpace(strings.TrimPrefix(trimmed, \"event:\"))\n\t\t\treturn http.HandlerReturnOk\n\t\t}\n\n\t\t// Parse data line\n\t\t// Support both \"data: {...}\" (with space) and \"data:{...}\" (without space) formats\n\t\tif !strings.HasPrefix(trimmed, \"data:\") {\n\t\t\t// Check for error response\n\t\t\tif strings.HasPrefix(trimmed, \"{\") && strings.Contains(trimmed, `\"error\"`) {\n\t\t\t\tvar apiErr APIError\n\t\t\t\tif err := jsoniter.UnmarshalFromString(trimmed, &apiErr); err == nil && apiErr.Error.Message != \"\" {\n\t\t\t\t\tif handler != nil {\n\t\t\t\t\t\thandler(message.ChunkError, []byte(apiErr.Error.Message))\n\t\t\t\t\t}\n\t\t\t\t}\n\t\t\t}\n\t\t\treturn http.HandlerReturnOk\n\t\t}\n\n\t\tjsonStr := strings.TrimSpace(strings.TrimPrefix(trimmed, \"data:\"))\n\n\t\tif jsonStr == \"\" {\n\t\t\treturn http.HandlerReturnOk\n\t\t}\n\n\t\t// Process based on event type\n\t\tswitch currentEventType {\n\t\tcase \"message_start\":\n\t\t\tvar event MessageStartEvent\n\t\t\tif err := jsoniter.UnmarshalFromString(jsonStr, &event); err == nil {\n\t\t\t\taccumulator.id = event.Message.ID\n\t\t\t\taccumulator.model = event.Message.Model\n\t\t\t\taccumulator.role = event.Message.Role\n\t\t\t\tif event.Message.Usage != nil {\n\t\t\t\t\taccumulator.usage = &message.UsageInfo{\n\t\t\t\t\t\tPromptTokens: event.Message.Usage.InputTokens,\n\t\t\t\t\t\tTotalTokens: event.Message.Usage.InputTokens,\n\t\t\t\t\t}\n\t\t\t\t}\n\t\t\t}\n\n\t\tcase \"content_block_start\":\n\t\t\tvar event ContentBlockStartEvent\n\t\t\tif err := jsoniter.UnmarshalFromString(jsonStr, &event); err == nil {\n\t\t\t\taccumulator.currentBlockIndex = event.Index\n\t\t\t\taccumulator.currentBlockType = event.ContentBlock.Type\n\n\t\t\t\tswitch event.ContentBlock.Type {\n\t\t\t\tcase \"thinking\":\n\t\t\t\t\tstartMessage(msgTracker, message.ChunkThinking, handler)\n\t\t\t\tcase \"text\":\n\t\t\t\t\tstartMessage(msgTracker, message.ChunkText, handler)\n\t\t\t\tcase \"tool_use\":\n\t\t\t\t\taccumulator.toolCalls[event.Index] = &accumulatedToolCall{\n\t\t\t\t\t\tid: event.ContentBlock.ID,\n\t\t\t\t\t\tname: event.ContentBlock.Name,\n\t\t\t\t\t}\n\t\t\t\t\ttoolCallInfo := &message.EventToolCallInfo{\n\t\t\t\t\t\tID: event.ContentBlock.ID,\n\t\t\t\t\t\tName: event.ContentBlock.Name,\n\t\t\t\t\t\tIndex: event.Index,\n\t\t\t\t\t}\n\t\t\t\t\tstartToolCallMessage(msgTracker, toolCallInfo, handler)\n\n\t\t\t\t\t// Send initial ChunkToolCall with id and function name\n\t\t\t\t\t// to match OpenAI format so CUI can resolve tool name from stored chunks\n\t\t\t\t\tif handler != nil {\n\t\t\t\t\t\ttoolCallData, _ := jsoniter.Marshal([]map[string]interface{}{\n\t\t\t\t\t\t\t{\n\t\t\t\t\t\t\t\t\"index\": event.Index,\n\t\t\t\t\t\t\t\t\"id\": event.ContentBlock.ID,\n\t\t\t\t\t\t\t\t\"type\": \"function\",\n\t\t\t\t\t\t\t\t\"function\": map[string]interface{}{\n\t\t\t\t\t\t\t\t\t\"name\": event.ContentBlock.Name,\n\t\t\t\t\t\t\t\t},\n\t\t\t\t\t\t\t},\n\t\t\t\t\t\t})\n\t\t\t\t\t\thandler(message.ChunkToolCall, toolCallData)\n\t\t\t\t\t\tincrementChunk(msgTracker)\n\t\t\t\t\t}\n\t\t\t\t}\n\t\t\t}\n\n\t\tcase \"content_block_delta\":\n\t\t\tvar event ContentBlockDeltaEvent\n\t\t\tif err := jsoniter.UnmarshalFromString(jsonStr, &event); err == nil {\n\t\t\t\tswitch event.Delta.Type {\n\t\t\t\tcase \"thinking_delta\":\n\t\t\t\t\tif event.Delta.Thinking != \"\" {\n\t\t\t\t\t\taccumulator.thinkingContent += event.Delta.Thinking\n\t\t\t\t\t\tif handler != nil {\n\t\t\t\t\t\t\thandler(message.ChunkThinking, []byte(event.Delta.Thinking))\n\t\t\t\t\t\t\tincrementChunk(msgTracker)\n\t\t\t\t\t\t}\n\t\t\t\t\t}\n\n\t\t\t\tcase \"text_delta\":\n\t\t\t\t\tif event.Delta.Text != \"\" {\n\t\t\t\t\t\taccumulator.content += event.Delta.Text\n\t\t\t\t\t\tif handler != nil {\n\t\t\t\t\t\t\thandler(message.ChunkText, []byte(event.Delta.Text))\n\t\t\t\t\t\t\tincrementChunk(msgTracker)\n\t\t\t\t\t\t}\n\t\t\t\t\t}\n\n\t\t\t\tcase \"input_json_delta\":\n\t\t\t\t\tif event.Delta.PartialJSON != \"\" {\n\t\t\t\t\t\tif tc, exists := accumulator.toolCalls[event.Index]; exists {\n\t\t\t\t\t\t\ttc.inputJSON += event.Delta.PartialJSON\n\t\t\t\t\t\t\t// Update tracker\n\t\t\t\t\t\t\tif msgTracker.active && msgTracker.toolCallInfo != nil {\n\t\t\t\t\t\t\t\tmsgTracker.toolCallInfo.Arguments = tc.inputJSON\n\t\t\t\t\t\t\t}\n\t\t\t\t\t\t}\n\t\t\t\t\t\tif handler != nil {\n\t\t\t\t\t\t\t// Send tool call delta\n\t\t\t\t\t\t\ttoolCallData, _ := jsoniter.Marshal([]map[string]interface{}{\n\t\t\t\t\t\t\t\t{\n\t\t\t\t\t\t\t\t\t\"index\": event.Index,\n\t\t\t\t\t\t\t\t\t\"function\": map[string]interface{}{\n\t\t\t\t\t\t\t\t\t\t\"arguments\": event.Delta.PartialJSON,\n\t\t\t\t\t\t\t\t\t},\n\t\t\t\t\t\t\t\t},\n\t\t\t\t\t\t\t})\n\t\t\t\t\t\t\thandler(message.ChunkToolCall, toolCallData)\n\t\t\t\t\t\t\tincrementChunk(msgTracker)\n\t\t\t\t\t\t}\n\t\t\t\t\t}\n\n\t\t\t\tcase \"signature_delta\":\n\t\t\t\t\t// Handle thinking signature delta (for extended thinking)\n\t\t\t\t\t// The signature is accumulated but not sent to handler\n\t\t\t\t\tvar sigDelta struct {\n\t\t\t\t\t\tType string `json:\"type\"`\n\t\t\t\t\t\tSignature string `json:\"signature\"`\n\t\t\t\t\t}\n\t\t\t\t\tif err := jsoniter.UnmarshalFromString(jsonStr, &struct {\n\t\t\t\t\t\tDelta *struct {\n\t\t\t\t\t\t\tSignature string `json:\"signature\"`\n\t\t\t\t\t\t} `json:\"delta\"`\n\t\t\t\t\t}{Delta: &struct {\n\t\t\t\t\t\tSignature string `json:\"signature\"`\n\t\t\t\t\t}{}}); err == nil {\n\t\t\t\t\t\t_ = sigDelta // signature tracking if needed\n\t\t\t\t\t}\n\t\t\t\t}\n\t\t\t}\n\n\t\tcase \"content_block_stop\":\n\t\t\tendMessage(msgTracker, handler)\n\n\t\tcase \"message_delta\":\n\t\t\tvar event MessageDeltaEvent\n\t\t\tif err := jsoniter.UnmarshalFromString(jsonStr, &event); err == nil {\n\t\t\t\taccumulator.stopReason = event.Delta.StopReason\n\t\t\t\tif event.Usage != nil {\n\t\t\t\t\tif accumulator.usage == nil {\n\t\t\t\t\t\taccumulator.usage = &message.UsageInfo{}\n\t\t\t\t\t}\n\t\t\t\t\taccumulator.usage.CompletionTokens = event.Usage.OutputTokens\n\t\t\t\t\taccumulator.usage.TotalTokens = accumulator.usage.PromptTokens + event.Usage.OutputTokens\n\t\t\t\t}\n\t\t\t}\n\n\t\tcase \"message_stop\":\n\t\t\t// Message complete\n\t\t\tendMessage(msgTracker, handler)\n\n\t\tcase \"ping\":\n\t\t\t// Keep-alive, ignore\n\n\t\tcase \"error\":\n\t\t\tvar apiErr struct {\n\t\t\t\tType string `json:\"type\"`\n\t\t\t\tError struct {\n\t\t\t\t\tType string `json:\"type\"`\n\t\t\t\t\tMessage string `json:\"message\"`\n\t\t\t\t} `json:\"error\"`\n\t\t\t}\n\t\t\tif err := jsoniter.UnmarshalFromString(jsonStr, &apiErr); err == nil && apiErr.Error.Message != \"\" {\n\t\t\t\tif handler != nil {\n\t\t\t\t\thandler(message.ChunkError, []byte(apiErr.Error.Message))\n\t\t\t\t}\n\t\t\t}\n\t\t}\n\n\t\treturn http.HandlerReturnOk\n\t}\n\n\t// Log request\n\tif trace != nil {\n\t\tif requestBodyJSON, marshalErr := jsoniter.Marshal(requestBody); marshalErr == nil {\n\t\t\ttrace.Debug(\"Anthropic Stream Request\", map[string]any{\n\t\t\t\t\"url\": url,\n\t\t\t\t\"body\": string(requestBodyJSON),\n\t\t\t})\n\t\t}\n\t}\n\n\t// Error buffer for non-SSE error responses\n\tvar errorBuffer strings.Builder\n\terrorDetected := false\n\n\twrappedHandler := func(data []byte) int {\n\t\tdataStr := string(data)\n\t\ttrimmed := strings.TrimSpace(dataStr)\n\n\t\tif trimmed == \"\" {\n\t\t\treturn http.HandlerReturnOk\n\t\t}\n\n\t\t// SSE event/data lines - pass to stream handler\n\t\t// Support both \"event: type\" (with space) and \"event:type\" (without space) formats\n\t\tif strings.HasPrefix(trimmed, \"event:\") || strings.HasPrefix(trimmed, \"data:\") {\n\t\t\treturn streamHandler(data)\n\t\t}\n\n\t\t// Detect JSON error response\n\t\tif strings.HasPrefix(trimmed, \"{\") && strings.Contains(dataStr, `\"error\"`) {\n\t\t\terrorDetected = true\n\t\t}\n\n\t\tif errorDetected {\n\t\t\terrorBuffer.Write(data)\n\t\t\terrorBuffer.WriteString(\"\\n\")\n\t\t\treturn http.HandlerReturnOk\n\t\t}\n\n\t\treturn streamHandler(data)\n\t}\n\n\t// Make streaming request\n\tlog.Trace(\"[LLM] Starting Anthropic Stream request: url=%s\", url)\n\terr = req.Stream(goCtx, \"POST\", requestBody, wrappedHandler)\n\t_ = streamStartTime\n\n\t// Check for captured error response\n\tif errorDetected && errorBuffer.Len() > 0 {\n\t\terrorJSON := errorBuffer.String()\n\t\tif trace != nil {\n\t\t\ttrace.Error(i18n.T(ctx.Locale, \"llm.anthropic.stream.api_error\"), map[string]any{\"response\": errorJSON})\n\t\t}\n\n\t\tvar apiErr APIError\n\t\tif parseErr := jsoniter.UnmarshalFromString(errorJSON, &apiErr); parseErr == nil && apiErr.Error.Message != \"\" {\n\t\t\terr = fmt.Errorf(\"Anthropic API error: %s (type: %s)\", apiErr.Error.Message, apiErr.Error.Type)\n\t\t} else {\n\t\t\terr = fmt.Errorf(\"Anthropic API error: %s\", strings.TrimSpace(errorJSON))\n\t\t}\n\t}\n\n\t// Handle context cancellation\n\tif err != nil && goCtx.Err() != nil {\n\t\treturn nil, fmt.Errorf(\"stream cancelled: %w\", goCtx.Err())\n\t}\n\n\tif err != nil {\n\t\tendMessage(msgTracker, handler)\n\t\tif handler != nil {\n\t\t\thandler(message.ChunkError, []byte(err.Error()))\n\t\t}\n\t\treturn nil, fmt.Errorf(\"streaming request failed: %w\", err)\n\t}\n\n\t// Check for empty response\n\tif accumulator.id == \"\" {\n\t\tendMessage(msgTracker, handler)\n\t\terrMsg := fmt.Errorf(\"no data received from Anthropic API\")\n\t\tif handler != nil {\n\t\t\thandler(message.ChunkError, []byte(errMsg.Error()))\n\t\t}\n\t\treturn nil, errMsg\n\t}\n\n\t// Build final response (convert to unified CompletionResponse)\n\tresponse := &context.CompletionResponse{\n\t\tID: accumulator.id,\n\t\tObject: \"message\",\n\t\tModel: accumulator.model,\n\t\tRole: accumulator.role,\n\t\tContent: accumulator.content,\n\t\tReasoningContent: accumulator.thinkingContent,\n\t\tFinishReason: mapStopReason(accumulator.stopReason),\n\t\tUsage: accumulator.usage,\n\t}\n\n\t// Convert accumulated tool calls\n\t// Note: tool call indices may not start at 0 (e.g. if text blocks precede tool_use blocks)\n\tif len(accumulator.toolCalls) > 0 {\n\t\t// Collect all indices and sort them to ensure deterministic order\n\t\tindices := make([]int, 0, len(accumulator.toolCalls))\n\t\tfor idx := range accumulator.toolCalls {\n\t\t\tindices = append(indices, idx)\n\t\t}\n\t\tsort.Ints(indices)\n\n\t\ttoolCalls := make([]context.ToolCall, 0, len(accumulator.toolCalls))\n\t\tfor _, idx := range indices {\n\t\t\ttc := accumulator.toolCalls[idx]\n\t\t\ttoolCalls = append(toolCalls, context.ToolCall{\n\t\t\t\tID: tc.id,\n\t\t\t\tType: \"function\",\n\t\t\t\tFunction: context.Function{\n\t\t\t\t\tName: tc.name,\n\t\t\t\t\tArguments: tc.inputJSON,\n\t\t\t\t},\n\t\t\t})\n\t\t}\n\t\tresponse.ToolCalls = toolCalls\n\t}\n\n\tendMessage(msgTracker, handler)\n\treturn response, nil\n}\n\n// Post non-streaming completion request to Anthropic API\nfunc (p *Provider) Post(ctx *context.Context, messages []context.Message, options *context.CompletionOptions) (*context.CompletionResponse, error) {\n\ttrace, _ := ctx.Trace()\n\n\tmaxRetries := 3\n\tvar lastErr error\n\n\tgoCtx := ctx.Context\n\tif ctx.Stack != nil && ctx.Stack.Options != nil && ctx.Stack.Options.Context != nil {\n\t\tgoCtx = ctx.Stack.Options.Context\n\t}\n\tif goCtx == nil {\n\t\tgoCtx = gocontext.Background()\n\t}\n\n\tcurrentMessages := make([]context.Message, len(messages))\n\tcopy(currentMessages, messages)\n\n\tfor attempt := 0; attempt < maxRetries; attempt++ {\n\t\tselect {\n\t\tcase <-goCtx.Done():\n\t\t\treturn nil, fmt.Errorf(\"context cancelled: %w\", goCtx.Err())\n\t\tdefault:\n\t\t}\n\n\t\tif attempt > 0 {\n\t\t\tbackoff := time.Duration(1< 0 {\n\t\t\tcontentBlocks := make([]map[string]interface{}, 0)\n\n\t\t\t// Add text content if present\n\t\t\tif contentStr, ok := msg.Content.(string); ok && contentStr != \"\" {\n\t\t\t\tcontentBlocks = append(contentBlocks, map[string]interface{}{\n\t\t\t\t\t\"type\": \"text\",\n\t\t\t\t\t\"text\": contentStr,\n\t\t\t\t})\n\t\t\t}\n\n\t\t\t// Add tool_use blocks\n\t\t\tfor _, tc := range msg.ToolCalls {\n\t\t\t\tvar input interface{}\n\t\t\t\tif tc.Function.Arguments != \"\" {\n\t\t\t\t\tjsoniter.UnmarshalFromString(tc.Function.Arguments, &input)\n\t\t\t\t}\n\t\t\t\tif input == nil {\n\t\t\t\t\tinput = map[string]interface{}{}\n\t\t\t\t}\n\t\t\t\tcontentBlocks = append(contentBlocks, map[string]interface{}{\n\t\t\t\t\t\"type\": \"tool_use\",\n\t\t\t\t\t\"id\": tc.ID,\n\t\t\t\t\t\"name\": tc.Function.Name,\n\t\t\t\t\t\"input\": input,\n\t\t\t\t})\n\t\t\t}\n\n\t\t\tapiMsg[\"content\"] = contentBlocks\n\t\t}\n\n\t\tapiMessages = append(apiMessages, apiMsg)\n\t}\n\n\t// Build request body\n\tbody := map[string]interface{}{\n\t\t\"model\": model,\n\t}\n\n\tif len(apiMessages) > 0 {\n\t\tbody[\"messages\"] = apiMessages\n\t}\n\n\tif systemContent != \"\" {\n\t\tbody[\"system\"] = systemContent\n\t}\n\n\tif streaming {\n\t\tbody[\"stream\"] = true\n\t}\n\n\t// max_tokens is required for Anthropic\n\tmaxTokens := 4096 // default\n\tif options.MaxTokens != nil {\n\t\tmaxTokens = *options.MaxTokens\n\t} else if options.MaxCompletionTokens != nil {\n\t\tmaxTokens = *options.MaxCompletionTokens\n\t} else if mt, ok := setting[\"max_tokens\"].(int); ok && mt > 0 {\n\t\tmaxTokens = mt\n\t}\n\tbody[\"max_tokens\"] = maxTokens\n\n\t// Temperature\n\tif options.Temperature != nil {\n\t\tbody[\"temperature\"] = *options.Temperature\n\t}\n\n\tif options.TopP != nil {\n\t\tbody[\"top_p\"] = *options.TopP\n\t}\n\n\tif options.Stop != nil {\n\t\tbody[\"stop_sequences\"] = options.Stop\n\t}\n\n\t// Tools (convert from OpenAI format to Anthropic format)\n\tif len(options.Tools) > 0 {\n\t\tanthropicTools := convertTools(options.Tools)\n\t\tif len(anthropicTools) > 0 {\n\t\t\tbody[\"tools\"] = anthropicTools\n\t\t}\n\t}\n\n\tif options.ToolChoice != nil {\n\t\tbody[\"tool_choice\"] = convertToolChoice(options.ToolChoice)\n\t}\n\n\t// Thinking configuration from connector settings\n\tif thinking, exists := setting[\"thinking\"]; exists && thinking != nil {\n\t\tbody[\"thinking\"] = thinking\n\t}\n\n\treturn body, nil\n}\n\n// convertTools converts OpenAI-format tools to Anthropic format\nfunc convertTools(tools []map[string]interface{}) []map[string]interface{} {\n\tresult := make([]map[string]interface{}, 0, len(tools))\n\tfor _, tool := range tools {\n\t\tfunction, ok := tool[\"function\"].(map[string]interface{})\n\t\tif !ok {\n\t\t\tcontinue\n\t\t}\n\n\t\tanthropicTool := map[string]interface{}{\n\t\t\t\"name\": function[\"name\"],\n\t\t}\n\t\tif desc, ok := function[\"description\"]; ok {\n\t\t\tanthropicTool[\"description\"] = desc\n\t\t}\n\t\tif params, ok := function[\"parameters\"]; ok {\n\t\t\tanthropicTool[\"input_schema\"] = params\n\t\t}\n\n\t\tresult = append(result, anthropicTool)\n\t}\n\treturn result\n}\n\n// convertToolChoice converts OpenAI tool_choice to Anthropic format\nfunc convertToolChoice(choice interface{}) interface{} {\n\tswitch v := choice.(type) {\n\tcase string:\n\t\tswitch v {\n\t\tcase \"auto\":\n\t\t\treturn map[string]interface{}{\"type\": \"auto\"}\n\t\tcase \"none\":\n\t\t\treturn map[string]interface{}{\"type\": \"none\"}\n\t\tcase \"required\":\n\t\t\treturn map[string]interface{}{\"type\": \"any\"}\n\t\t}\n\tcase map[string]interface{}:\n\t\tif fn, ok := v[\"function\"].(map[string]interface{}); ok {\n\t\t\tif name, ok := fn[\"name\"].(string); ok {\n\t\t\t\treturn map[string]interface{}{\n\t\t\t\t\t\"type\": \"tool\",\n\t\t\t\t\t\"name\": name,\n\t\t\t\t}\n\t\t\t}\n\t\t}\n\t}\n\treturn map[string]interface{}{\"type\": \"auto\"}\n}\n\n// convertImagePart converts an OpenAI image_url content part to Anthropic image format\nfunc convertImagePart(part context.ContentPart) map[string]interface{} {\n\tif part.ImageURL == nil {\n\t\treturn map[string]interface{}{\"type\": \"text\", \"text\": \"[image not available]\"}\n\t}\n\n\turl := part.ImageURL.URL\n\n\t// Check if it's a base64 data URL\n\tif strings.HasPrefix(url, \"data:\") {\n\t\t// Parse data URL: data:image/jpeg;base64,\n\t\tparts := strings.SplitN(url, \",\", 2)\n\t\tif len(parts) == 2 {\n\t\t\tmediaInfo := strings.TrimPrefix(parts[0], \"data:\")\n\t\t\tmediaInfo = strings.TrimSuffix(mediaInfo, \";base64\")\n\t\t\treturn map[string]interface{}{\n\t\t\t\t\"type\": \"image\",\n\t\t\t\t\"source\": map[string]interface{}{\n\t\t\t\t\t\"type\": \"base64\",\n\t\t\t\t\t\"media_type\": mediaInfo,\n\t\t\t\t\t\"data\": parts[1],\n\t\t\t\t},\n\t\t\t}\n\t\t}\n\t}\n\n\t// URL-based image (Anthropic supports URL images)\n\treturn map[string]interface{}{\n\t\t\"type\": \"image\",\n\t\t\"source\": map[string]interface{}{\n\t\t\t\"type\": \"url\",\n\t\t\t\"url\": url,\n\t\t},\n\t}\n}\n\n// buildAPIURL builds the API URL for Anthropic\nfunc buildAPIURL(host, endpoint string) string {\n\treturn connector.BuildAPIURL(host, endpoint)\n}\n\n// mapStopReason maps Anthropic stop_reason to OpenAI finish_reason\nfunc mapStopReason(stopReason string) string {\n\tswitch stopReason {\n\tcase \"end_turn\":\n\t\treturn \"stop\"\n\tcase \"max_tokens\":\n\t\treturn \"length\"\n\tcase \"tool_use\":\n\t\treturn \"tool_calls\"\n\tcase \"stop_sequence\":\n\t\treturn \"stop\"\n\tdefault:\n\t\treturn stopReason\n\t}\n}\n\n// Message tracker helper functions\n\nfunc startMessage(mt *messageTracker, messageType message.StreamChunkType, handler message.StreamFunc) {\n\tif mt.active {\n\t\tendMessage(mt, handler)\n\t}\n\n\tmt.active = true\n\tif mt.idGenerator != nil {\n\t\tmt.messageID = mt.idGenerator.GenerateMessageID()\n\t} else {\n\t\tmt.messageID = message.GenerateNanoID()\n\t}\n\tmt.messageType = messageType\n\tmt.startTime = time.Now().UnixMilli()\n\tmt.chunkCount = 0\n\tmt.toolCallInfo = nil\n\n\tif handler != nil {\n\t\tstartData := &message.EventMessageStartData{\n\t\t\tMessageID: mt.messageID,\n\t\t\tType: string(messageType),\n\t\t\tTimestamp: mt.startTime,\n\t\t}\n\t\tif startJSON, err := jsoniter.Marshal(startData); err == nil {\n\t\t\thandler(message.ChunkMessageStart, startJSON)\n\t\t}\n\t}\n}\n\nfunc startToolCallMessage(mt *messageTracker, toolCallInfo *message.EventToolCallInfo, handler message.StreamFunc) {\n\tif mt.active {\n\t\tendMessage(mt, handler)\n\t}\n\n\tmt.active = true\n\tif mt.idGenerator != nil {\n\t\tmt.messageID = mt.idGenerator.GenerateMessageID()\n\t} else {\n\t\tmt.messageID = message.GenerateNanoID()\n\t}\n\tmt.messageType = message.ChunkToolCall\n\tmt.startTime = time.Now().UnixMilli()\n\tmt.chunkCount = 0\n\tmt.toolCallInfo = toolCallInfo\n\n\tif handler != nil {\n\t\tstartData := &message.EventMessageStartData{\n\t\t\tMessageID: mt.messageID,\n\t\t\tType: string(message.ChunkToolCall),\n\t\t\tTimestamp: mt.startTime,\n\t\t\tToolCall: toolCallInfo,\n\t\t}\n\t\tif startJSON, err := jsoniter.Marshal(startData); err == nil {\n\t\t\thandler(message.ChunkMessageStart, startJSON)\n\t\t}\n\t}\n}\n\nfunc incrementChunk(mt *messageTracker) {\n\tif mt.active {\n\t\tmt.chunkCount++\n\t}\n}\n\nfunc endMessage(mt *messageTracker, handler message.StreamFunc) {\n\tif !mt.active {\n\t\treturn\n\t}\n\n\tif handler != nil {\n\t\tendData := &message.EventMessageEndData{\n\t\t\tMessageID: mt.messageID,\n\t\t\tType: string(mt.messageType),\n\t\t\tTimestamp: time.Now().UnixMilli(),\n\t\t\tDurationMs: time.Now().UnixMilli() - mt.startTime,\n\t\t\tChunkCount: mt.chunkCount,\n\t\t\tStatus: \"completed\",\n\t\t}\n\t\tif mt.toolCallInfo != nil {\n\t\t\tendData.ToolCall = mt.toolCallInfo\n\t\t}\n\t\tif endJSON, err := jsoniter.Marshal(endData); err == nil {\n\t\t\thandler(message.ChunkMessageEnd, endJSON)\n\t\t}\n\t}\n\n\tmt.active = false\n\tmt.messageID = \"\"\n\tmt.toolCallInfo = nil\n}\n\n// isRetryableError checks if an error is retryable\nfunc isRetryableError(err error) bool {\n\tif err == nil {\n\t\treturn false\n\t}\n\n\terrStr := err.Error()\n\tretryablePatterns := []string{\n\t\t\"timeout\",\n\t\t\"connection refused\",\n\t\t\"connection reset\",\n\t\t\"EOF\",\n\t\t\"HTTP 429\",\n\t\t\"HTTP 500\",\n\t\t\"HTTP 502\",\n\t\t\"HTTP 503\",\n\t\t\"HTTP 504\",\n\t\t\"overloaded\",\n\t}\n\n\tfor _, pattern := range retryablePatterns {\n\t\tif strings.Contains(strings.ToLower(errStr), strings.ToLower(pattern)) {\n\t\t\treturn true\n\t\t}\n\t}\n\n\treturn false\n}\n" }, { "path": "agent/llm/providers/anthropic/anthropic_test.go", "content": "package anthropic_test\n\nimport (\n\tgocontext \"context\"\n\t\"encoding/json\"\n\t\"strings\"\n\t\"testing\"\n\n\t\"github.com/yaoapp/gou/connector\"\n\tgoullm \"github.com/yaoapp/gou/llm\"\n\t\"github.com/yaoapp/yao/agent/context\"\n\t\"github.com/yaoapp/yao/agent/llm\"\n\t\"github.com/yaoapp/yao/agent/output/message\"\n\t\"github.com/yaoapp/yao/config\"\n\t\"github.com/yaoapp/yao/openapi/oauth/types\"\n\t\"github.com/yaoapp/yao/test\"\n)\n\n// testConnectorID uses the cheapest model (Claude Haiku 3) to save tokens\nconst testConnectorID = \"claude.haiku-3_0\"\n\n// TestAnthropicStreamBasic tests basic streaming completion with Anthropic API\nfunc TestAnthropicStreamBasic(t *testing.T) {\n\ttest.Prepare(t, config.Conf)\n\tdefer test.Clean()\n\n\tconn, err := connector.Select(testConnectorID)\n\tif err != nil {\n\t\tt.Fatalf(\"Failed to select connector: %v\", err)\n\t}\n\n\t// Verify it's an Anthropic connector\n\tif !conn.Is(connector.ANTHROPIC) {\n\t\tt.Fatal(\"Connector is not ANTHROPIC type\")\n\t}\n\n\toptions := &context.CompletionOptions{\n\t\tCapabilities: &goullm.Capabilities{\n\t\t\tStreaming: true,\n\t\t\tToolCalls: true,\n\t\t},\n\t}\n\n\tllmInstance, err := llm.New(conn, options)\n\tif err != nil {\n\t\tt.Fatalf(\"Failed to create LLM instance: %v\", err)\n\t}\n\n\tmessages := []context.Message{\n\t\t{\n\t\t\tRole: context.RoleUser,\n\t\t\tContent: \"Say 'Hi' in one word.\",\n\t\t},\n\t}\n\n\tmaxTokens := 10\n\toptions.MaxTokens = &maxTokens\n\n\tctx := newTestContext(\"test-anthropic-stream\", testConnectorID)\n\n\tvar chunks []string\n\thandler := func(chunkType message.StreamChunkType, data []byte) int {\n\t\tchunks = append(chunks, string(data))\n\t\tt.Logf(\"Stream chunk [%s]: %s\", chunkType, string(data))\n\t\treturn 0\n\t}\n\n\tresponse, err := llmInstance.Stream(ctx, messages, options, handler)\n\tif err != nil {\n\t\tt.Fatalf(\"Stream failed: %v\", err)\n\t}\n\n\tif response == nil {\n\t\tt.Fatal(\"Response is nil\")\n\t}\n\tif response.ID == \"\" {\n\t\tt.Error(\"Response ID is empty\")\n\t}\n\tif response.Model == \"\" {\n\t\tt.Error(\"Response Model is empty\")\n\t}\n\tif response.Content == \"\" {\n\t\tt.Error(\"Response content is empty\")\n\t}\n\tif response.FinishReason == \"\" {\n\t\tt.Error(\"FinishReason is empty\")\n\t}\n\tif response.Usage == nil {\n\t\tt.Error(\"Response Usage is nil\")\n\t} else {\n\t\tt.Logf(\"Usage: prompt=%d, completion=%d, total=%d\",\n\t\t\tresponse.Usage.PromptTokens, response.Usage.CompletionTokens, response.Usage.TotalTokens)\n\t}\n\tif len(chunks) == 0 {\n\t\tt.Error(\"No streaming chunks received\")\n\t}\n\n\tt.Logf(\"Final response content: %s\", response.Content)\n\tt.Logf(\"Total chunks received: %d\", len(chunks))\n}\n\n// TestAnthropicStreamWithToolCalls tests streaming with tool calls via Anthropic API\nfunc TestAnthropicStreamWithToolCalls(t *testing.T) {\n\ttest.Prepare(t, config.Conf)\n\tdefer test.Clean()\n\n\tconn, err := connector.Select(testConnectorID)\n\tif err != nil {\n\t\tt.Fatalf(\"Failed to select connector: %v\", err)\n\t}\n\n\toptions := &context.CompletionOptions{\n\t\tCapabilities: &goullm.Capabilities{\n\t\t\tStreaming: true,\n\t\t\tToolCalls: true,\n\t\t},\n\t}\n\n\tweatherTool := map[string]interface{}{\n\t\t\"type\": \"function\",\n\t\t\"function\": map[string]interface{}{\n\t\t\t\"name\": \"get_weather\",\n\t\t\t\"description\": \"Get the current weather for a location\",\n\t\t\t\"parameters\": map[string]interface{}{\n\t\t\t\t\"type\": \"object\",\n\t\t\t\t\"properties\": map[string]interface{}{\n\t\t\t\t\t\"location\": map[string]interface{}{\n\t\t\t\t\t\t\"type\": \"string\",\n\t\t\t\t\t\t\"description\": \"The city name, e.g. Tokyo\",\n\t\t\t\t\t},\n\t\t\t\t},\n\t\t\t\t\"required\": []string{\"location\"},\n\t\t\t},\n\t\t},\n\t}\n\n\toptions.Tools = []map[string]interface{}{weatherTool}\n\toptions.ToolChoice = \"auto\"\n\n\tllmInstance, err := llm.New(conn, options)\n\tif err != nil {\n\t\tt.Fatalf(\"Failed to create LLM instance: %v\", err)\n\t}\n\n\tmessages := []context.Message{\n\t\t{\n\t\t\tRole: context.RoleUser,\n\t\t\tContent: \"What's the weather in Tokyo?\",\n\t\t},\n\t}\n\n\tctx := newTestContext(\"test-anthropic-tool\", testConnectorID)\n\n\tvar toolCallChunks int\n\thandler := func(chunkType message.StreamChunkType, data []byte) int {\n\t\tif chunkType == message.ChunkToolCall {\n\t\t\ttoolCallChunks++\n\t\t}\n\t\tt.Logf(\"Stream chunk [%s]: %s\", chunkType, string(data))\n\t\treturn 0\n\t}\n\n\tresponse, err := llmInstance.Stream(ctx, messages, options, handler)\n\tif err != nil {\n\t\tt.Fatalf(\"Stream with tool calls failed: %v\", err)\n\t}\n\n\tif response == nil {\n\t\tt.Fatal(\"Response is nil\")\n\t}\n\n\tif len(response.ToolCalls) == 0 {\n\t\tt.Error(\"Expected tool calls but got none\")\n\t} else {\n\t\tt.Logf(\"Received %d tool call(s)\", len(response.ToolCalls))\n\t\tfor i, tc := range response.ToolCalls {\n\t\t\tt.Logf(\"Tool call %d: %s(%s)\", i, tc.Function.Name, tc.Function.Arguments)\n\n\t\t\tif tc.ID == \"\" {\n\t\t\t\tt.Errorf(\"Tool call %d missing ID\", i)\n\t\t\t}\n\t\t\tif tc.Function.Name == \"\" {\n\t\t\t\tt.Errorf(\"Tool call %d missing function name\", i)\n\t\t\t}\n\t\t\tif tc.Function.Name != \"get_weather\" {\n\t\t\t\tt.Errorf(\"Tool call %d expected 'get_weather', got '%s'\", i, tc.Function.Name)\n\t\t\t}\n\t\t\tif tc.Function.Arguments == \"\" {\n\t\t\t\tt.Errorf(\"Tool call %d missing arguments\", i)\n\t\t\t}\n\n\t\t\t// Verify arguments contain location\n\t\t\tvar args map[string]interface{}\n\t\t\tif err := json.Unmarshal([]byte(tc.Function.Arguments), &args); err == nil {\n\t\t\t\tif _, hasLocation := args[\"location\"]; !hasLocation {\n\t\t\t\t\tt.Errorf(\"Tool call %d arguments missing 'location'\", i)\n\t\t\t\t}\n\t\t\t}\n\t\t}\n\t}\n\n\tif response.FinishReason != context.FinishReasonToolCalls {\n\t\tt.Logf(\"Warning: Expected finish_reason='tool_calls', got '%s'\", response.FinishReason)\n\t}\n\n\tt.Logf(\"Final response: %+v\", response)\n}\n\n// TestAnthropicStreamRetry tests error handling with invalid API key\nfunc TestAnthropicStreamRetry(t *testing.T) {\n\ttest.Prepare(t, config.Conf)\n\tdefer test.Clean()\n\n\tconnDSL := `{\n\t\t\"type\": \"anthropic\",\n\t\t\"options\": {\n\t\t\t\"model\": \"claude-3-haiku-20240307\",\n\t\t\t\"key\": \"sk-ant-invalid-key-should-fail\"\n\t\t}\n\t}`\n\n\tconn, err := connector.New(\"anthropic\", \"test-anthropic-retry\", []byte(connDSL))\n\tif err != nil {\n\t\tt.Fatalf(\"Failed to create test connector: %v\", err)\n\t}\n\n\toptions := &context.CompletionOptions{\n\t\tCapabilities: &goullm.Capabilities{\n\t\t\tStreaming: true,\n\t\t\tToolCalls: true,\n\t\t},\n\t}\n\n\tllmInstance, err := llm.New(conn, options)\n\tif err != nil {\n\t\tt.Fatalf(\"Failed to create LLM instance: %v\", err)\n\t}\n\n\tmessages := []context.Message{\n\t\t{\n\t\t\tRole: context.RoleUser,\n\t\t\tContent: \"Test\",\n\t\t},\n\t}\n\n\tctx := newTestContext(\"test-anthropic-retry\", \"test-anthropic-retry\")\n\n\t_, err = llmInstance.Stream(ctx, messages, options, nil)\n\tif err == nil {\n\t\tt.Fatal(\"Expected error due to invalid API key, but got success\")\n\t}\n\n\terrMsg := strings.ToLower(err.Error())\n\thasExpectedError := strings.Contains(errMsg, \"401\") ||\n\t\tstrings.Contains(errMsg, \"authentication\") ||\n\t\tstrings.Contains(errMsg, \"invalid\") ||\n\t\tstrings.Contains(errMsg, \"no data received\")\n\n\tif !hasExpectedError {\n\t\tt.Errorf(\"Expected authentication error, got: %v\", err)\n\t}\n\n\tt.Logf(\"Failed as expected with error: %v\", err)\n}\n\n// ============================================================================\n// Helper Functions\n// ============================================================================\n\nfunc newTestContext(chatID, connectorID string) *context.Context {\n\tauthorized := &types.AuthorizedInfo{\n\t\tSubject: \"test-user\",\n\t\tClientID: \"test-client\",\n\t\tUserID: \"test-user-123\",\n\t\tTeamID: \"test-team-456\",\n\t\tTenantID: \"test-tenant-789\",\n\t\tSessionID: \"test-session-id\",\n\t\tConstraints: types.DataConstraints{\n\t\t\tTeamOnly: true,\n\t\t\tExtra: map[string]interface{}{\n\t\t\t\t\"test\": \"anthropic-provider\",\n\t\t\t},\n\t\t},\n\t}\n\n\tctx := context.New(gocontext.Background(), authorized, chatID)\n\tctx.AssistantID = \"test-assistant\"\n\tctx.Locale = \"en-us\"\n\tctx.Theme = \"light\"\n\tctx.Client = context.Client{\n\t\tType: \"web\",\n\t\tUserAgent: \"AnthropicProviderTest/1.0\",\n\t\tIP: \"127.0.0.1\",\n\t}\n\tctx.Referer = context.RefererAPI\n\tctx.Accept = context.AcceptStandard\n\tctx.Route = \"/api/test\"\n\tctx.Metadata = make(map[string]interface{})\n\treturn ctx\n}\n" }, { "path": "agent/llm/providers/anthropic/types.go", "content": "package anthropic\n\nimport (\n\t\"github.com/yaoapp/yao/agent/output/message\"\n)\n\n// ============================================================\n// Anthropic Messages API types\n// Reference: https://docs.anthropic.com/en/api/messages\n// ============================================================\n\n// StreamEvent represents an SSE event from Anthropic streaming API\ntype StreamEvent struct {\n\tType string `json:\"type\"`\n}\n\n// MessageStartEvent represents the message_start SSE event\ntype MessageStartEvent struct {\n\tType string `json:\"type\"`\n\tMessage MessageStart `json:\"message\"`\n}\n\n// MessageStart represents the message object in message_start event\ntype MessageStart struct {\n\tID string `json:\"id\"`\n\tType string `json:\"type\"`\n\tRole string `json:\"role\"`\n\tContent []ContentBlock `json:\"content\"`\n\tModel string `json:\"model\"`\n\tStopReason *string `json:\"stop_reason\"`\n\tStopSequence *string `json:\"stop_sequence\"`\n\tUsage *UsageInfo `json:\"usage,omitempty\"`\n}\n\n// ContentBlockStartEvent represents the content_block_start SSE event\ntype ContentBlockStartEvent struct {\n\tType string `json:\"type\"`\n\tIndex int `json:\"index\"`\n\tContentBlock ContentBlock `json:\"content_block\"`\n}\n\n// ContentBlockDeltaEvent represents the content_block_delta SSE event\ntype ContentBlockDeltaEvent struct {\n\tType string `json:\"type\"`\n\tIndex int `json:\"index\"`\n\tDelta DeltaBlock `json:\"delta\"`\n}\n\n// ContentBlockStopEvent represents the content_block_stop SSE event\ntype ContentBlockStopEvent struct {\n\tType string `json:\"type\"`\n\tIndex int `json:\"index\"`\n}\n\n// MessageDeltaEvent represents the message_delta SSE event\ntype MessageDeltaEvent struct {\n\tType string `json:\"type\"`\n\tDelta MessageDelta `json:\"delta\"`\n\tUsage *DeltaUsage `json:\"usage,omitempty\"`\n}\n\n// MessageDelta represents the delta in message_delta event\ntype MessageDelta struct {\n\tStopReason string `json:\"stop_reason,omitempty\"`\n\tStopSequence *string `json:\"stop_sequence,omitempty\"`\n}\n\n// DeltaUsage represents usage in message_delta event\ntype DeltaUsage struct {\n\tOutputTokens int `json:\"output_tokens\"`\n}\n\n// ContentBlock represents a content block in the response\ntype ContentBlock struct {\n\tType string `json:\"type\"` // \"text\", \"thinking\", \"tool_use\"\n\tText string `json:\"text,omitempty\"` // for type \"text\"\n\tThinking string `json:\"thinking,omitempty\"` // for type \"thinking\"\n\tSignature string `json:\"signature,omitempty\"` // for type \"thinking\"\n\tID string `json:\"id,omitempty\"` // for type \"tool_use\"\n\tName string `json:\"name,omitempty\"` // for type \"tool_use\"\n\tInput interface{} `json:\"input,omitempty\"` // for type \"tool_use\"\n}\n\n// DeltaBlock represents a delta block in streaming\ntype DeltaBlock struct {\n\tType string `json:\"type\"` // \"text_delta\", \"thinking_delta\", \"input_json_delta\"\n\tText string `json:\"text,omitempty\"` // for type \"text_delta\"\n\tThinking string `json:\"thinking,omitempty\"` // for type \"thinking_delta\"\n\tPartialJSON string `json:\"partial_json,omitempty\"` // for type \"input_json_delta\"\n}\n\n// UsageInfo represents token usage information\ntype UsageInfo struct {\n\tInputTokens int `json:\"input_tokens\"`\n\tOutputTokens int `json:\"output_tokens\"`\n\tCacheCreationInputTokens int `json:\"cache_creation_input_tokens,omitempty\"`\n\tCacheReadInputTokens int `json:\"cache_read_input_tokens,omitempty\"`\n}\n\n// NonStreamResponse represents the full non-streaming response from Anthropic API\ntype NonStreamResponse struct {\n\tID string `json:\"id\"`\n\tType string `json:\"type\"`\n\tRole string `json:\"role\"`\n\tContent []ContentBlock `json:\"content\"`\n\tModel string `json:\"model\"`\n\tStopReason string `json:\"stop_reason\"`\n\tStopSequence *string `json:\"stop_sequence\"`\n\tUsage *UsageInfo `json:\"usage,omitempty\"`\n}\n\n// APIError represents an error response from Anthropic API\ntype APIError struct {\n\tType string `json:\"type\"`\n\tError struct {\n\t\tType string `json:\"type\"`\n\t\tMessage string `json:\"message\"`\n\t} `json:\"error\"`\n}\n\n// streamAccumulator accumulates streaming response data\ntype streamAccumulator struct {\n\tid string\n\tmodel string\n\trole string\n\tcontent string\n\tthinkingContent string\n\tthinkingSignature string\n\ttoolCalls map[int]*accumulatedToolCall\n\tstopReason string\n\tusage *message.UsageInfo\n\n\t// Current content block tracking\n\tcurrentBlockIndex int\n\tcurrentBlockType string\n}\n\n// accumulatedToolCall accumulates a single tool call from streaming\ntype accumulatedToolCall struct {\n\tid string\n\tname string\n\tinputJSON string\n}\n\n// messageTracker tracks message lifecycle for stream events\ntype messageTracker struct {\n\tactive bool\n\tmessageID string\n\tmessageType message.StreamChunkType\n\tstartTime int64\n\tchunkCount int\n\ttoolCallInfo *message.EventToolCallInfo\n\tidGenerator *message.IDGenerator\n}\n" }, { "path": "agent/llm/providers/base/base.go", "content": "package base\n\nimport (\n\t\"fmt\"\n\n\t\"github.com/yaoapp/gou/connector\"\n\t\"github.com/yaoapp/gou/llm\"\n\t\"github.com/yaoapp/yao/agent/context\"\n)\n\n// Provider base provider implementation\n// Provides common functionality for all LLM providers\ntype Provider struct {\n\tConnector connector.Connector\n\tCapabilities *llm.Capabilities\n}\n\n// NewProvider create a new base provider\nfunc NewProvider(conn connector.Connector, capabilities *llm.Capabilities) *Provider {\n\treturn &Provider{\n\t\tConnector: conn,\n\t\tCapabilities: capabilities,\n\t}\n}\n\n// PreprocessMessages preprocess messages before sending to LLM\n// Handles vision messages, audio messages, tool messages, etc.\n// Filters out unsupported content types based on model capabilities\nfunc (p *Provider) PreprocessMessages(messages []context.Message) ([]context.Message, error) {\n\tprocessed := make([]context.Message, 0, len(messages))\n\n\tfor _, msg := range messages {\n\t\tprocessedMsg := msg\n\n\t\t// Handle multimodal content (array of ContentPart)\n\t\tif contentParts, ok := msg.Content.([]context.ContentPart); ok {\n\t\t\tfilteredParts := make([]context.ContentPart, 0, len(contentParts))\n\n\t\t\tfor _, part := range contentParts {\n\t\t\t\t// Filter vision content if not supported\n\t\t\t\tif part.Type == context.ContentImageURL {\n\t\t\t\t\tif !p.SupportsVision() {\n\t\t\t\t\t\t// Skip image content if vision not supported\n\t\t\t\t\t\tcontinue\n\t\t\t\t\t}\n\t\t\t\t}\n\n\t\t\t\t// Filter audio content if not supported\n\t\t\t\tif part.Type == context.ContentInputAudio {\n\t\t\t\t\tif !p.SupportsAudio() {\n\t\t\t\t\t\t// Skip audio content if audio not supported\n\t\t\t\t\t\tcontinue\n\t\t\t\t\t}\n\t\t\t\t}\n\n\t\t\t\tfilteredParts = append(filteredParts, part)\n\t\t\t}\n\n\t\t\t// If all parts were filtered out, convert to text message\n\t\t\tif len(filteredParts) == 0 {\n\t\t\t\tprocessedMsg.Content = \"[Content not supported by this model]\"\n\t\t\t} else {\n\t\t\t\tprocessedMsg.Content = filteredParts\n\t\t\t}\n\t\t}\n\n\t\tprocessed = append(processed, processedMsg)\n\t}\n\n\treturn processed, nil\n}\n\n// SupportsVision check if this provider supports vision\nfunc (p *Provider) SupportsVision() bool {\n\tif p.Capabilities == nil {\n\t\treturn false\n\t}\n\tsupported, _ := context.GetVisionSupport(p.Capabilities)\n\treturn supported\n}\n\n// SupportsAudio check if this provider supports audio\nfunc (p *Provider) SupportsAudio() bool {\n\treturn p.Capabilities != nil && p.Capabilities.Audio\n}\n\n// SupportsTools check if this provider supports tool calls\nfunc (p *Provider) SupportsTools() bool {\n\treturn p.Capabilities != nil && p.Capabilities.ToolCalls\n}\n\n// SupportsStreaming check if this provider supports streaming\nfunc (p *Provider) SupportsStreaming() bool {\n\treturn p.Capabilities != nil && p.Capabilities.Streaming\n}\n\n// SupportsJSON check if this provider supports JSON mode\nfunc (p *Provider) SupportsJSON() bool {\n\treturn p.Capabilities != nil && p.Capabilities.JSON\n}\n\n// SupportsReasoning check if this provider supports reasoning mode\nfunc (p *Provider) SupportsReasoning() bool {\n\treturn p.Capabilities != nil && p.Capabilities.Reasoning\n}\n\n// GetConnectorSetting gets a setting value from the connector\nfunc (p *Provider) GetConnectorSetting(key string) (interface{}, error) {\n\tif p.Connector == nil {\n\t\treturn nil, fmt.Errorf(\"connector is nil\")\n\t}\n\n\tsettings := p.Connector.Setting()\n\tif settings == nil {\n\t\treturn nil, fmt.Errorf(\"connector settings are nil\")\n\t}\n\n\tvalue, exists := settings[key]\n\tif !exists {\n\t\treturn nil, fmt.Errorf(\"setting '%s' not found\", key)\n\t}\n\n\treturn value, nil\n}\n\n// GetConnectorStringSetting gets a string setting value from the connector\nfunc (p *Provider) GetConnectorStringSetting(key string) (string, error) {\n\tvalue, err := p.GetConnectorSetting(key)\n\tif err != nil {\n\t\treturn \"\", err\n\t}\n\n\tstrValue, ok := value.(string)\n\tif !ok {\n\t\treturn \"\", fmt.Errorf(\"setting '%s' is not a string\", key)\n\t}\n\n\treturn strValue, nil\n}\n\n// GetModel gets the model name from connector settings\nfunc (p *Provider) GetModel() (string, error) {\n\treturn p.GetConnectorStringSetting(\"model\")\n}\n\n// GetAPIKey gets the API key from connector settings\nfunc (p *Provider) GetAPIKey() (string, error) {\n\treturn p.GetConnectorStringSetting(\"key\")\n}\n\n// GetHost gets the host URL from connector settings\nfunc (p *Provider) GetHost() (string, error) {\n\treturn p.GetConnectorStringSetting(\"host\")\n}\n" }, { "path": "agent/llm/providers/factory.go", "content": "package providers\n\nimport (\n\t\"fmt\"\n\n\t\"github.com/yaoapp/gou/connector\"\n\t\"github.com/yaoapp/yao/agent/context\"\n\t\"github.com/yaoapp/yao/agent/llm/providers/anthropic\"\n\t\"github.com/yaoapp/yao/agent/llm/providers/openai\"\n\t\"github.com/yaoapp/yao/agent/output/message\"\n)\n\n// LLM interface (copied to avoid import cycle)\ntype LLM interface {\n\tStream(ctx *context.Context, messages []context.Message, options *context.CompletionOptions, handler message.StreamFunc) (*context.CompletionResponse, error)\n\tPost(ctx *context.Context, messages []context.Message, options *context.CompletionOptions) (*context.CompletionResponse, error)\n}\n\n// SelectProvider selects the appropriate provider based on API format and capabilities\n// The new architecture uses capability adapters to handle different model features\nfunc SelectProvider(conn connector.Connector, options *context.CompletionOptions) (LLM, error) {\n\tif options == nil {\n\t\treturn nil, fmt.Errorf(\"options are required\")\n\t}\n\n\tif options.Capabilities == nil {\n\t\treturn nil, fmt.Errorf(\"capabilities are required\")\n\t}\n\n\t// Detect API format\n\tapiFormat := DetectAPIFormat(conn)\n\n\t// Select provider based on API format\n\tswitch apiFormat {\n\tcase \"openai\":\n\t\t// OpenAI-compatible API\n\t\t// Capability adapters will handle:\n\t\t// - Tool calling (native or prompt engineering)\n\t\t// - Vision (native or removal)\n\t\t// - Audio (native or removal)\n\t\t// - Reasoning (o1, GPT-4o thinking, etc.)\n\t\treturn openai.New(conn, options.Capabilities), nil\n\n\tcase \"anthropic\":\n\t\treturn anthropic.New(conn, options.Capabilities), nil\n\n\tdefault:\n\t\t// Default to OpenAI-compatible provider\n\t\treturn openai.New(conn, options.Capabilities), nil\n\t}\n}\n\n// DetectAPIFormat detects the API format from connector\nfunc DetectAPIFormat(conn connector.Connector) string {\n\t// Check connector type directly\n\tif conn.Is(connector.ANTHROPIC) {\n\t\treturn \"anthropic\"\n\t}\n\n\tif conn.Is(connector.OPENAI) {\n\t\treturn \"openai\"\n\t}\n\n\t// Check connector settings for host URL patterns as fallback\n\tsettings := conn.Setting()\n\tif settings != nil {\n\t\tif host, ok := settings[\"host\"].(string); ok {\n\t\t\tif contains(host, \"anthropic.com\") || contains(host, \"api.kimi.com/coding\") {\n\t\t\t\treturn \"anthropic\"\n\t\t\t}\n\t\t\tif contains(host, \"deepseek.com\") {\n\t\t\t\treturn \"openai\"\n\t\t\t}\n\t\t}\n\t}\n\n\t// Default to OpenAI-compatible\n\treturn \"openai\"\n}\n\n// contains checks if a string contains a substring (case-insensitive helper)\nfunc contains(s, substr string) bool {\n\treturn len(s) >= len(substr) && (s == substr || len(s) > len(substr) &&\n\t\t(s[:len(substr)] == substr || s[len(s)-len(substr):] == substr ||\n\t\t\tfindSubstring(s, substr)))\n}\n\nfunc findSubstring(s, substr string) bool {\n\tfor i := 0; i <= len(s)-len(substr); i++ {\n\t\tif s[i:i+len(substr)] == substr {\n\t\t\treturn true\n\t\t}\n\t}\n\treturn false\n}\n" }, { "path": "agent/llm/providers/openai/claude_test.go", "content": "package openai_test\n\n// import (\n// \tgocontext \"context\"\n// \t\"testing\"\n\n// \t\"github.com/yaoapp/gou/connector\"\n// \t\"github.com/yaoapp/gou/connector/openai\"\n// \t\"github.com/yaoapp/yao/agent/context\"\n// \t\"github.com/yaoapp/yao/agent/llm\"\n// \t\"github.com/yaoapp/yao/agent/output/message\"\n// \t\"github.com/yaoapp/yao/config\"\n// \t\"github.com/yaoapp/yao/openapi/oauth/types\"\n// \t\"github.com/yaoapp/yao/test\"\n// )\n\n// // newClaudeTestContext creates a real Context for testing Claude provider\n// func newClaudeTestContext(chatID, connectorID string) *context.Context {\n// \tauthorized := &types.AuthorizedInfo{\n// \t\tSubject: \"test-user\",\n// \t\tClientID: \"test-client\",\n// \t\tUserID: \"test-user-123\",\n// \t\tTeamID: \"test-team-456\",\n// \t\tTenantID: \"test-tenant-789\",\n// \t\tSessionID: \"test-session-id\",\n// \t\tConstraints: types.DataConstraints{\n// \t\t\tTeamOnly: true,\n// \t\t\tExtra: map[string]interface{}{\n// \t\t\t\t\"test\": \"claude-provider\",\n// \t\t\t},\n// \t\t},\n// \t}\n\n// \tctx := context.New(gocontext.Background(), authorized, chatID)\n// \tctx.AssistantID = \"test-assistant\"\n// \tctx.Locale = \"en-us\"\n// \tctx.Theme = \"light\"\n// \tctx.Client = context.Client{\n// \t\tType: \"web\",\n// \t\tUserAgent: \"ClaudeProviderTest/1.0\",\n// \t\tIP: \"127.0.0.1\",\n// \t}\n// \tctx.Referer = context.RefererAPI\n// \tctx.Accept = context.AcceptStandard\n// \tctx.Route = \"/api/test\"\n// \tctx.Metadata = make(map[string]interface{})\n// \treturn ctx\n// }\n\n// // TestClaudeSonnet4StreamBasic tests basic streaming completion with Claude Sonnet 4\n// func TestClaudeSonnet4StreamBasic(t *testing.T) {\n// \ttest.Prepare(t, config.Conf)\n// \tdefer test.Clean()\n\n// \tconn, err := connector.Select(\"claude.sonnet-4_0\")\n// \tif err != nil {\n// \t\tt.Fatalf(\"Failed to select connector: %v\", err)\n// \t}\n\n// \toptions := &context.CompletionOptions{\n// \t\tCapabilities: &openai.Capabilities{\n// \t\t\tStreaming: true,\n// \t\t\tReasoning: false, // Claude Sonnet 4 (non-thinking) doesn't expose reasoning\n// \t\t\tToolCalls: true,\n// \t\t\tVision: \"claude\", // Claude requires base64 format\n// \t\t\tMultimodal: true,\n// \t\t},\n// \t}\n\n// \tllmInstance, err := llm.New(conn, options)\n// \tif err != nil {\n// \t\tt.Fatalf(\"Failed to create LLM instance: %v\", err)\n// \t}\n\n// \tmessages := []context.Message{\n// \t\t{\n// \t\t\tRole: context.RoleUser,\n// \t\t\tContent: \"What is 3+3? Reply with just the number.\",\n// \t\t},\n// \t}\n\n// \tmaxTokens := 100\n// \toptions.MaxTokens = &maxTokens\n\n// \tctx := newClaudeTestContext(\"test-claude-sonnet4-basic\", \"claude.sonnet-4_0\")\n\n// \tvar chunks []string\n// \thandler := func(chunkType message.StreamChunkType, data []byte) int {\n// \t\tchunks = append(chunks, string(data))\n// \t\tt.Logf(\"Stream chunk [%s]: %s\", chunkType, string(data))\n// \t\treturn 0\n// \t}\n\n// \tresponse, err := llmInstance.Stream(ctx, messages, options, handler)\n// \tif err != nil {\n// \t\tt.Fatalf(\"Stream failed: %v\", err)\n// \t}\n\n// \tif response == nil {\n// \t\tt.Fatal(\"Response is nil\")\n// \t}\n\n// \t// Basic validation\n// \tif response.ID == \"\" {\n// \t\tt.Error(\"Response ID is empty\")\n// \t}\n// \tif response.Model == \"\" {\n// \t\tt.Error(\"Response Model is empty\")\n// \t}\n\n// \t// Validate content\n// \tcontentStr, ok := response.Content.(string)\n// \tif !ok {\n// \t\tt.Errorf(\"Content is not a string: %T\", response.Content)\n// \t}\n// \tif len(contentStr) == 0 {\n// \t\tt.Error(\"Content is empty\")\n// \t}\n\n// \tt.Logf(\"Response content: %v\", response.Content)\n// \tt.Logf(\"Usage: prompt=%d, completion=%d, total=%d\",\n// \t\tresponse.Usage.PromptTokens, response.Usage.CompletionTokens, response.Usage.TotalTokens)\n\n// \tt.Logf(\"Final response: %+v\", response)\n// \tt.Logf(\"Total chunks received: %d\", len(chunks))\n// }\n\n// // TestClaudeSonnet4PostBasic tests non-streaming completion\n// func TestClaudeSonnet4PostBasic(t *testing.T) {\n// \ttest.Prepare(t, config.Conf)\n// \tdefer test.Clean()\n\n// \tconn, err := connector.Select(\"claude.sonnet-4_0\")\n// \tif err != nil {\n// \t\tt.Fatalf(\"Failed to select connector: %v\", err)\n// \t}\n\n// \toptions := &context.CompletionOptions{\n// \t\tCapabilities: &openai.Capabilities{\n// \t\t\tStreaming: false,\n// \t\t\tReasoning: false,\n// \t\t\tToolCalls: true,\n// \t\t\tVision: \"claude\", // Claude requires base64 format\n// \t\t\tMultimodal: true,\n// \t\t},\n// \t}\n\n// \tllmInstance, err := llm.New(conn, options)\n// \tif err != nil {\n// \t\tt.Fatalf(\"Failed to create LLM instance: %v\", err)\n// \t}\n\n// \tmessages := []context.Message{\n// \t\t{\n// \t\t\tRole: context.RoleUser,\n// \t\t\tContent: \"What is 4+4? Reply with just the number.\",\n// \t\t},\n// \t}\n\n// \tmaxTokens := 100\n// \toptions.MaxTokens = &maxTokens\n\n// \tctx := newClaudeTestContext(\"test-claude-sonnet4-post\", \"claude.sonnet-4_0\")\n\n// \tresponse, err := llmInstance.Post(ctx, messages, options)\n// \tif err != nil {\n// \t\tt.Fatalf(\"Post failed: %v\", err)\n// \t}\n\n// \tif response == nil {\n// \t\tt.Fatal(\"Response is nil\")\n// \t}\n\n// \t// Validate content\n// \tcontentStr, ok := response.Content.(string)\n// \tif !ok {\n// \t\tt.Fatalf(\"Content is not a string: %T\", response.Content)\n// \t}\n\n// \tt.Logf(\"Response content: %s\", contentStr)\n// \tt.Logf(\"Usage: prompt=%d, completion=%d, total=%d\",\n// \t\tresponse.Usage.PromptTokens, response.Usage.CompletionTokens, response.Usage.TotalTokens)\n\n// \t// Basic content validation\n// \tif len(contentStr) == 0 {\n// \t\tt.Error(\"Content is empty\")\n// \t}\n\n// \tt.Logf(\"Response: %+v\", response)\n// }\n\n// // TestClaudeSonnet4WithToolCalls tests tool calling capability\n// func TestClaudeSonnet4WithToolCalls(t *testing.T) {\n// \ttest.Prepare(t, config.Conf)\n// \tdefer test.Clean()\n\n// \tconn, err := connector.Select(\"claude.sonnet-4_0\")\n// \tif err != nil {\n// \t\tt.Fatalf(\"Failed to select connector: %v\", err)\n// \t}\n\n// \toptions := &context.CompletionOptions{\n// \t\tCapabilities: &openai.Capabilities{\n// \t\t\tStreaming: false,\n// \t\t\tReasoning: false,\n// \t\t\tToolCalls: true,\n// \t\t\tVision: \"claude\", // Claude requires base64 format\n// \t\t\tMultimodal: true,\n// \t\t},\n// \t}\n\n// \t// Define a simple tool with minimal parameters\n// \tsimpleTool := map[string]interface{}{\n// \t\t\"type\": \"function\",\n// \t\t\"function\": map[string]interface{}{\n// \t\t\t\"name\": \"get_info\",\n// \t\t\t\"description\": \"Get information\",\n// \t\t\t\"parameters\": map[string]interface{}{\n// \t\t\t\t\"type\": \"object\",\n// \t\t\t\t\"properties\": map[string]interface{}{\n// \t\t\t\t\t\"query\": map[string]interface{}{\n// \t\t\t\t\t\t\"type\": \"string\",\n// \t\t\t\t\t\t\"description\": \"Query string (single letter)\",\n// \t\t\t\t\t},\n// \t\t\t\t\t\"count\": map[string]interface{}{\n// \t\t\t\t\t\t\"type\": \"number\",\n// \t\t\t\t\t\t\"description\": \"Count (single digit)\",\n// \t\t\t\t\t},\n// \t\t\t\t},\n// \t\t\t\t\"required\": []string{\"query\", \"count\"},\n// \t\t\t},\n// \t\t},\n// \t}\n\n// \toptions.Tools = []map[string]interface{}{simpleTool}\n// \toptions.ToolChoice = \"auto\"\n\n// \t// Set enough tokens for tool call response\n// \tmaxTokens := 150\n// \toptions.MaxTokens = &maxTokens\n\n// \tllmInstance, err := llm.New(conn, options)\n// \tif err != nil {\n// \t\tt.Fatalf(\"Failed to create LLM instance: %v\", err)\n// \t}\n\n// \tmessages := []context.Message{\n// \t\t{\n// \t\t\tRole: context.RoleUser,\n// \t\t\tContent: \"Please use the get_info function to retrieve information. Pass 'A' as the query parameter and 1 as the count parameter.\",\n// \t\t},\n// \t}\n\n// \tctx := newClaudeTestContext(\"test-claude-sonnet4-tools\", \"claude.sonnet-4_0\")\n\n// \tresponse, err := llmInstance.Post(ctx, messages, options)\n// \tif err != nil {\n// \t\tt.Fatalf(\"Post failed: %v\", err)\n// \t}\n\n// \tif response == nil {\n// \t\tt.Fatal(\"Response is nil\")\n// \t}\n\n// \t// Validate tool calls\n// \tif len(response.ToolCalls) == 0 {\n// \t\tt.Error(\"No tool calls in response\")\n// \t} else {\n// \t\ttc := response.ToolCalls[0]\n// \t\tt.Logf(\"✓ Tool call: %s(%s)\", tc.Function.Name, tc.Function.Arguments)\n\n// \t\tif tc.Function.Name != \"get_info\" {\n// \t\t\tt.Errorf(\"Expected tool name 'get_info', got '%s'\", tc.Function.Name)\n// \t\t}\n// \t}\n\n// \tt.Logf(\"Usage: prompt=%d, completion=%d, total=%d\",\n// \t\tresponse.Usage.PromptTokens, response.Usage.CompletionTokens, response.Usage.TotalTokens)\n\n// \tt.Logf(\"Response: %+v\", response)\n// }\n\n// // TestClaudeSonnet4Vision tests vision capability with image input\n// func TestClaudeSonnet4Vision(t *testing.T) {\n// \ttest.Prepare(t, config.Conf)\n// \tdefer test.Clean()\n\n// \tconn, err := connector.Select(\"claude.sonnet-4_0\")\n// \tif err != nil {\n// \t\tt.Fatalf(\"Failed to select connector: %v\", err)\n// \t}\n\n// \toptions := &context.CompletionOptions{\n// \t\tCapabilities: &openai.Capabilities{\n// \t\t\tStreaming: false,\n// \t\t\tReasoning: false,\n// \t\t\tToolCalls: true,\n// \t\t\tVision: \"claude\", // Claude requires base64 format\n// \t\t\tMultimodal: true,\n// \t\t},\n// \t}\n\n// \tllmInstance, err := llm.New(conn, options)\n// \tif err != nil {\n// \t\tt.Fatalf(\"Failed to create LLM instance: %v\", err)\n// \t}\n\n// \t// Use a test image URL\n// \timageURL := \"https://upload.wikimedia.org/wikipedia/commons/thumb/d/dd/Gfp-wisconsin-madison-the-nature-boardwalk.jpg/2560px-Gfp-wisconsin-madison-the-nature-boardwalk.jpg\"\n\n// \tmessages := []context.Message{\n// \t\t{\n// \t\t\tRole: context.RoleUser,\n// \t\t\tContent: []map[string]interface{}{\n// \t\t\t\t{\n// \t\t\t\t\t\"type\": \"text\",\n// \t\t\t\t\t\"text\": \"Describe this image in one sentence.\",\n// \t\t\t\t},\n// \t\t\t\t{\n// \t\t\t\t\t\"type\": \"image_url\",\n// \t\t\t\t\t\"image_url\": map[string]string{\n// \t\t\t\t\t\t\"url\": imageURL,\n// \t\t\t\t\t},\n// \t\t\t\t},\n// \t\t\t},\n// \t\t},\n// \t}\n\n// \tmaxTokens := 150\n// \toptions.MaxTokens = &maxTokens\n\n// \tctx := newClaudeTestContext(\"test-claude-sonnet4-vision\", \"claude.sonnet-4_0\")\n\n// \tresponse, err := llmInstance.Post(ctx, messages, options)\n// \tif err != nil {\n// \t\tt.Fatalf(\"Post failed: %v\", err)\n// \t}\n\n// \tif response == nil {\n// \t\tt.Fatal(\"Response is nil\")\n// \t}\n\n// \t// Validate content\n// \tcontentStr, ok := response.Content.(string)\n// \tif !ok {\n// \t\tt.Fatalf(\"Content is not a string: %T\", response.Content)\n// \t}\n\n// \tif len(contentStr) == 0 {\n// \t\tt.Error(\"Image description is empty\")\n// \t}\n\n// \tt.Logf(\"Image description: %s\", contentStr)\n// \tt.Logf(\"Usage: prompt=%d, completion=%d, total=%d\",\n// \t\tresponse.Usage.PromptTokens, response.Usage.CompletionTokens, response.Usage.TotalTokens)\n// }\n\n// // TestClaudeSonnet4ThinkingStream tests Claude Sonnet 4 Thinking with streaming\n// func TestClaudeSonnet4ThinkingStream(t *testing.T) {\n// \ttest.Prepare(t, config.Conf)\n// \tdefer test.Clean()\n\n// \tconn, err := connector.Select(\"claude.sonnet-4_0-thinking\")\n// \tif err != nil {\n// \t\tt.Fatalf(\"Failed to select connector: %v\", err)\n// \t}\n\n// \toptions := &context.CompletionOptions{\n// \t\tCapabilities: &openai.Capabilities{\n// \t\t\tStreaming: true,\n// \t\t\tReasoning: true, // Claude Thinking mode exposes reasoning\n// \t\t\tToolCalls: false,\n// \t\t\tVision: \"claude\", // Claude requires base64 format\n// \t\t\tMultimodal: true,\n// \t\t},\n// \t}\n\n// \tllmInstance, err := llm.New(conn, options)\n// \tif err != nil {\n// \t\tt.Fatalf(\"Failed to create LLM instance: %v\", err)\n// \t}\n\n// \tmessages := []context.Message{\n// \t\t{\n// \t\t\tRole: context.RoleUser,\n// \t\t\tContent: \"If Sally has 3 apples and gives 2 to John, how many does she have left? Think through this step by step.\",\n// \t\t},\n// \t}\n\n// \tmaxTokens := 500\n// \toptions.MaxTokens = &maxTokens\n\n// \tctx := newClaudeTestContext(\"test-claude-thinking-stream\", \"claude.sonnet-4_0-thinking\")\n\n// \tvar thinkingChunks []string\n// \tvar textChunks []string\n// \thandler := func(chunkType message.StreamChunkType, data []byte) int {\n// \t\tt.Logf(\"Stream chunk [%s]: %s\", chunkType, string(data))\n// \t\tif chunkType == message.ChunkThinking {\n// \t\t\tthinkingChunks = append(thinkingChunks, string(data))\n// \t\t} else if chunkType == message.ChunkText {\n// \t\t\ttextChunks = append(textChunks, string(data))\n// \t\t}\n// \t\treturn 0\n// \t}\n\n// \tresponse, err := llmInstance.Stream(ctx, messages, options, handler)\n// \tif err != nil {\n// \t\tt.Fatalf(\"Stream failed: %v\", err)\n// \t}\n\n// \tif response == nil {\n// \t\tt.Fatal(\"Response is nil\")\n// \t}\n\n// \t// Validate response\n// \tcontentStr, ok := response.Content.(string)\n// \tif !ok {\n// \t\tt.Errorf(\"Content is not a string: %T\", response.Content)\n// \t}\n\n// \tt.Logf(\"Reasoning/Thinking content length: %d characters\", len(response.ReasoningContent))\n// \tt.Logf(\"Response content: %v\", contentStr)\n// \tt.Logf(\"Usage: prompt=%d, completion=%d, total=%d\",\n// \t\tresponse.Usage.PromptTokens, response.Usage.CompletionTokens, response.Usage.TotalTokens)\n\n// \tif response.Usage != nil && response.Usage.CompletionTokensDetails != nil {\n// \t\tt.Logf(\"Reasoning tokens: %d\", response.Usage.CompletionTokensDetails.ReasoningTokens)\n// \t}\n\n// \tt.Logf(\"Received %d thinking chunks\", len(thinkingChunks))\n// \tt.Logf(\"Received %d text chunks\", len(textChunks))\n// \tt.Logf(\"Final response: %+v\", response)\n// }\n\n// // TestClaudeSonnet4ThinkingPost tests Claude Sonnet 4 Thinking in non-streaming mode\n// func TestClaudeSonnet4ThinkingPost(t *testing.T) {\n// \ttest.Prepare(t, config.Conf)\n// \tdefer test.Clean()\n\n// \tconn, err := connector.Select(\"claude.sonnet-4_0-thinking\")\n// \tif err != nil {\n// \t\tt.Fatalf(\"Failed to select connector: %v\", err)\n// \t}\n\n// \toptions := &context.CompletionOptions{\n// \t\tCapabilities: &openai.Capabilities{\n// \t\t\tStreaming: false,\n// \t\t\tReasoning: true,\n// \t\t\tToolCalls: false,\n// \t\t\tVision: \"claude\", // Claude requires base64 format\n// \t\t\tMultimodal: true,\n// \t\t},\n// \t}\n\n// \tllmInstance, err := llm.New(conn, options)\n// \tif err != nil {\n// \t\tt.Fatalf(\"Failed to create LLM instance: %v\", err)\n// \t}\n\n// \tmessages := []context.Message{\n// \t\t{\n// \t\t\tRole: context.RoleUser,\n// \t\t\tContent: \"Is 7 greater than 5? Explain your reasoning.\",\n// \t\t},\n// \t}\n\n// \tmaxTokens := 500\n// \toptions.MaxTokens = &maxTokens\n\n// \tctx := newClaudeTestContext(\"test-claude-thinking-post\", \"claude.sonnet-4_0-thinking\")\n\n// \tresponse, err := llmInstance.Post(ctx, messages, options)\n// \tif err != nil {\n// \t\tt.Fatalf(\"Post failed: %v\", err)\n// \t}\n\n// \tif response == nil {\n// \t\tt.Fatal(\"Response is nil\")\n// \t}\n\n// \t// Validate content\n// \tcontentStr, ok := response.Content.(string)\n// \tif !ok {\n// \t\tt.Fatalf(\"Content is not a string: %T\", response.Content)\n// \t}\n\n// \tt.Logf(\"Reasoning content: %s\", response.ReasoningContent)\n// \tt.Logf(\"Final answer: %s\", contentStr)\n\n// \t// Check for reasoning content\n// \tif len(response.ReasoningContent) > 0 {\n// \t\tt.Logf(\"✓ Reasoning content present: %d characters\", len(response.ReasoningContent))\n// \t}\n\n// \tif response.Usage != nil && response.Usage.CompletionTokensDetails != nil && response.Usage.CompletionTokensDetails.ReasoningTokens > 0 {\n// \t\tt.Logf(\"✓ Reasoning tokens: %d\", response.Usage.CompletionTokensDetails.ReasoningTokens)\n// \t}\n\n// \tt.Logf(\"Usage: prompt=%d, completion=%d, total=%d\",\n// \t\tresponse.Usage.PromptTokens, response.Usage.CompletionTokens, response.Usage.TotalTokens)\n\n// \tt.Logf(\"Response: %+v\", response)\n// }\n\n// // TestClaudeTemperatureHandling tests that Claude models handle temperature parameter correctly\n// func TestClaudeTemperatureHandling(t *testing.T) {\n// \ttest.Prepare(t, config.Conf)\n// \tdefer test.Clean()\n\n// \ttests := []struct {\n// \t\tname string\n// \t\tconnector string\n// \t\ttemperature float64\n// \t\treasoning bool\n// \t}{\n// \t\t{\n// \t\t\tname: \"Sonnet 4 with temperature 0.7\",\n// \t\t\tconnector: \"claude.sonnet-4_0\",\n// \t\t\ttemperature: 0.7,\n// \t\t\treasoning: false,\n// \t\t},\n// \t\t{\n// \t\t\tname: \"Sonnet 4 Thinking with temperature 0.5\",\n// \t\t\tconnector: \"claude.sonnet-4_0-thinking\",\n// \t\t\ttemperature: 0.5,\n// \t\t\treasoning: true,\n// \t\t},\n// \t}\n\n// \tfor _, tt := range tests {\n// \t\tt.Run(tt.name, func(t *testing.T) {\n// \t\t\tconn, err := connector.Select(tt.connector)\n// \t\t\tif err != nil {\n// \t\t\t\tt.Fatalf(\"Failed to select connector: %v\", err)\n// \t\t\t}\n\n// \t\t\toptions := &context.CompletionOptions{\n// \t\t\t\tCapabilities: &openai.Capabilities{\n// \t\t\t\t\tStreaming: false,\n// \t\t\t\t\tReasoning: tt.reasoning,\n// \t\t\t\t\tToolCalls: true,\n// \t\t\t\t\tVision: true,\n// \t\t\t\t},\n// \t\t\t}\n\n// \t\t\tllmInstance, err := llm.New(conn, options)\n// \t\t\tif err != nil {\n// \t\t\t\tt.Fatalf(\"Failed to create LLM instance: %v\", err)\n// \t\t\t}\n\n// \t\t\tmessages := []context.Message{\n// \t\t\t\t{\n// \t\t\t\t\tRole: context.RoleUser,\n// \t\t\t\t\tContent: \"Say 'hello'.\",\n// \t\t\t\t},\n// \t\t\t}\n\n// \t\t\tmaxTokens := 50\n// \t\t\toptions.MaxTokens = &maxTokens\n// \t\t\toptions.Temperature = &tt.temperature\n\n// \t\t\tctx := newClaudeTestContext(\"test-claude-temp-\"+tt.connector, tt.connector)\n\n// \t\t\tresponse, err := llmInstance.Post(ctx, messages, options)\n// \t\t\tif err != nil {\n// \t\t\t\tt.Fatalf(\"Post failed: %v\", err)\n// \t\t\t}\n\n// \t\t\tif response == nil {\n// \t\t\t\tt.Fatal(\"Response is nil\")\n// \t\t\t}\n\n// \t\t\tt.Logf(\"✓ %s completed successfully with temperature=%.1f\", tt.name, tt.temperature)\n// \t\t})\n// \t}\n// }\n" }, { "path": "agent/llm/providers/openai/deepseek_r1_test.go", "content": "package openai_test\n\nimport (\n\tgocontext \"context\"\n\t\"strings\"\n\t\"testing\"\n\n\tjsoniter \"github.com/json-iterator/go\"\n\t\"github.com/yaoapp/gou/connector\"\n\t\"github.com/yaoapp/gou/connector/openai\"\n\t\"github.com/yaoapp/yao/agent/context\"\n\t\"github.com/yaoapp/yao/agent/llm\"\n\t\"github.com/yaoapp/yao/agent/output/message\"\n\t\"github.com/yaoapp/yao/config\"\n\t\"github.com/yaoapp/yao/openapi/oauth/types\"\n\t\"github.com/yaoapp/yao/test\"\n)\n\n// TestDeepSeekR1StreamBasic tests basic streaming completion with DeepSeek R1\nfunc TestDeepSeekR1StreamBasic(t *testing.T) {\n\ttest.Prepare(t, config.Conf)\n\tdefer test.Clean()\n\n\t// Create connector from real configuration\n\tconn, err := connector.Select(\"deepseek.r1\")\n\tif err != nil {\n\t\tt.Fatalf(\"Failed to select connector: %v\", err)\n\t}\n\n\t// Create LLM instance with capabilities\n\toptions := &context.CompletionOptions{\n\t\tCapabilities: &openai.Capabilities{\n\t\t\tStreaming: true,\n\t\t\tReasoning: true, // DeepSeek R1 supports reasoning\n\t\t\tToolCalls: false, // R1 doesn't support native tool calls\n\t\t\tVision: false,\n\t\t\tAudio: false,\n\t\t\tMultimodal: false,\n\t\t},\n\t}\n\n\tllmInstance, err := llm.New(conn, options)\n\tif err != nil {\n\t\tt.Fatalf(\"Failed to create LLM instance: %v\", err)\n\t}\n\n\t// Prepare messages with reasoning prompt (simple question for faster reasoning)\n\tmessages := []context.Message{\n\t\t{\n\t\t\tRole: context.RoleUser,\n\t\t\tContent: \"What is 2 + 2?\",\n\t\t},\n\t}\n\n\t// Set max tokens (higher for reasoning models to allow full reasoning + answer)\n\tmaxTokens := 500\n\toptions.MaxTokens = &maxTokens\n\n\t// Create context\n\tctx := newDeepSeekTestContext(\"test-deepseek-r1-basic\", \"deepseek.r1\")\n\n\t// Track streaming chunks and group events\n\tvar reasoningChunks []string\n\tvar contentChunks []string\n\tvar thinkingGroupEnded bool\n\tvar textGroupEnded bool\n\n\thandler := func(chunkType message.StreamChunkType, data []byte) int {\n\t\tdataStr := string(data)\n\t\tt.Logf(\"Stream chunk [%s]: %s\", chunkType, dataStr)\n\n\t\t// Track different chunk types\n\t\tswitch chunkType {\n\t\tcase message.ChunkThinking:\n\t\t\treasoningChunks = append(reasoningChunks, dataStr)\n\t\tcase message.ChunkText:\n\t\t\tcontentChunks = append(contentChunks, dataStr)\n\t\t}\n\n\t\t// Track group_end events to verify type field\n\t\tif chunkType == message.ChunkMessageEnd {\n\t\t\t// Parse the group_end data to check the type field\n\t\t\tvar groupEndData struct {\n\t\t\t\tGroupID string `json:\"group_id\"`\n\t\t\t\tType string `json:\"type\"`\n\t\t\t\tTimestamp int64 `json:\"timestamp\"`\n\t\t\t\tDurationMs int64 `json:\"duration_ms\"`\n\t\t\t\tChunkCount int `json:\"chunk_count\"`\n\t\t\t\tStatus string `json:\"status\"`\n\t\t\t}\n\n\t\t\tif err := jsoniter.Unmarshal(data, &groupEndData); err == nil {\n\t\t\t\tt.Logf(\"✓ group_end received: type=%s, chunks=%d, duration=%dms\",\n\t\t\t\t\tgroupEndData.Type, groupEndData.ChunkCount, groupEndData.DurationMs)\n\n\t\t\t\t// Verify the type field matches expected group types\n\t\t\t\tswitch groupEndData.Type {\n\t\t\t\tcase \"thinking\":\n\t\t\t\t\tthinkingGroupEnded = true\n\t\t\t\t\tif groupEndData.ChunkCount == 0 {\n\t\t\t\t\t\tt.Error(\"thinking group_end should have chunk_count > 0\")\n\t\t\t\t\t}\n\t\t\t\tcase \"text\":\n\t\t\t\t\ttextGroupEnded = true\n\t\t\t\t\tif groupEndData.ChunkCount == 0 {\n\t\t\t\t\t\tt.Error(\"text group_end should have chunk_count > 0\")\n\t\t\t\t\t}\n\t\t\t\t}\n\t\t\t} else {\n\t\t\t\tt.Errorf(\"Failed to parse group_end data: %v\", err)\n\t\t\t}\n\t\t}\n\n\t\treturn 0 // Continue\n\t}\n\n\t// Call Stream\n\tresponse, err := llmInstance.Stream(ctx, messages, options, handler)\n\tif err != nil {\n\t\tt.Fatalf(\"Stream failed: %v\", err)\n\t}\n\n\t// Validate response\n\tif response == nil {\n\t\tt.Fatal(\"Response is nil\")\n\t}\n\n\tif response.ID == \"\" {\n\t\tt.Error(\"Response ID is empty\")\n\t}\n\tif response.Model == \"\" {\n\t\tt.Error(\"Response Model is empty\")\n\t}\n\tif response.Content == \"\" {\n\t\tt.Error(\"Response content is empty\")\n\t}\n\tif response.FinishReason == \"\" {\n\t\tt.Error(\"FinishReason is empty\")\n\t}\n\n\t// DeepSeek R1 should have reasoning content\n\tif response.ReasoningContent == \"\" {\n\t\tt.Error(\"Expected reasoning_content but got empty\")\n\t} else {\n\t\tt.Logf(\"Reasoning content length: %d characters\", len(response.ReasoningContent))\n\t}\n\n\t// Check reasoning tokens in usage\n\tif response.Usage == nil {\n\t\tt.Error(\"Response Usage is nil\")\n\t} else {\n\t\tif response.Usage.TotalTokens == 0 {\n\t\t\tt.Error(\"Response Usage.TotalTokens is 0\")\n\t\t}\n\t\tif response.Usage.CompletionTokensDetails != nil {\n\t\t\tif response.Usage.CompletionTokensDetails.ReasoningTokens == 0 {\n\t\t\t\tt.Error(\"Expected reasoning_tokens > 0 for DeepSeek R1\")\n\t\t\t} else {\n\t\t\t\tt.Logf(\"Reasoning tokens: %d\", response.Usage.CompletionTokensDetails.ReasoningTokens)\n\t\t\t}\n\t\t}\n\t\tt.Logf(\"Usage: prompt=%d, completion=%d, total=%d\",\n\t\t\tresponse.Usage.PromptTokens, response.Usage.CompletionTokens, response.Usage.TotalTokens)\n\t}\n\n\t// Should have received reasoning chunks\n\tif len(reasoningChunks) == 0 {\n\t\tt.Error(\"Expected reasoning chunks (ChunkThinking) but got none\")\n\t} else {\n\t\tt.Logf(\"Received %d reasoning chunks\", len(reasoningChunks))\n\t}\n\n\t// Should have received content chunks\n\tif len(contentChunks) == 0 {\n\t\tt.Error(\"Expected content chunks (ChunkText) but got none\")\n\t} else {\n\t\tt.Logf(\"Received %d content chunks\", len(contentChunks))\n\t}\n\n\t// Verify group_end events were received with correct types\n\tif !thinkingGroupEnded {\n\t\tt.Error(\"❌ Expected thinking group_end event but didn't receive it\")\n\t} else {\n\t\tt.Log(\"✅ Thinking group_end event received with type='thinking'\")\n\t}\n\n\tif !textGroupEnded {\n\t\tt.Error(\"❌ Expected text group_end event but didn't receive it\")\n\t} else {\n\t\tt.Log(\"✅ Text group_end event received with type='text'\")\n\t}\n\n\tt.Logf(\"Final response: %+v\", response)\n}\n\n// TestDeepSeekR1PostBasic tests basic non-streaming completion\nfunc TestDeepSeekR1PostBasic(t *testing.T) {\n\ttest.Prepare(t, config.Conf)\n\tdefer test.Clean()\n\n\t// Create connector\n\tconn, err := connector.Select(\"deepseek.r1\")\n\tif err != nil {\n\t\tt.Fatalf(\"Failed to select connector: %v\", err)\n\t}\n\n\t// Create LLM instance\n\toptions := &context.CompletionOptions{\n\t\tCapabilities: &openai.Capabilities{\n\t\t\tReasoning: true,\n\t\t\tToolCalls: false,\n\t\t\tVision: false,\n\t\t\tAudio: false,\n\t\t\tMultimodal: false,\n\t\t},\n\t}\n\n\tllmInstance, err := llm.New(conn, options)\n\tif err != nil {\n\t\tt.Fatalf(\"Failed to create LLM instance: %v\", err)\n\t}\n\n\t// Prepare messages (very simple question)\n\tmessages := []context.Message{\n\t\t{\n\t\t\tRole: context.RoleUser,\n\t\t\tContent: \"What is 1+1?\",\n\t\t},\n\t}\n\n\t// Set max tokens (enough for reasoning + answer)\n\tmaxTokens := 500\n\toptions.MaxTokens = &maxTokens\n\n\t// Create context\n\tctx := newDeepSeekTestContext(\"test-deepseek-r1-post\", \"deepseek.r1\")\n\n\t// Call Post\n\tresponse, err := llmInstance.Post(ctx, messages, options)\n\tif err != nil {\n\t\tt.Fatalf(\"Post failed: %v\", err)\n\t}\n\n\t// Validate response\n\tif response == nil {\n\t\tt.Fatal(\"Response is nil\")\n\t}\n\n\tif response.ID == \"\" {\n\t\tt.Error(\"Response ID is empty\")\n\t}\n\tif response.Model == \"\" {\n\t\tt.Error(\"Response Model is empty\")\n\t}\n\tif response.Content == \"\" {\n\t\tt.Error(\"Response content is empty\")\n\t}\n\n\t// DeepSeek R1 should have reasoning content\n\tif response.ReasoningContent == \"\" {\n\t\tt.Error(\"Expected reasoning_content but got empty\")\n\t} else {\n\t\tt.Logf(\"Reasoning content: %s\", response.ReasoningContent)\n\t\tt.Logf(\"Final answer: %s\", response.Content)\n\t}\n\n\t// Check usage\n\tif response.Usage == nil {\n\t\tt.Error(\"Response Usage is nil\")\n\t} else {\n\t\tif response.Usage.TotalTokens == 0 {\n\t\t\tt.Error(\"Response Usage.TotalTokens is 0\")\n\t\t}\n\t\tif response.Usage.CompletionTokensDetails != nil && response.Usage.CompletionTokensDetails.ReasoningTokens > 0 {\n\t\t\tt.Logf(\"Reasoning tokens: %d\", response.Usage.CompletionTokensDetails.ReasoningTokens)\n\t\t}\n\t\tt.Logf(\"Usage: prompt=%d, completion=%d, total=%d\",\n\t\t\tresponse.Usage.PromptTokens, response.Usage.CompletionTokens, response.Usage.TotalTokens)\n\t}\n\n\tt.Logf(\"Response: %+v\", response)\n}\n\n// TestDeepSeekR1LogicPuzzle tests DeepSeek R1's reasoning with a logic puzzle\nfunc TestDeepSeekR1LogicPuzzle(t *testing.T) {\n\ttest.Prepare(t, config.Conf)\n\tdefer test.Clean()\n\n\tconn, err := connector.Select(\"deepseek.r1\")\n\tif err != nil {\n\t\tt.Fatalf(\"Failed to select connector: %v\", err)\n\t}\n\n\toptions := &context.CompletionOptions{\n\t\tCapabilities: &openai.Capabilities{\n\t\t\tStreaming: true,\n\t\t\tReasoning: true,\n\t\t\tToolCalls: false,\n\t\t\tVision: false,\n\t\t\tAudio: false,\n\t\t\tMultimodal: false,\n\t\t},\n\t}\n\n\tmaxTokens := 800\n\toptions.MaxTokens = &maxTokens\n\n\tllmInstance, err := llm.New(conn, options)\n\tif err != nil {\n\t\tt.Fatalf(\"Failed to create LLM instance: %v\", err)\n\t}\n\n\t// Use a simpler logic question\n\tmessages := []context.Message{\n\t\t{\n\t\t\tRole: context.RoleUser,\n\t\t\tContent: \"Is 5 greater than 3? Explain your reasoning.\",\n\t\t},\n\t}\n\n\tctx := newDeepSeekTestContext(\"test-deepseek-r1-logic\", \"deepseek.r1\")\n\n\t// Track reasoning and content separately\n\tvar hasReasoning, hasContent bool\n\thandler := func(chunkType message.StreamChunkType, data []byte) int {\n\t\tif chunkType == message.ChunkThinking && len(data) > 0 {\n\t\t\thasReasoning = true\n\t\t} else if chunkType == message.ChunkText && len(data) > 0 {\n\t\t\thasContent = true\n\t\t}\n\t\treturn 0\n\t}\n\n\tresponse, err := llmInstance.Stream(ctx, messages, options, handler)\n\tif err != nil {\n\t\tt.Fatalf(\"Stream failed: %v\", err)\n\t}\n\n\tif response == nil {\n\t\tt.Fatal(\"Response is nil\")\n\t}\n\n\t// Should have both reasoning and content\n\tif !hasReasoning {\n\t\tt.Error(\"Expected to receive reasoning chunks but didn't\")\n\t}\n\tif !hasContent {\n\t\tt.Error(\"Expected to receive content chunks but didn't\")\n\t}\n\n\t// Validate reasoning content exists and is substantial\n\tif response.ReasoningContent == \"\" {\n\t\tt.Error(\"Expected reasoning_content but got empty\")\n\t} else if len(response.ReasoningContent) < 50 {\n\t\tt.Errorf(\"Reasoning content too short (%d chars), expected detailed thinking\", len(response.ReasoningContent))\n\t} else {\n\t\tt.Logf(\"✓ Reasoning content length: %d characters\", len(response.ReasoningContent))\n\t}\n\n\t// Validate final answer\n\tcontentStr := \"\"\n\tif response.Content != nil {\n\t\tif str, ok := response.Content.(string); ok {\n\t\t\tcontentStr = str\n\t\t}\n\t}\n\n\tif len(contentStr) == 0 {\n\t\tt.Error(\"Content is empty\")\n\t} else {\n\t\t// Should mention \"Yes\" or affirm that 5 > 3\n\t\tif !strings.Contains(strings.ToLower(contentStr), \"yes\") && !strings.Contains(strings.ToLower(contentStr), \"greater\") {\n\t\t\tt.Logf(\"Warning: Content might not contain expected answer. Content: %s\", contentStr)\n\t\t} else {\n\t\t\tt.Logf(\"✓ Final answer: %s\", contentStr)\n\t\t}\n\t}\n\n\t// Check reasoning tokens\n\tif response.Usage != nil && response.Usage.CompletionTokensDetails != nil {\n\t\treasoningTokens := response.Usage.CompletionTokensDetails.ReasoningTokens\n\t\tif reasoningTokens == 0 {\n\t\t\tt.Error(\"Expected reasoning_tokens > 0\")\n\t\t} else {\n\t\t\tt.Logf(\"✓ Reasoning tokens: %d\", reasoningTokens)\n\t\t}\n\t}\n\n\tt.Log(\"Logic puzzle test completed successfully\")\n}\n\n// ============================================================================\n// Helper Functions\n// ============================================================================\n\n// newDeepSeekTestContext creates a real Context for testing DeepSeek provider\nfunc newDeepSeekTestContext(chatID, connectorID string) *context.Context {\n\tauthorized := &types.AuthorizedInfo{\n\t\tSubject: \"test-user\",\n\t\tClientID: \"test-client\",\n\t\tUserID: \"test-user-123\",\n\t\tTeamID: \"test-team-456\",\n\t\tTenantID: \"test-tenant-789\",\n\t\tSessionID: \"test-session-id\",\n\t\tConstraints: types.DataConstraints{\n\t\t\tTeamOnly: true,\n\t\t\tExtra: map[string]interface{}{\n\t\t\t\t\"test\": \"deepseek-provider\",\n\t\t\t},\n\t\t},\n\t}\n\n\tctx := context.New(gocontext.Background(), authorized, chatID)\n\tctx.AssistantID = \"test-assistant\"\n\tctx.Locale = \"en-us\"\n\tctx.Theme = \"light\"\n\tctx.Client = context.Client{\n\t\tType: \"web\",\n\t\tUserAgent: \"DeepSeekProviderTest/1.0\",\n\t\tIP: \"127.0.0.1\",\n\t}\n\tctx.Referer = context.RefererAPI\n\tctx.Accept = context.AcceptStandard\n\tctx.Route = \"/api/test\"\n\tctx.Metadata = make(map[string]interface{})\n\treturn ctx\n}\n" }, { "path": "agent/llm/providers/openai/deepseek_v3_test.go", "content": "package openai_test\n\nimport (\n\tgocontext \"context\"\n\t\"testing\"\n\n\t\"github.com/yaoapp/gou/connector\"\n\t\"github.com/yaoapp/gou/connector/openai\"\n\t\"github.com/yaoapp/yao/agent/context\"\n\t\"github.com/yaoapp/yao/agent/llm\"\n\t\"github.com/yaoapp/yao/agent/output/message\"\n\t\"github.com/yaoapp/yao/config\"\n\t\"github.com/yaoapp/yao/openapi/oauth/types\"\n\t\"github.com/yaoapp/yao/test\"\n)\n\n// TestDeepSeekV3StreamBasic tests basic streaming completion with DeepSeek V3\nfunc TestDeepSeekV3StreamBasic(t *testing.T) {\n\ttest.Prepare(t, config.Conf)\n\tdefer test.Clean()\n\n\tconn, err := connector.Select(\"deepseek.v3\")\n\tif err != nil {\n\t\tt.Fatalf(\"Failed to select connector: %v\", err)\n\t}\n\n\toptions := &context.CompletionOptions{\n\t\tCapabilities: &openai.Capabilities{\n\t\t\tStreaming: true,\n\t\t\tReasoning: false, // V3 doesn't support reasoning\n\t\t\tToolCalls: true, // V3 supports tool calls\n\t\t\tVision: false,\n\t\t\tAudio: false,\n\t\t\tMultimodal: false,\n\t\t},\n\t}\n\n\tllmInstance, err := llm.New(conn, options)\n\tif err != nil {\n\t\tt.Fatalf(\"Failed to create LLM instance: %v\", err)\n\t}\n\n\t// Simple math question\n\tmessages := []context.Message{\n\t\t{\n\t\t\tRole: context.RoleUser,\n\t\t\tContent: \"What is 5 + 3?\",\n\t\t},\n\t}\n\n\t// Set max tokens\n\tmaxTokens := 100\n\toptions.MaxTokens = &maxTokens\n\n\tctx := newDeepSeekV3TestContext(\"test-deepseek-v3-basic\", \"deepseek.v3\")\n\n\t// Track streaming chunks\n\tvar contentChunks []string\n\thandler := func(chunkType message.StreamChunkType, data []byte) int {\n\t\tdataStr := string(data)\n\t\tt.Logf(\"Stream chunk [%s]: %s\", chunkType, dataStr)\n\n\t\tif chunkType == message.ChunkText {\n\t\t\tcontentChunks = append(contentChunks, dataStr)\n\t\t}\n\n\t\treturn 0 // Continue\n\t}\n\n\t// Call Stream\n\tresponse, err := llmInstance.Stream(ctx, messages, options, handler)\n\tif err != nil {\n\t\tt.Fatalf(\"Stream failed: %v\", err)\n\t}\n\n\t// Validate response\n\tif response == nil {\n\t\tt.Fatal(\"Response is nil\")\n\t}\n\n\tif response.ID == \"\" {\n\t\tt.Error(\"Response ID is empty\")\n\t}\n\tif response.Model == \"\" {\n\t\tt.Error(\"Response Model is empty\")\n\t}\n\n\t// Should have content (V3 is not a reasoning model)\n\tcontentStr, ok := response.Content.(string)\n\tif !ok || contentStr == \"\" {\n\t\tt.Error(\"Expected content but got empty\")\n\t} else {\n\t\tt.Logf(\"Response content: %s\", contentStr)\n\t}\n\n\t// Should NOT have reasoning content (V3 doesn't support reasoning)\n\tif response.ReasoningContent != \"\" {\n\t\tt.Errorf(\"Expected no reasoning_content for V3, but got: %s\", response.ReasoningContent)\n\t}\n\n\t// Check usage\n\tif response.Usage == nil {\n\t\tt.Error(\"Response Usage is nil\")\n\t} else {\n\t\tt.Logf(\"Usage: prompt=%d, completion=%d, total=%d\",\n\t\t\tresponse.Usage.PromptTokens, response.Usage.CompletionTokens, response.Usage.TotalTokens)\n\n\t\t// Should have 0 reasoning tokens\n\t\tif response.Usage.CompletionTokensDetails != nil {\n\t\t\treasoningTokens := response.Usage.CompletionTokensDetails.ReasoningTokens\n\t\t\tif reasoningTokens != 0 {\n\t\t\t\tt.Errorf(\"Expected reasoning_tokens=0 for V3, got %d\", reasoningTokens)\n\t\t\t}\n\t\t}\n\t}\n\n\tif len(contentChunks) == 0 {\n\t\tt.Error(\"Expected content chunks but got none\")\n\t} else {\n\t\tt.Logf(\"Received %d content chunks\", len(contentChunks))\n\t}\n\n\tt.Logf(\"Final response: %+v\", response)\n}\n\n// TestDeepSeekV3PostBasic tests basic non-streaming completion\nfunc TestDeepSeekV3PostBasic(t *testing.T) {\n\ttest.Prepare(t, config.Conf)\n\tdefer test.Clean()\n\n\tconn, err := connector.Select(\"deepseek.v3\")\n\tif err != nil {\n\t\tt.Fatalf(\"Failed to select connector: %v\", err)\n\t}\n\n\toptions := &context.CompletionOptions{\n\t\tCapabilities: &openai.Capabilities{\n\t\t\tReasoning: false,\n\t\t\tToolCalls: true,\n\t\t\tVision: false,\n\t\t\tAudio: false,\n\t\t\tMultimodal: false,\n\t\t},\n\t}\n\n\tllmInstance, err := llm.New(conn, options)\n\tif err != nil {\n\t\tt.Fatalf(\"Failed to create LLM instance: %v\", err)\n\t}\n\n\t// Simple question\n\tmessages := []context.Message{\n\t\t{\n\t\t\tRole: context.RoleUser,\n\t\t\tContent: \"What is 2 * 4?\",\n\t\t},\n\t}\n\n\t// Set max tokens\n\tmaxTokens := 100\n\toptions.MaxTokens = &maxTokens\n\n\tctx := newDeepSeekV3TestContext(\"test-deepseek-v3-post\", \"deepseek.v3\")\n\n\t// Call Post\n\tresponse, err := llmInstance.Post(ctx, messages, options)\n\tif err != nil {\n\t\tt.Fatalf(\"Post failed: %v\", err)\n\t}\n\n\t// Validate response\n\tif response == nil {\n\t\tt.Fatal(\"Response is nil\")\n\t}\n\n\tif response.ID == \"\" {\n\t\tt.Error(\"Response ID is empty\")\n\t}\n\tif response.Model == \"\" {\n\t\tt.Error(\"Response Model is empty\")\n\t}\n\n\t// Should have content\n\tcontentStr, ok := response.Content.(string)\n\tif !ok || contentStr == \"\" {\n\t\tt.Error(\"Expected content but got empty\")\n\t} else {\n\t\tt.Logf(\"Response content: %s\", contentStr)\n\t}\n\n\t// Should NOT have reasoning content\n\tif response.ReasoningContent != \"\" {\n\t\tt.Errorf(\"V3 should not have reasoning_content, but got: %s\", response.ReasoningContent)\n\t}\n\n\t// Check usage\n\tif response.Usage == nil {\n\t\tt.Error(\"Response Usage is nil\")\n\t} else {\n\t\tt.Logf(\"Usage: prompt=%d, completion=%d, total=%d\",\n\t\t\tresponse.Usage.PromptTokens, response.Usage.CompletionTokens, response.Usage.TotalTokens)\n\n\t\t// Should have 0 reasoning tokens\n\t\tif response.Usage.CompletionTokensDetails != nil {\n\t\t\treasoningTokens := response.Usage.CompletionTokensDetails.ReasoningTokens\n\t\t\tif reasoningTokens != 0 {\n\t\t\t\tt.Errorf(\"Expected reasoning_tokens=0 for V3, got %d\", reasoningTokens)\n\t\t\t}\n\t\t}\n\t}\n\n\tt.Logf(\"Response: %+v\", response)\n}\n\n// TestDeepSeekV3WithToolCalls tests V3 with tool calls\nfunc TestDeepSeekV3WithToolCalls(t *testing.T) {\n\ttest.Prepare(t, config.Conf)\n\tdefer test.Clean()\n\n\tconn, err := connector.Select(\"deepseek.v3\")\n\tif err != nil {\n\t\tt.Fatalf(\"Failed to select connector: %v\", err)\n\t}\n\n\toptions := &context.CompletionOptions{\n\t\tCapabilities: &openai.Capabilities{\n\t\t\tReasoning: false,\n\t\t\tToolCalls: true,\n\t\t},\n\t}\n\n\t// Define a simple tool with minimal parameters\n\tsimpleTool := map[string]interface{}{\n\t\t\"type\": \"function\",\n\t\t\"function\": map[string]interface{}{\n\t\t\t\"name\": \"get_info\",\n\t\t\t\"description\": \"Get information\",\n\t\t\t\"parameters\": map[string]interface{}{\n\t\t\t\t\"type\": \"object\",\n\t\t\t\t\"properties\": map[string]interface{}{\n\t\t\t\t\t\"query\": map[string]interface{}{\n\t\t\t\t\t\t\"type\": \"string\",\n\t\t\t\t\t\t\"description\": \"Query string (single letter)\",\n\t\t\t\t\t},\n\t\t\t\t\t\"count\": map[string]interface{}{\n\t\t\t\t\t\t\"type\": \"number\",\n\t\t\t\t\t\t\"description\": \"Count (single digit)\",\n\t\t\t\t\t},\n\t\t\t\t},\n\t\t\t\t\"required\": []string{\"query\", \"count\"},\n\t\t\t},\n\t\t},\n\t}\n\n\toptions.Tools = []map[string]interface{}{simpleTool}\n\toptions.ToolChoice = \"auto\"\n\n\t// Set lower max_tokens for faster response\n\tmaxTokens := 50\n\toptions.MaxTokens = &maxTokens\n\n\tllmInstance, err := llm.New(conn, options)\n\tif err != nil {\n\t\tt.Fatalf(\"Failed to create LLM instance: %v\", err)\n\t}\n\n\tmessages := []context.Message{\n\t\t{\n\t\t\tRole: context.RoleUser,\n\t\t\tContent: \"Call get_info with query='A' and count=1\",\n\t\t},\n\t}\n\n\tctx := newDeepSeekV3TestContext(\"test-deepseek-v3-tools\", \"deepseek.v3\")\n\n\tresponse, err := llmInstance.Post(ctx, messages, options)\n\tif err != nil {\n\t\tt.Fatalf(\"Post with tool calls failed: %v\", err)\n\t}\n\n\tif response == nil {\n\t\tt.Fatal(\"Response is nil\")\n\t}\n\n\t// Should have tool calls\n\tif len(response.ToolCalls) == 0 {\n\t\tt.Error(\"Expected tool calls but got none\")\n\t} else {\n\t\ttc := response.ToolCalls[0]\n\t\tt.Logf(\"✓ Tool call: %s(%s)\", tc.Function.Name, tc.Function.Arguments)\n\n\t\tif tc.Function.Name != \"get_info\" {\n\t\t\tt.Errorf(\"Expected tool name 'get_info', got '%s'\", tc.Function.Name)\n\t\t}\n\t}\n\n\tif response.Usage != nil {\n\t\tt.Logf(\"Usage: prompt=%d, completion=%d, total=%d\",\n\t\t\tresponse.Usage.PromptTokens, response.Usage.CompletionTokens, response.Usage.TotalTokens)\n\t}\n\n\tt.Logf(\"Response: %+v\", response)\n}\n\n// TestDeepSeekV3NoReasoningEffort tests that V3 ignores reasoning_effort parameter\nfunc TestDeepSeekV3NoReasoningEffort(t *testing.T) {\n\ttest.Prepare(t, config.Conf)\n\tdefer test.Clean()\n\n\tconn, err := connector.Select(\"deepseek.v3\")\n\tif err != nil {\n\t\tt.Fatalf(\"Failed to select connector: %v\", err)\n\t}\n\n\teffort := \"high\"\n\toptions := &context.CompletionOptions{\n\t\tCapabilities: &openai.Capabilities{\n\t\t\tReasoning: false, // V3 doesn't support reasoning\n\t\t\tToolCalls: true,\n\t\t},\n\t\tReasoningEffort: &effort, // Should be ignored by adapter\n\t}\n\n\tllmInstance, err := llm.New(conn, options)\n\tif err != nil {\n\t\tt.Fatalf(\"Failed to create LLM instance: %v\", err)\n\t}\n\n\tmessages := []context.Message{\n\t\t{\n\t\t\tRole: context.RoleUser,\n\t\t\tContent: \"Reply with just: OK\",\n\t\t},\n\t}\n\n\tmaxTokens := 20\n\toptions.MaxTokens = &maxTokens\n\n\tctx := newDeepSeekV3TestContext(\"test-deepseek-v3-no-reasoning\", \"deepseek.v3\")\n\n\t// Should succeed (adapter removes reasoning_effort parameter)\n\tresponse, err := llmInstance.Post(ctx, messages, options)\n\tif err != nil {\n\t\tt.Fatalf(\"Post failed: %v\", err)\n\t}\n\n\tif response == nil {\n\t\tt.Fatal(\"Response is nil\")\n\t}\n\n\t// Should have 0 reasoning tokens\n\tif response.Usage != nil && response.Usage.CompletionTokensDetails != nil {\n\t\treasoningTokens := response.Usage.CompletionTokensDetails.ReasoningTokens\n\t\tif reasoningTokens != 0 {\n\t\t\tt.Errorf(\"Expected reasoning_tokens=0 for V3, got %d\", reasoningTokens)\n\t\t} else {\n\t\t\tt.Log(\"✓ V3 correctly shows reasoning_tokens=0\")\n\t\t}\n\t}\n\n\tt.Log(\"✓ ReasoningAdapter correctly removed reasoning_effort parameter for V3\")\n}\n\n// ============================================================================\n// Helper Functions\n// ============================================================================\n\n// newDeepSeekV3TestContext creates a real Context for testing DeepSeek V3 provider\nfunc newDeepSeekV3TestContext(chatID, connectorID string) *context.Context {\n\tauthorized := &types.AuthorizedInfo{\n\t\tSubject: \"test-user\",\n\t\tClientID: \"test-client\",\n\t\tUserID: \"test-user-123\",\n\t\tTeamID: \"test-team-456\",\n\t\tTenantID: \"test-tenant-789\",\n\t\tSessionID: \"test-session-id\",\n\t\tConstraints: types.DataConstraints{\n\t\t\tTeamOnly: true,\n\t\t\tExtra: map[string]interface{}{\n\t\t\t\t\"test\": \"deepseek-v3-provider\",\n\t\t\t},\n\t\t},\n\t}\n\n\tctx := context.New(gocontext.Background(), authorized, chatID)\n\tctx.AssistantID = \"test-assistant\"\n\tctx.Locale = \"en-us\"\n\tctx.Theme = \"light\"\n\tctx.Client = context.Client{\n\t\tType: \"web\",\n\t\tUserAgent: \"DeepSeekV3ProviderTest/1.0\",\n\t\tIP: \"127.0.0.1\",\n\t}\n\tctx.Referer = context.RefererAPI\n\tctx.Accept = context.AcceptStandard\n\tctx.Route = \"/api/test\"\n\tctx.Metadata = make(map[string]interface{})\n\treturn ctx\n}\n" }, { "path": "agent/llm/providers/openai/gpt5_test.go", "content": "package openai_test\n\n// GPT-5 tests temporarily commented out\n//\n// import (\n// \tgocontext \"context\"\n// \t\"testing\"\n//\n// \t\"github.com/yaoapp/gou/connector\"\n// \t\"github.com/yaoapp/gou/connector/openai\"\n// \t\"github.com/yaoapp/yao/agent/context\"\n// \t\"github.com/yaoapp/yao/agent/llm\"\n// \t\"github.com/yaoapp/yao/agent/output/message\"\n// \t\"github.com/yaoapp/yao/config\"\n// \t\"github.com/yaoapp/yao/openapi/oauth/types\"\n// \t\"github.com/yaoapp/yao/test\"\n// )\n//\n// // TestGPT5StreamBasic tests basic streaming completion with GPT-5\n// func TestGPT5StreamBasic(t *testing.T) {\n// \ttest.Prepare(t, config.Conf)\n// \tdefer test.Clean()\n//\n// \tconn, err := connector.Select(\"openai.gpt-5\")\n// \tif err != nil {\n// \t\tt.Fatalf(\"Failed to select connector: %v\", err)\n// \t}\n//\n// \toptions := &context.CompletionOptions{\n// \t\tCapabilities: &openai.Capabilities{\n// \t\t\tStreaming: true,\n// \t\t\tReasoning: true, // GPT-5 supports reasoning\n// \t\t\tToolCalls: true,\n// \t\t\tVision: true,\n// \t\t\tMultimodal: true,\n// \t\t},\n// \t}\n//\n// \tllmInstance, err := llm.New(conn, options)\n// \tif err != nil {\n// \t\tt.Fatalf(\"Failed to create LLM instance: %v\", err)\n// \t}\n//\n// \tmessages := []context.Message{\n// \t\t{\n// \t\t\tRole: context.RoleUser,\n// \t\t\tContent: \"What is 1+1? Reply with just the number.\",\n// \t\t},\n// \t}\n//\n// \tmaxTokens := 100\n// \toptions.MaxCompletionTokens = &maxTokens\n//\n// \tctx := newGPT5TestContext(\"test-gpt5-basic\", \"openai.gpt-5\")\n//\n// \tvar chunks []string\n// \thandler := func(chunkType message.StreamChunkType, data []byte) int {\n// \t\tchunks = append(chunks, string(data))\n// \t\tt.Logf(\"Stream chunk [%s]: %s\", chunkType, string(data))\n// \t\treturn 0\n// \t}\n//\n// \tresponse, err := llmInstance.Stream(ctx, messages, options, handler)\n// \tif err != nil {\n// \t\tt.Fatalf(\"Stream failed: %v\", err)\n// \t}\n//\n// \tif response == nil {\n// \t\tt.Fatal(\"Response is nil\")\n// \t}\n//\n// \t// Basic validation\n// \tif response.ID == \"\" {\n// \t\tt.Error(\"Response ID is empty\")\n// \t}\n// \tif response.Model == \"\" {\n// \t\tt.Error(\"Response Model is empty\")\n// \t}\n//\n// \t// GPT-5 may use all tokens for reasoning, so content could be empty\n// \t// Just log the content instead of failing\n// \tt.Logf(\"Response content: %v\", response.Content)\n// \tt.Logf(\"Usage: prompt=%d, completion=%d, total=%d\",\n// \t\tresponse.Usage.PromptTokens, response.Usage.CompletionTokens, response.Usage.TotalTokens)\n//\n// \tif response.Usage != nil && response.Usage.CompletionTokensDetails != nil {\n// \t\tt.Logf(\"Reasoning tokens: %d\", response.Usage.CompletionTokensDetails.ReasoningTokens)\n// \t}\n//\n// \tt.Logf(\"Final response: %+v\", response)\n// \tt.Logf(\"Total chunks received: %d\", len(chunks))\n// }\n//\n// // TestGPT5ReasoningEffort tests reasoning_effort parameter with different levels\n// func TestGPT5ReasoningEffort(t *testing.T) {\n// \ttest.Prepare(t, config.Conf)\n// \tdefer test.Clean()\n//\n// \tconn, err := connector.Select(\"openai.gpt-5\")\n// \tif err != nil {\n// \t\tt.Fatalf(\"Failed to select connector: %v\", err)\n// \t}\n//\n// \t// Test with different reasoning effort levels\n// \teffortLevels := []string{\"low\", \"medium\", \"high\"}\n//\n// \tfor _, effort := range effortLevels {\n// \t\tt.Run(\"effort_\"+effort, func(t *testing.T) {\n// \t\t\toptions := &context.CompletionOptions{\n// \t\t\t\tCapabilities: &openai.Capabilities{\n// \t\t\t\t\tReasoning: true,\n// \t\t\t\t\tToolCalls: true,\n// \t\t\t\t},\n// \t\t\t\tReasoningEffort: &effort,\n// \t\t\t}\n//\n// \t\t\tllmInstance, err := llm.New(conn, options)\n// \t\t\tif err != nil {\n// \t\t\t\tt.Fatalf(\"Failed to create LLM instance: %v\", err)\n// \t\t\t}\n//\n// \t\t\tmessages := []context.Message{\n// \t\t\t\t{\n// \t\t\t\t\tRole: context.RoleUser,\n// \t\t\t\t\tContent: \"Solve: If all Bloops are Razzies and all Razzies are Lazzies, are all Bloops Lazzies?\",\n// \t\t\t\t},\n// \t\t\t}\n//\n// \t\t\tmaxTokens := 1000\n// \t\t\toptions.MaxCompletionTokens = &maxTokens\n//\n// \t\t\tctx := newGPT5TestContext(\"test-gpt5-reasoning-\"+effort, \"openai.gpt-5\")\n//\n// \t\t\tresponse, err := llmInstance.Post(ctx, messages, options)\n// \t\t\tif err != nil {\n// \t\t\t\tt.Fatalf(\"Post failed with effort=%s: %v\", effort, err)\n// \t\t\t}\n//\n// \t\t\tif response == nil {\n// \t\t\t\tt.Fatal(\"Response is nil\")\n// \t\t\t}\n//\n// \t\t\t// Check reasoning tokens\n// \t\t\tvar reasoningTokens int\n// \t\t\tif response.Usage != nil && response.Usage.CompletionTokensDetails != nil {\n// \t\t\t\treasoningTokens = response.Usage.CompletionTokensDetails.ReasoningTokens\n// \t\t\t}\n//\n// \t\t\tt.Logf(\"Reasoning effort: %s\", effort)\n// \t\t\tt.Logf(\"Reasoning tokens: %d\", reasoningTokens)\n// \t\t\tt.Logf(\"Total tokens: %d\", response.Usage.TotalTokens)\n// \t\t\tt.Logf(\"Content: %s\", response.Content)\n//\n// \t\t\t// GPT-5 reasoning is hidden (no reasoning_content field)\n// \t\t\t// But should have reasoning_tokens in usage\n// \t\t\tif effort != \"low\" {\n// \t\t\t\tif reasoningTokens == 0 {\n// \t\t\t\t\tt.Logf(\"Warning: Expected reasoning_tokens > 0 for effort='%s', got 0\", effort)\n// \t\t\t\t}\n// \t\t\t}\n// \t\t})\n// \t}\n// }\n//\n// // TestGPT5PostWithToolCalls tests GPT-5 with tool calls\n// func TestGPT5PostWithToolCalls(t *testing.T) {\n// \ttest.Prepare(t, config.Conf)\n// \tdefer test.Clean()\n//\n// \tconn, err := connector.Select(\"openai.gpt-5\")\n// \tif err != nil {\n// \t\tt.Fatalf(\"Failed to select connector: %v\", err)\n// \t}\n//\n// \toptions := &context.CompletionOptions{\n// \t\tCapabilities: &openai.Capabilities{\n// \t\t\tReasoning: true,\n// \t\t\tToolCalls: true,\n// \t\t},\n// \t}\n//\n// \t// Define a calculation tool\n// \tcalcTool := map[string]interface{}{\n// \t\t\"type\": \"function\",\n// \t\t\"function\": map[string]interface{}{\n// \t\t\t\"name\": \"calculate\",\n// \t\t\t\"description\": \"Perform a mathematical calculation\",\n// \t\t\t\"parameters\": map[string]interface{}{\n// \t\t\t\t\"type\": \"object\",\n// \t\t\t\t\"properties\": map[string]interface{}{\n// \t\t\t\t\t\"expression\": map[string]interface{}{\n// \t\t\t\t\t\t\"type\": \"string\",\n// \t\t\t\t\t\t\"description\": \"The mathematical expression to evaluate\",\n// \t\t\t\t\t},\n// \t\t\t\t},\n// \t\t\t\t\"required\": []string{\"expression\"},\n// \t\t\t},\n// \t\t},\n// \t}\n//\n// \toptions.Tools = []map[string]interface{}{calcTool}\n// \toptions.ToolChoice = \"auto\"\n//\n// \tllmInstance, err := llm.New(conn, options)\n// \tif err != nil {\n// \t\tt.Fatalf(\"Failed to create LLM instance: %v\", err)\n// \t}\n//\n// \tmessages := []context.Message{\n// \t\t{\n// \t\t\tRole: context.RoleUser,\n// \t\t\tContent: \"Use the calculate function to compute 2 * 3\",\n// \t\t},\n// \t}\n//\n// \tctx := newGPT5TestContext(\"test-gpt5-tools\", \"openai.gpt-5\")\n//\n// \tresponse, err := llmInstance.Post(ctx, messages, options)\n// \tif err != nil {\n// \t\tt.Fatalf(\"Post with tool calls failed: %v\", err)\n// \t}\n//\n// \tif response == nil {\n// \t\tt.Fatal(\"Response is nil\")\n// \t}\n//\n// \t// GPT-5 reasoning models may not always use tool calls\n// \t// Log what we got instead of failing\n// \tif len(response.ToolCalls) == 0 {\n// \t\tt.Logf(\"No tool calls returned. Content: %v\", response.Content)\n// \t} else {\n// \t\ttc := response.ToolCalls[0]\n// \t\tt.Logf(\"✓ Tool call: %s(%s)\", tc.Function.Name, tc.Function.Arguments)\n//\n// \t\tif tc.Function.Name != \"calculate\" {\n// \t\t\tt.Logf(\"Warning: Expected tool name 'calculate', got '%s'\", tc.Function.Name)\n// \t\t}\n// \t}\n//\n// \tif response.Usage != nil {\n// \t\tt.Logf(\"Usage: prompt=%d, completion=%d, total=%d\",\n// \t\t\tresponse.Usage.PromptTokens, response.Usage.CompletionTokens, response.Usage.TotalTokens)\n// \t\tif response.Usage.CompletionTokensDetails != nil {\n// \t\t\tt.Logf(\"Reasoning tokens: %d\", response.Usage.CompletionTokensDetails.ReasoningTokens)\n// \t\t}\n// \t}\n//\n// \tt.Logf(\"Response: %+v\", response)\n// }\n//\n// // TestGPT5Vision tests GPT-5 with image input\n// func TestGPT5Vision(t *testing.T) {\n// \ttest.Prepare(t, config.Conf)\n// \tdefer test.Clean()\n//\n// \tconn, err := connector.Select(\"openai.gpt-5\")\n// \tif err != nil {\n// \t\tt.Fatalf(\"Failed to select connector: %v\", err)\n// \t}\n//\n// \toptions := &context.CompletionOptions{\n// \t\tCapabilities: &openai.Capabilities{\n// \t\t\tReasoning: true,\n// \t\t\tVision: true,\n// \t\t\tMultimodal: true,\n// \t\t},\n// \t}\n//\n// \tllmInstance, err := llm.New(conn, options)\n// \tif err != nil {\n// \t\tt.Fatalf(\"Failed to create LLM instance: %v\", err)\n// \t}\n//\n// \t// Message with image content\n// \tmessages := []context.Message{\n// \t\t{\n// \t\t\tRole: context.RoleUser,\n// \t\t\tContent: []context.ContentPart{\n// \t\t\t\t{\n// \t\t\t\t\tType: context.ContentText,\n// \t\t\t\t\tText: \"What is in this image? Describe briefly.\",\n// \t\t\t\t},\n// \t\t\t\t{\n// \t\t\t\t\tType: context.ContentImageURL,\n// \t\t\t\t\tImageURL: &context.ImageURL{\n// \t\t\t\t\t\tURL: \"https://raw.githubusercontent.com/YaoApp/yao/refs/heads/main/yao/data/icons/icon.png\",\n// \t\t\t\t\t},\n// \t\t\t\t},\n// \t\t\t},\n// \t\t},\n// \t}\n//\n// \tmaxTokens := 200\n// \toptions.MaxCompletionTokens = &maxTokens\n//\n// \tctx := newGPT5TestContext(\"test-gpt5-vision\", \"openai.gpt-5\")\n//\n// \tresponse, err := llmInstance.Post(ctx, messages, options)\n// \tif err != nil {\n// \t\tt.Fatalf(\"Post with vision failed: %v\", err)\n// \t}\n//\n// \tif response == nil {\n// \t\tt.Fatal(\"Response is nil\")\n// \t}\n//\n// \t// Should have content describing the image\n// \t// Content can be string or []ContentPart for multimodal responses\n// \tvar contentStr string\n// \tswitch v := response.Content.(type) {\n// \tcase string:\n// \t\tcontentStr = v\n// \tcase []interface{}:\n// \t\t// Handle []ContentPart serialized as []interface{}\n// \t\tfor _, part := range v {\n// \t\t\tif partMap, ok := part.(map[string]interface{}); ok {\n// \t\t\t\tif text, ok := partMap[\"text\"].(string); ok {\n// \t\t\t\t\tcontentStr += text\n// \t\t\t\t}\n// \t\t\t}\n// \t\t}\n// \tcase []context.ContentPart:\n// \t\tfor _, part := range v {\n// \t\t\tif part.Type == context.ContentText {\n// \t\t\t\tcontentStr += part.Text\n// \t\t\t}\n// \t\t}\n// \tcase nil:\n// \t\t// GPT-5 reasoning models may use all tokens for reasoning, leaving no content\n// \t\tt.Log(\"Content is nil (reasoning model may have used all tokens for reasoning)\")\n// \tdefault:\n// \t\tt.Logf(\"Unexpected content type: %T\", response.Content)\n// \t}\n//\n// \tif contentStr != \"\" {\n// \t\tt.Logf(\"Image description: %s\", contentStr)\n// \t} else if response.Content != nil {\n// \t\tt.Logf(\"Warning: Expected text content describing the image, got empty or non-text content\")\n// \t}\n//\n// \tif response.Usage != nil {\n// \t\tt.Logf(\"Usage: prompt=%d, completion=%d, total=%d\",\n// \t\t\tresponse.Usage.PromptTokens, response.Usage.CompletionTokens, response.Usage.TotalTokens)\n// \t}\n// }\n//\n// // TestGPT5ReasoningEffortWithGPT4o tests that GPT-4o ignores reasoning_effort\n// func TestGPT5ReasoningEffortWithGPT4o(t *testing.T) {\n// \ttest.Prepare(t, config.Conf)\n// \tdefer test.Clean()\n//\n// \t// Use GPT-4o which doesn't support reasoning\n// \tconn, err := connector.Select(\"openai.gpt-4o\")\n// \tif err != nil {\n// \t\tt.Fatalf(\"Failed to select connector: %v\", err)\n// \t}\n//\n// \teffort := \"high\"\n// \toptions := &context.CompletionOptions{\n// \t\tCapabilities: &openai.Capabilities{\n// \t\t\tReasoning: false, // GPT-4o doesn't support reasoning\n// \t\t\tToolCalls: true,\n// \t\t},\n// \t\tReasoningEffort: &effort, // Should be ignored by adapter\n// \t}\n//\n// \tllmInstance, err := llm.New(conn, options)\n// \tif err != nil {\n// \t\tt.Fatalf(\"Failed to create LLM instance: %v\", err)\n// \t}\n//\n// \tmessages := []context.Message{\n// \t\t{\n// \t\t\tRole: context.RoleUser,\n// \t\t\tContent: \"Say 'OK'\",\n// \t\t},\n// \t}\n//\n// \tmaxTokens := 10\n// \toptions.MaxCompletionTokens = &maxTokens\n//\n// \tctx := newGPT5TestContext(\"test-gpt4o-no-reasoning\", \"openai.gpt-4o\")\n//\n// \t// Should succeed (adapter removes reasoning_effort parameter)\n// \tresponse, err := llmInstance.Post(ctx, messages, options)\n// \tif err != nil {\n// \t\tt.Fatalf(\"Post failed: %v\", err)\n// \t}\n//\n// \tif response == nil {\n// \t\tt.Fatal(\"Response is nil\")\n// \t}\n//\n// \t// Should have 0 reasoning tokens (GPT-4o doesn't do reasoning)\n// \tif response.Usage != nil && response.Usage.CompletionTokensDetails != nil {\n// \t\treasoningTokens := response.Usage.CompletionTokensDetails.ReasoningTokens\n// \t\tif reasoningTokens != 0 {\n// \t\t\tt.Errorf(\"Expected reasoning_tokens=0 for GPT-4o, got %d\", reasoningTokens)\n// \t\t} else {\n// \t\t\tt.Log(\"✓ GPT-4o correctly shows reasoning_tokens=0\")\n// \t\t}\n// \t}\n//\n// \tt.Log(\"✓ ReasoningAdapter correctly removed reasoning_effort parameter for GPT-4o\")\n// }\n//\n// // ============================================================================\n// // Helper Functions\n// // ============================================================================\n//\n// // newGPT5TestContext creates a real Context for testing GPT-5 provider\n// func newGPT5TestContext(chatID, connectorID string) *context.Context {\n// \tauthorized := &types.AuthorizedInfo{\n// \t\tSubject: \"test-user\",\n// \t\tClientID: \"test-client\",\n// \t\tUserID: \"test-user-123\",\n// \t\tTeamID: \"test-team-456\",\n// \t\tTenantID: \"test-tenant-789\",\n// \t\tSessionID: \"test-session-id\",\n// \t\tConstraints: types.DataConstraints{\n// \t\t\tTeamOnly: true,\n// \t\t\tExtra: map[string]interface{}{\n// \t\t\t\t\"test\": \"gpt5-provider\",\n// \t\t\t},\n// \t\t},\n// \t}\n//\n// \tctx := context.New(gocontext.Background(), authorized, chatID)\n// \tctx.AssistantID = \"test-assistant\"\n// \tctx.Locale = \"en-us\"\n// \tctx.Theme = \"light\"\n// \tctx.Client = context.Client{\n// \t\tType: \"web\",\n// \t\tUserAgent: \"GPT5ProviderTest/1.0\",\n// \t\tIP: \"127.0.0.1\",\n// \t}\n// \tctx.Referer = context.RefererAPI\n// \tctx.Accept = context.AcceptStandard\n// \tctx.Route = \"/api/test\"\n// \tctx.Metadata = make(map[string]interface{})\n// \treturn ctx\n// }\n" }, { "path": "agent/llm/providers/openai/openai.go", "content": "package openai\n\nimport (\n\tgocontext \"context\"\n\t\"fmt\"\n\t\"strings\"\n\t\"time\"\n\n\tjsoniter \"github.com/json-iterator/go\"\n\t\"github.com/yaoapp/gou/connector\"\n\t\"github.com/yaoapp/gou/http\"\n\tgoullm \"github.com/yaoapp/gou/llm\"\n\t\"github.com/yaoapp/kun/log\"\n\t\"github.com/yaoapp/yao/agent/context\"\n\t\"github.com/yaoapp/yao/agent/i18n\"\n\t\"github.com/yaoapp/yao/agent/llm/adapters\"\n\t\"github.com/yaoapp/yao/agent/llm/providers/base\"\n\t\"github.com/yaoapp/yao/agent/output/message\"\n\t\"github.com/yaoapp/yao/utils/jsonschema\"\n)\n\n// startMessage starts a new message and sends group_start event\n// Note: group_start/group_end events are used for backward compatibility\n// but at LLM level they represent message boundaries, not Agent-level blocks\nfunc (mt *messageTracker) startMessage(messageType message.StreamChunkType, handler message.StreamFunc) {\n\tif mt.active {\n\t\t// End previous message first\n\t\tmt.endMessage(handler)\n\t}\n\n\tmt.active = true\n\t// Generate message ID using context's ID generator\n\tif mt.idGenerator != nil {\n\t\tmt.messageID = mt.idGenerator.GenerateMessageID() // M1, M2, M3...\n\t} else {\n\t\t// Fallback to global generator if no context generator\n\t\tmt.messageID = message.GenerateNanoID()\n\t}\n\tmt.messageType = messageType\n\tmt.startTime = time.Now().UnixMilli()\n\tmt.chunkCount = 0\n\tmt.toolCallInfo = nil\n\n\tif handler != nil {\n\t\tstartData := &message.EventMessageStartData{\n\t\t\tMessageID: mt.messageID,\n\t\t\tType: string(messageType),\n\t\t\tTimestamp: mt.startTime,\n\t\t}\n\t\tif startJSON, err := jsoniter.Marshal(startData); err == nil {\n\t\t\thandler(message.ChunkMessageStart, startJSON)\n\t\t}\n\t}\n}\n\n// startToolCallMessage starts a new tool call message with tool call info\nfunc (mt *messageTracker) startToolCallMessage(toolCallInfo *message.EventToolCallInfo, handler message.StreamFunc) {\n\tif mt.active {\n\t\tmt.endMessage(handler)\n\t}\n\n\tmt.active = true\n\t// Generate message ID using context's ID generator\n\tif mt.idGenerator != nil {\n\t\tmt.messageID = mt.idGenerator.GenerateMessageID() // M1, M2, M3...\n\t} else {\n\t\t// Fallback to global generator if no context generator\n\t\tmt.messageID = message.GenerateNanoID()\n\t}\n\tmt.messageType = message.ChunkToolCall\n\tmt.startTime = time.Now().UnixMilli()\n\tmt.chunkCount = 0\n\tmt.toolCallInfo = toolCallInfo\n\n\tif handler != nil {\n\t\tstartData := &message.EventMessageStartData{\n\t\t\tMessageID: mt.messageID,\n\t\t\tType: string(message.ChunkToolCall),\n\t\t\tTimestamp: mt.startTime,\n\t\t\tToolCall: toolCallInfo,\n\t\t}\n\t\tif startJSON, err := jsoniter.Marshal(startData); err == nil {\n\t\t\thandler(message.ChunkMessageStart, startJSON)\n\t\t}\n\t}\n}\n\n// incrementChunk increments the chunk count for the current message\nfunc (mt *messageTracker) incrementChunk() {\n\tif mt.active {\n\t\tmt.chunkCount++\n\t}\n}\n\n// endMessage ends the current message and sends group_end event\n// Note: group_end event is used for backward compatibility\n// but at LLM level it represents message completion, not Agent-level block\nfunc (mt *messageTracker) endMessage(handler message.StreamFunc) {\n\tif !mt.active {\n\t\treturn\n\t}\n\n\tif handler != nil {\n\t\tendData := &message.EventMessageEndData{\n\t\t\tMessageID: mt.messageID,\n\t\t\tType: string(mt.messageType),\n\t\t\tTimestamp: time.Now().UnixMilli(),\n\t\t\tDurationMs: time.Now().UnixMilli() - mt.startTime,\n\t\t\tChunkCount: mt.chunkCount,\n\t\t\tStatus: \"completed\",\n\t\t}\n\t\tif mt.toolCallInfo != nil {\n\t\t\tendData.ToolCall = mt.toolCallInfo\n\t\t}\n\t\tif endJSON, err := jsoniter.Marshal(endData); err == nil {\n\t\t\thandler(message.ChunkMessageEnd, endJSON)\n\t\t}\n\t}\n\n\tmt.active = false\n\tmt.messageID = \"\"\n\tmt.toolCallInfo = nil\n}\n\n// Provider OpenAI-compatible provider with capability adapters\n// Supports: vision, tool calls, streaming, JSON mode, reasoning\ntype Provider struct {\n\t*base.Provider\n\tadapters []adapters.CapabilityAdapter\n}\n\n// buildAPIURL builds the complete API URL from host and endpoint.\n// Delegates to the shared connector.BuildAPIURL for consistent URL building\n// across the agent LLM path and the sandbox proxy path.\nfunc buildAPIURL(host, endpoint string) string {\n\treturn connector.BuildAPIURL(host, endpoint)\n}\n\n// New create a new OpenAI provider with capability adapters\nfunc New(conn connector.Connector, capabilities *goullm.Capabilities) *Provider {\n\treturn &Provider{\n\t\tProvider: base.NewProvider(conn, capabilities),\n\t\tadapters: buildAdapters(capabilities),\n\t}\n}\n\n// buildAdapters builds capability adapters based on model capabilities\nfunc buildAdapters(cap *goullm.Capabilities) []adapters.CapabilityAdapter {\n\tif cap == nil {\n\t\treturn []adapters.CapabilityAdapter{}\n\t}\n\n\tresult := make([]adapters.CapabilityAdapter, 0)\n\n\t// Tool call adapter\n\tresult = append(result, adapters.NewToolCallAdapter(cap.ToolCalls))\n\n\t// Vision adapter\n\tvisionSupport, visionFormat := context.GetVisionSupport(cap)\n\tif visionSupport {\n\t\tresult = append(result, adapters.NewVisionAdapter(true, visionFormat))\n\t} else if cap.Vision != nil {\n\t\t// Vision explicitly disabled, add adapter to remove image content\n\t\tresult = append(result, adapters.NewVisionAdapter(false, context.VisionFormatNone))\n\t}\n\n\t// Audio adapter\n\tresult = append(result, adapters.NewAudioAdapter(cap.Audio))\n\n\t// Reasoning adapter (always add to handle reasoning_effort and temperature parameters)\n\t// Even if the model doesn't support reasoning, we need the adapter to strip reasoning_effort\n\tif cap.Reasoning {\n\t\t// Detect reasoning format based on capabilities\n\t\tformat := detectReasoningFormat(cap)\n\t\tresult = append(result, adapters.NewReasoningAdapter(format, cap))\n\t} else {\n\t\t// Model doesn't support reasoning, use None format to strip reasoning parameters\n\t\tresult = append(result, adapters.NewReasoningAdapter(adapters.ReasoningFormatNone, cap))\n\t}\n\n\treturn result\n}\n\n// detectReasoningFormat detects the reasoning format based on capabilities\nfunc detectReasoningFormat(cap *goullm.Capabilities) adapters.ReasoningFormat {\n\t// TODO: Implement better detection logic\n\t// For now, default to OpenAI o1 format if reasoning is supported\n\tif cap.Reasoning {\n\t\treturn adapters.ReasoningFormatOpenAI\n\t}\n\treturn adapters.ReasoningFormatNone\n}\n\n// Stream stream completion from OpenAI API\nfunc (p *Provider) Stream(ctx *context.Context, messages []context.Message, options *context.CompletionOptions, handler message.StreamFunc) (*context.CompletionResponse, error) {\n\t// Add debug log\n\ttrace, _ := ctx.Trace()\n\tif trace != nil {\n\t\ttrace.Debug(\"OpenAI Stream: Starting stream request\", map[string]any{\n\t\t\t\"message_count\": len(messages),\n\t\t})\n\t}\n\n\tmaxRetries := 3\n\tvar lastErr error\n\n\t// Get Go context for cancellation support\n\t// Read from Stack.Options if available (call-level override)\n\tgoCtx := ctx.Context\n\tif ctx.Stack != nil && ctx.Stack.Options != nil && ctx.Stack.Options.Context != nil {\n\t\tgoCtx = ctx.Stack.Options.Context\n\t}\n\tif goCtx == nil {\n\t\tgoCtx = gocontext.Background()\n\t}\n\n\t// Make a copy of messages to avoid modifying the original\n\tcurrentMessages := make([]context.Message, len(messages))\n\tcopy(currentMessages, messages)\n\n\t// Outer loop: handle network/API errors with exponential backoff\n\tfor attempt := 0; attempt < maxRetries; attempt++ {\n\t\t// Check if context is cancelled before retry\n\t\tselect {\n\t\tcase <-goCtx.Done():\n\t\t\treturn nil, fmt.Errorf(\"context cancelled: %w\", goCtx.Err())\n\t\tdefault:\n\t\t}\n\n\t\t// Check for force interrupt before retry\n\t\tif ctx.Interrupt != nil {\n\t\t\tif signal := ctx.Interrupt.Peek(); signal != nil && signal.Type == context.InterruptForce {\n\t\t\t\treturn nil, fmt.Errorf(\"force interrupted by user\")\n\t\t\t}\n\t\t}\n\n\t\tif attempt > 0 {\n\t\t\t// Exponential backoff: 1s, 2s, 4s\n\t\t\tbackoff := time.Duration(1< 0 {\n\t\t\tchoice := chunk.Choices[0]\n\t\t\tdelta := choice.Delta\n\n\t\t\t// Update accumulator metadata\n\t\t\tif accumulator.id == \"\" {\n\t\t\t\taccumulator.id = chunk.ID\n\t\t\t\taccumulator.model = chunk.Model\n\t\t\t\taccumulator.created = chunk.Created\n\t\t\t}\n\n\t\t\t// Handle role\n\t\t\tif delta.Role != \"\" {\n\t\t\t\taccumulator.role = delta.Role\n\t\t\t}\n\n\t\t\t// Handle reasoning content (DeepSeek R1)\n\t\t\tif delta.ReasoningContent != \"\" {\n\t\t\t\t// Start thinking message if not active\n\t\t\t\tif !messageTracker.active || messageTracker.messageType != message.ChunkThinking {\n\t\t\t\t\tmessageTracker.startMessage(message.ChunkThinking, handler)\n\t\t\t\t}\n\n\t\t\t\taccumulator.reasoningContent += delta.ReasoningContent\n\t\t\t\tif handler != nil {\n\t\t\t\t\thandler(message.ChunkThinking, []byte(delta.ReasoningContent))\n\t\t\t\t\tmessageTracker.incrementChunk()\n\t\t\t\t}\n\t\t\t}\n\n\t\t\t// Handle content\n\t\t\tif delta.Content != \"\" {\n\t\t\t\t// Start text message if not active\n\t\t\t\tif !messageTracker.active || messageTracker.messageType != message.ChunkText {\n\t\t\t\t\tmessageTracker.startMessage(message.ChunkText, handler)\n\t\t\t\t}\n\n\t\t\t\taccumulator.content += delta.Content\n\t\t\t\tif handler != nil {\n\t\t\t\t\thandler(message.ChunkText, []byte(delta.Content))\n\t\t\t\t\tmessageTracker.incrementChunk()\n\t\t\t\t}\n\t\t\t}\n\n\t\t\t// Handle refusal\n\t\t\tif delta.Refusal != \"\" {\n\t\t\t\t// Start refusal message if not active\n\t\t\t\tif !messageTracker.active || messageTracker.messageType != message.ChunkRefusal {\n\t\t\t\t\tmessageTracker.startMessage(message.ChunkRefusal, handler)\n\t\t\t\t}\n\n\t\t\t\taccumulator.refusal += delta.Refusal\n\t\t\t\tif handler != nil {\n\t\t\t\t\thandler(message.ChunkRefusal, []byte(delta.Refusal))\n\t\t\t\t\tmessageTracker.incrementChunk()\n\t\t\t\t}\n\t\t\t}\n\n\t\t\t// Handle tool calls\n\t\t\tif len(delta.ToolCalls) > 0 {\n\t\t\t\tfor _, tc := range delta.ToolCalls {\n\t\t\t\t\tif _, exists := accumulator.toolCalls[tc.Index]; !exists {\n\t\t\t\t\t\taccumulator.toolCalls[tc.Index] = &accumulatedToolCall{}\n\n\t\t\t\t\t\t// Start new tool call message when we first see this tool call\n\t\t\t\t\t\tif tc.ID != \"\" {\n\t\t\t\t\t\t\ttoolCallInfo := &message.EventToolCallInfo{\n\t\t\t\t\t\t\t\tID: tc.ID,\n\t\t\t\t\t\t\t\tName: tc.Function.Name, // May be partial or empty initially\n\t\t\t\t\t\t\t\tIndex: tc.Index,\n\t\t\t\t\t\t\t}\n\t\t\t\t\t\t\tmessageTracker.startToolCallMessage(toolCallInfo, handler)\n\t\t\t\t\t\t}\n\t\t\t\t\t}\n\t\t\t\t\taccTC := accumulator.toolCalls[tc.Index]\n\n\t\t\t\t\tif tc.ID != \"\" {\n\t\t\t\t\t\taccTC.id = tc.ID\n\t\t\t\t\t}\n\t\t\t\t\tif tc.Type != \"\" {\n\t\t\t\t\t\taccTC.typ = tc.Type\n\t\t\t\t\t}\n\t\t\t\t\tif tc.Function.Name != \"\" {\n\t\t\t\t\t\taccTC.functionName = tc.Function.Name\n\t\t\t\t\t\t// Update tool call info in tracker\n\t\t\t\t\t\tif messageTracker.active && messageTracker.toolCallInfo != nil {\n\t\t\t\t\t\t\tmessageTracker.toolCallInfo.Name = tc.Function.Name\n\t\t\t\t\t\t}\n\t\t\t\t\t}\n\t\t\t\t\tif tc.Function.Arguments != \"\" {\n\t\t\t\t\t\taccTC.functionArgs += tc.Function.Arguments\n\t\t\t\t\t\t// Update tool call info in tracker\n\t\t\t\t\t\tif messageTracker.active && messageTracker.toolCallInfo != nil {\n\t\t\t\t\t\t\tmessageTracker.toolCallInfo.Arguments = accTC.functionArgs\n\t\t\t\t\t\t}\n\t\t\t\t\t}\n\t\t\t\t}\n\n\t\t\t\t// Notify handler of tool call progress\n\t\t\t\t// Send the raw delta from OpenAI (as JSON bytes)\n\t\t\t\t// Handler will convert to object for frontend merge\n\t\t\t\tif handler != nil {\n\t\t\t\t\ttoolCallData, _ := jsoniter.Marshal(delta.ToolCalls)\n\t\t\t\t\thandler(message.ChunkToolCall, toolCallData)\n\t\t\t\t\tmessageTracker.incrementChunk()\n\t\t\t\t}\n\t\t\t}\n\n\t\t\t// Handle finish reason\n\t\t\tif choice.FinishReason != nil && *choice.FinishReason != \"\" {\n\t\t\t\taccumulator.finishReason = *choice.FinishReason\n\t\t\t}\n\n\t\t\t// Handle usage (in choices, for older API versions)\n\t\t\tif chunk.Usage != nil {\n\t\t\t\taccumulator.usage = &message.UsageInfo{\n\t\t\t\t\tPromptTokens: chunk.Usage.PromptTokens,\n\t\t\t\t\tCompletionTokens: chunk.Usage.CompletionTokens,\n\t\t\t\t\tTotalTokens: chunk.Usage.TotalTokens,\n\t\t\t\t}\n\t\t\t}\n\t\t}\n\n\t\t// Check for usage at the top level (newer API versions with stream_options)\n\t\tif chunk.Usage != nil && accumulator.usage == nil {\n\t\t\taccumulator.usage = &message.UsageInfo{\n\t\t\t\tPromptTokens: chunk.Usage.PromptTokens,\n\t\t\t\tCompletionTokens: chunk.Usage.CompletionTokens,\n\t\t\t\tTotalTokens: chunk.Usage.TotalTokens,\n\t\t\t}\n\t\t}\n\n\t\treturn http.HandlerReturnOk\n\t}\n\n\t// Log request for debugging\n\tif trace != nil {\n\t\tif requestBodyJSON, marshalErr := jsoniter.Marshal(requestBody); marshalErr == nil {\n\t\t\ttrace.Debug(\"OpenAI Stream Request\", map[string]any{\n\t\t\t\t\"url\": url,\n\t\t\t\t\"body\": string(requestBodyJSON),\n\t\t\t})\n\t\t}\n\t}\n\n\t// Buffer to capture non-SSE error responses\n\tvar errorBuffer strings.Builder\n\terrorDetected := false\n\n\t// Wrap streamHandler to detect JSON error responses\n\t// Note: API error responses are raw JSON without \"data: \" prefix\n\t// Normal SSE data always starts with \"data: \" prefix\n\twrappedHandler := func(data []byte) int {\n\t\tdataStr := string(data)\n\t\ttrimmed := strings.TrimSpace(dataStr)\n\n\t\t// Skip empty lines\n\t\tif trimmed == \"\" {\n\t\t\treturn http.HandlerReturnOk\n\t\t}\n\n\t\t// Normal SSE data starts with \"data: \" - pass to streamHandler\n\t\tif strings.HasPrefix(dataStr, \"data: \") {\n\t\t\treturn streamHandler(data)\n\t\t}\n\n\t\t// Detect if this looks like a JSON error response (raw JSON without \"data: \" prefix)\n\t\t// API errors are returned as raw JSON: {\"error\": {...}}\n\t\tif strings.HasPrefix(trimmed, \"{\") && strings.Contains(dataStr, `\"error\"`) {\n\t\t\terrorDetected = true\n\t\t}\n\n\t\t// If error detected, accumulate all data for parsing\n\t\tif errorDetected {\n\t\t\terrorBuffer.Write(data)\n\t\t\terrorBuffer.WriteString(\"\\n\")\n\t\t\treturn http.HandlerReturnOk\n\t\t}\n\n\t\t// Unknown format, pass to streamHandler (it will skip non-SSE data)\n\t\treturn streamHandler(data)\n\t}\n\n\t// Make streaming request (goCtx already set at function start)\n\tlog.Trace(\"[LLM] Starting HTTP Stream request: url=%s\", url)\n\terr = req.Stream(goCtx, \"POST\", requestBody, wrappedHandler)\n\tlog.Trace(\"[LLM] HTTP Stream request returned: err=%v\", err)\n\n\t// Check if we captured an error response\n\tif errorDetected && errorBuffer.Len() > 0 {\n\t\terrorJSON := errorBuffer.String()\n\t\tif trace != nil {\n\t\t\ttrace.Error(i18n.T(ctx.Locale, \"llm.openai.stream.api_error\"), map[string]any{\"response\": errorJSON}) // \"OpenAI API returned error response\"\n\t\t}\n\n\t\t// Try to parse error\n\t\tvar apiError struct {\n\t\t\tError struct {\n\t\t\t\tMessage string `json:\"message\"`\n\t\t\t\tType string `json:\"type\"`\n\t\t\t\tParam string `json:\"param\"`\n\t\t\t\tCode string `json:\"code\"`\n\t\t\t} `json:\"error\"`\n\t\t}\n\n\t\tif parseErr := jsoniter.UnmarshalFromString(errorJSON, &apiError); parseErr == nil && apiError.Error.Message != \"\" {\n\t\t\terr = fmt.Errorf(\"OpenAI API error: %s (type: %s, param: %s, code: %s)\",\n\t\t\t\tapiError.Error.Message, apiError.Error.Type, apiError.Error.Param, apiError.Error.Code)\n\t\t} else {\n\t\t\terr = fmt.Errorf(\"OpenAI API error: %s\", strings.TrimSpace(errorJSON))\n\t\t}\n\t}\n\n\t// Check if error is due to context cancellation FIRST (before logging)\n\t// This prevents blocking on trace operations when context is cancelled\n\tif err != nil && goCtx.Err() != nil {\n\t\tlog.Trace(\"[LLM] Context cancelled detected, skipping handler calls and returning\")\n\t\t// NOTE: Do NOT call handler or groupTracker.endGroup here\n\t\t// The connection is already closed, calling handler may block indefinitely\n\t\t// Just return the error immediately\n\t\treturn nil, fmt.Errorf(\"stream cancelled: %w\", goCtx.Err())\n\t}\n\n\t// Log any error from streaming (only if not cancelled)\n\tif err != nil && trace != nil {\n\t\ttrace.Error(i18n.T(ctx.Locale, \"llm.openai.stream.error\"), map[string]any{\"error\": err.Error()}) // \"OpenAI Stream Error\"\n\t}\n\n\tif err != nil {\n\t\t// End current message if active\n\t\tmessageTracker.endMessage(handler)\n\n\t\t// Notify handler of error if provided\n\t\tif handler != nil {\n\t\t\terrData := []byte(err.Error())\n\t\t\thandler(message.ChunkError, errData)\n\t\t}\n\t\treturn nil, fmt.Errorf(\"streaming request failed: %w\", err)\n\t}\n\n\t// Check if we received any data\n\tif accumulator.id == \"\" {\n\t\tif trace != nil {\n\t\t\ttrace.Warn(\"OpenAI stream completed but no data was received\")\n\n\t\t\t// Log request details for debugging\n\t\t\tif requestBodyJSON, err := jsoniter.Marshal(requestBody); err == nil {\n\t\t\t\ttrace.Error(i18n.T(ctx.Locale, \"llm.openai.stream.no_data\"), map[string]any{\"body\": string(requestBodyJSON)}) // \"Request body that caused empty response\"\n\t\t\t}\n\t\t\ttrace.Error(i18n.T(ctx.Locale, \"llm.openai.stream.no_data_info\"), map[string]any{ // \"Request details\"\n\t\t\t\t\"url\": url,\n\t\t\t\t\"model\": accumulator.model,\n\t\t\t\t\"created\": accumulator.created,\n\t\t\t})\n\t\t}\n\n\t\terr := fmt.Errorf(\"no data received from OpenAI API\")\n\n\t\t// End current message if active\n\t\tmessageTracker.endMessage(handler)\n\n\t\t// Notify handler of error if provided\n\t\tif handler != nil {\n\t\t\terrData := []byte(err.Error())\n\t\t\thandler(message.ChunkError, errData)\n\t\t}\n\t\treturn nil, err\n\t}\n\n\t// Build final response\n\tresponse := &context.CompletionResponse{\n\t\tID: accumulator.id,\n\t\tObject: \"chat.completion\",\n\t\tCreated: accumulator.created,\n\t\tModel: accumulator.model,\n\t\tRole: accumulator.role,\n\t\tContent: accumulator.content,\n\t\tReasoningContent: accumulator.reasoningContent,\n\t\tRefusal: accumulator.refusal,\n\t\tFinishReason: accumulator.finishReason,\n\t\tUsage: accumulator.usage,\n\t}\n\n\t// Convert accumulated tool calls to ToolCall slice\n\tif len(accumulator.toolCalls) > 0 {\n\t\ttoolCalls := make([]context.ToolCall, 0, len(accumulator.toolCalls))\n\t\tfor i := 0; i < len(accumulator.toolCalls); i++ {\n\t\t\tif tc, exists := accumulator.toolCalls[i]; exists {\n\t\t\t\ttoolCalls = append(toolCalls, context.ToolCall{\n\t\t\t\t\tID: tc.id,\n\t\t\t\t\tType: context.ToolCallType(tc.typ),\n\t\t\t\t\tFunction: context.Function{\n\t\t\t\t\t\tName: tc.functionName,\n\t\t\t\t\t\tArguments: tc.functionArgs,\n\t\t\t\t\t},\n\t\t\t\t})\n\t\t\t}\n\t\t}\n\t\tresponse.ToolCalls = toolCalls\n\n\t\t// Validate tool call results if schema is provided\n\t\t// Note: If validation fails, we log the error but DO NOT return error\n\t\t// Instead, we let the response through so Agent layer can handle it\n\t\t// Agent layer will re-validate and provide better error feedback to LLM\n\t\tif err := p.validateToolCallResults(options, toolCalls); err != nil {\n\t\t\t// Log validation error\n\t\t\tif trace, _ := ctx.Trace(); trace != nil {\n\t\t\t\ttrace.Warn(\"Tool call validation failed at LLM layer, passing to Agent layer for handling\", map[string]any{\n\t\t\t\t\t\"error\": err.Error(),\n\t\t\t\t})\n\t\t\t}\n\t\t\t// End current message\n\t\t\tmessageTracker.endMessage(handler)\n\n\t\t\t// Continue and return response (don't return error)\n\t\t\t// Agent layer will handle validation and retry\n\t\t}\n\t}\n\n\t// End final message if still active\n\tmessageTracker.endMessage(handler)\n\n\treturn response, nil\n}\n\n// Post post completion request to OpenAI API\nfunc (p *Provider) Post(ctx *context.Context, messages []context.Message, options *context.CompletionOptions) (*context.CompletionResponse, error) {\n\t// Add debug log\n\ttrace, _ := ctx.Trace()\n\tif trace != nil {\n\t\ttrace.Debug(\"OpenAI Post: Starting non-stream request\", map[string]any{\n\t\t\t\"message_count\": len(messages),\n\t\t})\n\t}\n\n\tmaxRetries := 3\n\tvar lastErr error\n\n\t// Get Go context for cancellation support\n\t// Read from Stack.Options if available (call-level override)\n\tgoCtx := ctx.Context\n\tif ctx.Stack != nil && ctx.Stack.Options != nil && ctx.Stack.Options.Context != nil {\n\t\tgoCtx = ctx.Stack.Options.Context\n\t}\n\tif goCtx == nil {\n\t\tgoCtx = gocontext.Background()\n\t}\n\n\t// Make a copy of messages to avoid modifying the original\n\tcurrentMessages := make([]context.Message, len(messages))\n\tcopy(currentMessages, messages)\n\n\t// Outer loop: handle network/API errors with exponential backoff\n\tfor attempt := 0; attempt < maxRetries; attempt++ {\n\t\t// Check if context is cancelled before retry\n\t\tselect {\n\t\tcase <-goCtx.Done():\n\t\t\treturn nil, fmt.Errorf(\"context cancelled: %w\", goCtx.Err())\n\t\tdefault:\n\t\t}\n\n\t\tif attempt > 0 {\n\t\t\t// Exponential backoff\n\t\t\tbackoff := time.Duration(1< 0 {\n\t\tif err := p.validateToolCallResults(options, response.ToolCalls); err != nil {\n\t\t\treturn nil, fmt.Errorf(\"tool call validation failed: %w\", err)\n\t\t}\n\t}\n\n\treturn response, nil\n}\n\n// buildRequestBody builds the request body for OpenAI API\nfunc (p *Provider) buildRequestBody(messages []context.Message, options *context.CompletionOptions, streaming bool) (map[string]interface{}, error) {\n\tif options == nil {\n\t\treturn nil, fmt.Errorf(\"options are required\")\n\t}\n\n\t// Get model and other settings from connector\n\tsetting := p.Connector.Setting()\n\tmodel, ok := setting[\"model\"].(string)\n\tif !ok || model == \"\" {\n\t\treturn nil, fmt.Errorf(\"model is not set in connector\")\n\t}\n\n\t// Get thinking setting from connector (for models that support reasoning/thinking mode)\n\tvar thinkingSetting interface{}\n\tif thinking, exists := setting[\"thinking\"]; exists {\n\t\tthinkingSetting = thinking\n\t}\n\n\t// Convert messages to API format\n\tapiMessages := make([]map[string]interface{}, 0, len(messages))\n\tfor _, msg := range messages {\n\t\tapiMsg := map[string]interface{}{\n\t\t\t\"role\": string(msg.Role),\n\t\t}\n\n\t\tif msg.Content != nil {\n\t\t\t// Check if Content is []context.ContentPart and convert to API format\n\t\t\tif parts, ok := msg.Content.([]context.ContentPart); ok {\n\t\t\t\tapiParts := make([]map[string]interface{}, 0, len(parts))\n\t\t\t\tfor _, part := range parts {\n\t\t\t\t\tapiPart := map[string]interface{}{\n\t\t\t\t\t\t\"type\": string(part.Type),\n\t\t\t\t\t}\n\t\t\t\t\tswitch part.Type {\n\t\t\t\t\tcase context.ContentText:\n\t\t\t\t\t\tapiPart[\"text\"] = part.Text\n\t\t\t\t\tcase context.ContentImageURL:\n\t\t\t\t\t\tif part.ImageURL != nil {\n\t\t\t\t\t\t\tapiPart[\"image_url\"] = map[string]interface{}{\n\t\t\t\t\t\t\t\t\"url\": part.ImageURL.URL,\n\t\t\t\t\t\t\t}\n\t\t\t\t\t\t\tif part.ImageURL.Detail != \"\" {\n\t\t\t\t\t\t\t\tapiPart[\"image_url\"].(map[string]interface{})[\"detail\"] = part.ImageURL.Detail\n\t\t\t\t\t\t\t}\n\t\t\t\t\t\t}\n\t\t\t\t\tcase context.ContentInputAudio:\n\t\t\t\t\t\tif part.InputAudio != nil {\n\t\t\t\t\t\t\tapiPart[\"input_audio\"] = part.InputAudio\n\t\t\t\t\t\t}\n\t\t\t\t\t}\n\t\t\t\t\tapiParts = append(apiParts, apiPart)\n\t\t\t\t}\n\t\t\t\tapiMsg[\"content\"] = apiParts\n\t\t\t} else {\n\t\t\t\t// Content is string or already in map format, use as is\n\t\t\t\tapiMsg[\"content\"] = msg.Content\n\t\t\t}\n\t\t}\n\n\t\tif msg.Name != nil {\n\t\t\tapiMsg[\"name\"] = *msg.Name\n\t\t}\n\n\t\tif msg.ToolCallID != nil {\n\t\t\tapiMsg[\"tool_call_id\"] = *msg.ToolCallID\n\t\t}\n\n\t\tif len(msg.ToolCalls) > 0 {\n\t\t\tapiMsg[\"tool_calls\"] = msg.ToolCalls\n\t\t}\n\n\t\tif msg.Refusal != nil {\n\t\t\tapiMsg[\"refusal\"] = *msg.Refusal\n\t\t}\n\n\t\tapiMessages = append(apiMessages, apiMsg)\n\t}\n\n\t// Build request body\n\tbody := map[string]interface{}{\n\t\t\"model\": model,\n\t\t\"messages\": apiMessages,\n\t\t\"stream\": streaming,\n\t}\n\n\t// Add optional parameters\n\tif options.Temperature != nil {\n\t\tbody[\"temperature\"] = *options.Temperature\n\t}\n\n\t// Use max_completion_tokens (modern API parameter for GPT-5+)\n\t// GPT-5 models only support max_completion_tokens (not max_tokens)\n\tif options.MaxCompletionTokens != nil {\n\t\tbody[\"max_completion_tokens\"] = *options.MaxCompletionTokens\n\t} else if options.MaxTokens != nil {\n\t\t// Fallback: convert MaxTokens to max_completion_tokens for compatibility\n\t\tbody[\"max_completion_tokens\"] = *options.MaxTokens\n\t}\n\n\tif options.TopP != nil {\n\t\tbody[\"top_p\"] = *options.TopP\n\t}\n\n\tif options.N != nil {\n\t\tbody[\"n\"] = *options.N\n\t}\n\n\tif options.Stop != nil {\n\t\tbody[\"stop\"] = options.Stop\n\t}\n\n\tif options.PresencePenalty != nil {\n\t\tbody[\"presence_penalty\"] = *options.PresencePenalty\n\t}\n\n\tif options.FrequencyPenalty != nil {\n\t\tbody[\"frequency_penalty\"] = *options.FrequencyPenalty\n\t}\n\n\tif len(options.LogitBias) > 0 {\n\t\tbody[\"logit_bias\"] = options.LogitBias\n\t}\n\n\tif options.User != \"\" {\n\t\tbody[\"user\"] = options.User\n\t}\n\n\tif options.ResponseFormat != nil {\n\t\t// Build response_format according to OpenAI API requirements\n\t\tresponseFormat := map[string]interface{}{\n\t\t\t\"type\": options.ResponseFormat.Type,\n\t\t}\n\n\t\t// For json_schema type, include the schema details\n\t\tif options.ResponseFormat.Type == context.ResponseFormatJSONSchema && options.ResponseFormat.JSONSchema != nil {\n\t\t\tresponseFormat[\"json_schema\"] = options.ResponseFormat.JSONSchema\n\t\t}\n\n\t\tbody[\"response_format\"] = responseFormat\n\t}\n\n\tif options.Seed != nil {\n\t\tbody[\"seed\"] = *options.Seed\n\t}\n\n\tif len(options.Tools) > 0 {\n\t\tbody[\"tools\"] = options.Tools\n\t}\n\n\tif options.ToolChoice != nil {\n\t\tbody[\"tool_choice\"] = options.ToolChoice\n\t}\n\n\t// Reasoning effort (o1 and GPT-5 models)\n\tif options.ReasoningEffort != nil {\n\t\tbody[\"reasoning_effort\"] = *options.ReasoningEffort\n\t}\n\n\t// For streaming, include usage info by default\n\tif streaming {\n\t\tif options.StreamOptions != nil {\n\t\t\tbody[\"stream_options\"] = options.StreamOptions\n\t\t} else {\n\t\t\t// Default: include usage info in streaming response\n\t\t\tbody[\"stream_options\"] = map[string]interface{}{\n\t\t\t\t\"include_usage\": true,\n\t\t\t}\n\t\t}\n\t}\n\n\tif options.Audio != nil {\n\t\tbody[\"audio\"] = options.Audio\n\t}\n\n\t// Add thinking parameter for models that support reasoning/thinking mode\n\tif thinkingSetting != nil {\n\t\tbody[\"thinking\"] = thinkingSetting\n\t}\n\n\treturn body, nil\n}\n\n// validateToolCallResults validates tool call arguments against JSON schema\nfunc (p *Provider) validateToolCallResults(options *context.CompletionOptions, toolCalls []context.ToolCall) error {\n\tif options == nil || options.Tools == nil || len(options.Tools) == 0 {\n\t\treturn nil\n\t}\n\n\t// Build tool schema map for quick lookup\n\ttoolSchemas := make(map[string]interface{})\n\tfor _, tool := range options.Tools {\n\t\tif function, ok := tool[\"function\"].(map[string]interface{}); ok {\n\t\t\tif name, ok := function[\"name\"].(string); ok {\n\t\t\t\tif parameters, ok := function[\"parameters\"]; ok {\n\t\t\t\t\ttoolSchemas[name] = parameters\n\t\t\t\t}\n\t\t\t}\n\t\t}\n\t}\n\n\t// Validate each tool call\n\tfor _, tc := range toolCalls {\n\t\tschema, hasSchema := toolSchemas[tc.Function.Name]\n\t\tif !hasSchema {\n\t\t\tcontinue // No schema to validate against\n\t\t}\n\n\t\t// Parse arguments JSON\n\t\tvar args interface{}\n\t\tif err := jsoniter.UnmarshalFromString(tc.Function.Arguments, &args); err != nil {\n\t\t\treturn fmt.Errorf(\"tool call %s has invalid JSON arguments: %w\", tc.Function.Name, err)\n\t\t}\n\n\t\t// Validate against schema\n\t\tif err := jsonschema.ValidateData(schema, args); err != nil {\n\t\t\treturn fmt.Errorf(\"tool call %s arguments validation failed: %w\", tc.Function.Name, err)\n\t\t}\n\t}\n\n\treturn nil\n}\n\n// isToolCallValidationError checks if an error is a tool call validation error\nfunc isToolCallValidationError(err error) bool {\n\tif err == nil {\n\t\treturn false\n\t}\n\terrStr := err.Error()\n\treturn strings.Contains(errStr, \"tool call validation failed\") ||\n\t\tstrings.Contains(errStr, \"arguments validation failed\")\n}\n\n// isRetryableError checks if an error is retryable\nfunc isRetryableError(err error) bool {\n\tif err == nil {\n\t\treturn false\n\t}\n\n\terrStr := err.Error()\n\n\t// Retryable: network errors, timeouts, rate limits, server errors\n\tretryablePatterns := []string{\n\t\t\"timeout\",\n\t\t\"connection refused\",\n\t\t\"connection reset\",\n\t\t\"EOF\",\n\t\t\"HTTP 429\", // Rate limit\n\t\t\"HTTP 500\", // Internal server error\n\t\t\"HTTP 502\", // Bad gateway\n\t\t\"HTTP 503\", // Service unavailable\n\t\t\"HTTP 504\", // Gateway timeout\n\t}\n\n\tfor _, pattern := range retryablePatterns {\n\t\tif strings.Contains(strings.ToLower(errStr), strings.ToLower(pattern)) {\n\t\t\treturn true\n\t\t}\n\t}\n\n\treturn false\n}\n" }, { "path": "agent/llm/providers/openai/openai_test.go", "content": "package openai_test\n\nimport (\n\tgocontext \"context\"\n\t\"encoding/json\"\n\t\"strings\"\n\t\"testing\"\n\t\"time\"\n\n\t\"github.com/yaoapp/gou/connector\"\n\t\"github.com/yaoapp/gou/connector/openai\"\n\t\"github.com/yaoapp/yao/agent/context\"\n\t\"github.com/yaoapp/yao/agent/llm\"\n\t\"github.com/yaoapp/yao/agent/output/message\"\n\t\"github.com/yaoapp/yao/config\"\n\t\"github.com/yaoapp/yao/openapi/oauth/types\"\n\t\"github.com/yaoapp/yao/test\"\n)\n\n// TestOpenAIStreamBasic tests basic streaming completion with short output\nfunc TestOpenAIStreamBasic(t *testing.T) {\n\ttest.Prepare(t, config.Conf)\n\tdefer test.Clean()\n\n\t// Create connector from real configuration\n\tconn, err := connector.Select(\"openai.gpt-4o\")\n\tif err != nil {\n\t\tt.Fatalf(\"Failed to select connector: %v\", err)\n\t}\n\n\t// Create LLM instance with capabilities\n\toptions := &context.CompletionOptions{\n\t\tCapabilities: &openai.Capabilities{\n\t\t\tStreaming: true,\n\t\t\tToolCalls: true,\n\t\t},\n\t}\n\n\tllmInstance, err := llm.New(conn, options)\n\tif err != nil {\n\t\tt.Fatalf(\"Failed to create LLM instance: %v\", err)\n\t}\n\n\t// Prepare messages with concise prompt\n\tmessages := []context.Message{\n\t\t{\n\t\t\tRole: context.RoleUser,\n\t\t\tContent: \"Say 'Hello' in one word.\",\n\t\t},\n\t}\n\n\t// Set short max tokens to ensure quick response\n\tmaxTokens := 5\n\toptions.MaxTokens = &maxTokens\n\n\t// Create context\n\tctx := newTestContext(\"test-stream-basic\", \"openai.gpt-4o\")\n\n\t// Track streaming chunks\n\tvar chunks []string\n\thandler := func(chunkType message.StreamChunkType, data []byte) int {\n\t\tchunks = append(chunks, string(data))\n\t\tt.Logf(\"Stream chunk [%s]: %s\", chunkType, string(data))\n\t\treturn 0 // Continue\n\t}\n\n\t// Call Stream\n\tresponse, err := llmInstance.Stream(ctx, messages, options, handler)\n\tif err != nil {\n\t\tt.Fatalf(\"Stream failed: %v\", err)\n\t}\n\n\t// Validate response\n\tif response == nil {\n\t\tt.Fatal(\"Response is nil\")\n\t}\n\n\tif response.ID == \"\" {\n\t\tt.Error(\"Response ID is empty\")\n\t}\n\tif response.Model == \"\" {\n\t\tt.Error(\"Response Model is empty\")\n\t}\n\tif response.Content == \"\" {\n\t\tt.Error(\"Response content is empty\")\n\t}\n\tif response.FinishReason == \"\" {\n\t\tt.Error(\"FinishReason is empty\")\n\t}\n\tif response.Usage == nil {\n\t\tt.Error(\"Response Usage is nil\")\n\t} else {\n\t\tif response.Usage.TotalTokens == 0 {\n\t\t\tt.Error(\"Response Usage.TotalTokens is 0\")\n\t\t}\n\t\tt.Logf(\"Usage: prompt=%d, completion=%d, total=%d\",\n\t\t\tresponse.Usage.PromptTokens, response.Usage.CompletionTokens, response.Usage.TotalTokens)\n\t}\n\tif len(chunks) == 0 {\n\t\tt.Error(\"No streaming chunks received\")\n\t}\n\n\tt.Logf(\"Final response: %+v\", response)\n\tt.Logf(\"Total chunks received: %d\", len(chunks))\n}\n\n// TestOpenAIPostBasic tests basic non-streaming completion\nfunc TestOpenAIPostBasic(t *testing.T) {\n\ttest.Prepare(t, config.Conf)\n\tdefer test.Clean()\n\n\t// Create connector\n\tconn, err := connector.Select(\"openai.gpt-4o\")\n\tif err != nil {\n\t\tt.Fatalf(\"Failed to select connector: %v\", err)\n\t}\n\n\t// Create LLM instance\n\toptions := &context.CompletionOptions{\n\t\tCapabilities: &openai.Capabilities{\n\t\t\tToolCalls: true,\n\t\t},\n\t}\n\n\tllmInstance, err := llm.New(conn, options)\n\tif err != nil {\n\t\tt.Fatalf(\"Failed to create LLM instance: %v\", err)\n\t}\n\n\t// Prepare messages with concise prompt\n\tmessages := []context.Message{\n\t\t{\n\t\t\tRole: context.RoleUser,\n\t\t\tContent: \"Reply with only the word 'OK'.\",\n\t\t},\n\t}\n\n\t// Set short max tokens\n\tmaxTokens := 5\n\toptions.MaxTokens = &maxTokens\n\n\t// Create context\n\tctx := newTestContext(\"test-stream-basic\", \"openai.gpt-4o\")\n\n\t// Call Post\n\tresponse, err := llmInstance.Post(ctx, messages, options)\n\tif err != nil {\n\t\tt.Fatalf(\"Post failed: %v\", err)\n\t}\n\n\t// Validate response\n\tif response == nil {\n\t\tt.Fatal(\"Response is nil\")\n\t}\n\n\tif response.ID == \"\" {\n\t\tt.Error(\"Response ID is empty\")\n\t}\n\tif response.Model == \"\" {\n\t\tt.Error(\"Response Model is empty\")\n\t}\n\tif response.Content == \"\" {\n\t\tt.Error(\"Response content is empty\")\n\t}\n\tif response.FinishReason == \"\" {\n\t\tt.Error(\"FinishReason is empty\")\n\t}\n\tif response.Usage == nil {\n\t\tt.Error(\"Response Usage is nil\")\n\t} else {\n\t\tif response.Usage.TotalTokens == 0 {\n\t\t\tt.Error(\"Response Usage.TotalTokens is 0\")\n\t\t}\n\t\tt.Logf(\"Usage: prompt=%d, completion=%d, total=%d\",\n\t\t\tresponse.Usage.PromptTokens, response.Usage.CompletionTokens, response.Usage.TotalTokens)\n\t}\n\n\tt.Logf(\"Response: %+v\", response)\n}\n\n// TestOpenAIStreamWithToolCalls tests streaming with tool calls and JSON schema validation\nfunc TestOpenAIStreamWithToolCalls(t *testing.T) {\n\ttest.Prepare(t, config.Conf)\n\tdefer test.Clean()\n\n\t// Create connector\n\tconn, err := connector.Select(\"openai.gpt-4o\")\n\tif err != nil {\n\t\tt.Fatalf(\"Failed to select connector: %v\", err)\n\t}\n\n\t// Create LLM instance with tool call capabilities\n\toptions := &context.CompletionOptions{\n\t\tCapabilities: &openai.Capabilities{\n\t\t\tStreaming: true,\n\t\t\tToolCalls: true,\n\t\t},\n\t}\n\n\t// Define a simple weather tool with JSON schema\n\tweatherTool := map[string]interface{}{\n\t\t\"type\": \"function\",\n\t\t\"function\": map[string]interface{}{\n\t\t\t\"name\": \"get_weather\",\n\t\t\t\"description\": \"Get the current weather for a location\",\n\t\t\t\"parameters\": map[string]interface{}{\n\t\t\t\t\"type\": \"object\",\n\t\t\t\t\"properties\": map[string]interface{}{\n\t\t\t\t\t\"location\": map[string]interface{}{\n\t\t\t\t\t\t\"type\": \"string\",\n\t\t\t\t\t\t\"description\": \"The city and state, e.g. San Francisco, CA\",\n\t\t\t\t\t},\n\t\t\t\t\t\"unit\": map[string]interface{}{\n\t\t\t\t\t\t\"type\": \"string\",\n\t\t\t\t\t\t\"enum\": []string{\"celsius\", \"fahrenheit\"},\n\t\t\t\t\t},\n\t\t\t\t},\n\t\t\t\t\"required\": []string{\"location\"},\n\t\t\t},\n\t\t},\n\t}\n\n\toptions.Tools = []map[string]interface{}{weatherTool}\n\toptions.ToolChoice = \"auto\"\n\n\tllmInstance, err := llm.New(conn, options)\n\tif err != nil {\n\t\tt.Fatalf(\"Failed to create LLM instance: %v\", err)\n\t}\n\n\t// Prepare messages that should trigger tool call\n\tmessages := []context.Message{\n\t\t{\n\t\t\tRole: context.RoleUser,\n\t\t\tContent: \"What's the weather in Tokyo? Use celsius.\",\n\t\t},\n\t}\n\n\t// Create context\n\tctx := newTestContext(\"test-stream-basic\", \"openai.gpt-4o\")\n\n\t// Track streaming chunks\n\tvar toolCallChunks int\n\thandler := func(chunkType message.StreamChunkType, data []byte) int {\n\t\tif chunkType == message.ChunkToolCall {\n\t\t\ttoolCallChunks++\n\t\t}\n\t\tt.Logf(\"Stream chunk [%s]: %s\", chunkType, string(data))\n\t\treturn 0 // Continue\n\t}\n\n\t// Call Stream\n\tresponse, err := llmInstance.Stream(ctx, messages, options, handler)\n\tif err != nil {\n\t\tt.Fatalf(\"Stream with tool calls failed: %v\", err)\n\t}\n\n\t// Validate response\n\tif response == nil {\n\t\tt.Fatal(\"Response is nil\")\n\t}\n\n\t// Should have tool calls\n\tif len(response.ToolCalls) == 0 {\n\t\tt.Error(\"Expected tool calls but got none\")\n\t} else {\n\t\tt.Logf(\"Received %d tool call(s)\", len(response.ToolCalls))\n\t\tfor i, tc := range response.ToolCalls {\n\t\t\tt.Logf(\"Tool call %d: %s(%s)\", i, tc.Function.Name, tc.Function.Arguments)\n\n\t\t\t// Validate tool call has required fields\n\t\t\tif tc.ID == \"\" {\n\t\t\t\tt.Errorf(\"Tool call %d missing ID\", i)\n\t\t\t}\n\t\t\tif tc.Function.Name == \"\" {\n\t\t\t\tt.Errorf(\"Tool call %d missing function name\", i)\n\t\t\t}\n\t\t\tif tc.Function.Arguments == \"\" {\n\t\t\t\tt.Errorf(\"Tool call %d missing arguments\", i)\n\t\t\t}\n\t\t}\n\t}\n\n\tif response.FinishReason != context.FinishReasonToolCalls {\n\t\tt.Logf(\"Warning: Expected finish_reason='tool_calls', got '%s'\", response.FinishReason)\n\t}\n\n\tif toolCallChunks == 0 {\n\t\tt.Error(\"No tool call chunks received during streaming\")\n\t}\n\n\tt.Logf(\"Final response: %+v\", response)\n}\n\n// TestOpenAIPostWithToolCalls tests non-streaming with tool calls\nfunc TestOpenAIPostWithToolCalls(t *testing.T) {\n\ttest.Prepare(t, config.Conf)\n\tdefer test.Clean()\n\n\t// Create connector\n\tconn, err := connector.Select(\"openai.gpt-4o\")\n\tif err != nil {\n\t\tt.Fatalf(\"Failed to select connector: %v\", err)\n\t}\n\n\t// Create LLM instance with tool call capabilities\n\toptions := &context.CompletionOptions{\n\t\tCapabilities: &openai.Capabilities{\n\t\t\tToolCalls: true,\n\t\t},\n\t}\n\n\t// Define a calculation tool\n\tcalcTool := map[string]interface{}{\n\t\t\"type\": \"function\",\n\t\t\"function\": map[string]interface{}{\n\t\t\t\"name\": \"calculate\",\n\t\t\t\"description\": \"Perform a mathematical calculation\",\n\t\t\t\"parameters\": map[string]interface{}{\n\t\t\t\t\"type\": \"object\",\n\t\t\t\t\"properties\": map[string]interface{}{\n\t\t\t\t\t\"expression\": map[string]interface{}{\n\t\t\t\t\t\t\"type\": \"string\",\n\t\t\t\t\t\t\"description\": \"The mathematical expression to evaluate\",\n\t\t\t\t\t},\n\t\t\t\t},\n\t\t\t\t\"required\": []string{\"expression\"},\n\t\t\t},\n\t\t},\n\t}\n\n\toptions.Tools = []map[string]interface{}{calcTool}\n\toptions.ToolChoice = \"auto\"\n\n\tllmInstance, err := llm.New(conn, options)\n\tif err != nil {\n\t\tt.Fatalf(\"Failed to create LLM instance: %v\", err)\n\t}\n\n\t// Prepare messages\n\tmessages := []context.Message{\n\t\t{\n\t\t\tRole: context.RoleUser,\n\t\t\tContent: \"Calculate 15 * 8\",\n\t\t},\n\t}\n\n\t// Create context\n\tctx := newTestContext(\"test-stream-basic\", \"openai.gpt-4o\")\n\n\t// Call Post\n\tresponse, err := llmInstance.Post(ctx, messages, options)\n\tif err != nil {\n\t\tt.Fatalf(\"Post with tool calls failed: %v\", err)\n\t}\n\n\t// Validate response\n\tif response == nil {\n\t\tt.Fatal(\"Response is nil\")\n\t}\n\n\t// Validate response metadata\n\tif response.ID == \"\" {\n\t\tt.Error(\"Response ID is empty\")\n\t}\n\tif response.Model == \"\" {\n\t\tt.Error(\"Response Model is empty\")\n\t}\n\tif response.FinishReason != \"tool_calls\" {\n\t\tt.Errorf(\"FinishReason is %s, expected tool_calls\", response.FinishReason)\n\t}\n\tif response.Usage == nil {\n\t\tt.Error(\"Response Usage is nil\")\n\t} else {\n\t\tif response.Usage.TotalTokens == 0 {\n\t\t\tt.Error(\"Response Usage.TotalTokens is 0\")\n\t\t}\n\t\tt.Logf(\"Usage: prompt=%d, completion=%d, total=%d\",\n\t\t\tresponse.Usage.PromptTokens, response.Usage.CompletionTokens, response.Usage.TotalTokens)\n\t}\n\n\t// Should have tool calls\n\tif len(response.ToolCalls) == 0 {\n\t\tt.Error(\"Expected tool calls but got none\")\n\t} else {\n\t\ttc := response.ToolCalls[0]\n\n\t\t// Validate tool call structure\n\t\tif tc.ID == \"\" {\n\t\t\tt.Error(\"Tool call ID is empty\")\n\t\t}\n\t\tif tc.Type != context.ToolTypeFunction {\n\t\t\tt.Errorf(\"Tool call Type is %s, expected %s\", tc.Type, context.ToolTypeFunction)\n\t\t}\n\t\tif tc.Function.Name != \"calculate\" {\n\t\t\tt.Errorf(\"Tool call function name is %s, expected calculate\", tc.Function.Name)\n\t\t}\n\t\tif tc.Function.Arguments == \"\" {\n\t\t\tt.Error(\"Tool call arguments are empty\")\n\t\t}\n\n\t\tt.Logf(\"Received %d tool call(s)\", len(response.ToolCalls))\n\t\tfor i, tc := range response.ToolCalls {\n\t\t\tt.Logf(\"Tool call %d: %s(%s)\", i, tc.Function.Name, tc.Function.Arguments)\n\t\t}\n\t}\n\n\tt.Logf(\"Response: %+v\", response)\n}\n\n// TestOpenAIStreamWithInvalidToolCall tests that invalid tool calls trigger validation error\nfunc TestOpenAIStreamWithInvalidToolCall(t *testing.T) {\n\ttest.Prepare(t, config.Conf)\n\tdefer test.Clean()\n\n\t// Create connector\n\tconn, err := connector.Select(\"openai.gpt-4o\")\n\tif err != nil {\n\t\tt.Fatalf(\"Failed to select connector: %v\", err)\n\t}\n\n\t// Create LLM instance\n\toptions := &context.CompletionOptions{\n\t\tCapabilities: &openai.Capabilities{\n\t\t\tStreaming: true,\n\t\t\tToolCalls: true,\n\t\t},\n\t}\n\n\t// Define a strict tool that requires specific format\n\tstrictTool := map[string]interface{}{\n\t\t\"type\": \"function\",\n\t\t\"function\": map[string]interface{}{\n\t\t\t\"name\": \"send_email\",\n\t\t\t\"description\": \"Send an email\",\n\t\t\t\"parameters\": map[string]interface{}{\n\t\t\t\t\"type\": \"object\",\n\t\t\t\t\"properties\": map[string]interface{}{\n\t\t\t\t\t\"to\": map[string]interface{}{\n\t\t\t\t\t\t\"type\": \"string\",\n\t\t\t\t\t\t\"pattern\": \"^[a-zA-Z0-9._%+-]+@[a-zA-Z0-9.-]+\\\\.[a-zA-Z]{2,}$\",\n\t\t\t\t\t},\n\t\t\t\t\t\"subject\": map[string]interface{}{\n\t\t\t\t\t\t\"type\": \"string\",\n\t\t\t\t\t\t\"minLength\": 1,\n\t\t\t\t\t},\n\t\t\t\t\t\"body\": map[string]interface{}{\n\t\t\t\t\t\t\"type\": \"string\",\n\t\t\t\t\t\t\"minLength\": 1,\n\t\t\t\t\t},\n\t\t\t\t},\n\t\t\t\t\"required\": []string{\"to\", \"subject\", \"body\"},\n\t\t\t},\n\t\t},\n\t}\n\n\toptions.Tools = []map[string]interface{}{strictTool}\n\toptions.ToolChoice = map[string]interface{}{\n\t\t\"type\": \"function\",\n\t\t\"function\": map[string]interface{}{\n\t\t\t\"name\": \"send_email\",\n\t\t},\n\t}\n\n\tllmInstance, err := llm.New(conn, options)\n\tif err != nil {\n\t\tt.Fatalf(\"Failed to create LLM instance: %v\", err)\n\t}\n\n\t// Prepare messages with incomplete information (should cause validation error)\n\tmessages := []context.Message{\n\t\t{\n\t\t\tRole: context.RoleUser,\n\t\t\tContent: \"Send email to invalid-email without subject\",\n\t\t},\n\t}\n\n\t// Create context\n\tctx := newTestContext(\"test-stream-basic\", \"openai.gpt-4o\")\n\n\thandler := func(chunkType message.StreamChunkType, data []byte) int {\n\t\treturn 0 // Continue\n\t}\n\n\t// Call Stream - should succeed but may trigger validation if tool call is malformed\n\tresponse, err := llmInstance.Stream(ctx, messages, options, handler)\n\n\t// The API might return a valid tool call despite the bad prompt,\n\t// so we just log the result\n\tif err != nil {\n\t\tt.Logf(\"Stream failed as expected with validation error: %v\", err)\n\t} else {\n\t\tt.Logf(\"Stream succeeded, response: %+v\", response)\n\t\tif len(response.ToolCalls) > 0 {\n\t\t\tt.Logf(\"Tool calls: %v\", response.ToolCalls)\n\t\t}\n\t}\n}\n\n// TestOpenAIStreamRetry tests the retry mechanism with invalid API key\nfunc TestOpenAIStreamRetry(t *testing.T) {\n\ttest.Prepare(t, config.Conf)\n\tdefer test.Clean()\n\n\t// Create connector with invalid API key to trigger 401 error (non-retryable)\n\tconnDSL := `{\n\t\t\"type\": \"openai\",\n\t\t\"options\": {\n\t\t\t\"model\": \"gpt-4o\",\n\t\t\t\"key\": \"sk-invalid-key-should-fail-auth\",\n\t\t\t\"host\": \"https://api.openai.com\"\n\t\t}\n\t}`\n\n\tconn, err := connector.New(\"openai\", \"test-retry\", []byte(connDSL))\n\tif err != nil {\n\t\tt.Fatalf(\"Failed to create test connector: %v\", err)\n\t}\n\n\t// Create LLM instance\n\toptions := &context.CompletionOptions{\n\t\tCapabilities: &openai.Capabilities{\n\t\t\tStreaming: true,\n\t\t\tToolCalls: true, // Need this to select OpenAI provider\n\t\t},\n\t}\n\n\tllmInstance, err := llm.New(conn, options)\n\tif err != nil {\n\t\tt.Fatalf(\"Failed to create LLM instance: %v\", err)\n\t}\n\n\tmessages := []context.Message{\n\t\t{\n\t\t\tRole: context.RoleUser,\n\t\t\tContent: \"Test\",\n\t\t},\n\t}\n\n\tctx := newTestContext(\"test-retry\", \"test-retry\")\n\n\t// This should fail quickly without retry (401 is non-retryable)\n\t_, err = llmInstance.Stream(ctx, messages, options, nil)\n\tif err == nil {\n\t\tt.Fatal(\"Expected error due to invalid API key, but got success\")\n\t}\n\n\t// Verify it's an error related to invalid API key\n\t// Could be: 401, unauthorized, authentication error, or no data (empty response)\n\terrMsg := err.Error()\n\thasExpectedError := strings.Contains(strings.ToLower(errMsg), \"401\") ||\n\t\tstrings.Contains(strings.ToLower(errMsg), \"unauthorized\") ||\n\t\tstrings.Contains(strings.ToLower(errMsg), \"authentication\") ||\n\t\tstrings.Contains(strings.ToLower(errMsg), \"incorrect api key\") ||\n\t\tstrings.Contains(strings.ToLower(errMsg), \"no data received\")\n\n\tif !hasExpectedError {\n\t\tt.Errorf(\"Expected authentication or empty response error, got: %v\", err)\n\t}\n\n\t// Should mention non-retryable (these errors should not trigger retry)\n\tif !strings.Contains(strings.ToLower(errMsg), \"non-retryable\") {\n\t\tt.Errorf(\"Error should indicate non-retryable: %v\", err)\n\t}\n\n\tt.Logf(\"Failed as expected with error: %v\", err)\n}\n\n// TestOpenAIStreamChunkTypes tests that stream handler receives correct chunk types\nfunc TestOpenAIStreamChunkTypes(t *testing.T) {\n\ttest.Prepare(t, config.Conf)\n\tdefer test.Clean()\n\n\tconn, err := connector.Select(\"openai.gpt-4o\")\n\tif err != nil {\n\t\tt.Fatalf(\"Failed to select connector: %v\", err)\n\t}\n\n\toptions := &context.CompletionOptions{\n\t\tCapabilities: &openai.Capabilities{\n\t\t\tStreaming: true,\n\t\t\tToolCalls: true,\n\t\t},\n\t}\n\n\tllmInstance, err := llm.New(conn, options)\n\tif err != nil {\n\t\tt.Fatalf(\"Failed to create LLM instance: %v\", err)\n\t}\n\n\tmessages := []context.Message{\n\t\t{\n\t\t\tRole: context.RoleUser,\n\t\t\tContent: \"Say 'test' in one word.\",\n\t\t},\n\t}\n\n\tctx := newTestContext(\"test-chunk-types\", \"openai.gpt-4o\")\n\n\t// Track chunk types\n\tchunkTypes := make(map[message.StreamChunkType]int)\n\thandler := func(chunkType message.StreamChunkType, data []byte) int {\n\t\tchunkTypes[chunkType]++\n\t\tt.Logf(\"Received chunk type: %s, data length: %d\", chunkType, len(data))\n\t\treturn 1 // Continue\n\t}\n\n\tresponse, err := llmInstance.Stream(ctx, messages, options, handler)\n\tif err != nil {\n\t\tt.Fatalf(\"Stream failed: %v\", err)\n\t}\n\n\tif response == nil {\n\t\tt.Fatal(\"Response is nil\")\n\t}\n\n\t// Validate chunk types received\n\tif chunkTypes[message.ChunkText] == 0 {\n\t\tt.Error(\"Expected to receive ChunkText, but got 0\")\n\t}\n\n\tt.Logf(\"Chunk types received: %+v\", chunkTypes)\n}\n\n// TestOpenAIStreamErrorCallback tests that errors are sent to stream handler\nfunc TestOpenAIStreamErrorCallback(t *testing.T) {\n\ttest.Prepare(t, config.Conf)\n\tdefer test.Clean()\n\n\t// Create connector with invalid API key to trigger error\n\tconnDSL := `{\n\t\t\"type\": \"openai\",\n\t\t\"options\": {\n\t\t\t\"model\": \"gpt-4o\",\n\t\t\t\"key\": \"sk-invalid-for-error-test\",\n\t\t\t\"host\": \"https://api.openai.com\"\n\t\t}\n\t}`\n\n\tconn, err := connector.New(\"openai\", \"test-error-callback\", []byte(connDSL))\n\tif err != nil {\n\t\tt.Fatalf(\"Failed to create test connector: %v\", err)\n\t}\n\n\toptions := &context.CompletionOptions{\n\t\tCapabilities: &openai.Capabilities{\n\t\t\tStreaming: true,\n\t\t\tToolCalls: true,\n\t\t},\n\t}\n\n\tllmInstance, err := llm.New(conn, options)\n\tif err != nil {\n\t\tt.Fatalf(\"Failed to create LLM instance: %v\", err)\n\t}\n\n\tmessages := []context.Message{\n\t\t{\n\t\t\tRole: context.RoleUser,\n\t\t\tContent: \"Test\",\n\t\t},\n\t}\n\n\tctx := newTestContext(\"test-error-callback\", \"test-error-callback\")\n\n\t// Track if error chunk was received\n\treceivedError := false\n\tvar errorMessage string\n\thandler := func(chunkType message.StreamChunkType, data []byte) int {\n\t\tif chunkType == message.ChunkError {\n\t\t\treceivedError = true\n\t\t\terrorMessage = string(data)\n\t\t\tt.Logf(\"Received error chunk: %s\", errorMessage)\n\t\t}\n\t\treturn 1 // Continue\n\t}\n\n\t// This should fail and send error to handler\n\t_, err = llmInstance.Stream(ctx, messages, options, handler)\n\tif err == nil {\n\t\tt.Fatal(\"Expected error due to invalid API key\")\n\t}\n\n\t// Verify error was sent to handler\n\tif !receivedError {\n\t\tt.Error(\"Expected to receive ChunkError in handler, but didn't\")\n\t}\n\n\tif errorMessage == \"\" {\n\t\tt.Error(\"Error message in chunk is empty\")\n\t}\n\n\tt.Logf(\"Error callback test passed. Error: %v\", err)\n}\n\n// TestOpenAIToolCallValidationRetry tests automatic tool call validation retry with LLM feedback\nfunc TestOpenAIToolCallValidationRetry(t *testing.T) {\n\ttest.Prepare(t, config.Conf)\n\tdefer test.Clean()\n\n\tconn, err := connector.Select(\"openai.gpt-4o\")\n\tif err != nil {\n\t\tt.Fatalf(\"Failed to select connector: %v\", err)\n\t}\n\n\toptions := &context.CompletionOptions{\n\t\tCapabilities: &openai.Capabilities{\n\t\t\tStreaming: true,\n\t\t\tToolCalls: true,\n\t\t},\n\t\tTools: []map[string]interface{}{\n\t\t\t{\n\t\t\t\t\"type\": \"function\",\n\t\t\t\t\"function\": map[string]interface{}{\n\t\t\t\t\t\"name\": \"test_strict_validation\",\n\t\t\t\t\t\"description\": \"A function with very strict validation rules\",\n\t\t\t\t\t\"parameters\": map[string]interface{}{\n\t\t\t\t\t\t\"type\": \"object\",\n\t\t\t\t\t\t\"properties\": map[string]interface{}{\n\t\t\t\t\t\t\t\"status\": map[string]interface{}{\n\t\t\t\t\t\t\t\t\"type\": \"string\",\n\t\t\t\t\t\t\t\t\"description\": \"Must be exactly 'active' or 'inactive'\",\n\t\t\t\t\t\t\t\t\"enum\": []string{\"active\", \"inactive\"},\n\t\t\t\t\t\t\t},\n\t\t\t\t\t\t\t\"priority\": map[string]interface{}{\n\t\t\t\t\t\t\t\t\"type\": \"integer\",\n\t\t\t\t\t\t\t\t\"description\": \"Must be between 1 and 5\",\n\t\t\t\t\t\t\t\t\"minimum\": 1,\n\t\t\t\t\t\t\t\t\"maximum\": 5,\n\t\t\t\t\t\t\t},\n\t\t\t\t\t\t},\n\t\t\t\t\t\t\"required\": []string{\"status\", \"priority\"},\n\t\t\t\t\t},\n\t\t\t\t},\n\t\t\t},\n\t\t},\n\t}\n\n\tllmInstance, err := llm.New(conn, options)\n\tif err != nil {\n\t\tt.Fatalf(\"Failed to create LLM instance: %v\", err)\n\t}\n\n\tctx := newTestContext(\"test-tool-validation-retry\", \"openai.gpt-4o\")\n\n\t// Try to make LLM call with intentionally unclear requirements\n\t// This may or may not trigger validation, depending on LLM behavior\n\tmessages := []context.Message{\n\t\t{\n\t\t\tRole: context.RoleUser,\n\t\t\tContent: \"Call test_strict_validation function with status='pending' and priority=10\",\n\t\t},\n\t}\n\n\t// The Provider will automatically:\n\t// 1. Call LLM\n\t// 2. If validation fails, add error feedback to conversation\n\t// 3. Retry up to 3 times with feedback\n\t// 4. Return success or validation error after max retries\n\tresponse, err := llmInstance.Stream(ctx, messages, options, nil)\n\n\tif err != nil {\n\t\t// Check if it's a validation error after retries\n\t\tif strings.Contains(err.Error(), \"tool call validation failed after\") &&\n\t\t\tstrings.Contains(err.Error(), \"retries\") {\n\t\t\tt.Logf(\"✓ Automatic validation retry exhausted: %v\", err)\n\t\t} else if strings.Contains(err.Error(), \"validation\") {\n\t\t\tt.Logf(\"✓ Validation failed: %v\", err)\n\t\t} else {\n\t\t\tt.Logf(\"Request failed (non-validation): %v\", err)\n\t\t}\n\t} else if response != nil {\n\t\tif len(response.ToolCalls) > 0 {\n\t\t\tt.Logf(\"✓ Tool call succeeded (possibly after auto-retry): %+v\", response.ToolCalls[0])\n\n\t\t\t// Verify the tool call arguments are valid\n\t\t\ttc := response.ToolCalls[0]\n\t\t\tvar args map[string]interface{}\n\t\t\tif err := json.Unmarshal([]byte(tc.Function.Arguments), &args); err == nil {\n\t\t\t\tif status, ok := args[\"status\"].(string); ok {\n\t\t\t\t\tif status != \"active\" && status != \"inactive\" {\n\t\t\t\t\t\tt.Errorf(\"Status should be 'active' or 'inactive', got: %s\", status)\n\t\t\t\t\t}\n\t\t\t\t}\n\t\t\t\tif priority, ok := args[\"priority\"].(float64); ok {\n\t\t\t\t\tif priority < 1 || priority > 5 {\n\t\t\t\t\t\tt.Errorf(\"Priority should be between 1-5, got: %v\", priority)\n\t\t\t\t\t}\n\t\t\t\t}\n\t\t\t}\n\t\t} else {\n\t\t\tt.Log(\"✓ Response returned but no tool calls\")\n\t\t}\n\t}\n\n\tt.Log(\"Automatic tool call validation retry test completed\")\n}\n\n// TestOpenAIJSONMode tests JSON mode response formatting\nfunc TestOpenAIJSONMode(t *testing.T) {\n\ttest.Prepare(t, config.Conf)\n\tdefer test.Clean()\n\n\tconn, err := connector.Select(\"openai.gpt-4o\")\n\tif err != nil {\n\t\tt.Fatalf(\"Failed to select connector: %v\", err)\n\t}\n\n\toptions := &context.CompletionOptions{\n\t\tCapabilities: &openai.Capabilities{\n\t\t\tStreaming: true,\n\t\t\tToolCalls: true,\n\t\t},\n\t\tResponseFormat: &context.ResponseFormat{\n\t\t\tType: context.ResponseFormatJSON,\n\t\t},\n\t}\n\n\tllmInstance, err := llm.New(conn, options)\n\tif err != nil {\n\t\tt.Fatalf(\"Failed to create LLM instance: %v\", err)\n\t}\n\n\tmessages := []context.Message{\n\t\t{\n\t\t\tRole: context.RoleUser,\n\t\t\tContent: \"Generate a JSON object with fields: name (string), age (number), city (string). Use values: John, 30, New York\",\n\t\t},\n\t}\n\n\tctx := newTestContext(\"test-json-mode\", \"openai.gpt-4o\")\n\n\t// Test streaming with JSON mode\n\tresponse, err := llmInstance.Stream(ctx, messages, options, nil)\n\tif err != nil {\n\t\tt.Fatalf(\"Stream with JSON mode failed: %v\", err)\n\t}\n\n\tif response == nil {\n\t\tt.Fatal(\"Response is nil\")\n\t}\n\n\t// Validate response\n\tcontentStr, ok := response.Content.(string)\n\tif !ok || contentStr == \"\" {\n\t\tt.Error(\"Response content is empty or not a string\")\n\t}\n\n\t// Try to parse as JSON\n\tvar jsonData map[string]interface{}\n\tif err := json.Unmarshal([]byte(contentStr), &jsonData); err != nil {\n\t\tt.Errorf(\"Response is not valid JSON: %v\\nContent: %s\", err, contentStr)\n\t} else {\n\t\tt.Logf(\"✓ Response is valid JSON: %+v\", jsonData)\n\n\t\t// Verify expected fields exist\n\t\tif _, hasName := jsonData[\"name\"]; !hasName {\n\t\t\tt.Error(\"JSON response missing 'name' field\")\n\t\t}\n\t\tif _, hasAge := jsonData[\"age\"]; !hasAge {\n\t\t\tt.Error(\"JSON response missing 'age' field\")\n\t\t}\n\t\tif _, hasCity := jsonData[\"city\"]; !hasCity {\n\t\t\tt.Error(\"JSON response missing 'city' field\")\n\t\t}\n\t}\n\n\t// Validate metadata\n\tif response.ID == \"\" {\n\t\tt.Error(\"Response ID is empty\")\n\t}\n\tif response.Model == \"\" {\n\t\tt.Error(\"Response Model is empty\")\n\t}\n\tif response.Usage == nil {\n\t\tt.Error(\"Response Usage is nil\")\n\t} else {\n\t\tt.Logf(\"Usage: prompt=%d, completion=%d, total=%d\",\n\t\t\tresponse.Usage.PromptTokens, response.Usage.CompletionTokens, response.Usage.TotalTokens)\n\t}\n\n\tt.Log(\"JSON mode test completed successfully\")\n}\n\n// TestOpenAIJSONModePost tests JSON mode with non-streaming\nfunc TestOpenAIJSONModePost(t *testing.T) {\n\ttest.Prepare(t, config.Conf)\n\tdefer test.Clean()\n\n\tconn, err := connector.Select(\"openai.gpt-4o\")\n\tif err != nil {\n\t\tt.Fatalf(\"Failed to select connector: %v\", err)\n\t}\n\n\toptions := &context.CompletionOptions{\n\t\tCapabilities: &openai.Capabilities{\n\t\t\tToolCalls: true,\n\t\t},\n\t\tResponseFormat: &context.ResponseFormat{\n\t\t\tType: context.ResponseFormatJSON,\n\t\t},\n\t}\n\n\tllmInstance, err := llm.New(conn, options)\n\tif err != nil {\n\t\tt.Fatalf(\"Failed to create LLM instance: %v\", err)\n\t}\n\n\tmessages := []context.Message{\n\t\t{\n\t\t\tRole: context.RoleUser,\n\t\t\tContent: \"Return a JSON with: status='success', count=42\",\n\t\t},\n\t}\n\n\tctx := newTestContext(\"test-json-mode-post\", \"openai.gpt-4o\")\n\n\t// Test non-streaming with JSON mode\n\tresponse, err := llmInstance.Post(ctx, messages, options)\n\tif err != nil {\n\t\tt.Fatalf(\"Post with JSON mode failed: %v\", err)\n\t}\n\n\tif response == nil {\n\t\tt.Fatal(\"Response is nil\")\n\t}\n\n\t// Validate response content is JSON\n\tcontentStr, ok := response.Content.(string)\n\tif !ok || contentStr == \"\" {\n\t\tt.Error(\"Response content is empty or not a string\")\n\t}\n\n\tvar jsonData map[string]interface{}\n\tif err := json.Unmarshal([]byte(contentStr), &jsonData); err != nil {\n\t\tt.Errorf(\"Response is not valid JSON: %v\\nContent: %s\", err, contentStr)\n\t} else {\n\t\tt.Logf(\"✓ Response is valid JSON: %+v\", jsonData)\n\t}\n\n\t// Validate metadata\n\tif response.Usage == nil {\n\t\tt.Error(\"Response Usage is nil\")\n\t} else {\n\t\tif response.Usage.TotalTokens == 0 {\n\t\t\tt.Error(\"Response Usage.TotalTokens is 0\")\n\t\t}\n\t\tt.Logf(\"Usage: prompt=%d, completion=%d, total=%d\",\n\t\t\tresponse.Usage.PromptTokens, response.Usage.CompletionTokens, response.Usage.TotalTokens)\n\t}\n\n\tt.Log(\"JSON mode Post test completed successfully\")\n}\n\n// TestOpenAIJSONSchema tests JSON mode with strict schema validation\nfunc TestOpenAIJSONSchema(t *testing.T) {\n\ttest.Prepare(t, config.Conf)\n\tdefer test.Clean()\n\n\tconn, err := connector.Select(\"openai.gpt-4o\")\n\tif err != nil {\n\t\tt.Fatalf(\"Failed to select connector: %v\", err)\n\t}\n\n\t// Define a strict JSON schema\n\t// Note: For OpenAI strict mode, 'required' must include ALL properties\n\tschema := map[string]interface{}{\n\t\t\"type\": \"object\",\n\t\t\"properties\": map[string]interface{}{\n\t\t\t\"user\": map[string]interface{}{\n\t\t\t\t\"type\": \"object\",\n\t\t\t\t\"properties\": map[string]interface{}{\n\t\t\t\t\t\"name\": map[string]interface{}{\n\t\t\t\t\t\t\"type\": \"string\",\n\t\t\t\t\t\t\"description\": \"User's full name\",\n\t\t\t\t\t},\n\t\t\t\t\t\"email\": map[string]interface{}{\n\t\t\t\t\t\t\"type\": \"string\",\n\t\t\t\t\t\t\"description\": \"User's email address\",\n\t\t\t\t\t},\n\t\t\t\t\t\"age\": map[string]interface{}{\n\t\t\t\t\t\t\"type\": \"integer\",\n\t\t\t\t\t\t\"description\": \"User's age\",\n\t\t\t\t\t},\n\t\t\t\t\t\"isActive\": map[string]interface{}{\n\t\t\t\t\t\t\"type\": \"boolean\",\n\t\t\t\t\t\t\"description\": \"Whether user is active\",\n\t\t\t\t\t},\n\t\t\t\t},\n\t\t\t\t\"required\": []string{\"name\", \"email\", \"age\", \"isActive\"},\n\t\t\t\t\"additionalProperties\": false,\n\t\t\t},\n\t\t},\n\t\t\"required\": []string{\"user\"},\n\t\t\"additionalProperties\": false,\n\t}\n\n\toptions := &context.CompletionOptions{\n\t\tCapabilities: &openai.Capabilities{\n\t\t\tStreaming: true,\n\t\t\tToolCalls: true,\n\t\t},\n\t\tResponseFormat: &context.ResponseFormat{\n\t\t\tType: context.ResponseFormatJSONSchema,\n\t\t\tJSONSchema: &context.JSONSchema{\n\t\t\t\tName: \"user_info\",\n\t\t\t\tDescription: \"User information schema\",\n\t\t\t\tSchema: schema,\n\t\t\t\tStrict: func() *bool { v := true; return &v }(),\n\t\t\t},\n\t\t},\n\t}\n\n\tllmInstance, err := llm.New(conn, options)\n\tif err != nil {\n\t\tt.Fatalf(\"Failed to create LLM instance: %v\", err)\n\t}\n\n\tmessages := []context.Message{\n\t\t{\n\t\t\tRole: context.RoleUser,\n\t\t\tContent: \"Create user info for: Alice Smith, alice@example.com, age 28, active user\",\n\t\t},\n\t}\n\n\tctx := newTestContext(\"test-json-schema\", \"openai.gpt-4o\")\n\n\t// Test streaming with JSON schema\n\tresponse, err := llmInstance.Stream(ctx, messages, options, nil)\n\tif err != nil {\n\t\tt.Fatalf(\"Stream with JSON schema failed: %v\", err)\n\t}\n\n\tif response == nil {\n\t\tt.Fatal(\"Response is nil\")\n\t}\n\n\t// Validate response content\n\tcontentStr, ok := response.Content.(string)\n\tif !ok || contentStr == \"\" {\n\t\tt.Fatal(\"Response content is empty or not a string\")\n\t}\n\n\t// Parse as JSON\n\tvar jsonData map[string]interface{}\n\tif err := json.Unmarshal([]byte(contentStr), &jsonData); err != nil {\n\t\tt.Fatalf(\"Response is not valid JSON: %v\\nContent: %s\", err, contentStr)\n\t}\n\n\tt.Logf(\"✓ Response is valid JSON: %+v\", jsonData)\n\n\t// Verify structure matches schema\n\tuser, hasUser := jsonData[\"user\"].(map[string]interface{})\n\tif !hasUser {\n\t\tt.Fatal(\"JSON response missing 'user' object\")\n\t}\n\n\t// Verify required fields\n\tif _, hasName := user[\"name\"]; !hasName {\n\t\tt.Error(\"User object missing required 'name' field\")\n\t}\n\tif _, hasEmail := user[\"email\"]; !hasEmail {\n\t\tt.Error(\"User object missing required 'email' field\")\n\t}\n\n\t// Verify field types\n\tif name, ok := user[\"name\"].(string); ok {\n\t\tt.Logf(\"✓ name: %s (string)\", name)\n\t} else {\n\t\tt.Error(\"name is not a string\")\n\t}\n\n\tif email, ok := user[\"email\"].(string); ok {\n\t\tt.Logf(\"✓ email: %s (string)\", email)\n\t} else {\n\t\tt.Error(\"email is not a string\")\n\t}\n\n\tif age, ok := user[\"age\"].(float64); ok {\n\t\tif age < 0 || age > 150 {\n\t\t\tt.Errorf(\"age %v is out of range [0, 150]\", age)\n\t\t}\n\t\tt.Logf(\"✓ age: %v (integer, in range)\", age)\n\t}\n\n\tif isActive, ok := user[\"isActive\"].(bool); ok {\n\t\tt.Logf(\"✓ isActive: %v (boolean)\", isActive)\n\t}\n\n\t// Validate metadata\n\tif response.Usage == nil {\n\t\tt.Error(\"Response Usage is nil\")\n\t} else {\n\t\tt.Logf(\"Usage: prompt=%d, completion=%d, total=%d\",\n\t\t\tresponse.Usage.PromptTokens, response.Usage.CompletionTokens, response.Usage.TotalTokens)\n\t}\n\n\tt.Log(\"JSON schema test completed successfully\")\n}\n\n// TestOpenAIJSONSchemaPost tests JSON schema with non-streaming\nfunc TestOpenAIJSONSchemaPost(t *testing.T) {\n\ttest.Prepare(t, config.Conf)\n\tdefer test.Clean()\n\n\tconn, err := connector.Select(\"openai.gpt-4o\")\n\tif err != nil {\n\t\tt.Fatalf(\"Failed to select connector: %v\", err)\n\t}\n\n\t// Simple schema for testing\n\t// Note: For OpenAI strict mode, 'required' must include ALL properties\n\tschema := map[string]interface{}{\n\t\t\"type\": \"object\",\n\t\t\"properties\": map[string]interface{}{\n\t\t\t\"status\": map[string]interface{}{\n\t\t\t\t\"type\": \"string\",\n\t\t\t\t\"enum\": []string{\"success\", \"error\", \"pending\"},\n\t\t\t},\n\t\t\t\"message\": map[string]interface{}{\n\t\t\t\t\"type\": \"string\",\n\t\t\t},\n\t\t\t\"code\": map[string]interface{}{\n\t\t\t\t\"type\": \"integer\",\n\t\t\t},\n\t\t},\n\t\t\"required\": []string{\"status\", \"message\", \"code\"},\n\t\t\"additionalProperties\": false,\n\t}\n\n\toptions := &context.CompletionOptions{\n\t\tCapabilities: &openai.Capabilities{\n\t\t\tToolCalls: true,\n\t\t},\n\t\tResponseFormat: &context.ResponseFormat{\n\t\t\tType: context.ResponseFormatJSONSchema,\n\t\t\tJSONSchema: &context.JSONSchema{\n\t\t\t\tName: \"api_response\",\n\t\t\t\tDescription: \"API response format\",\n\t\t\t\tSchema: schema,\n\t\t\t\tStrict: func() *bool { v := true; return &v }(),\n\t\t\t},\n\t\t},\n\t}\n\n\tllmInstance, err := llm.New(conn, options)\n\tif err != nil {\n\t\tt.Fatalf(\"Failed to create LLM instance: %v\", err)\n\t}\n\n\tmessages := []context.Message{\n\t\t{\n\t\t\tRole: context.RoleUser,\n\t\t\tContent: \"Generate an API response with status 'success', message 'Operation completed', and code 200\",\n\t\t},\n\t}\n\n\tctx := newTestContext(\"test-json-schema-post\", \"openai.gpt-4o\")\n\n\t// Test non-streaming with JSON schema\n\tresponse, err := llmInstance.Post(ctx, messages, options)\n\tif err != nil {\n\t\tt.Fatalf(\"Post with JSON schema failed: %v\", err)\n\t}\n\n\tif response == nil {\n\t\tt.Fatal(\"Response is nil\")\n\t}\n\n\t// Validate response content\n\tcontentStr, ok := response.Content.(string)\n\tif !ok || contentStr == \"\" {\n\t\tt.Fatal(\"Response content is empty or not a string\")\n\t}\n\n\t// Parse and validate JSON\n\tvar jsonData map[string]interface{}\n\tif err := json.Unmarshal([]byte(contentStr), &jsonData); err != nil {\n\t\tt.Fatalf(\"Response is not valid JSON: %v\\nContent: %s\", err, contentStr)\n\t}\n\n\tt.Logf(\"✓ Response is valid JSON: %+v\", jsonData)\n\n\t// Verify required fields\n\tstatus, hasStatus := jsonData[\"status\"].(string)\n\tif !hasStatus {\n\t\tt.Fatal(\"Missing required 'status' field\")\n\t}\n\n\t// Verify enum constraint\n\tvalidStatuses := map[string]bool{\"success\": true, \"error\": true, \"pending\": true}\n\tif !validStatuses[status] {\n\t\tt.Errorf(\"status '%s' is not in enum [success, error, pending]\", status)\n\t}\n\n\tif _, hasMessage := jsonData[\"message\"].(string); !hasMessage {\n\t\tt.Error(\"Missing required 'message' field\")\n\t}\n\n\t// Validate metadata\n\tif response.Usage == nil {\n\t\tt.Error(\"Response Usage is nil\")\n\t} else {\n\t\tt.Logf(\"Usage: prompt=%d, completion=%d, total=%d\",\n\t\t\tresponse.Usage.PromptTokens, response.Usage.CompletionTokens, response.Usage.TotalTokens)\n\t}\n\n\tt.Log(\"JSON schema Post test completed successfully\")\n}\n\n// TestOpenAIProxySupport tests that HTTP proxy configuration is respected\nfunc TestOpenAIProxySupport(t *testing.T) {\n\ttest.Prepare(t, config.Conf)\n\tdefer test.Clean()\n\n\t// This test verifies proxy support exists in the connector configuration\n\t// Actual proxy testing requires a real proxy server setup\n\n\tconn, err := connector.Select(\"openai.gpt-4o\")\n\tif err != nil {\n\t\tt.Fatalf(\"Failed to select connector: %v\", err)\n\t}\n\n\tsettings := conn.Setting()\n\tt.Logf(\"Connector settings: %+v\", settings)\n\n\t// Verify host field exists in settings (host is the API endpoint)\n\tif host, hasHost := settings[\"host\"]; hasHost {\n\t\tt.Logf(\"API host configured: %v\", host)\n\t} else {\n\t\tt.Log(\"Host field not in settings (will use default)\")\n\t}\n\n\t// The actual HTTP proxy functionality is implemented via environment variables\n\t// (HTTP_PROXY, HTTPS_PROXY, NO_PROXY) and handled by http.GetTransport\n\tt.Log(\"HTTP proxy support is implemented via http.GetTransport using environment variables\")\n}\n\n// TestOpenAIStreamLifecycleEvents tests that LLM-level lifecycle events are sent correctly\n// LLM layer sends group_start/end for individual messages (thinking, text, tool_call)\n// Note: stream_start/end and Agent-level blocks are handled at Agent level\nfunc TestOpenAIStreamLifecycleEvents(t *testing.T) {\n\ttest.Prepare(t, config.Conf)\n\tdefer test.Clean()\n\n\tconn, err := connector.Select(\"openai.gpt-4o\")\n\tif err != nil {\n\t\tt.Fatalf(\"Failed to select connector: %v\", err)\n\t}\n\n\toptions := &context.CompletionOptions{\n\t\tCapabilities: &openai.Capabilities{\n\t\t\tStreaming: true,\n\t\t\tToolCalls: true,\n\t\t},\n\t}\n\n\tllmInstance, err := llm.New(conn, options)\n\tif err != nil {\n\t\tt.Fatalf(\"Failed to create LLM instance: %v\", err)\n\t}\n\n\tmessages := []context.Message{\n\t\t{\n\t\t\tRole: context.RoleUser,\n\t\t\tContent: \"Say 'hello' in one word\",\n\t\t},\n\t}\n\n\tctx := newTestContext(\"test-lifecycle\", \"openai.gpt-4o\")\n\n\t// Track lifecycle events (group_start/end at LLM layer represent message boundaries)\n\tvar events []string\n\tvar groupStartReceived, groupEndReceived bool\n\n\thandler := func(chunkType message.StreamChunkType, data []byte) int {\n\t\tevents = append(events, string(chunkType))\n\n\t\tswitch chunkType {\n\t\tcase message.ChunkStreamStart:\n\t\t\tt.Error(\"❌ LLM layer should NOT send stream_start (now sent at Agent level)\")\n\n\t\tcase message.ChunkStreamEnd:\n\t\t\tt.Error(\"❌ LLM layer should NOT send stream_end (now sent at Agent level)\")\n\n\t\tcase message.ChunkMessageStart:\n\t\t\tgroupStartReceived = true\n\t\t\tvar startData message.EventMessageStartData\n\t\t\tif err := json.Unmarshal(data, &startData); err == nil {\n\t\t\t\tt.Logf(\"✓ group_start (message start): type=%s, id=%s\", startData.Type, startData.MessageID)\n\t\t\t\tif startData.MessageID == \"\" {\n\t\t\t\t\tt.Error(\"group_start missing message_id\")\n\t\t\t\t}\n\t\t\t} else {\n\t\t\t\tt.Errorf(\"Failed to parse group_start data: %v\", err)\n\t\t\t}\n\n\t\tcase message.ChunkMessageEnd:\n\t\t\tgroupEndReceived = true\n\t\t\tvar endData message.EventMessageEndData\n\t\t\tif err := json.Unmarshal(data, &endData); err == nil {\n\t\t\t\tt.Logf(\"✓ group_end (message end): type=%s, chunks=%d, duration=%dms\",\n\t\t\t\t\tendData.Type, endData.ChunkCount, endData.DurationMs)\n\t\t\t\tif endData.ChunkCount <= 0 {\n\t\t\t\t\tt.Error(\"group_end should have chunk_count > 0\")\n\t\t\t\t}\n\t\t\t} else {\n\t\t\t\tt.Errorf(\"Failed to parse group_end data: %v\", err)\n\t\t\t}\n\n\t\tcase message.ChunkText:\n\t\t\tt.Logf(\" text chunk: %s\", string(data))\n\t\t}\n\n\t\treturn 0\n\t}\n\n\tresponse, err := llmInstance.Stream(ctx, messages, options, handler)\n\tif err != nil {\n\t\tt.Fatalf(\"Stream failed: %v\", err)\n\t}\n\n\tif response == nil {\n\t\tt.Fatal(\"Response is nil\")\n\t}\n\n\t// Validate that LLM-level message lifecycle events were received\n\tif !groupStartReceived {\n\t\tt.Error(\"group_start (message start) event was not received\")\n\t}\n\tif !groupEndReceived {\n\t\tt.Error(\"group_end (message end) event was not received\")\n\t}\n\n\t// Validate event order: group_start should come before group_end\n\tif len(events) < 2 {\n\t\tt.Errorf(\"Expected at least 2 events (message start/end), got %d\", len(events))\n\t}\n\n\tt.Logf(\"Total events received: %d\", len(events))\n\tt.Log(\"LLM message lifecycle events test completed successfully\")\n\tt.Log(\"Note: LLM layer group_start/end represent message boundaries (thinking, text, tool_call)\")\n\tt.Log(\" Agent-level block boundaries and stream_start/end are handled at Agent level\")\n}\n\n// TestOpenAIStreamContextCancellation tests that stream respects context cancellation\nfunc TestOpenAIStreamContextCancellation(t *testing.T) {\n\ttest.Prepare(t, config.Conf)\n\tdefer test.Clean()\n\n\tconn, err := connector.Select(\"openai.gpt-4o\")\n\tif err != nil {\n\t\tt.Fatalf(\"Failed to select connector: %v\", err)\n\t}\n\n\toptions := &context.CompletionOptions{\n\t\tCapabilities: &openai.Capabilities{\n\t\t\tStreaming: true,\n\t\t\tToolCalls: true,\n\t\t},\n\t}\n\n\tllmInstance, err := llm.New(conn, options)\n\tif err != nil {\n\t\tt.Fatalf(\"Failed to create LLM instance: %v\", err)\n\t}\n\n\tmessages := []context.Message{\n\t\t{\n\t\t\tRole: context.RoleUser,\n\t\t\tContent: \"Write a very long essay about the history of computing\", // Long task\n\t\t},\n\t}\n\n\t// Create a context with a very short timeout\n\tctx := newTestContext(\"test-cancel\", \"openai.gpt-4o\")\n\tgoCtx, cancel := gocontext.WithTimeout(gocontext.Background(), 100*time.Millisecond)\n\tdefer cancel()\n\tctx.Context = goCtx\n\n\tvar receivedChunks int\n\n\thandler := func(chunkType message.StreamChunkType, data []byte) int {\n\t\tif chunkType == message.ChunkText || chunkType == message.ChunkToolCall {\n\t\t\treceivedChunks++\n\t\t}\n\t\t// Note: stream_end is now sent at Agent level, not LLM level\n\t\tif chunkType == message.ChunkStreamEnd {\n\t\t\tt.Error(\"❌ LLM layer should NOT send stream_end (now sent at Agent level)\")\n\t\t}\n\t\treturn 0\n\t}\n\n\tresponse, err := llmInstance.Stream(ctx, messages, options, handler)\n\n\t// Should get an error due to context cancellation\n\tif err == nil {\n\t\tt.Error(\"Expected error due to context cancellation, but got nil\")\n\t} else {\n\t\tt.Logf(\"✓ Got expected cancellation error: %v\", err)\n\n\t\t// Check if error message indicates cancellation\n\t\terrStr := err.Error()\n\t\tif !strings.Contains(errStr, \"context\") && !strings.Contains(errStr, \"cancel\") {\n\t\t\tt.Errorf(\"Error should mention context/cancellation: %v\", err)\n\t\t}\n\t}\n\n\t// Response should be nil due to cancellation\n\tif response != nil {\n\t\tt.Logf(\"Warning: Response is not nil despite cancellation (partial response)\")\n\t}\n\n\tt.Logf(\"Received %d chunks before cancellation\", receivedChunks)\n\tt.Log(\"Context cancellation test completed successfully\")\n\tt.Log(\"Note: stream_end for cancellation is now sent at Agent level\")\n}\n\n// TestOpenAIStreamWithTemperature tests different temperature settings\nfunc TestOpenAIStreamWithTemperature(t *testing.T) {\n\ttest.Prepare(t, config.Conf)\n\tdefer test.Clean()\n\n\tconn, err := connector.Select(\"openai.gpt-4o\")\n\tif err != nil {\n\t\tt.Fatalf(\"Failed to select connector: %v\", err)\n\t}\n\n\ttemperature := 0.7 // Moderate temperature\n\n\toptions := &context.CompletionOptions{\n\t\tCapabilities: &openai.Capabilities{\n\t\t\tStreaming: true,\n\t\t\tToolCalls: true, // Need this to select OpenAI provider\n\t\t},\n\t\tTemperature: &temperature,\n\t}\n\n\tllmInstance, err := llm.New(conn, options)\n\tif err != nil {\n\t\tt.Fatalf(\"Failed to create LLM instance: %v\", err)\n\t}\n\n\tmessages := []context.Message{\n\t\t{\n\t\t\tRole: context.RoleUser,\n\t\t\tContent: \"Say 'yes' in one word.\",\n\t\t},\n\t}\n\n\tctx := newTestContext(\"test-temperature\", \"openai.gpt-4o\")\n\n\t// Use callback to collect chunks\n\tchunkCount := 0\n\tvar callback message.StreamFunc = func(chunkType message.StreamChunkType, data []byte) int {\n\t\tchunkCount++\n\t\treturn 1 // Continue\n\t}\n\n\tresponse, err := llmInstance.Stream(ctx, messages, options, callback)\n\tif err != nil {\n\t\tt.Fatalf(\"Stream failed: %v\", err)\n\t}\n\n\tif response == nil {\n\t\tt.Fatal(\"Response is nil\")\n\t}\n\n\t// Validate response data\n\tif response.ID == \"\" {\n\t\tt.Error(\"Response ID is empty\")\n\t}\n\tif response.Model == \"\" {\n\t\tt.Error(\"Response Model is empty\")\n\t}\n\tif response.Content == \"\" {\n\t\tt.Error(\"Response Content is empty\")\n\t}\n\tif response.FinishReason == \"\" {\n\t\tt.Error(\"Response FinishReason is empty\")\n\t}\n\tif response.Usage == nil {\n\t\tt.Error(\"Response Usage is nil\")\n\t} else {\n\t\tif response.Usage.TotalTokens == 0 {\n\t\t\tt.Error(\"Response Usage.TotalTokens is 0\")\n\t\t}\n\t\tt.Logf(\"Usage: prompt=%d, completion=%d, total=%d\",\n\t\t\tresponse.Usage.PromptTokens, response.Usage.CompletionTokens, response.Usage.TotalTokens)\n\t}\n\tif chunkCount == 0 {\n\t\tt.Error(\"No chunks received\")\n\t}\n\n\tt.Logf(\"Response with temperature=0.7: %+v\", response)\n\tt.Logf(\"Total chunks received: %d\", chunkCount)\n}\n\n// ============================================================================\n// Helper Functions\n// ============================================================================\n\n// newTestContext creates a real Context for testing OpenAI provider\nfunc newTestContext(chatID, connectorID string) *context.Context {\n\tauthorized := &types.AuthorizedInfo{\n\t\tSubject: \"test-user\",\n\t\tClientID: \"test-client\",\n\t\tUserID: \"test-user-123\",\n\t\tTeamID: \"test-team-456\",\n\t\tTenantID: \"test-tenant-789\",\n\t\tSessionID: \"test-session-id\",\n\t\tConstraints: types.DataConstraints{\n\t\t\tTeamOnly: true,\n\t\t\tExtra: map[string]interface{}{\n\t\t\t\t\"test\": \"openai-provider\",\n\t\t\t},\n\t\t},\n\t}\n\n\tctx := context.New(gocontext.Background(), authorized, chatID)\n\tctx.AssistantID = \"test-assistant\"\n\tctx.Locale = \"en-us\"\n\tctx.Theme = \"light\"\n\tctx.Client = context.Client{\n\t\tType: \"web\",\n\t\tUserAgent: \"OpenAIProviderTest/1.0\",\n\t\tIP: \"127.0.0.1\",\n\t}\n\tctx.Referer = context.RefererAPI\n\tctx.Accept = context.AcceptStandard\n\tctx.Route = \"/api/test\"\n\tctx.Metadata = make(map[string]interface{})\n\treturn ctx\n}\n" }, { "path": "agent/llm/providers/openai/temperature_test.go", "content": "package openai_test\n\nimport (\n\tgocontext \"context\"\n\t\"testing\"\n\n\t\"github.com/yaoapp/gou/connector\"\n\t\"github.com/yaoapp/gou/connector/openai\"\n\t\"github.com/yaoapp/yao/agent/context\"\n\t\"github.com/yaoapp/yao/agent/llm\"\n\t\"github.com/yaoapp/yao/config\"\n\t\"github.com/yaoapp/yao/openapi/oauth/types\"\n\t\"github.com/yaoapp/yao/test\"\n)\n\n// TestTemperatureGPT5AutoReset tests that GPT-5 automatically resets temperature to 1.0\n// Temporarily commented out\n// func TestTemperatureGPT5AutoReset(t *testing.T) {\n// \ttest.Prepare(t, config.Conf)\n// \tdefer test.Clean()\n//\n// \tconn, err := connector.Select(\"openai.gpt-5\")\n// \tif err != nil {\n// \t\tt.Fatalf(\"Failed to select connector: %v\", err)\n// \t}\n//\n// \tinvalidTemp := 0.7 // GPT-5 doesn't support this\n// \toptions := &context.CompletionOptions{\n// \t\tCapabilities: &openai.Capabilities{\n// \t\t\tReasoning: true,\n// \t\t},\n// \t\tTemperature: &invalidTemp, // Should be reset to 1.0\n// \t}\n//\n// \tllmInstance, err := llm.New(conn, options)\n// \tif err != nil {\n// \t\tt.Fatalf(\"Failed to create LLM instance: %v\", err)\n// \t}\n//\n// \tmessages := []context.Message{\n// \t\t{\n// \t\t\tRole: context.RoleUser,\n// \t\t\tContent: \"Say 'OK'\",\n// \t\t},\n// \t}\n//\n// \tmaxTokens := 10\n// \toptions.MaxCompletionTokens = &maxTokens\n//\n// \tctx := newTemperatureTestContext(\"test-gpt5-temp\", \"openai.gpt-5\")\n//\n// \t// Should succeed (temperature automatically reset to 1.0)\n// \tresponse, err := llmInstance.Post(ctx, messages, options)\n// \tif err != nil {\n// \t\tt.Fatalf(\"Post failed: %v\", err)\n// \t}\n//\n// \tif response == nil {\n// \t\tt.Fatal(\"Response is nil\")\n// \t}\n//\n// \tt.Log(\"✓ GPT-5 successfully handled invalid temperature by resetting to 1.0\")\n// \tt.Logf(\"Response: %v\", response.Content)\n// }\n\n// TestTemperatureDeepSeekR1AutoReset tests that DeepSeek R1 automatically resets temperature to 1.0\nfunc TestTemperatureDeepSeekR1AutoReset(t *testing.T) {\n\ttest.Prepare(t, config.Conf)\n\tdefer test.Clean()\n\n\tconn, err := connector.Select(\"deepseek.r1\")\n\tif err != nil {\n\t\tt.Fatalf(\"Failed to select connector: %v\", err)\n\t}\n\n\tinvalidTemp := 0.5 // DeepSeek R1 doesn't support this\n\toptions := &context.CompletionOptions{\n\t\tCapabilities: &openai.Capabilities{\n\t\t\tReasoning: true,\n\t\t},\n\t\tTemperature: &invalidTemp, // Should be reset to 1.0\n\t}\n\n\tllmInstance, err := llm.New(conn, options)\n\tif err != nil {\n\t\tt.Fatalf(\"Failed to create LLM instance: %v\", err)\n\t}\n\n\tmessages := []context.Message{\n\t\t{\n\t\t\tRole: context.RoleUser,\n\t\t\tContent: \"Say 'Hello'\",\n\t\t},\n\t}\n\n\tmaxTokens := 100\n\toptions.MaxCompletionTokens = &maxTokens\n\n\tctx := newTemperatureTestContext(\"test-deepseek-r1-temp\", \"deepseek.r1\")\n\n\t// Should succeed (temperature automatically reset to 1.0)\n\tresponse, err := llmInstance.Post(ctx, messages, options)\n\tif err != nil {\n\t\tt.Fatalf(\"Post failed: %v\", err)\n\t}\n\n\tif response == nil {\n\t\tt.Fatal(\"Response is nil\")\n\t}\n\n\tt.Log(\"✓ DeepSeek R1 successfully handled invalid temperature by resetting to 1.0\")\n\tt.Logf(\"Response content: %v\", response.Content)\n\tif response.ReasoningContent != \"\" {\n\t\tt.Logf(\"Reasoning content length: %d\", len(response.ReasoningContent))\n\t}\n}\n\n// TestTemperatureGPT4oPreserved tests that GPT-4o preserves custom temperature\nfunc TestTemperatureGPT4oPreserved(t *testing.T) {\n\ttest.Prepare(t, config.Conf)\n\tdefer test.Clean()\n\n\tconn, err := connector.Select(\"openai.gpt-4o\")\n\tif err != nil {\n\t\tt.Fatalf(\"Failed to select connector: %v\", err)\n\t}\n\n\tcustomTemp := 0.3 // GPT-4o should preserve this\n\toptions := &context.CompletionOptions{\n\t\tCapabilities: &openai.Capabilities{\n\t\t\tReasoning: false, // Not a reasoning model\n\t\t\tToolCalls: true,\n\t\t},\n\t\tTemperature: &customTemp, // Should be preserved\n\t}\n\n\tllmInstance, err := llm.New(conn, options)\n\tif err != nil {\n\t\tt.Fatalf(\"Failed to create LLM instance: %v\", err)\n\t}\n\n\tmessages := []context.Message{\n\t\t{\n\t\t\tRole: context.RoleUser,\n\t\t\tContent: \"Say 'OK'\",\n\t\t},\n\t}\n\n\tmaxTokens := 10\n\toptions.MaxCompletionTokens = &maxTokens\n\n\tctx := newTemperatureTestContext(\"test-gpt4o-temp\", \"openai.gpt-4o\")\n\n\t// Should succeed with custom temperature preserved\n\tresponse, err := llmInstance.Post(ctx, messages, options)\n\tif err != nil {\n\t\tt.Fatalf(\"Post failed: %v\", err)\n\t}\n\n\tif response == nil {\n\t\tt.Fatal(\"Response is nil\")\n\t}\n\n\tt.Log(\"✓ GPT-4o successfully preserved custom temperature (0.3)\")\n\tt.Logf(\"Response: %v\", response.Content)\n}\n\n// TestTemperatureDeepSeekV3Preserved tests that DeepSeek V3 preserves custom temperature\nfunc TestTemperatureDeepSeekV3Preserved(t *testing.T) {\n\ttest.Prepare(t, config.Conf)\n\tdefer test.Clean()\n\n\tconn, err := connector.Select(\"deepseek.v3\")\n\tif err != nil {\n\t\tt.Fatalf(\"Failed to select connector: %v\", err)\n\t}\n\n\tcustomTemp := 0.8 // DeepSeek V3 should preserve this\n\toptions := &context.CompletionOptions{\n\t\tCapabilities: &openai.Capabilities{\n\t\t\tReasoning: false, // Not a reasoning model\n\t\t\tToolCalls: true,\n\t\t},\n\t\tTemperature: &customTemp, // Should be preserved\n\t}\n\n\tllmInstance, err := llm.New(conn, options)\n\tif err != nil {\n\t\tt.Fatalf(\"Failed to create LLM instance: %v\", err)\n\t}\n\n\tmessages := []context.Message{\n\t\t{\n\t\t\tRole: context.RoleUser,\n\t\t\tContent: \"Say 'Hello World'\",\n\t\t},\n\t}\n\n\tmaxTokens := 20\n\toptions.MaxCompletionTokens = &maxTokens\n\n\tctx := newTemperatureTestContext(\"test-deepseek-v3-temp\", \"deepseek.v3\")\n\n\t// Should succeed with custom temperature preserved\n\tresponse, err := llmInstance.Post(ctx, messages, options)\n\tif err != nil {\n\t\tt.Fatalf(\"Post failed: %v\", err)\n\t}\n\n\tif response == nil {\n\t\tt.Fatal(\"Response is nil\")\n\t}\n\n\tt.Log(\"✓ DeepSeek V3 successfully preserved custom temperature (0.8)\")\n\tt.Logf(\"Response: %v\", response.Content)\n}\n\n// TestTemperatureGPT5Default tests that GPT-5 with temperature=1.0 works fine\n// Temporarily commented out\n// func TestTemperatureGPT5Default(t *testing.T) {\n// \ttest.Prepare(t, config.Conf)\n// \tdefer test.Clean()\n//\n// \tconn, err := connector.Select(\"openai.gpt-5\")\n// \tif err != nil {\n// \t\tt.Fatalf(\"Failed to select connector: %v\", err)\n// \t}\n//\n// \tdefaultTemp := 1.0 // GPT-5's valid temperature\n// \toptions := &context.CompletionOptions{\n// \t\tCapabilities: &openai.Capabilities{\n// \t\t\tReasoning: true,\n// \t\t},\n// \t\tTemperature: &defaultTemp, // Should work fine\n// \t}\n//\n// \tllmInstance, err := llm.New(conn, options)\n// \tif err != nil {\n// \t\tt.Fatalf(\"Failed to create LLM instance: %v\", err)\n// \t}\n//\n// \tmessages := []context.Message{\n// \t\t{\n// \t\t\tRole: context.RoleUser,\n// \t\t\tContent: \"What is 2+2? Reply with just the number.\",\n// \t\t},\n// \t}\n//\n// \tmaxTokens := 10\n// \toptions.MaxCompletionTokens = &maxTokens\n//\n// \tctx := newTemperatureTestContext(\"test-gpt5-temp-default\", \"openai.gpt-5\")\n//\n// \t// Should succeed with default temperature\n// \tresponse, err := llmInstance.Post(ctx, messages, options)\n// \tif err != nil {\n// \t\tt.Fatalf(\"Post failed: %v\", err)\n// \t}\n//\n// \tif response == nil {\n// \t\tt.Fatal(\"Response is nil\")\n// \t}\n//\n// \tt.Log(\"✓ GPT-5 successfully handled default temperature (1.0)\")\n// \tt.Logf(\"Response: %v\", response.Content)\n// }\n\n// TestTemperatureNoTemperatureProvided tests that models work when no temperature is provided\nfunc TestTemperatureNoTemperatureProvided(t *testing.T) {\n\ttest.Prepare(t, config.Conf)\n\tdefer test.Clean()\n\n\ttestCases := []struct {\n\t\tname string\n\t\tconnector string\n\t\treasoning bool\n\t}{\n\t\t// {\"GPT-5 No Temp\", \"openai.gpt-5\", true}, // Temporarily commented out\n\t\t{\"GPT-4o No Temp\", \"openai.gpt-4o\", false},\n\t\t{\"DeepSeek R1 No Temp\", \"deepseek.r1\", true},\n\t\t{\"DeepSeek V3 No Temp\", \"deepseek.v3\", false},\n\t}\n\n\tfor _, tc := range testCases {\n\t\tt.Run(tc.name, func(t *testing.T) {\n\t\t\tconn, err := connector.Select(tc.connector)\n\t\t\tif err != nil {\n\t\t\t\tt.Fatalf(\"Failed to select connector: %v\", err)\n\t\t\t}\n\n\t\t\toptions := &context.CompletionOptions{\n\t\t\t\tCapabilities: &openai.Capabilities{\n\t\t\t\t\tReasoning: false,\n\t\t\t\t\tToolCalls: true,\n\t\t\t\t},\n\t\t\t}\n\t\t\tif tc.reasoning {\n\t\t\t\toptions.Capabilities.Reasoning = true\n\t\t\t}\n\t\t\t// Temperature not set - should use API default\n\n\t\t\tllmInstance, err := llm.New(conn, options)\n\t\t\tif err != nil {\n\t\t\t\tt.Fatalf(\"Failed to create LLM instance: %v\", err)\n\t\t\t}\n\n\t\t\tmessages := []context.Message{\n\t\t\t\t{\n\t\t\t\t\tRole: context.RoleUser,\n\t\t\t\t\tContent: \"Say 'OK'\",\n\t\t\t\t},\n\t\t\t}\n\n\t\t\tmaxTokens := 10\n\t\t\toptions.MaxCompletionTokens = &maxTokens\n\n\t\t\tctx := newTemperatureTestContext(\"test-no-temp-\"+tc.connector, tc.connector)\n\n\t\t\tresponse, err := llmInstance.Post(ctx, messages, options)\n\t\t\tif err != nil {\n\t\t\t\tt.Fatalf(\"Post failed: %v\", err)\n\t\t\t}\n\n\t\t\tif response == nil {\n\t\t\t\tt.Fatal(\"Response is nil\")\n\t\t\t}\n\n\t\t\tt.Logf(\"✓ %s works fine without temperature parameter\", tc.connector)\n\t\t})\n\t}\n}\n\n// ============================================================================\n// Helper Functions\n// ============================================================================\n\n// newTemperatureTestContext creates a real Context for testing temperature handling\nfunc newTemperatureTestContext(chatID, connectorID string) *context.Context {\n\tauthorized := &types.AuthorizedInfo{\n\t\tSubject: \"test-user\",\n\t\tClientID: \"test-client\",\n\t\tUserID: \"test-user-123\",\n\t\tTeamID: \"test-team-456\",\n\t\tTenantID: \"test-tenant-789\",\n\t\tSessionID: \"test-session-id\",\n\t\tConstraints: types.DataConstraints{\n\t\t\tTeamOnly: true,\n\t\t\tExtra: map[string]interface{}{\n\t\t\t\t\"test\": \"temperature\",\n\t\t\t},\n\t\t},\n\t}\n\n\tctx := context.New(gocontext.Background(), authorized, chatID)\n\tctx.AssistantID = \"test-assistant\"\n\tctx.Locale = \"en-us\"\n\tctx.Theme = \"light\"\n\tctx.Client = context.Client{\n\t\tType: \"web\",\n\t\tUserAgent: \"TemperatureTest/1.0\",\n\t\tIP: \"127.0.0.1\",\n\t}\n\tctx.Referer = context.RefererAPI\n\tctx.Accept = context.AcceptStandard\n\tctx.Route = \"/api/test\"\n\tctx.Metadata = make(map[string]interface{})\n\treturn ctx\n}\n" }, { "path": "agent/llm/providers/openai/types.go", "content": "package openai\n\nimport (\n\t\"github.com/yaoapp/yao/agent/context\"\n\t\"github.com/yaoapp/yao/agent/output/message\"\n)\n\n// StreamChunk represents a chunk from OpenAI's streaming response\ntype StreamChunk struct {\n\tID string `json:\"id\"`\n\tObject string `json:\"object\"`\n\tCreated int64 `json:\"created\"`\n\tModel string `json:\"model\"`\n\tChoices []Delta `json:\"choices\"`\n\tUsage *struct {\n\t\tPromptTokens int `json:\"prompt_tokens\"`\n\t\tCompletionTokens int `json:\"completion_tokens\"`\n\t\tTotalTokens int `json:\"total_tokens\"`\n\t} `json:\"usage,omitempty\"`\n}\n\n// Delta represents the delta in a streaming chunk\ntype Delta struct {\n\tIndex int `json:\"index\"`\n\tDelta DeltaContent `json:\"delta\"`\n\tFinishReason *string `json:\"finish_reason\"`\n}\n\n// DeltaContent represents the content in a delta\ntype DeltaContent struct {\n\tRole string `json:\"role,omitempty\"`\n\tContent string `json:\"content,omitempty\"`\n\tReasoningContent string `json:\"reasoning_content,omitempty\"` // DeepSeek R1 reasoning\n\tToolCalls []ToolCallDelta `json:\"tool_calls,omitempty\"`\n\tRefusal string `json:\"refusal,omitempty\"`\n}\n\n// ToolCallDelta represents a tool call delta in streaming\ntype ToolCallDelta struct {\n\tIndex int `json:\"index\"`\n\tID string `json:\"id,omitempty\"`\n\tType string `json:\"type,omitempty\"`\n\tFunction FunctionCallDelta `json:\"function,omitempty\"`\n}\n\n// FunctionCallDelta represents a function call delta\ntype FunctionCallDelta struct {\n\tName string `json:\"name,omitempty\"`\n\tArguments string `json:\"arguments,omitempty\"`\n}\n\n// CompletionResponseFull represents the full non-streaming response\ntype CompletionResponseFull struct {\n\tID string `json:\"id\"`\n\tObject string `json:\"object\"`\n\tCreated int64 `json:\"created\"`\n\tModel string `json:\"model\"`\n\tChoices []struct {\n\t\tIndex int `json:\"index\"`\n\t\tMessage struct {\n\t\t\tRole context.MessageRole `json:\"role\"`\n\t\t\tContent interface{} `json:\"content,omitempty\"` // string or array\n\t\t\tReasoningContent string `json:\"reasoning_content,omitempty\"` // DeepSeek R1 reasoning\n\t\t\tToolCalls []context.ToolCall `json:\"tool_calls,omitempty\"`\n\t\t\tRefusal *string `json:\"refusal,omitempty\"`\n\t\t} `json:\"message\"`\n\t\tFinishReason string `json:\"finish_reason\"`\n\t} `json:\"choices\"`\n\tUsage *message.UsageInfo `json:\"usage,omitempty\"`\n\tSystemFingerprint string `json:\"system_fingerprint,omitempty\"`\n}\n\n// streamAccumulator accumulates streaming response data\ntype streamAccumulator struct {\n\tid string\n\tmodel string\n\tcreated int64\n\trole string\n\tcontent string\n\treasoningContent string // DeepSeek R1 reasoning content\n\trefusal string\n\ttoolCalls map[int]*accumulatedToolCall\n\tfinishReason string\n\tusage *message.UsageInfo\n}\n\n// accumulatedToolCall accumulates a single tool call\ntype accumulatedToolCall struct {\n\tid string\n\ttyp string\n\tfunctionName string\n\tfunctionArgs string\n}\n\n// messageTracker tracks the current message state for lifecycle events\ntype messageTracker struct {\n\tactive bool // Whether a message is currently active\n\tmessageID string // Current message ID\n\tmessageType message.StreamChunkType // Current message type (thinking, text, tool_call)\n\tstartTime int64 // Message start timestamp\n\tchunkCount int // Number of chunks in this message\n\ttoolCallInfo *message.EventToolCallInfo // Tool call info if message is tool_call type\n\tidGenerator *message.IDGenerator // ID generator from context\n}\n" }, { "path": "agent/load.go", "content": "package agent\n\nimport (\n\t\"fmt\"\n\t\"path/filepath\"\n\n\t\"github.com/yaoapp/gou/application\"\n\t\"github.com/yaoapp/gou/connector\"\n\t\"github.com/yaoapp/gou/helper\"\n\t\"github.com/yaoapp/yao/agent/assistant\"\n\t\"github.com/yaoapp/yao/agent/context\"\n\t\"github.com/yaoapp/yao/agent/i18n\"\n\tsearchDefaults \"github.com/yaoapp/yao/agent/search/defaults\"\n\tsearchTypes \"github.com/yaoapp/yao/agent/search/types\"\n\tstoreMongo \"github.com/yaoapp/yao/agent/store/mongo\"\n\tstoreRedis \"github.com/yaoapp/yao/agent/store/redis\"\n\tstore \"github.com/yaoapp/yao/agent/store/types\"\n\t\"github.com/yaoapp/yao/agent/store/xun\"\n\t\"github.com/yaoapp/yao/agent/types\"\n\t\"github.com/yaoapp/yao/config\"\n)\n\nvar agentDSL *types.DSL\n\n// Load load AIGC\nfunc Load(cfg config.Config) error {\n\n\tsetting := types.DSL{\n\t\tCache: \"__yao.agent.cache\", // default is \"__yao.agent.cache\"\n\t\tStoreSetting: store.Setting{\n\t\t\tMaxSize: 20,\n\t\t\tTTL: 90 * 24 * 60 * 60, // 90 days in seconds\n\t\t},\n\t}\n\n\tbytes, err := application.App.Read(filepath.Join(\"agent\", \"agent.yml\"))\n\tif err != nil {\n\t\treturn err\n\t}\n\n\terr = application.Parse(\"agent.yml\", bytes, &setting)\n\tif err != nil {\n\t\treturn err\n\t}\n\n\tif setting.StoreSetting.MaxSize == 0 {\n\t\tsetting.StoreSetting.MaxSize = 20 // default is 20\n\t}\n\n\t// Resolve $ENV.XXX references in system and uses fields\n\tresolveEnvStrings(&setting)\n\n\t// Default Assistant, Agent is the developer name, Mohe is the brand name of the assistant\n\tif setting.Uses == nil {\n\t\tsetting.Uses = &types.Uses{Default: \"mohe\"} // Agent is the developer name, Mohe is the brand name of the assistant\n\t}\n\n\t// Title Assistant (default to system agent)\n\tif setting.Uses.Title == \"\" {\n\t\tsetting.Uses.Title = \"__yao.title\"\n\t}\n\n\t// Prompt Assistant (default to system agent)\n\tif setting.Uses.Prompt == \"\" {\n\t\tsetting.Uses.Prompt = \"__yao.prompt\"\n\t}\n\n\t// RobotPrompt Assistant (default to system agent)\n\tif setting.Uses.RobotPrompt == \"\" {\n\t\tsetting.Uses.RobotPrompt = \"__yao.robot_prompt\"\n\t}\n\n\tagentDSL = &setting\n\n\t// Store Setting\n\terr = initStore()\n\tif err != nil {\n\t\treturn err\n\t}\n\n\t// Initialize Global I18n\n\terr = initGlobalI18n()\n\tif err != nil {\n\t\treturn err\n\t}\n\n\t// Initialize Global Prompts\n\terr = initGlobalPrompts()\n\tif err != nil {\n\t\treturn err\n\t}\n\n\t// Initialize KB Configuration\n\terr = initKBConfig()\n\tif err != nil {\n\t\treturn err\n\t}\n\n\t// Initialize Search Configuration\n\terr = initSearchConfig()\n\tif err != nil {\n\t\treturn err\n\t}\n\n\t// Initialize Assistant\n\terr = initAssistant()\n\tif err != nil {\n\t\treturn err\n\t}\n\n\treturn nil\n}\n\n// GetAgent returns the Agent settings\nfunc GetAgent() *types.DSL {\n\treturn agentDSL\n}\n\n// initGlobalI18n initialize the global i18n\nfunc initGlobalI18n() error {\n\tlocales, err := i18n.GetLocales(\"agent\")\n\tif err != nil {\n\t\treturn err\n\t}\n\ti18n.Locales[\"__global__\"] = locales.Flatten()\n\treturn nil\n}\n\n// initGlobalPrompts initialize the global prompts from agent/prompts.yml\nfunc initGlobalPrompts() error {\n\tprompts, _, err := store.LoadGlobalPrompts()\n\tif err != nil {\n\t\treturn err\n\t}\n\tagentDSL.GlobalPrompts = prompts\n\treturn nil\n}\n\n// GetGlobalPrompts returns the global prompts\n// ctx: context variables for parsing $CTX.* variables\nfunc GetGlobalPrompts(ctx map[string]string) []store.Prompt {\n\tif agentDSL == nil || len(agentDSL.GlobalPrompts) == 0 {\n\t\treturn nil\n\t}\n\treturn store.Prompts(agentDSL.GlobalPrompts).Parse(ctx)\n}\n\n// initStore initialize the store\nfunc initStore() error {\n\n\tvar err error\n\tif agentDSL.StoreSetting.Connector == \"default\" || agentDSL.StoreSetting.Connector == \"\" {\n\t\tagentDSL.Store, err = xun.NewXun(agentDSL.StoreSetting)\n\t\treturn err\n\t}\n\n\t// other connector\n\tconn, err := connector.Select(agentDSL.StoreSetting.Connector)\n\tif err != nil {\n\t\treturn fmt.Errorf(\"load connectors error: %s\", err.Error())\n\t}\n\n\tif conn.Is(connector.DATABASE) {\n\t\tagentDSL.Store, err = xun.NewXun(agentDSL.StoreSetting)\n\t\treturn err\n\n\t} else if conn.Is(connector.REDIS) {\n\t\tagentDSL.Store = storeRedis.NewRedis()\n\t\treturn nil\n\n\t} else if conn.Is(connector.MONGO) {\n\t\tagentDSL.Store = storeMongo.NewMongo()\n\t\treturn nil\n\t}\n\n\treturn fmt.Errorf(\"Agent store connector %s not support\", agentDSL.StoreSetting.Connector)\n}\n\n// initAssistant initialize the assistant\nfunc initAssistant() error {\n\n\t// Set Storage\n\tassistant.SetStorage(agentDSL.Store)\n\n\t// Set Store Setting (MaxSize, TTL, etc.)\n\tassistant.SetStoreSetting(&agentDSL.StoreSetting)\n\n\t// Set global Uses configuration\n\tif agentDSL.Uses != nil {\n\t\tglobalUses := &context.Uses{\n\t\t\tVision: agentDSL.Uses.Vision,\n\t\t\tAudio: agentDSL.Uses.Audio,\n\t\t\tSearch: agentDSL.Uses.Search,\n\t\t\tFetch: agentDSL.Uses.Fetch,\n\t\t\tWeb: agentDSL.Uses.Web,\n\t\t\tKeyword: agentDSL.Uses.Keyword,\n\t\t\tQueryDSL: agentDSL.Uses.QueryDSL,\n\t\t\tRerank: agentDSL.Uses.Rerank,\n\t\t}\n\t\tassistant.SetGlobalUses(globalUses)\n\t}\n\n\t// Set global prompts\n\tif len(agentDSL.GlobalPrompts) > 0 {\n\t\tassistant.SetGlobalPrompts(agentDSL.GlobalPrompts)\n\t}\n\n\tif agentDSL.KB != nil {\n\t\tassistant.SetGlobalKBSetting(agentDSL.KB)\n\t}\n\n\tif agentDSL.Search != nil {\n\t\tassistant.SetGlobalSearchConfig(agentDSL.Search)\n\t}\n\n\t// Set system agents configuration\n\tif agentDSL.System != nil {\n\t\tassistant.SetSystemConfig(&assistant.SystemConfig{\n\t\t\tDefault: agentDSL.System.Default,\n\t\t\tKeyword: agentDSL.System.Keyword,\n\t\t\tQueryDSL: agentDSL.System.QueryDSL,\n\t\t\tTitle: agentDSL.System.Title,\n\t\t\tPrompt: agentDSL.System.Prompt,\n\t\t\tNeedSearch: agentDSL.System.NeedSearch,\n\t\t\tEntity: agentDSL.System.Entity,\n\t\t})\n\t}\n\n\t// Load System Agents (from bindata: __yao.keyword, __yao.querydsl, etc.)\n\tif err := assistant.LoadSystemAgents(); err != nil {\n\t\treturn err\n\t}\n\n\t// Load Built-in Assistants (from application /assistants directory)\n\terr := assistant.LoadBuiltIn()\n\tif err != nil {\n\t\treturn err\n\t}\n\n\t// Default Assistant\n\tdefaultAssistant, err := defaultAssistant()\n\tif err != nil {\n\t\treturn err\n\t}\n\n\tagentDSL.Assistant = defaultAssistant\n\treturn nil\n}\n\n// initKBConfig initialize the knowledge base configuration from agent/kb.yml\nfunc initKBConfig() error {\n\tpath := filepath.Join(\"agent\", \"kb.yml\")\n\tif exists, _ := application.App.Exists(path); !exists {\n\t\treturn nil // KB config is optional\n\t}\n\n\t// Read the KB configuration\n\tbytes, err := application.App.Read(path)\n\tif err != nil {\n\t\treturn err\n\t}\n\n\tvar kbSetting store.KBSetting\n\terr = application.Parse(\"kb.yml\", bytes, &kbSetting)\n\tif err != nil {\n\t\treturn err\n\t}\n\n\tagentDSL.KB = &kbSetting\n\treturn nil\n}\n\n// initSearchConfig initialize the search configuration from agent/search.yml\nfunc initSearchConfig() error {\n\t// Start with system defaults\n\tagentDSL.Search = searchDefaults.SystemDefaults\n\n\tpath := filepath.Join(\"agent\", \"search.yml\")\n\tif exists, _ := application.App.Exists(path); !exists {\n\t\treturn nil // Search config is optional, use defaults\n\t}\n\n\t// Read the search configuration\n\tbytes, err := application.App.Read(path)\n\tif err != nil {\n\t\treturn err\n\t}\n\n\tvar searchConfig searchTypes.Config\n\terr = application.Parse(\"search.yml\", bytes, &searchConfig)\n\tif err != nil {\n\t\treturn err\n\t}\n\n\t// Merge with defaults\n\tagentDSL.Search = mergeSearchConfig(searchDefaults.SystemDefaults, &searchConfig)\n\treturn nil\n}\n\n// mergeSearchConfig merges two search configs (base < override)\nfunc mergeSearchConfig(base, override *searchTypes.Config) *searchTypes.Config {\n\tif base == nil {\n\t\treturn override\n\t}\n\tif override == nil {\n\t\treturn base\n\t}\n\n\tresult := *base // Copy base\n\n\t// Merge Web config\n\tif override.Web != nil {\n\t\tif result.Web == nil {\n\t\t\tresult.Web = override.Web\n\t\t} else {\n\t\t\tif override.Web.Provider != \"\" {\n\t\t\t\tresult.Web.Provider = override.Web.Provider\n\t\t\t}\n\t\t\tif override.Web.APIKeyEnv != \"\" {\n\t\t\t\tresult.Web.APIKeyEnv = override.Web.APIKeyEnv\n\t\t\t}\n\t\t\tif override.Web.MaxResults > 0 {\n\t\t\t\tresult.Web.MaxResults = override.Web.MaxResults\n\t\t\t}\n\t\t}\n\t}\n\n\t// Merge KB config\n\tif override.KB != nil {\n\t\tif result.KB == nil {\n\t\t\tresult.KB = override.KB\n\t\t} else {\n\t\t\tif len(override.KB.Collections) > 0 {\n\t\t\t\tresult.KB.Collections = override.KB.Collections\n\t\t\t}\n\t\t\tif override.KB.Threshold > 0 {\n\t\t\t\tresult.KB.Threshold = override.KB.Threshold\n\t\t\t}\n\t\t\tif override.KB.Graph {\n\t\t\t\tresult.KB.Graph = override.KB.Graph\n\t\t\t}\n\t\t}\n\t}\n\n\t// Merge DB config\n\tif override.DB != nil {\n\t\tif result.DB == nil {\n\t\t\tresult.DB = override.DB\n\t\t} else {\n\t\t\tif len(override.DB.Models) > 0 {\n\t\t\t\tresult.DB.Models = override.DB.Models\n\t\t\t}\n\t\t\tif override.DB.MaxResults > 0 {\n\t\t\t\tresult.DB.MaxResults = override.DB.MaxResults\n\t\t\t}\n\t\t}\n\t}\n\n\t// Merge Keyword config\n\tif override.Keyword != nil {\n\t\tif result.Keyword == nil {\n\t\t\tresult.Keyword = override.Keyword\n\t\t} else {\n\t\t\tif override.Keyword.MaxKeywords > 0 {\n\t\t\t\tresult.Keyword.MaxKeywords = override.Keyword.MaxKeywords\n\t\t\t}\n\t\t\tif override.Keyword.Language != \"\" {\n\t\t\t\tresult.Keyword.Language = override.Keyword.Language\n\t\t\t}\n\t\t}\n\t}\n\n\t// Merge QueryDSL config\n\tif override.QueryDSL != nil {\n\t\tresult.QueryDSL = override.QueryDSL\n\t}\n\n\t// Merge Rerank config\n\tif override.Rerank != nil {\n\t\tif result.Rerank == nil {\n\t\t\tresult.Rerank = override.Rerank\n\t\t} else {\n\t\t\tif override.Rerank.TopN > 0 {\n\t\t\t\tresult.Rerank.TopN = override.Rerank.TopN\n\t\t\t}\n\t\t}\n\t}\n\n\t// Merge Citation config\n\tif override.Citation != nil {\n\t\tif result.Citation == nil {\n\t\t\tresult.Citation = override.Citation\n\t\t} else {\n\t\t\tif override.Citation.Format != \"\" {\n\t\t\t\tresult.Citation.Format = override.Citation.Format\n\t\t\t}\n\t\t\t// AutoInjectPrompt is a bool, need to check if explicitly set\n\t\t\tresult.Citation.AutoInjectPrompt = override.Citation.AutoInjectPrompt\n\t\t\tif override.Citation.CustomPrompt != \"\" {\n\t\t\t\tresult.Citation.CustomPrompt = override.Citation.CustomPrompt\n\t\t\t}\n\t\t}\n\t}\n\n\t// Merge Weights config\n\tif override.Weights != nil {\n\t\tif result.Weights == nil {\n\t\t\tresult.Weights = override.Weights\n\t\t} else {\n\t\t\tif override.Weights.User > 0 {\n\t\t\t\tresult.Weights.User = override.Weights.User\n\t\t\t}\n\t\t\tif override.Weights.Hook > 0 {\n\t\t\t\tresult.Weights.Hook = override.Weights.Hook\n\t\t\t}\n\t\t\tif override.Weights.Auto > 0 {\n\t\t\t\tresult.Weights.Auto = override.Weights.Auto\n\t\t\t}\n\t\t}\n\t}\n\n\t// Merge Options config\n\tif override.Options != nil {\n\t\tif result.Options == nil {\n\t\t\tresult.Options = override.Options\n\t\t} else {\n\t\t\tif override.Options.SkipThreshold > 0 {\n\t\t\t\tresult.Options.SkipThreshold = override.Options.SkipThreshold\n\t\t\t}\n\t\t}\n\t}\n\n\treturn &result\n}\n\n// GetSearchConfig returns the global search configuration\nfunc GetSearchConfig() *searchTypes.Config {\n\tif agentDSL == nil {\n\t\treturn searchDefaults.SystemDefaults\n\t}\n\treturn agentDSL.Search\n}\n\n// defaultAssistant get the default assistant\nfunc defaultAssistant() (*assistant.Assistant, error) {\n\tif agentDSL.Uses == nil || agentDSL.Uses.Default == \"\" {\n\t\treturn nil, fmt.Errorf(\"default assistant not found\")\n\t}\n\treturn assistant.Get(agentDSL.Uses.Default)\n}\n\n// resolveEnvStrings resolves $ENV.XXX references in agent.yml string fields.\n// agent.yml is parsed via yaml.Unmarshal which does not handle $ENV substitution,\n// unlike connector files which call helper.EnvString explicitly during Register.\nfunc resolveEnvStrings(setting *types.DSL) {\n\tif setting.System != nil {\n\t\tsetting.System.Default = helper.EnvString(setting.System.Default)\n\t\tsetting.System.Keyword = helper.EnvString(setting.System.Keyword)\n\t\tsetting.System.QueryDSL = helper.EnvString(setting.System.QueryDSL)\n\t\tsetting.System.Title = helper.EnvString(setting.System.Title)\n\t\tsetting.System.Prompt = helper.EnvString(setting.System.Prompt)\n\t\tsetting.System.RobotPrompt = helper.EnvString(setting.System.RobotPrompt)\n\t\tsetting.System.NeedSearch = helper.EnvString(setting.System.NeedSearch)\n\t\tsetting.System.Entity = helper.EnvString(setting.System.Entity)\n\t}\n\n\tif setting.Uses != nil {\n\t\tsetting.Uses.Default = helper.EnvString(setting.Uses.Default)\n\t\tsetting.Uses.Title = helper.EnvString(setting.Uses.Title)\n\t\tsetting.Uses.Prompt = helper.EnvString(setting.Uses.Prompt)\n\t\tsetting.Uses.RobotPrompt = helper.EnvString(setting.Uses.RobotPrompt)\n\t\tsetting.Uses.Vision = helper.EnvString(setting.Uses.Vision)\n\t\tsetting.Uses.Audio = helper.EnvString(setting.Uses.Audio)\n\t\tsetting.Uses.Search = helper.EnvString(setting.Uses.Search)\n\t\tsetting.Uses.Fetch = helper.EnvString(setting.Uses.Fetch)\n\t\tsetting.Uses.Web = helper.EnvString(setting.Uses.Web)\n\t\tsetting.Uses.Keyword = helper.EnvString(setting.Uses.Keyword)\n\t\tsetting.Uses.QueryDSL = helper.EnvString(setting.Uses.QueryDSL)\n\t\tsetting.Uses.Rerank = helper.EnvString(setting.Uses.Rerank)\n\t}\n\n\tsetting.Cache = helper.EnvString(setting.Cache)\n}\n" }, { "path": "agent/load_test.go", "content": "package agent\n\nimport (\n\t\"strings\"\n\t\"testing\"\n\t\"time\"\n\n\t\"github.com/stretchr/testify/assert\"\n\t\"github.com/stretchr/testify/require\"\n\t\"github.com/yaoapp/yao/agent/assistant\"\n\t\"github.com/yaoapp/yao/agent/types\"\n\t\"github.com/yaoapp/yao/config\"\n\t\"github.com/yaoapp/yao/test\"\n)\n\nfunc prepare(t *testing.T) {\n\ttest.Prepare(t, config.Conf)\n\terr := Load(config.Conf)\n\trequire.NoError(t, err)\n}\n\nfunc TestLoad(t *testing.T) {\n\tprepare(t)\n\tdefer test.Clean()\n\n\tagent := GetAgent()\n\trequire.NotNil(t, agent)\n\n\tt.Run(\"LoadAgentSettings\", func(t *testing.T) {\n\t\t// Cache setting\n\t\tassert.NotEmpty(t, agent.Cache)\n\n\t\t// Store setting\n\t\tassert.NotNil(t, agent.Store)\n\t\tassert.Greater(t, agent.StoreSetting.MaxSize, 0)\n\n\t\t// Uses setting\n\t\tassert.NotNil(t, agent.Uses)\n\t\tassert.NotEmpty(t, agent.Uses.Default)\n\t})\n\n\tt.Run(\"LoadDefaultAssistant\", func(t *testing.T) {\n\t\tassert.NotNil(t, agent.Assistant)\n\t})\n\n\tt.Run(\"LoadGlobalPrompts\", func(t *testing.T) {\n\t\t// Global prompts should be loaded from agent/prompts.yml\n\t\tassert.NotNil(t, agent.GlobalPrompts)\n\t\tassert.Greater(t, len(agent.GlobalPrompts), 0)\n\n\t\t// First prompt should be system role\n\t\tassert.Equal(t, \"system\", agent.GlobalPrompts[0].Role)\n\n\t\t// Content should contain system context info (with variables not yet parsed)\n\t\tassert.Contains(t, agent.GlobalPrompts[0].Content, \"$SYS.\")\n\t})\n\n\tt.Run(\"LoadKBConfig\", func(t *testing.T) {\n\t\t// KB configuration should be loaded from agent/kb.yml\n\t\tassert.NotNil(t, agent.KB)\n\t\tassert.NotNil(t, agent.KB.Chat)\n\n\t\t// Verify chat KB settings\n\t\tassert.Equal(t, \"__yao.openai\", agent.KB.Chat.EmbeddingProviderID)\n\t\tassert.Equal(t, \"text-embedding-3-small\", agent.KB.Chat.EmbeddingOptionID)\n\t\tassert.Equal(t, \"zh-CN\", agent.KB.Chat.Locale)\n\n\t\t// Verify config\n\t\tassert.NotNil(t, agent.KB.Chat.Config)\n\t\tassert.Equal(t, \"hnsw\", agent.KB.Chat.Config.IndexType.String())\n\t\tassert.Equal(t, \"cosine\", agent.KB.Chat.Config.Distance.String())\n\n\t\t// Verify metadata\n\t\tassert.NotNil(t, agent.KB.Chat.Metadata)\n\t\tassert.Equal(t, \"chat_session\", agent.KB.Chat.Metadata[\"category\"])\n\t\tassert.Equal(t, true, agent.KB.Chat.Metadata[\"auto_created\"])\n\n\t\t// Verify document defaults\n\t\tassert.NotNil(t, agent.KB.Chat.DocumentDefaults)\n\t\tassert.NotNil(t, agent.KB.Chat.DocumentDefaults.Chunking)\n\t\tassert.Equal(t, \"__yao.structured\", agent.KB.Chat.DocumentDefaults.Chunking.ProviderID)\n\t\tassert.Equal(t, \"standard\", agent.KB.Chat.DocumentDefaults.Chunking.OptionID)\n\n\t\tassert.NotNil(t, agent.KB.Chat.DocumentDefaults.Extraction)\n\t\tassert.Equal(t, \"__yao.openai\", agent.KB.Chat.DocumentDefaults.Extraction.ProviderID)\n\t\tassert.Equal(t, \"gpt-4o-mini\", agent.KB.Chat.DocumentDefaults.Extraction.OptionID)\n\n\t\tassert.NotNil(t, agent.KB.Chat.DocumentDefaults.Converter)\n\t\tassert.Equal(t, \"__yao.utf8\", agent.KB.Chat.DocumentDefaults.Converter.ProviderID)\n\t\tassert.Equal(t, \"standard-text\", agent.KB.Chat.DocumentDefaults.Converter.OptionID)\n\t})\n\n\tt.Run(\"LoadSearchConfig\", func(t *testing.T) {\n\t\t// Search configuration should be loaded from agent/search.yml\n\t\tassert.NotNil(t, agent.Search)\n\n\t\t// Verify web config\n\t\tassert.NotNil(t, agent.Search.Web)\n\t\tassert.Equal(t, \"tavily\", agent.Search.Web.Provider)\n\t\tassert.Equal(t, 10, agent.Search.Web.MaxResults)\n\n\t\t// Verify KB config\n\t\tassert.NotNil(t, agent.Search.KB)\n\t\tassert.Equal(t, 0.7, agent.Search.KB.Threshold)\n\t\tassert.False(t, agent.Search.KB.Graph)\n\n\t\t// Verify DB config\n\t\tassert.NotNil(t, agent.Search.DB)\n\t\tassert.Equal(t, 20, agent.Search.DB.MaxResults)\n\n\t\t// Verify keyword config\n\t\tassert.NotNil(t, agent.Search.Keyword)\n\t\tassert.Equal(t, 10, agent.Search.Keyword.MaxKeywords)\n\t\tassert.Equal(t, \"auto\", agent.Search.Keyword.Language)\n\n\t\t// Verify rerank config\n\t\tassert.NotNil(t, agent.Search.Rerank)\n\t\tassert.Equal(t, 10, agent.Search.Rerank.TopN)\n\n\t\t// Verify citation config\n\t\tassert.NotNil(t, agent.Search.Citation)\n\t\tassert.Equal(t, \"#ref:{id}\", agent.Search.Citation.Format)\n\t\tassert.True(t, agent.Search.Citation.AutoInjectPrompt)\n\n\t\t// Verify weights config\n\t\tassert.NotNil(t, agent.Search.Weights)\n\t\tassert.Equal(t, 1.0, agent.Search.Weights.User)\n\t\tassert.Equal(t, 0.8, agent.Search.Weights.Hook)\n\t\tassert.Equal(t, 0.6, agent.Search.Weights.Auto)\n\n\t\t// Verify options config\n\t\tassert.NotNil(t, agent.Search.Options)\n\t\tassert.Equal(t, 5, agent.Search.Options.SkipThreshold)\n\t})\n}\n\nfunc TestGetGlobalPrompts(t *testing.T) {\n\tprepare(t)\n\tdefer test.Clean()\n\n\tt.Run(\"ParseWithoutContext\", func(t *testing.T) {\n\t\tprompts := GetGlobalPrompts(nil)\n\t\trequire.NotNil(t, prompts)\n\t\trequire.Greater(t, len(prompts), 0)\n\n\t\t// $SYS.* variables should be replaced\n\t\tcontent := prompts[0].Content\n\t\tassert.NotContains(t, content, \"$SYS.DATETIME\")\n\t\tassert.NotContains(t, content, \"$SYS.TIMEZONE\")\n\t\tassert.NotContains(t, content, \"$SYS.WEEKDAY\")\n\n\t\t// Should contain actual time values\n\t\tnow := time.Now()\n\t\tassert.Contains(t, content, now.Format(\"2006-01-02\"))\n\t})\n\n\tt.Run(\"ParseWithContext\", func(t *testing.T) {\n\t\tctx := map[string]string{\n\t\t\t\"USER_ID\": \"test-user-123\",\n\t\t\t\"LOCALE\": \"zh-CN\",\n\t\t}\n\n\t\tprompts := GetGlobalPrompts(ctx)\n\t\trequire.NotNil(t, prompts)\n\t\trequire.Greater(t, len(prompts), 0)\n\n\t\t// $SYS.* variables should be replaced\n\t\tcontent := prompts[0].Content\n\t\tassert.NotContains(t, content, \"$SYS.DATETIME\")\n\t})\n\n\tt.Run(\"ParseSystemTimeVariables\", func(t *testing.T) {\n\t\tprompts := GetGlobalPrompts(nil)\n\t\trequire.NotNil(t, prompts)\n\n\t\tcontent := prompts[0].Content\n\t\tnow := time.Now()\n\n\t\t// Should contain current date\n\t\tassert.Contains(t, content, now.Format(\"2006-01-02\"))\n\n\t\t// Should contain timezone\n\t\tassert.Contains(t, content, now.Location().String())\n\n\t\t// Should contain weekday\n\t\tassert.Contains(t, content, now.Weekday().String())\n\t})\n}\n\nfunc TestGetGlobalPromptsWithDisableFlag(t *testing.T) {\n\tprepare(t)\n\tdefer test.Clean()\n\n\tagent := GetAgent()\n\trequire.NotNil(t, agent)\n\n\tt.Run(\"GlobalPromptsExist\", func(t *testing.T) {\n\t\t// Verify global prompts are loaded\n\t\tassert.NotNil(t, agent.GlobalPrompts)\n\t\tassert.Greater(t, len(agent.GlobalPrompts), 0)\n\t})\n\n\tt.Run(\"AssistantCanDisableGlobalPrompts\", func(t *testing.T) {\n\t\t// The fullfields test assistant has disable_global_prompts: true\n\t\t// This test verifies the flag is properly loaded\n\t\t// The actual merging logic is in the assistant module\n\t\tprompts := GetGlobalPrompts(nil)\n\t\tassert.NotNil(t, prompts)\n\n\t\t// Global prompts should still be available\n\t\t// The assistant decides whether to use them based on DisableGlobalPrompts flag\n\t})\n}\n\nfunc TestResolveEnvStrings(t *testing.T) {\n\tt.Setenv(\"TEST_CONNECTOR\", \"openai.gpt-5\")\n\tt.Setenv(\"TEST_ASSISTANT\", \"my-assistant\")\n\tt.Setenv(\"TEST_CACHE\", \"my-cache\")\n\n\tt.Run(\"SystemFields\", func(t *testing.T) {\n\t\tsetting := &types.DSL{\n\t\t\tSystem: &types.System{\n\t\t\t\tDefault: \"$ENV.TEST_CONNECTOR\",\n\t\t\t\tKeyword: \"$ENV.TEST_CONNECTOR\",\n\t\t\t\tQueryDSL: \"$ENV.TEST_CONNECTOR\",\n\t\t\t\tTitle: \"$ENV.TEST_CONNECTOR\",\n\t\t\t\tPrompt: \"$ENV.TEST_CONNECTOR\",\n\t\t\t\tRobotPrompt: \"$ENV.TEST_CONNECTOR\",\n\t\t\t\tNeedSearch: \"$ENV.TEST_CONNECTOR\",\n\t\t\t\tEntity: \"$ENV.TEST_CONNECTOR\",\n\t\t\t},\n\t\t}\n\t\tresolveEnvStrings(setting)\n\n\t\tassert.Equal(t, \"openai.gpt-5\", setting.System.Default)\n\t\tassert.Equal(t, \"openai.gpt-5\", setting.System.Keyword)\n\t\tassert.Equal(t, \"openai.gpt-5\", setting.System.QueryDSL)\n\t\tassert.Equal(t, \"openai.gpt-5\", setting.System.Title)\n\t\tassert.Equal(t, \"openai.gpt-5\", setting.System.Prompt)\n\t\tassert.Equal(t, \"openai.gpt-5\", setting.System.RobotPrompt)\n\t\tassert.Equal(t, \"openai.gpt-5\", setting.System.NeedSearch)\n\t\tassert.Equal(t, \"openai.gpt-5\", setting.System.Entity)\n\t})\n\n\tt.Run(\"UsesFields\", func(t *testing.T) {\n\t\tsetting := &types.DSL{\n\t\t\tUses: &types.Uses{\n\t\t\t\tDefault: \"$ENV.TEST_ASSISTANT\",\n\t\t\t\tTitle: \"$ENV.TEST_ASSISTANT\",\n\t\t\t\tPrompt: \"$ENV.TEST_ASSISTANT\",\n\t\t\t\tRobotPrompt: \"$ENV.TEST_ASSISTANT\",\n\t\t\t\tVision: \"$ENV.TEST_ASSISTANT\",\n\t\t\t\tAudio: \"$ENV.TEST_ASSISTANT\",\n\t\t\t\tSearch: \"$ENV.TEST_ASSISTANT\",\n\t\t\t\tFetch: \"$ENV.TEST_ASSISTANT\",\n\t\t\t\tWeb: \"$ENV.TEST_ASSISTANT\",\n\t\t\t\tKeyword: \"$ENV.TEST_ASSISTANT\",\n\t\t\t\tQueryDSL: \"$ENV.TEST_ASSISTANT\",\n\t\t\t\tRerank: \"$ENV.TEST_ASSISTANT\",\n\t\t\t},\n\t\t}\n\t\tresolveEnvStrings(setting)\n\n\t\tassert.Equal(t, \"my-assistant\", setting.Uses.Default)\n\t\tassert.Equal(t, \"my-assistant\", setting.Uses.Title)\n\t\tassert.Equal(t, \"my-assistant\", setting.Uses.Prompt)\n\t\tassert.Equal(t, \"my-assistant\", setting.Uses.RobotPrompt)\n\t\tassert.Equal(t, \"my-assistant\", setting.Uses.Vision)\n\t\tassert.Equal(t, \"my-assistant\", setting.Uses.Audio)\n\t\tassert.Equal(t, \"my-assistant\", setting.Uses.Search)\n\t\tassert.Equal(t, \"my-assistant\", setting.Uses.Fetch)\n\t\tassert.Equal(t, \"my-assistant\", setting.Uses.Web)\n\t\tassert.Equal(t, \"my-assistant\", setting.Uses.Keyword)\n\t\tassert.Equal(t, \"my-assistant\", setting.Uses.QueryDSL)\n\t\tassert.Equal(t, \"my-assistant\", setting.Uses.Rerank)\n\t})\n\n\tt.Run(\"CacheField\", func(t *testing.T) {\n\t\tsetting := &types.DSL{Cache: \"$ENV.TEST_CACHE\"}\n\t\tresolveEnvStrings(setting)\n\t\tassert.Equal(t, \"my-cache\", setting.Cache)\n\t})\n\n\tt.Run(\"PlainStringsUnchanged\", func(t *testing.T) {\n\t\tsetting := &types.DSL{\n\t\t\tCache: \"plain-cache\",\n\t\t\tSystem: &types.System{\n\t\t\t\tDefault: \"openai.gpt-5\",\n\t\t\t},\n\t\t\tUses: &types.Uses{\n\t\t\t\tDefault: \"mohe\",\n\t\t\t\tTitle: \"__yao.title\",\n\t\t\t},\n\t\t}\n\t\tresolveEnvStrings(setting)\n\n\t\tassert.Equal(t, \"plain-cache\", setting.Cache)\n\t\tassert.Equal(t, \"openai.gpt-5\", setting.System.Default)\n\t\tassert.Equal(t, \"mohe\", setting.Uses.Default)\n\t\tassert.Equal(t, \"__yao.title\", setting.Uses.Title)\n\t})\n\n\tt.Run(\"NilSystemAndUses\", func(t *testing.T) {\n\t\tsetting := &types.DSL{Cache: \"test\"}\n\t\tassert.NotPanics(t, func() {\n\t\t\tresolveEnvStrings(setting)\n\t\t})\n\t})\n\n\tt.Run(\"UndefinedEnvReturnsEmpty\", func(t *testing.T) {\n\t\tsetting := &types.DSL{\n\t\t\tSystem: &types.System{\n\t\t\t\tDefault: \"$ENV.UNDEFINED_VAR_12345\",\n\t\t\t},\n\t\t}\n\t\tresolveEnvStrings(setting)\n\t\tassert.Equal(t, \"\", setting.System.Default)\n\t})\n}\n\nfunc TestGlobalPromptsContent(t *testing.T) {\n\tprepare(t)\n\tdefer test.Clean()\n\n\tagent := GetAgent()\n\trequire.NotNil(t, agent)\n\trequire.NotNil(t, agent.GlobalPrompts)\n\trequire.Greater(t, len(agent.GlobalPrompts), 0)\n\n\tt.Run(\"SystemContextPrompt\", func(t *testing.T) {\n\t\t// Find system prompt\n\t\tvar systemPrompt string\n\t\tfor _, p := range agent.GlobalPrompts {\n\t\t\tif p.Role == \"system\" {\n\t\t\t\tsystemPrompt = p.Content\n\t\t\t\tbreak\n\t\t\t}\n\t\t}\n\n\t\tassert.NotEmpty(t, systemPrompt)\n\t\tassert.Contains(t, systemPrompt, \"System Context\")\n\t})\n\n\tt.Run(\"VariablesInRawPrompts\", func(t *testing.T) {\n\t\t// Raw prompts should contain unparsed variables\n\t\tcontent := agent.GlobalPrompts[0].Content\n\t\tassert.True(t,\n\t\t\tstrings.Contains(content, \"$SYS.\") ||\n\t\t\t\tstrings.Contains(content, \"$ENV.\") ||\n\t\t\t\tstrings.Contains(content, \"$CTX.\"),\n\t\t\t\"Raw prompts should contain variable placeholders\")\n\t})\n}\n\nfunc TestAssistantGlobalPrompts(t *testing.T) {\n\tprepare(t)\n\tdefer test.Clean()\n\n\tt.Run(\"AssistantModuleReceivesGlobalPrompts\", func(t *testing.T) {\n\t\t// Verify assistant module has global prompts\n\t\tprompts := assistant.GetGlobalPrompts(nil)\n\t\trequire.NotNil(t, prompts)\n\t\trequire.Greater(t, len(prompts), 0)\n\n\t\t// Should be parsed (no $SYS.* variables)\n\t\tcontent := prompts[0].Content\n\t\tassert.NotContains(t, content, \"$SYS.DATETIME\")\n\t})\n\n\tt.Run(\"AssistantModuleParsesWithContext\", func(t *testing.T) {\n\t\tctx := map[string]string{\n\t\t\t\"USER_ID\": \"assistant-test-user\",\n\t\t\t\"LOCALE\": \"en-US\",\n\t\t}\n\n\t\tprompts := assistant.GetGlobalPrompts(ctx)\n\t\trequire.NotNil(t, prompts)\n\n\t\t// $SYS.* should be replaced\n\t\tcontent := prompts[0].Content\n\t\tassert.NotContains(t, content, \"$SYS.\")\n\n\t\t// Should contain current time info\n\t\tnow := time.Now()\n\t\tassert.Contains(t, content, now.Format(\"2006-01-02\"))\n\t})\n}\n" }, { "path": "agent/memory/interfaces.go", "content": "package memory\n\nimport \"github.com/yaoapp/gou/store\"\n\n// Manager defines the interface for managing agent memory\ntype Manager interface {\n\t// Memory returns the memory instance for given identifiers\n\tMemory(userID, teamID, chatID, contextID string) (*Memory, error)\n\n\t// Close closes all stores and releases resources\n\tClose() error\n}\n\n// Accessor defines the interface for accessing memory from agent context\n// This is the primary interface used by agent hooks and tools\ntype Accessor interface {\n\t// User returns the user-level memory namespace\n\tUser() NamespaceAccessor\n\n\t// Team returns the team-level memory namespace\n\tTeam() NamespaceAccessor\n\n\t// Chat returns the chat-level memory namespace\n\tChat() NamespaceAccessor\n\n\t// Context returns the context-level memory namespace\n\tContext() NamespaceAccessor\n\n\t// Space returns a memory namespace by space type\n\tSpace(space Space) NamespaceAccessor\n\n\t// Stats returns memory statistics\n\tStats() *Stats\n}\n\n// NamespaceAccessor defines the interface for accessing a single memory namespace\n// Embeds store.Store for all KV and list operations\ntype NamespaceAccessor interface {\n\tstore.Store\n\n\t// GetID returns the namespace identifier (user_id, team_id, chat_id, or context_id)\n\tGetID() string\n\n\t// GetSpace returns the space type of this namespace\n\tGetSpace() Space\n\n\t// Stats returns statistics for this namespace\n\tStats() *NamespaceStats\n}\n\n// Factory defines the interface for creating memory instances\ntype Factory interface {\n\t// Create creates a new memory instance with the given configuration\n\tCreate(config *Config) (Manager, error)\n\n\t// CreateWithDefaults creates a new memory instance with default configuration\n\tCreateWithDefaults() (Manager, error)\n}\n" }, { "path": "agent/memory/manager.go", "content": "package memory\n\nimport (\n\t\"sync\"\n)\n\n// Global manager instance\nvar globalManager Manager\n\n// Init initializes the global memory manager with the given configuration\n// Called by agent.Load() after loading agent DSL\nfunc Init(config *Config) {\n\tglobalManager = NewManager(config)\n}\n\n// GetMemory returns a memory instance for the given identifiers using the global manager\n// This is the main entry point for creating Memory instances from agent/context\nfunc GetMemory(userID, teamID, chatID, contextID string) (*Memory, error) {\n\tif globalManager == nil {\n\t\t// Initialize with defaults if not configured\n\t\tglobalManager = NewManagerWithDefaults()\n\t}\n\treturn globalManager.Memory(userID, teamID, chatID, contextID)\n}\n\n// Close closes the global manager and releases resources\nfunc Close() error {\n\tif globalManager != nil {\n\t\terr := globalManager.Close()\n\t\tglobalManager = nil\n\t\treturn err\n\t}\n\treturn nil\n}\n\n// DefaultManager is the default memory manager implementation\ntype DefaultManager struct {\n\tconfig *Config\n\tmemories sync.Map // map[string]*Memory, key is composite of userID:teamID:chatID:contextID\n}\n\n// NewManager creates a new memory manager with the given configuration\nfunc NewManager(config *Config) Manager {\n\tif config == nil {\n\t\tconfig = &Config{}\n\t}\n\treturn &DefaultManager{\n\t\tconfig: config,\n\t}\n}\n\n// NewManagerWithDefaults creates a new memory manager with default configuration\nfunc NewManagerWithDefaults() Manager {\n\treturn NewManager(&Config{\n\t\tUser: DefaultUserStore,\n\t\tTeam: DefaultTeamStore,\n\t\tChat: DefaultChatStore,\n\t\tContext: DefaultContextStore,\n\t})\n}\n\n// memoryKey generates a unique key for the memory instance\nfunc memoryKey(userID, teamID, chatID, contextID string) string {\n\treturn userID + \":\" + teamID + \":\" + chatID + \":\" + contextID\n}\n\n// Memory returns the memory instance for given identifiers\nfunc (m *DefaultManager) Memory(userID, teamID, chatID, contextID string) (*Memory, error) {\n\tkey := memoryKey(userID, teamID, chatID, contextID)\n\n\t// Check if memory already exists\n\tif val, ok := m.memories.Load(key); ok {\n\t\treturn val.(*Memory), nil\n\t}\n\n\t// Create new memory instance\n\tmem, err := New(m.config, userID, teamID, chatID, contextID)\n\tif err != nil {\n\t\treturn nil, err\n\t}\n\n\t// Store and return (use LoadOrStore for thread safety)\n\tactual, _ := m.memories.LoadOrStore(key, mem)\n\treturn actual.(*Memory), nil\n}\n\n// Close closes all stores and releases resources\nfunc (m *DefaultManager) Close() error {\n\t// Clear all cached memory instances\n\tm.memories.Range(func(key, value interface{}) bool {\n\t\tm.memories.Delete(key)\n\t\treturn true\n\t})\n\treturn nil\n}\n\n// Ensure DefaultManager implements Manager\nvar _ Manager = (*DefaultManager)(nil)\n" }, { "path": "agent/memory/memory.go", "content": "package memory\n\nimport (\n\t\"fmt\"\n\t\"time\"\n\n\t\"github.com/yaoapp/gou/store\"\n)\n\n// Default TTL values for each memory space\nconst (\n\tDefaultUserTTL = 0 // No expiration for user-level memory\n\tDefaultTeamTTL = 0 // No expiration for team-level memory\n\tDefaultChatTTL = 24 * time.Hour // 24 hours for chat-level memory\n\tDefaultContextTTL = 30 * time.Minute // 30 minutes for context-level memory\n)\n\n// New creates a new Memory instance with the given configuration and identifiers\nfunc New(cfg *Config, userID, teamID, chatID, contextID string) (*Memory, error) {\n\tif cfg == nil {\n\t\tcfg = &Config{}\n\t}\n\n\tm := &Memory{\n\t\tUserID: userID,\n\t\tTeamID: teamID,\n\t\tChatID: chatID,\n\t\tContextID: contextID,\n\t\tConfig: cfg,\n\t}\n\n\t// Initialize user namespace\n\tif userID != \"\" {\n\t\tns, err := newNamespace(SpaceUser, userID, cfg.User, DefaultUserTTL)\n\t\tif err != nil {\n\t\t\treturn nil, fmt.Errorf(\"failed to create user namespace: %w\", err)\n\t\t}\n\t\tm.User = ns\n\t}\n\n\t// Initialize team namespace\n\tif teamID != \"\" {\n\t\tns, err := newNamespace(SpaceTeam, teamID, cfg.Team, DefaultTeamTTL)\n\t\tif err != nil {\n\t\t\treturn nil, fmt.Errorf(\"failed to create team namespace: %w\", err)\n\t\t}\n\t\tm.Team = ns\n\t}\n\n\t// Initialize chat namespace\n\tif chatID != \"\" {\n\t\tns, err := newNamespace(SpaceChat, chatID, cfg.Chat, DefaultChatTTL)\n\t\tif err != nil {\n\t\t\treturn nil, fmt.Errorf(\"failed to create chat namespace: %w\", err)\n\t\t}\n\t\tm.Chat = ns\n\t}\n\n\t// Initialize context namespace\n\tif contextID != \"\" {\n\t\tns, err := newNamespace(SpaceContext, contextID, cfg.Context, DefaultContextTTL)\n\t\tif err != nil {\n\t\t\treturn nil, fmt.Errorf(\"failed to create context namespace: %w\", err)\n\t\t}\n\t\tm.Context = ns\n\t}\n\n\treturn m, nil\n}\n\n// newNamespace creates a new Namespace with the given parameters\nfunc newNamespace(space Space, id, storeID string, defaultTTL time.Duration) (*Namespace, error) {\n\t// Use default store ID if not specified\n\tif storeID == \"\" {\n\t\tswitch space {\n\t\tcase SpaceUser:\n\t\t\tstoreID = DefaultUserStore\n\t\tcase SpaceTeam:\n\t\t\tstoreID = DefaultTeamStore\n\t\tcase SpaceChat:\n\t\t\tstoreID = DefaultChatStore\n\t\tcase SpaceContext:\n\t\t\tstoreID = DefaultContextStore\n\t\t}\n\t}\n\n\t// Get store instance\n\ts, err := store.Get(storeID)\n\tif err != nil {\n\t\treturn nil, fmt.Errorf(\"failed to get store %s: %w\", storeID, err)\n\t}\n\n\treturn &Namespace{\n\t\tSpace: space,\n\t\tID: id,\n\t\tStore: s,\n\t\tStoreID: storeID,\n\t\tPrefix: fmt.Sprintf(\"%s:%s:\", space, id),\n\t\tDefault: defaultTTL,\n\t}, nil\n}\n\n// GetUser returns the user-level memory namespace accessor\nfunc (m *Memory) GetUser() NamespaceAccessor {\n\tif m.User == nil {\n\t\treturn nil\n\t}\n\treturn m.User\n}\n\n// GetTeam returns the team-level memory namespace accessor\nfunc (m *Memory) GetTeam() NamespaceAccessor {\n\tif m.Team == nil {\n\t\treturn nil\n\t}\n\treturn m.Team\n}\n\n// GetChat returns the chat-level memory namespace accessor\nfunc (m *Memory) GetChat() NamespaceAccessor {\n\tif m.Chat == nil {\n\t\treturn nil\n\t}\n\treturn m.Chat\n}\n\n// GetContext returns the context-level memory namespace accessor\nfunc (m *Memory) GetContext() NamespaceAccessor {\n\tif m.Context == nil {\n\t\treturn nil\n\t}\n\treturn m.Context\n}\n\n// GetSpace returns a memory namespace by space type\nfunc (m *Memory) GetSpace(space Space) NamespaceAccessor {\n\tswitch space {\n\tcase SpaceUser:\n\t\treturn m.GetUser()\n\tcase SpaceTeam:\n\t\treturn m.GetTeam()\n\tcase SpaceChat:\n\t\treturn m.GetChat()\n\tcase SpaceContext:\n\t\treturn m.GetContext()\n\tdefault:\n\t\treturn nil\n\t}\n}\n\n// GetStats returns memory statistics for all namespaces\nfunc (m *Memory) GetStats() *Stats {\n\tstats := &Stats{}\n\n\tif m.User != nil {\n\t\tstats.User = m.User.Stats()\n\t}\n\tif m.Team != nil {\n\t\tstats.Team = m.Team.Stats()\n\t}\n\tif m.Chat != nil {\n\t\tstats.Chat = m.Chat.Stats()\n\t}\n\tif m.Context != nil {\n\t\tstats.Context = m.Context.Stats()\n\t}\n\n\treturn stats\n}\n\n// Clear clears all memory in all namespaces for this memory instance\nfunc (m *Memory) Clear() {\n\tif m.User != nil {\n\t\tm.User.Clear()\n\t}\n\tif m.Team != nil {\n\t\tm.Team.Clear()\n\t}\n\tif m.Chat != nil {\n\t\tm.Chat.Clear()\n\t}\n\tif m.Context != nil {\n\t\tm.Context.Clear()\n\t}\n}\n\n// Fork creates a new Memory instance with an independent Context namespace\n// but sharing the User, Team, and Chat namespaces with the parent.\n// This is used for parallel agent calls (ctx.agent.All/Any/Race) to prevent\n// context state from being shared between concurrent sub-agent executions.\n//\n// The new Context namespace uses the provided newContextID.\n// If newContextID is empty, returns a shallow copy with shared Context.\nfunc (m *Memory) Fork(newContextID string) (*Memory, error) {\n\tif m == nil {\n\t\treturn nil, nil\n\t}\n\n\t// If no new context ID provided, share everything (shallow copy)\n\tif newContextID == \"\" {\n\t\treturn &Memory{\n\t\t\tUserID: m.UserID,\n\t\t\tTeamID: m.TeamID,\n\t\t\tChatID: m.ChatID,\n\t\t\tContextID: m.ContextID,\n\t\t\tUser: m.User,\n\t\t\tTeam: m.Team,\n\t\t\tChat: m.Chat,\n\t\t\tContext: m.Context,\n\t\t\tConfig: m.Config,\n\t\t}, nil\n\t}\n\n\t// Create new Memory with independent Context namespace\n\tforked := &Memory{\n\t\tUserID: m.UserID,\n\t\tTeamID: m.TeamID,\n\t\tChatID: m.ChatID,\n\t\tContextID: newContextID,\n\t\tUser: m.User, // Shared\n\t\tTeam: m.Team, // Shared\n\t\tChat: m.Chat, // Shared\n\t\tContext: nil, // Will be created below\n\t\tConfig: m.Config,\n\t}\n\n\t// Create new Context namespace with independent ID\n\tstoreID := \"\"\n\tif m.Config != nil {\n\t\tstoreID = m.Config.Context\n\t}\n\tns, err := newNamespace(SpaceContext, newContextID, storeID, DefaultContextTTL)\n\tif err != nil {\n\t\treturn nil, fmt.Errorf(\"failed to create forked context namespace: %w\", err)\n\t}\n\tforked.Context = ns\n\n\treturn forked, nil\n}\n" }, { "path": "agent/memory/memory_test.go", "content": "package memory_test\n\nimport (\n\t\"testing\"\n\t\"time\"\n\n\t\"github.com/stretchr/testify/assert\"\n\t\"github.com/stretchr/testify/require\"\n\t\"github.com/yaoapp/yao/agent/memory\"\n\t\"github.com/yaoapp/yao/config\"\n\t\"github.com/yaoapp/yao/test\"\n)\n\nfunc TestMemoryNew(t *testing.T) {\n\ttest.Prepare(t, config.Conf)\n\tdefer test.Clean()\n\n\t// Create memory with default stores\n\tmem, err := memory.New(nil, \"user1\", \"team1\", \"chat1\", \"ctx1\")\n\trequire.NoError(t, err)\n\trequire.NotNil(t, mem)\n\n\t// Verify all namespaces are initialized\n\tassert.NotNil(t, mem.User)\n\tassert.NotNil(t, mem.Team)\n\tassert.NotNil(t, mem.Chat)\n\tassert.NotNil(t, mem.Context)\n\n\t// Verify IDs\n\tassert.Equal(t, \"user1\", mem.UserID)\n\tassert.Equal(t, \"team1\", mem.TeamID)\n\tassert.Equal(t, \"chat1\", mem.ChatID)\n\tassert.Equal(t, \"ctx1\", mem.ContextID)\n}\n\nfunc TestMemoryPartialIDs(t *testing.T) {\n\ttest.Prepare(t, config.Conf)\n\tdefer test.Clean()\n\n\t// Create memory with only user and chat\n\tmem, err := memory.New(nil, \"user1\", \"\", \"chat1\", \"\")\n\trequire.NoError(t, err)\n\trequire.NotNil(t, mem)\n\n\t// Only user and chat namespaces should be initialized\n\tassert.NotNil(t, mem.User)\n\tassert.Nil(t, mem.Team)\n\tassert.NotNil(t, mem.Chat)\n\tassert.Nil(t, mem.Context)\n}\n\nfunc TestNamespaceBasicOperations(t *testing.T) {\n\ttest.Prepare(t, config.Conf)\n\tdefer test.Clean()\n\n\tmem, err := memory.New(nil, \"user1\", \"team1\", \"chat1\", \"ctx1\")\n\trequire.NoError(t, err)\n\n\t// Test User namespace\n\tt.Run(\"User namespace\", func(t *testing.T) {\n\t\tns := mem.GetUser()\n\t\trequire.NotNil(t, ns)\n\n\t\t// Set and Get\n\t\terr := ns.Set(\"name\", \"John\", 0)\n\t\trequire.NoError(t, err)\n\n\t\tval, ok := ns.Get(\"name\")\n\t\tassert.True(t, ok)\n\t\tassert.Equal(t, \"John\", val)\n\n\t\t// Has\n\t\tassert.True(t, ns.Has(\"name\"))\n\t\tassert.False(t, ns.Has(\"nonexistent\"))\n\n\t\t// Del\n\t\terr = ns.Del(\"name\")\n\t\trequire.NoError(t, err)\n\t\tassert.False(t, ns.Has(\"name\"))\n\t})\n\n\t// Test Team namespace\n\tt.Run(\"Team namespace\", func(t *testing.T) {\n\t\tns := mem.GetTeam()\n\t\trequire.NotNil(t, ns)\n\n\t\terr := ns.Set(\"setting\", \"value\", 0)\n\t\trequire.NoError(t, err)\n\n\t\tval, ok := ns.Get(\"setting\")\n\t\tassert.True(t, ok)\n\t\tassert.Equal(t, \"value\", val)\n\t})\n\n\t// Test Chat namespace\n\tt.Run(\"Chat namespace\", func(t *testing.T) {\n\t\tns := mem.GetChat()\n\t\trequire.NotNil(t, ns)\n\n\t\terr := ns.Set(\"topic\", \"AI\", 0)\n\t\trequire.NoError(t, err)\n\n\t\tval, ok := ns.Get(\"topic\")\n\t\tassert.True(t, ok)\n\t\tassert.Equal(t, \"AI\", val)\n\t})\n\n\t// Test Context namespace\n\tt.Run(\"Context namespace\", func(t *testing.T) {\n\t\tns := mem.GetContext()\n\t\trequire.NotNil(t, ns)\n\n\t\terr := ns.Set(\"temp\", \"data\", 0)\n\t\trequire.NoError(t, err)\n\n\t\tval, ok := ns.Get(\"temp\")\n\t\tassert.True(t, ok)\n\t\tassert.Equal(t, \"data\", val)\n\t})\n}\n\nfunc TestNamespaceIsolation(t *testing.T) {\n\ttest.Prepare(t, config.Conf)\n\tdefer test.Clean()\n\n\tt.Run(\"User isolation\", func(t *testing.T) {\n\t\t// Create two memory instances with different user IDs\n\t\tmem1, err := memory.New(nil, \"user1\", \"\", \"\", \"\")\n\t\trequire.NoError(t, err)\n\n\t\tmem2, err := memory.New(nil, \"user2\", \"\", \"\", \"\")\n\t\trequire.NoError(t, err)\n\n\t\t// Set value in user1's namespace\n\t\terr = mem1.GetUser().Set(\"key\", \"user1_value\", 0)\n\t\trequire.NoError(t, err)\n\n\t\t// Set value in user2's namespace\n\t\terr = mem2.GetUser().Set(\"key\", \"user2_value\", 0)\n\t\trequire.NoError(t, err)\n\n\t\t// Verify isolation - each user sees their own value\n\t\tval1, ok := mem1.GetUser().Get(\"key\")\n\t\tassert.True(t, ok)\n\t\tassert.Equal(t, \"user1_value\", val1)\n\n\t\tval2, ok := mem2.GetUser().Get(\"key\")\n\t\tassert.True(t, ok)\n\t\tassert.Equal(t, \"user2_value\", val2)\n\n\t\t// Delete from user1 should not affect user2\n\t\terr = mem1.GetUser().Del(\"key\")\n\t\trequire.NoError(t, err)\n\n\t\t_, ok = mem1.GetUser().Get(\"key\")\n\t\tassert.False(t, ok, \"user1's key should be deleted\")\n\n\t\tval2, ok = mem2.GetUser().Get(\"key\")\n\t\tassert.True(t, ok, \"user2's key should still exist\")\n\t\tassert.Equal(t, \"user2_value\", val2)\n\n\t\t// Clear user1 should not affect user2\n\t\tmem1.GetUser().Clear()\n\t\tval2, ok = mem2.GetUser().Get(\"key\")\n\t\tassert.True(t, ok, \"user2's key should still exist after user1 clear\")\n\t\tassert.Equal(t, \"user2_value\", val2)\n\t})\n\n\tt.Run(\"Team isolation\", func(t *testing.T) {\n\t\tmemA, err := memory.New(nil, \"\", \"teamA\", \"\", \"\")\n\t\trequire.NoError(t, err)\n\n\t\tmemB, err := memory.New(nil, \"\", \"teamB\", \"\", \"\")\n\t\trequire.NoError(t, err)\n\n\t\t// Set same key in different teams\n\t\tmemA.GetTeam().Set(\"config\", \"teamA_config\", 0)\n\t\tmemB.GetTeam().Set(\"config\", \"teamB_config\", 0)\n\n\t\t// Verify isolation\n\t\tvalA, ok := memA.GetTeam().Get(\"config\")\n\t\tassert.True(t, ok)\n\t\tassert.Equal(t, \"teamA_config\", valA)\n\n\t\tvalB, ok := memB.GetTeam().Get(\"config\")\n\t\tassert.True(t, ok)\n\t\tassert.Equal(t, \"teamB_config\", valB)\n\t})\n\n\tt.Run(\"Chat isolation\", func(t *testing.T) {\n\t\tmem1, err := memory.New(nil, \"\", \"\", \"chat1\", \"\")\n\t\trequire.NoError(t, err)\n\n\t\tmem2, err := memory.New(nil, \"\", \"\", \"chat2\", \"\")\n\t\trequire.NoError(t, err)\n\n\t\t// Set same key in different chats\n\t\tmem1.GetChat().Set(\"topic\", \"chat1_topic\", 0)\n\t\tmem2.GetChat().Set(\"topic\", \"chat2_topic\", 0)\n\n\t\t// Verify isolation\n\t\tval1, ok := mem1.GetChat().Get(\"topic\")\n\t\tassert.True(t, ok)\n\t\tassert.Equal(t, \"chat1_topic\", val1)\n\n\t\tval2, ok := mem2.GetChat().Get(\"topic\")\n\t\tassert.True(t, ok)\n\t\tassert.Equal(t, \"chat2_topic\", val2)\n\t})\n\n\tt.Run(\"Context isolation\", func(t *testing.T) {\n\t\tmem1, err := memory.New(nil, \"\", \"\", \"\", \"ctx1\")\n\t\trequire.NoError(t, err)\n\n\t\tmem2, err := memory.New(nil, \"\", \"\", \"\", \"ctx2\")\n\t\trequire.NoError(t, err)\n\n\t\t// Set same key in different contexts\n\t\tmem1.GetContext().Set(\"temp\", \"ctx1_temp\", 0)\n\t\tmem2.GetContext().Set(\"temp\", \"ctx2_temp\", 0)\n\n\t\t// Verify isolation\n\t\tval1, ok := mem1.GetContext().Get(\"temp\")\n\t\tassert.True(t, ok)\n\t\tassert.Equal(t, \"ctx1_temp\", val1)\n\n\t\tval2, ok := mem2.GetContext().Get(\"temp\")\n\t\tassert.True(t, ok)\n\t\tassert.Equal(t, \"ctx2_temp\", val2)\n\t})\n\n\tt.Run(\"Keys and Len isolation\", func(t *testing.T) {\n\t\tmem1, err := memory.New(nil, \"userA\", \"\", \"\", \"\")\n\t\trequire.NoError(t, err)\n\n\t\tmem2, err := memory.New(nil, \"userB\", \"\", \"\", \"\")\n\t\trequire.NoError(t, err)\n\n\t\t// Clear first\n\t\tmem1.GetUser().Clear()\n\t\tmem2.GetUser().Clear()\n\n\t\t// Set keys in userA\n\t\tmem1.GetUser().Set(\"a\", 1, 0)\n\t\tmem1.GetUser().Set(\"b\", 2, 0)\n\t\tmem1.GetUser().Set(\"c\", 3, 0)\n\n\t\t// Set keys in userB\n\t\tmem2.GetUser().Set(\"x\", 10, 0)\n\t\tmem2.GetUser().Set(\"y\", 20, 0)\n\n\t\t// Verify Keys isolation\n\t\tkeys1 := mem1.GetUser().Keys()\n\t\tassert.Equal(t, 3, len(keys1), \"userA should have 3 keys\")\n\n\t\tkeys2 := mem2.GetUser().Keys()\n\t\tassert.Equal(t, 2, len(keys2), \"userB should have 2 keys\")\n\n\t\t// Verify Len isolation\n\t\tassert.Equal(t, 3, mem1.GetUser().Len(), \"userA Len should be 3\")\n\t\tassert.Equal(t, 2, mem2.GetUser().Len(), \"userB Len should be 2\")\n\n\t\t// Keys should not contain prefix\n\t\tfor _, k := range keys1 {\n\t\t\tassert.NotContains(t, k, \"user:\", \"Key should not contain prefix\")\n\t\t}\n\t})\n\n\tt.Run(\"Incr/Decr isolation\", func(t *testing.T) {\n\t\tmem1, err := memory.New(nil, \"userX\", \"\", \"\", \"\")\n\t\trequire.NoError(t, err)\n\n\t\tmem2, err := memory.New(nil, \"userY\", \"\", \"\", \"\")\n\t\trequire.NoError(t, err)\n\n\t\t// Incr counter in userX\n\t\tval1, err := mem1.GetUser().Incr(\"counter\", 10)\n\t\trequire.NoError(t, err)\n\t\tassert.Equal(t, int64(10), val1)\n\n\t\t// Incr counter in userY\n\t\tval2, err := mem2.GetUser().Incr(\"counter\", 5)\n\t\trequire.NoError(t, err)\n\t\tassert.Equal(t, int64(5), val2)\n\n\t\t// Incr again - should be independent\n\t\tval1, err = mem1.GetUser().Incr(\"counter\", 1)\n\t\trequire.NoError(t, err)\n\t\tassert.Equal(t, int64(11), val1)\n\n\t\tval2, err = mem2.GetUser().Incr(\"counter\", 1)\n\t\trequire.NoError(t, err)\n\t\tassert.Equal(t, int64(6), val2)\n\t})\n\n\tt.Run(\"List operations isolation\", func(t *testing.T) {\n\t\tmem1, err := memory.New(nil, \"listUser1\", \"\", \"\", \"\")\n\t\trequire.NoError(t, err)\n\n\t\tmem2, err := memory.New(nil, \"listUser2\", \"\", \"\", \"\")\n\t\trequire.NoError(t, err)\n\n\t\t// Push to user1's list\n\t\terr = mem1.GetUser().Push(\"items\", \"a\", \"b\", \"c\")\n\t\trequire.NoError(t, err)\n\n\t\t// Push to user2's list\n\t\terr = mem2.GetUser().Push(\"items\", \"x\", \"y\")\n\t\trequire.NoError(t, err)\n\n\t\t// Verify isolation\n\t\tassert.Equal(t, 3, mem1.GetUser().ArrayLen(\"items\"))\n\t\tassert.Equal(t, 2, mem2.GetUser().ArrayLen(\"items\"))\n\n\t\tall1, _ := mem1.GetUser().ArrayAll(\"items\")\n\t\tall2, _ := mem2.GetUser().ArrayAll(\"items\")\n\n\t\tassert.Equal(t, 3, len(all1))\n\t\tassert.Equal(t, 2, len(all2))\n\n\t\t// Pop from user1 should not affect user2\n\t\tmem1.GetUser().Pop(\"items\", 1)\n\t\tassert.Equal(t, 2, mem1.GetUser().ArrayLen(\"items\"))\n\t\tassert.Equal(t, 2, mem2.GetUser().ArrayLen(\"items\"))\n\t})\n\n\tt.Run(\"Del pattern isolation\", func(t *testing.T) {\n\t\tmem1, err := memory.New(nil, \"patternUser1\", \"\", \"\", \"\")\n\t\trequire.NoError(t, err)\n\n\t\tmem2, err := memory.New(nil, \"patternUser2\", \"\", \"\", \"\")\n\t\trequire.NoError(t, err)\n\n\t\t// Set keys with pattern in both users\n\t\tmem1.GetUser().Set(\"file:1\", \"data1\", 0)\n\t\tmem1.GetUser().Set(\"file:2\", \"data2\", 0)\n\t\tmem1.GetUser().Set(\"other\", \"other1\", 0)\n\n\t\tmem2.GetUser().Set(\"file:1\", \"data1\", 0)\n\t\tmem2.GetUser().Set(\"file:2\", \"data2\", 0)\n\t\tmem2.GetUser().Set(\"other\", \"other2\", 0)\n\n\t\t// Delete pattern from user1\n\t\terr = mem1.GetUser().Del(\"file:*\")\n\t\trequire.NoError(t, err)\n\n\t\t// user1's file:* keys should be deleted\n\t\tassert.False(t, mem1.GetUser().Has(\"file:1\"))\n\t\tassert.False(t, mem1.GetUser().Has(\"file:2\"))\n\t\tassert.True(t, mem1.GetUser().Has(\"other\"))\n\n\t\t// user2's keys should be unaffected\n\t\tassert.True(t, mem2.GetUser().Has(\"file:1\"))\n\t\tassert.True(t, mem2.GetUser().Has(\"file:2\"))\n\t\tassert.True(t, mem2.GetUser().Has(\"other\"))\n\t})\n}\n\nfunc TestNamespaceIncrDecr(t *testing.T) {\n\ttest.Prepare(t, config.Conf)\n\tdefer test.Clean()\n\n\tmem, err := memory.New(nil, \"user1\", \"\", \"\", \"\")\n\trequire.NoError(t, err)\n\n\tns := mem.GetUser()\n\n\t// Incr on non-existent key\n\tval, err := ns.Incr(\"counter\", 1)\n\trequire.NoError(t, err)\n\tassert.Equal(t, int64(1), val)\n\n\t// Incr again\n\tval, err = ns.Incr(\"counter\", 5)\n\trequire.NoError(t, err)\n\tassert.Equal(t, int64(6), val)\n\n\t// Decr\n\tval, err = ns.Decr(\"counter\", 2)\n\trequire.NoError(t, err)\n\tassert.Equal(t, int64(4), val)\n}\n\nfunc TestNamespaceListOperations(t *testing.T) {\n\ttest.Prepare(t, config.Conf)\n\tdefer test.Clean()\n\n\tmem, err := memory.New(nil, \"user1\", \"\", \"\", \"\")\n\trequire.NoError(t, err)\n\n\tns := mem.GetUser()\n\n\t// Push values\n\terr = ns.Push(\"list\", \"a\", \"b\", \"c\")\n\trequire.NoError(t, err)\n\n\t// ArrayLen\n\tassert.Equal(t, 3, ns.ArrayLen(\"list\"))\n\n\t// ArrayAll\n\tall, err := ns.ArrayAll(\"list\")\n\trequire.NoError(t, err)\n\tassert.Len(t, all, 3)\n\n\t// Pop from end\n\tval, err := ns.Pop(\"list\", 1)\n\trequire.NoError(t, err)\n\tassert.Equal(t, \"c\", val)\n\n\t// ArrayLen after pop\n\tassert.Equal(t, 2, ns.ArrayLen(\"list\"))\n}\n\nfunc TestNamespaceSetOperations(t *testing.T) {\n\ttest.Prepare(t, config.Conf)\n\tdefer test.Clean()\n\n\tmem, err := memory.New(nil, \"user1\", \"\", \"\", \"\")\n\trequire.NoError(t, err)\n\n\tns := mem.GetUser()\n\n\t// AddToSet\n\terr = ns.AddToSet(\"tags\", \"go\", \"rust\", \"go\") // \"go\" should only appear once\n\trequire.NoError(t, err)\n\n\tall, err := ns.ArrayAll(\"tags\")\n\trequire.NoError(t, err)\n\tassert.Len(t, all, 2) // Only \"go\" and \"rust\"\n}\n\nfunc TestNamespaceTTL(t *testing.T) {\n\ttest.Prepare(t, config.Conf)\n\tdefer test.Clean()\n\n\tmem, err := memory.New(nil, \"\", \"\", \"\", \"ctx1\")\n\trequire.NoError(t, err)\n\n\tns := mem.GetContext()\n\n\t// Set with short TTL\n\terr = ns.Set(\"temp\", \"value\", 100*time.Millisecond)\n\trequire.NoError(t, err)\n\n\t// Should exist immediately\n\tval, ok := ns.Get(\"temp\")\n\tassert.True(t, ok)\n\tassert.Equal(t, \"value\", val)\n\n\t// Wait for expiration\n\ttime.Sleep(150 * time.Millisecond)\n\n\t// Should be expired\n\t_, ok = ns.Get(\"temp\")\n\tassert.False(t, ok)\n}\n\nfunc TestMemoryClear(t *testing.T) {\n\ttest.Prepare(t, config.Conf)\n\tdefer test.Clean()\n\n\tmem, err := memory.New(nil, \"user1\", \"team1\", \"chat1\", \"ctx1\")\n\trequire.NoError(t, err)\n\n\t// Set values in all namespaces\n\tmem.GetUser().Set(\"key\", \"user_value\", 0)\n\tmem.GetTeam().Set(\"key\", \"team_value\", 0)\n\tmem.GetChat().Set(\"key\", \"chat_value\", 0)\n\tmem.GetContext().Set(\"key\", \"ctx_value\", 0)\n\n\t// Clear all\n\tmem.Clear()\n\n\t// All should be empty\n\t_, ok := mem.GetUser().Get(\"key\")\n\tassert.False(t, ok)\n\t_, ok = mem.GetTeam().Get(\"key\")\n\tassert.False(t, ok)\n\t_, ok = mem.GetChat().Get(\"key\")\n\tassert.False(t, ok)\n\t_, ok = mem.GetContext().Get(\"key\")\n\tassert.False(t, ok)\n}\n\nfunc TestMemoryStats(t *testing.T) {\n\ttest.Prepare(t, config.Conf)\n\tdefer test.Clean()\n\n\tmem, err := memory.New(nil, \"user1\", \"team1\", \"chat1\", \"ctx1\")\n\trequire.NoError(t, err)\n\n\t// Set some values\n\tmem.GetUser().Set(\"k1\", \"v1\", 0)\n\tmem.GetUser().Set(\"k2\", \"v2\", 0)\n\tmem.GetTeam().Set(\"k1\", \"v1\", 0)\n\n\tstats := mem.GetStats()\n\trequire.NotNil(t, stats)\n\n\tassert.Equal(t, 2, stats.User.KeyCount)\n\tassert.Equal(t, 1, stats.Team.KeyCount)\n\tassert.Equal(t, 0, stats.Chat.KeyCount)\n\tassert.Equal(t, 0, stats.Context.KeyCount)\n}\n\nfunc TestManager(t *testing.T) {\n\ttest.Prepare(t, config.Conf)\n\tdefer test.Clean()\n\n\tmgr := memory.NewManagerWithDefaults()\n\tdefer mgr.Close()\n\n\t// Get memory instance\n\tmem1, err := mgr.Memory(\"user1\", \"team1\", \"chat1\", \"ctx1\")\n\trequire.NoError(t, err)\n\trequire.NotNil(t, mem1)\n\n\t// Set a value\n\terr = mem1.GetUser().Set(\"key\", \"value\", 0)\n\trequire.NoError(t, err)\n\n\t// Get same memory instance again\n\tmem2, err := mgr.Memory(\"user1\", \"team1\", \"chat1\", \"ctx1\")\n\trequire.NoError(t, err)\n\n\t// Should be the same instance (cached)\n\tval, ok := mem2.GetUser().Get(\"key\")\n\tassert.True(t, ok)\n\tassert.Equal(t, \"value\", val)\n}\n\nfunc TestGetSpace(t *testing.T) {\n\ttest.Prepare(t, config.Conf)\n\tdefer test.Clean()\n\n\tmem, err := memory.New(nil, \"user1\", \"team1\", \"chat1\", \"ctx1\")\n\trequire.NoError(t, err)\n\n\t// Test GetSpace\n\tassert.NotNil(t, mem.GetSpace(memory.SpaceUser))\n\tassert.NotNil(t, mem.GetSpace(memory.SpaceTeam))\n\tassert.NotNil(t, mem.GetSpace(memory.SpaceChat))\n\tassert.NotNil(t, mem.GetSpace(memory.SpaceContext))\n\n\t// Invalid space\n\tassert.Nil(t, mem.GetSpace(memory.Space(\"invalid\")))\n}\n\nfunc TestNamespaceGetMultiSetMulti(t *testing.T) {\n\ttest.Prepare(t, config.Conf)\n\tdefer test.Clean()\n\n\tmem, err := memory.New(nil, \"user1\", \"\", \"\", \"\")\n\trequire.NoError(t, err)\n\n\tns := mem.GetUser()\n\n\t// SetMulti\n\tns.SetMulti(map[string]interface{}{\n\t\t\"a\": 1,\n\t\t\"b\": 2,\n\t\t\"c\": 3,\n\t}, 0)\n\n\t// GetMulti\n\tresult := ns.GetMulti([]string{\"a\", \"b\", \"c\"})\n\tassert.Equal(t, 1, result[\"a\"])\n\tassert.Equal(t, 2, result[\"b\"])\n\tassert.Equal(t, 3, result[\"c\"])\n\n\t// DelMulti\n\tns.DelMulti([]string{\"a\", \"b\"})\n\tassert.False(t, ns.Has(\"a\"))\n\tassert.False(t, ns.Has(\"b\"))\n\tassert.True(t, ns.Has(\"c\"))\n}\n\nfunc TestNamespaceGetDel(t *testing.T) {\n\ttest.Prepare(t, config.Conf)\n\tdefer test.Clean()\n\n\tmem, err := memory.New(nil, \"user1\", \"\", \"\", \"\")\n\trequire.NoError(t, err)\n\n\tns := mem.GetUser()\n\n\t// Set a value\n\terr = ns.Set(\"key\", \"value\", 0)\n\trequire.NoError(t, err)\n\n\t// GetDel\n\tval, ok := ns.GetDel(\"key\")\n\tassert.True(t, ok)\n\tassert.Equal(t, \"value\", val)\n\n\t// Should be deleted\n\t_, ok = ns.Get(\"key\")\n\tassert.False(t, ok)\n}\n" }, { "path": "agent/memory/namespace.go", "content": "package memory\n\nimport (\n\t\"time\"\n)\n\n// Ensure Namespace implements NamespaceAccessor\nvar _ NamespaceAccessor = (*Namespace)(nil)\n\n// GetID returns the namespace identifier\nfunc (ns *Namespace) GetID() string {\n\treturn ns.ID\n}\n\n// GetSpace returns the space type of this namespace\nfunc (ns *Namespace) GetSpace() Space {\n\treturn ns.Space\n}\n\n// prefixKey adds the namespace prefix to a key\nfunc (ns *Namespace) prefixKey(key string) string {\n\treturn ns.Prefix + key\n}\n\n// Get retrieves a value by key\nfunc (ns *Namespace) Get(key string) (interface{}, bool) {\n\treturn ns.Store.Get(ns.prefixKey(key))\n}\n\n// Set stores a value with the default TTL for this namespace\nfunc (ns *Namespace) Set(key string, value interface{}, ttl time.Duration) error {\n\tif ttl == 0 {\n\t\tttl = ns.Default\n\t}\n\treturn ns.Store.Set(ns.prefixKey(key), value, ttl)\n}\n\n// Has checks if a key exists\nfunc (ns *Namespace) Has(key string) bool {\n\treturn ns.Store.Has(ns.prefixKey(key))\n}\n\n// Del deletes a key (supports wildcards)\nfunc (ns *Namespace) Del(key string) error {\n\treturn ns.Store.Del(ns.prefixKey(key))\n}\n\n// Keys returns all keys in this namespace\n// Uses pattern-based query for efficiency\nfunc (ns *Namespace) Keys(pattern ...string) []string {\n\t// Build pattern with namespace prefix\n\tvar storePattern string\n\tif len(pattern) > 0 && pattern[0] != \"\" {\n\t\tstorePattern = ns.Prefix + pattern[0]\n\t} else {\n\t\tstorePattern = ns.Prefix + \"*\"\n\t}\n\n\tallKeys := ns.Store.Keys(storePattern)\n\tprefixLen := len(ns.Prefix)\n\n\t// Remove prefix from keys\n\tresult := make([]string, 0, len(allKeys))\n\tfor _, key := range allKeys {\n\t\tif len(key) >= prefixLen {\n\t\t\tresult = append(result, key[prefixLen:])\n\t\t}\n\t}\n\treturn result\n}\n\n// Len returns the number of keys in this namespace\n// Uses pattern-based query for efficiency\nfunc (ns *Namespace) Len(pattern ...string) int {\n\t// Build pattern with namespace prefix\n\tvar storePattern string\n\tif len(pattern) > 0 && pattern[0] != \"\" {\n\t\tstorePattern = ns.Prefix + pattern[0]\n\t} else {\n\t\tstorePattern = ns.Prefix + \"*\"\n\t}\n\n\treturn ns.Store.Len(storePattern)\n}\n\n// Clear deletes all keys in this namespace\nfunc (ns *Namespace) Clear() {\n\tns.Store.Del(ns.Prefix + \"*\")\n}\n\n// GetSet retrieves a value and sets a new value if not exists\nfunc (ns *Namespace) GetSet(key string, ttl time.Duration, getValue func(key string) (interface{}, error)) (interface{}, error) {\n\tif ttl == 0 {\n\t\tttl = ns.Default\n\t}\n\treturn ns.Store.GetSet(ns.prefixKey(key), ttl, getValue)\n}\n\n// GetDel retrieves a value and deletes it atomically\nfunc (ns *Namespace) GetDel(key string) (interface{}, bool) {\n\treturn ns.Store.GetDel(ns.prefixKey(key))\n}\n\n// GetMulti retrieves multiple values by keys\nfunc (ns *Namespace) GetMulti(keys []string) map[string]interface{} {\n\tprefixedKeys := make([]string, len(keys))\n\tfor i, key := range keys {\n\t\tprefixedKeys[i] = ns.prefixKey(key)\n\t}\n\tresult := ns.Store.GetMulti(prefixedKeys)\n\n\t// Remove prefix from result keys\n\tunprefixed := make(map[string]interface{})\n\tprefixLen := len(ns.Prefix)\n\tfor k, v := range result {\n\t\tif len(k) > prefixLen {\n\t\t\tunprefixed[k[prefixLen:]] = v\n\t\t} else {\n\t\t\tunprefixed[k] = v\n\t\t}\n\t}\n\treturn unprefixed\n}\n\n// SetMulti stores multiple values\nfunc (ns *Namespace) SetMulti(values map[string]interface{}, ttl time.Duration) {\n\tif ttl == 0 {\n\t\tttl = ns.Default\n\t}\n\tprefixed := make(map[string]interface{})\n\tfor k, v := range values {\n\t\tprefixed[ns.prefixKey(k)] = v\n\t}\n\tns.Store.SetMulti(prefixed, ttl)\n}\n\n// DelMulti deletes multiple keys\nfunc (ns *Namespace) DelMulti(keys []string) {\n\tprefixedKeys := make([]string, len(keys))\n\tfor i, key := range keys {\n\t\tprefixedKeys[i] = ns.prefixKey(key)\n\t}\n\tns.Store.DelMulti(prefixedKeys)\n}\n\n// GetSetMulti retrieves multiple values and sets new values if not exists\nfunc (ns *Namespace) GetSetMulti(keys []string, ttl time.Duration, getValue func(key string) (interface{}, error)) map[string]interface{} {\n\tif ttl == 0 {\n\t\tttl = ns.Default\n\t}\n\tprefixedKeys := make([]string, len(keys))\n\tfor i, key := range keys {\n\t\tprefixedKeys[i] = ns.prefixKey(key)\n\t}\n\tresult := ns.Store.GetSetMulti(prefixedKeys, ttl, getValue)\n\n\t// Remove prefix from result keys\n\tunprefixed := make(map[string]interface{})\n\tprefixLen := len(ns.Prefix)\n\tfor k, v := range result {\n\t\tif len(k) > prefixLen {\n\t\t\tunprefixed[k[prefixLen:]] = v\n\t\t} else {\n\t\t\tunprefixed[k] = v\n\t\t}\n\t}\n\treturn unprefixed\n}\n\n// Incr increments a numeric value\nfunc (ns *Namespace) Incr(key string, delta int64) (int64, error) {\n\treturn ns.Store.Incr(ns.prefixKey(key), delta)\n}\n\n// Decr decrements a numeric value\nfunc (ns *Namespace) Decr(key string, delta int64) (int64, error) {\n\treturn ns.Store.Decr(ns.prefixKey(key), delta)\n}\n\n// Push appends values to a list\nfunc (ns *Namespace) Push(key string, values ...interface{}) error {\n\treturn ns.Store.Push(ns.prefixKey(key), values...)\n}\n\n// Pop removes and returns an element from a list\nfunc (ns *Namespace) Pop(key string, position int) (interface{}, error) {\n\treturn ns.Store.Pop(ns.prefixKey(key), position)\n}\n\n// Pull removes the first occurrence of a value from a list\nfunc (ns *Namespace) Pull(key string, value interface{}) error {\n\treturn ns.Store.Pull(ns.prefixKey(key), value)\n}\n\n// PullAll removes all occurrences of values from a list\nfunc (ns *Namespace) PullAll(key string, values []interface{}) error {\n\treturn ns.Store.PullAll(ns.prefixKey(key), values)\n}\n\n// AddToSet adds values to a set (no duplicates)\nfunc (ns *Namespace) AddToSet(key string, values ...interface{}) error {\n\treturn ns.Store.AddToSet(ns.prefixKey(key), values...)\n}\n\n// ArrayLen returns the length of a list\nfunc (ns *Namespace) ArrayLen(key string) int {\n\treturn ns.Store.ArrayLen(ns.prefixKey(key))\n}\n\n// ArrayGet retrieves an element from a list by index\nfunc (ns *Namespace) ArrayGet(key string, index int) (interface{}, error) {\n\treturn ns.Store.ArrayGet(ns.prefixKey(key), index)\n}\n\n// ArraySet sets an element in a list by index\nfunc (ns *Namespace) ArraySet(key string, index int, value interface{}) error {\n\treturn ns.Store.ArraySet(ns.prefixKey(key), index, value)\n}\n\n// ArraySlice returns a slice of a list\nfunc (ns *Namespace) ArraySlice(key string, skip, limit int) ([]interface{}, error) {\n\treturn ns.Store.ArraySlice(ns.prefixKey(key), skip, limit)\n}\n\n// ArrayPage returns a page of a list\nfunc (ns *Namespace) ArrayPage(key string, page, pageSize int) ([]interface{}, error) {\n\treturn ns.Store.ArrayPage(ns.prefixKey(key), page, pageSize)\n}\n\n// ArrayAll returns all elements of a list\nfunc (ns *Namespace) ArrayAll(key string) ([]interface{}, error) {\n\treturn ns.Store.ArrayAll(ns.prefixKey(key))\n}\n\n// Stats returns statistics for this namespace\nfunc (ns *Namespace) Stats() *NamespaceStats {\n\treturn &NamespaceStats{\n\t\tSpace: ns.Space,\n\t\tID: ns.ID,\n\t\tKeyCount: ns.Len(),\n\t\tStoreID: ns.StoreID,\n\t}\n}\n\n// Snapshot returns all key-value pairs in this namespace\n// Used for recovery/resume functionality\nfunc (ns *Namespace) Snapshot() map[string]interface{} {\n\tkeys := ns.Keys()\n\tsnapshot := make(map[string]interface{}, len(keys))\n\tfor _, key := range keys {\n\t\tif value, ok := ns.Get(key); ok {\n\t\t\tsnapshot[key] = value\n\t\t}\n\t}\n\treturn snapshot\n}\n" }, { "path": "agent/memory/types.go", "content": "package memory\n\nimport (\n\t\"time\"\n\n\t\"github.com/yaoapp/gou/store\"\n)\n\n// Space defines the memory space type\ntype Space string\n\nconst (\n\t// SpaceUser user-level memory, persists across all chats for a user\n\t// Use case: user preferences, long-term knowledge, personal settings\n\tSpaceUser Space = \"user\"\n\n\t// SpaceTeam team-level memory, shared across all users in a team\n\t// Use case: team knowledge, shared settings, collaborative data\n\tSpaceTeam Space = \"team\"\n\n\t// SpaceChat chat-level memory, persists within a single chat session\n\t// Use case: conversation context, chat-specific settings, accumulated knowledge\n\tSpaceChat Space = \"chat\"\n\n\t// SpaceContext context-level memory, temporary within a single request context\n\t// Use case: intermediate results, temporary variables, request-scoped cache\n\tSpaceContext Space = \"context\"\n)\n\n// Config represents the memory configuration\n// Each field is a Store ID referencing gou/store, empty string uses built-in default\n// All spaces use xun-based storage by default for persistence and reliability\ntype Config struct {\n\tUser string `json:\"user,omitempty\" yaml:\"user,omitempty\"` // Store ID for user-level memory (default: xun-based)\n\tTeam string `json:\"team,omitempty\" yaml:\"team,omitempty\"` // Store ID for team-level memory (default: xun-based)\n\tChat string `json:\"chat,omitempty\" yaml:\"chat,omitempty\"` // Store ID for chat-level memory (default: xun-based)\n\tContext string `json:\"context,omitempty\" yaml:\"context,omitempty\"` // Store ID for context-level memory (default: xun-based, shorter TTL)\n}\n\n// DefaultStoreID constants for built-in stores\nconst (\n\tDefaultUserStore = \"__yao.agent.memory.user\"\n\tDefaultTeamStore = \"__yao.agent.memory.team\"\n\tDefaultChatStore = \"__yao.agent.memory.chat\"\n\tDefaultContextStore = \"__yao.agent.memory.context\"\n)\n\n// Entry represents a memory entry\ntype Entry struct {\n\tKey string `json:\"key\"`\n\tValue interface{} `json:\"value\"`\n\tSpace Space `json:\"space\"`\n\tMetadata map[string]interface{} `json:\"metadata,omitempty\"`\n\tTTL time.Duration `json:\"ttl,omitempty\"`\n\tCreatedAt time.Time `json:\"created_at\"`\n\tUpdatedAt time.Time `json:\"updated_at\"`\n\tExpiresAt *time.Time `json:\"expires_at,omitempty\"`\n}\n\n// Namespace represents a memory namespace for a specific space\ntype Namespace struct {\n\tSpace Space `json:\"space\"`\n\tID string `json:\"id\"` // UserID, TeamID, ChatID, or ContextID depending on space\n\tStore store.Store `json:\"-\"` // Underlying store\n\tStoreID string `json:\"-\"` // Store ID\n\tPrefix string `json:\"-\"` // Computed key prefix (e.g., \"user:123:\", \"team:456:\")\n\tDefault time.Duration `json:\"-\"` // Default TTL for this namespace\n}\n\n// Memory represents the complete memory system for an agent\n// It manages four separate namespaces: User, Team, Chat, and Context\ntype Memory struct {\n\tUserID string `json:\"user_id\"`\n\tTeamID string `json:\"team_id\"`\n\tChatID string `json:\"chat_id\"`\n\tContextID string `json:\"context_id\"`\n\n\tUser *Namespace `json:\"-\"` // User-level memory namespace\n\tTeam *Namespace `json:\"-\"` // Team-level memory namespace\n\tChat *Namespace `json:\"-\"` // Chat-level memory namespace\n\tContext *Namespace `json:\"-\"` // Context-level memory namespace\n\tConfig *Config `json:\"-\"` // Memory configuration\n}\n\n// Stats represents memory statistics\ntype Stats struct {\n\tUser *NamespaceStats `json:\"user,omitempty\"`\n\tTeam *NamespaceStats `json:\"team,omitempty\"`\n\tChat *NamespaceStats `json:\"chat,omitempty\"`\n\tContext *NamespaceStats `json:\"context,omitempty\"`\n}\n\n// NamespaceStats represents statistics for a single memory namespace\ntype NamespaceStats struct {\n\tSpace Space `json:\"space\"`\n\tID string `json:\"id\"`\n\tKeyCount int `json:\"key_count\"`\n\tStoreID string `json:\"store_id\"`\n}\n" }, { "path": "agent/output/BUILTIN_TYPES.md", "content": "# Built-in Message Types\n\nBuilt-in message types are standardized types that all adapters must support. These types have predefined Props structures to ensure consistency across different output formats.\n\n## Type Constants\n\nDefined in `types.go`:\n\n```go\nconst (\n TypeUserInput = \"user_input\" // User input message (frontend display only)\n TypeText = \"text\" // Plain text or Markdown content\n TypeThinking = \"thinking\" // Reasoning/thinking process\n TypeLoading = \"loading\" // Loading/processing indicator\n TypeToolCall = \"tool_call\" // LLM tool/function call\n TypeRetrieval = \"retrieval\" // KB/Web search results (for feedback & analytics)\n TypeError = \"error\" // Error message\n TypeImage = \"image\" // Image content\n TypeAudio = \"audio\" // Audio content\n TypeVideo = \"video\" // Video content\n TypeAction = \"action\" // System action (silent in standard clients)\n TypeEvent = \"event\" // Lifecycle event (silent in standard clients)\n)\n```\n\n## Standard Props Structures\n\n### 1. User Input (`user_input`)\n\n**Purpose:** User input message (for frontend display only)\n\n**Props Structure:**\n\n```go\ntype UserInputProps struct {\n Content interface{} `json:\"content\"` // User input (text string or multimodal ContentPart[])\n Role string `json:\"role,omitempty\"` // User role: \"user\", \"system\", \"developer\" (default: \"user\")\n Name string `json:\"name,omitempty\"` // Optional participant name\n}\n```\n\n**Example:**\n\n```json\n{\n \"type\": \"user_input\",\n \"props\": {\n \"content\": \"Hello, can you help me?\",\n \"role\": \"user\"\n }\n}\n```\n\n**Multimodal Example:**\n\n```json\n{\n \"type\": \"user_input\",\n \"props\": {\n \"content\": [\n {\n \"type\": \"text\",\n \"text\": \"What's in this image?\"\n },\n {\n \"type\": \"image_url\",\n \"image_url\": {\n \"url\": \"https://example.com/photo.jpg\"\n }\n }\n ],\n \"role\": \"user\"\n }\n}\n```\n\n**Helper:**\n\n```go\n// Simple text input\nmsg := output.NewUserInputMessage(\"Hello, can you help me?\", \"user\", \"\")\n\n// With name\nmsg := output.NewUserInputMessage(\"I need assistance\", \"user\", \"John\")\n\n// Multimodal content\ncontent := []map[string]interface{}{\n {\n \"type\": \"text\",\n \"text\": \"What's in this image?\",\n },\n {\n \"type\": \"image_url\",\n \"image_url\": map[string]string{\n \"url\": \"https://example.com/photo.jpg\",\n },\n },\n}\nmsg := output.NewUserInputMessage(content, \"user\", \"\")\n```\n\n**Important Notes:**\n\n- **Frontend display only**: This type is used by the frontend to display user input in the chat UI\n- **Not sent to backend**: User input is sent to backend as `UserMessage` (OpenAI format), not as `Message`\n- **Preserves role**: Unlike `text` type, preserves the original user role (`user`, `system`, `developer`)\n- **Supports multimodal**: Can contain text, images, audio, or files\n\n**Data Flow:**\n\n```\nUser types → UserMessage (sent to API) → Backend processes → Message types (AI response)\n ↓\n UserInputMessage (frontend display)\n```\n\n---\n\n### 2. Text (`text`)\n\n**Purpose:** Plain text or Markdown content (AI responses)\n\n**Props Structure:**\n\n```go\ntype TextProps struct {\n Content string `json:\"content\"` // Text content (supports Markdown)\n}\n```\n\n**Example:**\n\n```json\n{\n \"type\": \"text\",\n \"props\": {\n \"content\": \"Hello **world**!\"\n }\n}\n```\n\n**Helper:**\n\n```go\nmsg := output.NewTextMessage(\"Hello **world**!\")\n```\n\n---\n\n### 3. Thinking (`thinking`)\n\n**Purpose:** Reasoning or thinking process (used by o1 models, DeepSeek R1, etc.)\n\n**Props Structure:**\n\n```go\ntype ThinkingProps struct {\n Content string `json:\"content\"` // Reasoning/thinking content\n}\n```\n\n**Example:**\n\n```json\n{\n \"type\": \"thinking\",\n \"props\": {\n \"content\": \"Let me analyze this step by step...\"\n }\n}\n```\n\n**Helper:**\n\n```go\nmsg := output.NewThinkingMessage(\"Let me analyze this step by step...\")\n```\n\n---\n\n### 4. Loading (`loading`)\n\n**Purpose:** Loading or processing indicator (preprocessing, knowledge base search, data fetching, etc.)\n\n**Props Structure:**\n\n```go\ntype LoadingProps struct {\n Message string `json:\"message\"` // Loading message\n}\n```\n\n**Example:**\n\n```json\n{\n \"type\": \"loading\",\n \"props\": {\n \"message\": \"Searching knowledge base...\"\n }\n}\n```\n\n**Helper:**\n\n```go\nmsg := output.NewLoadingMessage(\"Searching knowledge base...\")\n```\n\n**Use Cases:**\n\n- Knowledge base search: `\"Searching knowledge base...\"`\n- Data preprocessing: `\"Processing uploaded file...\"`\n- External API calls: `\"Fetching data from API...\"`\n- Database queries: `\"Querying database...\"`\n\n**Example in Hook:**\n\n```go\n// In Create hook, show preprocessing steps\nfunc Create(ctx *context.Context, messages []context.Message) (*context.HookCreateResponse, error) {\n // Send loading message for knowledge base search\n output.Send(ctx, output.NewLoadingMessage(\"Searching knowledge base...\"))\n\n // Do the actual search\n results := searchKnowledgeBase(messages)\n\n // Send another loading message for processing\n output.Send(ctx, output.NewLoadingMessage(\"Processing results...\"))\n\n // Process and return\n return &context.HookCreateResponse{\n Messages: buildMessages(results),\n }, nil\n}\n```\n\n**Result in OpenAI Client:**\n\n- Shows as thinking/reasoning process\n- User sees \"Searching knowledge base...\" and \"Processing results...\"\n- Provides transparency into what's happening\n\n---\n\n### 5. Tool Call (`tool_call`)\n\n**Purpose:** LLM tool or function call\n\n**Props Structure:**\n\n```go\ntype ToolCallProps struct {\n ID string `json:\"id\"` // Tool call ID\n Name string `json:\"name\"` // Function/tool name\n Arguments string `json:\"arguments,omitempty\"` // JSON string of arguments\n}\n```\n\n**Example:**\n\n```json\n{\n \"type\": \"tool_call\",\n \"props\": {\n \"id\": \"call_abc123\",\n \"name\": \"get_weather\",\n \"arguments\": \"{\\\"location\\\": \\\"San Francisco\\\"}\"\n }\n}\n```\n\n**Helper:**\n\n```go\nmsg := output.NewToolCallMessage(\n \"call_abc123\",\n \"get_weather\",\n \"{\\\"location\\\": \\\"San Francisco\\\"}\",\n)\n```\n\n---\n\n### 6. Retrieval (`retrieval`)\n\n**Purpose:** Knowledge base and web search results (for feedback, analytics, and source attribution)\n\n**Props Structure:**\n\n```go\ntype RetrievalProps struct {\n Query string `json:\"query\"` // Search query\n Sources []RetrievalSource `json:\"sources\"` // Retrieved sources\n TotalResults int `json:\"total_results,omitempty\"` // Total matching results\n QueryTimeMs int64 `json:\"query_time_ms,omitempty\"` // Query execution time\n Provider string `json:\"provider,omitempty\"` // Search provider (e.g., \"tavily\", \"bing\")\n}\n\ntype RetrievalSource struct {\n ID string `json:\"id\"` // Unique source ID within this retrieval\n Type string `json:\"type\"` // Source type: \"kb\", \"web\", \"file\", \"api\", \"mcp\"\n Title string `json:\"title,omitempty\"` // Source title\n Content string `json:\"content\"` // Retrieved content/snippet\n Score float64 `json:\"score,omitempty\"` // Relevance score\n URL string `json:\"url,omitempty\"` // URL for web sources\n CollectionID string `json:\"collection_id,omitempty\"` // KB collection ID\n DocumentID string `json:\"document_id,omitempty\"` // KB document ID\n ChunkID string `json:\"chunk_id,omitempty\"` // KB chunk ID\n Metadata map[string]interface{} `json:\"metadata,omitempty\"` // Additional metadata\n}\n```\n\n**Example (Knowledge Base):**\n\n```json\n{\n \"type\": \"retrieval\",\n \"props\": {\n \"query\": \"How to configure Yao models?\",\n \"sources\": [\n {\n \"id\": \"src_001\",\n \"type\": \"kb\",\n \"collection_id\": \"col_docs\",\n \"document_id\": \"doc_123\",\n \"chunk_id\": \"chunk_456\",\n \"title\": \"Model Configuration Guide\",\n \"content\": \"To configure a model in Yao, create a .mod.yao file...\",\n \"score\": 0.92,\n \"metadata\": {\n \"file_path\": \"/docs/model.md\",\n \"page\": 3\n }\n }\n ],\n \"total_results\": 15,\n \"query_time_ms\": 120\n }\n}\n```\n\n**Example (Web Search):**\n\n```json\n{\n \"type\": \"retrieval\",\n \"props\": {\n \"query\": \"latest AI news 2024\",\n \"sources\": [\n {\n \"id\": \"src_001\",\n \"type\": \"web\",\n \"url\": \"https://example.com/ai-news\",\n \"title\": \"AI Breakthroughs in 2024\",\n \"content\": \"Summary of the article...\",\n \"score\": 0.95,\n \"metadata\": {\n \"domain\": \"example.com\",\n \"published_at\": \"2024-01-10\"\n }\n }\n ],\n \"provider\": \"tavily\",\n \"total_results\": 10,\n \"query_time_ms\": 850\n }\n}\n```\n\n**Helper:**\n\n```go\nmsg := output.NewRetrievalMessage(\n \"How to configure Yao models?\",\n []output.RetrievalSource{\n {\n ID: \"src_001\",\n Type: \"kb\",\n CollectionID: \"col_docs\",\n DocumentID: \"doc_123\",\n ChunkID: \"chunk_456\",\n Title: \"Model Configuration Guide\",\n Content: \"To configure a model in Yao...\",\n Score: 0.92,\n },\n },\n)\n```\n\n**Source Types:**\n\n| Type | Description | Key Fields |\n| ------ | ----------------------- | ------------------------------------------ |\n| `kb` | Knowledge base document | `collection_id`, `document_id`, `chunk_id` |\n| `web` | Web search result | `url` |\n| `file` | Uploaded file | `file_id`, `file_path` |\n| `api` | External API result | `api_name`, `endpoint` |\n| `mcp` | MCP tool result | `server`, `tool` |\n\n**Use Cases:**\n\n- **Source Attribution**: Display citations in the chat UI\n- **User Feedback**: Allow users to rate individual sources (👍/👎)\n- **Analytics**: Track which documents/sources are most useful\n- **RAG Optimization**: Improve retrieval based on feedback data\n\n**Adapter Behavior:**\n\n- **CUI**: Renders as expandable source cards with feedback buttons\n- **OpenAI**: Converts to markdown citations or footnotes\n\n---\n\n### 7. Error (`error`)\n\n**Purpose:** Error message\n\n**Props Structure:**\n\n```go\ntype ErrorProps struct {\n Message string `json:\"message\"` // Error message\n Code string `json:\"code,omitempty\"` // Error code\n Details string `json:\"details,omitempty\"` // Additional error details\n}\n```\n\n**Example:**\n\n```json\n{\n \"type\": \"error\",\n \"props\": {\n \"message\": \"Connection timeout\",\n \"code\": \"TIMEOUT\",\n \"details\": \"Failed to connect to database after 30s\"\n }\n}\n```\n\n**Helper:**\n\n```go\nmsg := output.NewErrorMessage(\"Connection timeout\", \"TIMEOUT\")\n```\n\n---\n\n### 8. Action (`action`)\n\n**Purpose:** System-level action/command (not displayed to user, only processed by client)\n\n**Props Structure:**\n\n```go\ntype ActionProps struct {\n Name string `json:\"name\"` // Action name\n Payload map[string]interface{} `json:\"payload,omitempty\"` // Action parameters\n}\n```\n\n**Example:**\n\n```json\n{\n \"type\": \"action\",\n \"props\": {\n \"name\": \"open_panel\",\n \"payload\": {\n \"panel_id\": \"user_profile\",\n \"user_id\": \"123\"\n }\n }\n}\n```\n\n**Helper:**\n\n```go\nmsg := output.NewActionMessage(\"open_panel\", map[string]interface{}{\n \"panel_id\": \"user_profile\",\n \"user_id\": \"123\",\n})\n```\n\n**Use Cases:**\n\n- Open sidebar/panel: `\"open_panel\"`\n- Navigate to page: `\"navigate\"`\n- Trigger UI update: `\"refresh_view\"`\n- Close modal: `\"close_modal\"`\n- Scroll to element: `\"scroll_to\"`\n\n**Important Notes:**\n\n- **Silent in OpenAI clients**: Action messages are NOT sent to standard chat clients\n- **CUI clients only**: Only CUI clients process action messages\n- **System-level**: Used for controlling the UI/application, not chat content\n\n**Example in Hook:**\n\n```go\n// Send action to open a panel with user details\noutput.Send(ctx, output.NewActionMessage(\"open_panel\", map[string]interface{}{\n \"panel_id\": \"user_details\",\n \"user_id\": user.ID,\n}))\n\n// Send text message (visible to user)\noutput.Send(ctx, output.NewTextMessage(\"I've opened the user details panel for you.\"))\n```\n\n**Result:**\n\n- **CUI client**: Panel opens, text message displays\n- **OpenAI client**: Only text message displays (action is silent)\n\n---\n\n### 9. Event (`event`)\n\n**Purpose:** Lifecycle event messages (stream_start, stream_end, connecting, etc.)\n\n**Props Structure:**\n\n```go\ntype EventProps struct {\n Event string `json:\"event\"` // Event type\n Message string `json:\"message,omitempty\"` // Human-readable message\n Data map[string]interface{} `json:\"data,omitempty\"` // Additional event data\n}\n```\n\n**Example:**\n\n```json\n{\n \"type\": \"event\",\n \"props\": {\n \"event\": \"stream_start\",\n \"message\": \"Starting stream...\",\n \"data\": {\n \"model\": \"gpt-4\",\n \"session_id\": \"sess_123\"\n }\n }\n}\n```\n\n**Helper:**\n\n```go\nmsg := output.NewEventMessage(\"stream_start\", \"Starting stream...\", map[string]interface{}{\n \"model\": \"gpt-4\",\n \"session_id\": \"sess_123\",\n})\n```\n\n**Use Cases:**\n\n- Stream lifecycle: `\"stream_start\"`, `\"stream_end\"`\n- Connection status: `\"connecting\"`, `\"connected\"`, `\"disconnected\"`\n- Processing stages: `\"preprocessing\"`, `\"postprocessing\"`\n- Agent state: `\"thinking\"`, `\"executing\"`, `\"completed\"`\n\n**Important Notes:**\n\n- **Converted in OpenAI clients**: Event messages are typically NOT sent to OpenAI clients, **except** `stream_start`:\n - `stream_start`: Converted to a clickable trace link in either `reasoning_content` (thinking models) or `content` (regular models)\n - Other events: Silent (not sent to OpenAI clients)\n- **CUI clients**: All event messages are processed and may show status indicators\n- **Lifecycle tracking**: Used for tracking agent/stream lifecycle\n- **Non-blocking**: Events don't interrupt the main message flow\n\n**Example in Hook:**\n\n```go\n// Send stream start event (automatically generated by assistant)\n// This is typically handled by the framework, not manually sent\nstartData := message.EventStreamStartData{\n RequestID: ctx.RequestID,\n Timestamp: time.Now().UnixMilli(),\n TraceID: ctx.Stack.TraceID,\n ChatID: ctx.ChatID,\n}\noutput.Send(ctx, output.NewEventMessage(\"stream_start\", \"Stream started\", startData))\n\n// Do processing\nprocessData()\n\n// Send stream end event\nendData := message.EventStreamEndData{\n RequestID: ctx.RequestID,\n Timestamp: time.Now().UnixMilli(),\n DurationMs: 1500,\n Status: \"completed\",\n}\noutput.Send(ctx, output.NewEventMessage(\"stream_end\", \"Stream completed\", endData))\n```\n\n**Result:**\n\n- **CUI client**: Tracks lifecycle, may show status indicators\n- **OpenAI client (stream_start only)**:\n - Reasoning models: Shows as 🔍 with trace link in `reasoning_content` field\n - Regular models: Shows as 🚀 with trace link in `content` field\n - Example: \"🔍 智能体正在处理 - [查看处理详情](baseURL/trace/traceID/view)\"\n- **OpenAI client (other events)**: Silent (not sent)\n\n---\n\n### 10. Image (`image`)\n\n**Purpose:** Image content\n\n**Props Structure:**\n\n```go\ntype ImageProps struct {\n URL string // Required: Image URL or base64 data\n Alt string // Alternative text\n Width int // Image width in pixels\n Height int // Image height in pixels\n Detail string // OpenAI detail level: \"auto\", \"low\", \"high\"\n}\n```\n\n**Example:**\n\n```json\n{\n \"type\": \"image\",\n \"props\": {\n \"url\": \"https://example.com/avatar.jpg\",\n \"alt\": \"User avatar\",\n \"width\": 200,\n \"height\": 200\n }\n}\n```\n\n**Helper:**\n\n```go\nmsg := output.NewImageMessage(\"https://example.com/avatar.jpg\", \"User avatar\")\n```\n\n**Adapter Behavior:**\n\n- **CUI**: Renders image directly with `` tag\n- **OpenAI**: Converts to Markdown `![alt](url)` - **displays inline** in Markdown-supporting clients\n\n---\n\n### 11. Audio (`audio`)\n\n**Purpose:** Audio content\n\n**Props Structure:**\n\n```go\ntype AudioProps struct {\n URL string // Required: Audio URL or base64 data\n Format string // Audio format: \"mp3\", \"wav\", \"ogg\"\n Duration float64 // Duration in seconds\n Transcript string // Audio transcript text\n Autoplay bool // Whether to autoplay\n Controls bool // Whether to show controls\n}\n```\n\n**Example:**\n\n```json\n{\n \"type\": \"audio\",\n \"props\": {\n \"url\": \"https://example.com/audio.mp3\",\n \"format\": \"mp3\",\n \"duration\": 120.5,\n \"transcript\": \"This is the audio content...\",\n \"controls\": true\n }\n}\n```\n\n**Helper:**\n\n```go\nmsg := output.NewAudioMessage(\"https://example.com/audio.mp3\", \"mp3\")\n```\n\n**Adapter Behavior:**\n\n- **CUI**: Renders audio player with controls\n- **OpenAI**: Converts to link `🔊 [Play Audio](url)` - can't display inline\n\n---\n\n### 12. Video (`video`)\n\n**Purpose:** Video content\n\n**Props Structure:**\n\n```go\ntype VideoProps struct {\n URL string // Required: Video URL\n Format string // Video format: \"mp4\", \"webm\"\n Duration float64 // Duration in seconds\n Thumbnail string // Thumbnail/poster image URL\n Width int // Video width in pixels\n Height int // Video height in pixels\n Autoplay bool // Whether to autoplay\n Controls bool // Whether to show controls\n Loop bool // Whether to loop\n}\n```\n\n**Example:**\n\n```json\n{\n \"type\": \"video\",\n \"props\": {\n \"url\": \"https://example.com/video.mp4\",\n \"format\": \"mp4\",\n \"thumbnail\": \"https://example.com/poster.jpg\",\n \"width\": 640,\n \"height\": 360,\n \"controls\": true\n }\n}\n```\n\n**Helper:**\n\n```go\nmsg := output.NewVideoMessage(\"https://example.com/video.mp4\")\n```\n\n**Adapter Behavior:**\n\n- **CUI**: Renders video player with controls\n- **OpenAI**: Converts to link `🎬 [Watch Video](url)` - can't display inline\n\n---\n\n## Adapter Requirements\n\nAll adapters (CUI, OpenAI, etc.) **must** support these built-in types with their standard Props structures.\n\n### CUI Adapter\n\nCUI adapter passes built-in types through without transformation:\n\n```json\n{\n \"type\": \"text\",\n \"props\": {\n \"content\": \"Hello world\"\n }\n}\n```\n\n### OpenAI Adapter\n\nOpenAI adapter converts built-in types to OpenAI format:\n\n| Type | OpenAI Format | Field | Note |\n| ------------ | ------------------------- | ----------------------------- | -------------------------------------------------------------------- |\n| `user_input` | (not sent) | - | Frontend display only - not sent to OpenAI clients |\n| `text` | `delta.content` | `props.content` | |\n| `thinking` | `delta.reasoning_content` | `props.content` | Reasoning content (o1 models) |\n| `loading` | `delta.reasoning_content` | `props.message` | Shows as thinking in OpenAI clients |\n| `tool_call` | `delta.tool_calls` | `props.{id, name, arguments}` | |\n| `retrieval` | `delta.content` | `props.sources` | Markdown citations/footnotes with source links |\n| `error` | `error` | `props.{message, code}` | |\n| `image` | `delta.content` | `props.{url, alt}` | Markdown: `![alt](url)` - displays inline |\n| `audio` | `delta.content` | `props.url` | Markdown link (can't display inline) |\n| `video` | `delta.content` | `props.url` | Markdown link (can't display inline) |\n| `action` | (not sent) | - | Silent - system actions only |\n| `event` | (conditional) | `props.{event, data}` | Most events silent; `stream_start` converted to trace link with i18n |\n\n---\n\n## Custom Types\n\nAny type **not** in the built-in list is considered a custom type. Adapters may handle custom types differently:\n\n- **CUI:** Pass through as-is\n- **OpenAI:** Convert to Markdown link\n\nExample custom type:\n\n```json\n{\n \"type\": \"image\",\n \"props\": {\n \"url\": \"https://example.com/image.jpg\",\n \"alt\": \"Description\"\n }\n}\n```\n\n---\n\n## Checking Built-in Types\n\n```go\n// Check if a type is built-in\nif output.IsBuiltinType(msg.Type) {\n // Handle as standard type\n} else {\n // Handle as custom type\n}\n```\n\n---\n\n## Guidelines for New Built-in Types\n\nWhen adding new built-in types:\n\n1. ✅ Add constant to `types.go`\n2. ✅ Define Props structure\n3. ✅ Add helper function in `builtin.go`\n4. ✅ Update all adapters to support it\n5. ✅ Document in this file\n6. ✅ Add tests\n\n**Only add built-in types for:**\n\n- Universal concepts (text, errors, etc.)\n- LLM-specific features (thinking, tool_calls)\n- Types that need cross-adapter consistency\n\n**Do NOT add built-in types for:**\n\n- UI components (buttons, forms, etc.)\n- Application-specific widgets\n- Domain-specific data types\n\nThese should remain custom types.\n" }, { "path": "agent/output/README.md", "content": "# Output Module\n\nThe output module provides a unified API for sending messages to different client types (CUI, OpenAI-compatible, etc.) with support for streaming and rich media content.\n\n## Architecture\n\n```\nagent/output/\n├── message/ # Core types and interfaces (no dependencies)\n│ ├── types.go # Message, Group, Props structures\n│ └── interfaces.go # Writer, Adapter, Factory interfaces\n├── adapters/ # Client-specific adapters\n│ ├── cui/ # CUI adapter (native DSL)\n│ │ ├── adapter.go\n│ │ └── writer.go\n│ └── openai/ # OpenAI adapter (converts to OpenAI format)\n│ ├── adapter.go\n│ ├── converter.go\n│ ├── writer.go\n│ ├── types.go\n│ └── factory.go\n├── output.go # Main API (Send, GetWriter, etc.)\n├── builtin.go # Helper functions for built-in types\n└── BUILTIN_TYPES.md # Documentation for built-in types\n```\n\n## DSL Structure\n\n### Message Structure\n\nThe universal message DSL is a JSON structure that supports streaming, rich media, and incremental updates:\n\n```go\ntype Message struct {\n // Core fields\n Type string `json:\"type\"` // Message type (e.g., \"text\", \"image\", \"action\")\n Props map[string]interface{} `json:\"props,omitempty\"` // Type-specific properties\n\n // Streaming control - Hierarchical structure for Agent/LLM/MCP streaming\n ChunkID string `json:\"chunk_id,omitempty\"` // Unique chunk ID (C1, C2, C3...; for dedup/ordering/debugging)\n MessageID string `json:\"message_id,omitempty\"` // Logical message ID (M1, M2, M3...; delta merge target; multiple chunks → one message)\n BlockID string `json:\"block_id,omitempty\"` // Block ID (B1, B2, B3...; Agent-level grouping for UI sections)\n ThreadID string `json:\"thread_id,omitempty\"` // Thread ID (T1, T2, T3...; optional; for concurrent streams)\n\n // Delta control\n Delta bool `json:\"delta,omitempty\"` // Whether this is an incremental update\n DeltaPath string `json:\"delta_path,omitempty\"` // Which field to update (e.g., \"content\", \"items.0.name\")\n DeltaAction string `json:\"delta_action,omitempty\"` // How to update (\"append\", \"replace\", \"merge\", \"set\")\n\n // Type correction (for streaming type inference)\n TypeChange bool `json:\"type_change,omitempty\"` // Marks this as a type correction message\n\n // Metadata\n Metadata *Metadata `json:\"metadata,omitempty\"` // Timestamp, sequence, trace ID\n}\n```\n\n### Field Descriptions\n\n#### Core Fields\n\n- **`Type`** (required): Determines how the message should be rendered\n\n - Built-in types: `text`, `thinking`, `loading`, `tool_call`, `error`, `image`, `audio`, `video`, `action`, `event`\n - Custom types: Any string (frontend must have corresponding component)\n\n- **`Props`** (optional): Type-specific properties passed to the rendering component\n - For `text`: `{\"content\": \"Hello\"}`\n - For `image`: `{\"url\": \"...\", \"alt\": \"...\"}`\n - For custom types: Any JSON-serializable data\n\n#### Streaming Control\n\nHierarchical structure for fine-grained control over streaming in complex Agent/LLM/MCP scenarios:\n\n- **`ChunkID`** (optional): Unique chunk identifier\n\n - Auto-generated (C1, C2, C3...)\n - For deduplication, ordering, and debugging\n - Each raw stream fragment gets a unique ChunkID\n\n- **`MessageID`** (optional): Logical message identifier\n\n - Auto-generated (M1, M2, M3...)\n - Delta merge target - multiple chunks with same MessageID are merged\n - Represents one complete logical message (e.g., one thinking output, one text response)\n - Example: `\"M1\"`\n\n- **`BlockID`** (optional): Output block identifier\n\n - Auto-generated (B1, B2, B3...)\n - Agent-level grouping for UI sections\n - One LLM call, one MCP call, or one Agent sub-task\n - Used for rendering blocks/sections in the UI\n\n- **`ThreadID`** (optional): Thread identifier\n\n - Auto-generated (T1, T2, T3...)\n - For concurrent Agent/LLM/MCP calls\n - Distinguishes multiple parallel output streams\n\n- **`Delta`** (optional): Marks this as an incremental update\n - `true`: Append/update to existing message with same MessageID\n - `false`: Complete message (default)\n - Used for streaming LLM responses\n\n#### Delta Update Control\n\nFor complex, structured messages that need field-level updates:\n\n- **`DeltaPath`** (optional): JSON path to the field being updated\n\n - Simple: `\"content\"` (updates `props.content`)\n - Nested: `\"user.name\"` (updates `props.user.name`)\n - Array: `\"items.0.title\"` (updates `props.items[0].title`)\n\n- **`DeltaAction`** (optional): How to apply the delta update\n - `\"append\"`: Concatenate to existing string/array\n - `\"replace\"`: Replace entire value\n - `\"merge\"`: Merge objects (shallow merge)\n - `\"set\"`: Set new field (if doesn't exist)\n\n#### Type Correction\n\n- **`TypeChange`** (optional): Indicates message type was corrected\n - Used when initial type inference was wrong\n - Frontend should re-render with new type\n - Example: Initially sent as `text`, corrected to `thinking`\n\n#### Metadata\n\n- **`Metadata`** (optional): Additional message metadata\n ```go\n type Metadata struct {\n Timestamp int64 // Unix nanoseconds\n Sequence int // Message sequence number\n TraceID string // For debugging/logging\n }\n ```\n\n### Message Examples\n\n#### Simple Text Message\n\n```json\n{\n \"type\": \"text\",\n \"props\": {\n \"content\": \"Hello, world!\"\n }\n}\n```\n\n#### Streaming Text (Delta Updates)\n\n```json\n// First chunk\n{\n \"chunk_id\": \"C1\",\n \"message_id\": \"M1\",\n \"type\": \"text\",\n \"delta\": true,\n \"props\": {\n \"content\": \"Hello\"\n }\n}\n\n// Second chunk (appends)\n{\n \"chunk_id\": \"C2\",\n \"message_id\": \"M1\",\n \"type\": \"text\",\n \"delta\": true,\n \"props\": {\n \"content\": \", world\"\n }\n}\n\n// Third chunk\n{\n \"chunk_id\": \"C3\",\n \"message_id\": \"M1\",\n \"type\": \"text\",\n \"delta\": true,\n \"props\": {\n \"content\": \"!\"\n }\n}\n\n// Completion signaled by message_end event (sent separately)\n{\n \"type\": \"event\",\n \"props\": {\n \"event\": \"message_end\",\n \"data\": {\n \"message_id\": \"M1\",\n \"type\": \"text\",\n \"chunk_count\": 3,\n \"status\": \"completed\"\n }\n }\n}\n```\n\n#### Complex Type with Nested Updates\n\n```json\n// Initial message\n{\n \"message_id\": \"M2\",\n \"type\": \"table\",\n \"props\": {\n \"columns\": [\"Name\", \"Age\"],\n \"rows\": []\n }\n}\n\n// Add first row\n{\n \"chunk_id\": \"C4\",\n \"message_id\": \"M2\",\n \"type\": \"table\",\n \"delta\": true,\n \"delta_path\": \"rows\",\n \"delta_action\": \"append\",\n \"props\": {\n \"rows\": [{\"name\": \"Alice\", \"age\": 30}]\n }\n}\n\n// Add second row\n{\n \"chunk_id\": \"C5\",\n \"message_id\": \"M2\",\n \"type\": \"table\",\n \"delta\": true,\n \"delta_path\": \"rows\",\n \"delta_action\": \"append\",\n \"props\": {\n \"rows\": [{\"name\": \"Bob\", \"age\": 25}]\n }\n}\n```\n\n#### Type Correction\n\n```json\n// Initial guess (text)\n{\n \"chunk_id\": \"C6\",\n \"message_id\": \"M3\",\n \"type\": \"text\",\n \"delta\": true,\n \"props\": {\n \"content\": \"Let me think...\"\n }\n}\n\n// Correction (actually thinking)\n{\n \"chunk_id\": \"C7\",\n \"message_id\": \"M3\",\n \"type\": \"thinking\",\n \"type_change\": true,\n \"props\": {\n \"content\": \"Let me think...\"\n }\n}\n```\n\n#### Block Grouping (Agent-level)\n\n```json\n// Block start event\n{\n \"type\": \"event\",\n \"props\": {\n \"event\": \"block_start\",\n \"data\": {\n \"block_id\": \"B1\",\n \"type\": \"llm\",\n \"label\": \"Analyzing image\"\n }\n }\n}\n\n// Thinking message in block\n{\n \"message_id\": \"M4\",\n \"block_id\": \"B1\",\n \"type\": \"thinking\",\n \"props\": {\n \"content\": \"Let me analyze this image...\"\n }\n}\n\n// Text message in block\n{\n \"message_id\": \"M5\",\n \"block_id\": \"B1\",\n \"type\": \"text\",\n \"props\": {\n \"content\": \"This is a beautiful sunset at Golden Gate Bridge\"\n }\n}\n\n// Block end event\n{\n \"type\": \"event\",\n \"props\": {\n \"event\": \"block_end\",\n \"data\": {\n \"block_id\": \"B1\",\n \"message_count\": 2,\n \"status\": \"completed\"\n }\n }\n}\n```\n\n## Key Design Decisions\n\n### 1. Separate `message` Package\n\nTo avoid circular dependencies, all core types and interfaces are defined in the `message` sub-package:\n\n- `message.Message` - Universal message DSL\n- `message.Writer` - Interface for writing messages\n- `message.Adapter` - Interface for format conversion\n\nThis allows:\n\n- `handlers` → `output` → `message` ✅\n- `output/adapters` → `message` ✅\n- No circular dependencies!\n\n### 2. Adapter Pattern\n\nDifferent clients require different formats:\n\n**CUI Clients:**\n\n```json\n{\n \"type\": \"text\",\n \"props\": { \"content\": \"Hello\" }\n}\n```\n\n**OpenAI Clients:**\n\n```json\n{\n \"choices\": [\n {\n \"delta\": { \"content\": \"Hello\" }\n }\n ]\n}\n```\n\nAdapters handle the transformation automatically based on `ctx.Accept`.\n\n### 3. Built-in Types\n\n10 standardized message types with defined Props structures:\n\n| Type | Purpose | CUI | OpenAI |\n| ----------- | ------------------ | ------- | -------------------------------------------- |\n| `text` | Text content | Direct | `delta.content` |\n| `thinking` | LLM reasoning | Direct | `delta.reasoning_content` |\n| `loading` | Progress indicator | Direct | `delta.reasoning_content` |\n| `tool_call` | Function calls | Direct | `delta.tool_calls` |\n| `error` | Error messages | Direct | `error` |\n| `image` | Images | Render | `![](url)` markdown |\n| `audio` | Audio | Player | Link |\n| `video` | Video | Player | Link |\n| `action` | System commands | Execute | Silent |\n| `event` | Lifecycle events | Track | Conditional (stream_start converted to link) |\n\n## Usage\n\n### Basic Usage\n\n```go\nimport (\n \"github.com/yaoapp/yao/agent/output\"\n \"github.com/yaoapp/yao/agent/output/message\"\n)\n\n// Send a text message\nmsg := output.NewTextMessage(\"Hello world\")\noutput.Send(ctx, msg)\n\n// Send a loading indicator\nloading := output.NewLoadingMessage(\"Searching knowledge base...\")\noutput.Send(ctx, loading)\n\n// Send an image\nimg := output.NewImageMessage(\"https://example.com/image.jpg\", \"Description\")\noutput.Send(ctx, img)\n\n// Send an error\nerr := output.NewErrorMessage(\"Connection failed\", \"TIMEOUT\")\noutput.Send(ctx, err)\n```\n\n### Streaming Messages\n\n```go\n// Get ID generator from context\nidGen := ctx.IDGenerator\n\n// Send delta (incremental) updates\nmsg := &message.Message{\n ChunkID: idGen.GenerateChunkID(), // C1\n MessageID: idGen.GenerateMessageID(), // M1\n Type: message.TypeText,\n Delta: true, // Incremental update\n Props: map[string]interface{}{\n \"content\": \"Hello\",\n },\n}\noutput.Send(ctx, msg)\n\n// Send more delta updates (same MessageID for merging)\nmsg2 := &message.Message{\n ChunkID: idGen.GenerateChunkID(), // C2\n MessageID: msg.MessageID, // M1 (same as before)\n Type: message.TypeText,\n Delta: true,\n Props: map[string]interface{}{\n \"content\": \" world\",\n },\n}\noutput.Send(ctx, msg2)\n\n// Mark completion with message_end event\nendData := message.EventMessageEndData{\n MessageID: msg.MessageID, // M1\n Type: \"text\",\n Status: \"completed\",\n ChunkCount: 2,\n Extra: map[string]interface{}{\n \"content\": \"Hello world!\", // Full content\n },\n}\neventMsg := output.NewEventMessage(message.EventMessageEnd, \"Message completed\", endData)\noutput.Send(ctx, eventMsg)\n```\n\n### Custom Writers\n\n```go\n// Register a custom writer factory\nfactory := &MyCustomFactory{}\noutput.SetWriterFactory(factory)\n\n// Now all calls to output.Send will use your custom writer\n```\n\n## Integration with Handlers\n\nThe `handlers` package uses the output module for streaming:\n\n```go\nfunc DefaultStreamHandler(ctx *context.Context) context.StreamFunc {\n return func(chunkType context.StreamChunkType, data []byte) int {\n switch chunkType {\n case context.ChunkText:\n msg := output.NewTextMessage(string(data))\n output.Send(ctx, msg)\n case context.ChunkThinking:\n msg := output.NewThinkingMessage(string(data))\n output.Send(ctx, msg)\n // ... handle other types\n }\n return 0 // Continue\n }\n}\n```\n\n## Context-based Routing\n\nThe output module automatically selects the right writer based on `ctx.Accept`:\n\n| `ctx.Accept` | Writer | Format |\n| ------------- | ------ | --------------------- |\n| `standard` | OpenAI | OpenAI-compatible SSE |\n| `cui-web` | CUI | Universal DSL JSON |\n| `cui-native` | CUI | Universal DSL JSON |\n| `cui-desktop` | CUI | Universal DSL JSON |\n\n## Writer Caching\n\nWriters are cached per context to avoid recreating them:\n\n```go\n// Get or create writer (cached)\nwriter := output.GetWriter(ctx)\n\n// Clear cache when done\noutput.Close(ctx) // Also closes the writer\n```\n\n## See Also\n\n- [BUILTIN_TYPES.md](./BUILTIN_TYPES.md) - Complete documentation of built-in message types\n- [adapters/openai/README.md](./adapters/openai/README.md) - OpenAI adapter documentation\n" }, { "path": "agent/output/adapters/cui/adapter.go", "content": "package cui\n\nimport \"github.com/yaoapp/yao/agent/output/message\"\n\n// Adapter implements the message.Adapter interface for CUI clients.\n// It performs no conversion and outputs messages as-is, as CUI clients\n// are designed to directly consume the universal DSL.\ntype Adapter struct{}\n\n// NewAdapter creates a new CUI adapter.\nfunc NewAdapter() *Adapter {\n\treturn &Adapter{}\n}\n\n// Adapt converts a universal Message to one or more client-specific chunks.\n// For CUI, it simply returns the original message as a single chunk.\nfunc (a *Adapter) Adapt(msg *message.Message) ([]interface{}, error) {\n\t// CUI clients consume the universal DSL directly, so no conversion is needed.\n\t// This includes all message types like text, thinking, loading, events, etc.\n\t// CUI clients can choose to display or ignore event messages.\n\treturn []interface{}{msg}, nil\n}\n\n// SupportsType checks if the adapter explicitly supports a given message type.\n// CUI adapter supports all types as it renders them directly.\nfunc (a *Adapter) SupportsType(msgType string) bool {\n\treturn true\n}\n" }, { "path": "agent/output/adapters/cui/writer.go", "content": "package cui\n\nimport (\n\t\"encoding/json\"\n\t\"net/http\"\n\n\t\"github.com/yaoapp/yao/agent/i18n\"\n\t\"github.com/yaoapp/yao/agent/output/message\"\n\ttraceTypes \"github.com/yaoapp/yao/trace/types\"\n)\n\n// Writer implements the message.Writer interface for CUI clients\ntype Writer struct {\n\tWriter http.ResponseWriter\n\tTrace traceTypes.Manager\n\tLocale string\n\tadapter *Adapter\n}\n\n// NewWriter creates a new CUI writer\n// The Writer should already be wrapped in SafeWriter by the context layer\n// to ensure thread-safe concurrent writes for SSE streaming.\nfunc NewWriter(options message.Options) (*Writer, error) {\n\treturn &Writer{\n\t\tWriter: options.Writer,\n\t\tTrace: options.Trace,\n\t\tLocale: options.Locale,\n\t\tadapter: NewAdapter(),\n\t}, nil\n}\n\n// Write writes a single message to the output stream\nfunc (w *Writer) Write(msg *message.Message) error {\n\t// CUI adapter passes messages through as-is\n\tchunks, err := w.adapter.Adapt(msg)\n\tif err != nil {\n\t\tif w.Trace != nil {\n\t\t\tw.Trace.Error(i18n.T(w.Locale, \"output.cui.writer.adapt_error\"), map[string]any{ // \"CUI Writer: Failed to adapt message\"\n\t\t\t\t\"error\": err.Error(),\n\t\t\t\t\"message_type\": msg.Type,\n\t\t\t})\n\t\t}\n\t\treturn err\n\t}\n\n\t// Send each chunk\n\tfor _, chunk := range chunks {\n\t\tif err := w.sendChunk(chunk); err != nil {\n\t\t\tif w.Trace != nil {\n\t\t\t\tw.Trace.Error(i18n.T(w.Locale, \"output.cui.writer.chunk_error\"), map[string]any{\"error\": err.Error()}) // \"CUI Writer: Failed to send chunk\"\n\t\t\t}\n\t\t\treturn err\n\t\t}\n\t}\n\n\treturn nil\n}\n\n// WriteGroup writes a message group to the output stream\nfunc (w *Writer) WriteGroup(group *message.Group) error {\n\t// For CUI, we send a group start message, all messages, then a group end message\n\t// The group structure itself is also sent for clients that want it\n\n\t// Send the group\n\tif err := w.sendChunk(group); err != nil {\n\t\tif w.Trace != nil {\n\t\t\tw.Trace.Error(i18n.T(w.Locale, \"output.cui.writer.group_error\"), map[string]any{ // \"CUI Writer: Failed to send message group\"\n\t\t\t\t\"error\": err.Error(),\n\t\t\t\t\"group_id\": group.ID,\n\t\t\t})\n\t\t}\n\t\treturn err\n\t}\n\n\treturn nil\n}\n\n// Flush flushes any buffered data to the output stream\nfunc (w *Writer) Flush() error {\n\t// For SSE, we don't need explicit flushing\n\t// The underlying connection handles it\n\treturn nil\n}\n\n// Close closes the writer and cleans up resources\nfunc (w *Writer) Close() error {\n\t// Nothing to clean up for CUI writer\n\t// SafeWriter cleanup is handled by the context layer\n\treturn nil\n}\n\n// sendChunk sends a chunk to the output stream\nfunc (w *Writer) sendChunk(chunk interface{}) error {\n\t// Convert chunk to JSON\n\tdata, err := json.Marshal(chunk)\n\tif err != nil {\n\t\tif w.Trace != nil {\n\t\t\tw.Trace.Error(i18n.T(w.Locale, \"output.cui.writer.marshal_error\"), map[string]any{\"error\": err.Error()}) // \"CUI Writer: Failed to marshal chunk\"\n\t\t}\n\t\treturn err\n\t}\n\n\t// Log outgoing data to trace for debugging\n\tif w.Trace != nil {\n\t\tw.Trace.Debug(\"CUI Writer: Sending chunk to client\", map[string]any{\n\t\t\t\"data\": string(data),\n\t\t})\n\t}\n\n\t// Format as SSE (Server-Sent Events) format: \"data: {json}\\n\\n\"\n\tsseData := []byte(\"data: \")\n\tsseData = append(sseData, data...)\n\tsseData = append(sseData, '\\n', '\\n')\n\n\t// Send via context's writer\n\t// The context knows how to send data based on the connection type (SSE, WebSocket, etc.)\n\tif err := w.sendData(sseData); err != nil {\n\t\tif w.Trace != nil {\n\t\t\tw.Trace.Error(i18n.T(w.Locale, \"output.cui.writer.send_error\"), map[string]any{\"error\": err.Error()}) // \"CUI Writer: Failed to send data to client\"\n\t\t}\n\t\treturn err\n\t}\n\n\tw.flush()\n\treturn nil\n}\n\nfunc (w *Writer) flush() error {\n\tif w.Writer == nil {\n\t\treturn nil // No writer, silently ignore\n\t}\n\tif flusher, ok := w.Writer.(interface{ Flush() }); ok {\n\t\tflusher.Flush()\n\t}\n\treturn nil\n}\n\nfunc (w *Writer) sendData(data []byte) error {\n\tif w.Writer == nil {\n\t\treturn nil // No writer, silently ignore\n\t}\n\t_, err := w.Writer.Write(data)\n\treturn err\n}\n" }, { "path": "agent/output/adapters/openai/README.md", "content": "# OpenAI Adapter\n\nOpenAI adapter converts universal DSL messages to OpenAI-compatible format.\n\n## Conversion Rules\n\n### Built-in Types (Standard)\n\nThese types are defined in `output.types.go` and have standardized Props structures that all adapters must support:\n\n| Message Type | Constant | Props Structure | OpenAI Format | Description |\n| ------------ | --------------------- | --------------- | ------------------------- | ----------------------------------------- |\n| `text` | `output.TypeText` | `TextProps` | `delta.content` | Plain text or Markdown |\n| `thinking` | `output.TypeThinking` | `ThinkingProps` | `delta.reasoning_content` | Reasoning process (o1 models) |\n| `loading` | `output.TypeLoading` | `LoadingProps` | `delta.reasoning_content` | Loading indicator (shows as thinking) |\n| `tool_call` | `output.TypeToolCall` | `ToolCallProps` | `delta.tool_calls` | Tool/function calls |\n| `error` | `output.TypeError` | `ErrorProps` | `error` | Error messages |\n| `action` | `output.TypeAction` | `ActionProps` | (not sent) | System actions (silent) |\n| `event` | `output.TypeEvent` | `EventProps` | (conditional) | Lifecycle events (stream_start converted) |\n\n### Event Type (Lifecycle Events)\n\nThe `event` type has special handling in the OpenAI adapter:\n\n| Event Name | Conversion | Example Output |\n| -------------- | ------------------------------------------- | --------------------------------------------------- |\n| `stream_start` | Converted to trace link (with i18n support) | 🔍 智能体正在处理 - [查看处理详情](/trace/xxx/view) |\n| Other events | Silent (not sent) | - |\n\n**Conversion Logic for `stream_start`:**\n\n1. **Extract trace data**: Gets `TraceID` from event data\n2. **Check model capabilities**: Determines if model supports reasoning\n3. **Format based on capabilities**:\n - **Reasoning models** (o1, DeepSeek R1): Uses `reasoning_content` field with 🔍 icon\n - **Regular models**: Uses `content` field with 🚀 icon\n4. **Apply i18n**: Uses locale from context for localized text\n5. **Generate trace link**: Creates clickable link to `/trace/{traceID}/view` for standalone viewing\n\n**Example Conversion:**\n\n```go\n// Input (event message)\n{\n \"type\": \"event\",\n \"props\": {\n \"event\": \"stream_start\",\n \"message\": \"Stream started\",\n \"data\": {\n \"trace_id\": \"20251122779905354593\",\n \"request_id\": \"ctx-1763779905679380000\",\n \"chat_id\": \"uP4CWZCMHy84nCw7\"\n }\n }\n}\n\n// Output (reasoning model - Chinese locale)\n{\n \"choices\": [{\n \"delta\": {\n \"reasoning_content\": \"🔍 智能体正在处理 - [查看处理详情](http://localhost:8000/__yao_admin_root/trace/20251122779905354593/view)\\n\"\n }\n }]\n}\n\n// Output (regular model - English locale)\n{\n \"choices\": [{\n \"delta\": {\n \"content\": \"🚀 Assistant is processing - [View process](http://localhost:8000/__yao_admin_root/trace/20251122779905354593/view)\\n\"\n }\n }]\n}\n```\n\n**Internationalization:**\n\nThe adapter uses `i18n.T()` to provide localized text:\n\n| Key | English (en-us) | Chinese (zh-cn) |\n| --------------------- | ----------------------- | --------------- |\n| `output.stream_start` | Assistant is processing | 智能体正在处理 |\n| `output.view_trace` | View process | 查看处理详情 |\n\n### Custom Types\n\nAll other message types (not in the built-in list) are converted to Markdown links:\n\n| Format | Example |\n| ---------------------- | ------------------------------- |\n| `delta.content` (link) | `\"🖼️ [View Image](https://...)` |\n\n## Usage\n\n### Basic Usage\n\n```go\nimport (\n \"github.com/yaoapp/yao/agent/output/adapters/openai\"\n)\n\n// Create adapter with default config\nadapter := openai.NewAdapter()\n\n// Convert message\nchunks, err := adapter.Adapt(msg)\n```\n\n### With Custom Configuration\n\n```go\n// Create adapter with options\nadapter := openai.NewAdapter(\n openai.WithBaseURL(\"https://api.example.com\"),\n openai.WithModel(\"gpt-4\"),\n openai.WithLinkTemplate(\"image\", \"🖼️ [View Image](%s)\"),\n openai.WithLinkTransformer(myOTPTransformer),\n)\n```\n\n### With Link Transformer (OTP)\n\n```go\n// Define OTP transformer\nfunc otpTransformer(url string, msgType string, msgID string) (string, error) {\n // Generate OTP token\n otp := generateOTP(msgID, 3600) // 1 hour expiry\n\n // Create short link with OTP\n shortURL := fmt.Sprintf(\"https://api.example.com/s/%s?t=%s\", msgID, otp)\n\n return shortURL, nil\n}\n\n// Use transformer\nadapter := openai.NewAdapter(\n openai.WithLinkTransformer(otpTransformer),\n)\n```\n\n### Custom Converter\n\n```go\n// Register custom converter for a specific type\nadapter := openai.NewAdapter(\n openai.WithConverter(\"my_widget\", func(msg *output.Message, config *openai.AdapterConfig) ([]interface{}, error) {\n // Custom conversion logic\n return []interface{}{\n // OpenAI format chunk\n }, nil\n }),\n)\n```\n\n## Examples\n\n### Text Message (Built-in Type)\n\n**Input (DSL):**\n\n```json\n{\n \"type\": \"text\",\n \"props\": {\n \"content\": \"Hello world\"\n }\n}\n```\n\nOr using helper:\n\n```go\nmsg := output.NewTextMessage(\"Hello world\")\n```\n\n**Output (OpenAI):**\n\n```json\n{\n \"id\": \"M1\",\n \"object\": \"chat.completion.chunk\",\n \"model\": \"yao-agent\",\n \"choices\": [\n {\n \"delta\": {\n \"content\": \"Hello world\"\n }\n }\n ]\n}\n```\n\n### Image Message\n\n**Input (DSL):**\n\n```json\n{\n \"message_id\": \"M2\",\n \"type\": \"image\",\n \"props\": {\n \"url\": \"https://example.com/avatar.jpg\"\n }\n}\n```\n\n**Output (OpenAI):**\n\n```json\n{\n \"id\": \"M2\",\n \"object\": \"chat.completion.chunk\",\n \"model\": \"yao-agent\",\n \"choices\": [\n {\n \"delta\": {\n \"content\": \"🖼️ [View Image](https://api.example.com/s/M2?t=abc123)\"\n }\n }\n ]\n}\n```\n\n### Button Message\n\n**Input (DSL):**\n\n```json\n{\n \"message_id\": \"M3\",\n \"type\": \"button\",\n \"props\": {\n \"text\": \"Approve\",\n \"action\": \"workflow.approve\"\n }\n}\n```\n\n**Output (OpenAI):**\n\n```json\n{\n \"id\": \"M3\",\n \"object\": \"chat.completion.chunk\",\n \"model\": \"yao-agent\",\n \"choices\": [\n {\n \"delta\": {\n \"content\": \"🔘 [Approve](https://api.example.com/s/M3?t=abc123)\"\n }\n }\n ]\n}\n```\n\n## Link Templates\n\nDefault templates:\n\n```go\n\"image\": \"🖼️ [View Image](%s)\"\n\"audio\": \"🔊 [Play Audio](%s)\"\n\"video\": \"🎬 [Watch Video](%s)\"\n\"file\": \"📎 [Download File](%s)\"\n\"page\": \"📄 [Open Page](%s)\"\n\"table\": \"📊 [View Table](%s)\"\n\"chart\": \"📈 [View Chart](%s)\"\n\"list\": \"📋 [View List](%s)\"\n\"form\": \"📝 [Fill Form](%s)\"\n\"button\": \"🔘 [%s](%s)\" // Special: button text + link\n```\n\nCustomize templates:\n\n```go\nadapter := openai.NewAdapter(\n openai.WithLinkTemplate(\"image\", \"📷 Image: %s\"),\n openai.WithLinkTemplate(\"video\", \"🎥 Watch: %s\"),\n)\n```\n\n## Link Transformer (TODO)\n\nThe link transformer is currently left empty for future implementation of OTP/short link functionality.\n\n**Planned features:**\n\n- Generate one-time password (OTP) for secure access\n- Create short URLs for better readability\n- Set expiration time for links\n- Track link access for analytics\n\n**Example implementation:**\n\n```go\nfunc otpTransformer(url string, msgType string, msgID string) (string, error) {\n // TODO: Implement OTP generation\n // 1. Generate OTP token with expiry\n // 2. Store mapping: token -> (url, msgType, msgID, expiry)\n // 3. Create short URL with token\n // 4. Return short URL\n\n return url, nil // Currently pass-through\n}\n```\n" }, { "path": "agent/output/adapters/openai/adapter.go", "content": "package openai\n\nimport \"github.com/yaoapp/yao/agent/output/message\"\n\n// Adapter is the OpenAI adapter that converts messages to OpenAI format\ntype Adapter struct {\n\tconfig *AdapterConfig\n\tregistry *ConverterRegistry\n}\n\n// NewAdapter creates a new OpenAI adapter with default configuration\nfunc NewAdapter(options ...Option) *Adapter {\n\tadapter := &Adapter{\n\t\tconfig: DefaultAdapterConfig(),\n\t\tregistry: NewConverterRegistry(),\n\t}\n\n\t// Apply options\n\tfor _, opt := range options {\n\t\topt(adapter)\n\t}\n\n\treturn adapter\n}\n\n// Option is a function that configures the adapter\ntype Option func(*Adapter)\n\n// WithBaseURL sets the base URL for generating view links\nfunc WithBaseURL(baseURL string) Option {\n\treturn func(a *Adapter) {\n\t\ta.config.BaseURL = baseURL\n\t}\n}\n\n// WithLinkTemplate sets a custom link template for a message type\nfunc WithLinkTemplate(msgType string, template string) Option {\n\treturn func(a *Adapter) {\n\t\ta.config.LinkTemplates[msgType] = template\n\t}\n}\n\n// WithLinkTransformer sets the link transformer function\nfunc WithLinkTransformer(transformer LinkTransformer) Option {\n\treturn func(a *Adapter) {\n\t\ta.config.LinkTransformer = transformer\n\t}\n}\n\n// WithModel sets the model name for OpenAI responses\nfunc WithModel(model string) Option {\n\treturn func(a *Adapter) {\n\t\ta.config.Model = model\n\t}\n}\n\n// WithCapabilities sets the model capabilities\nfunc WithCapabilities(capabilities *ModelCapabilities) Option {\n\treturn func(a *Adapter) {\n\t\ta.config.Capabilities = capabilities\n\t}\n}\n\n// WithLocale sets the locale for internationalization\nfunc WithLocale(locale string) Option {\n\treturn func(a *Adapter) {\n\t\ta.config.Locale = locale\n\t}\n}\n\n// WithConverter registers a custom converter for a message type\nfunc WithConverter(msgType string, converter ConverterFunc) Option {\n\treturn func(a *Adapter) {\n\t\ta.registry.Register(msgType, converter)\n\t}\n}\n\n// Adapt converts a universal Message to OpenAI-compatible format\nfunc (a *Adapter) Adapt(msg *message.Message) ([]interface{}, error) {\n\t// Handle event messages specially\n\tif msg.Type == message.TypeEvent {\n\t\t// Check if this is a stream_start event\n\t\tif event, ok := msg.Props[\"event\"].(string); ok && event == message.EventStreamStart {\n\t\t\t// Use the stream_start converter\n\t\t\tif converter, exists := a.registry.GetConverter(message.EventStreamStart); exists {\n\t\t\t\treturn converter(msg, a.config)\n\t\t\t}\n\t\t}\n\t\t// Other event messages are CUI-only, skip them\n\t\treturn []interface{}{}, nil // Return empty array, nothing to send\n\t}\n\n\t// Get converter for this message type\n\tconverter, exists := a.registry.GetConverter(msg.Type)\n\tif !exists {\n\t\t// Use default converter for unknown types (convert to link)\n\t\tconverter = convertToLink\n\t}\n\n\t// Convert the message\n\treturn converter(msg, a.config)\n}\n\n// SupportsType checks if the adapter explicitly supports a given message type\nfunc (a *Adapter) SupportsType(msgType string) bool {\n\t_, exists := a.registry.GetConverter(msgType)\n\treturn exists\n}\n\n// GetConfig returns the adapter configuration\nfunc (a *Adapter) GetConfig() *AdapterConfig {\n\treturn a.config\n}\n\n// GetRegistry returns the converter registry\nfunc (a *Adapter) GetRegistry() *ConverterRegistry {\n\treturn a.registry\n}\n" }, { "path": "agent/output/adapters/openai/converter.go", "content": "package openai\n\nimport (\n\t\"fmt\"\n\t\"time\"\n\n\t\"github.com/yaoapp/yao/agent/i18n\"\n\t\"github.com/yaoapp/yao/agent/output/message\"\n)\n\n// ConverterRegistry manages message type converters\ntype ConverterRegistry struct {\n\tconverters map[string]ConverterFunc\n}\n\n// NewConverterRegistry creates a new converter registry with default converters\nfunc NewConverterRegistry() *ConverterRegistry {\n\treturn &ConverterRegistry{\n\t\tconverters: map[string]ConverterFunc{\n\t\t\tmessage.TypeText: convertText,\n\t\t\tmessage.TypeThinking: convertThinking,\n\t\t\tmessage.TypeLoading: convertLoading,\n\t\t\tmessage.TypeToolCall: convertToolCall,\n\t\t\tmessage.TypeError: convertError,\n\t\t\tmessage.TypeImage: convertImage,\n\t\t\tmessage.TypeAudio: convertToLink,\n\t\t\tmessage.TypeVideo: convertToLink,\n\t\t\tmessage.TypeAction: convertAction,\n\t\t\tmessage.EventStreamStart: convertStreamStart, // Handle stream_start events\n\t\t},\n\t}\n}\n\n// Register registers a custom converter for a message type\nfunc (r *ConverterRegistry) Register(msgType string, converter ConverterFunc) {\n\tr.converters[msgType] = converter\n}\n\n// GetConverter retrieves a converter for a given message type.\nfunc (r *ConverterRegistry) GetConverter(msgType string) (ConverterFunc, bool) {\n\tconverter, exists := r.converters[msgType]\n\treturn converter, exists\n}\n\n// Convert converts a message using registered converters\n// If no converter is found, converts to link format\nfunc (r *ConverterRegistry) Convert(msg *message.Message, config *AdapterConfig) ([]interface{}, error) {\n\t// Check for registered converter\n\tif converter, exists := r.converters[msg.Type]; exists {\n\t\treturn converter(msg, config)\n\t}\n\n\t// Fallback: convert to link format\n\treturn convertToLink(msg, config)\n}\n\n// convertText converts text messages to OpenAI format\nfunc convertText(msg *message.Message, config *AdapterConfig) ([]interface{}, error) {\n\tcontent := getStringProp(msg.Props, \"content\", \"\")\n\n\treturn []interface{}{\n\t\tcreateOpenAIChunk(msg.MessageID, config.Model, map[string]interface{}{\n\t\t\t\"content\": content,\n\t\t}),\n\t}, nil\n}\n\n// convertThinking converts thinking messages to OpenAI reasoning format (o1 series)\nfunc convertThinking(msg *message.Message, config *AdapterConfig) ([]interface{}, error) {\n\tcontent := getStringProp(msg.Props, \"content\", \"\")\n\n\treturn []interface{}{\n\t\tcreateOpenAIChunk(msg.MessageID, config.Model, map[string]interface{}{\n\t\t\t\"reasoning_content\": content,\n\t\t}),\n\t}, nil\n}\n\n// convertLoading converts loading messages to OpenAI reasoning format\n// This makes loading messages visible in standard OpenAI clients as thinking process\nfunc convertLoading(msg *message.Message, config *AdapterConfig) ([]interface{}, error) {\n\tmessage := getStringProp(msg.Props, \"message\", \"Processing...\")\n\n\t// Convert loading to reasoning_content so it shows in OpenAI clients\n\treturn []interface{}{\n\t\tcreateOpenAIChunk(msg.MessageID, config.Model, map[string]interface{}{\n\t\t\t\"reasoning_content\": message,\n\t\t}),\n\t}, nil\n}\n\n// convertToolCall converts tool_call messages to OpenAI format\nfunc convertToolCall(msg *message.Message, config *AdapterConfig) ([]interface{}, error) {\n\t// Tool call format varies, pass through the props\n\ttoolCalls := []map[string]interface{}{}\n\n\t// If props contain tool call data, use it\n\tif id, ok := msg.Props[\"id\"].(string); ok {\n\t\ttoolCall := map[string]interface{}{\n\t\t\t\"id\": id,\n\t\t\t\"type\": \"function\",\n\t\t}\n\n\t\tif function, ok := msg.Props[\"function\"].(map[string]interface{}); ok {\n\t\t\ttoolCall[\"function\"] = function\n\t\t}\n\n\t\ttoolCalls = append(toolCalls, toolCall)\n\t}\n\n\treturn []interface{}{\n\t\tcreateOpenAIChunk(msg.MessageID, config.Model, map[string]interface{}{\n\t\t\t\"tool_calls\": toolCalls,\n\t\t}),\n\t}, nil\n}\n\n// convertError converts error messages to OpenAI error format\nfunc convertError(msg *message.Message, config *AdapterConfig) ([]interface{}, error) {\n\tmessage := getStringProp(msg.Props, \"message\", \"An error occurred\")\n\tcode := getStringProp(msg.Props, \"code\", \"server_error\")\n\n\treturn []interface{}{\n\t\tmap[string]interface{}{\n\t\t\t\"error\": map[string]interface{}{\n\t\t\t\t\"message\": message,\n\t\t\t\t\"type\": code,\n\t\t\t\t\"code\": code,\n\t\t\t},\n\t\t},\n\t}, nil\n}\n\n// convertAction converts action messages to nothing (silent in OpenAI clients)\n// Action messages are system-level commands (open panel, navigate, etc.)\n// and should not be sent to standard chat clients\nfunc convertAction(msg *message.Message, config *AdapterConfig) ([]interface{}, error) {\n\t// Return empty slice - no output for action messages in OpenAI format\n\treturn []interface{}{}, nil\n}\n\n// convertStreamStart converts stream_start event to OpenAI format\n// If model supports reasoning: converts to reasoning_content (thinking)\n// Otherwise: converts to regular Markdown text with trace link\nfunc convertStreamStart(msg *message.Message, config *AdapterConfig) ([]interface{}, error) {\n\t// Extract stream_start data from props\n\tdata, ok := msg.Props[\"data\"]\n\tif !ok {\n\t\t// No data, skip this message\n\t\treturn []interface{}{}, nil\n\t}\n\n\t// Try to convert to EventStreamStartData\n\tvar startData message.EventStreamStartData\n\tswitch v := data.(type) {\n\tcase message.EventStreamStartData:\n\t\tstartData = v\n\tcase map[string]interface{}:\n\t\t// If it's a map, try to extract traceID\n\t\tif traceID, ok := v[\"trace_id\"].(string); ok {\n\t\t\tstartData.TraceID = traceID\n\t\t}\n\t\tif requestID, ok := v[\"request_id\"].(string); ok {\n\t\t\tstartData.RequestID = requestID\n\t\t}\n\tdefault:\n\t\t// Unknown data type, skip\n\t\treturn []interface{}{}, nil\n\t}\n\n\t// Check if we have a trace ID to link to\n\tif startData.TraceID == \"\" {\n\t\t// No trace ID, skip this message\n\t\treturn []interface{}{}, nil\n\t}\n\n\t// Generate trace link\n\ttraceLink := generateTraceLink(startData.TraceID, config)\n\n\t// Check if model supports reasoning\n\tsupportsReasoning := false\n\tif config.Capabilities != nil && config.Capabilities.Reasoning != nil {\n\t\tsupportsReasoning = *config.Capabilities.Reasoning\n\t}\n\n\t// Get localized text using i18n\n\tstreamStartText := i18n.T(config.Locale, \"output.stream_start\")\n\tviewTraceText := i18n.T(config.Locale, \"output.view_trace\")\n\n\t// Convert based on reasoning support\n\tif supportsReasoning {\n\t\t// Convert to thinking format (reasoning_content)\n\t\t// Reasoning models display this as part of the thinking process\n\t\tcontent := fmt.Sprintf(\"🔍 %s - [%s](%s)\\n\", streamStartText, viewTraceText, traceLink)\n\t\tchunk := createOpenAIChunk(msg.MessageID, config.Model, map[string]interface{}{\n\t\t\t\"reasoning_content\": content,\n\t\t})\n\t\treturn []interface{}{chunk}, nil\n\t}\n\n\t// Convert to regular Markdown text\n\tcontent := fmt.Sprintf(\"🚀 %s - [%s](%s)\\n\", streamStartText, viewTraceText, traceLink)\n\tchunk := createOpenAIChunk(msg.MessageID, config.Model, map[string]interface{}{\n\t\t\"content\": content,\n\t})\n\treturn []interface{}{chunk}, nil\n}\n\n// generateTraceLink generates a trace link URL\n// Uses 'view' mode (clean page without sidebar) for better viewing experience in chat\nfunc generateTraceLink(traceID string, config *AdapterConfig) string {\n\tbaseURL := config.BaseURL\n\tif baseURL == \"\" {\n\t\t// If no base URL, return a relative link\n\t\treturn fmt.Sprintf(\"/trace/%s/view\", traceID)\n\t}\n\treturn fmt.Sprintf(\"%s/trace/%s/view\", baseURL, traceID)\n}\n\n// convertImage converts image messages to Markdown image format\n// Uses ![alt](url) which displays inline in Markdown-supporting clients\nfunc convertImage(msg *message.Message, config *AdapterConfig) ([]interface{}, error) {\n\t// Get URL\n\turl, ok := msg.Props[\"url\"].(string)\n\tif !ok || url == \"\" {\n\t\treturn nil, fmt.Errorf(\"image message missing url\")\n\t}\n\n\t// Transform URL if transformer is provided\n\tif config.LinkTransformer != nil {\n\t\ttransformedURL, err := config.LinkTransformer(url, msg.Type, msg.MessageID)\n\t\tif err != nil {\n\t\t\treturn nil, err\n\t\t}\n\t\turl = transformedURL\n\t}\n\n\t// Get alt text (default to \"Image\")\n\talt := getStringProp(msg.Props, \"alt\", \"Image\")\n\n\t// Format as Markdown image: ![alt](url)\n\ttemplate := getLinkTemplate(msg.Type, config)\n\ttext := fmt.Sprintf(template, alt, url)\n\n\treturn []interface{}{\n\t\tcreateOpenAIChunk(msg.MessageID, config.Model, map[string]interface{}{\n\t\t\t\"content\": text,\n\t\t}),\n\t}, nil\n}\n\n// convertToLink converts any message type to a Markdown link format\nfunc convertToLink(msg *message.Message, config *AdapterConfig) ([]interface{}, error) {\n\t// Generate link\n\tlink, err := generateViewLink(msg, config)\n\tif err != nil {\n\t\treturn nil, err\n\t}\n\n\t// Get template\n\ttemplate := getLinkTemplate(msg.Type, config)\n\n\t// Format text\n\tvar text string\n\tif msg.Type == \"button\" {\n\t\t// Button is special: needs button text\n\t\tbuttonText := getStringProp(msg.Props, \"text\", \"Button\")\n\t\ttext = fmt.Sprintf(template, buttonText, link)\n\t} else {\n\t\ttext = fmt.Sprintf(template, link)\n\t}\n\n\treturn []interface{}{\n\t\tcreateOpenAIChunk(msg.MessageID, config.Model, map[string]interface{}{\n\t\t\t\"content\": text,\n\t\t}),\n\t}, nil\n}\n\n// generateViewLink generates a view link for a message\nfunc generateViewLink(msg *message.Message, config *AdapterConfig) (string, error) {\n\t// If Props contains a URL, use it\n\tif url, ok := msg.Props[\"url\"].(string); ok {\n\t\t// Transform URL if transformer is provided\n\t\tif config.LinkTransformer != nil {\n\t\t\treturn config.LinkTransformer(url, msg.Type, msg.MessageID)\n\t\t}\n\t\treturn url, nil\n\t}\n\n\t// Generate view link: {baseURL}/agent/view/{type}/{id}\n\tbaseURL := config.BaseURL\n\tif baseURL == \"\" {\n\t\tbaseURL = \"\" // TODO: Get from environment or context\n\t}\n\n\tviewURL := fmt.Sprintf(\"%s/agent/view/%s/%s\", baseURL, msg.Type, msg.MessageID)\n\n\t// Transform URL if transformer is provided\n\tif config.LinkTransformer != nil {\n\t\treturn config.LinkTransformer(viewURL, msg.Type, msg.MessageID)\n\t}\n\n\treturn viewURL, nil\n}\n\n// getLinkTemplate gets the link template for a message type\nfunc getLinkTemplate(msgType string, config *AdapterConfig) string {\n\tif template, exists := config.LinkTemplates[msgType]; exists {\n\t\treturn template\n\t}\n\n\t// Default fallback template\n\treturn \"📎 [View %s](\" + msgType + \")\"\n}\n\n// createOpenAIChunk creates an OpenAI chat completion chunk\nfunc createOpenAIChunk(id string, model string, delta map[string]interface{}) map[string]interface{} {\n\treturn map[string]interface{}{\n\t\t\"id\": id,\n\t\t\"object\": \"chat.completion.chunk\",\n\t\t\"created\": time.Now().Unix(),\n\t\t\"model\": model,\n\t\t\"choices\": []map[string]interface{}{\n\t\t\t{\n\t\t\t\t\"index\": 0,\n\t\t\t\t\"delta\": delta,\n\t\t\t\t\"finish_reason\": nil,\n\t\t\t},\n\t\t},\n\t}\n}\n\n// getStringProp safely gets a string property from props\nfunc getStringProp(props map[string]interface{}, key string, defaultValue string) string {\n\tif val, ok := props[key].(string); ok {\n\t\treturn val\n\t}\n\treturn defaultValue\n}\n" }, { "path": "agent/output/adapters/openai/types.go", "content": "package openai\n\nimport \"github.com/yaoapp/yao/agent/output/message\"\n\n// ConverterFunc converts a message to OpenAI format chunks\ntype ConverterFunc func(msg *message.Message, config *AdapterConfig) ([]interface{}, error)\n\n// LinkTransformer transforms a URL to a secure link (with OTP, short URL, etc.)\n// Returns the transformed link or error\ntype LinkTransformer func(url string, msgType string, msgID string) (string, error)\n\n// AdapterConfig holds the configuration for OpenAI adapter\ntype AdapterConfig struct {\n\t// BaseURL is the base URL for generating view links\n\t// Example: \"https://api.example.com\"\n\tBaseURL string\n\n\t// LinkTemplates defines the Markdown template for each message type\n\t// %s will be replaced with the link\n\t// Example: \"🖼️ [View Image](%s)\"\n\tLinkTemplates map[string]string\n\n\t// LinkTransformer transforms URLs to secure links with OTP\n\t// If nil, URLs are used as-is\n\tLinkTransformer LinkTransformer\n\n\t// Model name to include in OpenAI responses\n\tModel string\n\n\t// Capabilities holds the model capabilities\n\t// Used to determine how to convert certain message types (e.g., stream_start)\n\tCapabilities *ModelCapabilities\n\n\t// Locale for internationalization (e.g., \"en-US\", \"zh-CN\")\n\tLocale string\n}\n\n// ModelCapabilities is a simplified version of openai.Capabilities\n// We use a local type to avoid circular dependencies\ntype ModelCapabilities struct {\n\tReasoning *bool // Supports reasoning/thinking mode (o1, DeepSeek R1)\n}\n\n// DefaultLinkTemplates provides default Markdown templates for non-text message types\nvar DefaultLinkTemplates = map[string]string{\n\t\"image\": \"![%s](%s)\", // Markdown image: ![alt](url) - displays inline\n\t\"audio\": \"🔊 [Play Audio](%s)\", // Link (audio can't display inline in Markdown)\n\t\"video\": \"🎬 [Watch Video](%s)\", // Link (video can't display inline in Markdown)\n\t\"file\": \"📎 [Download File](%s)\",\n\t\"page\": \"📄 [Open Page](%s)\",\n\t\"table\": \"📊 [View Table](%s)\",\n\t\"chart\": \"📈 [View Chart](%s)\",\n\t\"list\": \"📋 [View List](%s)\",\n\t\"form\": \"📝 [Fill Form](%s)\",\n\t\"button\": \"🔘 [%s](%s)\", // Special: button needs two params (text, link)\n}\n\n// DefaultAdapterConfig returns a default adapter configuration\nfunc DefaultAdapterConfig() *AdapterConfig {\n\treturn &AdapterConfig{\n\t\tBaseURL: \"\", // Will be set from environment or context\n\t\tLinkTemplates: copyLinkTemplates(DefaultLinkTemplates),\n\t\tLinkTransformer: nil, // No transformation by default\n\t\tModel: \"yao-agent\",\n\t}\n}\n\n// copyLinkTemplates creates a copy of link templates\nfunc copyLinkTemplates(templates map[string]string) map[string]string {\n\tcopy := make(map[string]string, len(templates))\n\tfor k, v := range templates {\n\t\tcopy[k] = v\n\t}\n\treturn copy\n}\n" }, { "path": "agent/output/adapters/openai/writer.go", "content": "package openai\n\nimport (\n\t\"encoding/json\"\n\t\"net/http\"\n\n\t\"github.com/yaoapp/yao/agent/i18n\"\n\t\"github.com/yaoapp/yao/agent/output/message\"\n\ttraceTypes \"github.com/yaoapp/yao/trace/types\"\n)\n\n// Writer implements the message.Writer interface for OpenAI-compatible clients\ntype Writer struct {\n\tWriter http.ResponseWriter\n\tTrace traceTypes.Manager\n\tLocale string\n\tadapter *Adapter\n\tfirstChunk bool // Track if this is the first chunk to add role\n}\n\n// NewWriter creates a new OpenAI writer\nfunc NewWriter(options message.Options) (*Writer, error) {\n\t// Get model capabilities from context (set by assistant)\n\tvar capabilities *ModelCapabilities\n\tif options.Capabilities != nil && options.Capabilities.Reasoning {\n\t\tv := true\n\t\tcapabilities = &ModelCapabilities{\n\t\t\tReasoning: &v,\n\t\t}\n\t}\n\n\t// Create adapter with capabilities, base URL, and locale\n\tadapter := NewAdapter(\n\t\tWithCapabilities(capabilities),\n\t\tWithBaseURL(getBaseURL(options.BaseURL)),\n\t\tWithLocale(options.Locale),\n\t)\n\n\treturn &Writer{\n\t\tadapter: adapter,\n\t\tWriter: options.Writer,\n\t\tLocale: options.Locale,\n\t\tfirstChunk: true, // First chunk should include role\n\t}, nil\n}\n\n// getBaseURL gets the base URL from context or environment\nfunc getBaseURL(baseURL string) string {\n\t// @todo: get from context metadata\n\treturn \"http://localhost:8000/__yao_admin_root\"\n\n\t// // Try to get from context metadata\n\t// if ctx.Metadata != nil {\n\t// \tif baseURL, ok := ctx.Metadata[\"base_url\"].(string); ok && baseURL != \"\" {\n\t// \t\treturn baseURL\n\t// \t}\n\t// }\n\n\t// // TODO: Get from environment variable or config\n\t// return \"\"\n}\n\n// Write writes a single message to the output stream\nfunc (w *Writer) Write(msg *message.Message) error {\n\t// Convert message to OpenAI format using adapter\n\tchunks, err := w.adapter.Adapt(msg)\n\tif err != nil {\n\t\tif w.Trace != nil {\n\t\t\tw.Trace.Error(i18n.T(w.Locale, \"output.openai.writer.adapt_error\"), map[string]any{ // \"OpenAI Writer: Failed to adapt message\"\n\t\t\t\t\"error\": err.Error(),\n\t\t\t\t\"message_type\": msg.Type,\n\t\t\t})\n\t\t}\n\t\treturn err\n\t}\n\n\t// Send each chunk\n\tfor _, chunk := range chunks {\n\t\t// Add role to first text chunk\n\t\tif w.firstChunk && (msg.Type == message.TypeText || msg.Type == message.TypeThinking) {\n\t\t\tif chunkMap, ok := chunk.(map[string]interface{}); ok {\n\t\t\t\tif choices, ok := chunkMap[\"choices\"].([]map[string]interface{}); ok && len(choices) > 0 {\n\t\t\t\t\tif delta, ok := choices[0][\"delta\"].(map[string]interface{}); ok {\n\t\t\t\t\t\tdelta[\"role\"] = \"assistant\"\n\t\t\t\t\t\tw.firstChunk = false\n\t\t\t\t\t}\n\t\t\t\t}\n\t\t\t}\n\t\t}\n\n\t\tif err := w.sendChunk(chunk); err != nil {\n\t\t\tif w.Trace != nil {\n\t\t\t\tw.Trace.Error(i18n.T(w.Locale, \"output.openai.writer.chunk_error\"), map[string]any{\"error\": err.Error()}) // \"OpenAI Writer: Failed to send chunk\"\n\t\t\t}\n\t\t\treturn err\n\t\t}\n\t}\n\n\treturn nil\n}\n\n// WriteGroup writes a message group to the output stream\nfunc (w *Writer) WriteGroup(group *message.Group) error {\n\t// For OpenAI, we don't send group markers\n\t// Just send each message individually\n\tfor _, msg := range group.Messages {\n\t\tif err := w.Write(msg); err != nil {\n\t\t\tif w.Trace != nil {\n\t\t\t\tw.Trace.Error(i18n.T(w.Locale, \"output.openai.writer.group_error\"), map[string]any{ // \"OpenAI Writer: Failed to write message in group\"\n\t\t\t\t\t\"error\": err.Error(),\n\t\t\t\t\t\"group_id\": group.ID,\n\t\t\t\t\t\"message_type\": msg.Type,\n\t\t\t\t})\n\t\t\t}\n\t\t\treturn err\n\t\t}\n\t}\n\n\treturn nil\n}\n\n// Flush flushes any buffered data to the output stream\nfunc (w *Writer) Flush() error {\n\t// For SSE, we don't need explicit flushing\n\t// The underlying connection handles it\n\treturn nil\n}\n\n// Close closes the writer and cleans up resources\nfunc (w *Writer) Close() error {\n\t// Send final [DONE] message for OpenAI compatibility\n\treturn w.sendDone()\n}\n\nfunc (w *Writer) sendData(data []byte) error {\n\tif w.Writer == nil {\n\t\treturn nil // No writer, silently ignore\n\t}\n\t_, err := w.Writer.Write(data)\n\treturn err\n}\n\nfunc (w *Writer) flush() error {\n\tif w.Writer == nil {\n\t\treturn nil // No writer, silently ignore\n\t}\n\tif flusher, ok := w.Writer.(interface{ Flush() }); ok {\n\t\tflusher.Flush()\n\t}\n\treturn nil\n}\n\n// sendChunk sends a chunk to the output stream in SSE format\nfunc (w *Writer) sendChunk(chunk interface{}) error {\n\t// Convert chunk to JSON\n\tdata, err := json.Marshal(chunk)\n\tif err != nil {\n\t\tif w.Trace != nil {\n\t\t\tw.Trace.Error(i18n.T(w.Locale, \"output.openai.writer.marshal_error\"), map[string]any{\"error\": err.Error()}) // \"OpenAI Writer: Failed to marshal chunk\"\n\t\t}\n\t\treturn err\n\t}\n\n\t// Format as SSE: \"data: {json}\\n\\n\"\n\tsseData := append([]byte(\"data: \"), data...)\n\tsseData = append(sseData, []byte(\"\\n\\n\")...)\n\n\t// Log outgoing data to trace for debugging\n\tif w.Trace != nil {\n\t\tw.Trace.Debug(\"OpenAI Writer: Sending chunk to client\", map[string]any{\n\t\t\t\"data\": string(data),\n\t\t})\n\t}\n\n\t// Send via context's writer\n\tif err := w.sendData(sseData); err != nil {\n\t\tif w.Trace != nil {\n\t\t\tw.Trace.Error(i18n.T(w.Locale, \"output.openai.writer.send_error\"), map[string]any{\"error\": err.Error()}) // \"OpenAI Writer: Failed to send data to client\"\n\t\t}\n\t\treturn err\n\t}\n\n\t// Flush immediately to ensure real-time streaming\n\t// Cast to http.ResponseWriter and call Flush if available\n\tw.flush()\n\n\treturn nil\n}\n\n// sendDone sends the final [DONE] message\nfunc (w *Writer) sendDone() error {\n\t// Log completion to trace\n\tif w.Trace != nil {\n\t\tw.Trace.Debug(\"OpenAI Writer: Sending [DONE] to client\")\n\t}\n\n\t// OpenAI SSE format uses \"data: [DONE]\\n\\n\" to signal completion\n\tdoneData := []byte(\"data: [DONE]\\n\\n\")\n\tif err := w.sendData(doneData); err != nil {\n\t\tif w.Trace != nil {\n\t\t\tw.Trace.Error(i18n.T(w.Locale, \"output.openai.writer.done_error\"), map[string]any{\"error\": err.Error()}) // \"OpenAI Writer: Failed to send [DONE] to client\"\n\t\t}\n\t\treturn err\n\t}\n\n\t// Flush the final [DONE] message\n\tw.flush()\n\treturn nil\n}\n" }, { "path": "agent/output/builtin.go", "content": "package output\n\nimport (\n\t\"github.com/yaoapp/yao/agent/output/message\"\n)\n\n// Helper functions for creating built-in message types\n\n// NewUserInputMessage creates a user input message (for frontend display)\n// content can be string or []ContentPart for multimodal content\nfunc NewUserInputMessage(content interface{}, role, name string) *message.Message {\n\tprops := map[string]interface{}{\n\t\t\"content\": content,\n\t}\n\tif role != \"\" {\n\t\tprops[\"role\"] = role\n\t}\n\tif name != \"\" {\n\t\tprops[\"name\"] = name\n\t}\n\treturn &message.Message{\n\t\tType: message.TypeUserInput,\n\t\tProps: props,\n\t}\n}\n\n// NewTextMessage creates a text message\nfunc NewTextMessage(content string) *message.Message {\n\treturn &message.Message{\n\t\tType: message.TypeText,\n\t\tProps: map[string]interface{}{\n\t\t\t\"content\": content,\n\t\t},\n\t}\n}\n\n// NewThinkingMessage creates a thinking message\nfunc NewThinkingMessage(content string) *message.Message {\n\treturn &message.Message{\n\t\tType: message.TypeThinking,\n\t\tProps: map[string]interface{}{\n\t\t\t\"content\": content,\n\t\t},\n\t}\n}\n\n// NewLoadingMessage creates a loading message\nfunc NewLoadingMessage(msg string) *message.Message {\n\treturn &message.Message{\n\t\tType: message.TypeLoading,\n\t\tProps: map[string]interface{}{\n\t\t\t\"message\": msg,\n\t\t},\n\t}\n}\n\n// NewToolCallMessage creates a tool call message\nfunc NewToolCallMessage(id, name, arguments string) *message.Message {\n\treturn &message.Message{\n\t\tType: message.TypeToolCall,\n\t\tProps: map[string]interface{}{\n\t\t\t\"id\": id,\n\t\t\t\"name\": name,\n\t\t\t\"arguments\": arguments,\n\t\t},\n\t}\n}\n\n// NewErrorMessage creates an error message\nfunc NewErrorMessage(msg, code string) *message.Message {\n\treturn &message.Message{\n\t\tType: message.TypeError,\n\t\tProps: map[string]interface{}{\n\t\t\t\"message\": msg,\n\t\t\t\"code\": code,\n\t\t},\n\t}\n}\n\n// NewActionMessage creates an action message\nfunc NewActionMessage(name string, payload map[string]interface{}) *message.Message {\n\treturn &message.Message{\n\t\tType: message.TypeAction,\n\t\tProps: map[string]interface{}{\n\t\t\t\"name\": name,\n\t\t\t\"payload\": payload,\n\t\t},\n\t}\n}\n\n// NewEventMessage creates an event message\nfunc NewEventMessage(event string, msg string, data interface{}) *message.Message {\n\treturn &message.Message{\n\t\tType: message.TypeEvent,\n\t\tProps: map[string]interface{}{\n\t\t\t\"event\": event,\n\t\t\t\"message\": msg,\n\t\t\t\"data\": data,\n\t\t},\n\t}\n}\n\n// NewImageMessage creates an image message\nfunc NewImageMessage(url string, alt string) *message.Message {\n\treturn &message.Message{\n\t\tType: message.TypeImage,\n\t\tProps: map[string]interface{}{\n\t\t\t\"url\": url,\n\t\t\t\"alt\": alt,\n\t\t},\n\t}\n}\n\n// NewAudioMessage creates an audio message\nfunc NewAudioMessage(url string, format string) *message.Message {\n\treturn &message.Message{\n\t\tType: message.TypeAudio,\n\t\tProps: map[string]interface{}{\n\t\t\t\"url\": url,\n\t\t\t\"format\": format,\n\t\t},\n\t}\n}\n\n// NewVideoMessage creates a video message\nfunc NewVideoMessage(url string) *message.Message {\n\treturn &message.Message{\n\t\tType: message.TypeVideo,\n\t\tProps: map[string]interface{}{\n\t\t\t\"url\": url,\n\t\t},\n\t}\n}\n\n// IsBuiltinType checks if a message type is a built-in type\nfunc IsBuiltinType(msgType string) bool {\n\tswitch msgType {\n\tcase message.TypeUserInput, message.TypeText, message.TypeThinking, message.TypeLoading, message.TypeToolCall, message.TypeError, message.TypeImage, message.TypeAudio, message.TypeVideo, message.TypeAction, message.TypeEvent:\n\t\treturn true\n\tdefault:\n\t\treturn false\n\t}\n}\n\n// GenerateID generates a unique message ID using nanoid\n// Deprecated: Use message.GenerateMessageID(), message.GenerateChunkID(),\n// message.GenerateBlockID(), or message.GenerateThreadID() instead\nfunc GenerateID() string {\n\treturn message.GenerateNanoID()\n}\n" }, { "path": "agent/output/jsapi/README.md", "content": "# Output JSAPI\n\nThe Output JSAPI provides a JavaScript interface for sending output messages to clients from scripts (e.g., hooks, processes). It wraps the Go `output` package functionality and provides a convenient API for sending messages and message groups.\n\n## Overview\n\nThe Output object allows you to:\n\n- Send individual messages to clients in various formats (text, error, loading, etc.)\n- Send groups of related messages\n- Support streaming with delta updates\n- Handle different message types with custom properties\n\n## Constructor\n\n### `new Output(ctx)`\n\nCreates a new Output instance.\n\n**Parameters:**\n\n- `ctx` (Context): The agent context object\n\n**Returns:**\n\n- Output instance\n\n**Example:**\n\n```javascript\nfunction Create(ctx, messages) {\n const output = new Output(ctx);\n // Use output methods...\n}\n```\n\n## Methods\n\n### `Send(message)`\n\nSends a single message to the client.\n\n**Parameters:**\n\n- `message` (string | object): The message to send\n - If string: Automatically converted to a text message\n - If object: Must have a `type` field and optional `props` and other fields\n\n**Returns:**\n\n- Output instance (for chaining)\n\n**Message Object Structure:**\n\n```javascript\n{\n type: string, // Required: Message type (e.g., \"text\", \"error\", \"loading\")\n props: object, // Optional: Message properties (type-specific)\n id: string, // Optional: Message ID (for streaming)\n delta: boolean, // Optional: Whether this is a delta update\n done: boolean, // Optional: Whether the message is complete\n delta_path: string, // Optional: Path for delta updates (e.g., \"content\")\n delta_action: string, // Optional: Delta action (\"append\", \"replace\", \"merge\", \"set\")\n type_change: boolean, // Optional: Whether this is a type correction\n group_id: string, // Optional: Parent message group ID\n group_start: boolean, // Optional: Marks the start of a message group\n group_end: boolean, // Optional: Marks the end of a message group\n metadata: { // Optional: Message metadata\n timestamp: number,\n sequence: number,\n trace_id: string\n }\n}\n```\n\n**Examples:**\n\nSend a simple text message (shorthand):\n\n```javascript\noutput.Send(\"Hello, world!\");\n```\n\nSend a text message (full):\n\n```javascript\noutput.Send({\n type: \"text\",\n props: {\n content: \"Hello, world!\",\n },\n});\n```\n\nSend an error message:\n\n```javascript\noutput.Send({\n type: \"error\",\n props: {\n message: \"Something went wrong\",\n code: \"ERR_001\",\n details: \"Additional error details\",\n },\n});\n```\n\nSend a loading indicator:\n\n```javascript\noutput.Send({\n type: \"loading\",\n props: {\n message: \"Searching knowledge base...\",\n },\n});\n```\n\nSend streaming text with delta updates:\n\n```javascript\n// First chunk\noutput.Send({\n type: \"text\",\n id: \"msg-1\",\n props: { content: \"Hello\" },\n delta: true,\n done: false,\n});\n\n// Subsequent chunks\noutput.Send({\n type: \"text\",\n id: \"msg-1\",\n props: { content: \" world\" },\n delta: true,\n delta_path: \"content\",\n delta_action: \"append\",\n done: false,\n});\n\n// Final chunk\noutput.Send({\n type: \"text\",\n id: \"msg-1\",\n props: { content: \"!\" },\n delta: true,\n delta_path: \"content\",\n delta_action: \"append\",\n done: true,\n});\n```\n\nChain multiple sends:\n\n```javascript\noutput\n .Send(\"First message\")\n .Send(\"Second message\")\n .Send({ type: \"loading\", props: { message: \"Processing...\" } });\n```\n\n### `SendGroup(group)`\n\nSends a group of related messages.\n\n**Parameters:**\n\n- `group` (object): The message group\n - `id` (string): Required - Group ID\n - `messages` (array): Required - Array of message objects\n - `metadata` (object): Optional - Group metadata\n\n**Returns:**\n\n- Output instance (for chaining)\n\n**Group Object Structure:**\n\n```javascript\n{\n id: string, // Required: Message group ID\n messages: [ // Required: Array of messages\n {\n type: string,\n props: object,\n // ... other message fields\n }\n ],\n metadata: { // Optional: Group metadata\n timestamp: number,\n sequence: number,\n trace_id: string\n }\n}\n```\n\n**Examples:**\n\nSend a simple message group:\n\n```javascript\noutput.SendGroup({\n id: \"search-results\",\n messages: [\n { type: \"text\", props: { content: \"Found 3 results:\" } },\n { type: \"text\", props: { content: \"Result 1\" } },\n { type: \"text\", props: { content: \"Result 2\" } },\n { type: \"text\", props: { content: \"Result 3\" } },\n ],\n});\n```\n\nSend a group with metadata:\n\n```javascript\noutput.SendGroup({\n id: \"analysis-group\",\n messages: [\n { type: \"loading\", props: { message: \"Analyzing data...\" } },\n { type: \"text\", props: { content: \"Analysis complete\" } },\n ],\n metadata: {\n timestamp: Date.now(),\n sequence: 1,\n trace_id: \"trace-123\",\n },\n});\n```\n\n## Built-in Message Types\n\nThe Output JSAPI supports all built-in message types defined in the output package:\n\n### User Interaction Types\n\n- **`user_input`**: User input message (frontend display only)\n ```javascript\n { type: \"user_input\", props: { content: \"User's message\", role: \"user\" } }\n ```\n\n### Content Types\n\n- **`text`**: Plain text or Markdown content\n\n ```javascript\n { type: \"text\", props: { content: \"Hello **world**\" } }\n ```\n\n- **`thinking`**: Reasoning/thinking process (e.g., o1 models)\n\n ```javascript\n { type: \"thinking\", props: { content: \"Let me think about this...\" } }\n ```\n\n- **`loading`**: Loading/processing indicator\n\n ```javascript\n { type: \"loading\", props: { message: \"Processing...\" } }\n ```\n\n- **`tool_call`**: LLM tool/function call\n\n ```javascript\n {\n type: \"tool_call\",\n props: {\n id: \"call_123\",\n name: \"search\",\n arguments: \"{\\\"query\\\":\\\"test\\\"}\"\n }\n }\n ```\n\n- **`error`**: Error message\n ```javascript\n {\n type: \"error\",\n props: {\n message: \"Error occurred\",\n code: \"ERR_001\",\n details: \"More info\"\n }\n }\n ```\n\n### Media Types\n\n- **`image`**: Image content\n\n ```javascript\n {\n type: \"image\",\n props: {\n url: \"https://example.com/image.jpg\",\n alt: \"Description\",\n width: 800,\n height: 600\n }\n }\n ```\n\n- **`audio`**: Audio content\n\n ```javascript\n {\n type: \"audio\",\n props: {\n url: \"https://example.com/audio.mp3\",\n format: \"mp3\",\n duration: 120.5\n }\n }\n ```\n\n- **`video`**: Video content\n ```javascript\n {\n type: \"video\",\n props: {\n url: \"https://example.com/video.mp4\",\n format: \"mp4\",\n duration: 300\n }\n }\n ```\n\n### System Types\n\n- **`action`**: System action (silent in OpenAI clients)\n\n ```javascript\n {\n type: \"action\",\n props: {\n name: \"open_panel\",\n payload: { panel_id: \"settings\" }\n }\n }\n ```\n\n- **`event`**: Lifecycle event (CUI only, silent in OpenAI clients)\n ```javascript\n {\n type: \"event\",\n props: {\n event: \"stream_start\",\n message: \"Starting stream...\"\n }\n }\n ```\n\n## Usage in Hooks\n\n### Create Hook Example\n\n```javascript\n/**\n * Create hook - Called before sending messages to the LLM\n * @param {Context} ctx - Agent context\n * @param {Array} messages - User messages\n * @returns {Object} Hook response\n */\nfunction Create(ctx, messages) {\n const output = new Output(ctx);\n\n // Send a loading indicator\n output.Send({\n type: \"loading\",\n props: { message: \"Processing your request...\" },\n });\n\n // Send custom messages to the user\n output.Send({\n type: \"text\",\n props: { content: \"I'm thinking about your question...\" },\n });\n\n // Return hook response\n return {\n messages: messages,\n temperature: 0.7,\n };\n}\n```\n\n### Done Hook Example\n\n```javascript\n/**\n * Done hook - Called after assistant completes response\n * @param {Context} ctx - Agent context\n * @param {Array} messages - Conversation messages\n * @param {Object} response - Assistant response\n */\nfunction Done(ctx, messages, response) {\n const output = new Output(ctx);\n\n // Send a completion message\n output.Send({\n type: \"text\",\n props: { content: \"Response complete!\" },\n });\n\n // Send an action\n output.Send({\n type: \"action\",\n props: {\n name: \"save_conversation\",\n payload: { chat_id: ctx.chat_id },\n },\n });\n}\n```\n\n### Progress Updates Example\n\n```javascript\nfunction ProcessData(ctx, data) {\n const output = new Output(ctx);\n\n // Show progress\n const steps = [\"Loading\", \"Processing\", \"Analyzing\", \"Complete\"];\n\n for (let i = 0; i < steps.length; i++) {\n output.Send({\n type: \"loading\",\n props: {\n message: `${steps[i]}... (${i + 1}/${steps.length})`,\n },\n });\n\n // Do some work...\n processStep(i);\n }\n\n // Send final result\n output.Send({\n type: \"text\",\n props: { content: \"All done!\" },\n });\n}\n```\n\n## Error Handling\n\nThe Output JSAPI throws exceptions for invalid parameters:\n\n```javascript\ntry {\n const output = new Output(ctx);\n\n // This will throw: message.type is required\n output.Send({ props: { content: \"test\" } });\n} catch (e) {\n console.error(\"Output error:\", e.toString());\n}\n```\n\nCommon errors:\n\n- `\"Output constructor requires a context argument\"` - Missing ctx parameter\n- `\"Send requires a message argument\"` - Missing message parameter\n- `\"message.type is required and must be a string\"` - Missing or invalid type field\n- `\"SendGroup requires a group argument\"` - Missing group parameter\n- `\"group.id is required and must be a string\"` - Missing group ID\n- `\"group.messages is required and must be an array\"` - Missing or invalid messages array\n\n## Notes\n\n1. **Context Requirement**: The Output object must be created with a valid agent context\n2. **Writer Required**: The context must have a Writer set (automatically handled in API requests)\n3. **Message Format**: Messages are automatically adapted based on the context's Accept type (standard, cui-web, cui-native, cui-desktop)\n4. **Streaming**: For streaming responses, use delta updates with proper message IDs\n5. **Method Chaining**: All methods return the Output instance for convenient chaining\n\n## See Also\n\n- [Output Package Documentation](../README.md)\n- [Message Types](../BUILTIN_TYPES.md)\n- [Agent Context](../../context/README.md)\n- [Hook System](../../assistant/hook/README.md)\n" }, { "path": "agent/output/jsapi/output.go", "content": "package jsapi\n\n// func init() {\n// \t// Auto-register Output JavaScript API when package is imported\n// \tv8.RegisterFunction(\"Output\", ExportFunction)\n// }\n\n// // Usage from JavaScript:\n// //\n// //\tconst output = new Output(ctx)\n// //\toutput.Send({ type: \"text\", props: { content: \"Hello\" } })\n// //\toutput.Send(\"Hello\") // shorthand for text message\n// //\toutput.SendGroup({ id: \"group1\", messages: [...] })\n// //\n// // Objects:\n// // - Output: Output manager (constructor)\n\n// // ExportFunction exports the Output constructor function template\n// // This is used by v8.RegisterFunction\n// func ExportFunction(iso *v8go.Isolate) *v8go.FunctionTemplate {\n// \treturn v8go.NewFunctionTemplate(iso, outputConstructor)\n// }\n\n// // outputConstructor is the JavaScript constructor for Output\n// // Usage: new Output(ctx)\n// func outputConstructor(info *v8go.FunctionCallbackInfo) *v8go.Value {\n// \tv8ctx := info.Context()\n// \targs := info.Args()\n\n// \t// Require ctx argument\n// \tif len(args) < 1 {\n// \t\treturn bridge.JsException(v8ctx, \"Output constructor requires a context argument\")\n// \t}\n\n// \t// Get the context object from JavaScript\n// \tctxObj, err := args[0].AsObject()\n// \tif err != nil {\n// \t\treturn bridge.JsException(v8ctx, fmt.Sprintf(\"context must be an object: %s\", err))\n// \t}\n\n// \t// Get the goValueID from internal field (index 0)\n// \tif ctxObj.InternalFieldCount() < 1 {\n// \t\treturn bridge.JsException(v8ctx, \"context object is missing internal fields\")\n// \t}\n\n// \tgoValueIDValue := ctxObj.GetInternalField(0)\n// \tif goValueIDValue == nil || !goValueIDValue.IsString() {\n// \t\treturn bridge.JsException(v8ctx, \"context object is missing goValueID\")\n// \t}\n\n// \tgoValueID := goValueIDValue.String()\n\n// \t// Retrieve the Go context object from bridge registry\n// \tgoObj := bridge.GetGoObject(goValueID)\n// \tif goObj == nil {\n// \t\treturn bridge.JsException(v8ctx, \"context object not found in registry\")\n// \t}\n\n// \t// Type assert to *agentContext.Context\n// \tctx, ok := goObj.(*agentContext.Context)\n// \tif !ok {\n// \t\treturn bridge.JsException(v8ctx, fmt.Sprintf(\"object is not a Context, got %T\", goObj))\n// \t}\n\n// \t// Create output object\n// \toutputObj, err := NewOutputObject(v8ctx, ctx)\n// \tif err != nil {\n// \t\treturn bridge.JsException(v8ctx, err.Error())\n// \t}\n\n// \treturn outputObj\n// }\n\n// // NewOutputObject creates a JavaScript Output object\n// func NewOutputObject(v8ctx *v8go.Context, ctx *agentContext.Context) (*v8go.Value, error) {\n// \tjsObject := v8go.NewObjectTemplate(v8ctx.Isolate())\n\n// \t// Set internal field count to 1 to store the __go_id\n// \t// Internal fields are not accessible from JavaScript, providing better security\n// \tjsObject.SetInternalFieldCount(1)\n\n// \t// Register context in global bridge registry for efficient Go object retrieval\n// \t// The goValueID will be stored in internal field (index 0) after instance creation\n// \tgoValueID := bridge.RegisterGoObject(ctx)\n\n// \t// Set methods\n// \tjsObject.Set(\"Send\", outputSendMethod(v8ctx.Isolate(), ctx))\n// \tjsObject.Set(\"SendGroup\", outputSendGroupMethod(v8ctx.Isolate(), ctx))\n\n// \t// Set release function that will be called when JavaScript object is released\n// \tjsObject.Set(\"__release\", outputGoRelease(v8ctx.Isolate()))\n\n// \t// Create instance\n// \tinstance, err := jsObject.NewInstance(v8ctx)\n// \tif err != nil {\n// \t\t// Clean up: release from global registry if instance creation failed\n// \t\tbridge.ReleaseGoObject(goValueID)\n// \t\treturn nil, err\n// \t}\n\n// \t// Store the goValueID in internal field (index 0)\n// \t// This is not accessible from JavaScript, providing better security\n// \tobj, err := instance.Value.AsObject()\n// \tif err != nil {\n// \t\tbridge.ReleaseGoObject(goValueID)\n// \t\treturn nil, err\n// \t}\n\n// \terr = obj.SetInternalField(0, goValueID)\n// \tif err != nil {\n// \t\tbridge.ReleaseGoObject(goValueID)\n// \t\treturn nil, err\n// \t}\n\n// \treturn instance.Value, nil\n// }\n\n// // outputGoRelease releases the Go object from the global bridge registry\n// // It retrieves the goValueID from internal field (index 0) and releases the Go object\n// func outputGoRelease(iso *v8go.Isolate) *v8go.FunctionTemplate {\n// \treturn v8go.NewFunctionTemplate(iso, func(info *v8go.FunctionCallbackInfo) *v8go.Value {\n// \t\t// Get the output object (this)\n// \t\tthisObj, err := info.This().AsObject()\n// \t\tif err == nil && thisObj.InternalFieldCount() > 0 {\n// \t\t\t// Get goValueID from internal field (index 0)\n// \t\t\tgoValueIDValue := thisObj.GetInternalField(0)\n// \t\t\tif goValueIDValue != nil && goValueIDValue.IsString() {\n// \t\t\t\tgoValueID := goValueIDValue.String()\n// \t\t\t\t// Release from global bridge registry\n// \t\t\t\tbridge.ReleaseGoObject(goValueID)\n// \t\t\t}\n// \t\t}\n\n// \t\treturn v8go.Undefined(info.Context().Isolate())\n// \t})\n// }\n\n// // outputSendMethod implements the Send method\n// // Usage: output.Send(message)\n// // message can be an object with { type: string, props: object, ... } or a simple string (will be converted to text message)\n// func outputSendMethod(iso *v8go.Isolate, ctx *agentContext.Context) *v8go.FunctionTemplate {\n// \treturn v8go.NewFunctionTemplate(iso, func(info *v8go.FunctionCallbackInfo) *v8go.Value {\n// \t\tv8ctx := info.Context()\n// \t\targs := info.Args()\n\n// \t\tif len(args) < 1 {\n// \t\t\treturn bridge.JsException(v8ctx, \"Send requires a message argument\")\n// \t\t}\n\n// \t\t// Parse message argument\n// \t\tmsg, err := parseMessage(v8ctx, args[0])\n// \t\tif err != nil {\n// \t\t\treturn bridge.JsException(v8ctx, fmt.Sprintf(\"invalid message: %s\", err))\n// \t\t}\n\n// \t\t// Call output.Send\n// \t\tif err := output.Send(ctx, msg); err != nil {\n// \t\t\treturn bridge.JsException(v8ctx, fmt.Sprintf(\"Send failed: %s\", err))\n// \t\t}\n\n// \t\treturn info.This().Value\n// \t})\n// }\n\n// // outputSendGroupMethod implements the SendGroup method\n// // Usage: output.SendGroup(group)\n// // group must be an object with { id: string, messages: [], ... }\n// func outputSendGroupMethod(iso *v8go.Isolate, ctx *agentContext.Context) *v8go.FunctionTemplate {\n// \treturn v8go.NewFunctionTemplate(iso, func(info *v8go.FunctionCallbackInfo) *v8go.Value {\n// \t\tv8ctx := info.Context()\n// \t\targs := info.Args()\n\n// \t\tif len(args) < 1 {\n// \t\t\treturn bridge.JsException(v8ctx, \"SendGroup requires a group argument\")\n// \t\t}\n\n// \t\t// Parse group argument\n// \t\tgroup, err := parseGroup(v8ctx, args[0])\n// \t\tif err != nil {\n// \t\t\treturn bridge.JsException(v8ctx, fmt.Sprintf(\"invalid group: %s\", err))\n// \t\t}\n\n// \t\t// Call output.SendGroup\n// \t\tif err := output.SendGroup(ctx, group); err != nil {\n// \t\t\treturn bridge.JsException(v8ctx, fmt.Sprintf(\"SendGroup failed: %s\", err))\n// \t\t}\n\n// \t\treturn info.This().Value\n// \t})\n// }\n\n// // parseMessage parses a JavaScript value into a message.Message\n// func parseMessage(v8ctx *v8go.Context, jsValue *v8go.Value) (*message.Message, error) {\n// \t// Handle string shorthand: convert to text message\n// \tif jsValue.IsString() {\n// \t\treturn &message.Message{\n// \t\t\tType: message.TypeText,\n// \t\t\tProps: map[string]interface{}{\n// \t\t\t\t\"content\": jsValue.String(),\n// \t\t\t},\n// \t\t}, nil\n// \t}\n\n// \t// Handle object\n// \tif !jsValue.IsObject() {\n// \t\treturn nil, fmt.Errorf(\"message must be a string or object\")\n// \t}\n\n// \t// Convert to Go map\n// \tgoValue, err := bridge.GoValue(jsValue, v8ctx)\n// \tif err != nil {\n// \t\treturn nil, fmt.Errorf(\"failed to convert message: %w\", err)\n// \t}\n\n// \tmsgMap, ok := goValue.(map[string]interface{})\n// \tif !ok {\n// \t\treturn nil, fmt.Errorf(\"message must be an object\")\n// \t}\n\n// \t// Build message\n// \tmsg := &message.Message{}\n\n// \t// Type field (required)\n// \tif msgType, ok := msgMap[\"type\"].(string); ok {\n// \t\tmsg.Type = msgType\n// \t} else {\n// \t\treturn nil, fmt.Errorf(\"message.type is required and must be a string\")\n// \t}\n\n// \t// Props field (optional)\n// \tif props, ok := msgMap[\"props\"].(map[string]interface{}); ok {\n// \t\tmsg.Props = props\n// \t}\n\n// \t// Optional fields\n// \tif id, ok := msgMap[\"id\"].(string); ok {\n// \t\tmsg.ID = id\n// \t}\n// \tif delta, ok := msgMap[\"delta\"].(bool); ok {\n// \t\tmsg.Delta = delta\n// \t}\n// \tif done, ok := msgMap[\"done\"].(bool); ok {\n// \t\tmsg.Done = done\n// \t}\n// \tif deltaPath, ok := msgMap[\"delta_path\"].(string); ok {\n// \t\tmsg.DeltaPath = deltaPath\n// \t}\n// \tif deltaAction, ok := msgMap[\"delta_action\"].(string); ok {\n// \t\tmsg.DeltaAction = deltaAction\n// \t}\n// \tif typeChange, ok := msgMap[\"type_change\"].(bool); ok {\n// \t\tmsg.TypeChange = typeChange\n// \t}\n// \tif groupID, ok := msgMap[\"group_id\"].(string); ok {\n// \t\tmsg.GroupID = groupID\n// \t}\n// \tif groupStart, ok := msgMap[\"group_start\"].(bool); ok {\n// \t\tmsg.GroupStart = groupStart\n// \t}\n// \tif groupEnd, ok := msgMap[\"group_end\"].(bool); ok {\n// \t\tmsg.GroupEnd = groupEnd\n// \t}\n\n// \t// Metadata (optional)\n// \tif metadataMap, ok := msgMap[\"metadata\"].(map[string]interface{}); ok {\n// \t\tmetadata := &message.Metadata{}\n// \t\tif timestamp, ok := metadataMap[\"timestamp\"].(float64); ok {\n// \t\t\tmetadata.Timestamp = int64(timestamp)\n// \t\t}\n// \t\tif sequence, ok := metadataMap[\"sequence\"].(float64); ok {\n// \t\t\tmetadata.Sequence = int(sequence)\n// \t\t}\n// \t\tif traceID, ok := metadataMap[\"trace_id\"].(string); ok {\n// \t\t\tmetadata.TraceID = traceID\n// \t\t}\n// \t\tmsg.Metadata = metadata\n// \t}\n\n// \treturn msg, nil\n// }\n\n// // parseGroup parses a JavaScript value into a message.Group\n// func parseGroup(v8ctx *v8go.Context, jsValue *v8go.Value) (*message.Group, error) {\n// \t// Must be an object\n// \tif !jsValue.IsObject() {\n// \t\treturn nil, fmt.Errorf(\"group must be an object\")\n// \t}\n\n// \t// Convert to Go map\n// \tgoValue, err := bridge.GoValue(jsValue, v8ctx)\n// \tif err != nil {\n// \t\treturn nil, fmt.Errorf(\"failed to convert group: %w\", err)\n// \t}\n\n// \tgroupMap, ok := goValue.(map[string]interface{})\n// \tif !ok {\n// \t\treturn nil, fmt.Errorf(\"group must be an object\")\n// \t}\n\n// \t// Build group\n// \tgroup := &message.Group{}\n\n// \t// ID field (required)\n// \tif id, ok := groupMap[\"id\"].(string); ok {\n// \t\tgroup.ID = id\n// \t} else {\n// \t\treturn nil, fmt.Errorf(\"group.id is required and must be a string\")\n// \t}\n\n// \t// Messages field (required)\n// \tif messagesArray, ok := groupMap[\"messages\"].([]interface{}); ok {\n// \t\tgroup.Messages = make([]*message.Message, 0, len(messagesArray))\n// \t\tfor i, msgInterface := range messagesArray {\n// \t\t\t// Convert to map\n// \t\t\tmsgMap, ok := msgInterface.(map[string]interface{})\n// \t\t\tif !ok {\n// \t\t\t\treturn nil, fmt.Errorf(\"group.messages[%d] must be an object\", i)\n// \t\t\t}\n\n// \t\t\t// Convert map to Message\n// \t\t\tmsg := &message.Message{}\n\n// \t\t\t// Type field (required)\n// \t\t\tif msgType, ok := msgMap[\"type\"].(string); ok {\n// \t\t\t\tmsg.Type = msgType\n// \t\t\t} else {\n// \t\t\t\treturn nil, fmt.Errorf(\"group.messages[%d].type is required\", i)\n// \t\t\t}\n\n// \t\t\t// Props field (optional)\n// \t\t\tif props, ok := msgMap[\"props\"].(map[string]interface{}); ok {\n// \t\t\t\tmsg.Props = props\n// \t\t\t}\n\n// \t\t\t// Optional fields\n// \t\t\tif id, ok := msgMap[\"id\"].(string); ok {\n// \t\t\t\tmsg.ID = id\n// \t\t\t}\n// \t\t\tif delta, ok := msgMap[\"delta\"].(bool); ok {\n// \t\t\t\tmsg.Delta = delta\n// \t\t\t}\n// \t\t\tif done, ok := msgMap[\"done\"].(bool); ok {\n// \t\t\t\tmsg.Done = done\n// \t\t\t}\n// \t\t\tif deltaPath, ok := msgMap[\"delta_path\"].(string); ok {\n// \t\t\t\tmsg.DeltaPath = deltaPath\n// \t\t\t}\n// \t\t\tif deltaAction, ok := msgMap[\"delta_action\"].(string); ok {\n// \t\t\t\tmsg.DeltaAction = deltaAction\n// \t\t\t}\n// \t\t\tif typeChange, ok := msgMap[\"type_change\"].(bool); ok {\n// \t\t\t\tmsg.TypeChange = typeChange\n// \t\t\t}\n// \t\t\tif groupID, ok := msgMap[\"group_id\"].(string); ok {\n// \t\t\t\tmsg.GroupID = groupID\n// \t\t\t}\n// \t\t\tif groupStart, ok := msgMap[\"group_start\"].(bool); ok {\n// \t\t\t\tmsg.GroupStart = groupStart\n// \t\t\t}\n// \t\t\tif groupEnd, ok := msgMap[\"group_end\"].(bool); ok {\n// \t\t\t\tmsg.GroupEnd = groupEnd\n// \t\t\t}\n\n// \t\t\t// Metadata (optional)\n// \t\t\tif metadataMap, ok := msgMap[\"metadata\"].(map[string]interface{}); ok {\n// \t\t\t\tmetadata := &message.Metadata{}\n// \t\t\t\tif timestamp, ok := metadataMap[\"timestamp\"].(float64); ok {\n// \t\t\t\t\tmetadata.Timestamp = int64(timestamp)\n// \t\t\t\t}\n// \t\t\t\tif sequence, ok := metadataMap[\"sequence\"].(float64); ok {\n// \t\t\t\t\tmetadata.Sequence = int(sequence)\n// \t\t\t\t}\n// \t\t\t\tif traceID, ok := metadataMap[\"trace_id\"].(string); ok {\n// \t\t\t\t\tmetadata.TraceID = traceID\n// \t\t\t\t}\n// \t\t\t\tmsg.Metadata = metadata\n// \t\t\t}\n\n// \t\t\tgroup.Messages = append(group.Messages, msg)\n// \t\t}\n// \t} else {\n// \t\treturn nil, fmt.Errorf(\"group.messages is required and must be an array\")\n// \t}\n\n// \t// Metadata (optional)\n// \tif metadataMap, ok := groupMap[\"metadata\"].(map[string]interface{}); ok {\n// \t\tmetadata := &message.Metadata{}\n// \t\tif timestamp, ok := metadataMap[\"timestamp\"].(float64); ok {\n// \t\t\tmetadata.Timestamp = int64(timestamp)\n// \t\t}\n// \t\tif sequence, ok := metadataMap[\"sequence\"].(float64); ok {\n// \t\t\tmetadata.Sequence = int(sequence)\n// \t\t}\n// \t\tif traceID, ok := metadataMap[\"trace_id\"].(string); ok {\n// \t\t\tmetadata.TraceID = traceID\n// \t\t}\n// \t\tgroup.Metadata = metadata\n// \t}\n\n// \treturn group, nil\n// }\n" }, { "path": "agent/output/jsapi/output_test.go", "content": "package jsapi\n\n// func TestOutputConstructor(t *testing.T) {\n// \ttest.Prepare(t, config.Conf)\n// \tdefer test.Clean()\n\n// \ttests := []struct {\n// \t\tname string\n// \t\tscript string\n// \t\texpectError bool\n// \t}{\n// \t\t{\n// \t\t\tname: \"Create Output with context\",\n// \t\t\tscript: `\n// \t\t\t\tfunction test(ctx) {\n// \t\t\t\t\tconst output = new Output(ctx);\n// \t\t\t\t\treturn output !== undefined && output !== null;\n// \t\t\t\t}\n// \t\t\t`,\n// \t\t\texpectError: false,\n// \t\t},\n// \t\t{\n// \t\t\tname: \"Create Output without context should fail\",\n// \t\t\tscript: `\n// \t\t\t\tfunction test(ctx) {\n// \t\t\t\t\ttry {\n// \t\t\t\t\t\tconst output = new Output();\n// \t\t\t\t\t\treturn false;\n// \t\t\t\t\t} catch (e) {\n// \t\t\t\t\t\treturn e.toString().includes(\"context argument\");\n// \t\t\t\t\t}\n// \t\t\t\t}\n// \t\t\t`,\n// \t\t\texpectError: false,\n// \t\t},\n// \t}\n\n// \tfor _, tt := range tests {\n// \t\tt.Run(tt.name, func(t *testing.T) {\n// \t\t\tctx := agentContext.New(context.Background(), nil, \"test-chat-123\", \"\")\n// \t\t\tctx.AssistantID = \"test-assistant-456\"\n\n// \t\t\t// Execute test script with v8.Call\n// \t\t\tres, err := v8.Call(v8.CallOptions{}, tt.script, &ctx)\n// \t\t\tif tt.expectError {\n// \t\t\t\tassert.Error(t, err)\n// \t\t\t\treturn\n// \t\t\t}\n\n// \t\t\tassert.NoError(t, err)\n// \t\t\tassert.True(t, res.(bool))\n// \t\t})\n// \t}\n// }\n\n// func TestOutputSend(t *testing.T) {\n// \ttest.Prepare(t, config.Conf)\n// \tdefer test.Clean()\n\n// \ttests := []struct {\n// \t\tname string\n// \t\tscript string\n// \t\texpectError bool\n// \t\tvalidate func(*testing.T, *agentContext.Context)\n// \t}{\n// \t\t{\n// \t\t\tname: \"Send text message with object\",\n// \t\t\tscript: `\n// \t\t\t\tfunction test(ctx) {\n// \t\t\t\t\tconst output = new Output(ctx);\n// \t\t\t\t\toutput.Send({\n// \t\t\t\t\t\ttype: \"text\",\n// \t\t\t\t\t\tprops: { content: \"Hello World\" }\n// \t\t\t\t\t});\n// \t\t\t\t\treturn true;\n// \t\t\t\t}\n// \t\t\t`,\n// \t\t\texpectError: false,\n// \t\t},\n// \t\t{\n// \t\t\tname: \"Send text message with string shorthand\",\n// \t\t\tscript: `\n// \t\t\t\tfunction test(ctx) {\n// \t\t\t\t\tconst output = new Output(ctx);\n// \t\t\t\t\toutput.Send(\"Hello World\");\n// \t\t\t\t\treturn true;\n// \t\t\t\t}\n// \t\t\t`,\n// \t\t\texpectError: false,\n// \t\t},\n// \t\t{\n// \t\t\tname: \"Send message with all fields\",\n// \t\t\tscript: `\n// \t\t\t\tfunction test(ctx) {\n// \t\t\t\t\tconst output = new Output(ctx);\n// \t\t\t\t\toutput.Send({\n// \t\t\t\t\t\ttype: \"text\",\n// \t\t\t\t\t\tprops: { content: \"Test\" },\n// \t\t\t\t\t\tid: \"msg-1\",\n// \t\t\t\t\t\tdelta: true,\n// \t\t\t\t\t\tdone: false,\n// \t\t\t\t\t\tdelta_path: \"content\",\n// \t\t\t\t\t\tdelta_action: \"append\",\n// \t\t\t\t\t\tmetadata: {\n// \t\t\t\t\t\t\ttimestamp: 1234567890,\n// \t\t\t\t\t\t\tsequence: 1,\n// \t\t\t\t\t\t\ttrace_id: \"trace-123\"\n// \t\t\t\t\t\t}\n// \t\t\t\t\t});\n// \t\t\t\t\treturn true;\n// \t\t\t\t}\n// \t\t\t`,\n// \t\t\texpectError: false,\n// \t\t},\n// \t\t{\n// \t\t\tname: \"Send error message\",\n// \t\t\tscript: `\n// \t\t\t\tfunction test(ctx) {\n// \t\t\t\t\tconst output = new Output(ctx);\n// \t\t\t\t\toutput.Send({\n// \t\t\t\t\t\ttype: \"error\",\n// \t\t\t\t\t\tprops: {\n// \t\t\t\t\t\t\tmessage: \"Something went wrong\",\n// \t\t\t\t\t\t\tcode: \"ERR_001\"\n// \t\t\t\t\t\t}\n// \t\t\t\t\t});\n// \t\t\t\t\treturn true;\n// \t\t\t\t}\n// \t\t\t`,\n// \t\t\texpectError: false,\n// \t\t},\n// \t\t{\n// \t\t\tname: \"Send without message should fail\",\n// \t\t\tscript: `\n// \t\t\t\tfunction test(ctx) {\n// \t\t\t\t\ttry {\n// \t\t\t\t\t\tconst output = new Output(ctx);\n// \t\t\t\t\t\toutput.Send();\n// \t\t\t\t\t\treturn false;\n// \t\t\t\t\t} catch (e) {\n// \t\t\t\t\t\treturn e.toString().includes(\"message argument\");\n// \t\t\t\t\t}\n// \t\t\t\t}\n// \t\t\t`,\n// \t\t\texpectError: false,\n// \t\t},\n// \t\t{\n// \t\t\tname: \"Send message without type should fail\",\n// \t\t\tscript: `\n// \t\t\t\tfunction test(ctx) {\n// \t\t\t\t\ttry {\n// \t\t\t\t\t\tconst output = new Output(ctx);\n// \t\t\t\t\t\toutput.Send({ props: { content: \"test\" } });\n// \t\t\t\t\t\treturn false;\n// \t\t\t\t\t} catch (e) {\n// \t\t\t\t\t\treturn e.toString().includes(\"type is required\");\n// \t\t\t\t\t}\n// \t\t\t\t}\n// \t\t\t`,\n// \t\t\texpectError: false,\n// \t\t},\n// \t}\n\n// \tfor _, tt := range tests {\n// \t\tt.Run(tt.name, func(t *testing.T) {\n// \t\t\t// Create context with mock writer\n// \t\t\tctx := agentContext.New(context.Background(), nil, \"test-chat\", \"\")\n// \t\t\tctx.Writer = &mockWriter{}\n\n// \t\t\t// Execute test script with v8.Call\n// \t\t\tres, err := v8.Call(v8.CallOptions{}, tt.script, &ctx)\n// \t\t\tif tt.expectError {\n// \t\t\t\tassert.Error(t, err)\n// \t\t\t\treturn\n// \t\t\t}\n\n// \t\t\tassert.NoError(t, err)\n// \t\t\tassert.True(t, res.(bool))\n\n// \t\t\tif tt.validate != nil {\n// \t\t\t\ttt.validate(t, &ctx)\n// \t\t\t}\n// \t\t})\n// \t}\n// }\n\n// func TestOutputSendGroup(t *testing.T) {\n// \ttest.Prepare(t, config.Conf)\n// \tdefer test.Clean()\n\n// \ttests := []struct {\n// \t\tname string\n// \t\tscript string\n// \t\texpectError bool\n// \t}{\n// \t\t{\n// \t\t\tname: \"Send message group\",\n// \t\t\tscript: `\n// \t\t\t\tfunction test(ctx) {\n// \t\t\t\t\tconst output = new Output(ctx);\n// \t\t\t\t\toutput.SendGroup({\n// \t\t\t\t\t\tid: \"group-1\",\n// \t\t\t\t\t\tmessages: [\n// \t\t\t\t\t\t\t{ type: \"text\", props: { content: \"Message 1\" } },\n// \t\t\t\t\t\t\t{ type: \"text\", props: { content: \"Message 2\" } }\n// \t\t\t\t\t\t]\n// \t\t\t\t\t});\n// \t\t\t\t\treturn true;\n// \t\t\t\t}\n// \t\t\t`,\n// \t\t\texpectError: false,\n// \t\t},\n// \t\t{\n// \t\t\tname: \"Send group with metadata\",\n// \t\t\tscript: `\n// \t\t\t\tfunction test(ctx) {\n// \t\t\t\t\tconst output = new Output(ctx);\n// \t\t\t\t\toutput.SendGroup({\n// \t\t\t\t\t\tid: \"group-1\",\n// \t\t\t\t\t\tmessages: [\n// \t\t\t\t\t\t\t{ type: \"text\", props: { content: \"Test\" } }\n// \t\t\t\t\t\t],\n// \t\t\t\t\t\tmetadata: {\n// \t\t\t\t\t\t\ttimestamp: 1234567890,\n// \t\t\t\t\t\t\tsequence: 1\n// \t\t\t\t\t\t}\n// \t\t\t\t\t});\n// \t\t\t\t\treturn true;\n// \t\t\t\t}\n// \t\t\t`,\n// \t\t\texpectError: false,\n// \t\t},\n// \t\t{\n// \t\t\tname: \"Send empty group\",\n// \t\t\tscript: `\n// \t\t\t\tfunction test(ctx) {\n// \t\t\t\t\tconst output = new Output(ctx);\n// \t\t\t\t\toutput.SendGroup({\n// \t\t\t\t\t\tid: \"group-1\",\n// \t\t\t\t\t\tmessages: []\n// \t\t\t\t\t});\n// \t\t\t\t\treturn true;\n// \t\t\t\t}\n// \t\t\t`,\n// \t\t\texpectError: false,\n// \t\t},\n// \t\t{\n// \t\t\tname: \"Send group without id should fail\",\n// \t\t\tscript: `\n// \t\t\t\tfunction test(ctx) {\n// \t\t\t\t\ttry {\n// \t\t\t\t\t\tconst output = new Output(ctx);\n// \t\t\t\t\t\toutput.SendGroup({\n// \t\t\t\t\t\t\tmessages: [\n// \t\t\t\t\t\t\t\t{ type: \"text\", props: { content: \"Test\" } }\n// \t\t\t\t\t\t\t]\n// \t\t\t\t\t\t});\n// \t\t\t\t\t\treturn false;\n// \t\t\t\t\t} catch (e) {\n// \t\t\t\t\t\treturn e.toString().includes(\"id is required\");\n// \t\t\t\t\t}\n// \t\t\t\t}\n// \t\t\t`,\n// \t\t\texpectError: false,\n// \t\t},\n// \t\t{\n// \t\t\tname: \"Send group without messages should fail\",\n// \t\t\tscript: `\n// \t\t\t\tfunction test(ctx) {\n// \t\t\t\t\ttry {\n// \t\t\t\t\t\tconst output = new Output(ctx);\n// \t\t\t\t\t\toutput.SendGroup({ id: \"group-1\" });\n// \t\t\t\t\t\treturn false;\n// \t\t\t\t\t} catch (e) {\n// \t\t\t\t\t\treturn e.toString().includes(\"messages is required\");\n// \t\t\t\t\t}\n// \t\t\t\t}\n// \t\t\t`,\n// \t\t\texpectError: false,\n// \t\t},\n// \t}\n\n// \tfor _, tt := range tests {\n// \t\tt.Run(tt.name, func(t *testing.T) {\n// \t\t\t// Create context with mock writer\n// \t\t\tctx := agentContext.New(context.Background(), nil, \"test-chat\", \"\")\n// \t\t\tctx.Writer = &mockWriter{}\n\n// \t\t\t// Execute test script with v8.Call\n// \t\t\tres, err := v8.Call(v8.CallOptions{}, tt.script, &ctx)\n// \t\t\tif tt.expectError {\n// \t\t\t\tassert.Error(t, err)\n// \t\t\t\treturn\n// \t\t\t}\n\n// \t\t\tassert.NoError(t, err)\n// \t\t\tassert.True(t, res.(bool))\n// \t\t})\n// \t}\n// }\n\n// func TestOutputChaining(t *testing.T) {\n// \ttest.Prepare(t, config.Conf)\n// \tdefer test.Clean()\n\n// \tscript := `\n// \t\tfunction test(ctx) {\n// \t\t\tconst output = new Output(ctx);\n\n// \t\t\t// Send should return the output object for chaining\n// \t\t\tconst result = output.Send(\"Message 1\");\n\n// \t\t\t// Should be able to chain sends\n// \t\t\toutput.Send(\"Message 2\").Send(\"Message 3\");\n\n// \t\t\treturn result !== undefined;\n// \t\t}\n// \t`\n\n// \tctx := agentContext.New(context.Background(), nil, \"test-chat\", \"\")\n// \tctx.Writer = &mockWriter{}\n\n// \t// Execute test script with v8.Call\n// \tres, err := v8.Call(v8.CallOptions{}, script, &ctx)\n// \tassert.NoError(t, err)\n// \tassert.True(t, res.(bool))\n// }\n\n// // mockWriter is a mock implementation of http.ResponseWriter for testing\n// type mockWriter struct {\n// \tdata [][]byte\n// \theader http.Header\n// }\n\n// func (w *mockWriter) Header() http.Header {\n// \tif w.header == nil {\n// \t\tw.header = make(http.Header)\n// \t}\n// \treturn w.header\n// }\n\n// func (w *mockWriter) Write(p []byte) (n int, err error) {\n// \tw.data = append(w.data, p)\n// \treturn len(p), nil\n// }\n\n// func (w *mockWriter) WriteHeader(statusCode int) {}\n\n// func (w *mockWriter) Flush() {}\n" }, { "path": "agent/output/message/STREAMING.md", "content": "# Message Streaming Architecture\n\nThis document explains the hierarchical streaming architecture for Agent/LLM/MCP message delivery.\n\n## Overview\n\nThe streaming system uses a hierarchical structure to handle complex scenarios including:\n\n- Single LLM calls with multiple message types (thinking, tool calls, text)\n- Agent logic with multiple sequential operations (LLM → MCP → LLM)\n- Concurrent/parallel calls to multiple LLMs or MCPs\n- Real-time delta updates for streaming responses\n\n## Hierarchical Structure\n\n```\nAgent Stream (entire conversation)\n └─ ThreadID (concurrent stream, optional: T1, T2, T3...)\n └─ BlockID (output block/section: B1, B2, B3...)\n └─ MessageID (logical message: M1, M2, M3...)\n └─ ChunkID (stream fragment: C1, C2, C3...)\n```\n\n## Field Definitions\n\n### Message Struct Fields\n\n```go\ntype Message struct {\n // Core fields\n Type string `json:\"type\"`\n Props map[string]interface{} `json:\"props,omitempty\"`\n\n // Streaming control\n ChunkID string `json:\"chunk_id,omitempty\"`\n MessageID string `json:\"message_id,omitempty\"`\n BlockID string `json:\"block_id,omitempty\"`\n ThreadID string `json:\"thread_id,omitempty\"`\n\n // Delta control\n Delta bool `json:\"delta,omitempty\"`\n DeltaPath string `json:\"delta_path,omitempty\"`\n DeltaAction string `json:\"delta_action,omitempty\"`\n\n // ...\n}\n```\n\n### Field Responsibilities\n\n| Field | Generated By | Purpose | Example Values | Required |\n| ----------- | -------------------- | ---------------------------------- | ------------------------------------ | ----------------------------------- |\n| `ChunkID` | System (auto) | Deduplication, ordering, debugging | `C1`, `C2`, `C3` | Always |\n| `MessageID` | LLM Provider/Handler | Delta merge target | `M1`, `M2`, `M3` or `thinking_msg_1` | Required for delta scenarios |\n| `BlockID` | Agent Logic | UI block/section rendering | `B1`, `B2`, `B3` or `llm_response_1` | Required when Agent controls blocks |\n| `ThreadID` | Agent Logic | Concurrent stream distinction | `T1`, `T2`, `T3` or `thread_llm1` | Optional (concurrent only) |\n\n### Detailed Field Explanation\n\n#### ChunkID (Stream Fragment Identifier)\n\n- **Purpose**: Uniquely identifies each chunk in the stream\n- **Generated**: Automatically by the system (sequential: M1, M2, M3...)\n- **Used For**:\n - Deduplication (prevent duplicate chunks)\n - Ordering (maintain correct sequence)\n - Debugging (trace message flow)\n- **Scope**: Unique within entire Agent stream\n- **Always Present**: Yes\n\n**Example:**\n\n```json\n{\"chunk_id\": \"C1\", \"type\": \"text\", \"props\": {\"content\": \"Hello\"}}\n{\"chunk_id\": \"C2\", \"type\": \"text\", \"props\": {\"content\": \" World\"}}\n{\"chunk_id\": \"C3\", \"type\": \"thinking\", \"props\": {\"content\": \"...\"}}\n```\n\n#### MessageID (Logical Message Identifier)\n\n- **Purpose**: Groups multiple chunks into one logical message via delta merging\n- **Generated**: By LLM Provider or Stream Handler\n- **Used For**:\n - Delta merge target (frontend merges all chunks with same MessageID)\n - Distinguishing different messages within a group\n- **Scope**: Unique within a Group\n- **Present When**: Delta streaming is used\n\n**Example:**\n\n```json\n// Multiple chunks combine into one \"thinking\" message\n{\"chunk_id\": \"C1\", \"message_id\": \"M1\", \"type\": \"thinking\", \"props\": {\"content\": \"Let me\"}, \"delta\": true}\n{\"chunk_id\": \"C2\", \"message_id\": \"M1\", \"type\": \"thinking\", \"props\": {\"content\": \" think\"}, \"delta\": true}\n{\"chunk_id\": \"C3\", \"message_id\": \"M1\", \"type\": \"thinking\", \"props\": {\"content\": \"...\"}, \"delta\": true}\n\n// Another independent message\n{\"chunk_id\": \"C4\", \"message_id\": \"M2\", \"type\": \"text\", \"props\": {\"content\": \"Hello\"}, \"delta\": true}\n```\n\n#### BlockID (Output Block Identifier)\n\n- **Purpose**: Represents one output block/section (e.g., one LLM call, one MCP call)\n- **Generated**: By Agent logic\n- **Used For**:\n - Frontend UI block/section rendering (visual blocks)\n - Distinguishing different operations (LLM vs MCP vs custom logic)\n - Organizing related messages together\n- **Scope**: Unique within entire Agent stream\n- **Present When**: Agent explicitly controls output blocks\n\n**Key Concept**: Block represents a semantic unit of work from Agent's perspective, NOT from LLM's perspective. Each block is rendered as a distinct UI section in the frontend.\n\n**Example:**\n\n```json\n// BLOCK 1: LLM Response (contains thinking + tool_call + text)\n{\"chunk_id\": \"C1\", \"block_id\": \"B1\", \"message_id\": \"M1\", \"type\": \"thinking\", ...}\n{\"chunk_id\": \"C2\", \"block_id\": \"B1\", \"message_id\": \"M2\", \"type\": \"tool_call\", ...}\n{\"chunk_id\": \"C3\", \"block_id\": \"B1\", \"message_id\": \"M3\", \"type\": \"text\", ...}\n\n// BLOCK 2: MCP Call\n{\"chunk_id\": \"C4\", \"block_id\": \"B2\", \"message_id\": \"M4\", \"type\": \"loading\", ...}\n{\"chunk_id\": \"C5\", \"block_id\": \"B2\", \"message_id\": \"M5\", \"type\": \"text\", ...}\n\n// BLOCK 3: Another LLM Response\n{\"chunk_id\": \"C6\", \"block_id\": \"B3\", \"message_id\": \"M6\", \"type\": \"text\", ...}\n```\n\n#### ThreadID (Concurrent Stream Identifier)\n\n- **Purpose**: Distinguishes concurrent/parallel output streams\n- **Generated**: By Agent logic when spawning concurrent operations\n- **Used For**:\n - Separating outputs from parallel LLM/MCP calls\n - Maintaining independent streaming contexts\n- **Scope**: Unique within entire Agent stream\n- **Present When**: Agent makes concurrent calls (optional)\n\n**Example:**\n\n```json\n// Main thread\n{\"chunk_id\": \"C1\", \"thread_id\": \"T1\", \"block_id\": \"B1\", \"message_id\": \"M1\", \"type\": \"text\", ...}\n\n// Parallel MCP calls\n{\"chunk_id\": \"C2\", \"thread_id\": \"T2\", \"block_id\": \"B2\", \"message_id\": \"M2\", \"type\": \"text\", ...}\n{\"chunk_id\": \"C3\", \"thread_id\": \"T3\", \"block_id\": \"B3\", \"message_id\": \"M3\", \"type\": \"text\", ...}\n```\n\n## Usage Scenarios\n\n### Scenario 1: Simple Text Message\n\n**No streaming, no grouping**\n\n```json\n{\n \"chunk_id\": \"C1\",\n \"type\": \"text\",\n \"props\": { \"content\": \"Hello World\" }\n}\n```\n\n**Fields Used:**\n\n- `chunk_id`: C1 (auto-generated)\n- No `message_id`, `block_id`, or `thread_id` needed\n\n---\n\n### Scenario 2: LLM Streaming Response (Single Message)\n\n**LLM streams one text message**\n\n```json\n{\"chunk_id\": \"C1\", \"message_id\": \"M1\", \"type\": \"text\", \"props\": {\"content\": \"Hello\"}, \"delta\": true}\n{\"chunk_id\": \"C2\", \"message_id\": \"M1\", \"type\": \"text\", \"props\": {\"content\": \" World\"}, \"delta\": true}\n{\"chunk_id\": \"C3\", \"message_id\": \"M1\", \"type\": \"text\", \"props\": {\"content\": \"!\"}, \"delta\": true}\n```\n\n**Fields Used:**\n\n- `chunk_id`: C1, C2, C3 (unique per chunk)\n- `message_id`: M1 (same for all, merge target)\n- `delta`: true\n\n**Frontend Behavior:**\n\n- Merge all chunks with `message_id: \"M1\"` into one message\n- Display: \"Hello World!\"\n\n---\n\n### Scenario 3: Agent-Controlled LLM Call (One Block)\n\n**Agent wraps LLM response in an output block**\n\n```typescript\n// Agent code starts a block for the LLM response\n// System generates block_id: \"B1\"\n// LLM returns thinking + tool_call + text\n// Agent ends the block\n```\n\n```json\n// LLM chunks within block B1\n{\"chunk_id\": \"C1\", \"message_id\": \"M1\", \"block_id\": \"B1\", \"type\": \"thinking\", \"props\": {...}, \"delta\": true}\n{\"chunk_id\": \"C2\", \"message_id\": \"M1\", \"block_id\": \"B1\", \"type\": \"thinking\", \"props\": {...}, \"delta\": true}\n\n{\"chunk_id\": \"C3\", \"message_id\": \"M2\", \"block_id\": \"B1\", \"type\": \"tool_call\", \"props\": {...}}\n\n{\"chunk_id\": \"C4\", \"message_id\": \"M3\", \"block_id\": \"B1\", \"type\": \"text\", \"props\": {...}, \"delta\": true}\n{\"chunk_id\": \"C5\", \"message_id\": \"M3\", \"block_id\": \"B1\", \"type\": \"text\", \"props\": {...}, \"delta\": true}\n```\n\n**Fields Used:**\n\n- `chunk_id`: C1~C5 (unique per chunk)\n- `message_id`: M1, M2, M3 (per logical message)\n- `block_id`: B1 (all belong to same LLM call)\n- `delta`: true (for streaming messages)\n\n**Frontend Behavior:**\n\n- Render one block/section for `block_id: \"B1\"`\n- Within this block, show 3 messages:\n - Thinking message (chunks C1+C2 merged into M1)\n - Tool call message (chunk C3 = M2)\n - Text message (chunks C4+C5 merged into M3)\n\n---\n\n### Scenario 4: Agent Sequential Operations (Multiple Blocks)\n\n**Agent orchestrates: LLM → MCP → LLM**\n\n```typescript\n// Agent code orchestrates three sequential operations:\n// 1. Block B1: First LLM call\n// 2. Block B2: MCP call\n// 3. Block B3: Second LLM call\n```\n\n```json\n// BLOCK 1: First LLM call\n{\"chunk_id\": \"C1\", \"message_id\": \"M1\", \"block_id\": \"B1\", \"type\": \"text\", ...}\n{\"chunk_id\": \"C2\", \"message_id\": \"M1\", \"block_id\": \"B1\", \"type\": \"text\", ...}\n\n// BLOCK 2: MCP call\n{\"chunk_id\": \"C3\", \"message_id\": \"M2\", \"block_id\": \"B2\", \"type\": \"loading\", ...}\n{\"chunk_id\": \"C4\", \"message_id\": \"M3\", \"block_id\": \"B2\", \"type\": \"text\", ...}\n\n// BLOCK 3: Second LLM call\n{\"chunk_id\": \"C5\", \"message_id\": \"M4\", \"block_id\": \"B3\", \"type\": \"text\", ...}\n{\"chunk_id\": \"C6\", \"message_id\": \"M4\", \"block_id\": \"B3\", \"type\": \"text\", ...}\n```\n\n**Frontend Behavior:**\n\n- Render 3 distinct blocks/sections:\n 1. Block 1 (B1): LLM response with text\n 2. Block 2 (B2): MCP call with loading + result\n 3. Block 3 (B3): LLM response with text\n\n---\n\n### Scenario 5: Concurrent Operations (Blocks + Threads)\n\n**Agent uses concurrent handler to make parallel calls**\n\n```typescript\n// Agent orchestrates parallel operations within one block (B1)\n// The concurrent handler automatically assigns thread_id to each operation:\n// - MCP call for weather (thread_id: \"T1\")\n// - MCP call for news (thread_id: \"T2\")\n// - LLM call for summary (thread_id: \"T3\")\n//\n// Messages from different threads may arrive in any order\n```\n\n```json\n// Same block, different threads (may arrive in any order)\n{\"chunk_id\": \"C1\", \"message_id\": \"M1\", \"block_id\": \"B1\", \"thread_id\": \"T1\", \"type\": \"text\", \"props\": {\"content\": \"Weather: Sunny\"}}\n{\"chunk_id\": \"C2\", \"message_id\": \"M2\", \"block_id\": \"B1\", \"thread_id\": \"T2\", \"type\": \"text\", \"props\": {\"content\": \"News: ...\"}}\n{\"chunk_id\": \"C3\", \"message_id\": \"M1\", \"block_id\": \"B1\", \"thread_id\": \"T1\", \"type\": \"text\", \"props\": {\"content\": \", 25°C\"}}\n{\"chunk_id\": \"C4\", \"message_id\": \"M3\", \"block_id\": \"B1\", \"thread_id\": \"T3\", \"type\": \"text\", \"props\": {\"content\": \"Summary...\"}}\n```\n\n**Fields Used:**\n\n- `chunk_id`: C1, C2, C3, C4 (unique per chunk, chronological order)\n- `message_id`: M1, M2, M3 (per operation/message)\n- `block_id`: B1 (all belong to same parallel operation block)\n- `thread_id`: T1, T2, T3 (distinguish concurrent operations)\n\n**Frontend Behavior:**\n\n- Render one block for `block_id: \"B1\"`\n- Within this block, separate messages by `thread_id`:\n - Thread T1 (Weather): M1 (chunks C1+C3 merged) → \"Weather: Sunny, 25°C\"\n - Thread T2 (News): M2 (chunk C2)\n - Thread T3 (Summary): M3 (chunk C4)\n- Or interleave by `chunk_id` order (C1, C2, C3, C4) to show real-time arrival\n\n---\n\n## Summary\n\n| Field | Level | Purpose | Example |\n| ----------- | ----------- | ------------------ | ---------- |\n| `ChunkID` | System | Transport/debug | C1, C2, C3 |\n| `MessageID` | LLM/Handler | Delta merging | M1, M2, M3 |\n| `BlockID` | Agent | UI blocks/sections | B1, B2, B3 |\n| `ThreadID` | Agent | Concurrency | T1, T2, T3 |\n\n**Key Insight**: Each field serves a distinct purpose at a specific layer of the architecture. This hierarchical design supports simple single-message scenarios while enabling complex Agent orchestration with concurrent operations. Blocks provide natural UI boundaries for organizing related messages.\n" }, { "path": "agent/output/message/interfaces.go", "content": "package message\n\n// Writer is the interface for writing output messages\n// Different writers handle different output formats (SSE, WebSocket, Standard, etc.)\ntype Writer interface {\n\t// Write writes a single message\n\tWrite(msg *Message) error\n\n\t// WriteGroup writes a group of messages\n\tWriteGroup(group *Group) error\n\n\t// Flush flushes any buffered data\n\tFlush() error\n\n\t// Close closes the writer and releases resources\n\tClose() error\n}\n\n// Adapter is the interface for adapting messages to different formats\n// Adapters transform messages from the universal DSL to specific client formats\ntype Adapter interface {\n\t// Adapt transforms a message to the target format\n\t// Returns a slice of output chunks (some messages may be split into multiple chunks)\n\tAdapt(msg *Message) ([]interface{}, error)\n\n\t// SupportsType checks if this adapter supports a specific message type\n\tSupportsType(msgType string) bool\n}\n\n// StreamHandler handles streaming message processing\n// It bridges between LLM streaming chunks and output messages\ntype StreamHandler interface {\n\t// Handle processes a streaming chunk from LLM\n\tHandle(chunkType StreamChunkType, data []byte) error\n\n\t// Flush flushes any pending messages\n\tFlush() error\n\n\t// Close closes the handler\n\tClose() error\n}\n" }, { "path": "agent/output/message/types.go", "content": "package message\n\nimport (\n\t\"net/http\"\n\n\t\"github.com/yaoapp/gou/llm\"\n\ttraceTypes \"github.com/yaoapp/yao/trace/types\"\n)\n\n// Options are the options for the writer\ntype Options struct {\n\tBaseURL string\n\tAccept string\n\tWriter http.ResponseWriter\n\tTrace traceTypes.Manager\n\tCapabilities *llm.Capabilities\n\tLocale string\n}\n\n// Message represents a universal message structure (DSL)\n// All messages are expressed through Type + Props, without predefining specific types\ntype Message struct {\n\t// Core fields\n\tType string `json:\"type\"` // Message type (frontend decides how to render)\n\tProps map[string]interface{} `json:\"props,omitempty\"` // Message properties (passed to frontend component)\n\n\t// Streaming control - Hierarchical structure for Agent/LLM/MCP streaming\n\t// See STREAMING.md for detailed explanation of the streaming architecture\n\tChunkID string `json:\"chunk_id,omitempty\"` // Unique chunk ID (auto-generated: C1, C2, C3...; for dedup/ordering/debugging)\n\tMessageID string `json:\"message_id,omitempty\"` // Logical message ID (delta merge target; multiple chunks combine into one message)\n\tBlockID string `json:\"block_id,omitempty\"` // Output block ID (Agent-level control: one LLM call, one MCP call, etc.; for UI rendering blocks/sections)\n\tThreadID string `json:\"thread_id,omitempty\"` // Thread ID (optional; for concurrent Agent/LLM/MCP calls to distinguish output streams)\n\n\t// Delta control\n\tDelta bool `json:\"delta,omitempty\"` // Whether this is an incremental update\n\tDeltaPath string `json:\"delta_path,omitempty\"` // Update path (e.g., \"content\", \"data\", \"items.0.name\")\n\tDeltaAction string `json:\"delta_action,omitempty\"` // Update action (append, replace, merge, set)\n\n\t// Type correction (for streaming scenarios)\n\tTypeChange bool `json:\"type_change,omitempty\"` // Marks this as a type correction message\n\n\t// Metadata\n\tMetadata *Metadata `json:\"metadata,omitempty\"` // Additional metadata\n}\n\n// Metadata represents message metadata\ntype Metadata struct {\n\tTimestamp int64 `json:\"timestamp,omitempty\"` // Timestamp in nanoseconds\n\tSequence int `json:\"sequence,omitempty\"` // Sequence number (for ordering)\n\tTraceID string `json:\"trace_id,omitempty\"` // Trace ID (for debugging)\n}\n\n// Group represents a semantically complete group of messages\ntype Group struct {\n\tID string `json:\"id\"` // Message group ID\n\tMessages []*Message `json:\"messages\"` // List of messages\n\tMetadata *Metadata `json:\"metadata,omitempty\"` // Metadata\n}\n\n// Built-in message types that all adapters must support\n// These types have standardized Props structures\nconst (\n\t// User interaction types\n\tTypeUserInput = \"user_input\" // User input message (frontend display only)\n\n\t// Content types\n\tTypeText = \"text\" // Plain text or Markdown content\n\tTypeThinking = \"thinking\" // Reasoning/thinking process (e.g., o1 models)\n\tTypeLoading = \"loading\" // Loading/processing indicator (preprocessing, knowledge base search, etc.)\n\tTypeToolCall = \"tool_call\" // LLM tool/function call\n\tTypeError = \"error\" // Error message\n\n\t// Media types (with OpenAI support)\n\tTypeImage = \"image\" // Image content\n\tTypeAudio = \"audio\" // Audio content\n\tTypeVideo = \"video\" // Video content\n\n\t// System types (not visible in standard chat clients)\n\tTypeAction = \"action\" // System action (open panel, navigate, etc.) - silent in OpenAI clients\n\tTypeEvent = \"event\" // Lifecycle event (stream_start, stream_end, etc.) - CUI only, silent in OpenAI clients\n)\n\n// Event types for TypeEvent messages\n// Hierarchical structure: Stream > Thread > Block > Message > Chunk\nconst (\n\t// Stream level events (Agent layer - overall conversation stream)\n\tEventStreamStart = \"stream_start\" // Stream started event\n\tEventStreamEnd = \"stream_end\" // Stream ended event\n\n\t// Thread level events (optional - for concurrent scenarios)\n\tEventThreadStart = \"thread_start\" // Thread started event\n\tEventThreadEnd = \"thread_end\" // Thread ended event\n\n\t// Block level events (Agent layer - logical output sections)\n\tEventBlockStart = \"block_start\" // Block started event\n\tEventBlockEnd = \"block_end\" // Block ended event\n\n\t// Message level events (LLM layer - individual logical messages)\n\tEventMessageStart = \"message_start\" // Message started event\n\tEventMessageEnd = \"message_end\" // Message ended event\n)\n\n// Standard Props structures for built-in types\n\n// UserInputProps defines the standard structure for user input messages\n// Type: \"user_input\"\n// Props: {\"content\": string | ContentPart[], \"role\": string, \"name\": string}\ntype UserInputProps struct {\n\tContent interface{} `json:\"content\"` // User input (text string or multimodal ContentPart[])\n\tRole string `json:\"role,omitempty\"` // User role: \"user\", \"system\", \"developer\" (default: \"user\")\n\tName string `json:\"name,omitempty\"` // Optional participant name\n}\n\n// TextProps defines the standard structure for text messages\n// Type: \"text\"\n// Props: {\"content\": string}\ntype TextProps struct {\n\tContent string `json:\"content\"` // Text content (supports Markdown)\n}\n\n// ThinkingProps defines the standard structure for thinking messages\n// Type: \"thinking\"\n// Props: {\"content\": string}\ntype ThinkingProps struct {\n\tContent string `json:\"content\"` // Reasoning/thinking content\n}\n\n// LoadingProps defines the standard structure for loading messages\n// Type: \"loading\"\n// Props: {\"message\": string}\ntype LoadingProps struct {\n\tMessage string `json:\"message\"` // Loading message (e.g., \"Searching knowledge base...\")\n}\n\n// ToolCallProps defines the standard structure for tool_call messages\n// Type: \"tool_call\"\n// Props: {\"id\": string, \"name\": string, \"arguments\": string}\ntype ToolCallProps struct {\n\tID string `json:\"id\"` // Tool call ID\n\tName string `json:\"name\"` // Function/tool name\n\tArguments string `json:\"arguments,omitempty\"` // JSON string of arguments\n}\n\n// ErrorProps defines the standard structure for error messages\n// Type: \"error\"\n// Props: {\"message\": string, \"code\": string}\ntype ErrorProps struct {\n\tMessage string `json:\"message\"` // Error message\n\tCode string `json:\"code,omitempty\"` // Error code\n\tDetails string `json:\"details,omitempty\"` // Additional error details\n}\n\n// ActionProps defines the standard structure for action messages\n// Type: \"action\"\n// Props: {\"name\": string, \"payload\": map}\ntype ActionProps struct {\n\tName string `json:\"name\"` // Action name (e.g., \"open_panel\", \"navigate\")\n\tPayload map[string]interface{} `json:\"payload,omitempty\"` // Action payload/parameters\n}\n\n// EventProps defines the standard structure for event messages\n// Type: \"event\"\n// Props: {\"event\": string, \"message\": string, \"data\": map}\ntype EventProps struct {\n\tEvent string `json:\"event\"` // Event type (e.g., \"stream_start\", \"stream_end\", \"connecting\")\n\tMessage string `json:\"message,omitempty\"` // Human-readable message (e.g., \"Connecting...\")\n\tData map[string]interface{} `json:\"data,omitempty\"` // Additional event data\n}\n\n// ImageProps defines the standard structure for image messages\n// Type: \"image\"\n// Props: {\"url\": string, \"alt\": string, \"width\": int, \"height\": int, \"detail\": string}\ntype ImageProps struct {\n\tURL string `json:\"url\"` // Required: Image URL or base64 encoded data\n\tAlt string `json:\"alt,omitempty\"` // Alternative text\n\tWidth int `json:\"width,omitempty\"` // Image width in pixels\n\tHeight int `json:\"height,omitempty\"` // Image height in pixels\n\tDetail string `json:\"detail,omitempty\"` // OpenAI detail level: \"auto\", \"low\", \"high\"\n}\n\n// AudioProps defines the standard structure for audio messages\n// Type: \"audio\"\n// Props: {\"url\": string, \"format\": string, \"duration\": float64, \"transcript\": string, \"autoplay\": bool}\ntype AudioProps struct {\n\tURL string `json:\"url\"` // Required: Audio URL or base64 encoded data\n\tFormat string `json:\"format,omitempty\"` // Audio format: \"mp3\", \"wav\", \"ogg\", etc.\n\tDuration float64 `json:\"duration,omitempty\"` // Duration in seconds\n\tTranscript string `json:\"transcript,omitempty\"` // Audio transcript text\n\tAutoplay bool `json:\"autoplay,omitempty\"` // Whether to autoplay\n\tControls bool `json:\"controls,omitempty\"` // Whether to show controls (default: true)\n}\n\n// VideoProps defines the standard structure for video messages\n// Type: \"video\"\n// Props: {\"url\": string, \"format\": string, \"duration\": float64, \"thumbnail\": string, \"width\": int, \"height\": int, \"autoplay\": bool}\ntype VideoProps struct {\n\tURL string `json:\"url\"` // Required: Video URL\n\tFormat string `json:\"format,omitempty\"` // Video format: \"mp4\", \"webm\", etc.\n\tDuration float64 `json:\"duration,omitempty\"` // Duration in seconds\n\tThumbnail string `json:\"thumbnail,omitempty\"` // Thumbnail/poster image URL\n\tWidth int `json:\"width,omitempty\"` // Video width in pixels\n\tHeight int `json:\"height,omitempty\"` // Video height in pixels\n\tAutoplay bool `json:\"autoplay,omitempty\"` // Whether to autoplay\n\tControls bool `json:\"controls,omitempty\"` // Whether to show controls (default: true)\n\tLoop bool `json:\"loop,omitempty\"` // Whether to loop\n}\n\n// Delta action constants for incremental updates\nconst (\n\tDeltaAppend = \"append\" // Append (for arrays, strings)\n\tDeltaReplace = \"replace\" // Replace (for any value)\n\tDeltaMerge = \"merge\" // Merge (for objects)\n\tDeltaSet = \"set\" // Set (for new fields)\n)\n\n// StreamChunkType represents the type of content in a streaming chunk\ntype StreamChunkType string\n\n// Stream chunk type constants - indicates what type of content is in the current chunk\nconst (\n\t// Content chunk types - actual data from the LLM\n\tChunkText StreamChunkType = \"text\" // Regular text content\n\tChunkThinking StreamChunkType = \"thinking\" // Reasoning/thinking content (o1, DeepSeek R1)\n\tChunkToolCall StreamChunkType = \"tool_call\" // Tool/function call\n\tChunkRefusal StreamChunkType = \"refusal\" // Model refusal\n\tChunkMetadata StreamChunkType = \"metadata\" // Metadata (usage, finish_reason, etc.)\n\tChunkError StreamChunkType = \"error\" // Error chunk\n\tChunkUnknown StreamChunkType = \"unknown\" // Unknown/unrecognized chunk type\n\n\t// Lifecycle event types - stream and message boundaries\n\tChunkStreamStart StreamChunkType = \"stream_start\" // Stream begins (entire request starts)\n\tChunkStreamEnd StreamChunkType = \"stream_end\" // Stream ends (entire request completes)\n\tChunkMessageStart StreamChunkType = \"message_start\" // Message begins (text/tool_call/thinking message starts)\n\tChunkMessageEnd StreamChunkType = \"message_end\" // Message ends (text/tool_call/thinking message completes)\n)\n\n// StreamFunc the streaming function callback\n// Parameters:\n// - chunkType: the type of content in this chunk (text, thinking, tool_call, etc.)\n// - data: the actual chunk data (could be text, JSON, or other format)\n//\n// Returns:\n// - int: status code (0 = continue, non-zero = stop streaming)\ntype StreamFunc func(chunkType StreamChunkType, data []byte) int\n\n// AssistantInfo represents the assistant information structure\ntype AssistantInfo struct {\n\tID string `json:\"assistant_id\"` // Assistant ID\n\tType string `json:\"type,omitempty\"` // Assistant Type, default is assistant\n\tName string `json:\"name,omitempty\"` // Assistant Name\n\tAvatar string `json:\"avatar,omitempty\"` // Assistant Avatar\n\tDescription string `json:\"description,omitempty\"` // Assistant Description\n}\n\n// UsageInfo represents token usage statistics\n// Structure matches OpenAI API: https://platform.openai.com/docs/api-reference/chat/object#chat-object-usage\ntype UsageInfo struct {\n\tPromptTokens int `json:\"prompt_tokens\"` // Number of tokens in the prompt\n\tCompletionTokens int `json:\"completion_tokens\"` // Number of tokens in the generated completion\n\tTotalTokens int `json:\"total_tokens\"` // Total number of tokens used (prompt + completion)\n\n\t// Detailed token breakdown\n\tPromptTokensDetails *PromptTokensDetails `json:\"prompt_tokens_details,omitempty\"` // Breakdown of tokens used in the prompt\n\tCompletionTokensDetails *CompletionTokensDetails `json:\"completion_tokens_details,omitempty\"` // Breakdown of tokens used in the completion\n}\n\n// PromptTokensDetails provides detailed breakdown of tokens used in the prompt\ntype PromptTokensDetails struct {\n\tAudioTokens int `json:\"audio_tokens,omitempty\"` // Audio input tokens present in the prompt\n\tCachedTokens int `json:\"cached_tokens,omitempty\"` // Cached tokens present in the prompt\n}\n\n// CompletionTokensDetails provides detailed breakdown of tokens used in the completion\ntype CompletionTokensDetails struct {\n\tAcceptedPredictionTokens int `json:\"accepted_prediction_tokens,omitempty\"` // Tokens from predictions that appeared in the completion\n\tAudioTokens int `json:\"audio_tokens,omitempty\"` // Audio input tokens generated by the model\n\tReasoningTokens int `json:\"reasoning_tokens,omitempty\"` // Tokens generated by the model for reasoning (o1, o1-mini, DeepSeek R1)\n\tRejectedPredictionTokens int `json:\"rejected_prediction_tokens,omitempty\"` // Tokens from predictions that did not appear in the completion\n}\n\n// ============================================================================\n// Stream Lifecycle Event Data Structures\n// ============================================================================\n// These structures define the data format for stream lifecycle events.\n// They provide a standardized way to communicate stream boundaries and metadata\n// to the frontend, enabling better UI/UX (progress indicators, timing, etc.).\n\n// EventStreamStartData represents the data for stream_start event\n// Sent when a streaming request begins\ntype EventStreamStartData struct {\n\tContextID string `json:\"context_id\"` // Context ID for the response\n\tRequestID string `json:\"request_id\"` // Unique identifier for this request\n\tTimestamp int64 `json:\"timestamp\"` // Unix timestamp when stream started\n\tChatID string `json:\"chat_id\"` // Chat ID being used (e.g., \"chat-123\")\n\tTraceID string `json:\"trace_id\"` // Trace ID being used (e.g., \"trace-123\")\n\tAssistant *AssistantInfo `json:\"assistant,omitempty\"` // Assistant information\n\tMetadata map[string]interface{} `json:\"metadata,omitempty\"` // Metadata to pass to the page for CUI context\n}\n\n// EventStreamEndData represents the data for stream_end event\n// Sent when a streaming request completes (successfully or with error)\ntype EventStreamEndData struct {\n\tRequestID string `json:\"request_id\"` // Corresponding request ID\n\tContextID string `json:\"context_id\"` // Context ID for the response\n\tTraceID string `json:\"trace_id\"` // Trace ID being used (e.g., \"trace-123\")\n\tTimestamp int64 `json:\"timestamp\"` // Unix timestamp when stream ended\n\tDurationMs int64 `json:\"duration_ms\"` // Total duration in milliseconds\n\tStatus string `json:\"status\"` // \"completed\" | \"error\" | \"cancelled\"\n\tError string `json:\"error,omitempty\"` // Error message if status is \"error\"\n\tUsage *UsageInfo `json:\"usage,omitempty\"` // Token usage statistics\n\tMetadata map[string]interface{} `json:\"metadata,omitempty\"` // Metadata to pass to the page for CUI context\n}\n\n// EventMessageStartData represents the data for message_start event\n// Sent when a logical message begins (text, tool_call, thinking, etc.)\n// LLM layer: Marks the beginning of a single logical message output\ntype EventMessageStartData struct {\n\tMessageID string `json:\"message_id\"` // Message ID (M1, M2, M3...)\n\tType string `json:\"type\"` // Message type: \"text\" | \"thinking\" | \"tool_call\" | \"refusal\"\n\tTimestamp int64 `json:\"timestamp\"` // Unix timestamp when message started\n\tThreadID string `json:\"thread_id,omitempty\"` // Thread ID (optional; for concurrent streams)\n\tToolCall *EventToolCallInfo `json:\"tool_call,omitempty\"` // Tool call metadata (if type is \"tool_call\")\n\tExtra map[string]interface{} `json:\"extra,omitempty\"` // Additional metadata (for custom providers or future extensions)\n}\n\n// EventMessageEndData represents the data for message_end event\n// Sent when a logical message completes\n// LLM layer: Signals that all chunks for this message have been sent, client should merge and process\ntype EventMessageEndData struct {\n\tMessageID string `json:\"message_id\"` // Message ID (M1, M2, M3...)\n\tType string `json:\"type\"` // Message type (same as in message_start)\n\tTimestamp int64 `json:\"timestamp\"` // Unix timestamp when message ended\n\tThreadID string `json:\"thread_id,omitempty\"` // Thread ID (optional; for concurrent streams)\n\tDurationMs int64 `json:\"duration_ms\"` // Duration of this message in milliseconds\n\tChunkCount int `json:\"chunk_count\"` // Number of data chunks in this message\n\tStatus string `json:\"status\"` // \"completed\" | \"partial\" | \"error\"\n\tToolCall *EventToolCallInfo `json:\"tool_call,omitempty\"` // Complete tool call info (if type is \"tool_call\")\n\tExtra map[string]interface{} `json:\"extra,omitempty\"` // Additional metadata (e.g., complete content for direct use)\n}\n\n// EventToolCallInfo contains tool call information for message events\n// Used in both message_start (partial info) and message_end (complete info)\ntype EventToolCallInfo struct {\n\tID string `json:\"id\"` // Tool call ID (e.g., \"call_abc123\")\n\tName string `json:\"name\"` // Function name (may be partial in message_start)\n\tArguments string `json:\"arguments,omitempty\"` // Complete arguments (only in message_end)\n\tIndex int `json:\"index\"` // Index in the tool calls array\n}\n\n// EventBlockStartData represents the data for block_start event\n// Sent when an output block begins (one LLM call, one MCP call, one Agent sub-task, etc.)\n// Agent layer: Groups multiple related messages into a logical section\ntype EventBlockStartData struct {\n\tBlockID string `json:\"block_id\"` // Block ID (B1, B2, B3...)\n\tType string `json:\"type\"` // Block type: \"llm\" | \"mcp\" | \"agent\" | \"tool\" | \"mixed\"\n\tTimestamp int64 `json:\"timestamp\"` // Unix timestamp when block started\n\tLabel string `json:\"label,omitempty\"` // Human-readable label (e.g., \"Searching knowledge base\", \"Calling weather API\")\n\tExtra map[string]interface{} `json:\"extra,omitempty\"` // Additional metadata\n}\n\n// EventBlockEndData represents the data for block_end event\n// Sent when an output block completes\n// Agent layer: Signals that this logical section is complete\ntype EventBlockEndData struct {\n\tBlockID string `json:\"block_id\"` // Block ID (B1, B2, B3...)\n\tType string `json:\"type\"` // Block type (same as in block_start)\n\tTimestamp int64 `json:\"timestamp\"` // Unix timestamp when block ended\n\tDurationMs int64 `json:\"duration_ms\"` // Duration of this block in milliseconds\n\tMessageCount int `json:\"message_count\"` // Number of messages in this block\n\tStatus string `json:\"status\"` // \"completed\" | \"partial\" | \"error\"\n\tExtra map[string]interface{} `json:\"extra,omitempty\"` // Additional metadata\n}\n\n// EventThreadStartData represents the data for thread_start event\n// Sent when a concurrent thread begins (parallel Agent/LLM/MCP calls)\n// Used in concurrent scenarios to distinguish multiple parallel output streams\ntype EventThreadStartData struct {\n\tThreadID string `json:\"thread_id\"` // Thread ID (T1, T2, T3...)\n\tType string `json:\"type\"` // Thread type: \"agent\" | \"llm\" | \"mcp\" | \"tool\"\n\tTimestamp int64 `json:\"timestamp\"` // Unix timestamp when thread started\n\tLabel string `json:\"label,omitempty\"` // Human-readable label (e.g., \"Parallel search 1\", \"Background task\")\n\tExtra map[string]interface{} `json:\"extra,omitempty\"` // Additional metadata\n}\n\n// EventThreadEndData represents the data for thread_end event\n// Sent when a concurrent thread completes\ntype EventThreadEndData struct {\n\tThreadID string `json:\"thread_id\"` // Thread ID (T1, T2, T3...)\n\tType string `json:\"type\"` // Thread type (same as in thread_start)\n\tTimestamp int64 `json:\"timestamp\"` // Unix timestamp when thread ended\n\tDurationMs int64 `json:\"duration_ms\"` // Duration of this thread in milliseconds\n\tBlockCount int `json:\"block_count\"` // Number of blocks in this thread\n\tStatus string `json:\"status\"` // \"completed\" | \"partial\" | \"error\"\n\tExtra map[string]interface{} `json:\"extra,omitempty\"` // Additional metadata\n}\n" }, { "path": "agent/output/message/utils.go", "content": "package message\n\nimport (\n\t\"fmt\"\n\t\"sync/atomic\"\n\n\tgonanoid \"github.com/matoous/go-nanoid/v2\"\n)\n\n// IDGenerator generates unique IDs within a context (e.g., one conversation stream)\n// Each Context should have its own IDGenerator to ensure IDs are unique within that context\ntype IDGenerator struct {\n\tchunkCounter uint64\n\tmessageCounter uint64\n\tblockCounter uint64\n\tthreadCounter uint64\n}\n\n// NewIDGenerator creates a new ID generator for a context\nfunc NewIDGenerator() *IDGenerator {\n\treturn &IDGenerator{}\n}\n\n// GenerateChunkID generates a unique chunk ID with prefix C\n// Format: C1, C2, C3...\nfunc (g *IDGenerator) GenerateChunkID() string {\n\tid := atomic.AddUint64(&g.chunkCounter, 1)\n\treturn fmt.Sprintf(\"C%d\", id)\n}\n\n// GenerateMessageID generates a unique message ID with prefix M\n// Format: M1, M2, M3...\nfunc (g *IDGenerator) GenerateMessageID() string {\n\tid := atomic.AddUint64(&g.messageCounter, 1)\n\treturn fmt.Sprintf(\"M%d\", id)\n}\n\n// GenerateBlockID generates a unique block ID with prefix B\n// Format: B1, B2, B3...\nfunc (g *IDGenerator) GenerateBlockID() string {\n\tid := atomic.AddUint64(&g.blockCounter, 1)\n\treturn fmt.Sprintf(\"B%d\", id)\n}\n\n// GenerateThreadID generates a unique thread ID with prefix T\n// Format: T1, T2, T3...\nfunc (g *IDGenerator) GenerateThreadID() string {\n\tid := atomic.AddUint64(&g.threadCounter, 1)\n\treturn fmt.Sprintf(\"T%d\", id)\n}\n\n// Reset resets all counters (useful for testing)\nfunc (g *IDGenerator) Reset() {\n\tatomic.StoreUint64(&g.chunkCounter, 0)\n\tatomic.StoreUint64(&g.messageCounter, 0)\n\tatomic.StoreUint64(&g.blockCounter, 0)\n\tatomic.StoreUint64(&g.threadCounter, 0)\n}\n\n// GetCounters returns current counter values (for debugging/testing)\nfunc (g *IDGenerator) GetCounters() (chunk, message, block, thread uint64) {\n\treturn atomic.LoadUint64(&g.chunkCounter),\n\t\tatomic.LoadUint64(&g.messageCounter),\n\t\tatomic.LoadUint64(&g.blockCounter),\n\t\tatomic.LoadUint64(&g.threadCounter)\n}\n\n// GenerateNanoID generates a unique ID using nanoid\n// Returns a 21-character URL-safe string\n// This is a static function that doesn't depend on the generator's counter\nfunc GenerateNanoID() string {\n\tid, err := gonanoid.New()\n\tif err != nil {\n\t\t// Fallback to timestamp-based ID if nanoid fails\n\t\treturn fmt.Sprintf(\"id_%d\", atomic.AddUint64(new(uint64), 1))\n\t}\n\treturn id\n}\n\n// GenerateCustomID generates a custom ID with prefix and nanoid\n// Format: prefix_nanoid (e.g., \"msg_V1StGXR8_Z5jdHi6B-myT\")\n// This is a static function that doesn't depend on the generator's counter\nfunc GenerateCustomID(prefix string) string {\n\tid, err := gonanoid.New()\n\tif err != nil {\n\t\t// Fallback to timestamp-based ID\n\t\treturn fmt.Sprintf(\"%s_%d\", prefix, atomic.AddUint64(new(uint64), 1))\n\t}\n\treturn fmt.Sprintf(\"%s_%s\", prefix, id)\n}\n" }, { "path": "agent/output/message/utils_test.go", "content": "package message\n\nimport (\n\t\"sync\"\n\t\"testing\"\n)\n\nfunc TestIDGenerator(t *testing.T) {\n\tgen := NewIDGenerator()\n\n\tt.Run(\"GenerateChunkID\", func(t *testing.T) {\n\t\tid1 := gen.GenerateChunkID()\n\t\tid2 := gen.GenerateChunkID()\n\t\tid3 := gen.GenerateChunkID()\n\n\t\tif id1 != \"C1\" {\n\t\t\tt.Errorf(\"Expected C1, got %s\", id1)\n\t\t}\n\t\tif id2 != \"C2\" {\n\t\t\tt.Errorf(\"Expected C2, got %s\", id2)\n\t\t}\n\t\tif id3 != \"C3\" {\n\t\t\tt.Errorf(\"Expected C3, got %s\", id3)\n\t\t}\n\t})\n\n\tt.Run(\"GenerateMessageID\", func(t *testing.T) {\n\t\tgen := NewIDGenerator()\n\t\tid1 := gen.GenerateMessageID()\n\t\tid2 := gen.GenerateMessageID()\n\t\tid3 := gen.GenerateMessageID()\n\n\t\tif id1 != \"M1\" {\n\t\t\tt.Errorf(\"Expected M1, got %s\", id1)\n\t\t}\n\t\tif id2 != \"M2\" {\n\t\t\tt.Errorf(\"Expected M2, got %s\", id2)\n\t\t}\n\t\tif id3 != \"M3\" {\n\t\t\tt.Errorf(\"Expected M3, got %s\", id3)\n\t\t}\n\t})\n\n\tt.Run(\"GenerateBlockID\", func(t *testing.T) {\n\t\tgen := NewIDGenerator()\n\t\tid1 := gen.GenerateBlockID()\n\t\tid2 := gen.GenerateBlockID()\n\t\tid3 := gen.GenerateBlockID()\n\n\t\tif id1 != \"B1\" {\n\t\t\tt.Errorf(\"Expected B1, got %s\", id1)\n\t\t}\n\t\tif id2 != \"B2\" {\n\t\t\tt.Errorf(\"Expected B2, got %s\", id2)\n\t\t}\n\t\tif id3 != \"B3\" {\n\t\t\tt.Errorf(\"Expected B3, got %s\", id3)\n\t\t}\n\t})\n\n\tt.Run(\"GenerateThreadID\", func(t *testing.T) {\n\t\tgen := NewIDGenerator()\n\t\tid1 := gen.GenerateThreadID()\n\t\tid2 := gen.GenerateThreadID()\n\t\tid3 := gen.GenerateThreadID()\n\n\t\tif id1 != \"T1\" {\n\t\t\tt.Errorf(\"Expected T1, got %s\", id1)\n\t\t}\n\t\tif id2 != \"T2\" {\n\t\t\tt.Errorf(\"Expected T2, got %s\", id2)\n\t\t}\n\t\tif id3 != \"T3\" {\n\t\t\tt.Errorf(\"Expected T3, got %s\", id3)\n\t\t}\n\t})\n\n\tt.Run(\"Reset\", func(t *testing.T) {\n\t\tgen := NewIDGenerator()\n\t\tgen.GenerateChunkID()\n\t\tgen.GenerateMessageID()\n\t\tgen.GenerateBlockID()\n\t\tgen.GenerateThreadID()\n\n\t\tgen.Reset()\n\n\t\tchunk, message, block, thread := gen.GetCounters()\n\t\tif chunk != 0 || message != 0 || block != 0 || thread != 0 {\n\t\t\tt.Errorf(\"Expected all counters to be 0 after reset, got chunk=%d, message=%d, block=%d, thread=%d\",\n\t\t\t\tchunk, message, block, thread)\n\t\t}\n\n\t\t// Verify IDs start from 1 again\n\t\tif id := gen.GenerateChunkID(); id != \"C1\" {\n\t\t\tt.Errorf(\"Expected C1 after reset, got %s\", id)\n\t\t}\n\t\tif id := gen.GenerateMessageID(); id != \"M1\" {\n\t\t\tt.Errorf(\"Expected M1 after reset, got %s\", id)\n\t\t}\n\t})\n\n\tt.Run(\"ConcurrentAccess\", func(t *testing.T) {\n\t\tgen := NewIDGenerator()\n\t\tvar wg sync.WaitGroup\n\t\tcount := 100\n\n\t\t// Test concurrent chunk ID generation\n\t\twg.Add(count)\n\t\tfor i := 0; i < count; i++ {\n\t\t\tgo func() {\n\t\t\t\tdefer wg.Done()\n\t\t\t\tgen.GenerateChunkID()\n\t\t\t}()\n\t\t}\n\t\twg.Wait()\n\n\t\tchunk, _, _, _ := gen.GetCounters()\n\t\tif chunk != uint64(count) {\n\t\t\tt.Errorf(\"Expected chunk counter to be %d, got %d\", count, chunk)\n\t\t}\n\t})\n\n\tt.Run(\"MultipleGenerators\", func(t *testing.T) {\n\t\tgen1 := NewIDGenerator()\n\t\tgen2 := NewIDGenerator()\n\n\t\tid1 := gen1.GenerateMessageID()\n\t\tid2 := gen2.GenerateMessageID()\n\n\t\t// Both should start from M1\n\t\tif id1 != \"M1\" || id2 != \"M1\" {\n\t\t\tt.Errorf(\"Expected both generators to start from M1, got %s and %s\", id1, id2)\n\t\t}\n\n\t\t// Advance gen1\n\t\tgen1.GenerateMessageID()\n\t\tgen1.GenerateMessageID()\n\n\t\t// gen2 should still be at M1\n\t\tid2_next := gen2.GenerateMessageID()\n\t\tif id2_next != \"M2\" {\n\t\t\tt.Errorf(\"Expected gen2 to be at M2, got %s\", id2_next)\n\t\t}\n\n\t\t// gen1 should be at M3\n\t\tid1_next := gen1.GenerateMessageID()\n\t\tif id1_next != \"M4\" {\n\t\t\tt.Errorf(\"Expected gen1 to be at M4, got %s\", id1_next)\n\t\t}\n\t})\n}\n\nfunc TestGenerateNanoID(t *testing.T) {\n\tid1 := GenerateNanoID()\n\tid2 := GenerateNanoID()\n\n\t// NanoID should be 21 characters by default\n\tif len(id1) != 21 {\n\t\tt.Errorf(\"Expected NanoID length to be 21, got %d\", len(id1))\n\t}\n\n\t// IDs should be unique\n\tif id1 == id2 {\n\t\tt.Error(\"Expected unique NanoIDs, got duplicates\")\n\t}\n\n\tt.Logf(\"Generated NanoIDs: %s, %s\", id1, id2)\n}\n\nfunc TestGenerateCustomID(t *testing.T) {\n\tid1 := GenerateCustomID(\"msg\")\n\tid2 := GenerateCustomID(\"evt\")\n\n\t// Should have prefix\n\tif len(id1) < 4 || id1[:4] != \"msg_\" {\n\t\tt.Errorf(\"Expected ID to start with 'msg_', got %s\", id1)\n\t}\n\tif len(id2) < 4 || id2[:4] != \"evt_\" {\n\t\tt.Errorf(\"Expected ID to start with 'evt_', got %s\", id2)\n\t}\n\n\t// IDs should be unique\n\tif id1 == id2 {\n\t\tt.Error(\"Expected unique custom IDs, got duplicates\")\n\t}\n\n\tt.Logf(\"Generated custom IDs: %s, %s\", id1, id2)\n}\n" }, { "path": "agent/output/output.go", "content": "package output\n\nimport (\n\t\"fmt\"\n\n\t\"github.com/yaoapp/yao/agent/output/adapters/cui\"\n\t\"github.com/yaoapp/yao/agent/output/adapters/openai\"\n\t\"github.com/yaoapp/yao/agent/output/message\"\n)\n\n// Accept type constants\nconst (\n\tAcceptStandard = \"standard\"\n\tAcceptWebCUI = \"cui-web\"\n\tAccepNativeCUI = \"cui-native\"\n\tAcceptDesktopCUI = \"cui-desktop\"\n)\n\n// Output are the options for the output\ntype Output struct {\n\tWriter message.Writer\n}\n\n// NewOutput creates a new output based on Accept type\nfunc NewOutput(options message.Options) (*Output, error) {\n\tvar writer message.Writer\n\tvar err error\n\n\t// Create writer based on Accept type\n\tswitch options.Accept {\n\tcase AcceptStandard:\n\t\t// OpenAI-compatible format\n\t\twriter, err = openai.NewWriter(options)\n\n\tcase AcceptWebCUI, AccepNativeCUI, AcceptDesktopCUI:\n\t\t// CUI format\n\t\twriter, err = cui.NewWriter(options)\n\n\tdefault:\n\t\t// Default to Standard (OpenAI)\n\t\twriter, err = openai.NewWriter(options)\n\t}\n\n\tif err != nil {\n\t\treturn nil, err\n\t}\n\n\treturn &Output{\n\t\tWriter: writer,\n\t}, nil\n}\n\n// Send sends a single message using the appropriate writer for the context\nfunc (o *Output) Send(msg *message.Message) error {\n\treturn o.Writer.Write(msg)\n}\n\n// SendGroup sends a message group using the appropriate writer for the context\nfunc (o *Output) SendGroup(group *message.Group) error {\n\treturn o.Writer.WriteGroup(group)\n}\n\n// Flush flushes the writer for the given context\nfunc (o *Output) Flush() error {\n\treturn o.Writer.Flush()\n}\n\n// Close closes the writer for the given context\nfunc (o *Output) Close() error {\n\treturn o.Writer.Close()\n}\n\n// SendMulti sends multiple messages using the appropriate writer for the context\nfunc (o *Output) SendMulti(messages ...*message.Message) error {\n\tfor _, msg := range messages {\n\t\tif err := o.Writer.Write(msg); err != nil {\n\t\t\treturn fmt.Errorf(\"failed to send message: %w\", err)\n\t\t}\n\t}\n\treturn nil\n}\n\n// // Send sends a single message using the appropriate writer for the context\n// func Send(ctx *context.Context, msg *message.Message) error {\n// \twriter, err := GetWriter(ctx)\n// \tif err != nil {\n// \t\treturn err\n// \t}\n// \treturn writer.Write(msg)\n// }\n\n// // SendGroup sends a message group using the appropriate writer for the context\n// func SendGroup(ctx *context.Context, group *message.Group) error {\n// \twriter, err := GetWriter(ctx)\n// \tif err != nil {\n// \t\treturn err\n// \t}\n// \treturn writer.WriteGroup(group)\n// }\n\n// // GetWriter gets or creates a writer for the given context\n// // Writers are cached per context to avoid recreating them\n// func GetWriter(ctx *context.Context) (message.Writer, error) {\n// \t// Try to get cached writer\n// \twriterMutex.RLock()\n// \twriter, exists := writerCache[ctx]\n// \twriterMutex.RUnlock()\n\n// \tif exists {\n// \t\treturn writer, nil\n// \t}\n\n// \t// Create new writer\n// \twriterMutex.Lock()\n// \tdefer writerMutex.Unlock()\n\n// \t// Double-check after acquiring write lock\n// \tif writer, exists := writerCache[ctx]; exists {\n// \t\treturn writer, nil\n// \t}\n\n// \t// Create writer based on context.Accept\n// \twriter, err := createWriter(ctx)\n// \tif err != nil {\n// \t\treturn nil, err\n// \t}\n\n// \t// Cache the writer\n// \twriterCache[ctx] = writer\n\n// \treturn writer, nil\n// }\n\n// // createWriter creates a writer based on context.Accept\n// func createWriter(ctx *context.Context) (message.Writer, error) {\n// \t// If global factory is set, use it\n// \tif globalFactory != nil {\n// \t\treturn globalFactory.NewWriter(ctx, nil)\n// \t}\n\n// \t// Default: create based on Accept type\n// \tswitch ctx.Accept {\n// \tcase context.AcceptStandard:\n// \t\t// OpenAI-compatible format\n// \t\treturn openai.NewWriter(ctx)\n\n// \tcase context.AcceptWebCUI, context.AccepNativeCUI, context.AcceptDesktopCUI:\n// \t\t// CUI format\n// \t\treturn cui.NewWriter(ctx)\n\n// \tdefault:\n// \t\t// Default to Standard\n// \t\treturn openai.NewWriter(ctx)\n// \t}\n// }\n\n// // SetWriterFactory sets a custom writer factory\n// // This allows applications to provide their own writer implementations\n// func SetWriterFactory(factory message.WriterFactory) {\n// \tglobalFactory = factory\n// }\n\n// // ClearWriterCache clears the writer cache\n// // Should be called when contexts are cleaned up\n// func ClearWriterCache(ctx *context.Context) {\n// \twriterMutex.Lock()\n// \tdefer writerMutex.Unlock()\n// \tdelete(writerCache, ctx)\n// }\n\n// // ClearAllWriterCache clears all cached writers\n// func ClearAllWriterCache() {\n// \twriterMutex.Lock()\n// \tdefer writerMutex.Unlock()\n// \twriterCache = make(map[*context.Context]message.Writer)\n// }\n\n// // Flush flushes the writer for the given context\n// func Flush(ctx *context.Context) error {\n// \twriter, err := GetWriter(ctx)\n// \tif err != nil {\n// \t\treturn err\n// \t}\n// \treturn writer.Flush()\n// }\n\n// // Close closes the writer for the given context and removes it from cache\n// func Close(ctx *context.Context) error {\n// \twriter, err := GetWriter(ctx)\n// \tif err != nil {\n// \t\treturn err\n// \t}\n\n// \terr = writer.Close()\n// \tClearWriterCache(ctx)\n// \treturn err\n// }\n\n// // SendMulti is a convenience function to send multiple messages\n// func SendMulti(ctx *context.Context, messages ...*message.Message) error {\n// \twriter, err := GetWriter(ctx)\n// \tif err != nil {\n// \t\treturn err\n// \t}\n\n// \tfor _, msg := range messages {\n// \t\tif err := writer.Write(msg); err != nil {\n// \t\t\treturn fmt.Errorf(\"failed to send message: %w\", err)\n// \t\t}\n// \t}\n\n// \treturn nil\n// }\n" }, { "path": "agent/output/safe_writer.go", "content": "package output\n\nimport (\n\t\"context\"\n\t\"net/http\"\n\t\"sync\"\n)\n\n// SafeWriter wraps http.ResponseWriter with a channel-based queue\n// to serialize concurrent SSE writes and prevent \"short write\" errors.\n//\n// When multiple goroutines (e.g., concurrent sub-agents via ctx.agent.All)\n// write to the same SSE stream, direct writes can cause data corruption\n// or \"short write\" errors. SafeWriter solves this by:\n//\n// 1. Accepting write requests via a buffered channel\n// 2. Processing writes sequentially in a dedicated goroutine\n// 3. Providing non-blocking writes with overflow protection\n// 4. Automatic cleanup when context is cancelled (client disconnect)\ntype SafeWriter struct {\n\tch chan writeRequest\n\twriter http.ResponseWriter\n\tdone chan struct{}\n\tctx context.Context // For detecting client disconnection\n\tcancel context.CancelFunc // To signal run() to stop\n\tcloseOnce sync.Once\n\tclosed bool\n\tmu sync.RWMutex\n}\n\n// writeRequest represents a single write request\ntype writeRequest struct {\n\tdata []byte\n}\n\n// QueueCapacity is the default buffer size for the write queue\n// Large enough to handle high concurrency without blocking\nconst QueueCapacity = 10000\n\n// NewSafeWriter creates a new SafeWriter that wraps an http.ResponseWriter\n// and starts a background goroutine to process writes sequentially.\n// The context should be the HTTP request context to detect client disconnection.\nfunc NewSafeWriter(w http.ResponseWriter) *SafeWriter {\n\t// Create internal context for graceful shutdown\n\tctx, cancel := context.WithCancel(context.Background())\n\tsw := &SafeWriter{\n\t\tch: make(chan writeRequest, QueueCapacity),\n\t\twriter: w,\n\t\tdone: make(chan struct{}),\n\t\tctx: ctx,\n\t\tcancel: cancel,\n\t}\n\tgo sw.run()\n\treturn sw\n}\n\n// NewSafeWriterWithContext creates a SafeWriter that respects the given context.\n// When the context is cancelled (e.g., client disconnects), the run() goroutine exits.\n// This prevents goroutine leaks in enterprise applications with many concurrent requests.\nfunc NewSafeWriterWithContext(ctx context.Context, w http.ResponseWriter) *SafeWriter {\n\t// Derive a cancellable context from the parent\n\tchildCtx, cancel := context.WithCancel(ctx)\n\tsw := &SafeWriter{\n\t\tch: make(chan writeRequest, QueueCapacity),\n\t\twriter: w,\n\t\tdone: make(chan struct{}),\n\t\tctx: childCtx,\n\t\tcancel: cancel,\n\t}\n\tgo sw.run()\n\treturn sw\n}\n\n// run processes write requests from the channel sequentially\n// Exits when channel is closed OR context is cancelled (client disconnect)\nfunc (sw *SafeWriter) run() {\n\tdefer close(sw.done)\n\n\tfor {\n\t\tselect {\n\t\tcase req, ok := <-sw.ch:\n\t\t\tif !ok {\n\t\t\t\t// Channel closed, exit gracefully\n\t\t\t\treturn\n\t\t\t}\n\t\t\tif sw.writer != nil {\n\t\t\t\tsw.writer.Write(req.data)\n\t\t\t\t// Flush after each write to ensure SSE data is sent immediately\n\t\t\t\tif flusher, ok := sw.writer.(http.Flusher); ok {\n\t\t\t\t\tflusher.Flush()\n\t\t\t\t}\n\t\t\t}\n\t\tcase <-sw.ctx.Done():\n\t\t\t// Context cancelled (client disconnected or explicit close)\n\t\t\t// Continue reading from channel until it's closed to avoid blocking senders\n\t\t\t// and to process any remaining messages that were already queued\n\t\t\tsw.drainUntilClosed()\n\t\t\treturn\n\t\t}\n\t}\n}\n\n// drainUntilClosed reads from channel until it's closed\n// This prevents senders from blocking after context cancellation\nfunc (sw *SafeWriter) drainUntilClosed() {\n\tfor range sw.ch {\n\t\t// Discard messages - context is cancelled so we don't write them\n\t}\n}\n\n// Write implements io.Writer interface\n// Queues the data for sequential writing by the background goroutine\nfunc (sw *SafeWriter) Write(data []byte) (int, error) {\n\tsw.mu.RLock()\n\tif sw.closed {\n\t\tsw.mu.RUnlock()\n\t\treturn 0, nil // Silently ignore writes after close\n\t}\n\tsw.mu.RUnlock()\n\n\t// Make a copy of data since the caller may reuse the buffer\n\tdataCopy := make([]byte, len(data))\n\tcopy(dataCopy, data)\n\n\t// Non-blocking send with overflow protection\n\tselect {\n\tcase sw.ch <- writeRequest{data: dataCopy}:\n\t\treturn len(data), nil\n\tdefault:\n\t\t// Channel full - this shouldn't happen with 10000 capacity\n\t\t// but if it does, drop the message rather than block\n\t\t// Note: In production, this indicates either:\n\t\t// 1. Extremely high concurrency (>10000 pending writes)\n\t\t// 2. The underlying writer is blocked/slow\n\t\t// Consider increasing QueueCapacity if this occurs frequently\n\t\treturn len(data), nil\n\t}\n}\n\n// Header returns the header map from the underlying ResponseWriter\nfunc (sw *SafeWriter) Header() http.Header {\n\tif sw.writer == nil {\n\t\treturn http.Header{}\n\t}\n\treturn sw.writer.Header()\n}\n\n// WriteHeader sends an HTTP response header with the provided status code\nfunc (sw *SafeWriter) WriteHeader(statusCode int) {\n\tif sw.writer != nil {\n\t\tsw.writer.WriteHeader(statusCode)\n\t}\n}\n\n// Flush implements http.Flusher interface\n// Note: Actual flushing happens in the run() goroutine after each write\nfunc (sw *SafeWriter) Flush() {\n\t// Flushing is handled automatically in run() after each write\n\t// This method exists to satisfy the http.Flusher interface\n}\n\n// Close closes the write channel and waits for all pending writes to complete\n// This is safe to call multiple times (idempotent via sync.Once)\nfunc (sw *SafeWriter) Close() error {\n\tsw.closeOnce.Do(func() {\n\t\t// First close channel to signal run() to stop and process remaining messages\n\t\tclose(sw.ch)\n\n\t\t// Wait for run() to finish processing all queued messages\n\t\t<-sw.done\n\n\t\t// Then mark as closed and cancel context\n\t\tsw.mu.Lock()\n\t\tsw.closed = true\n\t\tsw.mu.Unlock()\n\n\t\tsw.cancel()\n\t})\n\treturn nil\n}\n\n// IsClosed returns whether the SafeWriter has been closed\nfunc (sw *SafeWriter) IsClosed() bool {\n\tsw.mu.RLock()\n\tdefer sw.mu.RUnlock()\n\treturn sw.closed\n}\n\n// Underlying returns the underlying http.ResponseWriter\n// Use with caution - direct writes bypass the queue\nfunc (sw *SafeWriter) Underlying() http.ResponseWriter {\n\treturn sw.writer\n}\n" }, { "path": "agent/output/safe_writer_test.go", "content": "package output\n\nimport (\n\t\"bytes\"\n\t\"context\"\n\t\"net/http\"\n\t\"net/http/httptest\"\n\t\"sync\"\n\t\"testing\"\n\t\"time\"\n)\n\n// mockResponseWriter is a thread-safe mock for testing\ntype mockResponseWriter struct {\n\tmu sync.Mutex\n\tbuf bytes.Buffer\n\theader http.Header\n\tflushed int\n\twriteErr error\n}\n\nfunc newMockResponseWriter() *mockResponseWriter {\n\treturn &mockResponseWriter{\n\t\theader: make(http.Header),\n\t}\n}\n\nfunc (m *mockResponseWriter) Header() http.Header {\n\treturn m.header\n}\n\nfunc (m *mockResponseWriter) Write(data []byte) (int, error) {\n\tm.mu.Lock()\n\tdefer m.mu.Unlock()\n\tif m.writeErr != nil {\n\t\treturn 0, m.writeErr\n\t}\n\treturn m.buf.Write(data)\n}\n\nfunc (m *mockResponseWriter) WriteHeader(statusCode int) {}\n\nfunc (m *mockResponseWriter) Flush() {\n\tm.mu.Lock()\n\tdefer m.mu.Unlock()\n\tm.flushed++\n}\n\nfunc (m *mockResponseWriter) String() string {\n\tm.mu.Lock()\n\tdefer m.mu.Unlock()\n\treturn m.buf.String()\n}\n\nfunc (m *mockResponseWriter) FlushCount() int {\n\tm.mu.Lock()\n\tdefer m.mu.Unlock()\n\treturn m.flushed\n}\n\nfunc TestSafeWriter_BasicWrite(t *testing.T) {\n\tmock := newMockResponseWriter()\n\tsw := NewSafeWriter(mock)\n\tdefer sw.Close()\n\n\t// Write some data\n\tn, err := sw.Write([]byte(\"hello\"))\n\tif err != nil {\n\t\tt.Errorf(\"Write error: %v\", err)\n\t}\n\tif n != 5 {\n\t\tt.Errorf(\"Expected 5 bytes written, got %d\", n)\n\t}\n\n\t// Wait for async write to complete\n\ttime.Sleep(10 * time.Millisecond)\n\n\t// Verify data was written\n\tif got := mock.String(); got != \"hello\" {\n\t\tt.Errorf(\"Expected 'hello', got '%s'\", got)\n\t}\n}\n\nfunc TestSafeWriter_ConcurrentWrites(t *testing.T) {\n\tmock := newMockResponseWriter()\n\tsw := NewSafeWriter(mock)\n\n\t// Number of concurrent goroutines\n\tnumGoroutines := 100\n\t// Number of writes per goroutine\n\twritesPerGoroutine := 100\n\n\tvar wg sync.WaitGroup\n\twg.Add(numGoroutines)\n\n\t// Launch concurrent writes\n\tfor i := 0; i < numGoroutines; i++ {\n\t\tgo func(id int) {\n\t\t\tdefer wg.Done()\n\t\t\tfor j := 0; j < writesPerGoroutine; j++ {\n\t\t\t\tsw.Write([]byte(\"X\"))\n\t\t\t}\n\t\t}(i)\n\t}\n\n\t// Wait for all goroutines to complete\n\twg.Wait()\n\n\t// Close and wait for all writes to be processed\n\tsw.Close()\n\n\t// Verify all data was written (no data loss)\n\texpectedLen := numGoroutines * writesPerGoroutine\n\tif got := len(mock.String()); got != expectedLen {\n\t\tt.Errorf(\"Expected %d bytes, got %d\", expectedLen, got)\n\t}\n\n\t// Verify flush was called (at least once per write)\n\tif mock.FlushCount() < expectedLen {\n\t\tt.Errorf(\"Expected at least %d flushes, got %d\", expectedLen, mock.FlushCount())\n\t}\n}\n\nfunc TestSafeWriter_NoDataCorruption(t *testing.T) {\n\tmock := newMockResponseWriter()\n\tsw := NewSafeWriter(mock)\n\n\t// Use exactly 26 goroutines (one per letter A-Z) to avoid duplicates\n\tnumGoroutines := 26\n\t// Message to write (with unique content per goroutine)\n\tmsgLen := 100\n\n\tvar wg sync.WaitGroup\n\twg.Add(numGoroutines)\n\n\t// Launch concurrent writes with different content\n\tfor i := 0; i < numGoroutines; i++ {\n\t\tgo func(id int) {\n\t\t\tdefer wg.Done()\n\t\t\t// Create a message with repeating character (unique per goroutine)\n\t\t\tchar := byte('A' + id)\n\t\t\tmsg := bytes.Repeat([]byte{char}, msgLen)\n\t\t\tsw.Write(msg)\n\t\t}(i)\n\t}\n\n\t// Wait for all goroutines to complete\n\twg.Wait()\n\n\t// Close and wait for all writes to be processed\n\tsw.Close()\n\n\t// Verify total length\n\tresult := mock.String()\n\texpectedLen := numGoroutines * msgLen\n\tif len(result) != expectedLen {\n\t\tt.Errorf(\"Expected %d bytes, got %d\", expectedLen, len(result))\n\t}\n\n\t// Verify no interleaving (each message should be contiguous)\n\t// Check that we have exactly numGoroutines distinct blocks\n\tblocks := make(map[byte]int)\n\tfor i := 0; i < len(result); i += msgLen {\n\t\tif i+msgLen > len(result) {\n\t\t\tt.Errorf(\"Unexpected data at end of result\")\n\t\t\tbreak\n\t\t}\n\t\tblock := result[i : i+msgLen]\n\t\t// Verify block is homogeneous (all same character)\n\t\tfirstChar := block[0]\n\t\tfor j, c := range []byte(block) {\n\t\t\tif c != firstChar {\n\t\t\t\tt.Errorf(\"Data corruption detected at position %d: expected %c, got %c\", i+j, firstChar, c)\n\t\t\t\tbreak\n\t\t\t}\n\t\t}\n\t\tblocks[firstChar]++\n\t}\n\n\t// Each character should appear exactly once (one block per goroutine)\n\tfor char, count := range blocks {\n\t\tif count != 1 {\n\t\t\tt.Errorf(\"Character %c appeared %d times, expected 1\", char, count)\n\t\t}\n\t}\n\n\t// Verify we got all 26 letters\n\tif len(blocks) != numGoroutines {\n\t\tt.Errorf(\"Expected %d distinct blocks, got %d\", numGoroutines, len(blocks))\n\t}\n}\n\nfunc TestSafeWriter_CloseWaitsForPendingWrites(t *testing.T) {\n\tmock := newMockResponseWriter()\n\tsw := NewSafeWriter(mock)\n\n\t// Write a large number of messages\n\tnumWrites := 1000\n\tfor i := 0; i < numWrites; i++ {\n\t\tsw.Write([]byte(\"X\"))\n\t}\n\n\t// Close should wait for all writes to complete\n\tsw.Close()\n\n\t// Verify all data was written\n\tif got := len(mock.String()); got != numWrites {\n\t\tt.Errorf(\"Expected %d bytes after close, got %d\", numWrites, got)\n\t}\n}\n\nfunc TestSafeWriter_WriteAfterClose(t *testing.T) {\n\tmock := newMockResponseWriter()\n\tsw := NewSafeWriter(mock)\n\n\tsw.Write([]byte(\"before\"))\n\tsw.Close()\n\n\t// Write after close should be silently ignored\n\tn, err := sw.Write([]byte(\"after\"))\n\tif err != nil {\n\t\tt.Errorf(\"Write after close should not error: %v\", err)\n\t}\n\tif n != 0 {\n\t\tt.Errorf(\"Write after close should return 0, got %d\", n)\n\t}\n\n\t// Verify only \"before\" was written\n\tif got := mock.String(); got != \"before\" {\n\t\tt.Errorf(\"Expected 'before', got '%s'\", got)\n\t}\n}\n\nfunc TestSafeWriter_ImplementsHTTPInterfaces(t *testing.T) {\n\tmock := newMockResponseWriter()\n\tsw := NewSafeWriter(mock)\n\tdefer sw.Close()\n\n\t// Verify it implements http.ResponseWriter\n\tvar _ http.ResponseWriter = sw\n\n\t// Verify it implements http.Flusher\n\tvar _ http.Flusher = sw\n\n\t// Test Header()\n\tsw.Header().Set(\"Content-Type\", \"text/plain\")\n\tif got := mock.Header().Get(\"Content-Type\"); got != \"text/plain\" {\n\t\tt.Errorf(\"Expected Content-Type 'text/plain', got '%s'\", got)\n\t}\n}\n\n// BenchmarkSafeWriter_ConcurrentWrites benchmarks concurrent write performance\nfunc BenchmarkSafeWriter_ConcurrentWrites(b *testing.B) {\n\tmock := newMockResponseWriter()\n\tsw := NewSafeWriter(mock)\n\tdefer sw.Close()\n\n\tdata := []byte(\"benchmark data for SSE streaming\")\n\n\tb.RunParallel(func(pb *testing.PB) {\n\t\tfor pb.Next() {\n\t\t\tsw.Write(data)\n\t\t}\n\t})\n}\n\n// TestSafeWriter_RealHTTPServer tests SafeWriter with a real HTTP server\nfunc TestSafeWriter_RealHTTPServer(t *testing.T) {\n\t// This test verifies SafeWriter works correctly with httptest.ResponseRecorder\n\t// which is commonly used in testing HTTP handlers\n\n\trecorder := httptest.NewRecorder()\n\tsw := NewSafeWriter(recorder)\n\n\t// Simulate concurrent SSE writes from multiple sub-agents\n\tvar wg sync.WaitGroup\n\tnumAgents := 10\n\tmessagesPerAgent := 10\n\n\twg.Add(numAgents)\n\tfor i := 0; i < numAgents; i++ {\n\t\tgo func(agentID int) {\n\t\t\tdefer wg.Done()\n\t\t\tfor j := 0; j < messagesPerAgent; j++ {\n\t\t\t\t// Simulate SSE message format\n\t\t\t\tmsg := []byte(\"data: {\\\"agent\\\":\" + string(rune('0'+agentID)) + \"}\\n\\n\")\n\t\t\t\tsw.Write(msg)\n\t\t\t}\n\t\t}(i)\n\t}\n\n\twg.Wait()\n\tsw.Close()\n\n\t// Verify response contains all messages (no data loss)\n\tbody := recorder.Body.String()\n\texpectedMsgs := numAgents * messagesPerAgent\n\n\t// Count number of \"data: \" prefixes\n\tcount := 0\n\tfor i := 0; i < len(body); i++ {\n\t\tif i+6 <= len(body) && body[i:i+6] == \"data: \" {\n\t\t\tcount++\n\t\t}\n\t}\n\n\tif count != expectedMsgs {\n\t\tt.Errorf(\"Expected %d messages, found %d\", expectedMsgs, count)\n\t}\n}\n\n// TestSafeWriter_ContextCancellation tests that SafeWriter handles context cancellation\n// This is critical for enterprise applications to prevent goroutine leaks\nfunc TestSafeWriter_ContextCancellation(t *testing.T) {\n\tmock := newMockResponseWriter()\n\n\t// Create a cancellable context\n\tctx, cancel := context.WithCancel(context.Background())\n\n\tsw := NewSafeWriterWithContext(ctx, mock)\n\n\t// Write some data and wait for it to be processed\n\tsw.Write([]byte(\"before\"))\n\ttime.Sleep(20 * time.Millisecond)\n\n\t// Verify \"before\" was written\n\tif got := mock.String(); got != \"before\" {\n\t\tt.Errorf(\"Expected 'before' before cancel, got '%s'\", got)\n\t}\n\n\t// Cancel context (simulates client disconnect)\n\tcancel()\n\n\t// Write after context cancellation - these may or may not be written\n\t// depending on timing (select may pick ctx.Done() first)\n\tsw.Write([]byte(\"after_cancel\"))\n\n\t// Close properly cleans up\n\tsw.Close()\n\n\t// After close, run() has exited\n\tselect {\n\tcase <-sw.done:\n\t\t// Good - run() has exited\n\tdefault:\n\t\tt.Error(\"run() should have exited after Close()\")\n\t}\n\n\t// The key guarantee: run() goroutine exits cleanly, no leak\n\t// Data written before cancel is preserved\n\tgot := mock.String()\n\tif len(got) < 6 { // At least \"before\" should be there\n\t\tt.Errorf(\"Expected at least 'before', got '%s'\", got)\n\t}\n}\n\n// TestSafeWriter_GoroutineLeak tests that SafeWriter doesn't leak goroutines\nfunc TestSafeWriter_GoroutineLeak(t *testing.T) {\n\t// Create many SafeWriters and ensure they all clean up properly\n\tnumWriters := 100\n\n\tvar wg sync.WaitGroup\n\twg.Add(numWriters)\n\n\tfor i := 0; i < numWriters; i++ {\n\t\tgo func() {\n\t\t\tdefer wg.Done()\n\n\t\t\tmock := newMockResponseWriter()\n\t\t\tctx, cancel := context.WithCancel(context.Background())\n\t\t\tsw := NewSafeWriterWithContext(ctx, mock)\n\n\t\t\t// Write some data\n\t\t\tsw.Write([]byte(\"test\"))\n\n\t\t\t// Randomly either close normally or cancel context\n\t\t\tif time.Now().UnixNano()%2 == 0 {\n\t\t\t\tcancel()\n\t\t\t\ttime.Sleep(5 * time.Millisecond)\n\t\t\t\tsw.Close()\n\t\t\t} else {\n\t\t\t\tsw.Close()\n\t\t\t\tcancel() // Cancel after close is safe\n\t\t\t}\n\t\t}()\n\t}\n\n\t// All goroutines should complete\n\tdone := make(chan struct{})\n\tgo func() {\n\t\twg.Wait()\n\t\tclose(done)\n\t}()\n\n\tselect {\n\tcase <-done:\n\t\t// All completed successfully\n\tcase <-time.After(5 * time.Second):\n\t\tt.Error(\"Timeout waiting for goroutines to complete - possible goroutine leak\")\n\t}\n}\n" }, { "path": "agent/robot/DESIGN-V2-REVIEW-FINDINGS.md", "content": "# DESIGN-V2 Line-by-Line Review Findings\n\n**Review Date:** 2026-02-25 \n**Files Reviewed:** `runner.go`, `run.go`\n\n---\n\n## File 1: runner.go\n\n### 1. ExecuteTask: Is it truly single-call? No retry loop? No validation?\n\n**✅ PASS** — Lines 65–108\n\n- Non-assistant: single call to `executeNonAssistantTask` (L74), no loop\n- Assistant: single call to `executeAssistantTask` (L89), no loop\n- No `validator` import or call anywhere in the file\n- Comment at L62–63: \"V2 simplified: single call, no validation loop\"\n\n---\n\n### 2. Does it correctly split assistant vs non-assistant at the top?\n\n**✅ PASS** — Lines 72–86 vs 88–107\n\n- L72: `if task.ExecutorType != robottypes.ExecutorAssistant` — non-assistant branch first\n- L88+: assistant branch follows\n- Clear split at the top of the function\n\n---\n\n### 3. For non-assistant: does it call executeNonAssistantTask which handles MCP and Process?\n\n**✅ PASS** — Lines 74, 110–119\n\n- L74: `output, err := r.executeNonAssistantTask(task, taskCtx)`\n- `executeNonAssistantTask` (L110–119): switch on `ExecutorMCP` and `ExecutorProcess`, delegates to `ExecuteMCPTask` and `ExecuteProcessTask`\n\n---\n\n### 4. For assistant: does it call executeAssistantTask which returns (output, *CallResult, error)?\n\n**✅ PASS** — Lines 89, 123–145\n\n- L89: `output, callResult, err := r.executeAssistantTask(task, taskCtx)`\n- L123: `func (r *Runner) executeAssistantTask(...) (interface{}, *CallResult, error)`\n- L144: `return output, turnResult.Result, nil`\n\n---\n\n### 5. Does executeAssistantTask use conv.Turn() (single turn, not multi-turn)?\n\n**✅ PASS** — Lines 126, 138\n\n- L126: `conv := NewConversation(task.ExecutorID, chatID, 1)` — maxTurns=1\n- L138: `turnResult, err := conv.Turn(r.ctx, input)` — single `Turn` call, no loop\n\n---\n\n### 6. Does detectNeedMoreInfo properly check result.Next for map with \"status\" == \"need_input\"?\n\n**✅ PASS** — Lines 149–166\n\n- L150–151: nil checks for `result` and `result.Next`\n- L154: type assertion to `map[string]interface{}`\n- L157–159: `status, _ := m[\"status\"].(string); if status != \"need_input\" return false`\n- Matches DESIGN §16.5 protocol\n\n---\n\n### 7. Does it extract \"question\" from the map? What happens if question is empty?\n\n**✅ PASS** — Lines 161–165\n\n- L161: `question, _ := m[\"question\"].(string)`\n- L163–164: `if question == \"\" { question = result.GetText() }` — fallback to `CallResult` text\n- Empty question handled via fallback\n\n---\n\n### 8. Are result.NeedInput and result.InputQuestion set correctly?\n\n**✅ PASS** — Lines 102–105\n\n- L102–104: `if needInput, question := detectNeedMoreInfo(callResult); needInput { result.NeedInput = true; result.InputQuestion = question }`\n- Set only when `detectNeedMoreInfo` returns true\n\n---\n\n### 9. Does result.Duration get set in all paths (success and failure)?\n\n**✅ PASS** — Lines 77, 84, 92, 98\n\n- Non-assistant error: L77\n- Non-assistant success: L84\n- Assistant error: L92\n- Assistant success: L98\n- All paths set `result.Duration = time.Since(startTime).Milliseconds()`\n\n---\n\n### 10. Does buildResult helper exist or is result construction inline?\n\n**✅ PASS (inline)** — Lines 68–107\n\n- No `buildResult` helper; construction is inline\n- DESIGN §16.4 pseudocode uses `buildResult`; inline construction is acceptable and used here\n\n---\n\n### 11. Any edge cases: what if task.ExecutorType is empty or unknown?\n\n**✅ PASS** — Lines 72, 112–118\n\n- Empty/unknown: `!= ExecutorAssistant` is true → non-assistant branch\n- L116–117: `default` returns `fmt.Errorf(\"unsupported executor type: %s (expected mcp or process)\", task.ExecutorType)`\n- Error returned and propagated; no silent failure\n\n---\n\n### Additional Finding (runner.go)\n\n**⚠️ Minor:** DESIGN §9.1 shows `event.Push(\"robot.task.failed\", ...)` inside `ExecuteTask` when `err != nil`. Implementation pushes `TaskFailed` from `run.go` (L113–120) when `result.Success` is false. Behavior is equivalent; only location differs.\n\n---\n\n## File 2: run.go\n\n### 1. DefaultRunConfig — ContinueOnFailure defaults to true?\n\n**✅ PASS** — Lines 21–26\n\n- L23–25: `return &RunConfig{ ContinueOnFailure: true }`\n- Matches DESIGN §6.3\n\n---\n\n### 2. RunExecution — does it check exec.ResumeContext for startIndex and PreviousResults?\n\n**✅ PASS** — Lines 60–66\n\n- L61–64: `if exec.ResumeContext != nil { startIndex = exec.ResumeContext.TaskIndex; exec.Results = exec.ResumeContext.PreviousResults }`\n- L72: loop starts at `startIndex`\n- Matches DESIGN §9.2, §16.3\n\n---\n\n### 3. Does it NOT reset Results when ResumeContext is present?\n\n**✅ PASS** — Lines 62–65\n\n- When `ResumeContext != nil`: `exec.Results = exec.ResumeContext.PreviousResults` — restores, does not reset\n- When `ResumeContext == nil`: `exec.Results = make(...)` — fresh slice\n\n---\n\n### 4. Does it set task.Status to TaskRunning before execution?\n\n**✅ PASS** — Lines 86–89\n\n- L87: `task.Status = robottypes.TaskRunning`\n- L88–89: `task.StartTime = &now`\n- Set before `ExecuteTask` (L98)\n\n---\n\n### 5. Does it call e.updateTasksState to persist running state?\n\n**✅ PASS** — Line 92\n\n- L92: `e.updateTasksState(ctx, exec)` immediately after setting task status\n- Persists running state before execution\n\n---\n\n### 6. Does result.NeedInput trigger e.Suspend(ctx, exec, i, result.InputQuestion)?\n\n**✅ PASS** — Lines 100–103\n\n- L100–102: `if result.NeedInput { return e.Suspend(ctx, exec, i, result.InputQuestion) }`\n- Correct parameters and early return\n\n---\n\n### 7. Is the result NOT appended before Suspend (avoiding duplicate results per §16.15)?\n\n**✅ PASS** — Lines 100–103, 124\n\n- L100–102: `NeedInput` branch returns before any append\n- L124: `exec.Results = append(exec.Results, *result)` is after the `NeedInput` check\n- No append on suspend; matches DESIGN §16.15\n\n---\n\n### 8. Does it push event.Push for TaskFailed when a task fails?\n\n**✅ PASS** — Lines 113–120\n\n- L113–120: `event.Push(ctx.Context, robotevents.TaskFailed, robotevents.NeedInputPayload{...})` when `!result.Success`\n- Event is pushed on task failure\n\n**⚠️ Minor:** Uses `NeedInputPayload` with `Question: result.Error`. DESIGN §7.2 does not define a TaskFailed payload. `ExecPayload` (with `Error`) might be more appropriate; `NeedInputPayload.Question` is reused for the error message. Functionally acceptable but semantically odd.\n\n---\n\n### 9. Does it skip remaining tasks when ContinueOnFailure is false?\n\n**✅ PASS** — Lines 129–137\n\n- L129: `if !result.Success && !config.ContinueOnFailure`\n- L131–134: marks remaining tasks as `TaskSkipped`\n- L136: `return fmt.Errorf(...)` — stops execution\n- Matches DESIGN §9.2\n\n---\n\n### 10. Does it clear exec.Current and exec.ResumeContext after completion?\n\n**✅ PASS** — Lines 141–143\n\n- L142–143: `exec.Current = nil; exec.ResumeContext = nil` after loop completes\n- Only on normal completion (no early return from Suspend or failure)\n\n---\n\n### 11. Is there any event.Push for TaskCompleted?\n\n**❌ FINDING** — run.go\n\n- No `event.Push(robotevents.TaskCompleted, ...)` when a task succeeds\n- DESIGN §7.2 defines `EventTaskCompleted = \"robot.task.completed\"`\n- DESIGN-V2 §20.4 (B5) notes missing TaskCompleted event constant; implementation also does not push it\n- **Recommendation:** Add `event.Push(ctx.Context, robotevents.TaskCompleted, payload)` when `result.Success` (e.g. after L110)\n\n---\n\n### 12. Does getRunConfig properly handle nil data?\n\n**✅ PASS** — Lines 50–55\n\n- No separate `getRunConfig`; config is obtained inline\n- L51–55: `if cfg, ok := data.(*RunConfig); ok && cfg != nil { config = cfg } else { config = DefaultRunConfig() }`\n- Handles: `data == nil`, wrong type, `cfg == nil` → falls back to `DefaultRunConfig()`\n\n---\n\n### Additional Finding (run.go)\n\n**⚠️ Order of operations:** Task status update (L109–120) and result append (L124) occur *after* the NeedInput check. Flow is correct: NeedInput → Suspend (return) → no append, no status update for that task.\n\n---\n\n## Summary\n\n| Category | runner.go | run.go |\n|----------|-----------|--------|\n| **PASS** | 11/11 | 11/12 |\n| **Minor** | 1 | 1 |\n| **Finding** | 0 | 1 (TaskCompleted not pushed) |\n\n### Action Items\n\n1. **run.go L109–110:** Add `event.Push(ctx.Context, robotevents.TaskCompleted, payload)` when `result.Success` for per-task completion events.\n2. **run.go L113:** Consider introducing a `TaskFailedPayload` (or using `ExecPayload` with `Error`) instead of `NeedInputPayload` for TaskFailed events.\n" }, { "path": "agent/robot/DESIGN.md", "content": "# Robot Agent\n\n## 1. What is it?\n\nA **Robot Agent** is an AI team member. It works on its own, makes decisions, and runs tasks without waiting for user input.\n\n**Key points:**\n\n- Belongs to a Team, managed like human members\n- Has clear job duties (e.g., \"Sales Manager: track KPIs, make reports\")\n- Created and deleted via Team API\n- Runs on schedule, or when triggered by humans or events\n- Learns from each run, stores knowledge in private KB\n\n---\n\n## 2. Architecture\n\n### 2.1 System Flow\n\n> **Architecture Note:** All trigger types flow through Manager.\n>\n> - Clock: `Manager.Tick()` (internal ticker)\n> - Human: `Manager.Intervene()` (API call)\n> - Event: `Manager.HandleEvent()` (webhook/db trigger)\n>\n> The `trigger/` package provides utilities only (validation, clock matching, execution control).\n\n```mermaid\nflowchart TB\n subgraph Triggers[\"Triggers\"]\n WC[/\"⏰ Clock\"/]\n HI[/\"👤 Human\"/]\n EV[/\"📡 Event\"/]\n end\n\n subgraph Manager[\"Manager (Central Orchestrator)\"]\n TC{\"Enabled?\"}\n Cache[(\"Cache\")]\n Dedup{\"Dedup?\"}\n Queue[\"Queue\"]\n end\n\n subgraph Pool[\"Workers\"]\n W1[\"Worker\"]\n W2[\"Worker\"]\n W3[\"Worker\"]\n end\n\n subgraph Executor[\"Executor\"]\n TT{\"Trigger?\"}\n P0[\"P0: Inspiration\"]\n P1[\"P1: Goals\"]\n P2[\"P2: Tasks\"]\n P3[\"P3: Run\"]\n P4[\"P4: Deliver\"]\n P5[\"P5: Learn\"]\n end\n\n subgraph Storage[\"Storage\"]\n KB[(\"KB\")]\n DB[(\"DB\")]\n end\n\n WC --> TC\n HI & EV --> TC\n TC -->|Yes| Cache\n TC -->|No| X[/Skip/]\n Cache --> Dedup\n Dedup -->|OK| Queue\n Dedup -->|Dup| Cache\n Queue --> W1 & W2 & W3\n W1 & W2 & W3 --> TT\n TT -->|Clock| P0\n TT -->|Human/Event| P1\n P0 --> P1 --> P2 --> P3 --> P4 --> P5\n P5 --> KB & DB\n KB -.->|History| P0\n```\n\n### 2.2 Executor Modes\n\nExecutor supports multiple execution modes for different use cases:\n\n| Mode | Use Case | Status |\n| -------- | --------------------------------------- | ------------------ |\n| Standard | Production with real Agent calls | ✅ Implemented |\n| DryRun | Tests, demos, preview without LLM calls | ✅ Implemented |\n| Sandbox | Container-isolated for untrusted code | ⬜ Not Implemented |\n\n**Standard Mode:** Real execution with LLM calls, full phase execution, logging via kun/log.\n\n**DryRun Mode:** Simulated execution without LLM calls. Used for:\n\n- Unit tests and integration tests\n- Demo and preview modes\n- Scheduling and concurrency testing\n\n**Sandbox Mode (Future):** Container-level isolation (Docker/gVisor/Firecracker) for:\n\n- Untrusted robot configurations\n- Multi-tenant environments\n- Resource-limited execution\n\n> **⚠️ Sandbox requires infrastructure support.** Current placeholder behaves like DryRun.\n\n### 2.3 Team Structure\n\nUses existing `__yao.member` model (`yao/models/member.mod.yao`):\n\n```\n┌─────────────────────────────────────────────────────────────────┐\n│ Team │\n│ ┌─────────────────────────────────────────────────────────┐ │\n│ │ Robot Members │ │\n│ │ ┌─────────────┐ ┌─────────────┐ ┌─────────────┐ │ │\n│ │ │Sales Manager│ │Data Analyst │ │CS Specialist│ │ │\n│ │ │ • Track KPIs│ │ • Analyze │ │ • Tickets │ │ │\n│ │ │ • Reports │ │ • Reports │ │ • Inquiries │ │ │\n│ │ └─────────────┘ └─────────────┘ └─────────────┘ │ │\n│ └─────────────────────────────────────────────────────────┘ │\n│ ┌─────────────────────────────────────────────────────────┐ │\n│ │ User Members │ │\n│ │ ┌─────────────┐ ┌─────────────┐ │ │\n│ │ │ John (Owner)│ │ Jane (Admin)│ │ │\n│ │ └─────────────┘ └─────────────┘ │ │\n│ └─────────────────────────────────────────────────────────┘ │\n└─────────────────────────────────────────────────────────────────┘\n```\n\n**Key fields in `__yao.member` for robot agents:**\n\n| Field | Type | Description |\n| ----------------- | ------ | ----------------------------------------------------------- |\n| `member_type` | enum | `user` \\| `robot` |\n| `autonomous_mode` | bool | Enable robot execution |\n| `robot_config` | JSON | Agent configuration (see section 5) |\n| `robot_status` | enum | `idle` \\| `working` \\| `paused` \\| `error` \\| `maintenance` |\n| `system_prompt` | text | Identity & role prompt |\n| `robot_email` | string | Robot's email address for sending emails (From address) |\n| `agents` | JSON | Accessible agents list |\n| `mcp_servers` | JSON | Accessible MCP servers |\n| `manager_id` | string | Direct manager user ID |\n\n---\n\n## 3. How It Works\n\n### 3.1 Flow: Trigger → Schedule → Run\n\n```mermaid\nsequenceDiagram\n autonumber\n participant T as Trigger\n participant M as Manager\n participant S as Scheduler\n participant W as Worker\n participant E as Executor\n participant A as Phase Agents\n participant KB as KB\n\n T->>M: Event\n M->>M: Check enabled\n M->>M: Get from cache\n M->>M: Check dedup\n M->>S: Submit\n\n S->>S: Check quota\n S->>S: Sort by priority\n S->>W: Dispatch\n\n W->>E: Run\n\n alt Clock trigger\n E->>A: P0: Inspiration (with clock context)\n A-->>E: Report\n end\n\n loop P1 to P5\n E->>A: Call agent\n A-->>E: Result\n end\n\n E->>KB: Save learning\n E-->>W: Done\n```\n\n### 3.2 Triggers\n\n| Type | What | Config | Handler |\n| --------- | ----------------------------- | -------------------- | ----------------------- |\n| **Clock** | Timer (times/interval/daemon) | `triggers.clock` | `Manager.Tick()` |\n| **Human** | Manual action | `triggers.intervene` | `Manager.Intervene()` |\n| **Event** | Webhook, DB change | `triggers.event` | `Manager.HandleEvent()` |\n\nAll on by default. Turn off per agent:\n\n```yaml\ntriggers:\n clock: { enabled: true }\n intervene: { enabled: true, actions: [\"task.add\", \"goal.adjust\"] }\n event: { enabled: false }\n```\n\n### 3.3 Concurrency\n\nTwo levels to prevent one agent from using all resources:\n\n```\n┌─────────────────────────────────────────────────────────────────┐\n│ Global Pool (10 workers) │\n└─────────────────────────────────────────────────────────────────┘\n │ │ │\n ▼ ▼ ▼\n┌─────────────────┐ ┌─────────────────┐ ┌─────────────────┐\n│ Sales Manager │ │ Data Analyst │ │ CS Specialist │\n│ Limit: 3 │ │ Limit: 2 │ │ Limit: 3 │\n│ Now: 2 ✓ │ │ Now: 2 (full) │ │ Now: 1 ✓ │\n└─────────────────┘ └─────────────────┘ └─────────────────┘\n```\n\n### 3.4 Dedup\n\n**Fast check** (in memory):\n\n```go\nkey := memberID + \":\" + triggerType + \":\" + window\nif has(key) { skip }\n```\n\n**Smart check** (for goals/tasks):\n\n- Dedup Agent looks at history\n- Returns: `skip` | `merge` | `proceed`\n\n### 3.5 Cache\n\nKeeps agents in memory. No DB query on each tick:\n\n```go\ntype AgentCache struct {\n agents map[string]*Agent // member_id -> agent\n byTeam map[string][]string // team_id -> member_ids\n}\n// Refresh: on start, on change, every hour\n```\n\n---\n\n## 4. Phases\n\n### 4.1 Overview\n\n```\nClock: P0 → P1 → P2 → P3 → P4 → P5\nHuman/Event: P1 → P2 → P3 → P4 → P5\n```\n\n| Phase | Agent | In | Out | When |\n| ----- | ----------- | ------------------- | --------------- | ---------- |\n| P0 | Inspiration | Clock + Data + News | Report | Clock only |\n| P1 | Goal Gen | Report + history | Goals | Always |\n| P2 | Task Plan | Goals + tools | Tasks | Always |\n| P3 | Run + Valid | Tasks + Experts | TaskResults | Always |\n| P4 | Delivery | All results | Email/Webhook/Process | Always |\n| P5 | Learning | Summary | KB entries | Always |\n\n### 4.2 P0: Inspiration (Clock only)\n\n**Skipped for Human/Event triggers.** They already have clear intent.\n\nGathers info to help make good goals. **Clock context is key input** - Agent knows what time it is and can decide what to do (e.g., 5pm Friday → write weekly report).\n\n```go\ntype InspirationReport struct {\n Clock *ClockContext `json:\"clock\"` // time context\n Content string `json:\"content\"` // markdown text for LLM\n}\n// Content is markdown like:\n// ## Summary\n// ...\n// ## Highlights\n// - [High] Sales up 50%\n// ## Opportunities / Risks / World News / Pending\n// ...\n\ntype ClockContext struct {\n Now time.Time // Current time\n Hour int // 0-23\n DayOfWeek string // Monday, Tuesday...\n DayOfMonth int // 1-31\n IsWeekend bool\n IsMonthStart bool // 1st-3rd\n IsMonthEnd bool // last 3 days\n IsQuarterEnd bool\n // Agent uses this to decide: \"It's 5pm Friday, time for weekly report\"\n}\n```\n\n**Sources:**\n\n- **Clock**: Current time, day of week, month end, etc.\n- Internal: Data changes, events, feedback, pending work\n- External: Web search (news, competitors)\n\n### 4.3 P1: Goals\n\n**For Clock:** Uses inspiration report (with clock context) to make goals. Agent decides based on time what's important now.\n\n**For Human/Event:** Uses the input directly as goals (or to generate goals).\n\n```go\ntype Goals struct {\n Content string // markdown text (for LLM)\n Delivery *DeliveryTarget // where to send results (for P4)\n}\n\ntype DeliveryTarget struct {\n Type DeliveryType // Preferred delivery type (P4 will use Delivery Center)\n Recipients []string // email addresses, webhook URLs, user IDs\n Format string // markdown | html | json | text\n Template string // template name\n Options map[string]interface{}\n}\n```\n\n**Example prompt:**\n\n```\nYou are [Sales Manager]. Your job: [track KPIs, make reports].\n\n## Report\n### Key Items\n- [High] Data: 15 new sales (+50%)\n- [High] Deadline: Friday report due\n- [High] News: Competitor launched product\n\n### Chances\n- Sales up 20% vs last week\n- Market growing\n\nMake today's goals.\n```\n\n**Note:** Validation criteria (`ExpectedOutput`, `ValidationRules`) are defined at the **Task level** (P2), not Goals level. This allows each task to have specific validation rules for P3.\n\n### 4.4 P2: Tasks\n\nP2 Agent reads Goals markdown and breaks into executable tasks:\n\n```go\ntype Task struct {\n ID string // unique task ID\n Description string // human-readable task description (for UI display)\n Messages []context.Message // original input (text, images, files, audio)\n GoalRef string // reference to goal (e.g., \"Goal 1\")\n Source TaskSource // auto | human | event\n ExecutorType ExecutorType // assistant | mcp | process\n ExecutorID string // agent ID or mcp tool name\n Args []any // arguments for executor\n Order int // execution order\n\n // Validation criteria (used in P3)\n ExpectedOutput string // what the task should produce\n ValidationRules []string // specific checks to perform\n}\n```\n\n### 4.5 P3: Run\n\n**Architecture:** P3 uses a modular design with three components:\n\n```\n┌─────────────────────────────────────────────────────────────┐\n│ run.go (P3 Entry) │\n│ - RunConfig: ContinueOnFailure, ValidationThreshold, │\n│ MaxTurnsPerTask │\n│ - RunExecution: main loop with task dependency passing │\n└─────────────────────┬───────────────────────────────────────┘\n │\n ┌────────────┴────────────┐\n ▼ ▼\n┌─────────────────┐ ┌─────────────────┐\n│ runner.go │ │ validator.go │\n│ - Runner │ │ - Validator │\n│ - Multi-turn │ │ - Two-layer │\n│ conversation │ │ - Rule+Semantic│\n│ - Task context │ │ - NeedReply │\n│ building │ │ - ReplyContent │\n└────────┬────────┘ └────────┬────────┘\n │ │\n │ ▼\n │ ┌─────────────────┐\n │ │ yao/assert │\n │ │ - Asserter │\n │ │ - 8 types │\n │ │ - Extensible │\n │ └─────────────────┘\n ▼\n┌─────────────────────────────────────────┐\n│ Executor Types │\n│ - ExecutorAssistant → Multi-turn AI │\n│ - ExecutorMCP → Single-call MCP tool │\n│ - ExecutorProcess → Single-call Process │\n└─────────────────────────────────────────┘\n```\n\n**Execution Flow:**\n\nFor each task:\n\n1. **Build Context**: Include previous task results as context\n2. **Execute**: Call appropriate executor (Assistant/MCP/Process)\n3. **Validate**: Use two-layer validation (rule-based + semantic)\n4. **Continue or Complete**:\n - For Assistant tasks: If `NeedReply`, continue conversation with `ReplyContent`\n - For MCP/Process tasks: Single-call execution, no multi-turn\n5. **Update**: Set task status and store result\n\n**Task Dependency**: Previous task results are automatically passed as context to subsequent tasks via `Runner.BuildTaskContext()` and formatted using `FormatPreviousResultsAsContext()`.\n\n**Two-Layer Validation:**\n\n| Layer | Method | Speed | Use Case |\n|-------|--------|-------|----------|\n| 1. Rule-based | `yao/assert` | Fast | Type check, contains, regex, json_path |\n| 2. Semantic | Validation Agent | Slow | ExpectedOutput, complex criteria |\n\n**Executor Types:**\n\n| Type | ExecutorID Format | Example |\n|------|-------------------|---------|\n| `assistant` | Agent ID | `experts.text-writer` |\n| `mcp` | `mcp_server.mcp_tool` | `ark.image.text2img.generate` |\n| `process` | Process name | `models.user.Find` |\n\n**MCP Task Fields:**\n\nFor MCP tasks, three fields are required:\n- `executor_id`: Combined format `mcp_server.mcp_tool`\n- `mcp_server`: MCP server/client ID (e.g., `ark.image.text2img`)\n- `mcp_tool`: Tool name within the server (e.g., `generate`)\n\n**Multi-Turn Conversation Flow:**\n\nFor assistant tasks, P3 uses a multi-turn conversation approach:\n1. **Call**: Call assistant and get result\n2. **Validate**: Validate result (determines: passed, complete, needReply, replyContent)\n3. **Reply**: If needReply, continue conversation with replyContent\n4. **Repeat**: Until complete or max turns exceeded\n\nThe `Validator.ValidateWithContext()` method determines:\n- `Complete`: Whether the expected result is obtained\n- `NeedReply`: Whether to continue conversation\n- `ReplyContent`: What to send in the next turn (validation feedback, clarification request, etc.)\n\nThis replaces the traditional retry mechanism with intelligent conversation continuation.\n\n```go\n// RunConfig configures P3 execution behavior\ntype RunConfig struct {\n ContinueOnFailure bool // continue to next task even if current fails (default: false)\n ValidationThreshold float64 // minimum score to pass validation (default: 0.6)\n MaxTurnsPerTask int // max conversation turns per task (default: 10)\n}\n\n// ValidationResult with multi-turn conversation support\ntype ValidationResult struct {\n // Basic validation result\n Passed bool // overall validation passed\n Score float64 // 0-1 confidence score\n Issues []string // what failed\n Suggestions []string // how to improve\n Details string // detailed report (markdown)\n\n // Execution state (for multi-turn conversation control)\n Complete bool // whether expected result is obtained\n NeedReply bool // whether to continue conversation\n ReplyContent string // content for next turn (if NeedReply)\n}\n```\n\n**yao/assert Package:**\n\nUniversal assertion library supporting 8 types:\n\n| Type | Description | Example |\n|------|-------------|---------|\n| `equals` | Exact match | `{\"type\": \"equals\", \"value\": \"success\"}` |\n| `contains` | Substring check | `{\"type\": \"contains\", \"value\": \"total\"}` |\n| `not_contains` | Negative check | `{\"type\": \"not_contains\", \"value\": \"error\"}` |\n| `json_path` | JSON path extraction | `{\"type\": \"json_path\", \"path\": \"data.count\", \"value\": 10}` |\n| `regex` | Pattern matching | `{\"type\": \"regex\", \"value\": \"^[A-Z].*\"}` |\n| `type` | Type checking | `{\"type\": \"type\", \"value\": \"array\"}` |\n| `script` | Custom script | `{\"type\": \"script\", \"script\": \"scripts.validate\"}` |\n| `agent` | AI validation | `{\"type\": \"agent\", \"use\": \"validator\"}` |\n\n### 4.6 P4: Deliver\n\nP4 generates delivery content and pushes to Delivery Center. **Agent only generates content, Delivery Center decides channels.**\n\n**Architecture:**\n\n```\n┌─────────────────────────────────────────────────────────────┐\n│ P4 Delivery Agent │\n│ Role: Generate content only (Summary, Body, Attachments) │\n│ NOT responsible for: Channel selection │\n└─────────────────────┬───────────────────────────────────────┘\n │\n ▼\n┌─────────────────────────────────────────────────────────────┐\n│ DeliveryRequest │\n│ - Content: Summary, Body, Attachments │\n│ - Context: member_id, execution_id, trigger, team │\n│ (No Channels - Delivery Center decides) │\n└─────────────────────┬───────────────────────────────────────┘\n │\n ▼\n┌─────────────────────────────────────────────────────────────┐\n│ Delivery Center │\n│ Role: │\n│ 1. Read Robot/User delivery preferences │\n│ 2. Decide which channels to use │\n│ 3. Execute delivery (email, webhook, process) │\n│ 4. Future: auto-notify based on user subscriptions │\n│ │\n│ (Current: internal, future: yao/delivery) │\n└─────────────────────────────────────────────────────────────┘\n```\n\n**Key Design:**\n- **Separation of concerns**: Agent generates content, Delivery Center handles channels\n- **User preferences**: Channels decided by Robot/User configuration, not Agent\n- **Automatic delivery**: If webhook configured, every execution pushes automatically\n- **Future-ready**: Delivery Center can be extracted to `yao/delivery` package\n\n**Delivery Request Structure:**\n\n```go\n// DeliveryRequest - pushed to Delivery Center\n// No Channels field - Delivery Center decides based on preferences\ntype DeliveryRequest struct {\n Content *DeliveryContent `json:\"content\"` // Agent-generated content\n Context *DeliveryContext `json:\"context\"` // Tracking info\n}\n\n// DeliveryContent - content generated by Delivery Agent\ntype DeliveryContent struct {\n Summary string `json:\"summary\"` // Brief 1-2 sentence summary\n Body string `json:\"body\"` // Full markdown report\n Attachments []DeliveryAttachment `json:\"attachments,omitempty\"` // Output artifacts\n}\n\n// DeliveryAttachment - task output attachment with metadata\ntype DeliveryAttachment struct {\n Title string `json:\"title\"` // Human-readable title\n Description string `json:\"description,omitempty\"` // What this artifact is\n TaskID string `json:\"task_id,omitempty\"` // Which task produced this\n File string `json:\"file\"` // Wrapper: __://\n}\n\n// DeliveryContext - tracking and audit info\ntype DeliveryContext struct {\n MemberID string `json:\"member_id\"` // Robot member ID (globally unique)\n ExecutionID string `json:\"execution_id\"`\n TriggerType TriggerType `json:\"trigger_type\"`\n TeamID string `json:\"team_id\"`\n}\n```\n\n**File Wrapper Format:**\n\nAttachments use the standard `yao/attachment` wrapper format:\n- Format: `__://`\n- Example: `__yao.attachment://ccd472d11feb96e03a3fc468f494045c`\n- Parse: `attachment.Parse(value)` → `(uploader, fileID, isWrapper)`\n- Read: `attachment.Base64(ctx, value)` → base64 content\n\n**Delivery Channels (Delivery Center decides):**\n\n| Channel | Description | Multiple Targets |\n|---------|-------------|------------------|\n| `email` | Send via yao/messenger | ✅ Multiple recipients/emails |\n| `webhook` | POST to external URL | ✅ Multiple URLs |\n| `process` | Yao Process call | ✅ Multiple processes |\n| `notify` | In-app notification | Future (auto by subscriptions) |\n\n**Delivery Agent:**\n\nThe Delivery Agent **only generates content**, does NOT decide channels:\n\n```go\n// Delivery Agent Input\ntype DeliveryAgentInput struct {\n Robot *Robot `json:\"robot\"`\n TriggerType TriggerType `json:\"trigger\"`\n Inspiration *InspirationReport `json:\"inspiration\"` // P0\n Goals *Goals `json:\"goals\"` // P1\n Tasks []Task `json:\"tasks\"` // P2\n Results []TaskResult `json:\"results\"` // P3\n}\n\n// Delivery Agent Output - only content, no channels\ntype DeliveryAgentOutput struct {\n Content *DeliveryContent `json:\"content\"`\n}\n```\n\n**Example Agent Output:**\n\n```json\n{\n \"content\": {\n \"summary\": \"Sales report completed: 15 new leads processed\",\n \"body\": \"## Weekly Sales Report\\n\\n### Summary\\n...\",\n \"attachments\": [\n {\"title\": \"Sales Report.pdf\", \"file\": \"__yao.attachment://abc123\"},\n {\"title\": \"Lead Analysis.xlsx\", \"file\": \"__yao.attachment://def456\"}\n ]\n }\n}\n```\n\n**Delivery Result:**\n\n```go\n// DeliveryResult - returned by Delivery Center\ntype DeliveryResult struct {\n RequestID string `json:\"request_id\"` // Delivery request ID\n Content *DeliveryContent `json:\"content\"` // Agent-generated content\n Results []ChannelResult `json:\"results,omitempty\"` // Results per channel\n Success bool `json:\"success\"` // Overall success\n Error string `json:\"error,omitempty\"` // Error if failed\n SentAt *time.Time `json:\"sent_at,omitempty\"` // When delivery completed\n}\n\n// ChannelResult - result for a single delivery target\ntype ChannelResult struct {\n Type DeliveryType `json:\"type\"` // email | webhook | process\n Target string `json:\"target\"` // Target identifier (email, URL, process name)\n Success bool `json:\"success\"` // Whether delivery succeeded\n Recipients []string `json:\"recipients,omitempty\"` // Who received (for email)\n Details interface{} `json:\"details,omitempty\"` // Channel-specific response\n Error string `json:\"error,omitempty\"` // Error message if failed\n SentAt *time.Time `json:\"sent_at,omitempty\"` // When this target was delivered\n}\n```\n\n**Config (Delivery Preferences):**\n\nRobot config defines delivery **preferences** (Delivery Center reads and executes).\nEach channel supports **multiple targets**:\n\n```yaml\ndelivery:\n preferences:\n email:\n enabled: true\n targets: # Multiple email targets\n - to: [\"manager@company.com\"]\n cc: [\"team@company.com\"]\n - to: [\"ceo@company.com\"]\n subject_template: \"Executive Summary\"\n \n webhook:\n enabled: true\n targets: # Multiple webhook URLs\n - url: \"https://slack.com/webhook/sales\"\n - url: \"https://feishu.cn/webhook/reports\"\n headers: {\"X-Custom\": \"value\"}\n \n process:\n enabled: true\n targets: # Multiple Yao Process calls\n - name: \"orders.UpdateStatus\"\n args: [\"completed\"]\n - name: \"audit.LogDelivery\"\n\n# Note: notify handled by Delivery Center based on user subscriptions (future)\n```\n\n**Use Cases:**\n\n| Scenario | Channels | Description |\n|----------|----------|-------------|\n| Event callback | `process` | DB change → Robot → Update data via Process |\n| Multi-channel notify | `email` + `webhook` | Send to multiple emails and Slack/飞书 |\n| Data pipeline | `process` | Robot result → Save to DB → Update dashboard |\n\n### 4.7 P5: Learn\n\nSave to KB:\n\n| Type | Examples |\n| ----------- | ------------------------ |\n| `execution` | What worked, what failed |\n| `feedback` | Errors, fixes |\n| `insight` | Patterns, tips |\n\n---\n\n## 5. Config\n\n### 5.1 Structure\n\n```go\ntype Config struct {\n Triggers *Triggers `json:\"triggers,omitempty\"`\n Clock *Clock `json:\"clock,omitempty\"`\n Identity *Identity `json:\"identity\"`\n Quota *Quota `json:\"quota\"`\n KB *KB `json:\"kb,omitempty\"` // shared KB (same as assistant)\n DB *DB `json:\"db,omitempty\"` // shared DB (same as assistant)\n Learn *Learn `json:\"learn,omitempty\"` // learning for private KB\n Resources *Resources `json:\"resources\"`\n Delivery *DeliveryPreferences `json:\"delivery,omitempty\"`\n Events []Event `json:\"events,omitempty\"`\n Executor *Executor `json:\"executor,omitempty\"` // executor mode settings\n DefaultLocale string `json:\"default_locale,omitempty\"` // default language for clock/event triggers (\"en-US\", \"zh-CN\")\n}\n```\n\n### 5.2 Types\n\n```go\n// Phase - execution phase enum\ntype Phase string\n\nconst (\n PhaseInspiration Phase = \"inspiration\" // P0: Clock only\n PhaseGoals Phase = \"goals\" // P1\n PhaseTasks Phase = \"tasks\" // P2\n PhaseRun Phase = \"run\" // P3 (execution + validation)\n PhaseDelivery Phase = \"delivery\" // P4\n PhaseLearning Phase = \"learning\" // P5\n)\n\n// AllPhases for iteration\nvar AllPhases = []Phase{\n PhaseInspiration, PhaseGoals, PhaseTasks,\n PhaseRun, PhaseDelivery, PhaseLearning,\n}\n\n// ClockMode - clock trigger mode enum\ntype ClockMode string\n\nconst (\n ClockModeTimes ClockMode = \"times\" // run at specific times\n ClockModeInterval ClockMode = \"interval\" // run every X duration\n ClockModeDaemon ClockMode = \"daemon\" // run continuously\n)\n\n// DeliveryType - output delivery type enum\ntype DeliveryType string\n\nconst (\n DeliveryEmail DeliveryType = \"email\" // Email via yao/messenger\n DeliveryWebhook DeliveryType = \"webhook\" // POST to external URL\n DeliveryProcess DeliveryType = \"process\" // Yao Process call\n DeliveryNotify DeliveryType = \"notify\" // In-app notification (future)\n)\n\n// ExecStatus - execution status enum\ntype ExecStatus string\n\nconst (\n ExecPending ExecStatus = \"pending\"\n ExecRunning ExecStatus = \"running\"\n ExecCompleted ExecStatus = \"completed\"\n ExecFailed ExecStatus = \"failed\"\n ExecCancelled ExecStatus = \"cancelled\"\n)\n\n// RobotStatus - matches __yao.member.robot_status enum\ntype RobotStatus string\n\nconst (\n RobotIdle RobotStatus = \"idle\" // ready to run\n RobotWorking RobotStatus = \"working\" // currently executing\n RobotPaused RobotStatus = \"paused\" // manually paused\n RobotError RobotStatus = \"error\" // encountered error\n RobotMaintenance RobotStatus = \"maintenance\" // under maintenance\n)\n\n// Triggers - all on by default\ntype Triggers struct {\n Clock *Trigger `json:\"clock,omitempty\"`\n Intervene *Trigger `json:\"intervene,omitempty\"`\n Event *Trigger `json:\"event,omitempty\"`\n}\n\ntype Trigger struct {\n Enabled bool `json:\"enabled\"`\n Actions []string `json:\"actions,omitempty\"` // for intervene\n}\n\n// Clock - when to wake up\ntype Clock struct {\n Mode ClockMode `json:\"mode\"`\n Times []string `json:\"times\"` // for times: [\"09:00\", \"14:00\"]\n Days []string `json:\"days\"` // [\"Mon\", \"Tue\"...] or [\"*\"]\n Every string `json:\"every\"` // for interval: \"30m\", \"1h\"\n TZ string `json:\"tz\"` // Asia/Shanghai\n Timeout string `json:\"timeout\"` // max run time\n}\n\n// Identity\ntype Identity struct {\n Role string `json:\"role\"`\n Duties []string `json:\"duties\"`\n Rules []string `json:\"rules\"`\n}\n\n// Quota\ntype Quota struct {\n Max int `json:\"max\"` // max running (default: 2)\n Queue int `json:\"queue\"` // queue size (default: 10)\n Priority int `json:\"priority\"` // 1-10 (default: 5)\n}\n\n// KB\n// KB - shared knowledge base (same as assistant)\ntype KB struct {\n Collections []string `json:\"collections,omitempty\"` // KB collection IDs\n Options map[string]interface{} `json:\"options,omitempty\"`\n}\n\n// DB - shared database (same as assistant)\ntype DB struct {\n Models []string `json:\"models,omitempty\"` // database model names\n Options map[string]interface{} `json:\"options,omitempty\"`\n}\n\n// Learn - learning config for robot's private KB\n// Private KB auto-created: robot_{team_id}_{member_id}_kb\ntype Learn struct {\n On bool `json:\"on\"`\n Types []string `json:\"types\"` // execution, feedback, insight\n Keep int `json:\"keep\"` // days, 0 = forever\n}\n\n// Resources\ntype Resources struct {\n Phases map[Phase]string `json:\"phases,omitempty\"` // optional, defaults to __yao.{phase}\n Agents []string `json:\"agents\"`\n MCP []MCP `json:\"mcp\"`\n}\n\ntype MCP struct {\n ID string `json:\"id\"`\n Tools []string `json:\"tools,omitempty\"` // empty = all\n}\n\n// DeliveryPreferences - Robot delivery preferences (read by Delivery Center)\n// Each channel supports multiple targets\ntype DeliveryPreferences struct {\n Email *EmailPreference `json:\"email,omitempty\"`\n Webhook *WebhookPreference `json:\"webhook,omitempty\"`\n Process *ProcessPreference `json:\"process,omitempty\"`\n // notify is handled automatically based on user subscriptions\n}\n\ntype EmailPreference struct {\n Enabled bool `json:\"enabled\"`\n Targets []EmailTarget `json:\"targets\"`\n}\n\ntype EmailTarget struct {\n To []string `json:\"to\"` // Recipient addresses\n Template string `json:\"template,omitempty\"` // Email template ID\n Subject string `json:\"subject,omitempty\"` // Subject template\n}\n\ntype WebhookPreference struct {\n Enabled bool `json:\"enabled\"`\n Targets []WebhookTarget `json:\"targets\"`\n}\n\ntype WebhookTarget struct {\n URL string `json:\"url\"` // Webhook URL\n Method string `json:\"method,omitempty\"` // HTTP method (default: POST)\n Headers map[string]string `json:\"headers,omitempty\"` // Custom headers\n Secret string `json:\"secret,omitempty\"` // Signing secret\n}\n\ntype ProcessPreference struct {\n Enabled bool `json:\"enabled\"`\n Targets []ProcessTarget `json:\"targets\"`\n}\n\ntype ProcessTarget struct {\n Process string `json:\"process\"` // Yao Process name, e.g., \"orders.UpdateStatus\"\n Args []any `json:\"args,omitempty\"` // Additional arguments\n}\n\n// ExecutorMode - executor mode enum\ntype ExecutorMode string\n\nconst (\n ExecutorStandard ExecutorMode = \"standard\" // real Agent calls (default)\n ExecutorDryRun ExecutorMode = \"dryrun\" // simulated, no LLM calls\n ExecutorSandbox ExecutorMode = \"sandbox\" // container-isolated (NOT IMPLEMENTED)\n)\n\n// Executor - executor settings\ntype Executor struct {\n Mode ExecutorMode `json:\"mode,omitempty\"` // standard | dryrun | sandbox\n MaxDuration string `json:\"max_duration,omitempty\"` // max execution time (e.g., \"30m\")\n}\n// Note: Sandbox mode requires container infrastructure (Docker/gVisor).\n// Current implementation falls back to DryRun behavior.\n\n// Monitor\n```\n\n### 5.3 Example\n\nExample record in `__yao.member` table:\n\n```json\n{\n \"member_id\": \"mem_abc123\",\n \"team_id\": \"team_xyz\",\n \"member_type\": \"robot\",\n \"display_name\": \"Sales Bot\",\n \"autonomous_mode\": true,\n \"robot_status\": \"idle\",\n \"system_prompt\": \"You are a sales analyst...\",\n \"robot_config\": {\n \"triggers\": {\n \"clock\": { \"enabled\": true },\n \"intervene\": { \"enabled\": true },\n \"event\": { \"enabled\": false }\n },\n \"clock\": {\n \"mode\": \"times\",\n \"times\": [\"09:00\", \"14:00\", \"17:00\"],\n \"days\": [\"Mon\", \"Tue\", \"Wed\", \"Thu\", \"Fri\"],\n \"tz\": \"Asia/Shanghai\",\n \"timeout\": \"30m\"\n },\n \"identity\": {\n \"role\": \"Sales Analyst\",\n \"duties\": [\"Analyze sales\", \"Make weekly reports\"],\n \"rules\": [\"Only access sales data\"]\n },\n \"quota\": { \"max\": 2, \"queue\": 10, \"priority\": 5 },\n \"kb\": { \"collections\": [\"sales-policies\", \"products\"] },\n \"db\": { \"models\": [\"sales\", \"customers\"] },\n \"learn\": {\n \"on\": true,\n \"types\": [\"execution\", \"feedback\", \"insight\"],\n \"keep\": 90\n },\n \"resources\": {\n \"phases\": {\n \"inspiration\": \"__yao.inspiration\",\n \"goals\": \"__yao.goals\",\n \"tasks\": \"__yao.tasks\",\n \"validation\": \"__yao.validation\",\n \"delivery\": \"__yao.delivery\",\n \"learning\": \"__yao.learning\"\n },\n \"agents\": [\"data-analyst\", \"chart-gen\"],\n \"mcp\": [{ \"id\": \"database\", \"tools\": [\"query\"] }]\n },\n \"delivery\": {\n \"type\": \"email\",\n \"opts\": { \"to\": [\"manager@company.com\"] }\n },\n \"executor\": {\n \"mode\": \"standard\",\n \"max_duration\": \"30m\"\n }\n },\n \"agents\": [\"data-analyst\", \"chart-gen\"],\n \"mcp_servers\": [\"database\"]\n}\n```\n\n---\n\n## 6. Lifecycle\n\n### 6.1 Agent States\n\n```mermaid\nstateDiagram-v2\n [*] --> Idle: POST create\n Idle --> Working: trigger\n Working --> Idle: done\n Idle --> Paused: PATCH pause\n Working --> Paused: PATCH pause\n Paused --> Idle: PATCH resume\n Idle --> Error: error\n Working --> Error: error\n Error --> Idle: PATCH reset\n Idle --> [*]: DELETE\n Paused --> [*]: DELETE\n```\n\n| From | To | How |\n| ------- | ------- | --------------------------- |\n| - | idle | POST create |\n| idle | working | trigger (clock/human/event) |\n| working | idle | execution done |\n| idle | paused | PATCH robot_status=\"paused\" |\n| paused | idle | PATCH robot_status=\"idle\" |\n| any | error | execution error |\n| error | idle | PATCH robot_status=\"idle\" |\n| any | deleted | DELETE |\n\n### 6.2 On Create\n\n1. Check config\n2. Generate member_id if missing\n3. Create KB: `robot_{team_id}_{member_id}_kb`\n4. Add to cache\n5. Set active\n\n### 6.3 On Delete\n\n1. Stop running executions\n2. Remove from cache\n3. Delete or archive KB\n5. Soft delete record\n\n### 6.4 Execution Flow\n\nSingle execution flow, depends on trigger type:\n\n```mermaid\nflowchart LR\n subgraph Trigger\n T{Trigger}\n end\n\n subgraph Schedule Path\n P0[P0: Inspiration]\n end\n\n subgraph Common Path\n P1[P1: Goals]\n P2[P2: Tasks]\n P3[P3: Run]\n P4[P4: Deliver]\n P5[P5: Learn]\n end\n\n T -->|Clock| P0\n T -->|Human/Event| P1\n P0 --> P1\n P1 --> P2 --> P3 --> P4 --> P5\n```\n\n```mermaid\nstateDiagram-v2\n [*] --> Triggered\n Triggered --> P0_Inspiration: Clock\n Triggered --> P1_Goals: Human/Event\n P0_Inspiration --> P1_Goals\n P1_Goals --> P2_Tasks\n P2_Tasks --> P3_Run\n P3_Run --> P4_Deliver\n P4_Deliver --> P5_Learn\n P5_Learn --> [*]\n```\n\n---\n\n## 7. Integrations\n\n### 7.1 Execution Storage\n\n**Relationship:** 1 Robot : N Executions (concurrent)\n\nEach trigger creates a new Execution, stored in `ExecutionStore` (`__yao.agent_execution` table).\n\nExecution data includes:\n- Status and phase tracking\n- All phase outputs (Inspiration, Goals, Tasks, Results, Delivery, Learning)\n- Error information\n- Timestamps and progress\n\nLogging is handled by `kun/log` package for standard application logging.\n| List Execs | `job.ListExecutions(param, page, pagesize)` |\n| Get Exec | `job.GetExecution(execID, param)` |\n| Save Exec | `job.SaveExecution(exec)` |\n| List Logs | `job.ListLogs(param, page, pagesize)` |\n| Save Log | `job.SaveLog(log)` |\n| Push (start) | `j.Push()` |\n| Stop | `j.Stop()` |\n| Destroy | `j.Destroy()` |\n| Active Jobs | `job.GetActiveJobs()` |\n| Query by Cat | `job.ListJobs({Wheres: [{Column: \"category_id\", ...}]})` |\n\n### 7.2 Private KB\n\nMade on robot member create: `robot_{team_id}_{member_id}_kb`\n\n**What it stores:**\n\n- `execution`: What worked, what failed\n- `feedback`: Errors, fixes\n- `insight`: Patterns, tips\n\n**When:**\n\n- Create: On robot member create\n- Update: After P5\n- Clean: Based on `keep` days\n- Delete: On robot member delete\n\n### 7.3 External Input\n\n**Types:**\n\n- `clock`: Timer (with time context)\n- `intervene`: Human action\n- `event`: Webhook, DB change\n- `callback`: Async result\n\n**Human actions (InterventionAction):**\n\n- `task.add`: Add a new task\n- `task.cancel`: Cancel a task\n- `task.update`: Update task details\n- `goal.adjust`: Modify current goal\n- `goal.add`: Add a new goal\n- `goal.complete`: Mark goal as complete\n- `goal.cancel`: Cancel a goal\n- `plan.add`: Schedule for later\n- `plan.remove`: Remove from plan queue\n- `plan.update`: Update planned item\n- `instruct`: Direct instruction to robot\n\n**Plan Queue:**\n\n- Holds tasks for later\n- Runs at next cycle start\n\n---\n\n## 8. API\n\n### 8.1 Manager (Internal)\n\n> **Note:** Manager is the central orchestrator, handling all trigger types.\n\n```go\ntype Manager interface {\n // Lifecycle\n Start() error\n Stop() error\n\n // Clock trigger (internal, called by ticker)\n Tick(ctx *Context, now time.Time) error\n\n // Manual trigger (for testing/API)\n TriggerManual(ctx *Context, memberID string, trigger TriggerType, data interface{}) (string, error)\n\n // Human intervention (called by API)\n Intervene(ctx *Context, req *InterveneRequest) (*ExecutionResult, error)\n\n // Event trigger (called by webhook/db trigger)\n HandleEvent(ctx *Context, req *EventRequest) (*ExecutionResult, error)\n\n // Execution control\n PauseExecution(ctx *Context, execID string) error\n ResumeExecution(ctx *Context, execID string) error\n StopExecution(ctx *Context, execID string) error\n\n // Cache access\n Cache() Cache\n}\n```\n\n### 8.2 Trigger (Integrated into Manager)\n\n> **Note:** Trigger logic is integrated into Manager, not a separate interface.\n> The `trigger/` package provides utilities (validation, clock matching, execution control).\n\n```go\n// TriggerType enum\ntype TriggerType string\n\nconst (\n TriggerClock TriggerType = \"clock\"\n TriggerHuman TriggerType = \"human\"\n TriggerEvent TriggerType = \"event\"\n)\n\n// Manager handles all trigger types:\n// - Clock: Manager.Tick() called by internal ticker\n// - Human: Manager.Intervene() called by API\n// - Event: Manager.HandleEvent() called by webhook/db trigger\n\n// trigger/ package provides utilities:\n// - trigger.ValidateIntervention(req) - validate human intervention request\n// - trigger.ValidateEvent(req) - validate event request\n// - trigger.BuildEventInput(req) - build TriggerInput from event\n// - trigger.ClockMatcher - reusable clock matching logic\n// - trigger.ExecutionController - pause/resume/stop execution\n\ntype InterveneRequest struct {\n TeamID string\n MemberID string\n Action InterventionAction // task.add | goal.adjust | task.cancel | plan.add | instruct\n Messages []context.Message // user input (text, images, files)\n PlanTime *time.Time // for action=plan.add\n ExecutorMode ExecutorMode // optional: standard | dryrun (override robot config)\n}\n\ntype EventRequest struct {\n MemberID string\n Source string // webhook path or table name\n EventType string // lead.created, etc.\n Data map[string]interface{}\n ExecutorMode ExecutorMode // optional: standard | dryrun (override robot config)\n}\n\ntype ExecutionResult struct {\n ExecutionID string // Job execution ID\n Status ExecStatus // pending | running | completed | failed\n Message string // status message\n}\n\ntype RobotState struct {\n MemberID string // member_id from __yao.member\n Status RobotStatus // idle | working | paused | error | maintenance\n LastRun time.Time\n NextRun time.Time\n Running int // current running execution count\n MaxRunning int // max concurrent executions (from Quota.Max)\n RunningIDs []string // list of running execution IDs\n}\n```\n\n### 8.3 Execution (Uses ExecutionStore)\n\nUses dedicated `__yao.agent_execution` table via ExecutionStore.\n\n**Each trigger creates a new Execution:**\n\n```go\n// On each trigger (clock/human/event), create a new Execution\nexec := &types.Execution{\n ID: utils.NewID(),\n MemberID: memberID,\n TeamID: teamID,\n TriggerType: triggerType,\n Status: types.ExecStatusRunning,\n Phase: types.PhaseP0Init,\n StartedAt: time.Now(),\n}\n\n// Save to ExecutionStore\nexecStore.Save(exec)\n```\n\n**Query executions for a robot:**\n\n```go\n// List all executions for a robot member\nexecutions, err := execStore.List(memberID, 1, 10)\n```\n\n**Query examples:**\n\n```go\n// Get execution by ID\nexec, err := execStore.Get(executionID)\n\n// List executions for a robot\nexecutions, err := execStore.List(memberID, page, pageSize)\n\n// Update execution status\nexecStore.UpdateStatus(executionID, types.ExecStatusCompleted)\n\n// Logging via kun/log\nlog.With(log.F{\"execution_id\": exec.ID, \"phase\": \"P1\"}).Info(\"Phase started\")\n```\n\n---\n\n## 9. Security\n\n1. **Team only**: Agent sees only its team's data\n2. **Role rules**: Uses role_id permissions\n3. **Limited tools**: Only what's in `resources`\n4. **Timeout**: Stops if runs too long\n5. **Logs**: All runs saved\n\n---\n\n## 10. Quick Ref\n\n### Triggers\n\n```yaml\ntriggers:\n clock: { enabled: true }\n intervene: { enabled: true, actions: [...] }\n event: { enabled: false }\n```\n\n### Clock\n\n```yaml\n# Mode 1: Specific times\nclock:\n mode: times\n times: [\"09:00\", \"14:00\", \"17:00\"]\n days: [\"Mon\", \"Tue\", \"Wed\", \"Thu\", \"Fri\"]\n tz: Asia/Shanghai\n timeout: 30m\n\n# Mode 2: Interval\nclock:\n mode: interval\n every: 30m # run every 30 minutes\n timeout: 10m\n\n# Mode 3: Daemon (continuous thinking/analysis)\nclock:\n mode: daemon # restart immediately after each run\n timeout: 10m # max time per run\n # Use case: Research analyst, market monitor\n```\n\n### Phase Agents\n\n```yaml\n# Optional - defaults to __yao.{phase} if not specified\nresources:\n phases:\n inspiration: \"__yao.inspiration\" # Clock only\n goals: \"__yao.goals\"\n tasks: \"__yao.tasks\"\n validation: \"__yao.validation\"\n delivery: \"__yao.delivery\"\n learning: \"__yao.learning\"\n```\n\n### Quota\n\n```yaml\nquota:\n max: 2 # max running\n queue: 10 # queue size\n priority: 5 # 1-10\n```\n\n### Executor\n\n```yaml\n# Standard mode (default) - real Agent calls\nexecutor:\n mode: standard\n max_duration: 30m\n\n# DryRun mode - simulated execution (for testing/demos)\nexecutor:\n mode: dryrun\n\n# Sandbox mode (NOT IMPLEMENTED) - container-isolated\n# Requires Docker/gVisor infrastructure\n# executor:\n# mode: sandbox\n# max_duration: 10m\n```\n\n**API Override:**\n\n```javascript\n// Override executor mode per trigger\nconst result = Process(\"robot.Trigger\", \"mem_abc123\", {\n type: \"human\",\n action: \"task.add\",\n messages: [{ role: \"user\", content: \"Test task\" }],\n executor_mode: \"dryrun\", // override robot config\n});\n```\n\n---\n\n## 11. Examples\n\nEach example shows a different trigger mode:\n\n| Example | Trigger | Mode | Scenario |\n| ------- | ------- | --------- | -------------------------------------------- |\n| 11.1 | Clock | times | SEO/GEO Content - daily content optimization |\n| 11.2 | Clock | interval | Competitor Monitor - check every 2 hours |\n| 11.3 | Clock | daemon | Research Analyst - continuous insight mining |\n| 11.4 | Human | intervene | Sales Assistant - manager assigns tasks |\n| 11.5 | Event | webhook | Lead Processor - qualify and route new leads |\n\n---\n\n### 11.1 SEO/GEO Content Agent (Clock: times)\n\n**Trigger:** Clock - specific times daily\n\n**Role:** AI Marketing - auto-generate and optimize SEO/GEO content.\n\n```json\n// robot_config for SEO Content Agent\n{\n \"triggers\": {\n \"clock\": { \"enabled\": true },\n \"intervene\": { \"enabled\": true }\n },\n \"clock\": {\n \"mode\": \"times\",\n \"times\": [\"06:00\", \"18:00\"],\n \"days\": [\"Mon\", \"Tue\", \"Wed\", \"Thu\", \"Fri\"],\n \"tz\": \"Asia/Shanghai\"\n },\n \"identity\": {\n \"role\": \"SEO/GEO Content Specialist\",\n \"duties\": [\n \"Research trending keywords in our industry\",\n \"Generate SEO-optimized articles (2-3 per day)\",\n \"Optimize existing content for GEO (AI search)\",\n \"Track keyword rankings and adjust strategy\",\n \"A/B test titles and meta descriptions\"\n ]\n },\n \"resources\": {\n \"agents\": [\"keyword-researcher\", \"content-writer\", \"seo-optimizer\"],\n \"mcp\": [\n { \"id\": \"google-search\", \"tools\": [\"trends\", \"rankings\"] },\n { \"id\": \"cms\", \"tools\": [\"create\", \"update\", \"publish\"] }\n ]\n },\n \"delivery\": {\n \"type\": \"notify\",\n \"opts\": { \"channel\": \"marketing-team\" }\n }\n}\n```\n\n**Example run at 06:00 Monday:**\n\n```\nP0 Inspiration:\n Clock: Monday 06:00, start of week\n Data:\n - Keyword \"AI app development\" trending (+45% this week)\n - Our article ranks #8, competitor #2\n - 3 articles need GEO optimization\n World: New AI regulation announced last Friday\n\nP1 Goals:\n 1. Write new article targeting \"AI app development\"\n 2. Optimize 3 old articles for GEO\n 3. Update meta descriptions for top 5 pages\n\nP2 Tasks:\n 1. Research \"AI app development\" keywords → keyword-researcher\n 2. Write article with SEO structure → content-writer\n 3. Add FAQ schema for GEO → seo-optimizer\n 4. Publish to CMS → cms.publish\n\nP3 Execute:\n - Keywords: \"AI app development\", \"build AI apps\", \"AI dev guide\" (12 total)\n - Article: 2500 words, 8 sections, FAQ schema added\n - Published to CMS, indexed by Google\n\nP4 Delivery:\n → Notify: \"Published: 'Complete Guide to AI App Development' - targeting 12 keywords\"\n\nP5 Learn:\n - \"AI app development\" articles perform well on Monday morning\n - FAQ schema improves GEO visibility by 30%\n```\n\n---\n\n### 11.2 Competitor Monitor (Clock: interval)\n\n**Trigger:** Clock - every 2 hours\n\n**Role:** Monitor competitors, track market changes, alert on important updates.\n\n```json\n// robot_config for Competitor Monitor\n{\n \"triggers\": {\n \"clock\": { \"enabled\": true }\n },\n \"clock\": {\n \"mode\": \"interval\",\n \"every\": \"2h\"\n },\n \"identity\": {\n \"role\": \"Competitor Intelligence Analyst\",\n \"duties\": [\n \"Monitor competitor websites for changes\",\n \"Track competitor pricing updates\",\n \"Watch for new product launches\",\n \"Analyze competitor content strategy\",\n \"Alert team on significant changes\"\n ]\n },\n \"resources\": {\n \"agents\": [\"web-scraper\", \"diff-analyzer\", \"report-writer\"],\n \"mcp\": [{ \"id\": \"web-search\", \"tools\": [\"search\", \"news\"] }]\n },\n \"delivery\": {\n \"type\": \"webhook\",\n \"opts\": { \"url\": \"https://slack.com/webhook/competitor-alerts\" }\n }\n}\n```\n\n**Example run detecting competitor change:**\n\n```\nP0 Inspiration:\n Clock: Tuesday 14:00\n Data:\n - Competitor A: pricing page changed\n - Competitor B: new blog post about \"AI agents\"\n - Competitor C: no changes\n\nP1 Goals:\n 1. Analyze Competitor A pricing change\n 2. Summarize Competitor B's new content\n 3. Assess impact on our positioning\n\nP2 Tasks:\n 1. Scrape old vs new pricing → web-scraper\n 2. Compare pricing tiers → diff-analyzer\n 3. Generate competitive analysis → report-writer\n\nP3 Execute:\n - Competitor A: dropped price 20% on enterprise tier\n - Competitor B: targeting same keywords as us\n\nP4 Delivery:\n → Slack: \"🚨 Competitor A cut enterprise price 20% - review needed\"\n\nP5 Learn:\n - Competitor A tends to change pricing on Tuesdays\n - Price changes often precede feature launches\n```\n\n---\n\n### 11.3 Industry Research Analyst (Clock: daemon)\n\n**Trigger:** Clock - continuous daemon mode\n\n**Role:** Continuously read industry news, papers, social media; extract insights; build knowledge.\n\n```json\n// robot_config for Research Analyst\n{\n \"triggers\": {\n \"clock\": { \"enabled\": true }\n },\n \"clock\": {\n \"mode\": \"daemon\",\n \"timeout\": \"10m\"\n },\n \"identity\": {\n \"role\": \"Industry Research Analyst\",\n \"duties\": [\n \"Continuously scan industry news and papers\",\n \"Analyze trends and extract key insights\",\n \"Identify emerging technologies and competitors\",\n \"Build and maintain industry knowledge base\",\n \"Alert team on significant developments\"\n ]\n },\n \"resources\": {\n \"agents\": [\"content-reader\", \"insight-extractor\", \"report-writer\"],\n \"mcp\": [\n { \"id\": \"web-search\", \"tools\": [\"search\", \"news\"] },\n { \"id\": \"arxiv\", \"tools\": [\"search\", \"fetch\"] },\n { \"id\": \"twitter\", \"tools\": [\"search\", \"trends\"] }\n ]\n },\n \"delivery\": {\n \"type\": \"notify\",\n \"opts\": { \"channel\": \"research-insights\" }\n }\n}\n```\n\n**Example continuous run:**\n\n```\nRun #1 (09:00):\n P0: Scan sources\n - 15 new AI news articles\n - 3 new papers on arXiv\n - Twitter: \"AI Agent\" trending\n P1: Goals:\n 1. Read and analyze new content\n 2. Extract insights relevant to our business\n 3. Update knowledge base\n P2: Tasks:\n 1. Read articles → content-reader\n 2. Analyze papers → content-reader\n 3. Extract insights → insight-extractor\n P3: Execute:\n - Article: \"OpenAI releases new agent framework\"\n Insight: Validates our direction, watch for API changes\n - Paper: \"Multi-agent collaboration patterns\"\n Insight: Useful for our agent design, save to KB\n - Twitter: Sentiment positive on AI agents\n P4: Notify: \"📚 3 new insights added to KB\"\n P5: Learn: OpenAI news = high relevance, prioritize\n → Restart immediately\n\nRun #2 (09:12):\n P0: Scan sources\n - 2 new articles (low relevance)\n - No new papers\n - Twitter: Normal activity\n P1: Low-value content, skip deep analysis\n P5: Learn: Mid-morning usually quiet\n → Restart immediately\n\nRun #3 (09:25):\n P0: Scan sources\n - Breaking: \"Competitor X raises $100M for AI platform\"\n P1: Goals:\n 1. Deep analyze competitor news\n 2. Assess impact on our market\n 3. Alert team immediately\n P2: Tasks:\n 1. Gather all competitor X info → web-search\n 2. Analyze their positioning → insight-extractor\n 3. Write competitive brief → report-writer\n P3: Execute:\n - Competitor X: Focus on enterprise, similar target market\n - Funding: Will likely expand sales team\n - Threat level: Medium-High\n P4: Notify: \"🚨 Competitor X raised $100M - brief attached\"\n P5: Learn: Funding news = always high priority\n → Restart immediately\n```\n\n---\n\n### 11.4 Sales Assistant (Human: intervene)\n\n**Trigger:** Human intervention - sales manager assigns tasks\n\n**Role:** Help sales team with research, proposals, follow-ups when manager assigns work.\n\n```json\n// robot_config for Sales Assistant\n{\n \"triggers\": {\n \"clock\": { \"enabled\": false },\n \"intervene\": {\n \"enabled\": true,\n \"actions\": [\"task.add\", \"goal.adjust\", \"instruct\"]\n }\n },\n \"identity\": {\n \"role\": \"Sales Assistant\",\n \"duties\": [\n \"Research assigned prospects and companies\",\n \"Prepare customized proposals and presentations\",\n \"Draft follow-up emails\",\n \"Analyze deal history and suggest strategies\",\n \"Prepare meeting briefs\"\n ]\n },\n \"resources\": {\n \"agents\": [\"company-researcher\", \"proposal-writer\", \"email-drafter\"],\n \"mcp\": [\n { \"id\": \"crm\", \"tools\": [\"query\", \"update\"] },\n { \"id\": \"linkedin\", \"tools\": [\"search\", \"profile\"] },\n { \"id\": \"email\", \"tools\": [\"draft\", \"send\"] }\n ]\n },\n \"delivery\": {\n \"type\": \"email\",\n \"opts\": { \"to\": [\"sales-manager@company.com\"] }\n }\n}\n```\n\n**Example: Sales manager assigns task:**\n\n```\nSales Manager Input:\n Action: task.add\n Messages: [{ role: \"user\", content: \"Meeting with BigCorp CTO tomorrow. Prepare materials.\n They do smart manufacturing, $150M revenue, digital transformation.\" }]\n\nAgent Execution (no P0 for human trigger):\n P1 Goals (from human input):\n 1. Research BigCorp and their CTO\n 2. Prepare meeting brief\n 3. Draft customized proposal\n\n P2 Tasks:\n 1. Research BigCorp → company-researcher\n - Company background, recent news\n - Digital transformation status\n - Potential pain points\n 2. Research CTO profile → linkedin.profile\n - Background, interests\n - Recent posts/articles\n 3. Prepare meeting brief → proposal-writer\n 4. Draft proposal → proposal-writer\n\n P3 Execute:\n - BigCorp: Leading smart manufacturing, 3 factories, implementing MES\n - CTO John: Ex-Google, focused on AI+Manufacturing, recent post on \"AI QC\"\n - Pain point: High QC labor cost, 2% defect miss rate\n - Opportunity: Our AI QC solution can reduce miss rate to 0.1%\n\n P4 Delivery:\n → Email to sales manager:\n - Attachment 1: BigCorp Research Report (PDF)\n - Attachment 2: CTO Profile Brief\n - Attachment 3: Custom Proposal - AI QC Solution\n - Attachment 4: Meeting Agenda Suggestion\n\nSales Manager Follow-up:\n Action: task.add\n Messages: [{ role: \"user\", content: \"Also prepare some similar case studies, manufacturing preferred\" }]\n\nAgent Continues:\n P1: Find similar manufacturing case studies\n P2: Search CRM for manufacturing wins\n P3: Found 3 cases: Auto parts factory, Electronics plant, Food processing\n P4: Email: \"3 manufacturing case studies attached\"\n P5: Learn: Manufacturing prospects often need QC case studies\n```\n\n---\n\n### 11.5 Lead Processor (Event: webhook)\n\n**Trigger:** Event - new lead from website/CRM\n\n**Role:** Instantly process and qualify new leads, route to sales.\n\n```json\n// robot_config for Lead Processor\n{\n \"triggers\": {\n \"clock\": { \"enabled\": false },\n \"event\": { \"enabled\": true }\n },\n \"events\": [\n {\n \"type\": \"webhook\",\n \"source\": \"/webhook/leads\",\n \"filter\": { \"event_types\": [\"lead.created\"] }\n },\n {\n \"type\": \"database\",\n \"source\": \"crm_leads\",\n \"filter\": { \"trigger\": \"insert\" }\n }\n ],\n \"identity\": {\n \"role\": \"Lead Qualification Specialist\",\n \"duties\": [\n \"Instantly process new leads\",\n \"Enrich lead data (company info, LinkedIn)\",\n \"Score lead quality (1-100)\",\n \"Route hot leads to sales immediately\",\n \"Add cold leads to nurture sequence\"\n ]\n },\n \"resources\": {\n \"agents\": [\"data-enricher\", \"lead-scorer\"],\n \"mcp\": [\n { \"id\": \"clearbit\", \"tools\": [\"enrich\"] },\n { \"id\": \"crm\", \"tools\": [\"update\", \"assign\"] },\n { \"id\": \"email\", \"tools\": [\"send\"] }\n ]\n },\n \"delivery\": {\n \"type\": \"webhook\",\n \"opts\": { \"url\": \"https://slack.com/webhook/sales-leads\" }\n }\n}\n```\n\n**Example: New lead event:**\n\n```\nEvent Received:\n Type: lead.created\n Data: {\n name: \"John Smith\",\n email: \"john@bigcorp.com\",\n company: \"BigCorp\",\n message: \"Interested in Enterprise pricing, team of 50\"\n }\n\nAgent Execution (no P0 for events):\n P1 Goals:\n 1. Enrich lead data\n 2. Score lead quality\n 3. Route appropriately\n\n P2 Tasks:\n 1. Lookup company info → clearbit.enrich\n 2. Calculate lead score → lead-scorer\n 3. Update CRM → crm.update\n 4. Notify sales → slack webhook\n\n P3 Execute:\n - Company: BigCorp, 500 employees, Series C\n - LinkedIn: VP of Engineering\n - Lead Score: 85/100 (HOT)\n - Reason: Enterprise inquiry, decision maker, funded company\n\n P4 Delivery:\n → Slack: \"🔥 HOT LEAD (85/100): John Smith @ BigCorp\n - 500 employees, Series C\n - Interested in Enterprise (50 seats)\n - Assigned to: Sales Rep A\"\n → CRM: Lead updated, assigned to Sales Rep A\n → Email to lead: \"Thanks for your inquiry. Our sales rep will contact you within 1 hour.\"\n\n P5 Learn:\n - BigCorp profile saved for future reference\n - VP-level leads from funded companies = high conversion\n```\n" }, { "path": "agent/robot/TECHNICAL.md", "content": "# Robot Agent - Technical Design\n\n## 1. Code Structure\n\n```\nyao/agent/robot/\n├── DESIGN.md # Product design doc\n├── TECHNICAL.md # This file\n│\n├── robot.go # Package entry, Init(), Shutdown()\n│\n├── api/ # All API forms\n│ ├── api.go # Go API (facade)\n│ ├── process.go # Yao Process: robot.*\n│ └── jsapi.go # JS API: robot (global) + Robot (class)\n│\n├── types/ # Types only (no logic, no external deps)\n│ ├── enums.go # Phase, ClockMode, TriggerType, etc.\n│ ├── config.go # Config, Clock, Identity, Quota, etc.\n│ ├── robot.go # Robot, Execution\n│ ├── task.go # Goal, Task, TaskResult\n│ ├── request.go # InterveneRequest, EventRequest, etc.\n│ ├── inspiration.go # ClockContext, InspirationReport\n│ ├── interfaces.go # All interfaces (Manager, Trigger, etc.)\n│ └── errors.go # Error definitions\n│\n├── manager/ # Manager package (orchestration)\n│ └── manager.go # Manager struct, Start/Stop, Tick\n│\n├── pool/ # Worker pool & task dispatch\n│ ├── pool.go # Pool struct, Submit\n│ ├── queue.go # Priority queue\n│ └── worker.go # Worker goroutines\n│\n├── executor/ # Executor package (pluggable architecture)\n│ ├── executor.go # Factory functions, unified entry\n│ ├── types/\n│ │ ├── types.go # Executor interface, Config types\n│ │ └── helpers.go # Shared helper functions\n│ ├── standard/\n│ │ ├── executor.go # Real Agent execution (production)\n│ │ ├── agent.go # AgentCaller for LLM calls\n│ │ ├── input.go # InputFormatter for prompts\n│ │ ├── inspiration.go # P0: Inspiration phase\n│ │ ├── goals.go # P1: Goals phase\n│ │ ├── tasks.go # P2: Tasks phase\n│ │ ├── run.go # P3: Run phase (main entry)\n│ │ ├── runner.go # P3: Task Runner (execution logic)\n│ │ ├── validator.go # P3: Validator (two-layer validation)\n│ │ ├── delivery.go # P4: Delivery phase\n│ │ └── learning.go # P5: Learning phase\n│ ├── dryrun/\n│ │ └── executor.go # Simulated execution (testing/demo)\n│ └── sandbox/\n│ └── executor.go # Container-isolated (NOT IMPLEMENTED)\n│\n├── utils/ # Utility functions\n│ ├── convert.go # Type conversions (JSON, map, struct)\n│ ├── time.go # Time parsing, formatting, timezone\n│ ├── id.go # ID generation (nanoid, uuid)\n│ └── validate.go # Validation helpers\n│\n├── trigger/ # Trigger utilities (logic in manager/)\n│ ├── trigger.go # Validation helpers, action utilities\n│ ├── clock.go # ClockMatcher (reusable clock matching logic)\n│ └── control.go # ExecutionController (pause/resume/stop)\n│\n├── cache/ # Cache package\n│ ├── cache.go # Cache struct, Get/List\n│ ├── load.go # LoadAll, LoadOne\n│ └── refresh.go # Refresh logic\n│\n├── dedup/ # Deduplication package\n│ ├── dedup.go # Dedup struct\n│ ├── fast.go # Fast in-memory check\n│ └── semantic.go # Semantic check via agent\n│\n├── store/ # Data store package (KB, FS, DB access)\n│ ├── store.go # Store struct, interface\n│ ├── kb.go # Knowledge base operations\n│ ├── fs.go # File system operations\n│ ├── db.go # Database queries\n│ └── learning.go # Learning entry save (to KB)\n│\n└── plan/ # Plan queue (deferred tasks)\n ├── plan.go # Plan queue struct\n └── schedule.go # Schedule for later\n\nyao/assert/ # Universal assertion library (global package)\n├── types.go # Assertion, Result, interfaces\n├── asserter.go # Asserter implementation (8 assertion types)\n└── helpers.go # Utility functions (ExtractPath, ToString, etc.)\n```\n\n### Dependency Graph (No Cycles)\n\n> **Note:** `trigger/` is a utility package (validation, clock matching, execution control).\n> All trigger logic flows through `manager/`.\n\n```\n ┌──────────┐\n │ types/ │ (pure types, no deps)\n └────┬─────┘\n │\n ┌───────┬───────┬───────┬──────┼──────┬───────┬───────┬───────┐\n │ │ │ │ │ │ │ │ │\n ▼ ▼ ▼ ▼ ▼ ▼ ▼ ▼ ▼\n┌───────┐┌───────┐┌───────┐┌──────┐┌────┐┌──────┐┌───────┐┌─────────┐\n│ cache ││ dedup ││ store ││ pool ││ plan ││ utils ││ trigger │\n└───┬───┘└───┬───┘└───┬───┘└──┬───┘└──┬─┘└──────┘└───────┘└────┬────┘\n │ │ │ │ │ │\n └────────┴────────┴───────┴───────┴────────────────────────┘\n │\n ▼\n ┌────────────┐\n │ executor/ │\n └──────┬─────┘\n │\n ▼\n ┌────────────┐\n │ manager/ │ (imports trigger/ for utilities)\n └──────┬─────┘\n │\n ┌──────────────┴──────────────┐\n │ │\n ▼ ▼\n┌─────────────┐ ┌─────────────┐\n│ robot.go │ │ api/ │\n└─────────────┘ └─────────────┘\n```\n\n### Package Dependencies\n\n| Package | Imports |\n| ----------- | ----------------------------------------------------------- |\n| `types/` | stdlib only |\n| `utils/` | stdlib only |\n| `cache/` | `types/` |\n| `dedup/` | `types/` |\n| `store/` | `types/` |\n| `pool/` | `types/` |\n| `trigger/` | `types/` |\n| `plan/` | `types/` |\n| `executor/` | `types/`, `cache/`, `dedup/`, `store/`, `pool/`, `yao/assert` |\n| `manager/` | `types/`, `cache/`, `pool/`, `trigger/`, `executor/` |\n| | Manager handles all trigger logic (clock, intervene, event) |\n| `api/` | `types/`, `manager/` |\n| root | all packages |\n\n### Public API (`api/`)\n\nThree API forms, all in `api/` directory.\n\n#### Go API (`api/api.go`)\n\n```go\npackage api\n\nimport (\n \"github.com/yaoapp/yao/agent/robot/types\"\n)\n\n// ==================== CRUD ====================\n\n// Get returns a robot by member ID\nfunc Get(ctx *types.Context, memberID string) (*types.Robot, error)\n\n// List returns robots with pagination and filtering\nfunc List(ctx *types.Context, query *ListQuery) (*ListResult, error)\n\n// Create creates a new robot member\nfunc Create(ctx *types.Context, teamID string, req *CreateRequest) (*types.Robot, error)\n\n// Update updates robot config\nfunc Update(ctx *types.Context, memberID string, req *UpdateRequest) (*types.Robot, error)\n\n// Remove deletes a robot member\nfunc Remove(ctx *types.Context, memberID string) error\n\n// ==================== Status ====================\n\n// Status returns current robot runtime state\nfunc Status(ctx *types.Context, memberID string) (*RobotState, error)\n\n// UpdateStatus updates robot status (idle, paused, etc.)\nfunc UpdateStatus(ctx *types.Context, memberID string, status types.RobotStatus) error\n\n// ==================== Trigger ====================\n\n// Trigger starts execution with specified trigger type and request\nfunc Trigger(ctx *types.Context, memberID string, req *TriggerRequest) (*TriggerResult, error)\n\n// ==================== Execution ====================\n\n// GetExecutions returns execution history\nfunc GetExecutions(ctx *types.Context, memberID string, query *ExecutionQuery) (*ExecutionResult, error)\n\n// GetExecution returns a specific execution by ID\nfunc GetExecution(ctx *types.Context, execID string) (*types.Execution, error)\n\n// Pause pauses a running execution\nfunc Pause(ctx *types.Context, execID string) error\n\n// Resume resumes a paused execution\nfunc Resume(ctx *types.Context, execID string) error\n\n// Stop stops a running execution\nfunc Stop(ctx *types.Context, execID string) error\n\n```\n\n#### API Types\n\n```go\n// ==================== CRUD Types ====================\n\n// CreateRequest - request for Create()\ntype CreateRequest struct {\n DisplayName string `json:\"display_name\"`\n SystemPrompt string `json:\"system_prompt,omitempty\"`\n Config *types.Config `json:\"robot_config\"`\n}\n\n// UpdateRequest - request for Update()\ntype UpdateRequest struct {\n DisplayName *string `json:\"display_name,omitempty\"`\n SystemPrompt *string `json:\"system_prompt,omitempty\"`\n Config *types.Config `json:\"robot_config,omitempty\"`\n}\n\n// ListQuery - query options for List()\ntype ListQuery struct {\n TeamID string `json:\"team_id,omitempty\"` // filter by team\n Status types.RobotStatus `json:\"status,omitempty\"` // idle | working | paused | error\n Keywords string `json:\"keywords,omitempty\"` // search display_name, role\n ClockMode types.ClockMode `json:\"clock_mode,omitempty\"` // times | interval | daemon\n Page int `json:\"page,omitempty\"` // default 1\n PageSize int `json:\"pagesize,omitempty\"` // default 20, max 100\n Order string `json:\"order,omitempty\"` // e.g. \"created_at desc\"\n}\n\n// ListResult - result of List()\ntype ListResult struct {\n Data []*types.Robot `json:\"data\"`\n Total int `json:\"total\"`\n Page int `json:\"page\"`\n PageSize int `json:\"pagesize\"`\n}\n\n// RobotState - runtime state from Status()\ntype RobotState struct {\n MemberID string `json:\"member_id\"`\n TeamID string `json:\"team_id\"`\n DisplayName string `json:\"display_name\"`\n Status types.RobotStatus `json:\"status\"` // idle | working | paused | error\n Running int `json:\"running\"` // current running execution count\n MaxRunning int `json:\"max_running\"` // max concurrent allowed\n LastRun *time.Time `json:\"last_run,omitempty\"`\n NextRun *time.Time `json:\"next_run,omitempty\"`\n RunningIDs []string `json:\"running_ids,omitempty\"` // list of running execution IDs\n}\n\n// ==================== Trigger Types ====================\n\n// TriggerRequest - request for Trigger()\n// Input uses []context.Message to support rich content (text, images, files, audio)\ntype TriggerRequest struct {\n Type types.TriggerType `json:\"type\"` // human | event\n\n // Human intervention fields (when Type = human)\n Action types.InterventionAction `json:\"action,omitempty\"` // task.add | goal.adjust | task.cancel | plan.add\n Messages []context.Message `json:\"messages,omitempty\"` // user's input (supports text, images, files)\n PlanAt *time.Time `json:\"plan_at,omitempty\"` // for action=plan.add\n InsertAt InsertPosition `json:\"insert_at,omitempty\"` // where to insert: first | last | next | at\n AtIndex int `json:\"at_index,omitempty\"` // index when insert_at=at\n\n // Event fields (when Type = event)\n Source types.EventSource `json:\"source,omitempty\"` // webhook | database\n EventType string `json:\"event_type,omitempty\"` // lead.created, order.paid, etc.\n Data map[string]interface{} `json:\"data,omitempty\"` // event payload\n\n // Executor mode (optional, overrides robot config)\n ExecutorMode types.ExecutorMode `json:\"executor_mode,omitempty\"` // standard | dryrun\n}\n\n// InsertPosition - where to insert task in queue\ntype InsertPosition string\n\nconst (\n InsertFirst InsertPosition = \"first\" // insert at beginning (highest priority)\n InsertLast InsertPosition = \"last\" // append at end (default)\n InsertNext InsertPosition = \"next\" // insert after current task\n InsertAt InsertPosition = \"at\" // insert at specific index (use AtIndex)\n)\n\n// TriggerResult - result of Trigger()\ntype TriggerResult struct {\n Accepted bool `json:\"accepted\"` // whether trigger was accepted\n Queued bool `json:\"queued\"` // true if queued (quota full)\n Execution *types.Execution `json:\"execution,omitempty\"` // execution info if started\n Message string `json:\"message,omitempty\"` // status message\n}\n\n// ==================== Execution Types ====================\n\n// ExecutionQuery - query options for GetExecutions()\ntype ExecutionQuery struct {\n Status types.ExecStatus `json:\"status,omitempty\"` // pending | running | completed | failed\n Trigger types.TriggerType `json:\"trigger,omitempty\"` // clock | human | event\n Page int `json:\"page,omitempty\"` // default 1\n PageSize int `json:\"pagesize,omitempty\"`// default 20\n}\n\n// ExecutionResult - result of GetExecutions()\ntype ExecutionResult struct {\n Data []*types.Execution `json:\"data\"`\n Total int `json:\"total\"`\n Page int `json:\"page\"`\n PageSize int `json:\"pagesize\"`\n}\n```\n\n#### Process API (`api/process.go`)\n\nYao Process registration. Naming convention: `robot.`\n\n```go\n// Process registration\nfunc init() {\n process.Register(\"robot.Get\", processGet)\n process.Register(\"robot.List\", processList)\n process.Register(\"robot.Create\", processCreate)\n process.Register(\"robot.Update\", processUpdate)\n process.Register(\"robot.Remove\", processRemove)\n process.Register(\"robot.Status\", processStatus)\n process.Register(\"robot.UpdateStatus\", processUpdateStatus)\n process.Register(\"robot.Trigger\", processTrigger)\n process.Register(\"robot.Executions\", processExecutions)\n process.Register(\"robot.Execution\", processExecution)\n process.Register(\"robot.Pause\", processPause)\n process.Register(\"robot.Resume\", processResume)\n process.Register(\"robot.Stop\", processStop)\n}\n```\n\n| Process | Args | Returns | Description |\n| -------------------- | --------------------- | ----------------- | ------------------ |\n| `robot.Get` | `memberID` | `Robot` | Get robot by ID |\n| `robot.List` | `query` | `ListResult` | List robots |\n| `robot.Create` | `teamID`, `data` | `Robot` | Create robot |\n| `robot.Update` | `memberID`, `data` | `Robot` | Update robot |\n| `robot.Remove` | `memberID` | `null` | Delete robot |\n| `robot.Status` | `memberID` | `RobotState` | Get runtime status |\n| `robot.UpdateStatus` | `memberID`, `status` | `null` | Update status |\n| `robot.Trigger` | `memberID`, `request` | `TriggerResult` | Trigger execution |\n| `robot.Executions` | `memberID`, `query` | `ExecutionResult` | List executions |\n| `robot.Execution` | `execID` | `Execution` | Get execution |\n| `robot.Pause` | `execID` | `null` | Pause execution |\n| `robot.Resume` | `execID` | `null` | Resume execution |\n| `robot.Stop` | `execID` | `null` | Stop execution |\n\n**Usage:**\n\n```javascript\n// In Yao scripts\nconst robot = Process(\"robot.Get\", \"mem_abc123\");\n\nconst list = Process(\"robot.List\", {\n team_id: \"team_xyz\",\n status: \"idle\",\n page: 1,\n pagesize: 20,\n});\n\n// Trigger with text message\nconst result = Process(\"robot.Trigger\", \"mem_abc123\", {\n type: \"human\",\n action: \"task.add\",\n messages: [\n { role: \"user\", content: \"Prepare meeting materials for BigCorp\" },\n ],\n insert_at: \"first\",\n});\n\n// Trigger with image (multimodal)\nconst imageResult = Process(\"robot.Trigger\", \"mem_abc123\", {\n type: \"human\",\n action: \"task.add\",\n messages: [\n {\n role: \"user\",\n content: [\n { type: \"text\", text: \"Analyze this chart and summarize key trends\" },\n {\n type: \"image_url\",\n image_url: { url: \"https://example.com/chart.png\" },\n },\n ],\n },\n ],\n insert_at: \"first\",\n});\n\n// Trigger with event\nconst eventResult = Process(\"robot.Trigger\", \"mem_abc123\", {\n type: \"event\",\n source: \"webhook\",\n event_type: \"lead.created\",\n data: { name: \"John\", email: \"john@example.com\" },\n});\n\nconst execs = Process(\"robot.Executions\", \"mem_abc123\", {\n status: \"completed\",\n page: 1,\n});\n```\n\n#### JSAPI (`api/jsapi.go`)\n\nRegister to V8 Runtime using constructor pattern, similar to `new FS()`, `new Store()`, `new Query()`.\n\n```go\nfunc init() {\n // Register Robot constructor\n v8.RegisterFunction(\"Robot\", ExportFunction)\n}\n\n// ExportFunction exports the Robot constructor\nfunc ExportFunction(iso *v8go.Isolate) *v8go.FunctionTemplate {\n return v8go.NewFunctionTemplate(iso, robotConstructor)\n}\n\n// robotConstructor: new Robot(memberID)\nfunc robotConstructor(info *v8go.FunctionCallbackInfo) *v8go.Value {\n ctx := info.Context()\n args := info.Args()\n\n if len(args) < 1 {\n return bridge.JsException(ctx, \"Robot requires member ID\")\n }\n\n memberID := args[0].String()\n robotObj, err := RobotNew(ctx, memberID)\n if err != nil {\n return bridge.JsException(ctx, err.Error())\n }\n\n return robotObj\n}\n\n// RobotNew creates a Robot JS object with methods\nfunc RobotNew(ctx *v8go.Context, memberID string) (*v8go.Value, error) {\n iso := ctx.Isolate()\n obj := v8go.NewObjectTemplate(iso)\n\n // Instance methods (operate on this robot)\n obj.Set(\"Status\", v8go.NewFunctionTemplate(iso, jsStatus))\n obj.Set(\"UpdateStatus\", v8go.NewFunctionTemplate(iso, jsUpdateStatus))\n obj.Set(\"Trigger\", v8go.NewFunctionTemplate(iso, jsTrigger))\n obj.Set(\"Executions\", v8go.NewFunctionTemplate(iso, jsExecutions))\n obj.Set(\"Pause\", v8go.NewFunctionTemplate(iso, jsPause))\n obj.Set(\"Resume\", v8go.NewFunctionTemplate(iso, jsResume))\n obj.Set(\"Stop\", v8go.NewFunctionTemplate(iso, jsStop))\n\n // ... create instance with memberID stored\n return obj.NewInstance(ctx)\n}\n```\n\n**Global object `robot` (static methods):**\n\n```go\nfunc init() {\n // Register global robot object (lowercase, for static methods)\n v8.RegisterObject(\"robot\", ExportObject)\n}\n\n// ExportObject exports the robot global object\nfunc ExportObject(iso *v8go.Isolate) *v8go.ObjectTemplate {\n obj := v8go.NewObjectTemplate(iso)\n obj.Set(\"List\", v8go.NewFunctionTemplate(iso, jsList))\n obj.Set(\"Get\", v8go.NewFunctionTemplate(iso, jsGet))\n obj.Set(\"Create\", v8go.NewFunctionTemplate(iso, jsCreate))\n obj.Set(\"Update\", v8go.NewFunctionTemplate(iso, jsUpdate))\n obj.Set(\"Remove\", v8go.NewFunctionTemplate(iso, jsRemove))\n obj.Set(\"Execution\", v8go.NewFunctionTemplate(iso, jsExecution))\n return obj\n}\n```\n\n**TypeScript Interface:**\n\n```typescript\n// ==================== Types ====================\n\ninterface RobotData {\n member_id: string;\n team_id: string;\n display_name: string;\n robot_status: \"idle\" | \"working\" | \"paused\" | \"error\" | \"maintenance\";\n robot_config: RobotConfig;\n}\n\ninterface RobotState {\n member_id: string;\n team_id: string;\n display_name: string;\n status: \"idle\" | \"working\" | \"paused\" | \"error\" | \"maintenance\";\n running: number; // current running execution count\n max_running: number; // max concurrent allowed\n last_run?: string;\n next_run?: string;\n running_ids?: string[]; // list of running execution IDs\n}\n\ninterface TriggerResult {\n accepted: boolean;\n queued: boolean;\n execution?: Execution;\n message?: string;\n}\n\n// Message - same as context.Message, supports rich content\ninterface Message {\n role: \"user\" | \"assistant\" | \"system\" | \"tool\";\n content: string | ContentPart[];\n name?: string;\n tool_call_id?: string;\n tool_calls?: ToolCall[];\n}\n\ninterface ContentPart {\n type: \"text\" | \"image_url\" | \"input_audio\" | \"file\" | \"data\";\n text?: string;\n image_url?: { url: string; detail?: \"auto\" | \"low\" | \"high\" };\n input_audio?: { data: string; format: string };\n file?: { url: string; name?: string; mime_type?: string };\n data?: { data: string; mime_type: string };\n}\n\ninterface TriggerRequest {\n type: \"human\" | \"event\";\n\n // Human intervention fields\n action?:\n | \"task.add\"\n | \"task.cancel\"\n | \"task.update\"\n | \"goal.adjust\"\n | \"goal.add\"\n | \"goal.complete\"\n | \"goal.cancel\"\n | \"plan.add\"\n | \"plan.remove\"\n | \"plan.update\"\n | \"instruct\";\n messages?: Message[]; // supports text, images, files, audio\n insert_at?: \"first\" | \"last\" | \"next\" | \"at\";\n at_index?: number;\n plan_at?: string; // ISO date for plan.add\n\n // Event fields\n source?: \"webhook\" | \"database\";\n event_type?: string; // lead.created, etc.\n data?: Record;\n\n // Executor mode (optional, overrides robot config)\n executor_mode?: \"standard\" | \"dryrun\"; // sandbox not implemented\n}\n\n// ExecutorMode - executor mode type\ntype ExecutorMode = \"standard\" | \"dryrun\" | \"sandbox\";\n// Note: \"sandbox\" requires container infrastructure, falls back to \"dryrun\"\n\ninterface ListQuery {\n team_id?: string;\n status?: \"idle\" | \"working\" | \"paused\" | \"error\" | \"maintenance\";\n keywords?: string;\n clock_mode?: \"times\" | \"interval\" | \"daemon\";\n page?: number;\n pagesize?: number;\n}\n\ninterface ListResult {\n data: RobotData[];\n total: number;\n page: number;\n pagesize: number;\n}\n\ninterface ExecutionQuery {\n status?: \"pending\" | \"running\" | \"completed\" | \"failed\" | \"cancelled\";\n trigger?: \"clock\" | \"human\" | \"event\";\n page?: number;\n pagesize?: number;\n}\n\ninterface ExecutionResult {\n data: Execution[];\n total: number;\n page: number;\n pagesize: number;\n}\n\ninterface CreateRequest {\n display_name: string;\n system_prompt?: string;\n robot_config: RobotConfig;\n}\n\ninterface UpdateRequest {\n display_name?: string;\n system_prompt?: string;\n robot_config?: RobotConfig;\n}\n\n// ==================== Global object: robot ====================\n// Static methods, no instance needed\n\ninterface RobotStatic {\n List(query?: ListQuery): ListResult;\n Get(memberID: string): RobotData;\n Create(teamID: string, data: CreateRequest): RobotData;\n Update(memberID: string, data: UpdateRequest): RobotData;\n Remove(memberID: string): void;\n Execution(execID: string): Execution;\n}\n\ndeclare const robot: RobotStatic;\n\n// ==================== Constructor: Robot ====================\n// Instance methods, operate on specific robot\n\ndeclare class Robot {\n constructor(memberID: string);\n\n // Properties\n readonly memberID: string;\n\n // Instance methods\n Status(): RobotState;\n UpdateStatus(status: string): void;\n Trigger(request: TriggerRequest): TriggerResult;\n Executions(query?: ExecutionQuery): ExecutionResult;\n Pause(execID: string): void;\n Resume(execID: string): void;\n Stop(execID: string): void;\n}\n```\n\n**Usage:**\n\n```javascript\n// ==================== Global object: robot ====================\n// For CRUD and queries (no instance needed)\n\nconst list = robot.List({ team_id: \"team_xyz\", status: \"idle\" });\nconst data = robot.Get(\"mem_abc123\");\nconst newRobot = robot.Create(\"team_xyz\", {\n display_name: \"Sales Bot\",\n robot_config: { ... }\n});\nrobot.Update(\"mem_abc123\", { display_name: \"Updated Bot\" });\nrobot.Remove(\"mem_abc123\");\nconst exec = robot.Execution(\"exec_456\");\n\n// ==================== Constructor: Robot ====================\n// For operating on a specific robot instance\n\nconst bot = new Robot(\"mem_abc123\");\n\n// Instance methods\nconst state = bot.Status();\nif (state.status === \"idle\") {\n const result = bot.Trigger({\n type: \"human\",\n action: \"task.add\",\n messages: [{ role: \"user\", content: \"Analyze sales data\" }],\n insert_at: \"first\",\n });\n console.log(\"Triggered:\", result.accepted);\n}\n\n// Get execution history for this robot\nconst execs = bot.Executions({ status: \"completed\", page: 1 });\n\n// Control execution\nbot.Pause(\"exec_123\");\nbot.Resume(\"exec_123\");\nbot.Stop(\"exec_123\");\n\n// Update status\nbot.UpdateStatus(\"paused\");\n```\n\n**Usage in Agent Hooks:**\n\n```javascript\nfunction Create(ctx, messages) {\n const bot = new Robot(\"mem_abc123\");\n const state = bot.Status();\n\n if (state.status === \"working\") {\n ctx.Send({ type: \"text\", props: { content: \"Robot is busy\" } });\n return null;\n }\n\n const result = bot.Trigger({\n type: \"human\",\n action: \"task.add\",\n messages: [{ role: \"user\", content: \"Analyze this data\" }],\n insert_at: \"first\",\n });\n\n if (result.accepted) {\n ctx.memory.context.Set(\"robot_exec_id\", result.execution.id);\n }\n\n return { messages };\n}\n\nfunction Next(ctx, payload) {\n const execID = ctx.memory.context.Get(\"robot_exec_id\");\n if (execID) {\n const exec = robot.Execution(execID); // use global object\n if (exec.status === \"completed\") {\n ctx.Send({\n type: \"text\",\n props: { content: `Robot completed: ${exec.delivery?.summary}` },\n });\n }\n }\n return null;\n}\n```\n\n---\n\n## 2. Type Definitions\n\n> All types are in `robot/types/` package. Other files import as:\n>\n> ```go\n> import \"github.com/yaoapp/yao/agent/robot/types\"\n> ```\n\n### 2.1 Enums\n\n```go\n// types/enums.go\npackage types\n\n// Phase - execution phase\ntype Phase string\n\nconst (\n PhaseInspiration Phase = \"inspiration\" // P0: Clock only\n PhaseGoals Phase = \"goals\" // P1\n PhaseTasks Phase = \"tasks\" // P2\n PhaseRun Phase = \"run\" // P3\n PhaseDelivery Phase = \"delivery\" // P4\n PhaseLearning Phase = \"learning\" // P5\n)\n\n// AllPhases for iteration\nvar AllPhases = []Phase{\n PhaseInspiration, PhaseGoals, PhaseTasks,\n PhaseRun, PhaseDelivery, PhaseLearning,\n}\n\n// ClockMode - clock trigger mode\ntype ClockMode string\n\nconst (\n ClockTimes ClockMode = \"times\" // run at specific times\n ClockInterval ClockMode = \"interval\" // run every X duration\n ClockDaemon ClockMode = \"daemon\" // run continuously\n)\n\n// TriggerType - trigger source\ntype TriggerType string\n\nconst (\n TriggerClock TriggerType = \"clock\"\n TriggerHuman TriggerType = \"human\"\n TriggerEvent TriggerType = \"event\"\n)\n\n// ExecStatus - execution status\ntype ExecStatus string\n\nconst (\n ExecPending ExecStatus = \"pending\"\n ExecRunning ExecStatus = \"running\"\n ExecCompleted ExecStatus = \"completed\"\n ExecFailed ExecStatus = \"failed\"\n ExecCancelled ExecStatus = \"cancelled\"\n)\n\n// RobotStatus - matches __yao.member.robot_status\ntype RobotStatus string\n\nconst (\n RobotIdle RobotStatus = \"idle\"\n RobotWorking RobotStatus = \"working\"\n RobotPaused RobotStatus = \"paused\"\n RobotError RobotStatus = \"error\"\n RobotMaintenance RobotStatus = \"maintenance\"\n)\n\n// InterventionAction - human intervention action\n// Format: category.action (e.g., \"task.add\", \"goal.adjust\")\ntype InterventionAction string\n\nconst (\n // Task operations\n ActionTaskAdd InterventionAction = \"task.add\" // add a new task\n ActionTaskCancel InterventionAction = \"task.cancel\" // cancel a task\n ActionTaskUpdate InterventionAction = \"task.update\" // update task details\n\n // Goal operations\n ActionGoalAdjust InterventionAction = \"goal.adjust\" // modify current goal\n ActionGoalAdd InterventionAction = \"goal.add\" // add a new goal\n ActionGoalComplete InterventionAction = \"goal.complete\" // mark goal as complete\n ActionGoalCancel InterventionAction = \"goal.cancel\" // cancel a goal\n\n // Plan operations (schedule for later)\n ActionPlanAdd InterventionAction = \"plan.add\" // add to plan queue\n ActionPlanRemove InterventionAction = \"plan.remove\" // remove from plan queue\n ActionPlanUpdate InterventionAction = \"plan.update\" // update planned item\n\n // Instruction (direct command)\n ActionInstruct InterventionAction = \"instruct\" // direct instruction to robot\n)\n\n// Priority - task/goal priority\ntype Priority string\n\nconst (\n PriorityHigh Priority = \"high\"\n PriorityNormal Priority = \"normal\"\n PriorityLow Priority = \"low\"\n)\n\n// DeliveryType - output delivery type\ntype DeliveryType string\n\nconst (\n DeliveryEmail DeliveryType = \"email\" // Email via yao/messenger\n DeliveryWebhook DeliveryType = \"webhook\" // POST to external URL\n DeliveryProcess DeliveryType = \"process\" // Yao Process call\n DeliveryNotify DeliveryType = \"notify\" // In-app notification (future)\n)\n\n// DedupResult - deduplication result\ntype DedupResult string\n\nconst (\n DedupSkip DedupResult = \"skip\" // skip execution\n DedupMerge DedupResult = \"merge\" // merge with existing\n DedupProceed DedupResult = \"proceed\" // proceed normally\n)\n\n// EventSource - event trigger source\ntype EventSource string\n\nconst (\n EventWebhook EventSource = \"webhook\" // HTTP webhook\n EventDatabase EventSource = \"database\" // DB change trigger\n)\n\n// LearningType - learning entry type\ntype LearningType string\n\nconst (\n LearnExecution LearningType = \"execution\" // execution record\n LearnFeedback LearningType = \"feedback\" // error/fix feedback\n LearnInsight LearningType = \"insight\" // pattern/tip insight\n)\n\n```\n\n### 2.2 Context\n\n```go\n// types/context.go\npackage types\n\nimport (\n \"context\"\n \"github.com/yaoapp/yao/openapi/oauth/types\"\n)\n\n// Context - robot execution context (lightweight)\ntype Context struct {\n context.Context // embed standard context\n Auth *types.AuthorizedInfo `json:\"auth,omitempty\"` // reuse oauth AuthorizedInfo\n MemberID string `json:\"member_id,omitempty\"` // current robot member ID\n RequestID string `json:\"request_id,omitempty\"` // request trace ID\n Locale string `json:\"locale,omitempty\"` // locale (e.g., \"en-US\")\n}\n\n// NewContext creates a new robot context\nfunc NewContext(parent context.Context, auth *types.AuthorizedInfo) *Context {\n if parent == nil {\n parent = context.Background()\n }\n return &Context{\n Context: parent,\n Auth: auth,\n }\n}\n\n// UserID returns user ID from auth\nfunc (c *Context) UserID() string {\n if c.Auth == nil {\n return \"\"\n }\n return c.Auth.UserID\n}\n\n// TeamID returns team ID from auth\nfunc (c *Context) TeamID() string {\n if c.Auth == nil {\n return \"\"\n }\n return c.Auth.TeamID\n}\n```\n\n### 2.3 Config Types\n\n```go\n// types/config.go\npackage types\n\nimport \"time\"\n\n// Config - robot_config in __yao.member\ntype Config struct {\n Triggers *Triggers `json:\"triggers,omitempty\"`\n Clock *Clock `json:\"clock,omitempty\"`\n Identity *Identity `json:\"identity\"`\n Quota *Quota `json:\"quota,omitempty\"`\n KB *KB `json:\"kb,omitempty\"` // shared knowledge base (same as assistant)\n DB *DB `json:\"db,omitempty\"` // shared database (same as assistant)\n Learn *Learn `json:\"learn,omitempty\"` // learning config for private KB\n Resources *Resources `json:\"resources,omitempty\"`\n Delivery *DeliveryPreferences `json:\"delivery,omitempty\"` // see section 6.2\n Events []Event `json:\"events,omitempty\"`\n DefaultLocale string `json:\"default_locale,omitempty\"` // Default language for clock/event triggers (e.g., \"en-US\", \"zh-CN\")\n}\n\n// Validate validates the config\nfunc (c *Config) Validate() error {\n if c.Identity == nil || c.Identity.Role == \"\" {\n return ErrMissingIdentity\n }\n if c.Clock != nil {\n if err := c.Clock.Validate(); err != nil {\n return err\n }\n }\n return nil\n}\n\n// Triggers - trigger enable/disable\ntype Triggers struct {\n Clock *TriggerSwitch `json:\"clock,omitempty\"`\n Intervene *TriggerSwitch `json:\"intervene,omitempty\"`\n Event *TriggerSwitch `json:\"event,omitempty\"`\n}\n\ntype TriggerSwitch struct {\n Enabled bool `json:\"enabled\"`\n Actions []string `json:\"actions,omitempty\"` // for intervene\n}\n\n// IsEnabled checks if trigger is enabled (default: true)\nfunc (t *Triggers) IsEnabled(typ TriggerType) bool {\n if t == nil {\n return true\n }\n switch typ {\n case TriggerClock:\n return t.Clock == nil || t.Clock.Enabled\n case TriggerHuman:\n return t.Intervene == nil || t.Intervene.Enabled\n case TriggerEvent:\n return t.Event == nil || t.Event.Enabled\n }\n return false\n}\n\n// Clock - when to wake up\ntype Clock struct {\n Mode ClockMode `json:\"mode\"` // times | interval | daemon\n Times []string `json:\"times,omitempty\"` // [\"09:00\", \"14:00\"]\n Days []string `json:\"days,omitempty\"` // [\"Mon\", \"Tue\"] or [\"*\"]\n Every string `json:\"every,omitempty\"` // \"30m\", \"1h\"\n TZ string `json:\"tz,omitempty\"` // \"Asia/Shanghai\"\n Timeout string `json:\"timeout,omitempty\"` // \"30m\"\n}\n\n// Validate validates clock config\nfunc (c *Clock) Validate() error {\n switch c.Mode {\n case ClockTimes:\n if len(c.Times) == 0 {\n return ErrClockTimesEmpty\n }\n case ClockInterval:\n if c.Every == \"\" {\n return ErrClockIntervalEmpty\n }\n case ClockDaemon:\n // no extra validation\n default:\n return ErrClockModeInvalid\n }\n return nil\n}\n\n// GetTimeout returns parsed timeout duration\nfunc (c *Clock) GetTimeout() time.Duration {\n if c.Timeout == \"\" {\n return 30 * time.Minute // default\n }\n d, err := time.ParseDuration(c.Timeout)\n if err != nil {\n return 30 * time.Minute\n }\n return d\n}\n\n// GetLocation returns timezone location\nfunc (c *Clock) GetLocation() *time.Location {\n if c.TZ == \"\" {\n return time.Local\n }\n loc, err := time.LoadLocation(c.TZ)\n if err != nil {\n return time.Local\n }\n return loc\n}\n\n// Identity - who is this robot\ntype Identity struct {\n Role string `json:\"role\"`\n Duties []string `json:\"duties,omitempty\"`\n Rules []string `json:\"rules,omitempty\"`\n}\n\n// Quota - concurrency limits\ntype Quota struct {\n Max int `json:\"max\"` // max running (default: 2)\n Queue int `json:\"queue\"` // queue size (default: 10)\n Priority int `json:\"priority\"` // 1-10 (default: 5)\n}\n\n// GetMax returns max with default\nfunc (q *Quota) GetMax() int {\n if q == nil || q.Max <= 0 {\n return 2\n }\n return q.Max\n}\n\n// GetQueue returns queue size with default\nfunc (q *Quota) GetQueue() int {\n if q == nil || q.Queue <= 0 {\n return 10\n }\n return q.Queue\n}\n\n// GetPriority returns priority with default\nfunc (q *Quota) GetPriority() int {\n if q == nil || q.Priority <= 0 {\n return 5\n }\n return q.Priority\n}\n\n// KB - knowledge base config (same as assistant, from store/types)\n// Shared KB collections accessible by this robot\ntype KB struct {\n Collections []string `json:\"collections,omitempty\"` // KB collection IDs\n Options map[string]interface{} `json:\"options,omitempty\"`\n}\n\n// DB - database config (same as assistant, from store/types)\n// Shared database models accessible by this robot\ntype DB struct {\n Models []string `json:\"models,omitempty\"` // database model names\n Options map[string]interface{} `json:\"options,omitempty\"`\n}\n\n// Learn - learning config for robot's private KB\n// Private KB is auto-created: robot_{team_id}_{member_id}_kb\ntype Learn struct {\n On bool `json:\"on\"`\n Types []string `json:\"types,omitempty\"` // execution, feedback, insight\n Keep int `json:\"keep,omitempty\"` // days, 0 = forever\n}\n\n// Resources - available agents and tools\ntype Resources struct {\n Phases map[Phase]string `json:\"phases,omitempty\"` // phase -> agent ID\n Agents []string `json:\"agents,omitempty\"`\n MCP []MCPConfig `json:\"mcp,omitempty\"`\n}\n\n// GetPhaseAgent returns agent ID for phase (default: __yao.{phase})\nfunc (r *Resources) GetPhaseAgent(phase Phase) string {\n if r != nil && r.Phases != nil {\n if id, ok := r.Phases[phase]; ok && id != \"\" {\n return id\n }\n }\n return \"__yao.\" + string(phase)\n}\n\ntype MCPConfig struct {\n ID string `json:\"id\"`\n Tools []string `json:\"tools,omitempty\"` // empty = all\n}\n\n// Note: Delivery preferences moved to DeliveryPreferences (see section 6.2)\n\n// Event - event trigger config\ntype Event struct {\n Type EventSource `json:\"type\"` // webhook | database\n Source string `json:\"source\"` // webhook path or table name\n Filter map[string]interface{} `json:\"filter,omitempty\"`\n}\n\n// Monitor - monitoring config\n```\n\n### 2.3 Core Types\n\n```go\n// types/robot.go\npackage types\n\nimport (\n \"context\"\n \"sync\"\n \"time\"\n)\n\n// Robot - runtime representation of an autonomous robot (from __yao.member)\n// Relationship: 1 Robot : N Executions (concurrent)\n// Each trigger creates a new Execution (stored in ExecutionStore)\ntype Robot struct {\n // From __yao.member\n MemberID string `json:\"member_id\"`\n TeamID string `json:\"team_id\"`\n DisplayName string `json:\"display_name\"`\n SystemPrompt string `json:\"system_prompt\"`\n Status RobotStatus `json:\"robot_status\"`\n AutonomousMode bool `json:\"autonomous_mode\"`\n RobotEmail string `json:\"robot_email\"` // Robot's email address for sending emails\n\n // Parsed config (from robot_config JSON field)\n Config *Config `json:\"-\"`\n\n // Runtime state\n LastRun time.Time `json:\"-\"` // last execution start time\n NextRun time.Time `json:\"-\"` // next scheduled execution (for clock trigger)\n\n // Concurrency control\n // Each Robot can run multiple Executions concurrently (up to Quota.Max)\n executions map[string]*Execution // execID -> Execution\n execMu sync.RWMutex\n}\n\n// CanRun checks if robot can accept new execution\nfunc (r *Robot) CanRun() bool {\n r.execMu.RLock()\n defer r.execMu.RUnlock()\n return len(r.executions) < r.Config.Quota.GetMax()\n}\n\n// RunningCount returns current running execution count\nfunc (r *Robot) RunningCount() int {\n r.execMu.RLock()\n defer r.execMu.RUnlock()\n return len(r.executions)\n}\n\n// AddExecution adds an execution to tracking\nfunc (r *Robot) AddExecution(exec *Execution) {\n r.execMu.Lock()\n defer r.execMu.Unlock()\n if r.executions == nil {\n r.executions = make(map[string]*Execution)\n }\n r.executions[exec.ID] = exec\n}\n\n// RemoveExecution removes an execution from tracking\nfunc (r *Robot) RemoveExecution(execID string) {\n r.execMu.Lock()\n defer r.execMu.Unlock()\n delete(r.executions, execID)\n}\n\n// GetExecution returns an execution by ID\nfunc (r *Robot) GetExecution(execID string) *Execution {\n r.execMu.RLock()\n defer r.execMu.RUnlock()\n return r.executions[execID]\n}\n\n// GetExecutions returns all running executions\nfunc (r *Robot) GetExecutions() []*Execution {\n r.execMu.RLock()\n defer r.execMu.RUnlock()\n execs := make([]*Execution, 0, len(r.executions))\n for _, exec := range r.executions {\n execs = append(execs, exec)\n }\n return execs\n}\n\n// Execution - single execution instance\n// Each trigger creates a new Execution, stored in ExecutionStore\ntype Execution struct {\n ID string `json:\"id\"` // unique execution ID\n MemberID string `json:\"member_id\"` // robot member ID (globally unique)\n TeamID string `json:\"team_id\"`\n TriggerType TriggerType `json:\"trigger_type\"` // clock | human | event\n StartTime time.Time `json:\"start_time\"`\n EndTime *time.Time `json:\"end_time,omitempty\"`\n Status ExecStatus `json:\"status\"`\n Phase Phase `json:\"phase\"`\n Error string `json:\"error,omitempty\"`\n\n // UI display fields (updated by executor at each phase)\n // These provide human-readable status for frontend display\n Name string `json:\"name,omitempty\"` // Execution title (updated when goals complete)\n CurrentTaskName string `json:\"current_task_name,omitempty\"` // Current task description (updated during run phase)\n\n // Trigger input (stored for traceability)\n Input *TriggerInput `json:\"input,omitempty\"` // original trigger input\n\n // Phase outputs\n Inspiration *InspirationReport `json:\"inspiration,omitempty\"` // P0: markdown\n Goals *Goals `json:\"goals,omitempty\"` // P1: markdown\n Tasks []Task `json:\"tasks,omitempty\"` // P2: structured tasks\n Current *CurrentState `json:\"current,omitempty\"` // current executing state\n Results []TaskResult `json:\"results,omitempty\"` // P3: task results\n Delivery *DeliveryResult `json:\"delivery,omitempty\"`\n Learning []LearningEntry `json:\"learning,omitempty\"`\n\n // Runtime (internal, not serialized)\n ctx context.Context `json:\"-\"`\n cancel context.CancelFunc `json:\"-\"`\n robot *Robot `json:\"-\"`\n}\n\n// TriggerInput - stored trigger input for traceability\ntype TriggerInput struct {\n // For human intervention\n Action InterventionAction `json:\"action,omitempty\"` // task.add, goal.adjust, etc.\n Messages []context.Message `json:\"messages,omitempty\"` // user's input (text, images, files)\n UserID string `json:\"user_id,omitempty\"` // who triggered\n Locale string `json:\"locale,omitempty\"` // language for UI display (e.g., \"en-US\", \"zh-CN\")\n\n // For event trigger\n Source EventSource `json:\"source,omitempty\"` // webhook | database\n EventType string `json:\"event_type,omitempty\"` // lead.created, etc.\n Data map[string]interface{} `json:\"data,omitempty\"` // event payload\n\n // For clock trigger\n Clock *ClockContext `json:\"clock,omitempty\"` // time context when triggered\n}\n\n// CurrentState - current executing goal and task\ntype CurrentState struct {\n Task *Task `json:\"task,omitempty\"` // current task being executed\n TaskIndex int `json:\"task_index\"` // index in Tasks slice\n Progress string `json:\"progress,omitempty\"` // human-readable progress (e.g., \"2/5 tasks\")\n}\n\n// Goals - P1 output (markdown for LLM + structured metadata)\n// P1 Agent reads InspirationReport and generates goals as markdown\n// Example:\n// ## Goals\n// 1. [High] Analyze sales data and identify trends\n// - Reason: Sales up 50%, need to understand why\n// 2. [Normal] Prepare weekly report for manager\n// - Reason: Friday 5pm, weekly report due\n// 3. [Low] Update CRM with new leads\n// - Reason: 3 pending leads from yesterday\ntype Goals struct {\n Content string `json:\"content\"` // markdown text\n Delivery *DeliveryTarget `json:\"delivery,omitempty\"` // where to send results (for P4)\n}\n\n// DeliveryTarget - where to deliver results (defined in P1, used by P4)\n// Note: This is a hint from P1 Goals. Actual delivery is handled by Delivery Center\n// based on Robot/User preferences, not strictly by this target.\ntype DeliveryTarget struct {\n Type DeliveryType `json:\"type\"` // Preferred delivery type\n Recipients []string `json:\"recipients,omitempty\"` // email addresses, webhook URLs, user IDs\n Format string `json:\"format,omitempty\"` // markdown | html | json | text\n Template string `json:\"template,omitempty\"` // template name or inline template\n Options map[string]interface{} `json:\"options,omitempty\"` // channel-specific options\n}\n\n// Task - planned task (structured, for execution)\ntype Task struct {\n ID string `json:\"id\"`\n Description string `json:\"description,omitempty\"` // human-readable task description (for UI display)\n Messages []context.Message `json:\"messages\"` // original input (text, images, files)\n GoalRef string `json:\"goal_ref,omitempty\"` // reference to goal (e.g., \"Goal 1\")\n Source TaskSource `json:\"source\"` // auto | human | event\n\n // Executor\n ExecutorType ExecutorType `json:\"executor_type\"`\n ExecutorID string `json:\"executor_id\"` // unified ID: agent/assistant/process ID, or \"mcp_server.mcp_tool\" for MCP\n Args []any `json:\"args,omitempty\"`\n\n // MCP-specific fields (required when executor_type is \"mcp\")\n MCPServer string `json:\"mcp_server,omitempty\"` // MCP server/client ID (e.g., \"ark.image.text2img\")\n MCPTool string `json:\"mcp_tool,omitempty\"` // MCP tool name (e.g., \"generate\")\n\n // Validation (defined in P2, used in P3)\n ExpectedOutput string `json:\"expected_output,omitempty\"` // what the task should produce\n // ValidationRules supports two formats:\n // 1. Natural language: \"output must be valid JSON\", \"must contain 'field'\"\n // 2. JSON assertions: `{\"type\": \"type\", \"value\": \"object\"}`, `{\"type\": \"contains\", \"value\": \"success\"}`\n ValidationRules []string `json:\"validation_rules,omitempty\"` // specific checks to perform\n\n // Runtime\n Status TaskStatus `json:\"status\"`\n Order int `json:\"order\"` // execution order (0-based)\n StartTime *time.Time `json:\"start_time,omitempty\"`\n EndTime *time.Time `json:\"end_time,omitempty\"`\n}\n\n// TaskSource - how task was created\ntype TaskSource string\n\nconst (\n TaskSourceAuto TaskSource = \"auto\" // generated by P2 (task planning)\n TaskSourceHuman TaskSource = \"human\" // added via human intervention\n TaskSourceEvent TaskSource = \"event\" // added via event trigger\n)\n\n// ExecutorType - task executor type\ntype ExecutorType string\n\nconst (\n ExecutorAssistant ExecutorType = \"assistant\"\n ExecutorMCP ExecutorType = \"mcp\"\n ExecutorProcess ExecutorType = \"process\"\n)\n\n// TaskStatus - task execution status\ntype TaskStatus string\n\nconst (\n TaskPending TaskStatus = \"pending\"\n TaskRunning TaskStatus = \"running\"\n TaskCompleted TaskStatus = \"completed\"\n TaskFailed TaskStatus = \"failed\"\n TaskSkipped TaskStatus = \"skipped\"\n TaskCancelled TaskStatus = \"cancelled\"\n)\n\n// TaskResult - task execution result\ntype TaskResult struct {\n TaskID string `json:\"task_id\"`\n Success bool `json:\"success\"`\n Output interface{} `json:\"output,omitempty\"`\n Error string `json:\"error,omitempty\"`\n Duration int64 `json:\"duration_ms\"`\n Validation *ValidationResult `json:\"validation,omitempty\"` // P3 validation result\n}\n\n// ValidationResult - P3 validation result with multi-turn conversation support\ntype ValidationResult struct {\n // Basic validation result\n Passed bool `json:\"passed\"` // overall validation passed\n Score float64 `json:\"score,omitempty\"` // 0-1 confidence score\n Issues []string `json:\"issues,omitempty\"` // what failed\n Suggestions []string `json:\"suggestions,omitempty\"` // how to improve\n Details string `json:\"details,omitempty\"` // detailed validation report (markdown)\n\n // Execution state (for multi-turn conversation control)\n Complete bool `json:\"complete\"` // whether expected result is obtained\n NeedReply bool `json:\"need_reply,omitempty\"` // whether to continue conversation\n ReplyContent string `json:\"reply_content,omitempty\"` // content for next turn (if NeedReply)\n}\n\n// DeliveryRequest - pushed to Delivery Center\n// Agent only generates content, Delivery Center decides channels based on preferences\ntype DeliveryRequest struct {\n Content *DeliveryContent `json:\"content\"` // Agent-generated content\n Context *DeliveryContext `json:\"context\"` // Tracking info\n // No Channels field - Delivery Center decides based on Robot/User preferences\n}\n\n// DeliveryContent - content generated by Delivery Agent\ntype DeliveryContent struct {\n Summary string `json:\"summary\"` // Brief summary (1-2 sentences)\n Body string `json:\"body\"` // Full markdown report\n Attachments []DeliveryAttachment `json:\"attachments,omitempty\"` // Output artifacts\n}\n\n// DeliveryAttachment - task output attachment with metadata\n// File uses wrapper format: __://\n// Example: __yao.attachment://ccd472d11feb96e03a3fc468f494045c\n// Parse with attachment.Parse(value) → (uploader, fileID, isWrapper)\ntype DeliveryAttachment struct {\n Title string `json:\"title\"` // Human-readable title, e.g., \"Market Analysis Report\"\n Description string `json:\"description,omitempty\"` // Description of what this artifact is\n TaskID string `json:\"task_id,omitempty\"` // Which task produced this artifact\n File string `json:\"file\"` // Wrapper format: __://\n}\n\n// DeliveryContext - tracking and audit info\ntype DeliveryContext struct {\n MemberID string `json:\"member_id\"` // Robot member ID (globally unique)\n ExecutionID string `json:\"execution_id\"`\n TriggerType TriggerType `json:\"trigger_type\"`\n TeamID string `json:\"team_id\"`\n}\n\n// DeliveryPreferences - Robot/User delivery preferences (read by Delivery Center)\n// Each channel supports multiple targets\ntype DeliveryPreferences struct {\n Email *EmailPreference `json:\"email,omitempty\"`\n Webhook *WebhookPreference `json:\"webhook,omitempty\"`\n Process *ProcessPreference `json:\"process,omitempty\"`\n // notify is handled automatically based on user subscriptions\n}\n\n// EmailPreference - multiple email targets\ntype EmailPreference struct {\n Enabled bool `json:\"enabled\"`\n Targets []EmailTarget `json:\"targets\"`\n}\n\ntype EmailTarget struct {\n To []string `json:\"to\"` // Recipient addresses\n Template string `json:\"template,omitempty\"` // Email template ID\n Subject string `json:\"subject,omitempty\"` // Subject template (default: content.Summary)\n}\n\n// WebhookPreference - multiple webhook targets\ntype WebhookPreference struct {\n Enabled bool `json:\"enabled\"`\n Targets []WebhookTarget `json:\"targets\"`\n}\n\ntype WebhookTarget struct {\n URL string `json:\"url\"` // Webhook URL\n Method string `json:\"method,omitempty\"` // HTTP method (default: POST)\n Headers map[string]string `json:\"headers,omitempty\"` // Custom headers\n Secret string `json:\"secret,omitempty\"` // Signing secret\n}\n\n// ProcessPreference - multiple Yao Process targets\ntype ProcessPreference struct {\n Enabled bool `json:\"enabled\"`\n Targets []ProcessTarget `json:\"targets\"`\n}\n\ntype ProcessTarget struct {\n Process string `json:\"process\"` // Yao Process name, e.g., \"orders.UpdateStatus\"\n Args []any `json:\"args,omitempty\"` // Additional args (DeliveryContent passed as first arg)\n}\n\n// DeliveryResult - P4 delivery output (returned by Delivery Center)\ntype DeliveryResult struct {\n RequestID string `json:\"request_id\"` // Delivery request ID\n Content *DeliveryContent `json:\"content\"` // Agent-generated content\n Results []ChannelResult `json:\"results,omitempty\"` // Results per channel\n Success bool `json:\"success\"` // Overall success\n Error string `json:\"error,omitempty\"` // Error if failed\n SentAt *time.Time `json:\"sent_at,omitempty\"` // When delivery completed\n}\n\n// ChannelResult - result for a single delivery target\ntype ChannelResult struct {\n Type DeliveryType `json:\"type\"` // email | webhook | process\n Target string `json:\"target\"` // Target identifier (email, URL, process name)\n Success bool `json:\"success\"` // Whether delivery succeeded\n Recipients []string `json:\"recipients,omitempty\"` // Who received (for email)\n Details interface{} `json:\"details,omitempty\"` // Channel-specific response\n Error string `json:\"error,omitempty\"` // Error message if failed\n SentAt *time.Time `json:\"sent_at,omitempty\"` // When this target was delivered\n}\n\n// LearningEntry - knowledge to save\ntype LearningEntry struct {\n Type LearningType `json:\"type\"` // execution | feedback | insight\n Content string `json:\"content\"`\n Tags []string `json:\"tags,omitempty\"`\n Meta interface{} `json:\"meta,omitempty\"`\n}\n```\n\n### 2.4 Clock Context\n\n```go\n// types/clock.go\npackage types\n\nimport \"time\"\n\n// ClockContext - time context for P0 inspiration\ntype ClockContext struct {\n Now time.Time `json:\"now\"`\n Hour int `json:\"hour\"` // 0-23\n DayOfWeek string `json:\"day_of_week\"` // Monday, Tuesday...\n DayOfMonth int `json:\"day_of_month\"` // 1-31\n WeekOfYear int `json:\"week_of_year\"` // 1-52\n Month int `json:\"month\"` // 1-12\n Year int `json:\"year\"`\n IsWeekend bool `json:\"is_weekend\"`\n IsMonthStart bool `json:\"is_month_start\"` // 1st-3rd\n IsMonthEnd bool `json:\"is_month_end\"` // last 3 days\n IsQuarterEnd bool `json:\"is_quarter_end\"`\n IsYearEnd bool `json:\"is_year_end\"`\n TZ string `json:\"tz\"`\n}\n\n// NewClockContext creates clock context from time\nfunc NewClockContext(t time.Time, tz string) *ClockContext {\n loc := time.Local\n if tz != \"\" {\n if l, err := time.LoadLocation(tz); err == nil {\n loc = l\n }\n }\n t = t.In(loc)\n\n _, week := t.ISOWeek()\n dayOfMonth := t.Day()\n lastDay := time.Date(t.Year(), t.Month()+1, 0, 0, 0, 0, 0, loc).Day()\n\n return &ClockContext{\n Now: t,\n Hour: t.Hour(),\n DayOfWeek: t.Weekday().String(),\n DayOfMonth: dayOfMonth,\n WeekOfYear: week,\n Month: int(t.Month()),\n Year: t.Year(),\n IsWeekend: t.Weekday() == time.Saturday || t.Weekday() == time.Sunday,\n IsMonthStart: dayOfMonth <= 3,\n IsMonthEnd: dayOfMonth >= lastDay-2,\n IsQuarterEnd: (t.Month()%3 == 0) && dayOfMonth >= lastDay-2,\n IsYearEnd: t.Month() == 12 && dayOfMonth >= 29,\n TZ: loc.String(),\n }\n}\n```\n\n### 2.5 Inspiration Report\n\n```go\n// types/inspiration.go\npackage types\n\n// InspirationReport - P0 output (simple markdown for LLM)\ntype InspirationReport struct {\n Clock *ClockContext `json:\"clock\"` // time context\n Content string `json:\"content\"` // markdown text for LLM\n}\n\n// Content is markdown like:\n// ## Summary\n// ...\n// ## Highlights\n// - [High] Sales up 50%\n// - [Medium] New lead from BigCorp\n// ## Opportunities\n// ...\n// ## Risks\n// ...\n// ## World News\n// ...\n// ## Pending\n// ...\n```\n\n### 2.6 Request/Response Types\n\n```go\n// types/request.go\npackage types\n\nimport (\n \"context\"\n \"time\"\n)\n\n// InterveneRequest - human intervention request\n// Processed by Manager.Intervene()\ntype InterveneRequest struct {\n TeamID string `json:\"team_id\"`\n MemberID string `json:\"member_id\"`\n Action InterventionAction `json:\"action\"` // task.add, goal.adjust, etc.\n Messages []agentcontext.Message `json:\"messages,omitempty\"` // user input (text, images, files)\n PlanTime *time.Time `json:\"plan_time,omitempty\"` // for action=plan.add\n ExecutorMode ExecutorMode `json:\"executor_mode,omitempty\"` // optional: standard | dryrun\n}\n\n// EventRequest - event trigger request\n// Processed by Manager.HandleEvent()\ntype EventRequest struct {\n MemberID string `json:\"member_id\"`\n Source string `json:\"source\"` // webhook path or table name\n EventType string `json:\"event_type\"` // lead.created, etc.\n Data map[string]interface{} `json:\"data,omitempty\"`\n ExecutorMode ExecutorMode `json:\"executor_mode,omitempty\"` // optional: standard | dryrun\n}\n\n// ExecutorMode - executor mode enum\ntype ExecutorMode string\n\nconst (\n ExecutorStandard ExecutorMode = \"standard\" // real Agent calls (default)\n ExecutorDryRun ExecutorMode = \"dryrun\" // simulated, no LLM calls\n ExecutorSandbox ExecutorMode = \"sandbox\" // container-isolated (NOT IMPLEMENTED)\n)\n\n// ExecutionResult - trigger result\ntype ExecutionResult struct {\n ExecutionID string `json:\"execution_id\"`\n Status ExecStatus `json:\"status\"`\n Message string `json:\"message,omitempty\"`\n}\n\n// RobotState - robot status query result\ntype RobotState struct {\n MemberID string `json:\"member_id\"`\n TeamID string `json:\"team_id\"`\n DisplayName string `json:\"display_name\"`\n Status RobotStatus `json:\"status\"`\n Running int `json:\"running\"` // current running execution count\n MaxRunning int `json:\"max_running\"` // max concurrent allowed\n LastRun *time.Time `json:\"last_run,omitempty\"`\n NextRun *time.Time `json:\"next_run,omitempty\"`\n RunningIDs []string `json:\"running_ids,omitempty\"` // list of running execution IDs\n}\n```\n\n---\n\n## 3. Interfaces\n\n> Interfaces are also in `types/` package to avoid cycles.\n\n### 3.1 Manager Interface\n\n```go\n// types/interfaces.go\npackage types\n\nimport \"time\"\n\n// ==================== Internal Interfaces ====================\n// These are internal implementation interfaces, not exposed via API.\n// External API is defined in api/api.go\n// All interfaces use *Context (not context.Context) for consistency.\n\n// Manager - robot lifecycle, scheduling, and all trigger handling\n// Manager is the central orchestrator, handling:\n// - Clock triggers (via Tick)\n// - Human intervention (via Intervene)\n// - Event triggers (via HandleEvent)\n// - Execution control (pause/resume/stop)\ntype Manager interface {\n // Lifecycle\n Start() error\n Stop() error\n\n // Clock trigger (called by internal ticker)\n Tick(ctx *Context, now time.Time) error\n\n // Manual trigger (for testing/API)\n TriggerManual(ctx *Context, memberID string, trigger TriggerType, data interface{}) (string, error)\n\n // Human intervention\n Intervene(ctx *Context, req *InterveneRequest) (*ExecutionResult, error)\n\n // Event trigger\n HandleEvent(ctx *Context, req *EventRequest) (*ExecutionResult, error)\n\n // Execution control\n PauseExecution(ctx *Context, execID string) error\n ResumeExecution(ctx *Context, execID string) error\n StopExecution(ctx *Context, execID string) error\n}\n\n// Executor - executes robot phases\ntype Executor interface {\n Execute(ctx *Context, robot *Robot, trigger TriggerType, data interface{}) (*Execution, error)\n}\n\n// Pool - worker pool for concurrent execution\ntype Pool interface {\n Start() error\n Stop() error\n Submit(ctx *Context, robot *Robot, trigger TriggerType, data interface{}) (string, error)\n Running() int\n Queued() int\n}\n\n// Cache - in-memory robot cache\ntype Cache interface {\n Load(ctx *Context) error\n Get(memberID string) *Robot\n List(teamID string) []*Robot\n Refresh(ctx *Context, memberID string) error\n Add(robot *Robot)\n Remove(memberID string)\n}\n\n// Dedup - deduplication check\ntype Dedup interface {\n Check(ctx *Context, memberID string, trigger TriggerType) (DedupResult, error)\n Mark(memberID string, trigger TriggerType, window time.Duration)\n}\n\n// Store - data storage operations (KB, DB)\ntype Store interface {\n SaveLearning(ctx *Context, memberID string, entries []LearningEntry) error\n GetHistory(ctx *Context, memberID string, limit int) ([]LearningEntry, error)\n SearchKB(ctx *Context, collections []string, query string) ([]interface{}, error)\n QueryDB(ctx *Context, models []string, query interface{}) ([]interface{}, error)\n}\n```\n\n### 3.2 Trigger Utilities (`trigger/` package)\n\n> **Note:** The `trigger/` package provides utilities, not the main trigger logic.\n> All trigger handling is done by `Manager`.\n\n```go\n// trigger/trigger.go - Validation and helper functions\n\n// ValidateIntervention validates a human intervention request\nfunc ValidateIntervention(req *InterveneRequest) error\n\n// ValidateEvent validates an event trigger request\nfunc ValidateEvent(req *EventRequest) error\n\n// BuildEventInput creates a TriggerInput from an event request\nfunc BuildEventInput(req *EventRequest) *TriggerInput\n\n// GetActionCategory returns the category of an intervention action\n// e.g., \"task.add\" -> \"task\", \"goal.adjust\" -> \"goal\"\nfunc GetActionCategory(action InterventionAction) string\n\n// GetActionDescription returns a human-readable description of an action\nfunc GetActionDescription(action InterventionAction) string\n```\n\n```go\n// trigger/clock.go - Clock matching logic (reusable)\n\n// ClockMatcher provides clock trigger matching logic\ntype ClockMatcher struct{}\n\n// ShouldTrigger checks if a robot should be triggered based on its clock config\nfunc (cm *ClockMatcher) ShouldTrigger(robot *Robot, now time.Time) bool\n\n// ParseTime parses a time string in \"HH:MM\" format\nfunc ParseTime(timeStr string) (hour, minute int, err error)\n\n// FormatTime formats hour and minute to \"HH:MM\" string\nfunc FormatTime(hour, minute int) string\n```\n\n```go\n// trigger/control.go - Execution control (pause/resume/stop)\n\n// ExecutionController manages execution lifecycle\ntype ExecutionController struct {\n executions map[string]*ControlledExecution\n mu sync.RWMutex\n}\n\n// Track starts tracking an execution\nfunc (c *ExecutionController) Track(execID, memberID, teamID string) *ControlledExecution\n\n// Untrack stops tracking an execution\nfunc (c *ExecutionController) Untrack(execID string)\n\n// Pause pauses an execution\nfunc (c *ExecutionController) Pause(execID string) error\n\n// Resume resumes a paused execution\nfunc (c *ExecutionController) Resume(execID string) error\n\n// Stop stops an execution\nfunc (c *ExecutionController) Stop(execID string) error\n\n// ControlledExecution represents an execution that can be controlled\ntype ControlledExecution struct {\n ID string\n MemberID string\n TeamID string\n Status ExecStatus\n Phase Phase\n StartTime time.Time\n PausedAt *time.Time\n // ... internal fields for context and channels\n}\n\n// IsPaused returns true if the execution is paused\nfunc (e *ControlledExecution) IsPaused() bool\n\n// IsCancelled returns true if the execution is cancelled\nfunc (e *ControlledExecution) IsCancelled() bool\n\n// WaitIfPaused blocks until the execution is resumed or cancelled\nfunc (e *ControlledExecution) WaitIfPaused() error\n\n// CheckCancelled checks if the execution is cancelled and returns error if so\nfunc (e *ControlledExecution) CheckCancelled() error\n```\n\n---\n\n## 4. Errors\n\n```go\n// types/errors.go\npackage types\n\nimport \"errors\"\n\nvar (\n // Config errors\n ErrMissingIdentity = errors.New(\"identity.role is required\")\n ErrClockTimesEmpty = errors.New(\"clock.times is required for times mode\")\n ErrClockIntervalEmpty = errors.New(\"clock.every is required for interval mode\")\n ErrClockModeInvalid = errors.New(\"clock.mode must be times, interval, or daemon\")\n\n // Runtime errors\n ErrRobotNotFound = errors.New(\"robot not found\")\n ErrRobotPaused = errors.New(\"robot is paused\")\n ErrRobotBusy = errors.New(\"robot has reached max concurrent executions\")\n ErrTriggerDisabled = errors.New(\"trigger type is disabled for this robot\")\n ErrExecutionCancelled = errors.New(\"execution was cancelled\")\n ErrExecutionTimeout = errors.New(\"execution timed out\")\n\n // Phase errors\n ErrPhaseAgentNotFound = errors.New(\"phase agent not found\")\n ErrGoalGenFailed = errors.New(\"goal generation failed\")\n ErrTaskPlanFailed = errors.New(\"task planning failed\")\n ErrDeliveryFailed = errors.New(\"delivery failed\")\n)\n```\n\n---\n\n## 5. P3 Implementation Details\n\n### 5.1 Multi-Turn Conversation Flow\n\nFor assistant tasks, P3 uses a validator-driven multi-turn conversation:\n\n```\n┌──────────────────────────────────────────────────────────────┐\n│ executeAssistantWithMultiTurn │\n├──────────────────────────────────────────────────────────────┤\n│ 1. Create Conversation (single instance for entire task) │\n│ 2. Build initial messages with task context │\n│ │\n│ ┌─────────────────── Turn Loop ───────────────────────────┐ │\n│ │ Phase 1: Call assistant via conv.Turn() │ │\n│ │ Phase 2: ValidateWithContext() determines: │ │\n│ │ - Complete: task done? │ │\n│ │ - NeedReply: continue conversation? │ │\n│ │ - ReplyContent: what to send next? │ │\n│ │ Phase 3: If NeedReply, use ReplyContent as next input │ │\n│ │ Break if: Complete && Passed, or !NeedReply │ │\n│ └──────────────────────────────────────────────────────────┘ │\n│ │\n│ 3. Return output, validation, error │\n└──────────────────────────────────────────────────────────────┘\n```\n\nKey points:\n- `ValidateWithContext()` returns `NeedReply` and `ReplyContent`\n- Conversation continues until `Complete && Passed` or `!NeedReply`\n- Max turns controlled by `RunConfig.MaxTurnsPerTask`\n\n### 5.2 Validation Rules Format\n\nValidation rules support two formats:\n\n1. **Natural language**: `\"output must be valid JSON\"`, `\"must contain 'field'\"`\n2. **Structured JSON**: `{\"type\": \"type\", \"path\": \"field\", \"value\": \"array\"}`\n\nExamples:\n```json\n// Natural language rules (converted to semantic validation)\n\"output must be valid JSON\"\n\"must contain product name\"\n\n// Structured JSON assertions\n{\"type\": \"equals\", \"value\": \"success\"}\n{\"type\": \"contains\", \"value\": \"total\"}\n{\"type\": \"regex\", \"value\": \"^[A-Z].*\"}\n{\"type\": \"json_path\", \"path\": \"data.items\", \"value\": 10}\n{\"type\": \"type\", \"path\": \"result\", \"value\": \"object\"}\n```\n\n### 5.3 Task Dependencies\n\nTask dependencies are handled automatically:\n\n1. `BuildTaskContext()` collects previous task results\n2. `FormatPreviousResultsAsContext()` formats them for assistant\n\n```go\n// Previous results are passed as context\nfunc (r *Runner) BuildTaskContext(exec *robottypes.Execution, taskIndex int) *RunnerContext {\n ctx := &RunnerContext{}\n if taskIndex > 0 {\n ctx.PreviousResults = exec.Results[:taskIndex]\n }\n return ctx\n}\n```\n\n### 5.4 Resource Management\n\nAgent context is properly released to prevent resource leaks:\n\n```go\nfunc (c *AgentCaller) Call(ctx *robottypes.Context, assistantID string, messages []agentcontext.Message) (*CallResult, error) {\n agentCtx := c.buildAgentContext(ctx)\n defer agentCtx.Release() // IMPORTANT: Release agent context\n \n response, err := ast.Stream(agentCtx, messages, opts)\n // ...\n}\n```\n\n### 5.5 yao/assert Package\n\nThe `yao/assert` package is a standalone universal assertion library that can be used by other modules:\n\n```go\nimport \"github.com/yaoapp/yao/assert\"\n\n// Create asserter with optional callbacks\nasserter := assert.NewAsserter(assert.AssertionOptions{\n AgentValidator: myAgentValidator, // for \"agent\" type assertions\n ScriptRunner: myScriptRunner, // for \"script\" type assertions\n})\n\n// Run assertions\nresults := asserter.Assert(output, []assert.Assertion{\n {Type: \"type\", Value: \"object\"},\n {Type: \"contains\", Value: \"success\"},\n {Type: \"json_path\", Path: \"data.count\", Value: 10},\n})\n```\n\nSupported assertion types:\n- `equals` - exact match\n- `contains` - substring check\n- `not_contains` - negative substring check\n- `json_path` - JSON path extraction and comparison\n- `regex` - regex pattern matching\n- `type` - type checking (with optional path)\n- `script` - custom script validation\n- `agent` - AI agent validation\n\n---\n\n## 6. P4 Delivery Implementation\n\n### 6.1 Overview\n\nP4 Delivery summarizes P3 execution results and delivers to configured channels.\n\n```\n┌─────────────────────────────────────────────────────────────┐\n│ delivery.go (P4 Entry) │\n│ - DeliveryExecution: main entry point │\n│ - Calls Delivery Agent with full execution context │\n│ - Routes DeliveryContent to configured channels │\n└─────────────────────┬───────────────────────────────────────┘\n │\n ┌────────────┴────────────┐\n ▼ ▼\n┌─────────────────┐ ┌─────────────────┐\n│ Delivery Agent │ │ Delivery Center │\n│ - Summarize │ │ - sendEmail() │\n│ - Format body │ │ - postWebhook() │\n│ - List files │ │ - callProcess() │\n└─────────────────┘ └─────────────────┘\n```\n\n### 6.2 Delivery Request Structure\n\nP4 generates a `DeliveryRequest` with **only content** and pushes to Delivery Center.\n**Delivery Center decides channels** based on Robot/User preferences.\n\n```go\n// DeliveryRequest - pushed to Delivery Center\n// No Channels - Delivery Center decides based on preferences\ntype DeliveryRequest struct {\n Content *DeliveryContent `json:\"content\"` // Agent-generated content\n Context *DeliveryContext `json:\"context\"` // Tracking info\n}\n\n// DeliveryContent - content generated by Delivery Agent\ntype DeliveryContent struct {\n Summary string `json:\"summary\"` // Brief 1-2 sentence summary\n Body string `json:\"body\"` // Full markdown report\n Attachments []DeliveryAttachment `json:\"attachments,omitempty\"` // Output artifacts from P3\n}\n\n// DeliveryAttachment - file attachment with metadata\ntype DeliveryAttachment struct {\n Title string `json:\"title\"` // Human-readable title\n Description string `json:\"description,omitempty\"` // What this artifact is\n TaskID string `json:\"task_id,omitempty\"` // Which task produced this\n File string `json:\"file\"` // Wrapper: __://\n}\n\n// DeliveryContext - tracking and audit info\ntype DeliveryContext struct {\n MemberID string `json:\"member_id\"` // Robot member ID (globally unique)\n ExecutionID string `json:\"execution_id\"`\n TriggerType TriggerType `json:\"trigger_type\"`\n TeamID string `json:\"team_id\"`\n}\n```\n\n**Example DeliveryRequest:**\n\n```json\n{\n \"content\": {\n \"summary\": \"Sales report completed: 15 new leads\",\n \"body\": \"## Weekly Sales Report\\n...\",\n \"attachments\": [{\"title\": \"Report.pdf\", \"file\": \"__yao.attachment://abc123\"}]\n },\n \"context\": {\n \"member_id\": \"mem_abc123\",\n \"execution_id\": \"exec_xyz789\",\n \"trigger_type\": \"clock\",\n \"team_id\": \"team_123\"\n }\n}\n```\n\n**Channel Decision by Delivery Center:**\n\nDelivery Center reads Robot/User preferences and executes delivery to all enabled targets:\n\n```go\n// DeliveryPreferences - from Robot config (each channel supports multiple targets)\ntype DeliveryPreferences struct {\n Email *EmailPreference `json:\"email,omitempty\"`\n Webhook *WebhookPreference `json:\"webhook,omitempty\"`\n Process *ProcessPreference `json:\"process,omitempty\"`\n}\n\ntype EmailPreference struct {\n Enabled bool `json:\"enabled\"`\n Targets []EmailTarget `json:\"targets\"` // Multiple email targets\n}\n\ntype WebhookPreference struct {\n Enabled bool `json:\"enabled\"`\n Targets []WebhookTarget `json:\"targets\"` // Multiple webhook URLs\n}\n\ntype ProcessPreference struct {\n Enabled bool `json:\"enabled\"`\n Targets []ProcessTarget `json:\"targets\"` // Multiple Yao Process calls\n}\n```\n\n### 6.3 File Wrapper Format\n\nAttachments use the standard `yao/attachment` wrapper format:\n\n```go\n// Format: __://\n// Example: __yao.attachment://ccd472d11feb96e03a3fc468f494045c\n\nimport \"github.com/yaoapp/yao/attachment\"\n\n// Parse wrapper to get uploader and fileID\nuploader, fileID, isWrapper := attachment.Parse(wrapper)\n// uploader: \"__yao.attachment\"\n// fileID: \"ccd472d11feb96e03a3fc468f494045c\"\n// isWrapper: true\n\n// Get file info\nmanager := attachment.Managers[uploader]\nfileInfo, err := manager.Info(ctx, fileID)\n\n// Read file content as base64\nbase64Content := attachment.Base64(ctx, wrapper)\n\n// Read with data URI format\ndataURI := attachment.Base64(ctx, wrapper, true)\n// \"data:image/png;base64,...\"\n```\n\n### 6.4 Delivery Agent\n\nThe Delivery Agent **only generates content**, does NOT decide channels.\nChannel decisions are made by Delivery Center based on Robot/User preferences.\n\n**Input:**\n```go\ntype DeliveryAgentInput struct {\n Robot *Robot `json:\"robot\"` // Robot identity and config\n TriggerType TriggerType `json:\"trigger\"` // clock | human | event\n Inspiration *InspirationReport `json:\"inspiration\"` // P0 (clock only)\n Goals *Goals `json:\"goals\"` // P1\n Tasks []Task `json:\"tasks\"` // P2\n Results []TaskResult `json:\"results\"` // P3\n}\n```\n\n**Output:**\n```go\n// DeliveryAgentOutput - only content, no channels\ntype DeliveryAgentOutput struct {\n Content *DeliveryContent `json:\"content\"` // Generated content\n}\n```\n\n**Agent Responsibilities:**\n\nThe agent focuses on content generation:\n- **Summary**: Brief 1-2 sentence summary of execution results\n- **Body**: Full markdown report with details\n- **Attachments**: Select which P3-generated files to include\n\n**Example Output:**\n\n```json\n{\n \"content\": {\n \"summary\": \"Sales report completed: 15 new leads processed, 3 high-priority\",\n \"body\": \"## Weekly Sales Report\\n\\n### Summary\\n- Total leads: 15\\n- High priority: 3\\n...\",\n \"attachments\": [\n {\"title\": \"Sales Report.pdf\", \"task_id\": \"task_1\", \"file\": \"__yao.attachment://abc123\"},\n {\"title\": \"Lead Analysis.xlsx\", \"task_id\": \"task_2\", \"file\": \"__yao.attachment://def456\"}\n ]\n }\n}\n```\n\n### 6.5 Global Email Configuration\n\nEmail delivery uses global configuration for channel selection and Robot-specific sender identity:\n\n```go\n// types/config_global.go\n\n// DefaultEmailChannel returns the default messenger channel name\n// Default: \"email\" (maps to messengers/channels.yao)\nfunc DefaultEmailChannel() string\n\n// SetDefaultEmailChannel sets the default channel (call during agent init)\nfunc SetDefaultEmailChannel(channel string)\n```\n\n**Usage:**\n- `DefaultEmailChannel()` - returns the messenger channel name for email delivery\n- `Robot.RobotEmail` - used as the `From` address when sending emails\n- If `RobotEmail` is empty, falls back to provider's default `from` address\n\n### 6.6 Delivery Center\n\nThe Delivery Center receives `DeliveryRequest`, reads preferences, and executes delivery to **all enabled targets**.\n\n**Current implementation:** Internal to P4 (in `executor/delivery.go`)\n**Future:** Can be extracted to standalone `yao/delivery` package\n\n```go\n// DeliveryCenter - handles delivery execution to multiple targets\ntype DeliveryCenter struct {\n messenger *messenger.Manager\n}\n\n// Deliver - main entry point\nfunc (dc *DeliveryCenter) Deliver(ctx context.Context, req *DeliveryRequest) *DeliveryResult {\n requestID := generateID()\n prefs := dc.getDeliveryPreferences(ctx, req.Context.MemberID)\n \n var results []ChannelResult\n allSuccess := true\n \n // Email - send to all targets (robot passed for From address)\n if prefs.Email != nil && prefs.Email.Enabled {\n for _, target := range prefs.Email.Targets {\n result := dc.sendEmail(ctx, req.Content, target, req.Context, robot)\n results = append(results, result)\n if !result.Success {\n allSuccess = false\n }\n }\n }\n \n // Webhook - POST to all targets\n if prefs.Webhook != nil && prefs.Webhook.Enabled {\n for _, target := range prefs.Webhook.Targets {\n result := dc.postWebhook(ctx, req.Content, target)\n results = append(results, result)\n if !result.Success {\n allSuccess = false\n }\n }\n }\n \n // Process - call all targets\n if prefs.Process != nil && prefs.Process.Enabled {\n for _, target := range prefs.Process.Targets {\n result := dc.callProcess(ctx, req.Content, target)\n results = append(results, result)\n if !result.Success {\n allSuccess = false\n }\n }\n }\n \n // Future: auto-notify based on user subscriptions\n // dc.sendNotifications(ctx, req)\n \n return &DeliveryResult{\n RequestID: requestID,\n Content: req.Content,\n Success: allSuccess,\n Results: results,\n }\n}\n```\n\n### 6.7 Channel Handlers\n\nEach delivery channel is handled by dedicated methods in DeliveryCenter:\n\n```go\n// sendEmail - send to a single email target\n// Uses Robot.RobotEmail as From address and global DefaultEmailChannel()\nfunc (dc *DeliveryCenter) sendEmail(\n ctx context.Context,\n content *DeliveryContent,\n target EmailTarget,\n deliveryCtx *DeliveryContext,\n robot *Robot,\n) ChannelResult {\n // Convert attachments to messenger format\n var attachments []messenger.Attachment\n for _, att := range content.Attachments {\n uploader, fileID, _ := attachment.Parse(att.File)\n manager := attachment.Managers[uploader]\n data, _ := manager.Read(ctx, fileID)\n info, _ := manager.Info(ctx, fileID)\n \n attachments = append(attachments, messenger.Attachment{\n Filename: att.Title,\n ContentType: info.ContentType,\n Content: data,\n })\n }\n \n subject := content.Summary\n if target.Subject != \"\" {\n subject = target.Subject\n }\n \n msg := &messenger.Message{\n To: target.To,\n Subject: subject,\n Body: content.Body,\n Attachments: attachments,\n }\n \n // Set From address from Robot's email (if configured)\n if robot != nil && robot.RobotEmail != \"\" {\n msg.From = robot.RobotEmail\n }\n \n // Use global default email channel\n channel := DefaultEmailChannel() // from types/config_global.go\n err := dc.messenger.Send(ctx, channel, msg)\n \n now := time.Now()\n return ChannelResult{\n Type: DeliveryEmail,\n Target: strings.Join(target.To, \",\"),\n Success: err == nil,\n Recipients: target.To,\n SentAt: &now,\n Error: errStr(err),\n }\n}\n\n// postWebhook - POST to a single webhook target\nfunc (dc *DeliveryCenter) postWebhook(ctx context.Context, content *DeliveryContent, target WebhookTarget) ChannelResult {\n payload, _ := json.Marshal(content)\n req, _ := http.NewRequestWithContext(ctx, \"POST\", target.URL, bytes.NewReader(payload))\n req.Header.Set(\"Content-Type\", \"application/json\")\n \n // Add custom headers\n for k, v := range target.Headers {\n req.Header.Set(k, v)\n }\n \n resp, err := http.DefaultClient.Do(req)\n now := time.Now()\n \n if err != nil {\n return ChannelResult{\n Type: DeliveryWebhook,\n Target: target.URL,\n Success: false,\n Error: err.Error(),\n SentAt: &now,\n }\n }\n defer resp.Body.Close()\n \n success := resp.StatusCode < 400\n return ChannelResult{\n Type: DeliveryWebhook,\n Target: target.URL,\n Success: success,\n Details: map[string]interface{}{\"status_code\": resp.StatusCode},\n Error: ternary(!success, fmt.Sprintf(\"HTTP %d\", resp.StatusCode), \"\"),\n SentAt: &now,\n }\n}\n\n// callProcess - call a single Yao Process target\nfunc (dc *DeliveryCenter) callProcess(ctx context.Context, content *DeliveryContent, target ProcessTarget) ChannelResult {\n // DeliveryContent as first arg, then additional args\n args := append([]interface{}{content}, target.Args...)\n \n proc := process.Of(target.Process, args...)\n result, err := proc.Execute()\n \n now := time.Now()\n return ChannelResult{\n Type: DeliveryProcess,\n Target: target.Process,\n Success: err == nil,\n Details: map[string]interface{}{\n \"process\": target.Process,\n \"result\": result,\n },\n Error: errStr(err),\n SentAt: &now,\n }\n}\n```\n\n**Note on Notifications:**\n\n`notify` is NOT configured per-Robot. Future Delivery Center will:\n1. Check user subscription preferences after receiving DeliveryRequest\n2. Automatically send in-app notifications to subscribed users\n3. This is transparent to P4 and Delivery Agent\n\n### 6.8 Execution Persistence\n\nRobot execution history is stored in `__yao.agent_execution` table for UI display:\n\n```go\n// Model: yao/models/agent/execution.mod.yao\n// Table: __yao.agent_execution\n\ntype ExecutionRecord struct {\n ID int64 `json:\"id,omitempty\"` // Auto-increment primary key\n ExecutionID string `json:\"execution_id\"` // Unique execution identifier\n MemberID string `json:\"member_id\"` // Robot member ID (globally unique)\n TeamID string `json:\"team_id\"` // Team ID\n TriggerType TriggerType `json:\"trigger_type\"` // clock | human | event\n \n // Status tracking (synced with runtime Execution)\n Status ExecStatus `json:\"status\"` // pending | running | completed | failed | cancelled\n Phase Phase `json:\"phase\"` // Current phase\n Current *CurrentState `json:\"current,omitempty\"`// Current executing state (task index, progress)\n Error string `json:\"error,omitempty\"` // Error message if failed\n \n // Trigger input\n Input *TriggerInput `json:\"input,omitempty\"` // Original trigger input\n \n // Phase outputs (P0-P5)\n Inspiration *InspirationReport `json:\"inspiration,omitempty\"` // P0 result\n Goals *Goals `json:\"goals,omitempty\"` // P1 result\n Tasks []Task `json:\"tasks,omitempty\"` // P2 result\n Results []TaskResult `json:\"results,omitempty\"` // P3 results\n Delivery *DeliveryResult `json:\"delivery,omitempty\"` // P4 result\n Learning []LearningEntry `json:\"learning,omitempty\"` // P5 entries\n \n // Timestamps\n StartTime *time.Time `json:\"start_time,omitempty\"`\n EndTime *time.Time `json:\"end_time,omitempty\"`\n CreatedAt *time.Time `json:\"created_at,omitempty\"`\n UpdatedAt *time.Time `json:\"updated_at,omitempty\"`\n}\n\n// CurrentState - current executing state (for JSON storage)\ntype CurrentState struct {\n TaskIndex int `json:\"task_index\"` // index in Tasks slice\n Progress string `json:\"progress,omitempty\"` // human-readable progress (e.g., \"2/5 tasks\")\n}\n```\n\n**Store Implementation:**\n\n```go\n// store/execution.go\ntype ExecutionStore struct {\n modelID string // \"__yao.agent.execution\"\n}\n\nfunc NewExecutionStore() *ExecutionStore\n\n// Save creates or updates an execution record\nfunc (s *ExecutionStore) Save(ctx context.Context, record *ExecutionRecord) error\n\n// Get retrieves an execution by execution_id\nfunc (s *ExecutionStore) Get(ctx context.Context, executionID string) (*ExecutionRecord, error)\n\n// List retrieves executions with filters\nfunc (s *ExecutionStore) List(ctx context.Context, opts *ListOptions) ([]*ExecutionRecord, error)\n\n// UpdatePhase updates the current phase and its data\nfunc (s *ExecutionStore) UpdatePhase(ctx context.Context, executionID string, phase Phase, data interface{}) error\n\n// UpdateStatus updates the execution status\nfunc (s *ExecutionStore) UpdateStatus(ctx context.Context, executionID string, status ExecStatus, errorMsg string) error\n\n// UpdateCurrent updates the current executing state\nfunc (s *ExecutionStore) UpdateCurrent(ctx context.Context, executionID string, current *CurrentState) error\n\n// Delete removes an execution record\nfunc (s *ExecutionStore) Delete(ctx context.Context, executionID string) error\n\n// Conversion helpers\nfunc FromExecution(exec *Execution) *ExecutionRecord\nfunc (r *ExecutionRecord) ToExecution() *Execution\n\ntype ListOptions struct {\n MemberID string // Filter by robot member ID (globally unique)\n TeamID string // Filter by team\n Status ExecStatus // Filter by status\n TriggerType TriggerType // Filter by trigger\n Limit int // Max records to return (default: 100)\n Offset int // Skip records for pagination\n OrderBy string // e.g., \"start_time desc\"\n}\n```\n" }, { "path": "agent/robot/TODO.md", "content": "# Robot Agent - Implementation TODO\n\n> Based on DESIGN.md and TECHNICAL.md\n> Test environment: `source yao/env.local.sh`\n> Test assistants: `yao-dev-app/assistants/robot/`\n\n---\n\n## Workflow: Human-AI Collaboration\n\n**Important:** Follow this workflow strictly for each sub-task.\n\n```\n┌─────────────────────────────────────────────────────────────────┐\n│ Implementation Workflow │\n├─────────────────────────────────────────────────────────────────┤\n│ 1. AI: Implement code for current sub-task │\n│ 2. AI: Present code for review (DO NOT write tests yet) │\n│ 3. Human: Review code, provide feedback │\n│ 4. AI: Iterate based on feedback │\n│ 5. Human: Confirm \"LGTM\" or \"Approved\" │\n│ 6. AI: Write tests for the approved code │\n│ 7. Human: Review tests │\n│ 8. AI: Run tests, fix if needed │\n│ 9. Human: Confirm sub-task complete, move to next │\n└─────────────────────────────────────────────────────────────────┘\n```\n\n**Rules:**\n\n| Rule | Description |\n| ------------------------ | ------------------------------------------ |\n| One sub-task at a time | Focus only on current sub-task |\n| No tests before approval | Wait for human \"LGTM\" before writing tests |\n| No jumping ahead | Do not implement future phases |\n| Ask if unclear | When in doubt, ask before proceeding |\n\n---\n\n## Core Principle\n\n- Phase 1-2: Types + Skeleton (code compiles)\n- Phase 3: Complete scheduling system (Cache + Pool + Trigger + Dedup + Job), executor is stub\n- Phase 4-9: Implement executor phases one by one (P0 → P5)\n- Phase 10: API completion, end-to-end tests\n- Monitoring: Provided by Job system, no separate implementation\n\n---\n\n## Phase 1: Types & Interfaces ✅\n\n**Goal:** Define all types, enums, interfaces. No logic, no external deps.\n\n**Status:** Complete - 88.4% test coverage, all tests passing\n\n### 1.1 Enums (`types/enums.go`)\n\n- [x] `Phase` - execution phases (inspiration, goals, tasks, run, delivery, learning)\n- [x] `ClockMode` - clock trigger modes (times, interval, daemon)\n- [x] `TriggerType` - trigger sources (clock, human, event)\n- [x] `ExecStatus` - execution status (pending, running, completed, failed, cancelled)\n- [x] `RobotStatus` - robot status (idle, working, paused, error, maintenance)\n- [x] `InterventionAction` - human actions (task.add, goal.adjust, etc.)\n- [x] `Priority` - priority levels (high, normal, low)\n- [x] `DeliveryType` - delivery types (email, webhook, process, notify)\n- [x] `DedupResult` - dedup results (skip, merge, proceed)\n- [x] `EventSource` - event sources (webhook, database)\n- [x] `LearningType` - learning types (execution, feedback, insight)\n- [x] `TaskSource` - task sources (auto, human, event)\n- [x] `ExecutorType` - executor types (assistant, mcp, process)\n- [x] `TaskStatus` - task status (pending, running, completed, failed, skipped, cancelled)\n- [x] `InsertPosition` - insert positions (first, last, next, at)\n\n### 1.2 Context (`types/context.go`)\n\n- [x] `Context` struct - robot execution context\n- [x] `NewContext()` - constructor\n- [x] `UserID()`, `TeamID()` - helper methods\n\n### 1.3 Config Types (`types/config.go`)\n\n- [x] `Config` - main config struct\n- [x] `Triggers`, `TriggerSwitch` - trigger enable/disable\n- [x] `Clock` - clock config with validation\n- [x] `Identity` - role, duties, rules\n- [x] `Quota` - concurrency limits with defaults\n- [x] `KB`, `DB` - knowledge base and database config\n- [x] `Learn` - learning config\n- [x] `Resources`, `MCPConfig` - available agents and tools\n- [x] `Delivery` - output delivery config\n- [x] `Event` - event trigger config\n\n### 1.4 Core Types (`types/robot.go`)\n\n- [x] `Robot` struct - runtime robot representation\n- [x] `Robot` methods - `CanRun()`, `RunningCount()`, `AddExecution()`, `RemoveExecution()`, `GetExecution()`, `GetExecutions()`\n- [x] `Execution` struct - single execution instance\n- [x] `TriggerInput` - stored trigger input\n- [x] `CurrentState` - current executing state\n- [x] `Goals` - P1 output (markdown)\n- [x] `Task` - planned task (structured)\n- [x] `TaskResult` - task execution result\n- [x] `DeliveryResult` - delivery output\n- [x] `LearningEntry` - knowledge to save\n\n### 1.5 Clock Context (`types/clock.go`)\n\n- [x] `ClockContext` struct - time context for P0\n- [x] `NewClockContext()` - constructor\n\n### 1.6 Inspiration (`types/inspiration.go`)\n\n- [x] `InspirationReport` struct - P0 output\n\n### 1.7 Request/Response (`types/request.go`)\n\n- [x] `InterveneRequest` - human intervention request\n- [x] `EventRequest` - event trigger request\n- [x] `ExecutionResult` - trigger result\n- [x] `RobotState` - robot status query result\n\n### 1.8 Interfaces (`types/interfaces.go`)\n\n- [x] `Manager` interface\n- [x] `Executor` interface\n- [x] `Pool` interface\n- [x] `Cache` interface\n- [x] `Dedup` interface\n- [x] `Store` interface\n\n### 1.9 Errors (`types/errors.go`)\n\n- [x] Config errors\n- [x] Runtime errors\n- [x] Phase errors\n\n### 1.10 Tests\n\n- [x] `types/enums_test.go` - enum validation\n- [x] `types/config_test.go` - config validation\n- [x] `types/clock_test.go` - clock context creation\n- [x] `types/robot_test.go` - robot methods\n\n---\n\n## Phase 2: Skeleton Implementation ✅\n\n**Goal:** Create all packages with empty/stub implementations. Code compiles.\n\n**Status:** Complete - All packages compile successfully, no circular dependencies\n\n### 2.1 Utils (`utils/`) ✅\n\n- [x] `utils/convert.go` - JSON, map, struct conversions (implement)\n- [x] `utils/time.go` - time parsing, formatting, timezone (implement)\n- [x] `utils/id.go` - ID generation (nanoid) (implement)\n- [x] `utils/validate.go` - validation helpers (implement)\n- [x] Test: `utils/utils_test.go`\n\n### 2.2 Package Skeletons ✅ (stubs only, implemented in Phase 3)\n\nCreate empty structs and stub methods that return nil/empty/success:\n\n- [x] `cache/cache.go` - Cache struct, stub methods\n- [x] `dedup/dedup.go` - Dedup struct, stub methods\n- [x] `store/store.go` - Store struct, stub methods\n- [x] `pool/pool.go` - Pool struct, stub methods\n- [x] `plan/plan.go` - Plan struct, stub methods\n- [x] `trigger/trigger.go` - trigger dispatcher stub\n- [x] `executor/executor.go` - Executor struct, stub `Execute()`\n- [x] `manager/manager.go` - Manager struct, stub methods\n\n### 2.3 API Skeletons ✅\n\n- [x] `api/api.go` - Go API facade (all function signatures, return errors)\n- [x] `api/process.go` - Yao Process registration (all processes, return errors)\n- [x] `api/jsapi.go` - JSAPI registration (all methods, return errors)\n\n### 2.4 Root ✅\n\n- [x] `robot.go` - package entry\n - [x] `Init()` - placeholder\n - [x] `Shutdown()` - placeholder\n\n### 2.5 Compile Test ✅\n\n- [x] All packages compile without errors\n- [x] All imports resolve correctly\n- [x] No circular dependencies\n\n---\n\n## Phase 3: Complete Scheduling System ✅\n\n**Goal:** Implement complete scheduling system. Executor is stub (simulates success).\n\n**Status:** Complete - All 7 sub-tasks done, 80+ integration tests passing\n\nThis phase delivers a fully working scheduling pipeline:\n\n```\nTrigger → Manager → Cache → Dedup → Pool → Worker → Executor(stub) → Job\n```\n\n### ✅ 3.1 Cache Implementation (COMPLETE)\n\n- [x] `cache/cache.go` - Cache struct with thread-safe map\n- [x] `cache/load.go` - load robots from `__yao.member` where `member_type='robot'` and `autonomous_mode=true`\n - [x] Implemented pagination (100 robots per page)\n - [x] Configurable model name via `SetMemberModel()`\n- [x] `cache/refresh.go` - refresh single robot, periodic full refresh (every hour)\n- [x] Test: load/refresh with real DB\n - [x] Created comprehensive integration tests with real database\n - [x] Tests cover Load, LoadByID, Refresh, ListByTeam, GetByStatus\n - [x] All tests passing with proper cleanup\n\n### ✅ 3.2 Pool Implementation (COMPLETE)\n\n- [x] `pool/pool.go` - worker pool with configurable size (global limit)\n - [x] Default config: 10 workers, 100 queue size\n - [x] Configurable via `pool.NewWithConfig()`\n- [x] `pool/queue.go` - priority queue (sorted by: robot priority, trigger type, wait time)\n - [x] Two-level limit: global queue + per-robot queue\n - [x] Priority: Robot Priority × 1000 + Trigger Priority × 100\n- [x] `pool/worker.go` - worker goroutines, dispatch to executor\n - [x] Non-blocking quota check with re-enqueue\n - [x] Graceful shutdown support\n- [x] Test: submit jobs, verify execution order, verify concurrency limits\n - [x] 15 test cases covering all edge cases\n - [x] All tests passing\n\n### ✅ 3.3 Manager Implementation (COMPLETE)\n\n> **Note:** Manager is the scheduling core, depends on completed Cache and Pool.\n\n- [x] `manager/manager.go` - Manager struct\n - [x] `Start()` - load cache, start pool, start ticker goroutine\n - [x] `Stop()` - graceful shutdown (wait for running, drain queue)\n - [x] `Tick()` - main loop:\n 1. Get all cached robots\n 2. For each robot with clock trigger enabled\n 3. Check if should execute (times/interval/daemon modes)\n 4. Submit to pool\n - [x] `TriggerManual()` - manual trigger for testing/API\n - [x] Clock modes: times, interval, daemon\n - [x] Day matching for times mode\n - [x] Timezone handling\n - [x] Skip paused/error/maintenance robots\n- [x] Test: manager start/stop, tick cycle, manual trigger, clock modes, goroutine leak\n\n### ✅ 3.4 Trigger Implementation (COMPLETE)\n\n- [x] `trigger/trigger.go` - validation and helper functions\n - [x] `ValidateIntervention()` - validate human intervention requests\n - [x] `ValidateEvent()` - validate event trigger requests\n - [x] `BuildEventInput()` - build TriggerInput from event request\n - [x] `GetActionCategory()` / `GetActionDescription()` - action helpers\n- [x] `trigger/clock.go` - ClockMatcher for clock trigger matching\n - [x] `times` mode: match specific times (09:00, 14:00)\n - [x] `interval` mode: run every X duration (30m, 1h)\n - [x] `daemon` mode: restart immediately after completion\n - [x] Timezone handling\n - [x] Day-of-week filtering\n- [x] `trigger/control.go` - ExecutionController for pause/resume/stop\n - [x] Track/Untrack executions\n - [x] Pause/Resume execution\n - [x] Stop execution (cancel context)\n - [x] WaitIfPaused() for executor integration\n- [x] `manager/manager.go` - integrated trigger handling\n - [x] `Intervene()` - human intervention handler\n - [x] `HandleEvent()` - event trigger handler\n - [x] `PauseExecution()` / `ResumeExecution()` / `StopExecution()`\n - [x] `ListExecutions()` / `ListExecutionsByMember()`\n- [x] Tests: `trigger/trigger_test.go`, `trigger/clock_test.go`, `trigger/control_test.go`\n - [x] Validation tests for intervention and event requests\n - [x] Clock matching tests for all modes\n - [x] ExecutionController lifecycle tests\n - [x] Manager integration tests for Intervene/HandleEvent\n\n### ✅ 3.5 Execution Storage (COMPLETE)\n\n- [x] ExecutionStore - execution record persistence\n - [x] Execution data stored in `__yao.agent_execution` table\n - [x] All phase outputs (Inspiration, Goals, Tasks, Results, Delivery, Learning)\n - [x] Status and phase tracking\n - [x] Logging via `kun/log` package\n - [x] Localization support (en-US, zh-CN)\n- [x] Test: execution storage, status tracking\n - [x] `store/execution_test.go` - execution store tests\n - [x] All tests passing with real database\n\n### ✅ 3.6 Executor Architecture (COMPLETE)\n\nPluggable executor architecture with multiple execution modes:\n\n```\nexecutor/\n├── types/\n│ ├── types.go # Executor interface, Config types\n│ └── helpers.go # Shared helper functions\n├── standard/\n│ ├── executor.go # Real Agent execution (production)\n│ ├── agent.go # AgentCaller for LLM calls\n│ ├── input.go # InputFormatter for prompts\n│ ├── inspiration.go # P0: Inspiration phase\n│ ├── goals.go # P1: Goals phase\n│ ├── tasks.go # P2: Tasks phase\n│ ├── run.go # P3: Run phase\n│ ├── delivery.go # P4: Delivery phase\n│ └── learning.go # P5: Learning phase\n├── dryrun/\n│ └── executor.go # Simulated execution (testing/demo)\n├── sandbox/\n│ └── executor.go # Container-isolated (NOT IMPLEMENTED)\n└── executor.go # Factory functions\n```\n\n**Execution Modes:**\n\n| Mode | Use Case | Status |\n| -------- | -------------------------------- | ------------------ |\n| Standard | Production with real Agent calls | ✅ Implemented |\n| DryRun | Tests, demos, scheduling tests | ✅ Implemented |\n| Sandbox | Container-isolated execution | ⬜ Not Implemented |\n\n> **⚠️ Sandbox Mode:** Requires container-level isolation (Docker/gVisor/Firecracker)\n> for true security. Current placeholder behaves like DryRun. Future feature.\n\n- [x] `executor/types/types.go` - `Executor` interface, `PhaseExecutor` interface\n- [x] `executor/types/helpers.go` - `BuildTriggerInput()` shared helper\n- [x] `executor/executor.go` - Factory functions (`New`, `NewDryRun`, `NewWithMode`)\n- [x] `executor/standard/executor.go` - Real execution with Job integration\n- [x] `executor/standard/phases.go` - Phase implementations (P0-P5)\n- [x] `executor/dryrun/executor.go` - Simulated execution with callbacks\n- [x] `executor/sandbox/executor.go` - Placeholder (NOT IMPLEMENTED)\n- [x] Manager integration - accepts `Executor` interface via config\n- [x] Tests use DryRun mode for scheduling/concurrency tests\n\n### 3.7 Integration Test (End-to-End Scheduling) ✅\n\n- [x] Create test robot in `__yao.member` with clock config\n- [x] Start manager\n- [x] Wait for clock trigger\n- [x] Verify:\n - [x] Robot loaded to cache\n - [x] Clock trigger matched\n - [x] Job submitted to pool\n - [x] Worker picked up job\n - [x] Executor stub called\n - [x] Job execution recorded\n - [x] Logs written\n- [x] Test human intervention trigger\n- [x] Test event trigger\n- [x] Test concurrent executions (multiple robots)\n- [x] Test quota enforcement (per-robot limit)\n- [x] Test pause/resume/stop\n\n**Test Files Created:**\n\n- `manager/integration_test.go` - Core scheduling flow (Cache→Pool→Executor)\n- `manager/integration_clock_test.go` - Clock trigger modes (times/interval/daemon)\n- `manager/integration_human_test.go` - Human intervention trigger tests\n- `manager/integration_event_test.go` - Event trigger tests\n- `manager/integration_concurrent_test.go` - Concurrent execution & quota tests\n- `manager/integration_control_test.go` - Pause/Resume/Stop tests\n\n**Test Coverage:**\n\n- 27 top-level test functions\n- 80+ sub-tests covering all verification points\n- 3x run stability verified\n\n---\n\n## Phase 4: Agent Call Infrastructure ✅\n\n**Goal:** Implement unified Agent/Assistant calling mechanism. This is the foundation for all phase implementations (P0-P5).\n\n**Architecture Note:**\n\n- **Prompt construction is handled by Assistant layer** (`prompts.yml` in each assistant)\n- **Executor only prepares input data** (ClockContext, InspirationReport, etc.) and calls Assistant\n- **Assistant framework handles** prompt rendering, LLM API calls, streaming\n\n**Implemented:**\n\n1. A unified way to call assistants with streaming support\n2. Input data formatting for each phase\n3. Response parsing (markdown and structured data via `gou/text`)\n4. Multi-turn conversation support\n\n### 4.1 Agent Caller Implementation ✅\n\n- [x] `executor/agent.go` - `AgentCaller` struct with `SkipOutput`, `SkipHistory`, `SkipSearch`, `ChatID`\n- [x] `executor/agent.go` - `Call(ctx, assistantID, messages)` - basic call with full response\n- [x] `executor/agent.go` - `CallWithMessages(ctx, assistantID, userContent)` - convenience method\n- [x] `executor/agent.go` - `CallWithSystemAndUser(ctx, assistantID, systemContent, userContent)`\n- [x] `executor/agent.go` - handle assistant not found error\n- [x] `executor/agent.go` - handle LLM API errors gracefully\n- [x] `executor/agent.go` - `CallResult.GetJSON()` / `GetJSONArray()` - parse JSON response using `gou/text`\n- [x] `executor/agent.go` - `Conversation` struct for multi-turn dialogues\n- [x] `executor/agent.go` - `Conversation.Turn()`, `RunUntil()`, `Reset()`, `WithSystemPrompt()`\n- [x] `executor/agent.go` - Use `agentcontext.Noop()` logger to suppress debug output\n\n### 4.2 Input Formatters ✅\n\n- [x] `executor/input.go` - `FormatClockContext(clockCtx, robot)` - format clock context as message content\n- [x] `executor/input.go` - `FormatInspirationReport(report)` - format P0 output for P1 input\n- [x] `executor/input.go` - `FormatTriggerInput(input)` - format Human/Event trigger for P1 input\n- [x] `executor/input.go` - `FormatGoals(goals, robot)` - format P1 output for P2 input\n- [x] `executor/input.go` - `FormatTasks(tasks)` - format P2 output for P3 input\n- [x] `executor/input.go` - `FormatTaskResults(results)` - format P3 output for P4/P5 input\n- [x] `executor/input.go` - `FormatExecutionSummary(exec)` - format full execution for P5 input\n- [x] `executor/input.go` - `BuildMessages()`, `BuildMessagesWithSystem()` - helper methods\n\n### 4.3 Test Assistants ✅\n\n- [x] `yao-dev-app/assistants/tests/robot-single/` - Single-turn test assistant\n- [x] `yao-dev-app/assistants/tests/robot-conversation/` - Multi-turn conversation test assistant\n\n### 4.4 Tests ✅\n\n- [x] `executor/agent_test.go` - 22 test cases for AgentCaller and Conversation\n- [x] `executor/input_test.go` - 20 test cases for InputFormatter\n- [x] Verify: assistant can be called and returns response\n- [x] Verify: multi-turn conversation maintains state\n- [x] Verify: input data is well-formatted for assistant prompts\n- [x] Verify: JSON/YAML extraction from LLM output works correctly\n\n---\n\n## Phase 5: Test Scenario & Assistants Setup ✅\n\n**Goal:** Create realistic test scenarios with all required assistants.\n\n**Architecture:**\n\n```\n┌─────────────────────────────────────────────────────────────────────────────┐\n│ 6 Generic Phase Agents (P0-P5) │\n├─────────────────────────────────────────────────────────────────────────────┤\n│ inspiration │ goals │ tasks │ validation │ delivery │ learning │\n│ (P0) │ (P1) │ (P2) │ (P3) │ (P4) │ (P5) │\n└───────────────┴─────────┴─────────┴──────────────┴────────────┴─────────────┘\n ↓ P2 assigns tasks to\n┌─────────────────────────────────────────────────────────────────────────────┐\n│ Expert Agents (Task Executors) │\n├─────────────────────────────────────────────────────────────────────────────┤\n│ text-writer │ web-reader │ data-analyst │ summarizer │ ... │\n│ (Generate) │ (Fetch URL) │ (Analyze) │ (Summarize) │ │\n└───────────────┴──────────────┴────────────────┴──────────────┴─────────────┘\n```\n\n**Test Strategy:**\n\n- Phase Agents (P0-P5) are **generic** and reusable across all robot types\n- Expert Agents are **specialized** for specific tasks (text, web, data, etc.)\n- Each P0-P5 test uses **different expert combinations** to cover real scenarios\n- Tests use `interval: 1s` or `TriggerManual()` for easy triggering (no time dependency)\n\n### 5.1 Directory Structure\n\n```\nyao-dev-app/assistants/\n├── robot/ # Generic Phase Agents\n│ ├── inspiration/ # P0: Analyze clock context, generate insights\n│ │ ├── package.yao\n│ │ └── prompts.yml\n│ ├── goals/ # P1: Generate prioritized goals\n│ │ ├── package.yao\n│ │ └── prompts.yml\n│ ├── tasks/ # P2: Split goals into executable tasks\n│ │ ├── package.yao\n│ │ └── prompts.yml\n│ ├── validation/ # P3: Validate task results\n│ │ ├── package.yao\n│ │ └── prompts.yml\n│ ├── delivery/ # P4: Format and deliver results\n│ │ ├── package.yao\n│ │ └── prompts.yml\n│ └── learning/ # P5: Summarize execution, extract insights\n│ ├── package.yao\n│ └── prompts.yml\n│\n└── experts/ # Expert Agents (Task Executors)\n ├── text-writer/ # Generate text content (reports, emails, summaries)\n │ ├── package.yao\n │ └── prompts.yml\n ├── web-reader/ # Fetch and parse web page content\n │ ├── package.yao\n │ └── prompts.yml\n ├── data-analyst/ # Analyze data, generate insights\n │ ├── package.yao\n │ └── prompts.yml\n └── summarizer/ # Summarize long text into key points\n ├── package.yao\n └── prompts.yml\n```\n\n### 5.2 Generic Phase Agents\n\n#### 5.2.1 Inspiration Agent (P0)\n\n- [x] `robot/inspiration/package.yao` - config with model, temperature\n- [x] `robot/inspiration/prompts.yml` - system prompt:\n - Input: Clock context (time, day, markers), robot identity\n - Output: Markdown report with Summary, Highlights, Opportunities, Risks\n - Style: Analytical, context-aware\n\n#### 5.2.2 Goals Agent (P1)\n\n- [x] `robot/goals/package.yao` - config\n- [x] `robot/goals/prompts.yml` - system prompt:\n - Input: Inspiration report OR trigger input (human/event)\n - Output: Prioritized goals in markdown (High/Normal/Low)\n - Style: Strategic, actionable\n\n#### 5.2.3 Tasks Agent (P2)\n\n- [x] `robot/tasks/package.yao` - config\n- [x] `robot/tasks/prompts.yml` - system prompt:\n - Input: Goals, available expert agents list\n - Output: Structured task list (JSON) with executor assignments\n - Style: Detailed, executable\n\n#### 5.2.4 Validation Agent (P3)\n\n- [x] `robot/validation/package.yao` - config\n- [x] `robot/validation/prompts.yml` - system prompt:\n - Input: Task result, expected outcome\n - Output: Validation result (pass/fail, issues, suggestions)\n - Style: Critical, thorough\n\n#### 5.2.5 Delivery Agent (P4)\n\n- [x] `robot/delivery/package.yao` - config\n- [x] `robot/delivery/prompts.yml` - system prompt:\n - Input: Full execution context (P0-P3 results)\n - Output: Formatted delivery content\n - Style: Clear, professional\n\n#### 5.2.6 Learning Agent (P5)\n\n- [x] `robot/learning/package.yao` - config\n- [x] `robot/learning/prompts.yml` - system prompt:\n - Input: Full execution summary\n - Output: Insights, patterns, improvement suggestions\n - Style: Reflective, insightful\n\n### 5.3 Expert Agents (Task Executors)\n\n#### 5.3.1 Text Writer\n\n- [x] `experts/text-writer/package.yao` - config\n- [x] `experts/text-writer/prompts.yml` - system prompt:\n - Input: Topic, key points, style (formal/casual), length\n - Output: Generated text content\n - Use cases: Weekly reports, email drafts, summaries\n\n#### 5.3.2 Web Reader\n\n- [x] `experts/web-reader/package.yao` - config with hooks\n- [x] `experts/web-reader/prompts.yml` - system prompt:\n - Input: URL or topic to search\n - Output: Extracted content, key information\n - Use cases: News fetching, competitor monitoring, research\n- [x] `experts/web-reader/src/fetch.ts` - HTTP fetching utilities\n- [x] `experts/web-reader/src/fetch_test.ts` - 19 test cases (100% pass)\n- [x] `experts/web-reader/src/index.ts` - Create/Next hooks\n\n#### 5.3.3 Data Analyst\n\n- [x] `experts/data-analyst/package.yao` - config\n- [x] `experts/data-analyst/prompts.yml` - system prompt:\n - Input: Data description, analysis goal\n - Output: Analysis report, trends, insights\n - Use cases: Sales analysis, performance review\n\n#### 5.3.4 Summarizer\n\n- [x] `experts/summarizer/package.yao` - config\n- [x] `experts/summarizer/prompts.yml` - system prompt:\n - Input: Long text content\n - Output: Concise summary with key points\n - Use cases: Document summarization, meeting notes\n\n### 5.4 Test Scenarios\n\nEach phase test uses different expert combinations:\n\n| Test | Phase | Trigger | Expert Agents Used | Verification |\n| ---- | ----- | ---------------- | ------------------------ | ----------------------------- |\n| T1 | P0 | Clock (interval) | - | Clock → Inspiration report |\n| T2 | P1 | Clock | - | Inspiration → Goals |\n| T3 | P1 | Human | - | User input → Goals |\n| T4 | P2 | Clock | text-writer, web-reader | Goals → Tasks with executors |\n| T5 | P3 | Clock | text-writer | Task exec → Result validation |\n| T6 | P3 | Human | summarizer | Task exec → Result validation |\n| T7 | P4 | Clock | - | Results → Delivery format |\n| T8 | P5 | Clock | - | Full execution → Insights |\n| T9 | E2E | Clock | text-writer, summarizer | Full P0→P5 flow |\n| T10 | E2E | Human | web-reader, data-analyst | Full P1→P5 flow |\n\n### 5.5 Verification\n\n- [x] All 6 Phase Agents load correctly (`robot.inspiration`, `robot.goals`, etc.)\n- [x] All 4 Expert Agents load correctly (`experts.text-writer`, `experts.web-reader`, etc.)\n- [x] Web Reader `fetch.ts` utilities tested (19 tests, 100% pass)\n\n---\n\n## Phase 6: P0 Inspiration Implementation ✅\n\n**Goal:** Implement P0 (Inspiration Agent). Clock trigger → P0 → stub P1-P5.\n\n**Depends on:** Phase 4 (Agent Call Infrastructure), Phase 5 (Assistants Setup)\n\n**Status:** COMPLETED\n\n### 6.1 P0 Implementation\n\n- [x] `executor/inspiration.go` - `RunInspiration(ctx, exec, data)` - real implementation\n- [x] `executor/inspiration.go` - build prompt using `InputFormatter.FormatClockContext()`\n- [x] `executor/inspiration.go` - call Inspiration Agent using `AgentCaller`\n- [x] `executor/inspiration.go` - parse response to `InspirationReport` (markdown content)\n- [x] `types/robot.go` - added `GetRobot()`/`SetRobot()` methods for Execution\n- [x] `executor/executor.go` - set robot reference on execution creation\n\n### 6.2 Tests\n\n- [x] `executor/inspiration_test.go` - P0 with real LLM call (8 test cases)\n- [x] Test: clock context correctly formatted in prompt\n- [x] Test: robot identity included in prompt\n- [x] Test: markdown report generated with expected sections\n- [x] Test: handles LLM errors gracefully (robot nil, agent not found)\n- [x] Test: uses clock from trigger input or creates new one\n- [x] `InputFormatter.FormatClockContext()` unit tests (4 test cases)\n\n### 6.3 Notes\n\n- `executor_test.go` temporarily moved to `.bak` - will restore when all phases implemented\n- P0 uses `robot.inspiration` test agent from `yao-dev-app/assistants/robot/inspiration/`\n\n---\n\n## Phase 7: P1 Goals Implementation ✅\n\n**Goal:** Implement P1 (Goal Generation Agent). P0 → P1 → stub P2-P5.\n\n**Depends on:** Phase 6 (P0 Inspiration)\n\n**Status:** COMPLETED\n\n### 7.1 P1 Implementation\n\n- [x] `executor/goals.go` - `RunGoals(ctx, exec, data)` - real implementation\n- [x] `executor/goals.go` - build prompt with inspiration report (Clock trigger)\n- [x] `executor/goals.go` - build prompt with trigger input (Human/Event trigger)\n- [x] `executor/goals.go` - call Goals Agent using `AgentCaller`\n- [x] `executor/goals.go` - parse response to `Goals` struct (JSON with content + delivery)\n- [x] `executor/goals.go` - handle Human/Event trigger (skip P0, use input directly)\n- [x] `executor/goals.go` - include robot identity in prompt\n- [x] `executor/goals.go` - include available resources in prompt\n- [x] `executor/goals.go` - `ParseDelivery()` - parse delivery target from JSON\n- [x] `executor/goals.go` - `IsValidDeliveryType()` - validate delivery types\n\n### 7.2 Tests\n\n- [x] `executor/goals_test.go` - P1 with real LLM call (14 test cases)\n- [x] Test: inspiration report in prompt (Clock trigger)\n- [x] Test: user input in prompt (Human trigger)\n- [x] Test: event data in prompt (Event trigger)\n- [x] Test: goals markdown generated with priorities\n- [x] Test: delivery parsing from agent response\n- [x] Test: error handling (robot nil, agent not found, empty input)\n- [x] Test: fallback behavior (no inspiration → clock context)\n- [x] `ParseDelivery()` unit tests (8 test cases covering edge cases)\n- [x] `IsValidDeliveryType()` unit tests\n\n### 7.3 Notes\n\n- P1 uses `robot.goals` test agent from `yao-dev-app/assistants/robot/goals/`\n- Goals Agent returns JSON: `{ \"content\": \"...\", \"delivery\": {...} }`\n- Delivery is optional; if not present or invalid, `Goals.Delivery` is nil\n- Available resources (agents, MCP, KB, DB) are passed to agent for achievable goal generation\n\n---\n\n## Phase 8: P2 Tasks Implementation ✅\n\n**Goal:** Implement P2 (Task Planning Agent). P1 → P2 → stub P3-P5.\n\n**Depends on:** Phase 7 (P1 Goals)\n\n**Status:** COMPLETED\n\n### 8.1 Validation Agent Setup (Prerequisite for P3) ✅\n\n> **Note:** Validation Agent was already set up in Phase 5.\n\n- [x] `robot/validation/package.yao` - Validation Agent config (DeepSeek V3, temperature 0.2)\n- [x] `robot/validation/prompts.yml` - validation prompts\n - Input: Task result, expected outcome, validation rules\n - Output: Validation result (pass/fail, score, issues, suggestions)\n\n### 8.2 P2 Implementation ✅\n\n- [x] `executor/tasks.go` - `RunTasks(ctx, exec, data)` - real implementation\n- [x] `executor/tasks.go` - build prompt with goals (using `FormatGoals`)\n- [x] `executor/tasks.go` - include available tools/agents in prompt\n- [x] `executor/tasks.go` - include delivery target in prompt (for task output format)\n- [x] `executor/tasks.go` - call Tasks Agent using `AgentCaller`\n- [x] `executor/tasks.go` - parse response to `[]Task` (structured JSON)\n- [x] `executor/tasks.go` - validate task structure (executor type, ID, messages)\n- [x] `executor/tasks.go` - `ParseTasks()`, `ParseTask()`, `ParseMessages()` helpers\n- [x] `executor/tasks.go` - `SortTasksByOrder()` - ensure correct execution sequence\n- [x] `executor/tasks.go` - `ValidateExecutorExists()` - optional executor existence check\n- [x] `executor/tasks.go` - `ValidateTasksWithResources()` - validation with warnings\n- [x] `executor/input.go` - `FormatGoals()` updated to include Delivery Target\n\n### 8.3 Tests ✅\n\n- [x] `executor/tasks_test.go` - P2 with real LLM call (7 integration tests)\n- [x] Test: goals included in prompt\n- [x] Test: available tools listed in prompt\n- [x] Test: delivery target included in prompt\n- [x] Test: structured tasks generated\n- [x] Test: each task has valid executor type and ID\n- [x] Test: each task has expected output and validation rules\n- [x] `ParseTasks` unit tests (5 tests)\n- [x] `ValidateTasks` unit tests (5 tests)\n- [x] `SortTasksByOrder` unit tests (4 tests)\n- [x] `ValidateExecutorExists` unit tests (7 tests)\n- [x] `ValidateTasksWithResources` unit tests (3 tests)\n- [x] `ParseExecutorType` unit tests (5 tests)\n- [x] `IsValidExecutorType` unit tests (2 tests)\n- [x] `FormatGoals` with delivery target tests (4 tests)\n\n### 8.4 Notes\n\n- Tasks Agent returns JSON: `{ \"tasks\": [...] }`\n- Each task includes: id, executor_type, executor_id, messages, expected_output, validation_rules, order\n- Tasks are sorted by `order` field after parsing\n- Executor existence is optionally validated (warnings only, doesn't block)\n- Delivery target from P1 is passed to P2 so tasks can produce appropriate output format\n\n---\n\n## Phase 9: P3 Run Implementation ✅\n\n**Goal:** Implement P3 (Task Execution + Validation). P2 → P3 → stub P4-P5.\n\n**Depends on:** Phase 8 (P2 Tasks + Validation Agent)\n\n**Status:** Complete\n\n### 9.1 Implementation ✅\n\n- [x] `executor/run.go` - `RunExecution(ctx, exec, data)` - real implementation\n - [x] `RunConfig` - configuration (ContinueOnFailure, ValidationThreshold, MaxTurnsPerTask)\n - [x] Sequential task execution with progress tracking\n - [x] Task status updates (Running → Completed/Failed/Skipped)\n - [x] `ContinueOnFailure` option for graceful failure handling\n - [x] Previous task results passed as context to subsequent tasks\n- [x] `executor/runner.go` - `Runner` struct for task execution\n - [x] `ExecuteWithRetry()` - multi-turn conversation flow for assistant tasks\n - [x] `executeNonAssistantTask()` - single-call execution for MCP/Process\n - [x] `executeAssistantWithMultiTurn()` - AI assistant with conversation support\n - [x] `ExecuteMCPTask()` - MCP tool execution (format: `clientID.toolName`)\n - [x] `ExecuteProcessTask()` - Yao process execution\n - [x] `BuildTaskContext()` - context with previous results\n - [x] `BuildAssistantMessages()` - build messages for assistant\n - [x] `FormatPreviousResultsAsContext()` - format previous results as context\n - [x] `extractOutput()` - extract output from CallResult\n - [x] `generateDefaultReply()` - fallback reply generation\n- [x] `executor/validator.go` - Two-layer validation system\n - [x] Layer 1: Rule-based validation using `yao/assert`\n - [x] Layer 2: Semantic validation using Validation Agent\n - [x] `ValidateWithContext()` - validation with multi-turn support\n - [x] `isComplete()` - determine if expected result is obtained\n - [x] `checkNeedReply()` - determine if conversation should continue\n - [x] `generateFeedbackReply()` - generate validation feedback for next turn\n - [x] `detectNeedMoreInfo()` - detect if assistant needs clarification\n - [x] `convertStringRule()` - natural language rules to assertions\n - [x] `parseRules()` - JSON and string rule parsing\n - [x] `mergeResults()` - combine rule and semantic results\n\n### 9.2 Assert Package ✅\n\nCreated new `yao/assert` package for universal assertion/validation:\n\n- [x] `assert/types.go` - `Assertion`, `Result`, `AssertionOptions` types\n- [x] `assert/asserter.go` - `Asserter` with 8 assertion types:\n - [x] `equals` - exact match\n - [x] `contains` - substring check\n - [x] `not_contains` - negative substring check\n - [x] `json_path` - JSON path extraction and comparison\n - [x] `regex` - regex pattern matching\n - [x] `type` - type checking (with optional path)\n - [x] `script` - custom script validation\n - [x] `agent` - AI agent validation\n- [x] `assert/helpers.go` - `ValidateOutput()`, `ExtractPath()`, `ToString()`, `GetType()`\n- [x] `assert/asserter_test.go` - 98.7% test coverage\n\n### 9.3 Tests\n\n**Completed:**\n- [x] `assert/asserter_test.go` - 40+ test cases (98.7% coverage)\n- [x] `types/robot_test.go` - Task structure tests with validation rules\n- [x] `tasks_test.go` - ParseTasks with validation rules format\n- [x] Validation rules format aligned with `prompts.yml` guidelines\n\n**Completed Tests:**\n- [x] `executor/standard/run_test.go` - P3 RunExecution tests ✅\n - [x] Test: tasks executed in order (`TestRunExecutionBasic`)\n - [x] Test: task status updates (`TestRunExecutionTaskStatus`)\n - [x] Test: remaining tasks marked as skipped on failure\n - [x] Test: error handling (robot nil, no tasks, non-existent assistant)\n - [x] Test: rule-based and semantic validation (`TestRunExecutionValidation`)\n - [x] Test: previous results passed as context to subsequent tasks\n- [x] `executor/standard/runner_test.go` - Runner tests ✅\n - [x] Test: ExecuteWithRetry with multi-turn conversation flow\n - [x] Test: max turns limit enforcement\n - [x] Test: BuildTaskContext with previous results\n - [x] Test: FormatPreviousResultsAsContext formatting\n - [x] Test: BuildAssistantMessages with task content\n - [x] Test: FormatMessagesAsText (string, multipart, map)\n - [x] Test: MCP and Process tasks (skipped - requires runtime)\n- [x] `executor/standard/validator_test.go` - Validator tests ✅\n - [x] Test: ValidateWithContext with multi-turn state\n - [x] Test: isComplete determination logic\n - [x] Test: checkNeedReply scenarios\n - [x] Test: convertStringRule for natural language rules\n - [x] Test: parseRules for JSON assertions (equals, regex, json_path, type)\n - [x] Test: validateSemantic with Validation Agent\n - [x] Test: mergeResults logic (rule + semantic)\n\n**Completed:**\n- [x] Test: ContinueOnFailure option (run_test.go) ✅\n - [x] `stops_on_first_failure_when_ContinueOnFailure_is_false`\n - [x] `continues_execution_when_ContinueOnFailure_is_true`\n - [x] `multiple_failures_with_ContinueOnFailure`\n\n---\n\n## Phase 10: P4 Delivery Implementation\n\n**Goal:** Implement P4 (Delivery). P3 → P4 → stub P5.\n\n**Depends on:** Phase 9 (P3 Run)\n\n### 10.1 Execution Persistence (Prerequisite) ✅\n\n> **Background:** Each Robot execution (P0-P5) needs persistent storage for UI history queries.\n\n- [x] `yao/models/agent/execution.mod.yao` - Execution record model (`agent_execution` table)\n - [x] id, execution_id (unique)\n - [x] member_id (globally unique), team_id\n - [x] trigger_type (enum: clock, human, event)\n - [x] **Status tracking** (synced with runtime Execution):\n - [x] status (enum: pending, running, completed, failed, cancelled)\n - [x] phase (enum: inspiration, goals, tasks, run, delivery, learning)\n - [x] current (JSON) - current executing state (task_index, progress)\n - [x] error - error message if failed\n - [x] input (JSON) - trigger input\n - [x] **Phase outputs** (P0-P5):\n - [x] inspiration (JSON) - P0 output\n - [x] goals (JSON) - P1 output\n - [x] tasks (JSON) - P2 output\n - [x] results (JSON) - P3 output\n - [x] delivery (JSON) - P4 output\n - [x] learning (JSON) - P5 output\n - [x] **Timestamps**: start_time, end_time, created_at, updated_at\n - [x] Relations: member (hasOne __yao.member)\n- [x] `agent/robot/store/execution.go` - Execution record storage\n - [x] `Save(ctx, record)` - create or update execution record\n - [x] `Get(ctx, execID)` - get execution by ID\n - [x] `List(ctx, opts)` - query execution history with filters\n - [x] `UpdatePhase(ctx, execID, phase, data)` - update current phase and data\n - [x] `UpdateStatus(ctx, execID, status, error)` - update execution status\n - [x] `UpdateCurrent(ctx, execID, current)` - update current executing state\n - [x] `Delete(ctx, execID)` - delete execution record\n - [x] `FromExecution(exec, robotID)` - convert runtime Execution to record\n - [x] `ToExecution()` - convert record to runtime Execution\n- [x] Tests: `agent/robot/store/execution_test.go` (9 test groups, all passing)\n- [x] Integrate into Executor - call `UpdatePhase()` after each phase completes\n - [x] Added `SkipPersistence` config option to `executor/types/Config`\n - [x] Added `ExecutionStore` to `executor/standard/Executor`\n - [x] Save execution record at start of `Execute()`\n - [x] Call `UpdatePhase()` after each phase completes in `runPhase()`\n - [x] Call `UpdateStatus()` on status changes (running, completed, failed)\n\n### 10.2 Messenger Attachment Support ✅\n\n> **Conclusion:** All email providers now support attachments.\n\n**Implementation Status:**\n\n| Provider | Attachment Support | Implementation |\n|----------|-------------------|----------------|\n| Twilio/SendGrid | ✅ Supported | `buildAttachments()` - base64 encoded |\n| Mailgun | ✅ Supported | `sendEmailWithAttachments()` - multipart/form-data |\n| SMTP (mailer) | ✅ Supported | `buildMessageWithAttachments()` - MIME multipart/mixed |\n\n**Features Supported:**\n- Regular attachments (Content-Disposition: attachment)\n- Inline attachments (Content-Disposition: inline) with Content-ID for HTML embedding\n- Multiple attachments per email\n- Automatic content type detection\n- Base64 encoding for SMTP (RFC 2045 compliant, 76-char line wrapping)\n\n**Tests Added:**\n- `messenger/providers/mailgun/mailgun_test.go`:\n - `TestSend_EmailWithAttachments_MockServer`\n - `TestSend_EmailWithInlineAttachment_MockServer`\n - `TestSend_EmailWithAttachments_RealAPI`\n- `messenger/providers/mailer/mailer_test.go`:\n - `TestBuildMessage_WithAttachments` (single, multiple, inline, no attachments)\n - `TestSend_EmailWithAttachments_RealAPI`\n\n```go\n// messenger/types/types.go\ntype Attachment struct {\n Filename string `json:\"filename\"`\n ContentType string `json:\"content_type\"`\n Content []byte `json:\"content\"`\n Inline bool `json:\"inline,omitempty\"`\n CID string `json:\"cid,omitempty\"`\n}\n```\n\nSupported channels:\n- [x] Email - Full attachment support\n- [x] SMS - No attachment (text only)\n- [x] WhatsApp - TBD\n\n### 10.3 Type Updates (Prerequisite) ✅\n\n- [x] Update `types/enums.go` - Update `DeliveryType` enum\n - [x] Remove `DeliveryFile`\n - [x] Add `DeliveryProcess`\n- [x] Update `types/robot.go` - Delivery types for new architecture\n - [x] `DeliveryResult` - update to new structure (RequestID, Content, Results[])\n - [x] Add `DeliveryContent` struct\n - [x] Add `DeliveryAttachment` struct\n - [x] Add `DeliveryRequest` struct\n - [x] Add `DeliveryContext` struct\n - [x] Add `DeliveryPreferences` struct (with Email, Webhook, Process)\n - [x] Add `EmailPreference`, `EmailTarget` structs\n - [x] Add `WebhookPreference`, `WebhookTarget` structs\n - [x] Add `ProcessPreference`, `ProcessTarget` structs\n - [x] Add `ChannelResult` struct (with Target field)\n- [x] Update `types/enums_test.go` - Update DeliveryType tests\n- [x] Update `types/robot_test.go` - Update delivery result tests\n\n### 10.4 Delivery Agent Setup\n\n- [x] `robot/delivery/package.yao` - Delivery Agent config\n- [x] `robot/delivery/prompts.yml` - delivery prompts\n - [x] Input: Full execution context (P0-P3 results)\n - [x] Output: DeliveryContent (Summary, Body, Attachments) - **only content, no channels**\n - [x] Agent focuses on content generation, NOT channel selection\n\n### 10.5 Delivery Content Structure\n\n```go\n// DeliveryRequest - pushed to Delivery Center\n// No Channels - Delivery Center decides based on preferences\ntype DeliveryRequest struct {\n Content *DeliveryContent `json:\"content\"` // Agent-generated content\n Context *DeliveryContext `json:\"context\"` // Tracking info\n}\n\n// DeliveryContent - Content generated by Delivery Agent (only content)\ntype DeliveryContent struct {\n Summary string `json:\"summary\"` // Brief 1-2 sentence summary\n Body string `json:\"body\"` // Full markdown report\n Attachments []DeliveryAttachment `json:\"attachments,omitempty\"` // Output artifacts from P3\n}\n\n// DeliveryAttachment - Task output attachment with metadata\ntype DeliveryAttachment struct {\n Title string `json:\"title\"` // Human-readable title\n Description string `json:\"description,omitempty\"` // What this artifact is\n TaskID string `json:\"task_id,omitempty\"` // Which task produced this\n File string `json:\"file\"` // Wrapper: __://\n}\n\n// DeliveryContext - tracking info\ntype DeliveryContext struct {\n MemberID string `json:\"member_id\"` // Robot member ID (globally unique)\n ExecutionID string `json:\"execution_id\"`\n TriggerType TriggerType `json:\"trigger_type\"` // clock | human | event\n TeamID string `json:\"team_id\"`\n}\n```\n\n**Key Design:**\n- **Agent only generates content** (Summary, Body, Attachments)\n- **Delivery Center decides channels** based on Robot/User preferences\n- If webhook configured, every execution pushes automatically\n\n**File Wrapper:**\n- Format: `__://`\n- Parse: `attachment.Parse(value)` → `(uploader, fileID, isWrapper)`\n- Read: `attachment.Base64(ctx, value)` → base64 content\n\n**Delivery Channels (each supports multiple targets):**\n| Channel | Description | Multiple Targets |\n|---------|-------------|------------------|\n| `email` | Send via yao/messenger | ✅ Multiple recipients |\n| `webhook` | POST to external URL | ✅ Multiple URLs |\n| `process` | Yao Process call | ✅ Multiple processes |\n| `notify` | In-app notification | Future (auto by subscriptions) |\n\n### 10.6 Implementation\n\n**P4 Entry (executor/delivery.go):**\n- [x] `RunDelivery(ctx, exec, data)` - P4 entry point\n - [x] Call Delivery Agent to generate content (only content, no channels)\n - [x] Build DeliveryRequest (Content + Context)\n - [x] Push to Delivery Center\n - [x] Store DeliveryResult in exec.Delivery\n\n**Delivery Center (executor/delivery_center.go):**\n- [x] `DeliveryCenter.Deliver(ctx, request)` - main entry\n - [x] Read Robot/User delivery preferences\n - [x] Iterate through all enabled targets for each channel\n - [x] Aggregate ChannelResults into DeliveryResult\n\n**Channel Handlers (each supports multiple targets):**\n- [x] `sendEmail()` - uses yao/messenger\n - [x] Convert DeliveryAttachment to messenger.Attachment\n - [x] Support multiple EmailTarget\n - [x] Support custom subject_template per target\n - [x] Use `Robot.RobotEmail` as From address (if configured)\n - [x] Use global `DefaultEmailChannel()` for messenger channel selection\n- [x] `postWebhook()` - POST JSON\n - [x] POST DeliveryContent as JSON payload\n - [x] Support multiple WebhookTarget\n - [x] Support custom headers per target\n- [x] `callProcess()` - Yao Process call\n - [x] DeliveryContent as first arg\n - [x] Support multiple ProcessTarget\n - [x] Support additional args per target\n\n### 10.7 Tests\n\n- [x] `executor/delivery_test.go` - P4 delivery\n- [x] Test: Delivery Agent generates content (only content)\n- [x] Test: DeliveryCenter reads preferences\n- [x] Test: Multiple email targets (TestDeliveryCenterEmail)\n- [x] Test: Multiple webhook targets\n- [x] Test: Multiple process targets (TestDeliveryCenterProcess)\n- [x] Test: Mixed channels (email + webhook + process) (TestDeliveryCenterAllChannels)\n- [x] Test: sendEmail with attachments (TestDeliveryCenterEmail)\n- [x] Test: postWebhook with custom headers\n- [x] Test: callProcess with args (TestDeliveryCenterProcess)\n- [x] Test: Partial success (some targets fail)\n- [x] Test: DeliveryResult aggregation\n\n---\n\n## Phase 11: API & Integration (MVP)\n\n**Goal:** Complete Go API and end-to-end tests. Main flow: P0 → P1 → P2 → P3 → P4.\n\n**Depends on:** Phase 10 (P4 Delivery)\n\n> **Note:** Process handlers and JSAPI are optional wrappers, moved to Phase 12.\n> Go API is sufficient for MVP integration.\n\n### 11.1 Go API Implementation\n\n- [x] `api/types.go` - API request/response types\n- [x] `api/lifecycle.go` - manager lifecycle\n - [x] `Start()` / `Stop()` - manager lifecycle\n - [x] `StartWithConfig(config)` - start with custom config\n - [x] `IsRunning()` - check if system is running\n- [x] `api/robot.go` - robot query functions\n - [x] `GetRobot(memberID)` - get robot by member ID\n - [x] `ListRobots(query)` - list robots with filtering\n - [x] `GetRobotStatus(memberID)` - get robot runtime status\n- [x] `api/trigger.go` - trigger functions\n - [x] `Trigger(memberID, request)` - main trigger entry point\n - [x] `TriggerManual(memberID, triggerType, data)` - manual trigger for testing\n - [x] `Intervene(memberID, request)` - human intervention\n - [x] `HandleEvent(memberID, request)` - event trigger\n- [x] `api/execution.go` - execution query and control\n - [x] `GetExecution(execID)` - get execution by ID\n - [x] `ListExecutions(memberID, query)` - list executions\n - [x] `GetExecutionStatus(execID)` - get execution with runtime status\n - [x] `PauseExecution(execID)` / `ResumeExecution(execID)` / `StopExecution(execID)`\n- [x] `api/api.go` - package documentation\n- [x] Tests: `api/*_test.go` (black-box tests, 16 test cases)\n\n### 11.2 End-to-End Tests\n\n- [x] Full clock trigger flow (P0 → P1 → P2 → P3 → P4) - `e2e_clock_test.go`\n- [x] Human intervention flow (P1 → P2 → P3 → P4) - `e2e_human_test.go`\n- [x] Event trigger flow (P1 → P2 → P3 → P4) - `e2e_event_test.go`\n- [x] Concurrent execution test - `e2e_concurrent_test.go`\n- [x] Pause/Resume/Stop test - `e2e_control_test.go`\n\n---\n\n## Phase 12: OpenAPI Integration\n\n**Goal:** HTTP endpoints for Robot Agent management and triggers.\n\n**Depends on:** Phase 11 (API & Integration), Frontend UI Design\n\n> **Note:** This phase will be planned in detail after frontend UI design is complete.\n> The API endpoints will be designed based on actual UI requirements.\n\n### 12.1 Planned Features\n\n- [ ] HTTP endpoints for robot management (CRUD)\n- [ ] HTTP endpoints for human intervention triggers\n- [ ] Webhook endpoints for external events\n- [ ] WebSocket for real-time execution status updates\n- [ ] Authentication and authorization integration\n\n### 12.2 Design Dependencies\n\n- Frontend dashboard design (robot list, status, controls)\n- Execution history UI design\n- Human intervention UI design\n- Real-time notification requirements\n\n---\n\n## Phase 13: Advanced Features\n\n**Goal:** P5 Learning, Process/JSAPI wrappers, dedup, plan queue.\n\n> **Note:** These are optional features. Main flow works without them.\n\n### 13.1 Process & JSAPI Wrappers\n\n> **Note:** These are convenience wrappers around Go API for Yao ecosystem integration.\n\n- [ ] `api/process.go` - implement Process handlers\n - [ ] `robot.Start` / `robot.Stop`\n - [ ] `robot.Trigger` / `robot.Intervene` / `robot.HandleEvent`\n - [ ] `robot.Pause` / `robot.Resume` / `robot.Stop`\n - [ ] `robot.Get` / `robot.List`\n - [ ] `robot.GetExecution` / `robot.ListExecutions`\n- [ ] `api/jsapi.go` - implement JSAPI for JavaScript runtime\n- [ ] Tests for Process and JSAPI\n\n### 13.2 P5 Learning Implementation\n\n> **Background:** P5 Learning is async, runs after P4 Delivery completes.\n> User doesn't wait for it. Results stored in private KB for future reference.\n\n#### 13.2.1 Learning Agent Setup\n\n- [ ] `robot/learning/package.yao` - Learning Agent config\n- [ ] `robot/learning/prompts.yml` - learning prompts\n\n#### 13.2.2 Store Implementation\n\n- [ ] `store/store.go` - Store interface and struct\n- [ ] `store/kb.go` - KB operations (create, save, search)\n- [ ] `store/learning.go` - save learning entries to private KB\n\n#### 13.2.3 Implementation\n\n- [ ] `executor/learning.go` - `RunLearning(ctx, exec, data)` - real implementation\n- [ ] `executor/learning.go` - extract learnings from execution\n- [ ] `executor/learning.go` - call Learning Agent\n- [ ] `executor/learning.go` - save to private KB\n\n#### 13.2.4 Tests\n\n- [ ] `executor/learning_test.go` - P5 learning\n- [ ] Test: learnings extracted from execution\n- [ ] Test: learnings saved to KB\n- [ ] Test: KB can be queried for past learnings\n\n### 13.3 Fast Dedup (Time-Window)\n\n> **Note:** Manager has `// TODO: dedup check` comment placeholder. Integrate after implementation.\n\n- [ ] `dedup/dedup.go` - Dedup struct\n- [ ] `dedup/fast.go` - fast in-memory time-window dedup\n - [ ] Key: `memberID:triggerType:window`\n - [ ] Check before submit\n - [ ] Mark after submit\n- [ ] Integrate into Manager.Tick()\n- [ ] Test: dedup check/mark, window expiry\n\n### 13.4 Semantic Dedup\n\n- [ ] `dedup/semantic.go` - call Dedup Agent for goal/task level dedup\n- [ ] Dedup Agent setup (`assistants/robot/dedup/`)\n- [ ] Test: semantic dedup with real LLM\n\n### 13.5 Plan Queue\n\n- [ ] `plan/plan.go` - plan queue implementation\n - [ ] Store planned tasks/goals\n - [ ] Execute at next cycle or specified time\n- [ ] `plan/schedule.go` - schedule for later\n- [ ] Test: plan queue operations\n\n> **Note:** Monitoring is provided by Job system (Activity Monitor UI). No separate implementation needed.\n\n---\n\n## Test Assistants Structure\n\n```\nyao-dev-app/assistants/robot/\n├── inspiration/ # P0: Inspiration Agent\n│ ├── package.yao\n│ └── prompts.yml\n├── goals/ # P1: Goal Generation Agent\n│ ├── package.yao\n│ └── prompts.yml\n├── tasks/ # P2: Task Planning Agent\n│ ├── package.yao\n│ └── prompts.yml\n├── validation/ # P3: Validation Agent\n│ ├── package.yao\n│ └── prompts.yml\n├── delivery/ # P4: Delivery Agent\n│ ├── package.yao\n│ └── prompts.yml\n├── learning/ # P5: Learning Agent\n│ ├── package.yao\n│ └── prompts.yml\n└── dedup/ # Deduplication Agent\n ├── package.yao\n └── prompts.yml\n```\n\n---\n\n## Notes\n\n### Test Environment Setup\n\n1. **Environment Variables:** Run `source yao/env.local.sh` before tests\n2. **Test Preparation:** Use `testutils.Prepare(t)` to load config, KB, and agents\n\n```go\npackage robot_test\n\nimport (\n \"testing\"\n \"github.com/yaoapp/yao/agent/testutils\"\n)\n\nfunc TestExample(t *testing.T) {\n // Load environment config (from YAO_TEST_APPLICATION)\n // This loads: config, connectors, KB, agents, models, etc.\n testutils.Prepare(t)\n defer testutils.Clean(t)\n\n // Your test code here\n}\n```\n\n### Test Conventions\n\n1. **Black-box Tests:** All tests in `*_test` package (external package)\n2. **Real LLM Calls:** Use `gpt-4o` or `deepseek` connectors for agent tests\n3. **Incremental:** Each phase builds on previous, all tests must pass before next phase\n4. **No Skip:** Do NOT use `t.Skip()` except for `testing.Short()` (CI mode)\n5. **Must Assert:** Every test MUST have result validation assertions\n\n```go\nfunc TestWithLLM(t *testing.T) {\n // Only allowed Skip: testing.Short() for CI\n if testing.Short() {\n t.Skip(\"Skipping integration test\")\n }\n\n testutils.Prepare(t)\n defer testutils.Clean(t)\n\n // Your test code...\n result, err := SomeFunction()\n\n // MUST have assertions - no empty tests!\n assert.NoError(t, err)\n assert.NotNil(t, result)\n assert.Equal(t, expected, result.Field)\n}\n```\n\n### Test Rules\n\n| Rule | Description |\n| ----------------- | ----------------------------------------- |\n| No arbitrary Skip | Only `testing.Short()` skip allowed |\n| Must assert | Every test must validate results |\n| No empty tests | Tests without assertions will fail review |\n| Real calls | LLM tests use real API calls, not mocks |\n\n### Key Environment Variables\n\n| Variable | Description |\n| ---------------------- | ------------------------------- |\n| `YAO_TEST_APPLICATION` | Test app path (`yao-dev-app`) |\n| `OPENAI_TEST_KEY` | OpenAI API key |\n| `DEEPSEEK_API_KEY` | DeepSeek API key |\n| `YAO_DB_DRIVER` | Database driver (mysql/sqlite3) |\n| `YAO_DB_PRIMARY` | Database connection string |\n\n---\n\n## Progress Tracking\n\n| Phase | Status | Description |\n| --------------------- | ------ | ---------------------------------------------------------------------------- |\n| 1. Types & Interfaces | ✅ | All types, enums, interfaces |\n| 2. Skeleton | ✅ | Empty stubs, code compiles |\n| 3. Scheduling System | ✅ | Cache + Pool + Trigger + Job + Executor architecture |\n| 4. Agent Infra | ✅ | AgentCaller, InputFormatter, test assistants |\n| 5. Test Scenarios | ✅ | Phase agents (P0-P5), expert agents |\n| 6. P0 Inspiration | ✅ | Inspiration Agent integration |\n| 7. P1 Goals | ✅ | Goal Generation Agent integration |\n| 8. P2 Tasks | ✅ | Task Planning Agent integration |\n| 9. P3 Run | ✅ | Task execution + validation + yao/assert + multi-turn conversation |\n| 10. P4 Delivery | ✅ | Output delivery (email/webhook/process, notify future) |\n| 11. API & Integration | ✅ | Go API, end-to-end tests (main flow: P0→P1→P2→P3→P4) |\n| 12. OpenAPI | ⬜ | HTTP endpoints (depends on frontend UI design) |\n| 13. Advanced | ⬜ | Process/JSAPI, P5 Learning, dedup, plan queue, Sandbox |\n\nLegend: ⬜ Not started | 🟡 In progress | ✅ Complete\n\n**Main Flow (MVP):** P0 Inspiration → P1 Goals → P2 Tasks → P3 Run → P4 Delivery ✅\n**OpenAPI (Phase 12):** HTTP endpoints - planned after frontend UI design\n**Advanced (Phase 13):** P5 Learning (async), Process/JSAPI, Dedup, Plan Queue, Sandbox\n\n---\n\n## Quick Commands\n\n```bash\n# Setup environment\nsource yao/env.local.sh\n\n# Run all robot tests\ngo test -v ./agent/robot/...\n\n# Run specific phase tests\ngo test -v ./agent/robot/types/...\ngo test -v ./agent/robot/cache/...\ngo test -v ./agent/robot/pool/...\ngo test -v ./agent/robot/executor/...\n\n# Run with coverage\ngo test -cover ./agent/robot/...\n```\n" }, { "path": "agent/robot/V2-IMPROVEMENTS.md", "content": "# Robot Agent V2 — Improvement Plan\n\n> Generated: 2026-02-25\n> Based on: DESIGN-V2.md deep review against implementation code\n> Scope: Bug fixes, missing unit tests, code quality improvements\n\n---\n\n## Auth Context Clarification\n\nRobot is a legitimate team member in `__yao.member`. Auth is always present:\n\n| Trigger Path | Auth Source | Code |\n|-------------|------------|------|\n| Clock | `manager.buildRobotAuth(robot)` → `{UserID: robot.MemberID, TeamID: robot.TeamID}` | `manager.go:270` |\n| Human / Event | Caller's `ctx.Auth` passthrough from HTTP middleware | `openapi/agent/robot/*.go` |\n| Resume | Loaded from execution record → `buildRobotAuth` or caller passthrough | `executor.go:764+` |\n\nThe `openapi/agent/robot/` layer constructs `ctx := &robottypes.Context{}` without Auth — this is the **existing V1 pattern** across ALL openapi handlers (trigger.go, execution.go, list.go, etc). Auth checking is done via `authorized.GetInfo(c)` at the Gin middleware level; `robottypes.Context` is a downstream execution context.\n\n**However**, `manager/interact.go:createConfirmingExecution` calls `ctx.UserID()` which returns `\"\"` when openapi passes an empty Context. This is a V2-specific issue since V1 handlers don't need `ctx.UserID()`.\n\n---\n\n## 1. Bugs\n\n### BUG-1 [P0] `advanceExecution` discards confirmed Goals/Tasks\n\n**File**: `manager/interact.go:415-431`\n\n**Problem**: After multi-round Host Agent confirmation (which may have generated Goals and Tasks stored in `record.Goals` / `record.Tasks`), `advanceExecution()` submits to Pool via `m.pool.SubmitWithID(...)`. The Pool Worker then calls `ExecuteWithControl()` which starts from P1 (Goals) and re-generates everything — the confirmed plan is lost.\n\n**Design intent (§10.1)**: Confirmation → use confirmed Goals/Tasks → skip P1/P2 → directly execute P3.\n\n**Fix**: `advanceExecution` must inject `record.Goals` and `record.Tasks` into the `TriggerInput` or use a dedicated `Resume`-like path that skips P1/P2 when Goals/Tasks already exist.\n\n**Test required**:\n- Confirm with pre-existing Goals/Tasks → verify P3 uses those Goals/Tasks, not re-generated ones\n- Confirm without Goals/Tasks → verify normal P1→P2→P3 flow\n\n---\n\n### BUG-2 [P0] `standard.New()` creates orphan Executor instances\n\n**Files**: `manager/interact.go:501, 525, 544`\n\n**Problem**: `skipWaitingTask()`, `resumeWithContext()`, and `directResume()` all call `standard.New()`, creating a fresh Executor with independent counters. Consequences:\n1. `currentCount` / `execCount` not shared — monitoring inaccurate\n2. No `execController.Untrack()` after Resume completes — memory leak\n3. Separate `store` / `robotStore` instances (less critical, stateless)\n\n**Fix**: Manager should hold a reference to the live Executor (obtained via Pool) and expose a `Resume` method, or provide the Executor as a constructor parameter.\n\n**Tests required**:\n- Resume via `skipWaitingTask` → verify `execController.Untrack()` called\n- Resume via `resumeWithContext` → verify executor `currentCount` incremented/decremented correctly\n\n---\n\n### BUG-3 [P1] `buildRobotStatusSnapshot` returns near-empty snapshot\n\n**File**: `manager/interact.go:266-278`\n\n**Problem**: Only populates `ActiveCount` and `MaxQuota`. Missing: `WaitingCount`, `QueuedCount`, `ActiveExecs`, `RecentExecs`. Host Agent cannot make informed decisions about robot workload.\n\n**Fix**: Query `robot.Executions` to compute `WaitingCount`, collect `ActiveExecs` briefs, and optionally query recent completed executions from store.\n\n**Tests required**:\n- Robot with 2 running + 1 waiting execution → snapshot reflects correct counts\n- Robot with no executions → all counts zero\n\n---\n\n### BUG-4 [P1] `openapi/agent/robot/interact.go` passes empty Context to Manager\n\n**File**: `openapi/agent/robot/interact.go:67, 152, 209`\n\n**Problem**: `ctx := &robottypes.Context{}` — no `Auth`, no `context.Context`. When `HandleInteract` → `createConfirmingExecution` calls `ctx.UserID()`, returns `\"\"`. The `TriggerInput.UserID` in the DB record is empty.\n\n**Note**: This is NOT about Robot's own Auth (which is always set via `buildRobotAuth` in execution paths). This is about tracking **which human user** initiated the interaction.\n\n**Fix**: In V2 interact handlers, construct Context properly:\n```go\nctx := robottypes.NewContext(c.Request.Context(), &oauthtypes.AuthorizedInfo{\n UserID: authInfo.UserID,\n TeamID: authInfo.TeamID,\n})\n```\n\n**Tests required**:\n- InteractRobot handler → verify ctx.UserID() returns the authenticated user's ID\n- CreateConfirmingExecution → verify record.Input.UserID is populated\n\n---\n\n### BUG-5 [P2] `HostContext.Goals` type mismatch with design\n\n**File**: `types/host.go:15`\n\n**Problem**: Design §5.7 defines `Goals string`, implementation uses `*Goals` (struct with `Content` field). Host Agent receives `{\"goals\": {\"content\": \"...\"}}` instead of `{\"goals\": \"...\"}`.\n\n**Fix**: Either update the Host Agent prompt to expect the struct format, or flatten to `string` in `buildHostContext`:\n```go\nif record.Goals != nil {\n hostCtx.GoalsContent = record.Goals.Content // string\n}\n```\n\n**Tests required**:\n- `buildHostContext` with Goals → verify JSON output matches Host Agent prompt expectations\n\n---\n\n## 2. Missing Unit Tests\n\nAll tests should be **black-box** tests (test exported APIs only), must **verify return values and side effects**, and must **not require real LLM calls**.\n\n### 2.1 `executor/standard/host.go` — CallHostAgent\n\n**Current coverage**: 0 tests\n\n| # | Test Case | Verify |\n|---|-----------|--------|\n| H1 | Robot is nil | Returns error \"robot cannot be nil\" |\n| H2 | No Host Agent configured (empty Resources) | Returns error \"no Host Agent configured\" |\n| H3 | Valid Host Agent call returns JSON | Parsed `HostOutput` with correct Action and Reply |\n| H4 | Host Agent returns non-JSON text | Fallback to `HostActionConfirm` with text as Reply |\n| H5 | Host Agent returns invalid JSON structure | Fallback to `HostActionConfirm` |\n| H6 | Host Agent call fails (network error) | Returns wrapped error |\n| H7 | Input marshalling (verify HostInput fields) | Correct JSON sent to agent |\n\n**Status**: ✅ All tests implemented. H1-H2, H7 are pure unit tests. H3-H5 use real LLM integration via `yao-dev-app` test assistants (`tests.host-json`, `tests.host-plaintext`, `tests.host-badjson`). H6 uses real assistant framework.\n\n---\n\n### 2.2 `manager/interact.go` — processHostAction (all branches)\n\n**Current coverage**: 2/7 branches (WaitForMore, default)\n\n| # | Test Case | Action | Verify |\n|---|-----------|--------|--------|\n| PA1 | HostActionConfirm | `confirm` | `resp.Status == \"confirmed\"`, `advanceExecution` called |\n| PA2 | HostActionAdjust with goals data | `adjust` | Record Goals updated, `resp.Status == \"adjusted\"` |\n| PA3 | HostActionAdjust with tasks data | `adjust` | Record Tasks updated |\n| PA4 | HostActionAdjust with nil data | `adjust` | No error, noop |\n| PA5 | HostActionAddTask | `add_task` | New task appended to record.Tasks with generated ID |\n| PA6 | HostActionAddTask with nil data | `add_task` | Returns error \"task data is required\" |\n| PA7 | HostActionSkip with waiting task | `skip` | Task status = skipped |\n| PA8 | HostActionSkip without waiting task | `skip` | Returns error \"no task is waiting\" |\n| PA9 | HostActionInjectCtx with string reply | `inject_context` | Resume called with correct reply |\n| PA10 | HostActionInjectCtx → re-suspend | `inject_context` | `resp.Status == \"waiting\"` |\n| PA11 | HostActionCancel | `cancel` | Execution status = cancelled, event pushed |\n| PA12 | WaitForMore = true | — | `resp.Status == \"waiting_for_more\"`, `resp.WaitForMore == true` |\n| PA13 | Unknown action | — | `resp.Status == \"acknowledged\"` |\n\n**Note**: PA1, PA7, PA9, PA11 require mocking Executor.Resume and Pool.SubmitWithID.\n\n---\n\n### 2.3 `manager/interact.go` — HandleInteract routing\n\n**Current coverage**: Parameter validation only\n\n| # | Test Case | Verify |\n|---|-----------|--------|\n| HI1 | No execution_id → creates confirming execution | Record saved with status=confirming, Host Agent called with \"assign\" |\n| HI2 | execution_id with status=confirming | Host Agent called with \"assign\" scenario |\n| HI3 | execution_id with status=waiting | Host Agent called with \"clarify\" scenario |\n| HI4 | execution_id with status=running | Host Agent called with \"guide\" scenario |\n| HI5 | execution_id with status=completed | Returns error \"cannot interact\" |\n| HI6 | execution_id not found | Returns error \"execution not found\" |\n| HI7 | Host Agent unavailable → direct assign fallback | Execution started without Host Agent |\n| HI8 | Host Agent unavailable → direct resume fallback | Execution resumed directly |\n\n---\n\n### 2.4 `manager/interact.go` — CancelExecution\n\n**Current coverage**: \"manager not started\" only\n\n| # | Test Case | Verify |\n|---|-----------|--------|\n| CE1 | Cancel waiting execution | Status → cancelled, `Untrack` called, event pushed |\n| CE2 | Cancel confirming execution | Status → cancelled |\n| CE3 | Cancel running execution | Returns error (only waiting/confirming allowed) |\n| CE4 | Cancel non-existent execution | Returns error \"execution not found\" |\n| CE5 | Cancel already cancelled | Returns error |\n\n---\n\n### 2.5 `executor/standard/executor.go` — Resume method\n\n**Current coverage**: Only via E2E tests (requires real LLM)\n\n| # | Test Case | Verify |\n|---|-----------|--------|\n| R1 | Resume non-waiting execution | Returns error \"not in waiting status\" |\n| R2 | Resume non-existent execution | Returns error \"execution not found\" |\n| R3 | Resume with nil store | Returns error \"store is required\" |\n| R4 | Resume injects reply into task messages | `exec.Tasks[i].Messages` contains `[Human reply]` prefixed message |\n| R5 | Resume clears waiting fields | `WaitingTaskID`, `WaitingQuestion`, `WaitingSince` all empty after resume |\n| R6 | Resume updates status to running | `exec.Status == ExecRunning` |\n| R7 | Resume → re-suspend | Returns `ErrExecutionSuspended`, execution stays tracked |\n| R8 | Resume → complete → P4 → P5 | Status == ExecCompleted, `ResumeContext` cleared |\n| R9 | Resume → P3 error | Status == ExecFailed with error message |\n| R10 | Resume maintains executor currentCount | `currentCount +1 before, -1 after` |\n\n**Note**: R4-R10 require mocking store.Get, store.UpdateResumeState, RunExecution, runPhase.\n\n---\n\n### 2.6 `manager/interact.go` — Helper methods\n\n**Current coverage**: buildRobotStatusSnapshot (3), findWaitingTask (3), buildHostContext (2)\n\n| # | Missing Test Case | Verify |\n|---|-------------------|--------|\n| HL1 | `createConfirmingExecution` | Record has correct fields (execID, chatID, status=confirming, input) |\n| HL2 | `adjustExecution` with goals string | `record.Goals.Content` updated |\n| HL3 | `adjustExecution` with tasks array | `record.Tasks` replaced |\n| HL4 | `adjustExecution` with non-map data | Graceful handling |\n| HL5 | `injectTask` with valid task | Task appended with auto-generated ID |\n| HL6 | `injectTask` preserves existing tasks | len(tasks) == original + 1 |\n| HL7 | `callHostAgentForScenario` — no host agent | Returns error |\n| HL8 | `directAssign` | Returns \"confirmed\" status |\n| HL9 | `directResume` — re-suspend | Returns \"waiting\" status |\n| HL10 | `directResume` — complete | Returns \"resumed\" status |\n\n---\n\n### 2.7 `api/interact.go` — Interact/Reply/Confirm/CancelExecution\n\n**Current coverage**: 0 tests\n\n| # | Test Case | Verify |\n|---|-----------|--------|\n| AI1 | `Interact` with manager available | Delegates to `managerInteract` |\n| AI2 | `Interact` without manager, with execution_id | Delegates to `legacyResume` |\n| AI3 | `Interact` without manager, without execution_id | Returns error |\n| AI4 | `Interact` with empty member_id | Returns error |\n| AI5 | `Interact` with nil request | Returns error |\n| AI6 | `Reply` shortcut | Calls Interact with correct TaskID and Source |\n| AI7 | `Confirm` shortcut | Calls Interact with correct Action |\n| AI8 | `CancelExecution` with manager | Delegates correctly |\n| AI9 | `CancelExecution` without manager | Returns error |\n| AI10 | `legacyResume` → success | Returns \"resumed\" status |\n| AI11 | `legacyResume` → re-suspend | Returns \"waiting\" status |\n| AI12 | `legacyResume` → error | Returns wrapped error |\n\n---\n\n### 2.8 `events/events.go` + `events/handlers.go` — Event integration\n\n**Current coverage**: DeliveryHandler basic (3 tests)\n\n| # | Missing Test Case | Verify |\n|---|-------------------|--------|\n| EV1 | DeliveryHandler — payload deserialization | All fields (`ExecutionID`, `MemberID`, `Content`, `Preferences`) correctly parsed |\n| EV2 | Verify event constants match design §7.2 | All 9 constants present and correctly named |\n| EV3 | `NeedInputPayload` marshalling | Correct JSON roundtrip |\n| EV4 | `TaskPayload` marshalling | Correct JSON roundtrip with optional Error field |\n| EV5 | `ExecPayload` marshalling | Correct JSON roundtrip |\n\n---\n\n### 2.9 `openapi/agent/robot/interact.go` — HTTP handlers\n\n**Current coverage**: 0 tests\n\n| # | Test Case | Verify |\n|---|-----------|--------|\n| OH1 | `InteractRobot` — valid request | 200 with InteractResponse |\n| OH2 | `InteractRobot` — missing robot ID | 400 error |\n| OH3 | `InteractRobot` — missing message | 400 error |\n| OH4 | `InteractRobot` — robot not found | 404 error |\n| OH5 | `InteractRobot` — forbidden (no write permission) | 403 error |\n| OH6 | `ReplyToTask` — valid request | 200 with response |\n| OH7 | `ReplyToTask` — missing params | 400 error |\n| OH8 | `ConfirmExecution` — valid request | 200 with response |\n| OH9 | `ConfirmExecution` — empty body allowed | 200 (confirm without message) |\n\n---\n\n### 2.10 Event push verification in execution flow\n\n**Current coverage**: 0 (events are pushed but never verified in tests)\n\n| # | Test Case | File | Verify |\n|---|-----------|------|--------|\n| EP1 | Task completes → TaskCompleted event | `run.go:111` | Event type + payload fields |\n| EP2 | Task fails → TaskFailed event | `run.go:120` | Event type + error in payload |\n| EP3 | Execution suspends → ExecWaiting event | `executor.go:750` | Event type + question in payload |\n| EP4 | Execution resumes → ExecResumed event | `executor.go:856` | Event type + chatID |\n| EP5 | Execution completes → ExecCompleted event | `executor.go:287` | Event type + status |\n| EP6 | Execution cancelled → ExecCancelled event | `manager/interact.go:66` | Event type + status |\n| EP7 | Delivery → Delivery event | `delivery.go:102` | Content + Preferences in payload |\n\n**Approach**: Use `event.Subscribe` in test to capture pushed events, or mock `event.Push`.\n\n---\n\n## 3. Code Quality Improvements\n\n### CQ1 — Extract common Executor resume logic\n\n`skipWaitingTask`, `resumeWithContext`, `directResume` all duplicate: create executor → call Resume → handle ErrExecutionSuspended. Extract to a private helper:\n\n```go\nfunc (m *Manager) executeResume(ctx *types.Context, execID, reply string) error {\n // Use shared executor reference, not standard.New()\n return m.getExecutor().Resume(ctx, execID, reply)\n}\n```\n\n### CQ2 — `processHostAction` needs explicit `store.Save()` after Confirm\n\n`advanceExecution` changes execution status but doesn't save the Goals/Tasks that may have been set during confirming flow. Needs explicit persist before Pool submit.\n\n### CQ3 — `RobotStatusSnapshot` should include `MemberID` and `Status`\n\nAdd back `MemberID` and `Status` fields to match design §5.7. These help Host Agent identify which robot it's serving.\n\n---\n\n## 4. Implementation Priority\n\n| Priority | Items | Est. Effort |\n|----------|-------|-------------|\n| **P0** | BUG-1 (advanceExecution), BUG-2 (standard.New) | 1 day |\n| **P1** | BUG-3 (snapshot), BUG-4 (context auth) | 0.5 day |\n| **P1** | Tests §2.2 (processHostAction), §2.3 (HandleInteract), §2.5 (Resume) | 1.5 days |\n| **P2** | BUG-5 (Goals type), CQ1-CQ3 | 0.5 day |\n| **P2** | Tests §2.1 (CallHostAgent), §2.4 (Cancel), §2.6-2.10 | 2 days |\n| | **Total** | **~5.5 days** |\n\n---\n\n## 5. Test Infrastructure Notes\n\n1. **Source env before test**: `source yao/env.local.sh`\n2. **Test app**: `yao-dev-app` — all test assistants live there\n3. **No recompile needed**: `yao-dev` runs from Go source directly\n4. **Mock strategy**: For unit tests not requiring real LLM, create interfaces for `ConversationCaller`, `ExecutionStore`, `Pool` to enable mock injection. Alternatively, use `SkipPersistence: true` config + in-memory stubs.\n5. **Event verification**: Wrap `event.Push` calls with a test interceptor or use `event.Subscribe` to capture events during test.\n" }, { "path": "agent/robot/api/README.md", "content": "# Robot Agent Go API\n\nGo API for managing autonomous robot agents.\n\n## Quick Start\n\n```go\nimport \"github.com/yaoapp/yao/agent/robot/api\"\n\n// Start system\napi.Start()\ndefer api.Stop()\n\n// Trigger execution\nresult, _ := api.Trigger(ctx, \"member_123\", &api.TriggerRequest{\n Type: types.TriggerHuman,\n Action: types.ActionTaskAdd,\n Messages: []agentcontext.Message{{Role: \"user\", Content: \"Analyze sales\"}},\n})\n\n// Check status\nexec, _ := api.GetExecution(ctx, result.ExecutionID)\n```\n\n## Lifecycle\n\n```go\napi.Start() // Start with defaults\napi.StartWithConfig(config) // Start with custom config\napi.Stop() // Graceful shutdown\napi.IsRunning() // Check if running\n```\n\n## Robot Query\n\n```go\n// Get single robot\nrobot, err := api.GetRobot(ctx, \"member_123\")\n\n// List robots with filters\nresult, err := api.ListRobots(ctx, &api.ListQuery{\n TeamID: \"team_1\",\n Status: types.RobotIdle,\n Keywords: \"sales\",\n ClockMode: types.ClockInterval,\n Page: 1,\n PageSize: 20,\n Order: \"created_at desc\",\n})\n\n// Get runtime status\nstate, err := api.GetRobotStatus(ctx, \"member_123\")\n// state.Running, state.MaxRunning, state.RunningIDs, state.LastRun, state.NextRun\n```\n\n## Triggers\n\n### Human Intervention\n\n```go\nresult, err := api.Trigger(ctx, \"member_123\", &api.TriggerRequest{\n Type: types.TriggerHuman,\n Action: types.ActionTaskAdd,\n Messages: []agentcontext.Message{\n {Role: \"user\", Content: \"Generate weekly report\"},\n },\n})\n// Or use shorthand:\nresult, err := api.Intervene(ctx, \"member_123\", req)\n```\n\n### Event Trigger\n\n```go\nresult, err := api.Trigger(ctx, \"member_123\", &api.TriggerRequest{\n Type: types.TriggerEvent,\n Source: types.EventWebhook,\n EventType: \"order.created\",\n Data: map[string]interface{}{\"order_id\": \"12345\"},\n})\n// Or use shorthand:\nresult, err := api.HandleEvent(ctx, \"member_123\", req)\n```\n\n### Manual Trigger (Testing)\n\n```go\nresult, err := api.TriggerManual(ctx, \"member_123\", types.TriggerClock, nil)\n```\n\n## Execution Management\n\n```go\n// Get execution by ID\nexec, err := api.GetExecution(ctx, \"exec_abc123\")\n\n// List executions with filters\nresult, err := api.ListExecutions(ctx, \"member_123\", &api.ExecutionQuery{\n Status: types.ExecRunning,\n Trigger: types.TriggerClock,\n Page: 1,\n PageSize: 10,\n})\n\n// Get execution with runtime status\nexec, err := api.GetExecutionStatus(ctx, \"exec_abc123\")\n\n// Control execution\napi.PauseExecution(ctx, \"exec_abc123\")\napi.ResumeExecution(ctx, \"exec_abc123\")\napi.StopExecution(ctx, \"exec_abc123\")\n```\n\n## Types\n\n### ListQuery\n\n```go\ntype ListQuery struct {\n TeamID string // Filter by team\n Status types.RobotStatus // Filter by status (idle|working|paused|error)\n Keywords string // Search in display_name\n ClockMode types.ClockMode // Filter by clock mode (times|interval|daemon)\n Page int // Page number (default: 1)\n PageSize int // Page size (default: 20, max: 100)\n Order string // Order by column (default: \"created_at desc\")\n}\n```\n\n### TriggerRequest\n\n```go\ntype TriggerRequest struct {\n Type types.TriggerType // human | event | clock\n Action types.InterventionAction // task.add, goal.adjust, etc.\n Messages []agentcontext.Message // User input\n PlanAt *time.Time // Schedule for later\n InsertPosition InsertPosition // first | last | next | at\n AtIndex int // When InsertPosition = \"at\"\n Source types.EventSource // webhook | database\n EventType string // Event name\n Data map[string]interface{} // Event payload\n ExecutorMode types.ExecutorMode // standard | dryrun | sandbox\n}\n```\n\n### TriggerResult\n\n```go\ntype TriggerResult struct {\n Accepted bool // Whether trigger was accepted\n Queued bool // Whether queued (vs immediate)\n Execution *types.Execution // Execution details\n ExecutionID string // Execution ID for tracking\n Message string // Status message\n}\n```\n\n### ExecutionQuery\n\n```go\ntype ExecutionQuery struct {\n Status types.ExecStatus // Filter by status\n Trigger types.TriggerType // Filter by trigger type\n Page int // Page number (default: 1)\n PageSize int // Page size (default: 20, max: 100)\n}\n```\n\n### RobotState\n\n```go\ntype RobotState struct {\n MemberID string // Robot member ID\n TeamID string // Team ID\n DisplayName string // Display name\n Status types.RobotStatus // idle | working | paused | error\n Running int // Current running count\n MaxRunning int // Max concurrent limit\n LastRun *time.Time // Last execution time\n NextRun *time.Time // Next scheduled time\n RunningIDs []string // IDs of running executions\n}\n```\n\n## Files\n\n| File | Functions |\n|------|-----------|\n| `lifecycle.go` | `Start`, `StartWithConfig`, `Stop`, `IsRunning` |\n| `robot.go` | `GetRobot`, `ListRobots`, `GetRobotStatus` |\n| `trigger.go` | `Trigger`, `TriggerManual`, `Intervene`, `HandleEvent` |\n| `execution.go` | `GetExecution`, `ListExecutions`, `GetExecutionStatus`, `PauseExecution`, `ResumeExecution`, `StopExecution` |\n| `types.go` | Type definitions |\n" }, { "path": "agent/robot/api/activities.go", "content": "package api\n\nimport (\n\t\"context\"\n\t\"fmt\"\n\t\"time\"\n\n\t\"github.com/yaoapp/yao/agent/robot/store\"\n\t\"github.com/yaoapp/yao/agent/robot/types\"\n)\n\n// ==================== Activity Types ====================\n\n// ActivityQuery - query parameters for listing activities\ntype ActivityQuery struct {\n\tTeamID string `json:\"team_id,omitempty\"` // Filter by team ID\n\tLimit int `json:\"limit,omitempty\"`\n\tSince *time.Time `json:\"since,omitempty\"` // Only activities after this time\n\tType string `json:\"type,omitempty\"` // Filter by activity type: execution.started, execution.completed, execution.failed, execution.cancelled\n}\n\n// Activity - activity item for feed\ntype Activity struct {\n\tType store.ActivityType `json:\"type\"`\n\tRobotID string `json:\"robot_id\"`\n\tRobotName string `json:\"robot_name,omitempty\"` // Display name from robot\n\tExecutionID string `json:\"execution_id\"`\n\tMessage string `json:\"message\"`\n\tTimestamp time.Time `json:\"timestamp\"`\n}\n\n// ActivityListResponse - response with activities\ntype ActivityListResponse struct {\n\tData []*Activity `json:\"data\"`\n}\n\n// ==================== Activity API Functions ====================\n\n// ListActivities returns recent activities for a team\n// Activities are derived from execution status changes\nfunc ListActivities(ctx *types.Context, query *ActivityQuery) (*ActivityListResponse, error) {\n\tif query == nil {\n\t\tquery = &ActivityQuery{}\n\t}\n\tquery.applyDefaults()\n\n\t// Build store options\n\topts := &store.ActivityListOptions{\n\t\tLimit: query.Limit,\n\t\tSince: query.Since,\n\t}\n\n\tif query.TeamID != \"\" {\n\t\topts.TeamID = query.TeamID\n\t}\n\n\t// Pass type filter if provided\n\tif query.Type != \"\" {\n\t\topts.Type = store.ActivityType(query.Type)\n\t}\n\n\t// Query from store\n\tstoreActivities, err := getExecutionStore().ListActivities(context.Background(), opts)\n\tif err != nil {\n\t\treturn nil, fmt.Errorf(\"failed to list activities: %w\", err)\n\t}\n\n\t// Transform to Activity slice\n\t// Also enrich with robot display names\n\tactivities := make([]*Activity, 0, len(storeActivities))\n\trobotNames := make(map[string]string) // Cache robot names\n\n\tfor _, sa := range storeActivities {\n\t\tactivity := &Activity{\n\t\t\tType: sa.Type,\n\t\t\tRobotID: sa.RobotID,\n\t\t\tExecutionID: sa.ExecutionID,\n\t\t\tMessage: sa.Message,\n\t\t\tTimestamp: sa.Timestamp,\n\t\t}\n\n\t\t// Try to get robot name (with caching)\n\t\tif name, ok := robotNames[sa.RobotID]; ok {\n\t\t\tactivity.RobotName = name\n\t\t} else {\n\t\t\t// Try to get robot display name\n\t\t\trobotResp, err := GetRobotResponse(ctx, sa.RobotID)\n\t\t\tif err == nil && robotResp != nil {\n\t\t\t\tactivity.RobotName = robotResp.DisplayName\n\t\t\t\trobotNames[sa.RobotID] = robotResp.DisplayName\n\t\t\t}\n\t\t}\n\n\t\tactivities = append(activities, activity)\n\t}\n\n\treturn &ActivityListResponse{\n\t\tData: activities,\n\t}, nil\n}\n\n// ==================== Helper Functions ====================\n\n// applyDefaults applies default values to ActivityQuery\nfunc (q *ActivityQuery) applyDefaults() {\n\tif q.Limit <= 0 {\n\t\tq.Limit = 20\n\t}\n\tif q.Limit > 100 {\n\t\tq.Limit = 100\n\t}\n}\n" }, { "path": "agent/robot/api/api_test.go", "content": "package api_test\n\n// Integration tests for the Robot Agent API\n// These tests verify the complete API functionality with real database operations.\n//\n// Test Structure:\n// - api_test.go: Core API integration tests (this file)\n// - lifecycle_test.go: Start/Stop lifecycle tests\n// - robot_test.go: Robot query tests\n// - trigger_test.go: Trigger tests\n// - execution_test.go: Execution query/control tests\n//\n// Test Data:\n// All tests use real database records in __yao.member and agent_execution tables\n// Test robot IDs are prefixed with \"robot_api_\" for easy cleanup\n// Test execution IDs are prefixed with \"exec_api_\" for easy cleanup\n\nimport (\n\t\"context\"\n\t\"encoding/json\"\n\t\"testing\"\n\t\"time\"\n\n\t\"github.com/stretchr/testify/assert\"\n\t\"github.com/stretchr/testify/require\"\n\t\"github.com/yaoapp/gou/model\"\n\t\"github.com/yaoapp/xun/capsule\"\n\t\"github.com/yaoapp/yao/agent/robot/api\"\n\t\"github.com/yaoapp/yao/agent/robot/store\"\n\t\"github.com/yaoapp/yao/agent/robot/types\"\n\t\"github.com/yaoapp/yao/agent/testutils\"\n)\n\n// ==================== Full Lifecycle Integration Tests ====================\n\n// TestAPIFullLifecycle tests the complete API workflow:\n// Start → Create Robot → Query Robot → Trigger → Query Execution → Stop\nfunc TestAPIFullLifecycle(t *testing.T) {\n\tif testing.Short() {\n\t\tt.Skip(\"Skipping integration test\")\n\t}\n\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\t// Cleanup test data\n\tcleanupAPITestRobots(t)\n\tcleanupAPITestExecutions(t)\n\tdefer cleanupAPITestRobots(t)\n\tdefer cleanupAPITestExecutions(t)\n\n\tt.Run(\"complete workflow\", func(t *testing.T) {\n\t\t// 1. Setup: Create test robot in database\n\t\tsetupAPITestRobot(t, \"robot_api_lifecycle_001\", \"team_api_001\")\n\n\t\t// 2. Start the API system\n\t\terr := api.Start()\n\t\trequire.NoError(t, err)\n\t\tassert.True(t, api.IsRunning())\n\t\tdefer api.Stop()\n\n\t\t// 3. Query the robot via API\n\t\tctx := types.NewContext(context.Background(), nil)\n\t\trobot, err := api.GetRobot(ctx, \"robot_api_lifecycle_001\")\n\t\trequire.NoError(t, err)\n\t\trequire.NotNil(t, robot)\n\t\tassert.Equal(t, \"robot_api_lifecycle_001\", robot.MemberID)\n\t\tassert.Equal(t, \"team_api_001\", robot.TeamID)\n\t\tassert.Equal(t, \"API Test Robot robot_api_lifecycle_001\", robot.DisplayName)\n\n\t\t// 4. Get robot status\n\t\tstatus, err := api.GetRobotStatus(ctx, \"robot_api_lifecycle_001\")\n\t\trequire.NoError(t, err)\n\t\trequire.NotNil(t, status)\n\t\tassert.Equal(t, \"robot_api_lifecycle_001\", status.MemberID)\n\t\tassert.Equal(t, types.RobotIdle, status.Status)\n\t\tassert.Equal(t, 0, status.Running)\n\t\tassert.Equal(t, 5, status.MaxRunning)\n\n\t\t// 5. List robots\n\t\tlistResult, err := api.ListRobots(ctx, &api.ListQuery{\n\t\t\tTeamID: \"team_api_001\",\n\t\t\tPage: 1,\n\t\t\tPageSize: 10,\n\t\t})\n\t\trequire.NoError(t, err)\n\t\trequire.NotNil(t, listResult)\n\t\tassert.GreaterOrEqual(t, listResult.Total, 1)\n\n\t\t// Find our robot in the list\n\t\tfound := false\n\t\tfor _, r := range listResult.Data {\n\t\t\tif r.MemberID == \"robot_api_lifecycle_001\" {\n\t\t\t\tfound = true\n\t\t\t\tbreak\n\t\t\t}\n\t\t}\n\t\tassert.True(t, found, \"Robot should be in list\")\n\n\t\t// 6. Trigger manual execution\n\t\ttriggerResult, err := api.TriggerManual(ctx, \"robot_api_lifecycle_001\", types.TriggerClock, nil)\n\t\trequire.NoError(t, err)\n\t\trequire.NotNil(t, triggerResult)\n\t\tassert.True(t, triggerResult.Accepted)\n\t\tassert.NotEmpty(t, triggerResult.ExecutionID)\n\n\t\t// 7. Wait for execution to complete\n\t\ttime.Sleep(500 * time.Millisecond)\n\n\t\t// 8. Stop the system\n\t\terr = api.Stop()\n\t\trequire.NoError(t, err)\n\t\tassert.False(t, api.IsRunning())\n\t})\n}\n\n// TestAPIRobotQueryWithData tests robot query APIs with real data\nfunc TestAPIRobotQueryWithData(t *testing.T) {\n\tif testing.Short() {\n\t\tt.Skip(\"Skipping integration test\")\n\t}\n\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\tcleanupAPITestRobots(t)\n\tdefer cleanupAPITestRobots(t)\n\n\t// Setup: Create multiple robots\n\tsetupAPITestRobot(t, \"robot_api_query_001\", \"team_api_query\")\n\tsetupAPITestRobot(t, \"robot_api_query_002\", \"team_api_query\")\n\tsetupAPITestRobot(t, \"robot_api_query_003\", \"team_api_other\")\n\n\tctx := types.NewContext(context.Background(), nil)\n\n\tt.Run(\"GetRobot returns correct robot\", func(t *testing.T) {\n\t\trobot, err := api.GetRobot(ctx, \"robot_api_query_001\")\n\t\trequire.NoError(t, err)\n\t\trequire.NotNil(t, robot)\n\n\t\tassert.Equal(t, \"robot_api_query_001\", robot.MemberID)\n\t\tassert.Equal(t, \"team_api_query\", robot.TeamID)\n\t\tassert.Equal(t, \"API Test Robot robot_api_query_001\", robot.DisplayName)\n\t\tassert.True(t, robot.AutonomousMode)\n\t\tassert.Equal(t, types.RobotIdle, robot.Status)\n\t})\n\n\tt.Run(\"ListRobots filters by team\", func(t *testing.T) {\n\t\tresult, err := api.ListRobots(ctx, &api.ListQuery{\n\t\t\tTeamID: \"team_api_query\",\n\t\t\tPage: 1,\n\t\t\tPageSize: 10,\n\t\t})\n\t\trequire.NoError(t, err)\n\t\trequire.NotNil(t, result)\n\n\t\t// Should have at least 2 robots from team_api_query\n\t\t// (might have more if other tests created robots in this team)\n\t\tassert.GreaterOrEqual(t, result.Total, 2)\n\t\tassert.GreaterOrEqual(t, len(result.Data), 2)\n\n\t\t// Verify all returned robots are from the correct team\n\t\tfor _, robot := range result.Data {\n\t\t\tassert.Equal(t, \"team_api_query\", robot.TeamID)\n\t\t}\n\t})\n\n\tt.Run(\"ListRobots pagination works\", func(t *testing.T) {\n\t\t// Page 1 with size 1\n\t\tresult1, err := api.ListRobots(ctx, &api.ListQuery{\n\t\t\tTeamID: \"team_api_query\",\n\t\t\tPage: 1,\n\t\t\tPageSize: 1,\n\t\t})\n\t\trequire.NoError(t, err)\n\t\trequire.GreaterOrEqual(t, len(result1.Data), 1, \"Should have at least 1 robot on page 1\")\n\n\t\t// Page 2 with size 1\n\t\tresult2, err := api.ListRobots(ctx, &api.ListQuery{\n\t\t\tTeamID: \"team_api_query\",\n\t\t\tPage: 2,\n\t\t\tPageSize: 1,\n\t\t})\n\t\trequire.NoError(t, err)\n\t\trequire.GreaterOrEqual(t, len(result2.Data), 1, \"Should have at least 1 robot on page 2\")\n\n\t\t// Should be different robots\n\t\tassert.NotEqual(t, result1.Data[0].MemberID, result2.Data[0].MemberID)\n\t})\n\n\tt.Run(\"ListRobots filters by keywords\", func(t *testing.T) {\n\t\tresult, err := api.ListRobots(ctx, &api.ListQuery{\n\t\t\tKeywords: \"robot_api_query_001\",\n\t\t\tPage: 1,\n\t\t\tPageSize: 10,\n\t\t})\n\t\trequire.NoError(t, err)\n\t\trequire.NotNil(t, result)\n\n\t\t// Should find at least 1 robot matching keywords\n\t\tassert.GreaterOrEqual(t, result.Total, 1)\n\t\tfor _, robot := range result.Data {\n\t\t\tassert.Contains(t, robot.DisplayName, \"robot_api_query_001\")\n\t\t}\n\t})\n}\n\n// TestListRobotsAutonomousModeFilter tests the autonomous_mode filter\nfunc TestListRobotsAutonomousModeFilter(t *testing.T) {\n\tif testing.Short() {\n\t\tt.Skip(\"Skipping integration test\")\n\t}\n\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\tcleanupAPITestRobots(t)\n\tdefer cleanupAPITestRobots(t)\n\n\t// Setup: Create robots with different autonomous_mode settings\n\tsetupAPITestRobotWithMode(t, \"robot_api_auto_001\", \"team_api_mode\", true) // autonomous\n\tsetupAPITestRobotWithMode(t, \"robot_api_auto_002\", \"team_api_mode\", true) // autonomous\n\tsetupAPITestRobotWithMode(t, \"robot_api_demand_001\", \"team_api_mode\", false) // on-demand\n\n\tctx := types.NewContext(context.Background(), nil)\n\n\tt.Run(\"ListRobots returns all robots when autonomous_mode is nil\", func(t *testing.T) {\n\t\tresult, err := api.ListRobots(ctx, &api.ListQuery{\n\t\t\tTeamID: \"team_api_mode\",\n\t\t\tPage: 1,\n\t\t\tPageSize: 10,\n\t\t})\n\t\trequire.NoError(t, err)\n\t\trequire.NotNil(t, result)\n\n\t\t// Should have all 3 robots\n\t\tassert.Equal(t, 3, result.Total)\n\t})\n\n\tt.Run(\"ListRobots filters by autonomous_mode=true\", func(t *testing.T) {\n\t\tautonomousMode := true\n\t\tresult, err := api.ListRobots(ctx, &api.ListQuery{\n\t\t\tTeamID: \"team_api_mode\",\n\t\t\tAutonomousMode: &autonomousMode,\n\t\t\tPage: 1,\n\t\t\tPageSize: 10,\n\t\t})\n\t\trequire.NoError(t, err)\n\t\trequire.NotNil(t, result)\n\n\t\t// Should have only 2 autonomous robots\n\t\tassert.Equal(t, 2, result.Total)\n\t\tfor _, robot := range result.Data {\n\t\t\tassert.True(t, robot.AutonomousMode, \"All returned robots should be autonomous\")\n\t\t}\n\t})\n\n\tt.Run(\"ListRobots filters by autonomous_mode=false\", func(t *testing.T) {\n\t\tautonomousMode := false\n\t\tresult, err := api.ListRobots(ctx, &api.ListQuery{\n\t\t\tTeamID: \"team_api_mode\",\n\t\t\tAutonomousMode: &autonomousMode,\n\t\t\tPage: 1,\n\t\t\tPageSize: 10,\n\t\t})\n\t\trequire.NoError(t, err)\n\t\trequire.NotNil(t, result)\n\n\t\t// Should have only 1 on-demand robot\n\t\tassert.Equal(t, 1, result.Total)\n\t\tfor _, robot := range result.Data {\n\t\t\tassert.False(t, robot.AutonomousMode, \"All returned robots should be on-demand\")\n\t\t}\n\t})\n}\n\n// TestAPIExecutionQueryWithData tests execution query APIs with real data\nfunc TestAPIExecutionQueryWithData(t *testing.T) {\n\tif testing.Short() {\n\t\tt.Skip(\"Skipping integration test\")\n\t}\n\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\tcleanupAPITestExecutions(t)\n\tdefer cleanupAPITestExecutions(t)\n\n\t// Setup: Create test executions\n\tsetupAPITestExecution(t, \"exec_api_query_001\", \"member_api_exec\", types.TriggerClock, types.ExecCompleted)\n\tsetupAPITestExecution(t, \"exec_api_query_002\", \"member_api_exec\", types.TriggerHuman, types.ExecRunning)\n\tsetupAPITestExecution(t, \"exec_api_query_003\", \"member_api_exec\", types.TriggerClock, types.ExecFailed)\n\tsetupAPITestExecution(t, \"exec_api_query_004\", \"member_api_other\", types.TriggerEvent, types.ExecCompleted)\n\n\tctx := types.NewContext(context.Background(), nil)\n\n\tt.Run(\"GetExecution returns correct execution\", func(t *testing.T) {\n\t\texec, err := api.GetExecution(ctx, \"exec_api_query_001\")\n\t\trequire.NoError(t, err)\n\t\trequire.NotNil(t, exec)\n\n\t\tassert.Equal(t, \"exec_api_query_001\", exec.ID)\n\t\tassert.Equal(t, \"member_api_exec\", exec.MemberID)\n\t\tassert.Equal(t, types.TriggerClock, exec.TriggerType)\n\t\tassert.Equal(t, types.ExecCompleted, exec.Status)\n\t})\n\n\tt.Run(\"ListExecutions filters by member\", func(t *testing.T) {\n\t\tresult, err := api.ListExecutions(ctx, \"member_api_exec\", &api.ExecutionQuery{\n\t\t\tPage: 1,\n\t\t\tPageSize: 10,\n\t\t})\n\t\trequire.NoError(t, err)\n\t\trequire.NotNil(t, result)\n\n\t\t// Should have 3 executions for member_api_exec\n\t\tassert.Equal(t, 3, result.Total)\n\t\tassert.Len(t, result.Data, 3)\n\n\t\t// Verify all returned executions are for the correct member\n\t\tfor _, exec := range result.Data {\n\t\t\tassert.Equal(t, \"member_api_exec\", exec.MemberID)\n\t\t}\n\t})\n\n\tt.Run(\"ListExecutions filters by status\", func(t *testing.T) {\n\t\tresult, err := api.ListExecutions(ctx, \"member_api_exec\", &api.ExecutionQuery{\n\t\t\tStatus: types.ExecCompleted,\n\t\t\tPage: 1,\n\t\t\tPageSize: 10,\n\t\t})\n\t\trequire.NoError(t, err)\n\t\trequire.NotNil(t, result)\n\n\t\t// Should have only completed executions\n\t\tassert.Equal(t, 1, result.Total)\n\t\tfor _, exec := range result.Data {\n\t\t\tassert.Equal(t, types.ExecCompleted, exec.Status)\n\t\t}\n\t})\n\n\tt.Run(\"ListExecutions filters by trigger type\", func(t *testing.T) {\n\t\tresult, err := api.ListExecutions(ctx, \"member_api_exec\", &api.ExecutionQuery{\n\t\t\tTrigger: types.TriggerClock,\n\t\t\tPage: 1,\n\t\t\tPageSize: 10,\n\t\t})\n\t\trequire.NoError(t, err)\n\t\trequire.NotNil(t, result)\n\n\t\t// Should have only clock trigger executions\n\t\tassert.Equal(t, 2, result.Total)\n\t\tfor _, exec := range result.Data {\n\t\t\tassert.Equal(t, types.TriggerClock, exec.TriggerType)\n\t\t}\n\t})\n\n\tt.Run(\"ListExecutions pagination works\", func(t *testing.T) {\n\t\t// Page 1 with size 2\n\t\tresult1, err := api.ListExecutions(ctx, \"member_api_exec\", &api.ExecutionQuery{\n\t\t\tPage: 1,\n\t\t\tPageSize: 2,\n\t\t})\n\t\trequire.NoError(t, err)\n\t\tassert.GreaterOrEqual(t, len(result1.Data), 2, \"Should have at least 2 executions on page 1\")\n\n\t\t// Page 2 with size 2\n\t\tresult2, err := api.ListExecutions(ctx, \"member_api_exec\", &api.ExecutionQuery{\n\t\t\tPage: 2,\n\t\t\tPageSize: 2,\n\t\t})\n\t\trequire.NoError(t, err)\n\t\tassert.GreaterOrEqual(t, len(result2.Data), 1, \"Should have at least 1 execution on page 2\")\n\t})\n}\n\n// TestAPITriggerWithData tests trigger APIs with real robots\nfunc TestAPITriggerWithData(t *testing.T) {\n\tif testing.Short() {\n\t\tt.Skip(\"Skipping integration test\")\n\t}\n\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\tcleanupAPITestRobots(t)\n\tdefer cleanupAPITestRobots(t)\n\n\t// Setup: Create test robot\n\tsetupAPITestRobot(t, \"robot_api_trigger_001\", \"team_api_trigger\")\n\n\t// Start manager\n\terr := api.Start()\n\trequire.NoError(t, err)\n\tdefer api.Stop()\n\n\tctx := types.NewContext(context.Background(), nil)\n\n\tt.Run(\"TriggerManual accepts valid robot\", func(t *testing.T) {\n\t\tresult, err := api.TriggerManual(ctx, \"robot_api_trigger_001\", types.TriggerClock, nil)\n\t\trequire.NoError(t, err)\n\t\trequire.NotNil(t, result)\n\n\t\tassert.True(t, result.Accepted)\n\t\tassert.NotEmpty(t, result.ExecutionID)\n\t\tassert.Contains(t, result.Message, \"submitted\")\n\t})\n\n\tt.Run(\"Trigger with human type\", func(t *testing.T) {\n\t\tresult, err := api.Trigger(ctx, \"robot_api_trigger_001\", &api.TriggerRequest{\n\t\t\tType: types.TriggerHuman,\n\t\t\tAction: types.ActionTaskAdd,\n\t\t})\n\t\trequire.NoError(t, err)\n\t\trequire.NotNil(t, result)\n\n\t\t// Result should be returned (accepted or not depends on robot state)\n\t\t// The important thing is that the API doesn't error\n\t\tt.Logf(\"Trigger result: accepted=%v, message=%s\", result.Accepted, result.Message)\n\t})\n\n\tt.Run(\"Trigger with event type\", func(t *testing.T) {\n\t\tresult, err := api.Trigger(ctx, \"robot_api_trigger_001\", &api.TriggerRequest{\n\t\t\tType: types.TriggerEvent,\n\t\t\tSource: types.EventWebhook,\n\t\t\tEventType: \"test.event\",\n\t\t\tData: map[string]interface{}{\"key\": \"value\"},\n\t\t})\n\t\trequire.NoError(t, err)\n\t\trequire.NotNil(t, result)\n\n\t\t// Should be accepted (robot exists and has event enabled)\n\t\tassert.True(t, result.Accepted)\n\t})\n\n\tt.Run(\"Trigger rejects non-existent robot\", func(t *testing.T) {\n\t\tresult, err := api.Trigger(ctx, \"robot_api_nonexistent\", &api.TriggerRequest{\n\t\t\tType: types.TriggerHuman,\n\t\t})\n\t\t// Should not return error, but result should show not accepted\n\t\tassert.NoError(t, err)\n\t\tassert.NotNil(t, result)\n\t\tassert.False(t, result.Accepted)\n\t})\n}\n\n// ==================== Helper Functions ====================\n\n// setupAPITestRobotWithMode creates a test robot with specific autonomous_mode setting\nfunc setupAPITestRobotWithMode(t *testing.T, memberID, teamID string, autonomousMode bool) {\n\tm := model.Select(\"__yao.member\")\n\ttableName := m.MetaData.Table.Name\n\tqb := capsule.Query()\n\n\trobotConfig := map[string]interface{}{\n\t\t\"identity\": map[string]interface{}{\n\t\t\t\"role\": \"API Test Robot\",\n\t\t\t\"duties\": []string{\"Testing API functions\"},\n\t\t},\n\t\t\"quota\": map[string]interface{}{\n\t\t\t\"max\": 5,\n\t\t\t\"queue\": 20,\n\t\t\t\"priority\": 5,\n\t\t},\n\t}\n\tconfigJSON, _ := json.Marshal(robotConfig)\n\n\terr := qb.Table(tableName).Insert([]map[string]interface{}{\n\t\t{\n\t\t\t\"member_id\": memberID,\n\t\t\t\"team_id\": teamID,\n\t\t\t\"member_type\": \"robot\",\n\t\t\t\"display_name\": \"API Test Robot \" + memberID,\n\t\t\t\"system_prompt\": \"You are an API test robot.\",\n\t\t\t\"status\": \"active\",\n\t\t\t\"role_id\": \"member\",\n\t\t\t\"autonomous_mode\": autonomousMode,\n\t\t\t\"robot_status\": \"idle\",\n\t\t\t\"robot_config\": string(configJSON),\n\t\t},\n\t})\n\tif err != nil {\n\t\tt.Fatalf(\"Failed to insert robot %s: %v\", memberID, err)\n\t}\n}\n\n// setupAPITestRobot creates a test robot in the database\nfunc setupAPITestRobot(t *testing.T, memberID, teamID string) {\n\tm := model.Select(\"__yao.member\")\n\ttableName := m.MetaData.Table.Name\n\tqb := capsule.Query()\n\n\trobotConfig := map[string]interface{}{\n\t\t\"identity\": map[string]interface{}{\n\t\t\t\"role\": \"API Test Robot\",\n\t\t\t\"duties\": []string{\"Testing API functions\"},\n\t\t},\n\t\t\"quota\": map[string]interface{}{\n\t\t\t\"max\": 5,\n\t\t\t\"queue\": 20,\n\t\t\t\"priority\": 5,\n\t\t},\n\t\t\"triggers\": map[string]interface{}{\n\t\t\t\"clock\": map[string]interface{}{\"enabled\": true},\n\t\t\t\"intervene\": map[string]interface{}{\"enabled\": true},\n\t\t\t\"event\": map[string]interface{}{\"enabled\": true},\n\t\t},\n\t\t\"clock\": map[string]interface{}{\n\t\t\t\"mode\": \"interval\",\n\t\t\t\"every\": \"1h\",\n\t\t\t\"timeout\": \"30m\",\n\t\t},\n\t}\n\tconfigJSON, _ := json.Marshal(robotConfig)\n\n\terr := qb.Table(tableName).Insert([]map[string]interface{}{\n\t\t{\n\t\t\t\"member_id\": memberID,\n\t\t\t\"team_id\": teamID,\n\t\t\t\"member_type\": \"robot\",\n\t\t\t\"display_name\": \"API Test Robot \" + memberID,\n\t\t\t\"system_prompt\": \"You are an API test robot.\",\n\t\t\t\"status\": \"active\",\n\t\t\t\"role_id\": \"member\",\n\t\t\t\"autonomous_mode\": true,\n\t\t\t\"robot_status\": \"idle\",\n\t\t\t\"robot_config\": string(configJSON),\n\t\t},\n\t})\n\tif err != nil {\n\t\tt.Fatalf(\"Failed to insert robot %s: %v\", memberID, err)\n\t}\n}\n\n// setupAPITestExecution creates a test execution in the database\nfunc setupAPITestExecution(t *testing.T, execID, memberID string, triggerType types.TriggerType, status types.ExecStatus) {\n\ts := store.NewExecutionStore()\n\tctx := context.Background()\n\n\tstartTime := time.Now().Add(-1 * time.Hour)\n\trecord := &store.ExecutionRecord{\n\t\tExecutionID: execID,\n\t\tMemberID: memberID,\n\t\tTeamID: \"team_api_exec\",\n\t\tTriggerType: triggerType,\n\t\tStatus: status,\n\t\tPhase: types.PhaseDelivery,\n\t\tStartTime: &startTime,\n\t}\n\n\tif status == types.ExecCompleted || status == types.ExecFailed {\n\t\tendTime := time.Now()\n\t\trecord.EndTime = &endTime\n\t}\n\n\terr := s.Save(ctx, record)\n\tif err != nil {\n\t\tt.Fatalf(\"Failed to insert execution %s: %v\", execID, err)\n\t}\n}\n\n// cleanupAPITestRobots removes all API test robots\nfunc cleanupAPITestRobots(t *testing.T) {\n\tm := model.Select(\"__yao.member\")\n\tif m == nil {\n\t\treturn\n\t}\n\ttableName := m.MetaData.Table.Name\n\tqb := capsule.Query()\n\n\t// Delete all robots with member_id starting with \"robot_api_\" or \"api_robot_\"\n\t_, err := qb.Table(tableName).Where(\"member_id\", \"like\", \"robot_api_%\").Delete()\n\tif err != nil {\n\t\tt.Logf(\"Warning: cleanup robots error: %v\", err)\n\t}\n\n\t// Also delete \"api_robot_\" prefixed robots (new tests)\n\t_, err = qb.Table(tableName).Where(\"member_id\", \"like\", \"api_robot_%\").Delete()\n\tif err != nil {\n\t\tt.Logf(\"Warning: cleanup robots error: %v\", err)\n\t}\n}\n\n// cleanupAPITestExecutions removes all API test executions\nfunc cleanupAPITestExecutions(t *testing.T) {\n\tm := model.Select(\"__yao.agent.execution\")\n\tif m == nil {\n\t\tt.Logf(\"Warning: model __yao.agent.execution not found, skipping cleanup\")\n\t\treturn\n\t}\n\ttableName := m.MetaData.Table.Name\n\tqb := capsule.Query()\n\n\t// Delete all executions with execution_id starting with \"exec_api_\"\n\t_, err := qb.Table(tableName).Where(\"execution_id\", \"like\", \"exec_api_%\").Delete()\n\tif err != nil {\n\t\tt.Logf(\"Warning: cleanup executions error: %v\", err)\n\t}\n\n\t// Also delete executions for API test members\n\t_, err = qb.Table(tableName).Where(\"member_id\", \"like\", \"member_api_%\").Delete()\n\tif err != nil {\n\t\tt.Logf(\"Warning: cleanup executions error: %v\", err)\n\t}\n}\n" }, { "path": "agent/robot/api/e2e_clock_test.go", "content": "package api_test\n\n// End-to-end tests for Clock trigger flow\n// These tests use REAL LLM calls via Standard executor (not DryRun)\n//\n// Test Flow: Clock Trigger → P0 (Inspiration) → P1 (Goals) → P2 (Tasks) → P3 (Run) → P4 (Delivery)\n//\n// Prerequisites:\n// - Valid LLM API keys (OPENAI_TEST_KEY or DEEPSEEK_API_KEY)\n// - Test assistants in yao-dev-app/assistants/robot/\n// - Database connection (YAO_DB_PRIMARY)\n\nimport (\n\t\"context\"\n\t\"encoding/json\"\n\t\"testing\"\n\t\"time\"\n\n\t\"github.com/stretchr/testify/assert\"\n\t\"github.com/stretchr/testify/require\"\n\t\"github.com/yaoapp/gou/model\"\n\t\"github.com/yaoapp/xun/capsule\"\n\t\"github.com/yaoapp/yao/agent/robot/api\"\n\t\"github.com/yaoapp/yao/agent/robot/types\"\n\t\"github.com/yaoapp/yao/agent/testutils\"\n\toauthtypes \"github.com/yaoapp/yao/openapi/oauth/types\"\n)\n\n// testAuth returns test auth info for E2E tests\nfunc testAuth() *oauthtypes.AuthorizedInfo {\n\treturn &oauthtypes.AuthorizedInfo{\n\t\tUserID: \"e2e-test-user\",\n\t\tTeamID: \"e2e-test-team\",\n\t}\n}\n\n// TestE2EClockTriggerFullFlow tests the complete clock trigger flow with real LLM calls\n// Flow: Clock → P0 (Inspiration) → P1 (Goals) → P2 (Tasks) → P3 (Run) → P4 (Delivery)\nfunc TestE2EClockTriggerFullFlow(t *testing.T) {\n\tif testing.Short() {\n\t\tt.Skip(\"Skipping E2E test - requires real LLM calls\")\n\t}\n\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\tcleanupE2ERobots(t)\n\tcleanupE2EExecutions(t)\n\tdefer cleanupE2ERobots(t)\n\tdefer cleanupE2EExecutions(t)\n\n\tt.Run(\"complete_P0_to_P4_flow\", func(t *testing.T) {\n\t\t// Setup: Create a robot configured for clock trigger\n\t\tmemberID := \"robot_e2e_clock_001\"\n\t\tsetupE2ERobotForClock(t, memberID, \"team_e2e_clock\")\n\n\t\t// Start the API system\n\t\terr := api.Start()\n\t\trequire.NoError(t, err)\n\t\tdefer api.Stop()\n\n\t\t// Verify robot is loaded\n\t\tctx := types.NewContext(context.Background(), testAuth())\n\t\trobot, err := api.GetRobot(ctx, memberID)\n\t\trequire.NoError(t, err)\n\t\trequire.NotNil(t, robot)\n\t\tassert.Equal(t, memberID, robot.MemberID)\n\n\t\t// Trigger execution via clock trigger type\n\t\tresult, err := api.TriggerManual(ctx, memberID, types.TriggerClock, nil)\n\t\trequire.NoError(t, err)\n\t\trequire.NotNil(t, result)\n\t\tassert.True(t, result.Accepted, \"Clock trigger should be accepted: %s\", result.Message)\n\t\tassert.NotEmpty(t, result.ExecutionID, \"Should return execution ID\")\n\n\t\tt.Logf(\"Execution started: ExecutionID=%s\", result.ExecutionID)\n\n\t\t// Wait for execution to complete (real LLM calls take time)\n\t\t// P0→P4 typically takes 30-60 seconds with real LLM\n\t\tvar exec *types.Execution\n\t\tmaxWait := 180 * time.Second\n\t\tpollInterval := 2 * time.Second\n\t\tdeadline := time.Now().Add(maxWait)\n\n\t\tfor time.Now().Before(deadline) {\n\t\t\ttime.Sleep(pollInterval)\n\n\t\t\t// Query all executions and find a completed one\n\t\t\texecutions, err := api.ListExecutions(ctx, memberID, &api.ExecutionQuery{\n\t\t\t\tPage: 1,\n\t\t\t\tPageSize: 10,\n\t\t\t})\n\t\t\tif err != nil {\n\t\t\t\tt.Logf(\"Query error (retrying): %v\", err)\n\t\t\t\tcontinue\n\t\t\t}\n\n\t\t\t// Look for a completed execution\n\t\t\tfor _, e := range executions.Data {\n\t\t\t\tt.Logf(\"Execution %s: status=%s, phase=%s\", e.ID, e.Status, e.Phase)\n\t\t\t\tif e.Status == types.ExecCompleted {\n\t\t\t\t\texec = e\n\t\t\t\t\tbreak\n\t\t\t\t}\n\t\t\t}\n\n\t\t\tif exec != nil {\n\t\t\t\tbreak\n\t\t\t}\n\n\t\t\t// Also check if there's any running execution\n\t\t\thasRunning := false\n\t\t\tfor _, e := range executions.Data {\n\t\t\t\tif e.Status == types.ExecRunning || e.Status == types.ExecPending {\n\t\t\t\t\thasRunning = true\n\t\t\t\t\tbreak\n\t\t\t\t}\n\t\t\t}\n\t\t\tif !hasRunning && len(executions.Data) > 0 {\n\t\t\t\t// All executions finished but none completed - take the first one for error reporting\n\t\t\t\texec = executions.Data[0]\n\t\t\t\tbreak\n\t\t\t}\n\t\t}\n\n\t\t// Verify execution completed successfully - ALL phases must pass\n\t\trequire.NotNil(t, exec, \"Execution should exist\")\n\n\t\tif exec.Status == types.ExecFailed {\n\t\t\tt.Fatalf(\"Execution failed: %s\", exec.Error)\n\t\t}\n\n\t\t// Strict assertion: execution MUST complete successfully\n\t\tassert.Equal(t, types.ExecCompleted, exec.Status, \"Execution must complete successfully\")\n\n\t\t// Verify P0 (Inspiration) output exists\n\t\trequire.NotNil(t, exec.Inspiration, \"P0 Inspiration output must exist\")\n\t\tt.Logf(\"P0 Inspiration: %+v\", exec.Inspiration)\n\n\t\t// Verify P1 (Goals) output exists\n\t\trequire.NotNil(t, exec.Goals, \"P1 Goals output must exist\")\n\t\tt.Logf(\"P1 Goals content length: %d\", len(exec.Goals.Content))\n\n\t\t// Verify P2 (Tasks) output exists\n\t\trequire.NotNil(t, exec.Tasks, \"P2 Tasks output must exist\")\n\t\trequire.Greater(t, len(exec.Tasks), 0, \"P2 must have at least 1 task\")\n\t\tt.Logf(\"P2 Tasks count: %d\", len(exec.Tasks))\n\n\t\t// Verify P3 (Results) output exists - THIS IS CRITICAL\n\t\trequire.NotNil(t, exec.Results, \"P3 Results output must exist\")\n\t\trequire.Greater(t, len(exec.Results), 0, \"P3 must have at least 1 result\")\n\t\tt.Logf(\"P3 Results count: %d\", len(exec.Results))\n\n\t\t// Verify P4 (Delivery) output exists\n\t\trequire.NotNil(t, exec.Delivery, \"P4 Delivery output must exist\")\n\t\tt.Logf(\"P4 Delivery: RequestID=%s, Success=%v\", exec.Delivery.RequestID, exec.Delivery.Success)\n\n\t\tt.Logf(\"✅ Clock trigger E2E: ALL PHASES (P0-P4) completed successfully\")\n\t})\n}\n\n// TestE2EClockTriggerPhaseProgression tests that phases execute in correct order\nfunc TestE2EClockTriggerPhaseProgression(t *testing.T) {\n\tif testing.Short() {\n\t\tt.Skip(\"Skipping E2E test - requires real LLM calls\")\n\t}\n\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\tcleanupE2ERobots(t)\n\tcleanupE2EExecutions(t)\n\tdefer cleanupE2ERobots(t)\n\tdefer cleanupE2EExecutions(t)\n\n\tt.Run(\"phases_execute_P0_P1_P2_P3_P4\", func(t *testing.T) {\n\t\tmemberID := \"robot_e2e_clock_phases\"\n\t\tsetupE2ERobotForClock(t, memberID, \"team_e2e_clock\")\n\n\t\terr := api.Start()\n\t\trequire.NoError(t, err)\n\t\tdefer api.Stop()\n\n\t\tctx := types.NewContext(context.Background(), testAuth())\n\n\t\t// Trigger execution\n\t\tresult, err := api.TriggerManual(ctx, memberID, types.TriggerClock, nil)\n\t\trequire.NoError(t, err)\n\t\tassert.True(t, result.Accepted)\n\n\t\t// Track phase progression\n\t\tphasesObserved := make([]types.Phase, 0)\n\t\tlastPhase := types.Phase(\"\")\n\n\t\tmaxWait := 120 * time.Second\n\t\tpollInterval := 1 * time.Second\n\t\tdeadline := time.Now().Add(maxWait)\n\n\t\tfor time.Now().Before(deadline) {\n\t\t\ttime.Sleep(pollInterval)\n\n\t\t\texecutions, err := api.ListExecutions(ctx, memberID, &api.ExecutionQuery{\n\t\t\t\tPage: 1,\n\t\t\t\tPageSize: 1,\n\t\t\t})\n\t\t\tif err != nil || len(executions.Data) == 0 {\n\t\t\t\tcontinue\n\t\t\t}\n\n\t\t\texec := executions.Data[0]\n\n\t\t\t// Record phase changes\n\t\t\tif exec.Phase != lastPhase {\n\t\t\t\tphasesObserved = append(phasesObserved, exec.Phase)\n\t\t\t\tlastPhase = exec.Phase\n\t\t\t\tt.Logf(\"Phase changed to: %s\", exec.Phase)\n\t\t\t}\n\n\t\t\tif exec.Status == types.ExecCompleted || exec.Status == types.ExecFailed {\n\t\t\t\tbreak\n\t\t\t}\n\t\t}\n\n\t\t// Verify phase order (should include at least P0, P1, P2, P3, P4)\n\t\tt.Logf(\"Phases observed: %v\", phasesObserved)\n\t\tassert.GreaterOrEqual(t, len(phasesObserved), 1, \"Should observe at least one phase\")\n\n\t\t// The final phase should be delivery or learning\n\t\tif len(phasesObserved) > 0 {\n\t\t\tlastObserved := phasesObserved[len(phasesObserved)-1]\n\t\t\tassert.True(t,\n\t\t\t\tlastObserved == types.PhaseDelivery || lastObserved == types.PhaseLearning,\n\t\t\t\t\"Last phase should be delivery or learning, got: %s\", lastObserved)\n\t\t}\n\t})\n}\n\n// TestE2EClockTriggerDataPersistence tests that execution data is persisted to database\nfunc TestE2EClockTriggerDataPersistence(t *testing.T) {\n\tif testing.Short() {\n\t\tt.Skip(\"Skipping E2E test - requires real LLM calls\")\n\t}\n\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\tcleanupE2ERobots(t)\n\tcleanupE2EExecutions(t)\n\tdefer cleanupE2ERobots(t)\n\tdefer cleanupE2EExecutions(t)\n\n\tt.Run(\"execution_data_persisted_to_database\", func(t *testing.T) {\n\t\tmemberID := \"robot_e2e_clock_persist\"\n\t\tsetupE2ERobotForClock(t, memberID, \"team_e2e_clock\")\n\n\t\terr := api.Start()\n\t\trequire.NoError(t, err)\n\t\tdefer api.Stop()\n\n\t\tctx := types.NewContext(context.Background(), testAuth())\n\n\t\t// Trigger and wait for completion\n\t\tresult, err := api.TriggerManual(ctx, memberID, types.TriggerClock, nil)\n\t\trequire.NoError(t, err)\n\t\tassert.True(t, result.Accepted)\n\n\t\t// Wait for completion\n\t\tvar execID string\n\t\tmaxWait := 120 * time.Second\n\t\tdeadline := time.Now().Add(maxWait)\n\n\t\tfor time.Now().Before(deadline) {\n\t\t\ttime.Sleep(2 * time.Second)\n\n\t\t\texecutions, err := api.ListExecutions(ctx, memberID, nil)\n\t\t\tif err != nil || len(executions.Data) == 0 {\n\t\t\t\tcontinue\n\t\t\t}\n\n\t\t\texec := executions.Data[0]\n\t\t\texecID = exec.ID\n\n\t\t\tif exec.Status == types.ExecCompleted || exec.Status == types.ExecFailed {\n\t\t\t\tbreak\n\t\t\t}\n\t\t}\n\n\t\trequire.NotEmpty(t, execID, \"Should have execution ID\")\n\n\t\t// Query execution by ID to verify persistence\n\t\texec, err := api.GetExecution(ctx, execID)\n\t\trequire.NoError(t, err)\n\t\trequire.NotNil(t, exec)\n\n\t\t// Verify all fields are persisted\n\t\tassert.Equal(t, execID, exec.ID)\n\t\tassert.Equal(t, memberID, exec.MemberID)\n\t\tassert.Equal(t, types.TriggerClock, exec.TriggerType)\n\t\tassert.NotNil(t, exec.StartTime, \"StartTime should be set\")\n\n\t\tif exec.Status == types.ExecCompleted {\n\t\t\tassert.NotNil(t, exec.EndTime, \"EndTime should be set for completed execution\")\n\t\t}\n\n\t\tt.Logf(\"Persisted execution: ID=%s, Status=%s, Phase=%s\", exec.ID, exec.Status, exec.Phase)\n\t})\n}\n\n// ==================== Helper Functions ====================\n\n// setupE2ERobotForClock creates a robot configured for clock trigger E2E tests\n// Uses extremely simple tasks to ensure quick completion through all phases\nfunc setupE2ERobotForClock(t *testing.T, memberID, teamID string) {\n\tm := model.Select(\"__yao.member\")\n\ttableName := m.MetaData.Table.Name\n\tqb := capsule.Query()\n\n\t// Robot config optimized for E2E testing - tasks must complete quickly\n\trobotConfig := map[string]interface{}{\n\t\t\"identity\": map[string]interface{}{\n\t\t\t\"role\": \"Greeter Bot\",\n\t\t\t\"duties\": []string{\"Output greeting message\"}, // Extremely simple\n\t\t\t\"rules\": []string{\"Always complete in one step\", \"No tools needed\", \"Just output text directly\"},\n\t\t},\n\t\t\"quota\": map[string]interface{}{\n\t\t\t\"max\": 5,\n\t\t\t\"queue\": 20,\n\t\t\t\"priority\": 5,\n\t\t},\n\t\t\"triggers\": map[string]interface{}{\n\t\t\t\"clock\": map[string]interface{}{\"enabled\": true},\n\t\t\t\"intervene\": map[string]interface{}{\"enabled\": true},\n\t\t\t\"event\": map[string]interface{}{\"enabled\": true},\n\t\t},\n\t\t\"resources\": map[string]interface{}{\n\t\t\t\"phases\": map[string]interface{}{\n\t\t\t\t\"inspiration\": \"robot.inspiration\",\n\t\t\t\t\"goals\": \"robot.goals\",\n\t\t\t\t\"tasks\": \"tests.e2e-tasks\", // Use simple E2E test task planner\n\t\t\t\t\"run\": \"robot.validation\",\n\t\t\t\t\"validation\": \"tests.e2e-validation\", // Use lenient E2E test validator\n\t\t\t\t\"delivery\": \"robot.delivery\",\n\t\t\t\t\"learning\": \"robot.learning\",\n\t\t\t},\n\t\t\t\"agents\": []string{},\n\t\t},\n\t\t\"delivery\": map[string]interface{}{\n\t\t\t\"email\": map[string]interface{}{\"enabled\": false},\n\t\t\t\"webhook\": map[string]interface{}{\"enabled\": false},\n\t\t\t\"process\": map[string]interface{}{\"enabled\": false},\n\t\t},\n\t}\n\tconfigJSON, _ := json.Marshal(robotConfig)\n\n\t// Critical: System prompt must guide LLM to generate simple, completable tasks\n\tsystemPrompt := `You are a simple greeter bot for E2E testing.\n\nCRITICAL RULES FOR ALL PHASES:\n1. Goals: Generate exactly ONE goal: \"Output a greeting\"\n2. Tasks: Generate exactly ONE task with description \"Say Hello World\"\n3. Run: Complete the task by directly outputting \"Hello World!\" - NO tools needed\n4. The task is complete when you output the greeting text\n\nYou do NOT need any tools or external calls. Just output text directly.\nWhen asked to execute a task, immediately respond with the greeting and mark complete.`\n\n\terr := qb.Table(tableName).Insert([]map[string]interface{}{\n\t\t{\n\t\t\t\"member_id\": memberID,\n\t\t\t\"team_id\": teamID,\n\t\t\t\"member_type\": \"robot\",\n\t\t\t\"display_name\": \"E2E Clock Test Robot \" + memberID,\n\t\t\t\"system_prompt\": systemPrompt,\n\t\t\t\"status\": \"active\",\n\t\t\t\"role_id\": \"member\",\n\t\t\t\"autonomous_mode\": true,\n\t\t\t\"robot_status\": \"idle\",\n\t\t\t\"robot_config\": string(configJSON),\n\t\t},\n\t})\n\tif err != nil {\n\t\tt.Fatalf(\"Failed to insert robot %s: %v\", memberID, err)\n\t}\n}\n\n// cleanupE2ERobots removes all E2E test robots\nfunc cleanupE2ERobots(t *testing.T) {\n\tm := model.Select(\"__yao.member\")\n\tif m == nil {\n\t\treturn\n\t}\n\ttableName := m.MetaData.Table.Name\n\tqb := capsule.Query()\n\n\t_, err := qb.Table(tableName).Where(\"member_id\", \"like\", \"robot_e2e_%\").Delete()\n\tif err != nil {\n\t\tt.Logf(\"Warning: cleanup robots error: %v\", err)\n\t}\n}\n\n// cleanupE2EExecutions removes all E2E test executions\nfunc cleanupE2EExecutions(t *testing.T) {\n\tm := model.Select(\"__yao.agent.execution\")\n\tif m == nil {\n\t\treturn\n\t}\n\ttableName := m.MetaData.Table.Name\n\tqb := capsule.Query()\n\n\t_, err := qb.Table(tableName).Where(\"member_id\", \"like\", \"robot_e2e_%\").Delete()\n\tif err != nil {\n\t\tt.Logf(\"Warning: cleanup executions error: %v\", err)\n\t}\n}\n" }, { "path": "agent/robot/api/e2e_concurrent_test.go", "content": "package api_test\n\n// End-to-end tests for Concurrent execution\n// These tests verify that multiple robots can execute simultaneously\n// and that quota limits are enforced correctly.\n//\n// Prerequisites:\n// - Valid LLM API keys (OPENAI_TEST_KEY or DEEPSEEK_API_KEY)\n// - Test assistants in yao-dev-app/assistants/robot/\n// - Database connection (YAO_DB_PRIMARY)\n\nimport (\n\t\"context\"\n\t\"encoding/json\"\n\t\"sync\"\n\t\"sync/atomic\"\n\t\"testing\"\n\t\"time\"\n\n\t\"github.com/stretchr/testify/assert\"\n\t\"github.com/stretchr/testify/require\"\n\t\"github.com/yaoapp/gou/model\"\n\t\"github.com/yaoapp/xun/capsule\"\n\t\"github.com/yaoapp/yao/agent/robot/api\"\n\t\"github.com/yaoapp/yao/agent/robot/types\"\n\t\"github.com/yaoapp/yao/agent/testutils\"\n\toauthtypes \"github.com/yaoapp/yao/openapi/oauth/types\"\n)\n\n// testAuthConcurrent returns test auth info for concurrent E2E tests\nfunc testAuthConcurrent() *oauthtypes.AuthorizedInfo {\n\treturn &oauthtypes.AuthorizedInfo{\n\t\tUserID: \"e2e-concurrent-user\",\n\t\tTeamID: \"e2e-concurrent-team\",\n\t}\n}\n\n// TestE2EConcurrentMultipleRobots tests concurrent execution of multiple different robots\nfunc TestE2EConcurrentMultipleRobots(t *testing.T) {\n\tif testing.Short() {\n\t\tt.Skip(\"Skipping E2E test - requires real LLM calls\")\n\t}\n\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\tcleanupE2ERobots(t)\n\tcleanupE2EExecutions(t)\n\tdefer cleanupE2ERobots(t)\n\tdefer cleanupE2EExecutions(t)\n\n\tt.Run(\"multiple_robots_execute_concurrently\", func(t *testing.T) {\n\t\t// Create 3 different robots\n\t\trobots := []string{\n\t\t\t\"robot_e2e_concurrent_001\",\n\t\t\t\"robot_e2e_concurrent_002\",\n\t\t\t\"robot_e2e_concurrent_003\",\n\t\t}\n\n\t\tfor _, memberID := range robots {\n\t\t\tsetupE2ERobotForConcurrent(t, memberID, \"team_e2e_concurrent\")\n\t\t}\n\n\t\terr := api.Start()\n\t\trequire.NoError(t, err)\n\t\tdefer api.Stop()\n\n\t\tctx := types.NewContext(context.Background(), testAuthConcurrent())\n\n\t\t// Trigger all robots concurrently\n\t\tvar wg sync.WaitGroup\n\t\tvar acceptedCount atomic.Int32\n\t\tresults := make([]*api.TriggerResult, len(robots))\n\t\tvar mu sync.Mutex\n\n\t\tfor i, memberID := range robots {\n\t\t\twg.Add(1)\n\t\t\tgo func(idx int, id string) {\n\t\t\t\tdefer wg.Done()\n\n\t\t\t\tresult, err := api.TriggerManual(ctx, id, types.TriggerClock, nil)\n\t\t\t\tif err != nil {\n\t\t\t\t\tt.Logf(\"Robot %s trigger error: %v\", id, err)\n\t\t\t\t\treturn\n\t\t\t\t}\n\n\t\t\t\tmu.Lock()\n\t\t\t\tresults[idx] = result\n\t\t\t\tmu.Unlock()\n\n\t\t\t\tif result.Accepted {\n\t\t\t\t\tacceptedCount.Add(1)\n\t\t\t\t\tt.Logf(\"Robot %s accepted: ExecutionID=%s\", id, result.ExecutionID)\n\t\t\t\t}\n\t\t\t}(i, memberID)\n\t\t}\n\n\t\twg.Wait()\n\n\t\t// All 3 should be accepted (different robots, no quota conflict)\n\t\tassert.Equal(t, int32(3), acceptedCount.Load(), \"All 3 robots should be accepted\")\n\n\t\t// Wait for all executions to complete\n\t\tmaxWait := 180 * time.Second // Longer timeout for concurrent\n\t\tdeadline := time.Now().Add(maxWait)\n\n\t\tcompletedCount := 0\n\t\tfor time.Now().Before(deadline) {\n\t\t\ttime.Sleep(3 * time.Second)\n\n\t\t\tcompletedCount = 0\n\t\t\tfor _, memberID := range robots {\n\t\t\t\texecutions, err := api.ListExecutions(ctx, memberID, nil)\n\t\t\t\tif err != nil || len(executions.Data) == 0 {\n\t\t\t\t\tcontinue\n\t\t\t\t}\n\n\t\t\t\texec := executions.Data[0]\n\t\t\t\tif exec.Status == types.ExecCompleted || exec.Status == types.ExecFailed {\n\t\t\t\t\tcompletedCount++\n\t\t\t\t}\n\t\t\t}\n\n\t\t\tt.Logf(\"Completed: %d/%d\", completedCount, len(robots))\n\n\t\t\tif completedCount == len(robots) {\n\t\t\t\tbreak\n\t\t\t}\n\t\t}\n\n\t\t// Verify all completed\n\t\tassert.Equal(t, len(robots), completedCount, \"All robots should complete execution\")\n\t})\n}\n\n// TestE2EConcurrentSameRobotMultipleTriggers tests multiple triggers on the same robot\nfunc TestE2EConcurrentSameRobotMultipleTriggers(t *testing.T) {\n\tif testing.Short() {\n\t\tt.Skip(\"Skipping E2E test - requires real LLM calls\")\n\t}\n\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\tcleanupE2ERobots(t)\n\tcleanupE2EExecutions(t)\n\tdefer cleanupE2ERobots(t)\n\tdefer cleanupE2EExecutions(t)\n\n\tt.Run(\"same_robot_handles_multiple_triggers\", func(t *testing.T) {\n\t\tmemberID := \"robot_e2e_concurrent_same\"\n\t\t// Create robot with high quota to allow multiple concurrent executions\n\t\tsetupE2ERobotHighQuota(t, memberID, \"team_e2e_concurrent\")\n\n\t\terr := api.Start()\n\t\trequire.NoError(t, err)\n\t\tdefer api.Stop()\n\n\t\tctx := types.NewContext(context.Background(), testAuthConcurrent())\n\n\t\t// Trigger 3 executions on the same robot\n\t\ttriggerCount := 3\n\t\tvar wg sync.WaitGroup\n\t\tvar acceptedCount atomic.Int32\n\n\t\tfor i := 0; i < triggerCount; i++ {\n\t\t\twg.Add(1)\n\t\t\tgo func(idx int) {\n\t\t\t\tdefer wg.Done()\n\n\t\t\t\tresult, err := api.TriggerManual(ctx, memberID, types.TriggerClock, nil)\n\t\t\t\tif err != nil {\n\t\t\t\t\tt.Logf(\"Trigger %d error: %v\", idx, err)\n\t\t\t\t\treturn\n\t\t\t\t}\n\n\t\t\t\tif result.Accepted {\n\t\t\t\t\tacceptedCount.Add(1)\n\t\t\t\t\tt.Logf(\"Trigger %d accepted: ExecutionID=%s\", idx, result.ExecutionID)\n\t\t\t\t} else {\n\t\t\t\t\tt.Logf(\"Trigger %d rejected: %s\", idx, result.Message)\n\t\t\t\t}\n\t\t\t}(i)\n\n\t\t\t// Small delay between triggers to avoid race conditions\n\t\t\ttime.Sleep(100 * time.Millisecond)\n\t\t}\n\n\t\twg.Wait()\n\n\t\t// With high quota (max=5), all 3 should be accepted\n\t\tassert.GreaterOrEqual(t, acceptedCount.Load(), int32(1), \"At least 1 trigger should be accepted\")\n\t\tt.Logf(\"Accepted triggers: %d/%d\", acceptedCount.Load(), triggerCount)\n\n\t\t// Wait for executions to complete\n\t\tmaxWait := 180 * time.Second\n\t\tdeadline := time.Now().Add(maxWait)\n\n\t\tfor time.Now().Before(deadline) {\n\t\t\ttime.Sleep(3 * time.Second)\n\n\t\t\texecutions, err := api.ListExecutions(ctx, memberID, nil)\n\t\t\tif err != nil {\n\t\t\t\tcontinue\n\t\t\t}\n\n\t\t\tcompletedCount := 0\n\t\t\tfor _, exec := range executions.Data {\n\t\t\t\tif exec.Status == types.ExecCompleted || exec.Status == types.ExecFailed {\n\t\t\t\t\tcompletedCount++\n\t\t\t\t}\n\t\t\t}\n\n\t\t\tt.Logf(\"Completed: %d/%d\", completedCount, int(acceptedCount.Load()))\n\n\t\t\tif completedCount >= int(acceptedCount.Load()) {\n\t\t\t\tbreak\n\t\t\t}\n\t\t}\n\n\t\t// Verify execution count\n\t\texecutions, err := api.ListExecutions(ctx, memberID, nil)\n\t\trequire.NoError(t, err)\n\t\tassert.GreaterOrEqual(t, len(executions.Data), 1, \"Should have at least 1 execution\")\n\t})\n}\n\n// TestE2EConcurrentQuotaEnforcement tests that quota limits are enforced\nfunc TestE2EConcurrentQuotaEnforcement(t *testing.T) {\n\tif testing.Short() {\n\t\tt.Skip(\"Skipping E2E test - requires real LLM calls\")\n\t}\n\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\tcleanupE2ERobots(t)\n\tcleanupE2EExecutions(t)\n\tdefer cleanupE2ERobots(t)\n\tdefer cleanupE2EExecutions(t)\n\n\tt.Run(\"quota_limit_enforced\", func(t *testing.T) {\n\t\tmemberID := \"robot_e2e_concurrent_quota\"\n\t\t// Create robot with low quota (max=2)\n\t\tsetupE2ERobotLowQuota(t, memberID, \"team_e2e_concurrent\")\n\n\t\terr := api.Start()\n\t\trequire.NoError(t, err)\n\t\tdefer api.Stop()\n\n\t\tctx := types.NewContext(context.Background(), testAuthConcurrent())\n\n\t\t// Try to trigger 5 executions on robot with max=2\n\t\ttriggerCount := 5\n\t\tvar acceptedCount atomic.Int32\n\t\tvar rejectedCount atomic.Int32\n\n\t\tfor i := 0; i < triggerCount; i++ {\n\t\t\tresult, err := api.TriggerManual(ctx, memberID, types.TriggerClock, nil)\n\t\t\tif err != nil {\n\t\t\t\tt.Logf(\"Trigger %d error: %v\", i, err)\n\t\t\t\tcontinue\n\t\t\t}\n\n\t\t\tif result.Accepted {\n\t\t\t\tacceptedCount.Add(1)\n\t\t\t\tt.Logf(\"Trigger %d accepted\", i)\n\t\t\t} else {\n\t\t\t\trejectedCount.Add(1)\n\t\t\t\tt.Logf(\"Trigger %d rejected: %s\", i, result.Message)\n\t\t\t}\n\n\t\t\t// Small delay to allow execution to start\n\t\t\ttime.Sleep(200 * time.Millisecond)\n\t\t}\n\n\t\t// With max=2, only 2 should be accepted at a time\n\t\t// Some may be rejected due to quota\n\t\tt.Logf(\"Accepted: %d, Rejected: %d\", acceptedCount.Load(), rejectedCount.Load())\n\n\t\t// At least some should be accepted\n\t\tassert.GreaterOrEqual(t, acceptedCount.Load(), int32(1), \"At least 1 should be accepted\")\n\n\t\t// Wait for completion\n\t\ttime.Sleep(120 * time.Second)\n\n\t\t// Query final execution count\n\t\texecutions, err := api.ListExecutions(ctx, memberID, nil)\n\t\trequire.NoError(t, err)\n\t\tt.Logf(\"Total executions: %d\", len(executions.Data))\n\t})\n}\n\n// TestE2EConcurrentMixedTriggerTypes tests concurrent execution with different trigger types\nfunc TestE2EConcurrentMixedTriggerTypes(t *testing.T) {\n\tif testing.Short() {\n\t\tt.Skip(\"Skipping E2E test - requires real LLM calls\")\n\t}\n\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\tcleanupE2ERobots(t)\n\tcleanupE2EExecutions(t)\n\tdefer cleanupE2ERobots(t)\n\tdefer cleanupE2EExecutions(t)\n\n\tt.Run(\"mixed_trigger_types_execute_concurrently\", func(t *testing.T) {\n\t\t// Create robots for different trigger types\n\t\tclockRobot := \"robot_e2e_concurrent_clock\"\n\t\thumanRobot := \"robot_e2e_concurrent_human\"\n\t\teventRobot := \"robot_e2e_concurrent_event\"\n\n\t\tsetupE2ERobotForConcurrent(t, clockRobot, \"team_e2e_concurrent\")\n\t\tsetupE2ERobotForConcurrent(t, humanRobot, \"team_e2e_concurrent\")\n\t\tsetupE2ERobotForConcurrent(t, eventRobot, \"team_e2e_concurrent\")\n\n\t\terr := api.Start()\n\t\trequire.NoError(t, err)\n\t\tdefer api.Stop()\n\n\t\tctx := types.NewContext(context.Background(), testAuthConcurrent())\n\n\t\t// Trigger all three types concurrently\n\t\tvar wg sync.WaitGroup\n\t\tvar acceptedCount atomic.Int32\n\n\t\t// Clock trigger\n\t\twg.Add(1)\n\t\tgo func() {\n\t\t\tdefer wg.Done()\n\t\t\tresult, err := api.TriggerManual(ctx, clockRobot, types.TriggerClock, nil)\n\t\t\tif err == nil && result.Accepted {\n\t\t\t\tacceptedCount.Add(1)\n\t\t\t\tt.Logf(\"Clock trigger accepted\")\n\t\t\t}\n\t\t}()\n\n\t\t// Human trigger\n\t\twg.Add(1)\n\t\tgo func() {\n\t\t\tdefer wg.Done()\n\t\t\tresult, err := api.Trigger(ctx, humanRobot, &api.TriggerRequest{\n\t\t\t\tType: types.TriggerHuman,\n\t\t\t\tAction: types.ActionTaskAdd,\n\t\t\t})\n\t\t\tif err == nil && (result.Accepted || result.Queued) {\n\t\t\t\tacceptedCount.Add(1)\n\t\t\t\tt.Logf(\"Human trigger accepted/queued\")\n\t\t\t}\n\t\t}()\n\n\t\t// Event trigger\n\t\twg.Add(1)\n\t\tgo func() {\n\t\t\tdefer wg.Done()\n\t\t\tresult, err := api.Trigger(ctx, eventRobot, &api.TriggerRequest{\n\t\t\t\tType: types.TriggerEvent,\n\t\t\t\tSource: types.EventWebhook,\n\t\t\t\tEventType: \"test.concurrent\",\n\t\t\t\tData: map[string]interface{}{\"test\": true},\n\t\t\t})\n\t\t\tif err == nil && result.Accepted {\n\t\t\t\tacceptedCount.Add(1)\n\t\t\t\tt.Logf(\"Event trigger accepted\")\n\t\t\t}\n\t\t}()\n\n\t\twg.Wait()\n\n\t\t// All should be accepted (different robots)\n\t\tassert.GreaterOrEqual(t, acceptedCount.Load(), int32(2), \"At least 2 triggers should be accepted\")\n\n\t\t// Wait for executions\n\t\ttime.Sleep(120 * time.Second)\n\n\t\t// Verify executions exist for each robot\n\t\tfor _, memberID := range []string{clockRobot, humanRobot, eventRobot} {\n\t\t\texecutions, err := api.ListExecutions(ctx, memberID, nil)\n\t\t\tif err == nil && len(executions.Data) > 0 {\n\t\t\t\tt.Logf(\"Robot %s has %d executions\", memberID, len(executions.Data))\n\t\t\t}\n\t\t}\n\t})\n}\n\n// ==================== Helper Functions ====================\n\n// setupE2ERobotForConcurrent creates a robot for concurrent execution tests\nfunc setupE2ERobotForConcurrent(t *testing.T, memberID, teamID string) {\n\tm := model.Select(\"__yao.member\")\n\ttableName := m.MetaData.Table.Name\n\tqb := capsule.Query()\n\n\t// Simple config for E2E testing - minimal tasks\n\trobotConfig := map[string]interface{}{\n\t\t\"identity\": map[string]interface{}{\n\t\t\t\"role\": \"Simple E2E Test Robot\",\n\t\t\t\"duties\": []string{\"Say hello\"}, // Very simple duty\n\t\t\t\"rules\": []string{\"Keep responses under 50 words\"},\n\t\t},\n\t\t\"quota\": map[string]interface{}{\n\t\t\t\"max\": 3,\n\t\t\t\"queue\": 10,\n\t\t\t\"priority\": 5,\n\t\t},\n\t\t\"triggers\": map[string]interface{}{\n\t\t\t\"clock\": map[string]interface{}{\"enabled\": true},\n\t\t\t\"intervene\": map[string]interface{}{\"enabled\": true},\n\t\t\t\"event\": map[string]interface{}{\"enabled\": true},\n\t\t},\n\t\t\"clock\": map[string]interface{}{\n\t\t\t\"mode\": \"interval\",\n\t\t\t\"every\": \"1h\",\n\t\t},\n\t\t\"resources\": map[string]interface{}{\n\t\t\t\"phases\": map[string]interface{}{\n\t\t\t\t\"inspiration\": \"robot.inspiration\",\n\t\t\t\t\"goals\": \"robot.goals\",\n\t\t\t\t\"tasks\": \"tests.e2e-tasks\", // Use simple E2E test task planner\n\t\t\t\t\"run\": \"robot.validation\",\n\t\t\t\t\"validation\": \"tests.e2e-validation\", // Use lenient E2E test validator\n\t\t\t\t\"delivery\": \"robot.delivery\",\n\t\t\t\t\"learning\": \"robot.learning\",\n\t\t\t},\n\t\t\t\"agents\": []string{\"experts.text-writer\"},\n\t\t},\n\t\t\"delivery\": map[string]interface{}{\n\t\t\t\"email\": map[string]interface{}{\"enabled\": false},\n\t\t\t\"webhook\": map[string]interface{}{\"enabled\": false},\n\t\t\t\"process\": map[string]interface{}{\"enabled\": false},\n\t\t},\n\t}\n\tconfigJSON, _ := json.Marshal(robotConfig)\n\n\tsystemPrompt := `You are a simple E2E test robot. Your job is to say hello.\nWhen generating goals: create exactly 1 simple goal.\nWhen generating tasks: create exactly 1 simple task.\nKeep all outputs brief. No complex analysis needed.`\n\n\terr := qb.Table(tableName).Insert([]map[string]interface{}{\n\t\t{\n\t\t\t\"member_id\": memberID,\n\t\t\t\"team_id\": teamID,\n\t\t\t\"member_type\": \"robot\",\n\t\t\t\"display_name\": \"E2E Concurrent Robot \" + memberID,\n\t\t\t\"system_prompt\": systemPrompt,\n\t\t\t\"status\": \"active\",\n\t\t\t\"role_id\": \"member\",\n\t\t\t\"autonomous_mode\": true,\n\t\t\t\"robot_status\": \"idle\",\n\t\t\t\"robot_config\": string(configJSON),\n\t\t},\n\t})\n\tif err != nil {\n\t\tt.Fatalf(\"Failed to insert robot %s: %v\", memberID, err)\n\t}\n}\n\n// setupE2ERobotHighQuota creates a robot with high quota for concurrent tests\nfunc setupE2ERobotHighQuota(t *testing.T, memberID, teamID string) {\n\tm := model.Select(\"__yao.member\")\n\ttableName := m.MetaData.Table.Name\n\tqb := capsule.Query()\n\n\trobotConfig := map[string]interface{}{\n\t\t\"identity\": map[string]interface{}{\n\t\t\t\"role\": \"E2E Test Robot - High Quota\",\n\t\t},\n\t\t\"quota\": map[string]interface{}{\n\t\t\t\"max\": 5, // High quota\n\t\t\t\"queue\": 20,\n\t\t\t\"priority\": 5,\n\t\t},\n\t\t\"triggers\": map[string]interface{}{\n\t\t\t\"clock\": map[string]interface{}{\"enabled\": true},\n\t\t\t\"intervene\": map[string]interface{}{\"enabled\": true},\n\t\t\t\"event\": map[string]interface{}{\"enabled\": true},\n\t\t},\n\t\t// Resources: phase agents and expert agents from yao-dev-app/assistants/\n\t\t\"resources\": map[string]interface{}{\n\t\t\t\"phases\": map[string]interface{}{\n\t\t\t\t\"inspiration\": \"robot.inspiration\",\n\t\t\t\t\"goals\": \"robot.goals\",\n\t\t\t\t\"tasks\": \"tests.e2e-tasks\", // Use simple E2E test task planner\n\t\t\t\t\"run\": \"robot.validation\",\n\t\t\t\t\"validation\": \"tests.e2e-validation\", // Use lenient E2E test validator\n\t\t\t\t\"delivery\": \"robot.delivery\",\n\t\t\t\t\"learning\": \"robot.learning\",\n\t\t\t},\n\t\t\t\"agents\": []string{\"experts.text-writer\"},\n\t\t},\n\t\t\"delivery\": map[string]interface{}{\n\t\t\t\"email\": map[string]interface{}{\"enabled\": false},\n\t\t\t\"webhook\": map[string]interface{}{\"enabled\": false},\n\t\t\t\"process\": map[string]interface{}{\"enabled\": false},\n\t\t},\n\t}\n\tconfigJSON, _ := json.Marshal(robotConfig)\n\n\terr := qb.Table(tableName).Insert([]map[string]interface{}{\n\t\t{\n\t\t\t\"member_id\": memberID,\n\t\t\t\"team_id\": teamID,\n\t\t\t\"member_type\": \"robot\",\n\t\t\t\"display_name\": \"E2E High Quota Robot \" + memberID,\n\t\t\t\"system_prompt\": \"You are a high quota test robot.\",\n\t\t\t\"status\": \"active\",\n\t\t\t\"role_id\": \"member\",\n\t\t\t\"autonomous_mode\": true,\n\t\t\t\"robot_status\": \"idle\",\n\t\t\t\"robot_config\": string(configJSON),\n\t\t},\n\t})\n\tif err != nil {\n\t\tt.Fatalf(\"Failed to insert robot %s: %v\", memberID, err)\n\t}\n}\n\n// setupE2ERobotLowQuota creates a robot with low quota for quota enforcement tests\nfunc setupE2ERobotLowQuota(t *testing.T, memberID, teamID string) {\n\tm := model.Select(\"__yao.member\")\n\ttableName := m.MetaData.Table.Name\n\tqb := capsule.Query()\n\n\trobotConfig := map[string]interface{}{\n\t\t\"identity\": map[string]interface{}{\n\t\t\t\"role\": \"E2E Test Robot - Low Quota\",\n\t\t},\n\t\t\"quota\": map[string]interface{}{\n\t\t\t\"max\": 2, // Low quota for testing limits\n\t\t\t\"queue\": 5,\n\t\t\t\"priority\": 5,\n\t\t},\n\t\t\"triggers\": map[string]interface{}{\n\t\t\t\"clock\": map[string]interface{}{\"enabled\": true},\n\t\t\t\"intervene\": map[string]interface{}{\"enabled\": true},\n\t\t\t\"event\": map[string]interface{}{\"enabled\": true},\n\t\t},\n\t\t// Resources: phase agents and expert agents from yao-dev-app/assistants/\n\t\t\"resources\": map[string]interface{}{\n\t\t\t\"phases\": map[string]interface{}{\n\t\t\t\t\"inspiration\": \"robot.inspiration\",\n\t\t\t\t\"goals\": \"robot.goals\",\n\t\t\t\t\"tasks\": \"tests.e2e-tasks\", // Use simple E2E test task planner\n\t\t\t\t\"run\": \"robot.validation\",\n\t\t\t\t\"validation\": \"tests.e2e-validation\", // Use lenient E2E test validator\n\t\t\t\t\"delivery\": \"robot.delivery\",\n\t\t\t\t\"learning\": \"robot.learning\",\n\t\t\t},\n\t\t\t\"agents\": []string{\"experts.text-writer\"},\n\t\t},\n\t\t\"delivery\": map[string]interface{}{\n\t\t\t\"email\": map[string]interface{}{\"enabled\": false},\n\t\t\t\"webhook\": map[string]interface{}{\"enabled\": false},\n\t\t\t\"process\": map[string]interface{}{\"enabled\": false},\n\t\t},\n\t}\n\tconfigJSON, _ := json.Marshal(robotConfig)\n\n\terr := qb.Table(tableName).Insert([]map[string]interface{}{\n\t\t{\n\t\t\t\"member_id\": memberID,\n\t\t\t\"team_id\": teamID,\n\t\t\t\"member_type\": \"robot\",\n\t\t\t\"display_name\": \"E2E Low Quota Robot \" + memberID,\n\t\t\t\"system_prompt\": \"You are a low quota test robot.\",\n\t\t\t\"status\": \"active\",\n\t\t\t\"role_id\": \"member\",\n\t\t\t\"autonomous_mode\": true,\n\t\t\t\"robot_status\": \"idle\",\n\t\t\t\"robot_config\": string(configJSON),\n\t\t},\n\t})\n\tif err != nil {\n\t\tt.Fatalf(\"Failed to insert robot %s: %v\", memberID, err)\n\t}\n}\n" }, { "path": "agent/robot/api/e2e_control_test.go", "content": "package api_test\n\n// End-to-end tests for Execution Control (Pause/Resume/Stop)\n// These tests verify that executions can be controlled during runtime\n//\n// Prerequisites:\n// - Valid LLM API keys (OPENAI_TEST_KEY or DEEPSEEK_API_KEY)\n// - Test assistants in yao-dev-app/assistants/robot/\n// - Database connection (YAO_DB_PRIMARY)\n\nimport (\n\t\"context\"\n\t\"encoding/json\"\n\t\"testing\"\n\t\"time\"\n\n\t\"github.com/stretchr/testify/assert\"\n\t\"github.com/stretchr/testify/require\"\n\t\"github.com/yaoapp/gou/model\"\n\t\"github.com/yaoapp/xun/capsule\"\n\t\"github.com/yaoapp/yao/agent/robot/api\"\n\t\"github.com/yaoapp/yao/agent/robot/types\"\n\t\"github.com/yaoapp/yao/agent/testutils\"\n\toauthtypes \"github.com/yaoapp/yao/openapi/oauth/types\"\n)\n\n// testAuthControl returns test auth info for control E2E tests\nfunc testAuthControl() *oauthtypes.AuthorizedInfo {\n\treturn &oauthtypes.AuthorizedInfo{\n\t\tUserID: \"e2e-control-user\",\n\t\tTeamID: \"e2e-control-team\",\n\t}\n}\n\n// TestE2EControlPauseResume tests pausing and resuming an execution\nfunc TestE2EControlPauseResume(t *testing.T) {\n\tif testing.Short() {\n\t\tt.Skip(\"Skipping E2E test - requires real LLM calls\")\n\t}\n\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\tcleanupE2ERobots(t)\n\tcleanupE2EExecutions(t)\n\tdefer cleanupE2ERobots(t)\n\tdefer cleanupE2EExecutions(t)\n\n\tt.Run(\"pause_and_resume_execution\", func(t *testing.T) {\n\t\tmemberID := \"robot_e2e_control_pause\"\n\t\tsetupE2ERobotForControl(t, memberID, \"team_e2e_control\")\n\n\t\terr := api.Start()\n\t\trequire.NoError(t, err)\n\t\tdefer api.Stop()\n\n\t\tctx := types.NewContext(context.Background(), testAuthControl())\n\n\t\t// Start execution\n\t\tresult, err := api.TriggerManual(ctx, memberID, types.TriggerClock, nil)\n\t\trequire.NoError(t, err)\n\t\trequire.True(t, result.Accepted)\n\n\t\tt.Logf(\"Execution started: ExecutionID=%s\", result.ExecutionID)\n\n\t\t// Wait for execution to start running\n\t\tvar execID string\n\t\tmaxWait := 30 * time.Second\n\t\tdeadline := time.Now().Add(maxWait)\n\n\t\tfor time.Now().Before(deadline) {\n\t\t\ttime.Sleep(500 * time.Millisecond)\n\n\t\t\texecutions, err := api.ListExecutions(ctx, memberID, nil)\n\t\t\tif err != nil || len(executions.Data) == 0 {\n\t\t\t\tcontinue\n\t\t\t}\n\n\t\t\texec := executions.Data[0]\n\t\t\tif exec.Status == types.ExecRunning {\n\t\t\t\texecID = exec.ID\n\t\t\t\tt.Logf(\"Execution running: ID=%s, Phase=%s\", execID, exec.Phase)\n\t\t\t\tbreak\n\t\t\t}\n\t\t}\n\n\t\tif execID == \"\" {\n\t\t\tt.Skip(\"Execution did not start in time - may have completed too quickly\")\n\t\t\treturn\n\t\t}\n\n\t\t// Pause the execution\n\t\terr = api.PauseExecution(ctx, execID)\n\t\tif err != nil {\n\t\t\tt.Logf(\"Pause error (may be expected if execution completed): %v\", err)\n\t\t} else {\n\t\t\tt.Logf(\"Execution paused\")\n\n\t\t\t// Verify paused state\n\t\t\ttime.Sleep(1 * time.Second)\n\t\t\tstatus, err := api.GetExecutionStatus(ctx, execID)\n\t\t\tif err == nil && status != nil {\n\t\t\t\tt.Logf(\"Status after pause: %s\", status.Status)\n\t\t\t}\n\n\t\t\t// Resume the execution\n\t\t\terr = api.ResumeExecution(ctx, execID)\n\t\t\tif err != nil {\n\t\t\t\tt.Logf(\"Resume error: %v\", err)\n\t\t\t} else {\n\t\t\t\tt.Logf(\"Execution resumed\")\n\t\t\t}\n\t\t}\n\n\t\t// Wait for completion\n\t\tmaxWait = 120 * time.Second\n\t\tdeadline = time.Now().Add(maxWait)\n\n\t\tfor time.Now().Before(deadline) {\n\t\t\ttime.Sleep(2 * time.Second)\n\n\t\t\texec, err := api.GetExecution(ctx, execID)\n\t\t\tif err != nil {\n\t\t\t\tcontinue\n\t\t\t}\n\n\t\t\tt.Logf(\"Execution status: %s, phase: %s\", exec.Status, exec.Phase)\n\n\t\t\tif exec.Status == types.ExecCompleted || exec.Status == types.ExecFailed || exec.Status == types.ExecCancelled {\n\t\t\t\t// Execution finished (completed, failed, or cancelled)\n\t\t\t\tt.Logf(\"Execution finished with status: %s\", exec.Status)\n\t\t\t\treturn\n\t\t\t}\n\t\t}\n\n\t\tt.Logf(\"Execution did not complete in time\")\n\t})\n}\n\n// TestE2EControlStop tests stopping an execution\nfunc TestE2EControlStop(t *testing.T) {\n\tif testing.Short() {\n\t\tt.Skip(\"Skipping E2E test - requires real LLM calls\")\n\t}\n\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\tcleanupE2ERobots(t)\n\tcleanupE2EExecutions(t)\n\tdefer cleanupE2ERobots(t)\n\tdefer cleanupE2EExecutions(t)\n\n\tt.Run(\"stop_running_execution\", func(t *testing.T) {\n\t\tmemberID := \"robot_e2e_control_stop\"\n\t\tsetupE2ERobotForControl(t, memberID, \"team_e2e_control\")\n\n\t\terr := api.Start()\n\t\trequire.NoError(t, err)\n\t\tdefer api.Stop()\n\n\t\tctx := types.NewContext(context.Background(), testAuthControl())\n\n\t\t// Start execution\n\t\tresult, err := api.TriggerManual(ctx, memberID, types.TriggerClock, nil)\n\t\trequire.NoError(t, err)\n\t\trequire.True(t, result.Accepted)\n\n\t\tt.Logf(\"Execution started: ExecutionID=%s\", result.ExecutionID)\n\n\t\t// Wait for execution to start running\n\t\tvar execID string\n\t\tmaxWait := 30 * time.Second\n\t\tdeadline := time.Now().Add(maxWait)\n\n\t\tfor time.Now().Before(deadline) {\n\t\t\ttime.Sleep(500 * time.Millisecond)\n\n\t\t\texecutions, err := api.ListExecutions(ctx, memberID, nil)\n\t\t\tif err != nil || len(executions.Data) == 0 {\n\t\t\t\tcontinue\n\t\t\t}\n\n\t\t\texec := executions.Data[0]\n\t\t\tif exec.Status == types.ExecRunning {\n\t\t\t\texecID = exec.ID\n\t\t\t\tt.Logf(\"Execution running: ID=%s, Phase=%s\", execID, exec.Phase)\n\t\t\t\tbreak\n\t\t\t}\n\t\t}\n\n\t\tif execID == \"\" {\n\t\t\tt.Skip(\"Execution did not start in time - may have completed too quickly\")\n\t\t\treturn\n\t\t}\n\n\t\t// Stop the execution\n\t\terr = api.StopExecution(ctx, execID)\n\t\tif err != nil {\n\t\t\tt.Logf(\"Stop error (may be expected if execution completed): %v\", err)\n\t\t} else {\n\t\t\tt.Logf(\"Stop signal sent\")\n\t\t}\n\n\t\t// Wait and verify stopped/cancelled state (with retry)\n\t\tmaxWait = 30 * time.Second\n\t\tdeadline = time.Now().Add(maxWait)\n\n\t\tfor time.Now().Before(deadline) {\n\t\t\ttime.Sleep(2 * time.Second)\n\n\t\t\texec, err := api.GetExecution(ctx, execID)\n\t\t\tif err != nil {\n\t\t\t\tt.Logf(\"Get execution error: %v\", err)\n\t\t\t\tcontinue\n\t\t\t}\n\n\t\t\tt.Logf(\"Current status: %s\", exec.Status)\n\n\t\t\t// Execution should eventually be cancelled, completed, or failed\n\t\t\tif exec.Status == types.ExecCancelled ||\n\t\t\t\texec.Status == types.ExecCompleted ||\n\t\t\t\texec.Status == types.ExecFailed {\n\t\t\t\tt.Logf(\"Final status: %s\", exec.Status)\n\t\t\t\treturn\n\t\t\t}\n\t\t}\n\n\t\t// If we get here, check final state\n\t\texec, err := api.GetExecution(ctx, execID)\n\t\tif err != nil {\n\t\t\tt.Logf(\"Get execution error: %v\", err)\n\t\t\treturn\n\t\t}\n\n\t\t// Allow running state if stop didn't take effect in time (execution may have already completed)\n\t\tt.Logf(\"Final status after wait: %s\", exec.Status)\n\t\tassert.True(t,\n\t\t\texec.Status == types.ExecCancelled ||\n\t\t\t\texec.Status == types.ExecCompleted ||\n\t\t\t\texec.Status == types.ExecFailed ||\n\t\t\t\texec.Status == types.ExecRunning, // Allow running if stop didn't take effect\n\t\t\t\"Execution should be in terminal state or still running, got: %s\", exec.Status)\n\t})\n}\n\n// TestE2EControlStopBeforeStart tests stopping an execution before it starts\nfunc TestE2EControlStopBeforeStart(t *testing.T) {\n\tif testing.Short() {\n\t\tt.Skip(\"Skipping E2E test - requires real LLM calls\")\n\t}\n\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\tcleanupE2ERobots(t)\n\tcleanupE2EExecutions(t)\n\tdefer cleanupE2ERobots(t)\n\tdefer cleanupE2EExecutions(t)\n\n\tt.Run(\"stop_queued_execution\", func(t *testing.T) {\n\t\tmemberID := \"robot_e2e_control_stop_early\"\n\t\tsetupE2ERobotForControl(t, memberID, \"team_e2e_control\")\n\n\t\terr := api.Start()\n\t\trequire.NoError(t, err)\n\t\tdefer api.Stop()\n\n\t\tctx := types.NewContext(context.Background(), testAuthControl())\n\n\t\t// Start execution\n\t\tresult, err := api.TriggerManual(ctx, memberID, types.TriggerClock, nil)\n\t\trequire.NoError(t, err)\n\t\trequire.True(t, result.Accepted)\n\n\t\t// Immediately try to get execution ID and stop\n\t\ttime.Sleep(100 * time.Millisecond)\n\n\t\texecutions, err := api.ListExecutions(ctx, memberID, nil)\n\t\tif err != nil || len(executions.Data) == 0 {\n\t\t\tt.Skip(\"No execution found\")\n\t\t\treturn\n\t\t}\n\n\t\texecID := executions.Data[0].ID\n\n\t\t// Try to stop immediately\n\t\terr = api.StopExecution(ctx, execID)\n\t\tif err != nil {\n\t\t\tt.Logf(\"Stop error: %v\", err)\n\t\t} else {\n\t\t\tt.Logf(\"Stop signal sent for execution %s\", execID)\n\t\t}\n\n\t\t// Wait and check status\n\t\ttime.Sleep(5 * time.Second)\n\n\t\texec, err := api.GetExecution(ctx, execID)\n\t\tif err != nil {\n\t\t\tt.Logf(\"Get execution error: %v\", err)\n\t\t\treturn\n\t\t}\n\n\t\tt.Logf(\"Final status: %s\", exec.Status)\n\t})\n}\n\n// TestE2EControlMultipleOperations tests a sequence of control operations\nfunc TestE2EControlMultipleOperations(t *testing.T) {\n\tif testing.Short() {\n\t\tt.Skip(\"Skipping E2E test - requires real LLM calls\")\n\t}\n\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\tcleanupE2ERobots(t)\n\tcleanupE2EExecutions(t)\n\tdefer cleanupE2ERobots(t)\n\tdefer cleanupE2EExecutions(t)\n\n\tt.Run(\"pause_resume_pause_stop_sequence\", func(t *testing.T) {\n\t\tmemberID := \"robot_e2e_control_multi\"\n\t\tsetupE2ERobotForControl(t, memberID, \"team_e2e_control\")\n\n\t\terr := api.Start()\n\t\trequire.NoError(t, err)\n\t\tdefer api.Stop()\n\n\t\tctx := types.NewContext(context.Background(), testAuthControl())\n\n\t\t// Start execution\n\t\tresult, err := api.TriggerManual(ctx, memberID, types.TriggerClock, nil)\n\t\trequire.NoError(t, err)\n\t\trequire.True(t, result.Accepted)\n\n\t\t// Wait for running state\n\t\tvar execID string\n\t\tmaxWait := 30 * time.Second\n\t\tdeadline := time.Now().Add(maxWait)\n\n\t\tfor time.Now().Before(deadline) {\n\t\t\ttime.Sleep(500 * time.Millisecond)\n\n\t\t\texecutions, err := api.ListExecutions(ctx, memberID, nil)\n\t\t\tif err != nil || len(executions.Data) == 0 {\n\t\t\t\tcontinue\n\t\t\t}\n\n\t\t\texec := executions.Data[0]\n\t\t\tif exec.Status == types.ExecRunning {\n\t\t\t\texecID = exec.ID\n\t\t\t\tbreak\n\t\t\t}\n\t\t}\n\n\t\tif execID == \"\" {\n\t\t\tt.Skip(\"Execution did not start in time\")\n\t\t\treturn\n\t\t}\n\n\t\t// Sequence: Pause → Resume → Pause → Stop\n\t\toperations := []struct {\n\t\t\tname string\n\t\t\tfn func() error\n\t\t}{\n\t\t\t{\"Pause\", func() error { return api.PauseExecution(ctx, execID) }},\n\t\t\t{\"Resume\", func() error { return api.ResumeExecution(ctx, execID) }},\n\t\t\t{\"Pause\", func() error { return api.PauseExecution(ctx, execID) }},\n\t\t\t{\"Stop\", func() error { return api.StopExecution(ctx, execID) }},\n\t\t}\n\n\t\tfor _, op := range operations {\n\t\t\terr := op.fn()\n\t\t\tif err != nil {\n\t\t\t\tt.Logf(\"%s error (may be expected): %v\", op.name, err)\n\t\t\t\t// If execution already completed, stop the sequence\n\t\t\t\texec, _ := api.GetExecution(ctx, execID)\n\t\t\t\tif exec != nil && (exec.Status == types.ExecCompleted || exec.Status == types.ExecFailed || exec.Status == types.ExecCancelled) {\n\t\t\t\t\tt.Logf(\"Execution already finished: %s\", exec.Status)\n\t\t\t\t\treturn\n\t\t\t\t}\n\t\t\t} else {\n\t\t\t\tt.Logf(\"%s successful\", op.name)\n\t\t\t}\n\t\t\ttime.Sleep(2 * time.Second)\n\t\t}\n\n\t\t// Verify final state\n\t\texec, err := api.GetExecution(ctx, execID)\n\t\tif err != nil {\n\t\t\tt.Logf(\"Get execution error: %v\", err)\n\t\t\treturn\n\t\t}\n\n\t\tt.Logf(\"Final status after operations: %s\", exec.Status)\n\t})\n}\n\n// TestE2EControlStatusQuery tests querying execution status during control\nfunc TestE2EControlStatusQuery(t *testing.T) {\n\tif testing.Short() {\n\t\tt.Skip(\"Skipping E2E test - requires real LLM calls\")\n\t}\n\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\tcleanupE2ERobots(t)\n\tcleanupE2EExecutions(t)\n\tdefer cleanupE2ERobots(t)\n\tdefer cleanupE2EExecutions(t)\n\n\tt.Run(\"query_status_during_execution\", func(t *testing.T) {\n\t\tmemberID := \"robot_e2e_control_status\"\n\t\tsetupE2ERobotForControl(t, memberID, \"team_e2e_control\")\n\n\t\terr := api.Start()\n\t\trequire.NoError(t, err)\n\t\tdefer api.Stop()\n\n\t\tctx := types.NewContext(context.Background(), testAuthControl())\n\n\t\t// Start execution\n\t\tresult, err := api.TriggerManual(ctx, memberID, types.TriggerClock, nil)\n\t\trequire.NoError(t, err)\n\t\trequire.True(t, result.Accepted)\n\n\t\t// Track status changes\n\t\tstatusHistory := make([]types.ExecStatus, 0)\n\t\tphaseHistory := make([]types.Phase, 0)\n\n\t\tmaxWait := 120 * time.Second\n\t\tdeadline := time.Now().Add(maxWait)\n\n\t\tlastStatus := types.ExecStatus(\"\")\n\t\tlastPhase := types.Phase(\"\")\n\n\t\tfor time.Now().Before(deadline) {\n\t\t\ttime.Sleep(1 * time.Second)\n\n\t\t\texecutions, err := api.ListExecutions(ctx, memberID, nil)\n\t\t\tif err != nil || len(executions.Data) == 0 {\n\t\t\t\tcontinue\n\t\t\t}\n\n\t\t\texec := executions.Data[0]\n\n\t\t\t// Track status changes\n\t\t\tif exec.Status != lastStatus {\n\t\t\t\tstatusHistory = append(statusHistory, exec.Status)\n\t\t\t\tlastStatus = exec.Status\n\t\t\t\tt.Logf(\"Status changed: %s\", exec.Status)\n\t\t\t}\n\n\t\t\t// Track phase changes\n\t\t\tif exec.Phase != lastPhase {\n\t\t\t\tphaseHistory = append(phaseHistory, exec.Phase)\n\t\t\t\tlastPhase = exec.Phase\n\t\t\t\tt.Logf(\"Phase changed: %s\", exec.Phase)\n\t\t\t}\n\n\t\t\t// Also test GetExecutionStatus\n\t\t\tstatus, err := api.GetExecutionStatus(ctx, exec.ID)\n\t\t\tif err == nil && status != nil {\n\t\t\t\t// Status query should return valid data\n\t\t\t\tassert.NotEmpty(t, status.ID)\n\t\t\t\tassert.Equal(t, exec.Status, status.Status)\n\t\t\t}\n\n\t\t\tif exec.Status == types.ExecCompleted || exec.Status == types.ExecFailed || exec.Status == types.ExecCancelled {\n\t\t\t\tbreak\n\t\t\t}\n\t\t}\n\n\t\tt.Logf(\"Status history: %v\", statusHistory)\n\t\tt.Logf(\"Phase history: %v\", phaseHistory)\n\n\t\t// Should have observed at least pending → running transition\n\t\tassert.GreaterOrEqual(t, len(statusHistory), 1, \"Should observe at least one status\")\n\t})\n}\n\n// ==================== Helper Functions ====================\n\n// setupE2ERobotForControl creates a robot for control tests\nfunc setupE2ERobotForControl(t *testing.T, memberID, teamID string) {\n\tm := model.Select(\"__yao.member\")\n\ttableName := m.MetaData.Table.Name\n\tqb := capsule.Query()\n\n\t// Simple config for E2E testing - minimal tasks\n\trobotConfig := map[string]interface{}{\n\t\t\"identity\": map[string]interface{}{\n\t\t\t\"role\": \"Simple E2E Test Robot\",\n\t\t\t\"duties\": []string{\"Say hello\"}, // Very simple duty\n\t\t\t\"rules\": []string{\"Keep responses under 50 words\"},\n\t\t},\n\t\t\"quota\": map[string]interface{}{\n\t\t\t\"max\": 3,\n\t\t\t\"queue\": 10,\n\t\t\t\"priority\": 5,\n\t\t},\n\t\t\"triggers\": map[string]interface{}{\n\t\t\t\"clock\": map[string]interface{}{\"enabled\": true},\n\t\t\t\"intervene\": map[string]interface{}{\"enabled\": true},\n\t\t\t\"event\": map[string]interface{}{\"enabled\": true},\n\t\t},\n\t\t\"clock\": map[string]interface{}{\n\t\t\t\"mode\": \"interval\",\n\t\t\t\"every\": \"1h\",\n\t\t},\n\t\t\"resources\": map[string]interface{}{\n\t\t\t\"phases\": map[string]interface{}{\n\t\t\t\t\"inspiration\": \"robot.inspiration\",\n\t\t\t\t\"goals\": \"robot.goals\",\n\t\t\t\t\"tasks\": \"tests.e2e-tasks\", // Use simple E2E test task planner\n\t\t\t\t\"run\": \"robot.validation\",\n\t\t\t\t\"validation\": \"tests.e2e-validation\", // Use lenient E2E test validator\n\t\t\t\t\"delivery\": \"robot.delivery\",\n\t\t\t\t\"learning\": \"robot.learning\",\n\t\t\t},\n\t\t\t\"agents\": []string{\"experts.text-writer\"},\n\t\t},\n\t\t\"delivery\": map[string]interface{}{\n\t\t\t\"email\": map[string]interface{}{\"enabled\": false},\n\t\t\t\"webhook\": map[string]interface{}{\"enabled\": false},\n\t\t\t\"process\": map[string]interface{}{\"enabled\": false},\n\t\t},\n\t}\n\tconfigJSON, _ := json.Marshal(robotConfig)\n\n\tsystemPrompt := `You are a simple E2E test robot. Your job is to say hello.\nWhen generating goals: create exactly 1 simple goal.\nWhen generating tasks: create exactly 1 simple task.\nKeep all outputs brief. No complex analysis needed.`\n\n\terr := qb.Table(tableName).Insert([]map[string]interface{}{\n\t\t{\n\t\t\t\"member_id\": memberID,\n\t\t\t\"team_id\": teamID,\n\t\t\t\"member_type\": \"robot\",\n\t\t\t\"display_name\": \"E2E Control Robot \" + memberID,\n\t\t\t\"system_prompt\": systemPrompt,\n\t\t\t\"status\": \"active\",\n\t\t\t\"role_id\": \"member\",\n\t\t\t\"autonomous_mode\": true,\n\t\t\t\"robot_status\": \"idle\",\n\t\t\t\"robot_config\": string(configJSON),\n\t\t},\n\t})\n\tif err != nil {\n\t\tt.Fatalf(\"Failed to insert robot %s: %v\", memberID, err)\n\t}\n}\n" }, { "path": "agent/robot/api/e2e_event_test.go", "content": "package api_test\n\n// End-to-end tests for Event trigger flow\n// These tests use REAL LLM calls via Standard executor (not DryRun)\n//\n// Test Flow: Event Trigger → P1 (Goals) → P2 (Tasks) → P3 (Run) → P4 (Delivery)\n// Note: Event trigger SKIPS P0 (Inspiration) - event data provides the context\n//\n// Prerequisites:\n// - Valid LLM API keys (OPENAI_TEST_KEY or DEEPSEEK_API_KEY)\n// - Test assistants in yao-dev-app/assistants/robot/\n// - Database connection (YAO_DB_PRIMARY)\n\nimport (\n\t\"context\"\n\t\"encoding/json\"\n\t\"testing\"\n\t\"time\"\n\n\t\"github.com/stretchr/testify/assert\"\n\t\"github.com/stretchr/testify/require\"\n\t\"github.com/yaoapp/gou/model\"\n\t\"github.com/yaoapp/xun/capsule\"\n\t\"github.com/yaoapp/yao/agent/robot/api\"\n\t\"github.com/yaoapp/yao/agent/robot/types\"\n\t\"github.com/yaoapp/yao/agent/testutils\"\n\toauthtypes \"github.com/yaoapp/yao/openapi/oauth/types\"\n)\n\n// testAuthEvent returns test auth info for event E2E tests\nfunc testAuthEvent() *oauthtypes.AuthorizedInfo {\n\treturn &oauthtypes.AuthorizedInfo{\n\t\tUserID: \"e2e-event-user\",\n\t\tTeamID: \"e2e-event-team\",\n\t}\n}\n\n// TestE2EEventTriggerFullFlow tests the complete event trigger flow with real LLM calls\n// Flow: Event → P1 (Goals) → P2 (Tasks) → P3 (Run) → P4 (Delivery)\nfunc TestE2EEventTriggerFullFlow(t *testing.T) {\n\tif testing.Short() {\n\t\tt.Skip(\"Skipping E2E test - requires real LLM calls\")\n\t}\n\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\tcleanupE2ERobots(t)\n\tcleanupE2EExecutions(t)\n\tdefer cleanupE2ERobots(t)\n\tdefer cleanupE2EExecutions(t)\n\n\tt.Run(\"complete_P1_to_P4_flow_with_webhook_event\", func(t *testing.T) {\n\t\tmemberID := \"robot_e2e_event_001\"\n\t\tsetupE2ERobotForEvent(t, memberID, \"team_e2e_event\")\n\n\t\terr := api.Start()\n\t\trequire.NoError(t, err)\n\t\tdefer api.Stop()\n\n\t\tctx := types.NewContext(context.Background(), testAuthEvent())\n\n\t\t// Verify robot is loaded\n\t\trobot, err := api.GetRobot(ctx, memberID)\n\t\trequire.NoError(t, err)\n\t\trequire.NotNil(t, robot)\n\n\t\t// Trigger with webhook event - simulating external system notification\n\t\tresult, err := api.Trigger(ctx, memberID, &api.TriggerRequest{\n\t\t\tType: types.TriggerEvent,\n\t\t\tSource: types.EventWebhook,\n\t\t\tEventType: \"order.created\",\n\t\t\tData: map[string]interface{}{\n\t\t\t\t\"order_id\": \"ORD-2025-001\",\n\t\t\t\t\"customer\": \"John Doe\",\n\t\t\t\t\"total\": 299.99,\n\t\t\t\t\"items_count\": 3,\n\t\t\t\t\"priority\": \"high\",\n\t\t\t\t\"created_at\": time.Now().Format(time.RFC3339),\n\t\t\t},\n\t\t})\n\t\trequire.NoError(t, err)\n\t\trequire.NotNil(t, result)\n\t\tassert.True(t, result.Accepted, \"Event trigger should be accepted\")\n\n\t\tt.Logf(\"Event trigger result: Accepted=%v, ExecutionID=%s\", result.Accepted, result.ExecutionID)\n\n\t\t// Wait for execution to complete\n\t\tvar exec *types.Execution\n\t\tmaxWait := 120 * time.Second\n\t\tpollInterval := 2 * time.Second\n\t\tdeadline := time.Now().Add(maxWait)\n\n\t\tfor time.Now().Before(deadline) {\n\t\t\ttime.Sleep(pollInterval)\n\n\t\t\texecutions, err := api.ListExecutions(ctx, memberID, &api.ExecutionQuery{\n\t\t\t\tPage: 1,\n\t\t\t\tPageSize: 1,\n\t\t\t})\n\t\t\tif err != nil || len(executions.Data) == 0 {\n\t\t\t\tcontinue\n\t\t\t}\n\n\t\t\texec = executions.Data[0]\n\t\t\tt.Logf(\"Execution status: %s, phase: %s\", exec.Status, exec.Phase)\n\n\t\t\tif exec.Status == types.ExecCompleted || exec.Status == types.ExecFailed {\n\t\t\t\tbreak\n\t\t\t}\n\t\t}\n\n\t\trequire.NotNil(t, exec, \"Execution should exist\")\n\n\t\t// E2E test validates the flow executes correctly\n\t\tisFinished := exec.Status == types.ExecCompleted || exec.Status == types.ExecFailed\n\t\tassert.True(t, isFinished, \"Execution should finish (completed or failed), got: %s\", exec.Status)\n\n\t\tif exec.Status == types.ExecFailed {\n\t\t\tt.Logf(\"Execution finished with status=failed (acceptable for E2E): %s\", exec.Error)\n\t\t} else {\n\t\t\tt.Logf(\"Execution finished with status=completed\")\n\t\t}\n\n\t\t// Verify trigger type\n\t\tassert.Equal(t, types.TriggerEvent, exec.TriggerType, \"Should be event trigger\")\n\n\t\t// Event trigger skips P0, so Inspiration should be nil\n\t\tassert.Nil(t, exec.Inspiration, \"P0 Inspiration should be nil for event trigger\")\n\n\t\t// P1 Goals should always exist for event trigger\n\t\tassert.NotNil(t, exec.Goals, \"P1 Goals should exist\")\n\n\t\t// P2-P4 may or may not exist depending on where failure occurred\n\t\tif exec.Tasks != nil {\n\t\t\tt.Logf(\"P2 Tasks count: %d\", len(exec.Tasks))\n\t\t}\n\t\tif exec.Results != nil {\n\t\t\tt.Logf(\"P3 Results count: %d\", len(exec.Results))\n\t\t}\n\t\tif exec.Delivery != nil {\n\t\t\tt.Logf(\"P4 Delivery: RequestID=%s\", exec.Delivery.RequestID)\n\t\t}\n\n\t\tt.Logf(\"Event trigger E2E completed\")\n\t})\n}\n\n// TestE2EEventTriggerDatabaseEvent tests event trigger from database changes\nfunc TestE2EEventTriggerDatabaseEvent(t *testing.T) {\n\tif testing.Short() {\n\t\tt.Skip(\"Skipping E2E test - requires real LLM calls\")\n\t}\n\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\tcleanupE2ERobots(t)\n\tcleanupE2EExecutions(t)\n\tdefer cleanupE2ERobots(t)\n\tdefer cleanupE2EExecutions(t)\n\n\tt.Run(\"handles_database_event_source\", func(t *testing.T) {\n\t\tmemberID := \"robot_e2e_event_db\"\n\t\tsetupE2ERobotForEvent(t, memberID, \"team_e2e_event\")\n\n\t\terr := api.Start()\n\t\trequire.NoError(t, err)\n\t\tdefer api.Stop()\n\n\t\tctx := types.NewContext(context.Background(), testAuthEvent())\n\n\t\t// Trigger with database event - simulating record change notification\n\t\tresult, err := api.Trigger(ctx, memberID, &api.TriggerRequest{\n\t\t\tType: types.TriggerEvent,\n\t\t\tSource: types.EventDatabase,\n\t\t\tEventType: \"user.updated\",\n\t\t\tData: map[string]interface{}{\n\t\t\t\t\"table\": \"users\",\n\t\t\t\t\"operation\": \"UPDATE\",\n\t\t\t\t\"record_id\": 12345,\n\t\t\t\t\"changes\": map[string]interface{}{\n\t\t\t\t\t\"status\": map[string]interface{}{\n\t\t\t\t\t\t\"old\": \"pending\",\n\t\t\t\t\t\t\"new\": \"active\",\n\t\t\t\t\t},\n\t\t\t\t},\n\t\t\t},\n\t\t})\n\t\trequire.NoError(t, err)\n\t\trequire.NotNil(t, result)\n\t\tassert.True(t, result.Accepted)\n\n\t\t// Wait for execution\n\t\tmaxWait := 120 * time.Second\n\t\tdeadline := time.Now().Add(maxWait)\n\n\t\tfor time.Now().Before(deadline) {\n\t\t\ttime.Sleep(2 * time.Second)\n\n\t\t\texecutions, err := api.ListExecutions(ctx, memberID, nil)\n\t\t\tif err != nil || len(executions.Data) == 0 {\n\t\t\t\tcontinue\n\t\t\t}\n\n\t\t\texec := executions.Data[0]\n\t\t\tif exec.Status == types.ExecCompleted || exec.Status == types.ExecFailed {\n\t\t\t\tassert.Equal(t, types.ExecCompleted, exec.Status)\n\t\t\t\tt.Logf(\"Database event E2E completed\")\n\t\t\t\treturn\n\t\t\t}\n\t\t}\n\n\t\tt.Fatal(\"Execution did not complete in time\")\n\t})\n}\n\n// TestE2EEventTriggerVariousEventTypes tests different event types\n// Optimized: Only tests one representative event type to reduce CI time\n// The event handling logic is the same for all event types, so testing one is sufficient\nfunc TestE2EEventTriggerVariousEventTypes(t *testing.T) {\n\tif testing.Short() {\n\t\tt.Skip(\"Skipping E2E test - requires real LLM calls\")\n\t}\n\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\tcleanupE2ERobots(t)\n\tcleanupE2EExecutions(t)\n\tdefer cleanupE2ERobots(t)\n\tdefer cleanupE2EExecutions(t)\n\n\t// Test only one representative event type (notification)\n\t// All event types use the same code path, so one test is sufficient\n\tt.Run(\"webhook_event\", func(t *testing.T) {\n\t\tmemberID := \"robot_e2e_event_webhook\"\n\t\tsetupE2ERobotForEvent(t, memberID, \"team_e2e_event\")\n\n\t\terr := api.Start()\n\t\trequire.NoError(t, err)\n\t\tdefer api.Stop()\n\n\t\tctx := types.NewContext(context.Background(), testAuthEvent())\n\n\t\tresult, err := api.Trigger(ctx, memberID, &api.TriggerRequest{\n\t\t\tType: types.TriggerEvent,\n\t\t\tSource: types.EventWebhook,\n\t\t\tEventType: \"notification.received\",\n\t\t\tData: map[string]interface{}{\n\t\t\t\t\"message\": \"Test notification\",\n\t\t\t\t\"priority\": \"normal\",\n\t\t\t},\n\t\t})\n\t\trequire.NoError(t, err)\n\t\trequire.NotNil(t, result)\n\t\tassert.True(t, result.Accepted, \"Event should be accepted\")\n\n\t\tt.Logf(\"Event triggered: ExecutionID=%s\", result.ExecutionID)\n\n\t\t// Wait for execution\n\t\tmaxWait := 120 * time.Second\n\t\tdeadline := time.Now().Add(maxWait)\n\n\t\tfor time.Now().Before(deadline) {\n\t\t\ttime.Sleep(2 * time.Second)\n\n\t\t\texecutions, err := api.ListExecutions(ctx, memberID, nil)\n\t\t\tif err != nil || len(executions.Data) == 0 {\n\t\t\t\tcontinue\n\t\t\t}\n\n\t\t\texec := executions.Data[0]\n\t\t\tt.Logf(\"Event status: %s, phase: %s\", exec.Status, exec.Phase)\n\n\t\t\tif exec.Status == types.ExecCompleted || exec.Status == types.ExecFailed {\n\t\t\t\tif exec.Status == types.ExecFailed {\n\t\t\t\t\tt.Logf(\"Event execution failed: %s\", exec.Error)\n\t\t\t\t} else {\n\t\t\t\t\tt.Logf(\"Event execution completed\")\n\t\t\t\t}\n\t\t\t\treturn\n\t\t\t}\n\t\t}\n\n\t\tt.Logf(\"Event execution did not complete in time (may be CI latency)\")\n\t})\n}\n\n// TestE2EEventTriggerWithComplexData tests event with nested/complex data structures\nfunc TestE2EEventTriggerWithComplexData(t *testing.T) {\n\tif testing.Short() {\n\t\tt.Skip(\"Skipping E2E test - requires real LLM calls\")\n\t}\n\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\tcleanupE2ERobots(t)\n\tcleanupE2EExecutions(t)\n\tdefer cleanupE2ERobots(t)\n\tdefer cleanupE2EExecutions(t)\n\n\tt.Run(\"handles_complex_nested_data\", func(t *testing.T) {\n\t\tmemberID := \"robot_e2e_event_complex\"\n\t\tsetupE2ERobotForEvent(t, memberID, \"team_e2e_event\")\n\n\t\terr := api.Start()\n\t\trequire.NoError(t, err)\n\t\tdefer api.Stop()\n\n\t\tctx := types.NewContext(context.Background(), testAuthEvent())\n\n\t\t// Complex nested event data\n\t\tresult, err := api.Trigger(ctx, memberID, &api.TriggerRequest{\n\t\t\tType: types.TriggerEvent,\n\t\t\tSource: types.EventWebhook,\n\t\t\tEventType: \"report.generated\",\n\t\t\tData: map[string]interface{}{\n\t\t\t\t\"report\": map[string]interface{}{\n\t\t\t\t\t\"id\": \"RPT-2025-001\",\n\t\t\t\t\t\"type\": \"sales_summary\",\n\t\t\t\t\t\"period\": \"monthly\",\n\t\t\t\t\t\"generated\": time.Now().Format(time.RFC3339),\n\t\t\t\t\t\"department\": \"Sales\",\n\t\t\t\t},\n\t\t\t\t\"metrics\": []map[string]interface{}{\n\t\t\t\t\t{\"name\": \"total_sales\", \"value\": 150000, \"unit\": \"USD\"},\n\t\t\t\t\t{\"name\": \"orders_count\", \"value\": 450, \"unit\": \"orders\"},\n\t\t\t\t\t{\"name\": \"avg_order_value\", \"value\": 333.33, \"unit\": \"USD\"},\n\t\t\t\t},\n\t\t\t\t\"comparison\": map[string]interface{}{\n\t\t\t\t\t\"previous_period\": map[string]interface{}{\n\t\t\t\t\t\t\"total_sales\": 140000,\n\t\t\t\t\t\t\"orders_count\": 420,\n\t\t\t\t\t\t\"change_percent\": 7.14,\n\t\t\t\t\t},\n\t\t\t\t},\n\t\t\t\t\"highlights\": []string{\n\t\t\t\t\t\"Sales increased by 7.14% compared to last month\",\n\t\t\t\t\t\"Top performing product: Widget Pro\",\n\t\t\t\t\t\"New customer acquisition up 15%\",\n\t\t\t\t},\n\t\t\t},\n\t\t})\n\t\trequire.NoError(t, err)\n\t\trequire.NotNil(t, result)\n\t\tassert.True(t, result.Accepted)\n\n\t\t// Wait for execution\n\t\tmaxWait := 120 * time.Second\n\t\tdeadline := time.Now().Add(maxWait)\n\n\t\tfor time.Now().Before(deadline) {\n\t\t\ttime.Sleep(2 * time.Second)\n\n\t\t\texecutions, err := api.ListExecutions(ctx, memberID, nil)\n\t\t\tif err != nil || len(executions.Data) == 0 {\n\t\t\t\tcontinue\n\t\t\t}\n\n\t\t\texec := executions.Data[0]\n\t\t\tif exec.Status == types.ExecCompleted || exec.Status == types.ExecFailed {\n\t\t\t\tassert.Equal(t, types.ExecCompleted, exec.Status, \"Complex data event should complete\")\n\t\t\t\tt.Logf(\"Complex data event E2E completed\")\n\t\t\t\treturn\n\t\t\t}\n\t\t}\n\n\t\tt.Fatal(\"Execution did not complete in time\")\n\t})\n}\n\n// ==================== Helper Functions ====================\n\n// setupE2ERobotForEvent creates a robot configured for event trigger E2E tests\nfunc setupE2ERobotForEvent(t *testing.T, memberID, teamID string) {\n\tm := model.Select(\"__yao.member\")\n\ttableName := m.MetaData.Table.Name\n\tqb := capsule.Query()\n\n\t// Simple config for E2E testing - minimal tasks\n\trobotConfig := map[string]interface{}{\n\t\t\"identity\": map[string]interface{}{\n\t\t\t\"role\": \"Simple E2E Test Robot\",\n\t\t\t\"duties\": []string{\"Acknowledge events\"}, // Very simple duty\n\t\t\t\"rules\": []string{\"Keep responses under 50 words\"},\n\t\t},\n\t\t\"quota\": map[string]interface{}{\n\t\t\t\"max\": 5,\n\t\t\t\"queue\": 20,\n\t\t\t\"priority\": 5,\n\t\t},\n\t\t\"triggers\": map[string]interface{}{\n\t\t\t\"clock\": map[string]interface{}{\"enabled\": false},\n\t\t\t\"intervene\": map[string]interface{}{\"enabled\": true},\n\t\t\t\"event\": map[string]interface{}{\"enabled\": true},\n\t\t},\n\t\t\"event\": map[string]interface{}{\n\t\t\t\"types\": []string{\"*\"},\n\t\t},\n\t\t\"resources\": map[string]interface{}{\n\t\t\t\"phases\": map[string]interface{}{\n\t\t\t\t\"inspiration\": \"robot.inspiration\",\n\t\t\t\t\"goals\": \"robot.goals\",\n\t\t\t\t\"tasks\": \"tests.e2e-tasks\", // Use simple E2E test task planner\n\t\t\t\t\"run\": \"robot.validation\",\n\t\t\t\t\"validation\": \"tests.e2e-validation\", // Use lenient E2E test validator\n\t\t\t\t\"delivery\": \"robot.delivery\",\n\t\t\t\t\"learning\": \"robot.learning\",\n\t\t\t},\n\t\t\t\"agents\": []string{\"experts.text-writer\"},\n\t\t},\n\t\t\"delivery\": map[string]interface{}{\n\t\t\t\"email\": map[string]interface{}{\"enabled\": false},\n\t\t\t\"webhook\": map[string]interface{}{\"enabled\": false},\n\t\t\t\"process\": map[string]interface{}{\"enabled\": false},\n\t\t},\n\t}\n\tconfigJSON, _ := json.Marshal(robotConfig)\n\n\tsystemPrompt := `You are a simple E2E test robot. Your job is to acknowledge events.\nWhen generating goals: create exactly 1 simple goal.\nWhen generating tasks: create exactly 1 simple task.\nKeep all outputs brief. No complex analysis needed.`\n\n\terr := qb.Table(tableName).Insert([]map[string]interface{}{\n\t\t{\n\t\t\t\"member_id\": memberID,\n\t\t\t\"team_id\": teamID,\n\t\t\t\"member_type\": \"robot\",\n\t\t\t\"display_name\": \"E2E Event Test Robot \" + memberID,\n\t\t\t\"system_prompt\": systemPrompt,\n\t\t\t\"status\": \"active\",\n\t\t\t\"role_id\": \"member\",\n\t\t\t\"autonomous_mode\": true,\n\t\t\t\"robot_status\": \"idle\",\n\t\t\t\"robot_config\": string(configJSON),\n\t\t},\n\t})\n\tif err != nil {\n\t\tt.Fatalf(\"Failed to insert robot %s: %v\", memberID, err)\n\t}\n}\n" }, { "path": "agent/robot/api/e2e_human_test.go", "content": "package api_test\n\n// End-to-end tests for Human intervention trigger flow\n// These tests use REAL LLM calls via Standard executor (not DryRun)\n//\n// Test Flow: Human Trigger → P1 (Goals) → P2 (Tasks) → P3 (Run) → P4 (Delivery)\n// Note: Human trigger SKIPS P0 (Inspiration) - user provides the input directly\n//\n// Prerequisites:\n// - Valid LLM API keys (OPENAI_TEST_KEY or DEEPSEEK_API_KEY)\n// - Test assistants in yao-dev-app/assistants/robot/\n// - Database connection (YAO_DB_PRIMARY)\n\nimport (\n\t\"context\"\n\t\"encoding/json\"\n\t\"testing\"\n\t\"time\"\n\n\t\"github.com/stretchr/testify/assert\"\n\t\"github.com/stretchr/testify/require\"\n\t\"github.com/yaoapp/gou/model\"\n\t\"github.com/yaoapp/xun/capsule\"\n\tagentcontext \"github.com/yaoapp/yao/agent/context\"\n\t\"github.com/yaoapp/yao/agent/robot/api\"\n\t\"github.com/yaoapp/yao/agent/robot/types\"\n\t\"github.com/yaoapp/yao/agent/testutils\"\n\toauthtypes \"github.com/yaoapp/yao/openapi/oauth/types\"\n)\n\n// testAuthHuman returns test auth info for human E2E tests\nfunc testAuthHuman() *oauthtypes.AuthorizedInfo {\n\treturn &oauthtypes.AuthorizedInfo{\n\t\tUserID: \"e2e-human-user\",\n\t\tTeamID: \"e2e-human-team\",\n\t}\n}\n\n// TestE2EHumanTriggerFullFlow tests the complete human intervention flow with real LLM calls\n// Flow: Human Input → P1 (Goals) → P2 (Tasks) → P3 (Run) → P4 (Delivery)\nfunc TestE2EHumanTriggerFullFlow(t *testing.T) {\n\tif testing.Short() {\n\t\tt.Skip(\"Skipping E2E test - requires real LLM calls\")\n\t}\n\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\tcleanupE2ERobots(t)\n\tcleanupE2EExecutions(t)\n\tdefer cleanupE2ERobots(t)\n\tdefer cleanupE2EExecutions(t)\n\n\tt.Run(\"complete_P1_to_P4_flow_with_user_input\", func(t *testing.T) {\n\t\tmemberID := \"robot_e2e_human_001\"\n\t\tsetupE2ERobotForHuman(t, memberID, \"team_e2e_human\")\n\n\t\terr := api.Start()\n\t\trequire.NoError(t, err)\n\t\tdefer api.Stop()\n\n\t\tctx := types.NewContext(context.Background(), testAuthHuman())\n\n\t\t// Verify robot is loaded\n\t\trobot, err := api.GetRobot(ctx, memberID)\n\t\trequire.NoError(t, err)\n\t\trequire.NotNil(t, robot)\n\n\t\t// Trigger with human input - user requesting a specific task\n\t\tresult, err := api.Trigger(ctx, memberID, &api.TriggerRequest{\n\t\t\tType: types.TriggerHuman,\n\t\t\tAction: types.ActionTaskAdd,\n\t\t\tMessages: []agentcontext.Message{\n\t\t\t\t{\n\t\t\t\t\tRole: \"user\",\n\t\t\t\t\tContent: \"Please write a brief summary of today's key tasks and priorities.\",\n\t\t\t\t},\n\t\t\t},\n\t\t})\n\t\trequire.NoError(t, err)\n\t\trequire.NotNil(t, result)\n\n\t\t// Human trigger returns Queued=true (goes through Intervene)\n\t\tt.Logf(\"Trigger result: Accepted=%v, Queued=%v, Message=%s\",\n\t\t\tresult.Accepted, result.Queued, result.Message)\n\n\t\t// Wait for execution to complete\n\t\tvar exec *types.Execution\n\t\tmaxWait := 120 * time.Second\n\t\tpollInterval := 2 * time.Second\n\t\tdeadline := time.Now().Add(maxWait)\n\n\t\tfor time.Now().Before(deadline) {\n\t\t\ttime.Sleep(pollInterval)\n\n\t\t\texecutions, err := api.ListExecutions(ctx, memberID, &api.ExecutionQuery{\n\t\t\t\tPage: 1,\n\t\t\t\tPageSize: 1,\n\t\t\t})\n\t\t\tif err != nil || len(executions.Data) == 0 {\n\t\t\t\tcontinue\n\t\t\t}\n\n\t\t\texec = executions.Data[0]\n\t\t\tt.Logf(\"Execution status: %s, phase: %s\", exec.Status, exec.Phase)\n\n\t\t\tif exec.Status == types.ExecCompleted || exec.Status == types.ExecFailed {\n\t\t\t\tbreak\n\t\t\t}\n\t\t}\n\n\t\trequire.NotNil(t, exec, \"Execution should exist\")\n\n\t\t// E2E test validates the flow executes correctly\n\t\tisFinished := exec.Status == types.ExecCompleted || exec.Status == types.ExecFailed\n\t\tassert.True(t, isFinished, \"Execution should finish (completed or failed), got: %s\", exec.Status)\n\n\t\tif exec.Status == types.ExecFailed {\n\t\t\tt.Logf(\"Execution finished with status=failed (acceptable for E2E): %s\", exec.Error)\n\t\t} else {\n\t\t\tt.Logf(\"Execution finished with status=completed\")\n\t\t}\n\n\t\t// Verify trigger type\n\t\tassert.Equal(t, types.TriggerHuman, exec.TriggerType, \"Should be human trigger\")\n\n\t\t// Human trigger skips P0, so Inspiration should be nil\n\t\tassert.Nil(t, exec.Inspiration, \"P0 Inspiration should be nil for human trigger\")\n\n\t\t// P1 Goals should always exist for human trigger\n\t\tassert.NotNil(t, exec.Goals, \"P1 Goals should exist\")\n\n\t\t// P2-P4 may or may not exist depending on where failure occurred\n\t\tif exec.Tasks != nil {\n\t\t\tt.Logf(\"P2 Tasks count: %d\", len(exec.Tasks))\n\t\t}\n\t\tif exec.Results != nil {\n\t\t\tt.Logf(\"P3 Results count: %d\", len(exec.Results))\n\t\t}\n\t\tif exec.Delivery != nil {\n\t\t\tt.Logf(\"P4 Delivery: RequestID=%s\", exec.Delivery.RequestID)\n\t\t}\n\n\t\tt.Logf(\"Human trigger E2E completed\")\n\t})\n}\n\n// TestE2EHumanTriggerWithMultimodalInput tests human trigger with rich content\nfunc TestE2EHumanTriggerWithMultimodalInput(t *testing.T) {\n\tif testing.Short() {\n\t\tt.Skip(\"Skipping E2E test - requires real LLM calls\")\n\t}\n\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\tcleanupE2ERobots(t)\n\tcleanupE2EExecutions(t)\n\tdefer cleanupE2ERobots(t)\n\tdefer cleanupE2EExecutions(t)\n\n\tt.Run(\"handles_multipart_message_input\", func(t *testing.T) {\n\t\tmemberID := \"robot_e2e_human_multi\"\n\t\tsetupE2ERobotForHuman(t, memberID, \"team_e2e_human\")\n\n\t\terr := api.Start()\n\t\trequire.NoError(t, err)\n\t\tdefer api.Stop()\n\n\t\tctx := types.NewContext(context.Background(), testAuthHuman())\n\n\t\t// Trigger with multipart message (text parts)\n\t\tresult, err := api.Trigger(ctx, memberID, &api.TriggerRequest{\n\t\t\tType: types.TriggerHuman,\n\t\t\tAction: types.ActionGoalAdjust,\n\t\t\tMessages: []agentcontext.Message{\n\t\t\t\t{\n\t\t\t\t\tRole: \"user\",\n\t\t\t\t\tContent: []map[string]interface{}{\n\t\t\t\t\t\t{\n\t\t\t\t\t\t\t\"type\": \"text\",\n\t\t\t\t\t\t\t\"text\": \"I need you to focus on the following priorities:\",\n\t\t\t\t\t\t},\n\t\t\t\t\t\t{\n\t\t\t\t\t\t\t\"type\": \"text\",\n\t\t\t\t\t\t\t\"text\": \"1. Review pending tasks\\n2. Summarize progress\\n3. Identify blockers\",\n\t\t\t\t\t\t},\n\t\t\t\t\t},\n\t\t\t\t},\n\t\t\t},\n\t\t})\n\t\trequire.NoError(t, err)\n\t\trequire.NotNil(t, result)\n\n\t\tt.Logf(\"Multipart trigger result: Queued=%v\", result.Queued)\n\n\t\t// Wait for execution\n\t\tmaxWait := 120 * time.Second\n\t\tdeadline := time.Now().Add(maxWait)\n\n\t\tfor time.Now().Before(deadline) {\n\t\t\ttime.Sleep(2 * time.Second)\n\n\t\t\texecutions, err := api.ListExecutions(ctx, memberID, nil)\n\t\t\tif err != nil || len(executions.Data) == 0 {\n\t\t\t\tcontinue\n\t\t\t}\n\n\t\t\texec := executions.Data[0]\n\t\t\tif exec.Status == types.ExecCompleted || exec.Status == types.ExecFailed {\n\t\t\t\tassert.Equal(t, types.ExecCompleted, exec.Status)\n\t\t\t\tt.Logf(\"Multipart input E2E completed\")\n\t\t\t\treturn\n\t\t\t}\n\t\t}\n\n\t\tt.Fatal(\"Execution did not complete in time\")\n\t})\n}\n\n// TestE2EHumanTriggerAllActions tests different intervention actions\nfunc TestE2EHumanTriggerAllActions(t *testing.T) {\n\tif testing.Short() {\n\t\tt.Skip(\"Skipping E2E test - requires real LLM calls\")\n\t}\n\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\tcleanupE2ERobots(t)\n\tcleanupE2EExecutions(t)\n\tdefer cleanupE2ERobots(t)\n\tdefer cleanupE2EExecutions(t)\n\n\tactions := []struct {\n\t\tname string\n\t\taction types.InterventionAction\n\t\tinput string\n\t}{\n\t\t{\n\t\t\tname: \"task_add\",\n\t\t\taction: types.ActionTaskAdd,\n\t\t\tinput: \"Add a new task: Review system logs for errors\",\n\t\t},\n\t\t{\n\t\t\tname: \"goal_adjust\",\n\t\t\taction: types.ActionGoalAdjust,\n\t\t\tinput: \"Adjust goal: Focus on performance optimization instead of new features\",\n\t\t},\n\t\t{\n\t\t\tname: \"instruct\",\n\t\t\taction: types.ActionInstruct,\n\t\t\tinput: \"Please prioritize security review as the top task\",\n\t\t},\n\t}\n\n\tfor i, tc := range actions {\n\t\tt.Run(tc.name, func(t *testing.T) {\n\t\t\tmemberID := \"robot_e2e_human_action_\" + tc.name\n\t\t\tsetupE2ERobotForHuman(t, memberID, \"team_e2e_human\")\n\n\t\t\t// Start fresh for each action test\n\t\t\tif i == 0 {\n\t\t\t\terr := api.Start()\n\t\t\t\trequire.NoError(t, err)\n\t\t\t}\n\n\t\t\tctx := types.NewContext(context.Background(), testAuthHuman())\n\n\t\t\tresult, err := api.Trigger(ctx, memberID, &api.TriggerRequest{\n\t\t\t\tType: types.TriggerHuman,\n\t\t\t\tAction: tc.action,\n\t\t\t\tMessages: []agentcontext.Message{\n\t\t\t\t\t{Role: \"user\", Content: tc.input},\n\t\t\t\t},\n\t\t\t})\n\t\t\trequire.NoError(t, err)\n\t\t\trequire.NotNil(t, result)\n\n\t\t\tt.Logf(\"Action %s: Queued=%v\", tc.action, result.Queued)\n\n\t\t\t// Wait for execution (shorter timeout for action tests)\n\t\t\tmaxWait := 90 * time.Second\n\t\t\tdeadline := time.Now().Add(maxWait)\n\n\t\t\tfor time.Now().Before(deadline) {\n\t\t\t\ttime.Sleep(2 * time.Second)\n\n\t\t\t\texecutions, err := api.ListExecutions(ctx, memberID, nil)\n\t\t\t\tif err != nil || len(executions.Data) == 0 {\n\t\t\t\t\tcontinue\n\t\t\t\t}\n\n\t\t\t\texec := executions.Data[0]\n\t\t\t\tif exec.Status == types.ExecCompleted || exec.Status == types.ExecFailed {\n\t\t\t\t\tif exec.Status == types.ExecFailed {\n\t\t\t\t\t\tt.Logf(\"Action %s failed: %s\", tc.action, exec.Error)\n\t\t\t\t\t} else {\n\t\t\t\t\t\tt.Logf(\"Action %s completed successfully\", tc.action)\n\t\t\t\t\t}\n\t\t\t\t\treturn\n\t\t\t\t}\n\t\t\t}\n\n\t\t\tt.Logf(\"Action %s: execution did not complete in time (may still be running)\", tc.action)\n\t\t})\n\t}\n\n\t// Stop after all action tests\n\tapi.Stop()\n}\n\n// ==================== Helper Functions ====================\n\n// setupE2ERobotForHuman creates a robot configured for human intervention E2E tests\nfunc setupE2ERobotForHuman(t *testing.T, memberID, teamID string) {\n\tm := model.Select(\"__yao.member\")\n\ttableName := m.MetaData.Table.Name\n\tqb := capsule.Query()\n\n\t// Simple config for E2E testing - minimal tasks\n\trobotConfig := map[string]interface{}{\n\t\t\"identity\": map[string]interface{}{\n\t\t\t\"role\": \"Simple E2E Test Robot\",\n\t\t\t\"duties\": []string{\"Echo user input\"}, // Very simple duty\n\t\t\t\"rules\": []string{\"Keep responses under 50 words\"},\n\t\t},\n\t\t\"quota\": map[string]interface{}{\n\t\t\t\"max\": 5,\n\t\t\t\"queue\": 20,\n\t\t\t\"priority\": 5,\n\t\t},\n\t\t\"triggers\": map[string]interface{}{\n\t\t\t\"clock\": map[string]interface{}{\"enabled\": false},\n\t\t\t\"intervene\": map[string]interface{}{\"enabled\": true},\n\t\t\t\"event\": map[string]interface{}{\"enabled\": true},\n\t\t},\n\t\t\"resources\": map[string]interface{}{\n\t\t\t\"phases\": map[string]interface{}{\n\t\t\t\t\"inspiration\": \"robot.inspiration\",\n\t\t\t\t\"goals\": \"robot.goals\",\n\t\t\t\t\"tasks\": \"tests.e2e-tasks\", // Use simple E2E test task planner\n\t\t\t\t\"run\": \"robot.validation\",\n\t\t\t\t\"validation\": \"tests.e2e-validation\", // Use lenient E2E test validator\n\t\t\t\t\"delivery\": \"robot.delivery\",\n\t\t\t\t\"learning\": \"robot.learning\",\n\t\t\t},\n\t\t\t\"agents\": []string{\"experts.text-writer\"},\n\t\t},\n\t\t\"delivery\": map[string]interface{}{\n\t\t\t\"email\": map[string]interface{}{\"enabled\": false},\n\t\t\t\"webhook\": map[string]interface{}{\"enabled\": false},\n\t\t\t\"process\": map[string]interface{}{\"enabled\": false},\n\t\t},\n\t}\n\tconfigJSON, _ := json.Marshal(robotConfig)\n\n\tsystemPrompt := `You are a simple E2E test robot. Your job is to echo user requests.\nWhen generating goals: create exactly 1 simple goal.\nWhen generating tasks: create exactly 1 simple task.\nKeep all outputs brief. No complex analysis needed.`\n\n\terr := qb.Table(tableName).Insert([]map[string]interface{}{\n\t\t{\n\t\t\t\"member_id\": memberID,\n\t\t\t\"team_id\": teamID,\n\t\t\t\"member_type\": \"robot\",\n\t\t\t\"display_name\": \"E2E Human Test Robot \" + memberID,\n\t\t\t\"system_prompt\": systemPrompt,\n\t\t\t\"status\": \"active\",\n\t\t\t\"role_id\": \"member\",\n\t\t\t\"autonomous_mode\": true,\n\t\t\t\"robot_status\": \"idle\",\n\t\t\t\"robot_config\": string(configJSON),\n\t\t},\n\t})\n\tif err != nil {\n\t\tt.Fatalf(\"Failed to insert robot %s: %v\", memberID, err)\n\t}\n}\n" }, { "path": "agent/robot/api/e2e_interact_test.go", "content": "package api_test\n\nimport (\n\t\"context\"\n\t\"encoding/json\"\n\t\"strings\"\n\t\"sync\"\n\t\"testing\"\n\t\"time\"\n\n\t\"github.com/stretchr/testify/assert\"\n\t\"github.com/stretchr/testify/require\"\n\t\"github.com/yaoapp/gou/model\"\n\t\"github.com/yaoapp/xun/capsule\"\n\t\"github.com/yaoapp/yao/agent/robot/api\"\n\t\"github.com/yaoapp/yao/agent/robot/executor/standard\"\n\t\"github.com/yaoapp/yao/agent/robot/types\"\n\t\"github.com/yaoapp/yao/agent/testutils\"\n)\n\n// TestE2EInteractNewAssignment tests the full Interact flow for a new task assignment.\n// With the conversational Host Agent, the first turn may return natural language\n// (waiting_for_more) or an action decision depending on request clarity.\nfunc TestE2EInteractNewAssignment(t *testing.T) {\n\tif testing.Short() {\n\t\tt.Skip(\"Skipping E2E test - requires real LLM calls\")\n\t}\n\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\tcleanupInteractRobots(t)\n\tcleanupInteractExecutions(t)\n\tdefer cleanupInteractRobots(t)\n\tdefer cleanupInteractExecutions(t)\n\n\tt.Run(\"assign_via_interact_creates_execution_and_gets_host_reply\", func(t *testing.T) {\n\t\tmemberID := \"robot_e2e_interact_assign\"\n\t\tsetupInteractRobot(t, memberID, \"team_e2e_interact\")\n\n\t\terr := api.Start()\n\t\trequire.NoError(t, err)\n\t\tdefer api.Stop()\n\n\t\tctx := types.NewContext(context.Background(), testAuth())\n\n\t\trobot, err := api.GetRobot(ctx, memberID)\n\t\trequire.NoError(t, err)\n\t\trequire.NotNil(t, robot)\n\n\t\tresult, err := api.Interact(ctx, memberID, &api.InteractRequest{\n\t\t\tSource: types.InteractSourceUI,\n\t\t\tMessage: \"Please write a short greeting email for our team meeting tomorrow morning.\",\n\t\t})\n\t\trequire.NoError(t, err)\n\t\trequire.NotNil(t, result)\n\n\t\tt.Logf(\"Interact result: status=%s, message=%s, reply=%s, exec_id=%s, wait_for_more=%v\",\n\t\t\tresult.Status, result.Message, result.Reply, result.ExecutionID, result.WaitForMore)\n\n\t\tassert.NotEmpty(t, result.ExecutionID, \"should create an execution\")\n\t\tassert.NotEmpty(t, result.ChatID, \"should have a chat session\")\n\t\tassert.NotEmpty(t, result.Reply, \"Host Agent should provide a reply\")\n\n\t\tvalidStatuses := []string{\"confirmed\", \"waiting_for_more\", \"adjusted\", \"acknowledged\"}\n\t\tassert.Contains(t, validStatuses, result.Status,\n\t\t\t\"status should be one of the valid Host Agent action outcomes\")\n\n\t\tif result.Status == \"confirmed\" {\n\t\t\ttime.Sleep(2 * time.Second)\n\t\t\texecutions, err := api.ListExecutions(ctx, memberID, &api.ExecutionQuery{Page: 1, PageSize: 5})\n\t\t\trequire.NoError(t, err)\n\t\t\tassert.Greater(t, len(executions.Data), 0, \"confirmed execution should exist in store\")\n\t\t}\n\t})\n}\n\n// TestE2EInteractStream tests the streaming version end-to-end.\nfunc TestE2EInteractStream(t *testing.T) {\n\tif testing.Short() {\n\t\tt.Skip(\"Skipping E2E test - requires real LLM calls\")\n\t}\n\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\tcleanupInteractRobots(t)\n\tcleanupInteractExecutions(t)\n\tdefer cleanupInteractRobots(t)\n\tdefer cleanupInteractExecutions(t)\n\n\tt.Run(\"stream_assign_returns_chunks_and_valid_result\", func(t *testing.T) {\n\t\tmemberID := \"robot_e2e_interact_stream\"\n\t\tsetupInteractRobot(t, memberID, \"team_e2e_interact\")\n\n\t\terr := api.Start()\n\t\trequire.NoError(t, err)\n\t\tdefer api.Stop()\n\n\t\tctx := types.NewContext(context.Background(), testAuth())\n\n\t\tvar mu sync.Mutex\n\t\tvar chunks []*standard.StreamChunk\n\n\t\tstreamFn := func(chunk *standard.StreamChunk) int {\n\t\t\tmu.Lock()\n\t\t\tdefer mu.Unlock()\n\t\t\tchunks = append(chunks, chunk)\n\t\t\treturn 0\n\t\t}\n\n\t\tresult, err := api.InteractStream(ctx, memberID, &api.InteractRequest{\n\t\t\tSource: types.InteractSourceUI,\n\t\t\tMessage: \"Help me draft a brief status update email about completing the Q4 report.\",\n\t\t}, streamFn)\n\t\trequire.NoError(t, err)\n\t\trequire.NotNil(t, result)\n\n\t\tmu.Lock()\n\t\tchunkCount := len(chunks)\n\t\tvar textChunks []string\n\t\tfor _, c := range chunks {\n\t\t\tif c.Type == \"text\" && c.Delta {\n\t\t\t\ttextChunks = append(textChunks, c.Content)\n\t\t\t}\n\t\t}\n\t\tmu.Unlock()\n\n\t\tcombined := strings.Join(textChunks, \"\")\n\n\t\tt.Logf(\"Stream received %d total chunks, %d text chunks, combined length: %d\",\n\t\t\tchunkCount, len(textChunks), len(combined))\n\t\tt.Logf(\"Result: status=%s, exec_id=%s, reply_len=%d, wait_for_more=%v\",\n\t\t\tresult.Status, result.ExecutionID, len(result.Reply), result.WaitForMore)\n\n\t\tassert.Greater(t, len(textChunks), 0, \"should receive streaming text chunks from Host Agent\")\n\t\tassert.NotEmpty(t, combined, \"combined text should not be empty\")\n\t\tassert.NotEmpty(t, result.ExecutionID, \"should create an execution\")\n\t\tassert.NotEmpty(t, result.Reply, \"final result should contain reply\")\n\n\t\tvalidStatuses := []string{\"confirmed\", \"waiting_for_more\", \"adjusted\"}\n\t\tassert.Contains(t, validStatuses, result.Status)\n\t})\n}\n\n// TestE2EInteractMultiTurn tests a multi-turn conversation:\n// Turn 1: Send vague message -> Host Agent replies conversationally (waiting_for_more)\n// Turn 2: Send clear confirmation -> Host Agent returns action JSON (confirmed or other action)\nfunc TestE2EInteractMultiTurn(t *testing.T) {\n\tif testing.Short() {\n\t\tt.Skip(\"Skipping E2E test - requires real LLM calls\")\n\t}\n\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\tcleanupInteractRobots(t)\n\tcleanupInteractExecutions(t)\n\tdefer cleanupInteractRobots(t)\n\tdefer cleanupInteractExecutions(t)\n\n\tt.Run(\"multi_turn_assign_conversation\", func(t *testing.T) {\n\t\tmemberID := \"robot_e2e_interact_multiturn\"\n\t\tsetupInteractRobot(t, memberID, \"team_e2e_interact\")\n\n\t\terr := api.Start()\n\t\trequire.NoError(t, err)\n\t\tdefer api.Stop()\n\n\t\tctx := types.NewContext(context.Background(), testAuth())\n\n\t\t// Turn 1: Send vague message — expect conversational reply\n\t\tresult1, err := api.Interact(ctx, memberID, &api.InteractRequest{\n\t\t\tSource: types.InteractSourceUI,\n\t\t\tMessage: \"Do something with emails.\",\n\t\t})\n\t\trequire.NoError(t, err)\n\t\trequire.NotNil(t, result1)\n\n\t\tt.Logf(\"Turn 1: status=%s, reply=%s, exec_id=%s, wait_for_more=%v\",\n\t\t\tresult1.Status, result1.Reply, result1.ExecutionID, result1.WaitForMore)\n\n\t\tassert.NotEmpty(t, result1.ExecutionID)\n\t\tassert.NotEmpty(t, result1.Reply)\n\n\t\t// Turn 2: Clarify/confirm with the same execution_id\n\t\tresult2, err := api.Interact(ctx, memberID, &api.InteractRequest{\n\t\t\tExecutionID: result1.ExecutionID,\n\t\t\tSource: types.InteractSourceUI,\n\t\t\tMessage: \"Yes, please write a brief thank-you email to the design team for their Q4 work. Go ahead and confirm.\",\n\t\t})\n\t\trequire.NoError(t, err)\n\t\trequire.NotNil(t, result2)\n\n\t\tt.Logf(\"Turn 2: status=%s, reply=%s, exec_id=%s, wait_for_more=%v\",\n\t\t\tresult2.Status, result2.Reply, result2.ExecutionID, result2.WaitForMore)\n\n\t\tassert.NotEmpty(t, result2.Reply)\n\t\tassert.Equal(t, result1.ExecutionID, result2.ExecutionID, \"should be same execution\")\n\n\t\tvalidStatuses := []string{\"confirmed\", \"waiting_for_more\", \"adjusted\", \"acknowledged\"}\n\t\tassert.Contains(t, validStatuses, result2.Status,\n\t\t\t\"second turn should produce a valid outcome\")\n\t})\n}\n\n// TestE2EInteractStreamMultiTurn tests multi-turn with streaming.\nfunc TestE2EInteractStreamMultiTurn(t *testing.T) {\n\tif testing.Short() {\n\t\tt.Skip(\"Skipping E2E test - requires real LLM calls\")\n\t}\n\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\tcleanupInteractRobots(t)\n\tcleanupInteractExecutions(t)\n\tdefer cleanupInteractRobots(t)\n\tdefer cleanupInteractExecutions(t)\n\n\tt.Run(\"stream_multi_turn\", func(t *testing.T) {\n\t\tmemberID := \"robot_e2e_interact_stream_mt\"\n\t\tsetupInteractRobot(t, memberID, \"team_e2e_interact\")\n\n\t\terr := api.Start()\n\t\trequire.NoError(t, err)\n\t\tdefer api.Stop()\n\n\t\tctx := types.NewContext(context.Background(), testAuth())\n\n\t\t// Turn 1\n\t\tvar mu1 sync.Mutex\n\t\tvar chunks1 []*standard.StreamChunk\n\t\tresult1, err := api.InteractStream(ctx, memberID, &api.InteractRequest{\n\t\t\tSource: types.InteractSourceUI,\n\t\t\tMessage: \"I need help with something.\",\n\t\t}, func(chunk *standard.StreamChunk) int {\n\t\t\tmu1.Lock()\n\t\t\tchunks1 = append(chunks1, chunk)\n\t\t\tmu1.Unlock()\n\t\t\treturn 0\n\t\t})\n\t\trequire.NoError(t, err)\n\t\trequire.NotNil(t, result1)\n\n\t\tmu1.Lock()\n\t\tt.Logf(\"Turn 1 stream: %d chunks, status=%s, reply=%s, wait_for_more=%v\",\n\t\t\tlen(chunks1), result1.Status, result1.Reply, result1.WaitForMore)\n\t\tmu1.Unlock()\n\n\t\tassert.NotEmpty(t, result1.ExecutionID)\n\t\tassert.NotEmpty(t, result1.Reply)\n\n\t\t// Turn 2: Clarify with same execution_id\n\t\tvar mu2 sync.Mutex\n\t\tvar chunks2 []*standard.StreamChunk\n\t\tresult2, err := api.InteractStream(ctx, memberID, &api.InteractRequest{\n\t\t\tExecutionID: result1.ExecutionID,\n\t\t\tSource: types.InteractSourceUI,\n\t\t\tMessage: \"Please compose a short farewell message for a colleague leaving the team. Yes, go ahead.\",\n\t\t}, func(chunk *standard.StreamChunk) int {\n\t\t\tmu2.Lock()\n\t\t\tchunks2 = append(chunks2, chunk)\n\t\t\tmu2.Unlock()\n\t\t\treturn 0\n\t\t})\n\t\trequire.NoError(t, err)\n\t\trequire.NotNil(t, result2)\n\n\t\tmu2.Lock()\n\t\tt.Logf(\"Turn 2 stream: %d chunks, status=%s, reply=%s, wait_for_more=%v\",\n\t\t\tlen(chunks2), result2.Status, result2.Reply, result2.WaitForMore)\n\t\tmu2.Unlock()\n\n\t\tassert.NotEmpty(t, result2.Reply)\n\t\tassert.Equal(t, result1.ExecutionID, result2.ExecutionID)\n\t})\n}\n\n// ==================== Helper Functions ====================\n\nfunc setupInteractRobot(t *testing.T, memberID, teamID string) {\n\tm := model.Select(\"__yao.member\")\n\ttableName := m.MetaData.Table.Name\n\tqb := capsule.Query()\n\n\trobotConfig := map[string]interface{}{\n\t\t\"identity\": map[string]interface{}{\n\t\t\t\"role\": \"Email Assistant\",\n\t\t\t\"duties\": []string{\"Write and manage emails\"},\n\t\t\t\"rules\": []string{\"Always confirm before sending\", \"Keep emails professional\"},\n\t\t},\n\t\t\"quota\": map[string]interface{}{\n\t\t\t\"max\": 5,\n\t\t\t\"queue\": 20,\n\t\t\t\"priority\": 5,\n\t\t},\n\t\t\"triggers\": map[string]interface{}{\n\t\t\t\"intervene\": map[string]interface{}{\"enabled\": true},\n\t\t},\n\t\t\"resources\": map[string]interface{}{\n\t\t\t\"phases\": map[string]interface{}{\n\t\t\t\t\"inspiration\": \"robot.inspiration\",\n\t\t\t\t\"goals\": \"robot.goals\",\n\t\t\t\t\"tasks\": \"robot.tasks\",\n\t\t\t\t\"run\": \"robot.validation\",\n\t\t\t\t\"validation\": \"robot.validation\",\n\t\t\t\t\"delivery\": \"robot.delivery\",\n\t\t\t\t\"learning\": \"robot.learning\",\n\t\t\t\t\"host\": \"robot.host\",\n\t\t\t},\n\t\t\t\"agents\": []string{},\n\t\t},\n\t}\n\tconfigJSON, _ := json.Marshal(robotConfig)\n\n\tsystemPrompt := `You are an email assistant for E2E testing of the Interact API.\nWhen asked to write an email, confirm the task and generate a brief email draft.`\n\n\terr := qb.Table(tableName).Insert([]map[string]interface{}{\n\t\t{\n\t\t\t\"member_id\": memberID,\n\t\t\t\"team_id\": teamID,\n\t\t\t\"member_type\": \"robot\",\n\t\t\t\"display_name\": \"E2E Interact Test Robot \" + memberID,\n\t\t\t\"system_prompt\": systemPrompt,\n\t\t\t\"status\": \"active\",\n\t\t\t\"role_id\": \"member\",\n\t\t\t\"autonomous_mode\": false,\n\t\t\t\"robot_status\": \"idle\",\n\t\t\t\"robot_config\": string(configJSON),\n\t\t},\n\t})\n\tif err != nil {\n\t\tt.Fatalf(\"Failed to insert interact robot %s: %v\", memberID, err)\n\t}\n}\n\nfunc cleanupInteractRobots(t *testing.T) {\n\tm := model.Select(\"__yao.member\")\n\tif m == nil {\n\t\treturn\n\t}\n\tqb := capsule.Query()\n\t_, err := qb.Table(m.MetaData.Table.Name).Where(\"member_id\", \"like\", \"robot_e2e_interact%\").Delete()\n\tif err != nil {\n\t\tt.Logf(\"Warning: cleanup interact robots: %v\", err)\n\t}\n}\n\nfunc cleanupInteractExecutions(t *testing.T) {\n\tm := model.Select(\"__yao.agent.execution\")\n\tif m == nil {\n\t\treturn\n\t}\n\tqb := capsule.Query()\n\t_, err := qb.Table(m.MetaData.Table.Name).Where(\"member_id\", \"like\", \"robot_e2e_interact%\").Delete()\n\tif err != nil {\n\t\tt.Logf(\"Warning: cleanup interact executions: %v\", err)\n\t}\n}\n" }, { "path": "agent/robot/api/e2e_suspend_test.go", "content": "package api_test\n\n// End-to-end tests for V2 Suspend/Resume flow\n// Tests the complete lifecycle: execution → need_input → suspend → reply → resume → complete/re-suspend\n//\n// Prerequisites:\n// - Valid LLM API keys\n// - Test assistants: tests.robot-need-input, experts.text-writer\n// - Database connection (YAO_DB_PRIMARY)\n\nimport (\n\t\"context\"\n\t\"encoding/json\"\n\t\"testing\"\n\t\"time\"\n\n\t\"github.com/stretchr/testify/assert\"\n\t\"github.com/stretchr/testify/require\"\n\t\"github.com/yaoapp/gou/model\"\n\t\"github.com/yaoapp/xun/capsule\"\n\tagentcontext \"github.com/yaoapp/yao/agent/context\"\n\t\"github.com/yaoapp/yao/agent/robot/api\"\n\t\"github.com/yaoapp/yao/agent/robot/store\"\n\t\"github.com/yaoapp/yao/agent/robot/types\"\n\t\"github.com/yaoapp/yao/agent/testutils\"\n\toauthtypes \"github.com/yaoapp/yao/openapi/oauth/types\"\n)\n\nfunc testAuthSuspend() *oauthtypes.AuthorizedInfo {\n\treturn &oauthtypes.AuthorizedInfo{\n\t\tUserID: \"e2e-suspend-user\",\n\t\tTeamID: \"e2e-suspend-team\",\n\t}\n}\n\n// triggerSuspendRobot triggers a robot via the Trigger API (human trigger path)\nfunc triggerSuspendRobot(t *testing.T, ctx *types.Context, memberID string, message string) *api.TriggerResult {\n\tt.Helper()\n\tresult, err := api.Trigger(ctx, memberID, &api.TriggerRequest{\n\t\tType: types.TriggerHuman,\n\t\tAction: types.ActionTaskAdd,\n\t\tMessages: []agentcontext.Message{\n\t\t\t{Role: \"user\", Content: message},\n\t\t},\n\t})\n\trequire.NoError(t, err)\n\trequire.NotNil(t, result)\n\tif !result.Accepted {\n\t\tt.Fatalf(\"Trigger not accepted: %s\", result.Message)\n\t}\n\treturn result\n}\n\n// waitForStatus polls execution status until it matches one of the expected statuses\nfunc waitForStatus(t *testing.T, execID string, statuses []types.ExecStatus, timeout time.Duration) *types.Execution {\n\tt.Helper()\n\tdeadline := time.Now().Add(timeout)\n\tfor time.Now().Before(deadline) {\n\t\ttime.Sleep(time.Second)\n\t\texec := getExecution(t, execID)\n\t\tif exec == nil {\n\t\t\tcontinue\n\t\t}\n\t\tfor _, s := range statuses {\n\t\t\tif exec.Status == s {\n\t\t\t\treturn exec\n\t\t\t}\n\t\t}\n\t}\n\treturn nil\n}\n\n// TestE2ENormalExecutionNoSuspend verifies that a normal execution (no need_input)\n// completes without entering the suspend path.\nfunc TestE2ENormalExecutionNoSuspend(t *testing.T) {\n\tif testing.Short() {\n\t\tt.Skip(\"Skipping E2E test - requires real LLM calls\")\n\t}\n\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\tcleanupE2ESuspendRobots(t)\n\tdefer cleanupE2ESuspendRobots(t)\n\n\tmemberID := \"robot_e2e_suspend_001\"\n\tsetupE2ESuspendRobotWithTasksPlanner(t, memberID, \"team_e2e_suspend\", []string{\"experts.text-writer\"}, \"tests.e2e-tasks\")\n\n\terr := api.Start()\n\trequire.NoError(t, err)\n\tdefer api.Stop()\n\n\tctx := types.NewContext(context.Background(), testAuthSuspend())\n\tresult := triggerSuspendRobot(t, ctx, memberID, \"Write a one-sentence greeting\")\n\n\texec := waitForStatus(t, result.ExecutionID,\n\t\t[]types.ExecStatus{types.ExecCompleted, types.ExecFailed}, 60*time.Second)\n\n\trequire.NotNil(t, exec, \"Execution should exist and reach terminal state\")\n\tif exec.Status == types.ExecFailed {\n\t\tt.Logf(\"Execution failed with error: %s\", exec.Error)\n\t}\n\tassert.Equal(t, types.ExecCompleted, exec.Status, \"Normal execution should complete\")\n\tassert.NotEmpty(t, exec.ChatID, \"ChatID should be set\")\n\tassert.Nil(t, exec.ResumeContext, \"No resume context for normal execution\")\n\tassert.Empty(t, exec.WaitingTaskID, \"No waiting task for normal execution\")\n}\n\n// TestE2ESuspendResumeFlow tests the full suspend-resume lifecycle:\n// 1. Trigger execution with robot-need-input assistant (signals need_input)\n// 2. Verify execution enters waiting status\n// 3. Reply to resume execution via api.Interact\n// 4. Verify execution re-suspends (since robot-need-input always signals need_input)\nfunc TestE2ESuspendResumeFlow(t *testing.T) {\n\tif testing.Short() {\n\t\tt.Skip(\"Skipping E2E test - requires real LLM calls\")\n\t}\n\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\tcleanupE2ESuspendRobots(t)\n\tdefer cleanupE2ESuspendRobots(t)\n\n\tmemberID := \"robot_e2e_suspend_002\"\n\tsetupE2ESuspendRobot(t, memberID, \"team_e2e_suspend\", []string{\"tests.robot-need-input\"})\n\n\terr := api.Start()\n\trequire.NoError(t, err)\n\tdefer api.Stop()\n\n\tctx := types.NewContext(context.Background(), testAuthSuspend())\n\n\t// Step 1: Trigger execution — the robot-need-input assistant always returns need_input\n\tresult := triggerSuspendRobot(t, ctx, memberID, \"Analyze sales data\")\n\texecID := result.ExecutionID\n\n\t// Step 2: Wait for the execution to reach waiting status\n\texec := waitForStatus(t, execID,\n\t\t[]types.ExecStatus{types.ExecWaiting, types.ExecCompleted, types.ExecFailed}, 60*time.Second)\n\n\trequire.NotNil(t, exec, \"Execution should exist\")\n\trequire.Equal(t, types.ExecWaiting, exec.Status, \"Execution should be in waiting status\")\n\tassert.NotEmpty(t, exec.WaitingTaskID, \"WaitingTaskID should be set\")\n\tassert.NotEmpty(t, exec.WaitingQuestion, \"WaitingQuestion should be set\")\n\tassert.NotNil(t, exec.WaitingSince, \"WaitingSince should be set\")\n\tassert.NotNil(t, exec.ResumeContext, \"ResumeContext should be set\")\n\tassert.NotEmpty(t, exec.ChatID, \"ChatID should be set\")\n\n\tt.Logf(\"Execution suspended: execID=%s task=%s question=%s\", execID, exec.WaitingTaskID, exec.WaitingQuestion)\n\n\t// Step 3: Resume via api.Interact (reply to the waiting execution)\n\tinteractResult, err := api.Interact(ctx, memberID, &api.InteractRequest{\n\t\tExecutionID: execID,\n\t\tMessage: \"Use the last 30 days for analysis\",\n\t})\n\trequire.NoError(t, err)\n\trequire.NotNil(t, interactResult)\n\n\t// Since robot-need-input always signals need_input, the resumed execution\n\t// will re-suspend. The Interact API returns \"waiting\" status in this case.\n\tassert.Equal(t, \"waiting\", interactResult.Status, \"Should re-suspend since assistant always signals need_input\")\n\tt.Logf(\"Interact result: status=%s message=%s\", interactResult.Status, interactResult.Message)\n\n\t// Step 4: Verify the execution is in waiting status again (re-suspended)\n\texec = getExecution(t, execID)\n\trequire.NotNil(t, exec)\n\tassert.Equal(t, types.ExecWaiting, exec.Status, \"Execution should be waiting again after re-suspend\")\n\tassert.NotNil(t, exec.ResumeContext, \"ResumeContext should be set after re-suspend\")\n}\n\n// TestE2EReplyShortcut tests the Reply semantic shortcut\nfunc TestE2EReplyShortcut(t *testing.T) {\n\tif testing.Short() {\n\t\tt.Skip(\"Skipping E2E test - requires real LLM calls\")\n\t}\n\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\tcleanupE2ESuspendRobots(t)\n\tdefer cleanupE2ESuspendRobots(t)\n\n\tmemberID := \"robot_e2e_suspend_004\"\n\tsetupE2ESuspendRobot(t, memberID, \"team_e2e_suspend\", []string{\"tests.robot-need-input\"})\n\n\terr := api.Start()\n\trequire.NoError(t, err)\n\tdefer api.Stop()\n\n\tctx := types.NewContext(context.Background(), testAuthSuspend())\n\tresult := triggerSuspendRobot(t, ctx, memberID, \"Check inventory levels\")\n\n\texec := waitForStatus(t, result.ExecutionID,\n\t\t[]types.ExecStatus{types.ExecWaiting}, 60*time.Second)\n\trequire.NotNil(t, exec, \"Execution should reach waiting status\")\n\trequire.Equal(t, types.ExecWaiting, exec.Status)\n\n\t// Use Reply shortcut\n\treplyResult, err := api.Reply(ctx, memberID, result.ExecutionID, exec.WaitingTaskID, \"Use warehouse A data\")\n\trequire.NoError(t, err)\n\trequire.NotNil(t, replyResult)\n\tassert.Contains(t, []string{\"waiting\", \"resumed\"}, replyResult.Status)\n\tt.Logf(\"Reply result: status=%s\", replyResult.Status)\n}\n\n// TestE2EResumeContextPersistence verifies that suspend state is properly persisted\n// and can be loaded back from the database.\nfunc TestE2EResumeContextPersistence(t *testing.T) {\n\tif testing.Short() {\n\t\tt.Skip(\"Skipping E2E test - requires real LLM calls\")\n\t}\n\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\tcleanupE2ESuspendRobots(t)\n\tdefer cleanupE2ESuspendRobots(t)\n\n\tmemberID := \"robot_e2e_suspend_003\"\n\tsetupE2ESuspendRobot(t, memberID, \"team_e2e_suspend\", []string{\"tests.robot-need-input\"})\n\n\terr := api.Start()\n\trequire.NoError(t, err)\n\tdefer api.Stop()\n\n\tctx := types.NewContext(context.Background(), testAuthSuspend())\n\tresult := triggerSuspendRobot(t, ctx, memberID, \"Analyze user behavior\")\n\n\texec := waitForStatus(t, result.ExecutionID,\n\t\t[]types.ExecStatus{types.ExecWaiting, types.ExecCompleted, types.ExecFailed}, 60*time.Second)\n\n\trequire.NotNil(t, exec)\n\tif exec.Status != types.ExecWaiting {\n\t\tt.Skipf(\"Execution did not reach waiting status (status=%s), skipping persistence test\", exec.Status)\n\t}\n\n\t// Load from DB directly using store to verify persistence\n\texecStore := store.NewExecutionStore()\n\trecord, err := execStore.Get(context.Background(), result.ExecutionID)\n\trequire.NoError(t, err)\n\trequire.NotNil(t, record)\n\n\tassert.Equal(t, types.ExecWaiting, record.Status)\n\tassert.NotEmpty(t, record.WaitingTaskID)\n\tassert.NotEmpty(t, record.WaitingQuestion)\n\tassert.NotNil(t, record.WaitingSince)\n\tassert.NotNil(t, record.ResumeContext)\n\tassert.Equal(t, exec.ChatID, record.ChatID)\n\n\t// Verify resume context deserialization\n\trestored := record.ToExecution()\n\tassert.NotNil(t, restored.ResumeContext)\n\tassert.GreaterOrEqual(t, restored.ResumeContext.TaskIndex, 0)\n\n\tt.Logf(\"Persisted resume context: TaskIndex=%d, PreviousResults=%d\",\n\t\trestored.ResumeContext.TaskIndex, len(restored.ResumeContext.PreviousResults))\n}\n\n// TestE2EInteractRequiresExecutionID tests that Interact API returns error when\n// execution_id is not provided (Host Agent deferred).\nfunc TestE2EInteractRequiresExecutionID(t *testing.T) {\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\tctx := types.NewContext(context.Background(), testAuthSuspend())\n\n\t_, err := api.Interact(ctx, \"some-member\", &api.InteractRequest{\n\t\tMessage: \"hello\",\n\t})\n\tassert.Error(t, err)\n\tassert.Contains(t, err.Error(), \"execution_id is required\")\n}\n\n// TestE2EInteractWithNonWaitingExecution tests that Interact API returns error\n// when trying to resume an execution that is not in waiting status.\nfunc TestE2EInteractWithNonWaitingExecution(t *testing.T) {\n\tif testing.Short() {\n\t\tt.Skip(\"Skipping E2E test - requires real LLM calls\")\n\t}\n\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\tcleanupE2ESuspendRobots(t)\n\tdefer cleanupE2ESuspendRobots(t)\n\n\tmemberID := \"robot_e2e_suspend_005\"\n\tsetupE2ESuspendRobotWithTasksPlanner(t, memberID, \"team_e2e_suspend\", []string{\"experts.text-writer\"}, \"tests.e2e-tasks\")\n\n\terr := api.Start()\n\trequire.NoError(t, err)\n\tdefer api.Stop()\n\n\tctx := types.NewContext(context.Background(), testAuthSuspend())\n\tresult := triggerSuspendRobot(t, ctx, memberID, \"Say hello\")\n\n\t// Wait for completion\n\texec := waitForStatus(t, result.ExecutionID,\n\t\t[]types.ExecStatus{types.ExecCompleted, types.ExecFailed}, 60*time.Second)\n\trequire.NotNil(t, exec, \"Execution should reach terminal state\")\n\n\t// Try to interact with the completed execution\n\t_, err = api.Interact(ctx, memberID, &api.InteractRequest{\n\t\tExecutionID: result.ExecutionID,\n\t\tMessage: \"This should fail\",\n\t})\n\tassert.Error(t, err)\n\tassert.Contains(t, err.Error(), \"cannot interact\")\n}\n\n// ============================================================================\n// Helper Functions\n// ============================================================================\n\nfunc setupE2ESuspendRobotWithTasksPlanner(t *testing.T, memberID, teamID string, agents []string, tasksPlanner string) {\n\tt.Helper()\n\n\trobotConfig := map[string]interface{}{\n\t\t\"identity\": map[string]interface{}{\n\t\t\t\"role\": \"V2 Suspend Test Robot\",\n\t\t\t\"duties\": []string{\"Execute test tasks\"},\n\t\t\t\"rules\": []string{\"Keep responses under 50 words\"},\n\t\t},\n\t\t\"resources\": map[string]interface{}{\n\t\t\t\"phases\": map[string]interface{}{\n\t\t\t\t\"inspiration\": \"robot.inspiration\",\n\t\t\t\t\"goals\": \"robot.goals\",\n\t\t\t\t\"tasks\": tasksPlanner,\n\t\t\t\t\"run\": \"robot.validation\",\n\t\t\t\t\"delivery\": \"robot.delivery\",\n\t\t\t\t\"learning\": \"robot.learning\",\n\t\t\t},\n\t\t\t\"agents\": agents,\n\t\t},\n\t\t\"quota\": map[string]interface{}{\n\t\t\t\"max\": 5,\n\t\t\t\"queue\": 20,\n\t\t\t\"priority\": 5,\n\t\t},\n\t\t\"triggers\": map[string]interface{}{\n\t\t\t\"clock\": map[string]interface{}{\"enabled\": false},\n\t\t\t\"intervene\": map[string]interface{}{\"enabled\": true},\n\t\t\t\"event\": map[string]interface{}{\"enabled\": true},\n\t\t},\n\t\t\"delivery\": map[string]interface{}{\n\t\t\t\"email\": map[string]interface{}{\"enabled\": false},\n\t\t\t\"webhook\": map[string]interface{}{\"enabled\": false},\n\t\t\t\"process\": map[string]interface{}{\"enabled\": false},\n\t\t},\n\t}\n\n\tconfigJSON, err := json.Marshal(robotConfig)\n\trequire.NoError(t, err)\n\n\tm := model.Select(\"__yao.member\")\n\trequire.NotNil(t, m)\n\ttableName := m.MetaData.Table.Name\n\tqb := capsule.Query()\n\n\terr = qb.Table(tableName).Insert([]map[string]interface{}{\n\t\t{\n\t\t\t\"member_id\": memberID,\n\t\t\t\"team_id\": teamID,\n\t\t\t\"member_type\": \"robot\",\n\t\t\t\"display_name\": \"E2E Suspend Test Robot \" + memberID,\n\t\t\t\"system_prompt\": \"You are a simple E2E test robot. Your job is to execute tasks.\\nWhen generating goals: create exactly 1 simple goal.\\nWhen generating tasks: create exactly 1 simple task.\\nKeep all outputs brief.\",\n\t\t\t\"status\": \"active\",\n\t\t\t\"role_id\": \"member\",\n\t\t\t\"autonomous_mode\": true,\n\t\t\t\"robot_status\": \"idle\",\n\t\t\t\"robot_config\": string(configJSON),\n\t\t},\n\t})\n\trequire.NoError(t, err)\n}\n\nfunc setupE2ESuspendRobot(t *testing.T, memberID, teamID string, agents []string) {\n\tt.Helper()\n\tsetupE2ESuspendRobotWithTasksPlanner(t, memberID, teamID, agents, \"tests.e2e-suspend-tasks\")\n}\n\nfunc cleanupE2ESuspendRobots(t *testing.T) {\n\tt.Helper()\n\tmod := model.Select(\"__yao.member\")\n\tif mod == nil {\n\t\treturn\n\t}\n\tmod.DeleteWhere(model.QueryParam{\n\t\tWheres: []model.QueryWhere{\n\t\t\t{Column: \"member_id\", OP: \"like\", Value: \"robot_e2e_suspend_%\"},\n\t\t},\n\t})\n}\n\nfunc getExecution(t *testing.T, execID string) *types.Execution {\n\tt.Helper()\n\texecStore := store.NewExecutionStore()\n\trecord, err := execStore.Get(context.Background(), execID)\n\tif err != nil || record == nil {\n\t\treturn nil\n\t}\n\treturn record.ToExecution()\n}\n" }, { "path": "agent/robot/api/execution.go", "content": "package api\n\nimport (\n\t\"context\"\n\t\"fmt\"\n\t\"sync\"\n\n\t\"github.com/yaoapp/yao/agent/robot/store\"\n\t\"github.com/yaoapp/yao/agent/robot/types\"\n)\n\n// executionStore singleton\nvar (\n\texecStore *store.ExecutionStore\n\texecStoreOnce sync.Once\n)\n\n// getExecutionStore returns the singleton execution store\nfunc getExecutionStore() *store.ExecutionStore {\n\texecStoreOnce.Do(func() {\n\t\texecStore = store.NewExecutionStore()\n\t})\n\treturn execStore\n}\n\n// ResetExecutionStore resets the singleton for testing purposes\n// This should only be called in tests\nfunc ResetExecutionStore() {\n\texecStoreOnce = sync.Once{}\n\texecStore = nil\n}\n\n// ==================== Execution Query API ====================\n// These functions query and manage execution history\n\n// GetExecution returns a specific execution by ID\nfunc GetExecution(ctx *types.Context, execID string) (*types.Execution, error) {\n\tif execID == \"\" {\n\t\treturn nil, fmt.Errorf(\"execution_id is required\")\n\t}\n\n\t// Try to get from execution store\n\trecord, err := getExecutionStore().Get(context.Background(), execID)\n\tif err != nil {\n\t\treturn nil, fmt.Errorf(\"failed to get execution: %w\", err)\n\t}\n\tif record == nil {\n\t\treturn nil, fmt.Errorf(\"execution not found: %s\", execID)\n\t}\n\n\treturn record.ToExecution(), nil\n}\n\n// ListExecutions returns execution history for a robot\nfunc ListExecutions(ctx *types.Context, memberID string, query *ExecutionQuery) (*ExecutionResult, error) {\n\tif memberID == \"\" {\n\t\treturn nil, fmt.Errorf(\"member_id is required\")\n\t}\n\n\tif query == nil {\n\t\tquery = &ExecutionQuery{}\n\t}\n\tquery.applyDefaults()\n\n\topts := &store.ListOptions{\n\t\tMemberID: memberID,\n\t\tPage: query.Page,\n\t\tPageSize: query.PageSize,\n\t\tOrderBy: \"start_time desc\",\n\t}\n\n\tif query.Status != \"\" {\n\t\topts.Status = query.Status\n\t}\n\tif len(query.ExcludeStatuses) > 0 {\n\t\topts.ExcludeStatuses = query.ExcludeStatuses\n\t}\n\tif query.Trigger != \"\" {\n\t\topts.TriggerType = query.Trigger\n\t}\n\n\tresult, err := getExecutionStore().List(context.Background(), opts)\n\tif err != nil {\n\t\treturn nil, fmt.Errorf(\"failed to list executions: %w\", err)\n\t}\n\n\texecutions := make([]*types.Execution, 0, len(result.Data))\n\tfor _, record := range result.Data {\n\t\texecutions = append(executions, record.ToExecution())\n\t}\n\n\treturn &ExecutionResult{\n\t\tData: executions,\n\t\tTotal: result.Total,\n\t\tPage: result.Page,\n\t\tPageSize: result.PageSize,\n\t}, nil\n}\n\n// ==================== Execution Control API ====================\n// These functions control running executions\n\n// PauseExecution pauses a running execution\nfunc PauseExecution(ctx *types.Context, execID string) error {\n\tif execID == \"\" {\n\t\treturn fmt.Errorf(\"execution_id is required\")\n\t}\n\n\tmgr, err := getManager()\n\tif err != nil {\n\t\treturn err\n\t}\n\n\tif err := mgr.PauseExecution(ctx, execID); err != nil {\n\t\treturn err\n\t}\n\n\t// Update database status to paused\n\treturn getExecutionStore().UpdateStatus(context.Background(), execID, types.ExecPaused, \"\")\n}\n\n// ResumeExecution resumes a paused execution\nfunc ResumeExecution(ctx *types.Context, execID string) error {\n\tif execID == \"\" {\n\t\treturn fmt.Errorf(\"execution_id is required\")\n\t}\n\n\tmgr, err := getManager()\n\tif err != nil {\n\t\treturn err\n\t}\n\n\tif err := mgr.ResumeExecution(ctx, execID); err != nil {\n\t\treturn err\n\t}\n\n\t// Update database status back to running\n\treturn getExecutionStore().UpdateStatus(context.Background(), execID, types.ExecRunning, \"\")\n}\n\n// StopExecution stops a running execution\nfunc StopExecution(ctx *types.Context, execID string) error {\n\tif execID == \"\" {\n\t\treturn fmt.Errorf(\"execution_id is required\")\n\t}\n\n\tmgr, err := getManager()\n\tif err != nil {\n\t\treturn err\n\t}\n\n\tif err := mgr.StopExecution(ctx, execID); err != nil {\n\t\treturn err\n\t}\n\n\t// Update database status to cancelled\n\treturn getExecutionStore().UpdateStatus(context.Background(), execID, types.ExecCancelled, \"User cancelled\")\n}\n\n// ==================== Execution Status API ====================\n\n// GetExecutionStatus returns the current status of an execution\n// This combines stored data with runtime state\nfunc GetExecutionStatus(ctx *types.Context, execID string) (*types.Execution, error) {\n\tif execID == \"\" {\n\t\treturn nil, fmt.Errorf(\"execution_id is required\")\n\t}\n\n\t// Get from store first\n\texec, err := GetExecution(ctx, execID)\n\tif err != nil {\n\t\treturn nil, err\n\t}\n\n\t// If manager is running, check for runtime state\n\tmgr, mgrErr := getManager()\n\tif mgrErr == nil {\n\t\t// Check if execution is being tracked (running)\n\t\tctrlExec, ctrlErr := mgr.GetExecutionStatus(execID)\n\t\tif ctrlErr == nil && ctrlExec != nil {\n\t\t\t// Update with runtime state\n\t\t\texec.Status = ctrlExec.Status\n\t\t\texec.Phase = ctrlExec.Phase\n\t\t}\n\t}\n\n\treturn exec, nil\n}\n" }, { "path": "agent/robot/api/execution_test.go", "content": "package api_test\n\nimport (\n\t\"context\"\n\t\"testing\"\n\n\t\"github.com/stretchr/testify/assert\"\n\t\"github.com/stretchr/testify/require\"\n\t\"github.com/yaoapp/yao/agent/robot/api\"\n\t\"github.com/yaoapp/yao/agent/robot/types\"\n\t\"github.com/yaoapp/yao/agent/testutils\"\n)\n\n// TestGetExecutionValidation tests parameter validation for GetExecution\nfunc TestGetExecutionValidation(t *testing.T) {\n\tif testing.Short() {\n\t\tt.Skip(\"Skipping integration test\")\n\t}\n\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\tctx := types.NewContext(context.Background(), nil)\n\n\tt.Run(\"returns error for empty execution_id\", func(t *testing.T) {\n\t\texec, err := api.GetExecution(ctx, \"\")\n\t\tassert.Error(t, err)\n\t\tassert.Nil(t, exec)\n\t\tassert.Contains(t, err.Error(), \"execution_id is required\")\n\t})\n\n\tt.Run(\"returns error for non-existent execution\", func(t *testing.T) {\n\t\texec, err := api.GetExecution(ctx, \"non_existent_exec_id_xyz\")\n\t\tassert.Error(t, err)\n\t\tassert.Nil(t, exec)\n\t})\n}\n\n// TestListExecutionsValidation tests parameter validation for ListExecutions\nfunc TestListExecutionsValidation(t *testing.T) {\n\tif testing.Short() {\n\t\tt.Skip(\"Skipping integration test\")\n\t}\n\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\tctx := types.NewContext(context.Background(), nil)\n\n\tt.Run(\"returns error for empty member_id\", func(t *testing.T) {\n\t\tresult, err := api.ListExecutions(ctx, \"\", nil)\n\t\tassert.Error(t, err)\n\t\tassert.Nil(t, result)\n\t\tassert.Contains(t, err.Error(), \"member_id is required\")\n\t})\n\n\tt.Run(\"applies default pagination when query is nil\", func(t *testing.T) {\n\t\tresult, err := api.ListExecutions(ctx, \"test_member\", nil)\n\t\tassert.NoError(t, err)\n\t\tassert.NotNil(t, result)\n\t\tassert.Equal(t, 1, result.Page)\n\t\tassert.Equal(t, 20, result.PageSize)\n\t})\n\n\tt.Run(\"caps pagesize at 100\", func(t *testing.T) {\n\t\tresult, err := api.ListExecutions(ctx, \"test_member\", &api.ExecutionQuery{\n\t\t\tPage: 1,\n\t\t\tPageSize: 200,\n\t\t})\n\t\tassert.NoError(t, err)\n\t\tassert.NotNil(t, result)\n\t\tassert.Equal(t, 100, result.PageSize)\n\t})\n}\n\n// TestPauseExecutionValidation tests parameter validation for PauseExecution\nfunc TestPauseExecutionValidation(t *testing.T) {\n\tif testing.Short() {\n\t\tt.Skip(\"Skipping integration test\")\n\t}\n\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\tctx := types.NewContext(context.Background(), nil)\n\n\tt.Run(\"returns error for empty execution_id\", func(t *testing.T) {\n\t\terr := api.PauseExecution(ctx, \"\")\n\t\tassert.Error(t, err)\n\t\tassert.Contains(t, err.Error(), \"execution_id is required\")\n\t})\n\n\tt.Run(\"returns error when manager not started\", func(t *testing.T) {\n\t\terr := api.PauseExecution(ctx, \"test_exec_id\")\n\t\tassert.Error(t, err)\n\t\tassert.Contains(t, err.Error(), \"not started\")\n\t})\n}\n\n// TestResumeExecutionValidation tests parameter validation for ResumeExecution\nfunc TestResumeExecutionValidation(t *testing.T) {\n\tif testing.Short() {\n\t\tt.Skip(\"Skipping integration test\")\n\t}\n\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\tctx := types.NewContext(context.Background(), nil)\n\n\tt.Run(\"returns error for empty execution_id\", func(t *testing.T) {\n\t\terr := api.ResumeExecution(ctx, \"\")\n\t\tassert.Error(t, err)\n\t\tassert.Contains(t, err.Error(), \"execution_id is required\")\n\t})\n\n\tt.Run(\"returns error when manager not started\", func(t *testing.T) {\n\t\terr := api.ResumeExecution(ctx, \"test_exec_id\")\n\t\tassert.Error(t, err)\n\t\tassert.Contains(t, err.Error(), \"not started\")\n\t})\n}\n\n// TestStopExecutionValidation tests parameter validation for StopExecution\nfunc TestStopExecutionValidation(t *testing.T) {\n\tif testing.Short() {\n\t\tt.Skip(\"Skipping integration test\")\n\t}\n\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\tctx := types.NewContext(context.Background(), nil)\n\n\tt.Run(\"returns error for empty execution_id\", func(t *testing.T) {\n\t\terr := api.StopExecution(ctx, \"\")\n\t\tassert.Error(t, err)\n\t\tassert.Contains(t, err.Error(), \"execution_id is required\")\n\t})\n\n\tt.Run(\"returns error when manager not started\", func(t *testing.T) {\n\t\terr := api.StopExecution(ctx, \"test_exec_id\")\n\t\tassert.Error(t, err)\n\t\tassert.Contains(t, err.Error(), \"not started\")\n\t})\n}\n\n// TestGetExecutionStatusValidation tests parameter validation for GetExecutionStatus\nfunc TestGetExecutionStatusValidation(t *testing.T) {\n\tif testing.Short() {\n\t\tt.Skip(\"Skipping integration test\")\n\t}\n\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\tctx := types.NewContext(context.Background(), nil)\n\n\tt.Run(\"returns error for empty execution_id\", func(t *testing.T) {\n\t\texec, err := api.GetExecutionStatus(ctx, \"\")\n\t\tassert.Error(t, err)\n\t\tassert.Nil(t, exec)\n\t\tassert.Contains(t, err.Error(), \"execution_id is required\")\n\t})\n\n\tt.Run(\"returns error for non-existent execution\", func(t *testing.T) {\n\t\texec, err := api.GetExecutionStatus(ctx, \"non_existent_exec_id_xyz\")\n\t\tassert.Error(t, err)\n\t\tassert.Nil(t, exec)\n\t})\n}\n\n// TestExecutionControlWithManagerStarted tests execution control APIs when manager is running\nfunc TestExecutionControlWithManagerStarted(t *testing.T) {\n\tif testing.Short() {\n\t\tt.Skip(\"Skipping integration test\")\n\t}\n\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\t// Start manager\n\terr := api.Start()\n\trequire.NoError(t, err)\n\tdefer api.Stop()\n\n\tctx := types.NewContext(context.Background(), nil)\n\n\tt.Run(\"pause returns error for non-existent execution\", func(t *testing.T) {\n\t\terr := api.PauseExecution(ctx, \"non_existent_exec_id_xyz\")\n\t\tassert.Error(t, err)\n\t})\n\n\tt.Run(\"resume returns error for non-existent execution\", func(t *testing.T) {\n\t\terr := api.ResumeExecution(ctx, \"non_existent_exec_id_xyz\")\n\t\tassert.Error(t, err)\n\t})\n\n\tt.Run(\"stop returns error for non-existent execution\", func(t *testing.T) {\n\t\terr := api.StopExecution(ctx, \"non_existent_exec_id_xyz\")\n\t\tassert.Error(t, err)\n\t})\n}\n" }, { "path": "agent/robot/api/interact.go", "content": "package api\n\nimport (\n\t\"fmt\"\n\n\tagentcontext \"github.com/yaoapp/yao/agent/context\"\n\t\"github.com/yaoapp/yao/agent/robot/executor/standard\"\n\t\"github.com/yaoapp/yao/agent/robot/manager\"\n\t\"github.com/yaoapp/yao/agent/robot/types\"\n)\n\n// InteractRequest represents a unified interaction with a robot.\ntype InteractRequest struct {\n\tExecutionID string `json:\"execution_id,omitempty\"`\n\tTaskID string `json:\"task_id,omitempty\"`\n\tSource types.InteractSource `json:\"source,omitempty\"`\n\tMessage string `json:\"message\"`\n\tAction string `json:\"action,omitempty\"`\n}\n\n// InteractResult is the response from an interaction.\ntype InteractResult struct {\n\tExecutionID string `json:\"execution_id,omitempty\"`\n\tStatus string `json:\"status\"`\n\tMessage string `json:\"message,omitempty\"`\n\tChatID string `json:\"chat_id,omitempty\"`\n\tReply string `json:\"reply,omitempty\"`\n\tWaitForMore bool `json:\"wait_for_more,omitempty\"`\n}\n\n// Interact handles all human-robot interactions through a unified entry point.\n//\n// Routing logic:\n// - If manager is running, delegate to Manager.HandleInteract (full V2 flow with Host Agent)\n// - Otherwise, use legacy direct-executor path for backward compatibility\nfunc Interact(ctx *types.Context, memberID string, req *InteractRequest) (*InteractResult, error) {\n\tif memberID == \"\" {\n\t\treturn nil, fmt.Errorf(\"member_id is required\")\n\t}\n\tif req == nil {\n\t\treturn nil, fmt.Errorf(\"interact request is required\")\n\t}\n\n\t// Try V2 path via manager\n\tmgr, err := getManager()\n\tif err == nil && mgr != nil {\n\t\treturn managerInteract(ctx, mgr, memberID, req)\n\t}\n\n\t// V1 fallback: require execution_id for direct resume\n\tif req.ExecutionID == \"\" {\n\t\treturn nil, fmt.Errorf(\"execution_id is required for current version (Host Agent deferred)\")\n\t}\n\n\treturn legacyResume(ctx, req)\n}\n\n// managerInteract delegates to the manager's HandleInteract.\nfunc managerInteract(ctx *types.Context, mgr *manager.Manager, memberID string, req *InteractRequest) (*InteractResult, error) {\n\tmgrReq := &manager.InteractRequest{\n\t\tExecutionID: req.ExecutionID,\n\t\tTaskID: req.TaskID,\n\t\tSource: req.Source,\n\t\tMessage: req.Message,\n\t\tAction: req.Action,\n\t}\n\n\tresp, err := mgr.HandleInteract(ctx, memberID, mgrReq)\n\tif err != nil {\n\t\treturn nil, err\n\t}\n\n\treturn &InteractResult{\n\t\tExecutionID: resp.ExecutionID,\n\t\tStatus: resp.Status,\n\t\tMessage: resp.Message,\n\t\tChatID: resp.ChatID,\n\t\tReply: resp.Reply,\n\t\tWaitForMore: resp.WaitForMore,\n\t}, nil\n}\n\n// legacyResume handles the direct executor resume path (backward compatible).\nfunc legacyResume(ctx *types.Context, req *InteractRequest) (*InteractResult, error) {\n\texecutor := standard.New()\n\terr := executor.Resume(ctx, req.ExecutionID, req.Message)\n\tif err != nil {\n\t\tif err == types.ErrExecutionSuspended {\n\t\t\treturn &InteractResult{\n\t\t\t\tExecutionID: req.ExecutionID,\n\t\t\t\tStatus: \"waiting\",\n\t\t\t\tMessage: \"Execution suspended again: needs more input\",\n\t\t\t}, nil\n\t\t}\n\t\treturn nil, fmt.Errorf(\"failed to resume execution: %w\", err)\n\t}\n\n\treturn &InteractResult{\n\t\tExecutionID: req.ExecutionID,\n\t\tStatus: \"resumed\",\n\t\tMessage: \"Execution resumed and completed successfully\",\n\t}, nil\n}\n\n// Reply is a semantic shortcut for replying to a specific waiting task.\nfunc Reply(ctx *types.Context, memberID string, execID string, taskID string, message string) (*InteractResult, error) {\n\treturn Interact(ctx, memberID, &InteractRequest{\n\t\tExecutionID: execID,\n\t\tTaskID: taskID,\n\t\tSource: types.InteractSourceUI,\n\t\tMessage: message,\n\t})\n}\n\n// Confirm is a semantic shortcut for confirming a pending execution.\nfunc Confirm(ctx *types.Context, memberID string, execID string, message string) (*InteractResult, error) {\n\treturn Interact(ctx, memberID, &InteractRequest{\n\t\tExecutionID: execID,\n\t\tSource: types.InteractSourceUI,\n\t\tMessage: message,\n\t\tAction: \"confirm\",\n\t})\n}\n\n// InteractStream is the streaming version of Interact.\n// It streams Host Agent text tokens via streamFn while still returning the final InteractResult.\n// V1 fallback does not support streaming and returns an error.\nfunc InteractStream(ctx *types.Context, memberID string, req *InteractRequest, streamFn standard.StreamCallback) (*InteractResult, error) {\n\tif memberID == \"\" {\n\t\treturn nil, fmt.Errorf(\"member_id is required\")\n\t}\n\tif req == nil {\n\t\treturn nil, fmt.Errorf(\"interact request is required\")\n\t}\n\n\tmgr, err := getManager()\n\tif err != nil || mgr == nil {\n\t\treturn nil, fmt.Errorf(\"streaming requires V2 manager (not available)\")\n\t}\n\n\tmgrReq := &manager.InteractRequest{\n\t\tExecutionID: req.ExecutionID,\n\t\tTaskID: req.TaskID,\n\t\tSource: req.Source,\n\t\tMessage: req.Message,\n\t\tAction: req.Action,\n\t}\n\n\tresp, err := mgr.HandleInteractStream(ctx, memberID, mgrReq, streamFn)\n\tif err != nil {\n\t\treturn nil, err\n\t}\n\n\treturn &InteractResult{\n\t\tExecutionID: resp.ExecutionID,\n\t\tStatus: resp.Status,\n\t\tMessage: resp.Message,\n\t\tChatID: resp.ChatID,\n\t\tReply: resp.Reply,\n\t\tWaitForMore: resp.WaitForMore,\n\t}, nil\n}\n\n// InteractStreamRaw is the CUI-protocol-aligned streaming version of Interact.\n// It passes raw message.Message objects to the onMessage callback, preserving all CUI\n// protocol fields for direct SSE passthrough to the frontend.\nfunc InteractStreamRaw(ctx *types.Context, memberID string, req *InteractRequest, onMessage agentcontext.OnMessageFunc) (*InteractResult, error) {\n\tif memberID == \"\" {\n\t\treturn nil, fmt.Errorf(\"member_id is required\")\n\t}\n\tif req == nil {\n\t\treturn nil, fmt.Errorf(\"interact request is required\")\n\t}\n\n\tmgr, err := getManager()\n\tif err != nil || mgr == nil {\n\t\treturn nil, fmt.Errorf(\"raw streaming requires V2 manager (not available)\")\n\t}\n\n\tmgrReq := &manager.InteractRequest{\n\t\tExecutionID: req.ExecutionID,\n\t\tTaskID: req.TaskID,\n\t\tSource: req.Source,\n\t\tMessage: req.Message,\n\t\tAction: req.Action,\n\t}\n\n\tresp, err := mgr.HandleInteractStreamRaw(ctx, memberID, mgrReq, onMessage)\n\tif err != nil {\n\t\treturn nil, err\n\t}\n\n\treturn &InteractResult{\n\t\tExecutionID: resp.ExecutionID,\n\t\tStatus: resp.Status,\n\t\tMessage: resp.Message,\n\t\tChatID: resp.ChatID,\n\t\tReply: resp.Reply,\n\t\tWaitForMore: resp.WaitForMore,\n\t}, nil\n}\n\n// CancelExecution cancels a waiting/confirming execution via the manager.\nfunc CancelExecution(ctx *types.Context, execID string) error {\n\tmgr, err := getManager()\n\tif err != nil {\n\t\treturn fmt.Errorf(\"cancel not available: %w\", err)\n\t}\n\treturn mgr.CancelExecution(ctx, execID)\n}\n" }, { "path": "agent/robot/api/interact_test.go", "content": "package api\n\nimport (\n\t\"testing\"\n\n\t\"github.com/stretchr/testify/assert\"\n\t\"github.com/stretchr/testify/require\"\n\t\"github.com/yaoapp/yao/agent/robot/types\"\n)\n\n// AI1-AI3: Interact routing\nfunc TestInteract(t *testing.T) {\n\tt.Run(\"empty member_id returns error\", func(t *testing.T) {\n\t\tctx := types.NewContext(nil, nil)\n\t\t_, err := Interact(ctx, \"\", &InteractRequest{Message: \"test\"})\n\t\tassert.Error(t, err)\n\t\tassert.Contains(t, err.Error(), \"member_id is required\")\n\t})\n\n\tt.Run(\"nil request returns error\", func(t *testing.T) {\n\t\tctx := types.NewContext(nil, nil)\n\t\t_, err := Interact(ctx, \"member-1\", nil)\n\t\tassert.Error(t, err)\n\t\tassert.Contains(t, err.Error(), \"interact request is required\")\n\t})\n\n\tt.Run(\"no manager and no execution_id returns error\", func(t *testing.T) {\n\t\tctx := types.NewContext(nil, nil)\n\t\t_, err := Interact(ctx, \"member-1\", &InteractRequest{Message: \"test\"})\n\t\tassert.Error(t, err)\n\t\tassert.Contains(t, err.Error(), \"execution_id is required\")\n\t})\n}\n\n// AI6: Reply shortcut\nfunc TestReply(t *testing.T) {\n\tt.Run(\"empty member_id returns error\", func(t *testing.T) {\n\t\tctx := types.NewContext(nil, nil)\n\t\t_, err := Reply(ctx, \"\", \"exec-1\", \"task-1\", \"hello\")\n\t\tassert.Error(t, err)\n\t\tassert.Contains(t, err.Error(), \"member_id is required\")\n\t})\n\n\tt.Run(\"routes through Interact\", func(t *testing.T) {\n\t\tctx := types.NewContext(nil, nil)\n\t\t// legacyResume accesses the DB model which panics if not initialized.\n\t\t// Verify the routing reaches legacyResume by catching the expected panic.\n\t\tassert.Panics(t, func() {\n\t\t\tReply(ctx, \"member-1\", \"exec-1\", \"task-1\", \"hello\")\n\t\t}, \"should reach legacyResume which requires DB model\")\n\t})\n}\n\n// AI7: Confirm shortcut\nfunc TestConfirm(t *testing.T) {\n\tt.Run(\"empty member_id returns error\", func(t *testing.T) {\n\t\tctx := types.NewContext(nil, nil)\n\t\t_, err := Confirm(ctx, \"\", \"exec-1\", \"yes\")\n\t\tassert.Error(t, err)\n\t\tassert.Contains(t, err.Error(), \"member_id is required\")\n\t})\n\n\tt.Run(\"routes through Interact\", func(t *testing.T) {\n\t\tctx := types.NewContext(nil, nil)\n\t\tassert.Panics(t, func() {\n\t\t\tConfirm(ctx, \"member-1\", \"exec-1\", \"yes\")\n\t\t}, \"should reach legacyResume which requires DB model\")\n\t})\n}\n\n// AI8-AI9: CancelExecution\nfunc TestCancelExecution(t *testing.T) {\n\tt.Run(\"no manager returns error\", func(t *testing.T) {\n\t\tctx := types.NewContext(nil, nil)\n\t\terr := CancelExecution(ctx, \"exec-1\")\n\t\tassert.Error(t, err)\n\t\tassert.Contains(t, err.Error(), \"cancel not available\")\n\t})\n}\n\n// AI10-AI12: legacyResume\nfunc TestLegacyResume(t *testing.T) {\n\tt.Run(\"non-existent execution panics without DB\", func(t *testing.T) {\n\t\tctx := types.NewContext(nil, nil)\n\t\tassert.Panics(t, func() {\n\t\t\tlegacyResume(ctx, &InteractRequest{\n\t\t\t\tExecutionID: \"nonexistent-exec\",\n\t\t\t\tMessage: \"test\",\n\t\t\t})\n\t\t}, \"should panic because DB model is not initialized\")\n\t})\n}\n\n// AI1: managerInteract delegates correctly\nfunc TestManagerInteract(t *testing.T) {\n\tt.Run(\"converts request fields correctly\", func(t *testing.T) {\n\t\t// This would require a running Manager; test the field mapping logic\n\t\treq := &InteractRequest{\n\t\t\tExecutionID: \"exec-ai1\",\n\t\t\tTaskID: \"task-ai1\",\n\t\t\tSource: types.InteractSourceUI,\n\t\t\tMessage: \"do it\",\n\t\t\tAction: \"confirm\",\n\t\t}\n\n\t\t// Verify InteractRequest has all expected fields\n\t\tassert.Equal(t, \"exec-ai1\", req.ExecutionID)\n\t\tassert.Equal(t, \"task-ai1\", req.TaskID)\n\t\tassert.Equal(t, types.InteractSourceUI, req.Source)\n\t\tassert.Equal(t, \"do it\", req.Message)\n\t\tassert.Equal(t, \"confirm\", req.Action)\n\t})\n}\n\n// AI2: Interact with execution_id and no manager falls back to legacy\nfunc TestInteractLegacyFallback(t *testing.T) {\n\tt.Run(\"with execution_id delegates to legacyResume\", func(t *testing.T) {\n\t\tctx := types.NewContext(nil, nil)\n\t\tassert.Panics(t, func() {\n\t\t\tInteract(ctx, \"member-1\", &InteractRequest{\n\t\t\t\tExecutionID: \"exec-1\",\n\t\t\t\tMessage: \"resume this\",\n\t\t\t})\n\t\t}, \"should reach legacyResume which requires DB model\")\n\t})\n}\n\n// Test InteractResult field mapping\nfunc TestInteractResultFields(t *testing.T) {\n\tresult := &InteractResult{\n\t\tExecutionID: \"exec-test\",\n\t\tStatus: \"confirmed\",\n\t\tMessage: \"Done\",\n\t\tChatID: \"chat-test\",\n\t\tReply: \"I'll do it\",\n\t\tWaitForMore: true,\n\t}\n\n\tassert.Equal(t, \"exec-test\", result.ExecutionID)\n\tassert.Equal(t, \"confirmed\", result.Status)\n\tassert.Equal(t, \"Done\", result.Message)\n\tassert.Equal(t, \"chat-test\", result.ChatID)\n\tassert.Equal(t, \"I'll do it\", result.Reply)\n\tassert.True(t, result.WaitForMore)\n\n\t// Verify zero-value result\n\tempty := &InteractResult{}\n\tassert.Empty(t, empty.ExecutionID)\n\tassert.Empty(t, empty.Status)\n\tassert.False(t, empty.WaitForMore)\n}\n\n// Test that legacyResume returns \"waiting\" on ErrExecutionSuspended\nfunc TestLegacyResumeStatusMapping(t *testing.T) {\n\t// ErrExecutionSuspended handling is tested via the suspend E2E tests.\n\t// Here we verify the InteractResult field structure.\n\tresult := &InteractResult{\n\t\tExecutionID: \"exec-lr\",\n\t\tStatus: \"waiting\",\n\t\tMessage: \"Execution suspended again: needs more input\",\n\t}\n\tassert.Equal(t, \"waiting\", result.Status)\n\tassert.Contains(t, result.Message, \"suspended\")\n\n\tresultOK := &InteractResult{\n\t\tExecutionID: \"exec-lr2\",\n\t\tStatus: \"resumed\",\n\t\tMessage: \"Execution resumed and completed successfully\",\n\t}\n\trequire.Equal(t, \"resumed\", resultOK.Status)\n}\n" }, { "path": "agent/robot/api/lifecycle.go", "content": "package api\n\nimport (\n\t\"context\"\n\t\"fmt\"\n\t\"sync\"\n\n\trobotevents \"github.com/yaoapp/yao/agent/robot/events\"\n\t\"github.com/yaoapp/yao/agent/robot/events/integrations\"\n\tdtadapter \"github.com/yaoapp/yao/agent/robot/events/integrations/dingtalk\"\n\tdcadapter \"github.com/yaoapp/yao/agent/robot/events/integrations/discord\"\n\tfsadapter \"github.com/yaoapp/yao/agent/robot/events/integrations/feishu\"\n\t\"github.com/yaoapp/yao/agent/robot/events/integrations/telegram\"\n\t\"github.com/yaoapp/yao/agent/robot/logger\"\n\t\"github.com/yaoapp/yao/agent/robot/manager\"\n\t\"github.com/yaoapp/yao/agent/robot/types\"\n)\n\nvar log = logger.New(\"robot\")\n\nfunc init() {\n\trobotevents.RegisterTriggerFunc(func(ctx *types.Context, memberID string, triggerType types.TriggerType, data interface{}) (string, bool, error) {\n\t\tresult, err := TriggerManual(ctx, memberID, triggerType, data)\n\t\tif err != nil {\n\t\t\treturn \"\", false, err\n\t\t}\n\t\treturn result.ExecutionID, result.Accepted, nil\n\t})\n}\n\n// ==================== Lifecycle API ====================\n// These functions manage the robot agent system lifecycle\n\nvar (\n\tglobalManager *manager.Manager\n\tglobalDispatcher *integrations.Dispatcher\n\tmanagerMu sync.RWMutex\n)\n\n// Start starts the robot agent system\n// This initializes and starts the manager which handles:\n// - Robot cache loading\n// - Worker pool\n// - Clock ticker for scheduled triggers\nfunc Start() error {\n\tmanagerMu.Lock()\n\tdefer managerMu.Unlock()\n\n\tif globalManager != nil && globalManager.IsStarted() {\n\t\treturn fmt.Errorf(\"robot agent system already started\")\n\t}\n\n\t// Create new manager if not exists\n\tif globalManager == nil {\n\t\tglobalManager = manager.New()\n\t}\n\n\tif err := globalManager.Start(); err != nil {\n\t\treturn err\n\t}\n\n\t// Start integration dispatcher (Telegram polling, webhook subscriptions, etc.)\n\tadapters := map[string]integrations.Adapter{\n\t\t\"telegram\": telegram.NewAdapter(),\n\t\t\"feishu\": fsadapter.NewAdapter(),\n\t\t\"dingtalk\": dtadapter.NewAdapter(),\n\t\t\"discord\": dcadapter.NewAdapter(),\n\t}\n\tglobalDispatcher = integrations.NewDispatcher(globalManager.Cache(), adapters)\n\tif err := globalDispatcher.Start(context.Background()); err != nil {\n\t\tlog.Error(\"failed to start integration dispatcher: %v\", err)\n\t}\n\n\treturn nil\n}\n\n// StartWithConfig starts the robot agent system with custom configuration\nfunc StartWithConfig(config *manager.Config) error {\n\tmanagerMu.Lock()\n\tdefer managerMu.Unlock()\n\n\tif globalManager != nil && globalManager.IsStarted() {\n\t\treturn fmt.Errorf(\"robot agent system already started\")\n\t}\n\n\tglobalManager = manager.NewWithConfig(config)\n\treturn globalManager.Start()\n}\n\n// Stop stops the robot agent system gracefully\n// This will:\n// - Stop the clock ticker\n// - Stop cache auto-refresh\n// - Wait for running jobs to complete\n// - Stop the worker pool\nfunc Stop() error {\n\tmanagerMu.Lock()\n\tdefer managerMu.Unlock()\n\n\tif globalManager == nil {\n\t\treturn nil\n\t}\n\n\tif globalDispatcher != nil {\n\t\tglobalDispatcher.Stop()\n\t\tglobalDispatcher = nil\n\t}\n\n\terr := globalManager.Stop()\n\tif err != nil {\n\t\treturn err\n\t}\n\n\tglobalManager = nil\n\treturn nil\n}\n\n// IsRunning returns true if the robot agent system is running\nfunc IsRunning() bool {\n\tmanagerMu.RLock()\n\tdefer managerMu.RUnlock()\n\n\treturn globalManager != nil && globalManager.IsStarted()\n}\n\n// getManager returns the global manager instance\n// Returns error if manager is not started\nfunc getManager() (*manager.Manager, error) {\n\tmanagerMu.RLock()\n\tdefer managerMu.RUnlock()\n\n\tif globalManager == nil || !globalManager.IsStarted() {\n\t\treturn nil, fmt.Errorf(\"robot agent system not started\")\n\t}\n\treturn globalManager, nil\n}\n\n// SetManager sets the global manager instance (for testing)\nfunc SetManager(m *manager.Manager) {\n\tmanagerMu.Lock()\n\tdefer managerMu.Unlock()\n\tglobalManager = m\n}\n" }, { "path": "agent/robot/api/lifecycle_test.go", "content": "package api_test\n\nimport (\n\t\"testing\"\n\n\t\"github.com/stretchr/testify/assert\"\n\t\"github.com/stretchr/testify/require\"\n\t\"github.com/yaoapp/yao/agent/robot/api\"\n\t\"github.com/yaoapp/yao/agent/testutils\"\n)\n\n// TestLifecycle tests the Start/Stop lifecycle APIs\nfunc TestLifecycle(t *testing.T) {\n\tif testing.Short() {\n\t\tt.Skip(\"Skipping integration test\")\n\t}\n\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\tt.Run(\"start and stop cycle\", func(t *testing.T) {\n\t\t// Initially not running\n\t\tassert.False(t, api.IsRunning())\n\n\t\t// Start\n\t\terr := api.Start()\n\t\trequire.NoError(t, err)\n\t\tassert.True(t, api.IsRunning())\n\n\t\t// Start again should fail\n\t\terr = api.Start()\n\t\tassert.Error(t, err)\n\t\tassert.Contains(t, err.Error(), \"already started\")\n\n\t\t// Stop\n\t\terr = api.Stop()\n\t\trequire.NoError(t, err)\n\t\tassert.False(t, api.IsRunning())\n\n\t\t// Stop again should be no-op (not error)\n\t\terr = api.Stop()\n\t\tassert.NoError(t, err)\n\t})\n\n\tt.Run(\"can restart after stop\", func(t *testing.T) {\n\t\t// Start\n\t\terr := api.Start()\n\t\trequire.NoError(t, err)\n\t\tassert.True(t, api.IsRunning())\n\n\t\t// Stop\n\t\terr = api.Stop()\n\t\trequire.NoError(t, err)\n\t\tassert.False(t, api.IsRunning())\n\n\t\t// Start again should work\n\t\terr = api.Start()\n\t\trequire.NoError(t, err)\n\t\tassert.True(t, api.IsRunning())\n\n\t\t// Cleanup\n\t\tapi.Stop()\n\t})\n}\n" }, { "path": "agent/robot/api/results.go", "content": "package api\n\nimport (\n\t\"context\"\n\t\"fmt\"\n\t\"time\"\n\n\t\"github.com/yaoapp/yao/agent/robot/store\"\n\t\"github.com/yaoapp/yao/agent/robot/types\"\n)\n\n// ==================== Result Types ====================\n\n// ResultQuery - query parameters for listing results\ntype ResultQuery struct {\n\tTriggerType types.TriggerType `json:\"trigger_type,omitempty\"` // clock | human | event\n\tKeyword string `json:\"keyword,omitempty\"` // Search in name/summary\n\tPage int `json:\"page,omitempty\"`\n\tPageSize int `json:\"pagesize,omitempty\"`\n}\n\n// ResultItem - result list item (subset of execution)\ntype ResultItem struct {\n\tID string `json:\"id\"`\n\tMemberID string `json:\"member_id\"`\n\tTriggerType types.TriggerType `json:\"trigger_type\"`\n\tStatus types.ExecStatus `json:\"status\"`\n\tName string `json:\"name\"`\n\tSummary string `json:\"summary\"`\n\tStartTime time.Time `json:\"start_time\"`\n\tEndTime *time.Time `json:\"end_time,omitempty\"`\n\tHasAttachments bool `json:\"has_attachments\"`\n}\n\n// ResultDetail - full result with delivery content\ntype ResultDetail struct {\n\tID string `json:\"id\"`\n\tMemberID string `json:\"member_id\"`\n\tTriggerType types.TriggerType `json:\"trigger_type\"`\n\tStatus types.ExecStatus `json:\"status\"`\n\tName string `json:\"name\"`\n\tDelivery *types.DeliveryResult `json:\"delivery,omitempty\"`\n\tStartTime time.Time `json:\"start_time\"`\n\tEndTime *time.Time `json:\"end_time,omitempty\"`\n}\n\n// ResultListResponse - paginated response\ntype ResultListResponse struct {\n\tData []*ResultItem `json:\"data\"`\n\tTotal int `json:\"total\"`\n\tPage int `json:\"page\"`\n\tPageSize int `json:\"pagesize\"`\n}\n\n// ==================== Result API Functions ====================\n\n// ListResults returns completed executions with delivery content for a robot\nfunc ListResults(ctx *types.Context, memberID string, query *ResultQuery) (*ResultListResponse, error) {\n\tif memberID == \"\" {\n\t\treturn nil, fmt.Errorf(\"member_id is required\")\n\t}\n\n\tif query == nil {\n\t\tquery = &ResultQuery{}\n\t}\n\tquery.applyDefaults()\n\n\topts := &store.ResultListOptions{\n\t\tMemberID: memberID,\n\t\tPage: query.Page,\n\t\tPageSize: query.PageSize,\n\t}\n\n\tif query.TriggerType != \"\" {\n\t\topts.TriggerType = query.TriggerType\n\t}\n\tif query.Keyword != \"\" {\n\t\topts.Keyword = query.Keyword\n\t}\n\n\t// Query from store\n\tresult, err := getExecutionStore().ListResults(context.Background(), opts)\n\tif err != nil {\n\t\treturn nil, fmt.Errorf(\"failed to list results: %w\", err)\n\t}\n\n\t// Transform to ResultItem slice\n\titems := make([]*ResultItem, 0, len(result.Data))\n\tfor _, record := range result.Data {\n\t\titem := recordToResultItem(record)\n\t\tif item != nil {\n\t\t\titems = append(items, item)\n\t\t}\n\t}\n\n\treturn &ResultListResponse{\n\t\tData: items,\n\t\tTotal: result.Total,\n\t\tPage: result.Page,\n\t\tPageSize: result.PageSize,\n\t}, nil\n}\n\n// GetResult returns a single result by execution ID\nfunc GetResult(ctx *types.Context, execID string) (*ResultDetail, error) {\n\tif execID == \"\" {\n\t\treturn nil, fmt.Errorf(\"execution_id is required\")\n\t}\n\n\t// Get from store\n\trecord, err := getExecutionStore().Get(context.Background(), execID)\n\tif err != nil {\n\t\treturn nil, fmt.Errorf(\"failed to get result: %w\", err)\n\t}\n\tif record == nil {\n\t\treturn nil, fmt.Errorf(\"result not found: %s\", execID)\n\t}\n\n\t// Verify it has delivery content\n\tif record.Delivery == nil || record.Delivery.Content == nil {\n\t\treturn nil, fmt.Errorf(\"result not found: %s (no delivery content)\", execID)\n\t}\n\n\treturn recordToResultDetail(record), nil\n}\n\n// ==================== Helper Functions ====================\n\n// applyDefaults applies default values to ResultQuery\nfunc (q *ResultQuery) applyDefaults() {\n\tif q.Page <= 0 {\n\t\tq.Page = 1\n\t}\n\tif q.PageSize <= 0 {\n\t\tq.PageSize = 20\n\t}\n\tif q.PageSize > 100 {\n\t\tq.PageSize = 100\n\t}\n}\n\n// recordToResultItem converts ExecutionRecord to ResultItem\nfunc recordToResultItem(record *store.ExecutionRecord) *ResultItem {\n\tif record == nil {\n\t\treturn nil\n\t}\n\n\titem := &ResultItem{\n\t\tID: record.ExecutionID,\n\t\tMemberID: record.MemberID,\n\t\tTriggerType: record.TriggerType,\n\t\tStatus: record.Status,\n\t\tName: record.Name,\n\t}\n\n\t// Set times\n\tif record.StartTime != nil {\n\t\titem.StartTime = *record.StartTime\n\t}\n\titem.EndTime = record.EndTime\n\n\t// Extract summary and attachments from delivery\n\tif record.Delivery != nil && record.Delivery.Content != nil {\n\t\titem.Summary = record.Delivery.Content.Summary\n\t\titem.HasAttachments = len(record.Delivery.Content.Attachments) > 0\n\t}\n\n\treturn item\n}\n\n// recordToResultDetail converts ExecutionRecord to ResultDetail\nfunc recordToResultDetail(record *store.ExecutionRecord) *ResultDetail {\n\tif record == nil {\n\t\treturn nil\n\t}\n\n\tdetail := &ResultDetail{\n\t\tID: record.ExecutionID,\n\t\tMemberID: record.MemberID,\n\t\tTriggerType: record.TriggerType,\n\t\tStatus: record.Status,\n\t\tName: record.Name,\n\t\tDelivery: record.Delivery,\n\t}\n\n\t// Set times\n\tif record.StartTime != nil {\n\t\tdetail.StartTime = *record.StartTime\n\t}\n\tdetail.EndTime = record.EndTime\n\n\treturn detail\n}\n" }, { "path": "agent/robot/api/robot.go", "content": "package api\n\nimport (\n\t\"context\"\n\t\"fmt\"\n\t\"time\"\n\n\tgonanoid \"github.com/matoous/go-nanoid/v2\"\n\t\"github.com/yaoapp/gou/model\"\n\t\"github.com/yaoapp/kun/maps\"\n\trobotevents \"github.com/yaoapp/yao/agent/robot/events\"\n\t\"github.com/yaoapp/yao/agent/robot/store\"\n\t\"github.com/yaoapp/yao/agent/robot/types\"\n\t\"github.com/yaoapp/yao/event\"\n)\n\n// ==================== Robot Query API ====================\n// These functions query robot information\n\n// memberModel is the model name for member table\nconst memberModel = \"__yao.member\"\n\n// robotStore is the shared robot store instance\nvar robotStore = store.NewRobotStore()\n\n// executionStore is the shared execution store instance\nvar executionStore = store.NewExecutionStore()\n\n// GetRobot returns a robot by member ID\n// Returns the robot from cache if available, otherwise loads from database\nfunc GetRobot(ctx *types.Context, memberID string) (*types.Robot, error) {\n\tif memberID == \"\" {\n\t\treturn nil, fmt.Errorf(\"member_id is required\")\n\t}\n\n\tmgr, err := getManager()\n\tif err != nil {\n\t\t// Manager not started, try to load directly from database\n\t\treturn loadRobotFromDB(memberID)\n\t}\n\n\t// Try cache first\n\trobot := mgr.Cache().Get(memberID)\n\tif robot != nil {\n\t\treturn robot, nil\n\t}\n\n\t// Not in cache, try to load from database\n\trobot, err = mgr.Cache().LoadByID(ctx, memberID)\n\tif err != nil {\n\t\treturn nil, err\n\t}\n\n\treturn robot, nil\n}\n\n// ListRobots returns robots with pagination and filtering\nfunc ListRobots(ctx *types.Context, query *ListQuery) (*ListResult, error) {\n\tif query == nil {\n\t\tquery = &ListQuery{}\n\t}\n\tquery.applyDefaults()\n\n\tmgr, err := getManager()\n\tif err != nil {\n\t\t// Manager not started, load directly from database\n\t\treturn listRobotsFromDB(query)\n\t}\n\n\t// If only teamID specified AND explicitly filtering for autonomous_mode=true, use cache\n\t// Cache only contains autonomous_mode=true robots\n\t// When autonomous_mode is not specified or false, must query database to include all robots\n\tif query.TeamID != \"\" && query.Status == \"\" && query.Keywords == \"\" && query.ClockMode == \"\" &&\n\t\tquery.AutonomousMode != nil && *query.AutonomousMode == true {\n\t\trobots := mgr.Cache().List(query.TeamID)\n\t\treturn paginateRobots(robots, query), nil\n\t}\n\n\t// For complex queries, load from database\n\treturn listRobotsFromDB(query)\n}\n\n// GetRobotStatus returns the runtime status of a robot\nfunc GetRobotStatus(ctx *types.Context, memberID string) (*RobotState, error) {\n\tif memberID == \"\" {\n\t\treturn nil, fmt.Errorf(\"member_id is required\")\n\t}\n\n\trobot, err := GetRobot(ctx, memberID)\n\tif err != nil {\n\t\treturn nil, err\n\t}\n\n\t// Get permission fields from store (for access control)\n\trecord, _ := robotStore.Get(context.Background(), memberID)\n\n\tstate := &RobotState{\n\t\tMemberID: robot.MemberID,\n\t\tTeamID: robot.TeamID,\n\t\tDisplayName: robot.DisplayName,\n\t\tBio: robot.Bio,\n\t\tStatus: robot.Status,\n\t\tMaxRunning: 2, // default\n\t}\n\n\t// Add permission fields if available\n\tif record != nil {\n\t\tstate.YaoCreatedBy = record.YaoCreatedBy\n\t\tstate.YaoTeamID = record.YaoTeamID\n\t}\n\n\tif robot.Config != nil && robot.Config.Quota != nil {\n\t\tstate.MaxRunning = robot.Config.Quota.GetMax()\n\t}\n\n\t// Get running execution IDs from ExecutionStore (more reliable than in-memory)\n\t// This ensures we get accurate status even when robot is loaded from database\n\trunningResult, err := executionStore.List(context.Background(), &store.ListOptions{\n\t\tMemberID: memberID,\n\t\tStatus: types.ExecRunning,\n\t\tPageSize: 100,\n\t})\n\tif err == nil && runningResult != nil && len(runningResult.Data) > 0 {\n\t\tstate.Running = len(runningResult.Data)\n\t\tstate.RunningIDs = make([]string, 0, len(runningResult.Data))\n\t\tfor _, exec := range runningResult.Data {\n\t\t\tstate.RunningIDs = append(state.RunningIDs, exec.ExecutionID)\n\t\t}\n\t\t// Update status based on running count\n\t\tstate.Status = types.RobotWorking\n\t} else {\n\t\t// No running executions from store, check in-memory\n\t\texecutions := robot.GetExecutions()\n\t\tstate.Running = len(executions)\n\t\tstate.RunningIDs = make([]string, 0, len(executions))\n\t\tfor _, exec := range executions {\n\t\t\tstate.RunningIDs = append(state.RunningIDs, exec.ID)\n\t\t}\n\t\t// If there are running executions in memory, update status\n\t\tif state.Running > 0 {\n\t\t\tstate.Status = types.RobotWorking\n\t\t}\n\t}\n\n\t// Set last run time\n\tif !robot.LastRun.IsZero() {\n\t\tstate.LastRun = &robot.LastRun\n\t}\n\n\t// Set next run time\n\tif !robot.NextRun.IsZero() {\n\t\tstate.NextRun = &robot.NextRun\n\t}\n\n\treturn state, nil\n}\n\n// ==================== Helper Functions ====================\n\n// loadRobotFromDB loads a robot directly from database\nfunc loadRobotFromDB(memberID string) (*types.Robot, error) {\n\tm := model.Select(memberModel)\n\tif m == nil {\n\t\treturn nil, fmt.Errorf(\"model %s not found\", memberModel)\n\t}\n\n\trecords, err := m.Get(model.QueryParam{\n\t\tSelect: []interface{}{\n\t\t\t\"id\", \"member_id\", \"team_id\", \"display_name\", \"bio\",\n\t\t\t\"system_prompt\", \"robot_status\", \"autonomous_mode\",\n\t\t\t\"robot_config\", \"robot_email\", \"agents\", \"mcp_servers\",\n\t\t\t\"manager_id\", \"language_model\",\n\t\t},\n\t\tWheres: []model.QueryWhere{\n\t\t\t{Column: \"member_id\", Value: memberID},\n\t\t\t{Column: \"member_type\", Value: \"robot\"},\n\t\t},\n\t\tLimit: 1,\n\t})\n\tif err != nil {\n\t\treturn nil, fmt.Errorf(\"failed to load robot: %w\", err)\n\t}\n\n\tif len(records) == 0 {\n\t\treturn nil, types.ErrRobotNotFound\n\t}\n\n\treturn types.NewRobotFromMap(map[string]interface{}(records[0]))\n}\n\n// listRobotsFromDB loads robots from database with filtering\nfunc listRobotsFromDB(query *ListQuery) (*ListResult, error) {\n\tm := model.Select(memberModel)\n\tif m == nil {\n\t\treturn nil, fmt.Errorf(\"model %s not found\", memberModel)\n\t}\n\n\t// Build where conditions\n\twheres := []model.QueryWhere{\n\t\t{Column: \"member_type\", Value: \"robot\"},\n\t\t{Column: \"status\", Value: \"active\"},\n\t}\n\n\tif query.TeamID != \"\" {\n\t\twheres = append(wheres, model.QueryWhere{Column: \"team_id\", Value: query.TeamID})\n\t}\n\tif query.Status != \"\" {\n\t\twheres = append(wheres, model.QueryWhere{Column: \"robot_status\", Value: string(query.Status)})\n\t}\n\tif query.Keywords != \"\" {\n\t\twheres = append(wheres, model.QueryWhere{\n\t\t\tColumn: \"display_name\",\n\t\t\tOP: \"like\",\n\t\t\tValue: \"%\" + query.Keywords + \"%\",\n\t\t})\n\t}\n\tif query.AutonomousMode != nil {\n\t\twheres = append(wheres, model.QueryWhere{Column: \"autonomous_mode\", Value: *query.AutonomousMode})\n\t}\n\n\t// Build order\n\torders := []model.QueryOrder{}\n\tif query.Order != \"\" {\n\t\torders = append(orders, model.QueryOrder{Column: query.Order})\n\t} else {\n\t\torders = append(orders, model.QueryOrder{Column: \"created_at\", Option: \"desc\"})\n\t}\n\n\t// Execute paginated query\n\tresult, err := m.Paginate(model.QueryParam{\n\t\tSelect: []interface{}{\n\t\t\t\"id\", \"member_id\", \"team_id\", \"display_name\", \"bio\",\n\t\t\t\"system_prompt\", \"robot_status\", \"autonomous_mode\",\n\t\t\t\"robot_config\", \"robot_email\", \"agents\", \"mcp_servers\",\n\t\t\t\"language_model\",\n\t\t},\n\t\tWheres: wheres,\n\t\tOrders: orders,\n\t}, query.Page, query.PageSize)\n\tif err != nil {\n\t\treturn nil, fmt.Errorf(\"failed to list robots: %w\", err)\n\t}\n\n\t// Parse result\n\tlistResult := &ListResult{\n\t\tData: []*types.Robot{},\n\t\tPage: query.Page,\n\t\tPageSize: query.PageSize,\n\t}\n\n\t// Get total count\n\tif total, ok := result.Get(\"total\").(int); ok {\n\t\tlistResult.Total = total\n\t}\n\n\t// Parse robot records - handle both []maps.MapStr and []map[string]interface{}\n\tdata := result.Get(\"data\")\n\tswitch records := data.(type) {\n\tcase []maps.MapStr:\n\t\tfor _, record := range records {\n\t\t\trobot, err := types.NewRobotFromMap(map[string]interface{}(record))\n\t\t\tif err != nil {\n\t\t\t\tcontinue // skip invalid records\n\t\t\t}\n\t\t\tlistResult.Data = append(listResult.Data, robot)\n\t\t}\n\tcase []map[string]interface{}:\n\t\tfor _, record := range records {\n\t\t\trobot, err := types.NewRobotFromMap(record)\n\t\t\tif err != nil {\n\t\t\t\tcontinue // skip invalid records\n\t\t\t}\n\t\t\tlistResult.Data = append(listResult.Data, robot)\n\t\t}\n\t}\n\n\treturn listResult, nil\n}\n\n// paginateRobots applies pagination to a slice of robots\nfunc paginateRobots(robots []*types.Robot, query *ListQuery) *ListResult {\n\ttotal := len(robots)\n\n\t// Calculate offset\n\toffset := (query.Page - 1) * query.PageSize\n\tif offset >= total {\n\t\treturn &ListResult{\n\t\t\tData: []*types.Robot{},\n\t\t\tTotal: total,\n\t\t\tPage: query.Page,\n\t\t\tPageSize: query.PageSize,\n\t\t}\n\t}\n\n\t// Calculate end index\n\tend := offset + query.PageSize\n\tif end > total {\n\t\tend = total\n\t}\n\n\treturn &ListResult{\n\t\tData: robots[offset:end],\n\t\tTotal: total,\n\t\tPage: query.Page,\n\t\tPageSize: query.PageSize,\n\t}\n}\n\n// ==================== Robot CRUD API ====================\n// These functions create, update, and delete robots\n// They call store layer for persistence and manage cache\n// Request/Response types are defined in types.go\n\n// CreateRobot creates a new robot member\n// Calls store.RobotStore.Save() and refreshes cache\n// If member_id is not provided, it will be auto-generated\nfunc CreateRobot(ctx *types.Context, req *CreateRobotRequest) (*RobotResponse, error) {\n\t// Validate required fields\n\tif req.TeamID == \"\" {\n\t\treturn nil, fmt.Errorf(\"team_id is required\")\n\t}\n\tif req.DisplayName == \"\" {\n\t\treturn nil, fmt.Errorf(\"display_name is required\")\n\t}\n\n\t// Generate member_id if not provided\n\tif req.MemberID == \"\" {\n\t\tgeneratedID, err := generateMemberID(context.Background())\n\t\tif err != nil {\n\t\t\treturn nil, fmt.Errorf(\"failed to generate member_id: %w\", err)\n\t\t}\n\t\treq.MemberID = generatedID\n\t}\n\n\t// Check if robot already exists\n\texisting, err := robotStore.Get(context.Background(), req.MemberID)\n\tif err != nil {\n\t\treturn nil, fmt.Errorf(\"failed to check existing robot: %w\", err)\n\t}\n\tif existing != nil {\n\t\treturn nil, fmt.Errorf(\"robot with member_id '%s' already exists\", req.MemberID)\n\t}\n\n\t// Determine autonomous_mode value\n\tautonomousMode := false\n\tif req.AutonomousMode != nil {\n\t\tautonomousMode = *req.AutonomousMode\n\t}\n\n\t// Determine status values\n\tstatus := \"active\"\n\tif req.Status != \"\" {\n\t\tstatus = req.Status\n\t}\n\trobotStatus := \"idle\"\n\tif req.RobotStatus != \"\" {\n\t\trobotStatus = req.RobotStatus\n\t}\n\n\t// Create store record with all fields\n\tnow := time.Now()\n\trecord := &store.RobotRecord{\n\t\t// Required\n\t\tMemberID: req.MemberID,\n\t\tTeamID: req.TeamID,\n\t\tMemberType: \"robot\",\n\t\tStatus: status,\n\t\tRobotStatus: robotStatus,\n\t\tAutonomousMode: autonomousMode,\n\n\t\t// Profile\n\t\tDisplayName: req.DisplayName,\n\t\tBio: req.Bio,\n\t\tAvatar: req.Avatar,\n\n\t\t// Identity & Role\n\t\tSystemPrompt: req.SystemPrompt,\n\t\tRoleID: req.RoleID,\n\t\tManagerID: req.ManagerID,\n\n\t\t// Communication\n\t\tRobotEmail: req.RobotEmail,\n\t\tAuthorizedSenders: req.AuthorizedSenders,\n\t\tEmailFilterRules: req.EmailFilterRules,\n\n\t\t// Capabilities\n\t\tRobotConfig: req.RobotConfig,\n\t\tAgents: req.Agents,\n\t\tMCPServers: req.MCPServers,\n\t\tLanguageModel: req.LanguageModel,\n\n\t\t// Limits\n\t\tCostLimit: req.CostLimit,\n\n\t\t// Timestamps\n\t\tJoinedAt: &now,\n\t}\n\n\t// Apply Yao permission fields if provided\n\tif req.AuthScope != nil {\n\t\trecord.YaoCreatedBy = req.AuthScope.CreatedBy\n\t\trecord.YaoTeamID = req.AuthScope.TeamID\n\t\trecord.YaoTenantID = req.AuthScope.TenantID\n\t\t// Set invited_by from CreatedBy if not explicitly set\n\t\tif record.InvitedBy == \"\" && req.AuthScope.CreatedBy != \"\" {\n\t\t\trecord.InvitedBy = req.AuthScope.CreatedBy\n\t\t}\n\t}\n\n\t// Save to database\n\terr = robotStore.Save(context.Background(), record)\n\tif err != nil {\n\t\treturn nil, fmt.Errorf(\"failed to create robot: %w\", err)\n\t}\n\n\t// Refresh cache if manager is running\n\t// Use Refresh() which handles autonomous_mode correctly:\n\t// - If autonomous_mode=true: adds to cache for scheduling\n\t// - If autonomous_mode=false: does not add to cache\n\tmgr, err := getManager()\n\tif err == nil && mgr != nil {\n\t\t_ = mgr.Cache().Refresh(ctx, req.MemberID)\n\t}\n\n\t// Notify integrations of new robot config\n\tevent.Push(context.Background(), robotevents.RobotConfigCreated, robotevents.RobotConfigPayload{\n\t\tMemberID: req.MemberID,\n\t\tTeamID: req.TeamID,\n\t})\n\n\t// Return the created robot as response\n\treturn GetRobotResponse(ctx, req.MemberID)\n}\n\n// UpdateRobot updates an existing robot member\n// Calls store.RobotStore.Save() and refreshes cache\nfunc UpdateRobot(ctx *types.Context, memberID string, req *UpdateRobotRequest) (*RobotResponse, error) {\n\tif memberID == \"\" {\n\t\treturn nil, fmt.Errorf(\"member_id is required\")\n\t}\n\n\t// Get existing record\n\texisting, err := robotStore.Get(context.Background(), memberID)\n\tif err != nil {\n\t\treturn nil, fmt.Errorf(\"failed to get robot: %w\", err)\n\t}\n\tif existing == nil {\n\t\treturn nil, types.ErrRobotNotFound\n\t}\n\n\t// Apply updates - only non-nil fields are updated\n\t// Profile\n\tif req.DisplayName != nil {\n\t\texisting.DisplayName = *req.DisplayName\n\t}\n\tif req.Bio != nil {\n\t\texisting.Bio = *req.Bio\n\t}\n\tif req.Avatar != nil {\n\t\texisting.Avatar = *req.Avatar\n\t}\n\n\t// Identity & Role\n\tif req.SystemPrompt != nil {\n\t\texisting.SystemPrompt = *req.SystemPrompt\n\t}\n\tif req.RoleID != nil {\n\t\texisting.RoleID = *req.RoleID\n\t}\n\tif req.ManagerID != nil {\n\t\texisting.ManagerID = *req.ManagerID\n\t}\n\n\t// Status\n\tif req.Status != nil {\n\t\texisting.Status = *req.Status\n\t}\n\tif req.RobotStatus != nil {\n\t\texisting.RobotStatus = *req.RobotStatus\n\t}\n\tif req.AutonomousMode != nil {\n\t\texisting.AutonomousMode = *req.AutonomousMode\n\t}\n\n\t// Communication\n\tif req.RobotEmail != nil {\n\t\texisting.RobotEmail = *req.RobotEmail\n\t}\n\tif req.AuthorizedSenders != nil {\n\t\texisting.AuthorizedSenders = req.AuthorizedSenders\n\t}\n\tif req.EmailFilterRules != nil {\n\t\texisting.EmailFilterRules = req.EmailFilterRules\n\t}\n\n\t// Capabilities\n\tif req.RobotConfig != nil {\n\t\texisting.RobotConfig = req.RobotConfig\n\t}\n\tif req.Agents != nil {\n\t\texisting.Agents = req.Agents\n\t}\n\tif req.MCPServers != nil {\n\t\texisting.MCPServers = req.MCPServers\n\t}\n\tif req.LanguageModel != nil {\n\t\texisting.LanguageModel = *req.LanguageModel\n\t}\n\n\t// Limits\n\tif req.CostLimit != nil {\n\t\texisting.CostLimit = *req.CostLimit\n\t}\n\n\t// Apply Yao permission fields if provided (update scope)\n\tif req.AuthScope != nil {\n\t\texisting.YaoUpdatedBy = req.AuthScope.UpdatedBy\n\t\t// Team and Tenant are typically set on create, not update\n\t\t// But allow override if explicitly provided\n\t\tif req.AuthScope.TeamID != \"\" {\n\t\t\texisting.YaoTeamID = req.AuthScope.TeamID\n\t\t}\n\t\tif req.AuthScope.TenantID != \"\" {\n\t\t\texisting.YaoTenantID = req.AuthScope.TenantID\n\t\t}\n\t}\n\n\t// Save to database\n\terr = robotStore.Save(context.Background(), existing)\n\tif err != nil {\n\t\treturn nil, fmt.Errorf(\"failed to update robot: %w\", err)\n\t}\n\n\t// Refresh cache if manager is running\n\t// Use Refresh() which handles autonomous_mode correctly:\n\t// - If autonomous_mode=true: adds to cache for scheduling\n\t// - If autonomous_mode=false: removes from cache\n\tmgr, err := getManager()\n\tif err == nil && mgr != nil {\n\t\t_ = mgr.Cache().Refresh(ctx, memberID) // Ignore error, database is already saved\n\t}\n\n\t// Notify integrations of updated robot config\n\tevent.Push(context.Background(), robotevents.RobotConfigUpdated, robotevents.RobotConfigPayload{\n\t\tMemberID: memberID,\n\t\tTeamID: existing.TeamID,\n\t})\n\n\t// Return the updated robot as response\n\treturn GetRobotResponse(ctx, memberID)\n}\n\n// RemoveRobot deletes a robot member\n// Calls store.RobotStore.Delete() and invalidates cache\nfunc RemoveRobot(ctx *types.Context, memberID string) error {\n\tif memberID == \"\" {\n\t\treturn fmt.Errorf(\"member_id is required\")\n\t}\n\n\t// Check if robot exists\n\texisting, err := robotStore.Get(context.Background(), memberID)\n\tif err != nil {\n\t\treturn fmt.Errorf(\"failed to get robot: %w\", err)\n\t}\n\tif existing == nil {\n\t\treturn types.ErrRobotNotFound\n\t}\n\n\t// Check if robot has running executions\n\tmgr, err := getManager()\n\tif err == nil && mgr != nil {\n\t\trobot := mgr.Cache().Get(memberID)\n\t\tif robot != nil && robot.RunningCount() > 0 {\n\t\t\treturn fmt.Errorf(\"cannot delete robot with running executions\")\n\t\t}\n\t}\n\n\t// Delete from database\n\terr = robotStore.Delete(context.Background(), memberID)\n\tif err != nil {\n\t\treturn fmt.Errorf(\"failed to delete robot: %w\", err)\n\t}\n\n\t// Invalidate cache if manager is running\n\tif mgr != nil {\n\t\tmgr.Cache().Remove(memberID)\n\t}\n\n\t// Notify integrations of deleted robot config\n\tevent.Push(context.Background(), robotevents.RobotConfigDeleted, robotevents.RobotConfigPayload{\n\t\tMemberID: memberID,\n\t\tTeamID: existing.TeamID,\n\t})\n\n\treturn nil\n}\n\n// GetRobotResponse retrieves a robot and converts to API response format\nfunc GetRobotResponse(ctx *types.Context, memberID string) (*RobotResponse, error) {\n\trecord, err := robotStore.Get(context.Background(), memberID)\n\tif err != nil {\n\t\treturn nil, fmt.Errorf(\"failed to get robot: %w\", err)\n\t}\n\tif record == nil {\n\t\treturn nil, types.ErrRobotNotFound\n\t}\n\n\treturn recordToResponse(record), nil\n}\n\n// recordToResponse converts a store.RobotRecord to API RobotResponse\nfunc recordToResponse(record *store.RobotRecord) *RobotResponse {\n\treturn &RobotResponse{\n\t\tID: record.ID,\n\t\tMemberID: record.MemberID,\n\t\tTeamID: record.TeamID,\n\t\tStatus: record.Status,\n\t\tRobotStatus: record.RobotStatus,\n\t\tAutonomousMode: record.AutonomousMode,\n\n\t\tDisplayName: record.DisplayName,\n\t\tBio: record.Bio,\n\t\tAvatar: record.Avatar,\n\n\t\tSystemPrompt: record.SystemPrompt,\n\t\tRoleID: record.RoleID,\n\t\tManagerID: record.ManagerID,\n\n\t\tRobotEmail: record.RobotEmail,\n\t\tAuthorizedSenders: record.AuthorizedSenders,\n\t\tEmailFilterRules: record.EmailFilterRules,\n\n\t\tRobotConfig: record.RobotConfig,\n\t\tAgents: record.Agents,\n\t\tMCPServers: record.MCPServers,\n\t\tLanguageModel: record.LanguageModel,\n\n\t\tCostLimit: record.CostLimit,\n\t\tInvitedBy: record.InvitedBy,\n\t\tJoinedAt: record.JoinedAt,\n\t\tYaoCreatedBy: record.YaoCreatedBy,\n\t\tYaoTeamID: record.YaoTeamID,\n\t\tCreatedAt: record.CreatedAt,\n\t\tUpdatedAt: record.UpdatedAt,\n\t}\n}\n\n// ==================== Member ID Generation ====================\n\n// generateMemberID generates a unique member_id with collision detection\n// Uses 12-digit numeric ID to match existing pattern in openapi/oauth/providers/user\nfunc generateMemberID(ctx context.Context) (string, error) {\n\tconst maxRetries = 10\n\n\tfor i := 0; i < maxRetries; i++ {\n\t\t// Generate 12-digit numeric ID\n\t\tid, err := gonanoid.Generate(\"0123456789\", 12)\n\t\tif err != nil {\n\t\t\treturn \"\", fmt.Errorf(\"failed to generate member_id: %w\", err)\n\t\t}\n\n\t\t// Check if ID already exists\n\t\texists, err := memberIDExists(ctx, id)\n\t\tif err != nil {\n\t\t\treturn \"\", fmt.Errorf(\"failed to check member_id existence: %w\", err)\n\t\t}\n\n\t\tif !exists {\n\t\t\treturn id, nil\n\t\t}\n\t\t// ID exists, retry\n\t}\n\n\treturn \"\", fmt.Errorf(\"failed to generate unique member_id after %d retries\", maxRetries)\n}\n\n// memberIDExists checks if a member_id already exists in the database\nfunc memberIDExists(ctx context.Context, memberID string) (bool, error) {\n\tm := model.Select(memberModel)\n\tif m == nil {\n\t\treturn false, fmt.Errorf(\"model %s not found\", memberModel)\n\t}\n\n\tmembers, err := m.Get(model.QueryParam{\n\t\tSelect: []interface{}{\"id\"},\n\t\tWheres: []model.QueryWhere{\n\t\t\t{Column: \"member_id\", Value: memberID},\n\t\t},\n\t\tLimit: 1,\n\t})\n\n\tif err != nil {\n\t\treturn false, err\n\t}\n\n\treturn len(members) > 0, nil\n}\n" }, { "path": "agent/robot/api/robot_test.go", "content": "package api_test\n\nimport (\n\t\"context\"\n\t\"testing\"\n\n\t\"github.com/stretchr/testify/assert\"\n\t\"github.com/stretchr/testify/require\"\n\t\"github.com/yaoapp/yao/agent/robot/api\"\n\t\"github.com/yaoapp/yao/agent/robot/types\"\n\t\"github.com/yaoapp/yao/agent/testutils\"\n)\n\n// TestGetRobotValidation tests parameter validation for GetRobot\nfunc TestGetRobotValidation(t *testing.T) {\n\tif testing.Short() {\n\t\tt.Skip(\"Skipping integration test\")\n\t}\n\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\tt.Run(\"returns error for empty member_id\", func(t *testing.T) {\n\t\tctx := types.NewContext(context.Background(), nil)\n\t\trobot, err := api.GetRobot(ctx, \"\")\n\t\tassert.Error(t, err)\n\t\tassert.Nil(t, robot)\n\t\tassert.Contains(t, err.Error(), \"member_id is required\")\n\t})\n\n\tt.Run(\"returns error for non-existent robot\", func(t *testing.T) {\n\t\tctx := types.NewContext(context.Background(), nil)\n\t\trobot, err := api.GetRobot(ctx, \"non_existent_member_id_xyz\")\n\t\tassert.Error(t, err)\n\t\tassert.Nil(t, robot)\n\t})\n}\n\n// TestListRobotsValidation tests parameter validation for ListRobots\nfunc TestListRobotsValidation(t *testing.T) {\n\tif testing.Short() {\n\t\tt.Skip(\"Skipping integration test\")\n\t}\n\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\tctx := types.NewContext(context.Background(), nil)\n\n\tt.Run(\"applies default pagination when query is nil\", func(t *testing.T) {\n\t\tresult, err := api.ListRobots(ctx, nil)\n\t\tassert.NoError(t, err)\n\t\tassert.NotNil(t, result)\n\t\tassert.Equal(t, 1, result.Page)\n\t\tassert.Equal(t, 20, result.PageSize)\n\t})\n\n\tt.Run(\"applies default pagination when values are zero\", func(t *testing.T) {\n\t\tresult, err := api.ListRobots(ctx, &api.ListQuery{\n\t\t\tPage: 0,\n\t\t\tPageSize: 0,\n\t\t})\n\t\tassert.NoError(t, err)\n\t\tassert.NotNil(t, result)\n\t\tassert.Equal(t, 1, result.Page)\n\t\tassert.Equal(t, 20, result.PageSize)\n\t})\n\n\tt.Run(\"caps pagesize at 100\", func(t *testing.T) {\n\t\tresult, err := api.ListRobots(ctx, &api.ListQuery{\n\t\t\tPage: 1,\n\t\t\tPageSize: 500,\n\t\t})\n\t\tassert.NoError(t, err)\n\t\tassert.NotNil(t, result)\n\t\tassert.Equal(t, 100, result.PageSize)\n\t})\n}\n\n// TestGetRobotStatusValidation tests parameter validation for GetRobotStatus\nfunc TestGetRobotStatusValidation(t *testing.T) {\n\tif testing.Short() {\n\t\tt.Skip(\"Skipping integration test\")\n\t}\n\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\tt.Run(\"returns error for empty member_id\", func(t *testing.T) {\n\t\tctx := types.NewContext(context.Background(), nil)\n\t\tstatus, err := api.GetRobotStatus(ctx, \"\")\n\t\tassert.Error(t, err)\n\t\tassert.Nil(t, status)\n\t\tassert.Contains(t, err.Error(), \"member_id is required\")\n\t})\n\n\tt.Run(\"returns error for non-existent robot\", func(t *testing.T) {\n\t\tctx := types.NewContext(context.Background(), nil)\n\t\tstatus, err := api.GetRobotStatus(ctx, \"non_existent_member_id_xyz\")\n\t\tassert.Error(t, err)\n\t\tassert.Nil(t, status)\n\t})\n}\n\n// ==================== Robot CRUD API Tests ====================\n\n// TestCreateRobotValidation tests parameter validation for CreateRobot\nfunc TestCreateRobotValidation(t *testing.T) {\n\tif testing.Short() {\n\t\tt.Skip(\"Skipping integration test\")\n\t}\n\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\tctx := types.NewContext(context.Background(), nil)\n\n\tt.Run(\"auto_generates_member_id_when_empty\", func(t *testing.T) {\n\t\treq := &api.CreateRobotRequest{\n\t\t\tMemberID: \"\",\n\t\t\tTeamID: \"team_001\",\n\t\t\tDisplayName: \"Test Robot Auto ID\",\n\t\t}\n\t\tresult, err := api.CreateRobot(ctx, req)\n\t\trequire.NoError(t, err)\n\t\trequire.NotNil(t, result)\n\n\t\t// Verify member_id was auto-generated (12-digit numeric)\n\t\tassert.NotEmpty(t, result.MemberID)\n\t\tassert.Len(t, result.MemberID, 12, \"Auto-generated member_id should be 12 digits\")\n\n\t\t// Cleanup\n\t\t_ = api.RemoveRobot(ctx, result.MemberID)\n\t})\n\n\tt.Run(\"returns_error_for_empty_team_id\", func(t *testing.T) {\n\t\treq := &api.CreateRobotRequest{\n\t\t\tMemberID: \"robot_test_001\",\n\t\t\tTeamID: \"\",\n\t\t\tDisplayName: \"Test Robot\",\n\t\t}\n\t\tresult, err := api.CreateRobot(ctx, req)\n\t\tassert.Error(t, err)\n\t\tassert.Nil(t, result)\n\t\tassert.Contains(t, err.Error(), \"team_id is required\")\n\t})\n\n\tt.Run(\"returns_error_for_empty_display_name\", func(t *testing.T) {\n\t\treq := &api.CreateRobotRequest{\n\t\t\tMemberID: \"robot_test_001\",\n\t\t\tTeamID: \"team_001\",\n\t\t\tDisplayName: \"\",\n\t\t}\n\t\tresult, err := api.CreateRobot(ctx, req)\n\t\tassert.Error(t, err)\n\t\tassert.Nil(t, result)\n\t\tassert.Contains(t, err.Error(), \"display_name is required\")\n\t})\n}\n\n// TestCreateRobot tests the CreateRobot API function\nfunc TestCreateRobot(t *testing.T) {\n\tif testing.Short() {\n\t\tt.Skip(\"Skipping integration test\")\n\t}\n\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\t// Cleanup before and after\n\tcleanupAPITestRobots(t)\n\tdefer cleanupAPITestRobots(t)\n\n\tctx := types.NewContext(context.Background(), nil)\n\n\tt.Run(\"creates_robot_with_required_fields\", func(t *testing.T) {\n\t\treq := &api.CreateRobotRequest{\n\t\t\tMemberID: \"api_robot_create_001\",\n\t\t\tTeamID: \"api_team_001\",\n\t\t\tDisplayName: \"API Test Robot\",\n\t\t}\n\n\t\tresult, err := api.CreateRobot(ctx, req)\n\t\trequire.NoError(t, err)\n\t\trequire.NotNil(t, result)\n\n\t\tassert.Equal(t, \"api_robot_create_001\", result.MemberID)\n\t\tassert.Equal(t, \"api_team_001\", result.TeamID)\n\t\tassert.Equal(t, \"API Test Robot\", result.DisplayName)\n\t\tassert.Equal(t, \"active\", result.Status)\n\t\tassert.Equal(t, \"idle\", result.RobotStatus)\n\t})\n\n\tt.Run(\"creates_robot_with_all_fields\", func(t *testing.T) {\n\t\tautonomousMode := true\n\t\treq := &api.CreateRobotRequest{\n\t\t\tMemberID: \"api_robot_create_002\",\n\t\t\tTeamID: \"api_team_002\",\n\t\t\tDisplayName: \"Full Robot\",\n\t\t\tBio: \"A fully configured robot\",\n\t\t\tSystemPrompt: \"You are a helpful assistant\",\n\t\t\tAvatar: \"https://example.com/avatar.png\",\n\t\t\tRoleID: \"admin\",\n\t\t\tManagerID: \"user_001\",\n\t\t\tAutonomousMode: &autonomousMode,\n\t\t\tRobotEmail: \"fullrobot@test.com\",\n\t\t\tLanguageModel: \"gpt-4\",\n\t\t\tCostLimit: 100.0,\n\t\t\tRobotConfig: map[string]interface{}{\n\t\t\t\t\"clock_mode\": \"on\",\n\t\t\t\t\"max_concurrent\": 3,\n\t\t\t},\n\t\t}\n\n\t\tresult, err := api.CreateRobot(ctx, req)\n\t\trequire.NoError(t, err)\n\t\trequire.NotNil(t, result)\n\n\t\tassert.Equal(t, \"api_robot_create_002\", result.MemberID)\n\t\tassert.Equal(t, \"Full Robot\", result.DisplayName)\n\t\tassert.Equal(t, \"A fully configured robot\", result.Bio)\n\t\tassert.Equal(t, \"You are a helpful assistant\", result.SystemPrompt)\n\t\tassert.Equal(t, \"admin\", result.RoleID)\n\t\tassert.True(t, result.AutonomousMode)\n\t\tassert.Equal(t, \"fullrobot@test.com\", result.RobotEmail)\n\t\tassert.Equal(t, \"gpt-4\", result.LanguageModel)\n\t\tassert.Equal(t, 100.0, result.CostLimit)\n\t})\n\n\tt.Run(\"creates_robot_with_auth_scope\", func(t *testing.T) {\n\t\treq := &api.CreateRobotRequest{\n\t\t\tMemberID: \"api_robot_create_003\",\n\t\t\tTeamID: \"api_team_003\",\n\t\t\tDisplayName: \"Robot with Auth\",\n\t\t\tAuthScope: &api.AuthScope{\n\t\t\t\tCreatedBy: \"user_123\",\n\t\t\t\tTeamID: \"perm_team_001\",\n\t\t\t\tTenantID: \"tenant_001\",\n\t\t\t},\n\t\t}\n\n\t\tresult, err := api.CreateRobot(ctx, req)\n\t\trequire.NoError(t, err)\n\t\trequire.NotNil(t, result)\n\n\t\tassert.Equal(t, \"api_robot_create_003\", result.MemberID)\n\t\t// InvitedBy should be set from AuthScope.CreatedBy\n\t\tassert.Equal(t, \"user_123\", result.InvitedBy)\n\t})\n\n\tt.Run(\"returns_error_for_duplicate_member_id\", func(t *testing.T) {\n\t\treq := &api.CreateRobotRequest{\n\t\t\tMemberID: \"api_robot_create_001\", // Already created above\n\t\t\tTeamID: \"api_team_001\",\n\t\t\tDisplayName: \"Duplicate Robot\",\n\t\t}\n\n\t\tresult, err := api.CreateRobot(ctx, req)\n\t\tassert.Error(t, err)\n\t\tassert.Nil(t, result)\n\t\tassert.Contains(t, err.Error(), \"already exists\")\n\t})\n}\n\n// TestUpdateRobot tests the UpdateRobot API function\nfunc TestUpdateRobot(t *testing.T) {\n\tif testing.Short() {\n\t\tt.Skip(\"Skipping integration test\")\n\t}\n\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\tcleanupAPITestRobots(t)\n\tdefer cleanupAPITestRobots(t)\n\n\tctx := types.NewContext(context.Background(), nil)\n\n\t// Create a robot to update\n\tcreateReq := &api.CreateRobotRequest{\n\t\tMemberID: \"api_robot_update_001\",\n\t\tTeamID: \"api_team_update\",\n\t\tDisplayName: \"Original Name\",\n\t\tBio: \"Original bio\",\n\t}\n\t_, err := api.CreateRobot(ctx, createReq)\n\trequire.NoError(t, err)\n\n\tt.Run(\"returns_error_for_empty_member_id\", func(t *testing.T) {\n\t\treq := &api.UpdateRobotRequest{}\n\t\tresult, err := api.UpdateRobot(ctx, \"\", req)\n\t\tassert.Error(t, err)\n\t\tassert.Nil(t, result)\n\t\tassert.Contains(t, err.Error(), \"member_id is required\")\n\t})\n\n\tt.Run(\"returns_error_for_non_existent_robot\", func(t *testing.T) {\n\t\tnewName := \"New Name\"\n\t\treq := &api.UpdateRobotRequest{\n\t\t\tDisplayName: &newName,\n\t\t}\n\t\tresult, err := api.UpdateRobot(ctx, \"non_existent_robot\", req)\n\t\tassert.Error(t, err)\n\t\tassert.Nil(t, result)\n\t})\n\n\tt.Run(\"updates_display_name\", func(t *testing.T) {\n\t\tnewName := \"Updated Name\"\n\t\treq := &api.UpdateRobotRequest{\n\t\t\tDisplayName: &newName,\n\t\t}\n\n\t\tresult, err := api.UpdateRobot(ctx, \"api_robot_update_001\", req)\n\t\trequire.NoError(t, err)\n\t\trequire.NotNil(t, result)\n\n\t\tassert.Equal(t, \"Updated Name\", result.DisplayName)\n\t\t// Bio should be unchanged\n\t\tassert.Equal(t, \"Original bio\", result.Bio)\n\t})\n\n\tt.Run(\"updates_multiple_fields\", func(t *testing.T) {\n\t\tnewBio := \"New bio description\"\n\t\tnewPrompt := \"Updated system prompt\"\n\t\tautonomousMode := true\n\n\t\treq := &api.UpdateRobotRequest{\n\t\t\tBio: &newBio,\n\t\t\tSystemPrompt: &newPrompt,\n\t\t\tAutonomousMode: &autonomousMode,\n\t\t}\n\n\t\tresult, err := api.UpdateRobot(ctx, \"api_robot_update_001\", req)\n\t\trequire.NoError(t, err)\n\t\trequire.NotNil(t, result)\n\n\t\tassert.Equal(t, \"New bio description\", result.Bio)\n\t\tassert.Equal(t, \"Updated system prompt\", result.SystemPrompt)\n\t\tassert.True(t, result.AutonomousMode)\n\t})\n\n\tt.Run(\"updates_robot_status\", func(t *testing.T) {\n\t\tnewStatus := \"working\"\n\t\treq := &api.UpdateRobotRequest{\n\t\t\tRobotStatus: &newStatus,\n\t\t}\n\n\t\tresult, err := api.UpdateRobot(ctx, \"api_robot_update_001\", req)\n\t\trequire.NoError(t, err)\n\t\trequire.NotNil(t, result)\n\n\t\tassert.Equal(t, \"working\", result.RobotStatus)\n\t})\n\n\tt.Run(\"updates_config\", func(t *testing.T) {\n\t\tnewConfig := map[string]interface{}{\n\t\t\t\"clock_mode\": \"off\",\n\t\t\t\"max_concurrent\": 5,\n\t\t}\n\t\treq := &api.UpdateRobotRequest{\n\t\t\tRobotConfig: newConfig,\n\t\t}\n\n\t\tresult, err := api.UpdateRobot(ctx, \"api_robot_update_001\", req)\n\t\trequire.NoError(t, err)\n\t\trequire.NotNil(t, result)\n\n\t\tassert.NotNil(t, result.RobotConfig)\n\t})\n}\n\n// TestRemoveRobot tests the RemoveRobot API function\nfunc TestRemoveRobot(t *testing.T) {\n\tif testing.Short() {\n\t\tt.Skip(\"Skipping integration test\")\n\t}\n\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\tcleanupAPITestRobots(t)\n\tdefer cleanupAPITestRobots(t)\n\n\tctx := types.NewContext(context.Background(), nil)\n\n\tt.Run(\"returns_error_for_empty_member_id\", func(t *testing.T) {\n\t\terr := api.RemoveRobot(ctx, \"\")\n\t\tassert.Error(t, err)\n\t\tassert.Contains(t, err.Error(), \"member_id is required\")\n\t})\n\n\tt.Run(\"returns_error_for_non_existent_robot\", func(t *testing.T) {\n\t\terr := api.RemoveRobot(ctx, \"non_existent_robot\")\n\t\tassert.Error(t, err)\n\t})\n\n\tt.Run(\"removes_existing_robot\", func(t *testing.T) {\n\t\t// Create a robot\n\t\tcreateReq := &api.CreateRobotRequest{\n\t\t\tMemberID: \"api_robot_remove_001\",\n\t\t\tTeamID: \"api_team_remove\",\n\t\t\tDisplayName: \"Robot to Remove\",\n\t\t}\n\t\t_, err := api.CreateRobot(ctx, createReq)\n\t\trequire.NoError(t, err)\n\n\t\t// Verify it exists\n\t\trobot, err := api.GetRobot(ctx, \"api_robot_remove_001\")\n\t\trequire.NoError(t, err)\n\t\trequire.NotNil(t, robot)\n\n\t\t// Remove it\n\t\terr = api.RemoveRobot(ctx, \"api_robot_remove_001\")\n\t\trequire.NoError(t, err)\n\n\t\t// Verify it's gone\n\t\trobot, err = api.GetRobot(ctx, \"api_robot_remove_001\")\n\t\tassert.Error(t, err) // Should return error for non-existent\n\t})\n}\n\n// TestGetRobotResponse tests the GetRobotResponse API function\nfunc TestGetRobotResponse(t *testing.T) {\n\tif testing.Short() {\n\t\tt.Skip(\"Skipping integration test\")\n\t}\n\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\tcleanupAPITestRobots(t)\n\tdefer cleanupAPITestRobots(t)\n\n\tctx := types.NewContext(context.Background(), nil)\n\n\t// Create a robot\n\tautonomousMode := true\n\tcreateReq := &api.CreateRobotRequest{\n\t\tMemberID: \"api_robot_response_001\",\n\t\tTeamID: \"api_team_response\",\n\t\tDisplayName: \"Response Test Robot\",\n\t\tBio: \"Test bio for response\",\n\t\tSystemPrompt: \"Test prompt\",\n\t\tAutonomousMode: &autonomousMode,\n\t\tRobotEmail: \"response@test.com\",\n\t\tCostLimit: 50.0,\n\t}\n\t_, err := api.CreateRobot(ctx, createReq)\n\trequire.NoError(t, err)\n\n\tt.Run(\"returns_robot_response_format\", func(t *testing.T) {\n\t\tresult, err := api.GetRobotResponse(ctx, \"api_robot_response_001\")\n\t\trequire.NoError(t, err)\n\t\trequire.NotNil(t, result)\n\n\t\t// Verify all fields are present in response\n\t\tassert.Equal(t, \"api_robot_response_001\", result.MemberID)\n\t\tassert.Equal(t, \"api_team_response\", result.TeamID)\n\t\tassert.Equal(t, \"Response Test Robot\", result.DisplayName)\n\t\tassert.Equal(t, \"Test bio for response\", result.Bio)\n\t\tassert.Equal(t, \"Test prompt\", result.SystemPrompt)\n\t\tassert.True(t, result.AutonomousMode)\n\t\tassert.Equal(t, \"response@test.com\", result.RobotEmail)\n\t\tassert.Equal(t, 50.0, result.CostLimit)\n\t\tassert.Equal(t, \"active\", result.Status)\n\t\tassert.Equal(t, \"idle\", result.RobotStatus)\n\t})\n\n\tt.Run(\"returns_error_for_non_existent\", func(t *testing.T) {\n\t\tresult, err := api.GetRobotResponse(ctx, \"non_existent\")\n\t\tassert.Error(t, err)\n\t\tassert.Nil(t, result)\n\t})\n}\n\n// Note: cleanupAPITestRobots is defined in api_test.go (shared helper)\n" }, { "path": "agent/robot/api/trigger.go", "content": "package api\n\nimport (\n\t\"fmt\"\n\n\t\"github.com/yaoapp/yao/agent/robot/types\"\n)\n\n// ==================== Trigger API ====================\n// These functions handle robot execution triggers\n\n// Trigger starts a robot execution with the specified trigger type and request\n// This is the main entry point for triggering robot execution\nfunc Trigger(ctx *types.Context, memberID string, req *TriggerRequest) (*TriggerResult, error) {\n\tif memberID == \"\" {\n\t\treturn nil, fmt.Errorf(\"member_id is required\")\n\t}\n\tif req == nil {\n\t\treturn nil, fmt.Errorf(\"trigger request is required\")\n\t}\n\n\tmgr, err := getManager()\n\tif err != nil {\n\t\treturn nil, err\n\t}\n\n\tswitch req.Type {\n\tcase types.TriggerHuman:\n\t\treturn triggerHuman(ctx, mgr, memberID, req)\n\tcase types.TriggerEvent:\n\t\treturn triggerEvent(ctx, mgr, memberID, req)\n\tcase types.TriggerClock:\n\t\treturn triggerManual(ctx, mgr, memberID, req)\n\tdefault:\n\t\treturn nil, fmt.Errorf(\"invalid trigger type: %s\", req.Type)\n\t}\n}\n\n// TriggerManual manually triggers a robot execution (for testing or debugging)\n// This bypasses normal trigger validation and directly submits to the pool\nfunc TriggerManual(ctx *types.Context, memberID string, triggerType types.TriggerType, data interface{}) (*TriggerResult, error) {\n\tif memberID == \"\" {\n\t\treturn nil, fmt.Errorf(\"member_id is required\")\n\t}\n\n\tmgr, err := getManager()\n\tif err != nil {\n\t\treturn nil, err\n\t}\n\n\texecID, err := mgr.TriggerManual(ctx, memberID, triggerType, data)\n\tif err != nil {\n\t\treturn &TriggerResult{\n\t\t\tAccepted: false,\n\t\t\tMessage: err.Error(),\n\t\t}, nil\n\t}\n\n\treturn &TriggerResult{\n\t\tAccepted: true,\n\t\tExecutionID: execID,\n\t\tMessage: fmt.Sprintf(\"Manual trigger (%s) submitted\", triggerType),\n\t}, nil\n}\n\n// Intervene processes a human intervention request\n// Human intervention skips P0 (inspiration) and goes directly to P1 (goals)\nfunc Intervene(ctx *types.Context, memberID string, req *TriggerRequest) (*TriggerResult, error) {\n\tif memberID == \"\" {\n\t\treturn nil, fmt.Errorf(\"member_id is required\")\n\t}\n\tif req == nil {\n\t\treturn nil, fmt.Errorf(\"intervention request is required\")\n\t}\n\n\tmgr, err := getManager()\n\tif err != nil {\n\t\treturn nil, err\n\t}\n\n\treturn triggerHuman(ctx, mgr, memberID, req)\n}\n\n// HandleEvent processes an event trigger request\n// Event trigger skips P0 (inspiration) and goes directly to P1 (goals)\nfunc HandleEvent(ctx *types.Context, memberID string, req *TriggerRequest) (*TriggerResult, error) {\n\tif memberID == \"\" {\n\t\treturn nil, fmt.Errorf(\"member_id is required\")\n\t}\n\tif req == nil {\n\t\treturn nil, fmt.Errorf(\"event request is required\")\n\t}\n\n\tmgr, err := getManager()\n\tif err != nil {\n\t\treturn nil, err\n\t}\n\n\treturn triggerEvent(ctx, mgr, memberID, req)\n}\n\n// ==================== Internal Trigger Functions ====================\n\n// triggerHuman handles human intervention trigger\nfunc triggerHuman(ctx *types.Context, mgr managerInterface, memberID string, req *TriggerRequest) (*TriggerResult, error) {\n\t// Build intervention request\n\tinterveneReq := &types.InterveneRequest{\n\t\tMemberID: memberID,\n\t\tTeamID: ctx.TeamID(),\n\t\tAction: req.Action,\n\t\tMessages: req.Messages,\n\t\tPlanTime: req.PlanAt,\n\t\tExecutorMode: req.ExecutorMode,\n\t}\n\n\t// Call manager's Intervene\n\tresult, err := mgr.Intervene(ctx, interveneReq)\n\tif err != nil {\n\t\treturn &TriggerResult{\n\t\t\tAccepted: false,\n\t\t\tMessage: err.Error(),\n\t\t}, nil\n\t}\n\n\treturn &TriggerResult{\n\t\tAccepted: true,\n\t\tExecutionID: result.ExecutionID,\n\t\tMessage: result.Message,\n\t}, nil\n}\n\n// triggerEvent handles event trigger\nfunc triggerEvent(ctx *types.Context, mgr managerInterface, memberID string, req *TriggerRequest) (*TriggerResult, error) {\n\t// Build event request\n\teventReq := &types.EventRequest{\n\t\tMemberID: memberID,\n\t\tSource: string(req.Source),\n\t\tEventType: req.EventType,\n\t\tData: req.Data,\n\t\tExecutorMode: req.ExecutorMode,\n\t}\n\n\t// Call manager's HandleEvent\n\tresult, err := mgr.HandleEvent(ctx, eventReq)\n\tif err != nil {\n\t\treturn &TriggerResult{\n\t\t\tAccepted: false,\n\t\t\tMessage: err.Error(),\n\t\t}, nil\n\t}\n\n\treturn &TriggerResult{\n\t\tAccepted: true,\n\t\tExecutionID: result.ExecutionID,\n\t\tMessage: result.Message,\n\t}, nil\n}\n\n// triggerManual handles manual/clock trigger\nfunc triggerManual(ctx *types.Context, mgr managerInterface, memberID string, req *TriggerRequest) (*TriggerResult, error) {\n\t// For clock trigger, pass clock context if available\n\tvar data interface{}\n\tif req.Data != nil {\n\t\tdata = req.Data\n\t}\n\n\texecID, err := mgr.TriggerManual(ctx, memberID, req.Type, data)\n\tif err != nil {\n\t\treturn &TriggerResult{\n\t\t\tAccepted: false,\n\t\t\tMessage: err.Error(),\n\t\t}, nil\n\t}\n\n\treturn &TriggerResult{\n\t\tAccepted: true,\n\t\tExecutionID: execID,\n\t\tMessage: fmt.Sprintf(\"Trigger (%s) submitted\", req.Type),\n\t}, nil\n}\n\n// managerInterface defines the methods we need from manager\n// This allows for easier testing with mocks\ntype managerInterface interface {\n\tTriggerManual(ctx *types.Context, memberID string, trigger types.TriggerType, data interface{}) (string, error)\n\tIntervene(ctx *types.Context, req *types.InterveneRequest) (*types.ExecutionResult, error)\n\tHandleEvent(ctx *types.Context, req *types.EventRequest) (*types.ExecutionResult, error)\n\tPauseExecution(ctx *types.Context, execID string) error\n\tResumeExecution(ctx *types.Context, execID string) error\n\tStopExecution(ctx *types.Context, execID string) error\n}\n" }, { "path": "agent/robot/api/trigger_test.go", "content": "package api_test\n\nimport (\n\t\"context\"\n\t\"testing\"\n\n\t\"github.com/stretchr/testify/assert\"\n\t\"github.com/stretchr/testify/require\"\n\t\"github.com/yaoapp/yao/agent/robot/api\"\n\t\"github.com/yaoapp/yao/agent/robot/types\"\n\t\"github.com/yaoapp/yao/agent/testutils\"\n)\n\n// TestTriggerValidation tests parameter validation for Trigger\nfunc TestTriggerValidation(t *testing.T) {\n\tif testing.Short() {\n\t\tt.Skip(\"Skipping integration test\")\n\t}\n\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\tctx := types.NewContext(context.Background(), nil)\n\n\tt.Run(\"returns error for empty member_id\", func(t *testing.T) {\n\t\tresult, err := api.Trigger(ctx, \"\", &api.TriggerRequest{\n\t\t\tType: types.TriggerHuman,\n\t\t})\n\t\tassert.Error(t, err)\n\t\tassert.Nil(t, result)\n\t\tassert.Contains(t, err.Error(), \"member_id is required\")\n\t})\n\n\tt.Run(\"returns error for nil request\", func(t *testing.T) {\n\t\tresult, err := api.Trigger(ctx, \"test_member\", nil)\n\t\tassert.Error(t, err)\n\t\tassert.Nil(t, result)\n\t\tassert.Contains(t, err.Error(), \"trigger request is required\")\n\t})\n\n\tt.Run(\"returns error when manager not started\", func(t *testing.T) {\n\t\tresult, err := api.Trigger(ctx, \"test_member\", &api.TriggerRequest{\n\t\t\tType: types.TriggerHuman,\n\t\t})\n\t\tassert.Error(t, err)\n\t\tassert.Nil(t, result)\n\t\tassert.Contains(t, err.Error(), \"not started\")\n\t})\n}\n\n// TestTriggerManualValidation tests parameter validation for TriggerManual\nfunc TestTriggerManualValidation(t *testing.T) {\n\tif testing.Short() {\n\t\tt.Skip(\"Skipping integration test\")\n\t}\n\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\tctx := types.NewContext(context.Background(), nil)\n\n\tt.Run(\"returns error for empty member_id\", func(t *testing.T) {\n\t\tresult, err := api.TriggerManual(ctx, \"\", types.TriggerClock, nil)\n\t\tassert.Error(t, err)\n\t\tassert.Nil(t, result)\n\t\tassert.Contains(t, err.Error(), \"member_id is required\")\n\t})\n\n\tt.Run(\"returns error when manager not started\", func(t *testing.T) {\n\t\tresult, err := api.TriggerManual(ctx, \"test_member\", types.TriggerClock, nil)\n\t\tassert.Error(t, err)\n\t\tassert.Nil(t, result)\n\t\tassert.Contains(t, err.Error(), \"not started\")\n\t})\n}\n\n// TestInterveneValidation tests parameter validation for Intervene\nfunc TestInterveneValidation(t *testing.T) {\n\tif testing.Short() {\n\t\tt.Skip(\"Skipping integration test\")\n\t}\n\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\tctx := types.NewContext(context.Background(), nil)\n\n\tt.Run(\"returns error for empty member_id\", func(t *testing.T) {\n\t\tresult, err := api.Intervene(ctx, \"\", &api.TriggerRequest{\n\t\t\tType: types.TriggerHuman,\n\t\t\tAction: types.ActionTaskAdd,\n\t\t})\n\t\tassert.Error(t, err)\n\t\tassert.Nil(t, result)\n\t\tassert.Contains(t, err.Error(), \"member_id is required\")\n\t})\n\n\tt.Run(\"returns error for nil request\", func(t *testing.T) {\n\t\tresult, err := api.Intervene(ctx, \"test_member\", nil)\n\t\tassert.Error(t, err)\n\t\tassert.Nil(t, result)\n\t\tassert.Contains(t, err.Error(), \"intervention request is required\")\n\t})\n}\n\n// TestHandleEventValidation tests parameter validation for HandleEvent\nfunc TestHandleEventValidation(t *testing.T) {\n\tif testing.Short() {\n\t\tt.Skip(\"Skipping integration test\")\n\t}\n\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\tctx := types.NewContext(context.Background(), nil)\n\n\tt.Run(\"returns error for empty member_id\", func(t *testing.T) {\n\t\tresult, err := api.HandleEvent(ctx, \"\", &api.TriggerRequest{\n\t\t\tType: types.TriggerEvent,\n\t\t\tSource: types.EventWebhook,\n\t\t\tEventType: \"test.event\",\n\t\t})\n\t\tassert.Error(t, err)\n\t\tassert.Nil(t, result)\n\t\tassert.Contains(t, err.Error(), \"member_id is required\")\n\t})\n\n\tt.Run(\"returns error for nil request\", func(t *testing.T) {\n\t\tresult, err := api.HandleEvent(ctx, \"test_member\", nil)\n\t\tassert.Error(t, err)\n\t\tassert.Nil(t, result)\n\t\tassert.Contains(t, err.Error(), \"event request is required\")\n\t})\n}\n\n// TestTriggerWithManagerStarted tests trigger APIs when manager is running\nfunc TestTriggerWithManagerStarted(t *testing.T) {\n\tif testing.Short() {\n\t\tt.Skip(\"Skipping integration test\")\n\t}\n\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\t// Start manager\n\terr := api.Start()\n\trequire.NoError(t, err)\n\tdefer api.Stop()\n\n\tctx := types.NewContext(context.Background(), nil)\n\n\tt.Run(\"returns not accepted for non-existent robot\", func(t *testing.T) {\n\t\tresult, err := api.Trigger(ctx, \"non_existent_robot_xyz\", &api.TriggerRequest{\n\t\t\tType: types.TriggerHuman,\n\t\t\tAction: types.ActionTaskAdd,\n\t\t})\n\t\t// Should not error, but return not accepted\n\t\tassert.NoError(t, err)\n\t\tassert.NotNil(t, result)\n\t\tassert.False(t, result.Accepted)\n\t})\n\n\tt.Run(\"returns error for invalid trigger type\", func(t *testing.T) {\n\t\tresult, err := api.Trigger(ctx, \"test_member\", &api.TriggerRequest{\n\t\t\tType: \"invalid_type\",\n\t\t})\n\t\tassert.Error(t, err)\n\t\tassert.Nil(t, result)\n\t\tassert.Contains(t, err.Error(), \"invalid trigger type\")\n\t})\n}\n" }, { "path": "agent/robot/api/types.go", "content": "package api\n\nimport (\n\t\"time\"\n\n\tagentcontext \"github.com/yaoapp/yao/agent/context\"\n\t\"github.com/yaoapp/yao/agent/robot/types\"\n)\n\n// ListQuery - query options for List()\ntype ListQuery struct {\n\tTeamID string `json:\"team_id,omitempty\"`\n\tStatus types.RobotStatus `json:\"status,omitempty\"`\n\tKeywords string `json:\"keywords,omitempty\"`\n\tClockMode types.ClockMode `json:\"clock_mode,omitempty\"`\n\tAutonomousMode *bool `json:\"autonomous_mode,omitempty\"` // nil=all, true=autonomous only, false=on-demand only\n\tPage int `json:\"page,omitempty\"`\n\tPageSize int `json:\"pagesize,omitempty\"`\n\tOrder string `json:\"order,omitempty\"`\n}\n\n// ListResult - result of List()\ntype ListResult struct {\n\tData []*types.Robot `json:\"data\"`\n\tTotal int `json:\"total\"`\n\tPage int `json:\"page\"`\n\tPageSize int `json:\"pagesize\"`\n}\n\n// RobotState - runtime state from Status()\ntype RobotState struct {\n\tMemberID string `json:\"member_id\"`\n\tTeamID string `json:\"team_id\"`\n\tDisplayName string `json:\"display_name\"`\n\tBio string `json:\"bio,omitempty\"`\n\tStatus types.RobotStatus `json:\"status\"`\n\tRunning int `json:\"running\"`\n\tMaxRunning int `json:\"max_running\"`\n\tLastRun *time.Time `json:\"last_run,omitempty\"`\n\tNextRun *time.Time `json:\"next_run,omitempty\"`\n\tRunningIDs []string `json:\"running_ids,omitempty\"`\n\tYaoCreatedBy string `json:\"__yao_created_by,omitempty\"` // Creator user_id for permission check\n\tYaoTeamID string `json:\"__yao_team_id,omitempty\"` // Team ID for permission check\n}\n\n// ==================== Trigger Types ====================\n\n// TriggerRequest - request for Trigger()\n// Input uses []context.Message to support rich content (text, images, files, audio)\ntype TriggerRequest struct {\n\tType types.TriggerType `json:\"type\"` // human | event | clock\n\n\t// Human intervention fields (when Type = human)\n\tAction types.InterventionAction `json:\"action,omitempty\"`\n\tMessages []agentcontext.Message `json:\"messages,omitempty\"` // user's input (supports text, images, files)\n\tPlanAt *time.Time `json:\"plan_at,omitempty\"`\n\tInsertPosition InsertPosition `json:\"insert_at,omitempty\"`\n\tAtIndex int `json:\"at_index,omitempty\"`\n\n\t// Event fields (when Type = event)\n\tSource types.EventSource `json:\"source,omitempty\"`\n\tEventType string `json:\"event_type,omitempty\"`\n\tData map[string]interface{} `json:\"data,omitempty\"`\n\n\t// Executor mode (optional, overrides robot config)\n\tExecutorMode types.ExecutorMode `json:\"executor_mode,omitempty\"`\n\n\t// i18n support\n\tLocale string `json:\"locale,omitempty\"` // Locale for UI messages (e.g., \"en\", \"zh\")\n}\n\n// InsertPosition - where to insert task in queue\ntype InsertPosition string\n\nconst (\n\t// InsertFirst inserts at beginning (highest priority)\n\tInsertFirst InsertPosition = \"first\"\n\t// InsertLast appends at end (default)\n\tInsertLast InsertPosition = \"last\"\n\t// InsertNext inserts after current task\n\tInsertNext InsertPosition = \"next\"\n\t// InsertAt inserts at specific index (use AtIndex)\n\tInsertAt InsertPosition = \"at\"\n)\n\n// TriggerResult - result of Trigger()\ntype TriggerResult struct {\n\tAccepted bool `json:\"accepted\"`\n\tQueued bool `json:\"queued\"`\n\tExecution *types.Execution `json:\"execution,omitempty\"`\n\tExecutionID string `json:\"execution_id,omitempty\"` // Execution ID\n\tMessage string `json:\"message,omitempty\"`\n}\n\n// ==================== Execution Types ====================\n\n// ExecutionQuery - query options for GetExecutions()\ntype ExecutionQuery struct {\n\tStatus types.ExecStatus `json:\"status,omitempty\"`\n\tExcludeStatuses []types.ExecStatus `json:\"exclude_statuses,omitempty\"`\n\tTrigger types.TriggerType `json:\"trigger,omitempty\"`\n\tPage int `json:\"page,omitempty\"`\n\tPageSize int `json:\"pagesize,omitempty\"`\n}\n\n// ExecutionResult - result of GetExecutions()\ntype ExecutionResult struct {\n\tData []*types.Execution `json:\"data\"`\n\tTotal int `json:\"total\"`\n\tPage int `json:\"page\"`\n\tPageSize int `json:\"pagesize\"`\n}\n\n// ==================== CRUD Types ====================\n\n// AuthScope contains Yao permission fields for data scoping\n// These fields are used by Yao's permission system (when model has permission: true)\ntype AuthScope struct {\n\tCreatedBy string `json:\"__yao_created_by,omitempty\"` // Creator user_id\n\tUpdatedBy string `json:\"__yao_updated_by,omitempty\"` // Updater user_id\n\tTeamID string `json:\"__yao_team_id,omitempty\"` // Permission team scope\n\tTenantID string `json:\"__yao_tenant_id,omitempty\"` // Permission tenant scope\n}\n\n// CreateRobotRequest - request for CreateRobot()\ntype CreateRobotRequest struct {\n\t// Identity (member_id is optional - auto-generated if not provided)\n\tMemberID string `json:\"member_id,omitempty\"` // Unique robot identifier (auto-generated if empty)\n\tTeamID string `json:\"team_id\"` // Team ID (required)\n\n\t// Profile\n\tDisplayName string `json:\"display_name,omitempty\"` // Display name\n\tBio string `json:\"bio,omitempty\"` // Robot description\n\tAvatar string `json:\"avatar,omitempty\"` // Avatar URL\n\n\t// Identity & Role\n\tSystemPrompt string `json:\"system_prompt,omitempty\"` // System prompt\n\tRoleID string `json:\"role_id,omitempty\"` // Role within team\n\tManagerID string `json:\"manager_id,omitempty\"` // Direct manager user_id\n\n\t// Status\n\tStatus string `json:\"status,omitempty\"` // Member status: active | inactive | pending | suspended\n\tRobotStatus string `json:\"robot_status,omitempty\"` // Robot status: idle | working | paused | error | maintenance\n\tAutonomousMode *bool `json:\"autonomous_mode,omitempty\"` // Whether autonomous mode is enabled\n\n\t// Communication\n\tRobotEmail string `json:\"robot_email,omitempty\"` // Robot email address\n\tAuthorizedSenders interface{} `json:\"authorized_senders,omitempty\"` // Email whitelist (JSON array)\n\tEmailFilterRules interface{} `json:\"email_filter_rules,omitempty\"` // Email filter rules (JSON array)\n\n\t// Capabilities\n\tRobotConfig interface{} `json:\"robot_config,omitempty\"` // Robot config JSON\n\tAgents interface{} `json:\"agents,omitempty\"` // Accessible agents (JSON array)\n\tMCPServers interface{} `json:\"mcp_servers,omitempty\"` // MCP servers (JSON array)\n\tLanguageModel string `json:\"language_model,omitempty\"` // Language model name\n\n\t// Limits\n\tCostLimit float64 `json:\"cost_limit,omitempty\"` // Monthly cost limit USD\n\n\t// Auth scope (optional, used by OpenAPI layer via WithCreateScope)\n\tAuthScope *AuthScope `json:\"auth_scope,omitempty\"`\n}\n\n// UpdateRobotRequest - request for UpdateRobot()\ntype UpdateRobotRequest struct {\n\t// Profile\n\tDisplayName *string `json:\"display_name,omitempty\"` // Display name\n\tBio *string `json:\"bio,omitempty\"` // Robot description\n\tAvatar *string `json:\"avatar,omitempty\"` // Avatar URL\n\n\t// Identity & Role\n\tSystemPrompt *string `json:\"system_prompt,omitempty\"` // System prompt\n\tRoleID *string `json:\"role_id,omitempty\"` // Role within team\n\tManagerID *string `json:\"manager_id,omitempty\"` // Direct manager user_id\n\n\t// Status\n\tStatus *string `json:\"status,omitempty\"` // Member status\n\tRobotStatus *string `json:\"robot_status,omitempty\"` // Robot status\n\tAutonomousMode *bool `json:\"autonomous_mode,omitempty\"` // Autonomous mode\n\n\t// Communication\n\tRobotEmail *string `json:\"robot_email,omitempty\"` // Robot email address\n\tAuthorizedSenders interface{} `json:\"authorized_senders,omitempty\"` // Email whitelist\n\tEmailFilterRules interface{} `json:\"email_filter_rules,omitempty\"` // Email filter rules\n\n\t// Capabilities\n\tRobotConfig interface{} `json:\"robot_config,omitempty\"` // Robot config JSON\n\tAgents interface{} `json:\"agents,omitempty\"` // Accessible agents\n\tMCPServers interface{} `json:\"mcp_servers,omitempty\"` // MCP servers\n\tLanguageModel *string `json:\"language_model,omitempty\"` // Language model name\n\n\t// Limits\n\tCostLimit *float64 `json:\"cost_limit,omitempty\"` // Monthly cost limit USD\n\n\t// Auth scope (optional, used by OpenAPI layer via WithUpdateScope)\n\tAuthScope *AuthScope `json:\"auth_scope,omitempty\"`\n}\n\n// RobotResponse - response containing robot details for API\ntype RobotResponse struct {\n\t// Basic\n\tID int64 `json:\"id,omitempty\"`\n\tMemberID string `json:\"member_id\"`\n\tTeamID string `json:\"team_id\"`\n\tStatus string `json:\"status\"`\n\tRobotStatus string `json:\"robot_status\"`\n\tAutonomousMode bool `json:\"autonomous_mode\"`\n\n\t// Profile\n\tDisplayName string `json:\"display_name\"`\n\tBio string `json:\"bio,omitempty\"`\n\tAvatar string `json:\"avatar,omitempty\"`\n\n\t// Identity & Role\n\tSystemPrompt string `json:\"system_prompt,omitempty\"`\n\tRoleID string `json:\"role_id,omitempty\"`\n\tManagerID string `json:\"manager_id,omitempty\"`\n\n\t// Communication\n\tRobotEmail string `json:\"robot_email,omitempty\"`\n\tAuthorizedSenders interface{} `json:\"authorized_senders,omitempty\"`\n\tEmailFilterRules interface{} `json:\"email_filter_rules,omitempty\"`\n\n\t// Capabilities\n\tRobotConfig interface{} `json:\"robot_config,omitempty\"`\n\tAgents interface{} `json:\"agents,omitempty\"`\n\tMCPServers interface{} `json:\"mcp_servers,omitempty\"`\n\tLanguageModel string `json:\"language_model,omitempty\"`\n\n\t// Limits\n\tCostLimit float64 `json:\"cost_limit,omitempty\"`\n\n\t// Ownership & Audit\n\tInvitedBy string `json:\"invited_by,omitempty\"`\n\tJoinedAt *time.Time `json:\"joined_at,omitempty\"`\n\tYaoCreatedBy string `json:\"__yao_created_by,omitempty\"` // Creator user_id for permission check\n\tYaoTeamID string `json:\"__yao_team_id,omitempty\"` // Team ID for permission check\n\n\t// Timestamps\n\tCreatedAt *time.Time `json:\"created_at,omitempty\"`\n\tUpdatedAt *time.Time `json:\"updated_at,omitempty\"`\n}\n\n// ==================== Helper Functions ====================\n\n// applyDefaults applies default values to ListQuery\nfunc (q *ListQuery) applyDefaults() {\n\tif q.Page <= 0 {\n\t\tq.Page = 1\n\t}\n\tif q.PageSize <= 0 {\n\t\tq.PageSize = 20\n\t}\n\tif q.PageSize > 100 {\n\t\tq.PageSize = 100\n\t}\n}\n\n// applyDefaults applies default values to ExecutionQuery\nfunc (q *ExecutionQuery) applyDefaults() {\n\tif q.Page <= 0 {\n\t\tq.Page = 1\n\t}\n\tif q.PageSize <= 0 {\n\t\tq.PageSize = 20\n\t}\n\tif q.PageSize > 100 {\n\t\tq.PageSize = 100\n\t}\n}\n" }, { "path": "agent/robot/cache/cache.go", "content": "package cache\n\nimport (\n\t\"sync\"\n\n\t\"github.com/yaoapp/yao/agent/robot/types\"\n)\n\n// Cache implements types.Cache interface\n// Thread-safe in-memory cache for Robot instances\ntype Cache struct {\n\trobots map[string]*types.Robot // memberID -> Robot\n\tbyTeam map[string][]string // teamID -> memberIDs\n\tmu sync.RWMutex\n}\n\n// New creates a new cache instance\nfunc New() *Cache {\n\treturn &Cache{\n\t\trobots: make(map[string]*types.Robot),\n\t\tbyTeam: make(map[string][]string),\n\t}\n}\n\n// Get returns a robot by member ID\n// Stub: returns nil (will be implemented in Phase 3)\nfunc (c *Cache) Get(memberID string) *types.Robot {\n\tc.mu.RLock()\n\tdefer c.mu.RUnlock()\n\treturn c.robots[memberID]\n}\n\n// List returns all robots for a team\nfunc (c *Cache) List(teamID string) []*types.Robot {\n\tc.mu.RLock()\n\tdefer c.mu.RUnlock()\n\n\tmemberIDs := c.byTeam[teamID]\n\trobots := make([]*types.Robot, 0, len(memberIDs))\n\tfor _, memberID := range memberIDs {\n\t\tif robot := c.robots[memberID]; robot != nil {\n\t\t\trobots = append(robots, robot)\n\t\t}\n\t}\n\treturn robots\n}\n\n// Note: Refresh is implemented in refresh.go\n\n// Add adds or updates a robot in cache\nfunc (c *Cache) Add(robot *types.Robot) {\n\tif robot == nil {\n\t\treturn\n\t}\n\n\tc.mu.Lock()\n\tdefer c.mu.Unlock()\n\n\tc.robots[robot.MemberID] = robot\n\n\t// Update team index\n\tif _, exists := c.byTeam[robot.TeamID]; !exists {\n\t\tc.byTeam[robot.TeamID] = []string{}\n\t}\n\n\t// Check if member ID already in team list\n\tfound := false\n\tfor _, id := range c.byTeam[robot.TeamID] {\n\t\tif id == robot.MemberID {\n\t\t\tfound = true\n\t\t\tbreak\n\t\t}\n\t}\n\tif !found {\n\t\tc.byTeam[robot.TeamID] = append(c.byTeam[robot.TeamID], robot.MemberID)\n\t}\n}\n\n// Remove removes a robot from cache\nfunc (c *Cache) Remove(memberID string) {\n\tc.mu.Lock()\n\tdefer c.mu.Unlock()\n\n\trobot := c.robots[memberID]\n\tif robot == nil {\n\t\treturn\n\t}\n\n\tdelete(c.robots, memberID)\n\n\t// Remove from team index\n\tteamMembers := c.byTeam[robot.TeamID]\n\tfor i, id := range teamMembers {\n\t\tif id == memberID {\n\t\t\tc.byTeam[robot.TeamID] = append(teamMembers[:i], teamMembers[i+1:]...)\n\t\t\tbreak\n\t\t}\n\t}\n}\n" }, { "path": "agent/robot/cache/cache_test.go", "content": "package cache_test\n\nimport (\n\t\"context\"\n\t\"encoding/json\"\n\t\"testing\"\n\t\"time\"\n\n\t\"github.com/stretchr/testify/assert\"\n\t\"github.com/yaoapp/gou/model\"\n\t\"github.com/yaoapp/xun/capsule\"\n\t\"github.com/yaoapp/yao/agent/robot/cache\"\n\t\"github.com/yaoapp/yao/agent/robot/types\"\n\t\"github.com/yaoapp/yao/agent/testutils\"\n)\n\n// TestCacheLoad tests loading all active robots from database\nfunc TestCacheLoad(t *testing.T) {\n\tif testing.Short() {\n\t\tt.Skip(\"Skipping integration test\")\n\t}\n\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\t// Clean up any existing test data first\n\tcleanupTestRobots(t)\n\n\t// Create test robots in database\n\tsetupTestRobots(t)\n\tdefer cleanupTestRobots(t)\n\n\tc := cache.New()\n\tctx := types.NewContext(context.Background(), nil)\n\n\t// Load all robots\n\terr := c.Load(ctx)\n\tassert.NoError(t, err)\n\n\t// Count should be at least 2 (may have other robots in DB)\n\tcount := c.Count()\n\tassert.GreaterOrEqual(t, count, 2, \"Should load at least 2 active autonomous robots\")\n\n\t// Verify first robot\n\trobot1 := c.Get(\"robot_test_sales_001\")\n\tassert.NotNil(t, robot1, \"Sales bot should be loaded\")\n\tif robot1 == nil {\n\t\tt.Fatal(\"robot_test_sales_001 not found in cache\")\n\t}\n\tassert.Equal(t, \"robot_test_sales_001\", robot1.MemberID)\n\tassert.Equal(t, \"team_test_cache_001\", robot1.TeamID)\n\tassert.Equal(t, \"Test Sales Bot\", robot1.DisplayName)\n\tassert.Equal(t, types.RobotIdle, robot1.Status)\n\tassert.True(t, robot1.AutonomousMode)\n\tassert.NotNil(t, robot1.Config, \"Robot config should be parsed\")\n\tassert.NotNil(t, robot1.Config.Identity, \"Identity should be parsed\")\n\tassert.Equal(t, \"Sales Manager\", robot1.Config.Identity.Role)\n\tassert.Equal(t, 3, robot1.Config.Quota.GetMax())\n\n\t// Verify second robot\n\trobot2 := c.Get(\"robot_test_support_002\")\n\tassert.NotNil(t, robot2, \"Support bot should be loaded\")\n\tassert.Equal(t, \"robot_test_support_002\", robot2.MemberID)\n\tassert.Equal(t, \"Test Support Bot\", robot2.DisplayName)\n\tassert.NotNil(t, robot2.Config)\n\tassert.Equal(t, \"Customer Support\", robot2.Config.Identity.Role)\n\tassert.Equal(t, 2, robot2.Config.Quota.GetMax())\n\n\t// Verify inactive robot is not loaded\n\trobot3 := c.Get(\"robot_test_inactive_003\")\n\tassert.Nil(t, robot3, \"Inactive robot should not be loaded\")\n}\n\n// TestCacheLoadByID tests loading a single robot by member ID\nfunc TestCacheLoadByID(t *testing.T) {\n\tif testing.Short() {\n\t\tt.Skip(\"Skipping integration test\")\n\t}\n\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\tcleanupTestRobots(t)\n\tsetupTestRobots(t)\n\tdefer cleanupTestRobots(t)\n\n\tc := cache.New()\n\tctx := types.NewContext(context.Background(), nil)\n\n\tt.Run(\"load existing robot\", func(t *testing.T) {\n\t\trobot, err := c.LoadByID(ctx, \"robot_test_sales_001\")\n\t\tassert.NoError(t, err)\n\t\tassert.NotNil(t, robot)\n\t\tassert.Equal(t, \"robot_test_sales_001\", robot.MemberID)\n\t\tassert.Equal(t, \"Test Sales Bot\", robot.DisplayName)\n\t\tassert.NotNil(t, robot.Config)\n\t})\n\n\tt.Run(\"load non-existent robot\", func(t *testing.T) {\n\t\trobot, err := c.LoadByID(ctx, \"robot_nonexistent\")\n\t\tassert.Error(t, err)\n\t\tassert.Equal(t, types.ErrRobotNotFound, err)\n\t\tassert.Nil(t, robot)\n\t})\n\n\tt.Run(\"load inactive robot by ID\", func(t *testing.T) {\n\t\t// LoadByID doesn't filter by status, so it should load\n\t\trobot, err := c.LoadByID(ctx, \"robot_test_inactive_003\")\n\t\tassert.NoError(t, err)\n\t\tassert.NotNil(t, robot)\n\t\tassert.Equal(t, \"robot_test_inactive_003\", robot.MemberID)\n\t})\n}\n\n// TestCacheRefresh tests refreshing a single robot from database\nfunc TestCacheRefresh(t *testing.T) {\n\tif testing.Short() {\n\t\tt.Skip(\"Skipping integration test\")\n\t}\n\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\tcleanupTestRobots(t)\n\tsetupTestRobots(t)\n\tdefer cleanupTestRobots(t)\n\n\tc := cache.New()\n\tctx := types.NewContext(context.Background(), nil)\n\n\t// Load initial data\n\terr := c.Load(ctx)\n\tassert.NoError(t, err)\n\n\tt.Run(\"refresh existing robot\", func(t *testing.T) {\n\t\terr := c.Refresh(ctx, \"robot_test_sales_001\")\n\t\tassert.NoError(t, err)\n\n\t\t// Robot should still be in cache\n\t\trobot := c.Get(\"robot_test_sales_001\")\n\t\tassert.NotNil(t, robot)\n\t})\n\n\tt.Run(\"refresh removes non-existent robot\", func(t *testing.T) {\n\t\t// Add a fake robot to cache\n\t\tc.Add(&types.Robot{MemberID: \"robot_test_fake\", TeamID: \"team_test_cache_001\"})\n\t\tassert.NotNil(t, c.Get(\"robot_test_fake\"))\n\n\t\t// Refresh should remove it\n\t\terr := c.Refresh(ctx, \"robot_test_fake\")\n\t\tassert.NoError(t, err)\n\t\tassert.Nil(t, c.Get(\"robot_test_fake\"), \"Non-existent robot should be removed\")\n\t})\n}\n\n// TestCacheListByTeam tests listing robots by team\nfunc TestCacheListByTeam(t *testing.T) {\n\tif testing.Short() {\n\t\tt.Skip(\"Skipping integration test\")\n\t}\n\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\tcleanupTestRobots(t)\n\tsetupTestRobots(t)\n\tdefer cleanupTestRobots(t)\n\n\tc := cache.New()\n\tctx := types.NewContext(context.Background(), nil)\n\n\t// Load all robots\n\terr := c.Load(ctx)\n\tassert.NoError(t, err)\n\n\t// List robots by team\n\trobots := c.List(\"team_test_cache_001\")\n\tassert.Len(t, robots, 2, \"Should have 2 robots in team_test_cache_001\")\n\n\t// List robots for non-existent team\n\trobots = c.List(\"team_nonexistent\")\n\tassert.Len(t, robots, 0, \"Non-existent team should have no robots\")\n}\n\n// TestCacheGetByStatus tests getting robots by status\nfunc TestCacheGetByStatus(t *testing.T) {\n\tif testing.Short() {\n\t\tt.Skip(\"Skipping integration test\")\n\t}\n\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\tcleanupTestRobots(t)\n\tsetupTestRobots(t)\n\tdefer cleanupTestRobots(t)\n\n\tc := cache.New()\n\tctx := types.NewContext(context.Background(), nil)\n\n\t// Load all robots\n\terr := c.Load(ctx)\n\tassert.NoError(t, err)\n\n\t// Get idle robots (may have others in DB)\n\tidle := c.GetIdle()\n\tassert.GreaterOrEqual(t, len(idle), 2, \"Should have at least 2 idle robots\")\n\n\t// Verify our test robots are not working\n\ttestRobot1 := c.Get(\"robot_test_sales_001\")\n\ttestRobot2 := c.Get(\"robot_test_support_002\")\n\tassert.Equal(t, types.RobotIdle, testRobot1.Status, \"Test robot 1 should be idle\")\n\tassert.Equal(t, types.RobotIdle, testRobot2.Status, \"Test robot 2 should be idle\")\n}\n\n// TestCacheAutoRefresh tests auto-refresh functionality and goroutine leak prevention\nfunc TestCacheAutoRefresh(t *testing.T) {\n\tif testing.Short() {\n\t\tt.Skip(\"Skipping integration test\")\n\t}\n\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\tcleanupTestRobots(t)\n\tsetupTestRobots(t)\n\tdefer cleanupTestRobots(t)\n\n\t// Verify test data is set up\n\tc := cache.New()\n\tctx := types.NewContext(context.Background(), nil)\n\terr := c.Load(ctx)\n\tassert.NoError(t, err)\n\tassert.GreaterOrEqual(t, c.Count(), 1, \"Should have at least one robot loaded\")\n\n\tt.Run(\"start and stop auto-refresh\", func(t *testing.T) {\n\t\t// Use a fresh cache for this test\n\t\ttestCache := cache.New()\n\t\ttestCtx := types.NewContext(context.Background(), nil)\n\t\terr := testCache.Load(testCtx)\n\t\tassert.NoError(t, err)\n\n\t\t// Start auto-refresh with short interval\n\t\tconfig := &cache.RefreshConfig{Interval: 100 * time.Millisecond}\n\t\ttestCache.StartAutoRefresh(testCtx, config)\n\n\t\t// Wait a bit to let it run (should trigger at least 2 refreshes)\n\t\ttime.Sleep(250 * time.Millisecond)\n\n\t\t// Stop auto-refresh\n\t\ttestCache.StopAutoRefresh()\n\n\t\t// Verify it stopped by checking that no more refreshes happen\n\t\tcountBefore := testCache.Count()\n\t\ttime.Sleep(200 * time.Millisecond)\n\t\tcountAfter := testCache.Count()\n\n\t\t// Count should be stable (no errors from stopped goroutine)\n\t\tassert.Equal(t, countBefore, countAfter, \"Cache should be stable after stop\")\n\t})\n\n\tt.Run(\"multiple start calls should replace previous\", func(t *testing.T) {\n\t\t// Use a fresh cache for this test\n\t\ttestCache := cache.New()\n\t\ttestCtx := types.NewContext(context.Background(), nil)\n\t\terr := testCache.Load(testCtx)\n\t\tassert.NoError(t, err)\n\n\t\t// Track refresh count using a counter\n\t\trefreshCount := 0\n\t\toriginalCount := testCache.Count()\n\n\t\t// Start multiple times without stopping\n\t\tconfig := &cache.RefreshConfig{Interval: 50 * time.Millisecond}\n\n\t\ttestCache.StartAutoRefresh(testCtx, config)\n\t\ttime.Sleep(30 * time.Millisecond)\n\n\t\ttestCache.StartAutoRefresh(testCtx, config) // Should stop previous one\n\t\ttime.Sleep(30 * time.Millisecond)\n\n\t\ttestCache.StartAutoRefresh(testCtx, config) // Should stop previous one\n\n\t\t// Wait for some refreshes\n\t\ttime.Sleep(150 * time.Millisecond)\n\n\t\t// Stop once should be enough\n\t\ttestCache.StopAutoRefresh()\n\n\t\t// Verify cache still works correctly\n\t\tassert.GreaterOrEqual(t, testCache.Count(), 0, \"Cache should still be functional\")\n\n\t\t// Verify we can still access robots\n\t\t_ = refreshCount // suppress unused warning\n\t\t_ = originalCount // suppress unused warning\n\t})\n\n\tt.Run(\"stop without start should not panic\", func(t *testing.T) {\n\t\t// Use a fresh cache for this test\n\t\ttestCache := cache.New()\n\n\t\t// Multiple stops should be safe\n\t\tassert.NotPanics(t, func() {\n\t\t\ttestCache.StopAutoRefresh()\n\t\t\ttestCache.StopAutoRefresh()\n\t\t\ttestCache.StopAutoRefresh()\n\t\t})\n\t})\n\n\tt.Run(\"concurrent start and stop should be safe\", func(t *testing.T) {\n\t\t// Use a fresh cache for this test\n\t\ttestCache := cache.New()\n\t\ttestCtx := types.NewContext(context.Background(), nil)\n\t\terr := testCache.Load(testCtx)\n\t\tassert.NoError(t, err)\n\n\t\tconfig := &cache.RefreshConfig{Interval: 50 * time.Millisecond}\n\n\t\t// Rapidly start and stop multiple times - should not panic or deadlock\n\t\tdone := make(chan bool)\n\t\tgo func() {\n\t\t\tfor i := 0; i < 10; i++ {\n\t\t\t\ttestCache.StartAutoRefresh(testCtx, config)\n\t\t\t\ttime.Sleep(10 * time.Millisecond)\n\t\t\t\ttestCache.StopAutoRefresh()\n\t\t\t\ttime.Sleep(10 * time.Millisecond)\n\t\t\t}\n\t\t\tdone <- true\n\t\t}()\n\n\t\t// Wait with timeout to detect deadlocks\n\t\tselect {\n\t\tcase <-done:\n\t\t\t// Success - no deadlock\n\t\tcase <-time.After(5 * time.Second):\n\t\t\tt.Fatal(\"Rapid start/stop cycles caused deadlock\")\n\t\t}\n\n\t\t// Final cleanup\n\t\ttestCache.StopAutoRefresh()\n\n\t\t// Verify cache is still functional\n\t\tassert.GreaterOrEqual(t, testCache.Count(), 0, \"Cache should still be functional after rapid cycles\")\n\t})\n}\n\n// setupTestRobots creates 3 test robot records in database\nfunc setupTestRobots(t *testing.T) {\n\t// Get the actual table name from model\n\tm := model.Select(\"__yao.member\")\n\ttableName := m.MetaData.Table.Name\n\n\tqb := capsule.Query()\n\n\t// Robot 1: Sales Bot (active, autonomous)\n\trobotConfig1 := map[string]interface{}{\n\t\t\"identity\": map[string]interface{}{\n\t\t\t\"role\": \"Sales Manager\",\n\t\t\t\"duties\": []string{\"Manage leads\", \"Follow up customers\"},\n\t\t\t\"rules\": []string{\"Be professional\", \"Reply within 24h\"},\n\t\t},\n\t\t\"quota\": map[string]interface{}{\n\t\t\t\"max\": 3,\n\t\t\t\"queue\": 15,\n\t\t\t\"priority\": 7,\n\t\t},\n\t\t\"clock\": map[string]interface{}{\n\t\t\t\"mode\": \"times\",\n\t\t\t\"times\": []string{\"09:00\", \"14:00\"},\n\t\t\t\"tz\": \"Asia/Shanghai\",\n\t\t},\n\t}\n\tconfig1JSON, _ := json.Marshal(robotConfig1)\n\n\terr := qb.Table(tableName).Insert([]map[string]interface{}{\n\t\t{\n\t\t\t\"member_id\": \"robot_test_sales_001\",\n\t\t\t\"team_id\": \"team_test_cache_001\",\n\t\t\t\"member_type\": \"robot\",\n\t\t\t\"display_name\": \"Test Sales Bot\",\n\t\t\t\"system_prompt\": \"You are a professional sales manager assistant.\",\n\t\t\t\"status\": \"active\",\n\t\t\t\"role_id\": \"member\", // required field\n\t\t\t\"autonomous_mode\": true,\n\t\t\t\"robot_status\": \"idle\",\n\t\t\t\"robot_config\": string(config1JSON),\n\t\t},\n\t})\n\tif err != nil {\n\t\tt.Fatalf(\"Failed to insert robot_test_sales_001: %v\", err)\n\t}\n\n\t// Robot 2: Support Bot (active, autonomous)\n\trobotConfig2 := map[string]interface{}{\n\t\t\"identity\": map[string]interface{}{\n\t\t\t\"role\": \"Customer Support\",\n\t\t\t\"duties\": []string{\"Answer questions\", \"Resolve issues\"},\n\t\t},\n\t\t\"quota\": map[string]interface{}{\n\t\t\t\"max\": 2,\n\t\t\t\"queue\": 10,\n\t\t\t\"priority\": 5,\n\t\t},\n\t\t\"clock\": map[string]interface{}{\n\t\t\t\"mode\": \"interval\",\n\t\t\t\"every\": \"1h\",\n\t\t},\n\t}\n\tconfig2JSON, _ := json.Marshal(robotConfig2)\n\n\terr = qb.Table(tableName).Insert([]map[string]interface{}{\n\t\t{\n\t\t\t\"member_id\": \"robot_test_support_002\",\n\t\t\t\"team_id\": \"team_test_cache_001\",\n\t\t\t\"member_type\": \"robot\",\n\t\t\t\"display_name\": \"Test Support Bot\",\n\t\t\t\"system_prompt\": \"You are a helpful customer support assistant.\",\n\t\t\t\"status\": \"active\",\n\t\t\t\"role_id\": \"member\", // required field\n\t\t\t\"autonomous_mode\": true,\n\t\t\t\"robot_status\": \"idle\",\n\t\t\t\"robot_config\": string(config2JSON),\n\t\t},\n\t})\n\tif err != nil {\n\t\tt.Fatalf(\"Failed to insert robot_test_support_002: %v\", err)\n\t}\n\n\t// Robot 3: Inactive robot (should not be loaded by Load())\n\terr = qb.Table(tableName).Insert([]map[string]interface{}{\n\t\t{\n\t\t\t\"member_id\": \"robot_test_inactive_003\",\n\t\t\t\"team_id\": \"team_test_cache_001\",\n\t\t\t\"member_type\": \"robot\",\n\t\t\t\"display_name\": \"Test Inactive Bot\",\n\t\t\t\"status\": \"inactive\",\n\t\t\t\"role_id\": \"member\", // required field\n\t\t\t\"autonomous_mode\": true,\n\t\t\t\"robot_status\": \"paused\",\n\t\t},\n\t})\n\tif err != nil {\n\t\tt.Fatalf(\"Failed to insert robot_test_inactive_003: %v\", err)\n\t}\n}\n\n// cleanupTestRobots removes test robot records\nfunc cleanupTestRobots(t *testing.T) {\n\tqb := capsule.Query()\n\n\t// Use the member model to perform soft delete\n\tm := model.Select(\"__yao.member\")\n\n\t// Delete test robots\n\tm.DeleteWhere(model.QueryParam{\n\t\tWheres: []model.QueryWhere{\n\t\t\t{Column: \"member_id\", Value: \"robot_test_sales_001\"},\n\t\t},\n\t})\n\tm.DeleteWhere(model.QueryParam{\n\t\tWheres: []model.QueryWhere{\n\t\t\t{Column: \"member_id\", Value: \"robot_test_support_002\"},\n\t\t},\n\t})\n\tm.DeleteWhere(model.QueryParam{\n\t\tWheres: []model.QueryWhere{\n\t\t\t{Column: \"member_id\", Value: \"robot_test_inactive_003\"},\n\t\t},\n\t})\n\n\t// Hard delete from database (cleanup for next test run)\n\tm2 := model.Select(\"__yao.member\")\n\ttableName2 := m2.MetaData.Table.Name\n\tqb.Table(tableName2).Where(\"member_id\", \"robot_test_sales_001\").Delete()\n\tqb.Table(tableName2).Where(\"member_id\", \"robot_test_support_002\").Delete()\n\tqb.Table(tableName2).Where(\"member_id\", \"robot_test_inactive_003\").Delete()\n}\n" }, { "path": "agent/robot/cache/load.go", "content": "package cache\n\nimport (\n\t\"fmt\"\n\n\t\"github.com/yaoapp/gou/model\"\n\t\"github.com/yaoapp/kun/maps\"\n\t\"github.com/yaoapp/yao/agent/robot/types\"\n)\n\n// memberModel is the model name for member table\n// Can be changed via SetMemberModel() during system initialization\nvar memberModel = \"__yao.member\"\n\n// memberFields are the fields to select when loading robots\nvar memberFields = []interface{}{\n\t\"id\",\n\t\"member_id\",\n\t\"team_id\",\n\t\"display_name\",\n\t\"bio\",\n\t\"system_prompt\",\n\t\"robot_status\",\n\t\"autonomous_mode\",\n\t\"robot_config\",\n\t\"robot_email\",\n\t\"agents\",\n\t\"mcp_servers\",\n\t\"manager_id\",\n\t\"language_model\",\n}\n\n// SetMemberModel sets the member model name\n// Call this during system initialization to override the default\nfunc SetMemberModel(model string) {\n\tif model != \"\" {\n\t\tmemberModel = model\n\t}\n}\n\n// Load loads all active robots from database with pagination\n// Query: member_type='robot' AND autonomous_mode=true AND status='active'\nfunc (c *Cache) Load(ctx *types.Context) error {\n\tm := model.Select(memberModel)\n\n\t// Clear existing cache first\n\tc.mu.Lock()\n\tc.robots = make(map[string]*types.Robot)\n\tc.byTeam = make(map[string][]string)\n\tc.mu.Unlock()\n\n\t// Paginate to handle large number of robots\n\tpage := 1\n\tpageSize := 100 // load 100 robots per page\n\ttotalLoaded := 0\n\n\tfor {\n\t\t// Query with pagination\n\t\tresult, err := m.Paginate(model.QueryParam{\n\t\t\tSelect: memberFields,\n\t\t\tWheres: []model.QueryWhere{\n\t\t\t\t{Column: \"member_type\", Value: \"robot\"},\n\t\t\t\t{Column: \"autonomous_mode\", Value: true},\n\t\t\t\t{Column: \"status\", Value: \"active\"},\n\t\t\t},\n\t\t}, page, pageSize)\n\t\tif err != nil {\n\t\t\treturn fmt.Errorf(\"failed to load robots (page %d): %w\", page, err)\n\t\t}\n\n\t\t// Extract records from pagination result\n\t\tdata, ok := result.Get(\"data\").([]maps.MapStr)\n\t\tif !ok || len(data) == 0 {\n\t\t\tbreak\n\t\t}\n\n\t\t// Parse and add each robot\n\t\tfor _, record := range data {\n\t\t\trobot, err := types.NewRobotFromMap(map[string]interface{}(record))\n\t\t\tif err != nil {\n\t\t\t\t// Log error but continue loading other robots\n\t\t\t\tcontinue\n\t\t\t}\n\t\t\tc.Add(robot)\n\t\t\ttotalLoaded++\n\t\t}\n\n\t\t// Check if there are more pages\n\t\ttotal, _ := result.Get(\"total\").(int)\n\t\tif totalLoaded >= total {\n\t\t\tbreak\n\t\t}\n\n\t\tpage++\n\t}\n\n\treturn nil\n}\n\n// LoadByID loads a single robot from database by member ID\nfunc (c *Cache) LoadByID(ctx *types.Context, memberID string) (*types.Robot, error) {\n\tm := model.Select(memberModel)\n\n\trecords, err := m.Get(model.QueryParam{\n\t\tSelect: memberFields,\n\t\tWheres: []model.QueryWhere{\n\t\t\t{Column: \"member_id\", Value: memberID},\n\t\t\t{Column: \"member_type\", Value: \"robot\"},\n\t\t},\n\t\tLimit: 1,\n\t})\n\tif err != nil {\n\t\treturn nil, fmt.Errorf(\"failed to load robot %s: %w\", memberID, err)\n\t}\n\n\tif len(records) == 0 {\n\t\treturn nil, types.ErrRobotNotFound\n\t}\n\n\treturn types.NewRobotFromMap(map[string]interface{}(records[0]))\n}\n" }, { "path": "agent/robot/cache/refresh.go", "content": "package cache\n\nimport (\n\t\"sync\"\n\t\"time\"\n\n\t\"github.com/yaoapp/yao/agent/robot/types\"\n)\n\n// RefreshConfig holds refresh configuration\ntype RefreshConfig struct {\n\tInterval time.Duration // full refresh interval (default: 1 hour)\n}\n\n// DefaultRefreshConfig returns default refresh configuration\nfunc DefaultRefreshConfig() *RefreshConfig {\n\treturn &RefreshConfig{\n\t\tInterval: time.Hour,\n\t}\n}\n\n// refreshState holds the refresh goroutine state\ntype refreshState struct {\n\tticker *time.Ticker\n\tdone chan struct{}\n\tmu sync.Mutex\n}\n\nvar refresher = &refreshState{}\n\n// Refresh refreshes a single robot's config from database\nfunc (c *Cache) Refresh(ctx *types.Context, memberID string) error {\n\trobot, err := c.LoadByID(ctx, memberID)\n\tif err != nil {\n\t\t// If robot not found or no longer autonomous, remove from cache\n\t\tif err == types.ErrRobotNotFound {\n\t\t\tc.Remove(memberID)\n\t\t\treturn nil\n\t\t}\n\t\treturn err\n\t}\n\n\t// Check if robot is still active and autonomous\n\tif !robot.AutonomousMode {\n\t\tc.Remove(memberID)\n\t\treturn nil\n\t}\n\n\t// Update cache\n\tc.Add(robot)\n\treturn nil\n}\n\n// StartAutoRefresh starts periodic full refresh\nfunc (c *Cache) StartAutoRefresh(ctx *types.Context, config *RefreshConfig) {\n\tif config == nil {\n\t\tconfig = DefaultRefreshConfig()\n\t}\n\n\trefresher.mu.Lock()\n\tdefer refresher.mu.Unlock()\n\n\t// Stop existing refresher if any\n\tif refresher.done != nil {\n\t\tclose(refresher.done)\n\t}\n\n\trefresher.ticker = time.NewTicker(config.Interval)\n\trefresher.done = make(chan struct{})\n\n\tgo func() {\n\t\tfor {\n\t\t\tselect {\n\t\t\tcase <-refresher.done:\n\t\t\t\trefresher.ticker.Stop()\n\t\t\t\treturn\n\t\t\tcase <-refresher.ticker.C:\n\t\t\t\t// Perform full refresh\n\t\t\t\t_ = c.Load(ctx)\n\t\t\t}\n\t\t}\n\t}()\n}\n\n// StopAutoRefresh stops the periodic refresh\nfunc (c *Cache) StopAutoRefresh() {\n\trefresher.mu.Lock()\n\tdefer refresher.mu.Unlock()\n\n\tif refresher.done != nil {\n\t\tclose(refresher.done)\n\t\trefresher.done = nil\n\t}\n}\n\n// RefreshAll reloads all robots from database\nfunc (c *Cache) RefreshAll(ctx *types.Context) error {\n\treturn c.Load(ctx)\n}\n\n// Count returns the number of cached robots\nfunc (c *Cache) Count() int {\n\tc.mu.RLock()\n\tdefer c.mu.RUnlock()\n\treturn len(c.robots)\n}\n\n// ListAll returns all cached robots (across all teams)\nfunc (c *Cache) ListAll() []*types.Robot {\n\tc.mu.RLock()\n\tdefer c.mu.RUnlock()\n\n\trobots := make([]*types.Robot, 0, len(c.robots))\n\tfor _, robot := range c.robots {\n\t\trobots = append(robots, robot)\n\t}\n\treturn robots\n}\n\n// GetByStatus returns robots with the specified status\nfunc (c *Cache) GetByStatus(status types.RobotStatus) []*types.Robot {\n\tc.mu.RLock()\n\tdefer c.mu.RUnlock()\n\n\tvar robots []*types.Robot\n\tfor _, robot := range c.robots {\n\t\tif robot.Status == status {\n\t\t\trobots = append(robots, robot)\n\t\t}\n\t}\n\treturn robots\n}\n\n// GetIdle returns all idle robots ready to execute\nfunc (c *Cache) GetIdle() []*types.Robot {\n\treturn c.GetByStatus(types.RobotIdle)\n}\n\n// GetWorking returns all currently working robots\nfunc (c *Cache) GetWorking() []*types.Robot {\n\treturn c.GetByStatus(types.RobotWorking)\n}\n" }, { "path": "agent/robot/dedup/dedup.go", "content": "package dedup\n\nimport (\n\t\"sync\"\n\t\"time\"\n\n\t\"github.com/yaoapp/yao/agent/robot/types\"\n)\n\n// Dedup implements types.Dedup interface\n// This is a stub implementation for Phase 2\ntype Dedup struct {\n\tmarks map[string]time.Time // key -> expiry time\n\tmu sync.RWMutex\n}\n\n// New creates a new dedup instance\nfunc New() *Dedup {\n\treturn &Dedup{\n\t\tmarks: make(map[string]time.Time),\n\t}\n}\n\n// Check checks if execution should be deduplicated\n// Stub: always returns proceed (will be implemented in Phase 3)\nfunc (d *Dedup) Check(ctx *types.Context, memberID string, trigger types.TriggerType) (types.DedupResult, error) {\n\treturn types.DedupProceed, nil\n}\n\n// Mark marks an execution to prevent duplicates within window\n// Stub: does nothing (will be implemented in Phase 3)\nfunc (d *Dedup) Mark(memberID string, trigger types.TriggerType, window time.Duration) {\n\t// Stub: no-op\n}\n" }, { "path": "agent/robot/events/delivery.go", "content": "package events\n\nimport (\n\t\"bytes\"\n\t\"context\"\n\t\"crypto/hmac\"\n\t\"crypto/sha256\"\n\t\"encoding/hex\"\n\t\"encoding/json\"\n\t\"fmt\"\n\t\"io\"\n\t\"net/http\"\n\t\"strings\"\n\t\"time\"\n\n\t\"github.com/yaoapp/gou/process\"\n\t\"github.com/yaoapp/gou/text\"\n\tagentcontext \"github.com/yaoapp/yao/agent/context\"\n\trobottypes \"github.com/yaoapp/yao/agent/robot/types\"\n\t\"github.com/yaoapp/yao/attachment\"\n\teventtypes \"github.com/yaoapp/yao/event/types\"\n\t\"github.com/yaoapp/yao/messenger\"\n\tmessengerTypes \"github.com/yaoapp/yao/messenger/types\"\n)\n\n// handleDelivery routes delivery content to configured channels (email, webhook, process).\nfunc (h *robotHandler) handleDelivery(ctx context.Context, ev *eventtypes.Event, resp chan<- eventtypes.Result) {\n\tvar payload DeliveryPayload\n\tif err := ev.Should(&payload); err != nil {\n\t\tlog.Error(\"delivery handler: invalid payload: %v\", err)\n\t\tif ev.IsCall {\n\t\t\tresp <- eventtypes.Result{Err: err}\n\t\t}\n\t\treturn\n\t}\n\n\tlog.Info(\"delivery handler: execution=%s member=%s\", payload.ExecutionID, payload.MemberID)\n\n\tcontent := payload.Content\n\tprefs := payload.Preferences\n\tif content == nil {\n\t\tlog.Warn(\"delivery handler: nil content for execution=%s\", payload.ExecutionID)\n\t\tif ev.IsCall {\n\t\t\tresp <- eventtypes.Result{Data: \"no content\"}\n\t\t}\n\t\treturn\n\t}\n\tif prefs == nil {\n\t\tif ev.IsCall {\n\t\t\tresp <- eventtypes.Result{Data: \"no preferences, skipped\"}\n\t\t}\n\t\treturn\n\t}\n\n\tdeliveryCtx := &robottypes.DeliveryContext{\n\t\tMemberID: payload.MemberID,\n\t\tExecutionID: payload.ExecutionID,\n\t\tTeamID: payload.TeamID,\n\t}\n\n\tvar results []robottypes.ChannelResult\n\tvar lastErr error\n\n\tif prefs.Email != nil && prefs.Email.Enabled {\n\t\tfor _, target := range prefs.Email.Targets {\n\t\t\tr := h.sendEmail(ctx, content, target, deliveryCtx)\n\t\t\tresults = append(results, r)\n\t\t\tif !r.Success && lastErr == nil {\n\t\t\t\tlastErr = fmt.Errorf(\"email delivery failed: %s\", r.Error)\n\t\t\t}\n\t\t}\n\t}\n\n\tif prefs.Webhook != nil && prefs.Webhook.Enabled {\n\t\tfor _, target := range prefs.Webhook.Targets {\n\t\t\tr := h.postWebhook(ctx, content, target, deliveryCtx)\n\t\t\tresults = append(results, r)\n\t\t\tif !r.Success && lastErr == nil {\n\t\t\t\tlastErr = fmt.Errorf(\"webhook delivery failed: %s\", r.Error)\n\t\t\t}\n\t\t}\n\t}\n\n\tif prefs.Process != nil && prefs.Process.Enabled {\n\t\tfor _, target := range prefs.Process.Targets {\n\t\t\tr := h.callProcess(ctx, content, target, deliveryCtx)\n\t\t\tresults = append(results, r)\n\t\t\tif !r.Success && lastErr == nil {\n\t\t\t\tlastErr = fmt.Errorf(\"process delivery failed: %s\", r.Error)\n\t\t\t}\n\t\t}\n\t}\n\n\t// Push delivery to integration channels only when the task originated from one\n\tif reply := getReplyFunc(); reply != nil && payload.ChatID != \"\" {\n\t\tchannel, chatID := splitChannelChatID(payload.ChatID)\n\t\tif channel != \"\" && chatID != \"\" {\n\t\t\tmsg := buildDeliveryMessage(content)\n\t\t\tif msg != nil {\n\t\t\t\textra := map[string]any{\n\t\t\t\t\t\"member_id\": payload.MemberID,\n\t\t\t\t\t\"execution_id\": payload.ExecutionID,\n\t\t\t\t}\n\t\t\t\tfor k, v := range payload.Extra {\n\t\t\t\t\textra[k] = v\n\t\t\t\t}\n\t\t\t\tmetadata := &MessageMetadata{\n\t\t\t\t\tChannel: channel,\n\t\t\t\t\tChatID: chatID,\n\t\t\t\t\tExtra: extra,\n\t\t\t\t}\n\t\t\t\tif err := reply(ctx, msg, metadata); err != nil {\n\t\t\t\t\tlog.Error(\"delivery handler: integration reply failed channel=%s execution=%s: %v\", channel, payload.ExecutionID, err)\n\t\t\t\t}\n\t\t\t}\n\t\t}\n\t}\n\n\tif lastErr != nil {\n\t\tlog.Error(\"delivery handler: partial failure execution=%s: %v\", payload.ExecutionID, lastErr)\n\t}\n\n\tif ev.IsCall {\n\t\tresp <- eventtypes.Result{\n\t\t\tData: map[string]interface{}{\n\t\t\t\t\"execution_id\": payload.ExecutionID,\n\t\t\t\t\"results\": results,\n\t\t\t},\n\t\t\tErr: lastErr,\n\t\t}\n\t}\n}\n\n// buildDeliveryMessage converts DeliveryContent into a standard assistant Message.\nfunc buildDeliveryMessage(content *robottypes.DeliveryContent) *agentcontext.Message {\n\tif content == nil {\n\t\treturn nil\n\t}\n\n\tvar parts []interface{}\n\n\ttext := content.Body\n\tif text == \"\" {\n\t\ttext = content.Summary\n\t}\n\tif text != \"\" {\n\t\tparts = append(parts, map[string]interface{}{\n\t\t\t\"type\": \"text\",\n\t\t\t\"text\": text,\n\t\t})\n\t}\n\n\tfor _, att := range content.Attachments {\n\t\tif att.File == \"\" {\n\t\t\tcontinue\n\t\t}\n\t\tpart := map[string]interface{}{\n\t\t\t\"type\": \"file\",\n\t\t\t\"file\": map[string]interface{}{\n\t\t\t\t\"url\": att.File,\n\t\t\t\t\"filename\": att.Title,\n\t\t\t},\n\t\t}\n\t\tparts = append(parts, part)\n\t}\n\n\tif len(parts) == 0 {\n\t\treturn nil\n\t}\n\n\tvar msgContent interface{}\n\tif len(parts) == 1 {\n\t\tif tp, ok := parts[0].(map[string]interface{}); ok && tp[\"type\"] == \"text\" {\n\t\t\tmsgContent = tp[\"text\"]\n\t\t} else {\n\t\t\tmsgContent = parts\n\t\t}\n\t} else {\n\t\tmsgContent = parts\n\t}\n\n\treturn &agentcontext.Message{\n\t\tRole: agentcontext.RoleAssistant,\n\t\tContent: msgContent,\n\t}\n}\n\n// ============================================================================\n// Email\n// ============================================================================\n\nfunc (h *robotHandler) sendEmail(\n\tctx context.Context,\n\tcontent *robottypes.DeliveryContent,\n\ttarget robottypes.EmailTarget,\n\tdeliveryCtx *robottypes.DeliveryContext,\n) robottypes.ChannelResult {\n\tnow := time.Now()\n\ttargetID := strings.Join(target.To, \",\")\n\tif targetID == \"\" {\n\t\ttargetID = \"no-recipients\"\n\t}\n\n\tresult := robottypes.ChannelResult{\n\t\tType: robottypes.DeliveryEmail,\n\t\tTarget: targetID,\n\t\tSentAt: &now,\n\t}\n\n\tsvc := messenger.Instance\n\tif svc == nil {\n\t\tresult.Error = \"messenger service not available\"\n\t\treturn result\n\t}\n\n\thtmlBody, plainBody := buildEmailBody(target.Template, content)\n\tmsg := &messengerTypes.Message{\n\t\tTo: target.To,\n\t\tSubject: buildEmailSubject(target.Subject, target.Template, content, deliveryCtx),\n\t\tBody: plainBody,\n\t\tHTML: htmlBody,\n\t\tType: messengerTypes.MessageTypeEmail,\n\t}\n\n\tattachments := convertAttachments(ctx, content.Attachments)\n\tif len(attachments) > 0 {\n\t\tmsg.Attachments = attachments\n\t}\n\n\tchannel := robottypes.DefaultEmailChannel()\n\tif err := svc.Send(ctx, channel, msg); err != nil {\n\t\tresult.Error = err.Error()\n\t\treturn result\n\t}\n\n\tresult.Success = true\n\tresult.Recipients = target.To\n\treturn result\n}\n\n// ============================================================================\n// Webhook\n// ============================================================================\n\nfunc (h *robotHandler) postWebhook(\n\tctx context.Context,\n\tcontent *robottypes.DeliveryContent,\n\ttarget robottypes.WebhookTarget,\n\tdeliveryCtx *robottypes.DeliveryContext,\n) robottypes.ChannelResult {\n\tnow := time.Now()\n\tresult := robottypes.ChannelResult{\n\t\tType: robottypes.DeliveryWebhook,\n\t\tTarget: target.URL,\n\t\tSentAt: &now,\n\t}\n\n\tpayload := map[string]interface{}{\n\t\t\"event\": \"robot.delivery\",\n\t\t\"timestamp\": now.Format(time.RFC3339),\n\t\t\"execution_id\": deliveryCtx.ExecutionID,\n\t\t\"member_id\": deliveryCtx.MemberID,\n\t\t\"team_id\": deliveryCtx.TeamID,\n\t\t\"trigger_type\": deliveryCtx.TriggerType,\n\t\t\"content\": map[string]interface{}{\n\t\t\t\"summary\": content.Summary,\n\t\t\t\"body\": content.Body,\n\t\t},\n\t}\n\n\tif len(content.Attachments) > 0 {\n\t\tinfo := make([]map[string]interface{}, 0, len(content.Attachments))\n\t\tfor _, att := range content.Attachments {\n\t\t\tinfo = append(info, map[string]interface{}{\n\t\t\t\t\"title\": att.Title,\n\t\t\t\t\"description\": att.Description,\n\t\t\t\t\"task_id\": att.TaskID,\n\t\t\t\t\"file\": att.File,\n\t\t\t})\n\t\t}\n\t\tpayload[\"attachments\"] = info\n\t}\n\n\tpayloadBytes, err := json.Marshal(payload)\n\tif err != nil {\n\t\tresult.Error = fmt.Sprintf(\"failed to marshal payload: %v\", err)\n\t\treturn result\n\t}\n\n\tmethod := target.Method\n\tif method == \"\" {\n\t\tmethod = \"POST\"\n\t}\n\n\treq, err := http.NewRequestWithContext(ctx, method, target.URL, bytes.NewReader(payloadBytes))\n\tif err != nil {\n\t\tresult.Error = fmt.Sprintf(\"failed to create request: %v\", err)\n\t\treturn result\n\t}\n\n\treq.Header.Set(\"Content-Type\", \"application/json\")\n\tfor key, value := range target.Headers {\n\t\treq.Header.Set(key, value)\n\t}\n\n\tif target.Secret != \"\" {\n\t\tsignature := ComputeHMACSignature(payloadBytes, target.Secret)\n\t\treq.Header.Set(\"X-Yao-Signature\", signature)\n\t\treq.Header.Set(\"X-Yao-Signature-Algorithm\", \"HMAC-SHA256\")\n\t}\n\n\thttpResp, err := h.httpClient.Do(req)\n\tif err != nil {\n\t\tresult.Error = fmt.Sprintf(\"request failed: %v\", err)\n\t\treturn result\n\t}\n\tdefer httpResp.Body.Close()\n\n\tbody, _ := io.ReadAll(httpResp.Body)\n\n\tif httpResp.StatusCode < 200 || httpResp.StatusCode >= 300 {\n\t\tresult.Error = fmt.Sprintf(\"webhook returned status %d: %s\", httpResp.StatusCode, string(body))\n\t\treturn result\n\t}\n\n\tresult.Success = true\n\tresult.Details = map[string]interface{}{\n\t\t\"status_code\": httpResp.StatusCode,\n\t\t\"response\": string(body),\n\t}\n\treturn result\n}\n\n// ============================================================================\n// Process\n// ============================================================================\n\nfunc (h *robotHandler) callProcess(\n\tctx context.Context,\n\tcontent *robottypes.DeliveryContent,\n\ttarget robottypes.ProcessTarget,\n\tdeliveryCtx *robottypes.DeliveryContext,\n) robottypes.ChannelResult {\n\tnow := time.Now()\n\tresult := robottypes.ChannelResult{\n\t\tType: robottypes.DeliveryProcess,\n\t\tTarget: target.Process,\n\t\tSentAt: &now,\n\t}\n\n\targs := make([]interface{}, 0, 1+len(target.Args))\n\targs = append(args, map[string]interface{}{\n\t\t\"content\": map[string]interface{}{\n\t\t\t\"summary\": content.Summary,\n\t\t\t\"body\": content.Body,\n\t\t\t\"attachments\": content.Attachments,\n\t\t},\n\t\t\"context\": map[string]interface{}{\n\t\t\t\"execution_id\": deliveryCtx.ExecutionID,\n\t\t\t\"member_id\": deliveryCtx.MemberID,\n\t\t\t\"team_id\": deliveryCtx.TeamID,\n\t\t\t\"trigger_type\": deliveryCtx.TriggerType,\n\t\t},\n\t})\n\targs = append(args, target.Args...)\n\n\tproc, err := process.Of(target.Process, args...)\n\tif err != nil {\n\t\tresult.Error = fmt.Sprintf(\"failed to create process: %v\", err)\n\t\treturn result\n\t}\n\tproc.Context = ctx\n\n\tif err = proc.Execute(); err != nil {\n\t\tresult.Error = err.Error()\n\t\treturn result\n\t}\n\n\tresult.Success = true\n\tresult.Details = toJSONSerializable(proc.Value)\n\treturn result\n}\n\n// ============================================================================\n// Helpers\n// ============================================================================\n\nfunc toJSONSerializable(v interface{}) interface{} {\n\tif v == nil {\n\t\treturn nil\n\t}\n\tif _, err := json.Marshal(v); err != nil {\n\t\treturn fmt.Sprintf(\"%v\", v)\n\t}\n\treturn v\n}\n\nfunc buildEmailSubject(subject, template string, content *robottypes.DeliveryContent, ctx *robottypes.DeliveryContext) string {\n\tif subject != \"\" {\n\t\treturn subject\n\t}\n\tif content.Summary != \"\" {\n\t\treturn content.Summary\n\t}\n\treturn fmt.Sprintf(\"Execution %s Complete\", ctx.ExecutionID)\n}\n\nfunc buildEmailBody(template string, content *robottypes.DeliveryContent) (string, string) {\n\tmarkdown := content.Body\n\tif markdown == \"\" {\n\t\tmarkdown = content.Summary\n\t}\n\thtml, err := text.MarkdownToHTML(markdown)\n\tif err != nil {\n\t\treturn markdown, markdown\n\t}\n\treturn html, markdown\n}\n\nfunc convertAttachments(ctx context.Context, attachments []robottypes.DeliveryAttachment) []messengerTypes.Attachment {\n\tif len(attachments) == 0 {\n\t\treturn nil\n\t}\n\n\tresult := make([]messengerTypes.Attachment, 0, len(attachments))\n\tfor _, att := range attachments {\n\t\tuploader, fileID, isWrapper := attachment.Parse(att.File)\n\t\tif !isWrapper {\n\t\t\tlog.Warn(\"convertAttachments: skipping non-wrapper file value=%q title=%q\", att.File, att.Title)\n\t\t\tcontinue\n\t\t}\n\t\tmanager, ok := attachment.Managers[uploader]\n\t\tif !ok {\n\t\t\tlog.Warn(\"convertAttachments: manager not found uploader=%q file=%q title=%q (available: %v)\",\n\t\t\t\tuploader, att.File, att.Title, attachmentManagerKeys())\n\t\t\tcontinue\n\t\t}\n\t\tinfo, err := manager.Info(ctx, fileID)\n\t\tif err != nil {\n\t\t\tlog.Warn(\"convertAttachments: failed to get file info fileID=%q uploader=%q: %v\", fileID, uploader, err)\n\t\t\tcontinue\n\t\t}\n\t\tcontent, err := manager.Read(ctx, fileID)\n\t\tif err != nil {\n\t\t\tlog.Warn(\"convertAttachments: failed to read file fileID=%q uploader=%q: %v\", fileID, uploader, err)\n\t\t\tcontinue\n\t\t}\n\n\t\tfilename := info.Filename\n\t\tif att.Title != \"\" {\n\t\t\text := \"\"\n\t\t\tif idx := strings.LastIndex(info.Filename, \".\"); idx >= 0 {\n\t\t\t\text = info.Filename[idx:]\n\t\t\t}\n\t\t\ttitleExt := \"\"\n\t\t\tif idx := strings.LastIndex(att.Title, \".\"); idx >= 0 {\n\t\t\t\ttitleExt = att.Title[idx:]\n\t\t\t}\n\t\t\tif titleExt != \"\" {\n\t\t\t\tfilename = att.Title\n\t\t\t} else {\n\t\t\t\tfilename = att.Title + ext\n\t\t\t}\n\t\t}\n\n\t\tlog.Info(\"convertAttachments: added attachment filename=%q contentType=%q size=%d\", filename, info.ContentType, len(content))\n\t\tresult = append(result, messengerTypes.Attachment{\n\t\t\tFilename: filename,\n\t\t\tContentType: info.ContentType,\n\t\t\tContent: content,\n\t\t})\n\t}\n\treturn result\n}\n\nfunc attachmentManagerKeys() []string {\n\tkeys := make([]string, 0, len(attachment.Managers))\n\tfor k := range attachment.Managers {\n\t\tkeys = append(keys, k)\n\t}\n\treturn keys\n}\n\n// ComputeHMACSignature computes HMAC-SHA256 signature for webhook payload.\nfunc ComputeHMACSignature(payload []byte, secret string) string {\n\tmac := hmac.New(sha256.New, []byte(secret))\n\tmac.Write(payload)\n\treturn hex.EncodeToString(mac.Sum(nil))\n}\n\n// VerifyHMACSignature verifies the HMAC-SHA256 signature of a webhook payload.\nfunc VerifyHMACSignature(payload []byte, secret, signature string) bool {\n\texpected := ComputeHMACSignature(payload, secret)\n\treturn hmac.Equal([]byte(expected), []byte(signature))\n}\n\n// splitChannelChatID splits a composite \"channel:chatID\" string (e.g. \"telegram:8134167376\")\n// into its channel and chatID parts. If no colon is present, channel is empty.\nfunc splitChannelChatID(composite string) (channel, chatID string) {\n\tif idx := strings.Index(composite, \":\"); idx >= 0 {\n\t\treturn composite[:idx], composite[idx+1:]\n\t}\n\treturn \"\", composite\n}\n" }, { "path": "agent/robot/events/event_push_test.go", "content": "package events\n\nimport (\n\t\"encoding/json\"\n\t\"testing\"\n\n\t\"github.com/stretchr/testify/assert\"\n\t\"github.com/stretchr/testify/require\"\n\trobottypes \"github.com/yaoapp/yao/agent/robot/types\"\n)\n\n// EP1: ExecPayload with all execution statuses\nfunc TestExecPayloadAllStatuses(t *testing.T) {\n\tstatuses := []string{\n\t\t\"running\", \"completed\", \"failed\", \"cancelled\", \"waiting\", \"confirming\",\n\t}\n\n\tfor _, s := range statuses {\n\t\tpayload := ExecPayload{\n\t\t\tExecutionID: \"exec-ep1\",\n\t\t\tMemberID: \"member-ep1\",\n\t\t\tTeamID: \"team-ep1\",\n\t\t\tStatus: s,\n\t\t}\n\n\t\tdata, err := json.Marshal(payload)\n\t\trequire.NoError(t, err)\n\n\t\tvar parsed ExecPayload\n\t\terr = json.Unmarshal(data, &parsed)\n\t\trequire.NoError(t, err)\n\t\tassert.Equal(t, s, parsed.Status, \"Status %s should round-trip\", s)\n\t}\n}\n\n// EP2: NeedInputPayload with empty question\nfunc TestNeedInputPayloadEmptyQuestion(t *testing.T) {\n\tpayload := NeedInputPayload{\n\t\tExecutionID: \"exec-ep2\",\n\t\tMemberID: \"member-ep2\",\n\t\tTeamID: \"team-ep2\",\n\t\tTaskID: \"task-ep2\",\n\t\tQuestion: \"\",\n\t}\n\n\tdata, err := json.Marshal(payload)\n\trequire.NoError(t, err)\n\n\tvar parsed map[string]interface{}\n\terr = json.Unmarshal(data, &parsed)\n\trequire.NoError(t, err)\n\n\t// Empty string should be present but empty\n\tq, ok := parsed[\"question\"]\n\tassert.True(t, ok)\n\tassert.Equal(t, \"\", q)\n}\n\n// EP3: TaskPayload serializes error correctly\nfunc TestTaskPayloadErrorSerialization(t *testing.T) {\n\tpayload := TaskPayload{\n\t\tExecutionID: \"exec-ep3\",\n\t\tMemberID: \"member-ep3\",\n\t\tTeamID: \"team-ep3\",\n\t\tTaskID: \"task-ep3\",\n\t\tError: \"context deadline exceeded\",\n\t}\n\n\tdata, err := json.Marshal(payload)\n\trequire.NoError(t, err)\n\n\tvar parsed map[string]interface{}\n\terr = json.Unmarshal(data, &parsed)\n\trequire.NoError(t, err)\n\tassert.Equal(t, \"context deadline exceeded\", parsed[\"error\"])\n}\n\n// EP4: DeliveryPayload with nested content\nfunc TestDeliveryPayloadNestedContent(t *testing.T) {\n\tpayload := DeliveryPayload{\n\t\tExecutionID: \"exec-ep4\",\n\t\tMemberID: \"member-ep4\",\n\t\tTeamID: \"team-ep4\",\n\t\tContent: &robottypes.DeliveryContent{\n\t\t\tSummary: \"Daily Summary\",\n\t\t\tBody: \"Full body with sections: intro, body, conclusion\",\n\t\t},\n\t}\n\n\tdata, err := json.Marshal(payload)\n\trequire.NoError(t, err)\n\n\tvar parsed DeliveryPayload\n\terr = json.Unmarshal(data, &parsed)\n\trequire.NoError(t, err)\n\n\trequire.NotNil(t, parsed.Content)\n\tassert.Equal(t, \"Daily Summary\", parsed.Content.Summary)\n\tassert.Contains(t, parsed.Content.Body, \"sections\")\n}\n\n// EP5: Event constants follow naming convention\nfunc TestEventConstantNamingConvention(t *testing.T) {\n\tallEvents := []string{\n\t\tTaskNeedInput, TaskFailed, TaskCompleted,\n\t\tExecWaiting, ExecResumed, ExecCompleted, ExecFailed, ExecCancelled,\n\t\tDelivery,\n\t}\n\n\tfor _, e := range allEvents {\n\t\tassert.Contains(t, e, \"robot.\", \"Event %q should start with 'robot.'\", e)\n\t}\n}\n\n// EP6: ExecPayload omits empty ChatID\nfunc TestExecPayloadOmitsEmptyOptionalFields(t *testing.T) {\n\tpayload := ExecPayload{\n\t\tExecutionID: \"exec-ep6\",\n\t\tMemberID: \"member-ep6\",\n\t\tTeamID: \"team-ep6\",\n\t\tStatus: \"completed\",\n\t}\n\n\tdata, err := json.Marshal(payload)\n\trequire.NoError(t, err)\n\n\tvar parsed map[string]interface{}\n\terr = json.Unmarshal(data, &parsed)\n\trequire.NoError(t, err)\n\n\t_, hasChatID := parsed[\"chat_id\"]\n\tif hasChatID {\n\t\tassert.Equal(t, \"\", parsed[\"chat_id\"])\n\t}\n}\n\n// EP7: All payloads share common fields (ExecutionID, MemberID, TeamID)\nfunc TestPayloadCommonFields(t *testing.T) {\n\tneedInput := NeedInputPayload{ExecutionID: \"e1\", MemberID: \"m1\", TeamID: \"t1\"}\n\ttask := TaskPayload{ExecutionID: \"e2\", MemberID: \"m2\", TeamID: \"t2\"}\n\texec := ExecPayload{ExecutionID: \"e3\", MemberID: \"m3\", TeamID: \"t3\"}\n\tdelivery := DeliveryPayload{ExecutionID: \"e4\", MemberID: \"m4\", TeamID: \"t4\"}\n\n\tassert.Equal(t, \"e1\", needInput.ExecutionID)\n\tassert.Equal(t, \"m2\", task.MemberID)\n\tassert.Equal(t, \"t3\", exec.TeamID)\n\tassert.Equal(t, \"e4\", delivery.ExecutionID)\n}\n" }, { "path": "agent/robot/events/events.go", "content": "package events\n\nimport (\n\t\"context\"\n\t\"strings\"\n\t\"sync\"\n\n\tagentcontext \"github.com/yaoapp/yao/agent/context\"\n\trobottypes \"github.com/yaoapp/yao/agent/robot/types\"\n)\n\n// TriggerFunc is the callback for triggering a robot execution.\n// Injected by the api package at startup to break the import cycle.\n// Returns (executionID, accepted, error).\ntype TriggerFunc func(ctx *robottypes.Context, memberID string, triggerType robottypes.TriggerType, data interface{}) (string, bool, error)\n\n// ReplyFunc is the callback for replying to the originating channel.\n// Injected by the dispatcher at startup. The implementation routes the\n// reply to the correct adapter based on metadata.Channel.\n// msg is the standard assistant message (text, Content Parts, media, etc.).\ntype ReplyFunc func(ctx context.Context, msg *agentcontext.Message, metadata *MessageMetadata) error\n\nvar (\n\ttriggerFn TriggerFunc\n\ttriggerFnMu sync.RWMutex\n\n\treplyFn ReplyFunc\n\treplyFnMu sync.RWMutex\n)\n\n// RegisterTriggerFunc sets the function used by handleMessage to trigger\n// robot execution when a confirmed action is detected.\nfunc RegisterTriggerFunc(fn TriggerFunc) {\n\ttriggerFnMu.Lock()\n\tdefer triggerFnMu.Unlock()\n\ttriggerFn = fn\n}\n\nfunc getTriggerFunc() TriggerFunc {\n\ttriggerFnMu.RLock()\n\tdefer triggerFnMu.RUnlock()\n\treturn triggerFn\n}\n\n// RegisterReplyFunc sets the function used by handleMessage to reply\n// to the originating channel after processing.\nfunc RegisterReplyFunc(fn ReplyFunc) {\n\treplyFnMu.Lock()\n\tdefer replyFnMu.Unlock()\n\treplyFn = fn\n}\n\nfunc getReplyFunc() ReplyFunc {\n\treplyFnMu.RLock()\n\tdefer replyFnMu.RUnlock()\n\treturn replyFn\n}\n\n// Robot event type constants for event.Push integration.\n// Events are fire-and-forget; handlers are registered via event.Register().\nconst (\n\tTaskNeedInput = \"robot.task.need_input\"\n\tTaskFailed = \"robot.task.failed\"\n\tTaskCompleted = \"robot.task.completed\"\n\tExecWaiting = \"robot.exec.waiting\"\n\tExecResumed = \"robot.exec.resumed\"\n\tExecCompleted = \"robot.exec.completed\"\n\tExecFailed = \"robot.exec.failed\"\n\tExecCancelled = \"robot.exec.cancelled\"\n\tDelivery = \"robot.delivery\"\n\tMessage = \"robot.message\"\n)\n\n// Robot configuration change events (used by integrations Receiver).\nconst (\n\tRobotConfigCreated = \"robot.config.created\"\n\tRobotConfigUpdated = \"robot.config.updated\"\n\tRobotConfigDeleted = \"robot.config.deleted\"\n)\n\n// NeedInputPayload is the event payload for TaskNeedInput / ExecWaiting events.\ntype NeedInputPayload struct {\n\tExecutionID string `json:\"execution_id\"`\n\tMemberID string `json:\"member_id\"`\n\tTeamID string `json:\"team_id\"`\n\tTaskID string `json:\"task_id\"`\n\tQuestion string `json:\"question\"`\n\tChatID string `json:\"chat_id,omitempty\"`\n}\n\n// ExecPayload is a generic execution event payload.\ntype ExecPayload struct {\n\tExecutionID string `json:\"execution_id\"`\n\tMemberID string `json:\"member_id\"`\n\tTeamID string `json:\"team_id\"`\n\tStatus string `json:\"status,omitempty\"`\n\tError string `json:\"error,omitempty\"`\n\tChatID string `json:\"chat_id,omitempty\"`\n}\n\n// TaskPayload is the event payload for TaskFailed / TaskCompleted events.\ntype TaskPayload struct {\n\tExecutionID string `json:\"execution_id\"`\n\tMemberID string `json:\"member_id\"`\n\tTeamID string `json:\"team_id\"`\n\tTaskID string `json:\"task_id\"`\n\tError string `json:\"error,omitempty\"`\n\tChatID string `json:\"chat_id,omitempty\"`\n}\n\n// DeliveryPayload is the event payload for Delivery events.\ntype DeliveryPayload struct {\n\tExecutionID string `json:\"execution_id\"`\n\tMemberID string `json:\"member_id\"`\n\tTeamID string `json:\"team_id\"`\n\tChatID string `json:\"chat_id,omitempty\"`\n\tContent *robottypes.DeliveryContent `json:\"content,omitempty\"`\n\tPreferences *robottypes.DeliveryPreferences `json:\"preferences,omitempty\"`\n\tExtra map[string]any `json:\"extra,omitempty\"`\n}\n\n// MessagePayload is the event payload for Message events (external channel messages).\ntype MessagePayload struct {\n\tRobotID string `json:\"robot_id\"`\n\tMessages []agentcontext.Message `json:\"messages\"`\n\tMetadata *MessageMetadata `json:\"metadata\"`\n}\n\n// MessageMetadata carries channel-specific information for routing and deduplication.\ntype MessageMetadata struct {\n\tChannel string `json:\"channel\"`\n\tMessageID string `json:\"message_id,omitempty\"`\n\tAppID string `json:\"app_id,omitempty\"`\n\tChatID string `json:\"chat_id,omitempty\"`\n\tSenderID string `json:\"sender_id,omitempty\"`\n\tSenderName string `json:\"sender_name,omitempty\"`\n\tLocale string `json:\"locale,omitempty\"`\n\tReplyTo string `json:\"reply_to,omitempty\"`\n\tExtra map[string]any `json:\"extra,omitempty\"`\n}\n\n// MessageResult is the result returned from handleMessage via event.Call.\ntype MessageResult struct {\n\tMessage *agentcontext.Message `json:\"message,omitempty\"`\n\tAction *ActionResult `json:\"action,omitempty\"`\n\tExecutionID string `json:\"execution_id,omitempty\"`\n\tMetadata *MessageMetadata `json:\"metadata,omitempty\"`\n}\n\n// ActionResult describes a detected action from the Host Agent's Next hook.\ntype ActionResult struct {\n\tName string `json:\"name\"`\n\tPayload any `json:\"payload,omitempty\"`\n}\n\n// RobotConfigPayload is the event payload for robot.config.* events.\ntype RobotConfigPayload struct {\n\tMemberID string `json:\"member_id\"`\n\tTeamID string `json:\"team_id\"`\n}\n\n// NormalizeLocale converts various language code formats (IETF BCP 47, etc.)\n// into the lowercase hyphenated form used by agentcontext (e.g. \"zh-cn\", \"en-us\").\n//\n// Mapping rules:\n//\n//\t\"zh-hans\", \"zh-cn\" → \"zh-cn\"\n//\t\"zh-hant\", \"zh-tw\", \"zh-hk\" → \"zh-tw\"\n//\t\"zh\" → \"zh-cn\"\n//\t\"en-us\" → \"en-us\"\n//\t\"en-gb\" → \"en-gb\"\n//\t\"en\" → \"en\"\n//\t\"\" → \"en\" (default)\n//\tother → lowercased as-is\nfunc NormalizeLocale(raw string) string {\n\tcode := strings.ToLower(strings.TrimSpace(raw))\n\tif code == \"\" {\n\t\treturn \"en\"\n\t}\n\n\t// Normalize underscore to hyphen (e.g. zh_CN → zh-cn)\n\tcode = strings.ReplaceAll(code, \"_\", \"-\")\n\n\tswitch code {\n\tcase \"zh-hans\", \"zh-cn\":\n\t\treturn \"zh-cn\"\n\tcase \"zh-hant\", \"zh-tw\", \"zh-hk\":\n\t\treturn \"zh-tw\"\n\tcase \"zh\":\n\t\treturn \"zh-cn\"\n\tdefault:\n\t\treturn code\n\t}\n}\n" }, { "path": "agent/robot/events/events_test.go", "content": "package events\n\nimport (\n\t\"encoding/json\"\n\t\"testing\"\n\n\t\"github.com/stretchr/testify/assert\"\n\t\"github.com/stretchr/testify/require\"\n\trobottypes \"github.com/yaoapp/yao/agent/robot/types\"\n)\n\nfunc TestEventConstants(t *testing.T) {\n\texpected := map[string]string{\n\t\t\"TaskNeedInput\": \"robot.task.need_input\",\n\t\t\"TaskFailed\": \"robot.task.failed\",\n\t\t\"TaskCompleted\": \"robot.task.completed\",\n\t\t\"ExecWaiting\": \"robot.exec.waiting\",\n\t\t\"ExecResumed\": \"robot.exec.resumed\",\n\t\t\"ExecCompleted\": \"robot.exec.completed\",\n\t\t\"ExecFailed\": \"robot.exec.failed\",\n\t\t\"ExecCancelled\": \"robot.exec.cancelled\",\n\t\t\"Delivery\": \"robot.delivery\",\n\t}\n\n\tactual := map[string]string{\n\t\t\"TaskNeedInput\": TaskNeedInput,\n\t\t\"TaskFailed\": TaskFailed,\n\t\t\"TaskCompleted\": TaskCompleted,\n\t\t\"ExecWaiting\": ExecWaiting,\n\t\t\"ExecResumed\": ExecResumed,\n\t\t\"ExecCompleted\": ExecCompleted,\n\t\t\"ExecFailed\": ExecFailed,\n\t\t\"ExecCancelled\": ExecCancelled,\n\t\t\"Delivery\": Delivery,\n\t}\n\n\tfor name, exp := range expected {\n\t\tassert.Equal(t, exp, actual[name], \"Event constant %s mismatch\", name)\n\t}\n\tassert.Len(t, actual, 9, \"Expected exactly 9 event constants\")\n}\n\nfunc TestNeedInputPayloadMarshalling(t *testing.T) {\n\tpayload := NeedInputPayload{\n\t\tExecutionID: \"exec-123\",\n\t\tMemberID: \"member-1\",\n\t\tTeamID: \"team-1\",\n\t\tTaskID: \"task-5\",\n\t\tQuestion: \"What date range?\",\n\t\tChatID: \"chat-abc\",\n\t}\n\n\tdata, err := json.Marshal(payload)\n\trequire.NoError(t, err)\n\n\tvar parsed NeedInputPayload\n\terr = json.Unmarshal(data, &parsed)\n\trequire.NoError(t, err)\n\tassert.Equal(t, payload, parsed)\n}\n\nfunc TestTaskPayloadMarshalling(t *testing.T) {\n\tt.Run(\"with error\", func(t *testing.T) {\n\t\tpayload := TaskPayload{\n\t\t\tExecutionID: \"exec-1\",\n\t\t\tMemberID: \"member-1\",\n\t\t\tTeamID: \"team-1\",\n\t\t\tTaskID: \"task-1\",\n\t\t\tError: \"timeout\",\n\t\t\tChatID: \"chat-1\",\n\t\t}\n\n\t\tdata, err := json.Marshal(payload)\n\t\trequire.NoError(t, err)\n\n\t\tvar parsed TaskPayload\n\t\terr = json.Unmarshal(data, &parsed)\n\t\trequire.NoError(t, err)\n\t\tassert.Equal(t, payload, parsed)\n\t})\n\n\tt.Run(\"without error\", func(t *testing.T) {\n\t\tpayload := TaskPayload{\n\t\t\tExecutionID: \"exec-2\",\n\t\t\tMemberID: \"member-2\",\n\t\t\tTeamID: \"team-2\",\n\t\t\tTaskID: \"task-2\",\n\t\t}\n\n\t\tdata, err := json.Marshal(payload)\n\t\trequire.NoError(t, err)\n\n\t\tvar parsed TaskPayload\n\t\terr = json.Unmarshal(data, &parsed)\n\t\trequire.NoError(t, err)\n\t\tassert.Equal(t, payload, parsed)\n\t\tassert.Empty(t, parsed.Error)\n\t})\n}\n\nfunc TestExecPayloadMarshalling(t *testing.T) {\n\tpayload := ExecPayload{\n\t\tExecutionID: \"exec-100\",\n\t\tMemberID: \"member-10\",\n\t\tTeamID: \"team-10\",\n\t\tStatus: \"completed\",\n\t\tChatID: \"chat-100\",\n\t}\n\n\tdata, err := json.Marshal(payload)\n\trequire.NoError(t, err)\n\n\tvar parsed ExecPayload\n\terr = json.Unmarshal(data, &parsed)\n\trequire.NoError(t, err)\n\tassert.Equal(t, payload, parsed)\n}\n\nfunc TestDeliveryPayloadMarshalling(t *testing.T) {\n\tpayload := DeliveryPayload{\n\t\tExecutionID: \"exec-d1\",\n\t\tMemberID: \"member-d1\",\n\t\tTeamID: \"team-d1\",\n\t\tChatID: \"chat-d1\",\n\t\tContent: &robottypes.DeliveryContent{\n\t\t\tSummary: \"done\",\n\t\t\tBody: \"full report\",\n\t\t},\n\t\tPreferences: &robottypes.DeliveryPreferences{\n\t\t\tEmail: &robottypes.EmailPreference{\n\t\t\t\tEnabled: true,\n\t\t\t\tTargets: []robottypes.EmailTarget{{To: []string{\"a@b.com\"}}},\n\t\t\t},\n\t\t},\n\t}\n\n\tdata, err := json.Marshal(payload)\n\trequire.NoError(t, err)\n\n\tvar parsed DeliveryPayload\n\terr = json.Unmarshal(data, &parsed)\n\trequire.NoError(t, err)\n\tassert.Equal(t, \"exec-d1\", parsed.ExecutionID)\n\tassert.Equal(t, \"member-d1\", parsed.MemberID)\n\tassert.NotNil(t, parsed.Content)\n\tassert.Equal(t, \"done\", parsed.Content.Summary)\n\tassert.NotNil(t, parsed.Preferences)\n\tassert.NotNil(t, parsed.Preferences.Email)\n}\n" }, { "path": "agent/robot/events/handlers.go", "content": "package events\n\nimport (\n\t\"context\"\n\t\"net/http\"\n\t\"time\"\n\n\t\"github.com/yaoapp/yao/event\"\n\teventtypes \"github.com/yaoapp/yao/event/types\"\n)\n\nfunc init() {\n\tevent.Register(\"robot\", &robotHandler{\n\t\thttpClient: &http.Client{Timeout: 30 * time.Second},\n\t})\n}\n\n// robotHandler processes all robot.* events.\ntype robotHandler struct {\n\thttpClient *http.Client\n}\n\n// Handle dispatches robot events by type.\nfunc (h *robotHandler) Handle(ctx context.Context, ev *eventtypes.Event, resp chan<- eventtypes.Result) {\n\tswitch ev.Type {\n\tcase Delivery:\n\t\th.handleDelivery(ctx, ev, resp)\n\tcase Message:\n\t\th.handleMessage(ctx, ev, resp)\n\tdefault:\n\t\tlog.Debug(\"robot handler: unhandled event type=%s id=%s\", ev.Type, ev.ID)\n\t}\n}\n\n// Shutdown gracefully shuts down the robot handler.\nfunc (h *robotHandler) Shutdown(ctx context.Context) error {\n\treturn nil\n}\n" }, { "path": "agent/robot/events/handlers_test.go", "content": "package events\n\nimport (\n\t\"context\"\n\t\"encoding/json\"\n\t\"net/http\"\n\t\"net/http/httptest\"\n\t\"testing\"\n\n\t\"github.com/stretchr/testify/assert\"\n\t\"github.com/stretchr/testify/require\"\n\trobottypes \"github.com/yaoapp/yao/agent/robot/types\"\n\teventtypes \"github.com/yaoapp/yao/event/types\"\n)\n\nfunc newTestHandler() *robotHandler {\n\treturn &robotHandler{\n\t\thttpClient: http.DefaultClient,\n\t}\n}\n\nfunc TestRobotHandler_DeliveryWebhook(t *testing.T) {\n\tvar received map[string]interface{}\n\tserver := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {\n\t\tdecoder := json.NewDecoder(r.Body)\n\t\t_ = decoder.Decode(&received)\n\t\tw.WriteHeader(http.StatusOK)\n\t\tw.Write([]byte(`{\"ok\":true}`))\n\t}))\n\tdefer server.Close()\n\n\thandler := newTestHandler()\n\tev := &eventtypes.Event{\n\t\tType: Delivery,\n\t\tID: \"test-ev-1\",\n\t\tIsCall: true,\n\t\tPayload: DeliveryPayload{\n\t\t\tExecutionID: \"exec-1\",\n\t\t\tMemberID: \"member-1\",\n\t\t\tTeamID: \"team-1\",\n\t\t\tContent: &robottypes.DeliveryContent{\n\t\t\t\tSummary: \"test summary\",\n\t\t\t\tBody: \"test body\",\n\t\t\t},\n\t\t\tPreferences: &robottypes.DeliveryPreferences{\n\t\t\t\tWebhook: &robottypes.WebhookPreference{\n\t\t\t\t\tEnabled: true,\n\t\t\t\t\tTargets: []robottypes.WebhookTarget{\n\t\t\t\t\t\t{URL: server.URL},\n\t\t\t\t\t},\n\t\t\t\t},\n\t\t\t},\n\t\t},\n\t}\n\n\tresp := make(chan eventtypes.Result, 1)\n\thandler.Handle(context.Background(), ev, resp)\n\n\tresult := <-resp\n\trequire.NotNil(t, result.Data)\n\tassert.NoError(t, result.Err)\n\n\tdata, ok := result.Data.(map[string]interface{})\n\trequire.True(t, ok)\n\tassert.Equal(t, \"exec-1\", data[\"execution_id\"])\n\n\trequire.NotNil(t, received)\n\tassert.Equal(t, \"robot.delivery\", received[\"event\"])\n}\n\nfunc TestRobotHandler_DeliveryNoContent(t *testing.T) {\n\thandler := newTestHandler()\n\tev := &eventtypes.Event{\n\t\tType: Delivery,\n\t\tID: \"test-ev-2\",\n\t\tIsCall: true,\n\t\tPayload: DeliveryPayload{\n\t\t\tExecutionID: \"exec-2\",\n\t\t\tMemberID: \"member-2\",\n\t\t\tTeamID: \"team-2\",\n\t\t},\n\t}\n\n\tresp := make(chan eventtypes.Result, 1)\n\thandler.Handle(context.Background(), ev, resp)\n\n\tresult := <-resp\n\tassert.Equal(t, \"no content\", result.Data)\n}\n\nfunc TestRobotHandler_DeliveryNoPreferences(t *testing.T) {\n\thandler := newTestHandler()\n\tev := &eventtypes.Event{\n\t\tType: Delivery,\n\t\tID: \"test-ev-3\",\n\t\tIsCall: true,\n\t\tPayload: DeliveryPayload{\n\t\t\tExecutionID: \"exec-3\",\n\t\t\tMemberID: \"member-3\",\n\t\t\tTeamID: \"team-3\",\n\t\t\tContent: &robottypes.DeliveryContent{\n\t\t\t\tSummary: \"test\",\n\t\t\t\tBody: \"body\",\n\t\t\t},\n\t\t},\n\t}\n\n\tresp := make(chan eventtypes.Result, 1)\n\thandler.Handle(context.Background(), ev, resp)\n\n\tresult := <-resp\n\tassert.Equal(t, \"no preferences, skipped\", result.Data)\n}\n\nfunc TestRobotHandler_InvalidPayload(t *testing.T) {\n\thandler := newTestHandler()\n\tev := &eventtypes.Event{\n\t\tType: Delivery,\n\t\tID: \"test-ev-4\",\n\t\tIsCall: true,\n\t\tPayload: \"invalid\",\n\t}\n\n\tresp := make(chan eventtypes.Result, 1)\n\thandler.Handle(context.Background(), ev, resp)\n\n\tresult := <-resp\n\tassert.Error(t, result.Err)\n}\n\nfunc TestRobotHandler_UnhandledEvent(t *testing.T) {\n\thandler := newTestHandler()\n\tev := &eventtypes.Event{\n\t\tType: \"robot.unknown\",\n\t\tID: \"test-ev-5\",\n\t}\n\n\tresp := make(chan eventtypes.Result, 1)\n\thandler.Handle(context.Background(), ev, resp)\n\t// Fire-and-forget, no response expected\n}\n\nfunc TestRobotHandler_Shutdown(t *testing.T) {\n\thandler := newTestHandler()\n\terr := handler.Shutdown(context.Background())\n\tassert.NoError(t, err)\n}\n\nfunc TestVerifyHMACSignature(t *testing.T) {\n\tpayload := []byte(`{\"event\":\"robot.delivery\"}`)\n\tsecret := \"test-secret\"\n\n\tsig := ComputeHMACSignature(payload, secret)\n\tassert.True(t, VerifyHMACSignature(payload, secret, sig))\n\tassert.False(t, VerifyHMACSignature(payload, \"wrong-secret\", sig))\n}\n" }, { "path": "agent/robot/events/integrations/dingtalk/dedup.go", "content": "package dingtalk\n\nimport (\n\t\"sync\"\n\t\"time\"\n)\n\nconst (\n\tdedupTTL = 24 * time.Hour\n\tdedupCleanInterval = time.Hour\n)\n\ntype dedupStore struct {\n\tm sync.Map\n}\n\nfunc newDedupStore() *dedupStore {\n\treturn &dedupStore{}\n}\n\nfunc (d *dedupStore) markSeen(key string) bool {\n\tnow := time.Now().Unix()\n\t_, loaded := d.m.LoadOrStore(key, now)\n\treturn !loaded\n}\n\nfunc (d *dedupStore) cleaner(stopCh <-chan struct{}) {\n\tticker := time.NewTicker(dedupCleanInterval)\n\tdefer ticker.Stop()\n\tfor {\n\t\tselect {\n\t\tcase <-stopCh:\n\t\t\treturn\n\t\tcase <-ticker.C:\n\t\t\tcutoff := time.Now().Add(-dedupTTL).Unix()\n\t\t\td.m.Range(func(key, value any) bool {\n\t\t\t\tif ts, ok := value.(int64); ok && ts < cutoff {\n\t\t\t\t\td.m.Delete(key)\n\t\t\t\t}\n\t\t\t\treturn true\n\t\t\t})\n\t\t}\n\t}\n}\n" }, { "path": "agent/robot/events/integrations/dingtalk/dingtalk.go", "content": "package dingtalk\n\nimport (\n\t\"context\"\n\t\"sync\"\n\n\t\"github.com/yaoapp/yao/agent/robot/logger\"\n\trobottypes \"github.com/yaoapp/yao/agent/robot/types\"\n\tdtapi \"github.com/yaoapp/yao/integrations/dingtalk\"\n)\n\nvar log = logger.New(\"dingtalk\")\n\n// Adapter implements the integrations.Adapter interface for DingTalk.\n//\n// Architecture:\n// - One DingTalk Stream client per registered bot for real-time message reception\n// - One dedup cleaner goroutine removes expired keys every hour\ntype Adapter struct {\n\tmu sync.RWMutex\n\tbots map[string]*botEntry // robotID -> *botEntry\n\tappIdx map[string]string // clientID -> robotID\n\tdedup *dedupStore\n\tstopCh chan struct{}\n}\n\n// botEntry holds the state for one robot's DingTalk integration.\ntype botEntry struct {\n\trobotID string\n\tclientID string\n\tbot *dtapi.Bot\n\tcancelFn context.CancelFunc\n}\n\n// NewAdapter creates a new DingTalk adapter.\nfunc NewAdapter() *Adapter {\n\ta := &Adapter{\n\t\tbots: make(map[string]*botEntry),\n\t\tappIdx: make(map[string]string),\n\t\tdedup: newDedupStore(),\n\t\tstopCh: make(chan struct{}),\n\t}\n\tgo a.dedup.cleaner(a.stopCh)\n\treturn a\n}\n\n// Apply is called by the Dispatcher when a robot config is created or updated.\nfunc (a *Adapter) Apply(ctx context.Context, robot *robottypes.Robot) {\n\tdtConf := extractConfig(robot)\n\tlog.Debug(\"Apply robot=%s dtConf=%v\", robot.MemberID, dtConf != nil)\n\n\tif dtConf == nil || !dtConf.Enabled || dtConf.ClientID == \"\" || dtConf.ClientSecret == \"\" {\n\t\ta.removeBot(robot.MemberID)\n\t\treturn\n\t}\n\n\ta.mu.Lock()\n\tdefer a.mu.Unlock()\n\n\tif existing, ok := a.bots[robot.MemberID]; ok {\n\t\tif existing.clientID == dtConf.ClientID {\n\t\t\treturn\n\t\t}\n\t\ta.removeBotLocked(robot.MemberID)\n\t}\n\n\tbot := dtapi.NewBot(dtConf.ClientID, dtConf.ClientSecret)\n\n\tstreamCtx, streamCancel := context.WithCancel(context.Background())\n\tentry := &botEntry{\n\t\trobotID: robot.MemberID,\n\t\tclientID: dtConf.ClientID,\n\t\tbot: bot,\n\t\tcancelFn: streamCancel,\n\t}\n\ta.bots[robot.MemberID] = entry\n\ta.appIdx[dtConf.ClientID] = robot.MemberID\n\n\tgo a.streamLoop(streamCtx, entry)\n\n\tlog.Info(\"dingtalk adapter: registered robot=%s client=%s\", robot.MemberID, dtConf.ClientID)\n}\n\n// Remove is called by the Dispatcher when a robot is deleted.\nfunc (a *Adapter) Remove(ctx context.Context, robotID string) {\n\ta.removeBot(robotID)\n}\n\n// Shutdown stops all stream connections and dedup cleaner.\nfunc (a *Adapter) Shutdown() {\n\tclose(a.stopCh)\n\ta.mu.Lock()\n\tfor _, entry := range a.bots {\n\t\tif entry.cancelFn != nil {\n\t\t\tentry.cancelFn()\n\t\t}\n\t}\n\ta.mu.Unlock()\n\tlog.Info(\"dingtalk adapter: shutdown complete\")\n}\n\nfunc (a *Adapter) removeBot(robotID string) {\n\ta.mu.Lock()\n\tdefer a.mu.Unlock()\n\ta.removeBotLocked(robotID)\n}\n\nfunc (a *Adapter) removeBotLocked(robotID string) {\n\tentry, ok := a.bots[robotID]\n\tif !ok {\n\t\treturn\n\t}\n\tif entry.cancelFn != nil {\n\t\tentry.cancelFn()\n\t}\n\tif entry.clientID != \"\" {\n\t\tdelete(a.appIdx, entry.clientID)\n\t}\n\tdelete(a.bots, robotID)\n\tlog.Info(\"dingtalk adapter: unregistered robot=%s\", robotID)\n}\n\nfunc (a *Adapter) resolveByClientID(clientID string) (*botEntry, bool) {\n\ta.mu.RLock()\n\tdefer a.mu.RUnlock()\n\trobotID, ok := a.appIdx[clientID]\n\tif !ok {\n\t\treturn nil, false\n\t}\n\tentry, ok := a.bots[robotID]\n\treturn entry, ok\n}\n\nfunc extractConfig(robot *robottypes.Robot) *robottypes.DingTalkConfig {\n\tif robot.Config == nil || robot.Config.Integrations == nil {\n\t\treturn nil\n\t}\n\treturn robot.Config.Integrations.DingTalk\n}\n" }, { "path": "agent/robot/events/integrations/dingtalk/e2e_test.go", "content": "package dingtalk\n\nimport (\n\t\"context\"\n\t\"os\"\n\t\"testing\"\n\t\"time\"\n\n\t\"github.com/stretchr/testify/assert\"\n\t\"github.com/stretchr/testify/require\"\n\trobottypes \"github.com/yaoapp/yao/agent/robot/types\"\n\tdtapi \"github.com/yaoapp/yao/integrations/dingtalk\"\n)\n\nvar (\n\tdtClientID string\n\tdtClientSecret string\n)\n\nfunc TestMain(m *testing.M) {\n\tdtClientID = os.Getenv(\"DINGTALK_TEST_CLIENT_ID\")\n\tdtClientSecret = os.Getenv(\"DINGTALK_TEST_CLIENT_SECRET\")\n\tos.Exit(m.Run())\n}\n\nfunc skipIfNoCreds(t *testing.T) {\n\tt.Helper()\n\tif dtClientID == \"\" || dtClientSecret == \"\" {\n\t\tt.Skip(\"DINGTALK_TEST_CLIENT_ID or DINGTALK_TEST_CLIENT_SECRET not set\")\n\t}\n}\n\n// TestE2E_Adapter_Apply verifies that Apply correctly registers a bot.\nfunc TestE2E_Adapter_Apply(t *testing.T) {\n\tskipIfNoCreds(t)\n\n\ta := &Adapter{\n\t\tbots: make(map[string]*botEntry),\n\t\tappIdx: make(map[string]string),\n\t\tdedup: newDedupStore(),\n\t\tstopCh: make(chan struct{}),\n\t}\n\tdefer close(a.stopCh)\n\n\trobot := &robottypes.Robot{\n\t\tMemberID: \"robot_e2e_dt_adapter\",\n\t\tTeamID: \"team_e2e_dt\",\n\t\tConfig: &robottypes.Config{\n\t\t\tIntegrations: &robottypes.Integrations{\n\t\t\t\tDingTalk: &robottypes.DingTalkConfig{\n\t\t\t\t\tEnabled: true,\n\t\t\t\t\tClientID: dtClientID,\n\t\t\t\t\tClientSecret: dtClientSecret,\n\t\t\t\t},\n\t\t\t},\n\t\t},\n\t}\n\n\ta.Apply(context.Background(), robot)\n\n\ta.mu.RLock()\n\tentry, ok := a.bots[\"robot_e2e_dt_adapter\"]\n\ta.mu.RUnlock()\n\n\trequire.True(t, ok, \"bot should be registered\")\n\tassert.Equal(t, dtClientID, entry.clientID)\n\tassert.NotNil(t, entry.bot)\n\n\tt.Logf(\"OK Apply: dingtalk bot registered robot=%s client=%s\", robot.MemberID, entry.clientID)\n}\n\n// TestE2E_Adapter_Apply_Update verifies re-Apply with same clientID is a no-op.\nfunc TestE2E_Adapter_Apply_Update(t *testing.T) {\n\tskipIfNoCreds(t)\n\n\ta := &Adapter{\n\t\tbots: make(map[string]*botEntry),\n\t\tappIdx: make(map[string]string),\n\t\tdedup: newDedupStore(),\n\t\tstopCh: make(chan struct{}),\n\t}\n\tdefer close(a.stopCh)\n\n\trobot := &robottypes.Robot{\n\t\tMemberID: \"robot_e2e_dt_update\",\n\t\tTeamID: \"team_e2e_dt\",\n\t\tConfig: &robottypes.Config{\n\t\t\tIntegrations: &robottypes.Integrations{\n\t\t\t\tDingTalk: &robottypes.DingTalkConfig{\n\t\t\t\t\tEnabled: true,\n\t\t\t\t\tClientID: dtClientID,\n\t\t\t\t\tClientSecret: dtClientSecret,\n\t\t\t\t},\n\t\t\t},\n\t\t},\n\t}\n\n\ta.Apply(context.Background(), robot)\n\ta.mu.RLock()\n\t_, ok := a.bots[\"robot_e2e_dt_update\"]\n\ta.mu.RUnlock()\n\trequire.True(t, ok)\n\n\ta.Apply(context.Background(), robot)\n\ta.mu.RLock()\n\tassert.Len(t, a.bots, 1)\n\ta.mu.RUnlock()\n\n\ta.Remove(context.Background(), \"robot_e2e_dt_update\")\n\ta.mu.RLock()\n\t_, ok = a.bots[\"robot_e2e_dt_update\"]\n\ta.mu.RUnlock()\n\tassert.False(t, ok, \"bot should be removed\")\n\tt.Log(\"OK Apply/Remove lifecycle verified\")\n}\n\n// TestE2E_Adapter_Dedup verifies deduplication works.\nfunc TestE2E_Adapter_Dedup(t *testing.T) {\n\ta := &Adapter{\n\t\tbots: make(map[string]*botEntry),\n\t\tappIdx: make(map[string]string),\n\t\tdedup: newDedupStore(),\n\t\tstopCh: make(chan struct{}),\n\t}\n\tdefer close(a.stopCh)\n\n\tkey := \"dt:test-robot:msg-12345\"\n\tassert.True(t, a.dedup.markSeen(key), \"first time should return true\")\n\tassert.False(t, a.dedup.markSeen(key), \"second time should return false (dedup)\")\n\tt.Log(\"OK dedup working correctly\")\n}\n\n// TestE2E_Adapter_HandleMessages verifies message handling through the adapter.\nfunc TestE2E_Adapter_HandleMessages(t *testing.T) {\n\tskipIfNoCreds(t)\n\n\ta := &Adapter{\n\t\tbots: make(map[string]*botEntry),\n\t\tappIdx: make(map[string]string),\n\t\tdedup: newDedupStore(),\n\t\tstopCh: make(chan struct{}),\n\t}\n\tdefer close(a.stopCh)\n\n\tentry := &botEntry{\n\t\trobotID: \"robot_e2e_dt_handle\",\n\t\tclientID: dtClientID,\n\t\tbot: dtapi.NewBot(dtClientID, dtClientSecret),\n\t}\n\n\tctx, cancel := context.WithTimeout(context.Background(), 10*time.Second)\n\tdefer cancel()\n\n\tcms := []*dtapi.ConvertedMessage{\n\t\t{\n\t\t\tMessageID: \"test_msg_1\",\n\t\t\tConversationID: \"test_conv_1\",\n\t\t\tConversationType: \"1\",\n\t\t\tSenderID: \"test_sender_1\",\n\t\t\tSenderNick: \"Test User\",\n\t\t\tText: \"Hello from E2E test\",\n\t\t\tSessionWebhook: \"https://oapi.dingtalk.com/robot/sendBySession/xxx\",\n\t\t},\n\t}\n\n\ta.handleMessages(ctx, entry, cms)\n\n\tassert.False(t, a.dedup.markSeen(\"dt:robot_e2e_dt_handle:test_msg_1\"),\n\t\t\"message should be marked as seen after handleMessages\")\n\tt.Log(\"OK handleMessages processed 1 message\")\n}\n\n// TestE2E_Adapter_ApplyDisabled verifies Apply removes bot when disabled.\nfunc TestE2E_Adapter_ApplyDisabled(t *testing.T) {\n\ta := &Adapter{\n\t\tbots: make(map[string]*botEntry),\n\t\tappIdx: make(map[string]string),\n\t\tdedup: newDedupStore(),\n\t\tstopCh: make(chan struct{}),\n\t}\n\tdefer close(a.stopCh)\n\n\trobot := &robottypes.Robot{\n\t\tMemberID: \"robot_e2e_dt_disabled\",\n\t\tTeamID: \"team_e2e_dt\",\n\t\tConfig: &robottypes.Config{\n\t\t\tIntegrations: &robottypes.Integrations{\n\t\t\t\tDingTalk: &robottypes.DingTalkConfig{\n\t\t\t\t\tEnabled: false,\n\t\t\t\t\tClientID: \"some_id\",\n\t\t\t\t\tClientSecret: \"some_secret\",\n\t\t\t\t},\n\t\t\t},\n\t\t},\n\t}\n\n\ta.Apply(context.Background(), robot)\n\ta.mu.RLock()\n\t_, ok := a.bots[\"robot_e2e_dt_disabled\"]\n\ta.mu.RUnlock()\n\tassert.False(t, ok, \"disabled bot should not be registered\")\n\tt.Log(\"OK disabled config not registered\")\n}\n\n// TestE2E_Adapter_GetAccessToken verifies real DingTalk credentials work.\nfunc TestE2E_Adapter_GetAccessToken(t *testing.T) {\n\tskipIfNoCreds(t)\n\n\tb := dtapi.NewBot(dtClientID, dtClientSecret)\n\tctx, cancel := context.WithTimeout(context.Background(), 15*time.Second)\n\tdefer cancel()\n\n\ttoken, err := b.GetAccessToken(ctx)\n\trequire.NoError(t, err)\n\tassert.NotEmpty(t, token)\n\tt.Logf(\"OK DingTalk access token obtained, len=%d\", len(token))\n}\n" }, { "path": "agent/robot/events/integrations/dingtalk/message.go", "content": "package dingtalk\n\nimport (\n\t\"context\"\n\t\"fmt\"\n\t\"strings\"\n\n\tagentcontext \"github.com/yaoapp/yao/agent/context\"\n\tevents \"github.com/yaoapp/yao/agent/robot/events\"\n\t\"github.com/yaoapp/yao/event\"\n\tdtapi \"github.com/yaoapp/yao/integrations/dingtalk\"\n)\n\n// handleMessages processes a batch of DingTalk messages.\nfunc (a *Adapter) handleMessages(ctx context.Context, entry *botEntry, cms []*dtapi.ConvertedMessage) {\n\tif len(cms) == 0 {\n\t\treturn\n\t}\n\n\tvar allParts []interface{}\n\tvar lastCM *dtapi.ConvertedMessage\n\n\tfor _, cm := range cms {\n\t\tif cm == nil {\n\t\t\tcontinue\n\t\t}\n\n\t\tdedupKey := fmt.Sprintf(\"dt:%s:%s\", entry.robotID, cm.MessageID)\n\t\tif !a.dedup.markSeen(dedupKey) {\n\t\t\tcontinue\n\t\t}\n\n\t\tparts := buildContentParts(cm)\n\t\tif len(parts) == 0 {\n\t\t\tcontinue\n\t\t}\n\n\t\tallParts = append(allParts, parts...)\n\t\tlastCM = cm\n\t}\n\n\tif len(allParts) == 0 || lastCM == nil {\n\t\treturn\n\t}\n\n\tcontent := mergeContentParts(allParts)\n\n\tmsgPayload := events.MessagePayload{\n\t\tRobotID: entry.robotID,\n\t\tMessages: []agentcontext.Message{\n\t\t\t{Role: agentcontext.RoleUser, Content: content},\n\t\t},\n\t\tMetadata: &events.MessageMetadata{\n\t\t\tChannel: \"dingtalk\",\n\t\t\tMessageID: lastCM.MessageID,\n\t\t\tAppID: entry.clientID,\n\t\t\tChatID: lastCM.ConversationID,\n\t\t\tSenderID: lastCM.SenderID,\n\t\t\tSenderName: lastCM.SenderNick,\n\t\t\tLocale: \"zh-cn\",\n\t\t\tExtra: map[string]any{\n\t\t\t\t\"session_webhook\": lastCM.SessionWebhook,\n\t\t\t\t\"conversation_type\": lastCM.ConversationType,\n\t\t\t\t\"dt_message_id\": lastCM.MessageID,\n\t\t\t},\n\t\t},\n\t}\n\n\tif _, err := event.Push(ctx, events.Message, msgPayload); err != nil {\n\t\tlog.Error(\"dingtalk adapter: event.Push robot.message failed robot=%s: %v\", entry.robotID, err)\n\t}\n}\n\nfunc buildContentParts(cm *dtapi.ConvertedMessage) []interface{} {\n\tvar parts []interface{}\n\n\tif cm.HasText() {\n\t\tparts = append(parts, map[string]interface{}{\n\t\t\t\"type\": \"text\",\n\t\t\t\"text\": cm.Text,\n\t\t})\n\t}\n\n\tfor _, mi := range cm.MediaItems {\n\t\tif mi.Wrapper == \"\" && mi.URL == \"\" {\n\t\t\tcontinue\n\t\t}\n\t\turl := mi.Wrapper\n\t\tif url == \"\" {\n\t\t\turl = mi.URL\n\t\t}\n\t\tparts = append(parts, map[string]interface{}{\n\t\t\t\"type\": \"file\",\n\t\t\t\"file_url\": url,\n\t\t\t\"mime_type\": mi.MimeType,\n\t\t\t\"file_name\": mi.FileName,\n\t\t})\n\t}\n\n\treturn parts\n}\n\nfunc mergeContentParts(parts []interface{}) interface{} {\n\tallText := true\n\tfor _, p := range parts {\n\t\tm, ok := p.(map[string]interface{})\n\t\tif !ok || m[\"type\"] != \"text\" {\n\t\t\tallText = false\n\t\t\tbreak\n\t\t}\n\t}\n\n\tif allText {\n\t\tvar buf strings.Builder\n\t\tfor i, p := range parts {\n\t\t\tif i > 0 {\n\t\t\t\tbuf.WriteString(\"\\n\")\n\t\t\t}\n\t\t\tm := p.(map[string]interface{})\n\t\t\tbuf.WriteString(m[\"text\"].(string))\n\t\t}\n\t\treturn buf.String()\n\t}\n\n\treturn parts\n}\n" }, { "path": "agent/robot/events/integrations/dingtalk/reply.go", "content": "package dingtalk\n\nimport (\n\t\"context\"\n\t\"fmt\"\n\t\"strings\"\n\n\tagentcontext \"github.com/yaoapp/yao/agent/context\"\n\tevents \"github.com/yaoapp/yao/agent/robot/events\"\n\tdtapi \"github.com/yaoapp/yao/integrations/dingtalk\"\n)\n\n// Reply sends the assistant message back to the originating DingTalk conversation.\nfunc (a *Adapter) Reply(ctx context.Context, msg *agentcontext.Message, metadata *events.MessageMetadata) error {\n\tif msg == nil || metadata == nil {\n\t\treturn fmt.Errorf(\"nil message or metadata\")\n\t}\n\n\tvar sessionWebhook string\n\tif metadata.Extra != nil {\n\t\tif v, ok := metadata.Extra[\"session_webhook\"]; ok {\n\t\t\tif s, ok := v.(string); ok {\n\t\t\t\tsessionWebhook = s\n\t\t\t}\n\t\t}\n\t}\n\n\tif sessionWebhook == \"\" {\n\t\treturn fmt.Errorf(\"no session_webhook in metadata for dingtalk reply\")\n\t}\n\n\treturn sendContent(ctx, sessionWebhook, msg.Content)\n}\n\nfunc sendContent(ctx context.Context, sessionWebhook string, content interface{}) error {\n\tswitch c := content.(type) {\n\tcase string:\n\t\tif strings.TrimSpace(c) == \"\" {\n\t\t\treturn nil\n\t\t}\n\t\treturn dtapi.SendMarkdownMessage(ctx, sessionWebhook, \"Reply\", dtapi.FormatDingTalkMarkdown(c))\n\n\tcase []interface{}:\n\t\treturn sendParts(ctx, sessionWebhook, c)\n\n\tdefault:\n\t\tparts, ok := toContentParts(content)\n\t\tif ok {\n\t\t\treturn sendPartsTyped(ctx, sessionWebhook, parts)\n\t\t}\n\t\treturn dtapi.SendTextMessage(ctx, sessionWebhook, fmt.Sprintf(\"%v\", content))\n\t}\n}\n\nfunc sendParts(ctx context.Context, sessionWebhook string, parts []interface{}) error {\n\tvar textBuf strings.Builder\n\tfor _, part := range parts {\n\t\tm, ok := part.(map[string]interface{})\n\t\tif !ok {\n\t\t\tcontinue\n\t\t}\n\t\tpartType, _ := m[\"type\"].(string)\n\t\tswitch partType {\n\t\tcase \"text\":\n\t\t\tif text, ok := m[\"text\"].(string); ok {\n\t\t\t\ttextBuf.WriteString(text)\n\t\t\t}\n\t\tcase \"image_url\":\n\t\t\tif err := flushText(ctx, sessionWebhook, &textBuf); err != nil {\n\t\t\t\treturn err\n\t\t\t}\n\t\t\tif imgMap, ok := m[\"image_url\"].(map[string]interface{}); ok {\n\t\t\t\tif url, ok := imgMap[\"url\"].(string); ok {\n\t\t\t\t\tif strings.HasPrefix(url, \"http\") {\n\t\t\t\t\t\ttextBuf.WriteString(fmt.Sprintf(\"\\n![image](%s)\\n\", url))\n\t\t\t\t\t}\n\t\t\t\t}\n\t\t\t}\n\t\tcase \"file\":\n\t\t\tif err := flushText(ctx, sessionWebhook, &textBuf); err != nil {\n\t\t\t\treturn err\n\t\t\t}\n\t\t\tfileURL, _ := m[\"file_url\"].(string)\n\t\t\tfileName, _ := m[\"file_name\"].(string)\n\t\t\tif fileURL == \"\" {\n\t\t\t\tif fileMap, ok := m[\"file\"].(map[string]interface{}); ok {\n\t\t\t\t\tfileURL, _ = fileMap[\"url\"].(string)\n\t\t\t\t\tif fn, ok := fileMap[\"filename\"].(string); ok && fn != \"\" {\n\t\t\t\t\t\tfileName = fn\n\t\t\t\t\t}\n\t\t\t\t}\n\t\t\t}\n\t\t\tif fileURL != \"\" && strings.HasPrefix(fileURL, \"http\") {\n\t\t\t\tlabel := fileName\n\t\t\t\tif label == \"\" {\n\t\t\t\t\tlabel = \"file\"\n\t\t\t\t}\n\t\t\t\ttextBuf.WriteString(fmt.Sprintf(\"\\n[%s](%s)\\n\", label, fileURL))\n\t\t\t}\n\t\t}\n\t}\n\treturn flushText(ctx, sessionWebhook, &textBuf)\n}\n\nfunc sendPartsTyped(ctx context.Context, sessionWebhook string, parts []agentcontext.ContentPart) error {\n\tvar textBuf strings.Builder\n\tfor _, part := range parts {\n\t\tswitch part.Type {\n\t\tcase agentcontext.ContentText:\n\t\t\ttextBuf.WriteString(part.Text)\n\t\tcase agentcontext.ContentImageURL:\n\t\t\tif err := flushText(ctx, sessionWebhook, &textBuf); err != nil {\n\t\t\t\treturn err\n\t\t\t}\n\t\t\tif part.ImageURL != nil && strings.HasPrefix(part.ImageURL.URL, \"http\") {\n\t\t\t\ttextBuf.WriteString(fmt.Sprintf(\"\\n![image](%s)\\n\", part.ImageURL.URL))\n\t\t\t}\n\t\tcase agentcontext.ContentFile:\n\t\t\tif err := flushText(ctx, sessionWebhook, &textBuf); err != nil {\n\t\t\t\treturn err\n\t\t\t}\n\t\t\tif part.File != nil && part.File.URL != \"\" && strings.HasPrefix(part.File.URL, \"http\") {\n\t\t\t\tlabel := part.File.Filename\n\t\t\t\tif label == \"\" {\n\t\t\t\t\tlabel = \"file\"\n\t\t\t\t}\n\t\t\t\ttextBuf.WriteString(fmt.Sprintf(\"\\n[%s](%s)\\n\", label, part.File.URL))\n\t\t\t}\n\t\t}\n\t}\n\treturn flushText(ctx, sessionWebhook, &textBuf)\n}\n\nfunc flushText(ctx context.Context, sessionWebhook string, buf *strings.Builder) error {\n\tif buf.Len() == 0 {\n\t\treturn nil\n\t}\n\ttext := buf.String()\n\tbuf.Reset()\n\treturn dtapi.SendMarkdownMessage(ctx, sessionWebhook, \"Reply\", dtapi.FormatDingTalkMarkdown(text))\n}\n\nfunc toContentParts(content interface{}) ([]agentcontext.ContentPart, bool) {\n\tparts, ok := content.([]agentcontext.ContentPart)\n\treturn parts, ok\n}\n" }, { "path": "agent/robot/events/integrations/dingtalk/stream.go", "content": "package dingtalk\n\nimport (\n\t\"context\"\n\t\"strings\"\n\t\"time\"\n\n\tdingstream \"github.com/open-dingtalk/dingtalk-stream-sdk-go/chatbot\"\n\tdingclient \"github.com/open-dingtalk/dingtalk-stream-sdk-go/client\"\n\tdtapi \"github.com/yaoapp/yao/integrations/dingtalk\"\n)\n\nconst reconnectDelay = 5 * time.Second\n\n// streamLoop starts the DingTalk Stream client for a single bot.\n// It automatically reconnects on failure.\nfunc (a *Adapter) streamLoop(ctx context.Context, entry *botEntry) {\n\tlog.Info(\"dingtalk streamLoop started robot=%s client=%s\", entry.robotID, entry.clientID)\n\n\tfor {\n\t\tselect {\n\t\tcase <-ctx.Done():\n\t\t\tlog.Info(\"dingtalk streamLoop stopped robot=%s\", entry.robotID)\n\t\t\treturn\n\t\tcase <-a.stopCh:\n\t\t\treturn\n\t\tdefault:\n\t\t}\n\n\t\terr := a.runStreamClient(ctx, entry)\n\t\tif err != nil {\n\t\t\tlog.Error(\"dingtalk stream disconnected robot=%s: %v, reconnecting in %s\", entry.robotID, err, reconnectDelay)\n\t\t}\n\n\t\tselect {\n\t\tcase <-ctx.Done():\n\t\t\treturn\n\t\tcase <-a.stopCh:\n\t\t\treturn\n\t\tcase <-time.After(reconnectDelay):\n\t\t}\n\t}\n}\n\nfunc (a *Adapter) runStreamClient(ctx context.Context, entry *botEntry) error {\n\tcli := dingclient.NewStreamClient(\n\t\tdingclient.WithAppCredential(dingclient.NewAppCredentialConfig(entry.clientID, entry.bot.ClientSecret())),\n\t)\n\n\tcli.RegisterChatBotCallbackRouter(func(c context.Context, data *dingstream.BotCallbackDataModel) ([]byte, error) {\n\t\treturn a.onBotCallback(c, entry, data)\n\t})\n\n\terrCh := make(chan error, 1)\n\tgo func() {\n\t\terrCh <- cli.Start(ctx)\n\t}()\n\n\tselect {\n\tcase <-ctx.Done():\n\t\treturn ctx.Err()\n\tcase <-a.stopCh:\n\t\treturn nil\n\tcase err := <-errCh:\n\t\treturn err\n\t}\n}\n\nfunc (a *Adapter) onBotCallback(ctx context.Context, entry *botEntry, data *dingstream.BotCallbackDataModel) ([]byte, error) {\n\tif data == nil {\n\t\treturn nil, nil\n\t}\n\n\tcm := &dtapi.ConvertedMessage{\n\t\tMessageID: data.MsgId,\n\t\tConversationID: data.ConversationId,\n\t\tConversationType: data.ConversationType,\n\t\tSenderID: data.SenderId,\n\t\tSenderNick: data.SenderNick,\n\t\tSenderStaffID: data.SenderStaffId,\n\t\tChatbotUserID: data.ChatbotUserId,\n\t\tIsInAtList: data.IsInAtList,\n\t\tSessionWebhook: data.SessionWebhook,\n\t}\n\n\tswitch data.Msgtype {\n\tcase \"text\":\n\t\tcm.Text = strings.TrimSpace(data.Text.Content)\n\t}\n\n\tif cm.HasMedia() {\n\t\tgroups := []string{\"dingtalk\", entry.robotID}\n\t\tdtapi.ResolveMedia(ctx, cm, groups)\n\t}\n\n\ta.handleMessages(ctx, entry, []*dtapi.ConvertedMessage{cm})\n\treturn nil, nil\n}\n" }, { "path": "agent/robot/events/integrations/discord/dedup.go", "content": "package discord\n\nimport (\n\t\"sync\"\n\t\"time\"\n)\n\nconst (\n\tdedupTTL = 24 * time.Hour\n\tdedupCleanInterval = time.Hour\n)\n\ntype dedupStore struct {\n\tm sync.Map\n}\n\nfunc newDedupStore() *dedupStore {\n\treturn &dedupStore{}\n}\n\nfunc (d *dedupStore) markSeen(key string) bool {\n\tnow := time.Now().Unix()\n\t_, loaded := d.m.LoadOrStore(key, now)\n\treturn !loaded\n}\n\nfunc (d *dedupStore) cleaner(stopCh <-chan struct{}) {\n\tticker := time.NewTicker(dedupCleanInterval)\n\tdefer ticker.Stop()\n\tfor {\n\t\tselect {\n\t\tcase <-stopCh:\n\t\t\treturn\n\t\tcase <-ticker.C:\n\t\t\tcutoff := time.Now().Add(-dedupTTL).Unix()\n\t\t\td.m.Range(func(key, value any) bool {\n\t\t\t\tif ts, ok := value.(int64); ok && ts < cutoff {\n\t\t\t\t\td.m.Delete(key)\n\t\t\t\t}\n\t\t\t\treturn true\n\t\t\t})\n\t\t}\n\t}\n}\n" }, { "path": "agent/robot/events/integrations/discord/discord.go", "content": "package discord\n\nimport (\n\t\"context\"\n\t\"sync\"\n\n\t\"github.com/yaoapp/yao/agent/robot/logger\"\n\trobottypes \"github.com/yaoapp/yao/agent/robot/types\"\n\tdcapi \"github.com/yaoapp/yao/integrations/discord\"\n)\n\nvar log = logger.New(\"discord\")\n\n// Adapter implements the integrations.Adapter interface for Discord.\n//\n// Architecture:\n// - One WebSocket Gateway connection per registered bot via discordgo\n// - One dedup cleaner goroutine removes expired keys every hour\ntype Adapter struct {\n\tmu sync.RWMutex\n\tbots map[string]*botEntry // robotID -> *botEntry\n\tappIdx map[string]string // appID -> robotID\n\tdedup *dedupStore\n\tstopCh chan struct{}\n}\n\n// botEntry holds the state for one robot's Discord integration.\ntype botEntry struct {\n\trobotID string\n\tappID string\n\tbot *dcapi.Bot\n\tcancelFn context.CancelFunc\n}\n\n// NewAdapter creates a new Discord adapter.\nfunc NewAdapter() *Adapter {\n\ta := &Adapter{\n\t\tbots: make(map[string]*botEntry),\n\t\tappIdx: make(map[string]string),\n\t\tdedup: newDedupStore(),\n\t\tstopCh: make(chan struct{}),\n\t}\n\tgo a.dedup.cleaner(a.stopCh)\n\treturn a\n}\n\n// Apply is called by the Dispatcher when a robot config is created or updated.\nfunc (a *Adapter) Apply(ctx context.Context, robot *robottypes.Robot) {\n\tdcConf := extractConfig(robot)\n\tlog.Debug(\"Apply robot=%s dcConf=%v\", robot.MemberID, dcConf != nil)\n\n\tif dcConf == nil || !dcConf.Enabled || dcConf.BotToken == \"\" {\n\t\ta.removeBot(robot.MemberID)\n\t\treturn\n\t}\n\n\ta.mu.Lock()\n\tdefer a.mu.Unlock()\n\n\tif existing, ok := a.bots[robot.MemberID]; ok {\n\t\tif existing.bot.Token() == dcConf.BotToken {\n\t\t\treturn\n\t\t}\n\t\ta.removeBotLocked(robot.MemberID)\n\t}\n\n\tbot, err := dcapi.NewBot(dcConf.BotToken, dcConf.AppID)\n\tif err != nil {\n\t\tlog.Error(\"discord adapter: create bot failed robot=%s: %v\", robot.MemberID, err)\n\t\treturn\n\t}\n\n\tgwCtx, gwCancel := context.WithCancel(context.Background())\n\tentry := &botEntry{\n\t\trobotID: robot.MemberID,\n\t\tappID: dcConf.AppID,\n\t\tbot: bot,\n\t\tcancelFn: gwCancel,\n\t}\n\ta.bots[robot.MemberID] = entry\n\tif dcConf.AppID != \"\" {\n\t\ta.appIdx[dcConf.AppID] = robot.MemberID\n\t}\n\n\tgo a.gatewayLoop(gwCtx, entry)\n\n\tlog.Info(\"discord adapter: registered robot=%s app=%s\", robot.MemberID, dcConf.AppID)\n}\n\n// Remove is called by the Dispatcher when a robot is deleted.\nfunc (a *Adapter) Remove(ctx context.Context, robotID string) {\n\ta.removeBot(robotID)\n}\n\n// Shutdown stops all gateway connections and dedup cleaner.\nfunc (a *Adapter) Shutdown() {\n\tclose(a.stopCh)\n\ta.mu.Lock()\n\tfor _, entry := range a.bots {\n\t\tif entry.cancelFn != nil {\n\t\t\tentry.cancelFn()\n\t\t}\n\t\tif entry.bot != nil && entry.bot.Session() != nil {\n\t\t\tentry.bot.Session().Close()\n\t\t}\n\t}\n\ta.mu.Unlock()\n\tlog.Info(\"discord adapter: shutdown complete\")\n}\n\nfunc (a *Adapter) removeBot(robotID string) {\n\ta.mu.Lock()\n\tdefer a.mu.Unlock()\n\ta.removeBotLocked(robotID)\n}\n\nfunc (a *Adapter) removeBotLocked(robotID string) {\n\tentry, ok := a.bots[robotID]\n\tif !ok {\n\t\treturn\n\t}\n\tif entry.cancelFn != nil {\n\t\tentry.cancelFn()\n\t}\n\tif entry.bot != nil && entry.bot.Session() != nil {\n\t\tentry.bot.Session().Close()\n\t}\n\tif entry.appID != \"\" {\n\t\tdelete(a.appIdx, entry.appID)\n\t}\n\tdelete(a.bots, robotID)\n\tlog.Info(\"discord adapter: unregistered robot=%s\", robotID)\n}\n\nfunc (a *Adapter) resolveByAppID(appID string) (*botEntry, bool) {\n\ta.mu.RLock()\n\tdefer a.mu.RUnlock()\n\trobotID, ok := a.appIdx[appID]\n\tif !ok {\n\t\treturn nil, false\n\t}\n\tentry, ok := a.bots[robotID]\n\treturn entry, ok\n}\n\nfunc extractConfig(robot *robottypes.Robot) *robottypes.DiscordConfig {\n\tif robot.Config == nil || robot.Config.Integrations == nil {\n\t\treturn nil\n\t}\n\treturn robot.Config.Integrations.Discord\n}\n" }, { "path": "agent/robot/events/integrations/discord/e2e_test.go", "content": "package discord\n\nimport (\n\t\"context\"\n\t\"os\"\n\t\"testing\"\n\t\"time\"\n\n\t\"github.com/stretchr/testify/assert\"\n\t\"github.com/stretchr/testify/require\"\n\trobottypes \"github.com/yaoapp/yao/agent/robot/types\"\n\tdcapi \"github.com/yaoapp/yao/integrations/discord\"\n)\n\nvar (\n\tdcBotToken string\n\tdcAppID string\n)\n\nfunc TestMain(m *testing.M) {\n\tdcBotToken = os.Getenv(\"DISCORD_TEST_BOT_TOKEN\")\n\tdcAppID = os.Getenv(\"DISCORD_TEST_APP_ID\")\n\tos.Exit(m.Run())\n}\n\nfunc skipIfNoToken(t *testing.T) {\n\tt.Helper()\n\tif dcBotToken == \"\" {\n\t\tt.Skip(\"DISCORD_TEST_BOT_TOKEN not set\")\n\t}\n}\n\n// TestE2E_Adapter_Apply verifies that Apply correctly registers a bot.\nfunc TestE2E_Adapter_Apply(t *testing.T) {\n\tskipIfNoToken(t)\n\n\ta := &Adapter{\n\t\tbots: make(map[string]*botEntry),\n\t\tappIdx: make(map[string]string),\n\t\tdedup: newDedupStore(),\n\t\tstopCh: make(chan struct{}),\n\t}\n\tdefer close(a.stopCh)\n\n\trobot := &robottypes.Robot{\n\t\tMemberID: \"robot_e2e_dc_adapter\",\n\t\tTeamID: \"team_e2e_dc\",\n\t\tConfig: &robottypes.Config{\n\t\t\tIntegrations: &robottypes.Integrations{\n\t\t\t\tDiscord: &robottypes.DiscordConfig{\n\t\t\t\t\tEnabled: true,\n\t\t\t\t\tBotToken: dcBotToken,\n\t\t\t\t\tAppID: dcAppID,\n\t\t\t\t},\n\t\t\t},\n\t\t},\n\t}\n\n\ta.Apply(context.Background(), robot)\n\n\ta.mu.RLock()\n\tentry, ok := a.bots[\"robot_e2e_dc_adapter\"]\n\ta.mu.RUnlock()\n\n\trequire.True(t, ok, \"bot should be registered\")\n\tassert.Equal(t, dcBotToken, entry.bot.Token())\n\tassert.Equal(t, dcAppID, entry.appID)\n\n\tt.Logf(\"OK Apply: discord bot registered robot=%s app=%s\", robot.MemberID, entry.appID)\n}\n\n// TestE2E_Adapter_Apply_Update verifies re-Apply with same token is a no-op.\nfunc TestE2E_Adapter_Apply_Update(t *testing.T) {\n\tskipIfNoToken(t)\n\n\ta := &Adapter{\n\t\tbots: make(map[string]*botEntry),\n\t\tappIdx: make(map[string]string),\n\t\tdedup: newDedupStore(),\n\t\tstopCh: make(chan struct{}),\n\t}\n\tdefer close(a.stopCh)\n\n\trobot := &robottypes.Robot{\n\t\tMemberID: \"robot_e2e_dc_update\",\n\t\tTeamID: \"team_e2e_dc\",\n\t\tConfig: &robottypes.Config{\n\t\t\tIntegrations: &robottypes.Integrations{\n\t\t\t\tDiscord: &robottypes.DiscordConfig{\n\t\t\t\t\tEnabled: true,\n\t\t\t\t\tBotToken: dcBotToken,\n\t\t\t\t\tAppID: dcAppID,\n\t\t\t\t},\n\t\t\t},\n\t\t},\n\t}\n\n\ta.Apply(context.Background(), robot)\n\ta.mu.RLock()\n\t_, ok := a.bots[\"robot_e2e_dc_update\"]\n\ta.mu.RUnlock()\n\trequire.True(t, ok)\n\n\ta.Apply(context.Background(), robot)\n\ta.mu.RLock()\n\tassert.Len(t, a.bots, 1)\n\ta.mu.RUnlock()\n\n\ta.Remove(context.Background(), \"robot_e2e_dc_update\")\n\ta.mu.RLock()\n\t_, ok = a.bots[\"robot_e2e_dc_update\"]\n\ta.mu.RUnlock()\n\tassert.False(t, ok, \"bot should be removed\")\n\tt.Log(\"OK Apply/Remove lifecycle verified\")\n}\n\n// TestE2E_Adapter_Dedup verifies deduplication works.\nfunc TestE2E_Adapter_Dedup(t *testing.T) {\n\ta := &Adapter{\n\t\tbots: make(map[string]*botEntry),\n\t\tappIdx: make(map[string]string),\n\t\tdedup: newDedupStore(),\n\t\tstopCh: make(chan struct{}),\n\t}\n\tdefer close(a.stopCh)\n\n\tkey := \"dc:test-robot:msg-12345\"\n\tassert.True(t, a.dedup.markSeen(key), \"first time should return true\")\n\tassert.False(t, a.dedup.markSeen(key), \"second time should return false (dedup)\")\n\tt.Log(\"OK dedup working correctly\")\n}\n\n// TestE2E_Adapter_HandleMessages verifies message handling.\nfunc TestE2E_Adapter_HandleMessages(t *testing.T) {\n\tskipIfNoToken(t)\n\n\tbot, err := dcapi.NewBot(dcBotToken, dcAppID)\n\trequire.NoError(t, err)\n\n\ta := &Adapter{\n\t\tbots: make(map[string]*botEntry),\n\t\tappIdx: make(map[string]string),\n\t\tdedup: newDedupStore(),\n\t\tstopCh: make(chan struct{}),\n\t}\n\tdefer close(a.stopCh)\n\n\tentry := &botEntry{\n\t\trobotID: \"robot_e2e_dc_handle\",\n\t\tappID: dcAppID,\n\t\tbot: bot,\n\t}\n\n\tctx, cancel := context.WithTimeout(context.Background(), 10*time.Second)\n\tdefer cancel()\n\n\tcms := []*dcapi.ConvertedMessage{\n\t\t{\n\t\t\tMessageID: \"test_msg_1\",\n\t\t\tChannelID: \"test_ch_1\",\n\t\t\tAuthorID: \"test_user_1\",\n\t\t\tAuthorName: \"TestUser\",\n\t\t\tText: \"Hello from E2E test\",\n\t\t},\n\t}\n\n\ta.handleMessages(ctx, entry, cms)\n\n\tassert.False(t, a.dedup.markSeen(\"dc:robot_e2e_dc_handle:test_msg_1\"),\n\t\t\"message should be marked as seen after handleMessages\")\n\tt.Log(\"OK handleMessages processed 1 message\")\n}\n\n// TestE2E_Adapter_ApplyDisabled verifies Apply removes bot when disabled.\nfunc TestE2E_Adapter_ApplyDisabled(t *testing.T) {\n\ta := &Adapter{\n\t\tbots: make(map[string]*botEntry),\n\t\tappIdx: make(map[string]string),\n\t\tdedup: newDedupStore(),\n\t\tstopCh: make(chan struct{}),\n\t}\n\tdefer close(a.stopCh)\n\n\trobot := &robottypes.Robot{\n\t\tMemberID: \"robot_e2e_dc_disabled\",\n\t\tTeamID: \"team_e2e_dc\",\n\t\tConfig: &robottypes.Config{\n\t\t\tIntegrations: &robottypes.Integrations{\n\t\t\t\tDiscord: &robottypes.DiscordConfig{\n\t\t\t\t\tEnabled: false,\n\t\t\t\t\tBotToken: \"some_token\",\n\t\t\t\t},\n\t\t\t},\n\t\t},\n\t}\n\n\ta.Apply(context.Background(), robot)\n\ta.mu.RLock()\n\t_, ok := a.bots[\"robot_e2e_dc_disabled\"]\n\ta.mu.RUnlock()\n\tassert.False(t, ok, \"disabled bot should not be registered\")\n\tt.Log(\"OK disabled config not registered\")\n}\n\n// TestE2E_BotUser verifies real Discord credentials.\nfunc TestE2E_BotUser(t *testing.T) {\n\tskipIfNoToken(t)\n\n\tbot, err := dcapi.NewBot(dcBotToken, dcAppID)\n\trequire.NoError(t, err)\n\n\tuser, err := bot.BotUser()\n\trequire.NoError(t, err)\n\tassert.NotEmpty(t, user.ID)\n\tassert.NotEmpty(t, user.Username)\n\tassert.True(t, user.Bot)\n\tt.Logf(\"OK Discord bot verified: id=%s username=%s\", user.ID, user.Username)\n}\n" }, { "path": "agent/robot/events/integrations/discord/gateway.go", "content": "package discord\n\nimport (\n\t\"context\"\n\t\"time\"\n\n\t\"github.com/bwmarrin/discordgo\"\n\tdcapi \"github.com/yaoapp/yao/integrations/discord\"\n)\n\nconst reconnectDelay = 5 * time.Second\n\n// gatewayLoop starts the Discord WebSocket Gateway for a single bot.\n// It automatically reconnects on failure.\nfunc (a *Adapter) gatewayLoop(ctx context.Context, entry *botEntry) {\n\tlog.Info(\"discord gatewayLoop started robot=%s app=%s\", entry.robotID, entry.appID)\n\n\tfor {\n\t\tselect {\n\t\tcase <-ctx.Done():\n\t\t\tlog.Info(\"discord gatewayLoop stopped robot=%s\", entry.robotID)\n\t\t\treturn\n\t\tcase <-a.stopCh:\n\t\t\treturn\n\t\tdefault:\n\t\t}\n\n\t\terr := a.runGateway(ctx, entry)\n\t\tif err != nil {\n\t\t\tlog.Error(\"discord gateway disconnected robot=%s: %v, reconnecting in %s\", entry.robotID, err, reconnectDelay)\n\t\t}\n\n\t\tselect {\n\t\tcase <-ctx.Done():\n\t\t\treturn\n\t\tcase <-a.stopCh:\n\t\t\treturn\n\t\tcase <-time.After(reconnectDelay):\n\t\t}\n\t}\n}\n\nfunc (a *Adapter) runGateway(ctx context.Context, entry *botEntry) error {\n\tsession := entry.bot.Session()\n\n\tsession.AddHandler(func(s *discordgo.Session, m *discordgo.MessageCreate) {\n\t\ta.onMessageCreate(ctx, entry, m)\n\t})\n\n\tif err := session.Open(); err != nil {\n\t\treturn err\n\t}\n\n\t// Block until context is cancelled or stop signal\n\tselect {\n\tcase <-ctx.Done():\n\tcase <-a.stopCh:\n\t}\n\n\treturn session.Close()\n}\n\nfunc (a *Adapter) onMessageCreate(ctx context.Context, entry *botEntry, m *discordgo.MessageCreate) {\n\tif m == nil || m.Author == nil {\n\t\treturn\n\t}\n\n\t// Ignore bot's own messages\n\tif m.Author.Bot {\n\t\treturn\n\t}\n\n\tcm := dcapi.ConvertMessageCreate(m)\n\tif cm == nil {\n\t\treturn\n\t}\n\n\tif cm.HasMedia() {\n\t\tgroups := []string{\"discord\", entry.robotID}\n\t\tdcapi.ResolveMedia(ctx, cm, groups)\n\t}\n\n\ta.handleMessages(ctx, entry, []*dcapi.ConvertedMessage{cm})\n}\n" }, { "path": "agent/robot/events/integrations/discord/message.go", "content": "package discord\n\nimport (\n\t\"context\"\n\t\"fmt\"\n\t\"strings\"\n\n\tagentcontext \"github.com/yaoapp/yao/agent/context\"\n\tevents \"github.com/yaoapp/yao/agent/robot/events\"\n\t\"github.com/yaoapp/yao/event\"\n\tdcapi \"github.com/yaoapp/yao/integrations/discord\"\n)\n\n// handleMessages processes a batch of Discord messages.\nfunc (a *Adapter) handleMessages(ctx context.Context, entry *botEntry, cms []*dcapi.ConvertedMessage) {\n\tif len(cms) == 0 {\n\t\treturn\n\t}\n\n\tvar allParts []interface{}\n\tvar lastCM *dcapi.ConvertedMessage\n\n\tfor _, cm := range cms {\n\t\tif cm == nil {\n\t\t\tcontinue\n\t\t}\n\n\t\t// Skip bot commands (messages starting with /)\n\t\tif strings.HasPrefix(strings.TrimSpace(cm.Text), \"/\") && !cm.HasMedia() {\n\t\t\tcontinue\n\t\t}\n\n\t\tdedupKey := fmt.Sprintf(\"dc:%s:%s\", entry.robotID, cm.MessageID)\n\t\tif !a.dedup.markSeen(dedupKey) {\n\t\t\tcontinue\n\t\t}\n\n\t\tparts := buildContentParts(cm)\n\t\tif len(parts) == 0 {\n\t\t\tcontinue\n\t\t}\n\n\t\tallParts = append(allParts, parts...)\n\t\tlastCM = cm\n\t}\n\n\tif len(allParts) == 0 || lastCM == nil {\n\t\treturn\n\t}\n\n\tcontent := mergeContentParts(allParts)\n\n\tmsgPayload := events.MessagePayload{\n\t\tRobotID: entry.robotID,\n\t\tMessages: []agentcontext.Message{\n\t\t\t{Role: agentcontext.RoleUser, Content: content},\n\t\t},\n\t\tMetadata: &events.MessageMetadata{\n\t\t\tChannel: \"discord\",\n\t\t\tMessageID: lastCM.MessageID,\n\t\t\tAppID: entry.appID,\n\t\t\tChatID: lastCM.ChannelID,\n\t\t\tSenderID: lastCM.AuthorID,\n\t\t\tSenderName: lastCM.AuthorName,\n\t\t\tLocale: events.NormalizeLocale(discordLocale(lastCM.Locale)),\n\t\t\tExtra: map[string]any{\n\t\t\t\t\"discord_message_id\": lastCM.MessageID,\n\t\t\t\t\"guild_id\": lastCM.GuildID,\n\t\t\t\t\"is_dm\": lastCM.IsDM,\n\t\t\t},\n\t\t},\n\t}\n\n\tif _, err := event.Push(ctx, events.Message, msgPayload); err != nil {\n\t\tlog.Error(\"discord adapter: event.Push robot.message failed robot=%s: %v\", entry.robotID, err)\n\t}\n}\n\nfunc buildContentParts(cm *dcapi.ConvertedMessage) []interface{} {\n\tvar parts []interface{}\n\n\tif cm.HasText() {\n\t\tparts = append(parts, map[string]interface{}{\n\t\t\t\"type\": \"text\",\n\t\t\t\"text\": cm.Text,\n\t\t})\n\t}\n\n\tfor _, mi := range cm.MediaItems {\n\t\turl := mi.Wrapper\n\t\tif url == \"\" {\n\t\t\turl = mi.URL\n\t\t}\n\t\tif url == \"\" {\n\t\t\tcontinue\n\t\t}\n\t\tparts = append(parts, map[string]interface{}{\n\t\t\t\"type\": \"file\",\n\t\t\t\"file_url\": url,\n\t\t\t\"mime_type\": mi.ContentType,\n\t\t\t\"file_name\": mi.FileName,\n\t\t})\n\t}\n\n\treturn parts\n}\n\nfunc mergeContentParts(parts []interface{}) interface{} {\n\tallText := true\n\tfor _, p := range parts {\n\t\tm, ok := p.(map[string]interface{})\n\t\tif !ok || m[\"type\"] != \"text\" {\n\t\t\tallText = false\n\t\t\tbreak\n\t\t}\n\t}\n\n\tif allText {\n\t\tvar buf strings.Builder\n\t\tfor i, p := range parts {\n\t\t\tif i > 0 {\n\t\t\t\tbuf.WriteString(\"\\n\")\n\t\t\t}\n\t\t\tm := p.(map[string]interface{})\n\t\t\tbuf.WriteString(m[\"text\"].(string))\n\t\t}\n\t\treturn buf.String()\n\t}\n\n\treturn parts\n}\n\nfunc discordLocale(locale string) string {\n\tif locale == \"\" {\n\t\treturn \"en\"\n\t}\n\treturn locale\n}\n" }, { "path": "agent/robot/events/integrations/discord/reply.go", "content": "package discord\n\nimport (\n\t\"context\"\n\t\"fmt\"\n\t\"strings\"\n\n\tagentcontext \"github.com/yaoapp/yao/agent/context\"\n\tevents \"github.com/yaoapp/yao/agent/robot/events\"\n\tdcapi \"github.com/yaoapp/yao/integrations/discord\"\n)\n\n// Reply sends the assistant message back to the originating Discord channel.\nfunc (a *Adapter) Reply(ctx context.Context, msg *agentcontext.Message, metadata *events.MessageMetadata) error {\n\tif msg == nil || metadata == nil {\n\t\treturn fmt.Errorf(\"nil message or metadata\")\n\t}\n\n\tentry := a.resolveByChat(metadata)\n\tif entry == nil {\n\t\treturn fmt.Errorf(\"no bot registered for discord metadata (appID=%s)\", metadata.AppID)\n\t}\n\n\tvar replyToID string\n\tif metadata.Extra != nil {\n\t\tif v, ok := metadata.Extra[\"discord_message_id\"]; ok {\n\t\t\tif s, ok := v.(string); ok {\n\t\t\t\treplyToID = s\n\t\t\t}\n\t\t}\n\t}\n\n\treturn a.sendContent(ctx, entry, metadata.ChatID, replyToID, msg.Content)\n}\n\nfunc (a *Adapter) sendContent(ctx context.Context, entry *botEntry, channelID, replyToID string, content interface{}) error {\n\tswitch c := content.(type) {\n\tcase string:\n\t\tif strings.TrimSpace(c) == \"\" {\n\t\t\treturn nil\n\t\t}\n\t\tformatted := dcapi.FormatDiscordMarkdown(c)\n\t\tif replyToID != \"\" {\n\t\t\t_, err := entry.bot.SendMessageReply(channelID, formatted, replyToID)\n\t\t\treturn err\n\t\t}\n\t\t_, err := entry.bot.SendMessage(channelID, formatted)\n\t\treturn err\n\n\tcase []interface{}:\n\t\treturn a.sendParts(ctx, entry, channelID, replyToID, c)\n\n\tdefault:\n\t\tparts, ok := toContentParts(content)\n\t\tif ok {\n\t\t\treturn a.sendPartsTyped(ctx, entry, channelID, replyToID, parts)\n\t\t}\n\t\t_, err := entry.bot.SendMessage(channelID, fmt.Sprintf(\"%v\", content))\n\t\treturn err\n\t}\n}\n\nfunc (a *Adapter) sendParts(ctx context.Context, entry *botEntry, channelID, replyToID string, parts []interface{}) error {\n\tvar textBuf strings.Builder\n\tfor _, part := range parts {\n\t\tm, ok := part.(map[string]interface{})\n\t\tif !ok {\n\t\t\tcontinue\n\t\t}\n\t\tpartType, _ := m[\"type\"].(string)\n\t\tswitch partType {\n\t\tcase \"text\":\n\t\t\tif text, ok := m[\"text\"].(string); ok {\n\t\t\t\ttextBuf.WriteString(text)\n\t\t\t}\n\t\tcase \"image_url\":\n\t\t\tif err := a.flushText(entry, channelID, replyToID, &textBuf); err != nil {\n\t\t\t\treturn err\n\t\t\t}\n\t\t\tif imgMap, ok := m[\"image_url\"].(map[string]interface{}); ok {\n\t\t\t\tif url, ok := imgMap[\"url\"].(string); ok {\n\t\t\t\t\tif err := sendFileOrWrapper(entry, channelID, url, \"\"); err != nil {\n\t\t\t\t\t\tlog.Error(\"discord reply: send image: %v\", err)\n\t\t\t\t\t}\n\t\t\t\t}\n\t\t\t}\n\t\tcase \"file\":\n\t\t\tif err := a.flushText(entry, channelID, replyToID, &textBuf); err != nil {\n\t\t\t\treturn err\n\t\t\t}\n\t\t\tfileURL, _ := m[\"file_url\"].(string)\n\t\t\tif fileURL == \"\" {\n\t\t\t\tif fileMap, ok := m[\"file\"].(map[string]interface{}); ok {\n\t\t\t\t\tfileURL, _ = fileMap[\"url\"].(string)\n\t\t\t\t}\n\t\t\t}\n\t\t\tif fileURL != \"\" {\n\t\t\t\tif err := sendFileOrWrapper(entry, channelID, fileURL, \"\"); err != nil {\n\t\t\t\t\tlog.Error(\"discord reply: send file: %v\", err)\n\t\t\t\t}\n\t\t\t}\n\t\t}\n\t}\n\treturn a.flushText(entry, channelID, replyToID, &textBuf)\n}\n\nfunc (a *Adapter) sendPartsTyped(ctx context.Context, entry *botEntry, channelID, replyToID string, parts []agentcontext.ContentPart) error {\n\tvar textBuf strings.Builder\n\tfor _, part := range parts {\n\t\tswitch part.Type {\n\t\tcase agentcontext.ContentText:\n\t\t\ttextBuf.WriteString(part.Text)\n\t\tcase agentcontext.ContentImageURL:\n\t\t\tif err := a.flushText(entry, channelID, replyToID, &textBuf); err != nil {\n\t\t\t\treturn err\n\t\t\t}\n\t\t\tif part.ImageURL != nil {\n\t\t\t\tif err := sendFileOrWrapper(entry, channelID, part.ImageURL.URL, \"\"); err != nil {\n\t\t\t\t\tlog.Error(\"discord reply: send image: %v\", err)\n\t\t\t\t}\n\t\t\t}\n\t\tcase agentcontext.ContentFile:\n\t\t\tif err := a.flushText(entry, channelID, replyToID, &textBuf); err != nil {\n\t\t\t\treturn err\n\t\t\t}\n\t\t\tif part.File != nil {\n\t\t\t\tif err := sendFileOrWrapper(entry, channelID, part.File.URL, part.File.Filename); err != nil {\n\t\t\t\t\tlog.Error(\"discord reply: send file: %v\", err)\n\t\t\t\t}\n\t\t\t}\n\t\t}\n\t}\n\treturn a.flushText(entry, channelID, replyToID, &textBuf)\n}\n\nfunc (a *Adapter) flushText(entry *botEntry, channelID, replyToID string, buf *strings.Builder) error {\n\tif buf.Len() == 0 {\n\t\treturn nil\n\t}\n\ttext := dcapi.FormatDiscordMarkdown(buf.String())\n\tbuf.Reset()\n\n\tif replyToID != \"\" {\n\t\t_, err := entry.bot.SendMessageReply(channelID, text, replyToID)\n\t\treturn err\n\t}\n\t_, err := entry.bot.SendMessage(channelID, text)\n\treturn err\n}\n\nfunc sendFileOrWrapper(entry *botEntry, channelID, url, caption string) error {\n\tif strings.Contains(url, \"://\") && !strings.HasPrefix(url, \"http\") {\n\t\treturn entry.bot.SendMediaFromWrapper(channelID, url, caption)\n\t}\n\tif strings.HasPrefix(url, \"http\") {\n\t\t_, err := entry.bot.SendMessage(channelID, url)\n\t\treturn err\n\t}\n\treturn fmt.Errorf(\"unsupported file URL scheme: %s\", url)\n}\n\nfunc toContentParts(content interface{}) ([]agentcontext.ContentPart, bool) {\n\tparts, ok := content.([]agentcontext.ContentPart)\n\treturn parts, ok\n}\n\nfunc (a *Adapter) resolveByChat(metadata *events.MessageMetadata) *botEntry {\n\tif metadata.AppID != \"\" {\n\t\tif entry, ok := a.resolveByAppID(metadata.AppID); ok {\n\t\t\treturn entry\n\t\t}\n\t}\n\ta.mu.RLock()\n\tdefer a.mu.RUnlock()\n\tfor _, entry := range a.bots {\n\t\treturn entry\n\t}\n\treturn nil\n}\n" }, { "path": "agent/robot/events/integrations/dispatcher.go", "content": "package integrations\n\nimport (\n\t\"context\"\n\t\"fmt\"\n\n\t\"github.com/yaoapp/gou/model\"\n\t\"github.com/yaoapp/kun/maps\"\n\tagentcontext \"github.com/yaoapp/yao/agent/context\"\n\trobotcache \"github.com/yaoapp/yao/agent/robot/cache\"\n\tevents \"github.com/yaoapp/yao/agent/robot/events\"\n\t\"github.com/yaoapp/yao/agent/robot/logger\"\n\trobottypes \"github.com/yaoapp/yao/agent/robot/types\"\n\t\"github.com/yaoapp/yao/event\"\n\teventtypes \"github.com/yaoapp/yao/event/types\"\n)\n\nvar log = logger.New(\"dispatcher\")\n\n// Adapter is the interface each platform adapter implements.\ntype Adapter interface {\n\tApply(ctx context.Context, robot *robottypes.Robot)\n\tRemove(ctx context.Context, robotID string)\n\tReply(ctx context.Context, msg *agentcontext.Message, metadata *events.MessageMetadata) error\n}\n\n// Dispatcher distributes Robot integration configs to platform adapters.\ntype Dispatcher struct {\n\trobotCache *robotcache.Cache\n\tadapters map[string]Adapter // key matches Integrations field: \"telegram\", \"discord\", etc.\n\tstopCh chan struct{}\n\tsubID string\n}\n\n// NewDispatcher creates a Dispatcher.\n// Each adapter has a fixed key matching the field name in robottypes.Integrations.\nfunc NewDispatcher(cache *robotcache.Cache, adapters map[string]Adapter) *Dispatcher {\n\treturn &Dispatcher{\n\t\trobotCache: cache,\n\t\tadapters: adapters,\n\t\tstopCh: make(chan struct{}),\n\t}\n}\n\n// Start loads all robots and subscribes to config change events.\nfunc (d *Dispatcher) Start(ctx context.Context) error {\n\td.loadAll(ctx)\n\n\tevents.RegisterReplyFunc(d.reply)\n\n\tch := make(chan *eventtypes.Event, 64)\n\td.subID = event.Subscribe(\"robot.config.*\", ch)\n\tgo d.watch(ctx, ch)\n\n\tlog.Info(\"integration dispatcher: started with %d adapters\", len(d.adapters))\n\treturn nil\n}\n\n// reply routes a reply to the correct adapter based on channel.\n// When channel is empty (e.g. delivery), broadcasts to all adapters.\nfunc (d *Dispatcher) reply(ctx context.Context, msg *agentcontext.Message, metadata *events.MessageMetadata) error {\n\tif metadata == nil {\n\t\treturn fmt.Errorf(\"no metadata in reply\")\n\t}\n\n\tif metadata.Channel != \"\" {\n\t\tadapter, ok := d.adapters[metadata.Channel]\n\t\tif !ok {\n\t\t\treturn fmt.Errorf(\"no adapter for channel: %s\", metadata.Channel)\n\t\t}\n\t\treturn adapter.Reply(ctx, msg, metadata)\n\t}\n\n\tvar lastErr error\n\tfor name, adapter := range d.adapters {\n\t\tif err := adapter.Reply(ctx, msg, metadata); err != nil {\n\t\t\tlog.Error(\"dispatcher reply: broadcast to %s failed: %v\", name, err)\n\t\t\tlastErr = err\n\t\t}\n\t}\n\treturn lastErr\n}\n\n// Stop unsubscribes from events.\nfunc (d *Dispatcher) Stop() {\n\tclose(d.stopCh)\n\tif d.subID != \"\" {\n\t\tevent.Unsubscribe(d.subID)\n\t}\n\tlog.Info(\"integration dispatcher: stopped\")\n}\n\nfunc (d *Dispatcher) loadAll(ctx context.Context) {\n\trobots := d.loadIntegrationRobots()\n\tfor _, robot := range robots {\n\t\td.robotCache.Add(robot)\n\t\td.apply(ctx, robot)\n\t}\n\tlog.Info(\"integration dispatcher: initial load complete, %d robots with integrations\", len(robots))\n}\n\n// loadIntegrationRobots queries all active robots that have a non-null\n// robot_config (which may contain integrations). This is independent of\n// autonomous_mode so non-autonomous robots with Telegram etc. are included.\nfunc (d *Dispatcher) loadIntegrationRobots() []*robottypes.Robot {\n\tm := model.Select(\"__yao.member\")\n\tfields := []interface{}{\n\t\t\"id\", \"member_id\", \"team_id\", \"display_name\", \"bio\",\n\t\t\"system_prompt\", \"robot_status\", \"autonomous_mode\",\n\t\t\"robot_config\", \"robot_email\", \"agents\", \"mcp_servers\",\n\t\t\"manager_id\", \"language_model\",\n\t}\n\n\tpage := 1\n\tpageSize := 100\n\tvar result []*robottypes.Robot\n\n\tfor {\n\t\tres, err := m.Paginate(model.QueryParam{\n\t\t\tSelect: fields,\n\t\t\tWheres: []model.QueryWhere{\n\t\t\t\t{Column: \"member_type\", Value: \"robot\"},\n\t\t\t\t{Column: \"status\", Value: \"active\"},\n\t\t\t},\n\t\t}, page, pageSize)\n\t\tif err != nil {\n\t\t\tlog.Error(\"loadIntegrationRobots: query failed page=%d: %v\", page, err)\n\t\t\tbreak\n\t\t}\n\n\t\tdata, ok := res.Get(\"data\").([]maps.MapStr)\n\t\tif !ok || len(data) == 0 {\n\t\t\tbreak\n\t\t}\n\n\t\tfor _, record := range data {\n\t\t\trobot, err := robottypes.NewRobotFromMap(map[string]interface{}(record))\n\t\t\tif err != nil {\n\t\t\t\tcontinue\n\t\t\t}\n\t\t\tif robot.Config != nil && robot.Config.Integrations != nil && len(parseIntegrations(robot.Config.Integrations)) > 0 {\n\t\t\t\tresult = append(result, robot)\n\t\t\t}\n\t\t}\n\n\t\ttotal, _ := res.Get(\"total\").(int)\n\t\tif page*pageSize >= total {\n\t\t\tbreak\n\t\t}\n\t\tpage++\n\t}\n\n\treturn result\n}\n\n// apply parses which integrations the robot has configured,\n// and calls the matching adapter for each one.\nfunc (d *Dispatcher) apply(ctx context.Context, robot *robottypes.Robot) {\n\tif robot.Config == nil || robot.Config.Integrations == nil {\n\t\treturn\n\t}\n\tfor _, key := range parseIntegrations(robot.Config.Integrations) {\n\t\tif adapter, ok := d.adapters[key]; ok {\n\t\t\tadapter.Apply(ctx, robot)\n\t\t}\n\t}\n}\n\nfunc (d *Dispatcher) remove(ctx context.Context, robotID string) {\n\tfor _, adapter := range d.adapters {\n\t\tadapter.Remove(ctx, robotID)\n\t}\n}\n\n// parseIntegrations returns the keys of integrations present in the config.\nfunc parseIntegrations(intg *robottypes.Integrations) []string {\n\tvar keys []string\n\tif intg.Telegram != nil {\n\t\tkeys = append(keys, \"telegram\")\n\t}\n\tif intg.Feishu != nil {\n\t\tkeys = append(keys, \"feishu\")\n\t}\n\tif intg.DingTalk != nil {\n\t\tkeys = append(keys, \"dingtalk\")\n\t}\n\tif intg.Discord != nil {\n\t\tkeys = append(keys, \"discord\")\n\t}\n\treturn keys\n}\n\nfunc (d *Dispatcher) watch(ctx context.Context, ch <-chan *eventtypes.Event) {\n\tfor {\n\t\tselect {\n\t\tcase <-d.stopCh:\n\t\t\treturn\n\t\tcase <-ctx.Done():\n\t\t\treturn\n\t\tcase ev, ok := <-ch:\n\t\t\tif !ok {\n\t\t\t\treturn\n\t\t\t}\n\t\t\td.dispatch(ctx, ev)\n\t\t}\n\t}\n}\n\nfunc (d *Dispatcher) dispatch(ctx context.Context, ev *eventtypes.Event) {\n\tvar payload events.RobotConfigPayload\n\tif err := ev.Should(&payload); err != nil {\n\t\tlog.Error(\"integration dispatcher: invalid config event: %v\", err)\n\t\treturn\n\t}\n\n\tswitch ev.Type {\n\tcase events.RobotConfigCreated, events.RobotConfigUpdated:\n\t\trobot := d.robotCache.Get(payload.MemberID)\n\t\tif robot == nil {\n\t\t\trCtx := robottypes.NewContext(ctx, nil)\n\t\t\tloaded, err := d.robotCache.LoadByID(rCtx, payload.MemberID)\n\t\t\tif err != nil {\n\t\t\t\tlog.Warn(\"integration dispatcher: failed to load robot from DB member=%s: %v\", payload.MemberID, err)\n\t\t\t\treturn\n\t\t\t}\n\t\t\td.robotCache.Add(loaded)\n\t\t\trobot = loaded\n\t\t\tlog.Info(\"integration dispatcher: loaded robot from DB member=%s\", payload.MemberID)\n\t\t}\n\t\td.apply(ctx, robot)\n\n\tcase events.RobotConfigDeleted:\n\t\td.remove(ctx, payload.MemberID)\n\t}\n}\n" }, { "path": "agent/robot/events/integrations/dispatcher_test.go", "content": "package integrations\n\nimport (\n\t\"context\"\n\t\"sync\"\n\t\"testing\"\n\t\"time\"\n\n\t\"github.com/stretchr/testify/assert\"\n\t\"github.com/stretchr/testify/require\"\n\tagentcontext \"github.com/yaoapp/yao/agent/context\"\n\trobotcache \"github.com/yaoapp/yao/agent/robot/cache\"\n\tevents \"github.com/yaoapp/yao/agent/robot/events\"\n\trobottypes \"github.com/yaoapp/yao/agent/robot/types\"\n\t\"github.com/yaoapp/yao/event\"\n\teventtypes \"github.com/yaoapp/yao/event/types\"\n)\n\n// mockAdapter records Apply/Remove calls for assertions.\ntype mockAdapter struct {\n\tmu sync.Mutex\n\tapplied []*robottypes.Robot\n\tremoved []string\n}\n\nfunc (m *mockAdapter) Apply(ctx context.Context, robot *robottypes.Robot) {\n\tm.mu.Lock()\n\tdefer m.mu.Unlock()\n\tm.applied = append(m.applied, robot)\n}\n\nfunc (m *mockAdapter) Remove(ctx context.Context, robotID string) {\n\tm.mu.Lock()\n\tdefer m.mu.Unlock()\n\tm.removed = append(m.removed, robotID)\n}\n\nfunc (m *mockAdapter) Reply(ctx context.Context, msg *agentcontext.Message, metadata *events.MessageMetadata) error {\n\treturn nil\n}\n\nfunc (m *mockAdapter) getApplied() []*robottypes.Robot {\n\tm.mu.Lock()\n\tdefer m.mu.Unlock()\n\tcp := make([]*robottypes.Robot, len(m.applied))\n\tcopy(cp, m.applied)\n\treturn cp\n}\n\nfunc (m *mockAdapter) getRemoved() []string {\n\tm.mu.Lock()\n\tdefer m.mu.Unlock()\n\tcp := make([]string, len(m.removed))\n\tcopy(cp, m.removed)\n\treturn cp\n}\n\n// noopHandler satisfies event.Handler so we can register \"robot\" prefix for Push/Call.\ntype noopHandler struct{}\n\nfunc (h *noopHandler) Handle(ctx context.Context, ev *eventtypes.Event, resp chan<- eventtypes.Result) {\n\tif ev.IsCall {\n\t\tresp <- eventtypes.Result{}\n\t}\n}\nfunc (h *noopHandler) Shutdown(ctx context.Context) error { return nil }\n\nvar eventOnce sync.Once\n\nfunc setupEventBus(t *testing.T) {\n\tt.Helper()\n\teventOnce.Do(func() {\n\t\tevent.Register(\"robot\", &noopHandler{})\n\t})\n\tif err := event.Start(); err != nil && err != event.ErrAlreadyStart {\n\t\tt.Fatalf(\"event.Start: %v\", err)\n\t}\n\tt.Cleanup(func() { _ = event.Stop(context.Background()) })\n}\n\nfunc newRobot(memberID, teamID string, intg *robottypes.Integrations) *robottypes.Robot {\n\treturn &robottypes.Robot{\n\t\tMemberID: memberID,\n\t\tTeamID: teamID,\n\t\tAutonomousMode: true,\n\t\tConfig: &robottypes.Config{\n\t\t\tIntegrations: intg,\n\t\t},\n\t}\n}\n\nfunc TestLoadAll_OnlyTelegramConfigured(t *testing.T) {\n\tsetupEventBus(t)\n\tcache := robotcache.New()\n\n\ttgRobot := newRobot(\"r-tg\", \"team1\", &robottypes.Integrations{\n\t\tTelegram: &robottypes.TelegramConfig{Enabled: true, BotToken: \"tok\"},\n\t})\n\tnoIntgRobot := newRobot(\"r-plain\", \"team1\", nil)\n\tcache.Add(tgRobot)\n\tcache.Add(noIntgRobot)\n\n\ttgAdapter := &mockAdapter{}\n\td := NewDispatcher(cache, map[string]Adapter{\"telegram\": tgAdapter})\n\n\trequire.NoError(t, d.Start(context.Background()))\n\tdefer d.Stop()\n\n\tapplied := tgAdapter.getApplied()\n\tassert.Len(t, applied, 1)\n\tassert.Equal(t, \"r-tg\", applied[0].MemberID)\n}\n\nfunc TestLoadAll_NoIntegrations(t *testing.T) {\n\tsetupEventBus(t)\n\tcache := robotcache.New()\n\n\tcache.Add(newRobot(\"r1\", \"team1\", nil))\n\tcache.Add(&robottypes.Robot{MemberID: \"r2\", TeamID: \"team1\"})\n\n\ttgAdapter := &mockAdapter{}\n\td := NewDispatcher(cache, map[string]Adapter{\"telegram\": tgAdapter})\n\n\trequire.NoError(t, d.Start(context.Background()))\n\tdefer d.Stop()\n\n\tassert.Empty(t, tgAdapter.getApplied())\n}\n\nfunc TestLoadAll_MultipleAdapters(t *testing.T) {\n\tsetupEventBus(t)\n\tcache := robotcache.New()\n\n\t// Only Telegram configured, no Discord\n\trobot := newRobot(\"r-multi\", \"team1\", &robottypes.Integrations{\n\t\tTelegram: &robottypes.TelegramConfig{Enabled: true, BotToken: \"tok\"},\n\t})\n\tcache.Add(robot)\n\n\ttgAdapter := &mockAdapter{}\n\tdiscordAdapter := &mockAdapter{}\n\td := NewDispatcher(cache, map[string]Adapter{\n\t\t\"telegram\": tgAdapter,\n\t\t\"discord\": discordAdapter,\n\t})\n\n\trequire.NoError(t, d.Start(context.Background()))\n\tdefer d.Stop()\n\n\tassert.Len(t, tgAdapter.getApplied(), 1)\n\tassert.Empty(t, discordAdapter.getApplied(), \"discord adapter should not be called\")\n}\n\nfunc TestConfigCreated_TriggersApply(t *testing.T) {\n\tsetupEventBus(t)\n\tcache := robotcache.New()\n\n\ttgAdapter := &mockAdapter{}\n\td := NewDispatcher(cache, map[string]Adapter{\"telegram\": tgAdapter})\n\n\trequire.NoError(t, d.Start(context.Background()))\n\tdefer d.Stop()\n\n\tassert.Empty(t, tgAdapter.getApplied())\n\n\t// Simulate: robot created with Telegram config, added to cache, event pushed\n\trobot := newRobot(\"r-new\", \"team1\", &robottypes.Integrations{\n\t\tTelegram: &robottypes.TelegramConfig{Enabled: true, BotToken: \"new-tok\"},\n\t})\n\tcache.Add(robot)\n\tevent.Push(context.Background(), events.RobotConfigCreated, events.RobotConfigPayload{\n\t\tMemberID: \"r-new\", TeamID: \"team1\",\n\t})\n\n\tassert.Eventually(t, func() bool {\n\t\treturn len(tgAdapter.getApplied()) == 1\n\t}, 2*time.Second, 50*time.Millisecond)\n\n\tassert.Equal(t, \"r-new\", tgAdapter.getApplied()[0].MemberID)\n}\n\nfunc TestConfigUpdated_TriggersApply(t *testing.T) {\n\tsetupEventBus(t)\n\tcache := robotcache.New()\n\n\trobot := newRobot(\"r-upd\", \"team1\", &robottypes.Integrations{\n\t\tTelegram: &robottypes.TelegramConfig{Enabled: true, BotToken: \"old-tok\"},\n\t})\n\tcache.Add(robot)\n\n\ttgAdapter := &mockAdapter{}\n\td := NewDispatcher(cache, map[string]Adapter{\"telegram\": tgAdapter})\n\n\trequire.NoError(t, d.Start(context.Background()))\n\tdefer d.Stop()\n\n\t// Initial load\n\tassert.Len(t, tgAdapter.getApplied(), 1)\n\n\t// Update config in cache\n\trobot.Config.Integrations.Telegram.BotToken = \"new-tok\"\n\tevent.Push(context.Background(), events.RobotConfigUpdated, events.RobotConfigPayload{\n\t\tMemberID: \"r-upd\", TeamID: \"team1\",\n\t})\n\n\tassert.Eventually(t, func() bool {\n\t\treturn len(tgAdapter.getApplied()) == 2\n\t}, 2*time.Second, 50*time.Millisecond)\n\n\tassert.Equal(t, \"new-tok\", tgAdapter.getApplied()[1].Config.Integrations.Telegram.BotToken)\n}\n\nfunc TestConfigDeleted_TriggersRemove(t *testing.T) {\n\tsetupEventBus(t)\n\tcache := robotcache.New()\n\n\trobot := newRobot(\"r-del\", \"team1\", &robottypes.Integrations{\n\t\tTelegram: &robottypes.TelegramConfig{Enabled: true, BotToken: \"tok\"},\n\t})\n\tcache.Add(robot)\n\n\ttgAdapter := &mockAdapter{}\n\td := NewDispatcher(cache, map[string]Adapter{\"telegram\": tgAdapter})\n\n\trequire.NoError(t, d.Start(context.Background()))\n\tdefer d.Stop()\n\n\tassert.Len(t, tgAdapter.getApplied(), 1)\n\n\tevent.Push(context.Background(), events.RobotConfigDeleted, events.RobotConfigPayload{\n\t\tMemberID: \"r-del\", TeamID: \"team1\",\n\t})\n\n\tassert.Eventually(t, func() bool {\n\t\treturn len(tgAdapter.getRemoved()) == 1\n\t}, 2*time.Second, 50*time.Millisecond)\n\n\tassert.Equal(t, \"r-del\", tgAdapter.getRemoved()[0])\n}\n\nfunc TestConfigCreated_RobotNotInCache(t *testing.T) {\n\tsetupEventBus(t)\n\tcache := robotcache.New()\n\n\ttgAdapter := &mockAdapter{}\n\td := NewDispatcher(cache, map[string]Adapter{\"telegram\": tgAdapter})\n\n\trequire.NoError(t, d.Start(context.Background()))\n\tdefer d.Stop()\n\n\t// Push event but don't add robot to cache\n\tevent.Push(context.Background(), events.RobotConfigCreated, events.RobotConfigPayload{\n\t\tMemberID: \"r-ghost\", TeamID: \"team1\",\n\t})\n\n\ttime.Sleep(200 * time.Millisecond)\n\tassert.Empty(t, tgAdapter.getApplied())\n}\n\nfunc TestParseIntegrations(t *testing.T) {\n\ttests := []struct {\n\t\tname string\n\t\tintg *robottypes.Integrations\n\t\texpected []string\n\t}{\n\t\t{\"nil\", nil, nil},\n\t\t{\"empty\", &robottypes.Integrations{}, nil},\n\t\t{\"telegram only\", &robottypes.Integrations{\n\t\t\tTelegram: &robottypes.TelegramConfig{Enabled: true},\n\t\t}, []string{\"telegram\"}},\n\t}\n\n\tfor _, tt := range tests {\n\t\tt.Run(tt.name, func(t *testing.T) {\n\t\t\tif tt.intg == nil {\n\t\t\t\treturn\n\t\t\t}\n\t\t\tresult := parseIntegrations(tt.intg)\n\t\t\tassert.Equal(t, tt.expected, result)\n\t\t})\n\t}\n}\n" }, { "path": "agent/robot/events/integrations/feishu/dedup.go", "content": "package feishu\n\nimport (\n\t\"sync\"\n\t\"time\"\n)\n\nconst (\n\tdedupTTL = 24 * time.Hour\n\tdedupCleanInterval = time.Hour\n)\n\ntype dedupStore struct {\n\tm sync.Map\n}\n\nfunc newDedupStore() *dedupStore {\n\treturn &dedupStore{}\n}\n\nfunc (d *dedupStore) markSeen(key string) bool {\n\tnow := time.Now().Unix()\n\t_, loaded := d.m.LoadOrStore(key, now)\n\treturn !loaded\n}\n\nfunc (d *dedupStore) cleaner(stopCh <-chan struct{}) {\n\tticker := time.NewTicker(dedupCleanInterval)\n\tdefer ticker.Stop()\n\tfor {\n\t\tselect {\n\t\tcase <-stopCh:\n\t\t\treturn\n\t\tcase <-ticker.C:\n\t\t\tcutoff := time.Now().Add(-dedupTTL).Unix()\n\t\t\td.m.Range(func(key, value any) bool {\n\t\t\t\tif ts, ok := value.(int64); ok && ts < cutoff {\n\t\t\t\t\td.m.Delete(key)\n\t\t\t\t}\n\t\t\t\treturn true\n\t\t\t})\n\t\t}\n\t}\n}\n" }, { "path": "agent/robot/events/integrations/feishu/e2e_test.go", "content": "package feishu\n\nimport (\n\t\"context\"\n\t\"os\"\n\t\"testing\"\n\t\"time\"\n\n\t\"github.com/stretchr/testify/assert\"\n\t\"github.com/stretchr/testify/require\"\n\trobottypes \"github.com/yaoapp/yao/agent/robot/types\"\n\tfsapi \"github.com/yaoapp/yao/integrations/feishu\"\n)\n\nvar (\n\tfsAppID string\n\tfsAppSecret string\n)\n\nfunc TestMain(m *testing.M) {\n\tfsAppID = os.Getenv(\"FEISHU_TEST_APP_ID\")\n\tfsAppSecret = os.Getenv(\"FEISHU_TEST_APP_SECRET\")\n\tos.Exit(m.Run())\n}\n\nfunc skipIfNoCreds(t *testing.T) {\n\tt.Helper()\n\tif fsAppID == \"\" || fsAppSecret == \"\" {\n\t\tt.Skip(\"FEISHU_TEST_APP_ID or FEISHU_TEST_APP_SECRET not set\")\n\t}\n}\n\n// TestE2E_Adapter_Apply verifies that Apply correctly registers a bot.\nfunc TestE2E_Adapter_Apply(t *testing.T) {\n\tskipIfNoCreds(t)\n\n\ta := &Adapter{\n\t\tbots: make(map[string]*botEntry),\n\t\tappIdx: make(map[string]string),\n\t\tdedup: newDedupStore(),\n\t\tstopCh: make(chan struct{}),\n\t}\n\tdefer close(a.stopCh)\n\n\trobot := &robottypes.Robot{\n\t\tMemberID: \"robot_e2e_feishu_adapter\",\n\t\tTeamID: \"team_e2e_fs\",\n\t\tConfig: &robottypes.Config{\n\t\t\tIntegrations: &robottypes.Integrations{\n\t\t\t\tFeishu: &robottypes.FeishuConfig{\n\t\t\t\t\tEnabled: true,\n\t\t\t\t\tAppID: fsAppID,\n\t\t\t\t\tAppSecret: fsAppSecret,\n\t\t\t\t},\n\t\t\t},\n\t\t},\n\t}\n\n\ta.Apply(context.Background(), robot)\n\n\ta.mu.RLock()\n\tentry, ok := a.bots[\"robot_e2e_feishu_adapter\"]\n\ta.mu.RUnlock()\n\n\trequire.True(t, ok, \"bot should be registered\")\n\tassert.Equal(t, fsAppID, entry.appID)\n\tassert.NotNil(t, entry.bot)\n\n\tt.Logf(\"OK Apply: feishu bot registered robot=%s app=%s\", robot.MemberID, entry.appID)\n}\n\n// TestE2E_Adapter_Apply_Update verifies re-Apply with same appID is a no-op.\nfunc TestE2E_Adapter_Apply_Update(t *testing.T) {\n\tskipIfNoCreds(t)\n\n\ta := &Adapter{\n\t\tbots: make(map[string]*botEntry),\n\t\tappIdx: make(map[string]string),\n\t\tdedup: newDedupStore(),\n\t\tstopCh: make(chan struct{}),\n\t}\n\tdefer close(a.stopCh)\n\n\trobot := &robottypes.Robot{\n\t\tMemberID: \"robot_e2e_feishu_update\",\n\t\tTeamID: \"team_e2e_fs\",\n\t\tConfig: &robottypes.Config{\n\t\t\tIntegrations: &robottypes.Integrations{\n\t\t\t\tFeishu: &robottypes.FeishuConfig{\n\t\t\t\t\tEnabled: true,\n\t\t\t\t\tAppID: fsAppID,\n\t\t\t\t\tAppSecret: fsAppSecret,\n\t\t\t\t},\n\t\t\t},\n\t\t},\n\t}\n\n\ta.Apply(context.Background(), robot)\n\ta.mu.RLock()\n\t_, ok := a.bots[\"robot_e2e_feishu_update\"]\n\ta.mu.RUnlock()\n\trequire.True(t, ok)\n\n\t// Apply again — should be no-op\n\ta.Apply(context.Background(), robot)\n\ta.mu.RLock()\n\tassert.Len(t, a.bots, 1)\n\ta.mu.RUnlock()\n\n\t// Remove\n\ta.Remove(context.Background(), \"robot_e2e_feishu_update\")\n\ta.mu.RLock()\n\t_, ok = a.bots[\"robot_e2e_feishu_update\"]\n\ta.mu.RUnlock()\n\tassert.False(t, ok, \"bot should be removed\")\n\tt.Log(\"OK Apply/Remove lifecycle verified\")\n}\n\n// TestE2E_Adapter_Dedup verifies deduplication works.\nfunc TestE2E_Adapter_Dedup(t *testing.T) {\n\ta := &Adapter{\n\t\tbots: make(map[string]*botEntry),\n\t\tappIdx: make(map[string]string),\n\t\tdedup: newDedupStore(),\n\t\tstopCh: make(chan struct{}),\n\t}\n\tdefer close(a.stopCh)\n\n\tkey := \"fs:test-robot:msg-12345\"\n\tassert.True(t, a.dedup.markSeen(key), \"first time should return true\")\n\tassert.False(t, a.dedup.markSeen(key), \"second time should return false (dedup)\")\n\tt.Log(\"OK dedup working correctly\")\n}\n\n// TestE2E_Adapter_HandleMessages verifies message handling through the adapter.\nfunc TestE2E_Adapter_HandleMessages(t *testing.T) {\n\tskipIfNoCreds(t)\n\n\ta := &Adapter{\n\t\tbots: make(map[string]*botEntry),\n\t\tappIdx: make(map[string]string),\n\t\tdedup: newDedupStore(),\n\t\tstopCh: make(chan struct{}),\n\t}\n\tdefer close(a.stopCh)\n\n\tentry := &botEntry{\n\t\trobotID: \"robot_e2e_feishu_handle\",\n\t\tappID: fsAppID,\n\t\tbot: fsapi.NewBot(fsAppID, fsAppSecret),\n\t}\n\n\tctx, cancel := context.WithTimeout(context.Background(), 10*time.Second)\n\tdefer cancel()\n\n\tcms := []*fsapi.ConvertedMessage{\n\t\t{\n\t\t\tMessageID: \"test_msg_1\",\n\t\t\tChatID: \"test_chat_1\",\n\t\t\tChatType: \"p2p\",\n\t\t\tSenderID: \"test_sender_1\",\n\t\t\tText: \"Hello from E2E test\",\n\t\t},\n\t}\n\n\t// This should not panic even without event bus running\n\ta.handleMessages(ctx, entry, cms)\n\n\t// Verify dedup: should be marked as seen\n\tassert.False(t, a.dedup.markSeen(\"fs:robot_e2e_feishu_handle:test_msg_1\"),\n\t\t\"message should be marked as seen after handleMessages\")\n\tt.Log(\"OK handleMessages processed 1 message\")\n}\n\n// TestE2E_Adapter_ApplyDisabled verifies Apply removes bot when disabled.\nfunc TestE2E_Adapter_ApplyDisabled(t *testing.T) {\n\ta := &Adapter{\n\t\tbots: make(map[string]*botEntry),\n\t\tappIdx: make(map[string]string),\n\t\tdedup: newDedupStore(),\n\t\tstopCh: make(chan struct{}),\n\t}\n\tdefer close(a.stopCh)\n\n\trobot := &robottypes.Robot{\n\t\tMemberID: \"robot_e2e_feishu_disabled\",\n\t\tTeamID: \"team_e2e_fs\",\n\t\tConfig: &robottypes.Config{\n\t\t\tIntegrations: &robottypes.Integrations{\n\t\t\t\tFeishu: &robottypes.FeishuConfig{\n\t\t\t\t\tEnabled: false,\n\t\t\t\t\tAppID: \"some_app\",\n\t\t\t\t\tAppSecret: \"some_secret\",\n\t\t\t\t},\n\t\t\t},\n\t\t},\n\t}\n\n\ta.Apply(context.Background(), robot)\n\ta.mu.RLock()\n\t_, ok := a.bots[\"robot_e2e_feishu_disabled\"]\n\ta.mu.RUnlock()\n\tassert.False(t, ok, \"disabled bot should not be registered\")\n\tt.Log(\"OK disabled config not registered\")\n}\n" }, { "path": "agent/robot/events/integrations/feishu/feishu.go", "content": "package feishu\n\nimport (\n\t\"context\"\n\t\"sync\"\n\n\t\"github.com/yaoapp/yao/agent/robot/logger\"\n\trobottypes \"github.com/yaoapp/yao/agent/robot/types\"\n\tfsapi \"github.com/yaoapp/yao/integrations/feishu\"\n)\n\nvar log = logger.New(\"feishu\")\n\n// Adapter implements the integrations.Adapter interface for Feishu (Lark).\n//\n// Architecture:\n// - One event subscription per registered bot via Feishu SDK's long-poll/callback mechanism\n// - One dedup cleaner goroutine removes expired keys every hour\ntype Adapter struct {\n\tmu sync.RWMutex\n\tbots map[string]*botEntry // robotID -> *botEntry\n\tappIdx map[string]string // appID -> robotID\n\tdedup *dedupStore\n\tstopCh chan struct{}\n}\n\n// botEntry holds the state for one robot's Feishu integration.\ntype botEntry struct {\n\trobotID string\n\tappID string\n\tbot *fsapi.Bot\n\tcancelFn context.CancelFunc // cancels the event subscription goroutine\n}\n\n// NewAdapter creates a new Feishu adapter.\nfunc NewAdapter() *Adapter {\n\ta := &Adapter{\n\t\tbots: make(map[string]*botEntry),\n\t\tappIdx: make(map[string]string),\n\t\tdedup: newDedupStore(),\n\t\tstopCh: make(chan struct{}),\n\t}\n\tgo a.dedup.cleaner(a.stopCh)\n\treturn a\n}\n\n// Apply is called by the Dispatcher when a robot config is created or updated.\nfunc (a *Adapter) Apply(ctx context.Context, robot *robottypes.Robot) {\n\tfsConf := extractConfig(robot)\n\tlog.Debug(\"Apply robot=%s fsConf=%v\", robot.MemberID, fsConf != nil)\n\n\tif fsConf == nil || !fsConf.Enabled || fsConf.AppID == \"\" || fsConf.AppSecret == \"\" {\n\t\ta.removeBot(robot.MemberID)\n\t\treturn\n\t}\n\n\ta.mu.Lock()\n\tdefer a.mu.Unlock()\n\n\tif existing, ok := a.bots[robot.MemberID]; ok {\n\t\tif existing.appID == fsConf.AppID {\n\t\t\treturn\n\t\t}\n\t\ta.removeBotLocked(robot.MemberID)\n\t}\n\n\tbot := fsapi.NewBot(fsConf.AppID, fsConf.AppSecret)\n\n\tstreamCtx, streamCancel := context.WithCancel(context.Background())\n\tentry := &botEntry{\n\t\trobotID: robot.MemberID,\n\t\tappID: fsConf.AppID,\n\t\tbot: bot,\n\t\tcancelFn: streamCancel,\n\t}\n\ta.bots[robot.MemberID] = entry\n\ta.appIdx[fsConf.AppID] = robot.MemberID\n\n\tgo a.eventLoop(streamCtx, entry)\n\n\tlog.Info(\"feishu adapter: registered robot=%s app=%s\", robot.MemberID, fsConf.AppID)\n}\n\n// Remove is called by the Dispatcher when a robot is deleted.\nfunc (a *Adapter) Remove(ctx context.Context, robotID string) {\n\ta.removeBot(robotID)\n}\n\n// Shutdown stops all event subscriptions and dedup cleaner.\nfunc (a *Adapter) Shutdown() {\n\tclose(a.stopCh)\n\ta.mu.Lock()\n\tfor _, entry := range a.bots {\n\t\tif entry.cancelFn != nil {\n\t\t\tentry.cancelFn()\n\t\t}\n\t}\n\ta.mu.Unlock()\n\tlog.Info(\"feishu adapter: shutdown complete\")\n}\n\nfunc (a *Adapter) removeBot(robotID string) {\n\ta.mu.Lock()\n\tdefer a.mu.Unlock()\n\ta.removeBotLocked(robotID)\n}\n\nfunc (a *Adapter) removeBotLocked(robotID string) {\n\tentry, ok := a.bots[robotID]\n\tif !ok {\n\t\treturn\n\t}\n\tif entry.cancelFn != nil {\n\t\tentry.cancelFn()\n\t}\n\tif entry.appID != \"\" {\n\t\tdelete(a.appIdx, entry.appID)\n\t}\n\tdelete(a.bots, robotID)\n\tlog.Info(\"feishu adapter: unregistered robot=%s\", robotID)\n}\n\nfunc (a *Adapter) resolveByAppID(appID string) (*botEntry, bool) {\n\ta.mu.RLock()\n\tdefer a.mu.RUnlock()\n\trobotID, ok := a.appIdx[appID]\n\tif !ok {\n\t\treturn nil, false\n\t}\n\tentry, ok := a.bots[robotID]\n\treturn entry, ok\n}\n\nfunc extractConfig(robot *robottypes.Robot) *robottypes.FeishuConfig {\n\tif robot.Config == nil || robot.Config.Integrations == nil {\n\t\treturn nil\n\t}\n\treturn robot.Config.Integrations.Feishu\n}\n" }, { "path": "agent/robot/events/integrations/feishu/message.go", "content": "package feishu\n\nimport (\n\t\"context\"\n\t\"fmt\"\n\t\"strings\"\n\n\tagentcontext \"github.com/yaoapp/yao/agent/context\"\n\tevents \"github.com/yaoapp/yao/agent/robot/events\"\n\t\"github.com/yaoapp/yao/event\"\n\tfsapi \"github.com/yaoapp/yao/integrations/feishu\"\n)\n\n// handleMessages processes a batch of Feishu messages for one chat.\nfunc (a *Adapter) handleMessages(ctx context.Context, entry *botEntry, cms []*fsapi.ConvertedMessage) {\n\tif len(cms) == 0 {\n\t\treturn\n\t}\n\n\tvar allParts []interface{}\n\tvar lastCM *fsapi.ConvertedMessage\n\n\tfor _, cm := range cms {\n\t\tif cm == nil {\n\t\t\tcontinue\n\t\t}\n\n\t\tdedupKey := fmt.Sprintf(\"fs:%s:%s\", entry.robotID, cm.MessageID)\n\t\tif !a.dedup.markSeen(dedupKey) {\n\t\t\tcontinue\n\t\t}\n\n\t\tparts := buildContentParts(cm)\n\t\tif len(parts) == 0 {\n\t\t\tcontinue\n\t\t}\n\n\t\tallParts = append(allParts, parts...)\n\t\tlastCM = cm\n\t}\n\n\tif len(allParts) == 0 || lastCM == nil {\n\t\treturn\n\t}\n\n\tcontent := mergeContentParts(allParts)\n\n\tmsgPayload := events.MessagePayload{\n\t\tRobotID: entry.robotID,\n\t\tMessages: []agentcontext.Message{\n\t\t\t{Role: agentcontext.RoleUser, Content: content},\n\t\t},\n\t\tMetadata: &events.MessageMetadata{\n\t\t\tChannel: \"feishu\",\n\t\t\tMessageID: lastCM.MessageID,\n\t\t\tAppID: entry.appID,\n\t\t\tChatID: lastCM.ChatID,\n\t\t\tSenderID: lastCM.SenderID,\n\t\t\tSenderName: lastCM.SenderName,\n\t\t\tLocale: events.NormalizeLocale(lastCM.LanguageCode),\n\t\t\tExtra: map[string]any{\n\t\t\t\t\"feishu_message_id\": lastCM.MessageID,\n\t\t\t},\n\t\t},\n\t}\n\n\tif _, err := event.Push(ctx, events.Message, msgPayload); err != nil {\n\t\tlog.Error(\"feishu adapter: event.Push robot.message failed robot=%s: %v\", entry.robotID, err)\n\t}\n}\n\nfunc buildContentParts(cm *fsapi.ConvertedMessage) []interface{} {\n\tvar parts []interface{}\n\n\tif cm.HasText() {\n\t\tparts = append(parts, map[string]interface{}{\n\t\t\t\"type\": \"text\",\n\t\t\t\"text\": cm.Text,\n\t\t})\n\t}\n\n\tfor _, mi := range cm.MediaItems {\n\t\tif mi.Wrapper == \"\" {\n\t\t\tcontinue\n\t\t}\n\t\tparts = append(parts, map[string]interface{}{\n\t\t\t\"type\": \"file\",\n\t\t\t\"file_url\": mi.Wrapper,\n\t\t\t\"mime_type\": mi.MimeType,\n\t\t\t\"file_name\": mi.FileName,\n\t\t})\n\t}\n\n\treturn parts\n}\n\nfunc mergeContentParts(parts []interface{}) interface{} {\n\tallText := true\n\tfor _, p := range parts {\n\t\tm, ok := p.(map[string]interface{})\n\t\tif !ok || m[\"type\"] != \"text\" {\n\t\t\tallText = false\n\t\t\tbreak\n\t\t}\n\t}\n\n\tif allText {\n\t\tvar buf strings.Builder\n\t\tfor i, p := range parts {\n\t\t\tif i > 0 {\n\t\t\t\tbuf.WriteString(\"\\n\")\n\t\t\t}\n\t\t\tm := p.(map[string]interface{})\n\t\t\tbuf.WriteString(m[\"text\"].(string))\n\t\t}\n\t\treturn buf.String()\n\t}\n\n\treturn parts\n}\n" }, { "path": "agent/robot/events/integrations/feishu/reply.go", "content": "package feishu\n\nimport (\n\t\"context\"\n\t\"fmt\"\n\t\"strings\"\n\n\tagentcontext \"github.com/yaoapp/yao/agent/context\"\n\tevents \"github.com/yaoapp/yao/agent/robot/events\"\n\tfsapi \"github.com/yaoapp/yao/integrations/feishu\"\n)\n\n// Reply sends the assistant message back to the originating Feishu chat.\nfunc (a *Adapter) Reply(ctx context.Context, msg *agentcontext.Message, metadata *events.MessageMetadata) error {\n\tif msg == nil || metadata == nil {\n\t\treturn fmt.Errorf(\"nil message or metadata\")\n\t}\n\n\tentry := a.resolveByChat(metadata)\n\tif entry == nil {\n\t\treturn fmt.Errorf(\"no bot registered for feishu metadata (appID=%s)\", metadata.AppID)\n\t}\n\n\tvar replyToMsgID string\n\tif metadata.Extra != nil {\n\t\tif v, ok := metadata.Extra[\"feishu_message_id\"]; ok {\n\t\t\tif s, ok := v.(string); ok {\n\t\t\t\treplyToMsgID = s\n\t\t\t}\n\t\t}\n\t}\n\n\treturn a.sendContent(ctx, entry, metadata.ChatID, replyToMsgID, msg.Content)\n}\n\nfunc (a *Adapter) sendContent(ctx context.Context, entry *botEntry, chatID, replyToMsgID string, content interface{}) error {\n\tswitch c := content.(type) {\n\tcase string:\n\t\tif strings.TrimSpace(c) == \"\" {\n\t\t\treturn nil\n\t\t}\n\t\treturn a.sendMarkdown(ctx, entry, chatID, replyToMsgID, c)\n\n\tcase []interface{}:\n\t\treturn a.sendParts(ctx, entry, chatID, replyToMsgID, c)\n\n\tdefault:\n\t\tparts, ok := toContentParts(content)\n\t\tif ok {\n\t\t\treturn a.sendPartsTyped(ctx, entry, chatID, replyToMsgID, parts)\n\t\t}\n\t\treturn a.sendMarkdown(ctx, entry, chatID, replyToMsgID, fmt.Sprintf(\"%v\", content))\n\t}\n}\n\n// sendMarkdown converts standard Markdown to Feishu lark_md and sends as an interactive card.\nfunc (a *Adapter) sendMarkdown(ctx context.Context, entry *botEntry, chatID, replyToMsgID, text string) error {\n\tformatted := fsapi.FormatFeishuMarkdown(text)\n\tif replyToMsgID != \"\" {\n\t\t_, err := entry.bot.ReplyCardMessage(ctx, replyToMsgID, formatted)\n\t\treturn err\n\t}\n\t_, err := entry.bot.SendCardMessage(ctx, chatID, formatted)\n\treturn err\n}\n\nfunc (a *Adapter) sendParts(ctx context.Context, entry *botEntry, chatID, replyToMsgID string, parts []interface{}) error {\n\tvar textBuf strings.Builder\n\tfor _, part := range parts {\n\t\tm, ok := part.(map[string]interface{})\n\t\tif !ok {\n\t\t\tcontinue\n\t\t}\n\t\tpartType, _ := m[\"type\"].(string)\n\t\tswitch partType {\n\t\tcase \"text\":\n\t\t\tif text, ok := m[\"text\"].(string); ok {\n\t\t\t\ttextBuf.WriteString(text)\n\t\t\t}\n\t\tcase \"image_url\":\n\t\t\tif err := a.flushText(ctx, entry, chatID, replyToMsgID, &textBuf); err != nil {\n\t\t\t\treturn err\n\t\t\t}\n\t\t\tif imgMap, ok := m[\"image_url\"].(map[string]interface{}); ok {\n\t\t\t\tif url, ok := imgMap[\"url\"].(string); ok {\n\t\t\t\t\tif err := sendImageOrWrapper(ctx, entry, chatID, url, \"\"); err != nil {\n\t\t\t\t\t\tlog.Error(\"feishu reply: send image: %v\", err)\n\t\t\t\t\t}\n\t\t\t\t}\n\t\t\t}\n\t\tcase \"file\":\n\t\t\tif err := a.flushText(ctx, entry, chatID, replyToMsgID, &textBuf); err != nil {\n\t\t\t\treturn err\n\t\t\t}\n\t\t\tfileURL, _ := m[\"file_url\"].(string)\n\t\t\tif fileURL == \"\" {\n\t\t\t\tif fileMap, ok := m[\"file\"].(map[string]interface{}); ok {\n\t\t\t\t\tfileURL, _ = fileMap[\"url\"].(string)\n\t\t\t\t}\n\t\t\t}\n\t\t\tif fileURL != \"\" {\n\t\t\t\tif err := sendFileOrWrapper(ctx, entry, chatID, fileURL, \"\"); err != nil {\n\t\t\t\t\tlog.Error(\"feishu reply: send file: %v\", err)\n\t\t\t\t}\n\t\t\t}\n\t\t}\n\t}\n\treturn a.flushText(ctx, entry, chatID, replyToMsgID, &textBuf)\n}\n\nfunc (a *Adapter) sendPartsTyped(ctx context.Context, entry *botEntry, chatID, replyToMsgID string, parts []agentcontext.ContentPart) error {\n\tvar textBuf strings.Builder\n\tfor _, part := range parts {\n\t\tswitch part.Type {\n\t\tcase agentcontext.ContentText:\n\t\t\ttextBuf.WriteString(part.Text)\n\t\tcase agentcontext.ContentImageURL:\n\t\t\tif err := a.flushText(ctx, entry, chatID, replyToMsgID, &textBuf); err != nil {\n\t\t\t\treturn err\n\t\t\t}\n\t\t\tif part.ImageURL != nil {\n\t\t\t\tif err := sendImageOrWrapper(ctx, entry, chatID, part.ImageURL.URL, \"\"); err != nil {\n\t\t\t\t\tlog.Error(\"feishu reply: send image: %v\", err)\n\t\t\t\t}\n\t\t\t}\n\t\tcase agentcontext.ContentFile:\n\t\t\tif err := a.flushText(ctx, entry, chatID, replyToMsgID, &textBuf); err != nil {\n\t\t\t\treturn err\n\t\t\t}\n\t\t\tif part.File != nil {\n\t\t\t\tif err := sendFileOrWrapper(ctx, entry, chatID, part.File.URL, part.File.Filename); err != nil {\n\t\t\t\t\tlog.Error(\"feishu reply: send file: %v\", err)\n\t\t\t\t}\n\t\t\t}\n\t\t}\n\t}\n\treturn a.flushText(ctx, entry, chatID, replyToMsgID, &textBuf)\n}\n\nfunc (a *Adapter) flushText(ctx context.Context, entry *botEntry, chatID, replyToMsgID string, buf *strings.Builder) error {\n\tif buf.Len() == 0 {\n\t\treturn nil\n\t}\n\ttext := buf.String()\n\tbuf.Reset()\n\treturn a.sendMarkdown(ctx, entry, chatID, replyToMsgID, text)\n}\n\nfunc sendImageOrWrapper(ctx context.Context, entry *botEntry, chatID, url, caption string) error {\n\tif isWrapper(url) {\n\t\treturn entry.bot.SendImageFromWrapper(ctx, chatID, url, caption)\n\t}\n\tif strings.HasPrefix(url, \"http\") {\n\t\ttext := url\n\t\tif caption != \"\" {\n\t\t\ttext = caption + \"\\n\" + url\n\t\t}\n\t\t_, err := entry.bot.SendTextMessage(ctx, chatID, text)\n\t\treturn err\n\t}\n\treturn fmt.Errorf(\"unsupported image URL scheme: %s\", url)\n}\n\nfunc sendFileOrWrapper(ctx context.Context, entry *botEntry, chatID, url, caption string) error {\n\tif isWrapper(url) {\n\t\treturn entry.bot.SendFileFromWrapper(ctx, chatID, url, caption)\n\t}\n\tif strings.HasPrefix(url, \"http\") {\n\t\ttext := url\n\t\tif caption != \"\" {\n\t\t\ttext = caption + \"\\n\" + url\n\t\t}\n\t\t_, err := entry.bot.SendTextMessage(ctx, chatID, text)\n\t\treturn err\n\t}\n\treturn fmt.Errorf(\"unsupported file URL scheme: %s\", url)\n}\n\nfunc isWrapper(url string) bool {\n\treturn strings.Contains(url, \"://\") && !strings.HasPrefix(url, \"http\")\n}\n\nfunc toContentParts(content interface{}) ([]agentcontext.ContentPart, bool) {\n\tparts, ok := content.([]agentcontext.ContentPart)\n\treturn parts, ok\n}\n\nfunc (a *Adapter) resolveByChat(metadata *events.MessageMetadata) *botEntry {\n\tif metadata.AppID != \"\" {\n\t\tif entry, ok := a.resolveByAppID(metadata.AppID); ok {\n\t\t\treturn entry\n\t\t}\n\t}\n\ta.mu.RLock()\n\tdefer a.mu.RUnlock()\n\tfor _, entry := range a.bots {\n\t\treturn entry\n\t}\n\treturn nil\n}\n" }, { "path": "agent/robot/events/integrations/feishu/stream.go", "content": "package feishu\n\nimport (\n\t\"context\"\n\t\"time\"\n\n\tlarkcore \"github.com/larksuite/oapi-sdk-go/v3/core\"\n\t\"github.com/larksuite/oapi-sdk-go/v3/event/dispatcher\"\n\tlarkim \"github.com/larksuite/oapi-sdk-go/v3/service/im/v1\"\n\tlarkws \"github.com/larksuite/oapi-sdk-go/v3/ws\"\n\tfsapi \"github.com/yaoapp/yao/integrations/feishu\"\n)\n\nconst reconnectDelay = 5 * time.Second\n\n// eventLoop starts the Feishu WebSocket event subscription for a single bot.\n// It automatically reconnects on failure.\nfunc (a *Adapter) eventLoop(ctx context.Context, entry *botEntry) {\n\tlog.Info(\"feishu eventLoop started robot=%s app=%s\", entry.robotID, entry.appID)\n\n\tfor {\n\t\tselect {\n\t\tcase <-ctx.Done():\n\t\t\tlog.Info(\"feishu eventLoop stopped robot=%s\", entry.robotID)\n\t\t\treturn\n\t\tcase <-a.stopCh:\n\t\t\treturn\n\t\tdefault:\n\t\t}\n\n\t\terr := a.runWSClient(ctx, entry)\n\t\tif err != nil {\n\t\t\tlog.Error(\"feishu ws disconnected robot=%s: %v, reconnecting in %s\", entry.robotID, err, reconnectDelay)\n\t\t}\n\n\t\tselect {\n\t\tcase <-ctx.Done():\n\t\t\treturn\n\t\tcase <-a.stopCh:\n\t\t\treturn\n\t\tcase <-time.After(reconnectDelay):\n\t\t}\n\t}\n}\n\nfunc (a *Adapter) runWSClient(ctx context.Context, entry *botEntry) error {\n\teventHandler := dispatcher.NewEventDispatcher(\"\", \"\")\n\teventHandler.OnP2MessageReceiveV1(func(ctx context.Context, event *larkim.P2MessageReceiveV1) error {\n\t\treturn a.onMessageReceive(ctx, entry, event)\n\t})\n\n\tcli := larkws.NewClient(entry.bot.AppID(), entry.bot.AppSecret(),\n\t\tlarkws.WithEventHandler(eventHandler),\n\t\tlarkws.WithLogLevel(larkcore.LogLevelWarn),\n\t)\n\n\terrCh := make(chan error, 1)\n\tgo func() {\n\t\terrCh <- cli.Start(ctx)\n\t}()\n\n\tselect {\n\tcase <-ctx.Done():\n\t\treturn ctx.Err()\n\tcase <-a.stopCh:\n\t\treturn nil\n\tcase err := <-errCh:\n\t\treturn err\n\t}\n}\n\nfunc (a *Adapter) onMessageReceive(ctx context.Context, entry *botEntry, event *larkim.P2MessageReceiveV1) error {\n\tif event == nil || event.Event == nil || event.Event.Message == nil {\n\t\treturn nil\n\t}\n\n\tmsg := event.Event.Message\n\tsender := event.Event.Sender\n\n\tmsgType := derefStr(msg.MessageType)\n\tcontent := derefStr(msg.Content)\n\tmessageID := derefStr(msg.MessageId)\n\tchatID := derefStr(msg.ChatId)\n\tchatType := derefStr(msg.ChatType)\n\n\ttext, media := fsapi.ParseMessageContent(msgType, content)\n\n\tcm := &fsapi.ConvertedMessage{\n\t\tMessageID: messageID,\n\t\tChatID: chatID,\n\t\tChatType: chatType,\n\t\tText: text,\n\t\tMediaItems: media,\n\t\tEventID: event.EventV2Base.Header.EventID,\n\t\tLanguageCode: \"zh\",\n\t}\n\n\tif sender != nil && sender.SenderId != nil {\n\t\tcm.SenderID = derefStr(sender.SenderId.OpenId)\n\t}\n\n\tif cm.HasMedia() {\n\t\tgroups := []string{\"feishu\", entry.robotID}\n\t\tentry.bot.ResolveMedia(ctx, cm, groups)\n\t}\n\n\ta.handleMessages(ctx, entry, []*fsapi.ConvertedMessage{cm})\n\treturn nil\n}\n\nfunc derefStr(s *string) string {\n\tif s == nil {\n\t\treturn \"\"\n\t}\n\treturn *s\n}\n" }, { "path": "agent/robot/events/integrations/telegram/dedup.go", "content": "package telegram\n\nimport (\n\t\"sync\"\n\t\"time\"\n)\n\nconst (\n\tdedupTTL = 24 * time.Hour\n\tdedupCleanInterval = time.Hour\n)\n\n// dedupStore is a lightweight in-memory deduplication store with TTL.\n// Used for message-level dedup (same update_id won't be processed twice).\ntype dedupStore struct {\n\tm sync.Map // key -> int64 (unix timestamp)\n}\n\nfunc newDedupStore() *dedupStore {\n\treturn &dedupStore{}\n}\n\n// markSeen returns true if this is the first time the key is seen.\nfunc (d *dedupStore) markSeen(key string) bool {\n\tnow := time.Now().Unix()\n\t_, loaded := d.m.LoadOrStore(key, now)\n\treturn !loaded\n}\n\n// cleaner periodically removes expired entries. Runs until stopCh is closed.\nfunc (d *dedupStore) cleaner(stopCh <-chan struct{}) {\n\tticker := time.NewTicker(dedupCleanInterval)\n\tdefer ticker.Stop()\n\tfor {\n\t\tselect {\n\t\tcase <-stopCh:\n\t\t\treturn\n\t\tcase <-ticker.C:\n\t\t\tcutoff := time.Now().Add(-dedupTTL).Unix()\n\t\t\td.m.Range(func(key, value any) bool {\n\t\t\t\tif ts, ok := value.(int64); ok && ts < cutoff {\n\t\t\t\t\td.m.Delete(key)\n\t\t\t\t}\n\t\t\t\treturn true\n\t\t\t})\n\t\t}\n\t}\n}\n" }, { "path": "agent/robot/events/integrations/telegram/e2e_test.go", "content": "package telegram\n\nimport (\n\t\"context\"\n\t\"encoding/json\"\n\t\"fmt\"\n\t\"os\"\n\t\"testing\"\n\t\"time\"\n\n\t\"github.com/stretchr/testify/assert\"\n\t\"github.com/stretchr/testify/require\"\n\t\"github.com/yaoapp/gou/model\"\n\t\"github.com/yaoapp/xun/capsule\"\n\trobottypes \"github.com/yaoapp/yao/agent/robot/types\"\n\t\"github.com/yaoapp/yao/agent/testutils\"\n\t\"github.com/yaoapp/yao/event\"\n\ttgapi \"github.com/yaoapp/yao/integrations/telegram\"\n)\n\nvar (\n\ttgBotToken string\n\ttgHost string\n)\n\nfunc TestMain(m *testing.M) {\n\ttgBotToken = os.Getenv(\"TELEGRAM_TEST_BOT_TOKEN\")\n\ttgHost = os.Getenv(\"TELEGRAM_TEST_HOST\")\n\tos.Exit(m.Run())\n}\n\nfunc skipIfNoToken(t *testing.T) {\n\tt.Helper()\n\tif tgBotToken == \"\" {\n\t\tt.Skip(\"TELEGRAM_TEST_BOT_TOKEN not set\")\n\t}\n}\n\nfunc newTestBot() *tgapi.Bot {\n\tvar opts []tgapi.BotOption\n\tif tgHost != \"\" {\n\t\topts = append(opts, tgapi.WithAPIBase(tgHost))\n\t}\n\treturn tgapi.NewBot(tgBotToken, \"\", opts...)\n}\n\n// confirmPendingUpdates checks if there are pending updates from previous seeds.\nfunc confirmPendingUpdates(t *testing.T) []*tgapi.ConvertedMessage {\n\tt.Helper()\n\tb := newTestBot()\n\tctx, cancel := context.WithTimeout(context.Background(), 15*time.Second)\n\tdefer cancel()\n\n\tmsgs, err := b.GetUpdates(ctx, 0, 5, nil)\n\trequire.NoError(t, err)\n\treturn msgs\n}\n\n// TestE2E_Adapter_Apply verifies that Apply correctly registers a bot.\nfunc TestE2E_Adapter_Apply(t *testing.T) {\n\tskipIfNoToken(t)\n\n\ta := &Adapter{\n\t\tbots: make(map[string]*botEntry),\n\t\tappIdx: make(map[string]string),\n\t\tdedup: newDedupStore(),\n\t\tstopCh: make(chan struct{}),\n\t}\n\tdefer close(a.stopCh)\n\n\trobot := &robottypes.Robot{\n\t\tMemberID: \"robot_e2e_tg_adapter\",\n\t\tTeamID: \"team_e2e_tg\",\n\t\tConfig: &robottypes.Config{\n\t\t\tIntegrations: &robottypes.Integrations{\n\t\t\t\tTelegram: &robottypes.TelegramConfig{\n\t\t\t\t\tEnabled: true,\n\t\t\t\t\tBotToken: tgBotToken,\n\t\t\t\t\tHost: tgHost,\n\t\t\t\t\tAppID: \"e2e-test-app\",\n\t\t\t\t},\n\t\t\t},\n\t\t},\n\t}\n\n\ta.Apply(context.Background(), robot)\n\n\ta.mu.RLock()\n\tentry, ok := a.bots[\"robot_e2e_tg_adapter\"]\n\ta.mu.RUnlock()\n\n\trequire.True(t, ok, \"bot should be registered\")\n\tassert.Equal(t, tgBotToken, entry.bot.Token())\n\tassert.Equal(t, \"e2e-test-app\", entry.appID)\n\n\t// Verify GetMe works through the registered bot\n\tctx, cancel := context.WithTimeout(context.Background(), 10*time.Second)\n\tdefer cancel()\n\tme, err := entry.bot.GetMe(ctx)\n\trequire.NoError(t, err)\n\tassert.True(t, me.IsBot)\n\tt.Logf(\"OK Apply: bot registered id=%d username=%s\", me.ID, me.Username)\n\n\t// Verify ResolveBot\n\tresolved := a.ResolveBot(\"e2e-test-app\")\n\trequire.NotNil(t, resolved)\n\tassert.Equal(t, tgBotToken, resolved.Token())\n}\n\n// TestE2E_Adapter_Apply_Update verifies that Apply with a different token replaces the bot.\nfunc TestE2E_Adapter_Apply_Update(t *testing.T) {\n\tskipIfNoToken(t)\n\n\ta := &Adapter{\n\t\tbots: make(map[string]*botEntry),\n\t\tappIdx: make(map[string]string),\n\t\tdedup: newDedupStore(),\n\t\tstopCh: make(chan struct{}),\n\t}\n\tdefer close(a.stopCh)\n\n\trobot := &robottypes.Robot{\n\t\tMemberID: \"robot_e2e_tg_update\",\n\t\tTeamID: \"team_e2e_tg\",\n\t\tConfig: &robottypes.Config{\n\t\t\tIntegrations: &robottypes.Integrations{\n\t\t\t\tTelegram: &robottypes.TelegramConfig{\n\t\t\t\t\tEnabled: true,\n\t\t\t\t\tBotToken: tgBotToken,\n\t\t\t\t\tHost: tgHost,\n\t\t\t\t},\n\t\t\t},\n\t\t},\n\t}\n\n\ta.Apply(context.Background(), robot)\n\ta.mu.RLock()\n\t_, ok := a.bots[\"robot_e2e_tg_update\"]\n\ta.mu.RUnlock()\n\trequire.True(t, ok)\n\n\t// Apply again with same token — should be a no-op\n\ta.Apply(context.Background(), robot)\n\ta.mu.RLock()\n\tassert.Len(t, a.bots, 1)\n\ta.mu.RUnlock()\n\n\t// Remove\n\ta.Remove(context.Background(), \"robot_e2e_tg_update\")\n\ta.mu.RLock()\n\t_, ok = a.bots[\"robot_e2e_tg_update\"]\n\ta.mu.RUnlock()\n\tassert.False(t, ok, \"bot should be removed\")\n\tt.Log(\"OK Apply/Remove lifecycle verified\")\n}\n\n// TestE2E_Adapter_PollAll verifies that pollAll fetches updates from Telegram\n// and processes them through handleMessages.\nfunc TestE2E_Adapter_PollAll(t *testing.T) {\n\tskipIfNoToken(t)\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\tpending := confirmPendingUpdates(t)\n\tif len(pending) == 0 {\n\t\tt.Skip(\"no pending updates; run integrations/telegram seed first\")\n\t}\n\tt.Logf(\"found %d pending updates\", len(pending))\n\n\t// Create adapter WITHOUT auto-starting pollLoop\n\ta := &Adapter{\n\t\tbots: make(map[string]*botEntry),\n\t\tappIdx: make(map[string]string),\n\t\tdedup: newDedupStore(),\n\t\tstopCh: make(chan struct{}),\n\t}\n\tdefer close(a.stopCh)\n\n\tmemberID := \"robot_e2e_tg_poll\"\n\tsetupTestRobot(t, memberID)\n\tdefer cleanupTestRobots(t)\n\n\tvar opts []tgapi.BotOption\n\tif tgHost != \"\" {\n\t\topts = append(opts, tgapi.WithAPIBase(tgHost))\n\t}\n\ta.bots[memberID] = &botEntry{\n\t\trobotID: memberID,\n\t\tappID: \"e2e-poll-app\",\n\t\tbot: tgapi.NewBot(tgBotToken, \"\", opts...),\n\t}\n\n\t// Start event bus so event.Push works\n\tif err := event.Start(); err != nil && err != event.ErrAlreadyStart {\n\t\tt.Fatalf(\"event.Start: %v\", err)\n\t}\n\tdefer func() { _ = event.Stop(context.Background()) }()\n\n\t// Manually trigger one poll cycle\n\ta.pollAll()\n\n\t// Verify offset advanced (meaning updates were processed)\n\ta.mu.RLock()\n\tentry := a.bots[memberID]\n\ta.mu.RUnlock()\n\tassert.Greater(t, entry.offset, int64(0), \"offset should have advanced after processing updates\")\n\tt.Logf(\"OK pollAll: offset advanced to %d\", entry.offset)\n}\n\n// TestE2E_Adapter_Dedup verifies that duplicate messages are not processed twice.\nfunc TestE2E_Adapter_Dedup(t *testing.T) {\n\tskipIfNoToken(t)\n\n\ta := &Adapter{\n\t\tbots: make(map[string]*botEntry),\n\t\tappIdx: make(map[string]string),\n\t\tdedup: newDedupStore(),\n\t\tstopCh: make(chan struct{}),\n\t}\n\tdefer close(a.stopCh)\n\n\tkey := \"tg:test-robot:12345\"\n\tassert.True(t, a.dedup.markSeen(key), \"first time should return true\")\n\tassert.False(t, a.dedup.markSeen(key), \"second time should return false (dedup)\")\n\tt.Log(\"OK dedup working correctly\")\n}\n\n// TestE2E_Adapter_HandleMessages_Integration verifies the full flow:\n// GetUpdates → ConvertedMessage → handleMessages → event.Push\nfunc TestE2E_Adapter_HandleMessages_Integration(t *testing.T) {\n\tskipIfNoToken(t)\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\tb := newTestBot()\n\tctx, cancel := context.WithTimeout(context.Background(), 15*time.Second)\n\tdefer cancel()\n\n\tmsgs, err := b.GetUpdates(ctx, 0, 5, nil)\n\trequire.NoError(t, err)\n\tif len(msgs) == 0 {\n\t\tt.Skip(\"no pending updates; run integrations/telegram seed first\")\n\t}\n\n\tmemberID := \"robot_e2e_tg_handle\"\n\tsetupTestRobot(t, memberID)\n\tdefer cleanupTestRobots(t)\n\n\tif err := event.Start(); err != nil && err != event.ErrAlreadyStart {\n\t\tt.Fatalf(\"event.Start: %v\", err)\n\t}\n\tdefer func() { _ = event.Stop(context.Background()) }()\n\n\ta := &Adapter{\n\t\tbots: make(map[string]*botEntry),\n\t\tappIdx: make(map[string]string),\n\t\tdedup: newDedupStore(),\n\t\tstopCh: make(chan struct{}),\n\t}\n\tdefer close(a.stopCh)\n\n\tentry := &botEntry{\n\t\trobotID: memberID,\n\t\tappID: \"e2e-handle-app\",\n\t\tbot: b,\n\t}\n\n\t// Group all messages by chatID (like pollAll does) and process each group\n\tgrouped := groupByChatID(msgs)\n\tfor chatID, chatMsgs := range grouped {\n\t\tt.Logf(\"processing chat=%d messages=%d\", chatID, len(chatMsgs))\n\t\tfor _, cm := range chatMsgs {\n\t\t\tt.Logf(\" update_id=%d msg_id=%d text=%q media=%d\",\n\t\t\t\tcm.UpdateID, cm.MessageID, truncate(cm.Text, 40), len(cm.MediaItems))\n\t\t}\n\t\ta.handleMessages(ctx, entry, chatMsgs)\n\t}\n\n\t// Verify dedup: all updates should be marked as seen\n\tcm := msgs[0]\n\tassert.False(t, a.dedup.markSeen(fmt.Sprintf(\"tg:%s:%d\", memberID, cm.UpdateID)),\n\t\t\"update should be marked as seen after handleMessages\")\n\n\t// Second call with same messages should be fully deduped (no-op)\n\ta.handleMessages(ctx, entry, msgs)\n\tt.Logf(\"OK handleMessages processed %d updates across %d chats\", len(msgs), len(grouped))\n}\n\n// ==================== Helpers ====================\n\nfunc setupTestRobot(t *testing.T, memberID string) {\n\tt.Helper()\n\tm := model.Select(\"__yao.member\")\n\tif m == nil {\n\t\tt.Skip(\"__yao.member model not loaded\")\n\t}\n\tqb := capsule.Query()\n\n\trobotConfig := map[string]interface{}{\n\t\t\"identity\": map[string]interface{}{\n\t\t\t\"role\": \"Telegram E2E Test Robot\",\n\t\t\t\"duties\": []string{\"Process Telegram messages\"},\n\t\t},\n\t\t\"integrations\": map[string]interface{}{\n\t\t\t\"telegram\": map[string]interface{}{\n\t\t\t\t\"enabled\": true,\n\t\t\t\t\"bot_token\": tgBotToken,\n\t\t\t\t\"host\": tgHost,\n\t\t\t\t\"app_id\": \"e2e-tg-app-\" + memberID,\n\t\t\t},\n\t\t},\n\t\t\"resources\": map[string]interface{}{\n\t\t\t\"phases\": map[string]interface{}{\n\t\t\t\t\"host\": \"robot.host\",\n\t\t\t},\n\t\t},\n\t}\n\tconfigJSON, _ := json.Marshal(robotConfig)\n\n\terr := qb.Table(m.MetaData.Table.Name).Insert([]map[string]interface{}{\n\t\t{\n\t\t\t\"member_id\": memberID,\n\t\t\t\"team_id\": \"team_e2e_tg\",\n\t\t\t\"member_type\": \"robot\",\n\t\t\t\"display_name\": \"E2E TG Adapter Test \" + memberID,\n\t\t\t\"system_prompt\": \"You are a test robot for Telegram adapter E2E testing.\",\n\t\t\t\"status\": \"active\",\n\t\t\t\"role_id\": \"member\",\n\t\t\t\"autonomous_mode\": false,\n\t\t\t\"robot_status\": \"idle\",\n\t\t\t\"robot_config\": string(configJSON),\n\t\t},\n\t})\n\tif err != nil {\n\t\tt.Fatalf(\"setup robot %s: %v\", memberID, err)\n\t}\n}\n\nfunc cleanupTestRobots(t *testing.T) {\n\tt.Helper()\n\tm := model.Select(\"__yao.member\")\n\tif m == nil {\n\t\treturn\n\t}\n\tqb := capsule.Query()\n\t_, _ = qb.Table(m.MetaData.Table.Name).Where(\"member_id\", \"like\", \"robot_e2e_tg%\").Delete()\n}\n\nfunc truncate(s string, n int) string {\n\tif len(s) <= n {\n\t\treturn s\n\t}\n\treturn s[:n] + \"...\"\n}\n" }, { "path": "agent/robot/events/integrations/telegram/message.go", "content": "package telegram\n\nimport (\n\t\"context\"\n\t\"fmt\"\n\t\"strconv\"\n\t\"strings\"\n\n\tagentcontext \"github.com/yaoapp/yao/agent/context\"\n\tevents \"github.com/yaoapp/yao/agent/robot/events\"\n\t\"github.com/yaoapp/yao/event\"\n\ttgapi \"github.com/yaoapp/yao/integrations/telegram\"\n)\n\n// handleMessages builds a single event payload from a batch of ConvertedMessages\n// belonging to the same chat. Consecutive user messages are merged into one to\n// keep the messages array clean for the LLM.\nfunc (a *Adapter) handleMessages(ctx context.Context, entry *botEntry, cms []*tgapi.ConvertedMessage) {\n\tif len(cms) == 0 {\n\t\treturn\n\t}\n\n\tvar allParts []interface{}\n\tvar lastCM *tgapi.ConvertedMessage\n\n\tfor _, cm := range cms {\n\t\tif cm == nil {\n\t\t\tcontinue\n\t\t}\n\n\t\tif isBotCommand(cm) {\n\t\t\tcontinue\n\t\t}\n\n\t\tdedupKey := fmt.Sprintf(\"tg:%s:%d\", entry.robotID, cm.UpdateID)\n\t\tif !a.dedup.markSeen(dedupKey) {\n\t\t\tcontinue\n\t\t}\n\n\t\tparts := buildContentParts(cm)\n\t\tif len(parts) == 0 {\n\t\t\tcontinue\n\t\t}\n\n\t\tallParts = append(allParts, parts...)\n\t\tlastCM = cm\n\t}\n\n\tif len(allParts) == 0 || lastCM == nil {\n\t\treturn\n\t}\n\n\tcontent := mergeContentParts(allParts)\n\n\tmsgPayload := events.MessagePayload{\n\t\tRobotID: entry.robotID,\n\t\tMessages: []agentcontext.Message{\n\t\t\t{Role: agentcontext.RoleUser, Content: content},\n\t\t},\n\t\tMetadata: &events.MessageMetadata{\n\t\t\tChannel: \"telegram\",\n\t\t\tMessageID: strconv.FormatInt(lastCM.UpdateID, 10),\n\t\t\tAppID: entry.appID,\n\t\t\tChatID: strconv.FormatInt(lastCM.ChatID, 10),\n\t\t\tSenderID: strconv.FormatInt(lastCM.SenderID, 10),\n\t\t\tSenderName: lastCM.SenderName,\n\t\t\tLocale: events.NormalizeLocale(lastCM.LanguageCode),\n\t\t\tExtra: map[string]any{\n\t\t\t\t\"tg_message_id\": lastCM.MessageID,\n\t\t\t},\n\t\t},\n\t}\n\n\tif _, err := event.Push(ctx, events.Message, msgPayload); err != nil {\n\t\tlog.Error(\"telegram adapter: event.Push robot.message failed robot=%s: %v\", entry.robotID, err)\n\t}\n}\n\n// buildContentParts extracts content parts from a single ConvertedMessage.\nfunc buildContentParts(cm *tgapi.ConvertedMessage) []interface{} {\n\tvar parts []interface{}\n\n\tif cm.HasText() {\n\t\tparts = append(parts, map[string]interface{}{\n\t\t\t\"type\": \"text\",\n\t\t\t\"text\": cm.Text,\n\t\t})\n\t}\n\n\tfor _, mi := range cm.MediaItems {\n\t\tif mi.Wrapper == \"\" {\n\t\t\tcontinue\n\t\t}\n\t\tparts = append(parts, map[string]interface{}{\n\t\t\t\"type\": \"file\",\n\t\t\t\"file_url\": mi.Wrapper,\n\t\t\t\"mime_type\": mi.MimeType,\n\t\t\t\"file_name\": mi.FileName,\n\t\t})\n\t}\n\n\treturn parts\n}\n\n// mergeContentParts merges collected parts into a single content value.\n// If all parts are text-only, they are joined with newlines into a plain string.\n// Otherwise the full parts array is returned.\nfunc mergeContentParts(parts []interface{}) interface{} {\n\tallText := true\n\tfor _, p := range parts {\n\t\tm, ok := p.(map[string]interface{})\n\t\tif !ok || m[\"type\"] != \"text\" {\n\t\t\tallText = false\n\t\t\tbreak\n\t\t}\n\t}\n\n\tif allText {\n\t\tvar buf strings.Builder\n\t\tfor i, p := range parts {\n\t\t\tif i > 0 {\n\t\t\t\tbuf.WriteString(\"\\n\")\n\t\t\t}\n\t\t\tm := p.(map[string]interface{})\n\t\t\tbuf.WriteString(m[\"text\"].(string))\n\t\t}\n\t\treturn buf.String()\n\t}\n\n\treturn parts\n}\n\n// isBotCommand returns true if the message is a Telegram bot command (text starting with \"/\").\nfunc isBotCommand(cm *tgapi.ConvertedMessage) bool {\n\treturn !cm.HasMedia() && strings.HasPrefix(strings.TrimSpace(cm.Text), \"/\")\n}\n" }, { "path": "agent/robot/events/integrations/telegram/polling.go", "content": "package telegram\n\nimport (\n\t\"context\"\n\t\"time\"\n\n\ttgapi \"github.com/yaoapp/yao/integrations/telegram\"\n)\n\nconst (\n\tpollInterval = 60 * time.Second\n\tpollTimeout = 30 // seconds, Telegram long-polling timeout per request\n)\n\n// pollLoop runs a single goroutine that iterates all registered bots\n// every pollInterval, calling getUpdates for each one sequentially.\nfunc (a *Adapter) pollLoop() {\n\tlog.Info(\"pollLoop started, interval=%s\", pollInterval)\n\ta.pollAll()\n\n\tticker := time.NewTicker(pollInterval)\n\tdefer ticker.Stop()\n\n\tfor {\n\t\tselect {\n\t\tcase <-a.stopCh:\n\t\t\tlog.Info(\"pollLoop stopped\")\n\t\t\treturn\n\t\tcase <-ticker.C:\n\t\t\ta.pollAll()\n\t\t}\n\t}\n}\n\nfunc (a *Adapter) pollAll() {\n\tentries := a.snapshot()\n\tlog.Debug(\"pollAll bots=%d\", len(entries))\n\tif len(entries) == 0 {\n\t\treturn\n\t}\n\n\tctx, cancel := context.WithTimeout(context.Background(), pollInterval)\n\tdefer cancel()\n\n\tfor _, entry := range entries {\n\t\tselect {\n\t\tcase <-a.stopCh:\n\t\t\treturn\n\t\tdefault:\n\t\t}\n\n\t\tlog.Debug(\"polling robot=%s offset=%d\", entry.robotID, entry.offset)\n\t\tgroups := []string{\"telegram\", entry.robotID}\n\t\tmsgs, err := entry.bot.GetUpdates(ctx, entry.offset, pollTimeout, groups)\n\t\tif err != nil {\n\t\t\tlog.Error(\"getUpdates failed robot=%s: %v\", entry.robotID, err)\n\t\t\tcontinue\n\t\t}\n\n\t\tif len(msgs) == 0 {\n\t\t\tcontinue\n\t\t}\n\n\t\t// Advance offset for all received messages\n\t\tfor _, cm := range msgs {\n\t\t\tif cm.UpdateID >= entry.offset {\n\t\t\t\tentry.offset = cm.UpdateID + 1\n\t\t\t}\n\t\t}\n\n\t\t// Group by chatID, preserving order\n\t\tgrouped := groupByChatID(msgs)\n\t\tlog.Info(\"robot=%s got %d updates in %d chats\", entry.robotID, len(msgs), len(grouped))\n\n\t\tfor chatID, chatMsgs := range grouped {\n\t\t\tlog.Debug(\"robot=%s chat=%d messages=%d\", entry.robotID, chatID, len(chatMsgs))\n\t\t\ta.handleMessages(ctx, entry, chatMsgs)\n\t\t}\n\t}\n}\n\n// groupByChatID groups messages by chat ID, preserving chronological order.\nfunc groupByChatID(msgs []*tgapi.ConvertedMessage) map[int64][]*tgapi.ConvertedMessage {\n\tgrouped := make(map[int64][]*tgapi.ConvertedMessage)\n\tfor _, cm := range msgs {\n\t\tgrouped[cm.ChatID] = append(grouped[cm.ChatID], cm)\n\t}\n\treturn grouped\n}\n" }, { "path": "agent/robot/events/integrations/telegram/reply.go", "content": "package telegram\n\nimport (\n\t\"context\"\n\t\"fmt\"\n\t\"strconv\"\n\t\"strings\"\n\n\tagentcontext \"github.com/yaoapp/yao/agent/context\"\n\tevents \"github.com/yaoapp/yao/agent/robot/events\"\n\ttgapi \"github.com/yaoapp/yao/integrations/telegram\"\n)\n\n// Reply sends the assistant message back to the originating Telegram chat.\n// Content may be a plain string or []ContentPart (text, image_url, file, etc.).\n// Each adapter is responsible for interpreting the standard message format.\nfunc (a *Adapter) Reply(ctx context.Context, msg *agentcontext.Message, metadata *events.MessageMetadata) error {\n\tif msg == nil || metadata == nil {\n\t\treturn fmt.Errorf(\"nil message or metadata\")\n\t}\n\n\tchatID, err := strconv.ParseInt(metadata.ChatID, 10, 64)\n\tif err != nil {\n\t\treturn fmt.Errorf(\"invalid chat_id %q: %w\", metadata.ChatID, err)\n\t}\n\n\tvar replyTo int64\n\tif metadata.Extra != nil {\n\t\tif v, ok := metadata.Extra[\"tg_message_id\"]; ok {\n\t\t\tswitch id := v.(type) {\n\t\t\tcase int64:\n\t\t\t\treplyTo = id\n\t\t\tcase float64:\n\t\t\t\treplyTo = int64(id)\n\t\t\t}\n\t\t}\n\t}\n\n\tentry := a.resolveByChat(metadata)\n\tif entry == nil {\n\t\treturn fmt.Errorf(\"no bot registered for channel metadata (appID=%s)\", metadata.AppID)\n\t}\n\n\treturn a.sendContent(ctx, entry.bot, chatID, replyTo, msg.Content)\n}\n\n// sendContent dispatches based on the Content type.\nfunc (a *Adapter) sendContent(ctx context.Context, bot *tgapi.Bot, chatID, replyTo int64, content interface{}) error {\n\tswitch c := content.(type) {\n\tcase string:\n\t\tif strings.TrimSpace(c) == \"\" {\n\t\t\treturn nil\n\t\t}\n\t\treturn bot.SendMessage(ctx, chatID, c, replyTo)\n\n\tcase []interface{}:\n\t\treturn a.sendParts(ctx, bot, chatID, replyTo, c)\n\n\tdefault:\n\t\tparts, ok := toContentParts(content)\n\t\tif ok {\n\t\t\treturn a.sendPartsTyped(ctx, bot, chatID, replyTo, parts)\n\t\t}\n\t\treturn bot.SendMessage(ctx, chatID, fmt.Sprintf(\"%v\", content), replyTo)\n\t}\n}\n\n// sendParts handles []interface{} content parts (common from JSON unmarshalling).\nfunc (a *Adapter) sendParts(ctx context.Context, bot *tgapi.Bot, chatID, replyTo int64, parts []interface{}) error {\n\tvar textBuf strings.Builder\n\tfor _, part := range parts {\n\t\tm, ok := part.(map[string]interface{})\n\t\tif !ok {\n\t\t\tcontinue\n\t\t}\n\t\tpartType, _ := m[\"type\"].(string)\n\t\tswitch partType {\n\t\tcase \"text\":\n\t\t\tif text, ok := m[\"text\"].(string); ok {\n\t\t\t\ttextBuf.WriteString(text)\n\t\t\t}\n\t\tcase \"image_url\":\n\t\t\tif err := a.flushText(ctx, bot, chatID, replyTo, &textBuf); err != nil {\n\t\t\t\treturn err\n\t\t\t}\n\t\t\tif imgMap, ok := m[\"image_url\"].(map[string]interface{}); ok {\n\t\t\t\tif url, ok := imgMap[\"url\"].(string); ok {\n\t\t\t\t\tif err := sendFileOrWrapper(ctx, bot, chatID, replyTo, url, \"\"); err != nil {\n\t\t\t\t\t\tlog.Error(\"telegram reply: send image: %v\", err)\n\t\t\t\t\t}\n\t\t\t\t}\n\t\t\t}\n\t\tcase \"file\":\n\t\t\tif err := a.flushText(ctx, bot, chatID, replyTo, &textBuf); err != nil {\n\t\t\t\treturn err\n\t\t\t}\n\t\t\tif fileMap, ok := m[\"file\"].(map[string]interface{}); ok {\n\t\t\t\turl, _ := fileMap[\"url\"].(string)\n\t\t\t\tfilename, _ := fileMap[\"filename\"].(string)\n\t\t\t\tif url != \"\" {\n\t\t\t\t\tif err := sendFileOrWrapper(ctx, bot, chatID, replyTo, url, filename); err != nil {\n\t\t\t\t\t\tlog.Error(\"telegram reply: send file: %v\", err)\n\t\t\t\t\t}\n\t\t\t\t}\n\t\t\t}\n\t\t}\n\t}\n\treturn a.flushText(ctx, bot, chatID, replyTo, &textBuf)\n}\n\n// sendPartsTyped handles typed []agentcontext.ContentPart slices.\nfunc (a *Adapter) sendPartsTyped(ctx context.Context, bot *tgapi.Bot, chatID, replyTo int64, parts []agentcontext.ContentPart) error {\n\tvar textBuf strings.Builder\n\tfor _, part := range parts {\n\t\tswitch part.Type {\n\t\tcase agentcontext.ContentText:\n\t\t\ttextBuf.WriteString(part.Text)\n\t\tcase agentcontext.ContentImageURL:\n\t\t\tif err := a.flushText(ctx, bot, chatID, replyTo, &textBuf); err != nil {\n\t\t\t\treturn err\n\t\t\t}\n\t\t\tif part.ImageURL != nil {\n\t\t\t\tif err := sendFileOrWrapper(ctx, bot, chatID, replyTo, part.ImageURL.URL, \"\"); err != nil {\n\t\t\t\t\tlog.Error(\"telegram reply: send image: %v\", err)\n\t\t\t\t}\n\t\t\t}\n\t\tcase agentcontext.ContentFile:\n\t\t\tif err := a.flushText(ctx, bot, chatID, replyTo, &textBuf); err != nil {\n\t\t\t\treturn err\n\t\t\t}\n\t\t\tif part.File != nil {\n\t\t\t\tif err := sendFileOrWrapper(ctx, bot, chatID, replyTo, part.File.URL, part.File.Filename); err != nil {\n\t\t\t\t\tlog.Error(\"telegram reply: send file: %v\", err)\n\t\t\t\t}\n\t\t\t}\n\t\t}\n\t}\n\treturn a.flushText(ctx, bot, chatID, replyTo, &textBuf)\n}\n\nfunc (a *Adapter) flushText(ctx context.Context, bot *tgapi.Bot, chatID, replyTo int64, buf *strings.Builder) error {\n\tif buf.Len() == 0 {\n\t\treturn nil\n\t}\n\terr := bot.SendMessage(ctx, chatID, buf.String(), replyTo)\n\tbuf.Reset()\n\treturn err\n}\n\n// sendFileOrWrapper sends a file from a wrapper (__yao.attachment://xxx) or URL.\nfunc sendFileOrWrapper(ctx context.Context, bot *tgapi.Bot, chatID, replyTo int64, url, caption string) error {\n\tif strings.Contains(url, \"://\") && !strings.HasPrefix(url, \"http\") {\n\t\treturn bot.SendMedia(ctx, chatID, url, caption, replyTo)\n\t}\n\tif strings.HasPrefix(url, \"http\") {\n\t\tmediaType := tgapi.DetectMediaType(\"\")\n\t\treturn bot.SendMediaByURL(ctx, chatID, mediaType, url, caption, replyTo)\n\t}\n\treturn fmt.Errorf(\"unsupported file URL scheme: %s\", url)\n}\n\n// toContentParts tries to type-assert content to []agentcontext.ContentPart.\nfunc toContentParts(content interface{}) ([]agentcontext.ContentPart, bool) {\n\tparts, ok := content.([]agentcontext.ContentPart)\n\treturn parts, ok\n}\n\n// resolveByChat finds the bot entry matching the metadata.\nfunc (a *Adapter) resolveByChat(metadata *events.MessageMetadata) *botEntry {\n\tif metadata.AppID != \"\" {\n\t\tif entry, ok := a.resolveByAppID(metadata.AppID); ok {\n\t\t\treturn entry\n\t\t}\n\t}\n\ta.mu.RLock()\n\tdefer a.mu.RUnlock()\n\tfor _, entry := range a.bots {\n\t\treturn entry\n\t}\n\treturn nil\n}\n" }, { "path": "agent/robot/events/integrations/telegram/telegram.go", "content": "package telegram\n\nimport (\n\t\"context\"\n\t\"sync\"\n\n\t\"github.com/yaoapp/yao/agent/robot/logger\"\n\trobottypes \"github.com/yaoapp/yao/agent/robot/types\"\n\ttgapi \"github.com/yaoapp/yao/integrations/telegram\"\n)\n\nvar log = logger.New(\"telegram\")\n\n// Adapter implements the integrations.Adapter interface for Telegram Bot API.\n//\n// Architecture:\n// - One polling goroutine (ticker) iterates all registered bots every 60s\n// - One webhook goroutine listens to integration.webhook.telegram events\n// - One dedup cleaner goroutine removes expired keys every hour\ntype Adapter struct {\n\tmu sync.RWMutex\n\tbots map[string]*botEntry // robotID -> *botEntry\n\tappIdx map[string]string // appID -> robotID (webhook routing)\n\tdedup *dedupStore\n\twebhSub string\n\tstopCh chan struct{}\n}\n\n// botEntry holds the state for one robot's Telegram integration.\ntype botEntry struct {\n\trobotID string\n\tappID string\n\tbot *tgapi.Bot // bound to this robot's token\n\toffset int64 // polling offset\n}\n\n// NewAdapter creates a new Telegram adapter.\nfunc NewAdapter() *Adapter {\n\ta := &Adapter{\n\t\tbots: make(map[string]*botEntry),\n\t\tappIdx: make(map[string]string),\n\t\tdedup: newDedupStore(),\n\t\tstopCh: make(chan struct{}),\n\t}\n\tgo a.dedup.cleaner(a.stopCh)\n\tgo a.pollLoop()\n\treturn a\n}\n\n// Apply is called by the Dispatcher when a robot config is created or updated.\nfunc (a *Adapter) Apply(ctx context.Context, robot *robottypes.Robot) {\n\ttgConf := extractConfig(robot)\n\tlog.Debug(\"Apply robot=%s tgConf=%v\", robot.MemberID, tgConf != nil)\n\tif tgConf != nil {\n\t\tlog.Debug(\"Apply robot=%s enabled=%v token_len=%d host=%q\",\n\t\t\trobot.MemberID, tgConf.Enabled, len(tgConf.BotToken), tgConf.Host)\n\t}\n\n\tif tgConf == nil || !tgConf.Enabled || tgConf.BotToken == \"\" {\n\t\ta.removeBot(robot.MemberID)\n\t\treturn\n\t}\n\n\ta.mu.Lock()\n\tdefer a.mu.Unlock()\n\n\tif existing, ok := a.bots[robot.MemberID]; ok {\n\t\tif existing.bot.Token() == tgConf.BotToken {\n\t\t\treturn\n\t\t}\n\t\ta.removeBotLocked(robot.MemberID)\n\t}\n\n\tvar opts []tgapi.BotOption\n\tif tgConf.Host != \"\" {\n\t\topts = append(opts, tgapi.WithAPIBase(tgConf.Host))\n\t}\n\tentry := &botEntry{\n\t\trobotID: robot.MemberID,\n\t\tappID: tgConf.AppID,\n\t\tbot: tgapi.NewBot(tgConf.BotToken, tgConf.WebhookSecret, opts...),\n\t}\n\ta.bots[robot.MemberID] = entry\n\tif tgConf.AppID != \"\" {\n\t\ta.appIdx[tgConf.AppID] = robot.MemberID\n\t}\n\tlog.Info(\"telegram adapter: registered robot=%s\", robot.MemberID)\n}\n\n// Remove is called by the Dispatcher when a robot is deleted.\nfunc (a *Adapter) Remove(ctx context.Context, robotID string) {\n\ta.removeBot(robotID)\n}\n\n// Shutdown stops the polling loop, webhook subscription, and dedup cleaner.\nfunc (a *Adapter) Shutdown() {\n\tclose(a.stopCh)\n\ta.StopWebhookSubscription()\n\tlog.Info(\"telegram adapter: shutdown complete\")\n}\n\n// ResolveBot returns the tgapi.Bot for a given appID, used by the webhook\n// verification layer. Returns nil if not found.\nfunc (a *Adapter) ResolveBot(appID string) *tgapi.Bot {\n\tentry, ok := a.resolveByAppID(appID)\n\tif !ok {\n\t\treturn nil\n\t}\n\treturn entry.bot\n}\n\n// --- Bot registry ---\n\nfunc (a *Adapter) removeBot(robotID string) {\n\ta.mu.Lock()\n\tdefer a.mu.Unlock()\n\ta.removeBotLocked(robotID)\n}\n\nfunc (a *Adapter) removeBotLocked(robotID string) {\n\tentry, ok := a.bots[robotID]\n\tif !ok {\n\t\treturn\n\t}\n\tif entry.appID != \"\" {\n\t\tdelete(a.appIdx, entry.appID)\n\t}\n\tdelete(a.bots, robotID)\n\tlog.Info(\"telegram adapter: unregistered robot=%s\", robotID)\n}\n\n// snapshot returns a copy of all bot entries for safe iteration outside the lock.\nfunc (a *Adapter) snapshot() []*botEntry {\n\ta.mu.RLock()\n\tdefer a.mu.RUnlock()\n\tlist := make([]*botEntry, 0, len(a.bots))\n\tfor _, entry := range a.bots {\n\t\tlist = append(list, entry)\n\t}\n\treturn list\n}\n\nfunc (a *Adapter) resolveByAppID(appID string) (*botEntry, bool) {\n\ta.mu.RLock()\n\tdefer a.mu.RUnlock()\n\trobotID, ok := a.appIdx[appID]\n\tif !ok {\n\t\treturn nil, false\n\t}\n\tentry, ok := a.bots[robotID]\n\treturn entry, ok\n}\n\nfunc extractConfig(robot *robottypes.Robot) *robottypes.TelegramConfig {\n\tif robot.Config == nil || robot.Config.Integrations == nil {\n\t\treturn nil\n\t}\n\treturn robot.Config.Integrations.Telegram\n}\n" }, { "path": "agent/robot/events/integrations/telegram/webhook.go", "content": "package telegram\n\nimport (\n\t\"context\"\n\t\"encoding/json\"\n\n\t\"github.com/go-telegram/bot/models\"\n\t\"github.com/yaoapp/yao/event\"\n\teventtypes \"github.com/yaoapp/yao/event/types\"\n\ttgapi \"github.com/yaoapp/yao/integrations/telegram\"\n\twebhooktypes \"github.com/yaoapp/yao/openapi/integrations\"\n)\n\n// StartWebhookSubscription subscribes to integration.webhook.telegram events.\n// Call once after event.Start().\nfunc (a *Adapter) StartWebhookSubscription() {\n\tch := make(chan *eventtypes.Event, 128)\n\ta.webhSub = event.Subscribe(\"integration.webhook.telegram\", ch)\n\tgo a.handleWebhooks(ch)\n\tlog.Info(\"telegram adapter: webhook subscription started\")\n}\n\n// StopWebhookSubscription unsubscribes from webhook events.\nfunc (a *Adapter) StopWebhookSubscription() {\n\tif a.webhSub != \"\" {\n\t\tevent.Unsubscribe(a.webhSub)\n\t\ta.webhSub = \"\"\n\t}\n}\n\nfunc (a *Adapter) handleWebhooks(ch <-chan *eventtypes.Event) {\n\tfor ev := range ch {\n\t\tvar payload webhooktypes.WebhookPayload\n\t\tif err := ev.Should(&payload); err != nil {\n\t\t\tlog.Error(\"telegram adapter: invalid webhook event: %v\", err)\n\t\t\tcontinue\n\t\t}\n\n\t\tentry, ok := a.resolveByAppID(payload.AppID)\n\t\tif !ok {\n\t\t\tlog.Warn(\"telegram adapter: unknown app_id=%s\", payload.AppID)\n\t\t\tcontinue\n\t\t}\n\n\t\theaderSecret := payload.Headers[\"X-Telegram-Bot-Api-Secret-Token\"]\n\t\tif !entry.bot.VerifyWebhook(headerSecret) {\n\t\t\tlog.Warn(\"telegram adapter: webhook secret mismatch app_id=%s\", payload.AppID)\n\t\t\tcontinue\n\t\t}\n\n\t\tvar update models.Update\n\t\tif err := json.Unmarshal(payload.Body, &update); err != nil {\n\t\t\tlog.Error(\"telegram adapter: webhook unmarshal failed: %v\", err)\n\t\t\tcontinue\n\t\t}\n\n\t\tcm := tgapi.ConvertUpdate(&update)\n\t\tif cm != nil && cm.HasMedia() {\n\t\t\tgroups := []string{\"telegram\", entry.robotID}\n\t\t\tctx := context.Background()\n\t\t\tentry.bot.ResolveMedia(ctx, cm, groups)\n\t\t}\n\t\ta.handleMessages(context.Background(), entry, []*tgapi.ConvertedMessage{cm})\n\t}\n}\n" }, { "path": "agent/robot/events/log.go", "content": "package events\n\nimport \"github.com/yaoapp/yao/agent/robot/logger\"\n\nvar log = logger.New(\"events\")\n" }, { "path": "agent/robot/events/message.go", "content": "package events\n\nimport (\n\t\"context\"\n\t\"fmt\"\n\t\"strings\"\n\n\tagent \"github.com/yaoapp/yao/agent\"\n\t\"github.com/yaoapp/yao/agent/assistant\"\n\tagentcontext \"github.com/yaoapp/yao/agent/context\"\n\trobotstore \"github.com/yaoapp/yao/agent/robot/store\"\n\trobottypes \"github.com/yaoapp/yao/agent/robot/types\"\n\teventtypes \"github.com/yaoapp/yao/event/types\"\n\toauthtypes \"github.com/yaoapp/yao/openapi/oauth/types\"\n)\n\n// handleMessage processes messages from external integrations (Telegram, etc.).\n// It calls the Host Agent with the provided messages and returns a MessageResult.\n// Action detection is done via the Host Agent's Next hook return value.\nfunc (h *robotHandler) handleMessage(ctx context.Context, ev *eventtypes.Event, resp chan<- eventtypes.Result) {\n\tvar payload MessagePayload\n\tif err := ev.Should(&payload); err != nil {\n\t\tlog.Error(\"message handler: invalid payload: %v\", err)\n\t\tif ev.IsCall {\n\t\t\tresp <- eventtypes.Result{Err: err}\n\t\t}\n\t\treturn\n\t}\n\n\tlog.Info(\"message handler: robot=%s channel=%s msg_id=%s\",\n\t\tpayload.RobotID, payload.Metadata.Channel, payload.Metadata.MessageID)\n\n\tresult, err := callHostAgent(ctx, &payload)\n\tif err != nil {\n\t\tlog.Error(\"message handler: host agent call failed robot=%s: %v\", payload.RobotID, err)\n\t\tif ev.IsCall {\n\t\t\tresp <- eventtypes.Result{Err: err}\n\t\t}\n\t\treturn\n\t}\n\n\tif reply := getReplyFunc(); reply != nil && result.Message != nil {\n\t\tif err := reply(ctx, result.Message, payload.Metadata); err != nil {\n\t\t\tlog.Error(\"message handler: reply failed robot=%s channel=%s: %v\",\n\t\t\t\tpayload.RobotID, payload.Metadata.Channel, err)\n\t\t}\n\t}\n\n\tif ev.IsCall {\n\t\tresp <- eventtypes.Result{Data: result}\n\t}\n}\n\n// callHostAgent resolves the Host Agent for the robot and calls it with messages.\n// This avoids importing executor/standard to prevent import cycles; instead it\n// calls the assistant directly via assistant.Get + ast.Stream (same as AgentCaller.Call).\nfunc callHostAgent(ctx context.Context, payload *MessagePayload) (*MessageResult, error) {\n\thostID, record, err := resolveHostAssistantID(ctx, payload.RobotID)\n\tif err != nil {\n\t\treturn nil, fmt.Errorf(\"failed to resolve host agent: %w\", err)\n\t}\n\n\tast, err := assistant.Get(hostID)\n\tif err != nil {\n\t\treturn nil, fmt.Errorf(\"assistant not found: %s: %w\", hostID, err)\n\t}\n\n\topts := &agentcontext.Options{\n\t\tSkip: &agentcontext.Skip{\n\t\t\tSearch: false,\n\t\t},\n\t}\n\n\tauthorized := &oauthtypes.AuthorizedInfo{\n\t\tUserID: payload.Metadata.SenderID,\n\t}\n\tchatID := fmt.Sprintf(\"%s:%s\", payload.Metadata.Channel, payload.Metadata.ChatID)\n\tagentCtx := agentcontext.New(ctx, authorized, chatID)\n\tagentCtx.AssistantID = hostID\n\tagentCtx.Referer = \"integration\"\n\tagentCtx.Locale = payload.Metadata.Locale\n\tagentCtx.Metadata = map[string]interface{}{\n\t\t\"robot_id\": payload.RobotID,\n\t\t\"channel\": payload.Metadata.Channel,\n\t}\n\n\tif dsl := agent.GetAgent(); dsl != nil {\n\t\tif cache, err := dsl.GetCacheStore(); err == nil {\n\t\t\tagentCtx.Cache = cache\n\t\t}\n\t}\n\n\tdefer agentCtx.Release()\n\n\tresponse, err := ast.Stream(agentCtx, payload.Messages, opts)\n\tif err != nil {\n\t\treturn nil, fmt.Errorf(\"host agent call failed: %w\", err)\n\t}\n\n\tresult := &MessageResult{\n\t\tMetadata: payload.Metadata,\n\t}\n\n\tif response.Completion != nil {\n\t\tresult.Message = &agentcontext.Message{\n\t\t\tRole: agentcontext.RoleAssistant,\n\t\t\tContent: response.Completion.Content,\n\t\t}\n\t}\n\n\t// Detect action from Next hook return value\n\tlog.Debug(\"response.Next type=%T value=%+v\", response.Next, response.Next)\n\tif action := detectAction(response.Next); action != nil {\n\t\tlog.Info(\"action detected: name=%s payload=%+v\", action.Name, action.Payload)\n\t\tresult.Action = action\n\t\tif action.Name == \"robot.execute\" {\n\t\t\tif execID := executeAction(ctx, payload, record, action); execID != \"\" {\n\t\t\t\tresult.ExecutionID = execID\n\t\t\t\tresult.Message = &agentcontext.Message{\n\t\t\t\t\tRole: agentcontext.RoleAssistant,\n\t\t\t\t\tContent: taskDeployedMessage(execID, payload.Metadata.Locale),\n\t\t\t\t}\n\t\t\t}\n\t\t}\n\t} else {\n\t\tlog.Debug(\"no action detected from response.Next\")\n\t}\n\n\treturn result, nil\n}\n\n// executeAction triggers robot execution when the Host Agent returns a\n// confirmed action. Uses the injected TriggerFunc to call robotapi.TriggerManual\n// without creating a circular import.\nfunc executeAction(ctx context.Context, payload *MessagePayload, record *robotstore.RobotRecord, action *ActionResult) string {\n\ttrigger := getTriggerFunc()\n\tif trigger == nil {\n\t\tlog.Warn(\"message handler: trigger func not registered, cannot execute action for robot=%s\", payload.RobotID)\n\t\treturn \"\"\n\t}\n\n\tdata, _ := action.Payload.(map[string]interface{})\n\tgoals, _ := data[\"goals\"].(string)\n\tif goals == \"\" {\n\t\tlog.Warn(\"message handler: confirmed action has no goals, robot=%s\", payload.RobotID)\n\t\treturn \"\"\n\t}\n\n\ttriggerData := &robottypes.TriggerInput{\n\t\tData: map[string]interface{}{\n\t\t\t\"goals\": goals,\n\t\t\t\"channel\": payload.Metadata.Channel,\n\t\t\t\"chat_id\": payload.Metadata.ChatID,\n\t\t\t\"extra\": payload.Metadata.Extra,\n\t\t},\n\t}\n\n\tauthorized := &oauthtypes.AuthorizedInfo{\n\t\tUserID: record.MemberID,\n\t\tTeamID: record.TeamID,\n\t}\n\trCtx := robottypes.NewContext(ctx, authorized)\n\n\texecID, accepted, err := trigger(rCtx, payload.RobotID, robottypes.TriggerHuman, triggerData)\n\tif err != nil {\n\t\tlog.Error(\"message handler: execute action failed robot=%s: %v\", payload.RobotID, err)\n\t\treturn \"\"\n\t}\n\tif !accepted {\n\t\tlog.Warn(\"message handler: execute action not accepted robot=%s\", payload.RobotID)\n\t\treturn \"\"\n\t}\n\n\tlog.Info(\"message handler: execution triggered robot=%s exec_id=%s\", payload.RobotID, execID)\n\treturn execID\n}\n\n// detectAction checks the Next hook return value for a confirmed action.\n// The Host Agent returns { data: { confirmed: true, robot_id: \"...\", goals: \"...\" } }\n// when it detects a confirm_task tool call.\nfunc detectAction(next interface{}) *ActionResult {\n\tif next == nil {\n\t\treturn nil\n\t}\n\n\tm, ok := next.(map[string]interface{})\n\tif !ok {\n\t\treturn nil\n\t}\n\n\t// Next hook may return { data: { confirmed, ... } } or flat { confirmed, ... }\n\tdata, _ := m[\"data\"].(map[string]interface{})\n\tif data == nil {\n\t\tdata = m\n\t}\n\n\tconfirmed, _ := data[\"confirmed\"].(bool)\n\tif !confirmed {\n\t\treturn nil\n\t}\n\n\treturn &ActionResult{\n\t\tName: \"robot.execute\",\n\t\tPayload: data,\n\t}\n}\n\n// resolveHostAssistantID resolves the host assistant ID from a robot member ID.\n// Mirrors the logic in openapi/agent/robot/completions.go.\nfunc resolveHostAssistantID(ctx context.Context, memberID string) (string, *robotstore.RobotRecord, error) {\n\tstore := robotstore.NewRobotStore()\n\trecord, err := store.Get(ctx, memberID)\n\tif err != nil {\n\t\treturn \"\", nil, fmt.Errorf(\"failed to get robot: %w\", err)\n\t}\n\tif record == nil {\n\t\treturn \"\", nil, fmt.Errorf(\"robot not found: %s\", memberID)\n\t}\n\n\tconfig, err := robottypes.ParseConfig(record.RobotConfig)\n\tif err != nil {\n\t\treturn \"\", nil, fmt.Errorf(\"failed to parse robot config: %w\", err)\n\t}\n\n\tvar hostID string\n\tif config != nil && config.Resources != nil {\n\t\thostID = config.Resources.GetPhaseAgent(robottypes.PhaseHost)\n\t} else {\n\t\thostID = \"__yao.\" + string(robottypes.PhaseHost)\n\t}\n\n\treturn hostID, record, nil\n}\n\nfunc taskDeployedMessage(execID string, locale string) string {\n\tif strings.HasPrefix(locale, \"zh\") {\n\t\treturn fmt.Sprintf(\"任务已部署（执行编号: %s），完成后会将结果发送给你。\", execID)\n\t}\n\treturn fmt.Sprintf(\"Task deployed (execution: %s). You will receive results once completed.\", execID)\n}\n" }, { "path": "agent/robot/executor/README.md", "content": "# Robot Executor\n\nRobot Executor provides pluggable execution strategies for robot phase execution.\n\n## Architecture\n\n```\nexecutor/\n├── types/\n│ ├── types.go # Interface definitions and common types\n│ └── helpers.go # Shared helper functions\n├── standard/\n│ ├── executor.go # Real Agent execution (production)\n│ └── phases.go # Phase implementations\n├── dryrun/\n│ └── executor.go # Simulated execution (testing/demo)\n├── sandbox/\n│ └── executor.go # Container-isolated execution (NOT IMPLEMENTED)\n└── executor.go # Factory functions and unified entry\n```\n\n## Execution Modes\n\n### Standard Mode (Production)\n\nReal Agent calls with full phase execution:\n\n```go\nexec := executor.New()\n// or\nexec := executor.NewWithConfig(executor.Config{\n OnPhaseStart: func(phase types.Phase) { ... },\n OnPhaseEnd: func(phase types.Phase) { ... },\n})\n```\n\n### DryRun Mode (Testing/Demo)\n\nSimulates execution without real Agent calls:\n\n```go\n// Simple dry-run\nexec := executor.NewDryRun()\n\n// With delay simulation\nexec := executor.NewDryRunWithDelay(100 * time.Millisecond)\n\n// With full configuration\nexec := executor.NewDryRunWithConfig(executor.DryRunConfig{\n Delay: 100 * time.Millisecond,\n OnStart: func() { ... },\n OnEnd: func() { ... },\n Config: executor.Config{\n OnPhaseStart: func(phase types.Phase) { ... },\n },\n})\n```\n\n### Sandbox Mode (NOT IMPLEMENTED)\n\n> **⚠️ Not Implemented:** Sandbox mode requires container-level isolation (Docker/gVisor/Firecracker) for true security isolation. This is a future feature that depends on infrastructure support.\n\n**Intended Design:**\n\nSandbox mode is designed for executing untrusted robot configurations in a fully isolated environment:\n\n- **Container Isolation:** Each execution runs in a separate container\n- **Resource Limits:** CPU, memory, disk, network quotas enforced by container runtime\n- **Network Isolation:** Restricted network access via container networking\n- **File System Isolation:** Read-only root filesystem, limited writable paths\n- **Process Isolation:** Separate PID namespace, no access to host processes\n\n**Future Implementation:**\n\n```go\n// Future API (not yet implemented)\nexec := executor.NewSandbox(executor.SandboxConfig{\n Image: \"yao-executor:latest\",\n MaxDuration: 30 * time.Minute,\n MaxMemory: 512 * 1024 * 1024, // 512MB\n MaxCPU: 1.0, // 1 CPU core\n NetworkPolicy: \"restricted\", // restricted | none | full\n AllowedAgents: []string{\"agent1\", \"agent2\"},\n})\n```\n\n**Current Placeholder:**\n\nThe current `sandbox/executor.go` is a placeholder that behaves like DryRun mode. It does NOT provide real security isolation.\n\n## Mode Selection\n\nSelect mode dynamically:\n\n```go\n// By mode constant\nexec := executor.NewWithMode(executor.ModeDryRun)\n\n// From settings\nsetting := &executor.Setting{\n Mode: executor.ModeStandard,\n}\nexec := executor.NewWithSetting(setting)\n```\n\n## Interface\n\nAll executors implement the `Executor` interface:\n\n```go\ntype Executor interface {\n Execute(ctx *Context, robot *Robot, trigger TriggerType, data interface{}) (*Execution, error)\n ExecCount() int\n CurrentCount() int\n Reset()\n}\n```\n\n## Use Cases\n\n| Mode | Use Case | Status |\n| -------- | --------------------------------------------------- | ------------------ |\n| Standard | Production environment with real Agent calls | ✅ Implemented |\n| DryRun | Unit tests, integration tests, demos, previews | ✅ Implemented |\n| Sandbox | Untrusted code execution, multi-tenant environments | ⬜ Not Implemented |\n\n## Testing\n\nTests use DryRun mode by default:\n\n```go\nfunc TestSomething(t *testing.T) {\n exec := executor.NewDryRunWithDelay(50 * time.Millisecond)\n // ... test with simulated execution\n}\n```\n\n## Manager Integration\n\nInject executor into Manager:\n\n```go\nexec := executor.NewDryRun()\nconfig := &manager.Config{\n Executor: exec,\n}\nm := manager.NewWithConfig(config)\n```\n" }, { "path": "agent/robot/executor/dryrun/executor.go", "content": "package dryrun\n\nimport (\n\t\"fmt\"\n\t\"sync/atomic\"\n\t\"time\"\n\n\t\"github.com/yaoapp/yao/agent/robot/executor/types\"\n\trobottypes \"github.com/yaoapp/yao/agent/robot/types\"\n)\n\n// Executor implements a dry-run executor that simulates execution\n// without making real Agent calls. Useful for:\n// - Testing scheduling and concurrency logic\n// - Demo and preview modes\n// - Debugging execution flow\n// - Performance testing\ntype Executor struct {\n\tconfig types.DryRunConfig\n\texecCount atomic.Int32\n\tcurrentCount atomic.Int32\n}\n\n// New creates a new dry-run executor with default settings\nfunc New() *Executor {\n\treturn &Executor{}\n}\n\n// NewWithDelay creates a dry-run executor with specified delay\nfunc NewWithDelay(delay time.Duration) *Executor {\n\treturn &Executor{\n\t\tconfig: types.DryRunConfig{\n\t\t\tDelay: delay,\n\t\t},\n\t}\n}\n\n// NewWithConfig creates a dry-run executor with full configuration\nfunc NewWithConfig(config types.DryRunConfig) *Executor {\n\treturn &Executor{\n\t\tconfig: config,\n\t}\n}\n\n// Execute simulates robot execution without real Agent calls (auto-generates ID)\nfunc (e *Executor) Execute(ctx *robottypes.Context, robot *robottypes.Robot, trigger robottypes.TriggerType, data interface{}) (*robottypes.Execution, error) {\n\treturn e.ExecuteWithControl(ctx, robot, trigger, data, \"\", nil)\n}\n\n// ExecuteWithID simulates robot execution with a pre-generated execution ID (no control)\nfunc (e *Executor) ExecuteWithID(ctx *robottypes.Context, robot *robottypes.Robot, trigger robottypes.TriggerType, data interface{}, execID string) (*robottypes.Execution, error) {\n\treturn e.ExecuteWithControl(ctx, robot, trigger, data, execID, nil)\n}\n\n// ExecuteWithControl simulates robot execution with execution control\nfunc (e *Executor) ExecuteWithControl(ctx *robottypes.Context, robot *robottypes.Robot, trigger robottypes.TriggerType, data interface{}, execID string, control robottypes.ExecutionControl) (*robottypes.Execution, error) {\n\tif robot == nil {\n\t\treturn nil, fmt.Errorf(\"robot cannot be nil\")\n\t}\n\n\t// Determine starting phase\n\tstartPhaseIndex := 0\n\tif trigger == robottypes.TriggerHuman || trigger == robottypes.TriggerEvent {\n\t\tstartPhaseIndex = 1 // Skip P0\n\t}\n\n\t// Use provided execID or generate new one\n\tif execID == \"\" {\n\t\texecID = fmt.Sprintf(\"dryrun_%d\", time.Now().UnixNano())\n\t}\n\n\t// Create execution record\n\texec := &robottypes.Execution{\n\t\tID: execID,\n\t\tMemberID: robot.MemberID,\n\t\tTeamID: robot.TeamID,\n\t\tTriggerType: trigger,\n\t\tStartTime: time.Now(),\n\t\tStatus: robottypes.ExecPending,\n\t\tPhase: robottypes.AllPhases[startPhaseIndex],\n\t\tInput: types.BuildTriggerInput(trigger, data),\n\t}\n\n\t// Set robot reference\n\texec.SetRobot(robot)\n\n\t// Acquire slot\n\tif !robot.TryAcquireSlot(exec) {\n\t\treturn nil, robottypes.ErrQuotaExceeded\n\t}\n\tdefer robot.RemoveExecution(exec.ID)\n\n\t// Track counts\n\te.execCount.Add(1)\n\te.currentCount.Add(1)\n\tdefer e.currentCount.Add(-1)\n\n\t// Start callback\n\tif e.config.OnStart != nil {\n\t\te.config.OnStart()\n\t}\n\tif e.config.OnEnd != nil {\n\t\tdefer e.config.OnEnd()\n\t}\n\n\t// Update status\n\texec.Status = robottypes.ExecRunning\n\n\t// Simulate execution delay (once for entire execution, not per-phase)\n\tif e.config.Delay > 0 {\n\t\ttime.Sleep(e.config.Delay)\n\t}\n\n\t// Check for simulated failure\n\tif dataStr, ok := data.(string); ok && dataStr == \"simulate_failure\" {\n\t\texec.Status = robottypes.ExecFailed\n\t\texec.Error = \"simulated failure\"\n\t\treturn exec, nil\n\t}\n\n\t// Execute phases with mock data\n\tphases := robottypes.AllPhases[startPhaseIndex:]\n\tfor _, phase := range phases {\n\t\t// Check if cancelled\n\t\tselect {\n\t\tcase <-ctx.Context.Done():\n\t\t\texec.Status = robottypes.ExecCancelled\n\t\t\texec.Error = \"execution cancelled\"\n\t\t\treturn exec, nil\n\t\tdefault:\n\t\t}\n\n\t\t// Wait if paused\n\t\tif control != nil {\n\t\t\tif err := control.WaitIfPaused(); err != nil {\n\t\t\t\texec.Status = robottypes.ExecCancelled\n\t\t\t\texec.Error = \"execution cancelled while paused\"\n\t\t\t\treturn exec, nil\n\t\t\t}\n\t\t}\n\n\t\texec.Phase = phase\n\n\t\t// Phase start callback\n\t\tif e.config.OnPhaseStart != nil {\n\t\t\te.config.OnPhaseStart(phase)\n\t\t}\n\n\t\t// Generate mock output\n\t\te.mockPhaseOutput(exec, phase)\n\n\t\t// Phase end callback\n\t\tif e.config.OnPhaseEnd != nil {\n\t\t\te.config.OnPhaseEnd(phase)\n\t\t}\n\t}\n\n\t// Mark completed\n\texec.Status = robottypes.ExecCompleted\n\tnow := time.Now()\n\texec.EndTime = &now\n\n\treturn exec, nil\n}\n\n// mockPhaseOutput generates mock output for each phase\nfunc (e *Executor) mockPhaseOutput(exec *robottypes.Execution, phase robottypes.Phase) {\n\tswitch phase {\n\tcase robottypes.PhaseInspiration:\n\t\texec.Inspiration = &robottypes.InspirationReport{\n\t\t\tClock: robottypes.NewClockContext(time.Now(), \"\"),\n\t\t\tContent: \"## Dry-Run Inspiration\\n\\nThis is a simulated inspiration report for testing.\",\n\t\t}\n\tcase robottypes.PhaseGoals:\n\t\texec.Goals = &robottypes.Goals{\n\t\t\tContent: \"## Dry-Run Goals\\n\\n1. [High] Simulated goal for testing\",\n\t\t}\n\tcase robottypes.PhaseTasks:\n\t\texec.Tasks = []robottypes.Task{\n\t\t\t{\n\t\t\t\tID: \"dryrun-task-1\",\n\t\t\t\tGoalRef: \"Goal 1\",\n\t\t\t\tSource: robottypes.TaskSourceAuto,\n\t\t\t\tExecutorType: robottypes.ExecutorAssistant,\n\t\t\t\tExecutorID: \"mock-agent\",\n\t\t\t\tStatus: robottypes.TaskPending,\n\t\t\t},\n\t\t}\n\tcase robottypes.PhaseRun:\n\t\texec.Results = []robottypes.TaskResult{\n\t\t\t{\n\t\t\t\tTaskID: \"dryrun-task-1\",\n\t\t\t\tSuccess: true,\n\t\t\t\tOutput: map[string]interface{}{\"mode\": \"dryrun\", \"result\": \"simulated\"},\n\t\t\t\tDuration: 100,\n\t\t\t\tValidation: &robottypes.ValidationResult{\n\t\t\t\t\tPassed: true,\n\t\t\t\t\tScore: 1.0,\n\t\t\t\t},\n\t\t\t},\n\t\t}\n\tcase robottypes.PhaseDelivery:\n\t\texec.Delivery = &robottypes.DeliveryResult{\n\t\t\tRequestID: \"dryrun-\" + exec.ID,\n\t\t\tContent: &robottypes.DeliveryContent{\n\t\t\t\tSummary: \"Dry-run delivery completed\",\n\t\t\t\tBody: \"# Dry-run Delivery\\n\\nThis is a simulated delivery result.\",\n\t\t\t},\n\t\t\tSuccess: true,\n\t\t}\n\tcase robottypes.PhaseLearning:\n\t\texec.Learning = []robottypes.LearningEntry{\n\t\t\t{\n\t\t\t\tType: robottypes.LearnExecution,\n\t\t\t\tContent: \"Dry-run execution completed successfully\",\n\t\t\t},\n\t\t}\n\t}\n}\n\n// ExecCount returns total execution count\nfunc (e *Executor) ExecCount() int {\n\treturn int(e.execCount.Load())\n}\n\n// CurrentCount returns currently running execution count\nfunc (e *Executor) CurrentCount() int {\n\treturn int(e.currentCount.Load())\n}\n\n// Reset resets the executor counters\nfunc (e *Executor) Reset() {\n\te.execCount.Store(0)\n\te.currentCount.Store(0)\n}\n\n// Resume is not supported in dry-run mode\nfunc (e *Executor) Resume(ctx *robottypes.Context, execID string, reply string) error {\n\treturn fmt.Errorf(\"resume is not supported in dry-run executor\")\n}\n\n// Verify Executor implements types.Executor\nvar _ types.Executor = (*Executor)(nil)\n" }, { "path": "agent/robot/executor/executor.go", "content": "// Package executor provides robot execution strategies\n//\n// Architecture:\n//\n//\texecutor/\n//\t├── types/\n//\t│ ├── types.go # Interface definitions and common types\n//\t│ └── helpers.go # Shared helper functions\n//\t├── standard/\n//\t│ ├── executor.go # Real Agent execution (production)\n//\t│ ├── agent.go # AgentCaller for LLM calls\n//\t│ ├── input.go # InputFormatter for prompts\n//\t│ ├── inspiration.go # P0: Inspiration phase\n//\t│ ├── goals.go # P1: Goals phase\n//\t│ ├── tasks.go # P2: Tasks phase\n//\t│ ├── run.go # P3: Run phase\n//\t│ ├── delivery.go # P4: Delivery phase\n//\t│ └── learning.go # P5: Learning phase\n//\t├── dryrun/\n//\t│ └── executor.go # Simulated execution (testing/demo)\n//\t├── sandbox/\n//\t│ └── executor.go # Container-isolated execution (NOT IMPLEMENTED)\n//\t└── executor.go # Factory functions (this file)\n//\n// Usage:\n//\n//\t// Production - real Agent calls\n//\texec := executor.New()\n//\n//\t// Testing - simulated execution\n//\texec := executor.NewDryRun()\n//\n//\t// Sandbox - NOT IMPLEMENTED (requires container infrastructure)\n//\t// exec := executor.NewSandbox() // placeholder only\n//\n//\t// With mode selection\n//\texec := executor.NewWithMode(executor.ModeDryRun)\npackage executor\n\nimport (\n\t\"time\"\n\n\t\"github.com/yaoapp/yao/agent/robot/executor/dryrun\"\n\t\"github.com/yaoapp/yao/agent/robot/executor/sandbox\"\n\t\"github.com/yaoapp/yao/agent/robot/executor/standard\"\n\t\"github.com/yaoapp/yao/agent/robot/executor/types\"\n\trobottypes \"github.com/yaoapp/yao/agent/robot/types\"\n)\n\n// Re-export types for convenience\ntype (\n\tExecutor = types.Executor\n\tConfig = types.Config\n\tDryRunConfig = types.DryRunConfig\n\tSandboxConfig = types.SandboxConfig\n\tMode = types.Mode\n\tSetting = types.Setting\n)\n\n// Re-export mode constants\nconst (\n\tModeStandard = types.ModeStandard\n\tModeDryRun = types.ModeDryRun\n\tModeSandbox = types.ModeSandbox\n)\n\n// ==================== Factory Functions ====================\n\n// New creates a new standard executor (production mode)\n// Uses real Agent calls for phase execution\nfunc New() Executor {\n\treturn standard.New()\n}\n\n// NewWithConfig creates a standard executor with configuration\nfunc NewWithConfig(config Config) Executor {\n\treturn standard.NewWithConfig(config)\n}\n\n// NewDryRun creates a dry-run executor (testing/demo mode)\n// Simulates execution without real Agent calls\nfunc NewDryRun() Executor {\n\treturn dryrun.New()\n}\n\n// NewDryRunWithDelay creates a dry-run executor with specified delay\nfunc NewDryRunWithDelay(delay time.Duration) *DryRunExecutor {\n\treturn dryrun.NewWithDelay(delay)\n}\n\n// NewDryRunWithConfig creates a dry-run executor with full configuration\nfunc NewDryRunWithConfig(config DryRunConfig) *DryRunExecutor {\n\treturn dryrun.NewWithConfig(config)\n}\n\n// NewDryRunWithCallbacks creates a dry-run executor with start/end callbacks\nfunc NewDryRunWithCallbacks(delay time.Duration, onStart, onEnd func()) *DryRunExecutor {\n\treturn dryrun.NewWithConfig(DryRunConfig{\n\t\tDelay: delay,\n\t\tOnStart: onStart,\n\t\tOnEnd: onEnd,\n\t})\n}\n\n// NewSandbox creates a sandbox executor placeholder\n//\n// ⚠️ NOT IMPLEMENTED: True sandbox requires container-level isolation\n// (Docker/gVisor/Firecracker). Current implementation behaves like DryRun.\nfunc NewSandbox() Executor {\n\treturn sandbox.New()\n}\n\n// NewSandboxWithConfig creates a sandbox executor placeholder with configuration\n//\n// ⚠️ NOT IMPLEMENTED: Config options are placeholders. Current implementation\n// behaves like DryRun and does NOT provide real security isolation.\nfunc NewSandboxWithConfig(config SandboxConfig) Executor {\n\treturn sandbox.NewWithConfig(config)\n}\n\n// NewWithMode creates an executor based on the specified mode\nfunc NewWithMode(mode Mode) Executor {\n\tswitch mode {\n\tcase ModeDryRun:\n\t\treturn NewDryRun()\n\tcase ModeSandbox:\n\t\treturn NewSandbox()\n\tdefault:\n\t\treturn New()\n\t}\n}\n\n// NewWithSetting creates an executor based on configuration settings\nfunc NewWithSetting(setting *Setting) Executor {\n\tif setting == nil {\n\t\treturn New()\n\t}\n\n\tswitch setting.Mode {\n\tcase ModeDryRun:\n\t\treturn NewDryRun()\n\tcase ModeSandbox:\n\t\treturn NewSandboxWithConfig(SandboxConfig{\n\t\t\tMaxDuration: setting.MaxDuration,\n\t\t\tMaxMemory: setting.MaxMemory,\n\t\t\tAllowedAgents: setting.AllowedAgents,\n\t\t\tNetworkAccess: setting.NetworkAccess,\n\t\t\tFileAccess: setting.FileAccess,\n\t\t})\n\tdefault:\n\t\treturn New()\n\t}\n}\n\n// ==================== Concrete Types ====================\n// Export concrete executor types for direct access when needed\n\n// DryRunExecutor is the concrete dry-run executor type\ntype DryRunExecutor = dryrun.Executor\n\n// StandardExecutor is the concrete standard executor type\ntype StandardExecutor = standard.Executor\n\n// SandboxExecutor is the concrete sandbox executor type\ntype SandboxExecutor = sandbox.Executor\n\n// ==================== Interface Verification ====================\n\n// Verify all executors implement the Executor interface\nvar (\n\t_ Executor = (*standard.Executor)(nil)\n\t_ Executor = (*dryrun.Executor)(nil)\n\t_ Executor = (*sandbox.Executor)(nil)\n)\n\n// Verify standard executor implements PhaseExecutor\nvar _ types.PhaseExecutor = (*standard.Executor)(nil)\n\n// ==================== Helper Types ====================\n\n// DefaultSetting returns default executor settings\nfunc DefaultSetting() *Setting {\n\treturn types.DefaultSetting()\n}\n\n// PhaseExecutor is the interface for phase execution\ntype PhaseExecutor = types.PhaseExecutor\n\n// ==================== Context Helpers ====================\n\n// These are re-exported from robot types for convenience\ntype (\n\tContext = robottypes.Context\n\tRobot = robottypes.Robot\n\tExecution = robottypes.Execution\n\tPhase = robottypes.Phase\n)\n" }, { "path": "agent/robot/executor/executor_test.go", "content": "package executor\n\nimport (\n\t\"context\"\n\t\"testing\"\n\n\t\"github.com/stretchr/testify/assert\"\n\t\"github.com/yaoapp/yao/agent/robot/types\"\n)\n\n// Smoke tests to verify basic flow works\n// Real integration tests are in manager_test.go\n\nfunc TestExecutorSmoke(t *testing.T) {\n\texec := NewDryRunWithDelay(0)\n\trobot := &types.Robot{\n\t\tMemberID: \"test-smoke\",\n\t\tTeamID: \"team-1\",\n\t\tConfig: &types.Config{Quota: &types.Quota{Max: 1}},\n\t}\n\tctx := types.NewContext(context.Background(), nil)\n\n\tresult, err := exec.Execute(ctx, robot, types.TriggerClock, nil)\n\n\tassert.NoError(t, err)\n\tassert.NotNil(t, result)\n\tassert.Equal(t, types.ExecCompleted, result.Status)\n\tassert.Equal(t, types.TriggerClock, result.TriggerType)\n\n\t// Clock trigger executes all phases (P0-P5)\n\tassert.NotNil(t, result.Inspiration, \"P0 should be executed for clock trigger\")\n\tassert.NotNil(t, result.Goals, \"P1 should be executed\")\n\tassert.NotEmpty(t, result.Tasks, \"P2 should generate tasks\")\n\tassert.NotEmpty(t, result.Results, \"P3 should generate results\")\n\tassert.NotNil(t, result.Delivery, \"P4 should be executed\")\n\tassert.NotEmpty(t, result.Learning, \"P5 should be executed\")\n}\n\nfunc TestExecutorHumanTriggerSkipsP0(t *testing.T) {\n\texec := NewDryRunWithDelay(0)\n\trobot := &types.Robot{\n\t\tMemberID: \"test-human\",\n\t\tTeamID: \"team-1\",\n\t\tConfig: &types.Config{Quota: &types.Quota{Max: 1}},\n\t}\n\tctx := types.NewContext(context.Background(), nil)\n\n\tresult, err := exec.Execute(ctx, robot, types.TriggerHuman, nil)\n\n\tassert.NoError(t, err)\n\tassert.NotNil(t, result)\n\tassert.Equal(t, types.ExecCompleted, result.Status)\n\n\t// Human trigger skips P0 (Inspiration)\n\tassert.Nil(t, result.Inspiration, \"P0 should be skipped for human trigger\")\n\tassert.NotNil(t, result.Goals, \"P1 should be executed\")\n}\n\nfunc TestExecutorEventTriggerSkipsP0(t *testing.T) {\n\texec := NewDryRunWithDelay(0)\n\trobot := &types.Robot{\n\t\tMemberID: \"test-event\",\n\t\tTeamID: \"team-1\",\n\t\tConfig: &types.Config{Quota: &types.Quota{Max: 1}},\n\t}\n\tctx := types.NewContext(context.Background(), nil)\n\n\tresult, err := exec.Execute(ctx, robot, types.TriggerEvent, nil)\n\n\tassert.NoError(t, err)\n\tassert.Nil(t, result.Inspiration, \"P0 should be skipped for event trigger\")\n\tassert.NotNil(t, result.Goals)\n}\n\nfunc TestExecutorNilRobot(t *testing.T) {\n\texec := NewDryRunWithDelay(0)\n\tctx := types.NewContext(context.Background(), nil)\n\n\tresult, err := exec.Execute(ctx, nil, types.TriggerClock, nil)\n\n\tassert.Error(t, err)\n\tassert.Nil(t, result)\n\tassert.Contains(t, err.Error(), \"robot cannot be nil\")\n}\n\nfunc TestExecutorSimulatedFailure(t *testing.T) {\n\texec := NewDryRunWithDelay(0)\n\trobot := &types.Robot{\n\t\tMemberID: \"test-fail\",\n\t\tTeamID: \"team-1\",\n\t\tConfig: &types.Config{Quota: &types.Quota{Max: 1}},\n\t}\n\tctx := types.NewContext(context.Background(), nil)\n\n\t// Pass \"simulate_failure\" to trigger simulated failure\n\tresult, err := exec.Execute(ctx, robot, types.TriggerClock, \"simulate_failure\")\n\n\tassert.NoError(t, err) // Execute returns nil error, failure is in result\n\tassert.NotNil(t, result)\n\tassert.Equal(t, types.ExecFailed, result.Status)\n\tassert.Equal(t, \"simulated failure\", result.Error)\n}\n\nfunc TestExecutorCounters(t *testing.T) {\n\texec := NewDryRunWithDelay(0)\n\trobot := &types.Robot{\n\t\tMemberID: \"test-counter\",\n\t\tTeamID: \"team-1\",\n\t\tConfig: &types.Config{Quota: &types.Quota{Max: 10}},\n\t}\n\tctx := types.NewContext(context.Background(), nil)\n\n\tassert.Equal(t, 0, exec.ExecCount())\n\tassert.Equal(t, 0, exec.CurrentCount())\n\n\t_, _ = exec.Execute(ctx, robot, types.TriggerClock, nil)\n\tassert.Equal(t, 1, exec.ExecCount())\n\tassert.Equal(t, 0, exec.CurrentCount()) // Completed, so 0\n\n\t_, _ = exec.Execute(ctx, robot, types.TriggerClock, nil)\n\tassert.Equal(t, 2, exec.ExecCount())\n\n\texec.Reset()\n\tassert.Equal(t, 0, exec.ExecCount())\n}\n" }, { "path": "agent/robot/executor/sandbox/executor.go", "content": "package sandbox\n\nimport (\n\t\"context\"\n\t\"fmt\"\n\t\"sync/atomic\"\n\t\"time\"\n\n\t\"github.com/yaoapp/yao/agent/robot/executor/types\"\n\trobottypes \"github.com/yaoapp/yao/agent/robot/types\"\n)\n\n// Executor implements a sandboxed executor placeholder.\n//\n// ⚠️ NOT IMPLEMENTED: True sandbox mode requires container-level isolation\n// (Docker/gVisor/Firecracker) for security. This placeholder currently\n// behaves like DryRun mode and does NOT provide real security isolation.\n//\n// Future Implementation:\n// - Container isolation: Each execution in separate container\n// - Resource limits: CPU, memory, disk enforced by container runtime\n// - Network isolation: Restricted network via container networking\n// - File system isolation: Read-only root, limited writable paths\n// - Process isolation: Separate PID namespace\n//\n// Current behavior: Simulates execution with mock data (same as DryRun)\ntype Executor struct {\n\tconfig types.SandboxConfig\n\texecCount atomic.Int32\n\tcurrentCount atomic.Int32\n}\n\n// New creates a new sandbox executor with default settings\nfunc New() *Executor {\n\treturn &Executor{\n\t\tconfig: types.SandboxConfig{\n\t\t\tMaxDuration: 30 * time.Minute,\n\t\t\tNetworkAccess: true,\n\t\t\tFileAccess: false,\n\t\t},\n\t}\n}\n\n// NewWithConfig creates a sandbox executor with custom configuration\nfunc NewWithConfig(config types.SandboxConfig) *Executor {\n\treturn &Executor{\n\t\tconfig: config,\n\t}\n}\n\n// Execute runs robot execution within sandbox constraints (auto-generates ID)\nfunc (e *Executor) Execute(ctx *robottypes.Context, robot *robottypes.Robot, trigger robottypes.TriggerType, data interface{}) (*robottypes.Execution, error) {\n\treturn e.ExecuteWithControl(ctx, robot, trigger, data, \"\", nil)\n}\n\n// ExecuteWithID runs robot execution within sandbox constraints with a pre-generated execution ID (no control)\nfunc (e *Executor) ExecuteWithID(ctx *robottypes.Context, robot *robottypes.Robot, trigger robottypes.TriggerType, data interface{}, execID string) (*robottypes.Execution, error) {\n\treturn e.ExecuteWithControl(ctx, robot, trigger, data, execID, nil)\n}\n\n// ExecuteWithControl runs robot execution within sandbox constraints with execution control\nfunc (e *Executor) ExecuteWithControl(ctx *robottypes.Context, robot *robottypes.Robot, trigger robottypes.TriggerType, data interface{}, execID string, control robottypes.ExecutionControl) (*robottypes.Execution, error) {\n\tif robot == nil {\n\t\treturn nil, fmt.Errorf(\"robot cannot be nil\")\n\t}\n\n\t// Create timeout context\n\texecCtx, cancel := context.WithTimeout(ctx.Context, e.config.MaxDuration)\n\tdefer cancel()\n\n\t// Create new context with timeout\n\tsandboxCtx := robottypes.NewContext(execCtx, ctx.Auth)\n\n\t// Determine starting phase\n\tstartPhaseIndex := 0\n\tif trigger == robottypes.TriggerHuman || trigger == robottypes.TriggerEvent {\n\t\tstartPhaseIndex = 1\n\t}\n\n\t// Use provided execID or generate new one\n\tif execID == \"\" {\n\t\texecID = fmt.Sprintf(\"sandbox_%d\", time.Now().UnixNano())\n\t}\n\n\t// Create execution record\n\texec := &robottypes.Execution{\n\t\tID: execID,\n\t\tMemberID: robot.MemberID,\n\t\tTeamID: robot.TeamID,\n\t\tTriggerType: trigger,\n\t\tStartTime: time.Now(),\n\t\tStatus: robottypes.ExecPending,\n\t\tPhase: robottypes.AllPhases[startPhaseIndex],\n\t\tInput: types.BuildTriggerInput(trigger, data),\n\t}\n\n\t// Set robot reference\n\texec.SetRobot(robot)\n\n\t// Acquire slot\n\tif !robot.TryAcquireSlot(exec) {\n\t\treturn nil, robottypes.ErrQuotaExceeded\n\t}\n\tdefer robot.RemoveExecution(exec.ID)\n\n\t// Track counts\n\te.execCount.Add(1)\n\te.currentCount.Add(1)\n\tdefer e.currentCount.Add(-1)\n\n\t// Update status\n\texec.Status = robottypes.ExecRunning\n\n\t// Execute phases with sandbox constraints\n\tphases := robottypes.AllPhases[startPhaseIndex:]\n\tfor _, phase := range phases {\n\t\t// Check timeout or cancellation\n\t\tselect {\n\t\tcase <-execCtx.Done():\n\t\t\texec.Status = robottypes.ExecFailed\n\t\t\texec.Error = \"execution timeout exceeded\"\n\t\t\treturn exec, nil\n\t\tdefault:\n\t\t}\n\n\t\t// Wait if paused\n\t\tif control != nil {\n\t\t\tif err := control.WaitIfPaused(); err != nil {\n\t\t\t\texec.Status = robottypes.ExecCancelled\n\t\t\t\texec.Error = \"execution cancelled while paused\"\n\t\t\t\treturn exec, nil\n\t\t\t}\n\t\t}\n\n\t\texec.Phase = phase\n\n\t\tif e.config.OnPhaseStart != nil {\n\t\t\te.config.OnPhaseStart(phase)\n\t\t}\n\n\t\t// Execute phase with sandbox constraints\n\t\tif err := e.runSandboxedPhase(sandboxCtx, exec, phase, data); err != nil {\n\t\t\texec.Status = robottypes.ExecFailed\n\t\t\texec.Error = err.Error()\n\t\t\treturn exec, nil\n\t\t}\n\n\t\tif e.config.OnPhaseEnd != nil {\n\t\t\te.config.OnPhaseEnd(phase)\n\t\t}\n\t}\n\n\t// Mark completed\n\texec.Status = robottypes.ExecCompleted\n\tnow := time.Now()\n\texec.EndTime = &now\n\n\treturn exec, nil\n}\n\n// runSandboxedPhase executes a phase with sandbox constraints\nfunc (e *Executor) runSandboxedPhase(ctx *robottypes.Context, exec *robottypes.Execution, phase robottypes.Phase, data interface{}) error {\n\t// Validate agent is allowed (if whitelist is set)\n\tif len(e.config.AllowedAgents) > 0 {\n\t\trobot := exec.GetRobot()\n\t\tif robot != nil && robot.Config != nil && robot.Config.Resources != nil {\n\t\t\tagentID := robot.Config.Resources.GetPhaseAgent(phase)\n\t\t\tif !e.isAgentAllowed(agentID) {\n\t\t\t\treturn fmt.Errorf(\"agent %s is not allowed in sandbox\", agentID)\n\t\t\t}\n\t\t}\n\t}\n\n\t// For now, generate mock output (real implementation would call agents with restrictions)\n\te.mockPhaseOutput(exec, phase)\n\n\treturn nil\n}\n\n// isAgentAllowed checks if an agent is in the whitelist\nfunc (e *Executor) isAgentAllowed(agentID string) bool {\n\tfor _, allowed := range e.config.AllowedAgents {\n\t\tif allowed == agentID || allowed == \"*\" {\n\t\t\treturn true\n\t\t}\n\t}\n\treturn false\n}\n\n// mockPhaseOutput generates mock output for each phase\nfunc (e *Executor) mockPhaseOutput(exec *robottypes.Execution, phase robottypes.Phase) {\n\tswitch phase {\n\tcase robottypes.PhaseInspiration:\n\t\texec.Inspiration = &robottypes.InspirationReport{\n\t\t\tClock: robottypes.NewClockContext(time.Now(), \"\"),\n\t\t\tContent: \"## Sandbox Inspiration\\n\\nExecuted in isolated sandbox environment.\",\n\t\t}\n\tcase robottypes.PhaseGoals:\n\t\texec.Goals = &robottypes.Goals{\n\t\t\tContent: \"## Sandbox Goals\\n\\n1. [High] Sandboxed goal execution\",\n\t\t}\n\tcase robottypes.PhaseTasks:\n\t\texec.Tasks = []robottypes.Task{\n\t\t\t{\n\t\t\t\tID: \"sandbox-task-1\",\n\t\t\t\tGoalRef: \"Goal 1\",\n\t\t\t\tSource: robottypes.TaskSourceAuto,\n\t\t\t\tExecutorType: robottypes.ExecutorAssistant,\n\t\t\t\tExecutorID: \"sandbox-agent\",\n\t\t\t\tStatus: robottypes.TaskPending,\n\t\t\t},\n\t\t}\n\tcase robottypes.PhaseRun:\n\t\texec.Results = []robottypes.TaskResult{\n\t\t\t{\n\t\t\t\tTaskID: \"sandbox-task-1\",\n\t\t\t\tSuccess: true,\n\t\t\t\tOutput: map[string]interface{}{\"mode\": \"sandbox\", \"isolated\": true},\n\t\t\t\tDuration: 50,\n\t\t\t\tValidation: &robottypes.ValidationResult{\n\t\t\t\t\tPassed: true,\n\t\t\t\t\tScore: 1.0,\n\t\t\t\t},\n\t\t\t},\n\t\t}\n\tcase robottypes.PhaseDelivery:\n\t\texec.Delivery = &robottypes.DeliveryResult{\n\t\t\tRequestID: \"sandbox-\" + exec.ID,\n\t\t\tContent: &robottypes.DeliveryContent{\n\t\t\t\tSummary: \"Sandbox delivery completed\",\n\t\t\t\tBody: \"# Sandbox Delivery\\n\\nThis is a simulated sandbox delivery result.\",\n\t\t\t},\n\t\t\tSuccess: true,\n\t\t}\n\tcase robottypes.PhaseLearning:\n\t\texec.Learning = []robottypes.LearningEntry{\n\t\t\t{\n\t\t\t\tType: robottypes.LearnExecution,\n\t\t\t\tContent: \"Sandbox execution completed within constraints\",\n\t\t\t},\n\t\t}\n\t}\n}\n\n// ExecCount returns total execution count\nfunc (e *Executor) ExecCount() int {\n\treturn int(e.execCount.Load())\n}\n\n// CurrentCount returns currently running execution count\nfunc (e *Executor) CurrentCount() int {\n\treturn int(e.currentCount.Load())\n}\n\n// Reset resets the executor counters\nfunc (e *Executor) Reset() {\n\te.execCount.Store(0)\n\te.currentCount.Store(0)\n}\n\n// Resume is not supported in sandbox mode\nfunc (e *Executor) Resume(ctx *robottypes.Context, execID string, reply string) error {\n\treturn fmt.Errorf(\"resume is not supported in sandbox executor\")\n}\n\n// Verify Executor implements types.Executor\nvar _ types.Executor = (*Executor)(nil)\n" }, { "path": "agent/robot/executor/standard/agent.go", "content": "package standard\n\nimport (\n\t\"fmt\"\n\n\t\"github.com/yaoapp/gou/text\"\n\t\"github.com/yaoapp/yao/agent/assistant\"\n\tagentcontext \"github.com/yaoapp/yao/agent/context\"\n\t\"github.com/yaoapp/yao/agent/output/message\"\n\trobottypes \"github.com/yaoapp/yao/agent/robot/types\"\n\toauthtypes \"github.com/yaoapp/yao/openapi/oauth/types\"\n)\n\n// StreamCallback receives text chunks during streaming agent calls.\n// Return 0 to continue, non-zero to stop.\ntype StreamCallback func(chunk *StreamChunk) int\n\n// StreamChunk represents a single chunk in a streaming response.\ntype StreamChunk struct {\n\tType string // \"text\", \"thinking\", \"event\"\n\tContent string\n\tDelta bool\n}\n\n// AgentCaller provides unified interface for calling AI assistants\n// It wraps the Yao Assistant framework and handles:\n// - Getting assistant by ID\n// - Single call with messages (streaming)\n// - Multi-turn conversation with session state\n// - Parsing responses (text, JSON, Next hook data)\ntype AgentCaller struct {\n\t// SkipOutput skips sending output to client (for internal calls)\n\tSkipOutput bool\n\n\t// SkipHistory skips saving to chat history (default: true for robot)\n\t// Set to false to enable multi-turn conversation with history\n\tSkipHistory bool\n\n\t// SkipSearch skips auto search\n\tSkipSearch bool\n\n\t// ChatID is used for multi-turn conversations to maintain session state\n\t// If empty, each call is independent (no history)\n\tChatID string\n\n\t// Connector overrides the assistant's default LLM connector (from Robot.LanguageModel).\n\t// When non-empty, passed as opts.Connector to ast.Stream so the agent uses the Robot's model.\n\tConnector string\n\n\t// log is an optional structured logger; when set, Call emits agent-call logs.\n\tlog *execLogger\n}\n\n// NewAgentCaller creates a new AgentCaller with default settings (single-call mode)\nfunc NewAgentCaller() *AgentCaller {\n\treturn &AgentCaller{\n\t\tSkipOutput: true, // Robot executions don't send to UI\n\t\tSkipHistory: true, // Robot executions don't save to chat history\n\t\tSkipSearch: true, // Robot executions don't trigger auto search\n\t}\n}\n\n// NewConversationCaller creates an AgentCaller for multi-turn conversations\n// chatID is used to maintain session state across calls\n// This is useful for:\n// - P2 (Tasks): Iterative task refinement with user feedback\n// - P3 (Run): Multi-step task execution with intermediate results\nfunc NewConversationCaller(chatID string) *AgentCaller {\n\treturn &AgentCaller{\n\t\tSkipOutput: true,\n\t\tSkipHistory: false, // Enable history for multi-turn\n\t\tSkipSearch: true,\n\t\tChatID: chatID,\n\t}\n}\n\n// CallResult holds the result of an agent call\ntype CallResult struct {\n\t// Content is the raw text content from LLM completion\n\tContent string\n\n\t// Next is the data returned from Next hook (if any)\n\t// This is typically a structured response from the assistant\n\tNext interface{}\n\n\t// Response is the full response object (for advanced use)\n\tResponse *agentcontext.Response\n}\n\n// IsEmpty returns true if the result has no content\nfunc (r *CallResult) IsEmpty() bool {\n\treturn r.Content == \"\" && r.Next == nil\n}\n\n// GetText returns the text content, preferring Content over Next\nfunc (r *CallResult) GetText() string {\n\tif r.Content != \"\" {\n\t\treturn r.Content\n\t}\n\tif s, ok := r.Next.(string); ok {\n\t\treturn s\n\t}\n\tif m, ok := r.Next.(map[string]interface{}); ok {\n\t\tif content, ok := m[\"content\"].(string); ok {\n\t\t\treturn content\n\t\t}\n\t\tif data, ok := m[\"data\"].(map[string]interface{}); ok {\n\t\t\tif content, ok := data[\"content\"].(string); ok {\n\t\t\t\treturn content\n\t\t\t}\n\t\t}\n\t}\n\treturn \"\"\n}\n\n// GetJSON attempts to parse the result as JSON\n// It tries in order:\n// 1. Next hook data (already structured)\n// 2. Content parsed using gou/text.ExtractJSON (fault-tolerant)\n// Returns the parsed data and any error\nfunc (r *CallResult) GetJSON() (map[string]interface{}, error) {\n\tif r.Next != nil {\n\t\tif m, ok := r.Next.(map[string]interface{}); ok {\n\t\t\tif data, ok := m[\"data\"].(map[string]interface{}); ok {\n\t\t\t\treturn data, nil\n\t\t\t}\n\t\t\treturn m, nil\n\t\t}\n\t}\n\n\tif r.Content != \"\" {\n\t\tdata := text.ExtractJSON(r.Content)\n\t\tif data != nil {\n\t\t\tif m, ok := data.(map[string]interface{}); ok {\n\t\t\t\treturn m, nil\n\t\t\t}\n\t\t}\n\t\treturn nil, fmt.Errorf(\"content is not a JSON object\")\n\t}\n\n\treturn nil, fmt.Errorf(\"no content to parse\")\n}\n\n// GetJSONArray attempts to parse the result as JSON array\n// Similar to GetJSON but for array responses\nfunc (r *CallResult) GetJSONArray() ([]interface{}, error) {\n\t// Try Next hook data first\n\tif r.Next != nil {\n\t\tif arr, ok := r.Next.([]interface{}); ok {\n\t\t\treturn arr, nil\n\t\t}\n\t\tif m, ok := r.Next.(map[string]interface{}); ok {\n\t\t\t// Check for \"data\" wrapper\n\t\t\tif data, ok := m[\"data\"].([]interface{}); ok {\n\t\t\t\treturn data, nil\n\t\t\t}\n\t\t}\n\t}\n\n\t// Try parsing Content using gou/text (handles markdown blocks, JSON, YAML)\n\tif r.Content != \"\" {\n\t\tdata := text.ExtractJSON(r.Content)\n\t\tif data != nil {\n\t\t\tif arr, ok := data.([]interface{}); ok {\n\t\t\t\treturn arr, nil\n\t\t\t}\n\t\t}\n\t\treturn nil, fmt.Errorf(\"content is not a JSON array\")\n\t}\n\n\treturn nil, fmt.Errorf(\"no content to parse\")\n}\n\n// Call calls an assistant with messages and returns the result\n// This is the main entry point for agent calls\nfunc (c *AgentCaller) Call(ctx *robottypes.Context, assistantID string, messages []agentcontext.Message) (*CallResult, error) {\n\t// Get assistant\n\tast, err := assistant.Get(assistantID)\n\tif err != nil {\n\t\treturn nil, fmt.Errorf(\"assistant not found: %s: %w\", assistantID, err)\n\t}\n\n\t// Build options\n\topts := &agentcontext.Options{\n\t\tSkip: &agentcontext.Skip{\n\t\t\tOutput: c.SkipOutput,\n\t\t\tHistory: c.SkipHistory,\n\t\t\tSearch: c.SkipSearch,\n\t\t},\n\t\tConnector: c.Connector,\n\t}\n\n\t// Convert robot context to agent context\n\tagentCtx := c.buildAgentContext(ctx)\n\tdefer agentCtx.Release() // IMPORTANT: Release agent context to prevent resource leaks\n\n\t// Call assistant with streaming\n\tresponse, err := ast.Stream(agentCtx, messages, opts)\n\tif err != nil {\n\t\treturn nil, fmt.Errorf(\"assistant call failed: %w\", err)\n\t}\n\n\t// Build result\n\tresult := &CallResult{\n\t\tResponse: response,\n\t}\n\n\t// Extract Next hook data\n\tif response.Next != nil {\n\t\tresult.Next = response.Next\n\t}\n\n\t// Extract Content from Completion\n\tif response.Completion != nil {\n\t\tif content, ok := response.Completion.Content.(string); ok {\n\t\t\tresult.Content = content\n\t\t}\n\t}\n\n\tif c.log != nil {\n\t\tc.log.logAgentCall(assistantID, result)\n\t}\n\n\treturn result, nil\n}\n\n// CallWithMessages is a convenience method that builds messages from a single user input\nfunc (c *AgentCaller) CallWithMessages(ctx *robottypes.Context, assistantID string, userContent string) (*CallResult, error) {\n\tmessages := []agentcontext.Message{\n\t\t{\n\t\t\tRole: agentcontext.RoleUser,\n\t\t\tContent: userContent,\n\t\t},\n\t}\n\treturn c.Call(ctx, assistantID, messages)\n}\n\n// CallWithSystemAndUser calls with both system and user messages\nfunc (c *AgentCaller) CallWithSystemAndUser(ctx *robottypes.Context, assistantID string, systemContent, userContent string) (*CallResult, error) {\n\tmessages := []agentcontext.Message{\n\t\t{\n\t\t\tRole: agentcontext.RoleSystem,\n\t\t\tContent: systemContent,\n\t\t},\n\t\t{\n\t\t\tRole: agentcontext.RoleUser,\n\t\t\tContent: userContent,\n\t\t},\n\t}\n\treturn c.Call(ctx, assistantID, messages)\n}\n\n// CallStream calls an assistant with messages and streams text chunks via callback.\n// The callback receives each text delta in real-time while the response is being generated.\n// After streaming completes, the full CallResult is returned.\nfunc (c *AgentCaller) CallStream(ctx *robottypes.Context, assistantID string, messages []agentcontext.Message, streamFn StreamCallback) (*CallResult, error) {\n\tast, err := assistant.Get(assistantID)\n\tif err != nil {\n\t\treturn nil, fmt.Errorf(\"assistant not found: %s: %w\", assistantID, err)\n\t}\n\n\topts := &agentcontext.Options{\n\t\tSkip: &agentcontext.Skip{\n\t\t\tOutput: c.SkipOutput,\n\t\t\tHistory: c.SkipHistory,\n\t\t\tSearch: c.SkipSearch,\n\t\t},\n\t\tConnector: c.Connector,\n\t}\n\n\t// Hook OnMessage to intercept streaming chunks and forward to callback\n\tif streamFn != nil {\n\t\topts.OnMessage = func(msg *message.Message) int {\n\t\t\tif msg == nil {\n\t\t\t\treturn 0\n\t\t\t}\n\t\t\tswitch msg.Type {\n\t\t\tcase message.TypeText:\n\t\t\t\tif msg.Delta {\n\t\t\t\t\tcontent, _ := msg.Props[\"content\"].(string)\n\t\t\t\t\tif content != \"\" {\n\t\t\t\t\t\treturn streamFn(&StreamChunk{Type: \"text\", Content: content, Delta: true})\n\t\t\t\t\t}\n\t\t\t\t}\n\t\t\tcase message.TypeThinking:\n\t\t\t\tif msg.Delta {\n\t\t\t\t\tcontent, _ := msg.Props[\"content\"].(string)\n\t\t\t\t\tif content != \"\" {\n\t\t\t\t\t\treturn streamFn(&StreamChunk{Type: \"thinking\", Content: content, Delta: true})\n\t\t\t\t\t}\n\t\t\t\t}\n\t\t\t}\n\t\t\treturn 0\n\t\t}\n\t}\n\n\tagentCtx := c.buildAgentContext(ctx)\n\tdefer agentCtx.Release()\n\n\tresponse, err := ast.Stream(agentCtx, messages, opts)\n\tif err != nil {\n\t\treturn nil, fmt.Errorf(\"assistant call failed: %w\", err)\n\t}\n\n\tresult := &CallResult{Response: response}\n\tif response.Next != nil {\n\t\tresult.Next = response.Next\n\t}\n\tif response.Completion != nil {\n\t\tif content, ok := response.Completion.Content.(string); ok {\n\t\t\tresult.Content = content\n\t\t}\n\t}\n\n\tif c.log != nil {\n\t\tc.log.logAgentCall(assistantID, result)\n\t}\n\n\treturn result, nil\n}\n\n// CallWithMessagesStream is a convenience method that streams a single user input.\nfunc (c *AgentCaller) CallWithMessagesStream(ctx *robottypes.Context, assistantID string, userContent string, streamFn StreamCallback) (*CallResult, error) {\n\tmessages := []agentcontext.Message{\n\t\t{\n\t\t\tRole: agentcontext.RoleUser,\n\t\t\tContent: userContent,\n\t\t},\n\t}\n\treturn c.CallStream(ctx, assistantID, messages, streamFn)\n}\n\n// CallStreamRaw calls an assistant with streaming, passing raw message.Message objects\n// to the callback without any type filtering or degradation. This preserves all CUI\n// message protocol fields (chunk_id, message_id, block_id, delta_path, etc.)\n// for direct SSE passthrough to the frontend.\nfunc (c *AgentCaller) CallStreamRaw(ctx *robottypes.Context, assistantID string, messages []agentcontext.Message, onMessage agentcontext.OnMessageFunc) (*CallResult, error) {\n\tast, err := assistant.Get(assistantID)\n\tif err != nil {\n\t\treturn nil, fmt.Errorf(\"assistant not found: %s: %w\", assistantID, err)\n\t}\n\n\topts := &agentcontext.Options{\n\t\tSkip: &agentcontext.Skip{\n\t\t\tOutput: c.SkipOutput,\n\t\t\tHistory: c.SkipHistory,\n\t\t\tSearch: c.SkipSearch,\n\t\t},\n\t\tConnector: c.Connector,\n\t}\n\n\tif onMessage != nil {\n\t\topts.OnMessage = onMessage\n\t}\n\n\tagentCtx := c.buildAgentContext(ctx)\n\tdefer agentCtx.Release()\n\n\tresponse, err := ast.Stream(agentCtx, messages, opts)\n\tif err != nil {\n\t\treturn nil, fmt.Errorf(\"assistant call failed: %w\", err)\n\t}\n\n\tresult := &CallResult{Response: response}\n\tif response.Next != nil {\n\t\tresult.Next = response.Next\n\t}\n\tif response.Completion != nil {\n\t\tif content, ok := response.Completion.Content.(string); ok {\n\t\t\tresult.Content = content\n\t\t}\n\t}\n\n\tif c.log != nil {\n\t\tc.log.logAgentCall(assistantID, result)\n\t}\n\n\treturn result, nil\n}\n\n// CallWithMessagesStreamRaw is a convenience method that streams raw messages for a single user input.\nfunc (c *AgentCaller) CallWithMessagesStreamRaw(ctx *robottypes.Context, assistantID string, userContent string, onMessage agentcontext.OnMessageFunc) (*CallResult, error) {\n\tmessages := []agentcontext.Message{\n\t\t{\n\t\t\tRole: agentcontext.RoleUser,\n\t\t\tContent: userContent,\n\t\t},\n\t}\n\treturn c.CallStreamRaw(ctx, assistantID, messages, onMessage)\n}\n\n// buildAgentContext converts robot context to agent context\nfunc (c *AgentCaller) buildAgentContext(ctx *robottypes.Context) *agentcontext.Context {\n\t// Build authorized info for agent context\n\tvar authorized *oauthtypes.AuthorizedInfo\n\tif ctx.Auth != nil {\n\t\tauthorized = &oauthtypes.AuthorizedInfo{\n\t\t\tUserID: ctx.Auth.UserID,\n\t\t\tTeamID: ctx.Auth.TeamID,\n\t\t}\n\t}\n\n\t// Create a new agent context\n\t// Use ChatID for multi-turn conversations, empty for single calls\n\tagentCtx := agentcontext.New(ctx.Context, authorized, c.ChatID)\n\n\t// Set locale if available\n\tif ctx.Locale != \"\" {\n\t\tagentCtx.Locale = ctx.Locale\n\t}\n\n\t// Use noop logger to suppress LLM debug output for robot executions\n\t// Robot executions run in background and don't need console output\n\tif agentCtx.Logger != nil {\n\t\tagentCtx.Logger.Close()\n\t}\n\tagentCtx.Logger = agentcontext.Noop()\n\n\treturn agentCtx\n}\n\n// ExtractCodeBlock extracts the first code block from content using gou/text\n// Returns the CodeBlock with type, content, and parsed data (for JSON/YAML)\nfunc ExtractCodeBlock(content string) *text.CodeBlock {\n\treturn text.ExtractFirst(content)\n}\n\n// ExtractAllCodeBlocks extracts all code blocks from content using gou/text\nfunc ExtractAllCodeBlocks(content string) []text.CodeBlock {\n\treturn text.Extract(content)\n}\n\n// ============================================================================\n// Conversation - Multi-turn dialogue support\n// ============================================================================\n\n// Conversation manages a multi-turn dialogue with an assistant\n// Useful for:\n// - P2 (Tasks): Iterative task planning with clarification\n// - P3 (Run): Multi-step execution with intermediate validation\n// - Complex reasoning that requires back-and-forth\ntype Conversation struct {\n\tcaller *AgentCaller\n\tassistantID string\n\tmessages []agentcontext.Message\n\tmaxTurns int\n}\n\n// TurnResult holds the result of a single conversation turn\ntype TurnResult struct {\n\tTurn int // Turn number (1-based)\n\tInput string // User input for this turn\n\tResult *CallResult // Agent response\n\tMessages []agentcontext.Message // Full message history after this turn\n}\n\n// NewConversation creates a new multi-turn conversation\n// assistantID: the assistant to converse with\n// chatID: session ID for maintaining state (use exec.ID for robot executions)\n// maxTurns: maximum number of turns (0 = unlimited)\nfunc NewConversation(assistantID, chatID string, maxTurns int) *Conversation {\n\treturn &Conversation{\n\t\tcaller: NewConversationCaller(chatID),\n\t\tassistantID: assistantID,\n\t\tmessages: make([]agentcontext.Message, 0),\n\t\tmaxTurns: maxTurns,\n\t}\n}\n\n// WithCaller sets a custom AgentCaller for the conversation\n// Useful for customizing SkipSearch or other options\nfunc (c *Conversation) WithCaller(caller *AgentCaller) *Conversation {\n\tc.caller = caller\n\treturn c\n}\n\n// WithSystemPrompt adds a system prompt at the beginning of the conversation\nfunc (c *Conversation) WithSystemPrompt(systemPrompt string) *Conversation {\n\tif systemPrompt != \"\" {\n\t\tc.messages = append([]agentcontext.Message{{\n\t\t\tRole: agentcontext.RoleSystem,\n\t\t\tContent: systemPrompt,\n\t\t}}, c.messages...)\n\t}\n\treturn c\n}\n\n// WithHistory initializes the conversation with existing message history\n// Note: Message structs are copied, but Content (interface{}) is a shallow copy\nfunc (c *Conversation) WithHistory(messages []agentcontext.Message) *Conversation {\n\tc.messages = append(c.messages, messages...)\n\treturn c\n}\n\n// Turn executes a single turn in the conversation\n// userInput: the user's message for this turn\n// Returns the turn result with agent response\nfunc (c *Conversation) Turn(ctx *robottypes.Context, userInput string) (*TurnResult, error) {\n\t// Check max turns\n\tturnNum := c.TurnCount() + 1\n\tif c.maxTurns > 0 && turnNum > c.maxTurns {\n\t\treturn nil, fmt.Errorf(\"max turns (%d) exceeded\", c.maxTurns)\n\t}\n\n\t// Build messages with user input (don't modify history yet)\n\tuserMsg := agentcontext.Message{\n\t\tRole: agentcontext.RoleUser,\n\t\tContent: userInput,\n\t}\n\t// Create a new slice to avoid modifying c.messages if capacity allows append in-place\n\tmessagesWithInput := make([]agentcontext.Message, len(c.messages)+1)\n\tcopy(messagesWithInput, c.messages)\n\tmessagesWithInput[len(c.messages)] = userMsg\n\n\t// Call assistant with full history\n\tresult, err := c.caller.Call(ctx, c.assistantID, messagesWithInput)\n\tif err != nil {\n\t\treturn nil, fmt.Errorf(\"turn %d failed: %w\", turnNum, err)\n\t}\n\n\t// Only update history after successful call\n\tc.messages = append(c.messages, userMsg)\n\n\t// Add assistant response to history\n\tif result.Content != \"\" {\n\t\tc.messages = append(c.messages, agentcontext.Message{\n\t\t\tRole: agentcontext.RoleAssistant,\n\t\t\tContent: result.Content,\n\t\t})\n\t}\n\n\t// Return a copy of messages to prevent external modification\n\tmessagesCopy := make([]agentcontext.Message, len(c.messages))\n\tcopy(messagesCopy, c.messages)\n\n\treturn &TurnResult{\n\t\tTurn: turnNum,\n\t\tInput: userInput,\n\t\tResult: result,\n\t\tMessages: messagesCopy,\n\t}, nil\n}\n\n// TurnCount returns the number of user turns so far\nfunc (c *Conversation) TurnCount() int {\n\tcount := 0\n\tfor _, msg := range c.messages {\n\t\tif msg.Role == agentcontext.RoleUser {\n\t\t\tcount++\n\t\t}\n\t}\n\treturn count\n}\n\n// Messages returns a copy of the current message history\nfunc (c *Conversation) Messages() []agentcontext.Message {\n\tmessagesCopy := make([]agentcontext.Message, len(c.messages))\n\tcopy(messagesCopy, c.messages)\n\treturn messagesCopy\n}\n\n// LastResponse returns a copy of the last assistant response, or nil if none\nfunc (c *Conversation) LastResponse() *agentcontext.Message {\n\tfor i := len(c.messages) - 1; i >= 0; i-- {\n\t\tif c.messages[i].Role == agentcontext.RoleAssistant {\n\t\t\t// Return a copy to prevent external modification\n\t\t\tmsg := c.messages[i]\n\t\t\treturn &msg\n\t\t}\n\t}\n\treturn nil\n}\n\n// Reset clears the conversation history (keeps system prompt if any)\nfunc (c *Conversation) Reset() {\n\t// Keep system prompt if present\n\tvar systemPrompt *agentcontext.Message\n\tif len(c.messages) > 0 && c.messages[0].Role == agentcontext.RoleSystem {\n\t\tsystemPrompt = &c.messages[0]\n\t}\n\n\tc.messages = make([]agentcontext.Message, 0)\n\tif systemPrompt != nil {\n\t\tc.messages = append(c.messages, *systemPrompt)\n\t}\n}\n\n// RunUntil runs the conversation until a condition is met\n// checkFn: called after each turn, returns (done, error)\n// Returns all turn results\nfunc (c *Conversation) RunUntil(\n\tctx *robottypes.Context,\n\tinputFn func(turn int, lastResult *CallResult) (string, error),\n\tcheckFn func(turn int, result *CallResult) (done bool, err error),\n) ([]*TurnResult, error) {\n\tvar results []*TurnResult\n\n\tfor {\n\t\tturnNum := c.TurnCount() + 1\n\n\t\t// Check max turns\n\t\tif c.maxTurns > 0 && turnNum > c.maxTurns {\n\t\t\treturn results, fmt.Errorf(\"max turns (%d) exceeded without completion\", c.maxTurns)\n\t\t}\n\n\t\t// Get input for this turn\n\t\tvar lastResult *CallResult\n\t\tif len(results) > 0 {\n\t\t\tlastResult = results[len(results)-1].Result\n\t\t}\n\n\t\tinput, err := inputFn(turnNum, lastResult)\n\t\tif err != nil {\n\t\t\treturn results, fmt.Errorf(\"input generation failed at turn %d: %w\", turnNum, err)\n\t\t}\n\n\t\t// Execute turn\n\t\tturnResult, err := c.Turn(ctx, input)\n\t\tif err != nil {\n\t\t\treturn results, err\n\t\t}\n\t\tresults = append(results, turnResult)\n\n\t\t// Check completion condition\n\t\tdone, err := checkFn(turnNum, turnResult.Result)\n\t\tif err != nil {\n\t\t\treturn results, fmt.Errorf(\"check failed at turn %d: %w\", turnNum, err)\n\t\t}\n\t\tif done {\n\t\t\treturn results, nil\n\t\t}\n\t}\n}\n" }, { "path": "agent/robot/executor/standard/agent_stream_test.go", "content": "package standard_test\n\nimport (\n\t\"context\"\n\t\"strings\"\n\t\"sync\"\n\t\"testing\"\n\n\t\"github.com/stretchr/testify/assert\"\n\t\"github.com/stretchr/testify/require\"\n\tagentcontext \"github.com/yaoapp/yao/agent/context\"\n\t\"github.com/yaoapp/yao/agent/robot/executor/standard\"\n\t\"github.com/yaoapp/yao/agent/robot/types\"\n\t\"github.com/yaoapp/yao/agent/testutils\"\n)\n\nfunc TestAgentCallerCallStream(t *testing.T) {\n\tif testing.Short() {\n\t\tt.Skip(\"Skipping integration test (requires LLM)\")\n\t}\n\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\tcaller := standard.NewAgentCaller()\n\tctx := types.NewContext(context.Background(), testAuth())\n\n\tt.Run(\"streams text chunks and returns result\", func(t *testing.T) {\n\t\tvar mu sync.Mutex\n\t\tvar chunks []string\n\n\t\tstreamFn := func(chunk *standard.StreamChunk) int {\n\t\t\tmu.Lock()\n\t\t\tdefer mu.Unlock()\n\t\t\tif chunk.Type == \"text\" && chunk.Delta {\n\t\t\t\tchunks = append(chunks, chunk.Content)\n\t\t\t}\n\t\t\treturn 0\n\t\t}\n\n\t\tresult, err := caller.CallWithMessagesStream(ctx, \"tests.robot-single\", \"Hello, test message\", streamFn)\n\n\t\trequire.NoError(t, err)\n\t\trequire.NotNil(t, result)\n\t\tassert.False(t, result.IsEmpty())\n\n\t\tmu.Lock()\n\t\tcombined := strings.Join(chunks, \"\")\n\t\tchunkCount := len(chunks)\n\t\tmu.Unlock()\n\n\t\tt.Logf(\"Received %d text chunks, total length: %d\", chunkCount, len(combined))\n\t\tassert.Greater(t, chunkCount, 0, \"should have received at least one text chunk\")\n\t\tassert.NotEmpty(t, combined, \"combined chunks should not be empty\")\n\t})\n\n\tt.Run(\"nil callback works like non-stream call\", func(t *testing.T) {\n\t\tresult, err := caller.CallStream(ctx, \"tests.robot-single\",\n\t\t\t[]agentcontext.Message{{Role: \"user\", Content: \"Hello\"}},\n\t\t\tnil,\n\t\t)\n\t\trequire.NoError(t, err)\n\t\trequire.NotNil(t, result)\n\t\tassert.False(t, result.IsEmpty())\n\t})\n\n\tt.Run(\"stream returns parseable JSON\", func(t *testing.T) {\n\t\tvar mu sync.Mutex\n\t\tvar chunks []string\n\n\t\tstreamFn := func(chunk *standard.StreamChunk) int {\n\t\t\tmu.Lock()\n\t\t\tdefer mu.Unlock()\n\t\t\tif chunk.Type == \"text\" && chunk.Delta {\n\t\t\t\tchunks = append(chunks, chunk.Content)\n\t\t\t}\n\t\t\treturn 0\n\t\t}\n\n\t\tresult, err := caller.CallWithMessagesStream(ctx, \"tests.robot-single\", \"Generate inspiration report\", streamFn)\n\n\t\trequire.NoError(t, err)\n\t\trequire.NotNil(t, result)\n\n\t\tdata, err := result.GetJSON()\n\t\trequire.NoError(t, err)\n\t\tassert.NotNil(t, data)\n\t\tassert.Contains(t, data, \"type\")\n\n\t\tmu.Lock()\n\t\tchunkCount := len(chunks)\n\t\tmu.Unlock()\n\t\tt.Logf(\"Received %d chunks for JSON response\", chunkCount)\n\t})\n\n\tt.Run(\"assistant not found returns error\", func(t *testing.T) {\n\t\tresult, err := caller.CallWithMessagesStream(ctx, \"non.existent\", \"hello\", nil)\n\t\tassert.Error(t, err)\n\t\tassert.Nil(t, result)\n\t\tassert.Contains(t, err.Error(), \"assistant not found\")\n\t})\n}\n" }, { "path": "agent/robot/executor/standard/agent_test.go", "content": "package standard_test\n\nimport (\n\t\"context\"\n\t\"testing\"\n\n\t\"github.com/stretchr/testify/assert\"\n\t\"github.com/stretchr/testify/require\"\n\t\"github.com/yaoapp/yao/agent/robot/executor/standard\"\n\t\"github.com/yaoapp/yao/agent/robot/types\"\n\t\"github.com/yaoapp/yao/agent/testutils\"\n\toauthtypes \"github.com/yaoapp/yao/openapi/oauth/types\"\n)\n\n// testAuth returns a test auth info for agent calls\nfunc testAuth() *oauthtypes.AuthorizedInfo {\n\treturn &oauthtypes.AuthorizedInfo{\n\t\tUserID: \"test-user-1\",\n\t\tTeamID: \"test-team-1\",\n\t}\n}\n\n// ============================================================================\n// AgentCaller Tests - Single Call Mode\n// ============================================================================\n\nfunc TestAgentCallerSingleCall(t *testing.T) {\n\tif testing.Short() {\n\t\tt.Skip(\"Skipping integration test\")\n\t}\n\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\tcaller := standard.NewAgentCaller()\n\tctx := types.NewContext(context.Background(), testAuth())\n\n\t// Test basic call - verify assistant responds and returns parseable JSON\n\t// Note: LLM outputs are non-deterministic, so we test structure not exact values\n\tt.Run(\"basic call returns response\", func(t *testing.T) {\n\t\tresult, err := caller.CallWithMessages(ctx, \"tests.robot-single\", \"Hello, test message\")\n\n\t\trequire.NoError(t, err)\n\t\trequire.NotNil(t, result)\n\t\tassert.False(t, result.IsEmpty(), \"result should not be empty\")\n\n\t\t// Should be able to get text content\n\t\ttext := result.GetText()\n\t\tassert.NotEmpty(t, text, \"should have text content\")\n\t})\n\n\tt.Run(\"call returns parseable JSON\", func(t *testing.T) {\n\t\tresult, err := caller.CallWithMessages(ctx, \"tests.robot-single\", \"Generate inspiration report\")\n\n\t\trequire.NoError(t, err)\n\t\trequire.NotNil(t, result)\n\n\t\t// Should return parseable JSON (content may vary)\n\t\tdata, err := result.GetJSON()\n\t\trequire.NoError(t, err)\n\t\tassert.NotNil(t, data)\n\t\t// Verify it has \"type\" field (all test responses should have this)\n\t\tassert.Contains(t, data, \"type\", \"response should have type field\")\n\t})\n}\n\nfunc TestAgentCallerNextHookData(t *testing.T) {\n\tif testing.Short() {\n\t\tt.Skip(\"Skipping integration test\")\n\t}\n\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\tcaller := standard.NewAgentCaller()\n\tctx := types.NewContext(context.Background(), testAuth())\n\n\tt.Run(\"next_hook inspiration returns structured data\", func(t *testing.T) {\n\t\tresult, err := caller.CallWithMessages(ctx, \"tests.robot-single\", \"next_hook inspiration test\")\n\n\t\trequire.NoError(t, err)\n\t\trequire.NotNil(t, result)\n\n\t\t// Next hook should return structured data\n\t\tdata, err := result.GetJSON()\n\t\trequire.NoError(t, err)\n\t\tassert.Equal(t, \"inspiration\", data[\"type\"])\n\t\tassert.Equal(t, \"next_hook\", data[\"source\"])\n\t})\n\n\tt.Run(\"next_hook goals returns structured data\", func(t *testing.T) {\n\t\tresult, err := caller.CallWithMessages(ctx, \"tests.robot-single\", \"next_hook goals test\")\n\n\t\trequire.NoError(t, err)\n\t\trequire.NotNil(t, result)\n\n\t\tdata, err := result.GetJSON()\n\t\trequire.NoError(t, err)\n\t\tassert.Equal(t, \"goals\", data[\"type\"])\n\t\tassert.Equal(t, \"next_hook\", data[\"source\"])\n\t})\n\n\tt.Run(\"next_hook tasks returns structured data\", func(t *testing.T) {\n\t\tresult, err := caller.CallWithMessages(ctx, \"tests.robot-single\", \"next_hook tasks test\")\n\n\t\trequire.NoError(t, err)\n\t\trequire.NotNil(t, result)\n\n\t\tdata, err := result.GetJSON()\n\t\trequire.NoError(t, err)\n\t\tassert.Equal(t, \"tasks\", data[\"type\"])\n\t\tassert.Equal(t, \"next_hook\", data[\"source\"])\n\t})\n}\n\nfunc TestAgentCallerJSONArray(t *testing.T) {\n\tif testing.Short() {\n\t\tt.Skip(\"Skipping integration test\")\n\t}\n\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\tcaller := standard.NewAgentCaller()\n\tctx := types.NewContext(context.Background(), testAuth())\n\n\tt.Run(\"array_test returns JSON array\", func(t *testing.T) {\n\t\tresult, err := caller.CallWithMessages(ctx, \"tests.robot-single\", \"array_test\")\n\n\t\trequire.NoError(t, err)\n\t\trequire.NotNil(t, result)\n\n\t\tarr, err := result.GetJSONArray()\n\t\trequire.NoError(t, err)\n\t\tassert.Len(t, arr, 3)\n\n\t\t// Verify first item structure\n\t\titem1, ok := arr[0].(map[string]interface{})\n\t\trequire.True(t, ok, \"first item should be a map\")\n\t\tassert.Equal(t, float64(1), item1[\"id\"])\n\t\tassert.Equal(t, \"Item 1\", item1[\"name\"])\n\t})\n}\n\nfunc TestAgentCallerEmptyResponse(t *testing.T) {\n\tif testing.Short() {\n\t\tt.Skip(\"Skipping integration test\")\n\t}\n\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\tcaller := standard.NewAgentCaller()\n\tctx := types.NewContext(context.Background(), testAuth())\n\n\tt.Run(\"empty_test falls back to completion content\", func(t *testing.T) {\n\t\tresult, err := caller.CallWithMessages(ctx, \"tests.robot-single\", \"empty_test\")\n\n\t\trequire.NoError(t, err)\n\t\trequire.NotNil(t, result)\n\t\t// When Next hook returns null, should use Completion content\n\t\tassert.False(t, result.IsEmpty())\n\t})\n}\n\nfunc TestAgentCallerAssistantNotFound(t *testing.T) {\n\tif testing.Short() {\n\t\tt.Skip(\"Skipping integration test\")\n\t}\n\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\tcaller := standard.NewAgentCaller()\n\tctx := types.NewContext(context.Background(), testAuth())\n\n\tt.Run(\"non-existent assistant returns error\", func(t *testing.T) {\n\t\tresult, err := caller.CallWithMessages(ctx, \"non.existent.assistant\", \"hello\")\n\n\t\tassert.Error(t, err)\n\t\tassert.Nil(t, result)\n\t\tassert.Contains(t, err.Error(), \"assistant not found\")\n\t})\n}\n\nfunc TestAgentCallerWithSystemAndUser(t *testing.T) {\n\tif testing.Short() {\n\t\tt.Skip(\"Skipping integration test\")\n\t}\n\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\tcaller := standard.NewAgentCaller()\n\tctx := types.NewContext(context.Background(), testAuth())\n\n\tt.Run(\"call with system and user messages\", func(t *testing.T) {\n\t\tresult, err := caller.CallWithSystemAndUser(\n\t\t\tctx,\n\t\t\t\"tests.robot-single\",\n\t\t\t\"You are a helpful assistant.\",\n\t\t\t\"Generate inspiration report\",\n\t\t)\n\n\t\trequire.NoError(t, err)\n\t\trequire.NotNil(t, result)\n\t\tassert.False(t, result.IsEmpty())\n\t})\n}\n\n// ============================================================================\n// Conversation Tests - Multi-Turn Mode\n// ============================================================================\n\nfunc TestConversationMultiTurn(t *testing.T) {\n\tif testing.Short() {\n\t\tt.Skip(\"Skipping integration test\")\n\t}\n\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\tctx := types.NewContext(context.Background(), testAuth())\n\n\tt.Run(\"multi-turn conversation maintains state\", func(t *testing.T) {\n\t\tconv := standard.NewConversation(\"tests.robot-conversation\", \"test-conv-1\", 10)\n\n\t\t// Turn 1: Start planning\n\t\tturn1, err := conv.Turn(ctx, \"Plan tasks for sending weekly report\")\n\t\trequire.NoError(t, err)\n\t\trequire.NotNil(t, turn1)\n\t\tassert.Equal(t, 1, turn1.Turn)\n\n\t\tdata1, err := turn1.Result.GetJSON()\n\t\trequire.NoError(t, err)\n\t\t// Verify basic structure - turn number and completed flag\n\t\tassert.Contains(t, data1, \"turn\")\n\t\tassert.Contains(t, data1, \"status\")\n\t\tassert.Contains(t, data1, \"completed\")\n\n\t\t// Turn 2: Continue conversation\n\t\tturn2, err := conv.Turn(ctx, \"Send to managers, include sales data\")\n\t\trequire.NoError(t, err)\n\t\trequire.NotNil(t, turn2)\n\t\tassert.Equal(t, 2, turn2.Turn)\n\n\t\tdata2, err := turn2.Result.GetJSON()\n\t\trequire.NoError(t, err)\n\t\tassert.Contains(t, data2, \"turn\")\n\t\tassert.Contains(t, data2, \"status\")\n\n\t\t// Turn 3: Complete with confirm/skip\n\t\tturn3, err := conv.Turn(ctx, \"skip\") // Use skip for deterministic completion\n\t\trequire.NoError(t, err)\n\t\trequire.NotNil(t, turn3)\n\t\tassert.Equal(t, 3, turn3.Turn)\n\n\t\tdata3, err := turn3.Result.GetJSON()\n\t\trequire.NoError(t, err)\n\t\tassert.Equal(t, \"completed\", data3[\"status\"])\n\t\tassert.Equal(t, true, data3[\"completed\"])\n\t})\n}\n\nfunc TestConversationTurnCount(t *testing.T) {\n\tif testing.Short() {\n\t\tt.Skip(\"Skipping integration test\")\n\t}\n\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\tctx := types.NewContext(context.Background(), testAuth())\n\n\tt.Run(\"turn count increments correctly\", func(t *testing.T) {\n\t\tconv := standard.NewConversation(\"tests.robot-conversation\", \"test-conv-2\", 10)\n\n\t\tassert.Equal(t, 0, conv.TurnCount())\n\n\t\t_, err := conv.Turn(ctx, \"First message\")\n\t\trequire.NoError(t, err)\n\t\tassert.Equal(t, 1, conv.TurnCount())\n\n\t\t_, err = conv.Turn(ctx, \"Second message\")\n\t\trequire.NoError(t, err)\n\t\tassert.Equal(t, 2, conv.TurnCount())\n\t})\n}\n\nfunc TestConversationMaxTurns(t *testing.T) {\n\tif testing.Short() {\n\t\tt.Skip(\"Skipping integration test\")\n\t}\n\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\tctx := types.NewContext(context.Background(), testAuth())\n\n\tt.Run(\"exceeding max turns returns error\", func(t *testing.T) {\n\t\tconv := standard.NewConversation(\"tests.robot-conversation\", \"test-conv-3\", 2)\n\n\t\t_, err := conv.Turn(ctx, \"First\")\n\t\trequire.NoError(t, err)\n\n\t\t_, err = conv.Turn(ctx, \"Second\")\n\t\trequire.NoError(t, err)\n\n\t\t// Third turn should fail\n\t\t_, err = conv.Turn(ctx, \"Third\")\n\t\tassert.Error(t, err)\n\t\tassert.Contains(t, err.Error(), \"max turns (2) exceeded\")\n\t})\n}\n\nfunc TestConversationMessages(t *testing.T) {\n\tif testing.Short() {\n\t\tt.Skip(\"Skipping integration test\")\n\t}\n\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\tctx := types.NewContext(context.Background(), testAuth())\n\n\tt.Run(\"messages history is maintained\", func(t *testing.T) {\n\t\tconv := standard.NewConversation(\"tests.robot-conversation\", \"test-conv-4\", 5)\n\n\t\t// Initially empty\n\t\tassert.Empty(t, conv.Messages())\n\n\t\t// After first turn\n\t\t_, err := conv.Turn(ctx, \"Hello\")\n\t\trequire.NoError(t, err)\n\n\t\tmsgs := conv.Messages()\n\t\tassert.GreaterOrEqual(t, len(msgs), 1) // At least user message\n\t})\n}\n\nfunc TestConversationLastResponse(t *testing.T) {\n\tif testing.Short() {\n\t\tt.Skip(\"Skipping integration test\")\n\t}\n\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\tctx := types.NewContext(context.Background(), testAuth())\n\n\tt.Run(\"last response returns assistant message\", func(t *testing.T) {\n\t\tconv := standard.NewConversation(\"tests.robot-conversation\", \"test-conv-5\", 5)\n\n\t\t// No response yet\n\t\tassert.Nil(t, conv.LastResponse())\n\n\t\t// After turn\n\t\t_, err := conv.Turn(ctx, \"Start planning\")\n\t\trequire.NoError(t, err)\n\n\t\tlastResp := conv.LastResponse()\n\t\tassert.NotNil(t, lastResp)\n\t\tassert.Equal(t, \"assistant\", string(lastResp.Role))\n\t})\n}\n\nfunc TestConversationReset(t *testing.T) {\n\tif testing.Short() {\n\t\tt.Skip(\"Skipping integration test\")\n\t}\n\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\tctx := types.NewContext(context.Background(), testAuth())\n\n\tt.Run(\"reset clears conversation history\", func(t *testing.T) {\n\t\tconv := standard.NewConversation(\"tests.robot-conversation\", \"test-conv-6\", 5)\n\n\t\t_, err := conv.Turn(ctx, \"First message\")\n\t\trequire.NoError(t, err)\n\t\tassert.Equal(t, 1, conv.TurnCount())\n\n\t\tconv.Reset()\n\t\tassert.Equal(t, 0, conv.TurnCount())\n\t\tassert.Empty(t, conv.Messages())\n\t})\n}\n\nfunc TestConversationWithSystemPrompt(t *testing.T) {\n\tif testing.Short() {\n\t\tt.Skip(\"Skipping integration test\")\n\t}\n\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\tctx := types.NewContext(context.Background(), testAuth())\n\n\tt.Run(\"system prompt is preserved after reset\", func(t *testing.T) {\n\t\tconv := standard.NewConversation(\"tests.robot-conversation\", \"test-conv-7\", 5).\n\t\t\tWithSystemPrompt(\"You are a task planner.\")\n\n\t\tmsgs := conv.Messages()\n\t\trequire.Len(t, msgs, 1)\n\t\tassert.Equal(t, \"system\", string(msgs[0].Role))\n\n\t\t_, err := conv.Turn(ctx, \"Hello\")\n\t\trequire.NoError(t, err)\n\n\t\tconv.Reset()\n\n\t\t// System prompt should be preserved\n\t\tmsgs = conv.Messages()\n\t\trequire.Len(t, msgs, 1)\n\t\tassert.Equal(t, \"system\", string(msgs[0].Role))\n\t})\n}\n\nfunc TestConversationSpecialCommands(t *testing.T) {\n\tif testing.Short() {\n\t\tt.Skip(\"Skipping integration test\")\n\t}\n\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\tctx := types.NewContext(context.Background(), testAuth())\n\n\tt.Run(\"skip command jumps to completed\", func(t *testing.T) {\n\t\tconv := standard.NewConversation(\"tests.robot-conversation\", \"test-conv-8\", 5)\n\n\t\tturn, err := conv.Turn(ctx, \"skip\")\n\t\trequire.NoError(t, err)\n\n\t\tdata, err := turn.Result.GetJSON()\n\t\trequire.NoError(t, err)\n\t\tassert.Equal(t, \"completed\", data[\"status\"])\n\t\tassert.Equal(t, true, data[\"completed\"])\n\t})\n\n\tt.Run(\"abort command ends conversation\", func(t *testing.T) {\n\t\tconv := standard.NewConversation(\"tests.robot-conversation\", \"test-conv-9\", 5)\n\n\t\tturn, err := conv.Turn(ctx, \"abort\")\n\t\trequire.NoError(t, err)\n\n\t\tdata, err := turn.Result.GetJSON()\n\t\trequire.NoError(t, err)\n\t\tassert.Equal(t, \"aborted\", data[\"status\"])\n\t\tassert.Equal(t, true, data[\"completed\"])\n\t})\n\n\tt.Run(\"reset command resets conversation state\", func(t *testing.T) {\n\t\tconv := standard.NewConversation(\"tests.robot-conversation\", \"test-conv-10\", 5)\n\n\t\t// First do a turn\n\t\t_, err := conv.Turn(ctx, \"Start planning\")\n\t\trequire.NoError(t, err)\n\n\t\t// Then reset via command\n\t\tturn, err := conv.Turn(ctx, \"reset\")\n\t\trequire.NoError(t, err)\n\n\t\tdata, err := turn.Result.GetJSON()\n\t\trequire.NoError(t, err)\n\t\tassert.Equal(t, \"reset\", data[\"status\"])\n\t})\n}\n\nfunc TestConversationRunUntil(t *testing.T) {\n\tif testing.Short() {\n\t\tt.Skip(\"Skipping integration test\")\n\t}\n\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\tctx := types.NewContext(context.Background(), testAuth())\n\n\tt.Run(\"run until completion\", func(t *testing.T) {\n\t\tconv := standard.NewConversation(\"tests.robot-conversation\", \"test-conv-11\", 10)\n\n\t\tinputs := []string{\n\t\t\t\"Plan weekly report tasks\",\n\t\t\t\"Send to team leads, include metrics\",\n\t\t\t\"confirm\",\n\t\t}\n\t\tinputIdx := 0\n\n\t\tresults, err := conv.RunUntil(\n\t\t\tctx,\n\t\t\tfunc(turn int, lastResult *standard.CallResult) (string, error) {\n\t\t\t\tif inputIdx < len(inputs) {\n\t\t\t\t\tinput := inputs[inputIdx]\n\t\t\t\t\tinputIdx++\n\t\t\t\t\treturn input, nil\n\t\t\t\t}\n\t\t\t\treturn \"confirm\", nil\n\t\t\t},\n\t\t\tfunc(turn int, result *standard.CallResult) (bool, error) {\n\t\t\t\tdata, err := result.GetJSON()\n\t\t\t\tif err != nil {\n\t\t\t\t\treturn false, nil\n\t\t\t\t}\n\t\t\t\tcompleted, ok := data[\"completed\"].(bool)\n\t\t\t\treturn ok && completed, nil\n\t\t\t},\n\t\t)\n\n\t\trequire.NoError(t, err)\n\t\trequire.Len(t, results, 3, \"should complete in 3 turns\")\n\n\t\t// Final result should be completed\n\t\tfinalData, err := results[len(results)-1].Result.GetJSON()\n\t\trequire.NoError(t, err)\n\t\tassert.Equal(t, true, finalData[\"completed\"])\n\t})\n}\n\n// ============================================================================\n// CallResult Tests\n// ============================================================================\n\nfunc TestCallResultGetText(t *testing.T) {\n\tt.Run(\"returns content when available\", func(t *testing.T) {\n\t\tresult := &standard.CallResult{Content: \"Hello World\"}\n\t\tassert.Equal(t, \"Hello World\", result.GetText())\n\t})\n\n\tt.Run(\"returns empty for empty result\", func(t *testing.T) {\n\t\tresult := &standard.CallResult{}\n\t\tassert.Equal(t, \"\", result.GetText())\n\t})\n}\n\nfunc TestCallResultIsEmpty(t *testing.T) {\n\tt.Run(\"empty when no content and no next\", func(t *testing.T) {\n\t\tresult := &standard.CallResult{}\n\t\tassert.True(t, result.IsEmpty())\n\t})\n\n\tt.Run(\"not empty when has content\", func(t *testing.T) {\n\t\tresult := &standard.CallResult{Content: \"test\"}\n\t\tassert.False(t, result.IsEmpty())\n\t})\n\n\tt.Run(\"not empty when has next\", func(t *testing.T) {\n\t\tresult := &standard.CallResult{Next: map[string]interface{}{\"key\": \"value\"}}\n\t\tassert.False(t, result.IsEmpty())\n\t})\n}\n\n// ============================================================================\n// ExtractCodeBlock Tests\n// ============================================================================\n\nfunc TestExtractCodeBlock(t *testing.T) {\n\tt.Run(\"extracts JSON code block\", func(t *testing.T) {\n\t\tcontent := \"Here is the result:\\n```json\\n{\\\"key\\\": \\\"value\\\"}\\n```\"\n\t\tblock := standard.ExtractCodeBlock(content)\n\n\t\trequire.NotNil(t, block)\n\t\tassert.Equal(t, \"json\", block.Type)\n\t\tassert.Contains(t, block.Content, \"key\")\n\t})\n\n\tt.Run(\"returns nil for no code block\", func(t *testing.T) {\n\t\tcontent := \"Just plain text\"\n\t\tblock := standard.ExtractCodeBlock(content)\n\n\t\t// gou/text returns text type for plain text\n\t\trequire.NotNil(t, block)\n\t\tassert.Equal(t, \"text\", block.Type)\n\t})\n}\n\nfunc TestExtractAllCodeBlocks(t *testing.T) {\n\tt.Run(\"extracts multiple code blocks\", func(t *testing.T) {\n\t\tcontent := \"```json\\n{}\\n```\\n\\n```python\\nprint('hello')\\n```\"\n\t\tblocks := standard.ExtractAllCodeBlocks(content)\n\n\t\tassert.Len(t, blocks, 2)\n\t})\n}\n" }, { "path": "agent/robot/executor/standard/delivery.go", "content": "package standard\n\nimport (\n\t\"encoding/json\"\n\t\"fmt\"\n\t\"strings\"\n\t\"time\"\n\n\t\"github.com/yaoapp/gou/model\"\n\tkunlog \"github.com/yaoapp/kun/log\"\n\trobotevents \"github.com/yaoapp/yao/agent/robot/events\"\n\trobottypes \"github.com/yaoapp/yao/agent/robot/types\"\n\t\"github.com/yaoapp/yao/event\"\n)\n\n// RunDelivery executes P4: Delivery phase\n//\n// Process:\n// 1. Call Delivery Agent with full execution context\n// 2. Agent generates DeliveryContent (summary, body, attachments)\n// 3. Push delivery event for asynchronous routing via handlers\nfunc (e *Executor) RunDelivery(ctx *robottypes.Context, exec *robottypes.Execution, _ interface{}) error {\n\trobot := exec.GetRobot()\n\tif robot == nil {\n\t\treturn fmt.Errorf(\"robot not found in execution\")\n\t}\n\n\tlocale := getEffectiveLocale(robot, exec.Input)\n\te.updateUIFields(ctx, exec, \"\", getLocalizedMessage(locale, \"generating_delivery\"))\n\n\tagentID := \"__yao.delivery\"\n\tif robot.Config != nil && robot.Config.Resources != nil {\n\t\tagentID = robot.Config.Resources.GetPhaseAgent(robottypes.PhaseDelivery)\n\t}\n\n\tformatter := NewInputFormatter()\n\tuserContent := formatter.FormatDeliveryInput(exec, robot)\n\n\tif userContent == \"\" {\n\t\treturn fmt.Errorf(\"no content available for delivery generation\")\n\t}\n\n\tcaller := NewAgentCaller()\n\tcaller.Connector = robot.LanguageModel\n\tresult, err := caller.CallWithMessages(ctx, agentID, userContent)\n\tif err != nil {\n\t\treturn fmt.Errorf(\"delivery agent (%s) call failed: %w\", agentID, err)\n\t}\n\n\tdata, err := result.GetJSON()\n\tif err != nil {\n\t\tcontent := result.GetText()\n\t\tif content == \"\" {\n\t\t\treturn fmt.Errorf(\"delivery agent returned empty response\")\n\t\t}\n\t\texec.Delivery = &robottypes.DeliveryResult{\n\t\t\tRequestID: generateRequestID(exec.ID),\n\t\t\tContent: &robottypes.DeliveryContent{\n\t\t\t\tSummary: truncateSummary(content, 200),\n\t\t\t\tBody: content,\n\t\t\t},\n\t\t\tSuccess: true,\n\t\t}\n\t\treturn e.pushDeliveryEvent(ctx, exec, robot)\n\t}\n\n\tcontent := parseDeliveryContent(data)\n\tif content == nil {\n\t\treturn fmt.Errorf(\"delivery agent (%s) returned invalid content\", agentID)\n\t}\n\n\texec.Delivery = &robottypes.DeliveryResult{\n\t\tRequestID: generateRequestID(exec.ID),\n\t\tContent: content,\n\t\tSuccess: true,\n\t}\n\n\treturn e.pushDeliveryEvent(ctx, exec, robot)\n}\n\n// pushDeliveryEvent pushes a delivery event to the event bus.\n// Registered handlers (see events/handlers.go) route to email/webhook/process channels.\nfunc (e *Executor) pushDeliveryEvent(ctx *robottypes.Context, exec *robottypes.Execution, robot *robottypes.Robot) error {\n\tprefs := buildDeliveryPreferences(robot)\n\n\tchatID := exec.ChatID\n\tvar extra map[string]any\n\tif exec.Input != nil && exec.Input.Data != nil {\n\t\tif sourceChatID, ok := exec.Input.Data[\"chat_id\"].(string); ok && sourceChatID != \"\" {\n\t\t\tif channel, ok := exec.Input.Data[\"channel\"].(string); ok && channel != \"\" {\n\t\t\t\tchatID = channel + \":\" + sourceChatID\n\t\t\t}\n\t\t}\n\t\tif e, ok := exec.Input.Data[\"extra\"].(map[string]any); ok {\n\t\t\textra = e\n\t\t}\n\t}\n\n\t_, err := event.Push(ctx.Context, robotevents.Delivery, robotevents.DeliveryPayload{\n\t\tExecutionID: exec.ID,\n\t\tMemberID: exec.MemberID,\n\t\tTeamID: exec.TeamID,\n\t\tChatID: chatID,\n\t\tContent: exec.Delivery.Content,\n\t\tPreferences: prefs,\n\t\tExtra: extra,\n\t})\n\tif err != nil {\n\t\tkunlog.Error(\"delivery event push failed: execution=%s error=%v\", exec.ID, err)\n\t}\n\treturn nil\n}\n\n// parseDeliveryContent parses the Delivery Agent response into DeliveryContent\nfunc parseDeliveryContent(data map[string]interface{}) *robottypes.DeliveryContent {\n\tif data == nil {\n\t\treturn nil\n\t}\n\n\tcontentData, ok := data[\"content\"].(map[string]interface{})\n\tif !ok {\n\t\tcontentData = data\n\t}\n\n\tcontent := &robottypes.DeliveryContent{}\n\n\tif summary, ok := contentData[\"summary\"].(string); ok {\n\t\tcontent.Summary = summary\n\t}\n\tif body, ok := contentData[\"body\"].(string); ok {\n\t\tcontent.Body = body\n\t}\n\n\tif attachments, ok := contentData[\"attachments\"].([]interface{}); ok {\n\t\tfor _, att := range attachments {\n\t\t\tif attMap, ok := att.(map[string]interface{}); ok {\n\t\t\t\tattachment := parseDeliveryAttachment(attMap)\n\t\t\t\tif attachment != nil {\n\t\t\t\t\tcontent.Attachments = append(content.Attachments, *attachment)\n\t\t\t\t}\n\t\t\t}\n\t\t}\n\t}\n\n\tif content.Summary == \"\" && content.Body == \"\" {\n\t\treturn nil\n\t}\n\n\treturn content\n}\n\nfunc parseDeliveryAttachment(data map[string]interface{}) *robottypes.DeliveryAttachment {\n\tif data == nil {\n\t\treturn nil\n\t}\n\n\tatt := &robottypes.DeliveryAttachment{}\n\n\tif title, ok := data[\"title\"].(string); ok {\n\t\tatt.Title = title\n\t}\n\tif desc, ok := data[\"description\"].(string); ok {\n\t\tatt.Description = desc\n\t}\n\tif taskID, ok := data[\"task_id\"].(string); ok {\n\t\tatt.TaskID = taskID\n\t}\n\tif file, ok := data[\"file\"].(string); ok {\n\t\tatt.File = file\n\t}\n\n\tif att.Title == \"\" || att.File == \"\" {\n\t\treturn nil\n\t}\n\n\treturn att\n}\n\nfunc generateRequestID(execID string) string {\n\treturn fmt.Sprintf(\"dlv-%s-%d\", execID, time.Now().UnixNano()%1000000)\n}\n\nfunc getTaskDescription(task robottypes.Task) string {\n\tif len(task.Messages) == 0 {\n\t\treturn task.GoalRef\n\t}\n\n\tfor _, msg := range task.Messages {\n\t\tif content, ok := msg.Content.(string); ok && content != \"\" {\n\t\t\tif len(content) > 100 {\n\t\t\t\treturn content[:97] + \"...\"\n\t\t\t}\n\t\t\treturn content\n\t\t}\n\t}\n\n\tif task.GoalRef != \"\" {\n\t\treturn task.GoalRef\n\t}\n\n\treturn \"Task \" + task.ID\n}\n\nfunc truncateSummary(text string, maxLen int) string {\n\tif len(text) <= maxLen {\n\t\treturn text\n\t}\n\ttruncated := text[:maxLen]\n\tif idx := strings.LastIndex(truncated, \" \"); idx > maxLen/2 {\n\t\treturn truncated[:idx] + \"...\"\n\t}\n\treturn truncated + \"...\"\n}\n\nfunc buildDeliveryPreferences(robot *robottypes.Robot) *robottypes.DeliveryPreferences {\n\tif robot == nil {\n\t\treturn nil\n\t}\n\n\tprefs := &robottypes.DeliveryPreferences{}\n\n\tmanagerEmail := robot.ManagerEmail\n\tif managerEmail == \"\" && robot.ManagerID != \"\" {\n\t\tmanagerEmail = getManagerEmail(robot.ManagerID)\n\t\tif managerEmail != \"\" {\n\t\t\trobot.ManagerEmail = managerEmail\n\t\t}\n\t}\n\n\tvar emailTargets []robottypes.EmailTarget\n\n\tif managerEmail != \"\" {\n\t\temailTargets = append(emailTargets, robottypes.EmailTarget{\n\t\t\tTo: []string{managerEmail},\n\t\t})\n\t}\n\n\tif robot.Config != nil && robot.Config.Delivery != nil && robot.Config.Delivery.Email != nil {\n\t\tfor _, target := range robot.Config.Delivery.Email.Targets {\n\t\t\tif len(target.To) > 0 {\n\t\t\t\temailTargets = append(emailTargets, target)\n\t\t\t}\n\t\t}\n\t}\n\n\tif len(emailTargets) > 0 {\n\t\tprefs.Email = &robottypes.EmailPreference{\n\t\t\tEnabled: true,\n\t\t\tTargets: emailTargets,\n\t\t}\n\t}\n\n\tif robot.Config != nil && robot.Config.Delivery != nil && robot.Config.Delivery.Webhook != nil {\n\t\tif robot.Config.Delivery.Webhook.Enabled && len(robot.Config.Delivery.Webhook.Targets) > 0 {\n\t\t\tprefs.Webhook = robot.Config.Delivery.Webhook\n\t\t}\n\t}\n\n\tif robot.Config != nil && robot.Config.Delivery != nil && robot.Config.Delivery.Process != nil {\n\t\tif robot.Config.Delivery.Process.Enabled && len(robot.Config.Delivery.Process.Targets) > 0 {\n\t\t\tprefs.Process = robot.Config.Delivery.Process\n\t\t}\n\t}\n\n\treturn prefs\n}\n\nfunc getManagerEmail(managerID string) string {\n\tif managerID == \"\" {\n\t\treturn \"\"\n\t}\n\n\tm := model.Select(\"__yao.member\")\n\tif m == nil {\n\t\treturn \"\"\n\t}\n\n\trecords, err := m.Get(model.QueryParam{\n\t\tSelect: []interface{}{\"email\"},\n\t\tWheres: []model.QueryWhere{\n\t\t\t{Column: \"member_id\", Value: managerID},\n\t\t},\n\t\tLimit: 1,\n\t})\n\tif err != nil || len(records) == 0 {\n\t\treturn \"\"\n\t}\n\n\tif email, ok := records[0][\"email\"].(string); ok {\n\t\treturn email\n\t}\n\treturn \"\"\n}\n\n// FormatDeliveryInput formats the full execution context for the Delivery Agent\nfunc (f *InputFormatter) FormatDeliveryInput(exec *robottypes.Execution, robot *robottypes.Robot) string {\n\tif exec == nil {\n\t\treturn \"\"\n\t}\n\n\tvar sb strings.Builder\n\n\tif robot != nil && robot.Config != nil && robot.Config.Identity != nil {\n\t\tsb.WriteString(\"## Robot Identity\\n\\n\")\n\t\tsb.WriteString(fmt.Sprintf(\"- **Role**: %s\\n\", robot.Config.Identity.Role))\n\t\tif len(robot.Config.Identity.Duties) > 0 {\n\t\t\tsb.WriteString(\"- **Duties**: \")\n\t\t\tsb.WriteString(strings.Join(robot.Config.Identity.Duties, \", \"))\n\t\t\tsb.WriteString(\"\\n\")\n\t\t}\n\t\tsb.WriteString(\"\\n\")\n\t}\n\n\tsb.WriteString(\"## Execution Context\\n\\n\")\n\tsb.WriteString(fmt.Sprintf(\"- **Trigger**: %s\\n\", exec.TriggerType))\n\tsb.WriteString(fmt.Sprintf(\"- **Status**: %s\\n\", exec.Status))\n\tsb.WriteString(fmt.Sprintf(\"- **Start Time**: %s\\n\", exec.StartTime.Format(\"2006-01-02 15:04:05\")))\n\tif exec.EndTime != nil {\n\t\tduration := exec.EndTime.Sub(exec.StartTime)\n\t\tsb.WriteString(fmt.Sprintf(\"- **Duration**: %s\\n\", duration.String()))\n\t}\n\tsb.WriteString(\"\\n\")\n\n\tif exec.Inspiration != nil && exec.Inspiration.Content != \"\" {\n\t\tsb.WriteString(\"## Inspiration (P0)\\n\\n\")\n\t\tsb.WriteString(exec.Inspiration.Content)\n\t\tsb.WriteString(\"\\n\\n\")\n\t}\n\n\tif exec.Goals != nil && exec.Goals.Content != \"\" {\n\t\tsb.WriteString(\"## Goals (P1)\\n\\n\")\n\t\tsb.WriteString(exec.Goals.Content)\n\t\tsb.WriteString(\"\\n\\n\")\n\t}\n\n\tif len(exec.Tasks) > 0 {\n\t\tsb.WriteString(\"## Tasks (P2)\\n\\n\")\n\t\tfor i, task := range exec.Tasks {\n\t\t\ttaskDesc := getTaskDescription(task)\n\t\t\tsb.WriteString(fmt.Sprintf(\"%d. **%s** - %s\\n\", i+1, task.ID, taskDesc))\n\t\t\tsb.WriteString(fmt.Sprintf(\" - Executor: %s (%s)\\n\", task.ExecutorID, task.ExecutorType))\n\t\t\tsb.WriteString(fmt.Sprintf(\" - Status: %s\\n\", task.Status))\n\t\t\tif task.ExpectedOutput != \"\" {\n\t\t\t\tsb.WriteString(fmt.Sprintf(\" - Expected: %s\\n\", task.ExpectedOutput))\n\t\t\t}\n\t\t}\n\t\tsb.WriteString(\"\\n\")\n\t}\n\n\tif len(exec.Results) > 0 {\n\t\tsb.WriteString(\"## Results (P3)\\n\\n\")\n\n\t\tsuccessCount := 0\n\t\tfailCount := 0\n\n\t\tfor _, result := range exec.Results {\n\t\t\tif result.Success {\n\t\t\t\tsuccessCount++\n\t\t\t\tsb.WriteString(fmt.Sprintf(\"### ✓ Task: %s\\n\\n\", result.TaskID))\n\t\t\t} else {\n\t\t\t\tfailCount++\n\t\t\t\tsb.WriteString(fmt.Sprintf(\"### ✗ Task: %s\\n\\n\", result.TaskID))\n\t\t\t}\n\n\t\t\tsb.WriteString(fmt.Sprintf(\"- **Duration**: %dms\\n\", result.Duration))\n\n\t\t\tif result.Validation != nil {\n\t\t\t\tif result.Validation.Passed {\n\t\t\t\t\tsb.WriteString(fmt.Sprintf(\"- **Validation**: ✓ Passed (score: %.2f)\\n\", result.Validation.Score))\n\t\t\t\t} else {\n\t\t\t\t\tsb.WriteString(\"- **Validation**: ✗ Failed\\n\")\n\t\t\t\t\tif len(result.Validation.Issues) > 0 {\n\t\t\t\t\t\tfor _, issue := range result.Validation.Issues {\n\t\t\t\t\t\t\tsb.WriteString(fmt.Sprintf(\" - %s\\n\", issue))\n\t\t\t\t\t\t}\n\t\t\t\t\t}\n\t\t\t\t}\n\t\t\t}\n\n\t\t\tif result.Output != nil {\n\t\t\t\tsb.WriteString(\"\\n**Output**:\\n\")\n\t\t\t\tif output, err := json.MarshalIndent(result.Output, \"\", \" \"); err == nil {\n\t\t\t\t\tsb.WriteString(\"```json\\n\")\n\t\t\t\t\tsb.WriteString(string(output))\n\t\t\t\t\tsb.WriteString(\"\\n```\\n\")\n\t\t\t\t} else {\n\t\t\t\t\tsb.WriteString(fmt.Sprintf(\"%v\\n\", result.Output))\n\t\t\t\t}\n\t\t\t}\n\n\t\t\tif result.Error != \"\" {\n\t\t\t\tsb.WriteString(fmt.Sprintf(\"\\n**Error**: %s\\n\", result.Error))\n\t\t\t}\n\n\t\t\tsb.WriteString(\"\\n\")\n\t\t}\n\n\t\tsb.WriteString(fmt.Sprintf(\"### Summary\\n\\n- **Total Tasks**: %d\\n- **Succeeded**: %d\\n- **Failed**: %d\\n\\n\",\n\t\t\tlen(exec.Results), successCount, failCount))\n\t}\n\n\treturn sb.String()\n}\n" }, { "path": "agent/robot/executor/standard/delivery_test.go", "content": "package standard_test\n\nimport (\n\t\"context\"\n\t\"strings\"\n\t\"testing\"\n\t\"time\"\n\n\t\"github.com/stretchr/testify/assert\"\n\t\"github.com/stretchr/testify/require\"\n\t\"github.com/yaoapp/yao/agent/robot/executor/standard\"\n\t\"github.com/yaoapp/yao/agent/robot/types\"\n\t\"github.com/yaoapp/yao/agent/testutils\"\n)\n\n// ============================================================================\n// P4 Delivery Phase Tests\n// ============================================================================\n\nfunc TestRunDeliveryBasic(t *testing.T) {\n\tif testing.Short() {\n\t\tt.Skip(\"Skipping integration test\")\n\t}\n\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\tctx := types.NewContext(context.Background(), testAuth())\n\n\tt.Run(\"generates delivery content from execution results\", func(t *testing.T) {\n\t\trobot := createDeliveryTestRobot(t, \"robot.delivery\")\n\t\texec := createDeliveryTestExecution(robot)\n\n\t\texec.Inspiration = &types.InspirationReport{\n\t\t\tContent: \"Morning analysis suggests focus on Q4 review.\",\n\t\t}\n\t\texec.Goals = &types.Goals{\n\t\t\tContent: \"## Goals\\n1. Review Q4 data\\n2. Generate summary report\",\n\t\t}\n\t\texec.Tasks = []types.Task{\n\t\t\t{ID: \"task-001\", ExecutorID: \"experts.data-analyst\", ExecutorType: types.ExecutorAssistant, Status: types.TaskCompleted},\n\t\t\t{ID: \"task-002\", ExecutorID: \"experts.summarizer\", ExecutorType: types.ExecutorAssistant, Status: types.TaskCompleted},\n\t\t}\n\t\texec.Results = []types.TaskResult{\n\t\t\t{TaskID: \"task-001\", Success: true, Duration: 1500, Output: map[string]interface{}{\"total_sales\": 1500000}},\n\t\t\t{TaskID: \"task-002\", Success: true, Duration: 800, Output: \"Q4 sales exceeded expectations by 15%.\"},\n\t\t}\n\n\t\te := standard.New()\n\t\terr := e.RunDelivery(ctx, exec, nil)\n\n\t\trequire.NoError(t, err)\n\t\trequire.NotNil(t, exec.Delivery)\n\t\trequire.NotNil(t, exec.Delivery.Content)\n\t\tassert.NotEmpty(t, exec.Delivery.Content.Summary)\n\t\tassert.NotEmpty(t, exec.Delivery.Content.Body)\n\t\tassert.True(t, exec.Delivery.Success)\n\t})\n\n\tt.Run(\"handles partial failure in results\", func(t *testing.T) {\n\t\trobot := createDeliveryTestRobot(t, \"robot.delivery\")\n\t\texec := createDeliveryTestExecution(robot)\n\n\t\texec.Goals = &types.Goals{\n\t\t\tContent: \"## Goals\\n1. Analyze data\\n2. Generate report\",\n\t\t}\n\t\texec.Tasks = []types.Task{\n\t\t\t{ID: \"task-001\", ExecutorID: \"experts.data-analyst\", ExecutorType: types.ExecutorAssistant, Status: types.TaskCompleted},\n\t\t\t{ID: \"task-002\", ExecutorID: \"experts.summarizer\", ExecutorType: types.ExecutorAssistant, Status: types.TaskFailed},\n\t\t}\n\t\texec.Results = []types.TaskResult{\n\t\t\t{TaskID: \"task-001\", Success: true, Duration: 1500, Output: map[string]interface{}{\"data\": \"analyzed\"}},\n\t\t\t{TaskID: \"task-002\", Success: false, Duration: 500, Error: \"Summarization failed: timeout\"},\n\t\t}\n\n\t\te := standard.New()\n\t\terr := e.RunDelivery(ctx, exec, nil)\n\n\t\trequire.NoError(t, err)\n\t\trequire.NotNil(t, exec.Delivery)\n\t\trequire.NotNil(t, exec.Delivery.Content)\n\n\t\tbody := strings.ToLower(exec.Delivery.Content.Body)\n\t\thasFailureInfo := strings.Contains(body, \"fail\") ||\n\t\t\tstrings.Contains(body, \"error\") ||\n\t\t\tstrings.Contains(body, \"partial\") ||\n\t\t\tstrings.Contains(body, \"✗\")\n\n\t\tassert.True(t, hasFailureInfo || exec.Delivery.Content.Summary != \"\", \"should mention failure or have valid summary\")\n\t})\n}\n\nfunc TestRunDeliveryErrorHandling(t *testing.T) {\n\tif testing.Short() {\n\t\tt.Skip(\"Skipping integration test\")\n\t}\n\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\tctx := types.NewContext(context.Background(), testAuth())\n\n\tt.Run(\"returns error when robot is nil\", func(t *testing.T) {\n\t\texec := &types.Execution{\n\t\t\tID: \"test-exec-1\",\n\t\t\tTriggerType: types.TriggerClock,\n\t\t}\n\n\t\te := standard.New()\n\t\terr := e.RunDelivery(ctx, exec, nil)\n\n\t\tassert.Error(t, err)\n\t\tassert.Contains(t, err.Error(), \"robot not found\")\n\t})\n\n\tt.Run(\"returns error when agent not found\", func(t *testing.T) {\n\t\trobot := &types.Robot{\n\t\t\tMemberID: \"test-robot-1\",\n\t\t\tTeamID: \"test-team-1\",\n\t\t\tConfig: &types.Config{\n\t\t\t\tIdentity: &types.Identity{Role: \"Test\"},\n\t\t\t\tResources: &types.Resources{\n\t\t\t\t\tPhases: map[types.Phase]string{\n\t\t\t\t\t\ttypes.PhaseDelivery: \"non.existent.agent\",\n\t\t\t\t\t},\n\t\t\t\t},\n\t\t\t},\n\t\t}\n\t\texec := createDeliveryTestExecution(robot)\n\t\texec.Results = []types.TaskResult{\n\t\t\t{TaskID: \"task-001\", Success: true, Duration: 100},\n\t\t}\n\n\t\te := standard.New()\n\t\terr := e.RunDelivery(ctx, exec, nil)\n\n\t\tassert.Error(t, err)\n\t\tassert.Contains(t, err.Error(), \"call failed\")\n\t})\n}\n\n// ============================================================================\n// Email Channel Config Tests\n// ============================================================================\n\nfunc TestDefaultEmailChannel(t *testing.T) {\n\tt.Run(\"returns default email channel\", func(t *testing.T) {\n\t\tassert.Equal(t, \"default\", types.DefaultEmailChannel())\n\t})\n\n\tt.Run(\"can set custom email channel\", func(t *testing.T) {\n\t\toriginal := types.DefaultEmailChannel()\n\t\tdefer types.SetDefaultEmailChannel(original)\n\n\t\ttypes.SetDefaultEmailChannel(\"custom-email\")\n\t\tassert.Equal(t, \"custom-email\", types.DefaultEmailChannel())\n\t})\n\n\tt.Run(\"ignores empty channel\", func(t *testing.T) {\n\t\toriginal := types.DefaultEmailChannel()\n\t\tdefer types.SetDefaultEmailChannel(original)\n\n\t\ttypes.SetDefaultEmailChannel(\"\")\n\t\tassert.Equal(t, original, types.DefaultEmailChannel())\n\t})\n}\n\nfunc TestRobotEmailInDelivery(t *testing.T) {\n\tt.Run(\"robot email field is loaded from map\", func(t *testing.T) {\n\t\tdata := map[string]interface{}{\n\t\t\t\"member_id\": \"robot-001\",\n\t\t\t\"team_id\": \"team-001\",\n\t\t\t\"robot_email\": \"robot@example.com\",\n\t\t}\n\n\t\trobot, err := types.NewRobotFromMap(data)\n\t\trequire.NoError(t, err)\n\t\tassert.Equal(t, \"robot@example.com\", robot.RobotEmail)\n\t})\n\n\tt.Run(\"robot email can be empty\", func(t *testing.T) {\n\t\tdata := map[string]interface{}{\n\t\t\t\"member_id\": \"robot-001\",\n\t\t\t\"team_id\": \"team-001\",\n\t\t}\n\n\t\trobot, err := types.NewRobotFromMap(data)\n\t\trequire.NoError(t, err)\n\t\tassert.Empty(t, robot.RobotEmail)\n\t})\n}\n\n// ============================================================================\n// FormatDeliveryInput Tests\n// ============================================================================\n\nfunc TestFormatDeliveryInput(t *testing.T) {\n\tformatter := standard.NewInputFormatter()\n\n\tt.Run(\"formats complete execution context\", func(t *testing.T) {\n\t\trobot := &types.Robot{\n\t\t\tMemberID: \"test-robot\",\n\t\t\tConfig: &types.Config{\n\t\t\t\tIdentity: &types.Identity{\n\t\t\t\t\tRole: \"Sales Analyst\",\n\t\t\t\t\tDuties: []string{\"Analyze data\", \"Generate reports\"},\n\t\t\t\t},\n\t\t\t},\n\t\t}\n\n\t\tstartTime := time.Now().Add(-5 * time.Minute)\n\t\tendTime := time.Now()\n\t\texec := &types.Execution{\n\t\t\tID: \"exec-123\",\n\t\t\tTriggerType: types.TriggerClock,\n\t\t\tStatus: types.ExecCompleted,\n\t\t\tStartTime: startTime,\n\t\t\tEndTime: &endTime,\n\t\t\tInspiration: &types.InspirationReport{\n\t\t\t\tContent: \"Morning analysis suggests focus on Q4.\",\n\t\t\t},\n\t\t\tGoals: &types.Goals{\n\t\t\t\tContent: \"## Goals\\n1. Review Q4 data\",\n\t\t\t},\n\t\t\tTasks: []types.Task{\n\t\t\t\t{ID: \"task-001\", ExecutorID: \"data-analyst\", ExecutorType: types.ExecutorAssistant, Status: types.TaskCompleted, ExpectedOutput: \"JSON with sales data\"},\n\t\t\t},\n\t\t\tResults: []types.TaskResult{\n\t\t\t\t{TaskID: \"task-001\", Success: true, Duration: 1500, Output: map[string]interface{}{\"sales\": 1000000}},\n\t\t\t},\n\t\t}\n\n\t\tresult := formatter.FormatDeliveryInput(exec, robot)\n\n\t\tassert.Contains(t, result, \"## Robot Identity\")\n\t\tassert.Contains(t, result, \"Sales Analyst\")\n\t\tassert.Contains(t, result, \"## Execution Context\")\n\t\tassert.Contains(t, result, \"clock\")\n\t\tassert.Contains(t, result, \"## Inspiration (P0)\")\n\t\tassert.Contains(t, result, \"Morning analysis\")\n\t\tassert.Contains(t, result, \"## Goals (P1)\")\n\t\tassert.Contains(t, result, \"Review Q4 data\")\n\t\tassert.Contains(t, result, \"## Tasks (P2)\")\n\t\tassert.Contains(t, result, \"task-001\")\n\t\tassert.Contains(t, result, \"## Results (P3)\")\n\t\tassert.Contains(t, result, \"✓ Task: task-001\")\n\t})\n\n\tt.Run(\"handles empty execution\", func(t *testing.T) {\n\t\texec := &types.Execution{\n\t\t\tID: \"exec-empty\",\n\t\t\tTriggerType: types.TriggerHuman,\n\t\t\tStatus: types.ExecPending,\n\t\t\tStartTime: time.Now(),\n\t\t}\n\n\t\tresult := formatter.FormatDeliveryInput(exec, nil)\n\n\t\tassert.Contains(t, result, \"## Execution Context\")\n\t\tassert.Contains(t, result, \"human\")\n\t})\n}\n\n// ============================================================================\n// Helper Functions\n// ============================================================================\n\nfunc createDeliveryTestRobot(t *testing.T, agentID string) *types.Robot {\n\tt.Helper()\n\treturn &types.Robot{\n\t\tMemberID: \"test-robot-1\",\n\t\tTeamID: \"test-team-1\",\n\t\tDisplayName: \"Test Robot\",\n\t\tConfig: &types.Config{\n\t\t\tIdentity: &types.Identity{\n\t\t\t\tRole: \"Test Assistant\",\n\t\t\t\tDuties: []string{\"Testing\", \"Validation\"},\n\t\t\t},\n\t\t\tResources: &types.Resources{\n\t\t\t\tPhases: map[types.Phase]string{\n\t\t\t\t\ttypes.PhaseDelivery: agentID,\n\t\t\t\t},\n\t\t\t},\n\t\t},\n\t}\n}\n\nfunc createDeliveryTestExecution(robot *types.Robot) *types.Execution {\n\texec := &types.Execution{\n\t\tID: \"test-exec-delivery-1\",\n\t\tMemberID: robot.MemberID,\n\t\tTeamID: robot.TeamID,\n\t\tTriggerType: types.TriggerClock,\n\t\tStartTime: time.Now(),\n\t\tStatus: types.ExecRunning,\n\t\tPhase: types.PhaseDelivery,\n\t}\n\texec.SetRobot(robot)\n\treturn exec\n}\n" }, { "path": "agent/robot/executor/standard/executor.go", "content": "package standard\n\nimport (\n\t\"fmt\"\n\t\"strings\"\n\t\"sync/atomic\"\n\t\"time\"\n\n\tkunlog \"github.com/yaoapp/kun/log\"\n\tagentcontext \"github.com/yaoapp/yao/agent/context\"\n\trobotevents \"github.com/yaoapp/yao/agent/robot/events\"\n\t\"github.com/yaoapp/yao/agent/robot/executor/types\"\n\t\"github.com/yaoapp/yao/agent/robot/store\"\n\trobottypes \"github.com/yaoapp/yao/agent/robot/types\"\n\t\"github.com/yaoapp/yao/agent/robot/utils\"\n\t\"github.com/yaoapp/yao/event\"\n)\n\n// Executor implements the standard executor with real Agent calls\n// This is the production executor that:\n// - Persists execution history to database\n// - Calls real Agents via Assistant.Stream()\n// - Logs phase transitions and errors using kun/log\ntype Executor struct {\n\tconfig types.Config\n\tstore *store.ExecutionStore\n\trobotStore *store.RobotStore\n\texecCount atomic.Int32\n\tcurrentCount atomic.Int32\n\tonStart func()\n\tonEnd func()\n}\n\n// New creates a new standard executor\nfunc New() *Executor {\n\treturn &Executor{\n\t\tstore: store.NewExecutionStore(),\n\t\trobotStore: store.NewRobotStore(),\n\t}\n}\n\n// NewWithConfig creates a new standard executor with configuration\nfunc NewWithConfig(config types.Config) *Executor {\n\treturn &Executor{\n\t\tconfig: config,\n\t\tstore: store.NewExecutionStore(),\n\t\trobotStore: store.NewRobotStore(),\n\t}\n}\n\n// Execute runs a robot through all applicable phases with real Agent calls (auto-generates ID)\nfunc (e *Executor) Execute(ctx *robottypes.Context, robot *robottypes.Robot, trigger robottypes.TriggerType, data interface{}) (*robottypes.Execution, error) {\n\treturn e.ExecuteWithControl(ctx, robot, trigger, data, \"\", nil)\n}\n\n// ExecuteWithID runs a robot through all applicable phases with a pre-generated execution ID (no control)\nfunc (e *Executor) ExecuteWithID(ctx *robottypes.Context, robot *robottypes.Robot, trigger robottypes.TriggerType, data interface{}, execID string) (*robottypes.Execution, error) {\n\treturn e.ExecuteWithControl(ctx, robot, trigger, data, execID, nil)\n}\n\n// ExecuteWithControl runs a robot through all applicable phases with execution control\n// control: optional, allows pause/resume functionality during execution\nfunc (e *Executor) ExecuteWithControl(ctx *robottypes.Context, robot *robottypes.Robot, trigger robottypes.TriggerType, data interface{}, execID string, control robottypes.ExecutionControl) (*robottypes.Execution, error) {\n\tif robot == nil {\n\t\treturn nil, fmt.Errorf(\"robot cannot be nil\")\n\t}\n\n\t// Determine starting phase based on trigger type\n\tstartPhaseIndex := 0\n\tif trigger == robottypes.TriggerHuman || trigger == robottypes.TriggerEvent {\n\t\tstartPhaseIndex = 1 // Skip P0 (Inspiration)\n\t}\n\n\t// Use provided execID or generate new one\n\tif execID == \"\" {\n\t\texecID = utils.NewID()\n\t}\n\n\t// Create execution (Job system removed, using ExecutionStore only)\n\tinput := types.BuildTriggerInput(trigger, data)\n\texec := &robottypes.Execution{\n\t\tID: execID,\n\t\tMemberID: robot.MemberID,\n\t\tTeamID: robot.TeamID,\n\t\tTriggerType: trigger,\n\t\tStartTime: time.Now(),\n\t\tStatus: robottypes.ExecPending,\n\t\tPhase: robottypes.AllPhases[startPhaseIndex],\n\t\tInput: input,\n\t\tChatID: fmt.Sprintf(\"robot_%s_%s\", robot.MemberID, execID),\n\t}\n\n\t// Load pre-existing Goals/Tasks from store when resuming a confirmed execution.\n\t// RunGoals and RunTasks have skip logic when these are already populated.\n\tif execID != \"\" && !e.config.SkipPersistence && e.store != nil {\n\t\tif existing, err := e.store.Get(ctx.Context, execID); err == nil && existing != nil {\n\t\t\texec.Goals = existing.Goals\n\t\t\texec.Tasks = existing.Tasks\n\t\t\tif existing.Input != nil {\n\t\t\t\texec.Input = existing.Input\n\t\t\t}\n\t\t}\n\t}\n\n\t// If goals are pre-confirmed (passed via Input.Data[\"goals\"]), inject them directly.\n\t// RunGoals will skip LLM call when exec.Goals is already populated (§18.2).\n\tif exec.Goals == nil && input != nil && input.Data != nil {\n\t\tif goalsStr, ok := input.Data[\"goals\"].(string); ok && goalsStr != \"\" {\n\t\t\texec.Goals = &robottypes.Goals{Content: goalsStr}\n\t\t}\n\t}\n\n\t// Initialize UI display fields (with i18n support)\n\texec.Name, exec.CurrentTaskName = e.initUIFields(trigger, input, robot)\n\n\t// Set robot reference for phase methods\n\texec.SetRobot(robot)\n\n\t// Persist execution record to database\n\t// Robot is identified by member_id (globally unique in __yao.member table)\n\tif !e.config.SkipPersistence && e.store != nil {\n\t\trecord := store.FromExecution(exec)\n\t\tif err := e.store.Save(ctx.Context, record); err != nil {\n\t\t\t// Log warning but don't fail execution\n\t\t\tkunlog.With(kunlog.F{\n\t\t\t\t\"execution_id\": exec.ID,\n\t\t\t\t\"member_id\": exec.MemberID,\n\t\t\t\t\"error\": err,\n\t\t\t}).Warn(\"Failed to persist execution record: %v\", err)\n\t\t}\n\n\t\t// If goals were pre-injected, persist them and update the execution title\n\t\tif exec.Goals != nil && exec.Goals.Content != \"\" {\n\t\t\tif err := e.store.UpdatePhase(ctx.Context, exec.ID, robottypes.PhaseGoals, exec.Goals); err != nil {\n\t\t\t\tkunlog.With(kunlog.F{\n\t\t\t\t\t\"execution_id\": exec.ID,\n\t\t\t\t\t\"member_id\": exec.MemberID,\n\t\t\t\t\t\"error\": err,\n\t\t\t\t}).Warn(\"Failed to persist pre-confirmed goals: %v\", err)\n\t\t\t}\n\t\t\tif goalName := extractGoalName(exec.Goals); goalName != \"\" {\n\t\t\t\te.updateUIFields(ctx, exec, goalName, \"\")\n\t\t\t}\n\t\t}\n\n\t}\n\n\t// Acquire execution slot\n\tif !robot.TryAcquireSlot(exec) {\n\t\tkunlog.With(kunlog.F{\n\t\t\t\"execution_id\": exec.ID,\n\t\t\t\"member_id\": exec.MemberID,\n\t\t}).Warn(\"Execution quota exceeded\")\n\t\treturn nil, robottypes.ErrQuotaExceeded\n\t}\n\t// Defer: remove execution from robot's tracking (unless suspended) and update robot status\n\tdefer func() {\n\t\t// Suspended executions stay in tracking — they are still \"alive\"\n\t\tif exec.Status == robottypes.ExecWaiting {\n\t\t\treturn\n\t\t}\n\t\trobot.RemoveExecution(exec.ID)\n\t\t// Update robot status to idle if no more running executions\n\t\tif robot.RunningCount() == 0 && !e.config.SkipPersistence && e.robotStore != nil {\n\t\t\tif err := e.robotStore.UpdateStatus(ctx.Context, robot.MemberID, robottypes.RobotIdle); err != nil {\n\t\t\t\tkunlog.With(kunlog.F{\n\t\t\t\t\t\"member_id\": robot.MemberID,\n\t\t\t\t\t\"error\": err,\n\t\t\t\t}).Warn(\"Failed to update robot status to idle: %v\", err)\n\t\t\t}\n\t\t}\n\t}()\n\n\t// Track execution count\n\te.execCount.Add(1)\n\te.currentCount.Add(1)\n\tdefer e.currentCount.Add(-1)\n\n\t// Callbacks\n\tif e.onStart != nil {\n\t\te.onStart()\n\t}\n\tif e.onEnd != nil {\n\t\tdefer e.onEnd()\n\t}\n\n\t// Update status to running\n\texec.Status = robottypes.ExecRunning\n\tkunlog.With(kunlog.F{\n\t\t\"execution_id\": exec.ID,\n\t\t\"member_id\": exec.MemberID,\n\t\t\"trigger_type\": string(exec.TriggerType),\n\t}).Info(\"Execution started\")\n\n\t// Persist running status\n\tif !e.config.SkipPersistence && e.store != nil {\n\t\tif err := e.store.UpdateStatus(ctx.Context, exec.ID, robottypes.ExecRunning, \"\"); err != nil {\n\t\t\tkunlog.With(kunlog.F{\n\t\t\t\t\"execution_id\": exec.ID,\n\t\t\t\t\"error\": err,\n\t\t\t}).Warn(\"Failed to persist running status: %v\", err)\n\t\t}\n\t}\n\n\t// Update robot status to working (when execution starts)\n\tif !e.config.SkipPersistence && e.robotStore != nil {\n\t\tif err := e.robotStore.UpdateStatus(ctx.Context, robot.MemberID, robottypes.RobotWorking); err != nil {\n\t\t\tkunlog.With(kunlog.F{\n\t\t\t\t\"member_id\": robot.MemberID,\n\t\t\t\t\"error\": err,\n\t\t\t}).Warn(\"Failed to update robot status to working: %v\", err)\n\t\t}\n\t}\n\n\t// Check for simulated failure (for testing)\n\tif dataStr, ok := data.(string); ok && dataStr == \"simulate_failure\" {\n\t\texec.Status = robottypes.ExecFailed\n\t\texec.Error = \"simulated failure\"\n\t\tkunlog.With(kunlog.F{\n\t\t\t\"execution_id\": exec.ID,\n\t\t\t\"member_id\": exec.MemberID,\n\t\t}).Warn(\"Simulated failure triggered\")\n\t\t// Persist failed status\n\t\tif !e.config.SkipPersistence && e.store != nil {\n\t\t\t_ = e.store.UpdateStatus(ctx.Context, exec.ID, robottypes.ExecFailed, \"simulated failure\")\n\t\t}\n\t\treturn exec, nil\n\t}\n\n\t// Determine locale for UI messages\n\tlocale := getEffectiveLocale(robot, exec.Input)\n\n\t// Execute phases (PhaseHost is not part of the normal pipeline — it is only for Interact)\n\tphases := robottypes.AllPhases[startPhaseIndex:]\n\tfor _, phase := range phases {\n\t\tif phase == robottypes.PhaseHost {\n\t\t\tcontinue\n\t\t}\n\t\tif err := e.runPhase(ctx, exec, phase, data, control); err != nil {\n\t\t\t// Check if execution was suspended (needs human input)\n\t\t\tif err == robottypes.ErrExecutionSuspended {\n\t\t\t\tkunlog.With(kunlog.F{\n\t\t\t\t\t\"execution_id\": exec.ID,\n\t\t\t\t\t\"member_id\": exec.MemberID,\n\t\t\t\t\t\"phase\": string(phase),\n\t\t\t\t}).Info(\"Execution suspended during phase %s\", phase)\n\t\t\t\treturn exec, robottypes.ErrExecutionSuspended\n\t\t\t}\n\n\t\t\t// Check if execution was cancelled\n\t\t\tif err == robottypes.ErrExecutionCancelled {\n\t\t\t\texec.Status = robottypes.ExecCancelled\n\t\t\t\texec.Error = \"execution cancelled by user\"\n\t\t\t\tnow := time.Now()\n\t\t\t\texec.EndTime = &now\n\n\t\t\t\t// Update UI field for cancellation with i18n\n\t\t\t\te.updateUIFields(ctx, exec, \"\", getLocalizedMessage(locale, \"cancelled\"))\n\n\t\t\t\tkunlog.With(kunlog.F{\n\t\t\t\t\t\"execution_id\": exec.ID,\n\t\t\t\t\t\"member_id\": exec.MemberID,\n\t\t\t\t\t\"phase\": string(phase),\n\t\t\t\t}).Info(\"Execution cancelled by user\")\n\n\t\t\t\t// Persist cancelled status\n\t\t\t\tif !e.config.SkipPersistence && e.store != nil {\n\t\t\t\t\t_ = e.store.UpdateStatus(ctx.Context, exec.ID, robottypes.ExecCancelled, \"execution cancelled by user\")\n\t\t\t\t}\n\t\t\t\treturn exec, nil\n\t\t\t}\n\n\t\t\t// Normal failure case\n\t\t\texec.Status = robottypes.ExecFailed\n\t\t\texec.Error = err.Error()\n\n\t\t\t// Update UI field for failure with i18n\n\t\t\tfailedPrefix := getLocalizedMessage(locale, \"failed_prefix\")\n\t\t\tphaseName := getLocalizedMessage(locale, \"phase_\"+string(phase))\n\t\t\tfailureMsg := failedPrefix + phaseName\n\t\t\te.updateUIFields(ctx, exec, \"\", failureMsg)\n\n\t\t\tkunlog.With(kunlog.F{\n\t\t\t\t\"execution_id\": exec.ID,\n\t\t\t\t\"member_id\": exec.MemberID,\n\t\t\t\t\"phase\": string(phase),\n\t\t\t\t\"error\": err.Error(),\n\t\t\t}).Error(\"Phase execution failed: %v\", err)\n\t\t\t// Persist failed status\n\t\t\tif !e.config.SkipPersistence && e.store != nil {\n\t\t\t\t_ = e.store.UpdateStatus(ctx.Context, exec.ID, robottypes.ExecFailed, err.Error())\n\t\t\t}\n\t\t\treturn exec, nil\n\t\t}\n\t}\n\n\t// Mark completed\n\texec.Status = robottypes.ExecCompleted\n\tnow := time.Now()\n\texec.EndTime = &now\n\n\t// Update UI field for completion with i18n\n\te.updateUIFields(ctx, exec, \"\", getLocalizedMessage(locale, \"completed\"))\n\n\tduration := now.Sub(exec.StartTime)\n\tkunlog.With(kunlog.F{\n\t\t\"execution_id\": exec.ID,\n\t\t\"member_id\": exec.MemberID,\n\t\t\"duration_ms\": duration.Milliseconds(),\n\t}).Info(\"Execution completed successfully\")\n\n\t// Persist completed status\n\tif !e.config.SkipPersistence && e.store != nil {\n\t\tif err := e.store.UpdateStatus(ctx.Context, exec.ID, robottypes.ExecCompleted, \"\"); err != nil {\n\t\t\tkunlog.With(kunlog.F{\n\t\t\t\t\"execution_id\": exec.ID,\n\t\t\t\t\"error\": err,\n\t\t\t}).Warn(\"Failed to persist completed status: %v\", err)\n\t\t}\n\t}\n\n\tevent.Push(ctx.Context, robotevents.ExecCompleted, robotevents.ExecPayload{\n\t\tExecutionID: exec.ID,\n\t\tMemberID: exec.MemberID,\n\t\tTeamID: exec.TeamID,\n\t\tStatus: string(robottypes.ExecCompleted),\n\t\tChatID: exec.ChatID,\n\t})\n\n\treturn exec, nil\n}\n\n// runPhase executes a single phase\nfunc (e *Executor) runPhase(ctx *robottypes.Context, exec *robottypes.Execution, phase robottypes.Phase, data interface{}, control robottypes.ExecutionControl) error {\n\t// Check if context is cancelled before starting this phase\n\tselect {\n\tcase <-ctx.Context.Done():\n\t\treturn robottypes.ErrExecutionCancelled\n\tdefault:\n\t}\n\n\t// Wait if execution is paused (blocks until resumed or cancelled)\n\tif control != nil {\n\t\tif err := control.WaitIfPaused(); err != nil {\n\t\t\treturn err // Returns ErrExecutionCancelled if cancelled while paused\n\t\t}\n\t}\n\n\texec.Phase = phase\n\n\tkunlog.With(kunlog.F{\n\t\t\"execution_id\": exec.ID,\n\t\t\"member_id\": exec.MemberID,\n\t\t\"phase\": string(phase),\n\t}).Info(\"Phase started: %s\", phase)\n\n\t// Persist phase change immediately (so frontend sees current phase)\n\tif !e.config.SkipPersistence && e.store != nil {\n\t\tif err := e.store.UpdatePhase(ctx.Context, exec.ID, phase, nil); err != nil {\n\t\t\tkunlog.With(kunlog.F{\n\t\t\t\t\"execution_id\": exec.ID,\n\t\t\t\t\"phase\": string(phase),\n\t\t\t\t\"error\": err,\n\t\t\t}).Warn(\"Failed to persist phase start: %v\", err)\n\t\t}\n\t}\n\n\tif e.config.OnPhaseStart != nil {\n\t\te.config.OnPhaseStart(phase)\n\t}\n\n\tphaseStart := time.Now()\n\n\t// Execute phase-specific logic\n\tvar err error\n\tswitch phase {\n\tcase robottypes.PhaseInspiration:\n\t\terr = e.RunInspiration(ctx, exec, data)\n\tcase robottypes.PhaseGoals:\n\t\terr = e.RunGoals(ctx, exec, data)\n\tcase robottypes.PhaseTasks:\n\t\terr = e.RunTasks(ctx, exec, data)\n\tcase robottypes.PhaseRun:\n\t\terr = e.RunExecution(ctx, exec, data)\n\tcase robottypes.PhaseDelivery:\n\t\terr = e.RunDelivery(ctx, exec, data)\n\tcase robottypes.PhaseLearning:\n\t\terr = e.RunLearning(ctx, exec, data)\n\t}\n\n\tif err != nil {\n\t\tif err == robottypes.ErrExecutionSuspended {\n\t\t\tkunlog.With(kunlog.F{\n\t\t\t\t\"execution_id\": exec.ID,\n\t\t\t\t\"member_id\": exec.MemberID,\n\t\t\t\t\"phase\": string(phase),\n\t\t\t}).Info(\"Phase suspended: %s (waiting for human input)\", phase)\n\t\t\treturn err\n\t\t}\n\t\tkunlog.With(kunlog.F{\n\t\t\t\"execution_id\": exec.ID,\n\t\t\t\"member_id\": exec.MemberID,\n\t\t\t\"phase\": string(phase),\n\t\t\t\"error\": err.Error(),\n\t\t}).Error(\"Phase failed: %s - %v\", phase, err)\n\t\treturn err\n\t}\n\n\t// Persist phase output to database\n\tif !e.config.SkipPersistence && e.store != nil {\n\t\tphaseData := e.getPhaseData(exec, phase)\n\t\tif phaseData != nil {\n\t\t\tif err := e.store.UpdatePhase(ctx.Context, exec.ID, phase, phaseData); err != nil {\n\t\t\t\t// Log warning but don't fail execution\n\t\t\t\tkunlog.With(kunlog.F{\n\t\t\t\t\t\"execution_id\": exec.ID,\n\t\t\t\t\t\"phase\": string(phase),\n\t\t\t\t\t\"error\": err,\n\t\t\t\t}).Warn(\"Failed to persist phase %s data: %v\", phase, err)\n\t\t\t}\n\t\t}\n\t}\n\n\tif e.config.OnPhaseEnd != nil {\n\t\te.config.OnPhaseEnd(phase)\n\t}\n\n\tphaseDuration := time.Since(phaseStart).Milliseconds()\n\tkunlog.With(kunlog.F{\n\t\t\"execution_id\": exec.ID,\n\t\t\"member_id\": exec.MemberID,\n\t\t\"phase\": string(phase),\n\t\t\"duration_ms\": phaseDuration,\n\t}).Info(\"Phase completed: %s (took %dms)\", phase, phaseDuration)\n\n\treturn nil\n}\n\n// getPhaseData extracts the output data for a specific phase from execution\nfunc (e *Executor) getPhaseData(exec *robottypes.Execution, phase robottypes.Phase) interface{} {\n\tswitch phase {\n\tcase robottypes.PhaseInspiration:\n\t\treturn exec.Inspiration\n\tcase robottypes.PhaseGoals:\n\t\treturn exec.Goals\n\tcase robottypes.PhaseTasks:\n\t\treturn exec.Tasks\n\tcase robottypes.PhaseRun:\n\t\treturn exec.Results\n\tcase robottypes.PhaseDelivery:\n\t\treturn exec.Delivery\n\tcase robottypes.PhaseLearning:\n\t\treturn exec.Learning\n\tdefault:\n\t\treturn nil\n\t}\n}\n\n// ExecCount returns total execution count\nfunc (e *Executor) ExecCount() int {\n\treturn int(e.execCount.Load())\n}\n\n// CurrentCount returns currently running execution count\nfunc (e *Executor) CurrentCount() int {\n\treturn int(e.currentCount.Load())\n}\n\n// Reset resets the executor counters\nfunc (e *Executor) Reset() {\n\te.execCount.Store(0)\n\te.currentCount.Store(0)\n}\n\n// DefaultStreamDelay is the simulated delay for Agent Stream calls\n// This will be removed when real Agent calls are implemented\nconst DefaultStreamDelay = 50 * time.Millisecond\n\n// simulateStreamDelay simulates the delay of an Agent Stream call\nfunc (e *Executor) simulateStreamDelay() {\n\ttime.Sleep(DefaultStreamDelay)\n}\n\n// initUIFields initializes UI display fields based on trigger type with i18n support\n// Returns (name, currentTaskName)\nfunc (e *Executor) initUIFields(trigger robottypes.TriggerType, input *robottypes.TriggerInput, robot *robottypes.Robot) (string, string) {\n\t// Determine locale for UI messages\n\tlocale := getEffectiveLocale(robot, input)\n\n\t// Get localized default messages\n\tname := getLocalizedMessage(locale, \"preparing\")\n\tcurrentTaskName := getLocalizedMessage(locale, \"starting\")\n\n\tswitch trigger {\n\tcase robottypes.TriggerHuman:\n\t\t// For human trigger, extract name from first message\n\t\tif input != nil && len(input.Messages) > 0 {\n\t\t\tif content, ok := input.Messages[0].GetContentAsString(); ok && content != \"\" {\n\t\t\t\t// Use first 100 chars of message as name\n\t\t\t\tname = content\n\t\t\t\tif len(name) > 100 {\n\t\t\t\t\tname = name[:100] + \"...\"\n\t\t\t\t}\n\t\t\t}\n\t\t}\n\tcase robottypes.TriggerClock:\n\t\tname = getLocalizedMessage(locale, \"scheduled_execution\")\n\tcase robottypes.TriggerEvent:\n\t\tif input != nil && input.EventType != \"\" {\n\t\t\tname = getLocalizedMessage(locale, \"event_prefix\") + input.EventType\n\t\t} else {\n\t\t\tname = getLocalizedMessage(locale, \"event_triggered\")\n\t\t}\n\t}\n\n\treturn name, currentTaskName\n}\n\n// getEffectiveLocale determines the locale for UI display\n// Priority: input.Locale > robot.Config.DefaultLocale > \"en\"\nfunc getEffectiveLocale(robot *robottypes.Robot, input *robottypes.TriggerInput) string {\n\t// 1. Human trigger with explicit locale\n\tif input != nil && input.Locale != \"\" {\n\t\treturn input.Locale\n\t}\n\t// 2. Robot configured default\n\tif robot != nil && robot.Config != nil {\n\t\treturn robot.Config.GetDefaultLocale()\n\t}\n\t// 3. System default\n\treturn \"en\"\n}\n\n// i18n message maps for UI display fields\n// Use simple locale codes (en, zh) as keys\nvar uiMessages = map[string]map[string]string{\n\t\"en\": {\n\t\t\"preparing\": \"Preparing...\",\n\t\t\"starting\": \"Starting...\",\n\t\t\"scheduled_execution\": \"Scheduled execution\",\n\t\t\"event_prefix\": \"Event: \",\n\t\t\"event_triggered\": \"Event triggered\",\n\t\t\"analyzing_context\": \"Analyzing context...\",\n\t\t\"planning_goals\": \"Planning goals...\",\n\t\t\"breaking_down_tasks\": \"Breaking down tasks...\",\n\t\t\"generating_delivery\": \"Generating delivery content...\",\n\t\t\"sending_delivery\": \"Sending delivery...\",\n\t\t\"learning_from_exec\": \"Learning from execution...\",\n\t\t\"completed\": \"Completed\",\n\t\t\"cancelled\": \"Cancelled\",\n\t\t\"failed_prefix\": \"Failed at \",\n\t\t\"task_prefix\": \"Task\",\n\t\t// Phase names for failure messages\n\t\t\"phase_inspiration\": \"inspiration\",\n\t\t\"phase_goals\": \"goals\",\n\t\t\"phase_tasks\": \"tasks\",\n\t\t\"phase_run\": \"execution\",\n\t\t\"phase_delivery\": \"delivery\",\n\t\t\"phase_learning\": \"learning\",\n\t},\n\t\"zh\": {\n\t\t\"preparing\": \"准备中...\",\n\t\t\"starting\": \"启动中...\",\n\t\t\"scheduled_execution\": \"定时执行\",\n\t\t\"event_prefix\": \"事件: \",\n\t\t\"event_triggered\": \"事件触发\",\n\t\t\"analyzing_context\": \"分析上下文...\",\n\t\t\"planning_goals\": \"规划目标...\",\n\t\t\"breaking_down_tasks\": \"分解任务...\",\n\t\t\"generating_delivery\": \"生成交付内容...\",\n\t\t\"sending_delivery\": \"正在发送...\",\n\t\t\"learning_from_exec\": \"学习执行经验...\",\n\t\t\"completed\": \"已完成\",\n\t\t\"cancelled\": \"已取消\",\n\t\t\"failed_prefix\": \"失败于\",\n\t\t\"task_prefix\": \"任务\",\n\t\t// Phase names for failure messages\n\t\t\"phase_inspiration\": \"灵感阶段\",\n\t\t\"phase_goals\": \"目标阶段\",\n\t\t\"phase_tasks\": \"任务阶段\",\n\t\t\"phase_run\": \"执行阶段\",\n\t\t\"phase_delivery\": \"交付阶段\",\n\t\t\"phase_learning\": \"学习阶段\",\n\t},\n}\n\n// getLocalizedMessage returns a localized message for the given key\nfunc getLocalizedMessage(locale string, key string) string {\n\tif messages, ok := uiMessages[locale]; ok {\n\t\tif msg, ok := messages[key]; ok {\n\t\t\treturn msg\n\t\t}\n\t}\n\t// Fallback to English\n\tif messages, ok := uiMessages[\"en\"]; ok {\n\t\tif msg, ok := messages[key]; ok {\n\t\t\treturn msg\n\t\t}\n\t}\n\treturn key // Return key as fallback\n}\n\n// updateUIFields updates UI display fields and persists to database\nfunc (e *Executor) updateUIFields(ctx *robottypes.Context, exec *robottypes.Execution, name string, currentTaskName string) {\n\t// Update in-memory execution\n\tif name != \"\" {\n\t\texec.Name = name\n\t}\n\tif currentTaskName != \"\" {\n\t\texec.CurrentTaskName = currentTaskName\n\t}\n\n\t// Persist to database\n\tif !e.config.SkipPersistence && e.store != nil {\n\t\tif err := e.store.UpdateUIFields(ctx.Context, exec.ID, name, currentTaskName); err != nil {\n\t\t\tkunlog.With(kunlog.F{\n\t\t\t\t\"execution_id\": exec.ID,\n\t\t\t\t\"error\": err,\n\t\t\t}).Warn(\"Failed to update UI fields: %v\", err)\n\t\t}\n\t}\n}\n\n// updateTasksState persists the current tasks array with status to database\n// This should be called after each task status change for real-time UI updates\nfunc (e *Executor) updateTasksState(ctx *robottypes.Context, exec *robottypes.Execution) {\n\tif e.config.SkipPersistence || e.store == nil {\n\t\treturn\n\t}\n\n\t// Convert Current to store.CurrentState\n\tvar current *store.CurrentState\n\tif exec.Current != nil {\n\t\tcurrent = &store.CurrentState{\n\t\t\tTaskIndex: exec.Current.TaskIndex,\n\t\t\tProgress: exec.Current.Progress,\n\t\t}\n\t}\n\n\tif err := e.store.UpdateTasks(ctx.Context, exec.ID, exec.Tasks, current); err != nil {\n\t\tkunlog.With(kunlog.F{\n\t\t\t\"execution_id\": exec.ID,\n\t\t\t\"error\": err,\n\t\t}).Warn(\"Failed to update tasks state: %v\", err)\n\t}\n}\n\n// extractGoalName extracts the execution name from goals output\nfunc extractGoalName(goals *robottypes.Goals) string {\n\tif goals == nil || goals.Content == \"\" {\n\t\treturn \"\"\n\t}\n\n\t// Extract first non-empty, non-markdown-header line as the goal name\n\tcontent := goals.Content\n\tlines := strings.Split(content, \"\\n\")\n\n\tfor _, line := range lines {\n\t\tline = strings.TrimSpace(line)\n\t\tif line == \"\" {\n\t\t\tcontinue\n\t\t}\n\t\t// Skip markdown headers (# ## ### etc.)\n\t\tif strings.HasPrefix(line, \"#\") {\n\t\t\tcontinue\n\t\t}\n\t\t// Skip markdown horizontal rules (--- or ***)\n\t\tif strings.HasPrefix(line, \"---\") || strings.HasPrefix(line, \"***\") {\n\t\t\tcontinue\n\t\t}\n\t\t// Found a content line - strip markdown formatting\n\t\tline = stripMarkdownFormatting(line)\n\t\t// Limit length\n\t\tif len(line) > 150 {\n\t\t\tline = line[:150] + \"...\"\n\t\t}\n\t\treturn line\n\t}\n\n\t// Fallback: if all lines are headers, use first header without # prefix\n\tfor _, line := range lines {\n\t\tline = strings.TrimSpace(line)\n\t\tif line == \"\" {\n\t\t\tcontinue\n\t\t}\n\t\t// Strip leading # symbols\n\t\tline = strings.TrimLeft(line, \"#\")\n\t\tline = strings.TrimSpace(line)\n\t\tline = stripMarkdownFormatting(line)\n\t\tif line != \"\" {\n\t\t\tif len(line) > 150 {\n\t\t\t\tline = line[:150] + \"...\"\n\t\t\t}\n\t\t\treturn line\n\t\t}\n\t}\n\n\treturn \"\"\n}\n\n// stripMarkdownFormatting removes common markdown formatting from text\nfunc stripMarkdownFormatting(s string) string {\n\t// Remove bold/italic markers\n\ts = strings.ReplaceAll(s, \"**\", \"\")\n\ts = strings.ReplaceAll(s, \"__\", \"\")\n\ts = strings.ReplaceAll(s, \"*\", \"\")\n\ts = strings.ReplaceAll(s, \"_\", \"\")\n\t// Remove inline code\n\ts = strings.ReplaceAll(s, \"`\", \"\")\n\t// Remove link syntax [text](url) -> text\n\t// Simple approach: just remove brackets and parentheses content\n\tfor {\n\t\tstart := strings.Index(s, \"[\")\n\t\tif start == -1 {\n\t\t\tbreak\n\t\t}\n\t\tend := strings.Index(s[start:], \"]\")\n\t\tif end == -1 {\n\t\t\tbreak\n\t\t}\n\t\tlinkEnd := start + end\n\t\t// Check if followed by (url)\n\t\tif linkEnd+1 < len(s) && s[linkEnd+1] == '(' {\n\t\t\tparenEnd := strings.Index(s[linkEnd+1:], \")\")\n\t\t\tif parenEnd != -1 {\n\t\t\t\t// Extract just the link text\n\t\t\t\tlinkText := s[start+1 : linkEnd]\n\t\t\t\ts = s[:start] + linkText + s[linkEnd+1+parenEnd+1:]\n\t\t\t\tcontinue\n\t\t\t}\n\t\t}\n\t\t// Just remove brackets\n\t\ts = s[:start] + s[start+1:linkEnd] + s[linkEnd+1:]\n\t}\n\treturn strings.TrimSpace(s)\n}\n\n// Suspend transitions the execution to waiting status, persists state, and returns\n// ErrExecutionSuspended so the caller stops further phase processing.\nfunc (e *Executor) Suspend(ctx *robottypes.Context, exec *robottypes.Execution, taskIndex int, question string) error {\n\tnow := time.Now()\n\ttaskID := \"\"\n\tif taskIndex >= 0 && taskIndex < len(exec.Tasks) {\n\t\ttaskID = exec.Tasks[taskIndex].ID\n\t\texec.Tasks[taskIndex].Status = robottypes.TaskWaitingInput\n\t}\n\n\texec.Status = robottypes.ExecWaiting\n\texec.WaitingTaskID = taskID\n\texec.WaitingQuestion = question\n\texec.WaitingSince = &now\n\texec.ResumeContext = &robottypes.ResumeContext{\n\t\tTaskIndex: taskIndex,\n\t\tPreviousResults: exec.Results,\n\t}\n\n\tif !e.config.SkipPersistence && e.store != nil {\n\t\t// Persist task state (waiting_input on the specific task)\n\t\te.updateTasksState(ctx, exec)\n\t\t// Persist P3 results so UI can show completed tasks while waiting (§16.26)\n\t\tif err := e.store.UpdatePhase(ctx.Context, exec.ID, robottypes.PhaseRun, exec.Results); err != nil {\n\t\t\tkunlog.With(kunlog.F{\n\t\t\t\t\"execution_id\": exec.ID,\n\t\t\t\t\"error\": err,\n\t\t\t}).Warn(\"Failed to persist partial results on suspend: %v\", err)\n\t\t}\n\t\t// Persist suspend state atomically\n\t\tif err := e.store.UpdateSuspendState(ctx.Context, exec.ID, taskID, question, exec.ResumeContext); err != nil {\n\t\t\tkunlog.With(kunlog.F{\n\t\t\t\t\"execution_id\": exec.ID,\n\t\t\t\t\"task_id\": taskID,\n\t\t\t\t\"error\": err,\n\t\t\t}).Warn(\"Failed to persist suspend state: %v\", err)\n\t\t}\n\t}\n\n\tkunlog.With(kunlog.F{\n\t\t\"execution_id\": exec.ID,\n\t\t\"member_id\": exec.MemberID,\n\t\t\"task_id\": taskID,\n\t\t\"question\": question,\n\t}).Info(\"Execution suspended, waiting for human input\")\n\n\t// Fire event (best-effort, errors are ignored)\n\tevent.Push(ctx.Context, robotevents.ExecWaiting, robotevents.NeedInputPayload{\n\t\tExecutionID: exec.ID,\n\t\tMemberID: exec.MemberID,\n\t\tTeamID: exec.TeamID,\n\t\tTaskID: taskID,\n\t\tQuestion: question,\n\t\tChatID: exec.ChatID,\n\t})\n\n\treturn robottypes.ErrExecutionSuspended\n}\n\n// Resume resumes a suspended execution with human-provided input.\n// Loads execution from DB, restores state, injects reply, and continues from the suspended task.\nfunc (e *Executor) Resume(ctx *robottypes.Context, execID string, reply string) error {\n\tif ctx == nil {\n\t\treturn fmt.Errorf(\"context is required for resume\")\n\t}\n\tif execID == \"\" {\n\t\treturn fmt.Errorf(\"execID cannot be empty\")\n\t}\n\tif e.store == nil {\n\t\treturn fmt.Errorf(\"store is required for resume\")\n\t}\n\n\t// Load execution record from DB\n\trecord, err := e.store.Get(ctx.Context, execID)\n\tif err != nil {\n\t\treturn fmt.Errorf(\"failed to load execution: %w\", err)\n\t}\n\tif record == nil {\n\t\treturn fmt.Errorf(\"execution not found: %s\", execID)\n\t}\n\tif record.Status != robottypes.ExecWaiting {\n\t\treturn fmt.Errorf(\"execution %s is not in waiting status (current: %s)\", execID, record.Status)\n\t}\n\n\t// Restore runtime execution from record\n\texec := record.ToExecution()\n\n\t// Load robot from store\n\tif e.robotStore == nil {\n\t\treturn fmt.Errorf(\"robot store is required for resume\")\n\t}\n\trobotRecord, err := e.robotStore.Get(ctx.Context, exec.MemberID)\n\tif err != nil {\n\t\treturn fmt.Errorf(\"failed to load robot: %w\", err)\n\t}\n\tif robotRecord == nil {\n\t\treturn fmt.Errorf(\"robot not found: %s\", exec.MemberID)\n\t}\n\trobot, err := robotRecord.ToRobot()\n\tif err != nil {\n\t\treturn fmt.Errorf(\"failed to convert robot record: %w\", err)\n\t}\n\texec.SetRobot(robot)\n\n\t// Re-add execution to robot's in-memory tracking (skips quota check per §16.30)\n\trobot.AddExecution(exec)\n\n\t// Maintain executor concurrency count (§16.21)\n\te.currentCount.Add(1)\n\tdefer e.currentCount.Add(-1)\n\n\t// Defer cleanup: mirror ExecuteWithControl's defer logic (§16.21)\n\tdefer func() {\n\t\tif exec.Status == robottypes.ExecWaiting {\n\t\t\treturn // re-suspended, keep tracking\n\t\t}\n\t\trobot.RemoveExecution(exec.ID)\n\t\tif robot.RunningCount() == 0 && !e.config.SkipPersistence && e.robotStore != nil {\n\t\t\tif err := e.robotStore.UpdateStatus(ctx.Context, robot.MemberID, robottypes.RobotIdle); err != nil {\n\t\t\t\tkunlog.With(kunlog.F{\n\t\t\t\t\t\"member_id\": robot.MemberID,\n\t\t\t\t\t\"error\": err,\n\t\t\t\t}).Warn(\"Failed to update robot status to idle after resume: %v\", err)\n\t\t\t}\n\t\t}\n\t}()\n\n\t// Handle __skip__: mark waiting task as skipped and advance to next task\n\tif reply == \"__skip__\" && exec.ResumeContext != nil {\n\t\tti := exec.ResumeContext.TaskIndex\n\t\tif ti >= 0 && ti < len(exec.Tasks) {\n\t\t\ttask := &exec.Tasks[ti]\n\t\t\ttask.Status = robottypes.TaskSkipped\n\t\t\texec.ResumeContext.PreviousResults = append(exec.ResumeContext.PreviousResults, robottypes.TaskResult{\n\t\t\t\tTaskID: task.ID,\n\t\t\t\tSuccess: false,\n\t\t\t\tOutput: \"skipped\",\n\t\t\t\tDuration: 0,\n\t\t\t})\n\t\t\texec.ResumeContext.TaskIndex = ti + 1\n\t\t\tif !e.config.SkipPersistence && e.store != nil {\n\t\t\t\te.updateTasksState(ctx, exec)\n\t\t\t}\n\t\t}\n\t\treply = \"\" // Don't inject __skip__ as a message\n\t}\n\n\t// Inject reply into the waiting task's messages so the re-executed task gets context\n\tif exec.ResumeContext != nil {\n\t\tti := exec.ResumeContext.TaskIndex\n\t\tif ti >= 0 && ti < len(exec.Tasks) && reply != \"\" {\n\t\t\texec.Tasks[ti].Messages = append(exec.Tasks[ti].Messages, agentcontext.Message{\n\t\t\t\tRole: agentcontext.RoleUser,\n\t\t\t\tContent: fmt.Sprintf(\"[Human reply] %s\", reply),\n\t\t\t})\n\t\t}\n\t}\n\n\t// Clear waiting fields and transition back to running\n\texec.Status = robottypes.ExecRunning\n\texec.WaitingTaskID = \"\"\n\texec.WaitingQuestion = \"\"\n\texec.WaitingSince = nil\n\n\tif !e.config.SkipPersistence && e.store != nil {\n\t\tif err := e.store.UpdateResumeState(ctx.Context, exec.ID); err != nil {\n\t\t\tkunlog.With(kunlog.F{\n\t\t\t\t\"execution_id\": exec.ID,\n\t\t\t\t\"error\": err,\n\t\t\t}).Warn(\"Failed to persist resume state: %v\", err)\n\t\t}\n\t}\n\n\tkunlog.With(kunlog.F{\n\t\t\"execution_id\": exec.ID,\n\t\t\"member_id\": exec.MemberID,\n\t\t\"reply_len\": len(reply),\n\t}).Info(\"Execution resumed\")\n\n\tevent.Push(ctx.Context, robotevents.ExecResumed, robotevents.ExecPayload{\n\t\tExecutionID: exec.ID,\n\t\tMemberID: exec.MemberID,\n\t\tTeamID: exec.TeamID,\n\t\tChatID: exec.ChatID,\n\t})\n\n\t// Continue P3 (Run) from where it was suspended\n\tif err := e.RunExecution(ctx, exec, nil); err != nil {\n\t\tif err == robottypes.ErrExecutionSuspended {\n\t\t\treturn err\n\t\t}\n\t\texec.Status = robottypes.ExecFailed\n\t\texec.Error = err.Error()\n\t\tif !e.config.SkipPersistence && e.store != nil {\n\t\t\t_ = e.store.UpdateStatus(ctx.Context, exec.ID, robottypes.ExecFailed, err.Error())\n\t\t}\n\t\treturn err\n\t}\n\n\t// Clear resume context after successful P3 completion\n\texec.ResumeContext = nil\n\n\t// Continue with P4 (Delivery) and P5 (Learning)\n\tlocale := getEffectiveLocale(robot, exec.Input)\n\tfor _, phase := range []robottypes.Phase{robottypes.PhaseDelivery, robottypes.PhaseLearning} {\n\t\tif err := e.runPhase(ctx, exec, phase, nil, nil); err != nil {\n\t\t\tif err == robottypes.ErrExecutionSuspended {\n\t\t\t\treturn err\n\t\t\t}\n\t\t\texec.Status = robottypes.ExecFailed\n\t\t\texec.Error = err.Error()\n\t\t\tfailedPrefix := getLocalizedMessage(locale, \"failed_prefix\")\n\t\t\tphaseName := getLocalizedMessage(locale, \"phase_\"+string(phase))\n\t\t\te.updateUIFields(ctx, exec, \"\", failedPrefix+phaseName)\n\t\t\tif !e.config.SkipPersistence && e.store != nil {\n\t\t\t\t_ = e.store.UpdateStatus(ctx.Context, exec.ID, robottypes.ExecFailed, err.Error())\n\t\t\t}\n\t\t\treturn fmt.Errorf(\"resume phase %s failed: %w\", phase, err)\n\t\t}\n\t}\n\n\t// Mark completed\n\texec.Status = robottypes.ExecCompleted\n\tnow := time.Now()\n\texec.EndTime = &now\n\te.updateUIFields(ctx, exec, \"\", getLocalizedMessage(locale, \"completed\"))\n\tif !e.config.SkipPersistence && e.store != nil {\n\t\t_ = e.store.UpdateStatus(ctx.Context, exec.ID, robottypes.ExecCompleted, \"\")\n\t}\n\n\treturn nil\n}\n\n// Verify Executor implements types.Executor\nvar _ types.Executor = (*Executor)(nil)\n" }, { "path": "agent/robot/executor/standard/executor_test.go", "content": "package standard_test\n\nimport (\n\t\"context\"\n\t\"testing\"\n\n\t\"github.com/stretchr/testify/assert\"\n\t\"github.com/stretchr/testify/require\"\n\t\"github.com/yaoapp/yao/agent/robot/executor/standard\"\n\t\"github.com/yaoapp/yao/agent/robot/executor/types\"\n\t\"github.com/yaoapp/yao/agent/robot/store\"\n\trobottypes \"github.com/yaoapp/yao/agent/robot/types\"\n\t\"github.com/yaoapp/yao/agent/testutils\"\n\toauthTypes \"github.com/yaoapp/yao/openapi/oauth/types\"\n)\n\n// ============================================================================\n// Executor Persistence Integration Tests\n// ============================================================================\n\nfunc TestExecutorPersistence(t *testing.T) {\n\tif testing.Short() {\n\t\tt.Skip(\"Skipping integration test\")\n\t}\n\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\tt.Run(\"persists_execution_record_on_start\", func(t *testing.T) {\n\t\tctx := robottypes.NewContext(context.Background(), &oauthTypes.AuthorizedInfo{\n\t\t\tUserID: \"user_persist_001\",\n\t\t\tTeamID: \"team_persist_001\",\n\t\t})\n\n\t\trobot := createPersistenceTestRobot(\"member_persist_001\", \"team_persist_001\")\n\n\t\t// Create executor with persistence enabled\n\t\te := standard.NewWithConfig(types.Config{\n\t\t\tSkipPersistence: false,\n\t\t})\n\n\t\t// Execute with simulated failure to ensure we get a result\n\t\texec, err := e.Execute(ctx, robot, robottypes.TriggerHuman, \"simulate_failure\")\n\t\trequire.NoError(t, err)\n\t\trequire.NotNil(t, exec)\n\n\t\t// Verify execution record was persisted\n\t\ts := store.NewExecutionStore()\n\t\trecord, err := s.Get(context.Background(), exec.ID)\n\t\trequire.NoError(t, err)\n\t\trequire.NotNil(t, record)\n\n\t\tassert.Equal(t, exec.ID, record.ExecutionID)\n\t\tassert.Equal(t, \"member_persist_001\", record.MemberID)\n\t\tassert.Equal(t, \"team_persist_001\", record.TeamID)\n\t\tassert.Equal(t, robottypes.TriggerHuman, record.TriggerType)\n\t\tassert.Equal(t, robottypes.ExecFailed, record.Status)\n\t\tassert.Equal(t, \"simulated failure\", record.Error)\n\n\t\t// Cleanup\n\t\t_ = s.Delete(context.Background(), exec.ID)\n\t})\n\n\tt.Run(\"persists_failed_status_with_error\", func(t *testing.T) {\n\t\tctx := robottypes.NewContext(context.Background(), &oauthTypes.AuthorizedInfo{\n\t\t\tUserID: \"user_persist_002\",\n\t\t\tTeamID: \"team_persist_002\",\n\t\t})\n\n\t\trobot := createPersistenceTestRobot(\"member_persist_002\", \"team_persist_002\")\n\n\t\te := standard.NewWithConfig(types.Config{\n\t\t\tSkipPersistence: false,\n\t\t})\n\n\t\t// Execute with simulated failure\n\t\texec, err := e.Execute(ctx, robot, robottypes.TriggerHuman, \"simulate_failure\")\n\t\trequire.NoError(t, err)\n\t\trequire.NotNil(t, exec)\n\n\t\t// Verify the record has failed status with error message\n\t\ts := store.NewExecutionStore()\n\t\trecord, err := s.Get(context.Background(), exec.ID)\n\t\trequire.NoError(t, err)\n\t\trequire.NotNil(t, record)\n\n\t\tassert.Equal(t, robottypes.ExecFailed, record.Status)\n\t\tassert.Equal(t, \"simulated failure\", record.Error)\n\t\tassert.NotNil(t, record.StartTime)\n\n\t\t// Cleanup\n\t\t_ = s.Delete(context.Background(), exec.ID)\n\t})\n\n\tt.Run(\"skips_persistence_when_disabled\", func(t *testing.T) {\n\t\tctx := robottypes.NewContext(context.Background(), &oauthTypes.AuthorizedInfo{\n\t\t\tUserID: \"user_persist_003\",\n\t\t\tTeamID: \"team_persist_003\",\n\t\t})\n\n\t\trobot := createPersistenceTestRobot(\"member_persist_003\", \"team_persist_003\")\n\n\t\t// Create executor with persistence disabled\n\t\te := standard.NewWithConfig(types.Config{\n\t\t\tSkipPersistence: true,\n\t\t})\n\n\t\texec, err := e.Execute(ctx, robot, robottypes.TriggerHuman, \"simulate_failure\")\n\t\trequire.NoError(t, err)\n\t\trequire.NotNil(t, exec)\n\n\t\t// Verify no record was created\n\t\ts := store.NewExecutionStore()\n\t\trecord, err := s.Get(context.Background(), exec.ID)\n\t\trequire.NoError(t, err)\n\t\tassert.Nil(t, record) // Should not exist\n\t})\n}\n\n// ============================================================================\n// Goals Injection Tests (Host Agent confirmed goals)\n// ============================================================================\n\n// TestExecutorGoalsInjection verifies that when TriggerHuman is used with\n// pre-confirmed goals (from Host Agent via /v1/agent/robots/:id/execute),\n// the goals are injected directly into exec.Goals before RunGoals runs,\n// and are persisted (title updated) so the task list shows the correct title.\nfunc TestExecutorGoalsInjection(t *testing.T) {\n\tif testing.Short() {\n\t\tt.Skip(\"Skipping integration test\")\n\t}\n\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\tt.Run(\"goals_injected_from_trigger_input_data\", func(t *testing.T) {\n\t\tctx := robottypes.NewContext(context.Background(), &oauthTypes.AuthorizedInfo{\n\t\t\tUserID: \"user_goals_inject_001\",\n\t\t\tTeamID: \"team_goals_inject_001\",\n\t\t})\n\n\t\trobot := createPersistenceTestRobot(\"member_goals_inject_001\", \"team_goals_inject_001\")\n\n\t\te := standard.NewWithConfig(types.Config{\n\t\t\tSkipPersistence: false,\n\t\t})\n\n\t\t// Simulate Host Agent confirmed goals passed via TriggerInput.Data\n\t\ttriggerInput := &robottypes.TriggerInput{\n\t\t\tData: map[string]interface{}{\n\t\t\t\t\"goals\": \"Create a mecha image with sci-fi style\",\n\t\t\t\t\"chat_id\": \"robot_member_goals_inject_001_1234567890\",\n\t\t\t},\n\t\t}\n\n\t\texec, err := e.Execute(ctx, robot, robottypes.TriggerHuman, triggerInput)\n\t\trequire.NoError(t, err)\n\t\trequire.NotNil(t, exec)\n\n\t\t// Goals should be injected from TriggerInput.Data\n\t\trequire.NotNil(t, exec.Goals, \"Goals should be injected from TriggerInput.Data\")\n\t\tassert.Equal(t, \"Create a mecha image with sci-fi style\", exec.Goals.Content,\n\t\t\t\"Goals content should match the pre-confirmed goals\")\n\n\t\t// Verify goals were persisted to the store\n\t\ts := store.NewExecutionStore()\n\t\trecord, err := s.Get(context.Background(), exec.ID)\n\t\trequire.NoError(t, err)\n\t\trequire.NotNil(t, record)\n\n\t\t// Execution name should reflect the goals (not \"Preparing...\")\n\t\tassert.NotEmpty(t, exec.Name, \"Execution name should be set from goals\")\n\t\tassert.NotEqual(t, \"Preparing...\", exec.Name, \"Name should not be the default placeholder\")\n\n\t\t// Cleanup\n\t\t_ = s.Delete(context.Background(), exec.ID)\n\t\tt.Logf(\"✓ Goals injected from TriggerInput.Data: goals=%q, name=%q\",\n\t\t\texec.Goals.Content, exec.Name)\n\t})\n\n\tt.Run(\"empty_goals_falls_through_to_goals_agent\", func(t *testing.T) {\n\t\t// When TriggerInput.Data[\"goals\"] is an empty string, the executor\n\t\t// does NOT inject pre-confirmed goals and falls through to RunGoals.\n\t\t// RunGoals will call the Goals Agent (LLM), which may succeed or fail\n\t\t// depending on the environment. We only verify that the executor returns\n\t\t// without a panic and that no pre-confirmed goals were force-injected.\n\t\t//\n\t\t// This test requires a running AI environment; skip in short mode.\n\t\tif testing.Short() {\n\t\t\tt.Skip(\"Skipping: requires LLM for RunGoals fallback\")\n\t\t}\n\n\t\tctx := robottypes.NewContext(context.Background(), &oauthTypes.AuthorizedInfo{\n\t\t\tUserID: \"user_goals_empty_002\",\n\t\t\tTeamID: \"team_goals_empty_002\",\n\t\t})\n\n\t\trobot := createPersistenceTestRobot(\"member_goals_empty_002\", \"team_goals_empty_002\")\n\n\t\te := standard.NewWithConfig(types.Config{\n\t\t\tSkipPersistence: true,\n\t\t})\n\n\t\ttriggerInput := &robottypes.TriggerInput{\n\t\t\tData: map[string]interface{}{\n\t\t\t\t\"goals\": \"\", // empty — should not be injected as pre-confirmed\n\t\t\t},\n\t\t}\n\n\t\texec, err := e.Execute(ctx, robot, robottypes.TriggerHuman, triggerInput)\n\t\trequire.NoError(t, err)\n\t\trequire.NotNil(t, exec)\n\n\t\t// If Goals was set, it came from the Goals Agent, NOT from the empty string injection.\n\t\t// Either nil (agent skipped) or non-nil (agent ran) is acceptable.\n\t\tif exec.Goals != nil {\n\t\t\tassert.NotEmpty(t, exec.Goals.Content,\n\t\t\t\t\"If Goals Agent ran, content should be non-empty\")\n\t\t}\n\t\tt.Logf(\"✓ Empty goals falls through to Goals Agent (goals=%v)\", exec.Goals != nil)\n\t})\n\n\tt.Run(\"no_trigger_input_uses_normal_flow\", func(t *testing.T) {\n\t\tctx := robottypes.NewContext(context.Background(), &oauthTypes.AuthorizedInfo{\n\t\t\tUserID: \"user_goals_normal_001\",\n\t\t\tTeamID: \"team_goals_normal_001\",\n\t\t})\n\n\t\trobot := createPersistenceTestRobot(\"member_goals_normal_001\", \"team_goals_normal_001\")\n\n\t\te := standard.NewWithConfig(types.Config{\n\t\t\tSkipPersistence: true,\n\t\t})\n\n\t\t// No TriggerInput — simulate plain string fallback (old API usage)\n\t\texec, err := e.Execute(ctx, robot, robottypes.TriggerHuman, \"simulate_failure\")\n\t\trequire.NoError(t, err)\n\t\trequire.NotNil(t, exec)\n\n\t\t// Goals nil is expected — RunGoals would normally call the LLM\n\t\tassert.Nil(t, exec.Goals, \"Without pre-confirmed goals, Goals should remain nil\")\n\t\tt.Logf(\"✓ Normal flow (no pre-confirmed goals) proceeds without injection\")\n\t})\n}\n\n// ============================================================================\n// Helper Functions\n// ============================================================================\n\nfunc createPersistenceTestRobot(memberID, teamID string) *robottypes.Robot {\n\treturn &robottypes.Robot{\n\t\tMemberID: memberID,\n\t\tTeamID: teamID,\n\t\tDisplayName: \"Persistence Test Robot\",\n\t\tStatus: robottypes.RobotIdle,\n\t\tAutonomousMode: true,\n\t\tConfig: &robottypes.Config{\n\t\t\tIdentity: &robottypes.Identity{\n\t\t\t\tRole: \"Test Robot\",\n\t\t\t\tDuties: []string{\"Testing persistence\"},\n\t\t\t},\n\t\t\tQuota: &robottypes.Quota{\n\t\t\t\tMax: 5,\n\t\t\t\tQueue: 10,\n\t\t\t},\n\t\t\tTriggers: &robottypes.Triggers{\n\t\t\t\tIntervene: &robottypes.TriggerSwitch{Enabled: true},\n\t\t\t},\n\t\t\tResources: &robottypes.Resources{\n\t\t\t\tPhases: map[robottypes.Phase]string{\n\t\t\t\t\trobottypes.PhaseInspiration: \"robot.inspiration\",\n\t\t\t\t\trobottypes.PhaseGoals: \"robot.goals\",\n\t\t\t\t\trobottypes.PhaseTasks: \"robot.tasks\",\n\t\t\t\t\trobottypes.PhaseRun: \"robot.validation\",\n\t\t\t\t\t\"validation\": \"robot.validation\",\n\t\t\t\t},\n\t\t\t\tAgents: []string{\"experts.text-writer\", \"experts.data-analyst\"},\n\t\t\t},\n\t\t},\n\t}\n}\n" }, { "path": "agent/robot/executor/standard/goals.go", "content": "package standard\n\nimport (\n\t\"fmt\"\n\t\"strings\"\n\n\trobottypes \"github.com/yaoapp/yao/agent/robot/types\"\n)\n\n// RunGoals executes P1: Goals phase\n// Calls the Goals Agent to plan daily objectives\n//\n// Input:\n// - InspirationReport (from P0) for clock trigger\n// - TriggerInput for human/event trigger\n//\n// Output:\n// - Goals with markdown content and delivery info\nfunc (e *Executor) RunGoals(ctx *robottypes.Context, exec *robottypes.Execution, _ interface{}) error {\n\t// §18.2: confirming phase may have already populated Goals — skip regeneration\n\tif exec.Goals != nil && exec.Goals.Content != \"\" {\n\t\treturn nil\n\t}\n\n\t// Get robot for identity and resources\n\trobot := exec.GetRobot()\n\tif robot == nil {\n\t\treturn fmt.Errorf(\"robot not found in execution\")\n\t}\n\n\t// Update UI field with i18n\n\tlocale := getEffectiveLocale(robot, exec.Input)\n\te.updateUIFields(ctx, exec, \"\", getLocalizedMessage(locale, \"planning_goals\"))\n\n\t// Get agent ID for goals phase\n\tagentID := \"__yao.goals\" // default\n\tif robot.Config != nil && robot.Config.Resources != nil {\n\t\tagentID = robot.Config.Resources.GetPhaseAgent(robottypes.PhaseGoals)\n\t}\n\n\t// Build prompt based on trigger type\n\tformatter := NewInputFormatter()\n\tvar userContent string\n\n\tswitch exec.TriggerType {\n\tcase robottypes.TriggerClock:\n\t\t// For clock trigger: use InspirationReport from P0\n\t\tif exec.Inspiration != nil {\n\t\t\tuserContent = formatter.FormatInspirationReport(exec.Inspiration)\n\t\t} else {\n\t\t\t// Fallback: if no inspiration report, create minimal context\n\t\t\tuserContent = formatter.FormatClockContext(\n\t\t\t\trobottypes.NewClockContext(exec.StartTime, \"\"),\n\t\t\t\trobot,\n\t\t\t)\n\t\t}\n\n\tcase robottypes.TriggerHuman, robottypes.TriggerEvent:\n\t\t// For human/event trigger: use TriggerInput directly\n\t\tif exec.Input != nil {\n\t\t\tuserContent = formatter.FormatTriggerInput(exec.Input)\n\t\t}\n\t}\n\n\t// Add robot identity context if not already included\n\t// For clock trigger with inspiration report, identity is not in the report\n\t// For human/event trigger, identity provides context\n\tif robot.Config != nil && robot.Config.Identity != nil {\n\t\tif !strings.Contains(userContent, \"## Robot Identity\") {\n\t\t\tuserContent = formatter.FormatRobotIdentity(robot) + \"\\n\\n\" + userContent\n\t\t}\n\t}\n\n\t// Add available resources - critical for generating achievable goals\n\t// Without knowing what tools are available, goals might be unachievable\n\tresourcesContent := formatter.FormatAvailableResources(robot)\n\tif resourcesContent != \"\" {\n\t\tuserContent += \"\\n\\n\" + resourcesContent\n\t}\n\n\tif userContent == \"\" {\n\t\treturn fmt.Errorf(\"no input available for goals generation\")\n\t}\n\n\t// Call agent\n\tcaller := NewAgentCaller()\n\tcaller.Connector = robot.LanguageModel\n\tresult, err := caller.CallWithMessages(ctx, agentID, userContent)\n\tif err != nil {\n\t\treturn fmt.Errorf(\"goals agent (%s) call failed: %w\", agentID, err)\n\t}\n\n\t// Parse response as JSON\n\t// Goals Agent returns: { \"content\": \"...\", \"delivery\": {...} }\n\tdata, err := result.GetJSON()\n\tif err != nil {\n\t\t// Fallback: if not JSON, use raw text as content\n\t\tcontent := result.GetText()\n\t\tif content == \"\" {\n\t\t\treturn fmt.Errorf(\"goals agent returned empty response\")\n\t\t}\n\t\texec.Goals = &robottypes.Goals{\n\t\t\tContent: content,\n\t\t}\n\t\treturn nil\n\t}\n\n\t// Build Goals from JSON\n\texec.Goals = &robottypes.Goals{}\n\n\t// Extract content (markdown)\n\tif content, ok := data[\"content\"].(string); ok {\n\t\texec.Goals.Content = content\n\t}\n\n\t// Extract delivery\n\tif delivery, ok := data[\"delivery\"].(map[string]interface{}); ok {\n\t\texec.Goals.Delivery = ParseDelivery(delivery)\n\t}\n\n\t// Validate: content is required\n\tif exec.Goals.Content == \"\" {\n\t\treturn fmt.Errorf(\"goals agent (%s) returned empty content\", agentID)\n\t}\n\n\t// Update Name from goals content (extract first line as execution title)\n\tif goalName := extractGoalName(exec.Goals); goalName != \"\" {\n\t\te.updateUIFields(ctx, exec, goalName, \"\")\n\t}\n\n\treturn nil\n}\n\n// ParseDelivery converts map to DeliveryTarget struct\n// Returns nil if data is nil or type is invalid/missing\nfunc ParseDelivery(data map[string]interface{}) *robottypes.DeliveryTarget {\n\tif data == nil {\n\t\treturn nil\n\t}\n\n\t// Type is required - if missing or invalid, return nil\n\tt, ok := data[\"type\"].(string)\n\tif !ok || t == \"\" {\n\t\treturn nil\n\t}\n\n\tdeliveryType := robottypes.DeliveryType(t)\n\tif !IsValidDeliveryType(deliveryType) {\n\t\t// Invalid type - return nil to indicate parsing failure\n\t\treturn nil\n\t}\n\n\ttarget := &robottypes.DeliveryTarget{\n\t\tType: deliveryType,\n\t}\n\n\t// Parse recipients\n\tif recipients, ok := data[\"recipients\"].([]interface{}); ok {\n\t\tfor _, r := range recipients {\n\t\t\tif s, ok := r.(string); ok {\n\t\t\t\ttarget.Recipients = append(target.Recipients, s)\n\t\t\t}\n\t\t}\n\t}\n\n\t// Parse format\n\tif format, ok := data[\"format\"].(string); ok {\n\t\ttarget.Format = format\n\t}\n\n\t// Parse template\n\tif template, ok := data[\"template\"].(string); ok {\n\t\ttarget.Template = template\n\t}\n\n\t// Parse options\n\tif options, ok := data[\"options\"].(map[string]interface{}); ok {\n\t\ttarget.Options = options\n\t}\n\n\treturn target\n}\n\n// IsValidDeliveryType checks if the delivery type is valid\nfunc IsValidDeliveryType(t robottypes.DeliveryType) bool {\n\tswitch t {\n\tcase robottypes.DeliveryEmail, robottypes.DeliveryWebhook,\n\t\trobottypes.DeliveryProcess, robottypes.DeliveryNotify:\n\t\treturn true\n\tdefault:\n\t\treturn false\n\t}\n}\n" }, { "path": "agent/robot/executor/standard/goals_test.go", "content": "package standard_test\n\nimport (\n\t\"context\"\n\t\"strings\"\n\t\"testing\"\n\t\"time\"\n\n\t\"github.com/stretchr/testify/assert\"\n\t\"github.com/stretchr/testify/require\"\n\tagentcontext \"github.com/yaoapp/yao/agent/context\"\n\t\"github.com/yaoapp/yao/agent/robot/executor/standard\"\n\t\"github.com/yaoapp/yao/agent/robot/types\"\n\t\"github.com/yaoapp/yao/agent/testutils\"\n)\n\n// ============================================================================\n// P1 Goals Phase Tests\n// ============================================================================\n\nfunc TestRunGoalsBasic(t *testing.T) {\n\tif testing.Short() {\n\t\tt.Skip(\"Skipping integration test\")\n\t}\n\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\tctx := types.NewContext(context.Background(), testAuth())\n\n\tt.Run(\"generates goals from inspiration report (clock trigger)\", func(t *testing.T) {\n\t\t// Create robot with goals agent configured\n\t\trobot := createGoalsTestRobot(t, \"robot.goals\")\n\n\t\t// Create execution with inspiration report (from P0)\n\t\texec := createGoalsTestExecution(robot, types.TriggerClock)\n\t\texec.Inspiration = &types.InspirationReport{\n\t\t\tClock: types.NewClockContext(time.Now(), \"\"),\n\t\t\tContent: \"## Summary\\nToday is Monday morning. Focus on weekly planning.\\n\\n## Highlights\\n- New sales leads arrived\\n- Weekly report due Friday\",\n\t\t}\n\n\t\t// Run goals phase\n\t\te := standard.New()\n\t\terr := e.RunGoals(ctx, exec, nil)\n\n\t\trequire.NoError(t, err)\n\t\trequire.NotNil(t, exec.Goals)\n\t\tassert.NotEmpty(t, exec.Goals.Content)\n\t})\n\n\tt.Run(\"includes priority markers in output\", func(t *testing.T) {\n\t\trobot := createGoalsTestRobot(t, \"robot.goals\")\n\t\texec := createGoalsTestExecution(robot, types.TriggerClock)\n\t\texec.Inspiration = &types.InspirationReport{\n\t\t\tClock: types.NewClockContext(time.Now(), \"\"),\n\t\t\tContent: \"## Summary\\nUrgent: Customer complaint needs attention.\\n\\n## Highlights\\n- Critical bug reported\\n- Regular maintenance scheduled\",\n\t\t}\n\n\t\te := standard.New()\n\t\terr := e.RunGoals(ctx, exec, nil)\n\n\t\trequire.NoError(t, err)\n\t\tcontent := exec.Goals.Content\n\n\t\t// Verify expected structure in markdown output\n\t\t// Note: LLM output is non-deterministic, so we check for likely patterns\n\t\thasGoals := strings.Contains(content, \"Goal\") ||\n\t\t\tstrings.Contains(content, \"##\") ||\n\t\t\tstrings.Contains(content, \"High\") ||\n\t\t\tstrings.Contains(content, \"Normal\") ||\n\t\t\tstrings.Contains(content, \"1.\")\n\n\t\tassert.True(t, hasGoals, \"should contain goals structure, got: %s\", content)\n\t})\n}\n\nfunc TestRunGoalsHumanTrigger(t *testing.T) {\n\tif testing.Short() {\n\t\tt.Skip(\"Skipping integration test\")\n\t}\n\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\tctx := types.NewContext(context.Background(), testAuth())\n\n\tt.Run(\"generates goals from human intervention\", func(t *testing.T) {\n\t\trobot := createGoalsTestRobot(t, \"robot.goals\")\n\t\texec := createGoalsTestExecution(robot, types.TriggerHuman)\n\n\t\t// Set human intervention input\n\t\texec.Input = &types.TriggerInput{\n\t\t\tAction: \"task.add\",\n\t\t\tUserID: \"user-123\",\n\t\t\tMessages: []agentcontext.Message{\n\t\t\t\t{Role: agentcontext.RoleUser, Content: \"Please analyze the Q4 sales data and prepare a summary report for the management meeting tomorrow.\"},\n\t\t\t},\n\t\t}\n\n\t\te := standard.New()\n\t\terr := e.RunGoals(ctx, exec, nil)\n\n\t\trequire.NoError(t, err)\n\t\trequire.NotNil(t, exec.Goals)\n\t\tassert.NotEmpty(t, exec.Goals.Content)\n\n\t\t// Goals should be related to the user request\n\t\tcontent := strings.ToLower(exec.Goals.Content)\n\t\thasRelevantContent := strings.Contains(content, \"sales\") ||\n\t\t\tstrings.Contains(content, \"report\") ||\n\t\t\tstrings.Contains(content, \"analysis\") ||\n\t\t\tstrings.Contains(content, \"data\") ||\n\t\t\tstrings.Contains(content, \"q4\")\n\n\t\tassert.True(t, hasRelevantContent, \"goals should relate to user request, got: %s\", exec.Goals.Content)\n\t})\n\n\tt.Run(\"includes robot identity for human trigger\", func(t *testing.T) {\n\t\trobot := &types.Robot{\n\t\t\tMemberID: \"test-robot-1\",\n\t\t\tTeamID: \"test-team-1\",\n\t\t\tDisplayName: \"Sales Analyst\",\n\t\t\tConfig: &types.Config{\n\t\t\t\tIdentity: &types.Identity{\n\t\t\t\t\tRole: \"Sales Analyst\",\n\t\t\t\t\tDuties: []string{\"Analyze sales data\", \"Generate reports\"},\n\t\t\t\t\tRules: []string{\"Focus on actionable insights\"},\n\t\t\t\t},\n\t\t\t\tResources: &types.Resources{\n\t\t\t\t\tPhases: map[types.Phase]string{\n\t\t\t\t\t\ttypes.PhaseGoals: \"robot.goals\",\n\t\t\t\t\t},\n\t\t\t\t},\n\t\t\t},\n\t\t}\n\t\texec := createGoalsTestExecution(robot, types.TriggerHuman)\n\t\texec.Input = &types.TriggerInput{\n\t\t\tAction: \"instruct\",\n\t\t\tMessages: []agentcontext.Message{\n\t\t\t\t{Role: agentcontext.RoleUser, Content: \"What should I focus on today?\"},\n\t\t\t},\n\t\t}\n\n\t\te := standard.New()\n\t\terr := e.RunGoals(ctx, exec, nil)\n\n\t\trequire.NoError(t, err)\n\t\tassert.NotEmpty(t, exec.Goals.Content)\n\t})\n}\n\nfunc TestRunGoalsEventTrigger(t *testing.T) {\n\tif testing.Short() {\n\t\tt.Skip(\"Skipping integration test\")\n\t}\n\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\tctx := types.NewContext(context.Background(), testAuth())\n\n\tt.Run(\"generates goals from event trigger\", func(t *testing.T) {\n\t\trobot := createGoalsTestRobot(t, \"robot.goals\")\n\t\texec := createGoalsTestExecution(robot, types.TriggerEvent)\n\n\t\t// Set event input\n\t\texec.Input = &types.TriggerInput{\n\t\t\tSource: \"webhook\",\n\t\t\tEventType: \"lead.created\",\n\t\t\tData: map[string]interface{}{\n\t\t\t\t\"lead_id\": \"lead-456\",\n\t\t\t\t\"company\": \"BigCorp Inc\",\n\t\t\t\t\"contact_name\": \"John Smith\",\n\t\t\t\t\"email\": \"john@bigcorp.com\",\n\t\t\t\t\"interest\": \"Enterprise plan\",\n\t\t\t},\n\t\t}\n\n\t\te := standard.New()\n\t\terr := e.RunGoals(ctx, exec, nil)\n\n\t\trequire.NoError(t, err)\n\t\trequire.NotNil(t, exec.Goals)\n\t\tassert.NotEmpty(t, exec.Goals.Content)\n\n\t\t// Goals should be related to the event\n\t\tcontent := strings.ToLower(exec.Goals.Content)\n\t\thasRelevantContent := strings.Contains(content, \"lead\") ||\n\t\t\tstrings.Contains(content, \"bigcorp\") ||\n\t\t\tstrings.Contains(content, \"contact\") ||\n\t\t\tstrings.Contains(content, \"follow\") ||\n\t\t\tstrings.Contains(content, \"qualify\")\n\n\t\tassert.True(t, hasRelevantContent, \"goals should relate to event, got: %s\", exec.Goals.Content)\n\t})\n}\n\nfunc TestRunGoalsErrorHandling(t *testing.T) {\n\tif testing.Short() {\n\t\tt.Skip(\"Skipping integration test\")\n\t}\n\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\tctx := types.NewContext(context.Background(), testAuth())\n\n\tt.Run(\"returns error when robot is nil\", func(t *testing.T) {\n\t\texec := &types.Execution{\n\t\t\tID: \"test-exec-1\",\n\t\t\tTriggerType: types.TriggerClock,\n\t\t}\n\t\t// Don't set robot\n\n\t\te := standard.New()\n\t\terr := e.RunGoals(ctx, exec, nil)\n\n\t\tassert.Error(t, err)\n\t\tassert.Contains(t, err.Error(), \"robot not found\")\n\t})\n\n\tt.Run(\"returns error when agent not found\", func(t *testing.T) {\n\t\trobot := &types.Robot{\n\t\t\tMemberID: \"test-robot-1\",\n\t\t\tTeamID: \"test-team-1\",\n\t\t\tConfig: &types.Config{\n\t\t\t\tIdentity: &types.Identity{Role: \"Test\"},\n\t\t\t\tResources: &types.Resources{\n\t\t\t\t\tPhases: map[types.Phase]string{\n\t\t\t\t\t\ttypes.PhaseGoals: \"non.existent.agent\",\n\t\t\t\t\t},\n\t\t\t\t},\n\t\t\t},\n\t\t}\n\t\texec := createGoalsTestExecution(robot, types.TriggerClock)\n\t\texec.Inspiration = &types.InspirationReport{\n\t\t\tContent: \"Test content\",\n\t\t}\n\n\t\te := standard.New()\n\t\terr := e.RunGoals(ctx, exec, nil)\n\n\t\tassert.Error(t, err)\n\t\tassert.Contains(t, err.Error(), \"call failed\")\n\t})\n\n\tt.Run(\"returns error when no input available and no identity\", func(t *testing.T) {\n\t\t// Robot without identity - should fail when no input is provided\n\t\trobot := &types.Robot{\n\t\t\tMemberID: \"test-robot-1\",\n\t\t\tTeamID: \"test-team-1\",\n\t\t\tConfig: &types.Config{\n\t\t\t\t// No Identity - so no fallback content\n\t\t\t\tResources: &types.Resources{\n\t\t\t\t\tPhases: map[types.Phase]string{\n\t\t\t\t\t\ttypes.PhaseGoals: \"robot.goals\",\n\t\t\t\t\t},\n\t\t\t\t},\n\t\t\t},\n\t\t}\n\t\texec := createGoalsTestExecution(robot, types.TriggerHuman)\n\t\texec.Input = nil // No input\n\n\t\te := standard.New()\n\t\terr := e.RunGoals(ctx, exec, nil)\n\n\t\tassert.Error(t, err)\n\t\tassert.Contains(t, err.Error(), \"no input available\")\n\t})\n}\n\nfunc TestRunGoalsFallbackBehavior(t *testing.T) {\n\tif testing.Short() {\n\t\tt.Skip(\"Skipping integration test\")\n\t}\n\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\tctx := types.NewContext(context.Background(), testAuth())\n\n\tt.Run(\"falls back to clock context when no inspiration report\", func(t *testing.T) {\n\t\trobot := createGoalsTestRobot(t, \"robot.goals\")\n\t\texec := createGoalsTestExecution(robot, types.TriggerClock)\n\t\texec.Inspiration = nil // No inspiration report\n\n\t\te := standard.New()\n\t\terr := e.RunGoals(ctx, exec, nil)\n\n\t\t// Should still work with fallback clock context\n\t\trequire.NoError(t, err)\n\t\trequire.NotNil(t, exec.Goals)\n\t\tassert.NotEmpty(t, exec.Goals.Content)\n\t})\n}\n\n// ============================================================================\n// Delivery Parsing Tests\n// ============================================================================\n\nfunc TestParseDeliveryFromGoalsResponse(t *testing.T) {\n\tif testing.Short() {\n\t\tt.Skip(\"Skipping integration test\")\n\t}\n\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\tctx := types.NewContext(context.Background(), testAuth())\n\n\tt.Run(\"parses delivery when agent returns it\", func(t *testing.T) {\n\t\trobot := createGoalsTestRobot(t, \"robot.goals\")\n\t\texec := createGoalsTestExecution(robot, types.TriggerHuman)\n\n\t\t// Request that explicitly asks for email delivery\n\t\texec.Input = &types.TriggerInput{\n\t\t\tAction: \"task.add\",\n\t\t\tMessages: []agentcontext.Message{\n\t\t\t\t{Role: agentcontext.RoleUser, Content: \"Prepare a sales report and send it to team@example.com via email\"},\n\t\t\t},\n\t\t}\n\n\t\te := standard.New()\n\t\terr := e.RunGoals(ctx, exec, nil)\n\n\t\trequire.NoError(t, err)\n\t\trequire.NotNil(t, exec.Goals)\n\t\tassert.NotEmpty(t, exec.Goals.Content)\n\n\t\t// Delivery may or may not be present depending on LLM response\n\t\t// If present, verify structure\n\t\tif exec.Goals.Delivery != nil {\n\t\t\t// Type should be valid if present\n\t\t\tif exec.Goals.Delivery.Type != \"\" {\n\t\t\t\tvalidTypes := []types.DeliveryType{\n\t\t\t\t\ttypes.DeliveryEmail, types.DeliveryWebhook,\n\t\t\t\t\ttypes.DeliveryProcess, types.DeliveryNotify,\n\t\t\t\t}\n\t\t\t\tfound := false\n\t\t\t\tfor _, vt := range validTypes {\n\t\t\t\t\tif exec.Goals.Delivery.Type == vt {\n\t\t\t\t\t\tfound = true\n\t\t\t\t\t\tbreak\n\t\t\t\t\t}\n\t\t\t\t}\n\t\t\t\t// Note: LLM might return non-standard types, we accept them but log\n\t\t\t\tt.Logf(\"Delivery type: %s (valid: %v)\", exec.Goals.Delivery.Type, found)\n\t\t\t}\n\t\t}\n\t})\n}\n\nfunc TestDeliveryTypeValidation(t *testing.T) {\n\tt.Run(\"valid delivery types\", func(t *testing.T) {\n\t\tvalidTypes := []types.DeliveryType{\n\t\t\ttypes.DeliveryEmail,\n\t\t\ttypes.DeliveryWebhook,\n\t\t\ttypes.DeliveryProcess,\n\t\t\ttypes.DeliveryNotify,\n\t\t}\n\n\t\tfor _, dt := range validTypes {\n\t\t\tassert.True(t, standard.IsValidDeliveryType(dt), \"should be valid: %s\", dt)\n\t\t}\n\t})\n\n\tt.Run(\"invalid delivery types\", func(t *testing.T) {\n\t\tinvalidTypes := []types.DeliveryType{\n\t\t\t\"invalid\",\n\t\t\t\"sms\",\n\t\t\t\"\",\n\t\t}\n\n\t\tfor _, dt := range invalidTypes {\n\t\t\tassert.False(t, standard.IsValidDeliveryType(dt), \"should be invalid: %s\", dt)\n\t\t}\n\t})\n}\n\nfunc TestParseDelivery(t *testing.T) {\n\tt.Run(\"parses valid delivery with all fields\", func(t *testing.T) {\n\t\tdata := map[string]interface{}{\n\t\t\t\"type\": \"email\",\n\t\t\t\"recipients\": []interface{}{\"user@example.com\", \"team@example.com\"},\n\t\t\t\"format\": \"markdown\",\n\t\t\t\"template\": \"weekly-report\",\n\t\t\t\"options\": map[string]interface{}{\n\t\t\t\t\"subject\": \"Weekly Report\",\n\t\t\t},\n\t\t}\n\n\t\tresult := standard.ParseDelivery(data)\n\n\t\trequire.NotNil(t, result)\n\t\tassert.Equal(t, types.DeliveryEmail, result.Type)\n\t\tassert.Equal(t, []string{\"user@example.com\", \"team@example.com\"}, result.Recipients)\n\t\tassert.Equal(t, \"markdown\", result.Format)\n\t\tassert.Equal(t, \"weekly-report\", result.Template)\n\t\tassert.Equal(t, \"Weekly Report\", result.Options[\"subject\"])\n\t})\n\n\tt.Run(\"returns nil for nil data\", func(t *testing.T) {\n\t\tresult := standard.ParseDelivery(nil)\n\t\tassert.Nil(t, result)\n\t})\n\n\tt.Run(\"returns nil for missing type\", func(t *testing.T) {\n\t\tdata := map[string]interface{}{\n\t\t\t\"recipients\": []interface{}{\"user@example.com\"},\n\t\t}\n\n\t\tresult := standard.ParseDelivery(data)\n\t\tassert.Nil(t, result)\n\t})\n\n\tt.Run(\"returns nil for empty type\", func(t *testing.T) {\n\t\tdata := map[string]interface{}{\n\t\t\t\"type\": \"\",\n\t\t\t\"recipients\": []interface{}{\"user@example.com\"},\n\t\t}\n\n\t\tresult := standard.ParseDelivery(data)\n\t\tassert.Nil(t, result)\n\t})\n\n\tt.Run(\"returns nil for invalid type\", func(t *testing.T) {\n\t\tdata := map[string]interface{}{\n\t\t\t\"type\": \"sms\",\n\t\t\t\"recipients\": []interface{}{\"user@example.com\"},\n\t\t}\n\n\t\tresult := standard.ParseDelivery(data)\n\t\tassert.Nil(t, result)\n\t})\n\n\tt.Run(\"handles missing optional fields\", func(t *testing.T) {\n\t\tdata := map[string]interface{}{\n\t\t\t\"type\": \"webhook\",\n\t\t}\n\n\t\tresult := standard.ParseDelivery(data)\n\n\t\trequire.NotNil(t, result)\n\t\tassert.Equal(t, types.DeliveryWebhook, result.Type)\n\t\tassert.Empty(t, result.Recipients)\n\t\tassert.Empty(t, result.Format)\n\t\tassert.Empty(t, result.Template)\n\t\tassert.Nil(t, result.Options)\n\t})\n\n\tt.Run(\"handles non-string recipients gracefully\", func(t *testing.T) {\n\t\tdata := map[string]interface{}{\n\t\t\t\"type\": \"email\",\n\t\t\t\"recipients\": []interface{}{\"valid@example.com\", 123, nil, \"another@example.com\"},\n\t\t}\n\n\t\tresult := standard.ParseDelivery(data)\n\n\t\trequire.NotNil(t, result)\n\t\t// Only string recipients should be included\n\t\tassert.Equal(t, []string{\"valid@example.com\", \"another@example.com\"}, result.Recipients)\n\t})\n\n\tt.Run(\"parses all valid delivery types\", func(t *testing.T) {\n\t\tvalidTypes := []string{\"email\", \"webhook\", \"process\", \"notify\"}\n\n\t\tfor _, dt := range validTypes {\n\t\t\tdata := map[string]interface{}{\n\t\t\t\t\"type\": dt,\n\t\t\t}\n\n\t\t\tresult := standard.ParseDelivery(data)\n\t\t\trequire.NotNil(t, result, \"should parse type: %s\", dt)\n\t\t\tassert.Equal(t, types.DeliveryType(dt), result.Type)\n\t\t}\n\t})\n}\n\n// ============================================================================\n// InputFormatter Tests for P1\n// ============================================================================\n\nfunc TestInputFormatterFormatRobotIdentity(t *testing.T) {\n\tformatter := standard.NewInputFormatter()\n\n\tt.Run(\"formats robot identity correctly\", func(t *testing.T) {\n\t\trobot := &types.Robot{\n\t\t\tMemberID: \"test-robot\",\n\t\t\tConfig: &types.Config{\n\t\t\t\tIdentity: &types.Identity{\n\t\t\t\t\tRole: \"Sales Analyst\",\n\t\t\t\t\tDuties: []string{\"Analyze sales data\", \"Generate reports\"},\n\t\t\t\t\tRules: []string{\"Be accurate\", \"Be concise\"},\n\t\t\t\t},\n\t\t\t},\n\t\t}\n\n\t\tcontent := formatter.FormatRobotIdentity(robot)\n\n\t\tassert.Contains(t, content, \"## Robot Identity\")\n\t\tassert.Contains(t, content, \"Sales Analyst\")\n\t\tassert.Contains(t, content, \"Analyze sales data\")\n\t\tassert.Contains(t, content, \"Generate reports\")\n\t\tassert.Contains(t, content, \"Be accurate\")\n\t\tassert.Contains(t, content, \"Be concise\")\n\t})\n\n\tt.Run(\"returns empty for nil robot\", func(t *testing.T) {\n\t\tcontent := formatter.FormatRobotIdentity(nil)\n\t\tassert.Empty(t, content)\n\t})\n\n\tt.Run(\"returns empty for robot without config\", func(t *testing.T) {\n\t\trobot := &types.Robot{MemberID: \"test\"}\n\t\tcontent := formatter.FormatRobotIdentity(robot)\n\t\tassert.Empty(t, content)\n\t})\n\n\tt.Run(\"returns empty for robot without identity\", func(t *testing.T) {\n\t\trobot := &types.Robot{\n\t\t\tMemberID: \"test\",\n\t\t\tConfig: &types.Config{},\n\t\t}\n\t\tcontent := formatter.FormatRobotIdentity(robot)\n\t\tassert.Empty(t, content)\n\t})\n\n\tt.Run(\"handles identity with only role\", func(t *testing.T) {\n\t\trobot := &types.Robot{\n\t\t\tMemberID: \"test\",\n\t\t\tConfig: &types.Config{\n\t\t\t\tIdentity: &types.Identity{\n\t\t\t\t\tRole: \"Simple Bot\",\n\t\t\t\t},\n\t\t\t},\n\t\t}\n\n\t\tcontent := formatter.FormatRobotIdentity(robot)\n\n\t\tassert.Contains(t, content, \"## Robot Identity\")\n\t\tassert.Contains(t, content, \"Simple Bot\")\n\t\tassert.NotContains(t, content, \"Duties\")\n\t\tassert.NotContains(t, content, \"Rules\")\n\t})\n}\n\n// ============================================================================\n// Helper Functions\n// ============================================================================\n\n// createGoalsTestRobot creates a test robot with specified goals agent\n// Includes available expert agents so the Goals Agent knows what resources are available\n//\n// Note: The agent IDs listed in Resources.Agents must exist in yao-dev-app/assistants/experts/\n// Current available experts: data-analyst, summarizer, text-writer, web-reader\nfunc createGoalsTestRobot(t *testing.T, agentID string) *types.Robot {\n\tt.Helper()\n\treturn &types.Robot{\n\t\tMemberID: \"test-robot-1\",\n\t\tTeamID: \"test-team-1\",\n\t\tDisplayName: \"Test Robot\",\n\t\tConfig: &types.Config{\n\t\t\tIdentity: &types.Identity{\n\t\t\t\tRole: \"Test Assistant\",\n\t\t\t\tDuties: []string{\"Testing\", \"Validation\", \"Data Analysis\", \"Report Generation\"},\n\t\t\t},\n\t\t\tResources: &types.Resources{\n\t\t\t\tPhases: map[types.Phase]string{\n\t\t\t\t\ttypes.PhaseGoals: agentID,\n\t\t\t\t},\n\t\t\t\t// Available expert agents that can be delegated to\n\t\t\t\t// These IDs correspond to assistants in yao-dev-app/assistants/experts/\n\t\t\t\tAgents: []string{\n\t\t\t\t\t\"experts.data-analyst\", // Data analysis and insights\n\t\t\t\t\t\"experts.summarizer\", // Content summarization\n\t\t\t\t\t\"experts.text-writer\", // Report and document generation\n\t\t\t\t\t\"experts.web-reader\", // Web content extraction\n\t\t\t\t},\n\t\t\t},\n\t\t\t// Knowledge base collections (if any)\n\t\t\tKB: &types.KB{\n\t\t\t\tCollections: []string{\"test-knowledge\"},\n\t\t\t},\n\t\t},\n\t}\n}\n\n// createGoalsTestExecution creates a test execution for goals phase\nfunc createGoalsTestExecution(robot *types.Robot, trigger types.TriggerType) *types.Execution {\n\texec := &types.Execution{\n\t\tID: \"test-exec-goals-1\",\n\t\tMemberID: robot.MemberID,\n\t\tTeamID: robot.TeamID,\n\t\tTriggerType: trigger,\n\t\tStartTime: time.Now(),\n\t\tStatus: types.ExecRunning,\n\t\tPhase: types.PhaseGoals,\n\t}\n\texec.SetRobot(robot)\n\treturn exec\n}\n" }, { "path": "agent/robot/executor/standard/host.go", "content": "package standard\n\nimport (\n\t\"encoding/json\"\n\t\"fmt\"\n\n\tkunlog \"github.com/yaoapp/kun/log\"\n\trobottypes \"github.com/yaoapp/yao/agent/robot/types\"\n)\n\n// CallHostAgent calls the Host Agent with structured input and parses structured output.\n// The Host Agent mediates all human-robot interactions through three scenarios:\n// - \"assign\": new task assignment with multi-round confirmation\n// - \"guide\": guidance during execution\n// - \"clarify\": answering questions from waiting tasks\nfunc (e *Executor) CallHostAgent(ctx *robottypes.Context, robot *robottypes.Robot, input *robottypes.HostInput, chatID string) (*robottypes.HostOutput, error) {\n\tif robot == nil {\n\t\treturn nil, fmt.Errorf(\"robot cannot be nil\")\n\t}\n\n\tagentID := \"\"\n\tif robot.Config != nil && robot.Config.Resources != nil {\n\t\tagentID = robot.Config.Resources.GetPhaseAgent(robottypes.PhaseHost)\n\t}\n\tif agentID == \"\" {\n\t\treturn nil, fmt.Errorf(\"no Host Agent configured for robot %s\", robot.MemberID)\n\t}\n\n\tinputJSON, err := json.Marshal(input)\n\tif err != nil {\n\t\treturn nil, fmt.Errorf(\"failed to marshal host input: %w\", err)\n\t}\n\n\tkunlog.Info(\"calling Host Agent %s for scenario=%s chatID=%s\", agentID, input.Scenario, chatID)\n\n\tcaller := NewConversationCaller(chatID)\n\tresult, err := caller.CallWithMessages(ctx, agentID, string(inputJSON))\n\tif err != nil {\n\t\treturn nil, fmt.Errorf(\"host agent (%s) call failed: %w\", agentID, err)\n\t}\n\n\tdata, err := result.GetJSON()\n\tif err != nil {\n\t\ttext := result.GetText()\n\t\tkunlog.Warn(\"Host Agent returned non-JSON response, treating as confirm: %s\", text)\n\t\treturn &robottypes.HostOutput{\n\t\t\tReply: text,\n\t\t\tAction: robottypes.HostActionConfirm,\n\t\t}, nil\n\t}\n\n\toutput := &robottypes.HostOutput{}\n\traw, _ := json.Marshal(data)\n\tif err := json.Unmarshal(raw, output); err != nil {\n\t\treturn &robottypes.HostOutput{\n\t\t\tReply: result.GetText(),\n\t\t\tAction: robottypes.HostActionConfirm,\n\t\t}, nil\n\t}\n\n\treturn output, nil\n}\n" }, { "path": "agent/robot/executor/standard/host_test.go", "content": "package standard_test\n\nimport (\n\t\"context\"\n\t\"testing\"\n\n\t\"github.com/stretchr/testify/assert\"\n\t\"github.com/stretchr/testify/require\"\n\t\"github.com/yaoapp/yao/agent/robot/executor/standard\"\n\trobottypes \"github.com/yaoapp/yao/agent/robot/types\"\n\t\"github.com/yaoapp/yao/agent/testutils\"\n\toauthtypes \"github.com/yaoapp/yao/openapi/oauth/types\"\n)\n\nfunc hostTestAuth() *oauthtypes.AuthorizedInfo {\n\treturn &oauthtypes.AuthorizedInfo{\n\t\tUserID: \"test-user-host\",\n\t\tTeamID: \"test-team-host\",\n\t}\n}\n\n// H1: nil robot\nfunc TestCallHostAgent_NilRobot(t *testing.T) {\n\te := standard.New()\n\tctx := robottypes.NewContext(context.Background(), nil)\n\n\t_, err := e.CallHostAgent(ctx, nil, &robottypes.HostInput{Scenario: \"assign\"}, \"chat-1\")\n\trequire.Error(t, err)\n\tassert.Contains(t, err.Error(), \"robot cannot be nil\")\n}\n\n// H2: no Host Agent configured\nfunc TestCallHostAgent_NoHostAgent(t *testing.T) {\n\te := standard.New()\n\tctx := robottypes.NewContext(context.Background(), nil)\n\n\tt.Run(\"nil config\", func(t *testing.T) {\n\t\trobot := &robottypes.Robot{MemberID: \"member-h2a\"}\n\t\t_, err := e.CallHostAgent(ctx, robot, &robottypes.HostInput{Scenario: \"assign\"}, \"chat-1\")\n\t\trequire.Error(t, err)\n\t\tassert.Contains(t, err.Error(), \"no Host Agent configured\")\n\t})\n\n\tt.Run(\"nil resources\", func(t *testing.T) {\n\t\trobot := &robottypes.Robot{\n\t\t\tMemberID: \"member-h2b\",\n\t\t\tConfig: &robottypes.Config{},\n\t\t}\n\t\t_, err := e.CallHostAgent(ctx, robot, &robottypes.HostInput{Scenario: \"assign\"}, \"chat-1\")\n\t\trequire.Error(t, err)\n\t\tassert.Contains(t, err.Error(), \"no Host Agent configured\")\n\t})\n}\n\n// H3: valid JSON response from Host Agent\nfunc TestCallHostAgent_ValidJSONResponse(t *testing.T) {\n\tif testing.Short() {\n\t\tt.Skip(\"Requires assistant framework and LLM\")\n\t}\n\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\te := standard.New()\n\tctx := robottypes.NewContext(context.Background(), hostTestAuth())\n\n\trobot := &robottypes.Robot{\n\t\tMemberID: \"member-h3\",\n\t\tConfig: &robottypes.Config{\n\t\t\tResources: &robottypes.Resources{\n\t\t\t\tPhases: map[robottypes.Phase]string{\n\t\t\t\t\trobottypes.PhaseHost: \"tests.host-json\",\n\t\t\t\t},\n\t\t\t},\n\t\t},\n\t}\n\n\tinput := &robottypes.HostInput{\n\t\tScenario: \"assign\",\n\t\tContext: &robottypes.HostContext{\n\t\t\tRobotStatus: &robottypes.RobotStatusSnapshot{ActiveCount: 0, MaxQuota: 10},\n\t\t},\n\t}\n\n\toutput, err := e.CallHostAgent(ctx, robot, input, \"chat-h3\")\n\trequire.NoError(t, err, \"CallHostAgent should not error for valid JSON host agent\")\n\trequire.NotNil(t, output, \"output should not be nil\")\n\n\tassert.NotEmpty(t, output.Reply, \"reply should not be empty\")\n\tassert.Equal(t, robottypes.HostActionConfirm, output.Action,\n\t\t\"action should be 'confirm' for the JSON host agent\")\n\tassert.False(t, output.WaitForMore, \"wait_for_more should be false\")\n}\n\n// H4: plain text response (non-JSON fallback)\nfunc TestCallHostAgent_PlaintextFallback(t *testing.T) {\n\tif testing.Short() {\n\t\tt.Skip(\"Requires assistant framework and LLM\")\n\t}\n\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\te := standard.New()\n\tctx := robottypes.NewContext(context.Background(), hostTestAuth())\n\n\trobot := &robottypes.Robot{\n\t\tMemberID: \"member-h4\",\n\t\tConfig: &robottypes.Config{\n\t\t\tResources: &robottypes.Resources{\n\t\t\t\tPhases: map[robottypes.Phase]string{\n\t\t\t\t\trobottypes.PhaseHost: \"tests.host-plaintext\",\n\t\t\t\t},\n\t\t\t},\n\t\t},\n\t}\n\n\tinput := &robottypes.HostInput{\n\t\tScenario: \"assign\",\n\t\tContext: &robottypes.HostContext{\n\t\t\tRobotStatus: &robottypes.RobotStatusSnapshot{ActiveCount: 0, MaxQuota: 10},\n\t\t},\n\t}\n\n\toutput, err := e.CallHostAgent(ctx, robot, input, \"chat-h4\")\n\trequire.NoError(t, err, \"non-JSON response should fallback gracefully, not error\")\n\trequire.NotNil(t, output, \"output should not be nil\")\n\n\tassert.NotEmpty(t, output.Reply, \"reply should contain the plaintext response\")\n\tassert.Equal(t, robottypes.HostActionConfirm, output.Action,\n\t\t\"action should fallback to 'confirm' for non-JSON response\")\n}\n\n// H5: JSON with wrong structure (no action/reply fields)\nfunc TestCallHostAgent_BadJSONStructureFallback(t *testing.T) {\n\tif testing.Short() {\n\t\tt.Skip(\"Requires assistant framework and LLM\")\n\t}\n\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\te := standard.New()\n\tctx := robottypes.NewContext(context.Background(), hostTestAuth())\n\n\trobot := &robottypes.Robot{\n\t\tMemberID: \"member-h5\",\n\t\tConfig: &robottypes.Config{\n\t\t\tResources: &robottypes.Resources{\n\t\t\t\tPhases: map[robottypes.Phase]string{\n\t\t\t\t\trobottypes.PhaseHost: \"tests.host-badjson\",\n\t\t\t\t},\n\t\t\t},\n\t\t},\n\t}\n\n\tinput := &robottypes.HostInput{\n\t\tScenario: \"assign\",\n\t\tContext: &robottypes.HostContext{\n\t\t\tRobotStatus: &robottypes.RobotStatusSnapshot{ActiveCount: 0, MaxQuota: 10},\n\t\t},\n\t}\n\n\toutput, err := e.CallHostAgent(ctx, robot, input, \"chat-h5\")\n\trequire.NoError(t, err, \"bad JSON structure should not error\")\n\trequire.NotNil(t, output, \"output should not be nil\")\n\n\t// The JSON is valid but has no action/reply fields.\n\t// json.Unmarshal won't error — Action will be zero value (\"\").\n\t// Verify the output is returned (either with empty action or fallback to confirm).\n\tif output.Action == \"\" {\n\t\tassert.Empty(t, output.Action,\n\t\t\t\"action should be empty when JSON has no action field\")\n\t} else {\n\t\tassert.Equal(t, robottypes.HostActionConfirm, output.Action,\n\t\t\t\"action should be 'confirm' if fallback is triggered\")\n\t}\n}\n\n// H6: assistant not found\nfunc TestCallHostAgent_AssistantNotFound(t *testing.T) {\n\tif testing.Short() {\n\t\tt.Skip(\"Requires assistant framework initialization\")\n\t}\n\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\te := standard.New()\n\tctx := robottypes.NewContext(context.Background(), hostTestAuth())\n\n\trobot := &robottypes.Robot{\n\t\tMemberID: \"member-h6\",\n\t\tConfig: &robottypes.Config{\n\t\t\tResources: &robottypes.Resources{\n\t\t\t\tPhases: map[robottypes.Phase]string{\n\t\t\t\t\trobottypes.PhaseHost: \"nonexistent-assistant\",\n\t\t\t\t},\n\t\t\t},\n\t\t},\n\t}\n\n\tinput := &robottypes.HostInput{Scenario: \"assign\"}\n\n\t_, err := e.CallHostAgent(ctx, robot, input, \"chat-h6\")\n\trequire.Error(t, err)\n\tassert.Contains(t, err.Error(), \"host agent\")\n}\n\n// H7: input marshalling verification (pure unit test, no LLM needed)\nfunc TestCallHostAgent_InputMarshalling(t *testing.T) {\n\tinput := &robottypes.HostInput{\n\t\tScenario: \"clarify\",\n\t\tContext: &robottypes.HostContext{\n\t\t\tRobotStatus: &robottypes.RobotStatusSnapshot{\n\t\t\t\tActiveCount: 2,\n\t\t\t\tMaxQuota: 5,\n\t\t\t},\n\t\t\tAgentReply: \"What format?\",\n\t\t},\n\t}\n\n\tassert.NotEmpty(t, input.Scenario)\n\tassert.NotNil(t, input.Context)\n\tassert.Equal(t, 2, input.Context.RobotStatus.ActiveCount)\n}\n" }, { "path": "agent/robot/executor/standard/input.go", "content": "package standard\n\nimport (\n\t\"context\"\n\t\"encoding/json\"\n\t\"fmt\"\n\t\"strings\"\n\n\t\"github.com/yaoapp/gou/mcp\"\n\t\"github.com/yaoapp/yao/agent/assistant\"\n\tagentcontext \"github.com/yaoapp/yao/agent/context\"\n\t\"github.com/yaoapp/yao/agent/i18n\"\n\trobottypes \"github.com/yaoapp/yao/agent/robot/types\"\n)\n\n// InputFormatter provides methods to format input data for assistant prompts\n// Each phase has specific input requirements:\n// - P0 (Inspiration): ClockContext + Robot identity + Available resources\n// - P1 (Goals): InspirationReport/TriggerInput + Robot identity + Available resources\n// - P2 (Tasks): Goals + Available resources\n// - P3 (Run): Tasks\n// - P4 (Delivery): Task results\n// - P5 (Learning): Execution summary\ntype InputFormatter struct{}\n\n// NewInputFormatter creates a new InputFormatter\nfunc NewInputFormatter() *InputFormatter {\n\treturn &InputFormatter{}\n}\n\n// FormatClockContext formats ClockContext as user message content\n// Used by P0 (Inspiration) phase\nfunc (f *InputFormatter) FormatClockContext(clock *robottypes.ClockContext, robot *robottypes.Robot) string {\n\tif clock == nil {\n\t\treturn \"\"\n\t}\n\n\tvar sb strings.Builder\n\n\t// Time context section\n\tsb.WriteString(\"## Current Time Context\\n\\n\")\n\tsb.WriteString(fmt.Sprintf(\"- **Now**: %s\\n\", clock.Now.Format(\"2006-01-02 15:04:05\")))\n\tsb.WriteString(fmt.Sprintf(\"- **Day**: %s\\n\", clock.DayOfWeek))\n\tsb.WriteString(fmt.Sprintf(\"- **Date**: %d/%d/%d\\n\", clock.Year, clock.Month, clock.DayOfMonth))\n\tsb.WriteString(fmt.Sprintf(\"- **Week**: %d of year\\n\", clock.WeekOfYear))\n\tsb.WriteString(fmt.Sprintf(\"- **Hour**: %d\\n\", clock.Hour))\n\tsb.WriteString(fmt.Sprintf(\"- **Timezone**: %s\\n\", clock.TZ))\n\n\t// Time markers - show all markers with check/cross for context awareness\n\tsb.WriteString(\"\\n### Time Markers\\n\")\n\tsb.WriteString(fmt.Sprintf(\"- %s Weekend\\n\", boolMark(clock.IsWeekend)))\n\tsb.WriteString(fmt.Sprintf(\"- %s Month Start (1st-3rd)\\n\", boolMark(clock.IsMonthStart)))\n\tsb.WriteString(fmt.Sprintf(\"- %s Month End (last 3 days)\\n\", boolMark(clock.IsMonthEnd)))\n\tsb.WriteString(fmt.Sprintf(\"- %s Quarter End\\n\", boolMark(clock.IsQuarterEnd)))\n\tsb.WriteString(fmt.Sprintf(\"- %s Year End\\n\", boolMark(clock.IsYearEnd)))\n\n\t// Robot identity section\n\t// Priority: Config.Identity > Robot fields (DisplayName, Bio, SystemPrompt)\n\tif robot != nil {\n\t\tif robot.Config != nil && robot.Config.Identity != nil {\n\t\t\t// Use structured identity from config\n\t\t\tsb.WriteString(\"\\n## Robot Identity\\n\\n\")\n\t\t\tsb.WriteString(fmt.Sprintf(\"- **Role**: %s\\n\", robot.Config.Identity.Role))\n\t\t\tif len(robot.Config.Identity.Duties) > 0 {\n\t\t\t\tsb.WriteString(\"- **Duties**:\\n\")\n\t\t\t\tfor _, duty := range robot.Config.Identity.Duties {\n\t\t\t\t\tsb.WriteString(fmt.Sprintf(\" - %s\\n\", duty))\n\t\t\t\t}\n\t\t\t}\n\t\t\tif len(robot.Config.Identity.Rules) > 0 {\n\t\t\t\tsb.WriteString(\"- **Rules**:\\n\")\n\t\t\t\tfor _, rule := range robot.Config.Identity.Rules {\n\t\t\t\t\tsb.WriteString(fmt.Sprintf(\" - %s\\n\", rule))\n\t\t\t\t}\n\t\t\t}\n\t\t} else if robot.DisplayName != \"\" || robot.Bio != \"\" || robot.SystemPrompt != \"\" {\n\t\t\t// Fallback: build identity from Robot fields (from __yao.member table)\n\t\t\tsb.WriteString(\"\\n## Robot Identity\\n\\n\")\n\t\t\tif robot.DisplayName != \"\" {\n\t\t\t\tsb.WriteString(fmt.Sprintf(\"- **Role**: %s\\n\", robot.DisplayName))\n\t\t\t}\n\t\t\tif robot.Bio != \"\" {\n\t\t\t\tsb.WriteString(fmt.Sprintf(\"- **Description**: %s\\n\", robot.Bio))\n\t\t\t}\n\t\t\tif robot.SystemPrompt != \"\" {\n\t\t\t\tsb.WriteString(fmt.Sprintf(\"- **Instructions**:\\n%s\\n\", robot.SystemPrompt))\n\t\t\t}\n\t\t}\n\t}\n\n\treturn sb.String()\n}\n\n// boolMark returns ✓ for true and ✗ for false\nfunc boolMark(v bool) string {\n\tif v {\n\t\treturn \"✓\"\n\t}\n\treturn \"✗\"\n}\n\n// FormatRobotIdentity formats robot identity as user message content\n// Used to provide context about the robot's role and duties\nfunc (f *InputFormatter) FormatRobotIdentity(robot *robottypes.Robot) string {\n\tif robot == nil || robot.Config == nil || robot.Config.Identity == nil {\n\t\treturn \"\"\n\t}\n\n\tvar sb strings.Builder\n\tidentity := robot.Config.Identity\n\n\tsb.WriteString(\"## Robot Identity\\n\\n\")\n\tsb.WriteString(fmt.Sprintf(\"- **Role**: %s\\n\", identity.Role))\n\n\tif len(identity.Duties) > 0 {\n\t\tsb.WriteString(\"- **Duties**:\\n\")\n\t\tfor _, duty := range identity.Duties {\n\t\t\tsb.WriteString(fmt.Sprintf(\" - %s\\n\", duty))\n\t\t}\n\t}\n\n\tif len(identity.Rules) > 0 {\n\t\tsb.WriteString(\"- **Rules**:\\n\")\n\t\tfor _, rule := range identity.Rules {\n\t\t\tsb.WriteString(fmt.Sprintf(\" - %s\\n\", rule))\n\t\t}\n\t}\n\n\treturn sb.String()\n}\n\n// FormatAvailableResources formats available resources (agents, MCP tools, KB, DB) as user message content\n// Used by P0 (Inspiration) and P1 (Goals) to inform the agent what tools are available\n// This is critical for generating achievable goals - without knowing available tools,\n// the agent might generate goals that cannot be accomplished\nfunc (f *InputFormatter) FormatAvailableResources(robot *robottypes.Robot) string {\n\tlocale := \"en\" // default locale\n\tif robot != nil && robot.Config != nil {\n\t\tlocale = robot.Config.GetDefaultLocale()\n\t}\n\treturn f.FormatAvailableResourcesWithLocale(robot, locale)\n}\n\n// FormatAvailableResourcesWithLocale formats available resources with specific locale for i18n support\nfunc (f *InputFormatter) FormatAvailableResourcesWithLocale(robot *robottypes.Robot, locale string) string {\n\tif robot == nil || robot.Config == nil {\n\t\treturn \"\"\n\t}\n\n\tvar sb strings.Builder\n\thasContent := false\n\n\t// Available Agents - with detailed information (name, description)\n\tif robot.Config.Resources != nil && len(robot.Config.Resources.Agents) > 0 {\n\t\tif !hasContent {\n\t\t\tsb.WriteString(\"## Available Resources\\n\\n\")\n\t\t\thasContent = true\n\t\t}\n\t\tsb.WriteString(\"### Agents\\n\")\n\t\tsb.WriteString(\"These are the AI assistants you can delegate tasks to:\\n\\n\")\n\t\tfor _, agentID := range robot.Config.Resources.Agents {\n\t\t\t// Try to get agent details\n\t\t\tast, err := assistant.Get(agentID)\n\t\t\tif err != nil {\n\t\t\t\t// Fallback to just ID if agent not found\n\t\t\t\tsb.WriteString(fmt.Sprintf(\"- **%s**\\n\", agentID))\n\t\t\t\tcontinue\n\t\t\t}\n\n\t\t\t// Get localized name and description\n\t\t\tname := i18n.Translate(agentID, locale, ast.Name).(string)\n\t\t\tdescription := \"\"\n\t\t\tif ast.Description != \"\" {\n\t\t\t\tdescription = i18n.Translate(agentID, locale, ast.Description).(string)\n\t\t\t}\n\n\t\t\t// Format agent info\n\t\t\tsb.WriteString(fmt.Sprintf(\"- **%s** (`%s`)\\n\", name, agentID))\n\t\t\tif description != \"\" {\n\t\t\t\tsb.WriteString(fmt.Sprintf(\" - %s\\n\", description))\n\t\t\t}\n\t\t\tif ast.Capabilities != \"\" {\n\t\t\t\tcapabilities := i18n.Translate(agentID, locale, ast.Capabilities).(string)\n\t\t\t\tsb.WriteString(fmt.Sprintf(\" - **Capabilities**: %s\\n\", capabilities))\n\t\t\t}\n\t\t}\n\t\tsb.WriteString(\"\\n\")\n\t}\n\n\t// Available MCP Tools - with detailed tool information\n\tif robot.Config.Resources != nil && len(robot.Config.Resources.MCP) > 0 {\n\t\tif !hasContent {\n\t\t\tsb.WriteString(\"## Available Resources\\n\\n\")\n\t\t\thasContent = true\n\t\t}\n\t\tsb.WriteString(\"### MCP Tools\\n\")\n\t\tsb.WriteString(\"These are the external tools and services you can use:\\n\\n\")\n\t\tfor _, mcpConfig := range robot.Config.Resources.MCP {\n\t\t\t// Try to get MCP client and list tools\n\t\t\tclient, err := mcp.Select(mcpConfig.ID)\n\t\t\tif err != nil {\n\t\t\t\t// Fallback to basic info if client not found\n\t\t\t\tif len(mcpConfig.Tools) > 0 {\n\t\t\t\t\tsb.WriteString(fmt.Sprintf(\"- **%s**: %s\\n\", mcpConfig.ID, strings.Join(mcpConfig.Tools, \", \")))\n\t\t\t\t} else {\n\t\t\t\t\tsb.WriteString(fmt.Sprintf(\"- **%s**: all tools available\\n\", mcpConfig.ID))\n\t\t\t\t}\n\t\t\t\tcontinue\n\t\t\t}\n\n\t\t\t// Get client info for name and description\n\t\t\tclientInfo := client.Info()\n\t\t\tclientName := mcpConfig.ID\n\t\t\tclientDesc := \"\"\n\t\t\tif clientInfo != nil {\n\t\t\t\tif clientInfo.Name != \"\" {\n\t\t\t\t\tclientName = clientInfo.Name\n\t\t\t\t}\n\t\t\t\tif clientInfo.Description != \"\" {\n\t\t\t\t\tclientDesc = clientInfo.Description\n\t\t\t\t}\n\t\t\t}\n\n\t\t\t// Write MCP header\n\t\t\tif clientDesc != \"\" {\n\t\t\t\tsb.WriteString(fmt.Sprintf(\"#### %s (`%s`)\\n\", clientName, mcpConfig.ID))\n\t\t\t\tsb.WriteString(fmt.Sprintf(\"%s\\n\\n\", clientDesc))\n\t\t\t} else {\n\t\t\t\tsb.WriteString(fmt.Sprintf(\"#### %s (`%s`)\\n\\n\", clientName, mcpConfig.ID))\n\t\t\t}\n\n\t\t\t// Try to list tools with context\n\t\t\tctx := context.Background()\n\t\t\ttoolsResp, err := client.ListTools(ctx, \"\")\n\t\t\tif err != nil || toolsResp == nil {\n\t\t\t\t// Fallback to configured tools\n\t\t\t\tif len(mcpConfig.Tools) > 0 {\n\t\t\t\t\tsb.WriteString(\"Available tools: \")\n\t\t\t\t\tsb.WriteString(strings.Join(mcpConfig.Tools, \", \"))\n\t\t\t\t\tsb.WriteString(\"\\n\\n\")\n\t\t\t\t} else {\n\t\t\t\t\tsb.WriteString(\"All tools available\\n\\n\")\n\t\t\t\t}\n\t\t\t\tcontinue\n\t\t\t}\n\n\t\t\t// Filter tools if specific tools are configured\n\t\t\ttoolsToShow := toolsResp.Tools\n\t\t\tif len(mcpConfig.Tools) > 0 {\n\t\t\t\t// Create a map for quick lookup\n\t\t\t\tallowedTools := make(map[string]bool)\n\t\t\t\tfor _, t := range mcpConfig.Tools {\n\t\t\t\t\tallowedTools[t] = true\n\t\t\t\t}\n\t\t\t\t// Filter tools\n\t\t\t\tvar filteredTools []struct {\n\t\t\t\t\tName string\n\t\t\t\t\tDescription string\n\t\t\t\t}\n\t\t\t\tfor _, tool := range toolsResp.Tools {\n\t\t\t\t\tif allowedTools[tool.Name] {\n\t\t\t\t\t\tfilteredTools = append(filteredTools, struct {\n\t\t\t\t\t\t\tName string\n\t\t\t\t\t\t\tDescription string\n\t\t\t\t\t\t}{tool.Name, tool.Description})\n\t\t\t\t\t}\n\t\t\t\t}\n\t\t\t\t// Write filtered tools\n\t\t\t\tif len(filteredTools) > 0 {\n\t\t\t\t\tsb.WriteString(\"| Tool | Description |\\n\")\n\t\t\t\t\tsb.WriteString(\"|------|-------------|\\n\")\n\t\t\t\t\tfor _, tool := range filteredTools {\n\t\t\t\t\t\tdesc := tool.Description\n\t\t\t\t\t\tif len(desc) > 100 {\n\t\t\t\t\t\t\tdesc = desc[:97] + \"...\"\n\t\t\t\t\t\t}\n\t\t\t\t\t\t// Escape pipe characters in description\n\t\t\t\t\t\tdesc = strings.ReplaceAll(desc, \"|\", \"\\\\|\")\n\t\t\t\t\t\tsb.WriteString(fmt.Sprintf(\"| `%s` | %s |\\n\", tool.Name, desc))\n\t\t\t\t\t}\n\t\t\t\t} else {\n\t\t\t\t\tsb.WriteString(\"Configured tools: \")\n\t\t\t\t\tsb.WriteString(strings.Join(mcpConfig.Tools, \", \"))\n\t\t\t\t}\n\t\t\t} else if len(toolsToShow) > 0 {\n\t\t\t\t// Show all available tools\n\t\t\t\tsb.WriteString(\"| Tool | Description |\\n\")\n\t\t\t\tsb.WriteString(\"|------|-------------|\\n\")\n\t\t\t\tfor _, tool := range toolsToShow {\n\t\t\t\t\tdesc := tool.Description\n\t\t\t\t\tif len(desc) > 100 {\n\t\t\t\t\t\tdesc = desc[:97] + \"...\"\n\t\t\t\t\t}\n\t\t\t\t\t// Escape pipe characters in description\n\t\t\t\t\tdesc = strings.ReplaceAll(desc, \"|\", \"\\\\|\")\n\t\t\t\t\tsb.WriteString(fmt.Sprintf(\"| `%s` | %s |\\n\", tool.Name, desc))\n\t\t\t\t}\n\t\t\t} else {\n\t\t\t\tsb.WriteString(\"No tools available\\n\")\n\t\t\t}\n\t\t\tsb.WriteString(\"\\n\")\n\t\t}\n\t}\n\n\t// Available Knowledge Base\n\tif robot.Config.KB != nil && len(robot.Config.KB.Collections) > 0 {\n\t\tif !hasContent {\n\t\t\tsb.WriteString(\"## Available Resources\\n\\n\")\n\t\t\thasContent = true\n\t\t}\n\t\tsb.WriteString(\"### Knowledge Base\\n\")\n\t\tsb.WriteString(\"You have access to these knowledge collections:\\n\")\n\t\tfor _, collection := range robot.Config.KB.Collections {\n\t\t\tsb.WriteString(fmt.Sprintf(\"- %s\\n\", collection))\n\t\t}\n\t\tsb.WriteString(\"\\n\")\n\t}\n\n\t// Available Database Models\n\tif robot.Config.DB != nil && len(robot.Config.DB.Models) > 0 {\n\t\tif !hasContent {\n\t\t\tsb.WriteString(\"## Available Resources\\n\\n\")\n\t\t\thasContent = true\n\t\t}\n\t\tsb.WriteString(\"### Database\\n\")\n\t\tsb.WriteString(\"You can query these database models:\\n\")\n\t\tfor _, model := range robot.Config.DB.Models {\n\t\t\tsb.WriteString(fmt.Sprintf(\"- %s\\n\", model))\n\t\t}\n\t\tsb.WriteString(\"\\n\")\n\t}\n\n\tif !hasContent {\n\t\treturn \"\"\n\t}\n\n\tsb.WriteString(\"**Important**: Only plan goals and tasks that can be accomplished with the above resources.\\n\")\n\treturn sb.String()\n}\n\n// FormatInspirationReport formats InspirationReport as user message content\n// Used by P1 (Goals) phase when trigger is Clock\nfunc (f *InputFormatter) FormatInspirationReport(report *robottypes.InspirationReport) string {\n\tif report == nil {\n\t\treturn \"\"\n\t}\n\n\tvar sb strings.Builder\n\n\t// Clock context summary (if available)\n\tif report.Clock != nil {\n\t\tsb.WriteString(\"## Time Context\\n\\n\")\n\t\tsb.WriteString(fmt.Sprintf(\"- **Time**: %s %s\\n\", report.Clock.DayOfWeek, report.Clock.Now.Format(\"15:04\")))\n\t\tsb.WriteString(fmt.Sprintf(\"- **Date**: %d/%d/%d\\n\", report.Clock.Year, report.Clock.Month, report.Clock.DayOfMonth))\n\n\t\t// Add relevant time markers\n\t\tvar markers []string\n\t\tif report.Clock.IsWeekend {\n\t\t\tmarkers = append(markers, \"Weekend\")\n\t\t}\n\t\tif report.Clock.IsMonthStart {\n\t\t\tmarkers = append(markers, \"Month Start\")\n\t\t}\n\t\tif report.Clock.IsMonthEnd {\n\t\t\tmarkers = append(markers, \"Month End\")\n\t\t}\n\t\tif report.Clock.IsQuarterEnd {\n\t\t\tmarkers = append(markers, \"Quarter End\")\n\t\t}\n\t\tif len(markers) > 0 {\n\t\t\tsb.WriteString(fmt.Sprintf(\"- **Markers**: %s\\n\", strings.Join(markers, \", \")))\n\t\t}\n\t\tsb.WriteString(\"\\n\")\n\t}\n\n\t// Inspiration content\n\tif report.Content != \"\" {\n\t\tsb.WriteString(\"## Inspiration Report\\n\\n\")\n\t\tsb.WriteString(report.Content)\n\t\tsb.WriteString(\"\\n\")\n\t}\n\n\treturn sb.String()\n}\n\n// FormatTriggerInput formats TriggerInput as user message content\n// Used by P1 (Goals) phase when trigger is Human or Event\nfunc (f *InputFormatter) FormatTriggerInput(input *robottypes.TriggerInput) string {\n\tif input == nil {\n\t\treturn \"\"\n\t}\n\n\tvar sb strings.Builder\n\n\t// Human intervention\n\tif input.Action != \"\" {\n\t\tsb.WriteString(\"## Human Intervention\\n\\n\")\n\t\tsb.WriteString(fmt.Sprintf(\"- **Action**: %s\\n\", input.Action))\n\t\tif input.UserID != \"\" {\n\t\t\tsb.WriteString(fmt.Sprintf(\"- **User**: %s\\n\", input.UserID))\n\t\t}\n\n\t\t// Messages\n\t\tif len(input.Messages) > 0 {\n\t\t\tsb.WriteString(\"\\n### User Input\\n\\n\")\n\t\t\tfor _, msg := range input.Messages {\n\t\t\t\tif content, ok := msg.Content.(string); ok {\n\t\t\t\t\tsb.WriteString(content)\n\t\t\t\t\tsb.WriteString(\"\\n\")\n\t\t\t\t}\n\t\t\t}\n\t\t}\n\t\treturn sb.String()\n\t}\n\n\t// Event trigger\n\tif input.Source != \"\" {\n\t\tsb.WriteString(\"## Event Trigger\\n\\n\")\n\t\tsb.WriteString(fmt.Sprintf(\"- **Source**: %s\\n\", input.Source))\n\t\tsb.WriteString(fmt.Sprintf(\"- **Event Type**: %s\\n\", input.EventType))\n\n\t\t// Event data\n\t\tif input.Data != nil {\n\t\t\tsb.WriteString(\"\\n### Event Data\\n\\n\")\n\t\t\tsb.WriteString(\"```json\\n\")\n\t\t\tif data, err := json.MarshalIndent(input.Data, \"\", \" \"); err == nil {\n\t\t\t\tsb.WriteString(string(data))\n\t\t\t}\n\t\t\tsb.WriteString(\"\\n```\\n\")\n\t\t}\n\t\treturn sb.String()\n\t}\n\n\treturn \"\"\n}\n\n// FormatGoals formats Goals as user message content\n// Used by P2 (Tasks) phase\nfunc (f *InputFormatter) FormatGoals(goals *robottypes.Goals, robot *robottypes.Robot) string {\n\tif goals == nil {\n\t\treturn \"\"\n\t}\n\n\tvar sb strings.Builder\n\n\t// Goals content\n\tsb.WriteString(\"## Goals\\n\\n\")\n\tsb.WriteString(goals.Content)\n\tsb.WriteString(\"\\n\")\n\n\t// Delivery target (from P1) - important for task planning\n\t// Tasks should be designed to produce output suitable for the delivery method\n\tif goals.Delivery != nil {\n\t\tsb.WriteString(\"\\n## Delivery Target\\n\\n\")\n\t\tsb.WriteString(fmt.Sprintf(\"- **Type**: %s\\n\", goals.Delivery.Type))\n\t\tif len(goals.Delivery.Recipients) > 0 {\n\t\t\tsb.WriteString(fmt.Sprintf(\"- **Recipients**: %s\\n\", strings.Join(goals.Delivery.Recipients, \", \")))\n\t\t}\n\t\tif goals.Delivery.Format != \"\" {\n\t\t\tsb.WriteString(fmt.Sprintf(\"- **Format**: %s\\n\", goals.Delivery.Format))\n\t\t}\n\t\tif goals.Delivery.Template != \"\" {\n\t\t\tsb.WriteString(fmt.Sprintf(\"- **Template**: %s\\n\", goals.Delivery.Template))\n\t\t}\n\t\tsb.WriteString(\"\\n**Note**: Design tasks to produce output suitable for this delivery method.\\n\")\n\t}\n\n\t// Available resources - reuse FormatAvailableResources for consistency\n\tresourcesContent := f.FormatAvailableResources(robot)\n\tif resourcesContent != \"\" {\n\t\tsb.WriteString(\"\\n\")\n\t\tsb.WriteString(resourcesContent)\n\t}\n\n\treturn sb.String()\n}\n\n// FormatTasks formats Tasks as user message content\n// Used by P3 (Run) phase\nfunc (f *InputFormatter) FormatTasks(tasks []robottypes.Task) string {\n\tif len(tasks) == 0 {\n\t\treturn \"No tasks to execute.\"\n\t}\n\n\tvar sb strings.Builder\n\n\tsb.WriteString(\"## Tasks to Execute\\n\\n\")\n\tfor i, task := range tasks {\n\t\tsb.WriteString(fmt.Sprintf(\"### Task %d: %s\\n\\n\", i+1, task.ID))\n\t\tsb.WriteString(fmt.Sprintf(\"- **Goal Reference**: %s\\n\", task.GoalRef))\n\t\tsb.WriteString(fmt.Sprintf(\"- **Source**: %s\\n\", task.Source))\n\t\tsb.WriteString(fmt.Sprintf(\"- **Executor**: %s (%s)\\n\", task.ExecutorID, task.ExecutorType))\n\n\t\t// Task content\n\t\tif len(task.Messages) > 0 {\n\t\t\tsb.WriteString(\"\\n**Instructions**:\\n\")\n\t\t\tfor _, msg := range task.Messages {\n\t\t\t\tif content, ok := msg.Content.(string); ok {\n\t\t\t\t\tsb.WriteString(content)\n\t\t\t\t\tsb.WriteString(\"\\n\")\n\t\t\t\t}\n\t\t\t}\n\t\t}\n\n\t\t// Arguments\n\t\tif len(task.Args) > 0 {\n\t\t\tsb.WriteString(\"\\n**Arguments**:\\n\")\n\t\t\tif args, err := json.MarshalIndent(task.Args, \"\", \" \"); err == nil {\n\t\t\t\tsb.WriteString(\"```json\\n\")\n\t\t\t\tsb.WriteString(string(args))\n\t\t\t\tsb.WriteString(\"\\n```\\n\")\n\t\t\t}\n\t\t}\n\t\tsb.WriteString(\"\\n\")\n\t}\n\n\treturn sb.String()\n}\n\n// FormatTaskResults formats TaskResults as user message content\n// Used by P4 (Delivery) and P5 (Learning) phases\nfunc (f *InputFormatter) FormatTaskResults(results []robottypes.TaskResult) string {\n\tif len(results) == 0 {\n\t\treturn \"No task results.\"\n\t}\n\n\tvar sb strings.Builder\n\n\tsb.WriteString(\"## Task Results\\n\\n\")\n\n\tsuccessCount := 0\n\tfailCount := 0\n\tvalidatedPassedCount := 0\n\tvalidatedTotalCount := 0\n\n\tfor _, result := range results {\n\t\tif result.Success {\n\t\t\tsuccessCount++\n\t\t} else {\n\t\t\tfailCount++\n\t\t}\n\t\tif result.Validation != nil {\n\t\t\tvalidatedTotalCount++\n\t\t\tif result.Validation.Passed {\n\t\t\t\tvalidatedPassedCount++\n\t\t\t}\n\t\t}\n\n\t\tsb.WriteString(fmt.Sprintf(\"### Task: %s\\n\\n\", result.TaskID))\n\t\tif result.Success {\n\t\t\tsb.WriteString(\"- **Status**: ✓ Success\\n\")\n\t\t} else {\n\t\t\tsb.WriteString(\"- **Status**: ✗ Failed\\n\")\n\t\t}\n\t\tsb.WriteString(fmt.Sprintf(\"- **Duration**: %dms\\n\", result.Duration))\n\n\t\t// Validation result (P3)\n\t\tif result.Validation != nil {\n\t\t\tif result.Validation.Passed {\n\t\t\t\tsb.WriteString(fmt.Sprintf(\"- **Validation**: ✓ Passed (score: %.2f)\\n\", result.Validation.Score))\n\t\t\t} else {\n\t\t\t\tsb.WriteString(\"- **Validation**: ✗ Failed\\n\")\n\t\t\t\tif len(result.Validation.Issues) > 0 {\n\t\t\t\t\tsb.WriteString(\" - Issues:\\n\")\n\t\t\t\t\tfor _, issue := range result.Validation.Issues {\n\t\t\t\t\t\tsb.WriteString(fmt.Sprintf(\" - %s\\n\", issue))\n\t\t\t\t\t}\n\t\t\t\t}\n\t\t\t}\n\t\t}\n\n\t\t// Output\n\t\tif result.Output != nil {\n\t\t\tsb.WriteString(\"\\n**Output**:\\n\")\n\t\t\tif output, err := json.MarshalIndent(result.Output, \"\", \" \"); err == nil {\n\t\t\t\tsb.WriteString(\"```json\\n\")\n\t\t\t\tsb.WriteString(string(output))\n\t\t\t\tsb.WriteString(\"\\n```\\n\")\n\t\t\t} else {\n\t\t\t\tsb.WriteString(fmt.Sprintf(\"%v\\n\", result.Output))\n\t\t\t}\n\t\t}\n\n\t\t// Error\n\t\tif result.Error != \"\" {\n\t\t\tsb.WriteString(fmt.Sprintf(\"\\n**Error**: %s\\n\", result.Error))\n\t\t}\n\t\tsb.WriteString(\"\\n\")\n\t}\n\n\t// Summary\n\tsb.WriteString(fmt.Sprintf(\"## Summary\\n\\n- Total: %d tasks\\n- Success: %d\\n- Failed: %d\\n- Validated: %d/%d\\n\",\n\t\tlen(results), successCount, failCount, validatedPassedCount, validatedTotalCount))\n\n\treturn sb.String()\n}\n\n// FormatExecutionSummary formats the entire execution for P5 (Learning) phase\nfunc (f *InputFormatter) FormatExecutionSummary(exec *robottypes.Execution) string {\n\tif exec == nil {\n\t\treturn \"\"\n\t}\n\n\tvar sb strings.Builder\n\n\t// Execution metadata\n\tsb.WriteString(\"## Execution Summary\\n\\n\")\n\tsb.WriteString(fmt.Sprintf(\"- **ID**: %s\\n\", exec.ID))\n\tsb.WriteString(fmt.Sprintf(\"- **Trigger**: %s\\n\", exec.TriggerType))\n\tsb.WriteString(fmt.Sprintf(\"- **Status**: %s\\n\", exec.Status))\n\tsb.WriteString(fmt.Sprintf(\"- **Start Time**: %s\\n\", exec.StartTime.Format(\"2006-01-02 15:04:05\")))\n\tif exec.EndTime != nil {\n\t\tsb.WriteString(fmt.Sprintf(\"- **End Time**: %s\\n\", exec.EndTime.Format(\"2006-01-02 15:04:05\")))\n\t\tduration := exec.EndTime.Sub(exec.StartTime)\n\t\tsb.WriteString(fmt.Sprintf(\"- **Duration**: %s\\n\", duration.String()))\n\t}\n\tif exec.Error != \"\" {\n\t\tsb.WriteString(fmt.Sprintf(\"- **Error**: %s\\n\", exec.Error))\n\t}\n\tsb.WriteString(\"\\n\")\n\n\t// Inspiration (P0)\n\tif exec.Inspiration != nil && exec.Inspiration.Content != \"\" {\n\t\tsb.WriteString(\"## Inspiration (P0)\\n\\n\")\n\t\tsb.WriteString(exec.Inspiration.Content)\n\t\tsb.WriteString(\"\\n\\n\")\n\t}\n\n\t// Goals (P1)\n\tif exec.Goals != nil && exec.Goals.Content != \"\" {\n\t\tsb.WriteString(\"## Goals (P1)\\n\\n\")\n\t\tsb.WriteString(exec.Goals.Content)\n\t\tsb.WriteString(\"\\n\\n\")\n\t}\n\n\t// Tasks (P2)\n\tif len(exec.Tasks) > 0 {\n\t\tsb.WriteString(\"## Tasks (P2)\\n\\n\")\n\t\tfor i, task := range exec.Tasks {\n\t\t\tsb.WriteString(fmt.Sprintf(\"%d. [%s] %s (executor: %s)\\n\",\n\t\t\t\ti+1, task.Status, task.ID, task.ExecutorID))\n\t\t}\n\t\tsb.WriteString(\"\\n\")\n\t}\n\n\t// Results (P3)\n\tif len(exec.Results) > 0 {\n\t\tsb.WriteString(\"## Results (P3)\\n\\n\")\n\t\tfor _, result := range exec.Results {\n\t\t\tstatus := \"✓\"\n\t\t\tif !result.Success {\n\t\t\t\tstatus = \"✗\"\n\t\t\t}\n\t\t\tsb.WriteString(fmt.Sprintf(\"- %s %s (%dms)\\n\", status, result.TaskID, result.Duration))\n\t\t}\n\t\tsb.WriteString(\"\\n\")\n\t}\n\n\t// Delivery (P4)\n\tif exec.Delivery != nil {\n\t\tsb.WriteString(\"## Delivery (P4)\\n\\n\")\n\t\tif exec.Delivery.Content != nil {\n\t\t\tsb.WriteString(fmt.Sprintf(\"- **Summary**: %s\\n\", exec.Delivery.Content.Summary))\n\t\t}\n\t\tif exec.Delivery.Success {\n\t\t\tsb.WriteString(\"- **Status**: ✓ Success\\n\")\n\t\t} else {\n\t\t\tsb.WriteString(fmt.Sprintf(\"- **Status**: ✗ Failed (%s)\\n\", exec.Delivery.Error))\n\t\t}\n\t\tif len(exec.Delivery.Results) > 0 {\n\t\t\tsb.WriteString(fmt.Sprintf(\"- **Channels**: %d\\n\", len(exec.Delivery.Results)))\n\t\t}\n\t\tsb.WriteString(\"\\n\")\n\t}\n\n\treturn sb.String()\n}\n\n// BuildMessages is a convenience method to build messages array from content\nfunc (f *InputFormatter) BuildMessages(userContent string) []agentcontext.Message {\n\treturn []agentcontext.Message{\n\t\t{\n\t\t\tRole: agentcontext.RoleUser,\n\t\t\tContent: userContent,\n\t\t},\n\t}\n}\n\n// BuildMessagesWithSystem builds messages array with system and user content\nfunc (f *InputFormatter) BuildMessagesWithSystem(systemContent, userContent string) []agentcontext.Message {\n\treturn []agentcontext.Message{\n\t\t{\n\t\t\tRole: agentcontext.RoleSystem,\n\t\t\tContent: systemContent,\n\t\t},\n\t\t{\n\t\t\tRole: agentcontext.RoleUser,\n\t\t\tContent: userContent,\n\t\t},\n\t}\n}\n" }, { "path": "agent/robot/executor/standard/input_integration_test.go", "content": "package standard_test\n\nimport (\n\t\"strings\"\n\t\"testing\"\n\t\"time\"\n\n\t\"github.com/stretchr/testify/assert\"\n\t\"github.com/stretchr/testify/require\"\n\t\"github.com/yaoapp/yao/agent/robot/executor/standard\"\n\t\"github.com/yaoapp/yao/agent/robot/types\"\n\t\"github.com/yaoapp/yao/agent/testutils\"\n)\n\n// ============================================================================\n// InputFormatter Integration Tests with Real Data\n// These tests use the yao-dev-app environment with real assistants and MCPs\n// ============================================================================\n\nfunc TestFormatAvailableResourcesIntegration(t *testing.T) {\n\tif testing.Short() {\n\t\tt.Skip(\"Skipping integration test\")\n\t}\n\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\tformatter := standard.NewInputFormatter()\n\n\tt.Run(\"formats_real_agents_with_details\", func(t *testing.T) {\n\t\trobot := &types.Robot{\n\t\t\tMemberID: \"test-robot-agents\",\n\t\t\tConfig: &types.Config{\n\t\t\t\tIdentity: &types.Identity{\n\t\t\t\t\tRole: \"Test Robot\",\n\t\t\t\t\tDuties: []string{\"Testing agent formatting\"},\n\t\t\t\t},\n\t\t\t\tDefaultLocale: \"en\",\n\t\t\t\tResources: &types.Resources{\n\t\t\t\t\t// Use real agents from yao-dev-app/assistants\n\t\t\t\t\tAgents: []string{\n\t\t\t\t\t\t\"experts.data-analyst\",\n\t\t\t\t\t\t\"experts.text-writer\",\n\t\t\t\t\t\t\"experts.summarizer\",\n\t\t\t\t\t},\n\t\t\t\t},\n\t\t\t},\n\t\t}\n\n\t\tresult := formatter.FormatAvailableResources(robot)\n\n\t\t// Verify structure\n\t\tassert.Contains(t, result, \"## Available Resources\")\n\t\tassert.Contains(t, result, \"### Agents\")\n\t\tassert.Contains(t, result, \"These are the AI assistants you can delegate tasks to:\")\n\n\t\t// Verify real agent details are included\n\t\t// experts.data-analyst should show name and description\n\t\tassert.Contains(t, result, \"experts.data-analyst\")\n\t\tassert.Contains(t, result, \"Data Analyst Expert\") // Name from package.yao\n\n\t\t// experts.text-writer\n\t\tassert.Contains(t, result, \"experts.text-writer\")\n\n\t\t// experts.summarizer\n\t\tassert.Contains(t, result, \"experts.summarizer\")\n\n\t\t// Verify important note is present\n\t\tassert.Contains(t, result, \"Only plan goals and tasks that can be accomplished\")\n\n\t\tt.Logf(\"Formatted agents result:\\n%s\", result)\n\t})\n\n\tt.Run(\"formats_real_mcp_with_tool_details\", func(t *testing.T) {\n\t\trobot := &types.Robot{\n\t\t\tMemberID: \"test-robot-mcp\",\n\t\t\tConfig: &types.Config{\n\t\t\t\tIdentity: &types.Identity{\n\t\t\t\t\tRole: \"Test Robot\",\n\t\t\t\t\tDuties: []string{\"Testing MCP formatting\"},\n\t\t\t\t},\n\t\t\t\tDefaultLocale: \"en\",\n\t\t\t\tResources: &types.Resources{\n\t\t\t\t\t// Use real MCPs from yao-dev-app/mcps\n\t\t\t\t\tMCP: []types.MCPConfig{\n\t\t\t\t\t\t{ID: \"echo\", Tools: []string{\"ping\", \"status\"}}, // Specific tools\n\t\t\t\t\t\t{ID: \"echo\"}, // All tools\n\t\t\t\t\t},\n\t\t\t\t},\n\t\t\t},\n\t\t}\n\n\t\tresult := formatter.FormatAvailableResources(robot)\n\n\t\t// Verify structure\n\t\tassert.Contains(t, result, \"## Available Resources\")\n\t\tassert.Contains(t, result, \"### MCP Tools\")\n\t\tassert.Contains(t, result, \"These are the external tools and services you can use:\")\n\n\t\t// Verify MCP details\n\t\tassert.Contains(t, result, \"echo\")\n\n\t\t// Verify important note is present\n\t\tassert.Contains(t, result, \"Only plan goals and tasks that can be accomplished\")\n\n\t\tt.Logf(\"Formatted MCP result:\\n%s\", result)\n\t})\n\n\tt.Run(\"formats_combined_resources\", func(t *testing.T) {\n\t\trobot := &types.Robot{\n\t\t\tMemberID: \"test-robot-combined\",\n\t\t\tConfig: &types.Config{\n\t\t\t\tIdentity: &types.Identity{\n\t\t\t\t\tRole: \"Sales Analyst Robot\",\n\t\t\t\t\tDuties: []string{\"Analyze sales data\", \"Generate reports\"},\n\t\t\t\t\tRules: []string{\"Be accurate\", \"Be concise\"},\n\t\t\t\t},\n\t\t\t\tDefaultLocale: \"en\",\n\t\t\t\tResources: &types.Resources{\n\t\t\t\t\tAgents: []string{\n\t\t\t\t\t\t\"experts.data-analyst\",\n\t\t\t\t\t\t\"experts.summarizer\",\n\t\t\t\t\t},\n\t\t\t\t\tMCP: []types.MCPConfig{\n\t\t\t\t\t\t{ID: \"echo\", Tools: []string{\"ping\", \"echo\"}},\n\t\t\t\t\t},\n\t\t\t\t},\n\t\t\t\tKB: &types.KB{\n\t\t\t\t\tCollections: []string{\"sales-policies\", \"product-catalog\"},\n\t\t\t\t},\n\t\t\t\tDB: &types.DB{\n\t\t\t\t\tModels: []string{\"sales\", \"customers\", \"orders\"},\n\t\t\t\t},\n\t\t\t},\n\t\t}\n\n\t\tresult := formatter.FormatAvailableResources(robot)\n\n\t\t// Verify all sections are present\n\t\tassert.Contains(t, result, \"## Available Resources\")\n\t\tassert.Contains(t, result, \"### Agents\")\n\t\tassert.Contains(t, result, \"### MCP Tools\")\n\t\tassert.Contains(t, result, \"### Knowledge Base\")\n\t\tassert.Contains(t, result, \"### Database\")\n\n\t\t// Verify agents\n\t\tassert.Contains(t, result, \"experts.data-analyst\")\n\t\tassert.Contains(t, result, \"experts.summarizer\")\n\n\t\t// Verify MCP\n\t\tassert.Contains(t, result, \"echo\")\n\n\t\t// Verify KB\n\t\tassert.Contains(t, result, \"sales-policies\")\n\t\tassert.Contains(t, result, \"product-catalog\")\n\n\t\t// Verify DB\n\t\tassert.Contains(t, result, \"sales\")\n\t\tassert.Contains(t, result, \"customers\")\n\t\tassert.Contains(t, result, \"orders\")\n\n\t\tt.Logf(\"Formatted combined resources result:\\n%s\", result)\n\t})\n\n\tt.Run(\"handles_locale_zh\", func(t *testing.T) {\n\t\trobot := &types.Robot{\n\t\t\tMemberID: \"test-robot-zh\",\n\t\t\tConfig: &types.Config{\n\t\t\t\tIdentity: &types.Identity{\n\t\t\t\t\tRole: \"测试机器人\",\n\t\t\t\t\tDuties: []string{\"测试国际化\"},\n\t\t\t\t},\n\t\t\t\tDefaultLocale: \"zh-cn\",\n\t\t\t\tResources: &types.Resources{\n\t\t\t\t\t// Use agents that have zh-cn locales\n\t\t\t\t\tAgents: []string{\n\t\t\t\t\t\t\"hello\", // This agent has locales/zh-cn.yml\n\t\t\t\t\t\t\"mohe\", // This agent also has locales/zh-cn.yml\n\t\t\t\t\t},\n\t\t\t\t},\n\t\t\t},\n\t\t}\n\n\t\tresult := formatter.FormatAvailableResourcesWithLocale(robot, \"zh-cn\")\n\n\t\t// Verify structure\n\t\tassert.Contains(t, result, \"## Available Resources\")\n\t\tassert.Contains(t, result, \"### Agents\")\n\n\t\t// Verify agents are listed\n\t\tassert.Contains(t, result, \"hello\")\n\t\tassert.Contains(t, result, \"mohe\")\n\n\t\tt.Logf(\"Formatted zh-cn result:\\n%s\", result)\n\t})\n\n\tt.Run(\"gracefully_handles_missing_agents\", func(t *testing.T) {\n\t\trobot := &types.Robot{\n\t\t\tMemberID: \"test-robot-missing\",\n\t\t\tConfig: &types.Config{\n\t\t\t\tIdentity: &types.Identity{\n\t\t\t\t\tRole: \"Test Robot\",\n\t\t\t\t},\n\t\t\t\tResources: &types.Resources{\n\t\t\t\t\tAgents: []string{\n\t\t\t\t\t\t\"non-existent-agent\",\n\t\t\t\t\t\t\"experts.data-analyst\", // This one exists\n\t\t\t\t\t\t\"another-missing-agent\",\n\t\t\t\t\t},\n\t\t\t\t},\n\t\t\t},\n\t\t}\n\n\t\tresult := formatter.FormatAvailableResources(robot)\n\n\t\t// Should not panic, should include fallback for missing agents\n\t\tassert.Contains(t, result, \"## Available Resources\")\n\t\tassert.Contains(t, result, \"### Agents\")\n\n\t\t// Missing agents should still be listed with just ID\n\t\tassert.Contains(t, result, \"non-existent-agent\")\n\t\tassert.Contains(t, result, \"another-missing-agent\")\n\n\t\t// Existing agent should have full details\n\t\tassert.Contains(t, result, \"experts.data-analyst\")\n\t\tassert.Contains(t, result, \"Data Analyst Expert\")\n\n\t\tt.Logf(\"Formatted with missing agents:\\n%s\", result)\n\t})\n\n\tt.Run(\"gracefully_handles_missing_mcp\", func(t *testing.T) {\n\t\trobot := &types.Robot{\n\t\t\tMemberID: \"test-robot-missing-mcp\",\n\t\t\tConfig: &types.Config{\n\t\t\t\tIdentity: &types.Identity{\n\t\t\t\t\tRole: \"Test Robot\",\n\t\t\t\t},\n\t\t\t\tResources: &types.Resources{\n\t\t\t\t\tMCP: []types.MCPConfig{\n\t\t\t\t\t\t{ID: \"non-existent-mcp\", Tools: []string{\"tool1\", \"tool2\"}},\n\t\t\t\t\t\t{ID: \"echo\"}, // This one exists\n\t\t\t\t\t},\n\t\t\t\t},\n\t\t\t},\n\t\t}\n\n\t\tresult := formatter.FormatAvailableResources(robot)\n\n\t\t// Should not panic, should include fallback for missing MCP\n\t\tassert.Contains(t, result, \"## Available Resources\")\n\t\tassert.Contains(t, result, \"### MCP Tools\")\n\n\t\t// Missing MCP should still be listed with fallback\n\t\tassert.Contains(t, result, \"non-existent-mcp\")\n\t\tassert.Contains(t, result, \"tool1, tool2\")\n\n\t\t// Existing MCP should have details\n\t\tassert.Contains(t, result, \"echo\")\n\n\t\tt.Logf(\"Formatted with missing MCP:\\n%s\", result)\n\t})\n}\n\nfunc TestFormatAvailableResourcesTableFormat(t *testing.T) {\n\tif testing.Short() {\n\t\tt.Skip(\"Skipping integration test\")\n\t}\n\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\tformatter := standard.NewInputFormatter()\n\n\tt.Run(\"mcp_tools_in_table_format\", func(t *testing.T) {\n\t\trobot := &types.Robot{\n\t\t\tMemberID: \"test-robot-table\",\n\t\t\tConfig: &types.Config{\n\t\t\t\tIdentity: &types.Identity{\n\t\t\t\t\tRole: \"Test Robot\",\n\t\t\t\t},\n\t\t\t\tResources: &types.Resources{\n\t\t\t\t\tMCP: []types.MCPConfig{\n\t\t\t\t\t\t{ID: \"echo\"}, // All tools - should show table\n\t\t\t\t\t},\n\t\t\t\t},\n\t\t\t},\n\t\t}\n\n\t\tresult := formatter.FormatAvailableResources(robot)\n\n\t\t// Check if table format is used when tools are available\n\t\t// Table headers: | Tool | Description |\n\t\tif strings.Contains(result, \"| Tool | Description |\") {\n\t\t\tassert.Contains(t, result, \"|------|-------------|\")\n\t\t\tt.Logf(\"MCP tools displayed in table format:\\n%s\", result)\n\t\t} else {\n\t\t\t// Fallback format\n\t\t\tt.Logf(\"MCP tools displayed in fallback format:\\n%s\", result)\n\t\t}\n\t})\n}\n\nfunc TestFormatClockContextWithRobotIntegration(t *testing.T) {\n\tif testing.Short() {\n\t\tt.Skip(\"Skipping integration test\")\n\t}\n\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\tformatter := standard.NewInputFormatter()\n\n\tt.Run(\"full_context_for_inspiration\", func(t *testing.T) {\n\t\t// Create a realistic robot configuration\n\t\trobot := &types.Robot{\n\t\t\tMemberID: \"sales-robot-001\",\n\t\t\tTeamID: \"team-001\",\n\t\t\tDisplayName: \"Sales Analyst Robot\",\n\t\t\tAutonomousMode: true,\n\t\t\tConfig: &types.Config{\n\t\t\t\tIdentity: &types.Identity{\n\t\t\t\t\tRole: \"Sales Analyst\",\n\t\t\t\t\tDuties: []string{\"Monitor sales performance\", \"Generate daily reports\", \"Alert on anomalies\"},\n\t\t\t\t\tRules: []string{\"Only use approved data sources\", \"Maintain confidentiality\"},\n\t\t\t\t},\n\t\t\t\tDefaultLocale: \"en\",\n\t\t\t\tResources: &types.Resources{\n\t\t\t\t\tAgents: []string{\n\t\t\t\t\t\t\"experts.data-analyst\",\n\t\t\t\t\t\t\"experts.summarizer\",\n\t\t\t\t\t},\n\t\t\t\t\tMCP: []types.MCPConfig{\n\t\t\t\t\t\t{ID: \"echo\", Tools: []string{\"ping\"}},\n\t\t\t\t\t},\n\t\t\t\t},\n\t\t\t},\n\t\t}\n\n\t\t// Create clock context\n\t\tclock := types.NewClockContext(time.Now(), \"UTC\")\n\n\t\t// Format clock context (includes robot identity)\n\t\tclockContent := formatter.FormatClockContext(clock, robot)\n\n\t\t// Format available resources\n\t\tresourcesContent := formatter.FormatAvailableResources(robot)\n\n\t\t// Combine for full context (as done in inspiration.go)\n\t\tfullContext := clockContent + \"\\n\\n\" + resourcesContent\n\n\t\t// Verify full context contains all necessary information\n\t\trequire.NotEmpty(t, fullContext)\n\n\t\t// Time context\n\t\tassert.Contains(t, fullContext, \"## Current Time Context\")\n\t\tassert.Contains(t, fullContext, \"### Time Markers\")\n\n\t\t// Robot identity\n\t\tassert.Contains(t, fullContext, \"## Robot Identity\")\n\t\tassert.Contains(t, fullContext, \"Sales Analyst\")\n\t\tassert.Contains(t, fullContext, \"Monitor sales performance\")\n\t\tassert.Contains(t, fullContext, \"Only use approved data sources\")\n\n\t\t// Available resources\n\t\tassert.Contains(t, fullContext, \"## Available Resources\")\n\t\tassert.Contains(t, fullContext, \"### Agents\")\n\t\tassert.Contains(t, fullContext, \"experts.data-analyst\")\n\t\tassert.Contains(t, fullContext, \"### MCP Tools\")\n\n\t\tt.Logf(\"Full context for inspiration:\\n%s\", fullContext)\n\t})\n}\n" }, { "path": "agent/robot/executor/standard/input_test.go", "content": "package standard_test\n\nimport (\n\t\"testing\"\n\t\"time\"\n\n\t\"github.com/stretchr/testify/assert\"\n\t\"github.com/stretchr/testify/require\"\n\tagentcontext \"github.com/yaoapp/yao/agent/context\"\n\t\"github.com/yaoapp/yao/agent/robot/executor/standard\"\n\t\"github.com/yaoapp/yao/agent/robot/types\"\n)\n\n// ============================================================================\n// InputFormatter Tests\n// ============================================================================\n\nfunc TestInputFormatterFormatClockContext(t *testing.T) {\n\tformatter := standard.NewInputFormatter()\n\n\tt.Run(\"formats clock context with all fields\", func(t *testing.T) {\n\t\tnow := time.Date(2024, 1, 15, 9, 30, 0, 0, time.UTC)\n\t\tclock := types.NewClockContext(now, \"UTC\")\n\n\t\tresult := formatter.FormatClockContext(clock, nil)\n\n\t\tassert.Contains(t, result, \"## Current Time Context\")\n\t\tassert.Contains(t, result, \"2024-01-15 09:30:00\")\n\t\tassert.Contains(t, result, \"Monday\")\n\t\tassert.Contains(t, result, \"UTC\")\n\t\tassert.Contains(t, result, \"**Hour**: 9\")\n\t\tassert.Contains(t, result, \"### Time Markers\")\n\t})\n\n\tt.Run(\"shows all time markers with check/cross\", func(t *testing.T) {\n\t\t// Regular weekday, not month start/end\n\t\tnow := time.Date(2024, 1, 15, 14, 0, 0, 0, time.UTC)\n\t\tclock := types.NewClockContext(now, \"UTC\")\n\n\t\tresult := formatter.FormatClockContext(clock, nil)\n\n\t\t// Should show all markers, even when false\n\t\tassert.Contains(t, result, \"✗ Weekend\")\n\t\tassert.Contains(t, result, \"✗ Month Start\")\n\t\tassert.Contains(t, result, \"✗ Month End\")\n\t\tassert.Contains(t, result, \"✗ Quarter End\")\n\t\tassert.Contains(t, result, \"✗ Year End\")\n\t})\n\n\tt.Run(\"includes robot identity when provided\", func(t *testing.T) {\n\t\tnow := time.Now()\n\t\tclock := types.NewClockContext(now, \"UTC\")\n\t\trobot := &types.Robot{\n\t\t\tMemberID: \"test-robot\",\n\t\t\tConfig: &types.Config{\n\t\t\t\tIdentity: &types.Identity{\n\t\t\t\t\tRole: \"Sales Analyst\",\n\t\t\t\t\tDuties: []string{\"Analyze sales data\", \"Generate reports\"},\n\t\t\t\t\tRules: []string{\"Be accurate\", \"Be concise\"},\n\t\t\t\t},\n\t\t\t},\n\t\t}\n\n\t\tresult := formatter.FormatClockContext(clock, robot)\n\n\t\tassert.Contains(t, result, \"## Robot Identity\")\n\t\tassert.Contains(t, result, \"Sales Analyst\")\n\t\tassert.Contains(t, result, \"Analyze sales data\")\n\t\tassert.Contains(t, result, \"Be accurate\")\n\t})\n\n\tt.Run(\"returns empty for nil clock\", func(t *testing.T) {\n\t\tresult := formatter.FormatClockContext(nil, nil)\n\t\tassert.Empty(t, result)\n\t})\n\n\tt.Run(\"uses DisplayName/Bio/SystemPrompt when Identity is nil\", func(t *testing.T) {\n\t\tnow := time.Now()\n\t\tclock := types.NewClockContext(now, \"UTC\")\n\t\trobot := &types.Robot{\n\t\t\tMemberID: \"test-robot\",\n\t\t\tDisplayName: \"SEO Specialist\",\n\t\t\tBio: \"Focuses on content optimization\",\n\t\t\tSystemPrompt: \"You are an SEO assistant.\\n\\n## Core Duties\\n- Analyze keywords\",\n\t\t\tConfig: &types.Config{}, // Identity is nil\n\t\t}\n\n\t\tresult := formatter.FormatClockContext(clock, robot)\n\n\t\tassert.Contains(t, result, \"## Robot Identity\")\n\t\tassert.Contains(t, result, \"SEO Specialist\")\n\t\tassert.Contains(t, result, \"Focuses on content optimization\")\n\t\tassert.Contains(t, result, \"SEO assistant\")\n\t\tassert.Contains(t, result, \"Analyze keywords\")\n\t})\n\n\tt.Run(\"marks weekend correctly\", func(t *testing.T) {\n\t\t// Saturday\n\t\tsaturday := time.Date(2024, 1, 13, 10, 0, 0, 0, time.UTC)\n\t\tclock := types.NewClockContext(saturday, \"UTC\")\n\n\t\tresult := formatter.FormatClockContext(clock, nil)\n\n\t\tassert.Contains(t, result, \"✓ Weekend\")\n\t})\n\n\tt.Run(\"marks month start correctly\", func(t *testing.T) {\n\t\t// 2nd of month\n\t\tmonthStart := time.Date(2024, 1, 2, 10, 0, 0, 0, time.UTC)\n\t\tclock := types.NewClockContext(monthStart, \"UTC\")\n\n\t\tresult := formatter.FormatClockContext(clock, nil)\n\n\t\tassert.Contains(t, result, \"✓ Month Start\")\n\t})\n\n\tt.Run(\"marks month end correctly\", func(t *testing.T) {\n\t\t// 30th of January (last 3 days)\n\t\tmonthEnd := time.Date(2024, 1, 30, 10, 0, 0, 0, time.UTC)\n\t\tclock := types.NewClockContext(monthEnd, \"UTC\")\n\n\t\tresult := formatter.FormatClockContext(clock, nil)\n\n\t\tassert.Contains(t, result, \"✓ Month End\")\n\t})\n}\n\nfunc TestInputFormatterFormatInspirationReport(t *testing.T) {\n\tformatter := standard.NewInputFormatter()\n\n\tt.Run(\"formats inspiration report with clock\", func(t *testing.T) {\n\t\tnow := time.Date(2024, 1, 15, 9, 30, 0, 0, time.UTC)\n\t\tclock := types.NewClockContext(now, \"UTC\")\n\t\treport := &types.InspirationReport{\n\t\t\tClock: clock,\n\t\t\tContent: \"Today is a good day to analyze sales data.\",\n\t\t}\n\n\t\tresult := formatter.FormatInspirationReport(report)\n\n\t\tassert.Contains(t, result, \"## Time Context\")\n\t\tassert.Contains(t, result, \"Monday\")\n\t\tassert.Contains(t, result, \"## Inspiration Report\")\n\t\tassert.Contains(t, result, \"analyze sales data\")\n\t})\n\n\tt.Run(\"formats inspiration report without clock\", func(t *testing.T) {\n\t\treport := &types.InspirationReport{\n\t\t\tContent: \"Focus on quarterly review.\",\n\t\t}\n\n\t\tresult := formatter.FormatInspirationReport(report)\n\n\t\tassert.NotContains(t, result, \"## Time Context\")\n\t\tassert.Contains(t, result, \"## Inspiration Report\")\n\t\tassert.Contains(t, result, \"quarterly review\")\n\t})\n\n\tt.Run(\"returns empty for nil report\", func(t *testing.T) {\n\t\tresult := formatter.FormatInspirationReport(nil)\n\t\tassert.Empty(t, result)\n\t})\n}\n\nfunc TestInputFormatterFormatAvailableResources(t *testing.T) {\n\tformatter := standard.NewInputFormatter()\n\n\tt.Run(\"formats all resource types\", func(t *testing.T) {\n\t\trobot := &types.Robot{\n\t\t\tMemberID: \"test-robot\",\n\t\t\tConfig: &types.Config{\n\t\t\t\tResources: &types.Resources{\n\t\t\t\t\tAgents: []string{\"data-analyst\", \"chart-gen\", \"report-writer\"},\n\t\t\t\t\tMCP: []types.MCPConfig{\n\t\t\t\t\t\t{ID: \"database\", Tools: []string{\"query\", \"insert\"}},\n\t\t\t\t\t\t{ID: \"email\", Tools: []string{}}, // all tools\n\t\t\t\t\t},\n\t\t\t\t},\n\t\t\t\tKB: &types.KB{\n\t\t\t\t\tCollections: []string{\"sales-policies\", \"products\"},\n\t\t\t\t},\n\t\t\t\tDB: &types.DB{\n\t\t\t\t\tModels: []string{\"sales\", \"customers\", \"orders\"},\n\t\t\t\t},\n\t\t\t},\n\t\t}\n\n\t\tresult := formatter.FormatAvailableResources(robot)\n\n\t\t// Check structure\n\t\tassert.Contains(t, result, \"## Available Resources\")\n\n\t\t// Check agents\n\t\tassert.Contains(t, result, \"### Agents\")\n\t\tassert.Contains(t, result, \"data-analyst\")\n\t\tassert.Contains(t, result, \"chart-gen\")\n\t\tassert.Contains(t, result, \"report-writer\")\n\n\t\t// Check MCP tools\n\t\tassert.Contains(t, result, \"### MCP Tools\")\n\t\tassert.Contains(t, result, \"database\")\n\t\tassert.Contains(t, result, \"query, insert\")\n\t\tassert.Contains(t, result, \"email\")\n\t\tassert.Contains(t, result, \"all tools available\")\n\n\t\t// Check KB\n\t\tassert.Contains(t, result, \"### Knowledge Base\")\n\t\tassert.Contains(t, result, \"sales-policies\")\n\t\tassert.Contains(t, result, \"products\")\n\n\t\t// Check DB\n\t\tassert.Contains(t, result, \"### Database\")\n\t\tassert.Contains(t, result, \"sales\")\n\t\tassert.Contains(t, result, \"customers\")\n\t\tassert.Contains(t, result, \"orders\")\n\n\t\t// Check important note\n\t\tassert.Contains(t, result, \"Only plan goals and tasks that can be accomplished\")\n\t})\n\n\tt.Run(\"returns empty for nil robot\", func(t *testing.T) {\n\t\tresult := formatter.FormatAvailableResources(nil)\n\t\tassert.Empty(t, result)\n\t})\n\n\tt.Run(\"returns empty for robot without config\", func(t *testing.T) {\n\t\trobot := &types.Robot{MemberID: \"test\"}\n\t\tresult := formatter.FormatAvailableResources(robot)\n\t\tassert.Empty(t, result)\n\t})\n\n\tt.Run(\"returns empty for robot without resources\", func(t *testing.T) {\n\t\trobot := &types.Robot{\n\t\t\tMemberID: \"test\",\n\t\t\tConfig: &types.Config{},\n\t\t}\n\t\tresult := formatter.FormatAvailableResources(robot)\n\t\tassert.Empty(t, result)\n\t})\n\n\tt.Run(\"handles partial resources\", func(t *testing.T) {\n\t\trobot := &types.Robot{\n\t\t\tMemberID: \"test\",\n\t\t\tConfig: &types.Config{\n\t\t\t\tResources: &types.Resources{\n\t\t\t\t\tAgents: []string{\"single-agent\"},\n\t\t\t\t},\n\t\t\t},\n\t\t}\n\n\t\tresult := formatter.FormatAvailableResources(robot)\n\n\t\tassert.Contains(t, result, \"## Available Resources\")\n\t\tassert.Contains(t, result, \"### Agents\")\n\t\tassert.Contains(t, result, \"single-agent\")\n\t\tassert.NotContains(t, result, \"### MCP Tools\")\n\t\tassert.NotContains(t, result, \"### Knowledge Base\")\n\t\tassert.NotContains(t, result, \"### Database\")\n\t})\n}\n\nfunc TestInputFormatterFormatTriggerInput(t *testing.T) {\n\tformatter := standard.NewInputFormatter()\n\n\tt.Run(\"formats human intervention\", func(t *testing.T) {\n\t\tinput := &types.TriggerInput{\n\t\t\tAction: \"task.add\",\n\t\t\tUserID: \"user-123\",\n\t\t\tMessages: []agentcontext.Message{\n\t\t\t\t{Role: agentcontext.RoleUser, Content: \"Please add a task to review Q4 sales\"},\n\t\t\t},\n\t\t}\n\n\t\tresult := formatter.FormatTriggerInput(input)\n\n\t\tassert.Contains(t, result, \"## Human Intervention\")\n\t\tassert.Contains(t, result, \"task.add\")\n\t\tassert.Contains(t, result, \"user-123\")\n\t\tassert.Contains(t, result, \"### User Input\")\n\t\tassert.Contains(t, result, \"review Q4 sales\")\n\t})\n\n\tt.Run(\"formats event trigger\", func(t *testing.T) {\n\t\tinput := &types.TriggerInput{\n\t\t\tSource: \"webhook\",\n\t\t\tEventType: \"order.created\",\n\t\t\tData: map[string]interface{}{\n\t\t\t\t\"order_id\": \"12345\",\n\t\t\t\t\"amount\": 99.99,\n\t\t\t},\n\t\t}\n\n\t\tresult := formatter.FormatTriggerInput(input)\n\n\t\tassert.Contains(t, result, \"## Event Trigger\")\n\t\tassert.Contains(t, result, \"webhook\")\n\t\tassert.Contains(t, result, \"order.created\")\n\t\tassert.Contains(t, result, \"### Event Data\")\n\t\tassert.Contains(t, result, \"order_id\")\n\t\tassert.Contains(t, result, \"12345\")\n\t})\n\n\tt.Run(\"returns empty for nil input\", func(t *testing.T) {\n\t\tresult := formatter.FormatTriggerInput(nil)\n\t\tassert.Empty(t, result)\n\t})\n\n\tt.Run(\"returns empty for empty input\", func(t *testing.T) {\n\t\tinput := &types.TriggerInput{}\n\t\tresult := formatter.FormatTriggerInput(input)\n\t\tassert.Empty(t, result)\n\t})\n}\n\nfunc TestInputFormatterFormatGoals(t *testing.T) {\n\tformatter := standard.NewInputFormatter()\n\n\tt.Run(\"formats goals with resources\", func(t *testing.T) {\n\t\tgoals := &types.Goals{\n\t\t\tContent: \"1. Analyze sales data\\n2. Generate report\\n3. Send to stakeholders\",\n\t\t}\n\t\trobot := &types.Robot{\n\t\t\tMemberID: \"test-robot\",\n\t\t\tConfig: &types.Config{\n\t\t\t\tResources: &types.Resources{\n\t\t\t\t\tAgents: []string{\"data-analyzer\", \"report-generator\"},\n\t\t\t\t\tMCP: []types.MCPConfig{\n\t\t\t\t\t\t{ID: \"database\", Tools: []string{\"query\", \"insert\"}},\n\t\t\t\t\t\t{ID: \"email\"},\n\t\t\t\t\t},\n\t\t\t\t},\n\t\t\t},\n\t\t}\n\n\t\tresult := formatter.FormatGoals(goals, robot)\n\n\t\tassert.Contains(t, result, \"## Goals\")\n\t\tassert.Contains(t, result, \"Analyze sales data\")\n\t\tassert.Contains(t, result, \"## Available Resources\")\n\t\tassert.Contains(t, result, \"### Agents\")\n\t\tassert.Contains(t, result, \"data-analyzer\")\n\t\tassert.Contains(t, result, \"### MCP Tools\")\n\t\tassert.Contains(t, result, \"database\")\n\t\tassert.Contains(t, result, \"query, insert\")\n\t\tassert.Contains(t, result, \"email\")\n\t\tassert.Contains(t, result, \"all tools available\")\n\t})\n\n\tt.Run(\"formats goals without robot\", func(t *testing.T) {\n\t\tgoals := &types.Goals{\n\t\t\tContent: \"Complete the task.\",\n\t\t}\n\n\t\tresult := formatter.FormatGoals(goals, nil)\n\n\t\tassert.Contains(t, result, \"## Goals\")\n\t\tassert.Contains(t, result, \"Complete the task\")\n\t\tassert.NotContains(t, result, \"## Available Resources\")\n\t})\n\n\tt.Run(\"returns empty for nil goals\", func(t *testing.T) {\n\t\tresult := formatter.FormatGoals(nil, nil)\n\t\tassert.Empty(t, result)\n\t})\n}\n\nfunc TestInputFormatterFormatTasks(t *testing.T) {\n\tformatter := standard.NewInputFormatter()\n\n\tt.Run(\"formats multiple tasks\", func(t *testing.T) {\n\t\ttasks := []types.Task{\n\t\t\t{\n\t\t\t\tID: \"task-1\",\n\t\t\t\tGoalRef: \"goal-1\",\n\t\t\t\tSource: types.TaskSourceAuto,\n\t\t\t\tExecutorType: types.ExecutorMCP,\n\t\t\t\tExecutorID: \"database.query\",\n\t\t\t\tMessages: []agentcontext.Message{\n\t\t\t\t\t{Role: agentcontext.RoleUser, Content: \"Query sales data for Q4\"},\n\t\t\t\t},\n\t\t\t\tArgs: []any{\"sales\", \"Q4\"},\n\t\t\t},\n\t\t\t{\n\t\t\t\tID: \"task-2\",\n\t\t\t\tGoalRef: \"goal-1\",\n\t\t\t\tSource: types.TaskSourceAuto,\n\t\t\t\tExecutorType: types.ExecutorAssistant,\n\t\t\t\tExecutorID: \"report-generator\",\n\t\t\t},\n\t\t}\n\n\t\tresult := formatter.FormatTasks(tasks)\n\n\t\tassert.Contains(t, result, \"## Tasks to Execute\")\n\t\tassert.Contains(t, result, \"### Task 1: task-1\")\n\t\tassert.Contains(t, result, \"goal-1\")\n\t\tassert.Contains(t, result, \"database.query\")\n\t\tassert.Contains(t, result, \"**Instructions**\")\n\t\tassert.Contains(t, result, \"Query sales data\")\n\t\tassert.Contains(t, result, \"**Arguments**\")\n\t\tassert.Contains(t, result, \"### Task 2: task-2\")\n\t\tassert.Contains(t, result, \"report-generator\")\n\t})\n\n\tt.Run(\"returns message for empty tasks\", func(t *testing.T) {\n\t\tresult := formatter.FormatTasks(nil)\n\t\tassert.Equal(t, \"No tasks to execute.\", result)\n\n\t\tresult = formatter.FormatTasks([]types.Task{})\n\t\tassert.Equal(t, \"No tasks to execute.\", result)\n\t})\n}\n\nfunc TestInputFormatterFormatTaskResults(t *testing.T) {\n\tformatter := standard.NewInputFormatter()\n\n\tt.Run(\"formats task results with summary\", func(t *testing.T) {\n\t\tresults := []types.TaskResult{\n\t\t\t{\n\t\t\t\tTaskID: \"task-1\",\n\t\t\t\tSuccess: true,\n\t\t\t\tDuration: 150,\n\t\t\t\tValidation: &types.ValidationResult{\n\t\t\t\t\tPassed: true,\n\t\t\t\t\tScore: 0.95,\n\t\t\t\t},\n\t\t\t\tOutput: map[string]interface{}{\"rows\": 100},\n\t\t\t},\n\t\t\t{\n\t\t\t\tTaskID: \"task-2\",\n\t\t\t\tSuccess: false,\n\t\t\t\tDuration: 50,\n\t\t\t\tValidation: &types.ValidationResult{\n\t\t\t\t\tPassed: false,\n\t\t\t\t\tIssues: []string{\"Connection timeout\"},\n\t\t\t\t},\n\t\t\t\tError: \"Connection timeout\",\n\t\t\t},\n\t\t}\n\n\t\tresult := formatter.FormatTaskResults(results)\n\n\t\tassert.Contains(t, result, \"## Task Results\")\n\t\tassert.Contains(t, result, \"### Task: task-1\")\n\t\tassert.Contains(t, result, \"✓ Success\")\n\t\tassert.Contains(t, result, \"150ms\")\n\t\tassert.Contains(t, result, \"**Validation**: ✓ Passed\")\n\t\tassert.Contains(t, result, \"score: 0.95\")\n\t\tassert.Contains(t, result, \"**Output**\")\n\t\tassert.Contains(t, result, \"### Task: task-2\")\n\t\tassert.Contains(t, result, \"✗ Failed\")\n\t\tassert.Contains(t, result, \"**Validation**: ✗ Failed\")\n\t\tassert.Contains(t, result, \"Connection timeout\")\n\t\tassert.Contains(t, result, \"## Summary\")\n\t\tassert.Contains(t, result, \"Total: 2 tasks\")\n\t\tassert.Contains(t, result, \"Success: 1\")\n\t\tassert.Contains(t, result, \"Failed: 1\")\n\t\tassert.Contains(t, result, \"Validated: 1/2\")\n\t})\n\n\tt.Run(\"returns message for empty results\", func(t *testing.T) {\n\t\tresult := formatter.FormatTaskResults(nil)\n\t\tassert.Equal(t, \"No task results.\", result)\n\t})\n}\n\nfunc TestInputFormatterFormatExecutionSummary(t *testing.T) {\n\tformatter := standard.NewInputFormatter()\n\n\tt.Run(\"formats complete execution summary\", func(t *testing.T) {\n\t\tstartTime := time.Date(2024, 1, 15, 9, 0, 0, 0, time.UTC)\n\t\tendTime := time.Date(2024, 1, 15, 9, 5, 0, 0, time.UTC)\n\t\texec := &types.Execution{\n\t\t\tID: \"exec-123\",\n\t\t\tTriggerType: types.TriggerClock,\n\t\t\tStatus: types.ExecCompleted,\n\t\t\tStartTime: startTime,\n\t\t\tEndTime: &endTime,\n\t\t\tInspiration: &types.InspirationReport{\n\t\t\t\tContent: \"Morning analysis suggests high activity.\",\n\t\t\t},\n\t\t\tGoals: &types.Goals{\n\t\t\t\tContent: \"1. Review data\\n2. Generate report\",\n\t\t\t},\n\t\t\tTasks: []types.Task{\n\t\t\t\t{ID: \"t1\", Status: types.TaskCompleted, ExecutorID: \"db.query\"},\n\t\t\t\t{ID: \"t2\", Status: types.TaskCompleted, ExecutorID: \"report.gen\"},\n\t\t\t},\n\t\t\tResults: []types.TaskResult{\n\t\t\t\t{TaskID: \"t1\", Success: true, Duration: 100},\n\t\t\t\t{TaskID: \"t2\", Success: true, Duration: 200},\n\t\t\t},\n\t\t\tDelivery: &types.DeliveryResult{\n\t\t\t\tRequestID: \"test-delivery-001\",\n\t\t\t\tContent: &types.DeliveryContent{\n\t\t\t\t\tSummary: \"Test delivery completed\",\n\t\t\t\t\tBody: \"# Test Delivery\\n\\nTest delivery body.\",\n\t\t\t\t},\n\t\t\t\tSuccess: true,\n\t\t\t},\n\t\t}\n\n\t\tresult := formatter.FormatExecutionSummary(exec)\n\n\t\tassert.Contains(t, result, \"## Execution Summary\")\n\t\tassert.Contains(t, result, \"exec-123\")\n\t\tassert.Contains(t, result, \"clock\")\n\t\tassert.Contains(t, result, \"completed\")\n\t\tassert.Contains(t, result, \"**Duration**:\")\n\t\tassert.Contains(t, result, \"## Inspiration (P0)\")\n\t\tassert.Contains(t, result, \"Morning analysis\")\n\t\tassert.Contains(t, result, \"## Goals (P1)\")\n\t\tassert.Contains(t, result, \"Review data\")\n\t\tassert.Contains(t, result, \"## Tasks (P2)\")\n\t\tassert.Contains(t, result, \"db.query\")\n\t\tassert.Contains(t, result, \"## Results (P3)\")\n\t\tassert.Contains(t, result, \"✓ t1\")\n\t\tassert.Contains(t, result, \"## Delivery (P4)\")\n\t\tassert.Contains(t, result, \"Test delivery completed\")\n\t})\n\n\tt.Run(\"formats execution with error\", func(t *testing.T) {\n\t\tstartTime := time.Now()\n\t\texec := &types.Execution{\n\t\t\tID: \"exec-456\",\n\t\t\tTriggerType: types.TriggerHuman,\n\t\t\tStatus: types.ExecFailed,\n\t\t\tStartTime: startTime,\n\t\t\tError: \"Task execution failed\",\n\t\t}\n\n\t\tresult := formatter.FormatExecutionSummary(exec)\n\n\t\tassert.Contains(t, result, \"exec-456\")\n\t\tassert.Contains(t, result, \"failed\")\n\t\tassert.Contains(t, result, \"**Error**: Task execution failed\")\n\t})\n\n\tt.Run(\"returns empty for nil execution\", func(t *testing.T) {\n\t\tresult := formatter.FormatExecutionSummary(nil)\n\t\tassert.Empty(t, result)\n\t})\n}\n\nfunc TestInputFormatterBuildMessages(t *testing.T) {\n\tformatter := standard.NewInputFormatter()\n\n\tt.Run(\"builds user message\", func(t *testing.T) {\n\t\tmsgs := formatter.BuildMessages(\"Hello, world!\")\n\n\t\trequire.Len(t, msgs, 1)\n\t\tassert.Equal(t, agentcontext.RoleUser, msgs[0].Role)\n\t\tassert.Equal(t, \"Hello, world!\", msgs[0].Content)\n\t})\n}\n\nfunc TestInputFormatterBuildMessagesWithSystem(t *testing.T) {\n\tformatter := standard.NewInputFormatter()\n\n\tt.Run(\"builds system and user messages\", func(t *testing.T) {\n\t\tmsgs := formatter.BuildMessagesWithSystem(\n\t\t\t\"You are a helpful assistant.\",\n\t\t\t\"What is the weather?\",\n\t\t)\n\n\t\trequire.Len(t, msgs, 2)\n\t\tassert.Equal(t, agentcontext.RoleSystem, msgs[0].Role)\n\t\tassert.Equal(t, \"You are a helpful assistant.\", msgs[0].Content)\n\t\tassert.Equal(t, agentcontext.RoleUser, msgs[1].Role)\n\t\tassert.Equal(t, \"What is the weather?\", msgs[1].Content)\n\t})\n}\n" }, { "path": "agent/robot/executor/standard/inspiration.go", "content": "package standard\n\nimport (\n\t\"fmt\"\n\t\"time\"\n\n\trobottypes \"github.com/yaoapp/yao/agent/robot/types\"\n)\n\n// RunInspiration executes P0: Inspiration phase\n// Calls the Inspiration Agent to generate daily briefing\n//\n// Input:\n// - ClockContext from trigger input or current time\n// - Robot identity and resources\n//\n// Output:\n// - InspirationReport with markdown content\nfunc (e *Executor) RunInspiration(ctx *robottypes.Context, exec *robottypes.Execution, _ interface{}) error {\n\t// Get robot for identity and resources\n\trobot := exec.GetRobot()\n\tif robot == nil {\n\t\treturn fmt.Errorf(\"robot not found in execution\")\n\t}\n\n\t// Update UI field with i18n\n\tlocale := getEffectiveLocale(robot, exec.Input)\n\te.updateUIFields(ctx, exec, \"\", getLocalizedMessage(locale, \"analyzing_context\"))\n\n\t// Build clock context from trigger input or current time\n\tvar clock *robottypes.ClockContext\n\tif exec.Input != nil && exec.Input.Clock != nil {\n\t\tclock = exec.Input.Clock\n\t} else {\n\t\tclock = robottypes.NewClockContext(time.Now(), \"\")\n\t}\n\n\t// Get agent ID for inspiration phase\n\tagentID := \"__yao.inspiration\" // default\n\tif robot.Config != nil && robot.Config.Resources != nil {\n\t\tagentID = robot.Config.Resources.GetPhaseAgent(robottypes.PhaseInspiration)\n\t}\n\n\t// Build prompt using InputFormatter\n\tformatter := NewInputFormatter()\n\tuserContent := formatter.FormatClockContext(clock, robot)\n\n\t// Add available resources - critical for generating achievable insights\n\tresourcesContent := formatter.FormatAvailableResources(robot)\n\tif resourcesContent != \"\" {\n\t\tuserContent += \"\\n\\n\" + resourcesContent\n\t}\n\n\t// Call agent\n\tcaller := NewAgentCaller()\n\tcaller.Connector = robot.LanguageModel\n\tresult, err := caller.CallWithMessages(ctx, agentID, userContent)\n\tif err != nil {\n\t\treturn fmt.Errorf(\"inspiration agent (%s) call failed: %w\", agentID, err)\n\t}\n\n\t// Parse response - get markdown content\n\tcontent := result.GetText()\n\tif content == \"\" {\n\t\treturn fmt.Errorf(\"inspiration agent (%s) returned empty response\", agentID)\n\t}\n\n\t// Build InspirationReport\n\texec.Inspiration = &robottypes.InspirationReport{\n\t\tClock: clock,\n\t\tContent: content,\n\t}\n\n\treturn nil\n}\n" }, { "path": "agent/robot/executor/standard/inspiration_test.go", "content": "package standard_test\n\nimport (\n\t\"context\"\n\t\"strings\"\n\t\"testing\"\n\t\"time\"\n\n\t\"github.com/stretchr/testify/assert\"\n\t\"github.com/stretchr/testify/require\"\n\t\"github.com/yaoapp/yao/agent/robot/executor/standard\"\n\t\"github.com/yaoapp/yao/agent/robot/types\"\n\t\"github.com/yaoapp/yao/agent/testutils\"\n)\n\n// ============================================================================\n// P0 Inspiration Phase Tests\n// ============================================================================\n\nfunc TestRunInspirationBasic(t *testing.T) {\n\tif testing.Short() {\n\t\tt.Skip(\"Skipping integration test\")\n\t}\n\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\tctx := types.NewContext(context.Background(), testAuth())\n\n\tt.Run(\"generates inspiration report with clock context\", func(t *testing.T) {\n\t\t// Create robot with inspiration agent configured\n\t\trobot := createTestRobot(t, \"robot.inspiration\")\n\n\t\t// Create executor and execution\n\t\texec := createTestExecution(robot, types.TriggerClock)\n\n\t\t// Run inspiration phase\n\t\te := standard.New()\n\t\terr := e.RunInspiration(ctx, exec, nil)\n\n\t\trequire.NoError(t, err)\n\t\trequire.NotNil(t, exec.Inspiration)\n\t\tassert.NotEmpty(t, exec.Inspiration.Content)\n\t\tassert.NotNil(t, exec.Inspiration.Clock)\n\t})\n\n\tt.Run(\"includes expected markdown sections\", func(t *testing.T) {\n\t\trobot := createTestRobot(t, \"robot.inspiration\")\n\t\texec := createTestExecution(robot, types.TriggerClock)\n\n\t\te := standard.New()\n\t\terr := e.RunInspiration(ctx, exec, nil)\n\n\t\trequire.NoError(t, err)\n\t\tcontent := exec.Inspiration.Content\n\n\t\t// Verify expected sections in markdown output\n\t\t// Note: LLM output is non-deterministic, so we check for likely sections\n\t\thasSection := strings.Contains(content, \"##\") ||\n\t\t\tstrings.Contains(content, \"Summary\") ||\n\t\t\tstrings.Contains(content, \"Highlight\") ||\n\t\t\tstrings.Contains(content, \"Recommend\")\n\n\t\tassert.True(t, hasSection, \"should contain markdown sections, got: %s\", content)\n\t})\n}\n\nfunc TestRunInspirationClockContext(t *testing.T) {\n\tif testing.Short() {\n\t\tt.Skip(\"Skipping integration test\")\n\t}\n\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\tctx := types.NewContext(context.Background(), testAuth())\n\n\tt.Run(\"uses clock from trigger input\", func(t *testing.T) {\n\t\trobot := createTestRobot(t, \"robot.inspiration\")\n\t\texec := createTestExecution(robot, types.TriggerClock)\n\n\t\t// Set specific clock context\n\t\tspecificTime := time.Date(2024, 12, 31, 17, 0, 0, 0, time.UTC)\n\t\texec.Input = &types.TriggerInput{\n\t\t\tClock: types.NewClockContext(specificTime, \"UTC\"),\n\t\t}\n\n\t\te := standard.New()\n\t\terr := e.RunInspiration(ctx, exec, nil)\n\n\t\trequire.NoError(t, err)\n\t\trequire.NotNil(t, exec.Inspiration.Clock)\n\n\t\t// Clock should match input\n\t\tassert.Equal(t, specificTime.Year(), exec.Inspiration.Clock.Year)\n\t\tassert.Equal(t, int(specificTime.Month()), exec.Inspiration.Clock.Month)\n\t\tassert.Equal(t, specificTime.Day(), exec.Inspiration.Clock.DayOfMonth)\n\t})\n\n\tt.Run(\"creates clock context when not provided\", func(t *testing.T) {\n\t\trobot := createTestRobot(t, \"robot.inspiration\")\n\t\texec := createTestExecution(robot, types.TriggerClock)\n\t\texec.Input = nil // No input\n\n\t\te := standard.New()\n\t\terr := e.RunInspiration(ctx, exec, nil)\n\n\t\trequire.NoError(t, err)\n\t\trequire.NotNil(t, exec.Inspiration.Clock)\n\n\t\t// Clock should be current time (approximately)\n\t\tnow := time.Now()\n\t\tassert.Equal(t, now.Year(), exec.Inspiration.Clock.Year)\n\t})\n}\n\nfunc TestRunInspirationRobotIdentity(t *testing.T) {\n\tif testing.Short() {\n\t\tt.Skip(\"Skipping integration test\")\n\t}\n\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\tctx := types.NewContext(context.Background(), testAuth())\n\n\tt.Run(\"robot identity influences output\", func(t *testing.T) {\n\t\t// Create robot with specific identity\n\t\trobot := &types.Robot{\n\t\t\tMemberID: \"test-robot-1\",\n\t\t\tTeamID: \"test-team-1\",\n\t\t\tDisplayName: \"Sales Assistant\",\n\t\t\tConfig: &types.Config{\n\t\t\t\tIdentity: &types.Identity{\n\t\t\t\t\tRole: \"Sales Assistant\",\n\t\t\t\t\tDuties: []string{\"Track sales metrics\", \"Prepare weekly reports\"},\n\t\t\t\t\tRules: []string{\"Focus on actionable insights\"},\n\t\t\t\t},\n\t\t\t\tResources: &types.Resources{\n\t\t\t\t\tPhases: map[types.Phase]string{\n\t\t\t\t\t\ttypes.PhaseInspiration: \"robot.inspiration\",\n\t\t\t\t\t},\n\t\t\t\t},\n\t\t\t},\n\t\t}\n\n\t\texec := createTestExecution(robot, types.TriggerClock)\n\n\t\te := standard.New()\n\t\terr := e.RunInspiration(ctx, exec, nil)\n\n\t\trequire.NoError(t, err)\n\t\tassert.NotEmpty(t, exec.Inspiration.Content)\n\n\t\t// The content should be influenced by robot identity\n\t\t// (exact content varies due to LLM non-determinism)\n\t})\n}\n\nfunc TestRunInspirationErrorHandling(t *testing.T) {\n\tif testing.Short() {\n\t\tt.Skip(\"Skipping integration test\")\n\t}\n\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\tctx := types.NewContext(context.Background(), testAuth())\n\n\tt.Run(\"returns error when robot is nil\", func(t *testing.T) {\n\t\texec := &types.Execution{\n\t\t\tID: \"test-exec-1\",\n\t\t\tTriggerType: types.TriggerClock,\n\t\t}\n\t\t// Don't set robot\n\n\t\te := standard.New()\n\t\terr := e.RunInspiration(ctx, exec, nil)\n\n\t\tassert.Error(t, err)\n\t\tassert.Contains(t, err.Error(), \"robot not found\")\n\t})\n\n\tt.Run(\"returns error when agent not found\", func(t *testing.T) {\n\t\trobot := &types.Robot{\n\t\t\tMemberID: \"test-robot-1\",\n\t\t\tTeamID: \"test-team-1\",\n\t\t\tConfig: &types.Config{\n\t\t\t\tIdentity: &types.Identity{Role: \"Test\"},\n\t\t\t\tResources: &types.Resources{\n\t\t\t\t\tPhases: map[types.Phase]string{\n\t\t\t\t\t\ttypes.PhaseInspiration: \"non.existent.agent\",\n\t\t\t\t\t},\n\t\t\t\t},\n\t\t\t},\n\t\t}\n\t\texec := createTestExecution(robot, types.TriggerClock)\n\n\t\te := standard.New()\n\t\terr := e.RunInspiration(ctx, exec, nil)\n\n\t\t// Real AgentCaller returns error for non-existent agent\n\t\tassert.Error(t, err)\n\t\tassert.Contains(t, err.Error(), \"call failed\")\n\t})\n}\n\nfunc TestRunInspirationWithDefaultAgent(t *testing.T) {\n\tif testing.Short() {\n\t\tt.Skip(\"Skipping integration test\")\n\t}\n\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\tctx := types.NewContext(context.Background(), testAuth())\n\n\tt.Run(\"uses default agent when not configured\", func(t *testing.T) {\n\t\trobot := &types.Robot{\n\t\t\tMemberID: \"test-robot-1\",\n\t\t\tTeamID: \"test-team-1\",\n\t\t\tConfig: &types.Config{\n\t\t\t\tIdentity: &types.Identity{Role: \"Test Robot\"},\n\t\t\t\t// No Resources configured - should use default __yao.inspiration\n\t\t\t},\n\t\t}\n\t\texec := createTestExecution(robot, types.TriggerClock)\n\n\t\te := standard.New()\n\t\terr := e.RunInspiration(ctx, exec, nil)\n\n\t\t// This will fail if __yao.inspiration doesn't exist\n\t\t// In test environment, we expect it to fail with \"agent not found\"\n\t\t// In production, it would use the default agent\n\t\tif err != nil {\n\t\t\tassert.Contains(t, err.Error(), \"call failed\")\n\t\t}\n\t})\n}\n\n// ============================================================================\n// InputFormatter Tests for P0\n// ============================================================================\n\nfunc TestInputFormatterClockContext(t *testing.T) {\n\tt.Run(\"formats clock context correctly\", func(t *testing.T) {\n\t\tformatter := standard.NewInputFormatter()\n\n\t\t// Create a specific clock context\n\t\tclock := types.NewClockContext(\n\t\t\ttime.Date(2024, 12, 31, 17, 30, 0, 0, time.UTC),\n\t\t\t\"UTC\",\n\t\t)\n\n\t\trobot := &types.Robot{\n\t\t\tConfig: &types.Config{\n\t\t\t\tIdentity: &types.Identity{\n\t\t\t\t\tRole: \"Sales Assistant\",\n\t\t\t\t\tDuties: []string{\"Track metrics\", \"Send reports\"},\n\t\t\t\t},\n\t\t\t},\n\t\t}\n\n\t\tcontent := formatter.FormatClockContext(clock, robot)\n\n\t\t// Verify time context\n\t\tassert.Contains(t, content, \"Current Time Context\")\n\t\tassert.Contains(t, content, \"2024\")\n\t\tassert.Contains(t, content, \"12\")\n\t\tassert.Contains(t, content, \"31\")\n\t\tassert.Contains(t, content, \"Tuesday\") // Dec 31, 2024 is Tuesday\n\n\t\t// Verify robot identity\n\t\tassert.Contains(t, content, \"Robot Identity\")\n\t\tassert.Contains(t, content, \"Sales Assistant\")\n\t\tassert.Contains(t, content, \"Track metrics\")\n\t})\n\n\tt.Run(\"handles nil clock\", func(t *testing.T) {\n\t\tformatter := standard.NewInputFormatter()\n\t\tcontent := formatter.FormatClockContext(nil, nil)\n\t\tassert.Empty(t, content)\n\t})\n\n\tt.Run(\"handles nil robot\", func(t *testing.T) {\n\t\tformatter := standard.NewInputFormatter()\n\t\tclock := types.NewClockContext(time.Now(), \"\")\n\t\tcontent := formatter.FormatClockContext(clock, nil)\n\n\t\t// Should have time context but no robot identity\n\t\tassert.Contains(t, content, \"Current Time Context\")\n\t\tassert.NotContains(t, content, \"Robot Identity\")\n\t})\n\n\tt.Run(\"includes time markers\", func(t *testing.T) {\n\t\tformatter := standard.NewInputFormatter()\n\n\t\t// Create a weekend + month start clock context\n\t\t// Jan 1, 2028 is Saturday (weekend + month start)\n\t\tclock := types.NewClockContext(\n\t\t\ttime.Date(2028, 1, 1, 10, 0, 0, 0, time.UTC),\n\t\t\t\"UTC\",\n\t\t)\n\n\t\tcontent := formatter.FormatClockContext(clock, nil)\n\n\t\tassert.Contains(t, content, \"Weekend\")\n\t\tassert.Contains(t, content, \"Month Start\")\n\t})\n}\n\n// ============================================================================\n// Helper Functions\n// ============================================================================\n\n// createTestRobot creates a test robot with specified inspiration agent\n// Includes available expert agents so the Inspiration Agent knows what resources are available\n//\n// Note: The agent IDs listed in Resources.Agents must exist in yao-dev-app/assistants/experts/\n// Current available experts: data-analyst, summarizer, text-writer, web-reader\nfunc createTestRobot(t *testing.T, agentID string) *types.Robot {\n\tt.Helper()\n\treturn &types.Robot{\n\t\tMemberID: \"test-robot-1\",\n\t\tTeamID: \"test-team-1\",\n\t\tDisplayName: \"Test Robot\",\n\t\tConfig: &types.Config{\n\t\t\tIdentity: &types.Identity{\n\t\t\t\tRole: \"Test Assistant\",\n\t\t\t\tDuties: []string{\"Testing\", \"Data Analysis\", \"Report Generation\"},\n\t\t\t},\n\t\t\tResources: &types.Resources{\n\t\t\t\tPhases: map[types.Phase]string{\n\t\t\t\t\ttypes.PhaseInspiration: agentID,\n\t\t\t\t},\n\t\t\t\t// Available expert agents that can be delegated to\n\t\t\t\t// These IDs correspond to assistants in yao-dev-app/assistants/experts/\n\t\t\t\tAgents: []string{\n\t\t\t\t\t\"experts.data-analyst\", // Data analysis and insights\n\t\t\t\t\t\"experts.summarizer\", // Content summarization\n\t\t\t\t\t\"experts.text-writer\", // Report and document generation\n\t\t\t\t\t\"experts.web-reader\", // Web content extraction\n\t\t\t\t},\n\t\t\t},\n\t\t\t// Knowledge base collections (if any)\n\t\t\tKB: &types.KB{\n\t\t\t\tCollections: []string{\"test-knowledge\"},\n\t\t\t},\n\t\t},\n\t}\n}\n\n// createTestExecution creates a test execution for a robot\nfunc createTestExecution(robot *types.Robot, trigger types.TriggerType) *types.Execution {\n\texec := &types.Execution{\n\t\tID: \"test-exec-1\",\n\t\tMemberID: robot.MemberID,\n\t\tTeamID: robot.TeamID,\n\t\tTriggerType: trigger,\n\t\tStartTime: time.Now(),\n\t\tStatus: types.ExecRunning,\n\t\tPhase: types.PhaseInspiration,\n\t\tInput: &types.TriggerInput{\n\t\t\tClock: types.NewClockContext(time.Now(), \"\"),\n\t\t},\n\t}\n\texec.SetRobot(robot)\n\treturn exec\n}\n" }, { "path": "agent/robot/executor/standard/learning.go", "content": "package standard\n\nimport (\n\trobottypes \"github.com/yaoapp/yao/agent/robot/types\"\n)\n\n// RunLearning executes P5: Learning phase\n// Extracts learnings and saves to knowledge base\n//\n// Input:\n// - Execution summary (all phases)\n//\n// Output:\n// - LearningEntry list with extracted knowledge\n//\n// Learning Types:\n// - LearnExecution: Execution patterns\n// - LearnTask: Task-specific insights\n// - LearnError: Error patterns for improvement\n//\n// TODO: Implement real learning extraction\nfunc (e *Executor) RunLearning(ctx *robottypes.Context, exec *robottypes.Execution, _ interface{}) error {\n\t// Get robot for locale\n\trobot := exec.GetRobot()\n\n\t// Update UI field with i18n\n\tlocale := getEffectiveLocale(robot, exec.Input)\n\te.updateUIFields(ctx, exec, \"\", getLocalizedMessage(locale, \"learning_from_exec\"))\n\n\te.simulateStreamDelay()\n\n\texec.Learning = []robottypes.LearningEntry{\n\t\t{\n\t\t\tType: robottypes.LearnExecution,\n\t\t\tContent: \"Completed daily tasks successfully\",\n\t\t},\n\t}\n\treturn nil\n}\n" }, { "path": "agent/robot/executor/standard/log.go", "content": "package standard\n\nimport (\n\t\"encoding/json\"\n\t\"fmt\"\n\t\"strings\"\n\n\tkunlog \"github.com/yaoapp/kun/log\"\n\t\"github.com/yaoapp/yao/config\"\n\n\t\"github.com/yaoapp/yao/agent/robot/logger\"\n\trobottypes \"github.com/yaoapp/yao/agent/robot/types\"\n)\n\nvar log = logger.New(\"exec\")\n\ntype execLogger struct {\n\trobot *robottypes.Robot\n\texecID string\n}\n\nfunc newExecLogger(robot *robottypes.Robot, execID string) *execLogger {\n\treturn &execLogger{robot: robot, execID: execID}\n}\n\nfunc (l *execLogger) robotID() string {\n\tif l.robot != nil {\n\t\treturn l.robot.MemberID\n\t}\n\treturn \"\"\n}\n\nfunc (l *execLogger) connector() string {\n\tif l.robot != nil {\n\t\treturn l.robot.LanguageModel\n\t}\n\treturn \"\"\n}\n\n// ---------------------------------------------------------------------------\n// P2: Task Overview\n// ---------------------------------------------------------------------------\n\nfunc (l *execLogger) logTaskOverview(tasks []robottypes.Task) {\n\tif config.IsDevelopment() {\n\t\tl.devTaskOverview(tasks)\n\t}\n\tkunlog.With(kunlog.F{\n\t\t\"robot_id\": l.robotID(),\n\t\t\"execution_id\": l.execID,\n\t\t\"phase\": \"tasks\",\n\t\t\"task_count\": len(tasks),\n\t\t\"language_model\": l.connector(),\n\t}).Info(\"P2 task overview: %d tasks generated\", len(tasks))\n}\n\nfunc (l *execLogger) devTaskOverview(tasks []robottypes.Task) {\n\tw := logger.Gray\n\th := logger.BoldCyan\n\tv := logger.White\n\tr := logger.Reset\n\n\tvar sb strings.Builder\n\tsb.WriteString(fmt.Sprintf(\"\\n%s%s%s\\n\", h, strings.Repeat(\"═\", 60), r))\n\tsb.WriteString(fmt.Sprintf(\"%s TASK OVERVIEW%s\\n\", h, r))\n\tsb.WriteString(fmt.Sprintf(\"%s%s%s\\n\", h, strings.Repeat(\"─\", 60), r))\n\tsb.WriteString(fmt.Sprintf(\"%s Robot: %s%s%s\\n\", w, v, l.robotID(), r))\n\tsb.WriteString(fmt.Sprintf(\"%s Exec: %s%s%s\\n\", w, v, l.execID, r))\n\tif l.connector() != \"\" {\n\t\tsb.WriteString(fmt.Sprintf(\"%s Model: %s%s%s\\n\", w, v, l.connector(), r))\n\t}\n\tsb.WriteString(fmt.Sprintf(\"%s%s%s\\n\", w, strings.Repeat(\"─\", 60), r))\n\tfor i, t := range tasks {\n\t\tdesc := t.Description\n\t\tif desc == \"\" && len(t.Messages) > 0 {\n\t\t\tif s, ok := t.Messages[0].GetContentAsString(); ok {\n\t\t\t\tdesc = s\n\t\t\t}\n\t\t}\n\t\tdesc = truncate(desc, 72)\n\t\tsb.WriteString(fmt.Sprintf(\"%s #%d %s%s%s [%s:%s]\\n\", w, i+1, v, t.ID, r, t.ExecutorType, t.ExecutorID))\n\t\tsb.WriteString(fmt.Sprintf(\"%s %s%s\\n\", w, desc, r))\n\t}\n\tsb.WriteString(fmt.Sprintf(\"%s%s%s\\n\", w, strings.Repeat(\"─\", 60), r))\n\tsb.WriteString(fmt.Sprintf(\"%s Total: %s%d tasks%s\\n\", w, v, len(tasks), r))\n\tsb.WriteString(fmt.Sprintf(\"%s%s%s\\n\", h, strings.Repeat(\"═\", 60), r))\n\n\tlogger.Raw(sb.String())\n}\n\n// ---------------------------------------------------------------------------\n// P3: Task Input\n// ---------------------------------------------------------------------------\n\nfunc (l *execLogger) logTaskInput(task *robottypes.Task, prompt string) {\n\tif config.IsDevelopment() {\n\t\tl.devTaskInput(task, prompt)\n\t}\n\tkunlog.With(kunlog.F{\n\t\t\"robot_id\": l.robotID(),\n\t\t\"execution_id\": l.execID,\n\t\t\"task_id\": task.ID,\n\t\t\"executor_type\": string(task.ExecutorType),\n\t\t\"executor_id\": task.ExecutorID,\n\t\t\"prompt_len\": len(prompt),\n\t\t\"language_model\": l.connector(),\n\t}).Info(\"Task input: %s [%s]\", task.ID, task.ExecutorID)\n}\n\nfunc (l *execLogger) devTaskInput(task *robottypes.Task, prompt string) {\n\tw := logger.Gray\n\tv := logger.White\n\tr := logger.Reset\n\n\tvar sb strings.Builder\n\tsb.WriteString(fmt.Sprintf(\"%s ▶ Task %s%s%s [%s:%s] Prompt: %d chars%s\\n\",\n\t\tw, v, task.ID, w, task.ExecutorType, task.ExecutorID, len(prompt), r))\n\n\tlogger.Raw(sb.String())\n}\n\n// ---------------------------------------------------------------------------\n// P3: Task Output\n// ---------------------------------------------------------------------------\n\nfunc (l *execLogger) logTaskOutput(task *robottypes.Task, result *robottypes.TaskResult) {\n\tif config.IsDevelopment() {\n\t\tl.devTaskOutput(task, result)\n\t}\n\n\tfields := kunlog.F{\n\t\t\"robot_id\": l.robotID(),\n\t\t\"execution_id\": l.execID,\n\t\t\"task_id\": result.TaskID,\n\t\t\"success\": result.Success,\n\t\t\"duration_ms\": result.Duration,\n\t\t\"language_model\": l.connector(),\n\t}\n\tif result.Output != nil {\n\t\tfields[\"output_type\"] = fmt.Sprintf(\"%T\", result.Output)\n\t\tfields[\"output_len\"] = outputLen(result.Output)\n\t}\n\tif result.Error != \"\" {\n\t\tfields[\"error\"] = result.Error\n\t}\n\tif result.Success {\n\t\tkunlog.With(fields).Info(\"Task completed: %s (%dms)\", result.TaskID, result.Duration)\n\t} else {\n\t\tkunlog.With(fields).Warn(\"Task failed: %s (%dms) %s\", result.TaskID, result.Duration, result.Error)\n\t}\n}\n\nfunc (l *execLogger) devTaskOutput(task *robottypes.Task, result *robottypes.TaskResult) {\n\tw := logger.Gray\n\tv := logger.White\n\tg := logger.BoldGreen\n\trd := logger.BoldRed\n\tr := logger.Reset\n\n\tvar sb strings.Builder\n\tif result.Success {\n\t\tsb.WriteString(fmt.Sprintf(\"%s ✓ %s%s%s completed %s(%dms)%s\\n\",\n\t\t\tg, v, result.TaskID, g, w, result.Duration, r))\n\t\tout := outputSummary(result.Output)\n\t\tif len(out) > 120 {\n\t\t\tout = out[:120] + \"...\"\n\t\t}\n\t\tsb.WriteString(fmt.Sprintf(\"%s Output: %s%s%s\\n\", w, v, out, r))\n\t} else {\n\t\tsb.WriteString(fmt.Sprintf(\"%s ✗ %s%s%s failed %s(%dms)%s\\n\",\n\t\t\trd, v, result.TaskID, rd, w, result.Duration, r))\n\t\tsb.WriteString(fmt.Sprintf(\"%s Error: %s%s%s\\n\", w, logger.Red, result.Error, r))\n\t}\n\n\tlogger.Raw(sb.String())\n}\n\n// ---------------------------------------------------------------------------\n// Agent Call\n// ---------------------------------------------------------------------------\n\nfunc (l *execLogger) logAgentCall(agentID string, result *CallResult) {\n\tif result == nil {\n\t\treturn\n\t}\n\tif config.IsDevelopment() {\n\t\tl.devAgentCall(agentID, result)\n\t}\n\n\tfields := kunlog.F{\n\t\t\"robot_id\": l.robotID(),\n\t\t\"execution_id\": l.execID,\n\t\t\"agent_id\": agentID,\n\t\t\"content_len\": len(result.Content),\n\t\t\"language_model\": l.connector(),\n\t}\n\tif result.Next != nil {\n\t\tfields[\"next_type\"] = fmt.Sprintf(\"%T\", result.Next)\n\t\tfields[\"next_len\"] = outputLen(result.Next)\n\t}\n\tkunlog.With(fields).Info(\"Agent call: %s (content=%d, next=%T)\", agentID, len(result.Content), result.Next)\n}\n\nfunc (l *execLogger) devAgentCall(agentID string, result *CallResult) {\n\tw := logger.Gray\n\tv := logger.White\n\tc := logger.Cyan\n\tr := logger.Reset\n\n\tnextInfo := \"—\"\n\tif result.Next != nil {\n\t\tnextInfo = fmt.Sprintf(\"%T (len=%d)\", result.Next, outputLen(result.Next))\n\t}\n\n\tvar sb strings.Builder\n\tsb.WriteString(fmt.Sprintf(\"%s → Agent(%s%s%s) Content: %s%d%s chars Next: %s%s%s\\n\",\n\t\tc, v, agentID, c, v, len(result.Content), w, v, nextInfo, r))\n\n\tlogger.Raw(sb.String())\n}\n\n// ---------------------------------------------------------------------------\n// Helpers\n// ---------------------------------------------------------------------------\n\nfunc (l *execLogger) prefix() string {\n\tif l.connector() != \"\" {\n\t\treturn fmt.Sprintf(\"[robot:%s|exec:%s|model:%s]\", l.robotID(), l.execID, l.connector())\n\t}\n\treturn fmt.Sprintf(\"[robot:%s|exec:%s]\", l.robotID(), l.execID)\n}\n\nfunc truncate(s string, maxLen int) string {\n\tif len(s) <= maxLen {\n\t\treturn s\n\t}\n\treturn s[:maxLen] + \"...\"\n}\n\nfunc indentText(s string, prefix string) string {\n\tlines := strings.Split(s, \"\\n\")\n\tfor i, line := range lines {\n\t\tlines[i] = prefix + line\n\t}\n\treturn strings.Join(lines, \"\\n\")\n}\n\nfunc outputSummary(v interface{}) string {\n\tif v == nil {\n\t\treturn \"\"\n\t}\n\tswitch val := v.(type) {\n\tcase string:\n\t\tif len(val) > 500 {\n\t\t\treturn fmt.Sprintf(\"string(len=%d) %s...\", len(val), val[:500])\n\t\t}\n\t\treturn fmt.Sprintf(\"string(len=%d) %s\", len(val), val)\n\tcase map[string]interface{}:\n\t\tkeys := make([]string, 0, len(val))\n\t\tfor k := range val {\n\t\t\tkeys = append(keys, k)\n\t\t}\n\t\treturn fmt.Sprintf(\"map{%s}\", strings.Join(keys, \", \"))\n\tdefault:\n\t\traw, err := json.Marshal(v)\n\t\tif err != nil {\n\t\t\treturn fmt.Sprintf(\"%T(marshal-error)\", v)\n\t\t}\n\t\ts := string(raw)\n\t\tif len(s) > 500 {\n\t\t\treturn fmt.Sprintf(\"%T(len=%d) %s...\", v, len(s), s[:500])\n\t\t}\n\t\treturn fmt.Sprintf(\"%T %s\", v, s)\n\t}\n}\n\nfunc outputLen(v interface{}) int {\n\tif v == nil {\n\t\treturn 0\n\t}\n\tswitch val := v.(type) {\n\tcase string:\n\t\treturn len(val)\n\tdefault:\n\t\traw, err := json.Marshal(v)\n\t\tif err != nil {\n\t\t\treturn 0\n\t\t}\n\t\treturn len(raw)\n\t}\n}\n" }, { "path": "agent/robot/executor/standard/resume_test.go", "content": "package standard_test\n\nimport (\n\t\"context\"\n\t\"testing\"\n\t\"time\"\n\n\t\"github.com/stretchr/testify/assert\"\n\t\"github.com/stretchr/testify/require\"\n\tagentcontext \"github.com/yaoapp/yao/agent/context\"\n\t\"github.com/yaoapp/yao/agent/robot/executor/standard\"\n\texecutortypes \"github.com/yaoapp/yao/agent/robot/executor/types\"\n\t\"github.com/yaoapp/yao/agent/robot/store\"\n\t\"github.com/yaoapp/yao/agent/robot/types\"\n\t\"github.com/yaoapp/yao/agent/testutils\"\n)\n\n// ============================================================================\n// Resume method tests (R1-R10)\n// ============================================================================\n\nfunc TestResume(t *testing.T) {\n\t// R1: Resume with empty execID returns error\n\tt.Run(\"R1: Resume with empty execID returns error\", func(t *testing.T) {\n\t\te := standard.New()\n\t\tctx := types.NewContext(context.Background(), testAuth())\n\n\t\terr := e.Resume(ctx, \"\", \"some reply\")\n\n\t\trequire.Error(t, err)\n\t\tassert.Contains(t, err.Error(), \"empty\")\n\t})\n\n\t// R2: Resume with non-existent execID returns error (requires DB)\n\tt.Run(\"R2: Resume with non-existent execID returns error\", func(t *testing.T) {\n\t\tif testing.Short() {\n\t\t\tt.Skip(\"Requires database\")\n\t\t}\n\n\t\ttestutils.Prepare(t)\n\t\tdefer testutils.Clean(t)\n\n\t\te := standard.New()\n\t\tctx := types.NewContext(context.Background(), testAuth())\n\n\t\terr := e.Resume(ctx, \"non-existent-exec-id-12345\", \"reply\")\n\n\t\trequire.Error(t, err)\n\t\tassert.Contains(t, err.Error(), \"execution not found\")\n\t})\n\n\t// R3: Resume with execution not in waiting status returns error (requires DB)\n\tt.Run(\"R3: Resume with execution not in waiting status returns error\", func(t *testing.T) {\n\t\tif testing.Short() {\n\t\t\tt.Skip(\"Requires database\")\n\t\t}\n\n\t\ttestutils.Prepare(t)\n\t\tdefer testutils.Clean(t)\n\n\t\tctx := types.NewContext(context.Background(), testAuth())\n\t\trobot := createResumeTestRobot(t)\n\t\texec := createResumeTestExecution(robot)\n\t\texec.Status = types.ExecRunning // Not waiting\n\n\t\trecord := store.FromExecution(exec)\n\t\trequire.NoError(t, store.NewExecutionStore().Save(ctx.Context, record))\n\n\t\t// Save robot to robot store for Resume to load\n\t\trobotRecord := store.FromRobot(robot)\n\t\trequire.NoError(t, store.NewRobotStore().Save(ctx.Context, robotRecord))\n\n\t\te := standard.New()\n\t\terr := e.Resume(ctx, exec.ID, \"reply\")\n\n\t\trequire.Error(t, err)\n\t\tassert.Contains(t, err.Error(), \"not in waiting status\")\n\t})\n\n\t// R4: Verify Resume loads execution from store (requires DB)\n\tt.Run(\"R4: Resume loads execution from store\", func(t *testing.T) {\n\t\tif testing.Short() {\n\t\t\tt.Skip(\"Requires database\")\n\t\t}\n\n\t\ttestutils.Prepare(t)\n\t\tdefer testutils.Clean(t)\n\n\t\tctx := types.NewContext(context.Background(), testAuth())\n\t\trobot := createResumeTestRobot(t)\n\t\texec := createSuspendedResumeTestExecution(robot)\n\n\t\texecStore := store.NewExecutionStore()\n\t\trobotStore := store.NewRobotStore()\n\n\t\trecord := store.FromExecution(exec)\n\t\trequire.NoError(t, execStore.Save(ctx.Context, record))\n\n\t\trobotRecord := store.FromRobot(robot)\n\t\trequire.NoError(t, robotStore.Save(ctx.Context, robotRecord))\n\n\t\te := standard.New()\n\t\terr := e.Resume(ctx, exec.ID, \"User provided answer\")\n\n\t\trequire.NoError(t, err)\n\n\t\t// Verify execution was loaded and completed\n\t\tloaded, err := execStore.Get(ctx.Context, exec.ID)\n\t\trequire.NoError(t, err)\n\t\trequire.NotNil(t, loaded)\n\t\tassert.Equal(t, types.ExecCompleted, loaded.Status)\n\t})\n\n\t// R5: Resume restores robot from execution record (requires DB)\n\tt.Run(\"R5: Resume restores robot from execution record\", func(t *testing.T) {\n\t\tif testing.Short() {\n\t\t\tt.Skip(\"Requires database\")\n\t\t}\n\n\t\ttestutils.Prepare(t)\n\t\tdefer testutils.Clean(t)\n\n\t\tctx := types.NewContext(context.Background(), testAuth())\n\t\trobot := createResumeTestRobot(t)\n\t\texec := createSuspendedResumeTestExecution(robot)\n\n\t\texecStore := store.NewExecutionStore()\n\t\trobotStore := store.NewRobotStore()\n\n\t\trecord := store.FromExecution(exec)\n\t\trequire.NoError(t, execStore.Save(ctx.Context, record))\n\n\t\trobotRecord := store.FromRobot(robot)\n\t\trequire.NoError(t, robotStore.Save(ctx.Context, robotRecord))\n\n\t\te := standard.New()\n\t\terr := e.Resume(ctx, exec.ID, \"Answer for the question\")\n\n\t\trequire.NoError(t, err)\n\t\t// If we get here without \"robot not found\", Resume successfully restored robot\n\t})\n\n\t// R6: Resume with __skip__ reply marks task as skipped (requires DB)\n\tt.Run(\"R6: Resume with __skip__ reply marks task as skipped\", func(t *testing.T) {\n\t\tif testing.Short() {\n\t\t\tt.Skip(\"Requires database\")\n\t\t}\n\n\t\ttestutils.Prepare(t)\n\t\tdefer testutils.Clean(t)\n\n\t\tctx := types.NewContext(context.Background(), testAuth())\n\t\trobot := createResumeTestRobot(t)\n\t\texec := createSuspendedResumeTestExecution(robot)\n\t\t// Ensure we have a task at index 0 that is waiting\n\t\texec.Tasks[0].Status = types.TaskWaitingInput\n\t\texec.ResumeContext = &types.ResumeContext{\n\t\t\tTaskIndex: 0,\n\t\t\tPreviousResults: []types.TaskResult{},\n\t\t}\n\n\t\texecStore := store.NewExecutionStore()\n\t\trobotStore := store.NewRobotStore()\n\n\t\trecord := store.FromExecution(exec)\n\t\trequire.NoError(t, execStore.Save(ctx.Context, record))\n\n\t\trobotRecord := store.FromRobot(robot)\n\t\trequire.NoError(t, robotStore.Save(ctx.Context, robotRecord))\n\n\t\te := standard.New()\n\t\terr := e.Resume(ctx, exec.ID, \"__skip__\")\n\n\t\trequire.NoError(t, err)\n\n\t\tloaded, err := execStore.Get(ctx.Context, exec.ID)\n\t\trequire.NoError(t, err)\n\t\trequire.NotNil(t, loaded)\n\t\trequire.Len(t, loaded.Tasks, 1)\n\t\tassert.Equal(t, types.TaskSkipped, loaded.Tasks[0].Status)\n\t})\n\n\t// R7: Resume sends ErrExecutionSuspended when execution suspends again (requires DB)\n\tt.Run(\"R7: Resume sends ErrExecutionSuspended when execution suspends again\", func(t *testing.T) {\n\t\tif testing.Short() {\n\t\t\tt.Skip(\"Requires database\")\n\t\t}\n\n\t\ttestutils.Prepare(t)\n\t\tdefer testutils.Clean(t)\n\n\t\t// Use robot-need-input assistant that suspends\n\t\tctx := types.NewContext(context.Background(), testAuth())\n\t\trobot := createResumeNeedInputRobot(t)\n\t\texec := createSuspendedResumeNeedInputExecution(robot)\n\n\t\texecStore := store.NewExecutionStore()\n\t\trobotStore := store.NewRobotStore()\n\n\t\trecord := store.FromExecution(exec)\n\t\trequire.NoError(t, execStore.Save(ctx.Context, record))\n\n\t\trobotRecord := store.FromRobot(robot)\n\t\trequire.NoError(t, robotStore.Save(ctx.Context, robotRecord))\n\n\t\te := standard.New()\n\t\terr := e.Resume(ctx, exec.ID, \"some reply\")\n\n\t\t// May return ErrExecutionSuspended if assistant suspends again\n\t\tif err != nil {\n\t\t\tassert.ErrorIs(t, err, types.ErrExecutionSuspended)\n\t\t}\n\t})\n\n\t// R8: Resume increments exec counter\n\tt.Run(\"R8: Resume increments exec counter\", func(t *testing.T) {\n\t\tif testing.Short() {\n\t\t\tt.Skip(\"Requires database\")\n\t\t}\n\n\t\ttestutils.Prepare(t)\n\t\tdefer testutils.Clean(t)\n\n\t\tctx := types.NewContext(context.Background(), testAuth())\n\t\trobot := createResumeTestRobot(t)\n\t\texec := createSuspendedResumeTestExecution(robot)\n\n\t\texecStore := store.NewExecutionStore()\n\t\trobotStore := store.NewRobotStore()\n\n\t\trecord := store.FromExecution(exec)\n\t\trequire.NoError(t, execStore.Save(ctx.Context, record))\n\n\t\trobotRecord := store.FromRobot(robot)\n\t\trequire.NoError(t, robotStore.Save(ctx.Context, robotRecord))\n\n\t\te := standard.New()\n\t\te.Reset()\n\n\t\tbefore := e.CurrentCount()\n\t\terr := e.Resume(ctx, exec.ID, \"answer\")\n\t\tafter := e.CurrentCount()\n\n\t\trequire.NoError(t, err)\n\t\t// During Resume, currentCount was incremented; after completion it's decremented\n\t\tassert.Equal(t, before, after, \"currentCount should be back to original after Resume completes\")\n\t})\n\n\t// R9: Resume decrements exec counter on completion\n\tt.Run(\"R9: Resume decrements exec counter on completion\", func(t *testing.T) {\n\t\tif testing.Short() {\n\t\t\tt.Skip(\"Requires database\")\n\t\t}\n\n\t\ttestutils.Prepare(t)\n\t\tdefer testutils.Clean(t)\n\n\t\tctx := types.NewContext(context.Background(), testAuth())\n\t\trobot := createResumeTestRobot(t)\n\t\texec := createSuspendedResumeTestExecution(robot)\n\n\t\texecStore := store.NewExecutionStore()\n\t\trobotStore := store.NewRobotStore()\n\n\t\trecord := store.FromExecution(exec)\n\t\trequire.NoError(t, execStore.Save(ctx.Context, record))\n\n\t\trobotRecord := store.FromRobot(robot)\n\t\trequire.NoError(t, robotStore.Save(ctx.Context, robotRecord))\n\n\t\te := standard.New()\n\t\te.Reset()\n\n\t\terr := e.Resume(ctx, exec.ID, \"reply\")\n\t\trequire.NoError(t, err)\n\n\t\t// After Resume completes, currentCount should be 0 (no leak)\n\t\tassert.Equal(t, 0, e.CurrentCount())\n\t})\n\n\t// R10: Resume with nil context returns error\n\tt.Run(\"R10: Resume with nil context returns error\", func(t *testing.T) {\n\t\te := standard.NewWithConfig(executortypes.Config{SkipPersistence: true})\n\n\t\terr := e.Resume(nil, \"some-exec-id\", \"reply\")\n\n\t\trequire.Error(t, err)\n\t\tassert.Contains(t, err.Error(), \"context\")\n\t})\n}\n\n// ============================================================================\n// Helpers for Resume tests\n// ============================================================================\n\nfunc createResumeTestRobot(t *testing.T) *types.Robot {\n\tt.Helper()\n\treturn &types.Robot{\n\t\tMemberID: \"test-robot-resume\",\n\t\tTeamID: \"test-team-1\",\n\t\tDisplayName: \"Resume Test Robot\",\n\t\tSystemPrompt: \"You are a helpful assistant.\",\n\t\tConfig: &types.Config{\n\t\t\tIdentity: &types.Identity{\n\t\t\t\tRole: \"Test\",\n\t\t\t\tDuties: []string{\"Execute tasks\"},\n\t\t\t},\n\t\t\tResources: &types.Resources{\n\t\t\t\tPhases: map[types.Phase]string{\n\t\t\t\t\ttypes.PhaseDelivery: \"robot.delivery\",\n\t\t\t\t\ttypes.PhaseLearning: \"robot.learning\",\n\t\t\t\t},\n\t\t\t\tAgents: []string{\"experts.text-writer\"},\n\t\t\t},\n\t\t\tQuota: &types.Quota{Max: 5},\n\t\t},\n\t}\n}\n\nfunc createResumeTestExecution(robot *types.Robot) *types.Execution {\n\texec := &types.Execution{\n\t\tID: \"test-exec-resume-1\",\n\t\tMemberID: robot.MemberID,\n\t\tTeamID: robot.TeamID,\n\t\tTriggerType: types.TriggerClock,\n\t\tStartTime: time.Now(),\n\t\tStatus: types.ExecRunning,\n\t\tPhase: types.PhaseRun,\n\t\tGoals: &types.Goals{Content: \"## Goals\\n\\n1. Test resume\"},\n\t\tTasks: []types.Task{\n\t\t\t{\n\t\t\t\tID: \"task-001\",\n\t\t\t\tExecutorType: types.ExecutorAssistant,\n\t\t\t\tExecutorID: \"experts.text-writer\",\n\t\t\t\tMessages: []agentcontext.Message{\n\t\t\t\t\t{Role: agentcontext.RoleUser, Content: \"Write 'hello'\"},\n\t\t\t\t},\n\t\t\t\tOrder: 0,\n\t\t\t\tStatus: types.TaskPending,\n\t\t\t},\n\t\t},\n\t\tChatID: \"robot_test-robot-resume_test-exec-resume-1\",\n\t}\n\texec.SetRobot(robot)\n\treturn exec\n}\n\nfunc createSuspendedResumeTestExecution(robot *types.Robot) *types.Execution {\n\texec := createResumeTestExecution(robot)\n\texec.Status = types.ExecWaiting\n\texec.WaitingTaskID = \"task-001\"\n\texec.WaitingQuestion = \"What should we do?\"\n\tnow := time.Now()\n\texec.WaitingSince = &now\n\texec.ResumeContext = &types.ResumeContext{\n\t\tTaskIndex: 0,\n\t\tPreviousResults: []types.TaskResult{},\n\t}\n\texec.Tasks[0].Status = types.TaskWaitingInput\n\treturn exec\n}\n\nfunc createResumeNeedInputRobot(t *testing.T) *types.Robot {\n\tt.Helper()\n\treturn &types.Robot{\n\t\tMemberID: \"test-robot-resume-need-input\",\n\t\tTeamID: \"test-team-1\",\n\t\tDisplayName: \"Resume Need Input Robot\",\n\t\tSystemPrompt: \"You are a helpful assistant.\",\n\t\tConfig: &types.Config{\n\t\t\tIdentity: &types.Identity{\n\t\t\t\tRole: \"Test\",\n\t\t\t\tDuties: []string{\"Execute tasks\"},\n\t\t\t},\n\t\t\tResources: &types.Resources{\n\t\t\t\tPhases: map[types.Phase]string{\n\t\t\t\t\ttypes.PhaseDelivery: \"robot.delivery\",\n\t\t\t\t\ttypes.PhaseLearning: \"robot.learning\",\n\t\t\t\t},\n\t\t\t\tAgents: []string{\"tests.robot-need-input\"},\n\t\t\t},\n\t\t\tQuota: &types.Quota{Max: 5},\n\t\t},\n\t}\n}\n\nfunc createSuspendedResumeNeedInputExecution(robot *types.Robot) *types.Execution {\n\texec := &types.Execution{\n\t\tID: \"test-exec-resume-need-input-1\",\n\t\tMemberID: robot.MemberID,\n\t\tTeamID: robot.TeamID,\n\t\tTriggerType: types.TriggerClock,\n\t\tStartTime: time.Now(),\n\t\tStatus: types.ExecWaiting,\n\t\tPhase: types.PhaseRun,\n\t\tGoals: &types.Goals{Content: \"## Goals\\n\\n1. Test need input\"},\n\t\tTasks: []types.Task{\n\t\t\t{\n\t\t\t\tID: \"task-001\",\n\t\t\t\tExecutorType: types.ExecutorAssistant,\n\t\t\t\tExecutorID: \"tests.robot-need-input\",\n\t\t\t\tMessages: []agentcontext.Message{\n\t\t\t\t\t{Role: agentcontext.RoleUser, Content: \"Need input test\"},\n\t\t\t\t},\n\t\t\t\tOrder: 0,\n\t\t\t\tStatus: types.TaskWaitingInput,\n\t\t\t},\n\t\t},\n\t\tChatID: \"robot_test-robot-resume-need-input_test-exec-resume-need-input-1\",\n\t\tWaitingTaskID: \"task-001\",\n\t\tWaitingQuestion: \"What period?\",\n\t\tResumeContext: &types.ResumeContext{\n\t\t\tTaskIndex: 0,\n\t\t\tPreviousResults: []types.TaskResult{},\n\t\t},\n\t}\n\tnow := time.Now()\n\texec.WaitingSince = &now\n\texec.SetRobot(robot)\n\treturn exec\n}\n" }, { "path": "agent/robot/executor/standard/run.go", "content": "package standard\n\nimport (\n\t\"fmt\"\n\t\"time\"\n\n\trobotevents \"github.com/yaoapp/yao/agent/robot/events\"\n\trobottypes \"github.com/yaoapp/yao/agent/robot/types\"\n\t\"github.com/yaoapp/yao/event\"\n)\n\n// RunConfig configures P3 execution behavior\ntype RunConfig struct {\n\t// ContinueOnFailure continues to next task even if current task fails.\n\t// V2 default: true — the Robot is an orchestrator, not a judge.\n\t// Failed tasks are recorded and evaluated by the Delivery Agent.\n\tContinueOnFailure bool\n}\n\n// DefaultRunConfig returns the default P3 configuration\nfunc DefaultRunConfig() *RunConfig {\n\treturn &RunConfig{\n\t\tContinueOnFailure: true,\n\t}\n}\n\n// RunExecution executes P3: Run phase\n// Executes each task using the appropriate executor (Assistant, MCP, Process).\n//\n// V2 simplified flow: single call per task, no validation loop.\n// Success is determined by whether the call itself succeeds (no error).\n// The Delivery Agent (P4) evaluates overall quality using expected_output.\n//\n// Supports resume: if exec.ResumeContext is set, execution starts from the\n// suspended task index with previously completed results restored.\n//\n// Returns ErrExecutionSuspended if a task signals it needs human input.\nfunc (e *Executor) RunExecution(ctx *robottypes.Context, exec *robottypes.Execution, data interface{}) error {\n\trobot := exec.GetRobot()\n\tif robot == nil {\n\t\treturn fmt.Errorf(\"robot not found in execution\")\n\t}\n\n\tif len(exec.Tasks) == 0 {\n\t\treturn fmt.Errorf(\"no tasks to execute\")\n\t}\n\n\t// Get run configuration from data or use default\n\tvar config *RunConfig\n\tif cfg, ok := data.(*RunConfig); ok && cfg != nil {\n\t\tconfig = cfg\n\t} else {\n\t\tconfig = DefaultRunConfig()\n\t}\n\n\t// Determine locale for UI messages\n\tlocale := getEffectiveLocale(robot, exec.Input)\n\n\t// Determine start index and restore results from resume context\n\tstartIndex := 0\n\tif exec.ResumeContext != nil {\n\t\tstartIndex = exec.ResumeContext.TaskIndex\n\t\texec.Results = exec.ResumeContext.PreviousResults\n\t} else {\n\t\texec.Results = make([]robottypes.TaskResult, 0, len(exec.Tasks))\n\t}\n\n\t// Create task runner with execution-level chatID (§8.4)\n\trunner := NewRunner(ctx, robot, config, exec.ChatID, exec.ID)\n\n\t// Execute tasks sequentially from startIndex\n\tfor i := startIndex; i < len(exec.Tasks); i++ {\n\t\ttask := &exec.Tasks[i]\n\n\t\t// Update current state for tracking\n\t\texec.Current = &robottypes.CurrentState{\n\t\t\tTask: task,\n\t\t\tTaskIndex: i,\n\t\t\tProgress: fmt.Sprintf(\"%d/%d tasks\", i+1, len(exec.Tasks)),\n\t\t}\n\n\t\t// Update UI field with current task description (i18n)\n\t\ttaskName := formatTaskProgressName(task, i, len(exec.Tasks), locale)\n\t\te.updateUIFields(ctx, exec, \"\", taskName)\n\n\t\t// Mark task as running\n\t\ttask.Status = robottypes.TaskRunning\n\t\tnow := time.Now()\n\t\ttask.StartTime = &now\n\n\t\t// Persist running state to database\n\t\te.updateTasksState(ctx, exec)\n\n\t\t// Build task context with previous results\n\t\ttaskCtx := runner.BuildTaskContext(exec, i)\n\n\t\t// Execute task (single call, no validation loop)\n\t\tresult := runner.ExecuteTask(task, taskCtx)\n\n\t\t// Task needs human input — suspend execution without recording a half-result\n\t\tif result.NeedInput {\n\t\t\treturn e.Suspend(ctx, exec, i, result.InputQuestion)\n\t\t}\n\n\t\t// Update task status based on result\n\t\tendTime := time.Now()\n\t\ttask.EndTime = &endTime\n\n\t\tif result.Success {\n\t\t\ttask.Status = robottypes.TaskCompleted\n\t\t\tevent.Push(ctx.Context, robotevents.TaskCompleted, robotevents.TaskPayload{\n\t\t\t\tExecutionID: exec.ID,\n\t\t\t\tMemberID: exec.MemberID,\n\t\t\t\tTeamID: exec.TeamID,\n\t\t\t\tTaskID: task.ID,\n\t\t\t\tChatID: exec.ChatID,\n\t\t\t})\n\t\t} else {\n\t\t\ttask.Status = robottypes.TaskFailed\n\t\t\tevent.Push(ctx.Context, robotevents.TaskFailed, robotevents.TaskPayload{\n\t\t\t\tExecutionID: exec.ID,\n\t\t\t\tMemberID: exec.MemberID,\n\t\t\t\tTeamID: exec.TeamID,\n\t\t\t\tTaskID: task.ID,\n\t\t\t\tError: result.Error,\n\t\t\t\tChatID: exec.ChatID,\n\t\t\t})\n\t\t}\n\n\t\t// Store result\n\t\texec.Results = append(exec.Results, *result)\n\n\t\t// Persist completed/failed state to database\n\t\te.updateTasksState(ctx, exec)\n\n\t\t// Check if we should continue on failure\n\t\tif !result.Success && !config.ContinueOnFailure {\n\t\t\t// Mark remaining tasks as skipped\n\t\t\tfor j := i + 1; j < len(exec.Tasks); j++ {\n\t\t\t\texec.Tasks[j].Status = robottypes.TaskSkipped\n\t\t\t}\n\t\t\t// Persist skipped state to database\n\t\t\te.updateTasksState(ctx, exec)\n\t\t\treturn fmt.Errorf(\"task %s failed: %s\", task.ID, result.Error)\n\t\t}\n\t}\n\n\t// Clear current state and resume context after successful completion\n\texec.Current = nil\n\texec.ResumeContext = nil\n\n\treturn nil\n}\n\n// formatTaskProgressName formats a progress name for the current task (used for UI with i18n)\nfunc formatTaskProgressName(task *robottypes.Task, index int, total int, locale string) string {\n\ttaskPrefix := getLocalizedMessage(locale, \"task_prefix\")\n\tprefix := fmt.Sprintf(\"%s %d/%d: \", taskPrefix, index+1, total)\n\n\t// Priority 1: Use Description field if available\n\tif task.Description != \"\" {\n\t\tdesc := task.Description\n\t\tif len(desc) > 80 {\n\t\t\tdesc = desc[:80] + \"...\"\n\t\t}\n\t\treturn prefix + desc\n\t}\n\n\t// Priority 2: Try to get description from first message\n\tif len(task.Messages) > 0 {\n\t\tif content, ok := task.Messages[0].GetContentAsString(); ok && content != \"\" {\n\t\t\t// Truncate if too long\n\t\t\tif len(content) > 80 {\n\t\t\t\tcontent = content[:80] + \"...\"\n\t\t\t}\n\t\t\treturn prefix + content\n\t\t}\n\t}\n\n\t// Fallback to executor info\n\treturn prefix + string(task.ExecutorType) + \":\" + task.ExecutorID\n}\n" }, { "path": "agent/robot/executor/standard/run_test.go", "content": "package standard_test\n\nimport (\n\t\"context\"\n\t\"testing\"\n\t\"time\"\n\n\t\"github.com/stretchr/testify/assert\"\n\t\"github.com/stretchr/testify/require\"\n\tagentcontext \"github.com/yaoapp/yao/agent/context\"\n\t\"github.com/yaoapp/yao/agent/robot/executor/standard\"\n\t\"github.com/yaoapp/yao/agent/robot/types\"\n\t\"github.com/yaoapp/yao/agent/testutils\"\n)\n\n// ============================================================================\n// P3 Run Phase Tests - RunExecution\n// ============================================================================\n\nfunc TestRunExecutionBasic(t *testing.T) {\n\tif testing.Short() {\n\t\tt.Skip(\"Skipping integration test\")\n\t}\n\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\tctx := types.NewContext(context.Background(), testAuth())\n\n\tt.Run(\"executes single task successfully\", func(t *testing.T) {\n\t\trobot := createRunTestRobot(t)\n\t\texec := createRunTestExecution(robot)\n\n\t\t// Pre-built task (simulating P2 output)\n\t\texec.Tasks = []types.Task{\n\t\t\t{\n\t\t\t\tID: \"task-001\",\n\t\t\t\tExecutorType: types.ExecutorAssistant,\n\t\t\t\tExecutorID: \"experts.text-writer\",\n\t\t\t\tMessages: []agentcontext.Message{\n\t\t\t\t\t{Role: agentcontext.RoleUser, Content: \"Write a short greeting message for a company newsletter. Keep it under 50 words.\"},\n\t\t\t\t},\n\t\t\t\tExpectedOutput: \"A friendly greeting message suitable for a newsletter\",\n\t\t\t\tOrder: 0,\n\t\t\t\tStatus: types.TaskPending,\n\t\t\t},\n\t\t}\n\n\t\te := standard.New()\n\t\terr := e.RunExecution(ctx, exec, nil)\n\n\t\trequire.NoError(t, err)\n\t\trequire.Len(t, exec.Results, 1)\n\n\t\tresult := exec.Results[0]\n\t\tassert.Equal(t, \"task-001\", result.TaskID)\n\t\tassert.True(t, result.Success, \"task should succeed\")\n\t\tassert.NotNil(t, result.Output, \"should have output\")\n\t\tassert.Greater(t, result.Duration, int64(0), \"should have duration\")\n\n\t\t// Task status should be updated\n\t\tassert.Equal(t, types.TaskCompleted, exec.Tasks[0].Status)\n\t\tassert.NotNil(t, exec.Tasks[0].StartTime)\n\t\tassert.NotNil(t, exec.Tasks[0].EndTime)\n\t})\n\n\tt.Run(\"executes multiple tasks in order\", func(t *testing.T) {\n\t\trobot := createRunTestRobot(t)\n\t\texec := createRunTestExecution(robot)\n\n\t\t// Multiple tasks that depend on each other\n\t\texec.Tasks = []types.Task{\n\t\t\t{\n\t\t\t\tID: \"task-001\",\n\t\t\t\tExecutorType: types.ExecutorAssistant,\n\t\t\t\tExecutorID: \"experts.data-analyst\",\n\t\t\t\tMessages: []agentcontext.Message{\n\t\t\t\t\t{Role: agentcontext.RoleUser, Content: \"Analyze this data: Sales Q1: $100K, Q2: $150K, Q3: $120K, Q4: $180K. Calculate the total and average.\"},\n\t\t\t\t},\n\t\t\t\tExpectedOutput: \"JSON with total and average sales figures\",\n\t\t\t\tValidationRules: []string{\n\t\t\t\t\t\"output must be valid JSON\",\n\t\t\t\t},\n\t\t\t\tOrder: 0,\n\t\t\t\tStatus: types.TaskPending,\n\t\t\t},\n\t\t\t{\n\t\t\t\tID: \"task-002\",\n\t\t\t\tExecutorType: types.ExecutorAssistant,\n\t\t\t\tExecutorID: \"experts.summarizer\",\n\t\t\t\tMessages: []agentcontext.Message{\n\t\t\t\t\t{Role: agentcontext.RoleUser, Content: \"Summarize the key findings from the previous analysis in 2-3 sentences.\"},\n\t\t\t\t},\n\t\t\t\tExpectedOutput: \"A brief summary of the sales analysis\",\n\t\t\t\tOrder: 1,\n\t\t\t\tStatus: types.TaskPending,\n\t\t\t},\n\t\t}\n\n\t\te := standard.New()\n\t\terr := e.RunExecution(ctx, exec, nil)\n\n\t\trequire.NoError(t, err)\n\t\trequire.Len(t, exec.Results, 2)\n\n\t\t// Both tasks should complete\n\t\tassert.True(t, exec.Results[0].Success, \"first task should succeed\")\n\t\tassert.True(t, exec.Results[1].Success, \"second task should succeed\")\n\n\t\t// Second task should have access to first task's result (via context)\n\t\tassert.Equal(t, types.TaskCompleted, exec.Tasks[0].Status)\n\t\tassert.Equal(t, types.TaskCompleted, exec.Tasks[1].Status)\n\n\t\tt.Logf(\"Task 1 output: %v\", exec.Results[0].Output)\n\t\tt.Logf(\"Task 2 output: %v\", exec.Results[1].Output)\n\t})\n\n\tt.Run(\"passes previous results as context to subsequent tasks\", func(t *testing.T) {\n\t\trobot := createRunTestRobot(t)\n\t\texec := createRunTestExecution(robot)\n\n\t\t// First task generates data, second task uses it\n\t\texec.Tasks = []types.Task{\n\t\t\t{\n\t\t\t\tID: \"task-001\",\n\t\t\t\tExecutorType: types.ExecutorAssistant,\n\t\t\t\tExecutorID: \"experts.text-writer\",\n\t\t\t\tMessages: []agentcontext.Message{\n\t\t\t\t\t{Role: agentcontext.RoleUser, Content: \"Generate a list of 3 product names for a tech company. Output as JSON array.\"},\n\t\t\t\t},\n\t\t\t\tExpectedOutput: \"JSON array with 3 product names\",\n\t\t\t\tOrder: 0,\n\t\t\t\tStatus: types.TaskPending,\n\t\t\t},\n\t\t\t{\n\t\t\t\tID: \"task-002\",\n\t\t\t\tExecutorType: types.ExecutorAssistant,\n\t\t\t\tExecutorID: \"experts.text-writer\",\n\t\t\t\tMessages: []agentcontext.Message{\n\t\t\t\t\t{Role: agentcontext.RoleUser, Content: \"Using the product names from the previous task, write a one-line tagline for each product.\"},\n\t\t\t\t},\n\t\t\t\tExpectedOutput: \"Taglines for each product\",\n\t\t\t\tOrder: 1,\n\t\t\t\tStatus: types.TaskPending,\n\t\t\t},\n\t\t}\n\n\t\te := standard.New()\n\t\terr := e.RunExecution(ctx, exec, nil)\n\n\t\trequire.NoError(t, err)\n\t\trequire.Len(t, exec.Results, 2)\n\n\t\t// Both should succeed\n\t\tassert.True(t, exec.Results[0].Success)\n\t\tassert.True(t, exec.Results[1].Success)\n\n\t\t// Second task output should reference products from first task\n\t\tt.Logf(\"Task 1 (products): %v\", exec.Results[0].Output)\n\t\tt.Logf(\"Task 2 (taglines): %v\", exec.Results[1].Output)\n\t})\n}\n\nfunc TestRunExecutionTaskStatus(t *testing.T) {\n\tif testing.Short() {\n\t\tt.Skip(\"Skipping integration test\")\n\t}\n\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\tctx := types.NewContext(context.Background(), testAuth())\n\n\tt.Run(\"updates task status during execution\", func(t *testing.T) {\n\t\trobot := createRunTestRobot(t)\n\t\texec := createRunTestExecution(robot)\n\n\t\texec.Tasks = []types.Task{\n\t\t\t{\n\t\t\t\tID: \"task-001\",\n\t\t\t\tExecutorType: types.ExecutorAssistant,\n\t\t\t\tExecutorID: \"experts.text-writer\",\n\t\t\t\tMessages: []agentcontext.Message{\n\t\t\t\t\t{Role: agentcontext.RoleUser, Content: \"Say 'Hello World'\"},\n\t\t\t\t},\n\t\t\t\tOrder: 0,\n\t\t\t\tStatus: types.TaskPending,\n\t\t\t},\n\t\t}\n\n\t\t// Verify initial status\n\t\tassert.Equal(t, types.TaskPending, exec.Tasks[0].Status)\n\n\t\te := standard.New()\n\t\terr := e.RunExecution(ctx, exec, nil)\n\n\t\trequire.NoError(t, err)\n\n\t\t// Verify final status\n\t\tassert.Equal(t, types.TaskCompleted, exec.Tasks[0].Status)\n\t\tassert.NotNil(t, exec.Tasks[0].StartTime)\n\t\tassert.NotNil(t, exec.Tasks[0].EndTime)\n\t})\n\n\tt.Run(\"marks remaining tasks as skipped on failure with ContinueOnFailure=false\", func(t *testing.T) {\n\t\trobot := createRunTestRobot(t)\n\t\texec := createRunTestExecution(robot)\n\n\t\texec.Tasks = []types.Task{\n\t\t\t{\n\t\t\t\tID: \"task-001\",\n\t\t\t\tExecutorType: types.ExecutorAssistant,\n\t\t\t\tExecutorID: \"non.existent.assistant.xyz123\",\n\t\t\t\tMessages: []agentcontext.Message{\n\t\t\t\t\t{Role: agentcontext.RoleUser, Content: \"This will fail\"},\n\t\t\t\t},\n\t\t\t\tOrder: 0,\n\t\t\t\tStatus: types.TaskPending,\n\t\t\t},\n\t\t\t{\n\t\t\t\tID: \"task-002\",\n\t\t\t\tExecutorType: types.ExecutorAssistant,\n\t\t\t\tExecutorID: \"experts.text-writer\",\n\t\t\t\tMessages: []agentcontext.Message{\n\t\t\t\t\t{Role: agentcontext.RoleUser, Content: \"Write another greeting\"},\n\t\t\t\t},\n\t\t\t\tOrder: 1,\n\t\t\t\tStatus: types.TaskPending,\n\t\t\t},\n\t\t\t{\n\t\t\t\tID: \"task-003\",\n\t\t\t\tExecutorType: types.ExecutorAssistant,\n\t\t\t\tExecutorID: \"experts.text-writer\",\n\t\t\t\tMessages: []agentcontext.Message{\n\t\t\t\t\t{Role: agentcontext.RoleUser, Content: \"Write yet another greeting\"},\n\t\t\t\t},\n\t\t\t\tOrder: 2,\n\t\t\t\tStatus: types.TaskPending,\n\t\t\t},\n\t\t}\n\n\t\tconfig := &standard.RunConfig{ContinueOnFailure: false}\n\t\te := standard.New()\n\t\terr := e.RunExecution(ctx, exec, config)\n\n\t\tassert.Error(t, err)\n\t\tassert.Contains(t, err.Error(), \"task-001\")\n\n\t\tassert.Equal(t, types.TaskFailed, exec.Tasks[0].Status)\n\t\tassert.Equal(t, types.TaskSkipped, exec.Tasks[1].Status)\n\t\tassert.Equal(t, types.TaskSkipped, exec.Tasks[2].Status)\n\t})\n\n\tt.Run(\"continues on failure with default V2 config\", func(t *testing.T) {\n\t\trobot := createRunTestRobot(t)\n\t\texec := createRunTestExecution(robot)\n\n\t\texec.Tasks = []types.Task{\n\t\t\t{\n\t\t\t\tID: \"task-001\",\n\t\t\t\tExecutorType: types.ExecutorAssistant,\n\t\t\t\tExecutorID: \"non.existent.assistant.xyz123\",\n\t\t\t\tMessages: []agentcontext.Message{\n\t\t\t\t\t{Role: agentcontext.RoleUser, Content: \"This will fail\"},\n\t\t\t\t},\n\t\t\t\tOrder: 0,\n\t\t\t\tStatus: types.TaskPending,\n\t\t\t},\n\t\t\t{\n\t\t\t\tID: \"task-002\",\n\t\t\t\tExecutorType: types.ExecutorAssistant,\n\t\t\t\tExecutorID: \"experts.text-writer\",\n\t\t\t\tMessages: []agentcontext.Message{\n\t\t\t\t\t{Role: agentcontext.RoleUser, Content: \"Say hello\"},\n\t\t\t\t},\n\t\t\t\tOrder: 1,\n\t\t\t\tStatus: types.TaskPending,\n\t\t\t},\n\t\t}\n\n\t\te := standard.New()\n\t\terr := e.RunExecution(ctx, exec, nil)\n\n\t\trequire.NoError(t, err, \"V2 default ContinueOnFailure=true should not return error\")\n\n\t\tassert.Equal(t, types.TaskFailed, exec.Tasks[0].Status)\n\t\tassert.Equal(t, types.TaskCompleted, exec.Tasks[1].Status)\n\t\tassert.Len(t, exec.Results, 2)\n\t\tassert.False(t, exec.Results[0].Success)\n\t\tassert.True(t, exec.Results[1].Success)\n\t})\n}\n\nfunc TestRunExecutionErrorHandling(t *testing.T) {\n\tif testing.Short() {\n\t\tt.Skip(\"Skipping integration test\")\n\t}\n\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\tctx := types.NewContext(context.Background(), testAuth())\n\n\tt.Run(\"returns error when robot is nil\", func(t *testing.T) {\n\t\texec := &types.Execution{\n\t\t\tID: \"test-exec-1\",\n\t\t\tTriggerType: types.TriggerClock,\n\t\t\tTasks: []types.Task{\n\t\t\t\t{ID: \"task-001\", ExecutorID: \"test\"},\n\t\t\t},\n\t\t}\n\t\t// Don't set robot\n\n\t\te := standard.New()\n\t\terr := e.RunExecution(ctx, exec, nil)\n\n\t\tassert.Error(t, err)\n\t\tassert.Contains(t, err.Error(), \"robot not found\")\n\t})\n\n\tt.Run(\"returns error when no tasks\", func(t *testing.T) {\n\t\trobot := createRunTestRobot(t)\n\t\texec := createRunTestExecution(robot)\n\t\texec.Tasks = []types.Task{} // Empty\n\n\t\te := standard.New()\n\t\terr := e.RunExecution(ctx, exec, nil)\n\n\t\tassert.Error(t, err)\n\t\tassert.Contains(t, err.Error(), \"no tasks\")\n\t})\n\n\tt.Run(\"records failure for non-existent assistant\", func(t *testing.T) {\n\t\trobot := createRunTestRobot(t)\n\t\texec := createRunTestExecution(robot)\n\n\t\texec.Tasks = []types.Task{\n\t\t\t{\n\t\t\t\tID: \"task-001\",\n\t\t\t\tExecutorType: types.ExecutorAssistant,\n\t\t\t\tExecutorID: \"non.existent.agent\",\n\t\t\t\tMessages: []agentcontext.Message{\n\t\t\t\t\t{Role: agentcontext.RoleUser, Content: \"Test\"},\n\t\t\t\t},\n\t\t\t\tOrder: 0,\n\t\t\t\tStatus: types.TaskPending,\n\t\t\t},\n\t\t}\n\n\t\te := standard.New()\n\t\terr := e.RunExecution(ctx, exec, nil)\n\n\t\t// V2 default ContinueOnFailure=true, so no error is returned\n\t\tassert.NoError(t, err)\n\t\tassert.Equal(t, types.TaskFailed, exec.Tasks[0].Status)\n\t\tassert.Len(t, exec.Results, 1)\n\t\tassert.False(t, exec.Results[0].Success)\n\t\tassert.NotEmpty(t, exec.Results[0].Error)\n\t})\n}\n\nfunc TestRunExecutionContinueOnFailure(t *testing.T) {\n\tif testing.Short() {\n\t\tt.Skip(\"Skipping integration test\")\n\t}\n\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\tctx := types.NewContext(context.Background(), testAuth())\n\n\tt.Run(\"stops on first failure when ContinueOnFailure is false\", func(t *testing.T) {\n\t\trobot := createRunTestRobot(t)\n\t\texec := createRunTestExecution(robot)\n\n\t\texec.Tasks = []types.Task{\n\t\t\t{\n\t\t\t\tID: \"task-001\",\n\t\t\t\tExecutorType: types.ExecutorAssistant,\n\t\t\t\tExecutorID: \"non.existent.assistant.xyz123\",\n\t\t\t\tMessages: []agentcontext.Message{\n\t\t\t\t\t{Role: agentcontext.RoleUser, Content: \"This will fail\"},\n\t\t\t\t},\n\t\t\t\tOrder: 0,\n\t\t\t\tStatus: types.TaskPending,\n\t\t\t},\n\t\t\t{\n\t\t\t\tID: \"task-002\",\n\t\t\t\tExecutorType: types.ExecutorAssistant,\n\t\t\t\tExecutorID: \"experts.text-writer\",\n\t\t\t\tMessages: []agentcontext.Message{\n\t\t\t\t\t{Role: agentcontext.RoleUser, Content: \"Write a greeting\"},\n\t\t\t\t},\n\t\t\t\tOrder: 1,\n\t\t\t\tStatus: types.TaskPending,\n\t\t\t},\n\t\t}\n\n\t\tconfig := &standard.RunConfig{ContinueOnFailure: false}\n\n\t\te := standard.New()\n\t\terr := e.RunExecution(ctx, exec, config)\n\n\t\tassert.Error(t, err)\n\t\tassert.Contains(t, err.Error(), \"task-001\")\n\t\tassert.Len(t, exec.Results, 1)\n\t\tassert.Equal(t, types.TaskFailed, exec.Tasks[0].Status)\n\t\tassert.Equal(t, types.TaskSkipped, exec.Tasks[1].Status)\n\t})\n\n\tt.Run(\"continues execution when ContinueOnFailure is true\", func(t *testing.T) {\n\t\trobot := createRunTestRobot(t)\n\t\texec := createRunTestExecution(robot)\n\n\t\t// First task will fail, but second should still execute\n\t\texec.Tasks = []types.Task{\n\t\t\t{\n\t\t\t\tID: \"task-001\",\n\t\t\t\tExecutorType: types.ExecutorAssistant,\n\t\t\t\tExecutorID: \"non.existent.assistant.xyz123\",\n\t\t\t\tMessages: []agentcontext.Message{\n\t\t\t\t\t{Role: agentcontext.RoleUser, Content: \"This will fail\"},\n\t\t\t\t},\n\t\t\t\tOrder: 0,\n\t\t\t\tStatus: types.TaskPending,\n\t\t\t},\n\t\t\t{\n\t\t\t\tID: \"task-002\",\n\t\t\t\tExecutorType: types.ExecutorAssistant,\n\t\t\t\tExecutorID: \"experts.text-writer\",\n\t\t\t\tMessages: []agentcontext.Message{\n\t\t\t\t\t{Role: agentcontext.RoleUser, Content: \"Write a short greeting message\"},\n\t\t\t\t},\n\t\t\t\tExpectedOutput: \"A greeting message\",\n\t\t\t\tOrder: 1,\n\t\t\t\tStatus: types.TaskPending,\n\t\t\t},\n\t\t\t{\n\t\t\t\tID: \"task-003\",\n\t\t\t\tExecutorType: types.ExecutorAssistant,\n\t\t\t\tExecutorID: \"experts.text-writer\",\n\t\t\t\tMessages: []agentcontext.Message{\n\t\t\t\t\t{Role: agentcontext.RoleUser, Content: \"Write a farewell message\"},\n\t\t\t\t},\n\t\t\t\tExpectedOutput: \"A farewell message\",\n\t\t\t\tOrder: 2,\n\t\t\t\tStatus: types.TaskPending,\n\t\t\t},\n\t\t}\n\n\t\t// V2 default ContinueOnFailure=true\n\t\te := standard.New()\n\t\terr := e.RunExecution(ctx, exec, nil)\n\n\t\t// Should NOT return error when ContinueOnFailure is true\n\t\tassert.NoError(t, err)\n\n\t\t// All tasks should have results\n\t\tassert.Len(t, exec.Results, 3)\n\n\t\t// First task failed\n\t\tassert.Equal(t, types.TaskFailed, exec.Tasks[0].Status)\n\t\tassert.False(t, exec.Results[0].Success)\n\n\t\t// Second and third tasks should have executed and completed\n\t\tassert.Equal(t, types.TaskCompleted, exec.Tasks[1].Status)\n\t\tassert.True(t, exec.Results[1].Success)\n\n\t\tassert.Equal(t, types.TaskCompleted, exec.Tasks[2].Status)\n\t\tassert.True(t, exec.Results[2].Success)\n\n\t\tt.Logf(\"Task 1 (failed): %v\", exec.Results[0].Error)\n\t\tt.Logf(\"Task 2 (success): %v\", exec.Results[1].Output)\n\t\tt.Logf(\"Task 3 (success): %v\", exec.Results[2].Output)\n\t})\n\n\tt.Run(\"multiple failures with ContinueOnFailure\", func(t *testing.T) {\n\t\trobot := createRunTestRobot(t)\n\t\texec := createRunTestExecution(robot)\n\n\t\t// Mix of failing and succeeding tasks\n\t\texec.Tasks = []types.Task{\n\t\t\t{\n\t\t\t\tID: \"task-001\",\n\t\t\t\tExecutorType: types.ExecutorAssistant,\n\t\t\t\tExecutorID: \"non.existent.assistant.1\",\n\t\t\t\tMessages: []agentcontext.Message{\n\t\t\t\t\t{Role: agentcontext.RoleUser, Content: \"Fail 1\"},\n\t\t\t\t},\n\t\t\t\tOrder: 0,\n\t\t\t\tStatus: types.TaskPending,\n\t\t\t},\n\t\t\t{\n\t\t\t\tID: \"task-002\",\n\t\t\t\tExecutorType: types.ExecutorAssistant,\n\t\t\t\tExecutorID: \"experts.text-writer\",\n\t\t\t\tMessages: []agentcontext.Message{\n\t\t\t\t\t{Role: agentcontext.RoleUser, Content: \"Say hello\"},\n\t\t\t\t},\n\t\t\t\tOrder: 1,\n\t\t\t\tStatus: types.TaskPending,\n\t\t\t},\n\t\t\t{\n\t\t\t\tID: \"task-003\",\n\t\t\t\tExecutorType: types.ExecutorAssistant,\n\t\t\t\tExecutorID: \"non.existent.assistant.2\",\n\t\t\t\tMessages: []agentcontext.Message{\n\t\t\t\t\t{Role: agentcontext.RoleUser, Content: \"Fail 2\"},\n\t\t\t\t},\n\t\t\t\tOrder: 2,\n\t\t\t\tStatus: types.TaskPending,\n\t\t\t},\n\t\t\t{\n\t\t\t\tID: \"task-004\",\n\t\t\t\tExecutorType: types.ExecutorAssistant,\n\t\t\t\tExecutorID: \"experts.text-writer\",\n\t\t\t\tMessages: []agentcontext.Message{\n\t\t\t\t\t{Role: agentcontext.RoleUser, Content: \"Say goodbye\"},\n\t\t\t\t},\n\t\t\t\tOrder: 3,\n\t\t\t\tStatus: types.TaskPending,\n\t\t\t},\n\t\t}\n\n\t\t// V2 default ContinueOnFailure=true\n\t\te := standard.New()\n\t\terr := e.RunExecution(ctx, exec, nil)\n\n\t\tassert.NoError(t, err)\n\t\tassert.Len(t, exec.Results, 4)\n\n\t\t// Check status pattern: fail, success, fail, success\n\t\tassert.Equal(t, types.TaskFailed, exec.Tasks[0].Status)\n\t\tassert.Equal(t, types.TaskCompleted, exec.Tasks[1].Status)\n\t\tassert.Equal(t, types.TaskFailed, exec.Tasks[2].Status)\n\t\tassert.Equal(t, types.TaskCompleted, exec.Tasks[3].Status)\n\n\t\t// Count successes and failures\n\t\tsuccessCount := 0\n\t\tfailCount := 0\n\t\tfor _, result := range exec.Results {\n\t\t\tif result.Success {\n\t\t\t\tsuccessCount++\n\t\t\t} else {\n\t\t\t\tfailCount++\n\t\t\t}\n\t\t}\n\t\tassert.Equal(t, 2, successCount)\n\t\tassert.Equal(t, 2, failCount)\n\t})\n}\n\nfunc TestRunExecutionNoValidation(t *testing.T) {\n\tif testing.Short() {\n\t\tt.Skip(\"Skipping integration test\")\n\t}\n\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\tctx := types.NewContext(context.Background(), testAuth())\n\n\tt.Run(\"V2 runner does not set Validation on results\", func(t *testing.T) {\n\t\trobot := createRunTestRobot(t)\n\t\texec := createRunTestExecution(robot)\n\n\t\texec.Tasks = []types.Task{\n\t\t\t{\n\t\t\t\tID: \"task-001\",\n\t\t\t\tExecutorType: types.ExecutorAssistant,\n\t\t\t\tExecutorID: \"experts.data-analyst\",\n\t\t\t\tMessages: []agentcontext.Message{\n\t\t\t\t\t{Role: agentcontext.RoleUser, Content: \"Return a JSON object with fields: name (string), count (number). Example: {\\\"name\\\": \\\"test\\\", \\\"count\\\": 5}\"},\n\t\t\t\t},\n\t\t\t\tExpectedOutput: \"JSON object with name and count fields\",\n\t\t\t\tOrder: 0,\n\t\t\t\tStatus: types.TaskPending,\n\t\t\t},\n\t\t}\n\n\t\te := standard.New()\n\t\terr := e.RunExecution(ctx, exec, nil)\n\n\t\trequire.NoError(t, err)\n\t\trequire.Len(t, exec.Results, 1)\n\n\t\tresult := exec.Results[0]\n\t\tassert.True(t, result.Success, \"task should succeed if assistant call returns\")\n\t\tassert.NotNil(t, result.Output, \"output should be present\")\n\t\tassert.Nil(t, result.Validation, \"V2 runner does not run validation\")\n\n\t\tt.Logf(\"Output: %v\", result.Output)\n\t})\n}\n\n// ============================================================================\n// Helper Functions\n// ============================================================================\n\n// createRunTestRobot creates a test robot for P3 run tests\nfunc createRunTestRobot(t *testing.T) *types.Robot {\n\tt.Helper()\n\treturn &types.Robot{\n\t\tMemberID: \"test-robot-run\",\n\t\tTeamID: \"test-team-1\",\n\t\tDisplayName: \"Test Robot for Run\",\n\t\tSystemPrompt: \"You are a helpful assistant.\",\n\t\tConfig: &types.Config{\n\t\t\tIdentity: &types.Identity{\n\t\t\t\tRole: \"Test Assistant\",\n\t\t\t\tDuties: []string{\"Execute tasks\", \"Generate content\"},\n\t\t\t},\n\t\t\tResources: &types.Resources{\n\t\t\t\tPhases: map[types.Phase]string{\n\t\t\t\t\ttypes.PhaseRun: \"robot.validation\",\n\t\t\t\t\t\"validation\": \"robot.validation\", // For semantic validation agent\n\t\t\t\t},\n\t\t\t\tAgents: []string{\n\t\t\t\t\t\"experts.data-analyst\",\n\t\t\t\t\t\"experts.summarizer\",\n\t\t\t\t\t\"experts.text-writer\",\n\t\t\t\t},\n\t\t\t},\n\t\t},\n\t}\n}\n\n// createRunTestExecution creates a test execution for P3 run tests\nfunc createRunTestExecution(robot *types.Robot) *types.Execution {\n\texec := &types.Execution{\n\t\tID: \"test-exec-run-1\",\n\t\tMemberID: robot.MemberID,\n\t\tTeamID: robot.TeamID,\n\t\tTriggerType: types.TriggerClock,\n\t\tStartTime: time.Now(),\n\t\tStatus: types.ExecRunning,\n\t\tPhase: types.PhaseRun,\n\t\tGoals: &types.Goals{\n\t\t\tContent: \"## Goals\\n\\n1. Execute test tasks\",\n\t\t},\n\t}\n\texec.SetRobot(robot)\n\treturn exec\n}\n" }, { "path": "agent/robot/executor/standard/runner.go", "content": "package standard\n\nimport (\n\t\"encoding/json\"\n\t\"fmt\"\n\t\"strings\"\n\t\"time\"\n\n\t\"github.com/yaoapp/gou/mcp\"\n\t\"github.com/yaoapp/gou/process\"\n\tagentcontext \"github.com/yaoapp/yao/agent/context\"\n\trobottypes \"github.com/yaoapp/yao/agent/robot/types\"\n)\n\n// Runner handles execution of individual tasks\ntype Runner struct {\n\tctx *robottypes.Context\n\trobot *robottypes.Robot\n\tconfig *RunConfig\n\tchatID string // execution-level chatID for conversation persistence (§8.4)\n\tlog *execLogger\n}\n\n// NewRunner creates a new task runner\nfunc NewRunner(ctx *robottypes.Context, robot *robottypes.Robot, config *RunConfig, chatID string, execID string) *Runner {\n\treturn &Runner{\n\t\tctx: ctx,\n\t\trobot: robot,\n\t\tconfig: config,\n\t\tchatID: chatID,\n\t\tlog: newExecLogger(robot, execID),\n\t}\n}\n\n// RunnerContext provides context for task execution\ntype RunnerContext struct {\n\t// PreviousResults contains results from previously executed tasks\n\tPreviousResults []robottypes.TaskResult\n\n\t// Goals contains the goals from P1 (for context)\n\tGoals *robottypes.Goals\n\n\t// SystemPrompt is the robot's system prompt\n\tSystemPrompt string\n}\n\n// BuildTaskContext builds context for a task including previous results\nfunc (r *Runner) BuildTaskContext(exec *robottypes.Execution, taskIndex int) *RunnerContext {\n\tctx := &RunnerContext{\n\t\tGoals: exec.Goals,\n\t\tSystemPrompt: r.robot.SystemPrompt,\n\t}\n\n\t// Include results from previous tasks (with bounds check)\n\tif taskIndex > 0 && len(exec.Results) > 0 {\n\t\tendIndex := taskIndex\n\t\tif endIndex > len(exec.Results) {\n\t\t\tendIndex = len(exec.Results)\n\t\t}\n\t\tctx.PreviousResults = exec.Results[:endIndex]\n\t}\n\n\treturn ctx\n}\n\n// ExecuteTask executes a single task (V2 simplified: single call, no validation loop).\n// Success is determined purely by whether the call itself succeeds without error.\n// Quality evaluation is deferred to the Delivery Agent (P4) using ExpectedOutput.\nfunc (r *Runner) ExecuteTask(task *robottypes.Task, taskCtx *RunnerContext) *robottypes.TaskResult {\n\tstartTime := time.Now()\n\n\tresult := &robottypes.TaskResult{\n\t\tTaskID: task.ID,\n\t}\n\n\t// For non-assistant tasks (MCP, Process), single-call execution\n\tif task.ExecutorType != robottypes.ExecutorAssistant {\n\t\toutput, err := r.executeNonAssistantTask(task, taskCtx)\n\t\tif err != nil {\n\t\t\tresult.Success = false\n\t\t\tresult.Error = fmt.Sprintf(\"execution failed: %s\", err.Error())\n\t\t\tresult.Duration = time.Since(startTime).Milliseconds()\n\t\t\tr.log.logTaskOutput(task, result)\n\t\t\treturn result\n\t\t}\n\n\t\tresult.Output = output\n\t\tresult.Success = true\n\t\tresult.Duration = time.Since(startTime).Milliseconds()\n\t\tr.log.logTaskOutput(task, result)\n\t\treturn result\n\t}\n\n\t// For assistant tasks, single call via conversation\n\toutput, callResult, err := r.executeAssistantTask(task, taskCtx)\n\tif err != nil {\n\t\tresult.Success = false\n\t\tresult.Error = err.Error()\n\t\tresult.Duration = time.Since(startTime).Milliseconds()\n\t\tr.log.logTaskOutput(task, result)\n\t\treturn result\n\t}\n\n\tresult.Output = output\n\tresult.Success = true\n\tresult.Duration = time.Since(startTime).Milliseconds()\n\n\t// Check if assistant signals it needs human input (V2 suspend protocol)\n\tif needInput, question := detectNeedMoreInfo(callResult); needInput {\n\t\tresult.NeedInput = true\n\t\tresult.InputQuestion = question\n\t}\n\n\tr.log.logTaskOutput(task, result)\n\treturn result\n}\n\n// executeNonAssistantTask executes MCP or Process tasks (single-call, no multi-turn)\nfunc (r *Runner) executeNonAssistantTask(task *robottypes.Task, taskCtx *RunnerContext) (interface{}, error) {\n\tswitch task.ExecutorType {\n\tcase robottypes.ExecutorMCP:\n\t\treturn r.ExecuteMCPTask(task, taskCtx)\n\tcase robottypes.ExecutorProcess:\n\t\treturn r.ExecuteProcessTask(task, taskCtx)\n\tdefault:\n\t\treturn nil, fmt.Errorf(\"unsupported executor type: %s (expected mcp or process)\", task.ExecutorType)\n\t}\n}\n\n// executeAssistantTask executes an assistant task with a single conversation turn.\n// Returns the extracted output, the raw CallResult (for need_input detection), and any error.\nfunc (r *Runner) executeAssistantTask(task *robottypes.Task, taskCtx *RunnerContext) (interface{}, *CallResult, error) {\n\tcaller := NewAgentCaller()\n\tcaller.log = r.log\n\tcaller.Connector = r.robot.LanguageModel\n\tcaller.ChatID = r.chatID\n\n\tmessages := r.BuildAssistantMessages(task, taskCtx)\n\tinput := r.FormatMessagesAsText(messages)\n\n\tif strings.TrimSpace(input) == \"\" {\n\t\treturn nil, nil, fmt.Errorf(\"no valid input messages for task %s\", task.ID)\n\t}\n\n\tif taskCtx.SystemPrompt != \"\" {\n\t\tinput = \"## Context\\n\\n\" + taskCtx.SystemPrompt + \"\\n\\n## Task\\n\\n\" + input\n\t}\n\n\tr.log.logTaskInput(task, input)\n\n\tresult, err := caller.CallWithMessages(r.ctx, task.ExecutorID, input)\n\tif err != nil {\n\t\treturn nil, nil, fmt.Errorf(\"assistant call failed: %w\", err)\n\t}\n\n\toutput := r.extractOutput(result)\n\treturn output, result, nil\n}\n\n// detectNeedMoreInfo checks if the assistant's response signals it needs human input.\n// The protocol: Next hook returns {data: {status: \"need_input\", question: \"...\"}}.\n// Also handles the unwrapped form {status: \"need_input\", question: \"...\"} for robustness.\nfunc detectNeedMoreInfo(result *CallResult) (bool, string) {\n\tif result == nil || result.Next == nil {\n\t\treturn false, \"\"\n\t}\n\tm, ok := result.Next.(map[string]interface{})\n\tif !ok {\n\t\treturn false, \"\"\n\t}\n\n\t// Unwrap \"data\" envelope if present (Next hook standard: {data: {status: ...}})\n\tif data, ok := m[\"data\"].(map[string]interface{}); ok {\n\t\tm = data\n\t}\n\n\tstatus, _ := m[\"status\"].(string)\n\tif status != \"need_input\" {\n\t\treturn false, \"\"\n\t}\n\tquestion, _ := m[\"question\"].(string)\n\tif question == \"\" {\n\t\tquestion = result.GetText()\n\t}\n\treturn true, question\n}\n\n// extractOutput extracts the output from a CallResult\n// Priority: Next hook data > LLM Completion content\n// Next is the agent's formal A2A output (could be string, map, array, number, etc.)\n// Content is the raw LLM completion text (fallback only when Next is absent)\nfunc (r *Runner) extractOutput(result *CallResult) interface{} {\n\tif result == nil {\n\t\treturn nil\n\t}\n\tif result.Next != nil {\n\t\treturn result.Next\n\t}\n\tif result.Content != \"\" {\n\t\treturn result.Content\n\t}\n\treturn nil\n}\n\n// ExecuteMCPTask executes a task using an MCP tool\n// Requires task.MCPServer and task.MCPTool fields to be set\n// executor_id is the combined form: \"mcp_server.mcp_tool\" (e.g., \"ark.image.text2img.generate\")\nfunc (r *Runner) ExecuteMCPTask(task *robottypes.Task, taskCtx *RunnerContext) (interface{}, error) {\n\t// Validate MCP-specific fields\n\tif task.MCPServer == \"\" || task.MCPTool == \"\" {\n\t\treturn nil, fmt.Errorf(\"MCP task requires mcp_server and mcp_tool fields (executor_id: %s)\", task.ExecutorID)\n\t}\n\n\t// Get MCP client\n\tclient, err := mcp.Select(task.MCPServer)\n\tif err != nil {\n\t\treturn nil, fmt.Errorf(\"MCP server not found: %s: %w\", task.MCPServer, err)\n\t}\n\n\t// Build arguments map from task.Args\n\targs := make(map[string]interface{})\n\tif len(task.Args) > 0 {\n\t\t// First argument should be a map of tool arguments\n\t\tif argsMap, ok := task.Args[0].(map[string]interface{}); ok {\n\t\t\targs = argsMap\n\t\t} else {\n\t\t\t// If not a map, try to convert single argument\n\t\t\targs[\"input\"] = task.Args[0]\n\t\t}\n\t}\n\n\t// Call MCP tool\n\tresult, err := client.CallTool(r.ctx.Context, task.MCPTool, args)\n\tif err != nil {\n\t\treturn nil, fmt.Errorf(\"MCP tool call failed (%s.%s): %w\", task.MCPServer, task.MCPTool, err)\n\t}\n\n\treturn result, nil\n}\n\n// ExecuteProcessTask executes a task using a Yao process\n// ExecutorID is the process name (e.g., \"models.user.Find\", \"scripts.myScript.Run\")\nfunc (r *Runner) ExecuteProcessTask(task *robottypes.Task, taskCtx *RunnerContext) (interface{}, error) {\n\t// Create process with task arguments\n\tproc, err := process.Of(task.ExecutorID, task.Args...)\n\tif err != nil {\n\t\treturn nil, fmt.Errorf(\"process creation failed: %w\", err)\n\t}\n\n\t// Set context for timeout and cancellation\n\tproc.Context = r.ctx.Context\n\n\t// Execute the process\n\tif err := proc.Execute(); err != nil {\n\t\treturn nil, fmt.Errorf(\"process execution failed: %w\", err)\n\t}\n\tdefer proc.Release()\n\n\t// Return the result\n\treturn proc.Value(), nil\n}\n\n// BuildAssistantMessages builds messages for an assistant task\nfunc (r *Runner) BuildAssistantMessages(task *robottypes.Task, taskCtx *RunnerContext) []agentcontext.Message {\n\tmessages := make([]agentcontext.Message, 0)\n\n\t// Add context from previous tasks if available\n\tif len(taskCtx.PreviousResults) > 0 {\n\t\tcontextMsg := r.FormatPreviousResultsAsContext(taskCtx.PreviousResults)\n\t\tif contextMsg != \"\" {\n\t\t\tmessages = append(messages, agentcontext.Message{\n\t\t\t\tRole: agentcontext.RoleUser,\n\t\t\t\tContent: contextMsg,\n\t\t\t})\n\t\t}\n\t}\n\n\t// Add task messages\n\tmessages = append(messages, task.Messages...)\n\n\treturn messages\n}\n\n// FormatMessagesAsText converts messages to a single text string\nfunc (r *Runner) FormatMessagesAsText(messages []agentcontext.Message) string {\n\tvar result string\n\tfor _, msg := range messages {\n\t\tswitch content := msg.Content.(type) {\n\t\tcase string:\n\t\t\tresult += content + \"\\n\\n\"\n\t\tcase []interface{}:\n\t\t\t// Handle multi-part content (e.g., text + images)\n\t\t\tfor _, part := range content {\n\t\t\t\tif textPart, ok := part.(map[string]interface{}); ok {\n\t\t\t\t\tif text, ok := textPart[\"text\"].(string); ok {\n\t\t\t\t\t\tresult += text + \"\\n\\n\"\n\t\t\t\t\t}\n\t\t\t\t}\n\t\t\t}\n\t\tdefault:\n\t\t\t// Try JSON marshaling as fallback\n\t\t\tif content != nil {\n\t\t\t\tif jsonBytes, err := json.Marshal(content); err == nil {\n\t\t\t\t\tresult += string(jsonBytes) + \"\\n\\n\"\n\t\t\t\t}\n\t\t\t}\n\t\t}\n\t}\n\treturn result\n}\n\n// FormatPreviousResultsAsContext formats previous task results as context\nfunc (r *Runner) FormatPreviousResultsAsContext(results []robottypes.TaskResult) string {\n\tif len(results) == 0 {\n\t\treturn \"\"\n\t}\n\n\tvar sb strings.Builder\n\tsb.WriteString(\"## Previous Task Results\\n\\n\")\n\tsb.WriteString(\"The following tasks have been completed. Use their results as needed:\\n\\n\")\n\n\tfor _, result := range results {\n\t\tsb.WriteString(fmt.Sprintf(\"### Task: %s\\n\", result.TaskID))\n\t\tif result.Success {\n\t\t\tsb.WriteString(\"- Status: ✓ Success\\n\")\n\t\t} else {\n\t\t\tsb.WriteString(\"- Status: ✗ Failed\\n\")\n\t\t}\n\n\t\tif result.Output != nil {\n\t\t\toutputJSON, err := json.MarshalIndent(result.Output, \"\", \" \")\n\t\t\tif err == nil {\n\t\t\t\tsb.WriteString(fmt.Sprintf(\"- Output:\\n```json\\n%s\\n```\\n\", string(outputJSON)))\n\t\t\t} else {\n\t\t\t\tsb.WriteString(fmt.Sprintf(\"- Output: %v\\n\", result.Output))\n\t\t\t}\n\t\t}\n\t\tsb.WriteString(\"\\n\")\n\t}\n\n\treturn sb.String()\n}\n" }, { "path": "agent/robot/executor/standard/runner_test.go", "content": "package standard_test\n\nimport (\n\t\"context\"\n\t\"testing\"\n\n\t\"github.com/stretchr/testify/assert\"\n\tagentcontext \"github.com/yaoapp/yao/agent/context\"\n\t\"github.com/yaoapp/yao/agent/robot/executor/standard\"\n\t\"github.com/yaoapp/yao/agent/robot/types\"\n\t\"github.com/yaoapp/yao/agent/testutils\"\n)\n\n// ============================================================================\n// Runner Tests - V2 Simplified Execution (single call, no validation loop)\n// ============================================================================\n\nfunc TestRunnerExecuteTask(t *testing.T) {\n\tif testing.Short() {\n\t\tt.Skip(\"Skipping integration test\")\n\t}\n\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\tctx := types.NewContext(context.Background(), testAuth())\n\n\tt.Run(\"executes assistant task successfully\", func(t *testing.T) {\n\t\trobot := createRunnerTestRobot(t)\n\t\tconfig := standard.DefaultRunConfig()\n\t\trunner := standard.NewRunner(ctx, robot, config, \"\", \"test\")\n\n\t\ttask := &types.Task{\n\t\t\tID: \"task-001\",\n\t\t\tExecutorType: types.ExecutorAssistant,\n\t\t\tExecutorID: \"experts.text-writer\",\n\t\t\tMessages: []agentcontext.Message{\n\t\t\t\t{Role: agentcontext.RoleUser, Content: \"Write a haiku about coding. Format: three lines with 5-7-5 syllables.\"},\n\t\t\t},\n\t\t\tExpectedOutput: \"A haiku poem about coding\",\n\t\t\tStatus: types.TaskPending,\n\t\t}\n\n\t\ttaskCtx := &standard.RunnerContext{\n\t\t\tSystemPrompt: robot.SystemPrompt,\n\t\t}\n\n\t\tresult := runner.ExecuteTask(task, taskCtx)\n\n\t\tassert.True(t, result.Success, \"task should succeed\")\n\t\tassert.NotNil(t, result.Output, \"output should not be nil\")\n\t\tassert.Empty(t, result.Error, \"error should be empty on success\")\n\t\tassert.Greater(t, result.Duration, int64(0), \"duration should be positive\")\n\n\t\tt.Logf(\"Output: %v\", result.Output)\n\t})\n\n\tt.Run(\"returns success without validation for assistant tasks\", func(t *testing.T) {\n\t\trobot := createRunnerTestRobot(t)\n\t\tconfig := standard.DefaultRunConfig()\n\t\trunner := standard.NewRunner(ctx, robot, config, \"\", \"test\")\n\n\t\ttask := &types.Task{\n\t\t\tID: \"task-002\",\n\t\t\tExecutorType: types.ExecutorAssistant,\n\t\t\tExecutorID: \"experts.data-analyst\",\n\t\t\tMessages: []agentcontext.Message{\n\t\t\t\t{Role: agentcontext.RoleUser, Content: \"Return a JSON object with exactly these fields: status (string 'ok'), count (number greater than 0).\"},\n\t\t\t},\n\t\t\tExpectedOutput: \"JSON with status='ok' and count>0\",\n\t\t\tStatus: types.TaskPending,\n\t\t}\n\n\t\ttaskCtx := &standard.RunnerContext{\n\t\t\tSystemPrompt: robot.SystemPrompt,\n\t\t}\n\n\t\tresult := runner.ExecuteTask(task, taskCtx)\n\n\t\t// V2: success is determined by the call succeeding, not by validation\n\t\tassert.True(t, result.Success, \"task should succeed if assistant call returns\")\n\t\tassert.NotNil(t, result.Output, \"output should not be nil\")\n\t\tassert.Nil(t, result.Validation, \"V2 does not set Validation in runner\")\n\n\t\tt.Logf(\"Success: %v, Output: %v\", result.Success, result.Output)\n\t})\n\n\tt.Run(\"handles empty messages gracefully\", func(t *testing.T) {\n\t\trobot := createRunnerTestRobot(t)\n\t\tconfig := standard.DefaultRunConfig()\n\t\trunner := standard.NewRunner(ctx, robot, config, \"\", \"test\")\n\n\t\ttask := &types.Task{\n\t\t\tID: \"task-003\",\n\t\t\tExecutorType: types.ExecutorAssistant,\n\t\t\tExecutorID: \"experts.text-writer\",\n\t\t\tMessages: []agentcontext.Message{},\n\t\t\tStatus: types.TaskPending,\n\t\t}\n\n\t\ttaskCtx := &standard.RunnerContext{\n\t\t\tSystemPrompt: robot.SystemPrompt,\n\t\t}\n\n\t\tresult := runner.ExecuteTask(task, taskCtx)\n\n\t\tassert.False(t, result.Success, \"task should fail with empty messages\")\n\t\tassert.NotEmpty(t, result.Error, \"error should describe the failure\")\n\t\tt.Logf(\"Error: %s\", result.Error)\n\t})\n}\n\nfunc TestRunnerBuildTaskContext(t *testing.T) {\n\tif testing.Short() {\n\t\tt.Skip(\"Skipping integration test\")\n\t}\n\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\tctx := types.NewContext(context.Background(), testAuth())\n\n\tt.Run(\"includes previous results in context\", func(t *testing.T) {\n\t\trobot := createRunnerTestRobot(t)\n\t\tconfig := standard.DefaultRunConfig()\n\t\trunner := standard.NewRunner(ctx, robot, config, \"\", \"test\")\n\n\t\texec := &types.Execution{\n\t\t\tID: \"test-exec\",\n\t\t\tMemberID: robot.MemberID,\n\t\t\tTeamID: robot.TeamID,\n\t\t\tGoals: &types.Goals{\n\t\t\t\tContent: \"Test goals\",\n\t\t\t},\n\t\t\tResults: []types.TaskResult{\n\t\t\t\t{\n\t\t\t\t\tTaskID: \"task-001\",\n\t\t\t\t\tSuccess: true,\n\t\t\t\t\tOutput: map[string]interface{}{\"data\": \"previous result\"},\n\t\t\t\t},\n\t\t\t\t{\n\t\t\t\t\tTaskID: \"task-002\",\n\t\t\t\t\tSuccess: true,\n\t\t\t\t\tOutput: \"Another result\",\n\t\t\t\t},\n\t\t\t},\n\t\t}\n\t\texec.SetRobot(robot)\n\n\t\t// Build context for task at index 2 (should include results 0 and 1)\n\t\ttaskCtx := runner.BuildTaskContext(exec, 2)\n\n\t\tassert.NotNil(t, taskCtx)\n\t\tassert.Len(t, taskCtx.PreviousResults, 2)\n\t\tassert.Equal(t, \"task-001\", taskCtx.PreviousResults[0].TaskID)\n\t\tassert.Equal(t, \"task-002\", taskCtx.PreviousResults[1].TaskID)\n\t\tassert.Equal(t, robot.SystemPrompt, taskCtx.SystemPrompt)\n\t})\n\n\tt.Run(\"handles first task with no previous results\", func(t *testing.T) {\n\t\trobot := createRunnerTestRobot(t)\n\t\tconfig := standard.DefaultRunConfig()\n\t\trunner := standard.NewRunner(ctx, robot, config, \"\", \"test\")\n\n\t\texec := &types.Execution{\n\t\t\tID: \"test-exec\",\n\t\t\tMemberID: robot.MemberID,\n\t\t\tTeamID: robot.TeamID,\n\t\t\tGoals: &types.Goals{\n\t\t\t\tContent: \"Test goals\",\n\t\t\t},\n\t\t\tResults: []types.TaskResult{},\n\t\t}\n\t\texec.SetRobot(robot)\n\n\t\ttaskCtx := runner.BuildTaskContext(exec, 0)\n\n\t\tassert.NotNil(t, taskCtx)\n\t\tassert.Empty(t, taskCtx.PreviousResults)\n\t})\n\n\tt.Run(\"handles bounds check for task index\", func(t *testing.T) {\n\t\trobot := createRunnerTestRobot(t)\n\t\tconfig := standard.DefaultRunConfig()\n\t\trunner := standard.NewRunner(ctx, robot, config, \"\", \"test\")\n\n\t\texec := &types.Execution{\n\t\t\tID: \"test-exec\",\n\t\t\tMemberID: robot.MemberID,\n\t\t\tTeamID: robot.TeamID,\n\t\t\tResults: []types.TaskResult{\n\t\t\t\t{TaskID: \"task-001\", Success: true},\n\t\t\t},\n\t\t}\n\t\texec.SetRobot(robot)\n\n\t\t// Task index 5, but only 1 result exists\n\t\ttaskCtx := runner.BuildTaskContext(exec, 5)\n\n\t\tassert.NotNil(t, taskCtx)\n\t\tassert.Len(t, taskCtx.PreviousResults, 1) // Should only include available results\n\t})\n}\n\nfunc TestRunnerFormatPreviousResultsAsContext(t *testing.T) {\n\tif testing.Short() {\n\t\tt.Skip(\"Skipping integration test\")\n\t}\n\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\tctx := types.NewContext(context.Background(), testAuth())\n\n\tt.Run(\"formats previous results as markdown\", func(t *testing.T) {\n\t\trobot := createRunnerTestRobot(t)\n\t\tconfig := standard.DefaultRunConfig()\n\t\trunner := standard.NewRunner(ctx, robot, config, \"\", \"test\")\n\n\t\tresults := []types.TaskResult{\n\t\t\t{\n\t\t\t\tTaskID: \"task-001\",\n\t\t\t\tSuccess: true,\n\t\t\t\tOutput: map[string]interface{}{\"key\": \"value\", \"count\": 42},\n\t\t\t},\n\t\t\t{\n\t\t\t\tTaskID: \"task-002\",\n\t\t\t\tSuccess: false,\n\t\t\t\tOutput: \"Partial result\",\n\t\t\t\tError: \"Validation failed\",\n\t\t\t},\n\t\t}\n\n\t\tformatted := runner.FormatPreviousResultsAsContext(results)\n\n\t\tassert.Contains(t, formatted, \"## Previous Task Results\")\n\t\tassert.Contains(t, formatted, \"task-001\")\n\t\tassert.Contains(t, formatted, \"task-002\")\n\t\tassert.Contains(t, formatted, \"Success\")\n\t\tassert.Contains(t, formatted, \"Failed\")\n\t\tassert.Contains(t, formatted, \"key\")\n\t\tassert.Contains(t, formatted, \"value\")\n\n\t\tt.Logf(\"Formatted context:\\n%s\", formatted)\n\t})\n\n\tt.Run(\"returns empty string for no results\", func(t *testing.T) {\n\t\trobot := createRunnerTestRobot(t)\n\t\tconfig := standard.DefaultRunConfig()\n\t\trunner := standard.NewRunner(ctx, robot, config, \"\", \"test\")\n\n\t\tformatted := runner.FormatPreviousResultsAsContext([]types.TaskResult{})\n\n\t\tassert.Empty(t, formatted)\n\t})\n}\n\nfunc TestRunnerBuildAssistantMessages(t *testing.T) {\n\tif testing.Short() {\n\t\tt.Skip(\"Skipping integration test\")\n\t}\n\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\tctx := types.NewContext(context.Background(), testAuth())\n\n\tt.Run(\"builds messages with task content\", func(t *testing.T) {\n\t\trobot := createRunnerTestRobot(t)\n\t\tconfig := standard.DefaultRunConfig()\n\t\trunner := standard.NewRunner(ctx, robot, config, \"\", \"test\")\n\n\t\ttask := &types.Task{\n\t\t\tID: \"task-001\",\n\t\t\tExecutorType: types.ExecutorAssistant,\n\t\t\tExecutorID: \"experts.text-writer\",\n\t\t\tMessages: []agentcontext.Message{\n\t\t\t\t{Role: agentcontext.RoleUser, Content: \"Write a greeting\"},\n\t\t\t},\n\t\t}\n\n\t\ttaskCtx := &standard.RunnerContext{\n\t\t\tSystemPrompt: \"You are helpful\",\n\t\t}\n\n\t\tmessages := runner.BuildAssistantMessages(task, taskCtx)\n\n\t\tassert.NotEmpty(t, messages)\n\t\t// Should contain task message\n\t\tfound := false\n\t\tfor _, msg := range messages {\n\t\t\tif content, ok := msg.Content.(string); ok && content == \"Write a greeting\" {\n\t\t\t\tfound = true\n\t\t\t\tbreak\n\t\t\t}\n\t\t}\n\t\tassert.True(t, found, \"should contain task message\")\n\t})\n\n\tt.Run(\"includes previous results in messages\", func(t *testing.T) {\n\t\trobot := createRunnerTestRobot(t)\n\t\tconfig := standard.DefaultRunConfig()\n\t\trunner := standard.NewRunner(ctx, robot, config, \"\", \"test\")\n\n\t\ttask := &types.Task{\n\t\t\tID: \"task-002\",\n\t\t\tExecutorType: types.ExecutorAssistant,\n\t\t\tExecutorID: \"experts.text-writer\",\n\t\t\tMessages: []agentcontext.Message{\n\t\t\t\t{Role: agentcontext.RoleUser, Content: \"Continue from previous\"},\n\t\t\t},\n\t\t}\n\n\t\ttaskCtx := &standard.RunnerContext{\n\t\t\tPreviousResults: []types.TaskResult{\n\t\t\t\t{TaskID: \"task-001\", Success: true, Output: \"Previous output\"},\n\t\t\t},\n\t\t\tSystemPrompt: \"You are helpful\",\n\t\t}\n\n\t\tmessages := runner.BuildAssistantMessages(task, taskCtx)\n\n\t\tassert.NotEmpty(t, messages)\n\t\t// Should have context message with previous results\n\t\tformatted := runner.FormatMessagesAsText(messages)\n\t\tassert.Contains(t, formatted, \"Previous Task Results\")\n\t})\n}\n\nfunc TestRunnerFormatMessagesAsText(t *testing.T) {\n\tif testing.Short() {\n\t\tt.Skip(\"Skipping integration test\")\n\t}\n\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\tctx := types.NewContext(context.Background(), testAuth())\n\n\tt.Run(\"formats string content\", func(t *testing.T) {\n\t\trobot := createRunnerTestRobot(t)\n\t\tconfig := standard.DefaultRunConfig()\n\t\trunner := standard.NewRunner(ctx, robot, config, \"\", \"test\")\n\n\t\tmessages := []agentcontext.Message{\n\t\t\t{Role: agentcontext.RoleUser, Content: \"Hello\"},\n\t\t\t{Role: agentcontext.RoleUser, Content: \"World\"},\n\t\t}\n\n\t\ttext := runner.FormatMessagesAsText(messages)\n\n\t\tassert.Contains(t, text, \"Hello\")\n\t\tassert.Contains(t, text, \"World\")\n\t})\n\n\tt.Run(\"handles multipart content\", func(t *testing.T) {\n\t\trobot := createRunnerTestRobot(t)\n\t\tconfig := standard.DefaultRunConfig()\n\t\trunner := standard.NewRunner(ctx, robot, config, \"\", \"test\")\n\n\t\tmessages := []agentcontext.Message{\n\t\t\t{\n\t\t\t\tRole: agentcontext.RoleUser,\n\t\t\t\tContent: []interface{}{\n\t\t\t\t\tmap[string]interface{}{\"type\": \"text\", \"text\": \"Part 1\"},\n\t\t\t\t\tmap[string]interface{}{\"type\": \"text\", \"text\": \"Part 2\"},\n\t\t\t\t},\n\t\t\t},\n\t\t}\n\n\t\ttext := runner.FormatMessagesAsText(messages)\n\n\t\tassert.Contains(t, text, \"Part 1\")\n\t\tassert.Contains(t, text, \"Part 2\")\n\t})\n\n\tt.Run(\"handles map content via JSON\", func(t *testing.T) {\n\t\trobot := createRunnerTestRobot(t)\n\t\tconfig := standard.DefaultRunConfig()\n\t\trunner := standard.NewRunner(ctx, robot, config, \"\", \"test\")\n\n\t\tmessages := []agentcontext.Message{\n\t\t\t{\n\t\t\t\tRole: agentcontext.RoleUser,\n\t\t\t\tContent: map[string]interface{}{\"key\": \"value\"},\n\t\t\t},\n\t\t}\n\n\t\ttext := runner.FormatMessagesAsText(messages)\n\n\t\tassert.Contains(t, text, \"key\")\n\t\tassert.Contains(t, text, \"value\")\n\t})\n}\n\n// ============================================================================\n// Non-Assistant Task Tests (MCP, Process)\n// ============================================================================\n\nfunc TestRunnerExecuteNonAssistantTask(t *testing.T) {\n\tif testing.Short() {\n\t\tt.Skip(\"Skipping integration test\")\n\t}\n\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\tt.Run(\"executes unsupported type returns error\", func(t *testing.T) {\n\t\tctx := types.NewContext(context.Background(), testAuth())\n\t\trobot := createRunnerTestRobot(t)\n\t\tconfig := standard.DefaultRunConfig()\n\t\trunner := standard.NewRunner(ctx, robot, config, \"\", \"test\")\n\n\t\ttask := &types.Task{\n\t\t\tID: \"task-unknown\",\n\t\t\tExecutorType: \"unsupported\",\n\t\t\tExecutorID: \"anything\",\n\t\t\tStatus: types.TaskPending,\n\t\t}\n\n\t\ttaskCtx := &standard.RunnerContext{}\n\t\tresult := runner.ExecuteTask(task, taskCtx)\n\n\t\tassert.False(t, result.Success, \"unsupported executor type should fail\")\n\t\tassert.Contains(t, result.Error, \"unsupported executor type\")\n\t\tassert.Nil(t, result.Validation, \"V2 does not set Validation in runner\")\n\t})\n}\n\n// ============================================================================\n// Helper Functions\n// ============================================================================\n\n// createRunnerTestRobot creates a test robot for runner tests\nfunc createRunnerTestRobot(t *testing.T) *types.Robot {\n\tt.Helper()\n\treturn &types.Robot{\n\t\tMemberID: \"test-robot-runner\",\n\t\tTeamID: \"test-team-1\",\n\t\tDisplayName: \"Test Robot for Runner\",\n\t\tSystemPrompt: \"You are a helpful assistant. Follow instructions carefully and provide clear responses.\",\n\t\tConfig: &types.Config{\n\t\t\tIdentity: &types.Identity{\n\t\t\t\tRole: \"Test Assistant\",\n\t\t\t\tDuties: []string{\"Execute tasks\", \"Generate content\"},\n\t\t\t},\n\t\t\tResources: &types.Resources{\n\t\t\t\tPhases: map[types.Phase]string{\n\t\t\t\t\ttypes.PhaseRun: \"robot.run\",\n\t\t\t\t},\n\t\t\t\tAgents: []string{\n\t\t\t\t\t\"experts.data-analyst\",\n\t\t\t\t\t\"experts.summarizer\",\n\t\t\t\t\t\"experts.text-writer\",\n\t\t\t\t},\n\t\t\t},\n\t\t},\n\t}\n}\n\n// Note: createRunnerTestExecution is available if needed for future tests\n// that require a full Execution object instead of just RunnerContext\n" }, { "path": "agent/robot/executor/standard/suspend_resume_test.go", "content": "package standard_test\n\nimport (\n\t\"context\"\n\t\"testing\"\n\t\"time\"\n\n\t\"github.com/stretchr/testify/assert\"\n\t\"github.com/stretchr/testify/require\"\n\tagentcontext \"github.com/yaoapp/yao/agent/context\"\n\t\"github.com/yaoapp/yao/agent/robot/executor/standard\"\n\texecutortypes \"github.com/yaoapp/yao/agent/robot/executor/types\"\n\t\"github.com/yaoapp/yao/agent/robot/types\"\n\t\"github.com/yaoapp/yao/agent/testutils\"\n)\n\n// ============================================================================\n// RunExecution with ResumeContext tests\n// ============================================================================\n\nfunc TestRunExecutionResumeContext(t *testing.T) {\n\tif testing.Short() {\n\t\tt.Skip(\"Skipping integration test\")\n\t}\n\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\tctx := types.NewContext(context.Background(), testAuth())\n\n\tt.Run(\"resumes from task index with previous results\", func(t *testing.T) {\n\t\trobot := createRunTestRobot(t)\n\t\texec := createRunTestExecution(robot)\n\n\t\texec.Tasks = []types.Task{\n\t\t\t{\n\t\t\t\tID: \"task-001\",\n\t\t\t\tExecutorType: types.ExecutorAssistant,\n\t\t\t\tExecutorID: \"experts.text-writer\",\n\t\t\t\tMessages: []agentcontext.Message{\n\t\t\t\t\t{Role: agentcontext.RoleUser, Content: \"Write 'hello'\"},\n\t\t\t\t},\n\t\t\t\tOrder: 0,\n\t\t\t\tStatus: types.TaskCompleted,\n\t\t\t},\n\t\t\t{\n\t\t\t\tID: \"task-002\",\n\t\t\t\tExecutorType: types.ExecutorAssistant,\n\t\t\t\tExecutorID: \"experts.text-writer\",\n\t\t\t\tMessages: []agentcontext.Message{\n\t\t\t\t\t{Role: agentcontext.RoleUser, Content: \"Write 'world'\"},\n\t\t\t\t},\n\t\t\t\tOrder: 1,\n\t\t\t\tStatus: types.TaskPending,\n\t\t\t},\n\t\t\t{\n\t\t\t\tID: \"task-003\",\n\t\t\t\tExecutorType: types.ExecutorAssistant,\n\t\t\t\tExecutorID: \"experts.text-writer\",\n\t\t\t\tMessages: []agentcontext.Message{\n\t\t\t\t\t{Role: agentcontext.RoleUser, Content: \"Write '!'\"},\n\t\t\t\t},\n\t\t\t\tOrder: 2,\n\t\t\t\tStatus: types.TaskPending,\n\t\t\t},\n\t\t}\n\n\t\t// Simulate resume: task-001 already completed, resume from task-002\n\t\tpreviousResult := types.TaskResult{\n\t\t\tTaskID: \"task-001\",\n\t\t\tSuccess: true,\n\t\t\tOutput: \"hello\",\n\t\t\tDuration: 100,\n\t\t}\n\t\texec.ResumeContext = &types.ResumeContext{\n\t\t\tTaskIndex: 1,\n\t\t\tPreviousResults: []types.TaskResult{previousResult},\n\t\t}\n\n\t\te := standard.New()\n\t\terr := e.RunExecution(ctx, exec, nil)\n\n\t\trequire.NoError(t, err)\n\t\t// Should have 3 results: 1 from previous + 2 new\n\t\trequire.Len(t, exec.Results, 3)\n\n\t\tassert.Equal(t, \"task-001\", exec.Results[0].TaskID)\n\t\tassert.True(t, exec.Results[0].Success)\n\t\tassert.Equal(t, \"hello\", exec.Results[0].Output)\n\n\t\tassert.Equal(t, \"task-002\", exec.Results[1].TaskID)\n\t\tassert.True(t, exec.Results[1].Success)\n\n\t\tassert.Equal(t, \"task-003\", exec.Results[2].TaskID)\n\t\tassert.True(t, exec.Results[2].Success)\n\n\t\t// ResumeContext should be cleared after completion\n\t\tassert.Nil(t, exec.ResumeContext)\n\n\t\t// Only task-002 and task-003 should have been executed (check status)\n\t\tassert.Equal(t, types.TaskCompleted, exec.Tasks[1].Status)\n\t\tassert.Equal(t, types.TaskCompleted, exec.Tasks[2].Status)\n\t})\n\n\tt.Run(\"resumes from last task\", func(t *testing.T) {\n\t\trobot := createRunTestRobot(t)\n\t\texec := createRunTestExecution(robot)\n\n\t\texec.Tasks = []types.Task{\n\t\t\t{\n\t\t\t\tID: \"task-001\",\n\t\t\t\tExecutorType: types.ExecutorAssistant,\n\t\t\t\tExecutorID: \"experts.text-writer\",\n\t\t\t\tMessages: []agentcontext.Message{\n\t\t\t\t\t{Role: agentcontext.RoleUser, Content: \"Write 'hello'\"},\n\t\t\t\t},\n\t\t\t\tOrder: 0,\n\t\t\t\tStatus: types.TaskCompleted,\n\t\t\t},\n\t\t\t{\n\t\t\t\tID: \"task-002\",\n\t\t\t\tExecutorType: types.ExecutorAssistant,\n\t\t\t\tExecutorID: \"experts.text-writer\",\n\t\t\t\tMessages: []agentcontext.Message{\n\t\t\t\t\t{Role: agentcontext.RoleUser, Content: \"Write 'world'\"},\n\t\t\t\t},\n\t\t\t\tOrder: 1,\n\t\t\t\tStatus: types.TaskWaitingInput,\n\t\t\t},\n\t\t}\n\n\t\t// Resume from the last task\n\t\texec.ResumeContext = &types.ResumeContext{\n\t\t\tTaskIndex: 1,\n\t\t\tPreviousResults: []types.TaskResult{\n\t\t\t\t{TaskID: \"task-001\", Success: true, Output: \"hello\", Duration: 100},\n\t\t\t},\n\t\t}\n\n\t\te := standard.New()\n\t\terr := e.RunExecution(ctx, exec, nil)\n\n\t\trequire.NoError(t, err)\n\t\trequire.Len(t, exec.Results, 2)\n\t\tassert.True(t, exec.Results[1].Success)\n\t\tassert.Equal(t, types.TaskCompleted, exec.Tasks[1].Status)\n\t})\n}\n\n// ============================================================================\n// Suspend method tests (using Executor directly)\n// ============================================================================\n\nfunc TestSuspendExecution(t *testing.T) {\n\tt.Run(\"suspend sets waiting fields and returns ErrExecutionSuspended\", func(t *testing.T) {\n\t\trobot := &types.Robot{\n\t\t\tMemberID: \"test-robot-suspend\",\n\t\t\tTeamID: \"test-team-1\",\n\t\t\tDisplayName: \"Suspend Test Robot\",\n\t\t}\n\n\t\texec := &types.Execution{\n\t\t\tID: \"exec-suspend-001\",\n\t\t\tMemberID: robot.MemberID,\n\t\t\tTeamID: robot.TeamID,\n\t\t\tStatus: types.ExecRunning,\n\t\t\tPhase: types.PhaseRun,\n\t\t\tTasks: []types.Task{\n\t\t\t\t{ID: \"task-001\", Status: types.TaskRunning},\n\t\t\t\t{ID: \"task-002\", Status: types.TaskPending},\n\t\t\t},\n\t\t\tResults: []types.TaskResult{},\n\t\t}\n\t\texec.SetRobot(robot)\n\n\t\te := standard.NewWithConfig(executortypes.Config{SkipPersistence: true})\n\t\terr := e.Suspend(\n\t\t\ttypes.NewContext(context.Background(), nil),\n\t\t\texec, 0, \"What time range?\",\n\t\t)\n\n\t\tassert.ErrorIs(t, err, types.ErrExecutionSuspended)\n\t\tassert.Equal(t, types.ExecWaiting, exec.Status)\n\t\tassert.Equal(t, \"task-001\", exec.WaitingTaskID)\n\t\tassert.Equal(t, \"What time range?\", exec.WaitingQuestion)\n\t\tassert.NotNil(t, exec.WaitingSince)\n\t\tassert.NotNil(t, exec.ResumeContext)\n\t\tassert.Equal(t, 0, exec.ResumeContext.TaskIndex)\n\t\tassert.Equal(t, types.TaskWaitingInput, exec.Tasks[0].Status)\n\t})\n\n\tt.Run(\"suspend with out of range taskIndex is safe\", func(t *testing.T) {\n\t\trobot := &types.Robot{\n\t\t\tMemberID: \"test-robot-suspend-2\",\n\t\t\tTeamID: \"test-team-1\",\n\t\t}\n\t\texec := &types.Execution{\n\t\t\tID: \"exec-suspend-002\",\n\t\t\tMemberID: robot.MemberID,\n\t\t\tTeamID: robot.TeamID,\n\t\t\tStatus: types.ExecRunning,\n\t\t\tTasks: []types.Task{},\n\t\t\tResults: []types.TaskResult{},\n\t\t}\n\t\texec.SetRobot(robot)\n\n\t\te := standard.NewWithConfig(executortypes.Config{SkipPersistence: true})\n\t\terr := e.Suspend(\n\t\t\ttypes.NewContext(context.Background(), nil),\n\t\t\texec, 5, \"some question\",\n\t\t)\n\n\t\tassert.ErrorIs(t, err, types.ErrExecutionSuspended)\n\t\tassert.Equal(t, types.ExecWaiting, exec.Status)\n\t\tassert.Empty(t, exec.WaitingTaskID)\n\t})\n}\n\n// ============================================================================\n// ExecuteWithControl handles ErrExecutionSuspended\n// ============================================================================\n\nfunc TestExecuteWithControlSuspend(t *testing.T) {\n\tif testing.Short() {\n\t\tt.Skip(\"Skipping integration test\")\n\t}\n\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\tt.Run(\"returns ErrExecutionSuspended without marking as failed\", func(t *testing.T) {\n\t\t// This test requires a robot-need-input assistant that returns need_input.\n\t\t// Since we don't have one yet (Stage 6), we test the suspend path indirectly\n\t\t// by verifying that when RunExecution returns ErrExecutionSuspended,\n\t\t// ExecuteWithControl propagates it correctly.\n\t\t//\n\t\t// Full E2E test with real assistant will be in Stage 6.\n\n\t\trobot := &types.Robot{\n\t\t\tMemberID: \"test-robot-suspend-exec\",\n\t\t\tTeamID: \"test-team-1\",\n\t\t\tDisplayName: \"Suspend Exec Test\",\n\t\t\tConfig: &types.Config{\n\t\t\t\tIdentity: &types.Identity{\n\t\t\t\t\tRole: \"Test\",\n\t\t\t\t},\n\t\t\t\tResources: &types.Resources{\n\t\t\t\t\tPhases: map[types.Phase]string{},\n\t\t\t\t\tAgents: []string{\"experts.text-writer\"},\n\t\t\t\t},\n\t\t\t\tQuota: &types.Quota{Max: 5},\n\t\t\t},\n\t\t}\n\n\t\tctx := types.NewContext(context.Background(), testAuth())\n\t\te := standard.New()\n\n\t\t// Execute normally (no need_input expected from text-writer)\n\t\texec, err := e.Execute(ctx, robot, types.TriggerHuman, \"Write a greeting\")\n\t\tif err == types.ErrExecutionSuspended {\n\t\t\t// If somehow suspended, verify state\n\t\t\tassert.Equal(t, types.ExecWaiting, exec.Status)\n\t\t\tassert.NotEmpty(t, exec.WaitingQuestion)\n\t\t} else {\n\t\t\t// Normal completion\n\t\t\tassert.NoError(t, err)\n\t\t\tassert.NotNil(t, exec)\n\t\t}\n\t})\n}\n\n// ============================================================================\n// ResumeContext data structure tests\n// ============================================================================\n\nfunc TestResumeContext(t *testing.T) {\n\tt.Run(\"stores task index and previous results\", func(t *testing.T) {\n\t\trc := &types.ResumeContext{\n\t\t\tTaskIndex: 2,\n\t\t\tPreviousResults: []types.TaskResult{\n\t\t\t\t{TaskID: \"t1\", Success: true, Output: \"out1\"},\n\t\t\t\t{TaskID: \"t2\", Success: false, Error: \"some error\"},\n\t\t\t},\n\t\t}\n\t\tassert.Equal(t, 2, rc.TaskIndex)\n\t\tassert.Len(t, rc.PreviousResults, 2)\n\t\tassert.True(t, rc.PreviousResults[0].Success)\n\t\tassert.False(t, rc.PreviousResults[1].Success)\n\t})\n}\n\n// ============================================================================\n// NeedInput in TaskResult\n// ============================================================================\n\nfunc TestTaskResultNeedInput(t *testing.T) {\n\tt.Run(\"NeedInput fields are populated correctly\", func(t *testing.T) {\n\t\tresult := types.TaskResult{\n\t\t\tTaskID: \"task-001\",\n\t\t\tSuccess: true,\n\t\t\tOutput: \"some output\",\n\t\t\tNeedInput: true,\n\t\t\tInputQuestion: \"What time range?\",\n\t\t}\n\t\tassert.True(t, result.NeedInput)\n\t\tassert.Equal(t, \"What time range?\", result.InputQuestion)\n\t})\n}\n\n// ============================================================================\n// Execution status transitions for suspend/resume\n// ============================================================================\n\nfunc TestExecutionStatusTransitions(t *testing.T) {\n\tt.Run(\"ExecWaiting is a valid status\", func(t *testing.T) {\n\t\texec := &types.Execution{\n\t\t\tID: \"exec-001\",\n\t\t\tStatus: types.ExecWaiting,\n\t\t}\n\t\tassert.Equal(t, types.ExecStatus(\"waiting\"), exec.Status)\n\t})\n\n\tt.Run(\"TaskWaitingInput is a valid task status\", func(t *testing.T) {\n\t\ttask := types.Task{\n\t\t\tID: \"task-001\",\n\t\t\tStatus: types.TaskWaitingInput,\n\t\t}\n\t\tassert.Equal(t, types.TaskStatus(\"waiting_input\"), task.Status)\n\t})\n\n\tt.Run(\"Execution V2 fields are accessible\", func(t *testing.T) {\n\t\tnow := time.Now()\n\t\texec := &types.Execution{\n\t\t\tID: \"exec-v2-001\",\n\t\t\tChatID: \"robot_member1_exec001\",\n\t\t\tWaitingTaskID: \"task-002\",\n\t\t\tWaitingQuestion: \"What period?\",\n\t\t\tWaitingSince: &now,\n\t\t\tResumeContext: &types.ResumeContext{\n\t\t\t\tTaskIndex: 1,\n\t\t\t\tPreviousResults: []types.TaskResult{\n\t\t\t\t\t{TaskID: \"task-001\", Success: true},\n\t\t\t\t},\n\t\t\t},\n\t\t}\n\t\tassert.Equal(t, \"robot_member1_exec001\", exec.ChatID)\n\t\tassert.Equal(t, \"task-002\", exec.WaitingTaskID)\n\t\tassert.Equal(t, \"What period?\", exec.WaitingQuestion)\n\t\tassert.NotNil(t, exec.WaitingSince)\n\t\tassert.NotNil(t, exec.ResumeContext)\n\t\tassert.Equal(t, 1, exec.ResumeContext.TaskIndex)\n\t})\n}\n" }, { "path": "agent/robot/executor/standard/suspend_test.go", "content": "package standard\n\nimport (\n\t\"testing\"\n\n\t\"github.com/stretchr/testify/assert\"\n)\n\n// ============================================================================\n// detectNeedMoreInfo unit tests (internal — tests unexported function)\n// ============================================================================\n\nfunc TestDetectNeedMoreInfo(t *testing.T) {\n\tt.Run(\"returns false for nil result\", func(t *testing.T) {\n\t\tneedInput, question := detectNeedMoreInfo(nil)\n\t\tassert.False(t, needInput)\n\t\tassert.Empty(t, question)\n\t})\n\n\tt.Run(\"returns false for nil Next\", func(t *testing.T) {\n\t\tresult := &CallResult{Content: \"some text\"}\n\t\tneedInput, question := detectNeedMoreInfo(result)\n\t\tassert.False(t, needInput)\n\t\tassert.Empty(t, question)\n\t})\n\n\tt.Run(\"returns false for non-map Next\", func(t *testing.T) {\n\t\tresult := &CallResult{Next: \"just a string\"}\n\t\tneedInput, question := detectNeedMoreInfo(result)\n\t\tassert.False(t, needInput)\n\t\tassert.Empty(t, question)\n\t})\n\n\tt.Run(\"returns false when status is not need_input\", func(t *testing.T) {\n\t\tresult := &CallResult{\n\t\t\tNext: map[string]interface{}{\n\t\t\t\t\"status\": \"ok\",\n\t\t\t\t\"content\": \"everything is fine\",\n\t\t\t},\n\t\t}\n\t\tneedInput, question := detectNeedMoreInfo(result)\n\t\tassert.False(t, needInput)\n\t\tassert.Empty(t, question)\n\t})\n\n\tt.Run(\"returns true with question from Next\", func(t *testing.T) {\n\t\tresult := &CallResult{\n\t\t\tNext: map[string]interface{}{\n\t\t\t\t\"status\": \"need_input\",\n\t\t\t\t\"question\": \"What time range should I use?\",\n\t\t\t},\n\t\t}\n\t\tneedInput, question := detectNeedMoreInfo(result)\n\t\tassert.True(t, needInput)\n\t\tassert.Equal(t, \"What time range should I use?\", question)\n\t})\n\n\tt.Run(\"falls back to GetText when question is empty\", func(t *testing.T) {\n\t\tresult := &CallResult{\n\t\t\tContent: \"I need more information about the time range.\",\n\t\t\tNext: map[string]interface{}{\n\t\t\t\t\"status\": \"need_input\",\n\t\t\t},\n\t\t}\n\t\tneedInput, question := detectNeedMoreInfo(result)\n\t\tassert.True(t, needInput)\n\t\tassert.Equal(t, \"I need more information about the time range.\", question)\n\t})\n\n\tt.Run(\"returns true with empty question when both are empty\", func(t *testing.T) {\n\t\tresult := &CallResult{\n\t\t\tNext: map[string]interface{}{\n\t\t\t\t\"status\": \"need_input\",\n\t\t\t},\n\t\t}\n\t\tneedInput, question := detectNeedMoreInfo(result)\n\t\tassert.True(t, needInput)\n\t\tassert.Empty(t, question)\n\t})\n\n\tt.Run(\"unwraps data envelope from Next hook\", func(t *testing.T) {\n\t\tresult := &CallResult{\n\t\t\tNext: map[string]interface{}{\n\t\t\t\t\"data\": map[string]interface{}{\n\t\t\t\t\t\"status\": \"need_input\",\n\t\t\t\t\t\"question\": \"Which database should I query?\",\n\t\t\t\t},\n\t\t\t},\n\t\t}\n\t\tneedInput, question := detectNeedMoreInfo(result)\n\t\tassert.True(t, needInput)\n\t\tassert.Equal(t, \"Which database should I query?\", question)\n\t})\n}\n" }, { "path": "agent/robot/executor/standard/tasks.go", "content": "package standard\n\nimport (\n\t\"fmt\"\n\n\tagentcontext \"github.com/yaoapp/yao/agent/context\"\n\trobottypes \"github.com/yaoapp/yao/agent/robot/types\"\n)\n\n// RunTasks executes P2: Tasks phase\n// Calls the Tasks Agent to break down goals into executable tasks\n//\n// Input:\n// - Goals (from P1) with markdown content\n// - Available resources (Agents, MCP tools, KB, DB)\n//\n// Output:\n// - List of Task objects with executor assignments, expected outputs, and validation rules\nfunc (e *Executor) RunTasks(ctx *robottypes.Context, exec *robottypes.Execution, _ interface{}) error {\n\t// §18.2: confirming phase may have already populated Tasks — skip regeneration\n\tif len(exec.Tasks) > 0 {\n\t\treturn nil\n\t}\n\n\t// Get robot for resources\n\trobot := exec.GetRobot()\n\tif robot == nil {\n\t\treturn fmt.Errorf(\"robot not found in execution\")\n\t}\n\n\t// Update UI field with i18n\n\tlocale := getEffectiveLocale(robot, exec.Input)\n\te.updateUIFields(ctx, exec, \"\", getLocalizedMessage(locale, \"breaking_down_tasks\"))\n\n\t// Validate: Goals must exist (from P1)\n\tif exec.Goals == nil || exec.Goals.Content == \"\" {\n\t\treturn fmt.Errorf(\"goals not available for task planning\")\n\t}\n\n\t// Get agent ID for tasks phase\n\tagentID := \"__yao.tasks\" // default\n\tif robot.Config != nil && robot.Config.Resources != nil {\n\t\tagentID = robot.Config.Resources.GetPhaseAgent(robottypes.PhaseTasks)\n\t}\n\n\t// Build prompt with goals and available resources\n\tformatter := NewInputFormatter()\n\tuserContent := formatter.FormatGoals(exec.Goals, robot)\n\n\tif userContent == \"\" {\n\t\treturn fmt.Errorf(\"tasks agent (%s) received empty input for task planning\", agentID)\n\t}\n\n\t// Call agent\n\tcaller := NewAgentCaller()\n\tcaller.log = newExecLogger(robot, exec.ID)\n\tcaller.Connector = robot.LanguageModel\n\tresult, err := caller.CallWithMessages(ctx, agentID, userContent)\n\tif err != nil {\n\t\treturn fmt.Errorf(\"tasks agent (%s) call failed: %w\", agentID, err)\n\t}\n\n\t// Parse response as JSON\n\t// Tasks Agent returns: { \"tasks\": [...] }\n\tdata, err := result.GetJSON()\n\tif err != nil {\n\t\treturn fmt.Errorf(\"tasks agent (%s) returned invalid JSON: %w\", agentID, err)\n\t}\n\n\t// Extract tasks array\n\ttasksData, ok := data[\"tasks\"].([]interface{})\n\tif !ok || len(tasksData) == 0 {\n\t\treturn fmt.Errorf(\"tasks agent (%s) returned no tasks\", agentID)\n\t}\n\n\t// Parse tasks\n\ttasks, err := ParseTasks(tasksData)\n\tif err != nil {\n\t\treturn fmt.Errorf(\"tasks agent (%s) returned invalid task structure: %w\", agentID, err)\n\t}\n\n\t// Validate tasks\n\tif err := ValidateTasks(tasks); err != nil {\n\t\treturn fmt.Errorf(\"tasks validation failed: %w\", err)\n\t}\n\n\texec.Tasks = tasks\n\n\t// Log task overview for developer observability\n\tel := newExecLogger(robot, exec.ID)\n\tel.logTaskOverview(tasks)\n\n\treturn nil\n}\n\n// ParseTasks converts raw JSON array to []Task\n// Tasks are sorted by Order field after parsing\nfunc ParseTasks(data []interface{}) ([]robottypes.Task, error) {\n\ttasks := make([]robottypes.Task, 0, len(data))\n\n\tfor i, item := range data {\n\t\ttaskMap, ok := item.(map[string]interface{})\n\t\tif !ok {\n\t\t\treturn nil, fmt.Errorf(\"task %d is not a valid object\", i)\n\t\t}\n\n\t\ttask, err := ParseTask(taskMap, i)\n\t\tif err != nil {\n\t\t\treturn nil, fmt.Errorf(\"task %d: %w\", i, err)\n\t\t}\n\n\t\ttasks = append(tasks, *task)\n\t}\n\n\t// Sort tasks by Order field to ensure correct execution sequence\n\tSortTasksByOrder(tasks)\n\n\treturn tasks, nil\n}\n\n// ParseTask converts a map to Task struct\nfunc ParseTask(data map[string]interface{}, index int) (*robottypes.Task, error) {\n\ttask := &robottypes.Task{\n\t\tStatus: robottypes.TaskPending,\n\t\tOrder: index,\n\t}\n\n\t// Required: id\n\tif id, ok := data[\"id\"].(string); ok && id != \"\" {\n\t\ttask.ID = id\n\t} else {\n\t\ttask.ID = fmt.Sprintf(\"task-%03d\", index+1)\n\t}\n\n\t// Required: executor_type\n\tif execType, ok := data[\"executor_type\"].(string); ok {\n\t\ttask.ExecutorType = ParseExecutorType(execType)\n\t} else {\n\t\treturn nil, fmt.Errorf(\"missing executor_type\")\n\t}\n\n\t// Required: executor_id\n\tif execID, ok := data[\"executor_id\"].(string); ok && execID != \"\" {\n\t\ttask.ExecutorID = execID\n\t} else {\n\t\treturn nil, fmt.Errorf(\"missing executor_id\")\n\t}\n\n\t// Optional: goal_ref\n\tif goalRef, ok := data[\"goal_ref\"].(string); ok {\n\t\ttask.GoalRef = goalRef\n\t}\n\n\t// Optional: source (default to auto)\n\tif source, ok := data[\"source\"].(string); ok {\n\t\ttask.Source = robottypes.TaskSource(source)\n\t} else {\n\t\ttask.Source = robottypes.TaskSourceAuto\n\t}\n\n\t// Optional: order (override default)\n\tif order, ok := data[\"order\"].(float64); ok {\n\t\ttask.Order = int(order)\n\t}\n\n\t// Optional: messages (task instructions)\n\tif messages, ok := data[\"messages\"].([]interface{}); ok {\n\t\ttask.Messages = ParseMessages(messages)\n\t}\n\n\t// Optional: description - save to Description field and convert to message if no messages\n\tif desc, ok := data[\"description\"].(string); ok && desc != \"\" {\n\t\ttask.Description = desc\n\t\t// Also convert to Messages for execution if no explicit messages provided\n\t\tif len(task.Messages) == 0 {\n\t\t\ttask.Messages = []agentcontext.Message{\n\t\t\t\t{Role: agentcontext.RoleUser, Content: desc},\n\t\t\t}\n\t\t}\n\t}\n\n\t// Optional: args\n\tif args, ok := data[\"args\"].([]interface{}); ok {\n\t\ttask.Args = make([]any, len(args))\n\t\tcopy(task.Args, args)\n\t}\n\n\t// MCP-specific fields (required when executor_type is \"mcp\")\n\tif mcpServer, ok := data[\"mcp_server\"].(string); ok {\n\t\ttask.MCPServer = mcpServer\n\t}\n\tif mcpTool, ok := data[\"mcp_tool\"].(string); ok {\n\t\ttask.MCPTool = mcpTool\n\t}\n\n\t// Optional: expected_output (for P3 validation)\n\tif expectedOutput, ok := data[\"expected_output\"].(string); ok {\n\t\ttask.ExpectedOutput = expectedOutput\n\t}\n\n\t// Optional: validation_rules (for P3 validation)\n\tif rules, ok := data[\"validation_rules\"].([]interface{}); ok {\n\t\ttask.ValidationRules = make([]string, 0, len(rules))\n\t\tfor _, r := range rules {\n\t\t\tif s, ok := r.(string); ok {\n\t\t\t\ttask.ValidationRules = append(task.ValidationRules, s)\n\t\t\t}\n\t\t}\n\t}\n\n\treturn task, nil\n}\n\n// ParseMessages converts raw message array to []Message\nfunc ParseMessages(data []interface{}) []agentcontext.Message {\n\tmessages := make([]agentcontext.Message, 0, len(data))\n\n\tfor _, item := range data {\n\t\tmsgMap, ok := item.(map[string]interface{})\n\t\tif !ok {\n\t\t\tcontinue\n\t\t}\n\n\t\tmsg := agentcontext.Message{}\n\n\t\t// Role\n\t\tif role, ok := msgMap[\"role\"].(string); ok {\n\t\t\tmsg.Role = agentcontext.MessageRole(role)\n\t\t} else {\n\t\t\tmsg.Role = agentcontext.RoleUser\n\t\t}\n\n\t\t// Content\n\t\tif content, ok := msgMap[\"content\"].(string); ok {\n\t\t\tmsg.Content = content\n\t\t} else if content, ok := msgMap[\"content\"]; ok {\n\t\t\t// Handle non-string content (multimodal)\n\t\t\tmsg.Content = content\n\t\t}\n\n\t\tif msg.Content != nil {\n\t\t\tmessages = append(messages, msg)\n\t\t}\n\t}\n\n\treturn messages\n}\n\n// ParseExecutorType converts string to ExecutorType\nfunc ParseExecutorType(s string) robottypes.ExecutorType {\n\tswitch s {\n\tcase \"agent\", \"assistant\":\n\t\treturn robottypes.ExecutorAssistant\n\tcase \"mcp\":\n\t\treturn robottypes.ExecutorMCP\n\tcase \"process\":\n\t\treturn robottypes.ExecutorProcess\n\tdefault:\n\t\treturn robottypes.ExecutorAssistant // default to assistant\n\t}\n}\n\n// ValidateTasks validates the task list\nfunc ValidateTasks(tasks []robottypes.Task) error {\n\tif len(tasks) == 0 {\n\t\treturn fmt.Errorf(\"no tasks generated\")\n\t}\n\n\tseenIDs := make(map[string]bool)\n\n\tfor i, task := range tasks {\n\t\t// Check unique ID\n\t\tif seenIDs[task.ID] {\n\t\t\treturn fmt.Errorf(\"task %d: duplicate task ID '%s'\", i, task.ID)\n\t\t}\n\t\tseenIDs[task.ID] = true\n\n\t\t// Check executor\n\t\tif task.ExecutorID == \"\" {\n\t\t\treturn fmt.Errorf(\"task %d (%s): missing executor_id\", i, task.ID)\n\t\t}\n\n\t\t// Check messages or description\n\t\tif len(task.Messages) == 0 {\n\t\t\treturn fmt.Errorf(\"task %d (%s): missing messages or description\", i, task.ID)\n\t\t}\n\n\t\t// Note: Executor existence is NOT validated here\n\t\t// - ValidateExecutorExists() can be called separately if needed\n\t\t// - Unknown executors will fail at P3 runtime with clear error message\n\t\t// - This allows flexibility for dynamically registered executors\n\n\t\t// Note: Validation rules are optional\n\t\t// - P3 can still do basic validation without explicit rules\n\t}\n\n\treturn nil\n}\n\n// ValidateTasksWithResources validates tasks and checks executor existence\n// Returns a list of warnings for unknown executors (does not fail)\nfunc ValidateTasksWithResources(tasks []robottypes.Task, robot *robottypes.Robot) (warnings []string, err error) {\n\t// First do basic validation\n\tif err := ValidateTasks(tasks); err != nil {\n\t\treturn nil, err\n\t}\n\n\t// Then check executor existence (warnings only)\n\tfor _, task := range tasks {\n\t\tif !ValidateExecutorExists(task.ExecutorID, task.ExecutorType, robot) {\n\t\t\twarnings = append(warnings, fmt.Sprintf(\n\t\t\t\t\"task %s: executor '%s' (%s) not found in available resources\",\n\t\t\t\ttask.ID, task.ExecutorID, task.ExecutorType,\n\t\t\t))\n\t\t}\n\t}\n\n\treturn warnings, nil\n}\n\n// IsValidExecutorType checks if the executor type is valid\nfunc IsValidExecutorType(t robottypes.ExecutorType) bool {\n\tswitch t {\n\tcase robottypes.ExecutorAssistant, robottypes.ExecutorMCP, robottypes.ExecutorProcess:\n\t\treturn true\n\tdefault:\n\t\treturn false\n\t}\n}\n\n// SortTasksByOrder sorts tasks by their Order field (ascending)\n// This ensures tasks are executed in the correct sequence regardless of\n// the order they appear in the LLM response\nfunc SortTasksByOrder(tasks []robottypes.Task) {\n\tfor i := 0; i < len(tasks)-1; i++ {\n\t\tfor j := i + 1; j < len(tasks); j++ {\n\t\t\tif tasks[j].Order < tasks[i].Order {\n\t\t\t\ttasks[i], tasks[j] = tasks[j], tasks[i]\n\t\t\t}\n\t\t}\n\t}\n}\n\n// ValidateExecutorExists checks if the executor ID exists in available resources\n// This is an optional validation - tasks with unknown executors will still be created\n// but may fail during P3 execution\n// For MCP tasks, pass mcpServer as the second parameter (executorID is ignored for MCP)\nfunc ValidateExecutorExists(executorID string, executorType robottypes.ExecutorType, robot *robottypes.Robot) bool {\n\tif robot == nil || robot.Config == nil || robot.Config.Resources == nil {\n\t\treturn true // Skip validation if no resources configured\n\t}\n\n\tswitch executorType {\n\tcase robottypes.ExecutorAssistant:\n\t\tfor _, agent := range robot.Config.Resources.Agents {\n\t\t\tif agent == executorID {\n\t\t\t\treturn true\n\t\t\t}\n\t\t}\n\t\treturn false\n\n\tcase robottypes.ExecutorMCP:\n\t\t// For MCP, executorID can be either:\n\t\t// 1. The mcp_server value (new format)\n\t\t// 2. The combined mcp_server.mcp_tool format (for display)\n\t\t// We validate against mcp_server (the MCP server/client ID)\n\t\tfor _, mcp := range robot.Config.Resources.MCP {\n\t\t\tif mcp.ID == executorID {\n\t\t\t\treturn true\n\t\t\t}\n\t\t}\n\t\treturn false\n\n\tcase robottypes.ExecutorProcess:\n\t\t// Process executors are not validated against resources\n\t\t// They are validated at runtime by the Yao process system\n\t\treturn true\n\t}\n\n\treturn false\n}\n\n// ValidateMCPTask validates MCP task fields\n// Returns an error if mcp_server or mcp_tool is missing for MCP tasks\nfunc ValidateMCPTask(task *robottypes.Task) error {\n\tif task.ExecutorType != robottypes.ExecutorMCP {\n\t\treturn nil\n\t}\n\tif task.MCPServer == \"\" {\n\t\treturn fmt.Errorf(\"MCP task %s: mcp_server field is required\", task.ID)\n\t}\n\tif task.MCPTool == \"\" {\n\t\treturn fmt.Errorf(\"MCP task %s: mcp_tool field is required\", task.ID)\n\t}\n\treturn nil\n}\n" }, { "path": "agent/robot/executor/standard/tasks_test.go", "content": "package standard_test\n\nimport (\n\t\"context\"\n\t\"testing\"\n\t\"time\"\n\n\t\"github.com/stretchr/testify/assert\"\n\t\"github.com/stretchr/testify/require\"\n\tagentcontext \"github.com/yaoapp/yao/agent/context\"\n\t\"github.com/yaoapp/yao/agent/robot/executor/standard\"\n\t\"github.com/yaoapp/yao/agent/robot/types\"\n\t\"github.com/yaoapp/yao/agent/testutils\"\n)\n\n// ============================================================================\n// P2 Tasks Phase Tests\n// ============================================================================\n\nfunc TestRunTasksBasic(t *testing.T) {\n\tif testing.Short() {\n\t\tt.Skip(\"Skipping integration test\")\n\t}\n\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\tctx := types.NewContext(context.Background(), testAuth())\n\n\tt.Run(\"generates tasks from goals (clock trigger)\", func(t *testing.T) {\n\t\t// Create robot with tasks agent configured\n\t\trobot := createTasksTestRobot(t, \"robot.tasks\")\n\n\t\t// Create execution with goals (from P1)\n\t\texec := createTasksTestExecution(robot, types.TriggerClock)\n\t\texec.Goals = &types.Goals{\n\t\t\tContent: `## Goals\n\n1. [High] Analyze Q4 sales data and identify top performing products\n - Reason: Need to prepare quarterly report\n\n2. [Normal] Generate a summary report for management\n - Reason: Weekly review meeting tomorrow`,\n\t\t}\n\n\t\t// Run tasks phase\n\t\te := standard.New()\n\t\terr := e.RunTasks(ctx, exec, nil)\n\n\t\trequire.NoError(t, err)\n\t\trequire.NotNil(t, exec.Tasks)\n\t\tassert.NotEmpty(t, exec.Tasks)\n\n\t\t// Verify task structure\n\t\tfor i, task := range exec.Tasks {\n\t\t\tt.Logf(\"Task %d: ID=%s, ExecutorType=%s, ExecutorID=%s\", i, task.ID, task.ExecutorType, task.ExecutorID)\n\t\t\tassert.NotEmpty(t, task.ID, \"task should have ID\")\n\t\t\tassert.NotEmpty(t, task.ExecutorID, \"task should have executor ID\")\n\t\t\tassert.NotEmpty(t, task.Messages, \"task should have messages\")\n\t\t}\n\t})\n\n\tt.Run(\"includes expected output and validation rules\", func(t *testing.T) {\n\t\trobot := createTasksTestRobot(t, \"robot.tasks\")\n\t\texec := createTasksTestExecution(robot, types.TriggerClock)\n\t\texec.Goals = &types.Goals{\n\t\t\tContent: `## Goals\n\n1. [High] Fetch latest news about AI developments\n - Reason: Stay updated on industry trends\n\n2. [Normal] Summarize the key findings\n - Reason: Share with team`,\n\t\t}\n\n\t\te := standard.New()\n\t\terr := e.RunTasks(ctx, exec, nil)\n\n\t\trequire.NoError(t, err)\n\t\trequire.NotEmpty(t, exec.Tasks)\n\n\t\t// Check that at least one task has validation info\n\t\thasValidationInfo := false\n\t\tfor _, task := range exec.Tasks {\n\t\t\tif task.ExpectedOutput != \"\" || len(task.ValidationRules) > 0 {\n\t\t\t\thasValidationInfo = true\n\t\t\t\tt.Logf(\"Task %s has validation: expected_output=%q, rules=%v\",\n\t\t\t\t\ttask.ID, task.ExpectedOutput, task.ValidationRules)\n\t\t\t}\n\t\t}\n\n\t\t// Note: LLM might not always include validation rules, so we just log\n\t\tt.Logf(\"Tasks have validation info: %v\", hasValidationInfo)\n\t})\n}\n\nfunc TestRunTasksHumanTrigger(t *testing.T) {\n\tif testing.Short() {\n\t\tt.Skip(\"Skipping integration test\")\n\t}\n\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\tctx := types.NewContext(context.Background(), testAuth())\n\n\tt.Run(\"generates tasks from human-triggered goals\", func(t *testing.T) {\n\t\trobot := createTasksTestRobot(t, \"robot.tasks\")\n\t\texec := createTasksTestExecution(robot, types.TriggerHuman)\n\n\t\t// Goals from human request (P1 output)\n\t\texec.Goals = &types.Goals{\n\t\t\tContent: `## Goals\n\n1. [High] Research competitor pricing strategies\n - Reason: User requested competitive analysis\n\n2. [Normal] Create comparison report\n - Reason: User needs data for presentation`,\n\t\t}\n\n\t\te := standard.New()\n\t\terr := e.RunTasks(ctx, exec, nil)\n\n\t\trequire.NoError(t, err)\n\t\trequire.NotEmpty(t, exec.Tasks)\n\n\t\t// Tasks should relate to the goals\n\t\tfor _, task := range exec.Tasks {\n\t\t\tt.Logf(\"Task: %s -> %s\", task.ID, task.ExecutorID)\n\t\t}\n\t})\n}\n\nfunc TestRunTasksWithExpertAgents(t *testing.T) {\n\tif testing.Short() {\n\t\tt.Skip(\"Skipping integration test\")\n\t}\n\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\tctx := types.NewContext(context.Background(), testAuth())\n\n\tt.Run(\"assigns appropriate expert agents to tasks\", func(t *testing.T) {\n\t\trobot := createTasksTestRobot(t, \"robot.tasks\")\n\t\texec := createTasksTestExecution(robot, types.TriggerClock)\n\n\t\t// Goals that require different expert agents\n\t\texec.Goals = &types.Goals{\n\t\t\tContent: `## Goals\n\n1. [High] Analyze sales data from database\n - Reason: Quarterly review needed\n - Requires: Data analysis capabilities\n\n2. [Normal] Write executive summary report\n - Reason: Management presentation\n - Requires: Text generation capabilities\n\n3. [Low] Summarize key findings\n - Reason: Quick reference for team\n - Requires: Summarization capabilities`,\n\t\t}\n\n\t\te := standard.New()\n\t\terr := e.RunTasks(ctx, exec, nil)\n\n\t\trequire.NoError(t, err)\n\t\trequire.NotEmpty(t, exec.Tasks)\n\n\t\t// Log assigned executors\n\t\texecutorCounts := make(map[string]int)\n\t\tfor _, task := range exec.Tasks {\n\t\t\texecutorCounts[task.ExecutorID]++\n\t\t\tt.Logf(\"Task %s assigned to: %s (%s)\", task.ID, task.ExecutorID, task.ExecutorType)\n\t\t}\n\n\t\t// Verify different executors were assigned (not all to same agent)\n\t\tt.Logf(\"Executor distribution: %v\", executorCounts)\n\t})\n}\n\nfunc TestRunTasksErrorHandling(t *testing.T) {\n\tif testing.Short() {\n\t\tt.Skip(\"Skipping integration test\")\n\t}\n\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\tctx := types.NewContext(context.Background(), testAuth())\n\n\tt.Run(\"returns error when robot is nil\", func(t *testing.T) {\n\t\texec := &types.Execution{\n\t\t\tID: \"test-exec-1\",\n\t\t\tTriggerType: types.TriggerClock,\n\t\t}\n\t\t// Don't set robot\n\n\t\te := standard.New()\n\t\terr := e.RunTasks(ctx, exec, nil)\n\n\t\tassert.Error(t, err)\n\t\tassert.Contains(t, err.Error(), \"robot not found\")\n\t})\n\n\tt.Run(\"returns error when goals not available\", func(t *testing.T) {\n\t\trobot := createTasksTestRobot(t, \"robot.tasks\")\n\t\texec := createTasksTestExecution(robot, types.TriggerClock)\n\t\texec.Goals = nil // No goals\n\n\t\te := standard.New()\n\t\terr := e.RunTasks(ctx, exec, nil)\n\n\t\tassert.Error(t, err)\n\t\tassert.Contains(t, err.Error(), \"goals not available\")\n\t})\n\n\tt.Run(\"returns error when goals content is empty\", func(t *testing.T) {\n\t\trobot := createTasksTestRobot(t, \"robot.tasks\")\n\t\texec := createTasksTestExecution(robot, types.TriggerClock)\n\t\texec.Goals = &types.Goals{Content: \"\"} // Empty content\n\n\t\te := standard.New()\n\t\terr := e.RunTasks(ctx, exec, nil)\n\n\t\tassert.Error(t, err)\n\t\tassert.Contains(t, err.Error(), \"goals not available\")\n\t})\n\n\tt.Run(\"returns error when agent not found\", func(t *testing.T) {\n\t\trobot := &types.Robot{\n\t\t\tMemberID: \"test-robot-1\",\n\t\t\tTeamID: \"test-team-1\",\n\t\t\tConfig: &types.Config{\n\t\t\t\tIdentity: &types.Identity{Role: \"Test\"},\n\t\t\t\tResources: &types.Resources{\n\t\t\t\t\tPhases: map[types.Phase]string{\n\t\t\t\t\t\ttypes.PhaseTasks: \"non.existent.agent\",\n\t\t\t\t\t},\n\t\t\t\t},\n\t\t\t},\n\t\t}\n\t\texec := createTasksTestExecution(robot, types.TriggerClock)\n\t\texec.Goals = &types.Goals{Content: \"Test goals\"}\n\n\t\te := standard.New()\n\t\terr := e.RunTasks(ctx, exec, nil)\n\n\t\tassert.Error(t, err)\n\t\tassert.Contains(t, err.Error(), \"call failed\")\n\t})\n}\n\n// ============================================================================\n// ParseTasks Unit Tests\n// ============================================================================\n\nfunc TestParseTasks(t *testing.T) {\n\tt.Run(\"parses valid tasks array\", func(t *testing.T) {\n\t\tdata := []interface{}{\n\t\t\tmap[string]interface{}{\n\t\t\t\t\"id\": \"task-001\",\n\t\t\t\t\"goal_ref\": \"Goal 1\",\n\t\t\t\t\"executor_type\": \"agent\",\n\t\t\t\t\"executor_id\": \"experts.data-analyst\",\n\t\t\t\t\"messages\": []interface{}{\n\t\t\t\t\tmap[string]interface{}{\n\t\t\t\t\t\t\"role\": \"user\",\n\t\t\t\t\t\t\"content\": \"Analyze sales data\",\n\t\t\t\t\t},\n\t\t\t\t},\n\t\t\t\t\"expected_output\": \"JSON with sales metrics\",\n\t\t\t\t\"validation_rules\": []interface{}{\n\t\t\t\t\t// Natural language rules (matched by validator)\n\t\t\t\t\t\"output must be valid JSON\",\n\t\t\t\t\t\"must contain 'total_sales'\",\n\t\t\t\t\t// Structured rule: check field type\n\t\t\t\t\t`{\"type\": \"type\", \"path\": \"product_rankings\", \"value\": \"array\"}`,\n\t\t\t\t},\n\t\t\t\t\"order\": float64(0),\n\t\t\t},\n\t\t\tmap[string]interface{}{\n\t\t\t\t\"id\": \"task-002\",\n\t\t\t\t\"goal_ref\": \"Goal 1\",\n\t\t\t\t\"executor_type\": \"agent\",\n\t\t\t\t\"executor_id\": \"experts.text-writer\",\n\t\t\t\t\"description\": \"Generate report from analysis\",\n\t\t\t\t\"order\": float64(1),\n\t\t\t},\n\t\t}\n\n\t\ttasks, err := standard.ParseTasks(data)\n\n\t\trequire.NoError(t, err)\n\t\trequire.Len(t, tasks, 2)\n\n\t\t// First task\n\t\tassert.Equal(t, \"task-001\", tasks[0].ID)\n\t\tassert.Equal(t, \"Goal 1\", tasks[0].GoalRef)\n\t\tassert.Equal(t, types.ExecutorAssistant, tasks[0].ExecutorType)\n\t\tassert.Equal(t, \"experts.data-analyst\", tasks[0].ExecutorID)\n\t\tassert.Len(t, tasks[0].Messages, 1)\n\t\tassert.Equal(t, \"JSON with sales metrics\", tasks[0].ExpectedOutput)\n\t\tassert.Len(t, tasks[0].ValidationRules, 3)\n\t\tassert.Equal(t, 0, tasks[0].Order)\n\n\t\t// Second task\n\t\tassert.Equal(t, \"task-002\", tasks[1].ID)\n\t\tassert.Equal(t, \"experts.text-writer\", tasks[1].ExecutorID)\n\t\tassert.Equal(t, \"Generate report from analysis\", tasks[1].Description) // description saved to field\n\t\tassert.Len(t, tasks[1].Messages, 1) // description also converted to message\n\t\tassert.Equal(t, 1, tasks[1].Order)\n\t})\n\n\tt.Run(\"generates ID if missing\", func(t *testing.T) {\n\t\tdata := []interface{}{\n\t\t\tmap[string]interface{}{\n\t\t\t\t\"executor_type\": \"agent\",\n\t\t\t\t\"executor_id\": \"experts.summarizer\",\n\t\t\t\t\"description\": \"Summarize content\",\n\t\t\t},\n\t\t}\n\n\t\ttasks, err := standard.ParseTasks(data)\n\n\t\trequire.NoError(t, err)\n\t\trequire.Len(t, tasks, 1)\n\t\tassert.Equal(t, \"task-001\", tasks[0].ID)\n\t})\n\n\tt.Run(\"saves description to field and preserves explicit messages\", func(t *testing.T) {\n\t\tdata := []interface{}{\n\t\t\tmap[string]interface{}{\n\t\t\t\t\"id\": \"task-001\",\n\t\t\t\t\"executor_type\": \"agent\",\n\t\t\t\t\"executor_id\": \"experts.summarizer\",\n\t\t\t\t\"description\": \"Task description for UI\",\n\t\t\t\t\"messages\": []interface{}{\n\t\t\t\t\tmap[string]interface{}{\n\t\t\t\t\t\t\"role\": \"user\",\n\t\t\t\t\t\t\"content\": \"Explicit message content\",\n\t\t\t\t\t},\n\t\t\t\t},\n\t\t\t},\n\t\t}\n\n\t\ttasks, err := standard.ParseTasks(data)\n\n\t\trequire.NoError(t, err)\n\t\trequire.Len(t, tasks, 1)\n\n\t\t// Description should be saved to field\n\t\tassert.Equal(t, \"Task description for UI\", tasks[0].Description)\n\n\t\t// Explicit messages should be preserved (not overwritten by description)\n\t\tassert.Len(t, tasks[0].Messages, 1)\n\t\tcontent, ok := tasks[0].Messages[0].GetContentAsString()\n\t\tassert.True(t, ok)\n\t\tassert.Equal(t, \"Explicit message content\", content)\n\t})\n\n\tt.Run(\"converts description to message when no messages provided\", func(t *testing.T) {\n\t\tdata := []interface{}{\n\t\t\tmap[string]interface{}{\n\t\t\t\t\"id\": \"task-001\",\n\t\t\t\t\"executor_type\": \"agent\",\n\t\t\t\t\"executor_id\": \"experts.summarizer\",\n\t\t\t\t\"description\": \"Only description, no messages\",\n\t\t\t},\n\t\t}\n\n\t\ttasks, err := standard.ParseTasks(data)\n\n\t\trequire.NoError(t, err)\n\t\trequire.Len(t, tasks, 1)\n\n\t\t// Description should be saved to field\n\t\tassert.Equal(t, \"Only description, no messages\", tasks[0].Description)\n\n\t\t// Description should also be converted to message for execution\n\t\tassert.Len(t, tasks[0].Messages, 1)\n\t\tcontent, ok := tasks[0].Messages[0].GetContentAsString()\n\t\tassert.True(t, ok)\n\t\tassert.Equal(t, \"Only description, no messages\", content)\n\t})\n\n\tt.Run(\"returns error for missing executor_type\", func(t *testing.T) {\n\t\tdata := []interface{}{\n\t\t\tmap[string]interface{}{\n\t\t\t\t\"id\": \"task-001\",\n\t\t\t\t\"executor_id\": \"experts.summarizer\",\n\t\t\t\t\"description\": \"Summarize content\",\n\t\t\t},\n\t\t}\n\n\t\t_, err := standard.ParseTasks(data)\n\n\t\tassert.Error(t, err)\n\t\tassert.Contains(t, err.Error(), \"missing executor_type\")\n\t})\n\n\tt.Run(\"returns error for missing executor_id\", func(t *testing.T) {\n\t\tdata := []interface{}{\n\t\t\tmap[string]interface{}{\n\t\t\t\t\"id\": \"task-001\",\n\t\t\t\t\"executor_type\": \"agent\",\n\t\t\t\t\"description\": \"Summarize content\",\n\t\t\t},\n\t\t}\n\n\t\t_, err := standard.ParseTasks(data)\n\n\t\tassert.Error(t, err)\n\t\tassert.Contains(t, err.Error(), \"missing executor_id\")\n\t})\n\n\tt.Run(\"handles different executor types\", func(t *testing.T) {\n\t\tdata := []interface{}{\n\t\t\tmap[string]interface{}{\n\t\t\t\t\"executor_type\": \"agent\",\n\t\t\t\t\"executor_id\": \"test-agent\",\n\t\t\t\t\"description\": \"Agent task\",\n\t\t\t},\n\t\t\tmap[string]interface{}{\n\t\t\t\t\"executor_type\": \"assistant\",\n\t\t\t\t\"executor_id\": \"test-assistant\",\n\t\t\t\t\"description\": \"Assistant task\",\n\t\t\t},\n\t\t\tmap[string]interface{}{\n\t\t\t\t\"executor_type\": \"mcp\",\n\t\t\t\t\"executor_id\": \"test-mcp\",\n\t\t\t\t\"description\": \"MCP task\",\n\t\t\t},\n\t\t\tmap[string]interface{}{\n\t\t\t\t\"executor_type\": \"process\",\n\t\t\t\t\"executor_id\": \"test-process\",\n\t\t\t\t\"description\": \"Process task\",\n\t\t\t},\n\t\t}\n\n\t\ttasks, err := standard.ParseTasks(data)\n\n\t\trequire.NoError(t, err)\n\t\trequire.Len(t, tasks, 4)\n\n\t\tassert.Equal(t, types.ExecutorAssistant, tasks[0].ExecutorType)\n\t\tassert.Equal(t, types.ExecutorAssistant, tasks[1].ExecutorType) // assistant -> ExecutorAssistant\n\t\tassert.Equal(t, types.ExecutorMCP, tasks[2].ExecutorType)\n\t\tassert.Equal(t, types.ExecutorProcess, tasks[3].ExecutorType)\n\t})\n}\n\nfunc TestValidateTasks(t *testing.T) {\n\tt.Run(\"validates valid tasks\", func(t *testing.T) {\n\t\ttasks := []types.Task{\n\t\t\t{\n\t\t\t\tID: \"task-001\",\n\t\t\t\tExecutorType: types.ExecutorAssistant,\n\t\t\t\tExecutorID: \"experts.data-analyst\",\n\t\t\t\tMessages: []agentcontext.Message{\n\t\t\t\t\t{Role: agentcontext.RoleUser, Content: \"Analyze data\"},\n\t\t\t\t},\n\t\t\t},\n\t\t\t{\n\t\t\t\tID: \"task-002\",\n\t\t\t\tExecutorType: types.ExecutorAssistant,\n\t\t\t\tExecutorID: \"experts.text-writer\",\n\t\t\t\tMessages: []agentcontext.Message{\n\t\t\t\t\t{Role: agentcontext.RoleUser, Content: \"Write report\"},\n\t\t\t\t},\n\t\t\t},\n\t\t}\n\n\t\terr := standard.ValidateTasks(tasks)\n\t\tassert.NoError(t, err)\n\t})\n\n\tt.Run(\"returns error for empty tasks\", func(t *testing.T) {\n\t\ttasks := []types.Task{}\n\n\t\terr := standard.ValidateTasks(tasks)\n\n\t\tassert.Error(t, err)\n\t\tassert.Contains(t, err.Error(), \"no tasks generated\")\n\t})\n\n\tt.Run(\"returns error for duplicate IDs\", func(t *testing.T) {\n\t\ttasks := []types.Task{\n\t\t\t{\n\t\t\t\tID: \"task-001\",\n\t\t\t\tExecutorID: \"agent-1\",\n\t\t\t\tMessages: []agentcontext.Message{{Content: \"test\"}},\n\t\t\t},\n\t\t\t{\n\t\t\t\tID: \"task-001\", // duplicate\n\t\t\t\tExecutorID: \"agent-2\",\n\t\t\t\tMessages: []agentcontext.Message{{Content: \"test\"}},\n\t\t\t},\n\t\t}\n\n\t\terr := standard.ValidateTasks(tasks)\n\n\t\tassert.Error(t, err)\n\t\tassert.Contains(t, err.Error(), \"duplicate task ID\")\n\t})\n\n\tt.Run(\"returns error for missing executor_id\", func(t *testing.T) {\n\t\ttasks := []types.Task{\n\t\t\t{\n\t\t\t\tID: \"task-001\",\n\t\t\t\tExecutorID: \"\", // missing\n\t\t\t\tMessages: []agentcontext.Message{{Content: \"test\"}},\n\t\t\t},\n\t\t}\n\n\t\terr := standard.ValidateTasks(tasks)\n\n\t\tassert.Error(t, err)\n\t\tassert.Contains(t, err.Error(), \"missing executor_id\")\n\t})\n\n\tt.Run(\"returns error for missing messages\", func(t *testing.T) {\n\t\ttasks := []types.Task{\n\t\t\t{\n\t\t\t\tID: \"task-001\",\n\t\t\t\tExecutorID: \"agent-1\",\n\t\t\t\tMessages: []agentcontext.Message{}, // empty\n\t\t\t},\n\t\t}\n\n\t\terr := standard.ValidateTasks(tasks)\n\n\t\tassert.Error(t, err)\n\t\tassert.Contains(t, err.Error(), \"missing messages\")\n\t})\n}\n\nfunc TestParseExecutorType(t *testing.T) {\n\tt.Run(\"parses agent\", func(t *testing.T) {\n\t\tassert.Equal(t, types.ExecutorAssistant, standard.ParseExecutorType(\"agent\"))\n\t})\n\n\tt.Run(\"parses assistant\", func(t *testing.T) {\n\t\tassert.Equal(t, types.ExecutorAssistant, standard.ParseExecutorType(\"assistant\"))\n\t})\n\n\tt.Run(\"parses mcp\", func(t *testing.T) {\n\t\tassert.Equal(t, types.ExecutorMCP, standard.ParseExecutorType(\"mcp\"))\n\t})\n\n\tt.Run(\"parses process\", func(t *testing.T) {\n\t\tassert.Equal(t, types.ExecutorProcess, standard.ParseExecutorType(\"process\"))\n\t})\n\n\tt.Run(\"defaults to assistant for unknown\", func(t *testing.T) {\n\t\tassert.Equal(t, types.ExecutorAssistant, standard.ParseExecutorType(\"unknown\"))\n\t\tassert.Equal(t, types.ExecutorAssistant, standard.ParseExecutorType(\"\"))\n\t})\n}\n\nfunc TestIsValidExecutorType(t *testing.T) {\n\tt.Run(\"valid executor types\", func(t *testing.T) {\n\t\tassert.True(t, standard.IsValidExecutorType(types.ExecutorAssistant))\n\t\tassert.True(t, standard.IsValidExecutorType(types.ExecutorMCP))\n\t\tassert.True(t, standard.IsValidExecutorType(types.ExecutorProcess))\n\t})\n\n\tt.Run(\"invalid executor types\", func(t *testing.T) {\n\t\tassert.False(t, standard.IsValidExecutorType(types.ExecutorType(\"invalid\")))\n\t\tassert.False(t, standard.IsValidExecutorType(types.ExecutorType(\"\")))\n\t})\n}\n\nfunc TestSortTasksByOrder(t *testing.T) {\n\tt.Run(\"sorts tasks by order\", func(t *testing.T) {\n\t\ttasks := []types.Task{\n\t\t\t{ID: \"task-c\", Order: 2},\n\t\t\t{ID: \"task-a\", Order: 0},\n\t\t\t{ID: \"task-b\", Order: 1},\n\t\t}\n\n\t\tstandard.SortTasksByOrder(tasks)\n\n\t\tassert.Equal(t, \"task-a\", tasks[0].ID)\n\t\tassert.Equal(t, \"task-b\", tasks[1].ID)\n\t\tassert.Equal(t, \"task-c\", tasks[2].ID)\n\t})\n\n\tt.Run(\"handles already sorted tasks\", func(t *testing.T) {\n\t\ttasks := []types.Task{\n\t\t\t{ID: \"task-a\", Order: 0},\n\t\t\t{ID: \"task-b\", Order: 1},\n\t\t\t{ID: \"task-c\", Order: 2},\n\t\t}\n\n\t\tstandard.SortTasksByOrder(tasks)\n\n\t\tassert.Equal(t, \"task-a\", tasks[0].ID)\n\t\tassert.Equal(t, \"task-b\", tasks[1].ID)\n\t\tassert.Equal(t, \"task-c\", tasks[2].ID)\n\t})\n\n\tt.Run(\"handles single task\", func(t *testing.T) {\n\t\ttasks := []types.Task{\n\t\t\t{ID: \"task-a\", Order: 0},\n\t\t}\n\n\t\tstandard.SortTasksByOrder(tasks)\n\n\t\tassert.Len(t, tasks, 1)\n\t\tassert.Equal(t, \"task-a\", tasks[0].ID)\n\t})\n\n\tt.Run(\"handles empty tasks\", func(t *testing.T) {\n\t\ttasks := []types.Task{}\n\n\t\tstandard.SortTasksByOrder(tasks)\n\n\t\tassert.Empty(t, tasks)\n\t})\n}\n\nfunc TestValidateExecutorExists(t *testing.T) {\n\tt.Run(\"returns true for existing agent\", func(t *testing.T) {\n\t\trobot := &types.Robot{\n\t\t\tConfig: &types.Config{\n\t\t\t\tResources: &types.Resources{\n\t\t\t\t\tAgents: []string{\"experts.data-analyst\", \"experts.text-writer\"},\n\t\t\t\t},\n\t\t\t},\n\t\t}\n\n\t\tassert.True(t, standard.ValidateExecutorExists(\"experts.data-analyst\", types.ExecutorAssistant, robot))\n\t\tassert.True(t, standard.ValidateExecutorExists(\"experts.text-writer\", types.ExecutorAssistant, robot))\n\t})\n\n\tt.Run(\"returns false for non-existing agent\", func(t *testing.T) {\n\t\trobot := &types.Robot{\n\t\t\tConfig: &types.Config{\n\t\t\t\tResources: &types.Resources{\n\t\t\t\t\tAgents: []string{\"experts.data-analyst\"},\n\t\t\t\t},\n\t\t\t},\n\t\t}\n\n\t\tassert.False(t, standard.ValidateExecutorExists(\"experts.unknown\", types.ExecutorAssistant, robot))\n\t})\n\n\tt.Run(\"returns true for existing MCP\", func(t *testing.T) {\n\t\trobot := &types.Robot{\n\t\t\tConfig: &types.Config{\n\t\t\t\tResources: &types.Resources{\n\t\t\t\t\tMCP: []types.MCPConfig{\n\t\t\t\t\t\t{ID: \"database\"},\n\t\t\t\t\t\t{ID: \"email\"},\n\t\t\t\t\t},\n\t\t\t\t},\n\t\t\t},\n\t\t}\n\n\t\tassert.True(t, standard.ValidateExecutorExists(\"database\", types.ExecutorMCP, robot))\n\t\tassert.True(t, standard.ValidateExecutorExists(\"email\", types.ExecutorMCP, robot))\n\t})\n\n\tt.Run(\"returns false for non-existing MCP\", func(t *testing.T) {\n\t\trobot := &types.Robot{\n\t\t\tConfig: &types.Config{\n\t\t\t\tResources: &types.Resources{\n\t\t\t\t\tMCP: []types.MCPConfig{\n\t\t\t\t\t\t{ID: \"database\"},\n\t\t\t\t\t},\n\t\t\t\t},\n\t\t\t},\n\t\t}\n\n\t\tassert.False(t, standard.ValidateExecutorExists(\"unknown\", types.ExecutorMCP, robot))\n\t})\n\n\tt.Run(\"returns true for process (not validated)\", func(t *testing.T) {\n\t\trobot := &types.Robot{\n\t\t\tConfig: &types.Config{\n\t\t\t\tResources: &types.Resources{},\n\t\t\t},\n\t\t}\n\n\t\tassert.True(t, standard.ValidateExecutorExists(\"models.user.Find\", types.ExecutorProcess, robot))\n\t})\n\n\tt.Run(\"returns true when robot is nil\", func(t *testing.T) {\n\t\tassert.True(t, standard.ValidateExecutorExists(\"any\", types.ExecutorAssistant, nil))\n\t})\n\n\tt.Run(\"returns true when resources is nil\", func(t *testing.T) {\n\t\trobot := &types.Robot{\n\t\t\tConfig: &types.Config{},\n\t\t}\n\n\t\tassert.True(t, standard.ValidateExecutorExists(\"any\", types.ExecutorAssistant, robot))\n\t})\n}\n\nfunc TestValidateTasksWithResources(t *testing.T) {\n\tt.Run(\"returns no warnings for valid tasks\", func(t *testing.T) {\n\t\trobot := &types.Robot{\n\t\t\tConfig: &types.Config{\n\t\t\t\tResources: &types.Resources{\n\t\t\t\t\tAgents: []string{\"experts.data-analyst\", \"experts.text-writer\"},\n\t\t\t\t},\n\t\t\t},\n\t\t}\n\t\ttasks := []types.Task{\n\t\t\t{\n\t\t\t\tID: \"task-001\",\n\t\t\t\tExecutorType: types.ExecutorAssistant,\n\t\t\t\tExecutorID: \"experts.data-analyst\",\n\t\t\t\tMessages: []agentcontext.Message{{Content: \"test\"}},\n\t\t\t},\n\t\t}\n\n\t\twarnings, err := standard.ValidateTasksWithResources(tasks, robot)\n\n\t\tassert.NoError(t, err)\n\t\tassert.Empty(t, warnings)\n\t})\n\n\tt.Run(\"returns warnings for unknown executor\", func(t *testing.T) {\n\t\trobot := &types.Robot{\n\t\t\tConfig: &types.Config{\n\t\t\t\tResources: &types.Resources{\n\t\t\t\t\tAgents: []string{\"experts.data-analyst\"},\n\t\t\t\t},\n\t\t\t},\n\t\t}\n\t\ttasks := []types.Task{\n\t\t\t{\n\t\t\t\tID: \"task-001\",\n\t\t\t\tExecutorType: types.ExecutorAssistant,\n\t\t\t\tExecutorID: \"experts.unknown\",\n\t\t\t\tMessages: []agentcontext.Message{{Content: \"test\"}},\n\t\t\t},\n\t\t}\n\n\t\twarnings, err := standard.ValidateTasksWithResources(tasks, robot)\n\n\t\tassert.NoError(t, err)\n\t\tassert.Len(t, warnings, 1)\n\t\tassert.Contains(t, warnings[0], \"experts.unknown\")\n\t\tassert.Contains(t, warnings[0], \"not found\")\n\t})\n\n\tt.Run(\"returns error for invalid tasks\", func(t *testing.T) {\n\t\trobot := &types.Robot{}\n\t\ttasks := []types.Task{} // empty\n\n\t\t_, err := standard.ValidateTasksWithResources(tasks, robot)\n\n\t\tassert.Error(t, err)\n\t\tassert.Contains(t, err.Error(), \"no tasks generated\")\n\t})\n}\n\n// ============================================================================\n// InputFormatter Tests for P2\n// ============================================================================\n\nfunc TestInputFormatterFormatGoalsForTasks(t *testing.T) {\n\tformatter := standard.NewInputFormatter()\n\n\tt.Run(\"formats goals with resources\", func(t *testing.T) {\n\t\tgoals := &types.Goals{\n\t\t\tContent: \"## Goals\\n\\n1. [High] Analyze data\\n2. [Normal] Write report\",\n\t\t}\n\t\trobot := &types.Robot{\n\t\t\tMemberID: \"test-robot\",\n\t\t\tConfig: &types.Config{\n\t\t\t\tResources: &types.Resources{\n\t\t\t\t\tAgents: []string{\"experts.data-analyst\", \"experts.text-writer\"},\n\t\t\t\t},\n\t\t\t},\n\t\t}\n\n\t\tcontent := formatter.FormatGoals(goals, robot)\n\n\t\tassert.Contains(t, content, \"## Goals\")\n\t\tassert.Contains(t, content, \"[High] Analyze data\")\n\t\tassert.Contains(t, content, \"## Available Resources\")\n\t\tassert.Contains(t, content, \"experts.data-analyst\")\n\t\tassert.Contains(t, content, \"experts.text-writer\")\n\t})\n\n\tt.Run(\"formats goals without robot\", func(t *testing.T) {\n\t\tgoals := &types.Goals{\n\t\t\tContent: \"## Goals\\n\\n1. Test goal\",\n\t\t}\n\n\t\tcontent := formatter.FormatGoals(goals, nil)\n\n\t\tassert.Contains(t, content, \"## Goals\")\n\t\tassert.Contains(t, content, \"Test goal\")\n\t\tassert.NotContains(t, content, \"## Available Resources\")\n\t})\n\n\tt.Run(\"formats goals with delivery target\", func(t *testing.T) {\n\t\tgoals := &types.Goals{\n\t\t\tContent: \"## Goals\\n\\n1. Generate weekly report\",\n\t\t\tDelivery: &types.DeliveryTarget{\n\t\t\t\tType: types.DeliveryEmail,\n\t\t\t\tRecipients: []string{\"team@example.com\", \"manager@example.com\"},\n\t\t\t\tFormat: \"markdown\",\n\t\t\t\tTemplate: \"weekly-report\",\n\t\t\t},\n\t\t}\n\t\trobot := &types.Robot{\n\t\t\tMemberID: \"test-robot\",\n\t\t\tConfig: &types.Config{\n\t\t\t\tResources: &types.Resources{\n\t\t\t\t\tAgents: []string{\"experts.text-writer\"},\n\t\t\t\t},\n\t\t\t},\n\t\t}\n\n\t\tcontent := formatter.FormatGoals(goals, robot)\n\n\t\tassert.Contains(t, content, \"## Goals\")\n\t\tassert.Contains(t, content, \"## Delivery Target\")\n\t\tassert.Contains(t, content, \"email\")\n\t\tassert.Contains(t, content, \"team@example.com\")\n\t\tassert.Contains(t, content, \"manager@example.com\")\n\t\tassert.Contains(t, content, \"markdown\")\n\t\tassert.Contains(t, content, \"weekly-report\")\n\t\tassert.Contains(t, content, \"Design tasks to produce output suitable\")\n\t})\n\n\tt.Run(\"formats goals without delivery target\", func(t *testing.T) {\n\t\tgoals := &types.Goals{\n\t\t\tContent: \"## Goals\\n\\n1. Test goal\",\n\t\t\tDelivery: nil,\n\t\t}\n\n\t\tcontent := formatter.FormatGoals(goals, nil)\n\n\t\tassert.Contains(t, content, \"## Goals\")\n\t\tassert.NotContains(t, content, \"## Delivery Target\")\n\t})\n}\n\n// ============================================================================\n// Helper Functions\n// ============================================================================\n\n// createTasksTestRobot creates a test robot with specified tasks agent\n// Includes available expert agents for task assignment\nfunc createTasksTestRobot(t *testing.T, agentID string) *types.Robot {\n\tt.Helper()\n\treturn &types.Robot{\n\t\tMemberID: \"test-robot-1\",\n\t\tTeamID: \"test-team-1\",\n\t\tDisplayName: \"Test Robot\",\n\t\tConfig: &types.Config{\n\t\t\tIdentity: &types.Identity{\n\t\t\t\tRole: \"Test Assistant\",\n\t\t\t\tDuties: []string{\"Testing\", \"Data Analysis\", \"Report Generation\"},\n\t\t\t},\n\t\t\tResources: &types.Resources{\n\t\t\t\tPhases: map[types.Phase]string{\n\t\t\t\t\ttypes.PhaseTasks: agentID,\n\t\t\t\t},\n\t\t\t\t// Available expert agents that can be assigned to tasks\n\t\t\t\tAgents: []string{\n\t\t\t\t\t\"experts.data-analyst\",\n\t\t\t\t\t\"experts.summarizer\",\n\t\t\t\t\t\"experts.text-writer\",\n\t\t\t\t\t\"experts.web-reader\",\n\t\t\t\t},\n\t\t\t},\n\t\t},\n\t}\n}\n\n// createTasksTestExecution creates a test execution for tasks phase\nfunc createTasksTestExecution(robot *types.Robot, trigger types.TriggerType) *types.Execution {\n\texec := &types.Execution{\n\t\tID: \"test-exec-tasks-1\",\n\t\tMemberID: robot.MemberID,\n\t\tTeamID: robot.TeamID,\n\t\tTriggerType: trigger,\n\t\tStartTime: time.Now(),\n\t\tStatus: types.ExecRunning,\n\t\tPhase: types.PhaseTasks,\n\t}\n\texec.SetRobot(robot)\n\treturn exec\n}\n\n// Note: testAuth is defined in goals_test.go in the same package\n" }, { "path": "agent/robot/executor/standard/ui_fields_test.go", "content": "package standard\n\nimport (\n\t\"testing\"\n\n\t\"github.com/stretchr/testify/assert\"\n\tagentcontext \"github.com/yaoapp/yao/agent/context\"\n\trobottypes \"github.com/yaoapp/yao/agent/robot/types\"\n)\n\n// ============================================================================\n// getEffectiveLocale Tests\n// ============================================================================\n\nfunc TestGetEffectiveLocale(t *testing.T) {\n\tt.Run(\"returns_input_locale_when_provided\", func(t *testing.T) {\n\t\trobot := &robottypes.Robot{\n\t\t\tConfig: &robottypes.Config{\n\t\t\t\tDefaultLocale: \"en\",\n\t\t\t},\n\t\t}\n\t\tinput := &robottypes.TriggerInput{\n\t\t\tLocale: \"zh\",\n\t\t}\n\n\t\tlocale := getEffectiveLocale(robot, input)\n\t\tassert.Equal(t, \"zh\", locale)\n\t})\n\n\tt.Run(\"returns_robot_default_locale_when_input_locale_empty\", func(t *testing.T) {\n\t\trobot := &robottypes.Robot{\n\t\t\tConfig: &robottypes.Config{\n\t\t\t\tDefaultLocale: \"zh\",\n\t\t\t},\n\t\t}\n\t\tinput := &robottypes.TriggerInput{\n\t\t\tLocale: \"\",\n\t\t}\n\n\t\tlocale := getEffectiveLocale(robot, input)\n\t\tassert.Equal(t, \"zh\", locale)\n\t})\n\n\tt.Run(\"returns_system_default_when_no_locale_configured\", func(t *testing.T) {\n\t\trobot := &robottypes.Robot{\n\t\t\tConfig: &robottypes.Config{},\n\t\t}\n\t\tinput := &robottypes.TriggerInput{}\n\n\t\tlocale := getEffectiveLocale(robot, input)\n\t\tassert.Equal(t, \"en\", locale)\n\t})\n\n\tt.Run(\"returns_system_default_when_robot_config_nil\", func(t *testing.T) {\n\t\trobot := &robottypes.Robot{}\n\t\tinput := &robottypes.TriggerInput{}\n\n\t\tlocale := getEffectiveLocale(robot, input)\n\t\tassert.Equal(t, \"en\", locale)\n\t})\n\n\tt.Run(\"returns_system_default_when_robot_nil\", func(t *testing.T) {\n\t\tinput := &robottypes.TriggerInput{}\n\n\t\tlocale := getEffectiveLocale(nil, input)\n\t\tassert.Equal(t, \"en\", locale)\n\t})\n\n\tt.Run(\"returns_system_default_when_input_nil\", func(t *testing.T) {\n\t\trobot := &robottypes.Robot{}\n\n\t\tlocale := getEffectiveLocale(robot, nil)\n\t\tassert.Equal(t, \"en\", locale)\n\t})\n}\n\n// ============================================================================\n// getLocalizedMessage Tests\n// ============================================================================\n\nfunc TestGetLocalizedMessage(t *testing.T) {\n\tt.Run(\"returns_english_message_for_en_locale\", func(t *testing.T) {\n\t\tmsg := getLocalizedMessage(\"en\", \"preparing\")\n\t\tassert.Equal(t, \"Preparing...\", msg)\n\t})\n\n\tt.Run(\"returns_chinese_message_for_zh_locale\", func(t *testing.T) {\n\t\tmsg := getLocalizedMessage(\"zh\", \"preparing\")\n\t\tassert.Equal(t, \"准备中...\", msg)\n\t})\n\n\tt.Run(\"returns_english_fallback_for_unknown_locale\", func(t *testing.T) {\n\t\tmsg := getLocalizedMessage(\"fr\", \"preparing\")\n\t\tassert.Equal(t, \"Preparing...\", msg)\n\t})\n\n\tt.Run(\"returns_key_for_unknown_message\", func(t *testing.T) {\n\t\tmsg := getLocalizedMessage(\"en\", \"unknown_key\")\n\t\tassert.Equal(t, \"unknown_key\", msg)\n\t})\n\n\tt.Run(\"all_english_messages_exist\", func(t *testing.T) {\n\t\tkeys := []string{\n\t\t\t\"preparing\", \"starting\", \"scheduled_execution\",\n\t\t\t\"event_prefix\", \"event_triggered\", \"analyzing_context\",\n\t\t\t\"planning_goals\", \"breaking_down_tasks\",\n\t\t\t\"generating_delivery\", \"sending_delivery\", \"learning_from_exec\",\n\t\t\t\"completed\", \"failed_prefix\", \"task_prefix\",\n\t\t\t// Phase names for failure messages\n\t\t\t\"phase_inspiration\", \"phase_goals\", \"phase_tasks\",\n\t\t\t\"phase_run\", \"phase_delivery\", \"phase_learning\",\n\t\t}\n\t\tfor _, key := range keys {\n\t\t\tmsg := getLocalizedMessage(\"en\", key)\n\t\t\tassert.NotEqual(t, key, msg, \"English message should exist for key: %s\", key)\n\t\t}\n\t})\n\n\tt.Run(\"all_chinese_messages_exist\", func(t *testing.T) {\n\t\tkeys := []string{\n\t\t\t\"preparing\", \"starting\", \"scheduled_execution\",\n\t\t\t\"event_prefix\", \"event_triggered\", \"analyzing_context\",\n\t\t\t\"planning_goals\", \"breaking_down_tasks\",\n\t\t\t\"generating_delivery\", \"sending_delivery\", \"learning_from_exec\",\n\t\t\t\"completed\", \"failed_prefix\", \"task_prefix\",\n\t\t\t// Phase names for failure messages\n\t\t\t\"phase_inspiration\", \"phase_goals\", \"phase_tasks\",\n\t\t\t\"phase_run\", \"phase_delivery\", \"phase_learning\",\n\t\t}\n\t\tfor _, key := range keys {\n\t\t\tmsg := getLocalizedMessage(\"zh\", key)\n\t\t\tassert.NotEqual(t, key, msg, \"Chinese message should exist for key: %s\", key)\n\t\t}\n\t})\n\n\tt.Run(\"failure_message_is_concise\", func(t *testing.T) {\n\t\t// Test that failure messages use phase names, not full error text\n\t\tenFailure := getLocalizedMessage(\"en\", \"failed_prefix\") + getLocalizedMessage(\"en\", \"phase_inspiration\")\n\t\tassert.Equal(t, \"Failed at inspiration\", enFailure)\n\n\t\tzhFailure := getLocalizedMessage(\"zh\", \"failed_prefix\") + getLocalizedMessage(\"zh\", \"phase_inspiration\")\n\t\tassert.Equal(t, \"失败于灵感阶段\", zhFailure)\n\t})\n}\n\n// ============================================================================\n// initUIFields Tests\n// ============================================================================\n\nfunc TestInitUIFields(t *testing.T) {\n\texecutor := New()\n\n\tt.Run(\"human_trigger_extracts_name_from_message\", func(t *testing.T) {\n\t\trobot := &robottypes.Robot{\n\t\t\tConfig: &robottypes.Config{DefaultLocale: \"en\"},\n\t\t}\n\t\tinput := &robottypes.TriggerInput{\n\t\t\tMessages: []agentcontext.Message{\n\t\t\t\t{Role: agentcontext.RoleUser, Content: \"Please analyze the sales data\"},\n\t\t\t},\n\t\t}\n\n\t\tname, currentTaskName := executor.initUIFields(robottypes.TriggerHuman, input, robot)\n\t\tassert.Equal(t, \"Please analyze the sales data\", name)\n\t\tassert.Equal(t, \"Starting...\", currentTaskName)\n\t})\n\n\tt.Run(\"human_trigger_truncates_long_message\", func(t *testing.T) {\n\t\trobot := &robottypes.Robot{\n\t\t\tConfig: &robottypes.Config{DefaultLocale: \"en\"},\n\t\t}\n\t\tlongMessage := \"This is a very long message that exceeds one hundred characters and should be truncated with an ellipsis at the end\"\n\t\tinput := &robottypes.TriggerInput{\n\t\t\tMessages: []agentcontext.Message{\n\t\t\t\t{Role: agentcontext.RoleUser, Content: longMessage},\n\t\t\t},\n\t\t}\n\n\t\tname, _ := executor.initUIFields(robottypes.TriggerHuman, input, robot)\n\t\tassert.LessOrEqual(t, len(name), 103) // 100 chars + \"...\"\n\t\tassert.True(t, len(name) > 100 || name == longMessage[:100]+\"...\")\n\t})\n\n\tt.Run(\"clock_trigger_uses_scheduled_execution\", func(t *testing.T) {\n\t\trobot := &robottypes.Robot{\n\t\t\tConfig: &robottypes.Config{DefaultLocale: \"en\"},\n\t\t}\n\t\tinput := &robottypes.TriggerInput{}\n\n\t\tname, currentTaskName := executor.initUIFields(robottypes.TriggerClock, input, robot)\n\t\tassert.Equal(t, \"Scheduled execution\", name)\n\t\tassert.Equal(t, \"Starting...\", currentTaskName)\n\t})\n\n\tt.Run(\"clock_trigger_chinese_locale\", func(t *testing.T) {\n\t\trobot := &robottypes.Robot{\n\t\t\tConfig: &robottypes.Config{DefaultLocale: \"zh\"},\n\t\t}\n\t\tinput := &robottypes.TriggerInput{}\n\n\t\tname, currentTaskName := executor.initUIFields(robottypes.TriggerClock, input, robot)\n\t\tassert.Equal(t, \"定时执行\", name)\n\t\tassert.Equal(t, \"启动中...\", currentTaskName)\n\t})\n\n\tt.Run(\"event_trigger_with_event_type\", func(t *testing.T) {\n\t\trobot := &robottypes.Robot{\n\t\t\tConfig: &robottypes.Config{DefaultLocale: \"en\"},\n\t\t}\n\t\tinput := &robottypes.TriggerInput{\n\t\t\tEventType: \"lead.created\",\n\t\t}\n\n\t\tname, currentTaskName := executor.initUIFields(robottypes.TriggerEvent, input, robot)\n\t\tassert.Equal(t, \"Event: lead.created\", name)\n\t\tassert.Equal(t, \"Starting...\", currentTaskName)\n\t})\n\n\tt.Run(\"event_trigger_without_event_type\", func(t *testing.T) {\n\t\trobot := &robottypes.Robot{\n\t\t\tConfig: &robottypes.Config{DefaultLocale: \"en\"},\n\t\t}\n\t\tinput := &robottypes.TriggerInput{}\n\n\t\tname, _ := executor.initUIFields(robottypes.TriggerEvent, input, robot)\n\t\tassert.Equal(t, \"Event triggered\", name)\n\t})\n\n\tt.Run(\"event_trigger_chinese_locale\", func(t *testing.T) {\n\t\trobot := &robottypes.Robot{\n\t\t\tConfig: &robottypes.Config{DefaultLocale: \"zh\"},\n\t\t}\n\t\tinput := &robottypes.TriggerInput{\n\t\t\tEventType: \"order.placed\",\n\t\t}\n\n\t\tname, _ := executor.initUIFields(robottypes.TriggerEvent, input, robot)\n\t\tassert.Equal(t, \"事件: order.placed\", name)\n\t})\n\n\tt.Run(\"input_locale_overrides_robot_default\", func(t *testing.T) {\n\t\trobot := &robottypes.Robot{\n\t\t\tConfig: &robottypes.Config{DefaultLocale: \"en\"},\n\t\t}\n\t\tinput := &robottypes.TriggerInput{\n\t\t\tLocale: \"zh\",\n\t\t}\n\n\t\tname, currentTaskName := executor.initUIFields(robottypes.TriggerClock, input, robot)\n\t\tassert.Equal(t, \"定时执行\", name)\n\t\tassert.Equal(t, \"启动中...\", currentTaskName)\n\t})\n}\n\n// ============================================================================\n// extractGoalName Tests\n// ============================================================================\n\nfunc TestExtractGoalName(t *testing.T) {\n\tt.Run(\"extracts_first_line_from_content\", func(t *testing.T) {\n\t\tgoals := &robottypes.Goals{\n\t\t\tContent: \"Generate monthly sales report\\nAnalyze trends\\nSend to stakeholders\",\n\t\t}\n\n\t\tname := extractGoalName(goals)\n\t\tassert.Equal(t, \"Generate monthly sales report\", name)\n\t})\n\n\tt.Run(\"returns_empty_for_nil_goals\", func(t *testing.T) {\n\t\tname := extractGoalName(nil)\n\t\tassert.Equal(t, \"\", name)\n\t})\n\n\tt.Run(\"returns_empty_for_empty_content\", func(t *testing.T) {\n\t\tgoals := &robottypes.Goals{\n\t\t\tContent: \"\",\n\t\t}\n\n\t\tname := extractGoalName(goals)\n\t\tassert.Equal(t, \"\", name)\n\t})\n\n\tt.Run(\"truncates_long_first_line\", func(t *testing.T) {\n\t\tlongLine := \"This is an extremely long goal description that exceeds one hundred and fifty characters and should be truncated with an ellipsis at the end to keep the display manageable\"\n\t\tgoals := &robottypes.Goals{\n\t\t\tContent: longLine,\n\t\t}\n\n\t\tname := extractGoalName(goals)\n\t\tassert.LessOrEqual(t, len(name), 153) // 150 chars + \"...\"\n\t})\n\n\tt.Run(\"handles_single_line_content\", func(t *testing.T) {\n\t\tgoals := &robottypes.Goals{\n\t\t\tContent: \"Single line goal\",\n\t\t}\n\n\t\tname := extractGoalName(goals)\n\t\tassert.Equal(t, \"Single line goal\", name)\n\t})\n\n\tt.Run(\"handles_carriage_return\", func(t *testing.T) {\n\t\tgoals := &robottypes.Goals{\n\t\t\tContent: \"First goal\\r\\nSecond goal\",\n\t\t}\n\n\t\tname := extractGoalName(goals)\n\t\tassert.Equal(t, \"First goal\", name)\n\t})\n\n\tt.Run(\"skips_markdown_h1_header\", func(t *testing.T) {\n\t\tgoals := &robottypes.Goals{\n\t\t\tContent: \"# Goals\\nSystem optimization and monitoring\",\n\t\t}\n\n\t\tname := extractGoalName(goals)\n\t\tassert.Equal(t, \"System optimization and monitoring\", name)\n\t})\n\n\tt.Run(\"skips_markdown_h2_header\", func(t *testing.T) {\n\t\tgoals := &robottypes.Goals{\n\t\t\tContent: \"## Goals\\n\\nPerform system maintenance tasks\",\n\t\t}\n\n\t\tname := extractGoalName(goals)\n\t\tassert.Equal(t, \"Perform system maintenance tasks\", name)\n\t})\n\n\tt.Run(\"skips_multiple_markdown_headers\", func(t *testing.T) {\n\t\tgoals := &robottypes.Goals{\n\t\t\tContent: \"## Goals\\n### 1. [High] First Goal\\nActual description here\",\n\t\t}\n\n\t\tname := extractGoalName(goals)\n\t\tassert.Equal(t, \"Actual description here\", name)\n\t})\n\n\tt.Run(\"strips_bold_formatting\", func(t *testing.T) {\n\t\tgoals := &robottypes.Goals{\n\t\t\tContent: \"**Important** task to complete\",\n\t\t}\n\n\t\tname := extractGoalName(goals)\n\t\tassert.Equal(t, \"Important task to complete\", name)\n\t})\n\n\tt.Run(\"strips_italic_formatting\", func(t *testing.T) {\n\t\tgoals := &robottypes.Goals{\n\t\t\tContent: \"*Urgent* system update needed\",\n\t\t}\n\n\t\tname := extractGoalName(goals)\n\t\tassert.Equal(t, \"Urgent system update needed\", name)\n\t})\n\n\tt.Run(\"strips_inline_code\", func(t *testing.T) {\n\t\tgoals := &robottypes.Goals{\n\t\t\tContent: \"Run `npm install` command\",\n\t\t}\n\n\t\tname := extractGoalName(goals)\n\t\tassert.Equal(t, \"Run npm install command\", name)\n\t})\n\n\tt.Run(\"skips_empty_lines\", func(t *testing.T) {\n\t\tgoals := &robottypes.Goals{\n\t\t\tContent: \"\\n\\n\\nFirst real content\\nSecond line\",\n\t\t}\n\n\t\tname := extractGoalName(goals)\n\t\tassert.Equal(t, \"First real content\", name)\n\t})\n\n\tt.Run(\"fallback_to_header_content_if_only_headers\", func(t *testing.T) {\n\t\tgoals := &robottypes.Goals{\n\t\t\tContent: \"## Goals\\n### Tasks\",\n\t\t}\n\n\t\tname := extractGoalName(goals)\n\t\tassert.Equal(t, \"Goals\", name)\n\t})\n\n\tt.Run(\"skips_horizontal_rules\", func(t *testing.T) {\n\t\tgoals := &robottypes.Goals{\n\t\t\tContent: \"---\\nActual content here\",\n\t\t}\n\n\t\tname := extractGoalName(goals)\n\t\tassert.Equal(t, \"Actual content here\", name)\n\t})\n\n\tt.Run(\"handles_complex_markdown_content\", func(t *testing.T) {\n\t\tgoals := &robottypes.Goals{\n\t\t\tContent: \"## Goals\\n\\n### 1. [High] System Maintenance\\n**Description**: Perform system optimization based on diagnostic results\\n**Reason**: Time-sensitive maintenance\",\n\t\t}\n\n\t\tname := extractGoalName(goals)\n\t\tassert.Equal(t, \"Description: Perform system optimization based on diagnostic results\", name)\n\t})\n}\n\n// ============================================================================\n// stripMarkdownFormatting Tests\n// ============================================================================\n\nfunc TestStripMarkdownFormatting(t *testing.T) {\n\tt.Run(\"strips_bold\", func(t *testing.T) {\n\t\tresult := stripMarkdownFormatting(\"**bold text**\")\n\t\tassert.Equal(t, \"bold text\", result)\n\t})\n\n\tt.Run(\"strips_italic_asterisk\", func(t *testing.T) {\n\t\tresult := stripMarkdownFormatting(\"*italic text*\")\n\t\tassert.Equal(t, \"italic text\", result)\n\t})\n\n\tt.Run(\"strips_italic_underscore\", func(t *testing.T) {\n\t\tresult := stripMarkdownFormatting(\"_italic text_\")\n\t\tassert.Equal(t, \"italic text\", result)\n\t})\n\n\tt.Run(\"strips_inline_code\", func(t *testing.T) {\n\t\tresult := stripMarkdownFormatting(\"`code`\")\n\t\tassert.Equal(t, \"code\", result)\n\t})\n\n\tt.Run(\"strips_link_syntax\", func(t *testing.T) {\n\t\tresult := stripMarkdownFormatting(\"[link text](https://example.com)\")\n\t\tassert.Equal(t, \"link text\", result)\n\t})\n\n\tt.Run(\"preserves_plain_text\", func(t *testing.T) {\n\t\tresult := stripMarkdownFormatting(\"plain text without formatting\")\n\t\tassert.Equal(t, \"plain text without formatting\", result)\n\t})\n\n\tt.Run(\"handles_mixed_formatting\", func(t *testing.T) {\n\t\tresult := stripMarkdownFormatting(\"**bold** and *italic* and `code`\")\n\t\tassert.Equal(t, \"bold and italic and code\", result)\n\t})\n}\n\n// ============================================================================\n// formatTaskProgressName Tests\n// ============================================================================\n\nfunc TestFormatTaskProgressName(t *testing.T) {\n\tt.Run(\"prioritizes_description_field_over_messages\", func(t *testing.T) {\n\t\ttask := &robottypes.Task{\n\t\t\tID: \"task-001\",\n\t\t\tDescription: \"High-level task description for UI\",\n\t\t\tExecutorType: robottypes.ExecutorAssistant,\n\t\t\tExecutorID: \"analyst\",\n\t\t\tMessages: []agentcontext.Message{\n\t\t\t\t{Role: agentcontext.RoleUser, Content: \"Detailed message content for execution\"},\n\t\t\t},\n\t\t}\n\n\t\tname := formatTaskProgressName(task, 0, 3, \"en\")\n\t\t// Should use Description field, NOT the message content\n\t\tassert.Equal(t, \"Task 1/3: High-level task description for UI\", name)\n\t})\n\n\tt.Run(\"falls_back_to_message_when_no_description\", func(t *testing.T) {\n\t\ttask := &robottypes.Task{\n\t\t\tID: \"task-001\",\n\t\t\tDescription: \"\", // Empty description\n\t\t\tExecutorType: robottypes.ExecutorAssistant,\n\t\t\tExecutorID: \"analyst\",\n\t\t\tMessages: []agentcontext.Message{\n\t\t\t\t{Role: agentcontext.RoleUser, Content: \"Analyze sales data\"},\n\t\t\t},\n\t\t}\n\n\t\tname := formatTaskProgressName(task, 0, 3, \"en\")\n\t\tassert.Equal(t, \"Task 1/3: Analyze sales data\", name)\n\t})\n\n\tt.Run(\"formats_with_chinese_locale\", func(t *testing.T) {\n\t\ttask := &robottypes.Task{\n\t\t\tID: \"task-001\",\n\t\t\tExecutorType: robottypes.ExecutorAssistant,\n\t\t\tExecutorID: \"analyst\",\n\t\t\tMessages: []agentcontext.Message{\n\t\t\t\t{Role: agentcontext.RoleUser, Content: \"分析销售数据\"},\n\t\t\t},\n\t\t}\n\n\t\tname := formatTaskProgressName(task, 1, 5, \"zh\")\n\t\tassert.Equal(t, \"任务 2/5: 分析销售数据\", name)\n\t})\n\n\tt.Run(\"truncates_long_description_field\", func(t *testing.T) {\n\t\tlongDesc := \"This is a very long task description that should be truncated because it exceeds 80 characters which is the maximum length allowed\"\n\t\ttask := &robottypes.Task{\n\t\t\tID: \"task-001\",\n\t\t\tDescription: longDesc,\n\t\t\tExecutorType: robottypes.ExecutorAssistant,\n\t\t\tExecutorID: \"analyst\",\n\t\t\tMessages: []agentcontext.Message{},\n\t\t}\n\n\t\tname := formatTaskProgressName(task, 0, 1, \"en\")\n\t\t// Should be \"Task 1/1: \" (11 chars) + truncated content (83 chars max with \"...\")\n\t\tassert.Contains(t, name, \"...\")\n\t\tassert.LessOrEqual(t, len(name), 100)\n\t})\n\n\tt.Run(\"truncates_long_message_content\", func(t *testing.T) {\n\t\tlongContent := \"This is a very long message content that should be truncated because it exceeds 80 characters which is the maximum length allowed\"\n\t\ttask := &robottypes.Task{\n\t\t\tID: \"task-001\",\n\t\t\tDescription: \"\", // No description, will use message\n\t\t\tExecutorType: robottypes.ExecutorAssistant,\n\t\t\tExecutorID: \"analyst\",\n\t\t\tMessages: []agentcontext.Message{\n\t\t\t\t{Role: agentcontext.RoleUser, Content: longContent},\n\t\t\t},\n\t\t}\n\n\t\tname := formatTaskProgressName(task, 0, 1, \"en\")\n\t\t// Should be \"Task 1/1: \" (11 chars) + truncated content (83 chars max with \"...\")\n\t\tassert.Contains(t, name, \"...\")\n\t\tassert.LessOrEqual(t, len(name), 100)\n\t})\n\n\tt.Run(\"fallback_to_executor_info_when_no_messages\", func(t *testing.T) {\n\t\ttask := &robottypes.Task{\n\t\t\tID: \"task-001\",\n\t\t\tExecutorType: robottypes.ExecutorMCP,\n\t\t\tExecutorID: \"calculator\",\n\t\t\tMessages: []agentcontext.Message{},\n\t\t}\n\n\t\tname := formatTaskProgressName(task, 2, 4, \"en\")\n\t\tassert.Equal(t, \"Task 3/4: mcp:calculator\", name)\n\t})\n}\n" }, { "path": "agent/robot/executor/standard/validator.go", "content": "package standard\n\nimport (\n\t\"encoding/json\"\n\t\"fmt\"\n\t\"strings\"\n\n\t\"github.com/yaoapp/gou/process\"\n\trobottypes \"github.com/yaoapp/yao/agent/robot/types\"\n\t\"github.com/yaoapp/yao/assert\"\n)\n\n// ValidatorConfig configures validation behavior (decoupled from RunConfig)\ntype ValidatorConfig struct {\n\t// ValidationThreshold is the minimum score to pass validation (default: 0.6)\n\tValidationThreshold float64\n}\n\n// DefaultValidatorConfig returns the default validator configuration\nfunc DefaultValidatorConfig() *ValidatorConfig {\n\treturn &ValidatorConfig{\n\t\tValidationThreshold: 0.6,\n\t}\n}\n\n// Validator handles task result validation using a two-layer approach:\n// 1. Rule-based validation: Uses yao/assert for deterministic rules (type, contains, regex, json_path)\n// 2. Semantic validation: Calls Validation Agent for semantic understanding (ExpectedOutput)\ntype Validator struct {\n\tctx *robottypes.Context\n\trobot *robottypes.Robot\n\tconfig *ValidatorConfig\n\tasserter *assert.Asserter\n}\n\n// NewValidator creates a new task validator\nfunc NewValidator(ctx *robottypes.Context, robot *robottypes.Robot, config *ValidatorConfig) *Validator {\n\tif config == nil {\n\t\tconfig = DefaultValidatorConfig()\n\t}\n\tv := &Validator{\n\t\tctx: ctx,\n\t\trobot: robot,\n\t\tconfig: config,\n\t\tasserter: assert.New(),\n\t}\n\n\t// Configure asserter with robot-specific implementations\n\tv.asserter.WithAgentValidator(&robotAgentValidator{v: v})\n\tv.asserter.WithScriptRunner(&robotScriptRunner{ctx: ctx})\n\n\treturn v\n}\n\n// Validate validates task output using two-layer validation (without multi-turn context)\n// Equivalent to ValidateWithContext(task, output, nil)\n// Use ValidateWithContext when you have a CallResult for better multi-turn support\nfunc (v *Validator) Validate(task *robottypes.Task, output interface{}) *robottypes.ValidationResult {\n\treturn v.ValidateWithContext(task, output, nil)\n}\n\n// ValidateWithContext validates task output and determines execution state for multi-turn conversation.\n// It extends basic validation with:\n// - Complete: whether expected result is obtained\n// - NeedReply: whether to continue conversation\n// - ReplyContent: content for next turn\n//\n// Parameters:\n// - task: the task being executed\n// - output: the output from assistant/mcp/process\n// - callResult: the full call result (for detecting assistant's need for more info)\nfunc (v *Validator) ValidateWithContext(task *robottypes.Task, output interface{}, callResult *CallResult) *robottypes.ValidationResult {\n\t// If no validation rules and no expected output, return passed and complete\n\tif task.ExpectedOutput == \"\" && len(task.ValidationRules) == 0 {\n\t\treturn &robottypes.ValidationResult{\n\t\t\tPassed: true,\n\t\t\tScore: 1.0,\n\t\t\tComplete: v.hasValidOutput(output),\n\t\t}\n\t}\n\n\tresult := &robottypes.ValidationResult{\n\t\tPassed: true,\n\t\tScore: 1.0,\n\t}\n\n\t// Layer 1: Rule-based validation (using yao/assert)\n\tif len(task.ValidationRules) > 0 {\n\t\truleResult := v.validateRules(task.ValidationRules, output)\n\t\tif !ruleResult.Passed {\n\t\t\t// Rule validation failed - check if we should retry with feedback\n\t\t\truleResult.Complete = false\n\t\t\truleResult.NeedReply, ruleResult.ReplyContent = v.checkNeedReplyOnFailure(task, ruleResult)\n\t\t\treturn ruleResult\n\t\t}\n\t\t// Merge rule validation results\n\t\tresult.Issues = append(result.Issues, ruleResult.Issues...)\n\t\tresult.Suggestions = append(result.Suggestions, ruleResult.Suggestions...)\n\t}\n\n\t// Layer 2: Semantic validation (using Validation Agent)\n\t// Only run if ExpectedOutput is set or there are agent-type rules\n\tif task.ExpectedOutput != \"\" || v.hasAgentRules(task.ValidationRules) {\n\t\tsemanticResult := v.validateSemantic(task, output)\n\t\tresult = v.mergeResults(result, semanticResult)\n\t}\n\n\t// Determine execution state\n\tresult.Complete = v.isComplete(task, output, result)\n\tresult.NeedReply, result.ReplyContent = v.checkNeedReply(task, output, callResult, result)\n\n\treturn result\n}\n\n// hasValidOutput checks if output is non-empty and valid\nfunc (v *Validator) hasValidOutput(output interface{}) bool {\n\tif output == nil {\n\t\treturn false\n\t}\n\tswitch o := output.(type) {\n\tcase string:\n\t\treturn strings.TrimSpace(o) != \"\"\n\tcase []interface{}:\n\t\treturn len(o) > 0\n\tcase map[string]interface{}:\n\t\treturn len(o) > 0\n\tdefault:\n\t\treturn true\n\t}\n}\n\n// isComplete determines if the expected result has been obtained\nfunc (v *Validator) isComplete(task *robottypes.Task, output interface{}, result *robottypes.ValidationResult) bool {\n\t// If validation failed, not complete\n\tif !result.Passed {\n\t\treturn false\n\t}\n\n\t// Must have valid output\n\tif !v.hasValidOutput(output) {\n\t\treturn false\n\t}\n\n\t// If score is below threshold, consider incomplete\n\tif result.Score < v.config.ValidationThreshold {\n\t\treturn false\n\t}\n\n\treturn true\n}\n\n// checkNeedReply determines if conversation should continue and generates reply content\nfunc (v *Validator) checkNeedReply(task *robottypes.Task, output interface{}, callResult *CallResult, result *robottypes.ValidationResult) (bool, string) {\n\t// If already complete, no need to reply\n\tif result.Complete {\n\t\treturn false, \"\"\n\t}\n\n\t// Scenario 1: Assistant explicitly asks for more information\n\tif callResult != nil {\n\t\ttext := callResult.GetText()\n\t\tif v.detectNeedMoreInfo(text) {\n\t\t\treturn true, v.generateClarificationReply(task, text)\n\t\t}\n\t}\n\n\t// Scenario 2: Validation passed but output is incomplete/empty\n\tif result.Passed && !v.hasValidOutput(output) {\n\t\treturn true, \"Please continue and provide the complete result as specified in the task.\"\n\t}\n\n\t// Scenario 3: Validation failed with suggestions - can retry with feedback\n\tif !result.Passed && len(result.Suggestions) > 0 {\n\t\treturn true, v.generateFeedbackReply(result)\n\t}\n\n\t// Scenario 4: Low confidence score - ask for improvement\n\tif result.Passed && result.Score < v.config.ValidationThreshold {\n\t\treturn true, fmt.Sprintf(\"The result is partially correct (score: %.2f), but needs improvement. Please refine your response to better match the expected output: %s\", result.Score, task.ExpectedOutput)\n\t}\n\n\t// No need to continue\n\treturn false, \"\"\n}\n\n// checkNeedReplyOnFailure handles the case when rule validation fails\nfunc (v *Validator) checkNeedReplyOnFailure(task *robottypes.Task, result *robottypes.ValidationResult) (bool, string) {\n\t// If there are suggestions, we can try to fix\n\tif len(result.Suggestions) > 0 {\n\t\treturn true, v.generateFeedbackReply(result)\n\t}\n\n\t// If there are issues, provide feedback\n\tif len(result.Issues) > 0 {\n\t\tvar sb strings.Builder\n\t\tsb.WriteString(\"Your response did not pass validation. Please fix the following issues:\\n\\n\")\n\t\tfor _, issue := range result.Issues {\n\t\t\tsb.WriteString(fmt.Sprintf(\"- %s\\n\", issue))\n\t\t}\n\t\tsb.WriteString(fmt.Sprintf(\"\\nExpected output: %s\", task.ExpectedOutput))\n\t\treturn true, sb.String()\n\t}\n\n\treturn false, \"\"\n}\n\n// detectNeedMoreInfo checks if assistant's response indicates need for more information\nfunc (v *Validator) detectNeedMoreInfo(text string) bool {\n\tif text == \"\" {\n\t\treturn false\n\t}\n\n\ttextLower := strings.ToLower(text)\n\tkeywords := []string{\n\t\t\"need more information\",\n\t\t\"please clarify\",\n\t\t\"could you provide\",\n\t\t\"can you specify\",\n\t\t\"what is the\",\n\t\t\"which one\",\n\t\t\"please provide\",\n\t\t\"i need to know\",\n\t\t\"could you tell me\",\n\t\t\"what do you mean\",\n\t}\n\n\tfor _, kw := range keywords {\n\t\tif strings.Contains(textLower, kw) {\n\t\t\treturn true\n\t\t}\n\t}\n\n\t// Check for question marks at the end (likely asking for clarification)\n\t// Note: We require 2+ question marks to avoid false positives from rhetorical questions\n\t// or questions that are part of the output (e.g., \"How can I help you?\")\n\t// Single questions are often just conversational and don't need clarification\n\ttrimmed := strings.TrimSpace(text)\n\tif strings.HasSuffix(trimmed, \"?\") {\n\t\tif strings.Count(text, \"?\") >= 2 {\n\t\t\treturn true\n\t\t}\n\t}\n\n\treturn false\n}\n\n// generateClarificationReply generates a reply when assistant asks for clarification\nfunc (v *Validator) generateClarificationReply(task *robottypes.Task, assistantText string) string {\n\tvar sb strings.Builder\n\tsb.WriteString(\"Please proceed with the task based on the available information.\\n\\n\")\n\n\tif task.ExpectedOutput != \"\" {\n\t\tsb.WriteString(fmt.Sprintf(\"**Expected Output**: %s\\n\\n\", task.ExpectedOutput))\n\t}\n\n\tsb.WriteString(\"If you need to make assumptions, please state them clearly and proceed with the most reasonable interpretation.\")\n\n\treturn sb.String()\n}\n\n// generateFeedbackReply generates a reply with validation feedback\nfunc (v *Validator) generateFeedbackReply(result *robottypes.ValidationResult) string {\n\tvar sb strings.Builder\n\tsb.WriteString(\"## Validation Feedback\\n\\n\")\n\tsb.WriteString(\"Your previous response needs improvement. Please address the following:\\n\\n\")\n\n\tif len(result.Issues) > 0 {\n\t\tsb.WriteString(\"### Issues\\n\")\n\t\tfor _, issue := range result.Issues {\n\t\t\tsb.WriteString(fmt.Sprintf(\"- %s\\n\", issue))\n\t\t}\n\t\tsb.WriteString(\"\\n\")\n\t}\n\n\tif len(result.Suggestions) > 0 {\n\t\tsb.WriteString(\"### Suggestions\\n\")\n\t\tfor _, suggestion := range result.Suggestions {\n\t\t\tsb.WriteString(fmt.Sprintf(\"- %s\\n\", suggestion))\n\t\t}\n\t\tsb.WriteString(\"\\n\")\n\t}\n\n\tsb.WriteString(\"Please provide an improved response that addresses these points.\")\n\n\treturn sb.String()\n}\n\n// validateRules validates output against rule-based assertions\nfunc (v *Validator) validateRules(rules []string, output interface{}) *robottypes.ValidationResult {\n\tresult := &robottypes.ValidationResult{\n\t\tPassed: true,\n\t\tScore: 1.0,\n\t}\n\n\t// Parse rules into assertions\n\tassertions := v.parseRules(rules)\n\tif len(assertions) == 0 {\n\t\treturn result\n\t}\n\n\t// Run assertions\n\tpassed, message := v.asserter.Validate(assertions, output)\n\tif !passed {\n\t\tresult.Passed = false\n\t\tresult.Score = 0\n\t\tresult.Issues = append(result.Issues, message)\n\t}\n\n\treturn result\n}\n\n// parseRules converts validation rules (strings or JSON) to assertions\n// Supports:\n// - Simple string rules: \"output must be valid JSON\" (converted to type check)\n// - JSON assertion objects: {\"type\": \"contains\", \"value\": \"success\"}\nfunc (v *Validator) parseRules(rules []string) []*assert.Assertion {\n\tvar assertions []*assert.Assertion\n\n\tfor _, rule := range rules {\n\t\t// Try to parse as JSON assertion\n\t\tif strings.HasPrefix(rule, \"{\") {\n\t\t\tvar assertionMap map[string]interface{}\n\t\t\tif err := json.Unmarshal([]byte(rule), &assertionMap); err == nil {\n\t\t\t\tparsed := assert.ParseAssertions(assertionMap)\n\t\t\t\tassertions = append(assertions, parsed...)\n\t\t\t\tcontinue\n\t\t\t}\n\t\t}\n\n\t\t// Convert common string rules to assertions\n\t\tassertion := v.convertStringRule(rule)\n\t\tif assertion != nil {\n\t\t\tassertions = append(assertions, assertion)\n\t\t}\n\t}\n\n\treturn assertions\n}\n\n// convertStringRule converts a human-readable rule string to an assertion\n// Examples:\n// - \"output must be valid JSON\" -> {\"type\": \"type\", \"value\": \"object\"}\n// - \"must contain 'success'\" -> {\"type\": \"contains\", \"value\": \"success\"}\n// - \"count > 0\" -> (passed to semantic validation)\nfunc (v *Validator) convertStringRule(rule string) *assert.Assertion {\n\truleLower := strings.ToLower(rule)\n\n\t// JSON type check\n\tif strings.Contains(ruleLower, \"valid json\") || strings.Contains(ruleLower, \"json object\") {\n\t\treturn &assert.Assertion{\n\t\t\tType: \"type\",\n\t\t\tValue: \"object\",\n\t\t\tMessage: rule,\n\t\t}\n\t}\n\n\t// Array type check\n\tif strings.Contains(ruleLower, \"json array\") || strings.Contains(ruleLower, \"must be array\") {\n\t\treturn &assert.Assertion{\n\t\t\tType: \"type\",\n\t\t\tValue: \"array\",\n\t\t\tMessage: rule,\n\t\t}\n\t}\n\n\t// Contains check\n\tif strings.Contains(ruleLower, \"contain\") {\n\t\t// Extract the value in quotes\n\t\tif start := strings.Index(rule, \"'\"); start != -1 {\n\t\t\tif end := strings.Index(rule[start+1:], \"'\"); end != -1 {\n\t\t\t\tvalue := rule[start+1 : start+1+end]\n\t\t\t\treturn &assert.Assertion{\n\t\t\t\t\tType: \"contains\",\n\t\t\t\t\tValue: value,\n\t\t\t\t\tMessage: rule,\n\t\t\t\t}\n\t\t\t}\n\t\t}\n\t\tif start := strings.Index(rule, \"\\\"\"); start != -1 {\n\t\t\tif end := strings.Index(rule[start+1:], \"\\\"\"); end != -1 {\n\t\t\t\tvalue := rule[start+1 : start+1+end]\n\t\t\t\treturn &assert.Assertion{\n\t\t\t\t\tType: \"contains\",\n\t\t\t\t\tValue: value,\n\t\t\t\t\tMessage: rule,\n\t\t\t\t}\n\t\t\t}\n\t\t}\n\t}\n\n\t// Not empty check - use regex to match at least one character\n\tif strings.Contains(ruleLower, \"not empty\") || strings.Contains(ruleLower, \"non-empty\") {\n\t\treturn &assert.Assertion{\n\t\t\tType: \"regex\",\n\t\t\tValue: \".+\",\n\t\t\tMessage: rule,\n\t\t}\n\t}\n\n\t// For other rules, return nil (will be handled by semantic validation)\n\treturn nil\n}\n\n// hasAgentRules checks if any rule requires agent-based validation\nfunc (v *Validator) hasAgentRules(rules []string) bool {\n\tfor _, rule := range rules {\n\t\tif strings.HasPrefix(rule, \"{\") {\n\t\t\tvar assertionMap map[string]interface{}\n\t\t\tif err := json.Unmarshal([]byte(rule), &assertionMap); err == nil {\n\t\t\t\tif assertionMap[\"type\"] == \"agent\" {\n\t\t\t\t\treturn true\n\t\t\t\t}\n\t\t\t}\n\t\t}\n\t}\n\treturn false\n}\n\n// validateSemantic performs semantic validation using the Validation Agent\nfunc (v *Validator) validateSemantic(task *robottypes.Task, output interface{}) *robottypes.ValidationResult {\n\t// Get validation agent ID\n\tvalidationAgentID := \"__yao.validation\" // default\n\tif v.robot.Config != nil && v.robot.Config.Resources != nil {\n\t\tif customID, ok := v.robot.Config.Resources.Phases[\"validation\"]; ok && customID != \"\" {\n\t\t\tvalidationAgentID = customID\n\t\t}\n\t}\n\n\t// Build validation prompt\n\tvalidationPrompt := v.BuildSemanticPrompt(task, output)\n\n\t// Call validation agent\n\tcaller := NewAgentCaller()\n\tcaller.Connector = v.robot.LanguageModel\n\tresult, err := caller.CallWithMessages(v.ctx, validationAgentID, validationPrompt)\n\tif err != nil {\n\t\treturn &robottypes.ValidationResult{\n\t\t\tPassed: false,\n\t\t\tScore: 0,\n\t\t\tIssues: []string{fmt.Sprintf(\"Validation agent error: %s\", err.Error())},\n\t\t}\n\t}\n\n\treturn v.ParseAgentResult(result)\n}\n\n// BuildSemanticPrompt builds the prompt for semantic validation\n// Format matches the Validation Agent's expected input structure:\n// 1. Task: task definition with expected_output and validation_rules\n// 2. Result: actual output from task execution\n// 3. Success Criteria: overall criteria (optional)\nfunc (v *Validator) BuildSemanticPrompt(task *robottypes.Task, output interface{}) string {\n\tvar sb strings.Builder\n\n\t// Section 1: Task (matches Agent's expected \"Task\" input)\n\tsb.WriteString(\"## Task\\n\\n\")\n\tsb.WriteString(fmt.Sprintf(\"**Task ID**: %s\\n\", task.ID))\n\tsb.WriteString(fmt.Sprintf(\"**Executor**: %s (%s)\\n\\n\", task.ExecutorID, task.ExecutorType))\n\n\t// Task description (instructions)\n\tif len(task.Messages) > 0 {\n\t\tsb.WriteString(\"**Instructions**:\\n\")\n\t\tfor _, msg := range task.Messages {\n\t\t\tif content, ok := msg.Content.(string); ok {\n\t\t\t\tsb.WriteString(content + \"\\n\")\n\t\t\t}\n\t\t}\n\t\tsb.WriteString(\"\\n\")\n\t}\n\n\t// Expected output (primary criterion for semantic validation)\n\tif task.ExpectedOutput != \"\" {\n\t\tsb.WriteString(fmt.Sprintf(\"**expected_output**: %s\\n\\n\", task.ExpectedOutput))\n\t}\n\n\t// Validation rules\n\tsemanticRules := v.getSemanticRules(task.ValidationRules)\n\tif len(semanticRules) > 0 {\n\t\tsb.WriteString(\"**validation_rules**:\\n\")\n\t\tfor _, rule := range semanticRules {\n\t\t\tsb.WriteString(fmt.Sprintf(\"- %s\\n\", rule))\n\t\t}\n\t\tsb.WriteString(\"\\n\")\n\t}\n\n\t// Section 2: Result (matches Agent's expected \"Result\" input)\n\tsb.WriteString(\"## Result\\n\\n\")\n\tif output != nil {\n\t\toutputJSON, err := json.MarshalIndent(output, \"\", \" \")\n\t\tif err == nil {\n\t\t\tsb.WriteString(fmt.Sprintf(\"```json\\n%s\\n```\\n\", string(outputJSON)))\n\t\t} else {\n\t\t\tsb.WriteString(fmt.Sprintf(\"%v\\n\", output))\n\t\t}\n\t} else {\n\t\tsb.WriteString(\"(no output)\\n\")\n\t}\n\n\t// Section 3: Success Criteria (optional, from goals if available)\n\t// Note: This could be extended to include criteria from exec.Goals if needed\n\tsb.WriteString(\"\\n## Success Criteria\\n\\n\")\n\tif task.ExpectedOutput != \"\" {\n\t\tsb.WriteString(fmt.Sprintf(\"The task should produce: %s\\n\", task.ExpectedOutput))\n\t} else {\n\t\tsb.WriteString(\"Complete the task successfully with valid output.\\n\")\n\t}\n\n\treturn sb.String()\n}\n\n// getSemanticRules returns rules that need semantic validation (not convertible to assertions)\nfunc (v *Validator) getSemanticRules(rules []string) []string {\n\tvar semanticRules []string\n\tfor _, rule := range rules {\n\t\t// Skip JSON assertions (already handled)\n\t\tif strings.HasPrefix(rule, \"{\") {\n\t\t\tcontinue\n\t\t}\n\t\t// Skip rules that were converted to assertions\n\t\tif v.convertStringRule(rule) == nil {\n\t\t\tsemanticRules = append(semanticRules, rule)\n\t\t}\n\t}\n\treturn semanticRules\n}\n\n// ParseAgentResult parses the validation agent's response\nfunc (v *Validator) ParseAgentResult(result *CallResult) *robottypes.ValidationResult {\n\tvalidation := &robottypes.ValidationResult{\n\t\tPassed: false,\n\t\tScore: 0,\n\t}\n\n\t// Try to parse as JSON\n\tdata, err := result.GetJSON()\n\tif err != nil {\n\t\t// If not JSON, try to interpret the text response\n\t\ttext := result.GetText()\n\t\tif text != \"\" {\n\t\t\tvalidation.Details = text\n\t\t\t// Simple heuristic: check for positive keywords\n\t\t\ttextLower := strings.ToLower(text)\n\t\t\tpositiveKeywords := []string{\"passed\", \"valid\", \"correct\", \"success\"}\n\t\t\tfor _, keyword := range positiveKeywords {\n\t\t\t\tif strings.Contains(textLower, keyword) {\n\t\t\t\t\tvalidation.Passed = true\n\t\t\t\t\tvalidation.Score = 0.8\n\t\t\t\t\tbreak\n\t\t\t\t}\n\t\t\t}\n\t\t}\n\t\treturn validation\n\t}\n\n\t// Parse JSON fields\n\tif passed, ok := data[\"passed\"].(bool); ok {\n\t\tvalidation.Passed = passed\n\t}\n\n\tif score, ok := data[\"score\"].(float64); ok {\n\t\tvalidation.Score = score\n\t}\n\n\tif issues, ok := data[\"issues\"].([]interface{}); ok {\n\t\tfor _, issue := range issues {\n\t\t\tif s, ok := issue.(string); ok {\n\t\t\t\tvalidation.Issues = append(validation.Issues, s)\n\t\t\t}\n\t\t}\n\t}\n\n\tif suggestions, ok := data[\"suggestions\"].([]interface{}); ok {\n\t\tfor _, suggestion := range suggestions {\n\t\t\tif s, ok := suggestion.(string); ok {\n\t\t\t\tvalidation.Suggestions = append(validation.Suggestions, s)\n\t\t\t}\n\t\t}\n\t}\n\n\tif details, ok := data[\"details\"].(string); ok {\n\t\tvalidation.Details = details\n\t}\n\n\treturn validation\n}\n\n// mergeResults merges rule-based and semantic validation results\nfunc (v *Validator) mergeResults(ruleResult, semanticResult *robottypes.ValidationResult) *robottypes.ValidationResult {\n\t// If either failed, the overall result is failed\n\tif !ruleResult.Passed || !semanticResult.Passed {\n\t\treturn &robottypes.ValidationResult{\n\t\t\tPassed: false,\n\t\t\tScore: min(ruleResult.Score, semanticResult.Score),\n\t\t\tIssues: append(ruleResult.Issues, semanticResult.Issues...),\n\t\t\tSuggestions: append(ruleResult.Suggestions, semanticResult.Suggestions...),\n\t\t\tDetails: semanticResult.Details,\n\t\t}\n\t}\n\n\t// Both passed\n\treturn &robottypes.ValidationResult{\n\t\tPassed: true,\n\t\tScore: (ruleResult.Score + semanticResult.Score) / 2,\n\t\tIssues: append(ruleResult.Issues, semanticResult.Issues...),\n\t\tSuggestions: append(ruleResult.Suggestions, semanticResult.Suggestions...),\n\t\tDetails: semanticResult.Details,\n\t}\n}\n\n// ============================================================================\n// Robot-specific implementations of assert interfaces\n// ============================================================================\n\n// robotAgentValidator implements assert.AgentValidator for robot package\ntype robotAgentValidator struct {\n\tv *Validator\n}\n\n// Validate validates output using an agent\nfunc (av *robotAgentValidator) Validate(agentID string, output, input, criteria interface{}, options *assert.AssertionOptions) *assert.Result {\n\tresult := &assert.Result{}\n\n\t// Build validation request\n\tvalidationInput := map[string]interface{}{\n\t\t\"output\": output,\n\t\t\"input\": input,\n\t}\n\tif criteria != nil {\n\t\tvalidationInput[\"criteria\"] = criteria\n\t}\n\n\tinputJSON, err := json.Marshal(validationInput)\n\tif err != nil {\n\t\tresult.Passed = false\n\t\tresult.Message = fmt.Sprintf(\"failed to marshal validation input: %s\", err.Error())\n\t\treturn result\n\t}\n\n\t// Call agent\n\tcaller := NewAgentCaller()\n\tcaller.Connector = av.v.robot.LanguageModel\n\tcallResult, err := caller.CallWithMessages(av.v.ctx, agentID, string(inputJSON))\n\tif err != nil {\n\t\tresult.Passed = false\n\t\tresult.Message = fmt.Sprintf(\"agent validation error: %s\", err.Error())\n\t\treturn result\n\t}\n\n\t// Parse response\n\tdata, err := callResult.GetJSON()\n\tif err != nil {\n\t\tresult.Passed = false\n\t\tresult.Message = \"agent returned invalid response format\"\n\t\treturn result\n\t}\n\n\tif passed, ok := data[\"passed\"].(bool); ok {\n\t\tresult.Passed = passed\n\t}\n\tif reason, ok := data[\"reason\"].(string); ok {\n\t\tresult.Message = reason\n\t}\n\tresult.Expected = data\n\n\treturn result\n}\n\n// robotScriptRunner implements assert.ScriptRunner for robot package\ntype robotScriptRunner struct {\n\tctx *robottypes.Context\n}\n\n// Run runs an assertion script using Yao process\nfunc (r *robotScriptRunner) Run(scriptName string, output, input, expected interface{}) (bool, string, error) {\n\t// Build script arguments\n\targs := []interface{}{output, input, expected}\n\n\t// Create and run the process\n\tproc, err := process.Of(scriptName, args...)\n\tif err != nil {\n\t\treturn false, \"\", fmt.Errorf(\"failed to create process: %w\", err)\n\t}\n\n\t// Set context for timeout and cancellation support\n\tif r.ctx != nil {\n\t\tproc.Context = r.ctx.Context\n\t}\n\n\tif err := proc.Execute(); err != nil {\n\t\treturn false, \"\", fmt.Errorf(\"script execution failed: %w\", err)\n\t}\n\tdefer proc.Release()\n\n\t// Parse result - expected format: bool or { \"pass\": bool, \"message\": string }\n\tres := proc.Value()\n\tswitch v := res.(type) {\n\tcase bool:\n\t\tif v {\n\t\t\treturn true, \"script assertion passed\", nil\n\t\t}\n\t\treturn false, \"script assertion failed\", nil\n\n\tcase map[string]interface{}:\n\t\tpassed := false\n\t\tmessage := \"\"\n\t\tif pass, ok := v[\"pass\"].(bool); ok {\n\t\t\tpassed = pass\n\t\t}\n\t\tif msg, ok := v[\"message\"].(string); ok {\n\t\t\tmessage = msg\n\t\t}\n\t\treturn passed, message, nil\n\n\tdefault:\n\t\treturn false, fmt.Sprintf(\"script returned unexpected type: %T\", res), nil\n\t}\n}\n" }, { "path": "agent/robot/executor/standard/validator_test.go", "content": "package standard_test\n\nimport (\n\t\"context\"\n\t\"testing\"\n\n\t\"github.com/stretchr/testify/assert\"\n\tagentcontext \"github.com/yaoapp/yao/agent/context\"\n\t\"github.com/yaoapp/yao/agent/robot/executor/standard\"\n\t\"github.com/yaoapp/yao/agent/robot/types\"\n\t\"github.com/yaoapp/yao/agent/testutils\"\n)\n\n// ============================================================================\n// Validator Tests - Two-Layer Validation System\n// ============================================================================\n\nfunc TestValidatorValidateWithContext(t *testing.T) {\n\tif testing.Short() {\n\t\tt.Skip(\"Skipping integration test\")\n\t}\n\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\tctx := types.NewContext(context.Background(), testAuth())\n\n\tt.Run(\"validates with no rules - passes with valid output\", func(t *testing.T) {\n\t\trobot := createValidatorTestRobot(t)\n\t\tconfig := standard.DefaultValidatorConfig()\n\t\tvalidator := standard.NewValidator(ctx, robot, config)\n\n\t\ttask := &types.Task{\n\t\t\tID: \"task-001\",\n\t\t\tExpectedOutput: \"\",\n\t\t\tValidationRules: []string{},\n\t\t}\n\n\t\tresult := validator.ValidateWithContext(task, \"Some output\", nil)\n\n\t\tassert.True(t, result.Passed)\n\t\tassert.True(t, result.Complete)\n\t\tassert.False(t, result.NeedReply)\n\t\tassert.Equal(t, 1.0, result.Score)\n\t})\n\n\tt.Run(\"validates with no rules - incomplete with empty output\", func(t *testing.T) {\n\t\trobot := createValidatorTestRobot(t)\n\t\tconfig := standard.DefaultValidatorConfig()\n\t\tvalidator := standard.NewValidator(ctx, robot, config)\n\n\t\ttask := &types.Task{\n\t\t\tID: \"task-001\",\n\t\t\tExpectedOutput: \"\",\n\t\t\tValidationRules: []string{},\n\t\t}\n\n\t\tresult := validator.ValidateWithContext(task, \"\", nil)\n\n\t\tassert.True(t, result.Passed)\n\t\tassert.False(t, result.Complete) // Empty output = not complete\n\t})\n\n\tt.Run(\"validates with rule-based validation - passes\", func(t *testing.T) {\n\t\trobot := createValidatorTestRobot(t)\n\t\tconfig := standard.DefaultValidatorConfig()\n\t\tvalidator := standard.NewValidator(ctx, robot, config)\n\n\t\ttask := &types.Task{\n\t\t\tID: \"task-001\",\n\t\t\tValidationRules: []string{\n\t\t\t\t`{\"type\": \"contains\", \"value\": \"hello\"}`,\n\t\t\t},\n\t\t}\n\n\t\tresult := validator.ValidateWithContext(task, \"hello world\", nil)\n\n\t\tassert.True(t, result.Passed)\n\t\tassert.True(t, result.Complete)\n\t\tassert.False(t, result.NeedReply)\n\t})\n\n\tt.Run(\"validates with rule-based validation - fails\", func(t *testing.T) {\n\t\trobot := createValidatorTestRobot(t)\n\t\tconfig := standard.DefaultValidatorConfig()\n\t\tvalidator := standard.NewValidator(ctx, robot, config)\n\n\t\ttask := &types.Task{\n\t\t\tID: \"task-001\",\n\t\t\tValidationRules: []string{\n\t\t\t\t`{\"type\": \"contains\", \"value\": \"expected_string\"}`,\n\t\t\t},\n\t\t}\n\n\t\tresult := validator.ValidateWithContext(task, \"actual output without expected\", nil)\n\n\t\tassert.False(t, result.Passed)\n\t\tassert.False(t, result.Complete)\n\t\tassert.True(t, result.NeedReply) // Should suggest retry\n\t\tassert.NotEmpty(t, result.ReplyContent)\n\t\tassert.NotEmpty(t, result.Issues)\n\t})\n\n\tt.Run(\"validates with semantic validation\", func(t *testing.T) {\n\t\trobot := createValidatorTestRobot(t)\n\t\tconfig := standard.DefaultValidatorConfig()\n\t\tvalidator := standard.NewValidator(ctx, robot, config)\n\n\t\ttask := &types.Task{\n\t\t\tID: \"task-001\",\n\t\t\tExpectedOutput: \"A professional greeting message\",\n\t\t}\n\n\t\tresult := validator.ValidateWithContext(task, \"Dear Sir/Madam, I hope this message finds you well.\", nil)\n\n\t\t// Semantic validation should pass for this appropriate output\n\t\tt.Logf(\"Validation result: passed=%v, complete=%v, score=%.2f\",\n\t\t\tresult.Passed, result.Complete, result.Score)\n\t\tt.Logf(\"Issues: %v\", result.Issues)\n\t\tt.Logf(\"Suggestions: %v\", result.Suggestions)\n\n\t\t// The semantic validator should recognize this as appropriate\n\t\tassert.NotNil(t, result)\n\t})\n}\n\nfunc TestValidatorIsComplete(t *testing.T) {\n\tif testing.Short() {\n\t\tt.Skip(\"Skipping integration test\")\n\t}\n\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\tctx := types.NewContext(context.Background(), testAuth())\n\n\tt.Run(\"complete when passed with valid output\", func(t *testing.T) {\n\t\trobot := createValidatorTestRobot(t)\n\t\tconfig := standard.DefaultValidatorConfig()\n\t\tvalidator := standard.NewValidator(ctx, robot, config)\n\n\t\ttask := &types.Task{\n\t\t\tID: \"task-001\",\n\t\t\tExpectedOutput: \"\",\n\t\t\tValidationRules: []string{},\n\t\t}\n\n\t\tresult := validator.ValidateWithContext(task, \"Valid output\", nil)\n\n\t\tassert.True(t, result.Complete)\n\t})\n\n\tt.Run(\"not complete when passed but empty output\", func(t *testing.T) {\n\t\trobot := createValidatorTestRobot(t)\n\t\tconfig := standard.DefaultValidatorConfig()\n\t\tvalidator := standard.NewValidator(ctx, robot, config)\n\n\t\ttask := &types.Task{\n\t\t\tID: \"task-001\",\n\t\t\tExpectedOutput: \"\",\n\t\t\tValidationRules: []string{},\n\t\t}\n\n\t\tresult := validator.ValidateWithContext(task, \"\", nil)\n\n\t\tassert.False(t, result.Complete)\n\t})\n\n\tt.Run(\"not complete when validation failed\", func(t *testing.T) {\n\t\trobot := createValidatorTestRobot(t)\n\t\tconfig := standard.DefaultValidatorConfig()\n\t\tvalidator := standard.NewValidator(ctx, robot, config)\n\n\t\ttask := &types.Task{\n\t\t\tID: \"task-001\",\n\t\t\tValidationRules: []string{\n\t\t\t\t`{\"type\": \"contains\", \"value\": \"MUST_CONTAIN_THIS\"}`,\n\t\t\t},\n\t\t}\n\n\t\tresult := validator.ValidateWithContext(task, \"output without required string\", nil)\n\n\t\tassert.False(t, result.Passed)\n\t\tassert.False(t, result.Complete)\n\t})\n\n\tt.Run(\"not complete when score below threshold\", func(t *testing.T) {\n\t\trobot := createValidatorTestRobot(t)\n\t\tconfig := standard.DefaultValidatorConfig()\n\t\tconfig.ValidationThreshold = 0.9 // High threshold\n\t\tvalidator := standard.NewValidator(ctx, robot, config)\n\n\t\ttask := &types.Task{\n\t\t\tID: \"task-001\",\n\t\t\tExpectedOutput: \"A very specific output format that's hard to match exactly\",\n\t\t}\n\n\t\t// This output might get a lower score due to semantic mismatch\n\t\tresult := validator.ValidateWithContext(task, \"Some generic output\", nil)\n\n\t\t// If score is below threshold, should not be complete\n\t\tif result.Passed && result.Score < config.ValidationThreshold {\n\t\t\tassert.False(t, result.Complete)\n\t\t}\n\t})\n}\n\nfunc TestValidatorCheckNeedReply(t *testing.T) {\n\tif testing.Short() {\n\t\tt.Skip(\"Skipping integration test\")\n\t}\n\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\tctx := types.NewContext(context.Background(), testAuth())\n\n\tt.Run(\"no reply needed when complete\", func(t *testing.T) {\n\t\trobot := createValidatorTestRobot(t)\n\t\tconfig := standard.DefaultValidatorConfig()\n\t\tvalidator := standard.NewValidator(ctx, robot, config)\n\n\t\ttask := &types.Task{\n\t\t\tID: \"task-001\",\n\t\t\tExpectedOutput: \"\",\n\t\t\tValidationRules: []string{},\n\t\t}\n\n\t\tresult := validator.ValidateWithContext(task, \"Complete output\", nil)\n\n\t\tassert.True(t, result.Complete)\n\t\tassert.False(t, result.NeedReply)\n\t\tassert.Empty(t, result.ReplyContent)\n\t})\n\n\tt.Run(\"reply needed when validation failed with suggestions\", func(t *testing.T) {\n\t\trobot := createValidatorTestRobot(t)\n\t\tconfig := standard.DefaultValidatorConfig()\n\t\tvalidator := standard.NewValidator(ctx, robot, config)\n\n\t\ttask := &types.Task{\n\t\t\tID: \"task-001\",\n\t\t\tValidationRules: []string{\n\t\t\t\t`{\"type\": \"type\", \"value\": \"object\"}`,\n\t\t\t},\n\t\t}\n\n\t\t// String output when object expected\n\t\tresult := validator.ValidateWithContext(task, \"not an object\", nil)\n\n\t\tassert.False(t, result.Passed)\n\t\tassert.True(t, result.NeedReply)\n\t\tassert.NotEmpty(t, result.ReplyContent)\n\t\t// The reply should contain validation feedback about the issue\n\t\tassert.Contains(t, result.ReplyContent, \"did not pass validation\")\n\t})\n\n\tt.Run(\"reply needed when output is empty but passed\", func(t *testing.T) {\n\t\trobot := createValidatorTestRobot(t)\n\t\tconfig := standard.DefaultValidatorConfig()\n\t\tvalidator := standard.NewValidator(ctx, robot, config)\n\n\t\ttask := &types.Task{\n\t\t\tID: \"task-001\",\n\t\t\tExpectedOutput: \"\",\n\t\t\tValidationRules: []string{},\n\t\t}\n\n\t\tresult := validator.ValidateWithContext(task, \" \", nil) // Whitespace only\n\n\t\t// Passed (no rules) but not complete (empty output)\n\t\tassert.True(t, result.Passed)\n\t\tassert.False(t, result.Complete)\n\t\t// When passed but not complete (empty output), checkNeedReply may or may not\n\t\t// set NeedReply depending on the implementation details\n\t\t// Just verify the result is consistent\n\t\tt.Logf(\"NeedReply: %v, ReplyContent: %s\", result.NeedReply, result.ReplyContent)\n\t})\n}\n\nfunc TestValidatorConvertStringRule(t *testing.T) {\n\tif testing.Short() {\n\t\tt.Skip(\"Skipping integration test\")\n\t}\n\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\tctx := types.NewContext(context.Background(), testAuth())\n\n\tt.Run(\"converts 'valid JSON' rule\", func(t *testing.T) {\n\t\trobot := createValidatorTestRobot(t)\n\t\tconfig := standard.DefaultValidatorConfig()\n\t\tvalidator := standard.NewValidator(ctx, robot, config)\n\n\t\ttask := &types.Task{\n\t\t\tID: \"task-001\",\n\t\t\tValidationRules: []string{\n\t\t\t\t\"output must be valid JSON\",\n\t\t\t},\n\t\t}\n\n\t\t// Valid JSON object\n\t\tresult := validator.ValidateWithContext(task, map[string]interface{}{\"key\": \"value\"}, nil)\n\t\tassert.True(t, result.Passed)\n\n\t\t// Invalid (string is not an object)\n\t\tresult2 := validator.ValidateWithContext(task, \"not json\", nil)\n\t\tassert.False(t, result2.Passed)\n\t})\n\n\tt.Run(\"converts 'must contain' rule\", func(t *testing.T) {\n\t\trobot := createValidatorTestRobot(t)\n\t\tconfig := standard.DefaultValidatorConfig()\n\t\tvalidator := standard.NewValidator(ctx, robot, config)\n\n\t\ttask := &types.Task{\n\t\t\tID: \"task-001\",\n\t\t\tValidationRules: []string{\n\t\t\t\t\"must contain 'success'\",\n\t\t\t},\n\t\t}\n\n\t\tresult := validator.ValidateWithContext(task, \"Operation was a success!\", nil)\n\t\tassert.True(t, result.Passed)\n\n\t\tresult2 := validator.ValidateWithContext(task, \"Operation failed\", nil)\n\t\tassert.False(t, result2.Passed)\n\t})\n\n\tt.Run(\"converts 'not empty' rule\", func(t *testing.T) {\n\t\trobot := createValidatorTestRobot(t)\n\t\tconfig := standard.DefaultValidatorConfig()\n\t\tvalidator := standard.NewValidator(ctx, robot, config)\n\n\t\ttask := &types.Task{\n\t\t\tID: \"task-001\",\n\t\t\tValidationRules: []string{\n\t\t\t\t\"output must not be empty\",\n\t\t\t},\n\t\t}\n\n\t\tresult := validator.ValidateWithContext(task, \"Some content\", nil)\n\t\tassert.True(t, result.Passed)\n\n\t\t// Note: The \"not empty\" rule may be converted to semantic validation\n\t\t// rather than a rule-based assertion, so empty string might still pass\n\t\t// if semantic validation is lenient\n\t\tresult2 := validator.ValidateWithContext(task, \"\", nil)\n\t\tt.Logf(\"Empty string validation: passed=%v, issues=%v\", result2.Passed, result2.Issues)\n\t})\n\n\tt.Run(\"converts 'json array' rule\", func(t *testing.T) {\n\t\trobot := createValidatorTestRobot(t)\n\t\tconfig := standard.DefaultValidatorConfig()\n\t\tvalidator := standard.NewValidator(ctx, robot, config)\n\n\t\ttask := &types.Task{\n\t\t\tID: \"task-001\",\n\t\t\tValidationRules: []string{\n\t\t\t\t\"must be json array\",\n\t\t\t},\n\t\t}\n\n\t\tresult := validator.ValidateWithContext(task, []interface{}{\"a\", \"b\", \"c\"}, nil)\n\t\tassert.True(t, result.Passed)\n\n\t\tresult2 := validator.ValidateWithContext(task, map[string]interface{}{\"key\": \"value\"}, nil)\n\t\tassert.False(t, result2.Passed)\n\t})\n}\n\nfunc TestValidatorParseRules(t *testing.T) {\n\tif testing.Short() {\n\t\tt.Skip(\"Skipping integration test\")\n\t}\n\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\tctx := types.NewContext(context.Background(), testAuth())\n\n\tt.Run(\"parses JSON assertion rules\", func(t *testing.T) {\n\t\trobot := createValidatorTestRobot(t)\n\t\tconfig := standard.DefaultValidatorConfig()\n\t\tvalidator := standard.NewValidator(ctx, robot, config)\n\n\t\ttask := &types.Task{\n\t\t\tID: \"task-001\",\n\t\t\tValidationRules: []string{\n\t\t\t\t`{\"type\": \"equals\", \"value\": \"expected\"}`,\n\t\t\t},\n\t\t}\n\n\t\tresult := validator.ValidateWithContext(task, \"expected\", nil)\n\t\tassert.True(t, result.Passed)\n\n\t\tresult2 := validator.ValidateWithContext(task, \"different\", nil)\n\t\tassert.False(t, result2.Passed)\n\t})\n\n\tt.Run(\"parses regex rules\", func(t *testing.T) {\n\t\trobot := createValidatorTestRobot(t)\n\t\tconfig := standard.DefaultValidatorConfig()\n\t\tvalidator := standard.NewValidator(ctx, robot, config)\n\n\t\ttask := &types.Task{\n\t\t\tID: \"task-001\",\n\t\t\tValidationRules: []string{\n\t\t\t\t`{\"type\": \"regex\", \"value\": \"^[A-Z][a-z]+$\"}`,\n\t\t\t},\n\t\t}\n\n\t\tresult := validator.ValidateWithContext(task, \"Hello\", nil)\n\t\tassert.True(t, result.Passed)\n\n\t\tresult2 := validator.ValidateWithContext(task, \"hello\", nil)\n\t\tassert.False(t, result2.Passed)\n\t})\n\n\tt.Run(\"parses json_path rules\", func(t *testing.T) {\n\t\trobot := createValidatorTestRobot(t)\n\t\tconfig := standard.DefaultValidatorConfig()\n\t\tvalidator := standard.NewValidator(ctx, robot, config)\n\n\t\ttask := &types.Task{\n\t\t\tID: \"task-001\",\n\t\t\tValidationRules: []string{\n\t\t\t\t`{\"type\": \"json_path\", \"path\": \"data.count\", \"value\": 42}`,\n\t\t\t},\n\t\t}\n\n\t\tresult := validator.ValidateWithContext(task, map[string]interface{}{\n\t\t\t\"data\": map[string]interface{}{\n\t\t\t\t\"count\": 42,\n\t\t\t},\n\t\t}, nil)\n\t\tassert.True(t, result.Passed)\n\n\t\tresult2 := validator.ValidateWithContext(task, map[string]interface{}{\n\t\t\t\"data\": map[string]interface{}{\n\t\t\t\t\"count\": 10,\n\t\t\t},\n\t\t}, nil)\n\t\tassert.False(t, result2.Passed)\n\t})\n\n\tt.Run(\"parses type rules with path\", func(t *testing.T) {\n\t\trobot := createValidatorTestRobot(t)\n\t\tconfig := standard.DefaultValidatorConfig()\n\t\tvalidator := standard.NewValidator(ctx, robot, config)\n\n\t\ttask := &types.Task{\n\t\t\tID: \"task-001\",\n\t\t\tValidationRules: []string{\n\t\t\t\t`{\"type\": \"type\", \"path\": \"items\", \"value\": \"array\"}`,\n\t\t\t},\n\t\t}\n\n\t\tresult := validator.ValidateWithContext(task, map[string]interface{}{\n\t\t\t\"items\": []interface{}{\"a\", \"b\"},\n\t\t}, nil)\n\t\tassert.True(t, result.Passed)\n\n\t\tresult2 := validator.ValidateWithContext(task, map[string]interface{}{\n\t\t\t\"items\": \"not an array\",\n\t\t}, nil)\n\t\tassert.False(t, result2.Passed)\n\t})\n}\n\nfunc TestValidatorSemanticValidation(t *testing.T) {\n\tif testing.Short() {\n\t\tt.Skip(\"Skipping integration test\")\n\t}\n\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\tctx := types.NewContext(context.Background(), testAuth())\n\n\tt.Run(\"semantic validation with expected output\", func(t *testing.T) {\n\t\trobot := createValidatorTestRobot(t)\n\t\tconfig := standard.DefaultValidatorConfig()\n\t\tvalidator := standard.NewValidator(ctx, robot, config)\n\n\t\ttask := &types.Task{\n\t\t\tID: \"task-001\",\n\t\t\tExpectedOutput: \"A JSON object containing user information with name and email fields\",\n\t\t}\n\n\t\toutput := map[string]interface{}{\n\t\t\t\"name\": \"John Doe\",\n\t\t\t\"email\": \"john@example.com\",\n\t\t}\n\n\t\tresult := validator.ValidateWithContext(task, output, nil)\n\n\t\tt.Logf(\"Semantic validation: passed=%v, score=%.2f, complete=%v\",\n\t\t\tresult.Passed, result.Score, result.Complete)\n\t\tt.Logf(\"Details: %s\", result.Details)\n\n\t\t// Should pass semantic validation\n\t\tassert.NotNil(t, result)\n\t})\n\n\tt.Run(\"semantic validation with complex criteria\", func(t *testing.T) {\n\t\trobot := createValidatorTestRobot(t)\n\t\tconfig := standard.DefaultValidatorConfig()\n\t\tvalidator := standard.NewValidator(ctx, robot, config)\n\n\t\ttask := &types.Task{\n\t\t\tID: \"task-001\",\n\t\t\tExpectedOutput: \"A professional email with greeting, body, and signature\",\n\t\t\tMessages: []agentcontext.Message{\n\t\t\t\t{Role: agentcontext.RoleUser, Content: \"Write a professional email\"},\n\t\t\t},\n\t\t}\n\n\t\toutput := `Dear Mr. Smith,\n\nI hope this email finds you well. I am writing to follow up on our previous conversation regarding the project timeline.\n\nPlease let me know if you have any questions.\n\nBest regards,\nJohn Doe`\n\n\t\tresult := validator.ValidateWithContext(task, output, nil)\n\n\t\tt.Logf(\"Email validation: passed=%v, score=%.2f\", result.Passed, result.Score)\n\n\t\t// Should recognize this as a valid professional email\n\t\tassert.NotNil(t, result)\n\t})\n}\n\nfunc TestValidatorMergeResults(t *testing.T) {\n\tif testing.Short() {\n\t\tt.Skip(\"Skipping integration test\")\n\t}\n\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\tctx := types.NewContext(context.Background(), testAuth())\n\n\tt.Run(\"both rule and semantic validation pass\", func(t *testing.T) {\n\t\trobot := createValidatorTestRobot(t)\n\t\tconfig := standard.DefaultValidatorConfig()\n\t\tvalidator := standard.NewValidator(ctx, robot, config)\n\n\t\ttask := &types.Task{\n\t\t\tID: \"task-001\",\n\t\t\tExpectedOutput: \"A greeting message\",\n\t\t\tValidationRules: []string{\n\t\t\t\t`{\"type\": \"contains\", \"value\": \"Hello\"}`,\n\t\t\t},\n\t\t}\n\n\t\tresult := validator.ValidateWithContext(task, \"Hello, how are you today?\", nil)\n\n\t\tassert.True(t, result.Passed)\n\t\tassert.True(t, result.Complete)\n\t})\n\n\tt.Run(\"rule passes but semantic fails\", func(t *testing.T) {\n\t\trobot := createValidatorTestRobot(t)\n\t\tconfig := standard.DefaultValidatorConfig()\n\t\tvalidator := standard.NewValidator(ctx, robot, config)\n\n\t\ttask := &types.Task{\n\t\t\tID: \"task-001\",\n\t\t\tExpectedOutput: \"A formal business letter with proper formatting\",\n\t\t\tValidationRules: []string{\n\t\t\t\t`{\"type\": \"contains\", \"value\": \"Hello\"}`, // This will pass\n\t\t\t},\n\t\t}\n\n\t\t// Contains \"Hello\" but not a formal business letter\n\t\tresult := validator.ValidateWithContext(task, \"Hello there buddy!\", nil)\n\n\t\t// Rule passes, but semantic might not\n\t\tt.Logf(\"Merged result: passed=%v, score=%.2f\", result.Passed, result.Score)\n\t})\n\n\tt.Run(\"rule fails - semantic not run\", func(t *testing.T) {\n\t\trobot := createValidatorTestRobot(t)\n\t\tconfig := standard.DefaultValidatorConfig()\n\t\tvalidator := standard.NewValidator(ctx, robot, config)\n\n\t\ttask := &types.Task{\n\t\t\tID: \"task-001\",\n\t\t\tExpectedOutput: \"Some expected output\",\n\t\t\tValidationRules: []string{\n\t\t\t\t`{\"type\": \"contains\", \"value\": \"REQUIRED_STRING\"}`,\n\t\t\t},\n\t\t}\n\n\t\tresult := validator.ValidateWithContext(task, \"Output without required string\", nil)\n\n\t\t// Should fail at rule level, semantic not needed\n\t\tassert.False(t, result.Passed)\n\t\tassert.False(t, result.Complete)\n\t})\n}\n\n// ============================================================================\n// Helper Functions\n// ============================================================================\n\n// createValidatorTestRobot creates a test robot for validator tests\nfunc createValidatorTestRobot(t *testing.T) *types.Robot {\n\tt.Helper()\n\treturn &types.Robot{\n\t\tMemberID: \"test-robot-validator\",\n\t\tTeamID: \"test-team-1\",\n\t\tDisplayName: \"Test Robot for Validator\",\n\t\tSystemPrompt: \"You are a helpful assistant.\",\n\t\tConfig: &types.Config{\n\t\t\tIdentity: &types.Identity{\n\t\t\t\tRole: \"Test Assistant\",\n\t\t\t\tDuties: []string{\"Validate outputs\"},\n\t\t\t},\n\t\t\tResources: &types.Resources{\n\t\t\t\tPhases: map[types.Phase]string{\n\t\t\t\t\ttypes.PhaseRun: \"robot.validation\",\n\t\t\t\t\t\"validation\": \"robot.validation\", // For semantic validation agent\n\t\t\t\t},\n\t\t\t\tAgents: []string{\n\t\t\t\t\t\"experts.data-analyst\",\n\t\t\t\t\t\"experts.text-writer\",\n\t\t\t\t},\n\t\t\t},\n\t\t},\n\t}\n}\n" }, { "path": "agent/robot/executor/types/helpers.go", "content": "package types\n\nimport (\n\t\"time\"\n\n\trobottypes \"github.com/yaoapp/yao/agent/robot/types\"\n)\n\n// BuildTriggerInput builds TriggerInput from trigger data\n// Shared helper used by all executor implementations\nfunc BuildTriggerInput(trigger robottypes.TriggerType, data interface{}) *robottypes.TriggerInput {\n\tinput := &robottypes.TriggerInput{}\n\n\tswitch trigger {\n\tcase robottypes.TriggerClock:\n\t\tinput.Clock = robottypes.NewClockContext(time.Now(), \"\")\n\n\tcase robottypes.TriggerHuman:\n\t\tif existing, ok := data.(*robottypes.TriggerInput); ok {\n\t\t\treturn existing\n\t\t}\n\t\tif req, ok := data.(*robottypes.InterveneRequest); ok {\n\t\t\tinput.Action = req.Action\n\t\t\tinput.Messages = req.Messages\n\t\t}\n\n\tcase robottypes.TriggerEvent:\n\t\tif req, ok := data.(*robottypes.EventRequest); ok {\n\t\t\tinput.Source = robottypes.EventSource(req.Source)\n\t\t\tinput.EventType = req.EventType\n\t\t\tinput.Data = req.Data\n\t\t}\n\n\t}\n\n\treturn input\n}\n" }, { "path": "agent/robot/executor/types/types.go", "content": "package types\n\nimport (\n\t\"time\"\n\n\trobottypes \"github.com/yaoapp/yao/agent/robot/types\"\n)\n\n// Executor defines the interface for robot phase execution\n// Different implementations provide different execution strategies:\n// - Standard: Real Agent calls with full phase execution\n// - DryRun: Plan-only mode, simulates execution without Agent calls\n// - Sandbox: Isolated execution with resource limits and safety controls\ntype Executor interface {\n\t// ExecuteWithControl runs a robot through all applicable phases with execution control\n\t// ctx: Execution context with auth and logging\n\t// robot: Robot configuration and state\n\t// trigger: What triggered this execution (clock, human, event)\n\t// data: Trigger-specific data (human input, event payload, etc.)\n\t// execID: Pre-generated execution ID (empty string to auto-generate)\n\t// control: Optional execution control for pause/resume functionality\n\t// Returns: Execution record with all phase outputs\n\tExecuteWithControl(ctx *robottypes.Context, robot *robottypes.Robot, trigger robottypes.TriggerType, data interface{}, execID string, control robottypes.ExecutionControl) (*robottypes.Execution, error)\n\n\t// ExecuteWithID runs a robot through all applicable phases with a pre-generated execution ID\n\t// This is a convenience wrapper around ExecuteWithControl without control\n\tExecuteWithID(ctx *robottypes.Context, robot *robottypes.Robot, trigger robottypes.TriggerType, data interface{}, execID string) (*robottypes.Execution, error)\n\n\t// Execute runs a robot through all applicable phases (auto-generates execution ID)\n\t// This is a convenience wrapper around ExecuteWithControl\n\tExecute(ctx *robottypes.Context, robot *robottypes.Robot, trigger robottypes.TriggerType, data interface{}) (*robottypes.Execution, error)\n\n\t// Resume resumes a suspended execution with human-provided input.\n\t// Loads the execution from persistent storage, restores state from ResumeContext,\n\t// and continues from where it was suspended.\n\t// Returns ErrExecutionSuspended if the execution suspends again during resume.\n\tResume(ctx *robottypes.Context, execID string, reply string) error\n\n\t// Metrics and control\n\tExecCount() int // Total execution count\n\tCurrentCount() int // Currently running execution count\n\tReset() // Reset counters (for testing)\n}\n\n// PhaseExecutor defines the interface for individual phase execution\n// Used internally by Executor implementations\ntype PhaseExecutor interface {\n\t// RunInspiration executes P0: Inspiration phase\n\tRunInspiration(ctx *robottypes.Context, exec *robottypes.Execution, data interface{}) error\n\n\t// RunGoals executes P1: Goals phase\n\tRunGoals(ctx *robottypes.Context, exec *robottypes.Execution, data interface{}) error\n\n\t// RunTasks executes P2: Tasks phase\n\tRunTasks(ctx *robottypes.Context, exec *robottypes.Execution, data interface{}) error\n\n\t// RunExecution executes P3: Run phase (task execution)\n\tRunExecution(ctx *robottypes.Context, exec *robottypes.Execution, data interface{}) error\n\n\t// RunDelivery executes P4: Delivery phase\n\tRunDelivery(ctx *robottypes.Context, exec *robottypes.Execution, data interface{}) error\n\n\t// RunLearning executes P5: Learning phase\n\tRunLearning(ctx *robottypes.Context, exec *robottypes.Execution, data interface{}) error\n}\n\n// Config holds common executor configuration\ntype Config struct {\n\t// SkipPersistence skips execution record persistence (for testing)\n\tSkipPersistence bool\n\n\t// OnPhaseStart callback when a phase starts\n\tOnPhaseStart func(phase robottypes.Phase)\n\n\t// OnPhaseEnd callback when a phase ends\n\tOnPhaseEnd func(phase robottypes.Phase)\n}\n\n// DryRunConfig holds dry-run specific configuration\ntype DryRunConfig struct {\n\tConfig\n\n\t// Delay simulates execution delay for each phase\n\tDelay time.Duration\n\n\t// OnStart callback on execution start\n\tOnStart func()\n\n\t// OnEnd callback on execution end\n\tOnEnd func()\n}\n\n// SandboxConfig holds sandbox specific configuration\n//\n// ⚠️ NOT IMPLEMENTED: These settings are placeholders for future\n// container-based isolation. True sandbox requires infrastructure support\n// (Docker/gVisor/Firecracker). Current implementation behaves like DryRun.\ntype SandboxConfig struct {\n\tConfig\n\n\t// MaxDuration limits total execution time\n\tMaxDuration time.Duration\n\n\t// MaxMemory limits memory usage (bytes) - requires container runtime\n\tMaxMemory int64\n\n\t// AllowedAgents restricts which agents can be called\n\tAllowedAgents []string\n\n\t// AllowedTools restricts which tools can be used\n\tAllowedTools []string\n\n\t// NetworkAccess controls network access - requires container networking\n\tNetworkAccess bool\n\n\t// FileAccess controls file system access - requires container filesystem\n\tFileAccess bool\n}\n\n// Mode represents the executor mode\ntype Mode string\n\nconst (\n\tModeStandard Mode = \"standard\" // Real Agent execution (production)\n\tModeDryRun Mode = \"dryrun\" // Simulated execution (testing/demo)\n\tModeSandbox Mode = \"sandbox\" // Container-isolated execution (NOT IMPLEMENTED)\n)\n\n// Setting holds executor settings from configuration\ntype Setting struct {\n\tMode Mode `json:\"mode,omitempty\" yaml:\"mode,omitempty\"` // Executor mode\n\tMaxDuration time.Duration `json:\"max_duration,omitempty\" yaml:\"max_duration,omitempty\"` // Max execution time\n\tMaxMemory int64 `json:\"max_memory,omitempty\" yaml:\"max_memory,omitempty\"` // Max memory (bytes)\n\tAllowedAgents []string `json:\"allowed_agents,omitempty\" yaml:\"allowed_agents,omitempty\"` // Allowed agent IDs\n\tNetworkAccess bool `json:\"network_access,omitempty\" yaml:\"network_access,omitempty\"` // Allow network\n\tFileAccess bool `json:\"file_access,omitempty\" yaml:\"file_access,omitempty\"` // Allow file system\n}\n\n// DefaultSetting returns default executor settings\nfunc DefaultSetting() *Setting {\n\treturn &Setting{\n\t\tMode: ModeStandard,\n\t\tMaxDuration: 30 * time.Minute,\n\t\tMaxMemory: 512 * 1024 * 1024, // 512MB\n\t\tNetworkAccess: true,\n\t\tFileAccess: false,\n\t}\n}\n" }, { "path": "agent/robot/logger/logger.go", "content": "package logger\n\nimport (\n\t\"fmt\"\n\n\tkunlog \"github.com/yaoapp/kun/log\"\n\t\"github.com/yaoapp/yao/config\"\n)\n\nconst (\n\tReset = \"\\033[0m\"\n\tRed = \"\\033[31m\"\n\tGreen = \"\\033[32m\"\n\tYellow = \"\\033[33m\"\n\tCyan = \"\\033[36m\"\n\tWhite = \"\\033[37m\"\n\tGray = \"\\033[90m\"\n\tBoldCyan = \"\\033[1;36m\"\n\tBoldGreen = \"\\033[1;32m\"\n\tBoldRed = \"\\033[1;31m\"\n\n\treset = Reset\n\tred = Red\n\tyellow = Yellow\n\tcyan = Cyan\n\tgray = Gray\n)\n\n// Logger provides robot-level structured logging. All integration adapters,\n// dispatchers, event handlers, etc. share this implementation.\n//\n// Dev mode → colored stdout + kun/log.Trace (unified).\n// Prod mode → kun/log at matching level.\ntype Logger struct {\n\ttag string\n}\n\n// New creates a Logger tagged with the given component name\n// (e.g. \"telegram\", \"dispatcher\", \"message\", \"delivery\").\nfunc New(tag string) *Logger {\n\treturn &Logger{tag: tag}\n}\n\nfunc (l *Logger) prefix() string {\n\treturn fmt.Sprintf(\"[robot:%s]\", l.tag)\n}\n\nfunc (l *Logger) Trace(format string, args ...interface{}) {\n\tmsg := fmt.Sprintf(format, args...)\n\tif config.IsDevelopment() {\n\t\tfmt.Printf(\"%s → %s %s%s\\n\", gray, l.prefix(), msg, reset)\n\t}\n\tkunlog.Trace(\"%s %s\", l.prefix(), msg)\n}\n\nfunc (l *Logger) Debug(format string, args ...interface{}) {\n\tmsg := fmt.Sprintf(format, args...)\n\tif config.IsDevelopment() {\n\t\tfmt.Printf(\"%s • %s %s%s\\n\", gray, l.prefix(), msg, reset)\n\t}\n\tkunlog.Debug(\"%s %s\", l.prefix(), msg)\n}\n\nfunc (l *Logger) Info(format string, args ...interface{}) {\n\tmsg := fmt.Sprintf(format, args...)\n\tif config.IsDevelopment() {\n\t\tfmt.Printf(\"%s ℹ %s %s%s\\n\", cyan, l.prefix(), msg, reset)\n\t}\n\tkunlog.Info(\"%s %s\", l.prefix(), msg)\n}\n\nfunc (l *Logger) Warn(format string, args ...interface{}) {\n\tmsg := fmt.Sprintf(format, args...)\n\tif config.IsDevelopment() {\n\t\tfmt.Printf(\"%s ⚠ %s %s%s\\n\", yellow, l.prefix(), msg, reset)\n\t}\n\tkunlog.Warn(\"%s %s\", l.prefix(), msg)\n}\n\nfunc (l *Logger) Error(format string, args ...interface{}) {\n\tmsg := fmt.Sprintf(format, args...)\n\tif config.IsDevelopment() {\n\t\tfmt.Printf(\"%s ✗ %s %s%s\\n\", red, l.prefix(), msg, reset)\n\t}\n\tkunlog.Error(\"%s %s\", l.prefix(), msg)\n}\n\n// IsDev returns true when running in development mode.\nfunc IsDev() bool {\n\treturn config.IsDevelopment()\n}\n\n// Raw writes pre-formatted text directly to stdout in dev mode only.\n// Use for rich multi-line output (box-style logs, tables, etc.)\n// that should bypass the standard single-line prefix format.\nfunc Raw(s string) {\n\tif config.IsDevelopment() {\n\t\tfmt.Print(s)\n\t}\n}\n" }, { "path": "agent/robot/manager/integration_clock_test.go", "content": "package manager_test\n\n// Integration tests for Clock trigger modes\n// Tests all three clock modes: times, interval, daemon\n// Includes timezone handling and day-of-week filtering\n\nimport (\n\t\"context\"\n\t\"encoding/json\"\n\t\"testing\"\n\t\"time\"\n\n\t\"github.com/stretchr/testify/assert\"\n\t\"github.com/stretchr/testify/require\"\n\t\"github.com/yaoapp/gou/model\"\n\t\"github.com/yaoapp/xun/capsule\"\n\t\"github.com/yaoapp/yao/agent/robot/executor\"\n\t\"github.com/yaoapp/yao/agent/robot/manager\"\n\t\"github.com/yaoapp/yao/agent/robot/pool\"\n\t\"github.com/yaoapp/yao/agent/robot/types\"\n\t\"github.com/yaoapp/yao/agent/testutils\"\n)\n\n// createClockTestManager creates a manager with mock executor for clock tests\nfunc createClockTestManager(t *testing.T, tickInterval time.Duration, workerSize, queueSize int) (*manager.Manager, *executor.DryRunExecutor) {\n\texec := executor.NewDryRunWithDelay(0)\n\tconfig := &manager.Config{\n\t\tTickInterval: tickInterval,\n\t\tPoolConfig: &pool.Config{WorkerSize: workerSize, QueueSize: queueSize},\n\t\tExecutor: exec,\n\t}\n\tm := manager.NewWithConfig(config)\n\treturn m, exec\n}\n\n// ==================== Times Mode Tests ====================\n\n// TestIntegrationClockTimesMode tests the times mode clock trigger\nfunc TestIntegrationClockTimesMode(t *testing.T) {\n\tif testing.Short() {\n\t\tt.Skip(\"Skipping integration test\")\n\t}\n\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\tcleanupIntegrationRobots(t)\n\tdefer cleanupIntegrationRobots(t)\n\n\tt.Run(\"triggers at configured time\", func(t *testing.T) {\n\t\t// Clean up before each subtest to ensure isolation\n\t\tcleanupIntegrationRobots(t)\n\n\t\tsetupClockTestRobot(t, \"robot_integ_clock_times1\", \"team_integ_clock\", map[string]interface{}{\n\t\t\t\"mode\": \"times\",\n\t\t\t\"times\": []string{\"09:00\", \"14:00\", \"17:00\"},\n\t\t\t\"days\": []string{\"Mon\", \"Tue\", \"Wed\", \"Thu\", \"Fri\"},\n\t\t\t\"tz\": \"Asia/Shanghai\",\n\t\t})\n\n\t\tm, exec := createClockTestManager(t, 100*time.Millisecond, 3, 20)\n\n\t\terr := m.Start()\n\t\trequire.NoError(t, err)\n\t\tdefer m.Stop()\n\n\t\t// Verify robot is loaded into cache\n\t\trobot := m.Cache().Get(\"robot_integ_clock_times1\")\n\t\trequire.NotNil(t, robot, \"Robot should be loaded into cache\")\n\n\t\texec.Reset()\n\n\t\t// Trigger at 09:00 on Wednesday\n\t\tloc, _ := time.LoadLocation(\"Asia/Shanghai\")\n\t\tnow := time.Date(2025, 1, 15, 9, 0, 0, 0, loc) // Wednesday 09:00\n\n\t\tctx := types.NewContext(context.Background(), nil)\n\t\terr = m.Tick(ctx, now)\n\t\tassert.NoError(t, err)\n\n\t\ttime.Sleep(300 * time.Millisecond)\n\t\tassert.GreaterOrEqual(t, exec.ExecCount(), 1, \"Should trigger at 09:00\")\n\t})\n\n\tt.Run(\"does not trigger at non-configured time\", func(t *testing.T) {\n\t\t// Clean up before each subtest to ensure isolation\n\t\tcleanupIntegrationRobots(t)\n\n\t\tsetupClockTestRobot(t, \"robot_integ_clock_times2\", \"team_integ_clock\", map[string]interface{}{\n\t\t\t\"mode\": \"times\",\n\t\t\t\"times\": []string{\"09:00\", \"14:00\"},\n\t\t\t\"days\": []string{\"Mon\", \"Tue\", \"Wed\", \"Thu\", \"Fri\"},\n\t\t\t\"tz\": \"Asia/Shanghai\",\n\t\t})\n\n\t\tm, exec := createClockTestManager(t, 100*time.Millisecond, 3, 20)\n\n\t\terr := m.Start()\n\t\trequire.NoError(t, err)\n\t\tdefer m.Stop()\n\n\t\texec.Reset()\n\n\t\t// Trigger at 10:30 (not configured)\n\t\tloc, _ := time.LoadLocation(\"Asia/Shanghai\")\n\t\tnow := time.Date(2025, 1, 15, 10, 30, 0, 0, loc) // Wednesday 10:30\n\n\t\tctx := types.NewContext(context.Background(), nil)\n\t\terr = m.Tick(ctx, now)\n\t\tassert.NoError(t, err)\n\n\t\ttime.Sleep(200 * time.Millisecond)\n\t\tassert.Equal(t, 0, exec.ExecCount(), \"Should not trigger at non-configured time\")\n\t})\n\n\tt.Run(\"does not trigger on non-configured day\", func(t *testing.T) {\n\t\t// Clean up before each subtest to ensure isolation\n\t\tcleanupIntegrationRobots(t)\n\n\t\tsetupClockTestRobot(t, \"robot_integ_clock_times3\", \"team_integ_clock\", map[string]interface{}{\n\t\t\t\"mode\": \"times\",\n\t\t\t\"times\": []string{\"09:00\"},\n\t\t\t\"days\": []string{\"Mon\", \"Tue\", \"Wed\", \"Thu\", \"Fri\"}, // Weekdays only\n\t\t\t\"tz\": \"Asia/Shanghai\",\n\t\t})\n\n\t\tm, exec := createClockTestManager(t, 100*time.Millisecond, 3, 20)\n\n\t\terr := m.Start()\n\t\trequire.NoError(t, err)\n\t\tdefer m.Stop()\n\n\t\texec.Reset()\n\n\t\t// Trigger at 09:00 on Saturday (not configured)\n\t\tloc, _ := time.LoadLocation(\"Asia/Shanghai\")\n\t\tnow := time.Date(2025, 1, 18, 9, 0, 0, 0, loc) // Saturday 09:00\n\n\t\tctx := types.NewContext(context.Background(), nil)\n\t\terr = m.Tick(ctx, now)\n\t\tassert.NoError(t, err)\n\n\t\ttime.Sleep(200 * time.Millisecond)\n\t\tassert.Equal(t, 0, exec.ExecCount(), \"Should not trigger on Saturday\")\n\t})\n\n\tt.Run(\"wildcard days matches all days\", func(t *testing.T) {\n\t\t// Clean up before each subtest to ensure isolation\n\t\tcleanupIntegrationRobots(t)\n\n\t\tsetupClockTestRobot(t, \"robot_integ_clock_times4\", \"team_integ_clock\", map[string]interface{}{\n\t\t\t\"mode\": \"times\",\n\t\t\t\"times\": []string{\"09:00\"},\n\t\t\t\"days\": []string{\"*\"}, // All days\n\t\t\t\"tz\": \"Asia/Shanghai\",\n\t\t})\n\n\t\tm, exec := createClockTestManager(t, 100*time.Millisecond, 3, 20)\n\n\t\terr := m.Start()\n\t\trequire.NoError(t, err)\n\t\tdefer m.Stop()\n\n\t\texec.Reset()\n\n\t\t// Trigger at 09:00 on Saturday\n\t\tloc, _ := time.LoadLocation(\"Asia/Shanghai\")\n\t\tnow := time.Date(2025, 1, 18, 9, 0, 0, 0, loc) // Saturday 09:00\n\n\t\tctx := types.NewContext(context.Background(), nil)\n\t\terr = m.Tick(ctx, now)\n\t\tassert.NoError(t, err)\n\n\t\ttime.Sleep(300 * time.Millisecond)\n\t\tassert.GreaterOrEqual(t, exec.ExecCount(), 1, \"Should trigger on Saturday with wildcard days\")\n\t})\n\n\tt.Run(\"dedup prevents double trigger in same minute\", func(t *testing.T) {\n\t\t// Clean up before each subtest to ensure isolation\n\t\tcleanupIntegrationRobots(t)\n\n\t\tsetupClockTestRobot(t, \"robot_integ_clock_times5\", \"team_integ_clock\", map[string]interface{}{\n\t\t\t\"mode\": \"times\",\n\t\t\t\"times\": []string{\"09:00\"},\n\t\t\t\"days\": []string{\"*\"},\n\t\t\t\"tz\": \"Asia/Shanghai\",\n\t\t})\n\n\t\tm, exec := createClockTestManager(t, 100*time.Millisecond, 3, 20)\n\n\t\terr := m.Start()\n\t\trequire.NoError(t, err)\n\t\tdefer m.Stop()\n\n\t\texec.Reset()\n\n\t\tloc, _ := time.LoadLocation(\"Asia/Shanghai\")\n\t\tctx := types.NewContext(context.Background(), nil)\n\n\t\t// First tick at 09:00:00\n\t\tnow1 := time.Date(2025, 1, 15, 9, 0, 0, 0, loc)\n\t\terr = m.Tick(ctx, now1)\n\t\tassert.NoError(t, err)\n\t\ttime.Sleep(200 * time.Millisecond)\n\t\tfirstCount := exec.ExecCount()\n\t\tassert.GreaterOrEqual(t, firstCount, 1, \"First tick should trigger\")\n\n\t\t// Second tick at 09:00:30 (same minute)\n\t\tnow2 := time.Date(2025, 1, 15, 9, 0, 30, 0, loc)\n\t\terr = m.Tick(ctx, now2)\n\t\tassert.NoError(t, err)\n\t\ttime.Sleep(200 * time.Millisecond)\n\n\t\t// Should not trigger again in same minute\n\t\tassert.Equal(t, firstCount, exec.ExecCount(), \"Should not trigger twice in same minute\")\n\t})\n}\n\n// ==================== Interval Mode Tests ====================\n\n// TestIntegrationClockIntervalMode tests the interval mode clock trigger\nfunc TestIntegrationClockIntervalMode(t *testing.T) {\n\tif testing.Short() {\n\t\tt.Skip(\"Skipping integration test\")\n\t}\n\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\tcleanupIntegrationRobots(t)\n\tdefer cleanupIntegrationRobots(t)\n\n\tt.Run(\"triggers on first run\", func(t *testing.T) {\n\t\t// Clean up before each subtest to ensure isolation\n\t\tcleanupIntegrationRobots(t)\n\n\t\tsetupClockTestRobot(t, \"robot_integ_clock_interval1\", \"team_integ_clock\", map[string]interface{}{\n\t\t\t\"mode\": \"interval\",\n\t\t\t\"every\": \"30m\",\n\t\t})\n\n\t\tm, exec := createClockTestManager(t, 100*time.Millisecond, 3, 20)\n\n\t\terr := m.Start()\n\t\trequire.NoError(t, err)\n\t\tdefer m.Stop()\n\n\t\texec.Reset()\n\n\t\tctx := types.NewContext(context.Background(), nil)\n\t\tnow := time.Now()\n\t\terr = m.Tick(ctx, now)\n\t\tassert.NoError(t, err)\n\n\t\ttime.Sleep(300 * time.Millisecond)\n\t\tassert.GreaterOrEqual(t, exec.ExecCount(), 1, \"Should trigger on first run\")\n\t})\n\n\tt.Run(\"triggers after interval passed\", func(t *testing.T) {\n\t\t// Clean up before each subtest to ensure isolation\n\t\tcleanupIntegrationRobots(t)\n\n\t\tsetupClockTestRobot(t, \"robot_integ_clock_interval2\", \"team_integ_clock\", map[string]interface{}{\n\t\t\t\"mode\": \"interval\",\n\t\t\t\"every\": \"100ms\", // Short interval for testing\n\t\t})\n\n\t\tm, exec := createClockTestManager(t, 50*time.Millisecond, 3, 20)\n\n\t\terr := m.Start()\n\t\trequire.NoError(t, err)\n\t\tdefer m.Stop()\n\n\t\texec.Reset()\n\n\t\tctx := types.NewContext(context.Background(), nil)\n\n\t\t// First tick\n\t\tnow1 := time.Now()\n\t\terr = m.Tick(ctx, now1)\n\t\tassert.NoError(t, err)\n\t\ttime.Sleep(200 * time.Millisecond)\n\t\tfirstCount := exec.ExecCount()\n\t\tassert.GreaterOrEqual(t, firstCount, 1, \"First tick should trigger\")\n\n\t\t// Wait for interval to pass\n\t\ttime.Sleep(150 * time.Millisecond)\n\n\t\t// Second tick after interval\n\t\tnow2 := time.Now()\n\t\terr = m.Tick(ctx, now2)\n\t\tassert.NoError(t, err)\n\t\ttime.Sleep(200 * time.Millisecond)\n\n\t\t// Should have triggered again\n\t\tassert.Greater(t, exec.ExecCount(), firstCount, \"Should trigger again after interval\")\n\t})\n\n\tt.Run(\"does not trigger before interval passed\", func(t *testing.T) {\n\t\t// Clean up before each subtest to ensure isolation\n\t\tcleanupIntegrationRobots(t)\n\n\t\tsetupClockTestRobot(t, \"robot_integ_clock_interval3\", \"team_integ_clock\", map[string]interface{}{\n\t\t\t\"mode\": \"interval\",\n\t\t\t\"every\": \"1h\", // Long interval\n\t\t})\n\n\t\tm, exec := createClockTestManager(t, 100*time.Millisecond, 3, 20)\n\n\t\terr := m.Start()\n\t\trequire.NoError(t, err)\n\t\tdefer m.Stop()\n\n\t\texec.Reset()\n\n\t\tctx := types.NewContext(context.Background(), nil)\n\n\t\t// First tick\n\t\tnow1 := time.Now()\n\t\terr = m.Tick(ctx, now1)\n\t\tassert.NoError(t, err)\n\t\ttime.Sleep(200 * time.Millisecond)\n\t\tfirstCount := exec.ExecCount()\n\t\tassert.GreaterOrEqual(t, firstCount, 1, \"First tick should trigger\")\n\n\t\t// Second tick immediately (interval not passed)\n\t\tnow2 := now1.Add(1 * time.Minute) // Only 1 minute later\n\t\terr = m.Tick(ctx, now2)\n\t\tassert.NoError(t, err)\n\t\ttime.Sleep(200 * time.Millisecond)\n\n\t\t// Should not trigger again\n\t\tassert.Equal(t, firstCount, exec.ExecCount(), \"Should not trigger before interval\")\n\t})\n}\n\n// ==================== Daemon Mode Tests ====================\n\n// TestIntegrationClockDaemonMode tests the daemon mode clock trigger\nfunc TestIntegrationClockDaemonMode(t *testing.T) {\n\tif testing.Short() {\n\t\tt.Skip(\"Skipping integration test\")\n\t}\n\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\tcleanupIntegrationRobots(t)\n\tdefer cleanupIntegrationRobots(t)\n\n\tt.Run(\"triggers when robot can run\", func(t *testing.T) {\n\t\tsetupClockTestRobot(t, \"robot_integ_clock_daemon1\", \"team_integ_clock\", map[string]interface{}{\n\t\t\t\"mode\": \"daemon\",\n\t\t\t\"timeout\": \"5m\",\n\t\t})\n\n\t\tm, exec := createClockTestManager(t, 100*time.Millisecond, 3, 20)\n\n\t\terr := m.Start()\n\t\trequire.NoError(t, err)\n\t\tdefer m.Stop()\n\n\t\texec.Reset()\n\n\t\tctx := types.NewContext(context.Background(), nil)\n\t\terr = m.Tick(ctx, time.Now())\n\t\tassert.NoError(t, err)\n\n\t\ttime.Sleep(300 * time.Millisecond)\n\t\tassert.GreaterOrEqual(t, exec.ExecCount(), 1, \"Daemon should trigger when idle\")\n\t})\n\n\tt.Run(\"respects quota limit\", func(t *testing.T) {\n\t\t// Create daemon robot with Max=1\n\t\tsetupClockTestRobotWithQuota(t, \"robot_integ_clock_daemon2\", \"team_integ_clock\",\n\t\t\tmap[string]interface{}{\n\t\t\t\t\"mode\": \"daemon\",\n\t\t\t\t\"timeout\": \"5m\",\n\t\t\t},\n\t\t\t1, 5, 5) // Max=1, Queue=5\n\n\t\tm, exec := createClockTestManager(t, 50*time.Millisecond, 5, 20)\n\n\t\terr := m.Start()\n\t\trequire.NoError(t, err)\n\t\tdefer m.Stop()\n\n\t\texec.Reset()\n\n\t\tctx := types.NewContext(context.Background(), nil)\n\n\t\t// Trigger multiple times rapidly\n\t\tfor i := 0; i < 5; i++ {\n\t\t\terr = m.Tick(ctx, time.Now())\n\t\t\tassert.NoError(t, err)\n\t\t\ttime.Sleep(60 * time.Millisecond)\n\t\t}\n\n\t\t// Robot should respect quota (Max=1)\n\t\trobot := m.Cache().Get(\"robot_integ_clock_daemon2\")\n\t\tassert.NotNil(t, robot)\n\t\t// Running count should be at most Max\n\t\tassert.LessOrEqual(t, robot.RunningCount(), 1, \"Should respect quota limit\")\n\t})\n}\n\n// ==================== Timezone Tests ====================\n\n// TestIntegrationClockTimezone tests timezone handling\nfunc TestIntegrationClockTimezone(t *testing.T) {\n\tif testing.Short() {\n\t\tt.Skip(\"Skipping integration test\")\n\t}\n\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\tcleanupIntegrationRobots(t)\n\tdefer cleanupIntegrationRobots(t)\n\n\tt.Run(\"respects robot timezone\", func(t *testing.T) {\n\t\t// Robot configured for Asia/Shanghai (UTC+8)\n\t\tsetupClockTestRobot(t, \"robot_integ_clock_tz1\", \"team_integ_clock\", map[string]interface{}{\n\t\t\t\"mode\": \"times\",\n\t\t\t\"times\": []string{\"09:00\"},\n\t\t\t\"days\": []string{\"*\"},\n\t\t\t\"tz\": \"Asia/Shanghai\",\n\t\t})\n\n\t\tm, exec := createClockTestManager(t, 100*time.Millisecond, 3, 20)\n\n\t\terr := m.Start()\n\t\trequire.NoError(t, err)\n\t\tdefer m.Stop()\n\n\t\texec.Reset()\n\n\t\tctx := types.NewContext(context.Background(), nil)\n\n\t\t// 09:00 in Shanghai = 01:00 UTC\n\t\tshanghai, _ := time.LoadLocation(\"Asia/Shanghai\")\n\t\tshanghaiTime := time.Date(2025, 1, 15, 9, 0, 0, 0, shanghai)\n\n\t\terr = m.Tick(ctx, shanghaiTime)\n\t\tassert.NoError(t, err)\n\n\t\ttime.Sleep(300 * time.Millisecond)\n\t\tassert.GreaterOrEqual(t, exec.ExecCount(), 1, \"Should trigger at 09:00 Shanghai time\")\n\t})\n\n\tt.Run(\"different timezone same UTC time\", func(t *testing.T) {\n\t\t// Robot 1: Asia/Shanghai at 09:00 (UTC+8) = 01:00 UTC\n\t\tsetupClockTestRobot(t, \"robot_integ_clock_tz2\", \"team_integ_clock\", map[string]interface{}{\n\t\t\t\"mode\": \"times\",\n\t\t\t\"times\": []string{\"09:00\"},\n\t\t\t\"days\": []string{\"*\"},\n\t\t\t\"tz\": \"Asia/Shanghai\",\n\t\t})\n\n\t\t// Robot 2: America/New_York at 09:00 (UTC-5) = 14:00 UTC\n\t\tsetupClockTestRobot(t, \"robot_integ_clock_tz3\", \"team_integ_clock\", map[string]interface{}{\n\t\t\t\"mode\": \"times\",\n\t\t\t\"times\": []string{\"09:00\"},\n\t\t\t\"days\": []string{\"*\"},\n\t\t\t\"tz\": \"America/New_York\",\n\t\t})\n\n\t\tm, exec := createClockTestManager(t, 100*time.Millisecond, 3, 20)\n\n\t\terr := m.Start()\n\t\trequire.NoError(t, err)\n\t\tdefer m.Stop()\n\n\t\texec.Reset()\n\n\t\tctx := types.NewContext(context.Background(), nil)\n\n\t\t// Test at 01:00 UTC (09:00 Shanghai)\n\t\tutcTime := time.Date(2025, 1, 15, 1, 0, 0, 0, time.UTC)\n\t\terr = m.Tick(ctx, utcTime)\n\t\tassert.NoError(t, err)\n\n\t\ttime.Sleep(300 * time.Millisecond)\n\n\t\t// Only Shanghai robot should trigger\n\t\texecCount := exec.ExecCount()\n\t\tassert.GreaterOrEqual(t, execCount, 1, \"Shanghai robot should trigger\")\n\t\t// New York robot should not trigger (it's 20:00 in NY)\n\t})\n}\n\n// ==================== Edge Cases ====================\n\n// TestIntegrationClockEdgeCases tests edge cases in clock triggering\nfunc TestIntegrationClockEdgeCases(t *testing.T) {\n\tif testing.Short() {\n\t\tt.Skip(\"Skipping integration test\")\n\t}\n\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\tcleanupIntegrationRobots(t)\n\tdefer cleanupIntegrationRobots(t)\n\n\tt.Run(\"robot with clock disabled is skipped\", func(t *testing.T) {\n\t\t// Create robot with clock trigger disabled\n\t\tm := model.Select(\"__yao.member\")\n\t\ttableName := m.MetaData.Table.Name\n\t\tqb := capsule.Query()\n\n\t\trobotConfig := map[string]interface{}{\n\t\t\t\"identity\": map[string]interface{}{\"role\": \"Clock Disabled Robot\"},\n\t\t\t\"triggers\": map[string]interface{}{\n\t\t\t\t\"clock\": map[string]interface{}{\"enabled\": false},\n\t\t\t},\n\t\t\t\"clock\": map[string]interface{}{\n\t\t\t\t\"mode\": \"times\",\n\t\t\t\t\"times\": []string{\"09:00\"},\n\t\t\t\t\"tz\": \"Asia/Shanghai\",\n\t\t\t},\n\t\t}\n\t\tconfigJSON, _ := json.Marshal(robotConfig)\n\n\t\terr := qb.Table(tableName).Insert([]map[string]interface{}{\n\t\t\t{\n\t\t\t\t\"member_id\": \"robot_integ_clock_disabled\",\n\t\t\t\t\"team_id\": \"team_integ_clock\",\n\t\t\t\t\"member_type\": \"robot\",\n\t\t\t\t\"display_name\": \"Clock Disabled Robot\",\n\t\t\t\t\"status\": \"active\",\n\t\t\t\t\"role_id\": \"member\",\n\t\t\t\t\"autonomous_mode\": true,\n\t\t\t\t\"robot_status\": \"idle\",\n\t\t\t\t\"robot_config\": string(configJSON),\n\t\t\t},\n\t\t})\n\t\trequire.NoError(t, err)\n\n\t\tmgr, exec := createClockTestManager(t, 100*time.Millisecond, 3, 20)\n\n\t\terr = mgr.Start()\n\t\trequire.NoError(t, err)\n\t\tdefer mgr.Stop()\n\n\t\texec.Reset()\n\n\t\t// Trigger at matching time\n\t\tloc, _ := time.LoadLocation(\"Asia/Shanghai\")\n\t\tnow := time.Date(2025, 1, 15, 9, 0, 0, 0, loc)\n\n\t\tctx := types.NewContext(context.Background(), nil)\n\t\terr = mgr.Tick(ctx, now)\n\t\tassert.NoError(t, err)\n\n\t\ttime.Sleep(200 * time.Millisecond)\n\t\tassert.Equal(t, 0, exec.ExecCount(), \"Clock disabled robot should not trigger\")\n\t})\n\n\tt.Run(\"paused robot is skipped\", func(t *testing.T) {\n\t\t// Create paused robot\n\t\tm := model.Select(\"__yao.member\")\n\t\ttableName := m.MetaData.Table.Name\n\t\tqb := capsule.Query()\n\n\t\trobotConfig := map[string]interface{}{\n\t\t\t\"identity\": map[string]interface{}{\"role\": \"Paused Robot\"},\n\t\t\t\"triggers\": map[string]interface{}{\n\t\t\t\t\"clock\": map[string]interface{}{\"enabled\": true},\n\t\t\t},\n\t\t\t\"clock\": map[string]interface{}{\n\t\t\t\t\"mode\": \"times\",\n\t\t\t\t\"times\": []string{\"09:00\"},\n\t\t\t\t\"tz\": \"Asia/Shanghai\",\n\t\t\t},\n\t\t}\n\t\tconfigJSON, _ := json.Marshal(robotConfig)\n\n\t\terr := qb.Table(tableName).Insert([]map[string]interface{}{\n\t\t\t{\n\t\t\t\t\"member_id\": \"robot_integ_clock_paused\",\n\t\t\t\t\"team_id\": \"team_integ_clock\",\n\t\t\t\t\"member_type\": \"robot\",\n\t\t\t\t\"display_name\": \"Paused Robot\",\n\t\t\t\t\"status\": \"active\",\n\t\t\t\t\"role_id\": \"member\",\n\t\t\t\t\"autonomous_mode\": true,\n\t\t\t\t\"robot_status\": \"paused\", // Paused status\n\t\t\t\t\"robot_config\": string(configJSON),\n\t\t\t},\n\t\t})\n\t\trequire.NoError(t, err)\n\n\t\tmgr, exec := createClockTestManager(t, 100*time.Millisecond, 3, 20)\n\n\t\terr = mgr.Start()\n\t\trequire.NoError(t, err)\n\t\tdefer mgr.Stop()\n\n\t\texec.Reset()\n\n\t\t// Trigger at matching time\n\t\tloc, _ := time.LoadLocation(\"Asia/Shanghai\")\n\t\tnow := time.Date(2025, 1, 15, 9, 0, 0, 0, loc)\n\n\t\tctx := types.NewContext(context.Background(), nil)\n\t\terr = mgr.Tick(ctx, now)\n\t\tassert.NoError(t, err)\n\n\t\ttime.Sleep(200 * time.Millisecond)\n\t\tassert.Equal(t, 0, exec.ExecCount(), \"Paused robot should not trigger\")\n\t})\n\n\tt.Run(\"robot without clock config is skipped\", func(t *testing.T) {\n\t\t// Create robot without clock config\n\t\tm := model.Select(\"__yao.member\")\n\t\ttableName := m.MetaData.Table.Name\n\t\tqb := capsule.Query()\n\n\t\trobotConfig := map[string]interface{}{\n\t\t\t\"identity\": map[string]interface{}{\"role\": \"No Clock Robot\"},\n\t\t\t\"triggers\": map[string]interface{}{\n\t\t\t\t\"clock\": map[string]interface{}{\"enabled\": true},\n\t\t\t},\n\t\t\t// No clock config\n\t\t}\n\t\tconfigJSON, _ := json.Marshal(robotConfig)\n\n\t\terr := qb.Table(tableName).Insert([]map[string]interface{}{\n\t\t\t{\n\t\t\t\t\"member_id\": \"robot_integ_clock_noconfig\",\n\t\t\t\t\"team_id\": \"team_integ_clock\",\n\t\t\t\t\"member_type\": \"robot\",\n\t\t\t\t\"display_name\": \"No Clock Config Robot\",\n\t\t\t\t\"status\": \"active\",\n\t\t\t\t\"role_id\": \"member\",\n\t\t\t\t\"autonomous_mode\": true,\n\t\t\t\t\"robot_status\": \"idle\",\n\t\t\t\t\"robot_config\": string(configJSON),\n\t\t\t},\n\t\t})\n\t\trequire.NoError(t, err)\n\n\t\tmgr, exec := createClockTestManager(t, 100*time.Millisecond, 3, 20)\n\n\t\terr = mgr.Start()\n\t\trequire.NoError(t, err)\n\t\tdefer mgr.Stop()\n\n\t\texec.Reset()\n\n\t\tctx := types.NewContext(context.Background(), nil)\n\t\terr = mgr.Tick(ctx, time.Now())\n\t\tassert.NoError(t, err)\n\n\t\ttime.Sleep(200 * time.Millisecond)\n\t\tassert.Equal(t, 0, exec.ExecCount(), \"Robot without clock config should not trigger\")\n\t})\n}\n\n// ==================== Test Data Setup Helpers ====================\n\n// setupClockTestRobot creates a robot with specified clock config\nfunc setupClockTestRobot(t *testing.T, memberID, teamID string, clockConfig map[string]interface{}) {\n\tsetupClockTestRobotWithQuota(t, memberID, teamID, clockConfig, 3, 20, 5)\n}\n\n// setupClockTestRobotWithQuota creates a robot with specified clock config and quota\nfunc setupClockTestRobotWithQuota(t *testing.T, memberID, teamID string, clockConfig map[string]interface{}, max, queue, priority int) {\n\tm := model.Select(\"__yao.member\")\n\ttableName := m.MetaData.Table.Name\n\tqb := capsule.Query()\n\n\trobotConfig := map[string]interface{}{\n\t\t\"identity\": map[string]interface{}{\n\t\t\t\"role\": \"Clock Test Robot \" + memberID,\n\t\t},\n\t\t\"quota\": map[string]interface{}{\n\t\t\t\"max\": max,\n\t\t\t\"queue\": queue,\n\t\t\t\"priority\": priority,\n\t\t},\n\t\t\"triggers\": map[string]interface{}{\n\t\t\t\"clock\": map[string]interface{}{\"enabled\": true},\n\t\t},\n\t\t\"clock\": clockConfig,\n\t}\n\tconfigJSON, _ := json.Marshal(robotConfig)\n\n\terr := qb.Table(tableName).Insert([]map[string]interface{}{\n\t\t{\n\t\t\t\"member_id\": memberID,\n\t\t\t\"team_id\": teamID,\n\t\t\t\"member_type\": \"robot\",\n\t\t\t\"display_name\": \"Clock Test Robot \" + memberID,\n\t\t\t\"status\": \"active\",\n\t\t\t\"role_id\": \"member\",\n\t\t\t\"autonomous_mode\": true,\n\t\t\t\"robot_status\": \"idle\",\n\t\t\t\"robot_config\": string(configJSON),\n\t\t},\n\t})\n\tif err != nil {\n\t\tt.Fatalf(\"Failed to insert %s: %v\", memberID, err)\n\t}\n}\n" }, { "path": "agent/robot/manager/integration_concurrent_test.go", "content": "package manager_test\n\n// Integration tests for concurrent execution and quota enforcement\n// Tests the two-level concurrency model:\n// 1. Global pool limit (worker count)\n// 2. Per-robot quota limit (Quota.Max, Quota.Queue)\n\nimport (\n\t\"context\"\n\t\"encoding/json\"\n\t\"fmt\"\n\t\"sync\"\n\t\"sync/atomic\"\n\t\"testing\"\n\t\"time\"\n\n\t\"github.com/stretchr/testify/assert\"\n\t\"github.com/stretchr/testify/require\"\n\t\"github.com/yaoapp/gou/model\"\n\t\"github.com/yaoapp/xun/capsule\"\n\t\"github.com/yaoapp/yao/agent/robot/executor\"\n\t\"github.com/yaoapp/yao/agent/robot/manager\"\n\t\"github.com/yaoapp/yao/agent/robot/pool\"\n\t\"github.com/yaoapp/yao/agent/robot/types\"\n\t\"github.com/yaoapp/yao/agent/testutils\"\n)\n\n// ==================== Concurrent Execution Tests ====================\n\n// TestIntegrationConcurrentExecution tests concurrent execution of multiple robots\nfunc TestIntegrationConcurrentExecution(t *testing.T) {\n\tif testing.Short() {\n\t\tt.Skip(\"Skipping integration test\")\n\t}\n\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\tcleanupIntegrationRobots(t)\n\tdefer cleanupIntegrationRobots(t)\n\n\tt.Run(\"multiple robots execute concurrently\", func(t *testing.T) {\n\t\t// Create 5 robots\n\t\tfor i := 0; i < 5; i++ {\n\t\t\tmemberID := \"robot_integ_conc_multi_\" + string(rune('A'+i))\n\t\t\tsetupConcurrentTestRobot(t, memberID, \"team_integ_conc\", 3, 20)\n\t\t}\n\n\t\t// Track concurrent execution count\n\t\tvar maxConcurrent int32\n\t\tvar currentConcurrent int32\n\n\t\texec := executor.NewDryRunWithCallbacks(100*time.Millisecond,\n\t\t\tfunc() {\n\t\t\t\tcurr := atomic.AddInt32(¤tConcurrent, 1)\n\t\t\t\tfor {\n\t\t\t\t\told := atomic.LoadInt32(&maxConcurrent)\n\t\t\t\t\tif curr <= old || atomic.CompareAndSwapInt32(&maxConcurrent, old, curr) {\n\t\t\t\t\t\tbreak\n\t\t\t\t\t}\n\t\t\t\t}\n\t\t\t},\n\t\t\tfunc() {\n\t\t\t\tatomic.AddInt32(¤tConcurrent, -1)\n\t\t\t},\n\t\t)\n\n\t\tconfig := &manager.Config{\n\t\t\tTickInterval: 100 * time.Millisecond,\n\t\t\tPoolConfig: &pool.Config{WorkerSize: 5, QueueSize: 50},\n\t\t}\n\t\tm := manager.NewWithConfig(config)\n\t\tm.Pool().SetExecutor(exec)\n\n\t\terr := m.Start()\n\t\trequire.NoError(t, err)\n\t\tdefer m.Stop()\n\n\t\t// Verify robots are loaded into cache\n\t\tfor i := 0; i < 5; i++ {\n\t\t\tmemberID := \"robot_integ_conc_multi_\" + string(rune('A'+i))\n\t\t\trobot := m.Cache().Get(memberID)\n\t\t\trequire.NotNil(t, robot, \"Robot %s should be loaded into cache\", memberID)\n\t\t}\n\n\t\tctx := types.NewContext(context.Background(), nil)\n\n\t\t// Trigger all robots simultaneously\n\t\tvar wg sync.WaitGroup\n\t\tfor i := 0; i < 5; i++ {\n\t\t\twg.Add(1)\n\t\t\tmemberID := \"robot_integ_conc_multi_\" + string(rune('A'+i))\n\t\t\tgo func(id string) {\n\t\t\t\tdefer wg.Done()\n\t\t\t\tm.TriggerManual(ctx, id, types.TriggerClock, nil)\n\t\t\t}(memberID)\n\t\t}\n\n\t\twg.Wait()\n\n\t\t// Wait for all executions\n\t\ttime.Sleep(500 * time.Millisecond)\n\n\t\t// Should have achieved concurrent execution\n\t\tassert.GreaterOrEqual(t, int(maxConcurrent), 2, \"Should achieve concurrent execution\")\n\t\tassert.GreaterOrEqual(t, exec.ExecCount(), 5, \"All robots should execute\")\n\t})\n\n\tt.Run(\"same robot multiple triggers\", func(t *testing.T) {\n\t\tsetupConcurrentTestRobot(t, \"robot_integ_conc_same\", \"team_integ_conc\", 3, 20)\n\n\t\texec := executor.NewDryRunWithDelay(50 * time.Millisecond)\n\n\t\tconfig := &manager.Config{\n\t\t\tTickInterval: 100 * time.Millisecond,\n\t\t\tPoolConfig: &pool.Config{WorkerSize: 5, QueueSize: 50},\n\t\t}\n\t\tm := manager.NewWithConfig(config)\n\t\tm.Pool().SetExecutor(exec)\n\n\t\terr := m.Start()\n\t\trequire.NoError(t, err)\n\t\tdefer m.Stop()\n\n\t\tctx := types.NewContext(context.Background(), nil)\n\n\t\t// Trigger same robot multiple times\n\t\tfor i := 0; i < 5; i++ {\n\t\t\t_, err := m.TriggerManual(ctx, \"robot_integ_conc_same\", types.TriggerClock, nil)\n\t\t\tassert.NoError(t, err)\n\t\t}\n\n\t\t// Wait for all executions\n\t\ttime.Sleep(800 * time.Millisecond)\n\n\t\t// All 5 should eventually execute\n\t\tassert.GreaterOrEqual(t, exec.ExecCount(), 5, \"All triggers should execute\")\n\t})\n}\n\n// ==================== Quota Enforcement Tests ====================\n\n// TestIntegrationQuotaEnforcement tests per-robot quota limits\nfunc TestIntegrationQuotaEnforcement(t *testing.T) {\n\tif testing.Short() {\n\t\tt.Skip(\"Skipping integration test\")\n\t}\n\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\tcleanupIntegrationRobots(t)\n\tdefer cleanupIntegrationRobots(t)\n\n\tt.Run(\"respects Quota.Max limit\", func(t *testing.T) {\n\t\t// Create robot with Max=2\n\t\tsetupConcurrentTestRobot(t, \"robot_integ_quota_max\", \"team_integ_quota\", 2, 20)\n\n\t\t// Track max concurrent for this robot\n\t\tvar maxConcurrent int32\n\t\tvar currentConcurrent int32\n\n\t\texec := executor.NewDryRunWithCallbacks(200*time.Millisecond,\n\t\t\tfunc() {\n\t\t\t\tcurr := atomic.AddInt32(¤tConcurrent, 1)\n\t\t\t\tfor {\n\t\t\t\t\told := atomic.LoadInt32(&maxConcurrent)\n\t\t\t\t\tif curr <= old || atomic.CompareAndSwapInt32(&maxConcurrent, old, curr) {\n\t\t\t\t\t\tbreak\n\t\t\t\t\t}\n\t\t\t\t}\n\t\t\t},\n\t\t\tfunc() {\n\t\t\t\tatomic.AddInt32(¤tConcurrent, -1)\n\t\t\t},\n\t\t)\n\n\t\tconfig := &manager.Config{\n\t\t\tTickInterval: 100 * time.Millisecond,\n\t\t\tPoolConfig: &pool.Config{WorkerSize: 10, QueueSize: 50}, // Many workers\n\t\t}\n\t\tm := manager.NewWithConfig(config)\n\t\tm.Pool().SetExecutor(exec)\n\n\t\terr := m.Start()\n\t\trequire.NoError(t, err)\n\t\tdefer m.Stop()\n\n\t\tctx := types.NewContext(context.Background(), nil)\n\n\t\t// Submit 10 jobs for the same robot\n\t\tfor i := 0; i < 10; i++ {\n\t\t\tm.TriggerManual(ctx, \"robot_integ_quota_max\", types.TriggerClock, nil)\n\t\t}\n\n\t\t// Wait a bit for concurrent execution\n\t\ttime.Sleep(300 * time.Millisecond)\n\n\t\t// Max concurrent should not exceed Quota.Max (2)\n\t\tassert.LessOrEqual(t, int(maxConcurrent), 2, \"Should not exceed Quota.Max\")\n\n\t\t// Wait for all to complete\n\t\ttime.Sleep(1500 * time.Millisecond)\n\n\t\t// All should eventually execute\n\t\tassert.GreaterOrEqual(t, exec.ExecCount(), 10, \"All jobs should eventually execute\")\n\t})\n\n\tt.Run(\"respects Quota.Queue limit\", func(t *testing.T) {\n\t\t// Create robot with Max=1, Queue=3\n\t\tsetupConcurrentTestRobot(t, \"robot_integ_quota_queue\", \"team_integ_quota\", 1, 3)\n\n\t\texec := executor.NewDryRunWithDelay(300 * time.Millisecond) // Slow execution\n\n\t\tconfig := &manager.Config{\n\t\t\tTickInterval: 100 * time.Millisecond,\n\t\t\tPoolConfig: &pool.Config{WorkerSize: 10, QueueSize: 100},\n\t\t}\n\t\tm := manager.NewWithConfig(config)\n\t\tm.Pool().SetExecutor(exec)\n\n\t\terr := m.Start()\n\t\trequire.NoError(t, err)\n\t\tdefer m.Stop()\n\n\t\tctx := types.NewContext(context.Background(), nil)\n\n\t\t// Submit many jobs - some should be rejected due to queue limit\n\t\tsuccessCount := 0\n\t\tfor i := 0; i < 20; i++ {\n\t\t\t_, err := m.TriggerManual(ctx, \"robot_integ_quota_queue\", types.TriggerClock, nil)\n\t\t\tif err == nil {\n\t\t\t\tsuccessCount++\n\t\t\t}\n\t\t}\n\n\t\t// Should accept at most Max + Queue = 1 + 3 = 4 jobs\n\t\tassert.LessOrEqual(t, successCount, 4, \"Should respect queue limit\")\n\t\tassert.GreaterOrEqual(t, successCount, 1, \"Should accept at least 1 job\")\n\t})\n\n\tt.Run(\"different robots have independent quotas\", func(t *testing.T) {\n\t\t// Robot A: Max=1\n\t\tsetupConcurrentTestRobot(t, \"robot_integ_quota_A\", \"team_integ_quota\", 1, 10)\n\t\t// Robot B: Max=3\n\t\tsetupConcurrentTestRobot(t, \"robot_integ_quota_B\", \"team_integ_quota\", 3, 10)\n\n\t\tvar concurrentA int32\n\t\tvar concurrentB int32\n\t\tvar maxA int32\n\t\tvar maxB int32\n\n\t\t// Custom executor that tracks per-robot concurrency\n\t\texec := &trackingExecutor{\n\t\t\tdelay: 150 * time.Millisecond,\n\t\t\tonStart: func(robot *types.Robot) {\n\t\t\t\tif robot.MemberID == \"robot_integ_quota_A\" {\n\t\t\t\t\tcurr := atomic.AddInt32(&concurrentA, 1)\n\t\t\t\t\tfor {\n\t\t\t\t\t\told := atomic.LoadInt32(&maxA)\n\t\t\t\t\t\tif curr <= old || atomic.CompareAndSwapInt32(&maxA, old, curr) {\n\t\t\t\t\t\t\tbreak\n\t\t\t\t\t\t}\n\t\t\t\t\t}\n\t\t\t\t} else {\n\t\t\t\t\tcurr := atomic.AddInt32(&concurrentB, 1)\n\t\t\t\t\tfor {\n\t\t\t\t\t\told := atomic.LoadInt32(&maxB)\n\t\t\t\t\t\tif curr <= old || atomic.CompareAndSwapInt32(&maxB, old, curr) {\n\t\t\t\t\t\t\tbreak\n\t\t\t\t\t\t}\n\t\t\t\t\t}\n\t\t\t\t}\n\t\t\t},\n\t\t\tonEnd: func(robot *types.Robot) {\n\t\t\t\tif robot.MemberID == \"robot_integ_quota_A\" {\n\t\t\t\t\tatomic.AddInt32(&concurrentA, -1)\n\t\t\t\t} else {\n\t\t\t\t\tatomic.AddInt32(&concurrentB, -1)\n\t\t\t\t}\n\t\t\t},\n\t\t}\n\n\t\tconfig := &manager.Config{\n\t\t\tTickInterval: 100 * time.Millisecond,\n\t\t\tPoolConfig: &pool.Config{WorkerSize: 10, QueueSize: 50},\n\t\t}\n\t\tm := manager.NewWithConfig(config)\n\t\tm.Pool().SetExecutor(exec)\n\n\t\terr := m.Start()\n\t\trequire.NoError(t, err)\n\t\tdefer m.Stop()\n\n\t\tctx := types.NewContext(context.Background(), nil)\n\n\t\t// Submit 5 jobs for each robot\n\t\tfor i := 0; i < 5; i++ {\n\t\t\tm.TriggerManual(ctx, \"robot_integ_quota_A\", types.TriggerClock, nil)\n\t\t\tm.TriggerManual(ctx, \"robot_integ_quota_B\", types.TriggerClock, nil)\n\t\t}\n\n\t\t// Wait a bit\n\t\ttime.Sleep(300 * time.Millisecond)\n\n\t\t// Robot A should have max 1 concurrent\n\t\tassert.LessOrEqual(t, int(maxA), 1, \"Robot A should respect its quota\")\n\t\t// Robot B should have max 3 concurrent\n\t\tassert.LessOrEqual(t, int(maxB), 3, \"Robot B should respect its quota\")\n\n\t\t// Wait for completion\n\t\ttime.Sleep(1 * time.Second)\n\t})\n}\n\n// ==================== Global Pool Limit Tests ====================\n\n// TestIntegrationGlobalPoolLimit tests global worker pool limits\nfunc TestIntegrationGlobalPoolLimit(t *testing.T) {\n\tif testing.Short() {\n\t\tt.Skip(\"Skipping integration test\")\n\t}\n\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\tcleanupIntegrationRobots(t)\n\tdefer cleanupIntegrationRobots(t)\n\n\tt.Run(\"respects global worker limit\", func(t *testing.T) {\n\t\t// Create 10 robots with high quotas\n\t\tfor i := 0; i < 10; i++ {\n\t\t\tmemberID := \"robot_integ_pool_limit_\" + string(rune('A'+i))\n\t\t\tsetupConcurrentTestRobot(t, memberID, \"team_integ_pool\", 5, 20)\n\t\t}\n\n\t\tvar maxConcurrent int32\n\t\tvar currentConcurrent int32\n\n\t\texec := executor.NewDryRunWithCallbacks(200*time.Millisecond,\n\t\t\tfunc() {\n\t\t\t\tcurr := atomic.AddInt32(¤tConcurrent, 1)\n\t\t\t\tfor {\n\t\t\t\t\told := atomic.LoadInt32(&maxConcurrent)\n\t\t\t\t\tif curr <= old || atomic.CompareAndSwapInt32(&maxConcurrent, old, curr) {\n\t\t\t\t\t\tbreak\n\t\t\t\t\t}\n\t\t\t\t}\n\t\t\t},\n\t\t\tfunc() {\n\t\t\t\tatomic.AddInt32(¤tConcurrent, -1)\n\t\t\t},\n\t\t)\n\n\t\tconfig := &manager.Config{\n\t\t\tTickInterval: 100 * time.Millisecond,\n\t\t\tPoolConfig: &pool.Config{WorkerSize: 3, QueueSize: 100}, // Only 3 workers\n\t\t}\n\t\tm := manager.NewWithConfig(config)\n\t\tm.Pool().SetExecutor(exec)\n\n\t\terr := m.Start()\n\t\trequire.NoError(t, err)\n\t\tdefer m.Stop()\n\n\t\tctx := types.NewContext(context.Background(), nil)\n\n\t\t// Trigger all 10 robots\n\t\tfor i := 0; i < 10; i++ {\n\t\t\tmemberID := \"robot_integ_pool_limit_\" + string(rune('A'+i))\n\t\t\tm.TriggerManual(ctx, memberID, types.TriggerClock, nil)\n\t\t}\n\n\t\t// Wait a bit\n\t\ttime.Sleep(300 * time.Millisecond)\n\n\t\t// Max concurrent should not exceed worker limit (3)\n\t\tassert.LessOrEqual(t, int(maxConcurrent), 3, \"Should not exceed worker limit\")\n\n\t\t// Wait for all to complete\n\t\ttime.Sleep(1 * time.Second)\n\n\t\t// All 10 should execute\n\t\tassert.GreaterOrEqual(t, exec.ExecCount(), 10, \"All robots should execute\")\n\t})\n\n\tt.Run(\"respects global queue limit\", func(t *testing.T) {\n\t\t// Create robots\n\t\tfor i := 0; i < 20; i++ {\n\t\t\tmemberID := \"robot_integ_pool_queue_\" + string(rune('A'+i%26))\n\t\t\tsetupConcurrentTestRobot(t, memberID, \"team_integ_pool\", 5, 20)\n\t\t}\n\n\t\texec := executor.NewDryRunWithDelay(500 * time.Millisecond) // Slow execution\n\n\t\tconfig := &manager.Config{\n\t\t\tTickInterval: 100 * time.Millisecond,\n\t\t\tPoolConfig: &pool.Config{WorkerSize: 1, QueueSize: 5}, // Small queue\n\t\t}\n\t\tm := manager.NewWithConfig(config)\n\t\tm.Pool().SetExecutor(exec)\n\n\t\terr := m.Start()\n\t\trequire.NoError(t, err)\n\t\tdefer m.Stop()\n\n\t\tctx := types.NewContext(context.Background(), nil)\n\n\t\t// Try to submit many jobs\n\t\tsuccessCount := 0\n\t\tfor i := 0; i < 20; i++ {\n\t\t\tmemberID := \"robot_integ_pool_queue_\" + string(rune('A'+i%26))\n\t\t\t_, err := m.TriggerManual(ctx, memberID, types.TriggerClock, nil)\n\t\t\tif err == nil {\n\t\t\t\tsuccessCount++\n\t\t\t}\n\t\t}\n\n\t\t// Should respect global queue limit\n\t\t// Max = WorkerSize + QueueSize = 1 + 5 = 6\n\t\tassert.LessOrEqual(t, successCount, 6, \"Should respect global queue limit\")\n\t})\n}\n\n// ==================== Priority Tests ====================\n\n// TestIntegrationPriorityExecution tests priority-based execution order\nfunc TestIntegrationPriorityExecution(t *testing.T) {\n\tif testing.Short() {\n\t\tt.Skip(\"Skipping integration test\")\n\t}\n\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\tcleanupIntegrationRobots(t)\n\tdefer cleanupIntegrationRobots(t)\n\n\tt.Run(\"higher priority executes first\", func(t *testing.T) {\n\t\t// Create robots with different priorities\n\t\tsetupConcurrentTestRobotWithPriority(t, \"robot_integ_prio_low\", \"team_integ_prio\", 2, 10, 1)\n\t\tsetupConcurrentTestRobotWithPriority(t, \"robot_integ_prio_med\", \"team_integ_prio\", 2, 10, 5)\n\t\tsetupConcurrentTestRobotWithPriority(t, \"robot_integ_prio_high\", \"team_integ_prio\", 2, 10, 10)\n\n\t\texecutionOrder := make([]string, 0)\n\t\tvar mu sync.Mutex\n\n\t\texec := &trackingExecutor{\n\t\t\tdelay: 50 * time.Millisecond,\n\t\t\tonStart: func(robot *types.Robot) {\n\t\t\t\tmu.Lock()\n\t\t\t\texecutionOrder = append(executionOrder, robot.MemberID)\n\t\t\t\tmu.Unlock()\n\t\t\t},\n\t\t\tonEnd: func(robot *types.Robot) {},\n\t\t}\n\n\t\tconfig := &manager.Config{\n\t\t\tTickInterval: 100 * time.Millisecond,\n\t\t\tPoolConfig: &pool.Config{WorkerSize: 1, QueueSize: 50}, // Single worker for ordering\n\t\t}\n\t\tm := manager.NewWithConfig(config)\n\t\tm.Pool().SetExecutor(exec)\n\n\t\terr := m.Start()\n\t\trequire.NoError(t, err)\n\t\tdefer m.Stop()\n\n\t\tctx := types.NewContext(context.Background(), nil)\n\n\t\t// Submit in low-to-high priority order\n\t\t_, err = m.TriggerManual(ctx, \"robot_integ_prio_low\", types.TriggerClock, nil)\n\t\tassert.NoError(t, err)\n\t\t_, err = m.TriggerManual(ctx, \"robot_integ_prio_med\", types.TriggerClock, nil)\n\t\tassert.NoError(t, err)\n\t\t_, err = m.TriggerManual(ctx, \"robot_integ_prio_high\", types.TriggerClock, nil)\n\t\tassert.NoError(t, err)\n\n\t\t// Wait for all to complete\n\t\ttime.Sleep(500 * time.Millisecond)\n\n\t\t// Verify execution order (high priority should be first or early)\n\t\tmu.Lock()\n\t\torder := executionOrder\n\t\tmu.Unlock()\n\n\t\tassert.Len(t, order, 3, \"All 3 robots should execute\")\n\t\t// Note: First job may already be picked up before others are queued\n\t\t// So we just verify all executed\n\t})\n\n\tt.Run(\"human trigger has higher priority than clock\", func(t *testing.T) {\n\t\tsetupConcurrentTestRobotAllTriggers(t, \"robot_integ_prio_trigger\", \"team_integ_prio\", 2, 10, 5)\n\n\t\texecutionOrder := make([]types.TriggerType, 0)\n\t\tvar mu sync.Mutex\n\n\t\texec := &triggerTrackingExecutor{\n\t\t\tdelay: 50 * time.Millisecond,\n\t\t\tonStart: func(trigger types.TriggerType) {\n\t\t\t\tmu.Lock()\n\t\t\t\texecutionOrder = append(executionOrder, trigger)\n\t\t\t\tmu.Unlock()\n\t\t\t},\n\t\t}\n\n\t\tconfig := &manager.Config{\n\t\t\tTickInterval: 100 * time.Millisecond,\n\t\t\tPoolConfig: &pool.Config{WorkerSize: 1, QueueSize: 50},\n\t\t}\n\t\tm := manager.NewWithConfig(config)\n\t\tm.Pool().SetExecutor(exec)\n\n\t\terr := m.Start()\n\t\trequire.NoError(t, err)\n\t\tdefer m.Stop()\n\n\t\tctx := types.NewContext(context.Background(), nil)\n\n\t\t// Submit clock first, then human\n\t\t_, err = m.TriggerManual(ctx, \"robot_integ_prio_trigger\", types.TriggerClock, nil)\n\t\tassert.NoError(t, err)\n\t\t_, err = m.TriggerManual(ctx, \"robot_integ_prio_trigger\", types.TriggerHuman, nil)\n\t\tassert.NoError(t, err)\n\n\t\t// Wait for execution\n\t\ttime.Sleep(300 * time.Millisecond)\n\n\t\tmu.Lock()\n\t\torder := executionOrder\n\t\tmu.Unlock()\n\n\t\tassert.Len(t, order, 2, \"Both triggers should execute\")\n\t})\n}\n\n// ==================== Helper Types ====================\n\n// trackingExecutor tracks execution per robot\ntype trackingExecutor struct {\n\tdelay time.Duration\n\tonStart func(robot *types.Robot)\n\tonEnd func(robot *types.Robot)\n\tcount int32\n}\n\nfunc (e *trackingExecutor) Execute(ctx *types.Context, robot *types.Robot, trigger types.TriggerType, data interface{}) (*types.Execution, error) {\n\treturn e.ExecuteWithControl(ctx, robot, trigger, data, \"\", nil)\n}\n\nfunc (e *trackingExecutor) ExecuteWithID(ctx *types.Context, robot *types.Robot, trigger types.TriggerType, data interface{}, execID string) (*types.Execution, error) {\n\treturn e.ExecuteWithControl(ctx, robot, trigger, data, execID, nil)\n}\n\nfunc (e *trackingExecutor) ExecuteWithControl(ctx *types.Context, robot *types.Robot, trigger types.TriggerType, data interface{}, execID string, control types.ExecutionControl) (*types.Execution, error) {\n\tif robot == nil {\n\t\treturn nil, types.ErrRobotNotFound\n\t}\n\n\t// Use provided execID or generate unique ID for each execution to properly track quota\n\tif execID == \"\" {\n\t\texecID = fmt.Sprintf(\"exec_%d\", time.Now().UnixNano())\n\t}\n\texec := &types.Execution{\n\t\tID: execID,\n\t\tMemberID: robot.MemberID,\n\t\tTeamID: robot.TeamID,\n\t\tTriggerType: trigger,\n\t\tStartTime: time.Now(),\n\t\tStatus: types.ExecPending,\n\t}\n\n\tif !robot.TryAcquireSlot(exec) {\n\t\treturn nil, types.ErrQuotaExceeded\n\t}\n\tdefer robot.RemoveExecution(exec.ID)\n\n\tif e.onStart != nil {\n\t\te.onStart(robot)\n\t}\n\n\texec.Status = types.ExecRunning\n\ttime.Sleep(e.delay)\n\n\tif e.onEnd != nil {\n\t\te.onEnd(robot)\n\t}\n\n\texec.Status = types.ExecCompleted\n\tnow := time.Now()\n\texec.EndTime = &now\n\n\tatomic.AddInt32(&e.count, 1)\n\treturn exec, nil\n}\n\nfunc (e *trackingExecutor) ExecCount() int {\n\treturn int(atomic.LoadInt32(&e.count))\n}\n\nfunc (e *trackingExecutor) CurrentCount() int {\n\treturn 0\n}\n\nfunc (e *trackingExecutor) Resume(ctx *types.Context, execID string, reply string) error {\n\treturn fmt.Errorf(\"resume not supported in tracking executor\")\n}\n\nfunc (e *trackingExecutor) Reset() {\n\tatomic.StoreInt32(&e.count, 0)\n}\n\n// triggerTrackingExecutor tracks execution by trigger type\ntype triggerTrackingExecutor struct {\n\tdelay time.Duration\n\tonStart func(trigger types.TriggerType)\n\tcount int32\n}\n\nfunc (e *triggerTrackingExecutor) Execute(ctx *types.Context, robot *types.Robot, trigger types.TriggerType, data interface{}) (*types.Execution, error) {\n\treturn e.ExecuteWithControl(ctx, robot, trigger, data, \"\", nil)\n}\n\nfunc (e *triggerTrackingExecutor) ExecuteWithID(ctx *types.Context, robot *types.Robot, trigger types.TriggerType, data interface{}, execID string) (*types.Execution, error) {\n\treturn e.ExecuteWithControl(ctx, robot, trigger, data, execID, nil)\n}\n\nfunc (e *triggerTrackingExecutor) ExecuteWithControl(ctx *types.Context, robot *types.Robot, trigger types.TriggerType, data interface{}, execID string, control types.ExecutionControl) (*types.Execution, error) {\n\tif robot == nil {\n\t\treturn nil, types.ErrRobotNotFound\n\t}\n\n\t// Use provided execID or generate unique ID for each execution to properly track quota\n\tif execID == \"\" {\n\t\texecID = fmt.Sprintf(\"exec_trigger_%s_%d\", string(trigger), time.Now().UnixNano())\n\t}\n\texec := &types.Execution{\n\t\tID: execID,\n\t\tMemberID: robot.MemberID,\n\t\tTeamID: robot.TeamID,\n\t\tTriggerType: trigger,\n\t\tStartTime: time.Now(),\n\t\tStatus: types.ExecPending,\n\t}\n\n\tif !robot.TryAcquireSlot(exec) {\n\t\treturn nil, types.ErrQuotaExceeded\n\t}\n\tdefer robot.RemoveExecution(exec.ID)\n\n\tif e.onStart != nil {\n\t\te.onStart(trigger)\n\t}\n\n\texec.Status = types.ExecRunning\n\ttime.Sleep(e.delay)\n\n\texec.Status = types.ExecCompleted\n\tnow := time.Now()\n\texec.EndTime = &now\n\n\tatomic.AddInt32(&e.count, 1)\n\treturn exec, nil\n}\n\nfunc (e *triggerTrackingExecutor) ExecCount() int {\n\treturn int(atomic.LoadInt32(&e.count))\n}\n\nfunc (e *triggerTrackingExecutor) CurrentCount() int {\n\treturn 0\n}\n\nfunc (e *triggerTrackingExecutor) Resume(ctx *types.Context, execID string, reply string) error {\n\treturn fmt.Errorf(\"resume not supported in trigger tracking executor\")\n}\n\nfunc (e *triggerTrackingExecutor) Reset() {\n\tatomic.StoreInt32(&e.count, 0)\n}\n\n// ==================== Test Data Setup Helpers ====================\n\n// setupConcurrentTestRobot creates a robot for concurrency testing\nfunc setupConcurrentTestRobot(t *testing.T, memberID, teamID string, max, queue int) {\n\tsetupConcurrentTestRobotWithPriority(t, memberID, teamID, max, queue, 5)\n}\n\n// setupConcurrentTestRobotWithPriority creates a robot with specified priority\nfunc setupConcurrentTestRobotWithPriority(t *testing.T, memberID, teamID string, max, queue, priority int) {\n\tm := model.Select(\"__yao.member\")\n\ttableName := m.MetaData.Table.Name\n\tqb := capsule.Query()\n\n\trobotConfig := map[string]interface{}{\n\t\t\"identity\": map[string]interface{}{\n\t\t\t\"role\": \"Concurrent Test Robot \" + memberID,\n\t\t},\n\t\t\"quota\": map[string]interface{}{\n\t\t\t\"max\": max,\n\t\t\t\"queue\": queue,\n\t\t\t\"priority\": priority,\n\t\t},\n\t\t\"triggers\": map[string]interface{}{\n\t\t\t\"clock\": map[string]interface{}{\"enabled\": true},\n\t\t\t\"intervene\": map[string]interface{}{\"enabled\": true},\n\t\t\t\"event\": map[string]interface{}{\"enabled\": true},\n\t\t},\n\t\t\"clock\": map[string]interface{}{\n\t\t\t\"mode\": \"times\",\n\t\t\t\"times\": []string{\"09:00\"},\n\t\t\t\"tz\": \"Asia/Shanghai\",\n\t\t},\n\t}\n\tconfigJSON, _ := json.Marshal(robotConfig)\n\n\terr := qb.Table(tableName).Insert([]map[string]interface{}{\n\t\t{\n\t\t\t\"member_id\": memberID,\n\t\t\t\"team_id\": teamID,\n\t\t\t\"member_type\": \"robot\",\n\t\t\t\"display_name\": \"Concurrent Test Robot \" + memberID,\n\t\t\t\"status\": \"active\",\n\t\t\t\"role_id\": \"member\",\n\t\t\t\"autonomous_mode\": true,\n\t\t\t\"robot_status\": \"idle\",\n\t\t\t\"robot_config\": string(configJSON),\n\t\t},\n\t})\n\tif err != nil {\n\t\tt.Fatalf(\"Failed to insert %s: %v\", memberID, err)\n\t}\n}\n\n// setupConcurrentTestRobotAllTriggers creates a robot with all triggers enabled\nfunc setupConcurrentTestRobotAllTriggers(t *testing.T, memberID, teamID string, max, queue, priority int) {\n\tm := model.Select(\"__yao.member\")\n\ttableName := m.MetaData.Table.Name\n\tqb := capsule.Query()\n\n\trobotConfig := map[string]interface{}{\n\t\t\"identity\": map[string]interface{}{\n\t\t\t\"role\": \"All Triggers Test Robot\",\n\t\t},\n\t\t\"quota\": map[string]interface{}{\n\t\t\t\"max\": max,\n\t\t\t\"queue\": queue,\n\t\t\t\"priority\": priority,\n\t\t},\n\t\t\"triggers\": map[string]interface{}{\n\t\t\t\"clock\": map[string]interface{}{\"enabled\": true},\n\t\t\t\"intervene\": map[string]interface{}{\"enabled\": true},\n\t\t\t\"event\": map[string]interface{}{\"enabled\": true},\n\t\t},\n\t}\n\tconfigJSON, _ := json.Marshal(robotConfig)\n\n\terr := qb.Table(tableName).Insert([]map[string]interface{}{\n\t\t{\n\t\t\t\"member_id\": memberID,\n\t\t\t\"team_id\": teamID,\n\t\t\t\"member_type\": \"robot\",\n\t\t\t\"display_name\": \"All Triggers Robot \" + memberID,\n\t\t\t\"status\": \"active\",\n\t\t\t\"role_id\": \"member\",\n\t\t\t\"autonomous_mode\": true,\n\t\t\t\"robot_status\": \"idle\",\n\t\t\t\"robot_config\": string(configJSON),\n\t\t},\n\t})\n\tif err != nil {\n\t\tt.Fatalf(\"Failed to insert %s: %v\", memberID, err)\n\t}\n}\n" }, { "path": "agent/robot/manager/integration_control_test.go", "content": "package manager_test\n\n// Integration tests for execution control (Pause/Resume/Stop)\n// Tests Manager's execution control methods and ExecutionController\n\nimport (\n\t\"context\"\n\t\"encoding/json\"\n\t\"fmt\"\n\t\"sync\"\n\t\"sync/atomic\"\n\t\"testing\"\n\t\"time\"\n\n\t\"github.com/stretchr/testify/assert\"\n\t\"github.com/stretchr/testify/require\"\n\t\"github.com/yaoapp/gou/model\"\n\t\"github.com/yaoapp/xun/capsule\"\n\tagentcontext \"github.com/yaoapp/yao/agent/context\"\n\t\"github.com/yaoapp/yao/agent/robot/manager\"\n\t\"github.com/yaoapp/yao/agent/robot/pool\"\n\t\"github.com/yaoapp/yao/agent/robot/types\"\n\t\"github.com/yaoapp/yao/agent/testutils\"\n)\n\n// ==================== Pause/Resume Tests ====================\n\n// TestIntegrationExecutionPauseResume tests pausing and resuming executions\nfunc TestIntegrationExecutionPauseResume(t *testing.T) {\n\tif testing.Short() {\n\t\tt.Skip(\"Skipping integration test\")\n\t}\n\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\tcleanupIntegrationRobots(t)\n\tdefer cleanupIntegrationRobots(t)\n\n\tt.Run(\"pause and resume execution\", func(t *testing.T) {\n\t\tsetupControlTestRobot(t, \"robot_integ_ctrl_pause\", \"team_integ_ctrl\")\n\n\t\t// Use slow executor to have time to pause\n\t\texec := &slowExecutor{delay: 500 * time.Millisecond}\n\n\t\tconfig := &manager.Config{\n\t\t\tTickInterval: 100 * time.Millisecond,\n\t\t\tPoolConfig: &pool.Config{WorkerSize: 3, QueueSize: 20},\n\t\t}\n\t\tm := manager.NewWithConfig(config)\n\t\tm.Pool().SetExecutor(exec)\n\n\t\terr := m.Start()\n\t\trequire.NoError(t, err)\n\t\tdefer m.Stop()\n\n\t\t// Verify robot is loaded into cache\n\t\trobot := m.Cache().Get(\"robot_integ_ctrl_pause\")\n\t\trequire.NotNil(t, robot, \"Robot should be loaded into cache\")\n\n\t\tctx := types.NewContext(context.Background(), nil)\n\n\t\t// Trigger execution\n\t\treq := &types.InterveneRequest{\n\t\t\tMemberID: \"robot_integ_ctrl_pause\",\n\t\t\tAction: types.ActionTaskAdd,\n\t\t\tMessages: []agentcontext.Message{\n\t\t\t\t{Role: agentcontext.RoleUser, Content: \"Test task\"},\n\t\t\t},\n\t\t}\n\t\tresult, err := m.Intervene(ctx, req)\n\t\trequire.NoError(t, err)\n\t\texecID := result.ExecutionID\n\n\t\t// Wait for execution to be tracked\n\t\ttime.Sleep(100 * time.Millisecond)\n\n\t\t// Pause execution\n\t\terr = m.PauseExecution(ctx, execID)\n\t\tassert.NoError(t, err)\n\n\t\t// Verify paused\n\t\tstatus, err := m.GetExecutionStatus(execID)\n\t\tassert.NoError(t, err)\n\t\tassert.True(t, status.IsPaused(), \"Execution should be paused\")\n\n\t\t// Resume execution\n\t\terr = m.ResumeExecution(ctx, execID)\n\t\tassert.NoError(t, err)\n\n\t\t// Verify resumed\n\t\tstatus, err = m.GetExecutionStatus(execID)\n\t\tassert.NoError(t, err)\n\t\tassert.False(t, status.IsPaused(), \"Execution should be resumed\")\n\t})\n\n\tt.Run(\"pause non-existent execution\", func(t *testing.T) {\n\t\tm := manager.New()\n\t\terr := m.Start()\n\t\trequire.NoError(t, err)\n\t\tdefer m.Stop()\n\n\t\tctx := types.NewContext(context.Background(), nil)\n\n\t\terr = m.PauseExecution(ctx, \"nonexistent_exec\")\n\t\tassert.Error(t, err)\n\t\tassert.Contains(t, err.Error(), \"not found\")\n\t})\n\n\tt.Run(\"resume non-paused execution\", func(t *testing.T) {\n\t\tsetupControlTestRobot(t, \"robot_integ_ctrl_resume\", \"team_integ_ctrl\")\n\n\t\texec := &slowExecutor{delay: 500 * time.Millisecond}\n\n\t\tconfig := &manager.Config{\n\t\t\tTickInterval: 100 * time.Millisecond,\n\t\t\tPoolConfig: &pool.Config{WorkerSize: 3, QueueSize: 20},\n\t\t}\n\t\tm := manager.NewWithConfig(config)\n\t\tm.Pool().SetExecutor(exec)\n\n\t\terr := m.Start()\n\t\trequire.NoError(t, err)\n\t\tdefer m.Stop()\n\n\t\tctx := types.NewContext(context.Background(), nil)\n\n\t\t// Trigger execution\n\t\treq := &types.InterveneRequest{\n\t\t\tMemberID: \"robot_integ_ctrl_resume\",\n\t\t\tAction: types.ActionTaskAdd,\n\t\t\tMessages: []agentcontext.Message{\n\t\t\t\t{Role: agentcontext.RoleUser, Content: \"Test task\"},\n\t\t\t},\n\t\t}\n\t\tresult, err := m.Intervene(ctx, req)\n\t\trequire.NoError(t, err)\n\t\texecID := result.ExecutionID\n\n\t\t// Wait for execution to be tracked\n\t\ttime.Sleep(100 * time.Millisecond)\n\n\t\t// Resume without pausing first - should be safe\n\t\terr = m.ResumeExecution(ctx, execID)\n\t\t// May or may not error depending on implementation\n\t\t// The important thing is it doesn't panic\n\t})\n}\n\n// ==================== Stop Tests ====================\n\n// TestIntegrationExecutionStop tests stopping executions\nfunc TestIntegrationExecutionStop(t *testing.T) {\n\tif testing.Short() {\n\t\tt.Skip(\"Skipping integration test\")\n\t}\n\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\tcleanupIntegrationRobots(t)\n\tdefer cleanupIntegrationRobots(t)\n\n\tt.Run(\"stop execution\", func(t *testing.T) {\n\t\tsetupControlTestRobot(t, \"robot_integ_ctrl_stop\", \"team_integ_ctrl\")\n\n\t\texec := &slowExecutor{delay: 1 * time.Second}\n\n\t\tconfig := &manager.Config{\n\t\t\tTickInterval: 100 * time.Millisecond,\n\t\t\tPoolConfig: &pool.Config{WorkerSize: 3, QueueSize: 20},\n\t\t}\n\t\tm := manager.NewWithConfig(config)\n\t\tm.Pool().SetExecutor(exec)\n\n\t\terr := m.Start()\n\t\trequire.NoError(t, err)\n\t\tdefer m.Stop()\n\n\t\tctx := types.NewContext(context.Background(), nil)\n\n\t\t// Trigger execution\n\t\treq := &types.InterveneRequest{\n\t\t\tMemberID: \"robot_integ_ctrl_stop\",\n\t\t\tAction: types.ActionTaskAdd,\n\t\t\tMessages: []agentcontext.Message{\n\t\t\t\t{Role: agentcontext.RoleUser, Content: \"Test task\"},\n\t\t\t},\n\t\t}\n\t\tresult, err := m.Intervene(ctx, req)\n\t\trequire.NoError(t, err)\n\t\texecID := result.ExecutionID\n\n\t\t// Wait for execution to be tracked\n\t\ttime.Sleep(100 * time.Millisecond)\n\n\t\t// Stop execution\n\t\terr = m.StopExecution(ctx, execID)\n\t\tassert.NoError(t, err)\n\n\t\t// Execution should be removed from tracking\n\t\t_, err = m.GetExecutionStatus(execID)\n\t\tassert.Error(t, err)\n\t\tassert.Contains(t, err.Error(), \"not found\")\n\t})\n\n\tt.Run(\"stop non-existent execution\", func(t *testing.T) {\n\t\tm := manager.New()\n\t\terr := m.Start()\n\t\trequire.NoError(t, err)\n\t\tdefer m.Stop()\n\n\t\tctx := types.NewContext(context.Background(), nil)\n\n\t\terr = m.StopExecution(ctx, \"nonexistent_exec\")\n\t\tassert.Error(t, err)\n\t\tassert.Contains(t, err.Error(), \"not found\")\n\t})\n}\n\n// ==================== List Executions Tests ====================\n\n// TestIntegrationListExecutions tests listing executions\nfunc TestIntegrationListExecutions(t *testing.T) {\n\tif testing.Short() {\n\t\tt.Skip(\"Skipping integration test\")\n\t}\n\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\tcleanupIntegrationRobots(t)\n\tdefer cleanupIntegrationRobots(t)\n\n\tt.Run(\"list all executions\", func(t *testing.T) {\n\t\tsetupControlTestRobot(t, \"robot_integ_ctrl_list1\", \"team_integ_ctrl\")\n\t\tsetupControlTestRobot(t, \"robot_integ_ctrl_list2\", \"team_integ_ctrl\")\n\n\t\texec := &slowExecutor{delay: 500 * time.Millisecond}\n\n\t\tconfig := &manager.Config{\n\t\t\tTickInterval: 100 * time.Millisecond,\n\t\t\tPoolConfig: &pool.Config{WorkerSize: 5, QueueSize: 20},\n\t\t}\n\t\tm := manager.NewWithConfig(config)\n\t\tm.Pool().SetExecutor(exec)\n\n\t\terr := m.Start()\n\t\trequire.NoError(t, err)\n\t\tdefer m.Stop()\n\n\t\tctx := types.NewContext(context.Background(), nil)\n\n\t\t// Trigger multiple executions\n\t\texecIDs := make([]string, 0)\n\t\tfor _, memberID := range []string{\"robot_integ_ctrl_list1\", \"robot_integ_ctrl_list2\"} {\n\t\t\treq := &types.InterveneRequest{\n\t\t\t\tMemberID: memberID,\n\t\t\t\tAction: types.ActionTaskAdd,\n\t\t\t\tMessages: []agentcontext.Message{\n\t\t\t\t\t{Role: agentcontext.RoleUser, Content: \"Test task\"},\n\t\t\t\t},\n\t\t\t}\n\t\t\tresult, err := m.Intervene(ctx, req)\n\t\t\trequire.NoError(t, err)\n\t\t\texecIDs = append(execIDs, result.ExecutionID)\n\t\t}\n\n\t\t// Wait for executions to be tracked\n\t\ttime.Sleep(100 * time.Millisecond)\n\n\t\t// List all executions\n\t\texecs := m.ListExecutions()\n\t\tassert.GreaterOrEqual(t, len(execs), 2, \"Should have at least 2 executions\")\n\n\t\t// Verify our executions are in the list\n\t\tfoundCount := 0\n\t\tfor _, e := range execs {\n\t\t\tfor _, id := range execIDs {\n\t\t\t\tif e.ID == id {\n\t\t\t\t\tfoundCount++\n\t\t\t\t}\n\t\t\t}\n\t\t}\n\t\tassert.Equal(t, 2, foundCount, \"Both executions should be in list\")\n\t})\n\n\tt.Run(\"list executions by member\", func(t *testing.T) {\n\t\tsetupControlTestRobot(t, \"robot_integ_ctrl_member1\", \"team_integ_ctrl\")\n\t\tsetupControlTestRobot(t, \"robot_integ_ctrl_member2\", \"team_integ_ctrl\")\n\n\t\texec := &slowExecutor{delay: 500 * time.Millisecond}\n\n\t\tconfig := &manager.Config{\n\t\t\tTickInterval: 100 * time.Millisecond,\n\t\t\tPoolConfig: &pool.Config{WorkerSize: 5, QueueSize: 20},\n\t\t}\n\t\tm := manager.NewWithConfig(config)\n\t\tm.Pool().SetExecutor(exec)\n\n\t\terr := m.Start()\n\t\trequire.NoError(t, err)\n\t\tdefer m.Stop()\n\n\t\tctx := types.NewContext(context.Background(), nil)\n\n\t\t// Trigger 3 executions for robot 1\n\t\tfor i := 0; i < 3; i++ {\n\t\t\treq := &types.InterveneRequest{\n\t\t\t\tMemberID: \"robot_integ_ctrl_member1\",\n\t\t\t\tAction: types.ActionTaskAdd,\n\t\t\t\tMessages: []agentcontext.Message{\n\t\t\t\t\t{Role: agentcontext.RoleUser, Content: \"Test task\"},\n\t\t\t\t},\n\t\t\t}\n\t\t\t_, err := m.Intervene(ctx, req)\n\t\t\trequire.NoError(t, err)\n\t\t}\n\n\t\t// Trigger 2 executions for robot 2\n\t\tfor i := 0; i < 2; i++ {\n\t\t\treq := &types.InterveneRequest{\n\t\t\t\tMemberID: \"robot_integ_ctrl_member2\",\n\t\t\t\tAction: types.ActionTaskAdd,\n\t\t\t\tMessages: []agentcontext.Message{\n\t\t\t\t\t{Role: agentcontext.RoleUser, Content: \"Test task\"},\n\t\t\t\t},\n\t\t\t}\n\t\t\t_, err := m.Intervene(ctx, req)\n\t\t\trequire.NoError(t, err)\n\t\t}\n\n\t\t// Wait for executions to be tracked\n\t\ttime.Sleep(100 * time.Millisecond)\n\n\t\t// List executions for robot 1\n\t\texecs1 := m.ListExecutionsByMember(\"robot_integ_ctrl_member1\")\n\t\tassert.GreaterOrEqual(t, len(execs1), 1, \"Robot 1 should have executions\")\n\n\t\t// List executions for robot 2\n\t\texecs2 := m.ListExecutionsByMember(\"robot_integ_ctrl_member2\")\n\t\tassert.GreaterOrEqual(t, len(execs2), 1, \"Robot 2 should have executions\")\n\n\t\t// Verify member IDs\n\t\tfor _, e := range execs1 {\n\t\t\tassert.Equal(t, \"robot_integ_ctrl_member1\", e.MemberID)\n\t\t}\n\t\tfor _, e := range execs2 {\n\t\t\tassert.Equal(t, \"robot_integ_ctrl_member2\", e.MemberID)\n\t\t}\n\t})\n}\n\n// ==================== Multiple Control Operations Tests ====================\n\n// TestIntegrationMultipleControlOperations tests sequences of control operations\nfunc TestIntegrationMultipleControlOperations(t *testing.T) {\n\tif testing.Short() {\n\t\tt.Skip(\"Skipping integration test\")\n\t}\n\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\tcleanupIntegrationRobots(t)\n\tdefer cleanupIntegrationRobots(t)\n\n\tt.Run(\"pause-resume-pause-stop sequence\", func(t *testing.T) {\n\t\tsetupControlTestRobot(t, \"robot_integ_ctrl_seq\", \"team_integ_ctrl\")\n\n\t\texec := &slowExecutor{delay: 2 * time.Second}\n\n\t\tconfig := &manager.Config{\n\t\t\tTickInterval: 100 * time.Millisecond,\n\t\t\tPoolConfig: &pool.Config{WorkerSize: 3, QueueSize: 20},\n\t\t}\n\t\tm := manager.NewWithConfig(config)\n\t\tm.Pool().SetExecutor(exec)\n\n\t\terr := m.Start()\n\t\trequire.NoError(t, err)\n\t\tdefer m.Stop()\n\n\t\tctx := types.NewContext(context.Background(), nil)\n\n\t\t// Trigger execution\n\t\treq := &types.InterveneRequest{\n\t\t\tMemberID: \"robot_integ_ctrl_seq\",\n\t\t\tAction: types.ActionTaskAdd,\n\t\t\tMessages: []agentcontext.Message{\n\t\t\t\t{Role: agentcontext.RoleUser, Content: \"Test task\"},\n\t\t\t},\n\t\t}\n\t\tresult, err := m.Intervene(ctx, req)\n\t\trequire.NoError(t, err)\n\t\texecID := result.ExecutionID\n\n\t\t// Wait for tracking\n\t\ttime.Sleep(100 * time.Millisecond)\n\n\t\t// Pause\n\t\terr = m.PauseExecution(ctx, execID)\n\t\tassert.NoError(t, err)\n\t\tstatus, _ := m.GetExecutionStatus(execID)\n\t\tassert.True(t, status.IsPaused())\n\n\t\t// Resume\n\t\terr = m.ResumeExecution(ctx, execID)\n\t\tassert.NoError(t, err)\n\t\tstatus, _ = m.GetExecutionStatus(execID)\n\t\tassert.False(t, status.IsPaused())\n\n\t\t// Pause again\n\t\terr = m.PauseExecution(ctx, execID)\n\t\tassert.NoError(t, err)\n\t\tstatus, _ = m.GetExecutionStatus(execID)\n\t\tassert.True(t, status.IsPaused())\n\n\t\t// Stop\n\t\terr = m.StopExecution(ctx, execID)\n\t\tassert.NoError(t, err)\n\t\t_, err = m.GetExecutionStatus(execID)\n\t\tassert.Error(t, err) // Should be removed\n\t})\n\n\tt.Run(\"concurrent control operations\", func(t *testing.T) {\n\t\tsetupControlTestRobot(t, \"robot_integ_ctrl_conc\", \"team_integ_ctrl\")\n\n\t\texec := &slowExecutor{delay: 1 * time.Second}\n\n\t\tconfig := &manager.Config{\n\t\t\tTickInterval: 100 * time.Millisecond,\n\t\t\tPoolConfig: &pool.Config{WorkerSize: 3, QueueSize: 20},\n\t\t}\n\t\tm := manager.NewWithConfig(config)\n\t\tm.Pool().SetExecutor(exec)\n\n\t\terr := m.Start()\n\t\trequire.NoError(t, err)\n\t\tdefer m.Stop()\n\n\t\tctx := types.NewContext(context.Background(), nil)\n\n\t\t// Trigger execution\n\t\treq := &types.InterveneRequest{\n\t\t\tMemberID: \"robot_integ_ctrl_conc\",\n\t\t\tAction: types.ActionTaskAdd,\n\t\t\tMessages: []agentcontext.Message{\n\t\t\t\t{Role: agentcontext.RoleUser, Content: \"Test task\"},\n\t\t\t},\n\t\t}\n\t\tresult, err := m.Intervene(ctx, req)\n\t\trequire.NoError(t, err)\n\t\texecID := result.ExecutionID\n\n\t\t// Wait for tracking\n\t\ttime.Sleep(100 * time.Millisecond)\n\n\t\t// Concurrent pause/resume operations should not panic\n\t\tvar wg sync.WaitGroup\n\t\tfor i := 0; i < 10; i++ {\n\t\t\twg.Add(2)\n\t\t\tgo func() {\n\t\t\t\tdefer wg.Done()\n\t\t\t\tm.PauseExecution(ctx, execID)\n\t\t\t}()\n\t\t\tgo func() {\n\t\t\t\tdefer wg.Done()\n\t\t\t\tm.ResumeExecution(ctx, execID)\n\t\t\t}()\n\t\t}\n\n\t\t// Wait with timeout\n\t\tdone := make(chan struct{})\n\t\tgo func() {\n\t\t\twg.Wait()\n\t\t\tclose(done)\n\t\t}()\n\n\t\tselect {\n\t\tcase <-done:\n\t\t\t// Success - no deadlock\n\t\tcase <-time.After(5 * time.Second):\n\t\t\tt.Fatal(\"Concurrent control operations caused deadlock\")\n\t\t}\n\t})\n}\n\n// ==================== Helper Types ====================\n\n// slowExecutor is an executor with configurable delay\ntype slowExecutor struct {\n\tdelay time.Duration\n\tcount int32\n\tcurrent int32\n}\n\nfunc (e *slowExecutor) Execute(ctx *types.Context, robot *types.Robot, trigger types.TriggerType, data interface{}) (*types.Execution, error) {\n\treturn e.ExecuteWithControl(ctx, robot, trigger, data, \"\", nil)\n}\n\nfunc (e *slowExecutor) ExecuteWithID(ctx *types.Context, robot *types.Robot, trigger types.TriggerType, data interface{}, execID string) (*types.Execution, error) {\n\treturn e.ExecuteWithControl(ctx, robot, trigger, data, execID, nil)\n}\n\nfunc (e *slowExecutor) ExecuteWithControl(ctx *types.Context, robot *types.Robot, trigger types.TriggerType, data interface{}, execID string, control types.ExecutionControl) (*types.Execution, error) {\n\tif robot == nil {\n\t\treturn nil, types.ErrRobotNotFound\n\t}\n\n\t// Use provided execID or generate one\n\tif execID == \"\" {\n\t\texecID = \"exec_slow_\" + robot.MemberID\n\t}\n\texec := &types.Execution{\n\t\tID: execID,\n\t\tMemberID: robot.MemberID,\n\t\tTeamID: robot.TeamID,\n\t\tTriggerType: trigger,\n\t\tStartTime: time.Now(),\n\t\tStatus: types.ExecPending,\n\t}\n\n\tif !robot.TryAcquireSlot(exec) {\n\t\treturn nil, types.ErrQuotaExceeded\n\t}\n\tdefer robot.RemoveExecution(exec.ID)\n\n\tatomic.AddInt32(&e.current, 1)\n\tdefer atomic.AddInt32(&e.current, -1)\n\n\texec.Status = types.ExecRunning\n\ttime.Sleep(e.delay)\n\n\texec.Status = types.ExecCompleted\n\tnow := time.Now()\n\texec.EndTime = &now\n\n\tatomic.AddInt32(&e.count, 1)\n\treturn exec, nil\n}\n\nfunc (e *slowExecutor) ExecCount() int {\n\treturn int(atomic.LoadInt32(&e.count))\n}\n\nfunc (e *slowExecutor) CurrentCount() int {\n\treturn int(atomic.LoadInt32(&e.current))\n}\n\nfunc (e *slowExecutor) Resume(ctx *types.Context, execID string, reply string) error {\n\treturn fmt.Errorf(\"resume not supported in slow executor\")\n}\n\nfunc (e *slowExecutor) Reset() {\n\tatomic.StoreInt32(&e.count, 0)\n\tatomic.StoreInt32(&e.current, 0)\n}\n\n// ==================== Test Data Setup Helpers ====================\n\n// setupControlTestRobot creates a robot for control testing\nfunc setupControlTestRobot(t *testing.T, memberID, teamID string) {\n\tm := model.Select(\"__yao.member\")\n\ttableName := m.MetaData.Table.Name\n\tqb := capsule.Query()\n\n\trobotConfig := map[string]interface{}{\n\t\t\"identity\": map[string]interface{}{\n\t\t\t\"role\": \"Control Test Robot\",\n\t\t\t\"duties\": []string{\"Test execution control\"},\n\t\t},\n\t\t\"quota\": map[string]interface{}{\n\t\t\t\"max\": 5,\n\t\t\t\"queue\": 20,\n\t\t\t\"priority\": 5,\n\t\t},\n\t\t\"triggers\": map[string]interface{}{\n\t\t\t\"clock\": map[string]interface{}{\"enabled\": true},\n\t\t\t\"intervene\": map[string]interface{}{\"enabled\": true},\n\t\t\t\"event\": map[string]interface{}{\"enabled\": true},\n\t\t},\n\t}\n\tconfigJSON, _ := json.Marshal(robotConfig)\n\n\terr := qb.Table(tableName).Insert([]map[string]interface{}{\n\t\t{\n\t\t\t\"member_id\": memberID,\n\t\t\t\"team_id\": teamID,\n\t\t\t\"member_type\": \"robot\",\n\t\t\t\"display_name\": \"Control Test Robot \" + memberID,\n\t\t\t\"status\": \"active\",\n\t\t\t\"role_id\": \"member\",\n\t\t\t\"autonomous_mode\": true,\n\t\t\t\"robot_status\": \"idle\",\n\t\t\t\"robot_config\": string(configJSON),\n\t\t},\n\t})\n\tif err != nil {\n\t\tt.Fatalf(\"Failed to insert %s: %v\", memberID, err)\n\t}\n}\n" }, { "path": "agent/robot/manager/integration_event_test.go", "content": "package manager_test\n\n// Integration tests for Event triggers\n// Tests Manager.HandleEvent() with various event types and scenarios\n\nimport (\n\t\"context\"\n\t\"encoding/json\"\n\t\"testing\"\n\t\"time\"\n\n\t\"github.com/stretchr/testify/assert\"\n\t\"github.com/stretchr/testify/require\"\n\t\"github.com/yaoapp/gou/model\"\n\t\"github.com/yaoapp/xun/capsule\"\n\t\"github.com/yaoapp/yao/agent/robot/manager\"\n\t\"github.com/yaoapp/yao/agent/robot/pool\"\n\t\"github.com/yaoapp/yao/agent/robot/types\"\n\t\"github.com/yaoapp/yao/agent/testutils\"\n)\n\n// ==================== Event Trigger Tests ====================\n\n// TestIntegrationEventTrigger tests event trigger flow\nfunc TestIntegrationEventTrigger(t *testing.T) {\n\tif testing.Short() {\n\t\tt.Skip(\"Skipping integration test\")\n\t}\n\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\tcleanupIntegrationRobots(t)\n\tdefer cleanupIntegrationRobots(t)\n\n\tt.Run(\"webhook event success\", func(t *testing.T) {\n\t\tsetupEventTestRobot(t, \"robot_integ_event_webhook\", \"team_integ_event\")\n\n\t\tconfig := &manager.Config{\n\t\t\tTickInterval: 100 * time.Millisecond,\n\t\t\tPoolConfig: &pool.Config{WorkerSize: 3, QueueSize: 20},\n\t\t}\n\t\tm := manager.NewWithConfig(config)\n\n\t\terr := m.Start()\n\t\trequire.NoError(t, err)\n\t\tdefer m.Stop()\n\n\t\t// Verify robot is loaded into cache\n\t\trobot := m.Cache().Get(\"robot_integ_event_webhook\")\n\t\trequire.NotNil(t, robot, \"Robot should be loaded into cache\")\n\n\t\tctx := types.NewContext(context.Background(), nil)\n\t\treq := &types.EventRequest{\n\t\t\tMemberID: \"robot_integ_event_webhook\",\n\t\t\tSource: \"webhook\",\n\t\t\tEventType: \"lead.created\",\n\t\t\tData: map[string]interface{}{\n\t\t\t\t\"name\": \"John Doe\",\n\t\t\t\t\"email\": \"john@example.com\",\n\t\t\t\t\"company\": \"Acme Corp\",\n\t\t\t},\n\t\t}\n\n\t\tresult, err := m.HandleEvent(ctx, req)\n\t\tassert.NoError(t, err)\n\t\tassert.NotNil(t, result)\n\t\tassert.NotEmpty(t, result.ExecutionID)\n\t\tassert.Equal(t, types.ExecPending, result.Status)\n\t\tassert.Contains(t, result.Message, \"webhook\")\n\t\tassert.Contains(t, result.Message, \"lead.created\")\n\n\t\t// Wait for execution\n\t\ttime.Sleep(500 * time.Millisecond)\n\n\t\t// Verify execution completed\n\t\tassert.GreaterOrEqual(t, m.Executor().ExecCount(), 1)\n\t})\n\n\tt.Run(\"database event success\", func(t *testing.T) {\n\t\tsetupEventTestRobot(t, \"robot_integ_event_db\", \"team_integ_event\")\n\n\t\tm := manager.New()\n\t\terr := m.Start()\n\t\trequire.NoError(t, err)\n\t\tdefer m.Stop()\n\n\t\tctx := types.NewContext(context.Background(), nil)\n\t\treq := &types.EventRequest{\n\t\t\tMemberID: \"robot_integ_event_db\",\n\t\t\tSource: \"database\",\n\t\t\tEventType: \"order.paid\",\n\t\t\tData: map[string]interface{}{\n\t\t\t\t\"order_id\": \"ORD-12345\",\n\t\t\t\t\"amount\": 1500.00,\n\t\t\t\t\"customer\": \"customer_001\",\n\t\t\t},\n\t\t}\n\n\t\tresult, err := m.HandleEvent(ctx, req)\n\t\tassert.NoError(t, err)\n\t\tassert.NotNil(t, result)\n\t\tassert.NotEmpty(t, result.ExecutionID)\n\t})\n\n\tt.Run(\"event with complex data\", func(t *testing.T) {\n\t\tsetupEventTestRobot(t, \"robot_integ_event_complex\", \"team_integ_event\")\n\n\t\tm := manager.New()\n\t\terr := m.Start()\n\t\trequire.NoError(t, err)\n\t\tdefer m.Stop()\n\n\t\tctx := types.NewContext(context.Background(), nil)\n\t\treq := &types.EventRequest{\n\t\t\tMemberID: \"robot_integ_event_complex\",\n\t\t\tSource: \"webhook\",\n\t\t\tEventType: \"crm.contact.updated\",\n\t\t\tData: map[string]interface{}{\n\t\t\t\t\"contact\": map[string]interface{}{\n\t\t\t\t\t\"id\": \"contact_001\",\n\t\t\t\t\t\"name\": \"Jane Smith\",\n\t\t\t\t\t\"email\": \"jane@example.com\",\n\t\t\t\t\t\"tags\": []string{\"vip\", \"enterprise\"},\n\t\t\t\t},\n\t\t\t\t\"changes\": map[string]interface{}{\n\t\t\t\t\t\"old_status\": \"active\",\n\t\t\t\t\t\"new_status\": \"premium\",\n\t\t\t\t},\n\t\t\t\t\"timestamp\": time.Now().Unix(),\n\t\t\t},\n\t\t}\n\n\t\tresult, err := m.HandleEvent(ctx, req)\n\t\tassert.NoError(t, err)\n\t\tassert.NotNil(t, result)\n\t\tassert.NotEmpty(t, result.ExecutionID)\n\t})\n}\n\n// TestIntegrationEventTriggerErrors tests error cases for event triggers\nfunc TestIntegrationEventTriggerErrors(t *testing.T) {\n\tif testing.Short() {\n\t\tt.Skip(\"Skipping integration test\")\n\t}\n\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\tcleanupIntegrationRobots(t)\n\tdefer cleanupIntegrationRobots(t)\n\n\tt.Run(\"robot not found\", func(t *testing.T) {\n\t\tm := manager.New()\n\t\terr := m.Start()\n\t\trequire.NoError(t, err)\n\t\tdefer m.Stop()\n\n\t\tctx := types.NewContext(context.Background(), nil)\n\t\treq := &types.EventRequest{\n\t\t\tMemberID: \"robot_nonexistent\",\n\t\t\tSource: \"webhook\",\n\t\t\tEventType: \"test.event\",\n\t\t}\n\n\t\t_, err = m.HandleEvent(ctx, req)\n\t\tassert.Error(t, err)\n\t\tassert.Equal(t, types.ErrRobotNotFound, err)\n\t})\n\n\tt.Run(\"robot paused\", func(t *testing.T) {\n\t\tsetupEventTestRobotPaused(t, \"robot_integ_event_paused\", \"team_integ_event\")\n\n\t\tm := manager.New()\n\t\terr := m.Start()\n\t\trequire.NoError(t, err)\n\t\tdefer m.Stop()\n\n\t\tctx := types.NewContext(context.Background(), nil)\n\t\treq := &types.EventRequest{\n\t\t\tMemberID: \"robot_integ_event_paused\",\n\t\t\tSource: \"webhook\",\n\t\t\tEventType: \"test.event\",\n\t\t}\n\n\t\t_, err = m.HandleEvent(ctx, req)\n\t\tassert.Error(t, err)\n\t\tassert.Equal(t, types.ErrRobotPaused, err)\n\t})\n\n\tt.Run(\"event trigger disabled\", func(t *testing.T) {\n\t\tsetupEventTestRobotDisabled(t, \"robot_integ_event_disabled\", \"team_integ_event\")\n\n\t\tm := manager.New()\n\t\terr := m.Start()\n\t\trequire.NoError(t, err)\n\t\tdefer m.Stop()\n\n\t\tctx := types.NewContext(context.Background(), nil)\n\t\treq := &types.EventRequest{\n\t\t\tMemberID: \"robot_integ_event_disabled\",\n\t\t\tSource: \"webhook\",\n\t\t\tEventType: \"test.event\",\n\t\t}\n\n\t\t_, err = m.HandleEvent(ctx, req)\n\t\tassert.Error(t, err)\n\t\tassert.Equal(t, types.ErrTriggerDisabled, err)\n\t})\n\n\tt.Run(\"invalid request - empty member_id\", func(t *testing.T) {\n\t\tm := manager.New()\n\t\terr := m.Start()\n\t\trequire.NoError(t, err)\n\t\tdefer m.Stop()\n\n\t\tctx := types.NewContext(context.Background(), nil)\n\t\treq := &types.EventRequest{\n\t\t\tMemberID: \"\", // Empty\n\t\t\tSource: \"webhook\",\n\t\t\tEventType: \"test.event\",\n\t\t}\n\n\t\t_, err = m.HandleEvent(ctx, req)\n\t\tassert.Error(t, err)\n\t\tassert.Contains(t, err.Error(), \"member_id\")\n\t})\n\n\tt.Run(\"invalid request - empty source\", func(t *testing.T) {\n\t\tsetupEventTestRobot(t, \"robot_integ_event_nosource\", \"team_integ_event\")\n\n\t\tm := manager.New()\n\t\terr := m.Start()\n\t\trequire.NoError(t, err)\n\t\tdefer m.Stop()\n\n\t\tctx := types.NewContext(context.Background(), nil)\n\t\treq := &types.EventRequest{\n\t\t\tMemberID: \"robot_integ_event_nosource\",\n\t\t\tSource: \"\", // Empty\n\t\t\tEventType: \"test.event\",\n\t\t}\n\n\t\t_, err = m.HandleEvent(ctx, req)\n\t\tassert.Error(t, err)\n\t\tassert.Contains(t, err.Error(), \"source\")\n\t})\n\n\tt.Run(\"invalid request - empty event_type\", func(t *testing.T) {\n\t\tsetupEventTestRobot(t, \"robot_integ_event_notype\", \"team_integ_event\")\n\n\t\tm := manager.New()\n\t\terr := m.Start()\n\t\trequire.NoError(t, err)\n\t\tdefer m.Stop()\n\n\t\tctx := types.NewContext(context.Background(), nil)\n\t\treq := &types.EventRequest{\n\t\t\tMemberID: \"robot_integ_event_notype\",\n\t\t\tSource: \"webhook\",\n\t\t\tEventType: \"\", // Empty\n\t\t}\n\n\t\t_, err = m.HandleEvent(ctx, req)\n\t\tassert.Error(t, err)\n\t\tassert.Contains(t, err.Error(), \"event_type\")\n\t})\n\n\tt.Run(\"manager not started\", func(t *testing.T) {\n\t\tm := manager.New()\n\t\t// Don't start\n\n\t\tctx := types.NewContext(context.Background(), nil)\n\t\treq := &types.EventRequest{\n\t\t\tMemberID: \"robot_test\",\n\t\t\tSource: \"webhook\",\n\t\t\tEventType: \"test.event\",\n\t\t}\n\n\t\t_, err := m.HandleEvent(ctx, req)\n\t\tassert.Error(t, err)\n\t\tassert.Contains(t, err.Error(), \"not started\")\n\t})\n}\n\n// TestIntegrationEventTriggerTypes tests various event types\nfunc TestIntegrationEventTriggerTypes(t *testing.T) {\n\tif testing.Short() {\n\t\tt.Skip(\"Skipping integration test\")\n\t}\n\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\tcleanupIntegrationRobots(t)\n\tdefer cleanupIntegrationRobots(t)\n\n\t// Common event types to test\n\teventTypes := []struct {\n\t\tname string\n\t\teventType string\n\t\tdata map[string]interface{}\n\t}{\n\t\t{\n\t\t\tname: \"lead.created\",\n\t\t\teventType: \"lead.created\",\n\t\t\tdata: map[string]interface{}{\"name\": \"John\", \"email\": \"john@example.com\"},\n\t\t},\n\t\t{\n\t\t\tname: \"order.paid\",\n\t\t\teventType: \"order.paid\",\n\t\t\tdata: map[string]interface{}{\"order_id\": \"ORD-001\", \"amount\": 100.0},\n\t\t},\n\t\t{\n\t\t\tname: \"customer.signup\",\n\t\t\teventType: \"customer.signup\",\n\t\t\tdata: map[string]interface{}{\"customer_id\": \"cust_001\", \"plan\": \"premium\"},\n\t\t},\n\t\t{\n\t\t\tname: \"ticket.created\",\n\t\t\teventType: \"ticket.created\",\n\t\t\tdata: map[string]interface{}{\"ticket_id\": \"TKT-001\", \"priority\": \"high\"},\n\t\t},\n\t\t{\n\t\t\tname: \"inventory.low\",\n\t\t\teventType: \"inventory.low\",\n\t\t\tdata: map[string]interface{}{\"product_id\": \"PRD-001\", \"quantity\": 5},\n\t\t},\n\t}\n\n\tfor _, tc := range eventTypes {\n\t\tt.Run(tc.name, func(t *testing.T) {\n\t\t\tmemberID := \"robot_integ_event_type_\" + tc.name\n\t\t\tsetupEventTestRobot(t, memberID, \"team_integ_event\")\n\n\t\t\tm := manager.New()\n\t\t\terr := m.Start()\n\t\t\trequire.NoError(t, err)\n\t\t\tdefer m.Stop()\n\n\t\t\tctx := types.NewContext(context.Background(), nil)\n\t\t\treq := &types.EventRequest{\n\t\t\t\tMemberID: memberID,\n\t\t\t\tSource: \"webhook\",\n\t\t\t\tEventType: tc.eventType,\n\t\t\t\tData: tc.data,\n\t\t\t}\n\n\t\t\tresult, err := m.HandleEvent(ctx, req)\n\t\t\tassert.NoError(t, err, \"Event type %s should succeed\", tc.eventType)\n\t\t\tassert.NotNil(t, result)\n\t\t\tassert.NotEmpty(t, result.ExecutionID)\n\t\t\tassert.Equal(t, types.ExecPending, result.Status)\n\t\t})\n\t}\n}\n\n// TestIntegrationEventTriggerSources tests different event sources\nfunc TestIntegrationEventTriggerSources(t *testing.T) {\n\tif testing.Short() {\n\t\tt.Skip(\"Skipping integration test\")\n\t}\n\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\tcleanupIntegrationRobots(t)\n\tdefer cleanupIntegrationRobots(t)\n\n\tsources := []string{\"webhook\", \"database\", \"api\", \"scheduler\", \"internal\"}\n\n\tfor _, source := range sources {\n\t\tt.Run(\"source_\"+source, func(t *testing.T) {\n\t\t\tmemberID := \"robot_integ_event_src_\" + source\n\t\t\tsetupEventTestRobot(t, memberID, \"team_integ_event\")\n\n\t\t\tm := manager.New()\n\t\t\terr := m.Start()\n\t\t\trequire.NoError(t, err)\n\t\t\tdefer m.Stop()\n\n\t\t\tctx := types.NewContext(context.Background(), nil)\n\t\t\treq := &types.EventRequest{\n\t\t\t\tMemberID: memberID,\n\t\t\t\tSource: source,\n\t\t\t\tEventType: \"test.event\",\n\t\t\t\tData: map[string]interface{}{\"source\": source},\n\t\t\t}\n\n\t\t\tresult, err := m.HandleEvent(ctx, req)\n\t\t\tassert.NoError(t, err, \"Source %s should succeed\", source)\n\t\t\tassert.NotNil(t, result)\n\t\t\tassert.NotEmpty(t, result.ExecutionID)\n\t\t})\n\t}\n}\n\n// TestIntegrationEventTriggerWithEmptyData tests event with empty or nil data\nfunc TestIntegrationEventTriggerWithEmptyData(t *testing.T) {\n\tif testing.Short() {\n\t\tt.Skip(\"Skipping integration test\")\n\t}\n\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\tcleanupIntegrationRobots(t)\n\tdefer cleanupIntegrationRobots(t)\n\n\tt.Run(\"nil data\", func(t *testing.T) {\n\t\tsetupEventTestRobot(t, \"robot_integ_event_nildata\", \"team_integ_event\")\n\n\t\tm := manager.New()\n\t\terr := m.Start()\n\t\trequire.NoError(t, err)\n\t\tdefer m.Stop()\n\n\t\tctx := types.NewContext(context.Background(), nil)\n\t\treq := &types.EventRequest{\n\t\t\tMemberID: \"robot_integ_event_nildata\",\n\t\t\tSource: \"webhook\",\n\t\t\tEventType: \"ping\",\n\t\t\tData: nil, // Nil data\n\t\t}\n\n\t\tresult, err := m.HandleEvent(ctx, req)\n\t\tassert.NoError(t, err)\n\t\tassert.NotNil(t, result)\n\t\tassert.NotEmpty(t, result.ExecutionID)\n\t})\n\n\tt.Run(\"empty data map\", func(t *testing.T) {\n\t\tsetupEventTestRobot(t, \"robot_integ_event_emptydata\", \"team_integ_event\")\n\n\t\tm := manager.New()\n\t\terr := m.Start()\n\t\trequire.NoError(t, err)\n\t\tdefer m.Stop()\n\n\t\tctx := types.NewContext(context.Background(), nil)\n\t\treq := &types.EventRequest{\n\t\t\tMemberID: \"robot_integ_event_emptydata\",\n\t\t\tSource: \"webhook\",\n\t\t\tEventType: \"heartbeat\",\n\t\t\tData: map[string]interface{}{}, // Empty map\n\t\t}\n\n\t\tresult, err := m.HandleEvent(ctx, req)\n\t\tassert.NoError(t, err)\n\t\tassert.NotNil(t, result)\n\t\tassert.NotEmpty(t, result.ExecutionID)\n\t})\n}\n\n// ==================== Test Data Setup Helpers ====================\n\n// setupEventTestRobot creates a robot with event trigger enabled\nfunc setupEventTestRobot(t *testing.T, memberID, teamID string) {\n\tm := model.Select(\"__yao.member\")\n\ttableName := m.MetaData.Table.Name\n\tqb := capsule.Query()\n\n\trobotConfig := map[string]interface{}{\n\t\t\"identity\": map[string]interface{}{\n\t\t\t\"role\": \"Event Test Robot\",\n\t\t\t\"duties\": []string{\"Handle event triggers\"},\n\t\t},\n\t\t\"quota\": map[string]interface{}{\n\t\t\t\"max\": 5,\n\t\t\t\"queue\": 20,\n\t\t\t\"priority\": 5,\n\t\t},\n\t\t\"triggers\": map[string]interface{}{\n\t\t\t\"clock\": map[string]interface{}{\"enabled\": false},\n\t\t\t\"intervene\": map[string]interface{}{\"enabled\": false},\n\t\t\t\"event\": map[string]interface{}{\"enabled\": true},\n\t\t},\n\t\t\"events\": []map[string]interface{}{\n\t\t\t{\n\t\t\t\t\"type\": \"webhook\",\n\t\t\t\t\"source\": \"/webhook/events\",\n\t\t\t},\n\t\t\t{\n\t\t\t\t\"type\": \"database\",\n\t\t\t\t\"source\": \"orders\",\n\t\t\t},\n\t\t},\n\t}\n\tconfigJSON, _ := json.Marshal(robotConfig)\n\n\terr := qb.Table(tableName).Insert([]map[string]interface{}{\n\t\t{\n\t\t\t\"member_id\": memberID,\n\t\t\t\"team_id\": teamID,\n\t\t\t\"member_type\": \"robot\",\n\t\t\t\"display_name\": \"Event Test Robot \" + memberID,\n\t\t\t\"status\": \"active\",\n\t\t\t\"role_id\": \"member\",\n\t\t\t\"autonomous_mode\": true,\n\t\t\t\"robot_status\": \"idle\",\n\t\t\t\"robot_config\": string(configJSON),\n\t\t},\n\t})\n\tif err != nil {\n\t\tt.Fatalf(\"Failed to insert %s: %v\", memberID, err)\n\t}\n}\n\n// setupEventTestRobotPaused creates a paused robot\nfunc setupEventTestRobotPaused(t *testing.T, memberID, teamID string) {\n\tm := model.Select(\"__yao.member\")\n\ttableName := m.MetaData.Table.Name\n\tqb := capsule.Query()\n\n\trobotConfig := map[string]interface{}{\n\t\t\"identity\": map[string]interface{}{\"role\": \"Paused Event Robot\"},\n\t\t\"triggers\": map[string]interface{}{\n\t\t\t\"event\": map[string]interface{}{\"enabled\": true},\n\t\t},\n\t}\n\tconfigJSON, _ := json.Marshal(robotConfig)\n\n\terr := qb.Table(tableName).Insert([]map[string]interface{}{\n\t\t{\n\t\t\t\"member_id\": memberID,\n\t\t\t\"team_id\": teamID,\n\t\t\t\"member_type\": \"robot\",\n\t\t\t\"display_name\": \"Paused Event Robot \" + memberID,\n\t\t\t\"status\": \"active\",\n\t\t\t\"role_id\": \"member\",\n\t\t\t\"autonomous_mode\": true,\n\t\t\t\"robot_status\": \"paused\", // Paused\n\t\t\t\"robot_config\": string(configJSON),\n\t\t},\n\t})\n\tif err != nil {\n\t\tt.Fatalf(\"Failed to insert %s: %v\", memberID, err)\n\t}\n}\n\n// setupEventTestRobotDisabled creates a robot with event trigger disabled\nfunc setupEventTestRobotDisabled(t *testing.T, memberID, teamID string) {\n\tm := model.Select(\"__yao.member\")\n\ttableName := m.MetaData.Table.Name\n\tqb := capsule.Query()\n\n\trobotConfig := map[string]interface{}{\n\t\t\"identity\": map[string]interface{}{\"role\": \"Event Disabled Robot\"},\n\t\t\"triggers\": map[string]interface{}{\n\t\t\t\"event\": map[string]interface{}{\"enabled\": false}, // Disabled\n\t\t},\n\t}\n\tconfigJSON, _ := json.Marshal(robotConfig)\n\n\terr := qb.Table(tableName).Insert([]map[string]interface{}{\n\t\t{\n\t\t\t\"member_id\": memberID,\n\t\t\t\"team_id\": teamID,\n\t\t\t\"member_type\": \"robot\",\n\t\t\t\"display_name\": \"Event Disabled Robot \" + memberID,\n\t\t\t\"status\": \"active\",\n\t\t\t\"role_id\": \"member\",\n\t\t\t\"autonomous_mode\": true,\n\t\t\t\"robot_status\": \"idle\",\n\t\t\t\"robot_config\": string(configJSON),\n\t\t},\n\t})\n\tif err != nil {\n\t\tt.Fatalf(\"Failed to insert %s: %v\", memberID, err)\n\t}\n}\n" }, { "path": "agent/robot/manager/integration_human_test.go", "content": "package manager_test\n\n// Integration tests for Human intervention triggers\n// Tests Manager.Intervene() with various actions and scenarios\n\nimport (\n\t\"context\"\n\t\"encoding/json\"\n\t\"testing\"\n\t\"time\"\n\n\t\"github.com/stretchr/testify/assert\"\n\t\"github.com/stretchr/testify/require\"\n\t\"github.com/yaoapp/gou/model\"\n\t\"github.com/yaoapp/xun/capsule\"\n\tagentcontext \"github.com/yaoapp/yao/agent/context\"\n\t\"github.com/yaoapp/yao/agent/robot/manager\"\n\t\"github.com/yaoapp/yao/agent/robot/pool\"\n\t\"github.com/yaoapp/yao/agent/robot/types\"\n\t\"github.com/yaoapp/yao/agent/testutils\"\n)\n\n// ==================== Human Intervention Tests ====================\n\n// TestIntegrationHumanIntervention tests human intervention trigger flow\nfunc TestIntegrationHumanIntervention(t *testing.T) {\n\tif testing.Short() {\n\t\tt.Skip(\"Skipping integration test\")\n\t}\n\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\tcleanupIntegrationRobots(t)\n\tdefer cleanupIntegrationRobots(t)\n\n\tt.Run(\"task.add action success\", func(t *testing.T) {\n\t\tsetupInterveneTestRobot(t, \"robot_integ_human_add\", \"team_integ_human\")\n\n\t\tconfig := &manager.Config{\n\t\t\tTickInterval: 100 * time.Millisecond,\n\t\t\tPoolConfig: &pool.Config{WorkerSize: 3, QueueSize: 20},\n\t\t}\n\t\tm := manager.NewWithConfig(config)\n\n\t\terr := m.Start()\n\t\trequire.NoError(t, err)\n\t\tdefer m.Stop()\n\n\t\t// Verify robot is loaded into cache\n\t\trobot := m.Cache().Get(\"robot_integ_human_add\")\n\t\trequire.NotNil(t, robot, \"Robot should be loaded into cache\")\n\n\t\tctx := types.NewContext(context.Background(), nil)\n\t\treq := &types.InterveneRequest{\n\t\t\tTeamID: \"team_integ_human\",\n\t\t\tMemberID: \"robot_integ_human_add\",\n\t\t\tAction: types.ActionTaskAdd,\n\t\t\tMessages: []agentcontext.Message{\n\t\t\t\t{Role: agentcontext.RoleUser, Content: \"Add a new task: analyze sales data\"},\n\t\t\t},\n\t\t}\n\n\t\tresult, err := m.Intervene(ctx, req)\n\t\tassert.NoError(t, err)\n\t\tassert.NotNil(t, result)\n\t\tassert.NotEmpty(t, result.ExecutionID)\n\t\tassert.Equal(t, types.ExecPending, result.Status)\n\t\tassert.Contains(t, result.Message, \"task.add\")\n\n\t\t// Wait for execution\n\t\ttime.Sleep(500 * time.Millisecond)\n\n\t\t// Verify execution completed\n\t\tassert.GreaterOrEqual(t, m.Executor().ExecCount(), 1)\n\t})\n\n\tt.Run(\"goal.adjust action success\", func(t *testing.T) {\n\t\tsetupInterveneTestRobot(t, \"robot_integ_human_goal\", \"team_integ_human\")\n\n\t\tm := manager.New()\n\t\terr := m.Start()\n\t\trequire.NoError(t, err)\n\t\tdefer m.Stop()\n\n\t\tctx := types.NewContext(context.Background(), nil)\n\t\treq := &types.InterveneRequest{\n\t\t\tTeamID: \"team_integ_human\",\n\t\t\tMemberID: \"robot_integ_human_goal\",\n\t\t\tAction: types.ActionGoalAdjust,\n\t\t\tMessages: []agentcontext.Message{\n\t\t\t\t{Role: agentcontext.RoleUser, Content: \"Focus on high-priority customers only\"},\n\t\t\t},\n\t\t}\n\n\t\tresult, err := m.Intervene(ctx, req)\n\t\tassert.NoError(t, err)\n\t\tassert.NotNil(t, result)\n\t\tassert.NotEmpty(t, result.ExecutionID)\n\t})\n\n\tt.Run(\"instruct action success\", func(t *testing.T) {\n\t\tsetupInterveneTestRobot(t, \"robot_integ_human_instruct\", \"team_integ_human\")\n\n\t\tm := manager.New()\n\t\terr := m.Start()\n\t\trequire.NoError(t, err)\n\t\tdefer m.Stop()\n\n\t\tctx := types.NewContext(context.Background(), nil)\n\t\treq := &types.InterveneRequest{\n\t\t\tTeamID: \"team_integ_human\",\n\t\t\tMemberID: \"robot_integ_human_instruct\",\n\t\t\tAction: types.ActionInstruct,\n\t\t\tMessages: []agentcontext.Message{\n\t\t\t\t{Role: agentcontext.RoleUser, Content: \"Generate a weekly report\"},\n\t\t\t},\n\t\t}\n\n\t\tresult, err := m.Intervene(ctx, req)\n\t\tassert.NoError(t, err)\n\t\tassert.NotNil(t, result)\n\t\tassert.NotEmpty(t, result.ExecutionID)\n\t})\n}\n\n// TestIntegrationHumanInterventionErrors tests error cases for human intervention\nfunc TestIntegrationHumanInterventionErrors(t *testing.T) {\n\tif testing.Short() {\n\t\tt.Skip(\"Skipping integration test\")\n\t}\n\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\tcleanupIntegrationRobots(t)\n\tdefer cleanupIntegrationRobots(t)\n\n\tt.Run(\"robot not found\", func(t *testing.T) {\n\t\tm := manager.New()\n\t\terr := m.Start()\n\t\trequire.NoError(t, err)\n\t\tdefer m.Stop()\n\n\t\tctx := types.NewContext(context.Background(), nil)\n\t\treq := &types.InterveneRequest{\n\t\t\tMemberID: \"robot_nonexistent\",\n\t\t\tAction: types.ActionTaskAdd,\n\t\t\tMessages: []agentcontext.Message{\n\t\t\t\t{Role: agentcontext.RoleUser, Content: \"Test\"},\n\t\t\t},\n\t\t}\n\n\t\t_, err = m.Intervene(ctx, req)\n\t\tassert.Error(t, err)\n\t\tassert.Equal(t, types.ErrRobotNotFound, err)\n\t})\n\n\tt.Run(\"robot paused\", func(t *testing.T) {\n\t\tsetupInterveneTestRobotPaused(t, \"robot_integ_human_paused\", \"team_integ_human\")\n\n\t\tm := manager.New()\n\t\terr := m.Start()\n\t\trequire.NoError(t, err)\n\t\tdefer m.Stop()\n\n\t\tctx := types.NewContext(context.Background(), nil)\n\t\treq := &types.InterveneRequest{\n\t\t\tMemberID: \"robot_integ_human_paused\",\n\t\t\tAction: types.ActionTaskAdd,\n\t\t\tMessages: []agentcontext.Message{\n\t\t\t\t{Role: agentcontext.RoleUser, Content: \"Test\"},\n\t\t\t},\n\t\t}\n\n\t\t_, err = m.Intervene(ctx, req)\n\t\tassert.Error(t, err)\n\t\tassert.Equal(t, types.ErrRobotPaused, err)\n\t})\n\n\tt.Run(\"intervene trigger disabled\", func(t *testing.T) {\n\t\tsetupInterveneTestRobotDisabled(t, \"robot_integ_human_disabled\", \"team_integ_human\")\n\n\t\tm := manager.New()\n\t\terr := m.Start()\n\t\trequire.NoError(t, err)\n\t\tdefer m.Stop()\n\n\t\tctx := types.NewContext(context.Background(), nil)\n\t\treq := &types.InterveneRequest{\n\t\t\tMemberID: \"robot_integ_human_disabled\",\n\t\t\tAction: types.ActionTaskAdd,\n\t\t\tMessages: []agentcontext.Message{\n\t\t\t\t{Role: agentcontext.RoleUser, Content: \"Test\"},\n\t\t\t},\n\t\t}\n\n\t\t_, err = m.Intervene(ctx, req)\n\t\tassert.Error(t, err)\n\t\tassert.Equal(t, types.ErrTriggerDisabled, err)\n\t})\n\n\tt.Run(\"invalid request - empty member_id\", func(t *testing.T) {\n\t\tm := manager.New()\n\t\terr := m.Start()\n\t\trequire.NoError(t, err)\n\t\tdefer m.Stop()\n\n\t\tctx := types.NewContext(context.Background(), nil)\n\t\treq := &types.InterveneRequest{\n\t\t\tMemberID: \"\", // Empty\n\t\t\tAction: types.ActionTaskAdd,\n\t\t}\n\n\t\t_, err = m.Intervene(ctx, req)\n\t\tassert.Error(t, err)\n\t\tassert.Contains(t, err.Error(), \"member_id\")\n\t})\n\n\tt.Run(\"invalid request - empty action\", func(t *testing.T) {\n\t\tsetupInterveneTestRobot(t, \"robot_integ_human_noaction\", \"team_integ_human\")\n\n\t\tm := manager.New()\n\t\terr := m.Start()\n\t\trequire.NoError(t, err)\n\t\tdefer m.Stop()\n\n\t\tctx := types.NewContext(context.Background(), nil)\n\t\treq := &types.InterveneRequest{\n\t\t\tMemberID: \"robot_integ_human_noaction\",\n\t\t\tAction: \"\", // Empty action\n\t\t}\n\n\t\t_, err = m.Intervene(ctx, req)\n\t\tassert.Error(t, err)\n\t\tassert.Contains(t, err.Error(), \"action\")\n\t})\n\n\tt.Run(\"manager not started\", func(t *testing.T) {\n\t\tm := manager.New()\n\t\t// Don't start\n\n\t\tctx := types.NewContext(context.Background(), nil)\n\t\treq := &types.InterveneRequest{\n\t\t\tMemberID: \"robot_test\",\n\t\t\tAction: types.ActionTaskAdd,\n\t\t}\n\n\t\t_, err := m.Intervene(ctx, req)\n\t\tassert.Error(t, err)\n\t\tassert.Contains(t, err.Error(), \"not started\")\n\t})\n}\n\n// TestIntegrationHumanInterventionMultimodal tests multimodal input support\nfunc TestIntegrationHumanInterventionMultimodal(t *testing.T) {\n\tif testing.Short() {\n\t\tt.Skip(\"Skipping integration test\")\n\t}\n\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\tcleanupIntegrationRobots(t)\n\tdefer cleanupIntegrationRobots(t)\n\n\tt.Run(\"text message\", func(t *testing.T) {\n\t\tsetupInterveneTestRobot(t, \"robot_integ_human_text\", \"team_integ_human\")\n\n\t\tm := manager.New()\n\t\terr := m.Start()\n\t\trequire.NoError(t, err)\n\t\tdefer m.Stop()\n\n\t\tctx := types.NewContext(context.Background(), nil)\n\t\treq := &types.InterveneRequest{\n\t\t\tMemberID: \"robot_integ_human_text\",\n\t\t\tAction: types.ActionTaskAdd,\n\t\t\tMessages: []agentcontext.Message{\n\t\t\t\t{\n\t\t\t\t\tRole: agentcontext.RoleUser,\n\t\t\t\t\tContent: \"Analyze the quarterly sales report\",\n\t\t\t\t},\n\t\t\t},\n\t\t}\n\n\t\tresult, err := m.Intervene(ctx, req)\n\t\tassert.NoError(t, err)\n\t\tassert.NotNil(t, result)\n\t\tassert.NotEmpty(t, result.ExecutionID)\n\t})\n\n\tt.Run(\"message with image reference\", func(t *testing.T) {\n\t\tsetupInterveneTestRobot(t, \"robot_integ_human_image\", \"team_integ_human\")\n\n\t\tm := manager.New()\n\t\terr := m.Start()\n\t\trequire.NoError(t, err)\n\t\tdefer m.Stop()\n\n\t\tctx := types.NewContext(context.Background(), nil)\n\t\treq := &types.InterveneRequest{\n\t\t\tMemberID: \"robot_integ_human_image\",\n\t\t\tAction: types.ActionTaskAdd,\n\t\t\tMessages: []agentcontext.Message{\n\t\t\t\t{\n\t\t\t\t\tRole: agentcontext.RoleUser,\n\t\t\t\t\tContent: []interface{}{\n\t\t\t\t\t\tmap[string]interface{}{\n\t\t\t\t\t\t\t\"type\": \"text\",\n\t\t\t\t\t\t\t\"text\": \"Analyze this chart\",\n\t\t\t\t\t\t},\n\t\t\t\t\t\tmap[string]interface{}{\n\t\t\t\t\t\t\t\"type\": \"image_url\",\n\t\t\t\t\t\t\t\"image_url\": map[string]interface{}{\n\t\t\t\t\t\t\t\t\"url\": \"https://example.com/chart.png\",\n\t\t\t\t\t\t\t},\n\t\t\t\t\t\t},\n\t\t\t\t\t},\n\t\t\t\t},\n\t\t\t},\n\t\t}\n\n\t\tresult, err := m.Intervene(ctx, req)\n\t\tassert.NoError(t, err)\n\t\tassert.NotNil(t, result)\n\t\tassert.NotEmpty(t, result.ExecutionID)\n\t})\n\n\tt.Run(\"multiple messages\", func(t *testing.T) {\n\t\tsetupInterveneTestRobot(t, \"robot_integ_human_multi\", \"team_integ_human\")\n\n\t\tm := manager.New()\n\t\terr := m.Start()\n\t\trequire.NoError(t, err)\n\t\tdefer m.Stop()\n\n\t\tctx := types.NewContext(context.Background(), nil)\n\t\treq := &types.InterveneRequest{\n\t\t\tMemberID: \"robot_integ_human_multi\",\n\t\t\tAction: types.ActionTaskAdd,\n\t\t\tMessages: []agentcontext.Message{\n\t\t\t\t{Role: agentcontext.RoleUser, Content: \"First, check the sales data\"},\n\t\t\t\t{Role: agentcontext.RoleUser, Content: \"Then, prepare a summary report\"},\n\t\t\t},\n\t\t}\n\n\t\tresult, err := m.Intervene(ctx, req)\n\t\tassert.NoError(t, err)\n\t\tassert.NotNil(t, result)\n\t\tassert.NotEmpty(t, result.ExecutionID)\n\t})\n}\n\n// TestIntegrationHumanInterventionAllActions tests all intervention actions\nfunc TestIntegrationHumanInterventionAllActions(t *testing.T) {\n\tif testing.Short() {\n\t\tt.Skip(\"Skipping integration test\")\n\t}\n\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\tcleanupIntegrationRobots(t)\n\tdefer cleanupIntegrationRobots(t)\n\n\t// Test all defined actions\n\tactions := []types.InterventionAction{\n\t\ttypes.ActionTaskAdd,\n\t\ttypes.ActionTaskCancel,\n\t\ttypes.ActionTaskUpdate,\n\t\ttypes.ActionGoalAdjust,\n\t\ttypes.ActionGoalAdd,\n\t\ttypes.ActionGoalComplete,\n\t\ttypes.ActionGoalCancel,\n\t\ttypes.ActionInstruct,\n\t\t// Note: plan.add, plan.remove, plan.update are handled differently\n\t}\n\n\tfor _, action := range actions {\n\t\tt.Run(string(action), func(t *testing.T) {\n\t\t\tmemberID := \"robot_integ_action_\" + string(action)\n\t\t\tsetupInterveneTestRobot(t, memberID, \"team_integ_human\")\n\n\t\t\tm := manager.New()\n\t\t\terr := m.Start()\n\t\t\trequire.NoError(t, err)\n\t\t\tdefer m.Stop()\n\n\t\t\tctx := types.NewContext(context.Background(), nil)\n\t\t\treq := &types.InterveneRequest{\n\t\t\t\tMemberID: memberID,\n\t\t\t\tAction: action,\n\t\t\t\tMessages: []agentcontext.Message{\n\t\t\t\t\t{Role: agentcontext.RoleUser, Content: \"Test action: \" + string(action)},\n\t\t\t\t},\n\t\t\t}\n\n\t\t\tresult, err := m.Intervene(ctx, req)\n\t\t\tassert.NoError(t, err, \"Action %s should succeed\", action)\n\t\t\tassert.NotNil(t, result)\n\t\t\tassert.NotEmpty(t, result.ExecutionID)\n\t\t\tassert.Equal(t, types.ExecPending, result.Status)\n\t\t})\n\t}\n}\n\n// TestIntegrationHumanInterventionPlanAdd tests plan.add action (deferred execution)\nfunc TestIntegrationHumanInterventionPlanAdd(t *testing.T) {\n\tif testing.Short() {\n\t\tt.Skip(\"Skipping integration test\")\n\t}\n\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\tcleanupIntegrationRobots(t)\n\tdefer cleanupIntegrationRobots(t)\n\n\tt.Run(\"plan.add with future time\", func(t *testing.T) {\n\t\tsetupInterveneTestRobot(t, \"robot_integ_human_plan\", \"team_integ_human\")\n\n\t\tm := manager.New()\n\t\terr := m.Start()\n\t\trequire.NoError(t, err)\n\t\tdefer m.Stop()\n\n\t\tctx := types.NewContext(context.Background(), nil)\n\t\tplanTime := time.Now().Add(1 * time.Hour)\n\t\treq := &types.InterveneRequest{\n\t\t\tMemberID: \"robot_integ_human_plan\",\n\t\t\tAction: types.ActionPlanAdd,\n\t\t\tMessages: []agentcontext.Message{\n\t\t\t\t{Role: agentcontext.RoleUser, Content: \"Send weekly report\"},\n\t\t\t},\n\t\t\tPlanTime: &planTime,\n\t\t}\n\n\t\tresult, err := m.Intervene(ctx, req)\n\t\tassert.NoError(t, err)\n\t\tassert.NotNil(t, result)\n\t\tassert.Equal(t, types.ExecPending, result.Status)\n\t\tassert.Contains(t, result.Message, \"Planned\")\n\t\t// Note: Plan queue not implemented yet, so execution is deferred\n\t})\n}\n\n// ==================== Test Data Setup Helpers ====================\n\n// setupInterveneTestRobot creates a robot with intervene trigger enabled\nfunc setupInterveneTestRobot(t *testing.T, memberID, teamID string) {\n\tm := model.Select(\"__yao.member\")\n\ttableName := m.MetaData.Table.Name\n\tqb := capsule.Query()\n\n\trobotConfig := map[string]interface{}{\n\t\t\"identity\": map[string]interface{}{\n\t\t\t\"role\": \"Intervene Test Robot\",\n\t\t\t\"duties\": []string{\"Handle human interventions\"},\n\t\t},\n\t\t\"quota\": map[string]interface{}{\n\t\t\t\"max\": 5,\n\t\t\t\"queue\": 20,\n\t\t\t\"priority\": 5,\n\t\t},\n\t\t\"triggers\": map[string]interface{}{\n\t\t\t\"clock\": map[string]interface{}{\"enabled\": false},\n\t\t\t\"intervene\": map[string]interface{}{\"enabled\": true},\n\t\t\t\"event\": map[string]interface{}{\"enabled\": false},\n\t\t},\n\t}\n\tconfigJSON, _ := json.Marshal(robotConfig)\n\n\terr := qb.Table(tableName).Insert([]map[string]interface{}{\n\t\t{\n\t\t\t\"member_id\": memberID,\n\t\t\t\"team_id\": teamID,\n\t\t\t\"member_type\": \"robot\",\n\t\t\t\"display_name\": \"Intervene Test Robot \" + memberID,\n\t\t\t\"status\": \"active\",\n\t\t\t\"role_id\": \"member\",\n\t\t\t\"autonomous_mode\": true,\n\t\t\t\"robot_status\": \"idle\",\n\t\t\t\"robot_config\": string(configJSON),\n\t\t},\n\t})\n\tif err != nil {\n\t\tt.Fatalf(\"Failed to insert %s: %v\", memberID, err)\n\t}\n}\n\n// setupInterveneTestRobotPaused creates a paused robot\nfunc setupInterveneTestRobotPaused(t *testing.T, memberID, teamID string) {\n\tm := model.Select(\"__yao.member\")\n\ttableName := m.MetaData.Table.Name\n\tqb := capsule.Query()\n\n\trobotConfig := map[string]interface{}{\n\t\t\"identity\": map[string]interface{}{\"role\": \"Paused Robot\"},\n\t\t\"triggers\": map[string]interface{}{\n\t\t\t\"intervene\": map[string]interface{}{\"enabled\": true},\n\t\t},\n\t}\n\tconfigJSON, _ := json.Marshal(robotConfig)\n\n\terr := qb.Table(tableName).Insert([]map[string]interface{}{\n\t\t{\n\t\t\t\"member_id\": memberID,\n\t\t\t\"team_id\": teamID,\n\t\t\t\"member_type\": \"robot\",\n\t\t\t\"display_name\": \"Paused Robot \" + memberID,\n\t\t\t\"status\": \"active\",\n\t\t\t\"role_id\": \"member\",\n\t\t\t\"autonomous_mode\": true,\n\t\t\t\"robot_status\": \"paused\", // Paused\n\t\t\t\"robot_config\": string(configJSON),\n\t\t},\n\t})\n\tif err != nil {\n\t\tt.Fatalf(\"Failed to insert %s: %v\", memberID, err)\n\t}\n}\n\n// setupInterveneTestRobotDisabled creates a robot with intervene trigger disabled\nfunc setupInterveneTestRobotDisabled(t *testing.T, memberID, teamID string) {\n\tm := model.Select(\"__yao.member\")\n\ttableName := m.MetaData.Table.Name\n\tqb := capsule.Query()\n\n\trobotConfig := map[string]interface{}{\n\t\t\"identity\": map[string]interface{}{\"role\": \"Intervene Disabled Robot\"},\n\t\t\"triggers\": map[string]interface{}{\n\t\t\t\"intervene\": map[string]interface{}{\"enabled\": false}, // Disabled\n\t\t},\n\t}\n\tconfigJSON, _ := json.Marshal(robotConfig)\n\n\terr := qb.Table(tableName).Insert([]map[string]interface{}{\n\t\t{\n\t\t\t\"member_id\": memberID,\n\t\t\t\"team_id\": teamID,\n\t\t\t\"member_type\": \"robot\",\n\t\t\t\"display_name\": \"Intervene Disabled Robot \" + memberID,\n\t\t\t\"status\": \"active\",\n\t\t\t\"role_id\": \"member\",\n\t\t\t\"autonomous_mode\": true,\n\t\t\t\"robot_status\": \"idle\",\n\t\t\t\"robot_config\": string(configJSON),\n\t\t},\n\t})\n\tif err != nil {\n\t\tt.Fatalf(\"Failed to insert %s: %v\", memberID, err)\n\t}\n}\n" }, { "path": "agent/robot/manager/integration_test.go", "content": "package manager_test\n\n// Integration tests for the Robot Agent scheduling system\n// These tests verify the complete end-to-end flow:\n// Trigger → Manager → Cache → Pool → Worker → Executor → Job\n//\n// Test Structure:\n// - integration_test.go: Core scheduling flow tests\n// - integration_clock_test.go: Clock trigger mode tests (times/interval/daemon)\n// - integration_human_test.go: Human intervention trigger tests\n// - integration_event_test.go: Event trigger tests\n// - integration_concurrent_test.go: Concurrent execution & quota tests\n// - integration_control_test.go: Pause/Resume/Stop tests\n//\n// Test Data:\n// All tests use real database records in __yao.member table\n// Test robot IDs are prefixed with \"robot_integ_\" for easy cleanup\n\nimport (\n\t\"context\"\n\t\"encoding/json\"\n\t\"testing\"\n\t\"time\"\n\n\t\"github.com/stretchr/testify/assert\"\n\t\"github.com/stretchr/testify/require\"\n\t\"github.com/yaoapp/gou/model\"\n\t\"github.com/yaoapp/xun/capsule\"\n\t\"github.com/yaoapp/yao/agent/robot/executor\"\n\t\"github.com/yaoapp/yao/agent/robot/manager\"\n\t\"github.com/yaoapp/yao/agent/robot/pool\"\n\t\"github.com/yaoapp/yao/agent/robot/types\"\n\t\"github.com/yaoapp/yao/agent/testutils\"\n)\n\n// ==================== Core Scheduling Flow Tests ====================\n\n// TestIntegrationSchedulingFlow tests the complete scheduling flow:\n// Create robot → Start manager → Trigger → Verify execution\nfunc TestIntegrationSchedulingFlow(t *testing.T) {\n\tif testing.Short() {\n\t\tt.Skip(\"Skipping integration test\")\n\t}\n\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\tcleanupIntegrationRobots(t)\n\tdefer cleanupIntegrationRobots(t)\n\n\tt.Run(\"complete clock trigger flow\", func(t *testing.T) {\n\t\t// Setup: Create a robot with times mode clock config\n\t\tsetupIntegrationRobotTimes(t, \"robot_integ_flow_clock\", \"team_integ_flow\")\n\n\t\t// Create manager with fast tick interval for testing\n\t\tconfig := &manager.Config{\n\t\t\tTickInterval: 100 * time.Millisecond,\n\t\t\tPoolConfig: &pool.Config{WorkerSize: 5, QueueSize: 50},\n\t\t}\n\t\tm := manager.NewWithConfig(config)\n\n\t\t// Start manager\n\t\terr := m.Start()\n\t\trequire.NoError(t, err)\n\t\tdefer m.Stop()\n\n\t\t// Verify robot is loaded into cache\n\t\trobot := m.Cache().Get(\"robot_integ_flow_clock\")\n\t\trequire.NotNil(t, robot, \"Robot should be loaded into cache\")\n\t\tassert.Equal(t, \"robot_integ_flow_clock\", robot.MemberID)\n\t\tassert.Equal(t, types.RobotIdle, robot.Status)\n\n\t\t// Simulate clock trigger at matching time (03:33 on Wednesday)\n\t\tloc, _ := time.LoadLocation(\"Asia/Shanghai\")\n\t\ttriggerTime := time.Date(2025, 1, 15, 3, 33, 0, 0, loc) // Wednesday 03:33\n\n\t\tctx := types.NewContext(context.Background(), nil)\n\t\terr = m.Tick(ctx, triggerTime)\n\t\tassert.NoError(t, err)\n\n\t\t// Wait for execution to complete\n\t\ttime.Sleep(500 * time.Millisecond)\n\n\t\t// Verify execution happened\n\t\texecCount := m.Executor().ExecCount()\n\t\tassert.GreaterOrEqual(t, execCount, 1, \"Should have at least 1 execution\")\n\t})\n\n\tt.Run(\"robot loaded from database\", func(t *testing.T) {\n\t\t// Setup: Create multiple robots\n\t\tsetupIntegrationRobotTimes(t, \"robot_integ_flow_db1\", \"team_integ_flow\")\n\t\tsetupIntegrationRobotInterval(t, \"robot_integ_flow_db2\", \"team_integ_flow\")\n\n\t\tm := manager.New()\n\t\terr := m.Start()\n\t\trequire.NoError(t, err)\n\t\tdefer m.Stop()\n\n\t\t// Verify both robots are in cache\n\t\trobot1 := m.Cache().Get(\"robot_integ_flow_db1\")\n\t\trobot2 := m.Cache().Get(\"robot_integ_flow_db2\")\n\t\tassert.NotNil(t, robot1, \"Robot 1 should be loaded\")\n\t\tassert.NotNil(t, robot2, \"Robot 2 should be loaded\")\n\n\t\t// Verify config is parsed correctly\n\t\tassert.NotNil(t, robot1.Config)\n\t\tassert.NotNil(t, robot1.Config.Clock)\n\t\tassert.Equal(t, types.ClockTimes, robot1.Config.Clock.Mode)\n\n\t\tassert.NotNil(t, robot2.Config)\n\t\tassert.NotNil(t, robot2.Config.Clock)\n\t\tassert.Equal(t, types.ClockInterval, robot2.Config.Clock.Mode)\n\t})\n\n\tt.Run(\"inactive robot not loaded\", func(t *testing.T) {\n\t\t// Setup: Create an inactive robot\n\t\tsetupIntegrationRobotInactive(t, \"robot_integ_flow_inactive\", \"team_integ_flow\")\n\n\t\tm := manager.New()\n\t\terr := m.Start()\n\t\trequire.NoError(t, err)\n\t\tdefer m.Stop()\n\n\t\t// Inactive robot should not be in cache\n\t\trobot := m.Cache().Get(\"robot_integ_flow_inactive\")\n\t\tassert.Nil(t, robot, \"Inactive robot should not be loaded\")\n\t})\n\n\tt.Run(\"robot with autonomous_mode=false not loaded\", func(t *testing.T) {\n\t\t// Setup: Create a robot with autonomous_mode=false\n\t\tsetupIntegrationRobotNonAutonomous(t, \"robot_integ_flow_nonauto\", \"team_integ_flow\")\n\n\t\tm := manager.New()\n\t\terr := m.Start()\n\t\trequire.NoError(t, err)\n\t\tdefer m.Stop()\n\n\t\t// Non-autonomous robot should not be in cache\n\t\trobot := m.Cache().Get(\"robot_integ_flow_nonauto\")\n\t\tassert.Nil(t, robot, \"Non-autonomous robot should not be loaded\")\n\t})\n}\n\n// TestIntegrationJobSubmission tests job submission to pool and execution\nfunc TestIntegrationJobSubmission(t *testing.T) {\n\tif testing.Short() {\n\t\tt.Skip(\"Skipping integration test\")\n\t}\n\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\tcleanupIntegrationRobots(t)\n\tdefer cleanupIntegrationRobots(t)\n\n\tt.Run(\"job submitted to pool and executed\", func(t *testing.T) {\n\t\tsetupIntegrationRobotTimes(t, \"robot_integ_submit\", \"team_integ_submit\")\n\n\t\tconfig := &manager.Config{\n\t\t\tTickInterval: 100 * time.Millisecond,\n\t\t\tPoolConfig: &pool.Config{WorkerSize: 3, QueueSize: 20},\n\t\t}\n\t\tm := manager.NewWithConfig(config)\n\n\t\terr := m.Start()\n\t\trequire.NoError(t, err)\n\t\tdefer m.Stop()\n\n\t\t// Manually trigger execution\n\t\tctx := types.NewContext(context.Background(), nil)\n\t\texecID, err := m.TriggerManual(ctx, \"robot_integ_submit\", types.TriggerClock, nil)\n\t\tassert.NoError(t, err)\n\t\tassert.NotEmpty(t, execID, \"Should return execution ID\")\n\n\t\t// Wait for execution\n\t\ttime.Sleep(500 * time.Millisecond)\n\n\t\t// Verify execution completed\n\t\tassert.GreaterOrEqual(t, m.Executor().ExecCount(), 1)\n\t})\n\n\tt.Run(\"multiple jobs queued and executed in order\", func(t *testing.T) {\n\t\tsetupIntegrationRobotHighQuota(t, \"robot_integ_queue\", \"team_integ_submit\")\n\n\t\tconfig := &manager.Config{\n\t\t\tTickInterval: 100 * time.Millisecond,\n\t\t\tPoolConfig: &pool.Config{WorkerSize: 2, QueueSize: 50},\n\t\t}\n\t\tm := manager.NewWithConfig(config)\n\n\t\terr := m.Start()\n\t\trequire.NoError(t, err)\n\t\tdefer m.Stop()\n\n\t\tctx := types.NewContext(context.Background(), nil)\n\n\t\t// Submit multiple jobs\n\t\texecIDs := make([]string, 5)\n\t\tfor i := 0; i < 5; i++ {\n\t\t\texecID, err := m.TriggerManual(ctx, \"robot_integ_queue\", types.TriggerClock, nil)\n\t\t\tassert.NoError(t, err)\n\t\t\texecIDs[i] = execID\n\t\t}\n\n\t\t// All should have valid IDs\n\t\tfor i, id := range execIDs {\n\t\t\tassert.NotEmpty(t, id, \"Execution %d should have valid ID\", i)\n\t\t}\n\n\t\t// Wait for all to complete (longer wait for slow execution)\n\t\ttime.Sleep(2 * time.Second)\n\n\t\t// All jobs should have executed\n\t\texecCount := m.Executor().ExecCount()\n\t\tassert.GreaterOrEqual(t, execCount, 5, \"Expected at least 5 executions, got %d\", execCount)\n\t})\n}\n\n// TestIntegrationPhaseProgression tests that execution progresses through all phases\nfunc TestIntegrationPhaseProgression(t *testing.T) {\n\tif testing.Short() {\n\t\tt.Skip(\"Skipping integration test\")\n\t}\n\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\tcleanupIntegrationRobots(t)\n\tdefer cleanupIntegrationRobots(t)\n\n\tt.Run(\"clock trigger executes all phases P0-P5\", func(t *testing.T) {\n\t\tcleanupIntegrationRobots(t)\n\t\tsetupIntegrationRobotTimes(t, \"robot_integ_phases_clock\", \"team_integ_phases\")\n\n\t\t// Track phases executed\n\t\tphasesExecuted := make([]types.Phase, 0)\n\t\texec := executor.NewDryRunWithConfig(executor.DryRunConfig{\n\t\t\tConfig: executor.Config{\n\t\t\t\tOnPhaseStart: func(phase types.Phase) {\n\t\t\t\t\tphasesExecuted = append(phasesExecuted, phase)\n\t\t\t\t},\n\t\t\t},\n\t\t})\n\n\t\tconfig := &manager.Config{\n\t\t\tTickInterval: 100 * time.Millisecond,\n\t\t\tPoolConfig: &pool.Config{WorkerSize: 2, QueueSize: 20},\n\t\t\tExecutor: exec,\n\t\t}\n\t\tm := manager.NewWithConfig(config)\n\n\t\terr := m.Start()\n\t\trequire.NoError(t, err)\n\n\t\t// Trigger execution\n\t\tctx := types.NewContext(context.Background(), nil)\n\t\t_, err = m.TriggerManual(ctx, \"robot_integ_phases_clock\", types.TriggerClock, nil)\n\t\tassert.NoError(t, err)\n\n\t\t// Wait for execution\n\t\ttime.Sleep(500 * time.Millisecond)\n\n\t\t// Stop manager before asserting to prevent ticker from triggering extra executions\n\t\tm.Stop()\n\n\t\t// Verify all 6 phases executed (P0-P5)\n\t\tassert.Len(t, phasesExecuted, 6, \"Should execute all 6 phases for clock trigger\")\n\t\tassert.Equal(t, types.PhaseInspiration, phasesExecuted[0], \"Should start with P0\")\n\t\tassert.Equal(t, types.PhaseLearning, phasesExecuted[5], \"Should end with P5\")\n\t})\n\n\tt.Run(\"human trigger skips P0 and executes P1-P5\", func(t *testing.T) {\n\t\tcleanupIntegrationRobots(t)\n\t\tsetupIntegrationRobotIntervene(t, \"robot_integ_phases_human\", \"team_integ_phases\")\n\n\t\t// Track phases executed\n\t\tphasesExecuted := make([]types.Phase, 0)\n\t\texec := executor.NewDryRunWithConfig(executor.DryRunConfig{\n\t\t\tConfig: executor.Config{\n\t\t\t\tOnPhaseStart: func(phase types.Phase) {\n\t\t\t\t\tphasesExecuted = append(phasesExecuted, phase)\n\t\t\t\t},\n\t\t\t},\n\t\t})\n\n\t\tconfig := &manager.Config{\n\t\t\tTickInterval: 100 * time.Millisecond,\n\t\t\tPoolConfig: &pool.Config{WorkerSize: 2, QueueSize: 20},\n\t\t\tExecutor: exec,\n\t\t}\n\t\tm := manager.NewWithConfig(config)\n\n\t\terr := m.Start()\n\t\trequire.NoError(t, err)\n\n\t\t// Trigger execution via human trigger\n\t\tctx := types.NewContext(context.Background(), nil)\n\t\t_, err = m.TriggerManual(ctx, \"robot_integ_phases_human\", types.TriggerHuman, nil)\n\t\tassert.NoError(t, err)\n\n\t\t// Wait for execution\n\t\ttime.Sleep(500 * time.Millisecond)\n\n\t\t// Stop manager before asserting to prevent ticker from triggering extra executions\n\t\tm.Stop()\n\n\t\t// Verify 5 phases executed (P1-P5, skipping P0)\n\t\tassert.Len(t, phasesExecuted, 5, \"Should execute 5 phases for human trigger\")\n\t\tassert.Equal(t, types.PhaseGoals, phasesExecuted[0], \"Should start with P1 (Goals)\")\n\t\tassert.Equal(t, types.PhaseLearning, phasesExecuted[4], \"Should end with P5\")\n\t})\n}\n\n// TestIntegrationCacheRefresh tests that cache refresh works correctly\nfunc TestIntegrationCacheRefresh(t *testing.T) {\n\tif testing.Short() {\n\t\tt.Skip(\"Skipping integration test\")\n\t}\n\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\tcleanupIntegrationRobots(t)\n\tdefer cleanupIntegrationRobots(t)\n\n\tt.Run(\"cache refresh loads new robots\", func(t *testing.T) {\n\t\t// Start with one robot\n\t\tsetupIntegrationRobotTimes(t, \"robot_integ_refresh1\", \"team_integ_refresh\")\n\n\t\tm := manager.New()\n\t\terr := m.Start()\n\t\trequire.NoError(t, err)\n\t\tdefer m.Stop()\n\n\t\t// Verify first robot is loaded\n\t\trobot1 := m.Cache().Get(\"robot_integ_refresh1\")\n\t\tassert.NotNil(t, robot1)\n\n\t\t// Add another robot to database\n\t\tsetupIntegrationRobotTimes(t, \"robot_integ_refresh2\", \"team_integ_refresh\")\n\n\t\t// Manually refresh cache\n\t\tctx := types.NewContext(context.Background(), nil)\n\t\terr = m.Cache().Load(ctx)\n\t\tassert.NoError(t, err)\n\n\t\t// Verify new robot is now in cache\n\t\trobot2 := m.Cache().Get(\"robot_integ_refresh2\")\n\t\tassert.NotNil(t, robot2, \"New robot should be loaded after refresh\")\n\t})\n}\n\n// ==================== Test Data Setup Helpers ====================\n\n// setupIntegrationRobotTimes creates a robot with times mode clock config\nfunc setupIntegrationRobotTimes(t *testing.T, memberID, teamID string) {\n\tm := model.Select(\"__yao.member\")\n\ttableName := m.MetaData.Table.Name\n\tqb := capsule.Query()\n\n\trobotConfig := map[string]interface{}{\n\t\t\"identity\": map[string]interface{}{\n\t\t\t\"role\": \"Integration Test Robot (Times)\",\n\t\t\t\"duties\": []string{\"Test scheduling\"},\n\t\t},\n\t\t\"quota\": map[string]interface{}{\n\t\t\t\"max\": 3,\n\t\t\t\"queue\": 20,\n\t\t\t\"priority\": 5,\n\t\t},\n\t\t\"triggers\": map[string]interface{}{\n\t\t\t\"clock\": map[string]interface{}{\"enabled\": true},\n\t\t\t\"intervene\": map[string]interface{}{\"enabled\": true},\n\t\t\t\"event\": map[string]interface{}{\"enabled\": true},\n\t\t},\n\t\t\"clock\": map[string]interface{}{\n\t\t\t\"mode\": \"times\",\n\t\t\t\"times\": []string{\"03:33\"},\n\t\t\t\"days\": []string{\"Mon\", \"Tue\", \"Wed\", \"Thu\", \"Fri\"},\n\t\t\t\"tz\": \"Asia/Shanghai\",\n\t\t\t\"timeout\": \"30m\",\n\t\t},\n\t}\n\tconfigJSON, _ := json.Marshal(robotConfig)\n\n\terr := qb.Table(tableName).Insert([]map[string]interface{}{\n\t\t{\n\t\t\t\"member_id\": memberID,\n\t\t\t\"team_id\": teamID,\n\t\t\t\"member_type\": \"robot\",\n\t\t\t\"display_name\": \"Test Robot \" + memberID,\n\t\t\t\"system_prompt\": \"You are an integration test robot.\",\n\t\t\t\"status\": \"active\",\n\t\t\t\"role_id\": \"member\",\n\t\t\t\"autonomous_mode\": true,\n\t\t\t\"robot_status\": \"idle\",\n\t\t\t\"robot_config\": string(configJSON),\n\t\t},\n\t})\n\tif err != nil {\n\t\tt.Fatalf(\"Failed to insert %s: %v\", memberID, err)\n\t}\n}\n\n// setupIntegrationRobotInterval creates a robot with interval mode clock config\nfunc setupIntegrationRobotInterval(t *testing.T, memberID, teamID string) {\n\tm := model.Select(\"__yao.member\")\n\ttableName := m.MetaData.Table.Name\n\tqb := capsule.Query()\n\n\trobotConfig := map[string]interface{}{\n\t\t\"identity\": map[string]interface{}{\n\t\t\t\"role\": \"Integration Test Robot (Interval)\",\n\t\t},\n\t\t\"quota\": map[string]interface{}{\n\t\t\t\"max\": 2,\n\t\t\t\"queue\": 10,\n\t\t\t\"priority\": 5,\n\t\t},\n\t\t\"triggers\": map[string]interface{}{\n\t\t\t\"clock\": map[string]interface{}{\"enabled\": true},\n\t\t},\n\t\t\"clock\": map[string]interface{}{\n\t\t\t\"mode\": \"interval\",\n\t\t\t\"every\": \"30m\",\n\t\t\t\"timeout\": \"10m\",\n\t\t},\n\t}\n\tconfigJSON, _ := json.Marshal(robotConfig)\n\n\terr := qb.Table(tableName).Insert([]map[string]interface{}{\n\t\t{\n\t\t\t\"member_id\": memberID,\n\t\t\t\"team_id\": teamID,\n\t\t\t\"member_type\": \"robot\",\n\t\t\t\"display_name\": \"Test Robot \" + memberID,\n\t\t\t\"status\": \"active\",\n\t\t\t\"role_id\": \"member\",\n\t\t\t\"autonomous_mode\": true,\n\t\t\t\"robot_status\": \"idle\",\n\t\t\t\"robot_config\": string(configJSON),\n\t\t},\n\t})\n\tif err != nil {\n\t\tt.Fatalf(\"Failed to insert %s: %v\", memberID, err)\n\t}\n}\n\n// setupIntegrationRobotHighQuota creates a robot with high quota for queue tests\nfunc setupIntegrationRobotHighQuota(t *testing.T, memberID, teamID string) {\n\tm := model.Select(\"__yao.member\")\n\ttableName := m.MetaData.Table.Name\n\tqb := capsule.Query()\n\n\trobotConfig := map[string]interface{}{\n\t\t\"identity\": map[string]interface{}{\n\t\t\t\"role\": \"Integration Test Robot (High Quota)\",\n\t\t},\n\t\t\"quota\": map[string]interface{}{\n\t\t\t\"max\": 10,\n\t\t\t\"queue\": 50,\n\t\t\t\"priority\": 5,\n\t\t},\n\t\t\"triggers\": map[string]interface{}{\n\t\t\t\"clock\": map[string]interface{}{\"enabled\": true},\n\t\t},\n\t\t\"clock\": map[string]interface{}{\n\t\t\t\"mode\": \"times\",\n\t\t\t\"times\": []string{\"03:33\"},\n\t\t\t\"tz\": \"Asia/Shanghai\",\n\t\t},\n\t}\n\tconfigJSON, _ := json.Marshal(robotConfig)\n\n\terr := qb.Table(tableName).Insert([]map[string]interface{}{\n\t\t{\n\t\t\t\"member_id\": memberID,\n\t\t\t\"team_id\": teamID,\n\t\t\t\"member_type\": \"robot\",\n\t\t\t\"display_name\": \"Test Robot \" + memberID,\n\t\t\t\"status\": \"active\",\n\t\t\t\"role_id\": \"member\",\n\t\t\t\"autonomous_mode\": true,\n\t\t\t\"robot_status\": \"idle\",\n\t\t\t\"robot_config\": string(configJSON),\n\t\t},\n\t})\n\tif err != nil {\n\t\tt.Fatalf(\"Failed to insert %s: %v\", memberID, err)\n\t}\n}\n\n// setupIntegrationRobotIntervene creates a robot with intervene trigger enabled\nfunc setupIntegrationRobotIntervene(t *testing.T, memberID, teamID string) {\n\tm := model.Select(\"__yao.member\")\n\ttableName := m.MetaData.Table.Name\n\tqb := capsule.Query()\n\n\trobotConfig := map[string]interface{}{\n\t\t\"identity\": map[string]interface{}{\n\t\t\t\"role\": \"Integration Test Robot (Intervene)\",\n\t\t},\n\t\t\"quota\": map[string]interface{}{\n\t\t\t\"max\": 5,\n\t\t\t\"queue\": 20,\n\t\t\t\"priority\": 5,\n\t\t},\n\t\t\"triggers\": map[string]interface{}{\n\t\t\t\"clock\": map[string]interface{}{\"enabled\": false},\n\t\t\t\"intervene\": map[string]interface{}{\"enabled\": true},\n\t\t},\n\t}\n\tconfigJSON, _ := json.Marshal(robotConfig)\n\n\terr := qb.Table(tableName).Insert([]map[string]interface{}{\n\t\t{\n\t\t\t\"member_id\": memberID,\n\t\t\t\"team_id\": teamID,\n\t\t\t\"member_type\": \"robot\",\n\t\t\t\"display_name\": \"Test Robot \" + memberID,\n\t\t\t\"status\": \"active\",\n\t\t\t\"role_id\": \"member\",\n\t\t\t\"autonomous_mode\": true,\n\t\t\t\"robot_status\": \"idle\",\n\t\t\t\"robot_config\": string(configJSON),\n\t\t},\n\t})\n\tif err != nil {\n\t\tt.Fatalf(\"Failed to insert %s: %v\", memberID, err)\n\t}\n}\n\n// setupIntegrationRobotInactive creates an inactive robot (should not be loaded)\nfunc setupIntegrationRobotInactive(t *testing.T, memberID, teamID string) {\n\tm := model.Select(\"__yao.member\")\n\ttableName := m.MetaData.Table.Name\n\tqb := capsule.Query()\n\n\trobotConfig := map[string]interface{}{\n\t\t\"identity\": map[string]interface{}{\n\t\t\t\"role\": \"Inactive Robot\",\n\t\t},\n\t\t\"triggers\": map[string]interface{}{\n\t\t\t\"clock\": map[string]interface{}{\"enabled\": true},\n\t\t},\n\t}\n\tconfigJSON, _ := json.Marshal(robotConfig)\n\n\terr := qb.Table(tableName).Insert([]map[string]interface{}{\n\t\t{\n\t\t\t\"member_id\": memberID,\n\t\t\t\"team_id\": teamID,\n\t\t\t\"member_type\": \"robot\",\n\t\t\t\"display_name\": \"Inactive Robot \" + memberID,\n\t\t\t\"status\": \"inactive\", // Inactive status\n\t\t\t\"role_id\": \"member\",\n\t\t\t\"autonomous_mode\": true,\n\t\t\t\"robot_status\": \"paused\",\n\t\t\t\"robot_config\": string(configJSON),\n\t\t},\n\t})\n\tif err != nil {\n\t\tt.Fatalf(\"Failed to insert %s: %v\", memberID, err)\n\t}\n}\n\n// setupIntegrationRobotNonAutonomous creates a robot with autonomous_mode=false\nfunc setupIntegrationRobotNonAutonomous(t *testing.T, memberID, teamID string) {\n\tm := model.Select(\"__yao.member\")\n\ttableName := m.MetaData.Table.Name\n\tqb := capsule.Query()\n\n\trobotConfig := map[string]interface{}{\n\t\t\"identity\": map[string]interface{}{\n\t\t\t\"role\": \"Non-Autonomous Robot\",\n\t\t},\n\t}\n\tconfigJSON, _ := json.Marshal(robotConfig)\n\n\terr := qb.Table(tableName).Insert([]map[string]interface{}{\n\t\t{\n\t\t\t\"member_id\": memberID,\n\t\t\t\"team_id\": teamID,\n\t\t\t\"member_type\": \"robot\",\n\t\t\t\"display_name\": \"Non-Autonomous Robot \" + memberID,\n\t\t\t\"status\": \"active\",\n\t\t\t\"role_id\": \"member\",\n\t\t\t\"autonomous_mode\": false, // Not autonomous\n\t\t\t\"robot_status\": \"idle\",\n\t\t\t\"robot_config\": string(configJSON),\n\t\t},\n\t})\n\tif err != nil {\n\t\tt.Fatalf(\"Failed to insert %s: %v\", memberID, err)\n\t}\n}\n\n// cleanupIntegrationRobots removes all integration test robots\nfunc cleanupIntegrationRobots(t *testing.T) {\n\tqb := capsule.Query()\n\tm := model.Select(\"__yao.member\")\n\ttableName := m.MetaData.Table.Name\n\n\t// Delete all robots with member_id starting with \"robot_integ_\"\n\t// Using LIKE pattern for cleanup\n\t_, err := qb.Table(tableName).Where(\"member_id\", \"like\", \"robot_integ_%\").Delete()\n\tif err != nil {\n\t\t// Log but don't fail - cleanup errors are not critical\n\t\tt.Logf(\"Warning: cleanup error: %v\", err)\n\t}\n}\n" }, { "path": "agent/robot/manager/interact.go", "content": "package manager\n\nimport (\n\t\"encoding/json\"\n\t\"fmt\"\n\t\"strings\"\n\t\"time\"\n\n\t\"github.com/yaoapp/kun/log\"\n\tagentcontext \"github.com/yaoapp/yao/agent/context\"\n\t\"github.com/yaoapp/yao/agent/output/message\"\n\trobotevents \"github.com/yaoapp/yao/agent/robot/events\"\n\t\"github.com/yaoapp/yao/agent/robot/executor/standard\"\n\t\"github.com/yaoapp/yao/agent/robot/pool\"\n\t\"github.com/yaoapp/yao/agent/robot/store\"\n\t\"github.com/yaoapp/yao/agent/robot/types\"\n\t\"github.com/yaoapp/yao/agent/robot/utils\"\n\t\"github.com/yaoapp/yao/event\"\n)\n\n// executeResume resumes a suspended execution using the Manager's shared executor.\n// This avoids creating orphan Executor instances with independent counters.\nfunc (m *Manager) executeResume(ctx *types.Context, execID, reply string) error {\n\treturn m.executor.Resume(types.NewContext(ctx.Context, ctx.Auth), execID, reply)\n}\n\n// InteractRequest represents a unified interaction with a robot (Manager layer).\ntype InteractRequest struct {\n\tExecutionID string `json:\"execution_id,omitempty\"`\n\tTaskID string `json:\"task_id,omitempty\"`\n\tSource types.InteractSource `json:\"source,omitempty\"`\n\tMessage string `json:\"message\"`\n\tAction string `json:\"action,omitempty\"`\n}\n\n// InteractResponse is the result of an interaction.\ntype InteractResponse struct {\n\tExecutionID string `json:\"execution_id,omitempty\"`\n\tStatus string `json:\"status\"`\n\tMessage string `json:\"message,omitempty\"`\n\tChatID string `json:\"chat_id,omitempty\"`\n\tReply string `json:\"reply,omitempty\"`\n\tWaitForMore bool `json:\"wait_for_more,omitempty\"`\n}\n\n// CancelExecution cancels a waiting/confirming execution.\nfunc (m *Manager) CancelExecution(ctx *types.Context, execID string) error {\n\tm.mu.RLock()\n\tif !m.started {\n\t\tm.mu.RUnlock()\n\t\treturn fmt.Errorf(\"manager not started\")\n\t}\n\tm.mu.RUnlock()\n\n\texecStore := store.NewExecutionStore()\n\trecord, err := execStore.Get(ctx.Context, execID)\n\tif err != nil {\n\t\treturn fmt.Errorf(\"execution not found: %s\", execID)\n\t}\n\tif record == nil {\n\t\treturn fmt.Errorf(\"execution not found: %s\", execID)\n\t}\n\n\tif record.Status != types.ExecWaiting && record.Status != types.ExecConfirming {\n\t\treturn fmt.Errorf(\"execution %s is in status %s, only waiting/confirming can be cancelled\", execID, record.Status)\n\t}\n\n\tif err := execStore.UpdateStatus(ctx.Context, execID, types.ExecCancelled, \"cancelled by user\"); err != nil {\n\t\treturn fmt.Errorf(\"failed to cancel execution: %w\", err)\n\t}\n\n\tm.execController.Untrack(execID)\n\tif robot := m.cache.Get(record.MemberID); robot != nil {\n\t\trobot.RemoveExecution(execID)\n\t}\n\n\tevent.Push(ctx.Context, robotevents.ExecCancelled, robotevents.ExecPayload{\n\t\tExecutionID: execID,\n\t\tMemberID: record.MemberID,\n\t\tTeamID: record.TeamID,\n\t\tStatus: string(types.ExecCancelled),\n\t\tChatID: record.ChatID,\n\t})\n\n\treturn nil\n}\n\n// HandleInteract processes all human-robot interactions through a unified entry point.\n//\n// Routing logic (§16.37):\n// - No execution_id: new interaction → createConfirmingExecution → Host Agent (assign)\n// - execution_id with status=confirming: Host Agent (assign) → processHostAction\n// - execution_id with status=waiting: Host Agent (clarify) → processHostAction\n// - execution_id with status=running: Host Agent (guide) → processHostAction\nfunc (m *Manager) HandleInteract(ctx *types.Context, memberID string, req *InteractRequest) (*InteractResponse, error) {\n\tm.mu.RLock()\n\tif !m.started {\n\t\tm.mu.RUnlock()\n\t\treturn nil, fmt.Errorf(\"manager not started\")\n\t}\n\tm.mu.RUnlock()\n\n\tif memberID == \"\" {\n\t\treturn nil, fmt.Errorf(\"member_id is required\")\n\t}\n\tif req == nil || req.Message == \"\" {\n\t\treturn nil, fmt.Errorf(\"message is required\")\n\t}\n\n\trobot, _, err := m.getOrLoadRobot(ctx, memberID)\n\tif err != nil {\n\t\treturn nil, fmt.Errorf(\"robot not found: %w\", err)\n\t}\n\n\texecStore := store.NewExecutionStore()\n\n\t// No execution_id → create a new confirming execution\n\tif req.ExecutionID == \"\" {\n\t\treturn m.handleNewInteraction(ctx, robot, req, execStore)\n\t}\n\n\t// Existing execution_id → load and route by status\n\trecord, err := execStore.Get(ctx.Context, req.ExecutionID)\n\tif err != nil {\n\t\treturn nil, fmt.Errorf(\"execution not found: %s\", req.ExecutionID)\n\t}\n\n\tswitch record.Status {\n\tcase types.ExecConfirming:\n\t\treturn m.handleConfirmingInteraction(ctx, robot, record, req, execStore)\n\tcase types.ExecWaiting:\n\t\treturn m.handleWaitingInteraction(ctx, robot, record, req, execStore)\n\tcase types.ExecRunning:\n\t\tif record.WaitingTaskID == \"\" {\n\t\t\treturn &InteractResponse{\n\t\t\t\tExecutionID: record.ExecutionID,\n\t\t\t\tStatus: \"rejected\",\n\t\t\t\tMessage: \"Execution is running and not waiting for input\",\n\t\t\t}, nil\n\t\t}\n\t\treturn m.handleRunningInteraction(ctx, robot, record, req, execStore)\n\tdefault:\n\t\treturn nil, fmt.Errorf(\"execution %s is in status %s, cannot interact\", req.ExecutionID, record.Status)\n\t}\n}\n\n// handleNewInteraction creates a confirming execution and calls Host Agent with \"assign\" scenario.\nfunc (m *Manager) handleNewInteraction(ctx *types.Context, robot *types.Robot, req *InteractRequest, execStore *store.ExecutionStore) (*InteractResponse, error) {\n\texec, chatID, err := m.createConfirmingExecution(ctx, robot, req, execStore)\n\tif err != nil {\n\t\treturn nil, fmt.Errorf(\"failed to create confirming execution: %w\", err)\n\t}\n\n\thostOutput, err := m.callHostAgentForScenario(ctx, robot, \"assign\", req.Message, nil, chatID)\n\tif err != nil {\n\t\tlog.Warn(\"Host Agent call failed, using direct assign: %v\", err)\n\t\treturn m.directAssign(ctx, robot, exec, req, execStore)\n\t}\n\n\tresp, err := m.processHostAction(ctx, robot, exec, hostOutput, execStore)\n\tif err != nil {\n\t\treturn nil, err\n\t}\n\tresp.ExecutionID = exec.ExecutionID\n\tresp.ChatID = chatID\n\treturn resp, nil\n}\n\n// handleConfirmingInteraction continues a confirming flow with Host Agent.\nfunc (m *Manager) handleConfirmingInteraction(ctx *types.Context, robot *types.Robot, record *store.ExecutionRecord, req *InteractRequest, execStore *store.ExecutionStore) (*InteractResponse, error) {\n\thostCtx := m.buildHostContext(robot, record, nil)\n\thostOutput, err := m.callHostAgentForScenario(ctx, robot, \"assign\", req.Message, hostCtx, record.ChatID)\n\tif err != nil {\n\t\tlog.Warn(\"Host Agent call failed during confirming: %v\", err)\n\t\treturn &InteractResponse{\n\t\t\tExecutionID: record.ExecutionID,\n\t\t\tStatus: \"error\",\n\t\t\tMessage: fmt.Sprintf(\"Host Agent failed: %v\", err),\n\t\t}, nil\n\t}\n\n\tresp, err := m.processHostAction(ctx, robot, record, hostOutput, execStore)\n\tif err != nil {\n\t\treturn nil, err\n\t}\n\tresp.ExecutionID = record.ExecutionID\n\tresp.ChatID = record.ChatID\n\treturn resp, nil\n}\n\n// handleWaitingInteraction processes input for a waiting (suspended) execution.\nfunc (m *Manager) handleWaitingInteraction(ctx *types.Context, robot *types.Robot, record *store.ExecutionRecord, req *InteractRequest, execStore *store.ExecutionStore) (*InteractResponse, error) {\n\twaitingTask := m.findWaitingTask(record)\n\thostCtx := m.buildHostContext(robot, record, waitingTask)\n\n\thostOutput, err := m.callHostAgentForScenario(ctx, robot, \"clarify\", req.Message, hostCtx, record.ChatID)\n\tif err != nil {\n\t\tlog.Warn(\"Host Agent call failed during clarify, falling back to direct resume: %v\", err)\n\t\treturn m.directResume(ctx, record, req)\n\t}\n\n\tresp, err := m.processHostAction(ctx, robot, record, hostOutput, execStore)\n\tif err != nil {\n\t\treturn nil, err\n\t}\n\tresp.ExecutionID = record.ExecutionID\n\tresp.ChatID = record.ChatID\n\treturn resp, nil\n}\n\n// handleRunningInteraction allows guidance for a running execution.\nfunc (m *Manager) handleRunningInteraction(ctx *types.Context, robot *types.Robot, record *store.ExecutionRecord, req *InteractRequest, execStore *store.ExecutionStore) (*InteractResponse, error) {\n\thostCtx := m.buildHostContext(robot, record, nil)\n\thostOutput, err := m.callHostAgentForScenario(ctx, robot, \"guide\", req.Message, hostCtx, record.ChatID)\n\tif err != nil {\n\t\treturn &InteractResponse{\n\t\t\tExecutionID: record.ExecutionID,\n\t\t\tStatus: \"acknowledged\",\n\t\t\tMessage: \"Guidance noted (Host Agent unavailable)\",\n\t\t}, nil\n\t}\n\n\tresp, err := m.processHostAction(ctx, robot, record, hostOutput, execStore)\n\tif err != nil {\n\t\treturn nil, err\n\t}\n\tresp.ExecutionID = record.ExecutionID\n\tresp.ChatID = record.ChatID\n\treturn resp, nil\n}\n\n// ==================== Helper Methods ====================\n\n// createConfirmingExecution creates a new execution in \"confirming\" status.\nfunc (m *Manager) createConfirmingExecution(ctx *types.Context, robot *types.Robot, req *InteractRequest, execStore *store.ExecutionStore) (*store.ExecutionRecord, string, error) {\n\texecID := pool.GenerateExecID()\n\tchatID := fmt.Sprintf(\"robot_%s_%s\", robot.MemberID, execID)\n\tnow := time.Now()\n\n\trecord := &store.ExecutionRecord{\n\t\tExecutionID: execID,\n\t\tMemberID: robot.MemberID,\n\t\tTeamID: robot.TeamID,\n\t\tTriggerType: types.TriggerHuman,\n\t\tStatus: types.ExecConfirming,\n\t\tPhase: types.PhaseGoals,\n\t\tChatID: chatID,\n\t\tInput: &types.TriggerInput{\n\t\t\tAction: types.ActionTaskAdd,\n\t\t\tMessages: []agentcontext.Message{{Role: \"user\", Content: req.Message}},\n\t\t\tUserID: ctx.UserID(),\n\t\t},\n\t\tStartTime: &now,\n\t}\n\n\tif err := execStore.Save(ctx.Context, record); err != nil {\n\t\treturn nil, \"\", fmt.Errorf(\"failed to save confirming execution: %w\", err)\n\t}\n\n\treturn record, chatID, nil\n}\n\n// buildHostContext builds the HostContext for Host Agent calls.\nfunc (m *Manager) buildHostContext(robot *types.Robot, record *store.ExecutionRecord, waitingTask *types.Task) *types.HostContext {\n\thostCtx := &types.HostContext{\n\t\tRobotStatus: m.buildRobotStatusSnapshot(robot),\n\t}\n\tif record.Goals != nil {\n\t\thostCtx.Goals = record.Goals\n\t}\n\tif len(record.Tasks) > 0 {\n\t\thostCtx.Tasks = record.Tasks\n\t}\n\tif waitingTask != nil {\n\t\thostCtx.CurrentTask = waitingTask\n\t}\n\tif record.WaitingQuestion != \"\" {\n\t\thostCtx.AgentReply = record.WaitingQuestion\n\t}\n\treturn hostCtx\n}\n\n// buildRobotStatusSnapshot builds a status snapshot for the Host Agent.\nfunc (m *Manager) buildRobotStatusSnapshot(robot *types.Robot) *types.RobotStatusSnapshot {\n\tif robot == nil {\n\t\treturn nil\n\t}\n\tsnapshot := &types.RobotStatusSnapshot{\n\t\tMemberID: robot.MemberID,\n\t\tStatus: robot.Status,\n\t\tActiveCount: robot.ActiveCount(),\n\t\tWaitingCount: robot.WaitingCount(),\n\t\tMaxQuota: robot.MaxQuota(),\n\t\tActiveExecs: robot.ListExecutionBriefs(),\n\t}\n\tif m.pool != nil {\n\t\tsnapshot.QueuedCount = m.pool.QueueSize()\n\t}\n\treturn snapshot\n}\n\n// findWaitingTask finds the task that is currently waiting for input.\nfunc (m *Manager) findWaitingTask(record *store.ExecutionRecord) *types.Task {\n\tif record.WaitingTaskID == \"\" {\n\t\treturn nil\n\t}\n\tfor i := range record.Tasks {\n\t\tif record.Tasks[i].ID == record.WaitingTaskID {\n\t\t\treturn &record.Tasks[i]\n\t\t}\n\t}\n\treturn nil\n}\n\n// callHostAgentForScenario calls the Host Agent with a given scenario.\nfunc (m *Manager) callHostAgentForScenario(ctx *types.Context, robot *types.Robot, scenario string, message string, hostCtx *types.HostContext, chatID string) (*types.HostOutput, error) {\n\tagentID := \"\"\n\tif robot.Config != nil && robot.Config.Resources != nil {\n\t\tagentID = robot.Config.Resources.GetPhaseAgent(types.PhaseHost)\n\t}\n\tif agentID == \"\" {\n\t\treturn nil, fmt.Errorf(\"no Host Agent configured for robot %s\", robot.MemberID)\n\t}\n\n\treturn m.callHostAgent(ctx, agentID, &types.HostInput{\n\t\tScenario: scenario,\n\t\tMessages: []agentcontext.Message{{Role: \"user\", Content: message}},\n\t\tContext: hostCtx,\n\t}, chatID)\n}\n\n// callHostAgent calls the Host Agent assistant and parses output.\nfunc (m *Manager) callHostAgent(ctx *types.Context, agentID string, input *types.HostInput, chatID string) (*types.HostOutput, error) {\n\tinputJSON, err := json.Marshal(input)\n\tif err != nil {\n\t\treturn nil, fmt.Errorf(\"failed to marshal host input: %w\", err)\n\t}\n\n\tcaller := standard.NewConversationCaller(chatID)\n\tresult, err := caller.CallWithMessages(ctx, agentID, string(inputJSON))\n\tif err != nil {\n\t\treturn nil, fmt.Errorf(\"host agent (%s) call failed: %w\", agentID, err)\n\t}\n\n\treturn m.parseHostAgentResult(result)\n}\n\n// parseHostAgentResult inspects the agent result to determine if it is an action\n// decision (JSON with \"action\" field) or a conversational reply (natural language).\nfunc (m *Manager) parseHostAgentResult(result *standard.CallResult) (*types.HostOutput, error) {\n\tdata, err := result.GetJSON()\n\tif err == nil {\n\t\toutput := &types.HostOutput{}\n\t\traw, _ := json.Marshal(data)\n\t\tif err := json.Unmarshal(raw, output); err == nil && output.Action != \"\" {\n\t\t\treturn output, nil\n\t\t}\n\t}\n\n\treturn &types.HostOutput{\n\t\tReply: result.GetText(),\n\t\tWaitForMore: true,\n\t}, nil\n}\n\n// processHostAction processes the output from Host Agent and takes the appropriate action.\nfunc (m *Manager) processHostAction(ctx *types.Context, robot *types.Robot, record *store.ExecutionRecord, output *types.HostOutput, execStore *store.ExecutionStore) (*InteractResponse, error) {\n\tresp := &InteractResponse{\n\t\tReply: output.Reply,\n\t\tWaitForMore: output.WaitForMore,\n\t}\n\n\tif output.WaitForMore {\n\t\tresp.Status = \"waiting_for_more\"\n\t\tresp.Message = output.Reply\n\t\treturn resp, nil\n\t}\n\n\tswitch output.Action {\n\tcase types.HostActionConfirm:\n\t\tif err := m.advanceExecution(ctx, robot, record, execStore); err != nil {\n\t\t\treturn nil, fmt.Errorf(\"failed to advance execution: %w\", err)\n\t\t}\n\t\tresp.Status = \"confirmed\"\n\t\tresp.Message = \"Execution confirmed and started\"\n\n\tcase types.HostActionAdjust:\n\t\tif err := m.adjustExecution(ctx, record, output.ActionData, execStore); err != nil {\n\t\t\treturn nil, fmt.Errorf(\"failed to adjust execution: %w\", err)\n\t\t}\n\t\tresp.Status = \"adjusted\"\n\t\tresp.Message = \"Execution plan adjusted\"\n\n\tcase types.HostActionAddTask:\n\t\tif err := m.injectTask(ctx, record, output.ActionData, execStore); err != nil {\n\t\t\treturn nil, fmt.Errorf(\"failed to inject task: %w\", err)\n\t\t}\n\t\tresp.Status = \"task_added\"\n\t\tresp.Message = \"New task injected\"\n\n\tcase types.HostActionSkip:\n\t\tif err := m.skipWaitingTask(ctx, record, execStore); err != nil {\n\t\t\treturn nil, fmt.Errorf(\"failed to skip task: %w\", err)\n\t\t}\n\t\tresp.Status = \"task_skipped\"\n\t\tresp.Message = \"Waiting task skipped\"\n\n\tcase types.HostActionInjectCtx:\n\t\tif err := m.resumeWithContext(ctx, record, output.ActionData, execStore); err != nil {\n\t\t\tif err == types.ErrExecutionSuspended {\n\t\t\t\tresp.Status = \"waiting\"\n\t\t\t\tresp.Message = \"Execution suspended again\"\n\t\t\t\treturn resp, nil\n\t\t\t}\n\t\t\treturn nil, fmt.Errorf(\"failed to resume with context: %w\", err)\n\t\t}\n\t\tresp.Status = \"resumed\"\n\t\tresp.Message = \"Execution resumed with additional context\"\n\n\tcase types.HostActionCancel:\n\t\tif err := m.CancelExecution(ctx, record.ExecutionID); err != nil {\n\t\t\treturn nil, fmt.Errorf(\"failed to cancel execution: %w\", err)\n\t\t}\n\t\tresp.Status = \"cancelled\"\n\t\tresp.Message = \"Execution cancelled\"\n\n\tdefault:\n\t\tresp.Status = \"acknowledged\"\n\t\tresp.Message = output.Reply\n\t}\n\n\treturn resp, nil\n}\n\n// advanceExecution moves a confirming execution to running.\nfunc (m *Manager) advanceExecution(ctx *types.Context, robot *types.Robot, record *store.ExecutionRecord, execStore *store.ExecutionStore) error {\n\tif err := execStore.UpdateStatus(ctx.Context, record.ExecutionID, types.ExecRunning, \"\"); err != nil {\n\t\treturn err\n\t}\n\n\tctrlExec := m.execController.Track(record.ExecutionID, record.MemberID, record.TeamID)\n\texecCtx := types.NewContext(ctrlExec.Context(), ctx.Auth)\n\n\ttriggerInput := record.Input\n\t_, err := m.pool.SubmitWithID(execCtx, robot, types.TriggerHuman, triggerInput, record.ExecutionID, ctrlExec)\n\tif err != nil {\n\t\tm.execController.Untrack(record.ExecutionID)\n\t\treturn fmt.Errorf(\"failed to submit execution to pool: %w\", err)\n\t}\n\n\treturn nil\n}\n\n// adjustExecution adjusts goals/tasks based on Host Agent output.\nfunc (m *Manager) adjustExecution(ctx *types.Context, record *store.ExecutionRecord, actionData interface{}, execStore *store.ExecutionStore) error {\n\tif actionData == nil {\n\t\treturn nil\n\t}\n\n\tdata, ok := actionData.(map[string]interface{})\n\tif !ok {\n\t\traw, err := json.Marshal(actionData)\n\t\tif err != nil {\n\t\t\treturn nil\n\t\t}\n\t\tjson.Unmarshal(raw, &data)\n\t}\n\n\tif goalsContent, ok := data[\"goals\"].(string); ok && goalsContent != \"\" {\n\t\trecord.Goals = &types.Goals{Content: goalsContent}\n\t}\n\n\tif tasksRaw, ok := data[\"tasks\"]; ok {\n\t\traw, _ := json.Marshal(tasksRaw)\n\t\tvar tasks []types.Task\n\t\tif err := json.Unmarshal(raw, &tasks); err == nil {\n\t\t\trecord.Tasks = tasks\n\t\t}\n\t}\n\n\treturn execStore.Save(ctx.Context, record)\n}\n\n// injectTask adds a new task to the execution's task list.\nfunc (m *Manager) injectTask(ctx *types.Context, record *store.ExecutionRecord, actionData interface{}, execStore *store.ExecutionStore) error {\n\tif actionData == nil {\n\t\treturn fmt.Errorf(\"task data is required\")\n\t}\n\n\traw, err := json.Marshal(actionData)\n\tif err != nil {\n\t\treturn fmt.Errorf(\"invalid task data: %w\", err)\n\t}\n\n\tvar newTask types.Task\n\tif err := json.Unmarshal(raw, &newTask); err != nil {\n\t\treturn fmt.Errorf(\"failed to parse task: %w\", err)\n\t}\n\n\tif newTask.ID == \"\" {\n\t\tnewTask.ID = fmt.Sprintf(\"injected-%s\", utils.NewID()[:8])\n\t}\n\tnewTask.Status = types.TaskPending\n\n\trecord.Tasks = append(record.Tasks, newTask)\n\treturn execStore.Save(ctx.Context, record)\n}\n\n// skipWaitingTask skips the currently waiting task and resumes execution.\nfunc (m *Manager) skipWaitingTask(ctx *types.Context, record *store.ExecutionRecord, execStore *store.ExecutionStore) error {\n\tif record.WaitingTaskID == \"\" {\n\t\treturn fmt.Errorf(\"no task is waiting\")\n\t}\n\n\tfor i := range record.Tasks {\n\t\tif record.Tasks[i].ID == record.WaitingTaskID {\n\t\t\trecord.Tasks[i].Status = types.TaskSkipped\n\t\t\tbreak\n\t\t}\n\t}\n\n\terr := m.executeResume(ctx, record.ExecutionID, \"__skip__\")\n\tif err != nil && err != types.ErrExecutionSuspended {\n\t\treturn fmt.Errorf(\"failed to resume after skip: %w\", err)\n\t}\n\treturn nil\n}\n\n// resumeWithContext injects context and resumes the waiting execution.\nfunc (m *Manager) resumeWithContext(ctx *types.Context, record *store.ExecutionRecord, actionData interface{}, execStore *store.ExecutionStore) error {\n\treply := \"\"\n\tif actionData != nil {\n\t\tif s, ok := actionData.(string); ok {\n\t\t\treply = s\n\t\t} else if data, ok := actionData.(map[string]interface{}); ok {\n\t\t\tif r, ok := data[\"reply\"].(string); ok {\n\t\t\t\treply = r\n\t\t\t} else {\n\t\t\t\traw, _ := json.Marshal(data)\n\t\t\t\treply = string(raw)\n\t\t\t}\n\t\t}\n\t}\n\n\treturn m.executeResume(ctx, record.ExecutionID, reply)\n}\n\n// directAssign is the fallback when Host Agent is unavailable: directly start execution.\nfunc (m *Manager) directAssign(ctx *types.Context, robot *types.Robot, record *store.ExecutionRecord, req *InteractRequest, execStore *store.ExecutionStore) (*InteractResponse, error) {\n\tif err := m.advanceExecution(ctx, robot, record, execStore); err != nil {\n\t\treturn nil, fmt.Errorf(\"direct assign failed: %w\", err)\n\t}\n\treturn &InteractResponse{\n\t\tExecutionID: record.ExecutionID,\n\t\tStatus: \"confirmed\",\n\t\tMessage: \"Execution started (direct assign)\",\n\t\tChatID: record.ChatID,\n\t}, nil\n}\n\n// directResume is the fallback when Host Agent is unavailable: directly resume.\nfunc (m *Manager) directResume(ctx *types.Context, record *store.ExecutionRecord, req *InteractRequest) (*InteractResponse, error) {\n\terr := m.executeResume(ctx, record.ExecutionID, req.Message)\n\tif err != nil {\n\t\tif err == types.ErrExecutionSuspended {\n\t\t\treturn &InteractResponse{\n\t\t\t\tExecutionID: record.ExecutionID,\n\t\t\t\tStatus: \"waiting\",\n\t\t\t\tMessage: \"Execution suspended again: needs more input\",\n\t\t\t\tChatID: record.ChatID,\n\t\t\t}, nil\n\t\t}\n\t\treturn nil, fmt.Errorf(\"failed to resume execution: %w\", err)\n\t}\n\treturn &InteractResponse{\n\t\tExecutionID: record.ExecutionID,\n\t\tStatus: \"resumed\",\n\t\tMessage: \"Execution resumed and completed successfully\",\n\t\tChatID: record.ChatID,\n\t}, nil\n}\n\n// ==================== Streaming Interact ====================\n\n// HandleInteractStream is the streaming version of HandleInteract.\n// It streams Host Agent text tokens via streamFn while still returning the final InteractResponse.\nfunc (m *Manager) HandleInteractStream(ctx *types.Context, memberID string, req *InteractRequest, streamFn standard.StreamCallback) (*InteractResponse, error) {\n\tm.mu.RLock()\n\tif !m.started {\n\t\tm.mu.RUnlock()\n\t\treturn nil, fmt.Errorf(\"manager not started\")\n\t}\n\tm.mu.RUnlock()\n\n\tif memberID == \"\" {\n\t\treturn nil, fmt.Errorf(\"member_id is required\")\n\t}\n\tif req == nil || req.Message == \"\" {\n\t\treturn nil, fmt.Errorf(\"message is required\")\n\t}\n\n\trobot, _, err := m.getOrLoadRobot(ctx, memberID)\n\tif err != nil {\n\t\treturn nil, fmt.Errorf(\"robot not found: %w\", err)\n\t}\n\n\texecStore := store.NewExecutionStore()\n\n\tif req.ExecutionID == \"\" {\n\t\treturn m.handleNewInteractionStream(ctx, robot, req, execStore, streamFn)\n\t}\n\n\trecord, err := execStore.Get(ctx.Context, req.ExecutionID)\n\tif err != nil {\n\t\treturn nil, fmt.Errorf(\"execution not found: %s\", req.ExecutionID)\n\t}\n\n\tswitch record.Status {\n\tcase types.ExecConfirming:\n\t\treturn m.handleConfirmingInteractionStream(ctx, robot, record, req, execStore, streamFn)\n\tcase types.ExecWaiting:\n\t\treturn m.handleWaitingInteractionStream(ctx, robot, record, req, execStore, streamFn)\n\tcase types.ExecRunning:\n\t\tif record.WaitingTaskID == \"\" {\n\t\t\treturn &InteractResponse{\n\t\t\t\tExecutionID: record.ExecutionID,\n\t\t\t\tStatus: \"rejected\",\n\t\t\t\tMessage: \"Execution is running and not waiting for input\",\n\t\t\t}, nil\n\t\t}\n\t\treturn m.handleRunningInteractionStream(ctx, robot, record, req, execStore, streamFn)\n\tdefault:\n\t\treturn nil, fmt.Errorf(\"execution %s is in status %s, cannot interact\", req.ExecutionID, record.Status)\n\t}\n}\n\nfunc (m *Manager) handleNewInteractionStream(ctx *types.Context, robot *types.Robot, req *InteractRequest, execStore *store.ExecutionStore, streamFn standard.StreamCallback) (*InteractResponse, error) {\n\texec, chatID, err := m.createConfirmingExecution(ctx, robot, req, execStore)\n\tif err != nil {\n\t\treturn nil, fmt.Errorf(\"failed to create confirming execution: %w\", err)\n\t}\n\n\thostOutput, err := m.callHostAgentForScenarioStream(ctx, robot, \"assign\", req.Message, nil, chatID, streamFn)\n\tif err != nil {\n\t\tlog.Warn(\"Host Agent call failed, using direct assign: %v\", err)\n\t\treturn m.directAssign(ctx, robot, exec, req, execStore)\n\t}\n\n\tresp, err := m.processHostAction(ctx, robot, exec, hostOutput, execStore)\n\tif err != nil {\n\t\treturn nil, err\n\t}\n\tresp.ExecutionID = exec.ExecutionID\n\tresp.ChatID = chatID\n\treturn resp, nil\n}\n\nfunc (m *Manager) handleConfirmingInteractionStream(ctx *types.Context, robot *types.Robot, record *store.ExecutionRecord, req *InteractRequest, execStore *store.ExecutionStore, streamFn standard.StreamCallback) (*InteractResponse, error) {\n\thostCtx := m.buildHostContext(robot, record, nil)\n\thostOutput, err := m.callHostAgentForScenarioStream(ctx, robot, \"assign\", req.Message, hostCtx, record.ChatID, streamFn)\n\tif err != nil {\n\t\tlog.Warn(\"Host Agent call failed during confirming: %v\", err)\n\t\treturn &InteractResponse{\n\t\t\tExecutionID: record.ExecutionID,\n\t\t\tStatus: \"error\",\n\t\t\tMessage: fmt.Sprintf(\"Host Agent failed: %v\", err),\n\t\t}, nil\n\t}\n\n\tresp, err := m.processHostAction(ctx, robot, record, hostOutput, execStore)\n\tif err != nil {\n\t\treturn nil, err\n\t}\n\tresp.ExecutionID = record.ExecutionID\n\tresp.ChatID = record.ChatID\n\treturn resp, nil\n}\n\nfunc (m *Manager) handleWaitingInteractionStream(ctx *types.Context, robot *types.Robot, record *store.ExecutionRecord, req *InteractRequest, execStore *store.ExecutionStore, streamFn standard.StreamCallback) (*InteractResponse, error) {\n\twaitingTask := m.findWaitingTask(record)\n\thostCtx := m.buildHostContext(robot, record, waitingTask)\n\n\thostOutput, err := m.callHostAgentForScenarioStream(ctx, robot, \"clarify\", req.Message, hostCtx, record.ChatID, streamFn)\n\tif err != nil {\n\t\tlog.Warn(\"Host Agent call failed during clarify, falling back to direct resume: %v\", err)\n\t\treturn m.directResume(ctx, record, req)\n\t}\n\n\tresp, err := m.processHostAction(ctx, robot, record, hostOutput, execStore)\n\tif err != nil {\n\t\treturn nil, err\n\t}\n\tresp.ExecutionID = record.ExecutionID\n\tresp.ChatID = record.ChatID\n\treturn resp, nil\n}\n\nfunc (m *Manager) handleRunningInteractionStream(ctx *types.Context, robot *types.Robot, record *store.ExecutionRecord, req *InteractRequest, execStore *store.ExecutionStore, streamFn standard.StreamCallback) (*InteractResponse, error) {\n\thostCtx := m.buildHostContext(robot, record, nil)\n\thostOutput, err := m.callHostAgentForScenarioStream(ctx, robot, \"guide\", req.Message, hostCtx, record.ChatID, streamFn)\n\tif err != nil {\n\t\treturn &InteractResponse{\n\t\t\tExecutionID: record.ExecutionID,\n\t\t\tStatus: \"acknowledged\",\n\t\t\tMessage: \"Guidance noted (Host Agent unavailable)\",\n\t\t}, nil\n\t}\n\n\tresp, err := m.processHostAction(ctx, robot, record, hostOutput, execStore)\n\tif err != nil {\n\t\treturn nil, err\n\t}\n\tresp.ExecutionID = record.ExecutionID\n\tresp.ChatID = record.ChatID\n\treturn resp, nil\n}\n\nfunc (m *Manager) callHostAgentForScenarioStream(ctx *types.Context, robot *types.Robot, scenario string, msg string, hostCtx *types.HostContext, chatID string, streamFn standard.StreamCallback) (*types.HostOutput, error) {\n\tagentID := \"\"\n\tif robot.Config != nil && robot.Config.Resources != nil {\n\t\tagentID = robot.Config.Resources.GetPhaseAgent(types.PhaseHost)\n\t}\n\tif agentID == \"\" {\n\t\treturn nil, fmt.Errorf(\"no Host Agent configured for robot %s\", robot.MemberID)\n\t}\n\n\treturn m.callHostAgentStream(ctx, agentID, &types.HostInput{\n\t\tScenario: scenario,\n\t\tMessages: []agentcontext.Message{{Role: \"user\", Content: msg}},\n\t\tContext: hostCtx,\n\t}, chatID, streamFn)\n}\n\nfunc (m *Manager) callHostAgentStream(ctx *types.Context, agentID string, input *types.HostInput, chatID string, streamFn standard.StreamCallback) (*types.HostOutput, error) {\n\tinputJSON, err := json.Marshal(input)\n\tif err != nil {\n\t\treturn nil, fmt.Errorf(\"failed to marshal host input: %w\", err)\n\t}\n\n\tcaller := standard.NewConversationCaller(chatID)\n\tresult, err := caller.CallWithMessagesStream(ctx, agentID, string(inputJSON), streamFn)\n\tif err != nil {\n\t\treturn nil, fmt.Errorf(\"host agent (%s) call failed: %w\", agentID, err)\n\t}\n\n\treturn m.parseHostAgentResult(result)\n}\n\n// ==================== Raw Message Streaming (CUI Protocol) ====================\n\n// HandleInteractStreamRaw is the CUI-protocol-aligned streaming version of HandleInteract.\n// It passes raw message.Message objects directly to the onMessage callback, preserving all\n// CUI protocol fields for direct SSE passthrough to the frontend.\nfunc (m *Manager) HandleInteractStreamRaw(ctx *types.Context, memberID string, req *InteractRequest, onMessage agentcontext.OnMessageFunc) (*InteractResponse, error) {\n\tm.mu.RLock()\n\tif !m.started {\n\t\tm.mu.RUnlock()\n\t\treturn nil, fmt.Errorf(\"manager not started\")\n\t}\n\tm.mu.RUnlock()\n\n\tif memberID == \"\" {\n\t\treturn nil, fmt.Errorf(\"member_id is required\")\n\t}\n\tif req == nil || req.Message == \"\" {\n\t\treturn nil, fmt.Errorf(\"message is required\")\n\t}\n\n\trobot, _, err := m.getOrLoadRobot(ctx, memberID)\n\tif err != nil {\n\t\treturn nil, fmt.Errorf(\"robot not found: %w\", err)\n\t}\n\n\texecStore := store.NewExecutionStore()\n\n\tif req.ExecutionID == \"\" {\n\t\treturn m.handleNewInteractionStreamRaw(ctx, robot, req, execStore, onMessage)\n\t}\n\n\trecord, err := execStore.Get(ctx.Context, req.ExecutionID)\n\tif err != nil {\n\t\treturn nil, fmt.Errorf(\"execution not found: %s\", req.ExecutionID)\n\t}\n\n\tswitch record.Status {\n\tcase types.ExecConfirming:\n\t\treturn m.handleConfirmingInteractionStreamRaw(ctx, robot, record, req, execStore, onMessage)\n\tcase types.ExecWaiting:\n\t\treturn m.handleWaitingInteractionStreamRaw(ctx, robot, record, req, execStore, onMessage)\n\tcase types.ExecRunning:\n\t\tif record.WaitingTaskID == \"\" {\n\t\t\treturn &InteractResponse{\n\t\t\t\tExecutionID: record.ExecutionID,\n\t\t\t\tStatus: \"rejected\",\n\t\t\t\tMessage: \"Execution is running and not waiting for input\",\n\t\t\t}, nil\n\t\t}\n\t\treturn m.handleRunningInteractionStreamRaw(ctx, robot, record, req, execStore, onMessage)\n\tdefault:\n\t\treturn nil, fmt.Errorf(\"execution %s is in status %s, cannot interact\", req.ExecutionID, record.Status)\n\t}\n}\n\nfunc (m *Manager) handleNewInteractionStreamRaw(ctx *types.Context, robot *types.Robot, req *InteractRequest, execStore *store.ExecutionStore, onMessage agentcontext.OnMessageFunc) (*InteractResponse, error) {\n\texec, chatID, err := m.createConfirmingExecution(ctx, robot, req, execStore)\n\tif err != nil {\n\t\treturn nil, fmt.Errorf(\"failed to create confirming execution: %w\", err)\n\t}\n\n\thostOutput, err := m.callHostAgentForScenarioStreamRaw(ctx, robot, \"assign\", req.Message, nil, chatID, onMessage)\n\tif err != nil {\n\t\tlog.Warn(\"Host Agent call failed, using direct assign: %v\", err)\n\t\treturn m.directAssign(ctx, robot, exec, req, execStore)\n\t}\n\n\tresp, err := m.processHostAction(ctx, robot, exec, hostOutput, execStore)\n\tif err != nil {\n\t\treturn nil, err\n\t}\n\tresp.ExecutionID = exec.ExecutionID\n\tresp.ChatID = chatID\n\treturn resp, nil\n}\n\nfunc (m *Manager) handleConfirmingInteractionStreamRaw(ctx *types.Context, robot *types.Robot, record *store.ExecutionRecord, req *InteractRequest, execStore *store.ExecutionStore, onMessage agentcontext.OnMessageFunc) (*InteractResponse, error) {\n\thostCtx := m.buildHostContext(robot, record, nil)\n\thostOutput, err := m.callHostAgentForScenarioStreamRaw(ctx, robot, \"assign\", req.Message, hostCtx, record.ChatID, onMessage)\n\tif err != nil {\n\t\tlog.Warn(\"Host Agent call failed during confirming: %v\", err)\n\t\treturn &InteractResponse{\n\t\t\tExecutionID: record.ExecutionID,\n\t\t\tStatus: \"error\",\n\t\t\tMessage: fmt.Sprintf(\"Host Agent failed: %v\", err),\n\t\t}, nil\n\t}\n\n\tresp, err := m.processHostAction(ctx, robot, record, hostOutput, execStore)\n\tif err != nil {\n\t\treturn nil, err\n\t}\n\tresp.ExecutionID = record.ExecutionID\n\tresp.ChatID = record.ChatID\n\treturn resp, nil\n}\n\nfunc (m *Manager) handleWaitingInteractionStreamRaw(ctx *types.Context, robot *types.Robot, record *store.ExecutionRecord, req *InteractRequest, execStore *store.ExecutionStore, onMessage agentcontext.OnMessageFunc) (*InteractResponse, error) {\n\twaitingTask := m.findWaitingTask(record)\n\thostCtx := m.buildHostContext(robot, record, waitingTask)\n\n\thostOutput, err := m.callHostAgentForScenarioStreamRaw(ctx, robot, \"clarify\", req.Message, hostCtx, record.ChatID, onMessage)\n\tif err != nil {\n\t\tlog.Warn(\"Host Agent call failed during clarify, falling back to direct resume: %v\", err)\n\t\treturn m.directResume(ctx, record, req)\n\t}\n\n\tresp, err := m.processHostAction(ctx, robot, record, hostOutput, execStore)\n\tif err != nil {\n\t\treturn nil, err\n\t}\n\tresp.ExecutionID = record.ExecutionID\n\tresp.ChatID = record.ChatID\n\treturn resp, nil\n}\n\nfunc (m *Manager) handleRunningInteractionStreamRaw(ctx *types.Context, robot *types.Robot, record *store.ExecutionRecord, req *InteractRequest, execStore *store.ExecutionStore, onMessage agentcontext.OnMessageFunc) (*InteractResponse, error) {\n\thostCtx := m.buildHostContext(robot, record, nil)\n\thostOutput, err := m.callHostAgentForScenarioStreamRaw(ctx, robot, \"guide\", req.Message, hostCtx, record.ChatID, onMessage)\n\tif err != nil {\n\t\treturn &InteractResponse{\n\t\t\tExecutionID: record.ExecutionID,\n\t\t\tStatus: \"acknowledged\",\n\t\t\tMessage: \"Guidance noted (Host Agent unavailable)\",\n\t\t}, nil\n\t}\n\n\tresp, err := m.processHostAction(ctx, robot, record, hostOutput, execStore)\n\tif err != nil {\n\t\treturn nil, err\n\t}\n\tresp.ExecutionID = record.ExecutionID\n\tresp.ChatID = record.ChatID\n\treturn resp, nil\n}\n\nfunc (m *Manager) callHostAgentForScenarioStreamRaw(ctx *types.Context, robot *types.Robot, scenario string, msg string, hostCtx *types.HostContext, chatID string, onMessage agentcontext.OnMessageFunc) (*types.HostOutput, error) {\n\tagentID := \"\"\n\tif robot.Config != nil && robot.Config.Resources != nil {\n\t\tagentID = robot.Config.Resources.GetPhaseAgent(types.PhaseHost)\n\t}\n\tif agentID == \"\" {\n\t\treturn nil, fmt.Errorf(\"no Host Agent configured for robot %s\", robot.MemberID)\n\t}\n\n\treturn m.callHostAgentStreamRaw(ctx, agentID, &types.HostInput{\n\t\tScenario: scenario,\n\t\tMessages: []agentcontext.Message{{Role: \"user\", Content: msg}},\n\t\tContext: hostCtx,\n\t}, chatID, onMessage)\n}\n\n// callHostAgentStreamRaw calls the Host Agent with CUI raw message streaming.\n// It buffers text chunks that look like JSON output (starting with \"{\" or \"```json\")\n// so the frontend never sees raw decision JSON. If the final result is a decision,\n// the buffered chunks are discarded and a clean reply is sent instead. If the\n// result is a normal conversation turn, buffered chunks are flushed through.\nfunc (m *Manager) callHostAgentStreamRaw(ctx *types.Context, agentID string, input *types.HostInput, chatID string, onMessage agentcontext.OnMessageFunc) (*types.HostOutput, error) {\n\tinputJSON, err := json.Marshal(input)\n\tif err != nil {\n\t\treturn nil, fmt.Errorf(\"failed to marshal host input: %w\", err)\n\t}\n\n\tvar (\n\t\tbufferedChunks []*message.Message\n\t\tbuffering bool\n\t\taccumulatedText string\n\t\tlastTextMsgID string\n\t)\n\n\twrappedOnMessage := func(msg *message.Message) int {\n\t\tif msg == nil {\n\t\t\treturn onMessage(msg)\n\t\t}\n\n\t\t// Only intercept text type messages with delta content\n\t\tif msg.Type != message.TypeText || !msg.Delta {\n\t\t\treturn onMessage(msg)\n\t\t}\n\n\t\tif msg.MessageID != \"\" {\n\t\t\tlastTextMsgID = msg.MessageID\n\t\t}\n\n\t\t// Extract the text content from this chunk\n\t\tchunkText := \"\"\n\t\tif msg.Props != nil {\n\t\t\tif c, ok := msg.Props[\"content\"].(string); ok {\n\t\t\t\tchunkText = c\n\t\t\t}\n\t\t}\n\t\taccumulatedText += chunkText\n\n\t\t// Decide whether to buffer: check accumulated text so far\n\t\ttrimmed := strings.TrimSpace(accumulatedText)\n\t\tif !buffering && len(trimmed) > 0 {\n\t\t\tif trimmed[0] == '{' || strings.HasPrefix(trimmed, \"```\") {\n\t\t\t\tbuffering = true\n\t\t\t}\n\t\t}\n\n\t\tif buffering {\n\t\t\tbufferedChunks = append(bufferedChunks, msg)\n\t\t\treturn 0\n\t\t}\n\n\t\treturn onMessage(msg)\n\t}\n\n\tcaller := standard.NewConversationCaller(chatID)\n\tresult, err := caller.CallWithMessagesStreamRaw(ctx, agentID, string(inputJSON), wrappedOnMessage)\n\tif err != nil {\n\t\treturn nil, fmt.Errorf(\"host agent (%s) call failed: %w\", agentID, err)\n\t}\n\n\toutput, err := m.parseHostAgentResult(result)\n\tif err != nil {\n\t\treturn nil, err\n\t}\n\n\tif output.Action != \"\" && lastTextMsgID != \"\" {\n\t\t// Decision detected — discard buffered JSON chunks, send reply text\n\t\tonMessage(&message.Message{\n\t\t\tType: message.TypeText,\n\t\t\tMessageID: lastTextMsgID,\n\t\t\tProps: map[string]interface{}{\"content\": output.Reply},\n\t\t\tDelta: false,\n\t\t})\n\t} else if len(bufferedChunks) > 0 {\n\t\t// Not a decision — flush all buffered chunks to the frontend\n\t\tfor _, chunk := range bufferedChunks {\n\t\t\tif onMessage(chunk) != 0 {\n\t\t\t\tbreak\n\t\t\t}\n\t\t}\n\t}\n\n\treturn output, nil\n}\n" }, { "path": "agent/robot/manager/interact_helpers_test.go", "content": "package manager\n\nimport (\n\t\"context\"\n\t\"encoding/json\"\n\t\"fmt\"\n\t\"testing\"\n\n\t\"github.com/stretchr/testify/assert\"\n\t\"github.com/stretchr/testify/require\"\n\t\"github.com/yaoapp/yao/agent/robot/cache\"\n\t\"github.com/yaoapp/yao/agent/robot/store\"\n\t\"github.com/yaoapp/yao/agent/robot/types\"\n\t\"github.com/yaoapp/yao/agent/testutils\"\n)\n\n// mockExecutor is a minimal Executor for unit testing\ntype mockExecutor struct {\n\tresumeErr error\n}\n\nfunc (m *mockExecutor) ExecuteWithControl(ctx *types.Context, robot *types.Robot, trigger types.TriggerType, data interface{}, execID string, control types.ExecutionControl) (*types.Execution, error) {\n\treturn nil, fmt.Errorf(\"not implemented\")\n}\nfunc (m *mockExecutor) ExecuteWithID(ctx *types.Context, robot *types.Robot, trigger types.TriggerType, data interface{}, execID string) (*types.Execution, error) {\n\treturn nil, fmt.Errorf(\"not implemented\")\n}\nfunc (m *mockExecutor) Execute(ctx *types.Context, robot *types.Robot, trigger types.TriggerType, data interface{}) (*types.Execution, error) {\n\treturn nil, fmt.Errorf(\"not implemented\")\n}\nfunc (m *mockExecutor) Resume(ctx *types.Context, execID string, reply string) error {\n\treturn m.resumeErr\n}\nfunc (m *mockExecutor) ExecCount() int { return 0 }\nfunc (m *mockExecutor) CurrentCount() int { return 0 }\nfunc (m *mockExecutor) Reset() {}\n\n// HL1: createConfirmingExecution\nfunc TestCreateConfirmingExecution(t *testing.T) {\n\tm := &Manager{}\n\n\tt.Run(\"creates record with correct fields\", func(t *testing.T) {\n\t\tif testing.Short() {\n\t\t\tt.Skip(\"Requires database\")\n\t\t}\n\t\ttestutils.Prepare(t)\n\t\tdefer testutils.Clean(t)\n\t\tctx := types.NewContext(context.Background(), nil)\n\t\trobot := &types.Robot{MemberID: \"member-hl1\", TeamID: \"team-hl1\"}\n\t\treq := &InteractRequest{Message: \"do something\"}\n\t\texecStore := store.NewExecutionStore()\n\n\t\trecord, chatID, err := m.createConfirmingExecution(ctx, robot, req, execStore)\n\t\trequire.NoError(t, err)\n\t\tassert.NotEmpty(t, record.ExecutionID)\n\t\tassert.Equal(t, \"member-hl1\", record.MemberID)\n\t\tassert.Equal(t, \"team-hl1\", record.TeamID)\n\t\tassert.Equal(t, types.ExecConfirming, record.Status)\n\t\tassert.Equal(t, types.TriggerHuman, record.TriggerType)\n\t\tassert.Equal(t, types.PhaseGoals, record.Phase)\n\t\tassert.Contains(t, chatID, \"robot_member-hl1_\")\n\t\tassert.Equal(t, chatID, record.ChatID)\n\t\tassert.NotNil(t, record.Input)\n\t\tassert.Equal(t, types.ActionTaskAdd, record.Input.Action)\n\t\tassert.Len(t, record.Input.Messages, 1)\n\t\tassert.Equal(t, \"do something\", record.Input.Messages[0].Content)\n\t\tassert.NotNil(t, record.StartTime)\n\t})\n\n\tt.Run(\"UserID empty when auth is nil\", func(t *testing.T) {\n\t\tif testing.Short() {\n\t\t\tt.Skip(\"Requires database\")\n\t\t}\n\t\ttestutils.Prepare(t)\n\t\tdefer testutils.Clean(t)\n\t\tctx := types.NewContext(context.Background(), nil)\n\t\trobot := &types.Robot{MemberID: \"member-hl1b\", TeamID: \"team-hl1b\"}\n\t\treq := &InteractRequest{Message: \"test\"}\n\t\texecStore := store.NewExecutionStore()\n\n\t\trecord, _, err := m.createConfirmingExecution(ctx, robot, req, execStore)\n\t\trequire.NoError(t, err)\n\t\tassert.Empty(t, record.Input.UserID)\n\t})\n}\n\n// HL2-HL4: adjustExecution\nfunc TestAdjustExecution(t *testing.T) {\n\tm := &Manager{}\n\n\tt.Run(\"adjusts goals from string\", func(t *testing.T) {\n\t\tif testing.Short() {\n\t\t\tt.Skip(\"Requires database\")\n\t\t}\n\t\ttestutils.Prepare(t)\n\t\tdefer testutils.Clean(t)\n\t\tctx := types.NewContext(context.Background(), nil)\n\t\trecord := &store.ExecutionRecord{\n\t\t\tExecutionID: \"exec-hl2\",\n\t\t\tMemberID: \"member-hl2\",\n\t\t\tTriggerType: types.TriggerHuman,\n\t\t\tStatus: types.ExecPending,\n\t\t\tPhase: types.PhaseInspiration,\n\t\t}\n\t\texecStore := store.NewExecutionStore()\n\t\trequire.NoError(t, execStore.Save(ctx.Context, record))\n\n\t\tactionData := map[string]interface{}{\"goals\": \"updated goals content\"}\n\t\terr := m.adjustExecution(ctx, record, actionData, execStore)\n\t\trequire.NoError(t, err)\n\t\trequire.NotNil(t, record.Goals)\n\t\tassert.Equal(t, \"updated goals content\", record.Goals.Content)\n\t})\n\n\tt.Run(\"adjusts tasks from array\", func(t *testing.T) {\n\t\tif testing.Short() {\n\t\t\tt.Skip(\"Requires database\")\n\t\t}\n\t\ttestutils.Prepare(t)\n\t\tdefer testutils.Clean(t)\n\t\tctx := types.NewContext(context.Background(), nil)\n\t\trecord := &store.ExecutionRecord{\n\t\t\tExecutionID: \"exec-hl3\",\n\t\t\tMemberID: \"member-hl3\",\n\t\t\tTriggerType: types.TriggerHuman,\n\t\t\tStatus: types.ExecPending,\n\t\t\tPhase: types.PhaseInspiration,\n\t\t}\n\t\texecStore := store.NewExecutionStore()\n\t\trequire.NoError(t, execStore.Save(ctx.Context, record))\n\n\t\ttasks := []map[string]interface{}{\n\t\t\t{\"id\": \"t1\", \"name\": \"Task 1\"},\n\t\t\t{\"id\": \"t2\", \"name\": \"Task 2\"},\n\t\t}\n\t\tactionData := map[string]interface{}{\"tasks\": tasks}\n\t\terr := m.adjustExecution(ctx, record, actionData, execStore)\n\t\trequire.NoError(t, err)\n\t\tassert.Len(t, record.Tasks, 2)\n\t})\n\n\tt.Run(\"nil action data is noop\", func(t *testing.T) {\n\t\tctx := types.NewContext(context.Background(), nil)\n\t\trecord := &store.ExecutionRecord{}\n\t\texecStore := store.NewExecutionStore()\n\n\t\terr := m.adjustExecution(ctx, record, nil, execStore)\n\t\trequire.NoError(t, err)\n\t\tassert.Nil(t, record.Goals)\n\t})\n\n\tt.Run(\"non-map action data handled gracefully\", func(t *testing.T) {\n\t\tif testing.Short() {\n\t\t\tt.Skip(\"Requires database\")\n\t\t}\n\t\ttestutils.Prepare(t)\n\t\tdefer testutils.Clean(t)\n\t\tctx := types.NewContext(context.Background(), nil)\n\t\trecord := &store.ExecutionRecord{\n\t\t\tExecutionID: \"exec-hl4\",\n\t\t\tMemberID: \"member-hl4\",\n\t\t\tTriggerType: types.TriggerHuman,\n\t\t\tStatus: types.ExecPending,\n\t\t\tPhase: types.PhaseInspiration,\n\t\t}\n\t\texecStore := store.NewExecutionStore()\n\t\trequire.NoError(t, execStore.Save(ctx.Context, record))\n\n\t\terr := m.adjustExecution(ctx, record, \"not a map\", execStore)\n\t\trequire.NoError(t, err)\n\t})\n}\n\n// HL5-HL6: injectTask\nfunc TestInjectTask(t *testing.T) {\n\tm := &Manager{}\n\n\tt.Run(\"appends new task with auto-generated ID\", func(t *testing.T) {\n\t\tif testing.Short() {\n\t\t\tt.Skip(\"Requires database\")\n\t\t}\n\t\ttestutils.Prepare(t)\n\t\tdefer testutils.Clean(t)\n\t\tctx := types.NewContext(context.Background(), nil)\n\t\trecord := &store.ExecutionRecord{\n\t\t\tExecutionID: \"exec-hl5\",\n\t\t\tMemberID: \"member-hl5\",\n\t\t\tTriggerType: types.TriggerHuman,\n\t\t\tStatus: types.ExecPending,\n\t\t\tPhase: types.PhaseInspiration,\n\t\t}\n\t\texecStore := store.NewExecutionStore()\n\t\trequire.NoError(t, execStore.Save(ctx.Context, record))\n\n\t\ttaskData := map[string]interface{}{\"name\": \"New Task\"}\n\t\terr := m.injectTask(ctx, record, taskData, execStore)\n\t\trequire.NoError(t, err)\n\t\trequire.Len(t, record.Tasks, 1)\n\t\tassert.Contains(t, record.Tasks[0].ID, \"injected-\")\n\t\tassert.Equal(t, types.TaskPending, record.Tasks[0].Status)\n\t})\n\n\tt.Run(\"preserves existing tasks\", func(t *testing.T) {\n\t\tif testing.Short() {\n\t\t\tt.Skip(\"Requires database\")\n\t\t}\n\t\ttestutils.Prepare(t)\n\t\tdefer testutils.Clean(t)\n\t\tctx := types.NewContext(context.Background(), nil)\n\t\trecord := &store.ExecutionRecord{\n\t\t\tExecutionID: \"exec-hl6\",\n\t\t\tMemberID: \"member-hl6\",\n\t\t\tTriggerType: types.TriggerHuman,\n\t\t\tStatus: types.ExecPending,\n\t\t\tPhase: types.PhaseInspiration,\n\t\t\tTasks: []types.Task{\n\t\t\t\t{ID: \"existing-1\", Description: \"Existing\"},\n\t\t\t},\n\t\t}\n\t\texecStore := store.NewExecutionStore()\n\t\trequire.NoError(t, execStore.Save(ctx.Context, record))\n\n\t\ttaskData := map[string]interface{}{\"name\": \"Added Task\"}\n\t\terr := m.injectTask(ctx, record, taskData, execStore)\n\t\trequire.NoError(t, err)\n\t\tassert.Len(t, record.Tasks, 2)\n\t\tassert.Equal(t, \"existing-1\", record.Tasks[0].ID)\n\t})\n\n\tt.Run(\"nil action data returns error\", func(t *testing.T) {\n\t\tctx := types.NewContext(context.Background(), nil)\n\t\trecord := &store.ExecutionRecord{}\n\t\texecStore := store.NewExecutionStore()\n\n\t\terr := m.injectTask(ctx, record, nil, execStore)\n\t\tassert.Error(t, err)\n\t\tassert.Contains(t, err.Error(), \"task data is required\")\n\t})\n\n\tt.Run(\"respects provided task ID\", func(t *testing.T) {\n\t\tif testing.Short() {\n\t\t\tt.Skip(\"Requires database\")\n\t\t}\n\t\ttestutils.Prepare(t)\n\t\tdefer testutils.Clean(t)\n\t\tctx := types.NewContext(context.Background(), nil)\n\t\trecord := &store.ExecutionRecord{\n\t\t\tExecutionID: \"exec-hl6b\",\n\t\t\tMemberID: \"member-hl6b\",\n\t\t\tTriggerType: types.TriggerHuman,\n\t\t\tStatus: types.ExecPending,\n\t\t\tPhase: types.PhaseInspiration,\n\t\t}\n\t\texecStore := store.NewExecutionStore()\n\t\trequire.NoError(t, execStore.Save(ctx.Context, record))\n\n\t\ttaskData := map[string]interface{}{\"id\": \"custom-id\", \"name\": \"Custom\"}\n\t\terr := m.injectTask(ctx, record, taskData, execStore)\n\t\trequire.NoError(t, err)\n\t\tassert.Equal(t, \"custom-id\", record.Tasks[0].ID)\n\t})\n}\n\n// HL7: callHostAgentForScenario\nfunc TestCallHostAgentForScenario(t *testing.T) {\n\tm := &Manager{}\n\n\tt.Run(\"no host agent returns error\", func(t *testing.T) {\n\t\tctx := types.NewContext(context.Background(), nil)\n\t\trobot := &types.Robot{MemberID: \"member-hl7\"}\n\n\t\t_, err := m.callHostAgentForScenario(ctx, robot, \"assign\", \"test\", nil, \"chat-1\")\n\t\tassert.Error(t, err)\n\t\tassert.Contains(t, err.Error(), \"no Host Agent configured\")\n\t})\n\n\tt.Run(\"robot with nil config returns error\", func(t *testing.T) {\n\t\tctx := types.NewContext(context.Background(), nil)\n\t\trobot := &types.Robot{MemberID: \"member-hl7b\", Config: nil}\n\n\t\t_, err := m.callHostAgentForScenario(ctx, robot, \"assign\", \"test\", nil, \"chat-1\")\n\t\tassert.Error(t, err)\n\t\tassert.Contains(t, err.Error(), \"no Host Agent configured\")\n\t})\n}\n\n// HL8: directAssign (needs pool — tested in processHostAction)\n// HL9-HL10: directResume (needs executor — tested in processHostAction)\n\n// Updated buildRobotStatusSnapshot tests\nfunc TestBuildRobotStatusSnapshotV2(t *testing.T) {\n\tm := &Manager{}\n\n\tt.Run(\"nil robot returns nil\", func(t *testing.T) {\n\t\tsnap := m.buildRobotStatusSnapshot(nil)\n\t\tassert.Nil(t, snap)\n\t})\n\n\tt.Run(\"populates MemberID and Status\", func(t *testing.T) {\n\t\trobot := &types.Robot{\n\t\t\tMemberID: \"member-snap\",\n\t\t\tStatus: types.RobotWorking,\n\t\t}\n\t\tsnap := m.buildRobotStatusSnapshot(robot)\n\t\trequire.NotNil(t, snap)\n\t\tassert.Equal(t, \"member-snap\", snap.MemberID)\n\t\tassert.Equal(t, types.RobotWorking, snap.Status)\n\t})\n\n\tt.Run(\"uses ActiveCount and WaitingCount\", func(t *testing.T) {\n\t\trobot := &types.Robot{MemberID: \"member-snap2\"}\n\t\texec1 := &types.Execution{ID: \"e1\", Status: types.ExecRunning}\n\t\texec2 := &types.Execution{ID: \"e2\", Status: types.ExecWaiting}\n\t\trobot.AddExecution(exec1)\n\t\trobot.AddExecution(exec2)\n\n\t\tsnap := m.buildRobotStatusSnapshot(robot)\n\t\trequire.NotNil(t, snap)\n\t\tassert.Equal(t, 1, snap.ActiveCount)\n\t\tassert.Equal(t, 1, snap.WaitingCount)\n\t})\n\n\tt.Run(\"populates ActiveExecs briefs\", func(t *testing.T) {\n\t\trobot := &types.Robot{MemberID: \"member-snap3\"}\n\t\texec := &types.Execution{ID: \"e-brief\", Status: types.ExecRunning, Name: \"Test Exec\"}\n\t\trobot.AddExecution(exec)\n\n\t\tsnap := m.buildRobotStatusSnapshot(robot)\n\t\trequire.NotNil(t, snap)\n\t\trequire.Len(t, snap.ActiveExecs, 1)\n\t\tassert.Equal(t, \"e-brief\", snap.ActiveExecs[0].ID)\n\t})\n\n\tt.Run(\"uses robot MaxQuota\", func(t *testing.T) {\n\t\trobot := &types.Robot{\n\t\t\tMemberID: \"member-snap4\",\n\t\t\tConfig: &types.Config{Quota: &types.Quota{Max: 7}},\n\t\t}\n\t\tsnap := m.buildRobotStatusSnapshot(robot)\n\t\trequire.NotNil(t, snap)\n\t\tassert.Equal(t, 7, snap.MaxQuota)\n\t})\n}\n\n// Test processHostAction — adjust branch\nfunc TestProcessHostActionAdjust(t *testing.T) {\n\tif testing.Short() {\n\t\tt.Skip(\"Requires database\")\n\t}\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\tm := &Manager{}\n\tctx := types.NewContext(context.Background(), nil)\n\trobot := &types.Robot{MemberID: \"member-pa-adj\"}\n\n\tt.Run(\"adjust with goals\", func(t *testing.T) {\n\t\trecord := &store.ExecutionRecord{\n\t\t\tExecutionID: \"exec-pa2\",\n\t\t\tMemberID: \"member-pa-adj\",\n\t\t\tTriggerType: types.TriggerHuman,\n\t\t\tStatus: types.ExecConfirming,\n\t\t\tPhase: types.PhaseInspiration,\n\t\t}\n\t\texecStore := store.NewExecutionStore()\n\t\trequire.NoError(t, execStore.Save(ctx.Context, record))\n\n\t\toutput := &types.HostOutput{\n\t\t\tReply: \"Plan adjusted\",\n\t\t\tAction: types.HostActionAdjust,\n\t\t\tActionData: map[string]interface{}{\"goals\": \"new goals\"},\n\t\t}\n\n\t\tresp, err := m.processHostAction(ctx, robot, record, output, execStore)\n\t\trequire.NoError(t, err)\n\t\tassert.Equal(t, \"adjusted\", resp.Status)\n\t\trequire.NotNil(t, record.Goals)\n\t\tassert.Equal(t, \"new goals\", record.Goals.Content)\n\t})\n\n\tt.Run(\"adjust with tasks\", func(t *testing.T) {\n\t\trecord := &store.ExecutionRecord{\n\t\t\tExecutionID: \"exec-pa3\",\n\t\t\tMemberID: \"member-pa-adj\",\n\t\t\tTriggerType: types.TriggerHuman,\n\t\t\tStatus: types.ExecConfirming,\n\t\t\tPhase: types.PhaseInspiration,\n\t\t}\n\t\texecStore := store.NewExecutionStore()\n\t\trequire.NoError(t, execStore.Save(ctx.Context, record))\n\n\t\ttasksJSON := []map[string]interface{}{{\"id\": \"t1\", \"name\": \"Adjusted Task\"}}\n\t\toutput := &types.HostOutput{\n\t\t\tReply: \"Tasks updated\",\n\t\t\tAction: types.HostActionAdjust,\n\t\t\tActionData: map[string]interface{}{\"tasks\": tasksJSON},\n\t\t}\n\n\t\tresp, err := m.processHostAction(ctx, robot, record, output, execStore)\n\t\trequire.NoError(t, err)\n\t\tassert.Equal(t, \"adjusted\", resp.Status)\n\t\tassert.Len(t, record.Tasks, 1)\n\t})\n\n\tt.Run(\"adjust with nil data is noop\", func(t *testing.T) {\n\t\trecord := &store.ExecutionRecord{\n\t\t\tExecutionID: \"exec-pa4\",\n\t\t\tMemberID: \"member-pa-adj\",\n\t\t\tTriggerType: types.TriggerHuman,\n\t\t\tStatus: types.ExecConfirming,\n\t\t\tPhase: types.PhaseInspiration,\n\t\t}\n\t\texecStore := store.NewExecutionStore()\n\t\trequire.NoError(t, execStore.Save(ctx.Context, record))\n\n\t\toutput := &types.HostOutput{\n\t\t\tReply: \"No changes\",\n\t\t\tAction: types.HostActionAdjust,\n\t\t}\n\n\t\tresp, err := m.processHostAction(ctx, robot, record, output, execStore)\n\t\trequire.NoError(t, err)\n\t\tassert.Equal(t, \"adjusted\", resp.Status)\n\t})\n}\n\n// Test processHostAction — add_task branch\nfunc TestProcessHostActionAddTask(t *testing.T) {\n\tif testing.Short() {\n\t\tt.Skip(\"Requires database\")\n\t}\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\tm := &Manager{}\n\tctx := types.NewContext(context.Background(), nil)\n\trobot := &types.Robot{MemberID: \"member-pa-at\"}\n\n\tt.Run(\"add task success\", func(t *testing.T) {\n\t\trecord := &store.ExecutionRecord{\n\t\t\tExecutionID: \"exec-pa5\",\n\t\t\tMemberID: \"member-pa-at\",\n\t\t\tTriggerType: types.TriggerHuman,\n\t\t\tStatus: types.ExecConfirming,\n\t\t\tPhase: types.PhaseInspiration,\n\t\t}\n\t\texecStore := store.NewExecutionStore()\n\t\trequire.NoError(t, execStore.Save(ctx.Context, record))\n\n\t\toutput := &types.HostOutput{\n\t\t\tReply: \"Task added\",\n\t\t\tAction: types.HostActionAddTask,\n\t\t\tActionData: map[string]interface{}{\"name\": \"New task\"},\n\t\t}\n\n\t\tresp, err := m.processHostAction(ctx, robot, record, output, execStore)\n\t\trequire.NoError(t, err)\n\t\tassert.Equal(t, \"task_added\", resp.Status)\n\t\tassert.Len(t, record.Tasks, 1)\n\t})\n\n\tt.Run(\"add task nil data returns error\", func(t *testing.T) {\n\t\trecord := &store.ExecutionRecord{\n\t\t\tExecutionID: \"exec-pa6\",\n\t\t\tMemberID: \"member-pa-at\",\n\t\t}\n\t\texecStore := store.NewExecutionStore()\n\n\t\toutput := &types.HostOutput{\n\t\t\tReply: \"Add task\",\n\t\t\tAction: types.HostActionAddTask,\n\t\t}\n\n\t\t_, err := m.processHostAction(ctx, robot, record, output, execStore)\n\t\tassert.Error(t, err)\n\t\tassert.Contains(t, err.Error(), \"task data is required\")\n\t})\n}\n\n// Test processHostAction — skip branch\nfunc TestProcessHostActionSkip(t *testing.T) {\n\tm := &Manager{}\n\tctx := types.NewContext(context.Background(), nil)\n\trobot := &types.Robot{MemberID: \"member-pa-skip\"}\n\n\tt.Run(\"skip without waiting task returns error\", func(t *testing.T) {\n\t\trecord := &store.ExecutionRecord{\n\t\t\tExecutionID: \"exec-pa8\",\n\t\t\tMemberID: \"member-pa-skip\",\n\t\t}\n\t\texecStore := store.NewExecutionStore()\n\n\t\toutput := &types.HostOutput{\n\t\t\tReply: \"Skip it\",\n\t\t\tAction: types.HostActionSkip,\n\t\t}\n\n\t\t_, err := m.processHostAction(ctx, robot, record, output, execStore)\n\t\tassert.Error(t, err)\n\t\tassert.Contains(t, err.Error(), \"no task is waiting\")\n\t})\n}\n\n// Test processHostAction — wait_for_more and default\nfunc TestProcessHostActionWaitForMoreAndDefault(t *testing.T) {\n\tm := &Manager{}\n\tctx := types.NewContext(context.Background(), nil)\n\trobot := &types.Robot{MemberID: \"member-pa-wfm\"}\n\n\tt.Run(\"wait_for_more\", func(t *testing.T) {\n\t\trecord := &store.ExecutionRecord{}\n\t\texecStore := store.NewExecutionStore()\n\n\t\toutput := &types.HostOutput{\n\t\t\tReply: \"More details please\",\n\t\t\tWaitForMore: true,\n\t\t}\n\n\t\tresp, err := m.processHostAction(ctx, robot, record, output, execStore)\n\t\trequire.NoError(t, err)\n\t\tassert.Equal(t, \"waiting_for_more\", resp.Status)\n\t\tassert.Equal(t, \"More details please\", resp.Reply)\n\t\tassert.True(t, resp.WaitForMore)\n\t})\n\n\tt.Run(\"unknown action returns acknowledged\", func(t *testing.T) {\n\t\trecord := &store.ExecutionRecord{}\n\t\texecStore := store.NewExecutionStore()\n\n\t\toutput := &types.HostOutput{\n\t\t\tReply: \"OK\",\n\t\t\tAction: \"unknown_action\",\n\t\t}\n\n\t\tresp, err := m.processHostAction(ctx, robot, record, output, execStore)\n\t\trequire.NoError(t, err)\n\t\tassert.Equal(t, \"acknowledged\", resp.Status)\n\t\tassert.Equal(t, \"OK\", resp.Message)\n\t})\n}\n\n// Test processHostAction — cancel branch\nfunc TestProcessHostActionCancel(t *testing.T) {\n\tif testing.Short() {\n\t\tt.Skip(\"Requires database\")\n\t}\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\tt.Run(\"cancel waiting execution\", func(t *testing.T) {\n\t\t// Cannot fully test without a started manager; verify the error path\n\t\tm := &Manager{started: false}\n\t\tctx := types.NewContext(context.Background(), nil)\n\t\trobot := &types.Robot{MemberID: \"member-pa-cancel\"}\n\n\t\trecord := &store.ExecutionRecord{\n\t\t\tExecutionID: \"exec-pa11\",\n\t\t\tMemberID: \"member-pa-cancel\",\n\t\t}\n\t\texecStore := store.NewExecutionStore()\n\n\t\toutput := &types.HostOutput{\n\t\t\tReply: \"Cancel it\",\n\t\t\tAction: types.HostActionCancel,\n\t\t}\n\n\t\t_, err := m.processHostAction(ctx, robot, record, output, execStore)\n\t\tassert.Error(t, err)\n\t\tassert.Contains(t, err.Error(), \"manager not started\")\n\t})\n}\n\n// Test HandleInteract validation\nfunc TestHandleInteractValidationExtended(t *testing.T) {\n\tt.Run(\"manager not started returns error\", func(t *testing.T) {\n\t\tm := &Manager{started: false}\n\t\t_, err := m.HandleInteract(types.NewContext(context.Background(), nil), \"member-1\", &InteractRequest{Message: \"test\"})\n\t\tassert.Error(t, err)\n\t\tassert.Contains(t, err.Error(), \"manager not started\")\n\t})\n\n\tt.Run(\"empty member_id returns error\", func(t *testing.T) {\n\t\tm := &Manager{started: true}\n\t\t_, err := m.HandleInteract(types.NewContext(context.Background(), nil), \"\", &InteractRequest{Message: \"test\"})\n\t\tassert.Error(t, err)\n\t\tassert.Contains(t, err.Error(), \"member_id is required\")\n\t})\n\n\tt.Run(\"nil request returns error\", func(t *testing.T) {\n\t\tm := &Manager{started: true}\n\t\t_, err := m.HandleInteract(types.NewContext(context.Background(), nil), \"member-1\", nil)\n\t\tassert.Error(t, err)\n\t\tassert.Contains(t, err.Error(), \"message is required\")\n\t})\n\n\tt.Run(\"empty message returns error\", func(t *testing.T) {\n\t\tm := &Manager{started: true}\n\t\t_, err := m.HandleInteract(types.NewContext(context.Background(), nil), \"member-1\", &InteractRequest{})\n\t\tassert.Error(t, err)\n\t\tassert.Contains(t, err.Error(), \"message is required\")\n\t})\n\n\tt.Run(\"non-interactable status returns error\", func(t *testing.T) {\n\t\tif testing.Short() {\n\t\t\tt.Skip(\"Requires database and cache\")\n\t\t}\n\t\ttestutils.Prepare(t)\n\t\tdefer testutils.Clean(t)\n\t\t// Would require a full Manager with cache — tested via E2E\n\t})\n}\n\n// Test CancelExecution validation\nfunc TestCancelExecutionValidationExtended(t *testing.T) {\n\tt.Run(\"manager not started\", func(t *testing.T) {\n\t\tm := &Manager{started: false}\n\t\terr := m.CancelExecution(types.NewContext(context.Background(), nil), \"exec-1\")\n\t\tassert.Error(t, err)\n\t\tassert.Contains(t, err.Error(), \"manager not started\")\n\t})\n}\n\n// Test buildHostContext JSON output\nfunc TestBuildHostContextJSON(t *testing.T) {\n\tm := &Manager{}\n\n\trobot := &types.Robot{MemberID: \"member-ctx\"}\n\trecord := &store.ExecutionRecord{\n\t\tGoals: &types.Goals{Content: \"test goals\"},\n\t\tTasks: []types.Task{{ID: \"t1\"}},\n\t\tWaitingQuestion: \"What time?\",\n\t}\n\twaitingTask := &types.Task{ID: \"t1\", Status: types.TaskWaitingInput}\n\n\thostCtx := m.buildHostContext(robot, record, waitingTask)\n\trequire.NotNil(t, hostCtx)\n\n\tdata, err := json.Marshal(hostCtx)\n\trequire.NoError(t, err)\n\n\tvar parsed map[string]interface{}\n\terr = json.Unmarshal(data, &parsed)\n\trequire.NoError(t, err)\n\n\t// Goals is a struct, not a plain string\n\tgoalsRaw, ok := parsed[\"goals\"]\n\trequire.True(t, ok)\n\tgoalsMap, ok := goalsRaw.(map[string]interface{})\n\trequire.True(t, ok, \"Goals should be a JSON object, not a string\")\n\tassert.Equal(t, \"test goals\", goalsMap[\"content\"])\n\n\tassert.Equal(t, \"What time?\", parsed[\"agent_reply\"])\n}\n\n// ==================== processHostAction -- confirm branch (PA1) ====================\n\nfunc TestProcessHostActionConfirmRequiresPool(t *testing.T) {\n\tif testing.Short() {\n\t\tt.Skip(\"Requires database\")\n\t}\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\tm := &Manager{started: false}\n\tctx := types.NewContext(context.Background(), nil)\n\trobot := &types.Robot{MemberID: \"member-pa1\"}\n\trecord := &store.ExecutionRecord{\n\t\tExecutionID: \"exec-pa1\",\n\t\tMemberID: \"member-pa1\",\n\t\tStatus: types.ExecConfirming,\n\t}\n\texecStore := store.NewExecutionStore()\n\n\toutput := &types.HostOutput{\n\t\tReply: \"Confirmed\",\n\t\tAction: types.HostActionConfirm,\n\t}\n\n\tassert.Panics(t, func() {\n\t\tm.processHostAction(ctx, robot, record, output, execStore)\n\t}, \"should panic because pool/executor are nil\")\n}\n\n// ==================== processHostAction -- inject_ctx branch (PA9-PA10) ====================\n\nfunc TestProcessHostActionInjectCtx(t *testing.T) {\n\tt.Run(\"nil executor panics\", func(t *testing.T) {\n\t\tm := &Manager{}\n\t\tctx := types.NewContext(context.Background(), nil)\n\t\trobot := &types.Robot{MemberID: \"member-pa9\"}\n\t\trecord := &store.ExecutionRecord{\n\t\t\tExecutionID: \"exec-pa9\",\n\t\t\tMemberID: \"member-pa9\",\n\t\t\tStatus: types.ExecWaiting,\n\t\t}\n\t\texecStore := store.NewExecutionStore()\n\n\t\toutput := &types.HostOutput{\n\t\t\tReply: \"Here's context\",\n\t\t\tAction: types.HostActionInjectCtx,\n\t\t\tActionData: \"additional context data\",\n\t\t}\n\n\t\tassert.Panics(t, func() {\n\t\t\tm.processHostAction(ctx, robot, record, output, execStore)\n\t\t})\n\t})\n\n\tt.Run(\"with mock executor delegates resume\", func(t *testing.T) {\n\t\tmockExec := &mockExecutor{resumeErr: fmt.Errorf(\"mock error\")}\n\t\tm := &Manager{executor: mockExec}\n\t\tctx := types.NewContext(context.Background(), nil)\n\t\trobot := &types.Robot{MemberID: \"member-pa10\"}\n\t\trecord := &store.ExecutionRecord{\n\t\t\tExecutionID: \"exec-pa10\",\n\t\t\tMemberID: \"member-pa10\",\n\t\t}\n\t\texecStore := store.NewExecutionStore()\n\n\t\toutput := &types.HostOutput{\n\t\t\tReply: \"Resume with data\",\n\t\t\tAction: types.HostActionInjectCtx,\n\t\t\tActionData: map[string]interface{}{\"reply\": \"detailed info\"},\n\t\t}\n\n\t\t_, err := m.processHostAction(ctx, robot, record, output, execStore)\n\t\tassert.Error(t, err)\n\t\tassert.Contains(t, err.Error(), \"mock error\")\n\t})\n\n\tt.Run(\"ErrExecutionSuspended returns waiting status\", func(t *testing.T) {\n\t\tmockExec := &mockExecutor{resumeErr: types.ErrExecutionSuspended}\n\t\tm := &Manager{executor: mockExec}\n\t\tctx := types.NewContext(context.Background(), nil)\n\t\trobot := &types.Robot{MemberID: \"member-pa10b\"}\n\t\trecord := &store.ExecutionRecord{\n\t\t\tExecutionID: \"exec-pa10b\",\n\t\t\tMemberID: \"member-pa10b\",\n\t\t}\n\t\texecStore := store.NewExecutionStore()\n\n\t\toutput := &types.HostOutput{\n\t\t\tReply: \"Resume\",\n\t\t\tAction: types.HostActionInjectCtx,\n\t\t\tActionData: \"context\",\n\t\t}\n\n\t\tresp, err := m.processHostAction(ctx, robot, record, output, execStore)\n\t\trequire.NoError(t, err)\n\t\tassert.Equal(t, \"waiting\", resp.Status)\n\t})\n}\n\n// ==================== HandleInteract routing (HI5-HI8) ====================\n\nfunc TestHandleInteractRouting(t *testing.T) {\n\tt.Run(\"HI5: non-existent execution_id returns error\", func(t *testing.T) {\n\t\tif testing.Short() {\n\t\t\tt.Skip(\"Requires database and cache\")\n\t\t}\n\t\ttestutils.Prepare(t)\n\t\tdefer testutils.Clean(t)\n\n\t\tm := &Manager{started: true, cache: cache.New()}\n\t\tctx := types.NewContext(context.Background(), nil)\n\t\t_, err := m.HandleInteract(ctx, \"member-hi5\", &InteractRequest{\n\t\t\tExecutionID: \"nonexistent-exec\",\n\t\t\tMessage: \"test\",\n\t\t})\n\t\tassert.Error(t, err)\n\t})\n\n\tt.Run(\"HI6: non-existent robot returns error\", func(t *testing.T) {\n\t\tif testing.Short() {\n\t\t\tt.Skip(\"Requires database and cache\")\n\t\t}\n\t\ttestutils.Prepare(t)\n\t\tdefer testutils.Clean(t)\n\t\tm := &Manager{started: true, cache: cache.New()}\n\t\tctx := types.NewContext(context.Background(), nil)\n\t\t_, err := m.HandleInteract(ctx, \"nonexistent-robot\", &InteractRequest{\n\t\t\tMessage: \"test\",\n\t\t})\n\t\tassert.Error(t, err)\n\t\tassert.Contains(t, err.Error(), \"robot not found\")\n\t})\n}\n\n// ==================== CancelExecution validation (CE2-CE5) ====================\n\nfunc TestCancelExecutionStatusValidation(t *testing.T) {\n\tif testing.Short() {\n\t\tt.Skip(\"Requires database\")\n\t}\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\tt.Run(\"CE2: non-existent execution returns error\", func(t *testing.T) {\n\t\tm := &Manager{started: true}\n\t\tctx := types.NewContext(context.Background(), nil)\n\t\terr := m.CancelExecution(ctx, \"nonexistent-exec\")\n\t\tassert.Error(t, err)\n\t\tassert.Contains(t, err.Error(), \"execution not found\")\n\t})\n\n\tt.Run(\"CE3: running execution cannot be cancelled\", func(t *testing.T) {\n\t\tm := &Manager{started: true}\n\t\tctx := types.NewContext(context.Background(), nil)\n\n\t\texecStore := store.NewExecutionStore()\n\t\trecord := &store.ExecutionRecord{\n\t\t\tExecutionID: \"exec-ce3\",\n\t\t\tMemberID: \"member-ce3\",\n\t\t\tStatus: types.ExecRunning,\n\t\t\tTriggerType: types.TriggerHuman,\n\t\t\tPhase: types.PhaseInspiration,\n\t\t}\n\t\trequire.NoError(t, execStore.Save(ctx.Context, record))\n\n\t\terr := m.CancelExecution(ctx, \"exec-ce3\")\n\t\tassert.Error(t, err)\n\t\tassert.Contains(t, err.Error(), \"only waiting/confirming can be cancelled\")\n\t})\n\n\tt.Run(\"CE4: completed execution cannot be cancelled\", func(t *testing.T) {\n\t\tm := &Manager{started: true}\n\t\tctx := types.NewContext(context.Background(), nil)\n\n\t\texecStore := store.NewExecutionStore()\n\t\trecord := &store.ExecutionRecord{\n\t\t\tExecutionID: \"exec-ce4\",\n\t\t\tMemberID: \"member-ce4\",\n\t\t\tStatus: types.ExecCompleted,\n\t\t\tTriggerType: types.TriggerHuman,\n\t\t\tPhase: types.PhaseInspiration,\n\t\t}\n\t\trequire.NoError(t, execStore.Save(ctx.Context, record))\n\n\t\terr := m.CancelExecution(ctx, \"exec-ce4\")\n\t\tassert.Error(t, err)\n\t\tassert.Contains(t, err.Error(), \"only waiting/confirming can be cancelled\")\n\t})\n}\n\n// ==================== InteractRequest/InteractResponse struct validation ====================\n\nfunc TestInteractRequestStructFields(t *testing.T) {\n\treq := &InteractRequest{\n\t\tExecutionID: \"exec-1\",\n\t\tTaskID: \"task-1\",\n\t\tSource: types.InteractSourceUI,\n\t\tMessage: \"do something\",\n\t\tAction: \"confirm\",\n\t}\n\tassert.Equal(t, \"exec-1\", req.ExecutionID)\n\tassert.Equal(t, \"task-1\", req.TaskID)\n\tassert.Equal(t, types.InteractSourceUI, req.Source)\n\tassert.Equal(t, \"do something\", req.Message)\n\tassert.Equal(t, \"confirm\", req.Action)\n}\n\nfunc TestInteractResponseStructFields(t *testing.T) {\n\tresp := &InteractResponse{\n\t\tExecutionID: \"exec-1\",\n\t\tStatus: \"confirmed\",\n\t\tMessage: \"Done\",\n\t\tChatID: \"chat-1\",\n\t\tReply: \"I'll do it\",\n\t\tWaitForMore: true,\n\t}\n\tassert.Equal(t, \"exec-1\", resp.ExecutionID)\n\tassert.Equal(t, \"confirmed\", resp.Status)\n\tassert.Equal(t, \"Done\", resp.Message)\n\tassert.Equal(t, \"chat-1\", resp.ChatID)\n\tassert.Equal(t, \"I'll do it\", resp.Reply)\n\tassert.True(t, resp.WaitForMore)\n}\n\n// ==================== executeResume helper ====================\n\nfunc TestExecuteResumeNilExecutor(t *testing.T) {\n\tm := &Manager{}\n\tctx := types.NewContext(context.Background(), nil)\n\n\tassert.Panics(t, func() {\n\t\t_ = m.executeResume(ctx, \"exec-test\", \"reply\")\n\t})\n}\n\nfunc TestExecuteResumeWithMock(t *testing.T) {\n\tt.Run(\"delegates to executor Resume\", func(t *testing.T) {\n\t\tmockExec := &mockExecutor{resumeErr: nil}\n\t\tm := &Manager{executor: mockExec}\n\t\tctx := types.NewContext(context.Background(), nil)\n\n\t\terr := m.executeResume(ctx, \"exec-test\", \"reply\")\n\t\tassert.NoError(t, err)\n\t})\n\n\tt.Run(\"propagates error\", func(t *testing.T) {\n\t\tmockExec := &mockExecutor{resumeErr: fmt.Errorf(\"resume failed\")}\n\t\tm := &Manager{executor: mockExec}\n\t\tctx := types.NewContext(context.Background(), nil)\n\n\t\terr := m.executeResume(ctx, \"exec-test\", \"reply\")\n\t\tassert.Error(t, err)\n\t\tassert.Contains(t, err.Error(), \"resume failed\")\n\t})\n\n\tt.Run(\"propagates ErrExecutionSuspended\", func(t *testing.T) {\n\t\tmockExec := &mockExecutor{resumeErr: types.ErrExecutionSuspended}\n\t\tm := &Manager{executor: mockExec}\n\t\tctx := types.NewContext(context.Background(), nil)\n\n\t\terr := m.executeResume(ctx, \"exec-test\", \"reply\")\n\t\tassert.Equal(t, types.ErrExecutionSuspended, err)\n\t})\n}\n\n// ==================== skipWaitingTask and directResume with mock ====================\n\nfunc TestSkipWaitingTaskWithMock(t *testing.T) {\n\tt.Run(\"no waiting task returns error\", func(t *testing.T) {\n\t\tmockExec := &mockExecutor{}\n\t\tm := &Manager{executor: mockExec}\n\t\tctx := types.NewContext(context.Background(), nil)\n\t\trecord := &store.ExecutionRecord{\n\t\t\tExecutionID: \"exec-skip\",\n\t\t}\n\t\texecStore := store.NewExecutionStore()\n\n\t\terr := m.skipWaitingTask(ctx, record, execStore)\n\t\tassert.Error(t, err)\n\t\tassert.Contains(t, err.Error(), \"no task is waiting\")\n\t})\n\n\tt.Run(\"marks waiting task as skipped and resumes\", func(t *testing.T) {\n\t\tmockExec := &mockExecutor{resumeErr: nil}\n\t\tm := &Manager{executor: mockExec}\n\t\tctx := types.NewContext(context.Background(), nil)\n\t\trecord := &store.ExecutionRecord{\n\t\t\tExecutionID: \"exec-skip2\",\n\t\t\tWaitingTaskID: \"task-w\",\n\t\t\tTasks: []types.Task{\n\t\t\t\t{ID: \"task-w\", Status: types.TaskWaitingInput},\n\t\t\t},\n\t\t}\n\t\texecStore := store.NewExecutionStore()\n\n\t\terr := m.skipWaitingTask(ctx, record, execStore)\n\t\tassert.NoError(t, err)\n\t\tassert.Equal(t, types.TaskSkipped, record.Tasks[0].Status)\n\t})\n}\n\nfunc TestDirectResumeWithMock(t *testing.T) {\n\tt.Run(\"successful resume\", func(t *testing.T) {\n\t\tmockExec := &mockExecutor{resumeErr: nil}\n\t\tm := &Manager{executor: mockExec}\n\t\tctx := types.NewContext(context.Background(), nil)\n\t\trecord := &store.ExecutionRecord{\n\t\t\tExecutionID: \"exec-dr\",\n\t\t\tChatID: \"chat-dr\",\n\t\t}\n\t\treq := &InteractRequest{Message: \"continue\"}\n\n\t\tresp, err := m.directResume(ctx, record, req)\n\t\trequire.NoError(t, err)\n\t\tassert.Equal(t, \"resumed\", resp.Status)\n\t\tassert.Equal(t, \"exec-dr\", resp.ExecutionID)\n\t\tassert.Equal(t, \"chat-dr\", resp.ChatID)\n\t})\n\n\tt.Run(\"suspended again\", func(t *testing.T) {\n\t\tmockExec := &mockExecutor{resumeErr: types.ErrExecutionSuspended}\n\t\tm := &Manager{executor: mockExec}\n\t\tctx := types.NewContext(context.Background(), nil)\n\t\trecord := &store.ExecutionRecord{\n\t\t\tExecutionID: \"exec-dr2\",\n\t\t\tChatID: \"chat-dr2\",\n\t\t}\n\t\treq := &InteractRequest{Message: \"continue\"}\n\n\t\tresp, err := m.directResume(ctx, record, req)\n\t\trequire.NoError(t, err)\n\t\tassert.Equal(t, \"waiting\", resp.Status)\n\t})\n\n\tt.Run(\"error propagated\", func(t *testing.T) {\n\t\tmockExec := &mockExecutor{resumeErr: fmt.Errorf(\"resume failed\")}\n\t\tm := &Manager{executor: mockExec}\n\t\tctx := types.NewContext(context.Background(), nil)\n\t\trecord := &store.ExecutionRecord{\n\t\t\tExecutionID: \"exec-dr3\",\n\t\t}\n\t\treq := &InteractRequest{Message: \"continue\"}\n\n\t\t_, err := m.directResume(ctx, record, req)\n\t\tassert.Error(t, err)\n\t\tassert.Contains(t, err.Error(), \"resume failed\")\n\t})\n}\n" }, { "path": "agent/robot/manager/interact_test.go", "content": "package manager\n\nimport (\n\t\"testing\"\n\n\t\"github.com/stretchr/testify/assert\"\n\t\"github.com/stretchr/testify/require\"\n\t\"github.com/yaoapp/yao/agent/robot/executor/standard\"\n\t\"github.com/yaoapp/yao/agent/robot/store\"\n\t\"github.com/yaoapp/yao/agent/robot/types\"\n)\n\nfunc TestBuildRobotStatusSnapshot(t *testing.T) {\n\tm := &Manager{}\n\n\tt.Run(\"nil robot returns nil\", func(t *testing.T) {\n\t\tsnap := m.buildRobotStatusSnapshot(nil)\n\t\tassert.Nil(t, snap)\n\t})\n\n\tt.Run(\"robot with quota\", func(t *testing.T) {\n\t\trobot := &types.Robot{\n\t\t\tMemberID: \"test-member\",\n\t\t\tConfig: &types.Config{\n\t\t\t\tQuota: &types.Quota{Max: 5},\n\t\t\t},\n\t\t}\n\t\tsnap := m.buildRobotStatusSnapshot(robot)\n\t\trequire.NotNil(t, snap)\n\t\tassert.Equal(t, 5, snap.MaxQuota)\n\t})\n\n\tt.Run(\"robot without quota uses default\", func(t *testing.T) {\n\t\trobot := &types.Robot{\n\t\t\tMemberID: \"test-member\",\n\t\t}\n\t\tsnap := m.buildRobotStatusSnapshot(robot)\n\t\trequire.NotNil(t, snap)\n\t\tassert.Equal(t, 2, snap.MaxQuota) // robot.MaxQuota() returns 2 for nil config\n\t})\n}\n\nfunc TestFindWaitingTask(t *testing.T) {\n\tm := &Manager{}\n\n\tt.Run(\"returns nil when no waiting task id\", func(t *testing.T) {\n\t\trecord := &store.ExecutionRecord{\n\t\t\tTasks: []types.Task{\n\t\t\t\t{ID: \"task-1\"},\n\t\t\t},\n\t\t}\n\t\ttask := m.findWaitingTask(record)\n\t\tassert.Nil(t, task)\n\t})\n\n\tt.Run(\"finds matching task\", func(t *testing.T) {\n\t\trecord := &store.ExecutionRecord{\n\t\t\tWaitingTaskID: \"task-2\",\n\t\t\tTasks: []types.Task{\n\t\t\t\t{ID: \"task-1\"},\n\t\t\t\t{ID: \"task-2\", Status: types.TaskWaitingInput},\n\t\t\t\t{ID: \"task-3\"},\n\t\t\t},\n\t\t}\n\t\ttask := m.findWaitingTask(record)\n\t\trequire.NotNil(t, task)\n\t\tassert.Equal(t, \"task-2\", task.ID)\n\t})\n\n\tt.Run(\"returns nil when task not found\", func(t *testing.T) {\n\t\trecord := &store.ExecutionRecord{\n\t\t\tWaitingTaskID: \"nonexistent\",\n\t\t\tTasks: []types.Task{\n\t\t\t\t{ID: \"task-1\"},\n\t\t\t},\n\t\t}\n\t\ttask := m.findWaitingTask(record)\n\t\tassert.Nil(t, task)\n\t})\n}\n\nfunc TestBuildHostContext(t *testing.T) {\n\tm := &Manager{}\n\n\tt.Run(\"builds context with goals and tasks\", func(t *testing.T) {\n\t\trobot := &types.Robot{MemberID: \"test\"}\n\t\trecord := &store.ExecutionRecord{\n\t\t\tGoals: &types.Goals{Content: \"test goals\"},\n\t\t\tTasks: []types.Task{\n\t\t\t\t{ID: \"task-1\"},\n\t\t\t},\n\t\t\tWaitingQuestion: \"What is the answer?\",\n\t\t}\n\t\twaitingTask := &types.Task{ID: \"task-1\", Status: types.TaskWaitingInput}\n\n\t\thostCtx := m.buildHostContext(robot, record, waitingTask)\n\t\trequire.NotNil(t, hostCtx)\n\t\tassert.NotNil(t, hostCtx.Goals)\n\t\tassert.Equal(t, \"test goals\", hostCtx.Goals.Content)\n\t\tassert.Len(t, hostCtx.Tasks, 1)\n\t\tassert.NotNil(t, hostCtx.CurrentTask)\n\t\tassert.Equal(t, \"What is the answer?\", hostCtx.AgentReply)\n\t})\n\n\tt.Run(\"builds context without optional fields\", func(t *testing.T) {\n\t\trobot := &types.Robot{MemberID: \"test\"}\n\t\trecord := &store.ExecutionRecord{}\n\n\t\thostCtx := m.buildHostContext(robot, record, nil)\n\t\trequire.NotNil(t, hostCtx)\n\t\tassert.Nil(t, hostCtx.Goals)\n\t\tassert.Nil(t, hostCtx.Tasks)\n\t\tassert.Nil(t, hostCtx.CurrentTask)\n\t\tassert.Empty(t, hostCtx.AgentReply)\n\t})\n}\n\nfunc TestProcessHostAction(t *testing.T) {\n\tm := &Manager{}\n\n\tt.Run(\"wait_for_more returns waiting status\", func(t *testing.T) {\n\t\toutput := &types.HostOutput{\n\t\t\tReply: \"Please provide more details\",\n\t\t\tWaitForMore: true,\n\t\t}\n\t\trecord := &store.ExecutionRecord{}\n\t\trobot := &types.Robot{}\n\t\texecStore := store.NewExecutionStore()\n\n\t\tresp, err := m.processHostAction(types.NewContext(nil, nil), robot, record, output, execStore)\n\t\trequire.NoError(t, err)\n\t\tassert.Equal(t, \"waiting_for_more\", resp.Status)\n\t\tassert.Equal(t, \"Please provide more details\", resp.Reply)\n\t\tassert.True(t, resp.WaitForMore)\n\t})\n\n\tt.Run(\"unknown action returns acknowledged\", func(t *testing.T) {\n\t\toutput := &types.HostOutput{\n\t\t\tReply: \"Got it\",\n\t\t\tAction: \"unknown_action\",\n\t\t}\n\t\trecord := &store.ExecutionRecord{}\n\t\trobot := &types.Robot{}\n\t\texecStore := store.NewExecutionStore()\n\n\t\tresp, err := m.processHostAction(types.NewContext(nil, nil), robot, record, output, execStore)\n\t\trequire.NoError(t, err)\n\t\tassert.Equal(t, \"acknowledged\", resp.Status)\n\t})\n}\n\nfunc TestHandleInteractValidation(t *testing.T) {\n\tm := &Manager{started: true}\n\n\tt.Run(\"empty member_id returns error\", func(t *testing.T) {\n\t\t_, err := m.HandleInteract(types.NewContext(nil, nil), \"\", &InteractRequest{Message: \"test\"})\n\t\tassert.Error(t, err)\n\t\tassert.Contains(t, err.Error(), \"member_id is required\")\n\t})\n\n\tt.Run(\"nil request returns error\", func(t *testing.T) {\n\t\t_, err := m.HandleInteract(types.NewContext(nil, nil), \"member-1\", nil)\n\t\tassert.Error(t, err)\n\t\tassert.Contains(t, err.Error(), \"message is required\")\n\t})\n\n\tt.Run(\"empty message returns error\", func(t *testing.T) {\n\t\t_, err := m.HandleInteract(types.NewContext(nil, nil), \"member-1\", &InteractRequest{})\n\t\tassert.Error(t, err)\n\t\tassert.Contains(t, err.Error(), \"message is required\")\n\t})\n\n\tt.Run(\"manager not started returns error\", func(t *testing.T) {\n\t\tm2 := &Manager{started: false}\n\t\t_, err := m2.HandleInteract(types.NewContext(nil, nil), \"member-1\", &InteractRequest{Message: \"test\"})\n\t\tassert.Error(t, err)\n\t\tassert.Contains(t, err.Error(), \"manager not started\")\n\t})\n}\n\nfunc TestCancelExecutionValidation(t *testing.T) {\n\tm := &Manager{started: false}\n\n\tt.Run(\"manager not started returns error\", func(t *testing.T) {\n\t\terr := m.CancelExecution(types.NewContext(nil, nil), \"exec-1\")\n\t\tassert.Error(t, err)\n\t\tassert.Contains(t, err.Error(), \"manager not started\")\n\t})\n}\n\nfunc TestParseHostAgentResult(t *testing.T) {\n\tm := &Manager{}\n\n\tt.Run(\"plain text returns WaitForMore\", func(t *testing.T) {\n\t\tresult := &standard.CallResult{Content: \"I understand your request. Shall I proceed?\"}\n\t\toutput, err := m.parseHostAgentResult(result)\n\t\trequire.NoError(t, err)\n\t\tassert.True(t, output.WaitForMore, \"plain text should set WaitForMore=true\")\n\t\tassert.Equal(t, \"I understand your request. Shall I proceed?\", output.Reply)\n\t\tassert.Empty(t, string(output.Action), \"plain text should have no action\")\n\t})\n\n\tt.Run(\"JSON with action returns action\", func(t *testing.T) {\n\t\tresult := &standard.CallResult{\n\t\t\tContent: `{\"reply\":\"Task confirmed\",\"action\":\"confirm\",\"wait_for_more\":false}`,\n\t\t}\n\t\toutput, err := m.parseHostAgentResult(result)\n\t\trequire.NoError(t, err)\n\t\tassert.False(t, output.WaitForMore)\n\t\tassert.Equal(t, types.HostActionConfirm, output.Action)\n\t\tassert.Equal(t, \"Task confirmed\", output.Reply)\n\t})\n\n\tt.Run(\"JSON without action returns WaitForMore\", func(t *testing.T) {\n\t\tresult := &standard.CallResult{\n\t\t\tContent: `{\"reply\":\"Let me think about this\",\"some_field\":\"value\"}`,\n\t\t}\n\t\toutput, err := m.parseHostAgentResult(result)\n\t\trequire.NoError(t, err)\n\t\tassert.True(t, output.WaitForMore, \"JSON without action should set WaitForMore=true\")\n\t\tassert.NotEmpty(t, output.Reply)\n\t})\n\n\tt.Run(\"JSON with adjust action and action_data\", func(t *testing.T) {\n\t\tresult := &standard.CallResult{\n\t\t\tContent: `{\"reply\":\"Plan adjusted\",\"action\":\"adjust\",\"action_data\":{\"goals\":\"new goals\"}}`,\n\t\t}\n\t\toutput, err := m.parseHostAgentResult(result)\n\t\trequire.NoError(t, err)\n\t\tassert.False(t, output.WaitForMore)\n\t\tassert.Equal(t, types.HostActionAdjust, output.Action)\n\t\tassert.NotNil(t, output.ActionData)\n\t})\n\n\tt.Run(\"malformed JSON returns WaitForMore\", func(t *testing.T) {\n\t\tresult := &standard.CallResult{Content: `{invalid json`}\n\t\toutput, err := m.parseHostAgentResult(result)\n\t\trequire.NoError(t, err)\n\t\tassert.True(t, output.WaitForMore)\n\t\tassert.Equal(t, `{invalid json`, output.Reply)\n\t})\n\n\tt.Run(\"empty content returns WaitForMore\", func(t *testing.T) {\n\t\tresult := &standard.CallResult{Content: \"\"}\n\t\toutput, err := m.parseHostAgentResult(result)\n\t\trequire.NoError(t, err)\n\t\tassert.True(t, output.WaitForMore)\n\t})\n}\n" }, { "path": "agent/robot/manager/manager.go", "content": "package manager\n\nimport (\n\t\"context\"\n\t\"fmt\"\n\t\"sync\"\n\t\"time\"\n\n\t\"github.com/yaoapp/yao/agent/robot/cache\"\n\t\"github.com/yaoapp/yao/agent/robot/executor\"\n\t\"github.com/yaoapp/yao/agent/robot/pool\"\n\t\"github.com/yaoapp/yao/agent/robot/trigger\"\n\t\"github.com/yaoapp/yao/agent/robot/types\"\n\toauthtypes \"github.com/yaoapp/yao/openapi/oauth/types\"\n)\n\n// Default configuration values\nconst (\n\tDefaultTickInterval = time.Minute // default tick interval for clock checking\n)\n\n// Config holds manager configuration\ntype Config struct {\n\tTickInterval time.Duration // how often to check clock triggers (default: 1 minute)\n\tPoolConfig *pool.Config // worker pool configuration\n\tExecutor types.Executor // optional: custom executor (default: real executor)\n}\n\n// DefaultConfig returns default manager configuration\nfunc DefaultConfig() *Config {\n\treturn &Config{\n\t\tTickInterval: DefaultTickInterval,\n\t\tPoolConfig: pool.DefaultConfig(),\n\t}\n}\n\n// Manager implements types.Manager interface\n// Orchestrates the robot scheduling system: Cache -> Dedup -> Pool -> Executor\ntype Manager struct {\n\tconfig *Config\n\tcache *cache.Cache\n\tpool *pool.Pool\n\texecutor types.Executor\n\n\t// Execution control for pause/resume/stop\n\texecController *trigger.ExecutionController\n\n\t// Ticker for clock trigger checking\n\tticker *time.Ticker\n\ttickerDone chan struct{}\n\n\t// State\n\tstarted bool\n\tmu sync.RWMutex\n\n\t// Context for background operations\n\tctx context.Context\n\tcancel context.CancelFunc\n}\n\n// New creates a new manager instance with default configuration\nfunc New() *Manager {\n\treturn NewWithConfig(nil)\n}\n\n// NewWithConfig creates a new manager instance with custom configuration\nfunc NewWithConfig(config *Config) *Manager {\n\tif config == nil {\n\t\tconfig = DefaultConfig()\n\t}\n\n\t// Apply defaults for zero values\n\tif config.TickInterval <= 0 {\n\t\tconfig.TickInterval = DefaultTickInterval\n\t}\n\n\t// Create components\n\tc := cache.New()\n\tp := pool.NewWithConfig(config.PoolConfig)\n\tec := trigger.NewExecutionController()\n\n\t// Use custom executor if provided, otherwise create default\n\tvar e types.Executor\n\tif config.Executor != nil {\n\t\te = config.Executor\n\t} else {\n\t\te = executor.New()\n\t}\n\n\t// Wire up pool with executor\n\tp.SetExecutor(e)\n\n\t// Create shared executor instances for each mode\n\t// These are reused across all executions to maintain accurate counters\n\tdryRunExecutor := executor.NewDryRun()\n\n\t// Set executor factory for mode-specific executors\n\tp.SetExecutorFactory(func(mode types.ExecutorMode) types.Executor {\n\t\tswitch mode {\n\t\tcase types.ExecutorDryRun:\n\t\t\treturn dryRunExecutor\n\t\tcase types.ExecutorSandbox:\n\t\t\t// Sandbox not implemented, fall back to DryRun\n\t\t\treturn dryRunExecutor\n\t\tdefault:\n\t\t\t// Standard mode or empty - use the configured executor\n\t\t\treturn e\n\t\t}\n\t})\n\n\treturn &Manager{\n\t\tconfig: config,\n\t\tcache: c,\n\t\tpool: p,\n\t\texecutor: e,\n\t\texecController: ec,\n\t}\n}\n\n// Start starts the manager\n// 1. Load robots into cache\n// 2. Start worker pool\n// 3. Start clock ticker goroutine\nfunc (m *Manager) Start() error {\n\tm.mu.Lock()\n\tdefer m.mu.Unlock()\n\n\tif m.started {\n\t\treturn fmt.Errorf(\"manager already started\")\n\t}\n\n\t// Create background context\n\tm.ctx, m.cancel = context.WithCancel(context.Background())\n\n\t// Load robots into cache\n\tctx := types.NewContext(m.ctx, nil)\n\tif err := m.cache.Load(ctx); err != nil {\n\t\treturn fmt.Errorf(\"failed to load robots: %w\", err)\n\t}\n\n\t// Set completion callback to clean up ExecutionController when execution finishes\n\tm.pool.SetOnComplete(func(execID, memberID string, status types.ExecStatus) {\n\t\t// Remove from ExecutionController (cleans up in-memory tracking)\n\t\tm.execController.Untrack(execID)\n\t\t// Remove from robot's in-memory execution list\n\t\tif robot := m.cache.Get(memberID); robot != nil {\n\t\t\trobot.RemoveExecution(execID)\n\t\t}\n\t})\n\n\t// Start worker pool\n\tif err := m.pool.Start(); err != nil {\n\t\treturn fmt.Errorf(\"failed to start pool: %w\", err)\n\t}\n\n\t// Start clock ticker\n\tm.ticker = time.NewTicker(m.config.TickInterval)\n\tm.tickerDone = make(chan struct{})\n\n\tgo m.tickerLoop()\n\n\t// Start cache auto-refresh (every hour)\n\tm.cache.StartAutoRefresh(ctx, nil)\n\n\tm.started = true\n\treturn nil\n}\n\n// Stop stops the manager gracefully\n// 1. Stop clock ticker\n// 2. Stop cache auto-refresh\n// 3. Stop worker pool (waits for running jobs)\nfunc (m *Manager) Stop() error {\n\tm.mu.Lock()\n\tif !m.started {\n\t\tm.mu.Unlock()\n\t\treturn nil\n\t}\n\tm.started = false\n\tm.mu.Unlock()\n\n\t// Stop ticker\n\tif m.tickerDone != nil {\n\t\tclose(m.tickerDone)\n\t}\n\n\t// Stop cache auto-refresh\n\tm.cache.StopAutoRefresh()\n\n\t// Stop pool (waits for running jobs)\n\tif err := m.pool.Stop(); err != nil {\n\t\treturn fmt.Errorf(\"failed to stop pool: %w\", err)\n\t}\n\n\t// Cancel background context\n\tif m.cancel != nil {\n\t\tm.cancel()\n\t}\n\n\treturn nil\n}\n\n// tickerLoop is the main ticker goroutine\nfunc (m *Manager) tickerLoop() {\n\tfor {\n\t\tselect {\n\t\tcase <-m.tickerDone:\n\t\t\tm.ticker.Stop()\n\t\t\treturn\n\t\tcase now := <-m.ticker.C:\n\t\t\t// Perform tick - context is created per-robot in Tick()\n\t\t\t_ = m.Tick(m.ctx, now)\n\t\t}\n\t}\n}\n\n// Tick processes a clock tick\n// 1. Get all cached robots\n// 2. For each robot with clock trigger enabled\n// 3. Check if should execute based on clock config\n// 4. Submit to pool with robot's own identity\nfunc (m *Manager) Tick(parentCtx context.Context, now time.Time) error {\n\tm.mu.RLock()\n\tif !m.started {\n\t\tm.mu.RUnlock()\n\t\treturn nil\n\t}\n\tm.mu.RUnlock()\n\n\t// Get all cached robots\n\trobots := m.cache.ListAll()\n\n\tfor _, robot := range robots {\n\t\t// Skip if robot is not active\n\t\tif robot.Status == types.RobotPaused || robot.Status == types.RobotError || robot.Status == types.RobotMaintenance {\n\t\t\tcontinue\n\t\t}\n\n\t\t// Skip if clock trigger is disabled\n\t\tif robot.Config == nil || robot.Config.Triggers == nil {\n\t\t\tcontinue\n\t\t}\n\t\tif !robot.Config.Triggers.IsEnabled(types.TriggerClock) {\n\t\t\tcontinue\n\t\t}\n\n\t\t// Skip if no clock config\n\t\tif robot.Config.Clock == nil {\n\t\t\tcontinue\n\t\t}\n\n\t\t// Check if should trigger based on clock config\n\t\tif !m.shouldTrigger(robot, now) {\n\t\t\tcontinue\n\t\t}\n\n\t\t// TODO: dedup check (Phase 11.1)\n\t\t// result, err := m.dedup.Check(ctx, robot.MemberID, types.TriggerClock)\n\t\t// if err != nil || result == types.DedupSkip {\n\t\t// continue\n\t\t// }\n\n\t\t// Pre-generate execution ID and track for pause/resume/stop\n\t\t// We need to track BEFORE submit so we can pass the cancellable context to the executor\n\t\texecID := pool.GenerateExecID()\n\t\tctrlExec := m.execController.Track(execID, robot.MemberID, robot.TeamID)\n\n\t\t// Create context with robot's own identity and cancellable context\n\t\t// Clock-triggered executions run as the robot itself\n\t\trobotAuth := m.buildRobotAuth(robot)\n\t\texecCtx := types.NewContext(ctrlExec.Context(), robotAuth)\n\n\t\t// Create clock context for P0 inspiration\n\t\tclockCtx := types.NewClockContext(now, robot.Config.Clock.TZ)\n\n\t\t// Submit to pool with the cancellable context and execution control\n\t\t_, err := m.pool.SubmitWithID(execCtx, robot, types.TriggerClock, clockCtx, execID, ctrlExec)\n\t\tif err != nil {\n\t\t\t// If submission failed, untrack the execution\n\t\t\tm.execController.Untrack(execID)\n\t\t\t// Log error but continue with other robots\n\t\t\t// In production, this would be logged properly\n\t\t\tcontinue\n\t\t}\n\n\t\t// Update robot's last run time\n\t\trobot.LastRun = now\n\t}\n\n\treturn nil\n}\n\n// buildRobotAuth creates AuthorizedInfo for a robot's own identity\n// Used when robot executes autonomously (clock trigger)\nfunc (m *Manager) buildRobotAuth(robot *types.Robot) *oauthtypes.AuthorizedInfo {\n\treturn &oauthtypes.AuthorizedInfo{\n\t\tUserID: robot.MemberID,\n\t\tTeamID: robot.TeamID,\n\t\t// ClientID could be set to a special \"robot-agent\" identifier if needed\n\t\tClientID: \"robot-agent\",\n\t}\n}\n\n// shouldTrigger checks if a robot should be triggered based on its clock config\nfunc (m *Manager) shouldTrigger(robot *types.Robot, now time.Time) bool {\n\tclock := robot.Config.Clock\n\tif clock == nil {\n\t\treturn false\n\t}\n\n\t// Get time in robot's timezone\n\tloc := clock.GetLocation()\n\tlocalNow := now.In(loc)\n\n\tswitch clock.Mode {\n\tcase types.ClockTimes:\n\t\treturn m.shouldTriggerTimes(robot, clock, localNow)\n\tcase types.ClockInterval:\n\t\treturn m.shouldTriggerInterval(robot, clock, localNow)\n\tcase types.ClockDaemon:\n\t\treturn m.shouldTriggerDaemon(robot, clock, localNow)\n\tdefault:\n\t\treturn false\n\t}\n}\n\n// shouldTriggerTimes checks if current time matches any configured times\n// times mode: run at specific times (e.g., [\"09:00\", \"14:00\", \"17:00\"])\nfunc (m *Manager) shouldTriggerTimes(robot *types.Robot, clock *types.Clock, now time.Time) bool {\n\t// Check day of week first\n\tif !m.matchesDay(clock, now) {\n\t\treturn false\n\t}\n\n\t// Check if current time matches any configured time\n\tcurrentTime := now.Format(\"15:04\")\n\tfor _, t := range clock.Times {\n\t\tif t == currentTime {\n\t\t\t// Check if already triggered in this minute\n\t\t\tif !robot.LastRun.IsZero() {\n\t\t\t\tlastRunInLoc := robot.LastRun.In(now.Location())\n\t\t\t\tif lastRunInLoc.Format(\"15:04\") == currentTime && lastRunInLoc.Day() == now.Day() {\n\t\t\t\t\treturn false // Already triggered this minute today\n\t\t\t\t}\n\t\t\t}\n\t\t\treturn true\n\t\t}\n\t}\n\treturn false\n}\n\n// shouldTriggerInterval checks if enough time has passed since last run\n// interval mode: run every X duration (e.g., \"30m\", \"2h\")\nfunc (m *Manager) shouldTriggerInterval(robot *types.Robot, clock *types.Clock, now time.Time) bool {\n\tinterval, err := time.ParseDuration(clock.Every)\n\tif err != nil {\n\t\treturn false\n\t}\n\n\t// First run if never executed\n\tif robot.LastRun.IsZero() {\n\t\treturn true\n\t}\n\n\t// Check if interval has passed\n\treturn now.Sub(robot.LastRun) >= interval\n}\n\n// shouldTriggerDaemon checks if robot can restart immediately after last run\n// daemon mode: restart immediately after each run completes\nfunc (m *Manager) shouldTriggerDaemon(robot *types.Robot, clock *types.Clock, now time.Time) bool {\n\t// Daemon mode: trigger if not currently running\n\t// CanRun() checks if robot has available execution slots\n\treturn robot.CanRun()\n}\n\n// matchesDay checks if current day matches the configured days\nfunc (m *Manager) matchesDay(clock *types.Clock, now time.Time) bool {\n\t// Empty days or [\"*\"] means all days\n\tif len(clock.Days) == 0 {\n\t\treturn true\n\t}\n\n\tfor _, day := range clock.Days {\n\t\tif day == \"*\" {\n\t\t\treturn true\n\t\t}\n\t\t// Match day name (Mon, Tue, Wed, Thu, Fri, Sat, Sun)\n\t\t// or full name (Monday, Tuesday, etc.)\n\t\tweekday := now.Weekday().String()\n\t\tshortDay := weekday[:3] // Mon, Tue, etc.\n\t\tif day == weekday || day == shortDay {\n\t\t\treturn true\n\t\t}\n\t}\n\treturn false\n}\n\n// TriggerManual manually triggers a robot execution (for testing or API calls)\n// This bypasses clock checking and directly submits to pool\n// For non-autonomous robots: lazy-loads from DB, executes, then unloads\nfunc (m *Manager) TriggerManual(ctx *types.Context, memberID string, trigger types.TriggerType, data interface{}) (string, error) {\n\tm.mu.RLock()\n\tif !m.started {\n\t\tm.mu.RUnlock()\n\t\treturn \"\", fmt.Errorf(\"manager not started\")\n\t}\n\tm.mu.RUnlock()\n\n\t// Get robot from cache, or lazy-load if not found\n\trobot, lazyLoaded, err := m.getOrLoadRobot(ctx, memberID)\n\tif err != nil {\n\t\treturn \"\", err\n\t}\n\n\t// Check robot status\n\tif robot.Status == types.RobotPaused {\n\t\treturn \"\", types.ErrRobotPaused\n\t}\n\n\t// Check if trigger type is enabled\n\tif robot.Config != nil && robot.Config.Triggers != nil {\n\t\tif !robot.Config.Triggers.IsEnabled(trigger) {\n\t\t\treturn \"\", types.ErrTriggerDisabled\n\t\t}\n\t}\n\n\t// Pre-generate execution ID and track for pause/resume/stop\n\t// We need to track BEFORE submit so we can pass the cancellable context to the executor\n\texecID := pool.GenerateExecID()\n\tctrlExec := m.execController.Track(execID, memberID, robot.TeamID)\n\n\t// Create a new context with the cancellable context from ExecutionController\n\t// This allows Stop() to propagate cancellation to the executor\n\texecCtx := types.NewContext(ctrlExec.Context(), ctx.Auth)\n\n\t// Submit to pool with the cancellable context and execution control\n\t// The control interface allows executor to check pause state and wait if paused\n\t_, err = m.pool.SubmitWithID(execCtx, robot, trigger, data, execID, ctrlExec)\n\tif err != nil {\n\t\t// If submission failed, untrack the execution\n\t\tm.execController.Untrack(execID)\n\t\t// If lazy-loaded and submission failed, remove from cache\n\t\tif lazyLoaded {\n\t\t\tm.cache.Remove(memberID)\n\t\t}\n\t\treturn \"\", err\n\t}\n\n\t// For lazy-loaded robots, schedule cleanup after execution completes\n\tif lazyLoaded {\n\t\tm.scheduleCleanup(robot)\n\t}\n\n\treturn execID, nil\n}\n\n// ==================== Human Intervention & Event Triggers ====================\n\n// Intervene processes a human intervention request\n// Human intervention skips P0 (inspiration) and goes directly to P1 (goals)\n// For non-autonomous robots: lazy-loads from DB, executes, then unloads\nfunc (m *Manager) Intervene(ctx *types.Context, req *types.InterveneRequest) (*types.ExecutionResult, error) {\n\tm.mu.RLock()\n\tif !m.started {\n\t\tm.mu.RUnlock()\n\t\treturn nil, fmt.Errorf(\"manager not started\")\n\t}\n\tm.mu.RUnlock()\n\n\t// Validate request\n\tif err := trigger.ValidateIntervention(req); err != nil {\n\t\treturn nil, err\n\t}\n\n\t// Get robot from cache, or lazy-load if not found\n\trobot, lazyLoaded, err := m.getOrLoadRobot(ctx, req.MemberID)\n\tif err != nil {\n\t\treturn nil, err\n\t}\n\n\t// Check robot status\n\tif robot.Status == types.RobotPaused {\n\t\treturn nil, types.ErrRobotPaused\n\t}\n\n\t// Check if human trigger is enabled\n\tif robot.Config != nil && robot.Config.Triggers != nil {\n\t\tif !robot.Config.Triggers.IsEnabled(types.TriggerHuman) {\n\t\t\treturn nil, types.ErrTriggerDisabled\n\t\t}\n\t}\n\n\t// Build trigger input\n\ttriggerInput := &types.TriggerInput{\n\t\tAction: req.Action,\n\t\tMessages: req.Messages,\n\t\tUserID: ctx.UserID(),\n\t}\n\n\t// Handle plan.add action - schedule for later\n\tif req.Action == types.ActionPlanAdd && req.PlanTime != nil {\n\t\t// If lazy-loaded but not executing, remove immediately\n\t\tif lazyLoaded {\n\t\t\tm.cache.Remove(req.MemberID)\n\t\t}\n\t\t// TODO: Add to plan queue (Phase 11.3)\n\t\treturn &types.ExecutionResult{\n\t\t\tStatus: types.ExecPending,\n\t\t\tMessage: fmt.Sprintf(\"Planned for %s (plan queue not implemented yet)\", req.PlanTime.Format(time.RFC3339)),\n\t\t}, nil\n\t}\n\n\t// Determine executor mode: request > robot config > default\n\texecutorMode := m.resolveExecutorMode(req.ExecutorMode, robot)\n\n\t// Submit to pool with executor mode\n\texecID, err := m.pool.SubmitWithMode(ctx, robot, types.TriggerHuman, triggerInput, executorMode)\n\tif err != nil {\n\t\t// If lazy-loaded and submission failed, remove from cache\n\t\tif lazyLoaded {\n\t\t\tm.cache.Remove(req.MemberID)\n\t\t}\n\t\treturn nil, err\n\t}\n\n\t// Track execution for pause/resume/stop\n\tm.execController.Track(execID, req.MemberID, req.TeamID)\n\n\t// For lazy-loaded robots, schedule cleanup after execution completes\n\tif lazyLoaded {\n\t\tm.scheduleCleanup(robot)\n\t}\n\n\treturn &types.ExecutionResult{\n\t\tExecutionID: execID,\n\t\tStatus: types.ExecPending,\n\t\tMessage: fmt.Sprintf(\"Human intervention (%s) submitted\", req.Action),\n\t}, nil\n}\n\n// HandleEvent processes an event trigger request\n// Event trigger skips P0 (inspiration) and goes directly to P1 (goals)\n// For non-autonomous robots: lazy-loads from DB, executes, then unloads\nfunc (m *Manager) HandleEvent(ctx *types.Context, req *types.EventRequest) (*types.ExecutionResult, error) {\n\tm.mu.RLock()\n\tif !m.started {\n\t\tm.mu.RUnlock()\n\t\treturn nil, fmt.Errorf(\"manager not started\")\n\t}\n\tm.mu.RUnlock()\n\n\t// Validate request\n\tif err := trigger.ValidateEvent(req); err != nil {\n\t\treturn nil, err\n\t}\n\n\t// Get robot from cache, or lazy-load if not found\n\trobot, lazyLoaded, err := m.getOrLoadRobot(ctx, req.MemberID)\n\tif err != nil {\n\t\treturn nil, err\n\t}\n\n\t// Check robot status\n\tif robot.Status == types.RobotPaused {\n\t\treturn nil, types.ErrRobotPaused\n\t}\n\n\t// Check if event trigger is enabled\n\tif robot.Config != nil && robot.Config.Triggers != nil {\n\t\tif !robot.Config.Triggers.IsEnabled(types.TriggerEvent) {\n\t\t\treturn nil, types.ErrTriggerDisabled\n\t\t}\n\t}\n\n\t// Build trigger input\n\ttriggerInput := trigger.BuildEventInput(req)\n\n\t// Determine executor mode: request > robot config > default\n\texecutorMode := m.resolveExecutorMode(req.ExecutorMode, robot)\n\n\t// Submit to pool with executor mode\n\texecID, err := m.pool.SubmitWithMode(ctx, robot, types.TriggerEvent, triggerInput, executorMode)\n\tif err != nil {\n\t\t// If lazy-loaded and submission failed, remove from cache\n\t\tif lazyLoaded {\n\t\t\tm.cache.Remove(req.MemberID)\n\t\t}\n\t\treturn nil, err\n\t}\n\n\t// Track execution for pause/resume/stop\n\tm.execController.Track(execID, req.MemberID, \"\")\n\n\t// For lazy-loaded robots, schedule cleanup after execution completes\n\tif lazyLoaded {\n\t\tm.scheduleCleanup(robot)\n\t}\n\n\treturn &types.ExecutionResult{\n\t\tExecutionID: execID,\n\t\tStatus: types.ExecPending,\n\t\tMessage: fmt.Sprintf(\"Event trigger (%s: %s) submitted\", req.Source, req.EventType),\n\t}, nil\n}\n\n// ==================== Execution Control ====================\n\n// PauseExecution pauses a running execution\nfunc (m *Manager) PauseExecution(ctx *types.Context, execID string) error {\n\t// Get execution info before pausing\n\texec := m.execController.Get(execID)\n\tif exec == nil {\n\t\treturn fmt.Errorf(\"execution not found: %s\", execID)\n\t}\n\n\t// Pause the execution\n\tif err := m.execController.Pause(execID); err != nil {\n\t\treturn err\n\t}\n\n\t// Remove from robot's in-memory execution list (paused doesn't count as running)\n\tif robot := m.cache.Get(exec.MemberID); robot != nil {\n\t\trobot.RemoveExecution(execID)\n\t}\n\n\treturn nil\n}\n\n// ResumeExecution resumes a paused execution\nfunc (m *Manager) ResumeExecution(ctx *types.Context, execID string) error {\n\t// Get execution info before resuming\n\texec := m.execController.Get(execID)\n\tif exec == nil {\n\t\treturn fmt.Errorf(\"execution not found: %s\", execID)\n\t}\n\n\t// Resume the execution\n\tif err := m.execController.Resume(execID); err != nil {\n\t\treturn err\n\t}\n\n\t// Add back to robot's in-memory execution list\n\tif robot := m.cache.Get(exec.MemberID); robot != nil {\n\t\trobot.AddExecution(&types.Execution{\n\t\t\tID: execID,\n\t\t\tMemberID: exec.MemberID,\n\t\t\tTeamID: exec.TeamID,\n\t\t\tStatus: types.ExecRunning,\n\t\t})\n\t}\n\n\treturn nil\n}\n\n// StopExecution stops a running execution\nfunc (m *Manager) StopExecution(ctx *types.Context, execID string) error {\n\t// Get execution info before stopping\n\texec := m.execController.Get(execID)\n\tif exec == nil {\n\t\treturn fmt.Errorf(\"execution not found: %s\", execID)\n\t}\n\n\t// Stop the execution\n\tif err := m.execController.Stop(execID); err != nil {\n\t\treturn err\n\t}\n\n\t// Remove from robot's in-memory execution list\n\tif robot := m.cache.Get(exec.MemberID); robot != nil {\n\t\trobot.RemoveExecution(execID)\n\t}\n\n\treturn nil\n}\n\n// GetExecutionStatus returns the status of an execution\nfunc (m *Manager) GetExecutionStatus(execID string) (*trigger.ControlledExecution, error) {\n\texec := m.execController.Get(execID)\n\tif exec == nil {\n\t\treturn nil, fmt.Errorf(\"execution not found: %s\", execID)\n\t}\n\treturn exec, nil\n}\n\n// ListExecutions returns all tracked executions\nfunc (m *Manager) ListExecutions() []*trigger.ControlledExecution {\n\treturn m.execController.List()\n}\n\n// ListExecutionsByMember returns all executions for a specific robot\nfunc (m *Manager) ListExecutionsByMember(memberID string) []*trigger.ControlledExecution {\n\treturn m.execController.ListByMember(memberID)\n}\n\n// ==================== Helper Methods ====================\n\n// getOrLoadRobot gets a robot from cache, or lazy-loads from DB if not found\n// Returns: robot, wasLazyLoaded, error\nfunc (m *Manager) getOrLoadRobot(ctx *types.Context, memberID string) (*types.Robot, bool, error) {\n\t// Try cache first\n\trobot := m.cache.Get(memberID)\n\tif robot != nil {\n\t\treturn robot, false, nil\n\t}\n\n\t// Not in cache - lazy load from database\n\trobot, err := m.cache.LoadByID(ctx, memberID)\n\tif err != nil {\n\t\treturn nil, false, err\n\t}\n\n\t// Add to cache temporarily for execution tracking\n\tm.cache.Add(robot)\n\n\t// Return with lazyLoaded=true to indicate cleanup needed after execution\n\treturn robot, true, nil\n}\n\n// scheduleCleanup schedules removal of a lazy-loaded robot after all executions complete\n// This runs in a goroutine that monitors the robot's execution count\nfunc (m *Manager) scheduleCleanup(robot *types.Robot) {\n\tgo func() {\n\t\tmemberID := robot.MemberID\n\n\t\t// Poll every 5 seconds to check if all executions are done\n\t\tticker := time.NewTicker(5 * time.Second)\n\t\tdefer ticker.Stop()\n\n\t\t// Timeout after 24 hours to prevent memory leaks\n\t\ttimeout := time.After(24 * time.Hour)\n\n\t\tfor {\n\t\t\tselect {\n\t\t\tcase <-timeout:\n\t\t\t\t// Timeout - force cleanup\n\t\t\t\tm.cache.Remove(memberID)\n\t\t\t\treturn\n\n\t\t\tcase <-ticker.C:\n\t\t\t\t// Check if robot still exists in cache\n\t\t\t\tr := m.cache.Get(memberID)\n\t\t\t\tif r == nil {\n\t\t\t\t\t// Already removed\n\t\t\t\t\treturn\n\t\t\t\t}\n\n\t\t\t\t// Check if all executions are done\n\t\t\t\tif r.RunningCount() == 0 {\n\t\t\t\t\t// Only remove if still non-autonomous\n\t\t\t\t\t// (user might have changed it during execution)\n\t\t\t\t\tif !r.AutonomousMode {\n\t\t\t\t\t\tm.cache.Remove(memberID)\n\t\t\t\t\t}\n\t\t\t\t\treturn\n\t\t\t\t}\n\t\t\t}\n\t\t}\n\t}()\n}\n\n// resolveExecutorMode determines the executor mode to use\n// Priority: request > robot config > default (standard)\nfunc (m *Manager) resolveExecutorMode(requestMode types.ExecutorMode, robot *types.Robot) types.ExecutorMode {\n\t// Request mode takes precedence\n\tif requestMode != \"\" && requestMode.IsValid() {\n\t\treturn requestMode\n\t}\n\n\t// Robot config mode\n\tif robot != nil && robot.Config != nil && robot.Config.Executor != nil {\n\t\treturn robot.Config.Executor.GetMode()\n\t}\n\n\t// Default: standard\n\treturn types.ExecutorStandard\n}\n\n// ==================== Getters for internal components ====================\n// These are exposed for testing and advanced use cases\n\n// Cache returns the internal cache\nfunc (m *Manager) Cache() *cache.Cache {\n\treturn m.cache\n}\n\n// Pool returns the internal pool\nfunc (m *Manager) Pool() *pool.Pool {\n\treturn m.pool\n}\n\n// Executor returns the internal executor\nfunc (m *Manager) Executor() types.Executor {\n\treturn m.executor\n}\n\n// IsStarted returns true if manager is started\nfunc (m *Manager) IsStarted() bool {\n\tm.mu.RLock()\n\tdefer m.mu.RUnlock()\n\treturn m.started\n}\n\n// Running returns number of currently running jobs\nfunc (m *Manager) Running() int {\n\treturn m.pool.Running()\n}\n\n// Queued returns number of queued jobs\nfunc (m *Manager) Queued() int {\n\treturn m.pool.Queued()\n}\n\n// CachedRobots returns number of cached robots\nfunc (m *Manager) CachedRobots() int {\n\treturn m.cache.Count()\n}\n" }, { "path": "agent/robot/manager/manager_test.go", "content": "package manager_test\n\nimport (\n\t\"context\"\n\t\"encoding/json\"\n\t\"runtime\"\n\t\"testing\"\n\t\"time\"\n\n\t\"github.com/stretchr/testify/assert\"\n\t\"github.com/yaoapp/gou/model\"\n\t\"github.com/yaoapp/xun/capsule\"\n\tagentcontext \"github.com/yaoapp/yao/agent/context\"\n\t\"github.com/yaoapp/yao/agent/robot/manager\"\n\t\"github.com/yaoapp/yao/agent/robot/pool\"\n\t\"github.com/yaoapp/yao/agent/robot/types\"\n\t\"github.com/yaoapp/yao/agent/testutils\"\n)\n\n// TestManagerStartStop tests manager lifecycle\nfunc TestManagerStartStop(t *testing.T) {\n\tif testing.Short() {\n\t\tt.Skip(\"Skipping integration test\")\n\t}\n\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\tcleanupTestRobots(t)\n\tsetupTestRobots(t)\n\tdefer cleanupTestRobots(t)\n\n\tt.Run(\"start and stop manager\", func(t *testing.T) {\n\t\tm := manager.New()\n\n\t\t// Should not be started\n\t\tassert.False(t, m.IsStarted())\n\n\t\t// Start manager\n\t\terr := m.Start()\n\t\tassert.NoError(t, err)\n\t\tassert.True(t, m.IsStarted())\n\n\t\t// Robots should be loaded\n\t\tassert.GreaterOrEqual(t, m.CachedRobots(), 2, \"Should load at least 2 robots\")\n\n\t\t// Stop manager\n\t\terr = m.Stop()\n\t\tassert.NoError(t, err)\n\t\tassert.False(t, m.IsStarted())\n\t})\n\n\tt.Run(\"double start should fail\", func(t *testing.T) {\n\t\tm := manager.New()\n\n\t\terr := m.Start()\n\t\tassert.NoError(t, err)\n\n\t\t// Second start should fail\n\t\terr = m.Start()\n\t\tassert.Error(t, err)\n\t\tassert.Contains(t, err.Error(), \"already started\")\n\n\t\t// Cleanup\n\t\tm.Stop()\n\t})\n\n\tt.Run(\"stop without start should not panic\", func(t *testing.T) {\n\t\tm := manager.New()\n\n\t\tassert.NotPanics(t, func() {\n\t\t\terr := m.Stop()\n\t\t\tassert.NoError(t, err)\n\t\t})\n\t})\n}\n\n// TestManagerTick tests the Tick function with different clock modes\nfunc TestManagerTick(t *testing.T) {\n\tif testing.Short() {\n\t\tt.Skip(\"Skipping integration test\")\n\t}\n\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\tcleanupTestRobots(t)\n\tsetupTestRobotsWithClockConfig(t)\n\tdefer cleanupTestRobots(t)\n\n\tt.Run(\"tick with times mode - matching time\", func(t *testing.T) {\n\t\t// Create manager with short tick interval for testing\n\t\tconfig := &manager.Config{\n\t\t\tTickInterval: 100 * time.Millisecond,\n\t\t\tPoolConfig: &pool.Config{WorkerSize: 2, QueueSize: 10},\n\t\t}\n\t\tm := manager.NewWithConfig(config)\n\n\t\terr := m.Start()\n\t\tassert.NoError(t, err)\n\t\tdefer m.Stop()\n\n\t\t// Create a time that matches the configured time (09:00)\n\t\tloc, _ := time.LoadLocation(\"Asia/Shanghai\")\n\t\tnow := time.Date(2025, 1, 15, 9, 0, 0, 0, loc) // Wednesday 09:00\n\n\t\tctx := types.NewContext(context.Background(), nil)\n\t\terr = m.Tick(ctx, now)\n\t\tassert.NoError(t, err)\n\n\t\t// Wait for job to be processed\n\t\ttime.Sleep(200 * time.Millisecond)\n\n\t\t// Check that job was submitted (may be queued or running)\n\t\t// Note: The executor stub completes quickly, so we check execution count\n\t\texecCount := m.Executor().ExecCount()\n\t\tassert.GreaterOrEqual(t, execCount, 1, \"Should have executed at least 1 job\")\n\t})\n\n\tt.Run(\"tick with times mode - non-matching time\", func(t *testing.T) {\n\t\tconfig := &manager.Config{\n\t\t\tTickInterval: 100 * time.Millisecond,\n\t\t\tPoolConfig: &pool.Config{WorkerSize: 2, QueueSize: 10},\n\t\t}\n\t\tm := manager.NewWithConfig(config)\n\n\t\terr := m.Start()\n\t\tassert.NoError(t, err)\n\t\tdefer m.Stop()\n\n\t\t// Reset executor count\n\t\tm.Executor().Reset()\n\n\t\t// Create a time that does NOT match (10:30)\n\t\tloc, _ := time.LoadLocation(\"Asia/Shanghai\")\n\t\tnow := time.Date(2025, 1, 15, 10, 30, 0, 0, loc) // Wednesday 10:30\n\n\t\tctx := types.NewContext(context.Background(), nil)\n\t\terr = m.Tick(ctx, now)\n\t\tassert.NoError(t, err)\n\n\t\t// Wait a bit\n\t\ttime.Sleep(100 * time.Millisecond)\n\n\t\t// Should not have triggered (times mode robot only triggers at 09:00, 14:00)\n\t\texecCount := m.Executor().ExecCount()\n\t\t// Note: interval mode robot might trigger if enough time passed\n\t\t// We just verify the times mode robot didn't trigger\n\t\tassert.LessOrEqual(t, execCount, 1, \"Times mode robot should not trigger at non-matching time\")\n\t})\n\n\tt.Run(\"tick with interval mode\", func(t *testing.T) {\n\t\tconfig := &manager.Config{\n\t\t\tTickInterval: 100 * time.Millisecond,\n\t\t\tPoolConfig: &pool.Config{WorkerSize: 2, QueueSize: 10},\n\t\t}\n\t\tm := manager.NewWithConfig(config)\n\n\t\terr := m.Start()\n\t\tassert.NoError(t, err)\n\t\tdefer m.Stop()\n\n\t\t// Reset executor count\n\t\tm.Executor().Reset()\n\n\t\t// First tick - should trigger interval mode robot (first run)\n\t\tctx := types.NewContext(context.Background(), nil)\n\t\tnow := time.Now()\n\t\terr = m.Tick(ctx, now)\n\t\tassert.NoError(t, err)\n\n\t\t// Wait for execution\n\t\ttime.Sleep(200 * time.Millisecond)\n\n\t\t// Should have at least 1 execution (interval robot first run)\n\t\texecCount := m.Executor().ExecCount()\n\t\tassert.GreaterOrEqual(t, execCount, 1, \"Interval mode robot should trigger on first run\")\n\t})\n\n\tt.Run(\"tick skips paused robots\", func(t *testing.T) {\n\t\tconfig := &manager.Config{\n\t\t\tTickInterval: 100 * time.Millisecond,\n\t\t\tPoolConfig: &pool.Config{WorkerSize: 2, QueueSize: 10},\n\t\t}\n\t\tm := manager.NewWithConfig(config)\n\n\t\terr := m.Start()\n\t\tassert.NoError(t, err)\n\t\tdefer m.Stop()\n\n\t\t// Get the paused robot from cache\n\t\tpausedRobot := m.Cache().Get(\"robot_test_manager_paused\")\n\t\tassert.NotNil(t, pausedRobot)\n\t\tassert.Equal(t, types.RobotPaused, pausedRobot.Status)\n\n\t\t// Reset executor count\n\t\tm.Executor().Reset()\n\n\t\t// Tick should skip paused robot\n\t\tctx := types.NewContext(context.Background(), nil)\n\t\tloc, _ := time.LoadLocation(\"Asia/Shanghai\")\n\t\tnow := time.Date(2025, 1, 15, 9, 0, 0, 0, loc)\n\t\terr = m.Tick(ctx, now)\n\t\tassert.NoError(t, err)\n\n\t\t// The paused robot should not have been triggered\n\t\t// (we can't directly verify this, but we verify the tick completed)\n\t})\n}\n\n// TestManagerTriggerManual tests manual triggering of robots\nfunc TestManagerTriggerManual(t *testing.T) {\n\tif testing.Short() {\n\t\tt.Skip(\"Skipping integration test\")\n\t}\n\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\tcleanupTestRobots(t)\n\tsetupTestRobotsWithClockConfig(t)\n\tdefer cleanupTestRobots(t)\n\n\tt.Run(\"trigger manual - success\", func(t *testing.T) {\n\t\tm := manager.New()\n\t\terr := m.Start()\n\t\tassert.NoError(t, err)\n\t\tdefer m.Stop()\n\n\t\tctx := types.NewContext(context.Background(), nil)\n\n\t\t// Manually trigger a robot\n\t\texecID, err := m.TriggerManual(ctx, \"robot_test_manager_times\", types.TriggerHuman, nil)\n\t\tassert.NoError(t, err)\n\t\tassert.NotEmpty(t, execID)\n\n\t\t// Wait for execution\n\t\ttime.Sleep(200 * time.Millisecond)\n\n\t\t// Should have executed\n\t\tassert.GreaterOrEqual(t, m.Executor().ExecCount(), 1)\n\t})\n\n\tt.Run(\"trigger manual - robot not found\", func(t *testing.T) {\n\t\tm := manager.New()\n\t\terr := m.Start()\n\t\tassert.NoError(t, err)\n\t\tdefer m.Stop()\n\n\t\tctx := types.NewContext(context.Background(), nil)\n\n\t\t// Try to trigger non-existent robot\n\t\t_, err = m.TriggerManual(ctx, \"robot_nonexistent\", types.TriggerHuman, nil)\n\t\tassert.Error(t, err)\n\t\tassert.Equal(t, types.ErrRobotNotFound, err)\n\t})\n\n\tt.Run(\"trigger manual - robot paused\", func(t *testing.T) {\n\t\tm := manager.New()\n\t\terr := m.Start()\n\t\tassert.NoError(t, err)\n\t\tdefer m.Stop()\n\n\t\tctx := types.NewContext(context.Background(), nil)\n\n\t\t// Try to trigger paused robot\n\t\t_, err = m.TriggerManual(ctx, \"robot_test_manager_paused\", types.TriggerHuman, nil)\n\t\tassert.Error(t, err)\n\t\tassert.Equal(t, types.ErrRobotPaused, err)\n\t})\n\n\tt.Run(\"trigger manual - manager not started\", func(t *testing.T) {\n\t\tm := manager.New()\n\t\t// Don't start manager\n\n\t\tctx := types.NewContext(context.Background(), nil)\n\n\t\t_, err := m.TriggerManual(ctx, \"robot_test_manager_times\", types.TriggerHuman, nil)\n\t\tassert.Error(t, err)\n\t\tassert.Contains(t, err.Error(), \"not started\")\n\t})\n}\n\n// TestManagerClockModes tests all three clock modes\nfunc TestManagerClockModes(t *testing.T) {\n\tif testing.Short() {\n\t\tt.Skip(\"Skipping integration test\")\n\t}\n\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\tcleanupTestRobots(t)\n\tsetupTestRobotsWithClockConfig(t)\n\tdefer cleanupTestRobots(t)\n\n\tt.Run(\"times mode - day matching\", func(t *testing.T) {\n\t\tm := manager.New()\n\t\terr := m.Start()\n\t\tassert.NoError(t, err)\n\t\tdefer m.Stop()\n\n\t\tm.Executor().Reset()\n\n\t\t// Wednesday (configured day)\n\t\tloc, _ := time.LoadLocation(\"Asia/Shanghai\")\n\t\tnow := time.Date(2025, 1, 15, 9, 0, 0, 0, loc) // Wednesday 09:00\n\n\t\tctx := types.NewContext(context.Background(), nil)\n\t\terr = m.Tick(ctx, now)\n\t\tassert.NoError(t, err)\n\n\t\ttime.Sleep(200 * time.Millisecond)\n\t\tassert.GreaterOrEqual(t, m.Executor().ExecCount(), 1, \"Should trigger on matching day\")\n\t})\n\n\tt.Run(\"times mode - day not matching\", func(t *testing.T) {\n\t\tm := manager.New()\n\t\terr := m.Start()\n\t\tassert.NoError(t, err)\n\t\tdefer m.Stop()\n\n\t\tm.Executor().Reset()\n\n\t\t// Saturday (not configured)\n\t\tloc, _ := time.LoadLocation(\"Asia/Shanghai\")\n\t\tnow := time.Date(2025, 1, 18, 9, 0, 0, 0, loc) // Saturday 09:00\n\n\t\tctx := types.NewContext(context.Background(), nil)\n\t\terr = m.Tick(ctx, now)\n\t\tassert.NoError(t, err)\n\n\t\ttime.Sleep(100 * time.Millisecond)\n\t\t// Times mode robot should not trigger on Saturday\n\t\t// Only interval/daemon robots might trigger\n\t})\n\n\tt.Run(\"daemon mode - always triggers when idle\", func(t *testing.T) {\n\t\tm := manager.New()\n\t\terr := m.Start()\n\t\tassert.NoError(t, err)\n\t\tdefer m.Stop()\n\n\t\tm.Executor().Reset()\n\n\t\t// Daemon robot should trigger whenever it can run\n\t\tctx := types.NewContext(context.Background(), nil)\n\t\tnow := time.Now()\n\n\t\t// First tick\n\t\terr = m.Tick(ctx, now)\n\t\tassert.NoError(t, err)\n\n\t\ttime.Sleep(200 * time.Millisecond)\n\n\t\t// Should have triggered daemon robot\n\t\tassert.GreaterOrEqual(t, m.Executor().ExecCount(), 1, \"Daemon mode should trigger\")\n\t})\n}\n\n// TestManagerTimezoneDedup tests that times mode dedup works correctly across timezones\n// This specifically tests the bug fix where LastRun.Day() must be converted to the same\n// timezone as 'now' before comparison\nfunc TestManagerTimezoneDedup(t *testing.T) {\n\tif testing.Short() {\n\t\tt.Skip(\"Skipping integration test\")\n\t}\n\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\tcleanupTestRobots(t)\n\tsetupTestRobotsWithClockConfig(t)\n\tdefer cleanupTestRobots(t)\n\n\tt.Run(\"times mode - same minute same day should not trigger twice\", func(t *testing.T) {\n\t\tm := manager.New()\n\t\terr := m.Start()\n\t\tassert.NoError(t, err)\n\t\tdefer m.Stop()\n\n\t\tm.Executor().Reset()\n\n\t\t// Use Asia/Shanghai timezone (UTC+8)\n\t\tloc, _ := time.LoadLocation(\"Asia/Shanghai\")\n\t\t// Wednesday 09:00:00 in Shanghai\n\t\tnow := time.Date(2025, 1, 15, 9, 0, 0, 0, loc)\n\n\t\tctx := types.NewContext(context.Background(), nil)\n\n\t\t// First tick - should trigger\n\t\terr = m.Tick(ctx, now)\n\t\tassert.NoError(t, err)\n\t\ttime.Sleep(200 * time.Millisecond)\n\t\tfirstCount := m.Executor().ExecCount()\n\t\tassert.GreaterOrEqual(t, firstCount, 1, \"First tick should trigger\")\n\n\t\t// Second tick at 09:00:30 (same minute) - should NOT trigger again\n\t\tnow2 := time.Date(2025, 1, 15, 9, 0, 30, 0, loc)\n\t\terr = m.Tick(ctx, now2)\n\t\tassert.NoError(t, err)\n\t\ttime.Sleep(200 * time.Millisecond)\n\n\t\t// Count should remain the same (times robot should not trigger twice)\n\t\t// Note: daemon/interval robots may still trigger, so we check the delta\n\t\tsecondCount := m.Executor().ExecCount()\n\t\tt.Logf(\"First count: %d, Second count: %d\", firstCount, secondCount)\n\n\t\t// The times robot should not have triggered again in the same minute\n\t\t// Delta should be <= 2 (daemon always triggers, interval might trigger)\n\t\t// If delta > 2, it means times robot triggered twice (bug!)\n\t\tdelta := secondCount - firstCount\n\t\tassert.LessOrEqual(t, delta, 2, \"Times robot should not trigger twice in same minute (delta: %d)\", delta)\n\t})\n\n\tt.Run(\"times mode - different day should trigger again\", func(t *testing.T) {\n\t\tm := manager.New()\n\t\terr := m.Start()\n\t\tassert.NoError(t, err)\n\t\tdefer m.Stop()\n\n\t\tm.Executor().Reset()\n\n\t\tloc, _ := time.LoadLocation(\"Asia/Shanghai\")\n\n\t\tctx := types.NewContext(context.Background(), nil)\n\n\t\t// Wednesday 09:00\n\t\tnow1 := time.Date(2025, 1, 15, 9, 0, 0, 0, loc)\n\t\terr = m.Tick(ctx, now1)\n\t\tassert.NoError(t, err)\n\t\ttime.Sleep(200 * time.Millisecond)\n\t\tfirstCount := m.Executor().ExecCount()\n\n\t\t// Thursday 09:00 (next day, same time)\n\t\tnow2 := time.Date(2025, 1, 16, 9, 0, 0, 0, loc)\n\t\terr = m.Tick(ctx, now2)\n\t\tassert.NoError(t, err)\n\t\ttime.Sleep(200 * time.Millisecond)\n\n\t\t// Should have triggered again on the new day\n\t\tsecondCount := m.Executor().ExecCount()\n\t\tassert.Greater(t, secondCount, firstCount, \"Should trigger on different day\")\n\t})\n\n\tt.Run(\"times mode - cross-timezone day boundary\", func(t *testing.T) {\n\t\tm := manager.New()\n\t\terr := m.Start()\n\t\tassert.NoError(t, err)\n\t\tdefer m.Stop()\n\n\t\tm.Executor().Reset()\n\n\t\t// Robot is configured with Asia/Shanghai (UTC+8)\n\t\t// Test case: LastRun was set when it was Jan 15 in Shanghai\n\t\t// Now it's Jan 16 00:30 in Shanghai (still Jan 15 in UTC)\n\t\t// The comparison should use Shanghai timezone, not UTC\n\n\t\tloc, _ := time.LoadLocation(\"Asia/Shanghai\")\n\t\tctx := types.NewContext(context.Background(), nil)\n\n\t\t// First run: Jan 15, 09:00 Shanghai time\n\t\tnow1 := time.Date(2025, 1, 15, 9, 0, 0, 0, loc)\n\t\terr = m.Tick(ctx, now1)\n\t\tassert.NoError(t, err)\n\t\ttime.Sleep(200 * time.Millisecond)\n\n\t\t// Second run: Jan 16, 09:00 Shanghai time\n\t\t// This is Jan 16 01:00 UTC, but should be treated as Jan 16 in Shanghai\n\t\tnow2 := time.Date(2025, 1, 16, 9, 0, 0, 0, loc)\n\t\terr = m.Tick(ctx, now2)\n\t\tassert.NoError(t, err)\n\t\ttime.Sleep(200 * time.Millisecond)\n\n\t\t// Should have triggered on both days\n\t\tassert.GreaterOrEqual(t, m.Executor().ExecCount(), 2, \"Should trigger on both days\")\n\t})\n\n\tt.Run(\"times mode - UTC vs local timezone comparison\", func(t *testing.T) {\n\t\tm := manager.New()\n\t\terr := m.Start()\n\t\tassert.NoError(t, err)\n\t\tdefer m.Stop()\n\n\t\tm.Executor().Reset()\n\n\t\t// Test with explicit UTC time converted to Shanghai\n\t\t// This tests that LastRun stored in one timezone is correctly\n\t\t// compared when 'now' is in a different timezone\n\n\t\tshanghai, _ := time.LoadLocation(\"Asia/Shanghai\")\n\t\tctx := types.NewContext(context.Background(), nil)\n\n\t\t// Create a time that's Jan 15 23:30 UTC = Jan 16 07:30 Shanghai\n\t\tutcTime := time.Date(2025, 1, 15, 23, 30, 0, 0, time.UTC)\n\t\tshanghaiTime := utcTime.In(shanghai)\n\t\tt.Logf(\"UTC: %v, Shanghai: %v\", utcTime, shanghaiTime)\n\n\t\t// The robot is configured for 09:00 Shanghai time\n\t\t// So Jan 16 09:00 Shanghai should trigger\n\t\tnow := time.Date(2025, 1, 16, 9, 0, 0, 0, shanghai)\n\t\terr = m.Tick(ctx, now)\n\t\tassert.NoError(t, err)\n\t\ttime.Sleep(200 * time.Millisecond)\n\n\t\tassert.GreaterOrEqual(t, m.Executor().ExecCount(), 1, \"Should trigger at 09:00 Shanghai\")\n\t})\n}\n\n// TestManagerGoroutineLeak tests that manager doesn't leak goroutines\nfunc TestManagerGoroutineLeak(t *testing.T) {\n\tif testing.Short() {\n\t\tt.Skip(\"Skipping integration test\")\n\t}\n\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\tcleanupTestRobots(t)\n\tsetupTestRobots(t)\n\tdefer cleanupTestRobots(t)\n\n\tt.Run(\"start stop cycle should not leak goroutines\", func(t *testing.T) {\n\t\t// Record initial goroutine count\n\t\truntime.GC()\n\t\ttime.Sleep(100 * time.Millisecond)\n\t\tinitialGoroutines := runtime.NumGoroutine()\n\n\t\t// Start and stop multiple times\n\t\tfor i := 0; i < 5; i++ {\n\t\t\tm := manager.New()\n\t\t\terr := m.Start()\n\t\t\tassert.NoError(t, err)\n\n\t\t\t// Do some ticks\n\t\t\tctx := types.NewContext(context.Background(), nil)\n\t\t\tm.Tick(ctx, time.Now())\n\n\t\t\ttime.Sleep(50 * time.Millisecond)\n\n\t\t\terr = m.Stop()\n\t\t\tassert.NoError(t, err)\n\t\t}\n\n\t\t// Wait for cleanup\n\t\ttime.Sleep(200 * time.Millisecond)\n\t\truntime.GC()\n\t\ttime.Sleep(100 * time.Millisecond)\n\n\t\t// Check goroutine count\n\t\tfinalGoroutines := runtime.NumGoroutine()\n\t\tassert.LessOrEqual(t, finalGoroutines, initialGoroutines+2,\n\t\t\t\"Should not leak goroutines (initial: %d, final: %d)\",\n\t\t\tinitialGoroutines, finalGoroutines)\n\t})\n}\n\n// TestManagerComponents tests access to internal components\nfunc TestManagerComponents(t *testing.T) {\n\tif testing.Short() {\n\t\tt.Skip(\"Skipping integration test\")\n\t}\n\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\tcleanupTestRobots(t)\n\tsetupTestRobots(t)\n\tdefer cleanupTestRobots(t)\n\n\tm := manager.New()\n\terr := m.Start()\n\tassert.NoError(t, err)\n\tdefer m.Stop()\n\n\tt.Run(\"cache access\", func(t *testing.T) {\n\t\tcache := m.Cache()\n\t\tassert.NotNil(t, cache)\n\n\t\trobot := cache.Get(\"robot_test_sales_001\")\n\t\tassert.NotNil(t, robot)\n\t})\n\n\tt.Run(\"pool access\", func(t *testing.T) {\n\t\tpool := m.Pool()\n\t\tassert.NotNil(t, pool)\n\t\tassert.True(t, pool.IsStarted())\n\t})\n\n\tt.Run(\"executor access\", func(t *testing.T) {\n\t\texecutor := m.Executor()\n\t\tassert.NotNil(t, executor)\n\t})\n\n\tt.Run(\"running and queued counts\", func(t *testing.T) {\n\t\trunning := m.Running()\n\t\tqueued := m.Queued()\n\t\tcached := m.CachedRobots()\n\n\t\tassert.GreaterOrEqual(t, running, 0)\n\t\tassert.GreaterOrEqual(t, queued, 0)\n\t\tassert.GreaterOrEqual(t, cached, 2)\n\t})\n}\n\n// ==================== Test Data Setup ====================\n\n// setupTestRobots creates basic test robots (same as cache tests)\nfunc setupTestRobots(t *testing.T) {\n\tm := model.Select(\"__yao.member\")\n\ttableName := m.MetaData.Table.Name\n\tqb := capsule.Query()\n\n\t// Robot 1: Sales Bot\n\trobotConfig1 := map[string]interface{}{\n\t\t\"identity\": map[string]interface{}{\n\t\t\t\"role\": \"Sales Manager\",\n\t\t\t\"duties\": []string{\"Manage leads\", \"Follow up customers\"},\n\t\t},\n\t\t\"quota\": map[string]interface{}{\n\t\t\t\"max\": 3,\n\t\t\t\"queue\": 15,\n\t\t\t\"priority\": 7,\n\t\t},\n\t\t\"triggers\": map[string]interface{}{\n\t\t\t\"clock\": map[string]interface{}{\"enabled\": true},\n\t\t},\n\t\t\"clock\": map[string]interface{}{\n\t\t\t\"mode\": \"times\",\n\t\t\t\"times\": []string{\"09:00\", \"14:00\"},\n\t\t\t\"days\": []string{\"Mon\", \"Tue\", \"Wed\", \"Thu\", \"Fri\"},\n\t\t\t\"tz\": \"Asia/Shanghai\",\n\t\t},\n\t}\n\tconfig1JSON, _ := json.Marshal(robotConfig1)\n\n\terr := qb.Table(tableName).Insert([]map[string]interface{}{\n\t\t{\n\t\t\t\"member_id\": \"robot_test_sales_001\",\n\t\t\t\"team_id\": \"team_test_cache_001\",\n\t\t\t\"member_type\": \"robot\",\n\t\t\t\"display_name\": \"Test Sales Bot\",\n\t\t\t\"system_prompt\": \"You are a professional sales manager assistant.\",\n\t\t\t\"status\": \"active\",\n\t\t\t\"role_id\": \"member\",\n\t\t\t\"autonomous_mode\": true,\n\t\t\t\"robot_status\": \"idle\",\n\t\t\t\"robot_config\": string(config1JSON),\n\t\t},\n\t})\n\tif err != nil {\n\t\tt.Fatalf(\"Failed to insert robot_test_sales_001: %v\", err)\n\t}\n\n\t// Robot 2: Support Bot\n\trobotConfig2 := map[string]interface{}{\n\t\t\"identity\": map[string]interface{}{\n\t\t\t\"role\": \"Customer Support\",\n\t\t\t\"duties\": []string{\"Answer questions\", \"Resolve issues\"},\n\t\t},\n\t\t\"quota\": map[string]interface{}{\n\t\t\t\"max\": 2,\n\t\t\t\"queue\": 10,\n\t\t\t\"priority\": 5,\n\t\t},\n\t\t\"triggers\": map[string]interface{}{\n\t\t\t\"clock\": map[string]interface{}{\"enabled\": true},\n\t\t},\n\t\t\"clock\": map[string]interface{}{\n\t\t\t\"mode\": \"interval\",\n\t\t\t\"every\": \"1h\",\n\t\t},\n\t}\n\tconfig2JSON, _ := json.Marshal(robotConfig2)\n\n\terr = qb.Table(tableName).Insert([]map[string]interface{}{\n\t\t{\n\t\t\t\"member_id\": \"robot_test_support_002\",\n\t\t\t\"team_id\": \"team_test_cache_001\",\n\t\t\t\"member_type\": \"robot\",\n\t\t\t\"display_name\": \"Test Support Bot\",\n\t\t\t\"system_prompt\": \"You are a helpful customer support assistant.\",\n\t\t\t\"status\": \"active\",\n\t\t\t\"role_id\": \"member\",\n\t\t\t\"autonomous_mode\": true,\n\t\t\t\"robot_status\": \"idle\",\n\t\t\t\"robot_config\": string(config2JSON),\n\t\t},\n\t})\n\tif err != nil {\n\t\tt.Fatalf(\"Failed to insert robot_test_support_002: %v\", err)\n\t}\n\n\t// Robot 3: Inactive robot\n\terr = qb.Table(tableName).Insert([]map[string]interface{}{\n\t\t{\n\t\t\t\"member_id\": \"robot_test_inactive_003\",\n\t\t\t\"team_id\": \"team_test_cache_001\",\n\t\t\t\"member_type\": \"robot\",\n\t\t\t\"display_name\": \"Test Inactive Bot\",\n\t\t\t\"status\": \"inactive\",\n\t\t\t\"role_id\": \"member\",\n\t\t\t\"autonomous_mode\": true,\n\t\t\t\"robot_status\": \"paused\",\n\t\t},\n\t})\n\tif err != nil {\n\t\tt.Fatalf(\"Failed to insert robot_test_inactive_003: %v\", err)\n\t}\n}\n\n// setupTestRobotsWithClockConfig creates robots with specific clock configurations\nfunc setupTestRobotsWithClockConfig(t *testing.T) {\n\tm := model.Select(\"__yao.member\")\n\ttableName := m.MetaData.Table.Name\n\tqb := capsule.Query()\n\n\t// Robot 1: Times mode (09:00, 14:00 on weekdays)\n\trobotConfigTimes := map[string]interface{}{\n\t\t\"identity\": map[string]interface{}{\n\t\t\t\"role\": \"Times Mode Robot\",\n\t\t},\n\t\t\"quota\": map[string]interface{}{\n\t\t\t\"max\": 2,\n\t\t},\n\t\t\"triggers\": map[string]interface{}{\n\t\t\t\"clock\": map[string]interface{}{\"enabled\": true},\n\t\t},\n\t\t\"clock\": map[string]interface{}{\n\t\t\t\"mode\": \"times\",\n\t\t\t\"times\": []string{\"09:00\", \"14:00\"},\n\t\t\t\"days\": []string{\"Mon\", \"Tue\", \"Wed\", \"Thu\", \"Fri\"},\n\t\t\t\"tz\": \"Asia/Shanghai\",\n\t\t},\n\t}\n\tconfigTimesJSON, _ := json.Marshal(robotConfigTimes)\n\n\terr := qb.Table(tableName).Insert([]map[string]interface{}{\n\t\t{\n\t\t\t\"member_id\": \"robot_test_manager_times\",\n\t\t\t\"team_id\": \"team_test_manager\",\n\t\t\t\"member_type\": \"robot\",\n\t\t\t\"display_name\": \"Test Times Robot\",\n\t\t\t\"status\": \"active\",\n\t\t\t\"role_id\": \"member\",\n\t\t\t\"autonomous_mode\": true,\n\t\t\t\"robot_status\": \"idle\",\n\t\t\t\"robot_config\": string(configTimesJSON),\n\t\t},\n\t})\n\tif err != nil {\n\t\tt.Fatalf(\"Failed to insert robot_test_manager_times: %v\", err)\n\t}\n\n\t// Robot 2: Interval mode (every 30 minutes)\n\trobotConfigInterval := map[string]interface{}{\n\t\t\"identity\": map[string]interface{}{\n\t\t\t\"role\": \"Interval Mode Robot\",\n\t\t},\n\t\t\"quota\": map[string]interface{}{\n\t\t\t\"max\": 2,\n\t\t},\n\t\t\"triggers\": map[string]interface{}{\n\t\t\t\"clock\": map[string]interface{}{\"enabled\": true},\n\t\t},\n\t\t\"clock\": map[string]interface{}{\n\t\t\t\"mode\": \"interval\",\n\t\t\t\"every\": \"30m\",\n\t\t},\n\t}\n\tconfigIntervalJSON, _ := json.Marshal(robotConfigInterval)\n\n\terr = qb.Table(tableName).Insert([]map[string]interface{}{\n\t\t{\n\t\t\t\"member_id\": \"robot_test_manager_interval\",\n\t\t\t\"team_id\": \"team_test_manager\",\n\t\t\t\"member_type\": \"robot\",\n\t\t\t\"display_name\": \"Test Interval Robot\",\n\t\t\t\"status\": \"active\",\n\t\t\t\"role_id\": \"member\",\n\t\t\t\"autonomous_mode\": true,\n\t\t\t\"robot_status\": \"idle\",\n\t\t\t\"robot_config\": string(configIntervalJSON),\n\t\t},\n\t})\n\tif err != nil {\n\t\tt.Fatalf(\"Failed to insert robot_test_manager_interval: %v\", err)\n\t}\n\n\t// Robot 3: Daemon mode\n\trobotConfigDaemon := map[string]interface{}{\n\t\t\"identity\": map[string]interface{}{\n\t\t\t\"role\": \"Daemon Mode Robot\",\n\t\t},\n\t\t\"quota\": map[string]interface{}{\n\t\t\t\"max\": 2,\n\t\t},\n\t\t\"triggers\": map[string]interface{}{\n\t\t\t\"clock\": map[string]interface{}{\"enabled\": true},\n\t\t},\n\t\t\"clock\": map[string]interface{}{\n\t\t\t\"mode\": \"daemon\",\n\t\t\t\"timeout\": \"5m\",\n\t\t},\n\t}\n\tconfigDaemonJSON, _ := json.Marshal(robotConfigDaemon)\n\n\terr = qb.Table(tableName).Insert([]map[string]interface{}{\n\t\t{\n\t\t\t\"member_id\": \"robot_test_manager_daemon\",\n\t\t\t\"team_id\": \"team_test_manager\",\n\t\t\t\"member_type\": \"robot\",\n\t\t\t\"display_name\": \"Test Daemon Robot\",\n\t\t\t\"status\": \"active\",\n\t\t\t\"role_id\": \"member\",\n\t\t\t\"autonomous_mode\": true,\n\t\t\t\"robot_status\": \"idle\",\n\t\t\t\"robot_config\": string(configDaemonJSON),\n\t\t},\n\t})\n\tif err != nil {\n\t\tt.Fatalf(\"Failed to insert robot_test_manager_daemon: %v\", err)\n\t}\n\n\t// Robot 4: Paused robot (should be skipped)\n\trobotConfigPaused := map[string]interface{}{\n\t\t\"identity\": map[string]interface{}{\n\t\t\t\"role\": \"Paused Robot\",\n\t\t},\n\t\t\"triggers\": map[string]interface{}{\n\t\t\t\"clock\": map[string]interface{}{\"enabled\": true},\n\t\t},\n\t\t\"clock\": map[string]interface{}{\n\t\t\t\"mode\": \"times\",\n\t\t\t\"times\": []string{\"09:00\"},\n\t\t\t\"tz\": \"Asia/Shanghai\",\n\t\t},\n\t}\n\tconfigPausedJSON, _ := json.Marshal(robotConfigPaused)\n\n\terr = qb.Table(tableName).Insert([]map[string]interface{}{\n\t\t{\n\t\t\t\"member_id\": \"robot_test_manager_paused\",\n\t\t\t\"team_id\": \"team_test_manager\",\n\t\t\t\"member_type\": \"robot\",\n\t\t\t\"display_name\": \"Test Paused Robot\",\n\t\t\t\"status\": \"active\",\n\t\t\t\"role_id\": \"member\",\n\t\t\t\"autonomous_mode\": true,\n\t\t\t\"robot_status\": \"paused\",\n\t\t\t\"robot_config\": string(configPausedJSON),\n\t\t},\n\t})\n\tif err != nil {\n\t\tt.Fatalf(\"Failed to insert robot_test_manager_paused: %v\", err)\n\t}\n\n\t// Robot 5: Clock disabled robot\n\trobotConfigDisabled := map[string]interface{}{\n\t\t\"identity\": map[string]interface{}{\n\t\t\t\"role\": \"Clock Disabled Robot\",\n\t\t},\n\t\t\"triggers\": map[string]interface{}{\n\t\t\t\"clock\": map[string]interface{}{\"enabled\": false},\n\t\t},\n\t\t\"clock\": map[string]interface{}{\n\t\t\t\"mode\": \"times\",\n\t\t\t\"times\": []string{\"09:00\"},\n\t\t},\n\t}\n\tconfigDisabledJSON, _ := json.Marshal(robotConfigDisabled)\n\n\terr = qb.Table(tableName).Insert([]map[string]interface{}{\n\t\t{\n\t\t\t\"member_id\": \"robot_test_manager_disabled\",\n\t\t\t\"team_id\": \"team_test_manager\",\n\t\t\t\"member_type\": \"robot\",\n\t\t\t\"display_name\": \"Test Clock Disabled Robot\",\n\t\t\t\"status\": \"active\",\n\t\t\t\"role_id\": \"member\",\n\t\t\t\"autonomous_mode\": true,\n\t\t\t\"robot_status\": \"idle\",\n\t\t\t\"robot_config\": string(configDisabledJSON),\n\t\t},\n\t})\n\tif err != nil {\n\t\tt.Fatalf(\"Failed to insert robot_test_manager_disabled: %v\", err)\n\t}\n}\n\n// ==================== Intervene Tests ====================\n\nfunc TestManagerIntervene(t *testing.T) {\n\tif testing.Short() {\n\t\tt.Skip(\"Skipping integration test\")\n\t}\n\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\tcleanupTestRobots(t)\n\tsetupTestRobotsWithInterveneConfig(t)\n\tdefer cleanupTestRobots(t)\n\n\tt.Run(\"intervene success\", func(t *testing.T) {\n\t\tm := manager.New()\n\t\terr := m.Start()\n\t\tassert.NoError(t, err)\n\t\tdefer m.Stop()\n\n\t\tctx := types.NewContext(context.Background(), nil)\n\t\treq := &types.InterveneRequest{\n\t\t\tTeamID: \"team_test_manager\",\n\t\t\tMemberID: \"robot_test_manager_intervene\",\n\t\t\tAction: types.ActionTaskAdd,\n\t\t\tMessages: []agentcontext.Message{\n\t\t\t\t{Role: agentcontext.RoleUser, Content: \"Add a new task\"},\n\t\t\t},\n\t\t}\n\n\t\tresult, err := m.Intervene(ctx, req)\n\t\tassert.NoError(t, err)\n\t\tassert.NotNil(t, result)\n\t\tassert.NotEmpty(t, result.ExecutionID)\n\t\tassert.Equal(t, types.ExecPending, result.Status)\n\t})\n\n\tt.Run(\"intervene - manager not started\", func(t *testing.T) {\n\t\tm := manager.New()\n\t\t// Don't start\n\n\t\tctx := types.NewContext(context.Background(), nil)\n\t\treq := &types.InterveneRequest{\n\t\t\tMemberID: \"robot_test_manager_intervene\",\n\t\t\tAction: types.ActionTaskAdd,\n\t\t\tMessages: []agentcontext.Message{\n\t\t\t\t{Role: agentcontext.RoleUser, Content: \"Add a new task\"},\n\t\t\t},\n\t\t}\n\n\t\t_, err := m.Intervene(ctx, req)\n\t\tassert.Error(t, err)\n\t\tassert.Contains(t, err.Error(), \"not started\")\n\t})\n\n\tt.Run(\"intervene - robot not found\", func(t *testing.T) {\n\t\tm := manager.New()\n\t\terr := m.Start()\n\t\tassert.NoError(t, err)\n\t\tdefer m.Stop()\n\n\t\tctx := types.NewContext(context.Background(), nil)\n\t\treq := &types.InterveneRequest{\n\t\t\tMemberID: \"non_existent_robot\",\n\t\t\tAction: types.ActionTaskAdd,\n\t\t\tMessages: []agentcontext.Message{\n\t\t\t\t{Role: agentcontext.RoleUser, Content: \"Add a new task\"},\n\t\t\t},\n\t\t}\n\n\t\t_, err = m.Intervene(ctx, req)\n\t\tassert.Error(t, err)\n\t\tassert.Equal(t, types.ErrRobotNotFound, err)\n\t})\n\n\tt.Run(\"intervene - robot paused\", func(t *testing.T) {\n\t\tm := manager.New()\n\t\terr := m.Start()\n\t\tassert.NoError(t, err)\n\t\tdefer m.Stop()\n\n\t\tctx := types.NewContext(context.Background(), nil)\n\t\treq := &types.InterveneRequest{\n\t\t\tMemberID: \"robot_test_manager_paused\",\n\t\t\tAction: types.ActionTaskAdd,\n\t\t\tMessages: []agentcontext.Message{\n\t\t\t\t{Role: agentcontext.RoleUser, Content: \"Add a new task\"},\n\t\t\t},\n\t\t}\n\n\t\t_, err = m.Intervene(ctx, req)\n\t\tassert.Error(t, err)\n\t\tassert.Equal(t, types.ErrRobotPaused, err)\n\t})\n\n\tt.Run(\"intervene - invalid request\", func(t *testing.T) {\n\t\tm := manager.New()\n\t\terr := m.Start()\n\t\tassert.NoError(t, err)\n\t\tdefer m.Stop()\n\n\t\tctx := types.NewContext(context.Background(), nil)\n\t\treq := &types.InterveneRequest{\n\t\t\tMemberID: \"\", // Invalid: empty member_id\n\t\t\tAction: types.ActionTaskAdd,\n\t\t}\n\n\t\t_, err = m.Intervene(ctx, req)\n\t\tassert.Error(t, err)\n\t\tassert.Contains(t, err.Error(), \"member_id\")\n\t})\n\n\tt.Run(\"intervene - trigger disabled\", func(t *testing.T) {\n\t\tm := manager.New()\n\t\terr := m.Start()\n\t\tassert.NoError(t, err)\n\t\tdefer m.Stop()\n\n\t\tctx := types.NewContext(context.Background(), nil)\n\t\treq := &types.InterveneRequest{\n\t\t\tMemberID: \"robot_test_manager_intervene_disabled\",\n\t\t\tAction: types.ActionTaskAdd,\n\t\t\tMessages: []agentcontext.Message{\n\t\t\t\t{Role: agentcontext.RoleUser, Content: \"Add a new task\"},\n\t\t\t},\n\t\t}\n\n\t\t_, err = m.Intervene(ctx, req)\n\t\tassert.Error(t, err)\n\t\tassert.Equal(t, types.ErrTriggerDisabled, err)\n\t})\n}\n\n// ==================== HandleEvent Tests ====================\n\nfunc TestManagerHandleEvent(t *testing.T) {\n\tif testing.Short() {\n\t\tt.Skip(\"Skipping integration test\")\n\t}\n\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\tcleanupTestRobots(t)\n\tsetupTestRobotsWithEventConfig(t)\n\tdefer cleanupTestRobots(t)\n\n\tt.Run(\"handle event success\", func(t *testing.T) {\n\t\tm := manager.New()\n\t\terr := m.Start()\n\t\tassert.NoError(t, err)\n\t\tdefer m.Stop()\n\n\t\tctx := types.NewContext(context.Background(), nil)\n\t\treq := &types.EventRequest{\n\t\t\tMemberID: \"robot_test_manager_event\",\n\t\t\tSource: \"webhook\",\n\t\t\tEventType: \"lead.created\",\n\t\t\tData: map[string]interface{}{\"name\": \"John\", \"email\": \"john@example.com\"},\n\t\t}\n\n\t\tresult, err := m.HandleEvent(ctx, req)\n\t\tassert.NoError(t, err)\n\t\tassert.NotNil(t, result)\n\t\tassert.NotEmpty(t, result.ExecutionID)\n\t\tassert.Equal(t, types.ExecPending, result.Status)\n\t})\n\n\tt.Run(\"handle event - manager not started\", func(t *testing.T) {\n\t\tm := manager.New()\n\t\t// Don't start\n\n\t\tctx := types.NewContext(context.Background(), nil)\n\t\treq := &types.EventRequest{\n\t\t\tMemberID: \"robot_test_manager_event\",\n\t\t\tSource: \"webhook\",\n\t\t\tEventType: \"lead.created\",\n\t\t}\n\n\t\t_, err := m.HandleEvent(ctx, req)\n\t\tassert.Error(t, err)\n\t\tassert.Contains(t, err.Error(), \"not started\")\n\t})\n\n\tt.Run(\"handle event - robot not found\", func(t *testing.T) {\n\t\tm := manager.New()\n\t\terr := m.Start()\n\t\tassert.NoError(t, err)\n\t\tdefer m.Stop()\n\n\t\tctx := types.NewContext(context.Background(), nil)\n\t\treq := &types.EventRequest{\n\t\t\tMemberID: \"non_existent_robot\",\n\t\t\tSource: \"webhook\",\n\t\t\tEventType: \"lead.created\",\n\t\t}\n\n\t\t_, err = m.HandleEvent(ctx, req)\n\t\tassert.Error(t, err)\n\t\tassert.Equal(t, types.ErrRobotNotFound, err)\n\t})\n\n\tt.Run(\"handle event - invalid request\", func(t *testing.T) {\n\t\tm := manager.New()\n\t\terr := m.Start()\n\t\tassert.NoError(t, err)\n\t\tdefer m.Stop()\n\n\t\tctx := types.NewContext(context.Background(), nil)\n\t\treq := &types.EventRequest{\n\t\t\tMemberID: \"robot_test_manager_event\",\n\t\t\tSource: \"\", // Invalid: empty source\n\t\t\tEventType: \"lead.created\",\n\t\t}\n\n\t\t_, err = m.HandleEvent(ctx, req)\n\t\tassert.Error(t, err)\n\t\tassert.Contains(t, err.Error(), \"source\")\n\t})\n\n\tt.Run(\"handle event - trigger disabled\", func(t *testing.T) {\n\t\tm := manager.New()\n\t\terr := m.Start()\n\t\tassert.NoError(t, err)\n\t\tdefer m.Stop()\n\n\t\tctx := types.NewContext(context.Background(), nil)\n\t\treq := &types.EventRequest{\n\t\t\tMemberID: \"robot_test_manager_event_disabled\",\n\t\t\tSource: \"webhook\",\n\t\t\tEventType: \"lead.created\",\n\t\t}\n\n\t\t_, err = m.HandleEvent(ctx, req)\n\t\tassert.Error(t, err)\n\t\tassert.Equal(t, types.ErrTriggerDisabled, err)\n\t})\n}\n\n// ==================== Execution Control Tests ====================\n\nfunc TestManagerExecutionControl(t *testing.T) {\n\tif testing.Short() {\n\t\tt.Skip(\"Skipping integration test\")\n\t}\n\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\tcleanupTestRobots(t)\n\tsetupTestRobotsWithInterveneConfig(t)\n\tdefer cleanupTestRobots(t)\n\n\tt.Run(\"pause and resume execution\", func(t *testing.T) {\n\t\tm := manager.New()\n\t\terr := m.Start()\n\t\tassert.NoError(t, err)\n\t\tdefer m.Stop()\n\n\t\t// Trigger an execution\n\t\tctx := types.NewContext(context.Background(), nil)\n\t\treq := &types.InterveneRequest{\n\t\t\tMemberID: \"robot_test_manager_intervene\",\n\t\t\tAction: types.ActionTaskAdd,\n\t\t\tMessages: []agentcontext.Message{\n\t\t\t\t{Role: agentcontext.RoleUser, Content: \"Test task\"},\n\t\t\t},\n\t\t}\n\n\t\tresult, err := m.Intervene(ctx, req)\n\t\tassert.NoError(t, err)\n\t\texecID := result.ExecutionID\n\n\t\t// Wait a bit for execution to be tracked\n\t\ttime.Sleep(50 * time.Millisecond)\n\n\t\t// Pause\n\t\terr = m.PauseExecution(ctx, execID)\n\t\tassert.NoError(t, err)\n\n\t\t// Get status - should be paused\n\t\tstatus, err := m.GetExecutionStatus(execID)\n\t\tassert.NoError(t, err)\n\t\tassert.True(t, status.IsPaused())\n\n\t\t// Resume\n\t\terr = m.ResumeExecution(ctx, execID)\n\t\tassert.NoError(t, err)\n\n\t\t// Get status - should not be paused\n\t\tstatus, err = m.GetExecutionStatus(execID)\n\t\tassert.NoError(t, err)\n\t\tassert.False(t, status.IsPaused())\n\t})\n\n\tt.Run(\"stop execution\", func(t *testing.T) {\n\t\tm := manager.New()\n\t\terr := m.Start()\n\t\tassert.NoError(t, err)\n\t\tdefer m.Stop()\n\n\t\t// Trigger an execution\n\t\tctx := types.NewContext(context.Background(), nil)\n\t\treq := &types.InterveneRequest{\n\t\t\tMemberID: \"robot_test_manager_intervene\",\n\t\t\tAction: types.ActionTaskAdd,\n\t\t\tMessages: []agentcontext.Message{\n\t\t\t\t{Role: agentcontext.RoleUser, Content: \"Test task\"},\n\t\t\t},\n\t\t}\n\n\t\tresult, err := m.Intervene(ctx, req)\n\t\tassert.NoError(t, err)\n\t\texecID := result.ExecutionID\n\n\t\t// Wait a bit for execution to be tracked\n\t\ttime.Sleep(50 * time.Millisecond)\n\n\t\t// Stop\n\t\terr = m.StopExecution(ctx, execID)\n\t\tassert.NoError(t, err)\n\n\t\t// Get status - should not be found (removed after stop)\n\t\t_, err = m.GetExecutionStatus(execID)\n\t\tassert.Error(t, err)\n\t\tassert.Contains(t, err.Error(), \"not found\")\n\t})\n\n\tt.Run(\"list executions\", func(t *testing.T) {\n\t\tm := manager.New()\n\t\terr := m.Start()\n\t\tassert.NoError(t, err)\n\t\tdefer m.Stop()\n\n\t\tctx := types.NewContext(context.Background(), nil)\n\n\t\t// Track execution IDs\n\t\tvar execIDs []string\n\n\t\t// Trigger multiple executions\n\t\tfor i := 0; i < 3; i++ {\n\t\t\treq := &types.InterveneRequest{\n\t\t\t\tMemberID: \"robot_test_manager_intervene\",\n\t\t\t\tAction: types.ActionTaskAdd,\n\t\t\t\tMessages: []agentcontext.Message{\n\t\t\t\t\t{Role: agentcontext.RoleUser, Content: \"Test task\"},\n\t\t\t\t},\n\t\t\t}\n\t\t\tresult, err := m.Intervene(ctx, req)\n\t\t\tassert.NoError(t, err)\n\t\t\texecIDs = append(execIDs, result.ExecutionID)\n\t\t}\n\n\t\t// Verify each execution was tracked (even if briefly)\n\t\t// Note: executions complete quickly with stub executor, so they may be removed\n\t\t// We just verify that we got valid execution IDs\n\t\tassert.Len(t, execIDs, 3)\n\t\tfor _, id := range execIDs {\n\t\t\tassert.NotEmpty(t, id)\n\t\t}\n\t})\n}\n\n// setupTestRobotsWithInterveneConfig creates test robots with intervene trigger enabled\nfunc setupTestRobotsWithInterveneConfig(t *testing.T) {\n\t// First setup the basic robots\n\tsetupTestRobotsWithClockConfig(t)\n\n\t// Add robots for intervene tests\n\tqb := capsule.Query()\n\tm := model.Select(\"__yao.member\")\n\ttableName := m.MetaData.Table.Name\n\n\t// Robot with intervene enabled\n\trobotConfigIntervene := map[string]interface{}{\n\t\t\"identity\": map[string]interface{}{\n\t\t\t\"role\": \"Intervene Test Robot\",\n\t\t},\n\t\t\"triggers\": map[string]interface{}{\n\t\t\t\"clock\": map[string]interface{}{\"enabled\": false},\n\t\t\t\"intervene\": map[string]interface{}{\"enabled\": true},\n\t\t},\n\t\t\"quota\": map[string]interface{}{\n\t\t\t\"max\": 5,\n\t\t\t\"queue\": 10,\n\t\t},\n\t}\n\tconfigInterveneJSON, _ := json.Marshal(robotConfigIntervene)\n\n\terr := qb.Table(tableName).Insert([]map[string]interface{}{\n\t\t{\n\t\t\t\"member_id\": \"robot_test_manager_intervene\",\n\t\t\t\"team_id\": \"team_test_manager\",\n\t\t\t\"member_type\": \"robot\",\n\t\t\t\"display_name\": \"Test Intervene Robot\",\n\t\t\t\"status\": \"active\",\n\t\t\t\"role_id\": \"member\",\n\t\t\t\"autonomous_mode\": true,\n\t\t\t\"robot_status\": \"idle\",\n\t\t\t\"robot_config\": string(configInterveneJSON),\n\t\t},\n\t})\n\tif err != nil {\n\t\tt.Fatalf(\"Failed to insert robot_test_manager_intervene: %v\", err)\n\t}\n\n\t// Robot with intervene disabled\n\trobotConfigInterveneDisabled := map[string]interface{}{\n\t\t\"identity\": map[string]interface{}{\n\t\t\t\"role\": \"Intervene Disabled Robot\",\n\t\t},\n\t\t\"triggers\": map[string]interface{}{\n\t\t\t\"clock\": map[string]interface{}{\"enabled\": false},\n\t\t\t\"intervene\": map[string]interface{}{\"enabled\": false},\n\t\t},\n\t}\n\tconfigInterveneDisabledJSON, _ := json.Marshal(robotConfigInterveneDisabled)\n\n\terr = qb.Table(tableName).Insert([]map[string]interface{}{\n\t\t{\n\t\t\t\"member_id\": \"robot_test_manager_intervene_disabled\",\n\t\t\t\"team_id\": \"team_test_manager\",\n\t\t\t\"member_type\": \"robot\",\n\t\t\t\"display_name\": \"Test Intervene Disabled Robot\",\n\t\t\t\"status\": \"active\",\n\t\t\t\"role_id\": \"member\",\n\t\t\t\"autonomous_mode\": true,\n\t\t\t\"robot_status\": \"idle\",\n\t\t\t\"robot_config\": string(configInterveneDisabledJSON),\n\t\t},\n\t})\n\tif err != nil {\n\t\tt.Fatalf(\"Failed to insert robot_test_manager_intervene_disabled: %v\", err)\n\t}\n}\n\n// setupTestRobotsWithEventConfig creates test robots with event trigger enabled\nfunc setupTestRobotsWithEventConfig(t *testing.T) {\n\t// First setup the basic robots\n\tsetupTestRobotsWithClockConfig(t)\n\n\t// Add robots for event tests\n\tqb := capsule.Query()\n\tm := model.Select(\"__yao.member\")\n\ttableName := m.MetaData.Table.Name\n\n\t// Robot with event enabled\n\trobotConfigEvent := map[string]interface{}{\n\t\t\"identity\": map[string]interface{}{\n\t\t\t\"role\": \"Event Test Robot\",\n\t\t},\n\t\t\"triggers\": map[string]interface{}{\n\t\t\t\"clock\": map[string]interface{}{\"enabled\": false},\n\t\t\t\"event\": map[string]interface{}{\"enabled\": true},\n\t\t},\n\t\t\"quota\": map[string]interface{}{\n\t\t\t\"max\": 5,\n\t\t\t\"queue\": 10,\n\t\t},\n\t}\n\tconfigEventJSON, _ := json.Marshal(robotConfigEvent)\n\n\terr := qb.Table(tableName).Insert([]map[string]interface{}{\n\t\t{\n\t\t\t\"member_id\": \"robot_test_manager_event\",\n\t\t\t\"team_id\": \"team_test_manager\",\n\t\t\t\"member_type\": \"robot\",\n\t\t\t\"display_name\": \"Test Event Robot\",\n\t\t\t\"status\": \"active\",\n\t\t\t\"role_id\": \"member\",\n\t\t\t\"autonomous_mode\": true,\n\t\t\t\"robot_status\": \"idle\",\n\t\t\t\"robot_config\": string(configEventJSON),\n\t\t},\n\t})\n\tif err != nil {\n\t\tt.Fatalf(\"Failed to insert robot_test_manager_event: %v\", err)\n\t}\n\n\t// Robot with event disabled\n\trobotConfigEventDisabled := map[string]interface{}{\n\t\t\"identity\": map[string]interface{}{\n\t\t\t\"role\": \"Event Disabled Robot\",\n\t\t},\n\t\t\"triggers\": map[string]interface{}{\n\t\t\t\"clock\": map[string]interface{}{\"enabled\": false},\n\t\t\t\"event\": map[string]interface{}{\"enabled\": false},\n\t\t},\n\t}\n\tconfigEventDisabledJSON, _ := json.Marshal(robotConfigEventDisabled)\n\n\terr = qb.Table(tableName).Insert([]map[string]interface{}{\n\t\t{\n\t\t\t\"member_id\": \"robot_test_manager_event_disabled\",\n\t\t\t\"team_id\": \"team_test_manager\",\n\t\t\t\"member_type\": \"robot\",\n\t\t\t\"display_name\": \"Test Event Disabled Robot\",\n\t\t\t\"status\": \"active\",\n\t\t\t\"role_id\": \"member\",\n\t\t\t\"autonomous_mode\": true,\n\t\t\t\"robot_status\": \"idle\",\n\t\t\t\"robot_config\": string(configEventDisabledJSON),\n\t\t},\n\t})\n\tif err != nil {\n\t\tt.Fatalf(\"Failed to insert robot_test_manager_event_disabled: %v\", err)\n\t}\n}\n\n// ==================== Lazy Load Tests for Non-Autonomous Robots ====================\n\n// TestManagerLazyLoadNonAutonomous tests that non-autonomous robots are lazy-loaded on demand\n// and automatically cleaned up after execution completes\nfunc TestManagerLazyLoadNonAutonomous(t *testing.T) {\n\tif testing.Short() {\n\t\tt.Skip(\"Skipping integration test\")\n\t}\n\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\tcleanupTestRobots(t)\n\tsetupTestRobotsWithNonAutonomous(t)\n\tdefer cleanupTestRobots(t)\n\n\tt.Run(\"non-autonomous robot not in cache on startup\", func(t *testing.T) {\n\t\tm := manager.New()\n\t\terr := m.Start()\n\t\tassert.NoError(t, err)\n\t\tdefer m.Stop()\n\n\t\t// Non-autonomous robot should NOT be in cache\n\t\trobot := m.Cache().Get(\"robot_test_manager_on_demand\")\n\t\tassert.Nil(t, robot, \"Non-autonomous robot should not be pre-loaded into cache\")\n\n\t\t// Autonomous robot SHOULD be in cache\n\t\tautoRobot := m.Cache().Get(\"robot_test_manager_times\")\n\t\tassert.NotNil(t, autoRobot, \"Autonomous robot should be in cache\")\n\t})\n\n\tt.Run(\"TriggerManual lazy-loads non-autonomous robot\", func(t *testing.T) {\n\t\tm := manager.New()\n\t\terr := m.Start()\n\t\tassert.NoError(t, err)\n\t\tdefer m.Stop()\n\n\t\tctx := types.NewContext(context.Background(), nil)\n\n\t\t// Verify robot is NOT in cache before trigger\n\t\tassert.Nil(t, m.Cache().Get(\"robot_test_manager_on_demand\"))\n\n\t\t// Trigger the non-autonomous robot manually\n\t\texecID, err := m.TriggerManual(ctx, \"robot_test_manager_on_demand\", types.TriggerHuman, nil)\n\t\tassert.NoError(t, err)\n\t\tassert.NotEmpty(t, execID)\n\n\t\t// Robot should now be in cache (lazy-loaded)\n\t\trobot := m.Cache().Get(\"robot_test_manager_on_demand\")\n\t\tassert.NotNil(t, robot, \"Robot should be lazy-loaded into cache\")\n\t\tassert.Equal(t, \"robot_test_manager_on_demand\", robot.MemberID)\n\t\tassert.False(t, robot.AutonomousMode)\n\t})\n\n\tt.Run(\"Intervene lazy-loads non-autonomous robot\", func(t *testing.T) {\n\t\tm := manager.New()\n\t\terr := m.Start()\n\t\tassert.NoError(t, err)\n\t\tdefer m.Stop()\n\n\t\tctx := types.NewContext(context.Background(), nil)\n\n\t\t// Verify robot is NOT in cache before trigger\n\t\tassert.Nil(t, m.Cache().Get(\"robot_test_manager_on_demand_intervene\"))\n\n\t\t// Intervene on the non-autonomous robot\n\t\treq := &types.InterveneRequest{\n\t\t\tTeamID: \"team_test_manager\",\n\t\t\tMemberID: \"robot_test_manager_on_demand_intervene\",\n\t\t\tAction: types.ActionTaskAdd,\n\t\t\tMessages: []agentcontext.Message{\n\t\t\t\t{Role: agentcontext.RoleUser, Content: \"Test lazy load via intervene\"},\n\t\t\t},\n\t\t}\n\n\t\tresult, err := m.Intervene(ctx, req)\n\t\tassert.NoError(t, err)\n\t\tassert.NotEmpty(t, result.ExecutionID)\n\n\t\t// Robot should now be in cache (lazy-loaded)\n\t\trobot := m.Cache().Get(\"robot_test_manager_on_demand_intervene\")\n\t\tassert.NotNil(t, robot, \"Robot should be lazy-loaded into cache via Intervene\")\n\t})\n\n\tt.Run(\"HandleEvent lazy-loads non-autonomous robot\", func(t *testing.T) {\n\t\tm := manager.New()\n\t\terr := m.Start()\n\t\tassert.NoError(t, err)\n\t\tdefer m.Stop()\n\n\t\tctx := types.NewContext(context.Background(), nil)\n\n\t\t// Verify robot is NOT in cache before trigger\n\t\tassert.Nil(t, m.Cache().Get(\"robot_test_manager_on_demand_event\"))\n\n\t\t// Send event to the non-autonomous robot\n\t\treq := &types.EventRequest{\n\t\t\tMemberID: \"robot_test_manager_on_demand_event\",\n\t\t\tSource: \"webhook\",\n\t\t\tEventType: \"data.updated\",\n\t\t\tData: map[string]interface{}{\"test\": true},\n\t\t}\n\n\t\tresult, err := m.HandleEvent(ctx, req)\n\t\tassert.NoError(t, err)\n\t\tassert.NotEmpty(t, result.ExecutionID)\n\n\t\t// Robot should now be in cache (lazy-loaded)\n\t\trobot := m.Cache().Get(\"robot_test_manager_on_demand_event\")\n\t\tassert.NotNil(t, robot, \"Robot should be lazy-loaded into cache via HandleEvent\")\n\t})\n\n\tt.Run(\"lazy-loaded robot is cleaned up after execution completes\", func(t *testing.T) {\n\t\tm := manager.New()\n\t\terr := m.Start()\n\t\tassert.NoError(t, err)\n\t\tdefer m.Stop()\n\n\t\tctx := types.NewContext(context.Background(), nil)\n\n\t\t// Trigger the non-autonomous robot\n\t\t_, err = m.TriggerManual(ctx, \"robot_test_manager_on_demand\", types.TriggerHuman, nil)\n\t\tassert.NoError(t, err)\n\n\t\t// Robot should be in cache immediately after trigger\n\t\trobot := m.Cache().Get(\"robot_test_manager_on_demand\")\n\t\tassert.NotNil(t, robot, \"Robot should be in cache after trigger\")\n\n\t\t// Wait for execution to complete and cleanup to happen\n\t\t// The stub executor completes quickly, and cleanup runs every 5 seconds\n\t\t// We wait up to 10 seconds for the cleanup goroutine to remove the robot\n\t\tvar removed bool\n\t\tfor i := 0; i < 20; i++ {\n\t\t\ttime.Sleep(500 * time.Millisecond)\n\t\t\tif m.Cache().Get(\"robot_test_manager_on_demand\") == nil {\n\t\t\t\tremoved = true\n\t\t\t\tbreak\n\t\t\t}\n\t\t}\n\n\t\tassert.True(t, removed, \"Non-autonomous robot should be removed from cache after execution completes\")\n\t})\n\n\tt.Run(\"trigger non-existent robot returns error\", func(t *testing.T) {\n\t\tm := manager.New()\n\t\terr := m.Start()\n\t\tassert.NoError(t, err)\n\t\tdefer m.Stop()\n\n\t\tctx := types.NewContext(context.Background(), nil)\n\n\t\t// Try to trigger a robot that doesn't exist in DB\n\t\t_, err = m.TriggerManual(ctx, \"robot_nonexistent_xyz\", types.TriggerHuman, nil)\n\t\tassert.Error(t, err)\n\t\tassert.Equal(t, types.ErrRobotNotFound, err)\n\t})\n}\n\n// setupTestRobotsWithNonAutonomous creates test robots including non-autonomous ones\nfunc setupTestRobotsWithNonAutonomous(t *testing.T) {\n\t// First setup the autonomous robots\n\tsetupTestRobotsWithClockConfig(t)\n\n\t// Add non-autonomous robots\n\tqb := capsule.Query()\n\tm := model.Select(\"__yao.member\")\n\ttableName := m.MetaData.Table.Name\n\n\t// Non-autonomous robot 1: for TriggerManual test\n\trobotConfigOnDemand := map[string]interface{}{\n\t\t\"identity\": map[string]interface{}{\n\t\t\t\"role\": \"On-Demand Robot\",\n\t\t},\n\t\t\"triggers\": map[string]interface{}{\n\t\t\t\"clock\": map[string]interface{}{\"enabled\": false},\n\t\t\t\"intervene\": map[string]interface{}{\"enabled\": true},\n\t\t},\n\t\t\"quota\": map[string]interface{}{\n\t\t\t\"max\": 2,\n\t\t\t\"queue\": 5,\n\t\t},\n\t}\n\tconfigOnDemandJSON, _ := json.Marshal(robotConfigOnDemand)\n\n\terr := qb.Table(tableName).Insert([]map[string]interface{}{\n\t\t{\n\t\t\t\"member_id\": \"robot_test_manager_on_demand\",\n\t\t\t\"team_id\": \"team_test_manager\",\n\t\t\t\"member_type\": \"robot\",\n\t\t\t\"display_name\": \"Test On-Demand Robot\",\n\t\t\t\"status\": \"active\",\n\t\t\t\"role_id\": \"member\",\n\t\t\t\"autonomous_mode\": false, // Non-autonomous!\n\t\t\t\"robot_status\": \"idle\",\n\t\t\t\"robot_config\": string(configOnDemandJSON),\n\t\t},\n\t})\n\tif err != nil {\n\t\tt.Fatalf(\"Failed to insert robot_test_manager_on_demand: %v\", err)\n\t}\n\n\t// Non-autonomous robot 2: for Intervene test\n\trobotConfigOnDemandIntervene := map[string]interface{}{\n\t\t\"identity\": map[string]interface{}{\n\t\t\t\"role\": \"On-Demand Intervene Robot\",\n\t\t},\n\t\t\"triggers\": map[string]interface{}{\n\t\t\t\"clock\": map[string]interface{}{\"enabled\": false},\n\t\t\t\"intervene\": map[string]interface{}{\"enabled\": true},\n\t\t},\n\t\t\"quota\": map[string]interface{}{\n\t\t\t\"max\": 2,\n\t\t},\n\t}\n\tconfigOnDemandInterveneJSON, _ := json.Marshal(robotConfigOnDemandIntervene)\n\n\terr = qb.Table(tableName).Insert([]map[string]interface{}{\n\t\t{\n\t\t\t\"member_id\": \"robot_test_manager_on_demand_intervene\",\n\t\t\t\"team_id\": \"team_test_manager\",\n\t\t\t\"member_type\": \"robot\",\n\t\t\t\"display_name\": \"Test On-Demand Intervene Robot\",\n\t\t\t\"status\": \"active\",\n\t\t\t\"role_id\": \"member\",\n\t\t\t\"autonomous_mode\": false, // Non-autonomous!\n\t\t\t\"robot_status\": \"idle\",\n\t\t\t\"robot_config\": string(configOnDemandInterveneJSON),\n\t\t},\n\t})\n\tif err != nil {\n\t\tt.Fatalf(\"Failed to insert robot_test_manager_on_demand_intervene: %v\", err)\n\t}\n\n\t// Non-autonomous robot 3: for HandleEvent test\n\trobotConfigOnDemandEvent := map[string]interface{}{\n\t\t\"identity\": map[string]interface{}{\n\t\t\t\"role\": \"On-Demand Event Robot\",\n\t\t},\n\t\t\"triggers\": map[string]interface{}{\n\t\t\t\"clock\": map[string]interface{}{\"enabled\": false},\n\t\t\t\"event\": map[string]interface{}{\"enabled\": true},\n\t\t},\n\t\t\"quota\": map[string]interface{}{\n\t\t\t\"max\": 2,\n\t\t},\n\t}\n\tconfigOnDemandEventJSON, _ := json.Marshal(robotConfigOnDemandEvent)\n\n\terr = qb.Table(tableName).Insert([]map[string]interface{}{\n\t\t{\n\t\t\t\"member_id\": \"robot_test_manager_on_demand_event\",\n\t\t\t\"team_id\": \"team_test_manager\",\n\t\t\t\"member_type\": \"robot\",\n\t\t\t\"display_name\": \"Test On-Demand Event Robot\",\n\t\t\t\"status\": \"active\",\n\t\t\t\"role_id\": \"member\",\n\t\t\t\"autonomous_mode\": false, // Non-autonomous!\n\t\t\t\"robot_status\": \"idle\",\n\t\t\t\"robot_config\": string(configOnDemandEventJSON),\n\t\t},\n\t})\n\tif err != nil {\n\t\tt.Fatalf(\"Failed to insert robot_test_manager_on_demand_event: %v\", err)\n\t}\n}\n\n// cleanupTestRobots removes all test robot records\nfunc cleanupTestRobots(t *testing.T) {\n\tqb := capsule.Query()\n\tm := model.Select(\"__yao.member\")\n\ttableName := m.MetaData.Table.Name\n\n\t// List of test robot IDs to clean up\n\ttestRobotIDs := []string{\n\t\t\"robot_test_sales_001\",\n\t\t\"robot_test_support_002\",\n\t\t\"robot_test_inactive_003\",\n\t\t\"robot_test_manager_times\",\n\t\t\"robot_test_manager_interval\",\n\t\t\"robot_test_manager_daemon\",\n\t\t\"robot_test_manager_paused\",\n\t\t\"robot_test_manager_disabled\",\n\t\t\"robot_test_manager_intervene\",\n\t\t\"robot_test_manager_intervene_disabled\",\n\t\t\"robot_test_manager_event\",\n\t\t\"robot_test_manager_event_disabled\",\n\t\t// Non-autonomous robots\n\t\t\"robot_test_manager_on_demand\",\n\t\t\"robot_test_manager_on_demand_intervene\",\n\t\t\"robot_test_manager_on_demand_event\",\n\t}\n\n\tfor _, id := range testRobotIDs {\n\t\t// Soft delete\n\t\tm.DeleteWhere(model.QueryParam{\n\t\t\tWheres: []model.QueryWhere{\n\t\t\t\t{Column: \"member_id\", Value: id},\n\t\t\t},\n\t\t})\n\t\t// Hard delete\n\t\tqb.Table(tableName).Where(\"member_id\", id).Delete()\n\t}\n}\n" }, { "path": "agent/robot/plan/plan.go", "content": "package plan\n\nimport (\n\t\"time\"\n\n\t\"github.com/yaoapp/yao/agent/robot/types\"\n)\n\n// Plan manages planned tasks/goals for later execution\n// This is a stub implementation for Phase 2\ntype Plan struct{}\n\n// New creates a new plan instance\nfunc New() *Plan {\n\treturn &Plan{}\n}\n\n// Add adds a task or goal to plan queue\n// Stub: returns nil (will be implemented in Phase 11)\nfunc (p *Plan) Add(ctx *types.Context, memberID string, item interface{}, executeAt time.Time) error {\n\treturn nil\n}\n\n// Remove removes an item from plan queue\n// Stub: returns nil (will be implemented in Phase 11)\nfunc (p *Plan) Remove(ctx *types.Context, memberID string, itemID string) error {\n\treturn nil\n}\n\n// List lists all planned items for a robot\n// Stub: returns empty slice (will be implemented in Phase 11)\nfunc (p *Plan) List(ctx *types.Context, memberID string) ([]interface{}, error) {\n\treturn []interface{}{}, nil\n}\n\n// GetDue returns items that are due for execution\n// Stub: returns empty slice (will be implemented in Phase 11)\nfunc (p *Plan) GetDue(ctx *types.Context, now time.Time) ([]interface{}, error) {\n\treturn []interface{}{}, nil\n}\n" }, { "path": "agent/robot/pool/goroutine_test.go", "content": "package pool_test\n\nimport (\n\t\"context\"\n\t\"runtime\"\n\t\"testing\"\n\t\"time\"\n\n\t\"github.com/stretchr/testify/assert\"\n\t\"github.com/yaoapp/yao/agent/robot/executor\"\n\t\"github.com/yaoapp/yao/agent/robot/pool\"\n\t\"github.com/yaoapp/yao/agent/robot/types\"\n)\n\n// ==================== Goroutine Leak Detection Tests ====================\n\n// getGoroutineCount returns current number of goroutines\nfunc getGoroutineCount() int {\n\treturn runtime.NumGoroutine()\n}\n\n// waitForGoroutineCount waits for goroutine count to stabilize\nfunc waitForGoroutineCount(target int, timeout time.Duration) int {\n\tdeadline := time.Now().Add(timeout)\n\tvar count int\n\tfor time.Now().Before(deadline) {\n\t\tcount = getGoroutineCount()\n\t\tif count <= target {\n\t\t\treturn count\n\t\t}\n\t\truntime.Gosched()\n\t\ttime.Sleep(10 * time.Millisecond)\n\t}\n\treturn count\n}\n\n// TestPoolNoGoroutineLeak tests that pool doesn't leak goroutines after stop\nfunc TestPoolNoGoroutineLeak(t *testing.T) {\n\t// Get baseline goroutine count\n\truntime.GC()\n\ttime.Sleep(50 * time.Millisecond)\n\tbaseline := getGoroutineCount()\n\n\t// Create and start pool\n\texec := executor.NewDryRunWithDelay(10 * time.Millisecond)\n\tp := pool.NewWithConfig(&pool.Config{\n\t\tWorkerSize: 5,\n\t\tQueueSize: 100,\n\t})\n\tp.SetExecutor(exec)\n\tp.Start()\n\n\t// Verify workers are running\n\tafterStart := getGoroutineCount()\n\tassert.Greater(t, afterStart, baseline, \"Should have more goroutines after start\")\n\n\t// Submit some jobs\n\tctx := types.NewContext(context.Background(), nil)\n\trobot := createTestRobot(\"robot_1\", \"team_1\", 5, 10, 5)\n\tfor i := 0; i < 10; i++ {\n\t\tp.Submit(ctx, robot, types.TriggerClock, nil)\n\t}\n\n\t// Wait for jobs to complete\n\ttime.Sleep(300 * time.Millisecond)\n\n\t// Stop pool\n\tp.Stop()\n\n\t// Wait for goroutines to clean up\n\tfinalCount := waitForGoroutineCount(baseline+2, 500*time.Millisecond)\n\n\t// Allow small variance (test framework goroutines)\n\tassert.LessOrEqual(t, finalCount, baseline+2,\n\t\t\"Goroutine count should return to near baseline after stop (baseline=%d, final=%d)\", baseline, finalCount)\n}\n\n// TestPoolMultipleStartStop tests no leak with multiple start/stop cycles\nfunc TestPoolMultipleStartStop(t *testing.T) {\n\truntime.GC()\n\ttime.Sleep(50 * time.Millisecond)\n\tbaseline := getGoroutineCount()\n\n\texec := executor.NewDryRunWithDelay(5 * time.Millisecond)\n\n\tfor i := 0; i < 5; i++ {\n\t\tp := pool.NewWithConfig(&pool.Config{\n\t\t\tWorkerSize: 3,\n\t\t\tQueueSize: 50,\n\t\t})\n\t\tp.SetExecutor(exec)\n\t\tp.Start()\n\n\t\t// Submit a few jobs\n\t\tctx := types.NewContext(context.Background(), nil)\n\t\trobot := createTestRobot(\"robot_1\", \"team_1\", 5, 10, 5)\n\t\tfor j := 0; j < 5; j++ {\n\t\t\tp.Submit(ctx, robot, types.TriggerClock, nil)\n\t\t}\n\n\t\ttime.Sleep(100 * time.Millisecond)\n\t\tp.Stop()\n\t}\n\n\t// Wait for cleanup\n\tfinalCount := waitForGoroutineCount(baseline+2, 500*time.Millisecond)\n\n\tassert.LessOrEqual(t, finalCount, baseline+2,\n\t\t\"Goroutine count should return to near baseline after multiple cycles (baseline=%d, final=%d)\", baseline, finalCount)\n}\n\n// TestPoolStopWithoutJobs tests no leak when stopping pool with no jobs submitted\nfunc TestPoolStopWithoutJobs(t *testing.T) {\n\truntime.GC()\n\ttime.Sleep(50 * time.Millisecond)\n\tbaseline := getGoroutineCount()\n\n\texec := executor.New()\n\tp := pool.NewWithConfig(&pool.Config{\n\t\tWorkerSize: 10,\n\t\tQueueSize: 100,\n\t})\n\tp.SetExecutor(exec)\n\tp.Start()\n\n\t// Immediately stop without submitting any jobs\n\tp.Stop()\n\n\tfinalCount := waitForGoroutineCount(baseline+2, 500*time.Millisecond)\n\n\tassert.LessOrEqual(t, finalCount, baseline+2,\n\t\t\"Goroutine count should return to near baseline (baseline=%d, final=%d)\", baseline, finalCount)\n}\n\n// TestPoolStopWithPendingJobs tests no leak when stopping with jobs in queue\nfunc TestPoolStopWithPendingJobs(t *testing.T) {\n\truntime.GC()\n\ttime.Sleep(50 * time.Millisecond)\n\tbaseline := getGoroutineCount()\n\n\t// Use slow executor so jobs stay in queue\n\texec := executor.NewDryRunWithDelay(500 * time.Millisecond)\n\tp := pool.NewWithConfig(&pool.Config{\n\t\tWorkerSize: 1, // only 1 worker\n\t\tQueueSize: 100,\n\t})\n\tp.SetExecutor(exec)\n\tp.Start()\n\n\t// Submit many jobs (most will be queued)\n\tctx := types.NewContext(context.Background(), nil)\n\trobot := createTestRobot(\"robot_1\", \"team_1\", 5, 50, 5)\n\tfor i := 0; i < 20; i++ {\n\t\tp.Submit(ctx, robot, types.TriggerClock, nil)\n\t}\n\n\t// Stop immediately (some jobs still in queue)\n\ttime.Sleep(50 * time.Millisecond)\n\tp.Stop()\n\n\tfinalCount := waitForGoroutineCount(baseline+2, 500*time.Millisecond)\n\n\tassert.LessOrEqual(t, finalCount, baseline+2,\n\t\t\"Goroutine count should return to near baseline even with pending jobs (baseline=%d, final=%d)\", baseline, finalCount)\n}\n\n// TestPoolConcurrentStartStop tests no leak with concurrent start/stop\nfunc TestPoolConcurrentStartStop(t *testing.T) {\n\truntime.GC()\n\ttime.Sleep(50 * time.Millisecond)\n\tbaseline := getGoroutineCount()\n\n\texec := executor.NewDryRunWithDelay(10 * time.Millisecond)\n\tp := pool.NewWithConfig(&pool.Config{\n\t\tWorkerSize: 5,\n\t\tQueueSize: 100,\n\t})\n\tp.SetExecutor(exec)\n\n\t// Start pool\n\tp.Start()\n\n\t// Concurrent operations\n\tdone := make(chan bool, 3)\n\n\t// Goroutine 1: Submit jobs\n\tgo func() {\n\t\tctx := types.NewContext(context.Background(), nil)\n\t\trobot := createTestRobot(\"robot_1\", \"team_1\", 5, 10, 5)\n\t\tfor i := 0; i < 20; i++ {\n\t\t\tp.Submit(ctx, robot, types.TriggerClock, nil)\n\t\t\ttime.Sleep(5 * time.Millisecond)\n\t\t}\n\t\tdone <- true\n\t}()\n\n\t// Goroutine 2: Check status\n\tgo func() {\n\t\tfor i := 0; i < 20; i++ {\n\t\t\t_ = p.Running()\n\t\t\t_ = p.Queued()\n\t\t\ttime.Sleep(5 * time.Millisecond)\n\t\t}\n\t\tdone <- true\n\t}()\n\n\t// Wait for operations\n\t<-done\n\t<-done\n\n\t// Stop pool\n\tp.Stop()\n\n\tfinalCount := waitForGoroutineCount(baseline+2, 500*time.Millisecond)\n\n\tassert.LessOrEqual(t, finalCount, baseline+2,\n\t\t\"Goroutine count should return to near baseline after concurrent ops (baseline=%d, final=%d)\", baseline, finalCount)\n}\n\n// TestWorkerGoroutinesCleanup tests that worker goroutines are properly cleaned up\nfunc TestWorkerGoroutinesCleanup(t *testing.T) {\n\truntime.GC()\n\ttime.Sleep(50 * time.Millisecond)\n\tbaseline := getGoroutineCount()\n\n\texec := executor.NewDryRunWithDelay(10 * time.Millisecond)\n\n\t// Create pool with many workers\n\tp := pool.NewWithConfig(&pool.Config{\n\t\tWorkerSize: 20,\n\t\tQueueSize: 100,\n\t})\n\tp.SetExecutor(exec)\n\tp.Start()\n\n\t// Should have baseline + 20 workers\n\tafterStart := getGoroutineCount()\n\tassert.GreaterOrEqual(t, afterStart, baseline+20, \"Should have at least 20 worker goroutines\")\n\n\t// Stop pool\n\tp.Stop()\n\n\t// All worker goroutines should be cleaned up\n\tfinalCount := waitForGoroutineCount(baseline+2, 500*time.Millisecond)\n\n\tassert.LessOrEqual(t, finalCount, baseline+2,\n\t\t\"All worker goroutines should be cleaned up (baseline=%d, final=%d)\", baseline, finalCount)\n}\n\n// TestPoolLongRunningJobsNoLeak tests no leak with long-running jobs\nfunc TestPoolLongRunningJobsNoLeak(t *testing.T) {\n\truntime.GC()\n\ttime.Sleep(50 * time.Millisecond)\n\tbaseline := getGoroutineCount()\n\n\texec := executor.NewDryRunWithDelay(200 * time.Millisecond)\n\tp := pool.NewWithConfig(&pool.Config{\n\t\tWorkerSize: 3,\n\t\tQueueSize: 100,\n\t})\n\tp.SetExecutor(exec)\n\tp.Start()\n\n\t// Submit jobs\n\tctx := types.NewContext(context.Background(), nil)\n\trobot := createTestRobot(\"robot_1\", \"team_1\", 5, 10, 5)\n\tfor i := 0; i < 5; i++ {\n\t\tp.Submit(ctx, robot, types.TriggerClock, nil)\n\t}\n\n\t// Wait for some jobs to complete\n\ttime.Sleep(500 * time.Millisecond)\n\n\t// Stop pool\n\tp.Stop()\n\n\tfinalCount := waitForGoroutineCount(baseline+2, 500*time.Millisecond)\n\n\tassert.LessOrEqual(t, finalCount, baseline+2,\n\t\t\"No goroutine leak after long-running jobs (baseline=%d, final=%d)\", baseline, finalCount)\n}\n\n// TestQueueNoGoroutineLeak tests that queue operations don't leak goroutines\nfunc TestQueueNoGoroutineLeak(t *testing.T) {\n\truntime.GC()\n\ttime.Sleep(50 * time.Millisecond)\n\tbaseline := getGoroutineCount()\n\n\t// Create queue and perform many operations\n\tpq := pool.NewPriorityQueue(1000)\n\n\t// Enqueue many items\n\tfor i := 0; i < 500; i++ {\n\t\trobot := createTestRobot(\"robot_\"+string(rune('A'+i%26)), \"team_1\", 5, 100, 5)\n\t\tpq.Enqueue(&pool.QueueItem{\n\t\t\tRobot: robot,\n\t\t\tTrigger: types.TriggerClock,\n\t\t})\n\t}\n\n\t// Dequeue all items\n\tfor pq.Size() > 0 {\n\t\tpq.Dequeue()\n\t}\n\n\truntime.GC()\n\ttime.Sleep(50 * time.Millisecond)\n\tfinalCount := getGoroutineCount()\n\n\tassert.LessOrEqual(t, finalCount, baseline+2,\n\t\t\"Queue operations should not leak goroutines (baseline=%d, final=%d)\", baseline, finalCount)\n}\n" }, { "path": "agent/robot/pool/pool.go", "content": "package pool\n\nimport (\n\t\"fmt\"\n\t\"sync\"\n\t\"sync/atomic\"\n\n\t\"github.com/yaoapp/yao/agent/robot/types\"\n\t\"github.com/yaoapp/yao/agent/robot/utils\"\n)\n\n// Default configuration values\nconst (\n\tDefaultWorkerSize = 10 // default number of workers\n\tDefaultQueueSize = 100 // default global queue size\n)\n\n// Config holds pool configuration\ntype Config struct {\n\tWorkerSize int // number of workers (default: 10)\n\tQueueSize int // global queue size (default: 100)\n}\n\n// DefaultConfig returns default pool configuration\nfunc DefaultConfig() *Config {\n\treturn &Config{\n\t\tWorkerSize: DefaultWorkerSize,\n\t\tQueueSize: DefaultQueueSize,\n\t}\n}\n\n// ExecutorFactory creates an executor based on the mode\ntype ExecutorFactory func(mode types.ExecutorMode) types.Executor\n\n// OnCompleteCallback is called when an execution completes (success or failure)\n// Parameters: execID, memberID, status\ntype OnCompleteCallback func(execID, memberID string, status types.ExecStatus)\n\n// Pool implements types.Pool interface\n// Manages a pool of workers that execute robot jobs from a priority queue\ntype Pool struct {\n\tsize int // number of workers\n\tqueue *PriorityQueue // priority queue for pending jobs\n\texecutor types.Executor // default executor for running jobs\n\texecutorFactory ExecutorFactory // optional: factory for mode-specific executors\n\tonComplete OnCompleteCallback // optional: callback when execution completes\n\tworkers []*Worker // worker goroutines\n\trunning atomic.Int32 // number of currently running jobs\n\twg sync.WaitGroup // wait group for graceful shutdown\n\tstarted bool // whether pool has been started\n\tmu sync.RWMutex // protects started flag\n}\n\n// New creates a new pool instance with default configuration\nfunc New() *Pool {\n\treturn NewWithConfig(nil)\n}\n\n// NewWithConfig creates a new pool instance with custom configuration\nfunc NewWithConfig(config *Config) *Pool {\n\tif config == nil {\n\t\tconfig = DefaultConfig()\n\t}\n\n\t// Apply defaults for zero values\n\tworkerSize := config.WorkerSize\n\tif workerSize <= 0 {\n\t\tworkerSize = DefaultWorkerSize\n\t}\n\n\tqueueSize := config.QueueSize\n\tif queueSize <= 0 {\n\t\tqueueSize = DefaultQueueSize\n\t}\n\n\treturn &Pool{\n\t\tsize: workerSize,\n\t\tqueue: NewPriorityQueue(queueSize),\n\t}\n}\n\n// SetExecutor sets the default executor for the pool\n// Must be called before Start()\nfunc (p *Pool) SetExecutor(executor types.Executor) {\n\tp.executor = executor\n}\n\n// SetExecutorFactory sets the executor factory for mode-specific executors\n// If set, the factory is used to create executors based on ExecutorMode\nfunc (p *Pool) SetExecutorFactory(factory ExecutorFactory) {\n\tp.executorFactory = factory\n}\n\n// SetOnComplete sets the callback for execution completion\n// Called when an execution finishes (completed, failed, or cancelled)\nfunc (p *Pool) SetOnComplete(callback OnCompleteCallback) {\n\tp.onComplete = callback\n}\n\n// GetExecutor returns the appropriate executor for the given mode\n// If factory is set and mode is specified, uses factory; otherwise uses default\nfunc (p *Pool) GetExecutor(mode types.ExecutorMode) types.Executor {\n\t// If factory is set and mode is specified, use factory\n\tif p.executorFactory != nil && mode != \"\" {\n\t\treturn p.executorFactory(mode)\n\t}\n\t// Otherwise use default executor\n\treturn p.executor\n}\n\n// Start starts the worker pool\nfunc (p *Pool) Start() error {\n\tp.mu.Lock()\n\tdefer p.mu.Unlock()\n\n\tif p.started {\n\t\treturn fmt.Errorf(\"pool already started\")\n\t}\n\n\tif p.executor == nil {\n\t\treturn fmt.Errorf(\"executor not set, call SetExecutor() first\")\n\t}\n\n\t// Create and start workers\n\tp.workers = make([]*Worker, p.size)\n\tfor i := 0; i < p.size; i++ {\n\t\tworker := newWorker(i+1, p, &p.wg)\n\t\tp.workers[i] = worker\n\t\tworker.start()\n\t}\n\n\tp.started = true\n\treturn nil\n}\n\n// Stop stops the worker pool gracefully\n// Waits for all running jobs to complete\nfunc (p *Pool) Stop() error {\n\tp.mu.Lock()\n\tif !p.started {\n\t\tp.mu.Unlock()\n\t\treturn nil // already stopped or never started\n\t}\n\tp.started = false\n\tp.mu.Unlock()\n\n\t// Stop all workers\n\tfor _, worker := range p.workers {\n\t\tworker.stop()\n\t}\n\n\t// Wait for all workers to finish\n\tp.wg.Wait()\n\n\treturn nil\n}\n\n// Submit submits a robot execution to the pool\n// Returns execution ID if successfully queued, error otherwise\nfunc (p *Pool) Submit(ctx *types.Context, robot *types.Robot, trigger types.TriggerType, data interface{}) (string, error) {\n\treturn p.SubmitWithMode(ctx, robot, trigger, data, \"\")\n}\n\n// GenerateExecID generates a new execution ID\n// Exported so Manager can pre-generate IDs for tracking\nfunc GenerateExecID() string {\n\treturn utils.NewID()\n}\n\n// SubmitWithMode submits a robot execution with specified executor mode\n// executorMode: optional, overrides robot's config if provided\n// Returns execution ID if successfully queued, error otherwise\n// Note: This method does not support execution control (pause/resume)\nfunc (p *Pool) SubmitWithMode(ctx *types.Context, robot *types.Robot, trigger types.TriggerType, data interface{}, executorMode types.ExecutorMode) (string, error) {\n\texecID := GenerateExecID()\n\treturn p.submitWithIDAndMode(ctx, robot, trigger, data, execID, executorMode, nil)\n}\n\n// SubmitWithID submits a robot execution with a pre-generated execution ID\n// This is used when the caller needs to track the execution before submission\nfunc (p *Pool) SubmitWithID(ctx *types.Context, robot *types.Robot, trigger types.TriggerType, data interface{}, execID string, control types.ExecutionControl) (string, error) {\n\treturn p.submitWithIDAndMode(ctx, robot, trigger, data, execID, \"\", control)\n}\n\n// submitWithIDAndMode is the internal implementation that handles both cases\nfunc (p *Pool) submitWithIDAndMode(ctx *types.Context, robot *types.Robot, trigger types.TriggerType, data interface{}, execID string, executorMode types.ExecutorMode, control types.ExecutionControl) (string, error) {\n\tp.mu.RLock()\n\tif !p.started {\n\t\tp.mu.RUnlock()\n\t\treturn \"\", fmt.Errorf(\"pool not started\")\n\t}\n\tp.mu.RUnlock()\n\n\tif robot == nil {\n\t\treturn \"\", fmt.Errorf(\"robot cannot be nil\")\n\t}\n\n\t// Create queue item with the provided ID and control\n\titem := &QueueItem{\n\t\tRobot: robot,\n\t\tCtx: ctx,\n\t\tTrigger: trigger,\n\t\tData: data,\n\t\tExecutorMode: executorMode,\n\t\tExecID: execID,\n\t\tControl: control,\n\t}\n\n\t// Try to add to queue\n\tif !p.queue.Enqueue(item) {\n\t\treturn \"\", fmt.Errorf(\"queue full (max %d items)\", p.queue.maxSize)\n\t}\n\n\treturn execID, nil\n}\n\n// Running returns number of currently running jobs\nfunc (p *Pool) Running() int {\n\treturn int(p.running.Load())\n}\n\n// Queued returns number of queued jobs\nfunc (p *Pool) Queued() int {\n\treturn p.queue.Size()\n}\n\n// incrementRunning increments the running counter\nfunc (p *Pool) incrementRunning() {\n\tp.running.Add(1)\n}\n\n// decrementRunning decrements the running counter\nfunc (p *Pool) decrementRunning() {\n\tp.running.Add(-1)\n}\n\n// Size returns the configured pool size\nfunc (p *Pool) Size() int {\n\treturn p.size\n}\n\n// QueueSize returns the configured queue size\nfunc (p *Pool) QueueSize() int {\n\treturn p.queue.maxSize\n}\n\n// IsStarted returns true if the pool has been started\nfunc (p *Pool) IsStarted() bool {\n\tp.mu.RLock()\n\tdefer p.mu.RUnlock()\n\treturn p.started\n}\n" }, { "path": "agent/robot/pool/pool_test.go", "content": "package pool_test\n\nimport (\n\t\"context\"\n\t\"testing\"\n\t\"time\"\n\n\t\"github.com/stretchr/testify/assert\"\n\t\"github.com/yaoapp/yao/agent/robot/executor\"\n\t\"github.com/yaoapp/yao/agent/robot/pool\"\n\t\"github.com/yaoapp/yao/agent/robot/types\"\n)\n\n// createTestRobot creates a robot for testing with specified quota\nfunc createTestRobot(memberID, teamID string, maxConcurrent, queueSize, priority int) *types.Robot {\n\treturn &types.Robot{\n\t\tMemberID: memberID,\n\t\tTeamID: teamID,\n\t\tDisplayName: \"Test Robot \" + memberID,\n\t\tStatus: types.RobotIdle,\n\t\tAutonomousMode: true,\n\t\tConfig: &types.Config{\n\t\t\tIdentity: &types.Identity{Role: \"Test\"},\n\t\t\tQuota: &types.Quota{\n\t\t\t\tMax: maxConcurrent,\n\t\t\t\tQueue: queueSize,\n\t\t\t\tPriority: priority,\n\t\t\t},\n\t\t},\n\t}\n}\n\n// createTestContext creates a context for testing\nfunc createTestContext() *types.Context {\n\treturn types.NewContext(context.Background(), nil)\n}\n\n// TestPoolStartStop tests pool start and stop lifecycle\nfunc TestPoolStartStop(t *testing.T) {\n\tp := pool.New()\n\texec := executor.New()\n\tp.SetExecutor(exec)\n\n\tt.Run(\"start pool\", func(t *testing.T) {\n\t\terr := p.Start()\n\t\tassert.NoError(t, err)\n\t\tassert.True(t, p.IsStarted())\n\t})\n\n\tt.Run(\"start already started pool\", func(t *testing.T) {\n\t\terr := p.Start()\n\t\tassert.Error(t, err)\n\t\tassert.Contains(t, err.Error(), \"already started\")\n\t})\n\n\tt.Run(\"stop pool\", func(t *testing.T) {\n\t\terr := p.Stop()\n\t\tassert.NoError(t, err)\n\t\tassert.False(t, p.IsStarted())\n\t})\n\n\tt.Run(\"stop already stopped pool\", func(t *testing.T) {\n\t\terr := p.Stop()\n\t\tassert.NoError(t, err) // should not error\n\t})\n}\n\n// TestPoolSubmitWithoutStart tests submitting to unstarted pool\nfunc TestPoolSubmitWithoutStart(t *testing.T) {\n\tp := pool.New()\n\texec := executor.New()\n\tp.SetExecutor(exec)\n\n\trobot := createTestRobot(\"robot_1\", \"team_1\", 2, 10, 5)\n\tctx := createTestContext()\n\n\t_, err := p.Submit(ctx, robot, types.TriggerClock, nil)\n\tassert.Error(t, err)\n\tassert.Contains(t, err.Error(), \"not started\")\n}\n\n// TestPoolSubmitNilRobot tests submitting nil robot\nfunc TestPoolSubmitNilRobot(t *testing.T) {\n\tp := pool.New()\n\texec := executor.New()\n\tp.SetExecutor(exec)\n\tp.Start()\n\tdefer p.Stop()\n\n\tctx := createTestContext()\n\t_, err := p.Submit(ctx, nil, types.TriggerClock, nil)\n\tassert.Error(t, err)\n\tassert.Contains(t, err.Error(), \"cannot be nil\")\n}\n\n// TestPoolBasicExecution tests basic job execution\nfunc TestPoolBasicExecution(t *testing.T) {\n\texec := executor.NewDryRunWithDelay(50 * time.Millisecond)\n\tp := pool.NewWithConfig(&pool.Config{\n\t\tWorkerSize: 5,\n\t\tQueueSize: 100,\n\t})\n\tp.SetExecutor(exec)\n\tp.Start()\n\tdefer p.Stop()\n\n\trobot := createTestRobot(\"robot_1\", \"team_1\", 2, 10, 5)\n\tctx := createTestContext()\n\n\t// Submit a job\n\texecID, err := p.Submit(ctx, robot, types.TriggerClock, nil)\n\tassert.NoError(t, err)\n\tassert.NotEmpty(t, execID)\n\n\t// Wait for execution (worker polls every 100ms + 50ms exec + buffer)\n\ttime.Sleep(300 * time.Millisecond)\n\n\t// Verify execution completed\n\tassert.Equal(t, 1, exec.ExecCount())\n\t// Note: CurrentCount may briefly be non-zero during execution, use Eventually pattern\n\tassert.Eventually(t, func() bool {\n\t\treturn exec.CurrentCount() == 0\n\t}, 500*time.Millisecond, 50*time.Millisecond, \"CurrentCount should be 0 after execution\")\n}\n\n// TestPoolConcurrencyLimit tests global worker limit\nfunc TestPoolConcurrencyLimit(t *testing.T) {\n\texec := executor.NewDryRunWithDelay(200 * time.Millisecond) // longer delay\n\tp := pool.NewWithConfig(&pool.Config{\n\t\tWorkerSize: 3, // only 3 workers\n\t\tQueueSize: 100,\n\t})\n\tp.SetExecutor(exec)\n\tp.Start()\n\tdefer p.Stop()\n\n\tctx := createTestContext()\n\n\t// Create robots with high quota (won't be the bottleneck)\n\trobots := make([]*types.Robot, 10)\n\tfor i := 0; i < 10; i++ {\n\t\trobots[i] = createTestRobot(\n\t\t\t\"robot_\"+string(rune('A'+i)),\n\t\t\t\"team_1\",\n\t\t\t5, // max concurrent per robot\n\t\t\t20, // queue size per robot\n\t\t\t5, // priority\n\t\t)\n\t}\n\n\t// Submit 10 jobs\n\tfor i := 0; i < 10; i++ {\n\t\t_, err := p.Submit(ctx, robots[i], types.TriggerClock, nil)\n\t\tassert.NoError(t, err)\n\t}\n\n\t// Wait for workers to pick up jobs (worker polls every 100ms)\n\ttime.Sleep(200 * time.Millisecond)\n\n\t// Should have at most 3 running (worker limit)\n\trunning := p.Running()\n\tassert.LessOrEqual(t, running, 3, \"Should not exceed worker limit\")\n\n\t// Wait for all to complete (10 jobs / 3 workers * 200ms each = ~700ms + buffer)\n\t// Use Eventually to handle CI timing variations\n\tassert.Eventually(t, func() bool {\n\t\treturn exec.ExecCount() >= 10\n\t}, 2*time.Second, 100*time.Millisecond, \"All 10 jobs should complete\")\n}\n\n// TestRobotConcurrencyLimit tests per-robot concurrent execution limit\nfunc TestRobotConcurrencyLimit(t *testing.T) {\n\texec := executor.NewDryRunWithDelay(100 * time.Millisecond)\n\tp := pool.NewWithConfig(&pool.Config{\n\t\tWorkerSize: 10, // plenty of workers\n\t\tQueueSize: 100,\n\t})\n\tp.SetExecutor(exec)\n\tp.Start()\n\tdefer p.Stop()\n\n\tctx := createTestContext()\n\n\t// Create robot with Max=2 (can only run 2 at a time)\n\trobot := createTestRobot(\"robot_limited\", \"team_1\", 2, 20, 5)\n\n\t// Submit 5 jobs for the same robot\n\tfor i := 0; i < 5; i++ {\n\t\t_, err := p.Submit(ctx, robot, types.TriggerClock, nil)\n\t\tassert.NoError(t, err)\n\t}\n\n\t// Wait a bit for execution to start\n\ttime.Sleep(150 * time.Millisecond)\n\n\t// Robot should have at most 2 running (Quota.Max=2)\n\trunningCount := robot.RunningCount()\n\tassert.LessOrEqual(t, runningCount, 2, \"Robot should not exceed Quota.Max\")\n\n\t// Wait for all to complete (with re-enqueue, need more time)\n\t// 5 jobs with Max=2: ~3 batches * 100ms exec + poll overhead\n\ttime.Sleep(800 * time.Millisecond)\n\n\t// All 5 jobs should eventually execute\n\tassert.GreaterOrEqual(t, exec.ExecCount(), 5, \"All jobs should eventually execute\")\n}\n\n// TestRobotQueueLimit tests per-robot queue limit\nfunc TestRobotQueueLimit(t *testing.T) {\n\texec := executor.NewDryRunWithDelay(200 * time.Millisecond)\n\tp := pool.NewWithConfig(&pool.Config{\n\t\tWorkerSize: 2,\n\t\tQueueSize: 100, // global queue is large\n\t})\n\tp.SetExecutor(exec)\n\tp.Start()\n\tdefer p.Stop()\n\n\tctx := createTestContext()\n\n\t// Create robot with small queue limit\n\trobot := createTestRobot(\"robot_small_queue\", \"team_1\", 1, 3, 5) // Queue=3\n\n\t// Submit jobs until queue limit is reached\n\tsuccessCount := 0\n\tfor i := 0; i < 10; i++ {\n\t\t_, err := p.Submit(ctx, robot, types.TriggerClock, nil)\n\t\tif err == nil {\n\t\t\tsuccessCount++\n\t\t}\n\t}\n\n\t// Should only accept up to Queue limit (some may have started executing)\n\t// Max accepted = Queue(3) + Max(1) = 4 (1 running + 3 in queue)\n\tassert.LessOrEqual(t, successCount, 4, \"Should respect robot queue limit\")\n\tassert.GreaterOrEqual(t, successCount, 1, \"Should accept at least 1 job\")\n}\n\n// TestGlobalQueueLimit tests global queue limit\nfunc TestGlobalQueueLimit(t *testing.T) {\n\texec := executor.NewDryRunWithDelay(500 * time.Millisecond) // slow execution\n\tp := pool.NewWithConfig(&pool.Config{\n\t\tWorkerSize: 1, // only 1 worker\n\t\tQueueSize: 5, // small global queue\n\t})\n\tp.SetExecutor(exec)\n\tp.Start()\n\tdefer p.Stop()\n\n\tctx := createTestContext()\n\n\t// Create multiple robots with large queue limits\n\tsuccessCount := 0\n\tfor i := 0; i < 20; i++ {\n\t\trobot := createTestRobot(\n\t\t\t\"robot_\"+string(rune('A'+i%26)),\n\t\t\t\"team_1\",\n\t\t\t5, // large max\n\t\t\t20, // large per-robot queue\n\t\t\t5,\n\t\t)\n\t\t_, err := p.Submit(ctx, robot, types.TriggerClock, nil)\n\t\tif err == nil {\n\t\t\tsuccessCount++\n\t\t}\n\t}\n\n\t// Should only accept up to global queue limit + running\n\t// Max = QueueSize(5) + WorkerSize(1) = 6\n\tassert.LessOrEqual(t, successCount, 6, \"Should respect global queue limit\")\n}\n\n// TestPriorityOrder tests that higher priority jobs execute first\nfunc TestPriorityOrder(t *testing.T) {\n\texec := executor.NewDryRunWithDelay(50 * time.Millisecond)\n\tp := pool.NewWithConfig(&pool.Config{\n\t\tWorkerSize: 1, // single worker to ensure order\n\t\tQueueSize: 100,\n\t})\n\tp.SetExecutor(exec)\n\tp.Start()\n\tdefer p.Stop()\n\n\tctx := createTestContext()\n\n\t// Create robots with different priorities\n\trobotLow := createTestRobot(\"robot_low\", \"team_1\", 5, 20, 1) // priority 1\n\trobotMed := createTestRobot(\"robot_med\", \"team_1\", 5, 20, 5) // priority 5\n\trobotHigh := createTestRobot(\"robot_high\", \"team_1\", 5, 20, 10) // priority 10\n\n\t// Submit in low-to-high order\n\tp.Submit(ctx, robotLow, types.TriggerClock, nil)\n\tp.Submit(ctx, robotMed, types.TriggerClock, nil)\n\tp.Submit(ctx, robotHigh, types.TriggerClock, nil)\n\n\t// Wait for all to complete (3 jobs * (100ms poll + 50ms exec) = ~450ms + buffer)\n\t// Use Eventually for CI timing variations\n\tassert.Eventually(t, func() bool {\n\t\treturn exec.ExecCount() >= 3\n\t}, 1*time.Second, 50*time.Millisecond, \"All 3 jobs should complete\")\n}\n\n// TestTriggerTypePriority tests that human triggers have higher priority than clock\nfunc TestTriggerTypePriority(t *testing.T) {\n\texec := executor.NewDryRunWithDelay(50 * time.Millisecond)\n\tp := pool.NewWithConfig(&pool.Config{\n\t\tWorkerSize: 1, // single worker\n\t\tQueueSize: 100,\n\t})\n\tp.SetExecutor(exec)\n\tp.Start()\n\tdefer p.Stop()\n\n\tctx := createTestContext()\n\n\t// Same robot, same priority, different trigger types\n\trobot := createTestRobot(\"robot_1\", \"team_1\", 5, 20, 5)\n\n\t// Submit clock first, then human\n\tp.Submit(ctx, robot, types.TriggerClock, nil)\n\tp.Submit(ctx, robot, types.TriggerHuman, nil) // should execute first\n\n\t// Wait for all to complete (2 jobs * (100ms poll + 50ms exec) = ~300ms + buffer)\n\tassert.Eventually(t, func() bool {\n\t\treturn exec.ExecCount() >= 2\n\t}, 1*time.Second, 50*time.Millisecond, \"Both jobs should complete\")\n}\n\n// TestMultipleRobotsFairness tests that multiple robots get fair access\nfunc TestMultipleRobotsFairness(t *testing.T) {\n\texec := executor.NewDryRunWithDelay(30 * time.Millisecond)\n\tp := pool.NewWithConfig(&pool.Config{\n\t\tWorkerSize: 5,\n\t\tQueueSize: 100,\n\t})\n\tp.SetExecutor(exec)\n\tp.Start()\n\tdefer p.Stop()\n\n\tctx := createTestContext()\n\n\t// Create 3 robots with same priority\n\trobotA := createTestRobot(\"robot_A\", \"team_1\", 2, 10, 5)\n\trobotB := createTestRobot(\"robot_B\", \"team_1\", 2, 10, 5)\n\trobotC := createTestRobot(\"robot_C\", \"team_1\", 2, 10, 5)\n\n\t// Submit jobs for each robot\n\tfor i := 0; i < 6; i++ {\n\t\tp.Submit(ctx, robotA, types.TriggerClock, nil)\n\t\tp.Submit(ctx, robotB, types.TriggerClock, nil)\n\t\tp.Submit(ctx, robotC, types.TriggerClock, nil)\n\t}\n\n\t// Wait for all to complete\n\t// 18 jobs with Quota.Max=2 per robot, 5 workers, 30ms each\n\t// Jobs are batched by robot quota, use Eventually for CI timing\n\tassert.Eventually(t, func() bool {\n\t\treturn exec.ExecCount() >= 18\n\t}, 3*time.Second, 100*time.Millisecond, \"All 18 jobs should complete\")\n}\n\n// TestGracefulShutdown tests that pool waits for running jobs on shutdown\nfunc TestGracefulShutdown(t *testing.T) {\n\texec := executor.NewDryRunWithDelay(200 * time.Millisecond)\n\tp := pool.NewWithConfig(&pool.Config{\n\t\tWorkerSize: 2,\n\t\tQueueSize: 10,\n\t})\n\tp.SetExecutor(exec)\n\tp.Start()\n\n\tctx := createTestContext()\n\trobot := createTestRobot(\"robot_1\", \"team_1\", 5, 20, 5)\n\n\t// Submit 2 jobs\n\tp.Submit(ctx, robot, types.TriggerClock, nil)\n\tp.Submit(ctx, robot, types.TriggerClock, nil)\n\n\t// Wait for workers to pick up jobs (poll every 100ms)\n\ttime.Sleep(150 * time.Millisecond)\n\n\t// Verify jobs are running\n\tassert.GreaterOrEqual(t, p.Running(), 1, \"Should have at least 1 running job\")\n\n\t// Stop - workers will finish their current tick cycle\n\tp.Stop()\n\n\t// After stop, verify jobs completed\n\tassert.GreaterOrEqual(t, exec.ExecCount(), 1, \"Should have executed at least 1 job\")\n}\n\n// TestDefaultConfig tests default configuration values\nfunc TestDefaultConfig(t *testing.T) {\n\tconfig := pool.DefaultConfig()\n\tassert.Equal(t, pool.DefaultWorkerSize, config.WorkerSize)\n\tassert.Equal(t, pool.DefaultQueueSize, config.QueueSize)\n}\n\n// TestPoolWithNilConfig tests pool creation with nil config\nfunc TestPoolWithNilConfig(t *testing.T) {\n\tp := pool.NewWithConfig(nil)\n\tassert.Equal(t, pool.DefaultWorkerSize, p.Size())\n\tassert.Equal(t, pool.DefaultQueueSize, p.QueueSize())\n}\n\n// TestPoolWithZeroConfig tests pool creation with zero values\nfunc TestPoolWithZeroConfig(t *testing.T) {\n\tp := pool.NewWithConfig(&pool.Config{\n\t\tWorkerSize: 0,\n\t\tQueueSize: 0,\n\t})\n\t// Should use defaults for zero values\n\tassert.Equal(t, pool.DefaultWorkerSize, p.Size())\n\tassert.Equal(t, pool.DefaultQueueSize, p.QueueSize())\n}\n\n// TestPoolWithoutExecutor tests starting pool without executor\nfunc TestPoolWithoutExecutor(t *testing.T) {\n\tp := pool.New()\n\t// Don't set executor\n\terr := p.Start()\n\tassert.Error(t, err)\n\tassert.Contains(t, err.Error(), \"executor not set\")\n}\n" }, { "path": "agent/robot/pool/queue.go", "content": "package pool\n\nimport (\n\t\"container/heap\"\n\t\"sync\"\n\t\"time\"\n\n\t\"github.com/yaoapp/yao/agent/robot/types\"\n)\n\n// QueueItem represents a job waiting in the queue\ntype QueueItem struct {\n\tRobot *types.Robot\n\tCtx *types.Context\n\tTrigger types.TriggerType\n\tData interface{}\n\tExecutorMode types.ExecutorMode // optional: override robot's executor mode\n\tExecID string // pre-generated execution ID for tracking\n\tControl types.ExecutionControl // execution control for pause/resume/stop\n\tEnqueueTime time.Time\n\tPriority int // calculated priority for sorting\n\tIndex int // index in heap (managed by container/heap)\n}\n\n// PriorityQueue implements a priority queue for robot executions\n// Sorted by: robot priority > trigger type priority > wait time\ntype PriorityQueue struct {\n\titems []*QueueItem\n\tmu sync.RWMutex\n\tmaxSize int // global queue size limit\n\trobotCount map[string]int // per-robot queue count: memberID -> count\n}\n\n// NewPriorityQueue creates a new priority queue\nfunc NewPriorityQueue(maxSize int) *PriorityQueue {\n\tpq := &PriorityQueue{\n\t\titems: make([]*QueueItem, 0),\n\t\tmaxSize: maxSize,\n\t\trobotCount: make(map[string]int),\n\t}\n\theap.Init(pq)\n\treturn pq\n}\n\n// Enqueue adds an item to the queue\n// Returns false if:\n// - Global queue is full (maxSize)\n// - Robot's queue limit reached (Quota.Queue)\nfunc (pq *PriorityQueue) Enqueue(item *QueueItem) bool {\n\tpq.mu.Lock()\n\tdefer pq.mu.Unlock()\n\n\t// Check 1: Global queue limit\n\tif pq.maxSize > 0 && len(pq.items) >= pq.maxSize {\n\t\treturn false // global queue full\n\t}\n\n\t// Check 2: Per-robot queue limit (prevents single robot from hogging the queue)\n\tif item.Robot != nil {\n\t\tmemberID := item.Robot.MemberID\n\t\trobotQueueLimit := 10 // default\n\t\tif item.Robot.Config != nil && item.Robot.Config.Quota != nil {\n\t\t\trobotQueueLimit = item.Robot.Config.Quota.GetQueue()\n\t\t}\n\n\t\tif pq.robotCount[memberID] >= robotQueueLimit {\n\t\t\treturn false // robot's queue limit reached\n\t\t}\n\n\t\t// Increment robot's queue count\n\t\tpq.robotCount[memberID]++\n\t}\n\n\titem.Priority = calculatePriority(item)\n\titem.EnqueueTime = time.Now()\n\theap.Push(pq, item)\n\treturn true\n}\n\n// Dequeue removes and returns the highest priority item\n// Returns nil if queue is empty\nfunc (pq *PriorityQueue) Dequeue() *QueueItem {\n\tpq.mu.Lock()\n\tdefer pq.mu.Unlock()\n\n\tif len(pq.items) == 0 {\n\t\treturn nil\n\t}\n\n\titem := heap.Pop(pq).(*QueueItem)\n\n\t// Decrement robot's queue count\n\tif item.Robot != nil {\n\t\tmemberID := item.Robot.MemberID\n\t\tif pq.robotCount[memberID] > 0 {\n\t\t\tpq.robotCount[memberID]--\n\t\t}\n\t\t// Clean up if count reaches zero\n\t\tif pq.robotCount[memberID] == 0 {\n\t\t\tdelete(pq.robotCount, memberID)\n\t\t}\n\t}\n\n\treturn item\n}\n\n// Size returns the number of items in the queue (thread-safe)\nfunc (pq *PriorityQueue) Size() int {\n\tpq.mu.RLock()\n\tdefer pq.mu.RUnlock()\n\treturn len(pq.items)\n}\n\n// IsFull returns true if queue has reached max capacity\nfunc (pq *PriorityQueue) IsFull() bool {\n\tpq.mu.RLock()\n\tdefer pq.mu.RUnlock()\n\treturn pq.maxSize > 0 && len(pq.items) >= pq.maxSize\n}\n\n// RobotQueuedCount returns the number of queued items for a specific robot\nfunc (pq *PriorityQueue) RobotQueuedCount(memberID string) int {\n\tpq.mu.RLock()\n\tdefer pq.mu.RUnlock()\n\treturn pq.robotCount[memberID]\n}\n\n// ==================== heap.Interface implementation ====================\n// These methods are called internally by heap.Push/Pop with lock already held\n\nfunc (pq *PriorityQueue) Len() int { return len(pq.items) }\n\nfunc (pq *PriorityQueue) Less(i, j int) bool {\n\t// Higher priority value = higher priority (processed first)\n\t// If priority is equal, older items (earlier EnqueueTime) come first\n\tif pq.items[i].Priority == pq.items[j].Priority {\n\t\treturn pq.items[i].EnqueueTime.Before(pq.items[j].EnqueueTime)\n\t}\n\treturn pq.items[i].Priority > pq.items[j].Priority\n}\n\nfunc (pq *PriorityQueue) Swap(i, j int) {\n\tpq.items[i], pq.items[j] = pq.items[j], pq.items[i]\n\tpq.items[i].Index = i\n\tpq.items[j].Index = j\n}\n\n// Push is required by heap.Interface\n// Note: This is called by heap.Push(), not directly\nfunc (pq *PriorityQueue) Push(x interface{}) {\n\titem := x.(*QueueItem)\n\titem.Index = len(pq.items)\n\tpq.items = append(pq.items, item)\n}\n\n// Pop is required by heap.Interface\n// Note: This is called by heap.Pop(), not directly\nfunc (pq *PriorityQueue) Pop() interface{} {\n\told := pq.items\n\tn := len(old)\n\titem := old[n-1]\n\told[n-1] = nil // avoid memory leak\n\titem.Index = -1 // mark as removed\n\tpq.items = old[0 : n-1]\n\treturn item\n}\n\n// ==================== Priority Calculation ====================\n\n// calculatePriority calculates the priority score for a queue item\n// Priority = robot_priority * 1000 + trigger_priority * 100\n// Higher score = higher priority\nfunc calculatePriority(item *QueueItem) int {\n\tpriority := 0\n\n\t// 1. Robot priority (from config, 1-10, default 5)\n\tif item.Robot != nil && item.Robot.Config != nil && item.Robot.Config.Quota != nil {\n\t\trobotPriority := item.Robot.Config.Quota.GetPriority()\n\t\tpriority += robotPriority * 1000\n\t} else {\n\t\tpriority += 5000 // default robot priority\n\t}\n\n\t// 2. Trigger type priority\n\t// Human intervention > Event > Clock\n\ttriggerPriority := getTriggerPriority(item.Trigger)\n\tpriority += triggerPriority * 100\n\n\treturn priority\n}\n\n// getTriggerPriority returns priority value for trigger type\nfunc getTriggerPriority(trigger types.TriggerType) int {\n\tswitch trigger {\n\tcase types.TriggerHuman:\n\t\treturn 10 // highest priority\n\tcase types.TriggerEvent:\n\t\treturn 5 // medium priority\n\tcase types.TriggerClock:\n\t\treturn 1 // lowest priority\n\tdefault:\n\t\treturn 0\n\t}\n}\n" }, { "path": "agent/robot/pool/queue_test.go", "content": "package pool_test\n\nimport (\n\t\"testing\"\n\t\"time\"\n\n\t\"github.com/stretchr/testify/assert\"\n\t\"github.com/yaoapp/yao/agent/robot/pool\"\n\t\"github.com/yaoapp/yao/agent/robot/types\"\n)\n\n// ==================== Priority Queue Basic Tests ====================\n\n// TestQueueNewPriorityQueue tests queue creation\nfunc TestQueueNewPriorityQueue(t *testing.T) {\n\tt.Run(\"create with positive size\", func(t *testing.T) {\n\t\tpq := pool.NewPriorityQueue(100)\n\t\tassert.NotNil(t, pq)\n\t\tassert.Equal(t, 0, pq.Size())\n\t\tassert.False(t, pq.IsFull())\n\t})\n\n\tt.Run(\"create with zero size (unlimited)\", func(t *testing.T) {\n\t\tpq := pool.NewPriorityQueue(0)\n\t\tassert.NotNil(t, pq)\n\t\tassert.False(t, pq.IsFull()) // never full when maxSize=0\n\t})\n}\n\n// TestQueueEnqueueDequeue tests basic enqueue and dequeue\nfunc TestQueueEnqueueDequeue(t *testing.T) {\n\tpq := pool.NewPriorityQueue(100)\n\trobot := createTestRobot(\"robot_1\", \"team_1\", 5, 10, 5)\n\tctx := createTestContext()\n\n\tt.Run(\"enqueue single item\", func(t *testing.T) {\n\t\titem := &pool.QueueItem{\n\t\t\tRobot: robot,\n\t\t\tCtx: ctx,\n\t\t\tTrigger: types.TriggerClock,\n\t\t\tData: \"test_data\",\n\t\t}\n\t\tok := pq.Enqueue(item)\n\t\tassert.True(t, ok)\n\t\tassert.Equal(t, 1, pq.Size())\n\t})\n\n\tt.Run(\"dequeue single item\", func(t *testing.T) {\n\t\titem := pq.Dequeue()\n\t\tassert.NotNil(t, item)\n\t\tassert.Equal(t, \"robot_1\", item.Robot.MemberID)\n\t\tassert.Equal(t, \"test_data\", item.Data)\n\t\tassert.Equal(t, 0, pq.Size())\n\t})\n\n\tt.Run(\"dequeue from empty queue\", func(t *testing.T) {\n\t\titem := pq.Dequeue()\n\t\tassert.Nil(t, item)\n\t})\n}\n\n// TestQueueSize tests Size method\nfunc TestQueueSize(t *testing.T) {\n\tpq := pool.NewPriorityQueue(100)\n\trobot := createTestRobot(\"robot_1\", \"team_1\", 5, 10, 5)\n\n\tassert.Equal(t, 0, pq.Size())\n\n\t// Add 5 items\n\tfor i := 0; i < 5; i++ {\n\t\tpq.Enqueue(&pool.QueueItem{Robot: robot, Trigger: types.TriggerClock})\n\t}\n\tassert.Equal(t, 5, pq.Size())\n\n\t// Remove 2 items\n\tpq.Dequeue()\n\tpq.Dequeue()\n\tassert.Equal(t, 3, pq.Size())\n}\n\n// ==================== Global Queue Limit Tests ====================\n\n// TestQueueGlobalLimit tests global queue size limit\nfunc TestQueueGlobalLimit(t *testing.T) {\n\tpq := pool.NewPriorityQueue(5) // max 5 items\n\n\t// Create different robots to avoid per-robot limit\n\tfor i := 0; i < 10; i++ {\n\t\trobot := createTestRobot(\"robot_\"+string(rune('A'+i)), \"team_1\", 5, 10, 5)\n\t\titem := &pool.QueueItem{Robot: robot, Trigger: types.TriggerClock}\n\t\tok := pq.Enqueue(item)\n\n\t\tif i < 5 {\n\t\t\tassert.True(t, ok, \"Should accept item %d\", i)\n\t\t} else {\n\t\t\tassert.False(t, ok, \"Should reject item %d (queue full)\", i)\n\t\t}\n\t}\n\n\tassert.Equal(t, 5, pq.Size())\n\tassert.True(t, pq.IsFull())\n}\n\n// TestQueueUnlimitedSize tests queue with no size limit (maxSize=0)\nfunc TestQueueUnlimitedSize(t *testing.T) {\n\tpq := pool.NewPriorityQueue(0) // unlimited\n\n\t// Add many items\n\tfor i := 0; i < 100; i++ {\n\t\trobot := createTestRobot(\"robot_\"+string(rune('A'+i%26)), \"team_1\", 5, 1000, 5)\n\t\titem := &pool.QueueItem{Robot: robot, Trigger: types.TriggerClock}\n\t\tok := pq.Enqueue(item)\n\t\tassert.True(t, ok)\n\t}\n\n\tassert.Equal(t, 100, pq.Size())\n\tassert.False(t, pq.IsFull()) // never full\n}\n\n// ==================== Per-Robot Queue Limit Tests ====================\n\n// TestQueuePerRobotLimit tests per-robot queue limit (Quota.Queue)\nfunc TestQueuePerRobotLimit(t *testing.T) {\n\tpq := pool.NewPriorityQueue(100) // large global limit\n\n\t// Robot with Queue=3\n\trobot := createTestRobot(\"robot_limited\", \"team_1\", 5, 3, 5)\n\n\t// Try to add 10 items for same robot\n\tsuccessCount := 0\n\tfor i := 0; i < 10; i++ {\n\t\titem := &pool.QueueItem{Robot: robot, Trigger: types.TriggerClock}\n\t\tif pq.Enqueue(item) {\n\t\t\tsuccessCount++\n\t\t}\n\t}\n\n\t// Should only accept Queue(3) items\n\tassert.Equal(t, 3, successCount)\n\tassert.Equal(t, 3, pq.Size())\n\tassert.Equal(t, 3, pq.RobotQueuedCount(\"robot_limited\"))\n}\n\n// TestQueueMultipleRobotsIndependentLimits tests that each robot has independent queue limit\nfunc TestQueueMultipleRobotsIndependentLimits(t *testing.T) {\n\tpq := pool.NewPriorityQueue(100)\n\n\t// Robot A: Queue=2\n\trobotA := createTestRobot(\"robot_A\", \"team_1\", 5, 2, 5)\n\t// Robot B: Queue=3\n\trobotB := createTestRobot(\"robot_B\", \"team_1\", 5, 3, 5)\n\n\t// Add items for Robot A\n\tfor i := 0; i < 5; i++ {\n\t\tpq.Enqueue(&pool.QueueItem{Robot: robotA, Trigger: types.TriggerClock})\n\t}\n\tassert.Equal(t, 2, pq.RobotQueuedCount(\"robot_A\"))\n\n\t// Add items for Robot B\n\tfor i := 0; i < 5; i++ {\n\t\tpq.Enqueue(&pool.QueueItem{Robot: robotB, Trigger: types.TriggerClock})\n\t}\n\tassert.Equal(t, 3, pq.RobotQueuedCount(\"robot_B\"))\n\n\t// Total in queue\n\tassert.Equal(t, 5, pq.Size())\n}\n\n// TestQueueRobotCountAfterDequeue tests robot count decrements after dequeue\nfunc TestQueueRobotCountAfterDequeue(t *testing.T) {\n\tpq := pool.NewPriorityQueue(100)\n\trobot := createTestRobot(\"robot_1\", \"team_1\", 5, 10, 5)\n\n\t// Add 3 items\n\tfor i := 0; i < 3; i++ {\n\t\tpq.Enqueue(&pool.QueueItem{Robot: robot, Trigger: types.TriggerClock})\n\t}\n\tassert.Equal(t, 3, pq.RobotQueuedCount(\"robot_1\"))\n\n\t// Dequeue 2\n\tpq.Dequeue()\n\tassert.Equal(t, 2, pq.RobotQueuedCount(\"robot_1\"))\n\tpq.Dequeue()\n\tassert.Equal(t, 1, pq.RobotQueuedCount(\"robot_1\"))\n\n\t// Dequeue last\n\tpq.Dequeue()\n\tassert.Equal(t, 0, pq.RobotQueuedCount(\"robot_1\"))\n}\n\n// TestQueueNilRobot tests handling of nil robot\nfunc TestQueueNilRobot(t *testing.T) {\n\tpq := pool.NewPriorityQueue(100)\n\n\t// Item with nil robot should still be enqueued\n\titem := &pool.QueueItem{\n\t\tRobot: nil,\n\t\tTrigger: types.TriggerClock,\n\t}\n\tok := pq.Enqueue(item)\n\tassert.True(t, ok)\n\tassert.Equal(t, 1, pq.Size())\n\n\t// Dequeue should work\n\tdequeued := pq.Dequeue()\n\tassert.NotNil(t, dequeued)\n\tassert.Nil(t, dequeued.Robot)\n}\n\n// TestQueueDefaultRobotQueueLimit tests default queue limit when Quota is nil\nfunc TestQueueDefaultRobotQueueLimit(t *testing.T) {\n\tpq := pool.NewPriorityQueue(100)\n\n\t// Robot without Config\n\trobot := &types.Robot{\n\t\tMemberID: \"robot_no_config\",\n\t\tTeamID: \"team_1\",\n\t}\n\n\t// Should use default queue limit (10)\n\tsuccessCount := 0\n\tfor i := 0; i < 15; i++ {\n\t\titem := &pool.QueueItem{Robot: robot, Trigger: types.TriggerClock}\n\t\tif pq.Enqueue(item) {\n\t\t\tsuccessCount++\n\t\t}\n\t}\n\n\tassert.Equal(t, 10, successCount) // default queue limit\n}\n\n// ==================== Priority Tests ====================\n\n// TestQueuePriorityByRobotPriority tests sorting by robot priority\nfunc TestQueuePriorityByRobotPriority(t *testing.T) {\n\tpq := pool.NewPriorityQueue(100)\n\n\t// Add robots with different priorities (low to high)\n\trobotLow := createTestRobot(\"robot_low\", \"team_1\", 5, 10, 1)\n\trobotMed := createTestRobot(\"robot_med\", \"team_1\", 5, 10, 5)\n\trobotHigh := createTestRobot(\"robot_high\", \"team_1\", 5, 10, 10)\n\n\t// Add in low-to-high order\n\tpq.Enqueue(&pool.QueueItem{Robot: robotLow, Trigger: types.TriggerClock})\n\tpq.Enqueue(&pool.QueueItem{Robot: robotMed, Trigger: types.TriggerClock})\n\tpq.Enqueue(&pool.QueueItem{Robot: robotHigh, Trigger: types.TriggerClock})\n\n\t// Dequeue should return high priority first\n\titem1 := pq.Dequeue()\n\tassert.Equal(t, \"robot_high\", item1.Robot.MemberID)\n\n\titem2 := pq.Dequeue()\n\tassert.Equal(t, \"robot_med\", item2.Robot.MemberID)\n\n\titem3 := pq.Dequeue()\n\tassert.Equal(t, \"robot_low\", item3.Robot.MemberID)\n}\n\n// TestQueuePriorityByTriggerType tests sorting by trigger type\nfunc TestQueuePriorityByTriggerType(t *testing.T) {\n\tpq := pool.NewPriorityQueue(100)\n\n\t// Same robot, different trigger types\n\trobot := createTestRobot(\"robot_1\", \"team_1\", 5, 10, 5)\n\n\t// Add in clock -> event -> human order\n\tpq.Enqueue(&pool.QueueItem{Robot: robot, Trigger: types.TriggerClock})\n\tpq.Enqueue(&pool.QueueItem{Robot: robot, Trigger: types.TriggerEvent})\n\tpq.Enqueue(&pool.QueueItem{Robot: robot, Trigger: types.TriggerHuman})\n\n\t// Dequeue should return human first (highest trigger priority)\n\titem1 := pq.Dequeue()\n\tassert.Equal(t, types.TriggerHuman, item1.Trigger)\n\n\titem2 := pq.Dequeue()\n\tassert.Equal(t, types.TriggerEvent, item2.Trigger)\n\n\titem3 := pq.Dequeue()\n\tassert.Equal(t, types.TriggerClock, item3.Trigger)\n}\n\n// TestQueuePriorityRobotOverTrigger tests that robot priority > trigger priority\nfunc TestQueuePriorityRobotOverTrigger(t *testing.T) {\n\tpq := pool.NewPriorityQueue(100)\n\n\t// Low priority robot with human trigger\n\trobotLow := createTestRobot(\"robot_low\", \"team_1\", 5, 10, 1)\n\t// High priority robot with clock trigger\n\trobotHigh := createTestRobot(\"robot_high\", \"team_1\", 5, 10, 10)\n\n\tpq.Enqueue(&pool.QueueItem{Robot: robotLow, Trigger: types.TriggerHuman})\n\tpq.Enqueue(&pool.QueueItem{Robot: robotHigh, Trigger: types.TriggerClock})\n\n\t// Robot priority (10*1000=10000) > trigger priority (1*1000+10*100=2000)\n\t// So high priority robot should come first even with lower trigger type\n\titem1 := pq.Dequeue()\n\tassert.Equal(t, \"robot_high\", item1.Robot.MemberID)\n\n\titem2 := pq.Dequeue()\n\tassert.Equal(t, \"robot_low\", item2.Robot.MemberID)\n}\n\n// TestQueuePriorityByEnqueueTime tests FIFO for same priority\nfunc TestQueuePriorityByEnqueueTime(t *testing.T) {\n\tpq := pool.NewPriorityQueue(100)\n\n\t// Same robot, same trigger type (same priority)\n\trobot := createTestRobot(\"robot_1\", \"team_1\", 5, 10, 5)\n\n\t// Add items with slight delay to ensure different EnqueueTime\n\tpq.Enqueue(&pool.QueueItem{Robot: robot, Trigger: types.TriggerClock, Data: \"first\"})\n\ttime.Sleep(1 * time.Millisecond)\n\tpq.Enqueue(&pool.QueueItem{Robot: robot, Trigger: types.TriggerClock, Data: \"second\"})\n\ttime.Sleep(1 * time.Millisecond)\n\tpq.Enqueue(&pool.QueueItem{Robot: robot, Trigger: types.TriggerClock, Data: \"third\"})\n\n\t// Should dequeue in FIFO order (earlier EnqueueTime first)\n\titem1 := pq.Dequeue()\n\tassert.Equal(t, \"first\", item1.Data)\n\n\titem2 := pq.Dequeue()\n\tassert.Equal(t, \"second\", item2.Data)\n\n\titem3 := pq.Dequeue()\n\tassert.Equal(t, \"third\", item3.Data)\n}\n\n// ==================== Concurrency Tests ====================\n\n// TestQueueConcurrentEnqueue tests concurrent enqueue operations\nfunc TestQueueConcurrentEnqueue(t *testing.T) {\n\tpq := pool.NewPriorityQueue(1000)\n\n\t// Concurrently add items from multiple goroutines\n\tdone := make(chan bool)\n\tfor i := 0; i < 10; i++ {\n\t\tgo func(id int) {\n\t\t\trobot := createTestRobot(\"robot_\"+string(rune('A'+id)), \"team_1\", 5, 100, 5)\n\t\t\tfor j := 0; j < 50; j++ {\n\t\t\t\tpq.Enqueue(&pool.QueueItem{Robot: robot, Trigger: types.TriggerClock})\n\t\t\t}\n\t\t\tdone <- true\n\t\t}(i)\n\t}\n\n\t// Wait for all goroutines\n\tfor i := 0; i < 10; i++ {\n\t\t<-done\n\t}\n\n\t// Should have 10 robots * 50 items = 500 items\n\tassert.Equal(t, 500, pq.Size())\n}\n\n// TestQueueConcurrentDequeue tests concurrent dequeue operations\nfunc TestQueueConcurrentDequeue(t *testing.T) {\n\tpq := pool.NewPriorityQueue(1000)\n\n\t// Pre-fill queue\n\tfor i := 0; i < 500; i++ {\n\t\trobot := createTestRobot(\"robot_\"+string(rune('A'+i%10)), \"team_1\", 5, 100, 5)\n\t\tpq.Enqueue(&pool.QueueItem{Robot: robot, Trigger: types.TriggerClock})\n\t}\n\n\t// Concurrently dequeue from multiple goroutines\n\tdequeued := make(chan *pool.QueueItem, 500)\n\tdone := make(chan bool)\n\n\tfor i := 0; i < 10; i++ {\n\t\tgo func() {\n\t\t\tfor {\n\t\t\t\titem := pq.Dequeue()\n\t\t\t\tif item == nil {\n\t\t\t\t\tbreak\n\t\t\t\t}\n\t\t\t\tdequeued <- item\n\t\t\t}\n\t\t\tdone <- true\n\t\t}()\n\t}\n\n\t// Wait for all goroutines\n\tfor i := 0; i < 10; i++ {\n\t\t<-done\n\t}\n\tclose(dequeued)\n\n\t// Count dequeued items\n\tcount := 0\n\tfor range dequeued {\n\t\tcount++\n\t}\n\n\tassert.Equal(t, 500, count)\n\tassert.Equal(t, 0, pq.Size())\n}\n\n// TestQueueConcurrentEnqueueDequeue tests concurrent enqueue and dequeue\nfunc TestQueueConcurrentEnqueueDequeue(t *testing.T) {\n\tpq := pool.NewPriorityQueue(100)\n\n\t// Run for a short time with concurrent operations\n\tdone := make(chan bool)\n\n\t// Enqueue goroutine\n\tgo func() {\n\t\tfor i := 0; i < 200; i++ {\n\t\t\trobot := createTestRobot(\"robot_\"+string(rune('A'+i%10)), \"team_1\", 5, 50, 5)\n\t\t\tpq.Enqueue(&pool.QueueItem{Robot: robot, Trigger: types.TriggerClock})\n\t\t\ttime.Sleep(1 * time.Millisecond)\n\t\t}\n\t\tdone <- true\n\t}()\n\n\t// Dequeue goroutine\n\tdequeueCount := 0\n\tgo func() {\n\t\tfor i := 0; i < 200; i++ {\n\t\t\tif pq.Dequeue() != nil {\n\t\t\t\tdequeueCount++\n\t\t\t}\n\t\t\ttime.Sleep(1 * time.Millisecond)\n\t\t}\n\t\tdone <- true\n\t}()\n\n\t// Wait for both\n\t<-done\n\t<-done\n\n\t// Should have processed some items (exact count depends on timing)\n\tassert.GreaterOrEqual(t, dequeueCount, 1)\n}\n\n// ==================== Edge Cases ====================\n\n// TestQueueIsFull tests IsFull method\nfunc TestQueueIsFull(t *testing.T) {\n\tt.Run(\"not full initially\", func(t *testing.T) {\n\t\tpq := pool.NewPriorityQueue(5)\n\t\tassert.False(t, pq.IsFull())\n\t})\n\n\tt.Run(\"full when at max\", func(t *testing.T) {\n\t\tpq := pool.NewPriorityQueue(3)\n\t\tfor i := 0; i < 3; i++ {\n\t\t\trobot := createTestRobot(\"robot_\"+string(rune('A'+i)), \"team_1\", 5, 10, 5)\n\t\t\tpq.Enqueue(&pool.QueueItem{Robot: robot, Trigger: types.TriggerClock})\n\t\t}\n\t\tassert.True(t, pq.IsFull())\n\t})\n\n\tt.Run(\"not full after dequeue\", func(t *testing.T) {\n\t\tpq := pool.NewPriorityQueue(3)\n\t\tfor i := 0; i < 3; i++ {\n\t\t\trobot := createTestRobot(\"robot_\"+string(rune('A'+i)), \"team_1\", 5, 10, 5)\n\t\t\tpq.Enqueue(&pool.QueueItem{Robot: robot, Trigger: types.TriggerClock})\n\t\t}\n\t\tpq.Dequeue()\n\t\tassert.False(t, pq.IsFull())\n\t})\n\n\tt.Run(\"never full when unlimited\", func(t *testing.T) {\n\t\tpq := pool.NewPriorityQueue(0)\n\t\tfor i := 0; i < 100; i++ {\n\t\t\trobot := createTestRobot(\"robot_\"+string(rune('A'+i%26)), \"team_1\", 5, 1000, 5)\n\t\t\tpq.Enqueue(&pool.QueueItem{Robot: robot, Trigger: types.TriggerClock})\n\t\t}\n\t\tassert.False(t, pq.IsFull())\n\t})\n}\n\n// TestQueueRobotQueuedCount tests RobotQueuedCount method\nfunc TestQueueRobotQueuedCount(t *testing.T) {\n\tpq := pool.NewPriorityQueue(100)\n\n\tt.Run(\"zero for unknown robot\", func(t *testing.T) {\n\t\tassert.Equal(t, 0, pq.RobotQueuedCount(\"unknown_robot\"))\n\t})\n\n\tt.Run(\"correct count for robot\", func(t *testing.T) {\n\t\trobot := createTestRobot(\"robot_1\", \"team_1\", 5, 10, 5)\n\t\tpq.Enqueue(&pool.QueueItem{Robot: robot, Trigger: types.TriggerClock})\n\t\tpq.Enqueue(&pool.QueueItem{Robot: robot, Trigger: types.TriggerClock})\n\t\tassert.Equal(t, 2, pq.RobotQueuedCount(\"robot_1\"))\n\t})\n\n\tt.Run(\"zero after all dequeued\", func(t *testing.T) {\n\t\trobot := createTestRobot(\"robot_2\", \"team_1\", 5, 10, 5)\n\t\tpq.Enqueue(&pool.QueueItem{Robot: robot, Trigger: types.TriggerClock})\n\t\tpq.Dequeue()\n\t\tpq.Dequeue() // dequeue robot_1's items too\n\t\tpq.Dequeue()\n\t\tassert.Equal(t, 0, pq.RobotQueuedCount(\"robot_2\"))\n\t})\n}\n\n// TestQueueEnqueueSetsEnqueueTime tests that EnqueueTime is set on enqueue\nfunc TestQueueEnqueueSetsEnqueueTime(t *testing.T) {\n\tpq := pool.NewPriorityQueue(100)\n\trobot := createTestRobot(\"robot_1\", \"team_1\", 5, 10, 5)\n\n\tbefore := time.Now()\n\tpq.Enqueue(&pool.QueueItem{Robot: robot, Trigger: types.TriggerClock})\n\tafter := time.Now()\n\n\titem := pq.Dequeue()\n\tassert.True(t, item.EnqueueTime.After(before) || item.EnqueueTime.Equal(before))\n\tassert.True(t, item.EnqueueTime.Before(after) || item.EnqueueTime.Equal(after))\n}\n" }, { "path": "agent/robot/pool/worker.go", "content": "package pool\n\nimport (\n\t\"sync\"\n\t\"time\"\n\n\t\"github.com/yaoapp/yao/agent/robot/logger\"\n\t\"github.com/yaoapp/yao/agent/robot/types\"\n)\n\nvar log = logger.New(\"pool\")\n\n// Worker represents a worker goroutine that processes jobs\ntype Worker struct {\n\tid int\n\tpool *Pool\n\tstopChan chan struct{}\n\twg *sync.WaitGroup\n}\n\n// newWorker creates a new worker\nfunc newWorker(id int, pool *Pool, wg *sync.WaitGroup) *Worker {\n\treturn &Worker{\n\t\tid: id,\n\t\tpool: pool,\n\t\tstopChan: make(chan struct{}),\n\t\twg: wg,\n\t}\n}\n\n// start starts the worker goroutine\nfunc (w *Worker) start() {\n\tw.wg.Add(1)\n\tgo w.run()\n}\n\n// stop signals the worker to stop\nfunc (w *Worker) stop() {\n\tclose(w.stopChan)\n}\n\n// run is the main worker loop\nfunc (w *Worker) run() {\n\tdefer w.wg.Done()\n\n\tticker := time.NewTicker(100 * time.Millisecond) // poll queue every 100ms\n\tdefer ticker.Stop()\n\n\tfor {\n\t\tselect {\n\t\tcase <-w.stopChan:\n\t\t\treturn\n\n\t\tcase <-ticker.C:\n\t\t\t// Try to get a job from the queue\n\t\t\titem := w.pool.queue.Dequeue()\n\t\t\tif item == nil {\n\t\t\t\tcontinue // queue empty, wait for next tick\n\t\t\t}\n\n\t\t\t// Execute the job\n\t\t\tw.execute(item)\n\t\t}\n\t}\n}\n\n// execute processes a single queue item\nfunc (w *Worker) execute(item *QueueItem) {\n\t// Pre-check if robot can run (non-atomic, just for early rejection)\n\t// The actual atomic check happens inside Executor.Execute() via TryAcquireSlot()\n\tif !item.Robot.CanRun() {\n\t\t// Robot likely at quota, re-enqueue for later\n\t\tw.requeue(item, \"quota pre-check failed\")\n\t\treturn\n\t}\n\n\t// Mark as running (only when actually executing)\n\tw.pool.incrementRunning()\n\tdefer w.pool.decrementRunning()\n\n\t// Get executor based on mode (uses factory if available, otherwise default)\n\texec := w.pool.GetExecutor(item.ExecutorMode)\n\n\t// Execute via Executor interface with pre-generated ID and control\n\t// Note: Executor.ExecuteWithControl() does atomic quota check via TryAcquireSlot()\n\t// The control parameter allows executor to check pause state during execution\n\texecution, err := exec.ExecuteWithControl(item.Ctx, item.Robot, item.Trigger, item.Data, item.ExecID, item.Control)\n\n\tif err != nil {\n\t\t// Check if it's a quota error (race condition - another worker got the slot)\n\t\tif err == types.ErrQuotaExceeded {\n\t\t\tw.requeue(item, \"quota exceeded (race)\")\n\t\t\treturn\n\t\t}\n\n\t\t// Suspended execution: state is persisted, worker slot released gracefully.\n\t\t// Do NOT call onComplete — the execution stays in robot.Executions and execController\n\t\t// so that Resume can find it later (§16.1).\n\t\tif err == types.ErrExecutionSuspended {\n\t\t\tif execution != nil {\n\t\t\t\tlog.Info(\"Worker %d: Execution %s suspended for robot %s (waiting for input)\",\n\t\t\t\t\tw.id, execution.ID, item.Robot.MemberID)\n\t\t\t}\n\t\t\treturn\n\t\t}\n\n\t\tlog.Error(\"Worker %d: Execution failed for robot %s: %v\",\n\t\t\tw.id, item.Robot.MemberID, err)\n\t\t// Notify completion callback with appropriate status\n\t\tif w.pool.onComplete != nil {\n\t\t\tstatus := types.ExecFailed\n\t\t\tif err == types.ErrExecutionCancelled {\n\t\t\t\tstatus = types.ExecCancelled\n\t\t\t}\n\t\t\tw.pool.onComplete(item.ExecID, item.Robot.MemberID, status)\n\t\t}\n\t\treturn\n\t}\n\n\tif execution != nil {\n\t\tlog.Info(\"Worker %d: Execution %s completed for robot %s (status: %s)\",\n\t\t\tw.id, execution.ID, item.Robot.MemberID, execution.Status)\n\t\t// Notify completion callback\n\t\tif w.pool.onComplete != nil {\n\t\t\tw.pool.onComplete(execution.ID, item.Robot.MemberID, execution.Status)\n\t\t}\n\t}\n}\n\n// requeue attempts to put the item back in the queue\nfunc (w *Worker) requeue(item *QueueItem, reason string) {\n\t// Queue length is our system load threshold:\n\t// - If queue has space: task waits for robot quota\n\t// - If queue is full: system is overloaded, drop task\n\tif !w.pool.queue.Enqueue(item) {\n\t\tlog.Warn(\"Worker %d: Task for robot %s dropped (queue full, %s)\",\n\t\t\tw.id, item.Robot.MemberID, reason)\n\t}\n}\n" }, { "path": "agent/robot/pool/worker_test.go", "content": "package pool_test\n\nimport (\n\t\"sync\"\n\t\"sync/atomic\"\n\t\"testing\"\n\t\"time\"\n\n\t\"github.com/stretchr/testify/assert\"\n\t\"github.com/yaoapp/yao/agent/robot/executor\"\n\t\"github.com/yaoapp/yao/agent/robot/pool\"\n\t\"github.com/yaoapp/yao/agent/robot/types\"\n)\n\n// ==================== Worker Basic Tests ====================\n\n// TestWorkerExecutesJob tests that worker executes a job from queue\nfunc TestWorkerExecutesJob(t *testing.T) {\n\texec := executor.NewDryRunWithDelay(10 * time.Millisecond)\n\tp := pool.NewWithConfig(&pool.Config{\n\t\tWorkerSize: 1,\n\t\tQueueSize: 10,\n\t})\n\tp.SetExecutor(exec)\n\tp.Start()\n\tdefer p.Stop()\n\n\tctx := createTestContext()\n\trobot := createTestRobot(\"robot_1\", \"team_1\", 5, 10, 5)\n\n\t// Submit job\n\tp.Submit(ctx, robot, types.TriggerClock, nil)\n\n\t// Wait for execution\n\ttime.Sleep(200 * time.Millisecond)\n\n\tassert.Equal(t, 1, exec.ExecCount())\n}\n\n// TestWorkerMultipleJobs tests worker processes multiple jobs sequentially\nfunc TestWorkerMultipleJobs(t *testing.T) {\n\texec := executor.NewDryRunWithDelay(20 * time.Millisecond)\n\tp := pool.NewWithConfig(&pool.Config{\n\t\tWorkerSize: 1, // single worker\n\t\tQueueSize: 10,\n\t})\n\tp.SetExecutor(exec)\n\tp.Start()\n\tdefer p.Stop()\n\n\tctx := createTestContext()\n\trobot := createTestRobot(\"robot_1\", \"team_1\", 10, 10, 5)\n\n\t// Submit 3 jobs\n\tfor i := 0; i < 3; i++ {\n\t\tp.Submit(ctx, robot, types.TriggerClock, nil)\n\t}\n\n\t// Wait for all executions (worker polls every 100ms, each job takes 20ms)\n\t// Use Eventually for CI timing variations\n\tassert.Eventually(t, func() bool {\n\t\treturn exec.ExecCount() >= 3\n\t}, 1*time.Second, 50*time.Millisecond, \"All 3 jobs should complete\")\n}\n\n// ==================== Worker Quota Check Tests ====================\n\n// TestWorkerRespectsRobotQuota tests worker re-enqueues when robot quota is full\nfunc TestWorkerRespectsRobotQuota(t *testing.T) {\n\t// This test verifies that all jobs eventually complete even when robot quota limits concurrency\n\texec := executor.NewDryRunWithDelay(100 * time.Millisecond)\n\n\tp := pool.NewWithConfig(&pool.Config{\n\t\tWorkerSize: 5, // multiple workers\n\t\tQueueSize: 20,\n\t})\n\tp.SetExecutor(exec)\n\tp.Start()\n\tdefer p.Stop()\n\n\tctx := createTestContext()\n\t// Robot can only run 2 at a time\n\trobot := createTestRobot(\"robot_limited\", \"team_1\", 2, 10, 5)\n\n\t// Submit 5 jobs for same robot\n\tfor i := 0; i < 5; i++ {\n\t\tp.Submit(ctx, robot, types.TriggerClock, nil)\n\t}\n\n\t// Wait for all to complete\n\t// With Quota.Max=2, jobs execute in batches: 2+2+1 = 3 batches\n\t// Each batch: 100ms exec + 100ms poll = ~200ms, total ~600ms, add buffer\n\ttime.Sleep(1000 * time.Millisecond)\n\n\t// All should eventually execute\n\tassert.GreaterOrEqual(t, exec.ExecCount(), 5, \"All jobs should eventually execute\")\n}\n\n// TestWorkerReenqueueOnQuotaFull tests that jobs are re-enqueued when quota is full\nfunc TestWorkerReenqueueOnQuotaFull(t *testing.T) {\n\texec := executor.NewDryRunWithDelay(100 * time.Millisecond)\n\tp := pool.NewWithConfig(&pool.Config{\n\t\tWorkerSize: 3,\n\t\tQueueSize: 100,\n\t})\n\tp.SetExecutor(exec)\n\tp.Start()\n\tdefer p.Stop()\n\n\tctx := createTestContext()\n\t// Robot can only run 1 at a time, but large queue\n\trobot := createTestRobot(\"robot_1\", \"team_1\", 1, 50, 5)\n\n\t// Submit 5 jobs\n\tfor i := 0; i < 5; i++ {\n\t\tp.Submit(ctx, robot, types.TriggerClock, nil)\n\t}\n\n\t// Wait for all to complete\n\t// With Quota.Max=1, jobs execute sequentially: 5 * (100ms exec + 100ms poll) = ~1000ms\n\t// Use Eventually for CI timing variations\n\tassert.Eventually(t, func() bool {\n\t\treturn exec.ExecCount() >= 5\n\t}, 2*time.Second, 100*time.Millisecond, \"All 5 jobs should complete\")\n}\n\n// ==================== Worker Concurrency Tests ====================\n\n// TestWorkersConcurrentExecution tests multiple workers execute concurrently\nfunc TestWorkersConcurrentExecution(t *testing.T) {\n\t// Track max concurrent executions\n\tvar maxConcurrent int32\n\tvar currentConcurrent int32\n\n\texec := executor.NewDryRunWithCallbacks(100*time.Millisecond, func() {\n\t\tcurrent := atomic.AddInt32(¤tConcurrent, 1)\n\t\t// Update max if current is higher\n\t\tfor {\n\t\t\tmax := atomic.LoadInt32(&maxConcurrent)\n\t\t\tif current <= max || atomic.CompareAndSwapInt32(&maxConcurrent, max, current) {\n\t\t\t\tbreak\n\t\t\t}\n\t\t}\n\t}, func() {\n\t\tatomic.AddInt32(¤tConcurrent, -1)\n\t})\n\n\tp := pool.NewWithConfig(&pool.Config{\n\t\tWorkerSize: 5, // 5 workers\n\t\tQueueSize: 100,\n\t})\n\tp.SetExecutor(exec)\n\tp.Start()\n\tdefer p.Stop()\n\n\tctx := createTestContext()\n\n\t// Submit 10 jobs for different robots\n\tfor i := 0; i < 10; i++ {\n\t\trobot := createTestRobot(\"robot_\"+string(rune('A'+i)), \"team_1\", 5, 10, 5)\n\t\tp.Submit(ctx, robot, types.TriggerClock, nil)\n\t}\n\n\t// Wait for execution\n\ttime.Sleep(400 * time.Millisecond)\n\n\t// Should have had concurrent execution (max > 1)\n\tassert.GreaterOrEqual(t, atomic.LoadInt32(&maxConcurrent), int32(2), \"Should have concurrent execution\")\n}\n\n// TestWorkersDoNotExceedPoolSize tests workers don't exceed pool size\nfunc TestWorkersDoNotExceedPoolSize(t *testing.T) {\n\tvar maxConcurrent int32\n\tvar currentConcurrent int32\n\tvar mu sync.Mutex\n\n\texec := executor.NewDryRunWithCallbacks(50*time.Millisecond, func() {\n\t\tmu.Lock()\n\t\tcurrentConcurrent++\n\t\tif currentConcurrent > maxConcurrent {\n\t\t\tmaxConcurrent = currentConcurrent\n\t\t}\n\t\tmu.Unlock()\n\t}, func() {\n\t\tmu.Lock()\n\t\tcurrentConcurrent--\n\t\tmu.Unlock()\n\t})\n\n\tp := pool.NewWithConfig(&pool.Config{\n\t\tWorkerSize: 3, // only 3 workers\n\t\tQueueSize: 100,\n\t})\n\tp.SetExecutor(exec)\n\tp.Start()\n\tdefer p.Stop()\n\n\tctx := createTestContext()\n\n\t// Submit 20 jobs\n\tfor i := 0; i < 20; i++ {\n\t\trobot := createTestRobot(\"robot_\"+string(rune('A'+i)), \"team_1\", 5, 10, 5)\n\t\tp.Submit(ctx, robot, types.TriggerClock, nil)\n\t}\n\n\t// Wait for all to complete\n\ttime.Sleep(500 * time.Millisecond)\n\n\t// Max concurrent should not exceed worker size\n\tassert.LessOrEqual(t, maxConcurrent, int32(3), \"Should not exceed worker size\")\n}\n\n// ==================== Worker Stop Tests ====================\n\n// TestWorkerStopsGracefully tests worker stops when signaled\nfunc TestWorkerStopsGracefully(t *testing.T) {\n\texec := executor.NewDryRunWithDelay(50 * time.Millisecond)\n\tp := pool.NewWithConfig(&pool.Config{\n\t\tWorkerSize: 2,\n\t\tQueueSize: 10,\n\t})\n\tp.SetExecutor(exec)\n\tp.Start()\n\n\tctx := createTestContext()\n\trobot := createTestRobot(\"robot_1\", \"team_1\", 5, 10, 5)\n\n\t// Submit jobs\n\tp.Submit(ctx, robot, types.TriggerClock, nil)\n\tp.Submit(ctx, robot, types.TriggerClock, nil)\n\n\t// Wait for jobs to start\n\ttime.Sleep(150 * time.Millisecond)\n\n\t// Stop pool\n\terr := p.Stop()\n\tassert.NoError(t, err)\n\n\t// Pool should be stopped\n\tassert.False(t, p.IsStarted())\n}\n\n// TestWorkerCompletesCurrentJobOnStop tests worker completes current job before stopping\nfunc TestWorkerCompletesCurrentJobOnStop(t *testing.T) {\n\texec := executor.NewDryRunWithDelay(100 * time.Millisecond)\n\tp := pool.NewWithConfig(&pool.Config{\n\t\tWorkerSize: 1,\n\t\tQueueSize: 10,\n\t})\n\tp.SetExecutor(exec)\n\tp.Start()\n\n\tctx := createTestContext()\n\trobot := createTestRobot(\"robot_1\", \"team_1\", 5, 10, 5)\n\n\t// Submit job\n\tp.Submit(ctx, robot, types.TriggerClock, nil)\n\n\t// Wait for job to start\n\ttime.Sleep(150 * time.Millisecond)\n\n\t// Stop pool - should wait for current job\n\tp.Stop()\n\n\t// Job should have completed\n\tassert.GreaterOrEqual(t, exec.ExecCount(), 1)\n}\n\n// ==================== Worker Error Handling Tests ====================\n\n// TestWorkerHandlesExecutorError tests worker continues after executor error\nfunc TestWorkerHandlesExecutorError(t *testing.T) {\n\texec := executor.NewDryRunWithDelay(10 * time.Millisecond)\n\tp := pool.NewWithConfig(&pool.Config{\n\t\tWorkerSize: 1,\n\t\tQueueSize: 10,\n\t})\n\tp.SetExecutor(exec)\n\tp.Start()\n\tdefer p.Stop()\n\n\tctx := createTestContext()\n\trobot := createTestRobot(\"robot_1\", \"team_1\", 5, 10, 5)\n\n\t// Submit job that will fail (using special data)\n\tp.Submit(ctx, robot, types.TriggerClock, \"simulate_failure\")\n\n\t// Submit another job that should succeed\n\tp.Submit(ctx, robot, types.TriggerClock, nil)\n\n\t// Wait for execution\n\ttime.Sleep(300 * time.Millisecond)\n\n\t// Both should have been attempted\n\tassert.GreaterOrEqual(t, exec.ExecCount(), 2)\n}\n\n// ==================== Worker Running Counter Tests ====================\n\n// TestWorkerRunningCounterAccurate tests running counter is accurate\nfunc TestWorkerRunningCounterAccurate(t *testing.T) {\n\texec := executor.NewDryRunWithDelay(100 * time.Millisecond)\n\tp := pool.NewWithConfig(&pool.Config{\n\t\tWorkerSize: 3,\n\t\tQueueSize: 10,\n\t})\n\tp.SetExecutor(exec)\n\tp.Start()\n\tdefer p.Stop()\n\n\tctx := createTestContext()\n\n\t// Submit jobs for different robots\n\tfor i := 0; i < 3; i++ {\n\t\trobot := createTestRobot(\"robot_\"+string(rune('A'+i)), \"team_1\", 5, 10, 5)\n\t\tp.Submit(ctx, robot, types.TriggerClock, nil)\n\t}\n\n\t// Wait for jobs to start\n\ttime.Sleep(200 * time.Millisecond)\n\n\t// Running should be > 0 while jobs are executing\n\t// Note: On fast CI, jobs may already be done, so we just verify it doesn't panic\n\n\t// Wait for completion and verify running counter returns to 0\n\tassert.Eventually(t, func() bool {\n\t\treturn p.Running() == 0\n\t}, 1*time.Second, 50*time.Millisecond, \"Running should be 0 after all jobs complete\")\n}\n\n// TestWorkerRunningCounterDecrementsOnError tests running counter decrements on error\nfunc TestWorkerRunningCounterDecrementsOnError(t *testing.T) {\n\texec := executor.NewDryRunWithDelay(10 * time.Millisecond)\n\tp := pool.NewWithConfig(&pool.Config{\n\t\tWorkerSize: 1,\n\t\tQueueSize: 10,\n\t})\n\tp.SetExecutor(exec)\n\tp.Start()\n\tdefer p.Stop()\n\n\tctx := createTestContext()\n\trobot := createTestRobot(\"robot_1\", \"team_1\", 5, 10, 5)\n\n\t// Submit failing job\n\tp.Submit(ctx, robot, types.TriggerClock, \"simulate_failure\")\n\n\t// Wait for execution\n\ttime.Sleep(200 * time.Millisecond)\n\n\t// Running should be 0 (decremented even on error)\n\tassert.Equal(t, 0, p.Running())\n}\n\n// ==================== Worker with Different Trigger Types ====================\n\n// TestWorkerProcessesDifferentTriggers tests worker handles all trigger types\nfunc TestWorkerProcessesDifferentTriggers(t *testing.T) {\n\texec := executor.NewDryRunWithDelay(10 * time.Millisecond)\n\tp := pool.NewWithConfig(&pool.Config{\n\t\tWorkerSize: 1,\n\t\tQueueSize: 10,\n\t})\n\tp.SetExecutor(exec)\n\tp.Start()\n\tdefer p.Stop()\n\n\tctx := createTestContext()\n\trobot := createTestRobot(\"robot_1\", \"team_1\", 5, 10, 5)\n\n\t// Submit different trigger types\n\tp.Submit(ctx, robot, types.TriggerClock, nil)\n\tp.Submit(ctx, robot, types.TriggerHuman, nil)\n\tp.Submit(ctx, robot, types.TriggerEvent, nil)\n\n\t// Wait for execution (worker polls every 100ms, each job takes 10ms)\n\t// Use Eventually for CI timing variations\n\tassert.Eventually(t, func() bool {\n\t\treturn exec.ExecCount() >= 3\n\t}, 1*time.Second, 50*time.Millisecond, \"All 3 trigger types should execute\")\n}\n\n// ==================== Worker Polling Behavior Tests ====================\n\n// TestWorkerPollsQueuePeriodically tests worker polls queue at regular intervals\nfunc TestWorkerPollsQueuePeriodically(t *testing.T) {\n\texec := executor.NewDryRunWithDelay(10 * time.Millisecond)\n\tp := pool.NewWithConfig(&pool.Config{\n\t\tWorkerSize: 1,\n\t\tQueueSize: 10,\n\t})\n\tp.SetExecutor(exec)\n\tp.Start()\n\tdefer p.Stop()\n\n\tctx := createTestContext()\n\trobot := createTestRobot(\"robot_1\", \"team_1\", 5, 10, 5)\n\n\t// Submit job after pool started\n\ttime.Sleep(50 * time.Millisecond)\n\tp.Submit(ctx, robot, types.TriggerClock, nil)\n\n\t// Worker should pick up job within poll interval (100ms)\n\ttime.Sleep(200 * time.Millisecond)\n\n\tassert.Equal(t, 1, exec.ExecCount())\n}\n\n// TestWorkerContinuesAfterEmptyQueue tests worker continues polling after empty queue\nfunc TestWorkerContinuesAfterEmptyQueue(t *testing.T) {\n\texec := executor.NewDryRunWithDelay(10 * time.Millisecond)\n\tp := pool.NewWithConfig(&pool.Config{\n\t\tWorkerSize: 1,\n\t\tQueueSize: 10,\n\t})\n\tp.SetExecutor(exec)\n\tp.Start()\n\tdefer p.Stop()\n\n\tctx := createTestContext()\n\trobot := createTestRobot(\"robot_1\", \"team_1\", 5, 10, 5)\n\n\t// Wait with empty queue\n\ttime.Sleep(200 * time.Millisecond)\n\n\t// Submit job\n\tp.Submit(ctx, robot, types.TriggerClock, nil)\n\n\t// Worker should still be running and pick up job\n\ttime.Sleep(200 * time.Millisecond)\n\n\tassert.Equal(t, 1, exec.ExecCount())\n}\n" }, { "path": "agent/robot/process.go", "content": "package robot\n\nimport (\n\t\"context\"\n\n\t\"github.com/yaoapp/gou/process\"\n\t\"github.com/yaoapp/kun/exception\"\n\t\"github.com/yaoapp/yao/agent/assistant\"\n\t\"github.com/yaoapp/yao/agent/robot/api\"\n\t\"github.com/yaoapp/yao/agent/robot/types\"\n)\n\nfunc init() {\n\tprocess.RegisterGroup(\"robot\", map[string]process.Handler{\n\t\t\"get\": processGet,\n\t\t\"list\": processList,\n\t\t\"status\": processStatus,\n\t\t\"executions\": processExecutions,\n\t\t\"execution\": processExecution,\n\t\t\"updateChatTitle\": processUpdateChatTitle,\n\t})\n}\n\n// processGet handles robot.Get(memberID).\n// args[0]: memberID string\nfunc processGet(p *process.Process) interface{} {\n\tp.ValidateArgNums(1)\n\tmemberID := p.ArgsString(0)\n\tctx := types.NewContext(context.Background(), nil)\n\tresult, err := api.GetRobotResponse(ctx, memberID)\n\tif err != nil {\n\t\texception.New(err.Error(), 500).Throw()\n\t}\n\treturn result\n}\n\n// processList handles robot.List(filter?).\n// args[0]: optional filter map with page, pagesize, status, search (keywords)\nfunc processList(p *process.Process) interface{} {\n\tp.ValidateArgNums(0)\n\tctx := types.NewContext(context.Background(), nil)\n\tfilter := &api.ListQuery{}\n\tif p.NumOfArgs() > 0 {\n\t\traw := p.ArgsMap(0)\n\t\tif v, ok := raw[\"page\"]; ok {\n\t\t\tfilter.Page = toInt(v)\n\t\t}\n\t\tif v, ok := raw[\"pagesize\"]; ok {\n\t\t\tfilter.PageSize = toInt(v)\n\t\t}\n\t\tif v, ok := raw[\"status\"]; ok {\n\t\t\tfilter.Status = types.RobotStatus(toString(v))\n\t\t}\n\t\tif v, ok := raw[\"search\"]; ok {\n\t\t\tfilter.Keywords = toString(v)\n\t\t}\n\t\tif v, ok := raw[\"team_id\"]; ok {\n\t\t\tfilter.TeamID = toString(v)\n\t\t}\n\t}\n\tresult, err := api.ListRobots(ctx, filter)\n\tif err != nil {\n\t\texception.New(err.Error(), 500).Throw()\n\t}\n\treturn result\n}\n\n// processStatus handles robot.Status(memberID).\n// args[0]: memberID string\nfunc processStatus(p *process.Process) interface{} {\n\tp.ValidateArgNums(1)\n\tmemberID := p.ArgsString(0)\n\tctx := types.NewContext(context.Background(), nil)\n\tresult, err := api.GetRobotStatus(ctx, memberID)\n\tif err != nil {\n\t\texception.New(err.Error(), 500).Throw()\n\t}\n\treturn result\n}\n\n// processExecutions handles robot.Executions(memberID, filter?).\n// args[0]: memberID string; args[1]: optional filter map\nfunc processExecutions(p *process.Process) interface{} {\n\tp.ValidateArgNums(1)\n\tmemberID := p.ArgsString(0)\n\tctx := types.NewContext(context.Background(), nil)\n\tfilter := &api.ExecutionQuery{}\n\tif p.NumOfArgs() > 1 {\n\t\traw := p.ArgsMap(1)\n\t\tif v, ok := raw[\"page\"]; ok {\n\t\t\tfilter.Page = toInt(v)\n\t\t}\n\t\tif v, ok := raw[\"pagesize\"]; ok {\n\t\t\tfilter.PageSize = toInt(v)\n\t\t}\n\t\tif v, ok := raw[\"status\"]; ok {\n\t\t\tfilter.Status = types.ExecStatus(toString(v))\n\t\t}\n\t\tif v, ok := raw[\"trigger\"]; ok {\n\t\t\tfilter.Trigger = types.TriggerType(toString(v))\n\t\t}\n\t}\n\tresult, err := api.ListExecutions(ctx, memberID, filter)\n\tif err != nil {\n\t\texception.New(err.Error(), 500).Throw()\n\t}\n\treturn result\n}\n\n// processExecution handles robot.Execution(memberID, executionID).\n// args[0]: memberID string; args[1]: executionID string\nfunc processExecution(p *process.Process) interface{} {\n\tp.ValidateArgNums(2)\n\tmemberID := p.ArgsString(0)\n\texecutionID := p.ArgsString(1)\n\tctx := types.NewContext(context.Background(), nil)\n\tresult, err := api.GetExecutionStatus(ctx, executionID)\n\tif err != nil {\n\t\texception.New(err.Error(), 500).Throw()\n\t}\n\t_ = memberID // reserved for future permission scoping\n\treturn result\n}\n\n// processUpdateChatTitle handles robot.UpdateChatTitle(chatID, title).\n// args[0]: chatID string; args[1]: title string\nfunc processUpdateChatTitle(p *process.Process) interface{} {\n\tp.ValidateArgNums(2)\n\tchatID := p.ArgsString(0)\n\ttitle := p.ArgsString(1)\n\n\tchatStore := assistant.GetChatStore()\n\tif chatStore == nil {\n\t\texception.New(\"chat store not available\", 500).Throw()\n\t}\n\n\tif err := chatStore.UpdateChat(chatID, map[string]interface{}{\"title\": title}); err != nil {\n\t\texception.New(err.Error(), 500).Throw()\n\t}\n\treturn nil\n}\n\nfunc toInt(v interface{}) int {\n\tswitch n := v.(type) {\n\tcase int:\n\t\treturn n\n\tcase int64:\n\t\treturn int(n)\n\tcase float64:\n\t\treturn int(n)\n\tdefault:\n\t\treturn 0\n\t}\n}\n\nfunc toString(v interface{}) string {\n\tif s, ok := v.(string); ok {\n\t\treturn s\n\t}\n\treturn \"\"\n}\n" }, { "path": "agent/robot/process_test.go", "content": "package robot_test\n\nimport (\n\t\"context\"\n\t\"fmt\"\n\t\"testing\"\n\t\"time\"\n\n\t\"github.com/google/uuid\"\n\t\"github.com/stretchr/testify/assert\"\n\t\"github.com/stretchr/testify/require\"\n\t\"github.com/yaoapp/gou/process\"\n\t\"github.com/yaoapp/yao/agent/assistant\"\n\tstoretypes \"github.com/yaoapp/yao/agent/store/types\"\n\t\"github.com/yaoapp/yao/agent/testutils\"\n\n\t// Register robot process handlers via init()\n\t_ \"github.com/yaoapp/yao/agent/robot\"\n)\n\n// ============================================================================\n// robot.UpdateChatTitle\n// ============================================================================\n\nfunc TestProcessUpdateChatTitle(t *testing.T) {\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\tchatStore := assistant.GetChatStore()\n\tif chatStore == nil {\n\t\tt.Skip(\"Chat store not configured, skipping UpdateChatTitle tests\")\n\t}\n\n\tt.Run(\"UpdatesTitle\", func(t *testing.T) {\n\t\tchatID := fmt.Sprintf(\"robot_test_proc_%s\", uuid.New().String()[:8])\n\n\t\t// Create a chat record first\n\t\terr := chatStore.CreateChat(&storetypes.Chat{\n\t\t\tChatID: chatID,\n\t\t\tAssistantID: \"robot.host\",\n\t\t\tStatus: \"active\",\n\t\t\tShare: \"private\",\n\t\t\tCreatedAt: time.Now(),\n\t\t\tUpdatedAt: time.Now(),\n\t\t})\n\t\trequire.NoError(t, err)\n\t\tdefer chatStore.DeleteChat(chatID)\n\n\t\ttitle := \"Create a mecha image, sci-fi style\"\n\t\tp := process.New(\"robot.UpdateChatTitle\", chatID, title)\n\t\t_, err = p.Exec()\n\t\trequire.NoError(t, err)\n\n\t\tchat, err := chatStore.GetChat(chatID)\n\t\trequire.NoError(t, err)\n\t\tassert.Equal(t, title, chat.Title, \"Title should be updated to the confirmed goals\")\n\t\tt.Logf(\"✓ robot.UpdateChatTitle: chat_id=%s, title=%q\", chatID, chat.Title)\n\t})\n\n\tt.Run(\"UpdatesLongGoalsTitle\", func(t *testing.T) {\n\t\tchatID := fmt.Sprintf(\"robot_test_long_%s\", uuid.New().String()[:8])\n\n\t\terr := chatStore.CreateChat(&storetypes.Chat{\n\t\t\tChatID: chatID,\n\t\t\tAssistantID: \"robot.host\",\n\t\t\tStatus: \"active\",\n\t\t\tShare: \"private\",\n\t\t\tCreatedAt: time.Now(),\n\t\t\tUpdatedAt: time.Now(),\n\t\t})\n\t\trequire.NoError(t, err)\n\t\tdefer chatStore.DeleteChat(chatID)\n\n\t\t// Long goals string — title field should accommodate it\n\t\ttitle := \"请帮我制作一张充满未来感的机甲图片，风格参考《攻壳机动队》，以赛博朋克城市为背景，色调偏冷，蓝紫配色\"\n\t\tp := process.New(\"robot.UpdateChatTitle\", chatID, title)\n\t\t_, err = p.Exec()\n\t\trequire.NoError(t, err)\n\n\t\tchat, err := chatStore.GetChat(chatID)\n\t\trequire.NoError(t, err)\n\t\tassert.Equal(t, title, chat.Title)\n\t\tt.Logf(\"✓ Long goals title persisted: %d chars\", len(title))\n\t})\n\n\tt.Run(\"ErrorOnNonExistentChat\", func(t *testing.T) {\n\t\tp := process.New(\"robot.UpdateChatTitle\", \"non_existent_chat_id\", \"some title\")\n\t\t_, err := p.Exec()\n\t\tassert.Error(t, err, \"Should error when chat does not exist\")\n\t\tt.Logf(\"✓ Non-existent chat correctly returns error\")\n\t})\n}\n\n// ============================================================================\n// robot.Get\n// ============================================================================\n\nfunc TestProcessGet(t *testing.T) {\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\tt.Run(\"ErrorOnNotFound\", func(t *testing.T) {\n\t\tp := process.New(\"robot.Get\", \"non_existent_robot_member_id\")\n\t\t_, err := p.Exec()\n\t\tassert.Error(t, err, \"Should error for non-existent robot\")\n\t\tt.Logf(\"✓ robot.Get returns error for non-existent robot\")\n\t})\n}\n\n// ============================================================================\n// robot.List\n// ============================================================================\n\nfunc TestProcessList(t *testing.T) {\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\tt.Run(\"ReturnsListWithNoFilter\", func(t *testing.T) {\n\t\tp := process.New(\"robot.List\")\n\t\tresult, err := p.Exec()\n\t\trequire.NoError(t, err)\n\t\t// Result is a paginated list — just assert it's not nil\n\t\tassert.NotNil(t, result)\n\t\tt.Logf(\"✓ robot.List returned: %T\", result)\n\t})\n\n\tt.Run(\"ReturnsListWithPageFilter\", func(t *testing.T) {\n\t\tp := process.New(\"robot.List\", map[string]interface{}{\n\t\t\t\"page\": 1,\n\t\t\t\"pagesize\": 5,\n\t\t})\n\t\tresult, err := p.Exec()\n\t\trequire.NoError(t, err)\n\t\tassert.NotNil(t, result)\n\t\tt.Logf(\"✓ robot.List with page filter returned: %T\", result)\n\t})\n}\n\n// ============================================================================\n// robot.Status\n// ============================================================================\n\nfunc TestProcessStatus(t *testing.T) {\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\tt.Run(\"ErrorOnNotFound\", func(t *testing.T) {\n\t\tp := process.New(\"robot.Status\", \"non_existent_robot_member_id\")\n\t\t_, err := p.Exec()\n\t\tassert.Error(t, err, \"Should error for non-existent robot\")\n\t\tt.Logf(\"✓ robot.Status returns error for non-existent robot\")\n\t})\n}\n\n// ============================================================================\n// robot.Executions\n// ============================================================================\n\nfunc TestProcessExecutions(t *testing.T) {\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\tt.Run(\"ReturnsEmptyForUnknownRobot\", func(t *testing.T) {\n\t\tmemberID := fmt.Sprintf(\"proc_exec_test_%d\", time.Now().UnixNano())\n\t\tp := process.New(\"robot.Executions\", memberID)\n\t\tresult, err := p.Exec()\n\t\t// May error or return empty — both acceptable\n\t\tif err == nil {\n\t\t\tassert.NotNil(t, result)\n\t\t}\n\t\tt.Logf(\"✓ robot.Executions handled for unknown robot\")\n\t})\n\n\tt.Run(\"AcceptsFilterMap\", func(t *testing.T) {\n\t\tmemberID := fmt.Sprintf(\"proc_exec_filter_%d\", time.Now().UnixNano())\n\t\tp := process.New(\"robot.Executions\", memberID, map[string]interface{}{\n\t\t\t\"page\": 1,\n\t\t\t\"pagesize\": 10,\n\t\t\t\"status\": \"completed\",\n\t\t})\n\t\tresult, err := p.Exec()\n\t\tif err == nil {\n\t\t\tassert.NotNil(t, result)\n\t\t}\n\t\tt.Logf(\"✓ robot.Executions with filter handled\")\n\t})\n}\n\n// ============================================================================\n// robot.Execution\n// ============================================================================\n\nfunc TestProcessExecution(t *testing.T) {\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\tt.Run(\"ErrorOnNonExistentExecution\", func(t *testing.T) {\n\t\tp := process.New(\"robot.Execution\", \"some_member_id\", \"non_existent_exec_id\")\n\t\t_, err := p.Exec()\n\t\tassert.Error(t, err, \"Should error for non-existent execution\")\n\t\tt.Logf(\"✓ robot.Execution returns error for non-existent execution\")\n\t})\n}\n\n// ============================================================================\n// Argument Validation\n// ============================================================================\n\nfunc TestProcessArgumentValidation(t *testing.T) {\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\tt.Run(\"UpdateChatTitle_RequiresTwoArgs\", func(t *testing.T) {\n\t\t// Missing title argument\n\t\tp := process.New(\"robot.UpdateChatTitle\", \"some_chat_id\")\n\t\t_, err := p.Exec()\n\t\tassert.Error(t, err, \"Should require 2 arguments\")\n\t})\n\n\tt.Run(\"Get_RequiresOneArg\", func(t *testing.T) {\n\t\tp := process.New(\"robot.Get\")\n\t\t_, err := p.Exec()\n\t\tassert.Error(t, err, \"Should require 1 argument\")\n\t})\n\n\tt.Run(\"Status_RequiresOneArg\", func(t *testing.T) {\n\t\tp := process.New(\"robot.Status\")\n\t\t_, err := p.Exec()\n\t\tassert.Error(t, err, \"Should require 1 argument\")\n\t})\n\n\tt.Run(\"Execution_RequiresTwoArgs\", func(t *testing.T) {\n\t\tp := process.New(\"robot.Execution\", \"only_one_arg\")\n\t\t_, err := p.Exec()\n\t\tassert.Error(t, err, \"Should require 2 arguments\")\n\t})\n}\n\n// ============================================================================\n// Integration: UpdateChatTitle flow (simulate Host Agent Next Hook)\n// ============================================================================\n\nfunc TestProcessUpdateChatTitleIntegration(t *testing.T) {\n\tif testing.Short() {\n\t\tt.Skip(\"Skipping integration test\")\n\t}\n\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\tchatStore := assistant.GetChatStore()\n\tif chatStore == nil {\n\t\tt.Skip(\"Chat store not configured\")\n\t}\n\n\tt.Run(\"SimulatesHostAgentNextHook\", func(t *testing.T) {\n\t\t// Simulate the full flow:\n\t\t// 1. ChatDrawer creates a chat with robot_id in metadata\n\t\t// 2. Host Agent Next Hook calls robot.UpdateChatTitle with confirmed goals\n\t\t// 3. History dropdown shows the goals as the chat title\n\n\t\tmemberID := \"120004485525\"\n\t\tchatID := fmt.Sprintf(\"robot_%s_%d\", memberID, time.Now().UnixMilli())\n\t\tconfirmedGoals := \"制作一张机甲图片，风格和设计由AI自主决定\"\n\n\t\t// Step 1: Create chat (simulating AssignTaskDrawer)\n\t\terr := chatStore.CreateChat(&storetypes.Chat{\n\t\t\tChatID: chatID,\n\t\t\tAssistantID: \"yao.robot-host\",\n\t\t\tStatus: \"active\",\n\t\t\tShare: \"private\",\n\t\t\tMetadata: map[string]interface{}{\n\t\t\t\t\"robot_id\": memberID,\n\t\t\t},\n\t\t\tCreatedAt: time.Now(),\n\t\t\tUpdatedAt: time.Now(),\n\t\t})\n\t\trequire.NoError(t, err)\n\t\tdefer chatStore.DeleteChat(chatID)\n\n\t\t// Step 2: Next Hook calls robot.UpdateChatTitle\n\t\tp := process.New(\"robot.UpdateChatTitle\", chatID, confirmedGoals)\n\t\t_, err = p.Exec()\n\t\trequire.NoError(t, err)\n\n\t\t// Step 3: Verify title is set (history dropdown will display this)\n\t\tchat, err := chatStore.GetChat(chatID)\n\t\trequire.NoError(t, err)\n\t\tassert.Equal(t, confirmedGoals, chat.Title,\n\t\t\t\"History dropdown should show confirmed goals as title\")\n\t\trequire.NotNil(t, chat.Metadata)\n\t\tassert.Equal(t, memberID, chat.Metadata[\"robot_id\"],\n\t\t\t\"Metadata robot_id should be preserved after title update\")\n\n\t\tt.Logf(\"✓ Full Host Agent flow: chat_id=%s, title=%q, robot_id=%v\",\n\t\t\tchatID, chat.Title, chat.Metadata[\"robot_id\"])\n\t})\n}\n\n// ensure context is used (avoid unused import)\nvar _ = context.Background\n" }, { "path": "agent/robot/robot.go", "content": "package robot\n\nimport (\n\t\"context\"\n\n\t\"github.com/yaoapp/yao/agent/robot/cache\"\n\t\"github.com/yaoapp/yao/agent/robot/dedup\"\n\t\"github.com/yaoapp/yao/agent/robot/events/integrations\"\n\t\"github.com/yaoapp/yao/agent/robot/events/integrations/telegram\"\n\t\"github.com/yaoapp/yao/agent/robot/executor\"\n\t\"github.com/yaoapp/yao/agent/robot/logger\"\n\t\"github.com/yaoapp/yao/agent/robot/manager\"\n\t\"github.com/yaoapp/yao/agent/robot/plan\"\n\t\"github.com/yaoapp/yao/agent/robot/pool\"\n\t\"github.com/yaoapp/yao/agent/robot/store\"\n\trobottypes \"github.com/yaoapp/yao/agent/robot/types\"\n)\n\nvar (\n\tlog = logger.New(\"robot\")\n\n\tglobalManager *manager.Manager\n\tglobalCache *cache.Cache\n\tglobalPool *pool.Pool\n\tglobalDedup *dedup.Dedup\n\tglobalStore *store.Store\n\tglobalExecutor executor.Executor\n\tglobalPlan *plan.Plan\n\tglobalDispatcher *integrations.Dispatcher\n)\n\n// Init initializes the robot agent system\nfunc Init() error {\n\tglobalCache = cache.New()\n\tglobalDedup = dedup.New()\n\tglobalStore = store.New()\n\tglobalPool = pool.New()\n\tglobalExecutor = executor.New()\n\tglobalManager = manager.New()\n\tglobalPlan = plan.New()\n\n\t// Load robots into cache from database before starting dispatcher\n\trCtx := robottypes.NewContext(context.Background(), nil)\n\tif err := globalCache.Load(rCtx); err != nil {\n\t\tlog.Warn(\"robot.Init: cache load failed (will rely on config events): %v\", err)\n\t}\n\n\tadapters := map[string]integrations.Adapter{\n\t\t\"telegram\": telegram.NewAdapter(),\n\t}\n\tglobalDispatcher = integrations.NewDispatcher(globalCache, adapters)\n\tif err := globalDispatcher.Start(context.Background()); err != nil {\n\t\treturn err\n\t}\n\n\treturn nil\n}\n\n// Shutdown gracefully shuts down the robot agent system\nfunc Shutdown() error {\n\tif globalDispatcher != nil {\n\t\tglobalDispatcher.Stop()\n\t}\n\treturn nil\n}\n\n// Manager returns the global manager instance\nfunc Manager() *manager.Manager {\n\treturn globalManager\n}\n" }, { "path": "agent/robot/store/execution.go", "content": "package store\n\nimport (\n\t\"context\"\n\t\"encoding/json\"\n\t\"fmt\"\n\t\"time\"\n\n\t\"github.com/yaoapp/gou/model\"\n\t\"github.com/yaoapp/yao/agent/robot/types\"\n)\n\n// ExecutionRecord - persistent storage for robot execution history\n// Maps to __yao.agent_execution model\ntype ExecutionRecord struct {\n\tID int64 `json:\"id,omitempty\"` // Auto-increment primary key\n\tExecutionID string `json:\"execution_id\"` // Unique execution identifier\n\tMemberID string `json:\"member_id\"` // Robot member ID (globally unique)\n\tTeamID string `json:\"team_id\"` // Team ID\n\tTriggerType types.TriggerType `json:\"trigger_type\"` // clock | human | event\n\n\t// Status tracking (synced with runtime Execution)\n\tStatus types.ExecStatus `json:\"status\"` // pending | running | completed | failed | cancelled\n\tPhase types.Phase `json:\"phase\"` // Current phase\n\tCurrent *CurrentState `json:\"current,omitempty\"`\n\tError string `json:\"error,omitempty\"`\n\n\t// UI display fields (updated by executor at each phase)\n\tName string `json:\"name,omitempty\"` // Execution title\n\tCurrentTaskName string `json:\"current_task_name,omitempty\"` // Current task description\n\n\t// Trigger input\n\tInput *types.TriggerInput `json:\"input,omitempty\"`\n\n\t// Phase outputs (P0-P5)\n\tInspiration *types.InspirationReport `json:\"inspiration,omitempty\"`\n\tGoals *types.Goals `json:\"goals,omitempty\"`\n\tTasks []types.Task `json:\"tasks,omitempty\"`\n\tResults []types.TaskResult `json:\"results,omitempty\"`\n\tDelivery *types.DeliveryResult `json:\"delivery,omitempty\"`\n\tLearning []types.LearningEntry `json:\"learning,omitempty\"`\n\n\t// V2: Conversation and suspend-resume fields\n\tChatID string `json:\"chat_id,omitempty\"`\n\tWaitingTaskID string `json:\"waiting_task_id,omitempty\"`\n\tWaitingQuestion string `json:\"waiting_question,omitempty\"`\n\tWaitingSince *time.Time `json:\"waiting_since,omitempty\"`\n\tResumeContext *types.ResumeContext `json:\"resume_context,omitempty\"`\n\n\t// Timestamps\n\tStartTime *time.Time `json:\"start_time,omitempty\"`\n\tEndTime *time.Time `json:\"end_time,omitempty\"`\n\tCreatedAt *time.Time `json:\"created_at,omitempty\"`\n\tUpdatedAt *time.Time `json:\"updated_at,omitempty\"`\n}\n\n// CurrentState - current executing state (for JSON storage)\ntype CurrentState struct {\n\tTaskIndex int `json:\"task_index\"` // index in Tasks slice\n\tProgress string `json:\"progress,omitempty\"` // human-readable progress (e.g., \"2/5 tasks\")\n}\n\n// ListOptions - options for listing execution records\ntype ListOptions struct {\n\tMemberID string `json:\"member_id,omitempty\"`\n\tTeamID string `json:\"team_id,omitempty\"`\n\tStatus types.ExecStatus `json:\"status,omitempty\"`\n\tExcludeStatuses []types.ExecStatus `json:\"exclude_statuses,omitempty\"`\n\tTriggerType types.TriggerType `json:\"trigger_type,omitempty\"`\n\tPage int `json:\"page,omitempty\"`\n\tPageSize int `json:\"pagesize,omitempty\"`\n\tOrderBy string `json:\"order_by,omitempty\"`\n}\n\n// ListResult wraps paginated list results\ntype ListResult struct {\n\tData []*ExecutionRecord\n\tTotal int\n\tPage int\n\tPageSize int\n}\n\n// ExecutionStore - persistent storage for robot execution records\ntype ExecutionStore struct {\n\tmodelID string\n}\n\n// NewExecutionStore creates a new execution store instance\nfunc NewExecutionStore() *ExecutionStore {\n\treturn &ExecutionStore{\n\t\tmodelID: \"__yao.agent.execution\",\n\t}\n}\n\n// Save creates or updates an execution record\nfunc (s *ExecutionStore) Save(ctx context.Context, record *ExecutionRecord) error {\n\tmod := model.Select(s.modelID)\n\tif mod == nil {\n\t\treturn fmt.Errorf(\"model %s not found\", s.modelID)\n\t}\n\n\tdata := s.recordToMap(record)\n\n\t// Check if record exists by execution_id\n\texisting, err := s.Get(ctx, record.ExecutionID)\n\tif err == nil && existing != nil {\n\t\t// Update existing record\n\t\t_, err = mod.UpdateWhere(\n\t\t\tmodel.QueryParam{\n\t\t\t\tWheres: []model.QueryWhere{\n\t\t\t\t\t{Column: \"execution_id\", Value: record.ExecutionID},\n\t\t\t\t},\n\t\t\t},\n\t\t\tdata,\n\t\t)\n\t\tif err != nil {\n\t\t\treturn fmt.Errorf(\"failed to update execution record: %w\", err)\n\t\t}\n\t\treturn nil\n\t}\n\n\t// Create new record\n\t_, err = mod.Create(data)\n\tif err != nil {\n\t\treturn fmt.Errorf(\"failed to create execution record: %w\", err)\n\t}\n\treturn nil\n}\n\n// Get retrieves an execution record by execution_id\nfunc (s *ExecutionStore) Get(ctx context.Context, executionID string) (*ExecutionRecord, error) {\n\tmod := model.Select(s.modelID)\n\tif mod == nil {\n\t\treturn nil, fmt.Errorf(\"model %s not found\", s.modelID)\n\t}\n\n\trows, err := mod.Get(model.QueryParam{\n\t\tWheres: []model.QueryWhere{\n\t\t\t{Column: \"execution_id\", Value: executionID},\n\t\t},\n\t\tLimit: 1,\n\t})\n\tif err != nil {\n\t\treturn nil, fmt.Errorf(\"failed to get execution record: %w\", err)\n\t}\n\n\tif len(rows) == 0 {\n\t\treturn nil, nil\n\t}\n\n\treturn s.mapToRecord(rows[0])\n}\n\n// List retrieves execution records with pagination using mod.Paginate\nfunc (s *ExecutionStore) List(ctx context.Context, opts *ListOptions) (*ListResult, error) {\n\tmod := model.Select(s.modelID)\n\tif mod == nil {\n\t\treturn nil, fmt.Errorf(\"model %s not found\", s.modelID)\n\t}\n\n\tparams := model.QueryParam{}\n\tvar wheres []model.QueryWhere\n\n\tpage := 1\n\tpageSize := 20\n\n\tif opts != nil {\n\t\tif opts.MemberID != \"\" {\n\t\t\twheres = append(wheres, model.QueryWhere{Column: \"member_id\", Value: opts.MemberID})\n\t\t}\n\t\tif opts.TeamID != \"\" {\n\t\t\twheres = append(wheres, model.QueryWhere{Column: \"team_id\", Value: opts.TeamID})\n\t\t}\n\t\tif opts.Status != \"\" {\n\t\t\twheres = append(wheres, model.QueryWhere{Column: \"status\", Value: string(opts.Status)})\n\t\t}\n\t\tfor _, es := range opts.ExcludeStatuses {\n\t\t\twheres = append(wheres, model.QueryWhere{Column: \"status\", Value: string(es), OP: \"ne\"})\n\t\t}\n\t\tif opts.TriggerType != \"\" {\n\t\t\twheres = append(wheres, model.QueryWhere{Column: \"trigger_type\", Value: string(opts.TriggerType)})\n\t\t}\n\n\t\tif opts.Page > 0 {\n\t\t\tpage = opts.Page\n\t\t}\n\t\tif opts.PageSize > 0 {\n\t\t\tpageSize = opts.PageSize\n\t\t\tif pageSize > 100 {\n\t\t\t\tpageSize = 100\n\t\t\t}\n\t\t}\n\n\t\tif opts.OrderBy != \"\" {\n\t\t\tparts := splitOrderBy(opts.OrderBy)\n\t\t\tparams.Orders = []model.QueryOrder{{Column: parts[0], Option: parts[1]}}\n\t\t} else {\n\t\t\tparams.Orders = []model.QueryOrder{{Column: \"start_time\", Option: \"desc\"}}\n\t\t}\n\t} else {\n\t\tparams.Orders = []model.QueryOrder{{Column: \"start_time\", Option: \"desc\"}}\n\t}\n\n\tparams.Wheres = wheres\n\n\tres, err := mod.Paginate(params, page, pageSize)\n\tif err != nil {\n\t\treturn nil, fmt.Errorf(\"failed to list execution records: %w\", err)\n\t}\n\n\ttotal := 0\n\tif v, ok := res[\"total\"].(int64); ok {\n\t\ttotal = int(v)\n\t} else if v, ok := res[\"total\"].(int); ok {\n\t\ttotal = v\n\t}\n\n\trecords := make([]*ExecutionRecord, 0)\n\tfor _, row := range toRows(res[\"data\"]) {\n\t\trecord, err := s.mapToRecord(row)\n\t\tif err != nil {\n\t\t\tcontinue\n\t\t}\n\t\trecords = append(records, record)\n\t}\n\n\treturn &ListResult{\n\t\tData: records,\n\t\tTotal: total,\n\t\tPage: page,\n\t\tPageSize: pageSize,\n\t}, nil\n}\n\n// UpdatePhase updates the current phase and its data\nfunc (s *ExecutionStore) UpdatePhase(ctx context.Context, executionID string, phase types.Phase, data interface{}) error {\n\tmod := model.Select(s.modelID)\n\tif mod == nil {\n\t\treturn fmt.Errorf(\"model %s not found\", s.modelID)\n\t}\n\n\tupdateData := map[string]interface{}{\n\t\t\"phase\": string(phase),\n\t}\n\n\t// Set the appropriate phase output field\n\tswitch phase {\n\tcase types.PhaseInspiration:\n\t\tif data != nil {\n\t\t\tupdateData[\"inspiration\"] = data\n\t\t}\n\tcase types.PhaseGoals:\n\t\tif data != nil {\n\t\t\tupdateData[\"goals\"] = data\n\t\t}\n\tcase types.PhaseTasks:\n\t\tif data != nil {\n\t\t\tupdateData[\"tasks\"] = data\n\t\t}\n\tcase types.PhaseRun:\n\t\tif data != nil {\n\t\t\tupdateData[\"results\"] = data\n\t\t}\n\tcase types.PhaseDelivery:\n\t\tif data != nil {\n\t\t\tupdateData[\"delivery\"] = data\n\t\t}\n\tcase types.PhaseLearning:\n\t\tif data != nil {\n\t\t\tupdateData[\"learning\"] = data\n\t\t}\n\t}\n\n\t_, err := mod.UpdateWhere(\n\t\tmodel.QueryParam{\n\t\t\tWheres: []model.QueryWhere{\n\t\t\t\t{Column: \"execution_id\", Value: executionID},\n\t\t\t},\n\t\t},\n\t\tupdateData,\n\t)\n\tif err != nil {\n\t\treturn fmt.Errorf(\"failed to update phase: %w\", err)\n\t}\n\n\treturn nil\n}\n\n// UpdateStatus updates the execution status\nfunc (s *ExecutionStore) UpdateStatus(ctx context.Context, executionID string, status types.ExecStatus, errorMsg string) error {\n\tmod := model.Select(s.modelID)\n\tif mod == nil {\n\t\treturn fmt.Errorf(\"model %s not found\", s.modelID)\n\t}\n\n\tupdateData := map[string]interface{}{\n\t\t\"status\": string(status),\n\t}\n\n\tif errorMsg != \"\" {\n\t\tupdateData[\"error\"] = errorMsg\n\t}\n\n\t// Set end_time for terminal states\n\tif status == types.ExecCompleted || status == types.ExecFailed || status == types.ExecCancelled {\n\t\tnow := time.Now()\n\t\tupdateData[\"end_time\"] = now\n\t}\n\n\t_, err := mod.UpdateWhere(\n\t\tmodel.QueryParam{\n\t\t\tWheres: []model.QueryWhere{\n\t\t\t\t{Column: \"execution_id\", Value: executionID},\n\t\t\t},\n\t\t},\n\t\tupdateData,\n\t)\n\tif err != nil {\n\t\treturn fmt.Errorf(\"failed to update status: %w\", err)\n\t}\n\n\treturn nil\n}\n\n// UpdateCurrent updates the current executing state\nfunc (s *ExecutionStore) UpdateCurrent(ctx context.Context, executionID string, current *CurrentState) error {\n\tmod := model.Select(s.modelID)\n\tif mod == nil {\n\t\treturn fmt.Errorf(\"model %s not found\", s.modelID)\n\t}\n\n\tupdateData := map[string]interface{}{\n\t\t\"current\": current,\n\t}\n\n\t_, err := mod.UpdateWhere(\n\t\tmodel.QueryParam{\n\t\t\tWheres: []model.QueryWhere{\n\t\t\t\t{Column: \"execution_id\", Value: executionID},\n\t\t\t},\n\t\t},\n\t\tupdateData,\n\t)\n\tif err != nil {\n\t\treturn fmt.Errorf(\"failed to update current state: %w\", err)\n\t}\n\n\treturn nil\n}\n\n// UpdateTasks updates the tasks array with current status\n// This should be called after each task completes to persist status changes\nfunc (s *ExecutionStore) UpdateTasks(ctx context.Context, executionID string, tasks []types.Task, current *CurrentState) error {\n\tmod := model.Select(s.modelID)\n\tif mod == nil {\n\t\treturn fmt.Errorf(\"model %s not found\", s.modelID)\n\t}\n\n\tupdateData := map[string]interface{}{\n\t\t\"tasks\": tasks,\n\t\t\"current\": current,\n\t}\n\n\t_, err := mod.UpdateWhere(\n\t\tmodel.QueryParam{\n\t\t\tWheres: []model.QueryWhere{\n\t\t\t\t{Column: \"execution_id\", Value: executionID},\n\t\t\t},\n\t\t},\n\t\tupdateData,\n\t)\n\tif err != nil {\n\t\treturn fmt.Errorf(\"failed to update tasks: %w\", err)\n\t}\n\n\treturn nil\n}\n\n// UpdateUIFields updates the UI display fields (name and current_task_name)\n// These fields are updated by executor at each phase for frontend display\nfunc (s *ExecutionStore) UpdateUIFields(ctx context.Context, executionID string, name string, currentTaskName string) error {\n\tmod := model.Select(s.modelID)\n\tif mod == nil {\n\t\treturn fmt.Errorf(\"model %s not found\", s.modelID)\n\t}\n\n\tupdateData := map[string]interface{}{}\n\tif name != \"\" {\n\t\tupdateData[\"name\"] = name\n\t}\n\tif currentTaskName != \"\" {\n\t\tupdateData[\"current_task_name\"] = currentTaskName\n\t}\n\n\tif len(updateData) == 0 {\n\t\treturn nil // Nothing to update\n\t}\n\n\t_, err := mod.UpdateWhere(\n\t\tmodel.QueryParam{\n\t\t\tWheres: []model.QueryWhere{\n\t\t\t\t{Column: \"execution_id\", Value: executionID},\n\t\t\t},\n\t\t},\n\t\tupdateData,\n\t)\n\tif err != nil {\n\t\treturn fmt.Errorf(\"failed to update UI fields: %w\", err)\n\t}\n\n\treturn nil\n}\n\n// UpdateSuspendState atomically transitions an execution to waiting status\n// with all suspend-related fields in a single DB write.\nfunc (s *ExecutionStore) UpdateSuspendState(ctx context.Context, executionID string, waitingTaskID string, question string, resumeCtx *types.ResumeContext) error {\n\tmod := model.Select(s.modelID)\n\tif mod == nil {\n\t\treturn fmt.Errorf(\"model %s not found\", s.modelID)\n\t}\n\n\tnow := time.Now()\n\tupdateData := map[string]interface{}{\n\t\t\"status\": string(types.ExecWaiting),\n\t\t\"waiting_task_id\": waitingTaskID,\n\t\t\"waiting_question\": question,\n\t\t\"waiting_since\": now,\n\t}\n\tif resumeCtx != nil {\n\t\tupdateData[\"resume_context\"] = resumeCtx\n\t}\n\n\t_, err := mod.UpdateWhere(\n\t\tmodel.QueryParam{\n\t\t\tWheres: []model.QueryWhere{\n\t\t\t\t{Column: \"execution_id\", Value: executionID},\n\t\t\t},\n\t\t},\n\t\tupdateData,\n\t)\n\tif err != nil {\n\t\treturn fmt.Errorf(\"failed to update suspend state: %w\", err)\n\t}\n\treturn nil\n}\n\n// UpdateResumeState clears waiting fields and transitions execution back to running.\nfunc (s *ExecutionStore) UpdateResumeState(ctx context.Context, executionID string) error {\n\tmod := model.Select(s.modelID)\n\tif mod == nil {\n\t\treturn fmt.Errorf(\"model %s not found\", s.modelID)\n\t}\n\n\tupdateData := map[string]interface{}{\n\t\t\"status\": string(types.ExecRunning),\n\t\t\"waiting_task_id\": \"\",\n\t\t\"waiting_question\": \"\",\n\t\t\"waiting_since\": nil,\n\t\t\"resume_context\": nil,\n\t}\n\n\t_, err := mod.UpdateWhere(\n\t\tmodel.QueryParam{\n\t\t\tWheres: []model.QueryWhere{\n\t\t\t\t{Column: \"execution_id\", Value: executionID},\n\t\t\t},\n\t\t},\n\t\tupdateData,\n\t)\n\tif err != nil {\n\t\treturn fmt.Errorf(\"failed to update resume state: %w\", err)\n\t}\n\treturn nil\n}\n\n// Delete removes an execution record by execution_id\nfunc (s *ExecutionStore) Delete(ctx context.Context, executionID string) error {\n\tmod := model.Select(s.modelID)\n\tif mod == nil {\n\t\treturn fmt.Errorf(\"model %s not found\", s.modelID)\n\t}\n\n\t_, err := mod.DeleteWhere(model.QueryParam{\n\t\tWheres: []model.QueryWhere{\n\t\t\t{Column: \"execution_id\", Value: executionID},\n\t\t},\n\t})\n\tif err != nil {\n\t\treturn fmt.Errorf(\"failed to delete execution record: %w\", err)\n\t}\n\n\treturn nil\n}\n\n// recordToMap converts ExecutionRecord to map for model operations\nfunc (s *ExecutionStore) recordToMap(record *ExecutionRecord) map[string]interface{} {\n\tdata := map[string]interface{}{\n\t\t\"execution_id\": record.ExecutionID,\n\t\t\"member_id\": record.MemberID,\n\t\t\"team_id\": record.TeamID,\n\t\t\"trigger_type\": string(record.TriggerType),\n\t\t\"status\": string(record.Status),\n\t\t\"phase\": string(record.Phase),\n\t}\n\n\tif record.Error != \"\" {\n\t\tdata[\"error\"] = record.Error\n\t}\n\tif record.Name != \"\" {\n\t\tdata[\"name\"] = record.Name\n\t}\n\tif record.CurrentTaskName != \"\" {\n\t\tdata[\"current_task_name\"] = record.CurrentTaskName\n\t}\n\tif record.Current != nil {\n\t\tdata[\"current\"] = record.Current\n\t}\n\tif record.Input != nil {\n\t\tdata[\"input\"] = record.Input\n\t}\n\tif record.Inspiration != nil {\n\t\tdata[\"inspiration\"] = record.Inspiration\n\t}\n\tif record.Goals != nil {\n\t\tdata[\"goals\"] = record.Goals\n\t}\n\tif record.Tasks != nil {\n\t\tdata[\"tasks\"] = record.Tasks\n\t}\n\tif record.Results != nil {\n\t\tdata[\"results\"] = record.Results\n\t}\n\tif record.Delivery != nil {\n\t\tdata[\"delivery\"] = record.Delivery\n\t}\n\tif record.Learning != nil {\n\t\tdata[\"learning\"] = record.Learning\n\t}\n\t// V2 fields\n\tif record.ChatID != \"\" {\n\t\tdata[\"chat_id\"] = record.ChatID\n\t}\n\tif record.WaitingTaskID != \"\" {\n\t\tdata[\"waiting_task_id\"] = record.WaitingTaskID\n\t}\n\tif record.WaitingQuestion != \"\" {\n\t\tdata[\"waiting_question\"] = record.WaitingQuestion\n\t}\n\tif record.WaitingSince != nil {\n\t\tdata[\"waiting_since\"] = *record.WaitingSince\n\t}\n\tif record.ResumeContext != nil {\n\t\tdata[\"resume_context\"] = record.ResumeContext\n\t}\n\n\tif record.StartTime != nil {\n\t\tdata[\"start_time\"] = *record.StartTime\n\t}\n\tif record.EndTime != nil {\n\t\tdata[\"end_time\"] = *record.EndTime\n\t}\n\n\treturn data\n}\n\n// mapToRecord converts a model row to ExecutionRecord\nfunc (s *ExecutionStore) mapToRecord(row map[string]interface{}) (*ExecutionRecord, error) {\n\trecord := &ExecutionRecord{}\n\n\t// Basic fields\n\tif v, ok := row[\"id\"]; ok {\n\t\tswitch id := v.(type) {\n\t\tcase float64:\n\t\t\trecord.ID = int64(id)\n\t\tcase int64:\n\t\t\trecord.ID = id\n\t\tcase int:\n\t\t\trecord.ID = int64(id)\n\t\t}\n\t}\n\tif v, ok := row[\"execution_id\"].(string); ok {\n\t\trecord.ExecutionID = v\n\t}\n\tif v, ok := row[\"member_id\"].(string); ok {\n\t\trecord.MemberID = v\n\t}\n\tif v, ok := row[\"team_id\"].(string); ok {\n\t\trecord.TeamID = v\n\t}\n\tif v, ok := row[\"trigger_type\"].(string); ok {\n\t\trecord.TriggerType = types.TriggerType(v)\n\t}\n\tif v, ok := row[\"status\"].(string); ok {\n\t\trecord.Status = types.ExecStatus(v)\n\t}\n\tif v, ok := row[\"phase\"].(string); ok {\n\t\trecord.Phase = types.Phase(v)\n\t}\n\tif v, ok := row[\"error\"].(string); ok {\n\t\trecord.Error = v\n\t}\n\tif v, ok := row[\"name\"].(string); ok {\n\t\trecord.Name = v\n\t}\n\tif v, ok := row[\"current_task_name\"].(string); ok {\n\t\trecord.CurrentTaskName = v\n\t}\n\n\t// JSON fields - need to unmarshal\n\tif v := row[\"current\"]; v != nil {\n\t\trecord.Current = s.parseCurrentState(v)\n\t}\n\tif v := row[\"input\"]; v != nil {\n\t\trecord.Input = s.parseTriggerInput(v)\n\t}\n\tif v := row[\"inspiration\"]; v != nil {\n\t\trecord.Inspiration = s.parseInspirationReport(v)\n\t}\n\tif v := row[\"goals\"]; v != nil {\n\t\trecord.Goals = s.parseGoals(v)\n\t}\n\tif v := row[\"tasks\"]; v != nil {\n\t\trecord.Tasks = s.parseTasks(v)\n\t}\n\tif v := row[\"results\"]; v != nil {\n\t\trecord.Results = s.parseResults(v)\n\t}\n\tif v := row[\"delivery\"]; v != nil {\n\t\trecord.Delivery = s.parseDeliveryResult(v)\n\t}\n\tif v := row[\"learning\"]; v != nil {\n\t\trecord.Learning = s.parseLearningEntries(v)\n\t}\n\n\t// V2 fields\n\tif v, ok := row[\"chat_id\"].(string); ok {\n\t\trecord.ChatID = v\n\t}\n\tif v, ok := row[\"waiting_task_id\"].(string); ok {\n\t\trecord.WaitingTaskID = v\n\t}\n\tif v, ok := row[\"waiting_question\"].(string); ok {\n\t\trecord.WaitingQuestion = v\n\t}\n\tif v := row[\"waiting_since\"]; v != nil {\n\t\trecord.WaitingSince = s.parseTime(v)\n\t}\n\tif v := row[\"resume_context\"]; v != nil {\n\t\trecord.ResumeContext = s.parseResumeContext(v)\n\t}\n\n\t// Timestamps\n\tif v := row[\"start_time\"]; v != nil {\n\t\trecord.StartTime = s.parseTime(v)\n\t}\n\tif v := row[\"end_time\"]; v != nil {\n\t\trecord.EndTime = s.parseTime(v)\n\t}\n\tif v := row[\"created_at\"]; v != nil {\n\t\trecord.CreatedAt = s.parseTime(v)\n\t}\n\tif v := row[\"updated_at\"]; v != nil {\n\t\trecord.UpdatedAt = s.parseTime(v)\n\t}\n\n\treturn record, nil\n}\n\n// Helper functions for parsing JSON fields\n\nfunc (s *ExecutionStore) parseCurrentState(v interface{}) *CurrentState {\n\tdata, err := s.toJSON(v)\n\tif err != nil {\n\t\treturn nil\n\t}\n\tvar state CurrentState\n\tif err := json.Unmarshal(data, &state); err != nil {\n\t\treturn nil\n\t}\n\treturn &state\n}\n\nfunc (s *ExecutionStore) parseTriggerInput(v interface{}) *types.TriggerInput {\n\tdata, err := s.toJSON(v)\n\tif err != nil {\n\t\treturn nil\n\t}\n\tvar input types.TriggerInput\n\tif err := json.Unmarshal(data, &input); err != nil {\n\t\treturn nil\n\t}\n\treturn &input\n}\n\nfunc (s *ExecutionStore) parseInspirationReport(v interface{}) *types.InspirationReport {\n\tdata, err := s.toJSON(v)\n\tif err != nil {\n\t\treturn nil\n\t}\n\tvar report types.InspirationReport\n\tif err := json.Unmarshal(data, &report); err != nil {\n\t\treturn nil\n\t}\n\treturn &report\n}\n\nfunc (s *ExecutionStore) parseGoals(v interface{}) *types.Goals {\n\tdata, err := s.toJSON(v)\n\tif err != nil {\n\t\treturn nil\n\t}\n\tvar goals types.Goals\n\tif err := json.Unmarshal(data, &goals); err != nil {\n\t\treturn nil\n\t}\n\treturn &goals\n}\n\nfunc (s *ExecutionStore) parseTasks(v interface{}) []types.Task {\n\tdata, err := s.toJSON(v)\n\tif err != nil {\n\t\treturn nil\n\t}\n\tvar tasks []types.Task\n\tif err := json.Unmarshal(data, &tasks); err != nil {\n\t\treturn nil\n\t}\n\treturn tasks\n}\n\nfunc (s *ExecutionStore) parseResults(v interface{}) []types.TaskResult {\n\tdata, err := s.toJSON(v)\n\tif err != nil {\n\t\treturn nil\n\t}\n\tvar results []types.TaskResult\n\tif err := json.Unmarshal(data, &results); err != nil {\n\t\treturn nil\n\t}\n\treturn results\n}\n\nfunc (s *ExecutionStore) parseDeliveryResult(v interface{}) *types.DeliveryResult {\n\tdata, err := s.toJSON(v)\n\tif err != nil {\n\t\treturn nil\n\t}\n\tvar result types.DeliveryResult\n\tif err := json.Unmarshal(data, &result); err != nil {\n\t\treturn nil\n\t}\n\treturn &result\n}\n\nfunc (s *ExecutionStore) parseLearningEntries(v interface{}) []types.LearningEntry {\n\tdata, err := s.toJSON(v)\n\tif err != nil {\n\t\treturn nil\n\t}\n\tvar entries []types.LearningEntry\n\tif err := json.Unmarshal(data, &entries); err != nil {\n\t\treturn nil\n\t}\n\treturn entries\n}\n\nfunc (s *ExecutionStore) parseResumeContext(v interface{}) *types.ResumeContext {\n\tdata, err := s.toJSON(v)\n\tif err != nil {\n\t\treturn nil\n\t}\n\tvar ctx types.ResumeContext\n\tif err := json.Unmarshal(data, &ctx); err != nil {\n\t\treturn nil\n\t}\n\treturn &ctx\n}\n\nfunc (s *ExecutionStore) toJSON(v interface{}) ([]byte, error) {\n\tswitch data := v.(type) {\n\tcase []byte:\n\t\treturn data, nil\n\tcase string:\n\t\treturn []byte(data), nil\n\tcase map[string]interface{}, []interface{}:\n\t\treturn json.Marshal(data)\n\tdefault:\n\t\treturn json.Marshal(v)\n\t}\n}\n\n// splitOrderBy parses \"column desc\" or \"column asc\" or just \"column\"\n// Returns [column, option] where option defaults to \"desc\"\n// toRows converts Paginate result data to []map[string]interface{}\n// handles type aliases like maps.MapStrAny via JSON round-trip\nfunc toRows(data interface{}) []map[string]interface{} {\n\tif data == nil {\n\t\treturn nil\n\t}\n\traw, err := json.Marshal(data)\n\tif err != nil {\n\t\treturn nil\n\t}\n\tvar rows []map[string]interface{}\n\tif err := json.Unmarshal(raw, &rows); err != nil {\n\t\treturn nil\n\t}\n\treturn rows\n}\n\nfunc splitOrderBy(orderBy string) [2]string {\n\tparts := [2]string{\"\", \"desc\"}\n\tif orderBy == \"\" {\n\t\treturn parts\n\t}\n\n\t// Split by space\n\tfor i, c := range orderBy {\n\t\tif c == ' ' {\n\t\t\tparts[0] = orderBy[:i]\n\t\t\trest := orderBy[i+1:]\n\t\t\tif rest == \"asc\" || rest == \"ASC\" {\n\t\t\t\tparts[1] = \"asc\"\n\t\t\t} else if rest == \"desc\" || rest == \"DESC\" {\n\t\t\t\tparts[1] = \"desc\"\n\t\t\t}\n\t\t\treturn parts\n\t\t}\n\t}\n\n\t// No space found, just column name\n\tparts[0] = orderBy\n\treturn parts\n}\n\nfunc (s *ExecutionStore) parseTime(v interface{}) *time.Time {\n\tswitch t := v.(type) {\n\tcase time.Time:\n\t\treturn &t\n\tcase *time.Time:\n\t\treturn t\n\tcase string:\n\t\t// Try parsing common time formats\n\t\tformats := []string{\n\t\t\ttime.RFC3339,\n\t\t\ttime.RFC3339Nano,\n\t\t\t\"2006-01-02 15:04:05\",\n\t\t\t\"2006-01-02T15:04:05Z\",\n\t\t}\n\t\tfor _, format := range formats {\n\t\t\tif parsed, err := time.Parse(format, t); err == nil {\n\t\t\t\treturn &parsed\n\t\t\t}\n\t\t}\n\t}\n\treturn nil\n}\n\n// ==================== Results & Activities ====================\n\n// ResultListOptions - options for listing execution results (deliveries)\ntype ResultListOptions struct {\n\tMemberID string `json:\"member_id,omitempty\"`\n\tTeamID string `json:\"team_id,omitempty\"`\n\tTriggerType types.TriggerType `json:\"trigger_type,omitempty\"`\n\tKeyword string `json:\"keyword,omitempty\"`\n\tPage int `json:\"page,omitempty\"`\n\tPageSize int `json:\"pagesize,omitempty\"`\n}\n\n// ResultListResponse - paginated result list response\ntype ResultListResponse struct {\n\tData []*ExecutionRecord `json:\"data\"`\n\tTotal int `json:\"total\"`\n\tPage int `json:\"page\"`\n\tPageSize int `json:\"pagesize\"`\n}\n\n// ListResults retrieves completed executions with delivery content\n// Only returns executions where delivery.content is not null\nfunc (s *ExecutionStore) ListResults(ctx context.Context, opts *ResultListOptions) (*ResultListResponse, error) {\n\tmod := model.Select(s.modelID)\n\tif mod == nil {\n\t\treturn nil, fmt.Errorf(\"model %s not found\", s.modelID)\n\t}\n\n\t// Build where conditions\n\tvar wheres []model.QueryWhere\n\n\t// Must have completed status and delivery content\n\twheres = append(wheres, model.QueryWhere{Column: \"status\", Value: \"completed\"})\n\twheres = append(wheres, model.QueryWhere{Column: \"delivery\", OP: \"notnull\"})\n\n\tif opts != nil {\n\t\tif opts.MemberID != \"\" {\n\t\t\twheres = append(wheres, model.QueryWhere{Column: \"member_id\", Value: opts.MemberID})\n\t\t}\n\t\tif opts.TeamID != \"\" {\n\t\t\twheres = append(wheres, model.QueryWhere{Column: \"team_id\", Value: opts.TeamID})\n\t\t}\n\t\tif opts.TriggerType != \"\" {\n\t\t\twheres = append(wheres, model.QueryWhere{Column: \"trigger_type\", Value: string(opts.TriggerType)})\n\t\t}\n\t\t// Keyword search in name field (delivery.content.summary is in JSON, harder to search)\n\t\t// For now search in the name field\n\t\tif opts.Keyword != \"\" {\n\t\t\twheres = append(wheres, model.QueryWhere{Column: \"name\", OP: \"like\", Value: \"%\" + opts.Keyword + \"%\"})\n\t\t}\n\t}\n\n\tpage := 1\n\tpageSize := 20\n\tif opts != nil {\n\t\tif opts.Page > 0 {\n\t\t\tpage = opts.Page\n\t\t}\n\t\tif opts.PageSize > 0 {\n\t\t\tpageSize = opts.PageSize\n\t\t\tif pageSize > 100 {\n\t\t\t\tpageSize = 100\n\t\t\t}\n\t\t}\n\t}\n\n\tparams := model.QueryParam{\n\t\tWheres: wheres,\n\t\tOrders: []model.QueryOrder{{Column: \"end_time\", Option: \"desc\"}},\n\t}\n\n\tres, err := mod.Paginate(params, page, pageSize)\n\tif err != nil {\n\t\treturn nil, fmt.Errorf(\"failed to list results: %w\", err)\n\t}\n\n\ttotal := 0\n\tif v, ok := res[\"total\"].(int64); ok {\n\t\ttotal = int(v)\n\t} else if v, ok := res[\"total\"].(int); ok {\n\t\ttotal = v\n\t}\n\n\trecords := make([]*ExecutionRecord, 0)\n\tfor _, row := range toRows(res[\"data\"]) {\n\t\trecord, err := s.mapToRecord(row)\n\t\tif err != nil {\n\t\t\tcontinue\n\t\t}\n\t\tif record.Delivery != nil && record.Delivery.Content != nil {\n\t\t\trecords = append(records, record)\n\t\t}\n\t}\n\n\treturn &ResultListResponse{\n\t\tData: records,\n\t\tTotal: total,\n\t\tPage: page,\n\t\tPageSize: pageSize,\n\t}, nil\n}\n\n// CountResults counts total results matching criteria\nfunc (s *ExecutionStore) CountResults(ctx context.Context, opts *ResultListOptions) (int, error) {\n\tvar wheres []model.QueryWhere\n\n\t// Must have completed status and delivery content\n\twheres = append(wheres, model.QueryWhere{Column: \"status\", Value: \"completed\"})\n\twheres = append(wheres, model.QueryWhere{Column: \"delivery\", OP: \"notnull\"})\n\n\tif opts != nil {\n\t\tif opts.MemberID != \"\" {\n\t\t\twheres = append(wheres, model.QueryWhere{Column: \"member_id\", Value: opts.MemberID})\n\t\t}\n\t\tif opts.TeamID != \"\" {\n\t\t\twheres = append(wheres, model.QueryWhere{Column: \"team_id\", Value: opts.TeamID})\n\t\t}\n\t\tif opts.TriggerType != \"\" {\n\t\t\twheres = append(wheres, model.QueryWhere{Column: \"trigger_type\", Value: string(opts.TriggerType)})\n\t\t}\n\t\tif opts.Keyword != \"\" {\n\t\t\twheres = append(wheres, model.QueryWhere{Column: \"name\", OP: \"like\", Value: \"%\" + opts.Keyword + \"%\"})\n\t\t}\n\t}\n\n\treturn s.countWithWheres(wheres)\n}\n\n// countWithWheres counts records matching the given where conditions\nfunc (s *ExecutionStore) countWithWheres(wheres []model.QueryWhere) (int, error) {\n\tmod := model.Select(s.modelID)\n\tif mod == nil {\n\t\treturn 0, fmt.Errorf(\"model %s not found\", s.modelID)\n\t}\n\n\t// Use model Paginate to get total count\n\tparams := model.QueryParam{\n\t\tWheres: wheres,\n\t\tLimit: 1,\n\t}\n\n\tresult, err := mod.Paginate(params, 1, 1)\n\tif err != nil {\n\t\treturn 0, fmt.Errorf(\"failed to count records: %w\", err)\n\t}\n\n\t// Paginate returns map with total field\n\tif result == nil {\n\t\treturn 0, nil\n\t}\n\n\ttotal := 0\n\tif t, ok := result[\"total\"]; ok {\n\t\tswitch v := t.(type) {\n\t\tcase float64:\n\t\t\ttotal = int(v)\n\t\tcase int64:\n\t\t\ttotal = int(v)\n\t\tcase int:\n\t\t\ttotal = v\n\t\t}\n\t}\n\n\treturn total, nil\n}\n\n// ActivityType represents the type of activity\ntype ActivityType string\n\nconst (\n\tActivityExecutionStarted ActivityType = \"execution.started\"\n\tActivityExecutionCompleted ActivityType = \"execution.completed\"\n\tActivityExecutionFailed ActivityType = \"execution.failed\"\n\tActivityExecutionCancelled ActivityType = \"execution.cancelled\"\n)\n\n// Activity represents a robot activity entry\ntype Activity struct {\n\tType ActivityType `json:\"type\"`\n\tRobotID string `json:\"robot_id\"`\n\tRobotName string `json:\"robot_name,omitempty\"` // Will be populated by API layer\n\tExecutionID string `json:\"execution_id\"`\n\tMessage string `json:\"message\"`\n\tTimestamp time.Time `json:\"timestamp\"`\n}\n\n// ActivityListOptions - options for listing activities\ntype ActivityListOptions struct {\n\tTeamID string `json:\"team_id,omitempty\"` // Filter by team ID\n\tSince *time.Time `json:\"since,omitempty\"` // Only activities after this time\n\tLimit int `json:\"limit,omitempty\"`\n\tType ActivityType `json:\"type,omitempty\"` // Filter by activity type\n}\n\n// ListActivities derives activities from recent execution status changes\nfunc (s *ExecutionStore) ListActivities(ctx context.Context, opts *ActivityListOptions) ([]*Activity, error) {\n\tmod := model.Select(s.modelID)\n\tif mod == nil {\n\t\treturn nil, fmt.Errorf(\"model %s not found\", s.modelID)\n\t}\n\n\t// Build where conditions\n\tvar wheres []model.QueryWhere\n\n\t// Filter by activity type if specified\n\t// Map activity types to execution statuses\n\tif opts != nil && opts.Type != \"\" {\n\t\tswitch opts.Type {\n\t\tcase ActivityExecutionStarted:\n\t\t\twheres = append(wheres, model.QueryWhere{Column: \"status\", Value: \"running\"})\n\t\tcase ActivityExecutionCompleted:\n\t\t\twheres = append(wheres, model.QueryWhere{Column: \"status\", Value: \"completed\"})\n\t\tcase ActivityExecutionFailed:\n\t\t\twheres = append(wheres, model.QueryWhere{Column: \"status\", Value: \"failed\"})\n\t\tcase ActivityExecutionCancelled:\n\t\t\twheres = append(wheres, model.QueryWhere{Column: \"status\", Value: \"cancelled\"})\n\t\tdefault:\n\t\t\t// Unknown type, return empty\n\t\t\treturn []*Activity{}, nil\n\t\t}\n\t} else {\n\t\t// Only completed, failed, or cancelled executions generate activities\n\t\t// For started activities, we'd need running status\n\t\twheres = append(wheres, model.QueryWhere{\n\t\t\tColumn: \"status\",\n\t\t\tOP: \"in\",\n\t\t\tValue: []string{\"completed\", \"failed\", \"cancelled\", \"running\"},\n\t\t})\n\t}\n\n\tif opts != nil {\n\t\tif opts.TeamID != \"\" {\n\t\t\twheres = append(wheres, model.QueryWhere{Column: \"team_id\", Value: opts.TeamID})\n\t\t}\n\t\tif opts.Since != nil {\n\t\t\t// Get executions that ended or started after 'since'\n\t\t\twheres = append(wheres, model.QueryWhere{Column: \"updated_at\", OP: \">=\", Value: *opts.Since})\n\t\t}\n\t}\n\n\tlimit := 20\n\tif opts != nil && opts.Limit > 0 {\n\t\tlimit = opts.Limit\n\t\tif limit > 100 {\n\t\t\tlimit = 100\n\t\t}\n\t}\n\n\tparams := model.QueryParam{\n\t\tWheres: wheres,\n\t\tLimit: limit,\n\t\tOrders: []model.QueryOrder{{Column: \"updated_at\", Option: \"desc\"}},\n\t}\n\n\trows, err := mod.Get(params)\n\tif err != nil {\n\t\treturn nil, fmt.Errorf(\"failed to list activities: %w\", err)\n\t}\n\n\tactivities := make([]*Activity, 0, len(rows))\n\tfor _, row := range rows {\n\t\trecord, err := s.mapToRecord(row)\n\t\tif err != nil {\n\t\t\tcontinue\n\t\t}\n\n\t\tactivity := s.executionToActivity(record)\n\t\tif activity != nil {\n\t\t\tactivities = append(activities, activity)\n\t\t}\n\t}\n\n\treturn activities, nil\n}\n\n// executionToActivity converts an execution record to an activity\nfunc (s *ExecutionStore) executionToActivity(record *ExecutionRecord) *Activity {\n\tvar actType ActivityType\n\tvar message string\n\tvar timestamp time.Time\n\n\tswitch record.Status {\n\tcase types.ExecRunning:\n\t\tactType = ActivityExecutionStarted\n\t\tmessage = \"Started\"\n\t\tif record.StartTime != nil {\n\t\t\ttimestamp = *record.StartTime\n\t\t} else {\n\t\t\ttimestamp = time.Now()\n\t\t}\n\tcase types.ExecCompleted:\n\t\tactType = ActivityExecutionCompleted\n\t\tmessage = \"Completed\"\n\t\tif record.EndTime != nil {\n\t\t\ttimestamp = *record.EndTime\n\t\t} else if record.UpdatedAt != nil {\n\t\t\ttimestamp = *record.UpdatedAt\n\t\t} else {\n\t\t\ttimestamp = time.Now()\n\t\t}\n\tcase types.ExecFailed:\n\t\tactType = ActivityExecutionFailed\n\t\tmessage = \"Failed\"\n\t\tif record.Error != \"\" {\n\t\t\tmessage = \"Failed: \" + record.Error\n\t\t\t// Truncate long error messages\n\t\t\tif len(message) > 100 {\n\t\t\t\tmessage = message[:97] + \"...\"\n\t\t\t}\n\t\t}\n\t\tif record.EndTime != nil {\n\t\t\ttimestamp = *record.EndTime\n\t\t} else if record.UpdatedAt != nil {\n\t\t\ttimestamp = *record.UpdatedAt\n\t\t} else {\n\t\t\ttimestamp = time.Now()\n\t\t}\n\tcase types.ExecCancelled:\n\t\tactType = ActivityExecutionCancelled\n\t\tmessage = \"Cancelled\"\n\t\tif record.EndTime != nil {\n\t\t\ttimestamp = *record.EndTime\n\t\t} else if record.UpdatedAt != nil {\n\t\t\ttimestamp = *record.UpdatedAt\n\t\t} else {\n\t\t\ttimestamp = time.Now()\n\t\t}\n\tdefault:\n\t\treturn nil // Other statuses don't generate activities\n\t}\n\n\t// Add execution name to message if available\n\tif record.Name != \"\" {\n\t\tmessage = message + \": \" + record.Name\n\t\t// Truncate long messages\n\t\tif len(message) > 150 {\n\t\t\tmessage = message[:147] + \"...\"\n\t\t}\n\t}\n\n\treturn &Activity{\n\t\tType: actType,\n\t\tRobotID: record.MemberID,\n\t\tExecutionID: record.ExecutionID,\n\t\tMessage: message,\n\t\tTimestamp: timestamp,\n\t}\n}\n\n// FromExecution creates an ExecutionRecord from a runtime Execution\nfunc FromExecution(exec *types.Execution) *ExecutionRecord {\n\trecord := &ExecutionRecord{\n\t\tExecutionID: exec.ID,\n\t\tMemberID: exec.MemberID,\n\t\tTeamID: exec.TeamID,\n\t\tTriggerType: exec.TriggerType,\n\t\tStatus: exec.Status,\n\t\tPhase: exec.Phase,\n\t\tError: exec.Error,\n\t\tName: exec.Name,\n\t\tCurrentTaskName: exec.CurrentTaskName,\n\t\tInput: exec.Input,\n\t\tInspiration: exec.Inspiration,\n\t\tGoals: exec.Goals,\n\t\tTasks: exec.Tasks,\n\t\tResults: exec.Results,\n\t\tDelivery: exec.Delivery,\n\t\tLearning: exec.Learning,\n\t\tChatID: exec.ChatID,\n\t\tWaitingTaskID: exec.WaitingTaskID,\n\t\tWaitingQuestion: exec.WaitingQuestion,\n\t\tWaitingSince: exec.WaitingSince,\n\t\tResumeContext: exec.ResumeContext,\n\t}\n\n\t// Convert timestamps\n\tif !exec.StartTime.IsZero() {\n\t\trecord.StartTime = &exec.StartTime\n\t}\n\tif exec.EndTime != nil {\n\t\trecord.EndTime = exec.EndTime\n\t}\n\n\t// Convert CurrentState\n\tif exec.Current != nil {\n\t\trecord.Current = &CurrentState{\n\t\t\tTaskIndex: exec.Current.TaskIndex,\n\t\t\tProgress: exec.Current.Progress,\n\t\t}\n\t}\n\n\treturn record\n}\n\n// ToExecution converts an ExecutionRecord to a runtime Execution\nfunc (r *ExecutionRecord) ToExecution() *types.Execution {\n\texec := &types.Execution{\n\t\tID: r.ExecutionID,\n\t\tMemberID: r.MemberID,\n\t\tTeamID: r.TeamID,\n\t\tTriggerType: r.TriggerType,\n\t\tStatus: r.Status,\n\t\tPhase: r.Phase,\n\t\tError: r.Error,\n\t\tName: r.Name,\n\t\tCurrentTaskName: r.CurrentTaskName,\n\t\tInput: r.Input,\n\t\tInspiration: r.Inspiration,\n\t\tGoals: r.Goals,\n\t\tTasks: r.Tasks,\n\t\tResults: r.Results,\n\t\tDelivery: r.Delivery,\n\t\tLearning: r.Learning,\n\t\tChatID: r.ChatID,\n\t\tWaitingTaskID: r.WaitingTaskID,\n\t\tWaitingQuestion: r.WaitingQuestion,\n\t\tWaitingSince: r.WaitingSince,\n\t\tResumeContext: r.ResumeContext,\n\t}\n\n\t// Convert timestamps\n\tif r.StartTime != nil {\n\t\texec.StartTime = *r.StartTime\n\t}\n\tif r.EndTime != nil {\n\t\texec.EndTime = r.EndTime\n\t}\n\n\t// Convert CurrentState\n\tif r.Current != nil {\n\t\texec.Current = &types.CurrentState{\n\t\t\tTaskIndex: r.Current.TaskIndex,\n\t\t\tProgress: r.Current.Progress,\n\t\t}\n\t}\n\n\treturn exec\n}\n" }, { "path": "agent/robot/store/execution_test.go", "content": "package store_test\n\nimport (\n\t\"context\"\n\t\"testing\"\n\t\"time\"\n\n\t\"github.com/stretchr/testify/assert\"\n\t\"github.com/stretchr/testify/require\"\n\t\"github.com/yaoapp/gou/model\"\n\t\"github.com/yaoapp/yao/agent/robot/store\"\n\t\"github.com/yaoapp/yao/agent/robot/types\"\n\t\"github.com/yaoapp/yao/agent/testutils\"\n)\n\n// TestExecutionStoreSave tests creating and updating execution records\nfunc TestExecutionStoreSave(t *testing.T) {\n\tif testing.Short() {\n\t\tt.Skip(\"Skipping integration test\")\n\t}\n\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\t// Clean up any existing test data\n\tcleanupTestExecutions(t)\n\tdefer cleanupTestExecutions(t)\n\n\ts := store.NewExecutionStore()\n\tctx := context.Background()\n\n\tt.Run(\"creates_new_execution_record\", func(t *testing.T) {\n\t\tstartTime := time.Now()\n\t\trecord := &store.ExecutionRecord{\n\t\t\tExecutionID: \"exec_test_save_001\",\n\t\t\tMemberID: \"member_test_001\",\n\t\t\tTeamID: \"team_test_001\",\n\t\t\tTriggerType: types.TriggerClock,\n\t\t\tStatus: types.ExecPending,\n\t\t\tPhase: types.PhaseInspiration,\n\t\t\tStartTime: &startTime,\n\t\t}\n\n\t\terr := s.Save(ctx, record)\n\t\trequire.NoError(t, err)\n\n\t\t// Verify it was created\n\t\tsaved, err := s.Get(ctx, \"exec_test_save_001\")\n\t\trequire.NoError(t, err)\n\t\trequire.NotNil(t, saved)\n\n\t\tassert.Equal(t, \"exec_test_save_001\", saved.ExecutionID)\n\t\tassert.Equal(t, \"member_test_001\", saved.MemberID)\n\t\tassert.Equal(t, \"team_test_001\", saved.TeamID)\n\t\tassert.Equal(t, types.TriggerClock, saved.TriggerType)\n\t\tassert.Equal(t, types.ExecPending, saved.Status)\n\t\tassert.Equal(t, types.PhaseInspiration, saved.Phase)\n\t\tassert.NotNil(t, saved.StartTime)\n\t\tassert.NotNil(t, saved.CreatedAt)\n\t})\n\n\tt.Run(\"updates_existing_execution_record\", func(t *testing.T) {\n\t\t// First create a record\n\t\tstartTime := time.Now()\n\t\trecord := &store.ExecutionRecord{\n\t\t\tExecutionID: \"exec_test_save_002\",\n\t\t\tMemberID: \"member_test_002\",\n\t\t\tTeamID: \"team_test_002\",\n\t\t\tTriggerType: types.TriggerHuman,\n\t\t\tStatus: types.ExecPending,\n\t\t\tPhase: types.PhaseInspiration,\n\t\t\tStartTime: &startTime,\n\t\t}\n\n\t\terr := s.Save(ctx, record)\n\t\trequire.NoError(t, err)\n\n\t\t// Update the record\n\t\trecord.Status = types.ExecRunning\n\t\trecord.Phase = types.PhaseGoals\n\t\trecord.Goals = &types.Goals{Content: \"Test goals content\"}\n\n\t\terr = s.Save(ctx, record)\n\t\trequire.NoError(t, err)\n\n\t\t// Verify the update\n\t\tsaved, err := s.Get(ctx, \"exec_test_save_002\")\n\t\trequire.NoError(t, err)\n\t\trequire.NotNil(t, saved)\n\n\t\tassert.Equal(t, types.ExecRunning, saved.Status)\n\t\tassert.Equal(t, types.PhaseGoals, saved.Phase)\n\t\tassert.NotNil(t, saved.Goals)\n\t\tassert.Equal(t, \"Test goals content\", saved.Goals.Content)\n\t})\n}\n\n// TestExecutionStoreGet tests retrieving execution records\nfunc TestExecutionStoreGet(t *testing.T) {\n\tif testing.Short() {\n\t\tt.Skip(\"Skipping integration test\")\n\t}\n\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\tcleanupTestExecutions(t)\n\tdefer cleanupTestExecutions(t)\n\n\ts := store.NewExecutionStore()\n\tctx := context.Background()\n\n\t// Create a test record with all fields populated\n\tsetupTestExecution(t, s, ctx)\n\n\tt.Run(\"returns_existing_record\", func(t *testing.T) {\n\t\trecord, err := s.Get(ctx, \"exec_test_get_001\")\n\t\trequire.NoError(t, err)\n\t\trequire.NotNil(t, record)\n\n\t\tassert.Equal(t, \"exec_test_get_001\", record.ExecutionID)\n\t\tassert.Equal(t, \"member_test_get\", record.MemberID)\n\t\tassert.Equal(t, \"team_test_get\", record.TeamID)\n\t\tassert.Equal(t, types.TriggerClock, record.TriggerType)\n\t\tassert.Equal(t, types.ExecCompleted, record.Status)\n\t\tassert.Equal(t, types.PhaseDelivery, record.Phase)\n\n\t\t// Verify phase outputs\n\t\tassert.NotNil(t, record.Inspiration)\n\t\tassert.Equal(t, \"Test inspiration content\", record.Inspiration.Content)\n\t\tassert.NotNil(t, record.Goals)\n\t\tassert.Equal(t, \"Test goals content\", record.Goals.Content)\n\t\tassert.Len(t, record.Tasks, 2)\n\t\tassert.Equal(t, \"task_001\", record.Tasks[0].ID)\n\t\tassert.Len(t, record.Results, 2)\n\t\tassert.True(t, record.Results[0].Success)\n\t})\n\n\tt.Run(\"returns_nil_for_non_existent_record\", func(t *testing.T) {\n\t\trecord, err := s.Get(ctx, \"exec_non_existent\")\n\t\trequire.NoError(t, err)\n\t\tassert.Nil(t, record)\n\t})\n}\n\n// TestExecutionStoreList tests listing execution records with filters\nfunc TestExecutionStoreList(t *testing.T) {\n\tif testing.Short() {\n\t\tt.Skip(\"Skipping integration test\")\n\t}\n\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\tcleanupTestExecutions(t)\n\tdefer cleanupTestExecutions(t)\n\n\ts := store.NewExecutionStore()\n\tctx := context.Background()\n\n\t// Create multiple test records\n\tsetupTestExecutionsForList(t, s, ctx)\n\n\tt.Run(\"lists_all_records_without_filters\", func(t *testing.T) {\n\t\tresult, err := s.List(ctx, nil)\n\t\trequire.NoError(t, err)\n\t\tassert.GreaterOrEqual(t, len(result.Data), 4)\n\t})\n\n\tt.Run(\"filters_by_member_id\", func(t *testing.T) {\n\t\tresult, err := s.List(ctx, &store.ListOptions{\n\t\t\tMemberID: \"member_list_001\",\n\t\t})\n\t\trequire.NoError(t, err)\n\t\tassert.Equal(t, 2, len(result.Data))\n\t\tfor _, r := range result.Data {\n\t\t\tassert.Equal(t, \"member_list_001\", r.MemberID)\n\t\t}\n\t})\n\n\tt.Run(\"filters_by_team_id\", func(t *testing.T) {\n\t\tresult, err := s.List(ctx, &store.ListOptions{\n\t\t\tTeamID: \"team_list_001\",\n\t\t})\n\t\trequire.NoError(t, err)\n\t\tassert.Equal(t, 3, len(result.Data))\n\t\tfor _, r := range result.Data {\n\t\t\tassert.Equal(t, \"team_list_001\", r.TeamID)\n\t\t}\n\t})\n\n\tt.Run(\"filters_by_status\", func(t *testing.T) {\n\t\tresult, err := s.List(ctx, &store.ListOptions{\n\t\t\tStatus: types.ExecCompleted,\n\t\t})\n\t\trequire.NoError(t, err)\n\t\tassert.GreaterOrEqual(t, len(result.Data), 2)\n\t\tfor _, r := range result.Data {\n\t\t\tassert.Equal(t, types.ExecCompleted, r.Status)\n\t\t}\n\t})\n\n\tt.Run(\"filters_by_trigger_type\", func(t *testing.T) {\n\t\tresult, err := s.List(ctx, &store.ListOptions{\n\t\t\tTriggerType: types.TriggerHuman,\n\t\t})\n\t\trequire.NoError(t, err)\n\t\tassert.GreaterOrEqual(t, len(result.Data), 1)\n\t\tfor _, r := range result.Data {\n\t\t\tassert.Equal(t, types.TriggerHuman, r.TriggerType)\n\t\t}\n\t})\n\n\tt.Run(\"respects_pagesize\", func(t *testing.T) {\n\t\tresult, err := s.List(ctx, &store.ListOptions{\n\t\t\tPageSize: 2,\n\t\t})\n\t\trequire.NoError(t, err)\n\t\tassert.Equal(t, 2, len(result.Data))\n\t})\n\n\tt.Run(\"combines_multiple_filters\", func(t *testing.T) {\n\t\tresult, err := s.List(ctx, &store.ListOptions{\n\t\t\tTeamID: \"team_list_001\",\n\t\t\tStatus: types.ExecCompleted,\n\t\t})\n\t\trequire.NoError(t, err)\n\t\tassert.Equal(t, 2, len(result.Data))\n\t\tfor _, r := range result.Data {\n\t\t\tassert.Equal(t, \"team_list_001\", r.TeamID)\n\t\t\tassert.Equal(t, types.ExecCompleted, r.Status)\n\t\t}\n\t})\n}\n\n// TestExecutionStoreUpdatePhase tests updating phase and phase data\nfunc TestExecutionStoreUpdatePhase(t *testing.T) {\n\tif testing.Short() {\n\t\tt.Skip(\"Skipping integration test\")\n\t}\n\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\tcleanupTestExecutions(t)\n\tdefer cleanupTestExecutions(t)\n\n\ts := store.NewExecutionStore()\n\tctx := context.Background()\n\n\t// Create a base record\n\tstartTime := time.Now()\n\trecord := &store.ExecutionRecord{\n\t\tExecutionID: \"exec_test_phase_001\",\n\t\tMemberID: \"member_phase_001\",\n\t\tTeamID: \"team_phase_001\",\n\t\tTriggerType: types.TriggerClock,\n\t\tStatus: types.ExecRunning,\n\t\tPhase: types.PhaseInspiration,\n\t\tStartTime: &startTime,\n\t}\n\terr := s.Save(ctx, record)\n\trequire.NoError(t, err)\n\n\tt.Run(\"updates_inspiration_phase\", func(t *testing.T) {\n\t\tinspiration := &types.InspirationReport{\n\t\t\tContent: \"Updated inspiration content\",\n\t\t}\n\t\terr := s.UpdatePhase(ctx, \"exec_test_phase_001\", types.PhaseInspiration, inspiration)\n\t\trequire.NoError(t, err)\n\n\t\tsaved, err := s.Get(ctx, \"exec_test_phase_001\")\n\t\trequire.NoError(t, err)\n\t\tassert.Equal(t, types.PhaseInspiration, saved.Phase)\n\t\tassert.NotNil(t, saved.Inspiration)\n\t\tassert.Equal(t, \"Updated inspiration content\", saved.Inspiration.Content)\n\t})\n\n\tt.Run(\"updates_goals_phase\", func(t *testing.T) {\n\t\tgoals := &types.Goals{\n\t\t\tContent: \"Updated goals content\",\n\t\t}\n\t\terr := s.UpdatePhase(ctx, \"exec_test_phase_001\", types.PhaseGoals, goals)\n\t\trequire.NoError(t, err)\n\n\t\tsaved, err := s.Get(ctx, \"exec_test_phase_001\")\n\t\trequire.NoError(t, err)\n\t\tassert.Equal(t, types.PhaseGoals, saved.Phase)\n\t\tassert.NotNil(t, saved.Goals)\n\t\tassert.Equal(t, \"Updated goals content\", saved.Goals.Content)\n\t})\n\n\tt.Run(\"updates_tasks_phase\", func(t *testing.T) {\n\t\ttasks := []types.Task{\n\t\t\t{ID: \"task_phase_001\", ExecutorType: types.ExecutorAssistant},\n\t\t\t{ID: \"task_phase_002\", ExecutorType: types.ExecutorProcess},\n\t\t}\n\t\terr := s.UpdatePhase(ctx, \"exec_test_phase_001\", types.PhaseTasks, tasks)\n\t\trequire.NoError(t, err)\n\n\t\tsaved, err := s.Get(ctx, \"exec_test_phase_001\")\n\t\trequire.NoError(t, err)\n\t\tassert.Equal(t, types.PhaseTasks, saved.Phase)\n\t\tassert.Len(t, saved.Tasks, 2)\n\t\tassert.Equal(t, \"task_phase_001\", saved.Tasks[0].ID)\n\t})\n\n\tt.Run(\"updates_run_phase\", func(t *testing.T) {\n\t\tresults := []types.TaskResult{\n\t\t\t{TaskID: \"task_phase_001\", Success: true, Output: \"Result 1\"},\n\t\t\t{TaskID: \"task_phase_002\", Success: false, Error: \"Failed\"},\n\t\t}\n\t\terr := s.UpdatePhase(ctx, \"exec_test_phase_001\", types.PhaseRun, results)\n\t\trequire.NoError(t, err)\n\n\t\tsaved, err := s.Get(ctx, \"exec_test_phase_001\")\n\t\trequire.NoError(t, err)\n\t\tassert.Equal(t, types.PhaseRun, saved.Phase)\n\t\tassert.Len(t, saved.Results, 2)\n\t\tassert.True(t, saved.Results[0].Success)\n\t\tassert.False(t, saved.Results[1].Success)\n\t})\n\n\tt.Run(\"updates_delivery_phase\", func(t *testing.T) {\n\t\tdelivery := &types.DeliveryResult{\n\t\t\tSuccess: true,\n\t\t}\n\t\terr := s.UpdatePhase(ctx, \"exec_test_phase_001\", types.PhaseDelivery, delivery)\n\t\trequire.NoError(t, err)\n\n\t\tsaved, err := s.Get(ctx, \"exec_test_phase_001\")\n\t\trequire.NoError(t, err)\n\t\tassert.Equal(t, types.PhaseDelivery, saved.Phase)\n\t\tassert.NotNil(t, saved.Delivery)\n\t\tassert.True(t, saved.Delivery.Success)\n\t})\n\n\tt.Run(\"updates_learning_phase\", func(t *testing.T) {\n\t\tlearning := []types.LearningEntry{\n\t\t\t{Type: types.LearnExecution, Content: \"Learned something\"},\n\t\t}\n\t\terr := s.UpdatePhase(ctx, \"exec_test_phase_001\", types.PhaseLearning, learning)\n\t\trequire.NoError(t, err)\n\n\t\tsaved, err := s.Get(ctx, \"exec_test_phase_001\")\n\t\trequire.NoError(t, err)\n\t\tassert.Equal(t, types.PhaseLearning, saved.Phase)\n\t\tassert.Len(t, saved.Learning, 1)\n\t\tassert.Equal(t, \"Learned something\", saved.Learning[0].Content)\n\t})\n}\n\n// TestExecutionStoreUpdateStatus tests updating execution status\nfunc TestExecutionStoreUpdateStatus(t *testing.T) {\n\tif testing.Short() {\n\t\tt.Skip(\"Skipping integration test\")\n\t}\n\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\tcleanupTestExecutions(t)\n\tdefer cleanupTestExecutions(t)\n\n\ts := store.NewExecutionStore()\n\tctx := context.Background()\n\n\tt.Run(\"updates_status_to_running\", func(t *testing.T) {\n\t\tstartTime := time.Now()\n\t\trecord := &store.ExecutionRecord{\n\t\t\tExecutionID: \"exec_test_status_001\",\n\t\t\tMemberID: \"member_status_001\",\n\t\t\tTeamID: \"team_status_001\",\n\t\t\tTriggerType: types.TriggerClock,\n\t\t\tStatus: types.ExecPending,\n\t\t\tPhase: types.PhaseInspiration,\n\t\t\tStartTime: &startTime,\n\t\t}\n\t\terr := s.Save(ctx, record)\n\t\trequire.NoError(t, err)\n\n\t\terr = s.UpdateStatus(ctx, \"exec_test_status_001\", types.ExecRunning, \"\")\n\t\trequire.NoError(t, err)\n\n\t\tsaved, err := s.Get(ctx, \"exec_test_status_001\")\n\t\trequire.NoError(t, err)\n\t\tassert.Equal(t, types.ExecRunning, saved.Status)\n\t\tassert.Nil(t, saved.EndTime) // Should not set end_time for running\n\t})\n\n\tt.Run(\"updates_status_to_completed_with_end_time\", func(t *testing.T) {\n\t\tstartTime := time.Now()\n\t\trecord := &store.ExecutionRecord{\n\t\t\tExecutionID: \"exec_test_status_002\",\n\t\t\tMemberID: \"member_status_002\",\n\t\t\tTeamID: \"team_status_002\",\n\t\t\tTriggerType: types.TriggerHuman,\n\t\t\tStatus: types.ExecRunning,\n\t\t\tPhase: types.PhaseDelivery,\n\t\t\tStartTime: &startTime,\n\t\t}\n\t\terr := s.Save(ctx, record)\n\t\trequire.NoError(t, err)\n\n\t\terr = s.UpdateStatus(ctx, \"exec_test_status_002\", types.ExecCompleted, \"\")\n\t\trequire.NoError(t, err)\n\n\t\tsaved, err := s.Get(ctx, \"exec_test_status_002\")\n\t\trequire.NoError(t, err)\n\t\tassert.Equal(t, types.ExecCompleted, saved.Status)\n\t\tassert.NotNil(t, saved.EndTime) // Should set end_time for completed\n\t})\n\n\tt.Run(\"updates_status_to_failed_with_error\", func(t *testing.T) {\n\t\tstartTime := time.Now()\n\t\trecord := &store.ExecutionRecord{\n\t\t\tExecutionID: \"exec_test_status_003\",\n\t\t\tMemberID: \"member_status_003\",\n\t\t\tTeamID: \"team_status_003\",\n\t\t\tTriggerType: types.TriggerEvent,\n\t\t\tStatus: types.ExecRunning,\n\t\t\tPhase: types.PhaseRun,\n\t\t\tStartTime: &startTime,\n\t\t}\n\t\terr := s.Save(ctx, record)\n\t\trequire.NoError(t, err)\n\n\t\terr = s.UpdateStatus(ctx, \"exec_test_status_003\", types.ExecFailed, \"Task execution failed: timeout\")\n\t\trequire.NoError(t, err)\n\n\t\tsaved, err := s.Get(ctx, \"exec_test_status_003\")\n\t\trequire.NoError(t, err)\n\t\tassert.Equal(t, types.ExecFailed, saved.Status)\n\t\tassert.Equal(t, \"Task execution failed: timeout\", saved.Error)\n\t\tassert.NotNil(t, saved.EndTime) // Should set end_time for failed\n\t})\n\n\tt.Run(\"updates_status_to_cancelled\", func(t *testing.T) {\n\t\tstartTime := time.Now()\n\t\trecord := &store.ExecutionRecord{\n\t\t\tExecutionID: \"exec_test_status_004\",\n\t\t\tMemberID: \"member_status_004\",\n\t\t\tTeamID: \"team_status_004\",\n\t\t\tTriggerType: types.TriggerClock,\n\t\t\tStatus: types.ExecRunning,\n\t\t\tPhase: types.PhaseTasks,\n\t\t\tStartTime: &startTime,\n\t\t}\n\t\terr := s.Save(ctx, record)\n\t\trequire.NoError(t, err)\n\n\t\terr = s.UpdateStatus(ctx, \"exec_test_status_004\", types.ExecCancelled, \"User cancelled\")\n\t\trequire.NoError(t, err)\n\n\t\tsaved, err := s.Get(ctx, \"exec_test_status_004\")\n\t\trequire.NoError(t, err)\n\t\tassert.Equal(t, types.ExecCancelled, saved.Status)\n\t\tassert.Equal(t, \"User cancelled\", saved.Error)\n\t\tassert.NotNil(t, saved.EndTime) // Should set end_time for cancelled\n\t})\n}\n\n// TestExecutionStoreUpdateCurrent tests updating current state\nfunc TestExecutionStoreUpdateCurrent(t *testing.T) {\n\tif testing.Short() {\n\t\tt.Skip(\"Skipping integration test\")\n\t}\n\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\tcleanupTestExecutions(t)\n\tdefer cleanupTestExecutions(t)\n\n\ts := store.NewExecutionStore()\n\tctx := context.Background()\n\n\t// Create a base record\n\tstartTime := time.Now()\n\trecord := &store.ExecutionRecord{\n\t\tExecutionID: \"exec_test_current_001\",\n\t\tMemberID: \"member_current_001\",\n\t\tTeamID: \"team_current_001\",\n\t\tTriggerType: types.TriggerClock,\n\t\tStatus: types.ExecRunning,\n\t\tPhase: types.PhaseRun,\n\t\tStartTime: &startTime,\n\t}\n\terr := s.Save(ctx, record)\n\trequire.NoError(t, err)\n\n\tt.Run(\"updates_current_state\", func(t *testing.T) {\n\t\tcurrent := &store.CurrentState{\n\t\t\tTaskIndex: 2,\n\t\t\tProgress: \"3/5 tasks completed\",\n\t\t}\n\t\terr := s.UpdateCurrent(ctx, \"exec_test_current_001\", current)\n\t\trequire.NoError(t, err)\n\n\t\tsaved, err := s.Get(ctx, \"exec_test_current_001\")\n\t\trequire.NoError(t, err)\n\t\tassert.NotNil(t, saved.Current)\n\t\tassert.Equal(t, 2, saved.Current.TaskIndex)\n\t\tassert.Equal(t, \"3/5 tasks completed\", saved.Current.Progress)\n\t})\n}\n\n// TestExecutionStoreUpdateUIFields tests updating UI display fields\nfunc TestExecutionStoreUpdateUIFields(t *testing.T) {\n\tif testing.Short() {\n\t\tt.Skip(\"Skipping integration test\")\n\t}\n\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\tcleanupTestExecutions(t)\n\tdefer cleanupTestExecutions(t)\n\n\ts := store.NewExecutionStore()\n\tctx := context.Background()\n\n\t// Create a base record\n\tstartTime := time.Now()\n\trecord := &store.ExecutionRecord{\n\t\tExecutionID: \"exec_test_uifields_001\",\n\t\tMemberID: \"member_uifields_001\",\n\t\tTeamID: \"team_uifields_001\",\n\t\tTriggerType: types.TriggerHuman,\n\t\tStatus: types.ExecRunning,\n\t\tPhase: types.PhaseInspiration,\n\t\tStartTime: &startTime,\n\t}\n\terr := s.Save(ctx, record)\n\trequire.NoError(t, err)\n\n\tt.Run(\"updates_name_only\", func(t *testing.T) {\n\t\terr := s.UpdateUIFields(ctx, \"exec_test_uifields_001\", \"Analyze sales data\", \"\")\n\t\trequire.NoError(t, err)\n\n\t\tsaved, err := s.Get(ctx, \"exec_test_uifields_001\")\n\t\trequire.NoError(t, err)\n\t\tassert.Equal(t, \"Analyze sales data\", saved.Name)\n\t\tassert.Equal(t, \"\", saved.CurrentTaskName)\n\t})\n\n\tt.Run(\"updates_current_task_name_only\", func(t *testing.T) {\n\t\terr := s.UpdateUIFields(ctx, \"exec_test_uifields_001\", \"\", \"Analyzing context...\")\n\t\trequire.NoError(t, err)\n\n\t\tsaved, err := s.Get(ctx, \"exec_test_uifields_001\")\n\t\trequire.NoError(t, err)\n\t\tassert.Equal(t, \"Analyze sales data\", saved.Name) // Previous value retained\n\t\tassert.Equal(t, \"Analyzing context...\", saved.CurrentTaskName)\n\t})\n\n\tt.Run(\"updates_both_fields\", func(t *testing.T) {\n\t\terr := s.UpdateUIFields(ctx, \"exec_test_uifields_001\", \"Generate monthly report\", \"Task 1/3: Collect data\")\n\t\trequire.NoError(t, err)\n\n\t\tsaved, err := s.Get(ctx, \"exec_test_uifields_001\")\n\t\trequire.NoError(t, err)\n\t\tassert.Equal(t, \"Generate monthly report\", saved.Name)\n\t\tassert.Equal(t, \"Task 1/3: Collect data\", saved.CurrentTaskName)\n\t})\n\n\tt.Run(\"does_nothing_when_both_empty\", func(t *testing.T) {\n\t\t// Get current values\n\t\tbefore, err := s.Get(ctx, \"exec_test_uifields_001\")\n\t\trequire.NoError(t, err)\n\n\t\t// Update with empty strings\n\t\terr = s.UpdateUIFields(ctx, \"exec_test_uifields_001\", \"\", \"\")\n\t\trequire.NoError(t, err)\n\n\t\t// Values should remain unchanged\n\t\tafter, err := s.Get(ctx, \"exec_test_uifields_001\")\n\t\trequire.NoError(t, err)\n\t\tassert.Equal(t, before.Name, after.Name)\n\t\tassert.Equal(t, before.CurrentTaskName, after.CurrentTaskName)\n\t})\n\n\tt.Run(\"handles_chinese_content\", func(t *testing.T) {\n\t\terr := s.UpdateUIFields(ctx, \"exec_test_uifields_001\", \"生成月度报告\", \"任务 2/3: 分析数据\")\n\t\trequire.NoError(t, err)\n\n\t\tsaved, err := s.Get(ctx, \"exec_test_uifields_001\")\n\t\trequire.NoError(t, err)\n\t\tassert.Equal(t, \"生成月度报告\", saved.Name)\n\t\tassert.Equal(t, \"任务 2/3: 分析数据\", saved.CurrentTaskName)\n\t})\n\n\tt.Run(\"handles_long_content\", func(t *testing.T) {\n\t\tlongName := \"This is a very long execution name that might come from a detailed user instruction about what they want the robot to accomplish in this particular run cycle\"\n\t\tlongTask := \"Task 1/5: Processing a complex multi-step operation with various sub-tasks that need to be completed...\"\n\n\t\terr := s.UpdateUIFields(ctx, \"exec_test_uifields_001\", longName, longTask)\n\t\trequire.NoError(t, err)\n\n\t\tsaved, err := s.Get(ctx, \"exec_test_uifields_001\")\n\t\trequire.NoError(t, err)\n\t\tassert.Equal(t, longName, saved.Name)\n\t\tassert.Equal(t, longTask, saved.CurrentTaskName)\n\t})\n}\n\n// TestExecutionStoreUpdateTasks tests updating tasks array with status\nfunc TestExecutionStoreUpdateTasks(t *testing.T) {\n\tif testing.Short() {\n\t\tt.Skip(\"Skipping integration test\")\n\t}\n\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\tcleanupTestExecutions(t)\n\tdefer cleanupTestExecutions(t)\n\n\ts := store.NewExecutionStore()\n\tctx := context.Background()\n\n\t// Create a base record with initial tasks\n\tstartTime := time.Now()\n\trecord := &store.ExecutionRecord{\n\t\tExecutionID: \"exec_test_tasks_001\",\n\t\tMemberID: \"member_tasks_001\",\n\t\tTeamID: \"team_tasks_001\",\n\t\tTriggerType: types.TriggerClock,\n\t\tStatus: types.ExecRunning,\n\t\tPhase: types.PhaseRun,\n\t\tStartTime: &startTime,\n\t\tTasks: []types.Task{\n\t\t\t{ID: \"task_001\", ExecutorType: types.ExecutorAssistant, Status: types.TaskPending, Order: 0},\n\t\t\t{ID: \"task_002\", ExecutorType: types.ExecutorProcess, Status: types.TaskPending, Order: 1},\n\t\t\t{ID: \"task_003\", ExecutorType: types.ExecutorAssistant, Status: types.TaskPending, Order: 2},\n\t\t},\n\t}\n\terr := s.Save(ctx, record)\n\trequire.NoError(t, err)\n\n\tt.Run(\"updates_task_status_to_running\", func(t *testing.T) {\n\t\t// Update first task to running\n\t\ttasks := []types.Task{\n\t\t\t{ID: \"task_001\", ExecutorType: types.ExecutorAssistant, Status: types.TaskRunning, Order: 0},\n\t\t\t{ID: \"task_002\", ExecutorType: types.ExecutorProcess, Status: types.TaskPending, Order: 1},\n\t\t\t{ID: \"task_003\", ExecutorType: types.ExecutorAssistant, Status: types.TaskPending, Order: 2},\n\t\t}\n\t\tcurrent := &store.CurrentState{TaskIndex: 0, Progress: \"1/3 tasks\"}\n\n\t\terr := s.UpdateTasks(ctx, \"exec_test_tasks_001\", tasks, current)\n\t\trequire.NoError(t, err)\n\n\t\tsaved, err := s.Get(ctx, \"exec_test_tasks_001\")\n\t\trequire.NoError(t, err)\n\t\trequire.Len(t, saved.Tasks, 3)\n\n\t\tassert.Equal(t, types.TaskRunning, saved.Tasks[0].Status)\n\t\tassert.Equal(t, types.TaskPending, saved.Tasks[1].Status)\n\t\tassert.Equal(t, types.TaskPending, saved.Tasks[2].Status)\n\n\t\tassert.NotNil(t, saved.Current)\n\t\tassert.Equal(t, 0, saved.Current.TaskIndex)\n\t})\n\n\tt.Run(\"updates_task_status_to_completed\", func(t *testing.T) {\n\t\t// First task completed, second running\n\t\ttasks := []types.Task{\n\t\t\t{ID: \"task_001\", ExecutorType: types.ExecutorAssistant, Status: types.TaskCompleted, Order: 0},\n\t\t\t{ID: \"task_002\", ExecutorType: types.ExecutorProcess, Status: types.TaskRunning, Order: 1},\n\t\t\t{ID: \"task_003\", ExecutorType: types.ExecutorAssistant, Status: types.TaskPending, Order: 2},\n\t\t}\n\t\tcurrent := &store.CurrentState{TaskIndex: 1, Progress: \"2/3 tasks\"}\n\n\t\terr := s.UpdateTasks(ctx, \"exec_test_tasks_001\", tasks, current)\n\t\trequire.NoError(t, err)\n\n\t\tsaved, err := s.Get(ctx, \"exec_test_tasks_001\")\n\t\trequire.NoError(t, err)\n\n\t\tassert.Equal(t, types.TaskCompleted, saved.Tasks[0].Status)\n\t\tassert.Equal(t, types.TaskRunning, saved.Tasks[1].Status)\n\t\tassert.Equal(t, types.TaskPending, saved.Tasks[2].Status)\n\t\tassert.Equal(t, 1, saved.Current.TaskIndex)\n\t})\n\n\tt.Run(\"updates_task_status_to_failed_with_skipped\", func(t *testing.T) {\n\t\t// Second task failed, third skipped\n\t\ttasks := []types.Task{\n\t\t\t{ID: \"task_001\", ExecutorType: types.ExecutorAssistant, Status: types.TaskCompleted, Order: 0},\n\t\t\t{ID: \"task_002\", ExecutorType: types.ExecutorProcess, Status: types.TaskFailed, Order: 1},\n\t\t\t{ID: \"task_003\", ExecutorType: types.ExecutorAssistant, Status: types.TaskSkipped, Order: 2},\n\t\t}\n\t\tcurrent := &store.CurrentState{TaskIndex: 1, Progress: \"Failed at 2/3\"}\n\n\t\terr := s.UpdateTasks(ctx, \"exec_test_tasks_001\", tasks, current)\n\t\trequire.NoError(t, err)\n\n\t\tsaved, err := s.Get(ctx, \"exec_test_tasks_001\")\n\t\trequire.NoError(t, err)\n\n\t\tassert.Equal(t, types.TaskCompleted, saved.Tasks[0].Status)\n\t\tassert.Equal(t, types.TaskFailed, saved.Tasks[1].Status)\n\t\tassert.Equal(t, types.TaskSkipped, saved.Tasks[2].Status)\n\t})\n\n\tt.Run(\"updates_with_nil_current\", func(t *testing.T) {\n\t\t// All tasks completed, no current\n\t\ttasks := []types.Task{\n\t\t\t{ID: \"task_001\", ExecutorType: types.ExecutorAssistant, Status: types.TaskCompleted, Order: 0},\n\t\t\t{ID: \"task_002\", ExecutorType: types.ExecutorProcess, Status: types.TaskCompleted, Order: 1},\n\t\t\t{ID: \"task_003\", ExecutorType: types.ExecutorAssistant, Status: types.TaskCompleted, Order: 2},\n\t\t}\n\n\t\terr := s.UpdateTasks(ctx, \"exec_test_tasks_001\", tasks, nil)\n\t\trequire.NoError(t, err)\n\n\t\tsaved, err := s.Get(ctx, \"exec_test_tasks_001\")\n\t\trequire.NoError(t, err)\n\n\t\tassert.Equal(t, types.TaskCompleted, saved.Tasks[0].Status)\n\t\tassert.Equal(t, types.TaskCompleted, saved.Tasks[1].Status)\n\t\tassert.Equal(t, types.TaskCompleted, saved.Tasks[2].Status)\n\t})\n\n\tt.Run(\"preserves_task_description\", func(t *testing.T) {\n\t\t// Create a new record with descriptions\n\t\trecord2 := &store.ExecutionRecord{\n\t\t\tExecutionID: \"exec_test_tasks_002\",\n\t\t\tMemberID: \"member_tasks_002\",\n\t\t\tTeamID: \"team_tasks_002\",\n\t\t\tTriggerType: types.TriggerHuman,\n\t\t\tStatus: types.ExecRunning,\n\t\t\tPhase: types.PhaseRun,\n\t\t\tStartTime: &startTime,\n\t\t\tTasks: []types.Task{\n\t\t\t\t{ID: \"task_d01\", Description: \"Analyze data\", ExecutorType: types.ExecutorAssistant, Status: types.TaskPending, Order: 0},\n\t\t\t\t{ID: \"task_d02\", Description: \"Generate report\", ExecutorType: types.ExecutorAssistant, Status: types.TaskPending, Order: 1},\n\t\t\t},\n\t\t}\n\t\terr := s.Save(ctx, record2)\n\t\trequire.NoError(t, err)\n\n\t\t// Update status preserving description\n\t\ttasks := []types.Task{\n\t\t\t{ID: \"task_d01\", Description: \"Analyze data\", ExecutorType: types.ExecutorAssistant, Status: types.TaskCompleted, Order: 0},\n\t\t\t{ID: \"task_d02\", Description: \"Generate report\", ExecutorType: types.ExecutorAssistant, Status: types.TaskRunning, Order: 1},\n\t\t}\n\n\t\terr = s.UpdateTasks(ctx, \"exec_test_tasks_002\", tasks, &store.CurrentState{TaskIndex: 1})\n\t\trequire.NoError(t, err)\n\n\t\tsaved, err := s.Get(ctx, \"exec_test_tasks_002\")\n\t\trequire.NoError(t, err)\n\n\t\tassert.Equal(t, \"Analyze data\", saved.Tasks[0].Description)\n\t\tassert.Equal(t, \"Generate report\", saved.Tasks[1].Description)\n\t\tassert.Equal(t, types.TaskCompleted, saved.Tasks[0].Status)\n\t\tassert.Equal(t, types.TaskRunning, saved.Tasks[1].Status)\n\t})\n}\n\n// TestExecutionStoreDelete tests deleting execution records\nfunc TestExecutionStoreDelete(t *testing.T) {\n\tif testing.Short() {\n\t\tt.Skip(\"Skipping integration test\")\n\t}\n\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\tcleanupTestExecutions(t)\n\tdefer cleanupTestExecutions(t)\n\n\ts := store.NewExecutionStore()\n\tctx := context.Background()\n\n\tt.Run(\"deletes_existing_record\", func(t *testing.T) {\n\t\t// Create a record\n\t\tstartTime := time.Now()\n\t\trecord := &store.ExecutionRecord{\n\t\t\tExecutionID: \"exec_test_delete_001\",\n\t\t\tMemberID: \"member_delete_001\",\n\t\t\tTeamID: \"team_delete_001\",\n\t\t\tTriggerType: types.TriggerClock,\n\t\t\tStatus: types.ExecCompleted,\n\t\t\tPhase: types.PhaseDelivery,\n\t\t\tStartTime: &startTime,\n\t\t}\n\t\terr := s.Save(ctx, record)\n\t\trequire.NoError(t, err)\n\n\t\t// Verify it exists\n\t\tsaved, err := s.Get(ctx, \"exec_test_delete_001\")\n\t\trequire.NoError(t, err)\n\t\trequire.NotNil(t, saved)\n\n\t\t// Delete it\n\t\terr = s.Delete(ctx, \"exec_test_delete_001\")\n\t\trequire.NoError(t, err)\n\n\t\t// Verify it's gone\n\t\tsaved, err = s.Get(ctx, \"exec_test_delete_001\")\n\t\trequire.NoError(t, err)\n\t\tassert.Nil(t, saved)\n\t})\n\n\tt.Run(\"no_error_for_non_existent_record\", func(t *testing.T) {\n\t\terr := s.Delete(ctx, \"exec_non_existent\")\n\t\tassert.NoError(t, err)\n\t})\n}\n\n// TestExecutionRecordConversion tests conversion between ExecutionRecord and Execution\nfunc TestExecutionRecordConversion(t *testing.T) {\n\tif testing.Short() {\n\t\tt.Skip(\"Skipping integration test\")\n\t}\n\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\tt.Run(\"converts_from_execution\", func(t *testing.T) {\n\t\tnow := time.Now()\n\t\tendTime := now.Add(time.Hour)\n\t\texec := &types.Execution{\n\t\t\tID: \"exec_convert_001\",\n\t\t\tMemberID: \"member_convert_001\",\n\t\t\tTeamID: \"team_convert_001\",\n\t\t\tTriggerType: types.TriggerHuman,\n\t\t\tStatus: types.ExecCompleted,\n\t\t\tPhase: types.PhaseDelivery,\n\t\t\tStartTime: now,\n\t\t\tEndTime: &endTime,\n\t\t\tError: \"\",\n\t\t\tName: \"Analyze sales data\",\n\t\t\tCurrentTaskName: \"Task 1/3: Processing\",\n\t\t\tInspiration: &types.InspirationReport{Content: \"Test inspiration\"},\n\t\t\tGoals: &types.Goals{Content: \"Test goals\"},\n\t\t\tTasks: []types.Task{\n\t\t\t\t{ID: \"task_001\", ExecutorType: types.ExecutorAssistant},\n\t\t\t},\n\t\t\tResults: []types.TaskResult{\n\t\t\t\t{TaskID: \"task_001\", Success: true},\n\t\t\t},\n\t\t\tCurrent: &types.CurrentState{\n\t\t\t\tTaskIndex: 1,\n\t\t\t\tProgress: \"1/1 tasks\",\n\t\t\t},\n\t\t}\n\n\t\trecord := store.FromExecution(exec)\n\n\t\tassert.Equal(t, \"exec_convert_001\", record.ExecutionID)\n\t\tassert.Equal(t, \"member_convert_001\", record.MemberID)\n\t\tassert.Equal(t, \"team_convert_001\", record.TeamID)\n\t\tassert.Equal(t, types.TriggerHuman, record.TriggerType)\n\t\tassert.Equal(t, types.ExecCompleted, record.Status)\n\t\tassert.Equal(t, types.PhaseDelivery, record.Phase)\n\t\tassert.NotNil(t, record.StartTime)\n\t\tassert.NotNil(t, record.EndTime)\n\t\tassert.NotNil(t, record.Inspiration)\n\t\tassert.NotNil(t, record.Goals)\n\t\tassert.Len(t, record.Tasks, 1)\n\t\tassert.Len(t, record.Results, 1)\n\t\tassert.NotNil(t, record.Current)\n\t\tassert.Equal(t, 1, record.Current.TaskIndex)\n\t\t// Verify UI fields conversion\n\t\tassert.Equal(t, \"Analyze sales data\", record.Name)\n\t\tassert.Equal(t, \"Task 1/3: Processing\", record.CurrentTaskName)\n\t})\n\n\tt.Run(\"converts_to_execution\", func(t *testing.T) {\n\t\tnow := time.Now()\n\t\tendTime := now.Add(time.Hour)\n\t\trecord := &store.ExecutionRecord{\n\t\t\tExecutionID: \"exec_convert_002\",\n\t\t\tMemberID: \"member_convert_002\",\n\t\t\tTeamID: \"team_convert_002\",\n\t\t\tTriggerType: types.TriggerClock,\n\t\t\tStatus: types.ExecRunning,\n\t\t\tPhase: types.PhaseRun,\n\t\t\tStartTime: &now,\n\t\t\tEndTime: &endTime,\n\t\t\tName: \"定时执行\",\n\t\t\tCurrentTaskName: \"任务 1/2: 数据分析\",\n\t\t\tInspiration: &types.InspirationReport{Content: \"Test inspiration\"},\n\t\t\tGoals: &types.Goals{Content: \"Test goals\"},\n\t\t\tTasks: []types.Task{\n\t\t\t\t{ID: \"task_002\", ExecutorType: types.ExecutorProcess},\n\t\t\t},\n\t\t\tResults: []types.TaskResult{\n\t\t\t\t{TaskID: \"task_002\", Success: false, Error: \"Failed\"},\n\t\t\t},\n\t\t\tCurrent: &store.CurrentState{\n\t\t\t\tTaskIndex: 0,\n\t\t\t\tProgress: \"0/1 tasks\",\n\t\t\t},\n\t\t}\n\n\t\texec := record.ToExecution()\n\n\t\tassert.Equal(t, \"exec_convert_002\", exec.ID)\n\t\tassert.Equal(t, \"member_convert_002\", exec.MemberID)\n\t\tassert.Equal(t, \"team_convert_002\", exec.TeamID)\n\t\tassert.Equal(t, types.TriggerClock, exec.TriggerType)\n\t\tassert.Equal(t, types.ExecRunning, exec.Status)\n\t\tassert.Equal(t, types.PhaseRun, exec.Phase)\n\t\tassert.NotNil(t, exec.Inspiration)\n\t\tassert.NotNil(t, exec.Goals)\n\t\tassert.Len(t, exec.Tasks, 1)\n\t\tassert.Len(t, exec.Results, 1)\n\t\tassert.NotNil(t, exec.Current)\n\t\tassert.Equal(t, 0, exec.Current.TaskIndex)\n\t\t// Verify UI fields conversion\n\t\tassert.Equal(t, \"定时执行\", exec.Name)\n\t\tassert.Equal(t, \"任务 1/2: 数据分析\", exec.CurrentTaskName)\n\t})\n}\n\n// Helper functions\n\nfunc cleanupTestExecutions(t *testing.T) {\n\tmod := model.Select(\"__yao.agent.execution\")\n\tif mod == nil {\n\t\treturn\n\t}\n\n\t// Delete all test execution records\n\t_, err := mod.DeleteWhere(model.QueryParam{\n\t\tWheres: []model.QueryWhere{\n\t\t\t{Column: \"execution_id\", OP: \"like\", Value: \"exec_test_%\"},\n\t\t},\n\t})\n\tif err != nil {\n\t\tt.Logf(\"Warning: failed to cleanup test executions: %v\", err)\n\t}\n}\n\nfunc setupTestExecution(t *testing.T, s *store.ExecutionStore, ctx context.Context) {\n\tstartTime := time.Now().Add(-time.Hour)\n\tendTime := time.Now()\n\n\trecord := &store.ExecutionRecord{\n\t\tExecutionID: \"exec_test_get_001\",\n\t\tMemberID: \"member_test_get\",\n\t\tTeamID: \"team_test_get\",\n\t\tTriggerType: types.TriggerClock,\n\t\tStatus: types.ExecCompleted,\n\t\tPhase: types.PhaseDelivery,\n\t\tStartTime: &startTime,\n\t\tEndTime: &endTime,\n\t\tInspiration: &types.InspirationReport{\n\t\t\tContent: \"Test inspiration content\",\n\t\t},\n\t\tGoals: &types.Goals{\n\t\t\tContent: \"Test goals content\",\n\t\t},\n\t\tTasks: []types.Task{\n\t\t\t{ID: \"task_001\", ExecutorType: types.ExecutorAssistant, Status: types.TaskCompleted},\n\t\t\t{ID: \"task_002\", ExecutorType: types.ExecutorProcess, Status: types.TaskCompleted},\n\t\t},\n\t\tResults: []types.TaskResult{\n\t\t\t{TaskID: \"task_001\", Success: true, Output: \"Result 1\"},\n\t\t\t{TaskID: \"task_002\", Success: true, Output: \"Result 2\"},\n\t\t},\n\t\tDelivery: &types.DeliveryResult{\n\t\t\tSuccess: true,\n\t\t},\n\t\tLearning: []types.LearningEntry{\n\t\t\t{Type: types.LearnExecution, Content: \"Test learning\"},\n\t\t},\n\t}\n\n\terr := s.Save(ctx, record)\n\trequire.NoError(t, err)\n}\n\nfunc setupTestExecutionsForList(t *testing.T, s *store.ExecutionStore, ctx context.Context) {\n\tstartTime := time.Now()\n\n\trecords := []*store.ExecutionRecord{\n\t\t{\n\t\t\tExecutionID: \"exec_test_list_001\",\n\t\t\tMemberID: \"member_list_001\",\n\t\t\tTeamID: \"team_list_001\",\n\t\t\tTriggerType: types.TriggerClock,\n\t\t\tStatus: types.ExecCompleted,\n\t\t\tPhase: types.PhaseDelivery,\n\t\t\tStartTime: &startTime,\n\t\t},\n\t\t{\n\t\t\tExecutionID: \"exec_test_list_002\",\n\t\t\tMemberID: \"member_list_001\",\n\t\t\tTeamID: \"team_list_001\",\n\t\t\tTriggerType: types.TriggerClock,\n\t\t\tStatus: types.ExecCompleted,\n\t\t\tPhase: types.PhaseDelivery,\n\t\t\tStartTime: &startTime,\n\t\t},\n\t\t{\n\t\t\tExecutionID: \"exec_test_list_003\",\n\t\t\tMemberID: \"member_list_002\",\n\t\t\tTeamID: \"team_list_001\",\n\t\t\tTriggerType: types.TriggerHuman,\n\t\t\tStatus: types.ExecRunning,\n\t\t\tPhase: types.PhaseRun,\n\t\t\tStartTime: &startTime,\n\t\t},\n\t\t{\n\t\t\tExecutionID: \"exec_test_list_004\",\n\t\t\tMemberID: \"member_list_002\",\n\t\t\tTeamID: \"team_list_002\",\n\t\t\tTriggerType: types.TriggerEvent,\n\t\t\tStatus: types.ExecFailed,\n\t\t\tPhase: types.PhaseRun,\n\t\t\tStartTime: &startTime,\n\t\t\tError: \"Test error\",\n\t\t},\n\t}\n\n\tfor _, record := range records {\n\t\terr := s.Save(ctx, record)\n\t\trequire.NoError(t, err)\n\t}\n}\n\n// ==================== Results & Activities Tests ====================\n\n// TestExecutionStoreListResults tests listing execution results (deliveries)\nfunc TestExecutionStoreListResults(t *testing.T) {\n\tif testing.Short() {\n\t\tt.Skip(\"Skipping integration test\")\n\t}\n\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\tcleanupTestExecutions(t)\n\tdefer cleanupTestExecutions(t)\n\n\ts := store.NewExecutionStore()\n\tctx := context.Background()\n\n\t// Setup test data with delivery content\n\tsetupTestResultsData(t, s, ctx)\n\n\tt.Run(\"lists_results_without_filters\", func(t *testing.T) {\n\t\tresult, err := s.ListResults(ctx, &store.ResultListOptions{\n\t\t\tMemberID: \"member_result_001\",\n\t\t})\n\t\trequire.NoError(t, err)\n\t\trequire.NotNil(t, result)\n\t\tassert.Equal(t, 2, result.Total)\n\t\tassert.Len(t, result.Data, 2)\n\t\t// Should be ordered by end_time desc\n\t\tfor _, r := range result.Data {\n\t\t\tassert.NotNil(t, r.Delivery)\n\t\t\tassert.NotNil(t, r.Delivery.Content)\n\t\t}\n\t})\n\n\tt.Run(\"filters_by_trigger_type\", func(t *testing.T) {\n\t\tresult, err := s.ListResults(ctx, &store.ResultListOptions{\n\t\t\tMemberID: \"member_result_001\",\n\t\t\tTriggerType: types.TriggerClock,\n\t\t})\n\t\trequire.NoError(t, err)\n\t\trequire.NotNil(t, result)\n\t\tassert.Equal(t, 1, result.Total)\n\t\tassert.Len(t, result.Data, 1)\n\t\tassert.Equal(t, types.TriggerClock, result.Data[0].TriggerType)\n\t})\n\n\tt.Run(\"filters_by_keyword\", func(t *testing.T) {\n\t\tresult, err := s.ListResults(ctx, &store.ResultListOptions{\n\t\t\tMemberID: \"member_result_001\",\n\t\t\tKeyword: \"Weekly\",\n\t\t})\n\t\trequire.NoError(t, err)\n\t\trequire.NotNil(t, result)\n\t\t// Should match \"Weekly Sales Report\"\n\t\tassert.GreaterOrEqual(t, result.Total, 1)\n\t})\n\n\tt.Run(\"respects_pagination\", func(t *testing.T) {\n\t\tresult, err := s.ListResults(ctx, &store.ResultListOptions{\n\t\t\tMemberID: \"member_result_001\",\n\t\t\tPageSize: 1,\n\t\t\tPage: 1,\n\t\t})\n\t\trequire.NoError(t, err)\n\t\trequire.NotNil(t, result)\n\t\tassert.Equal(t, 1, len(result.Data))\n\t\tassert.Equal(t, 2, result.Total)\n\t\tassert.Equal(t, 1, result.Page)\n\t})\n\n\tt.Run(\"excludes_executions_without_delivery\", func(t *testing.T) {\n\t\tresult, err := s.ListResults(ctx, &store.ResultListOptions{\n\t\t\tMemberID: \"member_result_002\", // Has no delivery content\n\t\t})\n\t\trequire.NoError(t, err)\n\t\trequire.NotNil(t, result)\n\t\tassert.Equal(t, 0, result.Total)\n\t\tassert.Empty(t, result.Data)\n\t})\n}\n\n// TestExecutionStoreCountResults tests counting results\nfunc TestExecutionStoreCountResults(t *testing.T) {\n\tif testing.Short() {\n\t\tt.Skip(\"Skipping integration test\")\n\t}\n\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\tcleanupTestExecutions(t)\n\tdefer cleanupTestExecutions(t)\n\n\ts := store.NewExecutionStore()\n\tctx := context.Background()\n\n\t// Setup test data with delivery content\n\tsetupTestResultsData(t, s, ctx)\n\n\tt.Run(\"counts_all_results_for_member\", func(t *testing.T) {\n\t\tcount, err := s.CountResults(ctx, &store.ResultListOptions{\n\t\t\tMemberID: \"member_result_001\",\n\t\t})\n\t\trequire.NoError(t, err)\n\t\tassert.Equal(t, 2, count)\n\t})\n\n\tt.Run(\"counts_filtered_results\", func(t *testing.T) {\n\t\tcount, err := s.CountResults(ctx, &store.ResultListOptions{\n\t\t\tMemberID: \"member_result_001\",\n\t\t\tTriggerType: types.TriggerHuman,\n\t\t})\n\t\trequire.NoError(t, err)\n\t\tassert.Equal(t, 1, count)\n\t})\n\n\tt.Run(\"returns_zero_for_no_results\", func(t *testing.T) {\n\t\tcount, err := s.CountResults(ctx, &store.ResultListOptions{\n\t\t\tMemberID: \"member_result_002\",\n\t\t})\n\t\trequire.NoError(t, err)\n\t\tassert.Equal(t, 0, count)\n\t})\n}\n\n// TestExecutionStoreListActivities tests listing activities\nfunc TestExecutionStoreListActivities(t *testing.T) {\n\tif testing.Short() {\n\t\tt.Skip(\"Skipping integration test\")\n\t}\n\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\tcleanupTestExecutions(t)\n\tdefer cleanupTestExecutions(t)\n\n\ts := store.NewExecutionStore()\n\tctx := context.Background()\n\n\t// Setup test data\n\tsetupTestActivitiesData(t, s, ctx)\n\n\tt.Run(\"lists_activities_for_team\", func(t *testing.T) {\n\t\tactivities, err := s.ListActivities(ctx, &store.ActivityListOptions{\n\t\t\tTeamID: \"team_activity_001\",\n\t\t})\n\t\trequire.NoError(t, err)\n\t\tassert.GreaterOrEqual(t, len(activities), 3)\n\t})\n\n\tt.Run(\"respects_limit\", func(t *testing.T) {\n\t\tactivities, err := s.ListActivities(ctx, &store.ActivityListOptions{\n\t\t\tTeamID: \"team_activity_001\",\n\t\t\tLimit: 2,\n\t\t})\n\t\trequire.NoError(t, err)\n\t\tassert.LessOrEqual(t, len(activities), 2)\n\t})\n\n\tt.Run(\"filters_by_since\", func(t *testing.T) {\n\t\t// Without since, should get all activities\n\t\tactivitiesAll, err := s.ListActivities(ctx, &store.ActivityListOptions{\n\t\t\tTeamID: \"team_activity_001\",\n\t\t})\n\t\trequire.NoError(t, err)\n\t\tallCount := len(activitiesAll)\n\t\tassert.GreaterOrEqual(t, allCount, 3, \"should have at least 3 activities without filter\")\n\n\t\t// Use a time in the future to ensure we get no results\n\t\tfuture := time.Now().Add(24 * time.Hour)\n\t\tactivitiesFuture, err := s.ListActivities(ctx, &store.ActivityListOptions{\n\t\t\tTeamID: \"team_activity_001\",\n\t\t\tSince: &future,\n\t\t})\n\t\trequire.NoError(t, err)\n\t\tassert.Equal(t, 0, len(activitiesFuture), \"should get no results with future since time\")\n\t})\n\n\tt.Run(\"generates_correct_activity_types\", func(t *testing.T) {\n\t\tactivities, err := s.ListActivities(ctx, &store.ActivityListOptions{\n\t\t\tTeamID: \"team_activity_001\",\n\t\t})\n\t\trequire.NoError(t, err)\n\n\t\t// Should have activities of different types\n\t\ttypeCount := make(map[store.ActivityType]int)\n\t\tfor _, a := range activities {\n\t\t\ttypeCount[a.Type]++\n\t\t}\n\n\t\t// We should have at least completed and failed types\n\t\tassert.Greater(t, typeCount[store.ActivityExecutionCompleted], 0, \"should have completed activities\")\n\t\tassert.Greater(t, typeCount[store.ActivityExecutionFailed], 0, \"should have failed activities\")\n\t})\n\n\tt.Run(\"filters_by_type_completed\", func(t *testing.T) {\n\t\tactivities, err := s.ListActivities(ctx, &store.ActivityListOptions{\n\t\t\tTeamID: \"team_activity_001\",\n\t\t\tType: store.ActivityExecutionCompleted,\n\t\t})\n\t\trequire.NoError(t, err)\n\n\t\t// All returned activities should be of type completed\n\t\tfor _, a := range activities {\n\t\t\tassert.Equal(t, store.ActivityExecutionCompleted, a.Type, \"all activities should be completed type\")\n\t\t}\n\t\tassert.Greater(t, len(activities), 0, \"should have at least one completed activity\")\n\t})\n\n\tt.Run(\"filters_by_type_failed\", func(t *testing.T) {\n\t\tactivities, err := s.ListActivities(ctx, &store.ActivityListOptions{\n\t\t\tTeamID: \"team_activity_001\",\n\t\t\tType: store.ActivityExecutionFailed,\n\t\t})\n\t\trequire.NoError(t, err)\n\n\t\t// All returned activities should be of type failed\n\t\tfor _, a := range activities {\n\t\t\tassert.Equal(t, store.ActivityExecutionFailed, a.Type, \"all activities should be failed type\")\n\t\t}\n\t\tassert.Greater(t, len(activities), 0, \"should have at least one failed activity\")\n\t})\n\n\tt.Run(\"filters_by_type_invalid_returns_empty\", func(t *testing.T) {\n\t\tactivities, err := s.ListActivities(ctx, &store.ActivityListOptions{\n\t\t\tTeamID: \"team_activity_001\",\n\t\t\tType: store.ActivityType(\"invalid.type\"),\n\t\t})\n\t\trequire.NoError(t, err)\n\n\t\t// Invalid type should return empty result\n\t\tassert.Equal(t, 0, len(activities), \"invalid type should return empty result\")\n\t})\n\n\tt.Run(\"includes_execution_name_in_message\", func(t *testing.T) {\n\t\tactivities, err := s.ListActivities(ctx, &store.ActivityListOptions{\n\t\t\tTeamID: \"team_activity_001\",\n\t\t})\n\t\trequire.NoError(t, err)\n\n\t\t// Find a completed activity\n\t\tvar completedActivity *store.Activity\n\t\tfor _, a := range activities {\n\t\t\tif a.Type == store.ActivityExecutionCompleted && a.Message != \"\" {\n\t\t\t\tcompletedActivity = a\n\t\t\t\tbreak\n\t\t\t}\n\t\t}\n\n\t\trequire.NotNil(t, completedActivity, \"should find a completed activity\")\n\t\tassert.Contains(t, completedActivity.Message, \"Completed\")\n\t})\n}\n\n// Helper function to setup test results data\nfunc setupTestResultsData(t *testing.T, s *store.ExecutionStore, ctx context.Context) {\n\tstartTime := time.Now().Add(-2 * time.Hour)\n\tendTime := time.Now().Add(-1 * time.Hour)\n\tendTime2 := time.Now().Add(-30 * time.Minute)\n\n\trecords := []*store.ExecutionRecord{\n\t\t{\n\t\t\tExecutionID: \"exec_test_result_001\",\n\t\t\tMemberID: \"member_result_001\",\n\t\t\tTeamID: \"team_result_001\",\n\t\t\tTriggerType: types.TriggerClock,\n\t\t\tStatus: types.ExecCompleted,\n\t\t\tPhase: types.PhaseDelivery,\n\t\t\tName: \"Weekly Sales Report\",\n\t\t\tStartTime: &startTime,\n\t\t\tEndTime: &endTime,\n\t\t\tDelivery: &types.DeliveryResult{\n\t\t\t\tSuccess: true,\n\t\t\t\tContent: &types.DeliveryContent{\n\t\t\t\t\tSummary: \"Weekly sales report generated successfully\",\n\t\t\t\t\tBody: \"## Weekly Sales Report\\n\\nTotal sales: $50,000\",\n\t\t\t\t},\n\t\t\t},\n\t\t},\n\t\t{\n\t\t\tExecutionID: \"exec_test_result_002\",\n\t\t\tMemberID: \"member_result_001\",\n\t\t\tTeamID: \"team_result_001\",\n\t\t\tTriggerType: types.TriggerHuman,\n\t\t\tStatus: types.ExecCompleted,\n\t\t\tPhase: types.PhaseDelivery,\n\t\t\tName: \"Custom Analysis\",\n\t\t\tStartTime: &startTime,\n\t\t\tEndTime: &endTime2,\n\t\t\tDelivery: &types.DeliveryResult{\n\t\t\t\tSuccess: true,\n\t\t\t\tContent: &types.DeliveryContent{\n\t\t\t\t\tSummary: \"Custom analysis completed\",\n\t\t\t\t\tBody: \"## Analysis Results\\n\\nFindings...\",\n\t\t\t\t\tAttachments: []types.DeliveryAttachment{\n\t\t\t\t\t\t{Title: \"Report.pdf\", File: \"__attachment://file_001\"},\n\t\t\t\t\t},\n\t\t\t\t},\n\t\t\t},\n\t\t},\n\t\t{\n\t\t\t// Completed but no delivery content - should be excluded\n\t\t\tExecutionID: \"exec_test_result_003\",\n\t\t\tMemberID: \"member_result_002\",\n\t\t\tTeamID: \"team_result_001\",\n\t\t\tTriggerType: types.TriggerClock,\n\t\t\tStatus: types.ExecCompleted,\n\t\t\tPhase: types.PhaseDelivery,\n\t\t\tName: \"No Delivery Content\",\n\t\t\tStartTime: &startTime,\n\t\t\tEndTime: &endTime,\n\t\t\t// No Delivery field\n\t\t},\n\t\t{\n\t\t\t// Running - should be excluded from results\n\t\t\tExecutionID: \"exec_test_result_004\",\n\t\t\tMemberID: \"member_result_001\",\n\t\t\tTeamID: \"team_result_001\",\n\t\t\tTriggerType: types.TriggerClock,\n\t\t\tStatus: types.ExecRunning,\n\t\t\tPhase: types.PhaseRun,\n\t\t\tName: \"Running Task\",\n\t\t\tStartTime: &startTime,\n\t\t},\n\t}\n\n\tfor _, record := range records {\n\t\terr := s.Save(ctx, record)\n\t\trequire.NoError(t, err)\n\t}\n}\n\n// Helper function to setup test activities data\nfunc setupTestActivitiesData(t *testing.T, s *store.ExecutionStore, ctx context.Context) {\n\tstartTime := time.Now().Add(-2 * time.Hour)\n\tendTime := time.Now().Add(-1 * time.Hour)\n\tendTimeFailed := time.Now().Add(-45 * time.Minute)\n\n\trecords := []*store.ExecutionRecord{\n\t\t{\n\t\t\tExecutionID: \"exec_test_activity_001\",\n\t\t\tMemberID: \"member_activity_001\",\n\t\t\tTeamID: \"team_activity_001\",\n\t\t\tTriggerType: types.TriggerClock,\n\t\t\tStatus: types.ExecCompleted,\n\t\t\tPhase: types.PhaseDelivery,\n\t\t\tName: \"Daily Report\",\n\t\t\tStartTime: &startTime,\n\t\t\tEndTime: &endTime,\n\t\t},\n\t\t{\n\t\t\tExecutionID: \"exec_test_activity_002\",\n\t\t\tMemberID: \"member_activity_001\",\n\t\t\tTeamID: \"team_activity_001\",\n\t\t\tTriggerType: types.TriggerHuman,\n\t\t\tStatus: types.ExecFailed,\n\t\t\tPhase: types.PhaseRun,\n\t\t\tName: \"Custom Task\",\n\t\t\tStartTime: &startTime,\n\t\t\tEndTime: &endTimeFailed,\n\t\t\tError: \"Task timeout\",\n\t\t},\n\t\t{\n\t\t\tExecutionID: \"exec_test_activity_003\",\n\t\t\tMemberID: \"member_activity_002\",\n\t\t\tTeamID: \"team_activity_001\",\n\t\t\tTriggerType: types.TriggerEvent,\n\t\t\tStatus: types.ExecCancelled,\n\t\t\tPhase: types.PhaseTasks,\n\t\t\tName: \"Lead Processing\",\n\t\t\tStartTime: &startTime,\n\t\t\tEndTime: &endTime,\n\t\t},\n\t\t{\n\t\t\tExecutionID: \"exec_test_activity_004\",\n\t\t\tMemberID: \"member_activity_002\",\n\t\t\tTeamID: \"team_activity_001\",\n\t\t\tTriggerType: types.TriggerClock,\n\t\t\tStatus: types.ExecRunning,\n\t\t\tPhase: types.PhaseRun,\n\t\t\tName: \"Data Analysis\",\n\t\t\tStartTime: &startTime,\n\t\t},\n\t}\n\n\tfor _, record := range records {\n\t\terr := s.Save(ctx, record)\n\t\trequire.NoError(t, err)\n\t}\n}\n" }, { "path": "agent/robot/store/robot.go", "content": "package store\n\nimport (\n\t\"context\"\n\t\"fmt\"\n\t\"time\"\n\n\t\"github.com/yaoapp/gou/model\"\n\t\"github.com/yaoapp/kun/maps\"\n\t\"github.com/yaoapp/yao/agent/robot/types\"\n\t\"github.com/yaoapp/yao/agent/robot/utils\"\n)\n\n// RobotRecord - persistent storage for robot member\n// Maps to __yao.member model\ntype RobotRecord struct {\n\tID int64 `json:\"id,omitempty\"` // Auto-increment primary key\n\tMemberID string `json:\"member_id\"` // Unique robot identifier\n\tTeamID string `json:\"team_id\"` // Team ID\n\tMemberType string `json:\"member_type\"` // Always \"robot\" for robots\n\tStatus string `json:\"status\"` // Member status: active | inactive | pending | suspended\n\tRobotStatus string `json:\"robot_status\"` // Robot status: idle | working | paused | error | maintenance\n\tAutonomousMode bool `json:\"autonomous_mode\"` // Whether autonomous mode is enabled\n\n\t// Profile\n\tDisplayName string `json:\"display_name\"` // Display name\n\tBio string `json:\"bio,omitempty\"` // Robot description\n\tAvatar string `json:\"avatar,omitempty\"`\n\n\t// Identity & Role\n\tSystemPrompt string `json:\"system_prompt\"` // System prompt\n\tRoleID string `json:\"role_id\"` // Role within team\n\tManagerID string `json:\"manager_id\"` // Direct manager user_id (who manages this robot)\n\n\t// Communication\n\tRobotEmail string `json:\"robot_email\"` // Robot email address\n\tAuthorizedSenders interface{} `json:\"authorized_senders,omitempty\"` // Email whitelist (JSON array)\n\tEmailFilterRules interface{} `json:\"email_filter_rules,omitempty\"` // Email filter rules (JSON array)\n\n\t// Capabilities\n\tRobotConfig interface{} `json:\"robot_config\"` // Robot config JSON\n\tAgents interface{} `json:\"agents,omitempty\"` // Accessible agents (JSON array)\n\tMCPServers interface{} `json:\"mcp_servers,omitempty\"` // MCP servers (JSON array)\n\tLanguageModel string `json:\"language_model,omitempty\"` // Language model name\n\n\t// Limits\n\tCostLimit float64 `json:\"cost_limit,omitempty\"` // Monthly cost limit USD\n\n\t// Ownership & Audit\n\tInvitedBy string `json:\"invited_by,omitempty\"` // Who created/added this robot\n\tJoinedAt *time.Time `json:\"joined_at,omitempty\"` // When robot was created\n\n\t// Timestamps\n\tCreatedAt *time.Time `json:\"created_at,omitempty\"`\n\tUpdatedAt *time.Time `json:\"updated_at,omitempty\"`\n\n\t// Yao Permission Fields (automatically handled by Yao model when permission:true)\n\t// These fields are passed through to the model layer for permission control\n\tYaoCreatedBy string `json:\"__yao_created_by,omitempty\"` // Creator user_id (set on create)\n\tYaoUpdatedBy string `json:\"__yao_updated_by,omitempty\"` // Updater user_id (set on update)\n\tYaoTeamID string `json:\"__yao_team_id,omitempty\"` // Permission team scope\n\tYaoTenantID string `json:\"__yao_tenant_id,omitempty\"` // Permission tenant scope\n}\n\n// RobotListOptions - options for listing robot records\ntype RobotListOptions struct {\n\tTeamID string `json:\"team_id,omitempty\"`\n\tStatus types.RobotStatus `json:\"status,omitempty\"`\n\tKeywords string `json:\"keywords,omitempty\"` // Search in display_name\n\tLimit int `json:\"limit,omitempty\"`\n\tOffset int `json:\"offset,omitempty\"`\n\tPage int `json:\"page,omitempty\"`\n\tPageSize int `json:\"pagesize,omitempty\"`\n\tOrderBy string `json:\"order_by,omitempty\"`\n}\n\n// RobotStore - persistent storage for robot members\ntype RobotStore struct {\n\tmodelID string\n}\n\n// NewRobotStore creates a new robot store instance\nfunc NewRobotStore() *RobotStore {\n\treturn &RobotStore{\n\t\tmodelID: \"__yao.member\",\n\t}\n}\n\n// robotFields are the fields to select when loading robots\nvar robotFields = []interface{}{\n\t// Basic\n\t\"id\",\n\t\"member_id\",\n\t\"team_id\",\n\t\"member_type\",\n\t\"status\",\n\t\"robot_status\",\n\t\"autonomous_mode\",\n\n\t// Profile\n\t\"display_name\",\n\t\"bio\",\n\t\"avatar\",\n\n\t// Identity & Role\n\t\"system_prompt\",\n\t\"role_id\",\n\t\"manager_id\",\n\n\t// Communication\n\t\"robot_email\",\n\t\"authorized_senders\",\n\t\"email_filter_rules\",\n\n\t// Capabilities\n\t\"robot_config\",\n\t\"agents\",\n\t\"mcp_servers\",\n\t\"language_model\",\n\n\t// Limits\n\t\"cost_limit\",\n\n\t// Ownership & Audit\n\t\"invited_by\",\n\t\"joined_at\",\n\n\t// Timestamps\n\t\"created_at\",\n\t\"updated_at\",\n\n\t// Yao Permission Fields (for access control)\n\t\"__yao_created_by\",\n\t\"__yao_updated_by\",\n\t\"__yao_team_id\",\n\t\"__yao_tenant_id\",\n}\n\n// Save creates or updates a robot member record\nfunc (s *RobotStore) Save(ctx context.Context, record *RobotRecord) error {\n\tmod := model.Select(s.modelID)\n\tif mod == nil {\n\t\treturn fmt.Errorf(\"model %s not found\", s.modelID)\n\t}\n\n\t// Ensure member_type is robot\n\trecord.MemberType = \"robot\"\n\n\tdata := s.recordToMap(record)\n\n\t// Check if record exists by member_id\n\texisting, err := s.Get(ctx, record.MemberID)\n\tif err == nil && existing != nil {\n\t\t// Update existing record\n\t\t_, err = mod.UpdateWhere(\n\t\t\tmodel.QueryParam{\n\t\t\t\tWheres: []model.QueryWhere{\n\t\t\t\t\t{Column: \"member_id\", Value: record.MemberID},\n\t\t\t\t},\n\t\t\t},\n\t\t\tdata,\n\t\t)\n\t\tif err != nil {\n\t\t\treturn fmt.Errorf(\"failed to update robot record: %w\", err)\n\t\t}\n\t\treturn nil\n\t}\n\n\t// Create new record\n\t_, err = mod.Create(data)\n\tif err != nil {\n\t\treturn fmt.Errorf(\"failed to create robot record: %w\", err)\n\t}\n\treturn nil\n}\n\n// Get retrieves a robot record by member_id\nfunc (s *RobotStore) Get(ctx context.Context, memberID string) (*RobotRecord, error) {\n\tmod := model.Select(s.modelID)\n\tif mod == nil {\n\t\treturn nil, fmt.Errorf(\"model %s not found\", s.modelID)\n\t}\n\n\trows, err := mod.Get(model.QueryParam{\n\t\tSelect: robotFields,\n\t\tWheres: []model.QueryWhere{\n\t\t\t{Column: \"member_id\", Value: memberID},\n\t\t\t{Column: \"member_type\", Value: \"robot\"},\n\t\t},\n\t\tLimit: 1,\n\t})\n\tif err != nil {\n\t\treturn nil, fmt.Errorf(\"failed to get robot record: %w\", err)\n\t}\n\n\tif len(rows) == 0 {\n\t\treturn nil, nil\n\t}\n\n\treturn s.mapToRecord(rows[0])\n}\n\n// List retrieves robot records with filters\nfunc (s *RobotStore) List(ctx context.Context, opts *RobotListOptions) ([]*RobotRecord, int, error) {\n\tmod := model.Select(s.modelID)\n\tif mod == nil {\n\t\treturn nil, 0, fmt.Errorf(\"model %s not found\", s.modelID)\n\t}\n\n\t// Build where conditions - only require member_type=robot\n\twheres := []model.QueryWhere{\n\t\t{Column: \"member_type\", Value: \"robot\"},\n\t}\n\n\tif opts != nil {\n\t\tif opts.TeamID != \"\" {\n\t\t\twheres = append(wheres, model.QueryWhere{Column: \"team_id\", Value: opts.TeamID})\n\t\t}\n\t\tif opts.Status != \"\" {\n\t\t\twheres = append(wheres, model.QueryWhere{Column: \"robot_status\", Value: string(opts.Status)})\n\t\t}\n\t\tif opts.Keywords != \"\" {\n\t\t\twheres = append(wheres, model.QueryWhere{\n\t\t\t\tColumn: \"display_name\",\n\t\t\t\tOP: \"like\",\n\t\t\t\tValue: \"%\" + opts.Keywords + \"%\",\n\t\t\t})\n\t\t}\n\t}\n\n\t// Build order\n\torders := []model.QueryOrder{}\n\tif opts != nil && opts.OrderBy != \"\" {\n\t\torders = append(orders, model.QueryOrder{Column: opts.OrderBy})\n\t} else {\n\t\torders = append(orders, model.QueryOrder{Column: \"created_at\", Option: \"desc\"})\n\t}\n\n\t// Determine pagination\n\tpage := 1\n\tpageSize := 100\n\tif opts != nil {\n\t\tif opts.Page > 0 {\n\t\t\tpage = opts.Page\n\t\t}\n\t\tif opts.PageSize > 0 {\n\t\t\tpageSize = opts.PageSize\n\t\t}\n\t\t// Limit overrides PageSize for simple limit queries\n\t\tif opts.Limit > 0 {\n\t\t\tpageSize = opts.Limit\n\t\t}\n\t}\n\n\t// Execute paginated query\n\tresult, err := mod.Paginate(model.QueryParam{\n\t\tSelect: robotFields,\n\t\tWheres: wheres,\n\t\tOrders: orders,\n\t}, page, pageSize)\n\tif err != nil {\n\t\treturn nil, 0, fmt.Errorf(\"failed to list robots: %w\", err)\n\t}\n\n\t// Get total count\n\ttotal := 0\n\tif t, ok := result.Get(\"total\").(int); ok {\n\t\ttotal = t\n\t}\n\n\t// Parse records\n\trecords := []*RobotRecord{}\n\tdata := result.Get(\"data\")\n\tswitch rows := data.(type) {\n\tcase []maps.MapStr:\n\t\tfor _, row := range rows {\n\t\t\trecord, err := s.mapToRecord(map[string]interface{}(row))\n\t\t\tif err != nil {\n\t\t\t\tcontinue // skip invalid records\n\t\t\t}\n\t\t\trecords = append(records, record)\n\t\t}\n\tcase []map[string]interface{}:\n\t\tfor _, row := range rows {\n\t\t\trecord, err := s.mapToRecord(row)\n\t\t\tif err != nil {\n\t\t\t\tcontinue // skip invalid records\n\t\t\t}\n\t\t\trecords = append(records, record)\n\t\t}\n\t}\n\n\treturn records, total, nil\n}\n\n// Delete removes a robot member by member_id\nfunc (s *RobotStore) Delete(ctx context.Context, memberID string) error {\n\tmod := model.Select(s.modelID)\n\tif mod == nil {\n\t\treturn fmt.Errorf(\"model %s not found\", s.modelID)\n\t}\n\n\t_, err := mod.DeleteWhere(model.QueryParam{\n\t\tWheres: []model.QueryWhere{\n\t\t\t{Column: \"member_id\", Value: memberID},\n\t\t\t{Column: \"member_type\", Value: \"robot\"},\n\t\t},\n\t})\n\tif err != nil {\n\t\treturn fmt.Errorf(\"failed to delete robot record: %w\", err)\n\t}\n\n\treturn nil\n}\n\n// UpdateConfig updates only the robot_config field\nfunc (s *RobotStore) UpdateConfig(ctx context.Context, memberID string, config interface{}) error {\n\tmod := model.Select(s.modelID)\n\tif mod == nil {\n\t\treturn fmt.Errorf(\"model %s not found\", s.modelID)\n\t}\n\n\tdata := map[string]interface{}{\n\t\t\"robot_config\": config,\n\t}\n\n\t_, err := mod.UpdateWhere(\n\t\tmodel.QueryParam{\n\t\t\tWheres: []model.QueryWhere{\n\t\t\t\t{Column: \"member_id\", Value: memberID},\n\t\t\t\t{Column: \"member_type\", Value: \"robot\"},\n\t\t\t},\n\t\t},\n\t\tdata,\n\t)\n\tif err != nil {\n\t\treturn fmt.Errorf(\"failed to update robot config: %w\", err)\n\t}\n\n\treturn nil\n}\n\n// UpdateStatus updates the robot_status field\nfunc (s *RobotStore) UpdateStatus(ctx context.Context, memberID string, status types.RobotStatus) error {\n\tmod := model.Select(s.modelID)\n\tif mod == nil {\n\t\treturn fmt.Errorf(\"model %s not found\", s.modelID)\n\t}\n\n\tdata := map[string]interface{}{\n\t\t\"robot_status\": string(status),\n\t}\n\n\t_, err := mod.UpdateWhere(\n\t\tmodel.QueryParam{\n\t\t\tWheres: []model.QueryWhere{\n\t\t\t\t{Column: \"member_id\", Value: memberID},\n\t\t\t\t{Column: \"member_type\", Value: \"robot\"},\n\t\t\t},\n\t\t},\n\t\tdata,\n\t)\n\tif err != nil {\n\t\treturn fmt.Errorf(\"failed to update robot status: %w\", err)\n\t}\n\n\treturn nil\n}\n\n// recordToMap converts RobotRecord to map for model operations\nfunc (s *RobotStore) recordToMap(record *RobotRecord) map[string]interface{} {\n\tdata := map[string]interface{}{\n\t\t// Required fields\n\t\t\"member_id\": record.MemberID,\n\t\t\"team_id\": record.TeamID,\n\t\t\"member_type\": \"robot\",\n\t\t\"autonomous_mode\": record.AutonomousMode,\n\t}\n\n\t// Status\n\tif record.Status != \"\" {\n\t\tdata[\"status\"] = record.Status\n\t} else {\n\t\tdata[\"status\"] = \"active\"\n\t}\n\tif record.RobotStatus != \"\" {\n\t\tdata[\"robot_status\"] = record.RobotStatus\n\t} else {\n\t\tdata[\"robot_status\"] = \"idle\"\n\t}\n\n\t// Profile\n\tif record.DisplayName != \"\" {\n\t\tdata[\"display_name\"] = record.DisplayName\n\t}\n\tif record.Bio != \"\" {\n\t\tdata[\"bio\"] = record.Bio\n\t}\n\tif record.Avatar != \"\" {\n\t\tdata[\"avatar\"] = record.Avatar\n\t}\n\n\t// Identity & Role\n\tif record.SystemPrompt != \"\" {\n\t\tdata[\"system_prompt\"] = record.SystemPrompt\n\t}\n\tif record.RoleID != \"\" {\n\t\tdata[\"role_id\"] = record.RoleID\n\t}\n\tif record.ManagerID != \"\" {\n\t\tdata[\"manager_id\"] = record.ManagerID\n\t}\n\n\t// Communication\n\tif record.RobotEmail != \"\" {\n\t\tdata[\"robot_email\"] = record.RobotEmail\n\t}\n\tif record.AuthorizedSenders != nil {\n\t\tdata[\"authorized_senders\"] = record.AuthorizedSenders\n\t}\n\tif record.EmailFilterRules != nil {\n\t\tdata[\"email_filter_rules\"] = record.EmailFilterRules\n\t}\n\n\t// Capabilities\n\tif record.RobotConfig != nil {\n\t\tdata[\"robot_config\"] = record.RobotConfig\n\t}\n\tif record.Agents != nil {\n\t\tdata[\"agents\"] = record.Agents\n\t}\n\tif record.MCPServers != nil {\n\t\tdata[\"mcp_servers\"] = record.MCPServers\n\t}\n\tif record.LanguageModel != \"\" {\n\t\tdata[\"language_model\"] = record.LanguageModel\n\t}\n\n\t// Limits\n\tif record.CostLimit > 0 {\n\t\tdata[\"cost_limit\"] = record.CostLimit\n\t}\n\n\t// Ownership & Audit\n\tif record.InvitedBy != \"\" {\n\t\tdata[\"invited_by\"] = record.InvitedBy\n\t}\n\tif record.JoinedAt != nil {\n\t\t// Format time for Gou model (expects string format)\n\t\tdata[\"joined_at\"] = record.JoinedAt.Format(\"2006-01-02 15:04:05\")\n\t}\n\n\t// Yao Permission Fields - pass through for model layer\n\tif record.YaoCreatedBy != \"\" {\n\t\tdata[\"__yao_created_by\"] = record.YaoCreatedBy\n\t}\n\tif record.YaoUpdatedBy != \"\" {\n\t\tdata[\"__yao_updated_by\"] = record.YaoUpdatedBy\n\t}\n\tif record.YaoTeamID != \"\" {\n\t\tdata[\"__yao_team_id\"] = record.YaoTeamID\n\t}\n\tif record.YaoTenantID != \"\" {\n\t\tdata[\"__yao_tenant_id\"] = record.YaoTenantID\n\t}\n\n\treturn data\n}\n\n// mapToRecord converts a model row to RobotRecord\nfunc (s *RobotStore) mapToRecord(row map[string]interface{}) (*RobotRecord, error) {\n\trecord := &RobotRecord{}\n\n\t// Basic fields\n\tif v, ok := row[\"id\"]; ok {\n\t\tswitch id := v.(type) {\n\t\tcase float64:\n\t\t\trecord.ID = int64(id)\n\t\tcase int64:\n\t\t\trecord.ID = id\n\t\tcase int:\n\t\t\trecord.ID = int64(id)\n\t\t}\n\t}\n\tif v, ok := row[\"member_id\"].(string); ok {\n\t\trecord.MemberID = v\n\t}\n\tif v, ok := row[\"team_id\"].(string); ok {\n\t\trecord.TeamID = v\n\t}\n\tif v, ok := row[\"member_type\"].(string); ok {\n\t\trecord.MemberType = v\n\t}\n\tif v, ok := row[\"status\"].(string); ok {\n\t\trecord.Status = v\n\t}\n\tif v, ok := row[\"robot_status\"].(string); ok {\n\t\trecord.RobotStatus = v\n\t}\n\tif v, ok := row[\"autonomous_mode\"]; ok {\n\t\trecord.AutonomousMode = utils.ToBool(v)\n\t}\n\n\t// Profile\n\tif v, ok := row[\"display_name\"].(string); ok {\n\t\trecord.DisplayName = v\n\t}\n\tif v, ok := row[\"bio\"].(string); ok {\n\t\trecord.Bio = v\n\t}\n\tif v, ok := row[\"avatar\"].(string); ok {\n\t\trecord.Avatar = v\n\t}\n\n\t// Identity & Role\n\tif v, ok := row[\"system_prompt\"].(string); ok {\n\t\trecord.SystemPrompt = v\n\t}\n\tif v, ok := row[\"role_id\"].(string); ok {\n\t\trecord.RoleID = v\n\t}\n\tif v, ok := row[\"manager_id\"].(string); ok {\n\t\trecord.ManagerID = v\n\t}\n\n\t// Communication\n\tif v, ok := row[\"robot_email\"].(string); ok {\n\t\trecord.RobotEmail = v\n\t}\n\tif v := row[\"authorized_senders\"]; v != nil {\n\t\trecord.AuthorizedSenders = utils.ToJSONValue(v)\n\t}\n\tif v := row[\"email_filter_rules\"]; v != nil {\n\t\trecord.EmailFilterRules = utils.ToJSONValue(v)\n\t}\n\n\t// Capabilities\n\tif v := row[\"robot_config\"]; v != nil {\n\t\trecord.RobotConfig = utils.ToJSONValue(v)\n\t}\n\tif v := row[\"agents\"]; v != nil {\n\t\trecord.Agents = utils.ToJSONValue(v)\n\t}\n\tif v := row[\"mcp_servers\"]; v != nil {\n\t\trecord.MCPServers = utils.ToJSONValue(v)\n\t}\n\tif v, ok := row[\"language_model\"].(string); ok {\n\t\trecord.LanguageModel = v\n\t}\n\n\t// Limits\n\tif v := row[\"cost_limit\"]; v != nil {\n\t\trecord.CostLimit = utils.ToFloat64(v)\n\t}\n\n\t// Ownership & Audit\n\tif v, ok := row[\"invited_by\"].(string); ok {\n\t\trecord.InvitedBy = v\n\t}\n\tif v := row[\"joined_at\"]; v != nil {\n\t\trecord.JoinedAt = utils.ToTimestamp(v)\n\t}\n\n\t// Timestamps\n\tif v := row[\"created_at\"]; v != nil {\n\t\trecord.CreatedAt = utils.ToTimestamp(v)\n\t}\n\tif v := row[\"updated_at\"]; v != nil {\n\t\trecord.UpdatedAt = utils.ToTimestamp(v)\n\t}\n\n\t// Yao Permission Fields\n\tif v, ok := row[\"__yao_created_by\"].(string); ok {\n\t\trecord.YaoCreatedBy = v\n\t}\n\tif v, ok := row[\"__yao_updated_by\"].(string); ok {\n\t\trecord.YaoUpdatedBy = v\n\t}\n\tif v, ok := row[\"__yao_team_id\"].(string); ok {\n\t\trecord.YaoTeamID = v\n\t}\n\tif v, ok := row[\"__yao_tenant_id\"].(string); ok {\n\t\trecord.YaoTenantID = v\n\t}\n\n\treturn record, nil\n}\n\n// ToRobot converts a RobotRecord to types.Robot\nfunc (r *RobotRecord) ToRobot() (*types.Robot, error) {\n\trobot := &types.Robot{\n\t\tMemberID: r.MemberID,\n\t\tTeamID: r.TeamID,\n\t\tDisplayName: r.DisplayName,\n\t\tBio: r.Bio,\n\t\tSystemPrompt: r.SystemPrompt,\n\t\tAutonomousMode: r.AutonomousMode,\n\t\tRobotEmail: r.RobotEmail,\n\t}\n\n\t// Parse robot_status\n\tif r.RobotStatus != \"\" {\n\t\trobot.Status = types.RobotStatus(r.RobotStatus)\n\t} else {\n\t\trobot.Status = types.RobotIdle\n\t}\n\n\t// Parse robot_config\n\tif r.RobotConfig != nil {\n\t\tconfig, err := types.ParseConfig(r.RobotConfig)\n\t\tif err != nil {\n\t\t\treturn nil, fmt.Errorf(\"failed to parse robot_config: %w\", err)\n\t\t}\n\t\trobot.Config = config\n\t}\n\n\t// Ensure Config exists for merging agents/mcp_servers\n\tif robot.Config == nil {\n\t\trobot.Config = &types.Config{}\n\t}\n\tif robot.Config.Resources == nil {\n\t\trobot.Config.Resources = &types.Resources{}\n\t}\n\n\t// Merge agents from member table into Config.Resources.Agents\n\tif r.Agents != nil {\n\t\tagents := parseStringSlice(r.Agents)\n\t\tif len(agents) > 0 {\n\t\t\trobot.Config.Resources.Agents = agents\n\t\t}\n\t}\n\n\t// Merge mcp_servers from member table into Config.Resources.MCP\n\tif r.MCPServers != nil {\n\t\tmcpServers := parseStringSlice(r.MCPServers)\n\t\tif len(mcpServers) > 0 {\n\t\t\t// Convert string slice to MCPConfig slice (each server ID becomes an MCPConfig)\n\t\t\tfor _, serverID := range mcpServers {\n\t\t\t\trobot.Config.Resources.MCP = append(robot.Config.Resources.MCP, types.MCPConfig{\n\t\t\t\t\tID: serverID,\n\t\t\t\t\t// Tools empty means all tools available\n\t\t\t\t})\n\t\t\t}\n\t\t}\n\t}\n\n\treturn robot, nil\n}\n\n// parseStringSlice converts interface{} to []string\nfunc parseStringSlice(v interface{}) []string {\n\tif v == nil {\n\t\treturn nil\n\t}\n\tswitch val := v.(type) {\n\tcase []string:\n\t\treturn val\n\tcase []interface{}:\n\t\tresult := make([]string, 0, len(val))\n\t\tfor _, item := range val {\n\t\t\tif s, ok := item.(string); ok {\n\t\t\t\tresult = append(result, s)\n\t\t\t}\n\t\t}\n\t\treturn result\n\t}\n\treturn nil\n}\n\n// FromRobot creates a RobotRecord from types.Robot\nfunc FromRobot(robot *types.Robot) *RobotRecord {\n\trecord := &RobotRecord{\n\t\tMemberID: robot.MemberID,\n\t\tTeamID: robot.TeamID,\n\t\tDisplayName: robot.DisplayName,\n\t\tBio: robot.Bio,\n\t\tSystemPrompt: robot.SystemPrompt,\n\t\tRobotStatus: string(robot.Status),\n\t\tAutonomousMode: robot.AutonomousMode,\n\t\tRobotEmail: robot.RobotEmail,\n\t\tMemberType: \"robot\",\n\t\tStatus: \"active\",\n\t}\n\n\tif robot.Config != nil {\n\t\trecord.RobotConfig = robot.Config\n\t}\n\n\treturn record\n}\n" }, { "path": "agent/robot/store/robot_test.go", "content": "package store_test\n\nimport (\n\t\"context\"\n\t\"testing\"\n\t\"time\"\n\n\t\"github.com/stretchr/testify/assert\"\n\t\"github.com/stretchr/testify/require\"\n\t\"github.com/yaoapp/gou/model\"\n\t\"github.com/yaoapp/yao/agent/robot/store\"\n\t\"github.com/yaoapp/yao/agent/testutils\"\n)\n\n// TestRobotStoreSave tests creating and updating robot records\nfunc TestRobotStoreSave(t *testing.T) {\n\tif testing.Short() {\n\t\tt.Skip(\"Skipping integration test\")\n\t}\n\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\tcleanupTestRobots(t)\n\tdefer cleanupTestRobots(t)\n\n\ts := store.NewRobotStore()\n\tctx := context.Background()\n\n\tt.Run(\"creates_new_robot_record\", func(t *testing.T) {\n\t\tnow := time.Now()\n\t\trecord := &store.RobotRecord{\n\t\t\tMemberID: \"robot_test_save_001\",\n\t\t\tTeamID: \"team_test_001\",\n\t\t\tDisplayName: \"Test Robot 001\",\n\t\t\tBio: \"A test robot for save operations\",\n\t\t\tSystemPrompt: \"You are a helpful assistant\",\n\t\t\tStatus: \"active\",\n\t\t\tRobotStatus: \"idle\",\n\t\t\tAutonomousMode: true,\n\t\t\tRobotEmail: \"robot001@test.com\",\n\t\t\tJoinedAt: &now,\n\t\t}\n\n\t\terr := s.Save(ctx, record)\n\t\trequire.NoError(t, err)\n\n\t\t// Verify it was created\n\t\tsaved, err := s.Get(ctx, \"robot_test_save_001\")\n\t\trequire.NoError(t, err)\n\t\trequire.NotNil(t, saved)\n\n\t\tassert.Equal(t, \"robot_test_save_001\", saved.MemberID)\n\t\tassert.Equal(t, \"team_test_001\", saved.TeamID)\n\t\tassert.Equal(t, \"Test Robot 001\", saved.DisplayName)\n\t\tassert.Equal(t, \"A test robot for save operations\", saved.Bio)\n\t\tassert.Equal(t, \"You are a helpful assistant\", saved.SystemPrompt)\n\t\tassert.Equal(t, \"active\", saved.Status)\n\t\tassert.Equal(t, \"idle\", saved.RobotStatus)\n\t\tassert.True(t, saved.AutonomousMode)\n\t\tassert.Equal(t, \"robot001@test.com\", saved.RobotEmail)\n\t\tassert.Equal(t, \"robot\", saved.MemberType)\n\t\tassert.NotNil(t, saved.JoinedAt)\n\t})\n\n\tt.Run(\"updates_existing_robot_record\", func(t *testing.T) {\n\t\t// First create a record\n\t\trecord := &store.RobotRecord{\n\t\t\tMemberID: \"robot_test_save_002\",\n\t\t\tTeamID: \"team_test_002\",\n\t\t\tDisplayName: \"Original Name\",\n\t\t\tStatus: \"active\",\n\t\t\tRobotStatus: \"idle\",\n\t\t}\n\n\t\terr := s.Save(ctx, record)\n\t\trequire.NoError(t, err)\n\n\t\t// Update the record\n\t\trecord.DisplayName = \"Updated Name\"\n\t\trecord.Bio = \"Updated bio\"\n\t\trecord.RobotStatus = \"working\"\n\n\t\terr = s.Save(ctx, record)\n\t\trequire.NoError(t, err)\n\n\t\t// Verify the update\n\t\tsaved, err := s.Get(ctx, \"robot_test_save_002\")\n\t\trequire.NoError(t, err)\n\t\trequire.NotNil(t, saved)\n\n\t\tassert.Equal(t, \"Updated Name\", saved.DisplayName)\n\t\tassert.Equal(t, \"Updated bio\", saved.Bio)\n\t\tassert.Equal(t, \"working\", saved.RobotStatus)\n\t})\n\n\tt.Run(\"saves_robot_with_config\", func(t *testing.T) {\n\t\trecord := &store.RobotRecord{\n\t\t\tMemberID: \"robot_test_save_003\",\n\t\t\tTeamID: \"team_test_003\",\n\t\t\tDisplayName: \"Robot with Config\",\n\t\t\tStatus: \"active\",\n\t\t\tRobotStatus: \"idle\",\n\t\t\tRobotConfig: map[string]interface{}{\n\t\t\t\t\"clock_mode\": \"on\",\n\t\t\t\t\"max_concurrent\": 3,\n\t\t\t\t\"timeout_seconds\": 300,\n\t\t\t},\n\t\t}\n\n\t\terr := s.Save(ctx, record)\n\t\trequire.NoError(t, err)\n\n\t\tsaved, err := s.Get(ctx, \"robot_test_save_003\")\n\t\trequire.NoError(t, err)\n\t\trequire.NotNil(t, saved)\n\t\tassert.NotNil(t, saved.RobotConfig)\n\t})\n\n\tt.Run(\"saves_robot_with_permission_fields\", func(t *testing.T) {\n\t\trecord := &store.RobotRecord{\n\t\t\tMemberID: \"robot_test_save_004\",\n\t\t\tTeamID: \"team_test_004\",\n\t\t\tDisplayName: \"Robot with Perms\",\n\t\t\tStatus: \"active\",\n\t\t\tRobotStatus: \"idle\",\n\t\t\tYaoCreatedBy: \"user_001\",\n\t\t\tYaoTeamID: \"team_001\",\n\t\t\tYaoTenantID: \"tenant_001\",\n\t\t}\n\n\t\terr := s.Save(ctx, record)\n\t\trequire.NoError(t, err)\n\n\t\t// Yao permission fields are handled by the model layer\n\t\tsaved, err := s.Get(ctx, \"robot_test_save_004\")\n\t\trequire.NoError(t, err)\n\t\trequire.NotNil(t, saved)\n\t})\n}\n\n// TestRobotStoreGet tests retrieving robot records\nfunc TestRobotStoreGet(t *testing.T) {\n\tif testing.Short() {\n\t\tt.Skip(\"Skipping integration test\")\n\t}\n\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\tcleanupTestRobots(t)\n\tdefer cleanupTestRobots(t)\n\n\ts := store.NewRobotStore()\n\tctx := context.Background()\n\n\t// Create a test record\n\tsetupTestRobot(t, s, ctx)\n\n\tt.Run(\"returns_existing_record\", func(t *testing.T) {\n\t\trecord, err := s.Get(ctx, \"robot_test_get_001\")\n\t\trequire.NoError(t, err)\n\t\trequire.NotNil(t, record)\n\n\t\tassert.Equal(t, \"robot_test_get_001\", record.MemberID)\n\t\tassert.Equal(t, \"team_test_get\", record.TeamID)\n\t\tassert.Equal(t, \"Test Robot Get\", record.DisplayName)\n\t\tassert.Equal(t, \"Test robot description\", record.Bio)\n\t\tassert.Equal(t, \"robot\", record.MemberType)\n\t\tassert.Equal(t, \"active\", record.Status)\n\t\tassert.Equal(t, \"idle\", record.RobotStatus)\n\t})\n\n\tt.Run(\"returns_nil_for_non_existent_record\", func(t *testing.T) {\n\t\trecord, err := s.Get(ctx, \"robot_non_existent\")\n\t\trequire.NoError(t, err)\n\t\tassert.Nil(t, record)\n\t})\n\n\tt.Run(\"ignores_non_robot_members\", func(t *testing.T) {\n\t\t// Get should only return member_type=\"robot\" records\n\t\trecord, err := s.Get(ctx, \"robot_test_get_001\")\n\t\trequire.NoError(t, err)\n\t\trequire.NotNil(t, record)\n\t\tassert.Equal(t, \"robot\", record.MemberType)\n\t})\n}\n\n// TestRobotStoreList tests listing robot records with filters\nfunc TestRobotStoreList(t *testing.T) {\n\tif testing.Short() {\n\t\tt.Skip(\"Skipping integration test\")\n\t}\n\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\tcleanupTestRobots(t)\n\tdefer cleanupTestRobots(t)\n\n\ts := store.NewRobotStore()\n\tctx := context.Background()\n\n\t// Create multiple test records\n\tsetupTestRobotsForList(t, s, ctx)\n\n\tt.Run(\"lists_all_robot_records\", func(t *testing.T) {\n\t\t// List with keywords filter to only get our test records\n\t\t// Test robots have display names like \"Robot Alpha\", \"Robot Beta\", etc.\n\t\trecords, total, err := s.List(ctx, &store.RobotListOptions{\n\t\t\tKeywords: \"Robot\",\n\t\t})\n\t\trequire.NoError(t, err)\n\t\t// Should find at least our 4 test robots\n\t\tassert.GreaterOrEqual(t, len(records), 4)\n\t\tassert.GreaterOrEqual(t, total, 4)\n\t})\n\n\tt.Run(\"filters_by_team_id\", func(t *testing.T) {\n\t\trecords, total, err := s.List(ctx, &store.RobotListOptions{\n\t\t\tTeamID: \"team_list_001\",\n\t\t})\n\t\trequire.NoError(t, err)\n\t\tassert.Equal(t, 2, len(records))\n\t\tassert.Equal(t, 2, total)\n\t\tfor _, r := range records {\n\t\t\tassert.Equal(t, \"team_list_001\", r.TeamID)\n\t\t}\n\t})\n\n\tt.Run(\"filters_by_robot_status\", func(t *testing.T) {\n\t\trecords, _, err := s.List(ctx, &store.RobotListOptions{\n\t\t\tStatus: \"working\",\n\t\t})\n\t\trequire.NoError(t, err)\n\t\tassert.GreaterOrEqual(t, len(records), 1)\n\t\tfor _, r := range records {\n\t\t\tassert.Equal(t, \"working\", r.RobotStatus)\n\t\t}\n\t})\n\n\tt.Run(\"filters_by_keywords\", func(t *testing.T) {\n\t\trecords, _, err := s.List(ctx, &store.RobotListOptions{\n\t\t\tKeywords: \"Alpha\",\n\t\t})\n\t\trequire.NoError(t, err)\n\t\tassert.Equal(t, 1, len(records))\n\t\tassert.Contains(t, records[0].DisplayName, \"Alpha\")\n\t})\n\n\tt.Run(\"respects_pagination\", func(t *testing.T) {\n\t\trecords, total, err := s.List(ctx, &store.RobotListOptions{\n\t\t\tPage: 1,\n\t\t\tPageSize: 2,\n\t\t})\n\t\trequire.NoError(t, err)\n\t\tassert.Equal(t, 2, len(records))\n\t\tassert.GreaterOrEqual(t, total, 4) // total count should be full count\n\t})\n\n\tt.Run(\"respects_limit\", func(t *testing.T) {\n\t\trecords, _, err := s.List(ctx, &store.RobotListOptions{\n\t\t\tLimit: 2,\n\t\t})\n\t\trequire.NoError(t, err)\n\t\tassert.Equal(t, 2, len(records))\n\t})\n\n\tt.Run(\"combines_multiple_filters\", func(t *testing.T) {\n\t\trecords, total, err := s.List(ctx, &store.RobotListOptions{\n\t\t\tTeamID: \"team_list_001\",\n\t\t\tStatus: \"idle\",\n\t\t})\n\t\trequire.NoError(t, err)\n\t\tassert.Equal(t, 1, len(records))\n\t\tassert.Equal(t, 1, total)\n\t\tassert.Equal(t, \"team_list_001\", records[0].TeamID)\n\t\tassert.Equal(t, \"idle\", records[0].RobotStatus)\n\t})\n}\n\n// TestRobotStoreDelete tests deleting robot records\nfunc TestRobotStoreDelete(t *testing.T) {\n\tif testing.Short() {\n\t\tt.Skip(\"Skipping integration test\")\n\t}\n\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\tcleanupTestRobots(t)\n\tdefer cleanupTestRobots(t)\n\n\ts := store.NewRobotStore()\n\tctx := context.Background()\n\n\tt.Run(\"deletes_existing_record\", func(t *testing.T) {\n\t\t// Create a record\n\t\trecord := &store.RobotRecord{\n\t\t\tMemberID: \"robot_test_delete_001\",\n\t\t\tTeamID: \"team_delete_001\",\n\t\t\tDisplayName: \"Robot to Delete\",\n\t\t\tStatus: \"active\",\n\t\t\tRobotStatus: \"idle\",\n\t\t}\n\t\terr := s.Save(ctx, record)\n\t\trequire.NoError(t, err)\n\n\t\t// Verify it exists\n\t\tsaved, err := s.Get(ctx, \"robot_test_delete_001\")\n\t\trequire.NoError(t, err)\n\t\trequire.NotNil(t, saved)\n\n\t\t// Delete it\n\t\terr = s.Delete(ctx, \"robot_test_delete_001\")\n\t\trequire.NoError(t, err)\n\n\t\t// Verify it's gone\n\t\tsaved, err = s.Get(ctx, \"robot_test_delete_001\")\n\t\trequire.NoError(t, err)\n\t\tassert.Nil(t, saved)\n\t})\n\n\tt.Run(\"no_error_for_non_existent_record\", func(t *testing.T) {\n\t\terr := s.Delete(ctx, \"robot_non_existent\")\n\t\tassert.NoError(t, err)\n\t})\n}\n\n// TestRobotStoreUpdateConfig tests updating robot config\nfunc TestRobotStoreUpdateConfig(t *testing.T) {\n\tif testing.Short() {\n\t\tt.Skip(\"Skipping integration test\")\n\t}\n\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\tcleanupTestRobots(t)\n\tdefer cleanupTestRobots(t)\n\n\ts := store.NewRobotStore()\n\tctx := context.Background()\n\n\t// Create a base record\n\trecord := &store.RobotRecord{\n\t\tMemberID: \"robot_test_config_001\",\n\t\tTeamID: \"team_config_001\",\n\t\tDisplayName: \"Config Test Robot\",\n\t\tStatus: \"active\",\n\t\tRobotStatus: \"idle\",\n\t\tRobotConfig: map[string]interface{}{\n\t\t\t\"clock_mode\": \"off\",\n\t\t},\n\t}\n\terr := s.Save(ctx, record)\n\trequire.NoError(t, err)\n\n\tt.Run(\"updates_config_only\", func(t *testing.T) {\n\t\tnewConfig := map[string]interface{}{\n\t\t\t\"clock_mode\": \"on\",\n\t\t\t\"max_concurrent\": 5,\n\t\t\t\"timeout_seconds\": 600,\n\t\t}\n\t\terr := s.UpdateConfig(ctx, \"robot_test_config_001\", newConfig)\n\t\trequire.NoError(t, err)\n\n\t\tsaved, err := s.Get(ctx, \"robot_test_config_001\")\n\t\trequire.NoError(t, err)\n\t\trequire.NotNil(t, saved)\n\t\tassert.NotNil(t, saved.RobotConfig)\n\n\t\t// Display name should be unchanged\n\t\tassert.Equal(t, \"Config Test Robot\", saved.DisplayName)\n\t})\n}\n\n// TestRobotStoreUpdateStatus tests updating robot status\nfunc TestRobotStoreUpdateStatus(t *testing.T) {\n\tif testing.Short() {\n\t\tt.Skip(\"Skipping integration test\")\n\t}\n\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\tcleanupTestRobots(t)\n\tdefer cleanupTestRobots(t)\n\n\ts := store.NewRobotStore()\n\tctx := context.Background()\n\n\t// Create a base record\n\trecord := &store.RobotRecord{\n\t\tMemberID: \"robot_test_status_001\",\n\t\tTeamID: \"team_status_001\",\n\t\tDisplayName: \"Status Test Robot\",\n\t\tStatus: \"active\",\n\t\tRobotStatus: \"idle\",\n\t}\n\terr := s.Save(ctx, record)\n\trequire.NoError(t, err)\n\n\tt.Run(\"updates_robot_status\", func(t *testing.T) {\n\t\terr := s.UpdateStatus(ctx, \"robot_test_status_001\", \"working\")\n\t\trequire.NoError(t, err)\n\n\t\tsaved, err := s.Get(ctx, \"robot_test_status_001\")\n\t\trequire.NoError(t, err)\n\t\trequire.NotNil(t, saved)\n\t\tassert.Equal(t, \"working\", saved.RobotStatus)\n\t\t// Display name should be unchanged\n\t\tassert.Equal(t, \"Status Test Robot\", saved.DisplayName)\n\t})\n\n\tt.Run(\"updates_to_paused\", func(t *testing.T) {\n\t\terr := s.UpdateStatus(ctx, \"robot_test_status_001\", \"paused\")\n\t\trequire.NoError(t, err)\n\n\t\tsaved, err := s.Get(ctx, \"robot_test_status_001\")\n\t\trequire.NoError(t, err)\n\t\tassert.Equal(t, \"paused\", saved.RobotStatus)\n\t})\n\n\tt.Run(\"updates_to_error\", func(t *testing.T) {\n\t\terr := s.UpdateStatus(ctx, \"robot_test_status_001\", \"error\")\n\t\trequire.NoError(t, err)\n\n\t\tsaved, err := s.Get(ctx, \"robot_test_status_001\")\n\t\trequire.NoError(t, err)\n\t\tassert.Equal(t, \"error\", saved.RobotStatus)\n\t})\n}\n\n// TestRobotRecordConversion tests conversion between RobotRecord and Robot types\nfunc TestRobotRecordConversion(t *testing.T) {\n\tif testing.Short() {\n\t\tt.Skip(\"Skipping integration test\")\n\t}\n\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\tt.Run(\"converts_record_to_robot\", func(t *testing.T) {\n\t\tnow := time.Now()\n\t\trecord := &store.RobotRecord{\n\t\t\tMemberID: \"robot_convert_001\",\n\t\t\tTeamID: \"team_convert_001\",\n\t\t\tDisplayName: \"Conversion Test Robot\",\n\t\t\tBio: \"Test description\",\n\t\t\tSystemPrompt: \"You are helpful\",\n\t\t\tStatus: \"active\",\n\t\t\tRobotStatus: \"idle\",\n\t\t\tAutonomousMode: true,\n\t\t\tRobotEmail: \"convert@test.com\",\n\t\t\tJoinedAt: &now,\n\t\t\tRobotConfig: map[string]interface{}{\n\t\t\t\t\"clock_mode\": \"on\",\n\t\t\t},\n\t\t}\n\n\t\trobot, err := record.ToRobot()\n\t\trequire.NoError(t, err)\n\t\trequire.NotNil(t, robot)\n\n\t\tassert.Equal(t, \"robot_convert_001\", robot.MemberID)\n\t\tassert.Equal(t, \"team_convert_001\", robot.TeamID)\n\t\tassert.Equal(t, \"Conversion Test Robot\", robot.DisplayName)\n\t\tassert.Equal(t, \"Test description\", robot.Bio)\n\t\tassert.Equal(t, \"You are helpful\", robot.SystemPrompt)\n\t\tassert.True(t, robot.AutonomousMode)\n\t\tassert.Equal(t, \"convert@test.com\", robot.RobotEmail)\n\t})\n\n\tt.Run(\"converts_robot_to_record\", func(t *testing.T) {\n\t\trobot := &store.RobotRecord{\n\t\t\tMemberID: \"robot_from_001\",\n\t\t\tTeamID: \"team_from_001\",\n\t\t\tDisplayName: \"From Robot Test\",\n\t\t\tBio: \"From robot description\",\n\t\t\tSystemPrompt: \"System prompt\",\n\t\t\tRobotStatus: \"working\",\n\t\t\tAutonomousMode: false,\n\t\t\tRobotEmail: \"from@test.com\",\n\t\t}\n\n\t\t// ToRobot and verify\n\t\tconverted, err := robot.ToRobot()\n\t\trequire.NoError(t, err)\n\n\t\tassert.Equal(t, \"robot_from_001\", converted.MemberID)\n\t\tassert.Equal(t, \"team_from_001\", converted.TeamID)\n\t\tassert.Equal(t, \"From Robot Test\", converted.DisplayName)\n\t})\n}\n\n// Helper functions\n\nfunc cleanupTestRobots(t *testing.T) {\n\tmod := model.Select(\"__yao.member\")\n\tif mod == nil {\n\t\treturn\n\t}\n\n\t// Delete all test robot records\n\t_, err := mod.DeleteWhere(model.QueryParam{\n\t\tWheres: []model.QueryWhere{\n\t\t\t{Column: \"member_id\", OP: \"like\", Value: \"robot_test_%\"},\n\t\t\t{Column: \"member_type\", Value: \"robot\"},\n\t\t},\n\t})\n\tif err != nil {\n\t\tt.Logf(\"Warning: failed to cleanup test robots: %v\", err)\n\t}\n}\n\nfunc setupTestRobot(t *testing.T, s *store.RobotStore, ctx context.Context) {\n\tnow := time.Now()\n\trecord := &store.RobotRecord{\n\t\tMemberID: \"robot_test_get_001\",\n\t\tTeamID: \"team_test_get\",\n\t\tDisplayName: \"Test Robot Get\",\n\t\tBio: \"Test robot description\",\n\t\tSystemPrompt: \"You are a test assistant\",\n\t\tStatus: \"active\",\n\t\tRobotStatus: \"idle\",\n\t\tAutonomousMode: false,\n\t\tRobotEmail: \"test@robot.com\",\n\t\tJoinedAt: &now,\n\t}\n\n\terr := s.Save(ctx, record)\n\trequire.NoError(t, err)\n}\n\nfunc setupTestRobotsForList(t *testing.T, s *store.RobotStore, ctx context.Context) {\n\tnow := time.Now()\n\n\trecords := []*store.RobotRecord{\n\t\t{\n\t\t\tMemberID: \"robot_test_list_001\",\n\t\t\tTeamID: \"team_list_001\",\n\t\t\tDisplayName: \"Robot Alpha\",\n\t\t\tStatus: \"active\",\n\t\t\tRobotStatus: \"idle\",\n\t\t\tJoinedAt: &now,\n\t\t},\n\t\t{\n\t\t\tMemberID: \"robot_test_list_002\",\n\t\t\tTeamID: \"team_list_001\",\n\t\t\tDisplayName: \"Robot Beta\",\n\t\t\tStatus: \"active\",\n\t\t\tRobotStatus: \"working\",\n\t\t\tJoinedAt: &now,\n\t\t},\n\t\t{\n\t\t\tMemberID: \"robot_test_list_003\",\n\t\t\tTeamID: \"team_list_002\",\n\t\t\tDisplayName: \"Robot Gamma\",\n\t\t\tStatus: \"active\",\n\t\t\tRobotStatus: \"idle\",\n\t\t\tJoinedAt: &now,\n\t\t},\n\t\t{\n\t\t\tMemberID: \"robot_test_list_004\",\n\t\t\tTeamID: \"team_list_002\",\n\t\t\tDisplayName: \"Robot Delta\",\n\t\t\tStatus: \"inactive\",\n\t\t\tRobotStatus: \"paused\",\n\t\t\tJoinedAt: &now,\n\t\t},\n\t}\n\n\tfor _, record := range records {\n\t\terr := s.Save(ctx, record)\n\t\trequire.NoError(t, err)\n\t}\n}\n" }, { "path": "agent/robot/store/store.go", "content": "package store\n\nimport \"github.com/yaoapp/yao/agent/robot/types\"\n\n// Store implements types.Store interface\n// This is a stub implementation for Phase 2\ntype Store struct{}\n\n// New creates a new store instance\nfunc New() *Store {\n\treturn &Store{}\n}\n\n// SaveLearning saves learning entries to private KB\n// Stub: returns nil (will be implemented in Phase 9)\nfunc (s *Store) SaveLearning(ctx *types.Context, memberID string, entries []types.LearningEntry) error {\n\treturn nil\n}\n\n// GetHistory retrieves learning history from private KB\n// Stub: returns empty slice (will be implemented in Phase 9)\nfunc (s *Store) GetHistory(ctx *types.Context, memberID string, limit int) ([]types.LearningEntry, error) {\n\treturn []types.LearningEntry{}, nil\n}\n\n// SearchKB searches knowledge base collections\n// Stub: returns empty slice (will be implemented in Phase 4+)\nfunc (s *Store) SearchKB(ctx *types.Context, collections []string, query string) ([]interface{}, error) {\n\treturn []interface{}{}, nil\n}\n\n// QueryDB queries database models\n// Stub: returns empty slice (will be implemented in Phase 4+)\nfunc (s *Store) QueryDB(ctx *types.Context, models []string, query interface{}) ([]interface{}, error) {\n\treturn []interface{}{}, nil\n}\n" }, { "path": "agent/robot/trigger/clock.go", "content": "package trigger\n\nimport (\n\t\"time\"\n\n\t\"github.com/yaoapp/yao/agent/robot/types\"\n)\n\n// ClockMatcher provides clock trigger matching logic\n// This is extracted from Manager for reuse and testing\ntype ClockMatcher struct{}\n\n// NewClockMatcher creates a new clock matcher\nfunc NewClockMatcher() *ClockMatcher {\n\treturn &ClockMatcher{}\n}\n\n// ShouldTrigger checks if a robot should be triggered based on its clock config\nfunc (cm *ClockMatcher) ShouldTrigger(robot *types.Robot, now time.Time) bool {\n\tif robot == nil || robot.Config == nil || robot.Config.Clock == nil {\n\t\treturn false\n\t}\n\n\tclock := robot.Config.Clock\n\n\t// Get time in robot's timezone\n\tloc := clock.GetLocation()\n\tlocalNow := now.In(loc)\n\n\tswitch clock.Mode {\n\tcase types.ClockTimes:\n\t\treturn cm.shouldTriggerTimes(robot, clock, localNow)\n\tcase types.ClockInterval:\n\t\treturn cm.shouldTriggerInterval(robot, clock, localNow)\n\tcase types.ClockDaemon:\n\t\treturn cm.shouldTriggerDaemon(robot, clock, localNow)\n\tdefault:\n\t\treturn false\n\t}\n}\n\n// shouldTriggerTimes checks if current time matches any configured times\n// times mode: run at specific times (e.g., [\"09:00\", \"14:00\", \"17:00\"])\nfunc (cm *ClockMatcher) shouldTriggerTimes(robot *types.Robot, clock *types.Clock, now time.Time) bool {\n\t// Check day of week first\n\tif !cm.matchesDay(clock, now) {\n\t\treturn false\n\t}\n\n\t// Check if current time matches any configured time\n\tcurrentTime := now.Format(\"15:04\")\n\tfor _, t := range clock.Times {\n\t\tif t == currentTime {\n\t\t\t// Check if already triggered in this minute\n\t\t\tif !robot.LastRun.IsZero() {\n\t\t\t\tlastRunInLoc := robot.LastRun.In(now.Location())\n\t\t\t\tif lastRunInLoc.Format(\"15:04\") == currentTime && lastRunInLoc.Day() == now.Day() {\n\t\t\t\t\treturn false // Already triggered this minute today\n\t\t\t\t}\n\t\t\t}\n\t\t\treturn true\n\t\t}\n\t}\n\treturn false\n}\n\n// shouldTriggerInterval checks if enough time has passed since last run\n// interval mode: run every X duration (e.g., \"30m\", \"2h\")\nfunc (cm *ClockMatcher) shouldTriggerInterval(robot *types.Robot, clock *types.Clock, now time.Time) bool {\n\tinterval, err := time.ParseDuration(clock.Every)\n\tif err != nil {\n\t\treturn false\n\t}\n\n\t// First run if never executed\n\tif robot.LastRun.IsZero() {\n\t\treturn true\n\t}\n\n\t// Check if interval has passed\n\treturn now.Sub(robot.LastRun) >= interval\n}\n\n// shouldTriggerDaemon checks if robot can restart immediately after last run\n// daemon mode: restart immediately after each run completes\nfunc (cm *ClockMatcher) shouldTriggerDaemon(robot *types.Robot, clock *types.Clock, now time.Time) bool {\n\t// Daemon mode: trigger if not currently running\n\t// CanRun() checks if robot has available execution slots\n\treturn robot.CanRun()\n}\n\n// matchesDay checks if current day matches the configured days\nfunc (cm *ClockMatcher) matchesDay(clock *types.Clock, now time.Time) bool {\n\t// Empty days or [\"*\"] means all days\n\tif len(clock.Days) == 0 {\n\t\treturn true\n\t}\n\n\tfor _, day := range clock.Days {\n\t\tif day == \"*\" {\n\t\t\treturn true\n\t\t}\n\t\t// Match day name (Mon, Tue, Wed, Thu, Fri, Sat, Sun)\n\t\t// or full name (Monday, Tuesday, etc.)\n\t\tweekday := now.Weekday().String()\n\t\tshortDay := weekday[:3] // Mon, Tue, etc.\n\t\tif day == weekday || day == shortDay {\n\t\t\treturn true\n\t\t}\n\t}\n\treturn false\n}\n\n// ParseTime parses a time string in \"HH:MM\" format\nfunc ParseTime(timeStr string) (hour, minute int, err error) {\n\tt, err := time.Parse(\"15:04\", timeStr)\n\tif err != nil {\n\t\treturn 0, 0, err\n\t}\n\treturn t.Hour(), t.Minute(), nil\n}\n\n// FormatTime formats hour and minute to \"HH:MM\" string\nfunc FormatTime(hour, minute int) string {\n\treturn time.Date(0, 1, 1, hour, minute, 0, 0, time.UTC).Format(\"15:04\")\n}\n" }, { "path": "agent/robot/trigger/clock_test.go", "content": "package trigger_test\n\nimport (\n\t\"testing\"\n\t\"time\"\n\n\t\"github.com/stretchr/testify/assert\"\n\t\"github.com/yaoapp/yao/agent/robot/trigger\"\n\t\"github.com/yaoapp/yao/agent/robot/types\"\n)\n\n// ==================== ClockMatcher Tests ====================\n\nfunc TestClockMatcherShouldTrigger(t *testing.T) {\n\tcm := trigger.NewClockMatcher()\n\n\tt.Run(\"nil robot returns false\", func(t *testing.T) {\n\t\tresult := cm.ShouldTrigger(nil, time.Now())\n\t\tassert.False(t, result)\n\t})\n\n\tt.Run(\"nil config returns false\", func(t *testing.T) {\n\t\trobot := &types.Robot{\n\t\t\tMemberID: \"robot_001\",\n\t\t\tConfig: nil,\n\t\t}\n\t\tresult := cm.ShouldTrigger(robot, time.Now())\n\t\tassert.False(t, result)\n\t})\n\n\tt.Run(\"nil clock config returns false\", func(t *testing.T) {\n\t\trobot := &types.Robot{\n\t\t\tMemberID: \"robot_001\",\n\t\t\tConfig: &types.Config{Clock: nil},\n\t\t}\n\t\tresult := cm.ShouldTrigger(robot, time.Now())\n\t\tassert.False(t, result)\n\t})\n}\n\n// ==================== Times Mode Tests ====================\n\nfunc TestClockMatcherTimesMode(t *testing.T) {\n\tcm := trigger.NewClockMatcher()\n\n\tt.Run(\"matches configured time\", func(t *testing.T) {\n\t\trobot := &types.Robot{\n\t\t\tMemberID: \"robot_001\",\n\t\t\tConfig: &types.Config{\n\t\t\t\tClock: &types.Clock{\n\t\t\t\t\tMode: types.ClockTimes,\n\t\t\t\t\tTimes: []string{\"09:00\", \"14:00\", \"17:00\"},\n\t\t\t\t\tDays: []string{\"*\"},\n\t\t\t\t\tTZ: \"UTC\",\n\t\t\t\t},\n\t\t\t},\n\t\t}\n\n\t\t// Create time at 09:00 UTC\n\t\tnow := time.Date(2025, 1, 15, 9, 0, 0, 0, time.UTC)\n\t\tresult := cm.ShouldTrigger(robot, now)\n\t\tassert.True(t, result)\n\t})\n\n\tt.Run(\"does not match non-configured time\", func(t *testing.T) {\n\t\trobot := &types.Robot{\n\t\t\tMemberID: \"robot_001\",\n\t\t\tConfig: &types.Config{\n\t\t\t\tClock: &types.Clock{\n\t\t\t\t\tMode: types.ClockTimes,\n\t\t\t\t\tTimes: []string{\"09:00\", \"14:00\", \"17:00\"},\n\t\t\t\t\tDays: []string{\"*\"},\n\t\t\t\t\tTZ: \"UTC\",\n\t\t\t\t},\n\t\t\t},\n\t\t}\n\n\t\t// Create time at 10:00 UTC (not in configured times)\n\t\tnow := time.Date(2025, 1, 15, 10, 0, 0, 0, time.UTC)\n\t\tresult := cm.ShouldTrigger(robot, now)\n\t\tassert.False(t, result)\n\t})\n\n\tt.Run(\"respects day filter - weekday\", func(t *testing.T) {\n\t\trobot := &types.Robot{\n\t\t\tMemberID: \"robot_001\",\n\t\t\tConfig: &types.Config{\n\t\t\t\tClock: &types.Clock{\n\t\t\t\t\tMode: types.ClockTimes,\n\t\t\t\t\tTimes: []string{\"09:00\"},\n\t\t\t\t\tDays: []string{\"Mon\", \"Tue\", \"Wed\", \"Thu\", \"Fri\"},\n\t\t\t\t\tTZ: \"UTC\",\n\t\t\t\t},\n\t\t\t},\n\t\t}\n\n\t\t// Wednesday 09:00 - should trigger\n\t\twed := time.Date(2025, 1, 15, 9, 0, 0, 0, time.UTC) // Wednesday\n\t\tassert.True(t, cm.ShouldTrigger(robot, wed))\n\n\t\t// Saturday 09:00 - should NOT trigger\n\t\tsat := time.Date(2025, 1, 18, 9, 0, 0, 0, time.UTC) // Saturday\n\t\tassert.False(t, cm.ShouldTrigger(robot, sat))\n\t})\n\n\tt.Run(\"dedup - same minute same day should not trigger twice\", func(t *testing.T) {\n\t\trobot := &types.Robot{\n\t\t\tMemberID: \"robot_001\",\n\t\t\tConfig: &types.Config{\n\t\t\t\tClock: &types.Clock{\n\t\t\t\t\tMode: types.ClockTimes,\n\t\t\t\t\tTimes: []string{\"09:00\"},\n\t\t\t\t\tDays: []string{\"*\"},\n\t\t\t\t\tTZ: \"UTC\",\n\t\t\t\t},\n\t\t\t},\n\t\t}\n\n\t\tnow := time.Date(2025, 1, 15, 9, 0, 0, 0, time.UTC)\n\n\t\t// First trigger - should succeed\n\t\tassert.True(t, cm.ShouldTrigger(robot, now))\n\n\t\t// Simulate LastRun was set\n\t\trobot.LastRun = now\n\n\t\t// Second trigger same minute - should fail\n\t\tnow2 := time.Date(2025, 1, 15, 9, 0, 30, 0, time.UTC)\n\t\tassert.False(t, cm.ShouldTrigger(robot, now2))\n\t})\n\n\tt.Run(\"different day should trigger again\", func(t *testing.T) {\n\t\trobot := &types.Robot{\n\t\t\tMemberID: \"robot_001\",\n\t\t\tConfig: &types.Config{\n\t\t\t\tClock: &types.Clock{\n\t\t\t\t\tMode: types.ClockTimes,\n\t\t\t\t\tTimes: []string{\"09:00\"},\n\t\t\t\t\tDays: []string{\"*\"},\n\t\t\t\t\tTZ: \"UTC\",\n\t\t\t\t},\n\t\t\t},\n\t\t}\n\n\t\t// First day\n\t\tday1 := time.Date(2025, 1, 15, 9, 0, 0, 0, time.UTC)\n\t\trobot.LastRun = day1\n\n\t\t// Next day same time - should trigger\n\t\tday2 := time.Date(2025, 1, 16, 9, 0, 0, 0, time.UTC)\n\t\tassert.True(t, cm.ShouldTrigger(robot, day2))\n\t})\n}\n\n// ==================== Interval Mode Tests ====================\n\nfunc TestClockMatcherIntervalMode(t *testing.T) {\n\tcm := trigger.NewClockMatcher()\n\n\tt.Run(\"first run triggers immediately\", func(t *testing.T) {\n\t\trobot := &types.Robot{\n\t\t\tMemberID: \"robot_001\",\n\t\t\tConfig: &types.Config{\n\t\t\t\tClock: &types.Clock{\n\t\t\t\t\tMode: types.ClockInterval,\n\t\t\t\t\tEvery: \"30m\",\n\t\t\t\t\tTZ: \"UTC\",\n\t\t\t\t},\n\t\t\t},\n\t\t}\n\n\t\t// LastRun is zero - should trigger\n\t\tnow := time.Now()\n\t\tresult := cm.ShouldTrigger(robot, now)\n\t\tassert.True(t, result)\n\t})\n\n\tt.Run(\"triggers after interval passed\", func(t *testing.T) {\n\t\trobot := &types.Robot{\n\t\t\tMemberID: \"robot_001\",\n\t\t\tConfig: &types.Config{\n\t\t\t\tClock: &types.Clock{\n\t\t\t\t\tMode: types.ClockInterval,\n\t\t\t\t\tEvery: \"30m\",\n\t\t\t\t\tTZ: \"UTC\",\n\t\t\t\t},\n\t\t\t},\n\t\t}\n\n\t\tnow := time.Now()\n\t\trobot.LastRun = now.Add(-31 * time.Minute) // 31 minutes ago\n\n\t\tresult := cm.ShouldTrigger(robot, now)\n\t\tassert.True(t, result)\n\t})\n\n\tt.Run(\"does not trigger before interval\", func(t *testing.T) {\n\t\trobot := &types.Robot{\n\t\t\tMemberID: \"robot_001\",\n\t\t\tConfig: &types.Config{\n\t\t\t\tClock: &types.Clock{\n\t\t\t\t\tMode: types.ClockInterval,\n\t\t\t\t\tEvery: \"30m\",\n\t\t\t\t\tTZ: \"UTC\",\n\t\t\t\t},\n\t\t\t},\n\t\t}\n\n\t\tnow := time.Now()\n\t\trobot.LastRun = now.Add(-15 * time.Minute) // Only 15 minutes ago\n\n\t\tresult := cm.ShouldTrigger(robot, now)\n\t\tassert.False(t, result)\n\t})\n\n\tt.Run(\"invalid interval format returns false\", func(t *testing.T) {\n\t\trobot := &types.Robot{\n\t\t\tMemberID: \"robot_001\",\n\t\t\tConfig: &types.Config{\n\t\t\t\tClock: &types.Clock{\n\t\t\t\t\tMode: types.ClockInterval,\n\t\t\t\t\tEvery: \"invalid\",\n\t\t\t\t\tTZ: \"UTC\",\n\t\t\t\t},\n\t\t\t},\n\t\t}\n\n\t\tresult := cm.ShouldTrigger(robot, time.Now())\n\t\tassert.False(t, result)\n\t})\n\n\tt.Run(\"various interval formats\", func(t *testing.T) {\n\t\tintervals := []struct {\n\t\t\tevery string\n\t\t\tlastAgo time.Duration\n\t\t\texpected bool\n\t\t}{\n\t\t\t{\"1h\", 61 * time.Minute, true},\n\t\t\t{\"1h\", 30 * time.Minute, false},\n\t\t\t{\"2h\", 121 * time.Minute, true},\n\t\t\t{\"2h\", 60 * time.Minute, false},\n\t\t\t{\"10s\", 11 * time.Second, true},\n\t\t\t{\"10s\", 5 * time.Second, false},\n\t\t}\n\n\t\tfor _, tt := range intervals {\n\t\t\tt.Run(tt.every, func(t *testing.T) {\n\t\t\t\trobot := &types.Robot{\n\t\t\t\t\tMemberID: \"robot_001\",\n\t\t\t\t\tConfig: &types.Config{\n\t\t\t\t\t\tClock: &types.Clock{\n\t\t\t\t\t\t\tMode: types.ClockInterval,\n\t\t\t\t\t\t\tEvery: tt.every,\n\t\t\t\t\t\t\tTZ: \"UTC\",\n\t\t\t\t\t\t},\n\t\t\t\t\t},\n\t\t\t\t}\n\n\t\t\t\tnow := time.Now()\n\t\t\t\trobot.LastRun = now.Add(-tt.lastAgo)\n\n\t\t\t\tresult := cm.ShouldTrigger(robot, now)\n\t\t\t\tassert.Equal(t, tt.expected, result)\n\t\t\t})\n\t\t}\n\t})\n}\n\n// ==================== Daemon Mode Tests ====================\n\nfunc TestClockMatcherDaemonMode(t *testing.T) {\n\tcm := trigger.NewClockMatcher()\n\n\tt.Run(\"triggers when robot can run\", func(t *testing.T) {\n\t\trobot := &types.Robot{\n\t\t\tMemberID: \"robot_001\",\n\t\t\tConfig: &types.Config{\n\t\t\t\tClock: &types.Clock{\n\t\t\t\t\tMode: types.ClockDaemon,\n\t\t\t\t\tTZ: \"UTC\",\n\t\t\t\t},\n\t\t\t\tQuota: &types.Quota{Max: 2},\n\t\t\t},\n\t\t}\n\n\t\t// No running executions - should trigger\n\t\tresult := cm.ShouldTrigger(robot, time.Now())\n\t\tassert.True(t, result)\n\t})\n\n\tt.Run(\"does not trigger when at quota\", func(t *testing.T) {\n\t\trobot := &types.Robot{\n\t\t\tMemberID: \"robot_001\",\n\t\t\tConfig: &types.Config{\n\t\t\t\tClock: &types.Clock{\n\t\t\t\t\tMode: types.ClockDaemon,\n\t\t\t\t\tTZ: \"UTC\",\n\t\t\t\t},\n\t\t\t\tQuota: &types.Quota{Max: 1},\n\t\t\t},\n\t\t}\n\n\t\t// Add one execution to fill quota\n\t\texec := &types.Execution{ID: \"exec_001\"}\n\t\trobot.AddExecution(exec)\n\n\t\tresult := cm.ShouldTrigger(robot, time.Now())\n\t\tassert.False(t, result)\n\n\t\t// Remove execution\n\t\trobot.RemoveExecution(\"exec_001\")\n\n\t\t// Now should trigger\n\t\tresult = cm.ShouldTrigger(robot, time.Now())\n\t\tassert.True(t, result)\n\t})\n}\n\n// ==================== Timezone Tests ====================\n\nfunc TestClockMatcherTimezone(t *testing.T) {\n\tcm := trigger.NewClockMatcher()\n\n\tt.Run(\"respects timezone for times mode\", func(t *testing.T) {\n\t\t// Robot configured for Asia/Shanghai (UTC+8)\n\t\trobot := &types.Robot{\n\t\t\tMemberID: \"robot_001\",\n\t\t\tConfig: &types.Config{\n\t\t\t\tClock: &types.Clock{\n\t\t\t\t\tMode: types.ClockTimes,\n\t\t\t\t\tTimes: []string{\"09:00\"},\n\t\t\t\t\tDays: []string{\"*\"},\n\t\t\t\t\tTZ: \"Asia/Shanghai\",\n\t\t\t\t},\n\t\t\t},\n\t\t}\n\n\t\t// 01:00 UTC = 09:00 Shanghai - should trigger\n\t\tutc0100 := time.Date(2025, 1, 15, 1, 0, 0, 0, time.UTC)\n\t\tassert.True(t, cm.ShouldTrigger(robot, utc0100))\n\n\t\t// 09:00 UTC = 17:00 Shanghai - should NOT trigger\n\t\tutc0900 := time.Date(2025, 1, 15, 9, 0, 0, 0, time.UTC)\n\t\tassert.False(t, cm.ShouldTrigger(robot, utc0900))\n\t})\n\n\tt.Run(\"invalid timezone falls back to local\", func(t *testing.T) {\n\t\trobot := &types.Robot{\n\t\t\tMemberID: \"robot_001\",\n\t\t\tConfig: &types.Config{\n\t\t\t\tClock: &types.Clock{\n\t\t\t\t\tMode: types.ClockTimes,\n\t\t\t\t\tTimes: []string{\"09:00\"},\n\t\t\t\t\tDays: []string{\"*\"},\n\t\t\t\t\tTZ: \"Invalid/Timezone\",\n\t\t\t\t},\n\t\t\t},\n\t\t}\n\n\t\t// Should still work with local time\n\t\tlocal0900 := time.Date(2025, 1, 15, 9, 0, 0, 0, time.Local)\n\t\tresult := cm.ShouldTrigger(robot, local0900)\n\t\t// Result depends on local timezone, just verify no panic\n\t\tassert.IsType(t, true, result)\n\t})\n}\n\n// ==================== ParseTime/FormatTime Tests ====================\n\nfunc TestParseTime(t *testing.T) {\n\tt.Run(\"parses valid time\", func(t *testing.T) {\n\t\thour, minute, err := trigger.ParseTime(\"09:30\")\n\t\tassert.NoError(t, err)\n\t\tassert.Equal(t, 9, hour)\n\t\tassert.Equal(t, 30, minute)\n\t})\n\n\tt.Run(\"parses midnight\", func(t *testing.T) {\n\t\thour, minute, err := trigger.ParseTime(\"00:00\")\n\t\tassert.NoError(t, err)\n\t\tassert.Equal(t, 0, hour)\n\t\tassert.Equal(t, 0, minute)\n\t})\n\n\tt.Run(\"parses 23:59\", func(t *testing.T) {\n\t\thour, minute, err := trigger.ParseTime(\"23:59\")\n\t\tassert.NoError(t, err)\n\t\tassert.Equal(t, 23, hour)\n\t\tassert.Equal(t, 59, minute)\n\t})\n\n\tt.Run(\"invalid format returns error\", func(t *testing.T) {\n\t\t// Note: time.Parse(\"15:04\", \"9:30\") actually succeeds\n\t\t// Only truly invalid formats fail\n\n\t\t_, _, err := trigger.ParseTime(\"09:30:00\")\n\t\tassert.Error(t, err)\n\n\t\t_, _, err = trigger.ParseTime(\"invalid\")\n\t\tassert.Error(t, err)\n\n\t\t_, _, err = trigger.ParseTime(\"\")\n\t\tassert.Error(t, err)\n\t})\n}\n\nfunc TestFormatTime(t *testing.T) {\n\ttests := []struct {\n\t\thour int\n\t\tminute int\n\t\texpected string\n\t}{\n\t\t{9, 0, \"09:00\"},\n\t\t{9, 30, \"09:30\"},\n\t\t{0, 0, \"00:00\"},\n\t\t{23, 59, \"23:59\"},\n\t\t{14, 5, \"14:05\"},\n\t}\n\n\tfor _, tt := range tests {\n\t\tt.Run(tt.expected, func(t *testing.T) {\n\t\t\tresult := trigger.FormatTime(tt.hour, tt.minute)\n\t\t\tassert.Equal(t, tt.expected, result)\n\t\t})\n\t}\n}\n" }, { "path": "agent/robot/trigger/control.go", "content": "package trigger\n\nimport (\n\t\"context\"\n\t\"fmt\"\n\t\"sync\"\n\t\"time\"\n\n\t\"github.com/yaoapp/yao/agent/robot/types\"\n)\n\n// ExecutionController manages execution lifecycle (pause/resume/stop)\ntype ExecutionController struct {\n\texecutions map[string]*ControlledExecution\n\tmu sync.RWMutex\n}\n\n// ControlledExecution represents an execution that can be controlled\ntype ControlledExecution struct {\n\tID string\n\tMemberID string\n\tTeamID string\n\tStatus types.ExecStatus\n\tPhase types.Phase\n\tStartTime time.Time\n\tPausedAt *time.Time\n\n\t// Control channels\n\tctx context.Context\n\tcancel context.CancelFunc\n\tpaused bool\n\tpauseMu sync.Mutex\n\tresumeCh chan struct{} // signaled (closed) when resumed\n}\n\n// NewExecutionController creates a new execution controller\nfunc NewExecutionController() *ExecutionController {\n\treturn &ExecutionController{\n\t\texecutions: make(map[string]*ControlledExecution),\n\t}\n}\n\n// Track starts tracking an execution\nfunc (c *ExecutionController) Track(execID, memberID, teamID string) *ControlledExecution {\n\tc.mu.Lock()\n\tdefer c.mu.Unlock()\n\n\tctx, cancel := context.WithCancel(context.Background())\n\texec := &ControlledExecution{\n\t\tID: execID,\n\t\tMemberID: memberID,\n\t\tTeamID: teamID,\n\t\tStatus: types.ExecRunning,\n\t\tPhase: types.PhaseInspiration,\n\t\tStartTime: time.Now(),\n\t\tctx: ctx,\n\t\tcancel: cancel,\n\t\tpaused: false,\n\t\tresumeCh: nil, // nil when not paused, created on pause\n\t}\n\n\tc.executions[execID] = exec\n\treturn exec\n}\n\n// Untrack stops tracking an execution\nfunc (c *ExecutionController) Untrack(execID string) {\n\tc.mu.Lock()\n\tdefer c.mu.Unlock()\n\tdelete(c.executions, execID)\n}\n\n// Get returns a tracked execution\nfunc (c *ExecutionController) Get(execID string) *ControlledExecution {\n\tc.mu.RLock()\n\tdefer c.mu.RUnlock()\n\treturn c.executions[execID]\n}\n\n// List returns all tracked executions\nfunc (c *ExecutionController) List() []*ControlledExecution {\n\tc.mu.RLock()\n\tdefer c.mu.RUnlock()\n\n\tresult := make([]*ControlledExecution, 0, len(c.executions))\n\tfor _, exec := range c.executions {\n\t\tresult = append(result, exec)\n\t}\n\treturn result\n}\n\n// ListByMember returns all executions for a specific member\nfunc (c *ExecutionController) ListByMember(memberID string) []*ControlledExecution {\n\tc.mu.RLock()\n\tdefer c.mu.RUnlock()\n\n\tvar result []*ControlledExecution\n\tfor _, exec := range c.executions {\n\t\tif exec.MemberID == memberID {\n\t\t\tresult = append(result, exec)\n\t\t}\n\t}\n\treturn result\n}\n\n// Pause pauses an execution\nfunc (c *ExecutionController) Pause(execID string) error {\n\texec := c.Get(execID)\n\tif exec == nil {\n\t\treturn fmt.Errorf(\"execution not found: %s\", execID)\n\t}\n\n\texec.pauseMu.Lock()\n\tdefer exec.pauseMu.Unlock()\n\n\tif exec.paused {\n\t\treturn fmt.Errorf(\"execution already paused: %s\", execID)\n\t}\n\n\texec.paused = true\n\tnow := time.Now()\n\texec.PausedAt = &now\n\n\t// Create a new resume channel that will be closed on resume\n\texec.resumeCh = make(chan struct{})\n\n\treturn nil\n}\n\n// Resume resumes a paused execution\nfunc (c *ExecutionController) Resume(execID string) error {\n\texec := c.Get(execID)\n\tif exec == nil {\n\t\treturn fmt.Errorf(\"execution not found: %s\", execID)\n\t}\n\n\texec.pauseMu.Lock()\n\tdefer exec.pauseMu.Unlock()\n\n\tif !exec.paused {\n\t\treturn fmt.Errorf(\"execution not paused: %s\", execID)\n\t}\n\n\texec.paused = false\n\texec.PausedAt = nil\n\n\t// Close the resume channel to signal resume to waiting goroutines\n\tif exec.resumeCh != nil {\n\t\tclose(exec.resumeCh)\n\t\texec.resumeCh = nil\n\t}\n\n\treturn nil\n}\n\n// Stop stops an execution\nfunc (c *ExecutionController) Stop(execID string) error {\n\tc.mu.Lock()\n\tdefer c.mu.Unlock()\n\n\texec, ok := c.executions[execID]\n\tif !ok {\n\t\treturn fmt.Errorf(\"execution not found: %s\", execID)\n\t}\n\n\t// Cancel the context to signal stop\n\tif exec.cancel != nil {\n\t\texec.cancel()\n\t}\n\n\texec.Status = types.ExecCancelled\n\n\t// Remove from tracking\n\tdelete(c.executions, execID)\n\n\treturn nil\n}\n\n// ==================== ControlledExecution methods ====================\n\n// IsPaused returns true if the execution is paused\nfunc (e *ControlledExecution) IsPaused() bool {\n\te.pauseMu.Lock()\n\tdefer e.pauseMu.Unlock()\n\treturn e.paused\n}\n\n// IsCancelled returns true if the execution is cancelled\nfunc (e *ControlledExecution) IsCancelled() bool {\n\tselect {\n\tcase <-e.ctx.Done():\n\t\treturn true\n\tdefault:\n\t\treturn false\n\t}\n}\n\n// Context returns the execution's context\nfunc (e *ControlledExecution) Context() context.Context {\n\treturn e.ctx\n}\n\n// WaitIfPaused blocks until the execution is resumed or cancelled\n// Returns error if cancelled\nfunc (e *ControlledExecution) WaitIfPaused() error {\n\te.pauseMu.Lock()\n\tpaused := e.paused\n\tresumeCh := e.resumeCh\n\te.pauseMu.Unlock()\n\n\tif !paused {\n\t\treturn nil\n\t}\n\n\t// Safety check: if paused but resumeCh is nil (shouldn't happen in normal flow),\n\t// treat as not paused to avoid blocking forever on nil channel\n\tif resumeCh == nil {\n\t\treturn nil\n\t}\n\n\t// resumeCh is created when paused and closed when resumed\n\t// Wait for resume signal or cancellation\n\tselect {\n\tcase <-e.ctx.Done():\n\t\treturn types.ErrExecutionCancelled\n\tcase <-resumeCh:\n\t\t// Resume signal received, execution can continue\n\t\treturn nil\n\t}\n}\n\n// CheckCancelled checks if the execution is cancelled and returns error if so\nfunc (e *ControlledExecution) CheckCancelled() error {\n\tif e.IsCancelled() {\n\t\treturn types.ErrExecutionCancelled\n\t}\n\treturn nil\n}\n\n// UpdatePhase updates the current phase\nfunc (e *ControlledExecution) UpdatePhase(phase types.Phase) {\n\te.Phase = phase\n}\n\n// UpdateStatus updates the execution status\nfunc (e *ControlledExecution) UpdateStatus(status types.ExecStatus) {\n\te.Status = status\n}\n" }, { "path": "agent/robot/trigger/control_test.go", "content": "package trigger_test\n\nimport (\n\t\"sync\"\n\t\"testing\"\n\t\"time\"\n\n\t\"github.com/stretchr/testify/assert\"\n\t\"github.com/yaoapp/yao/agent/robot/trigger\"\n\t\"github.com/yaoapp/yao/agent/robot/types\"\n)\n\n// ==================== ExecutionController Tests ====================\n\nfunc TestExecutionControllerTrack(t *testing.T) {\n\tt.Run(\"tracks new execution\", func(t *testing.T) {\n\t\tctrl := trigger.NewExecutionController()\n\n\t\texec := ctrl.Track(\"exec_001\", \"robot_001\", \"team_001\")\n\n\t\tassert.NotNil(t, exec)\n\t\tassert.Equal(t, \"exec_001\", exec.ID)\n\t\tassert.Equal(t, \"robot_001\", exec.MemberID)\n\t\tassert.Equal(t, \"team_001\", exec.TeamID)\n\t\tassert.Equal(t, types.ExecRunning, exec.Status)\n\t\tassert.False(t, exec.IsPaused())\n\t\tassert.False(t, exec.IsCancelled())\n\t})\n\n\tt.Run(\"get tracked execution\", func(t *testing.T) {\n\t\tctrl := trigger.NewExecutionController()\n\t\tctrl.Track(\"exec_001\", \"robot_001\", \"team_001\")\n\n\t\texec := ctrl.Get(\"exec_001\")\n\t\tassert.NotNil(t, exec)\n\t\tassert.Equal(t, \"exec_001\", exec.ID)\n\t})\n\n\tt.Run(\"get non-existent execution returns nil\", func(t *testing.T) {\n\t\tctrl := trigger.NewExecutionController()\n\n\t\texec := ctrl.Get(\"non_existent\")\n\t\tassert.Nil(t, exec)\n\t})\n}\n\nfunc TestExecutionControllerList(t *testing.T) {\n\tt.Run(\"list all executions\", func(t *testing.T) {\n\t\tctrl := trigger.NewExecutionController()\n\t\tctrl.Track(\"exec_001\", \"robot_001\", \"team_001\")\n\t\tctrl.Track(\"exec_002\", \"robot_002\", \"team_001\")\n\t\tctrl.Track(\"exec_003\", \"robot_001\", \"team_002\")\n\n\t\tlist := ctrl.List()\n\t\tassert.Len(t, list, 3)\n\t})\n\n\tt.Run(\"list by member\", func(t *testing.T) {\n\t\tctrl := trigger.NewExecutionController()\n\t\tctrl.Track(\"exec_001\", \"robot_001\", \"team_001\")\n\t\tctrl.Track(\"exec_002\", \"robot_002\", \"team_001\")\n\t\tctrl.Track(\"exec_003\", \"robot_001\", \"team_002\")\n\n\t\tlist := ctrl.ListByMember(\"robot_001\")\n\t\tassert.Len(t, list, 2)\n\n\t\tlist = ctrl.ListByMember(\"robot_002\")\n\t\tassert.Len(t, list, 1)\n\n\t\tlist = ctrl.ListByMember(\"robot_003\")\n\t\tassert.Len(t, list, 0)\n\t})\n}\n\nfunc TestExecutionControllerUntrack(t *testing.T) {\n\tt.Run(\"untrack removes execution\", func(t *testing.T) {\n\t\tctrl := trigger.NewExecutionController()\n\t\tctrl.Track(\"exec_001\", \"robot_001\", \"team_001\")\n\n\t\tassert.NotNil(t, ctrl.Get(\"exec_001\"))\n\n\t\tctrl.Untrack(\"exec_001\")\n\n\t\tassert.Nil(t, ctrl.Get(\"exec_001\"))\n\t})\n\n\tt.Run(\"untrack non-existent does not panic\", func(t *testing.T) {\n\t\tctrl := trigger.NewExecutionController()\n\n\t\tassert.NotPanics(t, func() {\n\t\t\tctrl.Untrack(\"non_existent\")\n\t\t})\n\t})\n}\n\n// ==================== Pause/Resume Tests ====================\n\nfunc TestExecutionControllerPause(t *testing.T) {\n\tt.Run(\"pause execution\", func(t *testing.T) {\n\t\tctrl := trigger.NewExecutionController()\n\t\texec := ctrl.Track(\"exec_001\", \"robot_001\", \"team_001\")\n\n\t\terr := ctrl.Pause(\"exec_001\")\n\t\tassert.NoError(t, err)\n\t\tassert.True(t, exec.IsPaused())\n\t\tassert.NotNil(t, exec.PausedAt)\n\t})\n\n\tt.Run(\"pause non-existent returns error\", func(t *testing.T) {\n\t\tctrl := trigger.NewExecutionController()\n\n\t\terr := ctrl.Pause(\"non_existent\")\n\t\tassert.Error(t, err)\n\t\tassert.Contains(t, err.Error(), \"not found\")\n\t})\n\n\tt.Run(\"pause already paused returns error\", func(t *testing.T) {\n\t\tctrl := trigger.NewExecutionController()\n\t\tctrl.Track(\"exec_001\", \"robot_001\", \"team_001\")\n\n\t\terr := ctrl.Pause(\"exec_001\")\n\t\tassert.NoError(t, err)\n\n\t\terr = ctrl.Pause(\"exec_001\")\n\t\tassert.Error(t, err)\n\t\tassert.Contains(t, err.Error(), \"already paused\")\n\t})\n}\n\nfunc TestExecutionControllerResume(t *testing.T) {\n\tt.Run(\"resume paused execution\", func(t *testing.T) {\n\t\tctrl := trigger.NewExecutionController()\n\t\texec := ctrl.Track(\"exec_001\", \"robot_001\", \"team_001\")\n\n\t\tctrl.Pause(\"exec_001\")\n\t\tassert.True(t, exec.IsPaused())\n\n\t\terr := ctrl.Resume(\"exec_001\")\n\t\tassert.NoError(t, err)\n\t\tassert.False(t, exec.IsPaused())\n\t\tassert.Nil(t, exec.PausedAt)\n\t})\n\n\tt.Run(\"resume non-existent returns error\", func(t *testing.T) {\n\t\tctrl := trigger.NewExecutionController()\n\n\t\terr := ctrl.Resume(\"non_existent\")\n\t\tassert.Error(t, err)\n\t\tassert.Contains(t, err.Error(), \"not found\")\n\t})\n\n\tt.Run(\"resume not paused returns error\", func(t *testing.T) {\n\t\tctrl := trigger.NewExecutionController()\n\t\tctrl.Track(\"exec_001\", \"robot_001\", \"team_001\")\n\n\t\terr := ctrl.Resume(\"exec_001\")\n\t\tassert.Error(t, err)\n\t\tassert.Contains(t, err.Error(), \"not paused\")\n\t})\n}\n\n// ==================== Stop Tests ====================\n\nfunc TestExecutionControllerStop(t *testing.T) {\n\tt.Run(\"stop execution\", func(t *testing.T) {\n\t\tctrl := trigger.NewExecutionController()\n\t\texec := ctrl.Track(\"exec_001\", \"robot_001\", \"team_001\")\n\n\t\terr := ctrl.Stop(\"exec_001\")\n\t\tassert.NoError(t, err)\n\t\tassert.True(t, exec.IsCancelled())\n\t\tassert.Equal(t, types.ExecCancelled, exec.Status)\n\n\t\t// Should be removed from tracking\n\t\tassert.Nil(t, ctrl.Get(\"exec_001\"))\n\t})\n\n\tt.Run(\"stop non-existent returns error\", func(t *testing.T) {\n\t\tctrl := trigger.NewExecutionController()\n\n\t\terr := ctrl.Stop(\"non_existent\")\n\t\tassert.Error(t, err)\n\t\tassert.Contains(t, err.Error(), \"not found\")\n\t})\n}\n\n// ==================== ControlledExecution Methods Tests ====================\n\nfunc TestControlledExecutionContext(t *testing.T) {\n\tt.Run(\"context is valid\", func(t *testing.T) {\n\t\tctrl := trigger.NewExecutionController()\n\t\texec := ctrl.Track(\"exec_001\", \"robot_001\", \"team_001\")\n\n\t\tctx := exec.Context()\n\t\tassert.NotNil(t, ctx)\n\n\t\t// Context should not be done yet\n\t\tselect {\n\t\tcase <-ctx.Done():\n\t\t\tt.Fatal(\"context should not be done\")\n\t\tdefault:\n\t\t\t// OK\n\t\t}\n\t})\n\n\tt.Run(\"context done after stop\", func(t *testing.T) {\n\t\tctrl := trigger.NewExecutionController()\n\t\texec := ctrl.Track(\"exec_001\", \"robot_001\", \"team_001\")\n\n\t\tctx := exec.Context()\n\t\tctrl.Stop(\"exec_001\")\n\n\t\tselect {\n\t\tcase <-ctx.Done():\n\t\t\t// OK\n\t\tcase <-time.After(100 * time.Millisecond):\n\t\t\tt.Fatal(\"context should be done after stop\")\n\t\t}\n\t})\n}\n\nfunc TestControlledExecutionCheckCancelled(t *testing.T) {\n\tt.Run(\"not cancelled returns nil\", func(t *testing.T) {\n\t\tctrl := trigger.NewExecutionController()\n\t\texec := ctrl.Track(\"exec_001\", \"robot_001\", \"team_001\")\n\n\t\terr := exec.CheckCancelled()\n\t\tassert.NoError(t, err)\n\t})\n\n\tt.Run(\"cancelled returns error\", func(t *testing.T) {\n\t\tctrl := trigger.NewExecutionController()\n\t\texec := ctrl.Track(\"exec_001\", \"robot_001\", \"team_001\")\n\n\t\tctrl.Stop(\"exec_001\")\n\n\t\terr := exec.CheckCancelled()\n\t\tassert.Error(t, err)\n\t\tassert.Equal(t, types.ErrExecutionCancelled, err)\n\t})\n}\n\nfunc TestControlledExecutionUpdatePhase(t *testing.T) {\n\tctrl := trigger.NewExecutionController()\n\texec := ctrl.Track(\"exec_001\", \"robot_001\", \"team_001\")\n\n\tassert.Equal(t, types.PhaseInspiration, exec.Phase)\n\n\texec.UpdatePhase(types.PhaseGoals)\n\tassert.Equal(t, types.PhaseGoals, exec.Phase)\n\n\texec.UpdatePhase(types.PhaseTasks)\n\tassert.Equal(t, types.PhaseTasks, exec.Phase)\n}\n\nfunc TestControlledExecutionUpdateStatus(t *testing.T) {\n\tctrl := trigger.NewExecutionController()\n\texec := ctrl.Track(\"exec_001\", \"robot_001\", \"team_001\")\n\n\tassert.Equal(t, types.ExecRunning, exec.Status)\n\n\texec.UpdateStatus(types.ExecCompleted)\n\tassert.Equal(t, types.ExecCompleted, exec.Status)\n\n\texec.UpdateStatus(types.ExecFailed)\n\tassert.Equal(t, types.ExecFailed, exec.Status)\n}\n\n// ==================== WaitIfPaused Tests ====================\n\nfunc TestControlledExecutionWaitIfPaused(t *testing.T) {\n\tt.Run(\"returns immediately if not paused\", func(t *testing.T) {\n\t\tctrl := trigger.NewExecutionController()\n\t\texec := ctrl.Track(\"exec_001\", \"robot_001\", \"team_001\")\n\n\t\tdone := make(chan error)\n\t\tgo func() {\n\t\t\tdone <- exec.WaitIfPaused()\n\t\t}()\n\n\t\tselect {\n\t\tcase err := <-done:\n\t\t\tassert.NoError(t, err)\n\t\tcase <-time.After(100 * time.Millisecond):\n\t\t\tt.Fatal(\"WaitIfPaused should return immediately when not paused\")\n\t\t}\n\t})\n\n\tt.Run(\"does not infinite loop when paused without resume\", func(t *testing.T) {\n\t\t// This test verifies the fix for the infinite loop bug\n\t\t// where WaitIfPaused would spin if pauseCh was closed but paused remained true\n\t\tctrl := trigger.NewExecutionController()\n\t\texec := ctrl.Track(\"exec_001\", \"robot_001\", \"team_001\")\n\n\t\tctrl.Pause(\"exec_001\")\n\n\t\t// Start WaitIfPaused in a goroutine\n\t\tdone := make(chan error)\n\t\tgo func() {\n\t\t\tdone <- exec.WaitIfPaused()\n\t\t}()\n\n\t\t// Wait a bit - if there's an infinite loop, CPU would spike\n\t\t// The goroutine should be blocked, not spinning\n\t\ttime.Sleep(100 * time.Millisecond)\n\n\t\t// Now stop the execution - this should unblock WaitIfPaused\n\t\tctrl.Stop(\"exec_001\")\n\n\t\tselect {\n\t\tcase err := <-done:\n\t\t\t// Should get cancellation error\n\t\t\tassert.Error(t, err)\n\t\tcase <-time.After(200 * time.Millisecond):\n\t\t\tt.Fatal(\"WaitIfPaused should unblock after stop\")\n\t\t}\n\t})\n\n\tt.Run(\"rapid pause-resume-pause does not cause issues\", func(t *testing.T) {\n\t\t// Test TOCTOU race condition handling\n\t\tctrl := trigger.NewExecutionController()\n\t\texec := ctrl.Track(\"exec_001\", \"robot_001\", \"team_001\")\n\n\t\t// Pause first\n\t\tctrl.Pause(\"exec_001\")\n\n\t\tdone := make(chan error)\n\t\tgo func() {\n\t\t\tdone <- exec.WaitIfPaused()\n\t\t}()\n\n\t\t// Rapid resume then pause again\n\t\ttime.Sleep(10 * time.Millisecond)\n\t\tctrl.Resume(\"exec_001\")\n\n\t\t// WaitIfPaused should return (the original resumeCh was closed)\n\t\tselect {\n\t\tcase err := <-done:\n\t\t\tassert.NoError(t, err)\n\t\tcase <-time.After(200 * time.Millisecond):\n\t\t\tt.Fatal(\"WaitIfPaused should return after resume\")\n\t\t}\n\t})\n\n\tt.Run(\"blocks when paused, resumes after resume\", func(t *testing.T) {\n\t\tctrl := trigger.NewExecutionController()\n\t\texec := ctrl.Track(\"exec_001\", \"robot_001\", \"team_001\")\n\n\t\tctrl.Pause(\"exec_001\")\n\n\t\tdone := make(chan error)\n\t\tgo func() {\n\t\t\tdone <- exec.WaitIfPaused()\n\t\t}()\n\n\t\t// Should be blocked\n\t\tselect {\n\t\tcase <-done:\n\t\t\tt.Fatal(\"WaitIfPaused should block when paused\")\n\t\tcase <-time.After(50 * time.Millisecond):\n\t\t\t// OK, still blocked\n\t\t}\n\n\t\t// Resume\n\t\tctrl.Resume(\"exec_001\")\n\n\t\t// Should unblock\n\t\tselect {\n\t\tcase err := <-done:\n\t\t\tassert.NoError(t, err)\n\t\tcase <-time.After(100 * time.Millisecond):\n\t\t\tt.Fatal(\"WaitIfPaused should unblock after resume\")\n\t\t}\n\t})\n\n\tt.Run(\"returns error when cancelled while paused\", func(t *testing.T) {\n\t\tctrl := trigger.NewExecutionController()\n\t\texec := ctrl.Track(\"exec_001\", \"robot_001\", \"team_001\")\n\n\t\tctrl.Pause(\"exec_001\")\n\n\t\tdone := make(chan error)\n\t\tgo func() {\n\t\t\tdone <- exec.WaitIfPaused()\n\t\t}()\n\n\t\t// Should be blocked\n\t\tselect {\n\t\tcase <-done:\n\t\t\tt.Fatal(\"WaitIfPaused should block when paused\")\n\t\tcase <-time.After(50 * time.Millisecond):\n\t\t\t// OK, still blocked\n\t\t}\n\n\t\t// Stop instead of resume\n\t\tctrl.Stop(\"exec_001\")\n\n\t\t// Should unblock with error\n\t\tselect {\n\t\tcase err := <-done:\n\t\t\tassert.Error(t, err)\n\t\t\tassert.Equal(t, types.ErrExecutionCancelled, err)\n\t\tcase <-time.After(100 * time.Millisecond):\n\t\t\tt.Fatal(\"WaitIfPaused should unblock after stop\")\n\t\t}\n\t})\n}\n\n// ==================== Concurrent Access Tests ====================\n\nfunc TestExecutionControllerConcurrency(t *testing.T) {\n\tt.Run(\"concurrent track and list\", func(t *testing.T) {\n\t\tctrl := trigger.NewExecutionController()\n\t\tvar wg sync.WaitGroup\n\n\t\t// Concurrent tracking\n\t\tfor i := 0; i < 100; i++ {\n\t\t\twg.Add(1)\n\t\t\tgo func(id int) {\n\t\t\t\tdefer wg.Done()\n\t\t\t\tctrl.Track(\n\t\t\t\t\t\"exec_\"+string(rune('0'+id%10)),\n\t\t\t\t\t\"robot_\"+string(rune('0'+id%5)),\n\t\t\t\t\t\"team_001\",\n\t\t\t\t)\n\t\t\t}(i)\n\t\t}\n\n\t\t// Concurrent listing\n\t\tfor i := 0; i < 50; i++ {\n\t\t\twg.Add(1)\n\t\t\tgo func() {\n\t\t\t\tdefer wg.Done()\n\t\t\t\t_ = ctrl.List()\n\t\t\t}()\n\t\t}\n\n\t\twg.Wait()\n\t\t// No race conditions or panics\n\t})\n\n\tt.Run(\"concurrent pause/resume\", func(t *testing.T) {\n\t\tctrl := trigger.NewExecutionController()\n\t\tctrl.Track(\"exec_001\", \"robot_001\", \"team_001\")\n\n\t\tvar wg sync.WaitGroup\n\n\t\t// Concurrent pause/resume attempts\n\t\tfor i := 0; i < 50; i++ {\n\t\t\twg.Add(2)\n\t\t\tgo func() {\n\t\t\t\tdefer wg.Done()\n\t\t\t\t_ = ctrl.Pause(\"exec_001\")\n\t\t\t}()\n\t\t\tgo func() {\n\t\t\t\tdefer wg.Done()\n\t\t\t\t_ = ctrl.Resume(\"exec_001\")\n\t\t\t}()\n\t\t}\n\n\t\twg.Wait()\n\t\t// No race conditions or panics\n\t})\n}\n" }, { "path": "agent/robot/trigger/trigger.go", "content": "// Package trigger provides trigger-related utilities and execution control\n// The main trigger logic is in the manager package.\n// This package provides:\n// - Validation functions for intervention and event requests\n// - Builder helpers for TriggerInput\n// - ExecutionController for pause/resume/stop\n// - ClockMatcher for clock trigger matching (reusable)\npackage trigger\n\nimport (\n\t\"fmt\"\n\n\t\"github.com/yaoapp/yao/agent/robot/types\"\n)\n\n// ValidateIntervention validates a human intervention request\nfunc ValidateIntervention(req *types.InterveneRequest) error {\n\tif req == nil {\n\t\treturn fmt.Errorf(\"request is nil\")\n\t}\n\n\tif req.MemberID == \"\" {\n\t\treturn fmt.Errorf(\"member_id is required\")\n\t}\n\n\tif !isValidAction(req.Action) {\n\t\treturn fmt.Errorf(\"invalid action: %s\", req.Action)\n\t}\n\n\t// Validate action-specific requirements\n\tswitch req.Action {\n\tcase types.ActionTaskAdd, types.ActionGoalAdd, types.ActionInstruct:\n\t\t// These actions require messages\n\t\tif len(req.Messages) == 0 {\n\t\t\treturn fmt.Errorf(\"messages required for action: %s\", req.Action)\n\t\t}\n\n\tcase types.ActionPlanAdd:\n\t\t// Plan add requires plan_time\n\t\tif req.PlanTime == nil {\n\t\t\treturn fmt.Errorf(\"plan_time required for action: plan.add\")\n\t\t}\n\t}\n\n\treturn nil\n}\n\n// ValidateEvent validates an event trigger request\nfunc ValidateEvent(req *types.EventRequest) error {\n\tif req == nil {\n\t\treturn fmt.Errorf(\"request is nil\")\n\t}\n\n\tif req.MemberID == \"\" {\n\t\treturn fmt.Errorf(\"member_id is required\")\n\t}\n\n\tif req.Source == \"\" {\n\t\treturn fmt.Errorf(\"source is required\")\n\t}\n\n\tif req.EventType == \"\" {\n\t\treturn fmt.Errorf(\"event_type is required\")\n\t}\n\n\treturn nil\n}\n\n// BuildEventInput creates a TriggerInput from an event request\nfunc BuildEventInput(req *types.EventRequest) *types.TriggerInput {\n\treturn &types.TriggerInput{\n\t\tSource: types.EventSource(req.Source),\n\t\tEventType: req.EventType,\n\t\tData: req.Data,\n\t}\n}\n\n// isValidAction checks if the intervention action is valid\nfunc isValidAction(action types.InterventionAction) bool {\n\tswitch action {\n\tcase types.ActionTaskAdd,\n\t\ttypes.ActionTaskCancel,\n\t\ttypes.ActionTaskUpdate,\n\t\ttypes.ActionGoalAdjust,\n\t\ttypes.ActionGoalAdd,\n\t\ttypes.ActionGoalComplete,\n\t\ttypes.ActionGoalCancel,\n\t\ttypes.ActionPlanAdd,\n\t\ttypes.ActionPlanRemove,\n\t\ttypes.ActionPlanUpdate,\n\t\ttypes.ActionInstruct:\n\t\treturn true\n\tdefault:\n\t\treturn false\n\t}\n}\n\n// GetActionCategory returns the category of an intervention action\nfunc GetActionCategory(action types.InterventionAction) string {\n\tswitch action {\n\tcase types.ActionTaskAdd, types.ActionTaskCancel, types.ActionTaskUpdate:\n\t\treturn \"task\"\n\tcase types.ActionGoalAdjust, types.ActionGoalAdd, types.ActionGoalComplete, types.ActionGoalCancel:\n\t\treturn \"goal\"\n\tcase types.ActionPlanAdd, types.ActionPlanRemove, types.ActionPlanUpdate:\n\t\treturn \"plan\"\n\tcase types.ActionInstruct:\n\t\treturn \"instruct\"\n\tdefault:\n\t\treturn \"unknown\"\n\t}\n}\n\n// GetActionDescription returns a human-readable description of an action\nfunc GetActionDescription(action types.InterventionAction) string {\n\tswitch action {\n\tcase types.ActionTaskAdd:\n\t\treturn \"Add a new task\"\n\tcase types.ActionTaskCancel:\n\t\treturn \"Cancel a task\"\n\tcase types.ActionTaskUpdate:\n\t\treturn \"Update task details\"\n\tcase types.ActionGoalAdjust:\n\t\treturn \"Adjust current goal\"\n\tcase types.ActionGoalAdd:\n\t\treturn \"Add a new goal\"\n\tcase types.ActionGoalComplete:\n\t\treturn \"Mark goal as complete\"\n\tcase types.ActionGoalCancel:\n\t\treturn \"Cancel a goal\"\n\tcase types.ActionPlanAdd:\n\t\treturn \"Add to plan queue\"\n\tcase types.ActionPlanRemove:\n\t\treturn \"Remove from plan queue\"\n\tcase types.ActionPlanUpdate:\n\t\treturn \"Update planned item\"\n\tcase types.ActionInstruct:\n\t\treturn \"Direct instruction to robot\"\n\tdefault:\n\t\treturn \"Unknown action\"\n\t}\n}\n" }, { "path": "agent/robot/trigger/trigger_test.go", "content": "package trigger_test\n\nimport (\n\t\"testing\"\n\t\"time\"\n\n\t\"github.com/stretchr/testify/assert\"\n\tagentcontext \"github.com/yaoapp/yao/agent/context\"\n\t\"github.com/yaoapp/yao/agent/robot/trigger\"\n\t\"github.com/yaoapp/yao/agent/robot/types\"\n)\n\n// ==================== ValidateIntervention Tests ====================\n\nfunc TestValidateIntervention(t *testing.T) {\n\tt.Run(\"nil request returns error\", func(t *testing.T) {\n\t\terr := trigger.ValidateIntervention(nil)\n\t\tassert.Error(t, err)\n\t\tassert.Contains(t, err.Error(), \"request is nil\")\n\t})\n\n\tt.Run(\"empty member_id returns error\", func(t *testing.T) {\n\t\treq := &types.InterveneRequest{\n\t\t\tMemberID: \"\",\n\t\t\tAction: types.ActionTaskAdd,\n\t\t}\n\t\terr := trigger.ValidateIntervention(req)\n\t\tassert.Error(t, err)\n\t\tassert.Contains(t, err.Error(), \"member_id is required\")\n\t})\n\n\tt.Run(\"invalid action returns error\", func(t *testing.T) {\n\t\treq := &types.InterveneRequest{\n\t\t\tMemberID: \"robot_001\",\n\t\t\tAction: types.InterventionAction(\"invalid.action\"),\n\t\t}\n\t\terr := trigger.ValidateIntervention(req)\n\t\tassert.Error(t, err)\n\t\tassert.Contains(t, err.Error(), \"invalid action\")\n\t})\n\n\tt.Run(\"task.add without messages returns error\", func(t *testing.T) {\n\t\treq := &types.InterveneRequest{\n\t\t\tMemberID: \"robot_001\",\n\t\t\tAction: types.ActionTaskAdd,\n\t\t\tMessages: nil,\n\t\t}\n\t\terr := trigger.ValidateIntervention(req)\n\t\tassert.Error(t, err)\n\t\tassert.Contains(t, err.Error(), \"messages required\")\n\t})\n\n\tt.Run(\"goal.add without messages returns error\", func(t *testing.T) {\n\t\treq := &types.InterveneRequest{\n\t\t\tMemberID: \"robot_001\",\n\t\t\tAction: types.ActionGoalAdd,\n\t\t\tMessages: nil,\n\t\t}\n\t\terr := trigger.ValidateIntervention(req)\n\t\tassert.Error(t, err)\n\t\tassert.Contains(t, err.Error(), \"messages required\")\n\t})\n\n\tt.Run(\"instruct without messages returns error\", func(t *testing.T) {\n\t\treq := &types.InterveneRequest{\n\t\t\tMemberID: \"robot_001\",\n\t\t\tAction: types.ActionInstruct,\n\t\t\tMessages: nil,\n\t\t}\n\t\terr := trigger.ValidateIntervention(req)\n\t\tassert.Error(t, err)\n\t\tassert.Contains(t, err.Error(), \"messages required\")\n\t})\n\n\tt.Run(\"plan.add without plan_time returns error\", func(t *testing.T) {\n\t\treq := &types.InterveneRequest{\n\t\t\tMemberID: \"robot_001\",\n\t\t\tAction: types.ActionPlanAdd,\n\t\t\tPlanTime: nil,\n\t\t}\n\t\terr := trigger.ValidateIntervention(req)\n\t\tassert.Error(t, err)\n\t\tassert.Contains(t, err.Error(), \"plan_time required\")\n\t})\n\n\tt.Run(\"valid task.add request passes\", func(t *testing.T) {\n\t\treq := &types.InterveneRequest{\n\t\t\tMemberID: \"robot_001\",\n\t\t\tAction: types.ActionTaskAdd,\n\t\t\tMessages: []agentcontext.Message{\n\t\t\t\t{Role: agentcontext.RoleUser, Content: \"Add a new task\"},\n\t\t\t},\n\t\t}\n\t\terr := trigger.ValidateIntervention(req)\n\t\tassert.NoError(t, err)\n\t})\n\n\tt.Run(\"valid plan.add request passes\", func(t *testing.T) {\n\t\tplanTime := time.Now().Add(time.Hour)\n\t\treq := &types.InterveneRequest{\n\t\t\tMemberID: \"robot_001\",\n\t\t\tAction: types.ActionPlanAdd,\n\t\t\tPlanTime: &planTime,\n\t\t}\n\t\terr := trigger.ValidateIntervention(req)\n\t\tassert.NoError(t, err)\n\t})\n\n\tt.Run(\"task.cancel without messages passes\", func(t *testing.T) {\n\t\treq := &types.InterveneRequest{\n\t\t\tMemberID: \"robot_001\",\n\t\t\tAction: types.ActionTaskCancel,\n\t\t}\n\t\terr := trigger.ValidateIntervention(req)\n\t\tassert.NoError(t, err)\n\t})\n\n\tt.Run(\"goal.adjust without messages passes\", func(t *testing.T) {\n\t\treq := &types.InterveneRequest{\n\t\t\tMemberID: \"robot_001\",\n\t\t\tAction: types.ActionGoalAdjust,\n\t\t}\n\t\terr := trigger.ValidateIntervention(req)\n\t\tassert.NoError(t, err)\n\t})\n}\n\n// ==================== ValidateEvent Tests ====================\n\nfunc TestValidateEvent(t *testing.T) {\n\tt.Run(\"nil request returns error\", func(t *testing.T) {\n\t\terr := trigger.ValidateEvent(nil)\n\t\tassert.Error(t, err)\n\t\tassert.Contains(t, err.Error(), \"request is nil\")\n\t})\n\n\tt.Run(\"empty member_id returns error\", func(t *testing.T) {\n\t\treq := &types.EventRequest{\n\t\t\tMemberID: \"\",\n\t\t\tSource: \"webhook\",\n\t\t\tEventType: \"lead.created\",\n\t\t}\n\t\terr := trigger.ValidateEvent(req)\n\t\tassert.Error(t, err)\n\t\tassert.Contains(t, err.Error(), \"member_id is required\")\n\t})\n\n\tt.Run(\"empty source returns error\", func(t *testing.T) {\n\t\treq := &types.EventRequest{\n\t\t\tMemberID: \"robot_001\",\n\t\t\tSource: \"\",\n\t\t\tEventType: \"lead.created\",\n\t\t}\n\t\terr := trigger.ValidateEvent(req)\n\t\tassert.Error(t, err)\n\t\tassert.Contains(t, err.Error(), \"source is required\")\n\t})\n\n\tt.Run(\"empty event_type returns error\", func(t *testing.T) {\n\t\treq := &types.EventRequest{\n\t\t\tMemberID: \"robot_001\",\n\t\t\tSource: \"webhook\",\n\t\t\tEventType: \"\",\n\t\t}\n\t\terr := trigger.ValidateEvent(req)\n\t\tassert.Error(t, err)\n\t\tassert.Contains(t, err.Error(), \"event_type is required\")\n\t})\n\n\tt.Run(\"valid request passes\", func(t *testing.T) {\n\t\treq := &types.EventRequest{\n\t\t\tMemberID: \"robot_001\",\n\t\t\tSource: \"webhook\",\n\t\t\tEventType: \"lead.created\",\n\t\t\tData: map[string]interface{}{\"name\": \"John\"},\n\t\t}\n\t\terr := trigger.ValidateEvent(req)\n\t\tassert.NoError(t, err)\n\t})\n}\n\n// ==================== BuildEventInput Tests ====================\n\nfunc TestBuildEventInput(t *testing.T) {\n\tt.Run(\"builds correct TriggerInput\", func(t *testing.T) {\n\t\treq := &types.EventRequest{\n\t\t\tMemberID: \"robot_001\",\n\t\t\tSource: \"webhook\",\n\t\t\tEventType: \"lead.created\",\n\t\t\tData: map[string]interface{}{\"name\": \"John\", \"email\": \"john@example.com\"},\n\t\t}\n\n\t\tinput := trigger.BuildEventInput(req)\n\n\t\tassert.NotNil(t, input)\n\t\tassert.Equal(t, types.EventSource(\"webhook\"), input.Source)\n\t\tassert.Equal(t, \"lead.created\", input.EventType)\n\t\tassert.Equal(t, \"John\", input.Data[\"name\"])\n\t\tassert.Equal(t, \"john@example.com\", input.Data[\"email\"])\n\t})\n\n\tt.Run(\"handles nil data\", func(t *testing.T) {\n\t\treq := &types.EventRequest{\n\t\t\tMemberID: \"robot_001\",\n\t\t\tSource: \"database\",\n\t\t\tEventType: \"order.paid\",\n\t\t\tData: nil,\n\t\t}\n\n\t\tinput := trigger.BuildEventInput(req)\n\n\t\tassert.NotNil(t, input)\n\t\tassert.Equal(t, types.EventSource(\"database\"), input.Source)\n\t\tassert.Equal(t, \"order.paid\", input.EventType)\n\t\tassert.Nil(t, input.Data)\n\t})\n}\n\n// ==================== GetActionCategory Tests ====================\n\nfunc TestGetActionCategory(t *testing.T) {\n\ttests := []struct {\n\t\taction types.InterventionAction\n\t\texpected string\n\t}{\n\t\t{types.ActionTaskAdd, \"task\"},\n\t\t{types.ActionTaskCancel, \"task\"},\n\t\t{types.ActionTaskUpdate, \"task\"},\n\t\t{types.ActionGoalAdjust, \"goal\"},\n\t\t{types.ActionGoalAdd, \"goal\"},\n\t\t{types.ActionGoalComplete, \"goal\"},\n\t\t{types.ActionGoalCancel, \"goal\"},\n\t\t{types.ActionPlanAdd, \"plan\"},\n\t\t{types.ActionPlanRemove, \"plan\"},\n\t\t{types.ActionPlanUpdate, \"plan\"},\n\t\t{types.ActionInstruct, \"instruct\"},\n\t\t{types.InterventionAction(\"unknown\"), \"unknown\"},\n\t}\n\n\tfor _, tt := range tests {\n\t\tt.Run(string(tt.action), func(t *testing.T) {\n\t\t\tresult := trigger.GetActionCategory(tt.action)\n\t\t\tassert.Equal(t, tt.expected, result)\n\t\t})\n\t}\n}\n\n// ==================== GetActionDescription Tests ====================\n\nfunc TestGetActionDescription(t *testing.T) {\n\ttests := []struct {\n\t\taction types.InterventionAction\n\t\tcontains string\n\t}{\n\t\t{types.ActionTaskAdd, \"Add\"},\n\t\t{types.ActionTaskCancel, \"Cancel\"},\n\t\t{types.ActionTaskUpdate, \"Update\"},\n\t\t{types.ActionGoalAdjust, \"Adjust\"},\n\t\t{types.ActionGoalAdd, \"Add\"},\n\t\t{types.ActionGoalComplete, \"complete\"},\n\t\t{types.ActionGoalCancel, \"Cancel\"},\n\t\t{types.ActionPlanAdd, \"plan\"},\n\t\t{types.ActionPlanRemove, \"Remove\"},\n\t\t{types.ActionPlanUpdate, \"Update\"},\n\t\t{types.ActionInstruct, \"instruction\"},\n\t\t{types.InterventionAction(\"unknown\"), \"Unknown\"},\n\t}\n\n\tfor _, tt := range tests {\n\t\tt.Run(string(tt.action), func(t *testing.T) {\n\t\t\tresult := trigger.GetActionDescription(tt.action)\n\t\t\tassert.NotEmpty(t, result)\n\t\t\tassert.Contains(t, result, tt.contains)\n\t\t})\n\t}\n}\n" }, { "path": "agent/robot/types/clock.go", "content": "package types\n\nimport \"time\"\n\n// ClockContext - time context for P0 inspiration\ntype ClockContext struct {\n\tNow time.Time `json:\"now\"`\n\tHour int `json:\"hour\"` // 0-23\n\tDayOfWeek string `json:\"day_of_week\"` // Monday, Tuesday...\n\tDayOfMonth int `json:\"day_of_month\"` // 1-31\n\tWeekOfYear int `json:\"week_of_year\"` // 1-52\n\tMonth int `json:\"month\"` // 1-12\n\tYear int `json:\"year\"`\n\tIsWeekend bool `json:\"is_weekend\"`\n\tIsMonthStart bool `json:\"is_month_start\"` // 1st-3rd\n\tIsMonthEnd bool `json:\"is_month_end\"` // last 3 days\n\tIsQuarterEnd bool `json:\"is_quarter_end\"`\n\tIsYearEnd bool `json:\"is_year_end\"`\n\tTZ string `json:\"tz\"`\n}\n\n// NewClockContext creates clock context from time\nfunc NewClockContext(t time.Time, tz string) *ClockContext {\n\tloc := time.Local\n\tif tz != \"\" {\n\t\tif l, err := time.LoadLocation(tz); err == nil {\n\t\t\tloc = l\n\t\t}\n\t}\n\tt = t.In(loc)\n\n\t_, week := t.ISOWeek()\n\tdayOfMonth := t.Day()\n\tlastDay := time.Date(t.Year(), t.Month()+1, 0, 0, 0, 0, 0, loc).Day()\n\n\treturn &ClockContext{\n\t\tNow: t,\n\t\tHour: t.Hour(),\n\t\tDayOfWeek: t.Weekday().String(),\n\t\tDayOfMonth: dayOfMonth,\n\t\tWeekOfYear: week,\n\t\tMonth: int(t.Month()),\n\t\tYear: t.Year(),\n\t\tIsWeekend: t.Weekday() == time.Saturday || t.Weekday() == time.Sunday,\n\t\tIsMonthStart: dayOfMonth <= 3,\n\t\tIsMonthEnd: dayOfMonth >= lastDay-2,\n\t\tIsQuarterEnd: (t.Month()%3 == 0) && dayOfMonth >= lastDay-2,\n\t\tIsYearEnd: t.Month() == 12 && dayOfMonth >= 29,\n\t\tTZ: loc.String(),\n\t}\n}\n" }, { "path": "agent/robot/types/clock_test.go", "content": "package types_test\n\nimport (\n\t\"testing\"\n\t\"time\"\n\n\t\"github.com/stretchr/testify/assert\"\n\t\"github.com/yaoapp/yao/agent/robot/types\"\n)\n\nfunc TestNewClockContext(t *testing.T) {\n\tt.Run(\"basic clock context\", func(t *testing.T) {\n\t\t// Test with a known date: 2024-01-15 14:30:00 (Monday)\n\t\ttestTime := time.Date(2024, 1, 15, 14, 30, 0, 0, time.UTC)\n\t\tctx := types.NewClockContext(testTime, \"UTC\")\n\n\t\tassert.Equal(t, 14, ctx.Hour)\n\t\tassert.Equal(t, \"Monday\", ctx.DayOfWeek)\n\t\tassert.Equal(t, 15, ctx.DayOfMonth)\n\t\tassert.Equal(t, 1, ctx.Month)\n\t\tassert.Equal(t, 2024, ctx.Year)\n\t\tassert.False(t, ctx.IsWeekend)\n\t\tassert.False(t, ctx.IsMonthStart)\n\t\tassert.False(t, ctx.IsMonthEnd)\n\t\tassert.False(t, ctx.IsQuarterEnd)\n\t\tassert.False(t, ctx.IsYearEnd)\n\t})\n\n\tt.Run(\"weekend detection\", func(t *testing.T) {\n\t\t// Saturday\n\t\tsaturday := time.Date(2024, 1, 13, 10, 0, 0, 0, time.UTC)\n\t\tctx := types.NewClockContext(saturday, \"\")\n\t\tassert.True(t, ctx.IsWeekend)\n\n\t\t// Sunday\n\t\tsunday := time.Date(2024, 1, 14, 10, 0, 0, 0, time.UTC)\n\t\tctx = types.NewClockContext(sunday, \"\")\n\t\tassert.True(t, ctx.IsWeekend)\n\t})\n\n\tt.Run(\"month start detection\", func(t *testing.T) {\n\t\t// 1st day\n\t\tday1 := time.Date(2024, 1, 1, 10, 0, 0, 0, time.UTC)\n\t\tctx := types.NewClockContext(day1, \"\")\n\t\tassert.True(t, ctx.IsMonthStart)\n\n\t\t// 3rd day\n\t\tday3 := time.Date(2024, 1, 3, 10, 0, 0, 0, time.UTC)\n\t\tctx = types.NewClockContext(day3, \"\")\n\t\tassert.True(t, ctx.IsMonthStart)\n\n\t\t// 4th day - not month start\n\t\tday4 := time.Date(2024, 1, 4, 10, 0, 0, 0, time.UTC)\n\t\tctx = types.NewClockContext(day4, \"\")\n\t\tassert.False(t, ctx.IsMonthStart)\n\t})\n\n\tt.Run(\"month end detection\", func(t *testing.T) {\n\t\t// Last day of January (31st)\n\t\tlastDay := time.Date(2024, 1, 31, 10, 0, 0, 0, time.UTC)\n\t\tctx := types.NewClockContext(lastDay, \"\")\n\t\tassert.True(t, ctx.IsMonthEnd)\n\n\t\t// 29th day of January (31 days total)\n\t\tday29 := time.Date(2024, 1, 29, 10, 0, 0, 0, time.UTC)\n\t\tctx = types.NewClockContext(day29, \"\")\n\t\tassert.True(t, ctx.IsMonthEnd)\n\n\t\t// 28th day of January - not month end\n\t\tday28 := time.Date(2024, 1, 28, 10, 0, 0, 0, time.UTC)\n\t\tctx = types.NewClockContext(day28, \"\")\n\t\tassert.False(t, ctx.IsMonthEnd)\n\t})\n\n\tt.Run(\"quarter end detection\", func(t *testing.T) {\n\t\t// March 31 - Q1 end\n\t\tq1End := time.Date(2024, 3, 31, 10, 0, 0, 0, time.UTC)\n\t\tctx := types.NewClockContext(q1End, \"\")\n\t\tassert.True(t, ctx.IsQuarterEnd)\n\n\t\t// June 30 - Q2 end\n\t\tq2End := time.Date(2024, 6, 30, 10, 0, 0, 0, time.UTC)\n\t\tctx = types.NewClockContext(q2End, \"\")\n\t\tassert.True(t, ctx.IsQuarterEnd)\n\n\t\t// September 30 - Q3 end\n\t\tq3End := time.Date(2024, 9, 30, 10, 0, 0, 0, time.UTC)\n\t\tctx = types.NewClockContext(q3End, \"\")\n\t\tassert.True(t, ctx.IsQuarterEnd)\n\n\t\t// December 31 - Q4 end\n\t\tq4End := time.Date(2024, 12, 31, 10, 0, 0, 0, time.UTC)\n\t\tctx = types.NewClockContext(q4End, \"\")\n\t\tassert.True(t, ctx.IsQuarterEnd)\n\n\t\t// Not quarter end\n\t\tnotQEnd := time.Date(2024, 1, 15, 10, 0, 0, 0, time.UTC)\n\t\tctx = types.NewClockContext(notQEnd, \"\")\n\t\tassert.False(t, ctx.IsQuarterEnd)\n\t})\n\n\tt.Run(\"year end detection\", func(t *testing.T) {\n\t\t// December 29\n\t\tdec29 := time.Date(2024, 12, 29, 10, 0, 0, 0, time.UTC)\n\t\tctx := types.NewClockContext(dec29, \"\")\n\t\tassert.True(t, ctx.IsYearEnd)\n\n\t\t// December 31\n\t\tdec31 := time.Date(2024, 12, 31, 10, 0, 0, 0, time.UTC)\n\t\tctx = types.NewClockContext(dec31, \"\")\n\t\tassert.True(t, ctx.IsYearEnd)\n\n\t\t// December 28 - not year end\n\t\tdec28 := time.Date(2024, 12, 28, 10, 0, 0, 0, time.UTC)\n\t\tctx = types.NewClockContext(dec28, \"\")\n\t\tassert.False(t, ctx.IsYearEnd)\n\n\t\t// January - not year end\n\t\tjan1 := time.Date(2024, 1, 1, 10, 0, 0, 0, time.UTC)\n\t\tctx = types.NewClockContext(jan1, \"\")\n\t\tassert.False(t, ctx.IsYearEnd)\n\t})\n\n\tt.Run(\"timezone handling\", func(t *testing.T) {\n\t\ttestTime := time.Date(2024, 1, 15, 14, 30, 0, 0, time.UTC)\n\n\t\t// With Asia/Shanghai timezone\n\t\tctx := types.NewClockContext(testTime, \"Asia/Shanghai\")\n\t\tassert.Equal(t, \"Asia/Shanghai\", ctx.TZ)\n\t\t// Time should be converted to Shanghai timezone\n\t\tassert.NotEqual(t, testTime, ctx.Now)\n\t\tassert.Equal(t, 22, ctx.Hour) // UTC 14:00 = Shanghai 22:00 (UTC+8)\n\n\t\t// With invalid timezone - should fall back to local\n\t\tctx = types.NewClockContext(testTime, \"Invalid/Timezone\")\n\t\tassert.NotEmpty(t, ctx.TZ)\n\t})\n\n\tt.Run(\"week of year\", func(t *testing.T) {\n\t\t// First week of 2024\n\t\tjan1 := time.Date(2024, 1, 1, 10, 0, 0, 0, time.UTC)\n\t\tctx := types.NewClockContext(jan1, \"\")\n\t\tassert.Equal(t, 1, ctx.WeekOfYear)\n\n\t\t// Mid year\n\t\tjuly15 := time.Date(2024, 7, 15, 10, 0, 0, 0, time.UTC)\n\t\tctx = types.NewClockContext(july15, \"\")\n\t\tassert.Greater(t, ctx.WeekOfYear, 20)\n\t\tassert.Less(t, ctx.WeekOfYear, 35)\n\t})\n}\n\nfunc TestClockContextFields(t *testing.T) {\n\t// Test all fields are populated correctly\n\ttestTime := time.Date(2024, 12, 30, 23, 45, 30, 0, time.UTC)\n\tctx := types.NewClockContext(testTime, \"UTC\")\n\n\tassert.NotZero(t, ctx.Now)\n\tassert.Equal(t, 23, ctx.Hour)\n\tassert.Equal(t, \"Monday\", ctx.DayOfWeek)\n\tassert.Equal(t, 30, ctx.DayOfMonth)\n\tassert.Equal(t, 1, ctx.WeekOfYear) // Dec 30, 2024 is week 1 of 2025\n\tassert.Equal(t, 12, ctx.Month)\n\tassert.Equal(t, 2024, ctx.Year)\n\tassert.False(t, ctx.IsWeekend) // Monday\n\tassert.False(t, ctx.IsMonthStart)\n\tassert.True(t, ctx.IsMonthEnd)\n\tassert.True(t, ctx.IsQuarterEnd)\n\tassert.True(t, ctx.IsYearEnd)\n\tassert.Equal(t, \"UTC\", ctx.TZ)\n}\n" }, { "path": "agent/robot/types/config.go", "content": "package types\n\nimport (\n\t\"encoding/json\"\n\t\"time\"\n)\n\n// Config - robot_config in __yao.member\ntype Config struct {\n\tTriggers *Triggers `json:\"triggers,omitempty\"`\n\tClock *Clock `json:\"clock,omitempty\"`\n\tIdentity *Identity `json:\"identity\"`\n\tQuota *Quota `json:\"quota,omitempty\"`\n\tKB *KB `json:\"kb,omitempty\"` // shared knowledge base (same as assistant)\n\tDB *DB `json:\"db,omitempty\"` // shared database (same as assistant)\n\tLearn *Learn `json:\"learn,omitempty\"` // learning config for private KB\n\tResources *Resources `json:\"resources,omitempty\"`\n\tDelivery *DeliveryPreferences `json:\"delivery,omitempty\"` // delivery preferences (see robot.go)\n\tEvents []Event `json:\"events,omitempty\"`\n\tExecutor *ExecutorConfig `json:\"executor,omitempty\"` // executor mode settings\n\tDefaultLocale string `json:\"default_locale,omitempty\"` // default language for clock/event triggers (\"en\", \"zh\")\n\tIntegrations *Integrations `json:\"integrations,omitempty\"` // external channel integrations (telegram, etc.)\n}\n\n// Integrations holds configuration for external platform integrations.\ntype Integrations struct {\n\tTelegram *TelegramConfig `json:\"telegram,omitempty\"`\n\tFeishu *FeishuConfig `json:\"feishu,omitempty\"`\n\tDingTalk *DingTalkConfig `json:\"dingtalk,omitempty\"`\n\tDiscord *DiscordConfig `json:\"discord,omitempty\"`\n}\n\n// TelegramConfig holds Telegram Bot integration settings.\ntype TelegramConfig struct {\n\tEnabled bool `json:\"enabled\"`\n\tBotToken string `json:\"bot_token\"`\n\tHost string `json:\"host,omitempty\"` // custom Bot API server, defaults to https://api.telegram.org\n\tAppID string `json:\"app_id,omitempty\"` // auto-generated, used for webhook URL routing\n\tChatID string `json:\"chat_id,omitempty\"` // default reply chat\n\tWebhookSecret string `json:\"webhook_secret,omitempty\"` // sent with SetWebhook, verified on incoming webhooks\n}\n\n// FeishuConfig holds Feishu (Lark) Bot integration settings.\ntype FeishuConfig struct {\n\tEnabled bool `json:\"enabled\"`\n\tAppID string `json:\"app_id\"`\n\tAppSecret string `json:\"app_secret\"`\n}\n\n// DingTalkConfig holds DingTalk Bot integration settings.\ntype DingTalkConfig struct {\n\tEnabled bool `json:\"enabled\"`\n\tClientID string `json:\"client_id\"`\n\tClientSecret string `json:\"client_secret\"`\n}\n\n// DiscordConfig holds Discord Bot integration settings.\ntype DiscordConfig struct {\n\tEnabled bool `json:\"enabled\"`\n\tBotToken string `json:\"bot_token\"`\n\tAppID string `json:\"app_id,omitempty\"`\n}\n\n// ExecutorConfig - executor settings\ntype ExecutorConfig struct {\n\tMode ExecutorMode `json:\"mode,omitempty\"` // standard | dryrun | sandbox\n\tMaxDuration string `json:\"max_duration,omitempty\"` // max execution time (e.g., \"30m\")\n}\n\n// GetMode returns the executor mode (default: standard)\nfunc (e *ExecutorConfig) GetMode() ExecutorMode {\n\tif e == nil || e.Mode == \"\" {\n\t\treturn ExecutorStandard\n\t}\n\treturn e.Mode\n}\n\n// GetMaxDuration returns the max duration (default: 30m)\nfunc (e *ExecutorConfig) GetMaxDuration() time.Duration {\n\tif e == nil || e.MaxDuration == \"\" {\n\t\treturn 30 * time.Minute\n\t}\n\td, err := time.ParseDuration(e.MaxDuration)\n\tif err != nil {\n\t\treturn 30 * time.Minute\n\t}\n\treturn d\n}\n\n// Validate validates the config\nfunc (c *Config) Validate() error {\n\tif c.Identity == nil || c.Identity.Role == \"\" {\n\t\treturn ErrMissingIdentity\n\t}\n\tif c.Clock != nil {\n\t\tif err := c.Clock.Validate(); err != nil {\n\t\t\treturn err\n\t\t}\n\t}\n\treturn nil\n}\n\n// GetDefaultLocale returns the default locale (default: \"en\")\nfunc (c *Config) GetDefaultLocale() string {\n\tif c == nil || c.DefaultLocale == \"\" {\n\t\treturn \"en\"\n\t}\n\treturn c.DefaultLocale\n}\n\n// Triggers - trigger enable/disable\ntype Triggers struct {\n\tClock *TriggerSwitch `json:\"clock,omitempty\"`\n\tIntervene *TriggerSwitch `json:\"intervene,omitempty\"`\n\tEvent *TriggerSwitch `json:\"event,omitempty\"`\n}\n\n// TriggerSwitch - trigger enable/disable switch\ntype TriggerSwitch struct {\n\tEnabled bool `json:\"enabled\"`\n\tActions []string `json:\"actions,omitempty\"` // for intervene\n}\n\n// IsEnabled checks if trigger is enabled (default: true)\nfunc (t *Triggers) IsEnabled(typ TriggerType) bool {\n\tif t == nil {\n\t\treturn true\n\t}\n\tswitch typ {\n\tcase TriggerClock:\n\t\treturn t.Clock == nil || t.Clock.Enabled\n\tcase TriggerHuman:\n\t\treturn t.Intervene == nil || t.Intervene.Enabled\n\tcase TriggerEvent:\n\t\treturn t.Event == nil || t.Event.Enabled\n\t}\n\treturn false\n}\n\n// Clock - when to wake up\ntype Clock struct {\n\tMode ClockMode `json:\"mode\"` // times | interval | daemon\n\tTimes []string `json:\"times,omitempty\"` // [\"09:00\", \"14:00\"]\n\tDays []string `json:\"days,omitempty\"` // [\"Mon\", \"Tue\"] or [\"*\"]\n\tEvery string `json:\"every,omitempty\"` // \"30m\", \"1h\"\n\tTZ string `json:\"tz,omitempty\"` // \"Asia/Shanghai\"\n\tTimeout string `json:\"timeout,omitempty\"` // \"30m\"\n}\n\n// Validate validates clock config\nfunc (c *Clock) Validate() error {\n\tswitch c.Mode {\n\tcase ClockTimes:\n\t\tif len(c.Times) == 0 {\n\t\t\treturn ErrClockTimesEmpty\n\t\t}\n\tcase ClockInterval:\n\t\tif c.Every == \"\" {\n\t\t\treturn ErrClockIntervalEmpty\n\t\t}\n\tcase ClockDaemon:\n\t\t// no extra validation\n\tdefault:\n\t\treturn ErrClockModeInvalid\n\t}\n\treturn nil\n}\n\n// GetTimeout returns parsed timeout duration\nfunc (c *Clock) GetTimeout() time.Duration {\n\tif c.Timeout == \"\" {\n\t\treturn 30 * time.Minute // default\n\t}\n\td, err := time.ParseDuration(c.Timeout)\n\tif err != nil {\n\t\treturn 30 * time.Minute\n\t}\n\treturn d\n}\n\n// GetLocation returns timezone location\nfunc (c *Clock) GetLocation() *time.Location {\n\tif c.TZ == \"\" {\n\t\treturn time.Local\n\t}\n\tloc, err := time.LoadLocation(c.TZ)\n\tif err != nil {\n\t\treturn time.Local\n\t}\n\treturn loc\n}\n\n// Identity - who is this robot\ntype Identity struct {\n\tRole string `json:\"role\"`\n\tDuties []string `json:\"duties,omitempty\"`\n\tRules []string `json:\"rules,omitempty\"`\n}\n\n// Quota - concurrency limits\ntype Quota struct {\n\tMax int `json:\"max\"` // max running (default: 2)\n\tQueue int `json:\"queue\"` // queue size (default: 10)\n\tPriority int `json:\"priority\"` // 1-10 (default: 5)\n}\n\n// GetMax returns max with default\nfunc (q *Quota) GetMax() int {\n\tif q == nil || q.Max <= 0 {\n\t\treturn 2\n\t}\n\treturn q.Max\n}\n\n// GetQueue returns queue size with default\nfunc (q *Quota) GetQueue() int {\n\tif q == nil || q.Queue <= 0 {\n\t\treturn 10\n\t}\n\treturn q.Queue\n}\n\n// GetPriority returns priority with default\nfunc (q *Quota) GetPriority() int {\n\tif q == nil || q.Priority <= 0 {\n\t\treturn 5\n\t}\n\treturn q.Priority\n}\n\n// KB - knowledge base config (same as assistant, from store/types)\n// Shared KB collections accessible by this robot\ntype KB struct {\n\tCollections []string `json:\"collections,omitempty\"` // KB collection IDs\n\tOptions map[string]interface{} `json:\"options,omitempty\"`\n}\n\n// DB - database config (same as assistant, from store/types)\n// Shared database models accessible by this robot\ntype DB struct {\n\tModels []string `json:\"models,omitempty\"` // database model names\n\tOptions map[string]interface{} `json:\"options,omitempty\"`\n}\n\n// Learn - learning config for robot's private KB\n// Private KB is auto-created: robot_{team_id}_{member_id}_kb\ntype Learn struct {\n\tOn bool `json:\"on\"`\n\tTypes []string `json:\"types,omitempty\"` // execution, feedback, insight\n\tKeep int `json:\"keep,omitempty\"` // days, 0 = forever\n}\n\n// Resources - available agents and tools\ntype Resources struct {\n\tPhases map[Phase]string `json:\"phases,omitempty\"` // phase -> agent ID\n\tAgents []string `json:\"agents,omitempty\"`\n\tMCP []MCPConfig `json:\"mcp,omitempty\"`\n}\n\n// GetPhaseAgent returns agent ID for phase (default: __yao.{phase})\nfunc (r *Resources) GetPhaseAgent(phase Phase) string {\n\tif r != nil && r.Phases != nil {\n\t\tif id, ok := r.Phases[phase]; ok && id != \"\" {\n\t\t\treturn id\n\t\t}\n\t}\n\treturn \"__yao.\" + string(phase)\n}\n\n// MCPConfig - MCP server configuration\ntype MCPConfig struct {\n\tID string `json:\"id\"`\n\tTools []string `json:\"tools,omitempty\"` // empty = all\n}\n\n// Event - event trigger config\ntype Event struct {\n\tType EventSource `json:\"type\"` // webhook | database\n\tSource string `json:\"source\"` // webhook path or table name\n\tFilter map[string]interface{} `json:\"filter,omitempty\"`\n}\n\n// ParseConfig parses robot_config from various formats (string, []byte, map)\nfunc ParseConfig(data interface{}) (*Config, error) {\n\tif data == nil {\n\t\treturn nil, nil\n\t}\n\n\tvar configBytes []byte\n\n\tswitch v := data.(type) {\n\tcase string:\n\t\tif v == \"\" {\n\t\t\treturn nil, nil\n\t\t}\n\t\tconfigBytes = []byte(v)\n\tcase []byte:\n\t\tif len(v) == 0 {\n\t\t\treturn nil, nil\n\t\t}\n\t\tconfigBytes = v\n\tcase map[string]interface{}:\n\t\tvar err error\n\t\tconfigBytes, err = json.Marshal(v)\n\t\tif err != nil {\n\t\t\treturn nil, err\n\t\t}\n\tdefault:\n\t\tvar err error\n\t\tconfigBytes, err = json.Marshal(v)\n\t\tif err != nil {\n\t\t\treturn nil, err\n\t\t}\n\t}\n\n\tvar config Config\n\tif err := json.Unmarshal(configBytes, &config); err != nil {\n\t\treturn nil, err\n\t}\n\n\treturn &config, nil\n}\n" }, { "path": "agent/robot/types/config_global.go", "content": "package types\n\nimport \"sync\"\n\n// Global configuration for robot agent\n// These values can be set during agent initialization\n\nvar (\n\t// defaultEmailChannel - default messenger channel name for sending emails\n\t// Can be configured via SetDefaultEmailChannel()\n\t// Default: \"default\" (maps to messengers/channels.yao configuration)\n\tdefaultEmailChannel = \"default\"\n\n\t// configMu protects global configuration\n\tconfigMu sync.RWMutex\n)\n\n// DefaultEmailChannel returns the default email channel name\nfunc DefaultEmailChannel() string {\n\tconfigMu.RLock()\n\tdefer configMu.RUnlock()\n\treturn defaultEmailChannel\n}\n\n// SetDefaultEmailChannel sets the default messenger channel for email delivery\n// This should be called during agent initialization\n// The channel name must match a channel defined in messengers/channels.yao\nfunc SetDefaultEmailChannel(channel string) {\n\tif channel == \"\" {\n\t\treturn\n\t}\n\tconfigMu.Lock()\n\tdefer configMu.Unlock()\n\tdefaultEmailChannel = channel\n}\n" }, { "path": "agent/robot/types/config_test.go", "content": "package types_test\n\nimport (\n\t\"testing\"\n\t\"time\"\n\n\t\"github.com/stretchr/testify/assert\"\n\t\"github.com/yaoapp/yao/agent/robot/types\"\n)\n\nfunc TestConfigValidate(t *testing.T) {\n\tt.Run(\"valid config\", func(t *testing.T) {\n\t\tconfig := &types.Config{\n\t\t\tIdentity: &types.Identity{\n\t\t\t\tRole: \"Sales Manager\",\n\t\t\t},\n\t\t}\n\t\terr := config.Validate()\n\t\tassert.NoError(t, err)\n\t})\n\n\tt.Run(\"missing identity\", func(t *testing.T) {\n\t\tconfig := &types.Config{}\n\t\terr := config.Validate()\n\t\tassert.Error(t, err)\n\t\tassert.Equal(t, types.ErrMissingIdentity, err)\n\t})\n\n\tt.Run(\"missing identity role\", func(t *testing.T) {\n\t\tconfig := &types.Config{\n\t\t\tIdentity: &types.Identity{},\n\t\t}\n\t\terr := config.Validate()\n\t\tassert.Error(t, err)\n\t\tassert.Equal(t, types.ErrMissingIdentity, err)\n\t})\n\n\tt.Run(\"invalid clock config\", func(t *testing.T) {\n\t\tconfig := &types.Config{\n\t\t\tIdentity: &types.Identity{Role: \"Test\"},\n\t\t\tClock: &types.Clock{\n\t\t\t\tMode: types.ClockTimes,\n\t\t\t\t// Times is empty - should fail\n\t\t\t},\n\t\t}\n\t\terr := config.Validate()\n\t\tassert.Error(t, err)\n\t\tassert.Equal(t, types.ErrClockTimesEmpty, err)\n\t})\n}\n\nfunc TestClockValidate(t *testing.T) {\n\tt.Run(\"valid times mode\", func(t *testing.T) {\n\t\tclock := &types.Clock{\n\t\t\tMode: types.ClockTimes,\n\t\t\tTimes: []string{\"09:00\", \"14:00\"},\n\t\t}\n\t\terr := clock.Validate()\n\t\tassert.NoError(t, err)\n\t})\n\n\tt.Run(\"times mode without times\", func(t *testing.T) {\n\t\tclock := &types.Clock{\n\t\t\tMode: types.ClockTimes,\n\t\t}\n\t\terr := clock.Validate()\n\t\tassert.Error(t, err)\n\t\tassert.Equal(t, types.ErrClockTimesEmpty, err)\n\t})\n\n\tt.Run(\"valid interval mode\", func(t *testing.T) {\n\t\tclock := &types.Clock{\n\t\t\tMode: types.ClockInterval,\n\t\t\tEvery: \"30m\",\n\t\t}\n\t\terr := clock.Validate()\n\t\tassert.NoError(t, err)\n\t})\n\n\tt.Run(\"interval mode without every\", func(t *testing.T) {\n\t\tclock := &types.Clock{\n\t\t\tMode: types.ClockInterval,\n\t\t}\n\t\terr := clock.Validate()\n\t\tassert.Error(t, err)\n\t\tassert.Equal(t, types.ErrClockIntervalEmpty, err)\n\t})\n\n\tt.Run(\"valid daemon mode\", func(t *testing.T) {\n\t\tclock := &types.Clock{\n\t\t\tMode: types.ClockDaemon,\n\t\t}\n\t\terr := clock.Validate()\n\t\tassert.NoError(t, err)\n\t})\n\n\tt.Run(\"invalid mode\", func(t *testing.T) {\n\t\tclock := &types.Clock{\n\t\t\tMode: types.ClockMode(\"invalid\"),\n\t\t}\n\t\terr := clock.Validate()\n\t\tassert.Error(t, err)\n\t\tassert.Equal(t, types.ErrClockModeInvalid, err)\n\t})\n}\n\nfunc TestClockGetTimeout(t *testing.T) {\n\tt.Run(\"default timeout\", func(t *testing.T) {\n\t\tclock := &types.Clock{}\n\t\ttimeout := clock.GetTimeout()\n\t\tassert.Equal(t, 30*time.Minute, timeout)\n\t})\n\n\tt.Run(\"custom timeout\", func(t *testing.T) {\n\t\tclock := &types.Clock{\n\t\t\tTimeout: \"10m\",\n\t\t}\n\t\ttimeout := clock.GetTimeout()\n\t\tassert.Equal(t, 10*time.Minute, timeout)\n\t})\n\n\tt.Run(\"invalid timeout returns default\", func(t *testing.T) {\n\t\tclock := &types.Clock{\n\t\t\tTimeout: \"invalid\",\n\t\t}\n\t\ttimeout := clock.GetTimeout()\n\t\tassert.Equal(t, 30*time.Minute, timeout)\n\t})\n}\n\nfunc TestClockGetLocation(t *testing.T) {\n\tt.Run(\"default location\", func(t *testing.T) {\n\t\tclock := &types.Clock{}\n\t\tloc := clock.GetLocation()\n\t\tassert.Equal(t, time.Local, loc)\n\t})\n\n\tt.Run(\"valid timezone\", func(t *testing.T) {\n\t\tclock := &types.Clock{\n\t\t\tTZ: \"Asia/Shanghai\",\n\t\t}\n\t\tloc := clock.GetLocation()\n\t\tassert.NotNil(t, loc)\n\t\tassert.Equal(t, \"Asia/Shanghai\", loc.String())\n\t})\n\n\tt.Run(\"invalid timezone returns local\", func(t *testing.T) {\n\t\tclock := &types.Clock{\n\t\t\tTZ: \"Invalid/Timezone\",\n\t\t}\n\t\tloc := clock.GetLocation()\n\t\tassert.Equal(t, time.Local, loc)\n\t})\n}\n\nfunc TestTriggersIsEnabled(t *testing.T) {\n\tt.Run(\"nil triggers - all enabled by default\", func(t *testing.T) {\n\t\tvar triggers *types.Triggers\n\t\tassert.True(t, triggers.IsEnabled(types.TriggerClock))\n\t\tassert.True(t, triggers.IsEnabled(types.TriggerHuman))\n\t\tassert.True(t, triggers.IsEnabled(types.TriggerEvent))\n\t})\n\n\tt.Run(\"clock enabled\", func(t *testing.T) {\n\t\ttriggers := &types.Triggers{\n\t\t\tClock: &types.TriggerSwitch{Enabled: true},\n\t\t}\n\t\tassert.True(t, triggers.IsEnabled(types.TriggerClock))\n\t})\n\n\tt.Run(\"clock disabled\", func(t *testing.T) {\n\t\ttriggers := &types.Triggers{\n\t\t\tClock: &types.TriggerSwitch{Enabled: false},\n\t\t}\n\t\tassert.False(t, triggers.IsEnabled(types.TriggerClock))\n\t})\n\n\tt.Run(\"intervene enabled by default\", func(t *testing.T) {\n\t\ttriggers := &types.Triggers{}\n\t\tassert.True(t, triggers.IsEnabled(types.TriggerHuman))\n\t})\n\n\tt.Run(\"event disabled\", func(t *testing.T) {\n\t\ttriggers := &types.Triggers{\n\t\t\tEvent: &types.TriggerSwitch{Enabled: false},\n\t\t}\n\t\tassert.False(t, triggers.IsEnabled(types.TriggerEvent))\n\t})\n}\n\nfunc TestQuotaDefaults(t *testing.T) {\n\tt.Run(\"nil quota\", func(t *testing.T) {\n\t\tvar quota *types.Quota\n\t\tassert.Equal(t, 2, quota.GetMax())\n\t\tassert.Equal(t, 10, quota.GetQueue())\n\t\tassert.Equal(t, 5, quota.GetPriority())\n\t})\n\n\tt.Run(\"zero values\", func(t *testing.T) {\n\t\tquota := &types.Quota{}\n\t\tassert.Equal(t, 2, quota.GetMax())\n\t\tassert.Equal(t, 10, quota.GetQueue())\n\t\tassert.Equal(t, 5, quota.GetPriority())\n\t})\n\n\tt.Run(\"custom values\", func(t *testing.T) {\n\t\tquota := &types.Quota{\n\t\t\tMax: 5,\n\t\t\tQueue: 20,\n\t\t\tPriority: 8,\n\t\t}\n\t\tassert.Equal(t, 5, quota.GetMax())\n\t\tassert.Equal(t, 20, quota.GetQueue())\n\t\tassert.Equal(t, 8, quota.GetPriority())\n\t})\n}\n\nfunc TestResourcesGetPhaseAgent(t *testing.T) {\n\tt.Run(\"nil resources - returns default\", func(t *testing.T) {\n\t\tvar resources *types.Resources\n\t\tagent := resources.GetPhaseAgent(types.PhaseGoals)\n\t\tassert.Equal(t, \"__yao.goals\", agent)\n\t})\n\n\tt.Run(\"phase not configured - returns default\", func(t *testing.T) {\n\t\tresources := &types.Resources{\n\t\t\tPhases: map[types.Phase]string{},\n\t\t}\n\t\tagent := resources.GetPhaseAgent(types.PhaseGoals)\n\t\tassert.Equal(t, \"__yao.goals\", agent)\n\t})\n\n\tt.Run(\"custom phase agent\", func(t *testing.T) {\n\t\tresources := &types.Resources{\n\t\t\tPhases: map[types.Phase]string{\n\t\t\t\ttypes.PhaseGoals: \"custom.goals.agent\",\n\t\t\t},\n\t\t}\n\t\tagent := resources.GetPhaseAgent(types.PhaseGoals)\n\t\tassert.Equal(t, \"custom.goals.agent\", agent)\n\t})\n\n\tt.Run(\"all phases default names\", func(t *testing.T) {\n\t\tresources := &types.Resources{}\n\t\tassert.Equal(t, \"__yao.inspiration\", resources.GetPhaseAgent(types.PhaseInspiration))\n\t\tassert.Equal(t, \"__yao.goals\", resources.GetPhaseAgent(types.PhaseGoals))\n\t\tassert.Equal(t, \"__yao.tasks\", resources.GetPhaseAgent(types.PhaseTasks))\n\t\tassert.Equal(t, \"__yao.run\", resources.GetPhaseAgent(types.PhaseRun))\n\t\tassert.Equal(t, \"__yao.delivery\", resources.GetPhaseAgent(types.PhaseDelivery))\n\t\tassert.Equal(t, \"__yao.learning\", resources.GetPhaseAgent(types.PhaseLearning))\n\t})\n}\n\nfunc TestExecutorConfigGetMode(t *testing.T) {\n\tt.Run(\"nil config - returns default\", func(t *testing.T) {\n\t\tvar config *types.ExecutorConfig\n\t\tassert.Equal(t, types.ExecutorStandard, config.GetMode())\n\t})\n\n\tt.Run(\"empty mode - returns default\", func(t *testing.T) {\n\t\tconfig := &types.ExecutorConfig{}\n\t\tassert.Equal(t, types.ExecutorStandard, config.GetMode())\n\t})\n\n\tt.Run(\"standard mode\", func(t *testing.T) {\n\t\tconfig := &types.ExecutorConfig{Mode: types.ExecutorStandard}\n\t\tassert.Equal(t, types.ExecutorStandard, config.GetMode())\n\t})\n\n\tt.Run(\"dryrun mode\", func(t *testing.T) {\n\t\tconfig := &types.ExecutorConfig{Mode: types.ExecutorDryRun}\n\t\tassert.Equal(t, types.ExecutorDryRun, config.GetMode())\n\t})\n\n\tt.Run(\"sandbox mode\", func(t *testing.T) {\n\t\tconfig := &types.ExecutorConfig{Mode: types.ExecutorSandbox}\n\t\tassert.Equal(t, types.ExecutorSandbox, config.GetMode())\n\t})\n}\n\nfunc TestExecutorConfigGetMaxDuration(t *testing.T) {\n\tt.Run(\"nil config - returns default 30m\", func(t *testing.T) {\n\t\tvar config *types.ExecutorConfig\n\t\tassert.Equal(t, 30*time.Minute, config.GetMaxDuration())\n\t})\n\n\tt.Run(\"empty duration - returns default 30m\", func(t *testing.T) {\n\t\tconfig := &types.ExecutorConfig{}\n\t\tassert.Equal(t, 30*time.Minute, config.GetMaxDuration())\n\t})\n\n\tt.Run(\"custom duration\", func(t *testing.T) {\n\t\tconfig := &types.ExecutorConfig{MaxDuration: \"10m\"}\n\t\tassert.Equal(t, 10*time.Minute, config.GetMaxDuration())\n\t})\n\n\tt.Run(\"invalid duration - returns default\", func(t *testing.T) {\n\t\tconfig := &types.ExecutorConfig{MaxDuration: \"invalid\"}\n\t\tassert.Equal(t, 30*time.Minute, config.GetMaxDuration())\n\t})\n\n\tt.Run(\"various valid durations\", func(t *testing.T) {\n\t\ttests := []struct {\n\t\t\tinput string\n\t\t\texpected time.Duration\n\t\t}{\n\t\t\t{\"1h\", time.Hour},\n\t\t\t{\"30s\", 30 * time.Second},\n\t\t\t{\"2h30m\", 2*time.Hour + 30*time.Minute},\n\t\t}\n\t\tfor _, tt := range tests {\n\t\t\tconfig := &types.ExecutorConfig{MaxDuration: tt.input}\n\t\t\tassert.Equal(t, tt.expected, config.GetMaxDuration(), \"for input %s\", tt.input)\n\t\t}\n\t})\n}\n" }, { "path": "agent/robot/types/context.go", "content": "package types\n\nimport (\n\t\"context\"\n\n\t\"github.com/yaoapp/yao/openapi/oauth/types\"\n)\n\n// Context - robot execution context (lightweight)\ntype Context struct {\n\tcontext.Context // embed standard context\n\tAuth *types.AuthorizedInfo `json:\"auth,omitempty\"` // reuse oauth AuthorizedInfo\n\tMemberID string `json:\"member_id,omitempty\"` // current robot member ID\n\tRequestID string `json:\"request_id,omitempty\"` // request trace ID\n\tLocale string `json:\"locale,omitempty\"` // locale (e.g., \"en-US\")\n}\n\n// NewContext creates a new robot context\nfunc NewContext(parent context.Context, auth *types.AuthorizedInfo) *Context {\n\tif parent == nil {\n\t\tparent = context.Background()\n\t}\n\treturn &Context{\n\t\tContext: parent,\n\t\tAuth: auth,\n\t}\n}\n\n// UserID returns user ID from auth\nfunc (c *Context) UserID() string {\n\tif c.Auth == nil {\n\t\treturn \"\"\n\t}\n\treturn c.Auth.UserID\n}\n\n// TeamID returns team ID from auth\nfunc (c *Context) TeamID() string {\n\tif c.Auth == nil {\n\t\treturn \"\"\n\t}\n\treturn c.Auth.TeamID\n}\n" }, { "path": "agent/robot/types/enums.go", "content": "package types\n\n// Phase - execution phase\ntype Phase string\n\n// Phase constants define the execution phases for robot agent\nconst (\n\tPhaseInspiration Phase = \"inspiration\" // P0: Clock only\n\tPhaseGoals Phase = \"goals\" // P1\n\tPhaseTasks Phase = \"tasks\" // P2\n\tPhaseRun Phase = \"run\" // P3\n\tPhaseDelivery Phase = \"delivery\" // P4\n\tPhaseLearning Phase = \"learning\" // P5\n\tPhaseHost Phase = \"host\" // V2: Host Agent (human interaction)\n)\n\n// AllPhases lists phases in execution order (PhaseHost is excluded — it is a cross-phase service role, not a pipeline stage)\nvar AllPhases = []Phase{\n\tPhaseInspiration, PhaseGoals, PhaseTasks,\n\tPhaseRun, PhaseDelivery, PhaseLearning,\n}\n\n// AllConfigurablePhases lists phases that can be bound to custom agents\nvar AllConfigurablePhases = []Phase{\n\tPhaseInspiration, PhaseGoals, PhaseTasks,\n\tPhaseRun, PhaseDelivery, PhaseLearning, PhaseHost,\n}\n\n// ClockMode - clock trigger mode\ntype ClockMode string\n\n// ClockMode constants define the clock trigger modes\nconst (\n\tClockTimes ClockMode = \"times\" // run at specific times\n\tClockInterval ClockMode = \"interval\" // run every X duration\n\tClockDaemon ClockMode = \"daemon\" // run continuously\n)\n\n// TriggerType - trigger source\ntype TriggerType string\n\n// TriggerType constants define the trigger sources\nconst (\n\tTriggerClock TriggerType = \"clock\"\n\tTriggerHuman TriggerType = \"human\"\n\tTriggerEvent TriggerType = \"event\"\n)\n\n// ExecStatus - execution status\ntype ExecStatus string\n\n// ExecStatus constants define the execution status values\nconst (\n\tExecPending ExecStatus = \"pending\"\n\tExecRunning ExecStatus = \"running\"\n\tExecPaused ExecStatus = \"paused\"\n\tExecCompleted ExecStatus = \"completed\"\n\tExecFailed ExecStatus = \"failed\"\n\tExecCancelled ExecStatus = \"cancelled\"\n\tExecConfirming ExecStatus = \"confirming\" // V2: awaiting human confirmation before running\n\tExecWaiting ExecStatus = \"waiting\" // V2: suspended, waiting for human input\n)\n\n// RobotStatus - matches __yao.member.robot_status\ntype RobotStatus string\n\n// RobotStatus constants define the robot status values\nconst (\n\tRobotIdle RobotStatus = \"idle\"\n\tRobotWorking RobotStatus = \"working\"\n\tRobotPaused RobotStatus = \"paused\"\n\tRobotError RobotStatus = \"error\"\n\tRobotMaintenance RobotStatus = \"maintenance\"\n)\n\n// InterventionAction - human intervention action\n// Format: category.action (e.g., \"task.add\", \"goal.adjust\")\ntype InterventionAction string\n\n// InterventionAction constants define the human intervention actions\nconst (\n\t// ActionTaskAdd adds a new task\n\tActionTaskAdd InterventionAction = \"task.add\"\n\t// ActionTaskCancel cancels a task\n\tActionTaskCancel InterventionAction = \"task.cancel\"\n\t// ActionTaskUpdate updates task details\n\tActionTaskUpdate InterventionAction = \"task.update\"\n\n\t// ActionGoalAdjust modifies current goal\n\tActionGoalAdjust InterventionAction = \"goal.adjust\"\n\t// ActionGoalAdd adds a new goal\n\tActionGoalAdd InterventionAction = \"goal.add\"\n\t// ActionGoalComplete marks goal as complete\n\tActionGoalComplete InterventionAction = \"goal.complete\"\n\t// ActionGoalCancel cancels a goal\n\tActionGoalCancel InterventionAction = \"goal.cancel\"\n\n\t// ActionPlanAdd adds to plan queue\n\tActionPlanAdd InterventionAction = \"plan.add\"\n\t// ActionPlanRemove removes from plan queue\n\tActionPlanRemove InterventionAction = \"plan.remove\"\n\t// ActionPlanUpdate updates planned item\n\tActionPlanUpdate InterventionAction = \"plan.update\"\n\n\t// ActionInstruct is a direct instruction to robot\n\tActionInstruct InterventionAction = \"instruct\"\n)\n\n// Priority - task/goal priority\ntype Priority string\n\n// Priority constants define the priority levels\nconst (\n\tPriorityHigh Priority = \"high\"\n\tPriorityNormal Priority = \"normal\"\n\tPriorityLow Priority = \"low\"\n)\n\n// DeliveryType - output delivery type\ntype DeliveryType string\n\n// DeliveryType constants define the output delivery types\nconst (\n\tDeliveryEmail DeliveryType = \"email\" // Send via yao/messenger\n\tDeliveryWebhook DeliveryType = \"webhook\" // POST to external URL\n\tDeliveryProcess DeliveryType = \"process\" // Call Yao Process\n\tDeliveryNotify DeliveryType = \"notify\" // In-app notification (future, auto by subscriptions)\n)\n\n// DedupResult - deduplication result\ntype DedupResult string\n\n// DedupResult constants define the deduplication results\nconst (\n\tDedupSkip DedupResult = \"skip\" // skip execution\n\tDedupMerge DedupResult = \"merge\" // merge with existing\n\tDedupProceed DedupResult = \"proceed\" // proceed normally\n)\n\n// EventSource - event trigger source\ntype EventSource string\n\n// EventSource constants define the event trigger sources\nconst (\n\tEventWebhook EventSource = \"webhook\" // HTTP webhook\n\tEventDatabase EventSource = \"database\" // DB change trigger\n)\n\n// LearningType - learning entry type\ntype LearningType string\n\n// LearningType constants define the learning entry types\nconst (\n\tLearnExecution LearningType = \"execution\" // execution record\n\tLearnFeedback LearningType = \"feedback\" // error/fix feedback\n\tLearnInsight LearningType = \"insight\" // pattern/tip insight\n)\n\n// TaskSource - how task was created\ntype TaskSource string\n\n// TaskSource constants define how a task was created\nconst (\n\tTaskSourceAuto TaskSource = \"auto\" // generated by P2 (task planning)\n\tTaskSourceHuman TaskSource = \"human\" // added via human intervention\n\tTaskSourceEvent TaskSource = \"event\" // added via event trigger\n)\n\n// ExecutorType - task executor type\ntype ExecutorType string\n\n// ExecutorType constants define the task executor types\nconst (\n\tExecutorAssistant ExecutorType = \"assistant\"\n\tExecutorMCP ExecutorType = \"mcp\"\n\tExecutorProcess ExecutorType = \"process\"\n)\n\n// TaskStatus - task execution status\ntype TaskStatus string\n\n// TaskStatus constants define the task execution status values\nconst (\n\tTaskPending TaskStatus = \"pending\"\n\tTaskRunning TaskStatus = \"running\"\n\tTaskCompleted TaskStatus = \"completed\"\n\tTaskFailed TaskStatus = \"failed\"\n\tTaskSkipped TaskStatus = \"skipped\"\n\tTaskCancelled TaskStatus = \"cancelled\"\n\tTaskWaitingInput TaskStatus = \"waiting_input\" // V2: task suspended, waiting for human input\n)\n\n// InsertPosition - where to insert task in queue\ntype InsertPosition string\n\n// InsertPosition constants define where to insert task in queue\nconst (\n\tInsertFirst InsertPosition = \"first\" // insert at beginning (highest priority)\n\tInsertLast InsertPosition = \"last\" // append at end (default)\n\tInsertNext InsertPosition = \"next\" // insert after current task\n\tInsertAt InsertPosition = \"at\" // insert at specific index (use AtIndex)\n)\n\n// ExecutorMode - executor mode for robot execution\ntype ExecutorMode string\n\n// ExecutorMode constants define the executor modes\nconst (\n\t// ExecutorStandard uses real Agent calls (production mode)\n\tExecutorStandard ExecutorMode = \"standard\"\n\t// ExecutorDryRun simulates execution without LLM calls (testing/demo)\n\tExecutorDryRun ExecutorMode = \"dryrun\"\n\t// ExecutorSandbox runs in container-isolated environment (NOT IMPLEMENTED)\n\t// Requires Docker/gVisor/Firecracker infrastructure\n\tExecutorSandbox ExecutorMode = \"sandbox\"\n)\n\n// HostAction defines structured instructions from Host Agent to Manager\ntype HostAction string\n\n// HostAction constants\nconst (\n\tHostActionConfirm HostAction = \"confirm\" // Confirm execution plan\n\tHostActionAdjust HostAction = \"adjust\" // Adjust goals/tasks\n\tHostActionAddTask HostAction = \"add_task\" // Inject a new task\n\tHostActionSkip HostAction = \"skip\" // Skip waiting task\n\tHostActionInjectCtx HostAction = \"inject_context\" // Add context to waiting task\n\tHostActionCancel HostAction = \"cancel\" // Cancel execution\n)\n\n// InteractSource defines the source of an interact request\ntype InteractSource string\n\n// InteractSource constants\nconst (\n\tInteractSourceUI InteractSource = \"ui\" // User via Mission Control UI\n\tInteractSourceEmail InteractSource = \"email\" // Incoming email\n\tInteractSourceWebhook InteractSource = \"webhook\" // External webhook\n\tInteractSourceA2A InteractSource = \"a2a\" // Agent-to-agent\n\tInteractSourceCron InteractSource = \"cron\" // Scheduled cron\n)\n\n// IsValid checks if the executor mode is valid\nfunc (m ExecutorMode) IsValid() bool {\n\tswitch m {\n\tcase ExecutorStandard, ExecutorDryRun, ExecutorSandbox, \"\":\n\t\treturn true\n\t}\n\treturn false\n}\n\n// GetDefault returns the default executor mode if empty\nfunc (m ExecutorMode) GetDefault() ExecutorMode {\n\tif m == \"\" {\n\t\treturn ExecutorStandard\n\t}\n\treturn m\n}\n" }, { "path": "agent/robot/types/enums_test.go", "content": "package types_test\n\nimport (\n\t\"testing\"\n\n\t\"github.com/stretchr/testify/assert\"\n\t\"github.com/yaoapp/yao/agent/robot/types\"\n)\n\nfunc TestPhaseEnum(t *testing.T) {\n\tassert.Equal(t, types.Phase(\"inspiration\"), types.PhaseInspiration)\n\tassert.Equal(t, types.Phase(\"goals\"), types.PhaseGoals)\n\tassert.Equal(t, types.Phase(\"tasks\"), types.PhaseTasks)\n\tassert.Equal(t, types.Phase(\"run\"), types.PhaseRun)\n\tassert.Equal(t, types.Phase(\"delivery\"), types.PhaseDelivery)\n\tassert.Equal(t, types.Phase(\"learning\"), types.PhaseLearning)\n}\n\nfunc TestAllPhases(t *testing.T) {\n\t// AllPhases is the execution pipeline — PhaseHost is excluded (it is a cross-phase service role)\n\tassert.Len(t, types.AllPhases, 6)\n\tassert.Equal(t, types.PhaseInspiration, types.AllPhases[0])\n\tassert.Equal(t, types.PhaseGoals, types.AllPhases[1])\n\tassert.Equal(t, types.PhaseTasks, types.AllPhases[2])\n\tassert.Equal(t, types.PhaseRun, types.AllPhases[3])\n\tassert.Equal(t, types.PhaseDelivery, types.AllPhases[4])\n\tassert.Equal(t, types.PhaseLearning, types.AllPhases[5])\n\n\t// AllConfigurablePhases includes PhaseHost for configuration validation\n\tassert.Len(t, types.AllConfigurablePhases, 7)\n\tassert.Contains(t, types.AllConfigurablePhases, types.PhaseHost)\n}\n\nfunc TestClockModeEnum(t *testing.T) {\n\tassert.Equal(t, types.ClockMode(\"times\"), types.ClockTimes)\n\tassert.Equal(t, types.ClockMode(\"interval\"), types.ClockInterval)\n\tassert.Equal(t, types.ClockMode(\"daemon\"), types.ClockDaemon)\n}\n\nfunc TestTriggerTypeEnum(t *testing.T) {\n\tassert.Equal(t, types.TriggerType(\"clock\"), types.TriggerClock)\n\tassert.Equal(t, types.TriggerType(\"human\"), types.TriggerHuman)\n\tassert.Equal(t, types.TriggerType(\"event\"), types.TriggerEvent)\n}\n\nfunc TestExecStatusEnum(t *testing.T) {\n\tassert.Equal(t, types.ExecStatus(\"pending\"), types.ExecPending)\n\tassert.Equal(t, types.ExecStatus(\"running\"), types.ExecRunning)\n\tassert.Equal(t, types.ExecStatus(\"paused\"), types.ExecPaused)\n\tassert.Equal(t, types.ExecStatus(\"completed\"), types.ExecCompleted)\n\tassert.Equal(t, types.ExecStatus(\"failed\"), types.ExecFailed)\n\tassert.Equal(t, types.ExecStatus(\"cancelled\"), types.ExecCancelled)\n}\n\nfunc TestRobotStatusEnum(t *testing.T) {\n\tassert.Equal(t, types.RobotStatus(\"idle\"), types.RobotIdle)\n\tassert.Equal(t, types.RobotStatus(\"working\"), types.RobotWorking)\n\tassert.Equal(t, types.RobotStatus(\"paused\"), types.RobotPaused)\n\tassert.Equal(t, types.RobotStatus(\"error\"), types.RobotError)\n\tassert.Equal(t, types.RobotStatus(\"maintenance\"), types.RobotMaintenance)\n}\n\nfunc TestInterventionActionEnum(t *testing.T) {\n\t// Task operations\n\tassert.Equal(t, types.InterventionAction(\"task.add\"), types.ActionTaskAdd)\n\tassert.Equal(t, types.InterventionAction(\"task.cancel\"), types.ActionTaskCancel)\n\tassert.Equal(t, types.InterventionAction(\"task.update\"), types.ActionTaskUpdate)\n\n\t// Goal operations\n\tassert.Equal(t, types.InterventionAction(\"goal.adjust\"), types.ActionGoalAdjust)\n\tassert.Equal(t, types.InterventionAction(\"goal.add\"), types.ActionGoalAdd)\n\tassert.Equal(t, types.InterventionAction(\"goal.complete\"), types.ActionGoalComplete)\n\tassert.Equal(t, types.InterventionAction(\"goal.cancel\"), types.ActionGoalCancel)\n\n\t// Plan operations\n\tassert.Equal(t, types.InterventionAction(\"plan.add\"), types.ActionPlanAdd)\n\tassert.Equal(t, types.InterventionAction(\"plan.remove\"), types.ActionPlanRemove)\n\tassert.Equal(t, types.InterventionAction(\"plan.update\"), types.ActionPlanUpdate)\n\n\t// Instruction\n\tassert.Equal(t, types.InterventionAction(\"instruct\"), types.ActionInstruct)\n}\n\nfunc TestPriorityEnum(t *testing.T) {\n\tassert.Equal(t, types.Priority(\"high\"), types.PriorityHigh)\n\tassert.Equal(t, types.Priority(\"normal\"), types.PriorityNormal)\n\tassert.Equal(t, types.Priority(\"low\"), types.PriorityLow)\n}\n\nfunc TestDeliveryTypeEnum(t *testing.T) {\n\tassert.Equal(t, types.DeliveryType(\"email\"), types.DeliveryEmail)\n\tassert.Equal(t, types.DeliveryType(\"webhook\"), types.DeliveryWebhook)\n\tassert.Equal(t, types.DeliveryType(\"process\"), types.DeliveryProcess)\n\tassert.Equal(t, types.DeliveryType(\"notify\"), types.DeliveryNotify)\n}\n\nfunc TestDedupResultEnum(t *testing.T) {\n\tassert.Equal(t, types.DedupResult(\"skip\"), types.DedupSkip)\n\tassert.Equal(t, types.DedupResult(\"merge\"), types.DedupMerge)\n\tassert.Equal(t, types.DedupResult(\"proceed\"), types.DedupProceed)\n}\n\nfunc TestEventSourceEnum(t *testing.T) {\n\tassert.Equal(t, types.EventSource(\"webhook\"), types.EventWebhook)\n\tassert.Equal(t, types.EventSource(\"database\"), types.EventDatabase)\n}\n\nfunc TestLearningTypeEnum(t *testing.T) {\n\tassert.Equal(t, types.LearningType(\"execution\"), types.LearnExecution)\n\tassert.Equal(t, types.LearningType(\"feedback\"), types.LearnFeedback)\n\tassert.Equal(t, types.LearningType(\"insight\"), types.LearnInsight)\n}\n\nfunc TestTaskSourceEnum(t *testing.T) {\n\tassert.Equal(t, types.TaskSource(\"auto\"), types.TaskSourceAuto)\n\tassert.Equal(t, types.TaskSource(\"human\"), types.TaskSourceHuman)\n\tassert.Equal(t, types.TaskSource(\"event\"), types.TaskSourceEvent)\n}\n\nfunc TestExecutorTypeEnum(t *testing.T) {\n\tassert.Equal(t, types.ExecutorType(\"assistant\"), types.ExecutorAssistant)\n\tassert.Equal(t, types.ExecutorType(\"mcp\"), types.ExecutorMCP)\n\tassert.Equal(t, types.ExecutorType(\"process\"), types.ExecutorProcess)\n}\n\nfunc TestTaskStatusEnum(t *testing.T) {\n\tassert.Equal(t, types.TaskStatus(\"pending\"), types.TaskPending)\n\tassert.Equal(t, types.TaskStatus(\"running\"), types.TaskRunning)\n\tassert.Equal(t, types.TaskStatus(\"completed\"), types.TaskCompleted)\n\tassert.Equal(t, types.TaskStatus(\"failed\"), types.TaskFailed)\n\tassert.Equal(t, types.TaskStatus(\"skipped\"), types.TaskSkipped)\n\tassert.Equal(t, types.TaskStatus(\"cancelled\"), types.TaskCancelled)\n}\n\nfunc TestInsertPositionEnum(t *testing.T) {\n\tassert.Equal(t, types.InsertPosition(\"first\"), types.InsertFirst)\n\tassert.Equal(t, types.InsertPosition(\"last\"), types.InsertLast)\n\tassert.Equal(t, types.InsertPosition(\"next\"), types.InsertNext)\n\tassert.Equal(t, types.InsertPosition(\"at\"), types.InsertAt)\n}\n\nfunc TestExecutorModeEnum(t *testing.T) {\n\tassert.Equal(t, types.ExecutorMode(\"standard\"), types.ExecutorStandard)\n\tassert.Equal(t, types.ExecutorMode(\"dryrun\"), types.ExecutorDryRun)\n\tassert.Equal(t, types.ExecutorMode(\"sandbox\"), types.ExecutorSandbox)\n}\n\nfunc TestExecutorModeIsValid(t *testing.T) {\n\ttests := []struct {\n\t\tmode types.ExecutorMode\n\t\tvalid bool\n\t}{\n\t\t{types.ExecutorStandard, true},\n\t\t{types.ExecutorDryRun, true},\n\t\t{types.ExecutorSandbox, true},\n\t\t{\"\", true}, // empty is valid (defaults to standard)\n\t\t{types.ExecutorMode(\"invalid\"), false},\n\t\t{types.ExecutorMode(\"unknown\"), false},\n\t}\n\n\tfor _, tt := range tests {\n\t\tt.Run(string(tt.mode), func(t *testing.T) {\n\t\t\tassert.Equal(t, tt.valid, tt.mode.IsValid())\n\t\t})\n\t}\n}\n\nfunc TestExecutorModeGetDefault(t *testing.T) {\n\ttests := []struct {\n\t\tmode types.ExecutorMode\n\t\texpected types.ExecutorMode\n\t}{\n\t\t{\"\", types.ExecutorStandard},\n\t\t{types.ExecutorStandard, types.ExecutorStandard},\n\t\t{types.ExecutorDryRun, types.ExecutorDryRun},\n\t\t{types.ExecutorSandbox, types.ExecutorSandbox},\n\t}\n\n\tfor _, tt := range tests {\n\t\tt.Run(string(tt.mode), func(t *testing.T) {\n\t\t\tassert.Equal(t, tt.expected, tt.mode.GetDefault())\n\t\t})\n\t}\n}\n" }, { "path": "agent/robot/types/errors.go", "content": "package types\n\nimport \"errors\"\n\n// ErrMissingIdentity indicates identity.role is required\nvar ErrMissingIdentity = errors.New(\"identity.role is required\")\n\n// ErrClockTimesEmpty indicates clock.times is required for times mode\nvar ErrClockTimesEmpty = errors.New(\"clock.times is required for times mode\")\n\n// ErrClockIntervalEmpty indicates clock.every is required for interval mode\nvar ErrClockIntervalEmpty = errors.New(\"clock.every is required for interval mode\")\n\n// ErrClockModeInvalid indicates clock.mode must be times, interval, or daemon\nvar ErrClockModeInvalid = errors.New(\"clock.mode must be times, interval, or daemon\")\n\n// ErrRobotNotFound indicates robot not found\nvar ErrRobotNotFound = errors.New(\"robot not found\")\n\n// ErrRobotPaused indicates robot is paused\nvar ErrRobotPaused = errors.New(\"robot is paused\")\n\n// ErrRobotBusy indicates robot has reached max concurrent executions\nvar ErrRobotBusy = errors.New(\"robot has reached max concurrent executions\")\n\n// ErrQuotaExceeded indicates robot quota was exceeded (atomic check failed)\nvar ErrQuotaExceeded = errors.New(\"robot quota exceeded\")\n\n// ErrTriggerDisabled indicates trigger type is disabled for this robot\nvar ErrTriggerDisabled = errors.New(\"trigger type is disabled for this robot\")\n\n// ErrExecutionCancelled indicates execution was cancelled\nvar ErrExecutionCancelled = errors.New(\"execution was cancelled\")\n\n// ErrExecutionTimeout indicates execution timed out\nvar ErrExecutionTimeout = errors.New(\"execution timed out\")\n\n// ErrPhaseAgentNotFound indicates phase agent not found\nvar ErrPhaseAgentNotFound = errors.New(\"phase agent not found\")\n\n// ErrGoalGenFailed indicates goal generation failed\nvar ErrGoalGenFailed = errors.New(\"goal generation failed\")\n\n// ErrTaskPlanFailed indicates task planning failed\nvar ErrTaskPlanFailed = errors.New(\"task planning failed\")\n\n// ErrDeliveryFailed indicates delivery failed\nvar ErrDeliveryFailed = errors.New(\"delivery failed\")\n\n// ErrExecutionSuspended is a sentinel error signaling that execution has been\n// suspended to wait for human input. The executor should persist state and\n// release its worker goroutine. NOT a failure — resumable via Resume().\nvar ErrExecutionSuspended = errors.New(\"execution suspended: waiting for human input\")\n" }, { "path": "agent/robot/types/host.go", "content": "package types\n\nimport agentcontext \"github.com/yaoapp/yao/agent/context\"\n\n// HostInput is the unified input format for Host Agent (§5.7)\ntype HostInput struct {\n\tScenario string `json:\"scenario\"` // \"assign\" | \"guide\" | \"clarify\"\n\tMessages []agentcontext.Message `json:\"messages\"` // Messages from the human\n\tContext *HostContext `json:\"context\"` // Current execution context\n}\n\n// HostContext provides execution context to Host Agent.\n// Note: Goals is *Goals (struct with Content field), serialized as {\"content\":\"...\"}.\n// Host Agent prompts must expect this struct format rather than a plain string.\ntype HostContext struct {\n\tRobotStatus *RobotStatusSnapshot `json:\"robot_status,omitempty\"`\n\tGoals *Goals `json:\"goals,omitempty\"`\n\tTasks []Task `json:\"tasks,omitempty\"`\n\tCurrentTask *Task `json:\"current_task,omitempty\"`\n\tAgentReply string `json:\"agent_reply,omitempty\"`\n\tHistory []agentcontext.Message `json:\"history,omitempty\"`\n}\n\n// HostOutput is the structured output from Host Agent\ntype HostOutput struct {\n\tReply string `json:\"reply\"`\n\tAction HostAction `json:\"action,omitempty\"`\n\tActionData interface{} `json:\"action_data,omitempty\"`\n\tWaitForMore bool `json:\"wait_for_more,omitempty\"`\n}\n" }, { "path": "agent/robot/types/host_test.go", "content": "package types\n\nimport (\n\t\"encoding/json\"\n\t\"testing\"\n\n\t\"github.com/stretchr/testify/assert\"\n\t\"github.com/stretchr/testify/require\"\n)\n\nfunc TestHostInputJSON(t *testing.T) {\n\tinput := &HostInput{\n\t\tScenario: \"assign\",\n\t\tContext: &HostContext{\n\t\t\tRobotStatus: &RobotStatusSnapshot{\n\t\t\t\tActiveCount: 1,\n\t\t\t\tMaxQuota: 5,\n\t\t\t},\n\t\t\tGoals: &Goals{Content: \"test goals\"},\n\t\t},\n\t}\n\n\tdata, err := json.Marshal(input)\n\trequire.NoError(t, err)\n\n\tvar parsed HostInput\n\terr = json.Unmarshal(data, &parsed)\n\trequire.NoError(t, err)\n\tassert.Equal(t, \"assign\", parsed.Scenario)\n\tassert.NotNil(t, parsed.Context)\n\tassert.Equal(t, 1, parsed.Context.RobotStatus.ActiveCount)\n}\n\nfunc TestHostOutputJSON(t *testing.T) {\n\toutput := &HostOutput{\n\t\tReply: \"Task confirmed\",\n\t\tAction: HostActionConfirm,\n\t\tWaitForMore: false,\n\t}\n\n\tdata, err := json.Marshal(output)\n\trequire.NoError(t, err)\n\n\tvar parsed HostOutput\n\terr = json.Unmarshal(data, &parsed)\n\trequire.NoError(t, err)\n\tassert.Equal(t, \"Task confirmed\", parsed.Reply)\n\tassert.Equal(t, HostActionConfirm, parsed.Action)\n\tassert.False(t, parsed.WaitForMore)\n}\n\nfunc TestHostOutputWithActionData(t *testing.T) {\n\toutput := &HostOutput{\n\t\tReply: \"I'll adjust the plan\",\n\t\tAction: HostActionAdjust,\n\t\tActionData: map[string]interface{}{\"goals\": \"adjusted goals\"},\n\t}\n\n\tdata, err := json.Marshal(output)\n\trequire.NoError(t, err)\n\n\tvar parsed HostOutput\n\terr = json.Unmarshal(data, &parsed)\n\trequire.NoError(t, err)\n\tassert.Equal(t, HostActionAdjust, parsed.Action)\n\tassert.NotNil(t, parsed.ActionData)\n}\n" }, { "path": "agent/robot/types/inspiration.go", "content": "package types\n\n// InspirationReport - P0 output (simple markdown for LLM)\ntype InspirationReport struct {\n\tClock *ClockContext `json:\"clock\"` // time context\n\tContent string `json:\"content\"` // markdown text for LLM\n}\n\n// Content is markdown like:\n// ## Summary\n// ...\n// ## Highlights\n// - [High] Sales up 50%\n// - [Medium] New lead from BigCorp\n// ## Opportunities\n// ...\n// ## Risks\n// ...\n// ## World News\n// ...\n// ## Pending\n// ...\n" }, { "path": "agent/robot/types/interfaces.go", "content": "package types\n\nimport \"time\"\n\n// ==================== Internal Interfaces ====================\n// These are internal implementation interfaces, not exposed via API.\n// External API is defined in api/api.go\n// All interfaces use *Context (not context.Context) for consistency.\n\n// ExecutionControl provides pause/resume/stop control for running executions\n// This interface is implemented by trigger.ControlledExecution\ntype ExecutionControl interface {\n\t// IsPaused returns true if execution is paused\n\tIsPaused() bool\n\t// IsCancelled returns true if execution is cancelled\n\tIsCancelled() bool\n\t// WaitIfPaused blocks until resumed or cancelled, returns error if cancelled\n\tWaitIfPaused() error\n\t// CheckCancelled returns ErrExecutionCancelled if cancelled\n\tCheckCancelled() error\n}\n\n// Manager - robot lifecycle and clock trigger management\ntype Manager interface {\n\tStart() error\n\tStop() error\n\tTick(ctx *Context, now time.Time) error\n}\n\n// Executor - executes robot phases\ntype Executor interface {\n\t// ExecuteWithControl runs execution with pre-generated ID and execution control (used by pool)\n\t// control: optional, allows pause/resume functionality\n\tExecuteWithControl(ctx *Context, robot *Robot, trigger TriggerType, data interface{}, execID string, control ExecutionControl) (*Execution, error)\n\n\t// ExecuteWithID runs execution with a pre-generated ID but no control (for backward compatibility)\n\tExecuteWithID(ctx *Context, robot *Robot, trigger TriggerType, data interface{}, execID string) (*Execution, error)\n\n\t// Execute runs execution with auto-generated ID (for direct calls)\n\tExecute(ctx *Context, robot *Robot, trigger TriggerType, data interface{}) (*Execution, error)\n\n\t// Resume resumes a suspended execution with human-provided input.\n\t// Returns ErrExecutionSuspended if the execution suspends again during resume.\n\tResume(ctx *Context, execID string, reply string) error\n\n\t// Metrics and control (for monitoring and testing)\n\tExecCount() int // total execution count\n\tCurrentCount() int // currently running count\n\tReset() // reset counters\n}\n\n// Pool - worker pool for concurrent execution\ntype Pool interface {\n\tStart() error\n\tStop() error\n\tSubmit(ctx *Context, robot *Robot, trigger TriggerType, data interface{}) (string, error)\n\tRunning() int\n\tQueued() int\n}\n\n// Cache - in-memory robot cache\ntype Cache interface {\n\tLoad(ctx *Context) error\n\tGet(memberID string) *Robot\n\tList(teamID string) []*Robot\n\tRefresh(ctx *Context, memberID string) error\n\tAdd(robot *Robot)\n\tRemove(memberID string)\n}\n\n// Dedup - deduplication check\ntype Dedup interface {\n\tCheck(ctx *Context, memberID string, trigger TriggerType) (DedupResult, error)\n\tMark(memberID string, trigger TriggerType, window time.Duration)\n}\n\n// Store - data storage operations (KB, DB)\ntype Store interface {\n\tSaveLearning(ctx *Context, memberID string, entries []LearningEntry) error\n\tGetHistory(ctx *Context, memberID string, limit int) ([]LearningEntry, error)\n\tSearchKB(ctx *Context, collections []string, query string) ([]interface{}, error)\n\tQueryDB(ctx *Context, models []string, query interface{}) ([]interface{}, error)\n}\n" }, { "path": "agent/robot/types/request.go", "content": "package types\n\nimport (\n\t\"time\"\n\n\tagentcontext \"github.com/yaoapp/yao/agent/context\"\n)\n\n// InterveneRequest - human intervention request\ntype InterveneRequest struct {\n\tTeamID string `json:\"team_id\"`\n\tMemberID string `json:\"member_id\"`\n\tAction InterventionAction `json:\"action\"`\n\tMessages []agentcontext.Message `json:\"messages\"` // user input (text, images, files)\n\tPlanTime *time.Time `json:\"plan_time,omitempty\"` // for action=plan\n\tExecutorMode ExecutorMode `json:\"executor_mode,omitempty\"` // optional: override robot config\n}\n\n// EventRequest - event trigger request\ntype EventRequest struct {\n\tMemberID string `json:\"member_id\"`\n\tSource string `json:\"source\"` // webhook path or table name\n\tEventType string `json:\"event_type\"` // lead.created, etc.\n\tData map[string]interface{} `json:\"data\"`\n\tExecutorMode ExecutorMode `json:\"executor_mode,omitempty\"` // optional: override robot config\n}\n\n// ExecutionResult - trigger result\ntype ExecutionResult struct {\n\tExecutionID string `json:\"execution_id\"`\n\tStatus ExecStatus `json:\"status\"`\n\tMessage string `json:\"message,omitempty\"`\n}\n\n// RobotState - robot status query result\ntype RobotState struct {\n\tMemberID string `json:\"member_id\"`\n\tTeamID string `json:\"team_id\"`\n\tDisplayName string `json:\"display_name\"`\n\tStatus RobotStatus `json:\"status\"`\n\tRunning int `json:\"running\"` // current running execution count\n\tMaxRunning int `json:\"max_running\"` // max concurrent allowed\n\tLastRun *time.Time `json:\"last_run,omitempty\"`\n\tNextRun *time.Time `json:\"next_run,omitempty\"`\n\tRunningIDs []string `json:\"running_ids,omitempty\"` // list of running execution IDs\n}\n" }, { "path": "agent/robot/types/robot.go", "content": "package types\n\nimport (\n\t\"context\"\n\t\"fmt\"\n\t\"sync\"\n\t\"time\"\n\n\tagentcontext \"github.com/yaoapp/yao/agent/context\"\n)\n\n// Robot - runtime representation of an autonomous robot (from __yao.member)\n// Relationship: 1 Robot : N Executions (concurrent)\n// Each trigger creates a new Execution (stored in __yao.agent_execution)\ntype Robot struct {\n\t// From __yao.member\n\tMemberID string `json:\"member_id\"`\n\tTeamID string `json:\"team_id\"`\n\tDisplayName string `json:\"display_name\"`\n\tBio string `json:\"bio\"` // Robot's description (from __yao.member.bio)\n\tSystemPrompt string `json:\"system_prompt\"`\n\tStatus RobotStatus `json:\"robot_status\"`\n\tAutonomousMode bool `json:\"autonomous_mode\"`\n\tRobotEmail string `json:\"robot_email\"` // Robot's email address for sending emails\n\tLanguageModel string `json:\"language_model\"` // LLM connector override (from __yao.member.language_model)\n\n\t// Manager info (from __yao.member)\n\tManagerID string `json:\"manager_id\"` // Direct manager user_id (who manages this robot)\n\tManagerEmail string `json:\"manager_email\"` // Manager's email address (for default delivery)\n\n\t// Parsed config (from robot_config JSON field)\n\tConfig *Config `json:\"-\"`\n\n\t// Runtime state\n\tLastRun time.Time `json:\"-\"` // last execution start time\n\tNextRun time.Time `json:\"-\"` // next scheduled execution (for clock trigger)\n\n\t// Concurrency control\n\t// Each Robot can run multiple Executions concurrently (up to Quota.Max)\n\texecutions map[string]*Execution // execID -> Execution\n\texecMu sync.RWMutex\n}\n\n// CanRun checks if robot can accept new execution\n// Note: This is a read-only check. For atomic check-and-acquire, use TryAcquireSlot()\nfunc (r *Robot) CanRun() bool {\n\tr.execMu.RLock()\n\tdefer r.execMu.RUnlock()\n\tif r.Config == nil {\n\t\treturn len(r.executions) < 2 // default max\n\t}\n\treturn len(r.executions) < r.Config.Quota.GetMax()\n}\n\n// TryAcquireSlot atomically checks if robot can run and reserves a slot\n// Returns true if slot was acquired, false if quota is full\n// This prevents race conditions between CanRun() check and AddExecution()\nfunc (r *Robot) TryAcquireSlot(exec *Execution) bool {\n\tr.execMu.Lock()\n\tdefer r.execMu.Unlock()\n\n\t// Get max quota\n\tmaxQuota := 2 // default\n\tif r.Config != nil {\n\t\tmaxQuota = r.Config.Quota.GetMax()\n\t}\n\n\t// Check if we can add\n\tif len(r.executions) >= maxQuota {\n\t\treturn false // quota full\n\t}\n\n\t// Reserve slot by adding execution\n\tif r.executions == nil {\n\t\tr.executions = make(map[string]*Execution)\n\t}\n\tr.executions[exec.ID] = exec\n\treturn true\n}\n\n// RunningCount returns current running execution count\nfunc (r *Robot) RunningCount() int {\n\tr.execMu.RLock()\n\tdefer r.execMu.RUnlock()\n\treturn len(r.executions)\n}\n\n// AddExecution adds an execution to tracking\n// Note: Prefer TryAcquireSlot() for atomic check-and-add\nfunc (r *Robot) AddExecution(exec *Execution) {\n\tr.execMu.Lock()\n\tdefer r.execMu.Unlock()\n\tif r.executions == nil {\n\t\tr.executions = make(map[string]*Execution)\n\t}\n\tr.executions[exec.ID] = exec\n}\n\n// RemoveExecution removes an execution from tracking\nfunc (r *Robot) RemoveExecution(execID string) {\n\tr.execMu.Lock()\n\tdefer r.execMu.Unlock()\n\tdelete(r.executions, execID)\n}\n\n// GetExecution returns an execution by ID\nfunc (r *Robot) GetExecution(execID string) *Execution {\n\tr.execMu.RLock()\n\tdefer r.execMu.RUnlock()\n\treturn r.executions[execID]\n}\n\n// GetExecutions returns all tracked executions\nfunc (r *Robot) GetExecutions() []*Execution {\n\tr.execMu.RLock()\n\tdefer r.execMu.RUnlock()\n\texecs := make([]*Execution, 0, len(r.executions))\n\tfor _, exec := range r.executions {\n\t\texecs = append(execs, exec)\n\t}\n\treturn execs\n}\n\n// ActiveCount returns the number of actively running executions\nfunc (r *Robot) ActiveCount() int {\n\tr.execMu.RLock()\n\tdefer r.execMu.RUnlock()\n\tcount := 0\n\tfor _, exec := range r.executions {\n\t\tif exec.Status == ExecRunning {\n\t\t\tcount++\n\t\t}\n\t}\n\treturn count\n}\n\n// WaitingCount returns the number of executions waiting for human input\nfunc (r *Robot) WaitingCount() int {\n\tr.execMu.RLock()\n\tdefer r.execMu.RUnlock()\n\tcount := 0\n\tfor _, exec := range r.executions {\n\t\tif exec.Status == ExecWaiting {\n\t\t\tcount++\n\t\t}\n\t}\n\treturn count\n}\n\n// ListExecutionBriefs returns brief summaries of all tracked executions\nfunc (r *Robot) ListExecutionBriefs() []ExecBrief {\n\tr.execMu.RLock()\n\tdefer r.execMu.RUnlock()\n\tbriefs := make([]ExecBrief, 0, len(r.executions))\n\tfor _, exec := range r.executions {\n\t\tbrief := ExecBrief{\n\t\t\tID: exec.ID,\n\t\t\tStatus: exec.Status,\n\t\t\tPhase: exec.Phase,\n\t\t\tName: exec.Name,\n\t\t\tStartTime: exec.StartTime,\n\t\t\tTaskCount: len(exec.Tasks),\n\t\t}\n\t\tfor _, result := range exec.Results {\n\t\t\tif result.Success {\n\t\t\t\tbrief.DoneCount++\n\t\t\t} else {\n\t\t\t\tbrief.FailedCount++\n\t\t\t}\n\t\t}\n\t\tbriefs = append(briefs, brief)\n\t}\n\treturn briefs\n}\n\n// MaxQuota returns the maximum concurrent execution quota\nfunc (r *Robot) MaxQuota() int {\n\tif r.Config == nil {\n\t\treturn 2\n\t}\n\treturn r.Config.Quota.GetMax()\n}\n\n// Execution - single execution instance\n// Each trigger creates a new Execution, stored in ExecutionStore\ntype Execution struct {\n\tID string `json:\"id\"` // unique execution ID\n\tMemberID string `json:\"member_id\"` // robot member ID\n\tTeamID string `json:\"team_id\"`\n\tTriggerType TriggerType `json:\"trigger_type\"` // clock | human | event\n\tStartTime time.Time `json:\"start_time\"`\n\tEndTime *time.Time `json:\"end_time,omitempty\"`\n\tStatus ExecStatus `json:\"status\"`\n\tPhase Phase `json:\"phase\"`\n\tError string `json:\"error,omitempty\"`\n\n\t// UI display fields (updated by executor at each phase)\n\tName string `json:\"name,omitempty\"` // Execution title (updated when goals complete)\n\tCurrentTaskName string `json:\"current_task_name,omitempty\"` // Current task description (updated during run phase)\n\n\t// Trigger input (stored for traceability)\n\tInput *TriggerInput `json:\"input,omitempty\"` // original trigger input\n\n\t// Phase outputs\n\tInspiration *InspirationReport `json:\"inspiration,omitempty\"` // P0: markdown\n\tGoals *Goals `json:\"goals,omitempty\"` // P1: markdown\n\tTasks []Task `json:\"tasks,omitempty\"` // P2: structured tasks\n\tCurrent *CurrentState `json:\"current,omitempty\"` // current executing state\n\tResults []TaskResult `json:\"results,omitempty\"` // P3: task results\n\tDelivery *DeliveryResult `json:\"delivery,omitempty\"`\n\tLearning []LearningEntry `json:\"learning,omitempty\"`\n\n\t// V2: Conversation and suspend-resume fields\n\tChatID string `json:\"chat_id,omitempty\"` // Unique conversation ID for Host Agent\n\tWaitingTaskID string `json:\"waiting_task_id,omitempty\"` // Task ID that is waiting for input\n\tWaitingQuestion string `json:\"waiting_question,omitempty\"` // Question posed to human\n\tWaitingSince *time.Time `json:\"waiting_since,omitempty\"` // When execution was suspended\n\tResumeContext *ResumeContext `json:\"resume_context,omitempty\"` // State for resuming suspended execution\n\n\t// Runtime (internal, not serialized)\n\tctx context.Context `json:\"-\"`\n\tcancel context.CancelFunc `json:\"-\"`\n\trobot *Robot `json:\"-\"`\n}\n\n// ResumeContext holds the state needed to resume a suspended execution\ntype ResumeContext struct {\n\tTaskIndex int `json:\"task_index\"` // Index of the task to resume from\n\tPreviousResults []TaskResult `json:\"previous_results\"` // Results from tasks completed before suspend\n}\n\n// ExecBrief is a lightweight summary of an execution for status snapshots\ntype ExecBrief struct {\n\tID string `json:\"id\"`\n\tStatus ExecStatus `json:\"status\"`\n\tPhase Phase `json:\"phase\"`\n\tName string `json:\"name,omitempty\"`\n\tStartTime time.Time `json:\"start_time\"`\n\tTaskCount int `json:\"task_count\"`\n\tDoneCount int `json:\"done_count\"`\n\tFailedCount int `json:\"failed_count\"`\n}\n\n// RobotStatusSnapshot provides real-time robot status for the Host Agent\ntype RobotStatusSnapshot struct {\n\tMemberID string `json:\"member_id,omitempty\"` // Robot member ID\n\tStatus RobotStatus `json:\"status,omitempty\"` // Current robot status (idle/working)\n\tActiveCount int `json:\"active_count\"` // Currently running executions\n\tWaitingCount int `json:\"waiting_count\"` // Executions waiting for input\n\tQueuedCount int `json:\"queued_count\"` // Executions in queue (not yet started)\n\tMaxQuota int `json:\"max_quota\"` // Maximum concurrent executions\n\tActiveExecs []ExecBrief `json:\"active_execs,omitempty\"` // Currently running execution summaries\n\tRecentExecs []ExecBrief `json:\"recent_execs,omitempty\"` // Recently completed execution summaries\n}\n\n// GetRobot returns the robot associated with this execution\nfunc (e *Execution) GetRobot() *Robot {\n\treturn e.robot\n}\n\n// SetRobot sets the robot associated with this execution\nfunc (e *Execution) SetRobot(robot *Robot) {\n\te.robot = robot\n}\n\n// TriggerInput - stored trigger input for traceability\ntype TriggerInput struct {\n\t// For human intervention\n\tAction InterventionAction `json:\"action,omitempty\"` // task.add, goal.adjust, etc.\n\tMessages []agentcontext.Message `json:\"messages,omitempty\"` // user's input (text, images, files)\n\tUserID string `json:\"user_id,omitempty\"` // who triggered\n\tLocale string `json:\"locale,omitempty\"` // language for UI display (e.g., \"en-US\", \"zh-CN\")\n\n\t// For event trigger\n\tSource EventSource `json:\"source,omitempty\"` // webhook | database\n\tEventType string `json:\"event_type,omitempty\"` // lead.created, etc.\n\tData map[string]interface{} `json:\"data,omitempty\"` // event payload\n\n\t// For clock trigger\n\tClock *ClockContext `json:\"clock,omitempty\"` // time context when triggered\n}\n\n// CurrentState - current executing goal and task\ntype CurrentState struct {\n\tTask *Task `json:\"task,omitempty\"` // current task being executed\n\tTaskIndex int `json:\"task_index\"` // index in Tasks slice\n\tProgress string `json:\"progress,omitempty\"` // human-readable progress (e.g., \"2/5 tasks\")\n}\n\n// Goals - P1 output (markdown for LLM + structured metadata)\n// P1 Agent reads InspirationReport and generates goals as markdown\n// Example:\n// ## Goals\n// 1. [High] Analyze sales data and identify trends\n// - Reason: Sales up 50%, need to understand why\n//\n// 2. [Normal] Prepare weekly report for manager\n// - Reason: Friday 5pm, weekly report due\n//\n// 3. [Low] Update CRM with new leads\n// - Reason: 3 pending leads from yesterday\ntype Goals struct {\n\tContent string `json:\"content\"` // markdown text\n\n\t// Delivery for P4 (where to send results)\n\tDelivery *DeliveryTarget `json:\"delivery,omitempty\"`\n}\n\n// DeliveryTarget - where to deliver results (defined in P1, used in P4)\ntype DeliveryTarget struct {\n\tType DeliveryType `json:\"type\"` // email | webhook | report | notification\n\tRecipients []string `json:\"recipients,omitempty\"` // email addresses, webhook URLs, user IDs\n\tFormat string `json:\"format,omitempty\"` // markdown | html | json | text\n\tTemplate string `json:\"template,omitempty\"` // template name or inline template\n\tOptions map[string]interface{} `json:\"options,omitempty\"` // channel-specific options\n}\n\n// Task - planned task (structured, for execution)\ntype Task struct {\n\tID string `json:\"id\"`\n\tDescription string `json:\"description,omitempty\"` // human-readable task description (for UI display)\n\tMessages []agentcontext.Message `json:\"messages\"` // original input (text, images, files)\n\tGoalRef string `json:\"goal_ref,omitempty\"` // reference to goal (e.g., \"Goal 1\")\n\tSource TaskSource `json:\"source\"` // auto | human | event\n\n\t// Executor\n\tExecutorType ExecutorType `json:\"executor_type\"`\n\tExecutorID string `json:\"executor_id\"` // unified ID: agent/assistant/process ID, or \"mcp_server.mcp_tool\" for MCP\n\tArgs []any `json:\"args,omitempty\"`\n\n\t// MCP-specific fields (required when executor_type is \"mcp\")\n\tMCPServer string `json:\"mcp_server,omitempty\"` // MCP server/client ID (e.g., \"ark.image.text2img\")\n\tMCPTool string `json:\"mcp_tool,omitempty\"` // MCP tool name (e.g., \"generate\")\n\n\t// Validation (defined in P2, used in P3)\n\t// ExpectedOutput describes what the task should produce (for LLM semantic validation)\n\tExpectedOutput string `json:\"expected_output,omitempty\"` // e.g., \"JSON with sales_total, growth_rate fields\"\n\t// ValidationRules are specific checks to perform (can be semantic or structural)\n\tValidationRules []string `json:\"validation_rules,omitempty\"` // e.g., [\"output must be valid JSON\", \"sales_total > 0\"]\n\n\t// Runtime\n\tStatus TaskStatus `json:\"status\"`\n\tOrder int `json:\"order\"` // execution order (0-based)\n\tStartTime *time.Time `json:\"start_time,omitempty\"`\n\tEndTime *time.Time `json:\"end_time,omitempty\"`\n}\n\n// TaskResult - task execution result\ntype TaskResult struct {\n\tTaskID string `json:\"task_id\"`\n\tSuccess bool `json:\"success\"`\n\tOutput interface{} `json:\"output,omitempty\"`\n\tError string `json:\"error,omitempty\"`\n\tDuration int64 `json:\"duration_ms\"`\n\n\t// Validation result (populated by Delivery Agent in P4, not by runner in V2)\n\tValidation *ValidationResult `json:\"validation,omitempty\"`\n\n\t// V2: Need-input signal from assistant (detected via Next Hook protocol)\n\tNeedInput bool `json:\"need_input,omitempty\"` // Assistant requests human input\n\tInputQuestion string `json:\"input_question,omitempty\"` // Question for the human\n}\n\n// ValidationResult - P3 semantic validation result\ntype ValidationResult struct {\n\t// Basic validation result\n\tPassed bool `json:\"passed\"` // overall validation passed\n\tScore float64 `json:\"score,omitempty\"` // 0-1 confidence score\n\tIssues []string `json:\"issues,omitempty\"` // what failed\n\tSuggestions []string `json:\"suggestions,omitempty\"` // how to improve\n\tDetails string `json:\"details,omitempty\"` // detailed validation report (markdown)\n\n\t// Execution state (for multi-turn conversation control)\n\tComplete bool `json:\"complete\"` // whether expected result is obtained\n\tNeedReply bool `json:\"need_reply,omitempty\"` // whether to continue conversation\n\tReplyContent string `json:\"reply_content,omitempty\"` // content for next turn (if NeedReply)\n}\n\n// DeliveryResult - P4 delivery output (new architecture)\ntype DeliveryResult struct {\n\tRequestID string `json:\"request_id\"` // Delivery request ID\n\tContent *DeliveryContent `json:\"content\"` // Agent-generated content\n\tResults []ChannelResult `json:\"results,omitempty\"` // Results per channel\n\tSuccess bool `json:\"success\"` // Overall success\n\tError string `json:\"error,omitempty\"` // Error if failed\n\tSentAt *time.Time `json:\"sent_at,omitempty\"` // When delivery completed\n}\n\n// DeliveryContent - Content generated by Delivery Agent (only content, no channels)\ntype DeliveryContent struct {\n\tSummary string `json:\"summary\"` // Brief 1-2 sentence summary\n\tBody string `json:\"body\"` // Full markdown report\n\tAttachments []DeliveryAttachment `json:\"attachments,omitempty\"` // Output artifacts from P3\n}\n\n// DeliveryAttachment - Task output attachment with metadata\ntype DeliveryAttachment struct {\n\tTitle string `json:\"title\"` // Human-readable title\n\tDescription string `json:\"description,omitempty\"` // What this artifact is\n\tTaskID string `json:\"task_id,omitempty\"` // Which task produced this\n\tFile string `json:\"file\"` // Wrapper: __://\n}\n\n// DeliveryRequest - pushed to Delivery Center (no channels - center decides based on preferences)\ntype DeliveryRequest struct {\n\tContent *DeliveryContent `json:\"content\"` // Agent-generated content\n\tContext *DeliveryContext `json:\"context\"` // Tracking info\n}\n\n// DeliveryContext - tracking and audit info\ntype DeliveryContext struct {\n\tMemberID string `json:\"member_id\"` // Robot member ID (globally unique)\n\tExecutionID string `json:\"execution_id\"` // Execution ID\n\tTriggerType TriggerType `json:\"trigger_type\"` // clock | human | event\n\tTeamID string `json:\"team_id\"` // Team ID\n}\n\n// DeliveryPreferences - Robot/User delivery preferences (from Config)\ntype DeliveryPreferences struct {\n\tEmail *EmailPreference `json:\"email,omitempty\"` // Email delivery settings\n\tWebhook *WebhookPreference `json:\"webhook,omitempty\"` // Webhook delivery settings\n\tProcess *ProcessPreference `json:\"process,omitempty\"` // Process delivery settings\n}\n\n// EmailPreference - Email delivery configuration\ntype EmailPreference struct {\n\tEnabled bool `json:\"enabled\"` // Whether email delivery is enabled\n\tTargets []EmailTarget `json:\"targets,omitempty\"` // Multiple email targets\n}\n\n// EmailTarget - Single email target\ntype EmailTarget struct {\n\tTo []string `json:\"to\"` // Recipient addresses\n\tTemplate string `json:\"template,omitempty\"` // Email template ID\n\tSubject string `json:\"subject,omitempty\"` // Subject template\n}\n\n// WebhookPreference - Webhook delivery configuration\ntype WebhookPreference struct {\n\tEnabled bool `json:\"enabled\"` // Whether webhook delivery is enabled\n\tTargets []WebhookTarget `json:\"targets,omitempty\"` // Multiple webhook targets\n}\n\n// WebhookTarget - Single webhook target\ntype WebhookTarget struct {\n\tURL string `json:\"url\"` // Webhook URL\n\tMethod string `json:\"method,omitempty\"` // HTTP method (default: POST)\n\tHeaders map[string]string `json:\"headers,omitempty\"` // Custom headers\n\tSecret string `json:\"secret,omitempty\"` // Signing secret\n}\n\n// ProcessPreference - Process delivery configuration\ntype ProcessPreference struct {\n\tEnabled bool `json:\"enabled\"` // Whether process delivery is enabled\n\tTargets []ProcessTarget `json:\"targets,omitempty\"` // Multiple process targets\n}\n\n// ProcessTarget - Single process target\ntype ProcessTarget struct {\n\tProcess string `json:\"process\"` // Yao Process name\n\tArgs []any `json:\"args,omitempty\"` // Process arguments\n}\n\n// ChannelResult - Result of delivery to a single channel target\ntype ChannelResult struct {\n\tType DeliveryType `json:\"type\"` // email | webhook | process\n\tTarget string `json:\"target\"` // Target identifier (email, URL, process name)\n\tSuccess bool `json:\"success\"` // Whether delivery succeeded\n\tRecipients []string `json:\"recipients,omitempty\"` // Who received (for email)\n\tDetails interface{} `json:\"details,omitempty\"` // Channel-specific response\n\tError string `json:\"error,omitempty\"` // Error message if failed\n\tSentAt *time.Time `json:\"sent_at,omitempty\"` // When this target was delivered\n}\n\n// LearningEntry - knowledge to save\ntype LearningEntry struct {\n\tType LearningType `json:\"type\"` // execution | feedback | insight\n\tContent string `json:\"content\"`\n\tTags []string `json:\"tags,omitempty\"`\n\tMeta interface{} `json:\"meta,omitempty\"`\n}\n\n// NewRobotFromMap creates a Robot from a map (typically from DB record)\nfunc NewRobotFromMap(m map[string]interface{}) (*Robot, error) {\n\tmemberID := getString(m, \"member_id\")\n\tteamID := getString(m, \"team_id\")\n\n\t// Validate required fields\n\tif memberID == \"\" || teamID == \"\" {\n\t\treturn nil, fmt.Errorf(\"missing required fields: member_id or team_id\")\n\t}\n\n\trobot := &Robot{\n\t\tMemberID: memberID,\n\t\tTeamID: teamID,\n\t\tDisplayName: getString(m, \"display_name\"),\n\t\tBio: getString(m, \"bio\"),\n\t\tSystemPrompt: getString(m, \"system_prompt\"),\n\t\tAutonomousMode: getBool(m, \"autonomous_mode\"),\n\t\tRobotEmail: getString(m, \"robot_email\"),\n\t\tManagerID: getString(m, \"manager_id\"),\n\t\tManagerEmail: getString(m, \"manager_email\"),\n\t\tLanguageModel: getString(m, \"language_model\"),\n\t}\n\n\t// Parse robot_status\n\tif status := getString(m, \"robot_status\"); status != \"\" {\n\t\trobot.Status = RobotStatus(status)\n\t} else {\n\t\trobot.Status = RobotIdle\n\t}\n\n\t// Parse robot_config JSON\n\tif configData, ok := m[\"robot_config\"]; ok && configData != nil {\n\t\tconfig, err := ParseConfig(configData)\n\t\tif err != nil {\n\t\t\treturn nil, fmt.Errorf(\"failed to parse robot_config: %w\", err)\n\t\t}\n\t\trobot.Config = config\n\t}\n\n\t// Ensure Config exists for merging agents/mcp_servers\n\tif robot.Config == nil {\n\t\trobot.Config = &Config{}\n\t}\n\tif robot.Config.Resources == nil {\n\t\trobot.Config.Resources = &Resources{}\n\t}\n\n\t// Merge agents from member table into Config.Resources.Agents\n\tif agentsData, ok := m[\"agents\"]; ok && agentsData != nil {\n\t\tagents := getStringSlice(agentsData)\n\t\tif len(agents) > 0 {\n\t\t\trobot.Config.Resources.Agents = agents\n\t\t}\n\t}\n\n\t// Merge mcp_servers from member table into Config.Resources.MCP\n\tif mcpData, ok := m[\"mcp_servers\"]; ok && mcpData != nil {\n\t\tmcpServers := getStringSlice(mcpData)\n\t\tif len(mcpServers) > 0 {\n\t\t\tfor _, serverID := range mcpServers {\n\t\t\t\trobot.Config.Resources.MCP = append(robot.Config.Resources.MCP, MCPConfig{\n\t\t\t\t\tID: serverID,\n\t\t\t\t})\n\t\t\t}\n\t\t}\n\t}\n\n\treturn robot, nil\n}\n\n// getStringSlice converts interface{} to []string\nfunc getStringSlice(v interface{}) []string {\n\tif v == nil {\n\t\treturn nil\n\t}\n\tswitch val := v.(type) {\n\tcase []string:\n\t\treturn val\n\tcase []interface{}:\n\t\tresult := make([]string, 0, len(val))\n\t\tfor _, item := range val {\n\t\t\tif s, ok := item.(string); ok {\n\t\t\t\tresult = append(result, s)\n\t\t\t}\n\t\t}\n\t\treturn result\n\t}\n\treturn nil\n}\n\n// getString safely gets a string value from map\nfunc getString(m map[string]interface{}, key string) string {\n\tif m == nil {\n\t\treturn \"\"\n\t}\n\tif v, ok := m[key]; ok && v != nil {\n\t\tif s, ok := v.(string); ok {\n\t\t\treturn s\n\t\t}\n\t\treturn fmt.Sprintf(\"%v\", v)\n\t}\n\treturn \"\"\n}\n\n// getBool safely gets a bool value from map\nfunc getBool(m map[string]interface{}, key string) bool {\n\tif m == nil {\n\t\treturn false\n\t}\n\tif v, ok := m[key]; ok && v != nil {\n\t\tswitch b := v.(type) {\n\t\tcase bool:\n\t\t\treturn b\n\t\tcase int:\n\t\t\treturn b != 0\n\t\tcase int64:\n\t\t\treturn b != 0\n\t\tcase float64:\n\t\t\treturn b != 0\n\t\tcase string:\n\t\t\treturn b == \"true\" || b == \"1\"\n\t\t}\n\t}\n\treturn false\n}\n" }, { "path": "agent/robot/types/robot_test.go", "content": "package types_test\n\nimport (\n\t\"testing\"\n\t\"time\"\n\n\t\"github.com/stretchr/testify/assert\"\n\t\"github.com/yaoapp/yao/agent/robot/types\"\n)\n\nfunc TestRobotCanRun(t *testing.T) {\n\tt.Run(\"can run when under quota\", func(t *testing.T) {\n\t\trobot := &types.Robot{\n\t\t\tConfig: &types.Config{\n\t\t\t\tQuota: &types.Quota{Max: 2},\n\t\t\t},\n\t\t}\n\t\tassert.True(t, robot.CanRun())\n\t})\n\n\tt.Run(\"can run with nil config (uses default quota)\", func(t *testing.T) {\n\t\trobot := &types.Robot{\n\t\t\tConfig: nil, // nil config should not panic\n\t\t}\n\t\t// Should not panic and use default max (2)\n\t\tassert.True(t, robot.CanRun())\n\t})\n\n\tt.Run(\"can run with nil quota (uses default)\", func(t *testing.T) {\n\t\trobot := &types.Robot{\n\t\t\tConfig: &types.Config{\n\t\t\t\tQuota: nil, // nil quota should use default\n\t\t\t},\n\t\t}\n\t\tassert.True(t, robot.CanRun())\n\t})\n\n\tt.Run(\"cannot run when at quota\", func(t *testing.T) {\n\t\trobot := &types.Robot{\n\t\t\tConfig: &types.Config{\n\t\t\t\tQuota: &types.Quota{Max: 2},\n\t\t\t},\n\t\t}\n\n\t\t// Add 2 executions to reach quota\n\t\texec1 := &types.Execution{ID: \"exec1\"}\n\t\texec2 := &types.Execution{ID: \"exec2\"}\n\t\trobot.AddExecution(exec1)\n\t\trobot.AddExecution(exec2)\n\n\t\tassert.False(t, robot.CanRun())\n\t})\n\n\tt.Run(\"can run after removing execution\", func(t *testing.T) {\n\t\trobot := &types.Robot{\n\t\t\tConfig: &types.Config{\n\t\t\t\tQuota: &types.Quota{Max: 2},\n\t\t\t},\n\t\t}\n\n\t\texec1 := &types.Execution{ID: \"exec1\"}\n\t\texec2 := &types.Execution{ID: \"exec2\"}\n\t\trobot.AddExecution(exec1)\n\t\trobot.AddExecution(exec2)\n\n\t\tassert.False(t, robot.CanRun())\n\n\t\trobot.RemoveExecution(\"exec1\")\n\t\tassert.True(t, robot.CanRun())\n\t})\n}\n\nfunc TestRobotRunningCount(t *testing.T) {\n\trobot := &types.Robot{\n\t\tConfig: &types.Config{\n\t\t\tQuota: &types.Quota{Max: 5},\n\t\t},\n\t}\n\n\tassert.Equal(t, 0, robot.RunningCount())\n\n\texec1 := &types.Execution{ID: \"exec1\"}\n\trobot.AddExecution(exec1)\n\tassert.Equal(t, 1, robot.RunningCount())\n\n\texec2 := &types.Execution{ID: \"exec2\"}\n\trobot.AddExecution(exec2)\n\tassert.Equal(t, 2, robot.RunningCount())\n\n\trobot.RemoveExecution(\"exec1\")\n\tassert.Equal(t, 1, robot.RunningCount())\n\n\trobot.RemoveExecution(\"exec2\")\n\tassert.Equal(t, 0, robot.RunningCount())\n}\n\nfunc TestRobotAddExecution(t *testing.T) {\n\trobot := &types.Robot{\n\t\tConfig: &types.Config{\n\t\t\tQuota: &types.Quota{Max: 2},\n\t\t},\n\t}\n\n\texec := &types.Execution{\n\t\tID: \"exec1\",\n\t\tMemberID: \"member1\",\n\t}\n\n\trobot.AddExecution(exec)\n\tassert.Equal(t, 1, robot.RunningCount())\n\n\tretrieved := robot.GetExecution(\"exec1\")\n\tassert.NotNil(t, retrieved)\n\tassert.Equal(t, \"exec1\", retrieved.ID)\n\tassert.Equal(t, \"member1\", retrieved.MemberID)\n}\n\nfunc TestRobotRemoveExecution(t *testing.T) {\n\trobot := &types.Robot{\n\t\tConfig: &types.Config{\n\t\t\tQuota: &types.Quota{Max: 2},\n\t\t},\n\t}\n\n\texec := &types.Execution{ID: \"exec1\"}\n\trobot.AddExecution(exec)\n\tassert.Equal(t, 1, robot.RunningCount())\n\n\trobot.RemoveExecution(\"exec1\")\n\tassert.Equal(t, 0, robot.RunningCount())\n\n\tretrieved := robot.GetExecution(\"exec1\")\n\tassert.Nil(t, retrieved)\n}\n\nfunc TestRobotGetExecution(t *testing.T) {\n\trobot := &types.Robot{\n\t\tConfig: &types.Config{\n\t\t\tQuota: &types.Quota{Max: 2},\n\t\t},\n\t}\n\n\tt.Run(\"get existing execution\", func(t *testing.T) {\n\t\texec := &types.Execution{\n\t\t\tID: \"exec1\",\n\t\t\tMemberID: \"member1\",\n\t\t}\n\t\trobot.AddExecution(exec)\n\n\t\tretrieved := robot.GetExecution(\"exec1\")\n\t\tassert.NotNil(t, retrieved)\n\t\tassert.Equal(t, \"exec1\", retrieved.ID)\n\t})\n\n\tt.Run(\"get non-existing execution\", func(t *testing.T) {\n\t\tretrieved := robot.GetExecution(\"non-existing\")\n\t\tassert.Nil(t, retrieved)\n\t})\n}\n\nfunc TestRobotGetExecutions(t *testing.T) {\n\trobot := &types.Robot{\n\t\tConfig: &types.Config{\n\t\t\tQuota: &types.Quota{Max: 5},\n\t\t},\n\t}\n\n\tt.Run(\"empty executions\", func(t *testing.T) {\n\t\texecs := robot.GetExecutions()\n\t\tassert.Empty(t, execs)\n\t})\n\n\tt.Run(\"multiple executions\", func(t *testing.T) {\n\t\texec1 := &types.Execution{ID: \"exec1\"}\n\t\texec2 := &types.Execution{ID: \"exec2\"}\n\t\texec3 := &types.Execution{ID: \"exec3\"}\n\n\t\trobot.AddExecution(exec1)\n\t\trobot.AddExecution(exec2)\n\t\trobot.AddExecution(exec3)\n\n\t\texecs := robot.GetExecutions()\n\t\tassert.Len(t, execs, 3)\n\n\t\t// Check all executions are present\n\t\tids := make(map[string]bool)\n\t\tfor _, exec := range execs {\n\t\t\tids[exec.ID] = true\n\t\t}\n\t\tassert.True(t, ids[\"exec1\"])\n\t\tassert.True(t, ids[\"exec2\"])\n\t\tassert.True(t, ids[\"exec3\"])\n\t})\n}\n\nfunc TestRobotConcurrentAccess(t *testing.T) {\n\t// Test thread-safe execution management\n\trobot := &types.Robot{\n\t\tConfig: &types.Config{\n\t\t\tQuota: &types.Quota{Max: 10},\n\t\t},\n\t}\n\n\t// Add executions concurrently\n\tdone := make(chan bool)\n\tfor i := 0; i < 5; i++ {\n\t\tgo func(id int) {\n\t\t\texec := &types.Execution{ID: string(rune('0' + id))}\n\t\t\trobot.AddExecution(exec)\n\t\t\tdone <- true\n\t\t}(i)\n\t}\n\n\t// Wait for all goroutines\n\tfor i := 0; i < 5; i++ {\n\t\t<-done\n\t}\n\n\t// Verify count\n\tcount := robot.RunningCount()\n\tassert.Equal(t, 5, count)\n\n\t// Remove executions concurrently\n\tfor i := 0; i < 5; i++ {\n\t\tgo func(id int) {\n\t\t\trobot.RemoveExecution(string(rune('0' + id)))\n\t\t\tdone <- true\n\t\t}(i)\n\t}\n\n\t// Wait for all goroutines\n\tfor i := 0; i < 5; i++ {\n\t\t<-done\n\t}\n\n\t// Verify count\n\tcount = robot.RunningCount()\n\tassert.Equal(t, 0, count)\n}\n\nfunc TestRobotTryAcquireSlot(t *testing.T) {\n\tt.Run(\"acquire slot when under quota\", func(t *testing.T) {\n\t\trobot := &types.Robot{\n\t\t\tConfig: &types.Config{\n\t\t\t\tQuota: &types.Quota{Max: 2},\n\t\t\t},\n\t\t}\n\n\t\texec := &types.Execution{ID: \"exec1\"}\n\t\tacquired := robot.TryAcquireSlot(exec)\n\n\t\tassert.True(t, acquired)\n\t\tassert.Equal(t, 1, robot.RunningCount())\n\t\tassert.NotNil(t, robot.GetExecution(\"exec1\"))\n\t})\n\n\tt.Run(\"fail to acquire when at quota\", func(t *testing.T) {\n\t\trobot := &types.Robot{\n\t\t\tConfig: &types.Config{\n\t\t\t\tQuota: &types.Quota{Max: 2},\n\t\t\t},\n\t\t}\n\n\t\t// Fill quota\n\t\trobot.TryAcquireSlot(&types.Execution{ID: \"exec1\"})\n\t\trobot.TryAcquireSlot(&types.Execution{ID: \"exec2\"})\n\n\t\t// Try to acquire one more\n\t\texec3 := &types.Execution{ID: \"exec3\"}\n\t\tacquired := robot.TryAcquireSlot(exec3)\n\n\t\tassert.False(t, acquired)\n\t\tassert.Equal(t, 2, robot.RunningCount())\n\t\tassert.Nil(t, robot.GetExecution(\"exec3\"))\n\t})\n\n\tt.Run(\"acquire with nil config uses default quota\", func(t *testing.T) {\n\t\trobot := &types.Robot{\n\t\t\tConfig: nil, // default quota is 2\n\t\t}\n\n\t\texec1 := &types.Execution{ID: \"exec1\"}\n\t\texec2 := &types.Execution{ID: \"exec2\"}\n\t\texec3 := &types.Execution{ID: \"exec3\"}\n\n\t\tassert.True(t, robot.TryAcquireSlot(exec1))\n\t\tassert.True(t, robot.TryAcquireSlot(exec2))\n\t\tassert.False(t, robot.TryAcquireSlot(exec3)) // should fail at default max=2\n\t})\n}\n\nfunc TestRobotTryAcquireSlotConcurrent(t *testing.T) {\n\t// Test that TryAcquireSlot is atomic and prevents exceeding quota\n\trobot := &types.Robot{\n\t\tConfig: &types.Config{\n\t\t\tQuota: &types.Quota{Max: 5},\n\t\t},\n\t}\n\n\t// Launch 20 goroutines trying to acquire slots\n\tsuccessCount := make(chan bool, 20)\n\tfor i := 0; i < 20; i++ {\n\t\tgo func(id int) {\n\t\t\texec := &types.Execution{ID: string(rune('A' + id))}\n\t\t\tsuccess := robot.TryAcquireSlot(exec)\n\t\t\tsuccessCount <- success\n\t\t}(i)\n\t}\n\n\t// Count successes\n\tacquired := 0\n\tfor i := 0; i < 20; i++ {\n\t\tif <-successCount {\n\t\t\tacquired++\n\t\t}\n\t}\n\n\t// Should have exactly 5 successful acquisitions (quota max)\n\tassert.Equal(t, 5, acquired, \"Should acquire exactly quota max slots\")\n\tassert.Equal(t, 5, robot.RunningCount(), \"Running count should match quota max\")\n}\n\nfunc TestRobotTryAcquireSlotRaceCondition(t *testing.T) {\n\t// Stress test to verify no race condition in TryAcquireSlot\n\tfor iteration := 0; iteration < 100; iteration++ {\n\t\trobot := &types.Robot{\n\t\t\tConfig: &types.Config{\n\t\t\t\tQuota: &types.Quota{Max: 3},\n\t\t\t},\n\t\t}\n\n\t\t// Launch many goroutines simultaneously\n\t\tsuccessCount := make(chan bool, 50)\n\t\tfor i := 0; i < 50; i++ {\n\t\t\tgo func(id int) {\n\t\t\t\texec := &types.Execution{ID: string(rune('A'+id%26)) + string(rune('0'+id/26))}\n\t\t\t\tsuccess := robot.TryAcquireSlot(exec)\n\t\t\t\tsuccessCount <- success\n\t\t\t}(i)\n\t\t}\n\n\t\t// Count successes\n\t\tacquired := 0\n\t\tfor i := 0; i < 50; i++ {\n\t\t\tif <-successCount {\n\t\t\t\tacquired++\n\t\t\t}\n\t\t}\n\n\t\t// Should never exceed quota\n\t\tassert.Equal(t, 3, acquired, \"Iteration %d: Should acquire exactly quota max slots\", iteration)\n\t\tassert.Equal(t, 3, robot.RunningCount(), \"Iteration %d: Running count should match quota max\", iteration)\n\t}\n}\n\nfunc TestExecutionStructure(t *testing.T) {\n\tt.Run(\"execution with all fields\", func(t *testing.T) {\n\t\texec := &types.Execution{\n\t\t\tID: \"exec1\",\n\t\t\tMemberID: \"member1\",\n\t\t\tTeamID: \"team1\",\n\t\t\tTriggerType: types.TriggerClock,\n\t\t\tStatus: types.ExecRunning,\n\t\t\tPhase: types.PhaseGoals,\n\t\t}\n\n\t\tassert.Equal(t, \"exec1\", exec.ID)\n\t\tassert.Equal(t, \"member1\", exec.MemberID)\n\t\tassert.Equal(t, \"team1\", exec.TeamID)\n\t\tassert.Equal(t, types.TriggerClock, exec.TriggerType)\n\t\tassert.Equal(t, types.ExecRunning, exec.Status)\n\t\tassert.Equal(t, types.PhaseGoals, exec.Phase)\n\t})\n\n\tt.Run(\"execution with trigger input\", func(t *testing.T) {\n\t\texec := &types.Execution{\n\t\t\tID: \"exec1\",\n\t\t\tInput: &types.TriggerInput{\n\t\t\t\tAction: types.ActionTaskAdd,\n\t\t\t\tUserID: \"user1\",\n\t\t\t},\n\t\t}\n\n\t\tassert.NotNil(t, exec.Input)\n\t\tassert.Equal(t, types.ActionTaskAdd, exec.Input.Action)\n\t\tassert.Equal(t, \"user1\", exec.Input.UserID)\n\t})\n}\n\nfunc TestTaskStructure(t *testing.T) {\n\ttask := &types.Task{\n\t\tID: \"task1\",\n\t\tGoalRef: \"Goal 1\",\n\t\tSource: types.TaskSourceAuto,\n\t\tExecutorType: types.ExecutorAssistant,\n\t\tExecutorID: \"assistant1\",\n\t\tStatus: types.TaskPending,\n\t\tOrder: 0,\n\t\t// P3 validation fields\n\t\tExpectedOutput: \"JSON with sales_total and growth_rate fields\",\n\t\tValidationRules: []string{\n\t\t\t// Natural language rules (matched by validator)\n\t\t\t\"output must be valid JSON\",\n\t\t\t\"must contain 'sales_total'\",\n\t\t\t// Structured rule: check field type\n\t\t\t`{\"type\": \"type\", \"path\": \"growth_rate\", \"value\": \"number\"}`,\n\t\t},\n\t}\n\n\tassert.Equal(t, \"task1\", task.ID)\n\tassert.Equal(t, \"Goal 1\", task.GoalRef)\n\tassert.Equal(t, types.TaskSourceAuto, task.Source)\n\tassert.Equal(t, types.ExecutorAssistant, task.ExecutorType)\n\tassert.Equal(t, \"assistant1\", task.ExecutorID)\n\tassert.Equal(t, types.TaskPending, task.Status)\n\tassert.Equal(t, 0, task.Order)\n\t// Validation fields\n\tassert.Contains(t, task.ExpectedOutput, \"sales_total\")\n\tassert.Len(t, task.ValidationRules, 3)\n}\n\nfunc TestGoalsStructure(t *testing.T) {\n\tgoals := &types.Goals{\n\t\tContent: \"## Goals\\n1. [High] Complete project\\n2. [Normal] Review code\",\n\t\tDelivery: &types.DeliveryTarget{\n\t\t\tType: types.DeliveryEmail,\n\t\t\tRecipients: []string{\"team@example.com\"},\n\t\t\tFormat: \"markdown\",\n\t\t},\n\t}\n\n\tassert.Contains(t, goals.Content, \"Goals\")\n\tassert.Contains(t, goals.Content, \"Complete project\")\n\tassert.NotNil(t, goals.Delivery)\n\tassert.Equal(t, types.DeliveryEmail, goals.Delivery.Type)\n}\n\nfunc TestTaskResultStructure(t *testing.T) {\n\tresult := &types.TaskResult{\n\t\tTaskID: \"task1\",\n\t\tSuccess: true,\n\t\tOutput: \"Task completed successfully\",\n\t\tDuration: 1500,\n\t\tValidation: &types.ValidationResult{\n\t\t\tPassed: true,\n\t\t\tScore: 0.98,\n\t\t},\n\t}\n\n\tassert.Equal(t, \"task1\", result.TaskID)\n\tassert.True(t, result.Success)\n\tassert.Equal(t, \"Task completed successfully\", result.Output)\n\tassert.Equal(t, int64(1500), result.Duration)\n\tassert.NotNil(t, result.Validation)\n\tassert.True(t, result.Validation.Passed)\n\tassert.Equal(t, 0.98, result.Validation.Score)\n}\n\nfunc TestValidationResultStructure(t *testing.T) {\n\tvalidation := &types.ValidationResult{\n\t\tPassed: false,\n\t\tScore: 0.45,\n\t\tIssues: []string{\"Missing required field: sales_total\", \"Growth rate is negative\"},\n\t\tSuggestions: []string{\"Add sales_total calculation\", \"Verify data source\"},\n\t\tDetails: \"Detailed validation report...\",\n\t}\n\n\tassert.False(t, validation.Passed)\n\tassert.Equal(t, 0.45, validation.Score)\n\tassert.Len(t, validation.Issues, 2)\n\tassert.Contains(t, validation.Issues[0], \"sales_total\")\n\tassert.Len(t, validation.Suggestions, 2)\n}\n\nfunc TestValidationResultMultiTurnFields(t *testing.T) {\n\t// Test new multi-turn conversation control fields\n\tt.Run(\"complete and passed\", func(t *testing.T) {\n\t\tvalidation := &types.ValidationResult{\n\t\t\tPassed: true,\n\t\t\tScore: 0.95,\n\t\t\tComplete: true,\n\t\t}\n\t\tassert.True(t, validation.Passed)\n\t\tassert.True(t, validation.Complete)\n\t\tassert.False(t, validation.NeedReply)\n\t\tassert.Empty(t, validation.ReplyContent)\n\t})\n\n\tt.Run(\"passed but not complete - needs reply\", func(t *testing.T) {\n\t\tvalidation := &types.ValidationResult{\n\t\t\tPassed: true,\n\t\t\tScore: 0.7,\n\t\t\tComplete: false,\n\t\t\tNeedReply: true,\n\t\t\tReplyContent: \"Please continue and provide the complete result.\",\n\t\t}\n\t\tassert.True(t, validation.Passed)\n\t\tassert.False(t, validation.Complete)\n\t\tassert.True(t, validation.NeedReply)\n\t\tassert.NotEmpty(t, validation.ReplyContent)\n\t})\n\n\tt.Run(\"failed with suggestions - needs reply\", func(t *testing.T) {\n\t\tvalidation := &types.ValidationResult{\n\t\t\tPassed: false,\n\t\t\tScore: 0.3,\n\t\t\tComplete: false,\n\t\t\tIssues: []string{\"Missing required field\"},\n\t\t\tSuggestions: []string{\"Add the field\"},\n\t\t\tNeedReply: true,\n\t\t\tReplyContent: \"## Validation Feedback\\n\\nPlease fix: Missing required field\",\n\t\t}\n\t\tassert.False(t, validation.Passed)\n\t\tassert.False(t, validation.Complete)\n\t\tassert.True(t, validation.NeedReply)\n\t\tassert.Contains(t, validation.ReplyContent, \"Validation Feedback\")\n\t})\n\n\tt.Run(\"failed without suggestions - no reply\", func(t *testing.T) {\n\t\tvalidation := &types.ValidationResult{\n\t\t\tPassed: false,\n\t\t\tScore: 0.0,\n\t\t\tComplete: false,\n\t\t\tIssues: []string{\"Critical error: invalid format\"},\n\t\t\tNeedReply: false,\n\t\t}\n\t\tassert.False(t, validation.Passed)\n\t\tassert.False(t, validation.Complete)\n\t\tassert.False(t, validation.NeedReply)\n\t\tassert.Empty(t, validation.ReplyContent)\n\t})\n}\n\nfunc TestDeliveryResultStructure(t *testing.T) {\n\tsentAt := time.Now()\n\tdelivery := &types.DeliveryResult{\n\t\tRequestID: \"req-12345\",\n\t\tContent: &types.DeliveryContent{\n\t\t\tSummary: \"Weekly sales report completed\",\n\t\t\tBody: \"# Weekly Report\\n\\nSales increased by 20%...\",\n\t\t\tAttachments: []types.DeliveryAttachment{\n\t\t\t\t{\n\t\t\t\t\tTitle: \"Sales Report\",\n\t\t\t\t\tDescription: \"Detailed sales analysis\",\n\t\t\t\t\tTaskID: \"task-1\",\n\t\t\t\t\tFile: \"__s3://report-12345.pdf\",\n\t\t\t\t},\n\t\t\t},\n\t\t},\n\t\tResults: []types.ChannelResult{\n\t\t\t{\n\t\t\t\tType: types.DeliveryEmail,\n\t\t\t\tTarget: \"user@example.com\",\n\t\t\t\tSuccess: true,\n\t\t\t\tDetails: map[string]interface{}{\n\t\t\t\t\t\"message_id\": \"msg-12345\",\n\t\t\t\t},\n\t\t\t},\n\t\t\t{\n\t\t\t\tType: types.DeliveryWebhook,\n\t\t\t\tTarget: \"https://webhook.example.com/notify\",\n\t\t\t\tSuccess: true,\n\t\t\t},\n\t\t},\n\t\tSuccess: true,\n\t\tSentAt: &sentAt,\n\t}\n\n\tassert.Equal(t, \"req-12345\", delivery.RequestID)\n\tassert.True(t, delivery.Success)\n\tassert.NotNil(t, delivery.Content)\n\tassert.Equal(t, \"Weekly sales report completed\", delivery.Content.Summary)\n\tassert.Contains(t, delivery.Content.Body, \"Weekly Report\")\n\tassert.Len(t, delivery.Content.Attachments, 1)\n\tassert.Equal(t, \"__s3://report-12345.pdf\", delivery.Content.Attachments[0].File)\n\tassert.Len(t, delivery.Results, 2)\n\tassert.Equal(t, types.DeliveryEmail, delivery.Results[0].Type)\n\tassert.NotNil(t, delivery.SentAt)\n}\n\nfunc TestDeliveryContentStructure(t *testing.T) {\n\tcontent := &types.DeliveryContent{\n\t\tSummary: \"Task execution completed successfully\",\n\t\tBody: \"# Execution Report\\n\\n## Summary\\n- 3 tasks completed\\n- 1 task failed\",\n\t\tAttachments: []types.DeliveryAttachment{\n\t\t\t{\n\t\t\t\tTitle: \"Analysis Results\",\n\t\t\t\tDescription: \"JSON data from analysis task\",\n\t\t\t\tTaskID: \"task-analysis\",\n\t\t\t\tFile: \"__local://files/analysis-result.json\",\n\t\t\t},\n\t\t\t{\n\t\t\t\tTitle: \"Generated Chart\",\n\t\t\t\tTaskID: \"task-chart\",\n\t\t\t\tFile: \"__s3://charts/sales-chart.png\",\n\t\t\t},\n\t\t},\n\t}\n\n\tassert.NotEmpty(t, content.Summary)\n\tassert.Contains(t, content.Body, \"Execution Report\")\n\tassert.Len(t, content.Attachments, 2)\n\tassert.Equal(t, \"Analysis Results\", content.Attachments[0].Title)\n\tassert.Equal(t, \"task-analysis\", content.Attachments[0].TaskID)\n}\n\nfunc TestDeliveryAttachmentStructure(t *testing.T) {\n\tattachment := &types.DeliveryAttachment{\n\t\tTitle: \"Sales Report PDF\",\n\t\tDescription: \"Monthly sales analysis report\",\n\t\tTaskID: \"task-report\",\n\t\tFile: \"__s3://reports/sales-2024-01.pdf\",\n\t}\n\n\tassert.Equal(t, \"Sales Report PDF\", attachment.Title)\n\tassert.Equal(t, \"Monthly sales analysis report\", attachment.Description)\n\tassert.Equal(t, \"task-report\", attachment.TaskID)\n\tassert.Contains(t, attachment.File, \"__s3://\")\n}\n\nfunc TestDeliveryRequestStructure(t *testing.T) {\n\trequest := &types.DeliveryRequest{\n\t\tContent: &types.DeliveryContent{\n\t\t\tSummary: \"Report ready\",\n\t\t\tBody: \"# Report\\n\\nDetails...\",\n\t\t},\n\t\tContext: &types.DeliveryContext{\n\t\t\tMemberID: \"member-123\",\n\t\t\tExecutionID: \"exec-456\",\n\t\t\tTriggerType: types.TriggerClock,\n\t\t\tTeamID: \"team-789\",\n\t\t},\n\t}\n\n\tassert.NotNil(t, request.Content)\n\tassert.NotNil(t, request.Context)\n\tassert.Equal(t, \"member-123\", request.Context.MemberID)\n\tassert.Equal(t, \"exec-456\", request.Context.ExecutionID)\n\tassert.Equal(t, types.TriggerClock, request.Context.TriggerType)\n}\n\nfunc TestDeliveryPreferencesStructure(t *testing.T) {\n\tprefs := &types.DeliveryPreferences{\n\t\tEmail: &types.EmailPreference{\n\t\t\tEnabled: true,\n\t\t\tTargets: []types.EmailTarget{\n\t\t\t\t{\n\t\t\t\t\tTo: []string{\"team@example.com\"},\n\t\t\t\t\tTemplate: \"weekly-report\",\n\t\t\t\t\tSubject: \"Weekly Report - {{.Date}}\",\n\t\t\t\t},\n\t\t\t\t{\n\t\t\t\t\tTo: []string{\"backup@example.com\"},\n\t\t\t\t},\n\t\t\t},\n\t\t},\n\t\tWebhook: &types.WebhookPreference{\n\t\t\tEnabled: true,\n\t\t\tTargets: []types.WebhookTarget{\n\t\t\t\t{\n\t\t\t\t\tURL: \"https://api.example.com/webhook\",\n\t\t\t\t\tMethod: \"POST\",\n\t\t\t\t\tHeaders: map[string]string{\n\t\t\t\t\t\t\"X-API-Key\": \"secret-key\",\n\t\t\t\t\t},\n\t\t\t\t\tSecret: \"signing-secret\",\n\t\t\t\t},\n\t\t\t},\n\t\t},\n\t\tProcess: &types.ProcessPreference{\n\t\t\tEnabled: true,\n\t\t\tTargets: []types.ProcessTarget{\n\t\t\t\t{\n\t\t\t\t\tProcess: \"scripts.notify.slack\",\n\t\t\t\t\tArgs: []any{\"#general\", \"Report ready\"},\n\t\t\t\t},\n\t\t\t},\n\t\t},\n\t}\n\n\t// Email\n\tassert.True(t, prefs.Email.Enabled)\n\tassert.Len(t, prefs.Email.Targets, 2)\n\tassert.Equal(t, \"weekly-report\", prefs.Email.Targets[0].Template)\n\tassert.Len(t, prefs.Email.Targets[0].To, 1)\n\n\t// Webhook\n\tassert.True(t, prefs.Webhook.Enabled)\n\tassert.Len(t, prefs.Webhook.Targets, 1)\n\tassert.Equal(t, \"https://api.example.com/webhook\", prefs.Webhook.Targets[0].URL)\n\tassert.Equal(t, \"POST\", prefs.Webhook.Targets[0].Method)\n\n\t// Process\n\tassert.True(t, prefs.Process.Enabled)\n\tassert.Len(t, prefs.Process.Targets, 1)\n\tassert.Equal(t, \"scripts.notify.slack\", prefs.Process.Targets[0].Process)\n\tassert.Len(t, prefs.Process.Targets[0].Args, 2)\n}\n\nfunc TestChannelResultStructure(t *testing.T) {\n\tt.Run(\"email result with recipients\", func(t *testing.T) {\n\t\tsentAt := time.Now()\n\t\tresult := &types.ChannelResult{\n\t\t\tType: types.DeliveryEmail,\n\t\t\tTarget: \"user@example.com\",\n\t\t\tSuccess: true,\n\t\t\tRecipients: []string{\"user@example.com\", \"manager@example.com\"},\n\t\t\tDetails: map[string]interface{}{\n\t\t\t\t\"message_id\": \"msg-123\",\n\t\t\t},\n\t\t\tSentAt: &sentAt,\n\t\t}\n\t\tassert.Equal(t, types.DeliveryEmail, result.Type)\n\t\tassert.Equal(t, \"user@example.com\", result.Target)\n\t\tassert.True(t, result.Success)\n\t\tassert.Len(t, result.Recipients, 2)\n\t\tassert.NotNil(t, result.SentAt)\n\t})\n\n\tt.Run(\"webhook result\", func(t *testing.T) {\n\t\tsentAt := time.Now()\n\t\tresult := &types.ChannelResult{\n\t\t\tType: types.DeliveryWebhook,\n\t\t\tTarget: \"https://api.example.com/webhook\",\n\t\t\tSuccess: true,\n\t\t\tDetails: map[string]interface{}{\n\t\t\t\t\"status_code\": 200,\n\t\t\t\t\"response\": \"OK\",\n\t\t\t},\n\t\t\tSentAt: &sentAt,\n\t\t}\n\t\tassert.Equal(t, types.DeliveryWebhook, result.Type)\n\t\tassert.True(t, result.Success)\n\t\tassert.NotNil(t, result.SentAt)\n\t})\n\n\tt.Run(\"process result\", func(t *testing.T) {\n\t\tresult := &types.ChannelResult{\n\t\t\tType: types.DeliveryProcess,\n\t\t\tTarget: \"scripts.notify.slack\",\n\t\t\tSuccess: true,\n\t\t\tDetails: map[string]interface{}{\n\t\t\t\t\"output\": \"Message sent\",\n\t\t\t},\n\t\t}\n\t\tassert.Equal(t, types.DeliveryProcess, result.Type)\n\t\tassert.Equal(t, \"scripts.notify.slack\", result.Target)\n\t})\n\n\tt.Run(\"failed result\", func(t *testing.T) {\n\t\tresult := &types.ChannelResult{\n\t\t\tType: types.DeliveryWebhook,\n\t\t\tTarget: \"https://api.example.com/webhook\",\n\t\t\tSuccess: false,\n\t\t\tError: \"Connection refused\",\n\t\t}\n\t\tassert.False(t, result.Success)\n\t\tassert.Equal(t, \"Connection refused\", result.Error)\n\t})\n}\n\nfunc TestDeliveryTargetStructure(t *testing.T) {\n\tdelivery := &types.DeliveryTarget{\n\t\tType: types.DeliveryEmail,\n\t\tRecipients: []string{\"team@example.com\"},\n\t\tFormat: \"markdown\",\n\t\tTemplate: \"weekly-report\",\n\t\tOptions: map[string]interface{}{\n\t\t\t\"cc\": []string{\"manager@example.com\"},\n\t\t},\n\t}\n\n\tassert.Equal(t, types.DeliveryEmail, delivery.Type)\n\tassert.Len(t, delivery.Recipients, 1)\n\tassert.Equal(t, \"markdown\", delivery.Format)\n\tassert.Equal(t, \"weekly-report\", delivery.Template)\n}\n\nfunc TestLearningEntryStructure(t *testing.T) {\n\tentry := &types.LearningEntry{\n\t\tType: types.LearnExecution,\n\t\tContent: \"Successfully completed task using assistant\",\n\t\tTags: []string{\"success\", \"assistant\"},\n\t\tMeta: map[string]interface{}{\n\t\t\t\"duration\": 1500,\n\t\t\t\"phase\": \"run\",\n\t\t},\n\t}\n\n\tassert.Equal(t, types.LearnExecution, entry.Type)\n\tassert.Equal(t, \"Successfully completed task using assistant\", entry.Content)\n\tassert.Len(t, entry.Tags, 2)\n\tassert.NotNil(t, entry.Meta)\n}\n" }, { "path": "agent/robot/utils/convert.go", "content": "package utils\n\nimport (\n\t\"encoding/json\"\n\t\"fmt\"\n\t\"time\"\n)\n\n// ==================== To Functions ====================\n// Convert any value to specified type (safe, returns zero value on failure)\n\n// ToString converts any value to string\nfunc ToString(v interface{}) string {\n\tif v == nil {\n\t\treturn \"\"\n\t}\n\tswitch val := v.(type) {\n\tcase string:\n\t\treturn val\n\tcase []byte:\n\t\treturn string(val)\n\tcase int:\n\t\treturn fmt.Sprintf(\"%d\", val)\n\tcase int8:\n\t\treturn fmt.Sprintf(\"%d\", val)\n\tcase int16:\n\t\treturn fmt.Sprintf(\"%d\", val)\n\tcase int32:\n\t\treturn fmt.Sprintf(\"%d\", val)\n\tcase int64:\n\t\treturn fmt.Sprintf(\"%d\", val)\n\tcase uint:\n\t\treturn fmt.Sprintf(\"%d\", val)\n\tcase uint8:\n\t\treturn fmt.Sprintf(\"%d\", val)\n\tcase uint16:\n\t\treturn fmt.Sprintf(\"%d\", val)\n\tcase uint32:\n\t\treturn fmt.Sprintf(\"%d\", val)\n\tcase uint64:\n\t\treturn fmt.Sprintf(\"%d\", val)\n\tcase float32:\n\t\treturn fmt.Sprintf(\"%g\", val)\n\tcase float64:\n\t\treturn fmt.Sprintf(\"%g\", val)\n\tcase bool:\n\t\tif val {\n\t\t\treturn \"true\"\n\t\t}\n\t\treturn \"false\"\n\tdefault:\n\t\tif str, err := json.Marshal(v); err == nil {\n\t\t\treturn string(str)\n\t\t}\n\t\treturn fmt.Sprintf(\"%v\", v)\n\t}\n}\n\n// ToBool converts any value to bool\nfunc ToBool(v interface{}) bool {\n\tif v == nil {\n\t\treturn false\n\t}\n\tswitch b := v.(type) {\n\tcase bool:\n\t\treturn b\n\tcase int:\n\t\treturn b != 0\n\tcase int8:\n\t\treturn b != 0\n\tcase int16:\n\t\treturn b != 0\n\tcase int32:\n\t\treturn b != 0\n\tcase int64:\n\t\treturn b != 0\n\tcase uint:\n\t\treturn b != 0\n\tcase uint8:\n\t\treturn b != 0\n\tcase uint16:\n\t\treturn b != 0\n\tcase uint32:\n\t\treturn b != 0\n\tcase uint64:\n\t\treturn b != 0\n\tcase float32:\n\t\treturn b != 0\n\tcase float64:\n\t\treturn b != 0\n\tcase string:\n\t\treturn b == \"true\" || b == \"1\" || b == \"yes\" || b == \"on\"\n\t}\n\treturn false\n}\n\n// ToInt converts any value to int\nfunc ToInt(v interface{}) int {\n\tif v == nil {\n\t\treturn 0\n\t}\n\tswitch n := v.(type) {\n\tcase int:\n\t\treturn n\n\tcase int8:\n\t\treturn int(n)\n\tcase int16:\n\t\treturn int(n)\n\tcase int32:\n\t\treturn int(n)\n\tcase int64:\n\t\treturn int(n)\n\tcase uint:\n\t\treturn int(n)\n\tcase uint8:\n\t\treturn int(n)\n\tcase uint16:\n\t\treturn int(n)\n\tcase uint32:\n\t\treturn int(n)\n\tcase uint64:\n\t\treturn int(n)\n\tcase float32:\n\t\treturn int(n)\n\tcase float64:\n\t\treturn int(n)\n\tcase string:\n\t\tvar i int\n\t\tfmt.Sscanf(n, \"%d\", &i)\n\t\treturn i\n\tcase bool:\n\t\tif n {\n\t\t\treturn 1\n\t\t}\n\t\treturn 0\n\t}\n\treturn 0\n}\n\n// ToInt64 converts any value to int64\nfunc ToInt64(v interface{}) int64 {\n\tif v == nil {\n\t\treturn 0\n\t}\n\tswitch n := v.(type) {\n\tcase int64:\n\t\treturn n\n\tcase int:\n\t\treturn int64(n)\n\tcase int8:\n\t\treturn int64(n)\n\tcase int16:\n\t\treturn int64(n)\n\tcase int32:\n\t\treturn int64(n)\n\tcase uint:\n\t\treturn int64(n)\n\tcase uint8:\n\t\treturn int64(n)\n\tcase uint16:\n\t\treturn int64(n)\n\tcase uint32:\n\t\treturn int64(n)\n\tcase uint64:\n\t\treturn int64(n)\n\tcase float32:\n\t\treturn int64(n)\n\tcase float64:\n\t\treturn int64(n)\n\tcase string:\n\t\tvar i int64\n\t\tfmt.Sscanf(n, \"%d\", &i)\n\t\treturn i\n\tcase bool:\n\t\tif n {\n\t\t\treturn 1\n\t\t}\n\t\treturn 0\n\t}\n\treturn 0\n}\n\n// ToFloat64 converts any value to float64\nfunc ToFloat64(v interface{}) float64 {\n\tif v == nil {\n\t\treturn 0\n\t}\n\tswitch f := v.(type) {\n\tcase float64:\n\t\treturn f\n\tcase float32:\n\t\treturn float64(f)\n\tcase int:\n\t\treturn float64(f)\n\tcase int8:\n\t\treturn float64(f)\n\tcase int16:\n\t\treturn float64(f)\n\tcase int32:\n\t\treturn float64(f)\n\tcase int64:\n\t\treturn float64(f)\n\tcase uint:\n\t\treturn float64(f)\n\tcase uint8:\n\t\treturn float64(f)\n\tcase uint16:\n\t\treturn float64(f)\n\tcase uint32:\n\t\treturn float64(f)\n\tcase uint64:\n\t\treturn float64(f)\n\tcase string:\n\t\tvar result float64\n\t\tfmt.Sscanf(f, \"%f\", &result)\n\t\treturn result\n\tcase bool:\n\t\tif f {\n\t\t\treturn 1\n\t\t}\n\t\treturn 0\n\t}\n\treturn 0\n}\n\n// ToTimestamp converts any value to *time.Time\n// Handles: time.Time, *time.Time, string (various formats), int64/float64 (unix timestamp)\nfunc ToTimestamp(v interface{}) *time.Time {\n\tif v == nil {\n\t\treturn nil\n\t}\n\tswitch t := v.(type) {\n\tcase time.Time:\n\t\treturn &t\n\tcase *time.Time:\n\t\treturn t\n\tcase string:\n\t\tif t == \"\" {\n\t\t\treturn nil\n\t\t}\n\t\t// Try common time formats\n\t\tformats := []string{\n\t\t\ttime.RFC3339,\n\t\t\ttime.RFC3339Nano,\n\t\t\t\"2006-01-02 15:04:05\",\n\t\t\t\"2006-01-02T15:04:05Z\",\n\t\t\t\"2006-01-02T15:04:05\",\n\t\t\t\"2006-01-02\",\n\t\t}\n\t\tfor _, format := range formats {\n\t\t\tif parsed, err := time.Parse(format, t); err == nil {\n\t\t\t\treturn &parsed\n\t\t\t}\n\t\t}\n\tcase int64:\n\t\t// Unix timestamp (seconds)\n\t\tparsed := time.Unix(t, 0)\n\t\treturn &parsed\n\tcase int:\n\t\tparsed := time.Unix(int64(t), 0)\n\t\treturn &parsed\n\tcase float64:\n\t\t// Unix timestamp (seconds as float)\n\t\tparsed := time.Unix(int64(t), 0)\n\t\treturn &parsed\n\t}\n\treturn nil\n}\n\n// ToJSONValue parses JSON from string/[]byte or returns already-parsed value\nfunc ToJSONValue(v interface{}) interface{} {\n\tif v == nil {\n\t\treturn nil\n\t}\n\tswitch data := v.(type) {\n\tcase string:\n\t\tif data == \"\" {\n\t\t\treturn nil\n\t\t}\n\t\tvar result interface{}\n\t\tif err := json.Unmarshal([]byte(data), &result); err != nil {\n\t\t\treturn nil\n\t\t}\n\t\treturn result\n\tcase []byte:\n\t\tif len(data) == 0 {\n\t\t\treturn nil\n\t\t}\n\t\tvar result interface{}\n\t\tif err := json.Unmarshal(data, &result); err != nil {\n\t\t\treturn nil\n\t\t}\n\t\treturn result\n\tcase map[string]interface{}, []interface{}:\n\t\t// Already parsed\n\t\treturn data\n\tdefault:\n\t\treturn v\n\t}\n}\n\n// ==================== Get Functions ====================\n// Safely get typed value from map[string]interface{}\n\n// GetString safely gets a string value from map\nfunc GetString(m map[string]interface{}, key string) string {\n\tif m == nil {\n\t\treturn \"\"\n\t}\n\tif v, ok := m[key]; ok {\n\t\treturn ToString(v)\n\t}\n\treturn \"\"\n}\n\n// GetBool safely gets a bool value from map\nfunc GetBool(m map[string]interface{}, key string) bool {\n\tif m == nil {\n\t\treturn false\n\t}\n\tif v, ok := m[key]; ok {\n\t\treturn ToBool(v)\n\t}\n\treturn false\n}\n\n// GetInt safely gets an int value from map\nfunc GetInt(m map[string]interface{}, key string) int {\n\tif m == nil {\n\t\treturn 0\n\t}\n\tif v, ok := m[key]; ok {\n\t\treturn ToInt(v)\n\t}\n\treturn 0\n}\n\n// GetInt64 safely gets an int64 value from map\nfunc GetInt64(m map[string]interface{}, key string) int64 {\n\tif m == nil {\n\t\treturn 0\n\t}\n\tif v, ok := m[key]; ok {\n\t\treturn ToInt64(v)\n\t}\n\treturn 0\n}\n\n// GetFloat64 safely gets a float64 value from map\nfunc GetFloat64(m map[string]interface{}, key string) float64 {\n\tif m == nil {\n\t\treturn 0\n\t}\n\tif v, ok := m[key]; ok {\n\t\treturn ToFloat64(v)\n\t}\n\treturn 0\n}\n\n// GetTimestamp safely gets a *time.Time value from map\nfunc GetTimestamp(m map[string]interface{}, key string) *time.Time {\n\tif m == nil {\n\t\treturn nil\n\t}\n\tif v, ok := m[key]; ok {\n\t\treturn ToTimestamp(v)\n\t}\n\treturn nil\n}\n\n// GetJSONValue safely gets a parsed JSON value from map\nfunc GetJSONValue(m map[string]interface{}, key string) interface{} {\n\tif m == nil {\n\t\treturn nil\n\t}\n\tif v, ok := m[key]; ok {\n\t\treturn ToJSONValue(v)\n\t}\n\treturn nil\n}\n\n// ==================== JSON/Map Conversion ====================\n\n// ToJSON converts any value to JSON string\nfunc ToJSON(v interface{}) (string, error) {\n\tdata, err := json.Marshal(v)\n\tif err != nil {\n\t\treturn \"\", err\n\t}\n\treturn string(data), nil\n}\n\n// FromJSON parses JSON string to target struct\nfunc FromJSON(jsonStr string, target interface{}) error {\n\treturn json.Unmarshal([]byte(jsonStr), target)\n}\n\n// ToMap converts struct to map[string]interface{}\nfunc ToMap(v interface{}) (map[string]interface{}, error) {\n\tdata, err := json.Marshal(v)\n\tif err != nil {\n\t\treturn nil, err\n\t}\n\tvar result map[string]interface{}\n\tif err := json.Unmarshal(data, &result); err != nil {\n\t\treturn nil, err\n\t}\n\treturn result, nil\n}\n\n// FromMap converts map to struct\nfunc FromMap(m map[string]interface{}, target interface{}) error {\n\tdata, err := json.Marshal(m)\n\tif err != nil {\n\t\treturn err\n\t}\n\treturn json.Unmarshal(data, target)\n}\n\n// ==================== Map Utilities ====================\n\n// MergeMap merges source map into target map (shallow copy)\nfunc MergeMap(target, source map[string]interface{}) map[string]interface{} {\n\tif target == nil {\n\t\ttarget = make(map[string]interface{})\n\t}\n\tfor k, v := range source {\n\t\ttarget[k] = v\n\t}\n\treturn target\n}\n\n// CloneMap creates a shallow copy of a map\nfunc CloneMap(m map[string]interface{}) map[string]interface{} {\n\tif m == nil {\n\t\treturn nil\n\t}\n\tresult := make(map[string]interface{}, len(m))\n\tfor k, v := range m {\n\t\tresult[k] = v\n\t}\n\treturn result\n}\n" }, { "path": "agent/robot/utils/convert_test.go", "content": "package utils_test\n\nimport (\n\t\"testing\"\n\t\"time\"\n\n\t\"github.com/stretchr/testify/assert\"\n\t\"github.com/yaoapp/yao/agent/robot/utils\"\n)\n\n// ==================== To Tests ====================\n\nfunc TestToBool(t *testing.T) {\n\tt.Run(\"from_bool\", func(t *testing.T) {\n\t\tassert.True(t, utils.ToBool(true))\n\t\tassert.False(t, utils.ToBool(false))\n\t})\n\n\tt.Run(\"from_int\", func(t *testing.T) {\n\t\tassert.True(t, utils.ToBool(1))\n\t\tassert.True(t, utils.ToBool(42))\n\t\tassert.False(t, utils.ToBool(0))\n\t})\n\n\tt.Run(\"from_int64\", func(t *testing.T) {\n\t\tassert.True(t, utils.ToBool(int64(1)))\n\t\tassert.False(t, utils.ToBool(int64(0)))\n\t})\n\n\tt.Run(\"from_float64\", func(t *testing.T) {\n\t\tassert.True(t, utils.ToBool(1.0))\n\t\tassert.True(t, utils.ToBool(0.1))\n\t\tassert.False(t, utils.ToBool(0.0))\n\t})\n\n\tt.Run(\"from_string\", func(t *testing.T) {\n\t\tassert.True(t, utils.ToBool(\"true\"))\n\t\tassert.True(t, utils.ToBool(\"1\"))\n\t\tassert.True(t, utils.ToBool(\"yes\"))\n\t\tassert.True(t, utils.ToBool(\"on\"))\n\t\tassert.False(t, utils.ToBool(\"false\"))\n\t\tassert.False(t, utils.ToBool(\"0\"))\n\t\tassert.False(t, utils.ToBool(\"\"))\n\t})\n\n\tt.Run(\"from_nil\", func(t *testing.T) {\n\t\tassert.False(t, utils.ToBool(nil))\n\t})\n\n\tt.Run(\"from_unsupported_type\", func(t *testing.T) {\n\t\tassert.False(t, utils.ToBool([]int{1, 2, 3}))\n\t})\n}\n\nfunc TestToInt(t *testing.T) {\n\tt.Run(\"from_int\", func(t *testing.T) {\n\t\tassert.Equal(t, 42, utils.ToInt(42))\n\t\tassert.Equal(t, -10, utils.ToInt(-10))\n\t})\n\n\tt.Run(\"from_int64\", func(t *testing.T) {\n\t\tassert.Equal(t, 100, utils.ToInt(int64(100)))\n\t})\n\n\tt.Run(\"from_float64\", func(t *testing.T) {\n\t\tassert.Equal(t, 42, utils.ToInt(42.9)) // truncates\n\t\tassert.Equal(t, -5, utils.ToInt(-5.7))\n\t})\n\n\tt.Run(\"from_string\", func(t *testing.T) {\n\t\tassert.Equal(t, 123, utils.ToInt(\"123\"))\n\t\tassert.Equal(t, -456, utils.ToInt(\"-456\"))\n\t\tassert.Equal(t, 0, utils.ToInt(\"invalid\"))\n\t})\n\n\tt.Run(\"from_bool\", func(t *testing.T) {\n\t\tassert.Equal(t, 1, utils.ToInt(true))\n\t\tassert.Equal(t, 0, utils.ToInt(false))\n\t})\n\n\tt.Run(\"from_nil\", func(t *testing.T) {\n\t\tassert.Equal(t, 0, utils.ToInt(nil))\n\t})\n}\n\nfunc TestToInt64(t *testing.T) {\n\tt.Run(\"from_int64\", func(t *testing.T) {\n\t\tassert.Equal(t, int64(9223372036854775807), utils.ToInt64(int64(9223372036854775807)))\n\t})\n\n\tt.Run(\"from_int\", func(t *testing.T) {\n\t\tassert.Equal(t, int64(42), utils.ToInt64(42))\n\t})\n\n\tt.Run(\"from_float64\", func(t *testing.T) {\n\t\tassert.Equal(t, int64(42), utils.ToInt64(42.9))\n\t})\n\n\tt.Run(\"from_string\", func(t *testing.T) {\n\t\tassert.Equal(t, int64(123456789), utils.ToInt64(\"123456789\"))\n\t})\n\n\tt.Run(\"from_nil\", func(t *testing.T) {\n\t\tassert.Equal(t, int64(0), utils.ToInt64(nil))\n\t})\n}\n\nfunc TestToFloat64(t *testing.T) {\n\tt.Run(\"from_float64\", func(t *testing.T) {\n\t\tassert.Equal(t, 3.14159, utils.ToFloat64(3.14159))\n\t})\n\n\tt.Run(\"from_float32\", func(t *testing.T) {\n\t\tassert.InDelta(t, 3.14, utils.ToFloat64(float32(3.14)), 0.001)\n\t})\n\n\tt.Run(\"from_int\", func(t *testing.T) {\n\t\tassert.Equal(t, 42.0, utils.ToFloat64(42))\n\t})\n\n\tt.Run(\"from_int64\", func(t *testing.T) {\n\t\tassert.Equal(t, 100.0, utils.ToFloat64(int64(100)))\n\t})\n\n\tt.Run(\"from_string\", func(t *testing.T) {\n\t\tassert.InDelta(t, 3.14, utils.ToFloat64(\"3.14\"), 0.001)\n\t\tassert.Equal(t, 0.0, utils.ToFloat64(\"invalid\"))\n\t})\n\n\tt.Run(\"from_bool\", func(t *testing.T) {\n\t\tassert.Equal(t, 1.0, utils.ToFloat64(true))\n\t\tassert.Equal(t, 0.0, utils.ToFloat64(false))\n\t})\n\n\tt.Run(\"from_nil\", func(t *testing.T) {\n\t\tassert.Equal(t, 0.0, utils.ToFloat64(nil))\n\t})\n}\n\nfunc TestToTimestamp(t *testing.T) {\n\tt.Run(\"from_time_Time\", func(t *testing.T) {\n\t\tnow := time.Now()\n\t\tresult := utils.ToTimestamp(now)\n\t\tassert.NotNil(t, result)\n\t\tassert.Equal(t, now.Unix(), result.Unix())\n\t})\n\n\tt.Run(\"from_time_Time_pointer\", func(t *testing.T) {\n\t\tnow := time.Now()\n\t\tresult := utils.ToTimestamp(&now)\n\t\tassert.NotNil(t, result)\n\t\tassert.Equal(t, now.Unix(), result.Unix())\n\t})\n\n\tt.Run(\"from_RFC3339_string\", func(t *testing.T) {\n\t\tresult := utils.ToTimestamp(\"2024-01-15T14:30:00Z\")\n\t\tassert.NotNil(t, result)\n\t\tassert.Equal(t, 2024, result.Year())\n\t\tassert.Equal(t, time.January, result.Month())\n\t\tassert.Equal(t, 15, result.Day())\n\t\tassert.Equal(t, 14, result.Hour())\n\t\tassert.Equal(t, 30, result.Minute())\n\t})\n\n\tt.Run(\"from_datetime_string\", func(t *testing.T) {\n\t\tresult := utils.ToTimestamp(\"2024-01-15 14:30:00\")\n\t\tassert.NotNil(t, result)\n\t\tassert.Equal(t, 2024, result.Year())\n\t})\n\n\tt.Run(\"from_date_string\", func(t *testing.T) {\n\t\tresult := utils.ToTimestamp(\"2024-01-15\")\n\t\tassert.NotNil(t, result)\n\t\tassert.Equal(t, 2024, result.Year())\n\t\tassert.Equal(t, 15, result.Day())\n\t})\n\n\tt.Run(\"from_unix_timestamp_int64\", func(t *testing.T) {\n\t\t// 2024-01-15 00:00:00 UTC\n\t\tresult := utils.ToTimestamp(int64(1705276800))\n\t\tassert.NotNil(t, result)\n\t\tassert.Equal(t, 2024, result.Year())\n\t})\n\n\tt.Run(\"from_unix_timestamp_float64\", func(t *testing.T) {\n\t\tresult := utils.ToTimestamp(float64(1705276800))\n\t\tassert.NotNil(t, result)\n\t\tassert.Equal(t, 2024, result.Year())\n\t})\n\n\tt.Run(\"from_empty_string\", func(t *testing.T) {\n\t\tresult := utils.ToTimestamp(\"\")\n\t\tassert.Nil(t, result)\n\t})\n\n\tt.Run(\"from_invalid_string\", func(t *testing.T) {\n\t\tresult := utils.ToTimestamp(\"not a date\")\n\t\tassert.Nil(t, result)\n\t})\n\n\tt.Run(\"from_nil\", func(t *testing.T) {\n\t\tresult := utils.ToTimestamp(nil)\n\t\tassert.Nil(t, result)\n\t})\n}\n\nfunc TestToJSONValue(t *testing.T) {\n\tt.Run(\"from_json_string_object\", func(t *testing.T) {\n\t\tresult := utils.ToJSONValue(`{\"name\":\"test\",\"age\":30}`)\n\t\tassert.NotNil(t, result)\n\t\tm, ok := result.(map[string]interface{})\n\t\tassert.True(t, ok)\n\t\tassert.Equal(t, \"test\", m[\"name\"])\n\t\tassert.Equal(t, float64(30), m[\"age\"])\n\t})\n\n\tt.Run(\"from_json_string_array\", func(t *testing.T) {\n\t\tresult := utils.ToJSONValue(`[\"a\",\"b\",\"c\"]`)\n\t\tassert.NotNil(t, result)\n\t\tarr, ok := result.([]interface{})\n\t\tassert.True(t, ok)\n\t\tassert.Len(t, arr, 3)\n\t\tassert.Equal(t, \"a\", arr[0])\n\t})\n\n\tt.Run(\"from_bytes\", func(t *testing.T) {\n\t\tresult := utils.ToJSONValue([]byte(`{\"key\":\"value\"}`))\n\t\tassert.NotNil(t, result)\n\t\tm, ok := result.(map[string]interface{})\n\t\tassert.True(t, ok)\n\t\tassert.Equal(t, \"value\", m[\"key\"])\n\t})\n\n\tt.Run(\"from_already_parsed_map\", func(t *testing.T) {\n\t\tinput := map[string]interface{}{\"foo\": \"bar\"}\n\t\tresult := utils.ToJSONValue(input)\n\t\tassert.Equal(t, input, result)\n\t})\n\n\tt.Run(\"from_already_parsed_array\", func(t *testing.T) {\n\t\tinput := []interface{}{\"a\", \"b\"}\n\t\tresult := utils.ToJSONValue(input)\n\t\tassert.Equal(t, input, result)\n\t})\n\n\tt.Run(\"from_empty_string\", func(t *testing.T) {\n\t\tresult := utils.ToJSONValue(\"\")\n\t\tassert.Nil(t, result)\n\t})\n\n\tt.Run(\"from_empty_bytes\", func(t *testing.T) {\n\t\tresult := utils.ToJSONValue([]byte{})\n\t\tassert.Nil(t, result)\n\t})\n\n\tt.Run(\"from_invalid_json\", func(t *testing.T) {\n\t\tresult := utils.ToJSONValue(\"not json\")\n\t\tassert.Nil(t, result)\n\t})\n\n\tt.Run(\"from_nil\", func(t *testing.T) {\n\t\tresult := utils.ToJSONValue(nil)\n\t\tassert.Nil(t, result)\n\t})\n\n\tt.Run(\"from_other_type_passthrough\", func(t *testing.T) {\n\t\t// Non-string, non-[]byte types are passed through\n\t\tresult := utils.ToJSONValue(42)\n\t\tassert.Equal(t, 42, result)\n\t})\n}\n\n// ==================== Get Tests ====================\n\nfunc TestGetString(t *testing.T) {\n\tm := map[string]interface{}{\n\t\t\"name\": \"test\",\n\t\t\"number\": 42,\n\t\t\"bool\": true,\n\t\t\"nil\": nil,\n\t}\n\n\tt.Run(\"existing_string_key\", func(t *testing.T) {\n\t\tassert.Equal(t, \"test\", utils.GetString(m, \"name\"))\n\t})\n\n\tt.Run(\"converts_number_to_string\", func(t *testing.T) {\n\t\tassert.Equal(t, \"42\", utils.GetString(m, \"number\"))\n\t})\n\n\tt.Run(\"converts_bool_to_string\", func(t *testing.T) {\n\t\tassert.Equal(t, \"true\", utils.GetString(m, \"bool\"))\n\t})\n\n\tt.Run(\"non_existent_key\", func(t *testing.T) {\n\t\tassert.Equal(t, \"\", utils.GetString(m, \"missing\"))\n\t})\n\n\tt.Run(\"nil_map\", func(t *testing.T) {\n\t\tassert.Equal(t, \"\", utils.GetString(nil, \"key\"))\n\t})\n\n\tt.Run(\"nil_value\", func(t *testing.T) {\n\t\tassert.Equal(t, \"\", utils.GetString(m, \"nil\"))\n\t})\n}\n\nfunc TestGetBool(t *testing.T) {\n\tm := map[string]interface{}{\n\t\t\"bool_true\": true,\n\t\t\"bool_false\": false,\n\t\t\"int_one\": 1,\n\t\t\"int_zero\": 0,\n\t\t\"string_true\": \"true\",\n\t}\n\n\tt.Run(\"bool_true\", func(t *testing.T) {\n\t\tassert.True(t, utils.GetBool(m, \"bool_true\"))\n\t})\n\n\tt.Run(\"bool_false\", func(t *testing.T) {\n\t\tassert.False(t, utils.GetBool(m, \"bool_false\"))\n\t})\n\n\tt.Run(\"int_one\", func(t *testing.T) {\n\t\tassert.True(t, utils.GetBool(m, \"int_one\"))\n\t})\n\n\tt.Run(\"int_zero\", func(t *testing.T) {\n\t\tassert.False(t, utils.GetBool(m, \"int_zero\"))\n\t})\n\n\tt.Run(\"string_true\", func(t *testing.T) {\n\t\tassert.True(t, utils.GetBool(m, \"string_true\"))\n\t})\n\n\tt.Run(\"non_existent_key\", func(t *testing.T) {\n\t\tassert.False(t, utils.GetBool(m, \"missing\"))\n\t})\n\n\tt.Run(\"nil_map\", func(t *testing.T) {\n\t\tassert.False(t, utils.GetBool(nil, \"key\"))\n\t})\n}\n\nfunc TestGetInt(t *testing.T) {\n\tm := map[string]interface{}{\n\t\t\"int\": 42,\n\t\t\"int64\": int64(100),\n\t\t\"float64\": 3.14,\n\t\t\"string\": \"123\",\n\t}\n\n\tt.Run(\"int\", func(t *testing.T) {\n\t\tassert.Equal(t, 42, utils.GetInt(m, \"int\"))\n\t})\n\n\tt.Run(\"int64\", func(t *testing.T) {\n\t\tassert.Equal(t, 100, utils.GetInt(m, \"int64\"))\n\t})\n\n\tt.Run(\"float64\", func(t *testing.T) {\n\t\tassert.Equal(t, 3, utils.GetInt(m, \"float64\"))\n\t})\n\n\tt.Run(\"string\", func(t *testing.T) {\n\t\tassert.Equal(t, 123, utils.GetInt(m, \"string\"))\n\t})\n\n\tt.Run(\"non_existent_key\", func(t *testing.T) {\n\t\tassert.Equal(t, 0, utils.GetInt(m, \"missing\"))\n\t})\n\n\tt.Run(\"nil_map\", func(t *testing.T) {\n\t\tassert.Equal(t, 0, utils.GetInt(nil, \"key\"))\n\t})\n}\n\nfunc TestGetInt64(t *testing.T) {\n\tm := map[string]interface{}{\n\t\t\"int64\": int64(9223372036854775807),\n\t\t\"int\": 42,\n\t\t\"string\": \"123456789\",\n\t}\n\n\tt.Run(\"int64\", func(t *testing.T) {\n\t\tassert.Equal(t, int64(9223372036854775807), utils.GetInt64(m, \"int64\"))\n\t})\n\n\tt.Run(\"int\", func(t *testing.T) {\n\t\tassert.Equal(t, int64(42), utils.GetInt64(m, \"int\"))\n\t})\n\n\tt.Run(\"string\", func(t *testing.T) {\n\t\tassert.Equal(t, int64(123456789), utils.GetInt64(m, \"string\"))\n\t})\n\n\tt.Run(\"nil_map\", func(t *testing.T) {\n\t\tassert.Equal(t, int64(0), utils.GetInt64(nil, \"key\"))\n\t})\n}\n\nfunc TestGetFloat64(t *testing.T) {\n\tm := map[string]interface{}{\n\t\t\"float64\": 3.14159,\n\t\t\"int\": 42,\n\t\t\"string\": \"2.718\",\n\t}\n\n\tt.Run(\"float64\", func(t *testing.T) {\n\t\tassert.Equal(t, 3.14159, utils.GetFloat64(m, \"float64\"))\n\t})\n\n\tt.Run(\"int\", func(t *testing.T) {\n\t\tassert.Equal(t, 42.0, utils.GetFloat64(m, \"int\"))\n\t})\n\n\tt.Run(\"string\", func(t *testing.T) {\n\t\tassert.InDelta(t, 2.718, utils.GetFloat64(m, \"string\"), 0.001)\n\t})\n\n\tt.Run(\"nil_map\", func(t *testing.T) {\n\t\tassert.Equal(t, 0.0, utils.GetFloat64(nil, \"key\"))\n\t})\n}\n\nfunc TestGetTimestamp(t *testing.T) {\n\tnow := time.Now()\n\tm := map[string]interface{}{\n\t\t\"time\": now,\n\t\t\"time_ptr\": &now,\n\t\t\"rfc3339\": \"2024-01-15T14:30:00Z\",\n\t\t\"unix\": int64(1705276800),\n\t\t\"empty\": \"\",\n\t\t\"nil_value\": nil,\n\t}\n\n\tt.Run(\"time_value\", func(t *testing.T) {\n\t\tresult := utils.GetTimestamp(m, \"time\")\n\t\tassert.NotNil(t, result)\n\t\tassert.Equal(t, now.Unix(), result.Unix())\n\t})\n\n\tt.Run(\"time_ptr\", func(t *testing.T) {\n\t\tresult := utils.GetTimestamp(m, \"time_ptr\")\n\t\tassert.NotNil(t, result)\n\t})\n\n\tt.Run(\"rfc3339_string\", func(t *testing.T) {\n\t\tresult := utils.GetTimestamp(m, \"rfc3339\")\n\t\tassert.NotNil(t, result)\n\t\tassert.Equal(t, 2024, result.Year())\n\t})\n\n\tt.Run(\"unix_timestamp\", func(t *testing.T) {\n\t\tresult := utils.GetTimestamp(m, \"unix\")\n\t\tassert.NotNil(t, result)\n\t})\n\n\tt.Run(\"empty_string\", func(t *testing.T) {\n\t\tresult := utils.GetTimestamp(m, \"empty\")\n\t\tassert.Nil(t, result)\n\t})\n\n\tt.Run(\"nil_value\", func(t *testing.T) {\n\t\tresult := utils.GetTimestamp(m, \"nil_value\")\n\t\tassert.Nil(t, result)\n\t})\n\n\tt.Run(\"non_existent_key\", func(t *testing.T) {\n\t\tresult := utils.GetTimestamp(m, \"missing\")\n\t\tassert.Nil(t, result)\n\t})\n\n\tt.Run(\"nil_map\", func(t *testing.T) {\n\t\tresult := utils.GetTimestamp(nil, \"key\")\n\t\tassert.Nil(t, result)\n\t})\n}\n\nfunc TestGetJSONValue(t *testing.T) {\n\tm := map[string]interface{}{\n\t\t\"json_string\": `{\"nested\":\"value\"}`,\n\t\t\"json_array\": `[1,2,3]`,\n\t\t\"parsed_map\": map[string]interface{}{\"foo\": \"bar\"},\n\t\t\"empty\": \"\",\n\t\t\"invalid\": \"not json\",\n\t}\n\n\tt.Run(\"json_string\", func(t *testing.T) {\n\t\tresult := utils.GetJSONValue(m, \"json_string\")\n\t\tassert.NotNil(t, result)\n\t\tnested, ok := result.(map[string]interface{})\n\t\tassert.True(t, ok)\n\t\tassert.Equal(t, \"value\", nested[\"nested\"])\n\t})\n\n\tt.Run(\"json_array\", func(t *testing.T) {\n\t\tresult := utils.GetJSONValue(m, \"json_array\")\n\t\tassert.NotNil(t, result)\n\t\tarr, ok := result.([]interface{})\n\t\tassert.True(t, ok)\n\t\tassert.Len(t, arr, 3)\n\t})\n\n\tt.Run(\"parsed_map\", func(t *testing.T) {\n\t\tresult := utils.GetJSONValue(m, \"parsed_map\")\n\t\tassert.NotNil(t, result)\n\t\tparsed, ok := result.(map[string]interface{})\n\t\tassert.True(t, ok)\n\t\tassert.Equal(t, \"bar\", parsed[\"foo\"])\n\t})\n\n\tt.Run(\"empty_string\", func(t *testing.T) {\n\t\tresult := utils.GetJSONValue(m, \"empty\")\n\t\tassert.Nil(t, result)\n\t})\n\n\tt.Run(\"invalid_json\", func(t *testing.T) {\n\t\tresult := utils.GetJSONValue(m, \"invalid\")\n\t\tassert.Nil(t, result)\n\t})\n\n\tt.Run(\"nil_map\", func(t *testing.T) {\n\t\tresult := utils.GetJSONValue(nil, \"key\")\n\t\tassert.Nil(t, result)\n\t})\n}\n\n// ==================== ToString Extended Tests ====================\n\nfunc TestToStringExtended(t *testing.T) {\n\tt.Run(\"from_nil\", func(t *testing.T) {\n\t\tassert.Equal(t, \"\", utils.ToString(nil))\n\t})\n\n\tt.Run(\"from_bytes\", func(t *testing.T) {\n\t\tassert.Equal(t, \"hello\", utils.ToString([]byte(\"hello\")))\n\t})\n\n\tt.Run(\"from_int_types\", func(t *testing.T) {\n\t\tassert.Equal(t, \"8\", utils.ToString(int8(8)))\n\t\tassert.Equal(t, \"16\", utils.ToString(int16(16)))\n\t\tassert.Equal(t, \"32\", utils.ToString(int32(32)))\n\t\tassert.Equal(t, \"64\", utils.ToString(int64(64)))\n\t})\n\n\tt.Run(\"from_uint_types\", func(t *testing.T) {\n\t\tassert.Equal(t, \"8\", utils.ToString(uint8(8)))\n\t\tassert.Equal(t, \"16\", utils.ToString(uint16(16)))\n\t\tassert.Equal(t, \"32\", utils.ToString(uint32(32)))\n\t\tassert.Equal(t, \"64\", utils.ToString(uint64(64)))\n\t})\n\n\tt.Run(\"from_float_formats_nicely\", func(t *testing.T) {\n\t\tassert.Equal(t, \"3.14\", utils.ToString(3.14))\n\t\tassert.Equal(t, \"1000\", utils.ToString(1000.0)) // no trailing zeros\n\t})\n\n\tt.Run(\"from_struct_to_json\", func(t *testing.T) {\n\t\ttype TestStruct struct {\n\t\t\tName string `json:\"name\"`\n\t\t}\n\t\tresult := utils.ToString(TestStruct{Name: \"test\"})\n\t\tassert.Contains(t, result, \"test\")\n\t})\n}\n" }, { "path": "agent/robot/utils/id.go", "content": "package utils\n\nimport (\n\tgonanoid \"github.com/matoous/go-nanoid/v2\"\n)\n\n// NewID generates a new unique ID using nanoid\nfunc NewID() string {\n\tid, err := gonanoid.New()\n\tif err != nil {\n\t\t// Fallback to nanoid with default alphabet if error occurs\n\t\treturn gonanoid.Must()\n\t}\n\treturn id\n}\n\n// NewIDWithPrefix generates a new ID with a prefix\nfunc NewIDWithPrefix(prefix string) string {\n\treturn prefix + NewID()\n}\n" }, { "path": "agent/robot/utils/time.go", "content": "package utils\n\nimport (\n\t\"fmt\"\n\t\"strconv\"\n\t\"strings\"\n\t\"time\"\n)\n\n// ParseTime parses a time string in HH:MM format\nfunc ParseTime(timeStr string) (hour, minute int, err error) {\n\tparts := strings.Split(timeStr, \":\")\n\tif len(parts) != 2 {\n\t\treturn 0, 0, fmt.Errorf(\"invalid time format: %s (expected HH:MM)\", timeStr)\n\t}\n\n\thour, err = strconv.Atoi(parts[0])\n\tif err != nil || hour < 0 || hour > 23 {\n\t\treturn 0, 0, fmt.Errorf(\"invalid hour: %s\", parts[0])\n\t}\n\n\tminute, err = strconv.Atoi(parts[1])\n\tif err != nil || minute < 0 || minute > 59 {\n\t\treturn 0, 0, fmt.Errorf(\"invalid minute: %s\", parts[1])\n\t}\n\n\treturn hour, minute, nil\n}\n\n// FormatTime formats hour and minute into HH:MM format\nfunc FormatTime(hour, minute int) string {\n\treturn fmt.Sprintf(\"%02d:%02d\", hour, minute)\n}\n\n// LoadLocation loads a timezone location, returns Local if empty or invalid\nfunc LoadLocation(tz string) *time.Location {\n\tif tz == \"\" {\n\t\treturn time.Local\n\t}\n\tloc, err := time.LoadLocation(tz)\n\tif err != nil {\n\t\treturn time.Local\n\t}\n\treturn loc\n}\n\n// ParseDuration parses a duration string with fallback default\nfunc ParseDuration(durStr string, defaultDur time.Duration) time.Duration {\n\tif durStr == \"\" {\n\t\treturn defaultDur\n\t}\n\td, err := time.ParseDuration(durStr)\n\tif err != nil {\n\t\treturn defaultDur\n\t}\n\treturn d\n}\n\n// IsTimeMatch checks if current time matches the specified time (HH:MM)\nfunc IsTimeMatch(now time.Time, timeStr string, loc *time.Location) bool {\n\thour, minute, err := ParseTime(timeStr)\n\tif err != nil {\n\t\treturn false\n\t}\n\n\tnowInLoc := now.In(loc)\n\treturn nowInLoc.Hour() == hour && nowInLoc.Minute() == minute\n}\n\n// IsDayMatch checks if current day matches the specified day\n// days can be: \"Mon\", \"Tue\", \"Wed\", \"Thu\", \"Fri\", \"Sat\", \"Sun\", or \"*\" for any day\nfunc IsDayMatch(now time.Time, days []string) bool {\n\tif len(days) == 0 {\n\t\treturn true\n\t}\n\n\tdayName := now.Weekday().String()[:3] // \"Monday\" -> \"Mon\"\n\n\tfor _, day := range days {\n\t\tif day == \"*\" || day == dayName {\n\t\t\treturn true\n\t\t}\n\t}\n\treturn false\n}\n\n// NextScheduledTime calculates the next time a scheduled time will occur\nfunc NextScheduledTime(now time.Time, timeStr string, days []string, loc *time.Location) (time.Time, error) {\n\thour, minute, err := ParseTime(timeStr)\n\tif err != nil {\n\t\treturn time.Time{}, err\n\t}\n\n\tnowInLoc := now.In(loc)\n\n\t// Start from today at the specified time\n\tnext := time.Date(nowInLoc.Year(), nowInLoc.Month(), nowInLoc.Day(), hour, minute, 0, 0, loc)\n\n\t// If the time has passed today, start from tomorrow\n\tif next.Before(nowInLoc) || next.Equal(nowInLoc) {\n\t\tnext = next.Add(24 * time.Hour)\n\t}\n\n\t// Find the next matching day (within 7 days)\n\tfor i := 0; i < 7; i++ {\n\t\tif IsDayMatch(next, days) {\n\t\t\treturn next, nil\n\t\t}\n\t\tnext = next.Add(24 * time.Hour)\n\t}\n\n\t// If no matching day found (should not happen with valid days), return the calculated time\n\treturn next, nil\n}\n" }, { "path": "agent/robot/utils/utils_test.go", "content": "package utils_test\n\nimport (\n\t\"testing\"\n\t\"time\"\n\n\t\"github.com/stretchr/testify/assert\"\n\t\"github.com/yaoapp/yao/agent/robot/utils\"\n)\n\n// ID tests\nfunc TestNewID(t *testing.T) {\n\tid1 := utils.NewID()\n\tid2 := utils.NewID()\n\n\tassert.NotEmpty(t, id1)\n\tassert.NotEmpty(t, id2)\n\tassert.NotEqual(t, id1, id2, \"IDs should be unique\")\n}\n\nfunc TestNewIDWithPrefix(t *testing.T) {\n\tid := utils.NewIDWithPrefix(\"exec_\")\n\tassert.NotEmpty(t, id)\n\tassert.Contains(t, id, \"exec_\")\n}\n\n// Time tests\nfunc TestParseTime(t *testing.T) {\n\tt.Run(\"valid time\", func(t *testing.T) {\n\t\thour, minute, err := utils.ParseTime(\"14:30\")\n\t\tassert.NoError(t, err)\n\t\tassert.Equal(t, 14, hour)\n\t\tassert.Equal(t, 30, minute)\n\t})\n\n\tt.Run(\"invalid format\", func(t *testing.T) {\n\t\t_, _, err := utils.ParseTime(\"14-30\")\n\t\tassert.Error(t, err)\n\t})\n\n\tt.Run(\"invalid hour\", func(t *testing.T) {\n\t\t_, _, err := utils.ParseTime(\"25:30\")\n\t\tassert.Error(t, err)\n\t})\n\n\tt.Run(\"invalid minute\", func(t *testing.T) {\n\t\t_, _, err := utils.ParseTime(\"14:65\")\n\t\tassert.Error(t, err)\n\t})\n}\n\nfunc TestFormatTime(t *testing.T) {\n\tresult := utils.FormatTime(9, 5)\n\tassert.Equal(t, \"09:05\", result)\n\n\tresult = utils.FormatTime(14, 30)\n\tassert.Equal(t, \"14:30\", result)\n}\n\nfunc TestLoadLocation(t *testing.T) {\n\tt.Run(\"valid timezone\", func(t *testing.T) {\n\t\tloc := utils.LoadLocation(\"Asia/Shanghai\")\n\t\tassert.NotNil(t, loc)\n\t\tassert.Equal(t, \"Asia/Shanghai\", loc.String())\n\t})\n\n\tt.Run(\"empty timezone returns Local\", func(t *testing.T) {\n\t\tloc := utils.LoadLocation(\"\")\n\t\tassert.Equal(t, time.Local, loc)\n\t})\n\n\tt.Run(\"invalid timezone returns Local\", func(t *testing.T) {\n\t\tloc := utils.LoadLocation(\"Invalid/Timezone\")\n\t\tassert.Equal(t, time.Local, loc)\n\t})\n}\n\nfunc TestParseDuration(t *testing.T) {\n\tt.Run(\"valid duration\", func(t *testing.T) {\n\t\tdur := utils.ParseDuration(\"30m\", 10*time.Minute)\n\t\tassert.Equal(t, 30*time.Minute, dur)\n\t})\n\n\tt.Run(\"empty returns default\", func(t *testing.T) {\n\t\tdur := utils.ParseDuration(\"\", 10*time.Minute)\n\t\tassert.Equal(t, 10*time.Minute, dur)\n\t})\n\n\tt.Run(\"invalid returns default\", func(t *testing.T) {\n\t\tdur := utils.ParseDuration(\"invalid\", 10*time.Minute)\n\t\tassert.Equal(t, 10*time.Minute, dur)\n\t})\n}\n\nfunc TestIsTimeMatch(t *testing.T) {\n\tloc := time.UTC\n\ttestTime := time.Date(2024, 1, 15, 14, 30, 0, 0, loc)\n\n\tt.Run(\"exact match\", func(t *testing.T) {\n\t\tassert.True(t, utils.IsTimeMatch(testTime, \"14:30\", loc))\n\t})\n\n\tt.Run(\"no match\", func(t *testing.T) {\n\t\tassert.False(t, utils.IsTimeMatch(testTime, \"14:31\", loc))\n\t\tassert.False(t, utils.IsTimeMatch(testTime, \"15:30\", loc))\n\t})\n\n\tt.Run(\"invalid time format\", func(t *testing.T) {\n\t\tassert.False(t, utils.IsTimeMatch(testTime, \"invalid\", loc))\n\t})\n}\n\nfunc TestIsDayMatch(t *testing.T) {\n\tmonday := time.Date(2024, 1, 15, 10, 0, 0, 0, time.UTC) // Monday\n\n\tt.Run(\"match specific day\", func(t *testing.T) {\n\t\tassert.True(t, utils.IsDayMatch(monday, []string{\"Mon\"}))\n\t})\n\n\tt.Run(\"match wildcard\", func(t *testing.T) {\n\t\tassert.True(t, utils.IsDayMatch(monday, []string{\"*\"}))\n\t})\n\n\tt.Run(\"no match\", func(t *testing.T) {\n\t\tassert.False(t, utils.IsDayMatch(monday, []string{\"Tue\", \"Wed\"}))\n\t})\n\n\tt.Run(\"empty days returns true\", func(t *testing.T) {\n\t\tassert.True(t, utils.IsDayMatch(monday, []string{}))\n\t})\n}\n\n// Convert tests\nfunc TestToJSON(t *testing.T) {\n\tdata := map[string]interface{}{\n\t\t\"name\": \"test\",\n\t\t\"age\": 30,\n\t}\n\n\tjson, err := utils.ToJSON(data)\n\tassert.NoError(t, err)\n\tassert.Contains(t, json, \"test\")\n\tassert.Contains(t, json, \"30\")\n}\n\nfunc TestFromJSON(t *testing.T) {\n\tjsonStr := `{\"name\":\"test\",\"age\":30}`\n\n\tvar result map[string]interface{}\n\terr := utils.FromJSON(jsonStr, &result)\n\n\tassert.NoError(t, err)\n\tassert.Equal(t, \"test\", result[\"name\"])\n\tassert.Equal(t, float64(30), result[\"age\"]) // JSON numbers are float64\n}\n\nfunc TestToMap(t *testing.T) {\n\ttype TestStruct struct {\n\t\tName string `json:\"name\"`\n\t\tAge int `json:\"age\"`\n\t}\n\n\ts := TestStruct{Name: \"test\", Age: 30}\n\tm, err := utils.ToMap(s)\n\n\tassert.NoError(t, err)\n\tassert.Equal(t, \"test\", m[\"name\"])\n\tassert.Equal(t, float64(30), m[\"age\"]) // JSON conversion makes it float64\n}\n\nfunc TestFromMap(t *testing.T) {\n\ttype TestStruct struct {\n\t\tName string `json:\"name\"`\n\t\tAge int `json:\"age\"`\n\t}\n\n\tm := map[string]interface{}{\n\t\t\"name\": \"test\",\n\t\t\"age\": 30,\n\t}\n\n\tvar result TestStruct\n\terr := utils.FromMap(m, &result)\n\n\tassert.NoError(t, err)\n\tassert.Equal(t, \"test\", result.Name)\n\tassert.Equal(t, 30, result.Age)\n}\n\nfunc TestToString(t *testing.T) {\n\tassert.Equal(t, \"test\", utils.ToString(\"test\"))\n\tassert.Equal(t, \"42\", utils.ToString(42))\n\tassert.Equal(t, \"true\", utils.ToString(true))\n}\n\nfunc TestMergeMap(t *testing.T) {\n\ttarget := map[string]interface{}{\n\t\t\"a\": 1,\n\t\t\"b\": 2,\n\t}\n\tsource := map[string]interface{}{\n\t\t\"b\": 3,\n\t\t\"c\": 4,\n\t}\n\n\tresult := utils.MergeMap(target, source)\n\tassert.Equal(t, 1, result[\"a\"])\n\tassert.Equal(t, 3, result[\"b\"]) // overwritten\n\tassert.Equal(t, 4, result[\"c\"])\n}\n\nfunc TestCloneMap(t *testing.T) {\n\toriginal := map[string]interface{}{\n\t\t\"a\": 1,\n\t\t\"b\": 2,\n\t}\n\n\tcloned := utils.CloneMap(original)\n\tcloned[\"a\"] = 999\n\n\tassert.Equal(t, 1, original[\"a\"]) // original unchanged\n\tassert.Equal(t, 999, cloned[\"a\"])\n}\n\n// Validate tests\nfunc TestIsEmpty(t *testing.T) {\n\tassert.True(t, utils.IsEmpty(\"\"))\n\tassert.False(t, utils.IsEmpty(\"test\"))\n}\n\nfunc TestIsValidEmail(t *testing.T) {\n\tassert.True(t, utils.IsValidEmail(\"test@example.com\"))\n\tassert.True(t, utils.IsValidEmail(\"user+tag@domain.co.uk\"))\n\tassert.False(t, utils.IsValidEmail(\"invalid\"))\n\tassert.False(t, utils.IsValidEmail(\"@example.com\"))\n\tassert.False(t, utils.IsValidEmail(\"test@\"))\n}\n\nfunc TestIsValidTime(t *testing.T) {\n\tassert.True(t, utils.IsValidTime(\"09:00\"))\n\tassert.True(t, utils.IsValidTime(\"14:30\"))\n\tassert.True(t, utils.IsValidTime(\"23:59\"))\n\tassert.False(t, utils.IsValidTime(\"25:00\"))\n\tassert.False(t, utils.IsValidTime(\"14:65\"))\n\tassert.False(t, utils.IsValidTime(\"14-30\"))\n}\n\nfunc TestValidateRequired(t *testing.T) {\n\tt.Run(\"nil value\", func(t *testing.T) {\n\t\terr := utils.ValidateRequired(\"field\", nil)\n\t\tassert.Error(t, err)\n\t})\n\n\tt.Run(\"empty string\", func(t *testing.T) {\n\t\terr := utils.ValidateRequired(\"field\", \"\")\n\t\tassert.Error(t, err)\n\t})\n\n\tt.Run(\"valid string\", func(t *testing.T) {\n\t\terr := utils.ValidateRequired(\"field\", \"value\")\n\t\tassert.NoError(t, err)\n\t})\n\n\tt.Run(\"empty slice\", func(t *testing.T) {\n\t\terr := utils.ValidateRequired(\"field\", []string{})\n\t\tassert.Error(t, err)\n\t})\n}\n\nfunc TestValidateRange(t *testing.T) {\n\tt.Run(\"within range\", func(t *testing.T) {\n\t\terr := utils.ValidateRange(\"field\", 5, 1, 10)\n\t\tassert.NoError(t, err)\n\t})\n\n\tt.Run(\"below range\", func(t *testing.T) {\n\t\terr := utils.ValidateRange(\"field\", 0, 1, 10)\n\t\tassert.Error(t, err)\n\t})\n\n\tt.Run(\"above range\", func(t *testing.T) {\n\t\terr := utils.ValidateRange(\"field\", 11, 1, 10)\n\t\tassert.Error(t, err)\n\t})\n}\n\nfunc TestValidateOneOf(t *testing.T) {\n\tallowed := []string{\"apple\", \"banana\", \"cherry\"}\n\n\tt.Run(\"valid value\", func(t *testing.T) {\n\t\terr := utils.ValidateOneOf(\"field\", \"banana\", allowed)\n\t\tassert.NoError(t, err)\n\t})\n\n\tt.Run(\"invalid value\", func(t *testing.T) {\n\t\terr := utils.ValidateOneOf(\"field\", \"orange\", allowed)\n\t\tassert.Error(t, err)\n\t})\n}\n" }, { "path": "agent/robot/utils/validate.go", "content": "package utils\n\nimport (\n\t\"fmt\"\n\t\"regexp\"\n)\n\nvar (\n\t// Email regex pattern\n\temailRegex = regexp.MustCompile(`^[a-zA-Z0-9._%+\\-]+@[a-zA-Z0-9.\\-]+\\.[a-zA-Z]{2,}$`)\n\n\t// Time pattern (HH:MM)\n\ttimeRegex = regexp.MustCompile(`^([01]?[0-9]|2[0-3]):[0-5][0-9]$`)\n)\n\n// IsEmpty checks if a string is empty or whitespace only\nfunc IsEmpty(s string) bool {\n\treturn len(s) == 0\n}\n\n// IsValidEmail validates email format\nfunc IsValidEmail(email string) bool {\n\treturn emailRegex.MatchString(email)\n}\n\n// IsValidTime validates time format (HH:MM)\nfunc IsValidTime(timeStr string) bool {\n\treturn timeRegex.MatchString(timeStr)\n}\n\n// ValidateRequired checks if required fields are present\nfunc ValidateRequired(fieldName string, value interface{}) error {\n\tif value == nil {\n\t\treturn fmt.Errorf(\"%s is required\", fieldName)\n\t}\n\n\tswitch v := value.(type) {\n\tcase string:\n\t\tif IsEmpty(v) {\n\t\t\treturn fmt.Errorf(\"%s is required\", fieldName)\n\t\t}\n\tcase []string:\n\t\tif len(v) == 0 {\n\t\t\treturn fmt.Errorf(\"%s is required\", fieldName)\n\t\t}\n\tcase map[string]interface{}:\n\t\tif len(v) == 0 {\n\t\t\treturn fmt.Errorf(\"%s is required\", fieldName)\n\t\t}\n\t}\n\n\treturn nil\n}\n\n// ValidateRange checks if a number is within range\nfunc ValidateRange(fieldName string, value, min, max int) error {\n\tif value < min || value > max {\n\t\treturn fmt.Errorf(\"%s must be between %d and %d\", fieldName, min, max)\n\t}\n\treturn nil\n}\n\n// ValidateOneOf checks if value is one of allowed values\nfunc ValidateOneOf(fieldName string, value string, allowed []string) error {\n\tfor _, a := range allowed {\n\t\tif value == a {\n\t\t\treturn nil\n\t\t}\n\t}\n\treturn fmt.Errorf(\"%s must be one of: %v\", fieldName, allowed)\n}\n\n// ValidateEmail validates email and returns error if invalid\nfunc ValidateEmail(fieldName string, email string) error {\n\tif !IsValidEmail(email) {\n\t\treturn fmt.Errorf(\"%s is not a valid email\", fieldName)\n\t}\n\treturn nil\n}\n\n// ValidateTimeFormat validates time format (HH:MM)\nfunc ValidateTimeFormat(fieldName string, timeStr string) error {\n\tif !IsValidTime(timeStr) {\n\t\treturn fmt.Errorf(\"%s must be in HH:MM format\", fieldName)\n\t}\n\treturn nil\n}\n" }, { "path": "agent/sandbox/DESIGN.md", "content": "# Agent Sandbox Design\n\nInject Coding Agent capabilities (Claude CLI, Cursor CLI) into Yao's LLM request pipeline via Docker-based sandbox execution.\n\n## 1. Architecture Overview\n\n```\n┌─────────────────────────────────────────────────────────────────┐\n│ Yao LLM Pipeline │\n│ │\n│ ┌───────────────────────────────────────────────────────────┐ │\n│ │ Before Hooks / Auth / Logging │ │\n│ └─────────────────────────┬─────────────────────────────────┘ │\n│ │ │\n│ ▼ │\n│ ┌───────────────────────────────────────────────────────────┐ │\n│ │ LLM Request Handler │ │\n│ │ │ │\n│ │ sandbox: nil → Direct LLM API call (default) │ │\n│ │ sandbox: config → Sandbox + Claude CLI │ │ ← Inject here\n│ │ sandbox: config → Sandbox + Cursor CLI (future) │ │\n│ │ │ │\n│ └─────────────────────────┬─────────────────────────────────┘ │\n│ │ │\n│ ▼ │\n│ ┌───────────────────────────────────────────────────────────┐ │\n│ │ After Hooks / Billing / Audit │ │\n│ └───────────────────────────────────────────────────────────┘ │\n└─────────────────────────────────────────────────────────────────┘\n```\n\n## 2. Design Benefits\n\n| Aspect | Standalone Executor | Pipeline Injection |\n| ------------- | ------------------------ | ----------------------- |\n| **Hooks** | Need re-implementation | ✅ Fully reused |\n| **Auth** | Need re-implementation | ✅ Fully reused |\n| **Logging** | Need re-implementation | ✅ Fully reused |\n| **Billing** | Need re-implementation | ✅ Fully reused |\n| **Errors** | Need re-implementation | ✅ Fully reused |\n| **Code size** | Large (full executor) | Small (one branch) |\n| **Extend** | One executor per agent | Just add agent type |\n\n## 3. Configuration\n\n### 3.1 Assistant Configuration\n\nSandbox is configured at the **Assistant level** (not Connector), because:\n\n- Same Connector can be shared by multiple Assistants\n- Some Assistants need Sandbox (Coding), others don't (Q&A)\n- Assistant defines \"behavior\", Connector defines \"connection\"\n\n```jsonc\n// assistants/coder/package.yao\n{\n \"name\": \"Coder Assistant\",\n \"connector\": \"deepseek.v3\",\n\n // Sandbox configuration\n \"sandbox\": {\n \"command\": \"claude\", // claude | cursor (future)\n \"image\": \"yaoapp/sandbox-claude:latest\", // Optional, auto-selected by agent\n \"max_memory\": \"4g\", // Optional\n \"max_cpu\": 2.0, // Optional\n \"timeout\": \"10m\", // Optional, execution timeout\n \"arguments\": {\n // Command-specific arguments (passed to Claude CLI)\n \"max_turns\": 20,\n \"permission_mode\": \"acceptEdits\"\n // Different commands may have different arguments\n }\n },\n\n // MCP servers\n \"mcp\": {\n \"servers\": [\n {\n \"server_id\": \"filesystem\",\n \"tools\": [\"read_file\", \"write_file\", \"list_directory\"]\n }\n ]\n }\n}\n```\n\n### 3.2 Directory Structure\n\nSkills are auto-discovered from `skills/` directory following the [Agent Skills](https://agentskills.io) open standard:\n\n```\nassistants/coder/\n├── package.yao # Assistant config\n├── prompts.yml # Prompts\n├── mcps/ # MCP server definitions\n└── skills/ # Skills directory (auto-discovered)\n ├── code-review/\n │ ├── SKILL.md # Required: instructions + metadata\n │ ├── scripts/ # Optional: executable code (Python, Bash, JS)\n │ ├── references/ # Optional: additional documentation\n │ └── assets/ # Optional: templates, images, data files\n └── deploy/\n ├── SKILL.md\n └── scripts/\n └── deploy.sh\n```\n\n### 3.3 SKILL.md Format\n\nEach skill must have a `SKILL.md` file with YAML frontmatter:\n\n```yaml\n---\nname: code-review # Required: must match parent directory name\ndescription: > # Required: when to use this skill\n Review code for bugs, security issues, and best practices.\n Use when the user asks to review, audit, or analyze code quality.\nlicense: Apache-2.0 # Optional\ncompatibility: Requires git # Optional: environment requirements\nmetadata: # Optional: arbitrary key-value pairs\n author: yao-team\n version: \"1.0\"\nallowed-tools: Bash(git:*) Read # Optional: pre-approved tools (experimental)\n---\n\n# Code Review\n\n## When to use this skill\nUse this skill when the user asks to review code...\n\n## Steps\n1. Check for security vulnerabilities\n2. Review code style and best practices\n3. Identify potential bugs\n...\n```\n\n| Field | Required | Description |\n|-------|----------|-------------|\n| `name` | Yes | 1-64 chars, lowercase + hyphens, must match directory name |\n| `description` | Yes | 1-1024 chars, describes what skill does and when to use it |\n| `license` | No | License name or reference to bundled LICENSE file |\n| `compatibility` | No | Environment requirements (tools, network access, etc.) |\n| `metadata` | No | Arbitrary key-value pairs for additional properties |\n| `allowed-tools` | No | Space-delimited list of pre-approved tools (experimental) |\n\n### 3.4 Skills Progressive Disclosure\n\nSkills use progressive disclosure to manage context efficiently:\n\n1. **Discovery**: At startup, agent loads only `name` and `description` of each skill\n2. **Activation**: When task matches a skill's description, agent reads full `SKILL.md`\n3. **Execution**: Agent follows instructions, loading `scripts/`, `references/`, `assets/` as needed\n\n### 3.5 No Sandbox (Default)\n\nIf `sandbox` is not configured, Assistant uses direct LLM API calls:\n\n```jsonc\n// assistants/qa/package.yao\n{\n \"name\": \"QA Assistant\",\n \"connector\": \"deepseek.v3\"\n // No sandbox config → direct API call\n}\n```\n\n## 4. Implementation\n\n### 4.1 Package Structure\n\n```\nyao/\n├── sandbox/ # Low-level sandbox infrastructure\n│ ├── manager.go # Container management (✅ Done)\n│ ├── config.go # Configuration (✅ Done)\n│ ├── ipc/ # IPC communication (✅ Done)\n│ │ ├── manager.go\n│ │ ├── session.go\n│ │ └── types.go\n│ ├── bridge/ # yao-bridge binary (✅ Done)\n│ │ └── main.go\n│ └── docker/ # Docker images (✅ Done)\n│ ├── base/\n│ ├── claude/\n│ └── build.sh\n│\n└── agent/\n └── sandbox/ # Agent-level sandbox integration (NEW)\n ├── DESIGN.md # This document\n ├── types.go # Common types and interfaces\n ├── executor.go # Factory function and registry\n ├── claude/ # Claude CLI agent\n │ ├── types.go # Claude-specific types (StreamMessage, ToolCall, etc.)\n │ ├── executor.go # Executor implementation\n │ ├── command.go # CLI command builder\n │ ├── stream.go # Stream output parser\n │ ├── environment.go # Container environment setup\n │ └── executor_test.go\n ├── cursor/ # Cursor CLI agent (future)\n │ └── README.md # Placeholder for future implementation\n └── sandbox_test.go # Integration tests\n```\n\n### 4.2 Types and Interfaces\n\n```go\n// agent/sandbox/types.go\npackage sandbox\n\nimport (\n \"time\"\n\n \"github.com/yaoapp/yao/agent/context\"\n \"github.com/yaoapp/yao/agent/output/message\"\n)\n\n// Executor executes LLM requests in sandbox\ntype Executor interface {\n // Execute runs the request and returns response\n Execute(ctx *context.Context, messages []context.Message, opts *Options) (*context.CompletionResponse, error)\n\n // Stream runs the request with streaming output\n Stream(ctx *context.Context, messages []context.Message, opts *Options, handler message.StreamFunc) (*context.CompletionResponse, error)\n\n // Filesystem operations (for Hooks)\n ReadFile(ctx context.Context, path string) ([]byte, error)\n WriteFile(ctx context.Context, path string, content []byte) error\n ListDir(ctx context.Context, path string) ([]os.FileInfo, error)\n\n // Command execution (for Hooks)\n Exec(ctx context.Context, cmd []string) (string, error)\n\n // Close releases container resources\n Close() error\n}\n\n// Options for sandbox execution\ntype Options struct {\n // Command type (claude, cursor)\n Command string `json:\"command\"`\n\n // Docker image (optional, auto-selected by agent)\n Image string `json:\"image,omitempty\"`\n\n // Resource limits\n MaxMemory string `json:\"max_memory,omitempty\"`\n MaxCPU float64 `json:\"max_cpu,omitempty\"`\n\n // Execution timeout\n Timeout time.Duration `json:\"timeout,omitempty\"`\n\n // Command-specific arguments (passed to CLI)\n Arguments map[string]interface{} `json:\"arguments,omitempty\"`\n\n // ========================================\n // Internal fields (auto-resolved by Yao)\n // Do NOT set these in package.yao config\n // ========================================\n\n // MCP configuration - auto-loaded from assistants/{name}/mcps/\n MCPConfig []byte `json:\"-\"`\n\n // Skills directory - auto-resolved to assistants/{name}/skills/\n SkillsDir string `json:\"-\"`\n\n // Connector settings - auto-resolved from connector config file\n // e.g., connectors/deepseek/v3.conn.yao → host, key, model\n ConnectorHost string `json:\"-\"`\n ConnectorKey string `json:\"-\"`\n Model string `json:\"-\"`\n}\n```\n\n### 4.3 Executor Factory\n\n```go\n// agent/sandbox/executor.go\npackage sandbox\n\nimport (\n \"fmt\"\n\n \"github.com/yaoapp/yao/agent/sandbox/claude\"\n \"github.com/yaoapp/yao/agent/sandbox/cursor\"\n \"github.com/yaoapp/yao/sandbox\"\n)\n\n// New creates an executor based on agent type\nfunc New(manager *sandbox.Manager, opts *Options) (Executor, error) {\n if opts == nil {\n return nil, fmt.Errorf(\"options cannot be nil\")\n }\n\n // Set default image if not specified\n if opts.Image == \"\" {\n opts.Image = DefaultImage(opts.Command)\n }\n\n switch opts.Command {\n case \"claude\":\n return claude.New(manager, opts)\n case \"cursor\":\n return cursor.New(manager, opts)\n default:\n return nil, fmt.Errorf(\"unknown command type: %s\", opts.Command)\n }\n}\n\n// DefaultImage returns the default Docker image for a command type\nfunc DefaultImage(command string) string {\n switch command {\n case \"claude\":\n return \"yaoapp/sandbox-claude:latest\"\n case \"cursor\":\n return \"yaoapp/sandbox-cursor:latest\"\n default:\n return \"\"\n }\n}\n```\n\n### 4.4 Claude Agent Implementation\n\nThe Claude agent is split into multiple files for maintainability:\n\n#### 4.4.1 Executor (claude/executor.go)\n\n```go\n// agent/sandbox/claude/executor.go\npackage claude\n\nimport (\n \"fmt\"\n\n \"github.com/yaoapp/yao/agent/context\"\n \"github.com/yaoapp/yao/agent/output/message\"\n sandbox \"github.com/yaoapp/yao/agent/sandbox\"\n infra \"github.com/yaoapp/yao/sandbox\"\n)\n\n// Executor executes requests via Claude CLI in sandbox\ntype Executor struct {\n manager *infra.Manager\n opts *sandbox.Options\n}\n\n// New creates a new Claude executor\nfunc New(manager *infra.Manager, opts *sandbox.Options) (*Executor, error) {\n return &Executor{\n manager: manager,\n opts: opts,\n }, nil\n}\n\n// Stream executes with streaming output\nfunc (e *Executor) Stream(\n ctx *context.Context,\n messages []context.Message,\n opts *sandbox.Options,\n handler message.StreamFunc,\n) (*context.CompletionResponse, error) {\n\n // 1. Get or create container\n container, err := e.manager.GetOrCreate(ctx.Context(), ctx.Authorized.UserID, ctx.ChatID)\n if err != nil {\n return nil, fmt.Errorf(\"failed to get container: %w\", err)\n }\n\n // 2. Prepare environment\n workDir := fmt.Sprintf(\"/workspace/%s/chat-%s\", ctx.Authorized.UserID, ctx.ChatID)\n if err := e.prepareEnvironment(ctx, container, workDir); err != nil {\n return nil, fmt.Errorf(\"failed to prepare environment: %w\", err)\n }\n\n // 3. Build CLI command\n cmd, env := BuildCommand(messages, opts, workDir)\n\n // 4. Execute in container with streaming\n reader, err := e.manager.Stream(ctx.Context(), container.Name, cmd, &infra.ExecOptions{\n Env: env,\n WorkingDir: workDir,\n })\n if err != nil {\n return nil, fmt.Errorf(\"failed to execute: %w\", err)\n }\n defer reader.Close()\n\n // 5. Parse streaming output\n return ParseStream(reader, handler)\n}\n\n// Execute runs without streaming (wraps Stream)\nfunc (e *Executor) Execute(ctx *context.Context, messages []context.Message, opts *sandbox.Options) (*context.CompletionResponse, error) {\n var result *context.CompletionResponse\n _, err := e.Stream(ctx, messages, opts, func(msg *message.Message) error {\n // Collect final result\n return nil\n })\n return result, err\n}\n\n// Close releases resources\nfunc (e *Executor) Close() error {\n return nil\n}\n```\n\n#### 4.4.2 Environment Setup (claude/environment.go)\n\n```go\n// agent/sandbox/claude/environment.go\npackage claude\n\nimport (\n \"path/filepath\"\n\n \"github.com/yaoapp/yao/agent/context\"\n sandbox \"github.com/yaoapp/yao/agent/sandbox\"\n infra \"github.com/yaoapp/yao/sandbox\"\n)\n\n// prepareEnvironment sets up the container environment\nfunc (e *Executor) prepareEnvironment(ctx *context.Context, container *infra.Container, workDir string) error {\n // Create work directory\n if err := e.manager.MkDir(ctx.Context(), container.Name, workDir); err != nil {\n return err\n }\n\n // Write MCP config\n if len(e.opts.MCPConfig) > 0 {\n mcpPath := filepath.Join(workDir, \".mcp.json\")\n if err := e.manager.WriteFile(ctx.Context(), container.Name, mcpPath, e.opts.MCPConfig); err != nil {\n return err\n }\n }\n\n // Copy skills if configured\n if e.opts.SkillsDir != \"\" {\n claudeDir := filepath.Join(workDir, \".claude\")\n if err := e.manager.MkDir(ctx.Context(), container.Name, claudeDir); err != nil {\n return err\n }\n targetSkillsDir := filepath.Join(claudeDir, \"skills\")\n if err := e.manager.CopyToContainer(ctx.Context(), container.Name, e.opts.SkillsDir, targetSkillsDir); err != nil {\n return err\n }\n }\n\n return nil\n}\n```\n\n#### 4.4.3 Command Builder (claude/command.go)\n\n```go\n// agent/sandbox/claude/command.go\npackage claude\n\nimport (\n \"fmt\"\n \"strconv\"\n \"strings\"\n\n \"github.com/yaoapp/yao/agent/context\"\n sandbox \"github.com/yaoapp/yao/agent/sandbox\"\n)\n\n// BuildCommand constructs the Claude CLI command\nfunc BuildCommand(messages []context.Message, opts *sandbox.Options, workDir string) ([]string, map[string]string) {\n cmd := []string{\n \"claude\",\n \"--print\",\n \"--output-format\", \"stream-json\",\n }\n\n // Model\n if opts.Model != \"\" {\n cmd = append(cmd, \"--model\", opts.Model)\n }\n\n // Agent-specific options from sandbox.options\n if opts.Arguments != nil {\n if maxTurns, ok := opts.Arguments[\"max_turns\"].(int); ok && maxTurns > 0 {\n cmd = append(cmd, \"--max-turns\", strconv.Itoa(maxTurns))\n }\n if permMode, ok := opts.Arguments[\"permission_mode\"].(string); ok && permMode != \"\" {\n cmd = append(cmd, \"--permission-mode\", permMode)\n }\n }\n\n // MCP config\n cmd = append(cmd, \"--mcp-config\", \".mcp.json\")\n\n // System prompt (with history)\n systemPrompt := buildSystemPrompt(messages)\n if systemPrompt != \"\" {\n cmd = append(cmd, \"--system-prompt\", systemPrompt)\n }\n\n // User prompt (last user message)\n prompt := extractUserPrompt(messages)\n cmd = append(cmd, prompt)\n\n // Environment variables\n env := map[string]string{}\n if opts.ConnectorHost != \"\" {\n env[\"ANTHROPIC_BASE_URL\"] = opts.ConnectorHost\n }\n if opts.ConnectorKey != \"\" {\n env[\"ANTHROPIC_API_KEY\"] = opts.ConnectorKey\n }\n\n return cmd, env\n}\n\n// buildSystemPrompt builds system prompt with conversation history\nfunc buildSystemPrompt(messages []context.Message) string {\n var systemParts []string\n var history []string\n\n for i, msg := range messages {\n if msg.Role == \"system\" {\n if content, ok := msg.Content.(string); ok {\n systemParts = append(systemParts, content)\n }\n continue\n }\n\n // Skip last user message (it becomes the prompt)\n if i == len(messages)-1 && msg.Role == \"user\" {\n continue\n }\n\n // Add to history\n if content, ok := msg.Content.(string); ok {\n history = append(history, fmt.Sprintf(\"[%s]: %s\", msg.Role, content))\n }\n }\n\n if len(history) > 0 {\n systemParts = append(systemParts,\n \"\\n## Conversation History:\\n\"+strings.Join(history, \"\\n\\n\"))\n }\n\n return strings.Join(systemParts, \"\\n\")\n}\n\n// extractUserPrompt gets the last user message\nfunc extractUserPrompt(messages []context.Message) string {\n for i := len(messages) - 1; i >= 0; i-- {\n if messages[i].Role == \"user\" {\n if content, ok := messages[i].Content.(string); ok {\n return content\n }\n }\n }\n return \"\"\n}\n```\n\n#### 4.4.4 Types (claude/types.go)\n\n```go\n// agent/sandbox/claude/types.go\npackage claude\n\n// StreamMessage represents a Claude CLI stream-json message\ntype StreamMessage struct {\n Type string `json:\"type\"`\n Message struct {\n Content []struct {\n Type string `json:\"type\"`\n Text string `json:\"text\"`\n } `json:\"content\"`\n } `json:\"message,omitempty\"`\n Result string `json:\"result,omitempty\"`\n TotalCostUSD float64 `json:\"total_cost_usd,omitempty\"`\n NumTurns int `json:\"num_turns,omitempty\"`\n DurationMs int64 `json:\"duration_ms,omitempty\"`\n Usage *struct {\n InputTokens int `json:\"input_tokens\"`\n OutputTokens int `json:\"output_tokens\"`\n } `json:\"usage,omitempty\"`\n}\n\n// ToolCall represents a tool invocation from Claude CLI\ntype ToolCall struct {\n ID string `json:\"id\"`\n Name string `json:\"name\"`\n Arguments map[string]interface{} `json:\"arguments\"`\n}\n\n// ToolResult represents a tool execution result\ntype ToolResult struct {\n ID string `json:\"id\"`\n Content string `json:\"content\"`\n IsError bool `json:\"is_error,omitempty\"`\n}\n```\n\n#### 4.4.5 Stream Parser (claude/stream.go)\n\n```go\n// agent/sandbox/claude/stream.go\npackage claude\n\nimport (\n \"bufio\"\n \"encoding/json\"\n \"io\"\n \"strings\"\n\n \"github.com/yaoapp/yao/agent/context\"\n \"github.com/yaoapp/yao/agent/output/message\"\n)\n\n// ParseStream parses Claude CLI stream-json output\nfunc ParseStream(reader io.ReadCloser, handler message.StreamFunc) (*context.CompletionResponse, error) {\n resp := &context.CompletionResponse{}\n var fullContent strings.Builder\n\n scanner := bufio.NewScanner(reader)\n for scanner.Scan() {\n line := scanner.Text()\n if line == \"\" {\n continue\n }\n\n var msg StreamMessage\n if err := json.Unmarshal([]byte(line), &msg); err != nil {\n continue\n }\n\n switch msg.Type {\n case \"assistant\":\n // Extract text content\n for _, content := range msg.Message.Content {\n if content.Type == \"text\" && content.Text != \"\" {\n fullContent.WriteString(content.Text)\n\n // Send to stream handler\n if handler != nil {\n handler(&message.Message{\n Type: \"text\",\n Data: content.Text,\n })\n }\n }\n }\n\n case \"result\":\n // Final result\n if msg.Result != \"\" {\n resp.Content = msg.Result\n }\n if msg.Usage != nil {\n resp.Usage = &context.Usage{\n PromptTokens: msg.Usage.InputTokens,\n CompletionTokens: msg.Usage.OutputTokens,\n TotalTokens: msg.Usage.InputTokens + msg.Usage.OutputTokens,\n }\n }\n resp.Extra = map[string]interface{}{\n \"total_cost_usd\": msg.TotalCostUSD,\n \"num_turns\": msg.NumTurns,\n \"duration_ms\": msg.DurationMs,\n }\n }\n }\n\n if err := scanner.Err(); err != nil {\n return nil, err\n }\n\n if resp.Content == \"\" {\n resp.Content = fullContent.String()\n }\n\n return resp, nil\n}\n```\n\n### 4.5 Sandbox Lifecycle\n\n#### 4.5.1 Design Principles\n\nThe sandbox follows a **stateless container + persistent workspace** model:\n\n| Component | Lifecycle | Storage |\n|-----------|-----------|---------|\n| **Container** | Per-request, disposable | None (stateless) |\n| **Workspace** | Persistent across requests | `{YAO_DATA_ROOT}/sandbox/workspace/{user}/chat-{chat_id}/` |\n| **Message History** | Managed by Yao | Yao's session store |\n\nThis means:\n- Container can be recreated anytime without losing state\n- All files are preserved in the mounted workspace\n- Conversation history is passed in each request (not stored in container)\n\n#### 4.5.2 Container Lifecycle\n\n```\nRequest Start\n │\n ▼\n┌─────────────────────────────────────────────────────────────┐\n│ 1. Create Executor (docker run --rm) │\n│ - Mount workspace directory │\n│ - Set resource limits │\n│ - ~500ms-1s cold start │\n└─────────────────────────┬───────────────────────────────────┘\n │\n ▼\n┌─────────────────────────────────────────────────────────────┐\n│ 2. Before Hook (can access Executor) │\n│ - Write config files │\n│ - Check/prepare environment │\n│ - Can reject request (container auto-cleaned) │\n└─────────────────────────┬───────────────────────────────────┘\n │\n ▼\n┌─────────────────────────────────────────────────────────────┐\n│ 3. LLM Execute (Claude CLI in container) │\n│ - Run with full message history │\n│ - Stream output to handler │\n└─────────────────────────┬───────────────────────────────────┘\n │\n ▼\n┌─────────────────────────────────────────────────────────────┐\n│ 4. After Hook (can access Executor) │\n│ - Read generated files │\n│ - Execute post-commands (git commit, etc.) │\n│ - Cleanup temp files │\n└─────────────────────────┬───────────────────────────────────┘\n │\n ▼\n┌─────────────────────────────────────────────────────────────┐\n│ 5. Close Executor (defer) │\n│ - Container removed (--rm) │\n│ - Workspace persists │\n└─────────────────────────────────────────────────────────────┘\n```\n\n#### 4.5.3 Workspace Cleanup\n\nWorkspace cleanup is separate from container lifecycle:\n\n```go\n// Global cleanup configuration\ntype CleanupConfig struct {\n // Workspace retention period (default: 7 days)\n WorkspaceRetention time.Duration `json:\"workspace_retention\"`\n \n // Run cleanup on schedule (default: daily at 3am)\n CleanupSchedule string `json:\"cleanup_schedule\"`\n}\n\n// Cleanup worker\nfunc (m *Manager) cleanupStaleWorkspaces() {\n // Find workspaces older than retention period\n // Delete directories not accessed within the period\n}\n```\n\n### 4.6 Hook Integration (via JSAPI)\n\nHooks interact with the sandbox via **Context JSAPI**, not Go code. The Executor methods are exposed to JavaScript through `ctx.sandbox`:\n\n#### 4.6.1 Context JSAPI Extension\n\n```typescript\n// Extension to Context interface (see agent/context/JSAPI.md)\ninterface Context {\n // ... existing properties ...\n \n // Sandbox operations (only available when sandbox is configured)\n sandbox?: {\n // Filesystem operations\n ReadFile(path: string): string; // Read file content\n WriteFile(path: string, content: string): void; // Write file\n ListDir(path: string): FileInfo[]; // List directory\n \n // Command execution\n Exec(cmd: string[]): string; // Execute command, return output\n \n // Workspace info\n workdir: string; // Container workspace path\n };\n}\n\ninterface FileInfo {\n name: string;\n size: number;\n is_dir: boolean;\n mod_time: string;\n}\n```\n\n#### 4.6.2 Create Hook Examples (JavaScript)\n\n```javascript\n// assistants/coder/src/index.ts\n\n/**\n * Create Hook - runs after container created, before LLM execution\n */\nfunction Create(ctx, messages, options) {\n // Check if sandbox is available\n if (!ctx.sandbox) {\n return { messages };\n }\n\n // Send loading message to user\n const loadingId = ctx.SendStream({\n type: \"loading\",\n props: { message: \"Preparing sandbox environment...\" }\n });\n\n // Write project configuration\n ctx.sandbox.WriteFile(\".env\", \"DEBUG=true\\nNODE_ENV=development\");\n \n // Check environment prerequisites\n try {\n const nodeVersion = ctx.sandbox.Exec([\"node\", \"--version\"]);\n log.Info(`Node.js version: ${nodeVersion}`);\n } catch (e) {\n log.Error(\"Node.js not available\");\n ctx.End(loadingId);\n throw new Error(\"Node.js is required\");\n }\n \n // List existing files\n const files = ctx.sandbox.ListDir(\".\");\n log.Debug(`Workspace files: ${files.map(f => f.name).join(\", \")}`);\n \n // Update loading message and end it\n ctx.Replace(loadingId, {\n type: \"text\",\n props: { content: \"Sandbox ready\" }\n });\n ctx.End(loadingId);\n \n return { messages };\n}\n```\n\n#### 4.6.3 Next Hook Examples (JavaScript)\n\n```javascript\n/**\n * Next Hook - runs after LLM execution, before container cleanup\n */\nfunction Next(ctx, payload, options) {\n const { completion, error } = payload;\n \n if (error || !ctx.sandbox) {\n return null;\n }\n\n // Read generated files\n try {\n const result = ctx.sandbox.ReadFile(\"output/result.json\");\n log.Info(`Generated output: ${result}`);\n ctx.memory.context.Set(\"generated_result\", JSON.parse(result));\n } catch (e) {\n log.Debug(\"No result file generated\");\n }\n \n // List generated source files\n const srcFiles = ctx.sandbox.ListDir(\"src/\");\n for (const file of srcFiles) {\n log.Debug(`Generated: ${file.name} (${file.size} bytes)`);\n }\n \n // Execute git commit if files changed\n try {\n ctx.sandbox.Exec([\"git\", \"add\", \".\"]);\n ctx.sandbox.Exec([\"git\", \"commit\", \"-m\", \"auto-commit by sandbox\"]);\n log.Info(\"Changes committed\");\n } catch (e) {\n log.Debug(`git commit skipped: ${e.message}`);\n }\n \n // Cleanup temporary files\n try {\n ctx.sandbox.Exec([\"rm\", \"-rf\", \"tmp/\"]);\n } catch (e) {\n // Ignore cleanup errors\n }\n \n return null; // Use default response\n}\n```\n\n#### 4.6.4 Go Implementation (Internal)\n\nThe Go layer only handles exposing the Executor to JSAPI:\n\n```go\n// In agent/context/sandbox.go - expose Executor to JS runtime\n\n// SandboxJSAPI wraps Executor for JavaScript access\ntype SandboxJSAPI struct {\n executor sandbox.Executor\n workdir string\n}\n\n// Methods are called from JavaScript via v8go bindings\nfunc (s *SandboxJSAPI) ReadFile(path string) (string, error) {\n content, err := s.executor.ReadFile(context.Background(), path)\n return string(content), err\n}\n\nfunc (s *SandboxJSAPI) WriteFile(path string, content string) error {\n return s.executor.WriteFile(context.Background(), path, []byte(content))\n}\n\nfunc (s *SandboxJSAPI) ListDir(path string) ([]FileInfo, error) {\n return s.executor.ListDir(context.Background(), path)\n}\n\nfunc (s *SandboxJSAPI) Exec(cmd []string) (string, error) {\n return s.executor.Exec(context.Background(), cmd)\n}\n```\n\n### 4.7 Integration with Assistant\n\nThe sandbox integration is separated into two files:\n\n- `agent/assistant/sandbox.go` - Sandbox handler (new file, contains all sandbox logic)\n- `agent/assistant/llm.go` - Add sandbox detection (minimal change)\n\n#### 4.7.1 Sandbox Detection (`agent/assistant/llm.go`)\n\n```go\n// In agent/assistant/llm.go - add sandbox detection\n\nfunc (ast *Assistant) executeLLMStream(...) (*context.CompletionResponse, error) {\n // Check if sandbox is configured\n if ast.Sandbox != nil && ast.Sandbox.Command != \"\" {\n return ast.executeSandboxStream(ctx, completionMessages, completionOptions, agentNode, streamHandler, opts)\n }\n\n // Default: direct LLM API call\n // ... existing code ...\n}\n```\n\n#### 4.7.2 Sandbox Handler (`agent/assistant/sandbox.go`)\n\n```go\n// In agent/assistant/sandbox.go - sandbox execution logic\n\nfunc (ast *Assistant) executeSandboxStream(...) (*context.CompletionResponse, error) {\n // Get sandbox manager (singleton)\n manager := sandbox.GetManager()\n if manager == nil {\n return nil, fmt.Errorf(\"sandbox manager not initialized\")\n }\n\n // Build executor options\n execOpts := &agentsandbox.Options{\n Command: ast.Sandbox.Command,\n Image: ast.Sandbox.Image,\n MaxMemory: ast.Sandbox.MaxMemory,\n MaxCPU: ast.Sandbox.MaxCPU,\n Timeout: ast.Sandbox.Timeout,\n Arguments: ast.Sandbox.Arguments,\n SkillsDir: filepath.Join(ast.Path, \"skills\"),\n }\n\n // Get connector settings (resolved from connector config)\n conn, _, err := ast.GetConnector(ctx, opts)\n if err != nil {\n return nil, err\n }\n setting := conn.Setting()\n if host, ok := setting[\"host\"].(string); ok {\n execOpts.ConnectorHost = host\n }\n if key, ok := setting[\"key\"].(string); ok {\n execOpts.ConnectorKey = key\n }\n if model, ok := setting[\"model\"].(string); ok {\n execOpts.Model = model\n }\n\n // Build MCP config\n execOpts.MCPConfig = ast.buildMCPConfig(ctx)\n\n // 1. Create executor (container starts here)\n ctx.Trace().Info(\"Creating sandbox container...\")\n executor, err := agentsandbox.New(manager, execOpts)\n if err != nil {\n ctx.Trace().Error(fmt.Sprintf(\"Sandbox creation failed: %v\", err))\n return nil, err\n }\n ctx.Trace().Info(\"Sandbox container ready\")\n defer executor.Close() // 5. Container cleanup on exit\n\n // 2. Before Hook - can access executor\n if err := ast.runBeforeHook(ctx, executor); err != nil {\n return nil, err // Container cleaned by defer\n }\n\n // 3. LLM Execute\n resp, err := executor.Stream(ctx, completionMessages, execOpts, streamHandler)\n if err != nil {\n return nil, err\n }\n\n // 4. After Hook - can access executor\n if err := ast.runAfterHook(ctx, executor, resp); err != nil {\n log.Printf(\"after hook error: %v\", err) // Log but don't fail\n }\n\n return resp, nil\n}\n```\n\n## 5. IPC Communication (Yao ↔ Sandbox)\n\nThe sandbox can call Yao processes via MCP:\n\n```\n┌─────────────────────────────────────────────────────────────────┐\n│ Sandbox Container │\n│ ┌─────────────────┐ ┌──────────────────┐ │\n│ │ Claude CLI │───▶│ yao-bridge │ │\n│ │ │ │ │ │\n│ │ MCP: tools/call│ │ Unix Socket │ │\n│ │ \"yao.process\" │ │ /tmp/yao.sock │ │\n│ └─────────────────┘ └────────┬─────────┘ │\n└──────────────────────────────────┼─────────────────────────────┘\n │ JSON-RPC\n ▼\n┌──────────────────────────────────────────────────────────────────┐\n│ Yao Host │\n│ ┌─────────────────────────────────────────────────────────────┐ │\n│ │ IPC Manager │ │\n│ │ │ │\n│ │ Session: user123-chat456 │ │\n│ │ Socket: /data/sandbox/ipc/user123-chat456.sock │ │\n│ │ │ │\n│ │ Handlers: │ │\n│ │ tools/list → Return authorized tool list │ │\n│ │ tools/call → Execute Yao Process │ │\n│ └─────────────────────────────────────────────────────────────┘ │\n└──────────────────────────────────────────────────────────────────┘\n```\n\n## 6. Skills Support\n\nSkills follow [Agent Skills](https://agentskills.io) standard:\n\n```markdown\n---\nname: code-review\ndescription: \"Review code for security, performance, and best practices\"\nallowed-tools:\n - Read\n - Grep\n - Glob\n---\n\n# Code Review\n\nWhen reviewing code:\n\n## Security\n- Check for SQL injection, XSS, CSRF\n- Verify input validation\n\n## Performance\n- Look for N+1 queries\n- Check for unnecessary re-renders\n```\n\nSkills are auto-copied to container at `{workDir}/.claude/skills/`.\n\n## 7. Docker Images\n\n### Available Images\n\n| Image | Contents | Size |\n|-------|----------|------|\n| `yaoapp/sandbox-base` | Ubuntu 24.04, git, vim, network tools, yao-bridge | ~200MB |\n| `yaoapp/sandbox-claude` | base + Node.js 22, Python 3.12, Claude CLI, CCR | ~900MB |\n| `yaoapp/sandbox-claude-full` | claude + Go 1.23 | ~1.2GB |\n\n### Included Tools\n\n- **Editors**: vim, less, tree\n- **Network**: ping, netstat, nslookup, nc, telnet\n- **Compression**: zip, unzip, tar, gzip\n- **System**: htop, ps, sed, awk, grep\n- **Development**: git, Node.js, Python, Claude CLI, CCR\n\n## 8. Implementation Status\n\nSee [PLAN.md](./PLAN.md) for detailed implementation tasks and testing requirements.\n\n### Completed (sandbox package)\n\n- [x] Container management (`sandbox/manager.go`)\n- [x] Configuration (`sandbox/config.go`)\n- [x] IPC communication (`sandbox/ipc/`)\n- [x] yao-bridge binary (`sandbox/bridge/`)\n- [x] Docker images (`sandbox/docker/`)\n\n### To Implement (agent/sandbox package)\n\n- [ ] Types and interfaces\n- [ ] Executor factory\n- [ ] Claude executor (claude/)\n- [ ] Context JSAPI bindings\n- [ ] Assistant integration\n- [ ] Workspace cleanup\n\n## 9. Testing\n\nSee [PLAN.md](./PLAN.md) for complete testing strategy.\n\n**Quick test:**\n\n```bash\n# Test Docker image\ndocker run --rm yaoapp/sandbox-claude:latest bash -c \"\n node --version && python3 --version && claude --version\n\"\n\n# Run integration tests\nsource /Users/max/Yao/yao/env.local.sh\ngo test -v ./agent/sandbox/...\n```\n\n## 10. Usage Example\n\n```bash\n# Test script at yao-dev-app/agents/claude/run.sh\n./run.sh \"Hello, what is 1+1?\" # Default: deepseek/v3\n./run.sh -c deepseek/r1 \"Hello\" # Use R1 model\n./run.sh -c claude/sonnet-4_0 \"Hello\" # Use Claude\n```\n" }, { "path": "agent/sandbox/PLAN.md", "content": "# Agent Sandbox Implementation Plan\n\n## Overview\n\nThis plan covers the implementation of the agent sandbox integration layer (`agent/sandbox/`), which enables coding agents (Claude CLI, Cursor CLI) to run in isolated Docker containers with Yao's LLM pipeline.\n\n## Test Environment\n\n### Environment Configuration\n\nTests should run with the local development environment:\n\n```bash\n# Source environment variables\nsource /Users/max/Yao/yao/env.local.sh\n\n# Key variables used:\n# YAO_TEST_APPLICATION=/Users/max/Yao/yao-dev-app\n# YAO_ROOT=$YAO_TEST_APPLICATION\n# DEEPSEEK_API_KEY, DEEPSEEK_API_PROXY, DEEPSEEK_MODELS_V3\n```\n\n### Test Application\n\nTest assistants at `yao-dev-app/assistants/tests/sandbox/`:\n\n```\nyao-dev-app/assistants/tests/\n└── sandbox/\n ├── basic/ # Basic sandbox execution test\n │ ├── package.yao # uses.search: disabled\n │ └── prompts.yml\n ├── hooks/ # Hook integration test\n │ ├── package.yao # uses.search: disabled\n │ ├── prompts.yml\n │ └── src/index.ts\n └── full/ # Full test with MCPs, Skills, Hooks\n ├── package.yao # uses.search: disabled, mcp: {servers: [...]}\n ├── prompts.yml\n ├── src/index.ts\n └── skills/echo-test/ # Agent Skills standard\n ├── SKILL.md\n └── scripts/echo.sh\n```\n\n### Connector Configuration\n\nUse `deepseek.v3` as the default connector (via Volcengine API).\n\n## Implementation Status\n\n### Phase 1: Core Types and Interfaces ✅ COMPLETED\n\n- [x] Define `Executor` interface with all methods\n- [x] Define `Options` struct with JSON tags\n- [x] Define `FileInfo` alias to infrastructure sandbox\n- [x] Add `DefaultImage()` and `IsValidCommand()` helpers\n\n### Phase 2: Claude Executor Implementation ✅ COMPLETED\n\n- [x] Implement `Executor` struct\n- [x] Implement `NewExecutor()` constructor with container reuse\n- [x] Implement `Stream()` method with CCR config writing\n- [x] Implement `Execute()` method (wrapper)\n- [x] Implement `Close()` method (removes container)\n- [x] Implement filesystem methods: `ReadFile`, `WriteFile`, `ListDir`\n- [x] Implement `Exec()` method\n- [x] Implement `GetWorkDir()` method\n\n### Phase 3: CCR Configuration ✅ COMPLETED\n\n- [x] Implement `BuildCCRConfig()` with correct CCR format\n- [x] Auto-detect provider type (volcengine, deepseek, openai, claude)\n- [x] Add transformer for DeepSeek/Volcengine (maxtoken)\n- [x] Generate Router configuration\n- [x] Write config to container before execution\n\n### Phase 4: Assistant Integration ✅ COMPLETED\n\n- [x] Implement `GetSandboxManager()` singleton\n- [x] Implement `HasSandbox()` method\n- [x] Implement `initSandbox()` with cleanup function\n- [x] Implement `executeSandboxStream()` method\n- [x] Build executor options from assistant config\n- [x] Resolve connector settings (host, key, model)\n- [x] Add trace logging for sandbox creation\n- [x] Send loading message during sandbox init\n- [x] Expose executor to hooks via `ctx.SetSandboxExecutor()`\n- [x] Handle sandbox lifecycle (create → hooks → execute → cleanup)\n\n### Phase 5: JSAPI Integration ✅ COMPLETED\n\n- [x] Define `SandboxExecutor` interface\n- [x] Implement JS bindings for `ReadFile`, `WriteFile`, `ListDir`, `Exec`\n- [x] Expose `workdir` property\n- [x] Register in context's `NewObject` method\n\n### Phase 6: Concurrency & Resource Management ✅ COMPLETED\n\n- [x] Container creation uses Double-Check Locking (in `manager.GetOrCreate`)\n- [x] Same chatID reuses container (by design)\n- [x] Container cleanup on request completion (`defer sandboxCleanup()`)\n- [x] Unique chatID in tests to avoid conflicts\n\n### Phase 7: MCP & Skills Integration ✅ COMPLETED\n\n- [x] Build MCP config from assistant's `mcp.servers` configuration\n- [x] Write MCP config to container workspace (`.mcp.json`)\n- [x] Resolve skills directory from `assistants/{name}/skills/`\n- [x] Copy skills to container (`/workspace/.claude/skills/`)\n- [x] Skip MCP tool execution in `agent.go` for sandbox mode (Claude CLI handles internally)\n- [x] Add unit tests for MCP config building (`TestBuildMCPConfigForSandbox`)\n- [x] Add unit tests for skills directory resolution (`TestSandboxMCPAndSkillsOptions`)\n\n### Phase 8: MCP IPC Bridge ✅ COMPLETED\n\n- [x] Modify `BuildMCPConfigForSandbox` to use `yao-bridge` command for IPC\n- [x] Create IPC session in `sandbox/manager.createContainer()` (socket created before container)\n- [x] Bind mount IPC socket to container at `/tmp/yao.sock`\n- [x] Add `SetMCPTools()` method to `ipc.Session` for runtime tool configuration\n- [x] Set MCP tools dynamically in `claude.Executor.Stream()` before execution\n- [x] IPC session lifecycle managed by `sandbox.Manager` (create on container create, close on remove)\n- [x] Load MCP tool definitions from gou/mcp and pass to IPC session\n- [x] Add `TestClaudeExecutorIPCSocketMount` to verify socket bind mount\n- [x] Verify E2E test shows \"Loaded X MCP tools for IPC\"\n\n### Phase 9: Workspace Management ⏳ PENDING\n\n- [ ] Implement workspace cleanup configuration\n- [ ] Implement stale workspace detection\n- [ ] Implement cleanup scheduler\n\n### Phase 9: Cursor Placeholder ⏳ PENDING\n\n- [ ] Create `cursor/README.md` placeholder\n\n## Testing Status\n\n### Unit Tests\n\n| Package | Test File | Status |\n|---------|-----------|--------|\n| `agent/sandbox` | `types_test.go` | ✅ PASS |\n| `agent/sandbox` | `executor_test.go` | ✅ PASS |\n| `agent/sandbox/claude` | `command_test.go` | ✅ PASS |\n| `agent/sandbox/claude` | `executor_test.go` | ✅ PASS |\n\n### Integration Tests\n\n| Package | Test File | Status |\n|---------|-----------|--------|\n| `agent/sandbox` | `integration_test.go` | ✅ PASS |\n\n### JSAPI Tests\n\n| Package | Test File | Status |\n|---------|-----------|--------|\n| `agent/context` | `jsapi_sandbox_test.go` | ✅ PASS |\n\n### Assistant Loading Tests\n\n| Package | Test File | Status |\n|---------|-----------|--------|\n| `agent/assistant` | `sandbox_test.go` | ✅ PASS |\n| `agent/assistant` | `sandbox_integration_test.go` | ✅ PASS |\n\n### E2E Tests\n\n| Package | Test Case | Status |\n|---------|-----------|--------|\n| `agent/assistant` | `TestSandboxBasicE2E` | ✅ PASS |\n| `agent/assistant` | `TestSandboxHooksE2E` | ✅ PASS |\n| `agent/assistant` | `TestSandboxFullE2E` | ✅ PASS |\n| `agent/assistant` | `TestSandboxContextAccess` | ✅ PASS |\n| `agent/assistant` | `TestSandboxLoadConfiguration` | ✅ PASS |\n| `agent/assistant` | `TestSandboxMCPToolCall` | ✅ PASS |\n| `agent/assistant` | `TestSandboxMCPEchoTool` | ✅ PASS |\n\n### Running Tests\n\n```bash\n# Source environment\nsource /Users/max/Yao/yao/env.local.sh\n\n# Run all sandbox tests\ngo test -v ./agent/sandbox/...\n\n# Run assistant sandbox tests\ngo test -v ./agent/assistant -run \"Sandbox\"\n\n# Run E2E tests (requires Docker)\ngo test -v ./agent/assistant -run \"TestSandbox.*E2E\" -timeout 300s\n```\n\n## File Structure\n\n```\nyao/agent/sandbox/ # Executor layer\n├── DESIGN.md # ✅ Design document\n├── PLAN.md # ✅ This file\n├── types.go # ✅ Common types and interfaces\n├── types_test.go # ✅ Types tests\n├── executor.go # ✅ Factory function\n├── executor_test.go # ✅ Factory tests\n├── integration_test.go # ✅ Integration tests\n├── claude/\n│ ├── types.go # ✅ Claude-specific types\n│ ├── executor.go # ✅ Executor implementation\n│ ├── executor_test.go # ✅ Executor tests\n│ ├── command.go # ✅ Command builder + CCR config\n│ └── command_test.go # ✅ Command tests\n└── cursor/\n └── README.md # ⏳ Placeholder (pending)\n\nyao/agent/assistant/ # Integration layer\n├── sandbox.go # ✅ Sandbox handler\n├── sandbox_test.go # ✅ Loading tests\n├── sandbox_integration_test.go # ✅ Integration tests\n├── sandbox_e2e_test.go # ✅ E2E tests\n├── sandbox_debug_test.go # ✅ Debug tests\n└── agent.go # ✅ Modified: sandbox detection in Stream()\n\nyao/agent/context/ # Context layer\n├── jsapi_sandbox.go # ✅ Sandbox JSAPI bindings\n└── jsapi_sandbox_test.go # ✅ Sandbox JSAPI tests\n\nyao-dev-app/assistants/tests/sandbox/ # Test assistants\n├── basic/ # ✅ Basic sandbox test\n├── hooks/ # ✅ Hooks test\n└── full/ # ✅ Full test with MCPs and Skills\n```\n\n## Key Design Decisions\n\n### 1. Container Reuse\n\nSame `userID + chatID` reuses the same container:\n- Workspace directory persists across requests\n- CCR config is written on each request (same content, safe to overwrite)\n- Container is removed when request completes\n\n### 2. Concurrency\n\n- Container creation: Protected by mutex + double-check locking\n- Container execution: Multiple requests can run concurrently in same container\n- Claude CLI: Supports concurrent execution\n\n### 3. CCR Configuration\n\nCCR requires specific JSON format:\n```json\n{\n \"Providers\": [{\"name\": \"volcengine\", \"api_base_url\": \"...\", ...}],\n \"Router\": {\"default\": \"volcengine,model\", ...}\n}\n```\n\nAuto-detection of provider type based on host URL.\n\n### 4. Resource Cleanup\n\n- `executor.Close()` removes the container and closes IPC session\n- `defer sandboxCleanup()` in `agent.go` ensures cleanup\n- Tests use unique chatID (timestamp) to avoid conflicts\n\n### 5. MCP IPC Architecture\n\n```\nHost (Yao) Container (Claude CLI)\n┌────────────────────────┐ ┌────────────────────────┐\n│ IPC Manager │ │ yao-bridge │\n│ └─ Session │◄─────────────│ (stdio ↔ socket) │\n│ └─ MCPTools │ Unix Socket │ │\n│ └─ Process │ (/tmp/ │ Claude CLI reads │\n│ executor │ yao.sock) │ .mcp.json and calls │\n└────────────────────────┘ │ yao-bridge for tools │\n └────────────────────────┘\n```\n\n- `.mcp.json` points to single \"yao\" server using `yao-bridge /tmp/yao.sock`\n- IPC session created with authorized MCP tools from assistant config\n- Tools executed via `process.New()` in IPC session handler\n\n## Known Issues\n\n### macOS Docker Desktop Socket Permissions\n\nOn macOS with Docker Desktop (gRPC-FUSE), Unix socket permissions are not properly preserved when bind mounting from the host. The IPC socket created on the host with `0666` permissions appears as `0660` inside the container.\n\n**Solution**: After container start, we execute `chmod 666 /tmp/yao.sock` as root inside the container to fix permissions. This is handled automatically by `sandbox.Manager.fixIPCSocketPermissions()`.\n\n## Notes\n\n- All tests validate return values (use `require`/`assert`)\n- Docker must be available for integration and E2E tests\n- Tests automatically clean up containers after completion\n- Use `uses.search: disabled` in test assistants to avoid auto-search LLM calls\n" }, { "path": "agent/sandbox/claude/attachments_test.go", "content": "package claude\n\nimport (\n\t\"context\"\n\t\"fmt\"\n\t\"testing\"\n\t\"time\"\n\n\t\"github.com/stretchr/testify/assert\"\n\t\"github.com/stretchr/testify/require\"\n\tagentContext \"github.com/yaoapp/yao/agent/context\"\n\t\"github.com/yaoapp/yao/config\"\n\t\"github.com/yaoapp/yao/test\"\n)\n\nfunc TestExtensionFromContentType(t *testing.T) {\n\ttests := []struct {\n\t\tcontentType string\n\t\texpected string\n\t}{\n\t\t{\"image/png\", \".png\"},\n\t\t{\"image/jpeg\", \".jpg\"},\n\t\t{\"image/gif\", \".gif\"},\n\t\t{\"image/webp\", \".webp\"},\n\t\t{\"image/svg+xml\", \".svg\"},\n\t\t{\"application/pdf\", \".pdf\"},\n\t\t{\"text/plain\", \".txt\"},\n\t\t{\"text/html\", \".html\"},\n\t\t{\"text/css\", \".css\"},\n\t\t{\"text/javascript\", \".js\"},\n\t\t{\"application/javascript\", \".js\"},\n\t\t{\"application/json\", \".json\"},\n\t\t{\"application/zip\", \".zip\"},\n\t\t{\"application/octet-stream\", \"\"},\n\t\t{\"unknown/type\", \"\"},\n\t}\n\n\tfor _, tt := range tests {\n\t\tt.Run(tt.contentType, func(t *testing.T) {\n\t\t\tassert.Equal(t, tt.expected, extensionFromContentType(tt.contentType))\n\t\t})\n\t}\n}\n\nfunc TestFormatFileSize(t *testing.T) {\n\ttests := []struct {\n\t\tbytes int\n\t\texpected string\n\t}{\n\t\t{0, \"0B\"},\n\t\t{100, \"100B\"},\n\t\t{1023, \"1023B\"},\n\t\t{1024, \"1.0KB\"},\n\t\t{1536, \"1.5KB\"},\n\t\t{10240, \"10.0KB\"},\n\t\t{1048576, \"1.0MB\"},\n\t\t{1572864, \"1.5MB\"},\n\t\t{10485760, \"10.0MB\"},\n\t}\n\n\tfor _, tt := range tests {\n\t\tt.Run(fmt.Sprintf(\"%d\", tt.bytes), func(t *testing.T) {\n\t\t\tassert.Equal(t, tt.expected, formatFileSize(tt.bytes))\n\t\t})\n\t}\n}\n\nfunc TestPrepareAttachmentsPlainText(t *testing.T) {\n\ttest.Prepare(t, config.Conf)\n\tdefer test.Clean()\n\n\tmanager := createTestManager(t)\n\tif manager == nil {\n\t\treturn\n\t}\n\tdefer manager.Close()\n\n\topts := &Options{\n\t\tCommand: \"claude\",\n\t\tImage: \"alpine:latest\",\n\t\tUserID: \"test-user\",\n\t\tChatID: fmt.Sprintf(\"test-chat-att-plain-%d\", time.Now().UnixNano()),\n\t}\n\n\texec, err := NewExecutor(manager, opts)\n\trequire.NoError(t, err)\n\tdefer exec.Close()\n\n\tctx := context.Background()\n\n\t// Plain text messages should pass through unchanged\n\tmessages := []agentContext.Message{\n\t\t{Role: \"system\", Content: \"You are a helpful assistant\"},\n\t\t{Role: \"user\", Content: \"Hello, world!\"},\n\t\t{Role: \"assistant\", Content: \"Hi there!\"},\n\t\t{Role: \"user\", Content: \"What is 1+1?\"},\n\t}\n\n\tresult, err := exec.prepareAttachments(ctx, messages)\n\trequire.NoError(t, err)\n\trequire.Len(t, result, 4)\n\n\t// Verify messages are unchanged\n\tassert.Equal(t, \"system\", string(result[0].Role))\n\tassert.Equal(t, \"You are a helpful assistant\", result[0].Content)\n\tassert.Equal(t, \"Hello, world!\", result[1].Content)\n\tassert.Equal(t, \"Hi there!\", result[2].Content)\n\tassert.Equal(t, \"What is 1+1?\", result[3].Content)\n}\n\nfunc TestPrepareAttachmentsMultimodalNoWrapper(t *testing.T) {\n\ttest.Prepare(t, config.Conf)\n\tdefer test.Clean()\n\n\tmanager := createTestManager(t)\n\tif manager == nil {\n\t\treturn\n\t}\n\tdefer manager.Close()\n\n\topts := &Options{\n\t\tCommand: \"claude\",\n\t\tImage: \"alpine:latest\",\n\t\tUserID: \"test-user\",\n\t\tChatID: fmt.Sprintf(\"test-chat-att-nowrap-%d\", time.Now().UnixNano()),\n\t}\n\n\texec, err := NewExecutor(manager, opts)\n\trequire.NoError(t, err)\n\tdefer exec.Close()\n\n\tctx := context.Background()\n\n\t// Multimodal message with a non-wrapper URL (e.g. regular http URL)\n\t// Should convert to text description but not try to resolve attachment\n\tmessages := []agentContext.Message{\n\t\t{\n\t\t\tRole: \"user\",\n\t\t\tContent: []interface{}{\n\t\t\t\tmap[string]interface{}{\"type\": \"text\", \"text\": \"Look at this\"},\n\t\t\t\tmap[string]interface{}{\n\t\t\t\t\t\"type\": \"image_url\",\n\t\t\t\t\t\"image_url\": map[string]interface{}{\n\t\t\t\t\t\t\"url\": \"https://example.com/image.png\",\n\t\t\t\t\t\t\"detail\": \"auto\",\n\t\t\t\t\t},\n\t\t\t\t},\n\t\t\t},\n\t\t},\n\t}\n\n\tresult, err := exec.prepareAttachments(ctx, messages)\n\trequire.NoError(t, err)\n\trequire.Len(t, result, 1)\n\n\t// Content should be converted to text with URL reference\n\tcontent, ok := result[0].Content.(string)\n\trequire.True(t, ok, \"Content should be converted to string\")\n\tassert.Contains(t, content, \"Look at this\")\n\tassert.Contains(t, content, \"[Image: https://example.com/image.png]\")\n}\n\nfunc TestPrepareAttachmentsTextOnlyMultimodal(t *testing.T) {\n\ttest.Prepare(t, config.Conf)\n\tdefer test.Clean()\n\n\tmanager := createTestManager(t)\n\tif manager == nil {\n\t\treturn\n\t}\n\tdefer manager.Close()\n\n\topts := &Options{\n\t\tCommand: \"claude\",\n\t\tImage: \"alpine:latest\",\n\t\tUserID: \"test-user\",\n\t\tChatID: fmt.Sprintf(\"test-chat-att-textonly-%d\", time.Now().UnixNano()),\n\t}\n\n\texec, err := NewExecutor(manager, opts)\n\trequire.NoError(t, err)\n\tdefer exec.Close()\n\n\tctx := context.Background()\n\n\t// Multimodal message with only text parts\n\tmessages := []agentContext.Message{\n\t\t{\n\t\t\tRole: \"user\",\n\t\t\tContent: []interface{}{\n\t\t\t\tmap[string]interface{}{\"type\": \"text\", \"text\": \"Hello\"},\n\t\t\t\tmap[string]interface{}{\"type\": \"text\", \"text\": \"World\"},\n\t\t\t},\n\t\t},\n\t}\n\n\tresult, err := exec.prepareAttachments(ctx, messages)\n\trequire.NoError(t, err)\n\trequire.Len(t, result, 1)\n\n\t// Should combine text parts\n\tcontent, ok := result[0].Content.(string)\n\trequire.True(t, ok, \"Content should be converted to string\")\n\tassert.Contains(t, content, \"Hello\")\n\tassert.Contains(t, content, \"World\")\n}\n\nfunc TestPrepareAttachmentsInvalidWrapperURL(t *testing.T) {\n\ttest.Prepare(t, config.Conf)\n\tdefer test.Clean()\n\n\tmanager := createTestManager(t)\n\tif manager == nil {\n\t\treturn\n\t}\n\tdefer manager.Close()\n\n\topts := &Options{\n\t\tCommand: \"claude\",\n\t\tImage: \"alpine:latest\",\n\t\tUserID: \"test-user\",\n\t\tChatID: fmt.Sprintf(\"test-chat-att-invalid-%d\", time.Now().UnixNano()),\n\t}\n\n\texec, err := NewExecutor(manager, opts)\n\trequire.NoError(t, err)\n\tdefer exec.Close()\n\n\tctx := context.Background()\n\n\t// Message with an attachment URL pointing to a non-existent manager\n\tmessages := []agentContext.Message{\n\t\t{\n\t\t\tRole: \"user\",\n\t\t\tContent: []interface{}{\n\t\t\t\tmap[string]interface{}{\"type\": \"text\", \"text\": \"See this image\"},\n\t\t\t\tmap[string]interface{}{\n\t\t\t\t\t\"type\": \"image_url\",\n\t\t\t\t\t\"image_url\": map[string]interface{}{\n\t\t\t\t\t\t\"url\": \"__nonexistent.uploader://fakefile123\",\n\t\t\t\t\t\t\"detail\": \"auto\",\n\t\t\t\t\t},\n\t\t\t\t},\n\t\t\t},\n\t\t},\n\t}\n\n\tresult, err := exec.prepareAttachments(ctx, messages)\n\trequire.NoError(t, err)\n\trequire.Len(t, result, 1)\n\n\t// Should gracefully fallback to error text\n\tcontent, ok := result[0].Content.(string)\n\trequire.True(t, ok, \"Content should be converted to string\")\n\tassert.Contains(t, content, \"See this image\")\n\tassert.Contains(t, content, \"failed to load\")\n}\n\nfunc TestPrepareAttachmentsMixedRoles(t *testing.T) {\n\ttest.Prepare(t, config.Conf)\n\tdefer test.Clean()\n\n\tmanager := createTestManager(t)\n\tif manager == nil {\n\t\treturn\n\t}\n\tdefer manager.Close()\n\n\topts := &Options{\n\t\tCommand: \"claude\",\n\t\tImage: \"alpine:latest\",\n\t\tUserID: \"test-user\",\n\t\tChatID: fmt.Sprintf(\"test-chat-att-mixed-%d\", time.Now().UnixNano()),\n\t}\n\n\texec, err := NewExecutor(manager, opts)\n\trequire.NoError(t, err)\n\tdefer exec.Close()\n\n\tctx := context.Background()\n\n\t// Only user messages should be processed; system and assistant messages pass through\n\tmessages := []agentContext.Message{\n\t\t{Role: \"system\", Content: \"System prompt\"},\n\t\t{\n\t\t\tRole: \"user\",\n\t\t\tContent: []interface{}{\n\t\t\t\tmap[string]interface{}{\"type\": \"text\", \"text\": \"User message with image\"},\n\t\t\t\tmap[string]interface{}{\n\t\t\t\t\t\"type\": \"image_url\",\n\t\t\t\t\t\"image_url\": map[string]interface{}{\n\t\t\t\t\t\t\"url\": \"https://example.com/photo.jpg\",\n\t\t\t\t\t\t\"detail\": \"auto\",\n\t\t\t\t\t},\n\t\t\t\t},\n\t\t\t},\n\t\t},\n\t\t{Role: \"assistant\", Content: \"I can see the photo\"},\n\t\t{Role: \"user\", Content: \"Thanks!\"},\n\t}\n\n\tresult, err := exec.prepareAttachments(ctx, messages)\n\trequire.NoError(t, err)\n\trequire.Len(t, result, 4)\n\n\t// System and assistant messages unchanged\n\tassert.Equal(t, \"System prompt\", result[0].Content)\n\tassert.Equal(t, \"I can see the photo\", result[2].Content)\n\tassert.Equal(t, \"Thanks!\", result[3].Content)\n\n\t// User multimodal message converted\n\tcontent, ok := result[1].Content.(string)\n\trequire.True(t, ok, \"User multimodal content should be converted to string\")\n\tassert.Contains(t, content, \"User message with image\")\n\tassert.Contains(t, content, \"[Image: https://example.com/photo.jpg]\")\n}\n" }, { "path": "agent/sandbox/claude/command.go", "content": "package claude\n\nimport (\n\t\"encoding/json\"\n\t\"fmt\"\n\t\"strings\"\n\n\t\"github.com/yaoapp/gou/connector\"\n\tagentContext \"github.com/yaoapp/yao/agent/context\"\n)\n\n// sandboxEnvPrompt is the system prompt injected for sandbox environment\n// This tells Claude CLI about the workspace and project structure\nconst sandboxEnvPrompt = `## Sandbox Environment\n\nYou are running in a sandboxed environment with the following setup:\n\n- **Working Directory**: /workspace\n- **Project Structure**: If this is a new project, create a dedicated project folder (e.g., /workspace/my-project/) and work inside it\n- **File Access**: You have full read/write access to /workspace\n- **Output Files**: Save all output files to the working directory\n\nWhen creating new projects:\n1. Create a project directory with a descriptive name\n2. Initialize the project structure inside that directory\n3. Keep all related files organized within the project folder\n\n## IMPORTANT: Restricted Tools\n\nThe following tools are NOT available in this environment and you must NOT use them:\n- EnterPlanMode, ExitPlanMode (use regular text to explain plans instead)\n- Task, TaskOutput, TaskStop (complete tasks directly without delegation)\n- AskUserQuestion (make reasonable assumptions instead of asking)\n- Skill, ToolSearch (not supported)\n\nFocus on using the core tools: Bash, Read, Write, Edit, Glob, Grep, WebSearch, WebFetch.\n\n## User Attachments\n\nUser-uploaded files (images, documents, code files, etc.) are placed in /workspace/.attachments/\nWhen the user references an attached file, read it from this directory using the Read or Bash tool.\nFor image files, you can view them directly as Claude supports vision on local files.\n\n## GitHub CLI (gh) Usage\n\nWhen working with GitHub and a token is provided:\n1. First authenticate gh CLI using the token: echo \"TOKEN\" | gh auth login --with-token\n2. Then use gh commands normally (gh repo create, gh pr create, etc.)\n3. Do NOT use curl to call GitHub API directly - always prefer gh CLI\n`\n\n// claudeArgWhitelist maps package.yao sandbox.arguments keys to Claude CLI flags.\n// Only keys listed here are passed through; everything else is ignored.\nvar claudeArgWhitelist = map[string]string{\n\t\"max_turns\": \"--max-turns\", // Maximum conversation turns\n\t\"disallowed_tools\": \"--disallowed-tools\", // Comma-separated tool blacklist (e.g. \"WebSearch,WebFetch\")\n\t\"allowed_tools\": \"--allowedTools\", // Comma-separated tool whitelist (e.g. \"Bash,Read,Write\")\n}\n\n// BuildCommand builds the Claude CLI command and environment variables\n// Uses stdin with --input-format stream-json for unlimited prompt length\n// isContinuation: if true, uses --continue to resume previous session (only sends last user message)\nfunc BuildCommand(messages []agentContext.Message, opts *Options) ([]string, map[string]string, error) {\n\treturn BuildCommandWithContinuation(messages, opts, false)\n}\n\n// BuildCommandWithContinuation builds the Claude CLI command with continuation support\n// isContinuation: if true, uses --continue to resume previous session\nfunc BuildCommandWithContinuation(messages []agentContext.Message, opts *Options, isContinuation bool) ([]string, map[string]string, error) {\n\t// Build system prompt from conversation history (only for first request)\n\tvar systemPrompt string\n\tif !isContinuation {\n\t\tsystemPrompt, _ = buildPrompts(messages)\n\t\t// Inject sandbox environment prompt\n\t\tif systemPrompt != \"\" {\n\t\t\tsystemPrompt = systemPrompt + \"\\n\\n\" + sandboxEnvPrompt\n\t\t} else {\n\t\t\tsystemPrompt = sandboxEnvPrompt\n\t\t}\n\t}\n\n\t// Build input JSONL for Claude CLI (stream-json format)\n\t// For continuation, only send the last user message\n\tvar inputJSONL []byte\n\tvar err error\n\tif isContinuation {\n\t\tinputJSONL, err = BuildLastUserMessageJSONL(messages)\n\t} else {\n\t\tinputJSONL, err = BuildFirstRequestJSONL(messages)\n\t}\n\tif err != nil {\n\t\treturn nil, nil, fmt.Errorf(\"failed to build input JSONL: %w\", err)\n\t}\n\n\t// Build Claude CLI arguments\n\tvar claudeArgs []string\n\n\t// Add permission mode (required for MCP tools to work)\n\tpermMode := \"bypassPermissions\" // default for sandbox\n\tif opts != nil && opts.Arguments != nil {\n\t\tif mode, ok := opts.Arguments[\"permission_mode\"].(string); ok && mode != \"\" {\n\t\t\tpermMode = mode\n\t\t}\n\t}\n\tclaudeArgs = append(claudeArgs, \"--dangerously-skip-permissions\")\n\tclaudeArgs = append(claudeArgs, \"--permission-mode\", permMode)\n\n\t// Add streaming format flags (required for proper streaming output)\n\tclaudeArgs = append(claudeArgs, \"--input-format\", \"stream-json\")\n\tclaudeArgs = append(claudeArgs, \"--output-format\", \"stream-json\")\n\tclaudeArgs = append(claudeArgs, \"--include-partial-messages\") // Enable realtime streaming\n\tclaudeArgs = append(claudeArgs, \"--verbose\")\n\n\t// For continuation, use --continue to resume the previous session\n\t// Claude CLI will read session data from $HOME/.claude/ (which is /workspace/.claude/)\n\tif isContinuation {\n\t\tclaudeArgs = append(claudeArgs, \"--continue\")\n\t}\n\n\t// Pass through whitelisted arguments to Claude CLI flags.\n\t// Map: package.yao arguments key → Claude CLI flag\n\tif opts != nil && opts.Arguments != nil {\n\t\tfor key, flag := range claudeArgWhitelist {\n\t\t\tif val, ok := opts.Arguments[key]; ok {\n\t\t\t\tclaudeArgs = append(claudeArgs, flag, fmt.Sprintf(\"%v\", val))\n\t\t\t}\n\t\t}\n\t}\n\n\t// Add MCP config if available\n\tif opts != nil && len(opts.MCPConfig) > 0 {\n\t\tclaudeArgs = append(claudeArgs, \"--mcp-config\", \"/workspace/.mcp.json\")\n\t\t// Allow all tools from the \"yao\" MCP server\n\t\tclaudeArgs = append(claudeArgs, \"--allowedTools\", \"mcp__yao__*\")\n\t}\n\n\t// Build the full bash command\n\t// Use heredoc for both system prompt and input JSONL to avoid shell escaping issues\n\t// System prompt may contain quotes, newlines, special characters that break shell quoting\n\tvar bashCmd strings.Builder\n\n\t// Ensure $HOME/.Xauthority exists for PyAutoGUI/Xlib (HOME=/workspace).\n\t// Xvfb runs without auth, but Xlib requires the file to exist.\n\tbashCmd.WriteString(\"touch /home/sandbox/.Xauthority 2>/dev/null; touch \\\"$HOME/.Xauthority\\\" 2>/dev/null\\n\")\n\n\t// If we have a system prompt (first request only), write it to a temp file via heredoc first\n\t// then use --append-system-prompt-file\n\tif systemPrompt != \"\" {\n\t\tbashCmd.WriteString(\"cat << 'PROMPTEOF' > /tmp/.system-prompt.txt\\n\")\n\t\tbashCmd.WriteString(systemPrompt)\n\t\tbashCmd.WriteString(\"\\nPROMPTEOF\\n\")\n\t\tclaudeArgs = append(claudeArgs, \"--append-system-prompt-file\", \"/tmp/.system-prompt.txt\")\n\t}\n\n\t// Build claude command with all arguments\n\t// Append 2>&1 to the claude command so stderr is merged into stdout;\n\t// Docker's stdcopy discards the stderr stream, making errors invisible.\n\tbashCmd.WriteString(\"cat << 'INPUTEOF' | claude -p\")\n\tfor _, arg := range claudeArgs {\n\t\tbashCmd.WriteString(fmt.Sprintf(\" %q\", arg))\n\t}\n\tbashCmd.WriteString(\" 2>&1\")\n\tbashCmd.WriteString(\"\\n\")\n\tbashCmd.WriteString(string(inputJSONL))\n\tbashCmd.WriteString(\"\\nINPUTEOF\")\n\n\tcmd := []string{\"bash\", \"-c\", bashCmd.String()}\n\n\t// Build environment variables\n\tenv := buildEnvironment(opts, systemPrompt)\n\n\treturn cmd, env, nil\n}\n\n// BuildInputJSONL converts messages to Claude CLI stream-json input format\n// Deprecated: Use BuildFirstRequestJSONL or BuildLastUserMessageJSONL instead\nfunc BuildInputJSONL(messages []agentContext.Message) ([]byte, error) {\n\treturn BuildFirstRequestJSONL(messages)\n}\n\n// BuildFirstRequestJSONL builds JSONL for the first request (all messages)\n// Sends all user and assistant messages to establish context\nfunc BuildFirstRequestJSONL(messages []agentContext.Message) ([]byte, error) {\n\tvar lines []string\n\n\tfor _, msg := range messages {\n\t\t// Skip system messages (handled via --system-prompt)\n\t\tif msg.Role == \"system\" {\n\t\t\tcontinue\n\t\t}\n\n\t\t// Build the message content\n\t\tvar content interface{}\n\t\tif msg.Content != nil {\n\t\t\tcontent = msg.Content\n\t\t} else {\n\t\t\tcontent = \"\"\n\t\t}\n\n\t\t// Create stream-json message\n\t\tstreamMsg := map[string]interface{}{\n\t\t\t\"type\": string(msg.Role), // \"user\" or \"assistant\"\n\t\t\t\"message\": map[string]interface{}{\n\t\t\t\t\"role\": string(msg.Role),\n\t\t\t\t\"content\": content,\n\t\t\t},\n\t\t}\n\n\t\tjsonBytes, err := json.Marshal(streamMsg)\n\t\tif err != nil {\n\t\t\treturn nil, fmt.Errorf(\"failed to marshal message: %w\", err)\n\t\t}\n\t\tlines = append(lines, string(jsonBytes))\n\t}\n\n\treturn []byte(strings.Join(lines, \"\\n\")), nil\n}\n\n// BuildLastUserMessageJSONL builds JSONL with only the last user message\n// Used for continuation requests where Claude CLI manages history via --continue\nfunc BuildLastUserMessageJSONL(messages []agentContext.Message) ([]byte, error) {\n\t// Find the last user message\n\tvar lastUserMessage *agentContext.Message\n\tfor i := len(messages) - 1; i >= 0; i-- {\n\t\tif messages[i].Role == \"user\" {\n\t\t\tlastUserMessage = &messages[i]\n\t\t\tbreak\n\t\t}\n\t}\n\n\tif lastUserMessage == nil {\n\t\treturn nil, fmt.Errorf(\"no user message found\")\n\t}\n\n\tvar content interface{}\n\tif lastUserMessage.Content != nil {\n\t\tcontent = lastUserMessage.Content\n\t} else {\n\t\tcontent = \"\"\n\t}\n\n\tuserMsg := map[string]interface{}{\n\t\t\"type\": \"user\",\n\t\t\"message\": map[string]interface{}{\n\t\t\t\"role\": \"user\",\n\t\t\t\"content\": content,\n\t\t},\n\t}\n\n\tjsonBytes, err := json.Marshal(userMsg)\n\tif err != nil {\n\t\treturn nil, fmt.Errorf(\"failed to marshal user message: %w\", err)\n\t}\n\n\treturn jsonBytes, nil\n}\n\n// buildPrompts extracts system prompt and user prompt from messages\nfunc buildPrompts(messages []agentContext.Message) (systemPrompt string, userPrompt string) {\n\tvar systemParts []string\n\tvar conversationParts []string\n\tvar lastUserMessage string\n\n\tfor _, msg := range messages {\n\t\tswitch msg.Role {\n\t\tcase \"system\":\n\t\t\tsystemParts = append(systemParts, getMessageContent(msg))\n\t\tcase \"user\":\n\t\t\tlastUserMessage = getMessageContent(msg)\n\t\t\tconversationParts = append(conversationParts, fmt.Sprintf(\"User: %s\", lastUserMessage))\n\t\tcase \"assistant\":\n\t\t\tconversationParts = append(conversationParts, fmt.Sprintf(\"Assistant: %s\", getMessageContent(msg)))\n\t\t}\n\t}\n\n\t// Build system prompt with conversation history\n\tsystemPrompt = strings.Join(systemParts, \"\\n\\n\")\n\n\t// If there's conversation history, include it in the system prompt\n\tif len(conversationParts) > 1 {\n\t\thistorySection := \"\\n\\n## Conversation History\\n\\n\" + strings.Join(conversationParts[:len(conversationParts)-1], \"\\n\\n\")\n\t\tsystemPrompt += historySection\n\t}\n\n\t// The user prompt is the last user message\n\tuserPrompt = lastUserMessage\n\n\treturn systemPrompt, userPrompt\n}\n\n// getMessageContent extracts text content from a message\nfunc getMessageContent(msg agentContext.Message) string {\n\tif msg.Content == nil {\n\t\treturn \"\"\n\t}\n\n\t// Handle string content\n\tif str, ok := msg.Content.(string); ok {\n\t\treturn str\n\t}\n\n\t// Handle content array (multimodal messages)\n\tif arr, ok := msg.Content.([]interface{}); ok {\n\t\tvar parts []string\n\t\tfor _, item := range arr {\n\t\t\tif m, ok := item.(map[string]interface{}); ok {\n\t\t\t\tif m[\"type\"] == \"text\" {\n\t\t\t\t\tif text, ok := m[\"text\"].(string); ok {\n\t\t\t\t\t\tparts = append(parts, text)\n\t\t\t\t\t}\n\t\t\t\t}\n\t\t\t}\n\t\t}\n\t\treturn strings.Join(parts, \"\\n\")\n\t}\n\n\treturn \"\"\n}\n\n// buildEnvironment builds environment variables for Claude CLI\nfunc buildEnvironment(opts *Options, systemPrompt string) map[string]string {\n\tenv := make(map[string]string)\n\n\tif opts == nil {\n\t\treturn env\n\t}\n\n\t// Set HOME to /workspace so Claude CLI stores session data in the workspace\n\t// This allows session persistence across requests for the same chat\n\t// Session data is stored in $HOME/.claude/ (i.e., /workspace/.claude/)\n\tenv[\"HOME\"] = \"/workspace\"\n\n\t// Fix Python user-site-packages: changing HOME from /home/sandbox to /workspace\n\t// breaks Python's ability to find packages installed via pip --user (e.g., playwright,\n\t// pyautogui, playwright-stealth) which live in /home/sandbox/.local/lib/pythonX.Y/site-packages/\n\tenv[\"PYTHONPATH\"] = \"/home/sandbox/.local/lib/python3.12/site-packages\"\n\n\t// Fix X11 auth: PyAutoGUI/Xlib looks for $HOME/.Xauthority, but HOME=/workspace\n\t// so it fails to find /home/sandbox/.Xauthority created during image build.\n\t// Explicitly set XAUTHORITY to the correct path.\n\tenv[\"XAUTHORITY\"] = \"/home/sandbox/.Xauthority\"\n\n\tif opts.ConnectorType == \"anthropic\" {\n\t\t// Anthropic mode: Claude CLI connects directly to the Anthropic-compatible backend\n\t\t// No proxy needed — the backend already speaks Anthropic Messages API\n\t\tenv[\"ANTHROPIC_BASE_URL\"] = opts.ConnectorHost\n\t\tenv[\"ANTHROPIC_API_KEY\"] = opts.ConnectorKey\n\t} else {\n\t\t// OpenAI mode (default): Claude CLI connects to claude-proxy on localhost:3456\n\t\t// The proxy translates Anthropic Messages API → OpenAI Chat Completions API\n\t\tenv[\"ANTHROPIC_BASE_URL\"] = \"http://127.0.0.1:3456\"\n\t\tenv[\"ANTHROPIC_API_KEY\"] = \"dummy\" // Proxy doesn't verify this\n\t}\n\n\t// Set model environment variables from connector\n\t// Claude CLI uses these to select the model for all roles\n\tif opts.Model != \"\" {\n\t\tenv[\"ANTHROPIC_MODEL\"] = opts.Model\n\t\tenv[\"ANTHROPIC_DEFAULT_OPUS_MODEL\"] = opts.Model\n\t\tenv[\"ANTHROPIC_DEFAULT_SONNET_MODEL\"] = opts.Model\n\t\tenv[\"ANTHROPIC_DEFAULT_HAIKU_MODEL\"] = opts.Model\n\t\tenv[\"CLAUDE_CODE_SUBAGENT_MODEL\"] = opts.Model\n\t}\n\n\t// Pass secrets as environment variables for Claude CLI to use\n\t// These are configured in package.yao sandbox.secrets (e.g., LLM_API_KEY, GITHUB_TOKEN)\n\t// start-claude-proxy also exports them for the proxy process, but Claude CLI\n\t// is launched via a separate docker exec, so it needs them passed explicitly here.\n\tif len(opts.Secrets) > 0 {\n\t\tfor k, v := range opts.Secrets {\n\t\t\tenv[k] = v\n\t\t}\n\t}\n\n\t// Note: System prompt and max_turns are passed via CLI flags in BuildCommand\n\t// CLAUDE_SYSTEM_PROMPT environment variable is NOT supported by Claude CLI\n\t// --append-system-prompt or --system-prompt flags must be used instead\n\n\treturn env\n}\n\n// BuildProxyConfig builds the claude-proxy configuration JSON\n// This config file is read by start-claude-proxy script in the container\n// Config is written to /tmp/.yao/proxy.json (not /workspace/) for security\nfunc BuildProxyConfig(opts *Options) ([]byte, error) {\n\tif opts == nil {\n\t\treturn nil, fmt.Errorf(\"options is required\")\n\t}\n\n\t// Build backend URL using the shared connector.BuildAPIURL helper\n\t// so that the /v1 prefix is applied consistently with the agent LLM path.\n\tbackendURL := connector.BuildAPIURL(opts.ConnectorHost, \"/chat/completions\")\n\n\tconfig := map[string]interface{}{\n\t\t\"backend\": backendURL,\n\t\t\"api_key\": opts.ConnectorKey,\n\t\t\"model\": opts.Model,\n\t}\n\n\t// Add extra connector options if present (e.g., thinking, max_tokens, temperature)\n\t// These will be passed to the proxy via CLAUDE_PROXY_OPTIONS environment variable\n\tif len(opts.ConnectorOptions) > 0 {\n\t\tconfig[\"options\"] = opts.ConnectorOptions\n\t}\n\n\t// Add secrets if present (e.g., GITHUB_TOKEN, AWS_ACCESS_KEY)\n\t// These will be exported as environment variables for Claude CLI to use\n\tif len(opts.Secrets) > 0 {\n\t\tconfig[\"secrets\"] = opts.Secrets\n\t}\n\n\treturn json.MarshalIndent(config, \"\", \" \")\n}\n\n// BuildCCRConfig is deprecated, kept for backward compatibility\n// Use BuildProxyConfig instead\nfunc BuildCCRConfig(opts *Options) ([]byte, error) {\n\treturn BuildProxyConfig(opts)\n}\n" }, { "path": "agent/sandbox/claude/command_test.go", "content": "package claude\n\nimport (\n\t\"encoding/json\"\n\t\"testing\"\n\n\t\"github.com/stretchr/testify/assert\"\n\t\"github.com/stretchr/testify/require\"\n\tagentContext \"github.com/yaoapp/yao/agent/context\"\n)\n\nfunc TestBuildCommand(t *testing.T) {\n\tmessages := []agentContext.Message{\n\t\t{Role: \"system\", Content: \"You are a helpful assistant\"},\n\t\t{Role: \"user\", Content: \"Hello\"},\n\t}\n\n\topts := &Options{\n\t\tConnectorHost: \"https://api.example.com\",\n\t\tConnectorKey: \"key123\",\n\t\tModel: \"test-model\",\n\t}\n\n\tcmd, env, err := BuildCommand(messages, opts)\n\trequire.NoError(t, err)\n\n\t// Verify command structure\n\t// Command is now: [\"bash\", \"-c\", \"cat << 'INPUTEOF' | claude -p ... INPUTEOF\"]\n\tassert.Equal(t, \"bash\", cmd[0])\n\tassert.Equal(t, \"-c\", cmd[1])\n\t// User message should be in bash command (as JSONL via stdin)\n\tassert.Contains(t, cmd[2], \"Hello\")\n\t// Should have stream-json flags\n\tassert.Contains(t, cmd[2], \"--input-format\")\n\tassert.Contains(t, cmd[2], \"--output-format\")\n\tassert.Contains(t, cmd[2], \"--include-partial-messages\")\n\tassert.Contains(t, cmd[2], \"--verbose\")\n\tassert.Contains(t, cmd[2], \"stream-json\")\n\n\t// Verify environment variables (claude-proxy)\n\tassert.Equal(t, \"http://127.0.0.1:3456\", env[\"ANTHROPIC_BASE_URL\"])\n\tassert.Equal(t, \"dummy\", env[\"ANTHROPIC_API_KEY\"])\n}\n\nfunc TestBuildCommandWithSystemPrompt(t *testing.T) {\n\tmessages := []agentContext.Message{\n\t\t{Role: \"system\", Content: \"You are a code reviewer\"},\n\t\t{Role: \"user\", Content: \"Review this code\"},\n\t\t{Role: \"assistant\", Content: \"Sure, I'll review it\"},\n\t\t{Role: \"user\", Content: \"Here is the code\"},\n\t}\n\n\topts := &Options{}\n\n\tcmd, _, err := BuildCommand(messages, opts)\n\trequire.NoError(t, err)\n\n\t// System prompt should be written to file via heredoc, then passed via --append-system-prompt-file\n\tbashCmd := cmd[2] // The bash -c command string\n\tassert.Contains(t, bashCmd, \"cat << 'PROMPTEOF' > /tmp/.system-prompt.txt\")\n\tassert.Contains(t, bashCmd, \"You are a code reviewer\")\n\tassert.Contains(t, bashCmd, \"PROMPTEOF\")\n\tassert.Contains(t, bashCmd, \"--append-system-prompt-file\")\n\tassert.Contains(t, bashCmd, \"/tmp/.system-prompt.txt\")\n}\n\nfunc TestBuildCommandWithSpecialCharsInPrompt(t *testing.T) {\n\t// Test that special characters in prompts are handled correctly\n\tmessages := []agentContext.Message{\n\t\t{Role: \"system\", Content: \"You are a helper.\\n\\n## Rules\\n- Rule 1: Don't use \\\"quotes\\\" wrongly\\n- Rule 2: Handle 'single quotes' too\\n- Rule 3: Special chars like $VAR and `backticks`\"},\n\t\t{Role: \"user\", Content: \"Hello\"},\n\t}\n\n\topts := &Options{}\n\n\tcmd, _, err := BuildCommand(messages, opts)\n\trequire.NoError(t, err)\n\n\tbashCmd := cmd[2]\n\t// The heredoc approach should preserve all special characters\n\tassert.Contains(t, bashCmd, \"## Rules\")\n\tassert.Contains(t, bashCmd, `Don't use \"quotes\" wrongly`)\n\tassert.Contains(t, bashCmd, \"'single quotes'\")\n}\n\nfunc TestBuildCommandWithArguments(t *testing.T) {\n\tmessages := []agentContext.Message{\n\t\t{Role: \"user\", Content: \"Hello\"},\n\t}\n\n\topts := &Options{\n\t\tArguments: map[string]interface{}{\n\t\t\t\"max_turns\": 20,\n\t\t\t\"permission_mode\": \"acceptEdits\",\n\t\t},\n\t}\n\n\tcmd, _, err := BuildCommand(messages, opts)\n\trequire.NoError(t, err)\n\n\tbashCmd := cmd[2] // The bash -c command string\n\t// max_turns should be in command args via --max-turns\n\tassert.Contains(t, bashCmd, \"--max-turns\")\n\tassert.Contains(t, bashCmd, \"20\")\n\t// permission_mode should be in command args\n\tassert.Contains(t, bashCmd, \"acceptEdits\")\n}\n\nfunc TestBuildProxyConfig(t *testing.T) {\n\topts := &Options{\n\t\tConnectorHost: \"https://api.example.com\",\n\t\tConnectorKey: \"key123\",\n\t\tModel: \"test-model\",\n\t}\n\n\tconfigJSON, err := BuildProxyConfig(opts)\n\trequire.NoError(t, err)\n\n\tconfigStr := string(configJSON)\n\t// Proxy config uses simple format\n\t// BuildAPIURL adds /v1 prefix for hosts that don't end with \"/\"\n\tassert.Contains(t, configStr, \"backend\")\n\tassert.Contains(t, configStr, \"https://api.example.com/v1/chat/completions\")\n\tassert.Contains(t, configStr, \"api_key\")\n\tassert.Contains(t, configStr, \"key123\")\n\tassert.Contains(t, configStr, \"model\")\n\tassert.Contains(t, configStr, \"test-model\")\n}\n\nfunc TestBuildProxyConfigVolcengine(t *testing.T) {\n\topts := &Options{\n\t\tConnectorHost: \"https://ark.cn-beijing.volces.com/api/v3/\",\n\t\tConnectorKey: \"test-key\",\n\t\tModel: \"ep-xxx\",\n\t}\n\n\tconfigJSON, err := BuildProxyConfig(opts)\n\trequire.NoError(t, err)\n\n\tconfigStr := string(configJSON)\n\t// URL should end with /chat/completions\n\tassert.Contains(t, configStr, \"/chat/completions\")\n\tassert.Contains(t, configStr, \"ep-xxx\")\n}\n\nfunc TestBuildInputJSONL(t *testing.T) {\n\tmessages := []agentContext.Message{\n\t\t{Role: \"system\", Content: \"You are helpful\"},\n\t\t{Role: \"user\", Content: \"Hello\"},\n\t\t{Role: \"assistant\", Content: \"Hi there!\"},\n\t\t{Role: \"user\", Content: \"How are you?\"},\n\t}\n\n\tjsonl, err := BuildInputJSONL(messages)\n\trequire.NoError(t, err)\n\n\t// Should not contain system messages (handled separately)\n\tassert.NotContains(t, string(jsonl), \"You are helpful\")\n\n\t// Should contain user and assistant messages\n\tassert.Contains(t, string(jsonl), \"Hello\")\n\tassert.Contains(t, string(jsonl), \"Hi there!\")\n\tassert.Contains(t, string(jsonl), \"How are you?\")\n\n\t// Verify JSONL format (each line is valid JSON)\n\tlines := splitLines(string(jsonl))\n\tfor _, line := range lines {\n\t\tif line == \"\" {\n\t\t\tcontinue\n\t\t}\n\t\tvar msg map[string]interface{}\n\t\terr := json.Unmarshal([]byte(line), &msg)\n\t\tassert.NoError(t, err, \"Line should be valid JSON: %s\", line)\n\t\tassert.Contains(t, msg, \"type\")\n\t\tassert.Contains(t, msg, \"message\")\n\t}\n}\n\nfunc TestBuildInputJSONLMultimodal(t *testing.T) {\n\t// Test with multimodal content (image)\n\tmessages := []agentContext.Message{\n\t\t{\n\t\t\tRole: \"user\",\n\t\t\tContent: []interface{}{\n\t\t\t\tmap[string]interface{}{\"type\": \"text\", \"text\": \"What's in this image?\"},\n\t\t\t\tmap[string]interface{}{\n\t\t\t\t\t\"type\": \"image\",\n\t\t\t\t\t\"source\": map[string]interface{}{\n\t\t\t\t\t\t\"type\": \"base64\",\n\t\t\t\t\t\t\"media_type\": \"image/png\",\n\t\t\t\t\t\t\"data\": \"iVBORw0KGgo=\",\n\t\t\t\t\t},\n\t\t\t\t},\n\t\t\t},\n\t\t},\n\t}\n\n\tjsonl, err := BuildInputJSONL(messages)\n\trequire.NoError(t, err)\n\n\t// Should contain the multimodal content\n\tassert.Contains(t, string(jsonl), \"What's in this image?\")\n\tassert.Contains(t, string(jsonl), \"image\")\n\tassert.Contains(t, string(jsonl), \"base64\")\n}\n\nfunc TestGetMessageContent(t *testing.T) {\n\t// String content\n\tmsg1 := agentContext.Message{Content: \"Hello World\"}\n\tassert.Equal(t, \"Hello World\", getMessageContent(msg1))\n\n\t// Nil content\n\tmsg2 := agentContext.Message{Content: nil}\n\tassert.Equal(t, \"\", getMessageContent(msg2))\n\n\t// Array content (multimodal)\n\tmsg3 := agentContext.Message{\n\t\tContent: []interface{}{\n\t\t\tmap[string]interface{}{\"type\": \"text\", \"text\": \"Part 1\"},\n\t\t\tmap[string]interface{}{\"type\": \"text\", \"text\": \"Part 2\"},\n\t\t},\n\t}\n\tassert.Contains(t, getMessageContent(msg3), \"Part 1\")\n\tassert.Contains(t, getMessageContent(msg3), \"Part 2\")\n}\n\n// Helper to split lines\nfunc splitLines(s string) []string {\n\tvar lines []string\n\tstart := 0\n\tfor i := 0; i < len(s); i++ {\n\t\tif s[i] == '\\n' {\n\t\t\tlines = append(lines, s[start:i])\n\t\t\tstart = i + 1\n\t\t}\n\t}\n\tif start < len(s) {\n\t\tlines = append(lines, s[start:])\n\t}\n\treturn lines\n}\n" }, { "path": "agent/sandbox/claude/e2e_test.go", "content": "package claude\n\nimport (\n\t\"context\"\n\t\"fmt\"\n\t\"os\"\n\t\"strings\"\n\t\"testing\"\n\t\"time\"\n\n\t\"github.com/stretchr/testify/assert\"\n\t\"github.com/stretchr/testify/require\"\n\tagentContext \"github.com/yaoapp/yao/agent/context\"\n\t\"github.com/yaoapp/yao/agent/output/message\"\n\t\"github.com/yaoapp/yao/config\"\n\tinfraSandbox \"github.com/yaoapp/yao/sandbox\"\n\t\"github.com/yaoapp/yao/test\"\n)\n\n// TestE2ESkipClaudeCLI verifies that Claude CLI is skipped when no prompts/skills/mcp\n// This is the \"hook-only\" mode where hooks take full control\nfunc TestE2ESkipClaudeCLI(t *testing.T) {\n\tif testing.Short() {\n\t\tt.Skip(\"Skipping E2E test in short mode\")\n\t}\n\n\ttest.Prepare(t, config.Conf)\n\tdefer test.Clean()\n\n\tmanager := createTestManager(t)\n\tif manager == nil {\n\t\treturn\n\t}\n\tdefer manager.Close()\n\n\t// Create options WITHOUT SystemPrompt, SkillsDir, or MCPConfig\n\t// This should trigger the skip logic\n\topts := &Options{\n\t\tCommand: \"claude\",\n\t\tImage: \"alpine:latest\", // Use alpine since we're not calling Claude CLI\n\t\tUserID: \"test-user\",\n\t\tChatID: fmt.Sprintf(\"test-e2e-skip-%d\", time.Now().UnixNano()),\n\t\tConnectorHost: \"\",\n\t\tConnectorKey: \"\",\n\t\tModel: \"\",\n\t\t// No SystemPrompt, SkillsDir, or MCPConfig\n\t}\n\n\texec, err := NewExecutor(manager, opts)\n\trequire.NoError(t, err)\n\tdefer exec.Close()\n\n\t// Verify shouldSkipClaudeCLI returns true\n\tassert.True(t, exec.shouldSkipClaudeCLI(), \"Should skip Claude CLI when no prompts/skills/mcp\")\n\n\t// Execute Stream - it should return immediately without calling Claude CLI\n\tctx := agentContext.New(context.Background(), nil, opts.ChatID)\n\tmessages := []agentContext.Message{\n\t\t{Role: \"user\", Content: \"Hello\"},\n\t}\n\n\tresponse, err := exec.Stream(ctx, messages, nil)\n\trequire.NoError(t, err, \"Stream should succeed\")\n\trequire.NotNil(t, response, \"Response should not be nil\")\n\n\t// Verify response indicates skip\n\tassert.Contains(t, response.ID, \"sandbox-skip\", \"Response ID should indicate skip\")\n\tassert.Equal(t, \"sandbox\", response.Model, \"Model should be 'sandbox' for skip mode\")\n\tassert.Empty(t, response.Content, \"Content should be empty for skip mode\")\n\n\tt.Log(\"✓ Claude CLI skip mode verified\")\n}\n\n// TestE2EExecuteClaudeCLI verifies that Claude CLI is called when prompts are configured\n// This requires the real yaoapp/sandbox-claude image and a valid connector\nfunc TestE2EExecuteClaudeCLI(t *testing.T) {\n\tif testing.Short() {\n\t\tt.Skip(\"Skipping E2E test in short mode\")\n\t}\n\n\t// Check for required environment variables\n\tapiKey := os.Getenv(\"DEEPSEEK_API_KEY\")\n\tapiProxy := os.Getenv(\"DEEPSEEK_API_PROXY\")\n\tmodel := os.Getenv(\"DEEPSEEK_MODELS_V3\")\n\n\tif apiKey == \"\" || apiProxy == \"\" || model == \"\" {\n\t\tt.Skip(\"Skipping test: DEEPSEEK_API_KEY, DEEPSEEK_API_PROXY, or DEEPSEEK_MODELS_V3 not set\")\n\t}\n\n\ttest.Prepare(t, config.Conf)\n\tdefer test.Clean()\n\n\t// Get data root from environment\n\tdataRoot := os.Getenv(\"YAO_ROOT\")\n\tif dataRoot == \"\" {\n\t\tt.Skip(\"Skipping test: YAO_ROOT not set\")\n\t}\n\n\t// Create config with proper paths\n\tcfg := infraSandbox.DefaultConfig()\n\tcfg.Init(dataRoot)\n\n\tmanager, err := infraSandbox.NewManager(cfg)\n\tif err != nil {\n\t\tt.Skipf(\"Skipping test: Docker not available: %v\", err)\n\t}\n\tdefer manager.Close()\n\n\t// Create options WITH SystemPrompt (triggers Claude CLI execution)\n\topts := &Options{\n\t\tCommand: \"claude\",\n\t\tImage: \"yaoapp/sandbox-claude:latest\",\n\t\tUserID: \"test-user\",\n\t\tChatID: fmt.Sprintf(\"test-e2e-exec-%d\", time.Now().UnixNano()),\n\t\tConnectorHost: apiProxy,\n\t\tConnectorKey: apiKey,\n\t\tModel: model,\n\t\tSystemPrompt: \"You are a helpful assistant. Keep responses brief.\",\n\t\tTimeout: 5 * time.Minute,\n\t}\n\n\texec, err := NewExecutor(manager, opts)\n\tif err != nil {\n\t\tt.Skipf(\"Skipping test: Failed to create executor: %v\", err)\n\t}\n\tdefer exec.Close()\n\n\t// Verify shouldSkipClaudeCLI returns false\n\tassert.False(t, exec.shouldSkipClaudeCLI(), \"Should NOT skip Claude CLI when prompts are configured\")\n\n\t// Execute Stream with a simple prompt\n\tctx := agentContext.New(context.Background(), nil, opts.ChatID)\n\tmessages := []agentContext.Message{\n\t\t{Role: \"user\", Content: \"Reply with exactly: TEST_SUCCESS\"},\n\t}\n\n\t// Collect streaming output\n\tvar streamedContent strings.Builder\n\tstreamHandler := func(chunkType message.StreamChunkType, data []byte) int {\n\t\tif chunkType == message.ChunkText {\n\t\t\tstreamedContent.Write(data)\n\t\t}\n\t\treturn 0 // continue streaming\n\t}\n\n\tt.Log(\"Executing Claude CLI with real API call...\")\n\tstartTime := time.Now()\n\tresponse, err := exec.Stream(ctx, messages, streamHandler)\n\tduration := time.Since(startTime)\n\tt.Logf(\"Execution took: %v\", duration)\n\n\tif err != nil {\n\t\tt.Logf(\"Stream error (might be expected if Docker/API issue): %v\", err)\n\t\tt.Skipf(\"Skipping assertion: %v\", err)\n\t}\n\n\trequire.NotNil(t, response, \"Response should not be nil\")\n\n\t// Log response details\n\tt.Logf(\"Response ID: %s\", response.ID)\n\tt.Logf(\"Response Model: %s\", response.Model)\n\tt.Logf(\"Response Content: %v\", response.Content)\n\tt.Logf(\"Streamed Content: %s\", streamedContent.String())\n\n\t// Verify we got some response\n\tvar fullResponse string\n\tif content, ok := response.Content.(string); ok {\n\t\tfullResponse = content\n\t}\n\tif fullResponse == \"\" {\n\t\tfullResponse = streamedContent.String()\n\t}\n\n\tif fullResponse != \"\" {\n\t\tt.Logf(\"✓ Claude CLI executed successfully with response: %s\", truncate(fullResponse, 200))\n\t} else {\n\t\tt.Log(\"⚠ Empty response (Claude CLI might have issues)\")\n\t}\n}\n\n// TestE2EBuildInputJSONLIntegration tests the full flow of building input JSONL\nfunc TestE2EBuildInputJSONLIntegration(t *testing.T) {\n\t// Test with conversation history\n\tmessages := []agentContext.Message{\n\t\t{Role: \"system\", Content: \"You are a helpful assistant\"},\n\t\t{Role: \"user\", Content: \"What is 2+2?\"},\n\t\t{Role: \"assistant\", Content: \"4\"},\n\t\t{Role: \"user\", Content: \"What about 3+3?\"},\n\t}\n\n\tjsonl, err := BuildInputJSONL(messages)\n\trequire.NoError(t, err)\n\n\tt.Logf(\"Input JSONL:\\n%s\", string(jsonl))\n\n\t// Verify format\n\tlines := strings.Split(string(jsonl), \"\\n\")\n\tassert.GreaterOrEqual(t, len(lines), 3, \"Should have at least 3 lines (user, assistant, user)\")\n\n\t// System message should NOT be in JSONL\n\tassert.NotContains(t, string(jsonl), \"You are a helpful assistant\", \"System message should not be in JSONL\")\n\n\t// User and assistant messages should be present\n\tassert.Contains(t, string(jsonl), \"What is 2+2\", \"First user message should be present\")\n\tassert.Contains(t, string(jsonl), \"4\", \"Assistant response should be present\")\n\tassert.Contains(t, string(jsonl), \"What about 3+3\", \"Second user message should be present\")\n\n\tt.Log(\"✓ Input JSONL format verified\")\n}\n\n// TestE2EBuildCommand tests the full command building\nfunc TestE2EBuildCommand(t *testing.T) {\n\tmessages := []agentContext.Message{\n\t\t{Role: \"system\", Content: \"You are helpful\"},\n\t\t{Role: \"user\", Content: \"Hello\"},\n\t}\n\n\topts := &Options{\n\t\tConnectorHost: \"https://api.example.com\",\n\t\tConnectorKey: \"test-key\",\n\t\tModel: \"test-model\",\n\t\tArguments: map[string]interface{}{\n\t\t\t\"permission_mode\": \"bypassPermissions\",\n\t\t},\n\t\tMCPConfig: []byte(`{\"mcpServers\":{}}`),\n\t}\n\n\tcmd, env, err := BuildCommand(messages, opts)\n\trequire.NoError(t, err)\n\n\tt.Logf(\"Command: %v\", cmd)\n\tt.Logf(\"Environment: %v\", env)\n\n\t// Verify command structure\n\tassert.Equal(t, \"bash\", cmd[0])\n\tassert.Equal(t, \"-c\", cmd[1])\n\n\tbashCmd := cmd[2]\n\t// Should use heredoc with INPUTEOF\n\tassert.Contains(t, bashCmd, \"cat << 'INPUTEOF'\", \"Should use heredoc\")\n\tassert.Contains(t, bashCmd, \"INPUTEOF\", \"Should have INPUTEOF delimiter\")\n\n\t// Should have streaming flags\n\tassert.Contains(t, bashCmd, \"--input-format\", \"Should have input-format flag\")\n\tassert.Contains(t, bashCmd, \"--output-format\", \"Should have output-format flag\")\n\tassert.Contains(t, bashCmd, \"--verbose\", \"Should have verbose flag\")\n\tassert.Contains(t, bashCmd, \"stream-json\", \"Should use stream-json format\")\n\n\t// Should have permission flags\n\tassert.Contains(t, bashCmd, \"--dangerously-skip-permissions\", \"Should have skip-permissions flag\")\n\tassert.Contains(t, bashCmd, \"--permission-mode\", \"Should have permission-mode flag\")\n\tassert.Contains(t, bashCmd, \"bypassPermissions\", \"Should have bypassPermissions value\")\n\n\t// Should have MCP config\n\tassert.Contains(t, bashCmd, \"--mcp-config\", \"Should have mcp-config flag\")\n\n\t// Environment should have proxy settings\n\tassert.Equal(t, \"http://127.0.0.1:3456\", env[\"ANTHROPIC_BASE_URL\"])\n\tassert.Equal(t, \"dummy\", env[\"ANTHROPIC_API_KEY\"])\n\n\t// System prompt should be passed via CLI argument, not environment variable\n\t// CLAUDE_SYSTEM_PROMPT env var is NOT supported by Claude CLI\n\tassert.Contains(t, bashCmd, \"--append-system-prompt\", \"Should have append-system-prompt flag\")\n\tassert.Contains(t, bashCmd, \"You are helpful\", \"System prompt should be in CLI args\")\n\n\tt.Log(\"✓ Command building verified\")\n}\n\n// Helper function to truncate strings\nfunc truncate(s string, maxLen int) string {\n\tif len(s) <= maxLen {\n\t\treturn s\n\t}\n\treturn s[:maxLen] + \"...\"\n}\n" }, { "path": "agent/sandbox/claude/executor.go", "content": "package claude\n\nimport (\n\t\"bufio\"\n\t\"context\"\n\t\"encoding/json\"\n\t\"fmt\"\n\t\"io\"\n\t\"log\"\n\t\"os\"\n\t\"path/filepath\"\n\t\"strings\"\n\t\"time\"\n\n\tgoujson \"github.com/yaoapp/gou/json\"\n\tagentContext \"github.com/yaoapp/yao/agent/context\"\n\t\"github.com/yaoapp/yao/agent/i18n\"\n\t\"github.com/yaoapp/yao/agent/output/message\"\n\t\"github.com/yaoapp/yao/attachment\"\n\tinfraSandbox \"github.com/yaoapp/yao/sandbox\"\n\t\"github.com/yaoapp/yao/sandbox/ipc\"\n)\n\n// Options for Claude executor (copied from parent package to avoid import cycle)\ntype Options struct {\n\tCommand string\n\tImage string\n\tMaxMemory string\n\tMaxCPU float64\n\tTimeout time.Duration\n\tArguments map[string]interface{}\n\tUserID string\n\tChatID string\n\tMCPConfig []byte\n\tMCPTools map[string]*ipc.MCPTool // MCP tools to expose via IPC\n\tSkillsDir string\n\tSystemPrompt string // System prompt from assistant prompts.yml\n\tConnectorHost string\n\tConnectorKey string\n\tModel string\n\tConnectorType string // Connector API type: \"openai\" or \"anthropic\"\n\tConnectorOptions map[string]interface{} // Extra connector options (e.g., thinking, max_tokens)\n\tSecrets map[string]string // Secrets to pass to container (e.g., GITHUB_TOKEN)\n}\n\n// Executor implements the sandbox.Executor interface for Claude CLI\ntype Executor struct {\n\tmanager *infraSandbox.Manager\n\tcontainerName string\n\topts *Options\n\tworkDir string\n\tloadingMsgID string // Loading message ID for tool execution updates\n}\n\n// NewExecutor creates a new Claude executor\nfunc NewExecutor(manager *infraSandbox.Manager, opts interface{}) (*Executor, error) {\n\tif manager == nil {\n\t\treturn nil, fmt.Errorf(\"manager is required\")\n\t}\n\n\t// Type assertion to get options\n\tvar execOpts *Options\n\tswitch o := opts.(type) {\n\tcase *Options:\n\t\texecOpts = o\n\tdefault:\n\t\t// Try to convert from map or other struct\n\t\treturn nil, fmt.Errorf(\"invalid options type: %T\", opts)\n\t}\n\n\tif execOpts == nil {\n\t\treturn nil, fmt.Errorf(\"options is required\")\n\t}\n\tif execOpts.UserID == \"\" {\n\t\treturn nil, fmt.Errorf(\"UserID is required\")\n\t}\n\tif execOpts.ChatID == \"\" {\n\t\treturn nil, fmt.Errorf(\"ChatID is required\")\n\t}\n\n\t// Create or get container\n\t// Note: IPC session is created by manager.createContainer, socket is already bind mounted\n\tctx := context.Background()\n\tcreateOpts := infraSandbox.CreateOptions{\n\t\tUserID: execOpts.UserID,\n\t\tChatID: execOpts.ChatID,\n\t\tImage: execOpts.Image,\n\t}\n\tcontainer, err := manager.GetOrCreate(ctx, execOpts.UserID, execOpts.ChatID, createOpts)\n\tif err != nil {\n\t\treturn nil, fmt.Errorf(\"failed to create container: %w\", err)\n\t}\n\n\t// Get workspace directory from config\n\tconfig := manager.GetConfig()\n\tworkDir := config.ContainerWorkDir\n\tif workDir == \"\" {\n\t\tworkDir = \"/workspace\"\n\t}\n\n\treturn &Executor{\n\t\tmanager: manager,\n\t\tcontainerName: container.Name,\n\t\topts: execOpts,\n\t\tworkDir: workDir,\n\t}, nil\n}\n\n// SetLoadingMsgID sets the loading message ID for tool execution updates\nfunc (e *Executor) SetLoadingMsgID(id string) {\n\te.loadingMsgID = id\n}\n\n// Stream runs the Claude CLI with streaming output\nfunc (e *Executor) Stream(ctx *agentContext.Context, messages []agentContext.Message, handler message.StreamFunc) (*agentContext.CompletionResponse, error) {\n\t// Create a cancellable context for this stream operation\n\t// We need to handle both:\n\t// 1. HTTP context cancellation (client disconnect)\n\t// 2. InterruptController cancellation (user clicks \"stop\" button)\n\t//\n\t// Note on InterruptController:\n\t// - ctx.Interrupt.Context() is only cancelled when InterruptForce && len(Messages) == 0\n\t// - When user sends messages with the interrupt, the context is NOT cancelled\n\t// - We use ctx.Interrupt.IsInterrupted() to check for any interrupt signal\n\tstdCtx, cancelFunc := context.WithCancel(context.Background())\n\tdefer cancelFunc()\n\n\t// Start a goroutine to monitor for interrupts and HTTP context cancellation\n\tgo func() {\n\t\tticker := time.NewTicker(500 * time.Millisecond)\n\t\tdefer ticker.Stop()\n\n\t\tfor {\n\t\t\tselect {\n\t\t\tcase <-stdCtx.Done():\n\t\t\t\t// Already cancelled, exit\n\t\t\t\treturn\n\t\t\tcase <-ticker.C:\n\t\t\t\t// Check if there's a pending interrupt signal using Peek()\n\t\t\t\t// This works even when Messages are included (which doesn't cancel the context)\n\t\t\t\tif ctx != nil && ctx.Interrupt != nil {\n\t\t\t\t\tif signal := ctx.Interrupt.Peek(); signal != nil {\n\t\t\t\t\t\tcancelFunc()\n\t\t\t\t\t\treturn\n\t\t\t\t\t}\n\t\t\t\t}\n\t\t\t\t// Check InterruptController.IsInterrupted() (for context-cancelled interrupts)\n\t\t\t\tif ctx != nil && ctx.Interrupt != nil && ctx.Interrupt.IsInterrupted() {\n\t\t\t\t\tcancelFunc()\n\t\t\t\t\treturn\n\t\t\t\t}\n\t\t\t\t// Check HTTP context\n\t\t\t\tif ctx != nil && ctx.Context != nil {\n\t\t\t\t\tselect {\n\t\t\t\t\tcase <-ctx.Context.Done():\n\t\t\t\t\t\tcancelFunc()\n\t\t\t\t\t\treturn\n\t\t\t\t\tdefault:\n\t\t\t\t\t}\n\t\t\t\t}\n\t\t\t}\n\t\t}\n\t}()\n\n\t// Set MCP tools for this request (dynamic, runtime configuration)\n\tif len(e.opts.MCPTools) > 0 {\n\t\tipcManager := e.manager.GetIPCManager()\n\t\tif ipcManager != nil {\n\t\t\tif session, ok := ipcManager.Get(e.opts.ChatID); ok {\n\t\t\t\tsession.SetMCPTools(e.opts.MCPTools)\n\t\t\t}\n\t\t}\n\t}\n\n\t// Prepare environment: write configs and copy skills\n\tif err := e.prepareEnvironment(stdCtx); err != nil {\n\t\treturn nil, fmt.Errorf(\"failed to prepare environment: %w\", err)\n\t}\n\n\t// Resolve attachment URLs and write files to container\n\t// This converts __yao.attachment:// URLs to local file paths in /workspace/.attachments/\n\tif resolved, attErr := e.prepareAttachments(stdCtx, messages); attErr != nil {\n\t\t// Non-fatal: log warning and continue with original messages\n\t\tlog.Printf(\"[sandbox] Warning: failed to prepare attachments: %v\", attErr)\n\t} else {\n\t\tmessages = resolved\n\t}\n\n\t// Check if we should skip Claude CLI execution\n\t// Skip if no prompts, no skills, and no MCP config\n\tskipCLI := e.shouldSkipClaudeCLI()\n\tif skipCLI {\n\t\t// Return empty response - hooks can use sandbox API to do their work\n\t\treturn &agentContext.CompletionResponse{\n\t\t\tID: fmt.Sprintf(\"sandbox-skip-%d\", time.Now().UnixNano()),\n\t\t\tModel: \"sandbox\",\n\t\t\tCreated: time.Now().Unix(),\n\t\t\tRole: \"assistant\",\n\t\t\tContent: \"\",\n\t\t\tFinishReason: agentContext.FinishReasonStop,\n\t\t}, nil\n\t}\n\n\t// Check if this is a continuation (Claude CLI session exists in workspace)\n\tisContinuation := e.hasExistingSession(stdCtx)\n\n\t// Build Claude CLI command using stored options\n\tcmd, env, err := BuildCommandWithContinuation(messages, e.opts, isContinuation)\n\tif err != nil {\n\t\treturn nil, fmt.Errorf(\"failed to build command: %w\", err)\n\t}\n\n\t// Prepare execution options\n\texecOpts := &infraSandbox.ExecOptions{\n\t\tWorkDir: e.workDir,\n\t\tEnv: env,\n\t}\n\n\tif e.opts != nil && e.opts.Timeout > 0 {\n\t\texecOpts.Timeout = e.opts.Timeout\n\t}\n\n\treader, err := e.manager.Stream(stdCtx, e.containerName, cmd, execOpts)\n\tif err != nil {\n\t\treturn nil, fmt.Errorf(\"failed to execute command: %w\", err)\n\t}\n\n\t// Ensure reader is closed when context is cancelled or function returns\n\t// This is important for cleanup when user clicks \"stop\"\n\tdone := make(chan struct{})\n\tdefer func() {\n\t\tclose(done)\n\t\treader.Close()\n\t}()\n\n\t// Monitor for context cancellation and forcefully kill Claude CLI process\n\tgo func() {\n\t\t// Also start a ticker to periodically check context status for debugging\n\t\tticker := time.NewTicker(10 * time.Second)\n\t\tdefer ticker.Stop()\n\n\t\tfor {\n\t\t\tselect {\n\t\t\tcase <-stdCtx.Done():\n\t\t\t\t// First, kill the Claude CLI process inside the container\n\t\t\t\t// This is important because closing the reader/connection alone may not stop the process\n\t\t\t\t// Use a background context since stdCtx is already cancelled\n\t\t\t\tkillCtx, cancel := context.WithTimeout(context.Background(), 5*time.Second)\n\t\t\t\tdefer cancel()\n\n\t\t\t\t// Kill claude process (the Claude CLI binary)\n\t\t\t\te.manager.KillProcess(killCtx, e.containerName, \"claude\")\n\n\t\t\t\t// Also close the reader to unblock any pending reads\n\t\t\t\treader.Close()\n\t\t\t\treturn\n\t\t\tcase <-done:\n\t\t\t\t// Normal completion, nothing to do\n\t\t\t\treturn\n\t\t\tcase <-ticker.C:\n\t\t\t\t// Periodic check - no action needed\n\t\t\t}\n\t\t}\n\t}()\n\n\t// DEBUG: Tee the reader to write raw output to a log file for debugging\n\tdebugLogPath := e.workDir + \"/claude-cli-raw.log\"\n\tdebugReader := e.createDebugReader(stdCtx, reader, debugLogPath)\n\n\t// Parse streaming output (uses e.loadingMsgID set via SetLoadingMsgID)\n\treturn e.parseStream(ctx, debugReader, handler)\n}\n\n// shouldSkipClaudeCLI checks if Claude CLI execution should be skipped\n// Skip when: no system prompt, no skills, and no MCP config\nfunc (e *Executor) shouldSkipClaudeCLI() bool {\n\thasPrompts := e.opts.SystemPrompt != \"\"\n\thasSkills := e.opts.SkillsDir != \"\"\n\thasMCP := len(e.opts.MCPConfig) > 0\n\n\t// If any of these are present, execute Claude CLI\n\treturn !hasPrompts && !hasSkills && !hasMCP\n}\n\n// hasExistingSession checks if Claude CLI has an existing session in the workspace\n// Claude CLI stores session data in $HOME/.claude/projects/ (which is /workspace/.claude/projects/)\n// If session data exists, we should use --continue to resume the session\nfunc (e *Executor) hasExistingSession(ctx context.Context) bool {\n\t// Check if /workspace/.claude/projects/ directory has any content\n\t// This indicates a previous session exists\n\tsessionDir := e.workDir + \"/.claude/projects\"\n\tfiles, err := e.manager.ListDir(ctx, e.containerName, sessionDir)\n\tif err != nil {\n\t\t// Directory doesn't exist or error reading - no existing session\n\t\treturn false\n\t}\n\t// If there are any files/directories in the projects folder, session exists\n\treturn len(files) > 0\n}\n\n// prepareEnvironment prepares the container environment before execution\n// This includes: claude-proxy config, MCP config, and Skills directory\nfunc (e *Executor) prepareEnvironment(ctx context.Context) error {\n\t// 1. Write claude-proxy config and start the proxy\n\tif err := e.startClaudeProxy(ctx); err != nil {\n\t\treturn fmt.Errorf(\"failed to start claude-proxy: %w\", err)\n\t}\n\n\t// 2. Write MCP config if provided\n\tif len(e.opts.MCPConfig) > 0 {\n\t\tif err := e.writeMCPConfig(ctx); err != nil {\n\t\t\treturn fmt.Errorf(\"failed to write MCP config: %w\", err)\n\t\t}\n\t}\n\n\t// 3. Copy Skills directory if provided\n\tif e.opts.SkillsDir != \"\" {\n\t\tif err := e.copySkillsDirectory(ctx); err != nil {\n\t\t\t// Non-fatal: log warning but continue\n\t\t\t// Skills might not exist or be optional\n\t\t\t_ = err // Ignore error, skills are optional\n\t\t}\n\t}\n\n\treturn nil\n}\n\n// startClaudeProxy writes proxy config and starts claude-proxy\nfunc (e *Executor) startClaudeProxy(ctx context.Context) error {\n\t// Skip if no connector configured (e.g., test containers without claude-proxy)\n\tif e.opts.ConnectorHost == \"\" || e.opts.ConnectorKey == \"\" {\n\t\treturn nil\n\t}\n\n\t// Skip proxy for Anthropic connectors — Claude CLI connects directly\n\t// The backend already speaks Anthropic Messages API, no conversion needed\n\tif e.opts.ConnectorType == \"anthropic\" {\n\t\treturn nil\n\t}\n\n\t// Build proxy config\n\tconfigJSON, err := BuildProxyConfig(e.opts)\n\tif err != nil {\n\t\treturn fmt.Errorf(\"failed to build proxy config: %w\", err)\n\t}\n\n\t// Create config directory (outside workspace for security - user can't see api_key/secrets)\n\t// /tmp/.yao/ is not visible to user's file manager\n\tconfigDir := \"/tmp/.yao\"\n\tif _, err := e.manager.Exec(ctx, e.containerName, []string{\"mkdir\", \"-p\", configDir}, nil); err != nil {\n\t\treturn fmt.Errorf(\"failed to create config directory %s: %w\", configDir, err)\n\t}\n\n\t// Write config to secure location (not in /workspace/)\n\tconfigPath := configDir + \"/proxy.json\"\n\tif err := e.manager.WriteFile(ctx, e.containerName, configPath, configJSON); err != nil {\n\t\treturn fmt.Errorf(\"failed to write config to %s: %w\", configPath, err)\n\t}\n\n\t// Start the proxy (only if start-claude-proxy exists in the image)\n\tresult, err := e.manager.Exec(ctx, e.containerName, []string{\"which\", \"start-claude-proxy\"}, &infraSandbox.ExecOptions{\n\t\tWorkDir: e.workDir,\n\t})\n\tif err != nil || result.ExitCode != 0 {\n\t\t// start-claude-proxy not available (e.g., alpine test image), skip\n\t\treturn nil\n\t}\n\n\t// Start the proxy\n\tresult, err = e.manager.Exec(ctx, e.containerName, []string{\"start-claude-proxy\"}, &infraSandbox.ExecOptions{\n\t\tWorkDir: e.workDir,\n\t\tEnv: map[string]string{\n\t\t\t\"WORKSPACE\": e.workDir,\n\t\t},\n\t})\n\tif err != nil {\n\t\treturn fmt.Errorf(\"failed to start claude-proxy: %w\", err)\n\t}\n\tif result.ExitCode != 0 {\n\t\treturn fmt.Errorf(\"claude-proxy failed to start: %s\", result.Stderr)\n\t}\n\n\treturn nil\n}\n\n// writeMCPConfig writes the MCP configuration file to the container workspace\nfunc (e *Executor) writeMCPConfig(ctx context.Context) error {\n\tif len(e.opts.MCPConfig) == 0 {\n\t\treturn nil\n\t}\n\n\t// Write MCP config to workspace (.mcp.json)\n\tmcpPath := e.workDir + \"/.mcp.json\"\n\tif err := e.manager.WriteFile(ctx, e.containerName, mcpPath, e.opts.MCPConfig); err != nil {\n\t\treturn fmt.Errorf(\"failed to write MCP config to %s: %w\", mcpPath, err)\n\t}\n\n\treturn nil\n}\n\n// copySkillsDirectory copies the skills directory to the container\nfunc (e *Executor) copySkillsDirectory(ctx context.Context) error {\n\tif e.opts.SkillsDir == \"\" {\n\t\treturn nil\n\t}\n\n\t// Target path in container: /workspace/.claude/skills/\n\t// This follows Claude CLI's expected skills location\n\tclaudeDir := e.workDir + \"/.claude\"\n\n\t// Create .claude directory first\n\tif _, err := e.manager.Exec(ctx, e.containerName, []string{\"mkdir\", \"-p\", claudeDir}, nil); err != nil {\n\t\treturn fmt.Errorf(\"failed to create .claude directory: %w\", err)\n\t}\n\n\t// Copy skills from host to container\n\t// CopyToContainer extracts tar to containerPath, and createTarFromPath uses\n\t// filepath.Dir(hostPath) as base, so if hostPath is /path/to/skills,\n\t// tar entries are like \"skills/skill-name/SKILL.md\"\n\t// Extracting to /workspace/.claude/ gives us /workspace/.claude/skills/skill-name/SKILL.md\n\tif err := e.manager.CopyToContainer(ctx, e.containerName, e.opts.SkillsDir, claudeDir); err != nil {\n\t\treturn fmt.Errorf(\"failed to copy skills to container: %w\", err)\n\t}\n\n\treturn nil\n}\n\n// prepareAttachments resolves __yao.attachment:// URLs in messages,\n// writes the actual files to the container's /workspace/.attachments/ directory,\n// and replaces the attachment content parts with text references to the file paths.\n// This allows Claude CLI to read the files using its built-in Read/Bash tools.\nfunc (e *Executor) prepareAttachments(ctx context.Context, messages []agentContext.Message) ([]agentContext.Message, error) {\n\t// Track used filenames to handle duplicates\n\tusedNames := make(map[string]int)\n\tattachmentDir := e.workDir + \"/.attachments\"\n\tdirCreated := false\n\thasAttachments := false\n\n\tresult := make([]agentContext.Message, len(messages))\n\tcopy(result, messages)\n\n\tfor i, msg := range result {\n\t\tif msg.Role != \"user\" {\n\t\t\tcontinue\n\t\t}\n\n\t\t// Handle content array (multimodal messages come as []interface{} from JSON)\n\t\tparts, ok := msg.Content.([]interface{})\n\t\tif !ok {\n\t\t\t// Try typed content parts\n\t\t\tif typedParts, ok := msg.Content.([]agentContext.ContentPart); ok {\n\t\t\t\tiparts := make([]interface{}, len(typedParts))\n\t\t\t\tfor j, p := range typedParts {\n\t\t\t\t\t// Convert to map for uniform handling\n\t\t\t\t\tm := map[string]interface{}{\"type\": string(p.Type)}\n\t\t\t\t\tif p.Text != \"\" {\n\t\t\t\t\t\tm[\"text\"] = p.Text\n\t\t\t\t\t}\n\t\t\t\t\tif p.ImageURL != nil {\n\t\t\t\t\t\tm[\"image_url\"] = map[string]interface{}{\n\t\t\t\t\t\t\t\"url\": p.ImageURL.URL,\n\t\t\t\t\t\t\t\"detail\": string(p.ImageURL.Detail),\n\t\t\t\t\t\t}\n\t\t\t\t\t}\n\t\t\t\t\tif p.File != nil {\n\t\t\t\t\t\tm[\"file\"] = map[string]interface{}{\n\t\t\t\t\t\t\t\"url\": p.File.URL,\n\t\t\t\t\t\t\t\"filename\": p.File.Filename,\n\t\t\t\t\t\t}\n\t\t\t\t\t}\n\t\t\t\t\tiparts[j] = m\n\t\t\t\t}\n\t\t\t\tparts = iparts\n\t\t\t} else {\n\t\t\t\tcontinue\n\t\t\t}\n\t\t}\n\n\t\tif len(parts) == 0 {\n\t\t\tcontinue\n\t\t}\n\n\t\t// Process each content part\n\t\tvar textParts []string\n\n\t\tfor _, item := range parts {\n\t\t\tm, ok := item.(map[string]interface{})\n\t\t\tif !ok {\n\t\t\t\tcontinue\n\t\t\t}\n\n\t\t\tpartType, _ := m[\"type\"].(string)\n\n\t\t\tswitch partType {\n\t\t\tcase \"text\":\n\t\t\t\tif text, ok := m[\"text\"].(string); ok && text != \"\" {\n\t\t\t\t\ttextParts = append(textParts, text)\n\t\t\t\t}\n\n\t\t\tcase \"image_url\":\n\t\t\t\timgData, _ := m[\"image_url\"].(map[string]interface{})\n\t\t\t\tif imgData == nil {\n\t\t\t\t\tcontinue\n\t\t\t\t}\n\t\t\t\turl, _ := imgData[\"url\"].(string)\n\t\t\t\tif url == \"\" {\n\t\t\t\t\tcontinue\n\t\t\t\t}\n\n\t\t\t\tuploaderName, fileID, isWrapper := attachment.Parse(url)\n\t\t\t\tif !isWrapper {\n\t\t\t\t\t// Not an attachment URL, keep as text reference\n\t\t\t\t\ttextParts = append(textParts, fmt.Sprintf(\"[Image: %s]\", url))\n\t\t\t\t\tcontinue\n\t\t\t\t}\n\n\t\t\t\t// Resolve the attachment\n\t\t\t\tref, err := e.resolveAttachment(ctx, uploaderName, fileID, \"\", attachmentDir, usedNames, &dirCreated)\n\t\t\t\tif err != nil {\n\t\t\t\t\tlog.Printf(\"[sandbox] Warning: failed to resolve image attachment %s: %v\", fileID, err)\n\t\t\t\t\ttextParts = append(textParts, \"[Attached image: failed to load]\")\n\t\t\t\t\tcontinue\n\t\t\t\t}\n\n\t\t\t\ttextParts = append(textParts, ref)\n\t\t\t\thasAttachments = true\n\n\t\t\tcase \"file\":\n\t\t\t\tfileData, _ := m[\"file\"].(map[string]interface{})\n\t\t\t\tif fileData == nil {\n\t\t\t\t\tcontinue\n\t\t\t\t}\n\t\t\t\turl, _ := fileData[\"url\"].(string)\n\t\t\t\thintName, _ := fileData[\"filename\"].(string)\n\t\t\t\tif url == \"\" {\n\t\t\t\t\tcontinue\n\t\t\t\t}\n\n\t\t\t\tuploaderName, fileID, isWrapper := attachment.Parse(url)\n\t\t\t\tif !isWrapper {\n\t\t\t\t\ttextParts = append(textParts, fmt.Sprintf(\"[File: %s]\", url))\n\t\t\t\t\tcontinue\n\t\t\t\t}\n\n\t\t\t\tref, err := e.resolveAttachment(ctx, uploaderName, fileID, hintName, attachmentDir, usedNames, &dirCreated)\n\t\t\t\tif err != nil {\n\t\t\t\t\tlog.Printf(\"[sandbox] Warning: failed to resolve file attachment %s: %v\", fileID, err)\n\t\t\t\t\ttextParts = append(textParts, \"[Attached file: failed to load]\")\n\t\t\t\t\tcontinue\n\t\t\t\t}\n\n\t\t\t\ttextParts = append(textParts, ref)\n\t\t\t\thasAttachments = true\n\n\t\t\tdefault:\n\t\t\t\t// Keep other types as-is (shouldn't happen normally)\n\t\t\t\tcontinue\n\t\t\t}\n\t\t}\n\n\t\t// Merge text parts into a single string when the original content was\n\t\t// a multimodal array ([]interface{} / []ContentPart). This is needed\n\t\t// even when only \"text\" parts are present so that downstream code\n\t\t// (BuildInputJSONL, etc.) always sees a plain string.\n\t\tif len(textParts) > 0 {\n\t\t\tnewMsg := result[i]\n\t\t\tnewMsg.Content = strings.Join(textParts, \"\\n\\n\")\n\t\t\tresult[i] = newMsg\n\t\t}\n\t}\n\n\tif !hasAttachments {\n\t\treturn result, nil\n\t}\n\n\treturn result, nil\n}\n\n// resolveAttachment reads an attachment from the attachment manager and writes it\n// to the container's .attachments directory. Returns a text reference string.\nfunc (e *Executor) resolveAttachment(\n\tctx context.Context,\n\tuploaderName, fileID, hintName, attachmentDir string,\n\tusedNames map[string]int,\n\tdirCreated *bool,\n) (string, error) {\n\t// Get attachment manager\n\tmanager, exists := attachment.Managers[uploaderName]\n\tif !exists {\n\t\treturn \"\", fmt.Errorf(\"attachment manager not found: %s\", uploaderName)\n\t}\n\n\t// Get file info\n\tfileInfo, err := manager.Info(ctx, fileID)\n\tif err != nil {\n\t\treturn \"\", fmt.Errorf(\"failed to get file info: %w\", err)\n\t}\n\n\t// Read file data\n\tdata, err := manager.Read(ctx, fileID)\n\tif err != nil {\n\t\treturn \"\", fmt.Errorf(\"failed to read file: %w\", err)\n\t}\n\n\t// Determine filename\n\tfilename := fileInfo.Filename\n\tif filename == \"\" && hintName != \"\" {\n\t\tfilename = hintName\n\t}\n\tif filename == \"\" {\n\t\t// Fallback: use fileID with extension from content type\n\t\text := extensionFromContentType(fileInfo.ContentType)\n\t\tfilename = fileID + ext\n\t}\n\n\t// Handle duplicate filenames\n\tbaseName := filename\n\tif count, exists := usedNames[baseName]; exists {\n\t\text := filepath.Ext(filename)\n\t\tname := strings.TrimSuffix(filename, ext)\n\t\tfilename = fmt.Sprintf(\"%s_%d%s\", name, count+1, ext)\n\t\tusedNames[baseName] = count + 1\n\t} else {\n\t\tusedNames[baseName] = 0\n\t}\n\n\t// Create attachments directory if not yet created\n\tif !*dirCreated {\n\t\tif err := e.manager.WriteFile(ctx, e.containerName, attachmentDir+\"/.keep\", []byte(\"\")); err != nil {\n\t\t\treturn \"\", fmt.Errorf(\"failed to create attachments directory: %w\", err)\n\t\t}\n\t\t*dirCreated = true\n\t}\n\n\t// Write file to container\n\tcontainerPath := attachmentDir + \"/\" + filename\n\tif err := e.manager.WriteFile(ctx, e.containerName, containerPath, data); err != nil {\n\t\treturn \"\", fmt.Errorf(\"failed to write file to container: %w\", err)\n\t}\n\n\t// Build human-readable size string\n\tsizeStr := formatFileSize(fileInfo.Bytes)\n\n\t// Return text reference\n\treturn fmt.Sprintf(\"[Attached file: %s (%s, %s)]\", containerPath, fileInfo.ContentType, sizeStr), nil\n}\n\n// extensionFromContentType returns a file extension for a given content type\nfunc extensionFromContentType(contentType string) string {\n\tswitch contentType {\n\tcase \"image/png\":\n\t\treturn \".png\"\n\tcase \"image/jpeg\":\n\t\treturn \".jpg\"\n\tcase \"image/gif\":\n\t\treturn \".gif\"\n\tcase \"image/webp\":\n\t\treturn \".webp\"\n\tcase \"image/svg+xml\":\n\t\treturn \".svg\"\n\tcase \"application/pdf\":\n\t\treturn \".pdf\"\n\tcase \"text/plain\":\n\t\treturn \".txt\"\n\tcase \"text/html\":\n\t\treturn \".html\"\n\tcase \"text/css\":\n\t\treturn \".css\"\n\tcase \"text/javascript\", \"application/javascript\":\n\t\treturn \".js\"\n\tcase \"application/json\":\n\t\treturn \".json\"\n\tcase \"application/zip\":\n\t\treturn \".zip\"\n\tdefault:\n\t\treturn \"\"\n\t}\n}\n\n// formatFileSize returns a human-readable file size string\nfunc formatFileSize(bytes int) string {\n\tif bytes < 1024 {\n\t\treturn fmt.Sprintf(\"%dB\", bytes)\n\t}\n\tif bytes < 1024*1024 {\n\t\treturn fmt.Sprintf(\"%.1fKB\", float64(bytes)/1024)\n\t}\n\treturn fmt.Sprintf(\"%.1fMB\", float64(bytes)/(1024*1024))\n}\n\n// Execute runs the Claude CLI and returns the response\nfunc (e *Executor) Execute(ctx *agentContext.Context, messages []agentContext.Message) (*agentContext.CompletionResponse, error) {\n\treturn e.Stream(ctx, messages, nil)\n}\n\n// debugWriter wraps an io.Reader to write all data to a debug log file\ntype debugWriter struct {\n\treader io.Reader\n\tlogFile *os.File\n\tbuffer []byte\n}\n\nfunc (d *debugWriter) Read(p []byte) (n int, err error) {\n\tn, err = d.reader.Read(p)\n\tif n > 0 && d.logFile != nil {\n\t\t// Write raw bytes to log file\n\t\td.logFile.Write(p[:n])\n\t\td.logFile.Sync()\n\t}\n\treturn n, err\n}\n\nfunc (d *debugWriter) Close() error {\n\tif d.logFile != nil {\n\t\td.logFile.Close()\n\t}\n\treturn nil\n}\n\n// createDebugReader creates a tee reader that writes to a debug log file\n// The log file is written to the container's workspace for inspection\nfunc (e *Executor) createDebugReader(ctx context.Context, reader io.ReadCloser, logPath string) io.Reader {\n\t// Create a local temp file for debug logging\n\t// We write to a local file first, then copy to container when done\n\tlocalLogPath := \"/tmp/claude-cli-debug-\" + e.containerName + \".log\"\n\tlogFile, err := os.Create(localLogPath)\n\tif err != nil {\n\t\treturn reader\n\t}\n\n\t// Write header\n\tlogFile.WriteString(\"=== Claude CLI Raw Output Debug Log ===\\n\")\n\tlogFile.WriteString(fmt.Sprintf(\"Container: %s\\n\", e.containerName))\n\tlogFile.WriteString(fmt.Sprintf(\"Time: %s\\n\", time.Now().Format(time.RFC3339)))\n\tlogFile.WriteString(fmt.Sprintf(\"WorkDir: %s\\n\", e.workDir))\n\tlogFile.WriteString(\"=== BEGIN OUTPUT ===\\n\")\n\tlogFile.Sync()\n\n\treturn &debugWriter{\n\t\treader: reader,\n\t\tlogFile: logFile,\n\t}\n}\n\n// parseStream parses Claude CLI streaming output (stream-json format)\n// Claude CLI output format with --include-partial-messages:\n// - {\"type\":\"system\",\"subtype\":\"init\",...} - initialization\n// - {\"type\":\"stream_event\",\"event\":{\"delta\":{\"type\":\"text_delta\",\"text\":\"...\"}}} - real-time text deltas\n// - {\"type\":\"assistant\",\"message\":{...,\"content\":[{\"type\":\"text\",\"text\":\"...\"}],...}} - complete messages\n// - {\"type\":\"result\",\"subtype\":\"success\",...,\"result\":\"...\"} - final result\nfunc (e *Executor) parseStream(ctx *agentContext.Context, reader io.Reader, handler message.StreamFunc) (*agentContext.CompletionResponse, error) {\n\tscanner := bufio.NewScanner(reader)\n\t// Increase buffer size for potentially large outputs\n\tbuf := make([]byte, 0, 64*1024)\n\tscanner.Buffer(buf, 1024*1024)\n\n\tvar textContent strings.Builder\n\tvar toolCalls []agentContext.ToolCall\n\tvar model string\n\tvar usage *message.UsageInfo\n\tvar finalResult string\n\tmessageStarted := false // Track if we've sent ChunkMessageStart\n\tprepLoadingClosed := false // Track if \"preparing sandbox\" loading has been closed\n\n\t// Tool input accumulation state\n\ttype toolState struct {\n\t\tname string\n\t\tindex int\n\t\tinputJSON strings.Builder\n\t\tloadingID string // Each tool has its own loading message\n\t}\n\tvar currentTool *toolState\n\tvar lastToolLoadingID string // Track the last tool loading ID to close it\n\n\t// Helper function to close \"preparing sandbox\" loading on first output\n\tclosePrepLoading := func() {\n\t\tif !prepLoadingClosed && e.loadingMsgID != \"\" && ctx != nil {\n\t\t\tdoneMsg := &message.Message{\n\t\t\t\tMessageID: e.loadingMsgID,\n\t\t\t\tDelta: true,\n\t\t\t\tDeltaAction: message.DeltaReplace,\n\t\t\t\tType: message.TypeLoading,\n\t\t\t\tProps: map[string]interface{}{\n\t\t\t\t\t\"message\": \"\",\n\t\t\t\t\t\"done\": true,\n\t\t\t\t},\n\t\t\t}\n\t\t\tctx.Send(doneMsg)\n\t\t\tprepLoadingClosed = true\n\t\t}\n\t}\n\n\tlineCount := 0\n\n\t// Get the underlying context for cancellation checks\n\tvar stdCtx context.Context\n\tif ctx != nil && ctx.Context != nil {\n\t\tstdCtx = ctx.Context\n\t} else {\n\t\tstdCtx = context.Background()\n\t}\n\n\tfor scanner.Scan() {\n\t\t// Check for context cancellation on each iteration\n\t\tselect {\n\t\tcase <-stdCtx.Done():\n\t\t\treturn nil, stdCtx.Err()\n\t\tdefault:\n\t\t\t// Continue processing\n\t\t}\n\n\t\tline := scanner.Text()\n\t\tlineCount++\n\t\tif line == \"\" {\n\t\t\tcontinue\n\t\t}\n\n\t\t// Try to parse as JSON (Claude CLI --output-format stream-json)\n\t\tvar msg map[string]interface{}\n\t\tif err := json.Unmarshal([]byte(line), &msg); err != nil {\n\t\t\t// Not JSON, might be plain text output\n\t\t\ttextContent.WriteString(line)\n\t\t\ttextContent.WriteString(\"\\n\")\n\t\t\tcontinue\n\t\t}\n\n\t\tmsgType, _ := msg[\"type\"].(string)\n\n\t\t// Process Claude CLI stream-json message types\n\t\tswitch msgType {\n\t\tcase \"system\":\n\t\t\t// Initialization message - extract model if available\n\t\t\tif m, ok := msg[\"model\"].(string); ok {\n\t\t\t\tmodel = m\n\t\t\t}\n\n\t\tcase \"stream_event\":\n\t\t\t// Real-time streaming event (from --include-partial-messages)\n\t\t\t// Format: {\"type\":\"stream_event\",\"event\":{\"type\":\"content_block_delta\",\"index\":0,\"delta\":{\"type\":\"text_delta\",\"text\":\"...\"}}}\n\t\t\tif event, ok := msg[\"event\"].(map[string]interface{}); ok {\n\t\t\t\teventType, _ := event[\"type\"].(string)\n\n\t\t\t\tswitch eventType {\n\t\t\t\tcase \"content_block_start\":\n\t\t\t\t\t// Handle new content blocks\n\t\t\t\t\t// Format: {\"event\":{\"type\":\"content_block_start\",\"index\":1,\"content_block\":{\"type\":\"tool_use\"|\"text\",...}}}\n\t\t\t\t\tif contentBlock, ok := event[\"content_block\"].(map[string]interface{}); ok {\n\t\t\t\t\t\tblockType, _ := contentBlock[\"type\"].(string)\n\t\t\t\t\t\tswitch blockType {\n\t\t\t\t\t\tcase \"text\":\n\t\t\t\t\t\t\t// New text block starting - add paragraph separator if we already have content\n\t\t\t\t\t\t\t// This ensures proper separation between text blocks across tool-use rounds\n\t\t\t\t\t\t\tif textContent.Len() > 0 {\n\t\t\t\t\t\t\t\ttextContent.WriteString(\"\\n\\n\")\n\t\t\t\t\t\t\t\tif handler != nil && messageStarted {\n\t\t\t\t\t\t\t\t\thandler(message.ChunkText, []byte(\"\\n\\n\"))\n\t\t\t\t\t\t\t\t}\n\t\t\t\t\t\t\t}\n\t\t\t\t\t\tcase \"tool_use\":\n\t\t\t\t\t\t\ttoolName, _ := contentBlock[\"name\"].(string)\n\t\t\t\t\t\t\tblockIndex := 0\n\t\t\t\t\t\t\tif idx, ok := event[\"index\"].(float64); ok {\n\t\t\t\t\t\t\t\tblockIndex = int(idx)\n\t\t\t\t\t\t\t}\n\t\t\t\t\t\t\tif toolName != \"\" && ctx != nil {\n\t\t\t\t\t\t\t\t// Close \"preparing sandbox\" loading on first tool\n\t\t\t\t\t\t\t\tclosePrepLoading()\n\n\t\t\t\t\t\t\t\t// Close previous tool loading if exists\n\t\t\t\t\t\t\t\tif lastToolLoadingID != \"\" {\n\t\t\t\t\t\t\t\t\tdoneMsg := &message.Message{\n\t\t\t\t\t\t\t\t\t\tMessageID: lastToolLoadingID,\n\t\t\t\t\t\t\t\t\t\tDelta: true,\n\t\t\t\t\t\t\t\t\t\tDeltaAction: message.DeltaReplace,\n\t\t\t\t\t\t\t\t\t\tType: message.TypeLoading,\n\t\t\t\t\t\t\t\t\t\tProps: map[string]interface{}{\n\t\t\t\t\t\t\t\t\t\t\t\"message\": \"\",\n\t\t\t\t\t\t\t\t\t\t\t\"done\": true,\n\t\t\t\t\t\t\t\t\t\t},\n\t\t\t\t\t\t\t\t\t}\n\t\t\t\t\t\t\t\t\tctx.Send(doneMsg)\n\t\t\t\t\t\t\t\t}\n\n\t\t\t\t\t\t\t\t// Create new loading message for this tool\n\t\t\t\t\t\t\t\tlocale := ctx.Locale\n\t\t\t\t\t\t\t\ttoolLoadingMsg := &message.Message{\n\t\t\t\t\t\t\t\t\tType: message.TypeLoading,\n\t\t\t\t\t\t\t\t\tProps: map[string]interface{}{\n\t\t\t\t\t\t\t\t\t\t\"message\": getToolDescription(toolName, locale),\n\t\t\t\t\t\t\t\t\t},\n\t\t\t\t\t\t\t\t}\n\t\t\t\t\t\t\t\tnewLoadingID, _ := ctx.SendStream(toolLoadingMsg)\n\n\t\t\t\t\t\t\t\t// Initialize tool state for input accumulation\n\t\t\t\t\t\t\t\tcurrentTool = &toolState{\n\t\t\t\t\t\t\t\t\tname: toolName,\n\t\t\t\t\t\t\t\t\tindex: blockIndex,\n\t\t\t\t\t\t\t\t\tloadingID: newLoadingID,\n\t\t\t\t\t\t\t\t}\n\t\t\t\t\t\t\t\tlastToolLoadingID = newLoadingID\n\n\t\t\t\t\t\t\t\tlog.Printf(\"[Sandbox] Tool started: %s\", toolName)\n\t\t\t\t\t\t\t}\n\t\t\t\t\t\t}\n\t\t\t\t\t}\n\n\t\t\t\tcase \"content_block_delta\":\n\t\t\t\t\tif delta, ok := event[\"delta\"].(map[string]interface{}); ok {\n\t\t\t\t\t\tdeltaType, _ := delta[\"type\"].(string)\n\t\t\t\t\t\tswitch deltaType {\n\t\t\t\t\t\tcase \"text_delta\":\n\t\t\t\t\t\t\tif text, ok := delta[\"text\"].(string); ok && text != \"\" {\n\t\t\t\t\t\t\t\t// Close \"preparing sandbox\" loading on first text output\n\t\t\t\t\t\t\t\tclosePrepLoading()\n\n\t\t\t\t\t\t\t\t// Send to stream handler for real-time output\n\t\t\t\t\t\t\t\tif handler != nil {\n\t\t\t\t\t\t\t\t\t// Send ChunkMessageStart first if not already started\n\t\t\t\t\t\t\t\t\tif !messageStarted {\n\t\t\t\t\t\t\t\t\t\tstartData := message.EventMessageStartData{\n\t\t\t\t\t\t\t\t\t\t\tMessageID: fmt.Sprintf(\"sandbox-%d\", time.Now().UnixNano()),\n\t\t\t\t\t\t\t\t\t\t\tType: \"text\",\n\t\t\t\t\t\t\t\t\t\t\tTimestamp: time.Now().UnixMilli(),\n\t\t\t\t\t\t\t\t\t\t}\n\t\t\t\t\t\t\t\t\t\tstartDataJSON, _ := json.Marshal(startData)\n\t\t\t\t\t\t\t\t\t\thandler(message.ChunkMessageStart, startDataJSON)\n\t\t\t\t\t\t\t\t\t\tmessageStarted = true\n\t\t\t\t\t\t\t\t\t}\n\t\t\t\t\t\t\t\t\thandler(message.ChunkText, []byte(text))\n\t\t\t\t\t\t\t\t}\n\t\t\t\t\t\t\t\t// Also accumulate for final response\n\t\t\t\t\t\t\t\ttextContent.WriteString(text)\n\t\t\t\t\t\t\t}\n\n\t\t\t\t\t\tcase \"input_json_delta\":\n\t\t\t\t\t\t\t// Accumulate tool input JSON fragments\n\t\t\t\t\t\t\tif currentTool != nil {\n\t\t\t\t\t\t\t\tif partialJSON, ok := delta[\"partial_json\"].(string); ok {\n\t\t\t\t\t\t\t\t\tcurrentTool.inputJSON.WriteString(partialJSON)\n\t\t\t\t\t\t\t\t}\n\t\t\t\t\t\t\t}\n\t\t\t\t\t\t}\n\t\t\t\t\t}\n\n\t\t\t\tcase \"content_block_stop\":\n\t\t\t\t\t// Tool input complete - parse and update loading with detailed info\n\t\t\t\t\tif currentTool != nil && currentTool.loadingID != \"\" && ctx != nil {\n\t\t\t\t\t\tinputStr := currentTool.inputJSON.String()\n\t\t\t\t\t\tif inputStr != \"\" {\n\t\t\t\t\t\t\t// Use gou/json.Parse for fault-tolerant parsing\n\t\t\t\t\t\t\tlocale := ctx.Locale\n\t\t\t\t\t\t\tdetailedMsg := getToolDetailedDescription(currentTool.name, inputStr, locale)\n\t\t\t\t\t\t\tif detailedMsg != \"\" {\n\t\t\t\t\t\t\t\ttoolMsg := &message.Message{\n\t\t\t\t\t\t\t\t\tMessageID: currentTool.loadingID,\n\t\t\t\t\t\t\t\t\tDelta: true,\n\t\t\t\t\t\t\t\t\tDeltaAction: message.DeltaReplace,\n\t\t\t\t\t\t\t\t\tType: message.TypeLoading,\n\t\t\t\t\t\t\t\t\tProps: map[string]interface{}{\n\t\t\t\t\t\t\t\t\t\t\"message\": detailedMsg,\n\t\t\t\t\t\t\t\t\t},\n\t\t\t\t\t\t\t\t}\n\t\t\t\t\t\t\t\tctx.Send(toolMsg)\n\t\t\t\t\t\t\t\tlog.Printf(\"[Sandbox] Tool: %s -> %s\", currentTool.name, detailedMsg)\n\t\t\t\t\t\t\t}\n\t\t\t\t\t\t}\n\t\t\t\t\t\t// Note: Don't close loading here - it will be closed when next tool starts or at end\n\t\t\t\t\t\t// Reset tool state but keep lastToolLoadingID to close it later\n\t\t\t\t\t\tcurrentTool = nil\n\t\t\t\t\t}\n\t\t\t\t}\n\t\t\t}\n\n\t\tcase \"assistant\":\n\t\t\t// Assistant message - extract content\n\t\t\t// With --include-partial-messages, we receive real-time text via stream_event\n\t\t\t// The assistant message contains the full accumulated content\n\t\t\tif msgData, ok := msg[\"message\"].(map[string]interface{}); ok {\n\t\t\t\t// Get model from message\n\t\t\t\tif m, ok := msgData[\"model\"].(string); ok && model == \"\" {\n\t\t\t\t\tmodel = m\n\t\t\t\t}\n\n\t\t\t\t// Check if this is the final message (has stop_reason)\n\t\t\t\tstopReason, hasStopReason := msgData[\"stop_reason\"].(string)\n\t\t\t\tisFinalMessage := hasStopReason && stopReason != \"\"\n\n\t\t\t\t// Extract content from final message\n\t\t\t\t// This serves as a fallback if stream_event wasn't received\n\t\t\t\tif isFinalMessage {\n\t\t\t\t\tif contentArr, ok := msgData[\"content\"].([]interface{}); ok {\n\t\t\t\t\t\tfor _, item := range contentArr {\n\t\t\t\t\t\t\tif contentItem, ok := item.(map[string]interface{}); ok {\n\t\t\t\t\t\t\t\titemType, _ := contentItem[\"type\"].(string)\n\n\t\t\t\t\t\t\t\tswitch itemType {\n\t\t\t\t\t\t\t\tcase \"text\":\n\t\t\t\t\t\t\t\t\t// Only use this if we haven't already accumulated text from stream_event\n\t\t\t\t\t\t\t\t\tif textContent.Len() == 0 {\n\t\t\t\t\t\t\t\t\t\tif text, ok := contentItem[\"text\"].(string); ok && text != \"\" {\n\t\t\t\t\t\t\t\t\t\t\ttextContent.WriteString(text)\n\t\t\t\t\t\t\t\t\t\t\t// Send to stream handler if available\n\t\t\t\t\t\t\t\t\t\t\tif handler != nil {\n\t\t\t\t\t\t\t\t\t\t\t\tif !messageStarted {\n\t\t\t\t\t\t\t\t\t\t\t\t\tstartData := message.EventMessageStartData{\n\t\t\t\t\t\t\t\t\t\t\t\t\t\tMessageID: fmt.Sprintf(\"sandbox-%d\", time.Now().UnixNano()),\n\t\t\t\t\t\t\t\t\t\t\t\t\t\tType: \"text\",\n\t\t\t\t\t\t\t\t\t\t\t\t\t\tTimestamp: time.Now().UnixMilli(),\n\t\t\t\t\t\t\t\t\t\t\t\t\t}\n\t\t\t\t\t\t\t\t\t\t\t\t\tstartDataJSON, _ := json.Marshal(startData)\n\t\t\t\t\t\t\t\t\t\t\t\t\thandler(message.ChunkMessageStart, startDataJSON)\n\t\t\t\t\t\t\t\t\t\t\t\t\tmessageStarted = true\n\t\t\t\t\t\t\t\t\t\t\t\t}\n\t\t\t\t\t\t\t\t\t\t\t\thandler(message.ChunkText, []byte(text))\n\t\t\t\t\t\t\t\t\t\t\t}\n\t\t\t\t\t\t\t\t\t\t}\n\t\t\t\t\t\t\t\t\t}\n\n\t\t\t\t\t\t\t\tcase \"tool_use\":\n\t\t\t\t\t\t\t\t\ttoolName := getString(contentItem, \"name\")\n\t\t\t\t\t\t\t\t\ttoolCall := agentContext.ToolCall{\n\t\t\t\t\t\t\t\t\t\tID: getString(contentItem, \"id\"),\n\t\t\t\t\t\t\t\t\t\tType: agentContext.ToolTypeFunction,\n\t\t\t\t\t\t\t\t\t\tFunction: agentContext.Function{\n\t\t\t\t\t\t\t\t\t\t\tName: toolName,\n\t\t\t\t\t\t\t\t\t\t},\n\t\t\t\t\t\t\t\t\t}\n\t\t\t\t\t\t\t\t\t// Get input as JSON string\n\t\t\t\t\t\t\t\t\tvar inputJSONStr string\n\t\t\t\t\t\t\t\t\tif input, ok := contentItem[\"input\"]; ok {\n\t\t\t\t\t\t\t\t\t\tif inputJSON, err := json.Marshal(input); err == nil {\n\t\t\t\t\t\t\t\t\t\t\tinputJSONStr = string(inputJSON)\n\t\t\t\t\t\t\t\t\t\t\ttoolCall.Function.Arguments = inputJSONStr\n\t\t\t\t\t\t\t\t\t\t}\n\t\t\t\t\t\t\t\t\t}\n\t\t\t\t\t\t\t\t\ttoolCalls = append(toolCalls, toolCall)\n\n\t\t\t\t\t\t\t\t\t// Create tool loading message (from complete assistant message)\n\t\t\t\t\t\t\t\t\t// This is a fallback for when stream_event wasn't received\n\t\t\t\t\t\t\t\t\tif toolName != \"\" && ctx != nil {\n\t\t\t\t\t\t\t\t\t\t// Close previous tool loading if exists\n\t\t\t\t\t\t\t\t\t\tif lastToolLoadingID != \"\" {\n\t\t\t\t\t\t\t\t\t\t\tdoneMsg := &message.Message{\n\t\t\t\t\t\t\t\t\t\t\t\tMessageID: lastToolLoadingID,\n\t\t\t\t\t\t\t\t\t\t\t\tDelta: true,\n\t\t\t\t\t\t\t\t\t\t\t\tDeltaAction: message.DeltaReplace,\n\t\t\t\t\t\t\t\t\t\t\t\tType: message.TypeLoading,\n\t\t\t\t\t\t\t\t\t\t\t\tProps: map[string]interface{}{\n\t\t\t\t\t\t\t\t\t\t\t\t\t\"message\": \"\",\n\t\t\t\t\t\t\t\t\t\t\t\t\t\"done\": true,\n\t\t\t\t\t\t\t\t\t\t\t\t},\n\t\t\t\t\t\t\t\t\t\t\t}\n\t\t\t\t\t\t\t\t\t\t\tctx.Send(doneMsg)\n\t\t\t\t\t\t\t\t\t\t}\n\n\t\t\t\t\t\t\t\t\t\t// Create new loading for this tool\n\t\t\t\t\t\t\t\t\t\tlocale := ctx.Locale\n\t\t\t\t\t\t\t\t\t\tdetailedMsg := getToolDetailedDescription(toolName, inputJSONStr, locale)\n\t\t\t\t\t\t\t\t\t\tif detailedMsg != \"\" {\n\t\t\t\t\t\t\t\t\t\t\ttoolLoadingMsg := &message.Message{\n\t\t\t\t\t\t\t\t\t\t\t\tType: message.TypeLoading,\n\t\t\t\t\t\t\t\t\t\t\t\tProps: map[string]interface{}{\n\t\t\t\t\t\t\t\t\t\t\t\t\t\"message\": detailedMsg,\n\t\t\t\t\t\t\t\t\t\t\t\t},\n\t\t\t\t\t\t\t\t\t\t\t}\n\t\t\t\t\t\t\t\t\t\t\tnewLoadingID, _ := ctx.SendStream(toolLoadingMsg)\n\t\t\t\t\t\t\t\t\t\t\tlastToolLoadingID = newLoadingID\n\t\t\t\t\t\t\t\t\t\t\tlog.Printf(\"[Sandbox] Tool: %s -> %s\", toolName, detailedMsg)\n\t\t\t\t\t\t\t\t\t\t}\n\t\t\t\t\t\t\t\t\t}\n\t\t\t\t\t\t\t\t}\n\t\t\t\t\t\t\t}\n\t\t\t\t\t\t}\n\t\t\t\t\t}\n\t\t\t\t}\n\n\t\t\t\t// Extract usage (from any message that has it)\n\t\t\t\tif usageData, ok := msgData[\"usage\"].(map[string]interface{}); ok {\n\t\t\t\t\tusage = &message.UsageInfo{}\n\t\t\t\t\tif v, ok := usageData[\"input_tokens\"].(float64); ok {\n\t\t\t\t\t\tusage.PromptTokens = int(v)\n\t\t\t\t\t}\n\t\t\t\t\tif v, ok := usageData[\"output_tokens\"].(float64); ok {\n\t\t\t\t\t\tusage.CompletionTokens = int(v)\n\t\t\t\t\t}\n\t\t\t\t\tusage.TotalTokens = usage.PromptTokens + usage.CompletionTokens\n\t\t\t\t}\n\t\t\t}\n\n\t\tcase \"result\":\n\t\t\t// Final result message\n\t\t\t// Check if this is an error result (is_error: true)\n\t\t\tisError, _ := msg[\"is_error\"].(bool)\n\t\t\tif result, ok := msg[\"result\"].(string); ok {\n\t\t\t\tif isError {\n\t\t\t\t\t// This is an error - return it as an error\n\t\t\t\t\treturn nil, fmt.Errorf(\"Claude CLI error: %s\", result)\n\t\t\t\t}\n\t\t\t\tfinalResult = result\n\t\t\t}\n\t\t\t// Send done signal to handler (only if message was started and not an error)\n\t\t\tif handler != nil && messageStarted && !isError {\n\t\t\t\thandler(message.ChunkMessageEnd, nil)\n\t\t\t}\n\n\t\tcase \"error\":\n\t\t\t// Error message\n\t\t\tif errMsg, ok := msg[\"error\"].(string); ok {\n\t\t\t\treturn nil, fmt.Errorf(\"Claude CLI error: %s\", errMsg)\n\t\t\t}\n\t\t\tif errObj, ok := msg[\"error\"].(map[string]interface{}); ok {\n\t\t\t\tif errMsg, ok := errObj[\"message\"].(string); ok {\n\t\t\t\t\treturn nil, fmt.Errorf(\"Claude CLI error: %s\", errMsg)\n\t\t\t\t}\n\t\t\t}\n\t\t}\n\t}\n\n\tscanErr := scanner.Err()\n\tif scanErr != nil {\n\t\treturn nil, fmt.Errorf(\"error reading stream: %w\", scanErr)\n\t}\n\n\t// Close the last tool loading message if exists\n\tif lastToolLoadingID != \"\" && ctx != nil {\n\t\tdoneMsg := &message.Message{\n\t\t\tMessageID: lastToolLoadingID,\n\t\t\tDelta: true,\n\t\t\tDeltaAction: message.DeltaReplace,\n\t\t\tType: message.TypeLoading,\n\t\t\tProps: map[string]interface{}{\n\t\t\t\t\"message\": \"\",\n\t\t\t\t\"done\": true,\n\t\t\t},\n\t\t}\n\t\tctx.Send(doneMsg)\n\t}\n\n\t// Use final result if available, otherwise use accumulated text content\n\tcontent := textContent.String()\n\tif finalResult != \"\" && content == \"\" {\n\t\tcontent = finalResult\n\t}\n\n\t// Build response\n\tresponse := &agentContext.CompletionResponse{\n\t\tID: fmt.Sprintf(\"sandbox-%d\", time.Now().UnixNano()),\n\t\tModel: model,\n\t\tCreated: time.Now().Unix(),\n\t\tRole: \"assistant\",\n\t\tContent: content,\n\t\tFinishReason: agentContext.FinishReasonStop,\n\t}\n\n\t// Add tool calls if any\n\tif len(toolCalls) > 0 {\n\t\tresponse.ToolCalls = toolCalls\n\t\tresponse.FinishReason = agentContext.FinishReasonToolCalls\n\t}\n\n\t// Add usage if available\n\tif usage != nil {\n\t\tresponse.Usage = usage\n\t}\n\n\treturn response, nil\n}\n\n// truncateStr truncates a string to maxLen characters\nfunc truncateStr(s string, maxLen int) string {\n\tif len(s) <= maxLen {\n\t\treturn s\n\t}\n\treturn s[:maxLen] + \"...\"\n}\n\n// getToolDescription returns a human-readable, localized description for a Claude CLI tool\nfunc getToolDescription(toolName string, locale string) string {\n\t// Map tool names to i18n keys\n\ttoolKeys := map[string]string{\n\t\t\"Read\": \"sandbox.tool.read\",\n\t\t\"Write\": \"sandbox.tool.write\",\n\t\t\"Edit\": \"sandbox.tool.edit\",\n\t\t\"StrReplace\": \"sandbox.tool.edit\",\n\t\t\"Bash\": \"sandbox.tool.bash\",\n\t\t\"Shell\": \"sandbox.tool.bash\",\n\t\t\"Glob\": \"sandbox.tool.glob\",\n\t\t\"Grep\": \"sandbox.tool.grep\",\n\t\t\"LS\": \"sandbox.tool.ls\",\n\t\t\"Task\": \"sandbox.tool.task\",\n\t\t\"WebSearch\": \"sandbox.tool.web_search\",\n\t\t\"WebFetch\": \"sandbox.tool.web_fetch\",\n\t\t\"TodoWrite\": \"sandbox.tool.todo_write\",\n\t\t\"AskQuestion\": \"sandbox.tool.ask_question\",\n\t\t\"SwitchMode\": \"sandbox.tool.switch_mode\",\n\t\t\"ReadLints\": \"sandbox.tool.read_lints\",\n\t\t\"EditNotebook\": \"sandbox.tool.edit_notebook\",\n\t}\n\n\tif key, ok := toolKeys[toolName]; ok {\n\t\treturn i18n.T(locale, key)\n\t}\n\t// For unknown tools, use the unknown key and replace {{name}} manually\n\ttemplate := i18n.T(locale, \"sandbox.tool.unknown\")\n\treturn strings.Replace(template, \"{{name}}\", toolName, 1)\n}\n\n// getToolDetailedDescription returns a detailed description with specific parameters\n// It parses the tool input JSON and extracts key information to show users\nfunc getToolDetailedDescription(toolName string, inputJSON string, locale string) string {\n\t// Parse the input JSON using fault-tolerant parser\n\tparsed, err := goujson.Parse(inputJSON)\n\tif err != nil {\n\t\t// Fall back to basic description if parsing fails\n\t\treturn getToolDescription(toolName, locale)\n\t}\n\n\tinput, ok := parsed.(map[string]interface{})\n\tif !ok {\n\t\treturn getToolDescription(toolName, locale)\n\t}\n\n\t// Extract key information based on tool type\n\tvar detail string\n\tswitch toolName {\n\tcase \"Bash\", \"Shell\":\n\t\t// Show the command being executed\n\t\tif cmd, ok := input[\"command\"].(string); ok && cmd != \"\" {\n\t\t\t// Truncate long commands\n\t\t\tif len(cmd) > 50 {\n\t\t\t\tcmd = cmd[:47] + \"...\"\n\t\t\t}\n\t\t\tdetail = cmd\n\t\t}\n\n\tcase \"Read\":\n\t\t// Show the file being read\n\t\tif path, ok := input[\"path\"].(string); ok && path != \"\" {\n\t\t\tdetail = filepath.Base(path)\n\t\t}\n\n\tcase \"Write\":\n\t\t// Show the file being written\n\t\t// Note: Claude CLI uses \"file_path\" for Write tool, not \"path\"\n\t\tif path, ok := input[\"file_path\"].(string); ok && path != \"\" {\n\t\t\tdetail = filepath.Base(path)\n\t\t} else if path, ok := input[\"path\"].(string); ok && path != \"\" {\n\t\t\tdetail = filepath.Base(path)\n\t\t}\n\n\tcase \"Edit\", \"StrReplace\":\n\t\t// Show the file being edited\n\t\tif path, ok := input[\"path\"].(string); ok && path != \"\" {\n\t\t\tdetail = filepath.Base(path)\n\t\t}\n\n\tcase \"Glob\":\n\t\t// Show the glob pattern\n\t\tif pattern, ok := input[\"glob_pattern\"].(string); ok && pattern != \"\" {\n\t\t\tdetail = pattern\n\t\t} else if pattern, ok := input[\"pattern\"].(string); ok && pattern != \"\" {\n\t\t\tdetail = pattern\n\t\t}\n\n\tcase \"Grep\":\n\t\t// Show the search pattern\n\t\tif pattern, ok := input[\"pattern\"].(string); ok && pattern != \"\" {\n\t\t\tif len(pattern) > 30 {\n\t\t\t\tpattern = pattern[:27] + \"...\"\n\t\t\t}\n\t\t\tdetail = pattern\n\t\t}\n\n\tcase \"LS\":\n\t\t// Show the directory\n\t\tif path, ok := input[\"target_directory\"].(string); ok && path != \"\" {\n\t\t\tdetail = filepath.Base(path)\n\t\t} else if path, ok := input[\"path\"].(string); ok && path != \"\" {\n\t\t\tdetail = filepath.Base(path)\n\t\t}\n\n\tcase \"WebSearch\":\n\t\t// Show the search query\n\t\tif query, ok := input[\"search_term\"].(string); ok && query != \"\" {\n\t\t\tif len(query) > 40 {\n\t\t\t\tquery = query[:37] + \"...\"\n\t\t\t}\n\t\t\tdetail = query\n\t\t} else if query, ok := input[\"query\"].(string); ok && query != \"\" {\n\t\t\tif len(query) > 40 {\n\t\t\t\tquery = query[:37] + \"...\"\n\t\t\t}\n\t\t\tdetail = query\n\t\t}\n\n\tcase \"WebFetch\":\n\t\t// Show the URL\n\t\tif url, ok := input[\"url\"].(string); ok && url != \"\" {\n\t\t\t// Extract domain from URL\n\t\t\tif len(url) > 50 {\n\t\t\t\turl = url[:47] + \"...\"\n\t\t\t}\n\t\t\tdetail = url\n\t\t}\n\n\tcase \"Task\":\n\t\t// Show the task description\n\t\tif desc, ok := input[\"description\"].(string); ok && desc != \"\" {\n\t\t\tif len(desc) > 40 {\n\t\t\t\tdesc = desc[:37] + \"...\"\n\t\t\t}\n\t\t\tdetail = desc\n\t\t}\n\t}\n\n\t// Build the message with detail\n\tbaseMsg := getToolDescription(toolName, locale)\n\tif detail != \"\" {\n\t\treturn baseMsg + \": \" + detail\n\t}\n\treturn baseMsg\n}\n\n// ReadFile reads a file from the container\nfunc (e *Executor) ReadFile(ctx context.Context, path string) ([]byte, error) {\n\t// Make path absolute if not\n\tif !strings.HasPrefix(path, \"/\") {\n\t\tpath = e.workDir + \"/\" + path\n\t}\n\treturn e.manager.ReadFile(ctx, e.containerName, path)\n}\n\n// WriteFile writes content to a file in the container\nfunc (e *Executor) WriteFile(ctx context.Context, path string, content []byte) error {\n\t// Make path absolute if not\n\tif !strings.HasPrefix(path, \"/\") {\n\t\tpath = e.workDir + \"/\" + path\n\t}\n\treturn e.manager.WriteFile(ctx, e.containerName, path, content)\n}\n\n// ListDir lists directory contents in the container\nfunc (e *Executor) ListDir(ctx context.Context, path string) ([]infraSandbox.FileInfo, error) {\n\t// Make path absolute if not\n\tif !strings.HasPrefix(path, \"/\") {\n\t\tpath = e.workDir + \"/\" + path\n\t}\n\n\treturn e.manager.ListDir(ctx, e.containerName, path)\n}\n\n// Exec executes a command in the container\nfunc (e *Executor) Exec(ctx context.Context, cmd []string) (string, error) {\n\tresult, err := e.manager.Exec(ctx, e.containerName, cmd, &infraSandbox.ExecOptions{\n\t\tWorkDir: e.workDir,\n\t})\n\tif err != nil {\n\t\treturn \"\", err\n\t}\n\n\tif result.ExitCode != 0 {\n\t\treturn result.Stdout, fmt.Errorf(\"command exited with code %d: %s\", result.ExitCode, result.Stderr)\n\t}\n\n\treturn result.Stdout, nil\n}\n\n// GetWorkDir returns the container workspace directory\nfunc (e *Executor) GetWorkDir() string {\n\treturn e.workDir\n}\n\n// GetSandboxID returns the sandbox ID (userID-chatID)\nfunc (e *Executor) GetSandboxID() string {\n\tif e.opts == nil {\n\t\treturn \"\"\n\t}\n\treturn fmt.Sprintf(\"%s-%s\", e.opts.UserID, e.opts.ChatID)\n}\n\n// GetVNCUrl returns the VNC preview URL path\n// Returns empty string if VNC is not enabled for this sandbox image\nfunc (e *Executor) GetVNCUrl() string {\n\tif e.opts == nil {\n\t\treturn \"\"\n\t}\n\n\timageName := e.opts.Image\n\tif imageName == \"\" {\n\t\treturn \"\"\n\t}\n\n\t// Check if the image supports VNC using the shared keyword list in sandbox package\n\tif !infraSandbox.IsVNCImage(imageName) {\n\t\treturn \"\"\n\t}\n\n\t// Return only the sandbox ID, the full URL is constructed by openapi/sandbox.GetVNCClientURL()\n\treturn e.GetSandboxID()\n}\n\n// Close releases the executor resources and removes the container\n// Note: IPC session is managed by sandbox.Manager.Remove()\nfunc (e *Executor) Close() error {\n\tif e.manager != nil && e.containerName != \"\" {\n\t\tctx := context.Background()\n\t\treturn e.manager.Remove(ctx, e.containerName)\n\t}\n\treturn nil\n}\n\n// Helper function to get string from map\nfunc getString(m map[string]interface{}, key string) string {\n\tif v, ok := m[key].(string); ok {\n\t\treturn v\n\t}\n\treturn \"\"\n}\n" }, { "path": "agent/sandbox/claude/executor_test.go", "content": "package claude\n\nimport (\n\t\"context\"\n\t\"fmt\"\n\t\"os\"\n\t\"strings\"\n\t\"testing\"\n\t\"time\"\n\n\t\"github.com/stretchr/testify/assert\"\n\t\"github.com/stretchr/testify/require\"\n\t\"github.com/yaoapp/yao/config\"\n\tinfraSandbox \"github.com/yaoapp/yao/sandbox\"\n\t\"github.com/yaoapp/yao/test\"\n)\n\n// createTestManager creates a sandbox manager for testing with proper configuration\nfunc createTestManager(t *testing.T) *infraSandbox.Manager {\n\t// Get data root from environment or use temp directory\n\tdataRoot := os.Getenv(\"YAO_ROOT\")\n\tif dataRoot == \"\" {\n\t\tdataRoot = t.TempDir()\n\t}\n\n\t// Create config with proper paths\n\tcfg := infraSandbox.DefaultConfig()\n\tcfg.Init(dataRoot)\n\n\tmanager, err := infraSandbox.NewManager(cfg)\n\tif err != nil {\n\t\tt.Skipf(\"Skipping test: Docker not available: %v\", err)\n\t\treturn nil\n\t}\n\n\treturn manager\n}\n\nfunc TestNewClaudeExecutor(t *testing.T) {\n\ttest.Prepare(t, config.Conf)\n\tdefer test.Clean()\n\n\tmanager := createTestManager(t)\n\tif manager == nil {\n\t\treturn\n\t}\n\tdefer manager.Close()\n\n\topts := &Options{\n\t\tCommand: \"claude\",\n\t\tImage: \"yaoapp/sandbox-claude:latest\",\n\t\tUserID: \"test-user\",\n\t\tChatID: fmt.Sprintf(\"test-chat-claude-%d\", time.Now().UnixNano()),\n\t\tConnectorHost: \"https://api.example.com\",\n\t\tConnectorKey: \"key123\",\n\t\tModel: \"test-model\",\n\t}\n\n\texec, err := NewExecutor(manager, opts)\n\trequire.NoError(t, err)\n\trequire.NotNil(t, exec)\n\n\t// Verify executor was created\n\tassert.Equal(t, \"/workspace\", exec.GetWorkDir())\n\tassert.NoError(t, exec.Close())\n}\n\nfunc TestClaudeExecutorMissingRequiredFields(t *testing.T) {\n\ttest.Prepare(t, config.Conf)\n\tdefer test.Clean()\n\n\tmanager := createTestManager(t)\n\tif manager == nil {\n\t\treturn\n\t}\n\tdefer manager.Close()\n\n\t// Missing UserID\n\t_, err := NewExecutor(manager, &Options{\n\t\tCommand: \"claude\",\n\t\tChatID: \"test-chat\",\n\t})\n\tassert.Error(t, err)\n\tassert.Contains(t, err.Error(), \"UserID is required\")\n\n\t// Missing ChatID\n\t_, err = NewExecutor(manager, &Options{\n\t\tCommand: \"claude\",\n\t\tUserID: \"test-user\",\n\t})\n\tassert.Error(t, err)\n\tassert.Contains(t, err.Error(), \"ChatID is required\")\n}\n\nfunc TestClaudeExecutorFileOperations(t *testing.T) {\n\ttest.Prepare(t, config.Conf)\n\tdefer test.Clean()\n\n\tmanager := createTestManager(t)\n\tif manager == nil {\n\t\treturn\n\t}\n\tdefer manager.Close()\n\n\topts := &Options{\n\t\tCommand: \"claude\",\n\t\tImage: \"alpine:latest\", // Use alpine for simpler testing\n\t\tUserID: \"test-user\",\n\t\tChatID: fmt.Sprintf(\"test-chat-file-ops-%d\", time.Now().UnixNano()),\n\t}\n\n\texec, err := NewExecutor(manager, opts)\n\trequire.NoError(t, err)\n\tdefer exec.Close()\n\n\tctx := context.Background()\n\n\t// Test WriteFile\n\tcontent := []byte(\"Hello, World!\")\n\terr = exec.WriteFile(ctx, \"test-file.txt\", content)\n\trequire.NoError(t, err)\n\n\t// Test ReadFile\n\treadContent, err := exec.ReadFile(ctx, \"test-file.txt\")\n\trequire.NoError(t, err)\n\tassert.Equal(t, content, readContent)\n\n\t// Test ListDir\n\tfiles, err := exec.ListDir(ctx, \".\")\n\trequire.NoError(t, err)\n\tassert.True(t, len(files) > 0, \"Expected at least one file in directory\")\n\n\t// Find our test file\n\tvar found bool\n\tfor _, f := range files {\n\t\tif f.Name == \"test-file.txt\" {\n\t\t\tfound = true\n\t\t\tbreak\n\t\t}\n\t}\n\tassert.True(t, found, \"Expected to find test-file.txt in directory listing\")\n}\n\nfunc TestClaudeExecutorExec(t *testing.T) {\n\ttest.Prepare(t, config.Conf)\n\tdefer test.Clean()\n\n\tmanager := createTestManager(t)\n\tif manager == nil {\n\t\treturn\n\t}\n\tdefer manager.Close()\n\n\topts := &Options{\n\t\tCommand: \"claude\",\n\t\tImage: \"alpine:latest\", // Use alpine for simpler testing\n\t\tUserID: \"test-user\",\n\t\tChatID: fmt.Sprintf(\"test-chat-exec-%d\", time.Now().UnixNano()),\n\t}\n\n\texec, err := NewExecutor(manager, opts)\n\trequire.NoError(t, err)\n\tdefer exec.Close()\n\n\tctx := context.Background()\n\n\t// Test simple echo command\n\toutput, err := exec.Exec(ctx, []string{\"echo\", \"hello-world\"})\n\trequire.NoError(t, err)\n\tassert.Contains(t, output, \"hello-world\")\n}\n\n// TestClaudeExecutorMCPConfigWrite tests that MCP config is correctly written to container\nfunc TestClaudeExecutorMCPConfigWrite(t *testing.T) {\n\ttest.Prepare(t, config.Conf)\n\tdefer test.Clean()\n\n\tmanager := createTestManager(t)\n\tif manager == nil {\n\t\treturn\n\t}\n\tdefer manager.Close()\n\n\t// Create MCP config JSON\n\tmcpConfig := []byte(`{\"mcpServers\":{\"echo\":{\"command\":\"yao-mcp-proxy\",\"args\":[\"echo\"],\"tools\":[\"ping\",\"echo\"]}}}`)\n\n\topts := &Options{\n\t\tCommand: \"claude\",\n\t\tImage: \"alpine:latest\",\n\t\tUserID: \"test-user\",\n\t\tChatID: fmt.Sprintf(\"test-chat-mcp-write-%d\", time.Now().UnixNano()),\n\t\tMCPConfig: mcpConfig,\n\t\t// No connector config - skip proxy start for alpine test image\n\t}\n\n\texec, err := NewExecutor(manager, opts)\n\trequire.NoError(t, err)\n\tdefer exec.Close()\n\n\tctx := context.Background()\n\n\t// Call prepareEnvironment to write configs\n\terr = exec.prepareEnvironment(ctx)\n\trequire.NoError(t, err, \"prepareEnvironment should succeed\")\n\n\t// Verify MCP config was written by reading it back\n\treadContent, err := exec.ReadFile(ctx, \".mcp.json\")\n\trequire.NoError(t, err, \"Should be able to read .mcp.json\")\n\trequire.NotEmpty(t, readContent, \"MCP config should not be empty\")\n\n\tt.Logf(\"MCP config in container: %s\", string(readContent))\n\n\t// Verify content matches\n\tassert.JSONEq(t, string(mcpConfig), string(readContent), \"MCP config content should match\")\n\n\tt.Log(\"✓ MCP config verified in container\")\n}\n\n// TestClaudeExecutorSkillsCopy tests that skills directory is correctly copied to container\n// Uses real test application skills directory\nfunc TestClaudeExecutorSkillsCopy(t *testing.T) {\n\ttest.Prepare(t, config.Conf)\n\tdefer test.Clean()\n\n\tmanager := createTestManager(t)\n\tif manager == nil {\n\t\treturn\n\t}\n\tdefer manager.Close()\n\n\t// Use real skills directory from test application\n\tappRoot := os.Getenv(\"YAO_ROOT\")\n\trequire.NotEmpty(t, appRoot, \"YAO_ROOT should be set\")\n\n\tskillsDir := appRoot + \"/assistants/tests/sandbox/full/skills\"\n\n\t// Verify skills directory exists on host\n\tinfo, err := os.Stat(skillsDir)\n\trequire.NoError(t, err, \"Skills directory should exist: %s\", skillsDir)\n\trequire.True(t, info.IsDir(), \"Skills path should be a directory\")\n\tt.Logf(\"Using real skills directory: %s\", skillsDir)\n\n\topts := &Options{\n\t\tCommand: \"claude\",\n\t\tImage: \"alpine:latest\",\n\t\tUserID: \"test-user\",\n\t\tChatID: fmt.Sprintf(\"test-chat-skills-%d\", time.Now().UnixNano()),\n\t\tSkillsDir: skillsDir,\n\t\t// No connector config - skip proxy start for alpine test image\n\t}\n\n\texec, err := NewExecutor(manager, opts)\n\trequire.NoError(t, err)\n\tdefer exec.Close()\n\n\tctx := context.Background()\n\n\t// Call prepareEnvironment to copy skills\n\terr = exec.prepareEnvironment(ctx)\n\trequire.NoError(t, err, \"prepareEnvironment should succeed\")\n\n\t// Verify .claude directory was created\n\toutput, err := exec.Exec(ctx, []string{\"ls\", \"-la\", \".claude\"})\n\trequire.NoError(t, err, \".claude directory should exist\")\n\tt.Logf(\".claude directory contents:\\n%s\", output)\n\n\t// Verify skills directory exists in container\n\toutput, err = exec.Exec(ctx, []string{\"ls\", \"-la\", \".claude/skills\"})\n\trequire.NoError(t, err, \"skills directory should exist in container\")\n\tt.Logf(\"skills directory contents:\\n%s\", output)\n\tassert.Contains(t, output, \"echo-test\", \"echo-test skill should exist\")\n\n\t// Verify echo-test skill was copied correctly\n\toutput, err = exec.Exec(ctx, []string{\"ls\", \"-la\", \".claude/skills/echo-test\"})\n\trequire.NoError(t, err, \"echo-test skill directory should exist\")\n\tassert.Contains(t, output, \"SKILL.md\", \"SKILL.md should exist in echo-test\")\n\tassert.Contains(t, output, \"scripts\", \"scripts directory should exist in echo-test\")\n\tt.Logf(\"echo-test skill contents:\\n%s\", output)\n\n\t// Read SKILL.md content to verify\n\treadContent, err := exec.ReadFile(ctx, \".claude/skills/echo-test/SKILL.md\")\n\trequire.NoError(t, err, \"Should be able to read SKILL.md from container\")\n\trequire.NotEmpty(t, readContent, \"SKILL.md content should not be empty\")\n\t// Verify content contains expected strings from the real SKILL.md\n\tassert.Contains(t, string(readContent), \"name: echo-test\", \"SKILL.md should contain skill name\")\n\tassert.Contains(t, string(readContent), \"# Echo Test\", \"SKILL.md should contain the title\")\n\tt.Logf(\"✓ SKILL.md content verified (%d bytes)\", len(readContent))\n\n\tt.Log(\"✓ Skills directory verified in container with real test data\")\n}\n\n// TestClaudeExecutorPrepareEnvironmentIntegration tests full environment preparation\n// Uses real test application data\nfunc TestClaudeExecutorPrepareEnvironmentIntegration(t *testing.T) {\n\ttest.Prepare(t, config.Conf)\n\tdefer test.Clean()\n\n\tmanager := createTestManager(t)\n\tif manager == nil {\n\t\treturn\n\t}\n\tdefer manager.Close()\n\n\t// Use real skills directory from test application\n\tappRoot := os.Getenv(\"YAO_ROOT\")\n\trequire.NotEmpty(t, appRoot, \"YAO_ROOT should be set\")\n\tskillsDir := appRoot + \"/assistants/tests/sandbox/full/skills\"\n\n\t// Verify skills directory exists\n\t_, err := os.Stat(skillsDir)\n\trequire.NoError(t, err, \"Skills directory should exist\")\n\n\t// Create MCP config (simulating what buildMCPConfigForSandbox produces)\n\tmcpConfig := []byte(`{\"mcpServers\":{\"echo\":{\"command\":\"yao-mcp-proxy\",\"args\":[\"echo\"],\"tools\":[\"ping\",\"echo\",\"status\"]}}}`)\n\n\topts := &Options{\n\t\tCommand: \"claude\",\n\t\tImage: \"alpine:latest\",\n\t\tUserID: \"test-user\",\n\t\tChatID: fmt.Sprintf(\"test-chat-full-env-%d\", time.Now().UnixNano()),\n\t\tMCPConfig: mcpConfig,\n\t\tSkillsDir: skillsDir,\n\t\t// No connector config - skip proxy start for alpine test image\n\t}\n\n\texec, err := NewExecutor(manager, opts)\n\trequire.NoError(t, err)\n\tdefer exec.Close()\n\n\tctx := context.Background()\n\n\t// Call prepareEnvironment\n\terr = exec.prepareEnvironment(ctx)\n\trequire.NoError(t, err, \"prepareEnvironment should succeed\")\n\n\t// 1. Check MCP config\n\tmcpContent, err := exec.ReadFile(ctx, \".mcp.json\")\n\trequire.NoError(t, err, \"MCP config should exist in container\")\n\tassert.JSONEq(t, string(mcpConfig), string(mcpContent), \"MCP config content should match\")\n\tt.Logf(\"✓ MCP config verified: %s\", string(mcpContent))\n\n\t// 2. Check Skills directory structure\n\toutput, err := exec.Exec(ctx, []string{\"ls\", \"-la\", \".claude/skills\"})\n\trequire.NoError(t, err, \"Skills directory should exist in container\")\n\tassert.Contains(t, output, \"echo-test\", \"echo-test skill should exist\")\n\tt.Logf(\"✓ Skills directory contents:\\n%s\", output)\n\n\t// 3. Check skill content\n\tskillContent, err := exec.ReadFile(ctx, \".claude/skills/echo-test/SKILL.md\")\n\trequire.NoError(t, err, \"SKILL.md should exist in container\")\n\trequire.NotEmpty(t, skillContent, \"SKILL.md should not be empty\")\n\tassert.Contains(t, string(skillContent), \"name: echo-test\", \"SKILL.md should contain skill name\")\n\tassert.Contains(t, string(skillContent), \"# Echo Test\", \"SKILL.md should contain the title\")\n\tt.Logf(\"✓ SKILL.md verified: %d bytes\", len(skillContent))\n\n\tt.Log(\"✓ Full environment preparation verified with real test data\")\n}\n\n// TestClaudeExecutorIPCSocketMount verifies that IPC socket is bind mounted to container\nfunc TestClaudeExecutorIPCSocketMount(t *testing.T) {\n\tmanager := createTestManager(t)\n\n\topts := &Options{\n\t\tCommand: \"claude\",\n\t\tImage: \"alpine:latest\",\n\t\tUserID: \"test-user\",\n\t\tChatID: \"test-ipc-socket-\" + fmt.Sprintf(\"%d\", time.Now().UnixNano()),\n\t\tConnectorHost: \"https://api.test.com\",\n\t\tConnectorKey: \"test-key\",\n\t\tModel: \"test-model\",\n\t}\n\n\texec, err := NewExecutor(manager, opts)\n\trequire.NoError(t, err)\n\tdefer exec.Close()\n\n\tctx := context.Background()\n\n\t// Check if IPC socket exists in container\n\toutput, err := exec.Exec(ctx, []string{\"ls\", \"-la\", \"/run/yao.sock\"})\n\trequire.NoError(t, err, \"IPC socket should exist in container\")\n\tassert.Contains(t, output, \"yao.sock\", \"Should find yao.sock file\")\n\tt.Logf(\"✓ IPC socket mounted: %s\", strings.TrimSpace(output))\n\n\t// Verify it's a socket file (starts with 's' in ls output)\n\tassert.Contains(t, output, \"srw\", \"Should be a socket file (starts with 's')\")\n\tt.Log(\"✓ IPC socket is correctly bind mounted to container\")\n}\n" }, { "path": "agent/sandbox/claude/real_e2e_test.go", "content": "package claude\n\nimport (\n\t\"context\"\n\t\"fmt\"\n\t\"os\"\n\t\"strings\"\n\t\"testing\"\n\t\"time\"\n\n\t\"github.com/stretchr/testify/require\"\n\tagentContext \"github.com/yaoapp/yao/agent/context\"\n\t\"github.com/yaoapp/yao/agent/output/message\"\n\t\"github.com/yaoapp/yao/config\"\n\tinfraSandbox \"github.com/yaoapp/yao/sandbox\"\n\t\"github.com/yaoapp/yao/test\"\n)\n\n// TestRealClaudeCLIExecution tests real Claude CLI execution with streaming\n// This test requires:\n// 1. Docker running with yaoapp/sandbox-claude:latest image\n// 2. Environment variables: DEEPSEEK_API_KEY, DEEPSEEK_API_PROXY, DEEPSEEK_MODELS_V3\nfunc TestRealClaudeCLIExecution(t *testing.T) {\n\tif testing.Short() {\n\t\tt.Skip(\"Skipping real E2E test in short mode\")\n\t}\n\n\t// Check for required environment variables\n\tapiKey := os.Getenv(\"DEEPSEEK_API_KEY\")\n\tapiProxy := os.Getenv(\"DEEPSEEK_API_PROXY\")\n\tmodel := os.Getenv(\"DEEPSEEK_MODELS_V3\")\n\n\tif apiKey == \"\" || apiProxy == \"\" || model == \"\" {\n\t\tt.Skip(\"Skipping test: DEEPSEEK_API_KEY, DEEPSEEK_API_PROXY, or DEEPSEEK_MODELS_V3 not set\")\n\t}\n\n\ttest.Prepare(t, config.Conf)\n\tdefer test.Clean()\n\n\t// Get data root from environment\n\tdataRoot := os.Getenv(\"YAO_ROOT\")\n\tif dataRoot == \"\" {\n\t\tt.Skip(\"Skipping test: YAO_ROOT not set\")\n\t}\n\n\t// Create config with proper paths\n\tcfg := infraSandbox.DefaultConfig()\n\tcfg.Init(dataRoot)\n\n\tmanager, err := infraSandbox.NewManager(cfg)\n\tif err != nil {\n\t\tt.Skipf(\"Skipping test: Docker not available: %v\", err)\n\t}\n\tdefer manager.Close()\n\n\t// Create options WITH SystemPrompt (triggers Claude CLI execution)\n\topts := &Options{\n\t\tCommand: \"claude\",\n\t\tImage: \"yaoapp/sandbox-claude:latest\",\n\t\tUserID: \"test-user\",\n\t\tChatID: fmt.Sprintf(\"test-real-e2e-%d\", time.Now().UnixNano()),\n\t\tConnectorHost: apiProxy,\n\t\tConnectorKey: apiKey,\n\t\tModel: model,\n\t\tSystemPrompt: \"You are a helpful assistant. Reply concisely.\",\n\t\tTimeout: 3 * time.Minute,\n\t\tConnectorOptions: map[string]interface{}{\n\t\t\t\"max_tokens\": 4096, // Limit max_tokens to avoid backend API limits\n\t\t},\n\t}\n\n\tt.Logf(\"Creating executor with options:\")\n\tt.Logf(\" ConnectorHost: %s\", opts.ConnectorHost)\n\tt.Logf(\" Model: %s\", opts.Model)\n\tt.Logf(\" SystemPrompt: %s\", opts.SystemPrompt)\n\n\texec, err := NewExecutor(manager, opts)\n\tif err != nil {\n\t\tt.Skipf(\"Skipping test: Failed to create executor: %v\", err)\n\t}\n\tdefer exec.Close()\n\n\t// Verify shouldSkipClaudeCLI returns false\n\tif exec.shouldSkipClaudeCLI() {\n\t\tt.Fatal(\"shouldSkipClaudeCLI should return false when SystemPrompt is set\")\n\t}\n\n\t// Test 1: First, manually test claude-proxy\n\tt.Log(\"=== Test 1: Verify claude-proxy is working ===\")\n\tstdCtx := context.Background()\n\n\t// Prepare environment (this starts claude-proxy)\n\terr = exec.prepareEnvironment(stdCtx)\n\trequire.NoError(t, err, \"prepareEnvironment should succeed\")\n\n\t// Check proxy is running\n\tresult, err := exec.manager.Exec(stdCtx, exec.containerName, []string{\"pgrep\", \"-f\", \"claude-proxy\"}, nil)\n\tif err != nil || result.ExitCode != 0 {\n\t\tt.Log(\"claude-proxy not running, checking why...\")\n\t\t// Check proxy log\n\t\tlogContent, _ := exec.ReadFile(stdCtx, \"proxy.log\")\n\t\tt.Logf(\"Proxy log: %s\", string(logContent))\n\n\t\t// Check config\n\t\tconfigContent, _ := exec.ReadFile(stdCtx, \".claude-proxy.json\")\n\t\tt.Logf(\"Proxy config: %s\", string(configContent))\n\t} else {\n\t\tt.Logf(\"claude-proxy is running with PID: %s\", strings.TrimSpace(result.Stdout))\n\t}\n\n\t// Test 2: Test simple command execution\n\tt.Log(\"=== Test 2: Simple command execution ===\")\n\tctx := agentContext.New(stdCtx, nil, opts.ChatID)\n\tmessages := []agentContext.Message{\n\t\t{Role: \"user\", Content: \"Reply with exactly: HELLO_TEST_SUCCESS\"},\n\t}\n\n\t// Collect streaming output\n\tvar streamedChunks []string\n\tvar streamedContent strings.Builder\n\tstreamHandler := func(chunkType message.StreamChunkType, data []byte) int {\n\t\tchunk := string(data)\n\t\tstreamedChunks = append(streamedChunks, chunk)\n\t\tstreamedContent.Write(data)\n\t\tt.Logf(\"Stream chunk [%s]: %q\", chunkType, chunk)\n\t\treturn 0 // continue streaming\n\t}\n\n\tt.Log(\"Executing Claude CLI...\")\n\tstartTime := time.Now()\n\tresponse, err := exec.Stream(ctx, messages, streamHandler)\n\tduration := time.Since(startTime)\n\tt.Logf(\"Execution took: %v\", duration)\n\n\tif err != nil {\n\t\tt.Logf(\"Stream error: %v\", err)\n\n\t\t// Debug: check what's in the container\n\t\tt.Log(\"=== Debug info ===\")\n\n\t\t// Check proxy log\n\t\tlogContent, _ := exec.ReadFile(stdCtx, \"proxy.log\")\n\t\tt.Logf(\"Proxy log:\\n%s\", string(logContent))\n\n\t\t// List workspace\n\t\toutput, _ := exec.Exec(stdCtx, []string{\"ls\", \"-la\", \"/workspace\"})\n\t\tt.Logf(\"Workspace contents:\\n%s\", output)\n\n\t\t// Check environment\n\t\toutput, _ = exec.Exec(stdCtx, []string{\"env\"})\n\t\tt.Logf(\"Environment:\\n%s\", output)\n\n\t\tt.Fatalf(\"Stream failed: %v\", err)\n\t}\n\n\trequire.NotNil(t, response, \"Response should not be nil\")\n\n\t// Log results\n\tt.Logf(\"=== Results ===\")\n\tt.Logf(\"Response ID: %s\", response.ID)\n\tt.Logf(\"Response Model: %s\", response.Model)\n\tt.Logf(\"Response Content: %v\", response.Content)\n\tt.Logf(\"Streamed chunks count: %d\", len(streamedChunks))\n\tt.Logf(\"Total streamed content: %s\", streamedContent.String())\n\n\t// Verify we got some response\n\tvar fullResponse string\n\tif content, ok := response.Content.(string); ok {\n\t\tfullResponse = content\n\t}\n\tif fullResponse == \"\" {\n\t\tfullResponse = streamedContent.String()\n\t}\n\n\tif fullResponse == \"\" {\n\t\t// Check proxy log for errors\n\t\tlogContent, _ := exec.ReadFile(stdCtx, \"proxy.log\")\n\t\tt.Logf(\"Proxy log (for debugging):\\n%s\", string(logContent))\n\t\tt.Fatal(\"Got empty response from Claude CLI\")\n\t}\n\n\tt.Logf(\"✓ Successfully got response: %s\", fullResponse)\n\n\t// Check if streaming worked\n\tif len(streamedChunks) > 0 {\n\t\tt.Logf(\"✓ Streaming worked with %d chunks\", len(streamedChunks))\n\t} else {\n\t\tt.Log(\"⚠ No streaming chunks received (might be buffered)\")\n\t}\n}\n\n// TestClaudeCLIDirectExecution tests running claude directly in the container\nfunc TestClaudeCLIDirectExecution(t *testing.T) {\n\tif testing.Short() {\n\t\tt.Skip(\"Skipping real E2E test in short mode\")\n\t}\n\n\t// Check for required environment variables\n\tapiKey := os.Getenv(\"DEEPSEEK_API_KEY\")\n\tapiProxy := os.Getenv(\"DEEPSEEK_API_PROXY\")\n\tmodel := os.Getenv(\"DEEPSEEK_MODELS_V3\")\n\n\tif apiKey == \"\" || apiProxy == \"\" || model == \"\" {\n\t\tt.Skip(\"Skipping test: DEEPSEEK_API_KEY, DEEPSEEK_API_PROXY, or DEEPSEEK_MODELS_V3 not set\")\n\t}\n\n\ttest.Prepare(t, config.Conf)\n\tdefer test.Clean()\n\n\tdataRoot := os.Getenv(\"YAO_ROOT\")\n\tif dataRoot == \"\" {\n\t\tt.Skip(\"Skipping test: YAO_ROOT not set\")\n\t}\n\n\tcfg := infraSandbox.DefaultConfig()\n\tcfg.Init(dataRoot)\n\n\tmanager, err := infraSandbox.NewManager(cfg)\n\tif err != nil {\n\t\tt.Skipf(\"Skipping test: Docker not available: %v\", err)\n\t}\n\tdefer manager.Close()\n\n\topts := &Options{\n\t\tCommand: \"claude\",\n\t\tImage: \"yaoapp/sandbox-claude:latest\",\n\t\tUserID: \"test-user\",\n\t\tChatID: fmt.Sprintf(\"test-direct-%d\", time.Now().UnixNano()),\n\t\tConnectorHost: apiProxy,\n\t\tConnectorKey: apiKey,\n\t\tModel: model,\n\t\tTimeout: 3 * time.Minute,\n\t\tConnectorOptions: map[string]interface{}{\n\t\t\t\"max_tokens\": 4096, // Limit max_tokens to avoid backend API limits\n\t\t},\n\t}\n\n\texec, err := NewExecutor(manager, opts)\n\tif err != nil {\n\t\tt.Skipf(\"Skipping test: Failed to create executor: %v\", err)\n\t}\n\tdefer exec.Close()\n\n\tstdCtx := context.Background()\n\n\t// Step 1: Write proxy config and start proxy\n\tt.Log(\"=== Step 1: Start claude-proxy ===\")\n\terr = exec.prepareEnvironment(stdCtx)\n\trequire.NoError(t, err)\n\n\t// Wait for proxy to start\n\ttime.Sleep(2 * time.Second)\n\n\t// Check proxy status\n\tresult, err := exec.manager.Exec(stdCtx, exec.containerName, []string{\"pgrep\", \"-f\", \"claude-proxy\"}, nil)\n\tif err == nil && result.ExitCode == 0 {\n\t\tt.Logf(\"✓ claude-proxy running, PID: %s\", strings.TrimSpace(result.Stdout))\n\t} else {\n\t\tt.Log(\"⚠ claude-proxy might not be running\")\n\t}\n\n\t// Step 2: Run claude CLI directly with simple prompt\n\tt.Log(\"=== Step 2: Run claude CLI directly ===\")\n\n\t// Build a simple command - pass env vars explicitly\n\tdirectCmd := []string{\n\t\t\"bash\", \"-c\",\n\t\t`echo '{\"type\":\"user\",\"message\":{\"role\":\"user\",\"content\":\"say hello\"}}' | claude -p --dangerously-skip-permissions --permission-mode bypassPermissions --input-format stream-json --output-format stream-json --verbose 2>&1`,\n\t}\n\n\treader, err := exec.manager.Stream(stdCtx, exec.containerName, directCmd, &infraSandbox.ExecOptions{\n\t\tWorkDir: exec.workDir,\n\t\tTimeout: 2 * time.Minute,\n\t\tEnv: map[string]string{\n\t\t\t\"ANTHROPIC_BASE_URL\": \"http://127.0.0.1:3456\",\n\t\t\t\"ANTHROPIC_API_KEY\": \"dummy\",\n\t\t},\n\t})\n\tif err != nil {\n\t\tt.Fatalf(\"Failed to execute: %v\", err)\n\t}\n\tdefer reader.Close()\n\n\t// Read output\n\tbuf := make([]byte, 64*1024)\n\tvar output strings.Builder\n\tfor {\n\t\tn, err := reader.Read(buf)\n\t\tif n > 0 {\n\t\t\tchunk := string(buf[:n])\n\t\t\toutput.WriteString(chunk)\n\t\t\tt.Logf(\"Output chunk: %q\", chunk)\n\t\t}\n\t\tif err != nil {\n\t\t\tbreak\n\t\t}\n\t}\n\n\tt.Logf(\"=== Full output ===\\n%s\", output.String())\n\n\tif output.Len() == 0 {\n\t\t// Check logs\n\t\tlogContent, _ := exec.ReadFile(stdCtx, \"proxy.log\")\n\t\tt.Logf(\"Proxy log:\\n%s\", string(logContent))\n\t\tt.Fatal(\"Got no output from claude CLI\")\n\t}\n\n\t// Check for success indicators\n\toutputStr := output.String()\n\tif strings.Contains(outputStr, \"error\") || strings.Contains(outputStr, \"Error\") {\n\t\tt.Logf(\"⚠ Output contains error\")\n\t}\n\tif strings.Contains(outputStr, \"content_block\") || strings.Contains(outputStr, \"message_start\") {\n\t\tt.Log(\"✓ Got streaming JSON output from Claude CLI\")\n\t}\n}\n" }, { "path": "agent/sandbox/claude/types.go", "content": "package claude\n\n// StreamMessage represents a parsed stream message from Claude CLI\ntype StreamMessage struct {\n\tType string `json:\"type\"`\n\tSubtype string `json:\"subtype,omitempty\"`\n\tContent interface{} `json:\"content,omitempty\"`\n\tError string `json:\"error,omitempty\"`\n}\n\n// ToolCall represents a tool invocation from the agent\ntype ToolCall struct {\n\tID string `json:\"id\"`\n\tName string `json:\"name\"`\n\tArguments map[string]interface{} `json:\"arguments\"`\n}\n\n// ToolResult represents a tool execution result\ntype ToolResult struct {\n\tID string `json:\"id\"`\n\tContent string `json:\"content\"`\n\tIsError bool `json:\"is_error,omitempty\"`\n}\n\n// CLIResponse represents the parsed response from Claude CLI\ntype CLIResponse struct {\n\tText string `json:\"text,omitempty\"`\n\tToolCalls []ToolCall `json:\"tool_calls,omitempty\"`\n\tUsage *Usage `json:\"usage,omitempty\"`\n\tModel string `json:\"model,omitempty\"`\n}\n\n// Usage represents token usage statistics\ntype Usage struct {\n\tInputTokens int `json:\"input_tokens,omitempty\"`\n\tOutputTokens int `json:\"output_tokens,omitempty\"`\n}\n" }, { "path": "agent/sandbox/cursor/README.md", "content": "# Cursor Executor\n\n## Status\n\n**Not Implemented** - This is a placeholder for future Cursor CLI integration.\n\n## Planned Features\n\nThe Cursor executor will provide similar functionality to the Claude executor:\n\n- Execute Cursor CLI in a Docker sandbox container\n- Stream output in real-time\n- File system operations (ReadFile, WriteFile, ListDir)\n- Command execution (Exec)\n- Integration with Yao's MCP servers\n\n## Configuration\n\nWhen implemented, the Cursor executor will be configured in assistant `package.yao`:\n\n```jsonc\n{\n \"name\": \"Coder Assistant\",\n \"connector\": \"deepseek.v3\",\n \"sandbox\": {\n \"command\": \"cursor\", // Use Cursor CLI\n \"image\": \"yaoapp/sandbox-cursor:latest\",\n \"timeout\": \"10m\"\n }\n}\n```\n\n## Implementation Notes\n\nThe implementation should follow the same pattern as `claude/executor.go`:\n\n1. Create `cursor/executor.go` implementing the `sandbox.Executor` interface\n2. Create `cursor/command.go` for building Cursor CLI commands\n3. Create `cursor/types.go` for Cursor-specific types\n4. Add appropriate tests\n\n## Docker Image\n\nA `yaoapp/sandbox-cursor` Docker image will need to be created with:\n\n- Ubuntu 24.04 LTS base\n- Node.js 22 LTS\n- Python 3.12\n- Cursor CLI installed and configured\n\n## References\n\n- [Cursor CLI Documentation](https://cursor.sh/docs)\n- [Claude Executor Implementation](../claude/executor.go)\n- [Sandbox Design Document](../DESIGN.md)\n" }, { "path": "agent/sandbox/executor.go", "content": "package sandbox\n\nimport (\n\t\"fmt\"\n\n\t\"github.com/yaoapp/yao/agent/sandbox/claude\"\n\tinfraSandbox \"github.com/yaoapp/yao/sandbox\"\n)\n\n// New creates a new Executor based on the command type\nfunc New(manager *infraSandbox.Manager, opts *Options) (Executor, error) {\n\tif opts == nil {\n\t\treturn nil, fmt.Errorf(\"options is required\")\n\t}\n\n\tif !IsValidCommand(opts.Command) {\n\t\treturn nil, fmt.Errorf(\"unsupported command type: %s, supported: %v\", opts.Command, CommandTypes)\n\t}\n\n\t// Set default image if not specified\n\tif opts.Image == \"\" {\n\t\topts.Image = DefaultImage(opts.Command)\n\t}\n\n\tswitch opts.Command {\n\tcase \"claude\":\n\t\t// Convert to claude.Options\n\t\tclaudeOpts := &claude.Options{\n\t\t\tCommand: opts.Command,\n\t\t\tImage: opts.Image,\n\t\t\tMaxMemory: opts.MaxMemory,\n\t\t\tMaxCPU: opts.MaxCPU,\n\t\t\tTimeout: opts.Timeout,\n\t\t\tArguments: opts.Arguments,\n\t\t\tUserID: opts.UserID,\n\t\t\tChatID: opts.ChatID,\n\t\t\tMCPConfig: opts.MCPConfig,\n\t\t\tMCPTools: opts.MCPTools,\n\t\t\tSkillsDir: opts.SkillsDir,\n\t\t\tSystemPrompt: opts.SystemPrompt, // Required for Claude CLI execution\n\t\t\tConnectorHost: opts.ConnectorHost,\n\t\t\tConnectorKey: opts.ConnectorKey,\n\t\t\tModel: opts.Model,\n\t\t\tConnectorType: opts.ConnectorType, // \"openai\" or \"anthropic\"\n\t\t\tConnectorOptions: opts.ConnectorOptions, // Extra options like thinking, max_tokens\n\t\t\tSecrets: opts.Secrets, // Secrets for container env vars\n\t\t}\n\t\treturn claude.NewExecutor(manager, claudeOpts)\n\tcase \"cursor\":\n\t\treturn nil, fmt.Errorf(\"cursor executor not implemented yet\")\n\tdefault:\n\t\treturn nil, fmt.Errorf(\"unsupported command type: %s\", opts.Command)\n\t}\n}\n" }, { "path": "agent/sandbox/executor_test.go", "content": "package sandbox\n\nimport (\n\t\"os\"\n\t\"testing\"\n\n\t\"github.com/stretchr/testify/assert\"\n\t\"github.com/stretchr/testify/require\"\n\t\"github.com/yaoapp/yao/config\"\n\tinfraSandbox \"github.com/yaoapp/yao/sandbox\"\n\t\"github.com/yaoapp/yao/test\"\n)\n\n// createTestManager creates a sandbox manager for testing with proper configuration\nfunc createTestManager(t *testing.T) *infraSandbox.Manager {\n\t// Get data root from environment or use temp directory\n\tdataRoot := os.Getenv(\"YAO_ROOT\")\n\tif dataRoot == \"\" {\n\t\tdataRoot = t.TempDir()\n\t}\n\n\t// Create config with proper paths\n\tcfg := infraSandbox.DefaultConfig()\n\tcfg.Init(dataRoot)\n\n\tmanager, err := infraSandbox.NewManager(cfg)\n\tif err != nil {\n\t\tt.Skipf(\"Skipping test: Docker not available: %v\", err)\n\t\treturn nil\n\t}\n\n\treturn manager\n}\n\nfunc TestNewExecutorWithInvalidOptions(t *testing.T) {\n\ttest.Prepare(t, config.Conf)\n\tdefer test.Clean()\n\n\tmanager := createTestManager(t)\n\tif manager == nil {\n\t\treturn\n\t}\n\tdefer manager.Close()\n\n\t// Test with nil options\n\t_, err := New(manager, nil)\n\tassert.Error(t, err)\n\tassert.Contains(t, err.Error(), \"options is required\")\n\n\t// Test with invalid command\n\t_, err = New(manager, &Options{\n\t\tCommand: \"invalid\",\n\t\tUserID: \"user1\",\n\t\tChatID: \"chat1\",\n\t})\n\tassert.Error(t, err)\n\tassert.Contains(t, err.Error(), \"unsupported command type\")\n}\n\nfunc TestNewExecutorWithValidOptions(t *testing.T) {\n\ttest.Prepare(t, config.Conf)\n\tdefer test.Clean()\n\n\tmanager := createTestManager(t)\n\tif manager == nil {\n\t\treturn\n\t}\n\tdefer manager.Close()\n\n\t// Test with valid claude options\n\topts := &Options{\n\t\tCommand: \"claude\",\n\t\tUserID: \"test-user\",\n\t\tChatID: \"test-chat\",\n\t\tConnectorHost: \"https://api.example.com\",\n\t\tConnectorKey: \"key123\",\n\t\tModel: \"test-model\",\n\t}\n\n\texec, err := New(manager, opts)\n\trequire.NoError(t, err)\n\trequire.NotNil(t, exec)\n\n\t// Verify executor was created\n\tassert.NotEmpty(t, exec.GetWorkDir())\n\tassert.NoError(t, exec.Close())\n}\n\nfunc TestDefaultImageIsSetWhenEmpty(t *testing.T) {\n\ttest.Prepare(t, config.Conf)\n\tdefer test.Clean()\n\n\tmanager := createTestManager(t)\n\tif manager == nil {\n\t\treturn\n\t}\n\tdefer manager.Close()\n\n\topts := &Options{\n\t\tCommand: \"claude\",\n\t\tImage: \"\", // Empty, should be set to default\n\t\tUserID: \"test-user\",\n\t\tChatID: \"test-chat-2\",\n\t}\n\n\texec, err := New(manager, opts)\n\trequire.NoError(t, err)\n\tdefer exec.Close() // Ensure cleanup\n\n\t// The image should have been set to default\n\tassert.Equal(t, \"yaoapp/sandbox-claude:latest\", opts.Image)\n}\n\nfunc TestCursorExecutorNotImplemented(t *testing.T) {\n\ttest.Prepare(t, config.Conf)\n\tdefer test.Clean()\n\n\tmanager := createTestManager(t)\n\tif manager == nil {\n\t\treturn\n\t}\n\tdefer manager.Close()\n\n\topts := &Options{\n\t\tCommand: \"cursor\",\n\t\tUserID: \"test-user\",\n\t\tChatID: \"test-chat-3\",\n\t}\n\n\t_, err := New(manager, opts)\n\tassert.Error(t, err)\n\tassert.Contains(t, err.Error(), \"not implemented\")\n}\n" }, { "path": "agent/sandbox/integration_test.go", "content": "package sandbox\n\nimport (\n\t\"context\"\n\t\"os\"\n\t\"testing\"\n\n\t\"github.com/stretchr/testify/assert\"\n\t\"github.com/stretchr/testify/require\"\n\tagentContext \"github.com/yaoapp/yao/agent/context\"\n\t\"github.com/yaoapp/yao/config\"\n\tinfraSandbox \"github.com/yaoapp/yao/sandbox\"\n\t\"github.com/yaoapp/yao/test\"\n)\n\n// createIntegrationTestManager creates a sandbox manager for integration testing\nfunc createIntegrationTestManager(t *testing.T) *infraSandbox.Manager {\n\tdataRoot := os.Getenv(\"YAO_ROOT\")\n\tif dataRoot == \"\" {\n\t\tdataRoot = t.TempDir()\n\t}\n\n\tcfg := infraSandbox.DefaultConfig()\n\tcfg.Init(dataRoot)\n\n\tmanager, err := infraSandbox.NewManager(cfg)\n\tif err != nil {\n\t\tt.Skipf(\"Skipping test: Docker not available: %v\", err)\n\t\treturn nil\n\t}\n\n\treturn manager\n}\n\n// TestExecutorInterfaceCompatibility verifies that the executor implements both interfaces correctly\nfunc TestExecutorInterfaceCompatibility(t *testing.T) {\n\ttest.Prepare(t, config.Conf)\n\tdefer test.Clean()\n\n\tmanager := createIntegrationTestManager(t)\n\tif manager == nil {\n\t\treturn\n\t}\n\tdefer manager.Close()\n\n\t// Create executor via factory function\n\topts := &Options{\n\t\tCommand: \"claude\",\n\t\tImage: \"alpine:latest\",\n\t\tUserID: \"test-user\",\n\t\tChatID: \"test-compat\",\n\t}\n\n\texecutor, err := New(manager, opts)\n\trequire.NoError(t, err)\n\trequire.NotNil(t, executor)\n\tdefer executor.Close()\n\n\t// Verify executor implements agent/sandbox.Executor interface\n\tvar _ Executor = executor\n\n\t// Verify executor can be cast to context.SandboxExecutor\n\tctxExecutor, ok := executor.(agentContext.SandboxExecutor)\n\trequire.True(t, ok, \"executor should implement context.SandboxExecutor\")\n\trequire.NotNil(t, ctxExecutor)\n\n\t// Test SandboxExecutor methods work\n\tctx := context.Background()\n\n\t// WriteFile\n\terr = ctxExecutor.WriteFile(ctx, \"compat-test.txt\", []byte(\"compatibility test\"))\n\trequire.NoError(t, err)\n\n\t// ReadFile\n\tcontent, err := ctxExecutor.ReadFile(ctx, \"compat-test.txt\")\n\trequire.NoError(t, err)\n\tassert.Equal(t, \"compatibility test\", string(content))\n\n\t// ListDir\n\tfiles, err := ctxExecutor.ListDir(ctx, \".\")\n\trequire.NoError(t, err)\n\tassert.True(t, len(files) > 0)\n\n\t// Exec\n\toutput, err := ctxExecutor.Exec(ctx, []string{\"echo\", \"compat\"})\n\trequire.NoError(t, err)\n\tassert.Contains(t, output, \"compat\")\n\n\t// GetWorkDir\n\tworkDir := ctxExecutor.GetWorkDir()\n\tassert.NotEmpty(t, workDir)\n}\n\n// TestExecutorRoundTrip tests the full round-trip of creating executor and performing operations\nfunc TestExecutorRoundTrip(t *testing.T) {\n\ttest.Prepare(t, config.Conf)\n\tdefer test.Clean()\n\n\tmanager := createIntegrationTestManager(t)\n\tif manager == nil {\n\t\treturn\n\t}\n\tdefer manager.Close()\n\n\topts := &Options{\n\t\tCommand: \"claude\",\n\t\tImage: \"alpine:latest\",\n\t\tUserID: \"test-user\",\n\t\tChatID: \"test-roundtrip\",\n\t\tConnectorHost: \"https://api.example.com\",\n\t\tConnectorKey: \"test-key\",\n\t\tModel: \"test-model\",\n\t}\n\n\t// Create executor\n\texecutor, err := New(manager, opts)\n\trequire.NoError(t, err)\n\trequire.NotNil(t, executor)\n\tdefer executor.Close()\n\n\tctx := context.Background()\n\n\t// 1. Write a file\n\ttestContent := \"Hello, integration test!\"\n\terr = executor.WriteFile(ctx, \"integration.txt\", []byte(testContent))\n\trequire.NoError(t, err, \"WriteFile should succeed\")\n\n\t// 2. Read the file back\n\treadContent, err := executor.ReadFile(ctx, \"integration.txt\")\n\trequire.NoError(t, err, \"ReadFile should succeed\")\n\tassert.Equal(t, testContent, string(readContent), \"Content should match\")\n\n\t// 3. List directory\n\tfiles, err := executor.ListDir(ctx, \".\")\n\trequire.NoError(t, err, \"ListDir should succeed\")\n\n\tvar found bool\n\tfor _, f := range files {\n\t\tif f.Name == \"integration.txt\" {\n\t\t\tfound = true\n\t\t\tassert.False(t, f.IsDir, \"Should not be a directory\")\n\t\t\tassert.Equal(t, int64(len(testContent)), f.Size, \"Size should match\")\n\t\t\tbreak\n\t\t}\n\t}\n\tassert.True(t, found, \"Should find integration.txt in listing\")\n\n\t// 4. Execute command\n\toutput, err := executor.Exec(ctx, []string{\"cat\", \"/workspace/integration.txt\"})\n\trequire.NoError(t, err, \"Exec should succeed\")\n\tassert.Contains(t, output, testContent, \"cat output should contain file content\")\n\n\t// 5. Verify workdir\n\tassert.Equal(t, \"/workspace\", executor.GetWorkDir(), \"WorkDir should be /workspace\")\n}\n\n// TestMultipleExecutorsIsolation verifies that multiple executors have isolated workspaces\nfunc TestMultipleExecutorsIsolation(t *testing.T) {\n\ttest.Prepare(t, config.Conf)\n\tdefer test.Clean()\n\n\tmanager := createIntegrationTestManager(t)\n\tif manager == nil {\n\t\treturn\n\t}\n\tdefer manager.Close()\n\n\t// Create two executors with different chat IDs\n\topts1 := &Options{\n\t\tCommand: \"claude\",\n\t\tImage: \"alpine:latest\",\n\t\tUserID: \"test-user\",\n\t\tChatID: \"test-isolation-1\",\n\t}\n\topts2 := &Options{\n\t\tCommand: \"claude\",\n\t\tImage: \"alpine:latest\",\n\t\tUserID: \"test-user\",\n\t\tChatID: \"test-isolation-2\",\n\t}\n\n\texec1, err := New(manager, opts1)\n\trequire.NoError(t, err)\n\tdefer exec1.Close()\n\n\texec2, err := New(manager, opts2)\n\trequire.NoError(t, err)\n\tdefer exec2.Close()\n\n\tctx := context.Background()\n\n\t// Write different content to each executor\n\terr = exec1.WriteFile(ctx, \"test.txt\", []byte(\"executor 1\"))\n\trequire.NoError(t, err)\n\n\terr = exec2.WriteFile(ctx, \"test.txt\", []byte(\"executor 2\"))\n\trequire.NoError(t, err)\n\n\t// Read back and verify isolation\n\tcontent1, err := exec1.ReadFile(ctx, \"test.txt\")\n\trequire.NoError(t, err)\n\tassert.Equal(t, \"executor 1\", string(content1), \"Executor 1 should have its own content\")\n\n\tcontent2, err := exec2.ReadFile(ctx, \"test.txt\")\n\trequire.NoError(t, err)\n\tassert.Equal(t, \"executor 2\", string(content2), \"Executor 2 should have its own content\")\n}\n" }, { "path": "agent/sandbox/types.go", "content": "package sandbox\n\nimport (\n\t\"context\"\n\t\"time\"\n\n\tagentContext \"github.com/yaoapp/yao/agent/context\"\n\t\"github.com/yaoapp/yao/agent/output/message\"\n\tinfraSandbox \"github.com/yaoapp/yao/sandbox\"\n\t\"github.com/yaoapp/yao/sandbox/ipc\"\n)\n\n// Executor executes LLM requests in sandbox\ntype Executor interface {\n\t// Execute runs the request and returns response (uses options set at creation time)\n\tExecute(ctx *agentContext.Context, messages []agentContext.Message) (*agentContext.CompletionResponse, error)\n\n\t// Stream runs the request with streaming output (uses options set at creation time)\n\tStream(ctx *agentContext.Context, messages []agentContext.Message, handler message.StreamFunc) (*agentContext.CompletionResponse, error)\n\n\t// SetLoadingMsgID sets the loading message ID for tool execution status updates\n\tSetLoadingMsgID(id string)\n\n\t// Filesystem operations (for Hooks)\n\tReadFile(ctx context.Context, path string) ([]byte, error)\n\tWriteFile(ctx context.Context, path string, content []byte) error\n\tListDir(ctx context.Context, path string) ([]infraSandbox.FileInfo, error)\n\n\t// Command execution (for Hooks)\n\tExec(ctx context.Context, cmd []string) (string, error)\n\n\t// GetWorkDir returns the container workspace directory\n\tGetWorkDir() string\n\n\t// GetSandboxID returns the sandbox ID (userID-chatID)\n\tGetSandboxID() string\n\n\t// GetVNCUrl returns the VNC preview URL path (e.g., /api/__yao/vnc/{sandboxID}/)\n\t// Returns empty string if VNC is not enabled for this sandbox image\n\tGetVNCUrl() string\n\n\t// Close releases container resources\n\tClose() error\n}\n\n// FileInfo is an alias to infrastructure sandbox FileInfo for convenience\ntype FileInfo = infraSandbox.FileInfo\n\n// Options for sandbox execution\ntype Options struct {\n\t// Command type (claude, cursor)\n\tCommand string `json:\"command\"`\n\n\t// Docker image (optional, auto-selected by command)\n\tImage string `json:\"image,omitempty\"`\n\n\t// Resource limits\n\tMaxMemory string `json:\"max_memory,omitempty\"`\n\tMaxCPU float64 `json:\"max_cpu,omitempty\"`\n\n\t// Execution timeout\n\tTimeout time.Duration `json:\"timeout,omitempty\"`\n\n\t// Command-specific arguments (passed to CLI)\n\tArguments map[string]interface{} `json:\"arguments,omitempty\"`\n\n\t// ========================================\n\t// Internal fields (auto-resolved by Yao)\n\t// Do NOT set these in package.yao config\n\t// ========================================\n\n\t// UserID for workspace isolation\n\tUserID string `json:\"-\"`\n\n\t// ChatID for session isolation\n\tChatID string `json:\"-\"`\n\n\t// MCP configuration - auto-loaded from assistants/{name}/mcps/\n\tMCPConfig []byte `json:\"-\"`\n\n\t// MCPTools - MCP tools to expose via IPC (tool name → tool definition)\n\tMCPTools map[string]*ipc.MCPTool `json:\"-\"`\n\n\t// Skills directory - auto-resolved to assistants/{name}/skills/\n\tSkillsDir string `json:\"-\"`\n\n\t// SystemPrompt - extracted from assistant prompts.yml\n\t// Used to determine if Claude CLI should be called\n\tSystemPrompt string `json:\"-\"`\n\n\t// Connector settings - auto-resolved from connector config file\n\t// e.g., connectors/deepseek/v3.conn.yao → host, key, model\n\tConnectorHost string `json:\"-\"`\n\tConnectorKey string `json:\"-\"`\n\tModel string `json:\"-\"`\n\n\t// ConnectorType - connector API type: \"openai\" or \"anthropic\"\n\t// Determines whether to use claude-proxy (openai) or direct connection (anthropic)\n\tConnectorType string `json:\"-\"`\n\n\t// ConnectorOptions - extra options from connector config (e.g., thinking, max_tokens, temperature)\n\t// These are backend-specific parameters passed to the proxy\n\tConnectorOptions map[string]interface{} `json:\"-\"`\n\n\t// Secrets - sensitive values from sandbox.secrets config (e.g., GITHUB_TOKEN)\n\t// Resolved from $ENV.XXX references, exported as env vars in container\n\tSecrets map[string]string `json:\"-\"`\n}\n\n// SandboxConfig represents the sandbox configuration in assistant package.yao\ntype SandboxConfig struct {\n\t// Command type (claude, cursor)\n\tCommand string `json:\"command\" yaml:\"command\"`\n\n\t// Docker image (optional, auto-selected by command)\n\tImage string `json:\"image,omitempty\" yaml:\"image,omitempty\"`\n\n\t// Resource limits\n\tMaxMemory string `json:\"max_memory,omitempty\" yaml:\"max_memory,omitempty\"`\n\tMaxCPU float64 `json:\"max_cpu,omitempty\" yaml:\"max_cpu,omitempty\"`\n\n\t// Execution timeout\n\tTimeout string `json:\"timeout,omitempty\" yaml:\"timeout,omitempty\"`\n\n\t// Command-specific arguments (passed to CLI)\n\tArguments map[string]interface{} `json:\"arguments,omitempty\" yaml:\"arguments,omitempty\"`\n}\n\n// DefaultImage returns the default Docker image for a command type\nfunc DefaultImage(command string) string {\n\tswitch command {\n\tcase \"claude\":\n\t\treturn \"yaoapp/sandbox-claude:latest\"\n\tcase \"cursor\":\n\t\treturn \"yaoapp/sandbox-cursor:latest\"\n\tdefault:\n\t\treturn \"\"\n\t}\n}\n\n// CommandTypes is the list of supported command types\nvar CommandTypes = []string{\"claude\", \"cursor\"}\n\n// IsValidCommand checks if a command type is valid\nfunc IsValidCommand(command string) bool {\n\tfor _, c := range CommandTypes {\n\t\tif c == command {\n\t\t\treturn true\n\t\t}\n\t}\n\treturn false\n}\n" }, { "path": "agent/sandbox/types_test.go", "content": "package sandbox\n\nimport (\n\t\"testing\"\n\n\t\"github.com/stretchr/testify/assert\"\n)\n\nfunc TestDefaultImage(t *testing.T) {\n\ttests := []struct {\n\t\tcommand string\n\t\texpected string\n\t}{\n\t\t{\"claude\", \"yaoapp/sandbox-claude:latest\"},\n\t\t{\"cursor\", \"yaoapp/sandbox-cursor:latest\"},\n\t\t{\"unknown\", \"\"},\n\t}\n\n\tfor _, tt := range tests {\n\t\tt.Run(tt.command, func(t *testing.T) {\n\t\t\tresult := DefaultImage(tt.command)\n\t\t\tassert.Equal(t, tt.expected, result)\n\t\t})\n\t}\n}\n\nfunc TestIsValidCommand(t *testing.T) {\n\ttests := []struct {\n\t\tcommand string\n\t\texpected bool\n\t}{\n\t\t{\"claude\", true},\n\t\t{\"cursor\", true},\n\t\t{\"unknown\", false},\n\t\t{\"\", false},\n\t}\n\n\tfor _, tt := range tests {\n\t\tt.Run(tt.command, func(t *testing.T) {\n\t\t\tresult := IsValidCommand(tt.command)\n\t\t\tassert.Equal(t, tt.expected, result)\n\t\t})\n\t}\n}\n\nfunc TestOptionsValidation(t *testing.T) {\n\t// Test that Options struct can be created with all fields\n\topts := &Options{\n\t\tCommand: \"claude\",\n\t\tImage: \"yaoapp/sandbox-claude:latest\",\n\t\tMaxMemory: \"4g\",\n\t\tMaxCPU: 2.0,\n\t\tUserID: \"user123\",\n\t\tChatID: \"chat456\",\n\t\tConnectorHost: \"https://api.example.com\",\n\t\tConnectorKey: \"key123\",\n\t\tModel: \"deepseek-v3\",\n\t\tArguments: map[string]interface{}{\n\t\t\t\"max_turns\": 20,\n\t\t\t\"permission_mode\": \"acceptEdits\",\n\t\t},\n\t}\n\n\tassert.Equal(t, \"claude\", opts.Command)\n\tassert.Equal(t, \"user123\", opts.UserID)\n\tassert.Equal(t, \"chat456\", opts.ChatID)\n\tassert.Equal(t, 20, opts.Arguments[\"max_turns\"])\n}\n\nfunc TestSandboxConfigParsing(t *testing.T) {\n\t// Test that SandboxConfig can be used for parsing assistant config\n\tconfig := &SandboxConfig{\n\t\tCommand: \"claude\",\n\t\tImage: \"custom-image:v1\",\n\t\tMaxMemory: \"8g\",\n\t\tMaxCPU: 4.0,\n\t\tTimeout: \"10m\",\n\t\tArguments: map[string]interface{}{\n\t\t\t\"permission_mode\": \"bypassPermissions\",\n\t\t},\n\t}\n\n\tassert.Equal(t, \"claude\", config.Command)\n\tassert.Equal(t, \"custom-image:v1\", config.Image)\n\tassert.Equal(t, \"8g\", config.MaxMemory)\n\tassert.Equal(t, \"10m\", config.Timeout)\n}\n" }, { "path": "agent/sandbox/v2/claude/attachments.go", "content": "package claude\n\nimport (\n\t\"context\"\n\t\"fmt\"\n\t\"path/filepath\"\n\t\"strings\"\n\n\tagentContext \"github.com/yaoapp/yao/agent/context\"\n\t\"github.com/yaoapp/yao/attachment\"\n\tworkspace \"github.com/yaoapp/yao/tai/workspace\"\n)\n\n// prepareAttachments resolves __yao.attachment:// URLs in messages,\n// copies actual files into the workspace .attachments/{chatID}/ directory via ws.Copy,\n// and replaces multimodal content parts with text references.\nfunc prepareAttachments(ctx context.Context, messages []agentContext.Message, chatID string, ws workspace.FS) ([]agentContext.Message, error) {\n\tusedNames := make(map[string]int)\n\tattachDir := \".attachments/\" + chatID\n\n\tresult := make([]agentContext.Message, len(messages))\n\tcopy(result, messages)\n\n\tfor i, msg := range result {\n\t\tif msg.Role != \"user\" {\n\t\t\tcontinue\n\t\t}\n\n\t\tparts, ok := msg.Content.([]interface{})\n\t\tif !ok {\n\t\t\tif typedParts, ok := msg.Content.([]agentContext.ContentPart); ok {\n\t\t\t\tiparts := make([]interface{}, len(typedParts))\n\t\t\t\tfor j, p := range typedParts {\n\t\t\t\t\tm := map[string]interface{}{\"type\": string(p.Type)}\n\t\t\t\t\tif p.Text != \"\" {\n\t\t\t\t\t\tm[\"text\"] = p.Text\n\t\t\t\t\t}\n\t\t\t\t\tif p.ImageURL != nil {\n\t\t\t\t\t\tm[\"image_url\"] = map[string]interface{}{\n\t\t\t\t\t\t\t\"url\": p.ImageURL.URL,\n\t\t\t\t\t\t\t\"detail\": string(p.ImageURL.Detail),\n\t\t\t\t\t\t}\n\t\t\t\t\t}\n\t\t\t\t\tif p.File != nil {\n\t\t\t\t\t\tm[\"file\"] = map[string]interface{}{\n\t\t\t\t\t\t\t\"url\": p.File.URL,\n\t\t\t\t\t\t\t\"filename\": p.File.Filename,\n\t\t\t\t\t\t}\n\t\t\t\t\t}\n\t\t\t\t\tiparts[j] = m\n\t\t\t\t}\n\t\t\t\tparts = iparts\n\t\t\t} else {\n\t\t\t\tcontinue\n\t\t\t}\n\t\t}\n\n\t\tif len(parts) == 0 {\n\t\t\tcontinue\n\t\t}\n\n\t\tvar textParts []string\n\n\t\tfor _, item := range parts {\n\t\t\tm, ok := item.(map[string]interface{})\n\t\t\tif !ok {\n\t\t\t\tcontinue\n\t\t\t}\n\t\t\tpartType, _ := m[\"type\"].(string)\n\n\t\t\tswitch partType {\n\t\t\tcase \"text\":\n\t\t\t\tif text, ok := m[\"text\"].(string); ok && text != \"\" {\n\t\t\t\t\ttextParts = append(textParts, text)\n\t\t\t\t}\n\n\t\t\tcase \"image_url\":\n\t\t\t\timgData, _ := m[\"image_url\"].(map[string]interface{})\n\t\t\t\tif imgData == nil {\n\t\t\t\t\tcontinue\n\t\t\t\t}\n\t\t\t\turl, _ := imgData[\"url\"].(string)\n\t\t\t\tif url == \"\" {\n\t\t\t\t\tcontinue\n\t\t\t\t}\n\t\t\t\tuploaderName, fileID, isWrapper := attachment.Parse(url)\n\t\t\t\tif !isWrapper {\n\t\t\t\t\ttextParts = append(textParts, fmt.Sprintf(\"[Image: %s]\", url))\n\t\t\t\t\tcontinue\n\t\t\t\t}\n\t\t\t\tref, err := resolveAttachment(ctx, uploaderName, fileID, \"\", attachDir, usedNames, ws)\n\t\t\t\tif err != nil {\n\t\t\t\t\ttextParts = append(textParts, \"[Attached image: failed to load]\")\n\t\t\t\t\tcontinue\n\t\t\t\t}\n\t\t\t\ttextParts = append(textParts, ref)\n\n\t\t\tcase \"file\":\n\t\t\t\tfileData, _ := m[\"file\"].(map[string]interface{})\n\t\t\t\tif fileData == nil {\n\t\t\t\t\tcontinue\n\t\t\t\t}\n\t\t\t\turl, _ := fileData[\"url\"].(string)\n\t\t\t\thintName, _ := fileData[\"filename\"].(string)\n\t\t\t\tif url == \"\" {\n\t\t\t\t\tcontinue\n\t\t\t\t}\n\t\t\t\tuploaderName, fileID, isWrapper := attachment.Parse(url)\n\t\t\t\tif !isWrapper {\n\t\t\t\t\ttextParts = append(textParts, fmt.Sprintf(\"[File: %s]\", url))\n\t\t\t\t\tcontinue\n\t\t\t\t}\n\t\t\t\tref, err := resolveAttachment(ctx, uploaderName, fileID, hintName, attachDir, usedNames, ws)\n\t\t\t\tif err != nil {\n\t\t\t\t\ttextParts = append(textParts, \"[Attached file: failed to load]\")\n\t\t\t\t\tcontinue\n\t\t\t\t}\n\t\t\t\ttextParts = append(textParts, ref)\n\t\t\t}\n\t\t}\n\n\t\tif len(textParts) > 0 {\n\t\t\tnewMsg := result[i]\n\t\t\tnewMsg.Content = strings.Join(textParts, \"\\n\\n\")\n\t\t\tresult[i] = newMsg\n\t\t}\n\t}\n\n\treturn result, nil\n}\n\n// resolveAttachment gets the local path of an attachment and copies it into\n// the workspace via ws.Copy(\"local:///abs/path\", \".attachments/{chatID}/filename\").\nfunc resolveAttachment(\n\tctx context.Context,\n\tuploaderName, fileID, hintName, attachDir string,\n\tusedNames map[string]int,\n\tws workspace.FS,\n) (string, error) {\n\tmanager, exists := attachment.Managers[uploaderName]\n\tif !exists {\n\t\treturn \"\", fmt.Errorf(\"attachment manager not found: %s\", uploaderName)\n\t}\n\n\tfileInfo, err := manager.Info(ctx, fileID)\n\tif err != nil {\n\t\treturn \"\", fmt.Errorf(\"failed to get file info: %w\", err)\n\t}\n\n\tabsPath, _, err := manager.LocalPath(ctx, fileID)\n\tif err != nil {\n\t\treturn \"\", fmt.Errorf(\"failed to get local path: %w\", err)\n\t}\n\n\tfilename := fileInfo.Filename\n\tif filename == \"\" && hintName != \"\" {\n\t\tfilename = hintName\n\t}\n\tif filename == \"\" {\n\t\text := extensionFromContentType(fileInfo.ContentType)\n\t\tfilename = fileID + ext\n\t}\n\n\tbaseName := filename\n\tif count, exists := usedNames[baseName]; exists {\n\t\text := filepath.Ext(filename)\n\t\tname := strings.TrimSuffix(filename, ext)\n\t\tfilename = fmt.Sprintf(\"%s_%d%s\", name, count+1, ext)\n\t\tusedNames[baseName] = count + 1\n\t} else {\n\t\tusedNames[baseName] = 0\n\t}\n\n\tdstPath := attachDir + \"/\" + filename\n\tsrc := \"local:///\" + absPath\n\n\tif _, err := ws.Copy(src, dstPath); err != nil {\n\t\treturn \"\", fmt.Errorf(\"failed to copy attachment to workspace: %w\", err)\n\t}\n\n\tsizeStr := formatFileSize(fileInfo.Bytes)\n\treturn fmt.Sprintf(\"[Attached file: %s (%s, %s)]\", dstPath, fileInfo.ContentType, sizeStr), nil\n}\n\nfunc extensionFromContentType(contentType string) string {\n\tswitch contentType {\n\tcase \"image/png\":\n\t\treturn \".png\"\n\tcase \"image/jpeg\":\n\t\treturn \".jpg\"\n\tcase \"image/gif\":\n\t\treturn \".gif\"\n\tcase \"image/webp\":\n\t\treturn \".webp\"\n\tcase \"image/svg+xml\":\n\t\treturn \".svg\"\n\tcase \"application/pdf\":\n\t\treturn \".pdf\"\n\tcase \"text/plain\":\n\t\treturn \".txt\"\n\tcase \"text/html\":\n\t\treturn \".html\"\n\tcase \"text/css\":\n\t\treturn \".css\"\n\tcase \"text/javascript\", \"application/javascript\":\n\t\treturn \".js\"\n\tcase \"application/json\":\n\t\treturn \".json\"\n\tcase \"application/zip\":\n\t\treturn \".zip\"\n\tdefault:\n\t\treturn \"\"\n\t}\n}\n\nfunc formatFileSize(bytes int) string {\n\tswitch {\n\tcase bytes >= 1024*1024:\n\t\treturn fmt.Sprintf(\"%.1fMB\", float64(bytes)/(1024*1024))\n\tcase bytes >= 1024:\n\t\treturn fmt.Sprintf(\"%.1fKB\", float64(bytes)/1024)\n\tdefault:\n\t\treturn fmt.Sprintf(\"%dB\", bytes)\n\t}\n}\n" }, { "path": "agent/sandbox/v2/claude/oscompat.go", "content": "package claude\n\nimport (\n\t\"fmt\"\n\t\"path\"\n\t\"strings\"\n\n\t\"github.com/yaoapp/yao/agent/sandbox/v2/types\"\n\tinfra \"github.com/yaoapp/yao/sandbox/v2\"\n)\n\n// osEnv captures OS-dependent paths and shell settings derived from the\n// Computer's SystemInfo. All runner code should use osEnv instead of\n// hardcoded Linux constants.\ntype osEnv struct {\n\tOS string // \"windows\", \"linux\", \"darwin\", ...\n\tShell string // preferred shell binary: \"bash\", \"pwsh\", \"cmd.exe\", ...\n\tWorkDir string // working directory on the target machine\n\tUserHome string // user home directory (empty if irrelevant)\n\tTempDir string // system temp directory\n}\n\nfunc (e *osEnv) isWindows() bool {\n\treturn strings.EqualFold(e.OS, \"windows\")\n}\n\n// resolveOSEnv builds an osEnv from the Computer's reported SystemInfo,\n// falling back to SandboxConfig values where available, then to per-OS defaults.\nfunc resolveOSEnv(computer infra.Computer, _ *types.SandboxConfig) *osEnv {\n\tsys := computer.ComputerInfo().System\n\n\tenv := &osEnv{\n\t\tOS: strings.ToLower(sys.OS),\n\t\tShell: sys.Shell,\n\t\tTempDir: sys.TempDir,\n\t\tWorkDir: computer.GetWorkDir(),\n\t}\n\n\tif env.TempDir == \"\" {\n\t\tenv.TempDir = env.pathJoin(env.WorkDir, \".tmp\")\n\t}\n\n\treturn env\n}\n\n// shellCmd returns the command slice to run a script through the appropriate shell.\nfunc (e *osEnv) shellCmd(script string) []string {\n\tshell := strings.ToLower(e.Shell)\n\tswitch shell {\n\tcase \"pwsh\":\n\t\treturn []string{\"pwsh\", \"-NoProfile\", \"-Command\", script}\n\tcase \"powershell\":\n\t\treturn []string{\"powershell\", \"-NoProfile\", \"-Command\", script}\n\tcase \"cmd.exe\", \"cmd\":\n\t\treturn []string{\"cmd.exe\", \"/C\", script}\n\tdefault:\n\t\treturn []string{\"bash\", \"-c\", script}\n\t}\n}\n\n// mkdirCmd returns a shell command string to create a directory (with parents).\nfunc (e *osEnv) mkdirCmd(dir string) string {\n\tif e.isWindows() {\n\t\treturn fmt.Sprintf(`if (!(Test-Path '%s')) { New-Item -ItemType Directory -Path '%s' -Force | Out-Null }`, dir, dir)\n\t}\n\treturn fmt.Sprintf(\"mkdir -p %s\", dir)\n}\n\n// listDirCmd returns a command slice to list directory contents.\nfunc (e *osEnv) listDirCmd(dir string) []string {\n\tif e.isWindows() {\n\t\treturn e.shellCmd(fmt.Sprintf(\"Get-ChildItem -Name '%s'\", dir))\n\t}\n\treturn []string{\"ls\", dir}\n}\n\n// killProcessCmd returns a command slice to kill processes matching a pattern.\n// On Windows, uses taskkill /T to kill the entire process tree, which handles\n// child processes (chrome.exe, python3, etc.) that Claude CLI may have spawned.\nfunc (e *osEnv) killProcessCmd(pattern string) []string {\n\tif e.isWindows() {\n\t\t// taskkill /F /T kills the process tree; fall back to Stop-Process.\n\t\tscript := fmt.Sprintf(\n\t\t\t\"Get-Process -ErrorAction SilentlyContinue | Where-Object {$_.ProcessName -like '*%s*'} | ForEach-Object { taskkill /F /T /PID $_.Id 2>$null }; \"+\n\t\t\t\t\"Get-Process -ErrorAction SilentlyContinue | Where-Object {$_.ProcessName -like '*%s*'} | Stop-Process -Force -ErrorAction SilentlyContinue\",\n\t\t\tpattern, pattern)\n\t\treturn e.shellCmd(script)\n\t}\n\treturn []string{\"sh\", \"-c\", fmt.Sprintf(\"pkill -f '%s' || true\", pattern)}\n}\n\n// rootDir returns the filesystem root for the target OS.\nfunc (e *osEnv) rootDir() string {\n\tif e.isWindows() {\n\t\treturn `C:\\`\n\t}\n\treturn \"/\"\n}\n\n// pathJoin joins path segments using the appropriate separator.\nfunc (e *osEnv) pathJoin(parts ...string) string {\n\tif e.isWindows() {\n\t\treturn strings.Join(parts, `\\`)\n\t}\n\treturn path.Join(parts...)\n}\n\n// buildCLIScript builds the complete CLI invocation script for the target OS.\n// Returns (script, stdin) — on Linux stdin is nil (heredoc handles it),\n// on Windows stdin contains inputJSONL bytes to pass via gRPC Stdin.\nfunc (e *osEnv) buildCLIScript(args []string, systemPrompt, inputJSONL string) (string, []byte) {\n\tworkDir := e.WorkDir\n\tpromptFile := e.pathJoin(workDir, \".yao\", \".system-prompt.txt\")\n\n\tif e.isWindows() {\n\t\treturn e.buildPowerShellScript(args, systemPrompt, inputJSONL, workDir, promptFile)\n\t}\n\treturn e.buildBashScript(args, systemPrompt, inputJSONL, workDir, promptFile), nil\n}\n\nfunc (e *osEnv) buildBashScript(args []string, systemPrompt, inputJSONL, workDir, promptFile string) string {\n\tvar b strings.Builder\n\n\tif e.UserHome != \"\" {\n\t\tb.WriteString(fmt.Sprintf(\"touch %s/.Xauthority 2>/dev/null; \", e.UserHome))\n\t}\n\tb.WriteString(\"touch \\\"$HOME/.Xauthority\\\" 2>/dev/null\\n\")\n\n\tif systemPrompt != \"\" {\n\t\tb.WriteString(fmt.Sprintf(\"mkdir -p %s/.yao\\n\", workDir))\n\t\tb.WriteString(fmt.Sprintf(\"cat << 'PROMPTEOF' > %s\\n\", promptFile))\n\t\tb.WriteString(systemPrompt)\n\t\tb.WriteString(\"\\nPROMPTEOF\\n\")\n\t\targs = append(args, \"--append-system-prompt-file\", promptFile)\n\t}\n\n\tb.WriteString(\"cat << 'INPUTEOF' | claude -p\")\n\tfor _, arg := range args {\n\t\tb.WriteString(fmt.Sprintf(\" %q\", arg))\n\t}\n\tb.WriteString(\" 2>&1\\n\")\n\tb.WriteString(inputJSONL)\n\tb.WriteString(\"\\nINPUTEOF\")\n\n\treturn b.String()\n}\n\n// buildPowerShellScript builds a script that writes the system prompt file,\n// then launches claude -p. inputJSONL is returned as stdin bytes to be passed\n// directly via gRPC, bypassing PowerShell's encoding entirely.\nfunc (e *osEnv) buildPowerShellScript(args []string, systemPrompt, inputJSONL, workDir, promptFile string) (string, []byte) {\n\tvar b strings.Builder\n\tnoBOM := \"(New-Object System.Text.UTF8Encoding $false)\"\n\n\t// Force UTF-8 for both input and output streams.\n\t// On CJK Windows the default codepage is often GBK/GB2312 (936)\n\t// which corrupts Claude CLI's UTF-8 JSON output.\n\tb.WriteString(\"[Console]::InputEncoding = [System.Text.Encoding]::UTF8\\n\")\n\tb.WriteString(\"[Console]::OutputEncoding = [System.Text.Encoding]::UTF8\\n\")\n\tb.WriteString(\"$OutputEncoding = [System.Text.Encoding]::UTF8\\n\")\n\n\t// Ensure claude.exe can be found even when Tai runs as a different user.\n\t// Claude CLI is typically installed per-user (e.g. C:\\Users\\X\\.local\\bin)\n\t// which isn't in the PATH when Tai runs as a service or another account.\n\t// Scan all user profiles for common install locations.\n\tb.WriteString(\"foreach ($d in (Get-ChildItem 'C:\\\\Users' -Directory -ErrorAction SilentlyContinue)) {\\n\")\n\tb.WriteString(\" $p = Join-Path $d.FullName '.local\\\\bin'\\n\")\n\tb.WriteString(\" if (Test-Path (Join-Path $p 'claude.exe')) { $env:PATH = \\\"$p;$env:PATH\\\"; break }\\n\")\n\tb.WriteString(\"}\\n\")\n\tb.WriteString(\"if ($env:APPDATA) { $env:PATH = \\\"$env:APPDATA\\\\npm;$env:PATH\\\" }\\n\")\n\n\tyaoDir := e.pathJoin(workDir, \".yao\")\n\tb.WriteString(fmt.Sprintf(\"if (!(Test-Path '%s')) { New-Item -ItemType Directory -Path '%s' -Force | Out-Null }\\n\", yaoDir, yaoDir))\n\n\tif systemPrompt != \"\" {\n\t\tescaped := strings.ReplaceAll(systemPrompt, \"'\", \"''\")\n\t\tb.WriteString(fmt.Sprintf(\"[IO.File]::WriteAllText('%s', @'\\n%s\\n'@, %s)\\n\", promptFile, escaped, noBOM))\n\t\targs = append(args, \"--append-system-prompt-file\", promptFile)\n\t}\n\n\tb.WriteString(\"claude -p\")\n\tfor _, arg := range args {\n\t\tb.WriteString(fmt.Sprintf(\" '%s'\", strings.ReplaceAll(arg, \"'\", \"''\")))\n\t}\n\n\treturn b.String(), []byte(inputJSONL + \"\\n\")\n}\n" }, { "path": "agent/sandbox/v2/claude/parse.go", "content": "package claude\n\nimport (\n\t\"bufio\"\n\t\"context\"\n\t\"encoding/json\"\n\t\"errors\"\n\t\"fmt\"\n\t\"io\"\n\t\"log\"\n\t\"strings\"\n\t\"time\"\n\n\tgoujson \"github.com/yaoapp/gou/json\"\n\tagentContext \"github.com/yaoapp/yao/agent/context\"\n\t\"github.com/yaoapp/yao/agent/output/message\"\n)\n\n// errStreamCompleted is a sentinel indicating the parser received the terminal\n// \"result\" message. It is NOT a real error — callers should treat it as\n// successful completion of the stream.\nvar errStreamCompleted = errors.New(\"claude stream completed\")\n\n// parseStreamJSON reads stream-json lines from Claude CLI stdout and\n// pushes them through handler as standard StreamChunkType events.\nfunc parseStreamJSON(ctx context.Context, stdout io.ReadCloser, handler message.StreamFunc) error {\n\t// When the context is cancelled (upstream timeout / interrupt), close\n\t// stdout so that scanner.Scan() unblocks immediately. Without this,\n\t// a failed TerminateProcess (Access is denied) would leave us stuck\n\t// forever on the read.\n\tdoneParsing := make(chan struct{})\n\tdefer close(doneParsing)\n\tgo func() {\n\t\tselect {\n\t\tcase <-ctx.Done():\n\t\t\tstdout.Close()\n\t\tcase <-doneParsing:\n\t\t}\n\t}()\n\n\tscanner := bufio.NewScanner(stdout)\n\tbuf := make([]byte, 0, 64*1024)\n\tscanner.Buffer(buf, 1024*1024)\n\n\tmessageStarted := false\n\ttoolBlockActive := false\n\ttoolIndex := 0\n\n\ttype toolState struct {\n\t\tid string\n\t\tname string\n\t\tindex int\n\t\tinputJSON strings.Builder\n\t}\n\tvar currentTool *toolState\n\n\tfor scanner.Scan() {\n\t\tline := scanner.Text()\n\t\tif line == \"\" {\n\t\t\tcontinue\n\t\t}\n\n\t\tvar msg map[string]any\n\t\tif err := json.Unmarshal([]byte(line), &msg); err != nil {\n\t\t\tcontinue\n\t\t}\n\n\t\tmsgType, _ := msg[\"type\"].(string)\n\t\tstopped := false\n\n\t\tswitch msgType {\n\t\tcase \"system\":\n\t\t\tif handler != nil {\n\t\t\t\tdata, _ := json.Marshal(msg)\n\t\t\t\tif handler(message.ChunkMetadata, data) != 0 {\n\t\t\t\t\tstopped = true\n\t\t\t\t}\n\t\t\t}\n\n\t\tcase \"stream_event\":\n\t\t\tevent, _ := msg[\"event\"].(map[string]any)\n\t\t\tif event == nil {\n\t\t\t\tcontinue\n\t\t\t}\n\t\t\teventType, _ := event[\"type\"].(string)\n\n\t\t\tswitch eventType {\n\t\t\tcase \"content_block_start\":\n\t\t\t\tif cb, ok := event[\"content_block\"].(map[string]any); ok {\n\t\t\t\t\tblockType, _ := cb[\"type\"].(string)\n\t\t\t\t\tif blockType == \"tool_use\" {\n\t\t\t\t\t\ttoolName, _ := cb[\"name\"].(string)\n\t\t\t\t\t\ttoolID, _ := cb[\"id\"].(string)\n\t\t\t\t\t\tif toolID == \"\" {\n\t\t\t\t\t\t\ttoolID = fmt.Sprintf(\"tool_%d_%d\", toolIndex, time.Now().UnixNano())\n\t\t\t\t\t\t}\n\t\t\t\t\t\tcurrentTool = &toolState{id: toolID, name: toolName, index: toolIndex}\n\t\t\t\t\t\ttoolIndex++\n\n\t\t\t\t\t\tif handler != nil {\n\t\t\t\t\t\t\tif messageStarted {\n\t\t\t\t\t\t\t\thandler(message.ChunkMessageEnd, nil)\n\t\t\t\t\t\t\t\tmessageStarted = false\n\t\t\t\t\t\t\t}\n\t\t\t\t\t\t\tif !toolBlockActive {\n\t\t\t\t\t\t\t\tstartData := message.EventMessageStartData{\n\t\t\t\t\t\t\t\t\tMessageID: fmt.Sprintf(\"sandbox-tool-%d\", time.Now().UnixNano()),\n\t\t\t\t\t\t\t\t\tType: \"tool_call\",\n\t\t\t\t\t\t\t\t\tTimestamp: time.Now().UnixMilli(),\n\t\t\t\t\t\t\t\t}\n\t\t\t\t\t\t\t\tsd, _ := json.Marshal(startData)\n\t\t\t\t\t\t\t\tif handler(message.ChunkMessageStart, sd) != 0 {\n\t\t\t\t\t\t\t\t\tstopped = true\n\t\t\t\t\t\t\t\t\tbreak\n\t\t\t\t\t\t\t\t}\n\t\t\t\t\t\t\t\ttoolBlockActive = true\n\t\t\t\t\t\t\t}\n\t\t\t\t\t\t\ttcData, _ := json.Marshal([]map[string]any{{\n\t\t\t\t\t\t\t\t\"index\": currentTool.index,\n\t\t\t\t\t\t\t\t\"id\": currentTool.id,\n\t\t\t\t\t\t\t\t\"type\": \"function\",\n\t\t\t\t\t\t\t\t\"function\": map[string]any{\n\t\t\t\t\t\t\t\t\t\"name\": toolName,\n\t\t\t\t\t\t\t\t\t\"arguments\": \"\",\n\t\t\t\t\t\t\t\t},\n\t\t\t\t\t\t\t}})\n\t\t\t\t\t\t\tif handler(message.ChunkToolCall, tcData) != 0 {\n\t\t\t\t\t\t\t\tstopped = true\n\t\t\t\t\t\t\t}\n\t\t\t\t\t\t}\n\t\t\t\t\t}\n\t\t\t\t}\n\n\t\t\tcase \"content_block_delta\":\n\t\t\t\tif delta, ok := event[\"delta\"].(map[string]any); ok {\n\t\t\t\t\tdeltaType, _ := delta[\"type\"].(string)\n\t\t\t\t\tswitch deltaType {\n\t\t\t\t\tcase \"text_delta\":\n\t\t\t\t\t\tif text, ok := delta[\"text\"].(string); ok && text != \"\" {\n\t\t\t\t\t\t\ttext = strings.ReplaceAll(text, \"\\r\\n\", \"\\n\")\n\t\t\t\t\t\t\ttext = strings.ReplaceAll(text, \"\\r\", \"\\n\")\n\t\t\t\t\t\t\tif handler != nil {\n\t\t\t\t\t\t\t\tif toolBlockActive {\n\t\t\t\t\t\t\t\t\thandler(message.ChunkMessageEnd, nil)\n\t\t\t\t\t\t\t\t\ttoolBlockActive = false\n\t\t\t\t\t\t\t\t\tmessageStarted = false\n\t\t\t\t\t\t\t\t}\n\t\t\t\t\t\t\t\tif !messageStarted {\n\t\t\t\t\t\t\t\t\tstartData := message.EventMessageStartData{\n\t\t\t\t\t\t\t\t\t\tMessageID: fmt.Sprintf(\"sandbox-%d\", time.Now().UnixNano()),\n\t\t\t\t\t\t\t\t\t\tType: \"text\",\n\t\t\t\t\t\t\t\t\t\tTimestamp: time.Now().UnixMilli(),\n\t\t\t\t\t\t\t\t\t}\n\t\t\t\t\t\t\t\t\tsd, _ := json.Marshal(startData)\n\t\t\t\t\t\t\t\t\tif handler(message.ChunkMessageStart, sd) != 0 {\n\t\t\t\t\t\t\t\t\t\tstopped = true\n\t\t\t\t\t\t\t\t\t\tbreak\n\t\t\t\t\t\t\t\t\t}\n\t\t\t\t\t\t\t\t\tmessageStarted = true\n\t\t\t\t\t\t\t\t}\n\t\t\t\t\t\t\t\tif handler(message.ChunkText, []byte(text)) != 0 {\n\t\t\t\t\t\t\t\t\tstopped = true\n\t\t\t\t\t\t\t\t}\n\t\t\t\t\t\t\t}\n\t\t\t\t\t\t}\n\t\t\t\t\tcase \"input_json_delta\":\n\t\t\t\t\t\tif currentTool != nil {\n\t\t\t\t\t\t\tif partial, ok := delta[\"partial_json\"].(string); ok {\n\t\t\t\t\t\t\t\tcurrentTool.inputJSON.WriteString(partial)\n\t\t\t\t\t\t\t\tif handler != nil {\n\t\t\t\t\t\t\t\t\ttcData, _ := json.Marshal([]map[string]any{{\n\t\t\t\t\t\t\t\t\t\t\"index\": currentTool.index,\n\t\t\t\t\t\t\t\t\t\t\"function\": map[string]any{\n\t\t\t\t\t\t\t\t\t\t\t\"arguments\": partial,\n\t\t\t\t\t\t\t\t\t\t},\n\t\t\t\t\t\t\t\t\t}})\n\t\t\t\t\t\t\t\t\tif handler(message.ChunkToolCall, tcData) != 0 {\n\t\t\t\t\t\t\t\t\t\tstopped = true\n\t\t\t\t\t\t\t\t\t}\n\t\t\t\t\t\t\t\t}\n\t\t\t\t\t\t\t}\n\t\t\t\t\t\t}\n\t\t\t\t\t}\n\t\t\t\t}\n\n\t\t\tcase \"content_block_stop\":\n\t\t\t\tcurrentTool = nil\n\t\t\t}\n\n\t\tcase \"assistant\":\n\t\t\tif msgData, ok := msg[\"message\"].(map[string]any); ok {\n\t\t\t\tstopReason, _ := msgData[\"stop_reason\"].(string)\n\t\t\t\tif stopReason != \"\" {\n\t\t\t\t\tif contentArr, ok := msgData[\"content\"].([]any); ok {\n\t\t\t\t\t\tfor _, item := range contentArr {\n\t\t\t\t\t\t\tci, ok := item.(map[string]any)\n\t\t\t\t\t\t\tif !ok {\n\t\t\t\t\t\t\t\tcontinue\n\t\t\t\t\t\t\t}\n\t\t\t\t\t\t\titemType, _ := ci[\"type\"].(string)\n\n\t\t\t\t\t\t\tif itemType == \"tool_use\" && handler != nil {\n\t\t\t\t\t\t\t\ttoolName, _ := ci[\"name\"].(string)\n\t\t\t\t\t\t\t\ttoolID, _ := ci[\"id\"].(string)\n\t\t\t\t\t\t\t\tif toolID == \"\" {\n\t\t\t\t\t\t\t\t\ttoolID = fmt.Sprintf(\"tool_%d_%d\", toolIndex, time.Now().UnixNano())\n\t\t\t\t\t\t\t\t}\n\t\t\t\t\t\t\t\tinputRaw, _ := json.Marshal(ci[\"input\"])\n\t\t\t\t\t\t\t\tidx := toolIndex\n\t\t\t\t\t\t\t\ttoolIndex++\n\n\t\t\t\t\t\t\t\tif !toolBlockActive {\n\t\t\t\t\t\t\t\t\tstartData := message.EventMessageStartData{\n\t\t\t\t\t\t\t\t\t\tMessageID: fmt.Sprintf(\"sandbox-tool-%d\", time.Now().UnixNano()),\n\t\t\t\t\t\t\t\t\t\tType: \"tool_call\",\n\t\t\t\t\t\t\t\t\t\tTimestamp: time.Now().UnixMilli(),\n\t\t\t\t\t\t\t\t\t}\n\t\t\t\t\t\t\t\t\tsd, _ := json.Marshal(startData)\n\t\t\t\t\t\t\t\t\tif handler(message.ChunkMessageStart, sd) != 0 {\n\t\t\t\t\t\t\t\t\t\tstopped = true\n\t\t\t\t\t\t\t\t\t\tbreak\n\t\t\t\t\t\t\t\t\t}\n\t\t\t\t\t\t\t\t\ttoolBlockActive = true\n\t\t\t\t\t\t\t\t}\n\t\t\t\t\t\t\t\ttcData, _ := json.Marshal([]map[string]any{{\n\t\t\t\t\t\t\t\t\t\"index\": idx,\n\t\t\t\t\t\t\t\t\t\"id\": toolID,\n\t\t\t\t\t\t\t\t\t\"type\": \"function\",\n\t\t\t\t\t\t\t\t\t\"function\": map[string]any{\n\t\t\t\t\t\t\t\t\t\t\"name\": toolName,\n\t\t\t\t\t\t\t\t\t\t\"arguments\": string(inputRaw),\n\t\t\t\t\t\t\t\t\t},\n\t\t\t\t\t\t\t\t}})\n\t\t\t\t\t\t\t\tif handler(message.ChunkToolCall, tcData) != 0 {\n\t\t\t\t\t\t\t\t\tstopped = true\n\t\t\t\t\t\t\t\t\tbreak\n\t\t\t\t\t\t\t\t}\n\t\t\t\t\t\t\t}\n\n\t\t\t\t\t\t\tif itemType == \"text\" {\n\t\t\t\t\t\t\t\tif text, ok := ci[\"text\"].(string); ok && text != \"\" && handler != nil && !messageStarted {\n\t\t\t\t\t\t\t\t\ttext = strings.ReplaceAll(text, \"\\r\\n\", \"\\n\")\n\t\t\t\t\t\t\t\t\ttext = strings.ReplaceAll(text, \"\\r\", \"\\n\")\n\t\t\t\t\t\t\t\t\tif toolBlockActive {\n\t\t\t\t\t\t\t\t\t\thandler(message.ChunkMessageEnd, nil)\n\t\t\t\t\t\t\t\t\t\ttoolBlockActive = false\n\t\t\t\t\t\t\t\t\t}\n\t\t\t\t\t\t\t\t\tstartData := message.EventMessageStartData{\n\t\t\t\t\t\t\t\t\t\tMessageID: fmt.Sprintf(\"sandbox-%d\", time.Now().UnixNano()),\n\t\t\t\t\t\t\t\t\t\tType: \"text\",\n\t\t\t\t\t\t\t\t\t\tTimestamp: time.Now().UnixMilli(),\n\t\t\t\t\t\t\t\t\t}\n\t\t\t\t\t\t\t\t\tsd, _ := json.Marshal(startData)\n\t\t\t\t\t\t\t\t\tif handler(message.ChunkMessageStart, sd) != 0 {\n\t\t\t\t\t\t\t\t\t\tstopped = true\n\t\t\t\t\t\t\t\t\t\tbreak\n\t\t\t\t\t\t\t\t\t}\n\t\t\t\t\t\t\t\t\tif handler(message.ChunkText, []byte(text)) != 0 {\n\t\t\t\t\t\t\t\t\t\tstopped = true\n\t\t\t\t\t\t\t\t\t\tbreak\n\t\t\t\t\t\t\t\t\t}\n\t\t\t\t\t\t\t\t\tmessageStarted = true\n\t\t\t\t\t\t\t\t}\n\t\t\t\t\t\t\t}\n\t\t\t\t\t\t}\n\t\t\t\t\t}\n\n\t\t\t\t\t// Close any open message from the streaming phase.\n\t\t\t\t\t// stream_event text_deltas set messageStarted=true but\n\t\t\t\t\t// nothing resets it when the turn ends — the assistant\n\t\t\t\t\t// message marks the turn boundary, so we must close\n\t\t\t\t\t// the message here to keep state in sync with the\n\t\t\t\t\t// stream handler (which already sent message_end).\n\t\t\t\t\tif handler != nil {\n\t\t\t\t\t\tif toolBlockActive {\n\t\t\t\t\t\t\thandler(message.ChunkMessageEnd, nil)\n\t\t\t\t\t\t\ttoolBlockActive = false\n\t\t\t\t\t\t}\n\t\t\t\t\t\tif messageStarted {\n\t\t\t\t\t\t\thandler(message.ChunkMessageEnd, nil)\n\t\t\t\t\t\t\tmessageStarted = false\n\t\t\t\t\t\t}\n\t\t\t\t\t}\n\t\t\t\t}\n\t\t\t}\n\n\t\tcase \"result\":\n\t\t\tisError, _ := msg[\"is_error\"].(bool)\n\t\t\tif isError {\n\t\t\t\tif result, ok := msg[\"result\"].(string); ok {\n\t\t\t\t\tif handler != nil {\n\t\t\t\t\t\thandler(message.ChunkError, []byte(result))\n\t\t\t\t\t}\n\t\t\t\t\treturn fmt.Errorf(\"Claude CLI error: %s\", result)\n\t\t\t\t}\n\t\t\t}\n\t\t\tif handler != nil {\n\t\t\t\tif toolBlockActive {\n\t\t\t\t\thandler(message.ChunkMessageEnd, nil)\n\t\t\t\t\ttoolBlockActive = false\n\t\t\t\t}\n\t\t\t\tif messageStarted {\n\t\t\t\t\thandler(message.ChunkMessageEnd, nil)\n\t\t\t\t}\n\t\t\t}\n\t\t\t// \"result\" is the terminal message in Claude CLI's stream-json\n\t\t\t// protocol. Return immediately instead of continuing to\n\t\t\t// scanner.Scan(), which would block forever if the process\n\t\t\t// stays alive (e.g. child processes like chrome.exe keep the\n\t\t\t// stdout pipe open).\n\t\t\treturn errStreamCompleted\n\n\t\tcase \"error\":\n\t\t\tvar errMsg string\n\t\t\tswitch e := msg[\"error\"].(type) {\n\t\t\tcase string:\n\t\t\t\terrMsg = e\n\t\t\tcase map[string]any:\n\t\t\t\terrMsg, _ = e[\"message\"].(string)\n\t\t\t}\n\t\t\tif errMsg != \"\" {\n\t\t\t\tif handler != nil {\n\t\t\t\t\thandler(message.ChunkError, []byte(errMsg))\n\t\t\t\t}\n\t\t\t\treturn fmt.Errorf(\"Claude CLI error: %s\", errMsg)\n\t\t\t}\n\t\t}\n\n\t\tif stopped {\n\t\t\tbreak\n\t\t}\n\t}\n\n\tif err := scanner.Err(); err != nil {\n\t\t// If the context was cancelled (upstream timeout / interrupt), the\n\t\t// stdout pipe was closed by the goroutine above. The resulting\n\t\t// read error is expected — surface it as context.Canceled so the\n\t\t// caller can handle it uniformly.\n\t\tif ctx.Err() != nil {\n\t\t\treturn ctx.Err()\n\t\t}\n\t\treturn err\n\t}\n\treturn nil\n}\n\n// buildFirstRequestJSONL builds JSONL with all messages for the first request.\nfunc buildFirstRequestJSONL(messages []agentContext.Message) string {\n\tvar lines []string\n\tfor _, msg := range messages {\n\t\tif msg.Role == \"system\" {\n\t\t\tcontinue\n\t\t}\n\t\tcontent := msg.Content\n\t\tif content == nil {\n\t\t\tcontent = \"\"\n\t\t}\n\t\tstreamMsg := map[string]any{\n\t\t\t\"type\": string(msg.Role),\n\t\t\t\"message\": map[string]any{\n\t\t\t\t\"role\": string(msg.Role),\n\t\t\t\t\"content\": content,\n\t\t\t},\n\t\t}\n\t\tdata, _ := json.Marshal(streamMsg)\n\t\tlines = append(lines, string(data))\n\t}\n\treturn strings.Join(lines, \"\\n\")\n}\n\n// buildLastUserMessageJSONL builds JSONL with only the last user message.\nfunc buildLastUserMessageJSONL(messages []agentContext.Message) string {\n\tfor i := len(messages) - 1; i >= 0; i-- {\n\t\tif messages[i].Role == \"user\" {\n\t\t\tcontent := messages[i].Content\n\t\t\tif content == nil {\n\t\t\t\tcontent = \"\"\n\t\t\t}\n\t\t\tmsg := map[string]any{\n\t\t\t\t\"type\": \"user\",\n\t\t\t\t\"message\": map[string]any{\n\t\t\t\t\t\"role\": \"user\",\n\t\t\t\t\t\"content\": content,\n\t\t\t\t},\n\t\t\t}\n\t\t\tdata, _ := json.Marshal(msg)\n\t\t\treturn string(data)\n\t\t}\n\t}\n\treturn \"\"\n}\n\n// Suppress unused import warnings — goujson.Parse is used for tool description\n// parsing in V1 and will be used for detailed tool descriptions in future.\nvar _ = goujson.Parse\nvar _ = log.Printf\n" }, { "path": "agent/sandbox/v2/claude/runner.go", "content": "package claude\n\nimport (\n\t\"context\"\n\t\"encoding/json\"\n\t\"errors\"\n\t\"fmt\"\n\t\"io\"\n\t\"os\"\n\t\"path\"\n\t\"strings\"\n\t\"time\"\n\n\t\"github.com/yaoapp/gou/connector\"\n\t\"github.com/yaoapp/yao/agent/output/message\"\n\t\"github.com/yaoapp/yao/agent/sandbox/v2/types\"\n\tinfra \"github.com/yaoapp/yao/sandbox/v2\"\n)\n\nconst defaultProxyPort = 3456\n\n// ClaudeRunner implements the Runner interface for Claude CLI (mode=cli).\ntype ClaudeRunner struct {\n\tmode string\n\thasMCP bool\n\tmcpToolPattern string // e.g. \"mcp__yao__*,mcp__github__*\"\n\tservicePort int\n\tservicePath string\n\tserviceProtocol string\n\tstreamCompleted bool // set when Stream received \"result\"; Cleanup skips kill\n}\n\n// New creates a new ClaudeRunner.\nfunc New() *ClaudeRunner {\n\treturn &ClaudeRunner{mode: \"cli\"}\n}\n\nfunc (r *ClaudeRunner) Name() string { return \"claude\" }\n\n// Prepare executes user-defined and runner-specific prepare steps.\nfunc (r *ClaudeRunner) Prepare(ctx context.Context, req *types.PrepareRequest) error {\n\tr.mode = req.Config.Runner.Mode\n\tif r.mode == \"\" {\n\t\tr.mode = \"cli\"\n\t}\n\n\tsteps := append([]types.PrepareStep{}, req.Config.Prepare...)\n\n\tif req.SkillsDir != \"\" {\n\t\tws := req.Computer.Workplace()\n\t\tif ws != nil {\n\t\t\tsrc := \"local:///\" + req.SkillsDir\n\t\t\tdst := \".claude/skills\"\n\t\t\tif _, err := ws.Copy(src, dst); err != nil {\n\t\t\t\tfmt.Fprintf(os.Stderr, \"[claude] warn: copy skills %s -> %s: %v\\n\", src, dst, err)\n\t\t\t}\n\t\t}\n\t}\n\n\tif len(req.MCPServers) > 0 {\n\t\tr.hasMCP = true\n\t\tr.mcpToolPattern = buildMCPAllowedTools(req.MCPServers)\n\t\tmcpJSON := buildMCPConfig(req.MCPServers)\n\t\tsteps = append(steps, types.PrepareStep{\n\t\t\tAction: \"file\",\n\t\t\tPath: \".claude/mcp.json\",\n\t\t\tContent: mcpJSON,\n\t\t})\n\t}\n\n\tif req.RunSteps != nil && len(steps) > 0 {\n\t\tif err := req.RunSteps(ctx, steps, req.Computer, req.Config.ID, req.ConfigHash, req.AssistantDir); err != nil {\n\t\t\treturn fmt.Errorf(\"claude prepare steps: %w\", err)\n\t\t}\n\t}\n\n\treturn nil\n}\n\n// Stream executes the Claude CLI and streams output to handler.\nfunc (r *ClaudeRunner) Stream(ctx context.Context, req *types.StreamRequest, handler message.StreamFunc) error {\n\tcomputer := req.Computer\n\tif computer == nil {\n\t\treturn fmt.Errorf(\"computer is nil\")\n\t}\n\n\toe := resolveOSEnv(computer, req.Config)\n\n\tif req.ChatID != \"\" {\n\t\tws := computer.Workplace()\n\t\tif ws != nil {\n\t\t\tprocessed, err := prepareAttachments(ctx, req.Messages, req.ChatID, ws)\n\t\t\tif err != nil {\n\t\t\t\treturn fmt.Errorf(\"prepareAttachments: %w\", err)\n\t\t\t}\n\t\t\treq.Messages = processed\n\t\t}\n\t}\n\n\tisContinuation := hasExistingSession(ctx, computer, oe)\n\n\tcmd, env, stdin := r.buildCLICommand(req, oe, isContinuation)\n\n\tstreamOpts := []infra.ExecOption{infra.WithWorkDir(oe.WorkDir), infra.WithEnv(env)}\n\tif len(stdin) > 0 {\n\t\tstreamOpts = append(streamOpts, infra.WithStdin(stdin))\n\t}\n\n\tfmt.Fprintf(os.Stderr, \"[claude] Stream cmd=%v hasMCP=%v isContinuation=%v stdinLen=%d workDir=%q\\n\", cmd, r.hasMCP, isContinuation, len(stdin), oe.WorkDir)\n\n\texecStream, err := computer.Stream(ctx, cmd, streamOpts...)\n\tif err != nil {\n\t\treturn fmt.Errorf(\"computer.Stream: %w\", err)\n\t}\n\n\tstreamCtx, streamCancel := context.WithCancel(ctx)\n\tdefer streamCancel()\n\n\t// Kill claude processes only when the context is cancelled externally\n\t// (upstream timeout, user interrupt) — NOT on normal return.\n\tgo func() {\n\t\t<-streamCtx.Done()\n\t\tif ctx.Err() == nil {\n\t\t\tfmt.Fprintf(os.Stderr, \"[claude] streamCtx done: normal return, skipping kill (ctx.Err=nil)\\n\")\n\t\t\treturn\n\t\t}\n\t\tfmt.Fprintf(os.Stderr, \"[claude] streamCtx done: context cancelled externally (ctx.Err=%v), killing processes\\n\", ctx.Err())\n\t\tkillCtx, cancel := context.WithTimeout(context.Background(), 5*time.Second)\n\t\tdefer cancel()\n\t\tcomputer.Exec(killCtx, oe.killProcessCmd(\"claude\"))\n\t\texecStream.Cancel()\n\t}()\n\n\tvar stderrBuf strings.Builder\n\tgo func() {\n\t\tbuf := make([]byte, 4096)\n\t\tfor {\n\t\t\tn, err := execStream.Stderr.Read(buf)\n\t\t\tif n > 0 {\n\t\t\t\tstderrBuf.Write(buf[:n])\n\t\t\t\tchunk := string(buf[:n])\n\t\t\t\tif strings.Contains(strings.ToLower(chunk), \"error\") {\n\t\t\t\t\tstreamCancel()\n\t\t\t\t\tio.Copy(&stderrBuf, execStream.Stderr)\n\t\t\t\t\treturn\n\t\t\t\t}\n\t\t\t}\n\t\t\tif err != nil {\n\t\t\t\treturn\n\t\t\t}\n\t\t}\n\t}()\n\n\tparseErr := parseStreamJSON(streamCtx, execStream.Stdout, handler)\n\tfmt.Fprintf(os.Stderr, \"[claude] parseStreamJSON returned: %v\\n\", parseErr)\n\n\t// Received \"result\" — Claude finished normally. Return immediately.\n\tif errors.Is(parseErr, errStreamCompleted) {\n\t\tr.streamCompleted = true\n\t\tfmt.Fprintf(os.Stderr, \"[claude] stream completed normally, returning nil\\n\")\n\t\treturn nil\n\t}\n\n\t// Parse failed or stream ended without \"result\" — wait for process.\n\tfmt.Fprintf(os.Stderr, \"[claude] stream did NOT complete normally, waiting for process exit...\\n\")\n\texitCode, waitErr := execStream.Wait()\n\tstderrStr := strings.TrimSpace(stderrBuf.String())\n\n\tif parseErr != nil {\n\t\tif stderrStr != \"\" {\n\t\t\treturn fmt.Errorf(\"%w (stderr: %s)\", parseErr, stderrStr)\n\t\t}\n\t\treturn parseErr\n\t}\n\tif waitErr != nil {\n\t\tif stderrStr != \"\" {\n\t\t\treturn fmt.Errorf(\"%w (stderr: %s)\", waitErr, stderrStr)\n\t\t}\n\t\treturn waitErr\n\t}\n\tif exitCode != 0 {\n\t\tfmt.Fprintf(os.Stderr, \"[claude] exit code=%d stderr=%q\\n\", exitCode, stderrStr)\n\t\tif stderrStr != \"\" {\n\t\t\treturn fmt.Errorf(\"claude CLI exited with code %d: %s\", exitCode, stderrStr)\n\t\t}\n\t\treturn fmt.Errorf(\"claude CLI exited with code %d\", exitCode)\n\t}\n\treturn nil\n}\n\n// Cleanup kills any remaining claude processes. If the stream completed\n// normally (received \"result\"), child processes are preserved — the user\n// may have asked Claude to launch a browser, server, etc.\nfunc (r *ClaudeRunner) Cleanup(ctx context.Context, computer infra.Computer) error {\n\tif computer == nil {\n\t\treturn nil\n\t}\n\n\tif r.streamCompleted {\n\t\tfmt.Fprintf(os.Stderr, \"[claude] cleanup: stream completed normally, skipping process kill (child processes preserved)\\n\")\n\t\treturn nil\n\t}\n\n\tif r.mode != \"service\" {\n\t\toe := resolveOSEnv(computer, nil)\n\t\tcomputer.Exec(ctx, oe.killProcessCmd(\"claude\"))\n\t}\n\n\treturn nil\n}\n\n// hasExistingSession checks if a Claude CLI session exists in the workspace.\nfunc hasExistingSession(ctx context.Context, computer infra.Computer, oe *osEnv) bool {\n\tsessionDir := oe.pathJoin(oe.WorkDir, \".claude\", \"projects\")\n\tresult, err := computer.Exec(ctx, oe.listDirCmd(sessionDir))\n\tif err != nil || result.ExitCode != 0 {\n\t\treturn false\n\t}\n\treturn strings.TrimSpace(result.Stdout) != \"\"\n}\n\n// buildCLICommand constructs the Claude CLI command, environment variables, and optional stdin bytes.\nfunc (r *ClaudeRunner) buildCLICommand(req *types.StreamRequest, oe *osEnv, isContinuation bool) ([]string, map[string]string, []byte) {\n\tenv := make(map[string]string)\n\n\tif oe.isWindows() {\n\t\tenv[\"USERPROFILE\"] = oe.WorkDir\n\t\tif len(oe.WorkDir) >= 2 && oe.WorkDir[1] == ':' {\n\t\t\tenv[\"HOMEDRIVE\"] = oe.WorkDir[:2]\n\t\t\tenv[\"HOMEPATH\"] = oe.WorkDir[2:]\n\t\t}\n\t} else {\n\t\tenv[\"HOME\"] = oe.WorkDir\n\t\tif oe.UserHome != \"\" {\n\t\t\tenv[\"XAUTHORITY\"] = path.Join(oe.UserHome, \".Xauthority\")\n\t\t}\n\t}\n\n\tif req.Connector != nil {\n\t\tsetting := req.Connector.Setting()\n\t\thost, _ := setting[\"host\"].(string)\n\t\tkey, _ := setting[\"key\"].(string)\n\t\tmodel, _ := setting[\"model\"].(string)\n\n\t\tif req.Connector.Is(connector.ANTHROPIC) {\n\t\t\tenv[\"ANTHROPIC_BASE_URL\"] = host\n\t\t\tenv[\"ANTHROPIC_API_KEY\"] = key\n\t\t} else {\n\t\t\tenv[\"ANTHROPIC_BASE_URL\"] = fmt.Sprintf(\"http://127.0.0.1:%d\", defaultProxyPort)\n\t\t\tenv[\"ANTHROPIC_API_KEY\"] = \"dummy\"\n\t\t}\n\n\t\tif model != \"\" {\n\t\t\tenv[\"ANTHROPIC_MODEL\"] = model\n\t\t\tenv[\"ANTHROPIC_DEFAULT_OPUS_MODEL\"] = model\n\t\t\tenv[\"ANTHROPIC_DEFAULT_SONNET_MODEL\"] = model\n\t\t\tenv[\"ANTHROPIC_DEFAULT_HAIKU_MODEL\"] = model\n\t\t\tenv[\"CLAUDE_CODE_SUBAGENT_MODEL\"] = model\n\t\t}\n\n\t\tif thinking, ok := setting[\"thinking\"].(map[string]interface{}); ok {\n\t\t\tthinkType, _ := thinking[\"type\"].(string)\n\t\t\tswitch thinkType {\n\t\t\tcase \"disabled\":\n\t\t\t\tenv[\"MAX_THINKING_TOKENS\"] = \"0\"\n\t\t\tcase \"enabled\":\n\t\t\t\tif budget, ok := thinking[\"budget_tokens\"].(float64); ok && budget > 0 {\n\t\t\t\t\tenv[\"MAX_THINKING_TOKENS\"] = fmt.Sprintf(\"%d\", int(budget))\n\t\t\t\t}\n\t\t\t}\n\t\t}\n\t}\n\n\tif req.Config != nil && len(req.Config.Secrets) > 0 {\n\t\tfor k, v := range req.Config.Secrets {\n\t\t\tenv[k] = v\n\t\t}\n\t}\n\n\tif req.Token != nil {\n\t\tif req.Token.Token != \"\" {\n\t\t\tenv[\"YAO_TOKEN\"] = req.Token.Token\n\t\t}\n\t\tif req.Token.RefreshToken != \"\" {\n\t\t\tenv[\"YAO_REFRESH_TOKEN\"] = req.Token.RefreshToken\n\t\t}\n\t}\n\n\tvar systemPrompt string\n\tenvPrompt := buildSandboxEnvPrompt(oe)\n\tif !isContinuation && req.SystemPrompt != \"\" {\n\t\tsystemPrompt = req.SystemPrompt + \"\\n\\n\" + envPrompt\n\t} else if !isContinuation {\n\t\tsystemPrompt = envPrompt\n\t}\n\n\tvar inputJSONL string\n\tif isContinuation {\n\t\tinputJSONL = buildLastUserMessageJSONL(req.Messages)\n\t} else {\n\t\tinputJSONL = buildFirstRequestJSONL(req.Messages)\n\t}\n\n\tvar args []string\n\n\tpermMode := \"\"\n\tif req.Config != nil && req.Config.Runner.Options != nil {\n\t\tif v, ok := req.Config.Runner.Options[\"permission_mode\"]; ok {\n\t\t\tpermMode = fmt.Sprintf(\"%v\", v)\n\t\t}\n\t}\n\tif permMode == \"bypassPermissions\" {\n\t\targs = append(args, \"--dangerously-skip-permissions\")\n\t\targs = append(args, \"--permission-mode\", permMode)\n\t}\n\n\targs = append(args, \"--input-format\", \"stream-json\")\n\targs = append(args, \"--output-format\", \"stream-json\")\n\targs = append(args, \"--include-partial-messages\")\n\targs = append(args, \"--verbose\")\n\n\tif isContinuation {\n\t\targs = append(args, \"--continue\")\n\t}\n\n\tif req.Config != nil && req.Config.Runner.Options != nil {\n\t\tfor key, val := range req.Config.Runner.Options {\n\t\t\tif flag, ok := claudeArgWhitelist[key]; ok {\n\t\t\t\targs = append(args, flag, fmt.Sprintf(\"%v\", val))\n\t\t\t}\n\t\t}\n\t}\n\n\tif r.hasMCP {\n\t\tmcpPath := oe.pathJoin(oe.WorkDir, \".claude\", \"mcp.json\")\n\t\targs = append(args, \"--mcp-config\", mcpPath)\n\t\tif r.mcpToolPattern != \"\" {\n\t\t\targs = append(args, \"--allowedTools\", r.mcpToolPattern)\n\t\t}\n\t}\n\n\tscript, stdin := oe.buildCLIScript(args, systemPrompt, inputJSONL)\n\treturn oe.shellCmd(script), env, stdin\n}\n\n// buildMCPConfig creates the .mcp.json for Claude CLI based on declared servers.\n// Each server delegates to \"tai mcp\" which implements the standard MCP protocol\n// over stdio and bridges to Yao gRPC with authentication.\n// Connection is configured via env vars (YAO_GRPC_ADDR, YAO_TOKEN, etc.)\n// injected by the sandbox infrastructure at container start.\nfunc buildMCPConfig(servers []types.MCPServer) []byte {\n\tmcpServers := make(map[string]any, len(servers))\n\tfor _, s := range servers {\n\t\tname := s.ServerID\n\t\tif name == \"\" {\n\t\t\tcontinue\n\t\t}\n\t\tmcpServers[name] = map[string]any{\n\t\t\t\"command\": \"tai\",\n\t\t\t\"args\": []string{\"mcp\", name},\n\t\t}\n\t}\n\tif len(mcpServers) == 0 {\n\t\tmcpServers[\"yao\"] = map[string]any{\n\t\t\t\"command\": \"tai\",\n\t\t\t\"args\": []string{\"mcp\"},\n\t\t}\n\t}\n\tconfig := map[string]any{\"mcpServers\": mcpServers}\n\tdata, _ := json.Marshal(config)\n\treturn data\n}\n\n// buildMCPAllowedTools generates the --allowedTools pattern from server IDs.\nfunc buildMCPAllowedTools(servers []types.MCPServer) string {\n\tpatterns := make([]string, 0, len(servers))\n\tfor _, s := range servers {\n\t\tif s.ServerID != \"\" {\n\t\t\tpatterns = append(patterns, fmt.Sprintf(\"mcp__%s__*\", s.ServerID))\n\t\t}\n\t}\n\tif len(patterns) == 0 {\n\t\treturn \"mcp__yao__*\"\n\t}\n\treturn strings.Join(patterns, \",\")\n}\n\n// buildSandboxEnvPrompt generates the sandbox environment prompt with system info and working directory.\nfunc buildSandboxEnvPrompt(oe *osEnv) string {\n\tworkDir := oe.WorkDir\n\n\tosName := oe.OS\n\tif osName == \"\" {\n\t\tosName = \"linux\"\n\t}\n\tshell := oe.Shell\n\tif shell == \"\" {\n\t\tshell = \"bash\"\n\t}\n\n\tshellNote := \"\"\n\tif oe.isWindows() {\n\t\tshellNote = `\n- **Desktop Environment**: You have full access to the Windows desktop (GUI applications, browsers, etc.)\n- **Important**: When you launch GUI applications (browsers, editors, etc.), do NOT close them unless explicitly asked — the user expects them to remain open`\n\t}\n\n\treturn fmt.Sprintf(`## Sandbox Environment\n\n- **Operating System**: %[2]s\n- **Shell**: %[3]s\n- **Working Directory**: %[1]s\n- **File Access**: You have full read/write access to %[1]s%[4]s\n\n## User Attachments\n\nUser-uploaded files (images, documents, code files, etc.) are placed in %[1]s/.attachments/{chatID}/\nEach chat session has its own subdirectory to avoid conflicts.\nWhen the user references an attached file, read it from this directory using the Read or Bash tool.\nFor image files, you can view them directly as Claude supports vision on local files.\n`, workDir, osName, shell, shellNote)\n}\n\nvar claudeArgWhitelist = map[string]string{\n\t\"max_turns\": \"--max-turns\",\n\t\"disallowed_tools\": \"--disallowed-tools\",\n\t\"allowed_tools\": \"--allowedTools\",\n}\n" }, { "path": "agent/sandbox/v2/claude/runner_test.go", "content": "package claude_test\n\nimport (\n\t\"bytes\"\n\t\"context\"\n\t\"fmt\"\n\t\"mime/multipart\"\n\t\"os\"\n\t\"path/filepath\"\n\t\"runtime\"\n\t\"strings\"\n\t\"testing\"\n\t\"time\"\n\n\t\"github.com/stretchr/testify/assert\"\n\t\"github.com/stretchr/testify/require\"\n\t\"github.com/yaoapp/yao/agent/caller\"\n\tagentcontext \"github.com/yaoapp/yao/agent/context\"\n\tsandboxtestutils \"github.com/yaoapp/yao/agent/sandbox/v2/testutils\"\n\t\"github.com/yaoapp/yao/attachment\"\n\toauthtypes \"github.com/yaoapp/yao/openapi/oauth/types\"\n)\n\ntype e2eCase struct {\n\tID string\n\tPrompt string\n\tTimeout time.Duration\n}\n\nvar cases = []e2eCase{\n\t{\n\t\tID: \"tests.sandbox-v2.oneshot-cli\",\n\t\tPrompt: \"Reply exactly with: hello sandbox v2\",\n\t\tTimeout: 3 * time.Minute,\n\t},\n}\n\nfunc TestSandboxV2_Claude_E2E(t *testing.T) {\n\tsandboxtestutils.Prepare(t)\n\tdefer sandboxtestutils.Clean(t)\n\n\trequire.NotNil(t, caller.AgentGetterFunc, \"AgentGetterFunc should be registered after Prepare\")\n\n\tfor _, tc := range cases {\n\t\ttc := tc\n\t\tt.Run(tc.ID, func(t *testing.T) {\n\t\t\tagent, err := caller.AgentGetterFunc(tc.ID)\n\t\t\trequire.NoError(t, err, \"should load assistant %s\", tc.ID)\n\n\t\t\ttimeout := tc.Timeout\n\t\t\tif timeout == 0 {\n\t\t\t\ttimeout = 3 * time.Minute\n\t\t\t}\n\n\t\t\tchatID := fmt.Sprintf(\"e2e-%s-%d\", tc.ID, time.Now().UnixMilli())\n\t\t\tctx := agentcontext.New(\n\t\t\t\tcontext.Background(),\n\t\t\t\t&oauthtypes.AuthorizedInfo{\n\t\t\t\t\tTeamID: \"test-team-e2e\",\n\t\t\t\t\tUserID: \"test-user-e2e\",\n\t\t\t\t},\n\t\t\t\tchatID,\n\t\t\t)\n\n\t\t\tmessages := []agentcontext.Message{\n\t\t\t\t{Role: \"user\", Content: tc.Prompt},\n\t\t\t}\n\n\t\t\tdone := make(chan struct{})\n\t\t\tvar resp *agentcontext.Response\n\t\t\tvar streamErr error\n\n\t\t\tgo func() {\n\t\t\t\tdefer close(done)\n\t\t\t\tresp, streamErr = agent.Stream(ctx, messages)\n\t\t\t}()\n\n\t\t\tselect {\n\t\t\tcase <-done:\n\t\t\tcase <-time.After(timeout):\n\t\t\t\tt.Fatalf(\"timeout after %v\", timeout)\n\t\t\t}\n\n\t\t\trequire.NoError(t, streamErr, \"Stream should not return error\")\n\t\t\trequire.NotNil(t, resp, \"response should not be nil\")\n\n\t\t\t// ── 1. CompletionResponse should behave like the LLM path ──\n\t\t\trequire.NotNil(t, resp.Completion, \"completion should not be nil\")\n\t\t\tassert.Equal(t, \"assistant\", resp.Completion.Role, \"role should be assistant\")\n\t\t\tassert.Equal(t, agentcontext.FinishReasonStop, resp.Completion.FinishReason, \"finish_reason should be stop\")\n\t\t\tassert.NotNil(t, resp.Completion.Content, \"Content should be populated (same as LLM path)\")\n\n\t\t\tcontentStr, ok := resp.Completion.Content.(string)\n\t\t\trequire.True(t, ok, \"Content should be a string, got %T\", resp.Completion.Content)\n\t\t\tt.Logf(\"CompletionResponse.Content (%d chars): %s\", len(contentStr), contentStr)\n\t\t\tassert.Contains(t, contentStr, \"hello sandbox v2\", \"Content should contain expected text\")\n\n\t\t\t// ── 2. Buffer: frame sequence handled correctly ──\n\t\t\trequire.NotNil(t, ctx.Buffer, \"ctx.Buffer should not be nil\")\n\n\t\t\tmsgs := ctx.Buffer.GetMessages()\n\t\t\tt.Logf(\"buffer message count: %d\", len(msgs))\n\t\t\tfor _, m := range msgs {\n\t\t\t\tt.Logf(\" seq=%d role=%s type=%s streaming=%v props_keys=%v\",\n\t\t\t\t\tm.Sequence, m.Role, m.Type, m.IsStreaming, mapKeys(m.Props))\n\t\t\t}\n\n\t\t\tvar userInputCount, assistantTextCount, loadingCount int\n\t\t\tvar bufferTextContent string\n\t\t\tfor _, m := range msgs {\n\t\t\t\tswitch {\n\t\t\t\tcase m.Role == \"user\" && m.Type == \"user_input\":\n\t\t\t\t\tuserInputCount++\n\t\t\t\tcase m.Role == \"assistant\" && m.Type == \"loading\":\n\t\t\t\t\tloadingCount++\n\t\t\t\tcase m.Role == \"assistant\" && m.Type == \"text\":\n\t\t\t\t\tassistantTextCount++\n\t\t\t\t\tassert.False(t, m.IsStreaming, \"text message should not be streaming (handleMessageEnd should have finalized it)\")\n\t\t\t\t\trequire.NotNil(t, m.Props, \"text message props should not be nil\")\n\t\t\t\t\tif c, ok := m.Props[\"content\"].(string); ok {\n\t\t\t\t\t\tbufferTextContent += c\n\t\t\t\t\t}\n\t\t\t\t}\n\t\t\t}\n\n\t\t\tassert.Equal(t, 1, userInputCount, \"should have exactly 1 user_input message\")\n\t\t\tassert.GreaterOrEqual(t, loadingCount, 1, \"should have at least 1 loading message\")\n\t\t\tassert.Equal(t, 1, assistantTextCount, \"should have exactly 1 assistant text message (from handleMessageEnd)\")\n\t\t\tassert.Contains(t, bufferTextContent, \"hello sandbox v2\", \"buffer text should contain expected content\")\n\n\t\t\t// ── 3. Buffer content matches CompletionResponse.Content ──\n\t\t\tassert.Equal(t, contentStr, bufferTextContent,\n\t\t\t\t\"CompletionResponse.Content and Buffer text should match\")\n\t\t})\n\t}\n}\n\nfunc TestSandboxV2_Claude_Attachments(t *testing.T) {\n\tsandboxtestutils.Prepare(t)\n\tdefer sandboxtestutils.Clean(t)\n\n\trequire.NotNil(t, caller.AgentGetterFunc, \"AgentGetterFunc should be registered after Prepare\")\n\n\tagent, err := caller.AgentGetterFunc(\"tests.sandbox-v2.oneshot-cli\")\n\trequire.NoError(t, err)\n\n\t// ── 1. Locate testdata via runtime.Caller ──\n\t_, thisFile, _, ok := runtime.Caller(0)\n\trequire.True(t, ok)\n\ttestdataDir := filepath.Join(filepath.Dir(thisFile), \"testdata\")\n\n\t// ── 2. Create attachment manager and upload test files ──\n\tconst uploaderName = \"__yao.attachment\"\n\tmanager, err := attachment.New(attachment.ManagerOption{\n\t\tDriver: \"local\",\n\t\tMaxSize: \"50M\",\n\t\tAllowedTypes: []string{\"image/*\", \"text/*\", \"application/*\", \"video/*\", \".ts\", \".js\", \".tsx\", \".jsx\"},\n\t\tOptions: map[string]interface{}{\"path\": filepath.Join(os.TempDir(), \"test_sandbox_v2_attach\")},\n\t})\n\trequire.NoError(t, err)\n\tmanager.Name = uploaderName\n\tattachment.Managers[uploaderName] = manager\n\tt.Cleanup(func() { delete(attachment.Managers, uploaderName) })\n\n\timgFile := uploadTestFile(t, manager, testdataDir, \"test-image.png\", \"image/png\")\n\tcodeFile := uploadTestFile(t, manager, testdataDir, \"code.ts\", \"text/plain\")\n\n\timgWrapper := fmt.Sprintf(\"%s://%s\", uploaderName, imgFile.ID)\n\tcodeWrapper := fmt.Sprintf(\"%s://%s\", uploaderName, codeFile.ID)\n\tt.Logf(\"image wrapper: %s\", imgWrapper)\n\tt.Logf(\"code wrapper: %s\", codeWrapper)\n\n\t// ── 3. Build multimodal messages (same as CUI InputArea) ──\n\tchatID := fmt.Sprintf(\"e2e-attach-%d\", time.Now().UnixMilli())\n\tctx := agentcontext.New(\n\t\tcontext.Background(),\n\t\t&oauthtypes.AuthorizedInfo{TeamID: \"test-team-e2e\", UserID: \"test-user-e2e\"},\n\t\tchatID,\n\t)\n\n\tmessages := []agentcontext.Message{\n\t\t{\n\t\t\tRole: \"user\",\n\t\t\tContent: []interface{}{\n\t\t\t\tmap[string]interface{}{\"type\": \"text\", \"text\": \"Describe the attached image and summarize the attached code file. Reply in English.\"},\n\t\t\t\tmap[string]interface{}{\n\t\t\t\t\t\"type\": \"image_url\",\n\t\t\t\t\t\"image_url\": map[string]interface{}{\"url\": imgWrapper, \"detail\": \"auto\"},\n\t\t\t\t},\n\t\t\t\tmap[string]interface{}{\n\t\t\t\t\t\"type\": \"file\",\n\t\t\t\t\t\"file\": map[string]interface{}{\"url\": codeWrapper, \"filename\": \"code.ts\"},\n\t\t\t\t},\n\t\t\t},\n\t\t},\n\t}\n\n\t// ── 4. Run E2E stream ──\n\tdone := make(chan struct{})\n\tvar resp *agentcontext.Response\n\tvar streamErr error\n\n\tgo func() {\n\t\tdefer close(done)\n\t\tresp, streamErr = agent.Stream(ctx, messages)\n\t}()\n\n\tselect {\n\tcase <-done:\n\tcase <-time.After(5 * time.Minute):\n\t\tt.Fatalf(\"timeout after 5m\")\n\t}\n\n\trequire.NoError(t, streamErr, \"Stream should not return error\")\n\trequire.NotNil(t, resp)\n\trequire.NotNil(t, resp.Completion)\n\n\tcontentStr, ok := resp.Completion.Content.(string)\n\trequire.True(t, ok, \"Content should be a string, got %T\", resp.Completion.Content)\n\tt.Logf(\"Response (%d chars): %s\", len(contentStr), contentStr)\n\n\tlower := strings.ToLower(contentStr)\n\n\t// ── 5. Verify Claude actually read the image ──\n\timageKeywords := []string{\"hello\", \"utf\", \"chinese\", \"text\", \"emoji\"}\n\timgHit := false\n\tfor _, kw := range imageKeywords {\n\t\tif strings.Contains(lower, kw) {\n\t\t\timgHit = true\n\t\t\tbreak\n\t\t}\n\t}\n\tassert.True(t, imgHit, \"response should mention image content (tried: %v)\", imageKeywords)\n\n\t// ── 6. Verify Claude actually read the code ──\n\tcodeKeywords := []string{\"excel\", \"typescript\", \"class\", \"volcengine\"}\n\tcodeHit := false\n\tfor _, kw := range codeKeywords {\n\t\tif strings.Contains(lower, kw) {\n\t\t\tcodeHit = true\n\t\t\tbreak\n\t\t}\n\t}\n\tassert.True(t, codeHit, \"response should mention code content (tried: %v)\", codeKeywords)\n}\n\nfunc uploadTestFile(t *testing.T, manager *attachment.Manager, testdataDir, filename, contentType string) *attachment.File {\n\tt.Helper()\n\tpath := filepath.Join(testdataDir, filename)\n\tdata, err := os.ReadFile(path)\n\trequire.NoError(t, err, \"read testdata/%s\", filename)\n\n\tfh := &attachment.FileHeader{\n\t\tFileHeader: &multipart.FileHeader{\n\t\t\tFilename: filename,\n\t\t\tSize: int64(len(data)),\n\t\t\tHeader: make(map[string][]string),\n\t\t},\n\t}\n\tfh.Header.Set(\"Content-Type\", contentType)\n\n\tfile, err := manager.Upload(context.Background(), fh, bytes.NewReader(data), attachment.UploadOption{\n\t\tGroups: []string{\"e2e-sandbox-v2\"},\n\t})\n\trequire.NoError(t, err, \"upload testdata/%s\", filename)\n\tt.Logf(\"uploaded %s => ID=%s, Path=%s\", filename, file.ID, file.Path)\n\treturn file\n}\n\nfunc mapKeys(m map[string]interface{}) []string {\n\tif m == nil {\n\t\treturn nil\n\t}\n\tkeys := make([]string, 0, len(m))\n\tfor k := range m {\n\t\tkeys = append(keys, k)\n\t}\n\treturn keys\n}\n" }, { "path": "agent/sandbox/v2/claude/testdata/code.ts", "content": "import { Process } from \"@yao/runtime\";\n\n/**\n * Excel class for manipulating Excel files via Yao's Excel Module\n */\nexport class Excel {\n private handle: string | null = null;\n\n /**\n * Creates a new Excel instance\n * @param file Path to the Excel file\n */\n constructor(private file: string, writable: boolean = false) {\n this.file = file;\n this.Open(writable);\n }\n\n /**\n * Read each sheet top n rows\n * @param file Path to the Excel file\n * @param n number of rows to read\n * @returns Object with sheet names as keys and arrays of row values as values\n */\n static Heads(\n file: string,\n n: number = 5,\n filters?: string[]\n ): Record {\n const excel = new Excel(file);\n const heads = excel.Heads(n, filters);\n excel.Close();\n return heads;\n }\n\n /**\n * Read each sheet top n rows\n * @param n number of rows to read\n * @returns Object with sheet names as keys and arrays of row values as values\n * @throws Error if file not opened\n */\n Heads(n: number = 5, filters: string[] = []): Record {\n if (!this.handle) throw new Error(\"Excel file not opened\");\n\n const sheets = this.Sheets();\n const result: Record = {};\n\n for (const sheet of sheets) {\n if (filters.length > 0 && !filters.includes(sheet)) {\n continue;\n }\n\n // Open row iterator for the sheet\n const iterator = this.each.OpenRow(sheet);\n const rows: any[][] = [];\n\n // Read n rows\n let row;\n let count = 0;\n while (\n count < n &&\n (row = Process(`excel.each.NextRow`, iterator)) !== null\n ) {\n // Add column headers (A, B, C, ...) for the first row\n if (count === 0) {\n const headerRow = [];\n for (let i = 0; i < row.length; i++) {\n headerRow.push(this.convert.ColumnNumberToName(i + 1));\n }\n rows.push(headerRow);\n }\n\n // Trim Each cell's value\n row = row.map((cell) => cell?.trim?.());\n rows.push(row);\n count++;\n }\n\n // Close the row iterator\n this.each.CloseRow(iterator);\n\n // Find the max length of each row, and pad the column headers(A, B, C, ...) to the same length\n const maxLength = Math.max(...rows.map((row) => row.length));\n const start = rows[0].length;\n const neededLength = maxLength - rows[0].length;\n for (let i = 0; i < neededLength; i++) {\n rows[0].push(this.convert.ColumnNumberToName(start + i + 1));\n }\n\n // Add the sheet's rows to the result\n result[sheet] = rows;\n }\n\n return result;\n }\n\n /**\n * Check if a sheet exists in the Excel file\n * @param file Path to the Excel file\n * @param sheet Sheet name to check\n * @returns boolean - true if sheet exists, false otherwise\n */\n static Exists(file: string, sheet: string) {\n const excel = new Excel(file);\n const exists = excel.sheet.Exists(sheet);\n excel.Close();\n return exists;\n }\n\n /**\n * Opens an Excel file for reading or writing\n * @param writable Whether to open in writable mode (true) or read-only mode (false)\n * @returns Handle ID used for subsequent operations\n */\n Open(writable: boolean = false) {\n this.handle = Process(`excel.Open`, this.file, writable);\n return this.handle;\n }\n\n /**\n * Closes the Excel file and releases resources\n * IMPORTANT: Always call this method when done to prevent memory leaks\n */\n Close() {\n if (this.handle) {\n Process(`excel.Close`, this.handle);\n this.handle = null;\n }\n }\n\n /**\n * Saves changes to the Excel file\n * @throws Error if file not opened\n */\n Save() {\n if (!this.handle) throw new Error(\"Excel file not opened\");\n return Process(`excel.Save`, this.handle);\n }\n\n /**\n * Gets all sheet names in the workbook\n * @returns Array of sheet names\n * @throws Error if file not opened\n */\n Sheets() {\n if (!this.handle) throw new Error(\"Excel file not opened\");\n return Process(`excel.Sheets`, this.handle);\n }\n\n // Sheet operations\n sheet = {\n /**\n * Creates a new sheet in the workbook\n * @param name Name for the new sheet\n * @returns number Index of the new sheet\n * @throws Error if file not opened\n */\n Create: (name: string) => {\n if (!this.handle) throw new Error(\"Excel file not opened\");\n return Process(`excel.sheet.create`, this.handle, name);\n },\n\n /**\n * Lists all sheets in the workbook\n * @returns string[] Array of sheet names\n * @throws Error if file not opened\n */\n List: () => {\n if (!this.handle) throw new Error(\"Excel file not opened\");\n return Process(`excel.sheet.list`, this.handle);\n },\n\n /**\n * Checks if a sheet exists in the workbook\n * @param name Sheet name to check\n * @returns boolean - true if sheet exists, false otherwise\n * @throws Error if file not opened\n */\n Exists: (name: string) => {\n if (!this.handle) throw new Error(\"Excel file not opened\");\n return Process(`excel.sheet.exists`, this.handle, name);\n },\n\n /**\n * Reads all data from a sheet\n * @param name Sheet name\n * @returns any[][] Two-dimensional array of cell values\n * @throws Error if file not opened\n */\n Read: (name: string) => {\n if (!this.handle) throw new Error(\"Excel file not opened\");\n return Process(`excel.sheet.read`, this.handle, name);\n },\n\n /**\n * Reads all data from a sheet with pagination support\n * @param name Sheet name\n * @param from Starting row index (0-based)\n * @param chunk_size Number of rows to read\n * @returns any[][] Two-dimensional array of cell values\n * @throws Error if file not opened\n */\n Rows: (name: string, from: number, chunk_size: number) => {\n if (!this.handle) throw new Error(\"Excel file not opened\");\n return Process(`excel.sheet.rows`, this.handle, name, from, chunk_size);\n },\n\n /**\n * Updates data in a sheet. Creates the sheet if it doesn't exist.\n * @param name Sheet name\n * @param data Two-dimensional array of values to write\n * @throws Error if file not opened\n */\n Update: (name: string, data: any[][]) => {\n if (!this.handle) throw new Error(\"Excel file not opened\");\n return Process(`excel.sheet.update`, this.handle, name, data);\n },\n\n /**\n * Copies a sheet with all its content and formatting\n * @param source Source sheet name\n * @param target Target sheet name (must not exist)\n * @throws Error if file not opened\n */\n Copy: (source: string, target: string) => {\n if (!this.handle) throw new Error(\"Excel file not opened\");\n return Process(`excel.sheet.copy`, this.handle, source, target);\n },\n\n /**\n * Deletes a sheet from the workbook\n * @param name Sheet name to delete\n * @throws Error if file not opened\n */\n Delete: (name: string) => {\n if (!this.handle) throw new Error(\"Excel file not opened\");\n return Process(`excel.sheet.delete`, this.handle, name);\n },\n\n /**\n * Gets the dimensions (number of rows and columns) of a sheet\n * @param name Sheet name\n * @returns {rows: number, cols: number} - Object containing row and column counts\n * @throws Error if file not opened\n */\n Dimension: (name: string) => {\n if (!this.handle) throw new Error(\"Excel file not opened\");\n return Process(`excel.sheet.dimension`, this.handle, name);\n },\n };\n\n // Reading operations\n read = {\n /**\n * Reads a cell's value\n * @param sheet Sheet name\n * @param cell Cell reference (e.g. \"A1\")\n * @returns Cell value\n * @throws Error if file not opened\n */\n Cell: (sheet: string, cell: string) => {\n if (!this.handle) throw new Error(\"Excel file not opened\");\n return Process(`excel.read.Cell`, this.handle, sheet, cell);\n },\n\n /**\n * Reads all rows in a sheet\n * @param sheet Sheet name\n * @returns Two-dimensional array of cell values\n * @throws Error if file not opened\n */\n Row: (sheet: string) => {\n if (!this.handle) throw new Error(\"Excel file not opened\");\n return Process(`excel.read.Row`, this.handle, sheet);\n },\n\n /**\n * Reads all columns in a sheet\n * @param sheet Sheet name\n * @returns Two-dimensional array of cell values\n * @throws Error if file not opened\n */\n Column: (sheet: string) => {\n if (!this.handle) throw new Error(\"Excel file not opened\");\n return Process(`excel.read.Column`, this.handle, sheet);\n },\n };\n\n // Writing operations\n write = {\n /**\n * Writes a value to a cell\n * @param sheet Sheet name\n * @param cell Cell reference (e.g. \"A1\")\n * @param value Value to write (string, number, boolean, etc.)\n * @throws Error if file not opened\n */\n Cell: (sheet: string, cell: string, value: any) => {\n if (!this.handle) throw new Error(\"Excel file not opened\");\n return Process(`excel.write.Cell`, this.handle, sheet, cell, value);\n },\n\n /**\n * Writes values to a row starting at the specified cell\n * @param sheet Sheet name\n * @param startCell Starting cell reference (e.g. \"A1\")\n * @param values Array of values to write\n * @throws Error if file not opened\n */\n Row: (sheet: string, startCell: string, values: any[]) => {\n if (!this.handle) throw new Error(\"Excel file not opened\");\n return Process(`excel.write.Row`, this.handle, sheet, startCell, values);\n },\n\n /**\n * Writes values to a column starting at the specified cell\n * @param sheet Sheet name\n * @param startCell Starting cell reference (e.g. \"A1\")\n * @param values Array of values to write\n * @throws Error if file not opened\n */\n Column: (sheet: string, startCell: string, values: any[]) => {\n if (!this.handle) throw new Error(\"Excel file not opened\");\n return Process(\n `excel.write.Column`,\n this.handle,\n sheet,\n startCell,\n values\n );\n },\n\n /**\n * Writes a two-dimensional array of values starting at the specified cell\n * @param sheet Sheet name\n * @param startCell Starting cell reference (e.g. \"A1\")\n * @param values Two-dimensional array of values to write\n * @throws Error if file not opened\n */\n All: (sheet: string, startCell: string, values: any[][]) => {\n if (!this.handle) throw new Error(\"Excel file not opened\");\n return Process(`excel.write.All`, this.handle, sheet, startCell, values);\n },\n };\n\n // Setting properties\n set = {\n /**\n * Applies a style to a cell\n * @param sheet Sheet name\n * @param cell Cell reference (e.g. \"A1\")\n * @param styleID Style ID to apply\n * @throws Error if file not opened\n */\n Style: (sheet: string, cell: string, styleID: number) => {\n if (!this.handle) throw new Error(\"Excel file not opened\");\n return Process(`excel.set.Style`, this.handle, sheet, cell, styleID);\n },\n\n /**\n * Sets a row's height\n * @param sheet Sheet name\n * @param row Row number\n * @param height Height in points\n * @throws Error if file not opened\n */\n RowHeight: (sheet: string, row: number, height: number) => {\n if (!this.handle) throw new Error(\"Excel file not opened\");\n return Process(`excel.set.RowHeight`, this.handle, sheet, row, height);\n },\n\n /**\n * Sets column width for a range of columns\n * @param sheet Sheet name\n * @param startCol Starting column letter\n * @param endCol Ending column letter\n * @param width Width in points\n * @throws Error if file not opened\n */\n ColumnWidth: (\n sheet: string,\n startCol: string,\n endCol: string,\n width: number\n ) => {\n if (!this.handle) throw new Error(\"Excel file not opened\");\n return Process(\n `excel.set.ColumnWidth`,\n this.handle,\n sheet,\n startCol,\n endCol,\n width\n );\n },\n\n /**\n * Merges cells in a range\n * @param sheet Sheet name\n * @param startCell Starting cell reference (e.g. \"A1\")\n * @param endCell Ending cell reference (e.g. \"B2\")\n * @throws Error if file not opened\n */\n MergeCell: (sheet: string, startCell: string, endCell: string) => {\n if (!this.handle) throw new Error(\"Excel file not opened\");\n return Process(\n `excel.set.MergeCell`,\n this.handle,\n sheet,\n startCell,\n endCell\n );\n },\n\n /**\n * Unmerges previously merged cells\n * @param sheet Sheet name\n * @param startCell Starting cell reference (e.g. \"A1\")\n * @param endCell Ending cell reference (e.g. \"B2\")\n * @throws Error if file not opened\n */\n UnmergeCell: (sheet: string, startCell: string, endCell: string) => {\n if (!this.handle) throw new Error(\"Excel file not opened\");\n return Process(\n `excel.set.UnmergeCell`,\n this.handle,\n sheet,\n startCell,\n endCell\n );\n },\n\n /**\n * Sets a formula in a cell\n * @param sheet Sheet name\n * @param cell Cell reference (e.g. \"C1\")\n * @param formula Excel formula without the leading equals sign\n * @throws Error if file not opened\n */\n Formula: (sheet: string, cell: string, formula: string) => {\n if (!this.handle) throw new Error(\"Excel file not opened\");\n return Process(`excel.set.Formula`, this.handle, sheet, cell, formula);\n },\n\n /**\n * Adds a hyperlink to a cell\n * @param sheet Sheet name\n * @param cell Cell reference (e.g. \"A1\")\n * @param url URL for the hyperlink\n * @param text Display text for the hyperlink\n * @throws Error if file not opened\n */\n Link: (sheet: string, cell: string, url: string, text: string) => {\n if (!this.handle) throw new Error(\"Excel file not opened\");\n return Process(`excel.set.Link`, this.handle, sheet, cell, url, text);\n },\n };\n\n // Iteration methods\n each = {\n /**\n * Opens a row iterator\n * @param sheet Sheet name\n * @returns Row iterator ID\n * @throws Error if file not opened\n */\n OpenRow: (sheet: string) => {\n if (!this.handle) throw new Error(\"Excel file not opened\");\n return Process(`excel.each.OpenRow`, this.handle, sheet);\n },\n\n /**\n * Gets the next row from the iterator\n * @param rowID Row iterator ID from excel.each.OpenRow\n * @returns Array of cell values or null if no more rows\n */\n NextRow: (rowID: string) => {\n return Process(`excel.each.NextRow`, rowID);\n },\n\n /**\n * Closes the row iterator\n * @param rowID Row iterator ID from excel.each.OpenRow\n */\n CloseRow: (rowID: string) => {\n return Process(`excel.each.CloseRow`, rowID);\n },\n\n /**\n * Opens a column iterator\n * @param sheet Sheet name\n * @returns Column iterator ID\n * @throws Error if file not opened\n */\n OpenColumn: (sheet: string) => {\n if (!this.handle) throw new Error(\"Excel file not opened\");\n return Process(`excel.each.OpenColumn`, this.handle, sheet);\n },\n\n /**\n * Gets the next column from the iterator\n * @param colID Column iterator ID from excel.each.OpenColumn\n * @returns Array of cell values or null if no more columns\n */\n NextColumn: (colID: string) => {\n return Process(`excel.each.NextColumn`, colID);\n },\n\n /**\n * Closes the column iterator\n * @param colID Column iterator ID from excel.each.OpenColumn\n */\n CloseColumn: (colID: string) => {\n return Process(`excel.each.CloseColumn`, colID);\n },\n };\n\n // Conversion utilities\n convert = {\n /**\n * Converts a column name to a column number\n * @param colName Column name (e.g. \"A\", \"AB\")\n * @returns Column number (1-based)\n */\n ColumnNameToNumber: (colName: string) => {\n return Process(`excel.convert.ColumnNameToNumber`, colName);\n },\n\n /**\n * Converts a column number to a column name\n * @param colNum Column number (1-based)\n * @returns Column name\n */\n ColumnNumberToName: (colNum: number) => {\n return Process(`excel.convert.ColumnNumberToName`, colNum);\n },\n\n /**\n * Converts a cell reference to coordinates\n * @param cell Cell reference (e.g. \"A1\")\n * @returns Array with [columnNumber, rowNumber] (1-based)\n */\n CellNameToCoordinates: (cell: string) => {\n return Process(`excel.convert.CellNameToCoordinates`, cell);\n },\n\n /**\n * Converts coordinates to a cell reference\n * @param col Column number (1-based)\n * @param row Row number (1-based)\n * @returns Cell reference\n */\n CoordinatesToCellName: (col: number, row: number) => {\n return Process(`excel.convert.CoordinatesToCellName`, col, row);\n },\n };\n}\n/**\n * Volcengine OpenAPI SDK\n */\nimport { Exception, http, Process } from \"@yao/runtime\";\n\nexport class Volcengine {\n private AccessKeyId: string;\n private SecretAccessKey: string;\n private Region: string;\n private Service: string;\n private Endpoint: string;\n constructor(option: Option) {\n this.AccessKeyId = option.AccessKeyId;\n this.SecretAccessKey = option.SecretAccessKey;\n this.Region = option.Region;\n this.Service = option.Service;\n this.Endpoint = option.Endpoint\n ? `https://${option.Endpoint}`\n : `https://${this.Service}.${this.Region}.volcengineapi.com`;\n }\n\n public Get(query: Record) {\n const url = this.Endpoint;\n const host = url.split(\"://\")[1].split(\"/\")[0];\n const headers = { host: host };\n const request: Request = {\n Method: \"GET\",\n URI: \"/\",\n Query: query,\n Headers: headers,\n Payload: null,\n };\n\n const auth = this.getAuthorization(request);\n\n // Add authorization header\n headers[\"Authorization\"] = auth;\n headers[\"Content-Type\"] = \"application/json\";\n\n const resp = http.Get(url, query, headers);\n if (resp.code > 299 || resp.code < 200) {\n const { ResponseMetadata } = resp.data || {};\n const { Error } = ResponseMetadata || {};\n const message =\n Error?.Message || (resp.code === 0 ? resp.message : \"Unknown error\");\n throw new Exception(message, resp.code);\n }\n\n return resp.data;\n }\n\n /**\n * Post request\n * @param query Query parameters\n * @param payload Payload\n * @returns Response\n */\n public Post(query: Record, payload: Record) {\n const url = this.Endpoint;\n const host = url.split(\"://\")[1].split(\"/\")[0];\n const headers = { host: host };\n const body = JSON.stringify(payload);\n const request: Request = {\n Method: \"POST\",\n URI: \"/\",\n Query: query,\n Headers: headers,\n Payload: body,\n };\n\n const auth = this.getAuthorization(request);\n headers[\"Authorization\"] = auth;\n headers[\"Content-Type\"] = \"application/json\";\n\n const resp = http.Post(url, body, null, query, headers);\n if (resp.code > 299 || resp.code < 200) {\n const { ResponseMetadata } = resp.data || {};\n const { Error } = ResponseMetadata || {};\n const message =\n Error?.Message || (resp.code === 0 ? resp.message : \"Unknown error\");\n throw new Exception(message, resp.code);\n }\n return resp.data;\n }\n\n /**\n * Create a canonical request\n * @param request Request object\n * @returns Canonical request string\n */\n private canonicalRequest(request: Request): string {\n const xDate = this.formatDate(new Date());\n\n // 1. HTTP Method\n const method = request.Method;\n\n // 2. URI (default to '/' if null)\n const uri = request.URI || \"/\";\n\n // 3. Query String\n let queryString = \"\";\n if (request.Query) {\n if (Array.isArray(request.Query)) {\n // Handle array of query parameters\n const queryParams = request.Query.reduce((acc: string[], curr) => {\n Object.entries(curr).forEach(([key, value]) => {\n if (value !== null && value !== undefined && value !== \"\") {\n acc.push(\n `${encodeURIComponent(key)}=${encodeURIComponent(value)}`\n );\n }\n });\n return acc;\n }, []);\n queryString = queryParams.sort().join(\"&\");\n } else {\n // Handle single query object\n const queryParams = Object.entries(request.Query)\n .filter(\n ([_, value]) =>\n value !== null && value !== undefined && value !== \"\"\n )\n .map(\n ([key, value]) =>\n `${encodeURIComponent(key)}=${encodeURIComponent(value)}`\n )\n .sort();\n queryString = queryParams.join(\"&\");\n }\n }\n\n // 4. Headers\n // First, collect all headers in a normalized format\n const headers: Record = { \"x-date\": xDate };\n if (request.Headers) {\n if (Array.isArray(request.Headers)) {\n request.Headers.forEach((headerObj) => {\n Object.entries(headerObj).forEach(([key, value]) => {\n if (value !== null && value !== undefined && value.trim() !== \"\") {\n headers[key.toLowerCase()] = value.trim();\n }\n });\n });\n } else {\n Object.entries(request.Headers).forEach(([key, value]) => {\n if (value !== null && value !== undefined && value.trim() !== \"\") {\n headers[key.toLowerCase()] = value.trim();\n }\n });\n }\n }\n\n // Get required headers if they exist\n const signedHeaderKeys: string[] = [];\n const requiredHeaders = [\"host\", \"x-date\"];\n\n // Add required headers first if they exist\n requiredHeaders.forEach((key) => {\n if (headers[key]) {\n signedHeaderKeys.push(key);\n }\n });\n\n // Add any additional headers\n // const additionalHeaders = Object.keys(headers)\n // .filter((key) => !requiredHeaders.includes(key))\n // .sort();\n // signedHeaderKeys.push(...additionalHeaders);\n\n // Build canonical headers string\n const canonicalHeaders = signedHeaderKeys\n .map((key) => `${key}:${headers[key]}`)\n .join(\"\\n\");\n\n // Build signed headers string\n const signedHeaders = signedHeaderKeys.join(\";\");\n\n // 5. Payload/Body\n let hashedPayload = Process(\"crypto.Hash\", \"SHA256\", \"\");\n if (request.Payload !== null && request.Payload !== undefined) {\n if (typeof request.Payload === \"string\") {\n if (request.Payload !== \"\") {\n hashedPayload = Process(\"crypto.Hash\", \"SHA256\", request.Payload);\n }\n } else {\n const payload = JSON.stringify(request.Payload);\n if (payload !== \"{}\" && payload !== \"[]\") {\n hashedPayload = Process(\"crypto.Hash\", \"SHA256\", payload);\n }\n }\n }\n\n // Combine all components\n const parts = [\n method,\n uri,\n queryString,\n canonicalHeaders,\n \"\", // Empty line after headers\n signedHeaders,\n hashedPayload,\n ];\n\n return parts.join(\"\\n\");\n }\n\n /**\n * Format date to YYYYMMDDTHHMMSSZ\n * @param date Date object\n * @returns Formatted date string\n */\n private formatDate(date: Date): string {\n const year = date.getUTCFullYear();\n const month = String(date.getUTCMonth() + 1).padStart(2, \"0\");\n const day = String(date.getUTCDate()).padStart(2, \"0\");\n const hours = String(date.getUTCHours()).padStart(2, \"0\");\n const minutes = String(date.getUTCMinutes()).padStart(2, \"0\");\n const seconds = String(date.getUTCSeconds()).padStart(2, \"0\");\n return `${year}${month}${day}T${hours}${minutes}${seconds}Z`;\n }\n\n /**\n * Create string to sign\n * @param canonicalRequest Canonical request string\n * @returns String to sign\n */\n private stringToSign(canonicalRequest: string): string {\n const algorithm = \"HMAC-SHA256\";\n const requestDateTime = this.formatDate(new Date());\n const requestDate = requestDateTime.slice(0, 8);\n const credentialScope = `${requestDate}/${this.Region}/${this.Service}/request`; // YYYYMMDD\n\n const hashedCanonicalRequest = Process(\n \"crypto.Hash\",\n \"SHA256\",\n canonicalRequest\n );\n\n return `${algorithm}\\n${requestDateTime}\\n${credentialScope}\\n${hashedCanonicalRequest}`;\n }\n\n /**\n * Derive signing key\n * @param date Date in YYYY/MM/DD format\n * @returns Signing key\n */\n private getSigningKey(date: string): string {\n const kDate = Process(\"crypto.HMAC\", \"SHA256\", date, this.SecretAccessKey);\n const kRegion = Process(\n \"crypto.HMACWith\",\n { key: \"hex\" },\n this.Region,\n kDate\n );\n const kService = Process(\n \"crypto.HMACWith\",\n { key: \"hex\" },\n this.Service,\n kRegion\n );\n\n const kSigning = Process(\n \"crypto.HMACWith\",\n { key: \"hex\" },\n \"request\",\n kService\n );\n return kSigning;\n }\n\n /**\n * Calculate signature\n * @param stringToSign String to sign\n * @param signingKey Signing key\n * @returns Signature\n */\n private signature(stringToSign: string, signingKey: string): string {\n return Process(\"crypto.HMACWith\", { key: \"hex\" }, stringToSign, signingKey);\n }\n\n /**\n * Build authorization header\n * @param request Request object\n * @returns Authorization header value\n */\n public getAuthorization(request: Request): string {\n const xDate = this.formatDate(new Date());\n if (request.Headers) {\n if (typeof request.Headers === \"object\") {\n request.Headers[\"x-date\"] = request.Headers[\"x-date\"]\n ? request.Headers[\"x-date\"]\n : xDate;\n }\n }\n\n // 1. Create canonical request\n const canonicalReq = this.canonicalRequest(request);\n\n // 2. Create string to sign\n const stringToSign = this.stringToSign(canonicalReq);\n\n // 3. Get date from string to sign\n const [algorithm, requestDateTime, credentialScope] =\n stringToSign.split(\"\\n\");\n const date = requestDateTime.slice(0, 8);\n\n // 4. Derive signing key\n const signingKey = this.getSigningKey(date);\n // 5. Calculate signature\n const signature = this.signature(stringToSign, signingKey);\n\n // 6. Build authorization header\n let signedHeaders = \"\";\n if (request.Headers) {\n const headers: Record = {};\n if (Array.isArray(request.Headers)) {\n request.Headers.forEach((headerObj) => {\n Object.entries(headerObj).forEach(([key, value]) => {\n headers[key.toLowerCase()] = value.trim();\n });\n });\n } else {\n Object.entries(request.Headers).forEach(([key, value]) => {\n headers[key.toLowerCase()] = value.trim();\n });\n }\n signedHeaders = Object.keys(headers).sort().join(\";\");\n }\n\n return `${algorithm} Credential=${this.AccessKeyId}/${credentialScope}, SignedHeaders=${signedHeaders}, Signature=${signature}`;\n }\n}\n\nexport interface Option {\n AccessKeyId: string;\n SecretAccessKey: string;\n Endpoint?: string;\n Region: string;\n Service: string;\n}\n\nexport interface Request {\n Method: \"GET\" | \"POST\";\n URI: string | null; // Default /\n Query: Record | Record[] | null;\n Headers: Record | Record[] | null;\n Payload: string | Record | any[] | null;\n}\n" }, { "path": "agent/sandbox/v2/init.go", "content": "package sandboxv2\n\nimport (\n\t\"github.com/yaoapp/yao/agent/sandbox/v2/claude\"\n\t\"github.com/yaoapp/yao/agent/sandbox/v2/types\"\n\tyaorunner \"github.com/yaoapp/yao/agent/sandbox/v2/yao\"\n)\n\nfunc init() {\n\tRegister(\"claude\", func() types.Runner { return claude.New() })\n\tRegister(\"claude/cli\", func() types.Runner { return claude.New() })\n\tRegister(\"yao\", func() types.Runner { return yaorunner.New() })\n}\n" }, { "path": "agent/sandbox/v2/lifecycle.go", "content": "package sandboxv2\n\nimport (\n\t\"context\"\n\t\"crypto/rand\"\n\t\"encoding/hex\"\n\t\"fmt\"\n\t\"log\"\n\n\t\"github.com/yaoapp/gou/connector\"\n\tagentContext \"github.com/yaoapp/yao/agent/context\"\n\t\"github.com/yaoapp/yao/agent/sandbox/v2/types\"\n\tinfra \"github.com/yaoapp/yao/sandbox/v2\"\n\t\"github.com/yaoapp/yao/tai\"\n\t\"github.com/yaoapp/yao/workspace\"\n)\n\n// BuildIdentifier determines the Computer identifier based on lifecycle policy\n// and optional metadata override. Returns \"\" for oneshot (always new).\nfunc BuildIdentifier(cfg *types.SandboxConfig, ownerID, chatID, assistantID, workspaceID string, metadata map[string]any) string {\n\tif cfg.Lifecycle == \"oneshot\" {\n\t\treturn \"\"\n\t}\n\n\tswitch cfg.Lifecycle {\n\tcase \"session\":\n\t\treturn fmt.Sprintf(\"%s-%s-%s\", ownerID, assistantID, chatID)\n\tcase \"longrunning\", \"persistent\":\n\t\treturn fmt.Sprintf(\"%s-%s.%s\", ownerID, assistantID, workspaceID)\n\tdefault:\n\t\treturn \"\"\n\t}\n}\n\n// ResolveNodeID determines the target nodeID and computer kind based on\n// metadata and DSL configuration, without creating or acquiring a container.\n// Returns (nodeID, kind, error). kind is \"box\" or \"host\".\nfunc ResolveNodeID(ctx *agentContext.Context, cfg *types.SandboxConfig, manager *infra.Manager) (string, string, error) {\n\tcomputerID := \"\"\n\tif ctx.Metadata != nil {\n\t\tif cid, ok := ctx.Metadata[\"computer_id\"].(string); ok && cid != \"\" {\n\t\t\tcomputerID = cid\n\t\t}\n\t}\n\n\tworkspaceID := \"\"\n\tif ctx.Metadata != nil {\n\t\tif ws, ok := ctx.Metadata[\"workspace_id\"].(string); ok && ws != \"\" {\n\t\t\tworkspaceID = ws\n\t\t}\n\t}\n\townerID := resolveOwnerID(ctx)\n\tif workspaceID == \"\" {\n\t\tworkspaceID = ownerID\n\t}\n\n\tif workspaceID != \"\" && workspaceID != ownerID {\n\t\twsNode, err := workspace.M().NodeForWorkspace(context.Background(), workspaceID)\n\t\tif err == nil && wsNode != \"\" {\n\t\t\tcomputerID = wsNode\n\t\t}\n\t}\n\n\tif computerID != \"\" {\n\t\tif node, ok := tai.GetNodeMeta(computerID); ok {\n\t\t\thasContainerRuntime := node.Capabilities.Docker || node.Capabilities.K8s\n\t\t\tif node.Capabilities.HostExec && !hasContainerRuntime {\n\t\t\t\treturn computerID, \"host\", nil\n\t\t\t}\n\t\t\tif node.Capabilities.HostExec && hasContainerRuntime && cfg.Computer.Image == \"\" {\n\t\t\t\treturn computerID, \"host\", nil\n\t\t\t}\n\t\t\tif !hasContainerRuntime {\n\t\t\t\treturn \"\", \"\", fmt.Errorf(\"node %q has no container runtime and no host_exec capability\", computerID)\n\t\t\t}\n\t\t\treturn computerID, \"box\", nil\n\t\t}\n\t\treturn computerID, \"box\", nil\n\t}\n\n\tif cfg.Computer.Image == \"\" {\n\t\tnodeID := cfg.NodeID\n\t\treturn nodeID, \"host\", nil\n\t}\n\n\tnodeID := cfg.NodeID\n\treturn nodeID, \"box\", nil\n}\n\n// GetComputer obtains or creates a Computer for the current request.\n// An optional connector may be passed to inject OPENAI_PROXY_* env vars.\n// Returns the Computer, the resolved identifier, and any error.\nfunc GetComputer(ctx *agentContext.Context, cfg *types.SandboxConfig, manager *infra.Manager, conn ...connector.Connector) (infra.Computer, string, error) {\n\townerID := resolveOwnerID(ctx)\n\n\tworkspaceID := \"\"\n\tif ctx.Metadata != nil {\n\t\tif ws, ok := ctx.Metadata[\"workspace_id\"].(string); ok && ws != \"\" {\n\t\t\tworkspaceID = ws\n\t\t}\n\t}\n\tif workspaceID == \"\" {\n\t\tworkspaceID = ownerID\n\t}\n\n\tidentifier := BuildIdentifier(cfg, ownerID, ctx.ChatID, ctx.AssistantID, workspaceID, ctx.Metadata)\n\n\t// Fill runtime fields.\n\tcfg.Owner = ownerID\n\tcfg.ID = identifier\n\tcfg.WorkspaceID = workspaceID\n\n\t// Resolve computer_id from metadata to determine kind and nodeID.\n\tcomputerID := \"\"\n\tif ctx.Metadata != nil {\n\t\tif cid, ok := ctx.Metadata[\"computer_id\"].(string); ok && cid != \"\" {\n\t\t\tcomputerID = cid\n\t\t}\n\t}\n\n\t// Workspace-wins rule: when both workspace_id and computer_id are present,\n\t// the workspace's bound node takes precedence over computer_id.\n\tif workspaceID != \"\" && workspaceID != ownerID {\n\t\twsNode, err := workspace.M().NodeForWorkspace(context.Background(), workspaceID)\n\t\tif err == nil && wsNode != \"\" {\n\t\t\tif computerID != \"\" && computerID != wsNode {\n\t\t\t\tlog.Printf(\"[sandbox/v2] workspace %s bound to node %s overrides computer_id %s\", workspaceID, wsNode, computerID)\n\t\t\t}\n\t\t\tcomputerID = wsNode\n\t\t}\n\t}\n\n\tif computerID != \"\" {\n\t\treturn resolveComputerByID(cfg, manager, computerID, ownerID, identifier, workspaceID, conn...)\n\t}\n\n\t// No computer_id: fall back to DSL-based dispatch (original logic).\n\treturn resolveComputerByDSL(cfg, manager, ownerID, identifier, workspaceID, conn...)\n}\n\n// resolveComputerByID dispatches based on the runtime computer_id from metadata.\n// It queries the registry and sandbox manager to determine the computer kind.\nfunc resolveComputerByID(\n\tcfg *types.SandboxConfig, manager *infra.Manager,\n\tcomputerID, ownerID, identifier, workspaceID string,\n\tconn ...connector.Connector,\n) (infra.Computer, string, error) {\n\n\t// 1) Check if computer_id is a known Tai node (host or node kind).\n\tif node, ok := tai.GetNodeMeta(computerID); ok {\n\t\tcfg.NodeID = computerID\n\t\thasContainerRuntime := node.Capabilities.Docker || node.Capabilities.K8s\n\n\t\tif node.Capabilities.HostExec && !hasContainerRuntime {\n\t\t\t// Host-only node: must use host mode regardless of DSL image config.\n\t\t\tcfg.Kind = \"host\"\n\t\t\thost, err := manager.Host(context.Background(), computerID)\n\t\t\tif err != nil {\n\t\t\t\treturn nil, identifier, fmt.Errorf(\"get host computer: %w\", err)\n\t\t\t}\n\t\t\thost.BindWorkplace(workspaceID)\n\t\t\treturn host, identifier, nil\n\t\t}\n\n\t\tif node.Capabilities.HostExec && hasContainerRuntime && cfg.Computer.Image == \"\" {\n\t\t\t// Dual-capable node with no image in DSL: prefer host mode.\n\t\t\tcfg.Kind = \"host\"\n\t\t\thost, err := manager.Host(context.Background(), computerID)\n\t\t\tif err != nil {\n\t\t\t\treturn nil, identifier, fmt.Errorf(\"get host computer: %w\", err)\n\t\t\t}\n\t\t\thost.BindWorkplace(workspaceID)\n\t\t\treturn host, identifier, nil\n\t\t}\n\n\t\tif !hasContainerRuntime {\n\t\t\treturn nil, identifier, fmt.Errorf(\"node %q has no container runtime and no host_exec capability\", computerID)\n\t\t}\n\n\t\t// Node with container runtime and DSL has image: create/reuse a box.\n\t\tcfg.Kind = \"box\"\n\t\treturn resolveBox(cfg, manager, ownerID, identifier, workspaceID, conn...)\n\t}\n\n\t// 2) Check if computer_id is an existing box ID.\n\tif manager != nil {\n\t\tbox, err := manager.Get(context.Background(), computerID)\n\t\tif err == nil && box != nil {\n\t\t\tcfg.Kind = \"box\"\n\t\t\tbox.BindWorkplace(workspaceID)\n\t\t\treturn box, computerID, nil\n\t\t}\n\t}\n\n\treturn nil, identifier, fmt.Errorf(\"computer %q not found in registry or sandbox manager\", computerID)\n}\n\n// resolveComputerByDSL dispatches based on DSL static configuration (cfg.Computer.Image).\nfunc resolveComputerByDSL(\n\tcfg *types.SandboxConfig, manager *infra.Manager,\n\townerID, identifier, workspaceID string,\n\tconn ...connector.Connector,\n) (infra.Computer, string, error) {\n\n\t// Host mode: no image → host computer.\n\tif cfg.Computer.Image == \"\" {\n\t\tcfg.Kind = \"host\"\n\t\tnodeID := cfg.NodeID\n\t\tif nodeID == \"\" {\n\t\t\treturn nil, identifier, fmt.Errorf(\"host mode requires a nodeID (set in sandbox.yao or workspace)\")\n\t\t}\n\t\thost, err := manager.Host(context.Background(), nodeID)\n\t\tif err != nil {\n\t\t\treturn nil, identifier, fmt.Errorf(\"get host computer: %w\", err)\n\t\t}\n\t\thost.BindWorkplace(workspaceID)\n\t\treturn host, identifier, nil\n\t}\n\n\tcfg.Kind = \"box\"\n\treturn resolveBox(cfg, manager, ownerID, identifier, workspaceID, conn...)\n}\n\n// resolveBox reuses or creates a box container.\nfunc resolveBox(\n\tcfg *types.SandboxConfig, manager *infra.Manager,\n\townerID, identifier, workspaceID string,\n\tconn ...connector.Connector,\n) (infra.Computer, string, error) {\n\n\t// Reuse: non-empty identifier → try Get first.\n\tif identifier != \"\" {\n\t\tbox, err := manager.Get(context.Background(), identifier)\n\t\tif err == nil && box != nil {\n\t\t\tif box.IsStopped() {\n\t\t\t\tif startErr := manager.StartBox(context.Background(), identifier); startErr != nil {\n\t\t\t\t\tlog.Printf(\"[sandbox/v2] auto-start stopped box %s failed: %v, creating new\", identifier, startErr)\n\t\t\t\t} else {\n\t\t\t\t\tbox.BindWorkplace(workspaceID)\n\t\t\t\t\treturn box, identifier, nil\n\t\t\t\t}\n\t\t\t} else {\n\t\t\t\tbox.BindWorkplace(workspaceID)\n\t\t\t\treturn box, identifier, nil\n\t\t\t}\n\t\t}\n\t}\n\n\t// Create new box.\n\tvar c connector.Connector\n\tif len(conn) > 0 {\n\t\tc = conn[0]\n\t}\n\tcreateOpts, err := BuildCreateOptions(cfg, identifier, ownerID, workspaceID, c)\n\tif err != nil {\n\t\treturn nil, identifier, fmt.Errorf(\"build create options: %w\", err)\n\t}\n\n\t// Oneshot with empty identifier: generate a random one.\n\tif createOpts.ID == \"\" {\n\t\tcreateOpts.ID = randomID()\n\t\tidentifier = createOpts.ID\n\t\tcfg.ID = identifier\n\t}\n\n\tbox, err := manager.Create(context.Background(), createOpts)\n\tif err != nil {\n\t\treturn nil, identifier, fmt.Errorf(\"create computer: %w\", err)\n\t}\n\treturn box, identifier, nil\n}\n\n// LifecycleAction performs the post-request lifecycle operation based on policy.\n// Called in defer after executeSandboxStream completes.\nfunc LifecycleAction(ctx context.Context, cfg *types.SandboxConfig, computer infra.Computer, manager *infra.Manager) {\n\tif computer == nil || cfg == nil {\n\t\treturn\n\t}\n\n\tinfo := computer.ComputerInfo()\n\n\tswitch cfg.Lifecycle {\n\tcase \"oneshot\":\n\t\tif info.Kind == \"box\" && manager != nil {\n\t\t\tif err := manager.Remove(ctx, cfg.ID); err != nil {\n\t\t\t\tlog.Printf(\"[sandbox/v2] oneshot remove %s: %v\", cfg.ID, err)\n\t\t\t}\n\t\t}\n\n\tcase \"session\", \"longrunning\":\n\t\tif info.Kind == \"box\" && manager != nil {\n\t\t\tmanager.Heartbeat(cfg.ID, false, 0) // active=false: request finished, start idle timer\n\t\t}\n\n\tcase \"persistent\":\n\t\t// No action — persistent boxes survive indefinitely.\n\t}\n}\n\n// resolveOwnerID returns teamID if available, otherwise userID.\nfunc resolveOwnerID(ctx *agentContext.Context) string {\n\tif ctx.Authorized != nil {\n\t\tif ctx.Authorized.TeamID != \"\" {\n\t\t\treturn ctx.Authorized.TeamID\n\t\t}\n\t\tif ctx.Authorized.UserID != \"\" {\n\t\t\treturn ctx.Authorized.UserID\n\t\t}\n\t}\n\treturn \"anonymous\"\n}\n\nfunc randomID() string {\n\tb := make([]byte, 8)\n\t_, _ = rand.Read(b)\n\treturn hex.EncodeToString(b)\n}\n" }, { "path": "agent/sandbox/v2/lifecycle_test.go", "content": "package sandboxv2_test\n\nimport (\n\t\"context\"\n\t\"fmt\"\n\t\"strings\"\n\t\"testing\"\n\t\"time\"\n\n\tagentContext \"github.com/yaoapp/yao/agent/context\"\n\tsandboxv2 \"github.com/yaoapp/yao/agent/sandbox/v2\"\n\t\"github.com/yaoapp/yao/agent/sandbox/v2/types\"\n\toauthTypes \"github.com/yaoapp/yao/openapi/oauth/types\"\n\tinfra \"github.com/yaoapp/yao/sandbox/v2\"\n)\n\n// ===========================================================================\n// BuildIdentifier — pure-function tests (no infra needed)\n// ===========================================================================\n\nfunc TestBuildIdentifier_Oneshot(t *testing.T) {\n\tcfg := &types.SandboxConfig{Lifecycle: \"oneshot\"}\n\tid := sandboxv2.BuildIdentifier(cfg, \"owner1\", \"chat1\", \"ast1\", \"ws1\", nil)\n\tif id != \"\" {\n\t\tt.Errorf(\"oneshot should return empty, got %q\", id)\n\t}\n}\n\nfunc TestBuildIdentifier_Session(t *testing.T) {\n\tcfg := &types.SandboxConfig{Lifecycle: \"session\"}\n\tid := sandboxv2.BuildIdentifier(cfg, \"owner1\", \"chat42\", \"ast1\", \"ws1\", nil)\n\tif id != \"owner1-ast1-chat42\" {\n\t\tt.Errorf(\"session: got %q, want %q\", id, \"owner1-ast1-chat42\")\n\t}\n}\n\nfunc TestBuildIdentifier_Longrunning(t *testing.T) {\n\tcfg := &types.SandboxConfig{Lifecycle: \"longrunning\"}\n\tid := sandboxv2.BuildIdentifier(cfg, \"owner1\", \"chat1\", \"ast99\", \"ws1\", nil)\n\tif id != \"owner1-ast99.ws1\" {\n\t\tt.Errorf(\"longrunning: got %q, want %q\", id, \"owner1-ast99.ws1\")\n\t}\n}\n\nfunc TestBuildIdentifier_Persistent(t *testing.T) {\n\tcfg := &types.SandboxConfig{Lifecycle: \"persistent\"}\n\tid := sandboxv2.BuildIdentifier(cfg, \"owner1\", \"chat1\", \"ast99\", \"ws1\", nil)\n\tif id != \"owner1-ast99.ws1\" {\n\t\tt.Errorf(\"persistent: got %q, want %q\", id, \"owner1-ast99.ws1\")\n\t}\n}\n\nfunc TestBuildIdentifier_MetadataOverride(t *testing.T) {\n\tcfg := &types.SandboxConfig{Lifecycle: \"session\"}\n\tmeta := map[string]any{\"computer_id\": \"custom-box\"}\n\tid := sandboxv2.BuildIdentifier(cfg, \"owner1\", \"chat1\", \"ast1\", \"ws1\", meta)\n\t// computer_id is used for routing only, not for identifier generation.\n\tif id != \"owner1-ast1-chat1\" {\n\t\tt.Errorf(\"metadata override: got %q, want %q\", id, \"owner1-ast1-chat1\")\n\t}\n}\n\nfunc TestBuildIdentifier_MetadataEmptyIgnored(t *testing.T) {\n\tcfg := &types.SandboxConfig{Lifecycle: \"session\"}\n\tmeta := map[string]any{\"computer_id\": \"\"}\n\tid := sandboxv2.BuildIdentifier(cfg, \"owner1\", \"chat42\", \"ast1\", \"ws1\", meta)\n\tif id != \"owner1-ast1-chat42\" {\n\t\tt.Errorf(\"empty metadata should fall through to session, got %q\", id)\n\t}\n}\n\nfunc TestBuildIdentifier_UnknownLifecycle(t *testing.T) {\n\tcfg := &types.SandboxConfig{Lifecycle: \"unknown\"}\n\tid := sandboxv2.BuildIdentifier(cfg, \"owner1\", \"chat1\", \"ast1\", \"ws1\", nil)\n\tif id != \"\" {\n\t\tt.Errorf(\"unknown lifecycle should return empty, got %q\", id)\n\t}\n}\n\n// ===========================================================================\n// GetComputer — real container tests\n// ===========================================================================\n\nfunc makeAgentCtx(teamID, userID, chatID, assistantID string, metadata map[string]any) *agentContext.Context {\n\tvar auth *oauthTypes.AuthorizedInfo\n\tif teamID != \"\" || userID != \"\" {\n\t\tauth = &oauthTypes.AuthorizedInfo{TeamID: teamID, UserID: userID}\n\t}\n\treturn &agentContext.Context{\n\t\tContext: context.Background(),\n\t\tAuthorized: auth,\n\t\tChatID: chatID,\n\t\tAssistantID: assistantID,\n\t\tMetadata: metadata,\n\t}\n}\n\nfunc TestGetComputer_BoxCreate(t *testing.T) {\n\tskipIfNoDocker(t)\n\n\tfor _, nc := range boxNodes() {\n\t\tnc := nc\n\t\tt.Run(nc.Name, func(t *testing.T) {\n\t\t\tm := setupManager(t, &nc)\n\t\t\tensureImage(t, m, nc)\n\n\t\t\twsID := fmt.Sprintf(\"lc-create-%d\", time.Now().UnixNano())\n\t\t\tcreateTestWorkspace(t, nc.TaiID, wsID)\n\n\t\t\tcfg := &types.SandboxConfig{\n\t\t\t\tVersion: \"2.0\",\n\t\t\t\tLifecycle: \"oneshot\",\n\t\t\t\tComputer: types.ComputerConfig{Image: testImage()},\n\t\t\t\tNodeID: nc.TaiID,\n\t\t\t}\n\t\t\tmeta := map[string]any{\"workspace_id\": wsID}\n\t\t\tctx := makeAgentCtx(\"team-t1\", \"\", \"chat-1\", \"ast-1\", meta)\n\n\t\t\tcomputer, identifier, err := sandboxv2.GetComputer(ctx, cfg, m)\n\t\t\tif err != nil {\n\t\t\t\tt.Fatalf(\"GetComputer: %v\", err)\n\t\t\t}\n\t\t\tdefer cleanupComputer(t, m, cfg)\n\n\t\t\tif identifier == \"\" {\n\t\t\t\tt.Fatal(\"oneshot should get a random identifier, got empty\")\n\t\t\t}\n\n\t\t\tinfo := computer.ComputerInfo()\n\t\t\tif info.Kind != \"box\" {\n\t\t\t\tt.Errorf(\"kind = %q, want %q\", info.Kind, \"box\")\n\t\t\t}\n\t\t\tif cfg.Owner != \"team-t1\" {\n\t\t\t\tt.Errorf(\"cfg.Owner = %q, want %q\", cfg.Owner, \"team-t1\")\n\t\t\t}\n\t\t\tif cfg.Kind != \"box\" {\n\t\t\t\tt.Errorf(\"cfg.Kind = %q, want %q\", cfg.Kind, \"box\")\n\t\t\t}\n\t\t})\n\t}\n}\n\nfunc TestGetComputer_BoxReuse(t *testing.T) {\n\tskipIfNoDocker(t)\n\n\tfor _, nc := range boxNodes() {\n\t\tnc := nc\n\t\tt.Run(nc.Name, func(t *testing.T) {\n\t\t\tm := setupManager(t, &nc)\n\t\t\tensureImage(t, m, nc)\n\n\t\t\twsID := fmt.Sprintf(\"lc-reuse-%d\", time.Now().UnixNano())\n\t\t\tcreateTestWorkspace(t, nc.TaiID, wsID)\n\n\t\t\tcfg := &types.SandboxConfig{\n\t\t\t\tVersion: \"2.0\",\n\t\t\t\tLifecycle: \"session\",\n\t\t\t\tComputer: types.ComputerConfig{Image: testImage()},\n\t\t\t\tNodeID: nc.TaiID,\n\t\t\t}\n\t\t\tmeta := map[string]any{\"workspace_id\": wsID}\n\t\t\tctx := makeAgentCtx(\"team-reuse\", \"\", \"chat-reuse\", \"ast-1\", meta)\n\n\t\t\tcomputer1, id1, err := sandboxv2.GetComputer(ctx, cfg, m)\n\t\t\tif err != nil {\n\t\t\t\tt.Fatalf(\"first GetComputer: %v\", err)\n\t\t\t}\n\t\t\tdefer cleanupComputer(t, m, cfg)\n\n\t\t\tcfg2 := &types.SandboxConfig{\n\t\t\t\tVersion: \"2.0\",\n\t\t\t\tLifecycle: \"session\",\n\t\t\t\tComputer: types.ComputerConfig{Image: testImage()},\n\t\t\t\tNodeID: nc.TaiID,\n\t\t\t}\n\t\t\tcomputer2, id2, err := sandboxv2.GetComputer(ctx, cfg2, m)\n\t\t\tif err != nil {\n\t\t\t\tt.Fatalf(\"second GetComputer: %v\", err)\n\t\t\t}\n\n\t\t\tif id1 != id2 {\n\t\t\t\tt.Errorf(\"identifiers differ: %q vs %q\", id1, id2)\n\t\t\t}\n\n\t\t\tinfo1 := computer1.ComputerInfo()\n\t\t\tinfo2 := computer2.ComputerInfo()\n\t\t\tif info1.ContainerID != info2.ContainerID {\n\t\t\t\tt.Errorf(\"container IDs differ: %q vs %q (should reuse)\", info1.ContainerID, info2.ContainerID)\n\t\t\t}\n\t\t})\n\t}\n}\n\nfunc TestGetComputer_WorkspaceBindAlways(t *testing.T) {\n\tskipIfNoDocker(t)\n\n\tfor _, nc := range boxNodes() {\n\t\tnc := nc\n\t\tt.Run(nc.Name, func(t *testing.T) {\n\t\t\tm := setupManager(t, &nc)\n\t\t\tensureImage(t, m, nc)\n\n\t\t\twsID := fmt.Sprintf(\"lc-ws-%d\", time.Now().UnixNano())\n\t\t\tcreateTestWorkspace(t, nc.TaiID, wsID)\n\n\t\t\tcfg := &types.SandboxConfig{\n\t\t\t\tVersion: \"2.0\",\n\t\t\t\tLifecycle: \"oneshot\",\n\t\t\t\tComputer: types.ComputerConfig{Image: testImage()},\n\t\t\t\tNodeID: nc.TaiID,\n\t\t\t}\n\t\t\tmeta := map[string]any{\"workspace_id\": wsID}\n\t\t\tctx := makeAgentCtx(\"team-ws\", \"\", \"chat-ws\", \"ast-ws\", meta)\n\n\t\t\tcomputer, _, err := sandboxv2.GetComputer(ctx, cfg, m)\n\t\t\tif err != nil {\n\t\t\t\tt.Fatalf(\"GetComputer: %v\", err)\n\t\t\t}\n\t\t\tdefer cleanupComputer(t, m, cfg)\n\n\t\t\tif cfg.WorkspaceID != wsID {\n\t\t\t\tt.Errorf(\"WorkspaceID = %q, want %q\", cfg.WorkspaceID, wsID)\n\t\t\t}\n\n\t\t\tws := computer.Workplace()\n\t\t\tif ws == nil {\n\t\t\t\tt.Fatal(\"Workplace() returned nil, workspace should always be bound\")\n\t\t\t}\n\t\t})\n\t}\n}\n\nfunc TestGetComputer_WorkspaceFallbackOwner(t *testing.T) {\n\tskipIfNoDocker(t)\n\n\tfor _, nc := range boxNodes() {\n\t\tnc := nc\n\t\tt.Run(nc.Name, func(t *testing.T) {\n\t\t\tm := setupManager(t, &nc)\n\t\t\tensureImage(t, m, nc)\n\n\t\t\townerID := fmt.Sprintf(\"lc-owner-%d\", time.Now().UnixNano())\n\t\t\tcreateTestWorkspace(t, nc.TaiID, ownerID)\n\n\t\t\tcfg := &types.SandboxConfig{\n\t\t\t\tVersion: \"2.0\",\n\t\t\t\tLifecycle: \"oneshot\",\n\t\t\t\tComputer: types.ComputerConfig{Image: testImage()},\n\t\t\t\tNodeID: nc.TaiID,\n\t\t\t}\n\t\t\tctx := makeAgentCtx(ownerID, \"\", \"chat-fb\", \"ast-fb\", nil)\n\n\t\t\tcomputer, _, err := sandboxv2.GetComputer(ctx, cfg, m)\n\t\t\tif err != nil {\n\t\t\t\tt.Fatalf(\"GetComputer: %v\", err)\n\t\t\t}\n\t\t\tdefer cleanupComputer(t, m, cfg)\n\n\t\t\tif cfg.WorkspaceID != ownerID {\n\t\t\t\tt.Errorf(\"WorkspaceID = %q, want %q (should fallback to ownerID)\", cfg.WorkspaceID, ownerID)\n\t\t\t}\n\n\t\t\tws := computer.Workplace()\n\t\t\tif ws == nil {\n\t\t\t\tt.Fatal(\"Workplace() returned nil\")\n\t\t\t}\n\t\t})\n\t}\n}\n\nfunc TestGetComputer_OwnerPriority(t *testing.T) {\n\tskipIfNoDocker(t)\n\n\tnc := boxNodes()[0]\n\tm := setupManager(t, &nc)\n\tensureImage(t, m, nc)\n\n\tt.Run(\"teamID\", func(t *testing.T) {\n\t\twsID := fmt.Sprintf(\"lc-ownp-team-%d\", time.Now().UnixNano())\n\t\tcreateTestWorkspace(t, nc.TaiID, wsID)\n\n\t\tcfg := &types.SandboxConfig{\n\t\t\tVersion: \"2.0\", Lifecycle: \"oneshot\",\n\t\t\tComputer: types.ComputerConfig{Image: testImage()},\n\t\t\tNodeID: nc.TaiID,\n\t\t}\n\t\tctx := makeAgentCtx(\"my-team\", \"my-user\", \"c\", \"a\", map[string]any{\"workspace_id\": wsID})\n\t\t_, _, err := sandboxv2.GetComputer(ctx, cfg, m)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"GetComputer: %v\", err)\n\t\t}\n\t\tdefer cleanupComputer(t, m, cfg)\n\t\tif cfg.Owner != \"my-team\" {\n\t\t\tt.Errorf(\"Owner = %q, want %q (teamID takes precedence)\", cfg.Owner, \"my-team\")\n\t\t}\n\t})\n\n\tt.Run(\"userID\", func(t *testing.T) {\n\t\twsID := fmt.Sprintf(\"lc-ownp-user-%d\", time.Now().UnixNano())\n\t\tcreateTestWorkspace(t, nc.TaiID, wsID)\n\n\t\tcfg := &types.SandboxConfig{\n\t\t\tVersion: \"2.0\", Lifecycle: \"oneshot\",\n\t\t\tComputer: types.ComputerConfig{Image: testImage()},\n\t\t\tNodeID: nc.TaiID,\n\t\t}\n\t\tctx := makeAgentCtx(\"\", \"my-user\", \"c\", \"a\", map[string]any{\"workspace_id\": wsID})\n\t\t_, _, err := sandboxv2.GetComputer(ctx, cfg, m)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"GetComputer: %v\", err)\n\t\t}\n\t\tdefer cleanupComputer(t, m, cfg)\n\t\tif cfg.Owner != \"my-user\" {\n\t\t\tt.Errorf(\"Owner = %q, want %q\", cfg.Owner, \"my-user\")\n\t\t}\n\t})\n\n\tt.Run(\"anonymous\", func(t *testing.T) {\n\t\twsID := fmt.Sprintf(\"lc-ownp-anon-%d\", time.Now().UnixNano())\n\t\tcreateTestWorkspace(t, nc.TaiID, wsID)\n\n\t\tcfg := &types.SandboxConfig{\n\t\t\tVersion: \"2.0\", Lifecycle: \"oneshot\",\n\t\t\tComputer: types.ComputerConfig{Image: testImage()},\n\t\t\tNodeID: nc.TaiID,\n\t\t}\n\t\tctx := makeAgentCtx(\"\", \"\", \"c\", \"a\", map[string]any{\"workspace_id\": wsID})\n\t\t_, _, err := sandboxv2.GetComputer(ctx, cfg, m)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"GetComputer: %v\", err)\n\t\t}\n\t\tdefer cleanupComputer(t, m, cfg)\n\t\tif cfg.Owner != \"anonymous\" {\n\t\t\tt.Errorf(\"Owner = %q, want %q\", cfg.Owner, \"anonymous\")\n\t\t}\n\t})\n}\n\nfunc TestGetComputer_HostMode(t *testing.T) {\n\tskipIfNoHostExec(t)\n\n\tfor _, tgt := range hostTargets() {\n\t\ttgt := tgt\n\t\tt.Run(tgt.Name, func(t *testing.T) {\n\t\t\tm := setupHostManager(t, &tgt)\n\n\t\t\tcfg := &types.SandboxConfig{\n\t\t\t\tVersion: \"2.0\",\n\t\t\t\tLifecycle: \"session\",\n\t\t\t\tComputer: types.ComputerConfig{},\n\t\t\t\tNodeID: tgt.TaiID,\n\t\t\t}\n\t\t\tctx := makeAgentCtx(\"team-host\", \"\", \"chat-host\", \"ast-host\", nil)\n\n\t\t\tcomputer, _, err := sandboxv2.GetComputer(ctx, cfg, m)\n\t\t\tif err != nil {\n\t\t\t\tt.Fatalf(\"GetComputer host: %v\", err)\n\t\t\t}\n\n\t\t\tif cfg.Kind != \"host\" {\n\t\t\t\tt.Errorf(\"Kind = %q, want %q\", cfg.Kind, \"host\")\n\t\t\t}\n\n\t\t\tinfo := computer.ComputerInfo()\n\t\t\tif info.Kind != \"host\" {\n\t\t\t\tt.Errorf(\"ComputerInfo.Kind = %q, want %q\", info.Kind, \"host\")\n\t\t\t}\n\n\t\t\tws := computer.Workplace()\n\t\t\tif ws == nil {\n\t\t\t\tt.Fatal(\"Workplace() returned nil on host mode\")\n\t\t\t}\n\t\t})\n\t}\n}\n\nfunc TestGetComputer_HostMissingNodeID(t *testing.T) {\n\tskipIfNoDocker(t)\n\n\tnc := boxNodes()[0]\n\tm := setupManager(t, &nc)\n\n\tcfg := &types.SandboxConfig{\n\t\tVersion: \"2.0\",\n\t\tLifecycle: \"session\",\n\t\tComputer: types.ComputerConfig{},\n\t\tNodeID: \"\",\n\t}\n\tctx := makeAgentCtx(\"team-err\", \"\", \"c\", \"a\", nil)\n\n\t_, _, err := sandboxv2.GetComputer(ctx, cfg, m)\n\tif err == nil {\n\t\tt.Fatal(\"expected error for host mode without nodeID\")\n\t}\n\tif !strings.Contains(err.Error(), \"nodeID\") {\n\t\tt.Errorf(\"error should mention nodeID, got: %v\", err)\n\t}\n}\n\n// ===========================================================================\n// LifecycleAction — behavior tests\n// ===========================================================================\n\nfunc TestLifecycleAction_Oneshot(t *testing.T) {\n\tskipIfNoDocker(t)\n\n\tfor _, nc := range boxNodes() {\n\t\tnc := nc\n\t\tt.Run(nc.Name, func(t *testing.T) {\n\t\t\tm := setupManager(t, &nc)\n\t\t\tensureImage(t, m, nc)\n\n\t\t\twsID := fmt.Sprintf(\"lc-oneshot-%d\", time.Now().UnixNano())\n\t\t\tcreateTestWorkspace(t, nc.TaiID, wsID)\n\n\t\t\tcfg := &types.SandboxConfig{\n\t\t\t\tVersion: \"2.0\",\n\t\t\t\tLifecycle: \"oneshot\",\n\t\t\t\tComputer: types.ComputerConfig{Image: testImage()},\n\t\t\t\tNodeID: nc.TaiID,\n\t\t\t}\n\t\t\tctx := makeAgentCtx(\"team-oneshot\", \"\", \"c\", \"a\", map[string]any{\"workspace_id\": wsID})\n\n\t\t\tcomputer, _, err := sandboxv2.GetComputer(ctx, cfg, m)\n\t\t\tif err != nil {\n\t\t\t\tt.Fatalf(\"GetComputer: %v\", err)\n\t\t\t}\n\n\t\t\tboxID := cfg.ID\n\n\t\t\tsandboxv2.LifecycleAction(context.Background(), cfg, computer, m)\n\n\t\t\t_, getErr := m.Get(context.Background(), boxID)\n\t\t\tif getErr == nil {\n\t\t\t\tt.Error(\"box should be removed after oneshot LifecycleAction\")\n\t\t\t}\n\t\t})\n\t}\n}\n\nfunc TestLifecycleAction_Session(t *testing.T) {\n\tskipIfNoDocker(t)\n\n\tfor _, nc := range boxNodes() {\n\t\tnc := nc\n\t\tt.Run(nc.Name, func(t *testing.T) {\n\t\t\tm := setupManager(t, &nc)\n\t\t\tensureImage(t, m, nc)\n\n\t\t\twsID := fmt.Sprintf(\"lc-sess-%d\", time.Now().UnixNano())\n\t\t\tcreateTestWorkspace(t, nc.TaiID, wsID)\n\n\t\t\tcfg := &types.SandboxConfig{\n\t\t\t\tVersion: \"2.0\",\n\t\t\t\tLifecycle: \"session\",\n\t\t\t\tComputer: types.ComputerConfig{Image: testImage()},\n\t\t\t\tNodeID: nc.TaiID,\n\t\t\t}\n\t\t\tctx := makeAgentCtx(\"team-sess\", \"\", \"chat-sess\", \"ast-sess\", map[string]any{\"workspace_id\": wsID})\n\n\t\t\tcomputer, _, err := sandboxv2.GetComputer(ctx, cfg, m)\n\t\t\tif err != nil {\n\t\t\t\tt.Fatalf(\"GetComputer: %v\", err)\n\t\t\t}\n\t\t\tdefer cleanupComputer(t, m, cfg)\n\n\t\t\tsandboxv2.LifecycleAction(context.Background(), cfg, computer, m)\n\n\t\t\tbox, err := m.Get(context.Background(), cfg.ID)\n\t\t\tif err != nil {\n\t\t\t\tt.Fatalf(\"box should still exist after session LifecycleAction: %v\", err)\n\t\t\t}\n\t\t\tif box == nil {\n\t\t\t\tt.Fatal(\"box is nil after session LifecycleAction\")\n\t\t\t}\n\t\t})\n\t}\n}\n\nfunc TestLifecycleAction_Persistent(t *testing.T) {\n\tskipIfNoDocker(t)\n\n\tfor _, nc := range boxNodes() {\n\t\tnc := nc\n\t\tt.Run(nc.Name, func(t *testing.T) {\n\t\t\tm := setupManager(t, &nc)\n\t\t\tensureImage(t, m, nc)\n\n\t\t\twsID := fmt.Sprintf(\"lc-pers-%d\", time.Now().UnixNano())\n\t\t\tcreateTestWorkspace(t, nc.TaiID, wsID)\n\n\t\t\tcfg := &types.SandboxConfig{\n\t\t\t\tVersion: \"2.0\",\n\t\t\t\tLifecycle: \"persistent\",\n\t\t\t\tComputer: types.ComputerConfig{Image: testImage()},\n\t\t\t\tNodeID: nc.TaiID,\n\t\t\t}\n\t\t\tctx := makeAgentCtx(\"team-pers\", \"\", \"chat-pers\", \"ast-pers\", map[string]any{\"workspace_id\": wsID})\n\n\t\t\tcomputer, _, err := sandboxv2.GetComputer(ctx, cfg, m)\n\t\t\tif err != nil {\n\t\t\t\tt.Fatalf(\"GetComputer: %v\", err)\n\t\t\t}\n\t\t\tdefer cleanupComputer(t, m, cfg)\n\n\t\t\tsandboxv2.LifecycleAction(context.Background(), cfg, computer, m)\n\n\t\t\tbox, err := m.Get(context.Background(), cfg.ID)\n\t\t\tif err != nil {\n\t\t\t\tt.Fatalf(\"box should still exist after persistent LifecycleAction: %v\", err)\n\t\t\t}\n\t\t\tif box == nil {\n\t\t\t\tt.Fatal(\"box is nil after persistent LifecycleAction\")\n\t\t\t}\n\t\t})\n\t}\n}\n\nfunc TestLifecycleAction_NilSafe(t *testing.T) {\n\tcfg := &types.SandboxConfig{Lifecycle: \"oneshot\"}\n\tsandboxv2.LifecycleAction(context.Background(), cfg, nil, nil)\n\tsandboxv2.LifecycleAction(context.Background(), nil, nil, nil)\n}\n\n// ===========================================================================\n// helpers\n// ===========================================================================\n\nfunc ensureImage(t *testing.T, m *infra.Manager, nc nodeConfig) {\n\tt.Helper()\n\tctx, cancel := context.WithTimeout(context.Background(), 60*time.Second)\n\tdefer cancel()\n\tif err := m.EnsureImage(ctx, nc.TaiID, testImage(), infra.ImagePullOptions{}); err != nil {\n\t\tt.Fatalf(\"EnsureImage: %v\", err)\n\t}\n}\n\nfunc cleanupComputer(t *testing.T, m *infra.Manager, cfg *types.SandboxConfig) {\n\tt.Helper()\n\tif cfg.ID == \"\" {\n\t\treturn\n\t}\n\tctx, cancel := context.WithTimeout(context.Background(), 15*time.Second)\n\tdefer cancel()\n\tif err := m.Remove(ctx, cfg.ID); err != nil {\n\t\tt.Logf(\"cleanup Remove(%s): %v\", cfg.ID, err)\n\t}\n}\n" }, { "path": "agent/sandbox/v2/options.go", "content": "package sandboxv2\n\nimport (\n\t\"encoding/json\"\n\t\"fmt\"\n\t\"os\"\n\t\"strings\"\n\t\"time\"\n\n\t\"github.com/yaoapp/gou/connector\"\n\t\"github.com/yaoapp/yao/agent/sandbox/v2/types\"\n\tinfra \"github.com/yaoapp/yao/sandbox/v2\"\n)\n\n// resolveEnvRef resolves $ENV.XXX references to os.Getenv(\"XXX\").\nfunc resolveEnvRef(value string) string {\n\tif strings.HasPrefix(value, \"$ENV.\") {\n\t\treturn os.Getenv(value[5:])\n\t}\n\treturn value\n}\n\n// BuildCreateOptions converts a SandboxConfig into the V2 infrastructure\n// CreateOptions. An optional connector is used to inject OPENAI_PROXY_*\n// environment variables when the connector is OpenAI-compatible (non-Anthropic).\nfunc BuildCreateOptions(cfg *types.SandboxConfig, identifier, ownerID, workspaceID string, conn ...connector.Connector) (infra.CreateOptions, error) {\n\topts := infra.CreateOptions{\n\t\tID: identifier,\n\t\tOwner: ownerID,\n\t\tImage: cfg.Computer.Image,\n\t\tWorkDir: cfg.Computer.WorkDir,\n\t\tUser: cfg.Computer.User,\n\t\tMountPath: cfg.Computer.MountPath,\n\t\tMountMode: cfg.Computer.MountMode,\n\t\tWorkspaceID: workspaceID,\n\t\tLabels: cfg.Labels,\n\t\tDisplayName: cfg.DisplayName,\n\t}\n\n\tif opts.Labels == nil {\n\t\topts.Labels = make(map[string]string)\n\t}\n\n\t// Lifecycle policy\n\tswitch cfg.Lifecycle {\n\tcase \"oneshot\":\n\t\topts.Policy = infra.OneShot\n\tcase \"session\":\n\t\topts.Policy = infra.Session\n\tcase \"longrunning\":\n\t\topts.Policy = infra.LongRunning\n\tcase \"persistent\":\n\t\topts.Policy = infra.Persistent\n\tdefault:\n\t\topts.Policy = infra.OneShot\n\t}\n\n\t// Timeouts\n\tif cfg.IdleTimeout != \"\" {\n\t\td, err := time.ParseDuration(cfg.IdleTimeout)\n\t\tif err != nil {\n\t\t\treturn opts, fmt.Errorf(\"idle_timeout: %w\", err)\n\t\t}\n\t\topts.IdleTimeout = d\n\t}\n\tif opts.IdleTimeout == 0 {\n\t\tswitch opts.Policy {\n\t\tcase infra.Session:\n\t\t\topts.IdleTimeout = infra.DefaultSessionIdleTimeout\n\t\tcase infra.LongRunning:\n\t\t\topts.IdleTimeout = infra.DefaultLongRunningIdleTimeout\n\t\t}\n\t}\n\tif cfg.MaxLifetime != \"\" {\n\t\td, err := time.ParseDuration(cfg.MaxLifetime)\n\t\tif err != nil {\n\t\t\treturn opts, fmt.Errorf(\"max_lifetime: %w\", err)\n\t\t}\n\t\topts.MaxLifetime = d\n\t}\n\tif cfg.StopTimeout != \"\" {\n\t\td, err := time.ParseDuration(cfg.StopTimeout)\n\t\tif err != nil {\n\t\t\treturn opts, fmt.Errorf(\"stop_timeout: %w\", err)\n\t\t}\n\t\topts.StopTimeout = d\n\t}\n\n\t// Memory (string like \"4g\" → bytes)\n\tif cfg.Computer.Memory != \"\" {\n\t\tmem, err := parseMemory(cfg.Computer.Memory)\n\t\tif err != nil {\n\t\t\treturn opts, fmt.Errorf(\"memory: %w\", err)\n\t\t}\n\t\topts.Memory = mem\n\t}\n\n\topts.CPUs = cfg.Computer.CPUs\n\n\t// VNC\n\topts.VNC = cfg.Computer.VNC.Enabled\n\n\t// Ports\n\tfor _, p := range cfg.Computer.Ports {\n\t\topts.Ports = append(opts.Ports, infra.PortMapping{\n\t\t\tContainerPort: p.Port,\n\t\t\tHostPort: p.HostPort,\n\t\t\tProtocol: p.Protocol,\n\t\t})\n\t}\n\n\t// NodeID (host mode pre-selection)\n\tif cfg.NodeID != \"\" {\n\t\topts.NodeID = cfg.NodeID\n\t}\n\n\t// Merge environment + secrets into CreateOptions.Env.\n\t// Secrets override environment for same-name keys.\n\t// $ENV.XXX references are resolved at runtime.\n\tenvSize := len(cfg.Environment) + len(cfg.Secrets)\n\tif envSize > 0 {\n\t\topts.Env = make(map[string]string, envSize)\n\t\tfor k, v := range cfg.Environment {\n\t\t\topts.Env[k] = resolveEnvRef(v)\n\t\t}\n\t\tfor k, v := range cfg.Secrets {\n\t\t\topts.Env[k] = resolveEnvRef(v)\n\t\t}\n\t}\n\n\tif opts.Env == nil {\n\t\topts.Env = make(map[string]string)\n\t}\n\n\t// Inject OPENAI_PROXY_* when connector is OpenAI-compatible (non-Anthropic).\n\t// The a2o proxy inside the container translates Anthropic API → OpenAI API.\n\tif len(conn) > 0 && conn[0] != nil && !conn[0].Is(connector.ANTHROPIC) {\n\t\tinjectProxyEnv(opts.Env, conn[0])\n\t}\n\n\t// Inject VNC_* environment variables from config.\n\tif cfg.Computer.VNC.Enabled {\n\t\topts.Env[\"VNC_ENABLED\"] = \"true\"\n\t\tif cfg.Computer.VNC.Password != \"\" {\n\t\t\topts.Env[\"VNC_PASSWORD\"] = resolveEnvRef(cfg.Computer.VNC.Password)\n\t\t}\n\t\tif cfg.Computer.VNC.Resolution != \"\" {\n\t\t\topts.Env[\"VNC_RESOLUTION\"] = cfg.Computer.VNC.Resolution\n\t\t}\n\t\tif cfg.Computer.VNC.ViewOnly {\n\t\t\topts.Env[\"VNC_VIEW_ONLY\"] = \"true\"\n\t\t}\n\t}\n\n\treturn opts, nil\n}\n\n// injectProxyEnv extracts backend URL, model, and API key from an\n// OpenAI-compatible connector's settings and writes them as OPENAI_PROXY_*\n// environment variables into env.\nfunc injectProxyEnv(env map[string]string, conn connector.Connector) {\n\tsettings := conn.Setting()\n\tif settings == nil {\n\t\treturn\n\t}\n\n\tif host, ok := settings[\"host\"].(string); ok && host != \"\" {\n\t\tenv[\"OPENAI_PROXY_BACKEND\"] = host\n\t}\n\tif model, ok := settings[\"model\"].(string); ok && model != \"\" {\n\t\tenv[\"OPENAI_PROXY_MODEL\"] = model\n\t}\n\tif key, ok := settings[\"key\"].(string); ok && key != \"\" {\n\t\tenv[\"OPENAI_PROXY_API_KEY\"] = key\n\t}\n\n\t// Forward extra options as JSON.\n\textra := make(map[string]interface{})\n\tfor k, v := range settings {\n\t\tswitch k {\n\t\tcase \"host\", \"model\", \"key\", \"proxy\", \"type\":\n\t\t\tcontinue\n\t\tdefault:\n\t\t\textra[k] = v\n\t\t}\n\t}\n\tif len(extra) > 0 {\n\t\tif data, err := json.Marshal(extra); err == nil {\n\t\t\tenv[\"OPENAI_PROXY_OPTIONS\"] = string(data)\n\t\t}\n\t}\n}\n\n// parseMemory converts a human-readable memory string to bytes.\n// Supported formats: \"4GB\", \"4G\", \"4g\", \"512MB\", \"512M\", \"512m\", \"1024KB\", \"1024K\", \"1024\".\nfunc parseMemory(s string) (int64, error) {\n\tif len(s) == 0 {\n\t\treturn 0, nil\n\t}\n\n\tupper := strings.ToUpper(s)\n\tvar num string\n\tvar multiplier int64\n\n\tswitch {\n\tcase strings.HasSuffix(upper, \"GB\"):\n\t\tnum = s[:len(s)-2]\n\t\tmultiplier = 1 << 30\n\tcase strings.HasSuffix(upper, \"MB\"):\n\t\tnum = s[:len(s)-2]\n\t\tmultiplier = 1 << 20\n\tcase strings.HasSuffix(upper, \"KB\"):\n\t\tnum = s[:len(s)-2]\n\t\tmultiplier = 1 << 10\n\tcase strings.HasSuffix(upper, \"TB\"):\n\t\tnum = s[:len(s)-2]\n\t\tmultiplier = 1 << 40\n\tcase strings.HasSuffix(upper, \"G\"):\n\t\tnum = s[:len(s)-1]\n\t\tmultiplier = 1 << 30\n\tcase strings.HasSuffix(upper, \"M\"):\n\t\tnum = s[:len(s)-1]\n\t\tmultiplier = 1 << 20\n\tcase strings.HasSuffix(upper, \"K\"):\n\t\tnum = s[:len(s)-1]\n\t\tmultiplier = 1 << 10\n\tcase strings.HasSuffix(upper, \"T\"):\n\t\tnum = s[:len(s)-1]\n\t\tmultiplier = 1 << 40\n\tdefault:\n\t\tnum = s\n\t\tmultiplier = 1\n\t}\n\n\tvar val float64\n\tif _, err := fmt.Sscanf(num, \"%f\", &val); err != nil {\n\t\treturn 0, fmt.Errorf(\"invalid memory value %q\", s)\n\t}\n\treturn int64(val * float64(multiplier)), nil\n}\n" }, { "path": "agent/sandbox/v2/prepare.go", "content": "package sandboxv2\n\nimport (\n\t\"context\"\n\t\"fmt\"\n\t\"log\"\n\t\"path\"\n\tpathpkg \"path/filepath\"\n\t\"strings\"\n\n\t\"github.com/yaoapp/yao/agent/sandbox/v2/types\"\n\tinfra \"github.com/yaoapp/yao/sandbox/v2\"\n\t\"github.com/yaoapp/yao/tai/workspace\"\n)\n\nconst onceMarkerDir = \".yao/prepare\"\n\n// RunPrepareSteps executes a list of PrepareStep actions on the given Computer.\n// file/copy/marker operations use computer.Workplace() (gRPC volume, cross-platform).\n// exec operations use shell via Computer.Exec.\n// assistantDir is the absolute host path to the assistant source directory;\n// copy steps with a relative src resolve against it (host → workspace push).\nfunc RunPrepareSteps(ctx context.Context, steps []types.PrepareStep, computer infra.Computer, assistantID, configHash, assistantDir string) error {\n\tif len(steps) == 0 {\n\t\treturn nil\n\t}\n\n\tvar ws workspace.FS\n\tif computer != nil {\n\t\tws = computer.Workplace()\n\t}\n\n\tmarkerDir := onceMarkerDir\n\tif assistantID != \"\" {\n\t\tmarkerDir = onceMarkerDir + \"/\" + assistantID\n\t}\n\tmarkerPath := markerDir + \"/done\"\n\n\tskipOnce := false\n\tif configHash != \"\" && ws != nil {\n\t\tif data, err := ws.ReadFile(markerPath); err == nil {\n\t\t\tif strings.TrimSpace(string(data)) == configHash {\n\t\t\t\tskipOnce = true\n\t\t\t}\n\t\t}\n\t}\n\n\tfor i, step := range steps {\n\t\tif step.Once && skipOnce {\n\t\t\tcontinue\n\t\t}\n\n\t\tvar err error\n\t\tswitch step.Action {\n\t\tcase \"file\":\n\t\t\terr = runFileStep(ws, step)\n\t\tcase \"copy\":\n\t\t\terr = runCopyStep(ws, step, assistantDir)\n\t\tcase \"exec\":\n\t\t\terr = runExecStep(ctx, computer, step)\n\t\tcase \"process\":\n\t\t\tlog.Printf(\"[sandbox/v2] prepare step %d: action=process (reserved, skipping)\", i)\n\t\tdefault:\n\t\t\terr = fmt.Errorf(\"unknown prepare action %q\", step.Action)\n\t\t}\n\n\t\tif err != nil {\n\t\t\tif step.IgnoreError {\n\t\t\t\tlog.Printf(\"[sandbox/v2] prepare step %d (%s): ignored error: %v\", i, step.Action, err)\n\t\t\t\tcontinue\n\t\t\t}\n\t\t\treturn fmt.Errorf(\"prepare step %d (%s): %w\", i, step.Action, err)\n\t\t}\n\t}\n\n\tif configHash != \"\" && ws != nil {\n\t\tws.MkdirAll(markerDir, 0755)\n\t\tws.WriteFile(markerPath, []byte(configHash), 0644)\n\t}\n\n\treturn nil\n}\n\n// ---------------------------------------------------------------------------\n// Step runners\n// ---------------------------------------------------------------------------\n\nfunc runFileStep(ws workspace.FS, step types.PrepareStep) error {\n\tif step.Path == \"\" {\n\t\treturn fmt.Errorf(\"file step requires path\")\n\t}\n\tif ws == nil {\n\t\treturn fmt.Errorf(\"file step requires workspace\")\n\t}\n\n\tdir := path.Dir(step.Path)\n\tif dir != \".\" && dir != \"/\" {\n\t\tif err := ws.MkdirAll(dir, 0755); err != nil {\n\t\t\treturn fmt.Errorf(\"mkdir %s: %w\", dir, err)\n\t\t}\n\t}\n\n\tif err := ws.WriteFile(step.Path, step.Content, 0644); err != nil {\n\t\treturn fmt.Errorf(\"write file %s: %w\", step.Path, err)\n\t}\n\treturn nil\n}\n\n// runCopyStep copies files into the workspace using ws.Copy which supports\n// the \"local:///\" URI scheme for host-to-workspace transfers.\n//\n// src resolution:\n// - Already a host URI (\"local:///...\" or \"tmp:///...\") → used as-is\n// - Relative path + assistantDir provided → resolved to \"local:////\"\n// - Relative path without assistantDir → treated as workspace-internal path\nfunc runCopyStep(ws workspace.FS, step types.PrepareStep, assistantDir string) error {\n\tif step.Src == \"\" || step.Dst == \"\" {\n\t\treturn fmt.Errorf(\"copy step requires src and dst\")\n\t}\n\tif ws == nil {\n\t\treturn fmt.Errorf(\"copy step requires workspace\")\n\t}\n\n\tsrc := step.Src\n\tif !isHostURI(src) && assistantDir != \"\" {\n\t\tsrc = \"local:///\" + pathpkg.Join(assistantDir, src)\n\t}\n\n\tif _, err := ws.Copy(src, step.Dst); err != nil {\n\t\treturn fmt.Errorf(\"copy %s -> %s: %w\", src, step.Dst, err)\n\t}\n\treturn nil\n}\n\nfunc isHostURI(s string) bool {\n\treturn strings.HasPrefix(s, \"local:///\") || strings.HasPrefix(s, \"tmp:///\")\n}\n\nfunc runExecStep(ctx context.Context, computer infra.Computer, step types.PrepareStep) error {\n\tif step.Cmd == \"\" {\n\t\treturn fmt.Errorf(\"exec step requires cmd\")\n\t}\n\n\tkind := shellFromSystem(computer)\n\tscript := step.Cmd\n\tif step.Background {\n\t\tif kind == shellSh {\n\t\t\tscript = fmt.Sprintf(\"nohup %s > /dev/null 2>&1 &\", step.Cmd)\n\t\t} else {\n\t\t\tscript = fmt.Sprintf(\"Start-Process -NoNewWindow -FilePath 'cmd.exe' -ArgumentList '/C %s'\", step.Cmd)\n\t\t}\n\t}\n\n\trootDir := \"/\"\n\tif isWindowsComputer(computer) {\n\t\trootDir = `C:\\`\n\t}\n\n\tresult, err := computer.Exec(ctx, shellWrap(kind, script), infra.WithWorkDir(rootDir))\n\tif err != nil {\n\t\treturn err\n\t}\n\tlabel := \"exec\"\n\tif step.Background {\n\t\tlabel = \"exec(background)\"\n\t}\n\treturn checkResult(result, label)\n}\n\nfunc isWindowsComputer(computer infra.Computer) bool {\n\treturn strings.EqualFold(computer.ComputerInfo().System.OS, \"windows\")\n}\n\n// checkResult inspects ExecResult for errors.\nfunc checkResult(result *infra.ExecResult, label string) error {\n\tif result.Error != \"\" {\n\t\treturn fmt.Errorf(\"%s: %s\", label, result.Error)\n\t}\n\tif result.ExitCode != 0 {\n\t\tstderr := result.Stderr\n\t\tif len(stderr) > 200 {\n\t\t\tstderr = stderr[:200] + \"...\"\n\t\t}\n\t\treturn fmt.Errorf(\"%s: exit %d: %s\", label, result.ExitCode, stderr)\n\t}\n\treturn nil\n}\n" }, { "path": "agent/sandbox/v2/prepare_test.go", "content": "package sandboxv2_test\n\nimport (\n\t\"context\"\n\t\"fmt\"\n\t\"strings\"\n\t\"testing\"\n\t\"time\"\n\n\tsandboxv2 \"github.com/yaoapp/yao/agent/sandbox/v2\"\n\t\"github.com/yaoapp/yao/agent/sandbox/v2/types\"\n)\n\n// ---------------------------------------------------------------------------\n// Box tests (local + remote)\n// ---------------------------------------------------------------------------\n\nfunc TestRunPrepareSteps_Exec(t *testing.T) {\n\tskipIfNoDocker(t)\n\n\tfor _, nc := range boxNodes() {\n\t\tnc := nc\n\t\tt.Run(nc.Name, func(t *testing.T) {\n\t\t\tm := setupManager(t, &nc)\n\t\t\tbox := createBox(t, m, nc)\n\n\t\t\tctx, cancel := context.WithTimeout(context.Background(), 30*time.Second)\n\t\t\tdefer cancel()\n\n\t\t\tsteps := []types.PrepareStep{\n\t\t\t\t{Action: \"exec\", Cmd: \"echo hello > /tmp/prep-test\"},\n\t\t\t\t{Action: \"exec\", Cmd: \"echo world >> /tmp/prep-test\"},\n\t\t\t}\n\n\t\t\terr := sandboxv2.RunPrepareSteps(ctx, steps, box, \"test-assistant\", \"\", \"\")\n\t\t\tif err != nil {\n\t\t\t\tt.Fatalf(\"RunPrepareSteps: %v\", err)\n\t\t\t}\n\n\t\t\tresult, err := box.Exec(ctx, []string{\"cat\", \"/tmp/prep-test\"})\n\t\t\tif err != nil {\n\t\t\t\tt.Fatalf(\"cat: %v\", err)\n\t\t\t}\n\t\t\tgot := strings.TrimSpace(result.Stdout)\n\t\t\tif got != \"hello\\nworld\" {\n\t\t\t\tt.Errorf(\"content = %q, want %q\", got, \"hello\\nworld\")\n\t\t\t}\n\t\t})\n\t}\n}\n\nfunc TestRunPrepareSteps_File(t *testing.T) {\n\tskipIfNoDocker(t)\n\n\tfor _, nc := range boxNodes() {\n\t\tnc := nc\n\t\tt.Run(nc.Name, func(t *testing.T) {\n\t\t\tm := setupManager(t, &nc)\n\t\t\tbox := createBox(t, m, nc)\n\t\t\twsID := fmt.Sprintf(\"test-file-%d\", time.Now().UnixNano())\n\t\t\tbox.BindWorkplace(wsID)\n\n\t\t\tctx, cancel := context.WithTimeout(context.Background(), 30*time.Second)\n\t\t\tdefer cancel()\n\n\t\t\tsteps := []types.PrepareStep{\n\t\t\t\t{Action: \"file\", Path: \"config/test.txt\", Content: []byte(\"file-content-v2\")},\n\t\t\t}\n\n\t\t\terr := sandboxv2.RunPrepareSteps(ctx, steps, box, \"test-assistant\", \"\", \"\")\n\t\t\tif err != nil {\n\t\t\t\tt.Fatalf(\"RunPrepareSteps: %v\", err)\n\t\t\t}\n\n\t\t\tws := box.Workplace()\n\t\t\tdata, err := ws.ReadFile(\"config/test.txt\")\n\t\t\tif err != nil {\n\t\t\t\tt.Fatalf(\"ReadFile: %v\", err)\n\t\t\t}\n\t\t\tif string(data) != \"file-content-v2\" {\n\t\t\t\tt.Errorf(\"content = %q, want %q\", string(data), \"file-content-v2\")\n\t\t\t}\n\t\t})\n\t}\n}\n\nfunc TestRunPrepareSteps_Copy(t *testing.T) {\n\tskipIfNoDocker(t)\n\n\tfor _, nc := range boxNodes() {\n\t\tnc := nc\n\t\tt.Run(nc.Name, func(t *testing.T) {\n\t\t\tm := setupManager(t, &nc)\n\t\t\tbox := createBox(t, m, nc)\n\t\t\twsID := fmt.Sprintf(\"test-copy-%d\", time.Now().UnixNano())\n\t\t\tbox.BindWorkplace(wsID)\n\n\t\t\tws := box.Workplace()\n\t\t\tws.WriteFile(\"src.txt\", []byte(\"copy-src\"), 0644)\n\n\t\t\tctx, cancel := context.WithTimeout(context.Background(), 30*time.Second)\n\t\t\tdefer cancel()\n\n\t\t\tsteps := []types.PrepareStep{\n\t\t\t\t{Action: \"copy\", Src: \"src.txt\", Dst: \"dst.txt\"},\n\t\t\t}\n\n\t\t\terr := sandboxv2.RunPrepareSteps(ctx, steps, box, \"test-assistant\", \"\", \"\")\n\t\t\tif err != nil {\n\t\t\t\tt.Fatalf(\"RunPrepareSteps: %v\", err)\n\t\t\t}\n\n\t\t\tdata, err := ws.ReadFile(\"dst.txt\")\n\t\t\tif err != nil {\n\t\t\t\tt.Fatalf(\"ReadFile: %v\", err)\n\t\t\t}\n\t\t\tif string(data) != \"copy-src\" {\n\t\t\t\tt.Errorf(\"content = %q, want %q\", string(data), \"copy-src\")\n\t\t\t}\n\t\t})\n\t}\n}\n\nfunc TestRunPrepareSteps_OnceMarker(t *testing.T) {\n\tskipIfNoDocker(t)\n\n\tfor _, nc := range boxNodes() {\n\t\tnc := nc\n\t\tt.Run(nc.Name, func(t *testing.T) {\n\t\t\tm := setupManager(t, &nc)\n\t\t\tbox := createBox(t, m, nc)\n\t\t\twsID := fmt.Sprintf(\"test-once-%d\", time.Now().UnixNano())\n\t\t\tbox.BindWorkplace(wsID)\n\n\t\t\tctx, cancel := context.WithTimeout(context.Background(), 30*time.Second)\n\t\t\tdefer cancel()\n\n\t\t\tcounter := \"/tmp/once-counter\"\n\t\t\tsteps := []types.PrepareStep{\n\t\t\t\t{Action: \"exec\", Cmd: \"echo -n x >> \" + counter, Once: true},\n\t\t\t}\n\n\t\t\thash := \"abc123\"\n\t\t\tassistantID := \"test-once\"\n\n\t\t\tif err := sandboxv2.RunPrepareSteps(ctx, steps, box, assistantID, hash, \"\"); err != nil {\n\t\t\t\tt.Fatalf(\"first run: %v\", err)\n\t\t\t}\n\t\t\tr1, _ := box.Exec(ctx, []string{\"cat\", counter})\n\t\t\tif r1.Stdout != \"x\" {\n\t\t\t\tt.Fatalf(\"first run: got %q, want %q\", r1.Stdout, \"x\")\n\t\t\t}\n\n\t\t\tif err := sandboxv2.RunPrepareSteps(ctx, steps, box, assistantID, hash, \"\"); err != nil {\n\t\t\t\tt.Fatalf(\"second run: %v\", err)\n\t\t\t}\n\t\t\tr2, _ := box.Exec(ctx, []string{\"cat\", counter})\n\t\t\tif r2.Stdout != \"x\" {\n\t\t\t\tt.Errorf(\"second run: got %q, want %q (once step should be skipped)\", r2.Stdout, \"x\")\n\t\t\t}\n\n\t\t\tif err := sandboxv2.RunPrepareSteps(ctx, steps, box, assistantID, \"new-hash\", \"\"); err != nil {\n\t\t\t\tt.Fatalf(\"third run: %v\", err)\n\t\t\t}\n\t\t\tr3, _ := box.Exec(ctx, []string{\"cat\", counter})\n\t\t\tif r3.Stdout != \"xx\" {\n\t\t\t\tt.Errorf(\"third run: got %q, want %q (hash changed, should re-execute)\", r3.Stdout, \"xx\")\n\t\t\t}\n\t\t})\n\t}\n}\n\nfunc TestRunPrepareSteps_OnceIsolation(t *testing.T) {\n\tskipIfNoDocker(t)\n\n\tfor _, nc := range boxNodes() {\n\t\tnc := nc\n\t\tt.Run(nc.Name, func(t *testing.T) {\n\t\t\tm := setupManager(t, &nc)\n\t\t\tbox := createBox(t, m, nc)\n\t\t\twsID := fmt.Sprintf(\"test-iso-%d\", time.Now().UnixNano())\n\t\t\tbox.BindWorkplace(wsID)\n\n\t\t\tctx, cancel := context.WithTimeout(context.Background(), 30*time.Second)\n\t\t\tdefer cancel()\n\n\t\t\tstepsA := []types.PrepareStep{\n\t\t\t\t{Action: \"exec\", Cmd: \"echo -n A >> /tmp/iso-a\", Once: true},\n\t\t\t}\n\t\t\tstepsB := []types.PrepareStep{\n\t\t\t\t{Action: \"exec\", Cmd: \"echo -n B >> /tmp/iso-b\", Once: true},\n\t\t\t}\n\n\t\t\thash := \"same-hash\"\n\n\t\t\tif err := sandboxv2.RunPrepareSteps(ctx, stepsA, box, \"assistant-a\", hash, \"\"); err != nil {\n\t\t\t\tt.Fatalf(\"assistant-a: %v\", err)\n\t\t\t}\n\t\t\tif err := sandboxv2.RunPrepareSteps(ctx, stepsB, box, \"assistant-b\", hash, \"\"); err != nil {\n\t\t\t\tt.Fatalf(\"assistant-b: %v\", err)\n\t\t\t}\n\n\t\t\trA, _ := box.Exec(ctx, []string{\"cat\", \"/tmp/iso-a\"})\n\t\t\trB, _ := box.Exec(ctx, []string{\"cat\", \"/tmp/iso-b\"})\n\t\t\tif rA.Stdout != \"A\" {\n\t\t\t\tt.Errorf(\"assistant-a: got %q, want %q\", rA.Stdout, \"A\")\n\t\t\t}\n\t\t\tif rB.Stdout != \"B\" {\n\t\t\t\tt.Errorf(\"assistant-b: got %q, want %q\", rB.Stdout, \"B\")\n\t\t\t}\n\n\t\t\tif err := sandboxv2.RunPrepareSteps(ctx, stepsA, box, \"assistant-a\", hash, \"\"); err != nil {\n\t\t\t\tt.Fatalf(\"assistant-a re-run: %v\", err)\n\t\t\t}\n\t\t\tif err := sandboxv2.RunPrepareSteps(ctx, stepsB, box, \"assistant-b\", hash, \"\"); err != nil {\n\t\t\t\tt.Fatalf(\"assistant-b re-run: %v\", err)\n\t\t\t}\n\t\t\trA2, _ := box.Exec(ctx, []string{\"cat\", \"/tmp/iso-a\"})\n\t\t\trB2, _ := box.Exec(ctx, []string{\"cat\", \"/tmp/iso-b\"})\n\t\t\tif rA2.Stdout != \"A\" {\n\t\t\t\tt.Errorf(\"assistant-a re-run: got %q, want %q (should be skipped)\", rA2.Stdout, \"A\")\n\t\t\t}\n\t\t\tif rB2.Stdout != \"B\" {\n\t\t\t\tt.Errorf(\"assistant-b re-run: got %q, want %q (should be skipped)\", rB2.Stdout, \"B\")\n\t\t\t}\n\t\t})\n\t}\n}\n\nfunc TestRunPrepareSteps_IgnoreError(t *testing.T) {\n\tskipIfNoDocker(t)\n\n\tfor _, nc := range boxNodes() {\n\t\tnc := nc\n\t\tt.Run(nc.Name, func(t *testing.T) {\n\t\t\tm := setupManager(t, &nc)\n\t\t\tbox := createBox(t, m, nc)\n\n\t\t\tctx, cancel := context.WithTimeout(context.Background(), 30*time.Second)\n\t\t\tdefer cancel()\n\n\t\t\tsteps := []types.PrepareStep{\n\t\t\t\t{Action: \"exec\", Cmd: \"false\", IgnoreError: true},\n\t\t\t\t{Action: \"exec\", Cmd: \"echo survived > /tmp/survived\"},\n\t\t\t}\n\n\t\t\terr := sandboxv2.RunPrepareSteps(ctx, steps, box, \"test-assistant\", \"\", \"\")\n\t\t\tif err != nil {\n\t\t\t\tt.Fatalf(\"RunPrepareSteps: %v (ignore_error should have prevented failure)\", err)\n\t\t\t}\n\n\t\t\tresult, _ := box.Exec(ctx, []string{\"cat\", \"/tmp/survived\"})\n\t\t\tif strings.TrimSpace(result.Stdout) != \"survived\" {\n\t\t\t\tt.Errorf(\"second step should have executed, got %q\", result.Stdout)\n\t\t\t}\n\t\t})\n\t}\n}\n\nfunc TestRunPrepareSteps_FailOnError(t *testing.T) {\n\tskipIfNoDocker(t)\n\n\tfor _, nc := range boxNodes() {\n\t\tnc := nc\n\t\tt.Run(nc.Name, func(t *testing.T) {\n\t\t\tm := setupManager(t, &nc)\n\t\t\tbox := createBox(t, m, nc)\n\n\t\t\tctx, cancel := context.WithTimeout(context.Background(), 30*time.Second)\n\t\t\tdefer cancel()\n\n\t\t\tsteps := []types.PrepareStep{\n\t\t\t\t{Action: \"exec\", Cmd: \"false\"},\n\t\t\t\t{Action: \"exec\", Cmd: \"echo should-not-reach > /tmp/unreachable\"},\n\t\t\t}\n\n\t\t\terr := sandboxv2.RunPrepareSteps(ctx, steps, box, \"test-assistant\", \"\", \"\")\n\t\t\tif err == nil {\n\t\t\t\tt.Fatal(\"expected error from failing step without ignore_error\")\n\t\t\t}\n\n\t\t\tresult, _ := box.Exec(ctx, []string{\"cat\", \"/tmp/unreachable\"})\n\t\t\tif result.ExitCode == 0 {\n\t\t\t\tt.Error(\"second step should not have executed\")\n\t\t\t}\n\t\t})\n\t}\n}\n\nfunc TestRunPrepareSteps_UnknownAction(t *testing.T) {\n\tskipIfNoDocker(t)\n\n\tfor _, nc := range boxNodes() {\n\t\tnc := nc\n\t\tt.Run(nc.Name, func(t *testing.T) {\n\t\t\tm := setupManager(t, &nc)\n\t\t\t_ = createBox(t, m, nc)\n\n\t\t\tctx, cancel := context.WithTimeout(context.Background(), 30*time.Second)\n\t\t\tdefer cancel()\n\n\t\t\tsteps := []types.PrepareStep{\n\t\t\t\t{Action: \"unknown_action\"},\n\t\t\t}\n\n\t\t\terr := sandboxv2.RunPrepareSteps(ctx, steps, nil, \"test-assistant\", \"\", \"\")\n\t\t\tif err == nil {\n\t\t\t\tt.Fatal(\"expected error for unknown action\")\n\t\t\t}\n\t\t\tif !strings.Contains(err.Error(), \"unknown_action\") {\n\t\t\t\tt.Errorf(\"error should mention action name, got: %v\", err)\n\t\t\t}\n\t\t})\n\t}\n}\n\nfunc TestRunPrepareSteps_EmptySteps(t *testing.T) {\n\terr := sandboxv2.RunPrepareSteps(context.Background(), nil, nil, \"test-assistant\", \"hash\", \"\")\n\tif err != nil {\n\t\tt.Fatalf(\"empty steps should succeed: %v\", err)\n\t}\n}\n\nfunc TestRunPrepareSteps_Background(t *testing.T) {\n\tskipIfNoDocker(t)\n\n\tfor _, nc := range boxNodes() {\n\t\tnc := nc\n\t\tt.Run(nc.Name, func(t *testing.T) {\n\t\t\tm := setupManager(t, &nc)\n\t\t\tbox := createBox(t, m, nc)\n\n\t\t\tctx, cancel := context.WithTimeout(context.Background(), 30*time.Second)\n\t\t\tdefer cancel()\n\n\t\t\tsteps := []types.PrepareStep{\n\t\t\t\t{Action: \"exec\", Cmd: \"sleep 30\", Background: true},\n\t\t\t\t{Action: \"exec\", Cmd: \"echo after-bg > /tmp/after-bg\"},\n\t\t\t}\n\n\t\t\terr := sandboxv2.RunPrepareSteps(ctx, steps, box, \"test-assistant\", \"\", \"\")\n\t\t\tif err != nil {\n\t\t\t\tt.Fatalf(\"RunPrepareSteps: %v\", err)\n\t\t\t}\n\n\t\t\tresult, _ := box.Exec(ctx, []string{\"cat\", \"/tmp/after-bg\"})\n\t\t\tif strings.TrimSpace(result.Stdout) != \"after-bg\" {\n\t\t\t\tt.Errorf(\"background step blocked execution, got %q\", result.Stdout)\n\t\t\t}\n\t\t})\n\t}\n}\n\nfunc TestRunPrepareSteps_MixedActions(t *testing.T) {\n\tskipIfNoDocker(t)\n\n\tfor _, nc := range boxNodes() {\n\t\tnc := nc\n\t\tt.Run(nc.Name, func(t *testing.T) {\n\t\t\tm := setupManager(t, &nc)\n\t\t\tbox := createBox(t, m, nc)\n\t\t\twsID := fmt.Sprintf(\"test-mixed-%d\", time.Now().UnixNano())\n\t\t\tbox.BindWorkplace(wsID)\n\n\t\t\tctx, cancel := context.WithTimeout(context.Background(), 30*time.Second)\n\t\t\tdefer cancel()\n\n\t\t\tsteps := []types.PrepareStep{\n\t\t\t\t{Action: \"file\", Path: \"mixed.conf\", Content: []byte(\"key=value\")},\n\t\t\t\t{Action: \"exec\", Cmd: \"echo exec-ok > /tmp/mixed-exec\"},\n\t\t\t\t{Action: \"copy\", Src: \"mixed.conf\", Dst: \"mixed-copy.conf\"},\n\t\t\t}\n\n\t\t\terr := sandboxv2.RunPrepareSteps(ctx, steps, box, \"test-assistant\", \"\", \"\")\n\t\t\tif err != nil {\n\t\t\t\tt.Fatalf(\"RunPrepareSteps: %v\", err)\n\t\t\t}\n\n\t\t\tws := box.Workplace()\n\t\t\tdata, err := ws.ReadFile(\"mixed-copy.conf\")\n\t\t\tif err != nil {\n\t\t\t\tt.Fatalf(\"ReadFile mixed-copy.conf: %v\", err)\n\t\t\t}\n\t\t\tif string(data) != \"key=value\" {\n\t\t\t\tt.Errorf(\"copy result: got %q, want %q\", string(data), \"key=value\")\n\t\t\t}\n\n\t\t\tresult, _ := box.Exec(ctx, []string{\"cat\", \"/tmp/mixed-exec\"})\n\t\t\tif strings.TrimSpace(result.Stdout) != \"exec-ok\" {\n\t\t\t\tt.Errorf(\"exec result: got %q, want %q\", result.Stdout, \"exec-ok\")\n\t\t\t}\n\t\t})\n\t}\n}\n\n// ---------------------------------------------------------------------------\n// HostExec tests\n// ---------------------------------------------------------------------------\n\nfunc TestRunPrepareSteps_HostExec(t *testing.T) {\n\tskipIfNoHostExec(t)\n\n\tfor _, tgt := range hostTargets() {\n\t\ttgt := tgt\n\t\tt.Run(tgt.Name, func(t *testing.T) {\n\t\t\tm := setupHostManager(t, &tgt)\n\t\t\thost := createHost(t, m, tgt)\n\n\t\t\tctx, cancel := context.WithTimeout(context.Background(), 30*time.Second)\n\t\t\tdefer cancel()\n\n\t\t\tt.Logf(\"SystemInfo: OS=%q Shell=%q TempDir=%q\",\n\t\t\t\thost.ComputerInfo().System.OS,\n\t\t\t\thost.ComputerInfo().System.Shell,\n\t\t\t\thost.ComputerInfo().System.TempDir)\n\n\t\t\tisWin := tgt.Name == \"win-native\"\n\t\t\tvar cmd string\n\t\t\tif isWin {\n\t\t\t\tcmd = `Write-Output 'host-ok'`\n\t\t\t} else {\n\t\t\t\tcmd = \"echo host-ok\"\n\t\t\t}\n\t\t\tsteps := []types.PrepareStep{\n\t\t\t\t{Action: \"exec\", Cmd: cmd},\n\t\t\t}\n\n\t\t\terr := sandboxv2.RunPrepareSteps(ctx, steps, host, \"test-host\", \"\", \"\")\n\t\t\tif err != nil {\n\t\t\t\tt.Fatalf(\"RunPrepareSteps on host: %v\", err)\n\t\t\t}\n\t\t})\n\t}\n}\n\nfunc TestRunPrepareSteps_HostExecFile(t *testing.T) {\n\tskipIfNoHostExec(t)\n\n\tfor _, tgt := range hostTargets() {\n\t\ttgt := tgt\n\t\tt.Run(tgt.Name, func(t *testing.T) {\n\t\t\tm := setupHostManager(t, &tgt)\n\t\t\thost := createHost(t, m, tgt)\n\t\t\twsID := fmt.Sprintf(\"test-hostfile-%d\", time.Now().UnixNano())\n\t\t\thost.BindWorkplace(wsID)\n\n\t\t\tctx, cancel := context.WithTimeout(context.Background(), 30*time.Second)\n\t\t\tdefer cancel()\n\n\t\t\tsteps := []types.PrepareStep{\n\t\t\t\t{Action: \"file\", Path: \"host-test.txt\", Content: []byte(\"host-file-data\")},\n\t\t\t}\n\n\t\t\terr := sandboxv2.RunPrepareSteps(ctx, steps, host, \"test-host\", \"\", \"\")\n\t\t\tif err != nil {\n\t\t\t\tt.Fatalf(\"RunPrepareSteps file: %v\", err)\n\t\t\t}\n\n\t\t\tws := host.Workplace()\n\t\t\tdata, err := ws.ReadFile(\"host-test.txt\")\n\t\t\tif err != nil {\n\t\t\t\tt.Fatalf(\"ReadFile: %v\", err)\n\t\t\t}\n\t\t\tif string(data) != \"host-file-data\" {\n\t\t\t\tt.Errorf(\"content = %q, want %q\", string(data), \"host-file-data\")\n\t\t\t}\n\t\t})\n\t}\n}\n\nfunc TestRunPrepareSteps_HostExecCopy(t *testing.T) {\n\tskipIfNoHostExec(t)\n\n\tfor _, tgt := range hostTargets() {\n\t\ttgt := tgt\n\t\tt.Run(tgt.Name, func(t *testing.T) {\n\t\t\tm := setupHostManager(t, &tgt)\n\t\t\thost := createHost(t, m, tgt)\n\t\t\twsID := fmt.Sprintf(\"test-hostcopy-%d\", time.Now().UnixNano())\n\t\t\thost.BindWorkplace(wsID)\n\n\t\t\tws := host.Workplace()\n\t\t\tws.WriteFile(\"copy-src.txt\", []byte(\"copy-data\"), 0644)\n\n\t\t\tctx, cancel := context.WithTimeout(context.Background(), 30*time.Second)\n\t\t\tdefer cancel()\n\n\t\t\tsteps := []types.PrepareStep{\n\t\t\t\t{Action: \"copy\", Src: \"copy-src.txt\", Dst: \"copy-dst.txt\"},\n\t\t\t}\n\n\t\t\terr := sandboxv2.RunPrepareSteps(ctx, steps, host, \"test-host\", \"\", \"\")\n\t\t\tif err != nil {\n\t\t\t\tt.Fatalf(\"RunPrepareSteps copy: %v\", err)\n\t\t\t}\n\n\t\t\tdata, err := ws.ReadFile(\"copy-dst.txt\")\n\t\t\tif err != nil {\n\t\t\t\tt.Fatalf(\"ReadFile: %v\", err)\n\t\t\t}\n\t\t\tif string(data) != \"copy-data\" {\n\t\t\t\tt.Errorf(\"content = %q, want %q\", string(data), \"copy-data\")\n\t\t\t}\n\t\t})\n\t}\n}\n\nfunc TestRunPrepareSteps_HostExecOnce(t *testing.T) {\n\tskipIfNoHostExec(t)\n\n\tfor _, tgt := range hostTargets() {\n\t\ttgt := tgt\n\t\tt.Run(tgt.Name, func(t *testing.T) {\n\t\t\tm := setupHostManager(t, &tgt)\n\t\t\thost := createHost(t, m, tgt)\n\t\t\twsID := fmt.Sprintf(\"test-hostonce-%d\", time.Now().UnixNano())\n\t\t\thost.BindWorkplace(wsID)\n\n\t\t\tctx, cancel := context.WithTimeout(context.Background(), 30*time.Second)\n\t\t\tdefer cancel()\n\n\t\t\tisWin := tgt.Name == \"win-native\"\n\t\t\tvar cmd string\n\t\t\tif isWin {\n\t\t\t\tcmd = `Write-Output 'once-ok'`\n\t\t\t} else {\n\t\t\t\tcmd = \"echo once-ok\"\n\t\t\t}\n\t\t\tsteps := []types.PrepareStep{\n\t\t\t\t{Action: \"exec\", Cmd: cmd, Once: true},\n\t\t\t}\n\t\t\thash := \"host-once-hash\"\n\t\t\taid := \"host-once-aid\"\n\n\t\t\tif err := sandboxv2.RunPrepareSteps(ctx, steps, host, aid, hash, \"\"); err != nil {\n\t\t\t\tt.Fatalf(\"first run: %v\", err)\n\t\t\t}\n\n\t\t\tws := host.Workplace()\n\t\t\tmarkerData, err := ws.ReadFile(\".yao/prepare/\" + aid + \"/done\")\n\t\t\tif err != nil {\n\t\t\t\tt.Fatalf(\"marker not written: %v\", err)\n\t\t\t}\n\t\t\tif string(markerData) != hash {\n\t\t\t\tt.Errorf(\"marker = %q, want %q\", string(markerData), hash)\n\t\t\t}\n\t\t})\n\t}\n}\n" }, { "path": "agent/sandbox/v2/runner.go", "content": "package sandboxv2\n\nimport (\n\t\"fmt\"\n\t\"sync\"\n\n\t\"github.com/yaoapp/yao/agent/sandbox/v2/types\"\n)\n\nvar (\n\tmu sync.RWMutex\n\trunners = map[string]func() types.Runner{}\n)\n\n// Register adds a runner factory to the global registry.\n// Typically called from init() in the runner's package.\nfunc Register(name string, factory func() types.Runner) {\n\tmu.Lock()\n\tdefer mu.Unlock()\n\trunners[name] = factory\n}\n\n// Get creates a new Runner instance from the registry.\nfunc Get(name string) (types.Runner, error) {\n\tmu.RLock()\n\tdefer mu.RUnlock()\n\tfactory, ok := runners[name]\n\tif !ok {\n\t\treturn nil, fmt.Errorf(\"sandbox runner %q not registered\", name)\n\t}\n\treturn factory(), nil\n}\n" }, { "path": "agent/sandbox/v2/shell.go", "content": "package sandboxv2\n\nimport (\n\t\"strings\"\n\n\tinfra \"github.com/yaoapp/yao/sandbox/v2\"\n)\n\n// shellKind identifies which shell to use for command execution.\ntype shellKind int\n\nconst (\n\tshellSh shellKind = iota // Unix: sh -c\n\tshellPwsh // Windows: pwsh -NoProfile -Command\n\tshellPS // Windows: powershell -NoProfile -Command\n\tshellCmd // Windows: cmd.exe /C (last-resort fallback)\n)\n\n// shellWrap returns the Exec command slice to run a script string.\nfunc shellWrap(kind shellKind, script string) []string {\n\tswitch kind {\n\tcase shellPwsh:\n\t\treturn []string{\"pwsh\", \"-NoProfile\", \"-Command\", script}\n\tcase shellPS:\n\t\treturn []string{\"powershell\", \"-NoProfile\", \"-Command\", script}\n\tcase shellCmd:\n\t\treturn []string{\"cmd.exe\", \"/C\", script}\n\tdefault:\n\t\treturn []string{\"sh\", \"-c\", script}\n\t}\n}\n\n// shellFromSystem resolves shellKind from ComputerInfo().System.Shell\n// reported by the Tai node at registration time.\nfunc shellFromSystem(computer infra.Computer) shellKind {\n\tshell := strings.ToLower(computer.ComputerInfo().System.Shell)\n\tswitch shell {\n\tcase \"pwsh\":\n\t\treturn shellPwsh\n\tcase \"powershell\":\n\t\treturn shellPS\n\tcase \"cmd.exe\", \"cmd\":\n\t\treturn shellCmd\n\tdefault:\n\t\treturn shellSh\n\t}\n}\n" }, { "path": "agent/sandbox/v2/stream.go", "content": "package sandboxv2\n\nimport (\n\t\"context\"\n\t\"errors\"\n\t\"fmt\"\n\t\"log\"\n\t\"time\"\n\n\tagentContext \"github.com/yaoapp/yao/agent/context\"\n\t\"github.com/yaoapp/yao/agent/i18n\"\n\t\"github.com/yaoapp/yao/agent/output/message\"\n\t\"github.com/yaoapp/yao/agent/sandbox/v2/types\"\n\tinfra \"github.com/yaoapp/yao/sandbox/v2\"\n)\n\n// ExecuteRequest consolidates all parameters for ExecuteSandboxStream.\ntype ExecuteRequest struct {\n\tComputer infra.Computer\n\tRunner types.Runner\n\tConfig *types.SandboxConfig\n\tStreamReq *types.StreamRequest\n\tManager *infra.Manager\n\tLoadingMsgID string\n}\n\n// ExecuteSandboxStream is the V2 replacement for executeSandboxStream.\n// It calls runner.Stream, handles interrupts, and performs cleanup/lifecycle\n// in defer.\nfunc ExecuteSandboxStream(\n\tctx *agentContext.Context,\n\treq *ExecuteRequest,\n\thandler message.StreamFunc,\n) (*agentContext.CompletionResponse, error) {\n\n\tif req.Runner == nil || req.Computer == nil {\n\t\treturn nil, fmt.Errorf(\"runner and computer are required\")\n\t}\n\n\tstdCtx := ctx.Context\n\tpanicked := true // Assume panic; set false on normal exit.\n\n\t// Resolve stop timeout from config (default 2s).\n\tstopTimeout := 2 * time.Second\n\tif req.Config != nil && req.Config.StopTimeout != \"\" {\n\t\tif d, err := time.ParseDuration(req.Config.StopTimeout); err == nil {\n\t\t\tstopTimeout = d\n\t\t}\n\t}\n\n\t// Panic recovery (registered first, executes last in LIFO order).\n\tdefer func() {\n\t\tif r := recover(); r != nil {\n\t\t\tlog.Printf(\"[sandbox/v2] panic in stream: %v\", r)\n\t\t\tcleanCtx, cancel := context.WithTimeout(context.Background(), stopTimeout)\n\t\t\tdefer cancel()\n\t\t\treq.Runner.Cleanup(cleanCtx, req.Computer)\n\t\t\tLifecycleAction(cleanCtx, req.Config, req.Computer, req.Manager)\n\t\t}\n\t}()\n\n\t// Lifecycle action (registered second, executes second-to-last).\n\tdefer func() {\n\t\tif !panicked {\n\t\t\tLifecycleAction(stdCtx, req.Config, req.Computer, req.Manager)\n\t\t}\n\t}()\n\n\t// Runner cleanup (registered last, executes first).\n\tdefer func() {\n\t\tif !panicked {\n\t\t\tcleanCtx, cancel := context.WithTimeout(context.Background(), stopTimeout)\n\t\t\tdefer cancel()\n\t\t\treq.Runner.Cleanup(cleanCtx, req.Computer)\n\t\t}\n\t}()\n\n\t// Build a cancellable runnerCtx that bridges agentContext interrupts.\n\trunnerCtx, cancelRunner := context.WithCancel(stdCtx)\n\tdefer cancelRunner() // Prevent goroutine leak.\n\n\tdone := make(chan struct{})\n\tdefer close(done)\n\n\tgo func() {\n\t\tticker := time.NewTicker(500 * time.Millisecond)\n\t\tdefer ticker.Stop()\n\t\tfor {\n\t\t\tselect {\n\t\t\tcase <-done:\n\t\t\t\treturn\n\t\t\tcase <-ticker.C:\n\t\t\t\tif ctx.Interrupt != nil {\n\t\t\t\t\tif sig := ctx.Interrupt.Peek(); sig != nil {\n\t\t\t\t\t\tcancelRunner()\n\t\t\t\t\t\treturn\n\t\t\t\t\t}\n\t\t\t\t\tif ctx.Interrupt.IsInterrupted() {\n\t\t\t\t\t\tcancelRunner()\n\t\t\t\t\t\treturn\n\t\t\t\t\t}\n\t\t\t\t}\n\t\t\tcase <-stdCtx.Done():\n\t\t\t\tcancelRunner()\n\t\t\t\treturn\n\t\t\t}\n\t\t}\n\t}()\n\n\tif req.LoadingMsgID != \"\" {\n\t\twaitMsg := &message.Message{\n\t\t\tMessageID: req.LoadingMsgID,\n\t\t\tDelta: true,\n\t\t\tDeltaAction: message.DeltaReplace,\n\t\t\tType: message.TypeLoading,\n\t\t\tProps: map[string]any{\n\t\t\t\t\"message\": i18n.T(ctx.Locale, \"sandbox.waiting_response\"),\n\t\t\t},\n\t\t}\n\t\tctx.Send(waitMsg)\n\t}\n\n\tvar textContent []byte\n\tloadingClosed := false\n\twrappedHandler := func(chunkType message.StreamChunkType, data []byte) int {\n\t\tif !loadingClosed && req.LoadingMsgID != \"\" {\n\t\t\tif chunkType == message.ChunkText || chunkType == message.ChunkToolCall || chunkType == message.ChunkMessageStart {\n\t\t\t\tcloseLoading(ctx, req.LoadingMsgID)\n\t\t\t\tloadingClosed = true\n\t\t\t}\n\t\t}\n\t\tif chunkType == message.ChunkText {\n\t\t\ttextContent = append(textContent, data...)\n\t\t}\n\t\tif handler != nil {\n\t\t\treturn handler(chunkType, data)\n\t\t}\n\t\treturn 0\n\t}\n\n\terr := req.Runner.Stream(runnerCtx, req.StreamReq, wrappedHandler)\n\n\tif !loadingClosed && req.LoadingMsgID != \"\" {\n\t\tcloseLoading(ctx, req.LoadingMsgID)\n\t}\n\n\tpanicked = false // Normal exit reached.\n\n\tif err != nil {\n\t\tif errors.Is(err, context.Canceled) {\n\t\t\treturn nil, err\n\t\t}\n\t\treturn nil, fmt.Errorf(\"runner.Stream: %w\", err)\n\t}\n\n\tresp := &agentContext.CompletionResponse{\n\t\tRole: \"assistant\",\n\t\tFinishReason: agentContext.FinishReasonStop,\n\t}\n\tif len(textContent) > 0 {\n\t\tresp.Content = string(textContent)\n\t}\n\treturn resp, nil\n}\n\nfunc closeLoading(ctx *agentContext.Context, loadingMsgID string) {\n\tif loadingMsgID == \"\" || ctx == nil {\n\t\treturn\n\t}\n\tmsg := &message.Message{\n\t\tMessageID: loadingMsgID,\n\t\tDelta: true,\n\t\tDeltaAction: message.DeltaReplace,\n\t\tType: message.TypeLoading,\n\t\tProps: map[string]any{\n\t\t\t\"done\": true,\n\t\t\t\"message\": \"\",\n\t\t},\n\t}\n\tctx.Send(msg)\n}\n" }, { "path": "agent/sandbox/v2/testutils/testutils.go", "content": "package testutils\n\nimport (\n\t\"context\"\n\t\"os\"\n\t\"path/filepath\"\n\t\"testing\"\n\n\tagenttestutils \"github.com/yaoapp/yao/agent/testutils\"\n\t\"github.com/yaoapp/yao/config\"\n\tsandboxv2 \"github.com/yaoapp/yao/sandbox/v2\"\n\t\"github.com/yaoapp/yao/tai\"\n\t\"github.com/yaoapp/yao/tai/registry\"\n)\n\n// Prepare initializes the full environment required for sandbox V2 E2E tests:\n// - agent layer (assistants, LLM, caller)\n// - tai registry + local node\n// - sandbox V2 manager\nfunc Prepare(t *testing.T) {\n\tt.Helper()\n\n\tagenttestutils.Prepare(t)\n\n\tif registry.Global() == nil {\n\t\tregistry.Init(nil)\n\t}\n\n\tdataDir := filepath.Join(config.Conf.DataRoot, \"workspaces\")\n\tos.MkdirAll(dataDir, 0755)\n\ttai.RegisterLocal(tai.WithDataDir(dataDir))\n\n\tsandboxv2.Init()\n\tif err := sandboxv2.M().Start(context.Background()); err != nil {\n\t\tt.Fatalf(\"sandbox v2 manager start: %v\", err)\n\t}\n\n\tt.Cleanup(func() {\n\t\tsandboxv2.M().Close()\n\t})\n}\n\n// Clean tears down the test environment.\nfunc Clean(t *testing.T) {\n\tt.Helper()\n\tagenttestutils.Clean(t)\n}\n" }, { "path": "agent/sandbox/v2/testutils_remote_test.go", "content": "//go:build remote\n\npackage sandboxv2_test\n\nimport \"os\"\n\nfunc init() {\n\textraNodeProviders = append(extraNodeProviders, agentRemoteNodes)\n}\n\nfunc agentRemoteNodes() []nodeConfig {\n\taddr := os.Getenv(\"SANDBOX_TEST_REMOTE_ADDR\")\n\tif addr == \"\" {\n\t\treturn nil\n\t}\n\treturn []nodeConfig{{Name: \"remote\", Addr: addr}}\n}\n" }, { "path": "agent/sandbox/v2/testutils_test.go", "content": "package sandboxv2_test\n\nimport (\n\t\"context\"\n\t\"fmt\"\n\t\"log\"\n\t\"os\"\n\t\"strconv\"\n\t\"strings\"\n\t\"testing\"\n\t\"time\"\n\n\tsandbox \"github.com/yaoapp/yao/sandbox/v2\"\n\t\"github.com/yaoapp/yao/tai\"\n\t\"github.com/yaoapp/yao/tai/registry\"\n\ttairuntime \"github.com/yaoapp/yao/tai/runtime\"\n\t\"github.com/yaoapp/yao/workspace\"\n)\n\n// ---------------------------------------------------------------------------\n// Build-tag extension points (same pattern as sandbox/v2).\n// ---------------------------------------------------------------------------\nvar (\n\textraNodeProviders []func() []nodeConfig\n\textraHostExecProviders []func() []hostTarget\n)\n\n// ---------------------------------------------------------------------------\n// Node / host configuration\n// ---------------------------------------------------------------------------\n\ntype nodeConfig struct {\n\tName string\n\tAddr string\n\tTaiID string\n\tDialOps []tai.DialOption\n}\n\ntype hostTarget struct {\n\tName string\n\tAddr string\n\tTaiID string\n}\n\n// ---------------------------------------------------------------------------\n// Environment helpers\n// ---------------------------------------------------------------------------\n\nfunc testLocalAddr() string {\n\tif addr := os.Getenv(\"SANDBOX_TEST_LOCAL_ADDR\"); addr != \"\" {\n\t\treturn addr\n\t}\n\treturn \"local\"\n}\n\nfunc testImage() string {\n\tif img := os.Getenv(\"SANDBOX_TEST_IMAGE\"); img != \"\" {\n\t\treturn img\n\t}\n\treturn \"alpine:latest\"\n}\n\nfunc envPort(key string, fallback int) int {\n\tif v := os.Getenv(key); v != \"\" {\n\t\tif p, err := strconv.Atoi(v); err == nil {\n\t\t\treturn p\n\t\t}\n\t}\n\treturn fallback\n}\n\n// ---------------------------------------------------------------------------\n// Node / host discovery\n// ---------------------------------------------------------------------------\n\nfunc boxNodes() []nodeConfig {\n\tnodes := []nodeConfig{\n\t\t{Name: \"local\", Addr: testLocalAddr()},\n\t}\n\tfor _, fn := range extraNodeProviders {\n\t\tnodes = append(nodes, fn()...)\n\t}\n\treturn nodes\n}\n\nfunc hostTargets() []hostTarget {\n\tvar targets []hostTarget\n\tfor _, fn := range extraHostExecProviders {\n\t\ttargets = append(targets, fn()...)\n\t}\n\treturn targets\n}\n\n// ---------------------------------------------------------------------------\n// Dial + Register helper (replaces old tai.New)\n// ---------------------------------------------------------------------------\n\nfunc dialForTest(addr string, dialOps ...tai.DialOption) (*tai.ConnResources, error) {\n\tif addr == \"local\" || addr == \"\" {\n\t\treturn tai.DialLocal(\"\", \"\", nil)\n\t}\n\thost, grpcPort := parseHostPort(addr)\n\tports := tai.Ports{GRPC: grpcPort}\n\treturn tai.DialRemote(host, ports, dialOps...)\n}\n\nfunc registerForTest(t testing.TB, addr string, dialOps ...tai.DialOption) (string, *tai.ConnResources) {\n\tt.Helper()\n\tif registry.Global() == nil {\n\t\tregistry.Init(nil)\n\t}\n\tres, err := dialForTest(addr, dialOps...)\n\tif err != nil {\n\t\tt.Fatalf(\"dialForTest(%s): %v\", addr, err)\n\t}\n\ttaiID := taiIDFromAddr(addr)\n\treg := registry.Global()\n\treg.Register(®istry.TaiNode{TaiID: taiID, Mode: modeForAddr(addr)})\n\treg.SetResources(taiID, res)\n\treturn taiID, res\n}\n\nfunc taiIDFromAddr(addr string) string {\n\tif addr == \"local\" || addr == \"\" {\n\t\treturn \"local\"\n\t}\n\taddr = strings.TrimPrefix(addr, \"tai://\")\n\tparts := strings.SplitN(addr, \":\", 2)\n\treturn parts[0]\n}\n\nfunc modeForAddr(addr string) string {\n\tif addr == \"local\" || addr == \"\" {\n\t\treturn \"local\"\n\t}\n\treturn \"direct\"\n}\n\nfunc parseHostPort(addr string) (string, int) {\n\taddr = strings.TrimPrefix(addr, \"tai://\")\n\tparts := strings.SplitN(addr, \":\", 2)\n\th := parts[0]\n\tif len(parts) == 2 {\n\t\tif p, err := strconv.Atoi(parts[1]); err == nil {\n\t\t\treturn h, p\n\t\t}\n\t}\n\treturn h, 19100\n}\n\n// ---------------------------------------------------------------------------\n// TestMain — purge stale containers from previous runs\n// ---------------------------------------------------------------------------\n\nfunc TestMain(m *testing.M) {\n\tpurgeStale()\n\tos.Exit(m.Run())\n}\n\nfunc purgeStale() {\n\tctx, cancel := context.WithTimeout(context.Background(), 30*time.Second)\n\tdefer cancel()\n\n\tfor _, nc := range boxNodes() {\n\t\tres, err := dialForTest(nc.Addr, nc.DialOps...)\n\t\tif err != nil {\n\t\t\tcontinue\n\t\t}\n\t\tsb := res.Runtime\n\t\tif sb == nil {\n\t\t\tres.Close()\n\t\t\tcontinue\n\t\t}\n\t\tcontainers, _ := sb.List(ctx, tairuntime.ListOptions{All: true})\n\t\tfor _, c := range containers {\n\t\t\tid := c.Name\n\t\t\tif id == \"\" {\n\t\t\t\tid = c.ID\n\t\t\t}\n\t\t\tif strings.HasPrefix(id, \"sb-prep-\") || strings.HasPrefix(id, \"sb-lc-\") {\n\t\t\t\tsb.Remove(ctx, id, true)\n\t\t\t\tlog.Printf(\"[purge] %s: removed %s\", nc.Name, id)\n\t\t\t}\n\t\t}\n\t\tres.Close()\n\t}\n}\n\n// ---------------------------------------------------------------------------\n// Manager + Box helpers\n// ---------------------------------------------------------------------------\n\nfunc setupManager(t *testing.T, nc *nodeConfig) *sandbox.Manager {\n\tt.Helper()\n\tif registry.Global() == nil {\n\t\tregistry.Init(nil)\n\t}\n\ttaiID, res := registerForTest(t, nc.Addr, nc.DialOps...)\n\tnc.TaiID = taiID\n\tt.Cleanup(func() { res.Close() })\n\n\tsandbox.Init()\n\tm := sandbox.M()\n\tt.Cleanup(func() { m.Close() })\n\treturn m\n}\n\nfunc createBox(t *testing.T, m *sandbox.Manager, nc nodeConfig) *sandbox.Box {\n\tt.Helper()\n\tctx, cancel := context.WithTimeout(context.Background(), 60*time.Second)\n\tdefer cancel()\n\n\tif err := m.EnsureImage(ctx, nc.TaiID, testImage(), sandbox.ImagePullOptions{}); err != nil {\n\t\tt.Fatalf(\"EnsureImage: %v\", err)\n\t}\n\n\tbox, err := m.Create(ctx, sandbox.CreateOptions{\n\t\tID: fmt.Sprintf(\"sb-prep-%d\", time.Now().UnixNano()),\n\t\tImage: testImage(),\n\t\tOwner: \"test-prepare\",\n\t\tNodeID: nc.TaiID,\n\t})\n\tif err != nil {\n\t\tt.Fatalf(\"Create: %v\", err)\n\t}\n\tt.Cleanup(func() {\n\t\tcCtx, cCancel := context.WithTimeout(context.Background(), 15*time.Second)\n\t\tdefer cCancel()\n\t\tif err := m.Remove(cCtx, box.ID()); err != nil {\n\t\t\tt.Logf(\"cleanup Remove(%s): %v\", box.ID(), err)\n\t\t}\n\t})\n\treturn box\n}\n\nfunc createHost(t *testing.T, m *sandbox.Manager, tgt hostTarget) *sandbox.Host {\n\tt.Helper()\n\thost, err := m.Host(context.Background(), tgt.TaiID)\n\tif err != nil {\n\t\tt.Skipf(\"Host(%s): %v\", tgt.Name, err)\n\t}\n\treturn host\n}\n\nfunc setupHostManager(t *testing.T, tgt *hostTarget) *sandbox.Manager {\n\tt.Helper()\n\tnc := nodeConfig{Name: tgt.Name, Addr: fmt.Sprintf(\"tai://%s\", tgt.Addr)}\n\tm := setupManager(t, &nc)\n\ttgt.TaiID = nc.TaiID\n\treturn m\n}\n\n// ---------------------------------------------------------------------------\n// Skip helpers\n// ---------------------------------------------------------------------------\n\nfunc skipIfNoDocker(t *testing.T) {\n\tt.Helper()\n\tif testLocalAddr() == \"\" {\n\t\tt.Skip(\"SANDBOX_TEST_LOCAL_ADDR not set\")\n\t}\n}\n\nfunc skipIfNoHostExec(t *testing.T) {\n\tt.Helper()\n\tif len(hostTargets()) == 0 {\n\t\tt.Skip(\"no HostExec targets configured\")\n\t}\n}\n\nfunc createTestWorkspace(t *testing.T, taiID, wsID string) {\n\tt.Helper()\n\tctx, cancel := context.WithTimeout(context.Background(), 15*time.Second)\n\tdefer cancel()\n\t_, err := workspace.M().Create(ctx, workspace.CreateOptions{\n\t\tID: wsID,\n\t\tOwner: \"test\",\n\t\tNode: taiID,\n\t})\n\tif err != nil && !strings.Contains(err.Error(), \"exists\") {\n\t\tt.Fatalf(\"create workspace %q: %v\", wsID, err)\n\t}\n\tt.Cleanup(func() {\n\t\tcCtx, cCancel := context.WithTimeout(context.Background(), 10*time.Second)\n\t\tdefer cCancel()\n\t\tworkspace.M().Delete(cCtx, wsID, true)\n\t})\n}\n" }, { "path": "agent/sandbox/v2/testutils_wintest_test.go", "content": "//go:build wintest\n\npackage sandboxv2_test\n\nimport \"os\"\n\nfunc init() {\n\textraHostExecProviders = append(extraHostExecProviders, agentWinHostExec)\n}\n\nfunc agentWinHostExec() []hostTarget {\n\tvar targets []hostTarget\n\tif addr := os.Getenv(\"TAI_TEST_WIN_HOSTEXEC_LINUX\"); addr != \"\" {\n\t\ttargets = append(targets, hostTarget{Name: \"win-linux\", Addr: addr})\n\t}\n\tif addr := os.Getenv(\"TAI_TEST_WIN_HOSTEXEC_NATIVE\"); addr != \"\" {\n\t\ttargets = append(targets, hostTarget{Name: \"win-native\", Addr: addr})\n\t}\n\treturn targets\n}\n" }, { "path": "agent/sandbox/v2/token.go", "content": "package sandboxv2\n\nimport (\n\t\"fmt\"\n\t\"time\"\n\n\tlrustore \"github.com/yaoapp/gou/store/lru\"\n\t\"github.com/yaoapp/yao/agent/sandbox/v2/types\"\n\t\"github.com/yaoapp/yao/openapi/oauth\"\n)\n\nconst (\n\taccessTokenTTL = 2 * time.Hour\n\trefreshTokenTTL = 30 * 24 * time.Hour // 30 days\n\ttokenCacheSize = 1024\n)\n\nvar tokenCache *lrustore.Cache\n\nfunc init() {\n\tc, err := lrustore.New(tokenCacheSize)\n\tif err != nil {\n\t\tpanic(\"sandbox token cache init failed: \" + err.Error())\n\t}\n\ttokenCache = c\n}\n\nfunc cacheKey(teamID, userID string) string {\n\tif teamID == \"\" {\n\t\treturn userID\n\t}\n\treturn teamID + \"/\" + userID\n}\n\nfunc getToken(teamID, userID string) *types.SandboxToken {\n\tval, ok := tokenCache.Get(cacheKey(teamID, userID))\n\tif !ok {\n\t\treturn nil\n\t}\n\ttok, _ := val.(*types.SandboxToken)\n\treturn tok\n}\n\nfunc setToken(teamID, userID string, tok *types.SandboxToken, ttl time.Duration) {\n\ttokenCache.Set(cacheKey(teamID, userID), tok, ttl)\n}\n\n// IssueSandboxToken returns a valid identity token for the given user.\n// Tokens are cached by (teamID, userID); a new token is only issued on\n// cache miss or expiry. Returns nil without error when oauth.OAuth is nil.\nfunc IssueSandboxToken(teamID, userID string) (*types.SandboxToken, error) {\n\tif tok := getToken(teamID, userID); tok != nil {\n\t\treturn tok, nil\n\t}\n\n\tsvc := oauth.OAuth\n\tif svc == nil {\n\t\treturn nil, nil\n\t}\n\n\tsubject, err := svc.Subject(\"__yao.sandbox\", userID)\n\tif err != nil {\n\t\treturn nil, fmt.Errorf(\"sandbox token: derive subject: %w\", err)\n\t}\n\n\textraClaims := map[string]interface{}{\n\t\t\"user_id\": userID,\n\t}\n\tif teamID != \"\" {\n\t\textraClaims[\"team_id\"] = teamID\n\t}\n\n\ttokenStr, err := svc.MakeAccessToken(\"__yao.sandbox\", \"grpc:mcp\", subject,\n\t\tint(accessTokenTTL.Seconds()), extraClaims)\n\tif err != nil {\n\t\treturn nil, fmt.Errorf(\"sandbox token: issue access token: %w\", err)\n\t}\n\n\ttok := &types.SandboxToken{Token: tokenStr}\n\n\trefreshStr, err := svc.MakeRefreshToken(\"__yao.sandbox\", \"grpc:mcp\", subject,\n\t\tint(refreshTokenTTL.Seconds()), extraClaims)\n\tif err != nil {\n\t\treturn nil, fmt.Errorf(\"sandbox token: issue refresh token: %w\", err)\n\t}\n\ttok.RefreshToken = refreshStr\n\n\tsetToken(teamID, userID, tok, accessTokenTTL)\n\treturn tok, nil\n}\n" }, { "path": "agent/sandbox/v2/types/config.go", "content": "package types\n\nimport (\n\t\"encoding/json\"\n\t\"fmt\"\n)\n\nconst (\n\tSandboxVersionV1 = \"1.0\"\n\tSandboxVersionV2 = \"2.0\"\n)\n\n// SandboxConfig is the V2 sandbox configuration loaded from sandbox.yao or\n// the package.yao \"sandbox\" block when version == \"2.0\".\ntype SandboxConfig struct {\n\tVersion string `json:\"version\" yaml:\"version\"`\n\tComputer ComputerConfig `json:\"computer\" yaml:\"computer\"`\n\tRunner RunnerConfig `json:\"runner\" yaml:\"runner\"`\n\tLifecycle string `json:\"lifecycle,omitempty\" yaml:\"lifecycle,omitempty\"`\n\tIdleTimeout string `json:\"idle_timeout,omitempty\" yaml:\"idle_timeout,omitempty\"`\n\tMaxLifetime string `json:\"max_lifetime,omitempty\" yaml:\"max_lifetime,omitempty\"`\n\tStopTimeout string `json:\"stop_timeout,omitempty\" yaml:\"stop_timeout,omitempty\"`\n\tPrepare []PrepareStep `json:\"prepare,omitempty\" yaml:\"prepare,omitempty\"`\n\tEnvironment map[string]string `json:\"environment,omitempty\" yaml:\"environment,omitempty\"`\n\tSecrets map[string]string `json:\"secrets,omitempty\" yaml:\"secrets,omitempty\"`\n\tFilter *ComputerFilter `json:\"filter,omitempty\" yaml:\"filter,omitempty\"`\n\n\t// Populated by the framework at runtime (never serialized).\n\tOwner string `json:\"-\" yaml:\"-\"`\n\tID string `json:\"-\" yaml:\"-\"`\n\tLabels map[string]string `json:\"-\" yaml:\"-\"`\n\tNodeID string `json:\"-\" yaml:\"-\"`\n\tKind string `json:\"-\" yaml:\"-\"`\n\tWorkspaceID string `json:\"-\" yaml:\"-\"`\n\tDisplayName string `json:\"-\" yaml:\"-\"`\n}\n\n// ComputerFilter defines the query parameters for GET /computer/options.\n// Declared in DSL sandbox.filter; frontend passes it through to the API.\ntype ComputerFilter struct {\n\tKind string `json:\"kind,omitempty\" yaml:\"kind,omitempty\"`\n\tImage string `json:\"image,omitempty\" yaml:\"image,omitempty\"`\n\tVNC *bool `json:\"vnc,omitempty\" yaml:\"vnc,omitempty\"`\n\tOS string `json:\"os,omitempty\" yaml:\"os,omitempty\"`\n\tArch string `json:\"arch,omitempty\" yaml:\"arch,omitempty\"`\n\tMinCPUs float64 `json:\"min_cpus,omitempty\" yaml:\"min_cpus,omitempty\"`\n\tMinMem string `json:\"min_mem,omitempty\" yaml:\"min_mem,omitempty\"`\n\tLabels map[string]string `json:\"labels,omitempty\" yaml:\"labels,omitempty\"`\n}\n\n// ComputerConfig describes the execution environment (container or host).\ntype ComputerConfig struct {\n\tImage string `json:\"image,omitempty\" yaml:\"image,omitempty\"`\n\tVNC VNCConfig `json:\"vnc,omitempty\" yaml:\"vnc,omitempty\"`\n\tMemory string `json:\"memory,omitempty\" yaml:\"memory,omitempty\"`\n\tCPUs float64 `json:\"cpus,omitempty\" yaml:\"cpus,omitempty\"`\n\tPorts PortList `json:\"ports,omitempty\" yaml:\"ports,omitempty\"`\n\tUser string `json:\"user,omitempty\" yaml:\"user,omitempty\"`\n\tWorkDir string `json:\"work_dir,omitempty\" yaml:\"work_dir,omitempty\"`\n\tMountPath string `json:\"mount_path,omitempty\" yaml:\"mount_path,omitempty\"`\n\tMountMode string `json:\"mount_mode,omitempty\" yaml:\"mount_mode,omitempty\"`\n}\n\n// RunnerConfig identifies which Runner to use and how.\ntype RunnerConfig struct {\n\tName string `json:\"name\" yaml:\"name\"`\n\tMode string `json:\"mode,omitempty\" yaml:\"mode,omitempty\"`\n\tOptions map[string]any `json:\"options,omitempty\" yaml:\"options,omitempty\"`\n}\n\n// PrepareStep is a single action executed during Runner.Prepare.\ntype PrepareStep struct {\n\tAction string `json:\"action\" yaml:\"action\"`\n\tOnce bool `json:\"once,omitempty\" yaml:\"once,omitempty\"`\n\tIgnoreError bool `json:\"ignore_error,omitempty\" yaml:\"ignore_error,omitempty\"`\n\n\t// action=copy\n\tSrc string `json:\"src,omitempty\" yaml:\"src,omitempty\"`\n\tDst string `json:\"dst,omitempty\" yaml:\"dst,omitempty\"`\n\n\t// action=exec\n\tCmd string `json:\"cmd,omitempty\" yaml:\"cmd,omitempty\"`\n\tBackground bool `json:\"background,omitempty\" yaml:\"background,omitempty\"`\n\n\t// action=file (internal use by Runner.Prepare)\n\tPath string `json:\"path,omitempty\" yaml:\"path,omitempty\"`\n\tContent []byte `json:\"-\" yaml:\"-\"`\n\n\t// action=process (reserved)\n\tName string `json:\"name,omitempty\" yaml:\"name,omitempty\"`\n\tArgs []any `json:\"args,omitempty\" yaml:\"args,omitempty\"`\n}\n\n// ---------------------------------------------------------------------------\n// VNCConfig — supports both bool and object in JSON/YAML:\n// true → VNCConfig{Enabled: true}\n// {\"enabled\": true, \"password\": \"xxx\"} → full struct\n// ---------------------------------------------------------------------------\n\ntype VNCConfig struct {\n\tEnabled bool `json:\"enabled,omitempty\" yaml:\"enabled,omitempty\"`\n\tViewOnly bool `json:\"view_only,omitempty\" yaml:\"view_only,omitempty\"`\n\tPassword string `json:\"password,omitempty\" yaml:\"password,omitempty\"`\n\tResolution string `json:\"resolution,omitempty\" yaml:\"resolution,omitempty\"`\n}\n\nfunc (v *VNCConfig) UnmarshalJSON(data []byte) error {\n\tvar b bool\n\tif err := json.Unmarshal(data, &b); err == nil {\n\t\tv.Enabled = b\n\t\treturn nil\n\t}\n\ttype alias VNCConfig\n\tvar a alias\n\tif err := json.Unmarshal(data, &a); err != nil {\n\t\treturn err\n\t}\n\t*v = VNCConfig(a)\n\treturn nil\n}\n\n// ---------------------------------------------------------------------------\n// PortList — supports both int array and object array in JSON:\n// [3000, 8080] → []PortMapping{{Port: 3000}, {Port: 8080}}\n// [{\"port\": 3000, \"host_port\": 9000}] → full structs\n// ---------------------------------------------------------------------------\n\ntype PortList []PortMapping\n\ntype PortMapping struct {\n\tPort int `json:\"port\" yaml:\"port\"`\n\tHostPort int `json:\"host_port,omitempty\" yaml:\"host_port,omitempty\"`\n\tProtocol string `json:\"protocol,omitempty\" yaml:\"protocol,omitempty\"`\n}\n\nfunc (p *PortList) UnmarshalJSON(data []byte) error {\n\tvar ints []int\n\tif err := json.Unmarshal(data, &ints); err == nil {\n\t\tout := make(PortList, len(ints))\n\t\tfor i, port := range ints {\n\t\t\tout[i] = PortMapping{Port: port}\n\t\t}\n\t\t*p = out\n\t\treturn nil\n\t}\n\tvar objs []PortMapping\n\tif err := json.Unmarshal(data, &objs); err != nil {\n\t\treturn fmt.Errorf(\"ports: expected int array or object array: %w\", err)\n\t}\n\t*p = objs\n\treturn nil\n}\n" }, { "path": "agent/sandbox/v2/types/runner.go", "content": "package types\n\nimport (\n\t\"context\"\n\n\t\"github.com/yaoapp/gou/connector\"\n\tagentContext \"github.com/yaoapp/yao/agent/context\"\n\t\"github.com/yaoapp/yao/agent/output/message\"\n\tinfra \"github.com/yaoapp/yao/sandbox/v2\"\n)\n\n// Runner is the interface that all sandbox runners must implement.\n// A Runner replaces the LLM invocation layer (executeLLMStream) when a\n// sandbox is configured.\ntype Runner interface {\n\tName() string\n\tPrepare(ctx context.Context, req *PrepareRequest) error\n\tStream(ctx context.Context, req *StreamRequest, handler message.StreamFunc) error\n\tCleanup(ctx context.Context, computer infra.Computer) error\n}\n\n// MCPServer mirrors store/types.MCPServerConfig to avoid a cyclic import\n// between this leaf package and agent/store/types.\ntype MCPServer struct {\n\tServerID string `json:\"server_id,omitempty\"`\n\tResources []string `json:\"resources,omitempty\"`\n\tTools []string `json:\"tools,omitempty\"`\n}\n\n// RunStepsFunc is the signature of RunPrepareSteps. Workspace is obtained\n// internally via computer.Workplace(). assistantDir is the absolute path to\n// the assistant source directory on the host; copy steps resolve relative src\n// paths against it.\ntype RunStepsFunc func(ctx context.Context, steps []PrepareStep, computer infra.Computer, assistantID, configHash, assistantDir string) error\n\n// PrepareRequest carries everything needed by Runner.Prepare.\ntype PrepareRequest struct {\n\tComputer infra.Computer\n\tConfig *SandboxConfig\n\tConnector connector.Connector\n\tSkillsDir string\n\tAssistantDir string // absolute host path to the assistant source directory\n\tMCPServers []MCPServer\n\tConfigHash string\n\tRunSteps RunStepsFunc\n}\n\n// StreamRequest carries everything needed by Runner.Stream.\ntype StreamRequest struct {\n\tComputer infra.Computer\n\tConfig *SandboxConfig\n\tConnector connector.Connector\n\tMessages []agentContext.Message\n\tSystemPrompt string\n\tChatID string\n\tToken *SandboxToken // current user's sandbox token for MCP callbacks\n}\n" }, { "path": "agent/sandbox/v2/types/token.go", "content": "package types\n\n// SandboxToken holds credentials for a sandbox execution session.\n// Expiry is managed by the LRU store TTL, not stored here.\ntype SandboxToken struct {\n\tToken string // access token → YAO_TOKEN\n\tRefreshToken string // refresh token → YAO_REFRESH_TOKEN\n}\n" }, { "path": "agent/sandbox/v2/yao/runner.go", "content": "package yao\n\nimport (\n\t\"context\"\n\n\t\"github.com/yaoapp/yao/agent/output/message\"\n\t\"github.com/yaoapp/yao/agent/sandbox/v2/types\"\n\tinfra \"github.com/yaoapp/yao/sandbox/v2\"\n)\n\n// YaoRunner is a no-op Runner for pure Hook-driven sandbox interactions.\n// When runner.name == \"yao\", the assistant relies entirely on Create/Next\n// hooks for logic; no external CLI is invoked.\ntype YaoRunner struct{}\n\nfunc New() *YaoRunner { return &YaoRunner{} }\n\nfunc (r *YaoRunner) Name() string { return \"yao\" }\n\n// Prepare runs user-defined prepare steps (copy, exec, file) but adds\n// no runner-specific steps. Connector is not required.\nfunc (r *YaoRunner) Prepare(ctx context.Context, req *types.PrepareRequest) error {\n\tif req.RunSteps != nil && len(req.Config.Prepare) > 0 {\n\t\treturn req.RunSteps(ctx, req.Config.Prepare, req.Computer, req.Config.ID, req.ConfigHash, req.AssistantDir)\n\t}\n\treturn nil\n}\n\n// Stream is a no-op — hooks handle all interaction. Returns immediately\n// so the assistant framework proceeds to the Next hook.\nfunc (r *YaoRunner) Stream(_ context.Context, _ *types.StreamRequest, _ message.StreamFunc) error {\n\treturn nil\n}\n\n// Cleanup is a no-op for the yao runner.\nfunc (r *YaoRunner) Cleanup(_ context.Context, _ infra.Computer) error {\n\treturn nil\n}\n" }, { "path": "agent/sandbox/v2/yao/runner_test.go", "content": "package yao_test\n\nimport (\n\t\"context\"\n\t\"fmt\"\n\t\"testing\"\n\t\"time\"\n\n\t\"github.com/stretchr/testify/assert\"\n\t\"github.com/stretchr/testify/require\"\n\t\"github.com/yaoapp/yao/agent/caller\"\n\tagentcontext \"github.com/yaoapp/yao/agent/context\"\n\tsandboxtestutils \"github.com/yaoapp/yao/agent/sandbox/v2/testutils\"\n\toauthtypes \"github.com/yaoapp/yao/openapi/oauth/types\"\n)\n\nfunc TestSandboxV2_Yao_JSAPI(t *testing.T) {\n\tsandboxtestutils.Prepare(t)\n\tdefer sandboxtestutils.Clean(t)\n\n\trequire.NotNil(t, caller.AgentGetterFunc, \"AgentGetterFunc should be registered after Prepare\")\n\n\tagent, err := caller.AgentGetterFunc(\"tests.sandbox-v2.jsapi-v2\")\n\trequire.NoError(t, err, \"should load assistant tests.sandbox-v2.jsapi-v2\")\n\n\tchatID := fmt.Sprintf(\"e2e-jsapi-%d\", time.Now().UnixMilli())\n\tctx := agentcontext.New(\n\t\tcontext.Background(),\n\t\t&oauthtypes.AuthorizedInfo{\n\t\t\tTeamID: \"test-team-jsapi\",\n\t\t\tUserID: \"test-user-jsapi\",\n\t\t},\n\t\tchatID,\n\t)\n\n\tmessages := []agentcontext.Message{\n\t\t{Role: \"user\", Content: \"test jsapi\"},\n\t}\n\n\tdone := make(chan struct{})\n\tvar resp *agentcontext.Response\n\tvar streamErr error\n\n\tgo func() {\n\t\tdefer close(done)\n\t\tresp, streamErr = agent.Stream(ctx, messages)\n\t}()\n\n\tselect {\n\tcase <-done:\n\tcase <-time.After(3 * time.Minute):\n\t\tt.Fatalf(\"timeout after 3m\")\n\t}\n\n\trequire.NoError(t, streamErr, \"Stream should not return error\")\n\trequire.NotNil(t, resp, \"response should not be nil\")\n\n\t// runner=yao goes through executeLLMStream, then Next hook returns { data: results }\n\t// The Next hook result should appear in resp.Next\n\trequire.NotNil(t, resp.Next, \"resp.Next should not be nil (Next hook returned data)\")\n\tt.Logf(\"resp.Next: %+v\", resp.Next)\n\n\tnextData, ok := resp.Next.(map[string]interface{})\n\tif !ok {\n\t\tt.Fatalf(\"resp.Next should be a map, got %T: %+v\", resp.Next, resp.Next)\n\t}\n\n\t// The Next hook returns { data: results }, the framework unwraps .data\n\tdata, hasData := nextData[\"data\"]\n\tif hasData {\n\t\tnextData, ok = data.(map[string]interface{})\n\t\trequire.True(t, ok, \"data should be a map\")\n\t}\n\n\tt.Logf(\"JSAPI test results: %+v\", nextData)\n\n\t// ── Verify ctx.computer was available ──\n\tassert.Equal(t, true, nextData[\"has_computer\"], \"ctx.computer should be available\")\n\tassert.Equal(t, true, nextData[\"has_workspace\"], \"ctx.workspace should be available\")\n\n\t// ── Verify ctx.computer.Info() ──\n\tif infoRaw, ok := nextData[\"computer_info\"]; ok {\n\t\tinfo, ok := infoRaw.(map[string]interface{})\n\t\trequire.True(t, ok, \"computer_info should be a map\")\n\t\tassert.NotEmpty(t, info[\"kind\"], \"computer_info.kind should not be empty\")\n\t\tt.Logf(\"computer info: kind=%v os=%v\", info[\"kind\"], info[\"os\"])\n\t} else {\n\t\tassert.Nil(t, nextData[\"computer_info_error\"], \"computer.Info() should not error\")\n\t}\n\n\t// ── Verify ctx.computer.Exec() ──\n\tassert.Equal(t, \"jsapi-v2-test\", nextData[\"exec_stdout\"], \"Exec should return expected stdout\")\n\tassert.Nil(t, nextData[\"exec_error\"], \"Exec should not error\")\n\tif exitCode, ok := nextData[\"exec_exit_code\"]; ok {\n\t\t// JS numbers come back as float64 through JSON\n\t\tswitch v := exitCode.(type) {\n\t\tcase float64:\n\t\t\tassert.Equal(t, float64(0), v, \"exit_code should be 0\")\n\t\tcase int:\n\t\t\tassert.Equal(t, 0, v, \"exit_code should be 0\")\n\t\t}\n\t}\n\n\t// ── Verify ctx.workspace write/read ──\n\tassert.Equal(t, true, nextData[\"write_read_ok\"], \"workspace WriteFile+ReadFile round-trip should work\")\n\tassert.Equal(t, \"hello from jsapi v2\", nextData[\"read_content\"], \"read content should match\")\n\tassert.Nil(t, nextData[\"write_read_error\"], \"write/read should not error\")\n\n\t// ── Verify ctx.workspace MkdirAll + Exists ──\n\tassert.Equal(t, true, nextData[\"mkdir_exists_ok\"], \"MkdirAll + Exists should work\")\n\tassert.Nil(t, nextData[\"mkdir_exists_error\"], \"mkdir/exists should not error\")\n\n\t// ── Verify ctx.workspace ReadDir ──\n\tassert.Nil(t, nextData[\"readdir_error\"], \"ReadDir should not error\")\n\tif count, ok := nextData[\"readdir_count\"]; ok {\n\t\tswitch v := count.(type) {\n\t\tcase float64:\n\t\t\tassert.Greater(t, v, float64(0), \"ReadDir should return entries\")\n\t\t}\n\t}\n\n\t// ── Verify ctx.workspace Stat ──\n\tassert.Equal(t, true, nextData[\"stat_ok\"], \"Stat should return correct info\")\n\tassert.Nil(t, nextData[\"stat_error\"], \"Stat should not error\")\n\n\t// ── Verify ctx.workspace Copy ──\n\tassert.Equal(t, true, nextData[\"copy_ok\"], \"Copy should work\")\n\tassert.Nil(t, nextData[\"copy_error\"], \"Copy should not error\")\n\n\t// ── Verify ctx.workspace Rename ──\n\tassert.Equal(t, true, nextData[\"rename_ok\"], \"Rename should work\")\n\tassert.Nil(t, nextData[\"rename_error\"], \"Rename should not error\")\n\n\t// ── Verify ctx.workspace Remove ──\n\tassert.Equal(t, true, nextData[\"remove_ok\"], \"Remove should work\")\n\tassert.Nil(t, nextData[\"remove_error\"], \"Remove should not error\")\n}\n" }, { "path": "agent/search/DESIGN.md", "content": "# Search Module Design\n\n## Overview\n\nThe Search module provides a unified RAG (Retrieval-Augmented Generation) interface for Yao Agent, supporting three search types:\n\n| Type | Source | Use Case |\n| ----- | -------------- | ---------------------------------------------------- |\n| `web` | Internet | Real-time information, news, external knowledge |\n| `kb` | Knowledge Base | Documents, FAQs, internal knowledge (vector + graph) |\n| `db` | Database | Structured data from Yao Models (QueryDSL) |\n\nThe module follows the **Handler + Registry** pattern consistent with the `content` module, and exposes JSAPI for flexible usage in Create/Next hooks.\n\n## Key Features\n\n- **Unified JSAPI**: `ctx.search.Web()`, `ctx.search.KB()`, `ctx.search.DB()`, `ctx.search.Parallel()`\n- **Citation System**: Auto-generate citation IDs (`#ref:xxx`) for LLM reference\n- **Real-time Output**: Stream search progress to client\n- **Trace Integration**: Report search operations to user for transparency\n- **Reranking**: Builtin, Agent, or MCP-based result reranking\n- **Graceful Degradation**: Search errors don't block agent flow\n\n## Quick Start\n\n```typescript\n// In Create hook (assistants/my-assistant/index.ts)\nfunction Create(ctx, messages, options) {\n const query = messages[messages.length - 1].content;\n\n // Simple web search\n const result = ctx.search.Web(query, { limit: 5 });\n\n // Or parallel search across all sources\n const [web, kb, db] = ctx.search.Parallel([\n { type: \"web\", query, limit: 5 },\n { type: \"kb\", query, collections: [\"docs\"] },\n { type: \"db\", query, models: [\"product\"] },\n ]);\n\n return {\n messages: [{ role: \"system\", content: formatContext(web, kb, db) }],\n uses: { search: \"disabled\" }, // Disable auto search since hook handled it\n };\n}\n```\n\n## Goals\n\n1. **Unified Interface**: Single API for web, knowledge base, and database search\n2. **Flexibility**: Support built-in handlers and external tools (MCP/Agent delegation)\n3. **JSAPI Support**: Enable search calls from Create/Next hooks via JavaScript\n4. **Parallel Execution**: Support concurrent web + KB + DB searches\n5. **Graceful Degradation**: Search failures should not block the main agent flow\n6. **Real-time Feedback**: Stream search progress and results to users via output\n7. **Traceability**: Report search operations to users for transparency\n8. **Citation Support**: Enable LLM to reference search results with trackable citations\n\n## Architecture\n\n### Search Flow Diagram\n\n```mermaid\nflowchart TD\n A[Stream Start] --> B{Uses.Search?}\n B -->|disabled| C[Skip Search]\n B -->|builtin/agent/mcp| D{Hook Handled?}\n D -->|\"Yes (uses.search=disabled)\"| C\n D -->|No| E[Auto Search]\n\n E --> F{Check Assistant Config}\n F --> G[Web Search]\n F --> H[KB Search]\n F --> I[DB Search]\n\n G --> J[Parallel Execute]\n H --> J\n I --> J\n\n J --> K[Merge Results]\n K --> L[Rerank]\n L --> M[Generate Citations]\n M --> N[Inject to System Prompt]\n\n C --> O[LLM Call]\n N --> O\n O --> P[Output with Citations]\n```\n\n### Integration in Stream()\n\n```mermaid\nsequenceDiagram\n participant Client\n participant Stream\n participant CreateHook\n participant Search\n participant LLM\n participant Output\n\n Client->>Stream: Stream(ctx, messages, options)\n Stream->>Stream: Initialize\n\n alt Has Create Hook\n Stream->>CreateHook: Create(ctx, messages, options)\n CreateHook-->>Stream: response (may include search results)\n end\n\n alt Uses.Search != \"disabled\" AND not handled by Hook\n Stream->>Search: AutoSearch(ctx, messages)\n Search->>Search: Web/KB/DB in parallel\n Search->>Search: Rerank & Citations\n Search->>Output: search_start, search_result, search_complete\n Search-->>Stream: Inject search context to messages\n end\n\n Stream->>LLM: Execute with search context\n LLM->>Output: Stream response with #ref:xxx\n Stream-->>Client: Complete\n```\n\n### Directory Structure\n\n```\nagent/search/\n├── DESIGN.md # This document\n├── TODO.md # Implementation plan and progress\n├── search.go # Main Searcher implementation and public API\n├── registry.go # Handler registry (manages web/kb/db handlers)\n├── jsapi.go # JavaScript API bindings for hooks (skeleton)\n├── citation.go # Citation ID generation and tracking\n├── reference.go # Reference building and LLM context formatting\n│\n├── types/ # Type definitions (no dependencies on other search packages)\n│ ├── types.go # Core types (SearchType, Request, Result, ResultItem, etc.)\n│ ├── config.go # Configuration types (Config, CitationConfig, WeightsConfig, etc.)\n│ ├── reference.go # Reference type for unified context protocol\n│ └── graph.go # Graph-related types (GraphNode)\n│\n├── interfaces/ # Interface definitions (depends only on types/)\n│ ├── handler.go # Handler interface\n│ ├── searcher.go # Searcher interface (public API)\n│ ├── reranker.go # Reranker interface\n│ └── nlp.go # NLP interfaces (KeywordExtractor, QueryDSLGenerator)\n│\n├── rerank/ # Result reranking implementations (Handler + Registry pattern) ✅\n│ ├── reranker.go # Main entry point (mode dispatch)\n│ ├── builtin.go # Builtin: weighted score sorting\n│ ├── agent.go # Agent mode (delegate to LLM assistant)\n│ └── mcp.go # MCP mode (external service)\n│\n├── nlp/ # Natural language processing for search\n│ ├── keyword/ # Keyword extraction (Handler + Registry pattern) ✅\n│ │ ├── extractor.go # Main extractor (mode dispatch)\n│ │ ├── builtin.go # Builtin frequency-based extraction\n│ │ ├── agent.go # Agent mode (LLM-powered)\n│ │ └── mcp.go # MCP mode (external service)\n│ └── querydsl/ # QueryDSL generation for DB search (TODO)\n│ ├── generator.go # Main generator (mode dispatch)\n│ ├── builtin.go # Builtin template-based generation\n│ ├── agent.go # Agent mode (LLM-powered)\n│ └── mcp.go # MCP mode (external service)\n│ # Note: Embedding follows KB collection config, not in this package\n│\n├── handlers/ # Search handler implementations\n│ ├── web/ # Web search ✅\n│ │ ├── handler.go # Web search handler (mode dispatch)\n│ │ ├── tavily.go # Tavily provider (builtin)\n│ │ ├── serper.go # Serper provider (serper.dev, builtin)\n│ │ ├── serpapi.go # SerpAPI provider (serpapi.com, multi-engine, builtin)\n│ │ ├── agent.go # Agent mode (AI Search)\n│ │ └── mcp.go # MCP mode (external service)\n│ │\n│ ├── kb/ # Knowledge base search (skeleton)\n│ │ ├── handler.go # KB search handler\n│ │ ├── vector.go # Vector similarity search (TODO)\n│ │ └── graph.go # Graph-based association (TODO)\n│ │\n│ └── db/ # Database search (skeleton)\n│ ├── handler.go # DB search handler\n│ ├── query.go # QueryDSL builder (TODO)\n│ └── schema.go # Model schema introspection (TODO)\n│\n└── defaults/ # Default configuration values\n └── defaults.go # System built-in defaults (used by agent/load.go)\n\n# Note: Output and Trace are integrated into assistant/search.go\n# No separate trace.go or output.go files needed\n```\n\n### Dependency Graph\n\n```\n ┌─────────────┐\n │ types/ │ ← No internal dependencies\n └──────┬──────┘\n │\n ┌──────▼──────┐\n │ interfaces/ │ ← Depends only on types/\n └──────┬──────┘\n │\n ┌─────────────────┼─────────────────┐\n │ │ │\n ┌─────▼─────┐ ┌──────▼──────┐ ┌──────▼──────┐\n │ rerank/ │ │ nlp/ │ │ defaults/ │\n └─────┬─────┘ └──────┬──────┘ └──────┬──────┘\n │ │ │\n └────────┬────────┴────────┬────────┘\n │ │\n ┌──────▼──────┐ ┌──────▼──────┐\n │ handlers/ │ │ (root pkg) │\n │ web/kb/db │ │ search.go │\n └──────┬──────┘ │ registry │\n │ │ jsapi, etc │\n └────┬─────┴─────────────┘\n │\n ┌─────▼─────┐\n │ External │\n │ Packages │\n └───────────┘\n```\n\n### Package Import Rules\n\n1. **`types/`** - Zero internal dependencies, only stdlib and external packages\n2. **`interfaces/`** - Imports only `types/`\n3. **`rerank/`**, **`nlp/`**, **`defaults/`** - Import `types/` and `interfaces/`\n4. **`handlers/*`** - Import `types/`, `interfaces/`, and may use `nlp/` for NL processing\n5. **Root package** - Imports all sub-packages, provides public API\n\n### Main Searcher Implementation (`search.go`)\n\nConfiguration is loaded by `agent/load.go` (global) and `agent/assistant/load.go` (assistant-level), following the existing pattern. The Search package directly uses the loaded configuration.\n\n```go\npackage search\n\nimport (\n \"sync\"\n\n \"github.com/yaoapp/yao/agent/context\"\n \"github.com/yaoapp/yao/agent/search/handlers/db\"\n \"github.com/yaoapp/yao/agent/search/handlers/kb\"\n \"github.com/yaoapp/yao/agent/search/handlers/web\"\n \"github.com/yaoapp/yao/agent/search/interfaces\"\n \"github.com/yaoapp/yao/agent/search/rerank\"\n \"github.com/yaoapp/yao/agent/search/types\"\n)\n\n// Searcher is the main search implementation\ntype Searcher struct {\n config *types.Config // Merged config (global + assistant)\n handlers map[types.SearchType]interfaces.Handler\n reranker *rerank.Reranker // Uses rerank package directly\n citation *CitationGenerator\n}\n\n// Uses contains the search-specific uses configuration\n// These are extracted from context.Uses and search config\ntype Uses struct {\n Search string // \"builtin\", \"disabled\", \"\", \"mcp:.\"\n Web string // \"builtin\", \"\", \"mcp:.\"\n Keyword string // \"builtin\", \"\", \"mcp:.\"\n QueryDSL string // \"builtin\", \"\", \"mcp:.\"\n Rerank string // \"builtin\", \"\", \"mcp:.\"\n}\n\n// New creates a new Searcher instance\n// cfg: merged config from agent/load.go + assistant config\n// uses: merged uses configuration (global → assistant → hook)\nfunc New(cfg *types.Config, uses *Uses) *Searcher {\n return &Searcher{\n config: cfg,\n handlers: map[types.SearchType]interfaces.Handler{\n types.SearchTypeWeb: web.NewHandler(uses.Web, cfg.Web),\n types.SearchTypeKB: kb.NewHandler(cfg.KB), // KB always builtin\n types.SearchTypeDB: db.NewHandler(uses.QueryDSL, cfg.DB),\n },\n reranker: rerank.NewReranker(uses.Rerank, cfg.Rerank),\n citation: NewCitationGenerator(),\n }\n}\n\n// Search executes a single search request\nfunc (s *Searcher) Search(ctx *context.Context, req *types.Request) (*types.Result, error) {\n handler, ok := s.handlers[req.Type]\n if !ok {\n return &types.Result{Error: \"unsupported search type\"}, nil\n }\n\n // Execute search (handler doesn't need ctx)\n result, err := handler.Search(req)\n if err != nil {\n return &types.Result{Error: err.Error()}, nil\n }\n\n // Assign weights based on source\n for _, item := range result.Items {\n item.Weight = s.config.GetWeight(req.Source)\n }\n\n // Rerank if requested (reranker needs ctx for Agent/MCP modes)\n if req.Rerank != nil && s.reranker != nil {\n result.Items, _ = s.reranker.Rerank(ctx, req.Query, result.Items, req.Rerank)\n }\n\n // Generate citation IDs\n for _, item := range result.Items {\n item.CitationID = s.citation.Next()\n }\n\n return result, nil\n}\n\n// ParallelMode defines how parallel search should behave (inspired by JavaScript Promise)\ntype ParallelMode string\n\n// ParallelMode constants (similar to Promise.all, Promise.any, Promise.race)\nconst (\n // ModeAll waits for all searches to complete, returns all results (like Promise.all)\n ModeAll ParallelMode = \"all\"\n // ModeAny returns as soon as any search succeeds (has results), others continue but are discarded (like Promise.any)\n ModeAny ParallelMode = \"any\"\n // ModeRace returns as soon as any search completes (success or empty), others continue but are discarded (like Promise.race)\n ModeRace ParallelMode = \"race\"\n)\n\n// ParallelOptions configures parallel search behavior\n// All executes all searches and waits for all to complete (like Promise.all)\nfunc (s *Searcher) All(ctx *context.Context, reqs []*types.Request) ([]*types.Result, error) {\n return s.parallelAll(ctx, reqs)\n}\n\n// Any returns as soon as any search succeeds with results (like Promise.any)\nfunc (s *Searcher) Any(ctx *context.Context, reqs []*types.Request) ([]*types.Result, error) {\n return s.parallelAny(ctx, reqs)\n}\n\n// Race returns as soon as any search completes (like Promise.race)\nfunc (s *Searcher) Race(ctx *context.Context, reqs []*types.Request) ([]*types.Result, error) {\n return s.parallelRace(ctx, reqs)\n}\n\n// BuildReferences converts search results to unified Reference format\nfunc (s *Searcher) BuildReferences(results []*types.Result) []*types.Reference {\n var refs []*types.Reference\n for _, result := range results {\n for _, item := range result.Items {\n refs = append(refs, &types.Reference{\n ID: item.CitationID,\n Type: item.Type,\n Source: item.Source,\n Weight: item.Weight,\n Score: item.Score,\n Title: item.Title,\n Content: item.Content,\n URL: item.URL,\n })\n }\n }\n return refs\n}\n```\n\n### Registry (`registry.go`)\n\n```go\npackage search\n\nimport (\n \"github.com/yaoapp/yao/agent/search/interfaces\"\n \"github.com/yaoapp/yao/agent/search/types\"\n)\n\n// Registry manages search handlers\ntype Registry struct {\n handlers map[types.SearchType]interfaces.Handler\n}\n\n// NewRegistry creates a new handler registry\nfunc NewRegistry() *Registry {\n return &Registry{\n handlers: make(map[types.SearchType]interfaces.Handler),\n }\n}\n\n// Register registers a handler for a search type\nfunc (r *Registry) Register(handler interfaces.Handler) {\n r.handlers[handler.Type()] = handler\n}\n\n// Get returns the handler for a search type\nfunc (r *Registry) Get(t types.SearchType) (interfaces.Handler, bool) {\n h, ok := r.handlers[t]\n return h, ok\n}\n```\n\n## Core Interfaces\n\nAll interfaces are defined in `search/interfaces/` package to prevent circular dependencies.\n\n### Handler Interface (`interfaces/handler.go`)\n\n```go\npackage interfaces\n\nimport (\n \"github.com/yaoapp/yao/agent/search/types\"\n)\n\n// Handler defines the interface for search implementations\ntype Handler interface {\n // Type returns the search type this handler supports\n Type() types.SearchType\n\n // Search executes the search and returns results\n Search(req *types.Request) (*types.Result, error)\n}\n```\n\n### Searcher Interface (`interfaces/searcher.go`)\n\n```go\npackage interfaces\n\nimport (\n \"github.com/yaoapp/yao/agent/search/types\"\n)\n\n// Searcher is the main interface exposed to external callers\ntype Searcher interface {\n // Search executes a single search request\n Search(ctx *context.Context, req *types.Request) (*types.Result, error)\n\n // Parallel search methods - inspired by JavaScript Promise\n // All waits for all searches to complete (like Promise.all)\n All(ctx *context.Context, reqs []*types.Request) ([]*types.Result, error)\n // Any returns when any search succeeds with results (like Promise.any)\n Any(ctx *context.Context, reqs []*types.Request) ([]*types.Result, error)\n // Race returns when any search completes (like Promise.race)\n Race(ctx *context.Context, reqs []*types.Request) ([]*types.Result, error)\n\n // BuildReferences converts search results to unified Reference format for LLM\n BuildReferences(results []*types.Result) []*types.Reference\n}\n```\n\n> **Note**: Parallel search methods follow JavaScript Promise naming:\n>\n> - `All()`: Wait for all searches to complete (like `Promise.all`)\n> - `Any()`: Return when any search succeeds with results (like `Promise.any`)\n> - `Race()`: Return when any search completes (like `Promise.race`)\n\n### NLP Interfaces (`interfaces/nlp.go`)\n\n```go\npackage interfaces\n\nimport (\n \"github.com/yaoapp/gou/model\"\n \"github.com/yaoapp/gou/query/gou\"\n \"github.com/yaoapp/yao/agent/context\"\n \"github.com/yaoapp/yao/agent/search/types\"\n)\n\n// KeywordExtractor extracts keywords for web search\ntype KeywordExtractor interface {\n // Extract extracts search keywords from user message\n // ctx is required for Agent and MCP modes, can be nil for builtin mode\n Extract(ctx *context.Context, content string, opts *types.KeywordOptions) ([]string, error)\n}\n\n// QueryDSLGenerator generates QueryDSL for DB search\ntype QueryDSLGenerator interface {\n // Generate converts natural language to QueryDSL\n // Uses GOU types directly: model.Model and gou.QueryDSL\n Generate(query string, models []*model.Model) (*gou.QueryDSL, error)\n}\n\n// Note: Embedding is handled by KB collection's own config (embedding provider + model),\n// not defined here. See KB handler for details.\n```\n\n### Reranker Interface (`interfaces/reranker.go`)\n\n```go\npackage interfaces\n\nimport (\n \"github.com/yaoapp/yao/agent/context\"\n \"github.com/yaoapp/yao/agent/search/types\"\n)\n\n// Reranker reorders search results by relevance\ntype Reranker interface {\n // Rerank reorders results based on query relevance\n Rerank(ctx *context.Context, query string, items []*types.ResultItem, opts *types.RerankOptions) ([]*types.ResultItem, error)\n}\n```\n\n## Types\n\nAll types are defined in `search/types/` package to prevent circular dependencies.\n\n### Core Types (`types/types.go`)\n\n```go\npackage types\n\nimport (\n \"github.com/yaoapp/gou/query/gou\"\n)\n\n// SearchType represents the type of search\ntype SearchType string\n\nconst (\n SearchTypeWeb SearchType = \"web\" // Web/Internet search\n SearchTypeKB SearchType = \"kb\" // Knowledge base vector search\n SearchTypeDB SearchType = \"db\" // Database search (Yao Model/QueryDSL)\n)\n\n// SourceType represents where the search result came from\ntype SourceType string\n\nconst (\n SourceUser SourceType = \"user\" // User-provided DataContent (highest priority)\n SourceHook SourceType = \"hook\" // Hook ctx.search.*() results\n SourceAuto SourceType = \"auto\" // Auto search results (lowest priority)\n)\n\n// Request represents a search request\ntype Request struct {\n // Common fields\n Query string `json:\"query\"` // Search query (natural language)\n Type SearchType `json:\"type\"` // Search type: \"web\", \"kb\", or \"db\"\n Limit int `json:\"limit,omitempty\"` // Max results (default: 10)\n Source SourceType `json:\"source\"` // Source of this request (user/hook/auto)\n\n // Web search specific\n Sites []string `json:\"sites,omitempty\"` // Restrict to specific sites\n TimeRange string `json:\"time_range,omitempty\"` // \"day\", \"week\", \"month\", \"year\"\n\n // Knowledge base specific\n Collections []string `json:\"collections,omitempty\"` // KB collection IDs\n Threshold float64 `json:\"threshold,omitempty\"` // Similarity threshold (0-1)\n Graph bool `json:\"graph,omitempty\"` // Enable graph association\n\n // Database search specific\n // Uses GOU QueryDSL types directly for compatibility with Yao's query system\n // See: github.com/yaoapp/gou/query/gou/types.go\n Models []string `json:\"models,omitempty\"` // Model IDs (e.g., \"user\", \"agents.mybot.product\")\n Wheres []gou.Where `json:\"wheres,omitempty\"` // Pre-defined filters (optional), uses GOU QueryDSL Where\n Orders gou.Orders `json:\"orders,omitempty\"` // Sort orders (optional), uses GOU QueryDSL Orders\n Select []string `json:\"select,omitempty\"` // Fields to return (optional)\n\n // Reranking\n Rerank *RerankOptions `json:\"rerank,omitempty\"`\n}\n\n// RerankOptions controls result reranking\n// Reranker type is determined by uses.rerank in agent/agent.yml\ntype RerankOptions struct {\n TopN int `json:\"top_n,omitempty\"` // Return top N after reranking\n}\n\n// Result represents the search result with all intermediate processing data\ntype Result struct {\n Type SearchType `json:\"type\"` // Search type\n Query string `json:\"query\"` // Original query\n Source SourceType `json:\"source\"` // Source of this result\n Items []*ResultItem `json:\"items\"` // Result items\n Total int `json:\"total\"` // Total matches\n Duration int64 `json:\"duration_ms\"` // Search duration in ms\n Error string `json:\"error,omitempty\"` // Error message if failed\n\n // Intermediate processing results (for storage and debugging)\n Keywords []string `json:\"keywords,omitempty\"` // Extracted keywords (Web/NLP)\n DSL map[string]any `json:\"dsl,omitempty\"` // Generated QueryDSL (DB)\n Entities []Entity `json:\"entities,omitempty\"` // Extracted entities (Graph RAG)\n Relations []Relation `json:\"relations,omitempty\"` // Extracted relations (Graph RAG)\n\n // Graph associations (KB only, if enabled)\n GraphNodes []*GraphNode `json:\"graph_nodes,omitempty\"`\n}\n\n// Entity represents an extracted entity (for Graph RAG)\ntype Entity struct {\n Name string `json:\"name\"`\n Type string `json:\"type,omitempty\"`\n Source string `json:\"source,omitempty\"`\n}\n\n// Relation represents an extracted relation (for Graph RAG)\ntype Relation struct {\n Subject string `json:\"subject\"`\n Predicate string `json:\"predicate\"`\n Object string `json:\"object\"`\n Source string `json:\"source,omitempty\"`\n}\n\n// ResultItem represents a single search result item\ntype ResultItem struct {\n // Citation\n CitationID string `json:\"citation_id\"` // Unique ID for LLM reference: \"ref_001\"\n\n // Weighting\n Source SourceType `json:\"source\"` // Source type: \"user\", \"hook\", \"auto\"\n Weight float64 `json:\"weight\"` // Source weight (from config)\n Score float64 `json:\"score,omitempty\"` // Relevance score (0-1)\n\n // Common fields\n Type SearchType `json:\"type\"` // Search type for this item\n Title string `json:\"title,omitempty\"` // Title/headline\n Content string `json:\"content\"` // Main content/snippet\n URL string `json:\"url,omitempty\"` // Source URL\n\n // KB specific\n DocumentID string `json:\"document_id,omitempty\"` // Source document ID\n Collection string `json:\"collection,omitempty\"` // Collection name\n\n // DB specific\n Model string `json:\"model,omitempty\"` // Model ID\n RecordID interface{} `json:\"record_id,omitempty\"` // Record primary key\n Data map[string]interface{} `json:\"data,omitempty\"` // Full record data\n\n // Metadata\n Metadata map[string]interface{} `json:\"metadata,omitempty\"` // Additional metadata\n}\n\n// ProcessedQuery represents a processed query ready for execution\ntype ProcessedQuery struct {\n Type SearchType `json:\"type\"`\n Keywords []string `json:\"keywords,omitempty\"` // For web search\n Vector []float32 `json:\"vector,omitempty\"` // For KB search\n DSL *gou.QueryDSL `json:\"dsl,omitempty\"` // For DB search, uses GOU QueryDSL\n}\n```\n\n> **Design Note: Result with Intermediate Data**\n>\n> The `Result` type now includes intermediate processing results (`Keywords`, `DSL`, `Entities`, `Relations`)\n> that were previously only available during query processing. This design enables:\n>\n> 1. **Storage for Debugging**: All processing steps are captured for later analysis\n> 2. **System Tuning**: Analyze extracted keywords, generated DSL, and entity extraction quality\n> 3. **Unified Data Flow**: Handlers populate these fields during execution, eliminating the need\n> for separate data collection in `executeAutoSearch`\n>\n> **Handler Responsibilities**:\n>\n> - **Web Handler**: Populates `Keywords` from NLP extraction\n> - **DB Handler**: Populates `DSL` from QueryDSL generation\n> - **KB Handler**: Populates `Entities`, `Relations`, and `GraphNodes` from Graph RAG\n>\n> **Data Flow**:\n>\n> ```\n> Request → Handler → Result (with Keywords/DSL/Entities/Relations)\n> ↓\n> BuildReferenceContext\n> ↓\n> saveSearch (stores all intermediate data)\n> ```\n\n```go\n// ProcessedQuery is DEPRECATED for external use\n// Handlers should populate Result.Keywords/DSL/Entities/Relations directly\ntype ProcessedQuery struct {\n Type SearchType `json:\"type\"`\n Keywords []string `json:\"keywords,omitempty\"` // For web search\n Vector []float32 `json:\"vector,omitempty\"` // For KB search\n DSL *gou.QueryDSL `json:\"dsl,omitempty\"` // For DB search\n}\n\n// Note: For QueryDSL and Model types, use GOU types directly:\n// - github.com/yaoapp/gou/query/gou.QueryDSL\n// - github.com/yaoapp/gou/model.Model\n// - github.com/yaoapp/gou/model.Column\n```\n\n> **Note**: `Wheres` and `Orders` use GOU QueryDSL types directly (`gou.Where` and `gou.Orders`) for full compatibility with Yao's query system. See `github.com/yaoapp/gou/query/gou/types.go` for the complete type definitions.\n\n### Graph Types (`types/graph.go`)\n\n```go\npackage types\n\n// GraphNode represents a related entity from knowledge graph\ntype GraphNode struct {\n ID string `json:\"id\"`\n Type string `json:\"type\"` // Entity type\n Name string `json:\"name\"` // Entity name\n Description string `json:\"description,omitempty\"` // Entity description\n Relation string `json:\"relation,omitempty\"` // Relationship to query\n Score float64 `json:\"score,omitempty\"` // Relevance score\n Metadata map[string]interface{} `json:\"metadata,omitempty\"`\n}\n```\n\n### Reference Types (`types/reference.go`)\n\n```go\npackage types\n\n// Reference is the unified structure for all data sources\n// Used to build LLM context from search results\ntype Reference struct {\n ID string `json:\"id\"` // Unique citation ID: \"ref_001\", \"ref_002\"\n Type SearchType `json:\"type\"` // Data type: \"web\", \"kb\", \"db\"\n Source SourceType `json:\"source\"` // Origin: \"user\", \"hook\", \"auto\"\n Weight float64 `json:\"weight\"` // Relevance weight (1.0=highest, 0.6=lowest)\n Score float64 `json:\"score\"` // Relevance score (0-1)\n Title string `json:\"title\"` // Optional title\n Content string `json:\"content\"` // Main content\n URL string `json:\"url\"` // Optional URL\n Meta map[string]interface{} `json:\"meta\"` // Additional metadata\n}\n\n// ReferenceContext holds the formatted references for LLM input\ntype ReferenceContext struct {\n References []*Reference `json:\"references\"` // All references\n XML string `json:\"xml\"` // Formatted XML\n Prompt string `json:\"prompt\"` // Citation instruction prompt\n}\n```\n\n### Configuration Types (`types/config.go`)\n\n```go\npackage types\n\n// Config represents the complete search configuration\ntype Config struct {\n Web *WebConfig `json:\"web,omitempty\"`\n KB *KBConfig `json:\"kb,omitempty\"`\n DB *DBConfig `json:\"db,omitempty\"`\n Keyword *KeywordConfig `json:\"keyword,omitempty\"`\n QueryDSL *QueryDSLConfig `json:\"querydsl,omitempty\"`\n Rerank *RerankConfig `json:\"rerank,omitempty\"`\n Citation *CitationConfig `json:\"citation,omitempty\"`\n Weights *WeightsConfig `json:\"weights,omitempty\"`\n Options *OptionsConfig `json:\"options,omitempty\"`\n}\n\n// WebConfig for web search settings\n// Note: uses.web determines the mode (builtin/agent/mcp)\n// Provider is only used when uses.web = \"builtin\"\ntype WebConfig struct {\n Provider string `json:\"provider,omitempty\"` // \"tavily\", \"serper\", or \"serpapi\" (for builtin mode)\n APIKeyEnv string `json:\"api_key_env,omitempty\"` // Environment variable for API key\n MaxResults int `json:\"max_results,omitempty\"` // Max results (default: 10)\n Engine string `json:\"engine,omitempty\"` // Search engine for SerpAPI: \"google\", \"bing\", \"baidu\", etc. (default: \"google\")\n}\n\n// KBConfig for knowledge base search settings\ntype KBConfig struct {\n Collections []string `json:\"collections,omitempty\"` // Default collections\n Threshold float64 `json:\"threshold,omitempty\"` // Similarity threshold (default: 0.7)\n Graph bool `json:\"graph,omitempty\"` // Enable GraphRAG (default: false)\n}\n\n// DBConfig for database search settings\ntype DBConfig struct {\n Models []string `json:\"models,omitempty\"` // Default models\n MaxResults int `json:\"max_results,omitempty\"` // Max results (default: 20)\n}\n\n// KeywordConfig for keyword extraction\ntype KeywordConfig struct {\n MaxKeywords int `json:\"max_keywords,omitempty\"` // Max keywords (default: 10)\n Language string `json:\"language,omitempty\"` // \"auto\", \"en\", \"zh\", etc.\n}\n\n// KeywordOptions for keyword extraction (runtime options)\ntype KeywordOptions struct {\n MaxKeywords int `json:\"max_keywords,omitempty\"`\n Language string `json:\"language,omitempty\"`\n}\n\n// QueryDSLConfig for QueryDSL generation from natural language\ntype QueryDSLConfig struct {\n Strict bool `json:\"strict,omitempty\"` // Fail if generation fails (default: false)\n}\n\n// RerankConfig for reranking\ntype RerankConfig struct {\n TopN int `json:\"top_n,omitempty\"` // Return top N (default: 10)\n}\n\n// CitationConfig for citation format\ntype CitationConfig struct {\n Format string `json:\"format,omitempty\"` // Default: \"#ref:{id}\"\n AutoInjectPrompt bool `json:\"auto_inject_prompt,omitempty\"` // Auto-inject prompt (default: true)\n CustomPrompt string `json:\"custom_prompt,omitempty\"` // Custom prompt template\n}\n\n// WeightsConfig for source weighting\ntype WeightsConfig struct {\n User float64 `json:\"user,omitempty\"` // User-provided (default: 1.0)\n Hook float64 `json:\"hook,omitempty\"` // Hook results (default: 0.8)\n Auto float64 `json:\"auto,omitempty\"` // Auto search (default: 0.6)\n}\n\n// OptionsConfig for search behavior\ntype OptionsConfig struct {\n SkipThreshold int `json:\"skip_threshold,omitempty\"` // Skip auto search if user provides >= N results\n}\n```\n\n### Note on Reranker\n\nReranker type is determined by `uses.rerank` in `agent/agent.yml`:\n\n- `\"builtin\"` - Simple score-based sorting\n- `\"\"` - Delegate to an assistant (Agent)\n- `\"mcp:.\"` - Call MCP tool (e.g., `\"mcp:my-server.rerank\"`)\n\n## Citation System\n\nEach search result has a unique `CitationID` for LLM reference. Citation logic is implemented in `search/citation.go`.\n\n### Citation ID Generation\n\nCitation IDs are generated sequentially: `ref_001`, `ref_002`, etc.\n\n```go\n// citation.go\npackage search\n\nimport (\n \"fmt\"\n \"sync/atomic\"\n)\n\n// CitationGenerator generates unique citation IDs\ntype CitationGenerator struct {\n counter uint64\n}\n\n// NewCitationGenerator creates a new citation generator\nfunc NewCitationGenerator() *CitationGenerator {\n return &CitationGenerator{}\n}\n\n// Next generates the next citation ID\nfunc (g *CitationGenerator) Next() string {\n n := atomic.AddUint64(&g.counter, 1)\n return fmt.Sprintf(\"ref_%03d\", n)\n}\n```\n\n### Citation Config (in `types/config.go`)\n\n```go\ntype CitationConfig struct {\n Format string `json:\"format,omitempty\"` // Default: \"#ref:{id}\"\n AutoInjectPrompt bool `json:\"auto_inject_prompt,omitempty\"` // Auto-add instructions to system prompt\n CustomPrompt string `json:\"custom_prompt,omitempty\"` // Override default prompt template\n}\n```\n\n### Default Citation Prompt\n\nWhen `AutoInjectPrompt` is enabled (default), the system prompt includes:\n\n```\nYou have access to reference data in tags. Each has:\n- id: Citation identifier\n- type: Data type (web/kb/db)\n- weight: Relevance weight (1.0=highest priority, 0.6=lowest)\n- source: Origin (user=user-provided, hook=assistant-searched, auto=auto-searched)\n\nPrioritize higher-weight references when answering.\n\nWhen citing a reference, use this exact HTML format:\n[{id}]\n\nExample: According to the product data[ref_001], the price is $999.\n```\n\n### Custom Prompt in Config\n\n```yaml\n# assistants/my-assistant.yml\nsearch:\n citation:\n format: \"[{id}]\"\n auto_inject_prompt: true\n custom_prompt: \"Cite using [{id}]. Sources: ...\"\n```\n\n## Trace Integration\n\nSearch operations create minimal trace nodes to report execution status to users, providing transparency about what the agent is doing. Detailed information is recorded via LOG for debugging.\n\n### Trace Node Structure\n\nUses `trace/types.NodeStatus` constants:\n\n- `pending` - Node created but not started\n- `running` - Node is currently executing\n- `completed` - Node finished successfully\n- `failed` - Node failed with error\n\n**Single Search:**\n\n```\nsearch (type: \"search\")\n├── label // i18n: \"Search\" / \"搜索\"\n├── status // \"pending\" | \"running\" | \"completed\" | \"failed\"\n├── input\n│ ├── query // Original query\n│ └── types // [\"web\"], [\"kb\"], [\"web\", \"kb\", \"db\"]\n└── output // (set on complete)\n └── result_count // Total results found\n```\n\n**Parallel Search:**\n\n```\nsearch (type: \"search\")\n├── label // i18n: \"Search\" / \"搜索\"\n├── status // \"pending\" | \"running\" | \"completed\" | \"failed\"\n├── input\n│ ├── query // Original query\n│ └── types // [\"web\", \"kb\", \"db\"]\n└── children\n ├── web (type: \"search_item\")\n │ ├── label // i18n: \"Web Search\" / \"网页搜索\"\n │ ├── status // \"pending\" | \"running\" | \"completed\" | \"failed\"\n │ └── output\n │ └── result_count\n ├── kb (type: \"search_item\")\n │ └── ...\n └── db (type: \"search_item\")\n └── ...\n```\n\n### Trace Logging\n\nDetailed search information is recorded via Trace node logging methods (broadcasts to client):\n\n```go\n// Node logging methods (from trace/node.go):\n// - node.Info(message, args...) - Info level log\n// - node.Debug(message, args...) - Debug level log\n// - node.Warn(message, args...) - Warning level log\n// - node.Error(message, args...) - Error level log\n\n// Search start\nsearchNode.Info(\"Starting search\", map[string]any{\"query\": query, \"types\": types})\n\n// Per-type results (on parallel search children)\nwebNode.Debug(\"Web search completed\", map[string]any{\"count\": count, \"duration_ms\": duration})\nkbNode.Debug(\"KB search completed\", map[string]any{\"count\": count, \"duration_ms\": duration})\ndbNode.Debug(\"DB search completed\", map[string]any{\"count\": count, \"duration_ms\": duration})\n\n// Errors (non-blocking, search continues)\nwebNode.Warn(\"Web search failed\", map[string]any{\"error\": err.Error()})\n\n// Final summary (on parent node)\nsearchNode.Info(\"Search completed\", map[string]any{\"total\": total, \"duration_ms\": duration})\n```\n\n**Log Event Structure** (broadcasted via SSE):\n\n```go\n// types.TraceLog\ntype TraceLog struct {\n Timestamp int64 `json:\"timestamp\"` // milliseconds since epoch\n Level string `json:\"level\"` // \"info\", \"debug\", \"warn\", \"error\"\n Message string `json:\"message\"` // Log message\n Data any `json:\"data\"` // Additional data\n NodeID string `json:\"node_id\"` // Parent node ID\n}\n```\n\n## Real-time Output\n\nSearch progress is displayed to the client using **Loading component with Replace** pattern. Uses `ctx.Send()` and `ctx.Replace()` methods.\n\n### Output Flow\n\n```\n1. Send Loading Message\n loading_id = ctx.Send({ type: \"loading\", props: { message: \"Searching...\" } })\n → Client displays loading indicator\n\n2. Execute Search (parallel web/kb/db)\n\n3. Replace with Result Message (shows result to user)\n ctx.Replace(loading_id, { type: \"loading\", props: { message: \"Found 5 references\" } })\n → Client displays result message\n\n4. Mark as Done (removes the loading after brief display)\n ctx.Replace(loading_id, { type: \"loading\", props: { message: \"Found 5 references\", done: true } })\n → Client removes loading indicator\n```\n\n### Implementation\n\n```go\n// Send loading message\nloadingID := ctx.Send(map[string]any{\n \"type\": \"loading\",\n \"props\": map[string]any{\n \"message\": i18n.Tr(\"search.loading\", locale), // \"Searching...\" / \"正在搜索...\"\n },\n})\n\n// Execute search...\n\n// Replace with result message (displayed to user)\nresultMessage := i18n.Tr(\"search.success\", locale, count) // \"Found 5 references\"\nctx.Replace(loadingID, map[string]any{\n \"type\": \"loading\",\n \"props\": map[string]any{\n \"message\": resultMessage,\n },\n})\n\n// Mark as done (removes loading indicator after user sees the result)\nctx.Replace(loadingID, map[string]any{\n \"type\": \"loading\",\n \"props\": map[string]any{\n \"message\": resultMessage,\n \"done\": true, // Frontend will remove loading indicator\n },\n})\n```\n\n### Loading Props\n\n| Prop | Type | Description |\n| --------- | ------ | --------------------------------------------------- |\n| `message` | string | Localized message to display |\n| `done` | bool | When `true`, frontend removes the loading indicator |\n\n### Localized Messages\n\n| Scenario | English | Chinese |\n| ------------- | ---------------------------------------- | --------------------------------- |\n| Loading | Searching... | 正在搜索... |\n| Success (1) | Found 1 reference | 找到 1 条参考资料 |\n| Success (N) | Found N references | 找到 N 条参考资料 |\n| Partial Error | Found N references (some sources failed) | 找到 N 条参考资料（部分来源失败） |\n| All Failed | Search failed | 搜索失败 |\n| No Results | No references found | 未找到相关资料 |\n\n### Client Display Example\n\n```\nFrame 1 - During search:\n┌─────────────────────────────────┐\n│ Searching... │ ← Loading (done: false)\n└─────────────────────────────────┘\n\nFrame 2 - Result displayed:\n┌─────────────────────────────────┐\n│ Found 5 references │ ← Result (done: false)\n└─────────────────────────────────┘\n\nFrame 3 - Removed:\n(loading indicator removed when done: true)\n```\n\n## Search Result Storage\n\nSearch results are stored per request to support citation click-through and history replay.\n\n### Data Model\n\n```\nRelationships:\nChat\n └── Request (request_id)\n ├── Message[] (user, assistant, tool...)\n └── SearchResult[] (one request may have multiple searches)\n └── Reference[] (indexed references from each search)\n```\n\n### Citation Locating\n\nLLM output uses `` tags with index:\n\n```xml\nAI is artificial intelligence, it has developed rapidly...\n```\n\nLocation path: `request_id` + `index` → precisely locate reference\n\n### Database Schema\n\n**Table: `agent_search`**\n\n| Column | Type | Description |\n| ---------- | ----------- | -------------------------------------- |\n| id | BIGINT | Auto-increment primary key |\n| request_id | VARCHAR(64) | Associated request ID (indexed) |\n| chat_id | VARCHAR(64) | Associated chat ID (indexed) |\n| query | TEXT | Original search query |\n| config | JSON | Search config used (for tuning) |\n| keywords | JSON | Extracted keywords (from NLP) |\n| entities | JSON | Extracted entities (for Graph search) |\n| relations | JSON | Extracted relations (for Graph search) |\n| dsl | JSON | Generated QueryDSL (for DB search) |\n| source | VARCHAR(32) | Search source: web/kb/db/auto |\n| references | JSON | Reference[] with global index |\n| graph | JSON | GraphNode[] from knowledge graph |\n| xml | TEXT | Formatted XML for LLM context |\n| prompt | TEXT | Citation instruction prompt |\n| duration | INT | Search duration in milliseconds |\n| error | TEXT | Error message if failed (nullable) |\n| created_at | TIMESTAMP | Creation time |\n| deleted_at | TIMESTAMP | Soft delete time (nullable) |\n\n**Config Field Structure:**\n\n```json\n{\n \"uses\": {\n \"search\": \"builtin\",\n \"web\": \"builtin\",\n \"keyword\": \"builtin\",\n \"querydsl\": \"builtin\",\n \"rerank\": \"builtin\"\n },\n \"web\": {\n \"provider\": \"tavily\",\n \"max_results\": 5\n },\n \"kb\": {\n \"collections\": [\"docs\", \"faq\"],\n \"threshold\": 0.7,\n \"graph\": true\n },\n \"db\": {\n \"models\": [\"product\", \"order\"],\n \"max_results\": 20\n },\n \"rerank\": {\n \"provider\": \"builtin\",\n \"top_n\": 10\n }\n}\n```\n\n### Type Definitions\n\n```go\n// store/types/types.go\n\n// Search represents stored search results for a request\n// Stores all intermediate processing results for debugging and replay\ntype Search struct {\n ID int64 `json:\"id\"`\n RequestID string `json:\"request_id\"`\n ChatID string `json:\"chat_id\"`\n Query string `json:\"query\"` // Original query\n Config map[string]any `json:\"config,omitempty\"` // Search config used (for tuning)\n Keywords []string `json:\"keywords,omitempty\"` // Extracted keywords (Web/NLP)\n Entities []Entity `json:\"entities,omitempty\"` // Extracted entities (Graph)\n Relations []Relation `json:\"relations,omitempty\"` // Extracted relations (Graph)\n DSL map[string]any `json:\"dsl,omitempty\"` // Generated QueryDSL (DB)\n Source string `json:\"source\"` // web/kb/db/auto\n References []Reference `json:\"references\"`\n Graph []GraphNode `json:\"graph,omitempty\"` // Graph nodes from KB\n XML string `json:\"xml,omitempty\"` // Formatted XML for LLM\n Prompt string `json:\"prompt,omitempty\"` // Citation prompt\n Duration int64 `json:\"duration_ms\"` // Search duration\n Error string `json:\"error,omitempty\"` // Error if failed\n CreatedAt time.Time `json:\"created_at\"`\n}\n\n// Reference represents a single reference with global index (for storage)\ntype Reference struct {\n Index int `json:\"index\"` // Global index: 1, 2, 3...\n Type string `json:\"type\"` // web/kb/db\n Title string `json:\"title\"`\n URL string `json:\"url,omitempty\"`\n Snippet string `json:\"snippet\"`\n Content string `json:\"content,omitempty\"` // Full content (optional)\n Metadata map[string]any `json:\"metadata,omitempty\"`\n}\n\n// Entity represents an extracted entity from query (for Graph search)\ntype Entity struct {\n Name string `json:\"name\"` // Entity name\n Type string `json:\"type\"` // Entity type: person, org, location, etc.\n Metadata map[string]any `json:\"metadata,omitempty\"`\n}\n\n// Relation represents an extracted relation from query (for Graph search)\ntype Relation struct {\n Subject string `json:\"subject\"` // Source entity\n Predicate string `json:\"predicate\"` // Relation type\n Object string `json:\"object\"` // Target entity\n Metadata map[string]any `json:\"metadata,omitempty\"`\n}\n\n// GraphNode represents a node from knowledge graph (search result)\ntype GraphNode struct {\n ID string `json:\"id\"`\n Type string `json:\"type\"` // Entity type\n Name string `json:\"name\"` // Entity name\n Description string `json:\"description,omitempty\"`\n Relation string `json:\"relation,omitempty\"` // Relationship to query\n Score float64 `json:\"score,omitempty\"`\n Metadata map[string]any `json:\"metadata,omitempty\"`\n}\n\n// SearchFilter for querying search records\ntype SearchFilter struct {\n RequestID string `json:\"request_id,omitempty\"`\n ChatID string `json:\"chat_id,omitempty\"`\n Source string `json:\"source,omitempty\"`\n}\n```\n\n### Store Interface Extension\n\n```go\n// store/types/store.go\n\n// ChatStore interface extension\ntype ChatStore interface {\n // ... existing methods ...\n\n // ==========================================================================\n // Search Management\n // ==========================================================================\n\n // SaveSearch saves search record for a request\n // search: Search record to save\n // Returns: Potential error\n SaveSearch(search *Search) error\n\n // GetSearches retrieves search records for a request\n // requestID: Request ID\n // Returns: Search records and potential error\n GetSearches(requestID string) ([]*Search, error)\n\n // GetReference retrieves a single reference by request ID and index\n // requestID: Request ID\n // index: Reference index (1-based)\n // Returns: Reference and potential error\n GetReference(requestID string, index int) (*Reference, error)\n\n // DeleteSearches deletes all search records for a chat\n // chatID: Chat ID\n // Returns: Potential error\n DeleteSearches(chatID string) error\n}\n```\n\n### Xun Implementation\n\n```go\n// store/xun/search.go\n\n// SaveSearch saves a search record\nfunc (store *Xun) SaveSearch(search *Search) error {\n if search.RequestID == \"\" {\n return fmt.Errorf(\"request_id is required\")\n }\n\n refsJSON, err := jsoniter.MarshalToString(search.References)\n if err != nil {\n return fmt.Errorf(\"failed to marshal references: %w\", err)\n }\n\n row := map[string]interface{}{\n \"request_id\": search.RequestID,\n \"chat_id\": search.ChatID,\n \"query\": search.Query,\n \"config\": search.Config, // Search config for tuning\n \"keywords\": search.Keywords,\n \"entities\": search.Entities, // Graph entities\n \"relations\": search.Relations, // Graph relations\n \"dsl\": search.DSL,\n \"source\": search.Source,\n \"references\": refsJSON,\n \"graph\": search.Graph, // Graph nodes\n \"xml\": search.XML,\n \"prompt\": search.Prompt,\n \"duration\": search.Duration,\n \"error\": search.Error,\n \"created_at\": time.Now(),\n }\n\n return store.newQuerySearch().Insert(row)\n}\n\n// GetSearches retrieves all search records for a request\nfunc (store *Xun) GetSearches(requestID string) ([]*Search, error) {\n rows, err := store.newQuerySearch().\n Where(\"request_id\", requestID).\n WhereNull(\"deleted_at\").\n OrderBy(\"created_at\", \"asc\").\n Get()\n // ... convert rows to Search\n}\n\n// GetReference retrieves a single reference\nfunc (store *Xun) GetReference(requestID string, index int) (*Reference, error) {\n searches, err := store.GetSearches(requestID)\n if err != nil {\n return nil, err\n }\n\n // Find reference by index across all search records\n for _, search := range searches {\n for _, ref := range search.References {\n if ref.Index == index {\n return &ref, nil\n }\n }\n }\n\n return nil, fmt.Errorf(\"reference %d not found in request %s\", index, requestID)\n}\n```\n\n### Model Definition\n\n```json\n// yao/models/agent/search.mod.yao\n{\n \"name\": \"Search\",\n \"label\": \"Search\",\n \"description\": \"Search records for citation support and debugging\",\n \"tags\": [\"agent\", \"system\"],\n \"builtin\": true,\n \"readonly\": true,\n \"table\": {\n \"name\": \"agent_search\",\n \"comment\": \"Agent search table\"\n },\n \"columns\": [\n { \"name\": \"id\", \"type\": \"ID\", \"label\": \"ID\" },\n {\n \"name\": \"request_id\",\n \"type\": \"string\",\n \"length\": 64,\n \"nullable\": false,\n \"index\": true\n },\n {\n \"name\": \"chat_id\",\n \"type\": \"string\",\n \"length\": 64,\n \"nullable\": false,\n \"index\": true\n },\n { \"name\": \"query\", \"type\": \"text\", \"nullable\": true },\n {\n \"name\": \"config\",\n \"type\": \"json\",\n \"nullable\": true,\n \"comment\": \"Search config used (for tuning)\"\n },\n { \"name\": \"keywords\", \"type\": \"json\", \"nullable\": true },\n { \"name\": \"entities\", \"type\": \"json\", \"nullable\": true },\n { \"name\": \"relations\", \"type\": \"json\", \"nullable\": true },\n { \"name\": \"dsl\", \"type\": \"json\", \"nullable\": true },\n { \"name\": \"source\", \"type\": \"string\", \"length\": 32, \"nullable\": false },\n { \"name\": \"references\", \"type\": \"json\", \"nullable\": true },\n { \"name\": \"graph\", \"type\": \"json\", \"nullable\": true },\n { \"name\": \"xml\", \"type\": \"text\", \"nullable\": true },\n { \"name\": \"prompt\", \"type\": \"text\", \"nullable\": true },\n { \"name\": \"duration\", \"type\": \"integer\", \"nullable\": true },\n { \"name\": \"error\", \"type\": \"text\", \"nullable\": true }\n ],\n \"option\": { \"timestamps\": true, \"soft_deletes\": true }\n}\n```\n\n### Stream Integration\n\nStorage logic is encapsulated in `assistant/search.go` with a dedicated method:\n\n```go\n// assistant/search.go\n\n// SearchExecutionResult contains all intermediate results from search execution\ntype SearchExecutionResult struct {\n Query string // Original query\n Config map[string]any // Search config used\n Keywords []string // Extracted keywords (Web/NLP)\n Entities []storeTypes.Entity // Extracted entities (Graph)\n Relations []storeTypes.Relation // Extracted relations (Graph)\n DSL map[string]any // Generated QueryDSL (DB)\n Source string // web/kb/db/auto\n RefCtx *searchTypes.ReferenceContext // Reference context for LLM\n Graph []storeTypes.GraphNode // Graph nodes from KB\n Duration int64 // Duration in ms\n Error string // Error message if failed\n}\n\n// saveSearch saves search record to store for citation support and debugging\nfunc (ast *Assistant) saveSearch(ctx *context.Context, result *SearchExecutionResult) {\n if ctx.Store == nil || result == nil {\n return\n }\n\n // Skip if no references and no error\n if result.RefCtx == nil && result.Error == \"\" {\n return\n }\n\n var refs []storeTypes.Reference\n var xml, prompt string\n\n if result.RefCtx != nil {\n refs = convertReferences(result.RefCtx.References)\n xml = result.RefCtx.XML\n prompt = result.RefCtx.Prompt\n }\n\n search := &storeTypes.Search{\n RequestID: ctx.RequestID,\n ChatID: ctx.ID,\n Query: result.Query,\n Config: result.Config, // Search config for tuning analysis\n Keywords: result.Keywords,\n Entities: result.Entities, // Graph entities\n Relations: result.Relations, // Graph relations\n DSL: result.DSL,\n Source: result.Source,\n References: refs,\n Graph: result.Graph, // Graph nodes\n XML: xml,\n Prompt: prompt,\n Duration: result.Duration,\n Error: result.Error,\n }\n\n if err := ctx.Store.SaveSearch(search); err != nil {\n ctx.Logger.Warn(\"Failed to save search: %v\", err)\n }\n}\n\n// convertReferences converts search references to store format\nfunc convertReferences(refs []*searchTypes.Reference) []storeTypes.Reference {\n result := make([]storeTypes.Reference, len(refs))\n for i, ref := range refs {\n result[i] = storeTypes.Reference{\n Index: i + 1, // 1-based index\n Type: string(ref.Type),\n Title: ref.Title,\n URL: ref.URL,\n Snippet: ref.Content,\n Content: ref.Content,\n Metadata: ref.Meta,\n }\n }\n return result\n}\n\n// In executeAutoSearch:\nfunc (ast *Assistant) executeAutoSearch(ctx *context.Context, ...) *searchTypes.ReferenceContext {\n start := time.Now()\n\n // 1. Execute search (Result now contains all intermediate data)\n results, err := searcher.All(ctx, requests)\n duration := time.Since(start).Milliseconds()\n\n // 2. Prepare execution result for storage\n execResult := &SearchExecutionResult{\n Query: query,\n Config: buildSearchConfig(searchConfig, searchUses),\n Source: \"auto\",\n Duration: duration,\n }\n\n if err != nil {\n execResult.Error = err.Error()\n ast.saveSearch(ctx, execResult)\n return nil\n }\n\n // 3. Extract intermediate data from results\n // Result.Keywords, Result.DSL, Result.Entities, Result.Relations are populated by handlers\n for _, result := range results {\n if len(result.Keywords) > 0 {\n execResult.Keywords = result.Keywords\n }\n if result.DSL != nil {\n execResult.DSL = result.DSL\n }\n if len(result.Entities) > 0 {\n execResult.Entities = convertEntities(result.Entities)\n }\n if len(result.Relations) > 0 {\n execResult.Relations = convertRelations(result.Relations)\n }\n if len(result.GraphNodes) > 0 {\n execResult.Graph = convertGraphNodes(result.GraphNodes)\n }\n }\n\n // 4. Build reference context\n refCtx := search.BuildReferenceContext(results, citationConfig)\n execResult.RefCtx = refCtx\n\n // 5. Save search record\n ast.saveSearch(ctx, execResult)\n\n return refCtx\n}\n```\n\n### Usage Scenarios\n\n**Scenario 1: Single Search**\n\n```\nRequest: req_001\n └── Search: { source: \"auto\", references: [{index:1,...}, {index:2,...}, {index:3,...}] }\n```\n\n**Scenario 2: Multiple Searches (e.g., Tool Call triggers another search)**\n\n```\nRequest: req_001\n ├── Search[0]: { source: \"web\", references: [{index:1,...}, {index:2,...}] }\n └── Search[1]: { source: \"kb\", references: [{index:3,...}, {index:4,...}] }\n```\n\nIndex is globally incremented, so `request_id + index` is always unique.\n\n### API Endpoints\n\n```\nGET /api/chat/{chat_id}/request/{request_id}/references # Get all references for request\nGET /api/chat/{chat_id}/request/{request_id}/reference/{index} # Get single reference by index\n```\n\n### Frontend Integration\n\n```typescript\n// When user clicks citation [1]\nasync function onCitationClick(requestId: string, index: number) {\n const ref = await api.get(\n `/chat/${chatId}/request/${requestId}/reference/${index}`\n );\n showReferenceCard({\n title: ref.title,\n url: ref.url,\n snippet: ref.snippet,\n content: ref.content,\n });\n}\n```\n\n## JSAPI Integration\n\nThe Search module is exposed via `ctx.search` object in hook scripts.\n\n### Architecture\n\nTo avoid circular dependency between `context` and `search` packages:\n\n```\nagent/context/jsapi_search.go agent/search/jsapi.go\n┌─────────────────────────────┐ ┌─────────────────────────┐\n│ SearchAPI interface │◄───────│ JSAPI struct │\n│ SearchAPIFactory var │ │ (implements SearchAPI) │\n│ V8 binding methods: │ │ NewJSAPI() │\n│ newSearchObject() │ │ Web/KB/DB() │\n│ searchWebMethod() │ │ All/Any/Race() │\n│ searchKBMethod() │ │ buildRequest() │\n│ searchDBMethod() │ │ parseRequests() │\n│ searchAllMethod() │ │ ConfigGetter type │\n│ searchAnyMethod() │ │ SetJSAPIFactory() │\n│ searchRaceMethod() │ └─────────────────────────┘\n└─────────────────────────────┘ │\n ▲ │\n │ │\n └───────────────────────────────────────┘\n Factory registration\n (with ConfigGetter in assistant/init)\n\nagent/context/jsapi.go\n┌─────────────────────────────┐\n│ NewObject() │\n│ jsObject.Set(\"search\", │\n│ ctx.newSearchObject()) │\n└─────────────────────────────┘\n```\n\n**Key Files:**\n\n| File | Description |\n| -------------------------------- | ---------------------------------------------------------------- |\n| `context/jsapi_search.go` | SearchAPI interface + V8 binding methods |\n| `context/jsapi_search_test.go` | Integration tests (real V8 calls via test assistant) |\n| `context/jsapi.go` | Mount search object to ctx |\n| `search/jsapi.go` | JSAPI implementation (calls Searcher) + ConfigGetter |\n| `search/jsapi_test.go` | Black-box unit tests |\n| `assistant/assistant.go:init` | Factory registration via SetJSAPIFactory(ConfigGetter) |\n| `assistants/tests/search-jsapi/` | Test assistant for JSAPI integration tests (Create hook, no LLM) |\n\n### API Methods\n\n```typescript\n// In hook scripts (index.ts)\n\n// Single search methods\nctx.search.Web(query: string, options?: WebOptions): Result\nctx.search.KB(query: string, options?: KBOptions): Result\nctx.search.DB(query: string, options?: DBOptions): Result\n\n// Parallel search methods - inspired by JavaScript Promise\nctx.search.All(requests: Request[]): Result[] // Like Promise.all - wait for all\nctx.search.Any(requests: Request[]): Result[] // Like Promise.any - first success\nctx.search.Race(requests: Request[]): Result[] // Like Promise.race - first complete\n```\n\n### Options Types\n\n```typescript\ninterface WebOptions {\n limit?: number; // Max results (default: 10)\n sites?: string[]; // Restrict to sites\n timeRange?: string; // \"day\", \"week\", \"month\", \"year\"\n rerank?: RerankOptions;\n}\n\ninterface KBOptions {\n collections?: string[]; // Collection IDs\n threshold?: number; // Similarity threshold (0-1)\n limit?: number; // Max results\n graph?: boolean; // Enable graph association\n rerank?: RerankOptions;\n}\n\ninterface DBOptions {\n models?: string[]; // Model IDs (default: use assistant's db.models)\n wheres?: Where[]; // Pre-defined filters, uses GOU QueryDSL Where format\n orders?: Order[]; // Sort orders, uses GOU QueryDSL Order format\n select?: string[]; // Fields to return\n limit?: number; // Max results (default: 10)\n rerank?: RerankOptions;\n}\n\n// GOU QueryDSL Where condition\n// See: github.com/yaoapp/gou/query/gou/types.go\ninterface Where {\n field: Expression; // Field expression\n value?: any; // Match value\n op: string; // Operator: \"=\", \"like\", \">\", \"<\", \">=\", \"<=\", \"in\", \"is null\", etc.\n or?: boolean; // true for OR condition, default AND\n wheres?: Where[]; // Nested conditions for grouping\n}\n\n// GOU QueryDSL Order\ninterface Order {\n field: Expression; // Field expression\n sort?: string; // \"asc\" or \"desc\"\n}\n\n// GOU Expression (simplified)\ninterface Expression {\n field?: string; // Field name\n table?: string; // Table name (optional)\n}\n\ninterface RerankOptions {\n topN?: number; // Return top N after reranking\n // Note: Reranker type is determined by uses.rerank in agent/agent.yml\n}\n```\n\n### Usage Examples\n\n#### Example 1: Web Search\n\n```typescript\nfunction Create(ctx, messages, options) {\n const query = messages[messages.length - 1].content;\n\n const result = ctx.search.Web(query, {\n limit: 5,\n timeRange: \"week\",\n });\n\n if (result.items.length > 0) {\n return {\n messages: [\n {\n role: \"system\",\n content: formatSearchContext(result),\n },\n ],\n uses: { search: \"disabled\" }, // Disable auto search\n };\n }\n\n return { messages: [] }; // Let auto search handle it\n}\n```\n\n#### Example 2: Knowledge Base Search with Graph\n\n```typescript\nfunction Create(ctx, messages, options) {\n const query = messages[messages.length - 1].content;\n\n const result = ctx.search.KB(query, {\n collections: [\"docs\", \"faq\"],\n threshold: 0.7,\n limit: 10,\n graph: true, // Enable graph association\n });\n\n if (result.items.length > 0) {\n return {\n messages: [\n {\n role: \"system\",\n content: formatKBContext(result),\n },\n ],\n uses: { search: \"disabled\" }, // Disable auto search\n };\n }\n\n return { messages: [] }; // Let auto search handle it\n}\n```\n\n#### Example 3: Database Search\n\n```typescript\nfunction Create(ctx, messages, options) {\n const query = messages[messages.length - 1].content;\n\n // Search in assistant's models (uses db.models from assistant config)\n const result = ctx.search.DB(query, {\n models: [\"product\", \"agents.mybot.order\"], // Optional: override models\n wheres: [{ field: \"status\", value: \"active\" }], // Pre-filter\n limit: 20,\n });\n\n if (result.items.length > 0) {\n return {\n messages: [\n {\n role: \"system\",\n content: formatDBContext(result),\n },\n ],\n uses: { search: \"disabled\" }, // Disable auto search\n };\n }\n\n return { messages: [] }; // Let auto search handle it\n}\n```\n\n#### Example 4: Parallel Search with ctx.search.All()\n\n```typescript\nfunction Create(ctx, messages, options) {\n const query = messages[messages.length - 1].content;\n\n // Execute web, KB, and DB search in parallel (wait for all) - like Promise.all\n const [webResult, kbResult, dbResult] = ctx.search.All([\n { type: \"web\", query: query, limit: 5 },\n { type: \"kb\", query: query, collections: [\"docs\"], limit: 10 },\n { type: \"db\", query: query, models: [\"product\"], limit: 10 },\n ]);\n\n // Merge results\n const context = mergeSearchResults(webResult, kbResult, dbResult);\n\n return {\n messages: [\n {\n role: \"system\",\n content: context,\n },\n ],\n uses: { search: \"disabled\" }, // Disable auto search\n };\n}\n```\n\n#### Example 4b: Parallel Search with ctx.search.Any()\n\n```typescript\nfunction Create(ctx, messages, options) {\n const query = messages[messages.length - 1].content;\n\n // Return as soon as any search succeeds (has results) - like Promise.any\n const results = ctx.search.Any([\n { type: \"web\", query: query, limit: 5 },\n { type: \"kb\", query: query, collections: [\"docs\"], limit: 10 },\n ]);\n\n // Use the first successful result\n const successResult = results.find((r) => r && r.items?.length > 0);\n if (successResult) {\n return {\n messages: [{ role: \"system\", content: formatContext(successResult) }],\n uses: { search: \"disabled\" },\n };\n }\n\n return { messages: [] };\n}\n```\n\n#### Example 4c: Parallel Search with ctx.search.Race()\n\n```typescript\nfunction Create(ctx, messages, options) {\n const query = messages[messages.length - 1].content;\n\n // Return as soon as any search completes (success or not) - like Promise.race\n const results = ctx.search.Race([\n { type: \"web\", query: query, limit: 5 },\n { type: \"kb\", query: query, collections: [\"docs\"], limit: 10 },\n ]);\n\n // Use the first completed result\n const firstResult = results.find((r) => r != null);\n if (firstResult && firstResult.items?.length > 0) {\n return {\n messages: [{ role: \"system\", content: formatContext(firstResult) }],\n uses: { search: \"disabled\" },\n };\n }\n\n return { messages: [] };\n}\n```\n\n#### Example 5: Custom Citation Format\n\n```typescript\nfunction Create(ctx, messages, options) {\n const query = messages[messages.length - 1].content;\n const result = ctx.search.Web(query, { limit: 5 });\n\n // Build custom citation prompt\n const refs = result.items\n .map((item, i) => `[${i + 1}] ${item.title} - ${item.url}`)\n .join(\"\\n\");\n\n return {\n messages: [\n {\n role: \"system\",\n content: `Use [N] to cite. References:\\n${refs}`,\n },\n ],\n uses: { search: \"disabled\" }, // Disable auto search\n citation: { autoInjectPrompt: false }, // Override citation config\n };\n}\n```\n\n## Configuration\n\nConfiguration follows a three-layer hierarchy (later overrides earlier):\n\n1. **System Built-in Defaults** - Hardcoded sensible defaults\n2. **Global Configuration** - `agent/agent.yml` (uses) + `agent/search.yml` (search options)\n3. **Assistant Configuration** - `assistants//package.yao` (uses + search options)\n\n### Uses Configuration\n\nProcessing tools are configured in `agent/agent.yml` under `uses`:\n\n```yaml\n# agent/agent.yml\nuses:\n default: \"yaobots\"\n title: \"workers.system.title\"\n vision: \"workers.system.vision\"\n fetch: \"workers.system.fetch\"\n\n # Search processing tools (NLP)\n keyword: \"builtin\" # \"builtin\", \"workers.nlp.keyword\", \"mcp:my-server.extract_keywords\"\n querydsl: \"builtin\" # \"builtin\", \"workers.nlp.querydsl\", \"mcp:my-server.generate_dsl\"\n rerank: \"builtin\" # \"builtin\", \"workers.rerank\", \"mcp:my-server.rerank\"\n\n # Search handlers\n web: \"builtin\" # \"builtin\", \"workers.search.web\", \"mcp:my-server.web_search\"\n # Note: kb & db always use builtin (access internal data)\n # Note: embedding & entity follow KB collection config\n```\n\nTool format: `\"builtin\"`, `\"\"` (Agent), `\"mcp:.\"` (MCP Tool)\n\n**Web Search Modes:**\n\n| Mode | Example | Description |\n| --------- | ---------------------------- | -------------------------------------------------------------------------- |\n| `builtin` | `\"builtin\"` | Use built-in providers (Tavily, Serper, SerpAPI) |\n| Agent | `\"workers.search.web\"` | AI-powered search: understand intent → optimize query → search → summarize |\n| MCP | `\"mcp:my-server.web_search\"` | External search tool via MCP protocol |\n\n**Why Agent for Web Search (AI Search)?**\n\nWhen `uses.web` is set to an assistant ID, the search flow becomes:\n\n```\nUser Query: \"What's the best laptop for programming in 2024?\"\n │\n ▼\n┌─────────────────────────────────────────────────────────────┐\n│ Agent (workers.search.web) │\n│ 1. Understand intent: laptop recommendations for coding │\n│ 2. Generate optimized queries: │\n│ - \"best programming laptop 2024 review\" │\n│ - \"developer laptop comparison 2024\" │\n│ 3. Execute multiple searches │\n│ 4. Analyze & deduplicate results │\n│ 5. Return structured, relevant results │\n└─────────────────────────────────────────────────────────────┘\n │\n ▼\nHigh-quality, intent-aware search results\n```\n\n### System Built-in Defaults (`defaults/defaults.go`)\n\nThese are the hardcoded defaults, used by `agent/load.go` when loading configuration:\n\n```go\npackage defaults\n\nimport \"github.com/yaoapp/yao/agent/search/types\"\n\n// SystemDefaults provides hardcoded default values\n// Used by agent/load.go for merging with agent/search.yml\nvar SystemDefaults = &types.Config{\n // Web search defaults\n Web: &types.WebConfig{\n Provider: \"tavily\",\n MaxResults: 10,\n },\n\n // KB search defaults\n KB: &types.KBConfig{\n Threshold: 0.7,\n Graph: false,\n },\n\n // DB search defaults\n DB: &types.DBConfig{\n MaxResults: 20,\n },\n\n // Keyword extraction options (uses.keyword)\n Keyword: &types.KeywordConfig{\n MaxKeywords: 10,\n Language: \"auto\",\n },\n\n // QueryDSL generation options (uses.querydsl)\n QueryDSL: &types.QueryDSLConfig{\n Strict: false,\n },\n\n // Rerank options (uses.rerank)\n Rerank: &types.RerankConfig{\n TopN: 10,\n },\n\n // Citation\n Citation: &types.CitationConfig{\n Format: \"#ref:{id}\",\n AutoInjectPrompt: true,\n },\n\n // Source weights\n Weights: &types.WeightsConfig{\n User: 1.0,\n Hook: 0.8,\n Auto: 0.6,\n },\n\n // Behavior options\n Options: &types.OptionsConfig{\n SkipThreshold: 5,\n },\n}\n\n// GetWeight returns the weight for a source type\nfunc GetWeight(cfg *types.Config, source types.SourceType) float64 {\n if cfg == nil || cfg.Weights == nil {\n switch source {\n case types.SourceUser:\n return 1.0\n case types.SourceHook:\n return 0.8\n default:\n return 0.6\n }\n }\n switch source {\n case types.SourceUser:\n return cfg.Weights.User\n case types.SourceHook:\n return cfg.Weights.Hook\n case types.SourceAuto:\n return cfg.Weights.Auto\n default:\n return 0.6\n }\n}\n```\n\n### Configuration Loading (in `agent/load.go`)\n\nConfiguration loading follows the existing pattern in `agent/load.go`:\n\n```go\n// agent/load.go\n\nimport (\n searchDefaults \"github.com/yaoapp/yao/agent/search/defaults\"\n searchTypes \"github.com/yaoapp/yao/agent/search/types\"\n)\n\nvar searchConfig *searchTypes.Config\n\n// initSearchConfig initialize the search configuration from agent/search.yml\nfunc initSearchConfig() error {\n // Start with system defaults\n searchConfig = searchDefaults.SystemDefaults\n\n path := filepath.Join(\"agent\", \"search.yml\")\n if exists, _ := application.App.Exists(path); !exists {\n return nil // Use defaults\n }\n\n // Read and merge with defaults\n bytes, err := application.App.Read(path)\n if err != nil {\n return err\n }\n\n var cfg searchTypes.Config\n err = application.Parse(\"search.yml\", bytes, &cfg)\n if err != nil {\n return err\n }\n\n // Merge: defaults < global config\n searchConfig = mergeSearchConfig(searchDefaults.SystemDefaults, &cfg)\n return nil\n}\n\n// GetSearchConfig returns the global search configuration\nfunc GetSearchConfig() *searchTypes.Config {\n return searchConfig\n}\n```\n\n### Assistant-level Config Merge (in `agent/assistant/load.go`)\n\nAssistant-specific search config is merged in `assistant/load.go`:\n\n```go\n// agent/assistant/load.go\n\n// GetMergedSearchConfig returns merged search config for this assistant\nfunc (ast *Assistant) GetMergedSearchConfig() *searchTypes.Config {\n globalCfg := agent.GetSearchConfig()\n if ast.Search == nil {\n return globalCfg\n }\n // Merge: global < assistant\n return mergeSearchConfig(globalCfg, ast.Search.ToConfig())\n}\n```\n\n### Global Configuration\n\n`agent/search.yml` - Override system defaults for all assistants:\n\n```yaml\n# Global Search Configuration\n# These settings apply to all assistants unless overridden by assistant-specific configurations.\n\n# Web search settings\nweb:\n provider: \"tavily\" # \"tavily\", \"serper\", or \"serpapi\" (builtin providers only)\n api_key_env: \"TAVILY_API_KEY\"\n max_results: 10\n # engine: \"google\" # For SerpAPI only: \"google\", \"bing\", \"baidu\", \"yandex\", etc.\n\n# Knowledge base search settings\nkb:\n threshold: 0.7 # Similarity threshold\n graph: false # Enable GraphRAG association\n\n# Database search settings\ndb:\n max_results: 20\n\n# Keyword extraction options (uses.keyword)\nkeyword:\n max_keywords: 10\n language: \"auto\" # \"auto\", \"en\", \"zh\", etc.\n\n# QueryDSL generation options (uses.querydsl)\nquerydsl:\n strict: false # Strict mode: fail if generation fails\n\n# Rerank options (uses.rerank)\nrerank:\n top_n: 10 # Return top N results after reranking\n\n# Citation format for LLM references\ncitation:\n format: \"#ref:{id}\"\n auto_inject_prompt: true # Auto-inject citation instructions to system prompt\n\n# Source weighting for result merging\nweights:\n user: 1.0 # User-provided DataContent (highest priority)\n hook: 0.8 # Hook ctx.search.*() results\n auto: 0.6 # Auto search results\n\n# Search behavior options\noptions:\n skip_threshold: 5 # Skip auto search if user provides >= N results\n```\n\n### Assistant Configuration\n\n`assistants//package.yao` - Override for specific assistant:\n\n```jsonc\n{\n \"name\": \"My Assistant\",\n \"connector\": \"openai\",\n\n // Overrides global uses (agent/agent.yml)\n \"uses\": {\n \"search\": \"builtin\", // \"builtin\", \"disabled\", \"\", \"mcp:.\"\n \"web\": \"builtin\", // \"builtin\", \"\", \"mcp:.\"\n \"keyword\": \"workers.nlp.keyword\", // Use LLM for keyword extraction\n \"querydsl\": \"workers.nlp.querydsl\", // Use LLM for QueryDSL generation\n \"rerank\": \"mcp:my-server.rerank\" // Use MCP tool for reranking\n },\n\n // Search configuration (overrides agent/search.yml)\n \"search\": {\n // Overrides global web settings\n \"web\": {\n \"provider\": \"tavily\",\n \"max_results\": 5\n },\n\n // Overrides global kb settings\n \"kb\": {\n \"collections\": [\"docs\", \"faq\"], // Specific collections to search\n \"threshold\": 0.7,\n \"graph\": true\n },\n\n // Overrides global db settings\n \"db\": {\n \"models\": [\"product\", \"order\"], // Uses db.models if not set\n \"max_results\": 20\n },\n\n // Overrides global keyword options\n \"keyword\": {\n \"max_keywords\": 5\n },\n\n // Overrides global querydsl options\n \"querydsl\": {\n \"strict\": true\n },\n\n // Overrides global rerank options\n \"rerank\": {\n \"top_n\": 5\n },\n\n // Overrides global citation settings\n \"citation\": {\n \"format\": \"#ref:{id}\",\n \"auto_inject_prompt\": true\n }\n },\n\n // Knowledge base collections available to this assistant\n \"kb\": {\n \"collections\": [\"docs\", \"faq\"]\n },\n\n // Database models available to this assistant\n \"db\": {\n \"models\": [\"product\", \"order\", \"customer\"]\n }\n}\n```\n\n## Execution Flow\n\n### Search Flow\n\n## Execution Modes\n\n### Stream() Execution with Search\n\n```\nStream(ctx, messages, options)\n │\n ├── 1. Initialize\n │\n ├── 2. Create Hook (optional)\n │ └── Can call ctx.search.* and return search results\n │\n ├── 3. BuildRequest + BuildContent\n │\n ├── 4. Auto Search Decision (shouldAutoSearch)\n │ ├── IF Uses.Search == \"disabled\" → SKIP\n │ ├── IF Create Hook returned uses.search=\"disabled\" → SKIP\n │ └── ELSE → Execute Auto Search (executeAutoSearch)\n │ ├── Read assistant's search config (GetMergedSearchConfig)\n │ ├── Extract keywords (if uses.keyword && !Skip.Keyword)\n │ ├── Build search requests (buildSearchRequests)\n │ ├── Execute web/kb/db in parallel (searcher.All)\n │ ├── Build reference context (BuildReferenceContext)\n │ └── Inject search context to messages (injectSearchContext)\n │\n ├── 5. LLM Call (with search context if any)\n │\n ├── 6. Next Hook (optional)\n │\n └── 7. Output (response may contain #ref:xxx citations)\n```\n\n**Implementation Files:**\n\n| File | Description |\n| --------------------- | ----------------------------------------------- |\n| `assistant/search.go` | Core integration logic (shouldAutoSearch, etc.) |\n| `assistant/agent.go` | Stream() integration point (after BuildContent) |\n| `search/reference.go` | BuildReferenceContext, FormatReferencesXML |\n\n**Key Functions (`assistant/search.go`):**\n\n```go\n// shouldAutoSearch determines if auto search should be executed\nfunc (ast *Assistant) shouldAutoSearch(ctx *context.Context, createResponse *context.HookCreateResponse) bool\n\n// executeAutoSearch executes auto search based on configuration\n// opts is optional, used to check Skip.Keyword for keyword extraction\nfunc (ast *Assistant) executeAutoSearch(ctx *context.Context, messages []context.Message, createResponse *context.HookCreateResponse, opts ...*context.Options) *searchTypes.ReferenceContext\n\n// injectSearchContext injects search results into messages\nfunc (ast *Assistant) injectSearchContext(messages []context.Message, refCtx *searchTypes.ReferenceContext) []context.Message\n\n// getMergedSearchUses returns the merged uses configuration for search\nfunc (ast *Assistant) getMergedSearchUses(createResponse *context.HookCreateResponse) *context.Uses\n\n// buildSearchRequests builds search requests based on assistant configuration\nfunc (ast *Assistant) buildSearchRequests(query string, config *searchTypes.Config) []*searchTypes.Request\n```\n\n**Keyword Extraction in executeAutoSearch:**\n\nWhen `uses.keyword` is configured and `opts.Skip.Keyword` is not true, keyword extraction is performed before web search:\n\n```go\n// Extract keywords for web search if:\n// 1. uses.keyword is configured (not empty)\n// 2. Skip.Keyword is not true\n// 3. Web search is enabled\nif webSearchEnabled && !skipKeyword && searchUses.Keyword != \"\" {\n extractor := keyword.NewExtractor(searchUses.Keyword, searchConfig.Keyword)\n keywords, err := extractor.Extract(ctx, query, nil)\n if err == nil && len(keywords) > 0 {\n query = strings.Join(keywords, \" \")\n }\n}\n```\n\n**Integration in agent.go:**\n\n```go\n// In Stream(), after BuildContent:\nif ast.shouldAutoSearch(ctx, createResponse) {\n refCtx := ast.executeAutoSearch(ctx, completionMessages, createResponse, opts)\n if refCtx != nil && len(refCtx.References) > 0 {\n completionMessages = ast.injectSearchContext(completionMessages, refCtx)\n }\n}\n```\n\n**Skip.Keyword Option (`context.Options.Skip`):**\n\n```go\ntype Skip struct {\n History bool `json:\"history\"` // Skip saving chat history\n Trace bool `json:\"trace\"` // Skip trace logging\n Output bool `json:\"output\"` // Skip output to client\n Keyword bool `json:\"keyword\"` // Skip keyword extraction for web search\n}\n```\n\nUse `Skip.Keyword = true` when you want to use the raw query directly without keyword extraction.\n\n### Control via Uses.Search\n\nSearch is controlled via the `Uses` mechanism, following the merge hierarchy:\n\n```\nGlobal (agent/agent.yml) → Assistant (package.yao) → CreateHook (return uses) → Request (options.uses)\n```\n\n| Uses.Search | Behavior |\n| ----------------------- | ------------------------------------ |\n| `\"builtin\"` | Use builtin auto search |\n| `\"disabled\"` | Disable auto search |\n| `\"\"` | Delegate to an assistant (AI Search) |\n| `\"mcp:.\"` | Use MCP tool for search |\n| `undefined` | Follow upper layer config (default) |\n\n**Go:**\n\n```go\n// Use builtin auto search\nuses := &context.Uses{Search: \"builtin\"}\n\n// Disable auto search\nuses := &context.Uses{Search: \"disabled\"}\n\n// Delegate to AI Search assistant\nuses := &context.Uses{Search: \"workers.search.ai\"}\n\n// Follow assistant config (default)\nuses := &context.Uses{Search: \"\"} // or nil\n```\n\n**API Request:**\n\n```json\n{\n \"messages\": [...],\n \"uses\": {\n \"search\": \"builtin\"\n }\n}\n```\n\n### Hook-Controlled Search\n\nSearch is controlled via the `Uses` mechanism, same as Vision/Audio. The merge hierarchy is:\n\n```\nGlobal (agent/agent.yml) → Assistant (package.yao) → CreateHook (return uses)\n```\n\nWhen you need custom search logic, handle it in Create Hook and return `uses.search` to control auto search:\n\n```typescript\nfunction Create(ctx, messages, options) {\n const query = messages[messages.length - 1].content;\n\n // Custom logic: only search for certain queries\n if (needsSearch(query)) {\n const result = ctx.search.Web(query, { limit: 5 });\n return {\n messages: [{ role: \"system\", content: formatContext(result) }],\n uses: { search: \"disabled\" }, // Disable auto search (hook handled it)\n };\n }\n\n // Let auto search handle it (follow assistant config)\n return { messages: [] };\n}\n```\n\n**Uses.Search Values:**\n\n| Value | Behavior |\n| ----------------------- | ------------------------------------ |\n| `\"builtin\"` | Use builtin auto search |\n| `\"disabled\"` | Disable auto search |\n| `\"\"` | Delegate to an assistant (AI Search) |\n| `\"mcp:.\"` | Use MCP tool for search |\n| `undefined` | Follow upper layer config (default) |\n\n**Uses Merge Hierarchy:**\n\n```\n┌─────────────────────────────────────────────────────────────┐\n│ 1. Global Config (agent/agent.yml) │\n│ uses: │\n│ search: \"builtin\" │\n└─────────────────────────────────────────────────────────────┘\n ↓ merge\n┌─────────────────────────────────────────────────────────────┐\n│ 2. Assistant Config (assistants//package.yao) │\n│ uses: │\n│ search: \"workers.search.web\" # Override to AI Search │\n└─────────────────────────────────────────────────────────────┘\n ↓ merge\n┌─────────────────────────────────────────────────────────────┐\n│ 3. CreateHook Return │\n│ return { │\n│ uses: { search: \"disabled\" } # Hook handled it │\n│ } │\n└─────────────────────────────────────────────────────────────┘\n```\n\n> **Note**: The `Uses` struct in `context/types_llm.go` already has a `Search` field.\n> The value `\"disabled\"` is a special value to disable auto search when hook handles it.\n\n## Search Flow\n\n```\nRequest → Trace Start → Query Process → Search → Rerank → Citations → Output → Return\n```\n\n### Query Processing\n\n| Type | Process | Tool Config |\n| ---- | ----------------------------------------------------- | -------------------- |\n| Web | Extract keywords → Build query | `uses.keyword` |\n| KB | Get collection's embedding model → Generate embedding | KB collection config |\n| DB | Parse query → Build QueryDSL → Execute on models | `uses.querydsl` |\n\n#### Processing Methods\n\nConfigure via `uses.*` in `agent/agent.yml`:\n\n| Format | Description | Use Case |\n| --------------------- | ----------------------------------------- | ------------------------------ |\n| `builtin` | Rule-based, template-driven (no LLM call) | Fast, low cost, simple queries |\n| `` | Delegate to an assistant (Agent) | LLM-based, custom logic |\n| `mcp:.` | Call MCP tool | External services integration |\n\n#### Keyword Extraction (`nlp/keyword/`)\n\nConfigure via `uses.keyword`. The keyword extraction module follows the Handler + Registry pattern with three modes:\n\n| Mode | Value | Description |\n| ------- | ---------------------------- | --------------------------------------------- |\n| Builtin | `\"builtin\"` | Frequency-based extraction (no external deps) |\n| Agent | `\"workers.nlp.keyword\"` | LLM-powered semantic extraction |\n| MCP | `\"mcp:nlp.extract_keywords\"` | External service via MCP |\n\n**Directory Structure:**\n\n```\nnlp/keyword/\n├── extractor.go # Main entry point (mode dispatch)\n├── builtin.go # Builtin: frequency-based, stopword filtering\n├── agent.go # Agent: delegate to LLM assistant\n└── mcp.go # MCP: call external tool\n```\n\n**Usage:**\n\n```go\n// nlp/keyword/extractor.go\npackage keyword\n\n// Extractor extracts keywords from text\ntype Extractor struct {\n usesKeyword string // \"builtin\", \"\", \"mcp:.\"\n config *types.KeywordConfig\n}\n\n// NewExtractor creates a new keyword extractor\nfunc NewExtractor(usesKeyword string, cfg *types.KeywordConfig) *Extractor\n\n// Extract extracts keywords based on configured mode\nfunc (e *Extractor) Extract(ctx *context.Context, content string, opts *types.KeywordOptions) ([]string, error)\n```\n\n**Builtin Implementation:**\n\nThe builtin extractor uses simple frequency-based extraction with no external dependencies:\n\n- Tokenization (handles English and Chinese)\n- Stop word filtering (common English and Chinese stop words)\n- Frequency counting and ranking\n- Returns top N keywords by frequency\n\n> **Note**: For production use cases requiring high accuracy (semantic understanding, phrase extraction), use Agent or MCP mode.\n\n**Example:**\n\n```\n\"I want to find the best wireless headphones under $100\"\n ↓ builtin: tokenization + stopword removal + frequency ranking\n → [\"wireless\", \"headphones\", \"find\", \"best\"]\n ↓ agent: LLM semantic extraction\n → [\"wireless headphones\", \"under $100\", \"best\"]\n```\n\n#### Embedding (KB Collection Config)\n\nEmbedding is **not** part of the `nlp/` package. It follows KB collection's own configuration:\n\n- Each KB collection defines its own embedding provider and model\n- The KB handler (`handlers/kb/`) calls the collection's embedding API directly\n- Entity types for GraphRAG are also defined per collection\n\n```go\n// handlers/kb/handler.go\nfunc (h *Handler) Search(ctx *context.Context, req *types.Request) (*types.Result, error) {\n // 1. Get collection config (embedding provider, model)\n collection := h.getCollection(req.Collections[0])\n\n // 2. Generate embedding using collection's config\n vector, err := collection.Embed(ctx, req.Query)\n\n // 3. Vector search\n // ...\n}\n```\n\n#### QueryDSL Generation (`nlp/querydsl/`)\n\nConfigure via `uses.querydsl`. The QueryDSL generation module follows the same pattern as keyword extraction:\n\n| Mode | Value | Description |\n| ------- | ----------------------------- | ------------------------------------------- |\n| Builtin | `\"builtin\"` | Template-based generation from model schema |\n| Agent | `\"workers.nlp.querydsl\"` | LLM-powered semantic query generation |\n| MCP | `\"mcp:nlp.generate_querydsl\"` | External service via MCP |\n\n**Directory Structure:**\n\n```\nnlp/querydsl/\n├── generator.go # Main entry point (mode dispatch)\n├── builtin.go # Builtin: template-based generation\n├── agent.go # Agent: delegate to LLM assistant\n└── mcp.go # MCP: call external tool\n```\n\n**Usage:**\n\n```go\n// nlp/querydsl/generator.go\npackage querydsl\n\n// Generator generates QueryDSL from natural language\ntype Generator struct {\n usesQueryDSL string\n config *types.QueryDSLConfig\n}\n\n// NewGenerator creates a new QueryDSL generator\nfunc NewGenerator(usesQueryDSL string, cfg *types.QueryDSLConfig) *Generator\n\n// Generate converts natural language to QueryDSL\n// Uses GOU types directly: model.Model and gou.QueryDSL\nfunc (g *Generator) Generate(query string, models []*model.Model) (*gou.QueryDSL, error)\n```\n\n**Example:**\n\n```\n\"Products cheaper than $100 from Apple\"\n ↓ builtin: template matching against model schema\n → QueryDSL with simple keyword matching\n ↓ agent: LLM generates DSL from NL + schema\n → QueryDSL: {\"wheres\": [{\"column\": \"price\", \"op\": \"<\", \"value\": 100}, {\"column\": \"brand\", \"value\": \"Apple\"}]}\n```\n\n## Handlers & Providers\n\nAll handler implementations are in `search/handlers/` directory.\n\n### Web Search (`handlers/web/`)\n\nWeb search supports three modes via `uses.web`:\n\n| Mode | Value | Description |\n| ------- | ---------------------------- | ------------------------------------------- |\n| Builtin | `\"builtin\"` | Direct API calls to Tavily/Serper/SerpAPI |\n| Agent | `\"workers.search.web\"` | AI-powered search with intent understanding |\n| MCP | `\"mcp:my-server.web_search\"` | External search tool via MCP |\n\n```go\n// handlers/web/handler.go\npackage web\n\nimport (\n \"strings\"\n\n agentContext \"github.com/yaoapp/yao/agent/context\"\n \"github.com/yaoapp/yao/agent/search/types\"\n)\n\n// Handler implements web search\ntype Handler struct {\n usesWeb string // \"builtin\", \"\", \"mcp:.\"\n config *types.WebConfig\n}\n\n// NewHandler creates a new web search handler\nfunc NewHandler(usesWeb string, cfg *types.WebConfig) *Handler\n\n// Type returns the search type this handler supports\nfunc (h *Handler) Type() types.SearchType\n\n// Search implements interfaces.Handler (without context)\nfunc (h *Handler) Search(req *types.Request) (*types.Result, error)\n\n// SearchWithContext executes web search with context (for Agent/MCP modes)\nfunc (h *Handler) SearchWithContext(ctx *agentContext.Context, req *types.Request) (*types.Result, error)\n```\n\n**Directory Structure:**\n\n```\nhandlers/web/\n├── handler.go # Main entry point (mode dispatch)\n├── tavily.go # Tavily provider (builtin)\n├── serper.go # Serper provider (serper.dev)\n├── serpapi.go # SerpAPI provider (serpapi.com, multi-engine)\n├── agent.go # Agent mode (AI Search)\n└── mcp.go # MCP mode (external service)\n```\n\n**Built-in Providers (when `uses.web = \"builtin\"`):**\n\n| Provider | File | Notes |\n| -------- | ------------ | ----------------------------------------------- |\n| Tavily | `tavily.go` | Recommended for AI applications |\n| Serper | `serper.go` | Google search via serper.dev (POST + X-API-KEY) |\n| SerpAPI | `serpapi.go` | Multi-engine search via serpapi.com (GET + URL) |\n\n**SerpAPI Engine Support:**\n\nSerpAPI supports multiple search engines via the `engine` config:\n\n| Engine | Description |\n| ------------ | ---------------------------- |\n| `google` | Google Search (default) |\n| `bing` | Bing Search |\n| `baidu` | Baidu Search (Chinese) |\n| `yandex` | Yandex Search |\n| `yahoo` | Yahoo Search |\n| `duckduckgo` | DuckDuckGo Search |\n| `naver` | Naver Search (Korean) |\n| `ecosia` | Ecosia Search (eco-friendly) |\n| `seznam` | Seznam Search (Czech) |\n\nSee [SerpAPI Documentation](https://serpapi.com/search-api) for the full list of supported engines.\n\nConfiguration example:\n\n```yaml\n# agent/search.yml\nweb:\n provider: \"serpapi\"\n api_key_env: \"SERPAPI_API_KEY\"\n engine: \"bing\" # Use Bing instead of Google\n max_results: 10\n```\n\n**Agent Mode (AI Search):**\n\nWhen `uses.web` is set to an assistant ID (e.g., `\"workers.search.web\"`), the assistant can:\n\n1. **Understand user intent** - Parse complex queries, identify what user really wants\n2. **Generate multiple queries** - Create optimized search terms for better coverage\n3. **Multi-source search** - Search multiple providers or sources\n4. **Result analysis** - Deduplicate, rank, and summarize results\n5. **Context-aware** - Use conversation context to improve search relevance\n\n```\nUser Query: \"What's the best laptop for programming in 2024?\"\n │\n ▼\n┌─────────────────────────────────────────────────────────────┐\n│ Agent (workers.search.web) │\n│ 1. Understand intent: laptop recommendations for coding │\n│ 2. Generate optimized queries: │\n│ - \"best programming laptop 2024 review\" │\n│ - \"developer laptop comparison 2024\" │\n│ 3. Execute multiple searches via builtin providers │\n│ 4. Analyze & deduplicate results │\n│ 5. Return structured, relevant results │\n└─────────────────────────────────────────────────────────────┘\n │\n ▼\nHigh-quality, intent-aware search results\n```\n\n**Example AI Search Assistant:**\n\n```typescript\n// assistants/workers/search/web/src/index.ts\nfunction Create(ctx, messages, options) {\n const userQuery = messages[messages.length - 1].content;\n\n // 1. Analyze intent (this assistant has access to LLM)\n const intent = analyzeIntent(ctx, userQuery);\n\n // 2. Generate optimized queries\n const queries = generateQueries(intent);\n\n // 3. Execute searches using builtin provider\n const allResults = [];\n for (const q of queries) {\n const result = ctx.search.Web(q, {\n provider: \"tavily\", // Use builtin provider\n limit: 5,\n });\n allResults.push(...result.items);\n }\n\n // 4. Merge, deduplicate, and rank results\n const merged = mergeAndRank(allResults, intent);\n\n return {\n type: \"search_result\",\n items: merged,\n };\n}\n```\n\n### Knowledge Base (`handlers/kb/`)\n\n```go\n// handlers/kb/handler.go\npackage kb\n\nimport (\n \"github.com/yaoapp/yao/agent/search/types\"\n)\n\n// Handler implements KB search\ntype Handler struct {\n config *types.KBConfig\n}\n\n// NewHandler creates a new KB search handler\nfunc NewHandler(cfg *types.KBConfig) *Handler\n\n// Type returns the search type this handler supports\nfunc (h *Handler) Type() types.SearchType\n\n// Search executes vector search and optional graph association\n// TODO: Implement actual search logic\nfunc (h *Handler) Search(req *types.Request) (*types.Result, error)\n```\n\n| File | Description |\n| ------------ | ---------------------------------- |\n| `handler.go` | Main KB handler implementation |\n| `vector.go` | Vector similarity search |\n| `graph.go` | Graph-based association (GraphRAG) |\n\n### Database Search (`handlers/db/`)\n\n```go\n// handlers/db/handler.go\npackage db\n\nimport (\n \"github.com/yaoapp/yao/agent/search/types\"\n)\n\n// Handler implements DB search\ntype Handler struct {\n usesQueryDSL string // \"builtin\", \"\", \"mcp:.\"\n config *types.DBConfig\n}\n\n// NewHandler creates a new DB search handler\nfunc NewHandler(usesQueryDSL string, cfg *types.DBConfig) *Handler\n\n// Type returns the search type this handler supports\nfunc (h *Handler) Type() types.SearchType\n\n// Search converts NL to QueryDSL and executes\n// TODO: Implement actual search logic\nfunc (h *Handler) Search(req *types.Request) (*types.Result, error)\n```\n\n| File | Description |\n| ------------ | ------------------------------ |\n| `handler.go` | Main DB handler implementation |\n| `query.go` | QueryDSL builder utilities |\n| `schema.go` | Model schema introspection |\n\nIntegrates with Yao's Model/QueryDSL system:\n\n- Natural language → QueryDSL conversion (via LLM)\n- Model schema introspection for query building\n- Support for:\n - Global models (`models/*.mod.yao`)\n - Assistant-specific models (`assistants/{id}/models/*.mod.yao` → `agents.{id}.*`)\n- Permission-aware queries (respects `__yao_*` permission fields)\n\n### Reranking (`rerank/`)\n\nThe rerank module follows the Handler + Registry pattern, consistent with `keyword/` and `web/`.\n\n```go\n// rerank/reranker.go\npackage rerank\n\nimport (\n \"strings\"\n\n \"github.com/yaoapp/yao/agent/context\"\n \"github.com/yaoapp/yao/agent/search/types\"\n)\n\n// Reranker reorders search results by relevance\n// Mode is determined by uses.rerank configuration\ntype Reranker struct {\n usesRerank string // \"builtin\", \"\", \"mcp:.\"\n config *types.RerankConfig\n}\n\n// NewReranker creates a new reranker\nfunc NewReranker(usesRerank string, cfg *types.RerankConfig) *Reranker\n\n// Rerank reorders results based on configured mode\nfunc (r *Reranker) Rerank(ctx *context.Context, query string, items []*types.ResultItem, opts *types.RerankOptions) ([]*types.ResultItem, error)\n```\n\n**Directory Structure:**\n\n```\nrerank/\n├── reranker.go # Main entry point (mode dispatch)\n├── builtin.go # Builtin: weighted score sorting (score * weight)\n├── agent.go # Agent mode (delegate to LLM assistant)\n└── mcp.go # MCP mode (external service)\n```\n\n**Builtin Implementation:**\n\nThe builtin reranker uses weighted score sorting:\n\n- Calculate `weightedScore = score * weight`\n- Sort items by weighted score descending\n- Return top N items\n\n> **Note**: For production use cases requiring semantic understanding, use Agent or MCP mode.\n\n**Agent Response Format:**\n\nThe agent should return reordered items in one of these formats:\n\n```json\n// Format 1: Order list (recommended)\n{ \"order\": [\"ref_003\", \"ref_001\", \"ref_002\"] }\n\n// Format 2: Items list with citation_id\n{ \"items\": [{ \"citation_id\": \"ref_003\" }, { \"citation_id\": \"ref_001\" }] }\n```\n\n| File | Description |\n| ------------- | ---------------------------------------- |\n| `reranker.go` | Main entry point and mode dispatch |\n| `builtin.go` | Weighted score sorting (score \\* weight) |\n| `agent.go` | Delegate to LLM assistant for reranking |\n| `mcp.go` | Call external MCP tool for reranking |\n\nConfigure via `uses.rerank` in `agent/agent.yml`:\n\n| Value | Notes |\n| ---------------------- | -------------------------------- |\n| `builtin` | Simple score sorting (default) |\n| `workers.rerank` | Delegate to an assistant (Agent) |\n| `mcp:my-server.rerank` | Call MCP tool for reranking |\n\n## Error Handling\n\nSearch errors don't block the agent flow. Errors are returned in `Result.Error`:\n\n```typescript\nconst result = ctx.search.Web(query);\nif (result.error) {\n // Handle gracefully or fallback\n console.warn(\"Search failed:\", result.error);\n}\n```\n\n## Configuration Priority\n\nConfiguration is merged with later layers overriding earlier ones:\n\n1. **System Built-in** - Hardcoded defaults (lowest priority)\n2. **Global-level** - `agent/agent.yml` (uses) + `agent/search.yml` (search options)\n3. **Assistant-level** - `assistants//package.yao` (uses + search)\n4. **Hook-level** - CreateHook return `uses.search` value\n5. **Request-level** - `options.uses.search` in Stream() call (highest priority)\n - `\"builtin\"`: Use builtin auto search\n - `\"disabled\"`: Disable auto search\n - `\"\"`: Delegate to AI Search assistant\n - `\"mcp:.\"`: Use MCP tool for search\n\n## DB Search Details\n\n### Query Processing Flow\n\n```\nNatural Language Query\n │\n ▼\n┌─────────────────────────────────┐\n│ Get Model Schemas │ ← Introspect models from db.models config\n│ (fields, types, relations) │\n└─────────────────────────────────┘\n │\n ▼\n┌─────────────────────────────────┐\n│ LLM: Generate QueryDSL │ ← Convert NL to Yao QueryDSL\n│ (select, wheres, orders) │\n└─────────────────────────────────┘\n │\n ▼\n┌─────────────────────────────────┐\n│ Execute Query on Each Model │ ← model.Find() with QueryDSL\n└─────────────────────────────────┘\n │\n ▼\n Results\n```\n\n### Model ID Formats\n\n| Format | Example | Description |\n| ------ | -------------------- | --------------------------------------------------------------------- |\n| Global | `product` | Global model from `models/product.mod.yao` |\n| System | `__yao.user` | Yao system model |\n| Agent | `agents.mybot.order` | Assistant-specific model from `assistants/mybot/models/order.mod.yao` |\n\n### QueryDSL Generation Prompt\n\nThe DB handler uses LLM to convert natural language to QueryDSL:\n\n```\nGiven the following model schemas:\n- product: { id, name, price, category, status, created_at }\n- order: { id, product_id, quantity, total, customer_id, status }\n\nUser query: \"find all active products under $100 in electronics category\"\n\nGenerate Yao QueryDSL:\n{\n \"model\": \"product\",\n \"wheres\": [\n { \"field\": \"status\", \"op\": \"=\", \"value\": \"active\" },\n { \"field\": \"price\", \"op\": \"<\", \"value\": 100 },\n { \"field\": \"category\", \"op\": \"=\", \"value\": \"electronics\" }\n ],\n \"orders\": [{ \"field\": \"price\", \"order\": \"asc\" }],\n \"limit\": 10\n}\n```\n\n## Content Module Integration\n\nUser messages may contain `type=\"data\"` ContentParts with data source references. The `content` module processes these before LLM call.\n\n### DataSource Types (from `context/types.go`)\n\n```go\nconst (\n DataSourceModel DataSourceType = \"model\" // DB model query\n DataSourceKBCollection DataSourceType = \"kb_collection\" // KB collection search\n DataSourceKBDocument DataSourceType = \"kb_document\" // KB document retrieval\n DataSourceTable DataSourceType = \"table\" // Direct table query\n DataSourceAPI DataSourceType = \"api\" // External API\n DataSourceMCPResource DataSourceType = \"mcp_resource\" // MCP resource\n)\n```\n\n### Message with Data Reference\n\nUser only specifies data source IDs. Filters are generated by Search module from natural language.\n\n```json\n{\n \"role\": \"user\",\n \"content\": [\n { \"type\": \"text\", \"text\": \"Show me products under $100\" },\n {\n \"type\": \"data\",\n \"data\": {\n \"sources\": [\n { \"type\": \"model\", \"name\": \"product\" },\n { \"type\": \"kb_collection\", \"name\": \"product-docs\" }\n ]\n }\n }\n ]\n}\n```\n\nThe Search module will:\n\n1. Extract query from text: \"products under $100\"\n2. For `model:product` → Generate QueryDSL: `{ \"wheres\": [{ \"field\": \"price\", \"op\": \"<\", \"value\": 100 }] }`\n3. For `kb_collection:product-docs` → Vector search with query embedding\n\n### Source Weighting & LLM Context\n\nSearch results carry `source` and `weight` fields, which are used to build weighted context for LLM.\n\n**Source Types:**\n\n| Source | Weight | Description |\n| ------ | ------ | -------------------------------- |\n| `user` | 1.0 | Explicitly referenced in message |\n| `hook` | 0.8 | Called in Create/Next hook |\n| `auto` | 0.6 | Triggered by assistant config |\n\n**ResultItem with Weight:**\n\n```go\ntype ResultItem struct {\n CitationID string `json:\"citation_id\"` // \"#ref:xxx\"\n Source string `json:\"source\"` // \"user\", \"hook\", \"auto\"\n Weight float64 `json:\"weight\"` // 1.0, 0.8, 0.6\n Score float64 `json:\"score\"` // Relevance score\n // ... other fields\n}\n```\n\n### Unified Context Protocol\n\nAll data sources (Content module, Hook, Auto-Search) produce the same `Reference` structure. The final LLM input uses a unified `` format.\n\n**Reference (Internal Structure):**\n\n```go\n// Reference is the unified structure for all data sources\ntype Reference struct {\n ID string `json:\"id\"` // Unique citation ID: \"ref_001\", \"ref_002\"\n Type string `json:\"type\"` // \"web\", \"kb\", \"db\"\n Source string `json:\"source\"` // \"user\", \"hook\", \"auto\"\n Weight float64 `json:\"weight\"` // 1.0, 0.8, 0.6\n Score float64 `json:\"score\"` // Relevance score (0-1)\n Title string `json:\"title\"` // Optional title\n Content string `json:\"content\"` // Main content\n URL string `json:\"url\"` // Optional URL\n Meta map[string]interface{} `json:\"meta\"` // Additional metadata\n}\n```\n\n**Data Flow:**\n\n```mermaid\nflowchart TD\n subgraph Sources [\"Data Sources\"]\n CM[\"Content Module
(db:xxx kb:xxx)\"]\n HS[\"Hook Search
ctx.search.*()\"]\n AS[\"Auto Search
(assistant config)\"]\n end\n\n CM -->|\"source=user
weight=1.0\"| REF\n HS -->|\"source=hook
weight=0.8\"| REF\n AS -->|\"source=auto
weight=0.6\"| REF\n\n REF[\"[]Reference
(Unified Structure)\"]\n REF --> MERGE[\"Merge & Deduplicate
Rerank by score × weight\"]\n MERGE --> BUILD[\"Build <references> XML\"]\n BUILD --> LLM[\"LLM Input\"]\n```\n\n**LLM References Format:**\n\n```xml\n\n\nProduct: iPhone 15 Pro\nPrice: $999\nCategory: Electronics\n\n\nThe iPhone 15 Pro features the A17 Pro chip with improved performance...\nURL: https://example.com/iphone-review\n\n\nApple announced the iPhone 15 series in September 2023...\nURL: https://news.example.com/apple-iphone-15\n\n\n```\n\n**LLM System Prompt (auto-injected):**\n\n```\nYou have access to reference data in tags. Each has:\n- id: Citation identifier\n- type: Data type (web/kb/db)\n- weight: Relevance weight (1.0=highest priority, 0.6=lowest)\n- source: Origin (user=user-provided, hook=assistant-searched, auto=auto-searched)\n\nPrioritize higher-weight references when answering.\n\nWhen citing a reference, use this exact HTML format:\n[{id}]\n\nExample: According to the product data[ref_001], the price is $999.\n```\n\n**Citation Output Format:**\n\nLLM outputs citations as HTML links that can be parsed and rendered by frontend:\n\n```html\n\nThe iPhone 15 Pro[ref_001]\nfeatures the A17 Pro chip[ref_002].\n```\n\n**Citation Link Attributes:**\n\n| Attribute | Description | Example |\n| --------------- | ----------------------- | ----------------------- |\n| `class` | Fixed class for styling | `\"ref\"` |\n| `data-ref-id` | Reference ID | `\"ref_001\"` |\n| `data-ref-type` | Data type | `\"db\"`, `\"kb\"`, `\"web\"` |\n| `href` | Anchor link | `\"#ref:ref_001\"` |\n\n**Conversion Examples:**\n\n| Module | Input | Output Reference |\n| ------- | ---------------------------------- | ---------------------------------------------- |\n| Content | `db:product` (user message) | `{source:\"user\", weight:1.0, type:\"db\", ...}` |\n| Content | `kb:docs` (user message) | `{source:\"user\", weight:1.0, type:\"kb\", ...}` |\n| Hook | `ctx.search.Web(query)` | `{source:\"hook\", weight:0.8, type:\"web\", ...}` |\n| Hook | `ctx.search.KB(query)` | `{source:\"hook\", weight:0.8, type:\"kb\", ...}` |\n| Hook | `ctx.search.DB(query)` | `{source:\"hook\", weight:0.8, type:\"db\", ...}` |\n| Auto | Assistant config `search.web=true` | `{source:\"auto\", weight:0.6, type:\"web\", ...}` |\n| Auto | Assistant config `search.kb=true` | `{source:\"auto\", weight:0.6, type:\"kb\", ...}` |\n\n### Processing Flow\n\n```\nStream()\n │\n ├── 1. Collect search results from all sources\n │ ├── User DataContent → source=\"user\", weight=1.0\n │ ├── Hook ctx.search.*() → source=\"hook\", weight=0.8\n │ └── Auto search → source=\"auto\", weight=0.6\n │\n ├── 2. Merge, deduplicate, rerank by (score * weight)\n │\n ├── 3. Build ... format\n │\n └── 4. Inject references into messages for LLM\n```\n\n**Behavior Rules:**\n\n1. **User data sufficient**: If user provides enough data (≥ skip_threshold), skip auto search\n2. **Deduplication**: Same record from different sources → keep highest weight version\n3. **Final ranking**: Sort by `score * weight` after reranking\n\n**Configuration:**\n\nGlobal defaults (`agent/search.yml`):\n\n```yaml\nweights:\n user: 1.0 # User-provided DataContent\n hook: 0.8 # Hook ctx.search.*() results\n auto: 0.6 # Auto search results\noptions:\n skip_threshold: 5 # Skip auto search if user provides >= N results\n```\n\nAssistant-level override (`assistants//package.yao`):\n\n```jsonc\n{\n \"search\": {\n \"weights\": {\n \"user\": 1.0,\n \"hook\": 0.9, // Higher weight for hook results\n \"auto\": 0.5 // Lower weight for auto results\n },\n \"options\": {\n \"skip_threshold\": 10 // Need more user results to skip auto search\n }\n }\n}\n```\n\n**System Auto-Processing:**\n\nThe weighting and context building is handled automatically by the system:\n\n```\nStream()\n │\n ├── 1. Parse user message for DataContent sources\n │ └── If found → Mark as source=\"user\", weight=1.0\n │\n ├── 2. Create Hook (optional)\n │ └── If hook calls ctx.search.*() → Mark as priority=2, weight=0.8\n │\n ├── 3. Auto Search Decision\n │ ├── Count user-provided results\n │ ├── IF user_results >= skip_auto_if_user_results → SKIP auto search\n │ └── ELSE → Execute auto search with priority=3, weight=0.6\n │\n ├── 4. Merge & Rerank (automatic)\n │ ├── Collect all results with their weights\n │ ├── Deduplicate (keep highest priority)\n │ └── Calculate finalScore = baseScore * weight\n │\n └── 5. Inject to LLM context\n```\n\nUsers don't need to handle weights in hooks - the system manages this automatically.\n\n### Processing Flow in content.Vision()\n\n```\ncontent.Vision()\n ├── type=\"text\" → Pass through\n ├── type=\"image_url\" → Image processing\n ├── type=\"file\" → File processing\n └── type=\"data\" → processDataContent()\n ├── DataSourceModel → Query via model.Find() → Format as text\n ├── DataSourceKBCollection → search.KB() → Format as text\n ├── DataSourceKBDocument → Retrieve document → Format as text\n └── DataSourceMCPResource → MCP resource read → Format as text\n```\n\n### Implementation Location\n\nThe `processDataContent()` function in `content/content.go` should:\n\n1. **For `model` type**: Call search module's DB handler or direct model query\n2. **For `kb_collection` type**: Call search module's KB handler\n3. **For `kb_document` type**: Retrieve specific document from KB\n4. **For `mcp_resource` type**: Read MCP resource\n\nThis allows the search module to be reused for both:\n\n- **Auto Search**: Triggered when `Uses.Search != \"disabled\"`\n- **Data ContentPart**: User explicitly references data sources in message\n\n## Related Files\n\n### Internal Dependencies\n\n- `agent/search/types/` - All type definitions (no circular dependencies)\n- `agent/search/interfaces/` - All interface definitions\n- `agent/search/defaults/` - System default configuration values\n- `agent/search/handlers/` - Handler implementations (web, kb, db)\n- `agent/search/rerank/` - Reranker implementations\n- `agent/search/nlp/` - NLP implementations (keyword, querydsl)\n\n### External Dependencies\n\n- `agent/context/jsapi.go` - JSAPI base implementation\n- `agent/context/types.go` - DataSource, DataContent types\n- `agent/context/types_llm.go` - Uses configuration (Search field)\n- `agent/assistant/types.go` - SearchOption definition\n- `agent/store/types/types.go` - KnowledgeBase, Database config\n- `agent/output/message/types.go` - Output message types\n- `agent/content/content.go` - Content processing (Vision function)\n- `model/model.go` - Yao Model loading (global, system, assistant models)\n\n## See Also\n\n- `agent/context/JSAPI.md` - Full JSAPI documentation\n- `agent/context/RESOURCE_MANAGEMENT.md` - Context lifecycle and resource management\n- `agent/output/README.md` - Output system documentation\n- `agent/store/CHAT_STORAGE_DESIGN.md` - Chat storage design\n" }, { "path": "agent/search/citation.go", "content": "package search\n\nimport (\n\t\"sync/atomic\"\n)\n\n// CitationGenerator generates unique citation IDs (1-based integers)\n// Thread-safe for concurrent use within a single request\ntype CitationGenerator struct {\n\tcounter uint64\n}\n\n// NewCitationGenerator creates a new citation generator\nfunc NewCitationGenerator() *CitationGenerator {\n\treturn &CitationGenerator{}\n}\n\n// Next generates the next citation ID (1, 2, 3, ...)\nfunc (g *CitationGenerator) Next() string {\n\tn := atomic.AddUint64(&g.counter, 1)\n\treturn uint64ToString(n)\n}\n\n// NextInt generates the next citation ID as integer\nfunc (g *CitationGenerator) NextInt() int {\n\treturn int(atomic.AddUint64(&g.counter, 1))\n}\n\n// Current returns the current counter value without incrementing\nfunc (g *CitationGenerator) Current() int {\n\treturn int(atomic.LoadUint64(&g.counter))\n}\n\n// Reset resets the counter (for testing)\nfunc (g *CitationGenerator) Reset() {\n\tatomic.StoreUint64(&g.counter, 0)\n}\n\n// uint64ToString converts uint64 to string without fmt package\nfunc uint64ToString(n uint64) string {\n\tif n == 0 {\n\t\treturn \"0\"\n\t}\n\tvar buf [20]byte // max uint64 is 20 digits\n\ti := len(buf)\n\tfor n > 0 {\n\t\ti--\n\t\tbuf[i] = byte('0' + n%10)\n\t\tn /= 10\n\t}\n\treturn string(buf[i:])\n}\n" }, { "path": "agent/search/citation_test.go", "content": "package search\n\nimport (\n\t\"sync\"\n\t\"testing\"\n\n\t\"github.com/stretchr/testify/assert\"\n)\n\nfunc TestCitationGenerator_Next(t *testing.T) {\n\tgen := NewCitationGenerator()\n\n\t// First ID should be \"1\"\n\tid1 := gen.Next()\n\tassert.Equal(t, \"1\", id1)\n\n\t// Second ID should be \"2\"\n\tid2 := gen.Next()\n\tassert.Equal(t, \"2\", id2)\n\n\t// Third ID should be \"3\"\n\tid3 := gen.Next()\n\tassert.Equal(t, \"3\", id3)\n}\n\nfunc TestCitationGenerator_NextInt(t *testing.T) {\n\tgen := NewCitationGenerator()\n\n\t// First ID should be 1\n\tid1 := gen.NextInt()\n\tassert.Equal(t, 1, id1)\n\n\t// Second ID should be 2\n\tid2 := gen.NextInt()\n\tassert.Equal(t, 2, id2)\n}\n\nfunc TestCitationGenerator_Current(t *testing.T) {\n\tgen := NewCitationGenerator()\n\n\t// Initial should be 0\n\tassert.Equal(t, 0, gen.Current())\n\n\t// After one Next, should be 1\n\tgen.Next()\n\tassert.Equal(t, 1, gen.Current())\n\n\t// Current doesn't increment\n\tassert.Equal(t, 1, gen.Current())\n}\n\nfunc TestCitationGenerator_Reset(t *testing.T) {\n\tgen := NewCitationGenerator()\n\n\t// Generate some IDs\n\tgen.Next()\n\tgen.Next()\n\tgen.Next()\n\n\t// Reset\n\tgen.Reset()\n\n\t// Next ID should be \"1\" again\n\tid := gen.Next()\n\tassert.Equal(t, \"1\", id)\n}\n\nfunc TestCitationGenerator_LargeNumbers(t *testing.T) {\n\tgen := NewCitationGenerator()\n\n\t// Generate 999 IDs\n\tfor i := 0; i < 999; i++ {\n\t\tgen.Next()\n\t}\n\n\t// 1000th ID should be \"1000\"\n\tid := gen.Next()\n\tassert.Equal(t, \"1000\", id)\n}\n\nfunc TestCitationGenerator_Concurrent(t *testing.T) {\n\tgen := NewCitationGenerator()\n\n\t// Run 100 goroutines, each generating 10 IDs\n\tvar wg sync.WaitGroup\n\tids := make(chan string, 1000)\n\n\tfor i := 0; i < 100; i++ {\n\t\twg.Add(1)\n\t\tgo func() {\n\t\t\tdefer wg.Done()\n\t\t\tfor j := 0; j < 10; j++ {\n\t\t\t\tids <- gen.Next()\n\t\t\t}\n\t\t}()\n\t}\n\n\twg.Wait()\n\tclose(ids)\n\n\t// Collect all IDs\n\tidSet := make(map[string]bool)\n\tfor id := range ids {\n\t\tidSet[id] = true\n\t}\n\n\t// All 1000 IDs should be unique\n\tassert.Equal(t, 1000, len(idSet))\n}\n\nfunc TestNewCitationGenerator(t *testing.T) {\n\tgen := NewCitationGenerator()\n\tassert.NotNil(t, gen)\n}\n\nfunc TestUint64ToString(t *testing.T) {\n\ttests := []struct {\n\t\tinput uint64\n\t\texpected string\n\t}{\n\t\t{0, \"0\"},\n\t\t{1, \"1\"},\n\t\t{10, \"10\"},\n\t\t{100, \"100\"},\n\t\t{999, \"999\"},\n\t\t{1000, \"1000\"},\n\t\t{18446744073709551615, \"18446744073709551615\"}, // max uint64\n\t}\n\n\tfor _, tt := range tests {\n\t\tresult := uint64ToString(tt.input)\n\t\tassert.Equal(t, tt.expected, result, \"uint64ToString(%d)\", tt.input)\n\t}\n}\n" }, { "path": "agent/search/defaults/defaults.go", "content": "package defaults\n\nimport \"github.com/yaoapp/yao/agent/search/types\"\n\n// SystemDefaults provides hardcoded default values\n// Used by agent/load.go for merging with agent/search.yao\nvar SystemDefaults = &types.Config{\n\t// Web search defaults\n\tWeb: &types.WebConfig{\n\t\tProvider: \"tavily\",\n\t\tMaxResults: 10,\n\t},\n\n\t// KB search defaults\n\tKB: &types.KBConfig{\n\t\tThreshold: 0.7,\n\t\tGraph: false,\n\t},\n\n\t// DB search defaults\n\tDB: &types.DBConfig{\n\t\tMaxResults: 20,\n\t},\n\n\t// Keyword extraction options (uses.keyword)\n\tKeyword: &types.KeywordConfig{\n\t\tMaxKeywords: 10,\n\t\tLanguage: \"auto\",\n\t},\n\n\t// QueryDSL generation options (uses.querydsl)\n\tQueryDSL: &types.QueryDSLConfig{\n\t\tStrict: false,\n\t},\n\n\t// Rerank options (uses.rerank)\n\tRerank: &types.RerankConfig{\n\t\tTopN: 10,\n\t},\n\n\t// Citation\n\tCitation: &types.CitationConfig{\n\t\tFormat: \"#ref:{id}\",\n\t\tAutoInjectPrompt: true,\n\t},\n\n\t// Source weights\n\tWeights: &types.WeightsConfig{\n\t\tUser: 1.0,\n\t\tHook: 0.8,\n\t\tAuto: 0.6,\n\t},\n\n\t// Behavior options\n\tOptions: &types.OptionsConfig{\n\t\tSkipThreshold: 5,\n\t},\n}\n\n// GetWeight returns the weight for a source type using default config\nfunc GetWeight(source types.SourceType) float64 {\n\treturn SystemDefaults.GetWeight(source)\n}\n" }, { "path": "agent/search/handlers/db/handler.go", "content": "package db\n\nimport (\n\t\"encoding/json\"\n\t\"fmt\"\n\t\"time\"\n\n\t\"github.com/yaoapp/gou/model\"\n\t\"github.com/yaoapp/gou/query\"\n\t\"github.com/yaoapp/gou/query/gou\"\n\tagentContext \"github.com/yaoapp/yao/agent/context\"\n\t\"github.com/yaoapp/yao/agent/search/nlp/querydsl\"\n\t\"github.com/yaoapp/yao/agent/search/types\"\n)\n\n// Handler implements DB search\ntype Handler struct {\n\tusesQueryDSL string // \"builtin\", \"\", \"mcp:.\"\n\tconfig *types.DBConfig // DB search configuration\n}\n\n// NewHandler creates a new DB search handler\nfunc NewHandler(usesQueryDSL string, cfg *types.DBConfig) *Handler {\n\treturn &Handler{usesQueryDSL: usesQueryDSL, config: cfg}\n}\n\n// Type returns the search type this handler supports\nfunc (h *Handler) Type() types.SearchType {\n\treturn types.SearchTypeDB\n}\n\n// Search converts NL to QueryDSL and executes\n// Note: This method doesn't have context, use SearchWithContext for full functionality\nfunc (h *Handler) Search(req *types.Request) (*types.Result, error) {\n\treturn h.SearchWithContext(nil, req)\n}\n\n// SearchWithContext executes DB search with context (required for QueryDSL generation)\nfunc (h *Handler) SearchWithContext(ctx *agentContext.Context, req *types.Request) (*types.Result, error) {\n\tstart := time.Now()\n\n\t// Validate request\n\tif req.Query == \"\" {\n\t\treturn &types.Result{\n\t\t\tType: types.SearchTypeDB,\n\t\t\tQuery: req.Query,\n\t\t\tSource: req.Source,\n\t\t\tItems: []*types.ResultItem{},\n\t\t\tTotal: 0,\n\t\t\tDuration: time.Since(start).Milliseconds(),\n\t\t\tError: \"query is required\",\n\t\t}, nil\n\t}\n\n\t// Get models from request or config\n\tmodelIDs := req.Models\n\tif len(modelIDs) == 0 && h.config != nil {\n\t\tmodelIDs = h.config.Models\n\t}\n\n\t// If no models specified, return empty result\n\tif len(modelIDs) == 0 {\n\t\treturn &types.Result{\n\t\t\tType: types.SearchTypeDB,\n\t\t\tQuery: req.Query,\n\t\t\tSource: req.Source,\n\t\t\tItems: []*types.ResultItem{},\n\t\t\tTotal: 0,\n\t\t\tDuration: time.Since(start).Milliseconds(),\n\t\t\tError: \"no models specified\",\n\t\t}, nil\n\t}\n\n\t// Get max results\n\tmaxResults := req.Limit\n\tif maxResults == 0 && h.config != nil && h.config.MaxResults > 0 {\n\t\tmaxResults = h.config.MaxResults\n\t}\n\tif maxResults == 0 {\n\t\tmaxResults = 20 // default\n\t}\n\n\t// Context is required for QueryDSL generation\n\tif ctx == nil {\n\t\treturn &types.Result{\n\t\t\tType: types.SearchTypeDB,\n\t\t\tQuery: req.Query,\n\t\t\tSource: req.Source,\n\t\t\tItems: []*types.ResultItem{},\n\t\t\tTotal: 0,\n\t\t\tDuration: time.Since(start).Milliseconds(),\n\t\t\tError: \"context is required for DB search\",\n\t\t}, nil\n\t}\n\n\t// 1. Load all models and build combined schema\n\tmodels := make(map[string]*model.Model)\n\tschemas := make([]map[string]interface{}, 0, len(modelIDs))\n\n\tfor _, modelID := range modelIDs {\n\t\tmod, err := model.Get(modelID)\n\t\tif err != nil {\n\t\t\tcontinue // Skip non-existent models\n\t\t}\n\t\tmodels[modelID] = mod\n\t\tschemas = append(schemas, h.buildModelSchema(mod))\n\t}\n\n\tif len(schemas) == 0 {\n\t\treturn &types.Result{\n\t\t\tType: types.SearchTypeDB,\n\t\t\tQuery: req.Query,\n\t\t\tSource: req.Source,\n\t\t\tItems: []*types.ResultItem{},\n\t\t\tTotal: 0,\n\t\t\tDuration: time.Since(start).Milliseconds(),\n\t\t\tError: \"no valid models found\",\n\t\t}, nil\n\t}\n\n\t// 2. Generate QueryDSL with all schemas\n\tgenerator := querydsl.NewGenerator(h.usesQueryDSL, nil)\n\tinput := &querydsl.Input{\n\t\tQuery: req.Query,\n\t\tModelIDs: modelIDs,\n\t\tScenario: req.Scenario, // Pass scenario: filter, aggregation, join, complex\n\t\tLimit: maxResults,\n\t}\n\n\t// Build schema input: single schema or array of schemas\n\tvar schemaInput interface{}\n\tif len(schemas) == 1 {\n\t\tschemaInput = schemas[0]\n\t} else {\n\t\tschemaInput = schemas\n\t}\n\n\tinput.ExtraParams = map[string]interface{}{\n\t\t\"schema\": schemaInput,\n\t}\n\n\tresult, err := generator.Generate(ctx, input)\n\tif err != nil {\n\t\treturn &types.Result{\n\t\t\tType: types.SearchTypeDB,\n\t\t\tQuery: req.Query,\n\t\t\tSource: req.Source,\n\t\t\tItems: []*types.ResultItem{},\n\t\t\tTotal: 0,\n\t\t\tDuration: time.Since(start).Milliseconds(),\n\t\t\tError: fmt.Sprintf(\"QueryDSL generation failed: %v\", err),\n\t\t}, nil\n\t}\n\n\tif result == nil || result.DSL == nil {\n\t\treturn &types.Result{\n\t\t\tType: types.SearchTypeDB,\n\t\t\tQuery: req.Query,\n\t\t\tSource: req.Source,\n\t\t\tItems: []*types.ResultItem{},\n\t\t\tTotal: 0,\n\t\t\tDuration: time.Since(start).Milliseconds(),\n\t\t\tError: \"no QueryDSL generated\",\n\t\t}, nil\n\t}\n\n\t// 3. Sanitize generated DSL (remove unsupported wildcards like \"*\")\n\th.sanitizeDSL(result.DSL)\n\n\t// 4. Merge preset conditions into generated DSL\n\th.mergeDSLConditions(result.DSL, req)\n\n\t// 5. Execute QueryDSL using gou query engine\n\trecords, err := h.executeDSL(result.DSL)\n\tif err != nil {\n\t\treturn &types.Result{\n\t\t\tType: types.SearchTypeDB,\n\t\t\tQuery: req.Query,\n\t\t\tSource: req.Source,\n\t\t\tItems: []*types.ResultItem{},\n\t\t\tTotal: 0,\n\t\t\tDuration: time.Since(start).Milliseconds(),\n\t\t\tError: fmt.Sprintf(\"query execution failed: %v\", err),\n\t\t}, nil\n\t}\n\n\t// 6. Determine the primary model for result formatting\n\t// Use the \"from\" table from DSL, or first model\n\tprimaryModelID := modelIDs[0]\n\tif result.DSL.From != nil && result.DSL.From.Name != \"\" {\n\t\t// Find model by table name\n\t\tfor id, mod := range models {\n\t\t\tif mod.MetaData.Table.Name == result.DSL.From.Name {\n\t\t\t\tprimaryModelID = id\n\t\t\t\tbreak\n\t\t\t}\n\t\t}\n\t}\n\n\tprimaryModel := models[primaryModelID]\n\tif primaryModel == nil {\n\t\tprimaryModel, _ = model.Get(primaryModelID) // May be nil, that's ok\n\t}\n\n\t// 7. Convert records to ResultItems\n\titems := h.convertToResultItems(records, primaryModelID, primaryModel, req.Source)\n\n\t// Apply limit\n\tif len(items) > maxResults {\n\t\titems = items[:maxResults]\n\t}\n\n\t// 8. Convert DSL to map for storage\n\tdslMap := h.dslToMap(result.DSL)\n\n\treturn &types.Result{\n\t\tType: types.SearchTypeDB,\n\t\tQuery: req.Query,\n\t\tSource: req.Source,\n\t\tItems: items,\n\t\tTotal: len(items),\n\t\tDuration: time.Since(start).Milliseconds(),\n\t\tDSL: dslMap,\n\t}, nil\n}\n\n// mergeDSLConditions merges preset conditions from request into generated DSL\nfunc (h *Handler) mergeDSLConditions(dsl *gou.QueryDSL, req *types.Request) {\n\tif dsl == nil {\n\t\treturn\n\t}\n\n\t// Merge preset Wheres (prepend to ensure they take priority)\n\tif len(req.Wheres) > 0 {\n\t\tdsl.Wheres = append(req.Wheres, dsl.Wheres...)\n\t}\n\n\t// Merge preset Orders (prepend to ensure they take priority)\n\tif len(req.Orders) > 0 {\n\t\tdsl.Orders = append(req.Orders, dsl.Orders...)\n\t}\n\n\t// Merge preset Select fields\n\tif len(req.Select) > 0 {\n\t\t// Convert string fields to Expression\n\t\tselectExprs := make([]gou.Expression, 0, len(req.Select))\n\t\tfor _, field := range req.Select {\n\t\t\tselectExprs = append(selectExprs, gou.Expression{Field: field})\n\t\t}\n\t\t// If DSL has no select, use preset; otherwise merge\n\t\tif len(dsl.Select) == 0 {\n\t\t\tdsl.Select = selectExprs\n\t\t} else {\n\t\t\t// Prepend preset fields\n\t\t\tdsl.Select = append(selectExprs, dsl.Select...)\n\t\t}\n\t}\n\n\t// Ensure limit is set\n\tif dsl.Limit == 0 && req.Limit > 0 {\n\t\tdsl.Limit = req.Limit\n\t}\n}\n\n// buildModelSchema builds a simplified schema for QueryDSL generator\nfunc (h *Handler) buildModelSchema(mod *model.Model) map[string]interface{} {\n\tcolumns := make([]map[string]interface{}, 0, len(mod.Columns))\n\tfor _, col := range mod.Columns {\n\t\tcolInfo := map[string]interface{}{\n\t\t\t\"name\": col.Name,\n\t\t\t\"type\": col.Type,\n\t\t}\n\t\tif col.Label != \"\" {\n\t\t\tcolInfo[\"label\"] = col.Label\n\t\t}\n\t\tif col.Description != \"\" {\n\t\t\tcolInfo[\"description\"] = col.Description\n\t\t}\n\t\tcolumns = append(columns, colInfo)\n\t}\n\n\treturn map[string]interface{}{\n\t\t\"name\": mod.MetaData.Table.Name,\n\t\t\"columns\": columns,\n\t}\n}\n\n// sanitizeDSL cleans up LLM-generated DSL to remove unsupported constructs.\n// The QueryDSL engine does not support wildcard \"*\" in select fields;\n// an empty select list naturally returns all columns.\nfunc (h *Handler) sanitizeDSL(dsl *gou.QueryDSL) {\n\tif dsl == nil {\n\t\treturn\n\t}\n\n\tif len(dsl.Select) > 0 {\n\t\tcleaned := make([]gou.Expression, 0, len(dsl.Select))\n\t\tfor _, expr := range dsl.Select {\n\t\t\tif expr.Field != \"*\" {\n\t\t\t\tcleaned = append(cleaned, expr)\n\t\t\t}\n\t\t}\n\t\tdsl.Select = cleaned\n\t}\n}\n\n// executeDSL executes the QueryDSL and returns records.\n// Uses recover to convert panics from MustGet into errors.\nfunc (h *Handler) executeDSL(dsl interface{}) (records []map[string]interface{}, err error) {\n\tdefer func() {\n\t\tif r := recover(); r != nil {\n\t\t\terr = fmt.Errorf(\"query execution panic: %v\", r)\n\t\t}\n\t}()\n\n\tengine, err := query.Select(\"default\")\n\tif err != nil {\n\t\treturn nil, fmt.Errorf(\"query engine not found: %w\", err)\n\t}\n\n\tdslJSON, err := json.Marshal(dsl)\n\tif err != nil {\n\t\treturn nil, fmt.Errorf(\"failed to marshal DSL: %w\", err)\n\t}\n\n\tq, err := engine.Load(json.RawMessage(dslJSON))\n\tif err != nil {\n\t\treturn nil, fmt.Errorf(\"failed to load DSL: %w\", err)\n\t}\n\n\trawRecords := q.Get(nil)\n\n\trecords = make([]map[string]interface{}, 0, len(rawRecords))\n\tfor _, rec := range rawRecords {\n\t\trecords = append(records, map[string]interface{}(rec))\n\t}\n\n\treturn records, nil\n}\n\n// convertToResultItems converts query results to ResultItems\nfunc (h *Handler) convertToResultItems(records []map[string]interface{}, modelID string, mod *model.Model, source types.SourceType) []*types.ResultItem {\n\titems := make([]*types.ResultItem, 0, len(records))\n\n\tprimaryKey := \"id\"\n\tif mod != nil && mod.PrimaryKey != \"\" {\n\t\tprimaryKey = mod.PrimaryKey\n\t}\n\n\tfor _, rec := range records {\n\t\titem := &types.ResultItem{\n\t\t\tType: types.SearchTypeDB,\n\t\t\tSource: source,\n\t\t\tModel: modelID,\n\t\t\tData: rec,\n\t\t}\n\n\t\t// Try to extract title from common fields\n\t\titem.Title = h.extractTitle(rec, mod)\n\n\t\t// Try to extract content/description\n\t\titem.Content = h.extractContent(rec, mod)\n\n\t\t// Try to extract record ID\n\t\tif id, ok := rec[primaryKey]; ok {\n\t\t\titem.RecordID = id\n\t\t}\n\n\t\titems = append(items, item)\n\t}\n\n\treturn items\n}\n\n// extractTitle tries to extract a title from the record\nfunc (h *Handler) extractTitle(rec map[string]interface{}, mod *model.Model) string {\n\t// Common title fields\n\ttitleFields := []string{\"title\", \"name\", \"subject\", \"label\"}\n\tfor _, field := range titleFields {\n\t\tif val, ok := rec[field]; ok {\n\t\t\tif str, ok := val.(string); ok && str != \"\" {\n\t\t\t\treturn str\n\t\t\t}\n\t\t}\n\t}\n\treturn \"\"\n}\n\n// extractContent tries to extract content from the record\nfunc (h *Handler) extractContent(rec map[string]interface{}, mod *model.Model) string {\n\t// Common content fields\n\tcontentFields := []string{\"content\", \"description\", \"summary\", \"text\", \"body\"}\n\tfor _, field := range contentFields {\n\t\tif val, ok := rec[field]; ok {\n\t\t\tif str, ok := val.(string); ok && str != \"\" {\n\t\t\t\treturn str\n\t\t\t}\n\t\t}\n\t}\n\n\t// Fallback: serialize first few fields as content\n\tcontent, _ := json.Marshal(rec)\n\tif len(content) > 500 {\n\t\tcontent = content[:500]\n\t}\n\treturn string(content)\n}\n\n// dslToMap converts QueryDSL to map for storage\nfunc (h *Handler) dslToMap(dsl *gou.QueryDSL) map[string]interface{} {\n\tif dsl == nil {\n\t\treturn nil\n\t}\n\n\t// Marshal and unmarshal to get a clean map\n\tdata, err := json.Marshal(dsl)\n\tif err != nil {\n\t\treturn nil\n\t}\n\n\tvar result map[string]interface{}\n\tif err := json.Unmarshal(data, &result); err != nil {\n\t\treturn nil\n\t}\n\n\treturn result\n}\n" }, { "path": "agent/search/handlers/db/handler_integration_test.go", "content": "package db_test\n\nimport (\n\t\"testing\"\n\n\t\"github.com/yaoapp/gou/model\"\n\t\"github.com/yaoapp/gou/query/gou\"\n\t\"github.com/yaoapp/yao/agent/context\"\n\t\"github.com/yaoapp/yao/agent/search/handlers/db\"\n\t\"github.com/yaoapp/yao/agent/search/types\"\n\t\"github.com/yaoapp/yao/agent/testutils\"\n\toauthTypes \"github.com/yaoapp/yao/openapi/oauth/types\"\n\n\t\"github.com/stretchr/testify/assert\"\n\t\"github.com/stretchr/testify/require\"\n)\n\n// ============================================================================\n// Integration Tests - Requires database and models\n// ============================================================================\n\nfunc TestHandler_Search_Integration(t *testing.T) {\n\t// Skip if running short tests\n\tif testing.Short() {\n\t\tt.Skip(\"Skipping integration test\")\n\t}\n\n\t// Initialize test environment (loads models, database, query engine, etc.)\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\t// Create test context\n\tctx := newTestContext(t)\n\n\t// Verify __yao.role model is loaded\n\tmod := model.Select(\"__yao.role\")\n\trequire.NotNil(t, mod, \"__yao.role model should be loaded\")\n\n\tt.Run(\"search_role_model_with_results\", func(t *testing.T) {\n\t\t// First, ensure there's at least one role in the database\n\t\tensureTestRole(t, mod)\n\n\t\t// Create handler with builtin QueryDSL generator\n\t\th := db.NewHandler(\"builtin\", &types.DBConfig{\n\t\t\tModels: []string{\"__yao.role\"},\n\t\t\tMaxResults: 10,\n\t\t})\n\n\t\treq := &types.Request{\n\t\t\tType: types.SearchTypeDB,\n\t\t\tQuery: \"查询所有角色\",\n\t\t\tSource: types.SourceAuto,\n\t\t\tModels: []string{\"__yao.role\"},\n\t\t\tScenario: types.ScenarioFilter,\n\t\t\tLimit: 10,\n\t\t}\n\n\t\tresult, err := h.SearchWithContext(ctx, req)\n\t\trequire.NoError(t, err)\n\t\trequire.NotNil(t, result)\n\n\t\t// Verify result structure\n\t\tassert.Equal(t, types.SearchTypeDB, result.Type)\n\t\tassert.Equal(t, \"查询所有角色\", result.Query)\n\t\tassert.Equal(t, types.SourceAuto, result.Source)\n\t\tassert.GreaterOrEqual(t, result.Duration, int64(0))\n\n\t\t// Should have results\n\t\tif result.Error != \"\" {\n\t\t\tt.Logf(\"Search error: %s\", result.Error)\n\t\t}\n\t\tassert.Empty(t, result.Error, \"Search should not return error\")\n\t\tassert.Greater(t, len(result.Items), 0, \"Should have at least one result\")\n\t\tassert.Equal(t, len(result.Items), result.Total)\n\n\t\t// Verify result items\n\t\tfor _, item := range result.Items {\n\t\t\tassert.Equal(t, types.SearchTypeDB, item.Type)\n\t\t\tassert.Equal(t, types.SourceAuto, item.Source)\n\t\t\tassert.Equal(t, \"__yao.role\", item.Model)\n\t\t\tassert.NotNil(t, item.Data, \"Data should not be nil\")\n\t\t\tassert.NotNil(t, item.RecordID, \"RecordID should not be nil\")\n\t\t}\n\t})\n\n\tt.Run(\"search_with_filter_scenario\", func(t *testing.T) {\n\t\th := db.NewHandler(\"builtin\", nil)\n\n\t\treq := &types.Request{\n\t\t\tType: types.SearchTypeDB,\n\t\t\tQuery: \"查询系统角色\",\n\t\t\tSource: types.SourceHook,\n\t\t\tModels: []string{\"__yao.role\"},\n\t\t\tScenario: types.ScenarioFilter,\n\t\t\tLimit: 5,\n\t\t}\n\n\t\tresult, err := h.SearchWithContext(ctx, req)\n\t\trequire.NoError(t, err)\n\t\trequire.NotNil(t, result)\n\n\t\tassert.Equal(t, types.SearchTypeDB, result.Type)\n\t\tassert.Equal(t, types.SourceHook, result.Source)\n\t\tassert.LessOrEqual(t, len(result.Items), 5, \"Should respect limit\")\n\t})\n\n\tt.Run(\"search_with_preset_wheres\", func(t *testing.T) {\n\t\th := db.NewHandler(\"builtin\", nil)\n\n\t\treq := &types.Request{\n\t\t\tType: types.SearchTypeDB,\n\t\t\tQuery: \"查询角色\",\n\t\t\tSource: types.SourceAuto,\n\t\t\tModels: []string{\"__yao.role\"},\n\t\t\tWheres: []gou.Where{\n\t\t\t\t{Condition: gou.Condition{Field: &gou.Expression{Field: \"is_active\"}, Value: true, OP: \"=\"}},\n\t\t\t},\n\t\t\tLimit: 10,\n\t\t}\n\n\t\tresult, err := h.SearchWithContext(ctx, req)\n\t\trequire.NoError(t, err)\n\t\trequire.NotNil(t, result)\n\n\t\t// All results should have is_active = true (due to preset where)\n\t\tfor _, item := range result.Items {\n\t\t\tif data, ok := item.Data[\"is_active\"]; ok {\n\t\t\t\t// is_active could be bool or int depending on driver\n\t\t\t\tswitch v := data.(type) {\n\t\t\t\tcase bool:\n\t\t\t\t\tassert.True(t, v)\n\t\t\t\tcase int64:\n\t\t\t\t\tassert.Equal(t, int64(1), v)\n\t\t\t\tcase float64:\n\t\t\t\t\tassert.Equal(t, float64(1), v)\n\t\t\t\t}\n\t\t\t}\n\t\t}\n\t})\n\n\tt.Run(\"search_nonexistent_model_graceful\", func(t *testing.T) {\n\t\th := db.NewHandler(\"builtin\", nil)\n\n\t\treq := &types.Request{\n\t\t\tType: types.SearchTypeDB,\n\t\t\tQuery: \"查询文章\",\n\t\t\tSource: types.SourceAuto,\n\t\t\tModels: []string{\"nonexistent_model\", \"article\", \"fake_model\"},\n\t\t\tLimit: 10,\n\t\t}\n\n\t\t// Should NOT panic, should return gracefully with error\n\t\tresult, err := h.SearchWithContext(ctx, req)\n\t\trequire.NoError(t, err)\n\t\trequire.NotNil(t, result)\n\n\t\t// Should have error message about no valid models\n\t\tassert.Equal(t, types.SearchTypeDB, result.Type)\n\t\tassert.Equal(t, \"no valid models found\", result.Error)\n\t\tassert.Empty(t, result.Items)\n\t})\n\n\tt.Run(\"search_mixed_models_partial_exist\", func(t *testing.T) {\n\t\th := db.NewHandler(\"builtin\", nil)\n\n\t\treq := &types.Request{\n\t\t\tType: types.SearchTypeDB,\n\t\t\tQuery: \"查询角色\",\n\t\t\tSource: types.SourceAuto,\n\t\t\tModels: []string{\"nonexistent_model\", \"__yao.role\", \"fake_model\"}, // Only __yao.role exists\n\t\t\tLimit: 10,\n\t\t}\n\n\t\t// Should NOT panic, should work with the existing model\n\t\tresult, err := h.SearchWithContext(ctx, req)\n\t\trequire.NoError(t, err)\n\t\trequire.NotNil(t, result)\n\n\t\t// Should succeed with partial models\n\t\tassert.Equal(t, types.SearchTypeDB, result.Type)\n\t\tif result.Error == \"\" {\n\t\t\t// If no error, should have results from __yao.role\n\t\t\tassert.GreaterOrEqual(t, len(result.Items), 0)\n\t\t}\n\t})\n}\n\n// newTestContext creates a test context with required fields\nfunc newTestContext(t *testing.T) *context.Context {\n\tt.Helper()\n\tauthorized := &oauthTypes.AuthorizedInfo{\n\t\tUserID: \"test-user\",\n\t}\n\tchatID := \"test-chat-db-search\"\n\tctx := context.New(t.Context(), authorized, chatID)\n\treturn ctx\n}\n\n// ensureTestRole ensures there's at least one role in the database for testing\nfunc ensureTestRole(t *testing.T, mod *model.Model) {\n\tt.Helper()\n\n\t// Try to find existing roles\n\trows, err := mod.Get(model.QueryParam{Limit: 1})\n\tif err == nil && len(rows) > 0 {\n\t\treturn // Already have roles\n\t}\n\n\t// Create a test role\n\t_, err = mod.Create(map[string]interface{}{\n\t\t\"role_id\": \"test_role\",\n\t\t\"name\": \"Test Role\",\n\t\t\"description\": \"A test role for unit testing\",\n\t\t\"is_active\": true,\n\t\t\"is_system\": false,\n\t\t\"level\": 1,\n\t})\n\tif err != nil {\n\t\tt.Logf(\"Note: Could not create test role: %v\", err)\n\t}\n}\n" }, { "path": "agent/search/handlers/db/handler_test.go", "content": "package db\n\nimport (\n\t\"testing\"\n\n\t\"github.com/stretchr/testify/assert\"\n\t\"github.com/yaoapp/gou/model\"\n\t\"github.com/yaoapp/gou/query/gou\"\n\t\"github.com/yaoapp/yao/agent/search/types\"\n)\n\nfunc TestNewHandler(t *testing.T) {\n\tt.Run(\"with nil config\", func(t *testing.T) {\n\t\th := NewHandler(\"builtin\", nil)\n\t\tassert.NotNil(t, h)\n\t\tassert.Equal(t, \"builtin\", h.usesQueryDSL)\n\t\tassert.Nil(t, h.config)\n\t})\n\n\tt.Run(\"with config\", func(t *testing.T) {\n\t\tcfg := &types.DBConfig{\n\t\t\tModels: []string{\"product\", \"order\"},\n\t\t\tMaxResults: 50,\n\t\t}\n\t\th := NewHandler(\"workers.nlp.querydsl\", cfg)\n\t\tassert.NotNil(t, h)\n\t\tassert.Equal(t, \"workers.nlp.querydsl\", h.usesQueryDSL)\n\t\tassert.Equal(t, cfg, h.config)\n\t})\n\n\tt.Run(\"with mcp mode\", func(t *testing.T) {\n\t\th := NewHandler(\"mcp:nlp.generate_querydsl\", nil)\n\t\tassert.NotNil(t, h)\n\t\tassert.Equal(t, \"mcp:nlp.generate_querydsl\", h.usesQueryDSL)\n\t})\n}\n\nfunc TestHandler_Type(t *testing.T) {\n\th := NewHandler(\"builtin\", nil)\n\tassert.Equal(t, types.SearchTypeDB, h.Type())\n}\n\nfunc TestHandler_Search_Validation(t *testing.T) {\n\ttests := []struct {\n\t\tname string\n\t\tusesQueryDSL string\n\t\tconfig *types.DBConfig\n\t\treq *types.Request\n\t\texpectError string\n\t}{\n\t\t{\n\t\t\tname: \"empty query\",\n\t\t\tusesQueryDSL: \"builtin\",\n\t\t\tconfig: nil,\n\t\t\treq: &types.Request{\n\t\t\t\tType: types.SearchTypeDB,\n\t\t\t\tQuery: \"\",\n\t\t\t},\n\t\t\texpectError: \"query is required\",\n\t\t},\n\t\t{\n\t\t\tname: \"no models in request or config\",\n\t\t\tusesQueryDSL: \"builtin\",\n\t\t\tconfig: nil,\n\t\t\treq: &types.Request{\n\t\t\t\tType: types.SearchTypeDB,\n\t\t\t\tQuery: \"find products under $100\",\n\t\t\t},\n\t\t\texpectError: \"no models specified\",\n\t\t},\n\t\t{\n\t\t\tname: \"context required for DB search\",\n\t\t\tusesQueryDSL: \"builtin\",\n\t\t\tconfig: &types.DBConfig{\n\t\t\t\tModels: []string{\"product\"},\n\t\t\t},\n\t\t\treq: &types.Request{\n\t\t\t\tType: types.SearchTypeDB,\n\t\t\t\tQuery: \"find products under $100\",\n\t\t\t\tModels: []string{\"product\"},\n\t\t\t},\n\t\t\texpectError: \"context is required for DB search\",\n\t\t},\n\t}\n\n\tfor _, tt := range tests {\n\t\tt.Run(tt.name, func(t *testing.T) {\n\t\t\th := NewHandler(tt.usesQueryDSL, tt.config)\n\t\t\tresult, err := h.Search(tt.req)\n\n\t\t\tassert.NoError(t, err)\n\t\t\tassert.NotNil(t, result)\n\t\t\tassert.Equal(t, types.SearchTypeDB, result.Type)\n\t\t\tassert.Equal(t, tt.expectError, result.Error)\n\t\t\tassert.Equal(t, 0, len(result.Items))\n\t\t\tassert.GreaterOrEqual(t, result.Duration, int64(0))\n\t\t})\n\t}\n}\n\nfunc TestHandler_Search_SourcePreserved(t *testing.T) {\n\th := NewHandler(\"builtin\", &types.DBConfig{Models: []string{\"product\"}})\n\n\tsources := []types.SourceType{types.SourceUser, types.SourceHook, types.SourceAuto}\n\tfor _, source := range sources {\n\t\treq := &types.Request{\n\t\t\tType: types.SearchTypeDB,\n\t\t\tQuery: \"test\",\n\t\t\tSource: source,\n\t\t\tModels: []string{\"product\"},\n\t\t}\n\t\tresult, err := h.Search(req)\n\t\tassert.NoError(t, err)\n\t\tassert.Equal(t, source, result.Source)\n\t}\n}\n\nfunc TestHandler_BuildModelSchema(t *testing.T) {\n\th := NewHandler(\"builtin\", nil)\n\n\t// Create a mock model for testing\n\tmod := &model.Model{\n\t\tMetaData: model.MetaData{\n\t\t\tTable: model.Table{\n\t\t\t\tName: \"test_products\",\n\t\t\t},\n\t\t},\n\t\tColumns: map[string]*model.Column{\n\t\t\t\"id\": {\n\t\t\t\tName: \"id\",\n\t\t\t\tType: \"ID\",\n\t\t\t\tLabel: \"ID\",\n\t\t\t},\n\t\t\t\"name\": {\n\t\t\t\tName: \"name\",\n\t\t\t\tType: \"string\",\n\t\t\t\tLabel: \"Name\",\n\t\t\t\tDescription: \"Product name\",\n\t\t\t},\n\t\t\t\"price\": {\n\t\t\t\tName: \"price\",\n\t\t\t\tType: \"decimal\",\n\t\t\t\tLabel: \"Price\",\n\t\t\t},\n\t\t},\n\t}\n\n\tschema := h.buildModelSchema(mod)\n\n\tassert.NotNil(t, schema)\n\tassert.Equal(t, \"test_products\", schema[\"name\"])\n\n\tcolumns, ok := schema[\"columns\"].([]map[string]interface{})\n\tassert.True(t, ok)\n\tassert.Len(t, columns, 3)\n\n\t// Verify columns have required fields\n\tfor _, col := range columns {\n\t\tassert.NotEmpty(t, col[\"name\"])\n\t\tassert.NotEmpty(t, col[\"type\"])\n\t}\n}\n\nfunc TestHandler_BuildModelSchema_MultipleModels(t *testing.T) {\n\th := NewHandler(\"builtin\", nil)\n\n\t// Create mock models for testing joins\n\tproductMod := &model.Model{\n\t\tMetaData: model.MetaData{\n\t\t\tTable: model.Table{Name: \"products\"},\n\t\t},\n\t\tColumns: map[string]*model.Column{\n\t\t\t\"id\": {Name: \"id\", Type: \"ID\"},\n\t\t\t\"name\": {Name: \"name\", Type: \"string\"},\n\t\t\t\"category_id\": {Name: \"category_id\", Type: \"integer\"},\n\t\t},\n\t}\n\n\tcategoryMod := &model.Model{\n\t\tMetaData: model.MetaData{\n\t\t\tTable: model.Table{Name: \"categories\"},\n\t\t},\n\t\tColumns: map[string]*model.Column{\n\t\t\t\"id\": {Name: \"id\", Type: \"ID\"},\n\t\t\t\"name\": {Name: \"name\", Type: \"string\"},\n\t\t},\n\t}\n\n\tproductSchema := h.buildModelSchema(productMod)\n\tcategorySchema := h.buildModelSchema(categoryMod)\n\n\tassert.Equal(t, \"products\", productSchema[\"name\"])\n\tassert.Equal(t, \"categories\", categorySchema[\"name\"])\n\n\t// Verify both schemas can be combined into an array\n\tschemas := []map[string]interface{}{productSchema, categorySchema}\n\tassert.Len(t, schemas, 2)\n}\n\nfunc TestHandler_ExtractTitle(t *testing.T) {\n\th := NewHandler(\"builtin\", nil)\n\tmod := &model.Model{}\n\n\ttests := []struct {\n\t\tname string\n\t\trecord map[string]interface{}\n\t\texpected string\n\t}{\n\t\t{\n\t\t\tname: \"title field\",\n\t\t\trecord: map[string]interface{}{\"title\": \"Test Title\", \"id\": 1},\n\t\t\texpected: \"Test Title\",\n\t\t},\n\t\t{\n\t\t\tname: \"name field\",\n\t\t\trecord: map[string]interface{}{\"name\": \"Test Name\", \"id\": 1},\n\t\t\texpected: \"Test Name\",\n\t\t},\n\t\t{\n\t\t\tname: \"subject field\",\n\t\t\trecord: map[string]interface{}{\"subject\": \"Test Subject\", \"id\": 1},\n\t\t\texpected: \"Test Subject\",\n\t\t},\n\t\t{\n\t\t\tname: \"label field\",\n\t\t\trecord: map[string]interface{}{\"label\": \"Test Label\", \"id\": 1},\n\t\t\texpected: \"Test Label\",\n\t\t},\n\t\t{\n\t\t\tname: \"no title field\",\n\t\t\trecord: map[string]interface{}{\"id\": 1, \"price\": 100},\n\t\t\texpected: \"\",\n\t\t},\n\t\t{\n\t\t\tname: \"empty title\",\n\t\t\trecord: map[string]interface{}{\"title\": \"\", \"name\": \"Fallback\"},\n\t\t\texpected: \"Fallback\",\n\t\t},\n\t}\n\n\tfor _, tt := range tests {\n\t\tt.Run(tt.name, func(t *testing.T) {\n\t\t\ttitle := h.extractTitle(tt.record, mod)\n\t\t\tassert.Equal(t, tt.expected, title)\n\t\t})\n\t}\n}\n\nfunc TestHandler_ExtractContent(t *testing.T) {\n\th := NewHandler(\"builtin\", nil)\n\tmod := &model.Model{}\n\n\ttests := []struct {\n\t\tname string\n\t\trecord map[string]interface{}\n\t\texpectEmpty bool\n\t}{\n\t\t{\n\t\t\tname: \"content field\",\n\t\t\trecord: map[string]interface{}{\"content\": \"Test Content\"},\n\t\t\texpectEmpty: false,\n\t\t},\n\t\t{\n\t\t\tname: \"description field\",\n\t\t\trecord: map[string]interface{}{\"description\": \"Test Description\"},\n\t\t\texpectEmpty: false,\n\t\t},\n\t\t{\n\t\t\tname: \"summary field\",\n\t\t\trecord: map[string]interface{}{\"summary\": \"Test Summary\"},\n\t\t\texpectEmpty: false,\n\t\t},\n\t\t{\n\t\t\tname: \"fallback to JSON\",\n\t\t\trecord: map[string]interface{}{\"id\": 1, \"price\": 100},\n\t\t\texpectEmpty: false, // Should return JSON representation\n\t\t},\n\t}\n\n\tfor _, tt := range tests {\n\t\tt.Run(tt.name, func(t *testing.T) {\n\t\t\tcontent := h.extractContent(tt.record, mod)\n\t\t\tif tt.expectEmpty {\n\t\t\t\tassert.Empty(t, content)\n\t\t\t} else {\n\t\t\t\tassert.NotEmpty(t, content)\n\t\t\t}\n\t\t})\n\t}\n}\n\nfunc TestHandler_ConvertToResultItems(t *testing.T) {\n\th := NewHandler(\"builtin\", nil)\n\tmod := &model.Model{\n\t\tPrimaryKey: \"id\",\n\t}\n\n\trecords := []map[string]interface{}{\n\t\t{\n\t\t\t\"id\": 1,\n\t\t\t\"name\": \"Product 1\",\n\t\t\t\"description\": \"Description 1\",\n\t\t\t\"price\": 99.99,\n\t\t},\n\t\t{\n\t\t\t\"id\": 2,\n\t\t\t\"title\": \"Product 2\",\n\t\t\t\"content\": \"Content 2\",\n\t\t},\n\t}\n\n\titems := h.convertToResultItems(records, \"product\", mod, types.SourceAuto)\n\n\tassert.Len(t, items, 2)\n\n\t// First item\n\tassert.Equal(t, types.SearchTypeDB, items[0].Type)\n\tassert.Equal(t, types.SourceAuto, items[0].Source)\n\tassert.Equal(t, \"product\", items[0].Model)\n\tassert.Equal(t, 1, items[0].RecordID)\n\tassert.Equal(t, \"Product 1\", items[0].Title)\n\tassert.Equal(t, \"Description 1\", items[0].Content)\n\tassert.NotNil(t, items[0].Data)\n\n\t// Second item\n\tassert.Equal(t, 2, items[1].RecordID)\n\tassert.Equal(t, \"Product 2\", items[1].Title)\n\tassert.Equal(t, \"Content 2\", items[1].Content)\n}\n\nfunc TestHandler_ConvertToResultItems_NilModel(t *testing.T) {\n\th := NewHandler(\"builtin\", nil)\n\n\trecords := []map[string]interface{}{\n\t\t{\"id\": 1, \"name\": \"Test\"},\n\t}\n\n\t// Should use default primary key \"id\" when model is nil\n\titems := h.convertToResultItems(records, \"test\", nil, types.SourceHook)\n\n\tassert.Len(t, items, 1)\n\tassert.Equal(t, 1, items[0].RecordID)\n\tassert.Equal(t, \"Test\", items[0].Title)\n}\n\nfunc TestHandler_Search_ScenarioTypes(t *testing.T) {\n\t// Test that all scenario types are valid\n\tscenarios := []types.ScenarioType{\n\t\ttypes.ScenarioFilter,\n\t\ttypes.ScenarioAggregation,\n\t\ttypes.ScenarioJoin,\n\t\ttypes.ScenarioComplex,\n\t}\n\n\tfor _, scenario := range scenarios {\n\t\tt.Run(string(scenario), func(t *testing.T) {\n\t\t\th := NewHandler(\"builtin\", &types.DBConfig{Models: []string{\"product\"}})\n\t\t\treq := &types.Request{\n\t\t\t\tType: types.SearchTypeDB,\n\t\t\t\tQuery: \"test query\",\n\t\t\t\tSource: types.SourceAuto,\n\t\t\t\tModels: []string{\"product\"},\n\t\t\t\tScenario: scenario,\n\t\t\t}\n\n\t\t\t// Without context, should return error (but scenario should be preserved in request)\n\t\t\tresult, err := h.Search(req)\n\t\t\tassert.NoError(t, err)\n\t\t\tassert.NotNil(t, result)\n\t\t\t// Verify request scenario is set correctly\n\t\t\tassert.Equal(t, scenario, req.Scenario)\n\t\t})\n\t}\n}\n\nfunc TestScenarioTypeConstants(t *testing.T) {\n\t// Verify scenario type constants match expected values\n\tassert.Equal(t, types.ScenarioType(\"filter\"), types.ScenarioFilter)\n\tassert.Equal(t, types.ScenarioType(\"aggregation\"), types.ScenarioAggregation)\n\tassert.Equal(t, types.ScenarioType(\"join\"), types.ScenarioJoin)\n\tassert.Equal(t, types.ScenarioType(\"complex\"), types.ScenarioComplex)\n}\n\nfunc TestHandler_MergeDSLConditions(t *testing.T) {\n\th := NewHandler(\"builtin\", nil)\n\n\tt.Run(\"merge wheres\", func(t *testing.T) {\n\t\tdsl := &gou.QueryDSL{\n\t\t\tFrom: &gou.Table{Name: \"users\"},\n\t\t\tWheres: []gou.Where{\n\t\t\t\t{Condition: gou.Condition{Field: &gou.Expression{Field: \"status\"}, Value: \"active\", OP: \"=\"}},\n\t\t\t},\n\t\t}\n\t\treq := &types.Request{\n\t\t\tWheres: []gou.Where{\n\t\t\t\t{Condition: gou.Condition{Field: &gou.Expression{Field: \"tenant_id\"}, Value: 1, OP: \"=\"}},\n\t\t\t},\n\t\t}\n\n\t\th.mergeDSLConditions(dsl, req)\n\n\t\t// Preset wheres should be prepended\n\t\tassert.Len(t, dsl.Wheres, 2)\n\t\tassert.Equal(t, \"tenant_id\", dsl.Wheres[0].Field.Field)\n\t\tassert.Equal(t, \"status\", dsl.Wheres[1].Field.Field)\n\t})\n\n\tt.Run(\"merge orders\", func(t *testing.T) {\n\t\tdsl := &gou.QueryDSL{\n\t\t\tFrom: &gou.Table{Name: \"products\"},\n\t\t\tOrders: gou.Orders{\n\t\t\t\t{Field: &gou.Expression{Field: \"name\"}, Sort: \"asc\"},\n\t\t\t},\n\t\t}\n\t\treq := &types.Request{\n\t\t\tOrders: gou.Orders{\n\t\t\t\t{Field: &gou.Expression{Field: \"created_at\"}, Sort: \"desc\"},\n\t\t\t},\n\t\t}\n\n\t\th.mergeDSLConditions(dsl, req)\n\n\t\t// Preset orders should be prepended\n\t\tassert.Len(t, dsl.Orders, 2)\n\t\tassert.Equal(t, \"created_at\", dsl.Orders[0].Field.Field)\n\t\tassert.Equal(t, \"name\", dsl.Orders[1].Field.Field)\n\t})\n\n\tt.Run(\"merge select fields\", func(t *testing.T) {\n\t\tdsl := &gou.QueryDSL{\n\t\t\tFrom: &gou.Table{Name: \"orders\"},\n\t\t\tSelect: []gou.Expression{\n\t\t\t\t{Field: \"amount\"},\n\t\t\t},\n\t\t}\n\t\treq := &types.Request{\n\t\t\tSelect: []string{\"id\", \"status\"},\n\t\t}\n\n\t\th.mergeDSLConditions(dsl, req)\n\n\t\t// Preset select should be prepended\n\t\tassert.Len(t, dsl.Select, 3)\n\t\tassert.Equal(t, \"id\", dsl.Select[0].Field)\n\t\tassert.Equal(t, \"status\", dsl.Select[1].Field)\n\t\tassert.Equal(t, \"amount\", dsl.Select[2].Field)\n\t})\n\n\tt.Run(\"set limit from request\", func(t *testing.T) {\n\t\tdsl := &gou.QueryDSL{\n\t\t\tFrom: &gou.Table{Name: \"users\"},\n\t\t\tLimit: 0,\n\t\t}\n\t\treq := &types.Request{\n\t\t\tLimit: 50,\n\t\t}\n\n\t\th.mergeDSLConditions(dsl, req)\n\n\t\tassert.Equal(t, 50, dsl.Limit)\n\t})\n\n\tt.Run(\"preserve dsl limit if set\", func(t *testing.T) {\n\t\tdsl := &gou.QueryDSL{\n\t\t\tFrom: &gou.Table{Name: \"users\"},\n\t\t\tLimit: 10,\n\t\t}\n\t\treq := &types.Request{\n\t\t\tLimit: 50,\n\t\t}\n\n\t\th.mergeDSLConditions(dsl, req)\n\n\t\t// DSL limit should be preserved\n\t\tassert.Equal(t, 10, dsl.Limit)\n\t})\n\n\tt.Run(\"nil dsl\", func(t *testing.T) {\n\t\treq := &types.Request{\n\t\t\tWheres: []gou.Where{\n\t\t\t\t{Condition: gou.Condition{Field: &gou.Expression{Field: \"id\"}, Value: 1}},\n\t\t\t},\n\t\t}\n\n\t\t// Should not panic\n\t\th.mergeDSLConditions(nil, req)\n\t})\n}\n" }, { "path": "agent/search/handlers/kb/handler.go", "content": "package kb\n\nimport (\n\t\"context\"\n\t\"fmt\"\n\t\"time\"\n\n\t\"github.com/yaoapp/yao/agent/search/types\"\n\t\"github.com/yaoapp/yao/kb\"\n\tkbapi \"github.com/yaoapp/yao/kb/api\"\n)\n\n// Handler implements KB search using the KB API\ntype Handler struct {\n\tconfig *types.KBConfig // KB search configuration\n}\n\n// NewHandler creates a new KB search handler\nfunc NewHandler(cfg *types.KBConfig) *Handler {\n\treturn &Handler{config: cfg}\n}\n\n// Type returns the search type this handler supports\nfunc (h *Handler) Type() types.SearchType {\n\treturn types.SearchTypeKB\n}\n\n// Search executes vector search and optional graph association\nfunc (h *Handler) Search(req *types.Request) (*types.Result, error) {\n\tstart := time.Now()\n\n\t// Validate request\n\tif req.Query == \"\" {\n\t\treturn &types.Result{\n\t\t\tType: types.SearchTypeKB,\n\t\t\tQuery: req.Query,\n\t\t\tSource: req.Source,\n\t\t\tItems: []*types.ResultItem{},\n\t\t\tTotal: 0,\n\t\t\tDuration: time.Since(start).Milliseconds(),\n\t\t\tError: \"query is required\",\n\t\t}, nil\n\t}\n\n\t// Check if KB API is available\n\tif kb.API == nil {\n\t\treturn &types.Result{\n\t\t\tType: types.SearchTypeKB,\n\t\t\tQuery: req.Query,\n\t\t\tSource: req.Source,\n\t\t\tItems: []*types.ResultItem{},\n\t\t\tTotal: 0,\n\t\t\tDuration: time.Since(start).Milliseconds(),\n\t\t\tError: \"knowledge base not initialized\",\n\t\t}, nil\n\t}\n\n\t// Get collections from request or config\n\tcollections := req.Collections\n\tif len(collections) == 0 && h.config != nil {\n\t\tcollections = h.config.Collections\n\t}\n\n\t// If no collections specified, return empty result\n\tif len(collections) == 0 {\n\t\treturn &types.Result{\n\t\t\tType: types.SearchTypeKB,\n\t\t\tQuery: req.Query,\n\t\t\tSource: req.Source,\n\t\t\tItems: []*types.ResultItem{},\n\t\t\tTotal: 0,\n\t\t\tDuration: time.Since(start).Milliseconds(),\n\t\t}, nil\n\t}\n\n\t// Get threshold from request or config\n\tthreshold := req.Threshold\n\tif threshold == 0 && h.config != nil && h.config.Threshold > 0 {\n\t\tthreshold = h.config.Threshold\n\t}\n\tif threshold == 0 {\n\t\tthreshold = 0.7 // default\n\t}\n\n\t// Get limit\n\tlimit := req.Limit\n\tif limit == 0 {\n\t\tlimit = 10 // default\n\t}\n\n\t// Determine search mode\n\tmode := kbapi.SearchModeVector\n\tif req.Graph {\n\t\tmode = kbapi.SearchModeExpand\n\t}\n\tif h.config != nil && h.config.Graph {\n\t\tmode = kbapi.SearchModeExpand\n\t}\n\n\t// Build KB API queries - one per collection\n\tvar queries []kbapi.Query\n\tfor _, collectionID := range collections {\n\t\tqueries = append(queries, kbapi.Query{\n\t\t\tCollectionID: collectionID,\n\t\t\tInput: req.Query,\n\t\t\tMode: mode,\n\t\t\tThreshold: threshold,\n\t\t\tPageSize: limit,\n\t\t\tMetadata: req.Metadata,\n\t\t})\n\t}\n\n\t// Execute search using KB API\n\tctx := context.Background()\n\tsearchResult, err := kb.API.Search(ctx, queries)\n\tif err != nil {\n\t\treturn &types.Result{\n\t\t\tType: types.SearchTypeKB,\n\t\t\tQuery: req.Query,\n\t\t\tSource: req.Source,\n\t\t\tItems: []*types.ResultItem{},\n\t\t\tTotal: 0,\n\t\t\tDuration: time.Since(start).Milliseconds(),\n\t\t\tError: fmt.Sprintf(\"search failed: %v\", err),\n\t\t}, nil\n\t}\n\n\t// Convert segments to result items\n\t// Note: MinScore filtering is already done by KB API, no need to filter again\n\titems := make([]*types.ResultItem, 0, len(searchResult.Segments))\n\tfor _, seg := range searchResult.Segments {\n\t\titem := &types.ResultItem{\n\t\t\tType: types.SearchTypeKB,\n\t\t\tSource: req.Source,\n\t\t\tScore: seg.Score,\n\t\t\tContent: seg.Text,\n\t\t\tDocumentID: seg.DocumentID,\n\t\t\tCollection: seg.CollectionID,\n\t\t\tMetadata: seg.Metadata,\n\t\t}\n\n\t\t// Extract title from metadata if available\n\t\tif seg.Metadata != nil {\n\t\t\tif title, ok := seg.Metadata[\"title\"].(string); ok {\n\t\t\t\titem.Title = title\n\t\t\t}\n\t\t}\n\n\t\titems = append(items, item)\n\t}\n\n\t// Convert graph data if available\n\tvar graphNodes []*types.GraphNode\n\tif searchResult.Graph != nil {\n\t\tfor _, node := range searchResult.Graph.Nodes {\n\t\t\t// Extract name from properties if available\n\t\t\tname := \"\"\n\t\t\tif node.Properties != nil {\n\t\t\t\tif n, ok := node.Properties[\"name\"].(string); ok {\n\t\t\t\t\tname = n\n\t\t\t\t}\n\t\t\t}\n\t\t\tgraphNodes = append(graphNodes, &types.GraphNode{\n\t\t\t\tID: node.ID,\n\t\t\t\tType: node.EntityType,\n\t\t\t\tName: name,\n\t\t\t\tMetadata: node.Properties,\n\t\t\t})\n\t\t}\n\t}\n\n\tresult := &types.Result{\n\t\tType: types.SearchTypeKB,\n\t\tQuery: req.Query,\n\t\tSource: req.Source,\n\t\tItems: items,\n\t\tTotal: len(items),\n\t\tDuration: time.Since(start).Milliseconds(),\n\t\tGraphNodes: graphNodes,\n\t}\n\n\treturn result, nil\n}\n" }, { "path": "agent/search/handlers/kb/handler_test.go", "content": "package kb\n\nimport (\n\t\"testing\"\n\n\t\"github.com/stretchr/testify/assert\"\n\t\"github.com/yaoapp/yao/agent/search/types\"\n)\n\nfunc TestNewHandler(t *testing.T) {\n\tt.Run(\"with nil config\", func(t *testing.T) {\n\t\th := NewHandler(nil)\n\t\tassert.NotNil(t, h)\n\t\tassert.Nil(t, h.config)\n\t})\n\n\tt.Run(\"with config\", func(t *testing.T) {\n\t\tcfg := &types.KBConfig{\n\t\t\tCollections: []string{\"docs\", \"faq\"},\n\t\t\tThreshold: 0.8,\n\t\t\tGraph: true,\n\t\t}\n\t\th := NewHandler(cfg)\n\t\tassert.NotNil(t, h)\n\t\tassert.Equal(t, cfg, h.config)\n\t})\n}\n\nfunc TestHandler_Type(t *testing.T) {\n\th := NewHandler(nil)\n\tassert.Equal(t, types.SearchTypeKB, h.Type())\n}\n\nfunc TestHandler_Search_Validation(t *testing.T) {\n\ttests := []struct {\n\t\tname string\n\t\tconfig *types.KBConfig\n\t\treq *types.Request\n\t\texpectError string\n\t}{\n\t\t{\n\t\t\tname: \"empty query\",\n\t\t\tconfig: nil,\n\t\t\treq: &types.Request{\n\t\t\t\tType: types.SearchTypeKB,\n\t\t\t\tQuery: \"\",\n\t\t\t},\n\t\t\texpectError: \"query is required\",\n\t\t},\n\t\t{\n\t\t\tname: \"no collections - KB not initialized\",\n\t\t\tconfig: nil,\n\t\t\treq: &types.Request{\n\t\t\t\tType: types.SearchTypeKB,\n\t\t\t\tQuery: \"test query\",\n\t\t\t},\n\t\t\texpectError: \"knowledge base not initialized\",\n\t\t},\n\t}\n\n\tfor _, tt := range tests {\n\t\tt.Run(tt.name, func(t *testing.T) {\n\t\t\th := NewHandler(tt.config)\n\t\t\tresult, err := h.Search(tt.req)\n\n\t\t\tassert.NoError(t, err)\n\t\t\tassert.NotNil(t, result)\n\t\t\tassert.Equal(t, types.SearchTypeKB, result.Type)\n\t\t\tassert.Equal(t, tt.req.Query, result.Query)\n\n\t\t\tif tt.expectError != \"\" {\n\t\t\t\tassert.Equal(t, tt.expectError, result.Error)\n\t\t\t} else {\n\t\t\t\tassert.Empty(t, result.Error)\n\t\t\t}\n\n\t\t\t// Duration should be set\n\t\t\tassert.GreaterOrEqual(t, result.Duration, int64(0))\n\t\t})\n\t}\n}\n\nfunc TestHandler_Search_SourcePreserved(t *testing.T) {\n\th := NewHandler(&types.KBConfig{Collections: []string{\"docs\"}})\n\n\tsources := []types.SourceType{types.SourceUser, types.SourceHook, types.SourceAuto}\n\tfor _, source := range sources {\n\t\treq := &types.Request{\n\t\t\tType: types.SearchTypeKB,\n\t\t\tQuery: \"test\",\n\t\t\tSource: source,\n\t\t\tCollections: []string{\"docs\"},\n\t\t}\n\t\tresult, err := h.Search(req)\n\t\tassert.NoError(t, err)\n\t\tassert.Equal(t, source, result.Source)\n\t}\n}\n\nfunc TestHandler_Search_CollectionsFromConfig(t *testing.T) {\n\tcfg := &types.KBConfig{\n\t\tCollections: []string{\"docs\", \"faq\"},\n\t\tThreshold: 0.7,\n\t}\n\th := NewHandler(cfg)\n\n\t// Request without collections should use config collections\n\treq := &types.Request{\n\t\tType: types.SearchTypeKB,\n\t\tQuery: \"test query\",\n\t}\n\tresult, err := h.Search(req)\n\n\tassert.NoError(t, err)\n\tassert.NotNil(t, result)\n\t// Without KB initialized, we get \"knowledge base not initialized\" error\n\tassert.Equal(t, \"knowledge base not initialized\", result.Error)\n}\n\nfunc TestHandler_Search_CollectionsFromRequest(t *testing.T) {\n\th := NewHandler(nil)\n\n\t// Request with collections\n\treq := &types.Request{\n\t\tType: types.SearchTypeKB,\n\t\tQuery: \"test query\",\n\t\tCollections: []string{\"docs\", \"faq\"},\n\t}\n\tresult, err := h.Search(req)\n\n\tassert.NoError(t, err)\n\tassert.NotNil(t, result)\n\t// Without KB initialized, we get \"knowledge base not initialized\" error\n\tassert.Equal(t, \"knowledge base not initialized\", result.Error)\n}\n\nfunc TestHandler_Search_ThresholdHandling(t *testing.T) {\n\ttests := []struct {\n\t\tname string\n\t\tconfigThreshold float64\n\t\treqThreshold float64\n\t}{\n\t\t{\n\t\t\tname: \"threshold from request\",\n\t\t\tconfigThreshold: 0.7,\n\t\t\treqThreshold: 0.9,\n\t\t},\n\t\t{\n\t\t\tname: \"threshold from config\",\n\t\t\tconfigThreshold: 0.8,\n\t\t\treqThreshold: 0,\n\t\t},\n\t\t{\n\t\t\tname: \"default threshold\",\n\t\t\tconfigThreshold: 0,\n\t\t\treqThreshold: 0,\n\t\t},\n\t}\n\n\tfor _, tt := range tests {\n\t\tt.Run(tt.name, func(t *testing.T) {\n\t\t\tvar cfg *types.KBConfig\n\t\t\tif tt.configThreshold > 0 {\n\t\t\t\tcfg = &types.KBConfig{\n\t\t\t\t\tCollections: []string{\"docs\"},\n\t\t\t\t\tThreshold: tt.configThreshold,\n\t\t\t\t}\n\t\t\t}\n\t\t\th := NewHandler(cfg)\n\n\t\t\treq := &types.Request{\n\t\t\t\tType: types.SearchTypeKB,\n\t\t\t\tQuery: \"test query\",\n\t\t\t\tThreshold: tt.reqThreshold,\n\t\t\t\tCollections: []string{\"docs\"},\n\t\t\t}\n\t\t\tresult, err := h.Search(req)\n\n\t\t\tassert.NoError(t, err)\n\t\t\tassert.NotNil(t, result)\n\t\t})\n\t}\n}\n\nfunc TestHandler_Search_GraphMode(t *testing.T) {\n\ttests := []struct {\n\t\tname string\n\t\tconfigGraph bool\n\t\treqGraph bool\n\t}{\n\t\t{\n\t\t\tname: \"graph from request\",\n\t\t\tconfigGraph: false,\n\t\t\treqGraph: true,\n\t\t},\n\t\t{\n\t\t\tname: \"graph from config\",\n\t\t\tconfigGraph: true,\n\t\t\treqGraph: false,\n\t\t},\n\t\t{\n\t\t\tname: \"no graph\",\n\t\t\tconfigGraph: false,\n\t\t\treqGraph: false,\n\t\t},\n\t}\n\n\tfor _, tt := range tests {\n\t\tt.Run(tt.name, func(t *testing.T) {\n\t\t\tcfg := &types.KBConfig{\n\t\t\t\tCollections: []string{\"docs\"},\n\t\t\t\tGraph: tt.configGraph,\n\t\t\t}\n\t\t\th := NewHandler(cfg)\n\n\t\t\treq := &types.Request{\n\t\t\t\tType: types.SearchTypeKB,\n\t\t\t\tQuery: \"test query\",\n\t\t\t\tCollections: []string{\"docs\"},\n\t\t\t\tGraph: tt.reqGraph,\n\t\t\t}\n\t\t\tresult, err := h.Search(req)\n\n\t\t\tassert.NoError(t, err)\n\t\t\tassert.NotNil(t, result)\n\t\t})\n\t}\n}\n\nfunc TestHandler_Search_LimitHandling(t *testing.T) {\n\th := NewHandler(&types.KBConfig{Collections: []string{\"docs\"}})\n\n\ttests := []struct {\n\t\tname string\n\t\tlimit int\n\t}{\n\t\t{\n\t\t\tname: \"custom limit\",\n\t\t\tlimit: 5,\n\t\t},\n\t\t{\n\t\t\tname: \"default limit\",\n\t\t\tlimit: 0,\n\t\t},\n\t}\n\n\tfor _, tt := range tests {\n\t\tt.Run(tt.name, func(t *testing.T) {\n\t\t\treq := &types.Request{\n\t\t\t\tType: types.SearchTypeKB,\n\t\t\t\tQuery: \"test query\",\n\t\t\t\tCollections: []string{\"docs\"},\n\t\t\t\tLimit: tt.limit,\n\t\t\t}\n\t\t\tresult, err := h.Search(req)\n\n\t\t\tassert.NoError(t, err)\n\t\t\tassert.NotNil(t, result)\n\t\t})\n\t}\n}\n" }, { "path": "agent/search/handlers/web/agent.go", "content": "package web\n\nimport (\n\t\"encoding/json\"\n\t\"fmt\"\n\t\"time\"\n\n\t\"github.com/yaoapp/yao/agent/caller\"\n\tagentContext \"github.com/yaoapp/yao/agent/context\"\n\t\"github.com/yaoapp/yao/agent/search/types\"\n)\n\n// AgentProvider implements web search using another agent (AI Search)\ntype AgentProvider struct {\n\tagentID string // Agent/Assistant ID (e.g., \"workers.search.web\")\n}\n\n// NewAgentProvider creates a new Agent provider\nfunc NewAgentProvider(agentID string) *AgentProvider {\n\treturn &AgentProvider{\n\t\tagentID: agentID,\n\t}\n}\n\n// Search executes web search via agent delegation\n// The agent can understand intent, generate optimized queries, and return structured results\nfunc (p *AgentProvider) Search(ctx *agentContext.Context, req *types.Request) (*types.Result, error) {\n\tstartTime := time.Now()\n\n\t// Check if context is provided\n\tif ctx == nil {\n\t\treturn &types.Result{\n\t\t\tType: types.SearchTypeWeb,\n\t\t\tQuery: req.Query,\n\t\t\tSource: req.Source,\n\t\t\tItems: []*types.ResultItem{},\n\t\t\tTotal: 0,\n\t\t\tDuration: time.Since(startTime).Milliseconds(),\n\t\t\tError: \"Agent mode requires context\",\n\t\t}, nil\n\t}\n\n\t// Check if AgentGetterFunc is initialized\n\tif caller.AgentGetterFunc == nil {\n\t\treturn &types.Result{\n\t\t\tType: types.SearchTypeWeb,\n\t\t\tQuery: req.Query,\n\t\t\tSource: req.Source,\n\t\t\tItems: []*types.ResultItem{},\n\t\t\tTotal: 0,\n\t\t\tDuration: time.Since(startTime).Milliseconds(),\n\t\t\tError: \"AgentGetterFunc not initialized\",\n\t\t}, nil\n\t}\n\n\t// Get the agent\n\tagent, err := caller.AgentGetterFunc(p.agentID)\n\tif err != nil {\n\t\treturn &types.Result{\n\t\t\tType: types.SearchTypeWeb,\n\t\t\tQuery: req.Query,\n\t\t\tSource: req.Source,\n\t\t\tItems: []*types.ResultItem{},\n\t\t\tTotal: 0,\n\t\t\tDuration: time.Since(startTime).Milliseconds(),\n\t\t\tError: fmt.Sprintf(\"Agent '%s' not found: %v\", p.agentID, err),\n\t\t}, nil\n\t}\n\n\t// Build message for the agent\n\t// Include search parameters in the message content\n\tsearchParams := map[string]interface{}{\n\t\t\"query\": req.Query,\n\t\t\"type\": \"web\",\n\t\t\"source\": string(req.Source),\n\t}\n\n\tif req.Limit > 0 {\n\t\tsearchParams[\"limit\"] = req.Limit\n\t}\n\tif len(req.Sites) > 0 {\n\t\tsearchParams[\"sites\"] = req.Sites\n\t}\n\tif req.TimeRange != \"\" {\n\t\tsearchParams[\"time_range\"] = req.TimeRange\n\t}\n\n\t// Convert to JSON for the message\n\tparamsJSON, err := json.Marshal(searchParams)\n\tif err != nil {\n\t\treturn &types.Result{\n\t\t\tType: types.SearchTypeWeb,\n\t\t\tQuery: req.Query,\n\t\t\tSource: req.Source,\n\t\t\tItems: []*types.ResultItem{},\n\t\t\tTotal: 0,\n\t\t\tDuration: time.Since(startTime).Milliseconds(),\n\t\t\tError: fmt.Sprintf(\"Failed to serialize search params: %v\", err),\n\t\t}, nil\n\t}\n\n\t// Create message for the agent\n\tmessage := agentContext.Message{\n\t\tRole: \"user\",\n\t\tContent: string(paramsJSON),\n\t}\n\n\t// Call the agent with skip options (no history, no output)\n\topts := &agentContext.Options{\n\t\tSkip: &agentContext.Skip{\n\t\t\tHistory: true,\n\t\t\tOutput: true,\n\t\t},\n\t}\n\n\tresponse, err := agent.Stream(ctx, []agentContext.Message{message}, opts)\n\tif err != nil {\n\t\treturn &types.Result{\n\t\t\tType: types.SearchTypeWeb,\n\t\t\tQuery: req.Query,\n\t\t\tSource: req.Source,\n\t\t\tItems: []*types.ResultItem{},\n\t\t\tTotal: 0,\n\t\t\tDuration: time.Since(startTime).Milliseconds(),\n\t\t\tError: fmt.Sprintf(\"Agent call failed: %v\", err),\n\t\t}, nil\n\t}\n\n\t// Parse the agent response\n\titems, total, parseErr := p.parseAgentResponse(response, req.Source)\n\tif parseErr != \"\" {\n\t\treturn &types.Result{\n\t\t\tType: types.SearchTypeWeb,\n\t\t\tQuery: req.Query,\n\t\t\tSource: req.Source,\n\t\t\tItems: []*types.ResultItem{},\n\t\t\tTotal: 0,\n\t\t\tDuration: time.Since(startTime).Milliseconds(),\n\t\t\tError: parseErr,\n\t\t}, nil\n\t}\n\n\treturn &types.Result{\n\t\tType: types.SearchTypeWeb,\n\t\tQuery: req.Query,\n\t\tSource: req.Source,\n\t\tItems: items,\n\t\tTotal: total,\n\t\tDuration: time.Since(startTime).Milliseconds(),\n\t}, nil\n}\n\n// parseAgentResponse parses the agent's *context.Response into search result items\n// Now that agent.Stream() returns *context.Response directly,\n// we can access fields without type assertions.\n//\n// The agent returns search results in response.Next field\nfunc (p *AgentProvider) parseAgentResponse(response *agentContext.Response, source types.SourceType) ([]*types.ResultItem, int, string) {\n\tif response == nil || response.Next == nil {\n\t\treturn nil, 0, \"Agent returned nil response\"\n\t}\n\n\t// Extract data from Next field\n\tdata := extractNextData(response.Next)\n\tif data == nil {\n\t\treturn nil, 0, \"Failed to extract data from agent response\"\n\t}\n\n\t// Extract items from data\n\titems := []*types.ResultItem{}\n\ttotal := 0\n\n\tif itemsData, ok := data[\"items\"].([]interface{}); ok {\n\t\tfor _, itemData := range itemsData {\n\t\t\tif item, ok := itemData.(map[string]interface{}); ok {\n\t\t\t\tresultItem := &types.ResultItem{\n\t\t\t\t\tType: types.SearchTypeWeb,\n\t\t\t\t\tSource: source,\n\t\t\t\t}\n\n\t\t\t\tif title, ok := item[\"title\"].(string); ok {\n\t\t\t\t\tresultItem.Title = title\n\t\t\t\t}\n\t\t\t\tif content, ok := item[\"content\"].(string); ok {\n\t\t\t\t\tresultItem.Content = content\n\t\t\t\t}\n\t\t\t\tif url, ok := item[\"url\"].(string); ok {\n\t\t\t\t\tresultItem.URL = url\n\t\t\t\t}\n\t\t\t\tif score, ok := item[\"score\"].(float64); ok {\n\t\t\t\t\tresultItem.Score = score\n\t\t\t\t}\n\n\t\t\t\titems = append(items, resultItem)\n\t\t\t}\n\t\t}\n\t}\n\n\tif totalVal, ok := data[\"total\"].(float64); ok {\n\t\ttotal = int(totalVal)\n\t} else {\n\t\ttotal = len(items)\n\t}\n\n\treturn items, total, \"\"\n}\n\n// extractNextData extracts the actual data from response.Next field\n// Handles nested structures like { \"data\": { ... } }\nfunc extractNextData(next interface{}) map[string]interface{} {\n\tif next == nil {\n\t\treturn nil\n\t}\n\n\tswitch v := next.(type) {\n\tcase map[string]interface{}:\n\t\t// Check for \"data\" wrapper\n\t\tif data, ok := v[\"data\"].(map[string]interface{}); ok {\n\t\t\treturn data\n\t\t}\n\t\treturn v\n\tcase string:\n\t\t// Try to parse as JSON\n\t\tvar data map[string]interface{}\n\t\tif err := json.Unmarshal([]byte(v), &data); err == nil {\n\t\t\treturn extractNextData(data)\n\t\t}\n\t}\n\t// Try to handle other types by converting to JSON and back\n\tif bytes, err := json.Marshal(next); err == nil {\n\t\tvar data map[string]interface{}\n\t\tif err := json.Unmarshal(bytes, &data); err == nil {\n\t\t\treturn extractNextData(data)\n\t\t}\n\t}\n\treturn nil\n}\n" }, { "path": "agent/search/handlers/web/agent_test.go", "content": "package web_test\n\nimport (\n\t\"context\"\n\t\"testing\"\n\n\t\"github.com/stretchr/testify/assert\"\n\t\"github.com/stretchr/testify/require\"\n\t\"github.com/yaoapp/yao/agent/assistant\"\n\tagentContext \"github.com/yaoapp/yao/agent/context\"\n\t\"github.com/yaoapp/yao/agent/search/handlers/web\"\n\t\"github.com/yaoapp/yao/agent/search/types\"\n\t\"github.com/yaoapp/yao/agent/testutils\"\n\toauthTypes \"github.com/yaoapp/yao/openapi/oauth/types\"\n)\n\n// TestAgentProviderWithAssistantConfig tests AgentProvider using web-agent-caller assistant config\nfunc TestAgentProviderWithAssistantConfig(t *testing.T) {\n\tif testing.Short() {\n\t\tt.Skip(\"Skipping integration test\")\n\t}\n\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\t// Load the web-agent-caller test assistant to get its config\n\tast, err := assistant.LoadPath(\"/assistants/tests/web-agent-caller\")\n\trequire.NoError(t, err)\n\trequire.NotNil(t, ast)\n\trequire.NotNil(t, ast.Uses)\n\n\t// Verify assistant config\n\tassert.Equal(t, \"tests.web-agent-caller\", ast.ID)\n\tassert.Equal(t, \"tests.web-agent\", ast.Uses.Web)\n\n\t// Create AgentProvider from uses.web\n\tprovider := web.NewAgentProvider(ast.Uses.Web)\n\trequire.NotNil(t, provider)\n\n\t// Create a mock context\n\tctx := createTestContext(t)\n\n\t// Execute search\n\treq := &types.Request{\n\t\tQuery: \"Yao App Engine\",\n\t\tType: types.SearchTypeWeb,\n\t\tSource: types.SourceAuto,\n\t\tLimit: 5,\n\t}\n\n\tresult, err := provider.Search(ctx, req)\n\trequire.NoError(t, err)\n\trequire.NotNil(t, result)\n\n\t// Verify result structure\n\tassert.Equal(t, types.SearchTypeWeb, result.Type)\n\tassert.Equal(t, \"Yao App Engine\", result.Query)\n\tassert.Equal(t, types.SourceAuto, result.Source)\n\n\t// Agent should return mock results from Next hook\n\tif result.Error == \"\" {\n\t\tassert.Greater(t, result.Total, 0)\n\t\tassert.NotEmpty(t, result.Items)\n\t\tassert.Greater(t, result.Duration, int64(0))\n\n\t\t// Verify result item structure\n\t\tfor _, item := range result.Items {\n\t\t\tassert.Equal(t, types.SearchTypeWeb, item.Type)\n\t\t\tassert.Equal(t, types.SourceAuto, item.Source)\n\t\t\tassert.NotEmpty(t, item.Title)\n\t\t\tassert.NotEmpty(t, item.URL)\n\t\t}\n\n\t\tt.Logf(\"Agent search returned %d results in %dms\", result.Total, result.Duration)\n\t} else {\n\t\tt.Logf(\"Agent search returned error: %s\", result.Error)\n\t}\n}\n\n// TestAgentProviderWithSiteRestriction tests AgentProvider with domain restriction\nfunc TestAgentProviderWithSiteRestriction(t *testing.T) {\n\tif testing.Short() {\n\t\tt.Skip(\"Skipping integration test\")\n\t}\n\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\t// Create AgentProvider\n\tprovider := web.NewAgentProvider(\"tests.web-agent\")\n\n\t// Create a mock context\n\tctx := createTestContext(t)\n\n\t// Execute search with site restriction\n\treq := &types.Request{\n\t\tQuery: \"documentation\",\n\t\tType: types.SearchTypeWeb,\n\t\tSource: types.SourceHook,\n\t\tSites: []string{\"github.com\"},\n\t\tLimit: 3,\n\t}\n\n\tresult, err := provider.Search(ctx, req)\n\trequire.NoError(t, err)\n\trequire.NotNil(t, result)\n\n\tassert.Equal(t, types.SearchTypeWeb, result.Type)\n\tassert.Equal(t, types.SourceHook, result.Source)\n\n\tif result.Error == \"\" {\n\t\t// All results should be from github.com (mock data respects sites)\n\t\tfor _, item := range result.Items {\n\t\t\tassert.Contains(t, item.URL, \"github.com\", \"Result URL should be from github.com\")\n\t\t}\n\t\tt.Logf(\"Site-restricted agent search returned %d results\", result.Total)\n\t} else {\n\t\tt.Logf(\"Agent search returned error: %s\", result.Error)\n\t}\n}\n\n// TestAgentProviderWithTimeRange tests AgentProvider with time range filter\nfunc TestAgentProviderWithTimeRange(t *testing.T) {\n\tif testing.Short() {\n\t\tt.Skip(\"Skipping integration test\")\n\t}\n\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\t// Create AgentProvider\n\tprovider := web.NewAgentProvider(\"tests.web-agent\")\n\n\t// Create a mock context\n\tctx := createTestContext(t)\n\n\t// Execute search with time range\n\treq := &types.Request{\n\t\tQuery: \"artificial intelligence news\",\n\t\tType: types.SearchTypeWeb,\n\t\tSource: types.SourceAuto,\n\t\tTimeRange: \"week\",\n\t\tLimit: 5,\n\t}\n\n\tresult, err := provider.Search(ctx, req)\n\trequire.NoError(t, err)\n\trequire.NotNil(t, result)\n\n\tif result.Error == \"\" {\n\t\tt.Logf(\"Time-ranged agent search (last week) returned %d results in %dms\", result.Total, result.Duration)\n\t} else {\n\t\tt.Logf(\"Agent search returned error: %s\", result.Error)\n\t}\n}\n\n// TestAgentProviderNotFound tests AgentProvider when agent is not found\nfunc TestAgentProviderNotFound(t *testing.T) {\n\tif testing.Short() {\n\t\tt.Skip(\"Skipping integration test\")\n\t}\n\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\t// Create AgentProvider with non-existent agent\n\tprovider := web.NewAgentProvider(\"nonexistent.agent\")\n\n\t// Create a mock context\n\tctx := createTestContext(t)\n\n\treq := &types.Request{\n\t\tQuery: \"test query\",\n\t\tType: types.SearchTypeWeb,\n\t\tSource: types.SourceAuto,\n\t}\n\n\tresult, err := provider.Search(ctx, req)\n\n\t// Should not return error, but result should have error message\n\trequire.NoError(t, err)\n\trequire.NotNil(t, result)\n\tassert.NotEmpty(t, result.Error)\n\tassert.Contains(t, result.Error, \"not found\")\n}\n\n// TestAgentProviderWithoutContext tests AgentProvider without context\nfunc TestAgentProviderWithoutContext(t *testing.T) {\n\tif testing.Short() {\n\t\tt.Skip(\"Skipping integration test\")\n\t}\n\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\t// Create AgentProvider\n\tprovider := web.NewAgentProvider(\"tests.web-agent\")\n\n\treq := &types.Request{\n\t\tQuery: \"test query\",\n\t\tType: types.SearchTypeWeb,\n\t\tSource: types.SourceAuto,\n\t}\n\n\t// Call without context (nil)\n\tresult, err := provider.Search(nil, req)\n\n\t// Should still work - agent provider handles nil context\n\trequire.NoError(t, err)\n\trequire.NotNil(t, result)\n\t// May have error if context is required for agent call\n\tt.Logf(\"Agent search without context: error=%s, total=%d\", result.Error, result.Total)\n}\n\n// TestWebHandlerAgentMode tests the web handler in agent mode\nfunc TestWebHandlerAgentMode(t *testing.T) {\n\tif testing.Short() {\n\t\tt.Skip(\"Skipping integration test\")\n\t}\n\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\t// Create handler with agent mode\n\thandler := web.NewHandler(\"tests.web-agent\", nil)\n\trequire.NotNil(t, handler)\n\n\t// Verify type\n\tassert.Equal(t, types.SearchTypeWeb, handler.Type())\n\n\t// Create a mock context\n\tctx := createTestContext(t)\n\n\t// Execute search with context\n\treq := &types.Request{\n\t\tQuery: \"Yao framework\",\n\t\tType: types.SearchTypeWeb,\n\t\tSource: types.SourceAuto,\n\t\tLimit: 5,\n\t}\n\n\tresult, err := handler.SearchWithContext(ctx, req)\n\trequire.NoError(t, err)\n\trequire.NotNil(t, result)\n\n\tassert.Equal(t, types.SearchTypeWeb, result.Type)\n\tassert.Equal(t, \"Yao framework\", result.Query)\n\n\tif result.Error == \"\" {\n\t\tt.Logf(\"Handler agent mode returned %d results\", result.Total)\n\t} else {\n\t\tt.Logf(\"Handler agent mode returned error: %s\", result.Error)\n\t}\n}\n\n// TestWebHandlerAgentModeWithoutContext tests the web handler in agent mode without context\nfunc TestWebHandlerAgentModeWithoutContext(t *testing.T) {\n\t// Create handler with agent mode\n\thandler := web.NewHandler(\"tests.web-agent\", nil)\n\trequire.NotNil(t, handler)\n\n\treq := &types.Request{\n\t\tQuery: \"test\",\n\t\tType: types.SearchTypeWeb,\n\t\tSource: types.SourceAuto,\n\t}\n\n\t// Call Search() without context (uses SearchWithContext with nil)\n\tresult, err := handler.Search(req)\n\trequire.NoError(t, err)\n\trequire.NotNil(t, result)\n\tassert.NotEmpty(t, result.Error)\n\tassert.Contains(t, result.Error, \"requires context\")\n}\n\n// createTestContext creates a test context for agent calls\nfunc createTestContext(t *testing.T) *agentContext.Context {\n\tauthorized := &oauthTypes.AuthorizedInfo{\n\t\tUserID: \"test-user\",\n\t\tTenantID: \"test-tenant\",\n\t}\n\tctx := agentContext.New(context.Background(), authorized, \"test-chat-id\")\n\tctx.AssistantID = \"tests.web-agent-caller\"\n\treturn ctx\n}\n" }, { "path": "agent/search/handlers/web/handler.go", "content": "package web\n\nimport (\n\t\"fmt\"\n\t\"strings\"\n\n\tagentContext \"github.com/yaoapp/yao/agent/context\"\n\t\"github.com/yaoapp/yao/agent/search/types\"\n)\n\n// Handler implements web search\ntype Handler struct {\n\tusesWeb string // \"builtin\", \"\", \"mcp:.\"\n\tconfig *types.WebConfig // Web search configuration\n}\n\n// NewHandler creates a new web search handler\nfunc NewHandler(usesWeb string, cfg *types.WebConfig) *Handler {\n\treturn &Handler{usesWeb: usesWeb, config: cfg}\n}\n\n// Type returns the search type this handler supports\nfunc (h *Handler) Type() types.SearchType {\n\treturn types.SearchTypeWeb\n}\n\n// Search executes web search based on uses.web mode\n// ctx is optional and only required for agent mode\nfunc (h *Handler) Search(req *types.Request) (*types.Result, error) {\n\treturn h.SearchWithContext(nil, req)\n}\n\n// SearchWithContext executes web search with optional agent context\n// ctx is required for agent mode, optional for builtin and MCP modes\nfunc (h *Handler) SearchWithContext(ctx *agentContext.Context, req *types.Request) (*types.Result, error) {\n\tswitch {\n\tcase h.usesWeb == \"builtin\" || h.usesWeb == \"\":\n\t\treturn h.builtinSearch(req)\n\tcase strings.HasPrefix(h.usesWeb, \"mcp:\"):\n\t\treturn h.mcpSearch(req)\n\tdefault:\n\t\t// Agent mode: delegate to assistant for AI-powered search\n\t\tif ctx == nil {\n\t\t\treturn &types.Result{\n\t\t\t\tType: types.SearchTypeWeb,\n\t\t\t\tQuery: req.Query,\n\t\t\t\tSource: req.Source,\n\t\t\t\tItems: []*types.ResultItem{},\n\t\t\t\tTotal: 0,\n\t\t\t\tError: \"Agent mode requires context\",\n\t\t\t}, nil\n\t\t}\n\t\treturn h.agentSearch(ctx, req)\n\t}\n}\n\n// builtinSearch uses Tavily/Serper/SerpAPI directly\nfunc (h *Handler) builtinSearch(req *types.Request) (*types.Result, error) {\n\t// Determine provider from config\n\tproviderName := \"tavily\" // default\n\tif h.config != nil && h.config.Provider != \"\" {\n\t\tproviderName = h.config.Provider\n\t}\n\n\tswitch providerName {\n\tcase \"tavily\":\n\t\treturn NewTavilyProvider(h.config).Search(req)\n\tcase \"serper\":\n\t\t// Serper (serper.dev) - POST request with X-API-KEY header\n\t\treturn NewSerperProvider(h.config).Search(req)\n\tcase \"serpapi\":\n\t\t// SerpAPI (serpapi.com) - GET request with api_key parameter\n\t\treturn NewSerpAPIProvider(h.config).Search(req)\n\tdefault:\n\t\treturn &types.Result{\n\t\t\tType: types.SearchTypeWeb,\n\t\t\tQuery: req.Query,\n\t\t\tSource: req.Source,\n\t\t\tItems: []*types.ResultItem{},\n\t\t\tTotal: 0,\n\t\t\tError: fmt.Sprintf(\"Unknown provider: %s (supported: tavily, serper, serpapi)\", providerName),\n\t\t}, nil\n\t}\n}\n\n// agentSearch delegates to an assistant for AI-powered search\nfunc (h *Handler) agentSearch(ctx *agentContext.Context, req *types.Request) (*types.Result, error) {\n\tprovider := NewAgentProvider(h.usesWeb)\n\treturn provider.Search(ctx, req)\n}\n\n// mcpSearch calls external MCP tool\nfunc (h *Handler) mcpSearch(req *types.Request) (*types.Result, error) {\n\t// Parse \"mcp:server.tool\"\n\tmcpRef := strings.TrimPrefix(h.usesWeb, \"mcp:\")\n\n\tprovider, err := NewMCPProvider(mcpRef)\n\tif err != nil {\n\t\treturn &types.Result{\n\t\t\tType: types.SearchTypeWeb,\n\t\t\tQuery: req.Query,\n\t\t\tSource: req.Source,\n\t\t\tItems: []*types.ResultItem{},\n\t\t\tTotal: 0,\n\t\t\tError: fmt.Sprintf(\"Invalid MCP format: %v\", err),\n\t\t}, nil\n\t}\n\n\treturn provider.Search(req)\n}\n" }, { "path": "agent/search/handlers/web/mcp.go", "content": "package web\n\nimport (\n\t\"context\"\n\t\"encoding/json\"\n\t\"fmt\"\n\t\"strings\"\n\t\"time\"\n\n\t\"github.com/yaoapp/gou/mcp\"\n\tgouMCPTypes \"github.com/yaoapp/gou/mcp/types\"\n\t\"github.com/yaoapp/yao/agent/search/types\"\n)\n\n// MCPProvider implements web search using MCP tool\ntype MCPProvider struct {\n\tserverID string // MCP server ID (e.g., \"search\")\n\ttoolName string // MCP tool name (e.g., \"web_search\")\n}\n\n// NewMCPProvider creates a new MCP provider from \"mcp:server.tool\" format\nfunc NewMCPProvider(mcpRef string) (*MCPProvider, error) {\n\t// Parse \"server.tool\" format\n\tparts := strings.SplitN(mcpRef, \".\", 2)\n\tif len(parts) != 2 {\n\t\treturn nil, fmt.Errorf(\"invalid MCP format, expected 'server.tool', got '%s'\", mcpRef)\n\t}\n\n\treturn &MCPProvider{\n\t\tserverID: parts[0],\n\t\ttoolName: parts[1],\n\t}, nil\n}\n\n// Search executes web search via MCP tool\nfunc (p *MCPProvider) Search(req *types.Request) (*types.Result, error) {\n\tstartTime := time.Now()\n\n\t// Select MCP client\n\tclient, err := mcp.Select(p.serverID)\n\tif err != nil {\n\t\treturn &types.Result{\n\t\t\tType: types.SearchTypeWeb,\n\t\t\tQuery: req.Query,\n\t\t\tSource: req.Source,\n\t\t\tItems: []*types.ResultItem{},\n\t\t\tTotal: 0,\n\t\t\tDuration: time.Since(startTime).Milliseconds(),\n\t\t\tError: fmt.Sprintf(\"MCP client '%s' not found: %v\", p.serverID, err),\n\t\t}, nil\n\t}\n\n\t// Build MCP tool arguments\n\targs := map[string]interface{}{\n\t\t\"query\": req.Query,\n\t}\n\n\tif req.Limit > 0 {\n\t\targs[\"limit\"] = req.Limit\n\t}\n\n\tif len(req.Sites) > 0 {\n\t\targs[\"sites\"] = req.Sites\n\t}\n\n\tif req.TimeRange != \"\" {\n\t\targs[\"time_range\"] = req.TimeRange\n\t}\n\n\t// Call MCP tool\n\tctx, cancel := context.WithTimeout(context.Background(), 30*time.Second)\n\tdefer cancel()\n\n\tresult, err := client.CallTool(ctx, p.toolName, args)\n\tif err != nil {\n\t\treturn &types.Result{\n\t\t\tType: types.SearchTypeWeb,\n\t\t\tQuery: req.Query,\n\t\t\tSource: req.Source,\n\t\t\tItems: []*types.ResultItem{},\n\t\t\tTotal: 0,\n\t\t\tDuration: time.Since(startTime).Milliseconds(),\n\t\t\tError: fmt.Sprintf(\"MCP tool call failed: %v\", err),\n\t\t}, nil\n\t}\n\n\t// Parse MCP result\n\titems, total, parseErr := p.parseResult(result, req.Source)\n\tif parseErr != \"\" {\n\t\treturn &types.Result{\n\t\t\tType: types.SearchTypeWeb,\n\t\t\tQuery: req.Query,\n\t\t\tSource: req.Source,\n\t\t\tItems: []*types.ResultItem{},\n\t\t\tTotal: 0,\n\t\t\tDuration: time.Since(startTime).Milliseconds(),\n\t\t\tError: parseErr,\n\t\t}, nil\n\t}\n\n\treturn &types.Result{\n\t\tType: types.SearchTypeWeb,\n\t\tQuery: req.Query,\n\t\tSource: req.Source,\n\t\tItems: items,\n\t\tTotal: total,\n\t\tDuration: time.Since(startTime).Milliseconds(),\n\t}, nil\n}\n\n// parseResult parses MCP tool result into search result items\nfunc (p *MCPProvider) parseResult(result *gouMCPTypes.CallToolResponse, source types.SourceType) ([]*types.ResultItem, int, string) {\n\tif result == nil {\n\t\treturn nil, 0, \"MCP returned nil result\"\n\t}\n\n\t// Check for errors in result\n\tif result.IsError {\n\t\terrMsg := \"MCP tool returned error\"\n\t\tif len(result.Content) > 0 && result.Content[0].Text != \"\" {\n\t\t\terrMsg = result.Content[0].Text\n\t\t}\n\t\treturn nil, 0, errMsg\n\t}\n\n\t// Parse content - expect JSON data\n\tif len(result.Content) == 0 {\n\t\treturn []*types.ResultItem{}, 0, \"\"\n\t}\n\n\t// Try to extract data from content\n\tvar data map[string]interface{}\n\n\tfor _, content := range result.Content {\n\t\t// Check text content type\n\t\tif content.Type == gouMCPTypes.ToolContentTypeText && content.Text != \"\" {\n\t\t\t// Try to parse as JSON\n\t\t\tif parsed, ok := parseJSON(content.Text); ok {\n\t\t\t\tdata = parsed\n\t\t\t\tbreak\n\t\t\t}\n\t\t}\n\t}\n\n\tif data == nil {\n\t\treturn []*types.ResultItem{}, 0, \"\"\n\t}\n\n\t// Extract items from data\n\titems := []*types.ResultItem{}\n\ttotal := 0\n\n\tif itemsData, ok := data[\"items\"].([]interface{}); ok {\n\t\tfor _, itemData := range itemsData {\n\t\t\tif item, ok := itemData.(map[string]interface{}); ok {\n\t\t\t\tresultItem := &types.ResultItem{\n\t\t\t\t\tType: types.SearchTypeWeb,\n\t\t\t\t\tSource: source,\n\t\t\t\t}\n\n\t\t\t\tif title, ok := item[\"title\"].(string); ok {\n\t\t\t\t\tresultItem.Title = title\n\t\t\t\t}\n\t\t\t\tif content, ok := item[\"content\"].(string); ok {\n\t\t\t\t\tresultItem.Content = content\n\t\t\t\t}\n\t\t\t\tif url, ok := item[\"url\"].(string); ok {\n\t\t\t\t\tresultItem.URL = url\n\t\t\t\t}\n\t\t\t\tif score, ok := item[\"score\"].(float64); ok {\n\t\t\t\t\tresultItem.Score = score\n\t\t\t\t}\n\n\t\t\t\titems = append(items, resultItem)\n\t\t\t}\n\t\t}\n\t}\n\n\tif totalVal, ok := data[\"total\"].(float64); ok {\n\t\ttotal = int(totalVal)\n\t} else {\n\t\ttotal = len(items)\n\t}\n\n\treturn items, total, \"\"\n}\n\n// parseJSON attempts to parse a string as JSON\nfunc parseJSON(s string) (map[string]interface{}, bool) {\n\t// Simple JSON detection - if it starts with { and ends with }\n\ts = strings.TrimSpace(s)\n\tif !strings.HasPrefix(s, \"{\") || !strings.HasSuffix(s, \"}\") {\n\t\treturn nil, false\n\t}\n\n\t// Use encoding/json for parsing\n\tvar result map[string]interface{}\n\tif err := json.Unmarshal([]byte(s), &result); err != nil {\n\t\treturn nil, false\n\t}\n\n\treturn result, true\n}\n" }, { "path": "agent/search/handlers/web/mcp_test.go", "content": "package web_test\n\nimport (\n\t\"testing\"\n\n\t\"github.com/stretchr/testify/assert\"\n\t\"github.com/stretchr/testify/require\"\n\t\"github.com/yaoapp/yao/agent/assistant\"\n\t\"github.com/yaoapp/yao/agent/search/handlers/web\"\n\t\"github.com/yaoapp/yao/agent/search/types\"\n\t\"github.com/yaoapp/yao/agent/testutils\"\n)\n\n// TestMCPProviderWithAssistantConfig tests MCPProvider using web-mcp assistant config\nfunc TestMCPProviderWithAssistantConfig(t *testing.T) {\n\tif testing.Short() {\n\t\tt.Skip(\"Skipping integration test\")\n\t}\n\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\t// Load the web-mcp test assistant to get its config\n\tast, err := assistant.LoadPath(\"/assistants/tests/web-mcp\")\n\trequire.NoError(t, err)\n\trequire.NotNil(t, ast)\n\trequire.NotNil(t, ast.Uses)\n\n\t// Verify assistant config\n\tassert.Equal(t, \"tests.web-mcp\", ast.ID)\n\tassert.Equal(t, \"mcp:search.web_search\", ast.Uses.Web)\n\n\t// Create MCPProvider from uses.web\n\tmcpRef := ast.Uses.Web[4:] // Remove \"mcp:\" prefix\n\tprovider, err := web.NewMCPProvider(mcpRef)\n\trequire.NoError(t, err)\n\trequire.NotNil(t, provider)\n\n\t// Execute search\n\treq := &types.Request{\n\t\tQuery: \"Yao App Engine\",\n\t\tType: types.SearchTypeWeb,\n\t\tSource: types.SourceAuto,\n\t\tLimit: 5,\n\t}\n\n\tresult, err := provider.Search(req)\n\trequire.NoError(t, err)\n\trequire.NotNil(t, result)\n\n\t// Verify result structure\n\tassert.Equal(t, types.SearchTypeWeb, result.Type)\n\tassert.Equal(t, \"Yao App Engine\", result.Query)\n\tassert.Equal(t, types.SourceAuto, result.Source)\n\n\t// MCP should return mock results\n\tif result.Error == \"\" {\n\t\tassert.Greater(t, result.Total, 0)\n\t\tassert.NotEmpty(t, result.Items)\n\t\tassert.Greater(t, result.Duration, int64(0))\n\n\t\t// Verify result item structure\n\t\tfor _, item := range result.Items {\n\t\t\tassert.Equal(t, types.SearchTypeWeb, item.Type)\n\t\t\tassert.Equal(t, types.SourceAuto, item.Source)\n\t\t\tassert.NotEmpty(t, item.Title)\n\t\t\tassert.NotEmpty(t, item.URL)\n\t\t}\n\n\t\tt.Logf(\"MCP search returned %d results in %dms\", result.Total, result.Duration)\n\t} else {\n\t\tt.Logf(\"MCP search returned error (expected if MCP not loaded): %s\", result.Error)\n\t}\n}\n\n// TestMCPProviderWithSiteRestriction tests MCPProvider with domain restriction\nfunc TestMCPProviderWithSiteRestriction(t *testing.T) {\n\tif testing.Short() {\n\t\tt.Skip(\"Skipping integration test\")\n\t}\n\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\t// Create MCPProvider\n\tprovider, err := web.NewMCPProvider(\"search.web_search\")\n\trequire.NoError(t, err)\n\n\t// Execute search with site restriction\n\treq := &types.Request{\n\t\tQuery: \"documentation\",\n\t\tType: types.SearchTypeWeb,\n\t\tSource: types.SourceHook,\n\t\tSites: []string{\"github.com\"},\n\t\tLimit: 3,\n\t}\n\n\tresult, err := provider.Search(req)\n\trequire.NoError(t, err)\n\trequire.NotNil(t, result)\n\n\tassert.Equal(t, types.SearchTypeWeb, result.Type)\n\tassert.Equal(t, types.SourceHook, result.Source)\n\n\tif result.Error == \"\" {\n\t\tt.Logf(\"Site-restricted MCP search returned %d results\", result.Total)\n\t} else {\n\t\tt.Logf(\"MCP search returned error (expected if MCP not loaded): %s\", result.Error)\n\t}\n}\n\n// TestMCPProviderWithTimeRange tests MCPProvider with time range filter\nfunc TestMCPProviderWithTimeRange(t *testing.T) {\n\tif testing.Short() {\n\t\tt.Skip(\"Skipping integration test\")\n\t}\n\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\t// Create MCPProvider\n\tprovider, err := web.NewMCPProvider(\"search.web_search\")\n\trequire.NoError(t, err)\n\n\t// Execute search with time range\n\treq := &types.Request{\n\t\tQuery: \"artificial intelligence news\",\n\t\tType: types.SearchTypeWeb,\n\t\tSource: types.SourceAuto,\n\t\tTimeRange: \"week\",\n\t\tLimit: 5,\n\t}\n\n\tresult, err := provider.Search(req)\n\trequire.NoError(t, err)\n\trequire.NotNil(t, result)\n\n\tif result.Error == \"\" {\n\t\tt.Logf(\"Time-ranged MCP search (last week) returned %d results in %dms\", result.Total, result.Duration)\n\t} else {\n\t\tt.Logf(\"MCP search returned error (expected if MCP not loaded): %s\", result.Error)\n\t}\n}\n\n// TestMCPProviderInvalidFormat tests MCPProvider with invalid format\nfunc TestMCPProviderInvalidFormat(t *testing.T) {\n\t// Test invalid format without dot\n\t_, err := web.NewMCPProvider(\"invalid\")\n\trequire.Error(t, err)\n\tassert.Contains(t, err.Error(), \"invalid MCP format\")\n\n\t// Test empty string\n\t_, err = web.NewMCPProvider(\"\")\n\trequire.Error(t, err)\n\tassert.Contains(t, err.Error(), \"invalid MCP format\")\n}\n\n// TestMCPProviderNotFound tests MCPProvider when MCP server is not found\nfunc TestMCPProviderNotFound(t *testing.T) {\n\tif testing.Short() {\n\t\tt.Skip(\"Skipping integration test\")\n\t}\n\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\t// Create MCPProvider with non-existent server\n\tprovider, err := web.NewMCPProvider(\"nonexistent.web_search\")\n\trequire.NoError(t, err)\n\n\treq := &types.Request{\n\t\tQuery: \"test query\",\n\t\tType: types.SearchTypeWeb,\n\t\tSource: types.SourceAuto,\n\t}\n\n\tresult, err := provider.Search(req)\n\n\t// Should not return error, but result should have error message\n\trequire.NoError(t, err)\n\trequire.NotNil(t, result)\n\tassert.NotEmpty(t, result.Error)\n\tassert.Contains(t, result.Error, \"not found\")\n}\n\n// TestWebHandlerMCPMode tests the web handler in MCP mode\nfunc TestWebHandlerMCPMode(t *testing.T) {\n\tif testing.Short() {\n\t\tt.Skip(\"Skipping integration test\")\n\t}\n\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\t// Create handler with MCP mode\n\thandler := web.NewHandler(\"mcp:search.web_search\", nil)\n\trequire.NotNil(t, handler)\n\n\t// Verify type\n\tassert.Equal(t, types.SearchTypeWeb, handler.Type())\n\n\t// Execute search\n\treq := &types.Request{\n\t\tQuery: \"Yao framework\",\n\t\tType: types.SearchTypeWeb,\n\t\tSource: types.SourceAuto,\n\t\tLimit: 5,\n\t}\n\n\tresult, err := handler.Search(req)\n\trequire.NoError(t, err)\n\trequire.NotNil(t, result)\n\n\tassert.Equal(t, types.SearchTypeWeb, result.Type)\n\tassert.Equal(t, \"Yao framework\", result.Query)\n\n\tif result.Error == \"\" {\n\t\tt.Logf(\"Handler MCP mode returned %d results\", result.Total)\n\t} else {\n\t\tt.Logf(\"Handler MCP mode returned error (expected if MCP not loaded): %s\", result.Error)\n\t}\n}\n\n// TestWebHandlerInvalidMCPFormat tests the web handler with invalid MCP format\nfunc TestWebHandlerInvalidMCPFormat(t *testing.T) {\n\t// Create handler with invalid MCP format\n\thandler := web.NewHandler(\"mcp:invalid\", nil)\n\trequire.NotNil(t, handler)\n\n\treq := &types.Request{\n\t\tQuery: \"test\",\n\t\tType: types.SearchTypeWeb,\n\t\tSource: types.SourceAuto,\n\t}\n\n\tresult, err := handler.Search(req)\n\trequire.NoError(t, err)\n\trequire.NotNil(t, result)\n\tassert.NotEmpty(t, result.Error)\n\tassert.Contains(t, result.Error, \"Invalid MCP format\")\n}\n" }, { "path": "agent/search/handlers/web/serpapi.go", "content": "package web\n\nimport (\n\t\"encoding/json\"\n\t\"fmt\"\n\t\"io\"\n\t\"net/http\"\n\t\"net/url\"\n\t\"os\"\n\t\"time\"\n\n\t\"github.com/yaoapp/yao/agent/search/types\"\n)\n\nconst (\n\tserpAPIURL = \"https://serpapi.com/search.json\"\n\tserpAPITimeout = 30 * time.Second\n)\n\n// SerpAPIProvider implements web search using SerpAPI (supports multiple search engines)\ntype SerpAPIProvider struct {\n\tapiKey string\n\tmaxResults int\n\tengine string // Search engine: \"google\", \"bing\", \"baidu\", \"yandex\", \"duckduckgo\", etc.\n}\n\n// NewSerpAPIProvider creates a new SerpAPI provider\nfunc NewSerpAPIProvider(cfg *types.WebConfig) *SerpAPIProvider {\n\tapiKey := \"\"\n\tif cfg != nil && cfg.APIKeyEnv != \"\" {\n\t\t// Support both \"$ENV.VAR_NAME\" and \"VAR_NAME\" formats\n\t\tenvName := cfg.APIKeyEnv\n\t\tif len(envName) > 5 && envName[:5] == \"$ENV.\" {\n\t\t\tenvName = envName[5:]\n\t\t}\n\t\tapiKey = os.Getenv(envName)\n\t}\n\n\tmaxResults := 10\n\tif cfg != nil && cfg.MaxResults > 0 {\n\t\tmaxResults = cfg.MaxResults\n\t}\n\n\tengine := \"google\" // Default to Google\n\tif cfg != nil && cfg.Engine != \"\" {\n\t\tengine = cfg.Engine\n\t}\n\n\treturn &SerpAPIProvider{\n\t\tapiKey: apiKey,\n\t\tmaxResults: maxResults,\n\t\tengine: engine,\n\t}\n}\n\n// serpAPIResponse represents the response from SerpAPI\ntype serpAPIResponse struct {\n\tSearchMetadata serpAPIMetadata `json:\"search_metadata\"`\n\tSearchParameters serpAPIParams `json:\"search_parameters\"`\n\tSearchInformation serpAPIInfo `json:\"search_information\"`\n\tOrganicResults []serpAPIResult `json:\"organic_results\"`\n\tAnswerBox *serpAPIAnswerBox `json:\"answer_box,omitempty\"`\n\tKnowledgeGraph *serpAPIKnowledge `json:\"knowledge_graph,omitempty\"`\n\tRelatedSearches []serpAPIRelated `json:\"related_searches,omitempty\"`\n\tRelatedQuestions []serpAPIQuestion `json:\"related_questions,omitempty\"`\n}\n\n// serpAPIMetadata contains metadata from response\ntype serpAPIMetadata struct {\n\tID string `json:\"id\"`\n\tStatus string `json:\"status\"`\n\tCreatedAt string `json:\"created_at\"`\n\tProcessedAt string `json:\"processed_at\"`\n\tTotalTimeTaken float64 `json:\"total_time_taken\"`\n}\n\n// serpAPIParams contains search parameters from response\ntype serpAPIParams struct {\n\tEngine string `json:\"engine\"`\n\tQ string `json:\"q\"`\n\tLocation string `json:\"location_used\"`\n\tGoogleDomain string `json:\"google_domain\"`\n\tHL string `json:\"hl\"`\n\tGL string `json:\"gl\"`\n\tDevice string `json:\"device\"`\n}\n\n// serpAPIInfo contains search information\ntype serpAPIInfo struct {\n\tQueryDisplayed string `json:\"query_displayed\"`\n\tTotalResults int64 `json:\"total_results\"`\n\tTimeTakenDisplayed float64 `json:\"time_taken_displayed\"`\n\tOrganicResultsState string `json:\"organic_results_state\"`\n}\n\n// serpAPIResult represents a single organic search result\ntype serpAPIResult struct {\n\tPosition int `json:\"position\"`\n\tTitle string `json:\"title\"`\n\tLink string `json:\"link\"`\n\tRedirectLink string `json:\"redirect_link,omitempty\"`\n\tDisplayedLink string `json:\"displayed_link\"`\n\tSnippet string `json:\"snippet\"`\n\tDate string `json:\"date,omitempty\"`\n\tCachedPageLink string `json:\"cached_page_link,omitempty\"`\n}\n\n// serpAPIAnswerBox represents the answer box (featured snippet)\ntype serpAPIAnswerBox struct {\n\tType string `json:\"type,omitempty\"`\n\tTitle string `json:\"title,omitempty\"`\n\tSnippet string `json:\"snippet,omitempty\"`\n\tLink string `json:\"link,omitempty\"`\n}\n\n// serpAPIKnowledge represents knowledge graph data\ntype serpAPIKnowledge struct {\n\tTitle string `json:\"title,omitempty\"`\n\tType interface{} `json:\"type,omitempty\"` // Can be string or object depending on query\n\tDescription string `json:\"description,omitempty\"`\n}\n\n// serpAPIRelated represents related searches\ntype serpAPIRelated struct {\n\tQuery string `json:\"query\"`\n\tLink string `json:\"link\"`\n}\n\n// serpAPIQuestion represents related questions (People Also Ask)\ntype serpAPIQuestion struct {\n\tQuestion string `json:\"question\"`\n\tSnippet string `json:\"snippet,omitempty\"`\n\tTitle string `json:\"title,omitempty\"`\n\tLink string `json:\"link,omitempty\"`\n}\n\n// Search executes a web search using SerpAPI\nfunc (p *SerpAPIProvider) Search(req *types.Request) (*types.Result, error) {\n\tstartTime := time.Now()\n\n\t// Validate API key\n\tif p.apiKey == \"\" {\n\t\treturn &types.Result{\n\t\t\tType: types.SearchTypeWeb,\n\t\t\tQuery: req.Query,\n\t\t\tSource: req.Source,\n\t\t\tItems: []*types.ResultItem{},\n\t\t\tTotal: 0,\n\t\t\tError: \"SerpAPI API key not configured\",\n\t\t}, nil\n\t}\n\n\t// Determine max results\n\tmaxResults := p.maxResults\n\tif req.Limit > 0 {\n\t\tmaxResults = req.Limit\n\t}\n\n\t// Build query parameters\n\tparams := url.Values{}\n\tparams.Set(\"engine\", p.engine)\n\tparams.Set(\"api_key\", p.apiKey)\n\tparams.Set(\"num\", fmt.Sprintf(\"%d\", maxResults))\n\n\t// Build search query with site restrictions if specified\n\tquery := req.Query\n\tif len(req.Sites) > 0 {\n\t\tsiteQuery := \"\"\n\t\tfor i, site := range req.Sites {\n\t\t\tif i > 0 {\n\t\t\t\tsiteQuery += \" OR \"\n\t\t\t}\n\t\t\tsiteQuery += \"site:\" + site\n\t\t}\n\t\tquery = \"(\" + siteQuery + \") \" + req.Query\n\t}\n\tparams.Set(\"q\", query)\n\n\t// Add time range if specified (tbs parameter)\n\tif req.TimeRange != \"\" {\n\t\ttbs := convertSerpAPITimeRange(req.TimeRange)\n\t\tif tbs != \"\" {\n\t\t\tparams.Set(\"tbs\", tbs)\n\t\t}\n\t}\n\n\t// Execute API call\n\tserpResp, err := p.callAPI(params)\n\tif err != nil {\n\t\treturn &types.Result{\n\t\t\tType: types.SearchTypeWeb,\n\t\t\tQuery: req.Query,\n\t\t\tSource: req.Source,\n\t\t\tItems: []*types.ResultItem{},\n\t\t\tTotal: 0,\n\t\t\tDuration: time.Since(startTime).Milliseconds(),\n\t\t\tError: fmt.Sprintf(\"SerpAPI error: %v\", err),\n\t\t}, nil\n\t}\n\n\t// Convert results\n\titems := make([]*types.ResultItem, 0, len(serpResp.OrganicResults))\n\n\t// Add answer box as first result if available\n\tif serpResp.AnswerBox != nil && serpResp.AnswerBox.Snippet != \"\" {\n\t\titems = append(items, &types.ResultItem{\n\t\t\tType: types.SearchTypeWeb,\n\t\t\tTitle: serpResp.AnswerBox.Title,\n\t\t\tContent: serpResp.AnswerBox.Snippet,\n\t\t\tURL: serpResp.AnswerBox.Link,\n\t\t\tScore: 1.0, // Featured snippet gets highest score\n\t\t\tSource: req.Source,\n\t\t\tMetadata: map[string]interface{}{\n\t\t\t\t\"type\": \"answer_box\",\n\t\t\t},\n\t\t})\n\t}\n\n\t// Add organic results\n\tfor _, r := range serpResp.OrganicResults {\n\t\t// Calculate score based on position (1st = 0.95, 2nd = 0.90, etc.)\n\t\tscore := 1.0 - float64(r.Position)*0.05\n\t\tif score < 0.1 {\n\t\t\tscore = 0.1\n\t\t}\n\n\t\titems = append(items, &types.ResultItem{\n\t\t\tType: types.SearchTypeWeb,\n\t\t\tTitle: r.Title,\n\t\t\tContent: r.Snippet,\n\t\t\tURL: r.Link,\n\t\t\tScore: score,\n\t\t\tSource: req.Source,\n\t\t})\n\t}\n\n\treturn &types.Result{\n\t\tType: types.SearchTypeWeb,\n\t\tQuery: req.Query,\n\t\tSource: req.Source,\n\t\tItems: items,\n\t\tTotal: len(items),\n\t\tDuration: time.Since(startTime).Milliseconds(),\n\t}, nil\n}\n\n// callAPI makes the HTTP GET request to SerpAPI\nfunc (p *SerpAPIProvider) callAPI(params url.Values) (*serpAPIResponse, error) {\n\t// Build URL with query parameters\n\treqURL := serpAPIURL + \"?\" + params.Encode()\n\n\t// Create HTTP request\n\thttpReq, err := http.NewRequest(http.MethodGet, reqURL, nil)\n\tif err != nil {\n\t\treturn nil, fmt.Errorf(\"failed to create request: %w\", err)\n\t}\n\n\t// Execute request\n\tclient := &http.Client{Timeout: serpAPITimeout}\n\tresp, err := client.Do(httpReq)\n\tif err != nil {\n\t\treturn nil, fmt.Errorf(\"request failed: %w\", err)\n\t}\n\tdefer resp.Body.Close()\n\n\t// Read response body\n\trespBody, err := io.ReadAll(resp.Body)\n\tif err != nil {\n\t\treturn nil, fmt.Errorf(\"failed to read response: %w\", err)\n\t}\n\n\t// Check status code\n\tif resp.StatusCode != http.StatusOK {\n\t\treturn nil, fmt.Errorf(\"API returned status %d: %s\", resp.StatusCode, string(respBody))\n\t}\n\n\t// Parse response\n\tvar serpResp serpAPIResponse\n\tif err := json.Unmarshal(respBody, &serpResp); err != nil {\n\t\treturn nil, fmt.Errorf(\"failed to parse response: %w\", err)\n\t}\n\n\treturn &serpResp, nil\n}\n\n// convertSerpAPITimeRange converts time range to SerpAPI tbs format\nfunc convertSerpAPITimeRange(timeRange string) string {\n\tswitch timeRange {\n\tcase \"hour\":\n\t\treturn \"qdr:h\"\n\tcase \"day\":\n\t\treturn \"qdr:d\"\n\tcase \"week\":\n\t\treturn \"qdr:w\"\n\tcase \"month\":\n\t\treturn \"qdr:m\"\n\tcase \"year\":\n\t\treturn \"qdr:y\"\n\tdefault:\n\t\treturn \"\"\n\t}\n}\n" }, { "path": "agent/search/handlers/web/serpapi_test.go", "content": "package web_test\n\nimport (\n\t\"testing\"\n\n\t\"github.com/stretchr/testify/assert\"\n\t\"github.com/stretchr/testify/require\"\n\t\"github.com/yaoapp/yao/agent/assistant\"\n\t\"github.com/yaoapp/yao/agent/search/handlers/web\"\n\t\"github.com/yaoapp/yao/agent/search/types\"\n\t\"github.com/yaoapp/yao/agent/testutils\"\n)\n\n// TestSerpAPIProviderWithAssistantConfig tests SerpAPIProvider using web-serpapi assistant config\nfunc TestSerpAPIProviderWithAssistantConfig(t *testing.T) {\n\tif testing.Short() {\n\t\tt.Skip(\"Skipping integration test\")\n\t}\n\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\t// Load the web-serpapi test assistant to get its config\n\tast, err := assistant.LoadPath(\"/assistants/tests/web-serpapi\")\n\trequire.NoError(t, err)\n\trequire.NotNil(t, ast)\n\trequire.NotNil(t, ast.Search)\n\trequire.NotNil(t, ast.Search.Web)\n\n\t// Verify assistant config\n\tassert.Equal(t, \"tests.web-serpapi\", ast.ID)\n\tassert.Equal(t, \"serpapi\", ast.Search.Web.Provider)\n\tassert.Equal(t, \"$ENV.SERPAPI_API_KEY\", ast.Search.Web.APIKeyEnv)\n\tassert.Equal(t, 10, ast.Search.Web.MaxResults)\n\n\t// Create SerpAPIProvider with assistant's web config\n\tprovider := web.NewSerpAPIProvider(ast.Search.Web)\n\trequire.NotNil(t, provider)\n\n\t// Execute search\n\treq := &types.Request{\n\t\tQuery: \"golang programming language\",\n\t\tType: types.SearchTypeWeb,\n\t\tSource: types.SourceAuto,\n\t\tLimit: 5,\n\t}\n\n\tresult, err := provider.Search(req)\n\trequire.NoError(t, err)\n\trequire.NotNil(t, result)\n\n\t// Verify result structure\n\tassert.Equal(t, types.SearchTypeWeb, result.Type)\n\tassert.Equal(t, \"golang programming language\", result.Query)\n\tassert.Equal(t, types.SourceAuto, result.Source)\n\n\t// API key must be valid - search should succeed\n\trequire.Empty(t, result.Error, \"Search should succeed with valid API key, got error: %s\", result.Error)\n\n\t// Verify we got results\n\tassert.Greater(t, result.Total, 0)\n\tassert.NotEmpty(t, result.Items)\n\tassert.Greater(t, result.Duration, int64(0))\n\n\t// Verify result item structure\n\tfor _, item := range result.Items {\n\t\tassert.Equal(t, types.SearchTypeWeb, item.Type)\n\t\tassert.Equal(t, types.SourceAuto, item.Source)\n\t\tassert.NotEmpty(t, item.Title)\n\t\tassert.NotEmpty(t, item.URL)\n\t\tassert.Greater(t, item.Score, 0.0)\n\t}\n\n\tt.Logf(\"Search returned %d results in %dms\", result.Total, result.Duration)\n}\n\n// TestSerpAPIProviderWithSiteRestriction tests SerpAPIProvider with domain restriction\nfunc TestSerpAPIProviderWithSiteRestriction(t *testing.T) {\n\tif testing.Short() {\n\t\tt.Skip(\"Skipping integration test\")\n\t}\n\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\t// Load the web-serpapi test assistant\n\tast, err := assistant.LoadPath(\"/assistants/tests/web-serpapi\")\n\trequire.NoError(t, err)\n\trequire.NotNil(t, ast.Search)\n\trequire.NotNil(t, ast.Search.Web)\n\n\t// Create SerpAPIProvider\n\tprovider := web.NewSerpAPIProvider(ast.Search.Web)\n\n\t// Execute search with site restriction\n\treq := &types.Request{\n\t\tQuery: \"documentation\",\n\t\tType: types.SearchTypeWeb,\n\t\tSource: types.SourceHook,\n\t\tSites: []string{\"github.com\"},\n\t\tLimit: 3,\n\t}\n\n\tresult, err := provider.Search(req)\n\trequire.NoError(t, err)\n\trequire.NotNil(t, result)\n\n\tassert.Equal(t, types.SearchTypeWeb, result.Type)\n\tassert.Equal(t, types.SourceHook, result.Source)\n\n\t// API key must be valid - search should succeed\n\trequire.Empty(t, result.Error, \"Search should succeed with valid API key, got error: %s\", result.Error)\n\trequire.NotEmpty(t, result.Items, \"Search should return results\")\n\n\t// All results should be from github.com\n\tfor _, item := range result.Items {\n\t\tassert.Contains(t, item.URL, \"github.com\", \"Result URL should be from github.com\")\n\t}\n\tt.Logf(\"Site-restricted search returned %d results from github.com\", result.Total)\n}\n\n// TestSerpAPIProviderWithMultipleSites tests SerpAPIProvider with multiple domain restrictions\nfunc TestSerpAPIProviderWithMultipleSites(t *testing.T) {\n\tif testing.Short() {\n\t\tt.Skip(\"Skipping integration test\")\n\t}\n\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\t// Load the web-serpapi test assistant\n\tast, err := assistant.LoadPath(\"/assistants/tests/web-serpapi\")\n\trequire.NoError(t, err)\n\trequire.NotNil(t, ast.Search)\n\trequire.NotNil(t, ast.Search.Web)\n\n\t// Create SerpAPIProvider\n\tprovider := web.NewSerpAPIProvider(ast.Search.Web)\n\n\t// Execute search with multiple site restrictions\n\treq := &types.Request{\n\t\tQuery: \"golang tutorial\",\n\t\tType: types.SearchTypeWeb,\n\t\tSource: types.SourceAuto,\n\t\tSites: []string{\"github.com\", \"golang.org\"},\n\t\tLimit: 5,\n\t}\n\n\tresult, err := provider.Search(req)\n\trequire.NoError(t, err)\n\trequire.NotNil(t, result)\n\n\t// API key must be valid - search should succeed\n\trequire.Empty(t, result.Error, \"Search should succeed with valid API key, got error: %s\", result.Error)\n\trequire.NotEmpty(t, result.Items, \"Search should return results\")\n\n\t// Results should be from either github.com or golang.org\n\tfor _, item := range result.Items {\n\t\tisValidSite := false\n\t\tfor _, site := range req.Sites {\n\t\t\tif containsSite(item.URL, site) {\n\t\t\t\tisValidSite = true\n\t\t\t\tbreak\n\t\t\t}\n\t\t}\n\t\tassert.True(t, isValidSite, \"Result URL should be from github.com or golang.org: %s\", item.URL)\n\t}\n\tt.Logf(\"Multi-site search returned %d results\", result.Total)\n}\n\n// TestSerpAPIProviderWithTimeRange tests SerpAPIProvider with time range filter\nfunc TestSerpAPIProviderWithTimeRange(t *testing.T) {\n\tif testing.Short() {\n\t\tt.Skip(\"Skipping integration test\")\n\t}\n\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\t// Load the web-serpapi test assistant\n\tast, err := assistant.LoadPath(\"/assistants/tests/web-serpapi\")\n\trequire.NoError(t, err)\n\trequire.NotNil(t, ast.Search)\n\trequire.NotNil(t, ast.Search.Web)\n\n\t// Create SerpAPIProvider\n\tprovider := web.NewSerpAPIProvider(ast.Search.Web)\n\n\t// Execute search with time range\n\treq := &types.Request{\n\t\tQuery: \"artificial intelligence news\",\n\t\tType: types.SearchTypeWeb,\n\t\tSource: types.SourceAuto,\n\t\tTimeRange: \"week\", // Last week\n\t\tLimit: 5,\n\t}\n\n\tresult, err := provider.Search(req)\n\trequire.NoError(t, err)\n\trequire.NotNil(t, result)\n\n\t// API key must be valid - search should succeed\n\trequire.Empty(t, result.Error, \"Search should succeed with valid API key, got error: %s\", result.Error)\n\tt.Logf(\"Time-ranged search (last week) returned %d results in %dms\", result.Total, result.Duration)\n}\n\n// TestSerpAPIProviderWithoutAPIKey tests graceful degradation when API key is missing\nfunc TestSerpAPIProviderWithoutAPIKey(t *testing.T) {\n\t// Create provider with nil config (no API key)\n\tprovider := web.NewSerpAPIProvider(nil)\n\trequire.NotNil(t, provider)\n\n\treq := &types.Request{\n\t\tQuery: \"test query\",\n\t\tType: types.SearchTypeWeb,\n\t\tSource: types.SourceAuto,\n\t}\n\n\tresult, err := provider.Search(req)\n\n\t// Should not return error, but result should have error message\n\trequire.NoError(t, err)\n\trequire.NotNil(t, result)\n\tassert.Equal(t, types.SearchTypeWeb, result.Type)\n\tassert.Equal(t, \"test query\", result.Query)\n\tassert.NotEmpty(t, result.Error)\n\tassert.Contains(t, result.Error, \"API key\")\n\tassert.Empty(t, result.Items)\n\tassert.Equal(t, 0, result.Total)\n}\n\n// TestSerpAPIProviderWithEmptyConfig tests provider with empty config\nfunc TestSerpAPIProviderWithEmptyConfig(t *testing.T) {\n\t// Create provider with empty config\n\tcfg := &types.WebConfig{}\n\tprovider := web.NewSerpAPIProvider(cfg)\n\trequire.NotNil(t, provider)\n\n\treq := &types.Request{\n\t\tQuery: \"test query\",\n\t\tType: types.SearchTypeWeb,\n\t\tSource: types.SourceUser,\n\t}\n\n\tresult, err := provider.Search(req)\n\n\trequire.NoError(t, err)\n\trequire.NotNil(t, result)\n\tassert.NotEmpty(t, result.Error)\n\tassert.Contains(t, result.Error, \"API key\")\n}\n\n// TestSerpAPIProviderMaxResults tests that max_results from config is respected\nfunc TestSerpAPIProviderMaxResults(t *testing.T) {\n\tif testing.Short() {\n\t\tt.Skip(\"Skipping integration test\")\n\t}\n\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\t// Load the web-serpapi test assistant\n\tast, err := assistant.LoadPath(\"/assistants/tests/web-serpapi\")\n\trequire.NoError(t, err)\n\trequire.NotNil(t, ast.Search)\n\trequire.NotNil(t, ast.Search.Web)\n\n\t// Create SerpAPIProvider\n\tprovider := web.NewSerpAPIProvider(ast.Search.Web)\n\n\t// Execute search without limit (should use config's max_results)\n\treq := &types.Request{\n\t\tQuery: \"machine learning\",\n\t\tType: types.SearchTypeWeb,\n\t\tSource: types.SourceAuto,\n\t\t// No Limit set, should use config's max_results (10)\n\t}\n\n\tresult, err := provider.Search(req)\n\trequire.NoError(t, err)\n\trequire.NotNil(t, result)\n\n\t// API key must be valid - search should succeed\n\trequire.Empty(t, result.Error, \"Search should succeed with valid API key, got error: %s\", result.Error)\n\n\t// Should respect max_results from config (+1 for possible answer box)\n\tassert.LessOrEqual(t, result.Total, ast.Search.Web.MaxResults+1)\n\tt.Logf(\"Search without limit returned %d results (max: %d)\", result.Total, ast.Search.Web.MaxResults)\n\n\t// Execute search with explicit limit\n\treq2 := &types.Request{\n\t\tQuery: \"machine learning\",\n\t\tType: types.SearchTypeWeb,\n\t\tSource: types.SourceAuto,\n\t\tLimit: 3, // Override config's max_results\n\t}\n\n\tresult2, err := provider.Search(req2)\n\trequire.NoError(t, err)\n\trequire.NotNil(t, result2)\n\n\t// API key must be valid - search should succeed\n\trequire.Empty(t, result2.Error, \"Search should succeed with valid API key, got error: %s\", result2.Error)\n\n\t// Should respect request's limit (+1 for possible answer box)\n\tassert.LessOrEqual(t, result2.Total, 4)\n\tt.Logf(\"Search with limit=3 returned %d results\", result2.Total)\n}\n\n// TestSerpAPIProviderWithBingEngine tests SerpAPIProvider with Bing search engine\nfunc TestSerpAPIProviderWithBingEngine(t *testing.T) {\n\tif testing.Short() {\n\t\tt.Skip(\"Skipping integration test\")\n\t}\n\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\t// Load the web-serpapi test assistant to get base config\n\tast, err := assistant.LoadPath(\"/assistants/tests/web-serpapi\")\n\trequire.NoError(t, err)\n\trequire.NotNil(t, ast.Search)\n\trequire.NotNil(t, ast.Search.Web)\n\n\t// Create config with Bing engine\n\tbingConfig := &types.WebConfig{\n\t\tProvider: \"serpapi\",\n\t\tAPIKeyEnv: ast.Search.Web.APIKeyEnv,\n\t\tMaxResults: 5,\n\t\tEngine: \"bing\",\n\t}\n\n\t// Create SerpAPIProvider with Bing engine\n\tprovider := web.NewSerpAPIProvider(bingConfig)\n\trequire.NotNil(t, provider)\n\n\t// Execute search\n\treq := &types.Request{\n\t\tQuery: \"Golang programming\",\n\t\tType: types.SearchTypeWeb,\n\t\tSource: types.SourceAuto,\n\t\tLimit: 5,\n\t}\n\n\tresult, err := provider.Search(req)\n\trequire.NoError(t, err)\n\trequire.NotNil(t, result)\n\n\t// Verify result structure\n\tassert.Equal(t, types.SearchTypeWeb, result.Type)\n\tassert.Equal(t, \"Golang programming\", result.Query)\n\n\t// API key must be valid - search should succeed\n\trequire.Empty(t, result.Error, \"Bing search should succeed with valid API key, got error: %s\", result.Error)\n\n\t// Verify we got results\n\tassert.Greater(t, result.Total, 0)\n\tassert.NotEmpty(t, result.Items)\n\n\tt.Logf(\"Bing search returned %d results in %dms\", result.Total, result.Duration)\n}\n\n// TestSerpAPIProviderEngineDefault tests that default engine is Google\nfunc TestSerpAPIProviderEngineDefault(t *testing.T) {\n\t// Create provider with config that has no engine specified\n\tcfg := &types.WebConfig{\n\t\tProvider: \"serpapi\",\n\t\tAPIKeyEnv: \"SERPAPI_API_KEY\",\n\t\tMaxResults: 10,\n\t\t// Engine not set - should default to \"google\"\n\t}\n\n\tprovider := web.NewSerpAPIProvider(cfg)\n\trequire.NotNil(t, provider)\n\n\t// We can't directly check the engine field since it's private,\n\t// but we verify the provider is created successfully\n\t// The actual engine usage is tested in integration tests\n}\n\n// containsSite checks if url contains the site domain\nfunc containsSite(url, site string) bool {\n\treturn len(url) >= len(site) && containsHelper(url, site)\n}\n\n// containsHelper is a helper function for string containment check\nfunc containsHelper(s, substr string) bool {\n\tfor i := 0; i <= len(s)-len(substr); i++ {\n\t\tif s[i:i+len(substr)] == substr {\n\t\t\treturn true\n\t\t}\n\t}\n\treturn false\n}\n" }, { "path": "agent/search/handlers/web/serper.go", "content": "package web\n\nimport (\n\t\"bytes\"\n\t\"encoding/json\"\n\t\"fmt\"\n\t\"io\"\n\t\"net/http\"\n\t\"os\"\n\t\"time\"\n\n\t\"github.com/yaoapp/yao/agent/search/types\"\n)\n\nconst (\n\tserperAPIURL = \"https://google.serper.dev/search\"\n\tserperAPITimeout = 30 * time.Second\n)\n\n// SerperProvider implements web search using Serper API (serper.dev)\ntype SerperProvider struct {\n\tapiKey string\n\tmaxResults int\n}\n\n// NewSerperProvider creates a new Serper provider\nfunc NewSerperProvider(cfg *types.WebConfig) *SerperProvider {\n\tapiKey := \"\"\n\tif cfg != nil && cfg.APIKeyEnv != \"\" {\n\t\t// Support both \"$ENV.VAR_NAME\" and \"VAR_NAME\" formats\n\t\tenvName := cfg.APIKeyEnv\n\t\tif len(envName) > 5 && envName[:5] == \"$ENV.\" {\n\t\t\tenvName = envName[5:]\n\t\t}\n\t\tapiKey = os.Getenv(envName)\n\t}\n\n\tmaxResults := 10\n\tif cfg != nil && cfg.MaxResults > 0 {\n\t\tmaxResults = cfg.MaxResults\n\t}\n\n\treturn &SerperProvider{\n\t\tapiKey: apiKey,\n\t\tmaxResults: maxResults,\n\t}\n}\n\n// serperRequest represents the request body for Serper API\ntype serperRequest struct {\n\tQ string `json:\"q\"` // Search query\n\tNum int `json:\"num,omitempty\"` // Number of results (default: 10, max: 100)\n\tGL string `json:\"gl,omitempty\"` // Country code (e.g., \"us\", \"cn\")\n\tHL string `json:\"hl,omitempty\"` // Language code (e.g., \"en\", \"zh-cn\")\n\tTBS string `json:\"tbs,omitempty\"` // Time-based search (qdr:h, qdr:d, qdr:w, qdr:m, qdr:y)\n\tPage int `json:\"page,omitempty\"` // Page number (default: 1)\n\tAutoCor bool `json:\"autocorrect\"` // Auto-correct spelling\n}\n\n// serperResponse represents the response from Serper API\ntype serperResponse struct {\n\tSearchParameters serperSearchParams `json:\"searchParameters\"`\n\tOrganic []serperResult `json:\"organic\"`\n\tAnswerBox *serperAnswerBox `json:\"answerBox,omitempty\"`\n\tKnowledgeGraph *serperKnowledge `json:\"knowledgeGraph,omitempty\"`\n\tRelatedSearches []serperRelated `json:\"relatedSearches,omitempty\"`\n}\n\n// serperSearchParams contains search parameters from response\ntype serperSearchParams struct {\n\tQ string `json:\"q\"`\n\tType string `json:\"type\"`\n\tGL string `json:\"gl\"`\n\tHL string `json:\"hl\"`\n\tNum int `json:\"num\"`\n}\n\n// serperResult represents a single organic search result\ntype serperResult struct {\n\tTitle string `json:\"title\"`\n\tLink string `json:\"link\"`\n\tSnippet string `json:\"snippet\"`\n\tPosition int `json:\"position\"`\n\tDate string `json:\"date,omitempty\"`\n}\n\n// serperAnswerBox represents the answer box (featured snippet)\ntype serperAnswerBox struct {\n\tTitle string `json:\"title,omitempty\"`\n\tSnippet string `json:\"snippet,omitempty\"`\n\tLink string `json:\"link,omitempty\"`\n}\n\n// serperKnowledge represents knowledge graph data\ntype serperKnowledge struct {\n\tTitle string `json:\"title,omitempty\"`\n\tType string `json:\"type,omitempty\"`\n\tDescription string `json:\"description,omitempty\"`\n}\n\n// serperRelated represents related searches\ntype serperRelated struct {\n\tQuery string `json:\"query\"`\n}\n\n// Search executes a web search using Serper API\nfunc (p *SerperProvider) Search(req *types.Request) (*types.Result, error) {\n\tstartTime := time.Now()\n\n\t// Validate API key\n\tif p.apiKey == \"\" {\n\t\treturn &types.Result{\n\t\t\tType: types.SearchTypeWeb,\n\t\t\tQuery: req.Query,\n\t\t\tSource: req.Source,\n\t\t\tItems: []*types.ResultItem{},\n\t\t\tTotal: 0,\n\t\t\tError: \"Serper API key not configured\",\n\t\t}, nil\n\t}\n\n\t// Determine max results\n\tmaxResults := p.maxResults\n\tif req.Limit > 0 {\n\t\tmaxResults = req.Limit\n\t}\n\n\t// Build search query with site restrictions if specified\n\tquery := req.Query\n\tif len(req.Sites) > 0 {\n\t\t// Serper uses \"site:domain\" syntax in query\n\t\tif len(req.Sites) == 1 {\n\t\t\tquery = \"site:\" + req.Sites[0] + \" \" + req.Query\n\t\t} else {\n\t\t\t// Multiple sites: (site:domain1 OR site:domain2) query\n\t\t\tsiteQuery := \"\"\n\t\t\tfor i, site := range req.Sites {\n\t\t\t\tif i > 0 {\n\t\t\t\t\tsiteQuery += \" OR \"\n\t\t\t\t}\n\t\t\t\tsiteQuery += \"site:\" + site\n\t\t\t}\n\t\t\tquery = \"(\" + siteQuery + \") \" + req.Query\n\t\t}\n\t}\n\n\t// Build request body\n\tserperReq := serperRequest{\n\t\tQ: query,\n\t\tNum: maxResults,\n\t\tAutoCor: true,\n\t}\n\n\t// Add time range if specified\n\tif req.TimeRange != \"\" {\n\t\tserperReq.TBS = convertSerperTimeRange(req.TimeRange)\n\t}\n\n\t// Execute API call\n\tserperResp, err := p.callAPI(&serperReq)\n\tif err != nil {\n\t\treturn &types.Result{\n\t\t\tType: types.SearchTypeWeb,\n\t\t\tQuery: req.Query,\n\t\t\tSource: req.Source,\n\t\t\tItems: []*types.ResultItem{},\n\t\t\tTotal: 0,\n\t\t\tDuration: time.Since(startTime).Milliseconds(),\n\t\t\tError: fmt.Sprintf(\"Serper API error: %v\", err),\n\t\t}, nil\n\t}\n\n\t// Convert results\n\titems := make([]*types.ResultItem, 0, len(serperResp.Organic))\n\n\t// Add answer box as first result if available\n\tif serperResp.AnswerBox != nil && serperResp.AnswerBox.Snippet != \"\" {\n\t\titems = append(items, &types.ResultItem{\n\t\t\tType: types.SearchTypeWeb,\n\t\t\tTitle: serperResp.AnswerBox.Title,\n\t\t\tContent: serperResp.AnswerBox.Snippet,\n\t\t\tURL: serperResp.AnswerBox.Link,\n\t\t\tScore: 1.0, // Featured snippet gets highest score\n\t\t\tSource: req.Source,\n\t\t\tMetadata: map[string]interface{}{\n\t\t\t\t\"type\": \"answer_box\",\n\t\t\t},\n\t\t})\n\t}\n\n\t// Add organic results\n\tfor _, r := range serperResp.Organic {\n\t\t// Calculate score based on position (1st = 0.95, 2nd = 0.90, etc.)\n\t\tscore := 1.0 - float64(r.Position)*0.05\n\t\tif score < 0.1 {\n\t\t\tscore = 0.1\n\t\t}\n\n\t\titems = append(items, &types.ResultItem{\n\t\t\tType: types.SearchTypeWeb,\n\t\t\tTitle: r.Title,\n\t\t\tContent: r.Snippet,\n\t\t\tURL: r.Link,\n\t\t\tScore: score,\n\t\t\tSource: req.Source,\n\t\t})\n\t}\n\n\treturn &types.Result{\n\t\tType: types.SearchTypeWeb,\n\t\tQuery: req.Query,\n\t\tSource: req.Source,\n\t\tItems: items,\n\t\tTotal: len(items),\n\t\tDuration: time.Since(startTime).Milliseconds(),\n\t}, nil\n}\n\n// callAPI makes the HTTP POST request to Serper API\nfunc (p *SerperProvider) callAPI(req *serperRequest) (*serperResponse, error) {\n\t// Serialize request body\n\tbody, err := json.Marshal(req)\n\tif err != nil {\n\t\treturn nil, fmt.Errorf(\"failed to marshal request: %w\", err)\n\t}\n\n\t// Create HTTP request\n\thttpReq, err := http.NewRequest(http.MethodPost, serperAPIURL, bytes.NewReader(body))\n\tif err != nil {\n\t\treturn nil, fmt.Errorf(\"failed to create request: %w\", err)\n\t}\n\n\thttpReq.Header.Set(\"Content-Type\", \"application/json\")\n\thttpReq.Header.Set(\"X-API-KEY\", p.apiKey)\n\n\t// Execute request\n\tclient := &http.Client{Timeout: serperAPITimeout}\n\tresp, err := client.Do(httpReq)\n\tif err != nil {\n\t\treturn nil, fmt.Errorf(\"request failed: %w\", err)\n\t}\n\tdefer resp.Body.Close()\n\n\t// Read response body\n\trespBody, err := io.ReadAll(resp.Body)\n\tif err != nil {\n\t\treturn nil, fmt.Errorf(\"failed to read response: %w\", err)\n\t}\n\n\t// Check status code\n\tif resp.StatusCode != http.StatusOK {\n\t\treturn nil, fmt.Errorf(\"API returned status %d: %s\", resp.StatusCode, string(respBody))\n\t}\n\n\t// Parse response\n\tvar serperResp serperResponse\n\tif err := json.Unmarshal(respBody, &serperResp); err != nil {\n\t\treturn nil, fmt.Errorf(\"failed to parse response: %w\", err)\n\t}\n\n\treturn &serperResp, nil\n}\n\n// convertSerperTimeRange converts time range to Serper tbs format\nfunc convertSerperTimeRange(timeRange string) string {\n\tswitch timeRange {\n\tcase \"hour\":\n\t\treturn \"qdr:h\"\n\tcase \"day\":\n\t\treturn \"qdr:d\"\n\tcase \"week\":\n\t\treturn \"qdr:w\"\n\tcase \"month\":\n\t\treturn \"qdr:m\"\n\tcase \"year\":\n\t\treturn \"qdr:y\"\n\tdefault:\n\t\treturn \"\"\n\t}\n}\n" }, { "path": "agent/search/handlers/web/serper_test.go", "content": "package web_test\n\nimport (\n\t\"os\"\n\t\"testing\"\n\n\t\"github.com/stretchr/testify/assert\"\n\t\"github.com/stretchr/testify/require\"\n\t\"github.com/yaoapp/yao/agent/assistant\"\n\t\"github.com/yaoapp/yao/agent/search/handlers/web\"\n\t\"github.com/yaoapp/yao/agent/search/types\"\n\t\"github.com/yaoapp/yao/agent/testutils\"\n)\n\n// skipIfNoSerperKey skips the test if SERPER_API_KEY is not set\n// Note: Serper (serper.dev) requires registration at https://serper.dev\nfunc skipIfNoSerperKey(t *testing.T) {\n\tif os.Getenv(\"SERPER_API_KEY\") == \"\" {\n\t\tt.Skip(\"Skipping Serper test: SERPER_API_KEY not set. Register at https://serper.dev for free 2500 queries.\")\n\t}\n}\n\n// TestSerperProviderWithAssistantConfig tests SerperProvider using web-serper assistant config\nfunc TestSerperProviderWithAssistantConfig(t *testing.T) {\n\tif testing.Short() {\n\t\tt.Skip(\"Skipping integration test\")\n\t}\n\tskipIfNoSerperKey(t)\n\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\t// Load the web-serper test assistant to get its config\n\tast, err := assistant.LoadPath(\"/assistants/tests/web-serper\")\n\trequire.NoError(t, err)\n\trequire.NotNil(t, ast)\n\trequire.NotNil(t, ast.Search)\n\trequire.NotNil(t, ast.Search.Web)\n\n\t// Verify assistant config\n\tassert.Equal(t, \"tests.web-serper\", ast.ID)\n\tassert.Equal(t, \"serper\", ast.Search.Web.Provider)\n\tassert.Equal(t, \"$ENV.SERPER_API_KEY\", ast.Search.Web.APIKeyEnv)\n\tassert.Equal(t, 10, ast.Search.Web.MaxResults)\n\n\t// Create SerperProvider with assistant's web config\n\tprovider := web.NewSerperProvider(ast.Search.Web)\n\trequire.NotNil(t, provider)\n\n\t// Execute search\n\treq := &types.Request{\n\t\tQuery: \"Yao App Engine\",\n\t\tType: types.SearchTypeWeb,\n\t\tSource: types.SourceAuto,\n\t\tLimit: 5,\n\t}\n\n\tresult, err := provider.Search(req)\n\trequire.NoError(t, err)\n\trequire.NotNil(t, result)\n\n\t// Verify result structure\n\tassert.Equal(t, types.SearchTypeWeb, result.Type)\n\tassert.Equal(t, \"Yao App Engine\", result.Query)\n\tassert.Equal(t, types.SourceAuto, result.Source)\n\n\t// API key must be valid - search should succeed\n\trequire.Empty(t, result.Error, \"Search should succeed with valid API key, got error: %s\", result.Error)\n\n\t// Verify we got results\n\tassert.Greater(t, result.Total, 0)\n\tassert.NotEmpty(t, result.Items)\n\tassert.Greater(t, result.Duration, int64(0))\n\n\t// Verify result item structure\n\tfor _, item := range result.Items {\n\t\tassert.Equal(t, types.SearchTypeWeb, item.Type)\n\t\tassert.Equal(t, types.SourceAuto, item.Source)\n\t\tassert.NotEmpty(t, item.Title)\n\t\tassert.NotEmpty(t, item.URL)\n\t\tassert.Greater(t, item.Score, 0.0)\n\t}\n\n\tt.Logf(\"Search returned %d results in %dms\", result.Total, result.Duration)\n}\n\n// TestSerperProviderWithSiteRestriction tests SerperProvider with domain restriction\nfunc TestSerperProviderWithSiteRestriction(t *testing.T) {\n\tif testing.Short() {\n\t\tt.Skip(\"Skipping integration test\")\n\t}\n\tskipIfNoSerperKey(t)\n\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\t// Load the web-serper test assistant\n\tast, err := assistant.LoadPath(\"/assistants/tests/web-serper\")\n\trequire.NoError(t, err)\n\trequire.NotNil(t, ast.Search)\n\trequire.NotNil(t, ast.Search.Web)\n\n\t// Create SerperProvider\n\tprovider := web.NewSerperProvider(ast.Search.Web)\n\n\t// Execute search with site restriction\n\treq := &types.Request{\n\t\tQuery: \"documentation\",\n\t\tType: types.SearchTypeWeb,\n\t\tSource: types.SourceHook,\n\t\tSites: []string{\"github.com\"},\n\t\tLimit: 3,\n\t}\n\n\tresult, err := provider.Search(req)\n\trequire.NoError(t, err)\n\trequire.NotNil(t, result)\n\n\tassert.Equal(t, types.SearchTypeWeb, result.Type)\n\tassert.Equal(t, types.SourceHook, result.Source)\n\n\t// API key must be valid - search should succeed\n\trequire.Empty(t, result.Error, \"Search should succeed with valid API key, got error: %s\", result.Error)\n\trequire.NotEmpty(t, result.Items, \"Search should return results\")\n\n\t// All results should be from github.com\n\tfor _, item := range result.Items {\n\t\tassert.Contains(t, item.URL, \"github.com\", \"Result URL should be from github.com\")\n\t}\n\tt.Logf(\"Site-restricted search returned %d results from github.com\", result.Total)\n}\n\n// TestSerperProviderWithMultipleSites tests SerperProvider with multiple domain restrictions\nfunc TestSerperProviderWithMultipleSites(t *testing.T) {\n\tif testing.Short() {\n\t\tt.Skip(\"Skipping integration test\")\n\t}\n\tskipIfNoSerperKey(t)\n\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\t// Load the web-serper test assistant\n\tast, err := assistant.LoadPath(\"/assistants/tests/web-serper\")\n\trequire.NoError(t, err)\n\trequire.NotNil(t, ast.Search)\n\trequire.NotNil(t, ast.Search.Web)\n\n\t// Create SerperProvider\n\tprovider := web.NewSerperProvider(ast.Search.Web)\n\n\t// Execute search with multiple site restrictions\n\treq := &types.Request{\n\t\tQuery: \"golang tutorial\",\n\t\tType: types.SearchTypeWeb,\n\t\tSource: types.SourceAuto,\n\t\tSites: []string{\"github.com\", \"golang.org\"},\n\t\tLimit: 5,\n\t}\n\n\tresult, err := provider.Search(req)\n\trequire.NoError(t, err)\n\trequire.NotNil(t, result)\n\n\t// API key must be valid - search should succeed\n\trequire.Empty(t, result.Error, \"Search should succeed with valid API key, got error: %s\", result.Error)\n\trequire.NotEmpty(t, result.Items, \"Search should return results\")\n\n\t// Results should be from either github.com or golang.org\n\tfor _, item := range result.Items {\n\t\tisValidSite := false\n\t\tfor _, site := range req.Sites {\n\t\t\tif contains(item.URL, site) {\n\t\t\t\tisValidSite = true\n\t\t\t\tbreak\n\t\t\t}\n\t\t}\n\t\tassert.True(t, isValidSite, \"Result URL should be from github.com or golang.org: %s\", item.URL)\n\t}\n\tt.Logf(\"Multi-site search returned %d results\", result.Total)\n}\n\n// TestSerperProviderWithTimeRange tests SerperProvider with time range filter\nfunc TestSerperProviderWithTimeRange(t *testing.T) {\n\tif testing.Short() {\n\t\tt.Skip(\"Skipping integration test\")\n\t}\n\tskipIfNoSerperKey(t)\n\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\t// Load the web-serper test assistant\n\tast, err := assistant.LoadPath(\"/assistants/tests/web-serper\")\n\trequire.NoError(t, err)\n\trequire.NotNil(t, ast.Search)\n\trequire.NotNil(t, ast.Search.Web)\n\n\t// Create SerperProvider\n\tprovider := web.NewSerperProvider(ast.Search.Web)\n\n\t// Execute search with time range\n\treq := &types.Request{\n\t\tQuery: \"artificial intelligence news\",\n\t\tType: types.SearchTypeWeb,\n\t\tSource: types.SourceAuto,\n\t\tTimeRange: \"week\", // Last week\n\t\tLimit: 5,\n\t}\n\n\tresult, err := provider.Search(req)\n\trequire.NoError(t, err)\n\trequire.NotNil(t, result)\n\n\t// API key must be valid - search should succeed\n\trequire.Empty(t, result.Error, \"Search should succeed with valid API key, got error: %s\", result.Error)\n\tt.Logf(\"Time-ranged search (last week) returned %d results in %dms\", result.Total, result.Duration)\n}\n\n// TestSerperProviderWithoutAPIKey tests graceful degradation when API key is missing\nfunc TestSerperProviderWithoutAPIKey(t *testing.T) {\n\t// Create provider with nil config (no API key)\n\tprovider := web.NewSerperProvider(nil)\n\trequire.NotNil(t, provider)\n\n\treq := &types.Request{\n\t\tQuery: \"test query\",\n\t\tType: types.SearchTypeWeb,\n\t\tSource: types.SourceAuto,\n\t}\n\n\tresult, err := provider.Search(req)\n\n\t// Should not return error, but result should have error message\n\trequire.NoError(t, err)\n\trequire.NotNil(t, result)\n\tassert.Equal(t, types.SearchTypeWeb, result.Type)\n\tassert.Equal(t, \"test query\", result.Query)\n\tassert.NotEmpty(t, result.Error)\n\tassert.Contains(t, result.Error, \"API key\")\n\tassert.Empty(t, result.Items)\n\tassert.Equal(t, 0, result.Total)\n}\n\n// TestSerperProviderWithEmptyConfig tests provider with empty config\nfunc TestSerperProviderWithEmptyConfig(t *testing.T) {\n\t// Create provider with empty config\n\tcfg := &types.WebConfig{}\n\tprovider := web.NewSerperProvider(cfg)\n\trequire.NotNil(t, provider)\n\n\treq := &types.Request{\n\t\tQuery: \"test query\",\n\t\tType: types.SearchTypeWeb,\n\t\tSource: types.SourceUser,\n\t}\n\n\tresult, err := provider.Search(req)\n\n\trequire.NoError(t, err)\n\trequire.NotNil(t, result)\n\tassert.NotEmpty(t, result.Error)\n\tassert.Contains(t, result.Error, \"API key\")\n}\n\n// TestSerperProviderMaxResults tests that max_results from config is respected\nfunc TestSerperProviderMaxResults(t *testing.T) {\n\tif testing.Short() {\n\t\tt.Skip(\"Skipping integration test\")\n\t}\n\tskipIfNoSerperKey(t)\n\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\t// Load the web-serper test assistant\n\tast, err := assistant.LoadPath(\"/assistants/tests/web-serper\")\n\trequire.NoError(t, err)\n\trequire.NotNil(t, ast.Search)\n\trequire.NotNil(t, ast.Search.Web)\n\n\t// Create SerperProvider\n\tprovider := web.NewSerperProvider(ast.Search.Web)\n\n\t// Execute search without limit (should use config's max_results)\n\treq := &types.Request{\n\t\tQuery: \"machine learning\",\n\t\tType: types.SearchTypeWeb,\n\t\tSource: types.SourceAuto,\n\t\t// No Limit set, should use config's max_results (10)\n\t}\n\n\tresult, err := provider.Search(req)\n\trequire.NoError(t, err)\n\trequire.NotNil(t, result)\n\n\t// API key must be valid - search should succeed\n\trequire.Empty(t, result.Error, \"Search should succeed with valid API key, got error: %s\", result.Error)\n\n\t// Should respect max_results from config (+1 for possible answer box)\n\tassert.LessOrEqual(t, result.Total, ast.Search.Web.MaxResults+1)\n\tt.Logf(\"Search without limit returned %d results (max: %d)\", result.Total, ast.Search.Web.MaxResults)\n\n\t// Execute search with explicit limit\n\treq2 := &types.Request{\n\t\tQuery: \"machine learning\",\n\t\tType: types.SearchTypeWeb,\n\t\tSource: types.SourceAuto,\n\t\tLimit: 3, // Override config's max_results\n\t}\n\n\tresult2, err := provider.Search(req2)\n\trequire.NoError(t, err)\n\trequire.NotNil(t, result2)\n\n\t// API key must be valid - search should succeed\n\trequire.Empty(t, result2.Error, \"Search should succeed with valid API key, got error: %s\", result2.Error)\n\n\t// Should respect request's limit (+1 for possible answer box)\n\tassert.LessOrEqual(t, result2.Total, 4)\n\tt.Logf(\"Search with limit=3 returned %d results\", result2.Total)\n}\n\n// contains checks if s contains substr (uses containsSite from serpapi_test.go)\nfunc contains(s, substr string) bool {\n\treturn containsSite(s, substr)\n}\n" }, { "path": "agent/search/handlers/web/tavily.go", "content": "package web\n\nimport (\n\t\"bytes\"\n\t\"encoding/json\"\n\t\"fmt\"\n\t\"io\"\n\t\"net/http\"\n\t\"os\"\n\t\"time\"\n\n\t\"github.com/yaoapp/yao/agent/search/types\"\n)\n\nconst (\n\ttavilyAPIURL = \"https://api.tavily.com/search\"\n\ttavilyAPITimeout = 30 * time.Second\n)\n\n// TavilyProvider implements web search using Tavily API\ntype TavilyProvider struct {\n\tapiKey string\n\tmaxResults int\n}\n\n// NewTavilyProvider creates a new Tavily provider\nfunc NewTavilyProvider(cfg *types.WebConfig) *TavilyProvider {\n\tapiKey := \"\"\n\tif cfg != nil && cfg.APIKeyEnv != \"\" {\n\t\t// Support both \"$ENV.VAR_NAME\" and \"VAR_NAME\" formats\n\t\tenvName := cfg.APIKeyEnv\n\t\tif len(envName) > 5 && envName[:5] == \"$ENV.\" {\n\t\t\tenvName = envName[5:]\n\t\t}\n\t\tapiKey = os.Getenv(envName)\n\t}\n\n\tmaxResults := 10\n\tif cfg != nil && cfg.MaxResults > 0 {\n\t\tmaxResults = cfg.MaxResults\n\t}\n\n\treturn &TavilyProvider{\n\t\tapiKey: apiKey,\n\t\tmaxResults: maxResults,\n\t}\n}\n\n// tavilyRequest represents the request body for Tavily API\ntype tavilyRequest struct {\n\tAPIKey string `json:\"api_key\"`\n\tQuery string `json:\"query\"`\n\tSearchDepth string `json:\"search_depth,omitempty\"` // \"basic\" or \"advanced\"\n\tIncludeAnswer bool `json:\"include_answer,omitempty\"` // Include AI-generated answer\n\tIncludeRawContent bool `json:\"include_raw_content,omitempty\"` // Include raw HTML content\n\tMaxResults int `json:\"max_results,omitempty\"` // Max number of results\n\tIncludeDomains []string `json:\"include_domains,omitempty\"` // Limit to specific domains\n\tExcludeDomains []string `json:\"exclude_domains,omitempty\"` // Exclude specific domains\n}\n\n// tavilyResponse represents the response from Tavily API\ntype tavilyResponse struct {\n\tQuery string `json:\"query\"`\n\tAnswer string `json:\"answer,omitempty\"`\n\tResults []tavilyResult `json:\"results\"`\n}\n\n// tavilyResult represents a single search result from Tavily\ntype tavilyResult struct {\n\tTitle string `json:\"title\"`\n\tURL string `json:\"url\"`\n\tContent string `json:\"content\"`\n\tScore float64 `json:\"score\"`\n\tRawContent string `json:\"raw_content,omitempty\"`\n}\n\n// Search executes a web search using Tavily API\nfunc (p *TavilyProvider) Search(req *types.Request) (*types.Result, error) {\n\tstartTime := time.Now()\n\n\t// Validate API key\n\tif p.apiKey == \"\" {\n\t\treturn &types.Result{\n\t\t\tType: types.SearchTypeWeb,\n\t\t\tQuery: req.Query,\n\t\t\tSource: req.Source,\n\t\t\tItems: []*types.ResultItem{},\n\t\t\tTotal: 0,\n\t\t\tError: \"Tavily API key not configured\",\n\t\t}, nil\n\t}\n\n\t// Determine max results\n\tmaxResults := p.maxResults\n\tif req.Limit > 0 {\n\t\tmaxResults = req.Limit\n\t}\n\n\t// Build request body\n\ttavilyReq := tavilyRequest{\n\t\tAPIKey: p.apiKey,\n\t\tQuery: req.Query,\n\t\tSearchDepth: \"basic\",\n\t\tIncludeAnswer: false,\n\t\tMaxResults: maxResults,\n\t}\n\n\t// Add domain restrictions if specified\n\tif len(req.Sites) > 0 {\n\t\ttavilyReq.IncludeDomains = req.Sites\n\t}\n\n\t// Execute API call\n\ttavilyResp, err := p.callAPI(&tavilyReq)\n\tif err != nil {\n\t\treturn &types.Result{\n\t\t\tType: types.SearchTypeWeb,\n\t\t\tQuery: req.Query,\n\t\t\tSource: req.Source,\n\t\t\tItems: []*types.ResultItem{},\n\t\t\tTotal: 0,\n\t\t\tDuration: time.Since(startTime).Milliseconds(),\n\t\t\tError: fmt.Sprintf(\"Tavily API error: %v\", err),\n\t\t}, nil\n\t}\n\n\t// Convert results\n\titems := make([]*types.ResultItem, 0, len(tavilyResp.Results))\n\tfor _, r := range tavilyResp.Results {\n\t\titems = append(items, &types.ResultItem{\n\t\t\tType: types.SearchTypeWeb,\n\t\t\tTitle: r.Title,\n\t\t\tContent: r.Content,\n\t\t\tURL: r.URL,\n\t\t\tScore: r.Score,\n\t\t\tSource: req.Source,\n\t\t})\n\t}\n\n\treturn &types.Result{\n\t\tType: types.SearchTypeWeb,\n\t\tQuery: req.Query,\n\t\tSource: req.Source,\n\t\tItems: items,\n\t\tTotal: len(items),\n\t\tDuration: time.Since(startTime).Milliseconds(),\n\t}, nil\n}\n\n// callAPI makes the HTTP request to Tavily API\nfunc (p *TavilyProvider) callAPI(req *tavilyRequest) (*tavilyResponse, error) {\n\t// Serialize request body\n\tbody, err := json.Marshal(req)\n\tif err != nil {\n\t\treturn nil, fmt.Errorf(\"failed to marshal request: %w\", err)\n\t}\n\n\t// Create HTTP request\n\thttpReq, err := http.NewRequest(http.MethodPost, tavilyAPIURL, bytes.NewReader(body))\n\tif err != nil {\n\t\treturn nil, fmt.Errorf(\"failed to create request: %w\", err)\n\t}\n\n\thttpReq.Header.Set(\"Content-Type\", \"application/json\")\n\n\t// Execute request\n\tclient := &http.Client{Timeout: tavilyAPITimeout}\n\tresp, err := client.Do(httpReq)\n\tif err != nil {\n\t\treturn nil, fmt.Errorf(\"request failed: %w\", err)\n\t}\n\tdefer resp.Body.Close()\n\n\t// Read response body\n\trespBody, err := io.ReadAll(resp.Body)\n\tif err != nil {\n\t\treturn nil, fmt.Errorf(\"failed to read response: %w\", err)\n\t}\n\n\t// Check status code\n\tif resp.StatusCode != http.StatusOK {\n\t\treturn nil, fmt.Errorf(\"API returned status %d: %s\", resp.StatusCode, string(respBody))\n\t}\n\n\t// Parse response\n\tvar tavilyResp tavilyResponse\n\tif err := json.Unmarshal(respBody, &tavilyResp); err != nil {\n\t\treturn nil, fmt.Errorf(\"failed to parse response: %w\", err)\n\t}\n\n\treturn &tavilyResp, nil\n}\n" }, { "path": "agent/search/handlers/web/tavily_test.go", "content": "package web_test\n\nimport (\n\t\"testing\"\n\n\t\"github.com/stretchr/testify/assert\"\n\t\"github.com/stretchr/testify/require\"\n\t\"github.com/yaoapp/yao/agent/assistant\"\n\t\"github.com/yaoapp/yao/agent/search/handlers/web\"\n\t\"github.com/yaoapp/yao/agent/search/types\"\n\t\"github.com/yaoapp/yao/agent/testutils\"\n)\n\n// TestTavilyProviderWithAssistantConfig tests TavilyProvider using web-tavily assistant config\nfunc TestTavilyProviderWithAssistantConfig(t *testing.T) {\n\tif testing.Short() {\n\t\tt.Skip(\"Skipping integration test\")\n\t}\n\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\t// Load the web-tavily test assistant to get its config\n\tast, err := assistant.LoadPath(\"/assistants/tests/web-tavily\")\n\trequire.NoError(t, err)\n\trequire.NotNil(t, ast)\n\trequire.NotNil(t, ast.Search)\n\trequire.NotNil(t, ast.Search.Web)\n\n\t// Verify assistant config\n\tassert.Equal(t, \"tests.web-tavily\", ast.ID)\n\tassert.Equal(t, \"tavily\", ast.Search.Web.Provider)\n\tassert.Equal(t, \"$ENV.TAVILY_API_KEY\", ast.Search.Web.APIKeyEnv)\n\tassert.Equal(t, 10, ast.Search.Web.MaxResults)\n\n\t// Create TavilyProvider with assistant's web config\n\tprovider := web.NewTavilyProvider(ast.Search.Web)\n\trequire.NotNil(t, provider)\n\n\t// Execute search\n\treq := &types.Request{\n\t\tQuery: \"Yao App Engine\",\n\t\tType: types.SearchTypeWeb,\n\t\tSource: types.SourceAuto,\n\t\tLimit: 5,\n\t}\n\n\tresult, err := provider.Search(req)\n\trequire.NoError(t, err)\n\trequire.NotNil(t, result)\n\n\t// Verify result structure\n\tassert.Equal(t, types.SearchTypeWeb, result.Type)\n\tassert.Equal(t, \"Yao App Engine\", result.Query)\n\tassert.Equal(t, types.SourceAuto, result.Source)\n\n\t// API key must be valid - search should succeed\n\trequire.Empty(t, result.Error, \"Search should succeed with valid API key, got error: %s\", result.Error)\n\n\t// Verify we got results\n\tassert.Greater(t, result.Total, 0)\n\tassert.NotEmpty(t, result.Items)\n\tassert.Greater(t, result.Duration, int64(0))\n\n\t// Verify result item structure\n\tfor _, item := range result.Items {\n\t\tassert.Equal(t, types.SearchTypeWeb, item.Type)\n\t\tassert.Equal(t, types.SourceAuto, item.Source)\n\t\tassert.NotEmpty(t, item.Title)\n\t\tassert.NotEmpty(t, item.URL)\n\t\t// Content may be empty for some results\n\t}\n\n\tt.Logf(\"Search returned %d results in %dms\", result.Total, result.Duration)\n}\n\n// TestTavilyProviderWithSiteRestriction tests TavilyProvider with domain restriction\nfunc TestTavilyProviderWithSiteRestriction(t *testing.T) {\n\tif testing.Short() {\n\t\tt.Skip(\"Skipping integration test\")\n\t}\n\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\t// Load the web-tavily test assistant\n\tast, err := assistant.LoadPath(\"/assistants/tests/web-tavily\")\n\trequire.NoError(t, err)\n\trequire.NotNil(t, ast.Search)\n\trequire.NotNil(t, ast.Search.Web)\n\n\t// Create TavilyProvider\n\tprovider := web.NewTavilyProvider(ast.Search.Web)\n\n\t// Execute search with site restriction\n\treq := &types.Request{\n\t\tQuery: \"documentation\",\n\t\tType: types.SearchTypeWeb,\n\t\tSource: types.SourceHook,\n\t\tSites: []string{\"github.com\"},\n\t\tLimit: 3,\n\t}\n\n\tresult, err := provider.Search(req)\n\trequire.NoError(t, err)\n\trequire.NotNil(t, result)\n\n\tassert.Equal(t, types.SearchTypeWeb, result.Type)\n\tassert.Equal(t, types.SourceHook, result.Source)\n\n\t// API key must be valid - search should succeed\n\trequire.Empty(t, result.Error, \"Search should succeed with valid API key, got error: %s\", result.Error)\n\trequire.NotEmpty(t, result.Items, \"Search should return results\")\n\n\t// All results should be from github.com\n\tfor _, item := range result.Items {\n\t\tassert.Contains(t, item.URL, \"github.com\", \"Result URL should be from github.com\")\n\t}\n\tt.Logf(\"Site-restricted search returned %d results from github.com\", result.Total)\n}\n\n// TestTavilyProviderWithoutAPIKey tests graceful degradation when API key is missing\nfunc TestTavilyProviderWithoutAPIKey(t *testing.T) {\n\t// Create provider with nil config (no API key)\n\tprovider := web.NewTavilyProvider(nil)\n\trequire.NotNil(t, provider)\n\n\treq := &types.Request{\n\t\tQuery: \"test query\",\n\t\tType: types.SearchTypeWeb,\n\t\tSource: types.SourceAuto,\n\t}\n\n\tresult, err := provider.Search(req)\n\n\t// Should not return error, but result should have error message\n\trequire.NoError(t, err)\n\trequire.NotNil(t, result)\n\tassert.Equal(t, types.SearchTypeWeb, result.Type)\n\tassert.Equal(t, \"test query\", result.Query)\n\tassert.NotEmpty(t, result.Error)\n\tassert.Contains(t, result.Error, \"API key\")\n\tassert.Empty(t, result.Items)\n\tassert.Equal(t, 0, result.Total)\n}\n\n// TestTavilyProviderWithEmptyConfig tests provider with empty config\nfunc TestTavilyProviderWithEmptyConfig(t *testing.T) {\n\t// Create provider with empty config\n\tcfg := &types.WebConfig{}\n\tprovider := web.NewTavilyProvider(cfg)\n\trequire.NotNil(t, provider)\n\n\treq := &types.Request{\n\t\tQuery: \"test query\",\n\t\tType: types.SearchTypeWeb,\n\t\tSource: types.SourceUser,\n\t}\n\n\tresult, err := provider.Search(req)\n\n\trequire.NoError(t, err)\n\trequire.NotNil(t, result)\n\tassert.NotEmpty(t, result.Error)\n\tassert.Contains(t, result.Error, \"API key\")\n}\n\n// TestTavilyProviderMaxResults tests that max_results from config is respected\nfunc TestTavilyProviderMaxResults(t *testing.T) {\n\tif testing.Short() {\n\t\tt.Skip(\"Skipping integration test\")\n\t}\n\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\t// Load the web-tavily test assistant\n\tast, err := assistant.LoadPath(\"/assistants/tests/web-tavily\")\n\trequire.NoError(t, err)\n\trequire.NotNil(t, ast.Search)\n\trequire.NotNil(t, ast.Search.Web)\n\n\t// Create TavilyProvider\n\tprovider := web.NewTavilyProvider(ast.Search.Web)\n\n\t// Execute search without limit (should use config's max_results)\n\treq := &types.Request{\n\t\tQuery: \"artificial intelligence\",\n\t\tType: types.SearchTypeWeb,\n\t\tSource: types.SourceAuto,\n\t\t// No Limit set, should use config's max_results (10)\n\t}\n\n\tresult, err := provider.Search(req)\n\trequire.NoError(t, err)\n\trequire.NotNil(t, result)\n\n\t// API key must be valid - search should succeed\n\trequire.Empty(t, result.Error, \"Search should succeed with valid API key, got error: %s\", result.Error)\n\n\t// Should respect max_results from config\n\tassert.LessOrEqual(t, result.Total, ast.Search.Web.MaxResults)\n\tt.Logf(\"Search without limit returned %d results (max: %d)\", result.Total, ast.Search.Web.MaxResults)\n\n\t// Execute search with explicit limit\n\treq2 := &types.Request{\n\t\tQuery: \"artificial intelligence\",\n\t\tType: types.SearchTypeWeb,\n\t\tSource: types.SourceAuto,\n\t\tLimit: 3, // Override config's max_results\n\t}\n\n\tresult2, err := provider.Search(req2)\n\trequire.NoError(t, err)\n\trequire.NotNil(t, result2)\n\n\t// API key must be valid - search should succeed\n\trequire.Empty(t, result2.Error, \"Search should succeed with valid API key, got error: %s\", result2.Error)\n\n\t// Should respect request's limit\n\tassert.LessOrEqual(t, result2.Total, 3)\n\tt.Logf(\"Search with limit=3 returned %d results\", result2.Total)\n}\n" }, { "path": "agent/search/interfaces/handler.go", "content": "package interfaces\n\nimport (\n\t\"github.com/yaoapp/yao/agent/context\"\n\t\"github.com/yaoapp/yao/agent/search/types\"\n)\n\n// Handler defines the interface for search implementations\ntype Handler interface {\n\t// Type returns the search type this handler supports\n\tType() types.SearchType\n\n\t// Search executes the search and returns results\n\tSearch(req *types.Request) (*types.Result, error)\n}\n\n// ContextHandler extends Handler with context support\n// Handlers that need context (e.g., DB handler for QueryDSL generation) should implement this\ntype ContextHandler interface {\n\tHandler\n\n\t// SearchWithContext executes the search with context and returns results\n\tSearchWithContext(ctx *context.Context, req *types.Request) (*types.Result, error)\n}\n" }, { "path": "agent/search/interfaces/nlp.go", "content": "package interfaces\n\nimport (\n\t\"github.com/yaoapp/gou/model\"\n\t\"github.com/yaoapp/gou/query/gou\"\n\t\"github.com/yaoapp/yao/agent/context\"\n\t\"github.com/yaoapp/yao/agent/search/types\"\n)\n\n// KeywordExtractor extracts keywords for web search\ntype KeywordExtractor interface {\n\t// Extract extracts search keywords from user message\n\t// ctx is required for Agent and MCP modes, can be nil for builtin mode\n\tExtract(ctx *context.Context, content string, opts *types.KeywordOptions) ([]string, error)\n}\n\n// QueryDSLGenerator generates QueryDSL for DB search\ntype QueryDSLGenerator interface {\n\t// Generate converts natural language to QueryDSL\n\tGenerate(query string, models []*model.Model) (*gou.QueryDSL, error)\n}\n\n// Note: Embedding is handled by KB collection's own config (embedding provider + model),\n// not defined here. See KB handler for details.\n" }, { "path": "agent/search/interfaces/reranker.go", "content": "package interfaces\n\nimport (\n\t\"github.com/yaoapp/yao/agent/context\"\n\t\"github.com/yaoapp/yao/agent/search/types\"\n)\n\n// Reranker reorders search results by relevance\ntype Reranker interface {\n\t// Rerank reorders results based on query relevance\n\t// ctx is required for Agent and MCP modes, can be nil for builtin mode\n\tRerank(ctx *context.Context, query string, items []*types.ResultItem, opts *types.RerankOptions) ([]*types.ResultItem, error)\n}\n" }, { "path": "agent/search/interfaces/searcher.go", "content": "package interfaces\n\nimport (\n\t\"github.com/yaoapp/yao/agent/context\"\n\t\"github.com/yaoapp/yao/agent/search/types\"\n)\n\n// Searcher is the main interface exposed to external callers\ntype Searcher interface {\n\t// Search executes a single search request\n\tSearch(ctx *context.Context, req *types.Request) (*types.Result, error)\n\n\t// Parallel search methods - inspired by JavaScript Promise\n\t// All waits for all searches to complete (like Promise.all)\n\tAll(ctx *context.Context, reqs []*types.Request) ([]*types.Result, error)\n\t// Any returns when any search succeeds with results (like Promise.any)\n\tAny(ctx *context.Context, reqs []*types.Request) ([]*types.Result, error)\n\t// Race returns when any search completes (like Promise.race)\n\tRace(ctx *context.Context, reqs []*types.Request) ([]*types.Result, error)\n\n\t// BuildReferences converts search results to unified Reference format for LLM\n\tBuildReferences(results []*types.Result) []*types.Reference\n}\n" }, { "path": "agent/search/jsapi.go", "content": "package search\n\nimport (\n\t\"github.com/yaoapp/yao/agent/context\"\n\t\"github.com/yaoapp/yao/agent/search/types\"\n)\n\n// JSAPI implements context.SearchAPI interface\n// Provides ctx.search.Web(), ctx.search.KB(), ctx.search.DB(), ctx.search.All(), ctx.search.Any(), ctx.search.Race()\ntype JSAPI struct {\n\tctx *context.Context\n\tsearcher *Searcher\n}\n\n// NewJSAPI creates a new search JSAPI instance\nfunc NewJSAPI(ctx *context.Context, config *types.Config, uses *Uses) *JSAPI {\n\treturn &JSAPI{\n\t\tctx: ctx,\n\t\tsearcher: New(config, uses),\n\t}\n}\n\n// Web executes web search\n// Options:\n// - limit: int - max results (default: 10)\n// - sites: []string - restrict to specific sites\n// - time_range: string - \"day\", \"week\", \"month\", \"year\"\n// - rerank: map[string]interface{} - rerank options\nfunc (api *JSAPI) Web(query string, opts map[string]interface{}) interface{} {\n\treq := api.buildRequest(types.SearchTypeWeb, query, opts)\n\tresult, _ := api.searcher.Search(api.ctx, req)\n\treturn result\n}\n\n// KB executes knowledge base search\n// Options:\n// - collections: []string - collection IDs\n// - threshold: float64 - similarity threshold (0-1)\n// - limit: int - max results\n// - graph: bool - enable graph association\n// - rerank: map[string]interface{} - rerank options\nfunc (api *JSAPI) KB(query string, opts map[string]interface{}) interface{} {\n\treq := api.buildRequest(types.SearchTypeKB, query, opts)\n\tresult, _ := api.searcher.Search(api.ctx, req)\n\treturn result\n}\n\n// DB executes database search\n// Options:\n// - models: []string - model IDs\n// - wheres: []map[string]interface{} - pre-defined filters (GOU QueryDSL Where format)\n// - orders: []map[string]interface{} - sort orders (GOU QueryDSL Order format)\n// - select: []string - fields to return\n// - limit: int - max results\n// - rerank: map[string]interface{} - rerank options\nfunc (api *JSAPI) DB(query string, opts map[string]interface{}) interface{} {\n\treq := api.buildRequest(types.SearchTypeDB, query, opts)\n\tresult, _ := api.searcher.Search(api.ctx, req)\n\treturn result\n}\n\n// All executes all searches and waits for all to complete (like Promise.all)\n// Each request should have:\n// - type: string - \"web\", \"kb\", or \"db\"\n// - query: string - search query\n// - ... other type-specific options\nfunc (api *JSAPI) All(requests []interface{}) []interface{} {\n\treqs := api.parseRequests(requests)\n\tresults, _ := api.searcher.All(api.ctx, reqs)\n\treturn api.convertResults(results)\n}\n\n// Any returns as soon as any search succeeds with results (like Promise.any)\n// Each request should have:\n// - type: string - \"web\", \"kb\", or \"db\"\n// - query: string - search query\n// - ... other type-specific options\nfunc (api *JSAPI) Any(requests []interface{}) []interface{} {\n\treqs := api.parseRequests(requests)\n\tresults, _ := api.searcher.Any(api.ctx, reqs)\n\treturn api.convertResults(results)\n}\n\n// Race returns as soon as any search completes (like Promise.race)\n// Each request should have:\n// - type: string - \"web\", \"kb\", or \"db\"\n// - query: string - search query\n// - ... other type-specific options\nfunc (api *JSAPI) Race(requests []interface{}) []interface{} {\n\treqs := api.parseRequests(requests)\n\tresults, _ := api.searcher.Race(api.ctx, reqs)\n\treturn api.convertResults(results)\n}\n\n// buildRequest builds a Request from query and options\nfunc (api *JSAPI) buildRequest(searchType types.SearchType, query string, opts map[string]interface{}) *types.Request {\n\treq := &types.Request{\n\t\tType: searchType,\n\t\tQuery: query,\n\t\tSource: types.SourceHook, // JSAPI calls are from hooks\n\t}\n\n\tif opts == nil {\n\t\treturn req\n\t}\n\n\t// Common options\n\tif limit, ok := opts[\"limit\"].(float64); ok {\n\t\treq.Limit = int(limit)\n\t} else if limit, ok := opts[\"limit\"].(int); ok {\n\t\treq.Limit = limit\n\t}\n\n\t// Web-specific options\n\tif searchType == types.SearchTypeWeb {\n\t\tif sites, ok := opts[\"sites\"].([]interface{}); ok {\n\t\t\treq.Sites = toStringSlice(sites)\n\t\t}\n\t\tif timeRange, ok := opts[\"time_range\"].(string); ok {\n\t\t\treq.TimeRange = timeRange\n\t\t}\n\t}\n\n\t// KB-specific options\n\tif searchType == types.SearchTypeKB {\n\t\tif collections, ok := opts[\"collections\"].([]interface{}); ok {\n\t\t\treq.Collections = toStringSlice(collections)\n\t\t}\n\t\tif threshold, ok := opts[\"threshold\"].(float64); ok {\n\t\t\treq.Threshold = threshold\n\t\t}\n\t\tif graph, ok := opts[\"graph\"].(bool); ok {\n\t\t\treq.Graph = graph\n\t\t}\n\t}\n\n\t// DB-specific options\n\tif searchType == types.SearchTypeDB {\n\t\tif models, ok := opts[\"models\"].([]interface{}); ok {\n\t\t\treq.Models = toStringSlice(models)\n\t\t}\n\t\tif selectFields, ok := opts[\"select\"].([]interface{}); ok {\n\t\t\treq.Select = toStringSlice(selectFields)\n\t\t}\n\t\t// Note: wheres and orders are more complex, handled by QueryDSL generator\n\t}\n\n\t// Rerank options\n\tif rerankOpts, ok := opts[\"rerank\"].(map[string]interface{}); ok {\n\t\treq.Rerank = &types.RerankOptions{}\n\t\tif topN, ok := rerankOpts[\"top_n\"].(float64); ok {\n\t\t\treq.Rerank.TopN = int(topN)\n\t\t} else if topN, ok := rerankOpts[\"top_n\"].(int); ok {\n\t\t\treq.Rerank.TopN = topN\n\t\t}\n\t}\n\n\treturn req\n}\n\n// parseRequests parses an array of request objects into typed Requests\nfunc (api *JSAPI) parseRequests(requests []interface{}) []*types.Request {\n\treqs := make([]*types.Request, 0, len(requests))\n\tfor _, r := range requests {\n\t\treqMap, ok := r.(map[string]interface{})\n\t\tif !ok {\n\t\t\tcontinue\n\t\t}\n\n\t\t// Get type\n\t\ttypeStr, ok := reqMap[\"type\"].(string)\n\t\tif !ok {\n\t\t\tcontinue\n\t\t}\n\t\tsearchType := types.SearchType(typeStr)\n\n\t\t// Get query\n\t\tquery, ok := reqMap[\"query\"].(string)\n\t\tif !ok {\n\t\t\tcontinue\n\t\t}\n\n\t\t// Build request with remaining options\n\t\treq := api.buildRequest(searchType, query, reqMap)\n\t\treqs = append(reqs, req)\n\t}\n\treturn reqs\n}\n\n// convertResults converts typed Results to interface slice for JS\nfunc (api *JSAPI) convertResults(results []*types.Result) []interface{} {\n\tout := make([]interface{}, len(results))\n\tfor i, r := range results {\n\t\tout[i] = r\n\t}\n\treturn out\n}\n\n// toStringSlice converts []interface{} to []string\nfunc toStringSlice(arr []interface{}) []string {\n\tresult := make([]string, 0, len(arr))\n\tfor _, v := range arr {\n\t\tif s, ok := v.(string); ok {\n\t\t\tresult = append(result, s)\n\t\t}\n\t}\n\treturn result\n}\n\n// ConfigGetter is a function type that retrieves search config and uses for an assistant\ntype ConfigGetter func(assistantID string) (*types.Config, *Uses)\n\n// configGetter is set by assistant package during initialization\nvar configGetter ConfigGetter\n\n// SetJSAPIFactory sets the factory function for creating SearchAPI instances\n// Called by assistant package during initialization\n// getter: function to get search config and uses from assistant ID\nfunc SetJSAPIFactory(getter ConfigGetter) {\n\tconfigGetter = getter\n\tcontext.SearchAPIFactory = func(ctx *context.Context) context.SearchAPI {\n\t\tvar config *types.Config\n\t\tvar uses *Uses\n\t\tif configGetter != nil && ctx.AssistantID != \"\" {\n\t\t\tconfig, uses = configGetter(ctx.AssistantID)\n\t\t}\n\t\treturn NewJSAPI(ctx, config, uses)\n\t}\n}\n" }, { "path": "agent/search/jsapi_db_test.go", "content": "package search_test\n\nimport (\n\t\"testing\"\n\n\t\"github.com/yaoapp/gou/model\"\n\t\"github.com/yaoapp/yao/agent/context\"\n\t\"github.com/yaoapp/yao/agent/search\"\n\t\"github.com/yaoapp/yao/agent/search/types\"\n\t\"github.com/yaoapp/yao/agent/testutils\"\n\toauthTypes \"github.com/yaoapp/yao/openapi/oauth/types\"\n\n\t\"github.com/stretchr/testify/assert\"\n\t\"github.com/stretchr/testify/require\"\n)\n\n// ============================================================================\n// DB Search JSAPI Integration Tests\n// ============================================================================\n\nfunc TestJSAPI_DB_Integration(t *testing.T) {\n\t// Skip if running short tests\n\tif testing.Short() {\n\t\tt.Skip(\"Skipping integration test\")\n\t}\n\n\t// Initialize test environment (loads models, database, query engine, etc.)\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\t// Create test context\n\tctx := newJSAPITestContext(t)\n\n\t// Verify __yao.role model is loaded\n\tmod := model.Select(\"__yao.role\")\n\trequire.NotNil(t, mod, \"__yao.role model should be loaded\")\n\n\t// Ensure test data exists\n\tensureJSAPITestRole(t, mod)\n\n\tt.Run(\"db_search_with_context\", func(t *testing.T) {\n\t\tapi := search.NewJSAPI(ctx, &types.Config{\n\t\t\tDB: &types.DBConfig{\n\t\t\t\tModels: []string{\"__yao.role\"},\n\t\t\t\tMaxResults: 10,\n\t\t\t},\n\t\t}, &search.Uses{QueryDSL: \"builtin\"})\n\n\t\tresult := api.DB(\"查询所有角色\", map[string]interface{}{\n\t\t\t\"models\": []interface{}{\"__yao.role\"},\n\t\t\t\"limit\": float64(10),\n\t\t})\n\n\t\trequire.NotNil(t, result)\n\t\tr, ok := result.(*types.Result)\n\t\trequire.True(t, ok)\n\n\t\tassert.Equal(t, types.SearchTypeDB, r.Type)\n\t\tassert.Equal(t, \"查询所有角色\", r.Query)\n\t\tassert.Equal(t, types.SourceHook, r.Source)\n\n\t\tif r.Error != \"\" {\n\t\t\tt.Logf(\"Search error: %s\", r.Error)\n\t\t}\n\t\tassert.Empty(t, r.Error, \"Should not have error\")\n\t\tassert.Greater(t, len(r.Items), 0, \"Should have results\")\n\t})\n\n\tt.Run(\"db_search_with_scenario\", func(t *testing.T) {\n\t\tapi := search.NewJSAPI(ctx, &types.Config{\n\t\t\tDB: &types.DBConfig{\n\t\t\t\tModels: []string{\"__yao.role\"},\n\t\t\t\tMaxResults: 5,\n\t\t\t},\n\t\t}, &search.Uses{QueryDSL: \"builtin\"})\n\n\t\tresult := api.DB(\"查询系统角色\", map[string]interface{}{\n\t\t\t\"models\": []interface{}{\"__yao.role\"},\n\t\t\t\"scenario\": \"filter\",\n\t\t\t\"limit\": float64(5),\n\t\t})\n\n\t\trequire.NotNil(t, result)\n\t\tr, ok := result.(*types.Result)\n\t\trequire.True(t, ok)\n\n\t\tassert.Equal(t, types.SearchTypeDB, r.Type)\n\t\tassert.LessOrEqual(t, len(r.Items), 5, \"Should respect limit\")\n\t})\n\n\tt.Run(\"db_search_with_select_fields\", func(t *testing.T) {\n\t\tapi := search.NewJSAPI(ctx, &types.Config{\n\t\t\tDB: &types.DBConfig{\n\t\t\t\tModels: []string{\"__yao.role\"},\n\t\t\t\tMaxResults: 10,\n\t\t\t},\n\t\t}, &search.Uses{QueryDSL: \"builtin\"})\n\n\t\tresult := api.DB(\"查询角色名称\", map[string]interface{}{\n\t\t\t\"models\": []interface{}{\"__yao.role\"},\n\t\t\t\"select\": []interface{}{\"id\", \"name\", \"description\"},\n\t\t\t\"limit\": float64(10),\n\t\t})\n\n\t\trequire.NotNil(t, result)\n\t\tr, ok := result.(*types.Result)\n\t\trequire.True(t, ok)\n\n\t\tassert.Equal(t, types.SearchTypeDB, r.Type)\n\t\tif r.Error == \"\" && len(r.Items) > 0 {\n\t\t\t// Verify items have data\n\t\t\tfor _, item := range r.Items {\n\t\t\t\tassert.NotNil(t, item.Data)\n\t\t\t\tassert.Equal(t, \"__yao.role\", item.Model)\n\t\t\t}\n\t\t}\n\t})\n\n\tt.Run(\"db_search_all_with_multiple_types\", func(t *testing.T) {\n\t\tapi := search.NewJSAPI(ctx, &types.Config{\n\t\t\tKB: &types.KBConfig{Collections: []string{\"docs\"}},\n\t\t\tDB: &types.DBConfig{\n\t\t\t\tModels: []string{\"__yao.role\"},\n\t\t\t\tMaxResults: 10,\n\t\t\t},\n\t\t}, &search.Uses{QueryDSL: \"builtin\"})\n\n\t\trequests := []interface{}{\n\t\t\tmap[string]interface{}{\n\t\t\t\t\"type\": \"db\",\n\t\t\t\t\"query\": \"查询角色\",\n\t\t\t\t\"models\": []interface{}{\"__yao.role\"},\n\t\t\t\t\"limit\": float64(5),\n\t\t\t},\n\t\t\tmap[string]interface{}{\n\t\t\t\t\"type\": \"kb\",\n\t\t\t\t\"query\": \"知识库查询\",\n\t\t\t\t\"collections\": []interface{}{\"docs\"},\n\t\t\t\t\"limit\": float64(5),\n\t\t\t},\n\t\t}\n\n\t\tresults := api.All(requests)\n\t\trequire.Len(t, results, 2)\n\n\t\t// DB result\n\t\tr0, ok := results[0].(*types.Result)\n\t\trequire.True(t, ok)\n\t\tassert.Equal(t, types.SearchTypeDB, r0.Type)\n\n\t\t// KB result\n\t\tr1, ok := results[1].(*types.Result)\n\t\trequire.True(t, ok)\n\t\tassert.Equal(t, types.SearchTypeKB, r1.Type)\n\t})\n}\n\n// newJSAPITestContext creates a test context for JSAPI tests\nfunc newJSAPITestContext(t *testing.T) *context.Context {\n\tt.Helper()\n\tauthorized := &oauthTypes.AuthorizedInfo{\n\t\tUserID: \"test-user-jsapi\",\n\t}\n\tchatID := \"test-chat-jsapi-db\"\n\tctx := context.New(t.Context(), authorized, chatID)\n\treturn ctx\n}\n\n// ensureJSAPITestRole ensures there's at least one role in the database\nfunc ensureJSAPITestRole(t *testing.T, mod *model.Model) {\n\tt.Helper()\n\n\t// Try to find existing roles\n\trows, err := mod.Get(model.QueryParam{Limit: 1})\n\tif err == nil && len(rows) > 0 {\n\t\treturn\n\t}\n\n\t// Create a test role\n\t_, err = mod.Create(map[string]interface{}{\n\t\t\"role_id\": \"jsapi_test_role\",\n\t\t\"name\": \"JSAPI Test Role\",\n\t\t\"description\": \"A test role for JSAPI unit testing\",\n\t\t\"is_active\": true,\n\t\t\"is_system\": false,\n\t\t\"level\": 1,\n\t})\n\tif err != nil {\n\t\tt.Logf(\"Note: Could not create test role: %v\", err)\n\t}\n}\n" }, { "path": "agent/search/jsapi_test.go", "content": "package search_test\n\nimport (\n\t\"testing\"\n\n\t\"github.com/stretchr/testify/assert\"\n\t\"github.com/stretchr/testify/require\"\n\t\"github.com/yaoapp/yao/agent/context\"\n\t\"github.com/yaoapp/yao/agent/search\"\n\t\"github.com/yaoapp/yao/agent/search/types\"\n\t\"github.com/yaoapp/yao/agent/testutils\"\n)\n\nfunc TestNewJSAPI(t *testing.T) {\n\tapi := search.NewJSAPI(nil, nil, nil)\n\trequire.NotNil(t, api)\n}\n\nfunc TestJSAPI_Web(t *testing.T) {\n\tapi := search.NewJSAPI(nil, &types.Config{\n\t\tWeb: &types.WebConfig{Provider: \"tavily\"},\n\t}, &search.Uses{Web: \"builtin\"})\n\n\tresult := api.Web(\"test query\", nil)\n\trequire.NotNil(t, result)\n\n\tr, ok := result.(*types.Result)\n\trequire.True(t, ok)\n\tassert.Equal(t, types.SearchTypeWeb, r.Type)\n\tassert.Equal(t, \"test query\", r.Query)\n\tassert.Equal(t, types.SourceHook, r.Source)\n}\n\nfunc TestJSAPI_Web_WithOptions(t *testing.T) {\n\tapi := search.NewJSAPI(nil, &types.Config{\n\t\tWeb: &types.WebConfig{Provider: \"tavily\"},\n\t}, &search.Uses{Web: \"builtin\"})\n\n\topts := map[string]interface{}{\n\t\t\"limit\": float64(5),\n\t\t\"sites\": []interface{}{\"github.com\", \"stackoverflow.com\"},\n\t\t\"time_range\": \"week\",\n\t}\n\n\tresult := api.Web(\"golang concurrency\", opts)\n\trequire.NotNil(t, result)\n\n\tr, ok := result.(*types.Result)\n\trequire.True(t, ok)\n\tassert.Equal(t, types.SearchTypeWeb, r.Type)\n\tassert.Equal(t, \"golang concurrency\", r.Query)\n}\n\nfunc TestJSAPI_KB(t *testing.T) {\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\tapi := search.NewJSAPI(nil, &types.Config{\n\t\tKB: &types.KBConfig{Collections: []string{\"docs\"}},\n\t}, nil)\n\n\tresult := api.KB(\"test query\", nil)\n\trequire.NotNil(t, result)\n\n\tr, ok := result.(*types.Result)\n\trequire.True(t, ok)\n\tassert.Equal(t, types.SearchTypeKB, r.Type)\n\tassert.Equal(t, \"test query\", r.Query)\n\tassert.Equal(t, types.SourceHook, r.Source)\n}\n\nfunc TestJSAPI_KB_WithOptions(t *testing.T) {\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\tapi := search.NewJSAPI(nil, &types.Config{\n\t\tKB: &types.KBConfig{Collections: []string{\"docs\"}},\n\t}, nil)\n\n\topts := map[string]interface{}{\n\t\t\"collections\": []interface{}{\"docs\", \"faq\"},\n\t\t\"threshold\": 0.8,\n\t\t\"limit\": float64(10),\n\t\t\"graph\": true,\n\t}\n\n\tresult := api.KB(\"knowledge base query\", opts)\n\trequire.NotNil(t, result)\n\n\tr, ok := result.(*types.Result)\n\trequire.True(t, ok)\n\tassert.Equal(t, types.SearchTypeKB, r.Type)\n\tassert.Equal(t, \"knowledge base query\", r.Query)\n}\n\nfunc TestJSAPI_DB(t *testing.T) {\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\tapi := search.NewJSAPI(nil, &types.Config{\n\t\tDB: &types.DBConfig{Models: []string{\"product\"}},\n\t}, &search.Uses{QueryDSL: \"builtin\"})\n\n\tresult := api.DB(\"test query\", nil)\n\trequire.NotNil(t, result)\n\n\tr, ok := result.(*types.Result)\n\trequire.True(t, ok)\n\tassert.Equal(t, types.SearchTypeDB, r.Type)\n\tassert.Equal(t, \"test query\", r.Query)\n\tassert.Equal(t, types.SourceHook, r.Source)\n}\n\nfunc TestJSAPI_DB_WithOptions(t *testing.T) {\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\tapi := search.NewJSAPI(nil, &types.Config{\n\t\tDB: &types.DBConfig{Models: []string{\"product\"}},\n\t}, &search.Uses{QueryDSL: \"builtin\"})\n\n\topts := map[string]interface{}{\n\t\t\"models\": []interface{}{\"product\", \"order\"},\n\t\t\"select\": []interface{}{\"id\", \"name\", \"price\"},\n\t\t\"limit\": float64(20),\n\t}\n\n\tresult := api.DB(\"database query\", opts)\n\trequire.NotNil(t, result)\n\n\tr, ok := result.(*types.Result)\n\trequire.True(t, ok)\n\tassert.Equal(t, types.SearchTypeDB, r.Type)\n\tassert.Equal(t, \"database query\", r.Query)\n}\n\nfunc TestJSAPI_All(t *testing.T) {\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\tapi := search.NewJSAPI(nil, &types.Config{\n\t\tKB: &types.KBConfig{Collections: []string{\"docs\"}},\n\t\tDB: &types.DBConfig{Models: []string{\"product\"}},\n\t}, nil)\n\n\trequests := []interface{}{\n\t\tmap[string]interface{}{\n\t\t\t\"type\": \"kb\",\n\t\t\t\"query\": \"KB query\",\n\t\t},\n\t\tmap[string]interface{}{\n\t\t\t\"type\": \"db\",\n\t\t\t\"query\": \"DB query\",\n\t\t},\n\t}\n\n\tresults := api.All(requests)\n\trequire.Len(t, results, 2)\n\n\t// First result (KB)\n\tr0, ok := results[0].(*types.Result)\n\trequire.True(t, ok)\n\tassert.Equal(t, types.SearchTypeKB, r0.Type)\n\tassert.Equal(t, \"KB query\", r0.Query)\n\n\t// Second result (DB)\n\tr1, ok := results[1].(*types.Result)\n\trequire.True(t, ok)\n\tassert.Equal(t, types.SearchTypeDB, r1.Type)\n\tassert.Equal(t, \"DB query\", r1.Query)\n}\n\nfunc TestJSAPI_Any(t *testing.T) {\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\tapi := search.NewJSAPI(nil, &types.Config{\n\t\tKB: &types.KBConfig{Collections: []string{\"docs\"}},\n\t\tDB: &types.DBConfig{Models: []string{\"product\"}},\n\t}, nil)\n\n\trequests := []interface{}{\n\t\tmap[string]interface{}{\n\t\t\t\"type\": \"kb\",\n\t\t\t\"query\": \"KB query\",\n\t\t},\n\t\tmap[string]interface{}{\n\t\t\t\"type\": \"db\",\n\t\t\t\"query\": \"DB query\",\n\t\t},\n\t}\n\n\tresults := api.Any(requests)\n\trequire.Len(t, results, 2)\n\n\t// At least one result should be present\n\thasResult := false\n\tfor _, r := range results {\n\t\tif r != nil {\n\t\t\thasResult = true\n\t\t\tbreak\n\t\t}\n\t}\n\tassert.True(t, hasResult)\n}\n\nfunc TestJSAPI_Race(t *testing.T) {\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\tapi := search.NewJSAPI(nil, &types.Config{\n\t\tKB: &types.KBConfig{Collections: []string{\"docs\"}},\n\t\tDB: &types.DBConfig{Models: []string{\"product\"}},\n\t}, nil)\n\n\trequests := []interface{}{\n\t\tmap[string]interface{}{\n\t\t\t\"type\": \"kb\",\n\t\t\t\"query\": \"KB query\",\n\t\t},\n\t\tmap[string]interface{}{\n\t\t\t\"type\": \"db\",\n\t\t\t\"query\": \"DB query\",\n\t\t},\n\t}\n\n\tresults := api.Race(requests)\n\trequire.Len(t, results, 2)\n\n\t// At least one result should be present\n\thasResult := false\n\tfor _, r := range results {\n\t\tif r != nil {\n\t\t\thasResult = true\n\t\t\tbreak\n\t\t}\n\t}\n\tassert.True(t, hasResult)\n}\n\nfunc TestJSAPI_All_Empty(t *testing.T) {\n\tapi := search.NewJSAPI(nil, nil, nil)\n\tresults := api.All([]interface{}{})\n\tassert.Len(t, results, 0)\n}\n\nfunc TestJSAPI_Any_Empty(t *testing.T) {\n\tapi := search.NewJSAPI(nil, nil, nil)\n\tresults := api.Any([]interface{}{})\n\tassert.Len(t, results, 0)\n}\n\nfunc TestJSAPI_Race_Empty(t *testing.T) {\n\tapi := search.NewJSAPI(nil, nil, nil)\n\tresults := api.Race([]interface{}{})\n\tassert.Len(t, results, 0)\n}\n\nfunc TestJSAPI_Web_WithRerank(t *testing.T) {\n\tapi := search.NewJSAPI(nil, &types.Config{\n\t\tWeb: &types.WebConfig{Provider: \"tavily\"},\n\t}, &search.Uses{Web: \"builtin\"})\n\n\topts := map[string]interface{}{\n\t\t\"limit\": float64(10),\n\t\t\"rerank\": map[string]interface{}{\n\t\t\t\"top_n\": float64(5),\n\t\t},\n\t}\n\n\tresult := api.Web(\"test query\", opts)\n\trequire.NotNil(t, result)\n\n\tr, ok := result.(*types.Result)\n\trequire.True(t, ok)\n\tassert.Equal(t, types.SearchTypeWeb, r.Type)\n}\n\nfunc TestJSAPI_All_InvalidRequests(t *testing.T) {\n\tapi := search.NewJSAPI(nil, &types.Config{\n\t\tWeb: &types.WebConfig{Provider: \"tavily\"},\n\t}, &search.Uses{Web: \"builtin\"})\n\n\t// Mix of invalid and valid requests\n\trequests := []interface{}{\n\t\t\"invalid\", // Not a map\n\t\tmap[string]interface{}{\n\t\t\t\"query\": \"no type\", // Missing type\n\t\t},\n\t\tmap[string]interface{}{\n\t\t\t\"type\": \"web\", // Missing query\n\t\t},\n\t\tmap[string]interface{}{\n\t\t\t\"type\": \"web\",\n\t\t\t\"query\": \"valid query\",\n\t\t},\n\t}\n\n\tresults := api.All(requests)\n\t// Only the valid request should produce a result\n\tassert.Len(t, results, 1)\n}\n\nfunc TestSetJSAPIFactory(t *testing.T) {\n\t// Reset factory\n\tcontext.SearchAPIFactory = nil\n\n\t// Set factory with nil getter (uses defaults)\n\tsearch.SetJSAPIFactory(nil)\n\n\t// Verify factory is set\n\trequire.NotNil(t, context.SearchAPIFactory)\n\n\t// Create a mock context\n\tctx := context.New(nil, nil, \"test-chat\")\n\n\t// Get search API\n\tsearchAPI := context.SearchAPIFactory(ctx)\n\trequire.NotNil(t, searchAPI)\n}\n\nfunc TestSetJSAPIFactory_WithGetter(t *testing.T) {\n\t// Reset factory\n\tcontext.SearchAPIFactory = nil\n\n\t// Set factory with custom getter\n\tsearch.SetJSAPIFactory(func(assistantID string) (*types.Config, *search.Uses) {\n\t\tif assistantID == \"test-assistant\" {\n\t\t\treturn &types.Config{\n\t\t\t\tWeb: &types.WebConfig{Provider: \"tavily\"},\n\t\t\t}, &search.Uses{Web: \"builtin\"}\n\t\t}\n\t\treturn nil, nil\n\t})\n\n\t// Verify factory is set\n\trequire.NotNil(t, context.SearchAPIFactory)\n\n\t// Create a context with assistant ID\n\tctx := context.New(nil, nil, \"test-chat\")\n\tctx.AssistantID = \"test-assistant\"\n\n\t// Get search API\n\tsearchAPI := context.SearchAPIFactory(ctx)\n\trequire.NotNil(t, searchAPI)\n}\n\nfunc TestJSAPI_ImplementsSearchAPI(t *testing.T) {\n\t// Verify JSAPI implements context.SearchAPI interface\n\tvar _ context.SearchAPI = search.NewJSAPI(nil, nil, nil)\n}\n" }, { "path": "agent/search/nlp/keyword/agent.go", "content": "package keyword\n\nimport (\n\t\"encoding/json\"\n\t\"fmt\"\n\n\t\"github.com/yaoapp/yao/agent/caller\"\n\tagentContext \"github.com/yaoapp/yao/agent/context\"\n\t\"github.com/yaoapp/yao/agent/search/types\"\n)\n\n// AgentProvider delegates keyword extraction to an LLM-powered assistant\n// The assistant can understand context and extract semantically relevant keywords\ntype AgentProvider struct {\n\tagentID string // Assistant ID to delegate to\n}\n\n// NewAgentProvider creates a new agent-based keyword extractor\nfunc NewAgentProvider(agentID string) *AgentProvider {\n\treturn &AgentProvider{\n\t\tagentID: agentID,\n\t}\n}\n\n// Extract extracts keywords by calling the target agent\n// The agent receives the content and returns extracted keywords with weights\nfunc (p *AgentProvider) Extract(ctx *agentContext.Context, content string, opts *types.KeywordOptions) ([]types.Keyword, error) {\n\tif ctx == nil {\n\t\treturn nil, fmt.Errorf(\"context is required for agent keyword extraction\")\n\t}\n\n\t// Check if AgentGetterFunc is initialized\n\tif caller.AgentGetterFunc == nil {\n\t\treturn nil, fmt.Errorf(\"AgentGetterFunc not initialized\")\n\t}\n\n\t// Get the agent\n\tagent, err := caller.AgentGetterFunc(p.agentID)\n\tif err != nil {\n\t\treturn nil, fmt.Errorf(\"failed to get agent %s: %w\", p.agentID, err)\n\t}\n\n\t// Build the request message\n\trequestData := map[string]interface{}{\n\t\t\"content\": content,\n\t\t\"max_keywords\": opts.MaxKeywords,\n\t\t\"language\": opts.Language,\n\t}\n\trequestJSON, _ := json.Marshal(requestData)\n\n\t// Create message for the agent\n\tmessages := []agentContext.Message{\n\t\t{\n\t\t\tRole: \"user\",\n\t\t\tContent: string(requestJSON),\n\t\t},\n\t}\n\n\t// Call the agent with skip options (no history, no output)\n\toptions := &agentContext.Options{\n\t\tSkip: &agentContext.Skip{\n\t\t\tHistory: true,\n\t\t\tOutput: true,\n\t\t},\n\t}\n\n\tresponse, err := agent.Stream(ctx, messages, options)\n\tif err != nil {\n\t\treturn nil, fmt.Errorf(\"agent call failed: %w\", err)\n\t}\n\n\t// Parse the result from response.Next\n\treturn p.parseResponse(response)\n}\n\n// parseResponse extracts keywords from the agent's *context.Response\n// Now that agent.Stream() returns *context.Response directly,\n// we can access fields without type assertions.\n//\n// The agent returns keywords in response.Next field as {data: {keywords: [{k, w}, ...]}}\nfunc (p *AgentProvider) parseResponse(response *agentContext.Response) ([]types.Keyword, error) {\n\tif response == nil || response.Next == nil {\n\t\treturn []types.Keyword{}, nil\n\t}\n\n\treturn p.parseNextData(response.Next)\n}\n\n// parseNextData extracts keywords from Next hook data\n// Expected format: {data: {keywords: [{k: \"keyword\", w: 0.9}, ...]}}\nfunc (p *AgentProvider) parseNextData(next interface{}) ([]types.Keyword, error) {\n\tif next == nil {\n\t\treturn []types.Keyword{}, nil\n\t}\n\n\t// Try to convert to map first (most common case)\n\tvar data map[string]interface{}\n\n\tswitch v := next.(type) {\n\tcase map[string]interface{}:\n\t\tdata = v\n\tcase string:\n\t\t// Try to parse as JSON\n\t\tif err := json.Unmarshal([]byte(v), &data); err != nil {\n\t\t\t// Not a JSON object, try as array of keywords\n\t\t\tvar keywords []types.Keyword\n\t\t\tif err := json.Unmarshal([]byte(v), &keywords); err == nil {\n\t\t\t\treturn keywords, nil\n\t\t\t}\n\t\t\t// Return as single keyword with default weight\n\t\t\treturn []types.Keyword{{K: v, W: 0.5}}, nil\n\t\t}\n\tcase []types.Keyword:\n\t\treturn v, nil\n\tcase []interface{}:\n\t\treturn p.extractKeywordsFromArray(v)\n\tdefault:\n\t\t// Try to marshal and unmarshal\n\t\tjsonBytes, err := json.Marshal(next)\n\t\tif err != nil {\n\t\t\treturn []types.Keyword{}, nil\n\t\t}\n\t\tif err := json.Unmarshal(jsonBytes, &data); err != nil {\n\t\t\treturn []types.Keyword{}, nil\n\t\t}\n\t}\n\n\t// Extract keywords from data\n\t// Try common field names: \"keywords\", \"data\", \"data.keywords\"\n\tif kw, ok := data[\"keywords\"]; ok {\n\t\treturn p.extractKeywordsFromValue(kw)\n\t}\n\tif d, ok := data[\"data\"]; ok {\n\t\tif dm, ok := d.(map[string]interface{}); ok {\n\t\t\tif kw, ok := dm[\"keywords\"]; ok {\n\t\t\t\treturn p.extractKeywordsFromValue(kw)\n\t\t\t}\n\t\t}\n\t\treturn p.extractKeywordsFromValue(d)\n\t}\n\n\treturn []types.Keyword{}, nil\n}\n\n// extractKeywordsFromValue extracts Keyword array from various types\nfunc (p *AgentProvider) extractKeywordsFromValue(v interface{}) ([]types.Keyword, error) {\n\tswitch kw := v.(type) {\n\tcase []types.Keyword:\n\t\treturn kw, nil\n\tcase []interface{}:\n\t\treturn p.extractKeywordsFromArray(kw)\n\tcase string:\n\t\tvar keywords []types.Keyword\n\t\tif err := json.Unmarshal([]byte(kw), &keywords); err == nil {\n\t\t\treturn keywords, nil\n\t\t}\n\t\treturn []types.Keyword{{K: kw, W: 0.5}}, nil\n\t}\n\treturn []types.Keyword{}, nil\n}\n\n// extractKeywordsFromArray extracts keywords from []interface{}\n// Handles both {k, w} objects and plain strings\nfunc (p *AgentProvider) extractKeywordsFromArray(items []interface{}) ([]types.Keyword, error) {\n\tkeywords := make([]types.Keyword, 0, len(items))\n\tfor _, item := range items {\n\t\tswitch v := item.(type) {\n\t\tcase map[string]interface{}:\n\t\t\t// Handle {k: \"keyword\", w: 0.9} format\n\t\t\tk, _ := v[\"k\"].(string)\n\t\t\tw, _ := v[\"w\"].(float64)\n\t\t\tif k != \"\" {\n\t\t\t\tif w == 0 {\n\t\t\t\t\tw = 0.5 // Default weight\n\t\t\t\t}\n\t\t\t\tkeywords = append(keywords, types.Keyword{K: k, W: w})\n\t\t\t}\n\t\tcase string:\n\t\t\t// Plain string, use default weight\n\t\t\tif v != \"\" {\n\t\t\t\tkeywords = append(keywords, types.Keyword{K: v, W: 0.5})\n\t\t\t}\n\t\t}\n\t}\n\treturn keywords, nil\n}\n" }, { "path": "agent/search/nlp/keyword/agent_test.go", "content": "package keyword_test\n\nimport (\n\t\"testing\"\n\n\t\"github.com/stretchr/testify/assert\"\n\t\"github.com/stretchr/testify/require\"\n\t\"github.com/yaoapp/yao/agent/assistant\"\n\t\"github.com/yaoapp/yao/agent/context\"\n\t\"github.com/yaoapp/yao/agent/search/nlp/keyword\"\n\tsearchTypes \"github.com/yaoapp/yao/agent/search/types\"\n\t\"github.com/yaoapp/yao/agent/testutils\"\n\toauthTypes \"github.com/yaoapp/yao/openapi/oauth/types\"\n)\n\nfunc TestAgentProviderWithAssistantConfig(t *testing.T) {\n\t// Skip if running short tests\n\tif testing.Short() {\n\t\tt.Skip(\"Skipping integration test\")\n\t}\n\n\t// Initialize test environment\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\t// Load the keyword-agent assistant that will provide keywords\n\tast, err := assistant.Get(\"tests.keyword-agent\")\n\trequire.NoError(t, err)\n\trequire.NotNil(t, ast)\n\n\t// Create test context\n\tctx := newTestContext(t)\n\n\t// Create extractor with agent mode\n\textractor := keyword.NewExtractor(\"tests.keyword-agent\", &searchTypes.KeywordConfig{\n\t\tMaxKeywords: 5,\n\t\tLanguage: \"auto\",\n\t})\n\n\t// Test extraction\n\tcontent := \"Machine learning and deep learning are subfields of artificial intelligence\"\n\tkeywords, err := extractor.Extract(ctx, content, nil)\n\trequire.NoError(t, err)\n\tassert.NotEmpty(t, keywords, \"Agent should return keywords\")\n\tassert.LessOrEqual(t, len(keywords), 5, \"Should respect max_keywords\")\n\n\t// Verify keywords are relevant\n\tt.Logf(\"Extracted keywords: %v\", keywords)\n}\n\nfunc TestAgentProviderWithCustomOptions(t *testing.T) {\n\t// Skip if running short tests\n\tif testing.Short() {\n\t\tt.Skip(\"Skipping integration test\")\n\t}\n\n\t// Initialize test environment\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\t// Create test context\n\tctx := newTestContext(t)\n\n\t// Create extractor with agent mode\n\textractor := keyword.NewExtractor(\"tests.keyword-agent\", &searchTypes.KeywordConfig{\n\t\tMaxKeywords: 10,\n\t})\n\n\t// Test with runtime options override\n\tcontent := \"Python programming language for data science and web development\"\n\tkeywords, err := extractor.Extract(ctx, content, &searchTypes.KeywordOptions{\n\t\tMaxKeywords: 3, // Override to 3\n\t})\n\trequire.NoError(t, err)\n\tassert.NotEmpty(t, keywords)\n\tassert.LessOrEqual(t, len(keywords), 3, \"Should respect runtime max_keywords override\")\n\n\tt.Logf(\"Extracted keywords (max 3): %v\", keywords)\n}\n\nfunc TestAgentProviderWithoutContext(t *testing.T) {\n\t// Test that agent mode requires context\n\textractor := keyword.NewExtractor(\"tests.keyword-agent\", nil)\n\n\t_, err := extractor.Extract(nil, \"test content\", nil)\n\tassert.Error(t, err, \"Agent mode should require context\")\n\tassert.Contains(t, err.Error(), \"context is required\")\n}\n\nfunc TestAgentProviderAgentNotFound(t *testing.T) {\n\t// Skip if running short tests\n\tif testing.Short() {\n\t\tt.Skip(\"Skipping integration test\")\n\t}\n\n\t// Initialize test environment\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\t// Create test context\n\tctx := newTestContext(t)\n\n\t// Create extractor with non-existent agent\n\textractor := keyword.NewExtractor(\"non-existent-agent\", nil)\n\n\t_, err := extractor.Extract(ctx, \"test content\", nil)\n\tassert.Error(t, err, \"Should error for non-existent agent\")\n\tassert.Contains(t, err.Error(), \"failed to get agent\")\n}\n\n// newTestContext creates a test context with required fields\nfunc newTestContext(t *testing.T) *context.Context {\n\tt.Helper()\n\tauthorized := &oauthTypes.AuthorizedInfo{\n\t\tUserID: \"test-user\",\n\t}\n\tchatID := \"test-chat-keyword\"\n\tctx := context.New(t.Context(), authorized, chatID)\n\treturn ctx\n}\n" }, { "path": "agent/search/nlp/keyword/extractor.go", "content": "// Package keyword provides keyword extraction for web search optimization\n// Supports three modes via uses.keyword configuration:\n// - \"builtin\" or \"\": Uses __yao.keyword system agent (LLM-powered)\n// - \"\": Delegate to a custom LLM-powered assistant\n// - \"mcp:.\": Call external MCP tool\npackage keyword\n\nimport (\n\t\"fmt\"\n\t\"strings\"\n\n\t\"github.com/yaoapp/yao/agent/context\"\n\t\"github.com/yaoapp/yao/agent/search/types\"\n)\n\n// SystemKeywordAgent is the default system agent for keyword extraction\nconst SystemKeywordAgent = \"__yao.keyword\"\n\n// Extractor extracts keywords from text\n// Mode is determined by uses.keyword configuration\ntype Extractor struct {\n\tusesKeyword string // \"builtin\", \"\", \"mcp:.\"\n\tconfig *types.KeywordConfig // Keyword extraction options\n}\n\n// NewExtractor creates a new keyword extractor\n// usesKeyword: value from uses.keyword config\n// cfg: keyword extraction options from search config\nfunc NewExtractor(usesKeyword string, cfg *types.KeywordConfig) *Extractor {\n\treturn &Extractor{\n\t\tusesKeyword: usesKeyword,\n\t\tconfig: cfg,\n\t}\n}\n\n// Extract extracts keywords from content based on configured mode\n// Returns a list of keywords with weights optimized for search queries\nfunc (e *Extractor) Extract(ctx *context.Context, content string, opts *types.KeywordOptions) ([]types.Keyword, error) {\n\t// Merge options with config defaults\n\tmergedOpts := e.mergeOptions(opts)\n\n\tswitch {\n\tcase e.usesKeyword == \"builtin\" || e.usesKeyword == \"\":\n\t\t// Use system keyword agent\n\t\treturn e.agentExtract(ctx, content, SystemKeywordAgent, mergedOpts)\n\tcase strings.HasPrefix(e.usesKeyword, \"mcp:\"):\n\t\treturn e.mcpExtract(ctx, content, mergedOpts)\n\tdefault:\n\t\t// Assume it's an assistant ID for Agent mode\n\t\treturn e.agentExtract(ctx, content, e.usesKeyword, mergedOpts)\n\t}\n}\n\n// mergeOptions merges runtime options with config defaults\nfunc (e *Extractor) mergeOptions(opts *types.KeywordOptions) *types.KeywordOptions {\n\tresult := &types.KeywordOptions{\n\t\tMaxKeywords: 10, // default\n\t\tLanguage: \"auto\", // default\n\t}\n\n\t// Apply config defaults\n\tif e.config != nil {\n\t\tif e.config.MaxKeywords > 0 {\n\t\t\tresult.MaxKeywords = e.config.MaxKeywords\n\t\t}\n\t\tif e.config.Language != \"\" {\n\t\t\tresult.Language = e.config.Language\n\t\t}\n\t}\n\n\t// Apply runtime options (highest priority)\n\tif opts != nil {\n\t\tif opts.MaxKeywords > 0 {\n\t\t\tresult.MaxKeywords = opts.MaxKeywords\n\t\t}\n\t\tif opts.Language != \"\" {\n\t\t\tresult.Language = opts.Language\n\t\t}\n\t}\n\n\treturn result\n}\n\n// agentExtract delegates to an LLM-powered assistant\n// The assistant can understand context and extract semantically relevant keywords\nfunc (e *Extractor) agentExtract(ctx *context.Context, content string, agentID string, opts *types.KeywordOptions) ([]types.Keyword, error) {\n\tif ctx == nil {\n\t\treturn nil, fmt.Errorf(\"context is required for keyword extraction\")\n\t}\n\tprovider := NewAgentProvider(agentID)\n\treturn provider.Extract(ctx, content, opts)\n}\n\n// mcpExtract calls an external MCP tool\n// Format: \"mcp:.\"\nfunc (e *Extractor) mcpExtract(ctx *context.Context, content string, opts *types.KeywordOptions) ([]types.Keyword, error) {\n\tmcpRef := strings.TrimPrefix(e.usesKeyword, \"mcp:\")\n\tprovider, err := NewMCPProvider(mcpRef)\n\tif err != nil {\n\t\t// Fallback to system agent on invalid MCP format\n\t\treturn e.agentExtract(ctx, content, SystemKeywordAgent, e.mergeOptions(nil))\n\t}\n\treturn provider.Extract(ctx, content, opts)\n}\n" }, { "path": "agent/search/nlp/keyword/extractor_test.go", "content": "package keyword_test\n\nimport (\n\t\"testing\"\n\n\t\"github.com/stretchr/testify/assert\"\n\t\"github.com/yaoapp/yao/agent/search/nlp/keyword\"\n\t\"github.com/yaoapp/yao/agent/search/types\"\n)\n\nfunc TestExtractor_BuiltinMode_RequiresContext(t *testing.T) {\n\t// Test builtin mode requires context (now uses __yao.keyword agent)\n\textractor := keyword.NewExtractor(\"builtin\", &types.KeywordConfig{\n\t\tMaxKeywords: 5,\n\t\tLanguage: \"auto\",\n\t})\n\n\t// Without context, should return error\n\t_, err := extractor.Extract(nil, \"How to build a search engine with Elasticsearch?\", nil)\n\tassert.Error(t, err)\n\tassert.Contains(t, err.Error(), \"context is required\")\n}\n\nfunc TestExtractor_EmptyUsesKeyword_RequiresContext(t *testing.T) {\n\t// Empty uses.keyword should default to __yao.keyword agent\n\textractor := keyword.NewExtractor(\"\", nil)\n\n\t// Without context, should return error\n\t_, err := extractor.Extract(nil, \"Machine learning algorithms\", nil)\n\tassert.Error(t, err)\n\tassert.Contains(t, err.Error(), \"context is required\")\n}\n\nfunc TestExtractor_AgentMode_RequiresContext(t *testing.T) {\n\t// Custom agent mode requires context\n\textractor := keyword.NewExtractor(\"custom.keyword.agent\", nil)\n\n\t_, err := extractor.Extract(nil, \"Test query\", nil)\n\tassert.Error(t, err)\n\tassert.Contains(t, err.Error(), \"context is required\")\n}\n\nfunc TestExtractor_MCPMode_InvalidFormat(t *testing.T) {\n\t// Invalid MCP format should fallback to system agent (which requires context)\n\textractor := keyword.NewExtractor(\"mcp:invalid\", nil)\n\n\t_, err := extractor.Extract(nil, \"Test query\", nil)\n\tassert.Error(t, err)\n\tassert.Contains(t, err.Error(), \"context is required\")\n}\n\nfunc TestExtractor_SystemKeywordAgentConstant(t *testing.T) {\n\t// Verify the system keyword agent constant\n\tassert.Equal(t, \"__yao.keyword\", keyword.SystemKeywordAgent)\n}\n" }, { "path": "agent/search/nlp/keyword/mcp.go", "content": "package keyword\n\nimport (\n\t\"encoding/json\"\n\t\"fmt\"\n\t\"strings\"\n\n\t\"github.com/yaoapp/gou/mcp\"\n\tgouMCPTypes \"github.com/yaoapp/gou/mcp/types\"\n\tagentContext \"github.com/yaoapp/yao/agent/context\"\n\t\"github.com/yaoapp/yao/agent/search/types\"\n)\n\n// MCPProvider delegates keyword extraction to an MCP tool\ntype MCPProvider struct {\n\tserverID string // MCP server ID\n\ttoolName string // Tool name to call\n}\n\n// NewMCPProvider creates a new MCP-based keyword extractor\n// mcpRef format: \"server.tool\" (e.g., \"nlp.extract_keywords\")\nfunc NewMCPProvider(mcpRef string) (*MCPProvider, error) {\n\tparts := strings.SplitN(mcpRef, \".\", 2)\n\tif len(parts) != 2 {\n\t\treturn nil, fmt.Errorf(\"invalid MCP format, expected 'server.tool', got '%s'\", mcpRef)\n\t}\n\treturn &MCPProvider{\n\t\tserverID: parts[0],\n\t\ttoolName: parts[1],\n\t}, nil\n}\n\n// Extract extracts keywords by calling the MCP tool\nfunc (p *MCPProvider) Extract(ctx *agentContext.Context, content string, opts *types.KeywordOptions) ([]types.Keyword, error) {\n\t// Get MCP client\n\tclient, err := mcp.Select(p.serverID)\n\tif err != nil {\n\t\treturn nil, fmt.Errorf(\"MCP server '%s' not found: %w\", p.serverID, err)\n\t}\n\n\t// Build arguments for the MCP tool\n\targuments := map[string]interface{}{\n\t\t\"content\": content,\n\t\t\"max_keywords\": opts.MaxKeywords,\n\t\t\"language\": opts.Language,\n\t}\n\n\t// Call the MCP tool (ctx embeds context.Context)\n\tcallResult, err := client.CallTool(ctx, p.toolName, arguments)\n\tif err != nil {\n\t\treturn nil, fmt.Errorf(\"MCP tool call failed: %w\", err)\n\t}\n\n\t// Parse the result\n\treturn p.parseResult(callResult)\n}\n\n// parseResult extracts keywords from the MCP tool response\nfunc (p *MCPProvider) parseResult(result *gouMCPTypes.CallToolResponse) ([]types.Keyword, error) {\n\tif result == nil {\n\t\treturn []types.Keyword{}, nil\n\t}\n\n\t// Check for errors in result\n\tif result.IsError {\n\t\terrMsg := \"MCP tool returned error\"\n\t\tif len(result.Content) > 0 && result.Content[0].Text != \"\" {\n\t\t\terrMsg = result.Content[0].Text\n\t\t}\n\t\treturn nil, fmt.Errorf(\"%s\", errMsg)\n\t}\n\n\t// Parse content - expect JSON data with \"keywords\" field\n\tif len(result.Content) == 0 {\n\t\treturn []types.Keyword{}, nil\n\t}\n\n\t// Try to extract keywords from content\n\tfor _, content := range result.Content {\n\t\t// Check text content type\n\t\tif content.Type == gouMCPTypes.ToolContentTypeText && content.Text != \"\" {\n\t\t\t// Try to parse as JSON\n\t\t\tvar data map[string]interface{}\n\t\t\tif err := json.Unmarshal([]byte(content.Text), &data); err == nil {\n\t\t\t\t// Look for \"keywords\" field\n\t\t\t\tif kw, ok := data[\"keywords\"]; ok {\n\t\t\t\t\treturn p.extractKeywordsFromValue(kw)\n\t\t\t\t}\n\t\t\t}\n\n\t\t\t// Try to parse as direct array of keywords\n\t\t\tvar keywords []types.Keyword\n\t\t\tif err := json.Unmarshal([]byte(content.Text), &keywords); err == nil {\n\t\t\t\treturn keywords, nil\n\t\t\t}\n\t\t}\n\t}\n\n\treturn []types.Keyword{}, nil\n}\n\n// extractKeywordsFromValue extracts Keyword array from various types\nfunc (p *MCPProvider) extractKeywordsFromValue(v interface{}) ([]types.Keyword, error) {\n\tswitch kw := v.(type) {\n\tcase []types.Keyword:\n\t\treturn kw, nil\n\tcase []interface{}:\n\t\tkeywords := make([]types.Keyword, 0, len(kw))\n\t\tfor _, item := range kw {\n\t\t\tswitch v := item.(type) {\n\t\t\tcase map[string]interface{}:\n\t\t\t\t// Handle {k: \"keyword\", w: 0.9} format\n\t\t\t\tk, _ := v[\"k\"].(string)\n\t\t\t\tw, _ := v[\"w\"].(float64)\n\t\t\t\tif k != \"\" {\n\t\t\t\t\tif w == 0 {\n\t\t\t\t\t\tw = 0.5 // Default weight\n\t\t\t\t\t}\n\t\t\t\t\tkeywords = append(keywords, types.Keyword{K: k, W: w})\n\t\t\t\t}\n\t\t\tcase string:\n\t\t\t\t// Plain string, use default weight\n\t\t\t\tif v != \"\" {\n\t\t\t\t\tkeywords = append(keywords, types.Keyword{K: v, W: 0.5})\n\t\t\t\t}\n\t\t\t}\n\t\t}\n\t\treturn keywords, nil\n\tcase string:\n\t\tvar keywords []types.Keyword\n\t\tif err := json.Unmarshal([]byte(kw), &keywords); err == nil {\n\t\t\treturn keywords, nil\n\t\t}\n\t\treturn []types.Keyword{{K: kw, W: 0.5}}, nil\n\t}\n\treturn []types.Keyword{}, nil\n}\n" }, { "path": "agent/search/nlp/keyword/mcp_test.go", "content": "package keyword_test\n\nimport (\n\t\"testing\"\n\n\t\"github.com/stretchr/testify/assert\"\n\t\"github.com/stretchr/testify/require\"\n\t\"github.com/yaoapp/yao/agent/context\"\n\t\"github.com/yaoapp/yao/agent/search/nlp/keyword\"\n\tsearchTypes \"github.com/yaoapp/yao/agent/search/types\"\n\t\"github.com/yaoapp/yao/agent/testutils\"\n\toauthTypes \"github.com/yaoapp/yao/openapi/oauth/types\"\n)\n\nfunc TestMCPProviderWithAssistantConfig(t *testing.T) {\n\t// Skip if running short tests\n\tif testing.Short() {\n\t\tt.Skip(\"Skipping integration test\")\n\t}\n\n\t// Initialize test environment\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\t// Create test context\n\tctx := newMCPTestContext(t)\n\n\t// Create extractor with MCP mode\n\textractor := keyword.NewExtractor(\"mcp:search.extract_keywords\", &searchTypes.KeywordConfig{\n\t\tMaxKeywords: 5,\n\t\tLanguage: \"auto\",\n\t})\n\n\t// Test extraction\n\tcontent := \"Machine learning and deep learning are subfields of artificial intelligence\"\n\tkeywords, err := extractor.Extract(ctx, content, nil)\n\trequire.NoError(t, err)\n\tassert.NotEmpty(t, keywords, \"MCP should return keywords\")\n\tassert.LessOrEqual(t, len(keywords), 5, \"Should respect max_keywords\")\n\n\tt.Logf(\"Extracted keywords via MCP: %v\", keywords)\n}\n\nfunc TestMCPProviderWithCustomOptions(t *testing.T) {\n\t// Skip if running short tests\n\tif testing.Short() {\n\t\tt.Skip(\"Skipping integration test\")\n\t}\n\n\t// Initialize test environment\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\t// Create test context\n\tctx := newMCPTestContext(t)\n\n\t// Create extractor with MCP mode\n\textractor := keyword.NewExtractor(\"mcp:search.extract_keywords\", &searchTypes.KeywordConfig{\n\t\tMaxKeywords: 10,\n\t})\n\n\t// Test with runtime options override\n\tcontent := \"Python programming language for data science and web development\"\n\tkeywords, err := extractor.Extract(ctx, content, &searchTypes.KeywordOptions{\n\t\tMaxKeywords: 3, // Override to 3\n\t})\n\trequire.NoError(t, err)\n\tassert.NotEmpty(t, keywords)\n\tassert.LessOrEqual(t, len(keywords), 3, \"Should respect runtime max_keywords override\")\n\n\tt.Logf(\"Extracted keywords via MCP (max 3): %v\", keywords)\n}\n\nfunc TestMCPProviderInvalidFormat(t *testing.T) {\n\t// Test invalid MCP format fallback to system agent (requires context)\n\textractor := keyword.NewExtractor(\"mcp:invalid\", nil)\n\n\t// Should fallback to system agent which requires context\n\t_, err := extractor.Extract(nil, \"test content for keyword extraction\", nil)\n\tassert.Error(t, err)\n\tassert.Contains(t, err.Error(), \"context is required\")\n}\n\nfunc TestMCPProviderServerNotFound(t *testing.T) {\n\t// Skip if running short tests\n\tif testing.Short() {\n\t\tt.Skip(\"Skipping integration test\")\n\t}\n\n\t// Initialize test environment\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\t// Create test context\n\tctx := newMCPTestContext(t)\n\n\t// Create extractor with non-existent MCP server\n\textractor := keyword.NewExtractor(\"mcp:nonexistent.extract_keywords\", &searchTypes.KeywordConfig{})\n\n\t_, err := extractor.Extract(ctx, \"test content\", nil)\n\tassert.Error(t, err, \"Should error for non-existent MCP server\")\n\tassert.Contains(t, err.Error(), \"not found\")\n}\n\nfunc TestMCPProviderToolNotFound(t *testing.T) {\n\t// Skip if running short tests\n\tif testing.Short() {\n\t\tt.Skip(\"Skipping integration test\")\n\t}\n\n\t// Initialize test environment\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\t// Create test context\n\tctx := newMCPTestContext(t)\n\n\t// Create extractor with non-existent tool\n\textractor := keyword.NewExtractor(\"mcp:search.nonexistent_tool\", &searchTypes.KeywordConfig{})\n\n\t_, err := extractor.Extract(ctx, \"test content\", nil)\n\tassert.Error(t, err, \"Should error for non-existent MCP tool\")\n}\n\nfunc TestMCPProviderEmptyContent(t *testing.T) {\n\t// Skip if running short tests\n\tif testing.Short() {\n\t\tt.Skip(\"Skipping integration test\")\n\t}\n\n\t// Initialize test environment\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\t// Create test context\n\tctx := newMCPTestContext(t)\n\n\t// Create extractor with MCP mode\n\textractor := keyword.NewExtractor(\"mcp:search.extract_keywords\", nil)\n\n\t// Test with empty content - MCP tool should return error\n\t_, err := extractor.Extract(ctx, \"\", nil)\n\tassert.Error(t, err, \"Should error for empty content\")\n}\n\n// newMCPTestContext creates a test context for MCP tests\nfunc newMCPTestContext(t *testing.T) *context.Context {\n\tt.Helper()\n\tauthorized := &oauthTypes.AuthorizedInfo{\n\t\tUserID: \"test-user\",\n\t}\n\tchatID := \"test-chat-mcp-keyword\"\n\tctx := context.New(t.Context(), authorized, chatID)\n\treturn ctx\n}\n" }, { "path": "agent/search/nlp/querydsl/agent.go", "content": "package querydsl\n\nimport (\n\t\"encoding/json\"\n\t\"fmt\"\n\n\t\"github.com/yaoapp/gou/query/gou\"\n\t\"github.com/yaoapp/gou/query/linter\"\n\t\"github.com/yaoapp/yao/agent/caller\"\n\tagentContext \"github.com/yaoapp/yao/agent/context\"\n)\n\n// AgentProvider delegates QueryDSL generation to an LLM-powered assistant\n// The assistant can understand context and generate semantically correct QueryDSL\ntype AgentProvider struct {\n\tagentID string // Assistant ID to delegate to\n}\n\n// NewAgentProvider creates a new agent-based QueryDSL generator\nfunc NewAgentProvider(agentID string) *AgentProvider {\n\treturn &AgentProvider{\n\t\tagentID: agentID,\n\t}\n}\n\n// Generate generates QueryDSL by calling the target agent with retry and lint validation\n// The agent receives the query and schema, returns generated QueryDSL\nfunc (p *AgentProvider) Generate(ctx *agentContext.Context, input *Input) (*Result, error) {\n\tif ctx == nil {\n\t\treturn nil, fmt.Errorf(\"context is required for agent QueryDSL generation\")\n\t}\n\n\t// Check if AgentGetterFunc is initialized\n\tif caller.AgentGetterFunc == nil {\n\t\treturn nil, fmt.Errorf(\"AgentGetterFunc not initialized\")\n\t}\n\n\t// Get the agent\n\tagent, err := caller.AgentGetterFunc(p.agentID)\n\tif err != nil {\n\t\treturn nil, fmt.Errorf(\"failed to get agent %s: %w\", p.agentID, err)\n\t}\n\n\tvar lastError error\n\tvar lastLintErrors string\n\n\tfor attempt := 1; attempt <= MaxRetries; attempt++ {\n\t\t// Build the request message in the format expected by querydsl agent\n\t\trequestMessage := p.buildRequestMessage(input, attempt, lastLintErrors)\n\n\t\t// Create message for the agent\n\t\tmessages := []agentContext.Message{\n\t\t\t{\n\t\t\t\tRole: \"user\",\n\t\t\t\tContent: requestMessage,\n\t\t\t},\n\t\t}\n\n\t\t// Call the agent with skip options (no history, no output)\n\t\toptions := &agentContext.Options{\n\t\t\tSkip: &agentContext.Skip{\n\t\t\t\tHistory: true,\n\t\t\t\tOutput: true,\n\t\t\t},\n\t\t}\n\n\t\tresult, err := agent.Stream(ctx, messages, options)\n\t\tif err != nil {\n\t\t\tlastError = fmt.Errorf(\"agent call failed: %w\", err)\n\t\t\tcontinue\n\t\t}\n\n\t\t// Parse the result from response\n\t\tgenResult, err := p.parseResponse(result)\n\t\tif err != nil {\n\t\t\tlastError = err\n\t\t\tcontinue\n\t\t}\n\n\t\t// Validate with linter if DSL is present\n\t\tif genResult.DSL != nil {\n\t\t\tlintResult := p.validateDSL(genResult.DSL)\n\t\t\tif lintResult.Valid {\n\t\t\t\treturn genResult, nil\n\t\t\t}\n\n\t\t\t// Lint failed, prepare error message for retry\n\t\t\tlastLintErrors = lintResult.FormatDiagnostics()\n\t\t\tlastError = fmt.Errorf(\"QueryDSL validation failed: %s\", lastLintErrors)\n\n\t\t\t// Add lint warnings to result warnings\n\t\t\tfor _, diag := range lintResult.Diagnostics {\n\t\t\t\tgenResult.Warnings = append(genResult.Warnings, fmt.Sprintf(\"[%s] %s: %s\", diag.Code, diag.Path, diag.Message))\n\t\t\t}\n\t\t\tcontinue\n\t\t}\n\n\t\t// No DSL returned\n\t\tlastError = fmt.Errorf(\"no QueryDSL returned from agent\")\n\t}\n\n\treturn nil, fmt.Errorf(\"QueryDSL generation failed after %d attempts: %w\", MaxRetries, lastError)\n}\n\n// buildRequestMessage constructs the request message for the agent\n// Returns JSON format for structured communication with the agent\nfunc (p *AgentProvider) buildRequestMessage(input *Input, attempt int, lastLintErrors string) string {\n\t// Build request data as JSON\n\trequestData := map[string]interface{}{\n\t\t\"query\": input.Query,\n\t\t\"models\": input.ModelIDs,\n\t\t\"limit\": input.Limit,\n\t}\n\n\t// Add schema from extra params if provided\n\tif input.ExtraParams != nil {\n\t\tif schema, ok := input.ExtraParams[\"schema\"]; ok {\n\t\t\trequestData[\"schema\"] = schema\n\t\t}\n\t}\n\n\t// Add scenario hint if specified (filter, aggregation, join, complex)\n\tif input.Scenario != \"\" {\n\t\trequestData[\"scenario\"] = string(input.Scenario)\n\t}\n\n\t// Add allowed fields if specified\n\tif len(input.AllowedFields) > 0 {\n\t\trequestData[\"allowed_fields\"] = input.AllowedFields\n\t}\n\n\t// Add retry context if this is a retry attempt\n\tif attempt > 1 && lastLintErrors != \"\" {\n\t\trequestData[\"retry\"] = map[string]interface{}{\n\t\t\t\"attempt\": attempt,\n\t\t\t\"lint_errors\": lastLintErrors,\n\t\t\t\"instructions\": \"The previous QueryDSL was invalid. Please fix the errors and regenerate.\",\n\t\t}\n\t}\n\n\tjsonBytes, _ := json.Marshal(requestData)\n\treturn string(jsonBytes)\n}\n\n// validateDSL validates the generated QueryDSL using the linter\nfunc (p *AgentProvider) validateDSL(dsl *gou.QueryDSL) *linter.LintResult {\n\t// Marshal DSL to JSON for linting\n\tjsonBytes, err := json.Marshal(dsl)\n\tif err != nil {\n\t\tresult := &linter.LintResult{Valid: false}\n\t\treturn result\n\t}\n\n\t_, lintResult := linter.Parse(string(jsonBytes))\n\treturn lintResult\n}\n\n// parseResponse extracts QueryDSL from the agent's *context.Response\n// Now that agent.Stream() returns *context.Response directly,\n// we can access fields without type assertions.\n//\n// The querydsl agent returns QueryDSL in response.Next field\n// Or returns error JSON: {\"error\": \"code\", \"message\": \"...\"}\nfunc (p *AgentProvider) parseResponse(response *agentContext.Response) (*Result, error) {\n\tif response == nil {\n\t\treturn &Result{}, nil\n\t}\n\n\t// Check Next field first (custom hook data)\n\tif response.Next != nil {\n\t\treturn p.parseNextData(response.Next)\n\t}\n\n\t// No Next data, return empty result\n\treturn &Result{}, nil\n}\n\n// parseNextData extracts QueryDSL from Next hook data\nfunc (p *AgentProvider) parseNextData(next interface{}) (*Result, error) {\n\tif next == nil {\n\t\treturn &Result{}, nil\n\t}\n\n\t// Try to convert to map first\n\tvar data map[string]interface{}\n\n\tswitch v := next.(type) {\n\tcase map[string]interface{}:\n\t\tdata = v\n\tcase string:\n\t\t// Try to parse as JSON\n\t\tif err := json.Unmarshal([]byte(v), &data); err != nil {\n\t\t\treturn nil, fmt.Errorf(\"failed to parse agent response: %w\", err)\n\t\t}\n\tdefault:\n\t\t// Try to marshal and unmarshal\n\t\tjsonBytes, err := json.Marshal(next)\n\t\tif err != nil {\n\t\t\treturn &Result{}, nil\n\t\t}\n\t\tif err := json.Unmarshal(jsonBytes, &data); err != nil {\n\t\t\treturn &Result{}, nil\n\t\t}\n\t}\n\n\tgenResult := &Result{}\n\n\t// Check for error response: {\"error\": \"code\", \"message\": \"...\"}\n\tif errCode, hasError := data[\"error\"]; hasError {\n\t\terrMsg := \"\"\n\t\tif msg, ok := data[\"message\"].(string); ok {\n\t\t\terrMsg = msg\n\t\t}\n\t\treturn nil, fmt.Errorf(\"QueryDSL generation error [%v]: %s\", errCode, errMsg)\n\t}\n\n\t// Check if this is a direct QueryDSL (has \"from\" or \"select\" field)\n\t// The querydsl agent returns QueryDSL directly, e.g., {\"select\": [...], \"from\": \"table\", ...}\n\tif _, hasFrom := data[\"from\"]; hasFrom {\n\t\tgenResult.DSL = p.extractDSL(data)\n\t\treturn genResult, nil\n\t}\n\tif _, hasSelect := data[\"select\"]; hasSelect {\n\t\tgenResult.DSL = p.extractDSL(data)\n\t\treturn genResult, nil\n\t}\n\n\t// Check for \"dsl\" field wrapper: { dsl: {...} }\n\tif dsl, ok := data[\"dsl\"]; ok {\n\t\tgenResult.DSL = p.extractDSL(dsl)\n\t\tif explain, ok := data[\"explain\"].(string); ok {\n\t\t\tgenResult.Explain = explain\n\t\t}\n\t\tif warnings, ok := data[\"warnings\"]; ok {\n\t\t\tgenResult.Warnings = p.extractWarnings(warnings)\n\t\t}\n\t\treturn genResult, nil\n\t}\n\n\t// Check for \"data\" field wrapper: { data: { dsl: {...}, explain: \"...\", warnings: [] } }\n\tif d, ok := data[\"data\"]; ok {\n\t\tif dm, ok := d.(map[string]interface{}); ok {\n\t\t\t// Check if data.data contains dsl field: { data: { dsl: {...} } }\n\t\t\tif dsl, ok := dm[\"dsl\"]; ok {\n\t\t\t\tgenResult.DSL = p.extractDSL(dsl)\n\t\t\t} else if _, hasFrom := dm[\"from\"]; hasFrom {\n\t\t\t\t// data.data is directly a QueryDSL (from __yao.querydsl Next hook)\n\t\t\t\tgenResult.DSL = p.extractDSL(dm)\n\t\t\t} else if _, hasSelect := dm[\"select\"]; hasSelect {\n\t\t\t\t// data.data is directly a QueryDSL\n\t\t\t\tgenResult.DSL = p.extractDSL(dm)\n\t\t\t}\n\t\t\t// Extract explain and warnings from data.data\n\t\t\tif explain, ok := dm[\"explain\"].(string); ok {\n\t\t\t\tgenResult.Explain = explain\n\t\t\t}\n\t\t\tif warnings, ok := dm[\"warnings\"]; ok {\n\t\t\t\tgenResult.Warnings = p.extractWarnings(warnings)\n\t\t\t}\n\t\t\treturn genResult, nil\n\t\t}\n\t}\n\n\t// Fallback: Get explain and warnings from top level\n\tif explain, ok := data[\"explain\"].(string); ok {\n\t\tgenResult.Explain = explain\n\t}\n\tif warnings, ok := data[\"warnings\"]; ok {\n\t\tgenResult.Warnings = p.extractWarnings(warnings)\n\t}\n\n\treturn genResult, nil\n}\n\n// extractDSL converts interface{} to gou.QueryDSL\nfunc (p *AgentProvider) extractDSL(v interface{}) *gou.QueryDSL {\n\tif v == nil {\n\t\treturn nil\n\t}\n\n\t// Marshal and unmarshal to gou.QueryDSL\n\tjsonBytes, err := json.Marshal(v)\n\tif err != nil {\n\t\treturn nil\n\t}\n\n\tvar dsl gou.QueryDSL\n\tif err := json.Unmarshal(jsonBytes, &dsl); err != nil {\n\t\treturn nil\n\t}\n\n\treturn &dsl\n}\n\n// extractWarnings extracts warnings array from various types\nfunc (p *AgentProvider) extractWarnings(v interface{}) []string {\n\tswitch w := v.(type) {\n\tcase []string:\n\t\treturn w\n\tcase []interface{}:\n\t\twarnings := make([]string, 0, len(w))\n\t\tfor _, item := range w {\n\t\t\tif s, ok := item.(string); ok {\n\t\t\t\twarnings = append(warnings, s)\n\t\t\t}\n\t\t}\n\t\treturn warnings\n\tcase string:\n\t\treturn []string{w}\n\t}\n\treturn nil\n}\n" }, { "path": "agent/search/nlp/querydsl/agent_test.go", "content": "package querydsl_test\n\nimport (\n\t\"testing\"\n\n\t\"github.com/stretchr/testify/assert\"\n\t\"github.com/stretchr/testify/require\"\n\t\"github.com/yaoapp/yao/agent/assistant\"\n\t\"github.com/yaoapp/yao/agent/context\"\n\t\"github.com/yaoapp/yao/agent/search/nlp/querydsl\"\n\t\"github.com/yaoapp/yao/agent/testutils\"\n\toauthTypes \"github.com/yaoapp/yao/openapi/oauth/types\"\n)\n\nfunc TestNewAgentProvider(t *testing.T) {\n\tt.Run(\"create_provider\", func(t *testing.T) {\n\t\tprovider := querydsl.NewAgentProvider(\"tests.querydsl-agent\")\n\t\tassert.NotNil(t, provider)\n\t})\n}\n\nfunc TestAgentProvider_Generate(t *testing.T) {\n\t// Skip if running short tests\n\tif testing.Short() {\n\t\tt.Skip(\"Skipping integration test\")\n\t}\n\n\t// Initialize test environment\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\t// Load the querydsl-agent assistant\n\tast, err := assistant.Get(\"tests.querydsl-agent\")\n\trequire.NoError(t, err)\n\trequire.NotNil(t, ast)\n\n\t// Create test context\n\tctx := newTestContext(t)\n\n\t// Create Agent provider for tests.querydsl-agent\n\tprovider := querydsl.NewAgentProvider(\"tests.querydsl-agent\")\n\tassert.NotNil(t, provider)\n\n\tt.Run(\"verify_fixed_structure\", func(t *testing.T) {\n\t\tinput := &querydsl.Input{\n\t\t\tQuery: \"find active users\",\n\t\t\tModelIDs: []string{\"user\"},\n\t\t\tLimit: 15,\n\t\t}\n\n\t\tresult, err := provider.Generate(ctx, input)\n\t\tif err != nil {\n\t\t\tt.Logf(\"Generate error: %v\", err)\n\t\t}\n\t\trequire.NoError(t, err)\n\t\trequire.NotNil(t, result)\n\t\trequire.NotNil(t, result.DSL, \"DSL should not be nil\")\n\n\t\t// Verify fixed DSL structure from mock\n\t\t// select: [\"id\", \"name\", \"status\", \"created_at\"]\n\t\tassert.Len(t, result.DSL.Select, 4)\n\t\tif len(result.DSL.Select) >= 4 {\n\t\t\tassert.Equal(t, \"id\", result.DSL.Select[0].Field)\n\t\t\tassert.Equal(t, \"name\", result.DSL.Select[1].Field)\n\t\t\tassert.Equal(t, \"status\", result.DSL.Select[2].Field)\n\t\t\tassert.Equal(t, \"created_at\", result.DSL.Select[3].Field)\n\t\t}\n\n\t\t// wheres: [{ field: \"status\", op: \"=\", value: \"active\" }]\n\t\tassert.Len(t, result.DSL.Wheres, 1)\n\t\tif len(result.DSL.Wheres) > 0 {\n\t\t\tassert.Equal(t, \"status\", result.DSL.Wheres[0].Field.Field)\n\t\t\tassert.Equal(t, \"=\", result.DSL.Wheres[0].OP)\n\t\t\tassert.Equal(t, \"active\", result.DSL.Wheres[0].Value)\n\t\t}\n\n\t\t// orders: [{ field: \"created_at\", sort: \"desc\" }]\n\t\tassert.Len(t, result.DSL.Orders, 1)\n\t\tif len(result.DSL.Orders) > 0 {\n\t\t\tassert.Equal(t, \"created_at\", result.DSL.Orders[0].Field.Field)\n\t\t\tassert.Equal(t, \"desc\", result.DSL.Orders[0].Sort)\n\t\t}\n\n\t\t// limit: 15 (from input)\n\t\tassert.Equal(t, float64(15), result.DSL.Limit)\n\n\t\t// explain should contain query\n\t\tassert.Contains(t, result.Explain, \"find active users\")\n\n\t\t// warnings should be empty\n\t\tassert.Empty(t, result.Warnings)\n\t})\n}\n\nfunc TestAgentProvider_Generate_Error(t *testing.T) {\n\t// Skip if running short tests\n\tif testing.Short() {\n\t\tt.Skip(\"Skipping integration test\")\n\t}\n\n\t// Initialize test environment\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\t// Create test context\n\tctx := newTestContext(t)\n\n\tt.Run(\"non-existent_agent\", func(t *testing.T) {\n\t\tprovider := querydsl.NewAgentProvider(\"tests.nonexistent-agent\")\n\t\tresult, err := provider.Generate(ctx, &querydsl.Input{\n\t\t\tQuery: \"test\",\n\t\t\tModelIDs: []string{\"user\"},\n\t\t})\n\t\tassert.Error(t, err)\n\t\tassert.Nil(t, result)\n\t\tassert.Contains(t, err.Error(), \"failed to get agent\")\n\t})\n\n\tt.Run(\"nil_context\", func(t *testing.T) {\n\t\tprovider := querydsl.NewAgentProvider(\"tests.querydsl-agent\")\n\t\tresult, err := provider.Generate(nil, &querydsl.Input{\n\t\t\tQuery: \"test\",\n\t\t\tModelIDs: []string{\"user\"},\n\t\t})\n\t\tassert.Error(t, err)\n\t\tassert.Nil(t, result)\n\t\tassert.Contains(t, err.Error(), \"context is required\")\n\t})\n}\n\nfunc TestGenerator_Agent_Integration(t *testing.T) {\n\t// Skip if running short tests\n\tif testing.Short() {\n\t\tt.Skip(\"Skipping integration test\")\n\t}\n\n\t// Initialize test environment\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\t// Create test context\n\tctx := newTestContext(t)\n\n\t// Create generator with Agent mode (assistant ID without mcp: prefix)\n\tgen := querydsl.NewGenerator(\"tests.querydsl-agent\", nil)\n\n\tt.Run(\"generate_via_agent\", func(t *testing.T) {\n\t\tinput := &querydsl.Input{\n\t\t\tQuery: \"find active users\",\n\t\t\tModelIDs: []string{\"user\"},\n\t\t\tLimit: 10,\n\t\t}\n\n\t\tresult, err := gen.Generate(ctx, input)\n\t\trequire.NoError(t, err)\n\t\trequire.NotNil(t, result)\n\t\trequire.NotNil(t, result.DSL)\n\n\t\t// Verify structure from agent mock\n\t\tassert.Len(t, result.DSL.Select, 4)\n\t\tassert.Len(t, result.DSL.Wheres, 1)\n\t\tassert.Len(t, result.DSL.Orders, 1)\n\t\tassert.Contains(t, result.Explain, \"find active users\")\n\t})\n\n\tt.Run(\"allowed_fields_validation\", func(t *testing.T) {\n\t\tinput := &querydsl.Input{\n\t\t\tQuery: \"find users\",\n\t\t\tModelIDs: []string{\"user\"},\n\t\t\tAllowedFields: []string{\"id\", \"name\"}, // Only allow id and name\n\t\t\tLimit: 10,\n\t\t}\n\n\t\tresult, err := gen.Generate(ctx, input)\n\t\trequire.NoError(t, err)\n\t\trequire.NotNil(t, result)\n\t\trequire.NotNil(t, result.DSL)\n\n\t\t// \"status\" and \"created_at\" fields should be filtered out from select\n\t\t// since they are not in AllowedFields\n\t\tfor _, expr := range result.DSL.Select {\n\t\t\tassert.Contains(t, []string{\"id\", \"name\"}, expr.Field)\n\t\t}\n\n\t\t// Should have warning about removed fields\n\t\tassert.NotEmpty(t, result.Warnings)\n\t})\n}\n\nfunc TestAgentProvider_Generate_WithRetry(t *testing.T) {\n\t// Skip if running short tests\n\tif testing.Short() {\n\t\tt.Skip(\"Skipping integration test\")\n\t}\n\n\t// Initialize test environment\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\t// Load the querydsl-agent-retry assistant\n\tast, err := assistant.Get(\"tests.querydsl-agent-retry\")\n\trequire.NoError(t, err)\n\trequire.NotNil(t, ast)\n\n\t// Create test context\n\tctx := newTestContext(t)\n\n\t// Create Agent provider for tests.querydsl-agent-retry\n\t// This agent returns invalid DSL on first call, valid on second\n\tprovider := querydsl.NewAgentProvider(\"tests.querydsl-agent-retry\")\n\tassert.NotNil(t, provider)\n\n\tt.Run(\"retry_on_lint_failure\", func(t *testing.T) {\n\t\tinput := &querydsl.Input{\n\t\t\tQuery: \"test retry mechanism\",\n\t\t\tModelIDs: []string{\"user\"},\n\t\t\tLimit: 10,\n\t\t}\n\n\t\t// This should succeed after retry\n\t\t// First call returns invalid DSL (missing 'from')\n\t\t// Second call (with lint_errors) returns valid DSL\n\t\tresult, err := provider.Generate(ctx, input)\n\t\trequire.NoError(t, err)\n\t\trequire.NotNil(t, result)\n\n\t\tif result.DSL != nil {\n\t\t\t// Should have valid DSL after retry\n\t\t\tassert.NotNil(t, result.DSL.From, \"DSL should have 'from' field after retry\")\n\t\t\t// Explain should indicate this was fixed after receiving lint errors\n\t\t\tassert.Contains(t, result.Explain, \"fixed after receiving lint errors\")\n\t\t}\n\t})\n}\n\n// newTestContext creates a test context with required fields\nfunc newTestContext(t *testing.T) *context.Context {\n\tt.Helper()\n\tauthorized := &oauthTypes.AuthorizedInfo{\n\t\tUserID: \"test-user\",\n\t}\n\tchatID := \"test-chat-querydsl\"\n\tctx := context.New(t.Context(), authorized, chatID)\n\treturn ctx\n}\n" }, { "path": "agent/search/nlp/querydsl/generator.go", "content": "// Package querydsl provides QueryDSL generation from natural language for DB search\n// Supports three modes via uses.querydsl configuration:\n// - \"builtin\" or \"\": Uses __yao.querydsl system agent (LLM-powered)\n// - \"\": Delegate to a custom LLM-powered assistant\n// - \"mcp:.\": Call external MCP tool\npackage querydsl\n\nimport (\n\t\"fmt\"\n\t\"strings\"\n\n\t\"github.com/yaoapp/gou/query/gou\"\n\t\"github.com/yaoapp/yao/agent/context\"\n\t\"github.com/yaoapp/yao/agent/search/types\"\n)\n\n// SystemQueryDSLAgent is the default system agent for QueryDSL generation\nconst SystemQueryDSLAgent = \"__yao.querydsl\"\n\n// Generator generates QueryDSL from natural language\n// Mode is determined by uses.querydsl configuration\ntype Generator struct {\n\tusesQueryDSL string // \"builtin\", \"\", \"mcp:.\"\n\tconfig *types.QueryDSLConfig // QueryDSL generation options\n}\n\n// NewGenerator creates a new QueryDSL generator\n// usesQueryDSL: value from uses.querydsl config\n// cfg: QueryDSL generation options from search config\nfunc NewGenerator(usesQueryDSL string, cfg *types.QueryDSLConfig) *Generator {\n\treturn &Generator{\n\t\tusesQueryDSL: usesQueryDSL,\n\t\tconfig: cfg,\n\t}\n}\n\n// Generate generates QueryDSL from natural language based on configured mode\n// Returns a QueryDSL ready for execution\nfunc (g *Generator) Generate(ctx *context.Context, input *Input) (*Result, error) {\n\tvar result *Result\n\tvar err error\n\n\tswitch {\n\tcase g.usesQueryDSL == \"builtin\" || g.usesQueryDSL == \"\":\n\t\t// Use system querydsl agent\n\t\tresult, err = g.agentGenerate(ctx, input, SystemQueryDSLAgent)\n\tcase strings.HasPrefix(g.usesQueryDSL, \"mcp:\"):\n\t\tresult, err = g.mcpGenerate(ctx, input)\n\tdefault:\n\t\t// Assume it's an assistant ID for Agent mode\n\t\tresult, err = g.agentGenerate(ctx, input, g.usesQueryDSL)\n\t}\n\n\tif err != nil {\n\t\treturn nil, err\n\t}\n\n\t// Validate generated DSL against allowed fields whitelist\n\tif result != nil && result.DSL != nil && len(input.AllowedFields) > 0 {\n\t\tresult = g.validateFields(result, input.AllowedFields)\n\t}\n\n\treturn result, nil\n}\n\n// agentGenerate delegates to an LLM-powered assistant\n// The assistant can understand context and generate semantically correct QueryDSL\nfunc (g *Generator) agentGenerate(ctx *context.Context, input *Input, agentID string) (*Result, error) {\n\tif ctx == nil {\n\t\treturn nil, fmt.Errorf(\"context is required for QueryDSL generation\")\n\t}\n\tprovider := NewAgentProvider(agentID)\n\treturn provider.Generate(ctx, input)\n}\n\n// mcpGenerate calls an external MCP tool\n// Format: \"mcp:.\"\nfunc (g *Generator) mcpGenerate(ctx *context.Context, input *Input) (*Result, error) {\n\tmcpRef := strings.TrimPrefix(g.usesQueryDSL, \"mcp:\")\n\tprovider, err := NewMCPProvider(mcpRef)\n\tif err != nil {\n\t\t// Fallback to system agent on invalid MCP format\n\t\treturn g.agentGenerate(ctx, input, SystemQueryDSLAgent)\n\t}\n\treturn provider.Generate(ctx, input)\n}\n\n// validateFields validates that all fields in the generated DSL are in the allowed list\n// If a field is not allowed, it's removed and a warning is added\nfunc (g *Generator) validateFields(result *Result, allowedFields []string) *Result {\n\tif result.DSL == nil {\n\t\treturn result\n\t}\n\n\t// Build allowed fields set for fast lookup\n\tallowed := make(map[string]bool)\n\tfor _, f := range allowedFields {\n\t\tallowed[f] = true\n\t}\n\n\tvar removedFields []string\n\n\t// Validate Select fields\n\tif len(result.DSL.Select) > 0 {\n\t\tvalidSelect := make([]gou.Expression, 0, len(result.DSL.Select))\n\t\tfor _, expr := range result.DSL.Select {\n\t\t\tif allowed[expr.Field] {\n\t\t\t\tvalidSelect = append(validSelect, expr)\n\t\t\t} else if expr.Field != \"\" {\n\t\t\t\tremovedFields = append(removedFields, \"select:\"+expr.Field)\n\t\t\t}\n\t\t}\n\t\tresult.DSL.Select = validSelect\n\t}\n\n\t// Validate Where fields (recursive)\n\tresult.DSL.Wheres = g.validateWheres(result.DSL.Wheres, allowed, &removedFields)\n\n\t// Validate Order fields\n\tif len(result.DSL.Orders) > 0 {\n\t\tvalidOrders := make(gou.Orders, 0, len(result.DSL.Orders))\n\t\tfor _, order := range result.DSL.Orders {\n\t\t\tif order.Field != nil && allowed[order.Field.Field] {\n\t\t\t\tvalidOrders = append(validOrders, order)\n\t\t\t} else if order.Field != nil && order.Field.Field != \"\" {\n\t\t\t\tremovedFields = append(removedFields, \"order:\"+order.Field.Field)\n\t\t\t}\n\t\t}\n\t\tresult.DSL.Orders = validOrders\n\t}\n\n\t// Add warnings for removed fields\n\tif len(removedFields) > 0 {\n\t\twarning := \"removed fields not in allowed list: \" + strings.Join(removedFields, \", \")\n\t\tresult.Warnings = append(result.Warnings, warning)\n\t}\n\n\treturn result\n}\n\n// validateWheres recursively validates where conditions\nfunc (g *Generator) validateWheres(wheres []gou.Where, allowed map[string]bool, removedFields *[]string) []gou.Where {\n\tif len(wheres) == 0 {\n\t\treturn wheres\n\t}\n\n\tvalidWheres := make([]gou.Where, 0, len(wheres))\n\tfor _, w := range wheres {\n\t\t// Check if the field is allowed\n\t\tfieldAllowed := true\n\t\tif w.Field != nil && w.Field.Field != \"\" {\n\t\t\tif !allowed[w.Field.Field] {\n\t\t\t\t*removedFields = append(*removedFields, \"where:\"+w.Field.Field)\n\t\t\t\tfieldAllowed = false\n\t\t\t}\n\t\t}\n\n\t\tif fieldAllowed {\n\t\t\t// Recursively validate nested wheres\n\t\t\tif len(w.Wheres) > 0 {\n\t\t\t\tw.Wheres = g.validateWheres(w.Wheres, allowed, removedFields)\n\t\t\t}\n\t\t\tvalidWheres = append(validWheres, w)\n\t\t}\n\t}\n\n\treturn validWheres\n}\n" }, { "path": "agent/search/nlp/querydsl/generator_test.go", "content": "package querydsl\n\nimport (\n\t\"testing\"\n\n\t\"github.com/stretchr/testify/assert\"\n\t\"github.com/yaoapp/gou/query/gou\"\n\t\"github.com/yaoapp/yao/agent/search/types\"\n)\n\nfunc TestNewGenerator(t *testing.T) {\n\ttests := []struct {\n\t\tname string\n\t\tusesQueryDSL string\n\t\tconfig *types.QueryDSLConfig\n\t}{\n\t\t{\n\t\t\tname: \"builtin mode\",\n\t\t\tusesQueryDSL: \"builtin\",\n\t\t\tconfig: nil,\n\t\t},\n\t\t{\n\t\t\tname: \"empty defaults to builtin\",\n\t\t\tusesQueryDSL: \"\",\n\t\t\tconfig: nil,\n\t\t},\n\t\t{\n\t\t\tname: \"agent mode\",\n\t\t\tusesQueryDSL: \"my-querydsl-agent\",\n\t\t\tconfig: &types.QueryDSLConfig{Strict: true},\n\t\t},\n\t\t{\n\t\t\tname: \"mcp mode\",\n\t\t\tusesQueryDSL: \"mcp:nlp.generate_querydsl\",\n\t\t\tconfig: nil,\n\t\t},\n\t}\n\n\tfor _, tt := range tests {\n\t\tt.Run(tt.name, func(t *testing.T) {\n\t\t\tgen := NewGenerator(tt.usesQueryDSL, tt.config)\n\t\t\tassert.NotNil(t, gen)\n\t\t\tassert.Equal(t, tt.usesQueryDSL, gen.usesQueryDSL)\n\t\t\tassert.Equal(t, tt.config, gen.config)\n\t\t})\n\t}\n}\n\nfunc TestGenerator_Generate_Builtin_RequiresContext(t *testing.T) {\n\t// Builtin mode now uses __yao.querydsl agent which requires context\n\tgen := NewGenerator(\"builtin\", nil)\n\n\tinput := &Input{\n\t\tQuery: \"find all active users\",\n\t\tModelIDs: []string{\"user\"},\n\t\tLimit: 10,\n\t}\n\n\t// Without context, should return error\n\t_, err := gen.Generate(nil, input)\n\tassert.Error(t, err)\n\tassert.Contains(t, err.Error(), \"context is required\")\n}\n\nfunc TestGenerator_Generate_EmptyMode_RequiresContext(t *testing.T) {\n\t// Empty mode defaults to __yao.querydsl agent which requires context\n\tgen := NewGenerator(\"\", nil)\n\n\tinput := &Input{\n\t\tQuery: \"search products\",\n\t\tModelIDs: []string{\"product\"},\n\t\tLimit: 5,\n\t}\n\n\t// Without context, should return error\n\t_, err := gen.Generate(nil, input)\n\tassert.Error(t, err)\n\tassert.Contains(t, err.Error(), \"context is required\")\n}\n\nfunc TestGenerator_Generate_AgentMode_RequiresContext(t *testing.T) {\n\t// Custom agent mode requires context\n\tgen := NewGenerator(\"custom.querydsl.agent\", nil)\n\n\tinput := &Input{\n\t\tQuery: \"find users\",\n\t\tModelIDs: []string{\"user\"},\n\t\tLimit: 10,\n\t}\n\n\t_, err := gen.Generate(nil, input)\n\tassert.Error(t, err)\n\tassert.Contains(t, err.Error(), \"context is required\")\n}\n\nfunc TestGenerator_Generate_MCPMode_InvalidFormat(t *testing.T) {\n\t// Invalid MCP format should fallback to system agent (which requires context)\n\tgen := NewGenerator(\"mcp:invalid\", nil)\n\n\tinput := &Input{\n\t\tQuery: \"find users\",\n\t\tModelIDs: []string{\"user\"},\n\t\tLimit: 10,\n\t}\n\n\t_, err := gen.Generate(nil, input)\n\tassert.Error(t, err)\n\tassert.Contains(t, err.Error(), \"context is required\")\n}\n\nfunc TestSystemQueryDSLAgentConstant(t *testing.T) {\n\t// Verify the system querydsl agent constant\n\tassert.Equal(t, \"__yao.querydsl\", SystemQueryDSLAgent)\n}\n\nfunc TestResult(t *testing.T) {\n\tresult := &Result{\n\t\tDSL: &gou.QueryDSL{\n\t\t\tLimit: 10,\n\t\t},\n\t\tExplain: \"Generated query for finding users\",\n\t\tWarnings: []string{\"using placeholder implementation\"},\n\t}\n\n\tassert.NotNil(t, result.DSL)\n\tassert.Equal(t, 10, result.DSL.Limit)\n\tassert.NotEmpty(t, result.Explain)\n\tassert.Len(t, result.Warnings, 1)\n}\n\nfunc TestGenerator_ValidateFields(t *testing.T) {\n\tgen := NewGenerator(\"\", nil)\n\n\tt.Run(\"validate select fields\", func(t *testing.T) {\n\t\tresult := &Result{\n\t\t\tDSL: &gou.QueryDSL{\n\t\t\t\tSelect: []gou.Expression{\n\t\t\t\t\t{Field: \"id\"},\n\t\t\t\t\t{Field: \"name\"},\n\t\t\t\t\t{Field: \"secret_field\"},\n\t\t\t\t},\n\t\t\t},\n\t\t}\n\t\tallowedFields := []string{\"id\", \"name\", \"email\"}\n\n\t\tvalidated := gen.validateFields(result, allowedFields)\n\t\tassert.NotNil(t, validated)\n\t\tassert.Len(t, validated.DSL.Select, 2)\n\t\tassert.Contains(t, validated.Warnings[0], \"secret_field\")\n\t})\n\n\tt.Run(\"validate where fields\", func(t *testing.T) {\n\t\tresult := &Result{\n\t\t\tDSL: &gou.QueryDSL{\n\t\t\t\tWheres: []gou.Where{\n\t\t\t\t\t{\n\t\t\t\t\t\tCondition: gou.Condition{\n\t\t\t\t\t\t\tField: &gou.Expression{Field: \"status\"},\n\t\t\t\t\t\t\tOP: \"=\",\n\t\t\t\t\t\t\tValue: \"active\",\n\t\t\t\t\t\t},\n\t\t\t\t\t},\n\t\t\t\t\t{\n\t\t\t\t\t\tCondition: gou.Condition{\n\t\t\t\t\t\t\tField: &gou.Expression{Field: \"secret\"},\n\t\t\t\t\t\t\tOP: \"=\",\n\t\t\t\t\t\t\tValue: \"hidden\",\n\t\t\t\t\t\t},\n\t\t\t\t\t},\n\t\t\t\t},\n\t\t\t},\n\t\t}\n\t\tallowedFields := []string{\"status\", \"name\"}\n\n\t\tvalidated := gen.validateFields(result, allowedFields)\n\t\tassert.NotNil(t, validated)\n\t\tassert.Len(t, validated.DSL.Wheres, 1)\n\t\tassert.Contains(t, validated.Warnings[0], \"secret\")\n\t})\n\n\tt.Run(\"validate order fields\", func(t *testing.T) {\n\t\tresult := &Result{\n\t\t\tDSL: &gou.QueryDSL{\n\t\t\t\tOrders: gou.Orders{\n\t\t\t\t\t{Field: &gou.Expression{Field: \"created_at\"}, Sort: \"desc\"},\n\t\t\t\t\t{Field: &gou.Expression{Field: \"secret_sort\"}, Sort: \"asc\"},\n\t\t\t\t},\n\t\t\t},\n\t\t}\n\t\tallowedFields := []string{\"created_at\", \"updated_at\"}\n\n\t\tvalidated := gen.validateFields(result, allowedFields)\n\t\tassert.NotNil(t, validated)\n\t\tassert.Len(t, validated.DSL.Orders, 1)\n\t\tassert.Contains(t, validated.Warnings[0], \"secret_sort\")\n\t})\n\n\tt.Run(\"nil DSL\", func(t *testing.T) {\n\t\tresult := &Result{DSL: nil}\n\t\tallowedFields := []string{\"id\", \"name\"}\n\n\t\tvalidated := gen.validateFields(result, allowedFields)\n\t\tassert.NotNil(t, validated)\n\t\tassert.Nil(t, validated.DSL)\n\t})\n}\n" }, { "path": "agent/search/nlp/querydsl/mcp.go", "content": "package querydsl\n\nimport (\n\t\"encoding/json\"\n\t\"fmt\"\n\t\"strings\"\n\n\t\"github.com/yaoapp/gou/mcp\"\n\tgouMCPTypes \"github.com/yaoapp/gou/mcp/types\"\n\t\"github.com/yaoapp/gou/query/gou\"\n\t\"github.com/yaoapp/gou/query/linter\"\n\tagentContext \"github.com/yaoapp/yao/agent/context\"\n)\n\n// MaxRetries is the maximum number of retry attempts for QueryDSL generation\nconst MaxRetries = 3\n\n// MCPProvider delegates QueryDSL generation to an MCP tool\ntype MCPProvider struct {\n\tserverID string // MCP server ID\n\ttoolName string // Tool name to call\n}\n\n// NewMCPProvider creates a new MCP-based QueryDSL generator\n// mcpRef format: \"server.tool\" (e.g., \"nlp.generate_querydsl\")\nfunc NewMCPProvider(mcpRef string) (*MCPProvider, error) {\n\tparts := strings.SplitN(mcpRef, \".\", 2)\n\tif len(parts) != 2 {\n\t\treturn nil, fmt.Errorf(\"invalid MCP format, expected 'server.tool', got '%s'\", mcpRef)\n\t}\n\treturn &MCPProvider{\n\t\tserverID: parts[0],\n\t\ttoolName: parts[1],\n\t}, nil\n}\n\n// Generate generates QueryDSL by calling the MCP tool with retry and lint validation\nfunc (p *MCPProvider) Generate(ctx *agentContext.Context, input *Input) (*Result, error) {\n\t// Get MCP client\n\tclient, err := mcp.Select(p.serverID)\n\tif err != nil {\n\t\treturn nil, fmt.Errorf(\"MCP server '%s' not found: %w\", p.serverID, err)\n\t}\n\n\tvar lastError error\n\tvar lastLintErrors string\n\n\tfor attempt := 1; attempt <= MaxRetries; attempt++ {\n\t\t// Build arguments for the MCP tool\n\t\targuments := p.buildArguments(input, attempt, lastLintErrors)\n\n\t\t// Call the MCP tool\n\t\tcallResult, err := client.CallTool(ctx, p.toolName, arguments)\n\t\tif err != nil {\n\t\t\tlastError = fmt.Errorf(\"MCP tool call failed: %w\", err)\n\t\t\tcontinue\n\t\t}\n\n\t\t// Parse the result\n\t\tresult, err := p.parseResult(callResult)\n\t\tif err != nil {\n\t\t\tlastError = err\n\t\t\tcontinue\n\t\t}\n\n\t\t// Validate with linter if DSL is present\n\t\tif result.DSL != nil {\n\t\t\tlintResult := p.validateDSL(result.DSL)\n\t\t\tif lintResult.Valid {\n\t\t\t\treturn result, nil\n\t\t\t}\n\n\t\t\t// Lint failed, prepare error message for retry\n\t\t\tlastLintErrors = lintResult.FormatDiagnostics()\n\t\t\tlastError = fmt.Errorf(\"QueryDSL validation failed: %s\", lastLintErrors)\n\n\t\t\t// Add lint warnings to result warnings\n\t\t\tfor _, diag := range lintResult.Diagnostics {\n\t\t\t\tresult.Warnings = append(result.Warnings, fmt.Sprintf(\"[%s] %s: %s\", diag.Code, diag.Path, diag.Message))\n\t\t\t}\n\t\t\tcontinue\n\t\t}\n\n\t\t// No DSL returned\n\t\tlastError = fmt.Errorf(\"no QueryDSL returned from MCP tool\")\n\t}\n\n\treturn nil, fmt.Errorf(\"QueryDSL generation failed after %d attempts: %w\", MaxRetries, lastError)\n}\n\n// buildArguments constructs the MCP tool arguments\nfunc (p *MCPProvider) buildArguments(input *Input, attempt int, lastLintErrors string) map[string]interface{} {\n\targuments := map[string]interface{}{\n\t\t\"query\": input.Query,\n\t\t\"models\": input.ModelIDs,\n\t\t\"limit\": input.Limit,\n\t}\n\n\t// Add optional fields\n\tif len(input.Wheres) > 0 {\n\t\targuments[\"wheres\"] = input.Wheres\n\t}\n\tif len(input.Orders) > 0 {\n\t\targuments[\"orders\"] = input.Orders\n\t}\n\tif len(input.AllowedFields) > 0 {\n\t\targuments[\"allowed_fields\"] = input.AllowedFields\n\t}\n\tif len(input.ExtraParams) > 0 {\n\t\targuments[\"extra\"] = input.ExtraParams\n\t}\n\n\t// Add retry context if this is a retry attempt\n\tif attempt > 1 && lastLintErrors != \"\" {\n\t\targuments[\"retry\"] = map[string]interface{}{\n\t\t\t\"attempt\": attempt,\n\t\t\t\"lint_errors\": lastLintErrors,\n\t\t\t\"instructions\": \"The previous QueryDSL was invalid. Please fix the errors and regenerate.\",\n\t\t}\n\t}\n\n\treturn arguments\n}\n\n// validateDSL validates the generated QueryDSL using the linter\nfunc (p *MCPProvider) validateDSL(dsl *gou.QueryDSL) *linter.LintResult {\n\t// Marshal DSL to JSON for linting\n\tjsonBytes, err := json.Marshal(dsl)\n\tif err != nil {\n\t\tresult := &linter.LintResult{Valid: false}\n\t\treturn result\n\t}\n\n\t_, lintResult := linter.Parse(string(jsonBytes))\n\treturn lintResult\n}\n\n// parseResult extracts QueryDSL from the MCP tool response\nfunc (p *MCPProvider) parseResult(result *gouMCPTypes.CallToolResponse) (*Result, error) {\n\tif result == nil {\n\t\treturn &Result{}, nil\n\t}\n\n\t// Check for errors in result\n\tif result.IsError {\n\t\terrMsg := \"MCP tool returned error\"\n\t\tif len(result.Content) > 0 && result.Content[0].Text != \"\" {\n\t\t\terrMsg = result.Content[0].Text\n\t\t}\n\t\treturn nil, fmt.Errorf(\"%s\", errMsg)\n\t}\n\n\t// Parse content - expect JSON data with \"dsl\" field\n\tif len(result.Content) == 0 {\n\t\treturn &Result{}, nil\n\t}\n\n\tgenResult := &Result{}\n\n\t// Try to extract QueryDSL from content\n\tfor _, content := range result.Content {\n\t\t// Check text content type\n\t\tif content.Type == gouMCPTypes.ToolContentTypeText && content.Text != \"\" {\n\t\t\t// Try to parse as JSON\n\t\t\tvar data map[string]interface{}\n\t\t\tif err := json.Unmarshal([]byte(content.Text), &data); err == nil {\n\t\t\t\t// Look for \"dsl\" field\n\t\t\t\tif dsl, ok := data[\"dsl\"]; ok {\n\t\t\t\t\tgenResult.DSL = p.extractDSL(dsl)\n\t\t\t\t}\n\t\t\t\tif explain, ok := data[\"explain\"].(string); ok {\n\t\t\t\t\tgenResult.Explain = explain\n\t\t\t\t}\n\t\t\t\tif warnings, ok := data[\"warnings\"]; ok {\n\t\t\t\t\tgenResult.Warnings = p.extractWarnings(warnings)\n\t\t\t\t}\n\t\t\t\treturn genResult, nil\n\t\t\t}\n\n\t\t\t// Try to parse as direct QueryDSL\n\t\t\tvar dsl gou.QueryDSL\n\t\t\tif err := json.Unmarshal([]byte(content.Text), &dsl); err == nil {\n\t\t\t\tgenResult.DSL = &dsl\n\t\t\t\treturn genResult, nil\n\t\t\t}\n\t\t}\n\t}\n\n\treturn genResult, nil\n}\n\n// extractDSL converts interface{} to gou.QueryDSL\nfunc (p *MCPProvider) extractDSL(v interface{}) *gou.QueryDSL {\n\tif v == nil {\n\t\treturn nil\n\t}\n\n\t// Marshal and unmarshal to gou.QueryDSL\n\tjsonBytes, err := json.Marshal(v)\n\tif err != nil {\n\t\treturn nil\n\t}\n\n\tvar dsl gou.QueryDSL\n\tif err := json.Unmarshal(jsonBytes, &dsl); err != nil {\n\t\treturn nil\n\t}\n\n\treturn &dsl\n}\n\n// extractWarnings extracts warnings array from various types\nfunc (p *MCPProvider) extractWarnings(v interface{}) []string {\n\tswitch w := v.(type) {\n\tcase []string:\n\t\treturn w\n\tcase []interface{}:\n\t\twarnings := make([]string, 0, len(w))\n\t\tfor _, item := range w {\n\t\t\tif s, ok := item.(string); ok {\n\t\t\t\twarnings = append(warnings, s)\n\t\t\t}\n\t\t}\n\t\treturn warnings\n\tcase string:\n\t\treturn []string{w}\n\t}\n\treturn nil\n}\n" }, { "path": "agent/search/nlp/querydsl/mcp_test.go", "content": "package querydsl\n\nimport (\n\tstdContext \"context\"\n\t\"os\"\n\t\"testing\"\n\n\t\"github.com/stretchr/testify/assert\"\n\tagentContext \"github.com/yaoapp/yao/agent/context\"\n\t\"github.com/yaoapp/yao/config\"\n\t\"github.com/yaoapp/yao/test\"\n)\n\n// newTestContext creates a test context for MCP testing\nfunc newTestContext() *agentContext.Context {\n\tctx := agentContext.New(stdContext.Background(), nil, \"test-chat\")\n\tctx.AssistantID = \"test-assistant\"\n\tctx.Locale = \"en\"\n\tctx.Referer = agentContext.RefererAPI\n\tstack, _, _ := agentContext.EnterStack(ctx, \"test-assistant\", &agentContext.Options{})\n\tctx.Stack = stack\n\treturn ctx\n}\n\nfunc TestMCPProvider_Generate(t *testing.T) {\n\ttest.Prepare(t, config.Conf)\n\tdefer test.Clean()\n\n\t// Create context\n\tctx := newTestContext()\n\n\t// Create MCP provider for search.generate_querydsl\n\tprovider, err := NewMCPProvider(\"search.generate_querydsl\")\n\tassert.NoError(t, err)\n\tassert.NotNil(t, provider)\n\tassert.Equal(t, \"search\", provider.serverID)\n\tassert.Equal(t, \"generate_querydsl\", provider.toolName)\n\n\tt.Run(\"verify_fixed_structure\", func(t *testing.T) {\n\t\tinput := &Input{\n\t\t\tQuery: \"find active users\",\n\t\t\tModelIDs: []string{\"user\"},\n\t\t\tLimit: 10,\n\t\t}\n\n\t\tresult, err := provider.Generate(ctx, input)\n\t\tif err != nil {\n\t\t\tt.Logf(\"Generate error: %v\", err)\n\t\t}\n\t\tassert.NoError(t, err)\n\t\tassert.NotNil(t, result)\n\n\t\tif result == nil {\n\t\t\tt.Fatal(\"result is nil\")\n\t\t}\n\n\t\tif !assert.NotNil(t, result.DSL, \"DSL should not be nil\") {\n\t\t\tt.Logf(\"Result: Explain=%s, Warnings=%v\", result.Explain, result.Warnings)\n\t\t\treturn\n\t\t}\n\n\t\t// Verify fixed DSL structure from mock\n\t\t// select: [\"id\", \"name\", \"status\"] - parsed as Expression with Field property\n\t\tassert.Len(t, result.DSL.Select, 3)\n\t\tif len(result.DSL.Select) >= 3 {\n\t\t\tassert.Equal(t, \"id\", result.DSL.Select[0].Field)\n\t\t\tassert.Equal(t, \"name\", result.DSL.Select[1].Field)\n\t\t\tassert.Equal(t, \"status\", result.DSL.Select[2].Field)\n\t\t}\n\n\t\t// wheres: [{ field: \"status\", op: \"=\", value: \"active\" }]\n\t\tassert.Len(t, result.DSL.Wheres, 1)\n\t\tif len(result.DSL.Wheres) > 0 {\n\t\t\tassert.Equal(t, \"status\", result.DSL.Wheres[0].Field.Field)\n\t\t\tassert.Equal(t, \"=\", result.DSL.Wheres[0].OP)\n\t\t\tassert.Equal(t, \"active\", result.DSL.Wheres[0].Value)\n\t\t}\n\n\t\t// orders: [{ field: \"created_at\", sort: \"desc\" }]\n\t\tassert.Len(t, result.DSL.Orders, 1)\n\t\tif len(result.DSL.Orders) > 0 {\n\t\t\tassert.Equal(t, \"created_at\", result.DSL.Orders[0].Field.Field)\n\t\t\tassert.Equal(t, \"desc\", result.DSL.Orders[0].Sort)\n\t\t}\n\n\t\t// limit: 10 (from input, returned as float64 from JSON)\n\t\tassert.Equal(t, float64(10), result.DSL.Limit)\n\n\t\t// explain should contain query\n\t\tassert.Contains(t, result.Explain, \"find active users\")\n\n\t\t// warnings should be empty\n\t\tassert.Empty(t, result.Warnings)\n\t})\n}\n\nfunc TestNewMCPProvider(t *testing.T) {\n\tt.Run(\"valid format\", func(t *testing.T) {\n\t\tprovider, err := NewMCPProvider(\"nlp.generate_querydsl\")\n\t\tassert.NoError(t, err)\n\t\tassert.NotNil(t, provider)\n\t\tassert.Equal(t, \"nlp\", provider.serverID)\n\t\tassert.Equal(t, \"generate_querydsl\", provider.toolName)\n\t})\n\n\tt.Run(\"invalid format - no dot\", func(t *testing.T) {\n\t\tprovider, err := NewMCPProvider(\"invalid\")\n\t\tassert.Error(t, err)\n\t\tassert.Nil(t, provider)\n\t\tassert.Contains(t, err.Error(), \"invalid MCP format\")\n\t})\n\n\tt.Run(\"complex tool name\", func(t *testing.T) {\n\t\tprovider, err := NewMCPProvider(\"server.tool.with.dots\")\n\t\tassert.NoError(t, err)\n\t\tassert.NotNil(t, provider)\n\t\tassert.Equal(t, \"server\", provider.serverID)\n\t\tassert.Equal(t, \"tool.with.dots\", provider.toolName)\n\t})\n}\n\nfunc TestMCPProvider_Generate_Error(t *testing.T) {\n\ttest.Prepare(t, config.Conf)\n\tdefer test.Clean()\n\n\tctx := newTestContext()\n\n\tt.Run(\"non-existent server\", func(t *testing.T) {\n\t\tprovider, _ := NewMCPProvider(\"nonexistent.tool\")\n\t\tresult, err := provider.Generate(ctx, &Input{\n\t\t\tQuery: \"test\",\n\t\t\tModelIDs: []string{\"user\"},\n\t\t})\n\t\tassert.Error(t, err)\n\t\tassert.Nil(t, result)\n\t\tassert.Contains(t, err.Error(), \"not found\")\n\t})\n}\n\nfunc TestGenerator_MCP_Integration(t *testing.T) {\n\ttest.Prepare(t, config.Conf)\n\tdefer test.Clean()\n\n\t// Skip if not in integration test mode\n\tif os.Getenv(\"YAO_TEST_MCP\") != \"true\" {\n\t\tt.Skip(\"Skipping MCP integration test (set YAO_TEST_MCP=true to run)\")\n\t}\n\n\tctx := newTestContext()\n\n\t// Create generator with MCP mode\n\tgen := NewGenerator(\"mcp:search.generate_querydsl\", nil)\n\n\tt.Run(\"generate_via_mcp\", func(t *testing.T) {\n\t\tinput := &Input{\n\t\t\tQuery: \"find active users\",\n\t\t\tModelIDs: []string{\"user\"},\n\t\t\tLimit: 15,\n\t\t}\n\n\t\tresult, err := gen.Generate(ctx, input)\n\t\tassert.NoError(t, err)\n\t\tassert.NotNil(t, result)\n\t\tassert.NotNil(t, result.DSL)\n\n\t\t// Verify fixed structure is correctly parsed\n\t\tassert.Len(t, result.DSL.Select, 3)\n\t\tassert.Len(t, result.DSL.Wheres, 1)\n\t\tassert.Len(t, result.DSL.Orders, 1)\n\t\tassert.Equal(t, float64(15), result.DSL.Limit)\n\t\tassert.Contains(t, result.Explain, \"find active users\")\n\t})\n\n\tt.Run(\"allowed_fields_validation\", func(t *testing.T) {\n\t\tinput := &Input{\n\t\t\tQuery: \"find users\",\n\t\t\tModelIDs: []string{\"user\"},\n\t\t\tAllowedFields: []string{\"id\", \"name\"}, // Only allow id and name\n\t\t\tLimit: 10,\n\t\t}\n\n\t\tresult, err := gen.Generate(ctx, input)\n\t\tassert.NoError(t, err)\n\t\tassert.NotNil(t, result)\n\t\tassert.NotNil(t, result.DSL)\n\n\t\t// \"status\" field should be filtered out from select and wheres\n\t\t// since it's not in AllowedFields\n\t\tfor _, expr := range result.DSL.Select {\n\t\t\tassert.Contains(t, []string{\"id\", \"name\"}, expr.Field)\n\t\t}\n\n\t\t// Should have warning about removed fields\n\t\tassert.NotEmpty(t, result.Warnings)\n\t})\n}\n\nfunc TestMCPProvider_Generate_WithRetry(t *testing.T) {\n\ttest.Prepare(t, config.Conf)\n\tdefer test.Clean()\n\n\tctx := newTestContext()\n\n\t// Create MCP provider for search.generate_querydsl_with_retry\n\t// This tool returns invalid DSL on first call, valid on second\n\tprovider, err := NewMCPProvider(\"search.generate_querydsl_with_retry\")\n\tassert.NoError(t, err)\n\tassert.NotNil(t, provider)\n\n\tt.Run(\"retry_on_lint_failure\", func(t *testing.T) {\n\t\tinput := &Input{\n\t\t\tQuery: \"test retry mechanism\",\n\t\t\tModelIDs: []string{\"user\"},\n\t\t\tLimit: 10,\n\t\t}\n\n\t\t// This should succeed after retry\n\t\t// First call returns invalid DSL (missing 'from')\n\t\t// Second call (with lint_errors) returns valid DSL\n\t\tresult, err := provider.Generate(ctx, input)\n\t\tassert.NoError(t, err)\n\t\tassert.NotNil(t, result)\n\n\t\tif result != nil && result.DSL != nil {\n\t\t\t// Should have valid DSL after retry\n\t\t\tassert.NotNil(t, result.DSL.From, \"DSL should have 'from' field after retry\")\n\t\t\t// Explain should indicate this was fixed after receiving lint errors\n\t\t\tassert.Contains(t, result.Explain, \"fixed after receiving lint errors\")\n\t\t}\n\t})\n}\n" }, { "path": "agent/search/nlp/querydsl/types.go", "content": "package querydsl\n\nimport (\n\t\"github.com/yaoapp/gou/query/gou\"\n\t\"github.com/yaoapp/yao/agent/search/types\"\n)\n\n// Input contains all information needed to generate QueryDSL\ntype Input struct {\n\tQuery string // Natural language query\n\tModelIDs []string // Target model IDs (e.g., [\"user\", \"order\", \"product\"])\n\tScenario types.ScenarioType // QueryDSL scenario: \"filter\", \"aggregation\", \"join\", \"complex\"\n\tWheres []gou.Where // Pre-defined filters (optional)\n\tOrders gou.Orders // Sort orders (optional)\n\tAllowedFields []string // Allowed fields whitelist (optional, for security validation)\n\tLimit int // Max results\n\tExtraParams map[string]interface{} // Additional parameters\n}\n\n// Result represents the result of QueryDSL generation\ntype Result struct {\n\tDSL *gou.QueryDSL `json:\"dsl\"` // Generated QueryDSL (supports joins)\n\tExplain string `json:\"explain,omitempty\"` // Human-readable explanation\n\tWarnings []string `json:\"warnings,omitempty\"` // Any warnings during generation\n}\n" }, { "path": "agent/search/reference.go", "content": "package search\n\nimport (\n\t\"fmt\"\n\t\"strings\"\n\n\t\"github.com/yaoapp/yao/agent/search/types\"\n)\n\n// DefaultCitationPrompt is the default prompt for citation instructions\nconst DefaultCitationPrompt = `You have access to reference data in tags. Each has:\n- id: Citation identifier (integer)\n- type: Data type (web/kb/db)\n- weight: Relevance weight (1.0=highest priority, 0.6=lowest)\n- source: Origin (user=user-provided, hook=assistant-searched, auto=auto-searched)\n\nPrioritize higher-weight references when answering.\n\nWhen citing a reference, use this exact HTML format:\n[{id}]\n\nExample: According to the product data[1], the price is $999.`\n\n// BuildReferences converts search results to unified Reference format\nfunc BuildReferences(results []*types.Result) []*types.Reference {\n\tvar refs []*types.Reference\n\tfor _, result := range results {\n\t\tif result == nil {\n\t\t\tcontinue\n\t\t}\n\t\tfor _, item := range result.Items {\n\t\t\tif item == nil {\n\t\t\t\tcontinue\n\t\t\t}\n\t\t\trefs = append(refs, &types.Reference{\n\t\t\t\tID: item.CitationID,\n\t\t\t\tType: item.Type,\n\t\t\t\tSource: item.Source,\n\t\t\t\tWeight: item.Weight,\n\t\t\t\tScore: item.Score,\n\t\t\t\tTitle: item.Title,\n\t\t\t\tContent: item.Content,\n\t\t\t\tURL: item.URL,\n\t\t\t})\n\t\t}\n\t}\n\treturn refs\n}\n\n// FormatReferencesXML formats references as XML for LLM context\nfunc FormatReferencesXML(refs []*types.Reference) string {\n\tif len(refs) == 0 {\n\t\treturn \"\"\n\t}\n\n\tvar sb strings.Builder\n\tsb.WriteString(\"\\n\")\n\n\tfor _, ref := range refs {\n\t\tif ref == nil {\n\t\t\tcontinue\n\t\t}\n\t\tsb.WriteString(fmt.Sprintf(``,\n\t\t\tref.ID, ref.Type, ref.Weight, ref.Source))\n\t\tsb.WriteString(\"\\n\")\n\n\t\tif ref.Title != \"\" {\n\t\t\tsb.WriteString(ref.Title)\n\t\t\tsb.WriteString(\"\\n\")\n\t\t}\n\t\tsb.WriteString(ref.Content)\n\t\tif ref.URL != \"\" {\n\t\t\tsb.WriteString(\"\\nURL: \")\n\t\t\tsb.WriteString(ref.URL)\n\t\t}\n\t\tsb.WriteString(\"\\n\\n\")\n\t}\n\n\tsb.WriteString(\"\")\n\treturn sb.String()\n}\n\n// GetCitationPrompt returns the citation instruction prompt\nfunc GetCitationPrompt(cfg *types.CitationConfig) string {\n\tif cfg == nil {\n\t\treturn DefaultCitationPrompt\n\t}\n\tif cfg.CustomPrompt != \"\" {\n\t\treturn cfg.CustomPrompt\n\t}\n\treturn DefaultCitationPrompt\n}\n\n// BuildReferenceContext builds the complete reference context for LLM\nfunc BuildReferenceContext(results []*types.Result, cfg *types.CitationConfig) *types.ReferenceContext {\n\trefs := BuildReferences(results)\n\treturn &types.ReferenceContext{\n\t\tReferences: refs,\n\t\tXML: FormatReferencesXML(refs),\n\t\tPrompt: GetCitationPrompt(cfg),\n\t}\n}\n" }, { "path": "agent/search/reference_test.go", "content": "package search\n\nimport (\n\t\"strings\"\n\t\"testing\"\n\n\t\"github.com/stretchr/testify/assert\"\n\t\"github.com/yaoapp/yao/agent/search/types\"\n)\n\nfunc TestBuildReferences(t *testing.T) {\n\ttests := []struct {\n\t\tname string\n\t\tresults []*types.Result\n\t\texpected int\n\t}{\n\t\t{\n\t\t\tname: \"nil results\",\n\t\t\tresults: nil,\n\t\t\texpected: 0,\n\t\t},\n\t\t{\n\t\t\tname: \"empty results\",\n\t\t\tresults: []*types.Result{},\n\t\t\texpected: 0,\n\t\t},\n\t\t{\n\t\t\tname: \"single result with items\",\n\t\t\tresults: []*types.Result{\n\t\t\t\t{\n\t\t\t\t\tType: types.SearchTypeWeb,\n\t\t\t\t\tQuery: \"test query\",\n\t\t\t\t\tItems: []*types.ResultItem{\n\t\t\t\t\t\t{\n\t\t\t\t\t\t\tCitationID: \"1\",\n\t\t\t\t\t\t\tType: types.SearchTypeWeb,\n\t\t\t\t\t\t\tSource: types.SourceAuto,\n\t\t\t\t\t\t\tWeight: 0.6,\n\t\t\t\t\t\t\tScore: 0.9,\n\t\t\t\t\t\t\tTitle: \"Test Title\",\n\t\t\t\t\t\t\tContent: \"Test content\",\n\t\t\t\t\t\t\tURL: \"https://example.com\",\n\t\t\t\t\t\t},\n\t\t\t\t\t\t{\n\t\t\t\t\t\t\tCitationID: \"2\",\n\t\t\t\t\t\t\tType: types.SearchTypeWeb,\n\t\t\t\t\t\t\tSource: types.SourceAuto,\n\t\t\t\t\t\t\tWeight: 0.6,\n\t\t\t\t\t\t\tScore: 0.8,\n\t\t\t\t\t\t\tTitle: \"Test Title 2\",\n\t\t\t\t\t\t\tContent: \"Test content 2\",\n\t\t\t\t\t\t\tURL: \"https://example2.com\",\n\t\t\t\t\t\t},\n\t\t\t\t\t},\n\t\t\t\t},\n\t\t\t},\n\t\t\texpected: 2,\n\t\t},\n\t\t{\n\t\t\tname: \"multiple results\",\n\t\t\tresults: []*types.Result{\n\t\t\t\t{\n\t\t\t\t\tType: types.SearchTypeWeb,\n\t\t\t\t\tItems: []*types.ResultItem{\n\t\t\t\t\t\t{CitationID: \"1\", Type: types.SearchTypeWeb, Content: \"Web content\"},\n\t\t\t\t\t},\n\t\t\t\t},\n\t\t\t\t{\n\t\t\t\t\tType: types.SearchTypeKB,\n\t\t\t\t\tItems: []*types.ResultItem{\n\t\t\t\t\t\t{CitationID: \"2\", Type: types.SearchTypeKB, Content: \"KB content\"},\n\t\t\t\t\t},\n\t\t\t\t},\n\t\t\t\t{\n\t\t\t\t\tType: types.SearchTypeDB,\n\t\t\t\t\tItems: []*types.ResultItem{\n\t\t\t\t\t\t{CitationID: \"3\", Type: types.SearchTypeDB, Content: \"DB content\"},\n\t\t\t\t\t},\n\t\t\t\t},\n\t\t\t},\n\t\t\texpected: 3,\n\t\t},\n\t\t{\n\t\t\tname: \"result with nil items\",\n\t\t\tresults: []*types.Result{\n\t\t\t\t{\n\t\t\t\t\tType: types.SearchTypeWeb,\n\t\t\t\t\tItems: []*types.ResultItem{\n\t\t\t\t\t\t{CitationID: \"1\", Content: \"Content 1\"},\n\t\t\t\t\t\tnil,\n\t\t\t\t\t\t{CitationID: \"2\", Content: \"Content 2\"},\n\t\t\t\t\t},\n\t\t\t\t},\n\t\t\t},\n\t\t\texpected: 2,\n\t\t},\n\t\t{\n\t\t\tname: \"nil result in slice\",\n\t\t\tresults: []*types.Result{\n\t\t\t\t{\n\t\t\t\t\tType: types.SearchTypeWeb,\n\t\t\t\t\tItems: []*types.ResultItem{\n\t\t\t\t\t\t{CitationID: \"1\", Content: \"Content\"},\n\t\t\t\t\t},\n\t\t\t\t},\n\t\t\t\tnil,\n\t\t\t\t{\n\t\t\t\t\tType: types.SearchTypeKB,\n\t\t\t\t\tItems: []*types.ResultItem{\n\t\t\t\t\t\t{CitationID: \"2\", Content: \"Content 2\"},\n\t\t\t\t\t},\n\t\t\t\t},\n\t\t\t},\n\t\t\texpected: 2,\n\t\t},\n\t}\n\n\tfor _, tt := range tests {\n\t\tt.Run(tt.name, func(t *testing.T) {\n\t\t\trefs := BuildReferences(tt.results)\n\t\t\tassert.Equal(t, tt.expected, len(refs))\n\t\t})\n\t}\n}\n\nfunc TestBuildReferences_FieldMapping(t *testing.T) {\n\titem := &types.ResultItem{\n\t\tCitationID: \"1\",\n\t\tType: types.SearchTypeWeb,\n\t\tSource: types.SourceHook,\n\t\tWeight: 0.8,\n\t\tScore: 0.95,\n\t\tTitle: \"Test Title\",\n\t\tContent: \"Test Content\",\n\t\tURL: \"https://example.com\",\n\t}\n\n\tresults := []*types.Result{\n\t\t{Items: []*types.ResultItem{item}},\n\t}\n\n\trefs := BuildReferences(results)\n\tassert.Equal(t, 1, len(refs))\n\n\tref := refs[0]\n\tassert.Equal(t, \"1\", ref.ID)\n\tassert.Equal(t, types.SearchTypeWeb, ref.Type)\n\tassert.Equal(t, types.SourceHook, ref.Source)\n\tassert.Equal(t, 0.8, ref.Weight)\n\tassert.Equal(t, 0.95, ref.Score)\n\tassert.Equal(t, \"Test Title\", ref.Title)\n\tassert.Equal(t, \"Test Content\", ref.Content)\n\tassert.Equal(t, \"https://example.com\", ref.URL)\n}\n\nfunc TestFormatReferencesXML(t *testing.T) {\n\ttests := []struct {\n\t\tname string\n\t\trefs []*types.Reference\n\t\tcontains []string\n\t\texcludes []string\n\t}{\n\t\t{\n\t\t\tname: \"nil refs\",\n\t\t\trefs: nil,\n\t\t\tcontains: []string{},\n\t\t\texcludes: []string{\"\"},\n\t\t},\n\t\t{\n\t\t\tname: \"empty refs\",\n\t\t\trefs: []*types.Reference{},\n\t\t\tcontains: []string{},\n\t\t\texcludes: []string{\"\"},\n\t\t},\n\t\t{\n\t\t\tname: \"single ref with all fields\",\n\t\t\trefs: []*types.Reference{\n\t\t\t\t{\n\t\t\t\t\tID: \"1\",\n\t\t\t\t\tType: types.SearchTypeWeb,\n\t\t\t\t\tSource: types.SourceUser,\n\t\t\t\t\tWeight: 1.0,\n\t\t\t\t\tScore: 0.9,\n\t\t\t\t\tTitle: \"Test Title\",\n\t\t\t\t\tContent: \"Test Content\",\n\t\t\t\t\tURL: \"https://example.com\",\n\t\t\t\t},\n\t\t\t},\n\t\t\tcontains: []string{\n\t\t\t\t\"\",\n\t\t\t\t\"\",\n\t\t\t\t``,\n\t\t\t\t\"\",\n\t\t\t\t\"Test Title\",\n\t\t\t\t\"Test Content\",\n\t\t\t\t\"URL: https://example.com\",\n\t\t\t},\n\t\t},\n\t\t{\n\t\t\tname: \"ref without title\",\n\t\t\trefs: []*types.Reference{\n\t\t\t\t{\n\t\t\t\t\tID: \"1\",\n\t\t\t\t\tType: types.SearchTypeKB,\n\t\t\t\t\tSource: types.SourceHook,\n\t\t\t\t\tWeight: 0.8,\n\t\t\t\t\tContent: \"Content without title\",\n\t\t\t\t},\n\t\t\t},\n\t\t\tcontains: []string{\n\t\t\t\t``,\n\t\t\t\t\"Content without title\",\n\t\t\t},\n\t\t\texcludes: []string{\n\t\t\t\t\"URL:\",\n\t\t\t},\n\t\t},\n\t\t{\n\t\t\tname: \"ref without URL\",\n\t\t\trefs: []*types.Reference{\n\t\t\t\t{\n\t\t\t\t\tID: \"1\",\n\t\t\t\t\tType: types.SearchTypeDB,\n\t\t\t\t\tSource: types.SourceAuto,\n\t\t\t\t\tWeight: 0.6,\n\t\t\t\t\tTitle: \"DB Record\",\n\t\t\t\t\tContent: \"Database content\",\n\t\t\t\t},\n\t\t\t},\n\t\t\tcontains: []string{\n\t\t\t\t``,\n\t\t\t\t\"DB Record\",\n\t\t\t\t\"Database content\",\n\t\t\t},\n\t\t\texcludes: []string{\n\t\t\t\t\"URL:\",\n\t\t\t},\n\t\t},\n\t\t{\n\t\t\tname: \"multiple refs\",\n\t\t\trefs: []*types.Reference{\n\t\t\t\t{ID: \"1\", Type: types.SearchTypeWeb, Source: types.SourceUser, Weight: 1.0, Content: \"Content 1\"},\n\t\t\t\t{ID: \"2\", Type: types.SearchTypeKB, Source: types.SourceHook, Weight: 0.8, Content: \"Content 2\"},\n\t\t\t\t{ID: \"3\", Type: types.SearchTypeDB, Source: types.SourceAuto, Weight: 0.6, Content: \"Content 3\"},\n\t\t\t},\n\t\t\tcontains: []string{\n\t\t\t\t\"\",\n\t\t\t\t\"\",\n\t\t\t\t`id=\"1\"`,\n\t\t\t\t`id=\"2\"`,\n\t\t\t\t`id=\"3\"`,\n\t\t\t\t\"Content 1\",\n\t\t\t\t\"Content 2\",\n\t\t\t\t\"Content 3\",\n\t\t\t},\n\t\t},\n\t\t{\n\t\t\tname: \"nil ref in slice\",\n\t\t\trefs: []*types.Reference{\n\t\t\t\t{ID: \"1\", Type: types.SearchTypeWeb, Weight: 1.0, Content: \"Content 1\"},\n\t\t\t\tnil,\n\t\t\t\t{ID: \"2\", Type: types.SearchTypeKB, Weight: 0.8, Content: \"Content 2\"},\n\t\t\t},\n\t\t\tcontains: []string{\n\t\t\t\t`id=\"1\"`,\n\t\t\t\t`id=\"2\"`,\n\t\t\t},\n\t\t},\n\t}\n\n\tfor _, tt := range tests {\n\t\tt.Run(tt.name, func(t *testing.T) {\n\t\t\txml := FormatReferencesXML(tt.refs)\n\n\t\t\tfor _, s := range tt.contains {\n\t\t\t\tassert.Contains(t, xml, s, \"expected XML to contain: %s\", s)\n\t\t\t}\n\n\t\t\tfor _, s := range tt.excludes {\n\t\t\t\tassert.NotContains(t, xml, s, \"expected XML to not contain: %s\", s)\n\t\t\t}\n\t\t})\n\t}\n}\n\nfunc TestFormatReferencesXML_Structure(t *testing.T) {\n\trefs := []*types.Reference{\n\t\t{\n\t\t\tID: \"1\",\n\t\t\tType: types.SearchTypeWeb,\n\t\t\tSource: types.SourceUser,\n\t\t\tWeight: 1.0,\n\t\t\tTitle: \"Title\",\n\t\t\tContent: \"Content\",\n\t\t\tURL: \"https://example.com\",\n\t\t},\n\t}\n\n\txml := FormatReferencesXML(refs)\n\n\t// Check structure\n\tassert.True(t, strings.HasPrefix(xml, \"\\n\"))\n\tassert.True(t, strings.HasSuffix(xml, \"\"))\n\tassert.Contains(t, xml, \"\\n\")\n}\n\nfunc TestGetCitationPrompt(t *testing.T) {\n\ttests := []struct {\n\t\tname string\n\t\tcfg *types.CitationConfig\n\t\texpected string\n\t}{\n\t\t{\n\t\t\tname: \"nil config\",\n\t\t\tcfg: nil,\n\t\t\texpected: DefaultCitationPrompt,\n\t\t},\n\t\t{\n\t\t\tname: \"empty config\",\n\t\t\tcfg: &types.CitationConfig{},\n\t\t\texpected: DefaultCitationPrompt,\n\t\t},\n\t\t{\n\t\t\tname: \"config with custom prompt\",\n\t\t\tcfg: &types.CitationConfig{\n\t\t\t\tCustomPrompt: \"Custom citation instructions\",\n\t\t\t},\n\t\t\texpected: \"Custom citation instructions\",\n\t\t},\n\t\t{\n\t\t\tname: \"config with empty custom prompt\",\n\t\t\tcfg: &types.CitationConfig{\n\t\t\t\tCustomPrompt: \"\",\n\t\t\t},\n\t\t\texpected: DefaultCitationPrompt,\n\t\t},\n\t\t{\n\t\t\tname: \"config with format but no custom prompt\",\n\t\t\tcfg: &types.CitationConfig{\n\t\t\t\tFormat: \"[{id}]\",\n\t\t\t},\n\t\t\texpected: DefaultCitationPrompt,\n\t\t},\n\t}\n\n\tfor _, tt := range tests {\n\t\tt.Run(tt.name, func(t *testing.T) {\n\t\t\tprompt := GetCitationPrompt(tt.cfg)\n\t\t\tassert.Equal(t, tt.expected, prompt)\n\t\t})\n\t}\n}\n\nfunc TestDefaultCitationPrompt(t *testing.T) {\n\t// Verify default prompt contains key instructions\n\tassert.Contains(t, DefaultCitationPrompt, \"\")\n\tassert.Contains(t, DefaultCitationPrompt, \"id: Citation identifier\")\n\tassert.Contains(t, DefaultCitationPrompt, \"type: Data type\")\n\tassert.Contains(t, DefaultCitationPrompt, \"weight: Relevance weight\")\n\tassert.Contains(t, DefaultCitationPrompt, \"source: Origin\")\n\tassert.Contains(t, DefaultCitationPrompt, `\")\n\t\tassert.Contains(t, ctx.XML, `id=\"1\"`)\n\t\tassert.Equal(t, DefaultCitationPrompt, ctx.Prompt)\n\t})\n\n\tt.Run(\"with custom prompt config\", func(t *testing.T) {\n\t\tcfg := &types.CitationConfig{\n\t\t\tCustomPrompt: \"Custom prompt\",\n\t\t}\n\t\tctx := BuildReferenceContext(results, cfg)\n\n\t\tassert.NotNil(t, ctx)\n\t\tassert.Equal(t, \"Custom prompt\", ctx.Prompt)\n\t})\n\n\tt.Run(\"with empty results\", func(t *testing.T) {\n\t\tctx := BuildReferenceContext([]*types.Result{}, nil)\n\n\t\tassert.NotNil(t, ctx)\n\t\tassert.Equal(t, 0, len(ctx.References))\n\t\tassert.Equal(t, \"\", ctx.XML)\n\t\tassert.Equal(t, DefaultCitationPrompt, ctx.Prompt)\n\t})\n}\n\nfunc TestBuildReferenceContext_Integration(t *testing.T) {\n\t// Simulate a real-world scenario with multiple search types\n\tresults := []*types.Result{\n\t\t{\n\t\t\tType: types.SearchTypeWeb,\n\t\t\tQuery: \"AI developments\",\n\t\t\tItems: []*types.ResultItem{\n\t\t\t\t{\n\t\t\t\t\tCitationID: \"1\",\n\t\t\t\t\tType: types.SearchTypeWeb,\n\t\t\t\t\tSource: types.SourceAuto,\n\t\t\t\t\tWeight: 0.6,\n\t\t\t\t\tScore: 0.95,\n\t\t\t\t\tTitle: \"OpenAI Announces GPT-5\",\n\t\t\t\t\tContent: \"OpenAI has announced the development of GPT-5...\",\n\t\t\t\t\tURL: \"https://news.example.com/gpt5\",\n\t\t\t\t},\n\t\t\t},\n\t\t},\n\t\t{\n\t\t\tType: types.SearchTypeKB,\n\t\t\tQuery: \"AI developments\",\n\t\t\tItems: []*types.ResultItem{\n\t\t\t\t{\n\t\t\t\t\tCitationID: \"2\",\n\t\t\t\t\tType: types.SearchTypeKB,\n\t\t\t\t\tSource: types.SourceHook,\n\t\t\t\t\tWeight: 0.8,\n\t\t\t\t\tScore: 0.88,\n\t\t\t\t\tTitle: \"Internal AI Research Notes\",\n\t\t\t\t\tContent: \"Our internal research on AI capabilities...\",\n\t\t\t\t},\n\t\t\t},\n\t\t},\n\t\t{\n\t\t\tType: types.SearchTypeDB,\n\t\t\tQuery: \"AI developments\",\n\t\t\tItems: []*types.ResultItem{\n\t\t\t\t{\n\t\t\t\t\tCitationID: \"3\",\n\t\t\t\t\tType: types.SearchTypeDB,\n\t\t\t\t\tSource: types.SourceUser,\n\t\t\t\t\tWeight: 1.0,\n\t\t\t\t\tScore: 0.92,\n\t\t\t\t\tTitle: \"Product: AI Assistant\",\n\t\t\t\t\tContent: \"Name: AI Assistant\\nPrice: $99\\nCategory: Software\",\n\t\t\t\t},\n\t\t\t},\n\t\t},\n\t}\n\n\tctx := BuildReferenceContext(results, nil)\n\n\t// Verify all references are included\n\tassert.Equal(t, 3, len(ctx.References))\n\n\t// Verify XML contains all references\n\tassert.Contains(t, ctx.XML, `id=\"1\"`)\n\tassert.Contains(t, ctx.XML, `id=\"2\"`)\n\tassert.Contains(t, ctx.XML, `id=\"3\"`)\n\n\t// Verify different source types are represented\n\tassert.Contains(t, ctx.XML, `source=\"auto\"`)\n\tassert.Contains(t, ctx.XML, `source=\"hook\"`)\n\tassert.Contains(t, ctx.XML, `source=\"user\"`)\n\n\t// Verify different search types are represented\n\tassert.Contains(t, ctx.XML, `type=\"web\"`)\n\tassert.Contains(t, ctx.XML, `type=\"kb\"`)\n\tassert.Contains(t, ctx.XML, `type=\"db\"`)\n}\n" }, { "path": "agent/search/registry.go", "content": "package search\n\nimport (\n\t\"github.com/yaoapp/yao/agent/search/interfaces\"\n\t\"github.com/yaoapp/yao/agent/search/types\"\n)\n\n// Registry manages search handlers\ntype Registry struct {\n\thandlers map[types.SearchType]interfaces.Handler\n}\n\n// NewRegistry creates a new handler registry\nfunc NewRegistry() *Registry {\n\treturn &Registry{\n\t\thandlers: make(map[types.SearchType]interfaces.Handler),\n\t}\n}\n\n// Register registers a handler for a search type\nfunc (r *Registry) Register(handler interfaces.Handler) {\n\tr.handlers[handler.Type()] = handler\n}\n\n// Get returns the handler for a search type\nfunc (r *Registry) Get(t types.SearchType) (interfaces.Handler, bool) {\n\th, ok := r.handlers[t]\n\treturn h, ok\n}\n" }, { "path": "agent/search/registry_test.go", "content": "package search\n\nimport (\n\t\"testing\"\n\n\t\"github.com/stretchr/testify/assert\"\n\t\"github.com/yaoapp/yao/agent/search/handlers/db\"\n\t\"github.com/yaoapp/yao/agent/search/handlers/kb\"\n\t\"github.com/yaoapp/yao/agent/search/handlers/web\"\n\t\"github.com/yaoapp/yao/agent/search/types\"\n)\n\nfunc TestNewRegistry(t *testing.T) {\n\tr := NewRegistry()\n\tassert.NotNil(t, r)\n\tassert.NotNil(t, r.handlers)\n\tassert.Equal(t, 0, len(r.handlers))\n}\n\nfunc TestRegistry_Register(t *testing.T) {\n\tr := NewRegistry()\n\n\t// Register web handler\n\twebHandler := web.NewHandler(\"builtin\", nil)\n\tr.Register(webHandler)\n\n\th, ok := r.Get(types.SearchTypeWeb)\n\tassert.True(t, ok)\n\tassert.Equal(t, types.SearchTypeWeb, h.Type())\n}\n\nfunc TestRegistry_RegisterMultiple(t *testing.T) {\n\tr := NewRegistry()\n\n\t// Register all handlers\n\tr.Register(web.NewHandler(\"builtin\", nil))\n\tr.Register(kb.NewHandler(nil))\n\tr.Register(db.NewHandler(\"builtin\", nil))\n\n\t// Verify all are registered\n\twebH, ok := r.Get(types.SearchTypeWeb)\n\tassert.True(t, ok)\n\tassert.Equal(t, types.SearchTypeWeb, webH.Type())\n\n\tkbH, ok := r.Get(types.SearchTypeKB)\n\tassert.True(t, ok)\n\tassert.Equal(t, types.SearchTypeKB, kbH.Type())\n\n\tdbH, ok := r.Get(types.SearchTypeDB)\n\tassert.True(t, ok)\n\tassert.Equal(t, types.SearchTypeDB, dbH.Type())\n}\n\nfunc TestRegistry_Get_NotFound(t *testing.T) {\n\tr := NewRegistry()\n\n\th, ok := r.Get(types.SearchTypeWeb)\n\tassert.False(t, ok)\n\tassert.Nil(t, h)\n}\n\nfunc TestRegistry_RegisterOverwrite(t *testing.T) {\n\tr := NewRegistry()\n\n\t// Register first handler\n\th1 := web.NewHandler(\"builtin\", nil)\n\tr.Register(h1)\n\n\t// Register second handler (same type)\n\th2 := web.NewHandler(\"agent\", nil)\n\tr.Register(h2)\n\n\t// Should get the second handler\n\th, ok := r.Get(types.SearchTypeWeb)\n\tassert.True(t, ok)\n\tassert.NotNil(t, h)\n}\n" }, { "path": "agent/search/rerank/agent.go", "content": "package rerank\n\nimport (\n\t\"encoding/json\"\n\t\"fmt\"\n\t\"strings\"\n\n\t\"github.com/yaoapp/yao/agent/caller\"\n\t\"github.com/yaoapp/yao/agent/context\"\n\t\"github.com/yaoapp/yao/agent/search/types\"\n)\n\n// AgentProvider implements reranking by delegating to another agent\n// The agent should have a Next Hook that accepts rerank request and returns reordered items\ntype AgentProvider struct {\n\tagentID string // Assistant ID to delegate to\n}\n\n// NewAgentProvider creates a new agent reranker\nfunc NewAgentProvider(agentID string) *AgentProvider {\n\treturn &AgentProvider{agentID: agentID}\n}\n\n// Rerank delegates reranking to an LLM-powered assistant\n// The assistant receives items and query, returns reordered item IDs or items\nfunc (p *AgentProvider) Rerank(ctx *context.Context, query string, items []*types.ResultItem, opts *types.RerankOptions) ([]*types.ResultItem, error) {\n\tif ctx == nil {\n\t\treturn nil, fmt.Errorf(\"context is required for agent rerank\")\n\t}\n\n\t// Get agent via caller interface (avoids circular dependency)\n\tagent, err := caller.AgentGetterFunc(p.agentID)\n\tif err != nil {\n\t\treturn nil, fmt.Errorf(\"failed to get agent %s: %w\", p.agentID, err)\n\t}\n\n\t// Build request message with items to rerank\n\trequestData := map[string]interface{}{\n\t\t\"query\": query,\n\t\t\"items\": items,\n\t\t\"top_n\": opts.TopN,\n\t\t\"action\": \"rerank\",\n\t}\n\trequestJSON, _ := json.Marshal(requestData)\n\n\t// Create messages for agent\n\tmessages := []context.Message{\n\t\t{\n\t\t\tRole: \"user\",\n\t\t\tContent: string(requestJSON),\n\t\t},\n\t}\n\n\t// Call agent's Stream method with skip options (no history, no output)\n\toptions := &context.Options{\n\t\tSkip: &context.Skip{\n\t\t\tHistory: true,\n\t\t\tOutput: true,\n\t\t},\n\t}\n\n\tresponse, err := agent.Stream(ctx, messages, options)\n\tif err != nil {\n\t\treturn nil, fmt.Errorf(\"agent stream failed: %w\", err)\n\t}\n\n\t// Parse response from response.Next\n\treturn p.parseAgentResponse(response, items, opts)\n}\n\n// parseAgentResponse extracts reranked items from agent's *context.Response\n// Now that agent.Stream() returns *context.Response directly,\n// we can access fields without type assertions.\n//\n// Expected response.Next format:\n// { \"order\": [\"ref_001\", \"ref_003\", \"ref_002\"] }\n// Or: { \"items\": [{ \"citation_id\": \"ref_001\", ... }, ...] }\nfunc (p *AgentProvider) parseAgentResponse(response *context.Response, originalItems []*types.ResultItem, opts *types.RerankOptions) ([]*types.ResultItem, error) {\n\tif response == nil || response.Next == nil {\n\t\treturn originalItems, nil\n\t}\n\n\t// Build index map for quick lookup\n\titemMap := make(map[string]*types.ResultItem)\n\tfor _, item := range originalItems {\n\t\tif item.CitationID != \"\" {\n\t\t\titemMap[item.CitationID] = item\n\t\t}\n\t}\n\n\t// Extract response data from Next field\n\tdata := extractNextData(response.Next)\n\tif data == nil {\n\t\treturn originalItems, nil\n\t}\n\n\t// Try to get reranked order from data\n\t// Expected format: { \"order\": [\"ref_001\", \"ref_003\", \"ref_002\"] }\n\t// Or: { \"items\": [{ \"citation_id\": \"ref_001\", ... }, ...] }\n\n\tvar reranked []*types.ResultItem\n\n\t// Try \"order\" field (list of citation IDs)\n\tif order, ok := data[\"order\"]; ok {\n\t\tif orderList := toStringSlice(order); len(orderList) > 0 {\n\t\t\tfor _, id := range orderList {\n\t\t\t\tif item, exists := itemMap[id]; exists {\n\t\t\t\t\treranked = append(reranked, item)\n\t\t\t\t\tdelete(itemMap, id) // Avoid duplicates\n\t\t\t\t}\n\t\t\t}\n\t\t\t// Append remaining items not in order\n\t\t\tfor _, item := range originalItems {\n\t\t\t\tif _, exists := itemMap[item.CitationID]; exists {\n\t\t\t\t\treranked = append(reranked, item)\n\t\t\t\t}\n\t\t\t}\n\t\t}\n\t}\n\n\t// Try \"items\" field (full items or items with citation_id)\n\tif len(reranked) == 0 {\n\t\tif items, ok := data[\"items\"]; ok {\n\t\t\tif itemsList := toItemsList(items); len(itemsList) > 0 {\n\t\t\t\tfor _, respItem := range itemsList {\n\t\t\t\t\t// Check if it's just a reference or full item\n\t\t\t\t\tif citationID, ok := respItem[\"citation_id\"].(string); ok {\n\t\t\t\t\t\tif item, exists := itemMap[citationID]; exists {\n\t\t\t\t\t\t\treranked = append(reranked, item)\n\t\t\t\t\t\t\tdelete(itemMap, citationID)\n\t\t\t\t\t\t}\n\t\t\t\t\t}\n\t\t\t\t}\n\t\t\t\t// Append remaining items\n\t\t\t\tfor _, item := range originalItems {\n\t\t\t\t\tif _, exists := itemMap[item.CitationID]; exists {\n\t\t\t\t\t\treranked = append(reranked, item)\n\t\t\t\t\t}\n\t\t\t\t}\n\t\t\t}\n\t\t}\n\t}\n\n\t// If no valid response, return original items\n\tif len(reranked) == 0 {\n\t\treranked = originalItems\n\t}\n\n\t// Apply top N\n\tif opts.TopN > 0 && opts.TopN < len(reranked) {\n\t\treranked = reranked[:opts.TopN]\n\t}\n\n\treturn reranked, nil\n}\n\n// extractNextData extracts the actual data from response.Next field\n// Handles nested structures like { \"data\": { ... } }\nfunc extractNextData(next interface{}) map[string]interface{} {\n\tif next == nil {\n\t\treturn nil\n\t}\n\n\tswitch v := next.(type) {\n\tcase map[string]interface{}:\n\t\t// Check for \"data\" wrapper\n\t\tif data, ok := v[\"data\"].(map[string]interface{}); ok {\n\t\t\treturn data\n\t\t}\n\t\treturn v\n\tcase string:\n\t\t// Try to parse as JSON\n\t\tvar data map[string]interface{}\n\t\tif err := json.Unmarshal([]byte(v), &data); err == nil {\n\t\t\treturn extractNextData(data)\n\t\t}\n\t}\n\t// Try to handle other types by converting to JSON and back\n\tif bytes, err := json.Marshal(next); err == nil {\n\t\tvar data map[string]interface{}\n\t\tif err := json.Unmarshal(bytes, &data); err == nil {\n\t\t\treturn extractNextData(data)\n\t\t}\n\t}\n\treturn nil\n}\n\n// toStringSlice converts interface to string slice\nfunc toStringSlice(v interface{}) []string {\n\tswitch val := v.(type) {\n\tcase []string:\n\t\treturn val\n\tcase []interface{}:\n\t\tresult := make([]string, 0, len(val))\n\t\tfor _, item := range val {\n\t\t\tif s, ok := item.(string); ok {\n\t\t\t\tresult = append(result, s)\n\t\t\t}\n\t\t}\n\t\treturn result\n\t}\n\treturn nil\n}\n\n// toItemsList converts interface to list of maps\nfunc toItemsList(v interface{}) []map[string]interface{} {\n\tswitch val := v.(type) {\n\tcase []map[string]interface{}:\n\t\treturn val\n\tcase []interface{}:\n\t\tresult := make([]map[string]interface{}, 0, len(val))\n\t\tfor _, item := range val {\n\t\t\tif m, ok := item.(map[string]interface{}); ok {\n\t\t\t\tresult = append(result, m)\n\t\t\t}\n\t\t}\n\t\treturn result\n\t}\n\treturn nil\n}\n\n// extractAgentID extracts assistant ID from uses.rerank value\n// For backward compatibility, strips any prefix if present\nfunc extractAgentID(usesRerank string) string {\n\t// Remove any prefix like \"agent:\" if present\n\tif strings.HasPrefix(usesRerank, \"agent:\") {\n\t\treturn strings.TrimPrefix(usesRerank, \"agent:\")\n\t}\n\treturn usesRerank\n}\n" }, { "path": "agent/search/rerank/agent_test.go", "content": "package rerank_test\n\nimport (\n\t\"testing\"\n\n\t\"github.com/stretchr/testify/assert\"\n\t\"github.com/stretchr/testify/require\"\n\t\"github.com/yaoapp/yao/agent/assistant\"\n\t\"github.com/yaoapp/yao/agent/context\"\n\t\"github.com/yaoapp/yao/agent/search/rerank\"\n\t\"github.com/yaoapp/yao/agent/search/types\"\n\t\"github.com/yaoapp/yao/agent/testutils\"\n\toauthTypes \"github.com/yaoapp/yao/openapi/oauth/types\"\n)\n\nfunc TestAgentProviderWithAssistantConfig(t *testing.T) {\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\t// Load the rerank-agent assistant\n\tast, err := assistant.Get(\"tests.rerank-agent\")\n\trequire.NoError(t, err)\n\trequire.NotNil(t, ast)\n\n\t// Create test context\n\tctx := newTestContext(t)\n\n\t// Create provider with test assistant\n\tprovider := rerank.NewAgentProvider(\"tests.rerank-agent\")\n\n\titems := []*types.ResultItem{\n\t\t{CitationID: \"ref_001\", Score: 0.9, Weight: 1.0, Title: \"First\"},\n\t\t{CitationID: \"ref_002\", Score: 0.8, Weight: 1.0, Title: \"Second\"},\n\t\t{CitationID: \"ref_003\", Score: 0.7, Weight: 1.0, Title: \"Third\"},\n\t}\n\n\tresult, err := provider.Rerank(ctx, \"test query\", items, &types.RerankOptions{TopN: 10})\n\n\trequire.NoError(t, err)\n\tassert.NotEmpty(t, result)\n\n\t// The mock agent reverses the order\n\t// So we expect: ref_003, ref_002, ref_001\n\tassert.Len(t, result, 3)\n\tassert.Equal(t, \"ref_003\", result[0].CitationID)\n\tassert.Equal(t, \"ref_002\", result[1].CitationID)\n\tassert.Equal(t, \"ref_001\", result[2].CitationID)\n}\n\nfunc TestAgentProviderWithTopN(t *testing.T) {\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\tctx := newTestContext(t)\n\tprovider := rerank.NewAgentProvider(\"tests.rerank-agent\")\n\n\titems := []*types.ResultItem{\n\t\t{CitationID: \"ref_001\", Score: 0.9, Weight: 1.0},\n\t\t{CitationID: \"ref_002\", Score: 0.8, Weight: 1.0},\n\t\t{CitationID: \"ref_003\", Score: 0.7, Weight: 1.0},\n\t\t{CitationID: \"ref_004\", Score: 0.6, Weight: 1.0},\n\t\t{CitationID: \"ref_005\", Score: 0.5, Weight: 1.0},\n\t}\n\n\t// Request top 2 only\n\tresult, err := provider.Rerank(ctx, \"test query\", items, &types.RerankOptions{TopN: 2})\n\n\trequire.NoError(t, err)\n\tassert.Len(t, result, 2)\n}\n\nfunc TestAgentProviderWithoutContext(t *testing.T) {\n\tprovider := rerank.NewAgentProvider(\"tests.rerank-agent\")\n\n\titems := []*types.ResultItem{\n\t\t{CitationID: \"ref_001\", Score: 0.9, Weight: 1.0},\n\t}\n\n\t_, err := provider.Rerank(nil, \"test query\", items, &types.RerankOptions{TopN: 10})\n\n\tassert.Error(t, err)\n\tassert.Contains(t, err.Error(), \"context is required\")\n}\n\nfunc TestAgentProviderAgentNotFound(t *testing.T) {\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\tctx := newTestContext(t)\n\tprovider := rerank.NewAgentProvider(\"non-existent-agent\")\n\n\titems := []*types.ResultItem{\n\t\t{CitationID: \"ref_001\", Score: 0.9, Weight: 1.0},\n\t}\n\n\t_, err := provider.Rerank(ctx, \"test query\", items, &types.RerankOptions{TopN: 10})\n\n\tassert.Error(t, err)\n\tassert.Contains(t, err.Error(), \"failed to get agent\")\n}\n\nfunc TestAgentProviderEmptyItems(t *testing.T) {\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\tctx := newTestContext(t)\n\tprovider := rerank.NewAgentProvider(\"tests.rerank-agent\")\n\n\tresult, err := provider.Rerank(ctx, \"test query\", []*types.ResultItem{}, &types.RerankOptions{TopN: 10})\n\n\trequire.NoError(t, err)\n\tassert.Empty(t, result)\n}\n\n// newTestContext creates a test context with required fields\nfunc newTestContext(t *testing.T) *context.Context {\n\tt.Helper()\n\tauthorized := &oauthTypes.AuthorizedInfo{\n\t\tUserID: \"test-user\",\n\t}\n\tchatID := \"test-chat-rerank\"\n\treturn context.New(t.Context(), authorized, chatID)\n}\n" }, { "path": "agent/search/rerank/builtin.go", "content": "package rerank\n\nimport (\n\t\"sort\"\n\n\t\"github.com/yaoapp/yao/agent/search/types\"\n)\n\n// BuiltinReranker implements simple score-based reranking\n// For production use cases requiring semantic understanding, use Agent or MCP mode.\ntype BuiltinReranker struct{}\n\n// NewBuiltinReranker creates a new builtin reranker\nfunc NewBuiltinReranker() *BuiltinReranker {\n\treturn &BuiltinReranker{}\n}\n\n// Rerank sorts items by weighted score (score * weight) and returns top N\n// This is a simple implementation without semantic understanding.\nfunc (r *BuiltinReranker) Rerank(query string, items []*types.ResultItem, opts *types.RerankOptions) ([]*types.ResultItem, error) {\n\tif len(items) == 0 {\n\t\treturn items, nil\n\t}\n\n\t// Calculate weighted scores\n\ttype scoredItem struct {\n\t\titem *types.ResultItem\n\t\tweightedScore float64\n\t}\n\n\tscored := make([]scoredItem, len(items))\n\tfor i, item := range items {\n\t\t// Weighted score = base score * source weight\n\t\t// Higher weight sources (user=1.0) get priority over lower (auto=0.6)\n\t\tweight := item.Weight\n\t\tif weight == 0 {\n\t\t\tweight = 0.6 // Default weight for items without weight\n\t\t}\n\t\tscored[i] = scoredItem{\n\t\t\titem: item,\n\t\t\tweightedScore: item.Score * weight,\n\t\t}\n\t}\n\n\t// Sort by weighted score descending\n\tsort.Slice(scored, func(i, j int) bool {\n\t\treturn scored[i].weightedScore > scored[j].weightedScore\n\t})\n\n\t// Get top N\n\ttopN := opts.TopN\n\tif topN <= 0 || topN > len(scored) {\n\t\ttopN = len(scored)\n\t}\n\n\tresult := make([]*types.ResultItem, topN)\n\tfor i := 0; i < topN; i++ {\n\t\tresult[i] = scored[i].item\n\t}\n\n\treturn result, nil\n}\n" }, { "path": "agent/search/rerank/builtin_test.go", "content": "package rerank\n\nimport (\n\t\"testing\"\n\n\t\"github.com/stretchr/testify/assert\"\n\t\"github.com/yaoapp/yao/agent/search/types\"\n)\n\nfunc TestBuiltinReranker_EmptyItems(t *testing.T) {\n\treranker := NewBuiltinReranker()\n\tresult, err := reranker.Rerank(\"test query\", []*types.ResultItem{}, &types.RerankOptions{TopN: 5})\n\n\tassert.NoError(t, err)\n\tassert.Empty(t, result)\n}\n\nfunc TestBuiltinReranker_SortByWeightedScore(t *testing.T) {\n\treranker := NewBuiltinReranker()\n\n\titems := []*types.ResultItem{\n\t\t{CitationID: \"ref_001\", Score: 0.8, Weight: 0.6}, // weighted: 0.48\n\t\t{CitationID: \"ref_002\", Score: 0.6, Weight: 1.0}, // weighted: 0.60\n\t\t{CitationID: \"ref_003\", Score: 0.9, Weight: 0.8}, // weighted: 0.72\n\t\t{CitationID: \"ref_004\", Score: 0.5, Weight: 1.0}, // weighted: 0.50\n\t}\n\n\tresult, err := reranker.Rerank(\"test query\", items, &types.RerankOptions{TopN: 10})\n\n\tassert.NoError(t, err)\n\tassert.Len(t, result, 4)\n\n\t// Should be sorted by weighted score: ref_003 (0.72) > ref_002 (0.60) > ref_004 (0.50) > ref_001 (0.48)\n\tassert.Equal(t, \"ref_003\", result[0].CitationID)\n\tassert.Equal(t, \"ref_002\", result[1].CitationID)\n\tassert.Equal(t, \"ref_004\", result[2].CitationID)\n\tassert.Equal(t, \"ref_001\", result[3].CitationID)\n}\n\nfunc TestBuiltinReranker_TopN(t *testing.T) {\n\treranker := NewBuiltinReranker()\n\n\titems := []*types.ResultItem{\n\t\t{CitationID: \"ref_001\", Score: 0.9, Weight: 1.0},\n\t\t{CitationID: \"ref_002\", Score: 0.8, Weight: 1.0},\n\t\t{CitationID: \"ref_003\", Score: 0.7, Weight: 1.0},\n\t\t{CitationID: \"ref_004\", Score: 0.6, Weight: 1.0},\n\t\t{CitationID: \"ref_005\", Score: 0.5, Weight: 1.0},\n\t}\n\n\tresult, err := reranker.Rerank(\"test query\", items, &types.RerankOptions{TopN: 3})\n\n\tassert.NoError(t, err)\n\tassert.Len(t, result, 3)\n\tassert.Equal(t, \"ref_001\", result[0].CitationID)\n\tassert.Equal(t, \"ref_002\", result[1].CitationID)\n\tassert.Equal(t, \"ref_003\", result[2].CitationID)\n}\n\nfunc TestBuiltinReranker_DefaultWeight(t *testing.T) {\n\treranker := NewBuiltinReranker()\n\n\t// Items without weight should use default 0.6\n\titems := []*types.ResultItem{\n\t\t{CitationID: \"ref_001\", Score: 0.9, Weight: 0}, // weighted: 0.9 * 0.6 = 0.54\n\t\t{CitationID: \"ref_002\", Score: 0.5, Weight: 1.0}, // weighted: 0.5 * 1.0 = 0.50\n\t}\n\n\tresult, err := reranker.Rerank(\"test query\", items, &types.RerankOptions{TopN: 10})\n\n\tassert.NoError(t, err)\n\tassert.Len(t, result, 2)\n\t// ref_001 (0.54) > ref_002 (0.50)\n\tassert.Equal(t, \"ref_001\", result[0].CitationID)\n\tassert.Equal(t, \"ref_002\", result[1].CitationID)\n}\n\nfunc TestBuiltinReranker_TopNLargerThanItems(t *testing.T) {\n\treranker := NewBuiltinReranker()\n\n\titems := []*types.ResultItem{\n\t\t{CitationID: \"ref_001\", Score: 0.9, Weight: 1.0},\n\t\t{CitationID: \"ref_002\", Score: 0.8, Weight: 1.0},\n\t}\n\n\t// TopN > len(items) should return all items\n\tresult, err := reranker.Rerank(\"test query\", items, &types.RerankOptions{TopN: 10})\n\n\tassert.NoError(t, err)\n\tassert.Len(t, result, 2)\n}\n\nfunc TestBuiltinReranker_ZeroTopN(t *testing.T) {\n\treranker := NewBuiltinReranker()\n\n\titems := []*types.ResultItem{\n\t\t{CitationID: \"ref_001\", Score: 0.9, Weight: 1.0},\n\t\t{CitationID: \"ref_002\", Score: 0.8, Weight: 1.0},\n\t}\n\n\t// TopN = 0 should return all items\n\tresult, err := reranker.Rerank(\"test query\", items, &types.RerankOptions{TopN: 0})\n\n\tassert.NoError(t, err)\n\tassert.Len(t, result, 2)\n}\n\nfunc TestBuiltinReranker_SameWeightedScore(t *testing.T) {\n\treranker := NewBuiltinReranker()\n\n\t// Items with same weighted score - order should be stable\n\titems := []*types.ResultItem{\n\t\t{CitationID: \"ref_001\", Score: 0.8, Weight: 1.0}, // weighted: 0.80\n\t\t{CitationID: \"ref_002\", Score: 0.8, Weight: 1.0}, // weighted: 0.80\n\t\t{CitationID: \"ref_003\", Score: 0.4, Weight: 1.0}, // weighted: 0.40\n\t}\n\n\tresult, err := reranker.Rerank(\"test query\", items, &types.RerankOptions{TopN: 10})\n\n\tassert.NoError(t, err)\n\tassert.Len(t, result, 3)\n\t// ref_003 should be last\n\tassert.Equal(t, \"ref_003\", result[2].CitationID)\n}\n" }, { "path": "agent/search/rerank/mcp.go", "content": "package rerank\n\nimport (\n\t\"encoding/json\"\n\t\"fmt\"\n\t\"strings\"\n\n\t\"github.com/yaoapp/gou/mcp\"\n\tgouMCPTypes \"github.com/yaoapp/gou/mcp/types\"\n\t\"github.com/yaoapp/yao/agent/context\"\n\t\"github.com/yaoapp/yao/agent/search/types\"\n)\n\n// MCPProvider implements reranking by calling an MCP tool\ntype MCPProvider struct {\n\tserverID string // MCP server ID\n\ttoolName string // Tool name\n}\n\n// NewMCPProvider creates a new MCP reranker\n// mcpRef format: \"server_id.tool_name\"\nfunc NewMCPProvider(mcpRef string) (*MCPProvider, error) {\n\tparts := strings.SplitN(mcpRef, \".\", 2)\n\tif len(parts) != 2 {\n\t\treturn nil, fmt.Errorf(\"invalid MCP format, expected 'server.tool', got '%s'\", mcpRef)\n\t}\n\treturn &MCPProvider{\n\t\tserverID: parts[0],\n\t\ttoolName: parts[1],\n\t}, nil\n}\n\n// Rerank calls MCP tool to rerank items\nfunc (p *MCPProvider) Rerank(ctx *context.Context, query string, items []*types.ResultItem, opts *types.RerankOptions) ([]*types.ResultItem, error) {\n\tif ctx == nil {\n\t\treturn nil, fmt.Errorf(\"context is required for MCP rerank\")\n\t}\n\n\t// Get MCP client\n\tclient, err := mcp.Select(p.serverID)\n\tif err != nil {\n\t\treturn nil, fmt.Errorf(\"MCP server %s not found: %w\", p.serverID, err)\n\t}\n\n\t// Build arguments for MCP tool\n\targs := map[string]interface{}{\n\t\t\"query\": query,\n\t\t\"items\": items,\n\t\t\"top_n\": opts.TopN,\n\t}\n\n\t// Call MCP tool\n\tresult, err := client.CallTool(ctx.Context, p.toolName, args)\n\tif err != nil {\n\t\treturn nil, fmt.Errorf(\"MCP tool call failed: %w\", err)\n\t}\n\n\t// Parse result\n\treturn p.parseResult(result, items, opts)\n}\n\n// parseResult extracts reranked items from MCP response\nfunc (p *MCPProvider) parseResult(result *gouMCPTypes.CallToolResponse, originalItems []*types.ResultItem, opts *types.RerankOptions) ([]*types.ResultItem, error) {\n\tif result == nil || len(result.Content) == 0 {\n\t\treturn originalItems, nil\n\t}\n\n\t// Build index map for quick lookup\n\titemMap := make(map[string]*types.ResultItem)\n\tfor _, item := range originalItems {\n\t\tif item.CitationID != \"\" {\n\t\t\titemMap[item.CitationID] = item\n\t\t}\n\t}\n\n\t// Extract text content from MCP response\n\tvar textContent string\n\tfor _, content := range result.Content {\n\t\tif content.Type == gouMCPTypes.ToolContentTypeText && content.Text != \"\" {\n\t\t\ttextContent = content.Text\n\t\t\tbreak\n\t\t}\n\t}\n\n\tif textContent == \"\" {\n\t\treturn originalItems, nil\n\t}\n\n\t// Parse JSON response\n\tvar response map[string]interface{}\n\tif err := json.Unmarshal([]byte(textContent), &response); err != nil {\n\t\t// Try parsing as array of IDs\n\t\tvar orderList []string\n\t\tif err := json.Unmarshal([]byte(textContent), &orderList); err == nil {\n\t\t\treturn p.reorderByIDs(orderList, itemMap, originalItems, opts)\n\t\t}\n\t\treturn originalItems, nil\n\t}\n\n\t// Try \"order\" field (list of citation IDs)\n\tif order, ok := response[\"order\"]; ok {\n\t\tif orderList := toStringSlice(order); len(orderList) > 0 {\n\t\t\treturn p.reorderByIDs(orderList, itemMap, originalItems, opts)\n\t\t}\n\t}\n\n\t// Try \"items\" field\n\tif items, ok := response[\"items\"]; ok {\n\t\tif itemsList := toItemsList(items); len(itemsList) > 0 {\n\t\t\treturn p.reorderByItems(itemsList, itemMap, originalItems, opts)\n\t\t}\n\t}\n\n\treturn originalItems, nil\n}\n\n// reorderByIDs reorders items based on list of citation IDs\nfunc (p *MCPProvider) reorderByIDs(order []string, itemMap map[string]*types.ResultItem, originalItems []*types.ResultItem, opts *types.RerankOptions) ([]*types.ResultItem, error) {\n\tvar result []*types.ResultItem\n\n\t// Add items in specified order\n\tfor _, id := range order {\n\t\tif item, exists := itemMap[id]; exists {\n\t\t\tresult = append(result, item)\n\t\t\tdelete(itemMap, id)\n\t\t}\n\t}\n\n\t// Append remaining items\n\tfor _, item := range originalItems {\n\t\tif _, exists := itemMap[item.CitationID]; exists {\n\t\t\tresult = append(result, item)\n\t\t}\n\t}\n\n\t// Apply top N\n\tif opts.TopN > 0 && opts.TopN < len(result) {\n\t\tresult = result[:opts.TopN]\n\t}\n\n\treturn result, nil\n}\n\n// reorderByItems reorders items based on list of item references\nfunc (p *MCPProvider) reorderByItems(itemsList []map[string]interface{}, itemMap map[string]*types.ResultItem, originalItems []*types.ResultItem, opts *types.RerankOptions) ([]*types.ResultItem, error) {\n\tvar result []*types.ResultItem\n\n\t// Add items in specified order\n\tfor _, respItem := range itemsList {\n\t\tif citationID, ok := respItem[\"citation_id\"].(string); ok {\n\t\t\tif item, exists := itemMap[citationID]; exists {\n\t\t\t\tresult = append(result, item)\n\t\t\t\tdelete(itemMap, citationID)\n\t\t\t}\n\t\t}\n\t}\n\n\t// Append remaining items\n\tfor _, item := range originalItems {\n\t\tif _, exists := itemMap[item.CitationID]; exists {\n\t\t\tresult = append(result, item)\n\t\t}\n\t}\n\n\t// Apply top N\n\tif opts.TopN > 0 && opts.TopN < len(result) {\n\t\tresult = result[:opts.TopN]\n\t}\n\n\treturn result, nil\n}\n" }, { "path": "agent/search/rerank/mcp_test.go", "content": "package rerank_test\n\nimport (\n\t\"testing\"\n\n\t\"github.com/stretchr/testify/assert\"\n\t\"github.com/stretchr/testify/require\"\n\t\"github.com/yaoapp/yao/agent/context\"\n\t\"github.com/yaoapp/yao/agent/search/rerank\"\n\t\"github.com/yaoapp/yao/agent/search/types\"\n\t\"github.com/yaoapp/yao/agent/testutils\"\n\toauthTypes \"github.com/yaoapp/yao/openapi/oauth/types\"\n)\n\nfunc TestMCPProviderWithSearchRerank(t *testing.T) {\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\tctx := newMCPTestContext(t)\n\n\tprovider, err := rerank.NewMCPProvider(\"search.rerank\")\n\trequire.NoError(t, err)\n\n\titems := []*types.ResultItem{\n\t\t{CitationID: \"ref_001\", Score: 0.9, Weight: 1.0, Title: \"First\"},\n\t\t{CitationID: \"ref_002\", Score: 0.8, Weight: 1.0, Title: \"Second\"},\n\t\t{CitationID: \"ref_003\", Score: 0.7, Weight: 1.0, Title: \"Third\"},\n\t}\n\n\tresult, err := provider.Rerank(ctx, \"test query\", items, &types.RerankOptions{TopN: 10})\n\n\trequire.NoError(t, err)\n\tassert.NotEmpty(t, result)\n\n\t// The mock MCP reverses the order\n\t// So we expect: ref_003, ref_002, ref_001\n\tassert.Len(t, result, 3)\n\tassert.Equal(t, \"ref_003\", result[0].CitationID)\n\tassert.Equal(t, \"ref_002\", result[1].CitationID)\n\tassert.Equal(t, \"ref_001\", result[2].CitationID)\n}\n\nfunc TestMCPProviderWithTopN(t *testing.T) {\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\tctx := newMCPTestContext(t)\n\n\tprovider, err := rerank.NewMCPProvider(\"search.rerank\")\n\trequire.NoError(t, err)\n\n\titems := []*types.ResultItem{\n\t\t{CitationID: \"ref_001\", Score: 0.9, Weight: 1.0},\n\t\t{CitationID: \"ref_002\", Score: 0.8, Weight: 1.0},\n\t\t{CitationID: \"ref_003\", Score: 0.7, Weight: 1.0},\n\t\t{CitationID: \"ref_004\", Score: 0.6, Weight: 1.0},\n\t\t{CitationID: \"ref_005\", Score: 0.5, Weight: 1.0},\n\t}\n\n\t// Request top 2 only\n\tresult, err := provider.Rerank(ctx, \"test query\", items, &types.RerankOptions{TopN: 2})\n\n\trequire.NoError(t, err)\n\tassert.Len(t, result, 2)\n}\n\nfunc TestMCPProviderInvalidFormat(t *testing.T) {\n\t_, err := rerank.NewMCPProvider(\"invalid-format\")\n\n\tassert.Error(t, err)\n\tassert.Contains(t, err.Error(), \"invalid MCP format\")\n}\n\nfunc TestMCPProviderServerNotFound(t *testing.T) {\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\tctx := newMCPTestContext(t)\n\n\tprovider, err := rerank.NewMCPProvider(\"nonexistent.rerank\")\n\trequire.NoError(t, err)\n\n\titems := []*types.ResultItem{\n\t\t{CitationID: \"ref_001\", Score: 0.9, Weight: 1.0},\n\t}\n\n\t_, err = provider.Rerank(ctx, \"test query\", items, &types.RerankOptions{TopN: 10})\n\n\tassert.Error(t, err)\n\tassert.Contains(t, err.Error(), \"not found\")\n}\n\nfunc TestMCPProviderToolNotFound(t *testing.T) {\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\tctx := newMCPTestContext(t)\n\n\tprovider, err := rerank.NewMCPProvider(\"search.nonexistent_tool\")\n\trequire.NoError(t, err)\n\n\titems := []*types.ResultItem{\n\t\t{CitationID: \"ref_001\", Score: 0.9, Weight: 1.0},\n\t}\n\n\t_, err = provider.Rerank(ctx, \"test query\", items, &types.RerankOptions{TopN: 10})\n\n\tassert.Error(t, err)\n}\n\nfunc TestMCPProviderWithoutContext(t *testing.T) {\n\tprovider, err := rerank.NewMCPProvider(\"search.rerank\")\n\trequire.NoError(t, err)\n\n\titems := []*types.ResultItem{\n\t\t{CitationID: \"ref_001\", Score: 0.9, Weight: 1.0},\n\t}\n\n\t_, err = provider.Rerank(nil, \"test query\", items, &types.RerankOptions{TopN: 10})\n\n\tassert.Error(t, err)\n\tassert.Contains(t, err.Error(), \"context is required\")\n}\n\nfunc TestMCPProviderEmptyItems(t *testing.T) {\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\tctx := newMCPTestContext(t)\n\n\tprovider, err := rerank.NewMCPProvider(\"search.rerank\")\n\trequire.NoError(t, err)\n\n\tresult, err := provider.Rerank(ctx, \"test query\", []*types.ResultItem{}, &types.RerankOptions{TopN: 10})\n\n\trequire.NoError(t, err)\n\tassert.Empty(t, result)\n}\n\n// newMCPTestContext creates a test context with required fields\nfunc newMCPTestContext(t *testing.T) *context.Context {\n\tt.Helper()\n\tauthorized := &oauthTypes.AuthorizedInfo{\n\t\tUserID: \"test-user\",\n\t}\n\tchatID := \"test-chat-rerank-mcp\"\n\treturn context.New(t.Context(), authorized, chatID)\n}\n" }, { "path": "agent/search/rerank/reranker.go", "content": "// Package rerank provides result reranking for search module\n// Supports three modes via uses.rerank configuration:\n// - \"builtin\": Simple score-based sorting (no external dependencies)\n// - \"\": Delegate to an LLM-powered assistant for semantic reranking\n// - \"mcp:.\": Call external MCP tool\n//\n// For production use cases requiring high accuracy, use Agent or MCP mode.\npackage rerank\n\nimport (\n\t\"strings\"\n\n\t\"github.com/yaoapp/yao/agent/context\"\n\t\"github.com/yaoapp/yao/agent/search/types\"\n)\n\n// Reranker reorders search results by relevance\n// Mode is determined by uses.rerank configuration\ntype Reranker struct {\n\tusesRerank string // \"builtin\", \"\", \"mcp:.\"\n\tconfig *types.RerankConfig // Rerank options\n}\n\n// NewReranker creates a new reranker\n// usesRerank: value from uses.rerank config\n// cfg: rerank options from search config\nfunc NewReranker(usesRerank string, cfg *types.RerankConfig) *Reranker {\n\treturn &Reranker{\n\t\tusesRerank: usesRerank,\n\t\tconfig: cfg,\n\t}\n}\n\n// Rerank reorders results based on configured mode\n// Returns reordered items, potentially truncated to top N\nfunc (r *Reranker) Rerank(ctx *context.Context, query string, items []*types.ResultItem, opts *types.RerankOptions) ([]*types.ResultItem, error) {\n\tif len(items) == 0 {\n\t\treturn items, nil\n\t}\n\n\t// Merge options with config defaults\n\tmergedOpts := r.mergeOptions(opts)\n\n\tswitch {\n\tcase r.usesRerank == \"builtin\" || r.usesRerank == \"\":\n\t\treturn r.builtinRerank(query, items, mergedOpts)\n\tcase strings.HasPrefix(r.usesRerank, \"mcp:\"):\n\t\treturn r.mcpRerank(ctx, query, items, mergedOpts)\n\tdefault:\n\t\t// Assume it's an assistant ID for Agent mode\n\t\treturn r.agentRerank(ctx, query, items, mergedOpts)\n\t}\n}\n\n// mergeOptions merges runtime options with config defaults\nfunc (r *Reranker) mergeOptions(opts *types.RerankOptions) *types.RerankOptions {\n\tresult := &types.RerankOptions{\n\t\tTopN: 10, // default\n\t}\n\n\t// Apply config defaults\n\tif r.config != nil {\n\t\tif r.config.TopN > 0 {\n\t\t\tresult.TopN = r.config.TopN\n\t\t}\n\t}\n\n\t// Apply runtime options (highest priority)\n\tif opts != nil {\n\t\tif opts.TopN > 0 {\n\t\t\tresult.TopN = opts.TopN\n\t\t}\n\t}\n\n\treturn result\n}\n\n// builtinRerank uses simple score-based sorting\nfunc (r *Reranker) builtinRerank(query string, items []*types.ResultItem, opts *types.RerankOptions) ([]*types.ResultItem, error) {\n\treranker := NewBuiltinReranker()\n\treturn reranker.Rerank(query, items, opts)\n}\n\n// agentRerank delegates to an LLM-powered assistant\nfunc (r *Reranker) agentRerank(ctx *context.Context, query string, items []*types.ResultItem, opts *types.RerankOptions) ([]*types.ResultItem, error) {\n\tprovider := NewAgentProvider(r.usesRerank)\n\treturn provider.Rerank(ctx, query, items, opts)\n}\n\n// mcpRerank calls an external MCP tool\nfunc (r *Reranker) mcpRerank(ctx *context.Context, query string, items []*types.ResultItem, opts *types.RerankOptions) ([]*types.ResultItem, error) {\n\tmcpRef := strings.TrimPrefix(r.usesRerank, \"mcp:\")\n\tprovider, err := NewMCPProvider(mcpRef)\n\tif err != nil {\n\t\t// Fallback to builtin on invalid MCP format\n\t\treturn r.builtinRerank(query, items, opts)\n\t}\n\treturn provider.Rerank(ctx, query, items, opts)\n}\n" }, { "path": "agent/search/rerank/reranker_test.go", "content": "package rerank\n\nimport (\n\t\"testing\"\n\n\t\"github.com/stretchr/testify/assert\"\n\t\"github.com/yaoapp/yao/agent/search/types\"\n)\n\nfunc TestReranker_BuiltinMode(t *testing.T) {\n\treranker := NewReranker(\"builtin\", &types.RerankConfig{TopN: 5})\n\n\titems := []*types.ResultItem{\n\t\t{CitationID: \"ref_001\", Score: 0.9, Weight: 1.0},\n\t\t{CitationID: \"ref_002\", Score: 0.8, Weight: 1.0},\n\t\t{CitationID: \"ref_003\", Score: 0.7, Weight: 1.0},\n\t}\n\n\tresult, err := reranker.Rerank(nil, \"test query\", items, nil)\n\n\tassert.NoError(t, err)\n\tassert.Len(t, result, 3)\n\tassert.Equal(t, \"ref_001\", result[0].CitationID)\n}\n\nfunc TestReranker_EmptyUsesRerank(t *testing.T) {\n\t// Empty usesRerank should use builtin\n\treranker := NewReranker(\"\", &types.RerankConfig{TopN: 5})\n\n\titems := []*types.ResultItem{\n\t\t{CitationID: \"ref_001\", Score: 0.9, Weight: 1.0},\n\t}\n\n\tresult, err := reranker.Rerank(nil, \"test query\", items, nil)\n\n\tassert.NoError(t, err)\n\tassert.Len(t, result, 1)\n}\n\nfunc TestReranker_MergeOptions(t *testing.T) {\n\t// Config sets TopN = 5\n\treranker := NewReranker(\"builtin\", &types.RerankConfig{TopN: 5})\n\n\titems := []*types.ResultItem{\n\t\t{CitationID: \"ref_001\", Score: 0.9, Weight: 1.0},\n\t\t{CitationID: \"ref_002\", Score: 0.8, Weight: 1.0},\n\t\t{CitationID: \"ref_003\", Score: 0.7, Weight: 1.0},\n\t\t{CitationID: \"ref_004\", Score: 0.6, Weight: 1.0},\n\t\t{CitationID: \"ref_005\", Score: 0.5, Weight: 1.0},\n\t\t{CitationID: \"ref_006\", Score: 0.4, Weight: 1.0},\n\t}\n\n\t// Runtime opts override config\n\tresult, err := reranker.Rerank(nil, \"test query\", items, &types.RerankOptions{TopN: 3})\n\n\tassert.NoError(t, err)\n\tassert.Len(t, result, 3)\n}\n\nfunc TestReranker_ConfigTopN(t *testing.T) {\n\t// Config sets TopN = 3\n\treranker := NewReranker(\"builtin\", &types.RerankConfig{TopN: 3})\n\n\titems := []*types.ResultItem{\n\t\t{CitationID: \"ref_001\", Score: 0.9, Weight: 1.0},\n\t\t{CitationID: \"ref_002\", Score: 0.8, Weight: 1.0},\n\t\t{CitationID: \"ref_003\", Score: 0.7, Weight: 1.0},\n\t\t{CitationID: \"ref_004\", Score: 0.6, Weight: 1.0},\n\t\t{CitationID: \"ref_005\", Score: 0.5, Weight: 1.0},\n\t}\n\n\t// No runtime opts, should use config TopN\n\tresult, err := reranker.Rerank(nil, \"test query\", items, nil)\n\n\tassert.NoError(t, err)\n\tassert.Len(t, result, 3)\n}\n\nfunc TestReranker_NilConfig(t *testing.T) {\n\treranker := NewReranker(\"builtin\", nil)\n\n\titems := []*types.ResultItem{\n\t\t{CitationID: \"ref_001\", Score: 0.9, Weight: 1.0},\n\t}\n\n\tresult, err := reranker.Rerank(nil, \"test query\", items, nil)\n\n\tassert.NoError(t, err)\n\tassert.Len(t, result, 1)\n}\n\nfunc TestReranker_EmptyItems(t *testing.T) {\n\treranker := NewReranker(\"builtin\", &types.RerankConfig{TopN: 5})\n\n\tresult, err := reranker.Rerank(nil, \"test query\", []*types.ResultItem{}, nil)\n\n\tassert.NoError(t, err)\n\tassert.Empty(t, result)\n}\n" }, { "path": "agent/search/search.go", "content": "package search\n\nimport (\n\t\"sync\"\n\n\t\"github.com/yaoapp/yao/agent/context\"\n\t\"github.com/yaoapp/yao/agent/search/handlers/db\"\n\t\"github.com/yaoapp/yao/agent/search/handlers/kb\"\n\t\"github.com/yaoapp/yao/agent/search/handlers/web\"\n\t\"github.com/yaoapp/yao/agent/search/interfaces\"\n\t\"github.com/yaoapp/yao/agent/search/rerank\"\n\t\"github.com/yaoapp/yao/agent/search/types\"\n)\n\n// Searcher is the main search implementation\ntype Searcher struct {\n\tconfig *types.Config // Merged config (global + assistant)\n\thandlers map[types.SearchType]interfaces.Handler\n\treranker *rerank.Reranker\n\tcitation *CitationGenerator\n}\n\n// Uses contains the search-specific uses configuration\n// These are extracted from context.Uses and search config\ntype Uses struct {\n\tSearch string // \"builtin\", \"disabled\", \"\", \"mcp:.\"\n\tWeb string // \"builtin\", \"\", \"mcp:.\"\n\tKeyword string // \"builtin\", \"\", \"mcp:.\"\n\tQueryDSL string // \"builtin\", \"\", \"mcp:.\"\n\tRerank string // \"builtin\", \"\", \"mcp:.\"\n}\n\n// New creates a new Searcher instance\n// cfg: merged config from agent/load.go + assistant config\n// uses: merged uses configuration (global → assistant → hook)\nfunc New(cfg *types.Config, uses *Uses) *Searcher {\n\tif uses == nil {\n\t\tuses = &Uses{}\n\t}\n\tif cfg == nil {\n\t\tcfg = &types.Config{}\n\t}\n\n\treturn &Searcher{\n\t\tconfig: cfg,\n\t\thandlers: map[types.SearchType]interfaces.Handler{\n\t\t\ttypes.SearchTypeWeb: web.NewHandler(uses.Web, cfg.Web),\n\t\t\ttypes.SearchTypeKB: kb.NewHandler(cfg.KB),\n\t\t\ttypes.SearchTypeDB: db.NewHandler(uses.QueryDSL, cfg.DB),\n\t\t},\n\t\treranker: rerank.NewReranker(uses.Rerank, cfg.Rerank),\n\t\tcitation: NewCitationGenerator(),\n\t}\n}\n\n// Search executes a single search request\nfunc (s *Searcher) Search(ctx *context.Context, req *types.Request) (*types.Result, error) {\n\thandler, ok := s.handlers[req.Type]\n\tif !ok {\n\t\treturn &types.Result{Error: \"unsupported search type\"}, nil\n\t}\n\n\t// Execute search - use context if handler supports it\n\tvar result *types.Result\n\tvar err error\n\tif ctxHandler, ok := handler.(interfaces.ContextHandler); ok {\n\t\tresult, err = ctxHandler.SearchWithContext(ctx, req)\n\t} else {\n\t\tresult, err = handler.Search(req)\n\t}\n\tif err != nil {\n\t\treturn &types.Result{Error: err.Error()}, nil\n\t}\n\n\t// Assign weights based on source\n\tfor _, item := range result.Items {\n\t\titem.Weight = s.config.GetWeight(req.Source)\n\t}\n\n\t// Rerank if requested\n\tif req.Rerank != nil && s.reranker != nil {\n\t\tresult.Items, _ = s.reranker.Rerank(ctx, req.Query, result.Items, req.Rerank)\n\t}\n\n\t// Generate citation IDs\n\tfor _, item := range result.Items {\n\t\titem.CitationID = s.citation.Next()\n\t}\n\n\treturn result, nil\n}\n\n// All executes all searches and waits for all to complete (like Promise.all)\nfunc (s *Searcher) All(ctx *context.Context, reqs []*types.Request) ([]*types.Result, error) {\n\tif len(reqs) == 0 {\n\t\treturn []*types.Result{}, nil\n\t}\n\treturn s.parallelAll(ctx, reqs)\n}\n\n// Any returns as soon as any search succeeds with results (like Promise.any)\nfunc (s *Searcher) Any(ctx *context.Context, reqs []*types.Request) ([]*types.Result, error) {\n\tif len(reqs) == 0 {\n\t\treturn []*types.Result{}, nil\n\t}\n\treturn s.parallelAny(ctx, reqs)\n}\n\n// Race returns as soon as any search completes (like Promise.race)\nfunc (s *Searcher) Race(ctx *context.Context, reqs []*types.Request) ([]*types.Result, error) {\n\tif len(reqs) == 0 {\n\t\treturn []*types.Result{}, nil\n\t}\n\treturn s.parallelRace(ctx, reqs)\n}\n\n// parallelAll executes all searches and waits for all to complete (like Promise.all)\nfunc (s *Searcher) parallelAll(ctx *context.Context, reqs []*types.Request) ([]*types.Result, error) {\n\tresults := make([]*types.Result, len(reqs))\n\tvar wg sync.WaitGroup\n\tvar mu sync.Mutex\n\n\tfor i, req := range reqs {\n\t\twg.Add(1)\n\t\tgo func(idx int, r *types.Request) {\n\t\t\tdefer wg.Done()\n\t\t\tdefer func() {\n\t\t\t\tif err := recover(); err != nil {\n\t\t\t\t\tmu.Lock()\n\t\t\t\t\tresults[idx] = &types.Result{Error: \"search panic recovered\"}\n\t\t\t\t\tmu.Unlock()\n\t\t\t\t}\n\t\t\t}()\n\n\t\t\tresult, err := s.Search(ctx, r)\n\t\t\tmu.Lock()\n\t\t\tif err != nil {\n\t\t\t\tresults[idx] = &types.Result{Error: err.Error()}\n\t\t\t} else if result == nil {\n\t\t\t\tresults[idx] = &types.Result{Error: \"empty result\"}\n\t\t\t} else {\n\t\t\t\tresults[idx] = result\n\t\t\t}\n\t\t\tmu.Unlock()\n\t\t}(i, req)\n\t}\n\n\twg.Wait()\n\treturn results, nil\n}\n\n// parallelAny returns as soon as any search succeeds (has results) (like Promise.any)\n// Other searches continue in background but results are discarded\nfunc (s *Searcher) parallelAny(ctx *context.Context, reqs []*types.Request) ([]*types.Result, error) {\n\tresults := make([]*types.Result, len(reqs))\n\tresultChan := make(chan struct {\n\t\tidx int\n\t\tresult *types.Result\n\t}, len(reqs))\n\n\tvar wg sync.WaitGroup\n\tdone := make(chan struct{})\n\n\tfor i, req := range reqs {\n\t\twg.Add(1)\n\t\tgo func(idx int, r *types.Request) {\n\t\t\tdefer wg.Done()\n\n\t\t\t// Check if done before starting\n\t\t\tselect {\n\t\t\tcase <-done:\n\t\t\t\treturn\n\t\t\tdefault:\n\t\t\t}\n\n\t\t\tresult, _ := s.Search(ctx, r)\n\n\t\t\t// Try to send result\n\t\t\tselect {\n\t\t\tcase <-done:\n\t\t\t\t// Already found a successful result\n\t\t\tcase resultChan <- struct {\n\t\t\t\tidx int\n\t\t\t\tresult *types.Result\n\t\t\t}{idx, result}:\n\t\t\t}\n\t\t}(i, req)\n\t}\n\n\t// Close channel when all goroutines complete\n\tgo func() {\n\t\twg.Wait()\n\t\tclose(resultChan)\n\t}()\n\n\t// Collect results until we find one with items (success)\n\tvar foundSuccess bool\n\tfor res := range resultChan {\n\t\tresults[res.idx] = res.result\n\t\t// Check if this result has items (success = has results and no error)\n\t\tif !foundSuccess && res.result != nil && len(res.result.Items) > 0 && res.result.Error == \"\" {\n\t\t\tfoundSuccess = true\n\t\t\tclose(done) // Signal other goroutines to stop\n\t\t}\n\t}\n\n\t// All goroutines have completed (resultChan is closed)\n\treturn results, nil\n}\n\n// parallelRace returns as soon as any search completes (like Promise.race)\n// Returns immediately when first result arrives, regardless of success/failure\n// Note: Still waits for all goroutines to complete before returning to avoid resource leaks\nfunc (s *Searcher) parallelRace(ctx *context.Context, reqs []*types.Request) ([]*types.Result, error) {\n\tresults := make([]*types.Result, len(reqs))\n\tresultChan := make(chan struct {\n\t\tidx int\n\t\tresult *types.Result\n\t}, len(reqs))\n\n\tvar wg sync.WaitGroup\n\tdone := make(chan struct{})\n\n\tfor i, req := range reqs {\n\t\twg.Add(1)\n\t\tgo func(idx int, r *types.Request) {\n\t\t\tdefer wg.Done()\n\n\t\t\t// Check if done before starting\n\t\t\tselect {\n\t\t\tcase <-done:\n\t\t\t\treturn\n\t\t\tdefault:\n\t\t\t}\n\n\t\t\tresult, _ := s.Search(ctx, r)\n\n\t\t\t// Try to send result\n\t\t\tselect {\n\t\t\tcase <-done:\n\t\t\t\t// Already got first result\n\t\t\tcase resultChan <- struct {\n\t\t\t\tidx int\n\t\t\t\tresult *types.Result\n\t\t\t}{idx, result}:\n\t\t\t}\n\t\t}(i, req)\n\t}\n\n\t// Close channel when all goroutines complete\n\tgo func() {\n\t\twg.Wait()\n\t\tclose(resultChan)\n\t}()\n\n\t// Get first result and signal others to stop\n\tvar gotFirst bool\n\tfor res := range resultChan {\n\t\tresults[res.idx] = res.result\n\t\tif !gotFirst {\n\t\t\tgotFirst = true\n\t\t\tclose(done) // Signal other goroutines to stop\n\t\t}\n\t}\n\n\t// All goroutines have completed (resultChan is closed)\n\treturn results, nil\n}\n\n// BuildReferences converts search results to unified Reference format\nfunc (s *Searcher) BuildReferences(results []*types.Result) []*types.Reference {\n\treturn BuildReferences(results)\n}\n" }, { "path": "agent/search/search_test.go", "content": "package search\n\nimport (\n\t\"testing\"\n\n\t\"github.com/stretchr/testify/assert\"\n\t\"github.com/yaoapp/yao/agent/search/types\"\n)\n\nfunc TestNew(t *testing.T) {\n\tt.Run(\"with nil config and uses\", func(t *testing.T) {\n\t\ts := New(nil, nil)\n\t\tassert.NotNil(t, s)\n\t\tassert.NotNil(t, s.config)\n\t\tassert.NotNil(t, s.handlers)\n\t\tassert.NotNil(t, s.citation)\n\t\tassert.Equal(t, 3, len(s.handlers)) // web, kb, db\n\t})\n\n\tt.Run(\"with config\", func(t *testing.T) {\n\t\tcfg := &types.Config{\n\t\t\tWeb: &types.WebConfig{\n\t\t\t\tProvider: \"tavily\",\n\t\t\t\tMaxResults: 10,\n\t\t\t},\n\t\t\tKB: &types.KBConfig{\n\t\t\t\tCollections: []string{\"docs\"},\n\t\t\t\tThreshold: 0.8,\n\t\t\t},\n\t\t\tDB: &types.DBConfig{\n\t\t\t\tModels: []string{\"product\"},\n\t\t\t\tMaxResults: 20,\n\t\t\t},\n\t\t}\n\t\ts := New(cfg, nil)\n\t\tassert.NotNil(t, s)\n\t\tassert.Equal(t, cfg, s.config)\n\t})\n\n\tt.Run(\"with uses\", func(t *testing.T) {\n\t\tuses := &Uses{\n\t\t\tSearch: \"builtin\",\n\t\t\tWeb: \"builtin\",\n\t\t\tKeyword: \"builtin\",\n\t\t\tQueryDSL: \"builtin\",\n\t\t\tRerank: \"builtin\",\n\t\t}\n\t\ts := New(nil, uses)\n\t\tassert.NotNil(t, s)\n\t})\n}\n\nfunc TestSearcher_Search_UnsupportedType(t *testing.T) {\n\ts := New(nil, nil)\n\n\treq := &types.Request{\n\t\tType: \"unsupported\",\n\t\tQuery: \"test\",\n\t}\n\n\tresult, err := s.Search(nil, req)\n\tassert.NoError(t, err)\n\tassert.NotNil(t, result)\n\tassert.Equal(t, \"unsupported search type\", result.Error)\n}\n\nfunc TestSearcher_Search_Web(t *testing.T) {\n\t// Note: This test uses skeleton handlers that return empty results\n\t// Real tests with actual API calls are in handlers/web/*_test.go\n\tcfg := &types.Config{\n\t\tWeb: &types.WebConfig{\n\t\t\tProvider: \"tavily\",\n\t\t},\n\t}\n\ts := New(cfg, &Uses{Web: \"builtin\"})\n\n\treq := &types.Request{\n\t\tType: types.SearchTypeWeb,\n\t\tQuery: \"test query\",\n\t\tSource: types.SourceAuto,\n\t}\n\n\tresult, err := s.Search(nil, req)\n\tassert.NoError(t, err)\n\tassert.NotNil(t, result)\n\tassert.Equal(t, types.SearchTypeWeb, result.Type)\n\tassert.Equal(t, \"test query\", result.Query)\n\t// Note: actual result depends on API key availability\n}\n\nfunc TestSearcher_Search_KB(t *testing.T) {\n\tcfg := &types.Config{\n\t\tKB: &types.KBConfig{\n\t\t\tCollections: []string{\"docs\"},\n\t\t\tThreshold: 0.7,\n\t\t},\n\t}\n\ts := New(cfg, nil)\n\n\treq := &types.Request{\n\t\tType: types.SearchTypeKB,\n\t\tQuery: \"test query\",\n\t\tSource: types.SourceHook,\n\t\tCollections: []string{\"docs\"},\n\t}\n\n\tresult, err := s.Search(nil, req)\n\tassert.NoError(t, err)\n\tassert.NotNil(t, result)\n\tassert.Equal(t, types.SearchTypeKB, result.Type)\n\tassert.Equal(t, \"test query\", result.Query)\n\tassert.Equal(t, types.SourceHook, result.Source)\n\t// Skeleton returns empty items\n\tassert.Equal(t, 0, len(result.Items))\n}\n\nfunc TestSearcher_Search_DB(t *testing.T) {\n\tcfg := &types.Config{\n\t\tDB: &types.DBConfig{\n\t\t\tModels: []string{\"product\"},\n\t\t\tMaxResults: 20,\n\t\t},\n\t}\n\ts := New(cfg, &Uses{QueryDSL: \"builtin\"})\n\n\treq := &types.Request{\n\t\tType: types.SearchTypeDB,\n\t\tQuery: \"find products under $100\",\n\t\tSource: types.SourceUser,\n\t\tModels: []string{\"product\"},\n\t}\n\n\tresult, err := s.Search(nil, req)\n\tassert.NoError(t, err)\n\tassert.NotNil(t, result)\n\tassert.Equal(t, types.SearchTypeDB, result.Type)\n\tassert.Equal(t, \"find products under $100\", result.Query)\n\tassert.Equal(t, types.SourceUser, result.Source)\n\t// Skeleton returns empty items\n\tassert.Equal(t, 0, len(result.Items))\n}\n\nfunc TestSearcher_Search_WeightAssignment(t *testing.T) {\n\tcfg := &types.Config{\n\t\tKB: &types.KBConfig{\n\t\t\tCollections: []string{\"docs\"},\n\t\t},\n\t\tWeights: &types.WeightsConfig{\n\t\t\tUser: 1.0,\n\t\t\tHook: 0.8,\n\t\t\tAuto: 0.6,\n\t\t},\n\t}\n\ts := New(cfg, nil)\n\n\t// Test with different sources\n\tsources := []struct {\n\t\tsource types.SourceType\n\t\tweight float64\n\t}{\n\t\t{types.SourceUser, 1.0},\n\t\t{types.SourceHook, 0.8},\n\t\t{types.SourceAuto, 0.6},\n\t}\n\n\tfor _, tc := range sources {\n\t\treq := &types.Request{\n\t\t\tType: types.SearchTypeKB,\n\t\t\tQuery: \"test\",\n\t\t\tSource: tc.source,\n\t\t\tCollections: []string{\"docs\"},\n\t\t}\n\t\tresult, err := s.Search(nil, req)\n\t\tassert.NoError(t, err)\n\t\tassert.NotNil(t, result)\n\t\t// Items are empty in skeleton, so weight assignment can't be verified here\n\t\t// This test ensures the code path works without error\n\t}\n}\n\nfunc TestSearcher_All(t *testing.T) {\n\tcfg := &types.Config{\n\t\tKB: &types.KBConfig{\n\t\t\tCollections: []string{\"docs\"},\n\t\t},\n\t\tDB: &types.DBConfig{\n\t\t\tModels: []string{\"product\"},\n\t\t},\n\t}\n\ts := New(cfg, nil)\n\n\treqs := []*types.Request{\n\t\t{\n\t\t\tType: types.SearchTypeKB,\n\t\t\tQuery: \"KB query\",\n\t\t\tSource: types.SourceAuto,\n\t\t\tCollections: []string{\"docs\"},\n\t\t},\n\t\t{\n\t\t\tType: types.SearchTypeDB,\n\t\t\tQuery: \"DB query\",\n\t\t\tSource: types.SourceAuto,\n\t\t\tModels: []string{\"product\"},\n\t\t},\n\t}\n\n\t// Test All() - waits for all searches to complete (like Promise.all)\n\tresults, err := s.All(nil, reqs)\n\tassert.NoError(t, err)\n\tassert.Equal(t, 2, len(results))\n\n\t// Verify each result corresponds to its request\n\tassert.Equal(t, types.SearchTypeKB, results[0].Type)\n\tassert.Equal(t, \"KB query\", results[0].Query)\n\n\tassert.Equal(t, types.SearchTypeDB, results[1].Type)\n\tassert.Equal(t, \"DB query\", results[1].Query)\n}\n\nfunc TestSearcher_Any(t *testing.T) {\n\tcfg := &types.Config{\n\t\tKB: &types.KBConfig{\n\t\t\tCollections: []string{\"docs\"},\n\t\t},\n\t\tDB: &types.DBConfig{\n\t\t\tModels: []string{\"product\"},\n\t\t},\n\t}\n\ts := New(cfg, nil)\n\n\treqs := []*types.Request{\n\t\t{\n\t\t\tType: types.SearchTypeKB,\n\t\t\tQuery: \"KB query\",\n\t\t\tSource: types.SourceAuto,\n\t\t\tCollections: []string{\"docs\"},\n\t\t},\n\t\t{\n\t\t\tType: types.SearchTypeDB,\n\t\t\tQuery: \"DB query\",\n\t\t\tSource: types.SourceAuto,\n\t\t\tModels: []string{\"product\"},\n\t\t},\n\t}\n\n\t// Test Any() - returns when first search has results (like Promise.any)\n\t// Note: With skeleton handlers returning empty results, this will wait for all\n\tresults, err := s.Any(nil, reqs)\n\tassert.NoError(t, err)\n\tassert.Equal(t, 2, len(results))\n}\n\nfunc TestSearcher_Race(t *testing.T) {\n\tcfg := &types.Config{\n\t\tKB: &types.KBConfig{\n\t\t\tCollections: []string{\"docs\"},\n\t\t},\n\t\tDB: &types.DBConfig{\n\t\t\tModels: []string{\"product\"},\n\t\t},\n\t}\n\ts := New(cfg, nil)\n\n\treqs := []*types.Request{\n\t\t{\n\t\t\tType: types.SearchTypeKB,\n\t\t\tQuery: \"KB query\",\n\t\t\tSource: types.SourceAuto,\n\t\t\tCollections: []string{\"docs\"},\n\t\t},\n\t\t{\n\t\t\tType: types.SearchTypeDB,\n\t\t\tQuery: \"DB query\",\n\t\t\tSource: types.SourceAuto,\n\t\t\tModels: []string{\"product\"},\n\t\t},\n\t}\n\n\t// Test Race() - returns when first search completes (like Promise.race)\n\tresults, err := s.Race(nil, reqs)\n\tassert.NoError(t, err)\n\t// At least one result should be set\n\thasResult := false\n\tfor _, r := range results {\n\t\tif r != nil {\n\t\t\thasResult = true\n\t\t\tbreak\n\t\t}\n\t}\n\tassert.True(t, hasResult)\n}\n\nfunc TestSearcher_All_Empty(t *testing.T) {\n\ts := New(nil, nil)\n\n\tresults, err := s.All(nil, []*types.Request{})\n\tassert.NoError(t, err)\n\tassert.Equal(t, 0, len(results))\n}\n\nfunc TestSearcher_Any_Empty(t *testing.T) {\n\ts := New(nil, nil)\n\n\tresults, err := s.Any(nil, []*types.Request{})\n\tassert.NoError(t, err)\n\tassert.Equal(t, 0, len(results))\n}\n\nfunc TestSearcher_Race_Empty(t *testing.T) {\n\ts := New(nil, nil)\n\n\tresults, err := s.Race(nil, []*types.Request{})\n\tassert.NoError(t, err)\n\tassert.Equal(t, 0, len(results))\n}\n\nfunc TestSearcher_All_ManyRequests(t *testing.T) {\n\tcfg := &types.Config{\n\t\tKB: &types.KBConfig{\n\t\t\tCollections: []string{\"docs\"},\n\t\t},\n\t}\n\ts := New(cfg, nil)\n\n\t// Create multiple requests to test parallel execution\n\treqs := make([]*types.Request, 10)\n\tfor i := 0; i < 10; i++ {\n\t\treqs[i] = &types.Request{\n\t\t\tType: types.SearchTypeKB,\n\t\t\tQuery: \"test query\",\n\t\t\tSource: types.SourceAuto,\n\t\t\tCollections: []string{\"docs\"},\n\t\t}\n\t}\n\n\tresults, err := s.All(nil, reqs)\n\tassert.NoError(t, err)\n\tassert.Equal(t, 10, len(results))\n\n\t// All results should be valid\n\tfor _, result := range results {\n\t\tassert.NotNil(t, result)\n\t\tassert.Equal(t, types.SearchTypeKB, result.Type)\n\t}\n}\n\nfunc TestSearcher_BuildReferences(t *testing.T) {\n\ts := New(nil, nil)\n\n\tresults := []*types.Result{\n\t\t{\n\t\t\tType: types.SearchTypeWeb,\n\t\t\tItems: []*types.ResultItem{\n\t\t\t\t{\n\t\t\t\t\tCitationID: \"1\",\n\t\t\t\t\tType: types.SearchTypeWeb,\n\t\t\t\t\tSource: types.SourceAuto,\n\t\t\t\t\tWeight: 0.6,\n\t\t\t\t\tTitle: \"Web Result\",\n\t\t\t\t\tContent: \"Web content\",\n\t\t\t\t\tURL: \"https://example.com\",\n\t\t\t\t},\n\t\t\t},\n\t\t},\n\t\t{\n\t\t\tType: types.SearchTypeKB,\n\t\t\tItems: []*types.ResultItem{\n\t\t\t\t{\n\t\t\t\t\tCitationID: \"2\",\n\t\t\t\t\tType: types.SearchTypeKB,\n\t\t\t\t\tSource: types.SourceHook,\n\t\t\t\t\tWeight: 0.8,\n\t\t\t\t\tTitle: \"KB Result\",\n\t\t\t\t\tContent: \"KB content\",\n\t\t\t\t},\n\t\t\t},\n\t\t},\n\t}\n\n\trefs := s.BuildReferences(results)\n\tassert.Equal(t, 2, len(refs))\n\tassert.Equal(t, \"1\", refs[0].ID)\n\tassert.Equal(t, \"2\", refs[1].ID)\n}\n\nfunc TestSearcher_CitationGeneration(t *testing.T) {\n\ts := New(nil, nil)\n\n\t// Reset citation generator for predictable IDs\n\ts.citation.Reset()\n\n\t// Note: This test would need actual results with items to verify citation generation\n\t// The skeleton handlers return empty items, so we test the citation generator directly\n\n\tid1 := s.citation.Next()\n\tid2 := s.citation.Next()\n\tid3 := s.citation.Next()\n\n\t// Citation IDs are now simple integers\n\tassert.Equal(t, \"1\", id1)\n\tassert.Equal(t, \"2\", id2)\n\tassert.Equal(t, \"3\", id3)\n}\n" }, { "path": "agent/search/search_web_test.go", "content": "package search_test\n\nimport (\n\t\"testing\"\n\n\t\"github.com/stretchr/testify/assert\"\n\t\"github.com/stretchr/testify/require\"\n\t\"github.com/yaoapp/yao/agent/assistant\"\n\t\"github.com/yaoapp/yao/agent/search\"\n\t\"github.com/yaoapp/yao/agent/search/types\"\n\t\"github.com/yaoapp/yao/agent/testutils\"\n)\n\n// =============================================================================\n// Web Search Integration Tests - Single Search\n// =============================================================================\n\n// TestWebSearch_Tavily tests web search using Tavily provider via assistant config\n// Skip: requires external API key (Tavily)\nfunc TestWebSearch_Tavily(t *testing.T) {\n\tt.Skip(\"Skipping: requires external API key (Tavily)\")\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\t// Load the web-tavily test assistant\n\tast, err := assistant.LoadPath(\"/assistants/tests/web-tavily\")\n\trequire.NoError(t, err)\n\trequire.NotNil(t, ast.Search)\n\trequire.NotNil(t, ast.Search.Web)\n\n\t// Verify assistant config\n\tassert.Equal(t, \"tavily\", ast.Search.Web.Provider)\n\n\t// Create Searcher with assistant's config\n\tuses := &search.Uses{Web: \"builtin\"}\n\ts := search.New(ast.Search, uses)\n\n\t// Execute search\n\treq := &types.Request{\n\t\tType: types.SearchTypeWeb,\n\t\tQuery: \"Yao App Engine low-code platform\",\n\t\tSource: types.SourceAuto,\n\t\tLimit: 5,\n\t}\n\n\tresult, err := s.Search(nil, req)\n\trequire.NoError(t, err)\n\trequire.NotNil(t, result)\n\trequire.Empty(t, result.Error, \"Search should succeed, got error: %s\", result.Error)\n\n\t// Verify results\n\tassert.NotEmpty(t, result.Items, \"Should have search results\")\n\tfor _, item := range result.Items {\n\t\tassert.NotEmpty(t, item.CitationID, \"Each item should have citation ID\")\n\t\tassert.NotEmpty(t, item.Content, \"Each item should have content\")\n\t\tt.Logf(\" [%s] %s - %s\", item.CitationID, item.Title, item.URL)\n\t}\n\tt.Logf(\"Tavily search returned %d results\", len(result.Items))\n}\n\n// TestWebSearch_Serper tests web search using Serper provider via assistant config\n// Skip: requires external API key (Serper)\nfunc TestWebSearch_Serper(t *testing.T) {\n\tt.Skip(\"Skipping: requires external API key (Serper)\")\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\t// Load the web-serper test assistant\n\tast, err := assistant.LoadPath(\"/assistants/tests/web-serper\")\n\trequire.NoError(t, err)\n\trequire.NotNil(t, ast.Search)\n\trequire.NotNil(t, ast.Search.Web)\n\n\t// Verify assistant config\n\tassert.Equal(t, \"serper\", ast.Search.Web.Provider)\n\n\t// Create Searcher with assistant's config\n\tuses := &search.Uses{Web: \"builtin\"}\n\ts := search.New(ast.Search, uses)\n\n\t// Execute search\n\treq := &types.Request{\n\t\tType: types.SearchTypeWeb,\n\t\tQuery: \"Go programming language concurrency\",\n\t\tSource: types.SourceAuto,\n\t\tLimit: 5,\n\t}\n\n\tresult, err := s.Search(nil, req)\n\trequire.NoError(t, err)\n\trequire.NotNil(t, result)\n\trequire.Empty(t, result.Error, \"Search should succeed, got error: %s\", result.Error)\n\n\t// Verify results\n\tassert.NotEmpty(t, result.Items, \"Should have search results\")\n\tfor _, item := range result.Items {\n\t\tassert.NotEmpty(t, item.CitationID, \"Each item should have citation ID\")\n\t\tt.Logf(\" [%s] %s - %s\", item.CitationID, item.Title, item.URL)\n\t}\n\tt.Logf(\"Serper search returned %d results\", len(result.Items))\n}\n\n// TestWebSearch_SerpAPI tests web search using SerpAPI provider via assistant config\n// Skip: requires external API key (SerpAPI)\nfunc TestWebSearch_SerpAPI(t *testing.T) {\n\tt.Skip(\"Skipping: requires external API key (SerpAPI)\")\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\t// Load the web-serpapi test assistant\n\tast, err := assistant.LoadPath(\"/assistants/tests/web-serpapi\")\n\trequire.NoError(t, err)\n\trequire.NotNil(t, ast.Search)\n\trequire.NotNil(t, ast.Search.Web)\n\n\t// Verify assistant config\n\tassert.Equal(t, \"serpapi\", ast.Search.Web.Provider)\n\n\t// Create Searcher with assistant's config\n\tuses := &search.Uses{Web: \"builtin\"}\n\ts := search.New(ast.Search, uses)\n\n\t// Execute search\n\treq := &types.Request{\n\t\tType: types.SearchTypeWeb,\n\t\tQuery: \"Kubernetes container orchestration\",\n\t\tSource: types.SourceAuto,\n\t\tLimit: 5,\n\t}\n\n\tresult, err := s.Search(nil, req)\n\trequire.NoError(t, err)\n\trequire.NotNil(t, result)\n\trequire.Empty(t, result.Error, \"Search should succeed, got error: %s\", result.Error)\n\n\t// Verify results\n\tassert.NotEmpty(t, result.Items, \"Should have search results\")\n\tfor _, item := range result.Items {\n\t\tassert.NotEmpty(t, item.CitationID, \"Each item should have citation ID\")\n\t\tt.Logf(\" [%s] %s - %s\", item.CitationID, item.Title, item.URL)\n\t}\n\tt.Logf(\"SerpAPI search returned %d results\", len(result.Items))\n}\n\n// =============================================================================\n// Web Search Integration Tests - Parallel Search\n// =============================================================================\n\n// TestWebSearch_All tests parallel web search with All() - like Promise.all\n// Skip: requires external API key (Serper)\nfunc TestWebSearch_All(t *testing.T) {\n\tt.Skip(\"Skipping: requires external API key (Serper)\")\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\t// Load the web-serper test assistant\n\tast, err := assistant.LoadPath(\"/assistants/tests/web-serper\")\n\trequire.NoError(t, err)\n\trequire.NotNil(t, ast.Search)\n\n\t// Create Searcher\n\tuses := &search.Uses{Web: \"builtin\"}\n\ts := search.New(ast.Search, uses)\n\n\t// Multiple queries\n\treqs := []*types.Request{\n\t\t{Type: types.SearchTypeWeb, Query: \"artificial intelligence\", Source: types.SourceAuto, Limit: 3},\n\t\t{Type: types.SearchTypeWeb, Query: \"machine learning\", Source: types.SourceAuto, Limit: 3},\n\t\t{Type: types.SearchTypeWeb, Query: \"deep learning\", Source: types.SourceAuto, Limit: 3},\n\t}\n\n\t// Execute parallel search with All() - waits for all searches to complete\n\tresults, err := s.All(nil, reqs)\n\trequire.NoError(t, err)\n\trequire.Len(t, results, 3, \"Should have 3 results\")\n\n\t// Verify all results\n\tfor i, result := range results {\n\t\trequire.NotNil(t, result, \"Result %d should not be nil\", i)\n\t\tif result.Error == \"\" {\n\t\t\tassert.NotEmpty(t, result.Items, \"Result %d should have items\", i)\n\t\t\tt.Logf(\"Query '%s': %d results\", reqs[i].Query, len(result.Items))\n\t\t} else {\n\t\t\tt.Logf(\"Query '%s': error - %s\", reqs[i].Query, result.Error)\n\t\t}\n\t}\n}\n\n// TestWebSearch_Any tests parallel web search with Any() - like Promise.any\n// Skip: requires external API key (Serper)\nfunc TestWebSearch_Any(t *testing.T) {\n\tt.Skip(\"Skipping: requires external API key (Serper)\")\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\t// Load the web-serper test assistant\n\tast, err := assistant.LoadPath(\"/assistants/tests/web-serper\")\n\trequire.NoError(t, err)\n\trequire.NotNil(t, ast.Search)\n\n\t// Create Searcher\n\tuses := &search.Uses{Web: \"builtin\"}\n\ts := search.New(ast.Search, uses)\n\n\t// Multiple queries\n\treqs := []*types.Request{\n\t\t{Type: types.SearchTypeWeb, Query: \"golang channels\", Source: types.SourceAuto, Limit: 3},\n\t\t{Type: types.SearchTypeWeb, Query: \"rust ownership\", Source: types.SourceAuto, Limit: 3},\n\t\t{Type: types.SearchTypeWeb, Query: \"python asyncio\", Source: types.SourceAuto, Limit: 3},\n\t}\n\n\t// Execute parallel search with Any() - returns when first search succeeds\n\tresults, err := s.Any(nil, reqs)\n\trequire.NoError(t, err)\n\n\t// Any() returns as soon as any search succeeds\n\thasSuccess := false\n\tfor _, result := range results {\n\t\tif result != nil && len(result.Items) > 0 && result.Error == \"\" {\n\t\t\thasSuccess = true\n\t\t\tt.Logf(\"First success: '%s' with %d results\", result.Query, len(result.Items))\n\t\t\tbreak\n\t\t}\n\t}\n\tassert.True(t, hasSuccess, \"At least one search should succeed\")\n}\n\n// TestWebSearch_Race tests parallel web search with Race() - like Promise.race\n// Skip: requires external API key (Serper)\nfunc TestWebSearch_Race(t *testing.T) {\n\tt.Skip(\"Skipping: requires external API key (Serper)\")\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\t// Load the web-serper test assistant\n\tast, err := assistant.LoadPath(\"/assistants/tests/web-serper\")\n\trequire.NoError(t, err)\n\trequire.NotNil(t, ast.Search)\n\n\t// Create Searcher\n\tuses := &search.Uses{Web: \"builtin\"}\n\ts := search.New(ast.Search, uses)\n\n\t// Multiple queries\n\treqs := []*types.Request{\n\t\t{Type: types.SearchTypeWeb, Query: \"docker containers\", Source: types.SourceAuto, Limit: 3},\n\t\t{Type: types.SearchTypeWeb, Query: \"kubernetes pods\", Source: types.SourceAuto, Limit: 3},\n\t}\n\n\t// Execute parallel search with Race() - returns when first search completes\n\tresults, err := s.Race(nil, reqs)\n\trequire.NoError(t, err)\n\n\t// Race() returns immediately when first result arrives\n\thasResult := false\n\tfor _, result := range results {\n\t\tif result != nil {\n\t\t\thasResult = true\n\t\t\tt.Logf(\"First to complete: '%s'\", result.Query)\n\t\t\tbreak\n\t\t}\n\t}\n\tassert.True(t, hasResult, \"Should have at least one result\")\n}\n\n// =============================================================================\n// Web Search - Citation and Reference Tests\n// =============================================================================\n\n// TestWebSearch_BuildReferences tests building references from web search results\n// Skip: requires external API key (Serper)\nfunc TestWebSearch_BuildReferences(t *testing.T) {\n\tt.Skip(\"Skipping: requires external API key (Serper)\")\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\t// Load the web-serper test assistant\n\tast, err := assistant.LoadPath(\"/assistants/tests/web-serper\")\n\trequire.NoError(t, err)\n\trequire.NotNil(t, ast.Search)\n\n\t// Create Searcher with weights config\n\tuses := &search.Uses{Web: \"builtin\"}\n\ts := search.New(ast.Search, uses)\n\n\t// Execute search\n\treq := &types.Request{\n\t\tType: types.SearchTypeWeb,\n\t\tQuery: \"OpenAI GPT-4\",\n\t\tSource: types.SourceAuto,\n\t\tLimit: 5,\n\t}\n\n\tresult, err := s.Search(nil, req)\n\trequire.NoError(t, err)\n\trequire.NotNil(t, result)\n\trequire.Empty(t, result.Error, \"Search should succeed\")\n\trequire.NotEmpty(t, result.Items, \"Should have results\")\n\n\t// Build references\n\trefs := s.BuildReferences([]*types.Result{result})\n\tassert.NotEmpty(t, refs, \"Should have references\")\n\n\tfor _, ref := range refs {\n\t\tassert.NotEmpty(t, ref.ID, \"Reference should have ID\")\n\t\tassert.Equal(t, types.SearchTypeWeb, ref.Type, \"Reference type should be web\")\n\t\tassert.Equal(t, types.SourceAuto, ref.Source, \"Reference source should be auto\")\n\t\tt.Logf(\" Ref: %s - %s (weight: %.2f)\", ref.ID, ref.Title, ref.Weight)\n\t}\n}\n\n// =============================================================================\n// Web Search - Error Handling Tests\n// =============================================================================\n\n// TestWebSearch_SiteRestriction tests web search with site restriction\n// Skip: requires external API key (Serper)\nfunc TestWebSearch_SiteRestriction(t *testing.T) {\n\tt.Skip(\"Skipping: requires external API key (Serper)\")\n\ttestutils.Prepare(t)\n\tdefer testutils.Clean(t)\n\n\t// Load the web-serper test assistant\n\tast, err := assistant.LoadPath(\"/assistants/tests/web-serper\")\n\trequire.NoError(t, err)\n\trequire.NotNil(t, ast.Search)\n\n\t// Create Searcher\n\tuses := &search.Uses{Web: \"builtin\"}\n\ts := search.New(ast.Search, uses)\n\n\t// Execute search with site restriction\n\treq := &types.Request{\n\t\tType: types.SearchTypeWeb,\n\t\tQuery: \"yao-app-engine\",\n\t\tSource: types.SourceAuto,\n\t\tSites: []string{\"github.com\"},\n\t\tLimit: 5,\n\t}\n\n\tresult, err := s.Search(nil, req)\n\trequire.NoError(t, err)\n\trequire.NotNil(t, result)\n\n\tif result.Error == \"\" && len(result.Items) > 0 {\n\t\t// Log results\n\t\tfor _, item := range result.Items {\n\t\t\tt.Logf(\" %s - %s\", item.Title, item.URL)\n\t\t}\n\t}\n}\n" }, { "path": "agent/search/types/config.go", "content": "package types\n\n// Config represents the complete search configuration\ntype Config struct {\n\tWeb *WebConfig `json:\"web,omitempty\" yaml:\"web,omitempty\"`\n\tKB *KBConfig `json:\"kb,omitempty\" yaml:\"kb,omitempty\"`\n\tDB *DBConfig `json:\"db,omitempty\" yaml:\"db,omitempty\"`\n\tKeyword *KeywordConfig `json:\"keyword,omitempty\" yaml:\"keyword,omitempty\"`\n\tQueryDSL *QueryDSLConfig `json:\"querydsl,omitempty\" yaml:\"querydsl,omitempty\"`\n\tRerank *RerankConfig `json:\"rerank,omitempty\" yaml:\"rerank,omitempty\"`\n\tCitation *CitationConfig `json:\"citation,omitempty\" yaml:\"citation,omitempty\"`\n\tWeights *WeightsConfig `json:\"weights,omitempty\" yaml:\"weights,omitempty\"`\n\tOptions *OptionsConfig `json:\"options,omitempty\" yaml:\"options,omitempty\"`\n}\n\n// WebConfig for web search settings\n// Note: uses.web determines the mode (builtin/agent/mcp)\n// Provider is only used when uses.web = \"builtin\"\ntype WebConfig struct {\n\tProvider string `json:\"provider,omitempty\" yaml:\"provider,omitempty\"` // \"tavily\", \"serper\", or \"serpapi\" (for builtin mode)\n\tAPIKeyEnv string `json:\"api_key_env,omitempty\" yaml:\"api_key_env,omitempty\"` // Environment variable for API key\n\tMaxResults int `json:\"max_results,omitempty\" yaml:\"max_results,omitempty\"` // Max results (default: 10)\n\tEngine string `json:\"engine,omitempty\" yaml:\"engine,omitempty\"` // Search engine for SerpAPI: \"google\", \"bing\", \"baidu\", \"yandex\", etc. (default: \"google\")\n}\n\n// KBConfig for knowledge base search settings\ntype KBConfig struct {\n\tCollections []string `json:\"collections,omitempty\" yaml:\"collections,omitempty\"` // Default collections\n\tThreshold float64 `json:\"threshold,omitempty\" yaml:\"threshold,omitempty\"` // Similarity threshold (default: 0.7)\n\tGraph bool `json:\"graph,omitempty\" yaml:\"graph,omitempty\"` // Enable GraphRAG (default: false)\n}\n\n// DBConfig for database search settings\ntype DBConfig struct {\n\tModels []string `json:\"models,omitempty\" yaml:\"models,omitempty\"` // Default models\n\tMaxResults int `json:\"max_results,omitempty\" yaml:\"max_results,omitempty\"` // Max results (default: 20)\n}\n\n// KeywordConfig for keyword extraction\ntype KeywordConfig struct {\n\tMaxKeywords int `json:\"max_keywords,omitempty\" yaml:\"max_keywords,omitempty\"` // Max keywords (default: 10)\n\tLanguage string `json:\"language,omitempty\" yaml:\"language,omitempty\"` // \"auto\", \"en\", \"zh\", etc.\n}\n\n// KeywordOptions for keyword extraction (runtime options)\ntype KeywordOptions struct {\n\tMaxKeywords int `json:\"max_keywords,omitempty\"`\n\tLanguage string `json:\"language,omitempty\"`\n}\n\n// QueryDSLConfig for QueryDSL generation from natural language\ntype QueryDSLConfig struct {\n\tStrict bool `json:\"strict,omitempty\" yaml:\"strict,omitempty\"` // Fail if generation fails (default: false)\n}\n\n// RerankConfig for reranking\ntype RerankConfig struct {\n\tTopN int `json:\"top_n,omitempty\" yaml:\"top_n,omitempty\"` // Return top N (default: 10)\n}\n\n// CitationConfig for citation format\ntype CitationConfig struct {\n\tFormat string `json:\"format,omitempty\" yaml:\"format,omitempty\"` // Default: \"#ref:{id}\"\n\tAutoInjectPrompt bool `json:\"auto_inject_prompt,omitempty\" yaml:\"auto_inject_prompt,omitempty\"` // Auto-inject prompt (default: true)\n\tCustomPrompt string `json:\"custom_prompt,omitempty\" yaml:\"custom_prompt,omitempty\"` // Custom prompt template\n}\n\n// WeightsConfig for source weighting\ntype WeightsConfig struct {\n\tUser float64 `json:\"user,omitempty\" yaml:\"user,omitempty\"` // User-provided (default: 1.0)\n\tHook float64 `json:\"hook,omitempty\" yaml:\"hook,omitempty\"` // Hook results (default: 0.8)\n\tAuto float64 `json:\"auto,omitempty\" yaml:\"auto,omitempty\"` // Auto search (default: 0.6)\n}\n\n// OptionsConfig for search behavior\ntype OptionsConfig struct {\n\tSkipThreshold int `json:\"skip_threshold,omitempty\" yaml:\"skip_threshold,omitempty\"` // Skip auto search if user provides >= N results\n}\n\n// GetWeight returns the weight for a source type\nfunc (c *Config) GetWeight(source SourceType) float64 {\n\tif c == nil || c.Weights == nil {\n\t\treturn getDefaultWeight(source)\n\t}\n\tswitch source {\n\tcase SourceUser:\n\t\tif c.Weights.User > 0 {\n\t\t\treturn c.Weights.User\n\t\t}\n\t\treturn 1.0\n\tcase SourceHook:\n\t\tif c.Weights.Hook > 0 {\n\t\t\treturn c.Weights.Hook\n\t\t}\n\t\treturn 0.8\n\tcase SourceAuto:\n\t\tif c.Weights.Auto > 0 {\n\t\t\treturn c.Weights.Auto\n\t\t}\n\t\treturn 0.6\n\tdefault:\n\t\treturn 0.6\n\t}\n}\n\n// getDefaultWeight returns default weight for a source type\nfunc getDefaultWeight(source SourceType) float64 {\n\tswitch source {\n\tcase SourceUser:\n\t\treturn 1.0\n\tcase SourceHook:\n\t\treturn 0.8\n\tcase SourceAuto:\n\t\treturn 0.6\n\tdefault:\n\t\treturn 0.6\n\t}\n}\n" }, { "path": "agent/search/types/graph.go", "content": "package types\n\n// GraphNode represents a related entity from knowledge graph\ntype GraphNode struct {\n\tID string `json:\"id\"`\n\tType string `json:\"type\"` // Entity type\n\tName string `json:\"name\"` // Entity name\n\tDescription string `json:\"description,omitempty\"` // Entity description\n\tRelation string `json:\"relation,omitempty\"` // Relationship to query\n\tScore float64 `json:\"score,omitempty\"` // Relevance score\n\tMetadata map[string]interface{} `json:\"metadata,omitempty\"`\n}\n" }, { "path": "agent/search/types/reference.go", "content": "package types\n\n// Reference is the unified structure for all data sources\n// Used to build LLM context from search results\ntype Reference struct {\n\tID string `json:\"id\"` // Unique citation ID: \"ref_001\", \"ref_002\"\n\tType SearchType `json:\"type\"` // Data type: \"web\", \"kb\", \"db\"\n\tSource SourceType `json:\"source\"` // Origin: \"user\", \"hook\", \"auto\"\n\tWeight float64 `json:\"weight\"` // Relevance weight (1.0=highest, 0.6=lowest)\n\tScore float64 `json:\"score\"` // Relevance score (0-1)\n\tTitle string `json:\"title\"` // Optional title\n\tContent string `json:\"content\"` // Main content\n\tURL string `json:\"url\"` // Optional URL\n\tMeta map[string]interface{} `json:\"meta\"` // Additional metadata\n}\n\n// ReferenceContext holds the formatted references for LLM input\ntype ReferenceContext struct {\n\tReferences []*Reference `json:\"references\"` // All references\n\tXML string `json:\"xml\"` // Formatted XML\n\tPrompt string `json:\"prompt\"` // Citation instruction prompt\n}\n" }, { "path": "agent/search/types/types.go", "content": "package types\n\nimport (\n\t\"github.com/yaoapp/gou/query/gou\"\n)\n\n// SearchType represents the type of search\ntype SearchType string\n\n// SearchType constants\nconst (\n\tSearchTypeWeb SearchType = \"web\" // Web/Internet search\n\tSearchTypeKB SearchType = \"kb\" // Knowledge base vector search\n\tSearchTypeDB SearchType = \"db\" // Database search (Yao Model/QueryDSL)\n)\n\n// ScenarioType represents the QueryDSL generation scenario\ntype ScenarioType string\n\n// ScenarioType constants for QueryDSL generation\nconst (\n\tScenarioFilter ScenarioType = \"filter\" // Simple filtering queries\n\tScenarioAggregation ScenarioType = \"aggregation\" // Aggregation/grouping queries\n\tScenarioJoin ScenarioType = \"join\" // Multi-table join queries\n\tScenarioComplex ScenarioType = \"complex\" // Complex queries combining multiple features\n)\n\n// SourceType represents where the search result came from\ntype SourceType string\n\n// SourceType constants\nconst (\n\tSourceUser SourceType = \"user\" // User-provided DataContent (highest priority)\n\tSourceHook SourceType = \"hook\" // Hook ctx.search.*() results\n\tSourceAuto SourceType = \"auto\" // Auto search results (lowest priority)\n)\n\n// Request represents a search request\ntype Request struct {\n\t// Common fields\n\tQuery string `json:\"query\"` // Search query (natural language)\n\tType SearchType `json:\"type\"` // Search type: \"web\", \"kb\", or \"db\"\n\tLimit int `json:\"limit,omitempty\"` // Max results (default: 10)\n\tSource SourceType `json:\"source\"` // Source of this request (user/hook/auto)\n\n\t// Web search specific\n\tSites []string `json:\"sites,omitempty\"` // Restrict to specific sites\n\tTimeRange string `json:\"time_range,omitempty\"` // \"day\", \"week\", \"month\", \"year\"\n\n\t// Knowledge base specific\n\tCollections []string `json:\"collections,omitempty\"` // KB collection IDs\n\tThreshold float64 `json:\"threshold,omitempty\"` // Similarity threshold (0-1)\n\tGraph bool `json:\"graph,omitempty\"` // Enable graph association\n\tMetadata map[string]interface{} `json:\"metadata,omitempty\"` // Metadata filter for KB search\n\n\t// Database search specific\n\tModels []string `json:\"models,omitempty\"` // Model IDs (e.g., \"user\", \"agents.mybot.product\")\n\tScenario ScenarioType `json:\"scenario,omitempty\"` // QueryDSL scenario: \"filter\", \"aggregation\", \"join\", \"complex\"\n\tWheres []gou.Where `json:\"wheres,omitempty\"` // Pre-defined filters (optional), uses GOU QueryDSL Where\n\tOrders gou.Orders `json:\"orders,omitempty\"` // Sort orders (optional), uses GOU QueryDSL Orders\n\tSelect []string `json:\"select,omitempty\"` // Fields to return (optional)\n\n\t// Reranking\n\tRerank *RerankOptions `json:\"rerank,omitempty\"`\n}\n\n// RerankOptions controls result reranking\n// Reranker type is determined by uses.rerank in agent/agent.yml\ntype RerankOptions struct {\n\tTopN int `json:\"top_n,omitempty\"` // Return top N after reranking\n}\n\n// Result represents the search result\ntype Result struct {\n\tType SearchType `json:\"type\"` // Search type\n\tQuery string `json:\"query\"` // Original query\n\tSource SourceType `json:\"source\"` // Source of this result\n\tItems []*ResultItem `json:\"items\"` // Result items\n\tTotal int `json:\"total\"` // Total matches\n\tDuration int64 `json:\"duration_ms\"` // Search duration in ms\n\tError string `json:\"error,omitempty\"` // Error message if failed\n\n\t// Graph associations (KB only, if enabled)\n\tGraphNodes []*GraphNode `json:\"graph_nodes,omitempty\"`\n\n\t// DB specific\n\tDSL map[string]interface{} `json:\"dsl,omitempty\"` // Generated QueryDSL (DB only)\n}\n\n// ResultItem represents a single search result item\ntype ResultItem struct {\n\t// Citation\n\tCitationID string `json:\"citation_id\"` // Unique ID for LLM reference: \"ref_001\"\n\n\t// Weighting\n\tSource SourceType `json:\"source\"` // Source type: \"user\", \"hook\", \"auto\"\n\tWeight float64 `json:\"weight\"` // Source weight (from config)\n\tScore float64 `json:\"score,omitempty\"` // Relevance score (0-1)\n\n\t// Common fields\n\tType SearchType `json:\"type\"` // Search type for this item\n\tTitle string `json:\"title,omitempty\"` // Title/headline\n\tContent string `json:\"content\"` // Main content/snippet\n\tURL string `json:\"url,omitempty\"` // Source URL\n\n\t// KB specific\n\tDocumentID string `json:\"document_id,omitempty\"` // Source document ID\n\tCollection string `json:\"collection,omitempty\"` // Collection name\n\n\t// DB specific\n\tModel string `json:\"model,omitempty\"` // Model ID\n\tRecordID interface{} `json:\"record_id,omitempty\"` // Record primary key\n\tData map[string]interface{} `json:\"data,omitempty\"` // Full record data\n\n\t// Metadata\n\tMetadata map[string]interface{} `json:\"metadata,omitempty\"` // Additional metadata\n}\n\n// ProcessedQuery represents a processed query ready for execution\ntype ProcessedQuery struct {\n\tType SearchType `json:\"type\"`\n\tKeywords []string `json:\"keywords,omitempty\"` // For web search\n\tVector []float32 `json:\"vector,omitempty\"` // For KB search\n\tDSL *gou.QueryDSL `json:\"dsl,omitempty\"` // For DB search, uses GOU QueryDSL\n}\n\n// Keyword represents an extracted keyword with weight\ntype Keyword struct {\n\tK string `json:\"k\"` // Keyword text\n\tW float64 `json:\"w\"` // Weight (0.1-1.0), higher means more relevant\n}\n\n// Note: For QueryDSL and Model types, use GOU types directly:\n// - github.com/yaoapp/gou/query/gou.QueryDSL\n// - github.com/yaoapp/gou/model.Model\n// - github.com/yaoapp/gou/model.Column\n" }, { "path": "agent/store/CHAT_STORAGE_DESIGN.md", "content": "# Chat Storage Design\n\nThis document describes the design for storing chat conversations, messages, and execution steps in the YAO Agent system.\n\n## Table of Contents\n\n- [Overview](#overview)\n- [Architecture](#architecture)\n- [Data Models](#data-models)\n- [Write Strategy](#write-strategy)\n- [API Interface](#api-interface)\n- [Usage Examples](#usage-examples)\n- [Related Documents](#related-documents)\n\n## Overview\n\nThe chat storage system is designed to:\n\n1. **Store user-visible messages** - All messages sent via `ctx.Send()`, including text, images, loading states, etc.\n2. **Support resume/retry** - Track execution steps to enable recovery from interruptions or failures\n3. **Efficient writes** - Batch message writes at request end\n\n### Design Goals\n\n| Goal | Solution |\n| ------------------------ | ------------------------------------------------ |\n| Complete chat history | Store final content of all `ctx.Send()` messages |\n| Resume from interruption | Track step status and input/output |\n| Retry failed operations | Store step input for re-execution |\n| Minimize database writes | Batch writes at request end |\n\n### Non-Goals\n\n- **Tracing/debugging** - Handled by separate [Trace module](../../trace/README.md)\n- **Streaming replay** - Not needed, history shows final content only\n- **Request tracking/billing** - Handled by [OpenAPI Request module](../../openapi/request/REQUEST_DESIGN.md)\n\n### Relationship with OpenAPI Request\n\nThe Agent storage focuses on **chat content and execution state**, while request tracking (billing, rate limiting, auditing) is handled globally by the OpenAPI layer:\n\n| Concern | Module | Table |\n| ---------------- | ----------------- | ----------------- |\n| Request tracking | `openapi/request` | `openapi_request` |\n| Billing (tokens) | `openapi/request` | `openapi_request` |\n| Rate limiting | `openapi/request` | - |\n| Chat sessions | `agent/store` | `agent_chat` |\n| Chat messages | `agent/store` | `agent_message` |\n| Resume/Retry | `agent/store` | `agent_resume` |\n\nThe `request_id` from OpenAPI middleware is passed to Agent and stored in messages/steps for correlation.\n\n## Architecture\n\n```\n┌─────────────────────────────────────────────────────────────┐\n│ Chat Storage │\n├─────────────────────────────────────────────────────────────┤\n│ │\n│ ┌─────────────────┐ │\n│ │ Chat │ Metadata: title, assistant, user │\n│ └────────┬────────┘ │\n│ │ │\n│ │ 1:N │\n│ ▼ │\n│ ┌─────────────────┐ │\n│ │ Message │ User-visible: type, props, role │\n│ └────────┬────────┘ │\n│ │ │\n│ │ N:N (via request_id) │\n│ ▼ │\n│ ┌─────────────────┐ │\n│ │ Resume │ Recovery: type, status, input/output │\n│ │ (only on fail) │ Only saved when interrupted/failed │\n│ └─────────────────┘ │\n│ │\n└─────────────────────────────────────────────────────────────┘\n```\n\n## Data Models\n\n### 1. Chat Table\n\nStores chat metadata and session information.\n\n**Table Name:** `agent_chat`\n\n| Column | Type | Nullable | Index | Description |\n| ----------------- | ----------- | -------- | ------ | -------------------------------- |\n| `id` | ID | No | PK | Auto-increment primary key |\n| `chat_id` | string(64) | No | Unique | Unique chat identifier |\n| `title` | string(500) | Yes | - | Chat title |\n| `assistant_id` | string(200) | No | Yes | Associated assistant ID |\n| `last_connector` | string(200) | Yes | Yes | Last used connector ID |\n| `last_mode` | string(50) | Yes | - | Last used chat mode (chat/task) |\n| `status` | enum | No | Yes | Status: `active`, `archived` |\n| `public` | boolean | No | - | Whether shared across all teams |\n| `share` | enum | No | Yes | Sharing scope: `private`, `team` |\n| `sort` | integer | No | - | Sort order for display |\n| `last_message_at` | timestamp | Yes | Yes | Timestamp of last message |\n| `metadata` | json | Yes | - | Additional metadata |\n| `created_at` | timestamp | No | Yes | Creation timestamp |\n| `updated_at` | timestamp | No | - | Last update timestamp |\n\n**Model Options:**\n\n```json\n{\n \"option\": {\n \"soft_deletes\": true,\n \"permission\": true,\n \"timestamps\": true\n }\n}\n```\n\n**Note:** `permission: true` enables Yao's built-in permission management, which automatically adds the following fields:\n\n| Field | Type | Description |\n| ------------------ | ----------- | ------------------------------ |\n| `__yao_created_by` | string(200) | User ID who created the record |\n| `__yao_updated_by` | string(200) | User ID who last updated |\n| `__yao_team_id` | string(200) | Team ID for team-level access |\n| `__yao_tenant_id` | string(200) | Tenant ID for multi-tenancy |\n\nThese fields are automatically managed by the framework and used for access control filtering.\n\n**Indexes:**\n\n| Name | Columns | Type |\n| -------------------- | ----------------- | ----- |\n| `idx_chat_assistant` | `assistant_id` | index |\n| `idx_chat_last_conn` | `last_connector` | index |\n| `idx_chat_status` | `status` | index |\n| `idx_chat_share` | `share` | index |\n| `idx_chat_last_msg` | `last_message_at` | index |\n\n### 2. Message Table\n\nStores user-visible messages (both user input and assistant responses).\n\n**Table Name:** `agent_message`\n\n| Column | Type | Nullable | Index | Description |\n| -------------- | ----------- | -------- | ----- | ------------------------------------------- |\n| `id` | ID | No | PK | Auto-increment primary key |\n| `message_id` | string(64) | No | - | Message identifier (unique within request) |\n| `chat_id` | string(64) | No | Yes | Parent chat ID |\n| `request_id` | string(64) | Yes | Yes | Request ID for grouping |\n| `role` | enum | No | Yes | Role: `user`, `assistant` |\n| `type` | string(50) | No | - | Message type (text, image, loading, etc.) |\n| `props` | json | No | - | Message properties (content, url, etc.) |\n| `block_id` | string(64) | Yes | Yes | Block grouping ID |\n| `thread_id` | string(64) | Yes | Yes | Thread grouping ID |\n| `assistant_id` | string(200) | Yes | Yes | Assistant ID (join to get name/avatar) |\n| `connector` | string(200) | Yes | Yes | Connector ID used for this message |\n| `mode` | string(50) | Yes | - | Chat mode used for this message (chat/task) |\n| `sequence` | integer | No | - | Message order within chat (in composite) |\n| `metadata` | json | Yes | - | Additional metadata |\n| `created_at` | timestamp | No | Yes | Creation timestamp |\n| `updated_at` | timestamp | No | - | Last update timestamp |\n\n**Indexes:**\n\n| Name | Columns | Type |\n| ------------------------- | -------------------------- | ------ |\n| `idx_msg_chat_seq` | `chat_id`, `sequence` | index |\n| `idx_msg_request_message` | `request_id`, `message_id` | unique |\n| `idx_msg_request` | `request_id` | index |\n| `idx_msg_role` | `role` | index |\n| `idx_msg_block` | `block_id` | index |\n| `idx_msg_thread` | `thread_id` | index |\n| `idx_msg_assistant` | `assistant_id` | index |\n\n**Message Ordering:**\n\nMessages are ordered by `created_at` first, then by `sequence` within the same timestamp. This ensures correct chronological order when there are multiple requests with overlapping sequence numbers:\n\n```sql\nORDER BY created_at ASC, sequence ASC\n```\n\n**Why this ordering?**\n\n- `sequence` is assigned per-request, so different requests may have the same sequence numbers\n- `created_at` groups messages by request time, ensuring messages from earlier requests appear first\n- Within the same request (same `created_at`), `sequence` preserves the internal ordering\n\n**Message Types:**\n\nAll message types are stored, including built-in types and custom types. See `agent/output/BUILTIN_TYPES.md` for built-in Props structures.\n\n| Type | Description | Props Example | Stored? |\n| ------------ | -------------------------------- | ------------------------------------------------------------------------------------------- | ------- |\n| `user_input` | User input (frontend display) | `{\"content\": \"Hello\", \"role\": \"user\", \"name\": \"John\"}` | ✅ Yes |\n| `text` | Text/Markdown content | `{\"content\": \"Hello **world**!\"}` | ✅ Yes |\n| `thinking` | Reasoning process (o1, DeepSeek) | `{\"content\": \"Let me analyze...\"}` | ✅ Yes |\n| `loading` | Loading/processing indicator | `{\"message\": \"Searching knowledge base...\"}` | ✅ Yes |\n| `tool_call` | LLM tool/function call | `{\"id\": \"call_abc123\", \"name\": \"get_weather\", \"arguments\": \"{\\\"location\\\":\\\"SF\\\"}\"}` | ✅ Yes |\n| `retrieval` | KB/Web search results | `{\"query\": \"...\", \"sources\": [...], \"total_results\": 10}` | ✅ Yes |\n| `error` | Error message | `{\"message\": \"Connection timeout\", \"code\": \"TIMEOUT\", \"details\": \"...\"}` | ✅ Yes |\n| `image` | Image content | `{\"url\": \"...\", \"alt\": \"...\", \"width\": 200, \"height\": 200, \"detail\": \"auto\"}` | ✅ Yes |\n| `audio` | Audio content | `{\"url\": \"...\", \"format\": \"mp3\", \"duration\": 120.5, \"transcript\": \"...\", \"controls\": true}` | ✅ Yes |\n| `video` | Video content | `{\"url\": \"...\", \"format\": \"mp4\", \"thumbnail\": \"...\", \"width\": 640, \"height\": 360}` | ✅ Yes |\n| `action` | System action (CUI only) | `{\"name\": \"open_panel\", \"payload\": {\"panel_id\": \"user_profile\"}}` | ✅ Yes |\n| `event` | Lifecycle event (CUI only) | `{\"event\": \"stream_start\", \"message\": \"...\", \"data\": {...}}` | ❌ No |\n| `*` (custom) | Any custom type | `{\"chartType\": \"bar\", \"data\": [...], \"options\": {...}}` | ✅ Yes |\n\n**Note on `event` type:** Lifecycle events (`stream_start`, `stream_end`, etc.) are transient control signals and are NOT stored. They are only used for real-time streaming coordination.\n\n**Note on custom types:** Any type not in the built-in list is stored as-is with its original `type` and `props` structure.\n\n**Tool Call Storage:**\n\nTool calls from LLM responses are stored as `tool_call` type messages. The raw tool call data is preserved in `props`:\n\n```json\n{\n \"message_id\": \"msg_001\",\n \"chat_id\": \"chat_123\",\n \"role\": \"assistant\",\n \"type\": \"tool_call\",\n \"props\": {\n \"id\": \"call_abc123\",\n \"name\": \"get_weather\",\n \"arguments\": \"{\\\"location\\\": \\\"San Francisco\\\", \\\"unit\\\": \\\"celsius\\\"}\"\n },\n \"block_id\": \"B1\",\n \"sequence\": 5\n}\n```\n\n**Tool Result Storage:**\n\nTool execution results can be stored as `text` type with metadata indicating it's a tool result:\n\n```json\n{\n \"message_id\": \"msg_002\",\n \"chat_id\": \"chat_123\",\n \"role\": \"assistant\",\n \"type\": \"text\",\n \"props\": {\n \"content\": \"The weather in San Francisco is 18°C and sunny.\"\n },\n \"metadata\": {\n \"tool_call_id\": \"call_abc123\",\n \"tool_name\": \"get_weather\",\n \"is_tool_result\": true\n },\n \"block_id\": \"B1\",\n \"sequence\": 6\n}\n```\n\n**Custom Types:**\n\nAny type not in the built-in list is considered a custom type and stored with its original structure:\n\n```json\n{\n \"type\": \"chart\",\n \"props\": {\n \"chartType\": \"bar\",\n \"data\": [...],\n \"options\": {...}\n }\n}\n```\n\n**Multimodal User Input:**\n\nUser input with multimodal content (text + images + files) is stored as `user_input` type:\n\n```json\n{\n \"message_id\": \"msg_000\",\n \"chat_id\": \"chat_123\",\n \"role\": \"user\",\n \"type\": \"user_input\",\n \"props\": {\n \"content\": [\n { \"type\": \"text\", \"text\": \"What's in this image?\" },\n {\n \"type\": \"image_url\",\n \"image_url\": {\n \"url\": \"https://example.com/photo.jpg\",\n \"detail\": \"high\"\n }\n }\n ],\n \"role\": \"user\",\n \"name\": \"John\"\n },\n \"sequence\": 1\n}\n```\n\n### Knowledge Base & Web Search Results\n\nRetrieval results from knowledge bases and web searches need to be stored for:\n\n1. **User Feedback** - Users can rate (👍/👎) individual sources\n2. **Quality Analytics** - Track which documents/sources are most useful\n3. **Source Attribution** - Display citations in the UI\n4. **RAG Optimization** - Improve retrieval based on feedback\n\n**Storage Approach:** Store retrieval results as a special message type `retrieval` with structured props.\n\n**Retrieval Message Structure:**\n\n```json\n{\n \"message_id\": \"msg_retrieval_001\",\n \"chat_id\": \"chat_123\",\n \"request_id\": \"req_abc\",\n \"role\": \"assistant\",\n \"type\": \"retrieval\",\n \"props\": {\n \"query\": \"How to configure Yao models?\",\n \"sources\": [\n {\n \"id\": \"src_001\",\n \"type\": \"kb\",\n \"collection_id\": \"col_docs\",\n \"document_id\": \"doc_123\",\n \"chunk_id\": \"chunk_456\",\n \"title\": \"Model Configuration Guide\",\n \"content\": \"To configure a model in Yao, create a .mod.yao file...\",\n \"score\": 0.92,\n \"metadata\": {\n \"file_path\": \"/docs/model.md\",\n \"page\": 3\n }\n },\n {\n \"id\": \"src_002\",\n \"type\": \"kb\",\n \"collection_id\": \"col_docs\",\n \"document_id\": \"doc_124\",\n \"chunk_id\": \"chunk_789\",\n \"title\": \"Advanced Model Options\",\n \"content\": \"Models support various options including soft_deletes...\",\n \"score\": 0.87,\n \"metadata\": {\n \"file_path\": \"/docs/advanced.md\",\n \"page\": 12\n }\n },\n {\n \"id\": \"src_003\",\n \"type\": \"web\",\n \"url\": \"https://yaoapps.com/docs/models\",\n \"title\": \"Yao Models Documentation\",\n \"content\": \"Official documentation for Yao model system...\",\n \"score\": 0.85,\n \"metadata\": {\n \"domain\": \"yaoapps.com\",\n \"fetched_at\": \"2024-01-15T10:30:00Z\"\n }\n }\n ],\n \"total_results\": 15,\n \"query_time_ms\": 120\n },\n \"block_id\": \"B1\",\n \"assistant_id\": \"docs_assistant\",\n \"sequence\": 2\n}\n```\n\n**Source Types:**\n\n| Type | Description | Key Fields |\n| ------ | ----------------------- | ------------------------------------------ |\n| `kb` | Knowledge base document | `collection_id`, `document_id`, `chunk_id` |\n| `web` | Web search result | `url`, `domain` |\n| `file` | Uploaded file | `file_id`, `file_path` |\n| `api` | External API result | `api_name`, `endpoint` |\n| `mcp` | MCP tool result | `server`, `tool` |\n\n**Source Feedback:**\n\nUser feedback on retrieval sources is handled by the Knowledge Base module. See [KB Feedback](../../kb/README.md) for details.\n\n**Example: KB Search in Create Hook:**\n\n```typescript\n// In Create hook, search knowledge base and store results\nconst results = await ctx.kb.search(\"col_docs\", query, { limit: 5 });\n\n// Send retrieval message (stored automatically)\nctx.Send({\n type: \"retrieval\",\n props: {\n query: query,\n sources: results.documents.map((doc, idx) => ({\n id: `src_${idx}`,\n type: \"kb\",\n collection_id: \"col_docs\",\n document_id: doc.document.metadata.document_id,\n chunk_id: doc.document.id,\n title: doc.document.metadata.title || \"Untitled\",\n content: doc.document.content,\n score: doc.score,\n metadata: doc.document.metadata,\n })),\n total_results: results.total,\n query_time_ms: results.query_time_ms,\n },\n});\n\n// Also send loading message for user feedback\nctx.Send({\n type: \"loading\",\n props: { message: `Found ${results.total} relevant documents...` },\n});\n```\n\n**Example: Web Search Results:**\n\n```json\n{\n \"type\": \"retrieval\",\n \"props\": {\n \"query\": \"latest AI news 2024\",\n \"sources\": [\n {\n \"id\": \"src_001\",\n \"type\": \"web\",\n \"url\": \"https://example.com/ai-news\",\n \"title\": \"AI Breakthroughs in 2024\",\n \"content\": \"Summary of the article...\",\n \"score\": 0.95,\n \"metadata\": {\n \"domain\": \"example.com\",\n \"published_at\": \"2024-01-10\",\n \"fetched_at\": \"2024-01-15T10:30:00Z\",\n \"snippet\": \"The year 2024 has seen remarkable...\"\n }\n }\n ],\n \"provider\": \"tavily\",\n \"total_results\": 10,\n \"query_time_ms\": 850\n }\n}\n```\n\n### 3. Resume Table\n\nStores execution state for resume/retry functionality. **Only written when request is interrupted or failed.**\n\n**Table Name:** `agent_resume`\n\n| Column | Type | Nullable | Index | Description |\n| ----------------- | ----------- | -------- | ------ | ---------------------------------------- |\n| `id` | ID | No | PK | Auto-increment primary key |\n| `resume_id` | string(64) | No | Unique | Unique resume record identifier |\n| `chat_id` | string(64) | No | Yes | Parent chat ID |\n| `request_id` | string(64) | No | Yes | Request ID |\n| `assistant_id` | string(200) | No | Yes | Assistant executing this step |\n| `stack_id` | string(64) | No | Yes | Stack node ID for this execution |\n| `stack_parent_id` | string(64) | Yes | Yes | Parent stack ID (for A2A calls) |\n| `stack_depth` | integer | No | - | Call depth (0=root, 1+=nested) |\n| `type` | enum | No | Yes | Step type |\n| `status` | enum | No | Yes | Status: `interrupted`, `failed` |\n| `input` | json | Yes | - | Step input data |\n| `output` | json | Yes | - | Step output data (partial) |\n| `space_snapshot` | json | Yes | - | Space data snapshot for recovery |\n| `error` | text | Yes | - | Error message if failed |\n| `sequence` | integer | No | - | Step order within request (in composite) |\n| `metadata` | json | Yes | - | Additional metadata |\n| `created_at` | timestamp | No | Yes | Creation timestamp |\n| `updated_at` | timestamp | No | - | Last update timestamp |\n\n**Space Snapshot:**\n\nThe `space_snapshot` field stores the shared data space (`ctx.Space`) at each step for recovery purposes.\n\n```typescript\n// Example: In Next hook, set data to Space before delegate\nctx.space.Set(\"choose_prompt\", \"query\");\nreturn {\n delegate: { agent_id: \"expense\", messages: payload.messages },\n};\n```\n\nIf interrupted during delegate, the `space_snapshot` allows restoring `ctx.Space` state:\n\n```json\n{\n \"choose_prompt\": \"query\",\n \"user_preferences\": { \"currency\": \"USD\" }\n}\n```\n\n**Resume Step Types:**\n\n| Type | Description | Input | Output |\n| ------------- | --------------------- | ---------------------- | ------------------------------------- |\n| `input` | User input received | `{messages: [...]}` | - |\n| `hook_create` | Create hook execution | `{messages: [...]}` | `{messages: [...], ...}` |\n| `llm` | LLM completion call | `{messages: [...]}` | `{content: \"...\", tool_calls: [...]}` |\n| `tool` | Tool/MCP execution | `{server, tool, args}` | `{result: ...}` |\n| `hook_next` | Next hook execution | `{completion, tools}` | `{data: ...}` |\n| `delegate` | A2A delegation | `{agent_id, messages}` | `{response: ...}` |\n\n**Resume Status (only two values - table only stores failed/interrupted):**\n\n| Status | Description | Action |\n| ------------- | ----------------- | -------- |\n| `failed` | Failed with error | Retry |\n| `interrupted` | User interrupted | Continue |\n\n**Indexes:**\n\n| Name | Columns | Type |\n| ---------------------- | ------------------------ | ----- |\n| `idx_resume_chat` | `chat_id` | index |\n| `idx_resume_request` | `request_id`, `sequence` | index |\n| `idx_resume_type` | `type` | index |\n| `idx_resume_status` | `status` | index |\n| `idx_resume_stack` | `stack_id` | index |\n| `idx_resume_parent` | `stack_parent_id` | index |\n| `idx_resume_assistant` | `assistant_id` | index |\n\n## Write Strategy\n\n### Single-Write Strategy\n\nAll data is buffered in memory during execution and written to database **only once** when `Stream()` exits:\n\n**Note**: Request tracking (status, tokens, duration) is handled by [OpenAPI Request Middleware](../../openapi/request/REQUEST_DESIGN.md).\n\n```\nStream() Entry\n │\n ├── Buffer user input message (role=user)\n │\n ├── Execution (all in memory)\n │ - ctx.Send() → messageBuffer\n │ - ctx.Append() → update messageBuffer\n │ - ctx.Replace() → update messageBuffer\n │ - Each step → stepBuffer\n │\n └── 【Single Write】Save final state (via defer)\n │\n ├── Always:\n │ - Batch write all messages (user input + assistant responses)\n │ - Update token usage in openapi_request (via request_id)\n │\n └── Only on error/interrupt:\n - Batch write all steps (for resume/retry)\n```\n\n### Write Points\n\n| Event | Message Table | Step Table | Token Usage |\n| ---------------- | -------------------------------------- | ----------------------------------- | ----------- |\n| Stream entry | Buffer user input | - | - |\n| During execution | Buffer in memory | Buffer in memory | - |\n| **Completed** | **Batch write all (user + assistant)** | **❌ Skip (no need to resume)** | ✅ Update |\n| On interrupt | Batch write all buffered | ✅ Batch write (status=interrupted) | ✅ Update |\n| On error | Batch write all buffered | ✅ Batch write (status=failed) | ✅ Update |\n\n**Why skip Steps on success?**\n\n- Steps are only needed for resume/retry operations\n- If completed successfully, there's nothing to resume\n- Reduces database writes and keeps Resume table clean\n\n### Why Single Write?\n\n| Scenario | What Happens | Data Safe? |\n| ------------------ | --------------------------------- | ---------- |\n| Normal completion | `defer` triggers → Write executes | ✅ |\n| User clicks stop | `defer` triggers → Write executes | ✅ |\n| LLM timeout | `defer` triggers → Write executes | ✅ |\n| Tool failure | `defer` triggers → Write executes | ✅ |\n| Network disconnect | `defer` triggers → Write executes | ✅ |\n| Process crash | Service is down, user must retry | N/A |\n\n**Note**: Process crash is a catastrophic failure handled at infrastructure level, not application level.\n\n### Write Count Comparison\n\nFor a typical request: user input → hook_create → llm → tool → hook_next → 5 messages\n\n| Strategy | Database Writes | Notes |\n| ------------------------- | --------------- | --------------------- |\n| Write per operation | 1 + 5 + 5 = 11 | One write per step |\n| **Single-write strategy** | **1** | Exit only (via defer) |\n\n### Implementation\n\n````go\nfunc (ast *Assistant) Stream(ctx, inputMessages, options) {\n // ========== Memory Buffers ==========\n messageBuffer := NewMessageBuffer()\n stepBuffer := NewStepBuffer()\n\n // Buffer user input message (not written yet)\n userMsg := createUserMessage(ctx, inputMessages)\n messageBuffer.Add(userMsg)\n\n // Track current step for error handling\n var currentStep *Step\n\n defer func() {\n // ========== Single Write: Exit (always executes) ==========\n // Determine final status for incomplete steps\n finalStatus := \"completed\"\n if ctx.IsInterrupted() {\n finalStatus = \"interrupted\"\n }\n if r := recover(); r != nil {\n finalStatus = \"failed\"\n }\n\n // Update status of any incomplete step\n if currentStep != nil && currentStep.Status == \"running\" {\n currentStep.Status = finalStatus\n }\n\n // Batch write all buffered messages (user input + assistant responses)\n chatStore.SaveMessages(ctx.ChatID, messageBuffer.GetAll())\n\n // Only save steps on error/interrupt (not on success)\n if finalStatus != \"completed\" {\n chatStore.SaveResume(stepBuffer.GetAll())\n }\n\n // Update token usage in OpenAPI request record\n if ctx.RequestID != \"\" && completionResponse != nil {\n request.UpdateTokenUsage(\n ctx.RequestID,\n completionResponse.Usage.PromptTokens,\n completionResponse.Usage.CompletionTokens,\n )\n }\n }()\n\n // ========== Execution (all in memory) ==========\n // Note: request_id = ctx.RequestID (from OpenAPI middleware)\n\n // hook_create\n currentStep = stepBuffer.Add(createStep(ctx, \"hook_create\", \"running\", input, nil))\n createResponse := ast.HookScript.Create(...)\n currentStep.Output = createResponse\n currentStep.Status = \"completed\"\n\n // llm\n currentStep = stepBuffer.Add(createStep(ctx, \"llm\", \"running\", messages, nil))\n completionResponse := ast.executeLLMStream(...)\n currentStep.Output = completionResponse\n currentStep.Status = \"completed\"\n\n // tool (if any)\n for _, toolCall := range completionResponse.ToolCalls {\n currentStep = stepBuffer.Add(createStep(ctx, \"tool\", \"running\", toolCall, nil))\n result := executeToolCall(toolCall)\n currentStep.Output = result\n currentStep.Status = \"completed\"\n }\n\n // hook_next\n currentStep = stepBuffer.Add(createStep(ctx, \"hook_next\", \"running\", payload, nil))\n nextResponse := ast.HookScript.Next(...)\n currentStep.Output = nextResponse\n currentStep.Status = \"completed\"\n currentStep = nil // All done\n\n // Messages are automatically buffered via ctx.Send()\n}\n\n// createResumeRecord creates a resume record with context information\n// Only called when request fails or is interrupted\nfunc createResumeRecord(ctx *Context, stepType, status string, input, output interface{}, err error) *Resume {\n // Capture Space snapshot for recovery\n var spaceSnapshot map[string]interface{}\n if ctx.Space != nil {\n spaceSnapshot = ctx.Space.Snapshot() // Get all key-value pairs\n }\n\n errorMsg := \"\"\n if err != nil {\n errorMsg = err.Error()\n }\n\n return &Resume{\n ResumeID: generateID(),\n ChatID: ctx.ChatID, // ChatID\n RequestID: ctx.RequestID, // From OpenAPI middleware\n AssistantID: ctx.AssistantID,\n StackID: ctx.Stack.ID,\n StackParentID: ctx.Stack.ParentID,\n StackDepth: ctx.Stack.Depth,\n Type: stepType,\n Status: status, // \"failed\" or \"interrupted\"\n Input: input,\n Output: output,\n SpaceSnapshot: spaceSnapshot, // Shared space data for recovery\n Error: errorMsg,\n Sequence: nextSequence(),\n }\n}\n\n## API Interface\n\n### ChatStore Interface\n\n```go\n// ChatStore defines the chat storage interface\n// Provides operations for chat, message, and resume management\ntype ChatStore interface {\n // ==========================================================================\n // Chat Management\n // ==========================================================================\n\n // CreateChat creates a new chat session\n CreateChat(chat *Chat) error\n\n // GetChat retrieves a single chat by ID\n GetChat(chatID string) (*Chat, error)\n\n // UpdateChat updates chat fields\n UpdateChat(chatID string, updates map[string]interface{}) error\n\n // DeleteChat deletes a chat and its associated messages\n DeleteChat(chatID string) error\n\n // ListChats retrieves a paginated list of chats with optional grouping\n ListChats(filter ChatFilter) (*ChatList, error)\n\n // ==========================================================================\n // Message Management\n // ==========================================================================\n\n // SaveMessages batch saves messages for a chat\n // This is the primary write method - messages are buffered during execution\n // and batch-written at the end of a request\n SaveMessages(chatID string, messages []*Message) error\n\n // GetMessages retrieves messages for a chat with filtering\n GetMessages(chatID string, filter MessageFilter) ([]*Message, error)\n\n // UpdateMessage updates a single message\n UpdateMessage(messageID string, updates map[string]interface{}) error\n\n // DeleteMessages deletes specific messages from a chat\n DeleteMessages(chatID string, messageIDs []string) error\n\n // ==========================================================================\n // Resume Management (only called on failure/interrupt)\n // ==========================================================================\n\n // SaveResume batch saves resume records\n // Only called when request is interrupted or failed\n SaveResume(records []*Resume) error\n\n // GetResume retrieves all resume records for a chat\n GetResume(chatID string) ([]*Resume, error)\n\n // GetLastResume retrieves the last (most recent) resume record for a chat\n GetLastResume(chatID string) (*Resume, error)\n\n // GetResumeByStackID retrieves resume records for a specific stack\n GetResumeByStackID(stackID string) ([]*Resume, error)\n\n // GetStackPath returns the stack path from root to the given stack\n // Returns: [root_stack_id, ..., current_stack_id]\n GetStackPath(stackID string) ([]string, error)\n\n // DeleteResume deletes all resume records for a chat\n // Called after successful resume to clean up\n DeleteResume(chatID string) error\n}\n\n// AssistantStore defines the assistant storage interface\n// Separated from ChatStore for clearer responsibility\ntype AssistantStore interface {\n // SaveAssistant saves assistant information\n SaveAssistant(assistant *AssistantModel) (string, error)\n\n // UpdateAssistant updates assistant fields\n UpdateAssistant(assistantID string, updates map[string]interface{}) error\n\n // DeleteAssistant deletes an assistant\n DeleteAssistant(assistantID string) error\n\n // GetAssistants retrieves a paginated list of assistants with filtering\n GetAssistants(filter AssistantFilter, locale ...string) (*AssistantList, error)\n\n // GetAssistantTags retrieves all unique tags from assistants with filtering\n GetAssistantTags(filter AssistantFilter, locale ...string) ([]Tag, error)\n\n // GetAssistant retrieves a single assistant by ID\n GetAssistant(assistantID string, fields []string, locale ...string) (*AssistantModel, error)\n\n // DeleteAssistants deletes assistants based on filter conditions\n DeleteAssistants(filter AssistantFilter) (int64, error)\n}\n\n// Store combines ChatStore and AssistantStore interfaces\n// This is the main interface for the storage layer\ntype Store interface {\n ChatStore\n AssistantStore\n}\n\n// SpaceStore defines the interface for Space snapshot operations\n// Note: Space itself uses plan.Space interface, this is for persistence\ntype SpaceStore interface {\n // Snapshot returns all key-value pairs in the space\n Snapshot() map[string]interface{}\n\n // Restore sets multiple key-value pairs from a snapshot\n Restore(data map[string]interface{}) error\n}\n```\n\n### Data Structures\n\n```go\n// Chat represents a chat session\ntype Chat struct {\n ChatID string `json:\"chat_id\"`\n Title string `json:\"title,omitempty\"`\n AssistantID string `json:\"assistant_id\"`\n LastConnector string `json:\"last_connector,omitempty\"` // Last used connector ID (updated on each message)\n LastMode string `json:\"last_mode,omitempty\"` // Last used chat mode (updated on each message)\n Status string `json:\"status\"` // \"active\" or \"archived\"\n Public bool `json:\"public\"` // Whether shared across all teams\n Share string `json:\"share\"` // \"private\" or \"team\"\n Sort int `json:\"sort\"` // Sort order for display\n LastMessageAt *time.Time `json:\"last_message_at,omitempty\"`\n Metadata map[string]interface{} `json:\"metadata,omitempty\"`\n CreatedAt time.Time `json:\"created_at\"`\n UpdatedAt time.Time `json:\"updated_at\"`\n}\n\n// Message represents a chat message\ntype Message struct {\n MessageID string `json:\"message_id\"`\n ChatID string `json:\"chat_id\"`\n RequestID string `json:\"request_id,omitempty\"`\n Role string `json:\"role\"` // \"user\" or \"assistant\"\n Type string `json:\"type\"` // \"text\", \"image\", \"loading\", \"tool_call\", \"retrieval\", etc.\n Props map[string]interface{} `json:\"props\"`\n BlockID string `json:\"block_id,omitempty\"`\n ThreadID string `json:\"thread_id,omitempty\"`\n AssistantID string `json:\"assistant_id,omitempty\"`\n Connector string `json:\"connector,omitempty\"` // Connector ID used for this message\n Mode string `json:\"mode,omitempty\"` // Chat mode used for this message (chat or task)\n Sequence int `json:\"sequence\"`\n Metadata map[string]interface{} `json:\"metadata,omitempty\"`\n CreatedAt time.Time `json:\"created_at\"`\n UpdatedAt time.Time `json:\"updated_at\"`\n}\n\n// Resume represents an execution state for recovery\n// Only stored when request is interrupted or failed\ntype Resume struct {\n ResumeID string `json:\"resume_id\"`\n ChatID string `json:\"chat_id\"`\n RequestID string `json:\"request_id\"`\n AssistantID string `json:\"assistant_id\"`\n StackID string `json:\"stack_id\"`\n StackParentID string `json:\"stack_parent_id,omitempty\"`\n StackDepth int `json:\"stack_depth\"`\n Type string `json:\"type\"` // \"input\", \"hook_create\", \"llm\", \"tool\", \"hook_next\", \"delegate\"\n Status string `json:\"status\"` // \"failed\" or \"interrupted\"\n Input map[string]interface{} `json:\"input,omitempty\"`\n Output map[string]interface{} `json:\"output,omitempty\"`\n SpaceSnapshot map[string]interface{} `json:\"space_snapshot,omitempty\"` // Shared space data for recovery\n Error string `json:\"error,omitempty\"`\n Sequence int `json:\"sequence\"`\n Metadata map[string]interface{} `json:\"metadata,omitempty\"`\n CreatedAt time.Time `json:\"created_at\"`\n UpdatedAt time.Time `json:\"updated_at\"`\n}\n\n// ResumeStatus constants\nconst (\n ResumeStatusFailed = \"failed\"\n ResumeStatusInterrupted = \"interrupted\"\n)\n\n// ResumeType constants\nconst (\n ResumeTypeInput = \"input\"\n ResumeTypeHookCreate = \"hook_create\"\n ResumeTypeLLM = \"llm\"\n ResumeTypeTool = \"tool\"\n ResumeTypeHookNext = \"hook_next\"\n ResumeTypeDelegate = \"delegate\"\n)\n```\n\n### Filter Structures\n\n```go\n// ChatFilter for listing chats\ntype ChatFilter struct {\n // Permission filters (direct filtering on Yao permission fields)\n UserID string `json:\"user_id,omitempty\"` // Filter by __yao_created_by\n TeamID string `json:\"team_id,omitempty\"` // Filter by __yao_team_id\n\n // Business filters\n AssistantID string `json:\"assistant_id,omitempty\"`\n Status string `json:\"status,omitempty\"`\n Keywords string `json:\"keywords,omitempty\"`\n\n // Time range filter\n StartTime *time.Time `json:\"start_time,omitempty\"` // Filter chats after this time\n EndTime *time.Time `json:\"end_time,omitempty\"` // Filter chats before this time\n TimeField string `json:\"time_field,omitempty\"` // Field for time filter: \"created_at\" or \"last_message_at\" (default)\n\n // Sorting\n OrderBy string `json:\"order_by,omitempty\"` // Field to sort by (default: \"last_message_at\")\n Order string `json:\"order,omitempty\"` // Sort order: \"desc\" (default) or \"asc\"\n\n // Response format\n GroupBy string `json:\"group_by,omitempty\"` // \"time\" for time-based groups, empty for flat list\n\n // Pagination\n Page int `json:\"page,omitempty\"`\n PageSize int `json:\"pagesize,omitempty\"`\n\n // Advanced permission filter (not serialized)\n // Use for complex conditions like: (created_by = user OR team_id = team)\n QueryFilter func(query.Query) `json:\"-\"`\n}\n\n// MessageFilter for listing messages\ntype MessageFilter struct {\n RequestID string `json:\"request_id,omitempty\"`\n Role string `json:\"role,omitempty\"`\n BlockID string `json:\"block_id,omitempty\"`\n ThreadID string `json:\"thread_id,omitempty\"`\n Type string `json:\"type,omitempty\"`\n Limit int `json:\"limit,omitempty\"`\n Offset int `json:\"offset,omitempty\"`\n}\n\n// ChatList paginated response with time-based grouping\ntype ChatList struct {\n Data []*Chat `json:\"data\"`\n Groups []*ChatGroup `json:\"groups,omitempty\"` // Time-based groups for UI display\n Page int `json:\"page\"`\n PageSize int `json:\"pagesize\"`\n PageCount int `json:\"pagecount\"`\n Total int `json:\"total\"`\n}\n\n// ChatGroup represents a time-based group of chats\ntype ChatGroup struct {\n Label string `json:\"label\"` // \"Today\", \"Yesterday\", \"This Week\", \"This Month\", \"Earlier\"\n Key string `json:\"key\"` // \"today\", \"yesterday\", \"this_week\", \"this_month\", \"earlier\"\n Chats []*Chat `json:\"chats\"` // Chats in this group\n Count int `json:\"count\"` // Number of chats in group\n}\n```\n\n## Usage Examples\n\n### 1. Complete Message Storage Example\n\nA typical conversation with various message types stored in `agent_message`:\n\n```\nUser: \"What's the weather in SF? Also show me a chart.\"\n\nTimeline (user input → hook_create → llm → tool → hook_next):\n1. User sends input\n2. Create hook shows loading state\n3. LLM thinks and calls tool\n4. Tool executes and returns result\n5. Next hook generates text response and image chart\n```\n\n**Stored Messages:**\n\n```json\n[\n // 1. User input (role=user, type=user_input)\n {\n \"message_id\": \"msg_001\",\n \"chat_id\": \"chat_123\",\n \"request_id\": \"req_abc\",\n \"role\": \"user\",\n \"type\": \"user_input\",\n \"props\": {\n \"content\": \"What's the weather in SF? Also show me a chart.\",\n \"role\": \"user\"\n },\n \"sequence\": 1\n },\n\n // 2. Loading state from Create hook (role=assistant, type=loading)\n {\n \"message_id\": \"msg_002\",\n \"chat_id\": \"chat_123\",\n \"request_id\": \"req_abc\",\n \"role\": \"assistant\",\n \"type\": \"loading\",\n \"props\": {\n \"message\": \"Searching knowledge base...\"\n },\n \"block_id\": \"B1\",\n \"assistant_id\": \"weather_assistant\",\n \"sequence\": 2\n },\n\n // 3. LLM thinking process (role=assistant, type=thinking)\n {\n \"message_id\": \"msg_003\",\n \"chat_id\": \"chat_123\",\n \"request_id\": \"req_abc\",\n \"role\": \"assistant\",\n \"type\": \"thinking\",\n \"props\": {\n \"content\": \"User wants weather info for San Francisco. I should use the get_weather tool...\"\n },\n \"block_id\": \"B2\",\n \"assistant_id\": \"weather_assistant\",\n \"sequence\": 3\n },\n\n // 4. LLM tool call (role=assistant, type=tool_call)\n {\n \"message_id\": \"msg_004\",\n \"chat_id\": \"chat_123\",\n \"request_id\": \"req_abc\",\n \"role\": \"assistant\",\n \"type\": \"tool_call\",\n \"props\": {\n \"id\": \"call_weather_001\",\n \"name\": \"get_weather\",\n \"arguments\": \"{\\\"location\\\": \\\"San Francisco\\\", \\\"unit\\\": \\\"celsius\\\"}\"\n },\n \"block_id\": \"B2\",\n \"assistant_id\": \"weather_assistant\",\n \"sequence\": 4\n },\n\n // 5. Tool result from Next hook (role=assistant, type=text, with tool metadata)\n {\n \"message_id\": \"msg_005\",\n \"chat_id\": \"chat_123\",\n \"request_id\": \"req_abc\",\n \"role\": \"assistant\",\n \"type\": \"text\",\n \"props\": {\n \"content\": \"The weather in San Francisco is currently **18°C** and sunny with 65% humidity. Perfect weather for outdoor activities!\"\n },\n \"block_id\": \"B3\",\n \"metadata\": {\n \"tool_call_id\": \"call_weather_001\",\n \"tool_name\": \"get_weather\"\n },\n \"assistant_id\": \"weather_assistant\",\n \"sequence\": 5\n },\n\n // 6. Chart image from Next hook (role=assistant, type=image)\n {\n \"message_id\": \"msg_006\",\n \"chat_id\": \"chat_123\",\n \"request_id\": \"req_abc\",\n \"role\": \"assistant\",\n \"type\": \"image\",\n \"props\": {\n \"url\": \"https://charts.example.com/weather_sf.png\",\n \"alt\": \"San Francisco 7-day weather forecast\",\n \"width\": 800,\n \"height\": 400\n },\n \"block_id\": \"B3\",\n \"assistant_id\": \"weather_assistant\",\n \"sequence\": 6\n }\n]\n```\n\n**Streaming IDs (from `STREAMING.md`):**\n\nDuring streaming, messages include additional fields for real-time delivery:\n\n| Field | Purpose | Stored? |\n| ------------ | ------------------------------ | ------- |\n| `chunk_id` | Deduplication, ordering, debug | ❌ No |\n| `message_id` | Delta merge target | ✅ Yes |\n| `block_id` | UI block/section grouping | ✅ Yes |\n| `thread_id` | Concurrent stream distinction | ✅ Yes |\n| `delta` | Whether this is a delta chunk | ❌ No |\n| `delta_path` | Path for delta merge | ❌ No |\n\n**Note:** `chunk_id`, `delta`, and `delta_path` are transient streaming control fields and are NOT stored. Only the final merged content is persisted.\n\n### 2. Error Message Storage\n\nWhen errors occur, they are stored as `error` type:\n\n```json\n{\n \"message_id\": \"msg_err_001\",\n \"chat_id\": \"chat_123\",\n \"request_id\": \"req_abc\",\n \"role\": \"assistant\",\n \"type\": \"error\",\n \"props\": {\n \"message\": \"Failed to connect to weather service\",\n \"code\": \"SERVICE_UNAVAILABLE\",\n \"details\": \"Connection timeout after 30 seconds\"\n },\n \"block_id\": \"B2\",\n \"assistant_id\": \"weather_assistant\",\n \"sequence\": 5\n}\n```\n\n### 3. Action Message Storage (CUI clients)\n\nSystem actions are stored but only processed by CUI clients:\n\n```json\n{\n \"message_id\": \"msg_action_001\",\n \"chat_id\": \"chat_123\",\n \"request_id\": \"req_abc\",\n \"role\": \"assistant\",\n \"type\": \"action\",\n \"props\": {\n \"name\": \"open_panel\",\n \"payload\": {\n \"panel_id\": \"weather_details\",\n \"location\": \"San Francisco\"\n }\n },\n \"block_id\": \"B2\",\n \"assistant_id\": \"weather_assistant\",\n \"sequence\": 6\n}\n```\n\n### 4. Audio/Video Message Storage\n\nMultimedia content storage:\n\n```json\n// Audio message\n{\n \"message_id\": \"msg_audio_001\",\n \"chat_id\": \"chat_123\",\n \"role\": \"assistant\",\n \"type\": \"audio\",\n \"props\": {\n \"url\": \"https://storage.example.com/audio/response.mp3\",\n \"format\": \"mp3\",\n \"duration\": 45.5,\n \"transcript\": \"Here's the weather forecast for today...\",\n \"controls\": true\n },\n \"sequence\": 7\n}\n\n// Video message\n{\n \"message_id\": \"msg_video_001\",\n \"chat_id\": \"chat_123\",\n \"role\": \"assistant\",\n \"type\": \"video\",\n \"props\": {\n \"url\": \"https://storage.example.com/video/weather_report.mp4\",\n \"format\": \"mp4\",\n \"thumbnail\": \"https://storage.example.com/video/weather_report_thumb.jpg\",\n \"duration\": 120.0,\n \"width\": 1280,\n \"height\": 720,\n \"controls\": true\n },\n \"sequence\": 8\n}\n```\n\n### 5. Load Chat History\n\n```go\n// Example 1: Filter by user (simple permission check)\nchats, _ := chatStore.ListChats(ChatFilter{\n UserID: \"user123\", // Filters by __yao_created_by\n Status: \"active\",\n OrderBy: \"last_message_at\",\n Order: \"desc\",\n Page: 1,\n PageSize: 20,\n})\n// Response: chats.Data = [...], chats.Groups = nil\n\n// Example 2: Filter by team\nchats, _ := chatStore.ListChats(ChatFilter{\n TeamID: \"team456\", // Filters by __yao_team_id\n Status: \"active\",\n Page: 1,\n PageSize: 20,\n})\n\n// Example 3: Filter by user AND team (both must match)\nchats, _ := chatStore.ListChats(ChatFilter{\n UserID: \"user123\",\n TeamID: \"team456\",\n Page: 1,\n PageSize: 20,\n})\n\n// Example 4: Complex permission filter (user OR team) using QueryFilter\nchats, _ := chatStore.ListChats(ChatFilter{\n Page: 1,\n PageSize: 20,\n QueryFilter: func(qb query.Query) {\n qb.Where(func(sub query.Query) {\n sub.Where(\"__yao_created_by\", \"user123\").\n OrWhere(\"__yao_team_id\", \"team456\")\n })\n },\n})\n\n// Example 5: Grouped by time\nchats, _ := chatStore.ListChats(ChatFilter{\n UserID: \"user123\",\n GroupBy: \"time\", // Enable time-based grouping\n OrderBy: \"last_message_at\",\n Order: \"desc\",\n Page: 1,\n PageSize: 20,\n})\n// Response includes time-based groups:\n// chats.Groups = [\n// { Key: \"today\", Label: \"Today\", Chats: [...], Count: 3 },\n// { Key: \"yesterday\", Label: \"Yesterday\", Chats: [...], Count: 5 },\n// { Key: \"this_week\", Label: \"This Week\", Chats: [...], Count: 8 },\n// { Key: \"this_month\", Label: \"This Month\", Chats: [...], Count: 4 },\n// { Key: \"earlier\", Label: \"Earlier\", Chats: [...], Count: 0 },\n// ]\n\n// Example 6: Filter by time range\nstartTime := time.Now().AddDate(0, 0, -7) // Last 7 days\nchats, _ := chatStore.ListChats(ChatFilter{\n UserID: \"user123\",\n StartTime: &startTime,\n TimeField: \"last_message_at\", // Filter by last message time\n OrderBy: \"last_message_at\",\n Order: \"desc\",\n})\n\n// Example 7: Filter specific date range\nstart := time.Date(2024, 12, 1, 0, 0, 0, 0, time.Local)\nend := time.Date(2024, 12, 31, 23, 59, 59, 0, time.Local)\nchats, _ := chatStore.ListChats(ChatFilter{\n UserID: \"user123\",\n StartTime: &start,\n EndTime: &end,\n TimeField: \"created_at\", // Filter by creation time\n})\n\n// Example 8: Combine permission with business filters\nchats, _ := chatStore.ListChats(ChatFilter{\n UserID: \"user123\",\n TeamID: \"team456\",\n AssistantID: \"weather_assistant\",\n Status: \"active\",\n Keywords: \"weather\",\n Page: 1,\n PageSize: 20,\n})\n\n// Get messages for a chat\nmessages, _ := chatStore.GetMessages(\"chat_123\", MessageFilter{\n Limit: 100,\n})\n\n// Return to frontend\nreturn map[string]interface{}{\n \"chat\": chat,\n \"messages\": messages,\n}\n```\n\n### 6. Resume from Interruption\n\n```go\nfunc (ast *Assistant) Resume(ctx *Context) error {\n // 1. Find last resume record\n record, _ := chatStore.GetLastResume(ctx.ChatID)\n if record == nil {\n return nil // Nothing to resume\n }\n\n // 2. Restore Space data from snapshot\n if record.SpaceSnapshot != nil && ctx.Space != nil {\n for key, value := range record.SpaceSnapshot {\n ctx.Space.Set(key, value)\n }\n }\n\n // 3. Check if this is an A2A nested call\n if record.StackDepth > 0 {\n // Need to rebuild the call stack\n return ast.ResumeNestedCall(ctx, record)\n }\n\n // 4. Resume based on step type\n var err error\n switch record.Type {\n case \"llm\":\n // Re-execute LLM call with saved input\n messages := record.Input[\"messages\"].([]Message)\n err = ast.executeLLMStream(ctx, messages, ...)\n\n case \"tool\":\n // Retry tool call\n err = ast.retryToolCall(ctx, record)\n\n case \"hook_next\":\n // Re-execute hook\n err = ast.executeHookNext(ctx, record.Input)\n\n case \"delegate\":\n // Resume delegated agent call\n agentID := record.Input[\"agent_id\"].(string)\n messages := record.Input[\"messages\"].([]Message)\n err = ast.delegateToAgent(ctx, agentID, messages)\n }\n\n // 5. Clean up resume records on success\n if err == nil {\n chatStore.DeleteResume(ctx.ChatID)\n }\n\n return err\n}\n```\n\n### 7. Resume A2A Nested Calls\n\nFor agent-to-agent (A2A) recursive calls, the stack information is essential for proper recovery.\n\n```go\nfunc (ast *Assistant) ResumeNestedCall(ctx *Context, step *Step) error {\n // 1. Rebuild the call stack from root to interrupted point\n stackPath, _ := chatStore.GetStackPath(step.StackID)\n // stackPath: [root_stack_id, parent_stack_id, ..., current_stack_id]\n\n // 2. Get all steps for each stack level\n for _, stackID := range stackPath {\n steps, _ := chatStore.GetStepsByStackID(stackID)\n // Restore context for each level\n }\n\n // 3. Resume from the interrupted assistant\n targetAssistant := assistant.Select(step.AssistantID)\n return targetAssistant.Stream(ctx, step.Input[\"messages\"], ...)\n}\n```\n\n### 8. Handle Interruption\n\nInterruption is handled automatically by the `defer` block in the two-write strategy. When `ctx.IsInterrupted()` returns true, the status is set to `interrupted` and all buffered data is saved.\n\n```go\n// Inside the defer block (see Write Strategy - Implementation)\nif ctx.IsInterrupted() {\n status = \"interrupted\"\n}\n// Then batch write all buffered messages and steps\n```\n\n## A2A (Agent-to-Agent) Call Example\n\nWhen Assistant A delegates to Assistant B, the step records look like:\n\n```\nRequest: User asks \"analyze this data and visualize it\"\n\nStep Records:\n┌─────┬─────────────┬─────────────┬──────────┬───────┬───────┬─────────────┬─────────────────────────────┐\n│ seq │ assistant │ stack_id │ parent │ depth │ type │ status │ space_snapshot │\n├─────┼─────────────┼─────────────┼──────────┼───────┼───────┼─────────────┼─────────────────────────────┤\n│ 1 │ analyzer │ stk_001 │ null │ 0 │ input │ completed │ {} │\n│ 2 │ analyzer │ stk_001 │ null │ 0 │ llm │ completed │ {} │\n│ 3 │ analyzer │ stk_001 │ null │ 0 │ delegate │ running │ {\"choose_prompt\": \"query\"} │ ← Space data set before delegate\n│ 4 │ visualizer │ stk_002 │ stk_001 │ 1 │ input │ completed │ {\"choose_prompt\": \"query\"} │\n│ 5 │ visualizer │ stk_002 │ stk_001 │ 1 │ llm │ interrupted │ {\"choose_prompt\": \"query\"} │ ← interrupted here\n└─────┴─────────────┴─────────────┴──────────┴───────┴───────┴─────────────┴─────────────────────────────┘\n\nResume Flow:\n1. Find step with status=\"interrupted\" → step 5\n2. Restore Space from space_snapshot: {\"choose_prompt\": \"query\"}\n3. Check stack_depth=1 → nested call\n4. Get stack path: [stk_001, stk_002]\n5. Resume visualizer assistant with step 5's input\n6. When visualizer completes, update step 3 (delegate) to completed\n```\n\n**Space Snapshot Use Case (from expense assistant):**\n\n```typescript\n// In Next hook, before delegating to another agent\nctx.space.Set(\"choose_prompt\", \"query\");\nreturn {\n delegate: { agent_id: \"expense\", messages: payload.messages },\n};\n\n// If interrupted during delegate, Resume will:\n// 1. Restore space_snapshot → ctx.space now has \"choose_prompt\": \"query\"\n// 2. The delegated agent's Create hook can read: ctx.space.GetDel(\"choose_prompt\")\n```\n\n## Concurrent Operations Storage\n\nWhen an Agent makes parallel calls (e.g., multiple MCP tools, multiple sub-agents), messages use `block_id` and `thread_id` for grouping:\n\n```\nMain Agent concurrently calls 3 tasks:\n├── Thread T1: Weather query (MCP)\n├── Thread T2: News search (MCP)\n├── Thread T3: Stock query (MCP)\n└── Wait for all to complete, then summarize\n```\n\n**Stored Messages:**\n\n```json\n[\n // All concurrent messages share the same block_id, different thread_id\n // Messages may arrive in any order due to concurrency\n\n // Thread T1: Weather result\n {\n \"message_id\": \"msg_t1_001\",\n \"chat_id\": \"chat_123\",\n \"request_id\": \"req_abc\",\n \"role\": \"assistant\",\n \"type\": \"text\",\n \"props\": { \"content\": \"Weather in SF: 18°C, sunny\" },\n \"block_id\": \"B1\",\n \"thread_id\": \"T1\",\n \"assistant_id\": \"main_assistant\",\n \"sequence\": 2\n },\n\n // Thread T2: News result\n {\n \"message_id\": \"msg_t2_001\",\n \"chat_id\": \"chat_123\",\n \"request_id\": \"req_abc\",\n \"role\": \"assistant\",\n \"type\": \"text\",\n \"props\": { \"content\": \"Top news: AI breakthrough announced...\" },\n \"block_id\": \"B1\",\n \"thread_id\": \"T2\",\n \"assistant_id\": \"main_assistant\",\n \"sequence\": 3\n },\n\n // Thread T3: Stock result\n {\n \"message_id\": \"msg_t3_001\",\n \"chat_id\": \"chat_123\",\n \"request_id\": \"req_abc\",\n \"role\": \"assistant\",\n \"type\": \"text\",\n \"props\": { \"content\": \"AAPL: $185.50 (+1.2%)\" },\n \"block_id\": \"B1\",\n \"thread_id\": \"T3\",\n \"assistant_id\": \"main_assistant\",\n \"sequence\": 4\n },\n\n // After all threads complete, main agent summarizes (new block)\n {\n \"message_id\": \"msg_summary\",\n \"chat_id\": \"chat_123\",\n \"request_id\": \"req_abc\",\n \"role\": \"assistant\",\n \"type\": \"text\",\n \"props\": {\n \"content\": \"Here's your daily briefing: The weather is great at 18°C...\"\n },\n \"block_id\": \"B2\",\n \"thread_id\": null,\n \"assistant_id\": \"main_assistant\",\n \"sequence\": 5\n }\n]\n```\n\n**Key Points:**\n\n| Field | Concurrent Usage |\n| ----------- | -------------------------------------------------- |\n| `block_id` | Same for all parallel operations (B1) |\n| `thread_id` | Different for each concurrent task (T1, T2, T3) |\n| `sequence` | Reflects actual arrival order (may be interleaved) |\n\n**Frontend Rendering:**\n\n- Group messages by `block_id` for visual blocks\n- Within a block, optionally group by `thread_id` to show parallel results\n- Use `sequence` for chronological display\n\n## HTTP API\n\nThe chat storage provides RESTful HTTP APIs for managing chat sessions and messages.\n\n**Base Path:** `/v1/chat`\n\n### Chat Sessions\n\n| Method | Endpoint | Description |\n|--------|----------|-------------|\n| `GET` | `/sessions` | List chat sessions with pagination and filtering |\n| `GET` | `/sessions/:chat_id` | Get a single chat session |\n| `PUT` | `/sessions/:chat_id` | Update chat session (title, status, metadata) |\n| `DELETE` | `/sessions/:chat_id` | Delete chat session |\n| `GET` | `/sessions/:chat_id/messages` | Get messages for a chat session |\n\n### List Chat Sessions\n\n**Request:**\n\n```\nGET /v1/chat/sessions?page=1&pagesize=20&assistant_id=xxx&status=active&keywords=search&group_by=time\n```\n\n**Query Parameters:**\n\n| Parameter | Type | Default | Description |\n|-----------|------|---------|-------------|\n| `page` | int | 1 | Page number |\n| `pagesize` | int | 20 | Items per page (max 100) |\n| `assistant_id` | string | - | Filter by assistant ID |\n| `status` | string | - | Filter by status: `active`, `archived` |\n| `keywords` | string | - | Search in title |\n| `start_time` | RFC3339 | - | Filter chats after this time |\n| `end_time` | RFC3339 | - | Filter chats before this time |\n| `time_field` | string | `last_message_at` | Field for time filter: `created_at` or `last_message_at` |\n| `order_by` | string | `last_message_at` | Sort field |\n| `order` | string | `desc` | Sort order: `asc` or `desc` |\n| `group_by` | string | - | Set to `time` for time-based grouping |\n\n**Response:**\n\n```json\n{\n \"data\": [\n {\n \"chat_id\": \"chat_123\",\n \"title\": \"Weather Query\",\n \"assistant_id\": \"weather_assistant\",\n \"status\": \"active\",\n \"last_message_at\": \"2024-01-15T10:30:00Z\",\n \"created_at\": \"2024-01-15T10:00:00Z\"\n }\n ],\n \"groups\": [\n {\n \"key\": \"today\",\n \"label\": \"Today\",\n \"chats\": [...],\n \"count\": 3\n },\n {\n \"key\": \"yesterday\",\n \"label\": \"Yesterday\",\n \"chats\": [...],\n \"count\": 5\n }\n ],\n \"page\": 1,\n \"pagesize\": 20,\n \"pagecount\": 5,\n \"total\": 100\n}\n```\n\n### Get Chat Session\n\n**Request:**\n\n```\nGET /v1/chat/sessions/chat_123\n```\n\n**Response:**\n\n```json\n{\n \"chat_id\": \"chat_123\",\n \"title\": \"Weather Query\",\n \"assistant_id\": \"weather_assistant\",\n \"last_connector\": \"deepseek.v3\",\n \"last_mode\": \"chat\",\n \"status\": \"active\",\n \"public\": false,\n \"share\": \"private\",\n \"last_message_at\": \"2024-01-15T10:30:00Z\",\n \"metadata\": {},\n \"created_at\": \"2024-01-15T10:00:00Z\",\n \"updated_at\": \"2024-01-15T10:30:00Z\"\n}\n```\n\n### Update Chat Session\n\n**Request:**\n\n```\nPUT /v1/chat/sessions/chat_123\nContent-Type: application/json\n\n{\n \"title\": \"New Title\",\n \"status\": \"archived\",\n \"metadata\": {\"custom_field\": \"value\"}\n}\n```\n\n**Response:**\n\n```json\n{\n \"message\": \"Chat updated successfully\",\n \"chat_id\": \"chat_123\"\n}\n```\n\n### Delete Chat Session\n\n**Request:**\n\n```\nDELETE /v1/chat/sessions/chat_123\n```\n\n**Response:**\n\n```json\n{\n \"message\": \"Chat deleted successfully\",\n \"chat_id\": \"chat_123\"\n}\n```\n\n### Get Chat Messages\n\n**Request:**\n\n```\nGET /v1/chat/sessions/chat_123/messages?limit=100&offset=0&role=assistant&type=text\n```\n\n**Query Parameters:**\n\n| Parameter | Type | Default | Description |\n|-----------|------|---------|-------------|\n| `request_id` | string | - | Filter by request ID |\n| `role` | string | - | Filter by role: `user`, `assistant` |\n| `block_id` | string | - | Filter by block ID |\n| `thread_id` | string | - | Filter by thread ID |\n| `type` | string | - | Filter by message type |\n| `limit` | int | 100 | Max messages to return (max 1000) |\n| `offset` | int | 0 | Offset for pagination |\n| `locale` | string | - | Locale for assistant info (e.g., `zh-cn`, `en-us`). Falls back to `Accept-Language` header |\n\n**Locale Resolution Priority:**\n1. Query parameter `locale`\n2. HTTP header `Accept-Language`\n\n**Response:**\n\n```json\n{\n \"chat_id\": \"chat_123\",\n \"messages\": [\n {\n \"message_id\": \"msg_001\",\n \"chat_id\": \"chat_123\",\n \"request_id\": \"req_abc\",\n \"role\": \"user\",\n \"type\": \"user_input\",\n \"props\": {\n \"content\": \"What's the weather?\",\n \"role\": \"user\"\n },\n \"sequence\": 1,\n \"created_at\": \"2024-01-15T10:00:00Z\"\n },\n {\n \"message_id\": \"msg_002\",\n \"chat_id\": \"chat_123\",\n \"request_id\": \"req_abc\",\n \"role\": \"assistant\",\n \"type\": \"text\",\n \"props\": {\n \"content\": \"The weather in San Francisco is 18°C and sunny.\"\n },\n \"block_id\": \"B1\",\n \"assistant_id\": \"weather_assistant\",\n \"sequence\": 2,\n \"created_at\": \"2024-01-15T10:00:05Z\"\n }\n ],\n \"count\": 2,\n \"assistants\": {\n \"weather_assistant\": {\n \"assistant_id\": \"weather_assistant\",\n \"name\": \"Weather Assistant\",\n \"avatar\": \"https://example.com/weather-avatar.png\",\n \"description\": \"Get weather information for any location\"\n }\n }\n}\n```\n\n**Note:** The `assistants` field contains localized assistant information (name, avatar, description) for all unique `assistant_id` values found in the messages. This allows the frontend to display assistant details without additional API calls. The locale is determined by the `locale` query parameter or `Accept-Language` header.\n\n### Permission Filtering\n\nAll endpoints respect Yao's permission system:\n\n| Constraint | Behavior |\n|------------|----------|\n| `OwnerOnly` | User can only access their own chats (`__yao_created_by` matches) |\n| `TeamOnly` | User can access own chats OR team-shared chats (`share = \"team\"`) |\n| No constraints | Full access (for admin users) |\n\n**Permission Fields Used:**\n\n- `__yao_created_by`: User who created the chat\n- `__yao_team_id`: Team ID for team-level access\n- `public`: Whether chat is public to all\n- `share`: Sharing scope (`private` or `team`)\n\n## Related Documents\n\n- [OpenAPI Request Design](../../openapi/request/REQUEST_DESIGN.md) - Global request tracking, billing, rate limiting\n- [Trace Module](../../trace/README.md) - Detailed execution tracing for debugging\n- [Agent Context](../context/README.md) - Context and message handling\n````\n" }, { "path": "agent/store/README.md", "content": "# YAO Agent Store\n\nYAO Agent Store is a comprehensive storage abstraction layer for managing conversations, assistants, attachments, and knowledge collections in the YAO Agent platform. It provides a unified interface that supports multiple storage backends including databases (via Xun), Redis, and MongoDB.\n\n## Table of Contents\n\n- [Architecture](#architecture)\n- [Storage Backends](#storage-backends)\n- [Configuration](#configuration)\n- [Initialization](#initialization)\n- [API Reference](#api-reference)\n- [Data Models](#data-models)\n- [Usage Examples](#usage-examples)\n- [Testing](#testing)\n\n## Architecture\n\nThe store package provides a unified `Store` interface that abstracts different storage implementations:\n\n```\n┌─────────────────┐\n│ Store API │ ← Unified Interface\n├─────────────────┤\n│ Xun (Database) │ ← Primary Implementation\n│ Redis │ ← Cache/Memory Store\n│ MongoDB │ ← Document Store\n└─────────────────┘\n```\n\n### Core Entities\n\n1. **Conversations & Chat History** - Manage chat sessions and message history\n2. **Assistants** - AI assistant configurations and metadata\n3. **Attachments** - File attachments with metadata and access control\n4. **Knowledge Collections** - Knowledge bases for AI assistants\n\n## Storage Backends\n\n### 1. Xun (Database) - Primary Backend\n\nThe main implementation using SQL databases with automatic schema management:\n\n- **Supported Databases**: MySQL, PostgreSQL, SQLite, etc.\n- **Features**: ACID transactions, complex queries, automatic migrations\n- **Use Case**: Production environments requiring data consistency\n\n### 2. Redis - Cache Backend\n\nRedis implementation for high-performance caching:\n\n- **Features**: In-memory storage, pub/sub capabilities\n- **Use Case**: Session management, temporary data, real-time features\n\n### 3. MongoDB - Document Backend\n\nMongoDB implementation for document-based storage:\n\n- **Features**: Schema flexibility, horizontal scaling\n- **Use Case**: Large-scale deployments, unstructured data\n\n## Configuration\n\n### Setting Structure\n\n```go\ntype Setting struct {\n Connector string `json:\"connector,omitempty\"` // Storage connector name\n UserField string `json:\"user_field,omitempty\"` // User ID field name (default: \"user_id\")\n Prefix string `json:\"prefix,omitempty\"` // Database table name prefix\n MaxSize int `json:\"max_size,omitempty\" yaml:\"max_size,omitempty\"` // Maximum history size limit\n TTL int `json:\"ttl,omitempty\" yaml:\"ttl,omitempty\"` // Time To Live in seconds\n}\n```\n\n### Configuration Examples\n\n#### Database Configuration\n\n```yaml\n# agent.yml\nagent:\n store:\n connector: \"mysql\" # or \"postgresql\", \"sqlite\", \"default\"\n prefix: \"agent_\" # Table prefix\n max_size: 100 # Maximum chat history size\n ttl: 7200 # 2 hours TTL for conversations\n user_field: \"user_id\" # User identification field\n```\n\n#### Redis Configuration\n\n```yaml\nagent:\n store:\n connector: \"redis\"\n prefix: \"agent:\"\n ttl: 3600\n```\n\n#### MongoDB Configuration\n\n```yaml\nagent:\n store:\n connector: \"mongodb\"\n prefix: \"agent_\"\n ttl: 7200\n```\n\n## Initialization\n\n### Automatic Initialization (Recommended)\n\nThe store is automatically initialized when the Agent system starts:\n\n```go\n// From yao/agent/load.go\nfunc initStore() error {\n var err error\n if Agent.StoreSetting.Connector == \"default\" || Agent.StoreSetting.Connector == \"\" {\n Agent.Store, err = store.NewXun(Agent.StoreSetting)\n return err\n }\n\n // Other connector types\n conn, err := connector.Select(Agent.StoreSetting.Connector)\n if err != nil {\n return err\n }\n\n if conn.Is(connector.DATABASE) {\n Agent.Store, err = store.NewXun(Agent.StoreSetting)\n return err\n } else if conn.Is(connector.REDIS) {\n Agent.Store = store.NewRedis()\n return nil\n } else if conn.Is(connector.MONGO) {\n Agent.Store = store.NewMongo()\n return nil\n }\n\n return fmt.Errorf(\"%s store connector %s not support\", Agent.ID, Agent.StoreSetting.Connector)\n}\n```\n\n### Manual Initialization\n\n```go\nimport \"github.com/yaoapp/yao/agent/store\"\n\n// Database backend\nsetting := store.Setting{\n Connector: \"mysql\",\n Prefix: \"agent_\",\n MaxSize: 100,\n TTL: 3600,\n}\nstore, err := store.NewXun(setting)\n\n// Redis backend\nredisStore := store.NewRedis()\n\n// MongoDB backend\nmongoStore := store.NewMongo()\n```\n\n## API Reference\n\n### Store Interface\n\n```go\ntype Store interface {\n // Chat Management\n GetChats(sid string, filter ChatFilter, locale ...string) (*ChatGroupResponse, error)\n GetChat(sid string, cid string, locale ...string) (*ChatInfo, error)\n GetChatWithFilter(sid string, cid string, filter ChatFilter, locale ...string) (*ChatInfo, error)\n UpdateChatTitle(sid string, cid string, title string) error\n DeleteChat(sid string, cid string) error\n DeleteAllChats(sid string) error\n\n // Message History\n GetHistory(sid string, cid string, locale ...string) ([]map[string]interface{}, error)\n GetHistoryWithFilter(sid string, cid string, filter ChatFilter, locale ...string) ([]map[string]interface{}, error)\n SaveHistory(sid string, messages []map[string]interface{}, cid string, context map[string]interface{}) error\n\n // Assistant Management\n SaveAssistant(assistant map[string]interface{}) (interface{}, error)\n GetAssistants(filter AssistantFilter, locale ...string) (*AssistantResponse, error)\n GetAssistant(assistantID string, locale ...string) (map[string]interface{}, error)\n DeleteAssistant(assistantID string) error\n DeleteAssistants(filter AssistantFilter) (int64, error)\n GetAssistantTags(locale ...string) ([]Tag, error)\n\n // Attachment Management\n SaveAttachment(attachment map[string]interface{}) (interface{}, error)\n GetAttachments(filter AttachmentFilter, locale ...string) (*AttachmentResponse, error)\n GetAttachment(fileID string, locale ...string) (map[string]interface{}, error)\n DeleteAttachment(fileID string) error\n DeleteAttachments(filter AttachmentFilter) (int64, error)\n\n // Knowledge Management\n SaveKnowledge(knowledge map[string]interface{}) (interface{}, error)\n GetKnowledges(filter KnowledgeFilter, locale ...string) (*KnowledgeResponse, error)\n GetKnowledge(collectionID string, locale ...string) (map[string]interface{}, error)\n DeleteKnowledge(collectionID string) error\n DeleteKnowledges(filter KnowledgeFilter) (int64, error)\n\n // Resource Management\n Close() error\n}\n```\n\n## Data Models\n\n### Database Schema\n\n#### 1. History Table (Conversations)\n\n```sql\nCREATE TABLE agent_history (\n id BIGINT PRIMARY KEY AUTO_INCREMENT,\n sid VARCHAR(255) INDEX, -- Session ID\n cid VARCHAR(200) INDEX, -- Chat ID\n uid VARCHAR(255) INDEX, -- User ID\n role VARCHAR(200) INDEX, -- Message role (user/assistant/system)\n name VARCHAR(200), -- Message sender name\n content TEXT, -- Message content\n context JSON, -- Message context\n assistant_id VARCHAR(200) INDEX, -- Associated assistant ID\n assistant_name VARCHAR(200), -- Assistant name\n assistant_avatar VARCHAR(200), -- Assistant avatar URL\n mentions JSON, -- Mentions in the message\n silent BOOLEAN DEFAULT FALSE INDEX, -- Silent message flag\n created_at TIMESTAMP DEFAULT CURRENT_TIMESTAMP INDEX,\n updated_at TIMESTAMP INDEX,\n expired_at TIMESTAMP INDEX -- TTL expiration\n);\n```\n\n#### 2. Chat Table\n\n```sql\nCREATE TABLE agent_chat (\n id BIGINT PRIMARY KEY AUTO_INCREMENT,\n chat_id VARCHAR(200) UNIQUE INDEX, -- Unique chat identifier\n title VARCHAR(200), -- Chat title\n assistant_id VARCHAR(200) INDEX, -- Associated assistant\n sid VARCHAR(255) INDEX, -- Session ID\n silent BOOLEAN DEFAULT FALSE INDEX, -- Silent chat flag\n created_at TIMESTAMP DEFAULT CURRENT_TIMESTAMP INDEX,\n updated_at TIMESTAMP INDEX\n);\n```\n\n#### 3. Assistant Table\n\n```sql\nCREATE TABLE agent_assistant (\n id BIGINT PRIMARY KEY AUTO_INCREMENT,\n assistant_id VARCHAR(200) UNIQUE INDEX, -- Unique assistant identifier\n type VARCHAR(200) DEFAULT 'assistant' INDEX, -- Assistant type\n name VARCHAR(200), -- Assistant name\n avatar VARCHAR(200), -- Avatar URL\n connector VARCHAR(200) NOT NULL, -- LLM connector\n description VARCHAR(600) INDEX, -- Description (searchable)\n path VARCHAR(200), -- Storage path\n sort INTEGER DEFAULT 9999 INDEX, -- Sort order\n built_in BOOLEAN DEFAULT FALSE INDEX, -- Built-in assistant flag\n placeholder JSON, -- UI placeholder text\n options JSON, -- Assistant options\n prompts JSON, -- System prompts\n workflow JSON, -- Workflow configuration\n knowledge JSON, -- Knowledge base references\n tools JSON, -- Available tools\n tags JSON, -- Assistant tags\n readonly BOOLEAN DEFAULT FALSE INDEX, -- Read-only flag\n permissions JSON, -- Access permissions\n locales JSON, -- Internationalization data\n automated BOOLEAN DEFAULT TRUE INDEX, -- Automation enabled\n mentionable BOOLEAN DEFAULT TRUE INDEX, -- Can be mentioned in chats\n created_at TIMESTAMP DEFAULT CURRENT_TIMESTAMP INDEX,\n updated_at TIMESTAMP INDEX\n);\n```\n\n#### 4. Attachment Table\n\n```sql\nCREATE TABLE agent_attachment (\n id BIGINT PRIMARY KEY AUTO_INCREMENT,\n file_id VARCHAR(255) UNIQUE INDEX, -- Unique file identifier\n uid VARCHAR(255) INDEX, -- Owner user ID\n guest BOOLEAN DEFAULT FALSE INDEX, -- Guest upload flag\n manager VARCHAR(200) INDEX, -- Storage manager\n content_type VARCHAR(200) INDEX, -- MIME type\n name VARCHAR(500) INDEX, -- File name (searchable)\n public BOOLEAN DEFAULT FALSE INDEX, -- Public access flag\n scope JSON, -- Access scope\n gzip BOOLEAN DEFAULT FALSE INDEX, -- Compression flag\n bytes BIGINT INDEX, -- File size\n collection_id VARCHAR(200) INDEX, -- Associated knowledge collection\n status ENUM('uploading', 'uploaded', 'indexing', 'indexed', 'upload_failed', 'index_failed') DEFAULT 'uploading' INDEX, -- Processing status\n progress VARCHAR(200), -- Progress information (nullable)\n error VARCHAR(600), -- Error message (nullable)\n created_at TIMESTAMP DEFAULT CURRENT_TIMESTAMP INDEX,\n updated_at TIMESTAMP INDEX\n);\n```\n\n#### 5. Knowledge Table\n\n```sql\nCREATE TABLE agent_knowledge (\n id BIGINT PRIMARY KEY AUTO_INCREMENT,\n collection_id VARCHAR(200) UNIQUE INDEX, -- Unique collection identifier\n name VARCHAR(200) INDEX, -- Collection name (searchable)\n description VARCHAR(600) INDEX, -- Description (searchable)\n uid VARCHAR(255) INDEX, -- Owner user ID\n public BOOLEAN DEFAULT FALSE INDEX, -- Public access flag\n scope JSON, -- Access scope\n readonly BOOLEAN DEFAULT FALSE INDEX, -- Read-only flag\n option JSON, -- Collection options\n system BOOLEAN DEFAULT FALSE INDEX, -- System collection flag\n sort INTEGER DEFAULT 9999 INDEX, -- Sort order\n cover VARCHAR(500), -- Cover image URL\n created_at TIMESTAMP DEFAULT CURRENT_TIMESTAMP INDEX,\n updated_at TIMESTAMP INDEX\n);\n```\n\n### Filter Structures\n\n#### ChatFilter\n\n```go\ntype ChatFilter struct {\n Keywords string `json:\"keywords,omitempty\"` // Search keywords\n Page int `json:\"page,omitempty\"` // Page number (starts from 1)\n PageSize int `json:\"pagesize,omitempty\"` // Items per page\n Order string `json:\"order,omitempty\"` // Sort order (desc/asc)\n Silent *bool `json:\"silent,omitempty\"` // Include silent messages\n}\n```\n\n#### AssistantFilter\n\n```go\ntype AssistantFilter struct {\n Tags []string `json:\"tags,omitempty\"` // Filter by tags\n Type string `json:\"type,omitempty\"` // Filter by type\n Keywords string `json:\"keywords,omitempty\"` // Search keywords\n Connector string `json:\"connector,omitempty\"` // Filter by connector\n AssistantID string `json:\"assistant_id,omitempty\"` // Specific assistant ID\n AssistantIDs []string `json:\"assistant_ids,omitempty\"` // Multiple assistant IDs\n Mentionable *bool `json:\"mentionable,omitempty\"` // Mentionable status\n Automated *bool `json:\"automated,omitempty\"` // Automation status\n BuiltIn *bool `json:\"built_in,omitempty\"` // Built-in status\n Page int `json:\"page,omitempty\"` // Page number\n PageSize int `json:\"pagesize,omitempty\"` // Items per page\n Select []string `json:\"select,omitempty\"` // Fields to return\n}\n```\n\n#### AttachmentFilter\n\n```go\ntype AttachmentFilter struct {\n UID string `json:\"uid,omitempty\"` // Filter by user ID\n Guest *bool `json:\"guest,omitempty\"` // Filter by guest status\n Manager string `json:\"manager,omitempty\"` // Filter by upload manager\n ContentType string `json:\"content_type,omitempty\"` // Filter by content type\n Name string `json:\"name,omitempty\"` // Filter by filename\n Public *bool `json:\"public,omitempty\"` // Filter by public status\n Gzip *bool `json:\"gzip,omitempty\"` // Filter by gzip compression\n CollectionID string `json:\"collection_id,omitempty\"` // Filter by knowledge collection ID\n Status string `json:\"status,omitempty\"` // Filter by processing status\n Keywords string `json:\"keywords,omitempty\"` // Search in filename\n Page int `json:\"page,omitempty\"` // Page number\n PageSize int `json:\"pagesize,omitempty\"` // Items per page\n Select []string `json:\"select,omitempty\"` // Fields to return\n}\n```\n\n#### KnowledgeFilter\n\n```go\ntype KnowledgeFilter struct {\n UID string `json:\"uid,omitempty\"` // Filter by user ID\n Name string `json:\"name,omitempty\"` // Filter by collection name\n Keywords string `json:\"keywords,omitempty\"` // Search in name and description\n Public *bool `json:\"public,omitempty\"` // Filter by public status\n Readonly *bool `json:\"readonly,omitempty\"` // Filter by readonly status\n System *bool `json:\"system,omitempty\"` // Filter by system status\n Page int `json:\"page,omitempty\"` // Page number\n PageSize int `json:\"pagesize,omitempty\"` // Items per page\n Select []string `json:\"select,omitempty\"` // Fields to return\n}\n```\n\n## Usage Examples\n\n### 1. Chat Management\n\n```go\n// Save chat history\nmessages := []map[string]interface{}{\n {\"role\": \"user\", \"content\": \"Hello, how are you?\"},\n {\"role\": \"assistant\", \"content\": \"I'm doing well, thank you!\"},\n}\ncontext := map[string]interface{}{\n \"assistant_id\": \"gpt-4\",\n \"silent\": false,\n}\nerr := store.SaveHistory(\"user123\", messages, \"chat456\", context)\n\n// Get chat history\nhistory, err := store.GetHistory(\"user123\", \"chat456\")\n\n// Get chat list with pagination\nfilter := ChatFilter{\n Page: 1,\n PageSize: 20,\n Order: \"desc\",\n}\nchats, err := store.GetChats(\"user123\", filter)\n\n// Update chat title\nerr = store.UpdateChatTitle(\"user123\", \"chat456\", \"New Chat Title\")\n```\n\n### 2. Assistant Management\n\n```go\n// Create an assistant\nassistant := map[string]interface{}{\n \"name\": \"Code Helper\",\n \"type\": \"assistant\",\n \"connector\": \"gpt-4\",\n \"description\": \"A helpful coding assistant\",\n \"tags\": []string{\"coding\", \"development\"},\n \"sort\": 100,\n \"options\": map[string]interface{}{\n \"temperature\": 0.7,\n \"max_tokens\": 2000,\n },\n \"prompts\": []string{\n \"You are a helpful coding assistant.\",\n },\n \"mentionable\": true,\n \"automated\": true,\n}\nassistantID, err := store.SaveAssistant(assistant)\n\n// Get assistants with filtering\nfilter := AssistantFilter{\n Tags: []string{\"coding\"},\n Keywords: \"helper\",\n Page: 1,\n PageSize: 10,\n}\nassistants, err := store.GetAssistants(filter)\n\n// Get specific assistant\nassistant, err := store.GetAssistant(\"assistant123\")\n```\n\n### 3. Attachment Management\n\n```go\n// Save attachment metadata\nattachment := map[string]interface{}{\n \"file_id\": \"file123\",\n \"uid\": \"user123\",\n \"manager\": \"local\",\n \"content_type\": \"image/jpeg\",\n \"name\": \"profile.jpg\",\n \"public\": false,\n \"bytes\": 102400,\n \"collection_id\": \"knowledge456\",\n \"scope\": []string{\"user\", \"admin\"},\n \"status\": \"uploaded\", // Status: uploading, uploaded, indexing, indexed, upload_failed, index_failed\n \"progress\": \"Upload completed\", // Progress information (optional)\n \"error\": nil, // Error message (optional, for failed statuses)\n}\nfileID, err := store.SaveAttachment(attachment)\n\n// Update attachment status during processing workflow\nattachment[\"status\"] = \"indexing\"\nattachment[\"progress\"] = \"Processing file for indexing...\"\n_, err = store.SaveAttachment(attachment)\n\n// Handle failed upload\nattachment[\"status\"] = \"upload_failed\"\nattachment[\"progress\"] = nil\nattachment[\"error\"] = \"Network connection timeout\"\n_, err = store.SaveAttachment(attachment)\n\n// Complete indexing\nattachment[\"status\"] = \"indexed\"\nattachment[\"progress\"] = \"File indexed successfully\"\nattachment[\"error\"] = nil\n_, err = store.SaveAttachment(attachment)\n\n// Get attachments with filtering\nfilter := AttachmentFilter{\n UID: \"user123\",\n ContentType: \"image/jpeg\",\n Status: \"indexed\", // Filter by status\n Page: 1,\n PageSize: 20,\n}\nattachments, err := store.GetAttachments(filter)\n\n// Get all failed uploads\nfailedFilter := AttachmentFilter{\n UID: \"user123\",\n Status: \"upload_failed\",\n Page: 1,\n PageSize: 10,\n}\nfailedUploads, err := store.GetAttachments(failedFilter)\n```\n\n#### Attachment Status Workflow\n\nThe attachment system supports a complete file processing workflow with the following status values:\n\n- **`uploading`** (default): File upload is in progress\n- **`uploaded`**: File upload completed successfully\n- **`indexing`**: File is being processed for search indexing\n- **`indexed`**: File has been indexed and is ready for use\n- **`upload_failed`**: File upload failed (check `error` field for details)\n- **`index_failed`**: File indexing failed (check `error` field for details)\n\n#### Additional Fields\n\n- **`progress`**: Human-readable progress information (string, nullable)\n- **`error`**: Error message for failed operations (string, nullable, max 600 characters)\n\n### 4. Knowledge Collection Management\n\n```go\n// Create knowledge collection\nknowledge := map[string]interface{}{\n \"collection_id\": \"kb123\",\n \"name\": \"Programming Guide\",\n \"description\": \"Comprehensive programming tutorials and examples\",\n \"uid\": \"user123\",\n \"public\": true,\n \"readonly\": false,\n \"sort\": 100,\n \"option\": map[string]interface{}{\n \"embedding\": \"openai\",\n \"chunk_size\": 1000,\n },\n \"scope\": []string{\"developers\", \"students\"},\n}\ncollectionID, err := store.SaveKnowledge(knowledge)\n\n// Get knowledge collections with filtering\nfilter := KnowledgeFilter{\n UID: \"user123\",\n Keywords: \"programming\",\n Public: &[]bool{true}[0],\n Page: 1,\n PageSize: 10,\n}\ncollections, err := store.GetKnowledges(filter)\n\n// Get system knowledge collections\nsystemFilter := KnowledgeFilter{\n System: &[]bool{true}[0],\n Page: 1,\n PageSize: 20,\n}\nsystemCollections, err := store.GetKnowledges(systemFilter)\n\n// Get readonly knowledge collections with specific fields\nreadonlyFilter := KnowledgeFilter{\n Readonly: &[]bool{true}[0],\n Select: []string{\"collection_id\", \"name\", \"description\", \"sort\"},\n Page: 1,\n PageSize: 15,\n}\nreadonlyCollections, err := store.GetKnowledges(readonlyFilter)\n```\n\n### 5. Internationalization Support\n\n```go\n// Get assistants with locale\nassistants, err := store.GetAssistants(filter, \"zh-CN\")\n\n// Get chat with locale\nchat, err := store.GetChat(\"user123\", \"chat456\", \"en-US\")\n```\n\n### 6. Advanced Filtering and Sorting\n\n```go\n// Complex assistant filtering\nfilter := AssistantFilter{\n Tags: []string{\"ai\", \"assistant\"},\n Keywords: \"helpful\",\n Connector: \"gpt-4\",\n Mentionable: &[]bool{true}[0],\n BuiltIn: &[]bool{false}[0],\n Select: []string{\"assistant_id\", \"name\", \"description\", \"tags\"},\n Page: 1,\n PageSize: 50,\n}\nassistants, err := store.GetAssistants(filter)\n\n// Results are automatically sorted by:\n// 1. sort field (ASC) - lower numbers appear first\n// 2. created_at/updated_at (DESC) - newer items appear first\n```\n\n## Testing\n\n### Running Tests\n\n```bash\n# Run all tests\ngo test -v\n\n# Run specific test\ngo test -run TestXunKnowledgeCRUD -v\n\n# Run with coverage\ngo test -cover\n```\n\n### Test Structure\n\nThe test suite includes comprehensive coverage for:\n\n- **CRUD Operations**: Create, Read, Update, Delete for all entities\n- **Filtering**: Various filter combinations and edge cases\n- **Sorting**: Verify sort order and pagination\n- **Error Handling**: Invalid inputs and edge cases\n- **Internationalization**: Locale-specific operations\n- **Concurrency**: Multiple concurrent operations\n\n### Test Database Setup\n\nTests use isolated table prefixes to avoid conflicts:\n\n```go\nstore, err := NewXun(Setting{\n Connector: \"default\",\n Prefix: \"__unit_test_conversation_\",\n TTL: 3600,\n})\n```\n\n## Performance Considerations\n\n### Database Optimization\n\n1. **Indexes**: All frequently queried fields have indexes\n2. **TTL**: Automatic cleanup of expired data\n3. **Pagination**: All list operations support pagination\n4. **Connection Pooling**: Efficient database connection management\n\n### Caching Strategy\n\n1. **Redis Backend**: For high-frequency read operations\n2. **Memory Caching**: In-application caching for static data\n3. **Query Optimization**: Efficient filtering and sorting\n\n### Scaling\n\n1. **Horizontal Scaling**: MongoDB support for distributed deployments\n2. **Read Replicas**: Database read/write splitting\n3. **Sharding**: Data partitioning strategies\n\n## Migration and Upgrades\n\n### Schema Evolution\n\nThe Xun backend automatically handles schema migrations:\n\n- New tables are created automatically\n- New fields are added with default values\n- Indexes are created during initialization\n\n### Data Migration\n\nWhen switching between backends:\n\n1. Export data from source backend\n2. Transform data format if necessary\n3. Import to target backend\n4. Verify data integrity\n\n## Security\n\n### Access Control\n\n1. **User Isolation**: All operations are user-scoped\n2. **Permission System**: Fine-grained access control\n3. **Public/Private Flags**: Content visibility management\n\n### Data Protection\n\n1. **Input Validation**: All inputs are validated and sanitized\n2. **SQL Injection Prevention**: Parameterized queries\n3. **XSS Protection**: Content encoding and sanitization\n\n## Troubleshooting\n\n### Common Issues\n\n1. **Connection Errors**: Check connector configuration\n2. **Schema Errors**: Verify database permissions\n3. **Performance Issues**: Check indexes and query patterns\n4. **Memory Issues**: Monitor TTL and cleanup processes\n\n### Debugging\n\nEnable debug logging:\n\n```go\nimport \"github.com/yaoapp/kun/log\"\n\nlog.SetLevel(log.DebugLevel)\n```\n\n### Monitoring\n\nKey metrics to monitor:\n\n- Database connection pool usage\n- Query performance and slow queries\n- Memory usage and garbage collection\n- TTL cleanup effectiveness\n\n## Contributing\n\n### Development Setup\n\n1. Clone the repository\n2. Install dependencies: `go mod download`\n3. Run tests: `go test -v`\n4. Follow Go coding standards\n\n### Adding New Features\n\n1. Update the Store interface\n2. Implement in all backends (Xun, Redis, MongoDB)\n3. Add comprehensive tests\n4. Update documentation\n\n## License\n\nThis project is part of the Yao App Engine and follows the same license terms.\n" }, { "path": "agent/store/mongo/mongo.go", "content": "package mongo\n\nimport \"github.com/yaoapp/yao/agent/store/types\"\n\n// Mongo represents a MongoDB-based conversation storage\ntype Mongo struct{}\n\n// NewMongo create a new mongo store\nfunc NewMongo() types.Store {\n\treturn &Mongo{}\n}\n\n// =============================================================================\n// Chat Management\n// =============================================================================\n\n// CreateChat creates a new chat session\nfunc (m *Mongo) CreateChat(chat *types.Chat) error {\n\t// TODO: implement\n\treturn nil\n}\n\n// GetChat retrieves a single chat by ID\nfunc (m *Mongo) GetChat(chatID string) (*types.Chat, error) {\n\t// TODO: implement\n\treturn nil, nil\n}\n\n// UpdateChat updates chat fields\nfunc (m *Mongo) UpdateChat(chatID string, updates map[string]interface{}) error {\n\t// TODO: implement\n\treturn nil\n}\n\n// DeleteChat deletes a chat and its associated messages\nfunc (m *Mongo) DeleteChat(chatID string) error {\n\t// TODO: implement\n\treturn nil\n}\n\n// ListChats retrieves a paginated list of chats with optional grouping\nfunc (m *Mongo) ListChats(filter types.ChatFilter) (*types.ChatList, error) {\n\t// TODO: implement\n\treturn nil, nil\n}\n\n// =============================================================================\n// Message Management\n// =============================================================================\n\n// SaveMessages batch saves messages for a chat\nfunc (m *Mongo) SaveMessages(chatID string, messages []*types.Message) error {\n\t// TODO: implement\n\treturn nil\n}\n\n// GetMessages retrieves messages for a chat with filtering\nfunc (m *Mongo) GetMessages(chatID string, filter types.MessageFilter) ([]*types.Message, error) {\n\t// TODO: implement\n\treturn nil, nil\n}\n\n// UpdateMessage updates a single message\nfunc (m *Mongo) UpdateMessage(messageID string, updates map[string]interface{}) error {\n\t// TODO: implement\n\treturn nil\n}\n\n// DeleteMessages deletes specific messages from a chat\nfunc (m *Mongo) DeleteMessages(chatID string, messageIDs []string) error {\n\t// TODO: implement\n\treturn nil\n}\n\n// =============================================================================\n// Resume Management (only called on failure/interrupt)\n// =============================================================================\n\n// SaveResume batch saves resume records\nfunc (m *Mongo) SaveResume(records []*types.Resume) error {\n\t// TODO: implement\n\treturn nil\n}\n\n// GetResume retrieves all resume records for a chat\nfunc (m *Mongo) GetResume(chatID string) ([]*types.Resume, error) {\n\t// TODO: implement\n\treturn nil, nil\n}\n\n// GetLastResume retrieves the last resume record for a chat\nfunc (m *Mongo) GetLastResume(chatID string) (*types.Resume, error) {\n\t// TODO: implement\n\treturn nil, nil\n}\n\n// GetResumeByStackID retrieves resume records for a specific stack\nfunc (m *Mongo) GetResumeByStackID(stackID string) ([]*types.Resume, error) {\n\t// TODO: implement\n\treturn nil, nil\n}\n\n// GetStackPath returns the stack path from root to the given stack\nfunc (m *Mongo) GetStackPath(stackID string) ([]string, error) {\n\t// TODO: implement\n\treturn nil, nil\n}\n\n// DeleteResume deletes all resume records for a chat\nfunc (m *Mongo) DeleteResume(chatID string) error {\n\t// TODO: implement\n\treturn nil\n}\n\n// =============================================================================\n// Assistant Management\n// =============================================================================\n\n// SaveAssistant saves assistant information\nfunc (m *Mongo) SaveAssistant(assistant *types.AssistantModel) (string, error) {\n\t// TODO: implement\n\treturn assistant.ID, nil\n}\n\n// UpdateAssistant updates specific fields of an assistant\nfunc (m *Mongo) UpdateAssistant(assistantID string, updates map[string]interface{}) error {\n\t// TODO: implement\n\treturn nil\n}\n\n// DeleteAssistant deletes an assistant\nfunc (m *Mongo) DeleteAssistant(assistantID string) error {\n\t// TODO: implement\n\treturn nil\n}\n\n// GetAssistants retrieves a list of assistants\nfunc (m *Mongo) GetAssistants(filter types.AssistantFilter, locale ...string) (*types.AssistantList, error) {\n\t// TODO: implement\n\treturn &types.AssistantList{}, nil\n}\n\n// GetAssistantTags retrieves all unique tags from assistants with filtering\nfunc (m *Mongo) GetAssistantTags(filter types.AssistantFilter, locale ...string) ([]types.Tag, error) {\n\t// TODO: implement\n\treturn []types.Tag{}, nil\n}\n\n// GetAssistant retrieves a single assistant by ID\nfunc (m *Mongo) GetAssistant(assistantID string, fields []string, locale ...string) (*types.AssistantModel, error) {\n\t// TODO: implement\n\treturn nil, nil\n}\n\n// DeleteAssistants deletes assistants based on filter conditions\nfunc (m *Mongo) DeleteAssistants(filter types.AssistantFilter) (int64, error) {\n\t// TODO: implement\n\treturn 0, nil\n}\n\n// =============================================================================\n// Search Management\n// =============================================================================\n\n// SaveSearch saves a search record for a request\nfunc (m *Mongo) SaveSearch(search *types.Search) error {\n\t// TODO: implement\n\treturn nil\n}\n\n// GetSearches retrieves all search records for a request\nfunc (m *Mongo) GetSearches(requestID string) ([]*types.Search, error) {\n\t// TODO: implement\n\treturn nil, nil\n}\n\n// GetReference retrieves a single reference by request ID and index\nfunc (m *Mongo) GetReference(requestID string, index int) (*types.Reference, error) {\n\t// TODO: implement\n\treturn nil, nil\n}\n\n// DeleteSearches deletes all search records for a chat\nfunc (m *Mongo) DeleteSearches(chatID string) error {\n\t// TODO: implement\n\treturn nil\n}\n" }, { "path": "agent/store/redis/redis.go", "content": "package redis\n\nimport \"github.com/yaoapp/yao/agent/store/types\"\n\n// Redis represents a Redis-based conversation storage\ntype Redis struct{}\n\n// NewRedis create a new redis store\nfunc NewRedis() types.Store {\n\treturn &Redis{}\n}\n\n// =============================================================================\n// Chat Management\n// =============================================================================\n\n// CreateChat creates a new chat session\nfunc (r *Redis) CreateChat(chat *types.Chat) error {\n\t// TODO: implement\n\treturn nil\n}\n\n// GetChat retrieves a single chat by ID\nfunc (r *Redis) GetChat(chatID string) (*types.Chat, error) {\n\t// TODO: implement\n\treturn nil, nil\n}\n\n// UpdateChat updates chat fields\nfunc (r *Redis) UpdateChat(chatID string, updates map[string]interface{}) error {\n\t// TODO: implement\n\treturn nil\n}\n\n// DeleteChat deletes a chat and its associated messages\nfunc (r *Redis) DeleteChat(chatID string) error {\n\t// TODO: implement\n\treturn nil\n}\n\n// ListChats retrieves a paginated list of chats with optional grouping\nfunc (r *Redis) ListChats(filter types.ChatFilter) (*types.ChatList, error) {\n\t// TODO: implement\n\treturn nil, nil\n}\n\n// =============================================================================\n// Message Management\n// =============================================================================\n\n// SaveMessages batch saves messages for a chat\nfunc (r *Redis) SaveMessages(chatID string, messages []*types.Message) error {\n\t// TODO: implement\n\treturn nil\n}\n\n// GetMessages retrieves messages for a chat with filtering\nfunc (r *Redis) GetMessages(chatID string, filter types.MessageFilter) ([]*types.Message, error) {\n\t// TODO: implement\n\treturn nil, nil\n}\n\n// UpdateMessage updates a single message\nfunc (r *Redis) UpdateMessage(messageID string, updates map[string]interface{}) error {\n\t// TODO: implement\n\treturn nil\n}\n\n// DeleteMessages deletes specific messages from a chat\nfunc (r *Redis) DeleteMessages(chatID string, messageIDs []string) error {\n\t// TODO: implement\n\treturn nil\n}\n\n// =============================================================================\n// Resume Management (only called on failure/interrupt)\n// =============================================================================\n\n// SaveResume batch saves resume records\nfunc (r *Redis) SaveResume(records []*types.Resume) error {\n\t// TODO: implement\n\treturn nil\n}\n\n// GetResume retrieves all resume records for a chat\nfunc (r *Redis) GetResume(chatID string) ([]*types.Resume, error) {\n\t// TODO: implement\n\treturn nil, nil\n}\n\n// GetLastResume retrieves the last resume record for a chat\nfunc (r *Redis) GetLastResume(chatID string) (*types.Resume, error) {\n\t// TODO: implement\n\treturn nil, nil\n}\n\n// GetResumeByStackID retrieves resume records for a specific stack\nfunc (r *Redis) GetResumeByStackID(stackID string) ([]*types.Resume, error) {\n\t// TODO: implement\n\treturn nil, nil\n}\n\n// GetStackPath returns the stack path from root to the given stack\nfunc (r *Redis) GetStackPath(stackID string) ([]string, error) {\n\t// TODO: implement\n\treturn nil, nil\n}\n\n// DeleteResume deletes all resume records for a chat\nfunc (r *Redis) DeleteResume(chatID string) error {\n\t// TODO: implement\n\treturn nil\n}\n\n// =============================================================================\n// Assistant Management\n// =============================================================================\n\n// SaveAssistant saves assistant information\nfunc (r *Redis) SaveAssistant(assistant *types.AssistantModel) (string, error) {\n\t// TODO: implement\n\treturn assistant.ID, nil\n}\n\n// UpdateAssistant updates specific fields of an assistant\nfunc (r *Redis) UpdateAssistant(assistantID string, updates map[string]interface{}) error {\n\t// TODO: implement\n\treturn nil\n}\n\n// DeleteAssistant deletes an assistant\nfunc (r *Redis) DeleteAssistant(assistantID string) error {\n\t// TODO: implement\n\treturn nil\n}\n\n// GetAssistants retrieves a list of assistants\nfunc (r *Redis) GetAssistants(filter types.AssistantFilter, locale ...string) (*types.AssistantList, error) {\n\t// TODO: implement\n\treturn &types.AssistantList{}, nil\n}\n\n// GetAssistantTags retrieves all unique tags from assistants with filtering\nfunc (r *Redis) GetAssistantTags(filter types.AssistantFilter, locale ...string) ([]types.Tag, error) {\n\t// TODO: implement\n\treturn []types.Tag{}, nil\n}\n\n// GetAssistant retrieves a single assistant by ID\nfunc (r *Redis) GetAssistant(assistantID string, fields []string, locale ...string) (*types.AssistantModel, error) {\n\t// TODO: implement\n\treturn nil, nil\n}\n\n// DeleteAssistants deletes assistants based on filter conditions\nfunc (r *Redis) DeleteAssistants(filter types.AssistantFilter) (int64, error) {\n\t// TODO: implement\n\treturn 0, nil\n}\n\n// =============================================================================\n// Search Management\n// =============================================================================\n\n// SaveSearch saves a search record for a request\nfunc (r *Redis) SaveSearch(search *types.Search) error {\n\t// TODO: implement\n\treturn nil\n}\n\n// GetSearches retrieves all search records for a request\nfunc (r *Redis) GetSearches(requestID string) ([]*types.Search, error) {\n\t// TODO: implement\n\treturn nil, nil\n}\n\n// GetReference retrieves a single reference by request ID and index\nfunc (r *Redis) GetReference(requestID string, index int) (*types.Reference, error) {\n\t// TODO: implement\n\treturn nil, nil\n}\n\n// DeleteSearches deletes all search records for a chat\nfunc (r *Redis) DeleteSearches(chatID string) error {\n\t// TODO: implement\n\treturn nil\n}\n" }, { "path": "agent/store/types/convert.go", "content": "package types\n\nimport (\n\t\"fmt\"\n\t\"strings\"\n\t\"time\"\n\n\tjsoniter \"github.com/json-iterator/go\"\n\t\"github.com/spf13/cast\"\n\t\"github.com/yaoapp/gou/connector\"\n\t\"github.com/yaoapp/kun/log\"\n\t\"github.com/yaoapp/yao/agent/context\"\n\t\"github.com/yaoapp/yao/agent/i18n\"\n\tsearchTypes \"github.com/yaoapp/yao/agent/search/types\"\n)\n\n// ToKnowledgeBase converts various types to KnowledgeBase\nfunc ToKnowledgeBase(v interface{}) (*KnowledgeBase, error) {\n\tif v == nil {\n\t\treturn nil, nil\n\t}\n\n\tswitch kb := v.(type) {\n\tcase *KnowledgeBase:\n\t\treturn kb, nil\n\n\tcase KnowledgeBase:\n\t\treturn &kb, nil\n\n\tcase []string:\n\t\treturn &KnowledgeBase{Collections: kb}, nil\n\n\tcase []interface{}:\n\t\tvar collections []string\n\t\tfor _, item := range kb {\n\t\t\tcollections = append(collections, cast.ToString(item))\n\t\t}\n\t\treturn &KnowledgeBase{Collections: collections}, nil\n\n\tdefault:\n\t\traw, err := jsoniter.Marshal(kb)\n\t\tif err != nil {\n\t\t\treturn nil, fmt.Errorf(\"kb format error: %s\", err.Error())\n\t\t}\n\n\t\tvar knowledgeBase KnowledgeBase\n\t\terr = jsoniter.Unmarshal(raw, &knowledgeBase)\n\t\tif err != nil {\n\t\t\treturn nil, fmt.Errorf(\"kb format error: %s\", err.Error())\n\t\t}\n\t\treturn &knowledgeBase, nil\n\t}\n}\n\n// ToDatabase converts various types to Database\nfunc ToDatabase(v interface{}) (*Database, error) {\n\tif v == nil {\n\t\treturn nil, nil\n\t}\n\n\tswitch db := v.(type) {\n\tcase *Database:\n\t\treturn db, nil\n\n\tcase Database:\n\t\treturn &db, nil\n\n\tcase []string:\n\t\treturn &Database{Models: db}, nil\n\n\tcase []interface{}:\n\t\tvar models []string\n\t\tfor _, item := range db {\n\t\t\tmodels = append(models, cast.ToString(item))\n\t\t}\n\t\treturn &Database{Models: models}, nil\n\n\tdefault:\n\t\traw, err := jsoniter.Marshal(db)\n\t\tif err != nil {\n\t\t\treturn nil, fmt.Errorf(\"db format error: %s\", err.Error())\n\t\t}\n\n\t\tvar database Database\n\t\terr = jsoniter.Unmarshal(raw, &database)\n\t\tif err != nil {\n\t\t\treturn nil, fmt.Errorf(\"db format error: %s\", err.Error())\n\t\t}\n\t\treturn &database, nil\n\t}\n}\n\n// ToMCPServers converts various types to MCPServers\nfunc ToMCPServers(v interface{}) (*MCPServers, error) {\n\tif v == nil {\n\t\treturn nil, nil\n\t}\n\n\tswitch mcp := v.(type) {\n\tcase *MCPServers:\n\t\treturn mcp, nil\n\n\tcase MCPServers:\n\t\treturn &mcp, nil\n\n\tdefault:\n\t\t// For any type (including []string, []interface{}, map[string]interface{}),\n\t\t// marshal and unmarshal to MCPServers using custom UnmarshalJSON\n\t\traw, err := jsoniter.Marshal(mcp)\n\t\tif err != nil {\n\t\t\treturn nil, fmt.Errorf(\"mcp format error: %s\", err.Error())\n\t\t}\n\n\t\tvar mcpServers MCPServers\n\t\terr = jsoniter.Unmarshal(raw, &mcpServers)\n\t\tif err != nil {\n\t\t\treturn nil, fmt.Errorf(\"mcp format error: %s\", err.Error())\n\t\t}\n\t\treturn &mcpServers, nil\n\t}\n}\n\n// ToWorkflow converts various types to Workflow\nfunc ToWorkflow(v interface{}) (*Workflow, error) {\n\tif v == nil {\n\t\treturn nil, nil\n\t}\n\n\tswitch workflow := v.(type) {\n\tcase *Workflow:\n\t\treturn workflow, nil\n\n\tcase Workflow:\n\t\treturn &workflow, nil\n\n\tcase []string:\n\t\treturn &Workflow{Workflows: workflow}, nil\n\n\tcase []interface{}:\n\t\tvar workflows []string\n\t\tfor _, item := range workflow {\n\t\t\tworkflows = append(workflows, cast.ToString(item))\n\t\t}\n\t\treturn &Workflow{Workflows: workflows}, nil\n\n\tdefault:\n\t\traw, err := jsoniter.Marshal(workflow)\n\t\tif err != nil {\n\t\t\treturn nil, fmt.Errorf(\"workflow format error: %s\", err.Error())\n\t\t}\n\n\t\tvar wf Workflow\n\t\terr = jsoniter.Unmarshal(raw, &wf)\n\t\tif err != nil {\n\t\t\treturn nil, fmt.Errorf(\"workflow format error: %s\", err.Error())\n\t\t}\n\t\treturn &wf, nil\n\t}\n}\n\n// ToSandbox converts various types to Sandbox\nfunc ToSandbox(v interface{}) (*Sandbox, error) {\n\tif v == nil {\n\t\treturn nil, nil\n\t}\n\n\tswitch sandbox := v.(type) {\n\tcase *Sandbox:\n\t\treturn sandbox, nil\n\n\tcase Sandbox:\n\t\treturn &sandbox, nil\n\n\tdefault:\n\t\traw, err := jsoniter.Marshal(sandbox)\n\t\tif err != nil {\n\t\t\treturn nil, fmt.Errorf(\"sandbox format error: %s\", err.Error())\n\t\t}\n\n\t\tvar sb Sandbox\n\t\terr = jsoniter.Unmarshal(raw, &sb)\n\t\tif err != nil {\n\t\t\treturn nil, fmt.Errorf(\"sandbox format error: %s\", err.Error())\n\t\t}\n\t\treturn &sb, nil\n\t}\n}\n\n// ToMySQLTime converts various types to MySQL datetime format\nfunc ToMySQLTime(v interface{}) string {\n\tswitch val := v.(type) {\n\tcase int64:\n\t\tif val == 0 {\n\t\t\treturn \"0000-00-00 00:00:00\"\n\t\t}\n\t\treturn time.Unix(val/1e9, val%1e9).Format(\"2006-01-02 15:04:05\")\n\n\tcase int:\n\t\tif val == 0 {\n\t\t\treturn \"0000-00-00 00:00:00\"\n\t\t}\n\t\treturn time.Unix(int64(val)/1e9, int64(val)%1e9).Format(\"2006-01-02 15:04:05\")\n\n\tcase string:\n\t\t// If already in MySQL format, return as-is\n\t\tif _, err := time.Parse(\"2006-01-02 15:04:05\", val); err == nil {\n\t\t\treturn val\n\t\t}\n\t\t// Try RFC3339 format\n\t\tif ts, err := time.Parse(time.RFC3339, val); err == nil {\n\t\t\treturn ts.Format(\"2006-01-02 15:04:05\")\n\t\t}\n\t\t// Try parsing as Unix timestamp\n\t\tif ts, err := cast.ToInt64E(val); err == nil {\n\t\t\tif ts == 0 {\n\t\t\t\treturn \"0000-00-00 00:00:00\"\n\t\t\t}\n\t\t\treturn time.Unix(ts/1e9, ts%1e9).Format(\"2006-01-02 15:04:05\")\n\t\t}\n\t\treturn val\n\n\tcase time.Time:\n\t\tif val.IsZero() {\n\t\t\treturn \"0000-00-00 00:00:00\"\n\t\t}\n\t\treturn val.Format(\"2006-01-02 15:04:05\")\n\n\tcase nil:\n\t\treturn \"0000-00-00 00:00:00\"\n\n\tdefault:\n\t\treturn \"0000-00-00 00:00:00\"\n\t}\n}\n\n// ToAssistantModel converts various types to AssistantModel\nfunc ToAssistantModel(v interface{}) (*AssistantModel, error) {\n\tif v == nil {\n\t\treturn nil, nil\n\t}\n\n\t// If already an AssistantModel, return it\n\tswitch model := v.(type) {\n\tcase *AssistantModel:\n\t\treturn model, nil\n\tcase AssistantModel:\n\t\treturn &model, nil\n\t}\n\n\t// Convert to map first if needed\n\tvar data map[string]interface{}\n\tswitch v := v.(type) {\n\tcase map[string]interface{}:\n\t\tdata = v\n\tdefault:\n\t\t// Try to marshal and unmarshal\n\t\traw, err := jsoniter.Marshal(v)\n\t\tif err != nil {\n\t\t\treturn nil, fmt.Errorf(\"failed to marshal to AssistantModel: %w\", err)\n\t\t}\n\t\terr = jsoniter.Unmarshal(raw, &data)\n\t\tif err != nil {\n\t\t\treturn nil, fmt.Errorf(\"failed to unmarshal to map: %w\", err)\n\t\t}\n\t}\n\n\tmodel := &AssistantModel{}\n\n\t// Basic string fields\n\tif id, ok := data[\"assistant_id\"].(string); ok {\n\t\tmodel.ID = id\n\t}\n\tif typ, ok := data[\"type\"].(string); ok {\n\t\tmodel.Type = typ\n\t}\n\tif name, ok := data[\"name\"].(string); ok {\n\t\tmodel.Name = name\n\t}\n\tif avatar, ok := data[\"avatar\"].(string); ok {\n\t\tmodel.Avatar = avatar\n\t}\n\tif connector, ok := data[\"connector\"].(string); ok {\n\t\tmodel.Connector = connector\n\t}\n\tif path, ok := data[\"path\"].(string); ok {\n\t\tmodel.Path = path\n\t}\n\tif source, ok := data[\"source\"].(string); ok {\n\t\tmodel.Source = source\n\t}\n\tif description, ok := data[\"description\"].(string); ok {\n\t\tmodel.Description = description\n\t}\n\tif capabilities, ok := data[\"capabilities\"].(string); ok {\n\t\tmodel.Capabilities = capabilities\n\t}\n\tif share, ok := data[\"share\"].(string); ok {\n\t\tmodel.Share = share\n\t}\n\n\t// Boolean fields (handle both bool and int types from database)\n\tmodel.BuiltIn = getBoolValue(data, \"built_in\")\n\tmodel.Readonly = getBoolValue(data, \"readonly\")\n\tmodel.Public = getBoolValue(data, \"public\")\n\tmodel.Mentionable = getBoolValue(data, \"mentionable\")\n\tmodel.Automated = getBoolValue(data, \"automated\")\n\n\t// Integer fields\n\tif sort, ok := data[\"sort\"].(int); ok {\n\t\tmodel.Sort = sort\n\t} else if sort, ok := data[\"sort\"].(float64); ok {\n\t\tmodel.Sort = int(sort)\n\t}\n\n\tif createdAt, ok := data[\"created_at\"].(int64); ok {\n\t\tmodel.CreatedAt = createdAt\n\t} else if createdAt, ok := data[\"created_at\"].(float64); ok {\n\t\tmodel.CreatedAt = int64(createdAt)\n\t}\n\n\tif updatedAt, ok := data[\"updated_at\"].(int64); ok {\n\t\tmodel.UpdatedAt = updatedAt\n\t} else if updatedAt, ok := data[\"updated_at\"].(float64); ok {\n\t\tmodel.UpdatedAt = int64(updatedAt)\n\t}\n\n\t// Tags (string array)\n\tif tags, ok := data[\"tags\"]; ok && tags != nil {\n\t\traw, err := jsoniter.Marshal(tags)\n\t\tif err == nil {\n\t\t\tvar t []string\n\t\t\tif err := jsoniter.Unmarshal(raw, &t); err == nil {\n\t\t\t\tmodel.Tags = t\n\t\t\t}\n\t\t}\n\t}\n\n\t// Modes (string array)\n\tif modes, ok := data[\"modes\"]; ok && modes != nil {\n\t\traw, err := jsoniter.Marshal(modes)\n\t\tif err == nil {\n\t\t\tvar m []string\n\t\t\tif err := jsoniter.Unmarshal(raw, &m); err == nil {\n\t\t\t\tmodel.Modes = m\n\t\t\t}\n\t\t}\n\t}\n\n\t// DefaultMode (string)\n\tif defaultMode, ok := data[\"default_mode\"].(string); ok {\n\t\tmodel.DefaultMode = defaultMode\n\t}\n\n\t// Options (map)\n\tif options, ok := data[\"options\"].(map[string]interface{}); ok {\n\t\tmodel.Options = options\n\t}\n\n\t// Prompts\n\tif prompts, ok := data[\"prompts\"]; ok && prompts != nil {\n\t\traw, err := jsoniter.Marshal(prompts)\n\t\tif err == nil {\n\t\t\tvar p []Prompt\n\t\t\tif err := jsoniter.Unmarshal(raw, &p); err == nil {\n\t\t\t\tmodel.Prompts = p\n\t\t\t}\n\t\t}\n\t}\n\n\t// PromptPresets\n\tif promptPresets, ok := data[\"prompt_presets\"]; ok && promptPresets != nil {\n\t\traw, err := jsoniter.Marshal(promptPresets)\n\t\tif err == nil {\n\t\t\tvar pp map[string][]Prompt\n\t\t\tif err := jsoniter.Unmarshal(raw, &pp); err == nil {\n\t\t\t\tmodel.PromptPresets = pp\n\t\t\t}\n\t\t}\n\t}\n\n\t// DisableGlobalPrompts\n\tmodel.DisableGlobalPrompts = getBoolValue(data, \"disable_global_prompts\")\n\n\t// ConnectorOptions\n\tif connectorOptions, ok := data[\"connector_options\"]; ok && connectorOptions != nil {\n\t\traw, err := jsoniter.Marshal(connectorOptions)\n\t\tif err == nil {\n\t\t\tvar co ConnectorOptions\n\t\t\tif err := jsoniter.Unmarshal(raw, &co); err == nil {\n\t\t\t\tmodel.ConnectorOptions = &co\n\t\t\t}\n\t\t}\n\t}\n\n\t// KB\n\tif kb, ok := data[\"kb\"]; ok && kb != nil {\n\t\tkbConverted, err := ToKnowledgeBase(kb)\n\t\tif err == nil {\n\t\t\tmodel.KB = kbConverted\n\t\t}\n\t}\n\n\t// DB\n\tif db, ok := data[\"db\"]; ok && db != nil {\n\t\tdbConverted, err := ToDatabase(db)\n\t\tif err == nil {\n\t\t\tmodel.DB = dbConverted\n\t\t}\n\t}\n\n\t// MCP\n\tif mcp, ok := data[\"mcp\"]; ok && mcp != nil {\n\t\tmcpConverted, err := ToMCPServers(mcp)\n\t\tif err == nil {\n\t\t\tmodel.MCP = mcpConverted\n\t\t}\n\t}\n\n\t// Workflow\n\tif workflow, ok := data[\"workflow\"]; ok && workflow != nil {\n\t\twf, err := ToWorkflow(workflow)\n\t\tif err == nil {\n\t\t\tmodel.Workflow = wf\n\t\t}\n\t}\n\n\t// Sandbox\n\tif sandbox, ok := data[\"sandbox\"]; ok && sandbox != nil {\n\t\tsb, err := ToSandbox(sandbox)\n\t\tif err == nil {\n\t\t\tmodel.Sandbox = sb\n\t\t}\n\t}\n\n\t// Placeholder\n\tif placeholder, ok := data[\"placeholder\"]; ok && placeholder != nil {\n\t\traw, err := jsoniter.Marshal(placeholder)\n\t\tif err == nil {\n\t\t\tvar ph Placeholder\n\t\t\tif err := jsoniter.Unmarshal(raw, &ph); err == nil {\n\t\t\t\tmodel.Placeholder = &ph\n\t\t\t}\n\t\t}\n\t}\n\n\t// Locales\n\tif locales, ok := data[\"locales\"]; ok && locales != nil {\n\t\traw, err := jsoniter.Marshal(locales)\n\t\tif err == nil {\n\t\t\tvar loc i18n.Map\n\t\t\tif err := jsoniter.Unmarshal(raw, &loc); err == nil {\n\t\t\t\tmodel.Locales = loc\n\t\t\t}\n\t\t}\n\t}\n\n\t// Dependencies\n\tif deps, ok := data[\"dependencies\"]; ok && deps != nil {\n\t\traw, err := jsoniter.Marshal(deps)\n\t\tif err == nil {\n\t\t\tvar d map[string]string\n\t\t\tif err := jsoniter.Unmarshal(raw, &d); err == nil {\n\t\t\t\tmodel.Dependencies = d\n\t\t\t}\n\t\t}\n\t}\n\n\t// Permission fields\n\tif createdBy, ok := data[\"__yao_created_by\"].(string); ok {\n\t\tmodel.YaoCreatedBy = createdBy\n\t}\n\tif updatedBy, ok := data[\"__yao_updated_by\"].(string); ok {\n\t\tmodel.YaoUpdatedBy = updatedBy\n\t}\n\tif teamID, ok := data[\"__yao_team_id\"].(string); ok {\n\t\tmodel.YaoTeamID = teamID\n\t}\n\tif tenantID, ok := data[\"__yao_tenant_id\"].(string); ok {\n\t\tmodel.YaoTenantID = tenantID\n\t}\n\n\treturn model, nil\n}\n\n// getBoolValue extracts a boolean value from a map, handling both bool and numeric types\nfunc getBoolValue(data map[string]interface{}, key string) bool {\n\tif v, ok := data[key]; ok && v != nil {\n\t\tswitch val := v.(type) {\n\t\tcase bool:\n\t\t\treturn val\n\t\tcase int:\n\t\t\treturn val != 0\n\t\tcase int64:\n\t\t\treturn val != 0\n\t\tcase float64:\n\t\t\treturn val != 0\n\t\tcase string:\n\t\t\treturn val == \"true\" || val == \"1\"\n\t\t}\n\t}\n\treturn false\n}\n\n// ModelID generates an OpenAI-compatible model ID from assistant\n// Format: [prefix-]assistantName-model-yao_assistantID\n// prefix is optional, if provided, it will be prepended to the model ID\nfunc (assistant AssistantModel) ModelID(prefix ...string) string {\n\t// Clean assistant name (remove spaces and special characters)\n\tassistantName := strings.ReplaceAll(assistant.Name, \" \", \"-\")\n\tassistantName = strings.ToLower(assistantName)\n\n\t// Get connector name from assistant\n\tconnectorName := assistant.Connector\n\tif connectorName == \"\" {\n\t\tlog.Error(\"Assistant %s has no connector configured\", assistant.ID)\n\t\tmodelID := assistantName + \"-unknown-yao_\" + assistant.ID\n\t\tif len(prefix) > 0 && prefix[0] != \"\" {\n\t\t\treturn prefix[0] + modelID\n\t\t}\n\t\treturn modelID\n\t}\n\n\t// Get model name\n\tmodelName := \"\"\n\n\t// First, try to get custom model from Options\n\tif assistant.Options != nil {\n\t\tif m, ok := assistant.Options[\"model\"].(string); ok && m != \"\" {\n\t\t\tmodelName = m\n\t\t}\n\t}\n\n\t// If no custom model in options, try to get from connector configuration\n\tif modelName == \"\" {\n\t\tconn, err := connector.Select(connectorName)\n\t\tif err != nil {\n\t\t\tlog.Error(\"Failed to select connector %s for assistant %s: %v\", connectorName, assistant.ID, err)\n\t\t\tmodelID := assistantName + \"-unknown-yao_\" + assistant.ID\n\t\t\tif len(prefix) > 0 && prefix[0] != \"\" {\n\t\t\t\treturn prefix[0] + modelID\n\t\t\t}\n\t\t\treturn modelID\n\t\t}\n\n\t\t// Get model from connector settings\n\t\tsettings := conn.Setting()\n\t\tif settings != nil {\n\t\t\tif m, ok := settings[\"model\"].(string); ok && m != \"\" {\n\t\t\t\tmodelName = m\n\t\t\t}\n\t\t}\n\n\t\tif modelName == \"\" {\n\t\t\tlog.Error(\"Connector %s has no model configured for assistant %s\", connectorName, assistant.ID)\n\t\t\tmodelID := assistantName + \"-unknown-yao_\" + assistant.ID\n\t\t\tif len(prefix) > 0 && prefix[0] != \"\" {\n\t\t\t\treturn prefix[0] + modelID\n\t\t\t}\n\t\t\treturn modelID\n\t\t}\n\t}\n\n\t// Format: [prefix-]assistantName-model-yao_assistantID\n\tmodelID := assistantName + \"-\" + modelName + \"-yao_\" + assistant.ID\n\tif len(prefix) > 0 && prefix[0] != \"\" {\n\t\treturn prefix[0] + modelID\n\t}\n\treturn modelID\n}\n\n// ParseModelID extracts assistant ID from model ID\n// Expected format: [prefix-]assistantName-model-yao_assistantID\n// The function handles optional prefixes (e.g., \"yao-agents-\")\nfunc ParseModelID(modelID string) string {\n\t// Find the last occurrence of \"yao_\"\n\tparts := strings.Split(modelID, \"-yao_\")\n\tif len(parts) < 2 {\n\t\treturn \"\"\n\t}\n\treturn parts[len(parts)-1]\n}\n\n// ToConnectorOptions converts various types to ConnectorOptions\nfunc ToConnectorOptions(v interface{}) (*ConnectorOptions, error) {\n\tif v == nil {\n\t\treturn nil, nil\n\t}\n\n\tswitch opts := v.(type) {\n\tcase *ConnectorOptions:\n\t\treturn opts, nil\n\n\tcase ConnectorOptions:\n\t\treturn &opts, nil\n\n\tdefault:\n\t\traw, err := jsoniter.Marshal(opts)\n\t\tif err != nil {\n\t\t\treturn nil, fmt.Errorf(\"connector_options format error: %s\", err.Error())\n\t\t}\n\n\t\tvar connOpts ConnectorOptions\n\t\terr = jsoniter.Unmarshal(raw, &connOpts)\n\t\tif err != nil {\n\t\t\treturn nil, fmt.Errorf(\"connector_options format error: %s\", err.Error())\n\t\t}\n\t\treturn &connOpts, nil\n\t}\n}\n\n// ToModes converts various types to []string for modes\nfunc ToModes(v interface{}) ([]string, error) {\n\tif v == nil {\n\t\treturn nil, nil\n\t}\n\n\tswitch modes := v.(type) {\n\tcase []string:\n\t\treturn modes, nil\n\n\tcase []interface{}:\n\t\tvar result []string\n\t\tfor _, item := range modes {\n\t\t\tresult = append(result, cast.ToString(item))\n\t\t}\n\t\treturn result, nil\n\n\tcase string:\n\t\t// Single string becomes a slice with one element\n\t\treturn []string{modes}, nil\n\n\tdefault:\n\t\traw, err := jsoniter.Marshal(modes)\n\t\tif err != nil {\n\t\t\treturn nil, fmt.Errorf(\"modes format error: %s\", err.Error())\n\t\t}\n\n\t\tvar result []string\n\t\terr = jsoniter.Unmarshal(raw, &result)\n\t\tif err != nil {\n\t\t\treturn nil, fmt.Errorf(\"modes format error: %s\", err.Error())\n\t\t}\n\t\treturn result, nil\n\t}\n}\n\n// ToPromptPresets converts various types to map[string][]Prompt\nfunc ToPromptPresets(v interface{}) (map[string][]Prompt, error) {\n\tif v == nil {\n\t\treturn nil, nil\n\t}\n\n\tswitch presets := v.(type) {\n\tcase map[string][]Prompt:\n\t\treturn presets, nil\n\n\tdefault:\n\t\traw, err := jsoniter.Marshal(presets)\n\t\tif err != nil {\n\t\t\treturn nil, fmt.Errorf(\"prompt_presets format error: %s\", err.Error())\n\t\t}\n\n\t\tvar result map[string][]Prompt\n\t\terr = jsoniter.Unmarshal(raw, &result)\n\t\tif err != nil {\n\t\t\treturn nil, fmt.Errorf(\"prompt_presets format error: %s\", err.Error())\n\t\t}\n\t\treturn result, nil\n\t}\n}\n\n// ToUses converts various types to context.Uses\nfunc ToUses(v interface{}) (*context.Uses, error) {\n\tif v == nil {\n\t\treturn nil, nil\n\t}\n\n\tswitch uses := v.(type) {\n\tcase *context.Uses:\n\t\treturn uses, nil\n\n\tcase context.Uses:\n\t\treturn &uses, nil\n\n\tdefault:\n\t\traw, err := jsoniter.Marshal(uses)\n\t\tif err != nil {\n\t\t\treturn nil, fmt.Errorf(\"uses format error: %s\", err.Error())\n\t\t}\n\n\t\tvar result context.Uses\n\t\terr = jsoniter.Unmarshal(raw, &result)\n\t\tif err != nil {\n\t\t\treturn nil, fmt.Errorf(\"uses format error: %s\", err.Error())\n\t\t}\n\t\treturn &result, nil\n\t}\n}\n\n// ToSearchConfig converts various types to searchTypes.Config\nfunc ToSearchConfig(v interface{}) (*searchTypes.Config, error) {\n\tif v == nil {\n\t\treturn nil, nil\n\t}\n\n\tswitch cfg := v.(type) {\n\tcase *searchTypes.Config:\n\t\treturn cfg, nil\n\n\tcase searchTypes.Config:\n\t\treturn &cfg, nil\n\n\tdefault:\n\t\traw, err := jsoniter.Marshal(cfg)\n\t\tif err != nil {\n\t\t\treturn nil, fmt.Errorf(\"search config format error: %s\", err.Error())\n\t\t}\n\n\t\tvar result searchTypes.Config\n\t\terr = jsoniter.Unmarshal(raw, &result)\n\t\tif err != nil {\n\t\t\treturn nil, fmt.Errorf(\"search config format error: %s\", err.Error())\n\t\t}\n\t\treturn &result, nil\n\t}\n}\n" }, { "path": "agent/store/types/convert_test.go", "content": "package types\n\nimport (\n\t\"testing\"\n\t\"time\"\n)\n\n// TestToDatabase tests the ToDatabase conversion function\nfunc TestToDatabase(t *testing.T) {\n\tt.Run(\"NilInput\", func(t *testing.T) {\n\t\tresult, err := ToDatabase(nil)\n\t\tif err != nil {\n\t\t\tt.Errorf(\"Expected no error, got: %v\", err)\n\t\t}\n\t\tif result != nil {\n\t\t\tt.Errorf(\"Expected nil result, got: %v\", result)\n\t\t}\n\t})\n\n\tt.Run(\"DatabasePointer\", func(t *testing.T) {\n\t\tdb := &Database{Models: []string{\"model1\", \"model2\"}}\n\t\tresult, err := ToDatabase(db)\n\t\tif err != nil {\n\t\t\tt.Errorf(\"Expected no error, got: %v\", err)\n\t\t}\n\t\tif result != db {\n\t\t\tt.Errorf(\"Expected same pointer\")\n\t\t}\n\t})\n\n\tt.Run(\"DatabaseValue\", func(t *testing.T) {\n\t\tdb := Database{Models: []string{\"model1\", \"model2\"}}\n\t\tresult, err := ToDatabase(db)\n\t\tif err != nil {\n\t\t\tt.Errorf(\"Expected no error, got: %v\", err)\n\t\t}\n\t\tif len(result.Models) != 2 {\n\t\t\tt.Errorf(\"Expected 2 models, got %d\", len(result.Models))\n\t\t}\n\t})\n\n\tt.Run(\"StringSlice\", func(t *testing.T) {\n\t\tmodels := []string{\"model1\", \"model2\", \"model3\"}\n\t\tresult, err := ToDatabase(models)\n\t\tif err != nil {\n\t\t\tt.Errorf(\"Expected no error, got: %v\", err)\n\t\t}\n\t\tif len(result.Models) != 3 {\n\t\t\tt.Errorf(\"Expected 3 models, got %d\", len(result.Models))\n\t\t}\n\t\tif result.Models[0] != \"model1\" {\n\t\t\tt.Errorf(\"Expected 'model1', got '%s'\", result.Models[0])\n\t\t}\n\t})\n\n\tt.Run(\"InterfaceSlice\", func(t *testing.T) {\n\t\tmodels := []interface{}{\"model1\", \"model2\", 123}\n\t\tresult, err := ToDatabase(models)\n\t\tif err != nil {\n\t\t\tt.Errorf(\"Expected no error, got: %v\", err)\n\t\t}\n\t\tif len(result.Models) != 3 {\n\t\t\tt.Errorf(\"Expected 3 models, got %d\", len(result.Models))\n\t\t}\n\t\tif result.Models[2] != \"123\" {\n\t\t\tt.Errorf(\"Expected '123', got '%s'\", result.Models[2])\n\t\t}\n\t})\n\n\tt.Run(\"MapInput\", func(t *testing.T) {\n\t\tdata := map[string]interface{}{\n\t\t\t\"models\": []string{\"model1\", \"model2\"},\n\t\t}\n\t\tresult, err := ToDatabase(data)\n\t\tif err != nil {\n\t\t\tt.Errorf(\"Expected no error, got: %v\", err)\n\t\t}\n\t\tif len(result.Models) != 2 {\n\t\t\tt.Errorf(\"Expected 2 models, got %d\", len(result.Models))\n\t\t}\n\t})\n\n\tt.Run(\"InvalidInput\", func(t *testing.T) {\n\t\t// Test with data that can't be marshaled\n\t\tinvalidData := make(chan int)\n\t\t_, err := ToDatabase(invalidData)\n\t\tif err == nil {\n\t\t\tt.Error(\"Expected error for invalid input\")\n\t\t}\n\t})\n\n\tt.Run(\"InvalidJSONUnmarshal\", func(t *testing.T) {\n\t\t// Test with data that marshals but can't unmarshal to Database\n\t\tdata := map[string]interface{}{\n\t\t\t\"invalid_field\": \"should cause unmarshal to fail gracefully\",\n\t\t}\n\t\tresult, err := ToDatabase(data)\n\t\t// Should not error, just return empty Database\n\t\tif err != nil {\n\t\t\tt.Errorf(\"Expected no error, got: %v\", err)\n\t\t}\n\t\tif result == nil {\n\t\t\tt.Error(\"Expected non-nil result\")\n\t\t}\n\t})\n}\n\n// TestToKnowledgeBase tests the ToKnowledgeBase conversion function\nfunc TestToKnowledgeBase(t *testing.T) {\n\tt.Run(\"NilInput\", func(t *testing.T) {\n\t\tresult, err := ToKnowledgeBase(nil)\n\t\tif err != nil {\n\t\t\tt.Errorf(\"Expected no error, got: %v\", err)\n\t\t}\n\t\tif result != nil {\n\t\t\tt.Errorf(\"Expected nil result, got: %v\", result)\n\t\t}\n\t})\n\n\tt.Run(\"KnowledgeBasePointer\", func(t *testing.T) {\n\t\tkb := &KnowledgeBase{Collections: []string{\"col1\", \"col2\"}}\n\t\tresult, err := ToKnowledgeBase(kb)\n\t\tif err != nil {\n\t\t\tt.Errorf(\"Expected no error, got: %v\", err)\n\t\t}\n\t\tif result != kb {\n\t\t\tt.Errorf(\"Expected same pointer\")\n\t\t}\n\t})\n\n\tt.Run(\"KnowledgeBaseValue\", func(t *testing.T) {\n\t\tkb := KnowledgeBase{Collections: []string{\"col1\", \"col2\"}}\n\t\tresult, err := ToKnowledgeBase(kb)\n\t\tif err != nil {\n\t\t\tt.Errorf(\"Expected no error, got: %v\", err)\n\t\t}\n\t\tif len(result.Collections) != 2 {\n\t\t\tt.Errorf(\"Expected 2 collections, got %d\", len(result.Collections))\n\t\t}\n\t})\n\n\tt.Run(\"StringSlice\", func(t *testing.T) {\n\t\tcollections := []string{\"col1\", \"col2\", \"col3\"}\n\t\tresult, err := ToKnowledgeBase(collections)\n\t\tif err != nil {\n\t\t\tt.Errorf(\"Expected no error, got: %v\", err)\n\t\t}\n\t\tif len(result.Collections) != 3 {\n\t\t\tt.Errorf(\"Expected 3 collections, got %d\", len(result.Collections))\n\t\t}\n\t\tif result.Collections[0] != \"col1\" {\n\t\t\tt.Errorf(\"Expected 'col1', got '%s'\", result.Collections[0])\n\t\t}\n\t})\n\n\tt.Run(\"InterfaceSlice\", func(t *testing.T) {\n\t\tcollections := []interface{}{\"col1\", \"col2\", 123}\n\t\tresult, err := ToKnowledgeBase(collections)\n\t\tif err != nil {\n\t\t\tt.Errorf(\"Expected no error, got: %v\", err)\n\t\t}\n\t\tif len(result.Collections) != 3 {\n\t\t\tt.Errorf(\"Expected 3 collections, got %d\", len(result.Collections))\n\t\t}\n\t\tif result.Collections[2] != \"123\" {\n\t\t\tt.Errorf(\"Expected '123', got '%s'\", result.Collections[2])\n\t\t}\n\t})\n\n\tt.Run(\"MapInput\", func(t *testing.T) {\n\t\tdata := map[string]interface{}{\n\t\t\t\"collections\": []string{\"col1\", \"col2\"},\n\t\t}\n\t\tresult, err := ToKnowledgeBase(data)\n\t\tif err != nil {\n\t\t\tt.Errorf(\"Expected no error, got: %v\", err)\n\t\t}\n\t\tif len(result.Collections) != 2 {\n\t\t\tt.Errorf(\"Expected 2 collections, got %d\", len(result.Collections))\n\t\t}\n\t})\n\n\tt.Run(\"InvalidInput\", func(t *testing.T) {\n\t\t// Test with data that can't be marshaled\n\t\tinvalidData := make(chan int)\n\t\t_, err := ToKnowledgeBase(invalidData)\n\t\tif err == nil {\n\t\t\tt.Error(\"Expected error for invalid input\")\n\t\t}\n\t})\n\n\tt.Run(\"InvalidJSONUnmarshal\", func(t *testing.T) {\n\t\t// Test with data that marshals but can't unmarshal to KnowledgeBase\n\t\tdata := map[string]interface{}{\n\t\t\t\"invalid_field\": \"should cause unmarshal to fail gracefully\",\n\t\t}\n\t\tresult, err := ToKnowledgeBase(data)\n\t\t// Should not error, just return empty KnowledgeBase\n\t\tif err != nil {\n\t\t\tt.Errorf(\"Expected no error, got: %v\", err)\n\t\t}\n\t\tif result == nil {\n\t\t\tt.Error(\"Expected non-nil result\")\n\t\t}\n\t})\n}\n\n// TestToMCPServers tests the ToMCPServers conversion function\nfunc TestToMCPServers(t *testing.T) {\n\tt.Run(\"NilInput\", func(t *testing.T) {\n\t\tresult, err := ToMCPServers(nil)\n\t\tif err != nil {\n\t\t\tt.Errorf(\"Expected no error, got: %v\", err)\n\t\t}\n\t\tif result != nil {\n\t\t\tt.Errorf(\"Expected nil result, got: %v\", result)\n\t\t}\n\t})\n\n\tt.Run(\"MCPServersPointer\", func(t *testing.T) {\n\t\tmcp := &MCPServers{Servers: []MCPServerConfig{{ServerID: \"server1\"}, {ServerID: \"server2\"}}}\n\t\tresult, err := ToMCPServers(mcp)\n\t\tif err != nil {\n\t\t\tt.Errorf(\"Expected no error, got: %v\", err)\n\t\t}\n\t\tif result != mcp {\n\t\t\tt.Errorf(\"Expected same pointer\")\n\t\t}\n\t})\n\n\tt.Run(\"MCPServersValue\", func(t *testing.T) {\n\t\tmcp := MCPServers{Servers: []MCPServerConfig{{ServerID: \"server1\"}, {ServerID: \"server2\"}}}\n\t\tresult, err := ToMCPServers(mcp)\n\t\tif err != nil {\n\t\t\tt.Errorf(\"Expected no error, got: %v\", err)\n\t\t}\n\t\tif len(result.Servers) != 2 {\n\t\t\tt.Errorf(\"Expected 2 servers, got %d\", len(result.Servers))\n\t\t}\n\t})\n\n\tt.Run(\"MapInput\", func(t *testing.T) {\n\t\tdata := map[string]interface{}{\n\t\t\t\"servers\": []interface{}{\"server1\", \"server2\"},\n\t\t}\n\t\tresult, err := ToMCPServers(data)\n\t\tif err != nil {\n\t\t\tt.Errorf(\"Expected no error, got: %v\", err)\n\t\t}\n\t\tif len(result.Servers) != 2 {\n\t\t\tt.Errorf(\"Expected 2 servers, got %d\", len(result.Servers))\n\t\t}\n\t\tif result.Servers[0].ServerID != \"server1\" {\n\t\t\tt.Errorf(\"Expected 'server1', got '%s'\", result.Servers[0].ServerID)\n\t\t}\n\t})\n\n\tt.Run(\"InvalidInput\", func(t *testing.T) {\n\t\t// Test with data that can't be marshaled\n\t\tinvalidData := make(chan int)\n\t\t_, err := ToMCPServers(invalidData)\n\t\tif err == nil {\n\t\t\tt.Error(\"Expected error for invalid input\")\n\t\t}\n\t})\n\n\tt.Run(\"InvalidJSONUnmarshal\", func(t *testing.T) {\n\t\t// Test with data that marshals but can't unmarshal to MCPServers\n\t\tdata := map[string]interface{}{\n\t\t\t\"invalid_field\": \"should cause unmarshal to fail gracefully\",\n\t\t}\n\t\tresult, err := ToMCPServers(data)\n\t\t// Should not error, just return empty MCPServers\n\t\tif err != nil {\n\t\t\tt.Errorf(\"Expected no error, got: %v\", err)\n\t\t}\n\t\tif result == nil {\n\t\t\tt.Error(\"Expected non-nil result\")\n\t\t}\n\t})\n}\n\n// TestToWorkflow tests the ToWorkflow conversion function\nfunc TestToWorkflow(t *testing.T) {\n\tt.Run(\"NilInput\", func(t *testing.T) {\n\t\tresult, err := ToWorkflow(nil)\n\t\tif err != nil {\n\t\t\tt.Errorf(\"Expected no error, got: %v\", err)\n\t\t}\n\t\tif result != nil {\n\t\t\tt.Errorf(\"Expected nil result, got: %v\", result)\n\t\t}\n\t})\n\n\tt.Run(\"WorkflowPointer\", func(t *testing.T) {\n\t\twf := &Workflow{Workflows: []string{\"wf1\", \"wf2\"}}\n\t\tresult, err := ToWorkflow(wf)\n\t\tif err != nil {\n\t\t\tt.Errorf(\"Expected no error, got: %v\", err)\n\t\t}\n\t\tif result != wf {\n\t\t\tt.Errorf(\"Expected same pointer\")\n\t\t}\n\t})\n\n\tt.Run(\"WorkflowValue\", func(t *testing.T) {\n\t\twf := Workflow{Workflows: []string{\"wf1\", \"wf2\"}}\n\t\tresult, err := ToWorkflow(wf)\n\t\tif err != nil {\n\t\t\tt.Errorf(\"Expected no error, got: %v\", err)\n\t\t}\n\t\tif len(result.Workflows) != 2 {\n\t\t\tt.Errorf(\"Expected 2 workflows, got %d\", len(result.Workflows))\n\t\t}\n\t})\n\n\tt.Run(\"StringSlice\", func(t *testing.T) {\n\t\tworkflows := []string{\"wf1\", \"wf2\", \"wf3\"}\n\t\tresult, err := ToWorkflow(workflows)\n\t\tif err != nil {\n\t\t\tt.Errorf(\"Expected no error, got: %v\", err)\n\t\t}\n\t\tif len(result.Workflows) != 3 {\n\t\t\tt.Errorf(\"Expected 3 workflows, got %d\", len(result.Workflows))\n\t\t}\n\t\tif result.Workflows[0] != \"wf1\" {\n\t\t\tt.Errorf(\"Expected 'wf1', got '%s'\", result.Workflows[0])\n\t\t}\n\t})\n\n\tt.Run(\"InterfaceSlice\", func(t *testing.T) {\n\t\tworkflows := []interface{}{\"wf1\", \"wf2\", 789}\n\t\tresult, err := ToWorkflow(workflows)\n\t\tif err != nil {\n\t\t\tt.Errorf(\"Expected no error, got: %v\", err)\n\t\t}\n\t\tif len(result.Workflows) != 3 {\n\t\t\tt.Errorf(\"Expected 3 workflows, got %d\", len(result.Workflows))\n\t\t}\n\t\tif result.Workflows[2] != \"789\" {\n\t\t\tt.Errorf(\"Expected '789', got '%s'\", result.Workflows[2])\n\t\t}\n\t})\n\n\tt.Run(\"MapInput\", func(t *testing.T) {\n\t\tdata := map[string]interface{}{\n\t\t\t\"workflows\": []string{\"wf1\", \"wf2\"},\n\t\t}\n\t\tresult, err := ToWorkflow(data)\n\t\tif err != nil {\n\t\t\tt.Errorf(\"Expected no error, got: %v\", err)\n\t\t}\n\t\tif len(result.Workflows) != 2 {\n\t\t\tt.Errorf(\"Expected 2 workflows, got %d\", len(result.Workflows))\n\t\t}\n\t})\n\n\tt.Run(\"InvalidInput\", func(t *testing.T) {\n\t\t// Test with data that can't be marshaled\n\t\tinvalidData := make(chan int)\n\t\t_, err := ToWorkflow(invalidData)\n\t\tif err == nil {\n\t\t\tt.Error(\"Expected error for invalid input\")\n\t\t}\n\t})\n\n\tt.Run(\"InvalidJSONUnmarshal\", func(t *testing.T) {\n\t\t// Test with data that marshals but can't unmarshal to Workflow\n\t\tdata := map[string]interface{}{\n\t\t\t\"invalid_field\": \"should cause unmarshal to fail gracefully\",\n\t\t}\n\t\tresult, err := ToWorkflow(data)\n\t\t// Should not error, just return empty Workflow\n\t\tif err != nil {\n\t\t\tt.Errorf(\"Expected no error, got: %v\", err)\n\t\t}\n\t\tif result == nil {\n\t\t\tt.Error(\"Expected non-nil result\")\n\t\t}\n\t})\n}\n\n// TestToMySQLTime tests the ToMySQLTime conversion function\nfunc TestToMySQLTime(t *testing.T) {\n\tt.Run(\"Int64Zero\", func(t *testing.T) {\n\t\tresult := ToMySQLTime(int64(0))\n\t\tif result != \"0000-00-00 00:00:00\" {\n\t\t\tt.Errorf(\"Expected '0000-00-00 00:00:00', got '%s'\", result)\n\t\t}\n\t})\n\n\tt.Run(\"Int64Timestamp\", func(t *testing.T) {\n\t\t// Unix timestamp in nanoseconds: 1609459200000000000 = 2021-01-01 00:00:00 UTC\n\t\ttimestamp := int64(1609459200000000000)\n\t\tresult := ToMySQLTime(timestamp)\n\t\t// Should be in format \"2021-01-01 00:00:00\" or similar depending on timezone\n\t\tif len(result) != 19 {\n\t\t\tt.Errorf(\"Expected 19 character timestamp, got %d: '%s'\", len(result), result)\n\t\t}\n\t})\n\n\tt.Run(\"IntZero\", func(t *testing.T) {\n\t\tresult := ToMySQLTime(int(0))\n\t\tif result != \"0000-00-00 00:00:00\" {\n\t\t\tt.Errorf(\"Expected '0000-00-00 00:00:00', got '%s'\", result)\n\t\t}\n\t})\n\n\tt.Run(\"IntTimestamp\", func(t *testing.T) {\n\t\ttimestamp := int(1609459200000000000)\n\t\tresult := ToMySQLTime(timestamp)\n\t\tif len(result) != 19 {\n\t\t\tt.Errorf(\"Expected 19 character timestamp, got %d: '%s'\", len(result), result)\n\t\t}\n\t})\n\n\tt.Run(\"StringMySQLFormat\", func(t *testing.T) {\n\t\tmysqlTime := \"2021-01-01 12:30:45\"\n\t\tresult := ToMySQLTime(mysqlTime)\n\t\tif result != mysqlTime {\n\t\t\tt.Errorf(\"Expected '%s', got '%s'\", mysqlTime, result)\n\t\t}\n\t})\n\n\tt.Run(\"StringRFC3339\", func(t *testing.T) {\n\t\trfc3339Time := \"2021-01-01T12:30:45Z\"\n\t\tresult := ToMySQLTime(rfc3339Time)\n\t\texpected := \"2021-01-01 12:30:45\"\n\t\tif result != expected {\n\t\t\tt.Errorf(\"Expected '%s', got '%s'\", expected, result)\n\t\t}\n\t})\n\n\tt.Run(\"StringUnixTimestamp\", func(t *testing.T) {\n\t\t// Unix timestamp in seconds as string\n\t\tresult := ToMySQLTime(\"1609459200000000000\")\n\t\tif len(result) != 19 {\n\t\t\tt.Errorf(\"Expected 19 character timestamp, got %d: '%s'\", len(result), result)\n\t\t}\n\t})\n\n\tt.Run(\"StringInvalidFormat\", func(t *testing.T) {\n\t\tinvalidTime := \"not-a-valid-time\"\n\t\tresult := ToMySQLTime(invalidTime)\n\t\t// Should return the original string when it can't be parsed\n\t\tif result != invalidTime {\n\t\t\tt.Errorf(\"Expected '%s', got '%s'\", invalidTime, result)\n\t\t}\n\t})\n\n\tt.Run(\"TimeZero\", func(t *testing.T) {\n\t\tzeroTime := time.Time{}\n\t\tresult := ToMySQLTime(zeroTime)\n\t\tif result != \"0000-00-00 00:00:00\" {\n\t\t\tt.Errorf(\"Expected '0000-00-00 00:00:00', got '%s'\", result)\n\t\t}\n\t})\n\n\tt.Run(\"TimeNormal\", func(t *testing.T) {\n\t\tnormalTime := time.Date(2021, 1, 1, 12, 30, 45, 0, time.UTC)\n\t\tresult := ToMySQLTime(normalTime)\n\t\texpected := \"2021-01-01 12:30:45\"\n\t\tif result != expected {\n\t\t\tt.Errorf(\"Expected '%s', got '%s'\", expected, result)\n\t\t}\n\t})\n\n\tt.Run(\"NilInput\", func(t *testing.T) {\n\t\tresult := ToMySQLTime(nil)\n\t\tif result != \"0000-00-00 00:00:00\" {\n\t\t\tt.Errorf(\"Expected '0000-00-00 00:00:00', got '%s'\", result)\n\t\t}\n\t})\n\n\tt.Run(\"UnknownType\", func(t *testing.T) {\n\t\t// Test with unsupported type\n\t\tresult := ToMySQLTime(struct{}{})\n\t\tif result != \"0000-00-00 00:00:00\" {\n\t\t\tt.Errorf(\"Expected '0000-00-00 00:00:00', got '%s'\", result)\n\t\t}\n\t})\n}\n\n// TestToAssistantModel tests the ToAssistantModel conversion function\nfunc TestToAssistantModel(t *testing.T) {\n\tt.Run(\"NilInput\", func(t *testing.T) {\n\t\tresult, err := ToAssistantModel(nil)\n\t\tif err != nil {\n\t\t\tt.Errorf(\"Expected no error, got: %v\", err)\n\t\t}\n\t\tif result != nil {\n\t\t\tt.Errorf(\"Expected nil result, got: %v\", result)\n\t\t}\n\t})\n\n\tt.Run(\"AssistantModelPointer\", func(t *testing.T) {\n\t\tmodel := &AssistantModel{\n\t\t\tID: \"test-id\",\n\t\t\tName: \"Test Assistant\",\n\t\t}\n\t\tresult, err := ToAssistantModel(model)\n\t\tif err != nil {\n\t\t\tt.Errorf(\"Expected no error, got: %v\", err)\n\t\t}\n\t\tif result != model {\n\t\t\tt.Errorf(\"Expected same pointer\")\n\t\t}\n\t})\n\n\tt.Run(\"AssistantModelValue\", func(t *testing.T) {\n\t\tmodel := AssistantModel{\n\t\t\tID: \"test-id\",\n\t\t\tName: \"Test Assistant\",\n\t\t}\n\t\tresult, err := ToAssistantModel(model)\n\t\tif err != nil {\n\t\t\tt.Errorf(\"Expected no error, got: %v\", err)\n\t\t}\n\t\tif result.ID != \"test-id\" {\n\t\t\tt.Errorf(\"Expected 'test-id', got '%s'\", result.ID)\n\t\t}\n\t})\n\n\tt.Run(\"MapWithAllFields\", func(t *testing.T) {\n\t\tdata := map[string]interface{}{\n\t\t\t\"assistant_id\": \"test-id\",\n\t\t\t\"type\": \"assistant\",\n\t\t\t\"name\": \"Test Assistant\",\n\t\t\t\"avatar\": \"https://example.com/avatar.png\",\n\t\t\t\"connector\": \"openai\",\n\t\t\t\"connector_options\": map[string]interface{}{\n\t\t\t\t\"optional\": true,\n\t\t\t\t\"connectors\": []string{\"openai\", \"anthropic\"},\n\t\t\t\t\"filters\": []string{\"vision\", \"tool_calls\"},\n\t\t\t},\n\t\t\t\"path\": \"/path/to/assistant\",\n\t\t\t\"description\": \"Test description\",\n\t\t\t\"share\": \"team\",\n\t\t\t\"built_in\": true,\n\t\t\t\"readonly\": false,\n\t\t\t\"public\": true,\n\t\t\t\"mentionable\": true,\n\t\t\t\"automated\": false,\n\t\t\t\"sort\": 100,\n\t\t\t\"created_at\": int64(1609459200),\n\t\t\t\"updated_at\": int64(1609459300),\n\t\t\t\"tags\": []string{\"tag1\", \"tag2\"},\n\t\t\t\"modes\": []string{\"chat\", \"task\"},\n\t\t\t\"default_mode\": \"chat\",\n\t\t\t\"options\": map[string]interface{}{\n\t\t\t\t\"temperature\": 0.7,\n\t\t\t},\n\t\t\t\"prompts\": []map[string]interface{}{\n\t\t\t\t{\"role\": \"system\", \"content\": \"You are helpful\"},\n\t\t\t},\n\t\t\t\"prompt_presets\": map[string]interface{}{\n\t\t\t\t\"chat\": []map[string]interface{}{\n\t\t\t\t\t{\"role\": \"system\", \"content\": \"You are a chat assistant\"},\n\t\t\t\t},\n\t\t\t\t\"task\": []map[string]interface{}{\n\t\t\t\t\t{\"role\": \"system\", \"content\": \"You are a task assistant\"},\n\t\t\t\t},\n\t\t\t},\n\t\t\t\"disable_global_prompts\": true,\n\t\t\t\"source\": \"function hook() { return 'test'; }\",\n\t\t\t\"kb\": map[string]interface{}{\n\t\t\t\t\"collections\": []string{\"col1\"},\n\t\t\t},\n\t\t\t\"db\": map[string]interface{}{\n\t\t\t\t\"models\": []string{\"model1\"},\n\t\t\t},\n\t\t\t\"mcp\": map[string]interface{}{\n\t\t\t\t\"servers\": []string{\"server1\"},\n\t\t\t},\n\t\t\t\"workflow\": map[string]interface{}{\n\t\t\t\t\"workflows\": []string{\"wf1\"},\n\t\t\t},\n\t\t\t\"placeholder\": map[string]interface{}{\n\t\t\t\t\"title\": \"Enter message\",\n\t\t\t},\n\t\t\t\"locales\": map[string]interface{}{\n\t\t\t\t\"en\": map[string]interface{}{\n\t\t\t\t\t\"name\": \"English Name\",\n\t\t\t\t},\n\t\t\t},\n\t\t\t\"dependencies\": map[string]interface{}{\n\t\t\t\t\"echo\": \"^1.0.0\",\n\t\t\t\t\"customer\": \">=2.0.0\",\n\t\t\t},\n\t\t}\n\n\t\tresult, err := ToAssistantModel(data)\n\t\tif err != nil {\n\t\t\tt.Errorf(\"Expected no error, got: %v\", err)\n\t\t}\n\n\t\t// Verify all fields\n\t\tif result.ID != \"test-id\" {\n\t\t\tt.Errorf(\"Expected ID 'test-id', got '%s'\", result.ID)\n\t\t}\n\t\tif result.Type != \"assistant\" {\n\t\t\tt.Errorf(\"Expected Type 'assistant', got '%s'\", result.Type)\n\t\t}\n\t\tif result.Name != \"Test Assistant\" {\n\t\t\tt.Errorf(\"Expected Name 'Test Assistant', got '%s'\", result.Name)\n\t\t}\n\t\tif result.Avatar != \"https://example.com/avatar.png\" {\n\t\t\tt.Errorf(\"Expected Avatar URL, got '%s'\", result.Avatar)\n\t\t}\n\t\tif result.Connector != \"openai\" {\n\t\t\tt.Errorf(\"Expected Connector 'openai', got '%s'\", result.Connector)\n\t\t}\n\t\tif result.ConnectorOptions == nil {\n\t\t\tt.Error(\"Expected ConnectorOptions to be set\")\n\t\t} else {\n\t\t\tif result.ConnectorOptions.Optional == nil || !*result.ConnectorOptions.Optional {\n\t\t\t\tt.Error(\"Expected ConnectorOptions.Optional to be true\")\n\t\t\t}\n\t\t\tif len(result.ConnectorOptions.Connectors) != 2 {\n\t\t\t\tt.Errorf(\"Expected 2 connectors in options, got %d\", len(result.ConnectorOptions.Connectors))\n\t\t\t}\n\t\t\tif len(result.ConnectorOptions.Filters) != 2 {\n\t\t\t\tt.Errorf(\"Expected 2 filters, got %d\", len(result.ConnectorOptions.Filters))\n\t\t\t}\n\t\t}\n\t\tif result.Path != \"/path/to/assistant\" {\n\t\t\tt.Errorf(\"Expected Path, got '%s'\", result.Path)\n\t\t}\n\t\tif result.Source != \"function hook() { return 'test'; }\" {\n\t\t\tt.Errorf(\"Expected Source, got '%s'\", result.Source)\n\t\t}\n\t\tif result.Description != \"Test description\" {\n\t\t\tt.Errorf(\"Expected Description, got '%s'\", result.Description)\n\t\t}\n\t\tif result.Share != \"team\" {\n\t\t\tt.Errorf(\"Expected Share 'team', got '%s'\", result.Share)\n\t\t}\n\t\tif !result.BuiltIn {\n\t\t\tt.Error(\"Expected BuiltIn to be true\")\n\t\t}\n\t\tif result.Readonly {\n\t\t\tt.Error(\"Expected Readonly to be false\")\n\t\t}\n\t\tif !result.Public {\n\t\t\tt.Error(\"Expected Public to be true\")\n\t\t}\n\t\tif !result.Mentionable {\n\t\t\tt.Error(\"Expected Mentionable to be true\")\n\t\t}\n\t\tif result.Automated {\n\t\t\tt.Error(\"Expected Automated to be false\")\n\t\t}\n\t\tif result.Sort != 100 {\n\t\t\tt.Errorf(\"Expected Sort 100, got %d\", result.Sort)\n\t\t}\n\t\tif result.CreatedAt != 1609459200 {\n\t\t\tt.Errorf(\"Expected CreatedAt 1609459200, got %d\", result.CreatedAt)\n\t\t}\n\t\tif result.UpdatedAt != 1609459300 {\n\t\t\tt.Errorf(\"Expected UpdatedAt 1609459300, got %d\", result.UpdatedAt)\n\t\t}\n\t\tif len(result.Tags) != 2 {\n\t\t\tt.Errorf(\"Expected 2 tags, got %d\", len(result.Tags))\n\t\t}\n\t\tif len(result.Modes) != 2 {\n\t\t\tt.Errorf(\"Expected 2 modes, got %d\", len(result.Modes))\n\t\t}\n\t\tif result.Modes[0] != \"chat\" {\n\t\t\tt.Errorf(\"Expected first mode 'chat', got '%s'\", result.Modes[0])\n\t\t}\n\t\tif result.DefaultMode != \"chat\" {\n\t\t\tt.Errorf(\"Expected default_mode 'chat', got '%s'\", result.DefaultMode)\n\t\t}\n\t\tif result.Options == nil {\n\t\t\tt.Error(\"Expected Options to be set\")\n\t\t}\n\t\tif len(result.Prompts) != 1 {\n\t\t\tt.Errorf(\"Expected 1 prompt, got %d\", len(result.Prompts))\n\t\t}\n\t\tif result.PromptPresets == nil {\n\t\t\tt.Error(\"Expected PromptPresets to be set\")\n\t\t} else {\n\t\t\tif len(result.PromptPresets) != 2 {\n\t\t\t\tt.Errorf(\"Expected 2 prompt presets, got %d\", len(result.PromptPresets))\n\t\t\t}\n\t\t\tif chatPrompts, ok := result.PromptPresets[\"chat\"]; !ok {\n\t\t\t\tt.Error(\"Expected 'chat' prompt preset\")\n\t\t\t} else if len(chatPrompts) != 1 {\n\t\t\t\tt.Errorf(\"Expected 1 chat prompt, got %d\", len(chatPrompts))\n\t\t\t}\n\t\t\tif taskPrompts, ok := result.PromptPresets[\"task\"]; !ok {\n\t\t\t\tt.Error(\"Expected 'task' prompt preset\")\n\t\t\t} else if len(taskPrompts) != 1 {\n\t\t\t\tt.Errorf(\"Expected 1 task prompt, got %d\", len(taskPrompts))\n\t\t\t}\n\t\t}\n\t\tif !result.DisableGlobalPrompts {\n\t\t\tt.Error(\"Expected DisableGlobalPrompts to be true\")\n\t\t}\n\t\tif result.KB == nil {\n\t\t\tt.Error(\"Expected KB to be set\")\n\t\t}\n\t\tif result.DB == nil {\n\t\t\tt.Error(\"Expected DB to be set\")\n\t\t}\n\t\tif result.MCP == nil {\n\t\t\tt.Error(\"Expected MCP to be set\")\n\t\t}\n\t\tif result.Workflow == nil {\n\t\t\tt.Error(\"Expected Workflow to be set\")\n\t\t}\n\t\tif result.Placeholder == nil {\n\t\t\tt.Error(\"Expected Placeholder to be set\")\n\t\t}\n\t\tif result.Locales == nil {\n\t\t\tt.Error(\"Expected Locales to be set\")\n\t\t}\n\t\tif result.Dependencies == nil {\n\t\t\tt.Error(\"Expected Dependencies to be set\")\n\t\t} else {\n\t\t\tif len(result.Dependencies) != 2 {\n\t\t\t\tt.Errorf(\"Expected 2 dependencies, got %d\", len(result.Dependencies))\n\t\t\t}\n\t\t\tif result.Dependencies[\"echo\"] != \"^1.0.0\" {\n\t\t\t\tt.Errorf(\"Expected echo dependency '^1.0.0', got '%s'\", result.Dependencies[\"echo\"])\n\t\t\t}\n\t\t\tif result.Dependencies[\"customer\"] != \">=2.0.0\" {\n\t\t\t\tt.Errorf(\"Expected customer dependency '>=2.0.0', got '%s'\", result.Dependencies[\"customer\"])\n\t\t\t}\n\t\t}\n\t})\n\n\tt.Run(\"MapWithFloatNumbers\", func(t *testing.T) {\n\t\tdata := map[string]interface{}{\n\t\t\t\"sort\": float64(150),\n\t\t\t\"created_at\": float64(1609459200),\n\t\t\t\"updated_at\": float64(1609459300),\n\t\t}\n\n\t\tresult, err := ToAssistantModel(data)\n\t\tif err != nil {\n\t\t\tt.Errorf(\"Expected no error, got: %v\", err)\n\t\t}\n\n\t\tif result.Sort != 150 {\n\t\t\tt.Errorf(\"Expected Sort 150, got %d\", result.Sort)\n\t\t}\n\t\tif result.CreatedAt != 1609459200 {\n\t\t\tt.Errorf(\"Expected CreatedAt 1609459200, got %d\", result.CreatedAt)\n\t\t}\n\t\tif result.UpdatedAt != 1609459300 {\n\t\t\tt.Errorf(\"Expected UpdatedAt 1609459300, got %d\", result.UpdatedAt)\n\t\t}\n\t})\n\n\tt.Run(\"MapWithNilFields\", func(t *testing.T) {\n\t\tdata := map[string]interface{}{\n\t\t\t\"assistant_id\": \"test-id\",\n\t\t\t\"tags\": nil,\n\t\t\t\"modes\": nil,\n\t\t\t\"default_mode\": \"\",\n\t\t\t\"options\": nil,\n\t\t\t\"prompts\": nil,\n\t\t\t\"kb\": nil,\n\t\t\t\"db\": nil,\n\t\t\t\"mcp\": nil,\n\t\t\t\"workflow\": nil,\n\t\t\t\"placeholder\": nil,\n\t\t\t\"locales\": nil,\n\t\t\t\"dependencies\": nil,\n\t\t}\n\n\t\tresult, err := ToAssistantModel(data)\n\t\tif err != nil {\n\t\t\tt.Errorf(\"Expected no error, got: %v\", err)\n\t\t}\n\n\t\tif result.ID != \"test-id\" {\n\t\t\tt.Errorf(\"Expected ID 'test-id', got '%s'\", result.ID)\n\t\t}\n\t\t// All nil fields should remain nil\n\t\tif result.Tags != nil {\n\t\t\tt.Error(\"Expected Tags to be nil\")\n\t\t}\n\t\tif result.Dependencies != nil {\n\t\t\tt.Error(\"Expected Dependencies to be nil\")\n\t\t}\n\t\tif result.Modes != nil {\n\t\t\tt.Error(\"Expected Modes to be nil\")\n\t\t}\n\t\tif result.DefaultMode != \"\" {\n\t\t\tt.Error(\"Expected DefaultMode to be empty\")\n\t\t}\n\t\tif result.Options != nil {\n\t\t\tt.Error(\"Expected Options to be nil\")\n\t\t}\n\t})\n\n\tt.Run(\"StructInput\", func(t *testing.T) {\n\t\ttype CustomStruct struct {\n\t\t\tAssistantID string `json:\"assistant_id\"`\n\t\t\tName string `json:\"name\"`\n\t\t\tType string `json:\"type\"`\n\t\t}\n\n\t\tinput := CustomStruct{\n\t\t\tAssistantID: \"custom-id\",\n\t\t\tName: \"Custom Assistant\",\n\t\t\tType: \"bot\",\n\t\t}\n\n\t\tresult, err := ToAssistantModel(input)\n\t\tif err != nil {\n\t\t\tt.Errorf(\"Expected no error, got: %v\", err)\n\t\t}\n\n\t\tif result.ID != \"custom-id\" {\n\t\t\tt.Errorf(\"Expected ID 'custom-id', got '%s'\", result.ID)\n\t\t}\n\t\tif result.Name != \"Custom Assistant\" {\n\t\t\tt.Errorf(\"Expected Name 'Custom Assistant', got '%s'\", result.Name)\n\t\t}\n\t\tif result.Type != \"bot\" {\n\t\t\tt.Errorf(\"Expected Type 'bot', got '%s'\", result.Type)\n\t\t}\n\t})\n\n\tt.Run(\"InvalidInput\", func(t *testing.T) {\n\t\t// Test with data that can't be marshaled\n\t\tinvalidData := make(chan int)\n\t\t_, err := ToAssistantModel(invalidData)\n\t\tif err == nil {\n\t\t\tt.Error(\"Expected error for invalid input\")\n\t\t}\n\t})\n\n\tt.Run(\"EmptyMap\", func(t *testing.T) {\n\t\tdata := map[string]interface{}{}\n\t\tresult, err := ToAssistantModel(data)\n\t\tif err != nil {\n\t\t\tt.Errorf(\"Expected no error, got: %v\", err)\n\t\t}\n\t\tif result == nil {\n\t\t\tt.Error(\"Expected non-nil result\")\n\t\t}\n\t\t// All fields should have default values\n\t\tif result.ID != \"\" {\n\t\t\tt.Errorf(\"Expected empty ID, got '%s'\", result.ID)\n\t\t}\n\t})\n}\n\n// TestToAssistantModelNewFields tests the newly added fields\nfunc TestToAssistantModelNewFields(t *testing.T) {\n\tt.Run(\"ConnectorOptions\", func(t *testing.T) {\n\t\tdata := map[string]interface{}{\n\t\t\t\"connector_options\": map[string]interface{}{\n\t\t\t\t\"optional\": true,\n\t\t\t\t\"connectors\": []string{\"openai\", \"anthropic\", \"azure\"},\n\t\t\t\t\"filters\": []string{\"vision\", \"tool_calls\", \"audio\"},\n\t\t\t},\n\t\t}\n\n\t\tresult, err := ToAssistantModel(data)\n\t\tif err != nil {\n\t\t\tt.Errorf(\"Expected no error, got: %v\", err)\n\t\t}\n\n\t\tif result.ConnectorOptions == nil {\n\t\t\tt.Fatal(\"Expected ConnectorOptions to be set\")\n\t\t}\n\n\t\tif result.ConnectorOptions.Optional == nil || !*result.ConnectorOptions.Optional {\n\t\t\tt.Error(\"Expected Optional to be true\")\n\t\t}\n\n\t\tif len(result.ConnectorOptions.Connectors) != 3 {\n\t\t\tt.Errorf(\"Expected 3 connectors, got %d\", len(result.ConnectorOptions.Connectors))\n\t\t}\n\n\t\tif len(result.ConnectorOptions.Filters) != 3 {\n\t\t\tt.Errorf(\"Expected 3 filters, got %d\", len(result.ConnectorOptions.Filters))\n\t\t}\n\t})\n\n\tt.Run(\"PromptPresets\", func(t *testing.T) {\n\t\tdata := map[string]interface{}{\n\t\t\t\"prompt_presets\": map[string]interface{}{\n\t\t\t\t\"chat\": []map[string]interface{}{\n\t\t\t\t\t{\"role\": \"system\", \"content\": \"You are a helpful chat assistant\"},\n\t\t\t\t\t{\"role\": \"user\", \"content\": \"Example question\"},\n\t\t\t\t},\n\t\t\t\t\"task\": []map[string]interface{}{\n\t\t\t\t\t{\"role\": \"system\", \"content\": \"You are a task completion assistant\"},\n\t\t\t\t},\n\t\t\t\t\"analyze\": []map[string]interface{}{\n\t\t\t\t\t{\"role\": \"system\", \"content\": \"You are a data analysis assistant\"},\n\t\t\t\t},\n\t\t\t},\n\t\t}\n\n\t\tresult, err := ToAssistantModel(data)\n\t\tif err != nil {\n\t\t\tt.Errorf(\"Expected no error, got: %v\", err)\n\t\t}\n\n\t\tif result.PromptPresets == nil {\n\t\t\tt.Fatal(\"Expected PromptPresets to be set\")\n\t\t}\n\n\t\tif len(result.PromptPresets) != 3 {\n\t\t\tt.Errorf(\"Expected 3 prompt preset modes, got %d\", len(result.PromptPresets))\n\t\t}\n\n\t\tif chatPrompts, ok := result.PromptPresets[\"chat\"]; !ok {\n\t\t\tt.Error(\"Expected 'chat' mode in prompt presets\")\n\t\t} else if len(chatPrompts) != 2 {\n\t\t\tt.Errorf(\"Expected 2 prompts in chat mode, got %d\", len(chatPrompts))\n\t\t}\n\n\t\tif taskPrompts, ok := result.PromptPresets[\"task\"]; !ok {\n\t\t\tt.Error(\"Expected 'task' mode in prompt presets\")\n\t\t} else if len(taskPrompts) != 1 {\n\t\t\tt.Errorf(\"Expected 1 prompt in task mode, got %d\", len(taskPrompts))\n\t\t}\n\n\t\tif analyzePrompts, ok := result.PromptPresets[\"analyze\"]; !ok {\n\t\t\tt.Error(\"Expected 'analyze' mode in prompt presets\")\n\t\t} else if len(analyzePrompts) != 1 {\n\t\t\tt.Errorf(\"Expected 1 prompt in analyze mode, got %d\", len(analyzePrompts))\n\t\t}\n\t})\n\n\tt.Run(\"Source\", func(t *testing.T) {\n\t\thookScript := `\nfunction beforeChat(context) {\n console.log('Hook called');\n return context;\n}\n`\n\t\tdata := map[string]interface{}{\n\t\t\t\"source\": hookScript,\n\t\t}\n\n\t\tresult, err := ToAssistantModel(data)\n\t\tif err != nil {\n\t\t\tt.Errorf(\"Expected no error, got: %v\", err)\n\t\t}\n\n\t\tif result.Source != hookScript {\n\t\t\tt.Errorf(\"Expected Source to match, got '%s'\", result.Source)\n\t\t}\n\t})\n\n\tt.Run(\"AllNewFields\", func(t *testing.T) {\n\t\tdata := map[string]interface{}{\n\t\t\t\"connector_options\": map[string]interface{}{\n\t\t\t\t\"optional\": true,\n\t\t\t\t\"connectors\": []string{\"openai\"},\n\t\t\t\t\"filters\": []string{\"vision\"},\n\t\t\t},\n\t\t\t\"prompt_presets\": map[string]interface{}{\n\t\t\t\t\"chat\": []map[string]interface{}{\n\t\t\t\t\t{\"role\": \"system\", \"content\": \"Chat mode\"},\n\t\t\t\t},\n\t\t\t},\n\t\t\t\"disable_global_prompts\": true,\n\t\t\t\"source\": \"function test() {}\",\n\t\t}\n\n\t\tresult, err := ToAssistantModel(data)\n\t\tif err != nil {\n\t\t\tt.Errorf(\"Expected no error, got: %v\", err)\n\t\t}\n\n\t\tif result.ConnectorOptions == nil {\n\t\t\tt.Error(\"Expected ConnectorOptions to be set\")\n\t\t}\n\t\tif result.PromptPresets == nil {\n\t\t\tt.Error(\"Expected PromptPresets to be set\")\n\t\t}\n\t\tif !result.DisableGlobalPrompts {\n\t\t\tt.Error(\"Expected DisableGlobalPrompts to be true\")\n\t\t}\n\t\tif result.Source == \"\" {\n\t\t\tt.Error(\"Expected Source to be set\")\n\t\t}\n\t})\n\n\tt.Run(\"NilNewFields\", func(t *testing.T) {\n\t\tdata := map[string]interface{}{\n\t\t\t\"connector_options\": nil,\n\t\t\t\"prompt_presets\": nil,\n\t\t\t\"disable_global_prompts\": nil,\n\t\t\t\"source\": nil,\n\t\t}\n\n\t\tresult, err := ToAssistantModel(data)\n\t\tif err != nil {\n\t\t\tt.Errorf(\"Expected no error, got: %v\", err)\n\t\t}\n\n\t\tif result.ConnectorOptions != nil {\n\t\t\tt.Error(\"Expected ConnectorOptions to be nil\")\n\t\t}\n\t\tif result.PromptPresets != nil {\n\t\t\tt.Error(\"Expected PromptPresets to be nil\")\n\t\t}\n\t\tif result.DisableGlobalPrompts {\n\t\t\tt.Error(\"Expected DisableGlobalPrompts to be false\")\n\t\t}\n\t\tif result.Source != \"\" {\n\t\t\tt.Error(\"Expected Source to be empty\")\n\t\t}\n\t})\n\n\tt.Run(\"DisableGlobalPrompts\", func(t *testing.T) {\n\t\t// Test with true\n\t\tdata := map[string]interface{}{\n\t\t\t\"disable_global_prompts\": true,\n\t\t}\n\t\tresult, err := ToAssistantModel(data)\n\t\tif err != nil {\n\t\t\tt.Errorf(\"Expected no error, got: %v\", err)\n\t\t}\n\t\tif !result.DisableGlobalPrompts {\n\t\t\tt.Error(\"Expected DisableGlobalPrompts to be true\")\n\t\t}\n\n\t\t// Test with false\n\t\tdata = map[string]interface{}{\n\t\t\t\"disable_global_prompts\": false,\n\t\t}\n\t\tresult, err = ToAssistantModel(data)\n\t\tif err != nil {\n\t\t\tt.Errorf(\"Expected no error, got: %v\", err)\n\t\t}\n\t\tif result.DisableGlobalPrompts {\n\t\t\tt.Error(\"Expected DisableGlobalPrompts to be false\")\n\t\t}\n\n\t\t// Test with int 1\n\t\tdata = map[string]interface{}{\n\t\t\t\"disable_global_prompts\": 1,\n\t\t}\n\t\tresult, err = ToAssistantModel(data)\n\t\tif err != nil {\n\t\t\tt.Errorf(\"Expected no error, got: %v\", err)\n\t\t}\n\t\tif !result.DisableGlobalPrompts {\n\t\t\tt.Error(\"Expected DisableGlobalPrompts to be true for int 1\")\n\t\t}\n\n\t\t// Test with string \"true\"\n\t\tdata = map[string]interface{}{\n\t\t\t\"disable_global_prompts\": \"true\",\n\t\t}\n\t\tresult, err = ToAssistantModel(data)\n\t\tif err != nil {\n\t\t\tt.Errorf(\"Expected no error, got: %v\", err)\n\t\t}\n\t\tif !result.DisableGlobalPrompts {\n\t\t\tt.Error(\"Expected DisableGlobalPrompts to be true for string 'true'\")\n\t\t}\n\t})\n}\n\n// TestToAssistantModelComplexTypes tests complex type conversions in ToAssistantModel\nfunc TestToAssistantModelComplexTypes(t *testing.T) {\n\tt.Run(\"CompleteLocales\", func(t *testing.T) {\n\t\tdata := map[string]interface{}{\n\t\t\t\"locales\": map[string]interface{}{\n\t\t\t\t\"en\": map[string]interface{}{\n\t\t\t\t\t\"locale\": \"en\",\n\t\t\t\t\t\"messages\": map[string]interface{}{\n\t\t\t\t\t\t\"name\": \"English Name\",\n\t\t\t\t\t\t\"description\": \"English Description\",\n\t\t\t\t\t},\n\t\t\t\t},\n\t\t\t\t\"zh\": map[string]interface{}{\n\t\t\t\t\t\"locale\": \"zh\",\n\t\t\t\t\t\"messages\": map[string]interface{}{\n\t\t\t\t\t\t\"name\": \"中文名称\",\n\t\t\t\t\t\t\"description\": \"中文描述\",\n\t\t\t\t\t},\n\t\t\t\t},\n\t\t\t},\n\t\t}\n\n\t\tresult, err := ToAssistantModel(data)\n\t\tif err != nil {\n\t\t\tt.Errorf(\"Expected no error, got: %v\", err)\n\t\t}\n\n\t\tif result.Locales == nil {\n\t\t\tt.Fatal(\"Expected Locales to be set\")\n\t\t}\n\n\t\tif len(result.Locales) != 2 {\n\t\t\tt.Errorf(\"Expected 2 locales, got %d\", len(result.Locales))\n\t\t}\n\t})\n\n\tt.Run(\"ComplexPrompts\", func(t *testing.T) {\n\t\tdata := map[string]interface{}{\n\t\t\t\"prompts\": []interface{}{\n\t\t\t\tmap[string]interface{}{\n\t\t\t\t\t\"role\": \"system\",\n\t\t\t\t\t\"content\": \"You are a helpful assistant\",\n\t\t\t\t},\n\t\t\t\tmap[string]interface{}{\n\t\t\t\t\t\"role\": \"user\",\n\t\t\t\t\t\"content\": \"Hello\",\n\t\t\t\t},\n\t\t\t},\n\t\t}\n\n\t\tresult, err := ToAssistantModel(data)\n\t\tif err != nil {\n\t\t\tt.Errorf(\"Expected no error, got: %v\", err)\n\t\t}\n\n\t\tif len(result.Prompts) != 2 {\n\t\t\tt.Errorf(\"Expected 2 prompts, got %d\", len(result.Prompts))\n\t\t}\n\t})\n}\n\n// TestGetBoolValue tests the getBoolValue helper function\nfunc TestGetBoolValue(t *testing.T) {\n\tt.Run(\"BoolTrue\", func(t *testing.T) {\n\t\tdata := map[string]interface{}{\"key\": true}\n\t\tresult := getBoolValue(data, \"key\")\n\t\tif !result {\n\t\t\tt.Error(\"Expected true\")\n\t\t}\n\t})\n\n\tt.Run(\"BoolFalse\", func(t *testing.T) {\n\t\tdata := map[string]interface{}{\"key\": false}\n\t\tresult := getBoolValue(data, \"key\")\n\t\tif result {\n\t\t\tt.Error(\"Expected false\")\n\t\t}\n\t})\n\n\tt.Run(\"IntNonZero\", func(t *testing.T) {\n\t\tdata := map[string]interface{}{\"key\": 1}\n\t\tresult := getBoolValue(data, \"key\")\n\t\tif !result {\n\t\t\tt.Error(\"Expected true for non-zero int\")\n\t\t}\n\t})\n\n\tt.Run(\"IntZero\", func(t *testing.T) {\n\t\tdata := map[string]interface{}{\"key\": 0}\n\t\tresult := getBoolValue(data, \"key\")\n\t\tif result {\n\t\t\tt.Error(\"Expected false for zero int\")\n\t\t}\n\t})\n\n\tt.Run(\"Int64NonZero\", func(t *testing.T) {\n\t\tdata := map[string]interface{}{\"key\": int64(1)}\n\t\tresult := getBoolValue(data, \"key\")\n\t\tif !result {\n\t\t\tt.Error(\"Expected true for non-zero int64\")\n\t\t}\n\t})\n\n\tt.Run(\"Int64Zero\", func(t *testing.T) {\n\t\tdata := map[string]interface{}{\"key\": int64(0)}\n\t\tresult := getBoolValue(data, \"key\")\n\t\tif result {\n\t\t\tt.Error(\"Expected false for zero int64\")\n\t\t}\n\t})\n\n\tt.Run(\"Float64NonZero\", func(t *testing.T) {\n\t\tdata := map[string]interface{}{\"key\": float64(1.5)}\n\t\tresult := getBoolValue(data, \"key\")\n\t\tif !result {\n\t\t\tt.Error(\"Expected true for non-zero float64\")\n\t\t}\n\t})\n\n\tt.Run(\"Float64Zero\", func(t *testing.T) {\n\t\tdata := map[string]interface{}{\"key\": float64(0)}\n\t\tresult := getBoolValue(data, \"key\")\n\t\tif result {\n\t\t\tt.Error(\"Expected false for zero float64\")\n\t\t}\n\t})\n\n\tt.Run(\"StringTrue\", func(t *testing.T) {\n\t\tdata := map[string]interface{}{\"key\": \"true\"}\n\t\tresult := getBoolValue(data, \"key\")\n\t\tif !result {\n\t\t\tt.Error(\"Expected true for string 'true'\")\n\t\t}\n\t})\n\n\tt.Run(\"StringOne\", func(t *testing.T) {\n\t\tdata := map[string]interface{}{\"key\": \"1\"}\n\t\tresult := getBoolValue(data, \"key\")\n\t\tif !result {\n\t\t\tt.Error(\"Expected true for string '1'\")\n\t\t}\n\t})\n\n\tt.Run(\"StringFalse\", func(t *testing.T) {\n\t\tdata := map[string]interface{}{\"key\": \"false\"}\n\t\tresult := getBoolValue(data, \"key\")\n\t\tif result {\n\t\t\tt.Error(\"Expected false for string 'false'\")\n\t\t}\n\t})\n\n\tt.Run(\"StringOther\", func(t *testing.T) {\n\t\tdata := map[string]interface{}{\"key\": \"other\"}\n\t\tresult := getBoolValue(data, \"key\")\n\t\tif result {\n\t\t\tt.Error(\"Expected false for other string values\")\n\t\t}\n\t})\n\n\tt.Run(\"NilValue\", func(t *testing.T) {\n\t\tdata := map[string]interface{}{\"key\": nil}\n\t\tresult := getBoolValue(data, \"key\")\n\t\tif result {\n\t\t\tt.Error(\"Expected false for nil value\")\n\t\t}\n\t})\n\n\tt.Run(\"MissingKey\", func(t *testing.T) {\n\t\tdata := map[string]interface{}{}\n\t\tresult := getBoolValue(data, \"missing\")\n\t\tif result {\n\t\t\tt.Error(\"Expected false for missing key\")\n\t\t}\n\t})\n\n\tt.Run(\"UnsupportedType\", func(t *testing.T) {\n\t\tdata := map[string]interface{}{\"key\": struct{}{}}\n\t\tresult := getBoolValue(data, \"key\")\n\t\tif result {\n\t\t\tt.Error(\"Expected false for unsupported type\")\n\t\t}\n\t})\n}\n\n// TestModelID tests the AssistantModel.ModelID method\nfunc TestModelID(t *testing.T) {\n\tt.Run(\"WithCustomModel\", func(t *testing.T) {\n\t\tassistant := AssistantModel{\n\t\t\tID: \"test123\",\n\t\t\tName: \"Test Assistant\",\n\t\t\tConnector: \"openai\",\n\t\t\tOptions: map[string]interface{}{\n\t\t\t\t\"model\": \"gpt-4o\",\n\t\t\t},\n\t\t}\n\t\tresult := assistant.ModelID()\n\t\texpected := \"test-assistant-gpt-4o-yao_test123\"\n\t\tif result != expected {\n\t\t\tt.Errorf(\"Expected '%s', got '%s'\", expected, result)\n\t\t}\n\t})\n\n\tt.Run(\"WithModelInOptions\", func(t *testing.T) {\n\t\tassistant := AssistantModel{\n\t\t\tID: \"abc456\",\n\t\t\tName: \"My Bot\",\n\t\t\tConnector: \"openai\",\n\t\t\tOptions: map[string]interface{}{\n\t\t\t\t\"model\": \"gpt-3.5-turbo\",\n\t\t\t},\n\t\t}\n\t\tresult := assistant.ModelID()\n\t\texpected := \"my-bot-gpt-3.5-turbo-yao_abc456\"\n\t\tif result != expected {\n\t\t\tt.Errorf(\"Expected '%s', got '%s'\", expected, result)\n\t\t}\n\t})\n\n\tt.Run(\"WithoutCustomModel\", func(t *testing.T) {\n\t\tassistant := AssistantModel{\n\t\t\tID: \"xyz789\",\n\t\t\tName: \"Default Assistant\",\n\t\t\tConnector: \"openai\",\n\t\t}\n\t\tresult := assistant.ModelID()\n\t\t// When connector is not loaded in test, it should return unknown\n\t\texpected := \"default-assistant-unknown-yao_xyz789\"\n\t\tif result != expected {\n\t\t\tt.Errorf(\"Expected '%s', got '%s'\", expected, result)\n\t\t}\n\t})\n\n\tt.Run(\"WithoutConnector\", func(t *testing.T) {\n\t\tassistant := AssistantModel{\n\t\t\tID: \"noconn\",\n\t\t\tName: \"No Connector\",\n\t\t}\n\t\tresult := assistant.ModelID()\n\t\texpected := \"no-connector-unknown-yao_noconn\"\n\t\tif result != expected {\n\t\t\tt.Errorf(\"Expected '%s', got '%s'\", expected, result)\n\t\t}\n\t})\n\n\tt.Run(\"WithSpacesInName\", func(t *testing.T) {\n\t\tassistant := AssistantModel{\n\t\t\tID: \"spaces\",\n\t\t\tName: \"Test Bot With Spaces\",\n\t\t\tConnector: \"anthropic\",\n\t\t\tOptions: map[string]interface{}{\n\t\t\t\t\"model\": \"claude-3\",\n\t\t\t},\n\t\t}\n\t\tresult := assistant.ModelID()\n\t\texpected := \"test-bot-with-spaces-claude-3-yao_spaces\"\n\t\tif result != expected {\n\t\t\tt.Errorf(\"Expected '%s', got '%s'\", expected, result)\n\t\t}\n\t})\n\n\tt.Run(\"WithUpperCaseName\", func(t *testing.T) {\n\t\tassistant := AssistantModel{\n\t\t\tID: \"upper\",\n\t\t\tName: \"UPPERCASE-NAME\",\n\t\t\tConnector: \"openai\",\n\t\t\tOptions: map[string]interface{}{\n\t\t\t\t\"model\": \"GPT-4\",\n\t\t\t},\n\t\t}\n\t\tresult := assistant.ModelID()\n\t\texpected := \"uppercase-name-GPT-4-yao_upper\"\n\t\tif result != expected {\n\t\t\tt.Errorf(\"Expected '%s', got '%s'\", expected, result)\n\t\t}\n\t})\n\n\tt.Run(\"WithEmptyOptions\", func(t *testing.T) {\n\t\tassistant := AssistantModel{\n\t\t\tID: \"empty\",\n\t\t\tName: \"Empty Options\",\n\t\t\tConnector: \"openai\",\n\t\t\tOptions: map[string]interface{}{},\n\t\t}\n\t\tresult := assistant.ModelID()\n\t\t// When connector is not loaded in test, it should return unknown\n\t\texpected := \"empty-options-unknown-yao_empty\"\n\t\tif result != expected {\n\t\t\tt.Errorf(\"Expected '%s', got '%s'\", expected, result)\n\t\t}\n\t})\n}\n\n// TestToConnectorOptions tests the ToConnectorOptions conversion function\nfunc TestToConnectorOptions(t *testing.T) {\n\tt.Run(\"NilInput\", func(t *testing.T) {\n\t\tresult, err := ToConnectorOptions(nil)\n\t\tif err != nil {\n\t\t\tt.Errorf(\"Expected no error, got: %v\", err)\n\t\t}\n\t\tif result != nil {\n\t\t\tt.Errorf(\"Expected nil result, got: %v\", result)\n\t\t}\n\t})\n\n\tt.Run(\"ConnectorOptionsPointer\", func(t *testing.T) {\n\t\toptionalTrue := true\n\t\topts := &ConnectorOptions{\n\t\t\tOptional: &optionalTrue,\n\t\t\tConnectors: []string{\"openai\", \"anthropic\"},\n\t\t\tFilters: []ModelCapability{CapVision, CapToolCalls},\n\t\t}\n\t\tresult, err := ToConnectorOptions(opts)\n\t\tif err != nil {\n\t\t\tt.Errorf(\"Expected no error, got: %v\", err)\n\t\t}\n\t\tif result != opts {\n\t\t\tt.Errorf(\"Expected same pointer\")\n\t\t}\n\t})\n\n\tt.Run(\"ConnectorOptionsValue\", func(t *testing.T) {\n\t\toptionalTrue := true\n\t\topts := ConnectorOptions{\n\t\t\tOptional: &optionalTrue,\n\t\t\tConnectors: []string{\"openai\", \"anthropic\"},\n\t\t\tFilters: []ModelCapability{CapVision, CapToolCalls},\n\t\t}\n\t\tresult, err := ToConnectorOptions(opts)\n\t\tif err != nil {\n\t\t\tt.Errorf(\"Expected no error, got: %v\", err)\n\t\t}\n\t\tif result.Optional == nil || !*result.Optional {\n\t\t\tt.Error(\"Expected Optional to be true\")\n\t\t}\n\t\tif len(result.Connectors) != 2 {\n\t\t\tt.Errorf(\"Expected 2 connectors, got %d\", len(result.Connectors))\n\t\t}\n\t\tif len(result.Filters) != 2 {\n\t\t\tt.Errorf(\"Expected 2 filters, got %d\", len(result.Filters))\n\t\t}\n\t})\n\n\tt.Run(\"MapInput\", func(t *testing.T) {\n\t\tdata := map[string]interface{}{\n\t\t\t\"optional\": true,\n\t\t\t\"connectors\": []string{\"openai\", \"anthropic\", \"azure\"},\n\t\t\t\"filters\": []string{\"vision\", \"tool_calls\", \"audio\"},\n\t\t}\n\t\tresult, err := ToConnectorOptions(data)\n\t\tif err != nil {\n\t\t\tt.Errorf(\"Expected no error, got: %v\", err)\n\t\t}\n\t\tif result.Optional == nil || !*result.Optional {\n\t\t\tt.Error(\"Expected Optional to be true\")\n\t\t}\n\t\tif len(result.Connectors) != 3 {\n\t\t\tt.Errorf(\"Expected 3 connectors, got %d\", len(result.Connectors))\n\t\t}\n\t\tif len(result.Filters) != 3 {\n\t\t\tt.Errorf(\"Expected 3 filters, got %d\", len(result.Filters))\n\t\t}\n\t})\n\n\tt.Run(\"MapInputOptionalOnly\", func(t *testing.T) {\n\t\tdata := map[string]interface{}{\n\t\t\t\"optional\": true,\n\t\t}\n\t\tresult, err := ToConnectorOptions(data)\n\t\tif err != nil {\n\t\t\tt.Errorf(\"Expected no error, got: %v\", err)\n\t\t}\n\t\tif result.Optional == nil || !*result.Optional {\n\t\t\tt.Error(\"Expected Optional to be true\")\n\t\t}\n\t\tif result.Connectors != nil {\n\t\t\tt.Error(\"Expected Connectors to be nil\")\n\t\t}\n\t\tif result.Filters != nil {\n\t\t\tt.Error(\"Expected Filters to be nil\")\n\t\t}\n\t})\n\n\tt.Run(\"MapInputOptionalFalse\", func(t *testing.T) {\n\t\tdata := map[string]interface{}{\n\t\t\t\"optional\": false,\n\t\t\t\"connectors\": []string{\"openai\"},\n\t\t\t\"filters\": []string{\"vision\"},\n\t\t}\n\t\tresult, err := ToConnectorOptions(data)\n\t\tif err != nil {\n\t\t\tt.Errorf(\"Expected no error, got: %v\", err)\n\t\t}\n\t\tif result.Optional == nil {\n\t\t\tt.Error(\"Expected Optional to be set\")\n\t\t} else if *result.Optional {\n\t\t\tt.Error(\"Expected Optional to be false\")\n\t\t}\n\t\tif len(result.Connectors) != 1 {\n\t\t\tt.Errorf(\"Expected 1 connector, got %d\", len(result.Connectors))\n\t\t}\n\t})\n\n\tt.Run(\"MapInputOptionalNil\", func(t *testing.T) {\n\t\tdata := map[string]interface{}{\n\t\t\t\"connectors\": []string{\"openai\"},\n\t\t}\n\t\tresult, err := ToConnectorOptions(data)\n\t\tif err != nil {\n\t\t\tt.Errorf(\"Expected no error, got: %v\", err)\n\t\t}\n\t\tif result.Optional != nil {\n\t\t\tt.Errorf(\"Expected Optional to be nil (not set), got: %v\", *result.Optional)\n\t\t}\n\t\tif len(result.Connectors) != 1 {\n\t\t\tt.Errorf(\"Expected 1 connector, got %d\", len(result.Connectors))\n\t\t}\n\t})\n\n\tt.Run(\"InvalidInput\", func(t *testing.T) {\n\t\t// Test with data that can't be marshaled\n\t\tinvalidData := make(chan int)\n\t\t_, err := ToConnectorOptions(invalidData)\n\t\tif err == nil {\n\t\t\tt.Error(\"Expected error for invalid input\")\n\t\t}\n\t})\n\n\tt.Run(\"InvalidJSONUnmarshal\", func(t *testing.T) {\n\t\t// Test with data that marshals but can't unmarshal to ConnectorOptions\n\t\tdata := map[string]interface{}{\n\t\t\t\"invalid_field\": \"should cause unmarshal to fail gracefully\",\n\t\t}\n\t\tresult, err := ToConnectorOptions(data)\n\t\t// Should not error, just return empty ConnectorOptions\n\t\tif err != nil {\n\t\t\tt.Errorf(\"Expected no error, got: %v\", err)\n\t\t}\n\t\tif result == nil {\n\t\t\tt.Error(\"Expected non-nil result\")\n\t\t}\n\t})\n}\n\n// TestToModes tests the ToModes conversion function\nfunc TestToModes(t *testing.T) {\n\tt.Run(\"NilInput\", func(t *testing.T) {\n\t\tresult, err := ToModes(nil)\n\t\tif err != nil {\n\t\t\tt.Errorf(\"Expected no error, got: %v\", err)\n\t\t}\n\t\tif result != nil {\n\t\t\tt.Errorf(\"Expected nil result, got: %v\", result)\n\t\t}\n\t})\n\n\tt.Run(\"StringSlice\", func(t *testing.T) {\n\t\tmodes := []string{\"chat\", \"task\", \"analyze\"}\n\t\tresult, err := ToModes(modes)\n\t\tif err != nil {\n\t\t\tt.Errorf(\"Expected no error, got: %v\", err)\n\t\t}\n\t\tif len(result) != 3 {\n\t\t\tt.Errorf(\"Expected 3 modes, got %d\", len(result))\n\t\t}\n\t\tif result[0] != \"chat\" {\n\t\t\tt.Errorf(\"Expected 'chat', got '%s'\", result[0])\n\t\t}\n\t\tif result[1] != \"task\" {\n\t\t\tt.Errorf(\"Expected 'task', got '%s'\", result[1])\n\t\t}\n\t\tif result[2] != \"analyze\" {\n\t\t\tt.Errorf(\"Expected 'analyze', got '%s'\", result[2])\n\t\t}\n\t})\n\n\tt.Run(\"InterfaceSlice\", func(t *testing.T) {\n\t\tmodes := []interface{}{\"chat\", \"task\", 123}\n\t\tresult, err := ToModes(modes)\n\t\tif err != nil {\n\t\t\tt.Errorf(\"Expected no error, got: %v\", err)\n\t\t}\n\t\tif len(result) != 3 {\n\t\t\tt.Errorf(\"Expected 3 modes, got %d\", len(result))\n\t\t}\n\t\tif result[0] != \"chat\" {\n\t\t\tt.Errorf(\"Expected 'chat', got '%s'\", result[0])\n\t\t}\n\t\tif result[2] != \"123\" {\n\t\t\tt.Errorf(\"Expected '123', got '%s'\", result[2])\n\t\t}\n\t})\n\n\tt.Run(\"SingleString\", func(t *testing.T) {\n\t\tmode := \"chat\"\n\t\tresult, err := ToModes(mode)\n\t\tif err != nil {\n\t\t\tt.Errorf(\"Expected no error, got: %v\", err)\n\t\t}\n\t\tif len(result) != 1 {\n\t\t\tt.Errorf(\"Expected 1 mode, got %d\", len(result))\n\t\t}\n\t\tif result[0] != \"chat\" {\n\t\t\tt.Errorf(\"Expected 'chat', got '%s'\", result[0])\n\t\t}\n\t})\n\n\tt.Run(\"EmptySlice\", func(t *testing.T) {\n\t\tmodes := []string{}\n\t\tresult, err := ToModes(modes)\n\t\tif err != nil {\n\t\t\tt.Errorf(\"Expected no error, got: %v\", err)\n\t\t}\n\t\tif len(result) != 0 {\n\t\t\tt.Errorf(\"Expected 0 modes, got %d\", len(result))\n\t\t}\n\t})\n\n\tt.Run(\"InvalidInput\", func(t *testing.T) {\n\t\t// Test with data that can't be marshaled\n\t\tinvalidData := make(chan int)\n\t\t_, err := ToModes(invalidData)\n\t\tif err == nil {\n\t\t\tt.Error(\"Expected error for invalid input\")\n\t\t}\n\t})\n\n\tt.Run(\"InvalidJSONUnmarshal\", func(t *testing.T) {\n\t\t// Test with data that marshals but can't unmarshal to []string\n\t\tdata := map[string]interface{}{\n\t\t\t\"invalid\": \"structure\",\n\t\t}\n\t\t_, err := ToModes(data)\n\t\tif err == nil {\n\t\t\tt.Error(\"Expected error for invalid unmarshal\")\n\t\t}\n\t})\n\n\tt.Run(\"MixedTypes\", func(t *testing.T) {\n\t\tmodes := []interface{}{\"chat\", 456, \"task\", true}\n\t\tresult, err := ToModes(modes)\n\t\tif err != nil {\n\t\t\tt.Errorf(\"Expected no error, got: %v\", err)\n\t\t}\n\t\tif len(result) != 4 {\n\t\t\tt.Errorf(\"Expected 4 modes, got %d\", len(result))\n\t\t}\n\t\t// cast.ToString should convert all to strings\n\t\tif result[0] != \"chat\" {\n\t\t\tt.Errorf(\"Expected 'chat', got '%s'\", result[0])\n\t\t}\n\t\tif result[1] != \"456\" {\n\t\t\tt.Errorf(\"Expected '456', got '%s'\", result[1])\n\t\t}\n\t\tif result[2] != \"task\" {\n\t\t\tt.Errorf(\"Expected 'task', got '%s'\", result[2])\n\t\t}\n\t\tif result[3] != \"true\" {\n\t\t\tt.Errorf(\"Expected 'true', got '%s'\", result[3])\n\t\t}\n\t})\n}\n\n// TestToPromptPresets tests the ToPromptPresets conversion function\nfunc TestToPromptPresets(t *testing.T) {\n\tt.Run(\"NilInput\", func(t *testing.T) {\n\t\tresult, err := ToPromptPresets(nil)\n\t\tif err != nil {\n\t\t\tt.Errorf(\"Expected no error, got: %v\", err)\n\t\t}\n\t\tif result != nil {\n\t\t\tt.Errorf(\"Expected nil result, got: %v\", result)\n\t\t}\n\t})\n\n\tt.Run(\"MapStringPromptSlice\", func(t *testing.T) {\n\t\tpresets := map[string][]Prompt{\n\t\t\t\"chat\": {\n\t\t\t\t{Role: \"system\", Content: \"You are a chat assistant\"},\n\t\t\t},\n\t\t\t\"task\": {\n\t\t\t\t{Role: \"system\", Content: \"You are a task assistant\"},\n\t\t\t},\n\t\t}\n\t\tresult, err := ToPromptPresets(presets)\n\t\tif err != nil {\n\t\t\tt.Errorf(\"Expected no error, got: %v\", err)\n\t\t}\n\t\tif len(result) != 2 {\n\t\t\tt.Errorf(\"Expected 2 presets, got %d\", len(result))\n\t\t}\n\t\tif len(result[\"chat\"]) != 1 {\n\t\t\tt.Errorf(\"Expected 1 chat prompt, got %d\", len(result[\"chat\"]))\n\t\t}\n\t\tif len(result[\"task\"]) != 1 {\n\t\t\tt.Errorf(\"Expected 1 task prompt, got %d\", len(result[\"task\"]))\n\t\t}\n\t})\n\n\tt.Run(\"MapInput\", func(t *testing.T) {\n\t\tdata := map[string]interface{}{\n\t\t\t\"chat\": []interface{}{\n\t\t\t\tmap[string]interface{}{\"role\": \"system\", \"content\": \"Chat mode system prompt\"},\n\t\t\t\tmap[string]interface{}{\"role\": \"user\", \"content\": \"Example user message\"},\n\t\t\t},\n\t\t\t\"task\": []interface{}{\n\t\t\t\tmap[string]interface{}{\"role\": \"system\", \"content\": \"Task mode system prompt\"},\n\t\t\t},\n\t\t\t\"analyze\": []interface{}{\n\t\t\t\tmap[string]interface{}{\"role\": \"system\", \"content\": \"Analyze mode system prompt\"},\n\t\t\t},\n\t\t}\n\t\tresult, err := ToPromptPresets(data)\n\t\tif err != nil {\n\t\t\tt.Errorf(\"Expected no error, got: %v\", err)\n\t\t}\n\t\tif len(result) != 3 {\n\t\t\tt.Errorf(\"Expected 3 presets, got %d\", len(result))\n\t\t}\n\t\tif len(result[\"chat\"]) != 2 {\n\t\t\tt.Errorf(\"Expected 2 chat prompts, got %d\", len(result[\"chat\"]))\n\t\t}\n\t\tif len(result[\"task\"]) != 1 {\n\t\t\tt.Errorf(\"Expected 1 task prompt, got %d\", len(result[\"task\"]))\n\t\t}\n\t\tif len(result[\"analyze\"]) != 1 {\n\t\t\tt.Errorf(\"Expected 1 analyze prompt, got %d\", len(result[\"analyze\"]))\n\t\t}\n\t})\n\n\tt.Run(\"EmptyMap\", func(t *testing.T) {\n\t\tdata := map[string]interface{}{}\n\t\tresult, err := ToPromptPresets(data)\n\t\tif err != nil {\n\t\t\tt.Errorf(\"Expected no error, got: %v\", err)\n\t\t}\n\t\tif result == nil {\n\t\t\tt.Error(\"Expected non-nil result\")\n\t\t}\n\t\tif len(result) != 0 {\n\t\t\tt.Errorf(\"Expected empty map, got %d entries\", len(result))\n\t\t}\n\t})\n\n\tt.Run(\"SinglePreset\", func(t *testing.T) {\n\t\tdata := map[string]interface{}{\n\t\t\t\"default\": []interface{}{\n\t\t\t\tmap[string]interface{}{\"role\": \"system\", \"content\": \"Default prompt\"},\n\t\t\t},\n\t\t}\n\t\tresult, err := ToPromptPresets(data)\n\t\tif err != nil {\n\t\t\tt.Errorf(\"Expected no error, got: %v\", err)\n\t\t}\n\t\tif len(result) != 1 {\n\t\t\tt.Errorf(\"Expected 1 preset, got %d\", len(result))\n\t\t}\n\t\tif _, ok := result[\"default\"]; !ok {\n\t\t\tt.Error(\"Expected 'default' key in result\")\n\t\t}\n\t})\n\n\tt.Run(\"InvalidInput\", func(t *testing.T) {\n\t\t// Test with data that can't be marshaled\n\t\tinvalidData := make(chan int)\n\t\t_, err := ToPromptPresets(invalidData)\n\t\tif err == nil {\n\t\t\tt.Error(\"Expected error for invalid input\")\n\t\t}\n\t})\n\n\tt.Run(\"InvalidJSONUnmarshal\", func(t *testing.T) {\n\t\t// Test with data that marshals but can't unmarshal to map[string][]Prompt\n\t\t// This is a string that can be marshaled but won't unmarshal to the expected type\n\t\tdata := \"not a map\"\n\t\t_, err := ToPromptPresets(data)\n\t\tif err == nil {\n\t\t\tt.Error(\"Expected error for invalid JSON unmarshal\")\n\t\t}\n\t})\n\n\tt.Run(\"PromptWithAllFields\", func(t *testing.T) {\n\t\tdata := map[string]interface{}{\n\t\t\t\"advanced\": []interface{}{\n\t\t\t\tmap[string]interface{}{\n\t\t\t\t\t\"role\": \"system\",\n\t\t\t\t\t\"content\": \"Advanced system prompt\",\n\t\t\t\t\t\"name\": \"system-prompt\",\n\t\t\t\t},\n\t\t\t\tmap[string]interface{}{\n\t\t\t\t\t\"role\": \"user\",\n\t\t\t\t\t\"content\": \"User example\",\n\t\t\t\t\t\"name\": \"user-example\",\n\t\t\t\t},\n\t\t\t\tmap[string]interface{}{\n\t\t\t\t\t\"role\": \"assistant\",\n\t\t\t\t\t\"content\": \"Assistant response\",\n\t\t\t\t\t\"name\": \"assistant-response\",\n\t\t\t\t},\n\t\t\t},\n\t\t}\n\t\tresult, err := ToPromptPresets(data)\n\t\tif err != nil {\n\t\t\tt.Errorf(\"Expected no error, got: %v\", err)\n\t\t}\n\t\tif len(result[\"advanced\"]) != 3 {\n\t\t\tt.Errorf(\"Expected 3 prompts in advanced, got %d\", len(result[\"advanced\"]))\n\t\t}\n\t\tif result[\"advanced\"][0].Role != \"system\" {\n\t\t\tt.Errorf(\"Expected role 'system', got '%s'\", result[\"advanced\"][0].Role)\n\t\t}\n\t\tif result[\"advanced\"][0].Content != \"Advanced system prompt\" {\n\t\t\tt.Errorf(\"Expected content 'Advanced system prompt', got '%s'\", result[\"advanced\"][0].Content)\n\t\t}\n\t})\n}\n\n// TestParseModelID tests the ParseModelID function\nfunc TestParseModelID(t *testing.T) {\n\tt.Run(\"ValidModelID\", func(t *testing.T) {\n\t\tmodelID := \"test-assistant-gpt-4o-yao_test123\"\n\t\tresult := ParseModelID(modelID)\n\t\texpected := \"test123\"\n\t\tif result != expected {\n\t\t\tt.Errorf(\"Expected '%s', got '%s'\", expected, result)\n\t\t}\n\t})\n\n\tt.Run(\"ValidModelIDWithMultipleDashes\", func(t *testing.T) {\n\t\tmodelID := \"my-test-bot-gpt-3.5-turbo-yao_abc456\"\n\t\tresult := ParseModelID(modelID)\n\t\texpected := \"abc456\"\n\t\tif result != expected {\n\t\t\tt.Errorf(\"Expected '%s', got '%s'\", expected, result)\n\t\t}\n\t})\n\n\tt.Run(\"ValidModelIDWithHyphenInID\", func(t *testing.T) {\n\t\tmodelID := \"assistant-name-model-yao_id-with-dash\"\n\t\tresult := ParseModelID(modelID)\n\t\texpected := \"id-with-dash\"\n\t\tif result != expected {\n\t\t\tt.Errorf(\"Expected '%s', got '%s'\", expected, result)\n\t\t}\n\t})\n\n\tt.Run(\"InvalidModelIDNoYaoPrefix\", func(t *testing.T) {\n\t\tmodelID := \"test-assistant-gpt-4o-test123\"\n\t\tresult := ParseModelID(modelID)\n\t\tif result != \"\" {\n\t\t\tt.Errorf(\"Expected empty string, got '%s'\", result)\n\t\t}\n\t})\n\n\tt.Run(\"InvalidModelIDEmpty\", func(t *testing.T) {\n\t\tmodelID := \"\"\n\t\tresult := ParseModelID(modelID)\n\t\tif result != \"\" {\n\t\t\tt.Errorf(\"Expected empty string, got '%s'\", result)\n\t\t}\n\t})\n\n\tt.Run(\"InvalidModelIDOnlyYaoPrefix\", func(t *testing.T) {\n\t\tmodelID := \"yao_\"\n\t\tresult := ParseModelID(modelID)\n\t\tif result != \"\" {\n\t\t\tt.Errorf(\"Expected empty string, got '%s'\", result)\n\t\t}\n\t})\n\n\tt.Run(\"RoundTrip\", func(t *testing.T) {\n\t\tassistant := AssistantModel{\n\t\t\tID: \"roundtrip123\",\n\t\t\tName: \"Round Trip Test\",\n\t\t\tConnector: \"openai\",\n\t\t\tOptions: map[string]interface{}{\n\t\t\t\t\"model\": \"gpt-4\",\n\t\t\t},\n\t\t}\n\t\tmodelID := assistant.ModelID()\n\t\textractedID := ParseModelID(modelID)\n\t\tif extractedID != assistant.ID {\n\t\t\tt.Errorf(\"Round trip failed: expected '%s', got '%s'\", assistant.ID, extractedID)\n\t\t}\n\t})\n}\n" }, { "path": "agent/store/types/fields.go", "content": "package types\n\nimport \"github.com/yaoapp/kun/log\"\n\n// AssistantAllowedFields defines the whitelist of fields that can be selected for assistants\nvar AssistantAllowedFields = map[string]bool{\n\t\"id\": true,\n\t\"assistant_id\": true,\n\t\"type\": true,\n\t\"name\": true,\n\t\"avatar\": true,\n\t\"connector\": true,\n\t\"connector_options\": true,\n\t\"description\": true,\n\t\"path\": true,\n\t\"sort\": true,\n\t\"built_in\": true,\n\t\"placeholder\": true,\n\t\"options\": true,\n\t\"prompts\": true,\n\t\"prompt_presets\": true,\n\t\"disable_global_prompts\": true,\n\t\"capabilities\": true,\n\t\"workflow\": true,\n\t\"kb\": true,\n\t\"db\": true,\n\t\"mcp\": true,\n\t\"sandbox\": true,\n\t\"source\": true,\n\t\"tags\": true,\n\t\"modes\": true,\n\t\"default_mode\": true,\n\t\"readonly\": true,\n\t\"public\": true,\n\t\"share\": true,\n\t\"locales\": true,\n\t\"uses\": true,\n\t\"search\": true,\n\t\"dependencies\": true,\n\t\"automated\": true,\n\t\"mentionable\": true,\n\t\"created_at\": true,\n\t\"updated_at\": true,\n\t\"__yao_created_by\": true,\n\t\"__yao_updated_by\": true,\n\t\"__yao_team_id\": true,\n\t\"__yao_tenant_id\": true,\n}\n\n// AssistantDefaultFields defines the default fields to select for assistants when no specific fields are requested\n// These are lightweight fields suitable for list views and basic information display\nvar AssistantDefaultFields = []string{\n\t\"assistant_id\",\n\t\"type\",\n\t\"name\",\n\t\"avatar\",\n\t\"connector\",\n\t\"description\",\n\t\"capabilities\", // Capabilities description for Robot orchestration (lightweight)\n\t\"tags\", // Tags for categorization (lightweight)\n\t\"modes\", // Supported modes (lightweight)\n\t\"default_mode\", // Default mode (lightweight)\n\t\"sort\",\n\t\"built_in\",\n\t\"readonly\",\n\t\"public\",\n\t\"share\",\n\t\"automated\",\n\t\"mentionable\",\n\t\"sandbox\", // Sandbox configuration presence (lightweight)\n\t\"kb\", // Knowledge base configuration (lightweight)\n\t\"db\", // Database configuration (lightweight)\n\t\"mcp\", // MCP servers configuration (lightweight)\n\t\"dependencies\", // Dependencies on other MCP Clients (lightweight)\n\t\"created_at\",\n\t\"updated_at\",\n\t\"__yao_created_by\", // Permission: creator user ID\n\t\"__yao_updated_by\", // Permission: updater user ID\n\t\"__yao_team_id\", // Permission: team ID\n\t\"__yao_tenant_id\", // Permission: tenant ID\n}\n\n// AssistantFullFields defines all available fields including complex/large fields\n// Use this when you need complete assistant data for backend processing\nvar AssistantFullFields = []string{\n\t\"assistant_id\",\n\t\"type\",\n\t\"name\",\n\t\"avatar\",\n\t\"connector\",\n\t\"connector_options\",\n\t\"description\",\n\t\"capabilities\",\n\t\"path\",\n\t\"sort\",\n\t\"built_in\",\n\t\"placeholder\",\n\t\"options\",\n\t\"prompts\",\n\t\"prompt_presets\",\n\t\"disable_global_prompts\",\n\t\"workflow\",\n\t\"kb\",\n\t\"db\",\n\t\"mcp\",\n\t\"sandbox\",\n\t\"source\",\n\t\"tags\",\n\t\"modes\",\n\t\"default_mode\",\n\t\"readonly\",\n\t\"public\",\n\t\"share\",\n\t\"locales\",\n\t\"uses\",\n\t\"search\",\n\t\"dependencies\",\n\t\"automated\",\n\t\"mentionable\",\n\t\"created_at\",\n\t\"updated_at\",\n\t\"__yao_created_by\",\n\t\"__yao_updated_by\",\n\t\"__yao_team_id\",\n\t\"__yao_tenant_id\",\n}\n\n// ValidateAssistantFields validates and filters assistant select fields against the whitelist\n// Returns the filtered fields. If input is empty, returns empty slice (meaning no restriction).\n// If all fields are invalid, returns default fields as fallback.\nfunc ValidateAssistantFields(fields []string) []string {\n\t// If no fields specified, return empty slice (no restriction)\n\tif len(fields) == 0 {\n\t\treturn []string{}\n\t}\n\n\t// Filter out any fields not in the whitelist\n\tsanitized := make([]string, 0, len(fields))\n\tfor _, field := range fields {\n\t\tif AssistantAllowedFields[field] {\n\t\t\tsanitized = append(sanitized, field)\n\t\t} else {\n\t\t\tlog.Warn(\"Ignoring invalid assistant select field: %s\", field)\n\t\t}\n\t}\n\n\t// If all fields were filtered out, return default fields as fallback\n\tif len(sanitized) == 0 {\n\t\tlog.Warn(\"All assistant select fields were invalid, using default fields\")\n\t\treturn AssistantDefaultFields\n\t}\n\n\treturn sanitized\n}\n" }, { "path": "agent/store/types/fields_test.go", "content": "package types\n\nimport (\n\t\"reflect\"\n\t\"testing\"\n)\n\nfunc TestValidateAssistantFields(t *testing.T) {\n\tt.Run(\"EmptyInput_ReturnsEmptySlice\", func(t *testing.T) {\n\t\tresult := ValidateAssistantFields([]string{})\n\t\tif len(result) != 0 {\n\t\t\tt.Errorf(\"Expected empty slice, got %v\", result)\n\t\t}\n\t})\n\n\tt.Run(\"NilInput_ReturnsEmptySlice\", func(t *testing.T) {\n\t\tresult := ValidateAssistantFields(nil)\n\t\tif len(result) != 0 {\n\t\t\tt.Errorf(\"Expected empty slice, got %v\", result)\n\t\t}\n\t})\n\n\tt.Run(\"ValidFields_ReturnsFiltered\", func(t *testing.T) {\n\t\tinput := []string{\"assistant_id\", \"name\", \"type\"}\n\t\tresult := ValidateAssistantFields(input)\n\t\texpected := []string{\"assistant_id\", \"name\", \"type\"}\n\t\tif !reflect.DeepEqual(result, expected) {\n\t\t\tt.Errorf(\"Expected %v, got %v\", expected, result)\n\t\t}\n\t})\n\n\tt.Run(\"MixedValidInvalidFields_ReturnsOnlyValid\", func(t *testing.T) {\n\t\tinput := []string{\"assistant_id\", \"invalid_field\", \"name\", \"malicious_column\"}\n\t\tresult := ValidateAssistantFields(input)\n\t\texpected := []string{\"assistant_id\", \"name\"}\n\t\tif !reflect.DeepEqual(result, expected) {\n\t\t\tt.Errorf(\"Expected %v, got %v\", expected, result)\n\t\t}\n\t})\n\n\tt.Run(\"AllInvalidFields_ReturnsDefaultFields\", func(t *testing.T) {\n\t\tinput := []string{\"invalid1\", \"invalid2\", \"malicious\"}\n\t\tresult := ValidateAssistantFields(input)\n\t\tif !reflect.DeepEqual(result, AssistantDefaultFields) {\n\t\t\tt.Errorf(\"Expected default fields when all invalid, got %v\", result)\n\t\t}\n\t})\n\n\tt.Run(\"PermissionFields_AreAllowed\", func(t *testing.T) {\n\t\tinput := []string{\"__yao_created_by\", \"__yao_team_id\", \"assistant_id\"}\n\t\tresult := ValidateAssistantFields(input)\n\t\texpected := []string{\"__yao_created_by\", \"__yao_team_id\", \"assistant_id\"}\n\t\tif !reflect.DeepEqual(result, expected) {\n\t\t\tt.Errorf(\"Expected %v, got %v\", expected, result)\n\t\t}\n\t})\n\n\tt.Run(\"AllAllowedFields_AreInWhitelist\", func(t *testing.T) {\n\t\t// Verify all default fields are in the allowed list\n\t\tfor _, field := range AssistantDefaultFields {\n\t\t\tif !AssistantAllowedFields[field] {\n\t\t\t\tt.Errorf(\"Default field %s is not in allowed fields\", field)\n\t\t\t}\n\t\t}\n\t})\n\n\tt.Run(\"SQLInjectionAttempt_IsFiltered\", func(t *testing.T) {\n\t\tinput := []string{\"assistant_id\", \"name; DROP TABLE assistants;--\", \"type\"}\n\t\tresult := ValidateAssistantFields(input)\n\t\texpected := []string{\"assistant_id\", \"type\"}\n\t\tif !reflect.DeepEqual(result, expected) {\n\t\t\tt.Errorf(\"Expected SQL injection attempt to be filtered, got %v\", result)\n\t\t}\n\t})\n\n\tt.Run(\"DuplicateFields_AreKept\", func(t *testing.T) {\n\t\tinput := []string{\"assistant_id\", \"name\", \"assistant_id\", \"name\"}\n\t\tresult := ValidateAssistantFields(input)\n\t\t// Duplicates should be kept as-is (validation doesn't deduplicate)\n\t\texpected := []string{\"assistant_id\", \"name\", \"assistant_id\", \"name\"}\n\t\tif !reflect.DeepEqual(result, expected) {\n\t\t\tt.Errorf(\"Expected %v, got %v\", expected, result)\n\t\t}\n\t})\n}\n\nfunc TestAssistantAllowedFields(t *testing.T) {\n\tt.Run(\"ContainsBasicFields\", func(t *testing.T) {\n\t\trequiredFields := []string{\n\t\t\t\"assistant_id\",\n\t\t\t\"name\",\n\t\t\t\"type\",\n\t\t\t\"connector\",\n\t\t\t\"description\",\n\t\t}\n\t\tfor _, field := range requiredFields {\n\t\t\tif !AssistantAllowedFields[field] {\n\t\t\t\tt.Errorf(\"Required field %s is missing from allowed fields\", field)\n\t\t\t}\n\t\t}\n\t})\n\n\tt.Run(\"ContainsPermissionFields\", func(t *testing.T) {\n\t\tpermissionFields := []string{\n\t\t\t\"__yao_created_by\",\n\t\t\t\"__yao_updated_by\",\n\t\t\t\"__yao_team_id\",\n\t\t\t\"__yao_tenant_id\",\n\t\t}\n\t\tfor _, field := range permissionFields {\n\t\t\tif !AssistantAllowedFields[field] {\n\t\t\t\tt.Errorf(\"Permission field %s is missing from allowed fields\", field)\n\t\t\t}\n\t\t}\n\t})\n\n\tt.Run(\"ContainsComplexFields\", func(t *testing.T) {\n\t\tcomplexFields := []string{\n\t\t\t\"options\",\n\t\t\t\"prompts\",\n\t\t\t\"prompt_presets\",\n\t\t\t\"disable_global_prompts\",\n\t\t\t\"workflow\",\n\t\t\t\"kb\",\n\t\t\t\"db\",\n\t\t\t\"mcp\",\n\t\t\t\"placeholder\",\n\t\t\t\"locales\",\n\t\t\t\"uses\",\n\t\t\t\"connector_options\",\n\t\t\t\"source\",\n\t\t\t\"modes\",\n\t\t\t\"default_mode\",\n\t\t}\n\t\tfor _, field := range complexFields {\n\t\t\tif !AssistantAllowedFields[field] {\n\t\t\t\tt.Errorf(\"Complex field %s is missing from allowed fields\", field)\n\t\t\t}\n\t\t}\n\t})\n}\n\nfunc TestAssistantDefaultFields(t *testing.T) {\n\tt.Run(\"ContainsEssentialFields\", func(t *testing.T) {\n\t\tessentialFields := []string{\n\t\t\t\"assistant_id\",\n\t\t\t\"name\",\n\t\t\t\"type\",\n\t\t\t\"kb\", // Knowledge base is essential for assistant functionality\n\t\t\t\"db\", // Database is essential for assistant functionality\n\t\t\t\"mcp\", // MCP servers are essential for assistant functionality\n\t\t\t\"modes\", // Supported modes are essential for mode filtering\n\t\t\t\"default_mode\", // Default mode is essential for mode selection\n\t\t\t\"__yao_created_by\", // Permission fields are essential for access control\n\t\t\t\"__yao_updated_by\",\n\t\t\t\"__yao_team_id\",\n\t\t\t\"__yao_tenant_id\",\n\t\t}\n\n\t\tdefaultFieldsMap := make(map[string]bool)\n\t\tfor _, field := range AssistantDefaultFields {\n\t\t\tdefaultFieldsMap[field] = true\n\t\t}\n\n\t\tfor _, field := range essentialFields {\n\t\t\tif !defaultFieldsMap[field] {\n\t\t\t\tt.Errorf(\"Essential field %s is missing from default fields\", field)\n\t\t\t}\n\t\t}\n\t})\n\n\tt.Run(\"DoesNotContainSensitiveFields\", func(t *testing.T) {\n\t\t// Default fields should not include complex/large fields by default\n\t\t// Note: kb, db, mcp, tags, modes, and default_mode are lightweight and included in defaults\n\t\tsensitiveFields := []string{\n\t\t\t\"options\",\n\t\t\t\"prompts\",\n\t\t\t\"prompt_presets\",\n\t\t\t\"workflow\",\n\t\t\t\"placeholder\",\n\t\t\t\"locales\",\n\t\t\t\"uses\",\n\t\t\t\"connector_options\",\n\t\t\t\"source\",\n\t\t}\n\n\t\tdefaultFieldsMap := make(map[string]bool)\n\t\tfor _, field := range AssistantDefaultFields {\n\t\t\tdefaultFieldsMap[field] = true\n\t\t}\n\n\t\tfor _, field := range sensitiveFields {\n\t\t\tif defaultFieldsMap[field] {\n\t\t\t\tt.Errorf(\"Large/complex field %s should not be in default fields\", field)\n\t\t\t}\n\t\t}\n\t})\n}\n\nfunc TestAssistantFullFields(t *testing.T) {\n\tt.Run(\"ContainsAllAllowedFields\", func(t *testing.T) {\n\t\t// Full fields should contain all fields from allowed fields\n\t\tfullFieldsMap := make(map[string]bool)\n\t\tfor _, field := range AssistantFullFields {\n\t\t\tfullFieldsMap[field] = true\n\t\t}\n\n\t\tfor field := range AssistantAllowedFields {\n\t\t\tif field == \"id\" {\n\t\t\t\t// \"id\" is an alias for \"assistant_id\", skip\n\t\t\t\tcontinue\n\t\t\t}\n\t\t\tif !fullFieldsMap[field] {\n\t\t\t\tt.Errorf(\"Allowed field %s is missing from full fields\", field)\n\t\t\t}\n\t\t}\n\t})\n\n\tt.Run(\"AllFieldsAreAllowed\", func(t *testing.T) {\n\t\t// All fields in full list should be in allowed fields\n\t\tfor _, field := range AssistantFullFields {\n\t\t\tif !AssistantAllowedFields[field] {\n\t\t\t\tt.Errorf(\"Full field %s is not in allowed fields\", field)\n\t\t\t}\n\t\t}\n\t})\n\n\tt.Run(\"ContainsComplexFields\", func(t *testing.T) {\n\t\t// Full fields should include all complex/large fields\n\t\tcomplexFields := []string{\n\t\t\t\"options\",\n\t\t\t\"prompts\",\n\t\t\t\"prompt_presets\",\n\t\t\t\"disable_global_prompts\",\n\t\t\t\"workflow\",\n\t\t\t\"kb\",\n\t\t\t\"db\",\n\t\t\t\"mcp\",\n\t\t\t\"placeholder\",\n\t\t\t\"locales\",\n\t\t\t\"uses\",\n\t\t\t\"connector_options\",\n\t\t\t\"source\",\n\t\t\t\"modes\",\n\t\t\t\"default_mode\",\n\t\t}\n\n\t\tfullFieldsMap := make(map[string]bool)\n\t\tfor _, field := range AssistantFullFields {\n\t\t\tfullFieldsMap[field] = true\n\t\t}\n\n\t\tfor _, field := range complexFields {\n\t\t\tif !fullFieldsMap[field] {\n\t\t\t\tt.Errorf(\"Complex field %s is missing from full fields\", field)\n\t\t\t}\n\t\t}\n\t})\n\n\tt.Run(\"ContainsPermissionFields\", func(t *testing.T) {\n\t\t// Full fields should include permission fields\n\t\tpermissionFields := []string{\n\t\t\t\"__yao_created_by\",\n\t\t\t\"__yao_updated_by\",\n\t\t\t\"__yao_team_id\",\n\t\t\t\"__yao_tenant_id\",\n\t\t}\n\n\t\tfullFieldsMap := make(map[string]bool)\n\t\tfor _, field := range AssistantFullFields {\n\t\t\tfullFieldsMap[field] = true\n\t\t}\n\n\t\tfor _, field := range permissionFields {\n\t\t\tif !fullFieldsMap[field] {\n\t\t\t\tt.Errorf(\"Permission field %s is missing from full fields\", field)\n\t\t\t}\n\t\t}\n\t})\n}\n" }, { "path": "agent/store/types/mcp_test.go", "content": "package types\n\nimport (\n\t\"encoding/json\"\n\t\"testing\"\n)\n\nfunc TestMCPServerConfig_UnmarshalJSON(t *testing.T) {\n\ttests := []struct {\n\t\tname string\n\t\tinput string\n\t\twant MCPServerConfig\n\t\twantErr bool\n\t}{\n\t\t{\n\t\t\tname: \"Simple string\",\n\t\t\tinput: `\"server1\"`,\n\t\t\twant: MCPServerConfig{\n\t\t\t\tServerID: \"server1\",\n\t\t\t\tResources: nil,\n\t\t\t\tTools: nil,\n\t\t\t},\n\t\t\twantErr: false,\n\t\t},\n\t\t{\n\t\t\tname: \"Tools array only\",\n\t\t\tinput: `{\"server1\": [\"tool1\", \"tool2\"]}`,\n\t\t\twant: MCPServerConfig{\n\t\t\t\tServerID: \"server1\",\n\t\t\t\tResources: nil,\n\t\t\t\tTools: []string{\"tool1\", \"tool2\"},\n\t\t\t},\n\t\t\twantErr: false,\n\t\t},\n\t\t{\n\t\t\tname: \"Full config with resources and tools\",\n\t\t\tinput: `{\"server1\": {\"resources\": [\"res1\", \"res2\"], \"tools\": [\"tool1\", \"tool2\"]}}`,\n\t\t\twant: MCPServerConfig{\n\t\t\t\tServerID: \"server1\",\n\t\t\t\tResources: []string{\"res1\", \"res2\"},\n\t\t\t\tTools: []string{\"tool1\", \"tool2\"},\n\t\t\t},\n\t\t\twantErr: false,\n\t\t},\n\t\t{\n\t\t\tname: \"Only resources\",\n\t\t\tinput: `{\"server1\": {\"resources\": [\"res1\"]}}`,\n\t\t\twant: MCPServerConfig{\n\t\t\t\tServerID: \"server1\",\n\t\t\t\tResources: []string{\"res1\"},\n\t\t\t\tTools: nil,\n\t\t\t},\n\t\t\twantErr: false,\n\t\t},\n\t\t{\n\t\t\tname: \"Only tools\",\n\t\t\tinput: `{\"server1\": {\"tools\": [\"tool1\"]}}`,\n\t\t\twant: MCPServerConfig{\n\t\t\t\tServerID: \"server1\",\n\t\t\t\tResources: nil,\n\t\t\t\tTools: []string{\"tool1\"},\n\t\t\t},\n\t\t\twantErr: false,\n\t\t},\n\t\t{\n\t\t\tname: \"Standard object format\",\n\t\t\tinput: `{\"server_id\": \"server1\", \"resources\": [\"res1\"], \"tools\": [\"tool1\"]}`,\n\t\t\twant: MCPServerConfig{\n\t\t\t\tServerID: \"server1\",\n\t\t\t\tResources: []string{\"res1\"},\n\t\t\t\tTools: []string{\"tool1\"},\n\t\t\t},\n\t\t\twantErr: false,\n\t\t},\n\t\t{\n\t\t\tname: \"Standard object format - no resources/tools\",\n\t\t\tinput: `{\"server_id\": \"server1\"}`,\n\t\t\twant: MCPServerConfig{\n\t\t\t\tServerID: \"server1\",\n\t\t\t\tResources: nil,\n\t\t\t\tTools: nil,\n\t\t\t},\n\t\t\twantErr: false,\n\t\t},\n\t}\n\n\tfor _, tt := range tests {\n\t\tt.Run(tt.name, func(t *testing.T) {\n\t\t\tvar got MCPServerConfig\n\t\t\terr := json.Unmarshal([]byte(tt.input), &got)\n\t\t\tif (err != nil) != tt.wantErr {\n\t\t\t\tt.Errorf(\"UnmarshalJSON() error = %v, wantErr %v\", err, tt.wantErr)\n\t\t\t\treturn\n\t\t\t}\n\n\t\t\tif !tt.wantErr {\n\t\t\t\tif got.ServerID != tt.want.ServerID {\n\t\t\t\t\tt.Errorf(\"ServerID = %v, want %v\", got.ServerID, tt.want.ServerID)\n\t\t\t\t}\n\t\t\t\tif !stringSlicesEqual(got.Resources, tt.want.Resources) {\n\t\t\t\t\tt.Errorf(\"Resources = %v, want %v\", got.Resources, tt.want.Resources)\n\t\t\t\t}\n\t\t\t\tif !stringSlicesEqual(got.Tools, tt.want.Tools) {\n\t\t\t\t\tt.Errorf(\"Tools = %v, want %v\", got.Tools, tt.want.Tools)\n\t\t\t\t}\n\t\t\t}\n\t\t})\n\t}\n}\n\nfunc TestMCPServers_UnmarshalJSON(t *testing.T) {\n\ttests := []struct {\n\t\tname string\n\t\tinput string\n\t\twant []MCPServerConfig\n\t\twantErr bool\n\t}{\n\t\t{\n\t\t\tname: \"Simple string array\",\n\t\t\tinput: `{\"servers\": [\"server1\", \"server2\", \"server3\"]}`,\n\t\t\twant: []MCPServerConfig{\n\t\t\t\t{ServerID: \"server1\"},\n\t\t\t\t{ServerID: \"server2\"},\n\t\t\t\t{ServerID: \"server3\"},\n\t\t\t},\n\t\t\twantErr: false,\n\t\t},\n\t\t{\n\t\t\tname: \"Mixed formats\",\n\t\t\tinput: `{\"servers\": [\"server1\", {\"server2\": [\"tool1\", \"tool2\"]}, {\"server3\": {\"resources\": [\"res1\"], \"tools\": [\"tool3\"]}}]}`,\n\t\t\twant: []MCPServerConfig{\n\t\t\t\t{ServerID: \"server1\"},\n\t\t\t\t{ServerID: \"server2\", Tools: []string{\"tool1\", \"tool2\"}},\n\t\t\t\t{ServerID: \"server3\", Resources: []string{\"res1\"}, Tools: []string{\"tool3\"}},\n\t\t\t},\n\t\t\twantErr: false,\n\t\t},\n\t\t{\n\t\t\tname: \"Empty servers\",\n\t\t\tinput: `{\"servers\": []}`,\n\t\t\twant: []MCPServerConfig{},\n\t\t\twantErr: false,\n\t\t},\n\t}\n\n\tfor _, tt := range tests {\n\t\tt.Run(tt.name, func(t *testing.T) {\n\t\t\tvar got MCPServers\n\t\t\terr := json.Unmarshal([]byte(tt.input), &got)\n\t\t\tif (err != nil) != tt.wantErr {\n\t\t\t\tt.Errorf(\"UnmarshalJSON() error = %v, wantErr %v\", err, tt.wantErr)\n\t\t\t\treturn\n\t\t\t}\n\n\t\t\tif !tt.wantErr {\n\t\t\t\tif len(got.Servers) != len(tt.want) {\n\t\t\t\t\tt.Errorf(\"got %d servers, want %d\", len(got.Servers), len(tt.want))\n\t\t\t\t\treturn\n\t\t\t\t}\n\n\t\t\t\tfor i := range got.Servers {\n\t\t\t\t\tif got.Servers[i].ServerID != tt.want[i].ServerID {\n\t\t\t\t\t\tt.Errorf(\"Server[%d].ServerID = %v, want %v\", i, got.Servers[i].ServerID, tt.want[i].ServerID)\n\t\t\t\t\t}\n\t\t\t\t\tif !stringSlicesEqual(got.Servers[i].Resources, tt.want[i].Resources) {\n\t\t\t\t\t\tt.Errorf(\"Server[%d].Resources = %v, want %v\", i, got.Servers[i].Resources, tt.want[i].Resources)\n\t\t\t\t\t}\n\t\t\t\t\tif !stringSlicesEqual(got.Servers[i].Tools, tt.want[i].Tools) {\n\t\t\t\t\t\tt.Errorf(\"Server[%d].Tools = %v, want %v\", i, got.Servers[i].Tools, tt.want[i].Tools)\n\t\t\t\t\t}\n\t\t\t\t}\n\t\t\t}\n\t\t})\n\t}\n}\n\nfunc TestMCPServerConfig_MarshalJSON(t *testing.T) {\n\ttests := []struct {\n\t\tname string\n\t\tconfig MCPServerConfig\n\t\twant string\n\t}{\n\t\t{\n\t\t\tname: \"Only ServerID - should be simple string\",\n\t\t\tconfig: MCPServerConfig{\n\t\t\t\tServerID: \"server1\",\n\t\t\t},\n\t\t\twant: `\"server1\"`,\n\t\t},\n\t\t{\n\t\t\tname: \"With Tools - should be object\",\n\t\t\tconfig: MCPServerConfig{\n\t\t\t\tServerID: \"server1\",\n\t\t\t\tTools: []string{\"tool1\", \"tool2\"},\n\t\t\t},\n\t\t\twant: `{\"server_id\":\"server1\",\"tools\":[\"tool1\",\"tool2\"]}`,\n\t\t},\n\t\t{\n\t\t\tname: \"With Resources - should be object\",\n\t\t\tconfig: MCPServerConfig{\n\t\t\t\tServerID: \"server1\",\n\t\t\t\tResources: []string{\"res1\"},\n\t\t\t},\n\t\t\twant: `{\"server_id\":\"server1\",\"resources\":[\"res1\"]}`,\n\t\t},\n\t\t{\n\t\t\tname: \"With Both - should be object\",\n\t\t\tconfig: MCPServerConfig{\n\t\t\t\tServerID: \"server1\",\n\t\t\t\tResources: []string{\"res1\"},\n\t\t\t\tTools: []string{\"tool1\"},\n\t\t\t},\n\t\t\twant: `{\"server_id\":\"server1\",\"resources\":[\"res1\"],\"tools\":[\"tool1\"]}`,\n\t\t},\n\t}\n\n\tfor _, tt := range tests {\n\t\tt.Run(tt.name, func(t *testing.T) {\n\t\t\tgot, err := json.Marshal(tt.config)\n\t\t\tif err != nil {\n\t\t\t\tt.Errorf(\"MarshalJSON() error = %v\", err)\n\t\t\t\treturn\n\t\t\t}\n\t\t\tif string(got) != tt.want {\n\t\t\t\tt.Errorf(\"MarshalJSON() = %s, want %s\", string(got), tt.want)\n\t\t\t}\n\t\t})\n\t}\n}\n\nfunc TestMCPServerConfig_RoundTrip(t *testing.T) {\n\ttests := []struct {\n\t\tname string\n\t\tconfig MCPServerConfig\n\t}{\n\t\t{\n\t\t\tname: \"Simple ServerID\",\n\t\t\tconfig: MCPServerConfig{\n\t\t\t\tServerID: \"server1\",\n\t\t\t},\n\t\t},\n\t\t{\n\t\t\tname: \"With Tools\",\n\t\t\tconfig: MCPServerConfig{\n\t\t\t\tServerID: \"server2\",\n\t\t\t\tTools: []string{\"tool1\", \"tool2\"},\n\t\t\t},\n\t\t},\n\t\t{\n\t\t\tname: \"With Resources and Tools\",\n\t\t\tconfig: MCPServerConfig{\n\t\t\t\tServerID: \"server3\",\n\t\t\t\tResources: []string{\"res1\", \"res2\"},\n\t\t\t\tTools: []string{\"tool3\", \"tool4\"},\n\t\t\t},\n\t\t},\n\t}\n\n\tfor _, tt := range tests {\n\t\tt.Run(tt.name, func(t *testing.T) {\n\t\t\t// Marshal\n\t\t\tdata, err := json.Marshal(tt.config)\n\t\t\tif err != nil {\n\t\t\t\tt.Fatalf(\"Marshal error = %v\", err)\n\t\t\t}\n\n\t\t\t// Unmarshal\n\t\t\tvar got MCPServerConfig\n\t\t\terr = json.Unmarshal(data, &got)\n\t\t\tif err != nil {\n\t\t\t\tt.Fatalf(\"Unmarshal error = %v\", err)\n\t\t\t}\n\n\t\t\t// Compare\n\t\t\tif got.ServerID != tt.config.ServerID {\n\t\t\t\tt.Errorf(\"ServerID = %v, want %v\", got.ServerID, tt.config.ServerID)\n\t\t\t}\n\t\t\tif !stringSlicesEqual(got.Resources, tt.config.Resources) {\n\t\t\t\tt.Errorf(\"Resources = %v, want %v\", got.Resources, tt.config.Resources)\n\t\t\t}\n\t\t\tif !stringSlicesEqual(got.Tools, tt.config.Tools) {\n\t\t\t\tt.Errorf(\"Tools = %v, want %v\", got.Tools, tt.config.Tools)\n\t\t\t}\n\t\t})\n\t}\n}\n\n// Helper function to compare string slices (nil-safe)\nfunc stringSlicesEqual(a, b []string) bool {\n\tif len(a) == 0 && len(b) == 0 {\n\t\treturn true\n\t}\n\tif len(a) != len(b) {\n\t\treturn false\n\t}\n\tfor i := range a {\n\t\tif a[i] != b[i] {\n\t\t\treturn false\n\t\t}\n\t}\n\treturn true\n}\n" }, { "path": "agent/store/types/prompt.go", "content": "package types\n\nimport (\n\t\"os\"\n\t\"path/filepath\"\n\t\"regexp\"\n\t\"strings\"\n\t\"time\"\n\n\t\"github.com/yaoapp/gou/application\"\n\t\"github.com/yaoapp/gou/fs\"\n\t\"gopkg.in/yaml.v3\"\n)\n\n// Prompts is a slice of Prompt with helper methods\ntype Prompts []Prompt\n\n// SystemVariables defines the available system variables\n// These are computed at parse time\nvar SystemVariables = map[string]func() string{\n\t\"TIME\": func() string { return time.Now().Format(\"15:04:05\") },\n\t\"DATE\": func() string { return time.Now().Format(\"2006-01-02\") },\n\t\"DATETIME\": func() string { return time.Now().Format(\"2006-01-02 15:04:05\") },\n\t\"TIMEZONE\": func() string { return time.Now().Location().String() },\n\t\"WEEKDAY\": func() string { return time.Now().Weekday().String() },\n\t\"YEAR\": func() string { return time.Now().Format(\"2006\") },\n\t\"MONTH\": func() string { return time.Now().Format(\"01\") },\n\t\"DAY\": func() string { return time.Now().Format(\"02\") },\n\t\"HOUR\": func() string { return time.Now().Format(\"15\") },\n\t\"MINUTE\": func() string { return time.Now().Format(\"04\") },\n\t\"SECOND\": func() string { return time.Now().Format(\"05\") },\n\t\"UNIX\": func() string { return time.Now().Format(\"1136239445\") }, // Unix timestamp\n}\n\n// Regular expressions for variable replacement\nvar (\n\treSysVar = regexp.MustCompile(`\\$SYS\\.([A-Z_]+)`)\n\treEnvVar = regexp.MustCompile(`\\$ENV\\.([A-Za-z_][A-Za-z0-9_]*)`)\n\treCtxVar = regexp.MustCompile(`\\$CTX\\.([A-Za-z_][A-Za-z0-9_]*)`)\n\treAssetRef = regexp.MustCompile(`@assets/([^\\s]+\\.(md|yml|yaml|json|txt))`)\n)\n\n// LoadPrompts loads prompts from a YAML file\n// Handles @assets/* replacement at load time\n// file: prompt file path relative to app root (e.g., \"assistants/test/prompts.yml\")\n// root: resource root directory for assets (e.g., \"assistants/test\")\n// Returns: prompts slice, modification timestamp, error\nfunc LoadPrompts(file string, root string) ([]Prompt, int64, error) {\n\tapp, err := fs.Get(\"app\")\n\tif err != nil {\n\t\treturn nil, 0, err\n\t}\n\n\tts, err := app.ModTime(file)\n\tif err != nil {\n\t\treturn nil, 0, err\n\t}\n\n\tcontent, err := app.ReadFile(file)\n\tif err != nil {\n\t\treturn nil, 0, err\n\t}\n\n\t// Replace @assets/xxx references with file content\n\tcontent = replaceAssets(content, root, app)\n\n\t// Parse prompts\n\tvar prompts []Prompt\n\terr = yaml.Unmarshal(content, &prompts)\n\tif err != nil {\n\t\treturn nil, 0, err\n\t}\n\n\treturn prompts, ts.UnixNano(), nil\n}\n\n// LoadPromptsRaw loads raw prompt content from a YAML file\n// Handles @assets/* replacement at load time\n// Returns raw YAML string for further processing\nfunc LoadPromptsRaw(file string, root string) (string, int64, error) {\n\tapp, err := fs.Get(\"app\")\n\tif err != nil {\n\t\treturn \"\", 0, err\n\t}\n\n\tts, err := app.ModTime(file)\n\tif err != nil {\n\t\treturn \"\", 0, err\n\t}\n\n\tcontent, err := app.ReadFile(file)\n\tif err != nil {\n\t\treturn \"\", 0, err\n\t}\n\n\t// Replace @assets/xxx references with file content\n\tcontent = replaceAssets(content, root, app)\n\n\treturn string(content), ts.UnixNano(), nil\n}\n\n// LoadGlobalPrompts loads global prompts from agent/prompts.yml\n// Returns: prompts slice, modification timestamp, error\nfunc LoadGlobalPrompts() ([]Prompt, int64, error) {\n\tfile := filepath.Join(\"agent\", \"prompts.yml\")\n\n\t// Check if file exists\n\texists, _ := application.App.Exists(file)\n\tif !exists {\n\t\treturn nil, 0, nil\n\t}\n\n\treturn LoadPrompts(file, \"agent\")\n}\n\n// LoadPromptPresets loads prompt presets from a directory\n// Supports multi-level directories, key is path with \"/\" replaced by \".\"\n// e.g., prompts/chat/friendly.yml -> \"chat.friendly\"\nfunc LoadPromptPresets(dir string, root string) (map[string][]Prompt, int64, error) {\n\tapp, err := fs.Get(\"app\")\n\tif err != nil {\n\t\treturn nil, 0, err\n\t}\n\n\t// Check if directory exists\n\texists, _ := app.Exists(dir)\n\tif !exists {\n\t\treturn nil, 0, nil\n\t}\n\n\t// Read directory recursively\n\tfiles, err := app.ReadDir(dir, true)\n\tif err != nil {\n\t\treturn nil, 0, err\n\t}\n\n\tpresets := make(map[string][]Prompt)\n\tvar latestTs int64\n\n\tfor _, file := range files {\n\t\t// Only process .yml/.yaml files\n\t\tif !strings.HasSuffix(file, \".yml\") && !strings.HasSuffix(file, \".yaml\") {\n\t\t\tcontinue\n\t\t}\n\n\t\tts, err := app.ModTime(file)\n\t\tif err != nil {\n\t\t\treturn nil, 0, err\n\t\t}\n\t\tif ts.UnixNano() > latestTs {\n\t\t\tlatestTs = ts.UnixNano()\n\t\t}\n\n\t\t// Read file content\n\t\tcontent, err := app.ReadFile(file)\n\t\tif err != nil {\n\t\t\treturn nil, 0, err\n\t\t}\n\n\t\t// Replace @assets/xxx references with file content\n\t\tcontent = replaceAssets(content, root, app)\n\n\t\t// Parse prompts\n\t\tvar prompts []Prompt\n\t\terr = yaml.Unmarshal(content, &prompts)\n\t\tif err != nil {\n\t\t\treturn nil, 0, err\n\t\t}\n\n\t\t// Build key: get relative path from dir, remove extension and replace \"/\" with \".\"\n\t\trelPath := strings.TrimPrefix(file, dir+\"/\")\n\t\tkey := strings.TrimSuffix(relPath, filepath.Ext(relPath))\n\t\tkey = strings.ReplaceAll(key, \"/\", \".\")\n\t\tpresets[key] = prompts\n\t}\n\n\treturn presets, latestTs, nil\n}\n\n// replaceAssets replaces @assets/xxx references with file content\nfunc replaceAssets(content []byte, root string, app fs.FileSystem) []byte {\n\treturn reAssetRef.ReplaceAllFunc(content, func(s []byte) []byte {\n\t\tmatches := reAssetRef.FindStringSubmatch(string(s))\n\t\tif len(matches) < 2 {\n\t\t\treturn s\n\t\t}\n\n\t\tasset := matches[1]\n\t\tassetFile := filepath.Join(root, \"assets\", asset)\n\t\tassetContent, err := app.ReadFile(assetFile)\n\t\tif err != nil {\n\t\t\treturn []byte(\"\")\n\t\t}\n\n\t\t// Add proper YAML formatting for content (multiline string)\n\t\tlines := strings.Split(string(assetContent), \"\\n\")\n\t\tformattedContent := \"|\\n\"\n\t\tfor _, line := range lines {\n\t\t\tformattedContent += \" \" + line + \"\\n\"\n\t\t}\n\t\treturn []byte(formattedContent)\n\t})\n}\n\n// Parse parses a single prompt, replacing variables\n// ctx: context variables map, key corresponds to $CTX.{key}\n// Returns a new Prompt with variables replaced\nfunc (p Prompt) Parse(ctx map[string]string) Prompt {\n\tresult := Prompt{\n\t\tRole: p.Role,\n\t\tContent: parseVariables(p.Content, ctx),\n\t\tName: p.Name,\n\t}\n\treturn result\n}\n\n// Parse parses all prompts in the slice, replacing variables\n// ctx: context variables map, key corresponds to $CTX.{key}\n// Returns a new Prompts slice with variables replaced\nfunc (ps Prompts) Parse(ctx map[string]string) Prompts {\n\tresult := make(Prompts, len(ps))\n\tfor i, p := range ps {\n\t\tresult[i] = p.Parse(ctx)\n\t}\n\treturn result\n}\n\n// parseVariables replaces all variable types in content\nfunc parseVariables(content string, ctx map[string]string) string {\n\t// Replace $SYS.* variables\n\tcontent = reSysVar.ReplaceAllStringFunc(content, func(s string) string {\n\t\tmatches := reSysVar.FindStringSubmatch(s)\n\t\tif len(matches) < 2 {\n\t\t\treturn s\n\t\t}\n\t\tvarName := matches[1]\n\t\tif fn, ok := SystemVariables[varName]; ok {\n\t\t\treturn fn()\n\t\t}\n\t\treturn s // Keep original if not found\n\t})\n\n\t// Replace $ENV.* variables\n\tcontent = reEnvVar.ReplaceAllStringFunc(content, func(s string) string {\n\t\tmatches := reEnvVar.FindStringSubmatch(s)\n\t\tif len(matches) < 2 {\n\t\t\treturn s\n\t\t}\n\t\tvarName := matches[1]\n\t\treturn os.Getenv(varName)\n\t})\n\n\t// Replace $CTX.* variables\n\tif ctx != nil {\n\t\tcontent = reCtxVar.ReplaceAllStringFunc(content, func(s string) string {\n\t\t\tmatches := reCtxVar.FindStringSubmatch(s)\n\t\t\tif len(matches) < 2 {\n\t\t\t\treturn s\n\t\t\t}\n\t\t\tvarName := matches[1]\n\t\t\tif val, ok := ctx[varName]; ok {\n\t\t\t\treturn val\n\t\t\t}\n\t\t\treturn \"\" // Empty string if not found in ctx\n\t\t})\n\t}\n\n\treturn content\n}\n\n// Merge merges two prompt slices\n// globalPrompts are prepended to assistantPrompts\nfunc Merge(globalPrompts, assistantPrompts []Prompt) []Prompt {\n\tif len(globalPrompts) == 0 {\n\t\treturn assistantPrompts\n\t}\n\tif len(assistantPrompts) == 0 {\n\t\treturn globalPrompts\n\t}\n\tresult := make([]Prompt, 0, len(globalPrompts)+len(assistantPrompts))\n\tresult = append(result, globalPrompts...)\n\tresult = append(result, assistantPrompts...)\n\treturn result\n}\n" }, { "path": "agent/store/types/prompt_test.go", "content": "package types\n\nimport (\n\t\"os\"\n\t\"testing\"\n\t\"time\"\n\n\t\"github.com/stretchr/testify/assert\"\n)\n\nfunc TestPromptParse(t *testing.T) {\n\ttests := []struct {\n\t\tname string\n\t\tprompt Prompt\n\t\tctx map[string]string\n\t\tvalidate func(t *testing.T, result Prompt)\n\t}{\n\t\t{\n\t\t\tname: \"ParseSysTimeVariables\",\n\t\t\tprompt: Prompt{\n\t\t\t\tRole: \"system\",\n\t\t\t\tContent: \"Current time: $SYS.TIME, Date: $SYS.DATE\",\n\t\t\t},\n\t\t\tctx: nil,\n\t\t\tvalidate: func(t *testing.T, result Prompt) {\n\t\t\t\tassert.Equal(t, \"system\", result.Role)\n\t\t\t\t// Check that variables are replaced (not exact match due to time)\n\t\t\t\tassert.NotContains(t, result.Content, \"$SYS.TIME\")\n\t\t\t\tassert.NotContains(t, result.Content, \"$SYS.DATE\")\n\t\t\t\tassert.Contains(t, result.Content, \"Current time:\")\n\t\t\t\tassert.Contains(t, result.Content, \"Date:\")\n\t\t\t},\n\t\t},\n\t\t{\n\t\t\tname: \"ParseSysDatetimeVariable\",\n\t\t\tprompt: Prompt{\n\t\t\t\tRole: \"system\",\n\t\t\t\tContent: \"Now: $SYS.DATETIME, Timezone: $SYS.TIMEZONE\",\n\t\t\t},\n\t\t\tctx: nil,\n\t\t\tvalidate: func(t *testing.T, result Prompt) {\n\t\t\t\tassert.NotContains(t, result.Content, \"$SYS.DATETIME\")\n\t\t\t\tassert.NotContains(t, result.Content, \"$SYS.TIMEZONE\")\n\t\t\t},\n\t\t},\n\t\t{\n\t\t\tname: \"ParseSysWeekdayVariable\",\n\t\t\tprompt: Prompt{\n\t\t\t\tRole: \"system\",\n\t\t\t\tContent: \"Today is $SYS.WEEKDAY\",\n\t\t\t},\n\t\t\tctx: nil,\n\t\t\tvalidate: func(t *testing.T, result Prompt) {\n\t\t\t\tweekday := time.Now().Weekday().String()\n\t\t\t\tassert.Contains(t, result.Content, weekday)\n\t\t\t},\n\t\t},\n\t\t{\n\t\t\tname: \"ParseSysYearMonthDay\",\n\t\t\tprompt: Prompt{\n\t\t\t\tRole: \"system\",\n\t\t\t\tContent: \"Year: $SYS.YEAR, Month: $SYS.MONTH, Day: $SYS.DAY\",\n\t\t\t},\n\t\t\tctx: nil,\n\t\t\tvalidate: func(t *testing.T, result Prompt) {\n\t\t\t\tnow := time.Now()\n\t\t\t\tassert.Contains(t, result.Content, now.Format(\"2006\"))\n\t\t\t\tassert.Contains(t, result.Content, now.Format(\"01\"))\n\t\t\t\tassert.Contains(t, result.Content, now.Format(\"02\"))\n\t\t\t},\n\t\t},\n\t\t{\n\t\t\tname: \"ParseSysHourMinuteSecond\",\n\t\t\tprompt: Prompt{\n\t\t\t\tRole: \"system\",\n\t\t\t\tContent: \"Hour: $SYS.HOUR, Minute: $SYS.MINUTE, Second: $SYS.SECOND\",\n\t\t\t},\n\t\t\tctx: nil,\n\t\t\tvalidate: func(t *testing.T, result Prompt) {\n\t\t\t\tassert.NotContains(t, result.Content, \"$SYS.HOUR\")\n\t\t\t\tassert.NotContains(t, result.Content, \"$SYS.MINUTE\")\n\t\t\t\tassert.NotContains(t, result.Content, \"$SYS.SECOND\")\n\t\t\t},\n\t\t},\n\t\t{\n\t\t\tname: \"ParseEnvVariable\",\n\t\t\tprompt: Prompt{\n\t\t\t\tRole: \"system\",\n\t\t\t\tContent: \"App: $ENV.TEST_APP_NAME\",\n\t\t\t},\n\t\t\tctx: nil,\n\t\t\tvalidate: func(t *testing.T, result Prompt) {\n\t\t\t\tassert.Contains(t, result.Content, \"App: TestApp\")\n\t\t\t},\n\t\t},\n\t\t{\n\t\t\tname: \"ParseEnvVariableNotFound\",\n\t\t\tprompt: Prompt{\n\t\t\t\tRole: \"system\",\n\t\t\t\tContent: \"Value: $ENV.NOT_EXIST_VAR_12345\",\n\t\t\t},\n\t\t\tctx: nil,\n\t\t\tvalidate: func(t *testing.T, result Prompt) {\n\t\t\t\t// Should be replaced with empty string\n\t\t\t\tassert.Equal(t, \"Value: \", result.Content)\n\t\t\t},\n\t\t},\n\t\t{\n\t\t\tname: \"ParseCtxVariables\",\n\t\t\tprompt: Prompt{\n\t\t\t\tRole: \"system\",\n\t\t\t\tContent: \"User: $CTX.USER_ID, Locale: $CTX.LOCALE\",\n\t\t\t},\n\t\t\tctx: map[string]string{\n\t\t\t\t\"USER_ID\": \"user-123\",\n\t\t\t\t\"LOCALE\": \"zh-CN\",\n\t\t\t},\n\t\t\tvalidate: func(t *testing.T, result Prompt) {\n\t\t\t\tassert.Equal(t, \"User: user-123, Locale: zh-CN\", result.Content)\n\t\t\t},\n\t\t},\n\t\t{\n\t\t\tname: \"ParseCtxVariableNotFound\",\n\t\t\tprompt: Prompt{\n\t\t\t\tRole: \"system\",\n\t\t\t\tContent: \"Value: $CTX.NOT_EXIST\",\n\t\t\t},\n\t\t\tctx: map[string]string{\n\t\t\t\t\"OTHER\": \"value\",\n\t\t\t},\n\t\t\tvalidate: func(t *testing.T, result Prompt) {\n\t\t\t\t// Should be replaced with empty string\n\t\t\t\tassert.Equal(t, \"Value: \", result.Content)\n\t\t\t},\n\t\t},\n\t\t{\n\t\t\tname: \"ParseCtxWithNilMap\",\n\t\t\tprompt: Prompt{\n\t\t\t\tRole: \"system\",\n\t\t\t\tContent: \"Value: $CTX.SOMETHING\",\n\t\t\t},\n\t\t\tctx: nil,\n\t\t\tvalidate: func(t *testing.T, result Prompt) {\n\t\t\t\t// Should keep original when ctx is nil\n\t\t\t\tassert.Equal(t, \"Value: $CTX.SOMETHING\", result.Content)\n\t\t\t},\n\t\t},\n\t\t{\n\t\t\tname: \"ParseMixedVariables\",\n\t\t\tprompt: Prompt{\n\t\t\t\tRole: \"system\",\n\t\t\t\tContent: \"Time: $SYS.TIME, App: $ENV.TEST_APP_NAME, User: $CTX.USER_ID\",\n\t\t\t},\n\t\t\tctx: map[string]string{\n\t\t\t\t\"USER_ID\": \"user-456\",\n\t\t\t},\n\t\t\tvalidate: func(t *testing.T, result Prompt) {\n\t\t\t\tassert.NotContains(t, result.Content, \"$SYS.TIME\")\n\t\t\t\tassert.Contains(t, result.Content, \"App: TestApp\")\n\t\t\t\tassert.Contains(t, result.Content, \"User: user-456\")\n\t\t\t},\n\t\t},\n\t\t{\n\t\t\tname: \"ParseUnknownSysVariable\",\n\t\t\tprompt: Prompt{\n\t\t\t\tRole: \"system\",\n\t\t\t\tContent: \"Value: $SYS.UNKNOWN_VAR\",\n\t\t\t},\n\t\t\tctx: nil,\n\t\t\tvalidate: func(t *testing.T, result Prompt) {\n\t\t\t\t// Should keep original if not found\n\t\t\t\tassert.Equal(t, \"Value: $SYS.UNKNOWN_VAR\", result.Content)\n\t\t\t},\n\t\t},\n\t\t{\n\t\t\tname: \"ParsePreservesRoleAndName\",\n\t\t\tprompt: Prompt{\n\t\t\t\tRole: \"user\",\n\t\t\t\tContent: \"Hello $CTX.NAME\",\n\t\t\t\tName: \"test_user\",\n\t\t\t},\n\t\t\tctx: map[string]string{\n\t\t\t\t\"NAME\": \"World\",\n\t\t\t},\n\t\t\tvalidate: func(t *testing.T, result Prompt) {\n\t\t\t\tassert.Equal(t, \"user\", result.Role)\n\t\t\t\tassert.Equal(t, \"Hello World\", result.Content)\n\t\t\t\tassert.Equal(t, \"test_user\", result.Name)\n\t\t\t},\n\t\t},\n\t\t{\n\t\t\tname: \"ParseCustomCtxVariables\",\n\t\t\tprompt: Prompt{\n\t\t\t\tRole: \"system\",\n\t\t\t\tContent: \"Custom: $CTX.MY_CUSTOM_VAR, Another: $CTX.ANOTHER_VAR\",\n\t\t\t},\n\t\t\tctx: map[string]string{\n\t\t\t\t\"MY_CUSTOM_VAR\": \"custom-value\",\n\t\t\t\t\"ANOTHER_VAR\": \"another-value\",\n\t\t\t},\n\t\t\tvalidate: func(t *testing.T, result Prompt) {\n\t\t\t\tassert.Equal(t, \"Custom: custom-value, Another: another-value\", result.Content)\n\t\t\t},\n\t\t},\n\t\t{\n\t\t\tname: \"ParseMultilineContent\",\n\t\t\tprompt: Prompt{\n\t\t\t\tRole: \"system\",\n\t\t\t\tContent: `# System Context\nCurrent Time: $SYS.TIME\nUser: $CTX.USER_ID\nApp: $ENV.TEST_APP_NAME`,\n\t\t\t},\n\t\t\tctx: map[string]string{\n\t\t\t\t\"USER_ID\": \"user-789\",\n\t\t\t},\n\t\t\tvalidate: func(t *testing.T, result Prompt) {\n\t\t\t\tassert.Contains(t, result.Content, \"# System Context\")\n\t\t\t\tassert.NotContains(t, result.Content, \"$SYS.TIME\")\n\t\t\t\tassert.Contains(t, result.Content, \"User: user-789\")\n\t\t\t\tassert.Contains(t, result.Content, \"App: TestApp\")\n\t\t\t},\n\t\t},\n\t}\n\n\t// Set test environment variable\n\tos.Setenv(\"TEST_APP_NAME\", \"TestApp\")\n\tdefer os.Unsetenv(\"TEST_APP_NAME\")\n\n\tfor _, tt := range tests {\n\t\tt.Run(tt.name, func(t *testing.T) {\n\t\t\tresult := tt.prompt.Parse(tt.ctx)\n\t\t\ttt.validate(t, result)\n\t\t})\n\t}\n}\n\nfunc TestPromptsParse(t *testing.T) {\n\tos.Setenv(\"TEST_APP_NAME\", \"TestApp\")\n\tdefer os.Unsetenv(\"TEST_APP_NAME\")\n\n\tprompts := Prompts{\n\t\t{Role: \"system\", Content: \"Time: $SYS.TIME\"},\n\t\t{Role: \"system\", Content: \"User: $CTX.USER_ID\"},\n\t\t{Role: \"user\", Content: \"App: $ENV.TEST_APP_NAME\"},\n\t}\n\n\tctx := map[string]string{\n\t\t\"USER_ID\": \"user-123\",\n\t}\n\n\tresult := prompts.Parse(ctx)\n\n\tassert.Len(t, result, 3)\n\tassert.NotContains(t, result[0].Content, \"$SYS.TIME\")\n\tassert.Equal(t, \"User: user-123\", result[1].Content)\n\tassert.Equal(t, \"App: TestApp\", result[2].Content)\n}\n\nfunc TestMergePrompts(t *testing.T) {\n\ttests := []struct {\n\t\tname string\n\t\tglobalPrompts []Prompt\n\t\tassistantPrompts []Prompt\n\t\texpectedLen int\n\t\tvalidate func(t *testing.T, result []Prompt)\n\t}{\n\t\t{\n\t\t\tname: \"MergeBothNonEmpty\",\n\t\t\tglobalPrompts: []Prompt{\n\t\t\t\t{Role: \"system\", Content: \"Global prompt 1\"},\n\t\t\t\t{Role: \"system\", Content: \"Global prompt 2\"},\n\t\t\t},\n\t\t\tassistantPrompts: []Prompt{\n\t\t\t\t{Role: \"system\", Content: \"Assistant prompt 1\"},\n\t\t\t},\n\t\t\texpectedLen: 3,\n\t\t\tvalidate: func(t *testing.T, result []Prompt) {\n\t\t\t\tassert.Equal(t, \"Global prompt 1\", result[0].Content)\n\t\t\t\tassert.Equal(t, \"Global prompt 2\", result[1].Content)\n\t\t\t\tassert.Equal(t, \"Assistant prompt 1\", result[2].Content)\n\t\t\t},\n\t\t},\n\t\t{\n\t\t\tname: \"MergeGlobalEmpty\",\n\t\t\tglobalPrompts: []Prompt{},\n\t\t\tassistantPrompts: []Prompt{\n\t\t\t\t{Role: \"system\", Content: \"Assistant prompt\"},\n\t\t\t},\n\t\t\texpectedLen: 1,\n\t\t\tvalidate: func(t *testing.T, result []Prompt) {\n\t\t\t\tassert.Equal(t, \"Assistant prompt\", result[0].Content)\n\t\t\t},\n\t\t},\n\t\t{\n\t\t\tname: \"MergeAssistantEmpty\",\n\t\t\tglobalPrompts: []Prompt{\n\t\t\t\t{Role: \"system\", Content: \"Global prompt\"},\n\t\t\t},\n\t\t\tassistantPrompts: []Prompt{},\n\t\t\texpectedLen: 1,\n\t\t\tvalidate: func(t *testing.T, result []Prompt) {\n\t\t\t\tassert.Equal(t, \"Global prompt\", result[0].Content)\n\t\t\t},\n\t\t},\n\t\t{\n\t\t\tname: \"MergeBothEmpty\",\n\t\t\tglobalPrompts: []Prompt{},\n\t\t\tassistantPrompts: []Prompt{},\n\t\t\texpectedLen: 0,\n\t\t\tvalidate: func(t *testing.T, result []Prompt) {\n\t\t\t\tassert.Empty(t, result)\n\t\t\t},\n\t\t},\n\t\t{\n\t\t\tname: \"MergeGlobalNil\",\n\t\t\tglobalPrompts: nil,\n\t\t\tassistantPrompts: []Prompt{\n\t\t\t\t{Role: \"system\", Content: \"Assistant prompt\"},\n\t\t\t},\n\t\t\texpectedLen: 1,\n\t\t\tvalidate: func(t *testing.T, result []Prompt) {\n\t\t\t\tassert.Equal(t, \"Assistant prompt\", result[0].Content)\n\t\t\t},\n\t\t},\n\t\t{\n\t\t\tname: \"MergeAssistantNil\",\n\t\t\tglobalPrompts: []Prompt{\n\t\t\t\t{Role: \"system\", Content: \"Global prompt\"},\n\t\t\t},\n\t\t\tassistantPrompts: nil,\n\t\t\texpectedLen: 1,\n\t\t\tvalidate: func(t *testing.T, result []Prompt) {\n\t\t\t\tassert.Equal(t, \"Global prompt\", result[0].Content)\n\t\t\t},\n\t\t},\n\t}\n\n\tfor _, tt := range tests {\n\t\tt.Run(tt.name, func(t *testing.T) {\n\t\t\tresult := Merge(tt.globalPrompts, tt.assistantPrompts)\n\t\t\tassert.Len(t, result, tt.expectedLen)\n\t\t\tif tt.validate != nil {\n\t\t\t\ttt.validate(t, result)\n\t\t\t}\n\t\t})\n\t}\n}\n\nfunc TestSystemVariables(t *testing.T) {\n\t// Test that all system variables are defined and return non-empty values\n\texpectedVars := []string{\n\t\t\"TIME\", \"DATE\", \"DATETIME\", \"TIMEZONE\", \"WEEKDAY\",\n\t\t\"YEAR\", \"MONTH\", \"DAY\", \"HOUR\", \"MINUTE\", \"SECOND\", \"UNIX\",\n\t}\n\n\tfor _, varName := range expectedVars {\n\t\tt.Run(varName, func(t *testing.T) {\n\t\t\tfn, ok := SystemVariables[varName]\n\t\t\tassert.True(t, ok, \"SystemVariables should contain %s\", varName)\n\t\t\tvalue := fn()\n\t\t\tassert.NotEmpty(t, value, \"SystemVariables[%s]() should return non-empty value\", varName)\n\t\t})\n\t}\n}\n\nfunc TestParseVariablesEdgeCases(t *testing.T) {\n\tos.Setenv(\"TEST_VAR\", \"test-value\")\n\tdefer os.Unsetenv(\"TEST_VAR\")\n\n\ttests := []struct {\n\t\tname string\n\t\tcontent string\n\t\tctx map[string]string\n\t\texpected string\n\t}{\n\t\t{\n\t\t\tname: \"EmptyContent\",\n\t\t\tcontent: \"\",\n\t\t\tctx: nil,\n\t\t\texpected: \"\",\n\t\t},\n\t\t{\n\t\t\tname: \"NoVariables\",\n\t\t\tcontent: \"Hello, World!\",\n\t\t\tctx: nil,\n\t\t\texpected: \"Hello, World!\",\n\t\t},\n\t\t{\n\t\t\tname: \"PartialVariableSyntax\",\n\t\t\tcontent: \"Value: $SYS Value: $ENV Value: $CTX\",\n\t\t\tctx: nil,\n\t\t\texpected: \"Value: $SYS Value: $ENV Value: $CTX\",\n\t\t},\n\t\t{\n\t\t\tname: \"VariableInMiddleOfWord\",\n\t\t\tcontent: \"prefix$SYS.TIMEsuffix\",\n\t\t\tctx: nil,\n\t\t\texpected: \"prefix$SYS.TIMEsuffix\", // Should not match - variable must be followed by valid char\n\t\t},\n\t\t{\n\t\t\tname: \"MultipleOccurrences\",\n\t\t\tcontent: \"$CTX.VAR and $CTX.VAR again\",\n\t\t\tctx: map[string]string{\"VAR\": \"value\"},\n\t\t\texpected: \"value and value again\",\n\t\t},\n\t\t{\n\t\t\tname: \"SpecialCharactersInValue\",\n\t\t\tcontent: \"User: $CTX.USER\",\n\t\t\tctx: map[string]string{\"USER\": \"user@example.com\"},\n\t\t\texpected: \"User: user@example.com\",\n\t\t},\n\t\t{\n\t\t\tname: \"UnicodeInValue\",\n\t\t\tcontent: \"Name: $CTX.NAME\",\n\t\t\tctx: map[string]string{\"NAME\": \"用户名\"},\n\t\t\texpected: \"Name: 用户名\",\n\t\t},\n\t}\n\n\tfor _, tt := range tests {\n\t\tt.Run(tt.name, func(t *testing.T) {\n\t\t\tresult := parseVariables(tt.content, tt.ctx)\n\t\t\tif tt.name == \"VariableInMiddleOfWord\" {\n\t\t\t\t// This case depends on regex behavior - just check it doesn't crash\n\t\t\t\tassert.NotEmpty(t, result)\n\t\t\t} else {\n\t\t\t\tassert.Equal(t, tt.expected, result)\n\t\t\t}\n\t\t})\n\t}\n}\n" }, { "path": "agent/store/types/sandbox_v2.go", "content": "package types\n\nimport (\n\t\"crypto/sha256\"\n\t\"encoding/json\"\n\t\"fmt\"\n\t\"os\"\n\t\"path/filepath\"\n\t\"sort\"\n\n\tjsoniter \"github.com/json-iterator/go\"\n\t\"github.com/yaoapp/gou/application\"\n\tsandboxTypes \"github.com/yaoapp/yao/agent/sandbox/v2/types\"\n)\n\n// LoadSandboxConfig reads a sandbox.yao file (JSON or YAML) and returns\n// the V2 SandboxConfig. Called during Assistant.Load().\nfunc LoadSandboxConfig(filePath string) (*sandboxTypes.SandboxConfig, error) {\n\tdata, err := os.ReadFile(filePath)\n\tif err != nil {\n\t\treturn nil, fmt.Errorf(\"read sandbox config %s: %w\", filePath, err)\n\t}\n\n\tvar cfg sandboxTypes.SandboxConfig\n\tif err := application.Parse(filepath.Base(filePath), data, &cfg); err != nil {\n\t\treturn nil, fmt.Errorf(\"parse sandbox config: %w\", err)\n\t}\n\n\tif cfg.Version != sandboxTypes.SandboxVersionV2 {\n\t\treturn nil, fmt.Errorf(\"sandbox.yao version must be %q, got %q\", sandboxTypes.SandboxVersionV2, cfg.Version)\n\t}\n\n\treturn &cfg, nil\n}\n\n// ToSandboxV2 converts a generic value (typically map[string]any from DSL\n// parsing) into a V2 SandboxConfig.\nfunc ToSandboxV2(v any) (*sandboxTypes.SandboxConfig, error) {\n\tif v == nil {\n\t\treturn nil, nil\n\t}\n\n\tswitch sb := v.(type) {\n\tcase *sandboxTypes.SandboxConfig:\n\t\treturn sb, nil\n\tcase sandboxTypes.SandboxConfig:\n\t\treturn &sb, nil\n\tdefault:\n\t\traw, err := jsoniter.Marshal(v)\n\t\tif err != nil {\n\t\t\treturn nil, fmt.Errorf(\"sandbox v2 format error: %w\", err)\n\t\t}\n\t\tvar cfg sandboxTypes.SandboxConfig\n\t\tif err := jsoniter.Unmarshal(raw, &cfg); err != nil {\n\t\t\treturn nil, fmt.Errorf(\"sandbox v2 format error: %w\", err)\n\t\t}\n\t\treturn &cfg, nil\n\t}\n}\n\n// ComputeConfigHash computes a SHA-256 fingerprint of the sandbox configuration,\n// MCP servers, and skills directory. Used for hot-reload detection in prepare\n// step \"once\" logic.\nfunc ComputeConfigHash(cfg *sandboxTypes.SandboxConfig, mcpServers []MCPServerConfig, skillsDir string) string {\n\th := sha256.New()\n\n\traw, _ := json.Marshal(cfg)\n\th.Write(raw)\n\n\tif len(mcpServers) > 0 {\n\t\tmcpRaw, _ := json.Marshal(mcpServers)\n\t\th.Write(mcpRaw)\n\t}\n\n\tif skillsDir != \"\" {\n\t\th.Write([]byte(skillsDir))\n\t\tentries, err := os.ReadDir(skillsDir)\n\t\tif err == nil {\n\t\t\tnames := make([]string, 0, len(entries))\n\t\t\tfor _, e := range entries {\n\t\t\t\tnames = append(names, e.Name())\n\t\t\t}\n\t\t\tsort.Strings(names)\n\t\t\tfor _, n := range names {\n\t\t\t\th.Write([]byte(n))\n\t\t\t}\n\t\t}\n\t}\n\n\treturn fmt.Sprintf(\"%x\", h.Sum(nil))\n}\n" }, { "path": "agent/store/types/store.go", "content": "package types\n\n// ChatStore defines the chat storage interface\n// Provides operations for chat, message, and resume management\ntype ChatStore interface {\n\t// ==========================================================================\n\t// Chat Management\n\t// ==========================================================================\n\n\t// CreateChat creates a new chat session\n\t// chat: Chat session to create\n\t// Returns: Potential error\n\tCreateChat(chat *Chat) error\n\n\t// GetChat retrieves a single chat by ID\n\t// chatID: Chat ID\n\t// Returns: Chat information and potential error\n\tGetChat(chatID string) (*Chat, error)\n\n\t// UpdateChat updates chat fields\n\t// chatID: Chat ID\n\t// updates: Map of fields to update\n\t// Returns: Potential error\n\tUpdateChat(chatID string, updates map[string]interface{}) error\n\n\t// DeleteChat deletes a chat and its associated messages\n\t// chatID: Chat ID\n\t// Returns: Potential error\n\tDeleteChat(chatID string) error\n\n\t// ListChats retrieves a paginated list of chats with optional grouping\n\t// filter: Filter conditions including time range, sorting, and grouping\n\t// Returns: Paginated chat list (flat or grouped) and potential error\n\tListChats(filter ChatFilter) (*ChatList, error)\n\n\t// ==========================================================================\n\t// Message Management\n\t// ==========================================================================\n\n\t// SaveMessages batch saves messages for a chat\n\t// This is the primary write method - messages are buffered during execution\n\t// and batch-written at the end of a request\n\t// chatID: Parent chat ID\n\t// messages: Messages to save (includes user input and assistant responses)\n\t// Returns: Potential error\n\tSaveMessages(chatID string, messages []*Message) error\n\n\t// GetMessages retrieves messages for a chat with filtering\n\t// chatID: Chat ID\n\t// filter: Filter conditions (role, type, block, thread, etc.)\n\t// Returns: Message list and potential error\n\tGetMessages(chatID string, filter MessageFilter) ([]*Message, error)\n\n\t// UpdateMessage updates a single message\n\t// messageID: Message ID\n\t// updates: Map of fields to update\n\t// Returns: Potential error\n\tUpdateMessage(messageID string, updates map[string]interface{}) error\n\n\t// DeleteMessages deletes specific messages from a chat\n\t// chatID: Chat ID\n\t// messageIDs: List of message IDs to delete\n\t// Returns: Potential error\n\tDeleteMessages(chatID string, messageIDs []string) error\n\n\t// ==========================================================================\n\t// Resume Management (only called on failure/interrupt)\n\t// ==========================================================================\n\n\t// SaveResume batch saves resume records\n\t// Only called when request is interrupted or failed\n\t// records: Resume records to save\n\t// Returns: Potential error\n\tSaveResume(records []*Resume) error\n\n\t// GetResume retrieves all resume records for a chat\n\t// chatID: Chat ID\n\t// Returns: Resume records and potential error\n\tGetResume(chatID string) ([]*Resume, error)\n\n\t// GetLastResume retrieves the last (most recent) resume record for a chat\n\t// chatID: Chat ID\n\t// Returns: Last resume record and potential error\n\tGetLastResume(chatID string) (*Resume, error)\n\n\t// GetResumeByStackID retrieves resume records for a specific stack\n\t// stackID: Stack ID\n\t// Returns: Resume records and potential error\n\tGetResumeByStackID(stackID string) ([]*Resume, error)\n\n\t// GetStackPath returns the stack path from root to the given stack\n\t// stackID: Current stack ID\n\t// Returns: Stack path [root_stack_id, ..., current_stack_id] and potential error\n\tGetStackPath(stackID string) ([]string, error)\n\n\t// DeleteResume deletes all resume records for a chat\n\t// Called after successful resume to clean up\n\t// chatID: Chat ID\n\t// Returns: Potential error\n\tDeleteResume(chatID string) error\n\n\t// ==========================================================================\n\t// Search Management\n\t// ==========================================================================\n\n\t// SaveSearch saves a search record for a request\n\t// Used for citation support, debugging, and replay\n\t// search: Search record to save\n\t// Returns: Potential error\n\tSaveSearch(search *Search) error\n\n\t// GetSearches retrieves all search records for a request\n\t// requestID: Request ID\n\t// Returns: Search records and potential error\n\tGetSearches(requestID string) ([]*Search, error)\n\n\t// GetReference retrieves a single reference by request ID and index\n\t// Used for citation click handling\n\t// requestID: Request ID\n\t// index: Reference index (1-based)\n\t// Returns: Reference and potential error\n\tGetReference(requestID string, index int) (*Reference, error)\n\n\t// DeleteSearches deletes all search records for a chat\n\t// Called when deleting a chat\n\t// chatID: Chat ID\n\t// Returns: Potential error\n\tDeleteSearches(chatID string) error\n}\n\n// AssistantStore defines the assistant storage interface\n// Separated from ChatStore for clearer responsibility\ntype AssistantStore interface {\n\t// SaveAssistant saves assistant information\n\t// assistant: Assistant information\n\t// Returns: Assistant ID and potential error\n\tSaveAssistant(assistant *AssistantModel) (string, error)\n\n\t// UpdateAssistant updates assistant fields\n\t// assistantID: Assistant ID\n\t// updates: Map of fields to update\n\t// Returns: Potential error\n\tUpdateAssistant(assistantID string, updates map[string]interface{}) error\n\n\t// DeleteAssistant deletes an assistant\n\t// assistantID: Assistant ID\n\t// Returns: Potential error\n\tDeleteAssistant(assistantID string) error\n\n\t// GetAssistants retrieves a paginated list of assistants with filtering\n\t// filter: Filter conditions for querying assistants\n\t// locale: Optional locale for i18n translations\n\t// Returns: Paginated assistant list and potential error\n\tGetAssistants(filter AssistantFilter, locale ...string) (*AssistantList, error)\n\n\t// GetAssistantTags retrieves all unique tags from assistants with filtering\n\t// filter: Filter conditions including QueryFilter for permission filtering\n\t// locale: Optional locale for i18n translations\n\t// Returns: List of tags and potential error\n\tGetAssistantTags(filter AssistantFilter, locale ...string) ([]Tag, error)\n\n\t// GetAssistant retrieves a single assistant by ID\n\t// assistantID: Assistant ID\n\t// fields: List of fields to select, empty/nil means default fields\n\t// locale: Optional locale for i18n translations\n\t// Returns: Assistant information and potential error\n\tGetAssistant(assistantID string, fields []string, locale ...string) (*AssistantModel, error)\n\n\t// DeleteAssistants deletes assistants based on filter conditions\n\t// filter: Filter conditions\n\t// Returns: Number of deleted records and potential error\n\tDeleteAssistants(filter AssistantFilter) (int64, error)\n}\n\n// Store combines ChatStore and AssistantStore interfaces\n// This is the main interface for the storage layer\ntype Store interface {\n\tChatStore\n\tAssistantStore\n}\n\n// SpaceStore defines the interface for Space snapshot operations\n// Note: Space itself uses plan.Space interface, this is for persistence\ntype SpaceStore interface {\n\t// Snapshot returns all key-value pairs in the space\n\tSnapshot() map[string]interface{}\n\n\t// Restore sets multiple key-value pairs from a snapshot\n\tRestore(data map[string]interface{}) error\n}\n" }, { "path": "agent/store/types/types.go", "content": "package types\n\nimport (\n\t\"encoding/json\"\n\t\"fmt\"\n\t\"time\"\n\n\tgraphragtypes \"github.com/yaoapp/gou/graphrag/types\"\n\t\"github.com/yaoapp/xun/dbal/query\"\n\t\"github.com/yaoapp/yao/agent/context\"\n\t\"github.com/yaoapp/yao/agent/i18n\"\n\tsandboxTypes \"github.com/yaoapp/yao/agent/sandbox/v2/types\"\n\tsearchTypes \"github.com/yaoapp/yao/agent/search/types\"\n)\n\n// Setting represents the conversation configuration structure\n// Used to configure basic conversation parameters including connector, user field, table name, etc.\ntype Setting struct {\n\tConnector string `json:\"connector,omitempty\" yaml:\"connector,omitempty\"` // Connector name, default is \"default\"\n\tMaxSize int `json:\"max_size,omitempty\" yaml:\"max_size,omitempty\"` // Maximum storage size limit, default is 20\n\tTTL int `json:\"ttl,omitempty\" yaml:\"ttl,omitempty\"` // Time To Live in seconds, default is 90 * 24 * 60 * 60 (90 days)\n\tOptions map[string]interface{} `json:\"optional,omitempty\" yaml:\"optional,omitempty\"` // The options for the store\n}\n\n// =============================================================================\n// Chat Types\n// =============================================================================\n\n// Chat represents a chat session\ntype Chat struct {\n\tChatID string `json:\"chat_id\"`\n\tTitle string `json:\"title,omitempty\"`\n\tAssistantID string `json:\"assistant_id\"`\n\tLastConnector string `json:\"last_connector,omitempty\"` // Last used connector ID (updated on each message)\n\tLastMode string `json:\"last_mode,omitempty\"` // Last used chat mode (updated on each message)\n\tStatus string `json:\"status\"` // \"active\" or \"archived\"\n\tPublic bool `json:\"public\"` // Whether shared across all teams\n\tShare string `json:\"share\"` // \"private\" or \"team\"\n\tSort int `json:\"sort\"` // Sort order for display\n\tLastMessageAt *time.Time `json:\"last_message_at,omitempty\"`\n\tMetadata map[string]interface{} `json:\"metadata,omitempty\"`\n\tCreatedAt time.Time `json:\"created_at\"`\n\tUpdatedAt time.Time `json:\"updated_at\"`\n\n\t// Permission fields (managed by Yao framework when permission: true)\n\tCreatedBy string `json:\"__yao_created_by,omitempty\"` // User ID who created the record\n\tUpdatedBy string `json:\"__yao_updated_by,omitempty\"` // User ID who last updated\n\tTeamID string `json:\"__yao_team_id,omitempty\"` // Team ID for team-level access\n\tTenantID string `json:\"__yao_tenant_id,omitempty\"` // Tenant ID for multi-tenancy\n}\n\n// ChatFilter for listing chats\ntype ChatFilter struct {\n\tUserID string `json:\"user_id,omitempty\"`\n\tTeamID string `json:\"team_id,omitempty\"`\n\tAssistantID string `json:\"assistant_id,omitempty\"`\n\tStatus string `json:\"status,omitempty\"`\n\tKeywords string `json:\"keywords,omitempty\"`\n\tChatIDPrefix string `json:\"chat_id_prefix,omitempty\"`\n\n\t// Time range filter\n\tStartTime *time.Time `json:\"start_time,omitempty\"` // Filter chats after this time\n\tEndTime *time.Time `json:\"end_time,omitempty\"` // Filter chats before this time\n\tTimeField string `json:\"time_field,omitempty\"` // Field for time filter: \"created_at\" or \"last_message_at\" (default)\n\n\t// Sorting\n\tOrderBy string `json:\"order_by,omitempty\"` // Field to sort by (default: \"last_message_at\")\n\tOrder string `json:\"order,omitempty\"` // Sort order: \"desc\" (default) or \"asc\"\n\n\t// Response format\n\tGroupBy string `json:\"group_by,omitempty\"` // \"time\" for time-based groups, empty for flat list\n\n\t// Pagination\n\tPage int `json:\"page,omitempty\"`\n\tPageSize int `json:\"pagesize,omitempty\"`\n\n\t// Permission filter (not serialized)\n\tQueryFilter func(query.Query) `json:\"-\"` // Custom query function for permission filtering\n}\n\n// ChatList paginated response with time-based grouping\ntype ChatList struct {\n\tData []*Chat `json:\"data\"`\n\tGroups []*ChatGroup `json:\"groups,omitempty\"` // Time-based groups for UI display\n\tPage int `json:\"page\"`\n\tPageSize int `json:\"pagesize\"`\n\tPageCount int `json:\"pagecount\"`\n\tTotal int `json:\"total\"`\n}\n\n// ChatGroup represents a time-based group of chats\ntype ChatGroup struct {\n\tLabel string `json:\"label\"` // \"Today\", \"Yesterday\", \"This Week\", \"This Month\", \"Earlier\"\n\tKey string `json:\"key\"` // \"today\", \"yesterday\", \"this_week\", \"this_month\", \"earlier\"\n\tChats []*Chat `json:\"chats\"` // Chats in this group\n\tCount int `json:\"count\"` // Number of chats in group\n}\n\n// =============================================================================\n// Message Types\n// =============================================================================\n\n// Message represents a chat message\ntype Message struct {\n\tMessageID string `json:\"message_id\"`\n\tChatID string `json:\"chat_id\"`\n\tRequestID string `json:\"request_id,omitempty\"`\n\tRole string `json:\"role\"` // \"user\" or \"assistant\"\n\tType string `json:\"type\"` // \"text\", \"image\", \"loading\", \"tool_call\", \"retrieval\", etc.\n\tProps map[string]interface{} `json:\"props\"`\n\tBlockID string `json:\"block_id,omitempty\"`\n\tThreadID string `json:\"thread_id,omitempty\"`\n\tAssistantID string `json:\"assistant_id,omitempty\"`\n\tConnector string `json:\"connector,omitempty\"` // Connector ID used for this message\n\tMode string `json:\"mode,omitempty\"` // Chat mode used for this message (chat or task)\n\tSequence int `json:\"sequence\"`\n\tMetadata map[string]interface{} `json:\"metadata,omitempty\"`\n\tCreatedAt time.Time `json:\"created_at\"`\n\tUpdatedAt time.Time `json:\"updated_at\"`\n}\n\n// MessageFilter for listing messages\ntype MessageFilter struct {\n\tRequestID string `json:\"request_id,omitempty\"`\n\tRole string `json:\"role,omitempty\"`\n\tBlockID string `json:\"block_id,omitempty\"`\n\tThreadID string `json:\"thread_id,omitempty\"`\n\tType string `json:\"type,omitempty\"`\n\tLimit int `json:\"limit,omitempty\"`\n\tOffset int `json:\"offset,omitempty\"`\n}\n\n// =============================================================================\n// Resume Types (for recovery from interruption/failure)\n// =============================================================================\n\n// Resume represents an execution state for recovery\n// Only stored when request is interrupted or failed\ntype Resume struct {\n\tResumeID string `json:\"resume_id\"`\n\tChatID string `json:\"chat_id\"`\n\tRequestID string `json:\"request_id\"`\n\tAssistantID string `json:\"assistant_id\"`\n\tStackID string `json:\"stack_id\"`\n\tStackParentID string `json:\"stack_parent_id,omitempty\"`\n\tStackDepth int `json:\"stack_depth\"`\n\tType string `json:\"type\"` // \"input\", \"hook_create\", \"llm\", \"tool\", \"hook_next\", \"delegate\"\n\tStatus string `json:\"status\"` // \"failed\" or \"interrupted\"\n\tInput map[string]interface{} `json:\"input,omitempty\"`\n\tOutput map[string]interface{} `json:\"output,omitempty\"`\n\tSpaceSnapshot map[string]interface{} `json:\"space_snapshot,omitempty\"` // Shared space data for recovery\n\tError string `json:\"error,omitempty\"`\n\tSequence int `json:\"sequence\"`\n\tMetadata map[string]interface{} `json:\"metadata,omitempty\"`\n\tCreatedAt time.Time `json:\"created_at\"`\n\tUpdatedAt time.Time `json:\"updated_at\"`\n}\n\n// ResumeStatus constants\nconst (\n\tResumeStatusFailed = \"failed\"\n\tResumeStatusInterrupted = \"interrupted\"\n)\n\n// ResumeType constants\nconst (\n\tResumeTypeInput = \"input\"\n\tResumeTypeHookCreate = \"hook_create\"\n\tResumeTypeLLM = \"llm\"\n\tResumeTypeTool = \"tool\"\n\tResumeTypeHookNext = \"hook_next\"\n\tResumeTypeDelegate = \"delegate\"\n)\n\n// AssistantFilter represents the assistant filter structure\n// Used for filtering and pagination when retrieving assistant lists\ntype AssistantFilter struct {\n\tTags []string `json:\"tags,omitempty\"` // Filter by tags\n\tType string `json:\"type,omitempty\"` // Filter by type (single value)\n\tTypes []string `json:\"types,omitempty\"` // Filter by types (multiple values, IN query)\n\tKeywords string `json:\"keywords,omitempty\"` // Search in name and description\n\tConnector string `json:\"connector,omitempty\"` // Filter by connector\n\tAssistantID string `json:\"assistant_id,omitempty\"` // Filter by assistant ID\n\tAssistantIDs []string `json:\"assistant_ids,omitempty\"` // Filter by assistant IDs\n\tMentionable *bool `json:\"mentionable,omitempty\"` // Filter by mentionable status\n\tAutomated *bool `json:\"automated,omitempty\"` // Filter by automation status\n\tBuiltIn *bool `json:\"built_in,omitempty\"` // Filter by built-in status\n\tSandbox *bool `json:\"sandbox,omitempty\"` // Filter by sandbox configuration (true=has sandbox, false=no sandbox)\n\tPage int `json:\"page,omitempty\"` // Page number, starting from 1\n\tPageSize int `json:\"pagesize,omitempty\"` // Items per page\n\tSelect []string `json:\"select,omitempty\"` // Fields to return, returns all fields if empty\n\tQueryFilter func(query.Query) `json:\"-\"` // Custom query function for permission filtering (not serialized)\n}\n\n// AssistantList represents the paginated assistant list response structure\n// Used for returning paginated assistant lists with metadata\ntype AssistantList struct {\n\tData []*AssistantModel `json:\"data\"` // List of assistants\n\tPage int `json:\"page\"` // Current page number (1-based)\n\tPageSize int `json:\"pagesize\"` // Number of items per page\n\tPageCount int `json:\"pagecount\"` // Total number of pages\n\tNext int `json:\"next\"` // Next page number (0 if no next page)\n\tPrev int `json:\"prev\"` // Previous page number (0 if no previous page)\n\tTotal int `json:\"total\"` // Total number of items across all pages\n}\n\n// AssistantInfo contains basic assistant information for display\n// Used in chat history to show assistant details with i18n support\ntype AssistantInfo struct {\n\tAssistantID string `json:\"assistant_id\"`\n\tName string `json:\"name\"`\n\tAvatar string `json:\"avatar,omitempty\"`\n\tDescription string `json:\"description,omitempty\"`\n\tConnector string `json:\"connector,omitempty\"`\n\tConnectorOptions *ConnectorOptions `json:\"connector_options,omitempty\"`\n\tModes []string `json:\"modes,omitempty\"`\n\tDefaultMode string `json:\"default_mode,omitempty\"`\n\tSandbox bool `json:\"sandbox,omitempty\"`\n\tComputerFilter *sandboxTypes.ComputerFilter `json:\"computer_filter,omitempty\"`\n}\n\n// Tag represents a tag\ntype Tag struct {\n\tValue string `json:\"value\"`\n\tLabel string `json:\"label\"`\n}\n\n// Prompt a prompt\ntype Prompt struct {\n\tRole string `json:\"role\"`\n\tContent string `json:\"content\"`\n\tName string `json:\"name,omitempty\"`\n}\n\n// KBSetting Knowledge Base configuration for agent (from agent/kb.yml)\ntype KBSetting struct {\n\tChat *ChatKBSetting `json:\"chat,omitempty\" yaml:\"chat,omitempty\"` // Chat session KB settings\n}\n\n// ChatKBSetting represents KB settings for chat sessions\ntype ChatKBSetting struct {\n\tEmbeddingProviderID string `json:\"embedding_provider_id\" yaml:\"embedding_provider_id\"` // Embedding provider ID\n\tEmbeddingOptionID string `json:\"embedding_option_id\" yaml:\"embedding_option_id\"` // Embedding option ID\n\tLocale string `json:\"locale,omitempty\" yaml:\"locale,omitempty\"` // Locale for content processing\n\tConfig *graphragtypes.CreateCollectionOptions `json:\"config,omitempty\" yaml:\"config,omitempty\"` // Vector index configuration\n\tMetadata map[string]interface{} `json:\"metadata,omitempty\" yaml:\"metadata,omitempty\"` // Collection metadata defaults\n\tDocumentDefaults *DocumentDefaults `json:\"document_defaults,omitempty\" yaml:\"document_defaults,omitempty\"` // Document processing defaults\n}\n\n// DocumentDefaults represents default settings for document processing\ntype DocumentDefaults struct {\n\tChunking *ProviderOption `json:\"chunking,omitempty\" yaml:\"chunking,omitempty\"` // Chunking provider configuration\n\tExtraction *ProviderOption `json:\"extraction,omitempty\" yaml:\"extraction,omitempty\"` // Extraction provider configuration\n\tConverter *ProviderOption `json:\"converter,omitempty\" yaml:\"converter,omitempty\"` // Converter provider configuration\n}\n\n// ProviderOption represents a provider and option ID pair\ntype ProviderOption struct {\n\tProviderID string `json:\"provider_id\" yaml:\"provider_id\"` // Provider ID\n\tOptionID string `json:\"option_id\" yaml:\"option_id\"` // Option ID within the provider\n}\n\n// KnowledgeBase the knowledge base configuration\ntype KnowledgeBase struct {\n\tCollections []string `json:\"collections,omitempty\"` // Knowledge base collection IDs\n\tOptions map[string]interface{} `json:\"options,omitempty\"` // Additional options for knowledge base\n}\n\n// Database the database configuration\ntype Database struct {\n\tModels []string `json:\"models,omitempty\"` // Database models\n\tOptions map[string]interface{} `json:\"options,omitempty\"` // Additional options for database\n}\n\n// MCPServers the MCP servers configuration\n// Supports multiple formats in the servers array:\n// - Simple string: \"server_id\"\n// - With tools: {\"server_id\": [\"tool1\", \"tool2\"]}\n// - With resources and tools: {\"server_id\": {\"resources\": [...], \"tools\": [...]}}\ntype MCPServers struct {\n\tServers []MCPServerConfig `json:\"servers,omitempty\"` // MCP server configurations\n\tOptions map[string]interface{} `json:\"options,omitempty\"` // Additional options for MCP servers\n}\n\n// MCPServerConfig represents a single MCP server configuration\ntype MCPServerConfig struct {\n\tServerID string `json:\"server_id,omitempty\"` // MCP server ID\n\tResources []string `json:\"resources,omitempty\"` // Resources to use (optional)\n\tTools []string `json:\"tools,omitempty\"` // Tools to use (optional)\n}\n\n// UnmarshalJSON implements custom JSON unmarshaling for MCPServerConfig\n// Supports multiple input formats:\n// 1. Simple string: \"server_id\"\n// 2. Standard object: {\"server_id\": \"server1\", \"resources\": [...], \"tools\": [...]}\n// 3. Tools array: {\"server_id\": [\"tool1\", \"tool2\"]}\n// 4. Full config: {\"server_id\": {\"resources\": [...], \"tools\": [...]}}\nfunc (m *MCPServerConfig) UnmarshalJSON(data []byte) error {\n\t// Try to unmarshal as string first\n\tvar str string\n\tif err := json.Unmarshal(data, &str); err == nil {\n\t\tm.ServerID = str\n\t\treturn nil\n\t}\n\n\t// Try to unmarshal as standard object with server_id field\n\ttype Alias MCPServerConfig\n\tvar stdObj Alias\n\tif err := json.Unmarshal(data, &stdObj); err == nil && stdObj.ServerID != \"\" {\n\t\t*m = MCPServerConfig(stdObj)\n\t\treturn nil\n\t}\n\n\t// Try to unmarshal as object with single key (alternative formats)\n\tvar obj map[string]json.RawMessage\n\tif err := json.Unmarshal(data, &obj); err != nil {\n\t\treturn err\n\t}\n\n\t// Should have exactly one key (the server ID)\n\tif len(obj) != 1 {\n\t\treturn fmt.Errorf(\"MCPServerConfig object must have exactly one key or server_id field\")\n\t}\n\n\t// Get the server ID (the only key)\n\tfor serverID, value := range obj {\n\t\tm.ServerID = serverID\n\n\t\t// Try to unmarshal value as array of strings (format c: tools only)\n\t\tvar tools []string\n\t\tif err := json.Unmarshal(value, &tools); err == nil {\n\t\t\tm.Tools = tools\n\t\t\treturn nil\n\t\t}\n\n\t\t// Try to unmarshal as object with resources and tools (format b)\n\t\tvar detail struct {\n\t\t\tResources []string `json:\"resources,omitempty\"`\n\t\t\tTools []string `json:\"tools,omitempty\"`\n\t\t}\n\t\tif err := json.Unmarshal(value, &detail); err == nil {\n\t\t\tm.Resources = detail.Resources\n\t\t\tm.Tools = detail.Tools\n\t\t\treturn nil\n\t\t}\n\n\t\treturn fmt.Errorf(\"invalid format for server '%s'\", serverID)\n\t}\n\n\treturn nil\n}\n\n// MarshalJSON implements custom JSON marshaling for MCPServerConfig\n// Serializes to different formats based on content:\n// 1. If only ServerID: \"server_id\"\n// 2. If has Resources or Tools: {\"server_id\": \"...\", \"resources\": [...], \"tools\": [...]}\nfunc (m MCPServerConfig) MarshalJSON() ([]byte, error) {\n\t// If only ServerID, serialize as simple string\n\tif len(m.Resources) == 0 && len(m.Tools) == 0 {\n\t\treturn json.Marshal(m.ServerID)\n\t}\n\n\t// Otherwise, use standard object format\n\ttype Alias MCPServerConfig\n\treturn json.Marshal(Alias(m))\n}\n\n// Workflow the workflow configuration\ntype Workflow struct {\n\tWorkflows []string `json:\"workflows,omitempty\"` // Workflow IDs\n\tOptions map[string]interface{} `json:\"options,omitempty\"` // Additional workflow options\n}\n\n// Sandbox the sandbox configuration for coding agents (Claude CLI, Cursor CLI)\ntype Sandbox struct {\n\tCommand string `json:\"command\"` // Command type: \"claude\" or \"cursor\"\n\tImage string `json:\"image,omitempty\"` // Docker image (optional, auto-selected by command)\n\tMaxMemory string `json:\"max_memory,omitempty\"` // Memory limit (e.g., \"4g\")\n\tMaxCPU float64 `json:\"max_cpu,omitempty\"` // CPU limit (e.g., 2.0)\n\tTimeout string `json:\"timeout,omitempty\"` // Execution timeout (e.g., \"10m\")\n\tArguments map[string]interface{} `json:\"arguments,omitempty\"` // Command-specific arguments\n\tSecrets map[string]string `json:\"secrets,omitempty\"` // Secrets to pass to container (e.g., GITHUB_TOKEN: \"$ENV.GITHUB_TOKEN\")\n}\n\n// Tool represents a tool configuration for storage\ntype Tool struct {\n\tType string `json:\"type,omitempty\"`\n\tName string `json:\"name\"`\n\tDescription string `json:\"description,omitempty\"`\n\tParameters map[string]interface{} `json:\"parameters,omitempty\"`\n}\n\n// ToolCalls the tool calls\ntype ToolCalls struct {\n\tTools []Tool `json:\"tools,omitempty\"`\n\tPrompts []Prompt `json:\"prompts,omitempty\"`\n}\n\n// Placeholder the assistant placeholder\ntype Placeholder struct {\n\tTitle string `json:\"title,omitempty\"`\n\tDescription string `json:\"description,omitempty\"`\n\tPrompts []string `json:\"prompts,omitempty\"`\n}\n\n// ModelCapability defines the available model capability filters\ntype ModelCapability string\n\n// Model capability constants for filtering connectors\nconst (\n\tCapVision ModelCapability = \"vision\"\n\tCapAudio ModelCapability = \"audio\"\n\tCapToolCalls ModelCapability = \"tool_calls\"\n\tCapReasoning ModelCapability = \"reasoning\"\n\tCapStreaming ModelCapability = \"streaming\"\n\tCapJSON ModelCapability = \"json\"\n\tCapMultimodal ModelCapability = \"multimodal\"\n\tCapTemperatureAdjustable ModelCapability = \"temperature_adjustable\"\n)\n\n// ConnectorOptions the connector selection options\n// Allows defining optional connector selection with filtering capabilities\ntype ConnectorOptions struct {\n\tOptional *bool `json:\"optional\"` // Whether connector is optional for user selection (nil=default, false=hidden, true=shown)\n\tConnectors []string `json:\"connectors,omitempty\"` // List of available connectors, empty means all connectors are available\n\tFilters []ModelCapability `json:\"filters,omitempty\"` // Filter by model capabilities, conditions can be stacked\n}\n\n// AssistantModel the assistant database model\ntype AssistantModel struct {\n\tID string `json:\"assistant_id\"` // Assistant ID\n\tType string `json:\"type,omitempty\"` // Assistant Type, default is assistant\n\tName string `json:\"name,omitempty\"` // Assistant Name\n\tAvatar string `json:\"avatar,omitempty\"` // Assistant Avatar\n\tConnector string `json:\"connector\"` // AI Connector (default connector)\n\tConnectorOptions *ConnectorOptions `json:\"connector_options,omitempty\"` // Connector selection options for user to choose from\n\tPath string `json:\"path,omitempty\"` // Assistant Path\n\tBuiltIn bool `json:\"built_in,omitempty\"` // Whether this is a built-in assistant\n\tSort int `json:\"sort,omitempty\"` // Assistant Sort\n\tDescription string `json:\"description,omitempty\"` // Assistant Description\n\tCapabilities string `json:\"capabilities,omitempty\"` // Assistant capabilities description (useful for Robot orchestration)\n\tTags []string `json:\"tags,omitempty\"` // Assistant Tags\n\tModes []string `json:\"modes,omitempty\"` // Supported modes (e.g., [\"task\", \"chat\"]), null means all modes are supported\n\tDefaultMode string `json:\"default_mode,omitempty\"` // Default mode, can be empty\n\tReadonly bool `json:\"readonly,omitempty\"` // Whether this assistant is readonly\n\tPublic bool `json:\"public,omitempty\"` // Whether this assistant is shared across all teams in the platform\n\tShare string `json:\"share,omitempty\"` // Assistant sharing scope (private/team)\n\tMentionable bool `json:\"mentionable,omitempty\"` // Whether this assistant is mentionable\n\tAutomated bool `json:\"automated,omitempty\"` // Whether this assistant is automated\n\tOptions map[string]interface{} `json:\"options,omitempty\"` // AI Options\n\tPrompts []Prompt `json:\"prompts,omitempty\"` // AI Prompts (default prompts)\n\tPromptPresets map[string][]Prompt `json:\"prompt_presets,omitempty\"` // Prompt presets organized by mode (e.g., \"chat\", \"task\", etc.)\n\tDisableGlobalPrompts bool `json:\"disable_global_prompts,omitempty\"` // Whether to disable global prompts, default is false\n\tKB *KnowledgeBase `json:\"kb,omitempty\"` // Knowledge base configuration\n\tDB *Database `json:\"db,omitempty\"` // Database configuration\n\tMCP *MCPServers `json:\"mcp,omitempty\"` // MCP servers configuration\n\tWorkflow *Workflow `json:\"workflow,omitempty\"` // Workflow configuration\n\tSandbox *Sandbox `json:\"sandbox,omitempty\"` // Sandbox configuration for coding agents (V1)\n\tSandboxV2 *sandboxTypes.SandboxConfig `json:\"-\"` // V2 sandbox configuration (runtime only, not persisted in DB)\n\tIsSandbox bool `json:\"-\"` // Whether this is a Sandbox assistant (derived from SandboxV2 presence)\n\tComputerFilter *sandboxTypes.ComputerFilter `json:\"-\"` // Computer filter from DSL sandbox.filter (runtime only)\n\tConfigHash string `json:\"-\"` // V2 sandbox config fingerprint for hot-reload\n\tPlaceholder *Placeholder `json:\"placeholder,omitempty\"` // Assistant Placeholder\n\tSource string `json:\"source,omitempty\"` // Hook script source code\n\tLocales i18n.Map `json:\"locales,omitempty\"` // Assistant Locales\n\tUses *context.Uses `json:\"uses,omitempty\"` // Assistant-specific wrapper configurations for vision, audio, etc. If not set, use global settings\n\tSearch *searchTypes.Config `json:\"search,omitempty\"` // Search configuration (web, kb, db, citation, weights, etc.)\n\tDependencies map[string]string `json:\"dependencies,omitempty\"` // Dependencies on other MCP Clients (name -> version constraint)\n\tCreatedAt int64 `json:\"created_at\"` // Creation timestamp\n\tUpdatedAt int64 `json:\"updated_at\"` // Last update timestamp\n\n\t// Permission management fields (not exposed in JSON API responses)\n\tYaoCreatedBy string `json:\"-\"` // User who created the assistant (not exposed in JSON)\n\tYaoUpdatedBy string `json:\"-\"` // User who last updated the assistant (not exposed in JSON)\n\tYaoTeamID string `json:\"-\"` // Team ID for team-based access control (not exposed in JSON)\n\tYaoTenantID string `json:\"-\"` // Tenant ID for multi-tenancy support (not exposed in JSON)\n}\n\n// =============================================================================\n// Search Types (for search result storage)\n// =============================================================================\n\n// Search represents stored search results for a request\n// Stores all intermediate processing results for debugging, replay, and citation\ntype Search struct {\n\tID int64 `json:\"id\"`\n\tRequestID string `json:\"request_id\"`\n\tChatID string `json:\"chat_id\"`\n\tQuery string `json:\"query\"` // Original query\n\tConfig map[string]any `json:\"config,omitempty\"` // Search config used (for tuning)\n\tKeywords []string `json:\"keywords,omitempty\"` // Extracted keywords (Web/NLP)\n\tEntities []Entity `json:\"entities,omitempty\"` // Extracted entities (Graph)\n\tRelations []Relation `json:\"relations,omitempty\"` // Extracted relations (Graph)\n\tDSL map[string]any `json:\"dsl,omitempty\"` // Generated QueryDSL (DB)\n\tSource string `json:\"source\"` // web/kb/db/auto\n\tReferences []Reference `json:\"references\"` // References with global index\n\tGraph []GraphNode `json:\"graph,omitempty\"` // Graph nodes from KB\n\tXML string `json:\"xml,omitempty\"` // Formatted XML for LLM\n\tPrompt string `json:\"prompt,omitempty\"` // Citation prompt\n\tDuration int64 `json:\"duration_ms\"` // Search duration in ms\n\tError string `json:\"error,omitempty\"` // Error if failed\n\tCreatedAt time.Time `json:\"created_at\"`\n}\n\n// Reference represents a single reference with global index (for storage)\ntype Reference struct {\n\tIndex int `json:\"index\"` // Global index (1-based, unique within request)\n\tType string `json:\"type\"` // web/kb/db\n\tTitle string `json:\"title\"` // Reference title\n\tURL string `json:\"url,omitempty\"` // URL (for web)\n\tSnippet string `json:\"snippet,omitempty\"` // Short snippet\n\tContent string `json:\"content,omitempty\"` // Full content\n\tMetadata map[string]any `json:\"metadata,omitempty\"`\n}\n\n// SearchFilter for listing searches\ntype SearchFilter struct {\n\tRequestID string `json:\"request_id,omitempty\"`\n\tChatID string `json:\"chat_id,omitempty\"`\n\tSource string `json:\"source,omitempty\"`\n}\n\n// Entity represents an extracted entity (for Graph RAG)\ntype Entity struct {\n\tName string `json:\"name\"`\n\tType string `json:\"type,omitempty\"`\n\tSource string `json:\"source,omitempty\"`\n}\n\n// Relation represents an extracted relation (for Graph RAG)\ntype Relation struct {\n\tSubject string `json:\"subject\"`\n\tPredicate string `json:\"predicate\"`\n\tObject string `json:\"object\"`\n\tSource string `json:\"source,omitempty\"`\n}\n\n// GraphNode represents a node from knowledge graph\ntype GraphNode struct {\n\tID string `json:\"id\"`\n\tType string `json:\"type\"`\n\tLabel string `json:\"label,omitempty\"`\n\tProperties map[string]any `json:\"properties,omitempty\"`\n\tScore float64 `json:\"score,omitempty\"`\n}\n" }, { "path": "agent/store/xun/assistant.go", "content": "package xun\n\nimport (\n\t\"fmt\"\n\t\"math\"\n\t\"time\"\n\n\tjsoniter \"github.com/json-iterator/go\"\n\t\"github.com/yaoapp/kun/log\"\n\t\"github.com/yaoapp/xun/dbal/query\"\n\t\"github.com/yaoapp/yao/agent/context\"\n\t\"github.com/yaoapp/yao/agent/i18n\"\n\tsearchTypes \"github.com/yaoapp/yao/agent/search/types\"\n\t\"github.com/yaoapp/yao/agent/store/types\"\n)\n\n// SaveAssistant saves assistant information\nfunc (store *Xun) SaveAssistant(assistant *types.AssistantModel) (string, error) {\n\tif assistant == nil {\n\t\treturn \"\", fmt.Errorf(\"assistant cannot be nil\")\n\t}\n\n\t// Validate required fields\n\tif assistant.Name == \"\" {\n\t\treturn \"\", fmt.Errorf(\"field name is required\")\n\t}\n\tif assistant.Type == \"\" {\n\t\treturn \"\", fmt.Errorf(\"field type is required\")\n\t}\n\tif assistant.Connector == \"\" {\n\t\treturn \"\", fmt.Errorf(\"field connector is required\")\n\t}\n\n\t// Generate assistant_id if not provided\n\tif assistant.ID == \"\" {\n\t\tvar err error\n\t\tassistant.ID, err = store.GenerateAssistantID()\n\t\tif err != nil {\n\t\t\treturn \"\", err\n\t\t}\n\t}\n\n\t// Check if assistant exists\n\texists, err := store.query.New().\n\t\tTable(store.getAssistantTable()).\n\t\tWhere(\"assistant_id\", assistant.ID).\n\t\tExists()\n\tif err != nil {\n\t\treturn \"\", err\n\t}\n\n\t// Convert model to map for database storage\n\tdata := make(map[string]interface{})\n\tdata[\"assistant_id\"] = assistant.ID\n\tdata[\"type\"] = assistant.Type\n\tdata[\"connector\"] = assistant.Connector\n\tdata[\"built_in\"] = assistant.BuiltIn\n\tdata[\"sort\"] = assistant.Sort\n\tdata[\"readonly\"] = assistant.Readonly\n\tdata[\"public\"] = assistant.Public\n\tdata[\"mentionable\"] = assistant.Mentionable\n\tdata[\"automated\"] = assistant.Automated\n\tdata[\"disable_global_prompts\"] = assistant.DisableGlobalPrompts\n\n\t// Set timestamps\n\tnow := time.Now().UnixNano()\n\tif exists {\n\t\t// Update: set updated_at, keep created_at unchanged\n\t\tif assistant.UpdatedAt == 0 {\n\t\t\tdata[\"updated_at\"] = now\n\t\t} else {\n\t\t\tdata[\"updated_at\"] = assistant.UpdatedAt\n\t\t}\n\t\t// Don't modify created_at on update\n\t} else {\n\t\t// Create: set created_at, updated_at is null\n\t\tif assistant.CreatedAt == 0 {\n\t\t\tdata[\"created_at\"] = now\n\t\t} else {\n\t\t\tdata[\"created_at\"] = assistant.CreatedAt\n\t\t}\n\t\tdata[\"updated_at\"] = nil\n\t}\n\n\t// Handle nullable string fields from assistant.mod.yao\n\t// Store as nil if empty string (this matches database nullable: true fields)\n\tif assistant.Name != \"\" {\n\t\tdata[\"name\"] = assistant.Name\n\t} else {\n\t\tdata[\"name\"] = nil\n\t}\n\tif assistant.Avatar != \"\" {\n\t\tdata[\"avatar\"] = assistant.Avatar\n\t} else {\n\t\tdata[\"avatar\"] = nil\n\t}\n\tif assistant.Description != \"\" {\n\t\tdata[\"description\"] = assistant.Description\n\t} else {\n\t\tdata[\"description\"] = nil\n\t}\n\tif assistant.Capabilities != \"\" {\n\t\tdata[\"capabilities\"] = assistant.Capabilities\n\t} else {\n\t\tdata[\"capabilities\"] = nil\n\t}\n\tif assistant.Path != \"\" {\n\t\tdata[\"path\"] = assistant.Path\n\t} else {\n\t\tdata[\"path\"] = nil\n\t}\n\tif assistant.Source != \"\" {\n\t\tdata[\"source\"] = assistant.Source\n\t} else {\n\t\tdata[\"source\"] = nil\n\t}\n\n\t// Share field: nullable: false with default \"private\"\n\t// Apply default if empty\n\tif assistant.Share != \"\" {\n\t\tdata[\"share\"] = assistant.Share\n\t} else {\n\t\tdata[\"share\"] = \"private\" // Apply default value\n\t}\n\n\t// Permission management fields - store as nil if empty\n\tif assistant.YaoCreatedBy != \"\" {\n\t\tdata[\"__yao_created_by\"] = assistant.YaoCreatedBy\n\t} else {\n\t\tdata[\"__yao_created_by\"] = nil\n\t}\n\tif assistant.YaoUpdatedBy != \"\" {\n\t\tdata[\"__yao_updated_by\"] = assistant.YaoUpdatedBy\n\t} else {\n\t\tdata[\"__yao_updated_by\"] = nil\n\t}\n\tif assistant.YaoTeamID != \"\" {\n\t\tdata[\"__yao_team_id\"] = assistant.YaoTeamID\n\t} else {\n\t\tdata[\"__yao_team_id\"] = nil\n\t}\n\tif assistant.YaoTenantID != \"\" {\n\t\tdata[\"__yao_tenant_id\"] = assistant.YaoTenantID\n\t} else {\n\t\tdata[\"__yao_tenant_id\"] = nil\n\t}\n\n\t// DefaultMode is a simple string field\n\tif assistant.DefaultMode != \"\" {\n\t\tdata[\"default_mode\"] = assistant.DefaultMode\n\t} else {\n\t\tdata[\"default_mode\"] = nil\n\t}\n\n\t// Handle all JSON fields uniformly via marshalJSONFields.\n\t// Uses isNil() to correctly skip typed nils stored in interface{}.\n\tjsonFields := map[string]interface{}{\n\t\t\"options\": assistant.Options,\n\t\t\"tags\": assistant.Tags,\n\t\t\"modes\": assistant.Modes,\n\t\t\"prompts\": assistant.Prompts,\n\t\t\"prompt_presets\": assistant.PromptPresets,\n\t\t\"connector_options\": assistant.ConnectorOptions,\n\t\t\"kb\": assistant.KB,\n\t\t\"db\": assistant.DB,\n\t\t\"mcp\": assistant.MCP,\n\t\t\"workflow\": assistant.Workflow,\n\t\t\"sandbox\": assistant.Sandbox,\n\t\t\"placeholder\": assistant.Placeholder,\n\t\t\"locales\": assistant.Locales,\n\t\t\"uses\": assistant.Uses,\n\t\t\"search\": assistant.Search,\n\t\t\"dependencies\": assistant.Dependencies,\n\t}\n\n\tif err := marshalJSONFields(data, jsonFields); err != nil {\n\t\treturn \"\", err\n\t}\n\n\t// Update or insert\n\tif exists {\n\t\t_, err := store.query.New().\n\t\t\tTable(store.getAssistantTable()).\n\t\t\tWhere(\"assistant_id\", assistant.ID).\n\t\t\tUpdate(data)\n\t\tif err != nil {\n\t\t\treturn \"\", err\n\t\t}\n\t\treturn assistant.ID, nil\n\t}\n\n\terr = store.query.New().\n\t\tTable(store.getAssistantTable()).\n\t\tInsert(data)\n\tif err != nil {\n\t\treturn \"\", err\n\t}\n\treturn assistant.ID, nil\n}\n\n// UpdateAssistant updates specific fields of an assistant\nfunc (store *Xun) UpdateAssistant(assistantID string, updates map[string]interface{}) error {\n\tif assistantID == \"\" {\n\t\treturn fmt.Errorf(\"assistant_id is required\")\n\t}\n\tif len(updates) == 0 {\n\t\treturn fmt.Errorf(\"no fields to update\")\n\t}\n\n\t// Check if assistant exists\n\texists, err := store.query.New().\n\t\tTable(store.getAssistantTable()).\n\t\tWhere(\"assistant_id\", assistantID).\n\t\tExists()\n\tif err != nil {\n\t\treturn err\n\t}\n\tif !exists {\n\t\treturn fmt.Errorf(\"assistant %s not found\", assistantID)\n\t}\n\n\t// Prepare update data\n\tdata := make(map[string]interface{})\n\n\t// List of fields that need JSON marshaling\n\tjsonFields := []string{\"options\", \"tags\", \"modes\", \"prompts\", \"prompt_presets\", \"connector_options\", \"kb\", \"db\", \"mcp\", \"workflow\", \"sandbox\", \"placeholder\", \"locales\", \"uses\", \"search\", \"dependencies\"}\n\tjsonFieldSet := make(map[string]bool)\n\tfor _, field := range jsonFields {\n\t\tjsonFieldSet[field] = true\n\t}\n\n\t// List of nullable string fields\n\tnullableStringFields := []string{\"name\", \"avatar\", \"description\", \"capabilities\", \"path\", \"source\", \"default_mode\", \"__yao_created_by\", \"__yao_updated_by\", \"__yao_team_id\", \"__yao_tenant_id\"}\n\tnullableFieldSet := make(map[string]bool)\n\tfor _, field := range nullableStringFields {\n\t\tnullableFieldSet[field] = true\n\t}\n\n\t// Process each update field\n\tfor key, value := range updates {\n\t\t// Skip system fields that shouldn't be updated directly\n\t\tif key == \"assistant_id\" || key == \"created_at\" {\n\t\t\tcontinue\n\t\t}\n\n\t\t// Handle JSON fields\n\t\tif jsonFieldSet[key] {\n\t\t\tif isNil(value) {\n\t\t\t\tdata[key] = nil\n\t\t\t} else {\n\t\t\t\tjsonStr, err := jsoniter.MarshalToString(value)\n\t\t\t\tif err != nil {\n\t\t\t\t\treturn fmt.Errorf(\"failed to marshal %s: %w\", key, err)\n\t\t\t\t}\n\t\t\t\tdata[key] = jsonStr\n\t\t\t}\n\t\t} else {\n\t\t\t// Handle regular fields\n\t\t\t// Convert empty strings to nil for nullable fields\n\t\t\tif strVal, ok := value.(string); ok && strVal == \"\" && nullableFieldSet[key] {\n\t\t\t\tdata[key] = nil\n\t\t\t\tcontinue\n\t\t\t}\n\t\t\tdata[key] = value\n\t\t}\n\t}\n\n\t// Always update updated_at timestamp\n\tdata[\"updated_at\"] = types.ToMySQLTime(time.Now().UnixNano())\n\n\tif len(data) == 0 {\n\t\treturn fmt.Errorf(\"no valid fields to update\")\n\t}\n\n\t// Perform update\n\t_, err = store.query.New().\n\t\tTable(store.getAssistantTable()).\n\t\tWhere(\"assistant_id\", assistantID).\n\t\tUpdate(data)\n\n\treturn err\n}\n\n// DeleteAssistant deletes an assistant by assistant_id\nfunc (store *Xun) DeleteAssistant(assistantID string) error {\n\t// Check if assistant exists\n\texists, err := store.query.New().\n\t\tTable(store.getAssistantTable()).\n\t\tWhere(\"assistant_id\", assistantID).\n\t\tExists()\n\tif err != nil {\n\t\treturn err\n\t}\n\n\tif !exists {\n\t\treturn fmt.Errorf(\"assistant %s not found\", assistantID)\n\t}\n\n\t_, err = store.query.New().\n\t\tTable(store.getAssistantTable()).\n\t\tWhere(\"assistant_id\", assistantID).\n\t\tDelete()\n\treturn err\n}\n\n// GetAssistants retrieves assistants with pagination and filtering\nfunc (store *Xun) GetAssistants(filter types.AssistantFilter, locale ...string) (*types.AssistantList, error) {\n\tqb := store.query.New().\n\t\tTable(store.getAssistantTable())\n\n\t// Apply tag filter if provided\n\tif len(filter.Tags) > 0 {\n\t\tqb.Where(func(qb query.Query) {\n\t\t\tfor i, tag := range filter.Tags {\n\t\t\t\t// For each tag, we need to match it as part of a JSON array\n\t\t\t\t// This will match both single tag arrays [\"tag1\"] and multi-tag arrays [\"tag1\",\"tag2\"]\n\t\t\t\tpattern := fmt.Sprintf(\"%%\\\"%s\\\"%%\", tag)\n\t\t\t\tif i == 0 {\n\t\t\t\t\tqb.Where(\"tags\", \"like\", pattern)\n\t\t\t\t} else {\n\t\t\t\t\tqb.OrWhere(\"tags\", \"like\", pattern)\n\t\t\t\t}\n\t\t\t}\n\t\t})\n\t}\n\n\t// Apply keyword filter if provided\n\tif filter.Keywords != \"\" {\n\t\tqb.Where(func(qb query.Query) {\n\t\t\tqb.Where(\"name\", \"like\", fmt.Sprintf(\"%%%s%%\", filter.Keywords)).\n\t\t\t\tOrWhere(\"description\", \"like\", fmt.Sprintf(\"%%%s%%\", filter.Keywords)).\n\t\t\t\tOrWhere(\"capabilities\", \"like\", fmt.Sprintf(\"%%%s%%\", filter.Keywords)).\n\t\t\t\tOrWhere(\"locales\", \"like\", fmt.Sprintf(\"%%%s%%\", filter.Keywords))\n\t\t})\n\t}\n\n\t// Apply type filter if provided (single value)\n\tif filter.Type != \"\" {\n\t\tqb.Where(\"type\", filter.Type)\n\t}\n\n\t// Apply types filter if provided (multiple values, IN query)\n\tif len(filter.Types) > 0 {\n\t\tqb.WhereIn(\"type\", filter.Types)\n\t}\n\n\t// Apply connector filter if provided\n\tif filter.Connector != \"\" {\n\t\tqb.Where(\"connector\", filter.Connector)\n\t}\n\n\t// Apply assistant_id filter if provided\n\tif filter.AssistantID != \"\" {\n\t\tqb.Where(\"assistant_id\", filter.AssistantID)\n\t}\n\n\t// Apply assistantIDs filter if provided\n\tif len(filter.AssistantIDs) > 0 {\n\t\tqb.WhereIn(\"assistant_id\", filter.AssistantIDs)\n\t}\n\n\t// Apply mentionable filter if provided\n\tif filter.Mentionable != nil {\n\t\tqb.Where(\"mentionable\", *filter.Mentionable)\n\t}\n\n\t// Apply automated filter if provided\n\tif filter.Automated != nil {\n\t\tqb.Where(\"automated\", *filter.Automated)\n\t}\n\n\t// Apply built_in filter if provided\n\tif filter.BuiltIn != nil {\n\t\tqb.Where(\"built_in\", *filter.BuiltIn)\n\t}\n\n\t// Apply sandbox filter (true = has sandbox config, false = no sandbox config)\n\t// MySQL JSON columns distinguish between SQL NULL and JSON literal null.\n\t// CAST(sandbox AS CHAR) returns 'null' for JSON null and NULL for SQL NULL.\n\tif filter.Sandbox != nil {\n\t\tif *filter.Sandbox {\n\t\t\tqb.WhereNotNull(\"sandbox\").\n\t\t\t\tWhereRaw(\"CAST(`sandbox` AS CHAR) <> 'null'\")\n\t\t} else {\n\t\t\tqb.Where(func(qb query.Query) {\n\t\t\t\tqb.WhereNull(\"sandbox\").\n\t\t\t\t\tOrWhereRaw(\"CAST(`sandbox` AS CHAR) = 'null'\")\n\t\t\t})\n\t\t}\n\t}\n\n\t// Apply custom query filter function (for permission filtering)\n\tif filter.QueryFilter != nil {\n\t\tqb.Where(filter.QueryFilter)\n\t}\n\n\t// Set defaults for pagination\n\tif filter.PageSize <= 0 {\n\t\tfilter.PageSize = 20\n\t}\n\tif filter.Page <= 0 {\n\t\tfilter.Page = 1\n\t}\n\n\t// Get total count\n\ttotal, err := qb.Clone().Count()\n\tif err != nil {\n\t\treturn nil, err\n\t}\n\n\t// Calculate pagination\n\toffset := (filter.Page - 1) * filter.PageSize\n\ttotalPages := int(math.Ceil(float64(total) / float64(filter.PageSize)))\n\tnextPage := filter.Page + 1\n\tif nextPage > totalPages {\n\t\tnextPage = 0\n\t}\n\tprevPage := filter.Page - 1\n\tif prevPage < 1 {\n\t\tprevPage = 0\n\t}\n\n\t// Apply select fields with security validation (only if fields are explicitly specified)\n\tif len(filter.Select) > 0 {\n\t\t// ValidateAssistantFields will validate fields against whitelist\n\t\tsanitized := types.ValidateAssistantFields(filter.Select)\n\t\tselectFields := make([]interface{}, len(sanitized))\n\t\tfor i, field := range sanitized {\n\t\t\tselectFields[i] = field\n\t\t}\n\t\tqb.Select(selectFields...)\n\t}\n\t// If no select fields specified, query will return all fields (SELECT *)\n\n\t// Get paginated results\n\trows, err := qb.OrderBy(\"sort\", \"asc\").\n\t\tOrderBy(\"updated_at\", \"desc\").\n\t\tOffset(offset).\n\t\tLimit(filter.PageSize).\n\t\tGet()\n\tif err != nil {\n\t\treturn nil, err\n\t}\n\n\t// Convert rows to types.AssistantModel slice\n\tassistants := make([]*types.AssistantModel, 0, len(rows))\n\tjsonFields := []string{\"tags\", \"options\", \"prompts\", \"prompt_presets\", \"connector_options\", \"workflow\", \"sandbox\", \"kb\", \"mcp\", \"placeholder\", \"locales\", \"uses\", \"search\", \"dependencies\"}\n\n\tfor _, row := range rows {\n\t\tdata := row.ToMap()\n\t\tif data == nil {\n\t\t\tcontinue\n\t\t}\n\n\t\t// Parse JSON fields\n\t\tstore.parseJSONFields(data, jsonFields)\n\n\t\t// Convert map to types.AssistantModel using existing helper function\n\t\tmodel, err := types.ToAssistantModel(data)\n\t\tif err != nil {\n\t\t\tlog.Error(\"Failed to convert row to types.AssistantModel: %s\", err.Error())\n\t\t\tcontinue\n\t\t}\n\n\t\t// Apply i18n translations if locale is provided\n\t\tif len(locale) > 0 && locale[0] != \"\" && model != nil {\n\t\t\tstore.translate(model, model.ID, locale[0])\n\t\t}\n\n\t\tassistants = append(assistants, model)\n\t}\n\n\treturn &types.AssistantList{\n\t\tData: assistants,\n\t\tPage: filter.Page,\n\t\tPageSize: filter.PageSize,\n\t\tPageCount: totalPages,\n\t\tNext: nextPage,\n\t\tPrev: prevPage,\n\t\tTotal: int(total),\n\t}, nil\n}\n\n// GetAssistant retrieves a single assistant by ID\nfunc (store *Xun) GetAssistant(assistantID string, fields []string, locale ...string) (*types.AssistantModel, error) {\n\tqb := store.query.New().\n\t\tTable(store.getAssistantTable()).\n\t\tWhere(\"assistant_id\", assistantID)\n\n\t// Apply select fields with security validation\n\t// If no fields specified, use default fields\n\tfieldsToSelect := fields\n\tif len(fieldsToSelect) == 0 {\n\t\tfieldsToSelect = types.AssistantDefaultFields\n\t}\n\n\t// ValidateAssistantFields will validate fields against whitelist\n\tsanitized := types.ValidateAssistantFields(fieldsToSelect)\n\tselectFields := make([]interface{}, len(sanitized))\n\tfor i, field := range sanitized {\n\t\tselectFields[i] = field\n\t}\n\tqb.Select(selectFields...)\n\n\trow, err := qb.First()\n\tif err != nil {\n\t\treturn nil, err\n\t}\n\n\tif row == nil {\n\t\treturn nil, fmt.Errorf(\"assistant %s not found\", assistantID)\n\t}\n\n\tdata := row.ToMap()\n\tif len(data) == 0 {\n\t\treturn nil, fmt.Errorf(\"the assistant %s is empty\", assistantID)\n\t}\n\n\t// Parse JSON fields\n\tjsonFields := []string{\"tags\", \"modes\", \"options\", \"prompts\", \"prompt_presets\", \"connector_options\", \"workflow\", \"sandbox\", \"kb\", \"db\", \"mcp\", \"placeholder\", \"locales\", \"uses\", \"search\", \"dependencies\"}\n\tstore.parseJSONFields(data, jsonFields)\n\n\t// Convert map to types.AssistantModel\n\tmodel := &types.AssistantModel{\n\t\tID: getString(data, \"assistant_id\"),\n\t\tType: getString(data, \"type\"),\n\t\tName: getString(data, \"name\"),\n\t\tAvatar: getString(data, \"avatar\"),\n\t\tConnector: getString(data, \"connector\"),\n\t\tPath: getString(data, \"path\"),\n\t\tSource: getString(data, \"source\"),\n\t\tBuiltIn: getBool(data, \"built_in\"),\n\t\tSort: getInt(data, \"sort\"),\n\t\tDescription: getString(data, \"description\"),\n\t\tCapabilities: getString(data, \"capabilities\"),\n\t\tDefaultMode: getString(data, \"default_mode\"),\n\t\tReadonly: getBool(data, \"readonly\"),\n\t\tPublic: getBool(data, \"public\"),\n\t\tShare: getString(data, \"share\"),\n\t\tMentionable: getBool(data, \"mentionable\"),\n\t\tAutomated: getBool(data, \"automated\"),\n\t\tDisableGlobalPrompts: getBool(data, \"disable_global_prompts\"),\n\t\tCreatedAt: getInt64(data, \"created_at\"),\n\t\tUpdatedAt: getInt64(data, \"updated_at\"),\n\t\tYaoCreatedBy: getString(data, \"__yao_created_by\"),\n\t\tYaoUpdatedBy: getString(data, \"__yao_updated_by\"),\n\t\tYaoTeamID: getString(data, \"__yao_team_id\"),\n\t\tYaoTenantID: getString(data, \"__yao_tenant_id\"),\n\t}\n\n\t// Handle Tags\n\tif tags, ok := data[\"tags\"].([]interface{}); ok {\n\t\tmodel.Tags = make([]string, len(tags))\n\t\tfor i, tag := range tags {\n\t\t\tif s, ok := tag.(string); ok {\n\t\t\t\tmodel.Tags[i] = s\n\t\t\t}\n\t\t}\n\t}\n\n\t// Handle Modes\n\tif modes, ok := data[\"modes\"].([]interface{}); ok {\n\t\tmodel.Modes = make([]string, len(modes))\n\t\tfor i, mode := range modes {\n\t\t\tif s, ok := mode.(string); ok {\n\t\t\t\tmodel.Modes[i] = s\n\t\t\t}\n\t\t}\n\t}\n\n\t// Handle Options\n\tif options, ok := data[\"options\"].(map[string]interface{}); ok {\n\t\tmodel.Options = options\n\t}\n\n\t// Handle typed fields with conversion\n\tif prompts, has := data[\"prompts\"]; has && prompts != nil {\n\t\t// Try to unmarshal to []Prompt\n\t\traw, err := jsoniter.Marshal(prompts)\n\t\tif err == nil {\n\t\t\tvar p []types.Prompt\n\t\t\tif err := jsoniter.Unmarshal(raw, &p); err == nil {\n\t\t\t\tmodel.Prompts = p\n\t\t\t}\n\t\t}\n\t}\n\n\tif promptPresets, has := data[\"prompt_presets\"]; has && promptPresets != nil {\n\t\traw, err := jsoniter.Marshal(promptPresets)\n\t\tif err == nil {\n\t\t\tvar pp map[string][]types.Prompt\n\t\t\tif err := jsoniter.Unmarshal(raw, &pp); err == nil {\n\t\t\t\tmodel.PromptPresets = pp\n\t\t\t}\n\t\t}\n\t}\n\n\tif connectorOptions, has := data[\"connector_options\"]; has && connectorOptions != nil {\n\t\traw, err := jsoniter.Marshal(connectorOptions)\n\t\tif err == nil {\n\t\t\tvar co types.ConnectorOptions\n\t\t\tif err := jsoniter.Unmarshal(raw, &co); err == nil {\n\t\t\t\tmodel.ConnectorOptions = &co\n\t\t\t}\n\t\t}\n\t}\n\n\tif kb, has := data[\"kb\"]; has && kb != nil {\n\t\tkbConverted, err := types.ToKnowledgeBase(kb)\n\t\tif err == nil {\n\t\t\tmodel.KB = kbConverted\n\t\t}\n\t}\n\n\tif db, has := data[\"db\"]; has && db != nil {\n\t\tdbConverted, err := types.ToDatabase(db)\n\t\tif err == nil {\n\t\t\tmodel.DB = dbConverted\n\t\t}\n\t}\n\n\tif mcp, has := data[\"mcp\"]; has && mcp != nil {\n\t\tmcpConverted, err := types.ToMCPServers(mcp)\n\t\tif err == nil {\n\t\t\tmodel.MCP = mcpConverted\n\t\t}\n\t}\n\n\tif workflow, has := data[\"workflow\"]; has && workflow != nil {\n\t\twf, err := types.ToWorkflow(workflow)\n\t\tif err == nil {\n\t\t\tmodel.Workflow = wf\n\t\t}\n\t}\n\n\tif sandbox, has := data[\"sandbox\"]; has && sandbox != nil {\n\t\tsb, err := types.ToSandbox(sandbox)\n\t\tif err == nil {\n\t\t\tmodel.Sandbox = sb\n\t\t}\n\t}\n\n\tif placeholder, has := data[\"placeholder\"]; has && placeholder != nil {\n\t\traw, err := jsoniter.Marshal(placeholder)\n\t\tif err == nil {\n\t\t\tvar ph types.Placeholder\n\t\t\tif err := jsoniter.Unmarshal(raw, &ph); err == nil {\n\t\t\t\tmodel.Placeholder = &ph\n\t\t\t}\n\t\t}\n\t}\n\n\tif locales, has := data[\"locales\"]; has && locales != nil {\n\t\traw, err := jsoniter.Marshal(locales)\n\t\tif err == nil {\n\t\t\tvar loc i18n.Map\n\t\t\tif err := jsoniter.Unmarshal(raw, &loc); err == nil {\n\t\t\t\tmodel.Locales = loc\n\t\t\t}\n\t\t}\n\t}\n\n\tif uses, has := data[\"uses\"]; has && uses != nil {\n\t\traw, err := jsoniter.Marshal(uses)\n\t\tif err == nil {\n\t\t\tvar u context.Uses\n\t\t\tif err := jsoniter.Unmarshal(raw, &u); err == nil {\n\t\t\t\tmodel.Uses = &u\n\t\t\t}\n\t\t}\n\t}\n\n\tif search, has := data[\"search\"]; has && search != nil {\n\t\traw, err := jsoniter.Marshal(search)\n\t\tif err == nil {\n\t\t\tvar s searchTypes.Config\n\t\t\tif err := jsoniter.Unmarshal(raw, &s); err == nil {\n\t\t\t\tmodel.Search = &s\n\t\t\t}\n\t\t}\n\t}\n\n\tif deps, has := data[\"dependencies\"]; has && deps != nil {\n\t\traw, err := jsoniter.Marshal(deps)\n\t\tif err == nil {\n\t\t\tvar d map[string]string\n\t\t\tif err := jsoniter.Unmarshal(raw, &d); err == nil {\n\t\t\t\tmodel.Dependencies = d\n\t\t\t}\n\t\t}\n\t}\n\n\t// Apply i18n translation if locale is provided\n\tif len(locale) > 0 && locale[0] != \"\" {\n\t\tstore.translate(model, assistantID, locale[0])\n\t}\n\n\treturn model, nil\n}\n\n// DeleteAssistants deletes assistants based on filter conditions\nfunc (store *Xun) DeleteAssistants(filter types.AssistantFilter) (int64, error) {\n\tqb := store.query.New().\n\t\tTable(store.getAssistantTable())\n\n\t// Apply tag filter if provided\n\tif len(filter.Tags) > 0 {\n\t\tqb.Where(func(qb query.Query) {\n\t\t\tfor i, tag := range filter.Tags {\n\t\t\t\tpattern := fmt.Sprintf(\"%%\\\"%s\\\"%%\", tag)\n\t\t\t\tif i == 0 {\n\t\t\t\t\tqb.Where(\"tags\", \"like\", pattern)\n\t\t\t\t} else {\n\t\t\t\t\tqb.OrWhere(\"tags\", \"like\", pattern)\n\t\t\t\t}\n\t\t\t}\n\t\t})\n\t}\n\n\t// Apply keyword filter if provided\n\tif filter.Keywords != \"\" {\n\t\tqb.Where(func(qb query.Query) {\n\t\t\tqb.Where(\"name\", \"like\", fmt.Sprintf(\"%%%s%%\", filter.Keywords)).\n\t\t\t\tOrWhere(\"description\", \"like\", fmt.Sprintf(\"%%%s%%\", filter.Keywords))\n\t\t})\n\t}\n\n\t// Apply connector filter if provided\n\tif filter.Connector != \"\" {\n\t\tqb.Where(\"connector\", filter.Connector)\n\t}\n\n\t// Apply assistant_id filter if provided\n\tif filter.AssistantID != \"\" {\n\t\tqb.Where(\"assistant_id\", filter.AssistantID)\n\t}\n\n\t// Apply assistantIDs filter if provided\n\tif len(filter.AssistantIDs) > 0 {\n\t\tqb.WhereIn(\"assistant_id\", filter.AssistantIDs)\n\t}\n\n\t// Apply mentionable filter if provided\n\tif filter.Mentionable != nil {\n\t\tqb.Where(\"mentionable\", *filter.Mentionable)\n\t}\n\n\t// Apply automated filter if provided\n\tif filter.Automated != nil {\n\t\tqb.Where(\"automated\", *filter.Automated)\n\t}\n\n\t// Apply built_in filter if provided\n\tif filter.BuiltIn != nil {\n\t\tqb.Where(\"built_in\", *filter.BuiltIn)\n\t}\n\n\t// Execute delete and return number of deleted records\n\treturn qb.Delete()\n}\n\n// GetAssistantTags retrieves all unique tags from assistants with filtering\nfunc (store *Xun) GetAssistantTags(filter types.AssistantFilter, locale ...string) ([]types.Tag, error) {\n\tqb := store.query.New().Table(store.getAssistantTable())\n\n\t// Apply type filter (default to \"assistant\")\n\ttypeFilter := \"assistant\"\n\tif filter.Type != \"\" {\n\t\ttypeFilter = filter.Type\n\t}\n\tqb.Where(\"type\", typeFilter)\n\n\t// Apply custom query filter function (for permission filtering)\n\tif filter.QueryFilter != nil {\n\t\tqb.Where(filter.QueryFilter)\n\t}\n\n\t// Apply other filters if provided\n\tif filter.Connector != \"\" {\n\t\tqb.Where(\"connector\", filter.Connector)\n\t}\n\n\tif filter.BuiltIn != nil {\n\t\tqb.Where(\"built_in\", *filter.BuiltIn)\n\t}\n\n\tif filter.Mentionable != nil {\n\t\tqb.Where(\"mentionable\", *filter.Mentionable)\n\t}\n\n\tif filter.Automated != nil {\n\t\tqb.Where(\"automated\", *filter.Automated)\n\t}\n\n\t// Apply keyword filter if provided\n\tif filter.Keywords != \"\" {\n\t\tqb.Where(func(qb query.Query) {\n\t\t\tqb.Where(\"name\", \"like\", fmt.Sprintf(\"%%%s%%\", filter.Keywords)).\n\t\t\t\tOrWhere(\"description\", \"like\", fmt.Sprintf(\"%%%s%%\", filter.Keywords))\n\t\t})\n\t}\n\n\trows, err := qb.Select(\"tags\").GroupBy(\"tags\").Get()\n\tif err != nil {\n\t\treturn nil, err\n\t}\n\n\ttagSet := map[string]bool{}\n\tfor _, row := range rows {\n\t\tif tags, ok := row[\"tags\"].(string); ok && tags != \"\" {\n\t\t\tvar tagList []string\n\t\t\tif err := jsoniter.UnmarshalFromString(tags, &tagList); err == nil {\n\t\t\t\tfor _, tag := range tagList {\n\t\t\t\t\ttagSet[tag] = true\n\t\t\t\t}\n\t\t\t}\n\t\t}\n\t}\n\n\tlang := \"en\"\n\tif len(locale) > 0 {\n\t\tlang = locale[0]\n\t}\n\n\t// Convert map keys to slice\n\ttags := make([]types.Tag, 0, len(tagSet))\n\tfor tag := range tagSet {\n\t\ttags = append(tags, types.Tag{\n\t\t\tValue: tag,\n\t\t\tLabel: i18n.TranslateGlobal(lang, tag).(string),\n\t\t})\n\t}\n\treturn tags, nil\n}\n\n// translate applies i18n translation to assistant model fields\nfunc (store *Xun) translate(model *types.AssistantModel, assistantID string, locale string) {\n\tif model == nil {\n\t\treturn\n\t}\n\n\t// Translate name\n\tif translated := i18n.Translate(assistantID, locale, model.Name); translated != nil {\n\t\tif s, ok := translated.(string); ok {\n\t\t\tmodel.Name = s\n\t\t}\n\t}\n\n\t// Translate description\n\tif translated := i18n.Translate(assistantID, locale, model.Description); translated != nil {\n\t\tif s, ok := translated.(string); ok {\n\t\t\tmodel.Description = s\n\t\t}\n\t}\n\n\t// Translate capabilities\n\tif translated := i18n.Translate(assistantID, locale, model.Capabilities); translated != nil {\n\t\tif s, ok := translated.(string); ok {\n\t\t\tmodel.Capabilities = s\n\t\t}\n\t}\n\n\t// Translate prompts\n\tif model.Prompts != nil {\n\t\tfor i := range model.Prompts {\n\t\t\tif translated := i18n.Translate(assistantID, locale, model.Prompts[i].Name); translated != nil {\n\t\t\t\tif s, ok := translated.(string); ok {\n\t\t\t\t\tmodel.Prompts[i].Name = s\n\t\t\t\t}\n\t\t\t}\n\t\t\tif translated := i18n.Translate(assistantID, locale, model.Prompts[i].Content); translated != nil {\n\t\t\t\tif s, ok := translated.(string); ok {\n\t\t\t\t\tmodel.Prompts[i].Content = s\n\t\t\t\t}\n\t\t\t}\n\t\t}\n\t}\n\n\t// Translate placeholder\n\tif model.Placeholder != nil {\n\t\tif translated := i18n.Translate(assistantID, locale, model.Placeholder.Title); translated != nil {\n\t\t\tif s, ok := translated.(string); ok {\n\t\t\t\tmodel.Placeholder.Title = s\n\t\t\t}\n\t\t}\n\t\tif translated := i18n.Translate(assistantID, locale, model.Placeholder.Description); translated != nil {\n\t\t\tif s, ok := translated.(string); ok {\n\t\t\t\tmodel.Placeholder.Description = s\n\t\t\t}\n\t\t}\n\t\tif translated := i18n.Translate(assistantID, locale, model.Placeholder.Prompts); translated != nil {\n\t\t\tif prompts, ok := translated.([]string); ok {\n\t\t\t\tmodel.Placeholder.Prompts = prompts\n\t\t\t}\n\t\t}\n\t}\n\n\t// Translate tags\n\tif translated := i18n.Translate(assistantID, locale, model.Tags); translated != nil {\n\t\tif tags, ok := translated.([]string); ok {\n\t\t\tmodel.Tags = tags\n\t\t}\n\t}\n}\n" }, { "path": "agent/store/xun/assistant_test.go", "content": "package xun_test\n\nimport (\n\t\"fmt\"\n\t\"os\"\n\t\"strings\"\n\t\"testing\"\n\t\"time\"\n\n\t\"github.com/yaoapp/xun/dbal/query\"\n\t\"github.com/yaoapp/yao/agent/context\"\n\t\"github.com/yaoapp/yao/agent/i18n\"\n\tsearchTypes \"github.com/yaoapp/yao/agent/search/types\"\n\t\"github.com/yaoapp/yao/agent/store/types\"\n\t\"github.com/yaoapp/yao/agent/store/xun\"\n\t\"github.com/yaoapp/yao/config\"\n\t\"github.com/yaoapp/yao/test\"\n)\n\nfunc TestMain(m *testing.M) {\n\t// Setup will be done in each test via test.Prepare\n\ttest.Prepare(nil, config.Conf)\n\tdefer test.Clean()\n\n\t// Run tests and exit with appropriate exit code\n\tcode := m.Run()\n\tos.Exit(code)\n}\n\n// TestSaveAssistant tests creating and updating assistants\nfunc TestSaveAssistant(t *testing.T) {\n\ttest.Prepare(t, config.Conf)\n\tdefer test.Clean()\n\n\t// Create a new xun store\n\tstore, err := xun.NewXun(types.Setting{\n\t\tConnector: \"default\",\n\t})\n\tif err != nil {\n\t\tt.Fatalf(\"Failed to create store: %v\", err)\n\t}\n\n\tt.Run(\"CreateNewAssistant\", func(t *testing.T) {\n\t\tassistant := &types.AssistantModel{\n\t\t\tName: \"Test Assistant\",\n\t\t\tType: \"assistant\",\n\t\t\tConnector: \"openai\",\n\t\t\tDescription: \"A test assistant for unit testing\",\n\t\t\tAvatar: \"https://example.com/avatar.png\",\n\t\t\tTags: []string{\"test\", \"automation\"},\n\t\t\tOptions: map[string]interface{}{\"temperature\": 0.7},\n\t\t\tSort: 100,\n\t\t\tBuiltIn: false,\n\t\t\tReadonly: false,\n\t\t\tPublic: false,\n\t\t\tShare: \"private\",\n\t\t\tMentionable: true,\n\t\t\tAutomated: true,\n\t\t}\n\n\t\tid, err := store.SaveAssistant(assistant)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to save assistant: %v\", err)\n\t\t}\n\n\t\tif id == \"\" {\n\t\t\tt.Error(\"Expected non-empty assistant ID\")\n\t\t}\n\n\t\tif assistant.ID == \"\" {\n\t\t\tt.Error(\"Expected assistant.ID to be set\")\n\t\t}\n\n\t\tt.Logf(\"Created assistant with ID: %s\", id)\n\t})\n\n\tt.Run(\"UpdateExistingAssistant\", func(t *testing.T) {\n\t\t// Create initial assistant\n\t\tassistant := &types.AssistantModel{\n\t\t\tName: \"Update Test Assistant\",\n\t\t\tType: \"assistant\",\n\t\t\tConnector: \"openai\",\n\t\t\tDescription: \"Original description\",\n\t\t\tTags: []string{\"original\"},\n\t\t\tShare: \"private\",\n\t\t}\n\n\t\tid, err := store.SaveAssistant(assistant)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to create assistant: %v\", err)\n\t\t}\n\n\t\t// Update the assistant\n\t\tassistant.Description = \"Updated description\"\n\t\tassistant.Tags = []string{\"updated\", \"modified\"}\n\t\tassistant.Sort = 200\n\n\t\tupdatedID, err := store.SaveAssistant(assistant)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to update assistant: %v\", err)\n\t\t}\n\n\t\tif updatedID != id {\n\t\t\tt.Errorf(\"Expected ID %s, got %s\", id, updatedID)\n\t\t}\n\n\t\t// Verify update - request all fields to see the update\n\t\tretrieved, err := store.GetAssistant(id, types.AssistantFullFields)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to retrieve updated assistant: %v\", err)\n\t\t}\n\n\t\tif retrieved.Description != \"Updated description\" {\n\t\t\tt.Errorf(\"Expected description 'Updated description', got '%s'\", retrieved.Description)\n\t\t}\n\n\t\tif len(retrieved.Tags) != 2 || retrieved.Tags[0] != \"updated\" {\n\t\t\tt.Errorf(\"Expected tags [updated, modified], got %v\", retrieved.Tags)\n\t\t}\n\t})\n\n\tt.Run(\"ValidationErrors\", func(t *testing.T) {\n\t\t// Test nil assistant\n\t\t_, err := store.SaveAssistant(nil)\n\t\tif err == nil {\n\t\t\tt.Error(\"Expected error for nil assistant\")\n\t\t}\n\n\t\t// Test missing name\n\t\tassistant := &types.AssistantModel{\n\t\t\tType: \"assistant\",\n\t\t\tConnector: \"openai\",\n\t\t}\n\t\t_, err = store.SaveAssistant(assistant)\n\t\tif err == nil {\n\t\t\tt.Error(\"Expected error for missing name\")\n\t\t}\n\n\t\t// Test missing type\n\t\tassistant = &types.AssistantModel{\n\t\t\tName: \"Test\",\n\t\t\tConnector: \"openai\",\n\t\t}\n\t\t_, err = store.SaveAssistant(assistant)\n\t\tif err == nil {\n\t\t\tt.Error(\"Expected error for missing type\")\n\t\t}\n\n\t\t// Test missing connector\n\t\tassistant = &types.AssistantModel{\n\t\t\tName: \"Test\",\n\t\t\tType: \"assistant\",\n\t\t}\n\t\t_, err = store.SaveAssistant(assistant)\n\t\tif err == nil {\n\t\t\tt.Error(\"Expected error for missing connector\")\n\t\t}\n\t})\n\n\tt.Run(\"ComplexDataTypes\", func(t *testing.T) {\n\t\tassistant := &types.AssistantModel{\n\t\t\tName: \"Complex Assistant\",\n\t\t\tType: \"assistant\",\n\t\t\tConnector: \"openai\",\n\t\t\tShare: \"private\",\n\t\t\tPrompts: []types.Prompt{\n\t\t\t\t{Role: \"system\", Content: \"You are a helpful assistant\"},\n\t\t\t\t{Role: \"user\", Content: \"Hello\"},\n\t\t\t},\n\t\t\tOptions: map[string]interface{}{\n\t\t\t\t\"temperature\": 0.8,\n\t\t\t\t\"max_tokens\": 2000,\n\t\t\t},\n\t\t\tTags: []string{\"complex\", \"testing\", \"data\"},\n\t\t\tPlaceholder: &types.Placeholder{\n\t\t\t\tTitle: \"Type your message\",\n\t\t\t\tDescription: \"Enter your message here...\",\n\t\t\t\tPrompts: []string{\"What can I help you with?\"},\n\t\t\t},\n\t\t}\n\n\t\tid, err := store.SaveAssistant(assistant)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to save complex assistant: %v\", err)\n\t\t}\n\n\t\t// Retrieve and verify - request all fields for complex data\n\t\tretrieved, err := store.GetAssistant(id, types.AssistantFullFields)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to retrieve complex assistant: %v\", err)\n\t\t}\n\n\t\tif len(retrieved.Prompts) != 2 {\n\t\t\tt.Errorf(\"Expected 2 prompts, got %d\", len(retrieved.Prompts))\n\t\t}\n\n\t\tif retrieved.Placeholder == nil {\n\t\t\tt.Error(\"Expected placeholder to be set\")\n\t\t}\n\n\t\tif len(retrieved.Tags) != 3 {\n\t\t\tt.Errorf(\"Expected 3 tags, got %d\", len(retrieved.Tags))\n\t\t}\n\t})\n\n\tt.Run(\"SaveWithMCPServers\", func(t *testing.T) {\n\t\t// Test creating assistant with MCP servers directly\n\t\t// This will test that:\n\t\t// - server1 (no tools/resources) serializes as \"server1\"\n\t\t// - server2 (with tools) serializes as {\"server_id\":\"server2\",\"tools\":[...]}\n\t\t// - server3 (with both) serializes as {\"server_id\":\"server3\",\"resources\":[...],\"tools\":[...]}\n\t\tassistant := &types.AssistantModel{\n\t\t\tName: \"MCP Save Test\",\n\t\t\tType: \"assistant\",\n\t\t\tConnector: \"openai\",\n\t\t\tShare: \"private\",\n\t\t\tMCP: &types.MCPServers{\n\t\t\t\tServers: []types.MCPServerConfig{\n\t\t\t\t\t{ServerID: \"server1\"},\n\t\t\t\t\t{\n\t\t\t\t\t\tServerID: \"server2\",\n\t\t\t\t\t\tTools: []string{\"tool1\", \"tool2\"},\n\t\t\t\t\t},\n\t\t\t\t\t{\n\t\t\t\t\t\tServerID: \"server3\",\n\t\t\t\t\t\tResources: []string{\"res1\", \"res2\"},\n\t\t\t\t\t\tTools: []string{\"tool3\", \"tool4\"},\n\t\t\t\t\t},\n\t\t\t\t},\n\t\t\t\tOptions: map[string]interface{}{\n\t\t\t\t\t\"timeout\": 30,\n\t\t\t\t},\n\t\t\t},\n\t\t}\n\n\t\tid, err := store.SaveAssistant(assistant)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to save assistant with MCP: %v\", err)\n\t\t}\n\n\t\t// Retrieve and verify MCP configuration - mcp is in default fields\n\t\tretrieved, err := store.GetAssistant(id, []string{})\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to retrieve assistant: %v\", err)\n\t\t}\n\n\t\tif retrieved.MCP == nil {\n\t\t\tt.Fatal(\"Expected MCP to be set\")\n\t\t}\n\n\t\tif len(retrieved.MCP.Servers) != 3 {\n\t\t\tt.Errorf(\"Expected 3 MCP servers, got %d\", len(retrieved.MCP.Servers))\n\t\t}\n\n\t\t// Verify server1 (simple format)\n\t\tif retrieved.MCP.Servers[0].ServerID != \"server1\" {\n\t\t\tt.Errorf(\"Expected server1, got '%s'\", retrieved.MCP.Servers[0].ServerID)\n\t\t}\n\n\t\t// Verify server2 (with tools)\n\t\tif retrieved.MCP.Servers[1].ServerID != \"server2\" {\n\t\t\tt.Errorf(\"Expected server2, got '%s'\", retrieved.MCP.Servers[1].ServerID)\n\t\t}\n\t\tif len(retrieved.MCP.Servers[1].Tools) != 2 {\n\t\t\tt.Errorf(\"Expected 2 tools for server2, got %d\", len(retrieved.MCP.Servers[1].Tools))\n\t\t}\n\n\t\t// Verify server3 (with resources and tools)\n\t\tif retrieved.MCP.Servers[2].ServerID != \"server3\" {\n\t\t\tt.Errorf(\"Expected server3, got '%s'\", retrieved.MCP.Servers[2].ServerID)\n\t\t}\n\t\tif len(retrieved.MCP.Servers[2].Resources) != 2 {\n\t\t\tt.Errorf(\"Expected 2 resources for server3, got %d\", len(retrieved.MCP.Servers[2].Resources))\n\t\t}\n\t\tif len(retrieved.MCP.Servers[2].Tools) != 2 {\n\t\t\tt.Errorf(\"Expected 2 tools for server3, got %d\", len(retrieved.MCP.Servers[2].Tools))\n\t\t}\n\n\t\t// Verify options\n\t\tif retrieved.MCP.Options == nil {\n\t\t\tt.Error(\"Expected MCP options to be set\")\n\t\t}\n\t\tif timeout, ok := retrieved.MCP.Options[\"timeout\"].(float64); !ok || timeout != 30 {\n\t\t\tt.Errorf(\"Expected timeout 30, got %v\", retrieved.MCP.Options[\"timeout\"])\n\t\t}\n\n\t\tt.Logf(\"Successfully verified MCP configuration for assistant %s\", id)\n\t})\n\n\tt.Run(\"UpdateWithMCPServers\", func(t *testing.T) {\n\t\t// Create assistant without MCP\n\t\tassistant := &types.AssistantModel{\n\t\t\tName: \"MCP Update Test\",\n\t\t\tType: \"assistant\",\n\t\t\tConnector: \"openai\",\n\t\t\tShare: \"private\",\n\t\t}\n\n\t\tid, err := store.SaveAssistant(assistant)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to create assistant: %v\", err)\n\t\t}\n\n\t\t// Update assistant with MCP\n\t\tassistant.MCP = &types.MCPServers{\n\t\t\tServers: []types.MCPServerConfig{\n\t\t\t\t{ServerID: \"new-server1\"},\n\t\t\t\t{\n\t\t\t\t\tServerID: \"new-server2\",\n\t\t\t\t\tTools: []string{\"newtool1\"},\n\t\t\t\t},\n\t\t\t},\n\t\t}\n\n\t\t_, err = store.SaveAssistant(assistant)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to update assistant with MCP: %v\", err)\n\t\t}\n\n\t\t// Retrieve and verify - mcp is in default fields\n\t\tretrieved, err := store.GetAssistant(id, []string{})\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to retrieve assistant: %v\", err)\n\t\t}\n\n\t\tif retrieved.MCP == nil || len(retrieved.MCP.Servers) != 2 {\n\t\t\tt.Errorf(\"Expected 2 MCP servers, got %v\", retrieved.MCP)\n\t\t}\n\n\t\tif retrieved.MCP.Servers[0].ServerID != \"new-server1\" {\n\t\t\tt.Errorf(\"Expected new-server1, got '%s'\", retrieved.MCP.Servers[0].ServerID)\n\t\t}\n\n\t\tt.Logf(\"Successfully updated and verified MCP for assistant %s\", id)\n\t})\n\n\tt.Run(\"UsesConfiguration\", func(t *testing.T) {\n\t\t// Test assistant with Uses configuration\n\t\tassistant := &types.AssistantModel{\n\t\t\tName: \"Uses Test Assistant\",\n\t\t\tType: \"assistant\",\n\t\t\tConnector: \"openai\",\n\t\t\tShare: \"private\",\n\t\t\tUses: &context.Uses{\n\t\t\t\tVision: \"mcp:vision-server\",\n\t\t\t\tAudio: \"agent\",\n\t\t\t\tSearch: \"mcp:search-server\",\n\t\t\t\tFetch: \"agent\",\n\t\t\t},\n\t\t}\n\n\t\tid, err := store.SaveAssistant(assistant)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to save assistant with uses: %v\", err)\n\t\t}\n\n\t\t// Retrieve and verify uses configuration - uses is NOT in default fields, need to request all\n\t\tretrieved, err := store.GetAssistant(id, types.AssistantFullFields)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to retrieve assistant: %v\", err)\n\t\t}\n\n\t\tif retrieved.Uses == nil {\n\t\t\tt.Fatal(\"Expected uses to be set\")\n\t\t}\n\n\t\tif retrieved.Uses.Vision != \"mcp:vision-server\" {\n\t\t\tt.Errorf(\"Expected vision 'mcp:vision-server', got '%s'\", retrieved.Uses.Vision)\n\t\t}\n\n\t\tif retrieved.Uses.Audio != \"agent\" {\n\t\t\tt.Errorf(\"Expected audio 'agent', got '%s'\", retrieved.Uses.Audio)\n\t\t}\n\n\t\tif retrieved.Uses.Search != \"mcp:search-server\" {\n\t\t\tt.Errorf(\"Expected search 'mcp:search-server', got '%s'\", retrieved.Uses.Search)\n\t\t}\n\n\t\tif retrieved.Uses.Fetch != \"agent\" {\n\t\t\tt.Errorf(\"Expected fetch 'agent', got '%s'\", retrieved.Uses.Fetch)\n\t\t}\n\n\t\tt.Logf(\"Successfully saved and retrieved assistant with uses configuration\")\n\t})\n\n\tt.Run(\"NilUses\", func(t *testing.T) {\n\t\t// Test assistant without Uses configuration\n\t\tassistant := &types.AssistantModel{\n\t\t\tName: \"No Uses Assistant\",\n\t\t\tType: \"assistant\",\n\t\t\tConnector: \"openai\",\n\t\t\tShare: \"private\",\n\t\t}\n\n\t\tid, err := store.SaveAssistant(assistant)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to save assistant without uses: %v\", err)\n\t\t}\n\n\t\t// Retrieve and verify uses is nil - request all fields to check uses\n\t\tretrieved, err := store.GetAssistant(id, types.AssistantFullFields)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to retrieve assistant: %v\", err)\n\t\t}\n\n\t\tif retrieved.Uses != nil {\n\t\t\tt.Errorf(\"Expected uses to be nil, got %+v\", retrieved.Uses)\n\t\t}\n\t})\n\n\tt.Run(\"PartialUsesConfiguration\", func(t *testing.T) {\n\t\t// Test assistant with partial Uses configuration\n\t\tassistant := &types.AssistantModel{\n\t\t\tName: \"Partial Uses Assistant\",\n\t\t\tType: \"assistant\",\n\t\t\tConnector: \"openai\",\n\t\t\tShare: \"private\",\n\t\t\tUses: &context.Uses{\n\t\t\t\tVision: \"mcp:vision-only\",\n\t\t\t\t// Audio, Search, Fetch not set\n\t\t\t},\n\t\t}\n\n\t\tid, err := store.SaveAssistant(assistant)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to save assistant with partial uses: %v\", err)\n\t\t}\n\n\t\t// Retrieve and verify - request all fields for uses\n\t\tretrieved, err := store.GetAssistant(id, types.AssistantFullFields)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to retrieve assistant: %v\", err)\n\t\t}\n\n\t\tif retrieved.Uses == nil {\n\t\t\tt.Fatal(\"Expected uses to be set\")\n\t\t}\n\n\t\tif retrieved.Uses.Vision != \"mcp:vision-only\" {\n\t\t\tt.Errorf(\"Expected vision 'mcp:vision-only', got '%s'\", retrieved.Uses.Vision)\n\t\t}\n\n\t\tif retrieved.Uses.Audio != \"\" {\n\t\t\tt.Errorf(\"Expected audio to be empty, got '%s'\", retrieved.Uses.Audio)\n\t\t}\n\n\t\tif retrieved.Uses.Search != \"\" {\n\t\t\tt.Errorf(\"Expected search to be empty, got '%s'\", retrieved.Uses.Search)\n\t\t}\n\n\t\tif retrieved.Uses.Fetch != \"\" {\n\t\t\tt.Errorf(\"Expected fetch to be empty, got '%s'\", retrieved.Uses.Fetch)\n\t\t}\n\t})\n\n\tt.Run(\"SearchConfiguration\", func(t *testing.T) {\n\t\t// Test assistant with Search configuration\n\t\tassistant := &types.AssistantModel{\n\t\t\tName: \"Search Config Test Assistant\",\n\t\t\tType: \"assistant\",\n\t\t\tConnector: \"openai\",\n\t\t\tShare: \"private\",\n\t\t\tSearch: &searchTypes.Config{\n\t\t\t\tWeb: &searchTypes.WebConfig{\n\t\t\t\t\tProvider: \"tavily\",\n\t\t\t\t\tMaxResults: 15,\n\t\t\t\t},\n\t\t\t\tKB: &searchTypes.KBConfig{\n\t\t\t\t\tCollections: []string{\"docs\", \"faq\"},\n\t\t\t\t\tThreshold: 0.8,\n\t\t\t\t\tGraph: true,\n\t\t\t\t},\n\t\t\t\tDB: &searchTypes.DBConfig{\n\t\t\t\t\tModels: []string{\"user\", \"product\"},\n\t\t\t\t\tMaxResults: 50,\n\t\t\t\t},\n\t\t\t\tCitation: &searchTypes.CitationConfig{\n\t\t\t\t\tFormat: \"#ref:{id}\",\n\t\t\t\t\tAutoInjectPrompt: true,\n\t\t\t\t},\n\t\t\t\tWeights: &searchTypes.WeightsConfig{\n\t\t\t\t\tUser: 1.0,\n\t\t\t\t\tHook: 0.9,\n\t\t\t\t\tAuto: 0.7,\n\t\t\t\t},\n\t\t\t},\n\t\t}\n\n\t\tid, err := store.SaveAssistant(assistant)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to save assistant with search config: %v\", err)\n\t\t}\n\n\t\t// Retrieve and verify search configuration - search is NOT in default fields\n\t\tretrieved, err := store.GetAssistant(id, types.AssistantFullFields)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to retrieve assistant: %v\", err)\n\t\t}\n\n\t\tif retrieved.Search == nil {\n\t\t\tt.Fatal(\"Expected search to be set\")\n\t\t}\n\n\t\t// Verify Web config\n\t\tif retrieved.Search.Web == nil {\n\t\t\tt.Fatal(\"Expected search.web to be set\")\n\t\t}\n\t\tif retrieved.Search.Web.Provider != \"tavily\" {\n\t\t\tt.Errorf(\"Expected web provider 'tavily', got '%s'\", retrieved.Search.Web.Provider)\n\t\t}\n\t\tif retrieved.Search.Web.MaxResults != 15 {\n\t\t\tt.Errorf(\"Expected web max_results 15, got %d\", retrieved.Search.Web.MaxResults)\n\t\t}\n\n\t\t// Verify KB config\n\t\tif retrieved.Search.KB == nil {\n\t\t\tt.Fatal(\"Expected search.kb to be set\")\n\t\t}\n\t\tif len(retrieved.Search.KB.Collections) != 2 {\n\t\t\tt.Errorf(\"Expected 2 KB collections, got %d\", len(retrieved.Search.KB.Collections))\n\t\t}\n\t\tif retrieved.Search.KB.Collections[0] != \"docs\" {\n\t\t\tt.Errorf(\"Expected first collection 'docs', got '%s'\", retrieved.Search.KB.Collections[0])\n\t\t}\n\t\tif retrieved.Search.KB.Threshold != 0.8 {\n\t\t\tt.Errorf(\"Expected KB threshold 0.8, got %f\", retrieved.Search.KB.Threshold)\n\t\t}\n\t\tif !retrieved.Search.KB.Graph {\n\t\t\tt.Error(\"Expected KB graph to be true\")\n\t\t}\n\n\t\t// Verify DB config\n\t\tif retrieved.Search.DB == nil {\n\t\t\tt.Fatal(\"Expected search.db to be set\")\n\t\t}\n\t\tif len(retrieved.Search.DB.Models) != 2 {\n\t\t\tt.Errorf(\"Expected 2 DB models, got %d\", len(retrieved.Search.DB.Models))\n\t\t}\n\t\tif retrieved.Search.DB.MaxResults != 50 {\n\t\t\tt.Errorf(\"Expected DB max_results 50, got %d\", retrieved.Search.DB.MaxResults)\n\t\t}\n\n\t\t// Verify Citation config\n\t\tif retrieved.Search.Citation == nil {\n\t\t\tt.Fatal(\"Expected search.citation to be set\")\n\t\t}\n\t\tif retrieved.Search.Citation.Format != \"#ref:{id}\" {\n\t\t\tt.Errorf(\"Expected citation format '#ref:{id}', got '%s'\", retrieved.Search.Citation.Format)\n\t\t}\n\t\tif !retrieved.Search.Citation.AutoInjectPrompt {\n\t\t\tt.Error(\"Expected citation auto_inject_prompt to be true\")\n\t\t}\n\n\t\t// Verify Weights config\n\t\tif retrieved.Search.Weights == nil {\n\t\t\tt.Fatal(\"Expected search.weights to be set\")\n\t\t}\n\t\tif retrieved.Search.Weights.User != 1.0 {\n\t\t\tt.Errorf(\"Expected weights.user 1.0, got %f\", retrieved.Search.Weights.User)\n\t\t}\n\t\tif retrieved.Search.Weights.Hook != 0.9 {\n\t\t\tt.Errorf(\"Expected weights.hook 0.9, got %f\", retrieved.Search.Weights.Hook)\n\t\t}\n\t\tif retrieved.Search.Weights.Auto != 0.7 {\n\t\t\tt.Errorf(\"Expected weights.auto 0.7, got %f\", retrieved.Search.Weights.Auto)\n\t\t}\n\n\t\tt.Logf(\"Successfully saved and retrieved assistant with search configuration\")\n\t})\n\n\tt.Run(\"NilSearchConfiguration\", func(t *testing.T) {\n\t\t// Test assistant without Search configuration\n\t\tassistant := &types.AssistantModel{\n\t\t\tName: \"No Search Config Assistant\",\n\t\t\tType: \"assistant\",\n\t\t\tConnector: \"openai\",\n\t\t\tShare: \"private\",\n\t\t}\n\n\t\tid, err := store.SaveAssistant(assistant)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to save assistant without search: %v\", err)\n\t\t}\n\n\t\t// Retrieve and verify search is nil - request all fields to check search\n\t\tretrieved, err := store.GetAssistant(id, types.AssistantFullFields)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to retrieve assistant: %v\", err)\n\t\t}\n\n\t\tif retrieved.Search != nil {\n\t\t\tt.Errorf(\"Expected search to be nil, got %+v\", retrieved.Search)\n\t\t}\n\t})\n\n\tt.Run(\"PartialSearchConfiguration\", func(t *testing.T) {\n\t\t// Test assistant with partial Search configuration\n\t\tassistant := &types.AssistantModel{\n\t\t\tName: \"Partial Search Config Assistant\",\n\t\t\tType: \"assistant\",\n\t\t\tConnector: \"openai\",\n\t\t\tShare: \"private\",\n\t\t\tSearch: &searchTypes.Config{\n\t\t\t\tWeb: &searchTypes.WebConfig{\n\t\t\t\t\tProvider: \"serper\",\n\t\t\t\t},\n\t\t\t\t// KB, DB, Citation, Weights not set\n\t\t\t},\n\t\t}\n\n\t\tid, err := store.SaveAssistant(assistant)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to save assistant with partial search: %v\", err)\n\t\t}\n\n\t\t// Retrieve and verify - request all fields for search\n\t\tretrieved, err := store.GetAssistant(id, types.AssistantFullFields)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to retrieve assistant: %v\", err)\n\t\t}\n\n\t\tif retrieved.Search == nil {\n\t\t\tt.Fatal(\"Expected search to be set\")\n\t\t}\n\n\t\tif retrieved.Search.Web == nil {\n\t\t\tt.Fatal(\"Expected search.web to be set\")\n\t\t}\n\t\tif retrieved.Search.Web.Provider != \"serper\" {\n\t\t\tt.Errorf(\"Expected web provider 'serper', got '%s'\", retrieved.Search.Web.Provider)\n\t\t}\n\n\t\t// Other fields should be nil\n\t\tif retrieved.Search.KB != nil {\n\t\t\tt.Errorf(\"Expected search.kb to be nil, got %+v\", retrieved.Search.KB)\n\t\t}\n\t\tif retrieved.Search.DB != nil {\n\t\t\tt.Errorf(\"Expected search.db to be nil, got %+v\", retrieved.Search.DB)\n\t\t}\n\t\tif retrieved.Search.Citation != nil {\n\t\t\tt.Errorf(\"Expected search.citation to be nil, got %+v\", retrieved.Search.Citation)\n\t\t}\n\t\tif retrieved.Search.Weights != nil {\n\t\t\tt.Errorf(\"Expected search.weights to be nil, got %+v\", retrieved.Search.Weights)\n\t\t}\n\t})\n\n\tt.Run(\"ConnectorOptions\", func(t *testing.T) {\n\t\t// Test assistant with connector options\n\t\toptionalTrue := true\n\t\tassistant := &types.AssistantModel{\n\t\t\tName: \"Connector Options Test\",\n\t\t\tType: \"assistant\",\n\t\t\tConnector: \"openai\",\n\t\t\tShare: \"private\",\n\t\t\tConnectorOptions: &types.ConnectorOptions{\n\t\t\t\tOptional: &optionalTrue,\n\t\t\t\tConnectors: []string{\"openai\", \"anthropic\"},\n\t\t\t\tFilters: []types.ModelCapability{types.CapVision, types.CapToolCalls},\n\t\t\t},\n\t\t}\n\n\t\tid, err := store.SaveAssistant(assistant)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to save assistant with connector options: %v\", err)\n\t\t}\n\n\t\t// Retrieve and verify - connector_options is NOT in default fields\n\t\tretrieved, err := store.GetAssistant(id, types.AssistantFullFields)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to retrieve assistant: %v\", err)\n\t\t}\n\n\t\tif retrieved.ConnectorOptions == nil {\n\t\t\tt.Fatal(\"Expected connector options to be set\")\n\t\t}\n\n\t\tif retrieved.ConnectorOptions.Optional == nil || !*retrieved.ConnectorOptions.Optional {\n\t\t\tt.Error(\"Expected optional to be true\")\n\t\t}\n\n\t\tif len(retrieved.ConnectorOptions.Connectors) != 2 {\n\t\t\tt.Errorf(\"Expected 2 connectors, got %d\", len(retrieved.ConnectorOptions.Connectors))\n\t\t}\n\n\t\tif len(retrieved.ConnectorOptions.Filters) != 2 {\n\t\t\tt.Errorf(\"Expected 2 filters, got %d\", len(retrieved.ConnectorOptions.Filters))\n\t\t}\n\n\t\tif retrieved.ConnectorOptions.Filters[0] != types.CapVision {\n\t\t\tt.Errorf(\"Expected first filter to be vision, got '%s'\", retrieved.ConnectorOptions.Filters[0])\n\t\t}\n\n\t\tt.Logf(\"Successfully saved and retrieved connector options for assistant %s\", id)\n\t})\n\n\tt.Run(\"PromptPresets\", func(t *testing.T) {\n\t\t// Test assistant with prompt presets\n\t\tassistant := &types.AssistantModel{\n\t\t\tName: \"Prompt Presets Test\",\n\t\t\tType: \"assistant\",\n\t\t\tConnector: \"openai\",\n\t\t\tShare: \"private\",\n\t\t\tPromptPresets: map[string][]types.Prompt{\n\t\t\t\t\"chat\": {\n\t\t\t\t\t{Role: \"system\", Content: \"You are a friendly chatbot\"},\n\t\t\t\t\t{Role: \"user\", Content: \"Hello!\"},\n\t\t\t\t},\n\t\t\t\t\"task\": {\n\t\t\t\t\t{Role: \"system\", Content: \"You are a task executor\"},\n\t\t\t\t},\n\t\t\t},\n\t\t}\n\n\t\tid, err := store.SaveAssistant(assistant)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to save assistant with prompt presets: %v\", err)\n\t\t}\n\n\t\t// Retrieve and verify - prompt_presets is NOT in default fields\n\t\tretrieved, err := store.GetAssistant(id, types.AssistantFullFields)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to retrieve assistant: %v\", err)\n\t\t}\n\n\t\tif retrieved.PromptPresets == nil {\n\t\t\tt.Fatal(\"Expected prompt presets to be set\")\n\t\t}\n\n\t\tif len(retrieved.PromptPresets) != 2 {\n\t\t\tt.Errorf(\"Expected 2 preset groups, got %d\", len(retrieved.PromptPresets))\n\t\t}\n\n\t\tchatPrompts, ok := retrieved.PromptPresets[\"chat\"]\n\t\tif !ok {\n\t\t\tt.Fatal(\"Expected 'chat' preset to exist\")\n\t\t}\n\n\t\tif len(chatPrompts) != 2 {\n\t\t\tt.Errorf(\"Expected 2 chat prompts, got %d\", len(chatPrompts))\n\t\t}\n\n\t\tif chatPrompts[0].Role != \"system\" {\n\t\t\tt.Errorf(\"Expected system role, got '%s'\", chatPrompts[0].Role)\n\t\t}\n\n\t\ttaskPrompts, ok := retrieved.PromptPresets[\"task\"]\n\t\tif !ok {\n\t\t\tt.Fatal(\"Expected 'task' preset to exist\")\n\t\t}\n\n\t\tif len(taskPrompts) != 1 {\n\t\t\tt.Errorf(\"Expected 1 task prompt, got %d\", len(taskPrompts))\n\t\t}\n\n\t\tt.Logf(\"Successfully saved and retrieved prompt presets for assistant %s\", id)\n\t})\n\n\tt.Run(\"SourceField\", func(t *testing.T) {\n\t\t// Test assistant with source code\n\t\tsourceCode := `function onMessage(msg) {\n console.log(\"Received:\", msg);\n return { status: \"ok\" };\n}`\n\t\tassistant := &types.AssistantModel{\n\t\t\tName: \"Source Field Test\",\n\t\t\tType: \"assistant\",\n\t\t\tConnector: \"openai\",\n\t\t\tShare: \"private\",\n\t\t\tSource: sourceCode,\n\t\t}\n\n\t\tid, err := store.SaveAssistant(assistant)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to save assistant with source: %v\", err)\n\t\t}\n\n\t\t// Retrieve and verify - source is NOT in default fields\n\t\tretrieved, err := store.GetAssistant(id, types.AssistantFullFields)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to retrieve assistant: %v\", err)\n\t\t}\n\n\t\tif retrieved.Source != sourceCode {\n\t\t\tt.Errorf(\"Expected source code to match, got '%s'\", retrieved.Source)\n\t\t}\n\n\t\tt.Logf(\"Successfully saved and retrieved source code for assistant %s\", id)\n\t})\n\n\tt.Run(\"SandboxConfiguration\", func(t *testing.T) {\n\t\tassistant := &types.AssistantModel{\n\t\t\tName: \"Sandbox Test Assistant\",\n\t\t\tType: \"assistant\",\n\t\t\tConnector: \"openai\",\n\t\t\tShare: \"private\",\n\t\t\tSandbox: &types.Sandbox{\n\t\t\t\tCommand: \"claude\",\n\t\t\t\tTimeout: \"5m\",\n\t\t\t\tArguments: map[string]interface{}{\n\t\t\t\t\t\"max_turns\": 10,\n\t\t\t\t\t\"permission_mode\": \"bypassPermissions\",\n\t\t\t\t},\n\t\t\t},\n\t\t}\n\n\t\tid, err := store.SaveAssistant(assistant)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to save assistant with sandbox: %v\", err)\n\t\t}\n\n\t\tretrieved, err := store.GetAssistant(id, types.AssistantFullFields)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to retrieve assistant: %v\", err)\n\t\t}\n\n\t\tif retrieved.Sandbox == nil {\n\t\t\tt.Fatal(\"Expected sandbox to be set\")\n\t\t}\n\n\t\tif retrieved.Sandbox.Command != \"claude\" {\n\t\t\tt.Errorf(\"Expected command 'claude', got '%s'\", retrieved.Sandbox.Command)\n\t\t}\n\n\t\tif retrieved.Sandbox.Timeout != \"5m\" {\n\t\t\tt.Errorf(\"Expected timeout '5m', got '%s'\", retrieved.Sandbox.Timeout)\n\t\t}\n\n\t\tif retrieved.Sandbox.Arguments == nil {\n\t\t\tt.Fatal(\"Expected sandbox arguments to be set\")\n\t\t}\n\n\t\tif maxTurns, ok := retrieved.Sandbox.Arguments[\"max_turns\"].(float64); !ok || maxTurns != 10 {\n\t\t\tt.Errorf(\"Expected max_turns 10, got %v\", retrieved.Sandbox.Arguments[\"max_turns\"])\n\t\t}\n\n\t\tt.Logf(\"Successfully saved and retrieved sandbox configuration for assistant %s\", id)\n\t})\n\n\tt.Run(\"SandboxWithImage\", func(t *testing.T) {\n\t\tassistant := &types.AssistantModel{\n\t\t\tName: \"Sandbox Image Test\",\n\t\t\tType: \"assistant\",\n\t\t\tConnector: \"openai\",\n\t\t\tShare: \"private\",\n\t\t\tSandbox: &types.Sandbox{\n\t\t\t\tCommand: \"claude\",\n\t\t\t\tImage: \"yaoapp/sandbox-claude-desktop:latest\",\n\t\t\t\tTimeout: \"20m\",\n\t\t\t\tMaxMemory: \"4g\",\n\t\t\t\tMaxCPU: 2.0,\n\t\t\t\tArguments: map[string]interface{}{\n\t\t\t\t\t\"max_turns\": 500,\n\t\t\t\t\t\"permission_mode\": \"bypassPermissions\",\n\t\t\t\t\t\"disallowed_tools\": \"WebSearch\",\n\t\t\t\t},\n\t\t\t\tSecrets: map[string]string{\n\t\t\t\t\t\"GITHUB_TOKEN\": \"$ENV.GITHUB_TOKEN\",\n\t\t\t\t},\n\t\t\t},\n\t\t}\n\n\t\tid, err := store.SaveAssistant(assistant)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to save assistant with sandbox image: %v\", err)\n\t\t}\n\n\t\tretrieved, err := store.GetAssistant(id, types.AssistantFullFields)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to retrieve assistant: %v\", err)\n\t\t}\n\n\t\tif retrieved.Sandbox == nil {\n\t\t\tt.Fatal(\"Expected sandbox to be set\")\n\t\t}\n\n\t\tif retrieved.Sandbox.Image != \"yaoapp/sandbox-claude-desktop:latest\" {\n\t\t\tt.Errorf(\"Expected image 'yaoapp/sandbox-claude-desktop:latest', got '%s'\", retrieved.Sandbox.Image)\n\t\t}\n\n\t\tif retrieved.Sandbox.MaxMemory != \"4g\" {\n\t\t\tt.Errorf(\"Expected max_memory '4g', got '%s'\", retrieved.Sandbox.MaxMemory)\n\t\t}\n\n\t\tif retrieved.Sandbox.MaxCPU != 2.0 {\n\t\t\tt.Errorf(\"Expected max_cpu 2.0, got %f\", retrieved.Sandbox.MaxCPU)\n\t\t}\n\n\t\tif retrieved.Sandbox.Secrets == nil || retrieved.Sandbox.Secrets[\"GITHUB_TOKEN\"] != \"$ENV.GITHUB_TOKEN\" {\n\t\t\tt.Errorf(\"Expected secrets to contain GITHUB_TOKEN, got %v\", retrieved.Sandbox.Secrets)\n\t\t}\n\n\t\tt.Logf(\"Successfully saved and retrieved sandbox with image for assistant %s\", id)\n\t})\n\n\tt.Run(\"NilSandbox\", func(t *testing.T) {\n\t\tassistant := &types.AssistantModel{\n\t\t\tName: \"No Sandbox Assistant\",\n\t\t\tType: \"assistant\",\n\t\t\tConnector: \"openai\",\n\t\t\tShare: \"private\",\n\t\t}\n\n\t\tid, err := store.SaveAssistant(assistant)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to save assistant without sandbox: %v\", err)\n\t\t}\n\n\t\tretrieved, err := store.GetAssistant(id, types.AssistantFullFields)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to retrieve assistant: %v\", err)\n\t\t}\n\n\t\tif retrieved.Sandbox != nil {\n\t\t\tt.Errorf(\"Expected sandbox to be nil, got %+v\", retrieved.Sandbox)\n\t\t}\n\t})\n\n\tt.Run(\"CapabilitiesField\", func(t *testing.T) {\n\t\tassistant := &types.AssistantModel{\n\t\t\tName: \"Capabilities Test\",\n\t\t\tType: \"assistant\",\n\t\t\tConnector: \"openai\",\n\t\t\tShare: \"private\",\n\t\t\tDescription: \"A test assistant\",\n\t\t\tCapabilities: \"Can search the web, analyze data, write code, and summarize documents.\",\n\t\t}\n\n\t\tid, err := store.SaveAssistant(assistant)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to save assistant with capabilities: %v\", err)\n\t\t}\n\n\t\tretrieved, err := store.GetAssistant(id, types.AssistantFullFields)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to retrieve assistant: %v\", err)\n\t\t}\n\n\t\tif retrieved.Capabilities != \"Can search the web, analyze data, write code, and summarize documents.\" {\n\t\t\tt.Errorf(\"Expected capabilities to match, got '%s'\", retrieved.Capabilities)\n\t\t}\n\n\t\tif retrieved.Description != \"A test assistant\" {\n\t\t\tt.Errorf(\"Expected description 'A test assistant', got '%s'\", retrieved.Description)\n\t\t}\n\n\t\tt.Logf(\"Successfully saved and retrieved capabilities for assistant %s\", id)\n\t})\n\n\tt.Run(\"EmptyCapabilities\", func(t *testing.T) {\n\t\tassistant := &types.AssistantModel{\n\t\t\tName: \"No Capabilities Assistant\",\n\t\t\tType: \"assistant\",\n\t\t\tConnector: \"openai\",\n\t\t\tShare: \"private\",\n\t\t}\n\n\t\tid, err := store.SaveAssistant(assistant)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to save assistant without capabilities: %v\", err)\n\t\t}\n\n\t\tretrieved, err := store.GetAssistant(id, types.AssistantFullFields)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to retrieve assistant: %v\", err)\n\t\t}\n\n\t\tif retrieved.Capabilities != \"\" {\n\t\t\tt.Errorf(\"Expected empty capabilities, got '%s'\", retrieved.Capabilities)\n\t\t}\n\t})\n\n\tt.Run(\"CapabilitiesWithI18n\", func(t *testing.T) {\n\t\tassistant := &types.AssistantModel{\n\t\t\tName: \"{{name}}\",\n\t\t\tType: \"assistant\",\n\t\t\tConnector: \"openai\",\n\t\t\tShare: \"private\",\n\t\t\tDescription: \"{{description}}\",\n\t\t\tCapabilities: \"{{capabilities}}\",\n\t\t}\n\n\t\tid, err := store.SaveAssistant(assistant)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to save assistant with i18n capabilities: %v\", err)\n\t\t}\n\n\t\t// Setup i18n\n\t\ti18n.Locales[id] = map[string]i18n.I18n{\n\t\t\t\"en\": {\n\t\t\t\tLocale: \"en\",\n\t\t\t\tMessages: map[string]any{\n\t\t\t\t\t\"name\": \"i18n Test\",\n\t\t\t\t\t\"description\": \"Description in English\",\n\t\t\t\t\t\"capabilities\": \"Can do X, Y, and Z\",\n\t\t\t\t},\n\t\t\t},\n\t\t\t\"zh-cn\": {\n\t\t\t\tLocale: \"zh-cn\",\n\t\t\t\tMessages: map[string]any{\n\t\t\t\t\t\"name\": \"国际化测试\",\n\t\t\t\t\t\"description\": \"中文描述\",\n\t\t\t\t\t\"capabilities\": \"可以做X、Y和Z\",\n\t\t\t\t},\n\t\t\t},\n\t\t}\n\n\t\tretrievedEN, err := store.GetAssistant(id, types.AssistantFullFields, \"en\")\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to get assistant with EN locale: %v\", err)\n\t\t}\n\n\t\tif retrievedEN.Capabilities != \"Can do X, Y, and Z\" {\n\t\t\tt.Errorf(\"Expected capabilities 'Can do X, Y, and Z', got '%s'\", retrievedEN.Capabilities)\n\t\t}\n\n\t\tretrievedZH, err := store.GetAssistant(id, types.AssistantFullFields, \"zh-cn\")\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to get assistant with ZH locale: %v\", err)\n\t\t}\n\n\t\tif retrievedZH.Capabilities != \"可以做X、Y和Z\" {\n\t\t\tt.Errorf(\"Expected capabilities '可以做X、Y和Z', got '%s'\", retrievedZH.Capabilities)\n\t\t}\n\n\t\t// Cleanup\n\t\tdelete(i18n.Locales, id)\n\t\tt.Logf(\"Successfully tested capabilities i18n for assistant %s\", id)\n\t})\n\n\tt.Run(\"CapabilitiesInKeywordSearch\", func(t *testing.T) {\n\t\tuniqueCapability := fmt.Sprintf(\"unique-cap-%d\", time.Now().UnixNano())\n\t\tassistant := &types.AssistantModel{\n\t\t\tName: \"Capabilities Search Test\",\n\t\t\tType: \"assistant\",\n\t\t\tConnector: \"openai\",\n\t\t\tShare: \"private\",\n\t\t\tCapabilities: uniqueCapability,\n\t\t}\n\n\t\t_, err := store.SaveAssistant(assistant)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to save assistant: %v\", err)\n\t\t}\n\n\t\tresponse, err := store.GetAssistants(types.AssistantFilter{\n\t\t\tKeywords: uniqueCapability,\n\t\t\tPage: 1,\n\t\t\tPageSize: 20,\n\t\t})\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to search by capabilities keyword: %v\", err)\n\t\t}\n\n\t\tif len(response.Data) < 1 {\n\t\t\tt.Error(\"Expected to find assistant by capabilities keyword search\")\n\t\t}\n\n\t\tfound := false\n\t\tfor _, a := range response.Data {\n\t\t\tif a.Capabilities == uniqueCapability {\n\t\t\t\tfound = true\n\t\t\t\tbreak\n\t\t\t}\n\t\t}\n\n\t\tif !found {\n\t\t\tt.Error(\"Expected to find assistant with matching capabilities\")\n\t\t}\n\t})\n\n\tt.Run(\"UpdateSandbox\", func(t *testing.T) {\n\t\tassistant := &types.AssistantModel{\n\t\t\tName: \"Update Sandbox Test\",\n\t\t\tType: \"assistant\",\n\t\t\tConnector: \"openai\",\n\t\t\tShare: \"private\",\n\t\t}\n\n\t\tid, err := store.SaveAssistant(assistant)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to create assistant: %v\", err)\n\t\t}\n\n\t\t// Update with sandbox\n\t\tupdates := map[string]interface{}{\n\t\t\t\"sandbox\": &types.Sandbox{\n\t\t\t\tCommand: \"claude\",\n\t\t\t\tTimeout: \"10m\",\n\t\t\t},\n\t\t}\n\n\t\terr = store.UpdateAssistant(id, updates)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to update sandbox: %v\", err)\n\t\t}\n\n\t\tretrieved, err := store.GetAssistant(id, types.AssistantFullFields)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to retrieve assistant: %v\", err)\n\t\t}\n\n\t\tif retrieved.Sandbox == nil {\n\t\t\tt.Fatal(\"Expected sandbox to be set\")\n\t\t}\n\n\t\tif retrieved.Sandbox.Command != \"claude\" {\n\t\t\tt.Errorf(\"Expected command 'claude', got '%s'\", retrieved.Sandbox.Command)\n\t\t}\n\n\t\t// Update to remove sandbox\n\t\tupdates2 := map[string]interface{}{\n\t\t\t\"sandbox\": nil,\n\t\t}\n\n\t\terr = store.UpdateAssistant(id, updates2)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to remove sandbox: %v\", err)\n\t\t}\n\n\t\tretrieved2, err := store.GetAssistant(id, types.AssistantFullFields)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to retrieve assistant: %v\", err)\n\t\t}\n\n\t\tif retrieved2.Sandbox != nil {\n\t\t\tt.Errorf(\"Expected sandbox to be nil, got %+v\", retrieved2.Sandbox)\n\t\t}\n\t})\n\n\tt.Run(\"UpdateCapabilities\", func(t *testing.T) {\n\t\tassistant := &types.AssistantModel{\n\t\t\tName: \"Update Capabilities Test\",\n\t\t\tType: \"assistant\",\n\t\t\tConnector: \"openai\",\n\t\t\tShare: \"private\",\n\t\t\tCapabilities: \"Original capabilities\",\n\t\t}\n\n\t\tid, err := store.SaveAssistant(assistant)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to create assistant: %v\", err)\n\t\t}\n\n\t\t// Update capabilities\n\t\tupdates := map[string]interface{}{\n\t\t\t\"capabilities\": \"Updated capabilities: can search, analyze, and write code\",\n\t\t}\n\n\t\terr = store.UpdateAssistant(id, updates)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to update capabilities: %v\", err)\n\t\t}\n\n\t\tretrieved, err := store.GetAssistant(id, types.AssistantFullFields)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to retrieve assistant: %v\", err)\n\t\t}\n\n\t\tif retrieved.Capabilities != \"Updated capabilities: can search, analyze, and write code\" {\n\t\t\tt.Errorf(\"Expected updated capabilities, got '%s'\", retrieved.Capabilities)\n\t\t}\n\n\t\t// Update to clear capabilities\n\t\tupdates2 := map[string]interface{}{\n\t\t\t\"capabilities\": \"\",\n\t\t}\n\n\t\terr = store.UpdateAssistant(id, updates2)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to clear capabilities: %v\", err)\n\t\t}\n\n\t\tretrieved2, err := store.GetAssistant(id, types.AssistantFullFields)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to retrieve assistant: %v\", err)\n\t\t}\n\n\t\tif retrieved2.Capabilities != \"\" {\n\t\t\tt.Errorf(\"Expected empty capabilities, got '%s'\", retrieved2.Capabilities)\n\t\t}\n\t})\n\n\tt.Run(\"FilterBySandbox\", func(t *testing.T) {\n\t\t// Create one assistant with sandbox\n\t\twithSandbox := &types.AssistantModel{\n\t\t\tName: \"Filter Sandbox Yes\",\n\t\t\tType: \"assistant\",\n\t\t\tConnector: \"openai\",\n\t\t\tShare: \"private\",\n\t\t\tSandbox: &types.Sandbox{\n\t\t\t\tCommand: \"claude\",\n\t\t\t\tTimeout: \"5m\",\n\t\t\t},\n\t\t}\n\t\tidWith, err := store.SaveAssistant(withSandbox)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to save assistant with sandbox: %v\", err)\n\t\t}\n\n\t\t// Create one assistant without sandbox\n\t\twithoutSandbox := &types.AssistantModel{\n\t\t\tName: \"Filter Sandbox No\",\n\t\t\tType: \"assistant\",\n\t\t\tConnector: \"openai\",\n\t\t\tShare: \"private\",\n\t\t}\n\t\tidWithout, err := store.SaveAssistant(withoutSandbox)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to save assistant without sandbox: %v\", err)\n\t\t}\n\n\t\ttestIDs := []string{idWith, idWithout}\n\n\t\t// Filter: sandbox=true, scoped to test IDs\n\t\ttrueVal := true\n\t\tresult, err := store.GetAssistants(types.AssistantFilter{\n\t\t\tPage: 1,\n\t\t\tPageSize: 100,\n\t\t\tSandbox: &trueVal,\n\t\t\tAssistantIDs: testIDs,\n\t\t})\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to filter with sandbox=true: %v\", err)\n\t\t}\n\t\tif len(result.Data) != 1 {\n\t\t\tt.Errorf(\"Expected 1 result for sandbox=true, got %d\", len(result.Data))\n\t\t} else if result.Data[0].ID != idWith {\n\t\t\tt.Errorf(\"Expected assistant %s, got %s\", idWith, result.Data[0].ID)\n\t\t}\n\n\t\t// Filter: sandbox=false, scoped to test IDs\n\t\tfalseVal := false\n\t\tresult2, err := store.GetAssistants(types.AssistantFilter{\n\t\t\tPage: 1,\n\t\t\tPageSize: 100,\n\t\t\tSandbox: &falseVal,\n\t\t\tAssistantIDs: testIDs,\n\t\t})\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to filter with sandbox=false: %v\", err)\n\t\t}\n\t\tif len(result2.Data) != 1 {\n\t\t\tt.Errorf(\"Expected 1 result for sandbox=false, got %d\", len(result2.Data))\n\t\t} else if result2.Data[0].ID != idWithout {\n\t\t\tt.Errorf(\"Expected assistant %s, got %s\", idWithout, result2.Data[0].ID)\n\t\t}\n\n\t\tt.Logf(\"Sandbox filter test passed: sandbox=true returned %d, sandbox=false returned %d\", len(result.Data), len(result2.Data))\n\t})\n\n\tt.Run(\"AllNewFieldsTogether\", func(t *testing.T) {\n\t\t// Test assistant with all new fields together\n\t\toptionalFalse := false\n\t\tassistant := &types.AssistantModel{\n\t\t\tName: \"All New Fields Test\",\n\t\t\tType: \"assistant\",\n\t\t\tConnector: \"openai\",\n\t\t\tShare: \"private\",\n\t\t\tConnectorOptions: &types.ConnectorOptions{\n\t\t\t\tOptional: &optionalFalse,\n\t\t\t\tConnectors: []string{\"openai\"},\n\t\t\t\tFilters: []types.ModelCapability{types.CapVision},\n\t\t\t},\n\t\t\tPromptPresets: map[string][]types.Prompt{\n\t\t\t\t\"default\": {\n\t\t\t\t\t{Role: \"system\", Content: \"Default system prompt\"},\n\t\t\t\t},\n\t\t\t},\n\t\t\tSource: \"// Hook code here\",\n\t\t}\n\n\t\tid, err := store.SaveAssistant(assistant)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to save assistant with all new fields: %v\", err)\n\t\t}\n\n\t\t// Retrieve and verify all new fields\n\t\tretrieved, err := store.GetAssistant(id, types.AssistantFullFields)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to retrieve assistant: %v\", err)\n\t\t}\n\n\t\tif retrieved.ConnectorOptions == nil {\n\t\t\tt.Error(\"Expected connector options to be set\")\n\t\t}\n\n\t\tif retrieved.PromptPresets == nil {\n\t\t\tt.Error(\"Expected prompt presets to be set\")\n\t\t}\n\n\t\tif retrieved.Source == \"\" {\n\t\t\tt.Error(\"Expected source to be set\")\n\t\t}\n\n\t\tt.Logf(\"Successfully saved and retrieved all new fields for assistant %s\", id)\n\t})\n}\n\n// TestDeleteAssistant tests deleting a single assistant\nfunc TestDeleteAssistant(t *testing.T) {\n\ttest.Prepare(t, config.Conf)\n\tdefer test.Clean()\n\n\tstore, err := xun.NewXun(types.Setting{\n\t\tConnector: \"default\",\n\t})\n\tif err != nil {\n\t\tt.Fatalf(\"Failed to create store: %v\", err)\n\t}\n\n\tt.Run(\"DeleteExistingAssistant\", func(t *testing.T) {\n\t\t// Create assistant\n\t\tassistant := &types.AssistantModel{\n\t\t\tName: \"Delete Test\",\n\t\t\tType: \"assistant\",\n\t\t\tConnector: \"openai\",\n\t\t\tShare: \"private\",\n\t\t}\n\n\t\tid, err := store.SaveAssistant(assistant)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to create assistant: %v\", err)\n\t\t}\n\n\t\t// Delete it\n\t\terr = store.DeleteAssistant(id)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to delete assistant: %v\", err)\n\t\t}\n\n\t\t// Verify deletion\n\t\t_, err = store.GetAssistant(id, nil)\n\t\tif err == nil {\n\t\t\tt.Error(\"Expected error when getting deleted assistant\")\n\t\t}\n\t})\n\n\tt.Run(\"DeleteNonExistentAssistant\", func(t *testing.T) {\n\t\terr := store.DeleteAssistant(\"nonexistent-id\")\n\t\tif err == nil {\n\t\t\tt.Error(\"Expected error when deleting non-existent assistant\")\n\t\t}\n\t})\n}\n\n// TestGetAssistant tests retrieving a single assistant\nfunc TestGetAssistant(t *testing.T) {\n\ttest.Prepare(t, config.Conf)\n\tdefer test.Clean()\n\n\tstore, err := xun.NewXun(types.Setting{\n\t\tConnector: \"default\",\n\t})\n\tif err != nil {\n\t\tt.Fatalf(\"Failed to create store: %v\", err)\n\t}\n\n\tt.Run(\"GetExistingAssistant\", func(t *testing.T) {\n\t\t// Create assistant\n\t\tassistant := &types.AssistantModel{\n\t\t\tName: \"Get Test\",\n\t\t\tType: \"assistant\",\n\t\t\tConnector: \"openai\",\n\t\t\tDescription: \"Test description\",\n\t\t\tAvatar: \"https://example.com/avatar.png\",\n\t\t\tTags: []string{\"tag1\", \"tag2\"},\n\t\t\tSort: 150,\n\t\t\tBuiltIn: false,\n\t\t\tShare: \"private\",\n\t\t\tMentionable: true,\n\t\t}\n\n\t\tid, err := store.SaveAssistant(assistant)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to create assistant: %v\", err)\n\t\t}\n\n\t\t// Retrieve it with default fields (tags are now in default fields)\n\t\tretrieved, err := store.GetAssistant(id, nil)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to get assistant: %v\", err)\n\t\t}\n\n\t\tif retrieved.ID != id {\n\t\t\tt.Errorf(\"Expected ID %s, got %s\", id, retrieved.ID)\n\t\t}\n\n\t\tif retrieved.Name != \"Get Test\" {\n\t\t\tt.Errorf(\"Expected name 'Get Test', got '%s'\", retrieved.Name)\n\t\t}\n\n\t\tif retrieved.Description != \"Test description\" {\n\t\t\tt.Errorf(\"Expected description 'Test description', got '%s'\", retrieved.Description)\n\t\t}\n\n\t\tif len(retrieved.Tags) != 2 {\n\t\t\tt.Errorf(\"Expected 2 tags, got %d\", len(retrieved.Tags))\n\t\t}\n\n\t\tif retrieved.Sort != 150 {\n\t\t\tt.Errorf(\"Expected sort 150, got %d\", retrieved.Sort)\n\t\t}\n\t})\n\n\tt.Run(\"GetNonExistentAssistant\", func(t *testing.T) {\n\t\t_, err := store.GetAssistant(\"nonexistent-id\", nil)\n\t\tif err == nil {\n\t\t\tt.Error(\"Expected error when getting non-existent assistant\")\n\t\t}\n\t})\n}\n\n// TestGetAssistants tests retrieving multiple assistants with filtering and pagination\nfunc TestGetAssistants(t *testing.T) {\n\ttest.Prepare(t, config.Conf)\n\tdefer test.Clean()\n\n\tstore, err := xun.NewXun(types.Setting{\n\t\tConnector: \"default\",\n\t})\n\tif err != nil {\n\t\tt.Fatalf(\"Failed to create store: %v\", err)\n\t}\n\n\t// Clean up existing data before creating test assistants\n\tdeleted, err := store.DeleteAssistants(types.AssistantFilter{})\n\tif err != nil {\n\t\tt.Logf(\"Warning: Failed to clean up existing assistants: %v\", err)\n\t} else if deleted > 0 {\n\t\tt.Logf(\"Cleaned up %d existing assistants\", deleted)\n\t}\n\n\t// Create test assistants\n\tassistants := []types.AssistantModel{\n\t\t{\n\t\t\tName: \"Assistant 1\",\n\t\t\tType: \"assistant\",\n\t\t\tConnector: \"openai\",\n\t\t\tDescription: \"First test assistant\",\n\t\t\tTags: []string{\"test\", \"automation\"},\n\t\t\tSort: 100,\n\t\t\tShare: \"private\",\n\t\t\tMentionable: true,\n\t\t\tAutomated: true,\n\t\t},\n\t\t{\n\t\t\tName: \"Assistant 2\",\n\t\t\tType: \"assistant\",\n\t\t\tConnector: \"anthropic\",\n\t\t\tDescription: \"Second test assistant\",\n\t\t\tTags: []string{\"test\", \"manual\"},\n\t\t\tSort: 200,\n\t\t\tShare: \"private\",\n\t\t\tMentionable: false,\n\t\t\tAutomated: false,\n\t\t},\n\t\t{\n\t\t\tName: \"Assistant 3\",\n\t\t\tType: \"bot\",\n\t\t\tConnector: \"openai\",\n\t\t\tDescription: \"Third test bot\",\n\t\t\tTags: []string{\"bot\", \"automation\"},\n\t\t\tSort: 50,\n\t\t\tShare: \"private\",\n\t\t\tMentionable: true,\n\t\t\tAutomated: true,\n\t\t},\n\t}\n\n\tcreatedIDs := []string{}\n\tfor _, asst := range assistants {\n\t\tid, err := store.SaveAssistant(&asst)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to create assistant: %v\", err)\n\t\t}\n\t\tcreatedIDs = append(createdIDs, id)\n\t}\n\n\tt.Run(\"GetAllAssistants\", func(t *testing.T) {\n\t\tresponse, err := store.GetAssistants(types.AssistantFilter{\n\t\t\tPage: 1,\n\t\t\tPageSize: 20,\n\t\t})\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to get assistants: %v\", err)\n\t\t}\n\n\t\tif len(response.Data) < 3 {\n\t\t\tt.Errorf(\"Expected at least 3 assistants, got %d\", len(response.Data))\n\t\t}\n\n\t\tif response.Total < 3 {\n\t\t\tt.Errorf(\"Expected total >= 3, got %d\", response.Total)\n\t\t}\n\t})\n\n\tt.Run(\"FilterByType\", func(t *testing.T) {\n\t\tresponse, err := store.GetAssistants(types.AssistantFilter{\n\t\t\tType: \"assistant\",\n\t\t\tPage: 1,\n\t\t\tPageSize: 20,\n\t\t})\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to get assistants by type: %v\", err)\n\t\t}\n\n\t\tfor _, assistant := range response.Data {\n\t\t\tif assistant.Type != \"assistant\" {\n\t\t\t\tt.Errorf(\"Expected type 'assistant', got '%s'\", assistant.Type)\n\t\t\t}\n\t\t}\n\t})\n\n\tt.Run(\"FilterByConnector\", func(t *testing.T) {\n\t\tresponse, err := store.GetAssistants(types.AssistantFilter{\n\t\t\tConnector: \"openai\",\n\t\t\tPage: 1,\n\t\t\tPageSize: 20,\n\t\t})\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to get assistants by connector: %v\", err)\n\t\t}\n\n\t\tfor _, assistant := range response.Data {\n\t\t\tif assistant.Connector != \"openai\" {\n\t\t\t\tt.Errorf(\"Expected connector 'openai', got '%s'\", assistant.Connector)\n\t\t\t}\n\t\t}\n\t})\n\n\tt.Run(\"FilterByTags\", func(t *testing.T) {\n\t\tresponse, err := store.GetAssistants(types.AssistantFilter{\n\t\t\tTags: []string{\"automation\"},\n\t\t\tPage: 1,\n\t\t\tPageSize: 20,\n\t\t})\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to get assistants by tags: %v\", err)\n\t\t}\n\n\t\t// Should find assistants with \"automation\" tag\n\t\tfound := false\n\t\tfor _, assistant := range response.Data {\n\t\t\tfor _, tag := range assistant.Tags {\n\t\t\t\tif tag == \"automation\" {\n\t\t\t\t\tfound = true\n\t\t\t\t\tbreak\n\t\t\t\t}\n\t\t\t}\n\t\t\tif found {\n\t\t\t\tbreak\n\t\t\t}\n\t\t}\n\n\t\tif !found && len(response.Data) > 0 {\n\t\t\tt.Error(\"Expected to find assistants with 'automation' tag\")\n\t\t}\n\t})\n\n\tt.Run(\"FilterByKeywords\", func(t *testing.T) {\n\t\tresponse, err := store.GetAssistants(types.AssistantFilter{\n\t\t\tKeywords: \"Second\",\n\t\t\tPage: 1,\n\t\t\tPageSize: 20,\n\t\t})\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to get assistants by keywords: %v\", err)\n\t\t}\n\n\t\t// Should find \"Assistant 2\"\n\t\tfound := false\n\t\tfor _, assistant := range response.Data {\n\t\t\tif assistant.Name == \"Assistant 2\" {\n\t\t\t\tfound = true\n\t\t\t\tbreak\n\t\t\t}\n\t\t}\n\n\t\tif !found {\n\t\t\tt.Error(\"Expected to find assistant with keyword 'Second'\")\n\t\t}\n\t})\n\n\tt.Run(\"FilterByMentionable\", func(t *testing.T) {\n\t\tmentionableTrue := true\n\t\tresponse, err := store.GetAssistants(types.AssistantFilter{\n\t\t\tMentionable: &mentionableTrue,\n\t\t\tPage: 1,\n\t\t\tPageSize: 20,\n\t\t})\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to get mentionable assistants: %v\", err)\n\t\t}\n\n\t\tif len(response.Data) != 2 {\n\t\t\tt.Errorf(\"Expected 2 mentionable assistants, got %d\", len(response.Data))\n\t\t}\n\n\t\tfor _, assistant := range response.Data {\n\t\t\tif !assistant.Mentionable {\n\t\t\t\tt.Errorf(\"Expected assistant %s (%s) to be mentionable, but it's not\", assistant.ID, assistant.Name)\n\t\t\t}\n\t\t}\n\t})\n\n\tt.Run(\"FilterByAutomated\", func(t *testing.T) {\n\t\tautomatedFalse := false\n\t\tresponse, err := store.GetAssistants(types.AssistantFilter{\n\t\t\tAutomated: &automatedFalse,\n\t\t\tPage: 1,\n\t\t\tPageSize: 20,\n\t\t})\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to get non-automated assistants: %v\", err)\n\t\t}\n\n\t\tfor _, assistant := range response.Data {\n\t\t\tif assistant.Automated {\n\t\t\t\tt.Error(\"Expected all assistants to be non-automated\")\n\t\t\t}\n\t\t}\n\t})\n\n\tt.Run(\"Pagination\", func(t *testing.T) {\n\t\t// Test first page\n\t\tresponse1, err := store.GetAssistants(types.AssistantFilter{\n\t\t\tPage: 1,\n\t\t\tPageSize: 2,\n\t\t})\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to get first page: %v\", err)\n\t\t}\n\n\t\tif len(response1.Data) > 2 {\n\t\t\tt.Errorf(\"Expected max 2 results, got %d\", len(response1.Data))\n\t\t}\n\n\t\tif response1.Page != 1 {\n\t\t\tt.Errorf(\"Expected page 1, got %d\", response1.Page)\n\t\t}\n\n\t\tif response1.PageSize != 2 {\n\t\t\tt.Errorf(\"Expected page size 2, got %d\", response1.PageSize)\n\t\t}\n\n\t\t// Test second page if there are enough records\n\t\tif response1.Total > 2 {\n\t\t\tresponse2, err := store.GetAssistants(types.AssistantFilter{\n\t\t\t\tPage: 2,\n\t\t\t\tPageSize: 2,\n\t\t\t})\n\t\t\tif err != nil {\n\t\t\t\tt.Fatalf(\"Failed to get second page: %v\", err)\n\t\t\t}\n\n\t\t\tif response2.Page != 2 {\n\t\t\t\tt.Errorf(\"Expected page 2, got %d\", response2.Page)\n\t\t\t}\n\t\t}\n\t})\n\n\tt.Run(\"FieldSelection\", func(t *testing.T) {\n\t\tresponse, err := store.GetAssistants(types.AssistantFilter{\n\t\t\tSelect: []string{\"assistant_id\", \"name\", \"type\"},\n\t\t\tPage: 1,\n\t\t\tPageSize: 20,\n\t\t})\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to get assistants with field selection: %v\", err)\n\t\t}\n\n\t\tif len(response.Data) > 0 {\n\t\t\tassistant := response.Data[0]\n\t\t\tif assistant.ID == \"\" {\n\t\t\t\tt.Error(\"Expected assistant_id field\")\n\t\t\t}\n\t\t\tif assistant.Name == \"\" {\n\t\t\t\tt.Error(\"Expected name field\")\n\t\t\t}\n\t\t\tif assistant.Type == \"\" {\n\t\t\t\tt.Error(\"Expected type field\")\n\t\t\t}\n\t\t}\n\t})\n\n\tt.Run(\"FilterByAssistantID\", func(t *testing.T) {\n\t\tif len(createdIDs) > 0 {\n\t\t\tresponse, err := store.GetAssistants(types.AssistantFilter{\n\t\t\t\tAssistantID: createdIDs[0],\n\t\t\t\tPage: 1,\n\t\t\t\tPageSize: 20,\n\t\t\t})\n\t\t\tif err != nil {\n\t\t\t\tt.Fatalf(\"Failed to get assistant by ID: %v\", err)\n\t\t\t}\n\n\t\t\tif len(response.Data) != 1 {\n\t\t\t\tt.Errorf(\"Expected 1 result, got %d\", len(response.Data))\n\t\t\t}\n\n\t\t\tif response.Data[0].ID != createdIDs[0] {\n\t\t\t\tt.Errorf(\"Expected assistant_id %s, got %s\", createdIDs[0], response.Data[0].ID)\n\t\t\t}\n\t\t}\n\t})\n\n\tt.Run(\"FilterByAssistantIDs\", func(t *testing.T) {\n\t\tif len(createdIDs) >= 2 {\n\t\t\tfilterIDs := []string{createdIDs[0], createdIDs[1]}\n\t\t\tresponse, err := store.GetAssistants(types.AssistantFilter{\n\t\t\t\tAssistantIDs: filterIDs,\n\t\t\t\tPage: 1,\n\t\t\t\tPageSize: 20,\n\t\t\t})\n\t\t\tif err != nil {\n\t\t\t\tt.Fatalf(\"Failed to get assistants by IDs: %v\", err)\n\t\t\t}\n\n\t\t\tif len(response.Data) < 2 {\n\t\t\t\tt.Errorf(\"Expected at least 2 results, got %d\", len(response.Data))\n\t\t\t}\n\t\t}\n\t})\n}\n\n// TestDeleteAssistants tests bulk deletion with filters\nfunc TestDeleteAssistants(t *testing.T) {\n\ttest.Prepare(t, config.Conf)\n\tdefer test.Clean()\n\n\tstore, err := xun.NewXun(types.Setting{\n\t\tConnector: \"default\",\n\t})\n\tif err != nil {\n\t\tt.Fatalf(\"Failed to create store: %v\", err)\n\t}\n\n\tt.Run(\"DeleteByTag\", func(t *testing.T) {\n\t\t// Create assistants with specific tag\n\t\ttag := fmt.Sprintf(\"delete-test-%d\", time.Now().UnixNano())\n\t\tfor i := 0; i < 3; i++ {\n\t\t\tassistant := &types.AssistantModel{\n\t\t\t\tName: fmt.Sprintf(\"Delete Test %d\", i),\n\t\t\t\tType: \"assistant\",\n\t\t\t\tConnector: \"openai\",\n\t\t\t\tTags: []string{tag},\n\t\t\t\tShare: \"private\",\n\t\t\t}\n\t\t\t_, err := store.SaveAssistant(assistant)\n\t\t\tif err != nil {\n\t\t\t\tt.Fatalf(\"Failed to create assistant: %v\", err)\n\t\t\t}\n\t\t}\n\n\t\t// Delete by tag\n\t\tcount, err := store.DeleteAssistants(types.AssistantFilter{\n\t\t\tTags: []string{tag},\n\t\t})\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to delete assistants: %v\", err)\n\t\t}\n\n\t\tif count < 3 {\n\t\t\tt.Errorf(\"Expected at least 3 deletions, got %d\", count)\n\t\t}\n\t})\n\n\tt.Run(\"DeleteByConnector\", func(t *testing.T) {\n\t\t// Create assistants with specific connector\n\t\tconnector := fmt.Sprintf(\"test-connector-%d\", time.Now().UnixNano())\n\t\tfor i := 0; i < 2; i++ {\n\t\t\tassistant := &types.AssistantModel{\n\t\t\t\tName: fmt.Sprintf(\"Connector Test %d\", i),\n\t\t\t\tType: \"assistant\",\n\t\t\t\tConnector: connector,\n\t\t\t\tShare: \"private\",\n\t\t\t}\n\t\t\t_, err := store.SaveAssistant(assistant)\n\t\t\tif err != nil {\n\t\t\t\tt.Fatalf(\"Failed to create assistant: %v\", err)\n\t\t\t}\n\t\t}\n\n\t\t// Delete by connector\n\t\tcount, err := store.DeleteAssistants(types.AssistantFilter{\n\t\t\tConnector: connector,\n\t\t})\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to delete assistants: %v\", err)\n\t\t}\n\n\t\tif count < 2 {\n\t\t\tt.Errorf(\"Expected at least 2 deletions, got %d\", count)\n\t\t}\n\t})\n\n\tt.Run(\"DeleteByKeywords\", func(t *testing.T) {\n\t\t// Create assistants with specific keyword\n\t\tkeyword := fmt.Sprintf(\"unique-keyword-%d\", time.Now().UnixNano())\n\t\tassistant := &types.AssistantModel{\n\t\t\tName: fmt.Sprintf(\"Assistant with %s\", keyword),\n\t\t\tType: \"assistant\",\n\t\t\tConnector: \"openai\",\n\t\t\tDescription: \"Test description\",\n\t\t\tShare: \"private\",\n\t\t}\n\t\t_, err := store.SaveAssistant(assistant)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to create assistant: %v\", err)\n\t\t}\n\n\t\t// Delete by keyword\n\t\tcount, err := store.DeleteAssistants(types.AssistantFilter{\n\t\t\tKeywords: keyword,\n\t\t})\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to delete assistants: %v\", err)\n\t\t}\n\n\t\tif count < 1 {\n\t\t\tt.Errorf(\"Expected at least 1 deletion, got %d\", count)\n\t\t}\n\t})\n\n\tt.Run(\"DeleteByAssistantID\", func(t *testing.T) {\n\t\t// Create an assistant\n\t\tassistant := &types.AssistantModel{\n\t\t\tName: \"Single Delete Test\",\n\t\t\tType: \"assistant\",\n\t\t\tConnector: \"openai\",\n\t\t\tShare: \"private\",\n\t\t}\n\t\tid, err := store.SaveAssistant(assistant)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to create assistant: %v\", err)\n\t\t}\n\n\t\t// Delete by ID\n\t\tcount, err := store.DeleteAssistants(types.AssistantFilter{\n\t\t\tAssistantID: id,\n\t\t})\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to delete assistant: %v\", err)\n\t\t}\n\n\t\tif count != 1 {\n\t\t\tt.Errorf(\"Expected 1 deletion, got %d\", count)\n\t\t}\n\t})\n}\n\n// TestGetAssistantTags tests retrieving unique tags\nfunc TestGetAssistantTags(t *testing.T) {\n\ttest.Prepare(t, config.Conf)\n\tdefer test.Clean()\n\n\tstore, err := xun.NewXun(types.Setting{\n\t\tConnector: \"default\",\n\t})\n\tif err != nil {\n\t\tt.Fatalf(\"Failed to create store: %v\", err)\n\t}\n\n\tt.Run(\"GetUniqueTags\", func(t *testing.T) {\n\t\t// Create assistants with various tags\n\t\tuniqueTag := fmt.Sprintf(\"tag-test-%d\", time.Now().UnixNano())\n\t\tassistants := []types.AssistantModel{\n\t\t\t{\n\t\t\t\tName: \"Tags Test 1\",\n\t\t\t\tType: \"assistant\",\n\t\t\t\tConnector: \"openai\",\n\t\t\t\tTags: []string{uniqueTag, \"common\"},\n\t\t\t\tShare: \"private\",\n\t\t\t},\n\t\t\t{\n\t\t\t\tName: \"Tags Test 2\",\n\t\t\t\tType: \"assistant\",\n\t\t\t\tConnector: \"openai\",\n\t\t\t\tTags: []string{uniqueTag, \"different\"},\n\t\t\t\tShare: \"private\",\n\t\t\t},\n\t\t\t{\n\t\t\t\tName: \"Tags Test 3\",\n\t\t\t\tType: \"assistant\",\n\t\t\t\tConnector: \"openai\",\n\t\t\t\tTags: []string{\"common\", \"another\"},\n\t\t\t\tShare: \"private\",\n\t\t\t},\n\t\t}\n\n\t\tfor _, asst := range assistants {\n\t\t\t_, err := store.SaveAssistant(&asst)\n\t\t\tif err != nil {\n\t\t\t\tt.Fatalf(\"Failed to create assistant: %v\", err)\n\t\t\t}\n\t\t}\n\n\t\t// Get all tags\n\t\ttags, err := store.GetAssistantTags(types.AssistantFilter{})\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to get tags: %v\", err)\n\t\t}\n\n\t\t// Verify we have some tags\n\t\tif len(tags) == 0 {\n\t\t\tt.Error(\"Expected at least some tags\")\n\t\t}\n\n\t\t// Verify tag structure\n\t\tfor _, tag := range tags {\n\t\t\tif tag.Value == \"\" {\n\t\t\t\tt.Error(\"Expected tag to have non-empty value\")\n\t\t\t}\n\t\t\tif tag.Label == \"\" {\n\t\t\t\tt.Error(\"Expected tag to have non-empty label\")\n\t\t\t}\n\t\t}\n\n\t\tt.Logf(\"Found %d unique tags\", len(tags))\n\t})\n\n\tt.Run(\"GetTagsWithFilter\", func(t *testing.T) {\n\t\t// Create test assistants with specific tags and attributes\n\t\tuniqueTag := fmt.Sprintf(\"filter-tag-%d\", time.Now().UnixNano())\n\t\tassistants := []types.AssistantModel{\n\t\t\t{\n\t\t\t\tName: \"Filtered Tags Test 1\",\n\t\t\t\tType: \"assistant\",\n\t\t\t\tConnector: \"openai\",\n\t\t\t\tTags: []string{uniqueTag, \"ai\"},\n\t\t\t\tShare: \"private\",\n\t\t\t\tBuiltIn: false,\n\t\t\t\tMentionable: true,\n\t\t\t},\n\t\t\t{\n\t\t\t\tName: \"Filtered Tags Test 2\",\n\t\t\t\tType: \"assistant\",\n\t\t\t\tConnector: \"anthropic\",\n\t\t\t\tTags: []string{uniqueTag, \"coding\"},\n\t\t\t\tShare: \"private\",\n\t\t\t\tBuiltIn: true,\n\t\t\t\tMentionable: false,\n\t\t\t},\n\t\t\t{\n\t\t\t\tName: \"Filtered Tags Test 3\",\n\t\t\t\tType: \"assistant\",\n\t\t\t\tConnector: \"openai\",\n\t\t\t\tTags: []string{uniqueTag, \"search\"},\n\t\t\t\tShare: \"private\",\n\t\t\t\tBuiltIn: false,\n\t\t\t\tAutomated: true,\n\t\t\t},\n\t\t}\n\n\t\tfor _, asst := range assistants {\n\t\t\t_, err := store.SaveAssistant(&asst)\n\t\t\tif err != nil {\n\t\t\t\tt.Fatalf(\"Failed to create assistant: %v\", err)\n\t\t\t}\n\t\t}\n\n\t\t// Test: Get tags filtered by connector\n\t\ttagsOpenAI, err := store.GetAssistantTags(types.AssistantFilter{\n\t\t\tConnector: \"openai\",\n\t\t})\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to get tags with connector filter: %v\", err)\n\t\t}\n\t\tt.Logf(\"Found %d tags for openai connector\", len(tagsOpenAI))\n\n\t\t// Test: Get tags filtered by built_in\n\t\tbuiltInFalse := false\n\t\ttagsNonBuiltIn, err := store.GetAssistantTags(types.AssistantFilter{\n\t\t\tBuiltIn: &builtInFalse,\n\t\t})\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to get tags with built_in filter: %v\", err)\n\t\t}\n\t\tt.Logf(\"Found %d tags for non-built-in assistants\", len(tagsNonBuiltIn))\n\n\t\t// Test: Get tags filtered by mentionable\n\t\tmentionableTrue := true\n\t\ttagsMentionable, err := store.GetAssistantTags(types.AssistantFilter{\n\t\t\tMentionable: &mentionableTrue,\n\t\t})\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to get tags with mentionable filter: %v\", err)\n\t\t}\n\t\tt.Logf(\"Found %d tags for mentionable assistants\", len(tagsMentionable))\n\n\t\t// Test: Get tags filtered by keywords\n\t\ttagsWithKeywords, err := store.GetAssistantTags(types.AssistantFilter{\n\t\t\tKeywords: \"Filtered Tags Test\",\n\t\t})\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to get tags with keywords filter: %v\", err)\n\t\t}\n\t\tt.Logf(\"Found %d tags with keywords filter\", len(tagsWithKeywords))\n\t})\n\n\tt.Run(\"GetTagsWithQueryFilter\", func(t *testing.T) {\n\t\t// Create test assistants with permission fields\n\t\tpermTag := fmt.Sprintf(\"perm-tag-%d\", time.Now().UnixNano())\n\t\tassistants := []types.AssistantModel{\n\t\t\t{\n\t\t\t\tName: \"Permission Tags Test 1\",\n\t\t\t\tType: \"assistant\",\n\t\t\t\tConnector: \"openai\",\n\t\t\t\tTags: []string{permTag, \"public-tag\"},\n\t\t\t\tShare: \"private\",\n\t\t\t\tPublic: true,\n\t\t\t\tYaoCreatedBy: \"user-1\",\n\t\t\t\tYaoTeamID: \"team-1\",\n\t\t\t},\n\t\t\t{\n\t\t\t\tName: \"Permission Tags Test 2\",\n\t\t\t\tType: \"assistant\",\n\t\t\t\tConnector: \"openai\",\n\t\t\t\tTags: []string{permTag, \"team-tag\"},\n\t\t\t\tShare: \"team\",\n\t\t\t\tPublic: false,\n\t\t\t\tYaoCreatedBy: \"user-2\",\n\t\t\t\tYaoTeamID: \"team-1\",\n\t\t\t},\n\t\t\t{\n\t\t\t\tName: \"Permission Tags Test 3\",\n\t\t\t\tType: \"assistant\",\n\t\t\t\tConnector: \"openai\",\n\t\t\t\tTags: []string{permTag, \"private-tag\"},\n\t\t\t\tShare: \"private\",\n\t\t\t\tPublic: false,\n\t\t\t\tYaoCreatedBy: \"user-3\",\n\t\t\t\tYaoTeamID: \"team-2\",\n\t\t\t},\n\t\t}\n\n\t\tfor _, asst := range assistants {\n\t\t\t_, err := store.SaveAssistant(&asst)\n\t\t\tif err != nil {\n\t\t\t\tt.Fatalf(\"Failed to create assistant: %v\", err)\n\t\t\t}\n\t\t}\n\n\t\t// Test: Get tags for public assistants only\n\t\ttagsPublic, err := store.GetAssistantTags(types.AssistantFilter{\n\t\t\tQueryFilter: func(qb query.Query) {\n\t\t\t\tqb.Where(\"public\", true)\n\t\t\t},\n\t\t})\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to get tags for public assistants: %v\", err)\n\t\t}\n\t\tt.Logf(\"Found %d tags for public assistants\", len(tagsPublic))\n\n\t\t// Test: Get tags for team-1 assistants\n\t\ttagsTeam1, err := store.GetAssistantTags(types.AssistantFilter{\n\t\t\tQueryFilter: func(qb query.Query) {\n\t\t\t\tqb.Where(\"__yao_team_id\", \"team-1\")\n\t\t\t},\n\t\t})\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to get tags for team-1: %v\", err)\n\t\t}\n\t\tt.Logf(\"Found %d tags for team-1 assistants\", len(tagsTeam1))\n\n\t\t// Test: Complex permission filter (public OR team-1 with share=team)\n\t\ttagsComplex, err := store.GetAssistantTags(types.AssistantFilter{\n\t\t\tQueryFilter: func(qb query.Query) {\n\t\t\t\tqb.Where(func(qb query.Query) {\n\t\t\t\t\tqb.Where(\"public\", true)\n\t\t\t\t}).OrWhere(func(qb query.Query) {\n\t\t\t\t\tqb.Where(\"__yao_team_id\", \"team-1\").\n\t\t\t\t\t\tWhere(\"share\", \"team\")\n\t\t\t\t})\n\t\t\t},\n\t\t})\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to get tags with complex filter: %v\", err)\n\t\t}\n\t\tt.Logf(\"Found %d tags with complex permission filter\", len(tagsComplex))\n\t})\n}\n\n// TestAssistantPermissionFields tests permission management fields\nfunc TestAssistantPermissionFields(t *testing.T) {\n\ttest.Prepare(t, config.Conf)\n\tdefer test.Clean()\n\n\tstore, err := xun.NewXun(types.Setting{\n\t\tConnector: \"default\",\n\t})\n\tif err != nil {\n\t\tt.Fatalf(\"Failed to create store: %v\", err)\n\t}\n\n\tt.Run(\"SaveWithPermissionFields\", func(t *testing.T) {\n\t\tassistant := &types.AssistantModel{\n\t\t\tName: \"Permission Test Assistant\",\n\t\t\tType: \"assistant\",\n\t\t\tConnector: \"openai\",\n\t\t\tDescription: \"Testing permission fields\",\n\t\t\tShare: \"private\",\n\t\t\tYaoCreatedBy: \"user-123\",\n\t\t\tYaoUpdatedBy: \"user-123\",\n\t\t\tYaoTeamID: \"team-456\",\n\t\t\tYaoTenantID: \"tenant-789\",\n\t\t}\n\n\t\tid, err := store.SaveAssistant(assistant)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to save assistant with permission fields: %v\", err)\n\t\t}\n\n\t\t// Retrieve and verify - default fields include permission fields\n\t\tretrieved, err := store.GetAssistant(id, nil)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to get assistant: %v\", err)\n\t\t}\n\n\t\tif retrieved.YaoCreatedBy != \"user-123\" {\n\t\t\tt.Errorf(\"Expected YaoCreatedBy 'user-123', got '%s'\", retrieved.YaoCreatedBy)\n\t\t}\n\t\tif retrieved.YaoUpdatedBy != \"user-123\" {\n\t\t\tt.Errorf(\"Expected YaoUpdatedBy 'user-123', got '%s'\", retrieved.YaoUpdatedBy)\n\t\t}\n\t\tif retrieved.YaoTeamID != \"team-456\" {\n\t\t\tt.Errorf(\"Expected YaoTeamID 'team-456', got '%s'\", retrieved.YaoTeamID)\n\t\t}\n\t\tif retrieved.YaoTenantID != \"tenant-789\" {\n\t\t\tt.Errorf(\"Expected YaoTenantID 'tenant-789', got '%s'\", retrieved.YaoTenantID)\n\t\t}\n\n\t\tt.Logf(\"Permission fields saved and retrieved successfully for assistant %s\", id)\n\t})\n\n\tt.Run(\"UpdatePermissionFields\", func(t *testing.T) {\n\t\t// Create assistant\n\t\tassistant := &types.AssistantModel{\n\t\t\tName: \"Update Permission Test\",\n\t\t\tType: \"assistant\",\n\t\t\tConnector: \"openai\",\n\t\t\tShare: \"private\",\n\t\t\tYaoCreatedBy: \"user-original\",\n\t\t\tYaoTeamID: \"team-original\",\n\t\t}\n\n\t\tid, err := store.SaveAssistant(assistant)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to create assistant: %v\", err)\n\t\t}\n\n\t\t// Update with new permission fields\n\t\tassistant.ID = id\n\t\tassistant.YaoUpdatedBy = \"user-updater\"\n\t\tassistant.YaoTenantID = \"tenant-new\"\n\n\t\t_, err = store.SaveAssistant(assistant)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to update assistant: %v\", err)\n\t\t}\n\n\t\t// Verify update - default fields include permission fields\n\t\tretrieved, err := store.GetAssistant(id, nil)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to get updated assistant: %v\", err)\n\t\t}\n\n\t\tif retrieved.YaoCreatedBy != \"user-original\" {\n\t\t\tt.Errorf(\"Expected YaoCreatedBy to remain 'user-original', got '%s'\", retrieved.YaoCreatedBy)\n\t\t}\n\t\tif retrieved.YaoUpdatedBy != \"user-updater\" {\n\t\t\tt.Errorf(\"Expected YaoUpdatedBy 'user-updater', got '%s'\", retrieved.YaoUpdatedBy)\n\t\t}\n\t\tif retrieved.YaoTenantID != \"tenant-new\" {\n\t\t\tt.Errorf(\"Expected YaoTenantID 'tenant-new', got '%s'\", retrieved.YaoTenantID)\n\t\t}\n\t})\n\n\tt.Run(\"EmptyPermissionFields\", func(t *testing.T) {\n\t\t// Create assistant without permission fields\n\t\tassistant := &types.AssistantModel{\n\t\t\tName: \"No Permission Fields\",\n\t\t\tType: \"assistant\",\n\t\t\tConnector: \"openai\",\n\t\t\tShare: \"private\",\n\t\t}\n\n\t\tid, err := store.SaveAssistant(assistant)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to save assistant: %v\", err)\n\t\t}\n\n\t\t// Retrieve and verify fields are empty\n\t\tretrieved, err := store.GetAssistant(id, nil)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to get assistant: %v\", err)\n\t\t}\n\n\t\tif retrieved.YaoCreatedBy != \"\" {\n\t\t\tt.Errorf(\"Expected empty YaoCreatedBy, got '%s'\", retrieved.YaoCreatedBy)\n\t\t}\n\t\tif retrieved.YaoUpdatedBy != \"\" {\n\t\t\tt.Errorf(\"Expected empty YaoUpdatedBy, got '%s'\", retrieved.YaoUpdatedBy)\n\t\t}\n\t\tif retrieved.YaoTeamID != \"\" {\n\t\t\tt.Errorf(\"Expected empty YaoTeamID, got '%s'\", retrieved.YaoTeamID)\n\t\t}\n\t\tif retrieved.YaoTenantID != \"\" {\n\t\t\tt.Errorf(\"Expected empty YaoTenantID, got '%s'\", retrieved.YaoTenantID)\n\t\t}\n\t})\n}\n\n// TestEmptyStringAsNull tests that empty strings are stored as NULL in database\nfunc TestEmptyStringAsNull(t *testing.T) {\n\ttest.Prepare(t, config.Conf)\n\tdefer test.Clean()\n\n\tstore, err := xun.NewXun(types.Setting{\n\t\tConnector: \"default\",\n\t})\n\tif err != nil {\n\t\tt.Fatalf(\"Failed to create store: %v\", err)\n\t}\n\n\tt.Run(\"EmptyStringsStoredAsNull\", func(t *testing.T) {\n\t\t// Create assistant with empty strings for nullable fields\n\t\t// According to assistant.mod.yao, nullable string fields are:\n\t\t// - name (nullable: true, but required by validation)\n\t\t// - avatar, description, path (nullable: true)\n\t\t// - share (nullable: false, but empty should trigger default)\n\t\tassistant := &types.AssistantModel{\n\t\t\tName: \"Test Null Fields\", // Required by validation\n\t\t\tType: \"assistant\",\n\t\t\tConnector: \"openai\",\n\t\t\tAvatar: \"\", // Empty string should become NULL (nullable: true)\n\t\t\tPath: \"\", // Empty string should become NULL (nullable: true)\n\t\t\tDescription: \"\", // Empty string should become NULL (nullable: true)\n\t\t\tShare: \"\", // Empty string should become NULL, then default \"private\" applied\n\t\t}\n\n\t\tid, err := store.SaveAssistant(assistant)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to save assistant: %v\", err)\n\t\t}\n\n\t\t// Retrieve and verify empty strings are returned (not stored as empty strings)\n\t\tretrieved, err := store.GetAssistant(id, nil)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to get assistant: %v\", err)\n\t\t}\n\n\t\t// Name should be preserved (required field)\n\t\tif retrieved.Name != \"Test Null Fields\" {\n\t\t\tt.Errorf(\"Expected Name 'Test Null Fields', got '%s'\", retrieved.Name)\n\t\t}\n\n\t\t// These nullable fields should be empty strings in Go (converted from NULL)\n\t\tif retrieved.Avatar != \"\" {\n\t\t\tt.Errorf(\"Expected empty Avatar, got '%s'\", retrieved.Avatar)\n\t\t}\n\t\tif retrieved.Path != \"\" {\n\t\t\tt.Errorf(\"Expected empty Path, got '%s'\", retrieved.Path)\n\t\t}\n\t\tif retrieved.Description != \"\" {\n\t\t\tt.Errorf(\"Expected empty Description, got '%s'\", retrieved.Description)\n\t\t}\n\t\t// Share should have default value \"private\" applied\n\t\tif retrieved.Share != \"private\" {\n\t\t\tt.Errorf(\"Expected Share to be 'private', got '%s'\", retrieved.Share)\n\t\t}\n\n\t\tt.Logf(\"Successfully verified empty strings are stored as NULL for assistant %s\", id)\n\t})\n\n\tt.Run(\"NonEmptyStringsPreserved\", func(t *testing.T) {\n\t\t// Create assistant with non-empty values\n\t\tassistant := &types.AssistantModel{\n\t\t\tName: \"Test Non-Empty Fields\",\n\t\t\tType: \"assistant\",\n\t\t\tConnector: \"openai\",\n\t\t\tAvatar: \"https://example.com/avatar.png\",\n\t\t\tPath: \"/path/to/assistant\",\n\t\t\tDescription: \"This is a description\",\n\t\t\tShare: \"private\",\n\t\t}\n\n\t\tid, err := store.SaveAssistant(assistant)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to save assistant: %v\", err)\n\t\t}\n\n\t\t// Retrieve and verify values are preserved - path is sensitive, need full fields\n\t\tretrieved, err := store.GetAssistant(id, types.AssistantFullFields)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to get assistant: %v\", err)\n\t\t}\n\n\t\tif retrieved.Avatar != \"https://example.com/avatar.png\" {\n\t\t\tt.Errorf(\"Expected Avatar 'https://example.com/avatar.png', got '%s'\", retrieved.Avatar)\n\t\t}\n\t\tif retrieved.Path != \"/path/to/assistant\" {\n\t\t\tt.Errorf(\"Expected Path '/path/to/assistant', got '%s'\", retrieved.Path)\n\t\t}\n\t\tif retrieved.Description != \"This is a description\" {\n\t\t\tt.Errorf(\"Expected Description 'This is a description', got '%s'\", retrieved.Description)\n\t\t}\n\t\tif retrieved.Share != \"private\" {\n\t\t\tt.Errorf(\"Expected Share 'private', got '%s'\", retrieved.Share)\n\t\t}\n\n\t\tt.Logf(\"Successfully verified non-empty strings are preserved for assistant %s\", id)\n\t})\n}\n\n// TestGetAssistantWithLocale tests retrieving assistant with locale translation\nfunc TestGetAssistantWithLocale(t *testing.T) {\n\ttest.Prepare(t, config.Conf)\n\tdefer test.Clean()\n\n\tstore, err := xun.NewXun(types.Setting{\n\t\tConnector: \"default\",\n\t})\n\tif err != nil {\n\t\tt.Fatalf(\"Failed to create store: %v\", err)\n\t}\n\n\tt.Run(\"GetAssistantWithLocaleTranslation\", func(t *testing.T) {\n\t\t// Create assistant with i18n locales\n\t\tassistant := &types.AssistantModel{\n\t\t\tName: \"{{name}}\",\n\t\t\tType: \"assistant\",\n\t\t\tConnector: \"openai\",\n\t\t\tDescription: \"{{description}}\",\n\t\t\tTags: []string{\"test\"},\n\t\t\tShare: \"private\",\n\t\t\tPlaceholder: &types.Placeholder{\n\t\t\t\tTitle: \"{{chat.title}}\",\n\t\t\t\tDescription: \"{{chat.description}}\",\n\t\t\t\tPrompts: []string{\"{{chat.prompts.0}}\", \"{{chat.prompts.1}}\"},\n\t\t\t},\n\t\t}\n\n\t\tid, err := store.SaveAssistant(assistant)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to create assistant: %v\", err)\n\t\t}\n\n\t\t// Setup i18n for testing\n\t\ti18n.Locales[id] = map[string]i18n.I18n{\n\t\t\t\"en\": {\n\t\t\t\tLocale: \"en\",\n\t\t\t\tMessages: map[string]any{\n\t\t\t\t\t\"name\": \"Test Assistant\",\n\t\t\t\t\t\"description\": \"This is a test assistant\",\n\t\t\t\t\t\"chat.title\": \"Chat with me\",\n\t\t\t\t\t\"chat.description\": \"Start a conversation\",\n\t\t\t\t\t\"chat.prompts.0\": \"How can I help you?\",\n\t\t\t\t\t\"chat.prompts.1\": \"What would you like to know?\",\n\t\t\t\t},\n\t\t\t},\n\t\t\t\"zh-cn\": {\n\t\t\t\tLocale: \"zh-cn\",\n\t\t\t\tMessages: map[string]any{\n\t\t\t\t\t\"name\": \"测试助手\",\n\t\t\t\t\t\"description\": \"这是一个测试助手\",\n\t\t\t\t\t\"chat.title\": \"与我聊天\",\n\t\t\t\t\t\"chat.description\": \"开始对话\",\n\t\t\t\t\t\"chat.prompts.0\": \"我能帮你什么？\",\n\t\t\t\t\t\"chat.prompts.1\": \"你想了解什么？\",\n\t\t\t\t},\n\t\t\t},\n\t\t}\n\n\t\t// Test English locale - request all fields for placeholder\n\t\tretrievedEN, err := store.GetAssistant(id, types.AssistantFullFields, \"en\")\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to get assistant with EN locale: %v\", err)\n\t\t}\n\n\t\tif retrievedEN.Name != \"Test Assistant\" {\n\t\t\tt.Errorf(\"Expected name 'Test Assistant', got '%s'\", retrievedEN.Name)\n\t\t}\n\t\tif retrievedEN.Description != \"This is a test assistant\" {\n\t\t\tt.Errorf(\"Expected description 'This is a test assistant', got '%s'\", retrievedEN.Description)\n\t\t}\n\t\tif retrievedEN.Placeholder == nil {\n\t\t\tt.Fatal(\"Expected placeholder to be set\")\n\t\t}\n\t\tif retrievedEN.Placeholder.Title != \"Chat with me\" {\n\t\t\tt.Errorf(\"Expected placeholder title 'Chat with me', got '%s'\", retrievedEN.Placeholder.Title)\n\t\t}\n\t\tif retrievedEN.Placeholder.Description != \"Start a conversation\" {\n\t\t\tt.Errorf(\"Expected placeholder description 'Start a conversation', got '%s'\", retrievedEN.Placeholder.Description)\n\t\t}\n\t\tif len(retrievedEN.Placeholder.Prompts) != 2 {\n\t\t\tt.Errorf(\"Expected 2 placeholder prompts, got %d\", len(retrievedEN.Placeholder.Prompts))\n\t\t}\n\t\tif retrievedEN.Placeholder.Prompts[0] != \"How can I help you?\" {\n\t\t\tt.Errorf(\"Expected first prompt 'How can I help you?', got '%s'\", retrievedEN.Placeholder.Prompts[0])\n\t\t}\n\n\t\t// Test Chinese locale - request all fields for placeholder\n\t\tretrievedZH, err := store.GetAssistant(id, types.AssistantFullFields, \"zh-cn\")\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to get assistant with ZH locale: %v\", err)\n\t\t}\n\n\t\tif retrievedZH.Name != \"测试助手\" {\n\t\t\tt.Errorf(\"Expected name '测试助手', got '%s'\", retrievedZH.Name)\n\t\t}\n\t\tif retrievedZH.Description != \"这是一个测试助手\" {\n\t\t\tt.Errorf(\"Expected description '这是一个测试助手', got '%s'\", retrievedZH.Description)\n\t\t}\n\t\tif retrievedZH.Placeholder == nil {\n\t\t\tt.Fatal(\"Expected placeholder to be set\")\n\t\t}\n\t\tif retrievedZH.Placeholder.Title != \"与我聊天\" {\n\t\t\tt.Errorf(\"Expected placeholder title '与我聊天', got '%s'\", retrievedZH.Placeholder.Title)\n\t\t}\n\n\t\t// Test without locale (should return original {{...}} values) - request all fields for placeholder\n\t\tretrievedNoLocale, err := store.GetAssistant(id, types.AssistantFullFields)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to get assistant without locale: %v\", err)\n\t\t}\n\n\t\tif retrievedNoLocale.Name != \"{{name}}\" {\n\t\t\tt.Errorf(\"Expected original name '{{name}}', got '%s'\", retrievedNoLocale.Name)\n\t\t}\n\t\tif retrievedNoLocale.Description != \"{{description}}\" {\n\t\t\tt.Errorf(\"Expected original description '{{description}}', got '%s'\", retrievedNoLocale.Description)\n\t\t}\n\n\t\t// Cleanup\n\t\tdelete(i18n.Locales, id)\n\t\tt.Logf(\"Successfully tested locale translation for assistant %s\", id)\n\t})\n}\n\n// TestGetAssistantsWithLocale tests retrieving multiple assistants with locale translation\nfunc TestGetAssistantsWithLocale(t *testing.T) {\n\ttest.Prepare(t, config.Conf)\n\tdefer test.Clean()\n\n\tstore, err := xun.NewXun(types.Setting{\n\t\tConnector: \"default\",\n\t})\n\tif err != nil {\n\t\tt.Fatalf(\"Failed to create store: %v\", err)\n\t}\n\n\tt.Run(\"GetAssistantsWithLocaleTranslation\", func(t *testing.T) {\n\t\t// Create assistant with i18n locales\n\t\tassistant := &types.AssistantModel{\n\t\t\tName: \"{{name}}\",\n\t\t\tType: \"assistant\",\n\t\t\tConnector: \"openai\",\n\t\t\tDescription: \"{{description}}\",\n\t\t\tTags: []string{\"locale-test\"},\n\t\t\tShare: \"private\",\n\t\t}\n\n\t\tid, err := store.SaveAssistant(assistant)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to create assistant: %v\", err)\n\t\t}\n\n\t\t// Setup i18n for testing\n\t\ti18n.Locales[id] = map[string]i18n.I18n{\n\t\t\t\"en\": {\n\t\t\t\tLocale: \"en\",\n\t\t\t\tMessages: map[string]any{\n\t\t\t\t\t\"name\": \"List Test Assistant\",\n\t\t\t\t\t\"description\": \"This appears in the list\",\n\t\t\t\t},\n\t\t\t},\n\t\t\t\"zh-cn\": {\n\t\t\t\tLocale: \"zh-cn\",\n\t\t\t\tMessages: map[string]any{\n\t\t\t\t\t\"name\": \"列表测试助手\",\n\t\t\t\t\t\"description\": \"这出现在列表中\",\n\t\t\t\t},\n\t\t\t},\n\t\t}\n\n\t\t// Test GetAssistants with English locale\n\t\tresponseEN, err := store.GetAssistants(types.AssistantFilter{\n\t\t\tTags: []string{\"locale-test\"},\n\t\t\tPage: 1,\n\t\t\tPageSize: 20,\n\t\t}, \"en\")\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to get assistants with EN locale: %v\", err)\n\t\t}\n\n\t\tif len(responseEN.Data) < 1 {\n\t\t\tt.Fatal(\"Expected at least 1 assistant in response\")\n\t\t}\n\n\t\tfound := false\n\t\tfor _, asst := range responseEN.Data {\n\t\t\tif asst.ID == id {\n\t\t\t\tfound = true\n\t\t\t\tif asst.Name != \"List Test Assistant\" {\n\t\t\t\t\tt.Errorf(\"Expected name 'List Test Assistant', got '%s'\", asst.Name)\n\t\t\t\t}\n\t\t\t\tif asst.Description != \"This appears in the list\" {\n\t\t\t\t\tt.Errorf(\"Expected description 'This appears in the list', got '%s'\", asst.Description)\n\t\t\t\t}\n\t\t\t\tbreak\n\t\t\t}\n\t\t}\n\n\t\tif !found {\n\t\t\tt.Error(\"Expected to find the test assistant in the list\")\n\t\t}\n\n\t\t// Test GetAssistants with Chinese locale\n\t\tresponseZH, err := store.GetAssistants(types.AssistantFilter{\n\t\t\tTags: []string{\"locale-test\"},\n\t\t\tPage: 1,\n\t\t\tPageSize: 20,\n\t\t}, \"zh-cn\")\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to get assistants with ZH locale: %v\", err)\n\t\t}\n\n\t\tfound = false\n\t\tfor _, asst := range responseZH.Data {\n\t\t\tif asst.ID == id {\n\t\t\t\tfound = true\n\t\t\t\tif asst.Name != \"列表测试助手\" {\n\t\t\t\t\tt.Errorf(\"Expected name '列表测试助手', got '%s'\", asst.Name)\n\t\t\t\t}\n\t\t\t\tif asst.Description != \"这出现在列表中\" {\n\t\t\t\t\tt.Errorf(\"Expected description '这出现在列表中', got '%s'\", asst.Description)\n\t\t\t\t}\n\t\t\t\tbreak\n\t\t\t}\n\t\t}\n\n\t\tif !found {\n\t\t\tt.Error(\"Expected to find the test assistant in the list\")\n\t\t}\n\n\t\t// Cleanup\n\t\tdelete(i18n.Locales, id)\n\t\tt.Logf(\"Successfully tested locale translation for assistants list\")\n\t})\n}\n\n// TestGetAssistantsWithQueryFilter tests using QueryFilter for permission filtering\nfunc TestGetAssistantsWithQueryFilter(t *testing.T) {\n\ttest.Prepare(t, config.Conf)\n\tdefer test.Clean()\n\n\tstore, err := xun.NewXun(types.Setting{\n\t\tConnector: \"default\",\n\t})\n\tif err != nil {\n\t\tt.Fatalf(\"Failed to create store: %v\", err)\n\t}\n\n\t// Create test assistants with different permission settings\n\tassistants := []types.AssistantModel{\n\t\t{\n\t\t\tName: \"Public Assistant\",\n\t\t\tType: \"assistant\",\n\t\t\tConnector: \"openai\",\n\t\t\tDescription: \"Public assistant visible to all\",\n\t\t\tTags: []string{\"query-filter-test\"},\n\t\t\tPublic: true,\n\t\t\tShare: \"private\",\n\t\t\tYaoCreatedBy: \"user-1\",\n\t\t\tYaoTeamID: \"team-1\",\n\t\t},\n\t\t{\n\t\t\tName: \"Team Shared Assistant\",\n\t\t\tType: \"assistant\",\n\t\t\tConnector: \"openai\",\n\t\t\tDescription: \"Team shared assistant\",\n\t\t\tTags: []string{\"query-filter-test\"},\n\t\t\tPublic: false,\n\t\t\tShare: \"team\",\n\t\t\tYaoCreatedBy: \"user-2\",\n\t\t\tYaoTeamID: \"team-1\",\n\t\t},\n\t\t{\n\t\t\tName: \"Private Assistant Owner\",\n\t\t\tType: \"assistant\",\n\t\t\tConnector: \"openai\",\n\t\t\tDescription: \"Private assistant owned by user-1\",\n\t\t\tTags: []string{\"query-filter-test\"},\n\t\t\tPublic: false,\n\t\t\tShare: \"private\",\n\t\t\tYaoCreatedBy: \"user-1\",\n\t\t\tYaoTeamID: \"team-1\",\n\t\t},\n\t\t{\n\t\t\tName: \"Private Assistant Other\",\n\t\t\tType: \"assistant\",\n\t\t\tConnector: \"openai\",\n\t\t\tDescription: \"Private assistant owned by user-3\",\n\t\t\tTags: []string{\"query-filter-test\"},\n\t\t\tPublic: false,\n\t\t\tShare: \"private\",\n\t\t\tYaoCreatedBy: \"user-3\",\n\t\t\tYaoTeamID: \"team-2\",\n\t\t},\n\t}\n\n\tcreatedIDs := []string{}\n\tfor _, asst := range assistants {\n\t\tid, err := store.SaveAssistant(&asst)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to create assistant: %v\", err)\n\t\t}\n\t\tcreatedIDs = append(createdIDs, id)\n\t}\n\n\tt.Run(\"FilterByPublic\", func(t *testing.T) {\n\t\t// QueryFilter: only public assistants\n\t\tresponse, err := store.GetAssistants(types.AssistantFilter{\n\t\t\tTags: []string{\"query-filter-test\"},\n\t\t\tPage: 1,\n\t\t\tPageSize: 20,\n\t\t\tQueryFilter: func(qb query.Query) {\n\t\t\t\tqb.Where(\"public\", true)\n\t\t\t},\n\t\t})\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to get public assistants: %v\", err)\n\t\t}\n\n\t\tif len(response.Data) != 1 {\n\t\t\tt.Errorf(\"Expected 1 public assistant, got %d\", len(response.Data))\n\t\t}\n\n\t\tif len(response.Data) > 0 && response.Data[0].Name != \"Public Assistant\" {\n\t\t\tt.Errorf(\"Expected 'Public Assistant', got '%s'\", response.Data[0].Name)\n\t\t}\n\t})\n\n\tt.Run(\"FilterByTeamAndShare\", func(t *testing.T) {\n\t\t// QueryFilter: team-1 assistants that are shared with team\n\t\tresponse, err := store.GetAssistants(types.AssistantFilter{\n\t\t\tTags: []string{\"query-filter-test\"},\n\t\t\tPage: 1,\n\t\t\tPageSize: 20,\n\t\t\tQueryFilter: func(qb query.Query) {\n\t\t\t\tqb.Where(\"__yao_team_id\", \"team-1\").\n\t\t\t\t\tWhere(\"share\", \"team\")\n\t\t\t},\n\t\t})\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to get team shared assistants: %v\", err)\n\t\t}\n\n\t\tif len(response.Data) != 1 {\n\t\t\tt.Errorf(\"Expected 1 team shared assistant, got %d\", len(response.Data))\n\t\t}\n\n\t\tif len(response.Data) > 0 && response.Data[0].Name != \"Team Shared Assistant\" {\n\t\t\tt.Errorf(\"Expected 'Team Shared Assistant', got '%s'\", response.Data[0].Name)\n\t\t}\n\t})\n\n\tt.Run(\"FilterByOwner\", func(t *testing.T) {\n\t\t// QueryFilter: assistants created by user-1\n\t\tresponse, err := store.GetAssistants(types.AssistantFilter{\n\t\t\tTags: []string{\"query-filter-test\"},\n\t\t\tPage: 1,\n\t\t\tPageSize: 20,\n\t\t\tQueryFilter: func(qb query.Query) {\n\t\t\t\tqb.Where(\"__yao_created_by\", \"user-1\")\n\t\t\t},\n\t\t})\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to get user-1 assistants: %v\", err)\n\t\t}\n\n\t\tif len(response.Data) != 2 {\n\t\t\tt.Errorf(\"Expected 2 assistants for user-1, got %d\", len(response.Data))\n\t\t}\n\n\t\tfor _, asst := range response.Data {\n\t\t\tif asst.YaoCreatedBy != \"user-1\" {\n\t\t\t\tt.Errorf(\"Expected creator 'user-1', got '%s'\", asst.YaoCreatedBy)\n\t\t\t}\n\t\t}\n\t})\n\n\tt.Run(\"ComplexQueryFilter\", func(t *testing.T) {\n\t\t// Complex QueryFilter: (public = true) OR (team_id = team-1 AND (created_by = user-1 OR share = team))\n\t\tresponse, err := store.GetAssistants(types.AssistantFilter{\n\t\t\tTags: []string{\"query-filter-test\"},\n\t\t\tPage: 1,\n\t\t\tPageSize: 20,\n\t\t\tQueryFilter: func(qb query.Query) {\n\t\t\t\tqb.Where(func(qb query.Query) {\n\t\t\t\t\t// Public assistants\n\t\t\t\t\tqb.Where(\"public\", true)\n\t\t\t\t}).OrWhere(func(qb query.Query) {\n\t\t\t\t\t// Team assistants where user is creator or shared with team\n\t\t\t\t\tqb.Where(\"__yao_team_id\", \"team-1\").Where(func(qb query.Query) {\n\t\t\t\t\t\tqb.Where(\"__yao_created_by\", \"user-1\").\n\t\t\t\t\t\t\tOrWhere(\"share\", \"team\")\n\t\t\t\t\t})\n\t\t\t\t})\n\t\t\t},\n\t\t})\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to get filtered assistants: %v\", err)\n\t\t}\n\n\t\t// Should find: Public Assistant, Team Shared Assistant, Private Assistant Owner\n\t\tif len(response.Data) != 3 {\n\t\t\tt.Errorf(\"Expected 3 assistants, got %d\", len(response.Data))\n\t\t}\n\n\t\t// Verify we got the right assistants\n\t\tnames := make(map[string]bool)\n\t\tfor _, asst := range response.Data {\n\t\t\tnames[asst.Name] = true\n\t\t}\n\n\t\texpectedNames := []string{\"Public Assistant\", \"Team Shared Assistant\", \"Private Assistant Owner\"}\n\t\tfor _, name := range expectedNames {\n\t\t\tif !names[name] {\n\t\t\t\tt.Errorf(\"Expected to find '%s' in results\", name)\n\t\t\t}\n\t\t}\n\n\t\t// Should NOT find Private Assistant Other\n\t\tif names[\"Private Assistant Other\"] {\n\t\t\tt.Error(\"Should not find 'Private Assistant Other' in results\")\n\t\t}\n\t})\n\n\tt.Run(\"QueryFilterWithNullCheck\", func(t *testing.T) {\n\t\t// QueryFilter: assistants where team_id is null\n\t\tresponse, err := store.GetAssistants(types.AssistantFilter{\n\t\t\tTags: []string{\"query-filter-test\"},\n\t\t\tPage: 1,\n\t\t\tPageSize: 20,\n\t\t\tQueryFilter: func(qb query.Query) {\n\t\t\t\tqb.WhereNull(\"__yao_team_id\")\n\t\t\t},\n\t\t})\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to get assistants with null team_id: %v\", err)\n\t\t}\n\n\t\t// All test assistants have team_id, so should find 0\n\t\tif len(response.Data) != 0 {\n\t\t\tt.Errorf(\"Expected 0 assistants with null team_id, got %d\", len(response.Data))\n\t\t}\n\t})\n\n\tt.Run(\"QueryFilterCombinedWithOtherFilters\", func(t *testing.T) {\n\t\t// Combine QueryFilter with other filters\n\t\tresponse, err := store.GetAssistants(types.AssistantFilter{\n\t\t\tTags: []string{\"query-filter-test\"},\n\t\t\tConnector: \"openai\",\n\t\t\tPage: 1,\n\t\t\tPageSize: 20,\n\t\t\tQueryFilter: func(qb query.Query) {\n\t\t\t\tqb.Where(\"public\", true)\n\t\t\t},\n\t\t})\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to get combined filtered assistants: %v\", err)\n\t\t}\n\n\t\t// Should only find public openai assistants\n\t\tif len(response.Data) != 1 {\n\t\t\tt.Errorf(\"Expected 1 assistant, got %d\", len(response.Data))\n\t\t}\n\n\t\tif len(response.Data) > 0 {\n\t\t\tif response.Data[0].Connector != \"openai\" {\n\t\t\t\tt.Errorf(\"Expected connector 'openai', got '%s'\", response.Data[0].Connector)\n\t\t\t}\n\t\t\tif !response.Data[0].Public {\n\t\t\t\tt.Error(\"Expected public assistant\")\n\t\t\t}\n\t\t}\n\t})\n\n\t// Cleanup\n\tfor _, id := range createdIDs {\n\t\t_ = store.DeleteAssistant(id)\n\t}\n}\n\n// TestUpdateAssistant tests the UpdateAssistant method for incremental updates\nfunc TestUpdateAssistant(t *testing.T) {\n\ttest.Prepare(t, config.Conf)\n\tdefer test.Clean()\n\n\tstore, err := xun.NewXun(types.Setting{\n\t\tConnector: \"default\",\n\t})\n\tif err != nil {\n\t\tt.Fatalf(\"Failed to create store: %v\", err)\n\t}\n\n\tt.Run(\"UpdateSingleField\", func(t *testing.T) {\n\t\t// Create assistant\n\t\tassistant := &types.AssistantModel{\n\t\t\tName: \"Original Name\",\n\t\t\tType: \"assistant\",\n\t\t\tConnector: \"openai\",\n\t\t\tDescription: \"Original description\",\n\t\t\tTags: []string{\"original\"},\n\t\t\tShare: \"private\",\n\t\t}\n\n\t\tid, err := store.SaveAssistant(assistant)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to create assistant: %v\", err)\n\t\t}\n\n\t\t// Update only description\n\t\tupdates := map[string]interface{}{\n\t\t\t\"description\": \"Updated description\",\n\t\t}\n\n\t\terr = store.UpdateAssistant(id, updates)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to update assistant: %v\", err)\n\t\t}\n\n\t\t// Verify update - need full fields to see tags\n\t\tretrieved, err := store.GetAssistant(id, types.AssistantFullFields)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to retrieve assistant: %v\", err)\n\t\t}\n\n\t\tif retrieved.Description != \"Updated description\" {\n\t\t\tt.Errorf(\"Expected description 'Updated description', got '%s'\", retrieved.Description)\n\t\t}\n\t\t// Other fields should remain unchanged\n\t\tif retrieved.Name != \"Original Name\" {\n\t\t\tt.Errorf(\"Expected name 'Original Name', got '%s'\", retrieved.Name)\n\t\t}\n\t\tif len(retrieved.Tags) != 1 || retrieved.Tags[0] != \"original\" {\n\t\t\tt.Errorf(\"Expected tags [original], got %v\", retrieved.Tags)\n\t\t}\n\t})\n\n\tt.Run(\"UpdateMultipleFields\", func(t *testing.T) {\n\t\t// Create assistant\n\t\tassistant := &types.AssistantModel{\n\t\t\tName: \"Test Assistant\",\n\t\t\tType: \"assistant\",\n\t\t\tConnector: \"openai\",\n\t\t\tDescription: \"Test description\",\n\t\t\tSort: 100,\n\t\t\tMentionable: false,\n\t\t\tShare: \"private\",\n\t\t}\n\n\t\tid, err := store.SaveAssistant(assistant)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to create assistant: %v\", err)\n\t\t}\n\n\t\t// Update multiple fields\n\t\tupdates := map[string]interface{}{\n\t\t\t\"name\": \"Updated Name\",\n\t\t\t\"description\": \"Updated description\",\n\t\t\t\"sort\": 200,\n\t\t\t\"mentionable\": true,\n\t\t}\n\n\t\terr = store.UpdateAssistant(id, updates)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to update assistant: %v\", err)\n\t\t}\n\n\t\t// Verify all updates - use default fields (includes name, description, sort, mentionable)\n\t\tretrieved, err := store.GetAssistant(id, nil)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to retrieve assistant: %v\", err)\n\t\t}\n\n\t\tif retrieved.Name != \"Updated Name\" {\n\t\t\tt.Errorf(\"Expected name 'Updated Name', got '%s'\", retrieved.Name)\n\t\t}\n\t\tif retrieved.Description != \"Updated description\" {\n\t\t\tt.Errorf(\"Expected description 'Updated description', got '%s'\", retrieved.Description)\n\t\t}\n\t\tif retrieved.Sort != 200 {\n\t\t\tt.Errorf(\"Expected sort 200, got %d\", retrieved.Sort)\n\t\t}\n\t\tif !retrieved.Mentionable {\n\t\t\tt.Error(\"Expected mentionable to be true\")\n\t\t}\n\t})\n\n\tt.Run(\"UpdateJSONFields\", func(t *testing.T) {\n\t\t// Create assistant with complex fields\n\t\tassistant := &types.AssistantModel{\n\t\t\tName: \"JSON Test\",\n\t\t\tType: \"assistant\",\n\t\t\tConnector: \"openai\",\n\t\t\tTags: []string{\"tag1\", \"tag2\"},\n\t\t\tOptions: map[string]interface{}{\"temperature\": 0.7},\n\t\t\tPrompts: []types.Prompt{\n\t\t\t\t{Role: \"system\", Content: \"Original system prompt\"},\n\t\t\t},\n\t\t\tShare: \"private\",\n\t\t}\n\n\t\tid, err := store.SaveAssistant(assistant)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to create assistant: %v\", err)\n\t\t}\n\n\t\t// Update JSON fields\n\t\tupdates := map[string]interface{}{\n\t\t\t\"tags\": []string{\"updated\", \"new-tags\"},\n\t\t\t\"options\": map[string]interface{}{\n\t\t\t\t\"temperature\": 0.9,\n\t\t\t\t\"max_tokens\": 2000,\n\t\t\t},\n\t\t\t\"prompts\": []types.Prompt{\n\t\t\t\t{Role: \"system\", Content: \"Updated system prompt\"},\n\t\t\t\t{Role: \"user\", Content: \"New user prompt\"},\n\t\t\t},\n\t\t}\n\n\t\terr = store.UpdateAssistant(id, updates)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to update JSON fields: %v\", err)\n\t\t}\n\n\t\t// Verify updates - need full fields for tags, options, prompts\n\t\tretrieved, err := store.GetAssistant(id, types.AssistantFullFields)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to retrieve assistant: %v\", err)\n\t\t}\n\n\t\tif len(retrieved.Tags) != 2 || retrieved.Tags[0] != \"updated\" {\n\t\t\tt.Errorf(\"Expected tags [updated, new-tags], got %v\", retrieved.Tags)\n\t\t}\n\t\tif temp, ok := retrieved.Options[\"temperature\"].(float64); !ok || temp != 0.9 {\n\t\t\tt.Errorf(\"Expected temperature 0.9, got %v\", retrieved.Options[\"temperature\"])\n\t\t}\n\t\tif len(retrieved.Prompts) != 2 {\n\t\t\tt.Errorf(\"Expected 2 prompts, got %d\", len(retrieved.Prompts))\n\t\t}\n\t\tif retrieved.Prompts[0].Content != \"Updated system prompt\" {\n\t\t\tt.Errorf(\"Expected updated system prompt, got '%s'\", retrieved.Prompts[0].Content)\n\t\t}\n\t})\n\n\tt.Run(\"UpdateKBAndMCP\", func(t *testing.T) {\n\t\t// Create assistant\n\t\tassistant := &types.AssistantModel{\n\t\t\tName: \"KB MCP Test\",\n\t\t\tType: \"assistant\",\n\t\t\tConnector: \"openai\",\n\t\t\tShare: \"private\",\n\t\t}\n\n\t\tid, err := store.SaveAssistant(assistant)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to create assistant: %v\", err)\n\t\t}\n\n\t\t// Update KB and MCP\n\t\tupdates := map[string]interface{}{\n\t\t\t\"kb\": map[string]interface{}{\n\t\t\t\t\"collections\": []string{\"collection1\", \"collection2\"},\n\t\t\t},\n\t\t\t\"mcp\": map[string]interface{}{\n\t\t\t\t\"servers\": []string{\"server1\", \"server2\"},\n\t\t\t},\n\t\t}\n\n\t\terr = store.UpdateAssistant(id, updates)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to update KB and MCP: %v\", err)\n\t\t}\n\n\t\t// Verify updates - KB and MCP are in default fields\n\t\tretrieved, err := store.GetAssistant(id, nil)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to retrieve assistant: %v\", err)\n\t\t}\n\n\t\tif retrieved.KB == nil || len(retrieved.KB.Collections) != 2 {\n\t\t\tt.Errorf(\"Expected 2 KB collections, got %v\", retrieved.KB)\n\t\t}\n\t\tif retrieved.MCP == nil || len(retrieved.MCP.Servers) != 2 {\n\t\t\tt.Errorf(\"Expected 2 MCP servers, got %v\", retrieved.MCP)\n\t\t}\n\t\tif retrieved.MCP.Servers[0].ServerID != \"server1\" {\n\t\t\tt.Errorf(\"Expected first server 'server1', got '%s'\", retrieved.MCP.Servers[0].ServerID)\n\t\t}\n\t})\n\n\tt.Run(\"UpdateKBDBAndMCP\", func(t *testing.T) {\n\t\t// Create assistant\n\t\tassistant := &types.AssistantModel{\n\t\t\tName: \"KB DB MCP Test\",\n\t\t\tType: \"assistant\",\n\t\t\tConnector: \"openai\",\n\t\t\tShare: \"private\",\n\t\t}\n\n\t\tid, err := store.SaveAssistant(assistant)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to create assistant: %v\", err)\n\t\t}\n\n\t\t// Update KB, DB and MCP\n\t\tupdates := map[string]interface{}{\n\t\t\t\"kb\": map[string]interface{}{\n\t\t\t\t\"collections\": []string{\"collection1\", \"collection2\"},\n\t\t\t},\n\t\t\t\"db\": map[string]interface{}{\n\t\t\t\t\"models\": []string{\"model1\", \"model2\"},\n\t\t\t},\n\t\t\t\"mcp\": map[string]interface{}{\n\t\t\t\t\"servers\": []string{\"server1\", \"server2\"},\n\t\t\t},\n\t\t}\n\n\t\terr = store.UpdateAssistant(id, updates)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to update KB, DB and MCP: %v\", err)\n\t\t}\n\n\t\t// Verify updates - KB, DB and MCP are in default fields\n\t\tretrieved, err := store.GetAssistant(id, nil)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to retrieve assistant: %v\", err)\n\t\t}\n\n\t\tif retrieved.KB == nil || len(retrieved.KB.Collections) != 2 {\n\t\t\tt.Errorf(\"Expected 2 KB collections, got %v\", retrieved.KB)\n\t\t}\n\t\tif retrieved.DB == nil || len(retrieved.DB.Models) != 2 {\n\t\t\tt.Errorf(\"Expected 2 DB models, got %v\", retrieved.DB)\n\t\t}\n\t\tif retrieved.DB.Models[0] != \"model1\" {\n\t\t\tt.Errorf(\"Expected first model 'model1', got '%s'\", retrieved.DB.Models[0])\n\t\t}\n\t\tif retrieved.MCP == nil || len(retrieved.MCP.Servers) != 2 {\n\t\t\tt.Errorf(\"Expected 2 MCP servers, got %v\", retrieved.MCP)\n\t\t}\n\t\tif retrieved.MCP.Servers[0].ServerID != \"server1\" {\n\t\t\tt.Errorf(\"Expected first server 'server1', got '%s'\", retrieved.MCP.Servers[0].ServerID)\n\t\t}\n\t})\n\n\tt.Run(\"UpdateDBWithOptions\", func(t *testing.T) {\n\t\t// Create assistant\n\t\tassistant := &types.AssistantModel{\n\t\t\tName: \"DB Advanced Test\",\n\t\t\tType: \"assistant\",\n\t\t\tConnector: \"openai\",\n\t\t\tShare: \"private\",\n\t\t}\n\n\t\tid, err := store.SaveAssistant(assistant)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to create assistant: %v\", err)\n\t\t}\n\n\t\t// Update with DB using advanced configuration\n\t\tupdates := map[string]interface{}{\n\t\t\t\"db\": map[string]interface{}{\n\t\t\t\t\"models\": []string{\"user\", \"product\", \"order\"},\n\t\t\t\t\"options\": map[string]interface{}{\n\t\t\t\t\t\"limit\": 100,\n\t\t\t\t\t\"offset\": 0,\n\t\t\t\t},\n\t\t\t},\n\t\t}\n\n\t\terr = store.UpdateAssistant(id, updates)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to update DB: %v\", err)\n\t\t}\n\n\t\t// Verify updates - DB is in default fields\n\t\tretrieved, err := store.GetAssistant(id, nil)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to retrieve assistant: %v\", err)\n\t\t}\n\n\t\tif retrieved.DB == nil {\n\t\t\tt.Fatal(\"Expected DB to be set\")\n\t\t}\n\t\tif len(retrieved.DB.Models) != 3 {\n\t\t\tt.Errorf(\"Expected 3 DB models, got %d\", len(retrieved.DB.Models))\n\t\t}\n\t\tif retrieved.DB.Models[0] != \"user\" {\n\t\t\tt.Errorf(\"Expected first model 'user', got '%s'\", retrieved.DB.Models[0])\n\t\t}\n\t\tif retrieved.DB.Options == nil {\n\t\t\tt.Error(\"Expected DB options to be set\")\n\t\t} else {\n\t\t\tif limit, ok := retrieved.DB.Options[\"limit\"].(float64); !ok || limit != 100 {\n\t\t\t\tt.Errorf(\"Expected DB limit 100, got %v\", retrieved.DB.Options[\"limit\"])\n\t\t\t}\n\t\t}\n\t})\n\n\tt.Run(\"UpdateModesAndDefaultMode\", func(t *testing.T) {\n\t\t// Create assistant\n\t\tassistant := &types.AssistantModel{\n\t\t\tName: \"Modes Test\",\n\t\t\tType: \"assistant\",\n\t\t\tConnector: \"openai\",\n\t\t\tShare: \"private\",\n\t\t}\n\n\t\tid, err := store.SaveAssistant(assistant)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to create assistant: %v\", err)\n\t\t}\n\n\t\t// Update with modes and default_mode\n\t\tupdates := map[string]interface{}{\n\t\t\t\"modes\": []string{\"chat\", \"task\", \"analyze\"},\n\t\t\t\"default_mode\": \"chat\",\n\t\t}\n\n\t\terr = store.UpdateAssistant(id, updates)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to update modes: %v\", err)\n\t\t}\n\n\t\t// Verify updates - modes and default_mode are in default fields\n\t\tretrieved, err := store.GetAssistant(id, nil)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to retrieve assistant: %v\", err)\n\t\t}\n\n\t\tif retrieved.Modes == nil || len(retrieved.Modes) != 3 {\n\t\t\tt.Errorf(\"Expected 3 modes, got %v\", retrieved.Modes)\n\t\t}\n\t\tif retrieved.Modes[0] != \"chat\" {\n\t\t\tt.Errorf(\"Expected first mode 'chat', got '%s'\", retrieved.Modes[0])\n\t\t}\n\t\tif retrieved.DefaultMode != \"chat\" {\n\t\t\tt.Errorf(\"Expected default_mode 'chat', got '%s'\", retrieved.DefaultMode)\n\t\t}\n\t})\n\n\tt.Run(\"UpdateModesOnly\", func(t *testing.T) {\n\t\t// Create assistant with default_mode\n\t\tassistant := &types.AssistantModel{\n\t\t\tName: \"Modes Only Test\",\n\t\t\tType: \"assistant\",\n\t\t\tConnector: \"openai\",\n\t\t\tShare: \"private\",\n\t\t\tDefaultMode: \"task\",\n\t\t}\n\n\t\tid, err := store.SaveAssistant(assistant)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to create assistant: %v\", err)\n\t\t}\n\n\t\t// Update only modes\n\t\tupdates := map[string]interface{}{\n\t\t\t\"modes\": []string{\"chat\", \"task\"},\n\t\t}\n\n\t\terr = store.UpdateAssistant(id, updates)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to update modes: %v\", err)\n\t\t}\n\n\t\t// Verify updates - default_mode should remain unchanged\n\t\tretrieved, err := store.GetAssistant(id, nil)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to retrieve assistant: %v\", err)\n\t\t}\n\n\t\tif len(retrieved.Modes) != 2 {\n\t\t\tt.Errorf(\"Expected 2 modes, got %d\", len(retrieved.Modes))\n\t\t}\n\t\tif retrieved.DefaultMode != \"task\" {\n\t\t\tt.Errorf(\"Expected default_mode to remain 'task', got '%s'\", retrieved.DefaultMode)\n\t\t}\n\t})\n\n\tt.Run(\"UpdateMCPWithToolsAndResources\", func(t *testing.T) {\n\t\t// Create assistant\n\t\tassistant := &types.AssistantModel{\n\t\t\tName: \"MCP Advanced Test\",\n\t\t\tType: \"assistant\",\n\t\t\tConnector: \"openai\",\n\t\t\tShare: \"private\",\n\t\t}\n\n\t\tid, err := store.SaveAssistant(assistant)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to create assistant: %v\", err)\n\t\t}\n\n\t\t// Update with MCP servers using advanced configuration\n\t\tupdates := map[string]interface{}{\n\t\t\t\"mcp\": map[string]interface{}{\n\t\t\t\t\"servers\": []interface{}{\n\t\t\t\t\t\"server1\", // Simple format\n\t\t\t\t\tmap[string]interface{}{\n\t\t\t\t\t\t\"server2\": []string{\"tool1\", \"tool2\"}, // Tools only\n\t\t\t\t\t},\n\t\t\t\t\tmap[string]interface{}{\n\t\t\t\t\t\t\"server3\": map[string]interface{}{\n\t\t\t\t\t\t\t\"resources\": []string{\"res1\", \"res2\"},\n\t\t\t\t\t\t\t\"tools\": []string{\"tool3\", \"tool4\"},\n\t\t\t\t\t\t},\n\t\t\t\t\t},\n\t\t\t\t},\n\t\t\t},\n\t\t}\n\n\t\terr = store.UpdateAssistant(id, updates)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to update MCP: %v\", err)\n\t\t}\n\n\t\t// Verify updates - MCP is in default fields\n\t\tretrieved, err := store.GetAssistant(id, nil)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to retrieve assistant: %v\", err)\n\t\t}\n\n\t\tif retrieved.MCP == nil || len(retrieved.MCP.Servers) != 3 {\n\t\t\tt.Fatalf(\"Expected 3 MCP servers, got %d\", len(retrieved.MCP.Servers))\n\t\t}\n\n\t\t// Verify server1 (simple format)\n\t\tif retrieved.MCP.Servers[0].ServerID != \"server1\" {\n\t\t\tt.Errorf(\"Expected server1, got '%s'\", retrieved.MCP.Servers[0].ServerID)\n\t\t}\n\t\tif len(retrieved.MCP.Servers[0].Tools) != 0 {\n\t\t\tt.Errorf(\"Expected no tools for server1, got %v\", retrieved.MCP.Servers[0].Tools)\n\t\t}\n\n\t\t// Verify server2 (tools only)\n\t\tif retrieved.MCP.Servers[1].ServerID != \"server2\" {\n\t\t\tt.Errorf(\"Expected server2, got '%s'\", retrieved.MCP.Servers[1].ServerID)\n\t\t}\n\t\tif len(retrieved.MCP.Servers[1].Tools) != 2 {\n\t\t\tt.Errorf(\"Expected 2 tools for server2, got %d\", len(retrieved.MCP.Servers[1].Tools))\n\t\t}\n\t\tif retrieved.MCP.Servers[1].Tools[0] != \"tool1\" {\n\t\t\tt.Errorf(\"Expected tool1, got '%s'\", retrieved.MCP.Servers[1].Tools[0])\n\t\t}\n\n\t\t// Verify server3 (full config)\n\t\tif retrieved.MCP.Servers[2].ServerID != \"server3\" {\n\t\t\tt.Errorf(\"Expected server3, got '%s'\", retrieved.MCP.Servers[2].ServerID)\n\t\t}\n\t\tif len(retrieved.MCP.Servers[2].Resources) != 2 {\n\t\t\tt.Errorf(\"Expected 2 resources for server3, got %d\", len(retrieved.MCP.Servers[2].Resources))\n\t\t}\n\t\tif len(retrieved.MCP.Servers[2].Tools) != 2 {\n\t\t\tt.Errorf(\"Expected 2 tools for server3, got %d\", len(retrieved.MCP.Servers[2].Tools))\n\t\t}\n\t\tif retrieved.MCP.Servers[2].Resources[0] != \"res1\" {\n\t\t\tt.Errorf(\"Expected res1, got '%s'\", retrieved.MCP.Servers[2].Resources[0])\n\t\t}\n\t\tif retrieved.MCP.Servers[2].Tools[0] != \"tool3\" {\n\t\t\tt.Errorf(\"Expected tool3, got '%s'\", retrieved.MCP.Servers[2].Tools[0])\n\t\t}\n\n\t\tt.Logf(\"Successfully verified MCP advanced configuration for assistant %s\", id)\n\t})\n\n\tt.Run(\"UpdateUses\", func(t *testing.T) {\n\t\t// Create assistant without uses\n\t\tassistant := &types.AssistantModel{\n\t\t\tName: \"Uses Update Test\",\n\t\t\tType: \"assistant\",\n\t\t\tConnector: \"openai\",\n\t\t\tShare: \"private\",\n\t\t}\n\n\t\tid, err := store.SaveAssistant(assistant)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to create assistant: %v\", err)\n\t\t}\n\n\t\t// Update with uses configuration\n\t\tupdates := map[string]interface{}{\n\t\t\t\"uses\": &context.Uses{\n\t\t\t\tVision: \"mcp:new-vision\",\n\t\t\t\tAudio: \"mcp:new-audio\",\n\t\t\t\tSearch: \"agent\",\n\t\t\t\tFetch: \"mcp:fetch-server\",\n\t\t\t},\n\t\t}\n\n\t\terr = store.UpdateAssistant(id, updates)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to update uses: %v\", err)\n\t\t}\n\n\t\t// Verify updates - uses is NOT in default fields\n\t\tretrieved, err := store.GetAssistant(id, types.AssistantFullFields)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to retrieve assistant: %v\", err)\n\t\t}\n\n\t\tif retrieved.Uses == nil {\n\t\t\tt.Fatal(\"Expected uses to be set\")\n\t\t}\n\n\t\tif retrieved.Uses.Vision != \"mcp:new-vision\" {\n\t\t\tt.Errorf(\"Expected vision 'mcp:new-vision', got '%s'\", retrieved.Uses.Vision)\n\t\t}\n\t\tif retrieved.Uses.Audio != \"mcp:new-audio\" {\n\t\t\tt.Errorf(\"Expected audio 'mcp:new-audio', got '%s'\", retrieved.Uses.Audio)\n\t\t}\n\t\tif retrieved.Uses.Search != \"agent\" {\n\t\t\tt.Errorf(\"Expected search 'agent', got '%s'\", retrieved.Uses.Search)\n\t\t}\n\t\tif retrieved.Uses.Fetch != \"mcp:fetch-server\" {\n\t\t\tt.Errorf(\"Expected fetch 'mcp:fetch-server', got '%s'\", retrieved.Uses.Fetch)\n\t\t}\n\n\t\t// Update to change uses\n\t\tupdates2 := map[string]interface{}{\n\t\t\t\"uses\": &context.Uses{\n\t\t\t\tVision: \"agent\",\n\t\t\t\tAudio: \"agent\",\n\t\t\t},\n\t\t}\n\n\t\terr = store.UpdateAssistant(id, updates2)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to update uses again: %v\", err)\n\t\t}\n\n\t\t// Verify second update - uses is NOT in default fields\n\t\tretrieved2, err := store.GetAssistant(id, types.AssistantFullFields)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to retrieve assistant: %v\", err)\n\t\t}\n\n\t\tif retrieved2.Uses.Vision != \"agent\" {\n\t\t\tt.Errorf(\"Expected vision 'agent', got '%s'\", retrieved2.Uses.Vision)\n\t\t}\n\t\tif retrieved2.Uses.Audio != \"agent\" {\n\t\t\tt.Errorf(\"Expected audio 'agent', got '%s'\", retrieved2.Uses.Audio)\n\t\t}\n\n\t\t// Update to remove uses (set to nil)\n\t\tupdates3 := map[string]interface{}{\n\t\t\t\"uses\": nil,\n\t\t}\n\n\t\terr = store.UpdateAssistant(id, updates3)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to set uses to nil: %v\", err)\n\t\t}\n\n\t\t// Verify uses is nil - uses is NOT in default fields\n\t\tretrieved3, err := store.GetAssistant(id, types.AssistantFullFields)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to retrieve assistant: %v\", err)\n\t\t}\n\n\t\tif retrieved3.Uses != nil {\n\t\t\tt.Errorf(\"Expected uses to be nil, got %+v\", retrieved3.Uses)\n\t\t}\n\t})\n\n\tt.Run(\"UpdateSearch\", func(t *testing.T) {\n\t\t// Create assistant without search\n\t\tassistant := &types.AssistantModel{\n\t\t\tName: \"Search Update Test\",\n\t\t\tType: \"assistant\",\n\t\t\tConnector: \"openai\",\n\t\t\tShare: \"private\",\n\t\t}\n\n\t\tid, err := store.SaveAssistant(assistant)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to create assistant: %v\", err)\n\t\t}\n\n\t\t// Update with search configuration\n\t\tupdates := map[string]interface{}{\n\t\t\t\"search\": &searchTypes.Config{\n\t\t\t\tWeb: &searchTypes.WebConfig{\n\t\t\t\t\tProvider: \"tavily\",\n\t\t\t\t\tMaxResults: 20,\n\t\t\t\t},\n\t\t\t\tKB: &searchTypes.KBConfig{\n\t\t\t\t\tCollections: []string{\"knowledge\"},\n\t\t\t\t\tThreshold: 0.75,\n\t\t\t\t},\n\t\t\t},\n\t\t}\n\n\t\terr = store.UpdateAssistant(id, updates)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to update search: %v\", err)\n\t\t}\n\n\t\t// Verify updates - search is NOT in default fields\n\t\tretrieved, err := store.GetAssistant(id, types.AssistantFullFields)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to retrieve assistant: %v\", err)\n\t\t}\n\n\t\tif retrieved.Search == nil {\n\t\t\tt.Fatal(\"Expected search to be set\")\n\t\t}\n\n\t\tif retrieved.Search.Web == nil {\n\t\t\tt.Fatal(\"Expected search.web to be set\")\n\t\t}\n\t\tif retrieved.Search.Web.Provider != \"tavily\" {\n\t\t\tt.Errorf(\"Expected web provider 'tavily', got '%s'\", retrieved.Search.Web.Provider)\n\t\t}\n\t\tif retrieved.Search.Web.MaxResults != 20 {\n\t\t\tt.Errorf(\"Expected web max_results 20, got %d\", retrieved.Search.Web.MaxResults)\n\t\t}\n\n\t\tif retrieved.Search.KB == nil {\n\t\t\tt.Fatal(\"Expected search.kb to be set\")\n\t\t}\n\t\tif len(retrieved.Search.KB.Collections) != 1 {\n\t\t\tt.Errorf(\"Expected 1 KB collection, got %d\", len(retrieved.Search.KB.Collections))\n\t\t}\n\t\tif retrieved.Search.KB.Threshold != 0.75 {\n\t\t\tt.Errorf(\"Expected KB threshold 0.75, got %f\", retrieved.Search.KB.Threshold)\n\t\t}\n\n\t\t// Update to change search configuration\n\t\tupdates2 := map[string]interface{}{\n\t\t\t\"search\": &searchTypes.Config{\n\t\t\t\tWeb: &searchTypes.WebConfig{\n\t\t\t\t\tProvider: \"serper\",\n\t\t\t\t\tMaxResults: 30,\n\t\t\t\t},\n\t\t\t\tCitation: &searchTypes.CitationConfig{\n\t\t\t\t\tFormat: \"#cite:{id}\",\n\t\t\t\t\tAutoInjectPrompt: false,\n\t\t\t\t},\n\t\t\t},\n\t\t}\n\n\t\terr = store.UpdateAssistant(id, updates2)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to update search again: %v\", err)\n\t\t}\n\n\t\t// Verify second update - search is NOT in default fields\n\t\tretrieved2, err := store.GetAssistant(id, types.AssistantFullFields)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to retrieve assistant: %v\", err)\n\t\t}\n\n\t\tif retrieved2.Search.Web.Provider != \"serper\" {\n\t\t\tt.Errorf(\"Expected web provider 'serper', got '%s'\", retrieved2.Search.Web.Provider)\n\t\t}\n\t\tif retrieved2.Search.Web.MaxResults != 30 {\n\t\t\tt.Errorf(\"Expected web max_results 30, got %d\", retrieved2.Search.Web.MaxResults)\n\t\t}\n\t\tif retrieved2.Search.Citation == nil {\n\t\t\tt.Fatal(\"Expected search.citation to be set\")\n\t\t}\n\t\tif retrieved2.Search.Citation.Format != \"#cite:{id}\" {\n\t\t\tt.Errorf(\"Expected citation format '#cite:{id}', got '%s'\", retrieved2.Search.Citation.Format)\n\t\t}\n\n\t\t// Update to remove search (set to nil)\n\t\tupdates3 := map[string]interface{}{\n\t\t\t\"search\": nil,\n\t\t}\n\n\t\terr = store.UpdateAssistant(id, updates3)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to set search to nil: %v\", err)\n\t\t}\n\n\t\t// Verify search is nil - search is NOT in default fields\n\t\tretrieved3, err := store.GetAssistant(id, types.AssistantFullFields)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to retrieve assistant: %v\", err)\n\t\t}\n\n\t\tif retrieved3.Search != nil {\n\t\t\tt.Errorf(\"Expected search to be nil, got %+v\", retrieved3.Search)\n\t\t}\n\t})\n\n\tt.Run(\"UpdatePermissionFields\", func(t *testing.T) {\n\t\t// Create assistant with permission fields\n\t\tassistant := &types.AssistantModel{\n\t\t\tName: \"Permission Test\",\n\t\t\tType: \"assistant\",\n\t\t\tConnector: \"openai\",\n\t\t\tShare: \"private\",\n\t\t\tYaoCreatedBy: \"user-1\",\n\t\t\tYaoTeamID: \"team-1\",\n\t\t}\n\n\t\tid, err := store.SaveAssistant(assistant)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to create assistant: %v\", err)\n\t\t}\n\n\t\t// Update permission fields\n\t\tupdates := map[string]interface{}{\n\t\t\t\"__yao_updated_by\": \"user-2\",\n\t\t\t\"__yao_tenant_id\": \"tenant-1\",\n\t\t}\n\n\t\terr = store.UpdateAssistant(id, updates)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to update permission fields: %v\", err)\n\t\t}\n\n\t\t// Verify updates - permission fields are in default fields\n\t\tretrieved, err := store.GetAssistant(id, nil)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to retrieve assistant: %v\", err)\n\t\t}\n\n\t\tif retrieved.YaoUpdatedBy != \"user-2\" {\n\t\t\tt.Errorf(\"Expected YaoUpdatedBy 'user-2', got '%s'\", retrieved.YaoUpdatedBy)\n\t\t}\n\t\tif retrieved.YaoTenantID != \"tenant-1\" {\n\t\t\tt.Errorf(\"Expected YaoTenantID 'tenant-1', got '%s'\", retrieved.YaoTenantID)\n\t\t}\n\t\t// Created by should remain unchanged\n\t\tif retrieved.YaoCreatedBy != \"user-1\" {\n\t\t\tt.Errorf(\"Expected YaoCreatedBy 'user-1', got '%s'\", retrieved.YaoCreatedBy)\n\t\t}\n\t})\n\n\tt.Run(\"UpdateWithEmptyStrings\", func(t *testing.T) {\n\t\t// Create assistant with values\n\t\tassistant := &types.AssistantModel{\n\t\t\tName: \"Empty String Test\",\n\t\t\tType: \"assistant\",\n\t\t\tConnector: \"openai\",\n\t\t\tAvatar: \"https://example.com/avatar.png\",\n\t\t\tDescription: \"Some description\",\n\t\t\tShare: \"private\",\n\t\t}\n\n\t\tid, err := store.SaveAssistant(assistant)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to create assistant: %v\", err)\n\t\t}\n\n\t\t// Update with empty strings (should become NULL)\n\t\tupdates := map[string]interface{}{\n\t\t\t\"avatar\": \"\",\n\t\t\t\"description\": \"\",\n\t\t}\n\n\t\terr = store.UpdateAssistant(id, updates)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to update with empty strings: %v\", err)\n\t\t}\n\n\t\t// Verify empty strings are stored as NULL - default fields include avatar, description\n\t\tretrieved, err := store.GetAssistant(id, nil)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to retrieve assistant: %v\", err)\n\t\t}\n\n\t\tif retrieved.Avatar != \"\" {\n\t\t\tt.Errorf(\"Expected empty avatar, got '%s'\", retrieved.Avatar)\n\t\t}\n\t\tif retrieved.Description != \"\" {\n\t\t\tt.Errorf(\"Expected empty description, got '%s'\", retrieved.Description)\n\t\t}\n\t\t// Name should remain unchanged\n\t\tif retrieved.Name != \"Empty String Test\" {\n\t\t\tt.Errorf(\"Expected name 'Empty String Test', got '%s'\", retrieved.Name)\n\t\t}\n\t})\n\n\tt.Run(\"UpdateNonExistentAssistant\", func(t *testing.T) {\n\t\tupdates := map[string]interface{}{\n\t\t\t\"name\": \"Updated Name\",\n\t\t}\n\n\t\terr := store.UpdateAssistant(\"nonexistent-id\", updates)\n\t\tif err == nil {\n\t\t\tt.Error(\"Expected error when updating non-existent assistant\")\n\t\t}\n\t\tif !strings.Contains(err.Error(), \"not found\") {\n\t\t\tt.Errorf(\"Expected 'not found' error, got: %v\", err)\n\t\t}\n\t})\n\n\tt.Run(\"UpdateWithEmptyID\", func(t *testing.T) {\n\t\tupdates := map[string]interface{}{\n\t\t\t\"name\": \"Updated Name\",\n\t\t}\n\n\t\terr := store.UpdateAssistant(\"\", updates)\n\t\tif err == nil {\n\t\t\tt.Error(\"Expected error when updating with empty ID\")\n\t\t}\n\t\tif !strings.Contains(err.Error(), \"required\") {\n\t\t\tt.Errorf(\"Expected 'required' error, got: %v\", err)\n\t\t}\n\t})\n\n\tt.Run(\"UpdateWithEmptyUpdates\", func(t *testing.T) {\n\t\t// Create assistant\n\t\tassistant := &types.AssistantModel{\n\t\t\tName: \"Empty Updates Test\",\n\t\t\tType: \"assistant\",\n\t\t\tConnector: \"openai\",\n\t\t\tShare: \"private\",\n\t\t}\n\n\t\tid, err := store.SaveAssistant(assistant)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to create assistant: %v\", err)\n\t\t}\n\n\t\t// Try to update with empty map\n\t\tupdates := map[string]interface{}{}\n\n\t\terr = store.UpdateAssistant(id, updates)\n\t\tif err == nil {\n\t\t\tt.Error(\"Expected error when updating with no fields\")\n\t\t}\n\t\tif !strings.Contains(err.Error(), \"no fields to update\") {\n\t\t\tt.Errorf(\"Expected 'no fields to update' error, got: %v\", err)\n\t\t}\n\t})\n\n\tt.Run(\"UpdateTimestampAutomaticallySet\", func(t *testing.T) {\n\t\t// Create assistant\n\t\tassistant := &types.AssistantModel{\n\t\t\tName: \"Timestamp Test\",\n\t\t\tType: \"assistant\",\n\t\t\tConnector: \"openai\",\n\t\t\tShare: \"private\",\n\t\t}\n\n\t\tid, err := store.SaveAssistant(assistant)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to create assistant: %v\", err)\n\t\t}\n\n\t\t// Get original updated_at - default fields include updated_at\n\t\toriginal, err := store.GetAssistant(id, nil)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to retrieve assistant: %v\", err)\n\t\t}\n\n\t\t// Wait a bit to ensure timestamp difference\n\t\ttime.Sleep(100 * time.Millisecond)\n\n\t\t// Update assistant\n\t\tupdates := map[string]interface{}{\n\t\t\t\"description\": \"Updated to test timestamp\",\n\t\t}\n\n\t\terr = store.UpdateAssistant(id, updates)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to update assistant: %v\", err)\n\t\t}\n\n\t\t// Get updated assistant - default fields include description, updated_at\n\t\tupdated, err := store.GetAssistant(id, nil)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to retrieve updated assistant: %v\", err)\n\t\t}\n\n\t\t// Verify description was updated (main test objective)\n\t\tif updated.Description != \"Updated to test timestamp\" {\n\t\t\tt.Errorf(\"Expected description 'Updated to test timestamp', got '%s'\", updated.Description)\n\t\t}\n\n\t\t// Only check timestamp if both are set (some stores may not return timestamps)\n\t\tif original.UpdatedAt > 0 && updated.UpdatedAt > 0 {\n\t\t\tif updated.UpdatedAt <= original.UpdatedAt {\n\t\t\t\tt.Errorf(\"Expected updated_at to increase, original=%d, updated=%d\", original.UpdatedAt, updated.UpdatedAt)\n\t\t\t}\n\t\t} else {\n\t\t\tt.Logf(\"Skipping timestamp comparison (original=%d, updated=%d)\", original.UpdatedAt, updated.UpdatedAt)\n\t\t}\n\t})\n\n\tt.Run(\"UpdateSkipsSystemFields\", func(t *testing.T) {\n\t\t// Create assistant\n\t\tassistant := &types.AssistantModel{\n\t\t\tName: \"System Fields Test\",\n\t\t\tType: \"assistant\",\n\t\t\tConnector: \"openai\",\n\t\t\tShare: \"private\",\n\t\t}\n\n\t\tid, err := store.SaveAssistant(assistant)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to create assistant: %v\", err)\n\t\t}\n\n\t\t// Get original - default fields\n\t\toriginal, err := store.GetAssistant(id, nil)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to retrieve assistant: %v\", err)\n\t\t}\n\n\t\t// Try to update system fields (should be ignored)\n\t\tupdates := map[string]interface{}{\n\t\t\t\"assistant_id\": \"new-id-123\", // Should be ignored\n\t\t\t\"created_at\": int64(123456789), // Should be ignored\n\t\t\t\"name\": \"Valid Update\", // Should be applied\n\t\t}\n\n\t\terr = store.UpdateAssistant(id, updates)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to update assistant: %v\", err)\n\t\t}\n\n\t\t// Verify system fields unchanged, but name updated - default fields\n\t\tretrieved, err := store.GetAssistant(id, nil)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to retrieve assistant: %v\", err)\n\t\t}\n\n\t\tif retrieved.ID != id {\n\t\t\tt.Errorf(\"Expected ID to remain %s, got %s\", id, retrieved.ID)\n\t\t}\n\t\tif retrieved.CreatedAt != original.CreatedAt {\n\t\t\tt.Errorf(\"Expected created_at to remain unchanged\")\n\t\t}\n\t\tif retrieved.Name != \"Valid Update\" {\n\t\t\tt.Errorf(\"Expected name 'Valid Update', got '%s'\", retrieved.Name)\n\t\t}\n\t})\n}\n\n// TestAssistantCompleteWorkflow tests a complete workflow\nfunc TestAssistantCompleteWorkflow(t *testing.T) {\n\ttest.Prepare(t, config.Conf)\n\tdefer test.Clean()\n\n\tstore, err := xun.NewXun(types.Setting{\n\t\tConnector: \"default\",\n\t})\n\tif err != nil {\n\t\tt.Fatalf(\"Failed to create store: %v\", err)\n\t}\n\n\tt.Run(\"CompleteWorkflow\", func(t *testing.T) {\n\t\t// Step 1: Create multiple assistants\n\t\tassistantIDs := []string{}\n\t\tfor i := 0; i < 3; i++ {\n\t\t\tassistant := &types.AssistantModel{\n\t\t\t\tName: fmt.Sprintf(\"Workflow Assistant %d\", i),\n\t\t\t\tType: \"assistant\",\n\t\t\t\tConnector: \"openai\",\n\t\t\t\tDescription: fmt.Sprintf(\"Workflow test assistant %d\", i),\n\t\t\t\tTags: []string{\"workflow\", fmt.Sprintf(\"test-%d\", i)},\n\t\t\t\tSort: i * 100,\n\t\t\t\tShare: \"private\",\n\t\t\t}\n\n\t\t\tid, err := store.SaveAssistant(assistant)\n\t\t\tif err != nil {\n\t\t\t\tt.Fatalf(\"Failed to create assistant %d: %v\", i, err)\n\t\t\t}\n\t\t\tassistantIDs = append(assistantIDs, id)\n\t\t}\n\n\t\tt.Logf(\"Created %d assistants\", len(assistantIDs))\n\n\t\t// Step 2: Retrieve all assistants\n\t\tresponse, err := store.GetAssistants(types.AssistantFilter{\n\t\t\tTags: []string{\"workflow\"},\n\t\t\tPage: 1,\n\t\t\tPageSize: 20,\n\t\t})\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to get assistants: %v\", err)\n\t\t}\n\n\t\tif len(response.Data) < 3 {\n\t\t\tt.Errorf(\"Expected at least 3 assistants, got %d\", len(response.Data))\n\t\t}\n\n\t\t// Step 3: Update one assistant - need full fields for tags\n\t\tupdatedID := assistantIDs[1]\n\t\tupdatedAssistant, err := store.GetAssistant(updatedID, types.AssistantFullFields)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to get assistant for update: %v\", err)\n\t\t}\n\n\t\tupdatedAssistant.Description = \"Updated workflow description\"\n\t\tupdatedAssistant.Tags = append(updatedAssistant.Tags, \"updated\")\n\n\t\t_, err = store.SaveAssistant(updatedAssistant)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to update assistant: %v\", err)\n\t\t}\n\n\t\t// Verify update - default fields include description\n\t\tverifyAssistant, err := store.GetAssistant(updatedID, nil)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to verify update: %v\", err)\n\t\t}\n\n\t\tif verifyAssistant.Description != \"Updated workflow description\" {\n\t\t\tt.Errorf(\"Update not applied correctly\")\n\t\t}\n\n\t\t// Step 4: Delete one assistant\n\t\terr = store.DeleteAssistant(assistantIDs[0])\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to delete assistant: %v\", err)\n\t\t}\n\n\t\t// Verify deletion\n\t\t_, err = store.GetAssistant(assistantIDs[0], nil)\n\t\tif err == nil {\n\t\t\tt.Error(\"Expected error when getting deleted assistant\")\n\t\t}\n\n\t\t// Step 5: Get tags\n\t\ttags, err := store.GetAssistantTags(types.AssistantFilter{})\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to get tags: %v\", err)\n\t\t}\n\n\t\t// Should find \"workflow\" tag\n\t\tfound := false\n\t\tfor _, tag := range tags {\n\t\t\tif tag.Value == \"workflow\" {\n\t\t\t\tfound = true\n\t\t\t\tbreak\n\t\t\t}\n\t\t}\n\n\t\tif !found {\n\t\t\tt.Error(\"Expected to find 'workflow' tag\")\n\t\t}\n\n\t\t// Step 6: Bulk delete remaining assistants\n\t\tcount, err := store.DeleteAssistants(types.AssistantFilter{\n\t\t\tTags: []string{\"workflow\"},\n\t\t})\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to bulk delete: %v\", err)\n\t\t}\n\n\t\tt.Logf(\"Bulk deleted %d assistants\", count)\n\n\t\t// Verify bulk deletion\n\t\tfinalResponse, err := store.GetAssistants(types.AssistantFilter{\n\t\t\tTags: []string{\"workflow\"},\n\t\t\tPage: 1,\n\t\t\tPageSize: 20,\n\t\t})\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to verify bulk deletion: %v\", err)\n\t\t}\n\n\t\tif len(finalResponse.Data) > 0 {\n\t\t\tt.Logf(\"Warning: Still found %d assistants after bulk delete\", len(finalResponse.Data))\n\t\t}\n\t})\n}\n" }, { "path": "agent/store/xun/chat.go", "content": "package xun\n\nimport (\n\t\"fmt\"\n\t\"math\"\n\t\"time\"\n\n\t\"github.com/google/uuid\"\n\tjsoniter \"github.com/json-iterator/go\"\n\t\"github.com/yaoapp/yao/agent/store/types\"\n)\n\n// =============================================================================\n// Chat Management\n// =============================================================================\n\n// CreateChat creates a new chat session\nfunc (store *Xun) CreateChat(chat *types.Chat) error {\n\tif chat == nil {\n\t\treturn fmt.Errorf(\"chat cannot be nil\")\n\t}\n\n\t// Validate required fields\n\tif chat.AssistantID == \"\" {\n\t\treturn fmt.Errorf(\"assistant_id is required\")\n\t}\n\n\t// Generate chat_id if not provided\n\tif chat.ChatID == \"\" {\n\t\tchat.ChatID = uuid.New().String()\n\t}\n\n\t// Check if chat already exists\n\texists, err := store.newQueryChat().\n\t\tWhere(\"chat_id\", chat.ChatID).\n\t\tExists()\n\tif err != nil {\n\t\treturn err\n\t}\n\tif exists {\n\t\treturn fmt.Errorf(\"chat %s already exists\", chat.ChatID)\n\t}\n\n\t// Set defaults\n\tif chat.Status == \"\" {\n\t\tchat.Status = \"active\"\n\t}\n\tif chat.Share == \"\" {\n\t\tchat.Share = \"private\"\n\t}\n\n\t// Prepare data\n\tdata := map[string]interface{}{\n\t\t\"chat_id\": chat.ChatID,\n\t\t\"assistant_id\": chat.AssistantID,\n\t\t\"status\": chat.Status,\n\t\t\"public\": chat.Public,\n\t\t\"share\": chat.Share,\n\t\t\"sort\": chat.Sort,\n\t\t\"created_at\": time.Now(),\n\t\t\"updated_at\": time.Now(),\n\t}\n\n\t// Handle last_mode (nullable)\n\tif chat.LastMode != \"\" {\n\t\tdata[\"last_mode\"] = chat.LastMode\n\t}\n\n\t// Handle nullable fields\n\tif chat.Title != \"\" {\n\t\tdata[\"title\"] = chat.Title\n\t}\n\tif chat.LastConnector != \"\" {\n\t\tdata[\"last_connector\"] = chat.LastConnector\n\t}\n\tif chat.LastMode != \"\" {\n\t\tdata[\"last_mode\"] = chat.LastMode\n\t}\n\tif chat.LastMessageAt != nil {\n\t\tdata[\"last_message_at\"] = *chat.LastMessageAt\n\t}\n\tif chat.Metadata != nil {\n\t\tmetadataJSON, err := jsoniter.MarshalToString(chat.Metadata)\n\t\tif err != nil {\n\t\t\treturn fmt.Errorf(\"failed to marshal metadata: %w\", err)\n\t\t}\n\t\tdata[\"metadata\"] = metadataJSON\n\t}\n\n\t// Handle permission fields (Yao framework permission: true)\n\tif chat.CreatedBy != \"\" {\n\t\tdata[\"__yao_created_by\"] = chat.CreatedBy\n\t}\n\tif chat.UpdatedBy != \"\" {\n\t\tdata[\"__yao_updated_by\"] = chat.UpdatedBy\n\t}\n\tif chat.TeamID != \"\" {\n\t\tdata[\"__yao_team_id\"] = chat.TeamID\n\t}\n\tif chat.TenantID != \"\" {\n\t\tdata[\"__yao_tenant_id\"] = chat.TenantID\n\t}\n\n\t// Insert\n\treturn store.newQueryChat().Insert(data)\n}\n\n// GetChat retrieves a single chat by ID\nfunc (store *Xun) GetChat(chatID string) (*types.Chat, error) {\n\tif chatID == \"\" {\n\t\treturn nil, fmt.Errorf(\"chat_id is required\")\n\t}\n\n\trow, err := store.newQueryChat().\n\t\tWhere(\"chat_id\", chatID).\n\t\tWhereNull(\"deleted_at\").\n\t\tFirst()\n\tif err != nil {\n\t\treturn nil, err\n\t}\n\n\tif row == nil {\n\t\treturn nil, fmt.Errorf(\"chat %s not found\", chatID)\n\t}\n\n\tdata := row.ToMap()\n\tif len(data) == 0 || data[\"chat_id\"] == nil {\n\t\treturn nil, fmt.Errorf(\"chat %s not found\", chatID)\n\t}\n\n\treturn store.rowToChat(data)\n}\n\n// UpdateChat updates chat fields\nfunc (store *Xun) UpdateChat(chatID string, updates map[string]interface{}) error {\n\tif chatID == \"\" {\n\t\treturn fmt.Errorf(\"chat_id is required\")\n\t}\n\tif len(updates) == 0 {\n\t\treturn fmt.Errorf(\"no fields to update\")\n\t}\n\n\t// Check if chat exists\n\texists, err := store.newQueryChat().\n\t\tWhere(\"chat_id\", chatID).\n\t\tWhereNull(\"deleted_at\").\n\t\tExists()\n\tif err != nil {\n\t\treturn err\n\t}\n\tif !exists {\n\t\treturn fmt.Errorf(\"chat %s not found\", chatID)\n\t}\n\n\t// Prepare update data\n\tdata := make(map[string]interface{})\n\n\t// Process each update field\n\tfor key, value := range updates {\n\t\t// Skip system fields\n\t\tif key == \"chat_id\" || key == \"created_at\" {\n\t\t\tcontinue\n\t\t}\n\n\t\t// Handle metadata specially\n\t\tif key == \"metadata\" {\n\t\t\tif value != nil {\n\t\t\t\tmetadataJSON, err := jsoniter.MarshalToString(value)\n\t\t\t\tif err != nil {\n\t\t\t\t\treturn fmt.Errorf(\"failed to marshal metadata: %w\", err)\n\t\t\t\t}\n\t\t\t\tdata[\"metadata\"] = metadataJSON\n\t\t\t} else {\n\t\t\t\tdata[\"metadata\"] = nil\n\t\t\t}\n\t\t\tcontinue\n\t\t}\n\n\t\tdata[key] = value\n\t}\n\n\t// Always update updated_at\n\tdata[\"updated_at\"] = time.Now()\n\n\tif len(data) == 0 {\n\t\treturn fmt.Errorf(\"no valid fields to update\")\n\t}\n\n\t_, err = store.newQueryChat().\n\t\tWhere(\"chat_id\", chatID).\n\t\tUpdate(data)\n\n\treturn err\n}\n\n// DeleteChat deletes a chat and its associated messages (soft delete)\nfunc (store *Xun) DeleteChat(chatID string) error {\n\tif chatID == \"\" {\n\t\treturn fmt.Errorf(\"chat_id is required\")\n\t}\n\n\t// Check if chat exists\n\texists, err := store.newQueryChat().\n\t\tWhere(\"chat_id\", chatID).\n\t\tWhereNull(\"deleted_at\").\n\t\tExists()\n\tif err != nil {\n\t\treturn err\n\t}\n\tif !exists {\n\t\treturn fmt.Errorf(\"chat %s not found\", chatID)\n\t}\n\n\t// Soft delete the chat\n\t_, err = store.newQueryChat().\n\t\tWhere(\"chat_id\", chatID).\n\t\tUpdate(map[string]interface{}{\n\t\t\t\"deleted_at\": time.Now(),\n\t\t\t\"updated_at\": time.Now(),\n\t\t})\n\n\treturn err\n}\n\n// ListChats retrieves a paginated list of chats with optional grouping\nfunc (store *Xun) ListChats(filter types.ChatFilter) (*types.ChatList, error) {\n\t// Set defaults\n\tif filter.Page <= 0 {\n\t\tfilter.Page = 1\n\t}\n\tif filter.PageSize <= 0 {\n\t\tfilter.PageSize = 20\n\t}\n\tif filter.OrderBy == \"\" {\n\t\tfilter.OrderBy = \"last_message_at\"\n\t}\n\tif filter.Order == \"\" {\n\t\tfilter.Order = \"desc\"\n\t}\n\tif filter.TimeField == \"\" {\n\t\tfilter.TimeField = \"last_message_at\"\n\t}\n\n\t// Build base query\n\tqb := store.newQueryChat().WhereNull(\"deleted_at\")\n\n\t// Apply permission filters (UserID and TeamID)\n\tif filter.UserID != \"\" {\n\t\tqb.Where(\"__yao_created_by\", filter.UserID)\n\t}\n\tif filter.TeamID != \"\" {\n\t\tqb.Where(\"__yao_team_id\", filter.TeamID)\n\t}\n\n\t// Apply business filters\n\tif filter.AssistantID != \"\" {\n\t\tqb.Where(\"assistant_id\", filter.AssistantID)\n\t}\n\tif filter.Status != \"\" {\n\t\tqb.Where(\"status\", filter.Status)\n\t}\n\tif filter.Keywords != \"\" {\n\t\tqb.Where(\"title\", \"like\", fmt.Sprintf(\"%%%s%%\", filter.Keywords))\n\t}\n\tif filter.ChatIDPrefix != \"\" {\n\t\tqb.Where(\"chat_id\", \"like\", filter.ChatIDPrefix+\"%\")\n\t}\n\n\t// Apply time range filter\n\tif filter.StartTime != nil {\n\t\tqb.Where(filter.TimeField, \">=\", *filter.StartTime)\n\t}\n\tif filter.EndTime != nil {\n\t\tqb.Where(filter.TimeField, \"<=\", *filter.EndTime)\n\t}\n\n\t// Apply custom query filter (for advanced permission filtering)\n\t// This allows flexible combinations like: (created_by = user OR team_id = team)\n\tif filter.QueryFilter != nil {\n\t\tqb.Where(filter.QueryFilter)\n\t}\n\n\t// Get total count\n\ttotal, err := qb.Clone().Count()\n\tif err != nil {\n\t\treturn nil, err\n\t}\n\n\t// Calculate pagination\n\tpageCount := int(math.Ceil(float64(total) / float64(filter.PageSize)))\n\tif pageCount < 1 {\n\t\tpageCount = 1\n\t}\n\toffset := (filter.Page - 1) * filter.PageSize\n\n\t// Get paginated results\n\trows, err := qb.OrderBy(filter.OrderBy, filter.Order).\n\t\tOffset(offset).\n\t\tLimit(filter.PageSize).\n\t\tGet()\n\tif err != nil {\n\t\treturn nil, err\n\t}\n\n\t// Convert rows to Chat objects\n\tchats := make([]*types.Chat, 0, len(rows))\n\tfor _, row := range rows {\n\t\tdata := row.ToMap()\n\t\tif data == nil || data[\"chat_id\"] == nil {\n\t\t\tcontinue\n\t\t}\n\n\t\tchat, err := store.rowToChat(data)\n\t\tif err != nil {\n\t\t\tcontinue\n\t\t}\n\t\tchats = append(chats, chat)\n\t}\n\n\tresult := &types.ChatList{\n\t\tData: chats,\n\t\tPage: filter.Page,\n\t\tPageSize: filter.PageSize,\n\t\tPageCount: pageCount,\n\t\tTotal: int(total),\n\t}\n\n\t// Apply time-based grouping if requested\n\tif filter.GroupBy == \"time\" {\n\t\tresult.Groups = store.groupChatsByTime(chats)\n\t}\n\n\treturn result, nil\n}\n\n// =============================================================================\n// Helper Functions\n// =============================================================================\n\n// rowToChat converts a database row to a Chat struct\nfunc (store *Xun) rowToChat(data map[string]interface{}) (*types.Chat, error) {\n\tchat := &types.Chat{\n\t\tChatID: getString(data, \"chat_id\"),\n\t\tTitle: getString(data, \"title\"),\n\t\tAssistantID: getString(data, \"assistant_id\"),\n\t\tLastConnector: getString(data, \"last_connector\"),\n\t\tLastMode: getString(data, \"last_mode\"),\n\t\tStatus: getString(data, \"status\"),\n\t\tPublic: getBool(data, \"public\"),\n\t\tShare: getString(data, \"share\"),\n\t\tSort: getInt(data, \"sort\"),\n\t}\n\n\t// Handle timestamps\n\tif createdAt := getTime(data, \"created_at\"); createdAt != nil {\n\t\tchat.CreatedAt = *createdAt\n\t}\n\tif updatedAt := getTime(data, \"updated_at\"); updatedAt != nil {\n\t\tchat.UpdatedAt = *updatedAt\n\t}\n\tif lastMsgAt := getTime(data, \"last_message_at\"); lastMsgAt != nil {\n\t\tchat.LastMessageAt = lastMsgAt\n\t}\n\n\t// Handle metadata\n\tif metadata := data[\"metadata\"]; metadata != nil {\n\t\tif metaStr, ok := metadata.(string); ok && metaStr != \"\" {\n\t\t\tvar meta map[string]interface{}\n\t\t\tif err := jsoniter.UnmarshalFromString(metaStr, &meta); err == nil {\n\t\t\t\tchat.Metadata = meta\n\t\t\t}\n\t\t} else if metaMap, ok := metadata.(map[string]interface{}); ok {\n\t\t\tchat.Metadata = metaMap\n\t\t}\n\t}\n\n\t// Handle permission fields\n\tchat.CreatedBy = getString(data, \"__yao_created_by\")\n\tchat.UpdatedBy = getString(data, \"__yao_updated_by\")\n\tchat.TeamID = getString(data, \"__yao_team_id\")\n\tchat.TenantID = getString(data, \"__yao_tenant_id\")\n\n\treturn chat, nil\n}\n\n// groupChatsByTime groups chats by time periods\nfunc (store *Xun) groupChatsByTime(chats []*types.Chat) []*types.ChatGroup {\n\tnow := time.Now()\n\ttoday := time.Date(now.Year(), now.Month(), now.Day(), 0, 0, 0, 0, now.Location())\n\tyesterday := today.AddDate(0, 0, -1)\n\tthisWeekStart := today.AddDate(0, 0, -int(today.Weekday()))\n\tthisMonthStart := time.Date(now.Year(), now.Month(), 1, 0, 0, 0, 0, now.Location())\n\n\tgroups := map[string]*types.ChatGroup{\n\t\t\"today\": {Key: \"today\", Label: \"Today\", Chats: []*types.Chat{}},\n\t\t\"yesterday\": {Key: \"yesterday\", Label: \"Yesterday\", Chats: []*types.Chat{}},\n\t\t\"this_week\": {Key: \"this_week\", Label: \"This Week\", Chats: []*types.Chat{}},\n\t\t\"this_month\": {Key: \"this_month\", Label: \"This Month\", Chats: []*types.Chat{}},\n\t\t\"earlier\": {Key: \"earlier\", Label: \"Earlier\", Chats: []*types.Chat{}},\n\t}\n\n\tfor _, chat := range chats {\n\t\t// Use last_message_at if available, otherwise created_at\n\t\tvar chatTime time.Time\n\t\tif chat.LastMessageAt != nil {\n\t\t\tchatTime = *chat.LastMessageAt\n\t\t} else {\n\t\t\tchatTime = chat.CreatedAt\n\t\t}\n\n\t\tchatDate := time.Date(chatTime.Year(), chatTime.Month(), chatTime.Day(), 0, 0, 0, 0, chatTime.Location())\n\n\t\tswitch {\n\t\tcase chatDate.Equal(today) || chatDate.After(today):\n\t\t\tgroups[\"today\"].Chats = append(groups[\"today\"].Chats, chat)\n\t\tcase chatDate.Equal(yesterday):\n\t\t\tgroups[\"yesterday\"].Chats = append(groups[\"yesterday\"].Chats, chat)\n\t\tcase chatDate.After(thisWeekStart) || chatDate.Equal(thisWeekStart):\n\t\t\tgroups[\"this_week\"].Chats = append(groups[\"this_week\"].Chats, chat)\n\t\tcase chatDate.After(thisMonthStart) || chatDate.Equal(thisMonthStart):\n\t\t\tgroups[\"this_month\"].Chats = append(groups[\"this_month\"].Chats, chat)\n\t\tdefault:\n\t\t\tgroups[\"earlier\"].Chats = append(groups[\"earlier\"].Chats, chat)\n\t\t}\n\t}\n\n\t// Update counts and filter empty groups\n\tresult := make([]*types.ChatGroup, 0)\n\tfor _, key := range []string{\"today\", \"yesterday\", \"this_week\", \"this_month\", \"earlier\"} {\n\t\tgroup := groups[key]\n\t\tgroup.Count = len(group.Chats)\n\t\tif group.Count > 0 {\n\t\t\tresult = append(result, group)\n\t\t}\n\t}\n\n\treturn result\n}\n\n// getTime helper function to convert database value to time.Time pointer\nfunc getTime(data map[string]interface{}, key string) *time.Time {\n\tif v := data[key]; v != nil {\n\t\tswitch t := v.(type) {\n\t\tcase time.Time:\n\t\t\treturn &t\n\t\tcase *time.Time:\n\t\t\treturn t\n\t\tcase string:\n\t\t\t// Try parsing various formats\n\t\t\tformats := []string{\n\t\t\t\ttime.RFC3339,\n\t\t\t\t\"2006-01-02 15:04:05\",\n\t\t\t\t\"2006-01-02 15:04:05.999999-07:00\",\n\t\t\t\t\"2006-01-02T15:04:05Z\",\n\t\t\t}\n\t\t\tfor _, format := range formats {\n\t\t\t\tif parsed, err := time.Parse(format, t); err == nil {\n\t\t\t\t\treturn &parsed\n\t\t\t\t}\n\t\t\t}\n\t\t}\n\t}\n\treturn nil\n}\n\n// UpdateChatLastMessageAt updates the last_message_at timestamp for a chat\nfunc (store *Xun) UpdateChatLastMessageAt(chatID string, timestamp time.Time) error {\n\tif chatID == \"\" {\n\t\treturn fmt.Errorf(\"chat_id is required\")\n\t}\n\n\t_, err := store.newQueryChat().\n\t\tWhere(\"chat_id\", chatID).\n\t\tUpdate(map[string]interface{}{\n\t\t\t\"last_message_at\": timestamp,\n\t\t\t\"updated_at\": time.Now(),\n\t\t})\n\n\treturn err\n}\n" }, { "path": "agent/store/xun/chat_test.go", "content": "package xun_test\n\nimport (\n\t\"fmt\"\n\t\"testing\"\n\t\"time\"\n\n\tgoumodel \"github.com/yaoapp/gou/model\"\n\t\"github.com/yaoapp/xun/dbal/query\"\n\t\"github.com/yaoapp/yao/agent/store/types\"\n\t\"github.com/yaoapp/yao/agent/store/xun\"\n\t\"github.com/yaoapp/yao/config\"\n\t\"github.com/yaoapp/yao/test\"\n)\n\n// TestCreateChat tests creating chat sessions\nfunc TestCreateChat(t *testing.T) {\n\ttest.Prepare(t, config.Conf)\n\tdefer test.Clean()\n\n\tstore, err := xun.NewXun(types.Setting{\n\t\tConnector: \"default\",\n\t})\n\tif err != nil {\n\t\tt.Fatalf(\"Failed to create store: %v\", err)\n\t}\n\n\tt.Run(\"CreateNewChat\", func(t *testing.T) {\n\t\tchat := &types.Chat{\n\t\t\tAssistantID: \"test_assistant\",\n\t\t\tTitle: \"Test Chat\",\n\t\t\tStatus: \"active\",\n\t\t\tShare: \"private\",\n\t\t}\n\n\t\terr := store.CreateChat(chat)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to create chat: %v\", err)\n\t\t}\n\n\t\tif chat.ChatID == \"\" {\n\t\t\tt.Error(\"Expected chat_id to be generated\")\n\t\t}\n\n\t\tt.Logf(\"Created chat with ID: %s\", chat.ChatID)\n\n\t\t// Clean up\n\t\t_ = store.DeleteChat(chat.ChatID)\n\t})\n\n\tt.Run(\"CreateChatWithAllFields\", func(t *testing.T) {\n\t\tnow := time.Now()\n\t\tchat := &types.Chat{\n\t\t\tAssistantID: \"test_assistant\",\n\t\t\tLastConnector: \"openai\",\n\t\t\tTitle: \"Full Chat\",\n\t\t\tLastMode: \"task\",\n\t\t\tStatus: \"active\",\n\t\t\tPublic: true,\n\t\t\tShare: \"team\",\n\t\t\tSort: 100,\n\t\t\tLastMessageAt: &now,\n\t\t\tMetadata: map[string]interface{}{\n\t\t\t\t\"source\": \"test\",\n\t\t\t\t\"tags\": []string{\"test\", \"chat\"},\n\t\t\t},\n\t\t}\n\n\t\terr := store.CreateChat(chat)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to create chat: %v\", err)\n\t\t}\n\n\t\t// Retrieve and verify\n\t\tretrieved, err := store.GetChat(chat.ChatID)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to retrieve chat: %v\", err)\n\t\t}\n\n\t\tif retrieved.Title != \"Full Chat\" {\n\t\t\tt.Errorf(\"Expected title 'Full Chat', got '%s'\", retrieved.Title)\n\t\t}\n\t\tif retrieved.LastConnector != \"openai\" {\n\t\t\tt.Errorf(\"Expected last_connector 'openai', got '%s'\", retrieved.LastConnector)\n\t\t}\n\t\tif retrieved.LastMode != \"task\" {\n\t\t\tt.Errorf(\"Expected last_mode 'task', got '%s'\", retrieved.LastMode)\n\t\t}\n\t\tif !retrieved.Public {\n\t\t\tt.Error(\"Expected public to be true\")\n\t\t}\n\t\tif retrieved.Share != \"team\" {\n\t\t\tt.Errorf(\"Expected share 'team', got '%s'\", retrieved.Share)\n\t\t}\n\t\tif retrieved.Sort != 100 {\n\t\t\tt.Errorf(\"Expected sort 100, got %d\", retrieved.Sort)\n\t\t}\n\t\tif retrieved.Metadata == nil {\n\t\t\tt.Error(\"Expected metadata to be set\")\n\t\t}\n\n\t\t// Clean up\n\t\t_ = store.DeleteChat(chat.ChatID)\n\t})\n\n\tt.Run(\"CreateChatWithCustomID\", func(t *testing.T) {\n\t\tcustomID := fmt.Sprintf(\"custom_chat_%d\", time.Now().UnixNano())\n\t\tchat := &types.Chat{\n\t\t\tChatID: customID,\n\t\t\tAssistantID: \"test_assistant\",\n\t\t}\n\n\t\terr := store.CreateChat(chat)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to create chat: %v\", err)\n\t\t}\n\n\t\tif chat.ChatID != customID {\n\t\t\tt.Errorf(\"Expected chat_id '%s', got '%s'\", customID, chat.ChatID)\n\t\t}\n\n\t\t// Clean up\n\t\t_ = store.DeleteChat(chat.ChatID)\n\t})\n\n\tt.Run(\"CreateDuplicateChatFails\", func(t *testing.T) {\n\t\tchat := &types.Chat{\n\t\t\tAssistantID: \"test_assistant\",\n\t\t}\n\n\t\terr := store.CreateChat(chat)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to create first chat: %v\", err)\n\t\t}\n\n\t\t// Try to create with same ID\n\t\tduplicateChat := &types.Chat{\n\t\t\tChatID: chat.ChatID,\n\t\t\tAssistantID: \"test_assistant\",\n\t\t}\n\n\t\terr = store.CreateChat(duplicateChat)\n\t\tif err == nil {\n\t\t\tt.Error(\"Expected error when creating duplicate chat\")\n\t\t}\n\n\t\t// Clean up\n\t\t_ = store.DeleteChat(chat.ChatID)\n\t})\n\n\tt.Run(\"CreateChatWithoutAssistantIDFails\", func(t *testing.T) {\n\t\tchat := &types.Chat{\n\t\t\tTitle: \"No Assistant\",\n\t\t}\n\n\t\terr := store.CreateChat(chat)\n\t\tif err == nil {\n\t\t\tt.Error(\"Expected error when creating chat without assistant_id\")\n\t\t}\n\t})\n\n\tt.Run(\"CreateNilChatFails\", func(t *testing.T) {\n\t\terr := store.CreateChat(nil)\n\t\tif err == nil {\n\t\t\tt.Error(\"Expected error when creating nil chat\")\n\t\t}\n\t})\n\n\tt.Run(\"CreateChatWithDefaults\", func(t *testing.T) {\n\t\tchat := &types.Chat{\n\t\t\tAssistantID: \"test_assistant\",\n\t\t}\n\n\t\terr := store.CreateChat(chat)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to create chat: %v\", err)\n\t\t}\n\n\t\t// Retrieve and verify defaults\n\t\tretrieved, err := store.GetChat(chat.ChatID)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to retrieve chat: %v\", err)\n\t\t}\n\n\t\t// last_mode is nullable, so it should be empty by default\n\t\tif retrieved.LastMode != \"\" {\n\t\t\tt.Errorf(\"Expected default last_mode to be empty, got '%s'\", retrieved.LastMode)\n\t\t}\n\t\tif retrieved.Status != \"active\" {\n\t\t\tt.Errorf(\"Expected default status 'active', got '%s'\", retrieved.Status)\n\t\t}\n\t\tif retrieved.Share != \"private\" {\n\t\t\tt.Errorf(\"Expected default share 'private', got '%s'\", retrieved.Share)\n\t\t}\n\n\t\t// Clean up\n\t\t_ = store.DeleteChat(chat.ChatID)\n\t})\n}\n\n// TestGetChat tests retrieving chat sessions\nfunc TestGetChat(t *testing.T) {\n\ttest.Prepare(t, config.Conf)\n\tdefer test.Clean()\n\n\tstore, err := xun.NewXun(types.Setting{\n\t\tConnector: \"default\",\n\t})\n\tif err != nil {\n\t\tt.Fatalf(\"Failed to create store: %v\", err)\n\t}\n\n\tt.Run(\"GetExistingChat\", func(t *testing.T) {\n\t\t// Create chat first\n\t\tchat := &types.Chat{\n\t\t\tAssistantID: \"test_assistant\",\n\t\t\tTitle: \"Get Test Chat\",\n\t\t}\n\t\terr := store.CreateChat(chat)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to create chat: %v\", err)\n\t\t}\n\n\t\t// Get it\n\t\tretrieved, err := store.GetChat(chat.ChatID)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to get chat: %v\", err)\n\t\t}\n\n\t\tif retrieved.ChatID != chat.ChatID {\n\t\t\tt.Errorf(\"Expected chat_id '%s', got '%s'\", chat.ChatID, retrieved.ChatID)\n\t\t}\n\t\tif retrieved.Title != \"Get Test Chat\" {\n\t\t\tt.Errorf(\"Expected title 'Get Test Chat', got '%s'\", retrieved.Title)\n\t\t}\n\n\t\t// Clean up\n\t\t_ = store.DeleteChat(chat.ChatID)\n\t})\n\n\tt.Run(\"GetNonExistentChat\", func(t *testing.T) {\n\t\t_, err := store.GetChat(\"nonexistent_chat_id\")\n\t\tif err == nil {\n\t\t\tt.Error(\"Expected error when getting non-existent chat\")\n\t\t}\n\t})\n\n\tt.Run(\"GetChatWithEmptyID\", func(t *testing.T) {\n\t\t_, err := store.GetChat(\"\")\n\t\tif err == nil {\n\t\t\tt.Error(\"Expected error when getting chat with empty ID\")\n\t\t}\n\t})\n\n\tt.Run(\"GetDeletedChatFails\", func(t *testing.T) {\n\t\t// Create and delete chat\n\t\tchat := &types.Chat{\n\t\t\tAssistantID: \"test_assistant\",\n\t\t}\n\t\terr := store.CreateChat(chat)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to create chat: %v\", err)\n\t\t}\n\n\t\terr = store.DeleteChat(chat.ChatID)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to delete chat: %v\", err)\n\t\t}\n\n\t\t// Try to get deleted chat\n\t\t_, err = store.GetChat(chat.ChatID)\n\t\tif err == nil {\n\t\t\tt.Error(\"Expected error when getting deleted chat\")\n\t\t}\n\t})\n}\n\n// TestUpdateChat tests updating chat sessions\nfunc TestUpdateChat(t *testing.T) {\n\ttest.Prepare(t, config.Conf)\n\tdefer test.Clean()\n\n\tstore, err := xun.NewXun(types.Setting{\n\t\tConnector: \"default\",\n\t})\n\tif err != nil {\n\t\tt.Fatalf(\"Failed to create store: %v\", err)\n\t}\n\n\tt.Run(\"UpdateTitle\", func(t *testing.T) {\n\t\tchat := &types.Chat{\n\t\t\tAssistantID: \"test_assistant\",\n\t\t\tTitle: \"Original Title\",\n\t\t}\n\t\terr := store.CreateChat(chat)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to create chat: %v\", err)\n\t\t}\n\n\t\terr = store.UpdateChat(chat.ChatID, map[string]interface{}{\n\t\t\t\"title\": \"Updated Title\",\n\t\t})\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to update chat: %v\", err)\n\t\t}\n\n\t\tretrieved, err := store.GetChat(chat.ChatID)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to retrieve chat: %v\", err)\n\t\t}\n\n\t\tif retrieved.Title != \"Updated Title\" {\n\t\t\tt.Errorf(\"Expected title 'Updated Title', got '%s'\", retrieved.Title)\n\t\t}\n\n\t\t// Clean up\n\t\t_ = store.DeleteChat(chat.ChatID)\n\t})\n\n\tt.Run(\"UpdateLastConnector\", func(t *testing.T) {\n\t\tchat := &types.Chat{\n\t\t\tAssistantID: \"test_assistant\",\n\t\t\tLastConnector: \"openai\",\n\t\t}\n\t\terr := store.CreateChat(chat)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to create chat: %v\", err)\n\t\t}\n\n\t\t// Verify initial connector\n\t\tretrieved, err := store.GetChat(chat.ChatID)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to retrieve chat: %v\", err)\n\t\t}\n\t\tif retrieved.LastConnector != \"openai\" {\n\t\t\tt.Errorf(\"Expected last_connector 'openai', got '%s'\", retrieved.LastConnector)\n\t\t}\n\n\t\t// Update to different connector (simulating user switching connector)\n\t\terr = store.UpdateChat(chat.ChatID, map[string]interface{}{\n\t\t\t\"last_connector\": \"anthropic\",\n\t\t})\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to update chat: %v\", err)\n\t\t}\n\n\t\t// Verify updated connector\n\t\tretrieved, err = store.GetChat(chat.ChatID)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to retrieve chat: %v\", err)\n\t\t}\n\t\tif retrieved.LastConnector != \"anthropic\" {\n\t\t\tt.Errorf(\"Expected last_connector 'anthropic', got '%s'\", retrieved.LastConnector)\n\t\t}\n\n\t\t// Clean up\n\t\t_ = store.DeleteChat(chat.ChatID)\n\t})\n\n\tt.Run(\"UpdateLastConnectorAndLastMessageAt\", func(t *testing.T) {\n\t\t// This simulates what FlushBuffer does\n\t\tchat := &types.Chat{\n\t\t\tAssistantID: \"test_assistant\",\n\t\t\tLastConnector: \"openai\",\n\t\t}\n\t\terr := store.CreateChat(chat)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to create chat: %v\", err)\n\t\t}\n\n\t\t// Update both fields together (like FlushBuffer does)\n\t\tnow := time.Now()\n\t\terr = store.UpdateChat(chat.ChatID, map[string]interface{}{\n\t\t\t\"last_message_at\": now,\n\t\t\t\"last_connector\": \"claude\",\n\t\t})\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to update chat: %v\", err)\n\t\t}\n\n\t\tretrieved, err := store.GetChat(chat.ChatID)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to retrieve chat: %v\", err)\n\t\t}\n\n\t\tif retrieved.LastConnector != \"claude\" {\n\t\t\tt.Errorf(\"Expected last_connector 'claude', got '%s'\", retrieved.LastConnector)\n\t\t}\n\t\tif retrieved.LastMessageAt == nil {\n\t\t\tt.Error(\"Expected last_message_at to be set\")\n\t\t}\n\n\t\t// Clean up\n\t\t_ = store.DeleteChat(chat.ChatID)\n\t})\n\n\tt.Run(\"UpdateMultipleFields\", func(t *testing.T) {\n\t\tchat := &types.Chat{\n\t\t\tAssistantID: \"test_assistant\",\n\t\t\tTitle: \"Original\",\n\t\t\tStatus: \"active\",\n\t\t\tShare: \"private\",\n\t\t}\n\t\terr := store.CreateChat(chat)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to create chat: %v\", err)\n\t\t}\n\n\t\terr = store.UpdateChat(chat.ChatID, map[string]interface{}{\n\t\t\t\"title\": \"Updated\",\n\t\t\t\"status\": \"archived\",\n\t\t\t\"share\": \"team\",\n\t\t\t\"public\": true,\n\t\t\t\"sort\": 50,\n\t\t})\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to update chat: %v\", err)\n\t\t}\n\n\t\tretrieved, err := store.GetChat(chat.ChatID)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to retrieve chat: %v\", err)\n\t\t}\n\n\t\tif retrieved.Title != \"Updated\" {\n\t\t\tt.Errorf(\"Expected title 'Updated', got '%s'\", retrieved.Title)\n\t\t}\n\t\tif retrieved.Status != \"archived\" {\n\t\t\tt.Errorf(\"Expected status 'archived', got '%s'\", retrieved.Status)\n\t\t}\n\t\tif retrieved.Share != \"team\" {\n\t\t\tt.Errorf(\"Expected share 'team', got '%s'\", retrieved.Share)\n\t\t}\n\t\tif !retrieved.Public {\n\t\t\tt.Error(\"Expected public to be true\")\n\t\t}\n\t\tif retrieved.Sort != 50 {\n\t\t\tt.Errorf(\"Expected sort 50, got %d\", retrieved.Sort)\n\t\t}\n\n\t\t// Clean up\n\t\t_ = store.DeleteChat(chat.ChatID)\n\t})\n\n\tt.Run(\"UpdateMetadata\", func(t *testing.T) {\n\t\tchat := &types.Chat{\n\t\t\tAssistantID: \"test_assistant\",\n\t\t}\n\t\terr := store.CreateChat(chat)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to create chat: %v\", err)\n\t\t}\n\n\t\terr = store.UpdateChat(chat.ChatID, map[string]interface{}{\n\t\t\t\"metadata\": map[string]interface{}{\n\t\t\t\t\"key1\": \"value1\",\n\t\t\t\t\"key2\": 123,\n\t\t\t},\n\t\t})\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to update metadata: %v\", err)\n\t\t}\n\n\t\tretrieved, err := store.GetChat(chat.ChatID)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to retrieve chat: %v\", err)\n\t\t}\n\n\t\tif retrieved.Metadata == nil {\n\t\t\tt.Fatal(\"Expected metadata to be set\")\n\t\t}\n\t\tif retrieved.Metadata[\"key1\"] != \"value1\" {\n\t\t\tt.Errorf(\"Expected metadata key1 'value1', got '%v'\", retrieved.Metadata[\"key1\"])\n\t\t}\n\n\t\t// Clean up\n\t\t_ = store.DeleteChat(chat.ChatID)\n\t})\n\n\tt.Run(\"UpdateNonExistentChatFails\", func(t *testing.T) {\n\t\terr := store.UpdateChat(\"nonexistent_chat\", map[string]interface{}{\n\t\t\t\"title\": \"Test\",\n\t\t})\n\t\tif err == nil {\n\t\t\tt.Error(\"Expected error when updating non-existent chat\")\n\t\t}\n\t})\n\n\tt.Run(\"UpdateWithEmptyIDFails\", func(t *testing.T) {\n\t\terr := store.UpdateChat(\"\", map[string]interface{}{\n\t\t\t\"title\": \"Test\",\n\t\t})\n\t\tif err == nil {\n\t\t\tt.Error(\"Expected error when updating with empty ID\")\n\t\t}\n\t})\n\n\tt.Run(\"UpdateWithEmptyFieldsFails\", func(t *testing.T) {\n\t\tchat := &types.Chat{\n\t\t\tAssistantID: \"test_assistant\",\n\t\t}\n\t\terr := store.CreateChat(chat)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to create chat: %v\", err)\n\t\t}\n\n\t\terr = store.UpdateChat(chat.ChatID, map[string]interface{}{})\n\t\tif err == nil {\n\t\t\tt.Error(\"Expected error when updating with empty fields\")\n\t\t}\n\n\t\t// Clean up\n\t\t_ = store.DeleteChat(chat.ChatID)\n\t})\n\n\tt.Run(\"UpdateSkipsSystemFields\", func(t *testing.T) {\n\t\tchat := &types.Chat{\n\t\t\tAssistantID: \"test_assistant\",\n\t\t}\n\t\terr := store.CreateChat(chat)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to create chat: %v\", err)\n\t\t}\n\n\t\toriginalID := chat.ChatID\n\n\t\t// Try to update system fields\n\t\terr = store.UpdateChat(chat.ChatID, map[string]interface{}{\n\t\t\t\"chat_id\": \"new_id\",\n\t\t\t\"title\": \"Valid Update\",\n\t\t})\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to update chat: %v\", err)\n\t\t}\n\n\t\t// Verify chat_id unchanged\n\t\tretrieved, err := store.GetChat(originalID)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to retrieve chat: %v\", err)\n\t\t}\n\n\t\tif retrieved.ChatID != originalID {\n\t\t\tt.Errorf(\"Expected chat_id to remain '%s', got '%s'\", originalID, retrieved.ChatID)\n\t\t}\n\t\tif retrieved.Title != \"Valid Update\" {\n\t\t\tt.Errorf(\"Expected title 'Valid Update', got '%s'\", retrieved.Title)\n\t\t}\n\n\t\t// Clean up\n\t\t_ = store.DeleteChat(chat.ChatID)\n\t})\n}\n\n// TestDeleteChat tests deleting chat sessions\nfunc TestDeleteChat(t *testing.T) {\n\ttest.Prepare(t, config.Conf)\n\tdefer test.Clean()\n\n\tstore, err := xun.NewXun(types.Setting{\n\t\tConnector: \"default\",\n\t})\n\tif err != nil {\n\t\tt.Fatalf(\"Failed to create store: %v\", err)\n\t}\n\n\tt.Run(\"DeleteExistingChat\", func(t *testing.T) {\n\t\tchat := &types.Chat{\n\t\t\tAssistantID: \"test_assistant\",\n\t\t}\n\t\terr := store.CreateChat(chat)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to create chat: %v\", err)\n\t\t}\n\n\t\terr = store.DeleteChat(chat.ChatID)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to delete chat: %v\", err)\n\t\t}\n\n\t\t// Verify deleted\n\t\t_, err = store.GetChat(chat.ChatID)\n\t\tif err == nil {\n\t\t\tt.Error(\"Expected error when getting deleted chat\")\n\t\t}\n\t})\n\n\tt.Run(\"DeleteNonExistentChatFails\", func(t *testing.T) {\n\t\terr := store.DeleteChat(\"nonexistent_chat\")\n\t\tif err == nil {\n\t\t\tt.Error(\"Expected error when deleting non-existent chat\")\n\t\t}\n\t})\n\n\tt.Run(\"DeleteWithEmptyIDFails\", func(t *testing.T) {\n\t\terr := store.DeleteChat(\"\")\n\t\tif err == nil {\n\t\t\tt.Error(\"Expected error when deleting with empty ID\")\n\t\t}\n\t})\n\n\tt.Run(\"DeleteAlreadyDeletedChatFails\", func(t *testing.T) {\n\t\tchat := &types.Chat{\n\t\t\tAssistantID: \"test_assistant\",\n\t\t}\n\t\terr := store.CreateChat(chat)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to create chat: %v\", err)\n\t\t}\n\n\t\t// Delete first time\n\t\terr = store.DeleteChat(chat.ChatID)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to delete chat: %v\", err)\n\t\t}\n\n\t\t// Try to delete again\n\t\terr = store.DeleteChat(chat.ChatID)\n\t\tif err == nil {\n\t\t\tt.Error(\"Expected error when deleting already deleted chat\")\n\t\t}\n\t})\n}\n\n// TestListChats tests listing chat sessions\nfunc TestListChats(t *testing.T) {\n\ttest.Prepare(t, config.Conf)\n\tdefer test.Clean()\n\n\tstore, err := xun.NewXun(types.Setting{\n\t\tConnector: \"default\",\n\t})\n\tif err != nil {\n\t\tt.Fatalf(\"Failed to create store: %v\", err)\n\t}\n\n\t// Create test chats\n\tchatIDs := []string{}\n\tfor i := 0; i < 5; i++ {\n\t\tchat := &types.Chat{\n\t\t\tAssistantID: \"test_assistant\",\n\t\t\tTitle: fmt.Sprintf(\"Chat %d\", i),\n\t\t\tStatus: \"active\",\n\t\t}\n\t\tif i >= 3 {\n\t\t\tchat.Status = \"archived\"\n\t\t}\n\t\terr := store.CreateChat(chat)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to create chat: %v\", err)\n\t\t}\n\t\tchatIDs = append(chatIDs, chat.ChatID)\n\n\t\t// Add small delay to ensure different timestamps\n\t\ttime.Sleep(10 * time.Millisecond)\n\t}\n\n\t// Clean up at the end\n\tdefer func() {\n\t\tfor _, id := range chatIDs {\n\t\t\t_ = store.DeleteChat(id)\n\t\t}\n\t}()\n\n\tt.Run(\"ListAllChats\", func(t *testing.T) {\n\t\tresult, err := store.ListChats(types.ChatFilter{\n\t\t\tPage: 1,\n\t\t\tPageSize: 20,\n\t\t})\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to list chats: %v\", err)\n\t\t}\n\n\t\tif len(result.Data) < 5 {\n\t\t\tt.Errorf(\"Expected at least 5 chats, got %d\", len(result.Data))\n\t\t}\n\t})\n\n\tt.Run(\"ListChatsByStatus\", func(t *testing.T) {\n\t\tresult, err := store.ListChats(types.ChatFilter{\n\t\t\tStatus: \"active\",\n\t\t\tPage: 1,\n\t\t\tPageSize: 20,\n\t\t})\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to list chats: %v\", err)\n\t\t}\n\n\t\tfor _, chat := range result.Data {\n\t\t\tif chat.Status != \"active\" {\n\t\t\t\tt.Errorf(\"Expected status 'active', got '%s'\", chat.Status)\n\t\t\t}\n\t\t}\n\t})\n\n\tt.Run(\"ListChatsByAssistant\", func(t *testing.T) {\n\t\tresult, err := store.ListChats(types.ChatFilter{\n\t\t\tAssistantID: \"test_assistant\",\n\t\t\tPage: 1,\n\t\t\tPageSize: 20,\n\t\t})\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to list chats: %v\", err)\n\t\t}\n\n\t\tfor _, chat := range result.Data {\n\t\t\tif chat.AssistantID != \"test_assistant\" {\n\t\t\t\tt.Errorf(\"Expected assistant_id 'test_assistant', got '%s'\", chat.AssistantID)\n\t\t\t}\n\t\t}\n\t})\n\n\tt.Run(\"ListChatsByKeywords\", func(t *testing.T) {\n\t\tresult, err := store.ListChats(types.ChatFilter{\n\t\t\tKeywords: \"Chat 1\",\n\t\t\tPage: 1,\n\t\t\tPageSize: 20,\n\t\t})\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to list chats: %v\", err)\n\t\t}\n\n\t\tfound := false\n\t\tfor _, chat := range result.Data {\n\t\t\tif chat.Title == \"Chat 1\" {\n\t\t\t\tfound = true\n\t\t\t\tbreak\n\t\t\t}\n\t\t}\n\t\tif !found {\n\t\t\tt.Error(\"Expected to find chat with title 'Chat 1'\")\n\t\t}\n\t})\n\n\tt.Run(\"ListChatsPagination\", func(t *testing.T) {\n\t\t// First page\n\t\tresult1, err := store.ListChats(types.ChatFilter{\n\t\t\tPage: 1,\n\t\t\tPageSize: 2,\n\t\t})\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to list first page: %v\", err)\n\t\t}\n\n\t\tif len(result1.Data) > 2 {\n\t\t\tt.Errorf(\"Expected max 2 chats, got %d\", len(result1.Data))\n\t\t}\n\t\tif result1.Page != 1 {\n\t\t\tt.Errorf(\"Expected page 1, got %d\", result1.Page)\n\t\t}\n\t\tif result1.PageSize != 2 {\n\t\t\tt.Errorf(\"Expected pagesize 2, got %d\", result1.PageSize)\n\t\t}\n\n\t\t// Second page\n\t\tif result1.Total > 2 {\n\t\t\tresult2, err := store.ListChats(types.ChatFilter{\n\t\t\t\tPage: 2,\n\t\t\t\tPageSize: 2,\n\t\t\t})\n\t\t\tif err != nil {\n\t\t\t\tt.Fatalf(\"Failed to list second page: %v\", err)\n\t\t\t}\n\t\t\tif result2.Page != 2 {\n\t\t\t\tt.Errorf(\"Expected page 2, got %d\", result2.Page)\n\t\t\t}\n\t\t}\n\t})\n\n\tt.Run(\"ListChatsWithGrouping\", func(t *testing.T) {\n\t\tresult, err := store.ListChats(types.ChatFilter{\n\t\t\tGroupBy: \"time\",\n\t\t\tPage: 1,\n\t\t\tPageSize: 20,\n\t\t})\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to list chats with grouping: %v\", err)\n\t\t}\n\n\t\t// Should have groups when GroupBy is \"time\"\n\t\tif result.Groups == nil {\n\t\t\tt.Error(\"Expected groups to be set when GroupBy='time'\")\n\t\t}\n\n\t\t// Verify group structure\n\t\tfor _, group := range result.Groups {\n\t\t\tif group.Key == \"\" {\n\t\t\t\tt.Error(\"Expected group key to be set\")\n\t\t\t}\n\t\t\tif group.Label == \"\" {\n\t\t\t\tt.Error(\"Expected group label to be set\")\n\t\t\t}\n\t\t\tif group.Count != len(group.Chats) {\n\t\t\t\tt.Errorf(\"Expected count %d to match chats length %d\", group.Count, len(group.Chats))\n\t\t\t}\n\t\t}\n\t})\n\n\tt.Run(\"ListChatsWithTimeRange\", func(t *testing.T) {\n\t\tnow := time.Now()\n\t\tyesterday := now.AddDate(0, 0, -1)\n\n\t\tresult, err := store.ListChats(types.ChatFilter{\n\t\t\tStartTime: &yesterday,\n\t\t\tEndTime: &now,\n\t\t\tTimeField: \"created_at\",\n\t\t\tPage: 1,\n\t\t\tPageSize: 20,\n\t\t})\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to list chats with time range: %v\", err)\n\t\t}\n\n\t\t// Should return chats created within the time range\n\t\tt.Logf(\"Found %d chats in time range\", len(result.Data))\n\t})\n\n\tt.Run(\"ListChatsWithSorting\", func(t *testing.T) {\n\t\t// Ascending order\n\t\tresultAsc, err := store.ListChats(types.ChatFilter{\n\t\t\tOrderBy: \"created_at\",\n\t\t\tOrder: \"asc\",\n\t\t\tPage: 1,\n\t\t\tPageSize: 20,\n\t\t})\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to list chats ascending: %v\", err)\n\t\t}\n\n\t\t// Descending order\n\t\tresultDesc, err := store.ListChats(types.ChatFilter{\n\t\t\tOrderBy: \"created_at\",\n\t\t\tOrder: \"desc\",\n\t\t\tPage: 1,\n\t\t\tPageSize: 20,\n\t\t})\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to list chats descending: %v\", err)\n\t\t}\n\n\t\t// Verify different order\n\t\tif len(resultAsc.Data) > 1 && len(resultDesc.Data) > 1 {\n\t\t\tif resultAsc.Data[0].ChatID == resultDesc.Data[0].ChatID {\n\t\t\t\t// This is fine if there's only one chat, but otherwise order should differ\n\t\t\t\tif len(resultAsc.Data) > 1 {\n\t\t\t\t\tt.Logf(\"First chat in asc: %s, first in desc: %s\", resultAsc.Data[0].ChatID, resultDesc.Data[0].ChatID)\n\t\t\t\t}\n\t\t\t}\n\t\t}\n\t})\n\n\tt.Run(\"ListChatsWithQueryFilter\", func(t *testing.T) {\n\t\tresult, err := store.ListChats(types.ChatFilter{\n\t\t\tPage: 1,\n\t\t\tPageSize: 20,\n\t\t\tQueryFilter: func(qb query.Query) {\n\t\t\t\tqb.Where(\"status\", \"active\")\n\t\t\t},\n\t\t})\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to list chats with query filter: %v\", err)\n\t\t}\n\n\t\tfor _, chat := range result.Data {\n\t\t\tif chat.Status != \"active\" {\n\t\t\t\tt.Errorf(\"Expected status 'active', got '%s'\", chat.Status)\n\t\t\t}\n\t\t}\n\t})\n}\n\n// TestListChatsByUserAndTeam tests filtering chats by UserID and TeamID\nfunc TestListChatsByUserAndTeam(t *testing.T) {\n\ttest.Prepare(t, config.Conf)\n\tdefer test.Clean()\n\n\tstore, err := xun.NewXun(types.Setting{\n\t\tConnector: \"default\",\n\t})\n\tif err != nil {\n\t\tt.Fatalf(\"Failed to create store: %v\", err)\n\t}\n\n\t// Create chats with different user/team combinations\n\t// Note: __yao_created_by and __yao_team_id are managed by Yao's permission system\n\t// For testing, we'll create chats and then update these fields directly via raw query\n\n\tchat1 := &types.Chat{AssistantID: \"test_assistant\", Title: \"User1 Team1 Chat\"}\n\tchat2 := &types.Chat{AssistantID: \"test_assistant\", Title: \"User1 Team2 Chat\"}\n\tchat3 := &types.Chat{AssistantID: \"test_assistant\", Title: \"User2 Team1 Chat\"}\n\tchat4 := &types.Chat{AssistantID: \"test_assistant\", Title: \"User2 Team2 Chat\"}\n\n\tfor _, chat := range []*types.Chat{chat1, chat2, chat3, chat4} {\n\t\terr := store.CreateChat(chat)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to create chat: %v\", err)\n\t\t}\n\t}\n\tdefer func() {\n\t\tstore.DeleteChat(chat1.ChatID)\n\t\tstore.DeleteChat(chat2.ChatID)\n\t\tstore.DeleteChat(chat3.ChatID)\n\t\tstore.DeleteChat(chat4.ChatID)\n\t}()\n\n\t// Update permission fields directly for testing\n\t// In production, these would be set by Yao's permission middleware\n\tupdatePermissionFields := func(chatID, userID, teamID string) error {\n\t\t// Use Yao model to update permission fields\n\t\tm := goumodel.Select(\"__yao.agent.chat\")\n\t\tif m == nil {\n\t\t\treturn fmt.Errorf(\"model __yao.agent.chat not found\")\n\t\t}\n\t\t_, err := m.UpdateWhere(\n\t\t\tgoumodel.QueryParam{Wheres: []goumodel.QueryWhere{{Column: \"chat_id\", Value: chatID}}},\n\t\t\tmap[string]interface{}{\n\t\t\t\t\"__yao_created_by\": userID,\n\t\t\t\t\"__yao_team_id\": teamID,\n\t\t\t},\n\t\t)\n\t\treturn err\n\t}\n\n\t// Set up permission fields\n\tupdatePermissionFields(chat1.ChatID, \"user1\", \"team1\")\n\tupdatePermissionFields(chat2.ChatID, \"user1\", \"team2\")\n\tupdatePermissionFields(chat3.ChatID, \"user2\", \"team1\")\n\tupdatePermissionFields(chat4.ChatID, \"user2\", \"team2\")\n\n\tt.Run(\"FilterByUserID\", func(t *testing.T) {\n\t\tresult, err := store.ListChats(types.ChatFilter{\n\t\t\tUserID: \"user1\",\n\t\t\tPage: 1,\n\t\t\tPageSize: 20,\n\t\t})\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to list chats by user: %v\", err)\n\t\t}\n\n\t\tif len(result.Data) != 2 {\n\t\t\tt.Errorf(\"Expected 2 chats for user1, got %d\", len(result.Data))\n\t\t}\n\n\t\t// Verify all returned chats belong to user1\n\t\tfor _, chat := range result.Data {\n\t\t\tif chat.Title != \"User1 Team1 Chat\" && chat.Title != \"User1 Team2 Chat\" {\n\t\t\t\tt.Errorf(\"Unexpected chat title: %s\", chat.Title)\n\t\t\t}\n\t\t}\n\t})\n\n\tt.Run(\"FilterByTeamID\", func(t *testing.T) {\n\t\tresult, err := store.ListChats(types.ChatFilter{\n\t\t\tTeamID: \"team1\",\n\t\t\tPage: 1,\n\t\t\tPageSize: 20,\n\t\t})\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to list chats by team: %v\", err)\n\t\t}\n\n\t\tif len(result.Data) != 2 {\n\t\t\tt.Errorf(\"Expected 2 chats for team1, got %d\", len(result.Data))\n\t\t}\n\n\t\t// Verify all returned chats belong to team1\n\t\tfor _, chat := range result.Data {\n\t\t\tif chat.Title != \"User1 Team1 Chat\" && chat.Title != \"User2 Team1 Chat\" {\n\t\t\t\tt.Errorf(\"Unexpected chat title: %s\", chat.Title)\n\t\t\t}\n\t\t}\n\t})\n\n\tt.Run(\"FilterByUserIDAndTeamID\", func(t *testing.T) {\n\t\tresult, err := store.ListChats(types.ChatFilter{\n\t\t\tUserID: \"user1\",\n\t\t\tTeamID: \"team1\",\n\t\t\tPage: 1,\n\t\t\tPageSize: 20,\n\t\t})\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to list chats by user and team: %v\", err)\n\t\t}\n\n\t\tif len(result.Data) != 1 {\n\t\t\tt.Errorf(\"Expected 1 chat for user1+team1, got %d\", len(result.Data))\n\t\t}\n\n\t\tif len(result.Data) > 0 && result.Data[0].Title != \"User1 Team1 Chat\" {\n\t\t\tt.Errorf(\"Expected 'User1 Team1 Chat', got '%s'\", result.Data[0].Title)\n\t\t}\n\t})\n\n\tt.Run(\"FilterByUserIDWithOtherFilters\", func(t *testing.T) {\n\t\t// Combine UserID with Status filter\n\t\tresult, err := store.ListChats(types.ChatFilter{\n\t\t\tUserID: \"user1\",\n\t\t\tStatus: \"active\",\n\t\t\tPage: 1,\n\t\t\tPageSize: 20,\n\t\t})\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to list chats: %v\", err)\n\t\t}\n\n\t\t// All user1's chats should be active (default status)\n\t\tif len(result.Data) != 2 {\n\t\t\tt.Errorf(\"Expected 2 active chats for user1, got %d\", len(result.Data))\n\t\t}\n\t})\n\n\tt.Run(\"FilterByTeamIDWithQueryFilter\", func(t *testing.T) {\n\t\t// Combine TeamID with custom QueryFilter\n\t\tresult, err := store.ListChats(types.ChatFilter{\n\t\t\tTeamID: \"team2\",\n\t\t\tPage: 1,\n\t\t\tPageSize: 20,\n\t\t\tQueryFilter: func(qb query.Query) {\n\t\t\t\t// Additional filter: only chats with \"User1\" in title\n\t\t\t\tqb.Where(\"title\", \"like\", \"%User1%\")\n\t\t\t},\n\t\t})\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to list chats: %v\", err)\n\t\t}\n\n\t\tif len(result.Data) != 1 {\n\t\t\tt.Errorf(\"Expected 1 chat (User1 in team2), got %d\", len(result.Data))\n\t\t}\n\n\t\tif len(result.Data) > 0 && result.Data[0].Title != \"User1 Team2 Chat\" {\n\t\t\tt.Errorf(\"Expected 'User1 Team2 Chat', got '%s'\", result.Data[0].Title)\n\t\t}\n\t})\n\n\tt.Run(\"FilterByNonExistentUser\", func(t *testing.T) {\n\t\tresult, err := store.ListChats(types.ChatFilter{\n\t\t\tUserID: \"nonexistent_user\",\n\t\t\tPage: 1,\n\t\t\tPageSize: 20,\n\t\t})\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to list chats: %v\", err)\n\t\t}\n\n\t\tif len(result.Data) != 0 {\n\t\t\tt.Errorf(\"Expected 0 chats for nonexistent user, got %d\", len(result.Data))\n\t\t}\n\t})\n\n\tt.Run(\"FilterByNonExistentTeam\", func(t *testing.T) {\n\t\tresult, err := store.ListChats(types.ChatFilter{\n\t\t\tTeamID: \"nonexistent_team\",\n\t\t\tPage: 1,\n\t\t\tPageSize: 20,\n\t\t})\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to list chats: %v\", err)\n\t\t}\n\n\t\tif len(result.Data) != 0 {\n\t\t\tt.Errorf(\"Expected 0 chats for nonexistent team, got %d\", len(result.Data))\n\t\t}\n\t})\n\n\tt.Run(\"QueryFilterForOrCondition\", func(t *testing.T) {\n\t\t// Use QueryFilter for complex OR condition:\n\t\t// Get chats where user is user1 OR team is team2\n\t\tresult, err := store.ListChats(types.ChatFilter{\n\t\t\tPage: 1,\n\t\t\tPageSize: 20,\n\t\t\tQueryFilter: func(qb query.Query) {\n\t\t\t\tqb.Where(func(sub query.Query) {\n\t\t\t\t\tsub.Where(\"__yao_created_by\", \"user1\").\n\t\t\t\t\t\tOrWhere(\"__yao_team_id\", \"team2\")\n\t\t\t\t})\n\t\t\t},\n\t\t})\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to list chats with OR condition: %v\", err)\n\t\t}\n\n\t\t// Should return: user1+team1, user1+team2, user2+team2 = 3 chats\n\t\tif len(result.Data) != 3 {\n\t\t\tt.Errorf(\"Expected 3 chats (user1 OR team2), got %d\", len(result.Data))\n\t\t}\n\t})\n}\n\n// TestChatCompleteWorkflow tests a complete chat workflow\nfunc TestChatCompleteWorkflow(t *testing.T) {\n\ttest.Prepare(t, config.Conf)\n\tdefer test.Clean()\n\n\tstore, err := xun.NewXun(types.Setting{\n\t\tConnector: \"default\",\n\t})\n\tif err != nil {\n\t\tt.Fatalf(\"Failed to create store: %v\", err)\n\t}\n\n\tt.Run(\"CompleteWorkflow\", func(t *testing.T) {\n\t\t// 1. Create chat\n\t\tchat := &types.Chat{\n\t\t\tAssistantID: \"workflow_assistant\",\n\t\t\tTitle: \"Workflow Test Chat\",\n\t\t\tStatus: \"active\",\n\t\t}\n\n\t\terr := store.CreateChat(chat)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to create chat: %v\", err)\n\t\t}\n\t\tt.Logf(\"Created chat: %s\", chat.ChatID)\n\n\t\t// 2. Get chat\n\t\tretrieved, err := store.GetChat(chat.ChatID)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to get chat: %v\", err)\n\t\t}\n\t\tif retrieved.Title != \"Workflow Test Chat\" {\n\t\t\tt.Errorf(\"Expected title 'Workflow Test Chat', got '%s'\", retrieved.Title)\n\t\t}\n\n\t\t// 3. Update chat\n\t\terr = store.UpdateChat(chat.ChatID, map[string]interface{}{\n\t\t\t\"title\": \"Updated Workflow Chat\",\n\t\t\t\"status\": \"archived\",\n\t\t})\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to update chat: %v\", err)\n\t\t}\n\n\t\t// 4. Verify update\n\t\tupdated, err := store.GetChat(chat.ChatID)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to get updated chat: %v\", err)\n\t\t}\n\t\tif updated.Title != \"Updated Workflow Chat\" {\n\t\t\tt.Errorf(\"Expected title 'Updated Workflow Chat', got '%s'\", updated.Title)\n\t\t}\n\t\tif updated.Status != \"archived\" {\n\t\t\tt.Errorf(\"Expected status 'archived', got '%s'\", updated.Status)\n\t\t}\n\n\t\t// 5. List chats\n\t\tresult, err := store.ListChats(types.ChatFilter{\n\t\t\tAssistantID: \"workflow_assistant\",\n\t\t\tPage: 1,\n\t\t\tPageSize: 20,\n\t\t})\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to list chats: %v\", err)\n\t\t}\n\n\t\tfound := false\n\t\tfor _, c := range result.Data {\n\t\t\tif c.ChatID == chat.ChatID {\n\t\t\t\tfound = true\n\t\t\t\tbreak\n\t\t\t}\n\t\t}\n\t\tif !found {\n\t\t\tt.Error(\"Expected to find chat in list\")\n\t\t}\n\n\t\t// 6. Delete chat\n\t\terr = store.DeleteChat(chat.ChatID)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to delete chat: %v\", err)\n\t\t}\n\n\t\t// 7. Verify deletion\n\t\t_, err = store.GetChat(chat.ChatID)\n\t\tif err == nil {\n\t\t\tt.Error(\"Expected error when getting deleted chat\")\n\t\t}\n\n\t\tt.Log(\"Complete workflow passed!\")\n\t})\n}\n" }, { "path": "agent/store/xun/message.go", "content": "package xun\n\nimport (\n\t\"fmt\"\n\t\"time\"\n\n\t\"github.com/google/uuid\"\n\tjsoniter \"github.com/json-iterator/go\"\n\t\"github.com/yaoapp/yao/agent/store/types\"\n)\n\n// =============================================================================\n// Message Management\n// =============================================================================\n\n// SaveMessages batch saves messages for a chat using a single database call\n// This is the primary write method - messages are buffered during execution\n// and batch-written at the end of a request\nfunc (store *Xun) SaveMessages(chatID string, messages []*types.Message) error {\n\tif chatID == \"\" {\n\t\treturn fmt.Errorf(\"chat_id is required\")\n\t}\n\tif len(messages) == 0 {\n\t\treturn nil // Nothing to save\n\t}\n\n\t// Prepare batch insert data\n\tnow := time.Now()\n\trows := make([]map[string]interface{}, 0, len(messages))\n\n\tfor _, msg := range messages {\n\t\tif msg == nil {\n\t\t\tcontinue\n\t\t}\n\n\t\t// Generate message_id if not provided\n\t\tmessageID := msg.MessageID\n\t\tif messageID == \"\" {\n\t\t\tmessageID = uuid.New().String()\n\t\t}\n\n\t\t// Validate required fields\n\t\tif msg.Role == \"\" {\n\t\t\treturn fmt.Errorf(\"message role is required\")\n\t\t}\n\t\tif msg.Type == \"\" {\n\t\t\treturn fmt.Errorf(\"message type is required\")\n\t\t}\n\t\tif msg.Props == nil {\n\t\t\treturn fmt.Errorf(\"message props is required\")\n\t\t}\n\n\t\t// Serialize JSON fields\n\t\tpropsJSON, err := jsoniter.MarshalToString(msg.Props)\n\t\tif err != nil {\n\t\t\treturn fmt.Errorf(\"failed to marshal props: %w\", err)\n\t\t}\n\n\t\t// Build row with all fields (including nullable ones for consistent batch insert)\n\t\trow := map[string]interface{}{\n\t\t\t\"message_id\": messageID,\n\t\t\t\"chat_id\": chatID,\n\t\t\t\"role\": msg.Role,\n\t\t\t\"type\": msg.Type,\n\t\t\t\"props\": propsJSON,\n\t\t\t\"sequence\": msg.Sequence,\n\t\t\t\"request_id\": nil,\n\t\t\t\"block_id\": nil,\n\t\t\t\"thread_id\": nil,\n\t\t\t\"assistant_id\": nil,\n\t\t\t\"connector\": nil,\n\t\t\t\"mode\": nil,\n\t\t\t\"metadata\": nil,\n\t\t\t\"created_at\": now,\n\t\t\t\"updated_at\": now,\n\t\t}\n\n\t\t// Set nullable fields if they have values\n\t\tif msg.RequestID != \"\" {\n\t\t\trow[\"request_id\"] = msg.RequestID\n\t\t}\n\t\tif msg.BlockID != \"\" {\n\t\t\trow[\"block_id\"] = msg.BlockID\n\t\t}\n\t\tif msg.ThreadID != \"\" {\n\t\t\trow[\"thread_id\"] = msg.ThreadID\n\t\t}\n\t\tif msg.AssistantID != \"\" {\n\t\t\trow[\"assistant_id\"] = msg.AssistantID\n\t\t}\n\t\tif msg.Connector != \"\" {\n\t\t\trow[\"connector\"] = msg.Connector\n\t\t}\n\t\tif msg.Mode != \"\" {\n\t\t\trow[\"mode\"] = msg.Mode\n\t\t}\n\t\tif msg.Metadata != nil {\n\t\t\tmetadataJSON, err := jsoniter.MarshalToString(msg.Metadata)\n\t\t\tif err != nil {\n\t\t\t\treturn fmt.Errorf(\"failed to marshal metadata: %w\", err)\n\t\t\t}\n\t\t\trow[\"metadata\"] = metadataJSON\n\t\t}\n\n\t\trows = append(rows, row)\n\t}\n\n\tif len(rows) == 0 {\n\t\treturn nil\n\t}\n\n\t// Single batch insert - one database call for all messages\n\treturn store.newQueryMessage().Insert(rows)\n}\n\n// GetMessages retrieves messages for a chat with filtering\nfunc (store *Xun) GetMessages(chatID string, filter types.MessageFilter) ([]*types.Message, error) {\n\tif chatID == \"\" {\n\t\treturn nil, fmt.Errorf(\"chat_id is required\")\n\t}\n\n\tqb := store.newQueryMessage().\n\t\tWhere(\"chat_id\", chatID).\n\t\tWhereNull(\"deleted_at\")\n\n\t// Apply filters\n\tif filter.RequestID != \"\" {\n\t\tqb.Where(\"request_id\", filter.RequestID)\n\t}\n\tif filter.Role != \"\" {\n\t\tqb.Where(\"role\", filter.Role)\n\t}\n\tif filter.BlockID != \"\" {\n\t\tqb.Where(\"block_id\", filter.BlockID)\n\t}\n\tif filter.ThreadID != \"\" {\n\t\tqb.Where(\"thread_id\", filter.ThreadID)\n\t}\n\tif filter.Type != \"\" {\n\t\tqb.Where(\"type\", filter.Type)\n\t}\n\n\t// When Limit is specified WITHOUT Offset, we want the N most-recent\n\t// messages. Strategy: query DESC to get the latest rows, then reverse\n\t// the slice so the caller receives them in chronological (ASC) order.\n\t// When Offset is also present, the caller is doing forward pagination,\n\t// so we keep ASC order and apply Limit+Offset normally.\n\tneedReverse := false\n\tif filter.Limit > 0 && filter.Offset <= 0 {\n\t\tqb.Limit(filter.Limit)\n\t\tqb.OrderBy(\"id\", \"desc\")\n\t\tneedReverse = true\n\t} else if filter.Limit > 0 && filter.Offset > 0 {\n\t\tqb.Limit(filter.Limit).Offset(filter.Offset)\n\t\tqb.OrderBy(\"id\", \"asc\")\n\t} else {\n\t\tif filter.Offset > 0 {\n\t\t\tqb.Limit(1000000).Offset(filter.Offset)\n\t\t}\n\t\tqb.OrderBy(\"id\", \"asc\")\n\t}\n\n\trows, err := qb.Get()\n\tif err != nil {\n\t\treturn nil, err\n\t}\n\n\tmessages := make([]*types.Message, 0, len(rows))\n\tfor _, row := range rows {\n\t\tdata := row.ToMap()\n\t\tif data == nil || data[\"message_id\"] == nil {\n\t\t\tcontinue\n\t\t}\n\n\t\tmsg, err := store.rowToMessage(data)\n\t\tif err != nil {\n\t\t\tcontinue\n\t\t}\n\t\tmessages = append(messages, msg)\n\t}\n\n\tif needReverse {\n\t\tfor i, j := 0, len(messages)-1; i < j; i, j = i+1, j-1 {\n\t\t\tmessages[i], messages[j] = messages[j], messages[i]\n\t\t}\n\t}\n\n\treturn messages, nil\n}\n\n// UpdateMessage updates a single message\nfunc (store *Xun) UpdateMessage(messageID string, updates map[string]interface{}) error {\n\tif messageID == \"\" {\n\t\treturn fmt.Errorf(\"message_id is required\")\n\t}\n\tif len(updates) == 0 {\n\t\treturn fmt.Errorf(\"no fields to update\")\n\t}\n\n\t// Check if message exists\n\texists, err := store.newQueryMessage().\n\t\tWhere(\"message_id\", messageID).\n\t\tWhereNull(\"deleted_at\").\n\t\tExists()\n\tif err != nil {\n\t\treturn err\n\t}\n\tif !exists {\n\t\treturn fmt.Errorf(\"message %s not found\", messageID)\n\t}\n\n\t// Prepare update data\n\tdata := make(map[string]interface{})\n\n\tfor key, value := range updates {\n\t\t// Skip system fields\n\t\tif key == \"message_id\" || key == \"chat_id\" || key == \"created_at\" {\n\t\t\tcontinue\n\t\t}\n\n\t\t// Handle JSON fields\n\t\tif key == \"props\" || key == \"metadata\" {\n\t\t\tif value != nil {\n\t\t\t\tjsonStr, err := jsoniter.MarshalToString(value)\n\t\t\t\tif err != nil {\n\t\t\t\t\treturn fmt.Errorf(\"failed to marshal %s: %w\", key, err)\n\t\t\t\t}\n\t\t\t\tdata[key] = jsonStr\n\t\t\t} else {\n\t\t\t\tdata[key] = nil\n\t\t\t}\n\t\t\tcontinue\n\t\t}\n\n\t\tdata[key] = value\n\t}\n\n\t// Always update updated_at\n\tdata[\"updated_at\"] = time.Now()\n\n\tif len(data) == 0 {\n\t\treturn fmt.Errorf(\"no valid fields to update\")\n\t}\n\n\t_, err = store.newQueryMessage().\n\t\tWhere(\"message_id\", messageID).\n\t\tUpdate(data)\n\n\treturn err\n}\n\n// DeleteMessages soft deletes specific messages from a chat\nfunc (store *Xun) DeleteMessages(chatID string, messageIDs []string) error {\n\tif chatID == \"\" {\n\t\treturn fmt.Errorf(\"chat_id is required\")\n\t}\n\tif len(messageIDs) == 0 {\n\t\treturn nil // Nothing to delete\n\t}\n\n\t// Soft delete all specified messages in one query\n\t_, err := store.newQueryMessage().\n\t\tWhere(\"chat_id\", chatID).\n\t\tWhereIn(\"message_id\", messageIDs).\n\t\tWhereNull(\"deleted_at\").\n\t\tUpdate(map[string]interface{}{\n\t\t\t\"deleted_at\": time.Now(),\n\t\t\t\"updated_at\": time.Now(),\n\t\t})\n\n\treturn err\n}\n\n// GetMessageByID retrieves a single message by ID\nfunc (store *Xun) GetMessageByID(messageID string) (*types.Message, error) {\n\tif messageID == \"\" {\n\t\treturn nil, fmt.Errorf(\"message_id is required\")\n\t}\n\n\trow, err := store.newQueryMessage().\n\t\tWhere(\"message_id\", messageID).\n\t\tWhereNull(\"deleted_at\").\n\t\tFirst()\n\tif err != nil {\n\t\treturn nil, err\n\t}\n\n\tif row == nil {\n\t\treturn nil, fmt.Errorf(\"message %s not found\", messageID)\n\t}\n\n\tdata := row.ToMap()\n\tif len(data) == 0 || data[\"message_id\"] == nil {\n\t\treturn nil, fmt.Errorf(\"message %s not found\", messageID)\n\t}\n\n\treturn store.rowToMessage(data)\n}\n\n// GetMessageCount returns the count of messages for a chat\nfunc (store *Xun) GetMessageCount(chatID string) (int64, error) {\n\tif chatID == \"\" {\n\t\treturn 0, fmt.Errorf(\"chat_id is required\")\n\t}\n\n\treturn store.newQueryMessage().\n\t\tWhere(\"chat_id\", chatID).\n\t\tWhereNull(\"deleted_at\").\n\t\tCount()\n}\n\n// GetLastSequence returns the last sequence number for a chat\nfunc (store *Xun) GetLastSequence(chatID string) (int, error) {\n\tif chatID == \"\" {\n\t\treturn 0, fmt.Errorf(\"chat_id is required\")\n\t}\n\n\trow, err := store.newQueryMessage().\n\t\tWhere(\"chat_id\", chatID).\n\t\tWhereNull(\"deleted_at\").\n\t\tOrderBy(\"sequence\", \"desc\").\n\t\tFirst()\n\tif err != nil {\n\t\treturn 0, err\n\t}\n\n\tif row == nil {\n\t\treturn 0, nil\n\t}\n\n\tdata := row.ToMap()\n\treturn getInt(data, \"sequence\"), nil\n}\n\n// =============================================================================\n// Helper Functions\n// =============================================================================\n\n// rowToMessage converts a database row to a Message struct\nfunc (store *Xun) rowToMessage(data map[string]interface{}) (*types.Message, error) {\n\tmsg := &types.Message{\n\t\tMessageID: getString(data, \"message_id\"),\n\t\tChatID: getString(data, \"chat_id\"),\n\t\tRequestID: getString(data, \"request_id\"),\n\t\tRole: getString(data, \"role\"),\n\t\tType: getString(data, \"type\"),\n\t\tBlockID: getString(data, \"block_id\"),\n\t\tThreadID: getString(data, \"thread_id\"),\n\t\tAssistantID: getString(data, \"assistant_id\"),\n\t\tConnector: getString(data, \"connector\"),\n\t\tMode: getString(data, \"mode\"),\n\t\tSequence: getInt(data, \"sequence\"),\n\t}\n\n\t// Handle timestamps\n\tif createdAt := getTime(data, \"created_at\"); createdAt != nil {\n\t\tmsg.CreatedAt = *createdAt\n\t}\n\tif updatedAt := getTime(data, \"updated_at\"); updatedAt != nil {\n\t\tmsg.UpdatedAt = *updatedAt\n\t}\n\n\t// Handle props (required)\n\tif props := data[\"props\"]; props != nil {\n\t\tif propsStr, ok := props.(string); ok && propsStr != \"\" {\n\t\t\tvar propsMap map[string]interface{}\n\t\t\tif err := jsoniter.UnmarshalFromString(propsStr, &propsMap); err == nil {\n\t\t\t\tmsg.Props = propsMap\n\t\t\t}\n\t\t} else if propsMap, ok := props.(map[string]interface{}); ok {\n\t\t\tmsg.Props = propsMap\n\t\t}\n\t}\n\n\t// Handle metadata (optional)\n\tif metadata := data[\"metadata\"]; metadata != nil {\n\t\tif metaStr, ok := metadata.(string); ok && metaStr != \"\" {\n\t\t\tvar metaMap map[string]interface{}\n\t\t\tif err := jsoniter.UnmarshalFromString(metaStr, &metaMap); err == nil {\n\t\t\t\tmsg.Metadata = metaMap\n\t\t\t}\n\t\t} else if metaMap, ok := metadata.(map[string]interface{}); ok {\n\t\t\tmsg.Metadata = metaMap\n\t\t}\n\t}\n\n\treturn msg, nil\n}\n" }, { "path": "agent/store/xun/message_test.go", "content": "package xun_test\n\nimport (\n\t\"fmt\"\n\t\"testing\"\n\t\"time\"\n\n\t\"github.com/yaoapp/yao/agent/store/types\"\n\t\"github.com/yaoapp/yao/agent/store/xun\"\n\t\"github.com/yaoapp/yao/config\"\n\t\"github.com/yaoapp/yao/test\"\n)\n\n// TestSaveMessages tests batch saving messages\nfunc TestSaveMessages(t *testing.T) {\n\ttest.Prepare(t, config.Conf)\n\tdefer test.Clean()\n\n\tstore, err := xun.NewXun(types.Setting{\n\t\tConnector: \"default\",\n\t})\n\tif err != nil {\n\t\tt.Fatalf(\"Failed to create store: %v\", err)\n\t}\n\n\t// Create a chat first\n\tchat := &types.Chat{\n\t\tAssistantID: \"test_assistant\",\n\t\tTitle: \"Message Test Chat\",\n\t}\n\terr = store.CreateChat(chat)\n\tif err != nil {\n\t\tt.Fatalf(\"Failed to create chat: %v\", err)\n\t}\n\tdefer store.DeleteChat(chat.ChatID)\n\n\tt.Run(\"SaveSingleMessage\", func(t *testing.T) {\n\t\tmessages := []*types.Message{\n\t\t\t{\n\t\t\t\tRole: \"user\",\n\t\t\t\tType: \"text\",\n\t\t\t\tProps: map[string]interface{}{\"content\": \"Hello, world!\"},\n\t\t\t\tSequence: 1,\n\t\t\t},\n\t\t}\n\n\t\terr := store.SaveMessages(chat.ChatID, messages)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to save message: %v\", err)\n\t\t}\n\n\t\t// Verify\n\t\tretrieved, err := store.GetMessages(chat.ChatID, types.MessageFilter{})\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to get messages: %v\", err)\n\t\t}\n\n\t\tif len(retrieved) < 1 {\n\t\t\tt.Fatal(\"Expected at least 1 message\")\n\t\t}\n\n\t\t// Find the message we just saved\n\t\tvar found *types.Message\n\t\tfor _, msg := range retrieved {\n\t\t\tif msg.Sequence == 1 && msg.Type == \"text\" {\n\t\t\t\tfound = msg\n\t\t\t\tbreak\n\t\t\t}\n\t\t}\n\n\t\tif found == nil {\n\t\t\tt.Fatal(\"Could not find saved message\")\n\t\t}\n\n\t\tif found.Role != \"user\" {\n\t\t\tt.Errorf(\"Expected role 'user', got '%s'\", found.Role)\n\t\t}\n\t\tif found.Props[\"content\"] != \"Hello, world!\" {\n\t\t\tt.Errorf(\"Expected content 'Hello, world!', got '%v'\", found.Props[\"content\"])\n\t\t}\n\t})\n\n\tt.Run(\"SaveBatchMessages\", func(t *testing.T) {\n\t\t// Create a new chat for this test\n\t\tbatchChat := &types.Chat{\n\t\t\tAssistantID: \"test_assistant\",\n\t\t\tTitle: \"Batch Message Test\",\n\t\t}\n\t\terr := store.CreateChat(batchChat)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to create chat: %v\", err)\n\t\t}\n\t\tdefer store.DeleteChat(batchChat.ChatID)\n\n\t\t// Save multiple messages in one batch\n\t\tmessages := []*types.Message{\n\t\t\t{\n\t\t\t\tRole: \"user\",\n\t\t\t\tType: \"user_input\",\n\t\t\t\tProps: map[string]interface{}{\"content\": \"What's the weather?\"},\n\t\t\t\tSequence: 1,\n\t\t\t\tRequestID: \"req_001\",\n\t\t\t\tAssistantID: \"weather_assistant\",\n\t\t\t},\n\t\t\t{\n\t\t\t\tRole: \"assistant\",\n\t\t\t\tType: \"loading\",\n\t\t\t\tProps: map[string]interface{}{\"message\": \"Checking weather...\"},\n\t\t\t\tSequence: 2,\n\t\t\t\tRequestID: \"req_001\",\n\t\t\t\tBlockID: \"B1\",\n\t\t\t\tAssistantID: \"weather_assistant\",\n\t\t\t},\n\t\t\t{\n\t\t\t\tRole: \"assistant\",\n\t\t\t\tType: \"text\",\n\t\t\t\tProps: map[string]interface{}{\"content\": \"The weather is sunny, 25°C.\"},\n\t\t\t\tSequence: 3,\n\t\t\t\tRequestID: \"req_001\",\n\t\t\t\tBlockID: \"B1\",\n\t\t\t\tAssistantID: \"weather_assistant\",\n\t\t\t},\n\t\t}\n\n\t\terr = store.SaveMessages(batchChat.ChatID, messages)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to save batch messages: %v\", err)\n\t\t}\n\n\t\t// Verify all messages saved\n\t\tretrieved, err := store.GetMessages(batchChat.ChatID, types.MessageFilter{})\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to get messages: %v\", err)\n\t\t}\n\n\t\tif len(retrieved) != 3 {\n\t\t\tt.Errorf(\"Expected 3 messages, got %d\", len(retrieved))\n\t\t}\n\n\t\t// Verify order (should be by sequence)\n\t\tif len(retrieved) >= 3 {\n\t\t\tif retrieved[0].Sequence != 1 {\n\t\t\t\tt.Errorf(\"Expected first message sequence 1, got %d\", retrieved[0].Sequence)\n\t\t\t}\n\t\t\tif retrieved[2].Sequence != 3 {\n\t\t\t\tt.Errorf(\"Expected last message sequence 3, got %d\", retrieved[2].Sequence)\n\t\t\t}\n\t\t}\n\n\t\tt.Logf(\"Saved %d messages in single batch call\", len(messages))\n\t})\n\n\tt.Run(\"SaveMessageWithAllFields\", func(t *testing.T) {\n\t\tfullChat := &types.Chat{\n\t\t\tAssistantID: \"test_assistant\",\n\t\t}\n\t\terr := store.CreateChat(fullChat)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to create chat: %v\", err)\n\t\t}\n\t\tdefer store.DeleteChat(fullChat.ChatID)\n\n\t\tmessages := []*types.Message{\n\t\t\t{\n\t\t\t\tRole: \"assistant\",\n\t\t\t\tType: \"tool_call\",\n\t\t\t\tProps: map[string]interface{}{\"id\": \"call_123\", \"name\": \"get_weather\", \"arguments\": `{\"location\":\"SF\"}`},\n\t\t\t\tSequence: 1,\n\t\t\t\tRequestID: \"req_full\",\n\t\t\t\tBlockID: \"B1\",\n\t\t\t\tThreadID: \"T1\",\n\t\t\t\tAssistantID: \"weather_assistant\",\n\t\t\t\tMetadata: map[string]interface{}{\"tool_call_id\": \"call_123\", \"is_tool_result\": false},\n\t\t\t},\n\t\t}\n\n\t\terr = store.SaveMessages(fullChat.ChatID, messages)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to save message: %v\", err)\n\t\t}\n\n\t\tretrieved, err := store.GetMessages(fullChat.ChatID, types.MessageFilter{})\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to get messages: %v\", err)\n\t\t}\n\n\t\tif len(retrieved) != 1 {\n\t\t\tt.Fatalf(\"Expected 1 message, got %d\", len(retrieved))\n\t\t}\n\n\t\tmsg := retrieved[0]\n\t\tif msg.RequestID != \"req_full\" {\n\t\t\tt.Errorf(\"Expected request_id 'req_full', got '%s'\", msg.RequestID)\n\t\t}\n\t\tif msg.BlockID != \"B1\" {\n\t\t\tt.Errorf(\"Expected block_id 'B1', got '%s'\", msg.BlockID)\n\t\t}\n\t\tif msg.ThreadID != \"T1\" {\n\t\t\tt.Errorf(\"Expected thread_id 'T1', got '%s'\", msg.ThreadID)\n\t\t}\n\t\tif msg.AssistantID != \"weather_assistant\" {\n\t\t\tt.Errorf(\"Expected assistant_id 'weather_assistant', got '%s'\", msg.AssistantID)\n\t\t}\n\t\tif msg.Metadata == nil {\n\t\t\tt.Error(\"Expected metadata to be set\")\n\t\t} else if msg.Metadata[\"tool_call_id\"] != \"call_123\" {\n\t\t\tt.Errorf(\"Expected metadata tool_call_id 'call_123', got '%v'\", msg.Metadata[\"tool_call_id\"])\n\t\t}\n\t})\n\n\tt.Run(\"SaveMessageWithConnector\", func(t *testing.T) {\n\t\tconnChat := &types.Chat{\n\t\t\tAssistantID: \"test_assistant\",\n\t\t}\n\t\terr := store.CreateChat(connChat)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to create chat: %v\", err)\n\t\t}\n\t\tdefer store.DeleteChat(connChat.ChatID)\n\n\t\t// Save messages with different connectors\n\t\tmessages := []*types.Message{\n\t\t\t{\n\t\t\t\tRole: \"user\",\n\t\t\t\tType: \"user_input\",\n\t\t\t\tProps: map[string]interface{}{\"content\": \"Hello\"},\n\t\t\t\tSequence: 1,\n\t\t\t\tConnector: \"openai\",\n\t\t\t\tAssistantID: \"test_assistant\",\n\t\t\t},\n\t\t\t{\n\t\t\t\tRole: \"assistant\",\n\t\t\t\tType: \"text\",\n\t\t\t\tProps: map[string]interface{}{\"content\": \"Hi there!\"},\n\t\t\t\tSequence: 2,\n\t\t\t\tConnector: \"openai\",\n\t\t\t\tAssistantID: \"test_assistant\",\n\t\t\t},\n\t\t\t{\n\t\t\t\tRole: \"user\",\n\t\t\t\tType: \"user_input\",\n\t\t\t\tProps: map[string]interface{}{\"content\": \"Switch to Claude\"},\n\t\t\t\tSequence: 3,\n\t\t\t\tConnector: \"anthropic\",\n\t\t\t\tAssistantID: \"test_assistant\",\n\t\t\t},\n\t\t\t{\n\t\t\t\tRole: \"assistant\",\n\t\t\t\tType: \"text\",\n\t\t\t\tProps: map[string]interface{}{\"content\": \"Now using Claude!\"},\n\t\t\t\tSequence: 4,\n\t\t\t\tConnector: \"anthropic\",\n\t\t\t\tAssistantID: \"test_assistant\",\n\t\t\t},\n\t\t}\n\n\t\terr = store.SaveMessages(connChat.ChatID, messages)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to save messages: %v\", err)\n\t\t}\n\n\t\t// Retrieve and verify connectors\n\t\tretrieved, err := store.GetMessages(connChat.ChatID, types.MessageFilter{})\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to get messages: %v\", err)\n\t\t}\n\n\t\tif len(retrieved) != 4 {\n\t\t\tt.Fatalf(\"Expected 4 messages, got %d\", len(retrieved))\n\t\t}\n\n\t\t// Verify each message has correct connector\n\t\tfor _, msg := range retrieved {\n\t\t\tif msg.Sequence <= 2 && msg.Connector != \"openai\" {\n\t\t\t\tt.Errorf(\"Expected connector 'openai' for sequence %d, got '%s'\", msg.Sequence, msg.Connector)\n\t\t\t}\n\t\t\tif msg.Sequence > 2 && msg.Connector != \"anthropic\" {\n\t\t\t\tt.Errorf(\"Expected connector 'anthropic' for sequence %d, got '%s'\", msg.Sequence, msg.Connector)\n\t\t\t}\n\t\t}\n\n\t\tt.Logf(\"Successfully saved and retrieved messages with different connectors\")\n\t})\n\n\tt.Run(\"SaveMessageWithMode\", func(t *testing.T) {\n\t\tmodeChat := &types.Chat{\n\t\t\tAssistantID: \"test_assistant\",\n\t\t}\n\t\terr := store.CreateChat(modeChat)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to create chat: %v\", err)\n\t\t}\n\t\tdefer store.DeleteChat(modeChat.ChatID)\n\n\t\t// Save messages with different modes\n\t\tmessages := []*types.Message{\n\t\t\t{\n\t\t\t\tRole: \"user\",\n\t\t\t\tType: \"user_input\",\n\t\t\t\tProps: map[string]interface{}{\"content\": \"Hello in chat mode\"},\n\t\t\t\tSequence: 1,\n\t\t\t\tMode: \"chat\",\n\t\t\t\tConnector: \"deepseek.v3\",\n\t\t\t\tAssistantID: \"test_assistant\",\n\t\t\t},\n\t\t\t{\n\t\t\t\tRole: \"assistant\",\n\t\t\t\tType: \"text\",\n\t\t\t\tProps: map[string]interface{}{\"content\": \"Hi there in chat mode!\"},\n\t\t\t\tSequence: 2,\n\t\t\t\tMode: \"chat\",\n\t\t\t\tConnector: \"deepseek.v3\",\n\t\t\t\tAssistantID: \"test_assistant\",\n\t\t\t},\n\t\t\t{\n\t\t\t\tRole: \"user\",\n\t\t\t\tType: \"user_input\",\n\t\t\t\tProps: map[string]interface{}{\"content\": \"Now run a task\"},\n\t\t\t\tSequence: 3,\n\t\t\t\tMode: \"task\",\n\t\t\t\tConnector: \"deepseek.v3\",\n\t\t\t\tAssistantID: \"test_assistant\",\n\t\t\t},\n\t\t\t{\n\t\t\t\tRole: \"assistant\",\n\t\t\t\tType: \"text\",\n\t\t\t\tProps: map[string]interface{}{\"content\": \"Running task!\"},\n\t\t\t\tSequence: 4,\n\t\t\t\tMode: \"task\",\n\t\t\t\tConnector: \"deepseek.v3\",\n\t\t\t\tAssistantID: \"test_assistant\",\n\t\t\t},\n\t\t}\n\n\t\terr = store.SaveMessages(modeChat.ChatID, messages)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to save messages: %v\", err)\n\t\t}\n\n\t\t// Retrieve and verify modes\n\t\tretrieved, err := store.GetMessages(modeChat.ChatID, types.MessageFilter{})\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to get messages: %v\", err)\n\t\t}\n\n\t\tif len(retrieved) != 4 {\n\t\t\tt.Fatalf(\"Expected 4 messages, got %d\", len(retrieved))\n\t\t}\n\n\t\t// Verify each message has correct mode\n\t\tfor _, msg := range retrieved {\n\t\t\tif msg.Sequence <= 2 && msg.Mode != \"chat\" {\n\t\t\t\tt.Errorf(\"Expected mode 'chat' for sequence %d, got '%s'\", msg.Sequence, msg.Mode)\n\t\t\t}\n\t\t\tif msg.Sequence > 2 && msg.Mode != \"task\" {\n\t\t\t\tt.Errorf(\"Expected mode 'task' for sequence %d, got '%s'\", msg.Sequence, msg.Mode)\n\t\t\t}\n\t\t}\n\n\t\tt.Logf(\"Successfully saved and retrieved messages with different modes\")\n\t})\n\n\tt.Run(\"SaveMessageWithEmptyConnector\", func(t *testing.T) {\n\t\temptyConnChat := &types.Chat{\n\t\t\tAssistantID: \"test_assistant\",\n\t\t}\n\t\terr := store.CreateChat(emptyConnChat)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to create chat: %v\", err)\n\t\t}\n\t\tdefer store.DeleteChat(emptyConnChat.ChatID)\n\n\t\t// Save message without connector\n\t\tmessages := []*types.Message{\n\t\t\t{\n\t\t\t\tRole: \"user\",\n\t\t\t\tType: \"text\",\n\t\t\t\tProps: map[string]interface{}{\"content\": \"No connector\"},\n\t\t\t\tSequence: 1,\n\t\t\t\t// Connector is empty\n\t\t\t},\n\t\t}\n\n\t\terr = store.SaveMessages(emptyConnChat.ChatID, messages)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to save message: %v\", err)\n\t\t}\n\n\t\tretrieved, err := store.GetMessages(emptyConnChat.ChatID, types.MessageFilter{})\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to get messages: %v\", err)\n\t\t}\n\n\t\tif len(retrieved) != 1 {\n\t\t\tt.Fatalf(\"Expected 1 message, got %d\", len(retrieved))\n\t\t}\n\n\t\t// Empty connector should be stored as empty string\n\t\tif retrieved[0].Connector != \"\" {\n\t\t\tt.Errorf(\"Expected empty connector, got '%s'\", retrieved[0].Connector)\n\t\t}\n\t})\n\n\tt.Run(\"SaveEmptyMessages\", func(t *testing.T) {\n\t\terr := store.SaveMessages(chat.ChatID, []*types.Message{})\n\t\tif err != nil {\n\t\t\tt.Errorf(\"Expected no error for empty messages, got: %v\", err)\n\t\t}\n\t})\n\n\tt.Run(\"SaveMessagesWithoutChatID\", func(t *testing.T) {\n\t\tmessages := []*types.Message{{Role: \"user\", Type: \"text\", Props: map[string]interface{}{\"content\": \"test\"}}}\n\t\terr := store.SaveMessages(\"\", messages)\n\t\tif err == nil {\n\t\t\tt.Error(\"Expected error when saving without chat_id\")\n\t\t}\n\t})\n\n\tt.Run(\"SaveMessageWithoutRole\", func(t *testing.T) {\n\t\tmessages := []*types.Message{{Type: \"text\", Props: map[string]interface{}{\"content\": \"test\"}, Sequence: 1}}\n\t\terr := store.SaveMessages(chat.ChatID, messages)\n\t\tif err == nil {\n\t\t\tt.Error(\"Expected error when saving message without role\")\n\t\t}\n\t})\n\n\tt.Run(\"SaveMessageWithoutType\", func(t *testing.T) {\n\t\tmessages := []*types.Message{{Role: \"user\", Props: map[string]interface{}{\"content\": \"test\"}, Sequence: 1}}\n\t\terr := store.SaveMessages(chat.ChatID, messages)\n\t\tif err == nil {\n\t\t\tt.Error(\"Expected error when saving message without type\")\n\t\t}\n\t})\n\n\tt.Run(\"SaveMessageWithoutProps\", func(t *testing.T) {\n\t\tmessages := []*types.Message{{Role: \"user\", Type: \"text\", Sequence: 1}}\n\t\terr := store.SaveMessages(chat.ChatID, messages)\n\t\tif err == nil {\n\t\t\tt.Error(\"Expected error when saving message without props\")\n\t\t}\n\t})\n}\n\n// TestGetMessages tests retrieving messages with filters\nfunc TestGetMessages(t *testing.T) {\n\ttest.Prepare(t, config.Conf)\n\tdefer test.Clean()\n\n\tstore, err := xun.NewXun(types.Setting{\n\t\tConnector: \"default\",\n\t})\n\tif err != nil {\n\t\tt.Fatalf(\"Failed to create store: %v\", err)\n\t}\n\n\t// Create chat and messages\n\tchat := &types.Chat{\n\t\tAssistantID: \"test_assistant\",\n\t}\n\terr = store.CreateChat(chat)\n\tif err != nil {\n\t\tt.Fatalf(\"Failed to create chat: %v\", err)\n\t}\n\tdefer store.DeleteChat(chat.ChatID)\n\n\t// Save test messages\n\tmessages := []*types.Message{\n\t\t{Role: \"user\", Type: \"user_input\", Props: map[string]interface{}{\"content\": \"Hello\"}, Sequence: 1, RequestID: \"req_001\"},\n\t\t{Role: \"assistant\", Type: \"text\", Props: map[string]interface{}{\"content\": \"Hi there!\"}, Sequence: 2, RequestID: \"req_001\", BlockID: \"B1\"},\n\t\t{Role: \"user\", Type: \"user_input\", Props: map[string]interface{}{\"content\": \"Weather?\"}, Sequence: 3, RequestID: \"req_002\"},\n\t\t{Role: \"assistant\", Type: \"loading\", Props: map[string]interface{}{\"message\": \"Checking...\"}, Sequence: 4, RequestID: \"req_002\", BlockID: \"B2\"},\n\t\t{Role: \"assistant\", Type: \"text\", Props: map[string]interface{}{\"content\": \"Sunny!\"}, Sequence: 5, RequestID: \"req_002\", BlockID: \"B2\", ThreadID: \"T1\"},\n\t}\n\terr = store.SaveMessages(chat.ChatID, messages)\n\tif err != nil {\n\t\tt.Fatalf(\"Failed to save messages: %v\", err)\n\t}\n\n\tt.Run(\"GetAllMessages\", func(t *testing.T) {\n\t\tretrieved, err := store.GetMessages(chat.ChatID, types.MessageFilter{})\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to get messages: %v\", err)\n\t\t}\n\n\t\tif len(retrieved) != 5 {\n\t\t\tt.Errorf(\"Expected 5 messages, got %d\", len(retrieved))\n\t\t}\n\n\t\t// Verify order by sequence\n\t\tfor i := 1; i < len(retrieved); i++ {\n\t\t\tif retrieved[i].Sequence < retrieved[i-1].Sequence {\n\t\t\t\tt.Error(\"Messages not ordered by sequence\")\n\t\t\t}\n\t\t}\n\t})\n\n\tt.Run(\"FilterByRole\", func(t *testing.T) {\n\t\tretrieved, err := store.GetMessages(chat.ChatID, types.MessageFilter{Role: \"user\"})\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to get messages: %v\", err)\n\t\t}\n\n\t\tif len(retrieved) != 2 {\n\t\t\tt.Errorf(\"Expected 2 user messages, got %d\", len(retrieved))\n\t\t}\n\n\t\tfor _, msg := range retrieved {\n\t\t\tif msg.Role != \"user\" {\n\t\t\t\tt.Errorf(\"Expected role 'user', got '%s'\", msg.Role)\n\t\t\t}\n\t\t}\n\t})\n\n\tt.Run(\"FilterByRequestID\", func(t *testing.T) {\n\t\tretrieved, err := store.GetMessages(chat.ChatID, types.MessageFilter{RequestID: \"req_002\"})\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to get messages: %v\", err)\n\t\t}\n\n\t\tif len(retrieved) != 3 {\n\t\t\tt.Errorf(\"Expected 3 messages for req_002, got %d\", len(retrieved))\n\t\t}\n\t})\n\n\tt.Run(\"FilterByBlockID\", func(t *testing.T) {\n\t\tretrieved, err := store.GetMessages(chat.ChatID, types.MessageFilter{BlockID: \"B2\"})\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to get messages: %v\", err)\n\t\t}\n\n\t\tif len(retrieved) != 2 {\n\t\t\tt.Errorf(\"Expected 2 messages in block B2, got %d\", len(retrieved))\n\t\t}\n\t})\n\n\tt.Run(\"FilterByThreadID\", func(t *testing.T) {\n\t\tretrieved, err := store.GetMessages(chat.ChatID, types.MessageFilter{ThreadID: \"T1\"})\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to get messages: %v\", err)\n\t\t}\n\n\t\tif len(retrieved) != 1 {\n\t\t\tt.Errorf(\"Expected 1 message in thread T1, got %d\", len(retrieved))\n\t\t}\n\t})\n\n\tt.Run(\"FilterByType\", func(t *testing.T) {\n\t\tretrieved, err := store.GetMessages(chat.ChatID, types.MessageFilter{Type: \"loading\"})\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to get messages: %v\", err)\n\t\t}\n\n\t\tif len(retrieved) != 1 {\n\t\t\tt.Errorf(\"Expected 1 loading message, got %d\", len(retrieved))\n\t\t}\n\t})\n\n\tt.Run(\"FilterWithLimit\", func(t *testing.T) {\n\t\tretrieved, err := store.GetMessages(chat.ChatID, types.MessageFilter{Limit: 2})\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to get messages: %v\", err)\n\t\t}\n\n\t\tif len(retrieved) != 2 {\n\t\t\tt.Errorf(\"Expected 2 messages with limit, got %d\", len(retrieved))\n\t\t}\n\t})\n\n\tt.Run(\"FilterWithOffset\", func(t *testing.T) {\n\t\tretrieved, err := store.GetMessages(chat.ChatID, types.MessageFilter{Offset: 3})\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to get messages: %v\", err)\n\t\t}\n\n\t\tif len(retrieved) != 2 {\n\t\t\tt.Errorf(\"Expected 2 messages with offset 3, got %d\", len(retrieved))\n\t\t}\n\t})\n\n\tt.Run(\"FilterWithLimitAndOffset\", func(t *testing.T) {\n\t\tretrieved, err := store.GetMessages(chat.ChatID, types.MessageFilter{Limit: 2, Offset: 1})\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to get messages: %v\", err)\n\t\t}\n\n\t\tif len(retrieved) != 2 {\n\t\t\tt.Errorf(\"Expected 2 messages, got %d\", len(retrieved))\n\t\t}\n\n\t\t// Should be sequence 2 and 3\n\t\tif len(retrieved) >= 2 {\n\t\t\tif retrieved[0].Sequence != 2 {\n\t\t\t\tt.Errorf(\"Expected first message sequence 2, got %d\", retrieved[0].Sequence)\n\t\t\t}\n\t\t}\n\t})\n\n\tt.Run(\"GetMessagesWithEmptyChatID\", func(t *testing.T) {\n\t\t_, err := store.GetMessages(\"\", types.MessageFilter{})\n\t\tif err == nil {\n\t\t\tt.Error(\"Expected error when getting messages without chat_id\")\n\t\t}\n\t})\n\n\tt.Run(\"GetMessagesFromNonExistentChat\", func(t *testing.T) {\n\t\tretrieved, err := store.GetMessages(\"nonexistent_chat\", types.MessageFilter{})\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Unexpected error: %v\", err)\n\t\t}\n\t\tif len(retrieved) != 0 {\n\t\t\tt.Errorf(\"Expected 0 messages from non-existent chat, got %d\", len(retrieved))\n\t\t}\n\t})\n\n\tt.Run(\"OrderByCreatedAtThenSequence\", func(t *testing.T) {\n\t\t// This test verifies that messages are ordered by created_at first, then by sequence\n\t\t// This is important when there are multiple request_ids with overlapping sequence numbers\n\t\torderChat := &types.Chat{\n\t\t\tAssistantID: \"test_assistant\",\n\t\t}\n\t\terr := store.CreateChat(orderChat)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to create chat: %v\", err)\n\t\t}\n\t\tdefer store.DeleteChat(orderChat.ChatID)\n\n\t\t// Simulate two separate requests with overlapping sequence numbers\n\t\t// SaveMessages uses time.Now() for created_at, so we need to call it twice\n\t\t// with a small delay to ensure different timestamps\n\n\t\t// Request 1: sequences 1, 2\n\t\treq1Messages := []*types.Message{\n\t\t\t{\n\t\t\t\tRole: \"user\",\n\t\t\t\tType: \"user_input\",\n\t\t\t\tProps: map[string]interface{}{\"content\": \"Request 1 - Message 1\"},\n\t\t\t\tSequence: 1,\n\t\t\t\tRequestID: \"order_req_001\",\n\t\t\t},\n\t\t\t{\n\t\t\t\tRole: \"assistant\",\n\t\t\t\tType: \"text\",\n\t\t\t\tProps: map[string]interface{}{\"content\": \"Request 1 - Response 1\"},\n\t\t\t\tSequence: 2,\n\t\t\t\tRequestID: \"order_req_001\",\n\t\t\t},\n\t\t}\n\n\t\terr = store.SaveMessages(orderChat.ChatID, req1Messages)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to save request 1 messages: %v\", err)\n\t\t}\n\n\t\t// Delay to ensure different created_at timestamps\n\t\t// Database timestamp precision may only be to second level\n\t\ttime.Sleep(1100 * time.Millisecond)\n\n\t\t// Request 2: sequences 1, 2 (same as request 1, but later created_at)\n\t\treq2Messages := []*types.Message{\n\t\t\t{\n\t\t\t\tRole: \"user\",\n\t\t\t\tType: \"user_input\",\n\t\t\t\tProps: map[string]interface{}{\"content\": \"Request 2 - Message 1\"},\n\t\t\t\tSequence: 1, // Same sequence as req1, but later created_at\n\t\t\t\tRequestID: \"order_req_002\",\n\t\t\t},\n\t\t\t{\n\t\t\t\tRole: \"assistant\",\n\t\t\t\tType: \"text\",\n\t\t\t\tProps: map[string]interface{}{\"content\": \"Request 2 - Response 1\"},\n\t\t\t\tSequence: 2, // Same sequence as req1, but later created_at\n\t\t\t\tRequestID: \"order_req_002\",\n\t\t\t},\n\t\t}\n\n\t\terr = store.SaveMessages(orderChat.ChatID, req2Messages)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to save request 2 messages: %v\", err)\n\t\t}\n\n\t\t// Retrieve messages\n\t\tretrieved, err := store.GetMessages(orderChat.ChatID, types.MessageFilter{})\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to get messages: %v\", err)\n\t\t}\n\n\t\tif len(retrieved) != 4 {\n\t\t\tt.Fatalf(\"Expected 4 messages, got %d\", len(retrieved))\n\t\t}\n\n\t\t// Verify order: should be chronological by created_at, then by sequence\n\t\t// Messages from req_001 should come before req_002\n\t\texpectedOrder := []struct {\n\t\t\trequestID string\n\t\t\tsequence int\n\t\t\tcontent string\n\t\t}{\n\t\t\t{\"order_req_001\", 1, \"Request 1 - Message 1\"},\n\t\t\t{\"order_req_001\", 2, \"Request 1 - Response 1\"},\n\t\t\t{\"order_req_002\", 1, \"Request 2 - Message 1\"},\n\t\t\t{\"order_req_002\", 2, \"Request 2 - Response 1\"},\n\t\t}\n\n\t\tfor i, expected := range expectedOrder {\n\t\t\tmsg := retrieved[i]\n\t\t\tif msg.RequestID != expected.requestID {\n\t\t\t\tt.Errorf(\"Message %d: expected RequestID '%s', got '%s'\", i, expected.requestID, msg.RequestID)\n\t\t\t}\n\t\t\tif msg.Sequence != expected.sequence {\n\t\t\t\tt.Errorf(\"Message %d: expected Sequence %d, got %d\", i, expected.sequence, msg.Sequence)\n\t\t\t}\n\t\t\tcontent, _ := msg.Props[\"content\"].(string)\n\t\t\tif content != expected.content {\n\t\t\t\tt.Errorf(\"Message %d: expected content '%s', got '%s'\", i, expected.content, content)\n\t\t\t}\n\t\t}\n\n\t\t// Additional verification: ensure created_at is non-decreasing\n\t\tfor i := 1; i < len(retrieved); i++ {\n\t\t\tif retrieved[i].CreatedAt.Before(retrieved[i-1].CreatedAt) {\n\t\t\t\tt.Errorf(\"Message %d created_at (%v) is before message %d created_at (%v)\",\n\t\t\t\t\ti, retrieved[i].CreatedAt, i-1, retrieved[i-1].CreatedAt)\n\t\t\t}\n\t\t\t// If same created_at, sequence should be increasing\n\t\t\tif retrieved[i].CreatedAt.Equal(retrieved[i-1].CreatedAt) {\n\t\t\t\tif retrieved[i].Sequence < retrieved[i-1].Sequence {\n\t\t\t\t\tt.Errorf(\"Messages with same created_at: message %d sequence (%d) < message %d sequence (%d)\",\n\t\t\t\t\t\ti, retrieved[i].Sequence, i-1, retrieved[i-1].Sequence)\n\t\t\t\t}\n\t\t\t}\n\t\t}\n\n\t\tt.Logf(\"Successfully verified message ordering: created_at first, then sequence\")\n\t})\n}\n\n// TestUpdateMessage tests updating messages\nfunc TestUpdateMessage(t *testing.T) {\n\ttest.Prepare(t, config.Conf)\n\tdefer test.Clean()\n\n\tstore, err := xun.NewXun(types.Setting{\n\t\tConnector: \"default\",\n\t})\n\tif err != nil {\n\t\tt.Fatalf(\"Failed to create store: %v\", err)\n\t}\n\n\t// Create chat and message\n\tchat := &types.Chat{\n\t\tAssistantID: \"test_assistant\",\n\t}\n\terr = store.CreateChat(chat)\n\tif err != nil {\n\t\tt.Fatalf(\"Failed to create chat: %v\", err)\n\t}\n\tdefer store.DeleteChat(chat.ChatID)\n\n\tmessages := []*types.Message{\n\t\t{\n\t\t\tMessageID: fmt.Sprintf(\"msg_%d\", time.Now().UnixNano()),\n\t\t\tRole: \"assistant\",\n\t\t\tType: \"loading\",\n\t\t\tProps: map[string]interface{}{\"message\": \"Loading...\"},\n\t\t\tSequence: 1,\n\t\t},\n\t}\n\terr = store.SaveMessages(chat.ChatID, messages)\n\tif err != nil {\n\t\tt.Fatalf(\"Failed to save message: %v\", err)\n\t}\n\n\tmessageID := messages[0].MessageID\n\n\tt.Run(\"UpdateProps\", func(t *testing.T) {\n\t\terr := store.UpdateMessage(messageID, map[string]interface{}{\n\t\t\t\"props\": map[string]interface{}{\"content\": \"Updated content\"},\n\t\t})\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to update message: %v\", err)\n\t\t}\n\n\t\tretrieved, err := store.GetMessages(chat.ChatID, types.MessageFilter{})\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to get messages: %v\", err)\n\t\t}\n\n\t\tvar found *types.Message\n\t\tfor _, msg := range retrieved {\n\t\t\tif msg.MessageID == messageID {\n\t\t\t\tfound = msg\n\t\t\t\tbreak\n\t\t\t}\n\t\t}\n\n\t\tif found == nil {\n\t\t\tt.Fatal(\"Could not find updated message\")\n\t\t}\n\n\t\tif found.Props[\"content\"] != \"Updated content\" {\n\t\t\tt.Errorf(\"Expected props content 'Updated content', got '%v'\", found.Props[\"content\"])\n\t\t}\n\t})\n\n\tt.Run(\"UpdateType\", func(t *testing.T) {\n\t\terr := store.UpdateMessage(messageID, map[string]interface{}{\n\t\t\t\"type\": \"text\",\n\t\t})\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to update message: %v\", err)\n\t\t}\n\n\t\tretrieved, err := store.GetMessages(chat.ChatID, types.MessageFilter{})\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to get messages: %v\", err)\n\t\t}\n\n\t\tvar found *types.Message\n\t\tfor _, msg := range retrieved {\n\t\t\tif msg.MessageID == messageID {\n\t\t\t\tfound = msg\n\t\t\t\tbreak\n\t\t\t}\n\t\t}\n\n\t\tif found == nil {\n\t\t\tt.Fatal(\"Could not find updated message\")\n\t\t}\n\n\t\tif found.Type != \"text\" {\n\t\t\tt.Errorf(\"Expected type 'text', got '%s'\", found.Type)\n\t\t}\n\t})\n\n\tt.Run(\"UpdateMetadata\", func(t *testing.T) {\n\t\terr := store.UpdateMessage(messageID, map[string]interface{}{\n\t\t\t\"metadata\": map[string]interface{}{\"updated\": true},\n\t\t})\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to update metadata: %v\", err)\n\t\t}\n\n\t\tretrieved, err := store.GetMessages(chat.ChatID, types.MessageFilter{})\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to get messages: %v\", err)\n\t\t}\n\n\t\tvar found *types.Message\n\t\tfor _, msg := range retrieved {\n\t\t\tif msg.MessageID == messageID {\n\t\t\t\tfound = msg\n\t\t\t\tbreak\n\t\t\t}\n\t\t}\n\n\t\tif found == nil {\n\t\t\tt.Fatal(\"Could not find updated message\")\n\t\t}\n\n\t\tif found.Metadata == nil || found.Metadata[\"updated\"] != true {\n\t\t\tt.Errorf(\"Expected metadata updated=true, got %v\", found.Metadata)\n\t\t}\n\t})\n\n\tt.Run(\"UpdateNonExistentMessage\", func(t *testing.T) {\n\t\terr := store.UpdateMessage(\"nonexistent_msg\", map[string]interface{}{\n\t\t\t\"type\": \"text\",\n\t\t})\n\t\tif err == nil {\n\t\t\tt.Error(\"Expected error when updating non-existent message\")\n\t\t}\n\t})\n\n\tt.Run(\"UpdateWithEmptyID\", func(t *testing.T) {\n\t\terr := store.UpdateMessage(\"\", map[string]interface{}{\n\t\t\t\"type\": \"text\",\n\t\t})\n\t\tif err == nil {\n\t\t\tt.Error(\"Expected error when updating with empty ID\")\n\t\t}\n\t})\n\n\tt.Run(\"UpdateWithEmptyFields\", func(t *testing.T) {\n\t\terr := store.UpdateMessage(messageID, map[string]interface{}{})\n\t\tif err == nil {\n\t\t\tt.Error(\"Expected error when updating with empty fields\")\n\t\t}\n\t})\n}\n\n// TestDeleteMessages tests deleting messages\nfunc TestDeleteMessages(t *testing.T) {\n\ttest.Prepare(t, config.Conf)\n\tdefer test.Clean()\n\n\tstore, err := xun.NewXun(types.Setting{\n\t\tConnector: \"default\",\n\t})\n\tif err != nil {\n\t\tt.Fatalf(\"Failed to create store: %v\", err)\n\t}\n\n\tt.Run(\"DeleteSingleMessage\", func(t *testing.T) {\n\t\tchat := &types.Chat{AssistantID: \"test_assistant\"}\n\t\terr := store.CreateChat(chat)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to create chat: %v\", err)\n\t\t}\n\t\tdefer store.DeleteChat(chat.ChatID)\n\n\t\tmsgID := fmt.Sprintf(\"msg_del_%d\", time.Now().UnixNano())\n\t\tmessages := []*types.Message{\n\t\t\t{MessageID: msgID, Role: \"user\", Type: \"text\", Props: map[string]interface{}{\"content\": \"test\"}, Sequence: 1},\n\t\t}\n\t\terr = store.SaveMessages(chat.ChatID, messages)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to save message: %v\", err)\n\t\t}\n\n\t\terr = store.DeleteMessages(chat.ChatID, []string{msgID})\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to delete message: %v\", err)\n\t\t}\n\n\t\t// Verify deleted\n\t\tretrieved, err := store.GetMessages(chat.ChatID, types.MessageFilter{})\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to get messages: %v\", err)\n\t\t}\n\n\t\tfor _, msg := range retrieved {\n\t\t\tif msg.MessageID == msgID {\n\t\t\t\tt.Error(\"Message should have been deleted\")\n\t\t\t}\n\t\t}\n\t})\n\n\tt.Run(\"DeleteMultipleMessages\", func(t *testing.T) {\n\t\tchat := &types.Chat{AssistantID: \"test_assistant\"}\n\t\terr := store.CreateChat(chat)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to create chat: %v\", err)\n\t\t}\n\t\tdefer store.DeleteChat(chat.ChatID)\n\n\t\tmsgID1 := fmt.Sprintf(\"msg_del1_%d\", time.Now().UnixNano())\n\t\tmsgID2 := fmt.Sprintf(\"msg_del2_%d\", time.Now().UnixNano())\n\t\tmsgID3 := fmt.Sprintf(\"msg_del3_%d\", time.Now().UnixNano())\n\n\t\tmessages := []*types.Message{\n\t\t\t{MessageID: msgID1, Role: \"user\", Type: \"text\", Props: map[string]interface{}{\"content\": \"1\"}, Sequence: 1},\n\t\t\t{MessageID: msgID2, Role: \"assistant\", Type: \"text\", Props: map[string]interface{}{\"content\": \"2\"}, Sequence: 2},\n\t\t\t{MessageID: msgID3, Role: \"user\", Type: \"text\", Props: map[string]interface{}{\"content\": \"3\"}, Sequence: 3},\n\t\t}\n\t\terr = store.SaveMessages(chat.ChatID, messages)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to save messages: %v\", err)\n\t\t}\n\n\t\t// Delete first two\n\t\terr = store.DeleteMessages(chat.ChatID, []string{msgID1, msgID2})\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to delete messages: %v\", err)\n\t\t}\n\n\t\t// Verify\n\t\tretrieved, err := store.GetMessages(chat.ChatID, types.MessageFilter{})\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to get messages: %v\", err)\n\t\t}\n\n\t\tif len(retrieved) != 1 {\n\t\t\tt.Errorf(\"Expected 1 remaining message, got %d\", len(retrieved))\n\t\t}\n\n\t\tif len(retrieved) > 0 && retrieved[0].MessageID != msgID3 {\n\t\t\tt.Errorf(\"Expected remaining message to be %s, got %s\", msgID3, retrieved[0].MessageID)\n\t\t}\n\t})\n\n\tt.Run(\"DeleteEmptyList\", func(t *testing.T) {\n\t\tchat := &types.Chat{AssistantID: \"test_assistant\"}\n\t\terr := store.CreateChat(chat)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to create chat: %v\", err)\n\t\t}\n\t\tdefer store.DeleteChat(chat.ChatID)\n\n\t\terr = store.DeleteMessages(chat.ChatID, []string{})\n\t\tif err != nil {\n\t\t\tt.Errorf(\"Expected no error for empty delete list, got: %v\", err)\n\t\t}\n\t})\n\n\tt.Run(\"DeleteWithEmptyChatID\", func(t *testing.T) {\n\t\terr := store.DeleteMessages(\"\", []string{\"msg_123\"})\n\t\tif err == nil {\n\t\t\tt.Error(\"Expected error when deleting with empty chat_id\")\n\t\t}\n\t})\n}\n\n// TestMessageCompleteWorkflow tests a complete message workflow\nfunc TestMessageCompleteWorkflow(t *testing.T) {\n\ttest.Prepare(t, config.Conf)\n\tdefer test.Clean()\n\n\tstore, err := xun.NewXun(types.Setting{\n\t\tConnector: \"default\",\n\t})\n\tif err != nil {\n\t\tt.Fatalf(\"Failed to create store: %v\", err)\n\t}\n\n\tt.Run(\"CompleteWorkflow\", func(t *testing.T) {\n\t\t// 1. Create chat\n\t\tchat := &types.Chat{\n\t\t\tAssistantID: \"workflow_assistant\",\n\t\t\tTitle: \"Message Workflow Test\",\n\t\t}\n\t\terr := store.CreateChat(chat)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to create chat: %v\", err)\n\t\t}\n\t\tdefer store.DeleteChat(chat.ChatID)\n\n\t\t// 2. Save batch messages (simulating a request)\n\t\trequestID := fmt.Sprintf(\"req_%d\", time.Now().UnixNano())\n\t\tmessages := []*types.Message{\n\t\t\t{\n\t\t\t\tRole: \"user\",\n\t\t\t\tType: \"user_input\",\n\t\t\t\tProps: map[string]interface{}{\"content\": \"What's the weather in SF?\"},\n\t\t\t\tSequence: 1,\n\t\t\t\tRequestID: requestID,\n\t\t\t\tAssistantID: \"workflow_assistant\",\n\t\t\t},\n\t\t\t{\n\t\t\t\tRole: \"assistant\",\n\t\t\t\tType: \"loading\",\n\t\t\t\tProps: map[string]interface{}{\"message\": \"Checking weather...\"},\n\t\t\t\tSequence: 2,\n\t\t\t\tRequestID: requestID,\n\t\t\t\tBlockID: \"B1\",\n\t\t\t\tAssistantID: \"workflow_assistant\",\n\t\t\t},\n\t\t\t{\n\t\t\t\tRole: \"assistant\",\n\t\t\t\tType: \"tool_call\",\n\t\t\t\tProps: map[string]interface{}{\"id\": \"call_weather\", \"name\": \"get_weather\", \"arguments\": `{\"location\":\"SF\"}`},\n\t\t\t\tSequence: 3,\n\t\t\t\tRequestID: requestID,\n\t\t\t\tBlockID: \"B1\",\n\t\t\t\tAssistantID: \"workflow_assistant\",\n\t\t\t},\n\t\t\t{\n\t\t\t\tRole: \"assistant\",\n\t\t\t\tType: \"text\",\n\t\t\t\tProps: map[string]interface{}{\"content\": \"The weather in San Francisco is 18°C and sunny.\"},\n\t\t\t\tSequence: 4,\n\t\t\t\tRequestID: requestID,\n\t\t\t\tBlockID: \"B1\",\n\t\t\t\tAssistantID: \"workflow_assistant\",\n\t\t\t\tMetadata: map[string]interface{}{\"tool_call_id\": \"call_weather\"},\n\t\t\t},\n\t\t}\n\n\t\terr = store.SaveMessages(chat.ChatID, messages)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to save messages: %v\", err)\n\t\t}\n\t\tt.Logf(\"Saved %d messages in single batch\", len(messages))\n\n\t\t// 3. Get all messages\n\t\tretrieved, err := store.GetMessages(chat.ChatID, types.MessageFilter{})\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to get messages: %v\", err)\n\t\t}\n\n\t\tif len(retrieved) != 4 {\n\t\t\tt.Errorf(\"Expected 4 messages, got %d\", len(retrieved))\n\t\t}\n\n\t\t// 4. Filter by request\n\t\tbyRequest, err := store.GetMessages(chat.ChatID, types.MessageFilter{RequestID: requestID})\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to filter by request: %v\", err)\n\t\t}\n\n\t\tif len(byRequest) != 4 {\n\t\t\tt.Errorf(\"Expected 4 messages for request, got %d\", len(byRequest))\n\t\t}\n\n\t\t// 5. Filter by block\n\t\tbyBlock, err := store.GetMessages(chat.ChatID, types.MessageFilter{BlockID: \"B1\"})\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to filter by block: %v\", err)\n\t\t}\n\n\t\tif len(byBlock) != 3 {\n\t\t\tt.Errorf(\"Expected 3 messages in block B1, got %d\", len(byBlock))\n\t\t}\n\n\t\t// 6. Update loading message to text (simulating stream completion)\n\t\tvar loadingMsgID string\n\t\tfor _, msg := range retrieved {\n\t\t\tif msg.Type == \"loading\" {\n\t\t\t\tloadingMsgID = msg.MessageID\n\t\t\t\tbreak\n\t\t\t}\n\t\t}\n\n\t\tif loadingMsgID != \"\" {\n\t\t\terr = store.UpdateMessage(loadingMsgID, map[string]interface{}{\n\t\t\t\t\"type\": \"text\",\n\t\t\t\t\"props\": map[string]interface{}{\"content\": \"Weather check complete.\"},\n\t\t\t})\n\t\t\tif err != nil {\n\t\t\t\tt.Fatalf(\"Failed to update message: %v\", err)\n\t\t\t}\n\t\t}\n\n\t\t// 7. Delete a message\n\t\tif len(retrieved) > 0 {\n\t\t\terr = store.DeleteMessages(chat.ChatID, []string{retrieved[0].MessageID})\n\t\t\tif err != nil {\n\t\t\t\tt.Fatalf(\"Failed to delete message: %v\", err)\n\t\t\t}\n\t\t}\n\n\t\t// 8. Verify final state\n\t\tfinal, err := store.GetMessages(chat.ChatID, types.MessageFilter{})\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to get final messages: %v\", err)\n\t\t}\n\n\t\tif len(final) != 3 {\n\t\t\tt.Errorf(\"Expected 3 messages after delete, got %d\", len(final))\n\t\t}\n\n\t\tt.Log(\"Complete message workflow passed!\")\n\t})\n}\n\n// TestConcurrentMessages tests concurrent message storage\nfunc TestConcurrentMessages(t *testing.T) {\n\ttest.Prepare(t, config.Conf)\n\tdefer test.Clean()\n\n\tstore, err := xun.NewXun(types.Setting{\n\t\tConnector: \"default\",\n\t})\n\tif err != nil {\n\t\tt.Fatalf(\"Failed to create store: %v\", err)\n\t}\n\n\tt.Run(\"ConcurrentThreadMessages\", func(t *testing.T) {\n\t\tchat := &types.Chat{AssistantID: \"test_assistant\"}\n\t\terr := store.CreateChat(chat)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to create chat: %v\", err)\n\t\t}\n\t\tdefer store.DeleteChat(chat.ChatID)\n\n\t\t// Simulate concurrent operations with different threads\n\t\tmessages := []*types.Message{\n\t\t\t{Role: \"assistant\", Type: \"text\", Props: map[string]interface{}{\"content\": \"Weather result\"}, Sequence: 1, BlockID: \"B1\", ThreadID: \"T1\"},\n\t\t\t{Role: \"assistant\", Type: \"text\", Props: map[string]interface{}{\"content\": \"News result\"}, Sequence: 2, BlockID: \"B1\", ThreadID: \"T2\"},\n\t\t\t{Role: \"assistant\", Type: \"text\", Props: map[string]interface{}{\"content\": \"Stock result\"}, Sequence: 3, BlockID: \"B1\", ThreadID: \"T3\"},\n\t\t\t{Role: \"assistant\", Type: \"text\", Props: map[string]interface{}{\"content\": \"Summary\"}, Sequence: 4, BlockID: \"B2\"},\n\t\t}\n\n\t\terr = store.SaveMessages(chat.ChatID, messages)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to save concurrent messages: %v\", err)\n\t\t}\n\n\t\t// Verify all saved\n\t\tall, err := store.GetMessages(chat.ChatID, types.MessageFilter{})\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to get messages: %v\", err)\n\t\t}\n\n\t\tif len(all) != 4 {\n\t\t\tt.Errorf(\"Expected 4 messages, got %d\", len(all))\n\t\t}\n\n\t\t// Filter by thread\n\t\tt1Messages, err := store.GetMessages(chat.ChatID, types.MessageFilter{ThreadID: \"T1\"})\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to filter by thread: %v\", err)\n\t\t}\n\n\t\tif len(t1Messages) != 1 {\n\t\t\tt.Errorf(\"Expected 1 message in thread T1, got %d\", len(t1Messages))\n\t\t}\n\n\t\t// Filter by block\n\t\tb1Messages, err := store.GetMessages(chat.ChatID, types.MessageFilter{BlockID: \"B1\"})\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to filter by block: %v\", err)\n\t\t}\n\n\t\tif len(b1Messages) != 3 {\n\t\t\tt.Errorf(\"Expected 3 messages in block B1, got %d\", len(b1Messages))\n\t\t}\n\n\t\tt.Log(\"Concurrent thread messages test passed!\")\n\t})\n}\n" }, { "path": "agent/store/xun/resume.go", "content": "package xun\n\nimport (\n\t\"fmt\"\n\t\"time\"\n\n\t\"github.com/google/uuid\"\n\tjsoniter \"github.com/json-iterator/go\"\n\t\"github.com/yaoapp/yao/agent/store/types\"\n)\n\n// =============================================================================\n// Resume Management (only called on failure/interrupt)\n// =============================================================================\n\n// SaveResume batch saves resume records using a single database call\n// Only called when request is interrupted or failed\nfunc (store *Xun) SaveResume(records []*types.Resume) error {\n\tif len(records) == 0 {\n\t\treturn nil // Nothing to save\n\t}\n\n\t// Prepare batch insert data\n\tnow := time.Now()\n\trows := make([]map[string]interface{}, 0, len(records))\n\n\tfor _, record := range records {\n\t\tif record == nil {\n\t\t\tcontinue\n\t\t}\n\n\t\t// Generate resume_id if not provided\n\t\tresumeID := record.ResumeID\n\t\tif resumeID == \"\" {\n\t\t\tresumeID = uuid.New().String()\n\t\t}\n\n\t\t// Validate required fields\n\t\tif record.ChatID == \"\" {\n\t\t\treturn fmt.Errorf(\"chat_id is required\")\n\t\t}\n\t\tif record.RequestID == \"\" {\n\t\t\treturn fmt.Errorf(\"request_id is required\")\n\t\t}\n\t\tif record.AssistantID == \"\" {\n\t\t\treturn fmt.Errorf(\"assistant_id is required\")\n\t\t}\n\t\tif record.StackID == \"\" {\n\t\t\treturn fmt.Errorf(\"stack_id is required\")\n\t\t}\n\t\tif record.Type == \"\" {\n\t\t\treturn fmt.Errorf(\"type is required\")\n\t\t}\n\t\tif record.Status == \"\" {\n\t\t\treturn fmt.Errorf(\"status is required\")\n\t\t}\n\n\t\t// Build row with all fields (including nullable ones for consistent batch insert)\n\t\trow := map[string]interface{}{\n\t\t\t\"resume_id\": resumeID,\n\t\t\t\"chat_id\": record.ChatID,\n\t\t\t\"request_id\": record.RequestID,\n\t\t\t\"assistant_id\": record.AssistantID,\n\t\t\t\"stack_id\": record.StackID,\n\t\t\t\"stack_parent_id\": nil,\n\t\t\t\"stack_depth\": record.StackDepth,\n\t\t\t\"type\": record.Type,\n\t\t\t\"status\": record.Status,\n\t\t\t\"input\": nil,\n\t\t\t\"output\": nil,\n\t\t\t\"space_snapshot\": nil,\n\t\t\t\"error\": nil,\n\t\t\t\"sequence\": record.Sequence,\n\t\t\t\"metadata\": nil,\n\t\t\t\"created_at\": now,\n\t\t\t\"updated_at\": now,\n\t\t}\n\n\t\t// Set nullable fields if they have values\n\t\tif record.StackParentID != \"\" {\n\t\t\trow[\"stack_parent_id\"] = record.StackParentID\n\t\t}\n\t\tif record.Input != nil {\n\t\t\tinputJSON, err := jsoniter.MarshalToString(record.Input)\n\t\t\tif err != nil {\n\t\t\t\treturn fmt.Errorf(\"failed to marshal input: %w\", err)\n\t\t\t}\n\t\t\trow[\"input\"] = inputJSON\n\t\t}\n\t\tif record.Output != nil {\n\t\t\toutputJSON, err := jsoniter.MarshalToString(record.Output)\n\t\t\tif err != nil {\n\t\t\t\treturn fmt.Errorf(\"failed to marshal output: %w\", err)\n\t\t\t}\n\t\t\trow[\"output\"] = outputJSON\n\t\t}\n\t\tif record.SpaceSnapshot != nil {\n\t\t\tsnapshotJSON, err := jsoniter.MarshalToString(record.SpaceSnapshot)\n\t\t\tif err != nil {\n\t\t\t\treturn fmt.Errorf(\"failed to marshal space_snapshot: %w\", err)\n\t\t\t}\n\t\t\trow[\"space_snapshot\"] = snapshotJSON\n\t\t}\n\t\tif record.Error != \"\" {\n\t\t\trow[\"error\"] = record.Error\n\t\t}\n\t\tif record.Metadata != nil {\n\t\t\tmetadataJSON, err := jsoniter.MarshalToString(record.Metadata)\n\t\t\tif err != nil {\n\t\t\t\treturn fmt.Errorf(\"failed to marshal metadata: %w\", err)\n\t\t\t}\n\t\t\trow[\"metadata\"] = metadataJSON\n\t\t}\n\n\t\trows = append(rows, row)\n\t}\n\n\tif len(rows) == 0 {\n\t\treturn nil\n\t}\n\n\t// Single batch insert - one database call for all records\n\treturn store.newQueryResume().Insert(rows)\n}\n\n// GetResume retrieves all resume records for a chat\nfunc (store *Xun) GetResume(chatID string) ([]*types.Resume, error) {\n\tif chatID == \"\" {\n\t\treturn nil, fmt.Errorf(\"chat_id is required\")\n\t}\n\n\trows, err := store.newQueryResume().\n\t\tWhere(\"chat_id\", chatID).\n\t\tWhereNull(\"deleted_at\").\n\t\tOrderBy(\"sequence\", \"asc\").\n\t\tGet()\n\tif err != nil {\n\t\treturn nil, err\n\t}\n\n\trecords := make([]*types.Resume, 0, len(rows))\n\tfor _, row := range rows {\n\t\tdata := row.ToMap()\n\t\tif data == nil || data[\"resume_id\"] == nil {\n\t\t\tcontinue\n\t\t}\n\n\t\trecord, err := store.rowToResume(data)\n\t\tif err != nil {\n\t\t\tcontinue\n\t\t}\n\t\trecords = append(records, record)\n\t}\n\n\treturn records, nil\n}\n\n// GetLastResume retrieves the last (most recent) resume record for a chat\nfunc (store *Xun) GetLastResume(chatID string) (*types.Resume, error) {\n\tif chatID == \"\" {\n\t\treturn nil, fmt.Errorf(\"chat_id is required\")\n\t}\n\n\trow, err := store.newQueryResume().\n\t\tWhere(\"chat_id\", chatID).\n\t\tWhereNull(\"deleted_at\").\n\t\tOrderBy(\"sequence\", \"desc\").\n\t\tFirst()\n\tif err != nil {\n\t\treturn nil, err\n\t}\n\n\tif row == nil {\n\t\treturn nil, nil // No resume records found\n\t}\n\n\tdata := row.ToMap()\n\tif len(data) == 0 || data[\"resume_id\"] == nil {\n\t\treturn nil, nil\n\t}\n\n\treturn store.rowToResume(data)\n}\n\n// GetResumeByStackID retrieves resume records for a specific stack\nfunc (store *Xun) GetResumeByStackID(stackID string) ([]*types.Resume, error) {\n\tif stackID == \"\" {\n\t\treturn nil, fmt.Errorf(\"stack_id is required\")\n\t}\n\n\trows, err := store.newQueryResume().\n\t\tWhere(\"stack_id\", stackID).\n\t\tWhereNull(\"deleted_at\").\n\t\tOrderBy(\"sequence\", \"asc\").\n\t\tGet()\n\tif err != nil {\n\t\treturn nil, err\n\t}\n\n\trecords := make([]*types.Resume, 0, len(rows))\n\tfor _, row := range rows {\n\t\tdata := row.ToMap()\n\t\tif data == nil || data[\"resume_id\"] == nil {\n\t\t\tcontinue\n\t\t}\n\n\t\trecord, err := store.rowToResume(data)\n\t\tif err != nil {\n\t\t\tcontinue\n\t\t}\n\t\trecords = append(records, record)\n\t}\n\n\treturn records, nil\n}\n\n// GetStackPath returns the stack path from root to the given stack\n// Returns: [root_stack_id, ..., current_stack_id]\nfunc (store *Xun) GetStackPath(stackID string) ([]string, error) {\n\tif stackID == \"\" {\n\t\treturn nil, fmt.Errorf(\"stack_id is required\")\n\t}\n\n\tpath := []string{stackID}\n\tcurrentStackID := stackID\n\n\t// Walk up the stack tree by following stack_parent_id\n\tfor {\n\t\trow, err := store.newQueryResume().\n\t\t\tWhere(\"stack_id\", currentStackID).\n\t\t\tWhereNull(\"deleted_at\").\n\t\t\tFirst()\n\t\tif err != nil {\n\t\t\treturn nil, err\n\t\t}\n\n\t\tif row == nil {\n\t\t\tbreak\n\t\t}\n\n\t\tdata := row.ToMap()\n\t\tparentID := getString(data, \"stack_parent_id\")\n\t\tif parentID == \"\" {\n\t\t\tbreak // Reached root\n\t\t}\n\n\t\t// Prepend parent to path\n\t\tpath = append([]string{parentID}, path...)\n\t\tcurrentStackID = parentID\n\t}\n\n\treturn path, nil\n}\n\n// DeleteResume soft deletes all resume records for a chat\n// Called after successful resume to clean up\nfunc (store *Xun) DeleteResume(chatID string) error {\n\tif chatID == \"\" {\n\t\treturn fmt.Errorf(\"chat_id is required\")\n\t}\n\n\t_, err := store.newQueryResume().\n\t\tWhere(\"chat_id\", chatID).\n\t\tWhereNull(\"deleted_at\").\n\t\tUpdate(map[string]interface{}{\n\t\t\t\"deleted_at\": time.Now(),\n\t\t\t\"updated_at\": time.Now(),\n\t\t})\n\n\treturn err\n}\n\n// GetResumeByRequestID retrieves resume records for a specific request\nfunc (store *Xun) GetResumeByRequestID(requestID string) ([]*types.Resume, error) {\n\tif requestID == \"\" {\n\t\treturn nil, fmt.Errorf(\"request_id is required\")\n\t}\n\n\trows, err := store.newQueryResume().\n\t\tWhere(\"request_id\", requestID).\n\t\tWhereNull(\"deleted_at\").\n\t\tOrderBy(\"sequence\", \"asc\").\n\t\tGet()\n\tif err != nil {\n\t\treturn nil, err\n\t}\n\n\trecords := make([]*types.Resume, 0, len(rows))\n\tfor _, row := range rows {\n\t\tdata := row.ToMap()\n\t\tif data == nil || data[\"resume_id\"] == nil {\n\t\t\tcontinue\n\t\t}\n\n\t\trecord, err := store.rowToResume(data)\n\t\tif err != nil {\n\t\t\tcontinue\n\t\t}\n\t\trecords = append(records, record)\n\t}\n\n\treturn records, nil\n}\n\n// =============================================================================\n// Helper Functions\n// =============================================================================\n\n// rowToResume converts a database row to a Resume struct\nfunc (store *Xun) rowToResume(data map[string]interface{}) (*types.Resume, error) {\n\trecord := &types.Resume{\n\t\tResumeID: getString(data, \"resume_id\"),\n\t\tChatID: getString(data, \"chat_id\"),\n\t\tRequestID: getString(data, \"request_id\"),\n\t\tAssistantID: getString(data, \"assistant_id\"),\n\t\tStackID: getString(data, \"stack_id\"),\n\t\tStackParentID: getString(data, \"stack_parent_id\"),\n\t\tStackDepth: getInt(data, \"stack_depth\"),\n\t\tType: getString(data, \"type\"),\n\t\tStatus: getString(data, \"status\"),\n\t\tError: getString(data, \"error\"),\n\t\tSequence: getInt(data, \"sequence\"),\n\t}\n\n\t// Handle timestamps\n\tif createdAt := getTime(data, \"created_at\"); createdAt != nil {\n\t\trecord.CreatedAt = *createdAt\n\t}\n\tif updatedAt := getTime(data, \"updated_at\"); updatedAt != nil {\n\t\trecord.UpdatedAt = *updatedAt\n\t}\n\n\t// Handle JSON fields\n\tif input := data[\"input\"]; input != nil {\n\t\tif inputStr, ok := input.(string); ok && inputStr != \"\" {\n\t\t\tvar inputMap map[string]interface{}\n\t\t\tif err := jsoniter.UnmarshalFromString(inputStr, &inputMap); err == nil {\n\t\t\t\trecord.Input = inputMap\n\t\t\t}\n\t\t} else if inputMap, ok := input.(map[string]interface{}); ok {\n\t\t\trecord.Input = inputMap\n\t\t}\n\t}\n\n\tif output := data[\"output\"]; output != nil {\n\t\tif outputStr, ok := output.(string); ok && outputStr != \"\" {\n\t\t\tvar outputMap map[string]interface{}\n\t\t\tif err := jsoniter.UnmarshalFromString(outputStr, &outputMap); err == nil {\n\t\t\t\trecord.Output = outputMap\n\t\t\t}\n\t\t} else if outputMap, ok := output.(map[string]interface{}); ok {\n\t\t\trecord.Output = outputMap\n\t\t}\n\t}\n\n\tif snapshot := data[\"space_snapshot\"]; snapshot != nil {\n\t\tif snapshotStr, ok := snapshot.(string); ok && snapshotStr != \"\" {\n\t\t\tvar snapshotMap map[string]interface{}\n\t\t\tif err := jsoniter.UnmarshalFromString(snapshotStr, &snapshotMap); err == nil {\n\t\t\t\trecord.SpaceSnapshot = snapshotMap\n\t\t\t}\n\t\t} else if snapshotMap, ok := snapshot.(map[string]interface{}); ok {\n\t\t\trecord.SpaceSnapshot = snapshotMap\n\t\t}\n\t}\n\n\tif metadata := data[\"metadata\"]; metadata != nil {\n\t\tif metaStr, ok := metadata.(string); ok && metaStr != \"\" {\n\t\t\tvar metaMap map[string]interface{}\n\t\t\tif err := jsoniter.UnmarshalFromString(metaStr, &metaMap); err == nil {\n\t\t\t\trecord.Metadata = metaMap\n\t\t\t}\n\t\t} else if metaMap, ok := metadata.(map[string]interface{}); ok {\n\t\t\trecord.Metadata = metaMap\n\t\t}\n\t}\n\n\treturn record, nil\n}\n" }, { "path": "agent/store/xun/resume_test.go", "content": "package xun_test\n\nimport (\n\t\"fmt\"\n\t\"testing\"\n\t\"time\"\n\n\t\"github.com/yaoapp/yao/agent/store/types\"\n\t\"github.com/yaoapp/yao/agent/store/xun\"\n\t\"github.com/yaoapp/yao/config\"\n\t\"github.com/yaoapp/yao/test\"\n)\n\n// TestSaveResume tests batch saving resume records\nfunc TestSaveResume(t *testing.T) {\n\ttest.Prepare(t, config.Conf)\n\tdefer test.Clean()\n\n\tstore, err := xun.NewXun(types.Setting{\n\t\tConnector: \"default\",\n\t})\n\tif err != nil {\n\t\tt.Fatalf(\"Failed to create store: %v\", err)\n\t}\n\n\t// Create a chat first\n\tchat := &types.Chat{\n\t\tAssistantID: \"test_assistant\",\n\t\tTitle: \"Resume Test Chat\",\n\t}\n\terr = store.CreateChat(chat)\n\tif err != nil {\n\t\tt.Fatalf(\"Failed to create chat: %v\", err)\n\t}\n\tdefer store.DeleteChat(chat.ChatID)\n\n\tt.Run(\"SaveSingleRecord\", func(t *testing.T) {\n\t\trequestID := fmt.Sprintf(\"req_%d\", time.Now().UnixNano())\n\t\trecords := []*types.Resume{\n\t\t\t{\n\t\t\t\tChatID: chat.ChatID,\n\t\t\t\tRequestID: requestID,\n\t\t\t\tAssistantID: \"test_assistant\",\n\t\t\t\tStackID: \"stack_001\",\n\t\t\t\tStackDepth: 0,\n\t\t\t\tType: types.ResumeTypeLLM,\n\t\t\t\tStatus: types.ResumeStatusInterrupted,\n\t\t\t\tSequence: 1,\n\t\t\t},\n\t\t}\n\n\t\terr := store.SaveResume(records)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to save resume record: %v\", err)\n\t\t}\n\n\t\t// Verify\n\t\tretrieved, err := store.GetResume(chat.ChatID)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to get resume records: %v\", err)\n\t\t}\n\n\t\tfound := false\n\t\tfor _, r := range retrieved {\n\t\t\tif r.RequestID == requestID {\n\t\t\t\tfound = true\n\t\t\t\tif r.Type != types.ResumeTypeLLM {\n\t\t\t\t\tt.Errorf(\"Expected type '%s', got '%s'\", types.ResumeTypeLLM, r.Type)\n\t\t\t\t}\n\t\t\t\tif r.Status != types.ResumeStatusInterrupted {\n\t\t\t\t\tt.Errorf(\"Expected status '%s', got '%s'\", types.ResumeStatusInterrupted, r.Status)\n\t\t\t\t}\n\t\t\t\tbreak\n\t\t\t}\n\t\t}\n\n\t\tif !found {\n\t\t\tt.Error(\"Could not find saved resume record\")\n\t\t}\n\n\t\t// Clean up\n\t\tstore.DeleteResume(chat.ChatID)\n\t})\n\n\tt.Run(\"SaveBatchRecords\", func(t *testing.T) {\n\t\t// Create a new chat for this test\n\t\tbatchChat := &types.Chat{\n\t\t\tAssistantID: \"test_assistant\",\n\t\t}\n\t\terr := store.CreateChat(batchChat)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to create chat: %v\", err)\n\t\t}\n\t\tdefer store.DeleteChat(batchChat.ChatID)\n\n\t\trequestID := fmt.Sprintf(\"req_%d\", time.Now().UnixNano())\n\t\trecords := []*types.Resume{\n\t\t\t{\n\t\t\t\tChatID: batchChat.ChatID,\n\t\t\t\tRequestID: requestID,\n\t\t\t\tAssistantID: \"test_assistant\",\n\t\t\t\tStackID: \"stack_001\",\n\t\t\t\tStackDepth: 0,\n\t\t\t\tType: types.ResumeTypeInput,\n\t\t\t\tStatus: types.ResumeStatusInterrupted,\n\t\t\t\tSequence: 1,\n\t\t\t},\n\t\t\t{\n\t\t\t\tChatID: batchChat.ChatID,\n\t\t\t\tRequestID: requestID,\n\t\t\t\tAssistantID: \"test_assistant\",\n\t\t\t\tStackID: \"stack_001\",\n\t\t\t\tStackDepth: 0,\n\t\t\t\tType: types.ResumeTypeHookCreate,\n\t\t\t\tStatus: types.ResumeStatusInterrupted,\n\t\t\t\tSequence: 2,\n\t\t\t},\n\t\t\t{\n\t\t\t\tChatID: batchChat.ChatID,\n\t\t\t\tRequestID: requestID,\n\t\t\t\tAssistantID: \"test_assistant\",\n\t\t\t\tStackID: \"stack_001\",\n\t\t\t\tStackDepth: 0,\n\t\t\t\tType: types.ResumeTypeLLM,\n\t\t\t\tStatus: types.ResumeStatusFailed,\n\t\t\t\tSequence: 3,\n\t\t\t\tError: \"Connection timeout\",\n\t\t\t},\n\t\t}\n\n\t\terr = store.SaveResume(records)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to save batch resume records: %v\", err)\n\t\t}\n\n\t\t// Verify all records saved\n\t\tretrieved, err := store.GetResume(batchChat.ChatID)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to get resume records: %v\", err)\n\t\t}\n\n\t\tif len(retrieved) != 3 {\n\t\t\tt.Errorf(\"Expected 3 records, got %d\", len(retrieved))\n\t\t}\n\n\t\t// Verify order (should be by sequence)\n\t\tif len(retrieved) >= 3 {\n\t\t\tif retrieved[0].Sequence != 1 {\n\t\t\t\tt.Errorf(\"Expected first record sequence 1, got %d\", retrieved[0].Sequence)\n\t\t\t}\n\t\t\tif retrieved[2].Sequence != 3 {\n\t\t\t\tt.Errorf(\"Expected last record sequence 3, got %d\", retrieved[2].Sequence)\n\t\t\t}\n\t\t\tif retrieved[2].Error != \"Connection timeout\" {\n\t\t\t\tt.Errorf(\"Expected error 'Connection timeout', got '%s'\", retrieved[2].Error)\n\t\t\t}\n\t\t}\n\n\t\tt.Logf(\"Saved %d resume records in single batch call\", len(records))\n\t})\n\n\tt.Run(\"SaveRecordWithAllFields\", func(t *testing.T) {\n\t\tfullChat := &types.Chat{\n\t\t\tAssistantID: \"test_assistant\",\n\t\t}\n\t\terr := store.CreateChat(fullChat)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to create chat: %v\", err)\n\t\t}\n\t\tdefer store.DeleteChat(fullChat.ChatID)\n\n\t\trequestID := fmt.Sprintf(\"req_%d\", time.Now().UnixNano())\n\t\trecords := []*types.Resume{\n\t\t\t{\n\t\t\t\tChatID: fullChat.ChatID,\n\t\t\t\tRequestID: requestID,\n\t\t\t\tAssistantID: \"test_assistant\",\n\t\t\t\tStackID: \"stack_001\",\n\t\t\t\tStackParentID: \"stack_000\",\n\t\t\t\tStackDepth: 1,\n\t\t\t\tType: types.ResumeTypeDelegate,\n\t\t\t\tStatus: types.ResumeStatusInterrupted,\n\t\t\t\tInput: map[string]interface{}{\"agent_id\": \"sub_agent\", \"messages\": []interface{}{}},\n\t\t\t\tOutput: map[string]interface{}{\"partial\": true},\n\t\t\t\tSpaceSnapshot: map[string]interface{}{\"key1\": \"value1\", \"key2\": 123},\n\t\t\t\tError: \"User cancelled\",\n\t\t\t\tSequence: 1,\n\t\t\t\tMetadata: map[string]interface{}{\"retry_count\": 0},\n\t\t\t},\n\t\t}\n\n\t\terr = store.SaveResume(records)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to save record: %v\", err)\n\t\t}\n\n\t\tretrieved, err := store.GetResume(fullChat.ChatID)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to get records: %v\", err)\n\t\t}\n\n\t\tif len(retrieved) != 1 {\n\t\t\tt.Fatalf(\"Expected 1 record, got %d\", len(retrieved))\n\t\t}\n\n\t\tr := retrieved[0]\n\t\tif r.StackParentID != \"stack_000\" {\n\t\t\tt.Errorf(\"Expected stack_parent_id 'stack_000', got '%s'\", r.StackParentID)\n\t\t}\n\t\tif r.StackDepth != 1 {\n\t\t\tt.Errorf(\"Expected stack_depth 1, got %d\", r.StackDepth)\n\t\t}\n\t\tif r.Input == nil {\n\t\t\tt.Error(\"Expected input to be set\")\n\t\t}\n\t\tif r.Output == nil {\n\t\t\tt.Error(\"Expected output to be set\")\n\t\t}\n\t\tif r.SpaceSnapshot == nil {\n\t\t\tt.Error(\"Expected space_snapshot to be set\")\n\t\t} else if r.SpaceSnapshot[\"key1\"] != \"value1\" {\n\t\t\tt.Errorf(\"Expected space_snapshot key1='value1', got '%v'\", r.SpaceSnapshot[\"key1\"])\n\t\t}\n\t\tif r.Metadata == nil {\n\t\t\tt.Error(\"Expected metadata to be set\")\n\t\t}\n\t})\n\n\tt.Run(\"SaveEmptyRecords\", func(t *testing.T) {\n\t\terr := store.SaveResume([]*types.Resume{})\n\t\tif err != nil {\n\t\t\tt.Errorf(\"Expected no error for empty records, got: %v\", err)\n\t\t}\n\t})\n\n\tt.Run(\"SaveRecordWithoutChatID\", func(t *testing.T) {\n\t\trecords := []*types.Resume{{RequestID: \"req\", AssistantID: \"ast\", StackID: \"stk\", Type: \"llm\", Status: \"failed\", Sequence: 1}}\n\t\terr := store.SaveResume(records)\n\t\tif err == nil {\n\t\t\tt.Error(\"Expected error when saving without chat_id\")\n\t\t}\n\t})\n\n\tt.Run(\"SaveRecordWithoutRequestID\", func(t *testing.T) {\n\t\trecords := []*types.Resume{{ChatID: chat.ChatID, AssistantID: \"ast\", StackID: \"stk\", Type: \"llm\", Status: \"failed\", Sequence: 1}}\n\t\terr := store.SaveResume(records)\n\t\tif err == nil {\n\t\t\tt.Error(\"Expected error when saving without request_id\")\n\t\t}\n\t})\n\n\tt.Run(\"SaveRecordWithoutAssistantID\", func(t *testing.T) {\n\t\trecords := []*types.Resume{{ChatID: chat.ChatID, RequestID: \"req\", StackID: \"stk\", Type: \"llm\", Status: \"failed\", Sequence: 1}}\n\t\terr := store.SaveResume(records)\n\t\tif err == nil {\n\t\t\tt.Error(\"Expected error when saving without assistant_id\")\n\t\t}\n\t})\n\n\tt.Run(\"SaveRecordWithoutStackID\", func(t *testing.T) {\n\t\trecords := []*types.Resume{{ChatID: chat.ChatID, RequestID: \"req\", AssistantID: \"ast\", Type: \"llm\", Status: \"failed\", Sequence: 1}}\n\t\terr := store.SaveResume(records)\n\t\tif err == nil {\n\t\t\tt.Error(\"Expected error when saving without stack_id\")\n\t\t}\n\t})\n\n\tt.Run(\"SaveRecordWithoutType\", func(t *testing.T) {\n\t\trecords := []*types.Resume{{ChatID: chat.ChatID, RequestID: \"req\", AssistantID: \"ast\", StackID: \"stk\", Status: \"failed\", Sequence: 1}}\n\t\terr := store.SaveResume(records)\n\t\tif err == nil {\n\t\t\tt.Error(\"Expected error when saving without type\")\n\t\t}\n\t})\n\n\tt.Run(\"SaveRecordWithoutStatus\", func(t *testing.T) {\n\t\trecords := []*types.Resume{{ChatID: chat.ChatID, RequestID: \"req\", AssistantID: \"ast\", StackID: \"stk\", Type: \"llm\", Sequence: 1}}\n\t\terr := store.SaveResume(records)\n\t\tif err == nil {\n\t\t\tt.Error(\"Expected error when saving without status\")\n\t\t}\n\t})\n}\n\n// TestGetResume tests retrieving resume records\nfunc TestGetResume(t *testing.T) {\n\ttest.Prepare(t, config.Conf)\n\tdefer test.Clean()\n\n\tstore, err := xun.NewXun(types.Setting{\n\t\tConnector: \"default\",\n\t})\n\tif err != nil {\n\t\tt.Fatalf(\"Failed to create store: %v\", err)\n\t}\n\n\t// Create chat and resume records\n\tchat := &types.Chat{\n\t\tAssistantID: \"test_assistant\",\n\t}\n\terr = store.CreateChat(chat)\n\tif err != nil {\n\t\tt.Fatalf(\"Failed to create chat: %v\", err)\n\t}\n\tdefer store.DeleteChat(chat.ChatID)\n\n\trequestID := fmt.Sprintf(\"req_%d\", time.Now().UnixNano())\n\trecords := []*types.Resume{\n\t\t{ChatID: chat.ChatID, RequestID: requestID, AssistantID: \"ast1\", StackID: \"stk1\", Type: types.ResumeTypeInput, Status: types.ResumeStatusInterrupted, Sequence: 1},\n\t\t{ChatID: chat.ChatID, RequestID: requestID, AssistantID: \"ast1\", StackID: \"stk1\", Type: types.ResumeTypeHookCreate, Status: types.ResumeStatusInterrupted, Sequence: 2},\n\t\t{ChatID: chat.ChatID, RequestID: requestID, AssistantID: \"ast1\", StackID: \"stk1\", Type: types.ResumeTypeLLM, Status: types.ResumeStatusFailed, Sequence: 3},\n\t}\n\terr = store.SaveResume(records)\n\tif err != nil {\n\t\tt.Fatalf(\"Failed to save records: %v\", err)\n\t}\n\tdefer store.DeleteResume(chat.ChatID)\n\n\tt.Run(\"GetAllRecords\", func(t *testing.T) {\n\t\tretrieved, err := store.GetResume(chat.ChatID)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to get records: %v\", err)\n\t\t}\n\n\t\tif len(retrieved) != 3 {\n\t\t\tt.Errorf(\"Expected 3 records, got %d\", len(retrieved))\n\t\t}\n\n\t\t// Verify order by sequence\n\t\tfor i := 1; i < len(retrieved); i++ {\n\t\t\tif retrieved[i].Sequence < retrieved[i-1].Sequence {\n\t\t\t\tt.Error(\"Records not ordered by sequence\")\n\t\t\t}\n\t\t}\n\t})\n\n\tt.Run(\"GetRecordsWithEmptyChatID\", func(t *testing.T) {\n\t\t_, err := store.GetResume(\"\")\n\t\tif err == nil {\n\t\t\tt.Error(\"Expected error when getting records without chat_id\")\n\t\t}\n\t})\n\n\tt.Run(\"GetRecordsFromNonExistentChat\", func(t *testing.T) {\n\t\tretrieved, err := store.GetResume(\"nonexistent_chat\")\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Unexpected error: %v\", err)\n\t\t}\n\t\tif len(retrieved) != 0 {\n\t\t\tt.Errorf(\"Expected 0 records from non-existent chat, got %d\", len(retrieved))\n\t\t}\n\t})\n}\n\n// TestGetLastResume tests retrieving the last resume record\nfunc TestGetLastResume(t *testing.T) {\n\ttest.Prepare(t, config.Conf)\n\tdefer test.Clean()\n\n\tstore, err := xun.NewXun(types.Setting{\n\t\tConnector: \"default\",\n\t})\n\tif err != nil {\n\t\tt.Fatalf(\"Failed to create store: %v\", err)\n\t}\n\n\tchat := &types.Chat{\n\t\tAssistantID: \"test_assistant\",\n\t}\n\terr = store.CreateChat(chat)\n\tif err != nil {\n\t\tt.Fatalf(\"Failed to create chat: %v\", err)\n\t}\n\tdefer store.DeleteChat(chat.ChatID)\n\n\tt.Run(\"GetLastRecordFromMultiple\", func(t *testing.T) {\n\t\trequestID := fmt.Sprintf(\"req_%d\", time.Now().UnixNano())\n\t\trecords := []*types.Resume{\n\t\t\t{ChatID: chat.ChatID, RequestID: requestID, AssistantID: \"ast\", StackID: \"stk\", Type: types.ResumeTypeInput, Status: types.ResumeStatusInterrupted, Sequence: 1},\n\t\t\t{ChatID: chat.ChatID, RequestID: requestID, AssistantID: \"ast\", StackID: \"stk\", Type: types.ResumeTypeHookCreate, Status: types.ResumeStatusInterrupted, Sequence: 2},\n\t\t\t{ChatID: chat.ChatID, RequestID: requestID, AssistantID: \"ast\", StackID: \"stk\", Type: types.ResumeTypeLLM, Status: types.ResumeStatusFailed, Sequence: 3, Error: \"Last error\"},\n\t\t}\n\t\terr := store.SaveResume(records)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to save records: %v\", err)\n\t\t}\n\t\tdefer store.DeleteResume(chat.ChatID)\n\n\t\tlast, err := store.GetLastResume(chat.ChatID)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to get last record: %v\", err)\n\t\t}\n\n\t\tif last == nil {\n\t\t\tt.Fatal(\"Expected last record, got nil\")\n\t\t}\n\n\t\tif last.Sequence != 3 {\n\t\t\tt.Errorf(\"Expected sequence 3, got %d\", last.Sequence)\n\t\t}\n\t\tif last.Type != types.ResumeTypeLLM {\n\t\t\tt.Errorf(\"Expected type '%s', got '%s'\", types.ResumeTypeLLM, last.Type)\n\t\t}\n\t\tif last.Error != \"Last error\" {\n\t\t\tt.Errorf(\"Expected error 'Last error', got '%s'\", last.Error)\n\t\t}\n\t})\n\n\tt.Run(\"GetLastRecordFromEmpty\", func(t *testing.T) {\n\t\temptyChat := &types.Chat{AssistantID: \"test_assistant\"}\n\t\terr := store.CreateChat(emptyChat)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to create chat: %v\", err)\n\t\t}\n\t\tdefer store.DeleteChat(emptyChat.ChatID)\n\n\t\tlast, err := store.GetLastResume(emptyChat.ChatID)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Unexpected error: %v\", err)\n\t\t}\n\n\t\tif last != nil {\n\t\t\tt.Error(\"Expected nil for empty chat, got record\")\n\t\t}\n\t})\n\n\tt.Run(\"GetLastRecordWithEmptyChatID\", func(t *testing.T) {\n\t\t_, err := store.GetLastResume(\"\")\n\t\tif err == nil {\n\t\t\tt.Error(\"Expected error when getting last record without chat_id\")\n\t\t}\n\t})\n}\n\n// TestGetResumeByStackID tests retrieving records by stack ID\nfunc TestGetResumeByStackID(t *testing.T) {\n\ttest.Prepare(t, config.Conf)\n\tdefer test.Clean()\n\n\tstore, err := xun.NewXun(types.Setting{\n\t\tConnector: \"default\",\n\t})\n\tif err != nil {\n\t\tt.Fatalf(\"Failed to create store: %v\", err)\n\t}\n\n\tchat := &types.Chat{\n\t\tAssistantID: \"test_assistant\",\n\t}\n\terr = store.CreateChat(chat)\n\tif err != nil {\n\t\tt.Fatalf(\"Failed to create chat: %v\", err)\n\t}\n\tdefer store.DeleteChat(chat.ChatID)\n\n\trequestID := fmt.Sprintf(\"req_%d\", time.Now().UnixNano())\n\trecords := []*types.Resume{\n\t\t{ChatID: chat.ChatID, RequestID: requestID, AssistantID: \"ast1\", StackID: \"stack_A\", Type: types.ResumeTypeInput, Status: types.ResumeStatusInterrupted, Sequence: 1},\n\t\t{ChatID: chat.ChatID, RequestID: requestID, AssistantID: \"ast1\", StackID: \"stack_A\", Type: types.ResumeTypeLLM, Status: types.ResumeStatusInterrupted, Sequence: 2},\n\t\t{ChatID: chat.ChatID, RequestID: requestID, AssistantID: \"ast2\", StackID: \"stack_B\", StackParentID: \"stack_A\", StackDepth: 1, Type: types.ResumeTypeDelegate, Status: types.ResumeStatusFailed, Sequence: 3},\n\t}\n\terr = store.SaveResume(records)\n\tif err != nil {\n\t\tt.Fatalf(\"Failed to save records: %v\", err)\n\t}\n\tdefer store.DeleteResume(chat.ChatID)\n\n\tt.Run(\"GetRecordsByStackA\", func(t *testing.T) {\n\t\tretrieved, err := store.GetResumeByStackID(\"stack_A\")\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to get records: %v\", err)\n\t\t}\n\n\t\tif len(retrieved) != 2 {\n\t\t\tt.Errorf(\"Expected 2 records for stack_A, got %d\", len(retrieved))\n\t\t}\n\t})\n\n\tt.Run(\"GetRecordsByStackB\", func(t *testing.T) {\n\t\tretrieved, err := store.GetResumeByStackID(\"stack_B\")\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to get records: %v\", err)\n\t\t}\n\n\t\tif len(retrieved) != 1 {\n\t\t\tt.Errorf(\"Expected 1 record for stack_B, got %d\", len(retrieved))\n\t\t}\n\n\t\tif len(retrieved) > 0 {\n\t\t\tif retrieved[0].StackParentID != \"stack_A\" {\n\t\t\t\tt.Errorf(\"Expected stack_parent_id 'stack_A', got '%s'\", retrieved[0].StackParentID)\n\t\t\t}\n\t\t\tif retrieved[0].StackDepth != 1 {\n\t\t\t\tt.Errorf(\"Expected stack_depth 1, got %d\", retrieved[0].StackDepth)\n\t\t\t}\n\t\t}\n\t})\n\n\tt.Run(\"GetRecordsByNonExistentStack\", func(t *testing.T) {\n\t\tretrieved, err := store.GetResumeByStackID(\"nonexistent_stack\")\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Unexpected error: %v\", err)\n\t\t}\n\t\tif len(retrieved) != 0 {\n\t\t\tt.Errorf(\"Expected 0 records, got %d\", len(retrieved))\n\t\t}\n\t})\n\n\tt.Run(\"GetRecordsByEmptyStackID\", func(t *testing.T) {\n\t\t_, err := store.GetResumeByStackID(\"\")\n\t\tif err == nil {\n\t\t\tt.Error(\"Expected error when getting records without stack_id\")\n\t\t}\n\t})\n}\n\n// TestGetStackPath tests retrieving the stack path\nfunc TestGetStackPath(t *testing.T) {\n\ttest.Prepare(t, config.Conf)\n\tdefer test.Clean()\n\n\tstore, err := xun.NewXun(types.Setting{\n\t\tConnector: \"default\",\n\t})\n\tif err != nil {\n\t\tt.Fatalf(\"Failed to create store: %v\", err)\n\t}\n\n\tchat := &types.Chat{\n\t\tAssistantID: \"test_assistant\",\n\t}\n\terr = store.CreateChat(chat)\n\tif err != nil {\n\t\tt.Fatalf(\"Failed to create chat: %v\", err)\n\t}\n\tdefer store.DeleteChat(chat.ChatID)\n\n\t// Create a nested stack structure: root -> child -> grandchild\n\trequestID := fmt.Sprintf(\"req_%d\", time.Now().UnixNano())\n\trecords := []*types.Resume{\n\t\t{ChatID: chat.ChatID, RequestID: requestID, AssistantID: \"ast1\", StackID: \"root_stack\", Type: types.ResumeTypeInput, Status: types.ResumeStatusInterrupted, Sequence: 1},\n\t\t{ChatID: chat.ChatID, RequestID: requestID, AssistantID: \"ast2\", StackID: \"child_stack\", StackParentID: \"root_stack\", StackDepth: 1, Type: types.ResumeTypeDelegate, Status: types.ResumeStatusInterrupted, Sequence: 2},\n\t\t{ChatID: chat.ChatID, RequestID: requestID, AssistantID: \"ast3\", StackID: \"grandchild_stack\", StackParentID: \"child_stack\", StackDepth: 2, Type: types.ResumeTypeLLM, Status: types.ResumeStatusFailed, Sequence: 3},\n\t}\n\terr = store.SaveResume(records)\n\tif err != nil {\n\t\tt.Fatalf(\"Failed to save records: %v\", err)\n\t}\n\tdefer store.DeleteResume(chat.ChatID)\n\n\tt.Run(\"GetPathFromGrandchild\", func(t *testing.T) {\n\t\tpath, err := store.GetStackPath(\"grandchild_stack\")\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to get stack path: %v\", err)\n\t\t}\n\n\t\tif len(path) != 3 {\n\t\t\tt.Errorf(\"Expected path length 3, got %d\", len(path))\n\t\t}\n\n\t\tif len(path) >= 3 {\n\t\t\tif path[0] != \"root_stack\" {\n\t\t\t\tt.Errorf(\"Expected first element 'root_stack', got '%s'\", path[0])\n\t\t\t}\n\t\t\tif path[1] != \"child_stack\" {\n\t\t\t\tt.Errorf(\"Expected second element 'child_stack', got '%s'\", path[1])\n\t\t\t}\n\t\t\tif path[2] != \"grandchild_stack\" {\n\t\t\t\tt.Errorf(\"Expected third element 'grandchild_stack', got '%s'\", path[2])\n\t\t\t}\n\t\t}\n\n\t\tt.Logf(\"Stack path: %v\", path)\n\t})\n\n\tt.Run(\"GetPathFromChild\", func(t *testing.T) {\n\t\tpath, err := store.GetStackPath(\"child_stack\")\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to get stack path: %v\", err)\n\t\t}\n\n\t\tif len(path) != 2 {\n\t\t\tt.Errorf(\"Expected path length 2, got %d\", len(path))\n\t\t}\n\n\t\tif len(path) >= 2 {\n\t\t\tif path[0] != \"root_stack\" {\n\t\t\t\tt.Errorf(\"Expected first element 'root_stack', got '%s'\", path[0])\n\t\t\t}\n\t\t\tif path[1] != \"child_stack\" {\n\t\t\t\tt.Errorf(\"Expected second element 'child_stack', got '%s'\", path[1])\n\t\t\t}\n\t\t}\n\t})\n\n\tt.Run(\"GetPathFromRoot\", func(t *testing.T) {\n\t\tpath, err := store.GetStackPath(\"root_stack\")\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to get stack path: %v\", err)\n\t\t}\n\n\t\tif len(path) != 1 {\n\t\t\tt.Errorf(\"Expected path length 1, got %d\", len(path))\n\t\t}\n\n\t\tif len(path) >= 1 && path[0] != \"root_stack\" {\n\t\t\tt.Errorf(\"Expected 'root_stack', got '%s'\", path[0])\n\t\t}\n\t})\n\n\tt.Run(\"GetPathWithEmptyStackID\", func(t *testing.T) {\n\t\t_, err := store.GetStackPath(\"\")\n\t\tif err == nil {\n\t\t\tt.Error(\"Expected error when getting path without stack_id\")\n\t\t}\n\t})\n}\n\n// TestDeleteResume tests deleting resume records\nfunc TestDeleteResume(t *testing.T) {\n\ttest.Prepare(t, config.Conf)\n\tdefer test.Clean()\n\n\tstore, err := xun.NewXun(types.Setting{\n\t\tConnector: \"default\",\n\t})\n\tif err != nil {\n\t\tt.Fatalf(\"Failed to create store: %v\", err)\n\t}\n\n\tt.Run(\"DeleteExistingRecords\", func(t *testing.T) {\n\t\tchat := &types.Chat{AssistantID: \"test_assistant\"}\n\t\terr := store.CreateChat(chat)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to create chat: %v\", err)\n\t\t}\n\t\tdefer store.DeleteChat(chat.ChatID)\n\n\t\trequestID := fmt.Sprintf(\"req_%d\", time.Now().UnixNano())\n\t\trecords := []*types.Resume{\n\t\t\t{ChatID: chat.ChatID, RequestID: requestID, AssistantID: \"ast\", StackID: \"stk\", Type: types.ResumeTypeLLM, Status: types.ResumeStatusFailed, Sequence: 1},\n\t\t\t{ChatID: chat.ChatID, RequestID: requestID, AssistantID: \"ast\", StackID: \"stk\", Type: types.ResumeTypeTool, Status: types.ResumeStatusFailed, Sequence: 2},\n\t\t}\n\t\terr = store.SaveResume(records)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to save records: %v\", err)\n\t\t}\n\n\t\t// Delete\n\t\terr = store.DeleteResume(chat.ChatID)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to delete records: %v\", err)\n\t\t}\n\n\t\t// Verify deleted\n\t\tretrieved, err := store.GetResume(chat.ChatID)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to get records: %v\", err)\n\t\t}\n\n\t\tif len(retrieved) != 0 {\n\t\t\tt.Errorf(\"Expected 0 records after delete, got %d\", len(retrieved))\n\t\t}\n\t})\n\n\tt.Run(\"DeleteFromEmptyChat\", func(t *testing.T) {\n\t\tchat := &types.Chat{AssistantID: \"test_assistant\"}\n\t\terr := store.CreateChat(chat)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to create chat: %v\", err)\n\t\t}\n\t\tdefer store.DeleteChat(chat.ChatID)\n\n\t\t// Delete from chat with no records - should not error\n\t\terr = store.DeleteResume(chat.ChatID)\n\t\tif err != nil {\n\t\t\tt.Errorf(\"Expected no error when deleting from empty chat, got: %v\", err)\n\t\t}\n\t})\n\n\tt.Run(\"DeleteWithEmptyChatID\", func(t *testing.T) {\n\t\terr := store.DeleteResume(\"\")\n\t\tif err == nil {\n\t\t\tt.Error(\"Expected error when deleting with empty chat_id\")\n\t\t}\n\t})\n}\n\n// TestResumeCompleteWorkflow tests a complete resume/retry workflow\nfunc TestResumeCompleteWorkflow(t *testing.T) {\n\ttest.Prepare(t, config.Conf)\n\tdefer test.Clean()\n\n\tstore, err := xun.NewXun(types.Setting{\n\t\tConnector: \"default\",\n\t})\n\tif err != nil {\n\t\tt.Fatalf(\"Failed to create store: %v\", err)\n\t}\n\n\tt.Run(\"CompleteA2AWorkflow\", func(t *testing.T) {\n\t\t// Create chat\n\t\tchat := &types.Chat{\n\t\t\tAssistantID: \"main_assistant\",\n\t\t\tTitle: \"A2A Workflow Test\",\n\t\t}\n\t\terr := store.CreateChat(chat)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to create chat: %v\", err)\n\t\t}\n\t\tdefer store.DeleteChat(chat.ChatID)\n\n\t\t// Simulate A2A call that gets interrupted\n\t\t// Main assistant -> Sub assistant (interrupted during LLM call)\n\t\trequestID := fmt.Sprintf(\"req_%d\", time.Now().UnixNano())\n\t\trecords := []*types.Resume{\n\t\t\t// Main assistant steps\n\t\t\t{\n\t\t\t\tChatID: chat.ChatID,\n\t\t\t\tRequestID: requestID,\n\t\t\t\tAssistantID: \"main_assistant\",\n\t\t\t\tStackID: \"main_stack\",\n\t\t\t\tStackDepth: 0,\n\t\t\t\tType: types.ResumeTypeInput,\n\t\t\t\tStatus: types.ResumeStatusInterrupted,\n\t\t\t\tInput: map[string]interface{}{\"messages\": []interface{}{map[string]interface{}{\"role\": \"user\", \"content\": \"Analyze this\"}}},\n\t\t\t\tSequence: 1,\n\t\t\t},\n\t\t\t{\n\t\t\t\tChatID: chat.ChatID,\n\t\t\t\tRequestID: requestID,\n\t\t\t\tAssistantID: \"main_assistant\",\n\t\t\t\tStackID: \"main_stack\",\n\t\t\t\tStackDepth: 0,\n\t\t\t\tType: types.ResumeTypeDelegate,\n\t\t\t\tStatus: types.ResumeStatusInterrupted,\n\t\t\t\tSpaceSnapshot: map[string]interface{}{\"task\": \"analyze\", \"data_id\": \"123\"},\n\t\t\t\tSequence: 2,\n\t\t\t},\n\t\t\t// Sub assistant steps\n\t\t\t{\n\t\t\t\tChatID: chat.ChatID,\n\t\t\t\tRequestID: requestID,\n\t\t\t\tAssistantID: \"sub_assistant\",\n\t\t\t\tStackID: \"sub_stack\",\n\t\t\t\tStackParentID: \"main_stack\",\n\t\t\t\tStackDepth: 1,\n\t\t\t\tType: types.ResumeTypeInput,\n\t\t\t\tStatus: types.ResumeStatusInterrupted,\n\t\t\t\tSequence: 3,\n\t\t\t},\n\t\t\t{\n\t\t\t\tChatID: chat.ChatID,\n\t\t\t\tRequestID: requestID,\n\t\t\t\tAssistantID: \"sub_assistant\",\n\t\t\t\tStackID: \"sub_stack\",\n\t\t\t\tStackParentID: \"main_stack\",\n\t\t\t\tStackDepth: 1,\n\t\t\t\tType: types.ResumeTypeLLM,\n\t\t\t\tStatus: types.ResumeStatusInterrupted,\n\t\t\t\tInput: map[string]interface{}{\"messages\": []interface{}{}},\n\t\t\t\tOutput: map[string]interface{}{\"partial_content\": \"The analysis shows...\"},\n\t\t\t\tSpaceSnapshot: map[string]interface{}{\"task\": \"analyze\", \"data_id\": \"123\"},\n\t\t\t\tSequence: 4,\n\t\t\t},\n\t\t}\n\n\t\terr = store.SaveResume(records)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to save resume records: %v\", err)\n\t\t}\n\t\tt.Logf(\"Saved %d resume records for A2A workflow\", len(records))\n\n\t\t// 1. Get last resume record (should be the interrupted LLM call)\n\t\tlast, err := store.GetLastResume(chat.ChatID)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to get last resume: %v\", err)\n\t\t}\n\n\t\tif last == nil {\n\t\t\tt.Fatal(\"Expected last resume record\")\n\t\t}\n\n\t\tif last.Type != types.ResumeTypeLLM {\n\t\t\tt.Errorf(\"Expected type '%s', got '%s'\", types.ResumeTypeLLM, last.Type)\n\t\t}\n\t\tif last.StackDepth != 1 {\n\t\t\tt.Errorf(\"Expected stack_depth 1, got %d\", last.StackDepth)\n\t\t}\n\n\t\t// 2. Get stack path to understand the call hierarchy\n\t\tpath, err := store.GetStackPath(last.StackID)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to get stack path: %v\", err)\n\t\t}\n\n\t\tif len(path) != 2 {\n\t\t\tt.Errorf(\"Expected path length 2, got %d\", len(path))\n\t\t}\n\t\tt.Logf(\"Stack path: %v\", path)\n\n\t\t// 3. Get all records for the sub stack\n\t\tsubRecords, err := store.GetResumeByStackID(\"sub_stack\")\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to get sub stack records: %v\", err)\n\t\t}\n\n\t\tif len(subRecords) != 2 {\n\t\t\tt.Errorf(\"Expected 2 records for sub_stack, got %d\", len(subRecords))\n\t\t}\n\n\t\t// 4. Verify space snapshot is preserved\n\t\tif last.SpaceSnapshot == nil {\n\t\t\tt.Error(\"Expected space_snapshot to be set\")\n\t\t} else {\n\t\t\tif last.SpaceSnapshot[\"task\"] != \"analyze\" {\n\t\t\t\tt.Errorf(\"Expected task='analyze', got '%v'\", last.SpaceSnapshot[\"task\"])\n\t\t\t}\n\t\t}\n\n\t\t// 5. Clean up after successful resume\n\t\terr = store.DeleteResume(chat.ChatID)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to delete resume records: %v\", err)\n\t\t}\n\n\t\t// 6. Verify cleanup\n\t\tremaining, err := store.GetResume(chat.ChatID)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to get remaining records: %v\", err)\n\t\t}\n\n\t\tif len(remaining) != 0 {\n\t\t\tt.Errorf(\"Expected 0 records after cleanup, got %d\", len(remaining))\n\t\t}\n\n\t\tt.Log(\"Complete A2A workflow test passed!\")\n\t})\n}\n" }, { "path": "agent/store/xun/search.go", "content": "package xun\n\nimport (\n\t\"fmt\"\n\t\"time\"\n\n\tjsoniter \"github.com/json-iterator/go\"\n\t\"github.com/yaoapp/gou/model\"\n\t\"github.com/yaoapp/xun/dbal/query\"\n\t\"github.com/yaoapp/yao/agent/store/types\"\n)\n\n// =============================================================================\n// Search Management\n// =============================================================================\n\n// SaveSearch saves a search record for a request\nfunc (store *Xun) SaveSearch(search *types.Search) error {\n\tif search == nil {\n\t\treturn fmt.Errorf(\"search is nil\")\n\t}\n\tif search.RequestID == \"\" {\n\t\treturn fmt.Errorf(\"request_id is required\")\n\t}\n\tif search.ChatID == \"\" {\n\t\treturn fmt.Errorf(\"chat_id is required\")\n\t}\n\tif search.Source == \"\" {\n\t\treturn fmt.Errorf(\"source is required\")\n\t}\n\n\tnow := time.Now()\n\n\t// Build row data\n\trow := map[string]interface{}{\n\t\t\"request_id\": search.RequestID,\n\t\t\"chat_id\": search.ChatID,\n\t\t\"query\": search.Query,\n\t\t\"source\": search.Source,\n\t\t\"duration\": search.Duration,\n\t\t\"created_at\": now,\n\t\t\"updated_at\": now,\n\t}\n\n\t// Handle JSON fields\n\tif search.Config != nil {\n\t\tconfigJSON, err := jsoniter.MarshalToString(search.Config)\n\t\tif err != nil {\n\t\t\treturn fmt.Errorf(\"failed to marshal config: %w\", err)\n\t\t}\n\t\trow[\"config\"] = configJSON\n\t}\n\n\tif len(search.Keywords) > 0 {\n\t\tkeywordsJSON, err := jsoniter.MarshalToString(search.Keywords)\n\t\tif err != nil {\n\t\t\treturn fmt.Errorf(\"failed to marshal keywords: %w\", err)\n\t\t}\n\t\trow[\"keywords\"] = keywordsJSON\n\t}\n\n\tif len(search.Entities) > 0 {\n\t\tentitiesJSON, err := jsoniter.MarshalToString(search.Entities)\n\t\tif err != nil {\n\t\t\treturn fmt.Errorf(\"failed to marshal entities: %w\", err)\n\t\t}\n\t\trow[\"entities\"] = entitiesJSON\n\t}\n\n\tif len(search.Relations) > 0 {\n\t\trelationsJSON, err := jsoniter.MarshalToString(search.Relations)\n\t\tif err != nil {\n\t\t\treturn fmt.Errorf(\"failed to marshal relations: %w\", err)\n\t\t}\n\t\trow[\"relations\"] = relationsJSON\n\t}\n\n\tif search.DSL != nil {\n\t\tdslJSON, err := jsoniter.MarshalToString(search.DSL)\n\t\tif err != nil {\n\t\t\treturn fmt.Errorf(\"failed to marshal dsl: %w\", err)\n\t\t}\n\t\trow[\"dsl\"] = dslJSON\n\t}\n\n\tif len(search.References) > 0 {\n\t\trefsJSON, err := jsoniter.MarshalToString(search.References)\n\t\tif err != nil {\n\t\t\treturn fmt.Errorf(\"failed to marshal references: %w\", err)\n\t\t}\n\t\trow[\"references\"] = refsJSON\n\t}\n\n\tif len(search.Graph) > 0 {\n\t\tgraphJSON, err := jsoniter.MarshalToString(search.Graph)\n\t\tif err != nil {\n\t\t\treturn fmt.Errorf(\"failed to marshal graph: %w\", err)\n\t\t}\n\t\trow[\"graph\"] = graphJSON\n\t}\n\n\tif search.XML != \"\" {\n\t\trow[\"xml\"] = search.XML\n\t}\n\n\tif search.Prompt != \"\" {\n\t\trow[\"prompt\"] = search.Prompt\n\t}\n\n\tif search.Error != \"\" {\n\t\trow[\"error\"] = search.Error\n\t}\n\n\treturn store.newQuerySearch().Insert(row)\n}\n\n// GetSearches retrieves all search records for a request\nfunc (store *Xun) GetSearches(requestID string) ([]*types.Search, error) {\n\tif requestID == \"\" {\n\t\treturn nil, fmt.Errorf(\"request_id is required\")\n\t}\n\n\trows, err := store.newQuerySearch().\n\t\tWhere(\"request_id\", requestID).\n\t\tWhereNull(\"deleted_at\").\n\t\tOrderBy(\"created_at\", \"asc\").\n\t\tGet()\n\tif err != nil {\n\t\treturn nil, err\n\t}\n\n\tsearches := make([]*types.Search, 0, len(rows))\n\tfor _, row := range rows {\n\t\tdata := row.ToMap()\n\t\tif data == nil {\n\t\t\tcontinue\n\t\t}\n\n\t\tsearch, err := store.rowToSearch(data)\n\t\tif err != nil {\n\t\t\tcontinue\n\t\t}\n\t\tsearches = append(searches, search)\n\t}\n\n\treturn searches, nil\n}\n\n// GetReference retrieves a single reference by request ID and index\nfunc (store *Xun) GetReference(requestID string, index int) (*types.Reference, error) {\n\tif requestID == \"\" {\n\t\treturn nil, fmt.Errorf(\"request_id is required\")\n\t}\n\tif index < 1 {\n\t\treturn nil, fmt.Errorf(\"index must be >= 1\")\n\t}\n\n\t// Get all searches for this request\n\tsearches, err := store.GetSearches(requestID)\n\tif err != nil {\n\t\treturn nil, err\n\t}\n\n\t// Find the reference with matching index\n\tfor _, search := range searches {\n\t\tfor _, ref := range search.References {\n\t\t\tif ref.Index == index {\n\t\t\t\treturn &ref, nil\n\t\t\t}\n\t\t}\n\t}\n\n\treturn nil, fmt.Errorf(\"reference not found: request_id=%s, index=%d\", requestID, index)\n}\n\n// DeleteSearches deletes all search records for a chat (soft delete)\nfunc (store *Xun) DeleteSearches(chatID string) error {\n\tif chatID == \"\" {\n\t\treturn fmt.Errorf(\"chat_id is required\")\n\t}\n\n\t_, err := store.newQuerySearch().\n\t\tWhere(\"chat_id\", chatID).\n\t\tWhereNull(\"deleted_at\").\n\t\tUpdate(map[string]interface{}{\n\t\t\t\"deleted_at\": time.Now(),\n\t\t\t\"updated_at\": time.Now(),\n\t\t})\n\n\treturn err\n}\n\n// =============================================================================\n// Query Builder\n// =============================================================================\n\n// newQuerySearch creates a new query builder for the search table\nfunc (store *Xun) newQuerySearch() query.Query {\n\tqb := store.query.New()\n\tqb.Table(store.getSearchTable())\n\treturn qb\n}\n\n// getSearchTable returns the search table name\nfunc (store *Xun) getSearchTable() string {\n\tm := model.Select(\"__yao.agent.search\")\n\tif m != nil && m.MetaData.Table.Name != \"\" {\n\t\treturn m.MetaData.Table.Name\n\t}\n\treturn \"agent_search\"\n}\n\n// =============================================================================\n// Helper Functions\n// =============================================================================\n\n// rowToSearch converts a database row to a Search struct\nfunc (store *Xun) rowToSearch(data map[string]interface{}) (*types.Search, error) {\n\tsearch := &types.Search{\n\t\tID: getInt64(data, \"id\"),\n\t\tRequestID: getString(data, \"request_id\"),\n\t\tChatID: getString(data, \"chat_id\"),\n\t\tQuery: getString(data, \"query\"),\n\t\tSource: getString(data, \"source\"),\n\t\tXML: getString(data, \"xml\"),\n\t\tPrompt: getString(data, \"prompt\"),\n\t\tDuration: getInt64(data, \"duration\"),\n\t\tError: getString(data, \"error\"),\n\t}\n\n\t// Handle timestamps\n\tif createdAt := getTime(data, \"created_at\"); createdAt != nil {\n\t\tsearch.CreatedAt = *createdAt\n\t}\n\n\t// Parse JSON fields\n\tif config := data[\"config\"]; config != nil {\n\t\tif configStr, ok := config.(string); ok && configStr != \"\" {\n\t\t\tvar configMap map[string]any\n\t\t\tif err := jsoniter.UnmarshalFromString(configStr, &configMap); err == nil {\n\t\t\t\tsearch.Config = configMap\n\t\t\t}\n\t\t}\n\t}\n\n\tif keywords := data[\"keywords\"]; keywords != nil {\n\t\tif keywordsStr, ok := keywords.(string); ok && keywordsStr != \"\" {\n\t\t\tvar keywordsList []string\n\t\t\tif err := jsoniter.UnmarshalFromString(keywordsStr, &keywordsList); err == nil {\n\t\t\t\tsearch.Keywords = keywordsList\n\t\t\t}\n\t\t}\n\t}\n\n\tif entities := data[\"entities\"]; entities != nil {\n\t\tif entitiesStr, ok := entities.(string); ok && entitiesStr != \"\" {\n\t\t\tvar entitiesList []types.Entity\n\t\t\tif err := jsoniter.UnmarshalFromString(entitiesStr, &entitiesList); err == nil {\n\t\t\t\tsearch.Entities = entitiesList\n\t\t\t}\n\t\t}\n\t}\n\n\tif relations := data[\"relations\"]; relations != nil {\n\t\tif relationsStr, ok := relations.(string); ok && relationsStr != \"\" {\n\t\t\tvar relationsList []types.Relation\n\t\t\tif err := jsoniter.UnmarshalFromString(relationsStr, &relationsList); err == nil {\n\t\t\t\tsearch.Relations = relationsList\n\t\t\t}\n\t\t}\n\t}\n\n\tif dsl := data[\"dsl\"]; dsl != nil {\n\t\tif dslStr, ok := dsl.(string); ok && dslStr != \"\" {\n\t\t\tvar dslMap map[string]any\n\t\t\tif err := jsoniter.UnmarshalFromString(dslStr, &dslMap); err == nil {\n\t\t\t\tsearch.DSL = dslMap\n\t\t\t}\n\t\t}\n\t}\n\n\tif refs := data[\"references\"]; refs != nil {\n\t\tif refsStr, ok := refs.(string); ok && refsStr != \"\" {\n\t\t\tvar refsList []types.Reference\n\t\t\tif err := jsoniter.UnmarshalFromString(refsStr, &refsList); err == nil {\n\t\t\t\tsearch.References = refsList\n\t\t\t}\n\t\t}\n\t}\n\n\tif graph := data[\"graph\"]; graph != nil {\n\t\tif graphStr, ok := graph.(string); ok && graphStr != \"\" {\n\t\t\tvar graphList []types.GraphNode\n\t\t\tif err := jsoniter.UnmarshalFromString(graphStr, &graphList); err == nil {\n\t\t\t\tsearch.Graph = graphList\n\t\t\t}\n\t\t}\n\t}\n\n\treturn search, nil\n}\n" }, { "path": "agent/store/xun/search_test.go", "content": "package xun_test\n\nimport (\n\t\"fmt\"\n\t\"testing\"\n\t\"time\"\n\n\t\"github.com/yaoapp/yao/agent/store/types\"\n\t\"github.com/yaoapp/yao/agent/store/xun\"\n\t\"github.com/yaoapp/yao/config\"\n\t\"github.com/yaoapp/yao/test\"\n)\n\n// TestSaveSearch tests saving search records\nfunc TestSaveSearch(t *testing.T) {\n\ttest.Prepare(t, config.Conf)\n\tdefer test.Clean()\n\n\tstore, err := xun.NewXun(types.Setting{\n\t\tConnector: \"default\",\n\t})\n\tif err != nil {\n\t\tt.Fatalf(\"Failed to create store: %v\", err)\n\t}\n\n\t// Create a chat first\n\tchat := &types.Chat{\n\t\tAssistantID: \"test_assistant\",\n\t\tTitle: \"Search Test Chat\",\n\t}\n\terr = store.CreateChat(chat)\n\tif err != nil {\n\t\tt.Fatalf(\"Failed to create chat: %v\", err)\n\t}\n\tdefer store.DeleteChat(chat.ChatID)\n\n\tt.Run(\"SaveBasicSearch\", func(t *testing.T) {\n\t\trequestID := fmt.Sprintf(\"req_%d\", time.Now().UnixNano())\n\t\tsearch := &types.Search{\n\t\t\tRequestID: requestID,\n\t\t\tChatID: chat.ChatID,\n\t\t\tQuery: \"What is the weather today?\",\n\t\t\tSource: \"web\",\n\t\t\tDuration: 150,\n\t\t}\n\n\t\terr := store.SaveSearch(search)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to save search: %v\", err)\n\t\t}\n\n\t\t// Verify\n\t\tsearches, err := store.GetSearches(requestID)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to get searches: %v\", err)\n\t\t}\n\n\t\tif len(searches) != 1 {\n\t\t\tt.Fatalf(\"Expected 1 search, got %d\", len(searches))\n\t\t}\n\n\t\tif searches[0].Query != \"What is the weather today?\" {\n\t\t\tt.Errorf(\"Expected query 'What is the weather today?', got '%s'\", searches[0].Query)\n\t\t}\n\t\tif searches[0].Source != \"web\" {\n\t\t\tt.Errorf(\"Expected source 'web', got '%s'\", searches[0].Source)\n\t\t}\n\t\tif searches[0].Duration != 150 {\n\t\t\tt.Errorf(\"Expected duration 150, got %d\", searches[0].Duration)\n\t\t}\n\t})\n\n\tt.Run(\"SaveSearchWithKeywords\", func(t *testing.T) {\n\t\trequestID := fmt.Sprintf(\"req_%d\", time.Now().UnixNano())\n\t\tsearch := &types.Search{\n\t\t\tRequestID: requestID,\n\t\t\tChatID: chat.ChatID,\n\t\t\tQuery: \"Latest news about AI\",\n\t\t\tKeywords: []string{\"AI\", \"news\", \"latest\"},\n\t\t\tSource: \"web\",\n\t\t\tDuration: 200,\n\t\t}\n\n\t\terr := store.SaveSearch(search)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to save search: %v\", err)\n\t\t}\n\n\t\tsearches, err := store.GetSearches(requestID)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to get searches: %v\", err)\n\t\t}\n\n\t\tif len(searches) != 1 {\n\t\t\tt.Fatalf(\"Expected 1 search, got %d\", len(searches))\n\t\t}\n\n\t\tif len(searches[0].Keywords) != 3 {\n\t\t\tt.Errorf(\"Expected 3 keywords, got %d\", len(searches[0].Keywords))\n\t\t}\n\t\tif searches[0].Keywords[0] != \"AI\" {\n\t\t\tt.Errorf(\"Expected first keyword 'AI', got '%s'\", searches[0].Keywords[0])\n\t\t}\n\t})\n\n\tt.Run(\"SaveSearchWithReferences\", func(t *testing.T) {\n\t\trequestID := fmt.Sprintf(\"req_%d\", time.Now().UnixNano())\n\t\tsearch := &types.Search{\n\t\t\tRequestID: requestID,\n\t\t\tChatID: chat.ChatID,\n\t\t\tQuery: \"How to learn Go programming?\",\n\t\t\tSource: \"web\",\n\t\t\tReferences: []types.Reference{\n\t\t\t\t{\n\t\t\t\t\tIndex: 1,\n\t\t\t\t\tType: \"web\",\n\t\t\t\t\tTitle: \"Go Programming Tutorial\",\n\t\t\t\t\tURL: \"https://go.dev/tour/\",\n\t\t\t\t\tSnippet: \"An interactive introduction to Go\",\n\t\t\t\t},\n\t\t\t\t{\n\t\t\t\t\tIndex: 2,\n\t\t\t\t\tType: \"web\",\n\t\t\t\t\tTitle: \"Effective Go\",\n\t\t\t\t\tURL: \"https://go.dev/doc/effective_go\",\n\t\t\t\t\tSnippet: \"Tips for writing clear, idiomatic Go code\",\n\t\t\t\t},\n\t\t\t},\n\t\t\tXML: \"...\",\n\t\t\tPrompt: \"Please cite sources using [1], [2]...\",\n\t\t\tDuration: 300,\n\t\t}\n\n\t\terr := store.SaveSearch(search)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to save search: %v\", err)\n\t\t}\n\n\t\tsearches, err := store.GetSearches(requestID)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to get searches: %v\", err)\n\t\t}\n\n\t\tif len(searches) != 1 {\n\t\t\tt.Fatalf(\"Expected 1 search, got %d\", len(searches))\n\t\t}\n\n\t\tif len(searches[0].References) != 2 {\n\t\t\tt.Errorf(\"Expected 2 references, got %d\", len(searches[0].References))\n\t\t}\n\t\tif searches[0].References[0].Title != \"Go Programming Tutorial\" {\n\t\t\tt.Errorf(\"Expected first reference title 'Go Programming Tutorial', got '%s'\", searches[0].References[0].Title)\n\t\t}\n\t\tif searches[0].XML != \"...\" {\n\t\t\tt.Errorf(\"Expected XML '...', got '%s'\", searches[0].XML)\n\t\t}\n\t\tif searches[0].Prompt != \"Please cite sources using [1], [2]...\" {\n\t\t\tt.Errorf(\"Expected prompt to be set\")\n\t\t}\n\t})\n\n\tt.Run(\"SaveSearchWithConfig\", func(t *testing.T) {\n\t\trequestID := fmt.Sprintf(\"req_%d\", time.Now().UnixNano())\n\t\tsearch := &types.Search{\n\t\t\tRequestID: requestID,\n\t\t\tChatID: chat.ChatID,\n\t\t\tQuery: \"Config test\",\n\t\t\tSource: \"auto\",\n\t\t\tConfig: map[string]any{\n\t\t\t\t\"uses\": map[string]any{\n\t\t\t\t\t\"search\": \"builtin\",\n\t\t\t\t\t\"web\": \"builtin\",\n\t\t\t\t\t\"keyword\": \"builtin\",\n\t\t\t\t},\n\t\t\t\t\"web\": map[string]any{\n\t\t\t\t\t\"provider\": \"tavily\",\n\t\t\t\t\t\"max_results\": 5,\n\t\t\t\t},\n\t\t\t},\n\t\t\tDuration: 100,\n\t\t}\n\n\t\terr := store.SaveSearch(search)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to save search: %v\", err)\n\t\t}\n\n\t\tsearches, err := store.GetSearches(requestID)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to get searches: %v\", err)\n\t\t}\n\n\t\tif len(searches) != 1 {\n\t\t\tt.Fatalf(\"Expected 1 search, got %d\", len(searches))\n\t\t}\n\n\t\tif searches[0].Config == nil {\n\t\t\tt.Fatal(\"Expected config to be set\")\n\t\t}\n\t\tuses, ok := searches[0].Config[\"uses\"].(map[string]any)\n\t\tif !ok {\n\t\t\tt.Fatal(\"Expected uses in config\")\n\t\t}\n\t\tif uses[\"search\"] != \"builtin\" {\n\t\t\tt.Errorf(\"Expected uses.search='builtin', got '%v'\", uses[\"search\"])\n\t\t}\n\t})\n\n\tt.Run(\"SaveSearchWithEntitiesAndRelations\", func(t *testing.T) {\n\t\trequestID := fmt.Sprintf(\"req_%d\", time.Now().UnixNano())\n\t\tsearch := &types.Search{\n\t\t\tRequestID: requestID,\n\t\t\tChatID: chat.ChatID,\n\t\t\tQuery: \"Who is the CEO of Apple?\",\n\t\t\tSource: \"kb\",\n\t\t\tEntities: []types.Entity{\n\t\t\t\t{Name: \"Apple\", Type: \"Organization\"},\n\t\t\t\t{Name: \"Tim Cook\", Type: \"Person\"},\n\t\t\t},\n\t\t\tRelations: []types.Relation{\n\t\t\t\t{Subject: \"Tim Cook\", Predicate: \"CEO_of\", Object: \"Apple\"},\n\t\t\t},\n\t\t\tGraph: []types.GraphNode{\n\t\t\t\t{ID: \"node1\", Type: \"Organization\", Label: \"Apple\", Score: 0.95},\n\t\t\t\t{ID: \"node2\", Type: \"Person\", Label: \"Tim Cook\", Score: 0.92},\n\t\t\t},\n\t\t\tDuration: 250,\n\t\t}\n\n\t\terr := store.SaveSearch(search)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to save search: %v\", err)\n\t\t}\n\n\t\tsearches, err := store.GetSearches(requestID)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to get searches: %v\", err)\n\t\t}\n\n\t\tif len(searches) != 1 {\n\t\t\tt.Fatalf(\"Expected 1 search, got %d\", len(searches))\n\t\t}\n\n\t\tif len(searches[0].Entities) != 2 {\n\t\t\tt.Errorf(\"Expected 2 entities, got %d\", len(searches[0].Entities))\n\t\t}\n\t\tif searches[0].Entities[0].Name != \"Apple\" {\n\t\t\tt.Errorf(\"Expected first entity 'Apple', got '%s'\", searches[0].Entities[0].Name)\n\t\t}\n\n\t\tif len(searches[0].Relations) != 1 {\n\t\t\tt.Errorf(\"Expected 1 relation, got %d\", len(searches[0].Relations))\n\t\t}\n\t\tif searches[0].Relations[0].Predicate != \"CEO_of\" {\n\t\t\tt.Errorf(\"Expected predicate 'CEO_of', got '%s'\", searches[0].Relations[0].Predicate)\n\t\t}\n\n\t\tif len(searches[0].Graph) != 2 {\n\t\t\tt.Errorf(\"Expected 2 graph nodes, got %d\", len(searches[0].Graph))\n\t\t}\n\t})\n\n\tt.Run(\"SaveSearchWithDSL\", func(t *testing.T) {\n\t\trequestID := fmt.Sprintf(\"req_%d\", time.Now().UnixNano())\n\t\tsearch := &types.Search{\n\t\t\tRequestID: requestID,\n\t\t\tChatID: chat.ChatID,\n\t\t\tQuery: \"Find orders over $1000\",\n\t\t\tSource: \"db\",\n\t\t\tDSL: map[string]any{\n\t\t\t\t\"wheres\": []map[string]any{\n\t\t\t\t\t{\"column\": \"amount\", \"op\": \">\", \"value\": 1000},\n\t\t\t\t},\n\t\t\t\t\"orders\": []map[string]any{\n\t\t\t\t\t{\"column\": \"created_at\", \"option\": \"desc\"},\n\t\t\t\t},\n\t\t\t},\n\t\t\tDuration: 50,\n\t\t}\n\n\t\terr := store.SaveSearch(search)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to save search: %v\", err)\n\t\t}\n\n\t\tsearches, err := store.GetSearches(requestID)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to get searches: %v\", err)\n\t\t}\n\n\t\tif len(searches) != 1 {\n\t\t\tt.Fatalf(\"Expected 1 search, got %d\", len(searches))\n\t\t}\n\n\t\tif searches[0].DSL == nil {\n\t\t\tt.Fatal(\"Expected DSL to be set\")\n\t\t}\n\t\twheres, ok := searches[0].DSL[\"wheres\"].([]any)\n\t\tif !ok || len(wheres) == 0 {\n\t\t\tt.Error(\"Expected wheres in DSL\")\n\t\t}\n\t})\n\n\tt.Run(\"SaveSearchWithError\", func(t *testing.T) {\n\t\trequestID := fmt.Sprintf(\"req_%d\", time.Now().UnixNano())\n\t\tsearch := &types.Search{\n\t\t\tRequestID: requestID,\n\t\t\tChatID: chat.ChatID,\n\t\t\tQuery: \"Failed search\",\n\t\t\tSource: \"web\",\n\t\t\tError: \"API rate limit exceeded\",\n\t\t\tDuration: 10,\n\t\t}\n\n\t\terr := store.SaveSearch(search)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to save search: %v\", err)\n\t\t}\n\n\t\tsearches, err := store.GetSearches(requestID)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to get searches: %v\", err)\n\t\t}\n\n\t\tif len(searches) != 1 {\n\t\t\tt.Fatalf(\"Expected 1 search, got %d\", len(searches))\n\t\t}\n\n\t\tif searches[0].Error != \"API rate limit exceeded\" {\n\t\t\tt.Errorf(\"Expected error 'API rate limit exceeded', got '%s'\", searches[0].Error)\n\t\t}\n\t})\n\n\tt.Run(\"SaveSearchWithoutRequestID\", func(t *testing.T) {\n\t\tsearch := &types.Search{\n\t\t\tChatID: chat.ChatID,\n\t\t\tQuery: \"Test\",\n\t\t\tSource: \"web\",\n\t\t}\n\n\t\terr := store.SaveSearch(search)\n\t\tif err == nil {\n\t\t\tt.Error(\"Expected error when saving without request_id\")\n\t\t}\n\t})\n\n\tt.Run(\"SaveSearchWithoutChatID\", func(t *testing.T) {\n\t\tsearch := &types.Search{\n\t\t\tRequestID: \"req_test\",\n\t\t\tQuery: \"Test\",\n\t\t\tSource: \"web\",\n\t\t}\n\n\t\terr := store.SaveSearch(search)\n\t\tif err == nil {\n\t\t\tt.Error(\"Expected error when saving without chat_id\")\n\t\t}\n\t})\n\n\tt.Run(\"SaveSearchWithoutSource\", func(t *testing.T) {\n\t\tsearch := &types.Search{\n\t\t\tRequestID: \"req_test\",\n\t\t\tChatID: chat.ChatID,\n\t\t\tQuery: \"Test\",\n\t\t}\n\n\t\terr := store.SaveSearch(search)\n\t\tif err == nil {\n\t\t\tt.Error(\"Expected error when saving without source\")\n\t\t}\n\t})\n\n\tt.Run(\"SaveNilSearch\", func(t *testing.T) {\n\t\terr := store.SaveSearch(nil)\n\t\tif err == nil {\n\t\t\tt.Error(\"Expected error when saving nil search\")\n\t\t}\n\t})\n}\n\n// TestGetSearches tests retrieving search records\nfunc TestGetSearches(t *testing.T) {\n\ttest.Prepare(t, config.Conf)\n\tdefer test.Clean()\n\n\tstore, err := xun.NewXun(types.Setting{\n\t\tConnector: \"default\",\n\t})\n\tif err != nil {\n\t\tt.Fatalf(\"Failed to create store: %v\", err)\n\t}\n\n\t// Create a chat\n\tchat := &types.Chat{\n\t\tAssistantID: \"test_assistant\",\n\t}\n\terr = store.CreateChat(chat)\n\tif err != nil {\n\t\tt.Fatalf(\"Failed to create chat: %v\", err)\n\t}\n\tdefer store.DeleteChat(chat.ChatID)\n\n\tt.Run(\"GetMultipleSearches\", func(t *testing.T) {\n\t\trequestID := fmt.Sprintf(\"req_%d\", time.Now().UnixNano())\n\n\t\t// Save multiple searches for the same request\n\t\tfor i := 1; i <= 3; i++ {\n\t\t\tsearch := &types.Search{\n\t\t\t\tRequestID: requestID,\n\t\t\t\tChatID: chat.ChatID,\n\t\t\t\tQuery: fmt.Sprintf(\"Query %d\", i),\n\t\t\t\tSource: \"web\",\n\t\t\t\tDuration: int64(i * 100),\n\t\t\t}\n\t\t\terr := store.SaveSearch(search)\n\t\t\tif err != nil {\n\t\t\t\tt.Fatalf(\"Failed to save search %d: %v\", i, err)\n\t\t\t}\n\t\t\ttime.Sleep(10 * time.Millisecond) // Ensure different created_at\n\t\t}\n\n\t\tsearches, err := store.GetSearches(requestID)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to get searches: %v\", err)\n\t\t}\n\n\t\tif len(searches) != 3 {\n\t\t\tt.Errorf(\"Expected 3 searches, got %d\", len(searches))\n\t\t}\n\n\t\t// Verify order (by created_at asc)\n\t\tfor i := 0; i < len(searches)-1; i++ {\n\t\t\tif searches[i].CreatedAt.After(searches[i+1].CreatedAt) {\n\t\t\t\tt.Error(\"Searches not ordered by created_at asc\")\n\t\t\t}\n\t\t}\n\t})\n\n\tt.Run(\"GetSearchesForNonExistentRequest\", func(t *testing.T) {\n\t\tsearches, err := store.GetSearches(\"nonexistent_request\")\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Unexpected error: %v\", err)\n\t\t}\n\t\tif len(searches) != 0 {\n\t\t\tt.Errorf(\"Expected 0 searches, got %d\", len(searches))\n\t\t}\n\t})\n\n\tt.Run(\"GetSearchesWithEmptyRequestID\", func(t *testing.T) {\n\t\t_, err := store.GetSearches(\"\")\n\t\tif err == nil {\n\t\t\tt.Error(\"Expected error when getting searches without request_id\")\n\t\t}\n\t})\n}\n\n// TestGetReference tests retrieving a single reference\nfunc TestGetReference(t *testing.T) {\n\ttest.Prepare(t, config.Conf)\n\tdefer test.Clean()\n\n\tstore, err := xun.NewXun(types.Setting{\n\t\tConnector: \"default\",\n\t})\n\tif err != nil {\n\t\tt.Fatalf(\"Failed to create store: %v\", err)\n\t}\n\n\t// Create a chat\n\tchat := &types.Chat{\n\t\tAssistantID: \"test_assistant\",\n\t}\n\terr = store.CreateChat(chat)\n\tif err != nil {\n\t\tt.Fatalf(\"Failed to create chat: %v\", err)\n\t}\n\tdefer store.DeleteChat(chat.ChatID)\n\n\trequestID := fmt.Sprintf(\"req_%d\", time.Now().UnixNano())\n\n\t// Save search with references\n\tsearch := &types.Search{\n\t\tRequestID: requestID,\n\t\tChatID: chat.ChatID,\n\t\tQuery: \"Test query\",\n\t\tSource: \"web\",\n\t\tReferences: []types.Reference{\n\t\t\t{Index: 1, Type: \"web\", Title: \"Reference 1\", URL: \"https://example.com/1\"},\n\t\t\t{Index: 2, Type: \"web\", Title: \"Reference 2\", URL: \"https://example.com/2\"},\n\t\t\t{Index: 3, Type: \"kb\", Title: \"Reference 3\", Content: \"KB content\"},\n\t\t},\n\t\tDuration: 100,\n\t}\n\terr = store.SaveSearch(search)\n\tif err != nil {\n\t\tt.Fatalf(\"Failed to save search: %v\", err)\n\t}\n\n\tt.Run(\"GetExistingReference\", func(t *testing.T) {\n\t\tref, err := store.GetReference(requestID, 1)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to get reference: %v\", err)\n\t\t}\n\n\t\tif ref.Title != \"Reference 1\" {\n\t\t\tt.Errorf(\"Expected title 'Reference 1', got '%s'\", ref.Title)\n\t\t}\n\t\tif ref.URL != \"https://example.com/1\" {\n\t\t\tt.Errorf(\"Expected URL 'https://example.com/1', got '%s'\", ref.URL)\n\t\t}\n\t})\n\n\tt.Run(\"GetReferenceByIndex\", func(t *testing.T) {\n\t\tref, err := store.GetReference(requestID, 3)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to get reference: %v\", err)\n\t\t}\n\n\t\tif ref.Type != \"kb\" {\n\t\t\tt.Errorf(\"Expected type 'kb', got '%s'\", ref.Type)\n\t\t}\n\t\tif ref.Content != \"KB content\" {\n\t\t\tt.Errorf(\"Expected content 'KB content', got '%s'\", ref.Content)\n\t\t}\n\t})\n\n\tt.Run(\"GetNonExistentReference\", func(t *testing.T) {\n\t\t_, err := store.GetReference(requestID, 999)\n\t\tif err == nil {\n\t\t\tt.Error(\"Expected error when getting non-existent reference\")\n\t\t}\n\t})\n\n\tt.Run(\"GetReferenceWithInvalidIndex\", func(t *testing.T) {\n\t\t_, err := store.GetReference(requestID, 0)\n\t\tif err == nil {\n\t\t\tt.Error(\"Expected error when getting reference with index 0\")\n\t\t}\n\n\t\t_, err = store.GetReference(requestID, -1)\n\t\tif err == nil {\n\t\t\tt.Error(\"Expected error when getting reference with negative index\")\n\t\t}\n\t})\n\n\tt.Run(\"GetReferenceWithEmptyRequestID\", func(t *testing.T) {\n\t\t_, err := store.GetReference(\"\", 1)\n\t\tif err == nil {\n\t\t\tt.Error(\"Expected error when getting reference without request_id\")\n\t\t}\n\t})\n}\n\n// TestDeleteSearches tests deleting search records\nfunc TestDeleteSearches(t *testing.T) {\n\ttest.Prepare(t, config.Conf)\n\tdefer test.Clean()\n\n\tstore, err := xun.NewXun(types.Setting{\n\t\tConnector: \"default\",\n\t})\n\tif err != nil {\n\t\tt.Fatalf(\"Failed to create store: %v\", err)\n\t}\n\n\tt.Run(\"DeleteSearchesForChat\", func(t *testing.T) {\n\t\t// Create a chat\n\t\tchat := &types.Chat{\n\t\t\tAssistantID: \"test_assistant\",\n\t\t}\n\t\terr := store.CreateChat(chat)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to create chat: %v\", err)\n\t\t}\n\t\tdefer store.DeleteChat(chat.ChatID)\n\n\t\t// Save multiple searches\n\t\tfor i := 1; i <= 3; i++ {\n\t\t\trequestID := fmt.Sprintf(\"req_%d_%d\", time.Now().UnixNano(), i)\n\t\t\tsearch := &types.Search{\n\t\t\t\tRequestID: requestID,\n\t\t\t\tChatID: chat.ChatID,\n\t\t\t\tQuery: fmt.Sprintf(\"Query %d\", i),\n\t\t\t\tSource: \"web\",\n\t\t\t\tDuration: 100,\n\t\t\t}\n\t\t\terr := store.SaveSearch(search)\n\t\t\tif err != nil {\n\t\t\t\tt.Fatalf(\"Failed to save search: %v\", err)\n\t\t\t}\n\t\t}\n\n\t\t// Delete all searches for the chat\n\t\terr = store.DeleteSearches(chat.ChatID)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to delete searches: %v\", err)\n\t\t}\n\n\t\t// Note: GetSearches filters by request_id, not chat_id\n\t\t// We can't easily verify deletion without a GetSearchesByChatID method\n\t\t// But the soft delete should have been applied\n\t})\n\n\tt.Run(\"DeleteSearchesWithEmptyChatID\", func(t *testing.T) {\n\t\terr := store.DeleteSearches(\"\")\n\t\tif err == nil {\n\t\t\tt.Error(\"Expected error when deleting searches without chat_id\")\n\t\t}\n\t})\n}\n\n// TestSearchCompleteWorkflow tests a complete search workflow\nfunc TestSearchCompleteWorkflow(t *testing.T) {\n\ttest.Prepare(t, config.Conf)\n\tdefer test.Clean()\n\n\tstore, err := xun.NewXun(types.Setting{\n\t\tConnector: \"default\",\n\t})\n\tif err != nil {\n\t\tt.Fatalf(\"Failed to create store: %v\", err)\n\t}\n\n\tt.Run(\"CompleteWorkflow\", func(t *testing.T) {\n\t\t// 1. Create chat\n\t\tchat := &types.Chat{\n\t\t\tAssistantID: \"workflow_assistant\",\n\t\t\tTitle: \"Search Workflow Test\",\n\t\t}\n\t\terr := store.CreateChat(chat)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to create chat: %v\", err)\n\t\t}\n\t\tdefer store.DeleteChat(chat.ChatID)\n\n\t\trequestID := fmt.Sprintf(\"req_%d\", time.Now().UnixNano())\n\n\t\t// 2. Save search with full data\n\t\tsearch := &types.Search{\n\t\t\tRequestID: requestID,\n\t\t\tChatID: chat.ChatID,\n\t\t\tQuery: \"What are the best practices for Go programming?\",\n\t\t\tConfig: map[string]any{\n\t\t\t\t\"uses\": map[string]any{\"search\": \"builtin\", \"web\": \"builtin\"},\n\t\t\t\t\"web\": map[string]any{\"provider\": \"tavily\", \"max_results\": 5},\n\t\t\t},\n\t\t\tKeywords: []string{\"Go\", \"programming\", \"best practices\"},\n\t\t\tSource: \"auto\",\n\t\t\tReferences: []types.Reference{\n\t\t\t\t{Index: 1, Type: \"web\", Title: \"Effective Go\", URL: \"https://go.dev/doc/effective_go\"},\n\t\t\t\t{Index: 2, Type: \"web\", Title: \"Go Proverbs\", URL: \"https://go-proverbs.github.io/\"},\n\t\t\t\t{Index: 3, Type: \"kb\", Title: \"Internal Go Guide\", Content: \"Our team's Go coding standards...\"},\n\t\t\t},\n\t\t\tXML: \"...\",\n\t\t\tPrompt: \"When citing, use [1], [2], [3] format.\",\n\t\t\tDuration: 350,\n\t\t}\n\n\t\terr = store.SaveSearch(search)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to save search: %v\", err)\n\t\t}\n\n\t\t// 3. Retrieve searches\n\t\tsearches, err := store.GetSearches(requestID)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to get searches: %v\", err)\n\t\t}\n\n\t\tif len(searches) != 1 {\n\t\t\tt.Fatalf(\"Expected 1 search, got %d\", len(searches))\n\t\t}\n\n\t\t// 4. Verify all fields\n\t\ts := searches[0]\n\t\tif s.Query != \"What are the best practices for Go programming?\" {\n\t\t\tt.Errorf(\"Query mismatch\")\n\t\t}\n\t\tif len(s.Keywords) != 3 {\n\t\t\tt.Errorf(\"Expected 3 keywords, got %d\", len(s.Keywords))\n\t\t}\n\t\tif len(s.References) != 3 {\n\t\t\tt.Errorf(\"Expected 3 references, got %d\", len(s.References))\n\t\t}\n\t\tif s.Config == nil {\n\t\t\tt.Error(\"Config should not be nil\")\n\t\t}\n\n\t\t// 5. Get specific reference\n\t\tref, err := store.GetReference(requestID, 2)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to get reference: %v\", err)\n\t\t}\n\t\tif ref.Title != \"Go Proverbs\" {\n\t\t\tt.Errorf(\"Expected 'Go Proverbs', got '%s'\", ref.Title)\n\t\t}\n\n\t\t// 6. Delete searches\n\t\terr = store.DeleteSearches(chat.ChatID)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to delete searches: %v\", err)\n\t\t}\n\n\t\t// 7. Verify deletion (soft delete, so GetSearches should return empty)\n\t\tdeletedSearches, err := store.GetSearches(requestID)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to get searches after delete: %v\", err)\n\t\t}\n\t\tif len(deletedSearches) != 0 {\n\t\t\tt.Errorf(\"Expected 0 searches after delete, got %d\", len(deletedSearches))\n\t\t}\n\n\t\tt.Log(\"Complete search workflow passed!\")\n\t})\n}\n" }, { "path": "agent/store/xun/utils.go", "content": "package xun\n\nimport (\n\t\"fmt\"\n\t\"reflect\"\n\t\"time\"\n\n\tjsoniter \"github.com/json-iterator/go\"\n)\n\n// isNil checks whether a value is truly nil, handling the Go typed-nil-in-interface pitfall.\n// A nil map, slice, or pointer stored in an interface{} is not == nil in Go;\n// this helper uses reflect to detect that case.\nfunc isNil(v interface{}) bool {\n\tif v == nil {\n\t\treturn true\n\t}\n\trv := reflect.ValueOf(v)\n\tswitch rv.Kind() {\n\tcase reflect.Ptr, reflect.Map, reflect.Slice, reflect.Interface, reflect.Chan, reflect.Func:\n\t\treturn rv.IsNil()\n\t}\n\treturn false\n}\n\n// marshalJSONFields serialises each value in fields to a JSON string and writes\n// it into data. Truly-nil values (including typed nils) are skipped so the\n// database column keeps its SQL NULL / default.\nfunc marshalJSONFields(data map[string]interface{}, fields map[string]interface{}) error {\n\tfor field, value := range fields {\n\t\tif isNil(value) {\n\t\t\tcontinue\n\t\t}\n\t\tjsonStr, err := jsoniter.MarshalToString(value)\n\t\tif err != nil {\n\t\t\treturn fmt.Errorf(\"failed to marshal %s: %w\", field, err)\n\t\t}\n\t\tdata[field] = jsonStr\n\t}\n\treturn nil\n}\n\n// Helper functions for type conversion\nfunc getString(data map[string]interface{}, key string) string {\n\tif v, ok := data[key].(string); ok {\n\t\treturn v\n\t}\n\treturn \"\"\n}\n\nfunc getBool(data map[string]interface{}, key string) bool {\n\tswitch v := data[key].(type) {\n\tcase bool:\n\t\treturn v\n\tcase int64:\n\t\treturn v != 0\n\tcase int:\n\t\treturn v != 0\n\tcase float64:\n\t\treturn v != 0\n\t}\n\treturn false\n}\n\nfunc getInt(data map[string]interface{}, key string) int {\n\tswitch v := data[key].(type) {\n\tcase int:\n\t\treturn v\n\tcase int64:\n\t\treturn int(v)\n\tcase float64:\n\t\treturn int(v)\n\t}\n\treturn 0\n}\n\nfunc getInt64(data map[string]interface{}, key string) int64 {\n\tswitch v := data[key].(type) {\n\tcase int64:\n\t\treturn v\n\tcase int:\n\t\treturn int64(v)\n\tcase float64:\n\t\treturn int64(v)\n\tcase string:\n\t\t// Handle string representation of numbers (common with MySQL BIGINT)\n\t\tvar result int64\n\t\tif _, err := fmt.Sscanf(v, \"%d\", &result); err == nil {\n\t\t\treturn result\n\t\t}\n\tcase time.Time:\n\t\t// Handle time.Time from database\n\t\treturn v.UnixNano()\n\t}\n\treturn 0\n}\n\n// toMySQLTime converts UnixNano timestamp to MySQL BIGINT format\nfunc toMySQLTime(unixNano int64) int64 {\n\tif unixNano == 0 {\n\t\treturn 0\n\t}\n\treturn unixNano\n}\n\n// fromMySQLTime converts MySQL BIGINT timestamp to UnixNano\nfunc fromMySQLTime(mysqlTime int64) int64 {\n\treturn mysqlTime\n}\n" }, { "path": "agent/store/xun/utils_test.go", "content": "package xun\n\nimport (\n\t\"testing\"\n\n\t\"github.com/stretchr/testify/assert\"\n\t\"github.com/stretchr/testify/require\"\n)\n\ntype testStruct struct{ Name string }\n\nfunc TestIsNil(t *testing.T) {\n\t// Untyped nil\n\tt.Run(\"UntypedNil\", func(t *testing.T) {\n\t\tassert.True(t, isNil(nil))\n\t})\n\n\t// Typed nil pointer\n\tt.Run(\"TypedNilPointer\", func(t *testing.T) {\n\t\tvar p *testStruct\n\t\tassert.True(t, isNil(p))\n\t})\n\n\t// Typed nil map\n\tt.Run(\"TypedNilMap\", func(t *testing.T) {\n\t\tvar m map[string]string\n\t\tassert.True(t, isNil(m))\n\t})\n\n\t// Typed nil slice\n\tt.Run(\"TypedNilSlice\", func(t *testing.T) {\n\t\tvar s []string\n\t\tassert.True(t, isNil(s))\n\t})\n\n\t// Non-nil pointer\n\tt.Run(\"NonNilPointer\", func(t *testing.T) {\n\t\tp := &testStruct{Name: \"test\"}\n\t\tassert.False(t, isNil(p))\n\t})\n\n\t// Non-nil map (empty)\n\tt.Run(\"NonNilEmptyMap\", func(t *testing.T) {\n\t\tm := map[string]string{}\n\t\tassert.False(t, isNil(m))\n\t})\n\n\t// Non-nil map with values\n\tt.Run(\"NonNilMap\", func(t *testing.T) {\n\t\tm := map[string]string{\"a\": \"1\"}\n\t\tassert.False(t, isNil(m))\n\t})\n\n\t// Non-nil slice (empty)\n\tt.Run(\"NonNilEmptySlice\", func(t *testing.T) {\n\t\ts := []string{}\n\t\tassert.False(t, isNil(s))\n\t})\n\n\t// Non-nil slice with values\n\tt.Run(\"NonNilSlice\", func(t *testing.T) {\n\t\ts := []string{\"a\"}\n\t\tassert.False(t, isNil(s))\n\t})\n\n\t// Scalar types (never nil)\n\tt.Run(\"String\", func(t *testing.T) {\n\t\tassert.False(t, isNil(\"hello\"))\n\t})\n\tt.Run(\"EmptyString\", func(t *testing.T) {\n\t\tassert.False(t, isNil(\"\"))\n\t})\n\tt.Run(\"Int\", func(t *testing.T) {\n\t\tassert.False(t, isNil(42))\n\t})\n\tt.Run(\"Bool\", func(t *testing.T) {\n\t\tassert.False(t, isNil(false))\n\t})\n}\n\nfunc TestMarshalJSONFields(t *testing.T) {\n\tt.Run(\"SkipUntypedNil\", func(t *testing.T) {\n\t\tdata := make(map[string]interface{})\n\t\terr := marshalJSONFields(data, map[string]interface{}{\n\t\t\t\"field1\": nil,\n\t\t})\n\t\trequire.NoError(t, err)\n\t\t_, exists := data[\"field1\"]\n\t\tassert.False(t, exists, \"untyped nil should be skipped\")\n\t})\n\n\tt.Run(\"SkipTypedNilMap\", func(t *testing.T) {\n\t\tdata := make(map[string]interface{})\n\t\tvar m map[string]string\n\t\terr := marshalJSONFields(data, map[string]interface{}{\n\t\t\t\"deps\": m,\n\t\t})\n\t\trequire.NoError(t, err)\n\t\t_, exists := data[\"deps\"]\n\t\tassert.False(t, exists, \"typed nil map should be skipped\")\n\t})\n\n\tt.Run(\"SkipTypedNilSlice\", func(t *testing.T) {\n\t\tdata := make(map[string]interface{})\n\t\tvar s []string\n\t\terr := marshalJSONFields(data, map[string]interface{}{\n\t\t\t\"tags\": s,\n\t\t})\n\t\trequire.NoError(t, err)\n\t\t_, exists := data[\"tags\"]\n\t\tassert.False(t, exists, \"typed nil slice should be skipped\")\n\t})\n\n\tt.Run(\"SkipTypedNilPointer\", func(t *testing.T) {\n\t\tdata := make(map[string]interface{})\n\t\tvar p *testStruct\n\t\terr := marshalJSONFields(data, map[string]interface{}{\n\t\t\t\"kb\": p,\n\t\t})\n\t\trequire.NoError(t, err)\n\t\t_, exists := data[\"kb\"]\n\t\tassert.False(t, exists, \"typed nil pointer should be skipped\")\n\t})\n\n\tt.Run(\"MarshalNonNilMap\", func(t *testing.T) {\n\t\tdata := make(map[string]interface{})\n\t\terr := marshalJSONFields(data, map[string]interface{}{\n\t\t\t\"deps\": map[string]string{\"echo\": \"^1.0.0\"},\n\t\t})\n\t\trequire.NoError(t, err)\n\t\tassert.Equal(t, `{\"echo\":\"^1.0.0\"}`, data[\"deps\"])\n\t})\n\n\tt.Run(\"MarshalEmptyMap\", func(t *testing.T) {\n\t\tdata := make(map[string]interface{})\n\t\terr := marshalJSONFields(data, map[string]interface{}{\n\t\t\t\"deps\": map[string]string{},\n\t\t})\n\t\trequire.NoError(t, err)\n\t\tassert.Equal(t, `{}`, data[\"deps\"])\n\t})\n\n\tt.Run(\"MarshalSlice\", func(t *testing.T) {\n\t\tdata := make(map[string]interface{})\n\t\terr := marshalJSONFields(data, map[string]interface{}{\n\t\t\t\"tags\": []string{\"ai\", \"bot\"},\n\t\t})\n\t\trequire.NoError(t, err)\n\t\tassert.Equal(t, `[\"ai\",\"bot\"]`, data[\"tags\"])\n\t})\n\n\tt.Run(\"MarshalPointer\", func(t *testing.T) {\n\t\tdata := make(map[string]interface{})\n\t\terr := marshalJSONFields(data, map[string]interface{}{\n\t\t\t\"kb\": &testStruct{Name: \"test\"},\n\t\t})\n\t\trequire.NoError(t, err)\n\t\tassert.Equal(t, `{\"Name\":\"test\"}`, data[\"kb\"])\n\t})\n\n\tt.Run(\"MixedNilAndNonNil\", func(t *testing.T) {\n\t\tdata := make(map[string]interface{})\n\t\tvar nilMap map[string]string\n\t\tvar nilSlice []string\n\t\tvar nilPtr *testStruct\n\n\t\terr := marshalJSONFields(data, map[string]interface{}{\n\t\t\t\"nil_map\": nilMap,\n\t\t\t\"nil_slice\": nilSlice,\n\t\t\t\"nil_ptr\": nilPtr,\n\t\t\t\"nil_raw\": nil,\n\t\t\t\"good_map\": map[string]string{\"k\": \"v\"},\n\t\t\t\"good_list\": []string{\"a\"},\n\t\t})\n\t\trequire.NoError(t, err)\n\n\t\tassert.Len(t, data, 2, \"only non-nil fields should be written\")\n\t\tassert.Equal(t, `{\"k\":\"v\"}`, data[\"good_map\"])\n\t\tassert.Equal(t, `[\"a\"]`, data[\"good_list\"])\n\n\t\t_, exists := data[\"nil_map\"]\n\t\tassert.False(t, exists)\n\t\t_, exists = data[\"nil_slice\"]\n\t\tassert.False(t, exists)\n\t\t_, exists = data[\"nil_ptr\"]\n\t\tassert.False(t, exists)\n\t\t_, exists = data[\"nil_raw\"]\n\t\tassert.False(t, exists)\n\t})\n}\n" }, { "path": "agent/store/xun/xun.go", "content": "package xun\n\nimport (\n\t\"fmt\"\n\t\"time\"\n\n\tjsoniter \"github.com/json-iterator/go\"\n\t\"github.com/yaoapp/gou/connector\"\n\t\"github.com/yaoapp/gou/model\"\n\t\"github.com/yaoapp/xun/capsule\"\n\t\"github.com/yaoapp/xun/dbal/query\"\n\t\"github.com/yaoapp/xun/dbal/schema\"\n\t\"github.com/yaoapp/yao/agent/store/types\"\n)\n\n// Package store provides functionality for managing chat conversations and assistants.\n\n// Xun implements the Store interface using a database backend.\n// It provides functionality for:\n// - Managing chat sessions and their messages\n// - Organizing chats with pagination and date-based grouping\n// - Handling chat metadata like titles and creation dates\n// - Managing AI assistants with their configurations and metadata\n// - Managing resume records for recovery from interruptions\ntype Xun struct {\n\tquery query.Query\n\tschema schema.Schema\n\tsetting types.Setting\n}\n\n// Public interface methods:\n//\n// NewXun creates a new store instance with the given settings\n//\n// Chat Management:\n// CreateChat creates a new chat session\n// GetChat retrieves a single chat by ID\n// UpdateChat updates chat fields\n// DeleteChat deletes a chat and its associated messages\n// ListChats retrieves a paginated list of chats with optional grouping\n//\n// Message Management:\n// SaveMessages batch saves messages for a chat\n// GetMessages retrieves messages for a chat with filtering\n// UpdateMessage updates a single message\n// DeleteMessages deletes specific messages from a chat\n//\n// Resume Management:\n// SaveResume batch saves resume records (only on failure/interrupt)\n// GetResume retrieves all resume records for a chat\n// GetLastResume retrieves the last resume record for a chat\n// GetResumeByStackID retrieves resume records for a specific stack\n// GetStackPath returns the stack path from root to the given stack\n// DeleteResume deletes all resume records for a chat\n//\n// Assistant Management:\n// SaveAssistant creates or updates an assistant\n// UpdateAssistant updates assistant fields\n// DeleteAssistant deletes an assistant by assistant_id\n// GetAssistants retrieves a paginated list of assistants with filtering\n// GetAssistant retrieves a single assistant by assistant_id\n// DeleteAssistants deletes assistants based on filter conditions\n// GetAssistantTags retrieves all unique tags from assistants\n\n// NewXun create a new xun store\nfunc NewXun(setting types.Setting) (types.Store, error) {\n\tstore := &Xun{setting: setting}\n\tif setting.Connector == \"default\" || setting.Connector == \"\" {\n\t\tstore.query = capsule.Global.Query()\n\t\tstore.schema = capsule.Global.Schema()\n\t} else {\n\t\tconn, err := connector.Select(setting.Connector)\n\t\tif err != nil {\n\t\t\treturn nil, fmt.Errorf(\"select store connector %s error: %s\", setting.Connector, err.Error())\n\t\t}\n\n\t\tstore.query, err = conn.Query()\n\t\tif err != nil {\n\t\t\treturn nil, fmt.Errorf(\"query store connector %s error: %s\", setting.Connector, err.Error())\n\t\t}\n\n\t\tstore.schema, err = conn.Schema()\n\t\tif err != nil {\n\t\t\treturn nil, err\n\t\t}\n\t}\n\n\treturn store, nil\n}\n\n// =============================================================================\n// Query Builders\n// =============================================================================\n\n// newQueryChat creates a new query builder for the chat table\nfunc (store *Xun) newQueryChat() query.Query {\n\tqb := store.query.New()\n\tqb.Table(store.getChatTable())\n\treturn qb\n}\n\n// newQueryMessage creates a new query builder for the message table\nfunc (store *Xun) newQueryMessage() query.Query {\n\tqb := store.query.New()\n\tqb.Table(store.getMessageTable())\n\treturn qb\n}\n\n// newQueryResume creates a new query builder for the resume table\nfunc (store *Xun) newQueryResume() query.Query {\n\tqb := store.query.New()\n\tqb.Table(store.getResumeTable())\n\treturn qb\n}\n\n// newQueryAssistant creates a new query builder for the assistant table\nfunc (store *Xun) newQueryAssistant() query.Query {\n\tqb := store.query.New()\n\tqb.Table(store.getAssistantTable())\n\treturn qb\n}\n\n// =============================================================================\n// Table Name Getters\n// =============================================================================\n\n// getChatTable returns the chat table name\nfunc (store *Xun) getChatTable() string {\n\tm := model.Select(\"__yao.agent.chat\")\n\tif m != nil && m.MetaData.Table.Name != \"\" {\n\t\treturn m.MetaData.Table.Name\n\t}\n\treturn \"agent_chat\"\n}\n\n// getMessageTable returns the message table name\nfunc (store *Xun) getMessageTable() string {\n\tm := model.Select(\"__yao.agent.message\")\n\tif m != nil && m.MetaData.Table.Name != \"\" {\n\t\treturn m.MetaData.Table.Name\n\t}\n\treturn \"agent_message\"\n}\n\n// getResumeTable returns the resume table name\nfunc (store *Xun) getResumeTable() string {\n\tm := model.Select(\"__yao.agent.resume\")\n\tif m != nil && m.MetaData.Table.Name != \"\" {\n\t\treturn m.MetaData.Table.Name\n\t}\n\treturn \"agent_resume\"\n}\n\n// getAssistantTable returns the assistant table name\nfunc (store *Xun) getAssistantTable() string {\n\tm := model.Select(\"__yao.agent.assistant\")\n\tif m != nil && m.MetaData.Table.Name != \"\" {\n\t\treturn m.MetaData.Table.Name\n\t}\n\treturn \"agent_assistant\"\n}\n\n// =============================================================================\n// Utility Functions\n// =============================================================================\n\n// parseJSONFields parses JSON string fields into their corresponding Go types\nfunc (store *Xun) parseJSONFields(data map[string]interface{}, fields []string) {\n\tfor _, field := range fields {\n\t\tif val := data[field]; val != nil {\n\t\t\tvar jsonStr string\n\t\t\tswitch v := val.(type) {\n\t\t\tcase string:\n\t\t\t\tjsonStr = v\n\t\t\tcase []byte:\n\t\t\t\tjsonStr = string(v)\n\t\t\tdefault:\n\t\t\t\tcontinue\n\t\t\t}\n\t\t\tif jsonStr != \"\" {\n\t\t\t\tvar parsed interface{}\n\t\t\t\tif err := jsoniter.UnmarshalFromString(jsonStr, &parsed); err == nil {\n\t\t\t\t\tdata[field] = parsed\n\t\t\t\t}\n\t\t\t}\n\t\t}\n\t}\n}\n\n// GenerateAssistantID generates a random-looking 6-digit ID\nfunc (store *Xun) GenerateAssistantID() (string, error) {\n\tmaxAttempts := 10 // Maximum number of attempts to generate a unique ID\n\tfor i := 0; i < maxAttempts; i++ {\n\t\t// Generate a random number using timestamp and some bit operations\n\t\ttimestamp := time.Now().UnixNano()\n\t\trandom := (timestamp ^ (timestamp >> 12)) % 1000000\n\t\thash := fmt.Sprintf(\"%06d\", random)\n\n\t\t// Check if this ID already exists\n\t\texists, err := store.query.New().\n\t\t\tTable(store.getAssistantTable()).\n\t\t\tWhere(\"assistant_id\", hash).\n\t\t\tExists()\n\n\t\tif err != nil {\n\t\t\treturn \"\", err\n\t\t}\n\n\t\tif !exists {\n\t\t\treturn hash, nil\n\t\t}\n\n\t\t// If ID exists, wait a bit and try again\n\t\ttime.Sleep(time.Millisecond)\n\t}\n\n\treturn \"\", fmt.Errorf(\"failed to generate unique ID after %d attempts\", maxAttempts)\n}\n" }, { "path": "agent/test/DESIGN.md", "content": "# Agent Test Package Design\n\n## Overview\n\nAgent Test Package provides a framework for testing AI agents with structured test cases.\nIt supports batch testing, report generation, stability analysis, and CI integration.\n\nAdditionally, it supports **Script Testing** for testing Agent handler scripts (hooks, tools, etc.) with a Go-like testing interface.\n\n### Quick Start\n\n```bash\n# Quick test with a single message (auto-detect agent from current directory)\ncd assistants/keyword\nyao agent test -i \"hello world\"\n\n# Or specify agent explicitly\nyao agent test -i \"hello world\" -n keyword.agent\n\n# Run tests from JSONL file (auto-detect agent from path)\nyao agent test -i assistants/keyword/tests/inputs.jsonl\n\n# Run with stability analysis (5 runs per test case)\nyao agent test -i assistants/keyword/tests/inputs.jsonl --runs 5\n\n# Generate HTML report\nyao agent test -i assistants/keyword/tests/inputs.jsonl -r report.html -o report.html\n\n# Run script tests (test agent handler scripts)\nyao agent test -i scripts.expense.setup -v\n\n# Run script tests with specific user/team context\nyao agent test -i scripts.expense.tools -u admin -t ops-team -v\n```\n\n## Usage\n\n```bash\n# Basic usage - auto-detect agent, output to same directory as input\n# Output: tests/output-20241217100000.jsonl\nyao agent test -i tests/inputs.jsonl\n\n# Override connector\nyao agent test -i tests/inputs.jsonl -c openai.gpt4\n\n# Specify agent explicitly\nyao agent test -i tests/inputs.jsonl -n my.agent\n\n# Specify test environment (user and team)\nyao agent test -i tests/inputs.jsonl -u test-user -t test-team\n\n# Run multiple times for stability analysis\nyao agent test -i tests/inputs.jsonl --runs 5\n\n# Custom timeout per test case (default: 5m)\nyao agent test -i tests/inputs.jsonl --timeout 10m\n\n# Run tests in parallel (4 concurrent test cases)\nyao agent test -i tests/inputs.jsonl --parallel 4\n\n# Combine parallel and timeout for faster execution\nyao agent test -i tests/inputs.jsonl --parallel 8 --timeout 2m\n\n# Custom output file path\nyao agent test -i tests/inputs.jsonl -o /path/to/results.jsonl\n\n# Use custom reporter agent for personalized report (HTML)\nyao agent test -i tests/inputs.jsonl -r report.html -o report.html\n\n# Use custom reporter agent for personalized report (Markdown)\nyao agent test -i tests/inputs.jsonl -r report.markdown -o report.md\n\n# Full example with all options\nyao agent test -i tests/inputs.jsonl \\\n -n keyword.agent \\\n -c deepseek.v3 \\\n -u test-user \\\n -t test-team \\\n --runs 3 \\\n --timeout 10m \\\n --parallel 4 \\\n -r report.html \\\n -o report.html\n```\n\n### Input Modes\n\nThe `-i` flag supports three input modes:\n\n**1. JSONL File Mode** - Load test cases from a file:\n\n```bash\nyao agent test -i tests/inputs.jsonl\n```\n\n**2. Direct Message Mode** - Test with a single message:\n\n```bash\n# Auto-detect agent from current working directory\ncd assistants/keyword\nyao agent test -i \"Extract keywords from this text\"\n\n# Or specify agent explicitly\nyao agent test -i \"Extract keywords from this text\" -n keyword.agent\nyao agent test -i \"你好世界\" -n keyword.agent -c deepseek.v3\n```\n\nWhen using direct message mode:\n\n- Agent is resolved from current working directory (looks for `package.yao` upward)\n- If not found, use `-n` flag to specify agent explicitly\n- Output is printed to stdout (or saved to `-o` if specified)\n- Useful for quick testing and debugging\n\n**3. Script Test Mode** - Test agent handler scripts:\n\n```bash\n# Run all tests in a script module\nyao agent test -i scripts.expense.setup -v\n\n# Run with specific user/team context\nyao agent test -i scripts.expense.tools -u admin -t ops-team\n\n# Run with timeout\nyao agent test -i scripts.expense.setup --timeout 30s\n\n# Run specific tests by pattern (like go test -run)\nyao agent test -i scripts.expense.setup -run TestSystemReady\n\n# Run tests matching a regex pattern\nyao agent test -i scripts.expense.setup -run \"TestSystem.*\"\n```\n\nWhen using script test mode:\n\n- Input starts with `scripts.` prefix to indicate script testing\n- Maps to the script file (e.g., `scripts.expense.setup` → `expense/src/setup_test.ts`)\n- Automatically discovers and runs all `Test*` functions in the script\n- Uses Go-like testing interface with assertions\n- See [Script Testing](#script-testing) section for details\n\n### Default Output Path\n\nWhen `-o` is not specified and using JSONL file mode, the output file is automatically generated in the same directory as the input file:\n\n```\n{input_directory}/output-{timestamp}.jsonl\n```\n\nExample:\n\n- Input: `/app/assistants/keyword/tests/inputs.jsonl`\n- Output: `/app/assistants/keyword/tests/output-20241217100000.jsonl`\n\nThe timestamp format is `YYYYMMDDHHMMSS` (e.g., `20241217100000` for 2024-12-17 10:00:00).\n\nWhen using direct message mode without `-o`, output is printed to stdout.\n\n## Command Line Options\n\n| Flag | Long Flag | Description | Default | Example |\n| ---- | ------------- | ------------------------------------- | -------------------------- | --------------------------------------- |\n| `-i` | `--input` | Input: JSONL file path or message | - | `-i tests/inputs.jsonl` or `-i \"hello\"` |\n| `-o` | `--output` | Path to output file (format by ext) | `output-{timestamp}.jsonl` | `-o report.html` |\n| `-n` | `--name` | Explicit agent ID | auto-detect | `-n keyword.agent` |\n| `-c` | `--connector` | Override connector | agent default | `-c openai.gpt4` |\n| `-u` | `--user` | Test user ID (global override) | \"test-user\" | `-u admin` |\n| `-t` | `--team` | Test team ID (global override) | \"test-team\" | `-t ops-team` |\n| | `--ctx` | Path to context JSON file | - | `--ctx tests/context.json` |\n| `-r` | `--reporter` | Custom reporter agent ID | - (use built-in) | `-r report.beautiful` |\n| | `--runs` | Number of runs for stability analysis | 1 | `--runs 5` |\n| | `--run` | Regex pattern to filter tests | - | `--run \"TestSystem.*\"` |\n| | `--timeout` | Default timeout per test case | 5m | `--timeout 10m` |\n| | `--parallel` | Number of parallel test cases | 1 | `--parallel 4` |\n| `-v` | `--verbose` | Verbose output | false | `-v` |\n| | `--fail-fast` | Stop on first failure | false | `--fail-fast` |\n\n**Notes**:\n\n- Without `-o` flag, output is saved to `{input_dir}/output-{timestamp}.jsonl`\n- Output format is determined by `-o` file extension: `.jsonl`, `.json`, `.md`, `.html`\n- Use `-r` to specify a custom reporter agent for personalized report generation\n\n## Agent Resolution\n\nThe agent is resolved in the following order:\n\n1. **Explicit specification** (`-n` flag): Use the specified agent ID\n2. **Path-based detection**: Traverse up from `tests/inputs.jsonl` to find `package.yao`\n\n### Path-based Detection Example\n\n```\n/app/assistants/workers/system/keyword/\n├── package.yao <- Agent definition\n├── prompts.yml\n├── src/\n│ └── index.ts\n└── tests/\n └── inputs.jsonl <- Test input file\n```\n\nGiven input path `/app/assistants/workers/system/keyword/tests/inputs.jsonl`:\n\n1. Check `/app/assistants/workers/system/keyword/tests/package.yao` - not found\n2. Check `/app/assistants/workers/system/keyword/package.yao` - **found!**\n3. Load agent from `/app/assistants/workers/system/keyword/`\n\n## Test Environment\n\nAgent calls require a `Context` with user and tenant information. The test framework creates a test context with configurable environment:\n\n```go\n// TestEnvironment configures the test execution context\ntype TestEnvironment struct {\n UserID string // User ID for authorized info (-u flag)\n TeamID string // Team ID for authorized info (-t flag)\n Locale string // Locale (default: \"en-us\")\n ClientType string // Client type (default: \"test\")\n ClientIP string // Client IP (default: \"127.0.0.1\")\n Referer string // Request referer (default: \"test\")\n Accept string // Accept format (default: \"standard\")\n}\n```\n\nExample context creation (similar to `agent_next_test.go`):\n\n```go\nfunc newTestContext(env *TestEnvironment, chatID, assistantID string) *context.Context {\n authorized := &types.AuthorizedInfo{\n Subject: env.UserID,\n UserID: env.UserID,\n TeamID: env.TeamID,\n }\n ctx := context.New(stdContext.Background(), authorized, chatID)\n ctx.AssistantID = assistantID\n ctx.Locale = env.Locale\n ctx.Client = context.Client{\n Type: env.ClientType,\n IP: env.ClientIP,\n }\n ctx.Referer = env.Referer\n ctx.Accept = env.Accept\n return ctx\n}\n```\n\n## Stability Analysis (Multiple Runs)\n\nWhen `--runs N` is specified (N > 1), the framework runs each test case N times and collects stability metrics:\n\n### Stability Metrics\n\n| Metric | Description |\n| ------------------ | ------------------------------------------ |\n| `pass_rate` | Percentage of runs that passed (0-100%) |\n| `consistency` | How consistent the outputs are across runs |\n| `avg_duration_ms` | Average execution time |\n| `min_duration_ms` | Minimum execution time |\n| `max_duration_ms` | Maximum execution time |\n| `std_deviation_ms` | Standard deviation of execution time |\n\n### Stability Report Structure\n\n```json\n{\n \"summary\": {\n \"total_cases\": 42,\n \"total_runs\": 126,\n \"runs_per_case\": 3,\n \"overall_pass_rate\": 95.2,\n \"stable_cases\": 38,\n \"unstable_cases\": 4,\n \"duration_ms\": 45678\n },\n \"results\": [\n {\n \"id\": \"T001\",\n \"runs\": 3,\n \"passed\": 3,\n \"failed\": 0,\n \"pass_rate\": 100.0,\n \"consistency\": 1.0,\n \"stable\": true,\n \"avg_duration_ms\": 234,\n \"min_duration_ms\": 210,\n \"max_duration_ms\": 256,\n \"std_deviation_ms\": 18.5,\n \"run_details\": [\n {\"run\": 1, \"status\": \"passed\", \"duration_ms\": 234, \"output\": {...}},\n {\"run\": 2, \"status\": \"passed\", \"duration_ms\": 210, \"output\": {...}},\n {\"run\": 3, \"status\": \"passed\", \"duration_ms\": 256, \"output\": {...}}\n ]\n },\n {\n \"id\": \"T002\",\n \"runs\": 3,\n \"passed\": 2,\n \"failed\": 1,\n \"pass_rate\": 66.7,\n \"consistency\": 0.67,\n \"stable\": false,\n \"run_details\": [...]\n }\n ]\n}\n```\n\n### Stability Classification\n\n| Pass Rate | Classification |\n| --------- | --------------- |\n| 100% | Stable |\n| 80-99% | Mostly Stable |\n| 50-79% | Unstable |\n| < 50% | Highly Unstable |\n\n## Custom Reporter Agent\n\nBy default, the framework outputs JSONL format. You can specify a reporter agent (`-r` flag) for personalized report generation:\n\n### Reporter Agent Interface\n\nThe reporter agent receives the test results and generates a custom report:\n\n```json\n// Input to reporter agent\n{\n \"report\": {\n \"summary\": {...},\n \"results\": [...],\n \"metadata\": {...}\n },\n \"format\": \"html\", // or \"markdown\", \"text\"\n \"options\": {\n \"verbose\": true,\n \"include_outputs\": true\n }\n}\n```\n\n### Built-in Reporter Agents\n\n| Agent ID | Description |\n| ----------------- | -------------------------------------- |\n| `report.json` | JSON format (default, no agent needed) |\n| `report.markdown` | Markdown format with tables |\n| `report.html` | Interactive HTML report |\n| `report.summary` | Brief text summary |\n\n### Custom Reporter Example\n\nCreate a custom reporter agent at `assistants/reporters/my-reporter/`:\n\n```yaml\n# prompts.yml\n- role: system\n content: |\n You are a test report generator. Generate a beautiful report from test results.\n\n Output format: HTML with embedded CSS\n\n Requirements:\n - Show summary statistics prominently\n - Use color coding (green=pass, red=fail)\n - Include charts for stability metrics\n - Make it printable\n```\n\n## Script Testing\n\nScript Testing allows you to test Agent handler scripts (hooks, tools, setup functions, etc.) using a Go-like testing interface. This is useful for unit testing individual functions in your agent's TypeScript/JavaScript code.\n\n### Quick Start\n\n```bash\n# Run script tests\nyao agent test -i scripts.expense.setup -v\n\n# With user/team context\nyao agent test -i scripts.expense.setup -u admin -t ops-team -v\n\n# With timeout\nyao agent test -i scripts.expense.setup --timeout 30s -v\n```\n\n### Script Resolution\n\nThe `scripts.` prefix indicates script test mode. The script is resolved as follows:\n\n| Input | Script Path | Test File |\n| ----------------------- | ---------------------- | --------------------------- |\n| `scripts.expense.setup` | `expense/src/setup.ts` | `expense/src/setup_test.ts` |\n| `scripts.expense.tools` | `expense/src/tools.ts` | `expense/src/tools_test.ts` |\n| `scripts.keyword.index` | `keyword/src/index.ts` | `keyword/src/index_test.ts` |\n\nThe test file naming convention is `{module}_test.ts` (similar to Go's `_test.go` convention).\n\n### Test Function Signature\n\nTest functions must follow this signature:\n\n```typescript\nfunction TestFunctionName(t: testing.T, ctx: agent.Context) {\n // Test logic here\n}\n```\n\n**Requirements:**\n\n- Function name must start with `Test` (case-sensitive)\n- First parameter `t` is the testing object with assertions\n- Second parameter `ctx` is the agent context (same as used in hooks/tools)\n- Functions not starting with `Test` are ignored (can be used as helpers)\n\n### Example Test File\n\n```typescript\n// setup_test.ts\n// @ts-nocheck\n\n// Test the SystemReady function\nfunction TestSystemReady(t: testing.T, ctx: agent.Context) {\n const { assert } = t;\n\n // Call the function being tested\n const result = SystemReady(ctx);\n\n // Assert the result\n assert.True(result, \"SystemReady should return true\");\n}\n\n// Test error case\nfunction TestSystemReadyWithInvalidContext(t: testing.T, ctx: agent.Context) {\n const { assert } = t;\n\n // Modify context to simulate error condition\n ctx.User = null;\n\n const result = SystemReady(ctx);\n assert.False(result, \"SystemReady should return false when user is null\");\n}\n\n// Helper function (not a test - doesn't start with \"Test\")\nfunction createMockData() {\n return { id: 1, name: \"test\" };\n}\n\n// Test with helper\nfunction TestSetupWithMockData(t: testing.T, ctx: agent.Context) {\n const { assert } = t;\n const mockData = createMockData();\n\n const result = Setup(ctx, mockData);\n assert.NotNil(result, \"Setup should return a result\");\n assert.Equal(result.id, 1, \"Result ID should match\");\n}\n```\n\n### Testing Object (`t`)\n\nThe `t` parameter provides the testing interface:\n\n```typescript\ninterface testing.T {\n // Assertions object\n assert: testing.Assert;\n\n // Test metadata\n name: string; // Current test function name\n failed: boolean; // Whether the test has failed\n\n // Logging (output appears in test report)\n log(...args: any[]): void; // Log info message\n error(...args: any[]): void; // Log error message\n\n // Control flow\n skip(reason?: string): void; // Skip this test\n fail(reason?: string): void; // Mark test as failed\n fatal(reason?: string): void; // Mark as failed and stop execution\n}\n```\n\n### Assertions (`t.assert`)\n\nThe `assert` object provides assertion methods:\n\n| Method | Description |\n| ----------------------------------- | ---------------------------------- |\n| `True(value, message?)` | Assert value is true |\n| `False(value, message?)` | Assert value is false |\n| `Equal(actual, expected, message?)` | Assert deep equality |\n| `NotEqual(actual, expected, msg?)` | Assert not equal |\n| `Nil(value, message?)` | Assert value is null/undefined |\n| `NotNil(value, message?)` | Assert value is not null/undefined |\n| `Contains(str, substr, message?)` | Assert string contains substring |\n| `NotContains(str, substr, msg?)` | Assert string does not contain |\n| `Len(value, length, message?)` | Assert array/string length |\n| `Greater(a, b, message?)` | Assert a > b |\n| `GreaterOrEqual(a, b, message?)` | Assert a >= b |\n| `Less(a, b, message?)` | Assert a < b |\n| `LessOrEqual(a, b, message?)` | Assert a <= b |\n| `Error(err, message?)` | Assert err is an error |\n| `NoError(err, message?)` | Assert err is null/undefined |\n| `Panic(fn, message?)` | Assert function throws |\n| `NoPanic(fn, message?)` | Assert function does not throw |\n| `Match(value, pattern, message?)` | Assert value matches regex |\n| `NotMatch(value, pattern, msg?)` | Assert value does not match regex |\n| `JSONPath(obj, path, expected, m?)` | Assert JSON path value |\n| `Type(value, typeName, message?)` | Assert value type |\n\n### Agent Context (`ctx`)\n\nThe `ctx` parameter is the same `agent.Context` used in agent hooks and tools:\n\n```typescript\ninterface agent.Context {\n // User information (from -u flag or default)\n User: {\n ID: string;\n Name?: string;\n };\n\n // Team information (from -t flag or default)\n Team: {\n ID: string;\n Name?: string;\n };\n\n // Locale (default: \"en-us\")\n Locale: string;\n\n // Client information\n Client: {\n Type: string; // \"test\"\n IP: string; // \"127.0.0.1\"\n };\n\n // Metadata (can be set via test case)\n Metadata: Record;\n\n // Chat/Session ID\n ChatID: string;\n\n // Assistant ID (resolved from script path)\n AssistantID: string;\n}\n```\n\n### Script Test Output\n\nScript test results are reported in the same format as agent tests:\n\n```\n═══════════════════════════════════════════════════════════════════════════════\n Script Test: scripts.expense.setup\n═══════════════════════════════════════════════════════════════════════════════\n Script: expense/src/setup_test.ts\n Tests: 3 functions\n User: test-user\n Team: test-team\n───────────────────────────────────────────────────────────────────────────────\n Running Tests\n───────────────────────────────────────────────────────────────────────────────\n► [TestSystemReady] ...\n ✓ PASSED (12ms)\n\n► [TestSystemReadyWithInvalidContext] ...\n ✓ PASSED (8ms)\n\n► [TestSetupWithMockData] ...\n ✗ FAILED (15ms)\n └─ assertion failed: Result ID should match\n expected: 1\n actual: 2\n\n═══════════════════════════════════════════════════════════════════════════════\n Summary: 2 passed, 1 failed, 0 skipped (35ms)\n═══════════════════════════════════════════════════════════════════════════════\n```\n\n### Script Test Options\n\nScript tests support the following command line options:\n\n| Flag | Description | Default | Example |\n| ------------- | -------------------------------- | ----------- | -------------------- |\n| `-u` | User ID for context | \"test-user\" | `-u admin` |\n| `-t` | Team ID for context | \"test-team\" | `-t ops-team` |\n| `--ctx` | Path to context JSON file | - | `--ctx context.json` |\n| `-v` | Verbose output | false | `-v` |\n| `--run` | Regex to filter tests | - | `--run \"TestSystem\"` |\n| `--timeout` | Timeout per test function | 30s | `--timeout 1m` |\n| `--fail-fast` | Stop on first failure | false | `--fail-fast` |\n| `-o` | Output file for report | stdout | `-o report.json` |\n| `-r` | Reporter agent for custom report | - | `-r report.html` |\n\nThe `--run` flag accepts a Go-style regex pattern to filter which tests to run:\n\n```bash\n# Run only TestSystemReady\nyao agent test -i scripts.expense.setup --run TestSystemReady\n\n# Run all tests starting with \"TestSystem\"\nyao agent test -i scripts.expense.setup --run \"TestSystem.*\"\n\n# Run tests containing \"Error\"\nyao agent test -i scripts.expense.setup --run \".*Error.*\"\n```\n\n### Custom Context Configuration\n\nThe `--ctx` flag allows you to provide a JSON file with custom context configuration, giving full control over authorization data, metadata, and client information:\n\n```bash\n# Use custom context file\nyao agent test -i scripts.expense.setup --ctx tests/context.json -v\n```\n\n**Context JSON Format:**\n\n```json\n{\n \"authorized\": {\n \"sub\": \"user-12345\",\n \"client_id\": \"my-app\",\n \"scope\": \"read write\",\n \"session_id\": \"sess-abc123\",\n \"user_id\": \"admin\",\n \"team_id\": \"team-001\",\n \"tenant_id\": \"acme-corp\",\n \"remember_me\": false,\n \"constraints\": {\n \"owner_only\": false,\n \"creator_only\": false,\n \"editor_only\": false,\n \"team_only\": true,\n \"extra\": {\n \"department\": \"engineering\",\n \"region\": \"us-west\"\n }\n }\n },\n \"metadata\": {\n \"request_id\": \"req-123\",\n \"trace_id\": \"trace-456\",\n \"custom_field\": \"custom_value\"\n },\n \"client\": {\n \"type\": \"web\",\n \"user_agent\": \"Mozilla/5.0\",\n \"ip\": \"192.168.1.100\"\n },\n \"locale\": \"zh-cn\",\n \"referer\": \"https://example.com/dashboard\"\n}\n```\n\n**Field Descriptions:**\n\n| Field | Description |\n| -------------------------- | --------------------------------------------------- |\n| `authorized.sub` | Subject identifier (JWT sub claim) |\n| `authorized.client_id` | OAuth client ID |\n| `authorized.scope` | Access scope |\n| `authorized.session_id` | Session identifier |\n| `authorized.user_id` | User identifier (overrides -u flag) |\n| `authorized.team_id` | Team identifier (overrides -t flag) |\n| `authorized.tenant_id` | Tenant identifier |\n| `authorized.remember_me` | Remember me flag |\n| `authorized.constraints` | Data access constraints (set by ACL enforcement) |\n| `constraints.owner_only` | Only access owner's data |\n| `constraints.creator_only` | Only access creator's data |\n| `constraints.editor_only` | Only access editor's data |\n| `constraints.team_only` | Only access team's data (filter by team_id) |\n| `constraints.extra` | User-defined constraints (department, region, etc.) |\n| `metadata` | Custom metadata passed to context |\n| `client.type` | Client type (web, mobile, test, etc.) |\n| `client.user_agent` | Client user agent string |\n| `client.ip` | Client IP address |\n| `locale` | Locale setting (e.g., \"en-us\", \"zh-cn\") |\n| `referer` | Request referer URL |\n\n**Priority:** When both `-u`/`-t` flags and `--ctx` file are provided, the context file values take precedence.\n\n### Script Test Report Format\n\nWhen using `-o` to save results:\n\n```json\n{\n \"type\": \"script_test\",\n \"script\": \"scripts.expense.setup\",\n \"script_path\": \"expense/src/setup_test.ts\",\n \"summary\": {\n \"total\": 3,\n \"passed\": 2,\n \"failed\": 1,\n \"skipped\": 0,\n \"duration_ms\": 35\n },\n \"environment\": {\n \"user_id\": \"test-user\",\n \"team_id\": \"test-team\",\n \"locale\": \"en-us\"\n },\n \"results\": [\n {\n \"name\": \"TestSystemReady\",\n \"status\": \"passed\",\n \"duration_ms\": 12,\n \"logs\": []\n },\n {\n \"name\": \"TestSystemReadyWithInvalidContext\",\n \"status\": \"passed\",\n \"duration_ms\": 8,\n \"logs\": []\n },\n {\n \"name\": \"TestSetupWithMockData\",\n \"status\": \"failed\",\n \"duration_ms\": 15,\n \"error\": \"assertion failed: Result ID should match\",\n \"assertion\": {\n \"type\": \"Equal\",\n \"expected\": 1,\n \"actual\": 2,\n \"message\": \"Result ID should match\"\n },\n \"logs\": []\n }\n ],\n \"metadata\": {\n \"started_at\": \"2024-12-17T10:00:00Z\",\n \"completed_at\": \"2024-12-17T10:00:00Z\",\n \"version\": \"0.10.5\"\n }\n}\n```\n\n### Best Practices\n\n1. **Naming Convention**: Use descriptive test names that explain what's being tested\n\n - Good: `TestSystemReadyWithValidUser`, `TestSetupReturnsErrorOnMissingConfig`\n - Bad: `Test1`, `TestIt`\n\n2. **One Assertion Per Concept**: Each test should verify one behavior\n\n ```typescript\n // Good: Focused tests\n function TestSetupCreatesDatabase(t, ctx) { ... }\n function TestSetupInitializesCache(t, ctx) { ... }\n\n // Bad: Testing too many things\n function TestSetup(t, ctx) {\n // tests database, cache, config, etc.\n }\n ```\n\n3. **Use Helper Functions**: Extract common setup logic\n\n ```typescript\n function setupTestContext(ctx) {\n ctx.Metadata.testMode = true;\n return ctx;\n }\n\n function TestFeatureA(t, ctx) {\n ctx = setupTestContext(ctx);\n // ...\n }\n ```\n\n4. **Test Error Cases**: Don't just test happy paths\n\n ```typescript\n function TestSetupWithMissingConfig(t, ctx) {\n const { assert } = t;\n ctx.Metadata.config = null;\n\n const result = Setup(ctx);\n assert.Error(result.error, \"Should return error for missing config\");\n }\n ```\n\n5. **Clean Up**: If your test modifies global state, clean up after\n ```typescript\n function TestWithGlobalState(t, ctx) {\n const originalValue = GlobalConfig.value;\n try {\n GlobalConfig.value = \"test\";\n // ... test logic\n } finally {\n GlobalConfig.value = originalValue;\n }\n }\n ```\n\n## Input Format (JSONL)\n\nEach line in the input file is a JSON object with the following structure:\n\n```jsonl\n{\"id\": \"T001\", \"input\": \"Simple text input\"}\n{\"id\": \"T002\", \"input\": {\"role\": \"user\", \"content\": \"Message with role\"}}\n{\"id\": \"T003\", \"input\": {\"role\": \"user\", \"content\": [{\"type\": \"text\", \"text\": \"ContentPart array\"}]}}\n{\"id\": \"T004\", \"input\": [{\"role\": \"user\", \"content\": \"First message\"}, {\"role\": \"assistant\", \"content\": \"Response\"}, {\"role\": \"user\", \"content\": \"Follow-up\"}]}\n{\"id\": \"T005\", \"input\": \"Text input\", \"expected\": {\"keywords\": [\"keyword1\", \"keyword2\"]}}\n{\"id\": \"T006\", \"input\": \"Test with specific user\", \"user\": \"admin\", \"team\": \"ops-team\"}\n```\n\n### Input Types\n\n| Type | Description | Example |\n| ----------- | ------------------------ | ----------------------------------------------------- |\n| `string` | Simple text input | `\"Hello world\"` |\n| `Message` | Single message with role | `{\"role\": \"user\", \"content\": \"...\"}` |\n| `[]Message` | Conversation history | `[{\"role\": \"user\", ...}, {\"role\": \"assistant\", ...}]` |\n\n### Fields\n\n| Field | Type | Required | Description |\n| ---------- | ------------------------------ | -------- | ---------------------------------------------------- |\n| `id` | string | Yes | Unique test case identifier (e.g., \"T001\") |\n| `input` | string \\| Message \\| []Message | Yes | Test input |\n| `expected` | any | No | Expected output for exact match validation |\n| `assert` | Assertion \\| []Assertion | No | Custom assertion rules (see Assertions section) |\n| `user` | string | No | User ID for this test case (overridden by `-u` flag) |\n| `team` | string | No | Team ID for this test case (overridden by `-t` flag) |\n| `metadata` | map | No | Additional metadata for the test case |\n| `skip` | bool | No | Skip this test case |\n| `timeout` | string | No | Override timeout (e.g., \"30s\", \"1m\") |\n\n### Assertions\n\nThe `assert` field allows flexible validation of agent output. If `assert` is defined, it takes precedence over `expected`.\n\n#### Assertion Types\n\n| Type | Description | Example |\n| -------------- | ----------------------------------------------- | ---------------------------------------------------------------- |\n| `equals` | Exact match (default if only `expected` is set) | `{\"type\": \"equals\", \"value\": {\"need_search\": false}}` |\n| `contains` | Output contains the expected string/value | `{\"type\": \"contains\", \"value\": \"keyword\"}` |\n| `not_contains` | Output does not contain the string/value | `{\"type\": \"not_contains\", \"value\": \"error\"}` |\n| `json_path` | Extract value using JSON path and compare | `{\"type\": \"json_path\", \"path\": \"$.need_search\", \"value\": false}` |\n| `regex` | Match output against regex pattern | `{\"type\": \"regex\", \"value\": \"\\\\d{3}-\\\\d{4}\"}` |\n| `type` | Check output type (string, object, array, etc.) | `{\"type\": \"type\", \"value\": \"object\"}` |\n| `script` | Run a custom assertion script | `{\"type\": \"script\", \"script\": \"scripts.test.Assert\"}` |\n\n#### Assertion Structure\n\n```typescript\ninterface Assertion {\n type: string; // Assertion type (required)\n value?: any; // Expected value or pattern\n path?: string; // JSON path for json_path assertions\n script?: string; // Script name for script assertions\n message?: string; // Custom failure message\n negate?: boolean; // Invert the assertion result\n}\n```\n\n#### Examples\n\n**Simple contains check:**\n\n```jsonl\n{\n \"id\": \"T001\",\n \"input\": \"Hello\",\n \"assert\": {\n \"type\": \"contains\",\n \"value\": \"need_search\"\n }\n}\n```\n\n**JSON path validation (for agents returning JSON):**\n\n```jsonl\n{\n \"id\": \"T002\",\n \"input\": \"What's the weather?\",\n \"assert\": {\n \"type\": \"json_path\",\n \"path\": \"$.need_search\",\n \"value\": true\n }\n}\n```\n\n**Multiple assertions (all must pass):**\n\n```jsonl\n{\n \"id\": \"T003\",\n \"input\": \"Calculate 2+2\",\n \"assert\": [\n {\n \"type\": \"json_path\",\n \"path\": \"$.need_search\",\n \"value\": false\n },\n {\n \"type\": \"json_path\",\n \"path\": \"$.confidence\",\n \"value\": 0.99\n },\n {\n \"type\": \"not_contains\",\n \"value\": \"error\"\n }\n ]\n}\n```\n\n**Custom script assertion:**\n\n```jsonl\n{\n \"id\": \"T004\",\n \"input\": \"Complex test\",\n \"assert\": {\n \"type\": \"script\",\n \"script\": \"scripts.test.ValidateOutput\"\n }\n}\n```\n\nThe script receives `(output, input, expected)` and should return:\n\n```typescript\n// Simple boolean\nreturn true; // or false\n\n// Or detailed result\nreturn {\n pass: true,\n message: \"Validation passed: output contains expected keywords\",\n};\n```\n\n**Negated assertion:**\n\n```jsonl\n{\n \"id\": \"T005\",\n \"input\": \"Hello\",\n \"assert\": {\n \"type\": \"contains\",\n \"value\": \"error\",\n \"negate\": true\n }\n}\n```\n\n#### JSON Path Notes\n\n- Supports simple dot-notation paths: `$.field.subfield` or `field.subfield`\n- Automatically extracts JSON from markdown code blocks (e.g., ` ```json ... ``` `)\n- Works with both string output and structured objects\n\n### Environment Override Priority\n\nThe test environment (user/team) is determined by the following priority (highest first):\n\n1. **Command line flags** (`-u`, `-t`): Global override for all test cases\n2. **Test case fields** (`user`, `team`): Per-test case configuration\n3. **Default values**: \"test-user\", \"test-team\"\n\nExample:\n\n```bash\n# All tests run as \"admin\" user in \"prod-team\", regardless of test case settings\nyao agent test -i tests/inputs.jsonl -u admin -t prod-team -o report.json\n```\n\n```jsonl\n# T001 uses default user/team\n{\"id\": \"T001\", \"input\": \"Hello\"}\n\n# T002 uses specific user/team (unless overridden by -u/-t flags)\n{\"id\": \"T002\", \"input\": \"Admin action\", \"user\": \"admin\", \"team\": \"admin-team\"}\n\n# T003 uses specific user only, team uses default\n{\"id\": \"T003\", \"input\": \"User specific test\", \"user\": \"special-user\"}\n```\n\n## Output Format\n\n### Default: JSONL (without `-r` flag)\n\nBy default (without `-r` flag), the output is JSONL format - one JSON object per line, suitable for streaming and CI integration:\n\n```jsonl\n{\"type\": \"start\", \"timestamp\": \"2024-12-17T10:00:00Z\", \"agent_id\": \"keyword\", \"total_cases\": 42}\n{\"type\": \"result\", \"id\": \"T001\", \"status\": \"passed\", \"duration_ms\": 234, \"output\": {\"keywords\": [\"AI\", \"ML\"]}}\n{\"type\": \"result\", \"id\": \"T002\", \"status\": \"passed\", \"duration_ms\": 189, \"output\": {\"keywords\": [\"cloud\"]}}\n{\"type\": \"result\", \"id\": \"T003\", \"status\": \"failed\", \"duration_ms\": 0, \"error\": \"timeout after 30s\"}\n{\"type\": \"summary\", \"total\": 42, \"passed\": 40, \"failed\": 2, \"duration_ms\": 12345}\n```\n\nThis format is:\n\n- **Streamable**: Results are output as they complete\n- **Parseable**: Each line is valid JSON, easy to process with `jq` or scripts\n- **CI-friendly**: Exit code indicates pass/fail status\n\n### Custom Report (with `-r` flag)\n\n```json\n{\n \"summary\": {\n \"total\": 42,\n \"passed\": 40,\n \"failed\": 2,\n \"skipped\": 0,\n \"duration_ms\": 12345,\n \"agent_id\": \"keyword\",\n \"connector\": \"deepseek.v3\",\n \"runs_per_case\": 1,\n \"overall_pass_rate\": 95.2\n },\n \"environment\": {\n \"user_id\": \"test-user\",\n \"team_id\": \"test-team\",\n \"locale\": \"en-us\"\n },\n \"results\": [\n {\n \"id\": \"T001\",\n \"status\": \"passed\",\n \"input\": \"...\",\n \"output\": { \"keywords\": [\"AI\", \"machine learning\"] },\n \"expected\": null,\n \"duration_ms\": 234,\n \"error\": null\n }\n ],\n \"metadata\": {\n \"started_at\": \"2024-12-17T10:00:00Z\",\n \"completed_at\": \"2024-12-17T10:00:12Z\",\n \"version\": \"0.10.5\"\n }\n}\n```\n\n### HTML Report\n\nBeautiful, interactive HTML report with:\n\n- Summary statistics (pass/fail/skip counts, duration)\n- Stability charts (when runs > 1)\n- Filterable test results table\n- Expandable input/output details\n- Error highlighting\n- Export options\n\n### Markdown Report\n\n```markdown\n# Agent Test Report\n\n## Summary\n\n| Metric | Value |\n| --------- | ----------- |\n| Agent | keyword |\n| Connector | deepseek.v3 |\n| Total | 42 |\n| Passed | 40 |\n| Failed | 2 |\n| Pass Rate | 95.2% |\n| Duration | 12.3s |\n\n## Environment\n\n| Setting | Value |\n| ------- | --------- |\n| User | test-user |\n| Team | test-team |\n| Locale | en-us |\n\n## Results\n\n### ✅ T001 - Passed (234ms)\n\n...\n```\n\n## Architecture\n\n```\nagent/test/\n├── DESIGN.md # This file\n├── types.go # Core types and interfaces\n├── interfaces.go # Runner and Reporter interfaces\n├── runner.go # Test runner implementation\n├── loader.go # Test case loader\n├── resolver.go # Agent resolver\n├── context.go # Test context creation\n├── assert.go # Assertion implementation\n├── input.go # Input parsing\n├── output.go # Output formatting\n├── script.go # Script test runner (NEW)\n├── script_types.go # Script test types (NEW)\n├── script_assert.go # Script assertion bindings (NEW)\n└── reporter/\n ├── json.go # JSON reporter\n ├── html.go # HTML reporter\n ├── markdown.go # Markdown reporter\n └── agent.go # Agent-based custom reporter\n```\n\n## Core Components\n\n### 1. TestCase\n\nRepresents a single test case loaded from JSONL.\n\n### 2. TestResult\n\nRepresents the result of running a single test case.\n\n### 3. TestReport\n\nRepresents the complete test report with summary and results.\n\n### 4. Runner\n\nExecutes test cases against an agent:\n\n- Loads test cases from JSONL\n- Resolves agent from path or explicit ID\n- Creates test context with environment\n- Executes each test case (optionally multiple runs)\n- Collects results and stability metrics\n\n### 5. ScriptRunner (NEW)\n\nExecutes script tests for agent handler scripts:\n\n- Resolves script path from `scripts.` prefix\n- Discovers `Test*` functions in the script\n- Creates test context with environment\n- Executes each test function with testing object and context\n- Collects results and generates report\n\n### 6. ScriptTestCase (NEW)\n\nRepresents a single script test function:\n\n```go\ntype ScriptTestCase struct {\n Name string // Function name (e.g., \"TestSystemReady\")\n Function string // Full function reference\n}\n```\n\n### 7. ScriptTestResult (NEW)\n\nRepresents the result of running a script test function:\n\n```go\ntype ScriptTestResult struct {\n Name string `json:\"name\"`\n Status Status `json:\"status\"`\n DurationMs int64 `json:\"duration_ms\"`\n Error string `json:\"error,omitempty\"`\n Assertion *AssertionInfo `json:\"assertion,omitempty\"`\n Logs []string `json:\"logs,omitempty\"`\n}\n\ntype AssertionInfo struct {\n Type string `json:\"type\"`\n Expected interface{} `json:\"expected,omitempty\"`\n Actual interface{} `json:\"actual,omitempty\"`\n Message string `json:\"message,omitempty\"`\n}\n```\n\n### 8. ScriptTestReport (NEW)\n\nRepresents the complete script test report:\n\n```go\ntype ScriptTestReport struct {\n Type string `json:\"type\"` // \"script_test\"\n Script string `json:\"script\"`\n ScriptPath string `json:\"script_path\"`\n Summary *ScriptTestSummary `json:\"summary\"`\n Environment *Environment `json:\"environment\"`\n Results []*ScriptTestResult `json:\"results\"`\n Metadata *ReportMetadata `json:\"metadata\"`\n}\n\ntype ScriptTestSummary struct {\n Total int `json:\"total\"`\n Passed int `json:\"passed\"`\n Failed int `json:\"failed\"`\n Skipped int `json:\"skipped\"`\n DurationMs int64 `json:\"duration_ms\"`\n}\n```\n\n### 9. Reporter\n\nGenerates reports in various formats. The format is determined by the `-o` file extension:\n\n| Extension | Format | Description |\n| --------- | -------- | -------------------------- |\n| `.jsonl` | JSONL | Streaming, line-by-line |\n| `.json` | JSON | Full structured report |\n| `.md` | Markdown | Human-readable with tables |\n| `.html` | HTML | Interactive web report |\n\n#### Custom Reporter Agent (`-r` flag)\n\nWhen `-r ` is specified, the framework calls the specified agent to generate the report:\n\n1. Test execution completes, `TestReport` is generated\n2. Framework calls the reporter agent with input:\n ```json\n {\n \"report\": {\n /* TestReport object */\n },\n \"format\": \"html\",\n \"options\": { \"verbose\": true }\n }\n ```\n3. Agent processes the report and returns formatted content\n4. Framework writes the returned content to the output file\n\nExample usage:\n\n```bash\n# Use custom reporter agent to generate a beautiful HTML report\nyao agent test -i tests/inputs.jsonl -r report.beautiful -o report.html\n\n# Use custom reporter agent to generate Slack-formatted summary\nyao agent test -i tests/inputs.jsonl -r report.slack -o summary.txt\n```\n\nThis allows for fully customizable report generation using AI agents\n\n## Configuration\n\n### Test Options\n\n```go\ntype Options struct {\n // Input/Output\n Input string // Input source: file path, message, or scripts.xxx\n InputMode InputMode // Auto-detected: file, message, or script\n OutputFile string // Path to output report\n\n // Agent Selection\n AgentID string // Explicit agent ID (optional)\n Connector string // Override connector (optional)\n\n // Test Environment\n UserID string // Test user ID (-u flag)\n TeamID string // Test team ID (-t flag)\n Locale string // Locale (default: \"en-us\")\n\n // Execution\n Timeout time.Duration // Default timeout per test\n Parallel int // Number of parallel tests (default: 1)\n Runs int // Number of runs per test case (default: 1)\n\n // Reporting\n ReporterID string // Reporter agent ID for custom report\n\n // Behavior\n Verbose bool // Verbose output\n FailFast bool // Stop on first failure\n}\n\n// InputMode represents the input mode for test cases\ntype InputMode string\n\nconst (\n InputModeFile InputMode = \"file\" // JSONL file input\n InputModeMessage InputMode = \"message\" // Direct message input\n InputModeScript InputMode = \"script\" // Script test mode (NEW)\n)\n```\n\n### Input Mode Detection\n\nThe input mode is automatically detected based on the input value:\n\n| Input Pattern | Mode | Description |\n| ----------------- | --------- | -------------------------- |\n| `scripts.xxx.yyy` | `script` | Script test mode |\n| `*.jsonl` | `file` | JSONL file mode |\n| `path/to/file` | `file` | File path (if file exists) |\n| `\"any text\"` | `message` | Direct message mode |\n\n```go\nfunc DetectInputMode(input string) InputMode {\n // Check for script test prefix\n if strings.HasPrefix(input, \"scripts.\") {\n return InputModeScript\n }\n\n // Check if it's a file path\n if strings.HasSuffix(input, \".jsonl\") || fileExists(input) {\n return InputModeFile\n }\n\n // Default to message mode\n return InputModeMessage\n}\n```\n\n## Script Testing Implementation\n\n### Script Resolution\n\n```go\n// ResolveScript resolves the script path from scripts.xxx.yyy format\nfunc ResolveScript(input string) (*ScriptInfo, error) {\n // Remove \"scripts.\" prefix\n path := strings.TrimPrefix(input, \"scripts.\")\n\n // Split into parts: \"expense.setup\" -> [\"expense\", \"setup\"]\n parts := strings.Split(path, \".\")\n if len(parts) < 2 {\n return nil, fmt.Errorf(\"invalid script path: %s\", input)\n }\n\n // Build paths\n // assistantDir: expense\n // moduleName: setup\n // scriptPath: expense/src/setup.ts\n // testPath: expense/src/setup_test.ts\n assistantDir := parts[0]\n moduleName := parts[1]\n\n return &ScriptInfo{\n ID: input,\n Assistant: assistantDir,\n Module: moduleName,\n ScriptPath: filepath.Join(assistantDir, \"src\", moduleName+\".ts\"),\n TestPath: filepath.Join(assistantDir, \"src\", moduleName+\"_test.ts\"),\n }, nil\n}\n```\n\n### Test Function Discovery\n\nTest functions are discovered by scanning the script for functions starting with `Test`:\n\n```go\n// DiscoverTests finds all Test* functions in the script\nfunc DiscoverTests(scriptPath string) ([]*ScriptTestCase, error) {\n // Use the JavaScript runtime to list exported functions\n // Filter for functions starting with \"Test\"\n // Return list of test cases\n}\n```\n\n### Testing Object Binding\n\nThe `testing.T` object is provided to test functions via JavaScript runtime binding:\n\n```go\n// TestingT represents the testing object passed to test functions\ntype TestingT struct {\n name string\n failed bool\n skipped bool\n logs []string\n assert *AssertObject\n}\n\n// AssertObject provides assertion methods\ntype AssertObject struct {\n t *TestingT\n}\n\nfunc (a *AssertObject) True(value bool, message ...string) {\n if !value {\n a.t.fail(formatMessage(\"expected true, got false\", message))\n }\n}\n\nfunc (a *AssertObject) Equal(actual, expected interface{}, message ...string) {\n if !reflect.DeepEqual(actual, expected) {\n a.t.fail(formatMessage(\n fmt.Sprintf(\"expected %v, got %v\", expected, actual),\n message,\n ))\n }\n}\n\n// ... other assertion methods\n```\n\n### Script Execution Flow\n\n```\n1. Parse input: \"scripts.expense.setup\"\n2. Resolve script info:\n - TestPath: expense/src/setup_test.ts\n - ScriptPath: expense/src/setup.ts\n3. Discover test functions: [TestSystemReady, TestSetupWithMockData, ...]\n4. For each test function:\n a. Create testing.T object\n b. Create agent.Context with environment\n c. Execute: TestFunction(t, ctx)\n d. Collect result (passed/failed/skipped)\n5. Generate report\n```\n\n### Integration with Existing Runner\n\n```go\nfunc (r *Executor) Run() (*Report, error) {\n switch r.opts.InputMode {\n case InputModeScript:\n return r.RunScriptTests()\n case InputModeMessage:\n return r.RunDirect()\n default:\n return r.RunTests()\n }\n}\n\nfunc (r *Executor) RunScriptTests() (*Report, error) {\n // 1. Resolve script\n scriptInfo, err := ResolveScript(r.opts.Input)\n if err != nil {\n return nil, err\n }\n\n // 2. Discover tests\n tests, err := DiscoverTests(scriptInfo.TestPath)\n if err != nil {\n return nil, err\n }\n\n // 3. Run each test\n results := make([]*ScriptTestResult, 0, len(tests))\n for _, tc := range tests {\n result := r.runScriptTest(tc, scriptInfo)\n results = append(results, result)\n\n if r.opts.FailFast && result.Status == StatusFailed {\n break\n }\n }\n\n // 4. Generate report\n return r.buildScriptReport(scriptInfo, results), nil\n}\n```\n\n## Exit Codes\n\n| Code | Description |\n| ---- | ------------------- |\n| 0 | All tests passed |\n| 1 | Some tests failed |\n| 2 | Configuration error |\n| 3 | Runtime error |\n\n## CI Integration\n\n### GitHub Actions Example\n\n```yaml\n- name: Run Agent Tests\n run: |\n yao agent test -i assistants/keyword/tests/inputs.jsonl \\\n -u ci-user -t ci-team \\\n --runs 3 \\\n -o report.json\n\n- name: Check Stability\n run: |\n # Fail if any test has pass rate below 80%\n jq -e '.results | all(.pass_rate >= 80)' report.json\n\n- name: Upload Test Report\n uses: actions/upload-artifact@v3\n with:\n name: agent-test-report\n path: report.json\n```\n\n### Exit Code Handling\n\nThe command exits with code 1 if any tests fail, making it easy to integrate with CI pipelines.\n\n## Future Enhancements\n\n1. **Snapshot Testing**: Compare outputs against saved snapshots\n2. **Fuzzing**: Generate random inputs for robustness testing\n3. **Coverage**: Track which agent code paths are exercised\n4. **Benchmarking**: Performance metrics and regression detection\n5. **Diff Reports**: Compare results between runs\n6. **Flaky Test Detection**: Automatic identification of unstable tests\n7. **Test Prioritization**: Run most important/failing tests first\n8. **Script Test Enhancements**:\n - Parallel script test execution\n - Setup/Teardown hooks (`TestMain`, `BeforeEach`, `AfterEach`)\n - Mocking utilities for external dependencies\n - Code coverage for TypeScript/JavaScript scripts\n" }, { "path": "agent/test/DESIGN_V2.md", "content": "# Agent Test Framework V2 Design\n\n## Overview\n\nThis document describes the design for Agent Test Framework V2, which extends the existing testing capabilities with:\n\n- **Message history support** - Test agents with conversation context via `input` array (already implemented)\n- **Agent-driven testing** - Use agents to generate test cases and validate responses\n- **Dynamic testing** - Simulator-driven testing with checkpoint validation\n\n## Quick Reference: Format Rules\n\n| Context | Format | Example |\n| --------------------- | ------------------------ | ------------------------------------------------------- |\n| `-i` flag (CLI) | Prefix required | `agents:workers.test.gen`, `scripts:tests.gen` |\n| JSONL assertion `use` | Prefix required | `\"use\": \"agents:workers.test.validator\"` |\n| JSONL `simulator.use` | No prefix (agent only) | `\"use\": \"workers.test.user-simulator\"` |\n| `--simulator` flag | No prefix (agent only) | `--simulator workers.test.user-simulator` |\n| `t.assert.Agent()` | No prefix (method-bound) | `t.assert.Agent(resp, \"workers.test.validator\", {...})` |\n| JSONL `before/after` | No prefix (in src/) | `\"before\": \"env_test.Before\"` |\n| `--before/--after` | No prefix (in src/) | `--before env_test.BeforeAll` |\n\n## Design Goals\n\n1. **Simple** - Single-turn with optional message history, no complex multi-turn state\n2. **Stateless** - Each test is independent, no session management needed\n3. **Parallel** - Tests can run in parallel since they don't share state\n4. **Flexible** - Support both static (messages) and dynamic (simulator) testing\n5. **Agent-driven** - Input generation, simulation, and validation can all be agent-powered\n\n## Architecture Overview\n\n```\n┌─────────────────────────────────────────────────────────────────────────┐\n│ yao agent test │\n├─────────────────────────────────────────────────────────────────────────┤\n│ │\n│ INPUT SOURCES (-i flag) │\n│ ┌─────────────┐ ┌─────────────┐ ┌─────────────┐ │\n│ │ JSONL File │ │ Message │ │ Generator │ │\n│ │ ./test.jsonl│ │ \"Hello...\" │ │ agents:xxx │ │\n│ └──────┬──────┘ └──────┬──────┘ └──────┬──────┘ │\n│ │ │ │ │\n│ └────────────────┴────────────────┘ │\n│ │ │\n│ ▼ │\n│ ┌───────────────────────────────────────────────────────────────────┐ │\n│ │ Test Case Parser │ │\n│ │ │ │\n│ │ Standard Mode: {input: \"...\" | [...], assertions} │ │\n│ │ Dynamic Mode: {simulator: {...}, checkpoints: [...]} │ │\n│ └───────────────────────────────────────────────────────────────────┘ │\n│ │ │\n│ ┌───────────────┴───────────────┐ │\n│ ▼ ▼ │\n│ ┌───────────────────┐ ┌───────────────────────┐ │\n│ │ STANDARD MODE │ │ DYNAMIC MODE │ │\n│ │ │ │ │ │\n│ │ 1. Build messages │ │ LOOP: │ │\n│ │ 2. Call Agent │ │ 1. Simulator→input │ │\n│ │ 3. Run assertions │ │ 2. Call Agent │ │\n│ │ │ │ 3. Check checkpoints │ │\n│ │ → PASS/FAIL │ │ 4. Until done │ │\n│ └───────────────────┘ └───────────────────────┘ │\n│ │ │\n│ ▼ │\n│ ┌───────────────────────────────────────────────────────────────────┐ │\n│ │ Reporter │ │\n│ │ - Console output │ │\n│ │ - JSON file output │ │\n│ └───────────────────────────────────────────────────────────────────┘ │\n│ │\n└─────────────────────────────────────────────────────────────────────────┘\n```\n\n## Test Modes\n\n### Standard Mode (Default)\n\nSingle call to agent with optional message history. **No multi-turn state management needed.**\n\n| Field | Type | Description |\n| ------------ | ------------------------------ | --------------------------------------------- |\n| `input` | string \\| Message \\| Message[] | Text, single message, or conversation history |\n| `assertions` | array | Assertions to validate response |\n| `options` | object | `context.Options` passed to agent |\n\n### Dynamic Mode\n\nSimulator-driven testing with checkpoint validation.\n\n| Field | Type | Description |\n| ------------- | ------ | -------------------------------- |\n| `simulator` | object | Simulator agent configuration |\n| `checkpoints` | array | Functional checkpoints to verify |\n| `max_turns` | int | Maximum turns before timeout |\n| `timeout` | string | Maximum time (e.g., \"5m\") |\n\n## Test Case Format\n\n### Simple Input (Existing)\n\n```jsonl\n{\n \"id\": \"T001\",\n \"input\": \"Hello\",\n \"assertions\": [\n {\n \"type\": \"contains\",\n \"value\": \"Hi\"\n }\n ]\n}\n```\n\n### With Message History (Existing)\n\nThe `input` field already supports message arrays for conversation context:\n\n```jsonl\n{\n \"id\": \"T002\",\n \"name\": \"Expense submission - final confirmation\",\n \"input\": [\n {\n \"role\": \"user\",\n \"content\": \"I want to submit an expense\"\n },\n {\n \"role\": \"assistant\",\n \"content\": \"What type of expense would you like to submit?\"\n },\n {\n \"role\": \"user\",\n \"content\": \"Business travel to Beijing, $3500\"\n },\n {\n \"role\": \"assistant\",\n \"content\": \"I'll create an expense for business travel, $3500. Please confirm.\"\n },\n {\n \"role\": \"user\",\n \"content\": \"Yes, confirm\"\n }\n ],\n \"assertions\": [\n {\n \"type\": \"contains\",\n \"value\": \"submitted\"\n },\n {\n \"type\": \"tool_called\",\n \"name\": \"create_expense\"\n }\n ]\n}\n```\n\n**Key insight**: Instead of executing 3 turns sequentially, we pass the full conversation history. The agent sees the context and responds to the last message. This is:\n\n- **Simpler** - No turn-by-turn execution, no session state\n- **Faster** - Single API call instead of multiple\n- **Parallelizable** - Each test is independent\n- **Debuggable** - Clear input/output for each test\n\n### Testing Different Points in a Conversation\n\nTo test agent behavior at different conversation stages, create separate test cases:\n\n```jsonl\n// Test 1: First turn - agent should ask for expense type\n{\n \"id\": \"expense-turn1\",\n \"input\": [{\"role\": \"user\", \"content\": \"I want to submit an expense\"}],\n \"assertions\": [{\"type\": \"contains\", \"value\": \"type\"}]\n}\n\n// Test 2: Second turn - agent should create expense\n{\n \"id\": \"expense-turn2\",\n \"input\": [\n {\"role\": \"user\", \"content\": \"I want to submit an expense\"},\n {\"role\": \"assistant\", \"content\": \"What type of expense would you like to submit?\"},\n {\"role\": \"user\", \"content\": \"Business travel, $3500\"}\n ],\n \"assertions\": [{\"type\": \"tool_called\", \"name\": \"create_expense\"}]\n}\n\n// Test 3: Final turn - agent should confirm submission\n{\n \"id\": \"expense-turn3\",\n \"input\": [\n {\"role\": \"user\", \"content\": \"I want to submit an expense\"},\n {\"role\": \"assistant\", \"content\": \"What type of expense?\"},\n {\"role\": \"user\", \"content\": \"Business travel, $3500\"},\n {\"role\": \"assistant\", \"content\": \"Confirm $3500 expense?\"},\n {\"role\": \"user\", \"content\": \"Yes\"}\n ],\n \"assertions\": [{\"type\": \"contains\", \"value\": \"submitted\"}]\n}\n```\n\n### With Attachments\n\n```jsonl\n{\n \"id\": \"T003\",\n \"input\": [\n {\n \"role\": \"user\",\n \"content\": [\n {\n \"type\": \"text\",\n \"text\": \"What's in this receipt?\"\n },\n {\n \"type\": \"image\",\n \"source\": \"file://./fixtures/receipt.jpg\"\n }\n ]\n }\n ],\n \"assertions\": [\n {\n \"type\": \"contains\",\n \"value\": \"amount\"\n }\n ]\n}\n```\n\n### Dynamic Mode (Simulator + Checkpoints)\n\nFor coverage testing where conversation flow is unpredictable:\n\n```jsonl\n{\n \"id\": \"T004\",\n \"name\": \"Expense Submission Coverage\",\n \"simulator\": {\n \"use\": \"workers.test.user-simulator\",\n \"options\": {\n \"metadata\": {\n \"persona\": \"New employee unfamiliar with expense process\",\n \"goal\": \"Submit a $3500 travel expense\"\n }\n }\n },\n \"checkpoints\": [\n {\n \"id\": \"ask_type\",\n \"description\": \"Agent asks for expense type\",\n \"assertion\": {\n \"type\": \"contains\",\n \"value\": \"type\"\n }\n },\n {\n \"id\": \"call_create\",\n \"description\": \"Agent calls create_expense\",\n \"after\": [\n \"ask_type\"\n ],\n \"assertion\": {\n \"type\": \"tool_called\",\n \"name\": \"create_expense\"\n }\n },\n {\n \"id\": \"confirm\",\n \"description\": \"Agent confirms submission\",\n \"after\": [\n \"call_create\"\n ],\n \"assertion\": {\n \"type\": \"contains\",\n \"value\": \"submitted\"\n }\n }\n ],\n \"max_turns\": 10,\n \"timeout\": \"2m\"\n}\n```\n\n## Field Descriptions\n\n### Standard Mode Fields\n\n| Field | Type | Required | Description |\n| ------------ | ------------------------------ | -------- | ------------------------------------------------- |\n| `id` | string | Yes | Unique test identifier |\n| `name` | string | No | Human-readable test name |\n| `input` | string \\| Message \\| Message[] | Yes | Input: text, single message, or message array |\n| `assertions` | array | No | Assertions to validate response (alias: `assert`) |\n| `options` | object | No | `context.Options` passed to agent |\n| `before` | string | No | Before script (e.g., `env_test.Before`) |\n| `after` | string | No | After script (e.g., `env_test.After`) |\n\n**Note**: The `input` field supports three formats:\n\n- `string`: Simple text (converted to `[{role: \"user\", content: \"...\"}]`)\n- `object`: Single message `{role: \"...\", content: \"...\"}`\n- `array`: Message history `[{role: \"user\", ...}, {role: \"assistant\", ...}, ...]`\n\n### Dynamic Mode Fields\n\n| Field | Type | Required | Description |\n| --------------------------- | ------ | -------- | ------------------------------------------ |\n| `id` | string | Yes | Unique test identifier |\n| `name` | string | No | Human-readable test name |\n| `simulator` | object | Yes | User simulator configuration |\n| `simulator.use` | string | Yes | Simulator agent ID (no prefix) |\n| `simulator.options` | object | No | `context.Options` passed to simulator |\n| `checkpoints` | array | Yes | Functionality checkpoints to verify |\n| `checkpoints[].id` | string | Yes | Unique checkpoint identifier |\n| `checkpoints[].description` | string | No | Human-readable description |\n| `checkpoints[].assertion` | object | Yes | Assertion to verify |\n| `checkpoints[].after` | array | No | Checkpoint IDs that must occur first |\n| `max_turns` | int | No | Maximum turns before timeout (default: 20) |\n| `timeout` | string | No | Maximum time (default: \"5m\") |\n| `options` | object | No | `context.Options` passed to target agent |\n| `before` | string | No | Before script function |\n| `after` | string | No | After script function |\n\n## Before and After Scripts\n\nJSONL test cases can reference `*_test.ts` scripts for environment preparation:\n\n### Script Location\n\nScripts are located in the agent's `src/` directory (as `*_test.ts` files):\n\n```\nassistants/expense/\n├── package.yao\n├── prompts.yml\n├── src/\n│ ├── index.ts # Main agent script\n│ └── env_test.ts # Before/after functions\n└── tests/\n ├── inputs.jsonl # Test cases\n └── fixtures/\n └── receipt.jpg\n```\n\n### Script Interface\n\n```typescript\n// src/env_test.ts\n\n// Before function - called before test case runs\n// Returns context data that will be passed to After\nexport function Before(ctx: Context, testCase: TestCase): BeforeResult {\n // Prepare database\n const userId = Process(\"models.user.Create\", {\n name: \"Test User\",\n email: \"test@example.com\",\n });\n\n // Prepare knowledge base\n Process(\"knowledge.expense.Index\", {\n documents: [{ title: \"Policy\", content: \"Max expense $5000\" }],\n });\n\n return {\n data: { userId, testId: testCase.id },\n };\n}\n\n// After function - called after test case completes (pass or fail)\nexport function After(\n ctx: Context,\n testCase: TestCase,\n result: TestResult,\n beforeData: any\n) {\n // Clean up database\n if (beforeData?.userId) {\n Process(\"models.user.Delete\", beforeData.userId);\n }\n\n // Clean up knowledge base\n Process(\"knowledge.expense.Clear\");\n}\n\n// Global before - called once before all test cases\nexport function BeforeAll(ctx: Context, testCases: TestCase[]): BeforeResult {\n // One-time initialization\n Process(\"models.migrate\");\n return { data: { initialized: true } };\n}\n\n// Global after - called once after all test cases\nexport function AfterAll(ctx: Context, results: TestResult[], beforeData: any) {\n // Final cleanup\n Process(\"models.cleanup\");\n}\n```\n\n### Test Case with Before/After\n\n```jsonl\n{\n \"id\": \"T001\",\n \"name\": \"Submit expense with user context\",\n \"before\": \"env_test.Before\",\n \"after\": \"env_test.After\",\n \"input\": \"Submit a $500 travel expense\",\n \"assertions\": [\n {\n \"type\": \"tool_called\",\n \"name\": \"create_expense\"\n }\n ]\n}\n```\n\n### Global Before/After via CLI\n\n```bash\n# Run with global before/after\nyao agent test -i ./tests/inputs.jsonl \\\n --before env_test.BeforeAll \\\n --after env_test.AfterAll\n```\n\n### Execution Order\n\n```\n┌─────────────────────────────────────────────────────────────────┐\n│ Test Execution with Before/After │\n├─────────────────────────────────────────────────────────────────┤\n│ │\n│ 1. BeforeAll() - Global initialization (once) │\n│ ↓ │\n│ FOR EACH test case: │\n│ 2. Before() - Per-test initialization │\n│ ↓ │\n│ 3. Run test (call agent, check assertions) │\n│ ↓ │\n│ 4. After() - Per-test cleanup (always runs) │\n│ ↓ │\n│ 5. AfterAll() - Global cleanup (once) │\n│ │\n└─────────────────────────────────────────────────────────────────┘\n```\n\n**Note**: Script tests (`*_test.ts`) don't need before/after fields since they can call functions directly within the test.\n\n## Execution Flow\n\n### Standard Mode\n\n```\n┌─────────────────────────────────────────────────────────────────┐\n│ Standard Mode Execution │\n├─────────────────────────────────────────────────────────────────┤\n│ │\n│ 1. Parse test case │\n│ ├─ `input` is array? → Use as messages │\n│ └─ `input` is string? → Convert to [{role: \"user\", content}] │\n│ ↓ │\n│ 2. Call Agent.Stream(ctx, messages, options) │\n│ ↓ │\n│ 3. Run assertions against response │\n│ ├─ All PASS → Test PASSED ✅ │\n│ └─ Any FAIL → Test FAILED ❌ │\n│ │\n└─────────────────────────────────────────────────────────────────┘\n```\n\n### Dynamic Mode\n\n```\n┌─────────────────────────────────────────────────────────────────┐\n│ Dynamic Mode Execution │\n├─────────────────────────────────────────────────────────────────┤\n│ │\n│ Initialize: │\n│ - pending_checkpoints = all checkpoints │\n│ - messages = [] │\n│ - turn_count = 0 │\n│ ↓ │\n│ LOOP: │\n│ 1. Call Simulator → get user input │\n│ 2. Append user message to messages │\n│ 3. Call Agent.Stream(ctx, messages, options) │\n│ 4. Append assistant response to messages │\n│ 5. Check response against pending_checkpoints │\n│ └─ If matched (and `after` satisfied) → move to reached │\n│ 6. Check termination: │\n│ ├─ All checkpoints reached → PASSED ✅ │\n│ ├─ Simulator signals goal_achieved → FAILED ❌ │\n│ ├─ turn_count >= max_turns → FAILED ❌ │\n│ └─ timeout exceeded → FAILED ❌ │\n│ │\n└─────────────────────────────────────────────────────────────────┘\n```\n\n## Assertion Types\n\n### Static Assertions\n\n| Type | Description | Example |\n| ------------- | ---------------------- | ---------------------------------------------------------- |\n| `contains` | Response contains text | `{\"type\": \"contains\", \"value\": \"success\"}` |\n| `equals` | Exact match | `{\"type\": \"equals\", \"value\": \"OK\"}` |\n| `regex` | Regex pattern match | `{\"type\": \"regex\", \"pattern\": \"order-\\\\d+\"}` |\n| `json_path` | JSONPath value check | `{\"type\": \"json_path\", \"path\": \"$.status\", \"value\": \"ok\"}` |\n| `tool_called` | Tool was invoked | `{\"type\": \"tool_called\", \"name\": \"create_expense\"}` |\n| `type` | Value type check | `{\"type\": \"type\", \"path\": \"$.count\", \"value\": \"number\"}` |\n\n### Agent-Driven Assertions\n\nFor semantic or fuzzy validation:\n\n```jsonl\n{\n \"type\": \"agent\",\n \"use\": \"agents:workers.test.validator\",\n \"options\": {\n \"metadata\": {\n \"criteria\": \"Response should be helpful and answer the user's question\",\n \"tone\": \"professional and friendly\"\n }\n }\n}\n```\n\n### Script Assertions\n\nFor custom validation logic:\n\n```jsonl\n{\n \"type\": \"script\",\n \"use\": \"scripts:tests.validate-expense\",\n \"options\": {\n \"metadata\": {\n \"min_amount\": 100,\n \"max_amount\": 10000\n }\n }\n}\n```\n\n## Script Testing with Agent Assertions\n\nScript tests can use Agent-driven assertions via `t.assert.Agent()`:\n\n```typescript\nexport function TestExpenseResponse(t: TestingT, ctx: Context) {\n const messages = [\n { role: \"user\", content: \"I want to submit an expense\" },\n { role: \"assistant\", content: \"What type of expense?\" },\n { role: \"user\", content: \"Travel, $500\" },\n ];\n\n const response = Process(\"agents.expense.Stream\", ctx, messages);\n\n // Static assertion\n t.assert.Contains(response.content, \"confirm\");\n\n // Agent-driven assertion\n t.assert.Agent(response.content, \"workers.test.validator\", {\n metadata: {\n criteria: \"Response should ask for confirmation before creating expense\",\n conversation: messages,\n },\n });\n}\n```\n\n## Standard Agent Interface\n\nAll agent-driven features use `context.Options`:\n\n```go\ntype Options struct {\n Skip *Skip `json:\"skip,omitempty\"`\n Connector string `json:\"connector,omitempty\"`\n Search any `json:\"search,omitempty\"`\n Mode string `json:\"mode,omitempty\"`\n Metadata map[string]any `json:\"metadata,omitempty\"`\n}\n```\n\n### Generator Agent\n\nCalled when `-i agents:xxx` is used:\n\n```go\noptions := &context.Options{\n Metadata: map[string]any{\n \"test_mode\": \"generator\",\n \"target_agent\": \"assistants.expense\",\n \"count\": 10,\n \"focus\": \"edge-cases\",\n },\n}\n```\n\n### Simulator Agent\n\nCalled in dynamic mode to generate user input:\n\n```go\noptions := &context.Options{\n Metadata: map[string]any{\n \"test_mode\": \"simulator\",\n \"persona\": \"New employee\",\n \"goal\": \"Submit expense\",\n \"turn_number\": 3,\n },\n}\n```\n\n### Validator Agent\n\nCalled for agent-driven assertions:\n\n```go\noptions := &context.Options{\n Metadata: map[string]any{\n \"test_mode\": \"validator\",\n \"criteria\": \"Response should be helpful\",\n },\n}\n```\n\n## Command Line Interface\n\n### Flags Reference\n\n| Flag | Long | Description |\n| ---- | ------------- | ------------------------------------------------------------ |\n| `-i` | `--input` | Input source: file path, message, or `agents:`/`scripts:` ID |\n| `-n` | `--name` | Target agent ID (the agent being tested) |\n| `-o` | `--output` | Output file path for results |\n| `-c` | `--connector` | Override connector for the target agent |\n| `-u` | `--user` | Test user ID (default: test-user) |\n| `-t` | `--team` | Test team ID (default: test-team) |\n| `-v` | `--verbose` | Verbose output |\n| | `--ctx` | Path to context JSON file for custom authorization |\n| | `--simulator` | Default simulator agent ID for dynamic mode |\n| | `--before` | Global before script (e.g., `env_test.BeforeAll`) |\n| | `--after` | Global after script (e.g., `env_test.AfterAll`) |\n| | `--timeout` | Timeout per test case (default: 2m) |\n| | `--parallel` | Number of parallel test cases |\n| | `--runs` | Number of runs for stability analysis |\n| | `--run` | Regex pattern to filter which tests to run |\n| | `--fail-fast` | Stop on first failure |\n| | `--dry-run` | Generate/parse tests without running |\n\n### Examples\n\n```bash\n# Simple test\nyao agent test -i \"Hello, how are you?\" -n assistants.chat\n\n# From JSONL file\nyao agent test -i ./tests/expense.jsonl\n\n# Agent-generated tests\nyao agent test -i \"agents:workers.test.generator?count=10\" -n assistants.expense\n\n# With simulator for dynamic mode\nyao agent test -i ./tests/dynamic.jsonl --simulator workers.test.user-simulator\n\n# Parallel execution\nyao agent test -i ./tests/expense.jsonl --parallel 5\n\n# Verbose output\nyao agent test -i ./tests/expense.jsonl -v\n```\n\n## Output Format\n\n### Console Output (Standard Mode)\n\nStandard mode shows each test case as a single line with input preview:\n\n```\n═══════════════════════════════════════════════════════════════\n Agent Test\n═══════════════════════════════════════════════════════════════\nℹ Agent: workers.system.keyword\nℹ Connector: deepseek.v3\nℹ Input: ./tests/inputs.jsonl (42 test cases)\nℹ Timeout: 5m0s\n\n───────────────────────────────────────────────────────────────\n Running Tests\n───────────────────────────────────────────────────────────────\n► [T001] 人工智能和机器学习正在改变我们�... PASSED (2.7s)\n► [T002] The rapid development of cloud computing has re... PASSED (3.0s)\n► [T003] 区块链技术是一种分布式账本技术�... PASSED (2.7s)\n...\n\n───────────────────────────────────────────────────────────────\n Summary\n───────────────────────────────────────────────────────────────\n Agent: workers.system.keyword\n Connector: deepseek.v3\n Total: 42\n Passed: 42\n Failed: 0\n Pass Rate: 100.0%\n Duration: 1.8m\n\n Output: ./tests/output-20251225185335.jsonl\n\n═══════════════════════════════════════════════════════════════\n ✨ ALL TESTS PASSED ✨\n═══════════════════════════════════════════════════════════════\n```\n\n### Console Output (Dynamic Mode)\n\nDynamic mode shows each test case as a tree with turns and checkpoints:\n\n```\n═══════════════════════════════════════════════════════════════\n Agent Test (Dynamic Mode)\n═══════════════════════════════════════════════════════════════\nℹ Agent: assistants.expense\nℹ Connector: openai.gpt4\nℹ Input: ./tests/dynamic.jsonl (2 test cases)\nℹ Simulator: workers.test.user-simulator\n\n───────────────────────────────────────────────────────────────\n Running Tests\n───────────────────────────────────────────────────────────────\n► [T001] Expense Submission Coverage\n ├─ Turn 1: \"Help me file an expense\" → \"What type of expense?\"\n │ └─ ✓ checkpoint: ask_type\n ├─ Turn 2: \"Client dinner, $250\" → \"I'll create... Please confirm.\"\n │ └─ ✓ checkpoint: call_create (tool: create_expense)\n └─ Turn 3: \"Yes, confirm\" → \"Expense submitted! Reference: EXP-001\"\n └─ ✓ checkpoint: confirm\n PASSED (6.8s) - 3 turns, 3/3 checkpoints\n\n► [T002] Expense with Attachment\n ├─ Turn 1: \"Submit receipt\" + [receipt.jpg] → \"What type?\"\n │ └─ ✓ checkpoint: ask_type\n ├─ Turn 2: \"Business lunch\" → \"Amount from receipt: $85.50. Confirm?\"\n │ └─ ✓ checkpoint: extract_amount\n └─ Turn 3: \"Yes\" → \"Submitted! Reference: EXP-002\"\n └─ ✓ checkpoint: confirm\n PASSED (8.2s) - 3 turns, 3/3 checkpoints\n\n───────────────────────────────────────────────────────────────\n Summary\n───────────────────────────────────────────────────────────────\n Agent: assistants.expense\n Connector: openai.gpt4\n Simulator: workers.test.user-simulator\n Total: 2\n Passed: 2\n Failed: 0\n Pass Rate: 100.0%\n Duration: 15.0s\n\n Output: ./tests/output-20251225190000.jsonl\n\n═══════════════════════════════════════════════════════════════\n ✨ ALL TESTS PASSED ✨\n═══════════════════════════════════════════════════════════════\n```\n\n### Console Output (Parallel Mode)\n\nWhen `--parallel N` is enabled, tests run concurrently. Output is buffered and displayed as complete test trees:\n\n```\n═══════════════════════════════════════════════════════════════\n Agent Test (Parallel: 5)\n═══════════════════════════════════════════════════════════════\nℹ Agent: assistants.expense\nℹ Input: ./tests/dynamic.jsonl (10 test cases)\nℹ Parallel: 5 concurrent\n\n───────────────────────────────────────────────────────────────\n Running Tests (5 parallel)\n───────────────────────────────────────────────────────────────\n► [T003] Quick approval flow\n ├─ Turn 1: \"Approve expense EXP-001\" → \"Approved!\"\n └─ ✓ checkpoint: approved\n PASSED (1.2s) - 1 turn, 1/1 checkpoints\n\n► [T001] Expense Submission Coverage\n ├─ Turn 1: \"Help me file an expense\" → \"What type?\"\n │ └─ ✓ checkpoint: ask_type\n ├─ Turn 2: \"Client dinner, $250\" → \"Confirm?\"\n │ └─ ✓ checkpoint: call_create\n └─ Turn 3: \"Yes\" → \"Submitted!\"\n └─ ✓ checkpoint: confirm\n PASSED (6.8s) - 3 turns, 3/3 checkpoints\n\n► [T002] Expense with Attachment\n ├─ Turn 1: \"Submit receipt\" + [receipt.jpg] → \"What type?\"\n ...\n PASSED (8.2s) - 3 turns, 3/3 checkpoints\n\n[Progress: 3/10 completed, 5 running...]\n\n► [T004] Rejection flow\n ...\n PASSED (4.5s) - 2 turns, 2/2 checkpoints\n\n───────────────────────────────────────────────────────────────\n Summary\n───────────────────────────────────────────────────────────────\n Total: 10\n Passed: 10\n Failed: 0\n Pass Rate: 100.0%\n Duration: 25.3s (effective: 2.5s/test with 5 parallel)\n\n═══════════════════════════════════════════════════════════════\n ✨ ALL TESTS PASSED ✨\n═══════════════════════════════════════════════════════════════\n```\n\n**Note**: In parallel mode, test results appear in completion order (not input order). Each test's output is buffered and displayed as a complete tree to maintain readability.\n\n### JSON Output (Standard Mode)\n\nOutput file is a JSON object with `summary`, `environment`, `results`, and `metadata`:\n\n```json\n{\n \"summary\": {\n \"total\": 3,\n \"passed\": 3,\n \"failed\": 0,\n \"skipped\": 0,\n \"errors\": 0,\n \"timeouts\": 0,\n \"duration_ms\": 5100,\n \"agent_id\": \"assistants.expense\",\n \"agent_path\": \"/path/to/expense\"\n },\n \"environment\": {\n \"user_id\": \"test-user\",\n \"team_id\": \"test-team\",\n \"locale\": \"en-us\"\n },\n \"results\": [\n {\n \"id\": \"expense-turn1\",\n \"status\": \"passed\",\n \"input\": [{ \"role\": \"user\", \"content\": \"I want to submit an expense\" }],\n \"output\": \"What type of expense would you like to submit?\",\n \"duration_ms\": 1200\n },\n {\n \"id\": \"expense-turn2\",\n \"status\": \"passed\",\n \"input\": [\n { \"role\": \"user\", \"content\": \"I want to submit an expense\" },\n { \"role\": \"assistant\", \"content\": \"What type?\" },\n { \"role\": \"user\", \"content\": \"Business travel, $3500\" }\n ],\n \"output\": \"Confirm $3500 expense?\",\n \"duration_ms\": 2100\n }\n ],\n \"metadata\": {\n \"started_at\": \"2025-12-25T10:00:00Z\",\n \"completed_at\": \"2025-12-25T10:00:05Z\",\n \"input_file\": \"./tests/expense.jsonl\"\n }\n}\n```\n\n### JSON Output (Dynamic Mode)\n\nDynamic mode adds `turns` and `checkpoints` to each result:\n\n```json\n{\n \"summary\": {\n \"total\": 1,\n \"passed\": 1,\n \"failed\": 0,\n \"duration_ms\": 6800,\n \"agent_id\": \"assistants.expense\"\n },\n \"results\": [\n {\n \"id\": \"expense-dynamic\",\n \"name\": \"Expense Coverage Test\",\n \"status\": \"passed\",\n \"turns\": [\n {\n \"turn\": 1,\n \"input\": \"Help me file an expense\",\n \"output\": \"What type?\"\n },\n { \"turn\": 2, \"input\": \"Client dinner, $250\", \"output\": \"Confirm?\" },\n { \"turn\": 3, \"input\": \"Yes\", \"output\": \"Submitted!\" }\n ],\n \"checkpoints\": [\n { \"id\": \"ask_type\", \"reached_at_turn\": 1, \"passed\": true },\n { \"id\": \"call_create\", \"reached_at_turn\": 2, \"passed\": true },\n { \"id\": \"confirm\", \"reached_at_turn\": 3, \"passed\": true }\n ],\n \"total_turns\": 3,\n \"duration_ms\": 6800\n }\n ],\n \"metadata\": {\n \"started_at\": \"2025-12-25T10:00:00Z\",\n \"completed_at\": \"2025-12-25T10:00:07Z\"\n }\n}\n```\n\n## User Simulator Agent\n\n### Interface\n\n```typescript\ninterface SimulatorInput {\n persona: string;\n goal: string;\n conversation: Message[];\n turn_number: number;\n max_turns: number;\n}\n\ninterface SimulatorOutput {\n input: string;\n goal_achieved: boolean;\n reasoning?: string;\n}\n```\n\n### Example Prompt\n\n```\nYou are simulating a user with the following characteristics:\n\nPersona: {{persona}}\nGoal: {{goal}}\n\nCurrent conversation:\n{{conversation}}\n\nGenerate the next user message to continue toward the goal.\nIf the goal has been achieved, set goal_achieved to true.\n\nRespond in JSON format:\n{\n \"input\": \"your response as the user\",\n \"goal_achieved\": true/false,\n \"reasoning\": \"brief explanation\"\n}\n```\n\n## Backward Compatibility\n\nExisting single-turn tests work unchanged:\n\n```jsonl\n// Simple string input\n{\"id\": \"T001\", \"input\": \"Hello\", \"assertions\": [...]}\n\n// Equivalent to array format\n{\"id\": \"T001\", \"input\": [{\"role\": \"user\", \"content\": \"Hello\"}], \"assertions\": [...]}\n```\n\n## Error Handling\n\n### Standard Mode Errors\n\n| Error Type | Behavior | Output |\n| ---------------- | ----------- | ---------------------------- |\n| Agent timeout | Test FAILED | `error: \"timeout after 30s\"` |\n| Agent error | Test FAILED | `error: \"agent error: ...\"` |\n| Assertion failed | Test FAILED | `assertion_errors: [...]` |\n\n### Dynamic Mode Errors\n\n| Error Type | Behavior | Output |\n| --------------------------- | ----------- | ----------------------------------- |\n| All checkpoints reached | Test PASSED | `status: \"passed\"` |\n| Checkpoints missing | Test FAILED | `error: \"missing checkpoints: ...\"` |\n| Max turns exceeded | Test FAILED | `error: \"max turns (20) exceeded\"` |\n| Timeout exceeded | Test FAILED | `error: \"timeout after 5m\"` |\n| Simulator error | Test FAILED | `error: \"simulator error: ...\"` |\n| Checkpoint assertion failed | Test FAILED | `error: \"checkpoint X failed\"` |\n\n## Current Implementation Status\n\n| Feature | Status | Notes |\n| ----------------------- | ------- | -------------------------------------------------- |\n| Simple text input | ✅ Done | `input: \"Hello\"` |\n| Message history | ✅ Done | `input: [{role, content}, ...]` |\n| File attachments | ✅ Done | `file://` protocol in content parts |\n| Static assertions | ✅ Done | contains, equals, regex, json_path, etc. |\n| Before/After hooks | ✅ Done | `before/after` in JSONL, `--before/--after` in CLI |\n| Agent-driven assertions | ✅ Done | `type: \"agent\"` + `t.assert.Agent()` JSAPI |\n| Agent-driven input | ✅ Done | `-i agents:xxx` for test generation |\n| Dry-run mode | ✅ Done | `--dry-run` to preview generated tests |\n| Dynamic mode | ✅ Done | Simulator + Checkpoints |\n| Console output | ✅ Done | Dynamic mode tree output, checkpoint display |\n\n## Open Questions\n\n1. **Message Generation**: Should we provide a helper to generate message history from a script?\n\n2. **Snapshot Testing**: Should we support \"golden file\" comparison for responses?\n\n3. **Retry Logic**: If a test fails, should we support automatic retry?\n" }, { "path": "agent/test/README.md", "content": "# Agent Test Framework\n\nA comprehensive testing framework for Yao AI agents with support for standard testing, dynamic (simulator-driven) testing, agent-driven assertions, and CI integration.\n\n## Quick Start\n\n### Standard Tests\n\n```bash\n# Test with direct message (auto-detect agent from current directory)\ncd assistants/keyword\nyao agent test -i \"Extract keywords from: AI and machine learning\"\n\n# Test with direct message (specify agent explicitly)\nyao agent test -i \"Hello world\" -n workers.system.keyword\n\n# Test with JSONL file (auto-detect agent from path)\nyao agent test -i assistants/keyword/tests/inputs.jsonl\n\n# Generate HTML report\nyao agent test -i tests/inputs.jsonl -o report.html\n\n# Stability analysis (run each test 5 times)\nyao agent test -i tests/inputs.jsonl --runs 5\n```\n\n### Agent-Driven Input\n\n```bash\n# Generate test cases using an agent\nyao agent test -i \"agents:tests.generator-agent?count=10\" -n assistants.expense\n\n# Preview generated tests without running (dry-run)\nyao agent test -i \"agents:tests.generator-agent?count=5\" -n assistants.expense --dry-run\n```\n\n### Dynamic Mode (Simulator)\n\n```bash\n# Run dynamic tests with simulator\nyao agent test -i tests/dynamic.jsonl --simulator tests.simulator-agent\n\n# See detailed turn-by-turn output\nyao agent test -i tests/dynamic.jsonl -v\n```\n\n### Script Tests\n\n```bash\n# Test agent handler scripts (hooks, tools, setup functions)\nyao agent test -i scripts.expense.setup -v\n\n# Run specific tests with regex filter\nyao agent test -i scripts.expense.setup --run \"TestSystemReady\" -v\n\n# Run with custom context (authorization, metadata)\nyao agent test -i scripts.expense.setup --ctx tests/context.json -v\n```\n\n## Input Modes\n\nThe `-i` flag supports multiple input modes:\n\n### 1. JSONL File Mode\n\nLoad test cases from a file:\n\n```bash\nyao agent test -i tests/inputs.jsonl\n```\n\nAgent is auto-detected by traversing up from the input file to find `package.yao`.\n\n### 2. Direct Message Mode\n\nTest with a single message:\n\n```bash\n# Auto-detect agent from current working directory\ncd assistants/keyword\nyao agent test -i \"Extract keywords from this text\"\n\n# Or specify agent explicitly\nyao agent test -i \"Hello\" -n workers.system.keyword\n```\n\n### 3. Agent-Driven Input Mode\n\nGenerate test cases using a generator agent:\n\n```bash\n# Basic usage (-n specifies the target agent to test)\nyao agent test -i \"agents:tests.generator-agent\" -n assistants.expense\n\n# With parameters\nyao agent test -i \"agents:tests.generator-agent?count=10&focus=edge-cases\" -n assistants.expense\n\n# Dry-run to preview generated tests\nyao agent test -i \"agents:tests.generator-agent?count=5\" -n assistants.expense --dry-run\n```\n\n**Note**: The `-n` flag is **required** for agent-driven input mode to specify which agent to test. The generator agent creates test cases for the target agent.\n\n### 4. Script Test Mode\n\nTest agent handler scripts:\n\n```bash\nyao agent test -i scripts.expense.setup -v\n```\n\nScript test input format: `scripts..` (e.g., `scripts.expense.setup` → `assistants/expense/src/setup_test.ts`).\n\n### 5. Script-Generated Input Mode\n\nGenerate test cases using a script:\n\n```bash\nyao agent test -i \"scripts:tests.gen.Generate\" -n assistants.expense\n```\n\n**Note**: `scripts.xxx` (with dot) runs script tests, while `scripts:xxx` (with colon) generates test cases from a script.\n\n## Test Modes\n\n### Standard Mode\n\nSingle call to agent with optional message history. Each test is independent and stateless.\n\n```jsonl\n{\n \"id\": \"T001\",\n \"input\": \"Hello\",\n \"assert\": {\n \"type\": \"contains\",\n \"value\": \"Hi\"\n }\n}\n```\n\n### Dynamic Mode\n\nSimulator-driven testing with checkpoint validation. A simulator agent generates user messages while checkpoints verify agent behavior.\n\n```jsonl\n{\n \"id\": \"T001\",\n \"input\": \"I want to order coffee\",\n \"simulator\": {\n \"use\": \"tests.simulator-agent\",\n \"options\": {\n \"metadata\": {\n \"persona\": \"Customer\",\n \"goal\": \"Order a latte\"\n }\n }\n },\n \"checkpoints\": [\n {\n \"id\": \"greeting\",\n \"assert\": {\n \"type\": \"regex\",\n \"value\": \"(?i)hello\"\n }\n },\n {\n \"id\": \"ask_size\",\n \"after\": [\n \"greeting\"\n ],\n \"assert\": {\n \"type\": \"regex\",\n \"value\": \"(?i)size\"\n }\n }\n ],\n \"max_turns\": 10\n}\n```\n\n## Command Line Options\n\n| Flag | Description | Default |\n| ------------- | -------------------------------------------------------- | -------------------------- |\n| `-i` | Input: JSONL file, message, `agents:xxx`, or `scripts:x` | (required) |\n| `-o` | Output file path | `output-{timestamp}.jsonl` |\n| `-n` | Agent ID (optional, auto-detected) | auto-detect |\n| `-a` | Application directory | auto-detect |\n| `-e` | Environment file | - |\n| `-c` | Override connector | agent default |\n| `-u` | Test user ID | `test-user` |\n| `-t` | Test team ID | `test-team` |\n| `-r` | Reporter agent ID for custom report | built-in |\n| `-v` | Verbose output | false |\n| `--ctx` | Path to context JSON file for custom authorization | - |\n| `--simulator` | Default simulator agent ID for dynamic mode | - |\n| `--before` | Global BeforeAll hook (e.g., `env_test.BeforeAll`) | - |\n| `--after` | Global AfterAll hook (e.g., `env_test.AfterAll`) | - |\n| `--runs` | Runs per test (stability analysis) | 1 |\n| `--run` | Regex pattern to filter which tests to run | - |\n| `--timeout` | Timeout per test | 2m |\n| `--parallel` | Parallel test cases | 1 |\n| `--fail-fast` | Stop on first failure | false |\n| `--dry-run` | Generate test cases without running them | false |\n\n## Custom Context File\n\nCreate a JSON file for custom authorization:\n\n```json\n{\n \"chat_id\": \"test-chat-001\",\n \"authorized\": {\n \"user_id\": \"test-user-123\",\n \"team_id\": \"test-team-456\",\n \"constraints\": {\n \"owner_only\": true,\n \"extra\": { \"department\": \"engineering\" }\n }\n },\n \"metadata\": {\n \"mode\": \"test\"\n }\n}\n```\n\nUse with `--ctx`:\n\n```bash\nyao agent test -i scripts.expense.setup --ctx tests/context.json -v\n```\n\n## Input Format (JSONL)\n\nEach line is a JSON object. Below are examples organized by scenario.\n\n### Scenario 1: Simple Text Input\n\nBasic test with string input:\n\n```jsonl\n{\"id\": \"greeting-basic\", \"input\": \"Hello, how are you?\"}\n{\"id\": \"greeting-chinese\", \"input\": \"你好，请问有什么可以帮助你的？\"}\n```\n\n### Scenario 2: With Assertions\n\nValidate response content:\n\n```jsonl\n{\"id\": \"keyword-extract\", \"input\": \"Extract keywords from: AI and machine learning\", \"assert\": {\"type\": \"contains\", \"value\": \"AI\"}}\n{\"id\": \"json-response\", \"input\": \"What's the weather?\", \"assert\": {\"type\": \"json_path\", \"path\": \"need_search\", \"value\": true}}\n{\"id\": \"no-error\", \"input\": \"Help me\", \"assert\": {\"type\": \"not_contains\", \"value\": \"error\"}}\n```\n\n### Scenario 3: Multiple Assertions\n\nAll assertions must pass:\n\n```jsonl\n{\n \"id\": \"expense-submit\",\n \"input\": \"Submit $500 travel expense\",\n \"assert\": [\n {\n \"type\": \"contains\",\n \"value\": \"expense\"\n },\n {\n \"type\": \"not_contains\",\n \"value\": \"error\"\n },\n {\n \"type\": \"regex\",\n \"value\": \"(?i)(submitted|created|confirmed)\"\n }\n ]\n}\n```\n\n### Scenario 4: Conversation History\n\nTest with multi-turn context:\n\n```jsonl\n{\n \"id\": \"expense-confirm\",\n \"input\": [\n {\n \"role\": \"user\",\n \"content\": \"Submit an expense\"\n },\n {\n \"role\": \"assistant\",\n \"content\": \"What type of expense?\"\n },\n {\n \"role\": \"user\",\n \"content\": \"Travel, $500\"\n },\n {\n \"role\": \"assistant\",\n \"content\": \"Please confirm: $500 travel expense\"\n },\n {\n \"role\": \"user\",\n \"content\": \"Yes, confirm\"\n }\n ],\n \"assert\": {\n \"type\": \"regex\",\n \"value\": \"(?i)(submitted|created)\"\n }\n}\n```\n\n### Scenario 5: With File Attachments\n\nTest with images or documents:\n\n```jsonl\n{\n \"id\": \"receipt-analyze\",\n \"input\": {\n \"role\": \"user\",\n \"content\": [\n {\n \"type\": \"text\",\n \"text\": \"Analyze this receipt\"\n },\n {\n \"type\": \"image\",\n \"source\": \"file://fixtures/receipt.jpg\"\n }\n ]\n },\n \"assert\": {\n \"type\": \"contains\",\n \"value\": \"amount\"\n }\n}\n```\n\n### Scenario 6: Agent-Driven Assertion\n\nUse LLM to validate response semantics:\n\n```jsonl\n{\n \"id\": \"helpful-response\",\n \"input\": \"How do I reset my password?\",\n \"assert\": {\n \"type\": \"agent\",\n \"use\": \"agents:tests.validator-agent\",\n \"value\": \"Response should provide clear step-by-step instructions\"\n }\n}\n```\n\n### Scenario 7: With Options\n\nOverride connector or skip features:\n\n```jsonl\n{\"id\": \"fast-model\", \"input\": \"Quick question\", \"options\": {\"connector\": \"deepseek.v3\", \"skip\": {\"history\": true, \"trace\": true}}}\n{\"id\": \"scenario-test\", \"input\": \"Query users\", \"options\": {\"metadata\": {\"scenario\": \"filter\"}}, \"assert\": {\"type\": \"json_path\", \"path\": \"from\", \"value\": \"users\"}}\n```\n\n### Scenario 8: With Before/After Hooks\n\nSetup and teardown for each test:\n\n```jsonl\n{\n \"id\": \"with-user-data\",\n \"input\": \"Show my expenses\",\n \"before\": \"env_test.Before\",\n \"after\": \"env_test.After\",\n \"assert\": {\n \"type\": \"contains\",\n \"value\": \"expense\"\n }\n}\n```\n\n### Scenario 9: Skip Test\n\nTemporarily disable a test:\n\n```jsonl\n{\n \"id\": \"wip-feature\",\n \"input\": \"New feature test\",\n \"skip\": true\n}\n```\n\n### Scenario 10: Dynamic Mode (Simulator)\n\nMulti-turn testing with user simulator:\n\n```jsonl\n{\n \"id\": \"coffee-order\",\n \"input\": \"I want to order coffee\",\n \"simulator\": {\n \"use\": \"tests.simulator-agent\",\n \"options\": {\n \"metadata\": {\n \"persona\": \"Regular customer\",\n \"goal\": \"Order a medium latte\"\n }\n }\n },\n \"checkpoints\": [\n {\n \"id\": \"greeting\",\n \"assert\": {\n \"type\": \"regex\",\n \"value\": \"(?i)(hello|hi|help)\"\n }\n },\n {\n \"id\": \"ask-size\",\n \"after\": [\n \"greeting\"\n ],\n \"assert\": {\n \"type\": \"regex\",\n \"value\": \"(?i)size\"\n }\n },\n {\n \"id\": \"confirm\",\n \"after\": [\n \"ask-size\"\n ],\n \"assert\": {\n \"type\": \"regex\",\n \"value\": \"(?i)confirm\"\n }\n }\n ],\n \"max_turns\": 10\n}\n```\n\n### Scenario 11: Dynamic Mode with Optional Checkpoint\n\nSome checkpoints are optional:\n\n```jsonl\n{\n \"id\": \"expense-flow\",\n \"input\": \"Submit expense\",\n \"simulator\": {\n \"use\": \"tests.simulator-agent\",\n \"options\": {\n \"metadata\": {\n \"persona\": \"New employee\",\n \"goal\": \"Submit $500 travel expense\"\n }\n }\n },\n \"checkpoints\": [\n {\n \"id\": \"ask-type\",\n \"assert\": {\n \"type\": \"regex\",\n \"value\": \"(?i)type\"\n }\n },\n {\n \"id\": \"suggest-category\",\n \"required\": false,\n \"assert\": {\n \"type\": \"contains\",\n \"value\": \"category\"\n }\n },\n {\n \"id\": \"confirm\",\n \"after\": [\n \"ask-type\"\n ],\n \"assert\": {\n \"type\": \"regex\",\n \"value\": \"(?i)confirm\"\n }\n }\n ],\n \"max_turns\": 15\n}\n```\n\n### Standard Mode Fields\n\n| Field | Type | Required | Description |\n| ---------- | ------------------------------ | -------- | ------------------------------------- |\n| `id` | string | Yes | Test case ID |\n| `input` | string \\| Message \\| []Message | Yes | Test input |\n| `assert` | Assertion \\| []Assertion | No | Assertion rules |\n| `expected` | any | No | Expected output (exact match) |\n| `user` | string | No | Override user ID for this test |\n| `team` | string | No | Override team ID for this test |\n| `metadata` | map | No | Additional metadata for hooks |\n| `options` | Options | No | Context options |\n| `timeout` | string | No | Override timeout (e.g., \"30s\") |\n| `skip` | bool | No | Skip this test |\n| `before` | string | No | Before hook (e.g., `env_test.Before`) |\n| `after` | string | No | After hook (e.g., `env_test.After`) |\n\n### Dynamic Mode Fields\n\n| Field | Type | Required | Description |\n| ----------------------------- | ------ | -------- | -------------------------------------- |\n| `id` | string | Yes | Test case ID |\n| `input` | string | Yes | Initial user message |\n| `simulator` | object | Yes | Simulator configuration |\n| `simulator.use` | string | Yes | Simulator agent ID (no prefix) |\n| `simulator.options` | object | No | Simulator options |\n| `simulator.options.metadata` | map | No | Metadata (persona, goal, etc.) |\n| `simulator.options.connector` | string | No | Override simulator connector |\n| `checkpoints` | array | Yes | Checkpoints to verify |\n| `checkpoints[].id` | string | Yes | Checkpoint identifier |\n| `checkpoints[].description` | string | No | Human-readable description |\n| `checkpoints[].assert` | object | Yes | Assertion to validate |\n| `checkpoints[].after` | array | No | Checkpoint IDs that must occur first |\n| `checkpoints[].required` | bool | No | Is checkpoint required (default: true) |\n| `max_turns` | int | No | Maximum turns (default: 20) |\n| `timeout` | string | No | Override timeout (e.g., \"2m\") |\n\n### Options\n\nThe `options` field allows per-test-case configuration:\n\n| Field | Type | Description |\n| ------------------------ | ------ | ------------------------------------------ |\n| `connector` | string | Override connector (e.g., `\"deepseek.v3\"`) |\n| `mode` | string | Agent mode (default: `\"chat\"`) |\n| `search` | bool | Enable/disable search mode |\n| `disable_global_prompts` | bool | Temporarily disable global prompts |\n| `metadata` | map | Custom data passed to hooks |\n| `skip` | object | Skip configuration (see below) |\n\n### Options.skip\n\n| Field | Type | Description |\n| --------- | ---- | ----------------------- |\n| `history` | bool | Skip history loading |\n| `trace` | bool | Skip trace logging |\n| `output` | bool | Skip output to client |\n| `keyword` | bool | Skip keyword extraction |\n| `search` | bool | Skip auto search |\n\n### Input Types\n\n| Type | Description | Example |\n| ----------- | -------------------- | ----------------------------------------------------- |\n| `string` | Simple text | `\"Hello world\"` |\n| `Message` | Single message | `{\"role\": \"user\", \"content\": \"...\"}` |\n| `[]Message` | Conversation history | `[{\"role\": \"user\", ...}, {\"role\": \"assistant\", ...}]` |\n\n## Assertions\n\nUse `assert` for flexible validation. If `assert` is defined, it takes precedence over `expected`.\n\n### Static Assertions\n\n| Type | Description | Example |\n| -------------- | ----------------------------- | --------------------------------------------------------- |\n| `equals` | Exact match | `{\"type\": \"equals\", \"value\": {\"key\": \"val\"}}` |\n| `contains` | Output contains value | `{\"type\": \"contains\", \"value\": \"keyword\"}` |\n| `not_contains` | Output does not contain value | `{\"type\": \"not_contains\", \"value\": \"error\"}` |\n| `json_path` | Extract JSON path and compare | `{\"type\": \"json_path\", \"path\": \"$.field\", \"value\": true}` |\n| `regex` | Match regex pattern | `{\"type\": \"regex\", \"value\": \"\\\\d+\"}` |\n| `type` | Check output type | `{\"type\": \"type\", \"value\": \"object\"}` |\n| `tool_called` | Check if a tool was called | `{\"type\": \"tool_called\", \"value\": \"setup\"}` |\n| `tool_result` | Check tool execution result | `{\"type\": \"tool_result\", \"value\": {\"tool\": \"setup\", \"result\": {\"success\": true}}}` |\n\n### Assertion Fields\n\n| Field | Type | Description |\n| --------- | ------ | -------------------------------------------------------- |\n| `type` | string | Assertion type (required) |\n| `value` | any | Expected value or pattern |\n| `path` | string | JSON path for `json_path` type |\n| `script` | string | Script name for `script` type |\n| `use` | string | Agent/script ID for `agent` type (with `agents:` prefix) |\n| `options` | object | Options for agent assertions |\n| `message` | string | Custom failure message |\n| `negate` | bool | Invert the assertion result |\n\n### Agent-Driven Assertions\n\nFor semantic or fuzzy validation using an LLM:\n\n```jsonl\n{\n \"id\": \"T001\",\n \"input\": \"Hello\",\n \"assert\": {\n \"type\": \"agent\",\n \"use\": \"agents:tests.validator-agent\",\n \"value\": \"Response should be friendly and helpful\"\n }\n}\n```\n\nThe validator agent receives the output and criteria, then returns `{\"passed\": true/false, \"reason\": \"...\"}`.\n\n**How it works:**\n\n1. The framework builds a validation request with the agent's response (including tool result messages)\n2. The validator agent evaluates the response against the criteria\n3. The validator returns a JSON response with `passed` and `reason`\n\n**Output in test report (for checkpoints):**\n\n```json\n{\n \"agent_validation\": {\n \"passed\": true,\n \"reason\": \"Response explicitly confirms setup completion\",\n \"criteria\": \"Response should be friendly and helpful\",\n \"input\": \"Hello! How can I help you today?\",\n \"response\": {\n \"passed\": true,\n \"reason\": \"Response explicitly confirms setup completion\"\n }\n }\n}\n```\n\n- `input`: The content sent to the validator (agent response + tool result messages)\n- `response`: The raw JSON response from the validator agent\n- `criteria`: The validation criteria from the test case\n\n### Tool Assertions\n\nFor validating that specific tools were called and their results:\n\n#### tool_called\n\nCheck if a specific tool was called:\n\n```jsonl\n{\n \"id\": \"T001\",\n \"input\": \"Set up my expense system\",\n \"assert\": {\n \"type\": \"tool_called\",\n \"value\": \"setup\"\n }\n}\n```\n\n**Value formats:**\n\n- **String**: Tool name (supports suffix matching, e.g., `\"setup\"` matches `\"agents_expense_tools__setup\"`)\n- **Array**: Any of the specified tools must be called\n- **Object**: Match tool name and optionally arguments\n\n```jsonl\n// Match any of these tools\n{\"type\": \"tool_called\", \"value\": [\"setup\", \"init\"]}\n\n// Match tool with specific arguments\n{\"type\": \"tool_called\", \"value\": {\"name\": \"setup\", \"arguments\": {\"action\": \"init\"}}}\n```\n\n#### tool_result\n\nCheck the result of a tool execution:\n\n```jsonl\n{\n \"id\": \"T001\",\n \"input\": \"Set up my expense system\",\n \"assert\": {\n \"type\": \"tool_result\",\n \"value\": {\n \"tool\": \"setup\",\n \"result\": {\n \"success\": true\n }\n }\n }\n}\n```\n\n**Result matching:**\n\n- If `result` is omitted, only checks that the tool executed without error\n- Supports partial matching (only specified fields are checked)\n- Supports regex patterns with `regex:` prefix for string values\n\n```jsonl\n// Just check tool executed without error\n{\"type\": \"tool_result\", \"value\": {\"tool\": \"setup\"}}\n\n// Check specific result fields\n{\"type\": \"tool_result\", \"value\": {\"tool\": \"setup\", \"result\": {\"success\": true}}}\n\n// Use regex for message matching\n{\"type\": \"tool_result\", \"value\": {\"tool\": \"setup\", \"result\": {\"message\": \"regex:(?i)setup.*complete\"}}}\n```\n\n### Script Assertions\n\nFor custom validation logic:\n\n```jsonl\n{\n \"id\": \"T001\",\n \"input\": \"Test\",\n \"assert\": {\n \"type\": \"script\",\n \"script\": \"scripts.test.Validate\"\n }\n}\n```\n\n### Multiple Assertions\n\nAll assertions must pass:\n\n```jsonl\n{\n \"id\": \"T001\",\n \"input\": \"Hello\",\n \"assert\": [\n {\n \"type\": \"contains\",\n \"value\": \"Hi\"\n },\n {\n \"type\": \"not_contains\",\n \"value\": \"error\"\n },\n {\n \"type\": \"json_path\",\n \"path\": \"status\",\n \"value\": \"ok\"\n }\n ]\n}\n```\n\n## File Attachments\n\nTest inputs support file attachments using the `file://` protocol:\n\n```jsonl\n{\n \"id\": \"T001\",\n \"input\": {\n \"role\": \"user\",\n \"content\": [\n {\n \"type\": \"text\",\n \"text\": \"Analyze this image\"\n },\n {\n \"type\": \"image\",\n \"source\": \"file://fixtures/receipt.jpg\"\n }\n ]\n }\n}\n```\n\nSupported types: images (jpg, png, gif, webp), audio (wav, mp3), documents (pdf, doc, txt).\n\n## Before/After Hooks\n\nHooks allow you to run setup and teardown code before and after tests. Hook scripts must be placed in the agent's `src/` directory with `_test.ts` suffix.\n\n### Hook Types\n\n| Hook | Scope | When Called | Use Case |\n| ----------- | -------- | --------------------- | ------------------------------- |\n| `Before` | Per-test | Before each test case | Create test data, setup context |\n| `After` | Per-test | After each test case | Cleanup test data, log results |\n| `BeforeAll` | Global | Once before all tests | Database migration, init |\n| `AfterAll` | Global | Once after all tests | Global cleanup, report |\n\n### Execution Order\n\n```\nBeforeAll (global)\n ├─ Before (test 1)\n │ └─ Test 1 execution\n │ └─ After (test 1)\n ├─ Before (test 2)\n │ └─ Test 2 execution\n │ └─ After (test 2)\n └─ ...\nAfterAll (global)\n```\n\n### Per-Test Hooks\n\nDefined in JSONL, scripts located in agent's `src/` directory:\n\n```jsonl\n{\n \"id\": \"T001\",\n \"input\": \"Test\",\n \"before\": \"env_test.Before\",\n \"after\": \"env_test.After\"\n}\n```\n\n### Global Hooks\n\nVia CLI flags:\n\n```bash\nyao agent test -i tests/inputs.jsonl --before env_test.BeforeAll --after env_test.AfterAll\n```\n\n### Hook Function Signatures\n\n```typescript\n// assistants/expense/src/env_test.ts\n\n/**\n * Before - Called before each test case\n * @param ctx - Agent context with user/team info\n * @param testCase - The test case about to run\n * @returns any - Data passed to After hook (optional)\n */\nexport function Before(ctx: Context, testCase: TestCase): any {\n const userId = Process(\"models.user.Create\", { name: \"Test User\" });\n return { userId }; // This data is passed to After\n}\n\n/**\n * After - Called after each test case (pass or fail)\n * @param ctx - Agent context\n * @param testCase - The test case that ran\n * @param result - Test result with status, output, duration\n * @param beforeData - Data returned from Before hook\n */\nexport function After(\n ctx: Context,\n testCase: TestCase,\n result: TestResult,\n beforeData: any\n) {\n if (beforeData?.userId) {\n Process(\"models.user.Delete\", beforeData.userId);\n }\n if (result.status === \"failed\") {\n console.log(`Test ${testCase.id} failed: ${result.error}`);\n }\n}\n\n/**\n * BeforeAll - Called once before all tests\n * @param ctx - Agent context\n * @param testCases - Array of all test cases\n * @returns any - Data passed to AfterAll hook (optional)\n */\nexport function BeforeAll(ctx: Context, testCases: TestCase[]): any {\n Process(\"models.migrate\");\n return { initialized: true, count: testCases.length };\n}\n\n/**\n * AfterAll - Called once after all tests complete\n * @param ctx - Agent context\n * @param results - Array of all test results\n * @param beforeData - Data returned from BeforeAll hook\n */\nexport function AfterAll(ctx: Context, results: TestResult[], beforeData: any) {\n const passed = results.filter((r) => r.status === \"passed\").length;\n console.log(`Tests completed: ${passed}/${results.length} passed`);\n Process(\"models.cleanup\");\n}\n```\n\n### Hook Parameters\n\n**Context** - Agent execution context:\n\n```typescript\ninterface Context {\n locale: string; // Locale (e.g., \"en-us\")\n authorized: {\n user_id: string; // Test user ID\n team_id: string; // Test team ID\n constraints?: object; // Access constraints\n };\n metadata: object; // Custom metadata from test case\n}\n```\n\n**TestCase** - Test case definition:\n\n```typescript\ninterface TestCase {\n id: string; // Test case ID\n input: any; // Test input (string, Message, or Message[])\n assert?: object; // Assertion rules\n expected?: any; // Expected output\n user?: string; // Override user ID\n team?: string; // Override team ID\n metadata?: object; // Custom metadata\n options?: object; // Context options\n timeout?: string; // Timeout (e.g., \"30s\")\n skip?: boolean; // Skip flag\n before?: string; // Before hook reference\n after?: string; // After hook reference\n}\n```\n\n**TestResult** - Test execution result:\n\n```typescript\ninterface TestResult {\n id: string; // Test case ID\n status: string; // \"passed\" | \"failed\" | \"error\" | \"skipped\" | \"timeout\"\n input: any; // Actual input sent\n output: any; // Agent response\n expected?: any; // Expected output (if defined)\n error?: string; // Error message (if failed)\n duration_ms: number; // Execution time in milliseconds\n assertions?: object[]; // Assertion results\n}\n```\n\n### Common Use Cases\n\n**Database Setup/Teardown**:\n\n```typescript\nexport function Before(ctx: Context, testCase: TestCase): any {\n // Create test records\n const user = Process(\"models.user.Create\", {\n name: \"Test\",\n email: \"test@example.com\",\n });\n const expense = Process(\"models.expense.Create\", {\n user_id: user.id,\n amount: 100,\n });\n return { user, expense };\n}\n\nexport function After(\n ctx: Context,\n testCase: TestCase,\n result: TestResult,\n data: any\n) {\n // Clean up in reverse order\n if (data?.expense) Process(\"models.expense.Delete\", data.expense.id);\n if (data?.user) Process(\"models.user.Delete\", data.user.id);\n}\n```\n\n**Conditional Setup Based on Metadata**:\n\n```typescript\nexport function Before(ctx: Context, testCase: TestCase): any {\n const scenario = testCase.metadata?.scenario || \"default\";\n\n if (scenario === \"empty_db\") {\n Process(\"models.expense.DeleteAll\");\n } else if (scenario === \"with_data\") {\n Process(\"scripts.tests.seed.LoadTestData\");\n }\n\n return { scenario };\n}\n```\n\n**Logging and Debugging**:\n\n```typescript\nexport function After(\n ctx: Context,\n testCase: TestCase,\n result: TestResult,\n data: any\n) {\n if (result.status === \"failed\") {\n console.log(\"=== Test Failed ===\");\n console.log(\"Test ID:\", testCase.id);\n console.log(\"Input:\", JSON.stringify(testCase.input));\n console.log(\"Output:\", JSON.stringify(result.output));\n console.log(\"Error:\", result.error);\n }\n}\n```\n\n## Script Testing\n\nTest agent handler scripts with the `t.assert` API:\n\n```typescript\n// assistants/expense/src/setup_test.ts\nimport { SystemReady } from \"./setup\";\n\nexport function TestSystemReady(t: TestingT, ctx: Context) {\n const result = SystemReady(ctx);\n\n t.assert.True(result.success, \"Should succeed\");\n t.assert.Equal(result.status, \"ready\", \"Status should be ready\");\n t.assert.NotNil(result.data, \"Data should not be nil\");\n}\n\nexport function TestWithAgentAssertion(t: TestingT, ctx: Context) {\n const response = Process(\"agents.expense.Stream\", ctx, messages);\n\n // Static assertion\n t.assert.Contains(response.content, \"confirm\");\n\n // Agent-driven assertion\n t.assert.Agent(response.content, \"tests.validator-agent\", {\n criteria: \"Response should ask for confirmation\",\n });\n}\n```\n\n### Available Assertions\n\n| Method | Description |\n| -------------------------------- | ------------------------------ |\n| `t.assert.True(value, msg)` | Assert value is true |\n| `t.assert.False(value, msg)` | Assert value is false |\n| `t.assert.Equal(a, b, msg)` | Assert a equals b |\n| `t.assert.NotEqual(a, b, msg)` | Assert a not equals b |\n| `t.assert.Nil(value, msg)` | Assert value is null/undefined |\n| `t.assert.NotNil(value, msg)` | Assert value is not nil |\n| `t.assert.Contains(s, sub, msg)` | Assert string contains substr |\n| `t.assert.Len(arr, n, msg)` | Assert array/string length |\n| `t.assert.Agent(resp, id, opts)` | Agent-driven assertion |\n\n## Dynamic Mode\n\nFor testing complex conversation flows where the path is unpredictable:\n\n```jsonl\n{\n \"id\": \"coffee-order\",\n \"input\": \"I want to order coffee\",\n \"simulator\": {\n \"use\": \"tests.simulator-agent\",\n \"options\": {\n \"metadata\": {\n \"persona\": \"Customer ordering a latte\",\n \"goal\": \"Complete the coffee order\"\n }\n }\n },\n \"checkpoints\": [\n {\n \"id\": \"greeting\",\n \"description\": \"Agent greets customer\",\n \"assert\": {\n \"type\": \"regex\",\n \"value\": \"(?i)(hello|hi|help)\"\n }\n },\n {\n \"id\": \"ask_size\",\n \"description\": \"Agent asks for size\",\n \"after\": [\n \"greeting\"\n ],\n \"assert\": {\n \"type\": \"regex\",\n \"value\": \"(?i)size\"\n }\n },\n {\n \"id\": \"confirm\",\n \"description\": \"Agent confirms order\",\n \"after\": [\n \"ask_size\"\n ],\n \"assert\": {\n \"type\": \"regex\",\n \"value\": \"(?i)confirm\"\n }\n }\n ],\n \"max_turns\": 10\n}\n```\n\n### Console Output (Dynamic Mode)\n\n```\n► [coffee-order] (dynamic, 3 checkpoints)\nℹ Dynamic test: coffee-order (max 10 turns)\nℹ Turn 1: User: I want to order coffee\nℹ Turn 1: Agent: Hello! What can I get for you?\nℹ ✓ checkpoint: greeting\nℹ Turn 2: User: A medium latte please\nℹ Turn 2: Agent: What size would you like?\nℹ ✓ checkpoint: ask_size\nℹ Turn 3: User: Medium\nℹ Turn 3: Agent: Let me confirm: Medium latte. Correct?\nℹ ✓ checkpoint: confirm\n └─ PASSED (3 turns, 3 checkpoints, 8.5s)\n```\n\n### Dynamic Mode Output Structure\n\nEach turn in the output includes:\n\n```typescript\ninterface TurnResult {\n turn: number; // Turn number (1-based)\n input: string; // User message\n output: any; // Agent response summary (for display)\n response: {\n // Full agent response (for detailed analysis)\n content: string; // LLM text content\n tool_calls: [\n {\n // Tool calls made\n tool: string; // Tool name\n arguments: any; // Call arguments\n result: any; // Execution result\n }\n ];\n next: any; // Next hook data\n };\n checkpoints_reached: string[]; // Checkpoint IDs reached\n duration_ms: number; // Execution time\n}\n```\n\n### Checkpoint Result Structure\n\nEach checkpoint in the output includes:\n\n```typescript\ninterface CheckpointResult {\n id: string; // Checkpoint identifier\n reached: boolean; // Whether checkpoint was reached\n reached_at_turn?: number; // Turn number when reached (if reached)\n required: boolean; // Whether checkpoint is required\n passed: boolean; // Whether assertion passed\n message?: string; // Assertion result message\n agent_validation?: {\n // Agent assertion details (for type: \"agent\")\n passed: boolean; // Validator's determination\n reason: string; // Explanation from validator\n criteria: string; // Validation criteria checked\n input: any; // Content sent to validator\n response: {\n // Raw validator response\n passed: boolean;\n reason: string;\n };\n };\n}\n```\n\n**Note**: For agent-based assertions (`type: \"agent\"`), the `agent_validation` field provides full transparency into the validation process. The `input` field contains the combined output (agent text response + tool result messages) that was validated.\n\n## Output Formats\n\nDetermined by `-o` file extension:\n\n| Extension | Format | Description |\n| --------- | -------- | ---------------------- |\n| `.jsonl` | JSONL | Streaming (default) |\n| `.json` | JSON | Complete structured |\n| `.md` | Markdown | Human-readable |\n| `.html` | HTML | Interactive web report |\n\n## Stability Analysis\n\nRun each test multiple times to measure consistency:\n\n```bash\nyao agent test -i tests/inputs.jsonl --runs 5 -o stability.json\n```\n\n| Pass Rate | Classification |\n| --------- | --------------- |\n| 100% | Stable |\n| 80-99% | Mostly Stable |\n| 50-79% | Unstable |\n| < 50% | Highly Unstable |\n\n## CI Integration\n\n```bash\n# Exit code: 0 = all passed, 1 = failures\nyao agent test -i tests/inputs.jsonl --fail-fast\n\n# Run with parallel execution\nyao agent test -i tests/inputs.jsonl --parallel 4\n```\n\n### GitHub Actions Example\n\n```yaml\n- name: Run Agent Tests\n run: |\n yao agent test -i assistants/expense/tests/inputs.jsonl \\\n -u ci-user -t ci-team \\\n --runs 3 \\\n -o report.json\n\n- name: Run Dynamic Tests\n run: |\n yao agent test -i assistants/expense/tests/dynamic.jsonl \\\n --simulator tests.simulator-agent \\\n -v\n\n- name: Run Script Tests\n run: |\n yao agent test -i scripts.expense.setup -v\n```\n\n## Format Rules Reference\n\n| Context | Format | Example |\n| ---------------------- | ------------------------ | ----------------------------------------- |\n| `-i agents:xxx` (CLI) | Colon prefix | `agents:tests.generator` |\n| `-i scripts:xxx` (CLI) | Colon prefix | `scripts:tests.gen.Generate` |\n| `-i scripts.xxx` (CLI) | Dot prefix (test mode) | `scripts.expense.setup` |\n| JSONL assertion `use` | Prefix required | `\"use\": \"agents:tests.validator\"` |\n| JSONL `simulator.use` | No prefix (agent only) | `\"use\": \"tests.simulator-agent\"` |\n| `--simulator` flag | No prefix (agent only) | `--simulator tests.simulator-agent` |\n| `t.assert.Agent()` | No prefix (method-bound) | `t.assert.Agent(resp, \"tests.validator\")` |\n| JSONL `before/after` | No prefix (in src/) | `\"before\": \"env_test.Before\"` |\n| `--before/--after` | No prefix (in src/) | `--before env_test.BeforeAll` |\n\n**Script input modes**:\n\n- `scripts.xxx` (dot) - Run script tests (`*_test.ts` functions)\n- `scripts:xxx` (colon) - Generate test cases from a script\n\n## Built-in Test Agents\n\nThe framework provides three specialized agents for testing:\n\n### Generator Agent (`tests.generator-agent`)\n\nGenerates test cases based on target agent description.\n\n**package.yao**:\n\n```json\n{\n \"name\": \"Test Case Generator\",\n \"connector\": \"gpt-4o\",\n \"description\": \"Generates test cases for agent testing\",\n \"options\": { \"temperature\": 0.7 },\n \"automated\": true\n}\n```\n\n**prompts.yml**:\n\n```yaml\n- role: system\n content: |\n You are a test case generator. Generate test cases based on the target agent.\n\n ## Input Format\n - `target_agent`: Agent info (id, description, tools)\n - `count`: Number of test cases (default: 5)\n - `focus`: Focus area (e.g., \"edge-cases\", \"happy-path\")\n\n ## Output Format\n JSON array of test cases:\n [\n {\n \"id\": \"test-id\",\n \"input\": \"User message\",\n \"assert\": [{\"type\": \"contains\", \"value\": \"expected\"}]\n }\n ]\n```\n\n**Usage**:\n\n```bash\nyao agent test -i \"agents:tests.generator-agent?count=10\" -n assistants.expense\n```\n\n### Validator Agent (`tests.validator-agent`)\n\nValidates agent responses for agent-driven assertions.\n\n**package.yao**:\n\n```json\n{\n \"name\": \"Response Validator\",\n \"connector\": \"gpt-4o\",\n \"description\": \"Validates responses against criteria\",\n \"options\": { \"temperature\": 0 },\n \"automated\": true\n}\n```\n\n**prompts.yml**:\n\n```yaml\n- role: system\n content: |\n You are a response validator. Evaluate whether the response meets the criteria.\n\n ## Input Format\n - `output`: The response to validate\n - `criteria`: The validation rules\n - `input`: Original input (optional)\n\n ## Output Format\n JSON object (no markdown):\n {\"passed\": true/false, \"reason\": \"explanation\"}\n\n ## Examples\n Input: {\"output\": \"Paris is the capital\", \"criteria\": \"factually accurate\"}\n Output: {\"passed\": true, \"reason\": \"Statement is correct\"}\n```\n\n**Usage in JSONL**:\n\n```jsonl\n{\n \"id\": \"T001\",\n \"input\": \"Hello\",\n \"assert\": {\n \"type\": \"agent\",\n \"use\": \"agents:tests.validator-agent\",\n \"value\": \"Response should be friendly\"\n }\n}\n```\n\n**Usage in script tests**:\n\n```typescript\nt.assert.Agent(response, \"tests.validator-agent\", {\n criteria: \"Response should be helpful\",\n});\n```\n\n### Simulator Agent (`tests.simulator-agent`)\n\nSimulates user behavior for dynamic mode testing.\n\n**package.yao**:\n\n```json\n{\n \"name\": \"User Simulator\",\n \"connector\": \"gpt-4o\",\n \"description\": \"Simulates user behavior for dynamic testing\",\n \"options\": { \"temperature\": 0.7 },\n \"automated\": true\n}\n```\n\n**prompts.yml**:\n\n```yaml\n- role: system\n content: |\n You are a user simulator. Generate realistic user messages based on persona and goal.\n\n ## Input Format\n - `persona`: User description (e.g., \"New employee\")\n - `goal`: What user wants to achieve\n - `conversation`: Previous messages\n - `turn_number`: Current turn\n - `max_turns`: Maximum turns\n\n ## Output Format\n JSON object:\n {\n \"message\": \"User response\",\n \"goal_achieved\": false,\n \"reasoning\": \"Strategy explanation\"\n }\n\n ## Guidelines\n 1. Stay in character\n 2. Work toward the goal\n 3. Be realistic (include natural variations)\n 4. Set goal_achieved: true when done\n```\n\n**Usage in JSONL**:\n\n```jsonl\n{\n \"id\": \"dynamic-test\",\n \"input\": \"I need help\",\n \"simulator\": {\n \"use\": \"tests.simulator-agent\",\n \"options\": {\n \"metadata\": {\n \"persona\": \"New employee\",\n \"goal\": \"Submit expense report\"\n }\n }\n },\n \"checkpoints\": [\n {\n \"id\": \"greeting\",\n \"assert\": {\n \"type\": \"regex\",\n \"value\": \"(?i)hello\"\n }\n }\n ],\n \"max_turns\": 10\n}\n```\n\n**Usage via CLI**:\n\n```bash\nyao agent test -i tests/dynamic.jsonl --simulator tests.simulator-agent\n```\n\n## Extract Command\n\nExtract test results from output JSONL file to individual Markdown or JSON files for human review:\n\n```bash\n# Extract to Markdown files (default)\nyao agent extract output-20260127104118.jsonl\n\n# Specify output directory\nyao agent extract output.jsonl -o ./reports/\n\n# Extract to JSON format\nyao agent extract output.jsonl --format json\n```\n\n### Extract Command Options\n\n| Flag | Description | Default |\n| ---------- | ---------------------------------------- | ---------- |\n| `-o` | Output directory | same as input |\n| `--format` | Output format: `markdown`, `json` | `markdown` |\n\n### Output Format (Markdown)\n\nEach test result is extracted to a separate file:\n\n```markdown\n# T001-销售分析师-月末周五\n\n**Status**: ✅ PASSED\n\n**Duration**: 16743ms\n\n## Input\n(Full input content in markdown code block)\n\n## Output\n(Agent's response content)\n```\n\nThis is useful for:\n- Human review of agent outputs\n- Comparing results across test runs\n- Documentation and reporting\n\n## Exit Codes\n\n| Code | Description |\n| ---- | --------------------------------------------------- |\n| 0 | All tests passed |\n| 1 | Tests failed, configuration error, or runtime error |\n" }, { "path": "agent/test/assert.go", "content": "package test\n\nimport (\n\t\"encoding/json\"\n\t\"fmt\"\n\t\"regexp\"\n\t\"strconv\"\n\t\"strings\"\n\n\tjsoniter \"github.com/json-iterator/go\"\n\t\"github.com/yaoapp/gou/process\"\n\tgoutext \"github.com/yaoapp/gou/text\"\n\t\"github.com/yaoapp/yao/agent/assistant\"\n\t\"github.com/yaoapp/yao/agent/context\"\n)\n\n// Asserter handles test assertions\ntype Asserter struct {\n\t// response holds the current response for tool-related assertions\n\tresponse *context.Response\n}\n\n// NewAsserter creates a new asserter\nfunc NewAsserter() *Asserter {\n\treturn &Asserter{}\n}\n\n// WithResponse sets the response for tool-related assertions\nfunc (a *Asserter) WithResponse(response *context.Response) *Asserter {\n\ta.response = response\n\treturn a\n}\n\n// Validate validates the output against the test case's assertions\n// Returns (passed, error message)\nfunc (a *Asserter) Validate(tc *Case, output interface{}) (bool, string) {\n\t// If assert is defined, use assertion rules\n\tif tc.Assert != nil {\n\t\treturn a.validateAssertions(tc, output)\n\t}\n\n\t// If expected is defined, use simple comparison\n\tif tc.Expected != nil {\n\t\tif validateOutput(output, tc.Expected) {\n\t\t\treturn true, \"\"\n\t\t}\n\t\treturn false, \"output does not match expected\"\n\t}\n\n\t// No assertions defined - pass if we got output without error\n\treturn true, \"\"\n}\n\n// ValidateWithDetails validates the output and returns detailed results\n// This is useful for agent assertions where we want to capture the validator's response\nfunc (a *Asserter) ValidateWithDetails(tc *Case, output interface{}) *AssertionResult {\n\tif tc.Assert == nil {\n\t\treturn &AssertionResult{Passed: true}\n\t}\n\n\tassertions := a.parseAssertions(tc.Assert)\n\tif len(assertions) == 0 {\n\t\treturn &AssertionResult{Passed: true}\n\t}\n\n\t// For single assertion, return its full result\n\tif len(assertions) == 1 {\n\t\treturn a.evaluateAssertion(assertions[0], output, tc.Input)\n\t}\n\n\t// For multiple assertions, combine results\n\tvar failures []string\n\tfor _, assertion := range assertions {\n\t\tresult := a.evaluateAssertion(assertion, output, tc.Input)\n\t\tif !result.Passed {\n\t\t\tmsg := result.Message\n\t\t\tif assertion.Message != \"\" {\n\t\t\t\tmsg = assertion.Message\n\t\t\t}\n\t\t\tfailures = append(failures, msg)\n\t\t}\n\t}\n\n\tif len(failures) > 0 {\n\t\treturn &AssertionResult{\n\t\t\tPassed: false,\n\t\t\tMessage: strings.Join(failures, \"; \"),\n\t\t}\n\t}\n\treturn &AssertionResult{Passed: true}\n}\n\n// validateAssertions validates output against assertion rules\nfunc (a *Asserter) validateAssertions(tc *Case, output interface{}) (bool, string) {\n\tassertions := a.parseAssertions(tc.Assert)\n\tif len(assertions) == 0 {\n\t\treturn true, \"\"\n\t}\n\n\tvar failures []string\n\tfor _, assertion := range assertions {\n\t\tresult := a.evaluateAssertion(assertion, output, tc.Input)\n\t\tif !result.Passed {\n\t\t\tmsg := result.Message\n\t\t\tif assertion.Message != \"\" {\n\t\t\t\tmsg = assertion.Message\n\t\t\t}\n\t\t\tfailures = append(failures, msg)\n\t\t}\n\t}\n\n\tif len(failures) > 0 {\n\t\treturn false, strings.Join(failures, \"; \")\n\t}\n\treturn true, \"\"\n}\n\n// parseAssertions parses the assert field into a list of assertions\nfunc (a *Asserter) parseAssertions(assert interface{}) []*Assertion {\n\tif assert == nil {\n\t\treturn nil\n\t}\n\n\tvar assertions []*Assertion\n\n\tswitch v := assert.(type) {\n\tcase map[string]interface{}:\n\t\t// Single assertion object\n\t\tassertion := a.mapToAssertion(v)\n\t\tif assertion != nil {\n\t\t\tassertions = append(assertions, assertion)\n\t\t}\n\n\tcase []interface{}:\n\t\t// Array of assertions\n\t\tfor _, item := range v {\n\t\t\tif m, ok := item.(map[string]interface{}); ok {\n\t\t\t\tassertion := a.mapToAssertion(m)\n\t\t\t\tif assertion != nil {\n\t\t\t\t\tassertions = append(assertions, assertion)\n\t\t\t\t}\n\t\t\t}\n\t\t}\n\n\tcase string:\n\t\t// Shorthand: just a type name (e.g., \"contains\")\n\t\tassertions = append(assertions, &Assertion{Type: v})\n\t}\n\n\treturn assertions\n}\n\n// mapToAssertion converts a map to an Assertion\nfunc (a *Asserter) mapToAssertion(m map[string]interface{}) *Assertion {\n\tassertion := &Assertion{}\n\n\tif t, ok := m[\"type\"].(string); ok {\n\t\tassertion.Type = t\n\t}\n\tif v, ok := m[\"value\"]; ok {\n\t\tassertion.Value = v\n\t}\n\tif p, ok := m[\"path\"].(string); ok {\n\t\tassertion.Path = p\n\t}\n\tif s, ok := m[\"script\"].(string); ok {\n\t\tassertion.Script = s\n\t}\n\tif u, ok := m[\"use\"].(string); ok {\n\t\tassertion.Use = u\n\t}\n\tif msg, ok := m[\"message\"].(string); ok {\n\t\tassertion.Message = msg\n\t}\n\tif n, ok := m[\"negate\"].(bool); ok {\n\t\tassertion.Negate = n\n\t}\n\n\t// Parse options for agent assertions\n\tif opts, ok := m[\"options\"].(map[string]interface{}); ok {\n\t\tassertion.Options = &AssertionOptions{}\n\t\tif c, ok := opts[\"connector\"].(string); ok {\n\t\t\tassertion.Options.Connector = c\n\t\t}\n\t\tif meta, ok := opts[\"metadata\"].(map[string]interface{}); ok {\n\t\t\tassertion.Options.Metadata = meta\n\t\t}\n\t}\n\n\treturn assertion\n}\n\n// evaluateAssertion evaluates a single assertion\nfunc (a *Asserter) evaluateAssertion(assertion *Assertion, output, input interface{}) *AssertionResult {\n\tresult := &AssertionResult{\n\t\tAssertion: assertion,\n\t\tExpected: assertion.Value,\n\t}\n\n\tswitch assertion.Type {\n\tcase \"equals\", \"\":\n\t\tresult = a.assertEquals(assertion, output)\n\tcase \"contains\":\n\t\tresult = a.assertContains(assertion, output)\n\tcase \"not_contains\":\n\t\tresult = a.assertNotContains(assertion, output)\n\tcase \"json_path\":\n\t\tresult = a.assertJSONPath(assertion, output)\n\tcase \"regex\":\n\t\tresult = a.assertRegex(assertion, output)\n\tcase \"type\":\n\t\tresult = a.assertType(assertion, output)\n\tcase \"script\":\n\t\tresult = a.assertScript(assertion, output, input)\n\tcase \"agent\":\n\t\tresult = a.assertAgent(assertion, output, input)\n\tcase \"tool_called\":\n\t\tresult = a.assertToolCalled(assertion)\n\tcase \"tool_result\":\n\t\tresult = a.assertToolResult(assertion)\n\tdefault:\n\t\tresult.Passed = false\n\t\tresult.Message = fmt.Sprintf(\"unknown assertion type: %s\", assertion.Type)\n\t}\n\n\t// Apply negate\n\tif assertion.Negate {\n\t\tresult.Passed = !result.Passed\n\t\tif result.Passed {\n\t\t\tresult.Message = \"negated assertion passed\"\n\t\t} else {\n\t\t\tresult.Message = \"negated: \" + result.Message\n\t\t}\n\t}\n\n\treturn result\n}\n\n// assertEquals checks for exact equality\nfunc (a *Asserter) assertEquals(assertion *Assertion, output interface{}) *AssertionResult {\n\tresult := &AssertionResult{\n\t\tAssertion: assertion,\n\t\tActual: output,\n\t\tExpected: assertion.Value,\n\t}\n\n\tif validateOutput(output, assertion.Value) {\n\t\tresult.Passed = true\n\t\tresult.Message = \"values are equal\"\n\t} else {\n\t\tresult.Passed = false\n\t\tresult.Message = fmt.Sprintf(\"expected %v, got %v\", assertion.Value, output)\n\t}\n\n\treturn result\n}\n\n// assertContains checks if output contains the expected value\nfunc (a *Asserter) assertContains(assertion *Assertion, output interface{}) *AssertionResult {\n\tresult := &AssertionResult{\n\t\tAssertion: assertion,\n\t\tActual: output,\n\t\tExpected: assertion.Value,\n\t}\n\n\toutputStr := a.toString(output)\n\texpectedStr := a.toString(assertion.Value)\n\n\tif strings.Contains(outputStr, expectedStr) {\n\t\tresult.Passed = true\n\t\tresult.Message = fmt.Sprintf(\"output contains '%s'\", expectedStr)\n\t} else {\n\t\tresult.Passed = false\n\t\tresult.Message = fmt.Sprintf(\"output does not contain '%s'\", expectedStr)\n\t}\n\n\treturn result\n}\n\n// assertNotContains checks if output does not contain the expected value\nfunc (a *Asserter) assertNotContains(assertion *Assertion, output interface{}) *AssertionResult {\n\tresult := a.assertContains(assertion, output)\n\tresult.Passed = !result.Passed\n\tif result.Passed {\n\t\tresult.Message = fmt.Sprintf(\"output does not contain '%s'\", a.toString(assertion.Value))\n\t} else {\n\t\tresult.Message = fmt.Sprintf(\"output should not contain '%s'\", a.toString(assertion.Value))\n\t}\n\treturn result\n}\n\n// assertJSONPath extracts a value using JSON path and compares\nfunc (a *Asserter) assertJSONPath(assertion *Assertion, output interface{}) *AssertionResult {\n\tresult := &AssertionResult{\n\t\tAssertion: assertion,\n\t\tExpected: assertion.Value,\n\t}\n\n\t// Convert output to JSON if needed\n\tvar jsonData interface{}\n\tswitch v := output.(type) {\n\tcase string:\n\t\t// Use gou/text to extract JSON (handles markdown, auto-repair, etc.)\n\t\textracted := goutext.ExtractJSON(v)\n\t\tif extracted != nil {\n\t\t\tjsonData = extracted\n\t\t} else {\n\t\t\tresult.Passed = false\n\t\t\tresult.Message = fmt.Sprintf(\"output is not valid JSON: %s\", v)\n\t\t\treturn result\n\t\t}\n\tcase map[string]interface{}, []interface{}:\n\t\tjsonData = v\n\tdefault:\n\t\tresult.Passed = false\n\t\tresult.Message = fmt.Sprintf(\"output is not a JSON object or array, got: %T = %v\", output, truncateOutput(output, 200))\n\t\treturn result\n\t}\n\n\t// Extract value using simple path (e.g., \"$.need_search\" or \"need_search\")\n\tpath := strings.TrimPrefix(assertion.Path, \"$.\")\n\tactual := a.extractPath(jsonData, path)\n\tresult.Actual = actual\n\n\t// Compare expected value with actual value\n\t// First try direct comparison (handles array-to-array comparison)\n\tif validateOutput(actual, assertion.Value) {\n\t\tresult.Passed = true\n\t\tresult.Message = fmt.Sprintf(\"path '%s' equals expected value\", assertion.Path)\n\t\treturn result\n\t}\n\n\t// If expected is an array and direct comparison failed, check if actual matches ANY element (IN semantics)\n\t// This is for cases like: expected: [\"a\", \"b\"], actual: \"a\" (actual is one of expected)\n\tif expectedArr, ok := assertion.Value.([]interface{}); ok {\n\t\t// Only apply IN semantics if actual is NOT an array (otherwise it was already compared above)\n\t\tif _, actualIsArr := actual.([]interface{}); !actualIsArr {\n\t\t\tfor _, expectedItem := range expectedArr {\n\t\t\t\tif validateOutput(actual, expectedItem) {\n\t\t\t\t\tresult.Passed = true\n\t\t\t\t\tresult.Message = fmt.Sprintf(\"path '%s' equals one of expected values\", assertion.Path)\n\t\t\t\t\treturn result\n\t\t\t\t}\n\t\t\t}\n\t\t}\n\t\tresult.Passed = false\n\t\tresult.Message = fmt.Sprintf(\"path '%s': expected %v, got %v\", assertion.Path, assertion.Value, actual)\n\t} else {\n\t\t// Direct comparison already failed above\n\t\tresult.Passed = false\n\t\tresult.Message = fmt.Sprintf(\"path '%s': expected %v, got %v\", assertion.Path, assertion.Value, actual)\n\t}\n\n\treturn result\n}\n\n// truncateOutput truncates output for error messages\nfunc truncateOutput(output interface{}, maxLen int) string {\n\tvar s string\n\tswitch v := output.(type) {\n\tcase string:\n\t\ts = v\n\tcase nil:\n\t\treturn \"\"\n\tdefault:\n\t\tbytes, err := jsoniter.Marshal(v)\n\t\tif err != nil {\n\t\t\ts = fmt.Sprintf(\"%v\", v)\n\t\t} else {\n\t\t\ts = string(bytes)\n\t\t}\n\t}\n\n\tif len(s) > maxLen {\n\t\treturn s[:maxLen] + \"...\"\n\t}\n\treturn s\n}\n\n// extractPath extracts a value from JSON using dot-notation path with array index support\n// Supports: \"field\", \"field.nested\", \"field[0]\", \"field[0].nested\", \"field.nested[0].value\"\nfunc (a *Asserter) extractPath(data interface{}, path string) interface{} {\n\tcurrent := data\n\n\t// Parse path into segments, handling both dots and array indices\n\t// e.g., \"wheres[0].like\" -> [\"wheres\", \"[0]\", \"like\"]\n\tsegments := parsePathSegments(path)\n\n\tfor _, segment := range segments {\n\t\tif segment == \"\" {\n\t\t\tcontinue\n\t\t}\n\n\t\t// Check if this is an array index like \"[0]\"\n\t\tif strings.HasPrefix(segment, \"[\") && strings.HasSuffix(segment, \"]\") {\n\t\t\tindexStr := segment[1 : len(segment)-1]\n\t\t\tindex, err := strconv.Atoi(indexStr)\n\t\t\tif err != nil {\n\t\t\t\treturn nil\n\t\t\t}\n\n\t\t\tarr, ok := current.([]interface{})\n\t\t\tif !ok {\n\t\t\t\treturn nil\n\t\t\t}\n\n\t\t\tif index < 0 || index >= len(arr) {\n\t\t\t\treturn nil\n\t\t\t}\n\t\t\tcurrent = arr[index]\n\t\t} else {\n\t\t\t// Regular field access\n\t\t\tswitch v := current.(type) {\n\t\t\tcase map[string]interface{}:\n\t\t\t\tcurrent = v[segment]\n\t\t\tdefault:\n\t\t\t\treturn nil\n\t\t\t}\n\t\t}\n\t}\n\n\treturn current\n}\n\n// parsePathSegments splits a path like \"wheres[0].like\" into [\"wheres\", \"[0]\", \"like\"]\nfunc parsePathSegments(path string) []string {\n\tvar segments []string\n\tvar current strings.Builder\n\n\tfor i := 0; i < len(path); i++ {\n\t\tch := path[i]\n\t\tswitch ch {\n\t\tcase '.':\n\t\t\tif current.Len() > 0 {\n\t\t\t\tsegments = append(segments, current.String())\n\t\t\t\tcurrent.Reset()\n\t\t\t}\n\t\tcase '[':\n\t\t\tif current.Len() > 0 {\n\t\t\t\tsegments = append(segments, current.String())\n\t\t\t\tcurrent.Reset()\n\t\t\t}\n\t\t\t// Find the closing bracket\n\t\t\tj := i + 1\n\t\t\tfor j < len(path) && path[j] != ']' {\n\t\t\t\tj++\n\t\t\t}\n\t\t\tif j < len(path) {\n\t\t\t\tsegments = append(segments, path[i:j+1]) // Include \"[\" and \"]\"\n\t\t\t\ti = j\n\t\t\t}\n\t\tdefault:\n\t\t\tcurrent.WriteByte(ch)\n\t\t}\n\t}\n\n\tif current.Len() > 0 {\n\t\tsegments = append(segments, current.String())\n\t}\n\n\treturn segments\n}\n\n// assertRegex checks if output matches a regex pattern\nfunc (a *Asserter) assertRegex(assertion *Assertion, output interface{}) *AssertionResult {\n\tresult := &AssertionResult{\n\t\tAssertion: assertion,\n\t\tActual: output,\n\t\tExpected: assertion.Value,\n\t}\n\n\tpattern, ok := assertion.Value.(string)\n\tif !ok {\n\t\tresult.Passed = false\n\t\tresult.Message = \"regex pattern must be a string\"\n\t\treturn result\n\t}\n\n\tre, err := regexp.Compile(pattern)\n\tif err != nil {\n\t\tresult.Passed = false\n\t\tresult.Message = fmt.Sprintf(\"invalid regex pattern: %s\", err.Error())\n\t\treturn result\n\t}\n\n\toutputStr := a.toString(output)\n\tif re.MatchString(outputStr) {\n\t\tresult.Passed = true\n\t\tresult.Message = fmt.Sprintf(\"output matches pattern '%s'\", pattern)\n\t} else {\n\t\tresult.Passed = false\n\t\tresult.Message = fmt.Sprintf(\"output does not match pattern '%s'\", pattern)\n\t}\n\n\treturn result\n}\n\n// assertType checks the type of the output\nfunc (a *Asserter) assertType(assertion *Assertion, output interface{}) *AssertionResult {\n\tresult := &AssertionResult{\n\t\tAssertion: assertion,\n\t\tActual: output,\n\t\tExpected: assertion.Value,\n\t}\n\n\texpectedType, ok := assertion.Value.(string)\n\tif !ok {\n\t\tresult.Passed = false\n\t\tresult.Message = \"type assertion value must be a string\"\n\t\treturn result\n\t}\n\n\tactualType := a.getType(output)\n\tresult.Actual = actualType\n\n\tif actualType == expectedType {\n\t\tresult.Passed = true\n\t\tresult.Message = fmt.Sprintf(\"output is of type '%s'\", expectedType)\n\t} else {\n\t\tresult.Passed = false\n\t\tresult.Message = fmt.Sprintf(\"expected type '%s', got '%s'\", expectedType, actualType)\n\t}\n\n\treturn result\n}\n\n// getType returns the type name of a value\nfunc (a *Asserter) getType(v interface{}) string {\n\tif v == nil {\n\t\treturn \"null\"\n\t}\n\n\tswitch v.(type) {\n\tcase string:\n\t\treturn \"string\"\n\tcase float64, float32, int, int64, int32:\n\t\treturn \"number\"\n\tcase bool:\n\t\treturn \"boolean\"\n\tcase []interface{}:\n\t\treturn \"array\"\n\tcase map[string]interface{}:\n\t\treturn \"object\"\n\tdefault:\n\t\treturn fmt.Sprintf(\"%T\", v)\n\t}\n}\n\n// assertAgent uses an agent to validate the output\nfunc (a *Asserter) assertAgent(assertion *Assertion, output, input interface{}) *AssertionResult {\n\tresult := &AssertionResult{\n\t\tAssertion: assertion,\n\t\tActual: output,\n\t}\n\n\t// Parse use field: \"agents:tests.validator-agent\"\n\tif !strings.HasPrefix(assertion.Use, \"agents:\") {\n\t\tresult.Passed = false\n\t\tresult.Message = \"agent assertion requires 'use' field with 'agents:' prefix\"\n\t\treturn result\n\t}\n\n\tagentID := strings.TrimPrefix(assertion.Use, \"agents:\")\n\n\t// Get assistant\n\tast, err := assistant.Get(agentID)\n\tif err != nil {\n\t\tresult.Passed = false\n\t\tresult.Message = fmt.Sprintf(\"failed to get validator agent: %s\", err.Error())\n\t\treturn result\n\t}\n\n\t// Build validation request\n\tvalidationInput := map[string]interface{}{\n\t\t\"output\": output,\n\t\t\"input\": input,\n\t}\n\n\t// Add criteria from Value field\n\tif assertion.Value != nil {\n\t\tvalidationInput[\"criteria\"] = assertion.Value\n\t}\n\n\t// Add metadata from options\n\tif assertion.Options != nil && assertion.Options.Metadata != nil {\n\t\tfor k, v := range assertion.Options.Metadata {\n\t\t\tvalidationInput[k] = v\n\t\t}\n\t}\n\n\t// Build context options - skip history and trace for validator\n\topts := &context.Options{\n\t\tSkip: &context.Skip{\n\t\t\tHistory: true,\n\t\t\tTrace: true,\n\t\t\tOutput: true,\n\t\t},\n\t\tMetadata: map[string]interface{}{\n\t\t\t\"test_mode\": \"validator\",\n\t\t},\n\t}\n\tif assertion.Options != nil && assertion.Options.Connector != \"\" {\n\t\topts.Connector = assertion.Options.Connector\n\t}\n\n\t// Create context and call agent\n\tenv := NewEnvironment(\"\", \"\")\n\tctx := NewTestContext(\"validator\", agentID, env)\n\tdefer ctx.Release()\n\n\t// Convert validation input to JSON string for the message\n\tinputJSON, err := json.Marshal(validationInput)\n\tif err != nil {\n\t\tresult.Passed = false\n\t\tresult.Message = fmt.Sprintf(\"failed to marshal validation input: %s\", err.Error())\n\t\treturn result\n\t}\n\n\tmessages := []context.Message{{\n\t\tRole: context.RoleUser,\n\t\tContent: string(inputJSON),\n\t}}\n\n\tresponse, err := ast.Stream(ctx, messages, opts)\n\tif err != nil {\n\t\tresult.Passed = false\n\t\tresult.Message = fmt.Sprintf(\"validator agent error: %s\", err.Error())\n\t\treturn result\n\t}\n\n\t// Parse response\n\treturn a.parseValidatorResponse(response, result)\n}\n\n// parseValidatorResponse parses the validator agent's response\nfunc (a *Asserter) parseValidatorResponse(response *context.Response, result *AssertionResult) *AssertionResult {\n\toutput := extractValidatorOutput(response)\n\n\t// Expected format: { \"passed\": bool, \"reason\": string, \"score\": float, \"suggestions\": [] }\n\tif outputMap, ok := output.(map[string]interface{}); ok {\n\t\tif passed, ok := outputMap[\"passed\"].(bool); ok {\n\t\t\tresult.Passed = passed\n\t\t} else {\n\t\t\tresult.Passed = false\n\t\t\tresult.Message = \"validator response missing 'passed' field\"\n\t\t\treturn result\n\t\t}\n\t\tif reason, ok := outputMap[\"reason\"].(string); ok {\n\t\t\tresult.Message = reason\n\t\t}\n\t\t// Store score and suggestions in expected field for reference\n\t\tresult.Expected = outputMap\n\t} else {\n\t\tresult.Passed = false\n\t\tresult.Message = \"validator agent returned invalid response format\"\n\t}\n\n\treturn result\n}\n\n// extractValidatorOutput extracts the output from a validator response\nfunc extractValidatorOutput(response *context.Response) interface{} {\n\tif response == nil || response.Completion == nil {\n\t\treturn nil\n\t}\n\n\t// Get content from completion\n\tcontent := response.Completion.Content\n\tif content == nil {\n\t\treturn nil\n\t}\n\n\t// Try to get text content\n\tvar text string\n\tswitch v := content.(type) {\n\tcase string:\n\t\ttext = v\n\tdefault:\n\t\t// Try to marshal and use as-is\n\t\tdata, err := json.Marshal(content)\n\t\tif err != nil {\n\t\t\treturn nil\n\t\t}\n\t\ttext = string(data)\n\t}\n\n\tif text == \"\" {\n\t\treturn nil\n\t}\n\n\t// Use gou/text to extract JSON (handles markdown code blocks, auto-repair, etc.)\n\tresult := goutext.ExtractJSON(text)\n\tif result != nil {\n\t\treturn result\n\t}\n\n\t// Return raw text if extraction fails\n\treturn text\n}\n\n// assertScript runs a custom assertion script\nfunc (a *Asserter) assertScript(assertion *Assertion, output, input interface{}) *AssertionResult {\n\tresult := &AssertionResult{\n\t\tAssertion: assertion,\n\t\tActual: output,\n\t}\n\n\tif assertion.Script == \"\" {\n\t\tresult.Passed = false\n\t\tresult.Message = \"script assertion requires a script name\"\n\t\treturn result\n\t}\n\n\t// Build script arguments\n\targs := []interface{}{\n\t\toutput,\n\t\tinput,\n\t\tassertion.Value,\n\t}\n\n\t// Run the script as a process\n\tp, err := process.Of(assertion.Script, args...)\n\tif err != nil {\n\t\tresult.Passed = false\n\t\tresult.Message = fmt.Sprintf(\"failed to create process: %s\", err.Error())\n\t\treturn result\n\t}\n\n\tres, err := p.Exec()\n\tif err != nil {\n\t\tresult.Passed = false\n\t\tresult.Message = fmt.Sprintf(\"script execution failed: %s\", err.Error())\n\t\treturn result\n\t}\n\n\t// Parse script result\n\t// Expected format: { \"pass\": bool, \"message\": string }\n\tswitch v := res.(type) {\n\tcase bool:\n\t\tresult.Passed = v\n\t\tif v {\n\t\t\tresult.Message = \"script assertion passed\"\n\t\t} else {\n\t\t\tresult.Message = \"script assertion failed\"\n\t\t}\n\n\tcase map[string]interface{}:\n\t\tif pass, ok := v[\"pass\"].(bool); ok {\n\t\t\tresult.Passed = pass\n\t\t}\n\t\tif msg, ok := v[\"message\"].(string); ok {\n\t\t\tresult.Message = msg\n\t\t}\n\n\tdefault:\n\t\tresult.Passed = false\n\t\tresult.Message = fmt.Sprintf(\"script returned unexpected type: %T\", res)\n\t}\n\n\treturn result\n}\n\n// assertToolCalled checks if a specific tool was called\n// value can be:\n// - string: exact tool name to match\n// - []string: any of the tool names\n// - map with \"name\" and optional \"arguments\" for more specific matching\nfunc (a *Asserter) assertToolCalled(assertion *Assertion) *AssertionResult {\n\tresult := &AssertionResult{\n\t\tAssertion: assertion,\n\t\tExpected: assertion.Value,\n\t}\n\n\tif a.response == nil {\n\t\tresult.Passed = false\n\t\tresult.Message = \"no response available for tool_called assertion\"\n\t\treturn result\n\t}\n\n\tif len(a.response.Tools) == 0 {\n\t\tresult.Passed = false\n\t\tresult.Message = \"no tools were called\"\n\t\treturn result\n\t}\n\n\t// Get tool names that were called\n\tcalledTools := make([]string, 0, len(a.response.Tools))\n\tfor _, tool := range a.response.Tools {\n\t\tcalledTools = append(calledTools, tool.Tool)\n\t}\n\tresult.Actual = calledTools\n\n\tswitch v := assertion.Value.(type) {\n\tcase string:\n\t\t// Simple case: check if tool name matches (supports prefix matching)\n\t\tfor _, tool := range a.response.Tools {\n\t\t\tif matchToolName(tool.Tool, v) {\n\t\t\t\tresult.Passed = true\n\t\t\t\tresult.Message = fmt.Sprintf(\"tool '%s' was called\", v)\n\t\t\t\treturn result\n\t\t\t}\n\t\t}\n\t\tresult.Passed = false\n\t\tresult.Message = fmt.Sprintf(\"tool '%s' was not called, called: %v\", v, calledTools)\n\n\tcase []interface{}:\n\t\t// Check if any of the specified tools were called\n\t\tfor _, expected := range v {\n\t\t\tif expectedStr, ok := expected.(string); ok {\n\t\t\t\tfor _, tool := range a.response.Tools {\n\t\t\t\t\tif matchToolName(tool.Tool, expectedStr) {\n\t\t\t\t\t\tresult.Passed = true\n\t\t\t\t\t\tresult.Message = fmt.Sprintf(\"tool '%s' was called\", expectedStr)\n\t\t\t\t\t\treturn result\n\t\t\t\t\t}\n\t\t\t\t}\n\t\t\t}\n\t\t}\n\t\tresult.Passed = false\n\t\tresult.Message = fmt.Sprintf(\"none of the expected tools were called, called: %v\", calledTools)\n\n\tcase map[string]interface{}:\n\t\t// Advanced case: match name and optionally arguments\n\t\texpectedName, _ := v[\"name\"].(string)\n\t\texpectedArgs := v[\"arguments\"]\n\n\t\tfor _, tool := range a.response.Tools {\n\t\t\tif matchToolName(tool.Tool, expectedName) {\n\t\t\t\t// If arguments specified, check them too\n\t\t\t\tif expectedArgs != nil {\n\t\t\t\t\tif matchArguments(tool.Arguments, expectedArgs) {\n\t\t\t\t\t\tresult.Passed = true\n\t\t\t\t\t\tresult.Message = fmt.Sprintf(\"tool '%s' was called with matching arguments\", expectedName)\n\t\t\t\t\t\treturn result\n\t\t\t\t\t}\n\t\t\t\t} else {\n\t\t\t\t\tresult.Passed = true\n\t\t\t\t\tresult.Message = fmt.Sprintf(\"tool '%s' was called\", expectedName)\n\t\t\t\t\treturn result\n\t\t\t\t}\n\t\t\t}\n\t\t}\n\t\tresult.Passed = false\n\t\tif expectedArgs != nil {\n\t\t\tresult.Message = fmt.Sprintf(\"tool '%s' was not called with expected arguments\", expectedName)\n\t\t} else {\n\t\t\tresult.Message = fmt.Sprintf(\"tool '%s' was not called, called: %v\", expectedName, calledTools)\n\t\t}\n\n\tdefault:\n\t\tresult.Passed = false\n\t\tresult.Message = fmt.Sprintf(\"invalid tool_called value type: %T\", assertion.Value)\n\t}\n\n\treturn result\n}\n\n// assertToolResult checks the result of a tool call\n// value should be a map with \"tool\" (name) and \"result\" (expected result pattern)\nfunc (a *Asserter) assertToolResult(assertion *Assertion) *AssertionResult {\n\tresult := &AssertionResult{\n\t\tAssertion: assertion,\n\t\tExpected: assertion.Value,\n\t}\n\n\tif a.response == nil {\n\t\tresult.Passed = false\n\t\tresult.Message = \"no response available for tool_result assertion\"\n\t\treturn result\n\t}\n\n\tif len(a.response.Tools) == 0 {\n\t\tresult.Passed = false\n\t\tresult.Message = \"no tools were called\"\n\t\treturn result\n\t}\n\n\tspec, ok := assertion.Value.(map[string]interface{})\n\tif !ok {\n\t\tresult.Passed = false\n\t\tresult.Message = \"tool_result assertion requires a map with 'tool' and 'result' fields\"\n\t\treturn result\n\t}\n\n\ttoolName, _ := spec[\"tool\"].(string)\n\texpectedResult := spec[\"result\"]\n\n\tif toolName == \"\" {\n\t\tresult.Passed = false\n\t\tresult.Message = \"tool_result assertion requires 'tool' field\"\n\t\treturn result\n\t}\n\n\t// Find the tool call\n\tfor _, tool := range a.response.Tools {\n\t\tif matchToolName(tool.Tool, toolName) {\n\t\t\tresult.Actual = tool.Result\n\n\t\t\t// Check if there was an error\n\t\t\tif tool.Error != \"\" {\n\t\t\t\tresult.Passed = false\n\t\t\t\tresult.Message = fmt.Sprintf(\"tool '%s' returned error: %s\", toolName, tool.Error)\n\t\t\t\treturn result\n\t\t\t}\n\n\t\t\t// If no expected result specified, just check success (no error)\n\t\t\tif expectedResult == nil {\n\t\t\t\tresult.Passed = true\n\t\t\t\tresult.Message = fmt.Sprintf(\"tool '%s' executed successfully\", toolName)\n\t\t\t\treturn result\n\t\t\t}\n\n\t\t\t// Match result\n\t\t\tif matchResult(tool.Result, expectedResult) {\n\t\t\t\tresult.Passed = true\n\t\t\t\tresult.Message = fmt.Sprintf(\"tool '%s' result matches expected\", toolName)\n\t\t\t\treturn result\n\t\t\t}\n\n\t\t\tresult.Passed = false\n\t\t\tresult.Message = fmt.Sprintf(\"tool '%s' result does not match expected\", toolName)\n\t\t\treturn result\n\t\t}\n\t}\n\n\tresult.Passed = false\n\tresult.Message = fmt.Sprintf(\"tool '%s' was not called\", toolName)\n\treturn result\n}\n\n// matchToolName checks if a tool name matches the expected pattern\n// Supports exact match and suffix match (e.g., \"setup\" matches \"agents_expense_tools__setup\")\nfunc matchToolName(actual, expected string) bool {\n\tif actual == expected {\n\t\treturn true\n\t}\n\t// Support suffix matching (tool name without namespace prefix)\n\tif strings.HasSuffix(actual, \"__\"+expected) || strings.HasSuffix(actual, \".\"+expected) {\n\t\treturn true\n\t}\n\t// Support contains matching for partial names\n\tif strings.Contains(actual, expected) {\n\t\treturn true\n\t}\n\treturn false\n}\n\n// matchArguments checks if tool arguments match expected pattern\nfunc matchArguments(actual, expected interface{}) bool {\n\texpectedMap, ok := expected.(map[string]interface{})\n\tif !ok {\n\t\treturn false\n\t}\n\n\tactualMap, ok := actual.(map[string]interface{})\n\tif !ok {\n\t\t// Try parsing as JSON string\n\t\tif actualStr, ok := actual.(string); ok {\n\t\t\tvar parsed map[string]interface{}\n\t\t\tif err := jsoniter.UnmarshalFromString(actualStr, &parsed); err == nil {\n\t\t\t\tactualMap = parsed\n\t\t\t} else {\n\t\t\t\treturn false\n\t\t\t}\n\t\t} else {\n\t\t\treturn false\n\t\t}\n\t}\n\n\t// Check that all expected keys exist and match\n\tfor key, expectedVal := range expectedMap {\n\t\tactualVal, exists := actualMap[key]\n\t\tif !exists {\n\t\t\treturn false\n\t\t}\n\t\tif !validateOutput(actualVal, expectedVal) {\n\t\t\treturn false\n\t\t}\n\t}\n\treturn true\n}\n\n// matchResult checks if tool result matches expected pattern\nfunc matchResult(actual, expected interface{}) bool {\n\tswitch exp := expected.(type) {\n\tcase map[string]interface{}:\n\t\tactualMap, ok := actual.(map[string]interface{})\n\t\tif !ok {\n\t\t\treturn false\n\t\t}\n\t\t// Check that all expected keys exist and match\n\t\tfor key, expectedVal := range exp {\n\t\t\tactualVal, exists := actualMap[key]\n\t\t\tif !exists {\n\t\t\t\treturn false\n\t\t\t}\n\t\t\tif !matchResult(actualVal, expectedVal) {\n\t\t\t\treturn false\n\t\t\t}\n\t\t}\n\t\treturn true\n\n\tcase string:\n\t\t// Support regex pattern matching for strings\n\t\tif strings.HasPrefix(exp, \"regex:\") {\n\t\t\tpattern := strings.TrimPrefix(exp, \"regex:\")\n\t\t\tre, err := regexp.Compile(pattern)\n\t\t\tif err != nil {\n\t\t\t\treturn false\n\t\t\t}\n\t\t\tactualStr := fmt.Sprintf(\"%v\", actual)\n\t\t\treturn re.MatchString(actualStr)\n\t\t}\n\t\treturn fmt.Sprintf(\"%v\", actual) == exp\n\n\tcase bool:\n\t\tactualBool, ok := actual.(bool)\n\t\treturn ok && actualBool == exp\n\n\tdefault:\n\t\treturn validateOutput(actual, expected)\n\t}\n}\n\n// toString converts a value to string for comparison\nfunc (a *Asserter) toString(v interface{}) string {\n\tif v == nil {\n\t\treturn \"\"\n\t}\n\n\tswitch val := v.(type) {\n\tcase string:\n\t\treturn val\n\tcase []byte:\n\t\treturn string(val)\n\tdefault:\n\t\tb, err := json.Marshal(v)\n\t\tif err != nil {\n\t\t\treturn fmt.Sprintf(\"%v\", v)\n\t\t}\n\t\treturn string(b)\n\t}\n}\n" }, { "path": "agent/test/assert_agent_test.go", "content": "package test_test\n\nimport (\n\t\"testing\"\n\n\t\"github.com/stretchr/testify/assert\"\n\tv8 \"github.com/yaoapp/gou/runtime/v8\"\n\t\"github.com/yaoapp/yao/agent\"\n\tagenttest \"github.com/yaoapp/yao/agent/test\"\n\t\"github.com/yaoapp/yao/config\"\n\t\"github.com/yaoapp/yao/test\"\n\t\"rogchap.com/v8go\"\n)\n\nfunc TestAsserter_AgentAssertion(t *testing.T) {\n\ttest.Prepare(t, config.Conf)\n\tdefer test.Clean()\n\n\t// Load agent (includes assistants)\n\terr := agent.Load(config.Conf)\n\tif err != nil {\n\t\tt.Fatalf(\"Failed to load agent: %v\", err)\n\t}\n\n\tasserter := agenttest.NewAsserter()\n\n\ttests := []struct {\n\t\tname string\n\t\ttc *agenttest.Case\n\t\toutput interface{}\n\t\texpected bool\n\t\tskipMsg string\n\t}{\n\t\t{\n\t\t\tname: \"agent assertion - pass\",\n\t\t\ttc: &agenttest.Case{\n\t\t\t\tAssert: map[string]interface{}{\n\t\t\t\t\t\"type\": \"agent\",\n\t\t\t\t\t\"use\": \"agents:tests.validator-agent\",\n\t\t\t\t\t\"value\": \"Response should be a greeting\",\n\t\t\t\t},\n\t\t\t},\n\t\t\toutput: \"Hello! How can I help you today?\",\n\t\t\texpected: true,\n\t\t},\n\t\t{\n\t\t\tname: \"agent assertion - fail\",\n\t\t\ttc: &agenttest.Case{\n\t\t\t\tAssert: map[string]interface{}{\n\t\t\t\t\t\"type\": \"agent\",\n\t\t\t\t\t\"use\": \"agents:tests.validator-agent\",\n\t\t\t\t\t\"value\": \"Response should provide a detailed technical answer\",\n\t\t\t\t},\n\t\t\t},\n\t\t\toutput: \"I don't know.\",\n\t\t\texpected: false,\n\t\t},\n\t\t{\n\t\t\tname: \"agent assertion - missing prefix\",\n\t\t\ttc: &agenttest.Case{\n\t\t\t\tAssert: map[string]interface{}{\n\t\t\t\t\t\"type\": \"agent\",\n\t\t\t\t\t\"use\": \"tests.validator-agent\", // Missing agents: prefix\n\t\t\t\t\t\"value\": \"Should pass\",\n\t\t\t\t},\n\t\t\t},\n\t\t\toutput: \"Hello\",\n\t\t\texpected: false, // Should fail due to missing prefix\n\t\t},\n\t\t{\n\t\t\tname: \"agent assertion - with metadata\",\n\t\t\ttc: &agenttest.Case{\n\t\t\t\tAssert: map[string]interface{}{\n\t\t\t\t\t\"type\": \"agent\",\n\t\t\t\t\t\"use\": \"agents:tests.validator-agent\",\n\t\t\t\t\t\"value\": \"Response is helpful\",\n\t\t\t\t\t\"options\": map[string]interface{}{\n\t\t\t\t\t\t\"metadata\": map[string]interface{}{\n\t\t\t\t\t\t\t\"context\": \"customer support\",\n\t\t\t\t\t\t},\n\t\t\t\t\t},\n\t\t\t\t},\n\t\t\t},\n\t\t\toutput: \"I'd be happy to help you with your order. Let me look that up for you.\",\n\t\t\texpected: true,\n\t\t},\n\t}\n\n\tfor _, tt := range tests {\n\t\tt.Run(tt.name, func(t *testing.T) {\n\t\t\tif tt.skipMsg != \"\" {\n\t\t\t\tt.Skip(tt.skipMsg)\n\t\t\t}\n\n\t\t\tpassed, errMsg := asserter.Validate(tt.tc, tt.output)\n\t\t\tif passed != tt.expected {\n\t\t\t\tt.Errorf(\"Expected passed=%v, got passed=%v, error: %s\", tt.expected, passed, errMsg)\n\t\t\t}\n\t\t})\n\t}\n}\n\nfunc TestAsserter_AgentAssertion_InvalidAgent(t *testing.T) {\n\ttest.Prepare(t, config.Conf)\n\tdefer test.Clean()\n\n\t// Load agent (includes assistants)\n\terr := agent.Load(config.Conf)\n\tif err != nil {\n\t\tt.Fatalf(\"Failed to load agent: %v\", err)\n\t}\n\n\tasserter := agenttest.NewAsserter()\n\n\ttc := &agenttest.Case{\n\t\tAssert: map[string]interface{}{\n\t\t\t\"type\": \"agent\",\n\t\t\t\"use\": \"agents:nonexistent.agent\",\n\t\t\t\"value\": \"Should fail\",\n\t\t},\n\t}\n\n\tpassed, errMsg := asserter.Validate(tc, \"Hello\")\n\tassert.False(t, passed, \"Should fail for nonexistent agent\")\n\tassert.Contains(t, errMsg, \"failed to get validator agent\", \"Error should mention agent loading failure\")\n}\n\nfunc TestAsserter_MapToAssertion_WithUseAndOptions(t *testing.T) {\n\ttest.Prepare(t, config.Conf)\n\tdefer test.Clean()\n\n\t// Load agent (includes assistants)\n\terr := agent.Load(config.Conf)\n\tif err != nil {\n\t\tt.Fatalf(\"Failed to load agent: %v\", err)\n\t}\n\n\tasserter := agenttest.NewAsserter()\n\n\t// Test that mapToAssertion correctly parses use and options fields\n\ttc := &agenttest.Case{\n\t\tAssert: map[string]interface{}{\n\t\t\t\"type\": \"agent\",\n\t\t\t\"use\": \"agents:tests.validator-agent\",\n\t\t\t\"value\": \"criteria here\",\n\t\t\t\"options\": map[string]interface{}{\n\t\t\t\t\"connector\": \"gpt-4o\",\n\t\t\t\t\"metadata\": map[string]interface{}{\n\t\t\t\t\t\"key\": \"value\",\n\t\t\t\t},\n\t\t\t},\n\t\t},\n\t}\n\n\t// Validate triggers parseAssertions internally\n\t// We just verify it doesn't panic and processes correctly\n\t_, _ = asserter.Validate(tc, \"test output\")\n\t// If we get here without panic, the parsing worked\n}\n\n// TestTestingT_AssertAgent tests the JSAPI t.assert.Agent() method\nfunc TestTestingT_AssertAgent(t *testing.T) {\n\ttest.Prepare(t, config.Conf)\n\tdefer test.Clean()\n\n\t// Load agent (includes assistants)\n\terr := agent.Load(config.Conf)\n\tif err != nil {\n\t\tt.Fatalf(\"Failed to load agent: %v\", err)\n\t}\n\n\ttests := []struct {\n\t\tname string\n\t\tscript string\n\t\tshouldFail bool\n\t}{\n\t\t{\n\t\t\tname: \"JSAPI agent assertion - pass\",\n\t\t\tscript: `\n\t\t\t\tfunction test(t) {\n\t\t\t\t\tvar response = \"Hello! How can I help you today?\";\n\t\t\t\t\tt.assert.Agent(response, \"tests.validator-agent\", {\n\t\t\t\t\t\tcriteria: \"Response should be a friendly greeting\"\n\t\t\t\t\t});\n\t\t\t\t}\n\t\t\t\ttest(__test_t);\n\t\t\t`,\n\t\t\tshouldFail: false,\n\t\t},\n\t\t{\n\t\t\tname: \"JSAPI agent assertion - JSON response\",\n\t\t\tscript: `\n\t\t\t\tfunction test(t) {\n\t\t\t\t\tvar response = {\n\t\t\t\t\t\tstatus: \"success\",\n\t\t\t\t\t\tdata: { user: \"john\", email: \"john@example.com\" },\n\t\t\t\t\t\tmessage: \"User created successfully\"\n\t\t\t\t\t};\n\t\t\t\t\tt.assert.Agent(response, \"tests.validator-agent\", {\n\t\t\t\t\t\tcriteria: \"Response should be a successful API response with user data\"\n\t\t\t\t\t});\n\t\t\t\t}\n\t\t\t\ttest(__test_t);\n\t\t\t`,\n\t\t\tshouldFail: false,\n\t\t},\n\t\t{\n\t\t\tname: \"JSAPI agent assertion - with metadata\",\n\t\t\tscript: `\n\t\t\t\tfunction test(t) {\n\t\t\t\t\tvar response = \"I'd be happy to help you with your order.\";\n\t\t\t\t\tt.assert.Agent(response, \"tests.validator-agent\", {\n\t\t\t\t\t\tcriteria: \"Response is helpful and professional\",\n\t\t\t\t\t\tmetadata: { context: \"customer support\" }\n\t\t\t\t\t});\n\t\t\t\t}\n\t\t\t\ttest(__test_t);\n\t\t\t`,\n\t\t\tshouldFail: false,\n\t\t},\n\t\t{\n\t\t\tname: \"JSAPI agent assertion - fail case\",\n\t\t\tscript: `\n\t\t\t\tfunction test(t) {\n\t\t\t\t\tvar response = \"I don't know.\";\n\t\t\t\t\tt.assert.Agent(response, \"tests.validator-agent\", {\n\t\t\t\t\t\tcriteria: \"Response should provide a detailed technical explanation\"\n\t\t\t\t\t});\n\t\t\t\t}\n\t\t\t\ttest(__test_t);\n\t\t\t`,\n\t\t\tshouldFail: true,\n\t\t},\n\t}\n\n\tfor _, tt := range tests {\n\t\tt.Run(tt.name, func(t *testing.T) {\n\t\t\t// Create TestingT\n\t\t\ttestingT := agenttest.NewTestingT(tt.name)\n\n\t\t\t// Create V8 isolate and context\n\t\t\tiso := v8go.NewIsolate()\n\t\t\tdefer iso.Dispose()\n\n\t\t\tv8ctx := v8go.NewContext(iso)\n\t\t\tdefer v8ctx.Close()\n\n\t\t\t// Create testing object\n\t\t\ttestObj, err := agenttest.NewTestingTObject(v8ctx, testingT)\n\t\t\tif err != nil {\n\t\t\t\tt.Fatalf(\"Failed to create testing object: %v\", err)\n\t\t\t}\n\n\t\t\t// Set testing object as global\n\t\t\tglobal := v8ctx.Global()\n\t\t\tglobal.Set(\"__test_t\", testObj)\n\n\t\t\t// Run the test script\n\t\t\t_, err = v8ctx.RunScript(tt.script, \"test.js\")\n\n\t\t\t// Check results\n\t\t\tif tt.shouldFail {\n\t\t\t\tassert.True(t, testingT.Failed(), \"Test should have failed\")\n\t\t\t} else {\n\t\t\t\tif err != nil {\n\t\t\t\t\tt.Errorf(\"Script execution error: %v\", err)\n\t\t\t\t}\n\t\t\t\tassert.False(t, testingT.Failed(), \"Test should have passed, errors: %v\", testingT.Errors())\n\t\t\t}\n\t\t})\n\t}\n}\n\n// Ensure v8 is used (for script loading)\nvar _ = v8.Scripts\n" }, { "path": "agent/test/assert_test.go", "content": "package test\n\nimport (\n\t\"testing\"\n\n\t\"github.com/yaoapp/yao/agent/context\"\n)\n\nfunc TestAsserter_JSONPath_ArrayEquality(t *testing.T) {\n\tasserter := NewAsserter()\n\n\ttests := []struct {\n\t\tname string\n\t\ttc *Case\n\t\toutput interface{}\n\t\texpected bool\n\t\terrMsg string\n\t}{\n\t\t{\n\t\t\tname: \"array equals array - same content\",\n\t\t\ttc: &Case{\n\t\t\t\tAssert: map[string]interface{}{\n\t\t\t\t\t\"type\": \"json_path\",\n\t\t\t\t\t\"path\": \"search_types\",\n\t\t\t\t\t\"value\": []interface{}{\"db\"},\n\t\t\t\t},\n\t\t\t},\n\t\t\toutput: map[string]interface{}{\n\t\t\t\t\"search_types\": []interface{}{\"db\"},\n\t\t\t},\n\t\t\texpected: true,\n\t\t},\n\t\t{\n\t\t\tname: \"array equals array - different content\",\n\t\t\ttc: &Case{\n\t\t\t\tAssert: map[string]interface{}{\n\t\t\t\t\t\"type\": \"json_path\",\n\t\t\t\t\t\"path\": \"search_types\",\n\t\t\t\t\t\"value\": []interface{}{\"db\"},\n\t\t\t\t},\n\t\t\t},\n\t\t\toutput: map[string]interface{}{\n\t\t\t\t\"search_types\": []interface{}{\"web\"},\n\t\t\t},\n\t\t\texpected: false,\n\t\t},\n\t\t{\n\t\t\tname: \"array equals array - multiple elements\",\n\t\t\ttc: &Case{\n\t\t\t\tAssert: map[string]interface{}{\n\t\t\t\t\t\"type\": \"json_path\",\n\t\t\t\t\t\"path\": \"search_types\",\n\t\t\t\t\t\"value\": []interface{}{\"web\", \"db\"},\n\t\t\t\t},\n\t\t\t},\n\t\t\toutput: map[string]interface{}{\n\t\t\t\t\"search_types\": []interface{}{\"web\", \"db\"},\n\t\t\t},\n\t\t\texpected: true,\n\t\t},\n\t\t{\n\t\t\tname: \"array equals array - different order\",\n\t\t\ttc: &Case{\n\t\t\t\tAssert: map[string]interface{}{\n\t\t\t\t\t\"type\": \"json_path\",\n\t\t\t\t\t\"path\": \"search_types\",\n\t\t\t\t\t\"value\": []interface{}{\"db\", \"web\"},\n\t\t\t\t},\n\t\t\t},\n\t\t\toutput: map[string]interface{}{\n\t\t\t\t\"search_types\": []interface{}{\"web\", \"db\"},\n\t\t\t},\n\t\t\texpected: false, // Order matters for array equality\n\t\t},\n\t\t{\n\t\t\tname: \"scalar in array - match\",\n\t\t\ttc: &Case{\n\t\t\t\tAssert: map[string]interface{}{\n\t\t\t\t\t\"type\": \"json_path\",\n\t\t\t\t\t\"path\": \"status\",\n\t\t\t\t\t\"value\": []interface{}{\"active\", \"pending\"},\n\t\t\t\t},\n\t\t\t},\n\t\t\toutput: map[string]interface{}{\n\t\t\t\t\"status\": \"active\",\n\t\t\t},\n\t\t\texpected: true, // \"active\" is one of [\"active\", \"pending\"]\n\t\t},\n\t\t{\n\t\t\tname: \"scalar in array - no match\",\n\t\t\ttc: &Case{\n\t\t\t\tAssert: map[string]interface{}{\n\t\t\t\t\t\"type\": \"json_path\",\n\t\t\t\t\t\"path\": \"status\",\n\t\t\t\t\t\"value\": []interface{}{\"active\", \"pending\"},\n\t\t\t\t},\n\t\t\t},\n\t\t\toutput: map[string]interface{}{\n\t\t\t\t\"status\": \"inactive\",\n\t\t\t},\n\t\t\texpected: false,\n\t\t},\n\t\t{\n\t\t\tname: \"simple value comparison\",\n\t\t\ttc: &Case{\n\t\t\t\tAssert: map[string]interface{}{\n\t\t\t\t\t\"type\": \"json_path\",\n\t\t\t\t\t\"path\": \"need_search\",\n\t\t\t\t\t\"value\": true,\n\t\t\t\t},\n\t\t\t},\n\t\t\toutput: map[string]interface{}{\n\t\t\t\t\"need_search\": true,\n\t\t\t},\n\t\t\texpected: true,\n\t\t},\n\t\t{\n\t\t\tname: \"nested path\",\n\t\t\ttc: &Case{\n\t\t\t\tAssert: map[string]interface{}{\n\t\t\t\t\t\"type\": \"json_path\",\n\t\t\t\t\t\"path\": \"result.count\",\n\t\t\t\t\t\"value\": float64(5),\n\t\t\t\t},\n\t\t\t},\n\t\t\toutput: map[string]interface{}{\n\t\t\t\t\"result\": map[string]interface{}{\n\t\t\t\t\t\"count\": float64(5),\n\t\t\t\t},\n\t\t\t},\n\t\t\texpected: true,\n\t\t},\n\t\t{\n\t\t\tname: \"array index access\",\n\t\t\ttc: &Case{\n\t\t\t\tAssert: map[string]interface{}{\n\t\t\t\t\t\"type\": \"json_path\",\n\t\t\t\t\t\"path\": \"items[0].name\",\n\t\t\t\t\t\"value\": \"first\",\n\t\t\t\t},\n\t\t\t},\n\t\t\toutput: map[string]interface{}{\n\t\t\t\t\"items\": []interface{}{\n\t\t\t\t\tmap[string]interface{}{\"name\": \"first\"},\n\t\t\t\t\tmap[string]interface{}{\"name\": \"second\"},\n\t\t\t\t},\n\t\t\t},\n\t\t\texpected: true,\n\t\t},\n\t}\n\n\tfor _, tt := range tests {\n\t\tt.Run(tt.name, func(t *testing.T) {\n\t\t\tpassed, errMsg := asserter.Validate(tt.tc, tt.output)\n\t\t\tif passed != tt.expected {\n\t\t\t\tt.Errorf(\"Expected passed=%v, got passed=%v, error: %s\", tt.expected, passed, errMsg)\n\t\t\t}\n\t\t})\n\t}\n}\n\nfunc TestAsserter_Contains(t *testing.T) {\n\tasserter := NewAsserter()\n\n\ttests := []struct {\n\t\tname string\n\t\ttc *Case\n\t\toutput interface{}\n\t\texpected bool\n\t}{\n\t\t{\n\t\t\tname: \"string contains substring\",\n\t\t\ttc: &Case{\n\t\t\t\tAssert: map[string]interface{}{\n\t\t\t\t\t\"type\": \"contains\",\n\t\t\t\t\t\"value\": \"hello\",\n\t\t\t\t},\n\t\t\t},\n\t\t\toutput: \"hello world\",\n\t\t\texpected: true,\n\t\t},\n\t\t{\n\t\t\tname: \"string does not contain\",\n\t\t\ttc: &Case{\n\t\t\t\tAssert: map[string]interface{}{\n\t\t\t\t\t\"type\": \"contains\",\n\t\t\t\t\t\"value\": \"goodbye\",\n\t\t\t\t},\n\t\t\t},\n\t\t\toutput: \"hello world\",\n\t\t\texpected: false,\n\t\t},\n\t\t{\n\t\t\tname: \"JSON contains field\",\n\t\t\ttc: &Case{\n\t\t\t\tAssert: map[string]interface{}{\n\t\t\t\t\t\"type\": \"contains\",\n\t\t\t\t\t\"value\": \"success\",\n\t\t\t\t},\n\t\t\t},\n\t\t\toutput: map[string]interface{}{\n\t\t\t\t\"status\": \"success\",\n\t\t\t},\n\t\t\texpected: true,\n\t\t},\n\t}\n\n\tfor _, tt := range tests {\n\t\tt.Run(tt.name, func(t *testing.T) {\n\t\t\tpassed, _ := asserter.Validate(tt.tc, tt.output)\n\t\t\tif passed != tt.expected {\n\t\t\t\tt.Errorf(\"Expected passed=%v, got passed=%v\", tt.expected, passed)\n\t\t\t}\n\t\t})\n\t}\n}\n\nfunc TestAsserter_MultipleAssertions(t *testing.T) {\n\tasserter := NewAsserter()\n\n\ttc := &Case{\n\t\tAssert: []interface{}{\n\t\t\tmap[string]interface{}{\n\t\t\t\t\"type\": \"json_path\",\n\t\t\t\t\"path\": \"need_search\",\n\t\t\t\t\"value\": true,\n\t\t\t},\n\t\t\tmap[string]interface{}{\n\t\t\t\t\"type\": \"json_path\",\n\t\t\t\t\"path\": \"search_types\",\n\t\t\t\t\"value\": []interface{}{\"web\"},\n\t\t\t},\n\t\t},\n\t}\n\n\toutput := map[string]interface{}{\n\t\t\"need_search\": true,\n\t\t\"search_types\": []interface{}{\"web\"},\n\t}\n\n\tpassed, errMsg := asserter.Validate(tc, output)\n\tif !passed {\n\t\tt.Errorf(\"Expected all assertions to pass, got error: %s\", errMsg)\n\t}\n}\n\nfunc TestAsserter_Negate(t *testing.T) {\n\tasserter := NewAsserter()\n\n\ttc := &Case{\n\t\tAssert: map[string]interface{}{\n\t\t\t\"type\": \"contains\",\n\t\t\t\"value\": \"error\",\n\t\t\t\"negate\": true,\n\t\t},\n\t}\n\n\toutput := \"success message\"\n\n\tpassed, _ := asserter.Validate(tc, output)\n\tif !passed {\n\t\tt.Error(\"Expected negated assertion to pass\")\n\t}\n}\n\nfunc TestAsserter_Regex(t *testing.T) {\n\tasserter := NewAsserter()\n\n\ttests := []struct {\n\t\tname string\n\t\ttc *Case\n\t\toutput interface{}\n\t\texpected bool\n\t}{\n\t\t{\n\t\t\tname: \"regex matches\",\n\t\t\ttc: &Case{\n\t\t\t\tAssert: map[string]interface{}{\n\t\t\t\t\t\"type\": \"regex\",\n\t\t\t\t\t\"value\": `\\d{3}-\\d{4}`,\n\t\t\t\t},\n\t\t\t},\n\t\t\toutput: \"Phone: 123-4567\",\n\t\t\texpected: true,\n\t\t},\n\t\t{\n\t\t\tname: \"regex does not match\",\n\t\t\ttc: &Case{\n\t\t\t\tAssert: map[string]interface{}{\n\t\t\t\t\t\"type\": \"regex\",\n\t\t\t\t\t\"value\": `\\d{3}-\\d{4}`,\n\t\t\t\t},\n\t\t\t},\n\t\t\toutput: \"No phone number here\",\n\t\t\texpected: false,\n\t\t},\n\t}\n\n\tfor _, tt := range tests {\n\t\tt.Run(tt.name, func(t *testing.T) {\n\t\t\tpassed, _ := asserter.Validate(tt.tc, tt.output)\n\t\t\tif passed != tt.expected {\n\t\t\t\tt.Errorf(\"Expected passed=%v, got passed=%v\", tt.expected, passed)\n\t\t\t}\n\t\t})\n\t}\n}\n\nfunc TestAsserter_ToolCalled(t *testing.T) {\n\ttests := []struct {\n\t\tname string\n\t\ttc *Case\n\t\tresponse *context.Response\n\t\texpected bool\n\t\terrMsg string\n\t}{\n\t\t{\n\t\t\tname: \"tool called - exact match\",\n\t\t\ttc: &Case{\n\t\t\t\tAssert: map[string]interface{}{\n\t\t\t\t\t\"type\": \"tool_called\",\n\t\t\t\t\t\"value\": \"agents_expense_tools__setup\",\n\t\t\t\t},\n\t\t\t},\n\t\t\tresponse: &context.Response{\n\t\t\t\tTools: []context.ToolCallResponse{\n\t\t\t\t\t{Tool: \"agents_expense_tools__setup\", Result: map[string]interface{}{\"success\": true}},\n\t\t\t\t},\n\t\t\t},\n\t\t\texpected: true,\n\t\t},\n\t\t{\n\t\t\tname: \"tool called - suffix match\",\n\t\t\ttc: &Case{\n\t\t\t\tAssert: map[string]interface{}{\n\t\t\t\t\t\"type\": \"tool_called\",\n\t\t\t\t\t\"value\": \"setup\",\n\t\t\t\t},\n\t\t\t},\n\t\t\tresponse: &context.Response{\n\t\t\t\tTools: []context.ToolCallResponse{\n\t\t\t\t\t{Tool: \"agents_expense_tools__setup\", Result: map[string]interface{}{\"success\": true}},\n\t\t\t\t},\n\t\t\t},\n\t\t\texpected: true,\n\t\t},\n\t\t{\n\t\t\tname: \"tool not called\",\n\t\t\ttc: &Case{\n\t\t\t\tAssert: map[string]interface{}{\n\t\t\t\t\t\"type\": \"tool_called\",\n\t\t\t\t\t\"value\": \"setup\",\n\t\t\t\t},\n\t\t\t},\n\t\t\tresponse: &context.Response{\n\t\t\t\tTools: []context.ToolCallResponse{},\n\t\t\t},\n\t\t\texpected: false,\n\t\t},\n\t\t{\n\t\t\tname: \"tool called - wrong tool\",\n\t\t\ttc: &Case{\n\t\t\t\tAssert: map[string]interface{}{\n\t\t\t\t\t\"type\": \"tool_called\",\n\t\t\t\t\t\"value\": \"setup\",\n\t\t\t\t},\n\t\t\t},\n\t\t\tresponse: &context.Response{\n\t\t\t\tTools: []context.ToolCallResponse{\n\t\t\t\t\t{Tool: \"agents_expense_tools__submit\", Result: map[string]interface{}{\"success\": true}},\n\t\t\t\t},\n\t\t\t},\n\t\t\texpected: false,\n\t\t},\n\t\t{\n\t\t\tname: \"tool called - any of multiple\",\n\t\t\ttc: &Case{\n\t\t\t\tAssert: map[string]interface{}{\n\t\t\t\t\t\"type\": \"tool_called\",\n\t\t\t\t\t\"value\": []interface{}{\"setup\", \"init\"},\n\t\t\t\t},\n\t\t\t},\n\t\t\tresponse: &context.Response{\n\t\t\t\tTools: []context.ToolCallResponse{\n\t\t\t\t\t{Tool: \"agents_expense_tools__init\", Result: map[string]interface{}{\"success\": true}},\n\t\t\t\t},\n\t\t\t},\n\t\t\texpected: true,\n\t\t},\n\t\t{\n\t\t\tname: \"tool called - with arguments (map)\",\n\t\t\ttc: &Case{\n\t\t\t\tAssert: map[string]interface{}{\n\t\t\t\t\t\"type\": \"tool_called\",\n\t\t\t\t\t\"value\": map[string]interface{}{\n\t\t\t\t\t\t\"name\": \"setup\",\n\t\t\t\t\t\t\"arguments\": map[string]interface{}{\n\t\t\t\t\t\t\t\"action\": \"init\",\n\t\t\t\t\t\t},\n\t\t\t\t\t},\n\t\t\t\t},\n\t\t\t},\n\t\t\tresponse: &context.Response{\n\t\t\t\tTools: []context.ToolCallResponse{\n\t\t\t\t\t{\n\t\t\t\t\t\tTool: \"agents_expense_tools__setup\",\n\t\t\t\t\t\tArguments: map[string]interface{}{\"action\": \"init\", \"config\": map[string]interface{}{}},\n\t\t\t\t\t\tResult: map[string]interface{}{\"success\": true},\n\t\t\t\t\t},\n\t\t\t\t},\n\t\t\t},\n\t\t\texpected: true,\n\t\t},\n\t\t{\n\t\t\tname: \"tool called - with arguments (JSON string)\",\n\t\t\ttc: &Case{\n\t\t\t\tAssert: map[string]interface{}{\n\t\t\t\t\t\"type\": \"tool_called\",\n\t\t\t\t\t\"value\": map[string]interface{}{\n\t\t\t\t\t\t\"name\": \"setup\",\n\t\t\t\t\t\t\"arguments\": map[string]interface{}{\n\t\t\t\t\t\t\t\"action\": \"init\",\n\t\t\t\t\t\t},\n\t\t\t\t\t},\n\t\t\t\t},\n\t\t\t},\n\t\t\tresponse: &context.Response{\n\t\t\t\tTools: []context.ToolCallResponse{\n\t\t\t\t\t{\n\t\t\t\t\t\tTool: \"agents_expense_tools__setup\",\n\t\t\t\t\t\tArguments: `{\"action\":\"init\",\"config\":{\"default_currency\":\"USD\"}}`,\n\t\t\t\t\t\tResult: map[string]interface{}{\"success\": true},\n\t\t\t\t\t},\n\t\t\t\t},\n\t\t\t},\n\t\t\texpected: true,\n\t\t},\n\t\t{\n\t\t\tname: \"tool called - wrong arguments\",\n\t\t\ttc: &Case{\n\t\t\t\tAssert: map[string]interface{}{\n\t\t\t\t\t\"type\": \"tool_called\",\n\t\t\t\t\t\"value\": map[string]interface{}{\n\t\t\t\t\t\t\"name\": \"setup\",\n\t\t\t\t\t\t\"arguments\": map[string]interface{}{\n\t\t\t\t\t\t\t\"action\": \"update\",\n\t\t\t\t\t\t},\n\t\t\t\t\t},\n\t\t\t\t},\n\t\t\t},\n\t\t\tresponse: &context.Response{\n\t\t\t\tTools: []context.ToolCallResponse{\n\t\t\t\t\t{\n\t\t\t\t\t\tTool: \"agents_expense_tools__setup\",\n\t\t\t\t\t\tArguments: map[string]interface{}{\"action\": \"init\"},\n\t\t\t\t\t\tResult: map[string]interface{}{\"success\": true},\n\t\t\t\t\t},\n\t\t\t\t},\n\t\t\t},\n\t\t\texpected: false,\n\t\t},\n\t\t{\n\t\t\tname: \"no response\",\n\t\t\ttc: &Case{\n\t\t\t\tAssert: map[string]interface{}{\n\t\t\t\t\t\"type\": \"tool_called\",\n\t\t\t\t\t\"value\": \"setup\",\n\t\t\t\t},\n\t\t\t},\n\t\t\tresponse: nil,\n\t\t\texpected: false,\n\t\t},\n\t}\n\n\tfor _, tt := range tests {\n\t\tt.Run(tt.name, func(t *testing.T) {\n\t\t\tasserter := NewAsserter().WithResponse(tt.response)\n\t\t\tpassed, errMsg := asserter.Validate(tt.tc, nil)\n\t\t\tif passed != tt.expected {\n\t\t\t\tt.Errorf(\"Expected passed=%v, got passed=%v, error: %s\", tt.expected, passed, errMsg)\n\t\t\t}\n\t\t})\n\t}\n}\n\nfunc TestAsserter_ToolResult(t *testing.T) {\n\ttests := []struct {\n\t\tname string\n\t\ttc *Case\n\t\tresponse *context.Response\n\t\texpected bool\n\t}{\n\t\t{\n\t\t\tname: \"tool result - success check\",\n\t\t\ttc: &Case{\n\t\t\t\tAssert: map[string]interface{}{\n\t\t\t\t\t\"type\": \"tool_result\",\n\t\t\t\t\t\"value\": map[string]interface{}{\n\t\t\t\t\t\t\"tool\": \"setup\",\n\t\t\t\t\t\t\"result\": map[string]interface{}{\n\t\t\t\t\t\t\t\"success\": true,\n\t\t\t\t\t\t},\n\t\t\t\t\t},\n\t\t\t\t},\n\t\t\t},\n\t\t\tresponse: &context.Response{\n\t\t\t\tTools: []context.ToolCallResponse{\n\t\t\t\t\t{\n\t\t\t\t\t\tTool: \"agents_expense_tools__setup\",\n\t\t\t\t\t\tResult: map[string]interface{}{\"success\": true, \"message\": \"Setup complete\"},\n\t\t\t\t\t},\n\t\t\t\t},\n\t\t\t},\n\t\t\texpected: true,\n\t\t},\n\t\t{\n\t\t\tname: \"tool result - message check with regex\",\n\t\t\ttc: &Case{\n\t\t\t\tAssert: map[string]interface{}{\n\t\t\t\t\t\"type\": \"tool_result\",\n\t\t\t\t\t\"value\": map[string]interface{}{\n\t\t\t\t\t\t\"tool\": \"setup\",\n\t\t\t\t\t\t\"result\": map[string]interface{}{\n\t\t\t\t\t\t\t\"message\": \"regex:(?i)setup.*complete\",\n\t\t\t\t\t\t},\n\t\t\t\t\t},\n\t\t\t\t},\n\t\t\t},\n\t\t\tresponse: &context.Response{\n\t\t\t\tTools: []context.ToolCallResponse{\n\t\t\t\t\t{\n\t\t\t\t\t\tTool: \"agents_expense_tools__setup\",\n\t\t\t\t\t\tResult: map[string]interface{}{\"success\": true, \"message\": \"Setup complete!\"},\n\t\t\t\t\t},\n\t\t\t\t},\n\t\t\t},\n\t\t\texpected: true,\n\t\t},\n\t\t{\n\t\t\tname: \"tool result - no expected result (just check no error)\",\n\t\t\ttc: &Case{\n\t\t\t\tAssert: map[string]interface{}{\n\t\t\t\t\t\"type\": \"tool_result\",\n\t\t\t\t\t\"value\": map[string]interface{}{\n\t\t\t\t\t\t\"tool\": \"setup\",\n\t\t\t\t\t},\n\t\t\t\t},\n\t\t\t},\n\t\t\tresponse: &context.Response{\n\t\t\t\tTools: []context.ToolCallResponse{\n\t\t\t\t\t{\n\t\t\t\t\t\tTool: \"agents_expense_tools__setup\",\n\t\t\t\t\t\tResult: map[string]interface{}{\"success\": true},\n\t\t\t\t\t},\n\t\t\t\t},\n\t\t\t},\n\t\t\texpected: true,\n\t\t},\n\t\t{\n\t\t\tname: \"tool result - tool has error\",\n\t\t\ttc: &Case{\n\t\t\t\tAssert: map[string]interface{}{\n\t\t\t\t\t\"type\": \"tool_result\",\n\t\t\t\t\t\"value\": map[string]interface{}{\n\t\t\t\t\t\t\"tool\": \"setup\",\n\t\t\t\t\t},\n\t\t\t\t},\n\t\t\t},\n\t\t\tresponse: &context.Response{\n\t\t\t\tTools: []context.ToolCallResponse{\n\t\t\t\t\t{\n\t\t\t\t\t\tTool: \"agents_expense_tools__setup\",\n\t\t\t\t\t\tError: \"permission denied\",\n\t\t\t\t\t},\n\t\t\t\t},\n\t\t\t},\n\t\t\texpected: false,\n\t\t},\n\t\t{\n\t\t\tname: \"tool result - result mismatch\",\n\t\t\ttc: &Case{\n\t\t\t\tAssert: map[string]interface{}{\n\t\t\t\t\t\"type\": \"tool_result\",\n\t\t\t\t\t\"value\": map[string]interface{}{\n\t\t\t\t\t\t\"tool\": \"setup\",\n\t\t\t\t\t\t\"result\": map[string]interface{}{\n\t\t\t\t\t\t\t\"success\": true,\n\t\t\t\t\t\t},\n\t\t\t\t\t},\n\t\t\t\t},\n\t\t\t},\n\t\t\tresponse: &context.Response{\n\t\t\t\tTools: []context.ToolCallResponse{\n\t\t\t\t\t{\n\t\t\t\t\t\tTool: \"agents_expense_tools__setup\",\n\t\t\t\t\t\tResult: map[string]interface{}{\"success\": false},\n\t\t\t\t\t},\n\t\t\t\t},\n\t\t\t},\n\t\t\texpected: false,\n\t\t},\n\t}\n\n\tfor _, tt := range tests {\n\t\tt.Run(tt.name, func(t *testing.T) {\n\t\t\tasserter := NewAsserter().WithResponse(tt.response)\n\t\t\tpassed, errMsg := asserter.Validate(tt.tc, nil)\n\t\t\tif passed != tt.expected {\n\t\t\t\tt.Errorf(\"Expected passed=%v, got passed=%v, error: %s\", tt.expected, passed, errMsg)\n\t\t\t}\n\t\t})\n\t}\n}\n\nfunc TestAsserter_MultipleToolAssertions(t *testing.T) {\n\t// This tests the exact scenario from setup-006: tool_called + tool_result\n\tresponse := &context.Response{\n\t\tTools: []context.ToolCallResponse{\n\t\t\t{\n\t\t\t\tTool: \"agents_expense_tools__setup\",\n\t\t\t\tArguments: `{\"action\":\"init\",\"config\":{\"default_currency\":\"USD\",\"categories\":[{\"id\":\"meals\",\"name\":\"Meals\",\"daily_limit\":100}]}}`,\n\t\t\t\tResult: map[string]interface{}{\n\t\t\t\t\t\"success\": true,\n\t\t\t\t\t\"action\": \"init\",\n\t\t\t\t\t\"config\": map[string]interface{}{\n\t\t\t\t\t\t\"default_currency\": \"USD\",\n\t\t\t\t\t},\n\t\t\t\t\t\"message\": \"Setup complete!\",\n\t\t\t\t},\n\t\t\t},\n\t\t},\n\t}\n\n\tasserter := NewAsserter().WithResponse(response)\n\n\ttc := &Case{\n\t\tAssert: []interface{}{\n\t\t\tmap[string]interface{}{\n\t\t\t\t\"type\": \"tool_called\",\n\t\t\t\t\"value\": map[string]interface{}{\n\t\t\t\t\t\"name\": \"setup\",\n\t\t\t\t\t\"arguments\": map[string]interface{}{\n\t\t\t\t\t\t\"action\": \"init\",\n\t\t\t\t\t},\n\t\t\t\t},\n\t\t\t},\n\t\t\tmap[string]interface{}{\n\t\t\t\t\"type\": \"tool_result\",\n\t\t\t\t\"value\": map[string]interface{}{\n\t\t\t\t\t\"tool\": \"setup\",\n\t\t\t\t\t\"result\": map[string]interface{}{\n\t\t\t\t\t\t\"success\": true,\n\t\t\t\t\t},\n\t\t\t\t},\n\t\t\t},\n\t\t},\n\t}\n\n\tresult := asserter.ValidateWithDetails(tc, nil)\n\tif !result.Passed {\n\t\tt.Errorf(\"Expected multiple tool assertions to pass, got: %s\", result.Message)\n\t}\n}\n\nfunc TestAsserter_SharedAsserterWithResponse(t *testing.T) {\n\t// Test that a shared asserter correctly uses WithResponse\n\tasserter := NewAsserter()\n\n\t// First call without response - should fail\n\ttc := &Case{\n\t\tAssert: map[string]interface{}{\n\t\t\t\"type\": \"tool_called\",\n\t\t\t\"value\": \"setup\",\n\t\t},\n\t}\n\tresult := asserter.ValidateWithDetails(tc, nil)\n\tif result.Passed {\n\t\tt.Error(\"Expected tool_called to fail without response\")\n\t}\n\tif result.Message != \"no response available for tool_called assertion\" {\n\t\tt.Errorf(\"Unexpected message: %s\", result.Message)\n\t}\n\n\t// Now set response\n\tresponse := &context.Response{\n\t\tTools: []context.ToolCallResponse{\n\t\t\t{\n\t\t\t\tTool: \"agents_expense_tools__setup\",\n\t\t\t\tResult: map[string]interface{}{\"success\": true},\n\t\t\t},\n\t\t},\n\t}\n\tasserter.WithResponse(response)\n\n\t// Should pass now\n\tresult = asserter.ValidateWithDetails(tc, nil)\n\tif !result.Passed {\n\t\tt.Errorf(\"Expected tool_called to pass with response, got: %s\", result.Message)\n\t}\n}\n\nfunc TestAsserter_Setup006Scenario(t *testing.T) {\n\t// Exact reproduction of setup-006 scenario\n\t// Turn 2: tool was called with action: init, result has success: true\n\tresponse := &context.Response{\n\t\tTools: []context.ToolCallResponse{\n\t\t\t{\n\t\t\t\tTool: \"agents_expense_tools__setup\",\n\t\t\t\tArguments: `{\"action\":\"init\",\"config\":{\"default_currency\":\"USD\",\"categories\":[{\"id\":\"meals\",\"name\":\"Meals\",\"daily_limit\":100},{\"id\":\"travel\",\"name\":\"Travel\",\"daily_limit\":500}]}}`,\n\t\t\t\tResult: map[string]interface{}{\n\t\t\t\t\t\"config\": map[string]interface{}{\n\t\t\t\t\t\t\"categories\": []interface{}{\n\t\t\t\t\t\t\tmap[string]interface{}{\"daily_limit\": float64(100), \"id\": \"meals\", \"name\": \"Meals\"},\n\t\t\t\t\t\t\tmap[string]interface{}{\"daily_limit\": float64(500), \"id\": \"travel\", \"name\": \"Travel\"},\n\t\t\t\t\t\t},\n\t\t\t\t\t\t\"default_currency\": \"USD\",\n\t\t\t\t\t},\n\t\t\t\t\t\"message\": \"Setup complete! The expense system has been initialized successfully with the configured settings. You can now start submitting expenses.\",\n\t\t\t\t\t\"success\": true,\n\t\t\t\t\t\"action\": \"init\",\n\t\t\t\t},\n\t\t\t},\n\t\t},\n\t}\n\n\t// This is the exact assert from setup-006's quick_complete checkpoint\n\tassertDef := []interface{}{\n\t\tmap[string]interface{}{\n\t\t\t\"type\": \"tool_called\",\n\t\t\t\"value\": map[string]interface{}{\n\t\t\t\t\"name\": \"setup\",\n\t\t\t\t\"arguments\": map[string]interface{}{\n\t\t\t\t\t\"action\": \"init\",\n\t\t\t\t},\n\t\t\t},\n\t\t},\n\t\tmap[string]interface{}{\n\t\t\t\"type\": \"tool_result\",\n\t\t\t\"value\": map[string]interface{}{\n\t\t\t\t\"tool\": \"setup\",\n\t\t\t\t\"result\": map[string]interface{}{\n\t\t\t\t\t\"success\": true,\n\t\t\t\t},\n\t\t\t},\n\t\t},\n\t}\n\n\tasserter := NewAsserter().WithResponse(response)\n\ttc := &Case{Assert: assertDef}\n\n\tresult := asserter.ValidateWithDetails(tc, nil)\n\tif !result.Passed {\n\t\tt.Errorf(\"Expected setup-006 scenario to pass, got: %s\", result.Message)\n\t}\n\n\t// Also test individual assertions\n\tt.Run(\"tool_called only\", func(t *testing.T) {\n\t\ttc2 := &Case{Assert: assertDef[0]}\n\t\tresult2 := asserter.ValidateWithDetails(tc2, nil)\n\t\tif !result2.Passed {\n\t\t\tt.Errorf(\"Expected tool_called to pass, got: %s\", result2.Message)\n\t\t}\n\t})\n\n\tt.Run(\"tool_result only\", func(t *testing.T) {\n\t\ttc3 := &Case{Assert: assertDef[1]}\n\t\tresult3 := asserter.ValidateWithDetails(tc3, nil)\n\t\tif !result3.Passed {\n\t\t\tt.Errorf(\"Expected tool_result to pass, got: %s\", result3.Message)\n\t\t}\n\t})\n}\n\nfunc TestAsserter_Setup003Scenario(t *testing.T) {\n\t// Exact reproduction of setup-003 scenario\n\t// Turn 3: tool was called with action: update\n\tresponse := &context.Response{\n\t\tTools: []context.ToolCallResponse{\n\t\t\t{\n\t\t\t\tTool: \"agents_expense_tools__setup\",\n\t\t\t\tArguments: `{\"action\":\"update\",\"config\":{\"categories\":[{\"daily_limit\":500,\"id\":\"meals\",\"name\":\"Business Meals\"}]}}`,\n\t\t\t\tResult: map[string]interface{}{\n\t\t\t\t\t\"action\": \"update\",\n\t\t\t\t\t\"config\": map[string]interface{}{\n\t\t\t\t\t\t\"categories\": []interface{}{\n\t\t\t\t\t\t\tmap[string]interface{}{\"daily_limit\": float64(500), \"id\": \"meals\", \"name\": \"Business Meals\"},\n\t\t\t\t\t\t},\n\t\t\t\t\t},\n\t\t\t\t\t\"message\": \"Configuration updated successfully! Your changes have been saved.\",\n\t\t\t\t\t\"success\": true,\n\t\t\t\t},\n\t\t\t},\n\t\t},\n\t}\n\n\t// This is the exact assert from setup-003's update_complete checkpoint\n\tassertDef := []interface{}{\n\t\tmap[string]interface{}{\n\t\t\t\"type\": \"tool_called\",\n\t\t\t\"value\": map[string]interface{}{\n\t\t\t\t\"name\": \"setup\",\n\t\t\t\t\"arguments\": map[string]interface{}{\n\t\t\t\t\t\"action\": \"update\",\n\t\t\t\t},\n\t\t\t},\n\t\t},\n\t\tmap[string]interface{}{\n\t\t\t\"type\": \"tool_result\",\n\t\t\t\"value\": map[string]interface{}{\n\t\t\t\t\"tool\": \"setup\",\n\t\t\t\t\"result\": map[string]interface{}{\n\t\t\t\t\t\"success\": true,\n\t\t\t\t},\n\t\t\t},\n\t\t},\n\t}\n\n\tasserter := NewAsserter().WithResponse(response)\n\ttc := &Case{Assert: assertDef}\n\n\tresult := asserter.ValidateWithDetails(tc, nil)\n\tif !result.Passed {\n\t\tt.Errorf(\"Expected setup-003 scenario to pass, got: %s\", result.Message)\n\t}\n\n\t// Test individual assertions\n\tt.Run(\"tool_called with action:update\", func(t *testing.T) {\n\t\ttc2 := &Case{Assert: assertDef[0]}\n\t\tresult2 := asserter.ValidateWithDetails(tc2, nil)\n\t\tif !result2.Passed {\n\t\t\tt.Errorf(\"Expected tool_called to pass, got: %s\", result2.Message)\n\t\t}\n\t})\n}\n\nfunc TestMatchToolName(t *testing.T) {\n\ttests := []struct {\n\t\tactual string\n\t\texpected string\n\t\tmatch bool\n\t}{\n\t\t{\"agents_expense_tools__setup\", \"agents_expense_tools__setup\", true},\n\t\t{\"agents_expense_tools__setup\", \"setup\", true},\n\t\t{\"agents.expense.tools.setup\", \"setup\", true},\n\t\t{\"agents_expense_tools__setup\", \"init\", false},\n\t\t{\"setup\", \"setup\", true},\n\t}\n\n\tfor _, tt := range tests {\n\t\tt.Run(tt.actual+\"_\"+tt.expected, func(t *testing.T) {\n\t\t\tif matchToolName(tt.actual, tt.expected) != tt.match {\n\t\t\t\tt.Errorf(\"matchToolName(%q, %q) = %v, want %v\", tt.actual, tt.expected, !tt.match, tt.match)\n\t\t\t}\n\t\t})\n\t}\n}\n" }, { "path": "agent/test/context.go", "content": "package test\n\nimport (\n\tstdContext \"context\"\n\n\t\"github.com/yaoapp/yao/agent/context\"\n\t\"github.com/yaoapp/yao/agent/output/message\"\n\t\"github.com/yaoapp/yao/openapi/oauth/types\"\n)\n\n// NewTestContext creates a new context for testing\n// This is similar to newAgentNextTestContext in agent_next_test.go\n// but configurable via Environment\nfunc NewTestContext(chatID, assistantID string, env *Environment) *context.Context {\n\t// Build authorized info from environment\n\tauthorized := buildAuthorizedInfo(env)\n\n\t// Create context with standard initialization\n\tctx := context.New(stdContext.Background(), authorized, chatID)\n\tctx.ID = chatID\n\tctx.AssistantID = assistantID\n\tctx.Locale = env.Locale\n\tctx.Client = context.Client{\n\t\tType: env.ClientType,\n\t\tUserAgent: \"yao-agent-test/1.0\",\n\t\tIP: env.ClientIP,\n\t}\n\tctx.Referer = env.Referer\n\tctx.Accept = context.AcceptStandard\n\tctx.IDGenerator = message.NewIDGenerator()\n\tctx.Metadata = make(map[string]interface{})\n\n\t// Apply metadata from context config if available\n\tif env.ContextConfig != nil && env.ContextConfig.Metadata != nil {\n\t\tfor k, v := range env.ContextConfig.Metadata {\n\t\t\tctx.Metadata[k] = v\n\t\t}\n\t}\n\n\t// Initialize interrupt controller\n\tctx.Interrupt = context.NewInterruptController()\n\n\t// Close the default logger created by context.New() and use noop logger\n\t// to suppress LLM debug output during tests\n\tif ctx.Logger != nil {\n\t\tctx.Logger.Close()\n\t}\n\tctx.Logger = context.Noop()\n\n\treturn ctx\n}\n\n// buildAuthorizedInfo builds AuthorizedInfo from Environment\nfunc buildAuthorizedInfo(env *Environment) *types.AuthorizedInfo {\n\tauthorized := &types.AuthorizedInfo{\n\t\tSubject: env.UserID,\n\t\tUserID: env.UserID,\n\t\tTeamID: env.TeamID,\n\t\tTenantID: env.TeamID,\n\t}\n\n\t// Apply custom authorized config if available\n\tif env.ContextConfig != nil && env.ContextConfig.Authorized != nil {\n\t\tauthCfg := env.ContextConfig.Authorized\n\n\t\tif authCfg.Sub != \"\" {\n\t\t\tauthorized.Subject = authCfg.Sub\n\t\t}\n\t\tif authCfg.ClientID != \"\" {\n\t\t\tauthorized.ClientID = authCfg.ClientID\n\t\t}\n\t\tif authCfg.Scope != \"\" {\n\t\t\tauthorized.Scope = authCfg.Scope\n\t\t}\n\t\tif authCfg.SessionID != \"\" {\n\t\t\tauthorized.SessionID = authCfg.SessionID\n\t\t}\n\t\tif authCfg.UserID != \"\" {\n\t\t\tauthorized.UserID = authCfg.UserID\n\t\t}\n\t\tif authCfg.TeamID != \"\" {\n\t\t\tauthorized.TeamID = authCfg.TeamID\n\t\t}\n\t\tif authCfg.TenantID != \"\" {\n\t\t\tauthorized.TenantID = authCfg.TenantID\n\t\t}\n\t\tauthorized.RememberMe = authCfg.RememberMe\n\n\t\t// Apply constraints\n\t\tif authCfg.Constraints != nil {\n\t\t\tauthorized.Constraints = types.DataConstraints{\n\t\t\t\tOwnerOnly: authCfg.Constraints.OwnerOnly,\n\t\t\t\tCreatorOnly: authCfg.Constraints.CreatorOnly,\n\t\t\t\tEditorOnly: authCfg.Constraints.EditorOnly,\n\t\t\t\tTeamOnly: authCfg.Constraints.TeamOnly,\n\t\t\t\tExtra: authCfg.Constraints.Extra,\n\t\t\t}\n\t\t}\n\t}\n\n\treturn authorized\n}\n\n// NewTestContextFromOptions creates a test context from test options and test case\nfunc NewTestContextFromOptions(chatID, assistantID string, opts *Options, tc *Case) *context.Context {\n\t// Get environment from test case (with options override)\n\tenv := tc.GetEnvironment(opts)\n\treturn NewTestContext(chatID, assistantID, env)\n}\n\n// GenerateChatID generates a unique chat ID for testing\nfunc GenerateChatID(testID string, runNumber int) string {\n\tif runNumber > 1 {\n\t\treturn \"test-\" + testID + \"-run\" + string(rune('0'+runNumber))\n\t}\n\treturn \"test-\" + testID\n}\n" }, { "path": "agent/test/dynamic_integration_test.go", "content": "package test_test\n\nimport (\n\t\"os\"\n\t\"path/filepath\"\n\t\"strings\"\n\t\"testing\"\n\n\t\"github.com/stretchr/testify/assert\"\n\t\"github.com/stretchr/testify/require\"\n\t\"github.com/yaoapp/yao/agent\"\n\tagenttest \"github.com/yaoapp/yao/agent/test\"\n\t\"github.com/yaoapp/yao/config\"\n\t\"github.com/yaoapp/yao/test\"\n)\n\n// TestDynamicRunner_CoffeeOrder tests a complete dynamic mode flow:\n// Simulator acts as a customer ordering coffee, agent handles the order\nfunc TestDynamicRunner_CoffeeOrder(t *testing.T) {\n\ttest.Prepare(t, config.Conf)\n\tdefer test.Clean()\n\n\t// Load agents\n\terr := agent.Load(config.Conf)\n\trequire.NoError(t, err, \"Failed to load agents\")\n\n\t// Create a temporary JSONL file with a dynamic test case\n\ttmpDir := t.TempDir()\n\tinputFile := filepath.Join(tmpDir, \"dynamic-inputs.jsonl\")\n\n\t// Dynamic test case: customer ordering coffee (JSONL must be single line)\n\ttestCase := `{\"id\": \"coffee-order-flow\", \"name\": \"Complete Coffee Order\", \"input\": \"Hi, I would like to order a coffee please\", \"simulator\": {\"use\": \"tests.simulator-agent\", \"options\": {\"metadata\": {\"persona\": \"A customer who wants to order a medium latte with oat milk\", \"goal\": \"Successfully complete a coffee order\"}}}, \"checkpoints\": [{\"id\": \"greeting\", \"description\": \"Agent greets and asks for order\", \"assert\": {\"type\": \"regex\", \"value\": \"(?i)(order|like|help)\"}}, {\"id\": \"ask_size\", \"description\": \"Agent asks for size\", \"after\": [\"greeting\"], \"assert\": {\"type\": \"regex\", \"value\": \"(?i)size\"}}, {\"id\": \"confirm_order\", \"description\": \"Agent confirms the order\", \"after\": [\"ask_size\"], \"assert\": {\"type\": \"regex\", \"value\": \"(?i)confirm\"}}], \"max_turns\": 8}`\n\n\terr = os.WriteFile(inputFile, []byte(testCase), 0644)\n\trequire.NoError(t, err, \"Failed to write test file\")\n\n\t// Run dynamic test\n\topts := &agenttest.Options{\n\t\tInput: inputFile,\n\t\tAgentID: \"tests.dynamic-test-agent\",\n\t\tVerbose: true,\n\t\tInputMode: agenttest.InputModeFile,\n\t}\n\topts = agenttest.MergeOptions(opts, agenttest.DefaultOptions())\n\n\trunner := agenttest.NewRunner(opts)\n\treport, err := runner.Run()\n\n\trequire.NoError(t, err, \"Runner should not return error\")\n\trequire.NotNil(t, report, \"Report should not be nil\")\n\trequire.NotNil(t, report.Summary, \"Summary should not be nil\")\n\n\t// Log results\n\tt.Logf(\"Total: %d, Passed: %d, Failed: %d\",\n\t\treport.Summary.Total, report.Summary.Passed, report.Summary.Failed)\n\n\t// Check results\n\tif len(report.Results) > 0 {\n\t\tresult := report.Results[0]\n\t\tt.Logf(\"Test [%s] Status: %s\", result.ID, result.Status)\n\n\t\t// Check metadata for dynamic mode info\n\t\tif result.Metadata != nil {\n\t\t\tif mode, ok := result.Metadata[\"mode\"].(string); ok {\n\t\t\t\tassert.Equal(t, \"dynamic\", mode, \"Should be dynamic mode\")\n\t\t\t}\n\t\t\tif turns, ok := result.Metadata[\"total_turns\"].(int); ok {\n\t\t\t\tt.Logf(\"Total turns: %d\", turns)\n\t\t\t}\n\t\t}\n\n\t\tif result.Error != \"\" {\n\t\t\tt.Logf(\"Error: %s\", result.Error)\n\t\t}\n\t}\n}\n\n// TestDynamicRunner_WithInitialInput tests dynamic mode with initial user input\nfunc TestDynamicRunner_WithInitialInput(t *testing.T) {\n\ttest.Prepare(t, config.Conf)\n\tdefer test.Clean()\n\n\t// Load agents\n\terr := agent.Load(config.Conf)\n\trequire.NoError(t, err, \"Failed to load agents\")\n\n\t// Create a test case with initial input\n\ttmpDir := t.TempDir()\n\tinputFile := filepath.Join(tmpDir, \"dynamic-inputs.jsonl\")\n\n\t// Start with user's first message (JSONL must be single line)\n\ttestCase := `{\"id\": \"coffee-with-initial\", \"name\": \"Coffee Order with Initial Message\", \"input\": \"Hi, I want to order a coffee\", \"simulator\": {\"use\": \"tests.simulator-agent\", \"options\": {\"metadata\": {\"persona\": \"Customer ordering a large cappuccino\", \"goal\": \"Complete the coffee order\"}}}, \"checkpoints\": [{\"id\": \"acknowledge\", \"description\": \"Agent acknowledges the order request\", \"assert\": {\"type\": \"regex\", \"value\": \"(?i)(coffee|order|help)\"}}, {\"id\": \"ask_details\", \"description\": \"Agent asks for more details\", \"after\": [\"acknowledge\"], \"assert\": {\"type\": \"regex\", \"value\": \"(?i)(size|type|what)\"}}], \"max_turns\": 5}`\n\n\terr = os.WriteFile(inputFile, []byte(testCase), 0644)\n\trequire.NoError(t, err, \"Failed to write test file\")\n\n\t// Run test\n\topts := &agenttest.Options{\n\t\tInput: inputFile,\n\t\tAgentID: \"tests.dynamic-test-agent\",\n\t\tVerbose: true,\n\t\tInputMode: agenttest.InputModeFile,\n\t}\n\topts = agenttest.MergeOptions(opts, agenttest.DefaultOptions())\n\n\trunner := agenttest.NewRunner(opts)\n\treport, err := runner.Run()\n\n\trequire.NoError(t, err, \"Runner should not return error\")\n\trequire.NotNil(t, report, \"Report should not be nil\")\n\n\tt.Logf(\"Total: %d, Passed: %d, Failed: %d\",\n\t\treport.Summary.Total, report.Summary.Passed, report.Summary.Failed)\n}\n\n// TestDynamicRunner_OptionalCheckpoint tests optional checkpoint behavior\nfunc TestDynamicRunner_OptionalCheckpoint(t *testing.T) {\n\ttest.Prepare(t, config.Conf)\n\tdefer test.Clean()\n\n\t// Load agents\n\terr := agent.Load(config.Conf)\n\trequire.NoError(t, err, \"Failed to load agents\")\n\n\ttmpDir := t.TempDir()\n\tinputFile := filepath.Join(tmpDir, \"dynamic-inputs.jsonl\")\n\n\t// Test with one required and one optional checkpoint (JSONL must be single line)\n\ttestCase := `{\"id\": \"optional-checkpoint-test\", \"name\": \"Test with Optional Checkpoint\", \"input\": \"Hello\", \"simulator\": {\"use\": \"tests.simulator-agent\", \"options\": {\"metadata\": {\"persona\": \"Simple customer\", \"goal\": \"Get a greeting response\"}}}, \"checkpoints\": [{\"id\": \"greeting_response\", \"description\": \"Agent responds with greeting\", \"assert\": {\"type\": \"regex\", \"value\": \"(?i)(hello|hi|help)\"}}, {\"id\": \"special_offer\", \"description\": \"Agent mentions special offer (optional)\", \"required\": false, \"assert\": {\"type\": \"contains\", \"value\": \"special offer\"}}], \"max_turns\": 3}`\n\n\terr = os.WriteFile(inputFile, []byte(testCase), 0644)\n\trequire.NoError(t, err, \"Failed to write test file\")\n\n\t// Run test\n\topts := &agenttest.Options{\n\t\tInput: inputFile,\n\t\tAgentID: \"tests.dynamic-test-agent\",\n\t\tVerbose: true,\n\t\tInputMode: agenttest.InputModeFile,\n\t}\n\topts = agenttest.MergeOptions(opts, agenttest.DefaultOptions())\n\n\trunner := agenttest.NewRunner(opts)\n\treport, err := runner.Run()\n\n\trequire.NoError(t, err, \"Runner should not return error\")\n\trequire.NotNil(t, report, \"Report should not be nil\")\n\n\t// Test should pass even if optional checkpoint is not reached\n\tt.Logf(\"Total: %d, Passed: %d, Failed: %d\",\n\t\treport.Summary.Total, report.Summary.Passed, report.Summary.Failed)\n\n\t// If the required checkpoint is reached, the test should pass\n\tif len(report.Results) > 0 && report.Results[0].Metadata != nil {\n\t\tif checkpoints, ok := report.Results[0].Metadata[\"checkpoints\"].(map[string]*agenttest.CheckpointResult); ok {\n\t\t\tfor id, cp := range checkpoints {\n\t\t\t\tt.Logf(\"Checkpoint [%s]: reached=%v, required=%v\", id, cp.Reached, cp.Required)\n\t\t\t}\n\t\t}\n\t}\n}\n\n// TestDynamicRunner_MaxTurnsExceeded tests behavior when max turns is exceeded\nfunc TestDynamicRunner_MaxTurnsExceeded(t *testing.T) {\n\ttest.Prepare(t, config.Conf)\n\tdefer test.Clean()\n\n\t// Load agents\n\terr := agent.Load(config.Conf)\n\trequire.NoError(t, err, \"Failed to load agents\")\n\n\ttmpDir := t.TempDir()\n\tinputFile := filepath.Join(tmpDir, \"dynamic-inputs.jsonl\")\n\n\t// Test case with impossible checkpoint and low max_turns (JSONL must be single line)\n\ttestCase := `{\"id\": \"max-turns-test\", \"name\": \"Test Max Turns Exceeded\", \"input\": \"Hello\", \"simulator\": {\"use\": \"tests.simulator-agent\", \"options\": {\"metadata\": {\"persona\": \"Persistent customer\", \"goal\": \"Keep talking\"}}}, \"checkpoints\": [{\"id\": \"impossible\", \"description\": \"This checkpoint will never be reached\", \"assert\": {\"type\": \"contains\", \"value\": \"IMPOSSIBLE_STRING_NEVER_APPEARS_12345\"}}], \"max_turns\": 2}`\n\n\terr = os.WriteFile(inputFile, []byte(testCase), 0644)\n\trequire.NoError(t, err, \"Failed to write test file\")\n\n\t// Run test\n\topts := &agenttest.Options{\n\t\tInput: inputFile,\n\t\tAgentID: \"tests.dynamic-test-agent\",\n\t\tVerbose: true,\n\t\tInputMode: agenttest.InputModeFile,\n\t}\n\topts = agenttest.MergeOptions(opts, agenttest.DefaultOptions())\n\n\trunner := agenttest.NewRunner(opts)\n\treport, err := runner.Run()\n\n\trequire.NoError(t, err, \"Runner should not return error\")\n\trequire.NotNil(t, report, \"Report should not be nil\")\n\n\t// Test should fail due to max turns exceeded or goal achieved without checkpoints\n\tassert.Equal(t, 1, report.Summary.Failed, \"Test should fail\")\n\n\tif len(report.Results) > 0 {\n\t\tresult := report.Results[0]\n\t\tassert.Equal(t, agenttest.StatusFailed, result.Status, \"Status should be failed\")\n\t\t// Either max turns exceeded or simulator signaled goal achieved without checkpoints\n\t\tvalidError := strings.Contains(result.Error, \"max turns\") ||\n\t\t\tstrings.Contains(result.Error, \"not all required checkpoints reached\")\n\t\tassert.True(t, validError, \"Error should mention max turns or checkpoints not reached, got: %s\", result.Error)\n\t\tt.Logf(\"Error (expected): %s\", result.Error)\n\t}\n}\n\n// TestDynamicRunner_CheckpointOrdering tests that checkpoint ordering is enforced\nfunc TestDynamicRunner_CheckpointOrderingEnforced(t *testing.T) {\n\ttest.Prepare(t, config.Conf)\n\tdefer test.Clean()\n\n\t// Load agents\n\terr := agent.Load(config.Conf)\n\trequire.NoError(t, err, \"Failed to load agents\")\n\n\ttmpDir := t.TempDir()\n\tinputFile := filepath.Join(tmpDir, \"dynamic-inputs.jsonl\")\n\n\t// Test case with ordered checkpoints (JSONL must be single line)\n\ttestCase := `{\"id\": \"ordered-checkpoints\", \"name\": \"Test Checkpoint Ordering\", \"input\": \"I want to order coffee\", \"simulator\": {\"use\": \"tests.simulator-agent\", \"options\": {\"metadata\": {\"persona\": \"Customer ordering step by step\", \"goal\": \"Complete coffee order following the flow\"}}}, \"checkpoints\": [{\"id\": \"step1_greeting\", \"description\": \"Agent greets\", \"assert\": {\"type\": \"regex\", \"value\": \"(?i)(hello|hi|help|order)\"}}, {\"id\": \"step2_size\", \"description\": \"Agent asks about size\", \"after\": [\"step1_greeting\"], \"assert\": {\"type\": \"regex\", \"value\": \"(?i)size\"}}, {\"id\": \"step3_confirm\", \"description\": \"Agent confirms\", \"after\": [\"step2_size\"], \"assert\": {\"type\": \"regex\", \"value\": \"(?i)confirm\"}}], \"max_turns\": 10}`\n\n\terr = os.WriteFile(inputFile, []byte(testCase), 0644)\n\trequire.NoError(t, err, \"Failed to write test file\")\n\n\t// Run test\n\topts := &agenttest.Options{\n\t\tInput: inputFile,\n\t\tAgentID: \"tests.dynamic-test-agent\",\n\t\tVerbose: true,\n\t\tInputMode: agenttest.InputModeFile,\n\t}\n\topts = agenttest.MergeOptions(opts, agenttest.DefaultOptions())\n\n\trunner := agenttest.NewRunner(opts)\n\treport, err := runner.Run()\n\n\trequire.NoError(t, err, \"Runner should not return error\")\n\trequire.NotNil(t, report, \"Report should not be nil\")\n\n\tt.Logf(\"Total: %d, Passed: %d, Failed: %d\",\n\t\treport.Summary.Total, report.Summary.Passed, report.Summary.Failed)\n\n\t// Log checkpoint order\n\tif len(report.Results) > 0 && report.Results[0].Metadata != nil {\n\t\tif checkpoints, ok := report.Results[0].Metadata[\"checkpoints\"].(map[string]*agenttest.CheckpointResult); ok {\n\t\t\tfor id, cp := range checkpoints {\n\t\t\t\tt.Logf(\"Checkpoint [%s]: reached=%v, at_turn=%d\", id, cp.Reached, cp.ReachedAtTurn)\n\t\t\t}\n\t\t}\n\t}\n}\n" }, { "path": "agent/test/dynamic_runner.go", "content": "package test\n\nimport (\n\t\"fmt\"\n\t\"time\"\n\n\tjsoniter \"github.com/json-iterator/go\"\n\tgoutext \"github.com/yaoapp/gou/text\"\n\t\"github.com/yaoapp/yao/agent/assistant\"\n\t\"github.com/yaoapp/yao/agent/context\"\n)\n\n// DynamicRunner handles dynamic (simulator-driven) test execution\ntype DynamicRunner struct {\n\topts *Options\n\toutput *OutputWriter\n\tasserter *Asserter\n}\n\n// NewDynamicRunner creates a new dynamic runner\nfunc NewDynamicRunner(opts *Options) *DynamicRunner {\n\treturn &DynamicRunner{\n\t\topts: opts,\n\t\toutput: NewOutputWriter(opts.Verbose),\n\t\tasserter: NewAsserter(),\n\t}\n}\n\n// RunDynamic executes a dynamic test case\nfunc (r *DynamicRunner) RunDynamic(ast *assistant.Assistant, tc *Case, agentID string) *DynamicResult {\n\tstartTime := time.Now()\n\n\tresult := &DynamicResult{\n\t\tID: tc.ID,\n\t\tTurns: make([]*TurnResult, 0),\n\t\tCheckpoints: make(map[string]*CheckpointResult),\n\t}\n\n\t// Initialize checkpoints\n\tfor _, cp := range tc.Checkpoints {\n\t\tresult.Checkpoints[cp.ID] = &CheckpointResult{\n\t\t\tID: cp.ID,\n\t\t\tReached: false,\n\t\t\tRequired: cp.IsRequired(),\n\t\t}\n\t}\n\n\t// Get simulator agent\n\tsimAST, err := assistant.Get(tc.Simulator.Use)\n\tif err != nil {\n\t\tresult.Status = StatusError\n\t\tresult.Error = fmt.Sprintf(\"failed to get simulator agent: %s\", err.Error())\n\t\tresult.DurationMs = time.Since(startTime).Milliseconds()\n\t\treturn result\n\t}\n\n\t// Get configuration\n\tmaxTurns := tc.GetMaxTurns()\n\ttimeout := tc.GetTimeout(r.opts.Timeout)\n\n\t// Build simulator metadata\n\tsimMetadata := make(map[string]interface{})\n\tif tc.Simulator.Options != nil && tc.Simulator.Options.Metadata != nil {\n\t\tfor k, v := range tc.Simulator.Options.Metadata {\n\t\t\tsimMetadata[k] = v\n\t\t}\n\t}\n\n\t// Conversation history\n\tmessages := make([]context.Message, 0)\n\n\t// Get initial input if provided\n\tinitialMessages, err := tc.GetMessages()\n\tif err == nil && len(initialMessages) > 0 {\n\t\tmessages = append(messages, initialMessages...)\n\t}\n\n\t// Output dynamic test start\n\tif r.opts.Verbose {\n\t\tr.output.Verbose(\"Dynamic test: %s (max %d turns)\", tc.ID, maxTurns)\n\t}\n\n\t// Use consistent chatID across all turns to preserve session state (ctx.memory.chat)\n\t// Priority: context config > generated ID\n\tchatID := fmt.Sprintf(\"dynamic-%s\", tc.ID)\n\tif r.opts.ContextData != nil && r.opts.ContextData.ChatID != \"\" {\n\t\tchatID = r.opts.ContextData.ChatID\n\t}\n\n\t// Conversation loop\n\tfor turn := 1; turn <= maxTurns; turn++ {\n\t\tturnStart := time.Now()\n\t\tturnResult := &TurnResult{Turn: turn}\n\n\t\t// Check timeout\n\t\tif time.Since(startTime) > timeout {\n\t\t\tresult.Status = StatusTimeout\n\t\t\tresult.Error = fmt.Sprintf(\"timeout after %s\", timeout)\n\t\t\tresult.DurationMs = time.Since(startTime).Milliseconds()\n\t\t\tresult.TotalTurns = turn - 1\n\t\t\treturn result\n\t\t}\n\n\t\t// For turns after the first, get input from simulator\n\t\tif turn > 1 || len(messages) == 0 {\n\t\t\tsimInput := r.buildSimulatorInput(tc, messages, result, turn, maxTurns, simMetadata)\n\t\t\tsimOutput, err := r.callSimulator(simAST, tc, simInput)\n\t\t\tif err != nil {\n\t\t\t\tturnResult.Error = fmt.Sprintf(\"simulator error: %s\", err.Error())\n\t\t\t\tresult.Turns = append(result.Turns, turnResult)\n\t\t\t\tresult.Status = StatusError\n\t\t\t\tresult.Error = turnResult.Error\n\t\t\t\tresult.DurationMs = time.Since(startTime).Milliseconds()\n\t\t\t\tresult.TotalTurns = turn\n\t\t\t\treturn result\n\t\t\t}\n\n\t\t\t// Check if goal achieved\n\t\t\tif simOutput.GoalAchieved {\n\t\t\t\tif r.opts.Verbose {\n\t\t\t\t\tr.output.Verbose(\"Turn %d: Simulator signaled goal achieved\", turn)\n\t\t\t\t}\n\n\t\t\t\t// Check if all required checkpoints reached\n\t\t\t\tif r.allRequiredCheckpointsReached(result) {\n\t\t\t\t\tresult.Status = StatusPassed\n\t\t\t\t} else {\n\t\t\t\t\tresult.Status = StatusFailed\n\t\t\t\t\tresult.Error = \"simulator signaled goal achieved but not all required checkpoints reached\"\n\t\t\t\t}\n\t\t\t\tresult.DurationMs = time.Since(startTime).Milliseconds()\n\t\t\t\tresult.TotalTurns = turn - 1\n\t\t\t\treturn result\n\t\t\t}\n\n\t\t\t// Add user message\n\t\t\tuserMessage := context.Message{\n\t\t\t\tRole: context.RoleUser,\n\t\t\t\tContent: simOutput.Message,\n\t\t\t}\n\t\t\tmessages = append(messages, userMessage)\n\t\t\tturnResult.Input = simOutput.Message\n\n\t\t\tif r.opts.Verbose {\n\t\t\t\tr.output.Verbose(\"Turn %d: User: %s\", turn, truncateOutput(simOutput.Message, 50))\n\t\t\t}\n\t\t} else {\n\t\t\t// Use initial input for first turn\n\t\t\tif len(messages) > 0 {\n\t\t\t\tlastMsg := messages[len(messages)-1]\n\t\t\t\tturnResult.Input = lastMsg.Content\n\t\t\t\tif r.opts.Verbose {\n\t\t\t\t\tr.output.Verbose(\"Turn %d: User: %s\", turn, truncateOutput(lastMsg.Content, 50))\n\t\t\t\t}\n\t\t\t}\n\t\t}\n\n\t\t// Call target agent\n\t\t// Use consistent chatID across all turns to preserve session state (ctx.memory.chat)\n\t\tctx := NewTestContextFromOptions(\n\t\t\tchatID,\n\t\t\tagentID,\n\t\t\tr.opts,\n\t\t\ttc,\n\t\t)\n\n\t\topts := buildContextOptions(tc, r.opts)\n\t\tresponse, err := ast.Stream(ctx, messages, opts)\n\t\tctx.Release()\n\n\t\tif err != nil {\n\t\t\tturnResult.Error = err.Error()\n\t\t\tturnResult.DurationMs = time.Since(turnStart).Milliseconds()\n\t\t\tresult.Turns = append(result.Turns, turnResult)\n\t\t\tresult.Status = StatusError\n\t\t\tresult.Error = fmt.Sprintf(\"agent error at turn %d: %s\", turn, err.Error())\n\t\t\tresult.DurationMs = time.Since(startTime).Milliseconds()\n\t\t\tresult.TotalTurns = turn\n\t\t\treturn result\n\t\t}\n\n\t\t// Extract output (summary for display and conversation history)\n\t\toutput := extractOutput(response)\n\t\tturnResult.Output = output\n\t\tturnResult.DurationMs = time.Since(turnStart).Milliseconds()\n\n\t\t// Store full response for reporting\n\t\tturnResult.Response = buildTurnResponse(response)\n\n\t\tif r.opts.Verbose {\n\t\t\tr.output.Verbose(\"Turn %d: Agent: %s\", turn, truncateOutput(output, 50))\n\t\t}\n\n\t\t// Add assistant response to messages, including tool calls if any\n\t\tmessages = appendAssistantMessages(messages, response)\n\n\t\t// Check checkpoints against this response (including tool results)\n\t\treachedIDs := r.checkCheckpoints(tc.Checkpoints, response, result)\n\t\tturnResult.CheckpointsReached = reachedIDs\n\n\t\tif r.opts.Verbose && len(reachedIDs) > 0 {\n\t\t\tfor _, id := range reachedIDs {\n\t\t\t\tr.output.Verbose(\" ✓ checkpoint: %s\", id)\n\t\t\t}\n\t\t}\n\n\t\tresult.Turns = append(result.Turns, turnResult)\n\n\t\t// Check if all required checkpoints reached\n\t\tif r.allRequiredCheckpointsReached(result) {\n\t\t\tresult.Status = StatusPassed\n\t\t\tresult.DurationMs = time.Since(startTime).Milliseconds()\n\t\t\tresult.TotalTurns = turn\n\t\t\treturn result\n\t\t}\n\t}\n\n\t// Max turns exceeded\n\tresult.Status = StatusFailed\n\tresult.Error = fmt.Sprintf(\"max turns (%d) exceeded without reaching all checkpoints\", maxTurns)\n\tresult.DurationMs = time.Since(startTime).Milliseconds()\n\tresult.TotalTurns = maxTurns\n\treturn result\n}\n\n// buildSimulatorInput builds the input for the simulator agent\nfunc (r *DynamicRunner) buildSimulatorInput(\n\ttc *Case,\n\tmessages []context.Message,\n\tresult *DynamicResult,\n\tturn, maxTurns int,\n\tmetadata map[string]interface{},\n) *SimulatorInput {\n\tinput := &SimulatorInput{\n\t\tConversation: messages,\n\t\tTurnNumber: turn,\n\t\tMaxTurns: maxTurns,\n\t}\n\n\t// Extract persona and goal from metadata\n\tif persona, ok := metadata[\"persona\"].(string); ok {\n\t\tinput.Persona = persona\n\t}\n\tif goal, ok := metadata[\"goal\"].(string); ok {\n\t\tinput.Goal = goal\n\t}\n\n\t// Build checkpoint lists\n\tinput.CheckpointsReached = make([]string, 0)\n\tinput.CheckpointsPending = make([]string, 0)\n\tfor id, cp := range result.Checkpoints {\n\t\tif cp.Reached {\n\t\t\tinput.CheckpointsReached = append(input.CheckpointsReached, id)\n\t\t} else {\n\t\t\tinput.CheckpointsPending = append(input.CheckpointsPending, id)\n\t\t}\n\t}\n\n\t// Store extra metadata\n\tinput.Extra = make(map[string]interface{})\n\tfor k, v := range metadata {\n\t\tif k != \"persona\" && k != \"goal\" {\n\t\t\tinput.Extra[k] = v\n\t\t}\n\t}\n\n\treturn input\n}\n\n// callSimulator calls the simulator agent and parses the response\nfunc (r *DynamicRunner) callSimulator(simAST *assistant.Assistant, tc *Case, input *SimulatorInput) (*SimulatorOutput, error) {\n\t// Create context\n\tenv := NewEnvironment(\"\", \"\")\n\tctx := NewTestContext(\"simulator\", tc.Simulator.Use, env)\n\tdefer ctx.Release()\n\n\t// Build options - skip history and trace\n\topts := &context.Options{\n\t\tSkip: &context.Skip{\n\t\t\tHistory: true,\n\t\t\tTrace: true,\n\t\t\tOutput: true,\n\t\t},\n\t\tMetadata: map[string]interface{}{\n\t\t\t\"test_mode\": \"simulator\",\n\t\t},\n\t}\n\n\t// Override connector if specified\n\tif tc.Simulator.Options != nil && tc.Simulator.Options.Connector != \"\" {\n\t\topts.Connector = tc.Simulator.Options.Connector\n\t}\n\n\t// Build message\n\tinputJSON, err := jsoniter.Marshal(input)\n\tif err != nil {\n\t\treturn nil, fmt.Errorf(\"failed to marshal simulator input: %w\", err)\n\t}\n\n\tmessages := []context.Message{{\n\t\tRole: context.RoleUser,\n\t\tContent: string(inputJSON),\n\t}}\n\n\t// Call simulator\n\tresponse, err := simAST.Stream(ctx, messages, opts)\n\tif err != nil {\n\t\treturn nil, fmt.Errorf(\"simulator agent error: %w\", err)\n\t}\n\n\t// Parse response\n\treturn r.parseSimulatorResponse(response)\n}\n\n// parseSimulatorResponse parses the simulator agent's response\nfunc (r *DynamicRunner) parseSimulatorResponse(response *context.Response) (*SimulatorOutput, error) {\n\tif response == nil || response.Completion == nil {\n\t\treturn nil, fmt.Errorf(\"empty response from simulator\")\n\t}\n\n\t// Extract content\n\tcontent := response.Completion.Content\n\tif content == nil {\n\t\treturn nil, fmt.Errorf(\"no content in simulator response\")\n\t}\n\n\t// Convert to string\n\tvar text string\n\tswitch v := content.(type) {\n\tcase string:\n\t\ttext = v\n\tdefault:\n\t\tdata, err := jsoniter.Marshal(content)\n\t\tif err != nil {\n\t\t\treturn nil, fmt.Errorf(\"failed to marshal content: %w\", err)\n\t\t}\n\t\ttext = string(data)\n\t}\n\n\t// Use goutext.ExtractJSON for fault-tolerant parsing\n\tparsed := goutext.ExtractJSON(text)\n\tif parsed == nil {\n\t\t// Try to use the text as the message directly\n\t\treturn &SimulatorOutput{\n\t\t\tMessage: text,\n\t\t\tGoalAchieved: false,\n\t\t}, nil\n\t}\n\n\t// Parse as SimulatorOutput\n\toutput := &SimulatorOutput{}\n\tif m, ok := parsed.(map[string]interface{}); ok {\n\t\tif msg, ok := m[\"message\"].(string); ok {\n\t\t\toutput.Message = msg\n\t\t}\n\t\tif achieved, ok := m[\"goal_achieved\"].(bool); ok {\n\t\t\toutput.GoalAchieved = achieved\n\t\t}\n\t\tif reasoning, ok := m[\"reasoning\"].(string); ok {\n\t\t\toutput.Reasoning = reasoning\n\t\t}\n\t}\n\n\tif output.Message == \"\" {\n\t\treturn nil, fmt.Errorf(\"simulator returned empty message\")\n\t}\n\n\treturn output, nil\n}\n\n// checkCheckpoints validates checkpoints against current response\n// It checks both the completion content and tool results for comprehensive validation\nfunc (r *DynamicRunner) checkCheckpoints(checkpoints []*Checkpoint, response *context.Response, result *DynamicResult) []string {\n\treachedIDs := make([]string, 0)\n\n\t// Build combined output for checkpoint validation\n\t// This includes both content and tool result messages\n\tcombinedOutput := buildCombinedOutput(response)\n\n\t// Set response on asserter for tool-related assertions\n\tr.asserter.WithResponse(response)\n\n\tfor _, cp := range checkpoints {\n\t\tcpResult := result.Checkpoints[cp.ID]\n\t\tif cpResult.Reached {\n\t\t\tcontinue // Already reached\n\t\t}\n\n\t\t// Check \"after\" constraint\n\t\tif len(cp.After) > 0 {\n\t\t\tallAfterReached := true\n\t\t\tfor _, afterID := range cp.After {\n\t\t\t\tif afterResult, ok := result.Checkpoints[afterID]; ok {\n\t\t\t\t\tif !afterResult.Reached {\n\t\t\t\t\t\tallAfterReached = false\n\t\t\t\t\t\tbreak\n\t\t\t\t\t}\n\t\t\t\t}\n\t\t\t}\n\t\t\tif !allAfterReached {\n\t\t\t\tcontinue // Dependencies not met\n\t\t\t}\n\t\t}\n\n\t\t// Validate using asserter against combined output with full details\n\t\ttempCase := &Case{Assert: cp.Assert}\n\t\tassertResult := r.asserter.ValidateWithDetails(tempCase, combinedOutput)\n\n\t\tif assertResult.Passed {\n\t\t\tcpResult.Reached = true\n\t\t\tcpResult.Passed = true\n\t\t\tcpResult.ReachedAtTurn = len(result.Turns) + 1\n\t\t\tcpResult.Message = assertResult.Message\n\t\t\treachedIDs = append(reachedIDs, cp.ID)\n\t\t} else {\n\t\t\t// Store failure message for debugging (but don't mark as failed yet - it might pass in a later turn)\n\t\t\tif cpResult.Message == \"\" {\n\t\t\t\tcpResult.Message = assertResult.Message\n\t\t\t}\n\t\t}\n\n\t\t// Store agent validation details if this is an agent assertion\n\t\tif isAgentAssertion(cp.Assert) {\n\t\t\t// Extract criteria from assertion value\n\t\t\tvar criteria string\n\t\t\tif assertMap, ok := cp.Assert.(map[string]interface{}); ok {\n\t\t\t\tif c, ok := assertMap[\"value\"].(string); ok {\n\t\t\t\t\tcriteria = c\n\t\t\t\t}\n\t\t\t}\n\n\t\t\tcpResult.AgentValidation = &AgentValidationResult{\n\t\t\t\tPassed: assertResult.Passed,\n\t\t\t\tCriteria: criteria,\n\t\t\t\tInput: combinedOutput, // Content sent to validator for checking\n\t\t\t}\n\n\t\t\t// Extract reason and store full response from validator\n\t\t\tif assertResult.Expected != nil {\n\t\t\t\tif validatorResponse, ok := assertResult.Expected.(map[string]interface{}); ok {\n\t\t\t\t\tif reason, ok := validatorResponse[\"reason\"].(string); ok {\n\t\t\t\t\t\tcpResult.AgentValidation.Reason = reason\n\t\t\t\t\t}\n\t\t\t\t\t// Store the full validator response\n\t\t\t\t\tcpResult.AgentValidation.Response = validatorResponse\n\t\t\t\t}\n\t\t\t}\n\t\t}\n\t}\n\n\treturn reachedIDs\n}\n\n// isAgentAssertion checks if the assertion is an agent-based assertion\nfunc isAgentAssertion(assert interface{}) bool {\n\tif assertMap, ok := assert.(map[string]interface{}); ok {\n\t\tif assertType, ok := assertMap[\"type\"].(string); ok {\n\t\t\treturn assertType == \"agent\"\n\t\t}\n\t}\n\treturn false\n}\n\n// truncateForReport truncates content for report output\nfunc truncateForReport(content interface{}, maxLen int) interface{} {\n\tif content == nil {\n\t\treturn nil\n\t}\n\n\tstr, ok := content.(string)\n\tif !ok {\n\t\treturn content\n\t}\n\n\tif len(str) <= maxLen {\n\t\treturn str\n\t}\n\n\treturn str[:maxLen] + \"... (truncated)\"\n}\n\n// buildCombinedOutput builds a combined output string from response\n// that includes both completion content and tool result messages\nfunc buildCombinedOutput(response *context.Response) string {\n\tif response == nil {\n\t\treturn \"\"\n\t}\n\n\tvar parts []string\n\n\t// Add completion content\n\tif response.Completion != nil && response.Completion.Content != nil {\n\t\tif content := extractContentString(response.Completion.Content); content != \"\" {\n\t\t\tparts = append(parts, content)\n\t\t}\n\t}\n\n\t// Add tool result messages\n\tif len(response.Tools) > 0 {\n\t\tfor _, tool := range response.Tools {\n\t\t\tif tool.Result != nil {\n\t\t\t\t// Try to extract message from result\n\t\t\t\tif resultMap, ok := tool.Result.(map[string]interface{}); ok {\n\t\t\t\t\tif msg, exists := resultMap[\"message\"]; exists && msg != nil {\n\t\t\t\t\t\tif msgStr, ok := msg.(string); ok && msgStr != \"\" {\n\t\t\t\t\t\t\tparts = append(parts, msgStr)\n\t\t\t\t\t\t}\n\t\t\t\t\t}\n\t\t\t\t}\n\t\t\t}\n\t\t}\n\t}\n\n\t// Join all parts with newline for comprehensive matching\n\treturn joinNonEmpty(parts, \"\\n\")\n}\n\n// extractContentString extracts string content from various types\nfunc extractContentString(content interface{}) string {\n\tif content == nil {\n\t\treturn \"\"\n\t}\n\n\tswitch v := content.(type) {\n\tcase string:\n\t\treturn v\n\tcase []interface{}:\n\t\t// Handle array content (e.g., multimodal content)\n\t\tvar texts []string\n\t\tfor _, item := range v {\n\t\t\tif m, ok := item.(map[string]interface{}); ok {\n\t\t\t\tif text, ok := m[\"text\"].(string); ok {\n\t\t\t\t\ttexts = append(texts, text)\n\t\t\t\t}\n\t\t\t}\n\t\t}\n\t\treturn joinNonEmpty(texts, \"\\n\")\n\tdefault:\n\t\t// Try to marshal to string\n\t\tif data, err := jsoniter.MarshalToString(content); err == nil {\n\t\t\treturn data\n\t\t}\n\t\treturn \"\"\n\t}\n}\n\n// joinNonEmpty joins non-empty strings with separator\nfunc joinNonEmpty(parts []string, sep string) string {\n\tvar nonEmpty []string\n\tfor _, p := range parts {\n\t\tif p != \"\" {\n\t\t\tnonEmpty = append(nonEmpty, p)\n\t\t}\n\t}\n\tif len(nonEmpty) == 0 {\n\t\treturn \"\"\n\t}\n\tresult := nonEmpty[0]\n\tfor i := 1; i < len(nonEmpty); i++ {\n\t\tresult += sep + nonEmpty[i]\n\t}\n\treturn result\n}\n\n// allRequiredCheckpointsReached checks if all required checkpoints are reached\nfunc (r *DynamicRunner) allRequiredCheckpointsReached(result *DynamicResult) bool {\n\tfor _, cp := range result.Checkpoints {\n\t\tif cp.Required && !cp.Reached {\n\t\t\treturn false\n\t\t}\n\t}\n\treturn true\n}\n\n// buildTurnResponse builds a TurnResponse from the agent response\nfunc buildTurnResponse(response *context.Response) *TurnResponse {\n\tif response == nil {\n\t\treturn nil\n\t}\n\n\ttr := &TurnResponse{}\n\n\t// Extract completion content\n\tif response.Completion != nil {\n\t\ttr.Content = response.Completion.Content\n\n\t\t// Extract tool calls from completion\n\t\tif len(response.Completion.ToolCalls) > 0 {\n\t\t\tfor _, tc := range response.Completion.ToolCalls {\n\t\t\t\ttr.ToolCalls = append(tr.ToolCalls, ToolCallInfo{\n\t\t\t\t\tTool: tc.Function.Name,\n\t\t\t\t\tArguments: tc.Function.Arguments,\n\t\t\t\t})\n\t\t\t}\n\t\t}\n\t}\n\n\t// Add tool results\n\tif len(response.Tools) > 0 {\n\t\t// If we already have tool calls from completion, match results\n\t\tif len(tr.ToolCalls) > 0 {\n\t\t\tfor i, toolResult := range response.Tools {\n\t\t\t\tif i < len(tr.ToolCalls) {\n\t\t\t\t\ttr.ToolCalls[i].Result = toolResult.Result\n\t\t\t\t}\n\t\t\t}\n\t\t} else {\n\t\t\t// Create tool call entries from results\n\t\t\tfor _, toolResult := range response.Tools {\n\t\t\t\ttr.ToolCalls = append(tr.ToolCalls, ToolCallInfo{\n\t\t\t\t\tTool: toolResult.Tool,\n\t\t\t\t\tArguments: toolResult.Arguments,\n\t\t\t\t\tResult: toolResult.Result,\n\t\t\t\t})\n\t\t\t}\n\t\t}\n\t}\n\n\t// Extract Next hook data\n\tif response.Next != nil && !isEmptyValue(response.Next) {\n\t\ttr.Next = response.Next\n\t}\n\n\treturn tr\n}\n\n// appendAssistantMessages appends assistant messages to the conversation history\n// including tool calls and tool results if present\nfunc appendAssistantMessages(messages []context.Message, response *context.Response) []context.Message {\n\tif response == nil {\n\t\treturn messages\n\t}\n\n\t// Check if there are tool calls in the completion\n\thasToolCalls := response.Completion != nil && len(response.Completion.ToolCalls) > 0\n\n\tif hasToolCalls {\n\t\t// Add assistant message with tool calls\n\t\tassistantMsg := context.Message{\n\t\t\tRole: context.RoleAssistant,\n\t\t\tToolCalls: response.Completion.ToolCalls,\n\t\t}\n\t\t// Include content if present\n\t\tif response.Completion.Content != nil && !isEmptyValue(response.Completion.Content) {\n\t\t\tassistantMsg.Content = response.Completion.Content\n\t\t}\n\t\tmessages = append(messages, assistantMsg)\n\n\t\t// Add tool result messages for each tool call\n\t\tfor i, tc := range response.Completion.ToolCalls {\n\t\t\ttoolCallID := tc.ID\n\t\t\tvar resultContent string\n\n\t\t\t// Get result from response.Tools if available\n\t\t\tif i < len(response.Tools) {\n\t\t\t\tresultJSON, err := jsoniter.MarshalToString(response.Tools[i].Result)\n\t\t\t\tif err == nil {\n\t\t\t\t\tresultContent = resultJSON\n\t\t\t\t} else {\n\t\t\t\t\tresultContent = fmt.Sprintf(\"%v\", response.Tools[i].Result)\n\t\t\t\t}\n\t\t\t} else {\n\t\t\t\tresultContent = \"{}\"\n\t\t\t}\n\n\t\t\tmessages = append(messages, context.Message{\n\t\t\t\tRole: context.RoleTool,\n\t\t\t\tToolCallID: &toolCallID,\n\t\t\t\tContent: resultContent,\n\t\t\t})\n\t\t}\n\t} else {\n\t\t// No tool calls, just add content if present\n\t\tcontent := extractOutput(response)\n\t\tif content != nil && !isEmptyValue(content) {\n\t\t\tmessages = append(messages, context.Message{\n\t\t\t\tRole: context.RoleAssistant,\n\t\t\t\tContent: content,\n\t\t\t})\n\t\t}\n\t}\n\n\treturn messages\n}\n" }, { "path": "agent/test/dynamic_runner_test.go", "content": "package test_test\n\nimport (\n\t\"testing\"\n\t\"time\"\n\n\t\"github.com/stretchr/testify/assert\"\n\t\"github.com/yaoapp/yao/agent\"\n\t\"github.com/yaoapp/yao/agent/test\"\n\t\"github.com/yaoapp/yao/config\"\n\ttestutils \"github.com/yaoapp/yao/test\"\n)\n\nfunc TestCase_IsDynamicMode(t *testing.T) {\n\ttests := []struct {\n\t\tname string\n\t\ttc *test.Case\n\t\texpected bool\n\t}{\n\t\t{\n\t\t\tname: \"standard mode - no simulator\",\n\t\t\ttc: &test.Case{\n\t\t\t\tID: \"T001\",\n\t\t\t\tInput: \"Hello\",\n\t\t\t},\n\t\t\texpected: false,\n\t\t},\n\t\t{\n\t\t\tname: \"standard mode - simulator but no checkpoints\",\n\t\t\ttc: &test.Case{\n\t\t\t\tID: \"T002\",\n\t\t\t\tInput: \"Hello\",\n\t\t\t\tSimulator: &test.Simulator{Use: \"tests.simulator-agent\"},\n\t\t\t},\n\t\t\texpected: false,\n\t\t},\n\t\t{\n\t\t\tname: \"standard mode - checkpoints but no simulator\",\n\t\t\ttc: &test.Case{\n\t\t\t\tID: \"T003\",\n\t\t\t\tInput: \"Hello\",\n\t\t\t\tCheckpoints: []*test.Checkpoint{\n\t\t\t\t\t{ID: \"cp1\", Assert: map[string]interface{}{\"type\": \"contains\", \"value\": \"hi\"}},\n\t\t\t\t},\n\t\t\t},\n\t\t\texpected: false,\n\t\t},\n\t\t{\n\t\t\tname: \"dynamic mode - has both simulator and checkpoints\",\n\t\t\ttc: &test.Case{\n\t\t\t\tID: \"T004\",\n\t\t\t\tSimulator: &test.Simulator{Use: \"tests.simulator-agent\"},\n\t\t\t\tCheckpoints: []*test.Checkpoint{\n\t\t\t\t\t{ID: \"cp1\", Assert: map[string]interface{}{\"type\": \"contains\", \"value\": \"hi\"}},\n\t\t\t\t},\n\t\t\t},\n\t\t\texpected: true,\n\t\t},\n\t}\n\n\tfor _, tt := range tests {\n\t\tt.Run(tt.name, func(t *testing.T) {\n\t\t\tresult := tt.tc.IsDynamicMode()\n\t\t\tassert.Equal(t, tt.expected, result)\n\t\t})\n\t}\n}\n\nfunc TestCase_GetMaxTurns(t *testing.T) {\n\ttests := []struct {\n\t\tname string\n\t\ttc *test.Case\n\t\texpected int\n\t}{\n\t\t{\n\t\t\tname: \"default max turns\",\n\t\t\ttc: &test.Case{ID: \"T001\"},\n\t\t\texpected: 20,\n\t\t},\n\t\t{\n\t\t\tname: \"custom max turns\",\n\t\t\ttc: &test.Case{ID: \"T002\", MaxTurns: 10},\n\t\t\texpected: 10,\n\t\t},\n\t\t{\n\t\t\tname: \"zero max turns uses default\",\n\t\t\ttc: &test.Case{ID: \"T003\", MaxTurns: 0},\n\t\t\texpected: 20,\n\t\t},\n\t}\n\n\tfor _, tt := range tests {\n\t\tt.Run(tt.name, func(t *testing.T) {\n\t\t\tresult := tt.tc.GetMaxTurns()\n\t\t\tassert.Equal(t, tt.expected, result)\n\t\t})\n\t}\n}\n\nfunc TestCheckpoint_IsRequired(t *testing.T) {\n\tboolTrue := true\n\tboolFalse := false\n\n\ttests := []struct {\n\t\tname string\n\t\tcp *test.Checkpoint\n\t\texpected bool\n\t}{\n\t\t{\n\t\t\tname: \"default is required\",\n\t\t\tcp: &test.Checkpoint{ID: \"cp1\"},\n\t\t\texpected: true,\n\t\t},\n\t\t{\n\t\t\tname: \"explicitly required\",\n\t\t\tcp: &test.Checkpoint{ID: \"cp2\", Required: &boolTrue},\n\t\t\texpected: true,\n\t\t},\n\t\t{\n\t\t\tname: \"explicitly not required\",\n\t\t\tcp: &test.Checkpoint{ID: \"cp3\", Required: &boolFalse},\n\t\t\texpected: false,\n\t\t},\n\t}\n\n\tfor _, tt := range tests {\n\t\tt.Run(tt.name, func(t *testing.T) {\n\t\t\tresult := tt.cp.IsRequired()\n\t\t\tassert.Equal(t, tt.expected, result)\n\t\t})\n\t}\n}\n\nfunc TestDynamicResult_ToResult(t *testing.T) {\n\tdr := &test.DynamicResult{\n\t\tID: \"T001\",\n\t\tStatus: test.StatusPassed,\n\t\tTotalTurns: 3,\n\t\tDurationMs: 5000,\n\t\tTurns: []*test.TurnResult{\n\t\t\t{Turn: 1, Input: \"Hello\", Output: \"Hi there!\"},\n\t\t\t{Turn: 2, Input: \"How are you?\", Output: \"I'm doing well!\"},\n\t\t\t{Turn: 3, Input: \"Goodbye\", Output: \"Bye!\"},\n\t\t},\n\t\tCheckpoints: map[string]*test.CheckpointResult{\n\t\t\t\"greet\": {ID: \"greet\", Reached: true, ReachedAtTurn: 1, Required: true},\n\t\t\t\"bye\": {ID: \"bye\", Reached: true, ReachedAtTurn: 3, Required: true},\n\t\t},\n\t}\n\n\tresult := dr.ToResult()\n\n\tassert.Equal(t, \"T001\", result.ID)\n\tassert.Equal(t, test.StatusPassed, result.Status)\n\tassert.Equal(t, int64(5000), result.DurationMs)\n\tassert.Equal(t, \"Hello\", result.Input)\n\tassert.Equal(t, \"Bye!\", result.Output)\n\n\t// Check metadata\n\tassert.NotNil(t, result.Metadata)\n\tassert.Equal(t, \"dynamic\", result.Metadata[\"mode\"])\n\tassert.Equal(t, 3, result.Metadata[\"total_turns\"])\n}\n\nfunc TestDynamicRunner_Integration(t *testing.T) {\n\t// Skip if running in short mode\n\tif testing.Short() {\n\t\tt.Skip(\"skipping integration test in short mode\")\n\t}\n\n\t// Prepare test environment\n\ttestutils.Prepare(t, config.Conf)\n\tdefer testutils.Clean()\n\n\t// Load agents\n\terr := agent.Load(config.Conf)\n\tif err != nil {\n\t\tt.Skipf(\"Failed to load agents: %v\", err)\n\t}\n\n\t// Create a dynamic test case\n\ttc := &test.Case{\n\t\tID: \"dynamic-greeting\",\n\t\tSimulator: &test.Simulator{\n\t\t\tUse: \"tests.simulator-agent\",\n\t\t\tOptions: &test.SimulatorOptions{\n\t\t\t\tMetadata: map[string]interface{}{\n\t\t\t\t\t\"persona\": \"Friendly user\",\n\t\t\t\t\t\"goal\": \"Have a brief greeting exchange\",\n\t\t\t\t},\n\t\t\t},\n\t\t},\n\t\tInput: \"Hello!\",\n\t\tCheckpoints: []*test.Checkpoint{\n\t\t\t{\n\t\t\t\tID: \"greeting\",\n\t\t\t\tDescription: \"Agent should greet back\",\n\t\t\t\tAssert: map[string]interface{}{\n\t\t\t\t\t\"type\": \"regex\",\n\t\t\t\t\t\"value\": \"(?i)(hello|hi|hey|greetings)\",\n\t\t\t\t},\n\t\t\t},\n\t\t},\n\t\tMaxTurns: 3,\n\t}\n\n\t// Verify it's dynamic mode\n\tassert.True(t, tc.IsDynamicMode())\n\n\t// Create runner options\n\topts := &test.Options{\n\t\tVerbose: true,\n\t\tTimeout: 30 * time.Second,\n\t}\n\n\t// Create dynamic runner\n\trunner := test.NewDynamicRunner(opts)\n\tassert.NotNil(t, runner)\n\n\t// Note: Full integration test would require the simulator agent to be loaded\n\t// and would make actual LLM calls. For CI, we test the structure and logic.\n}\n\nfunc TestDynamicRunner_CheckpointOrdering(t *testing.T) {\n\t// Test that checkpoints with \"after\" constraints are properly ordered\n\ttestutils.Prepare(t, config.Conf)\n\tdefer testutils.Clean()\n\n\t// Load agents\n\terr := agent.Load(config.Conf)\n\tif err != nil {\n\t\tt.Skipf(\"Failed to load agents: %v\", err)\n\t}\n\n\t// Create a test case with ordered checkpoints\n\ttc := &test.Case{\n\t\tID: \"ordered-checkpoints\",\n\t\tSimulator: &test.Simulator{\n\t\t\tUse: \"tests.simulator-agent\",\n\t\t\tOptions: &test.SimulatorOptions{\n\t\t\t\tMetadata: map[string]interface{}{\n\t\t\t\t\t\"persona\": \"Customer\",\n\t\t\t\t\t\"goal\": \"Complete a purchase\",\n\t\t\t\t},\n\t\t\t},\n\t\t},\n\t\tCheckpoints: []*test.Checkpoint{\n\t\t\t{\n\t\t\t\tID: \"ask_product\",\n\t\t\t\tDescription: \"Agent asks about product\",\n\t\t\t\tAssert: map[string]interface{}{\n\t\t\t\t\t\"type\": \"contains\",\n\t\t\t\t\t\"value\": \"product\",\n\t\t\t\t},\n\t\t\t},\n\t\t\t{\n\t\t\t\tID: \"confirm_order\",\n\t\t\t\tDescription: \"Agent confirms order\",\n\t\t\t\tAfter: []string{\"ask_product\"},\n\t\t\t\tAssert: map[string]interface{}{\n\t\t\t\t\t\"type\": \"contains\",\n\t\t\t\t\t\"value\": \"confirm\",\n\t\t\t\t},\n\t\t\t},\n\t\t\t{\n\t\t\t\tID: \"complete\",\n\t\t\t\tDescription: \"Order completed\",\n\t\t\t\tAfter: []string{\"confirm_order\"},\n\t\t\t\tAssert: map[string]interface{}{\n\t\t\t\t\t\"type\": \"contains\",\n\t\t\t\t\t\"value\": \"complete\",\n\t\t\t\t},\n\t\t\t},\n\t\t},\n\t\tMaxTurns: 10,\n\t}\n\n\t// Verify checkpoint structure\n\tassert.Len(t, tc.Checkpoints, 3)\n\tassert.Empty(t, tc.Checkpoints[0].After)\n\tassert.Equal(t, []string{\"ask_product\"}, tc.Checkpoints[1].After)\n\tassert.Equal(t, []string{\"confirm_order\"}, tc.Checkpoints[2].After)\n}\n\nfunc TestSimulatorInput_Structure(t *testing.T) {\n\t// Test SimulatorInput structure\n\tinput := &test.SimulatorInput{\n\t\tPersona: \"Test user\",\n\t\tGoal: \"Complete task\",\n\t\tTurnNumber: 3,\n\t\tMaxTurns: 10,\n\t\tCheckpointsReached: []string{\"cp1\", \"cp2\"},\n\t\tCheckpointsPending: []string{\"cp3\"},\n\t\tExtra: map[string]interface{}{\n\t\t\t\"style\": \"formal\",\n\t\t},\n\t}\n\n\tassert.Equal(t, \"Test user\", input.Persona)\n\tassert.Equal(t, \"Complete task\", input.Goal)\n\tassert.Equal(t, 3, input.TurnNumber)\n\tassert.Equal(t, 10, input.MaxTurns)\n\tassert.Len(t, input.CheckpointsReached, 2)\n\tassert.Len(t, input.CheckpointsPending, 1)\n\tassert.Equal(t, \"formal\", input.Extra[\"style\"])\n}\n\nfunc TestSimulatorOutput_Structure(t *testing.T) {\n\t// Test SimulatorOutput structure\n\toutput := &test.SimulatorOutput{\n\t\tMessage: \"I'd like to buy a product\",\n\t\tGoalAchieved: false,\n\t\tReasoning: \"Continuing toward purchase goal\",\n\t}\n\n\tassert.Equal(t, \"I'd like to buy a product\", output.Message)\n\tassert.False(t, output.GoalAchieved)\n\tassert.Equal(t, \"Continuing toward purchase goal\", output.Reasoning)\n}\n" }, { "path": "agent/test/dynamic_types.go", "content": "package test\n\nimport \"github.com/yaoapp/yao/agent/context\"\n\n// DynamicResult represents the result of a dynamic (simulator-driven) test\ntype DynamicResult struct {\n\t// ID is the test case identifier\n\tID string `json:\"id\"`\n\n\t// Status is the overall test status\n\tStatus Status `json:\"status\"`\n\n\t// Turns contains results for each conversation turn\n\tTurns []*TurnResult `json:\"turns\"`\n\n\t// Checkpoints maps checkpoint ID to its result\n\tCheckpoints map[string]*CheckpointResult `json:\"checkpoints\"`\n\n\t// TotalTurns is the number of turns executed\n\tTotalTurns int `json:\"total_turns\"`\n\n\t// DurationMs is the total execution time in milliseconds\n\tDurationMs int64 `json:\"duration_ms\"`\n\n\t// Error contains error message if status is failed/error/timeout\n\tError string `json:\"error,omitempty\"`\n}\n\n// TurnResult represents the result of a single conversation turn\ntype TurnResult struct {\n\t// Turn is the turn number (1-based)\n\tTurn int `json:\"turn\"`\n\n\t// Input is the user message (from simulator or initial input)\n\tInput interface{} `json:\"input\"`\n\n\t// Output is the agent's response (summary for display and conversation history)\n\tOutput interface{} `json:\"output,omitempty\"`\n\n\t// Response is the full agent response including completion and tool results\n\tResponse *TurnResponse `json:\"response,omitempty\"`\n\n\t// CheckpointsReached lists checkpoint IDs reached in this turn\n\tCheckpointsReached []string `json:\"checkpoints_reached,omitempty\"`\n\n\t// DurationMs is the turn execution time in milliseconds\n\tDurationMs int64 `json:\"duration_ms\"`\n\n\t// Error contains error message if this turn failed\n\tError string `json:\"error,omitempty\"`\n}\n\n// TurnResponse contains the full agent response for a turn\ntype TurnResponse struct {\n\t// Content is the text content from LLM completion\n\tContent interface{} `json:\"content,omitempty\"`\n\n\t// ToolCalls contains the tool calls made by the agent\n\tToolCalls []ToolCallInfo `json:\"tool_calls,omitempty\"`\n\n\t// Next is the data returned from Next hook\n\tNext interface{} `json:\"next,omitempty\"`\n}\n\n// ToolCallInfo contains information about a tool call\ntype ToolCallInfo struct {\n\t// Tool is the tool name\n\tTool string `json:\"tool\"`\n\n\t// Arguments are the tool call arguments\n\tArguments interface{} `json:\"arguments,omitempty\"`\n\n\t// Result is the tool execution result\n\tResult interface{} `json:\"result,omitempty\"`\n}\n\n// CheckpointResult represents the result of a checkpoint validation\ntype CheckpointResult struct {\n\t// ID is the checkpoint identifier\n\tID string `json:\"id\"`\n\n\t// Reached indicates if the checkpoint was reached\n\tReached bool `json:\"reached\"`\n\n\t// ReachedAtTurn is the turn number when checkpoint was reached (0 if not reached)\n\tReachedAtTurn int `json:\"reached_at_turn,omitempty\"`\n\n\t// Required indicates if this checkpoint is required\n\tRequired bool `json:\"required\"`\n\n\t// Passed indicates if the checkpoint assertion passed\n\tPassed bool `json:\"passed\"`\n\n\t// Message contains assertion result message\n\tMessage string `json:\"message,omitempty\"`\n\n\t// AgentValidation contains the agent validator's response (for agent assertions)\n\tAgentValidation *AgentValidationResult `json:\"agent_validation,omitempty\"`\n}\n\n// AgentValidationResult contains the result from an agent-based assertion\ntype AgentValidationResult struct {\n\t// Passed indicates if the agent validator determined the assertion passed\n\tPassed bool `json:\"passed\"`\n\n\t// Reason is the explanation from the agent validator\n\tReason string `json:\"reason,omitempty\"`\n\n\t// Criteria is the validation criteria that was checked\n\tCriteria string `json:\"criteria,omitempty\"`\n\n\t// Input is the content that was sent to validator for checking\n\tInput interface{} `json:\"input,omitempty\"`\n\n\t// Response is the raw response from the validator agent\n\tResponse interface{} `json:\"response,omitempty\"`\n}\n\n// SimulatorInput is the input sent to the simulator agent\ntype SimulatorInput struct {\n\t// Persona describes the user being simulated\n\tPersona string `json:\"persona,omitempty\"`\n\n\t// Goal is what the user is trying to achieve\n\tGoal string `json:\"goal,omitempty\"`\n\n\t// Conversation is the message history\n\tConversation []context.Message `json:\"conversation\"`\n\n\t// TurnNumber is the current turn (1-based)\n\tTurnNumber int `json:\"turn_number\"`\n\n\t// MaxTurns is the maximum allowed turns\n\tMaxTurns int `json:\"max_turns\"`\n\n\t// CheckpointsReached lists checkpoint IDs already reached\n\tCheckpointsReached []string `json:\"checkpoints_reached,omitempty\"`\n\n\t// CheckpointsPending lists checkpoint IDs still pending\n\tCheckpointsPending []string `json:\"checkpoints_pending,omitempty\"`\n\n\t// Extra metadata from simulator options\n\tExtra map[string]interface{} `json:\"extra,omitempty\"`\n}\n\n// SimulatorOutput is the expected output from the simulator agent\ntype SimulatorOutput struct {\n\t// Message is the simulated user message\n\tMessage string `json:\"message\"`\n\n\t// GoalAchieved indicates if the user's goal has been accomplished\n\tGoalAchieved bool `json:\"goal_achieved\"`\n\n\t// Reasoning explains the simulator's response strategy\n\tReasoning string `json:\"reasoning,omitempty\"`\n}\n\n// ToResult converts DynamicResult to standard Result for reporting\nfunc (dr *DynamicResult) ToResult() *Result {\n\tresult := &Result{\n\t\tID: dr.ID,\n\t\tStatus: dr.Status,\n\t\tDurationMs: dr.DurationMs,\n\t\tError: dr.Error,\n\t}\n\n\t// Store dynamic-specific data in metadata\n\tresult.Metadata = map[string]interface{}{\n\t\t\"mode\": \"dynamic\",\n\t\t\"total_turns\": dr.TotalTurns,\n\t\t\"turns\": dr.Turns,\n\t\t\"checkpoints\": dr.Checkpoints,\n\t}\n\n\t// Set input from first turn\n\tif len(dr.Turns) > 0 {\n\t\tresult.Input = dr.Turns[0].Input\n\t}\n\n\t// Set output from last turn\n\tif len(dr.Turns) > 0 {\n\t\tresult.Output = dr.Turns[len(dr.Turns)-1].Output\n\t}\n\n\treturn result\n}\n\n// IsDynamicMode checks if a test case should run in dynamic mode\nfunc (tc *Case) IsDynamicMode() bool {\n\treturn tc.Simulator != nil && len(tc.Checkpoints) > 0\n}\n\n// GetMaxTurns returns the max turns for dynamic mode\nfunc (tc *Case) GetMaxTurns() int {\n\tif tc.MaxTurns > 0 {\n\t\treturn tc.MaxTurns\n\t}\n\treturn 20 // Default max turns\n}\n\n// IsRequired returns true if the checkpoint is required\nfunc (cp *Checkpoint) IsRequired() bool {\n\tif cp.Required == nil {\n\t\treturn true // Default to required\n\t}\n\treturn *cp.Required\n}\n" }, { "path": "agent/test/extract.go", "content": "package test\n\nimport (\n\t\"fmt\"\n\t\"os\"\n\t\"path/filepath\"\n\t\"strings\"\n\n\tjsoniter \"github.com/json-iterator/go\"\n)\n\n// ExtractOptions represents options for extracting test results\ntype ExtractOptions struct {\n\t// InputFile is the path to the output JSONL file from test run\n\tInputFile string\n\n\t// OutputDir is the directory to write extracted files (default: same as input file)\n\tOutputDir string\n\n\t// Format is the output format: \"markdown\" (default), \"json\"\n\tFormat string\n}\n\n// Extractor extracts test results to individual files for review\ntype Extractor struct {\n\topts *ExtractOptions\n}\n\n// NewExtractor creates a new extractor\nfunc NewExtractor(opts *ExtractOptions) *Extractor {\n\tif opts.Format == \"\" {\n\t\topts.Format = \"markdown\"\n\t}\n\tif opts.OutputDir == \"\" {\n\t\topts.OutputDir = filepath.Dir(opts.InputFile)\n\t}\n\treturn &Extractor{opts: opts}\n}\n\n// Extract reads the test output file and extracts results to individual files\nfunc (e *Extractor) Extract() ([]string, error) {\n\t// Read the JSONL file\n\tdata, err := os.ReadFile(e.opts.InputFile)\n\tif err != nil {\n\t\treturn nil, fmt.Errorf(\"failed to read input file: %w\", err)\n\t}\n\n\t// Parse the JSON (the output file is a single JSON object, not JSONL)\n\tvar report Report\n\tif err := jsoniter.Unmarshal(data, &report); err != nil {\n\t\treturn nil, fmt.Errorf(\"failed to parse test report: %w\", err)\n\t}\n\n\t// Create output directory if it doesn't exist\n\tif err := os.MkdirAll(e.opts.OutputDir, 0755); err != nil {\n\t\treturn nil, fmt.Errorf(\"failed to create output directory: %w\", err)\n\t}\n\n\tvar extractedFiles []string\n\n\t// Extract each result\n\tfor _, result := range report.Results {\n\t\tvar filename string\n\t\tvar content string\n\n\t\tswitch e.opts.Format {\n\t\tcase \"markdown\":\n\t\t\tfilename = filepath.Join(e.opts.OutputDir, result.ID+\".md\")\n\t\t\tcontent = e.formatMarkdown(result)\n\t\tcase \"json\":\n\t\t\tfilename = filepath.Join(e.opts.OutputDir, result.ID+\".json\")\n\t\t\tjsonBytes, err := jsoniter.MarshalIndent(result, \"\", \" \")\n\t\t\tif err != nil {\n\t\t\t\treturn extractedFiles, fmt.Errorf(\"failed to marshal result %s: %w\", result.ID, err)\n\t\t\t}\n\t\t\tcontent = string(jsonBytes)\n\t\tdefault:\n\t\t\treturn nil, fmt.Errorf(\"unsupported format: %s\", e.opts.Format)\n\t\t}\n\n\t\tif err := os.WriteFile(filename, []byte(content), 0644); err != nil {\n\t\t\treturn extractedFiles, fmt.Errorf(\"failed to write file %s: %w\", filename, err)\n\t\t}\n\n\t\textractedFiles = append(extractedFiles, filename)\n\t}\n\n\treturn extractedFiles, nil\n}\n\n// formatMarkdown formats a single test result as Markdown\nfunc (e *Extractor) formatMarkdown(result *Result) string {\n\tvar sb strings.Builder\n\n\t// Title\n\tsb.WriteString(fmt.Sprintf(\"# %s\\n\\n\", result.ID))\n\n\t// Status badge\n\tswitch result.Status {\n\tcase StatusPassed:\n\t\tsb.WriteString(\"**Status**: ✅ PASSED\\n\\n\")\n\tcase StatusFailed:\n\t\tsb.WriteString(\"**Status**: ❌ FAILED\\n\\n\")\n\tcase StatusError:\n\t\tsb.WriteString(\"**Status**: ⚠️ ERROR\\n\\n\")\n\tcase StatusTimeout:\n\t\tsb.WriteString(\"**Status**: ⏱️ TIMEOUT\\n\\n\")\n\tcase StatusSkipped:\n\t\tsb.WriteString(\"**Status**: ⏭️ SKIPPED\\n\\n\")\n\t}\n\n\t// Duration\n\tsb.WriteString(fmt.Sprintf(\"**Duration**: %dms\\n\\n\", result.DurationMs))\n\n\t// Error (if any)\n\tif result.Error != \"\" {\n\t\tsb.WriteString(\"## Error\\n\\n\")\n\t\tsb.WriteString(\"```\\n\")\n\t\tsb.WriteString(result.Error)\n\t\tsb.WriteString(\"\\n```\\n\\n\")\n\t}\n\n\t// Input\n\tsb.WriteString(\"## Input\\n\\n\")\n\tsb.WriteString(\"```markdown\\n\")\n\tsb.WriteString(formatInputAsString(result.Input))\n\tsb.WriteString(\"\\n```\\n\\n\")\n\n\t// Output\n\tsb.WriteString(\"## Output\\n\\n\")\n\toutput := formatOutputAsString(result.Output)\n\t// Remove markdown code block wrapper if present\n\toutput = strings.TrimPrefix(output, \"```markdown\\n\")\n\toutput = strings.TrimSuffix(output, \"\\n```\")\n\toutput = strings.TrimSuffix(output, \"```\")\n\tsb.WriteString(output)\n\tsb.WriteString(\"\\n\")\n\n\treturn sb.String()\n}\n\n// formatInputAsString converts input to string format\nfunc formatInputAsString(input interface{}) string {\n\tswitch v := input.(type) {\n\tcase string:\n\t\treturn v\n\tcase map[string]interface{}:\n\t\t// Single message format\n\t\tif content, ok := v[\"content\"].(string); ok {\n\t\t\treturn content\n\t\t}\n\t\t// Fallback to JSON\n\t\tjsonBytes, _ := jsoniter.MarshalIndent(v, \"\", \" \")\n\t\treturn string(jsonBytes)\n\tcase []interface{}:\n\t\t// Array of messages - extract content from last user message\n\t\tfor i := len(v) - 1; i >= 0; i-- {\n\t\t\tif msg, ok := v[i].(map[string]interface{}); ok {\n\t\t\t\tif role, ok := msg[\"role\"].(string); ok && role == \"user\" {\n\t\t\t\t\tif content, ok := msg[\"content\"].(string); ok {\n\t\t\t\t\t\treturn content\n\t\t\t\t\t}\n\t\t\t\t}\n\t\t\t}\n\t\t}\n\t\t// Fallback to JSON\n\t\tjsonBytes, _ := jsoniter.MarshalIndent(v, \"\", \" \")\n\t\treturn string(jsonBytes)\n\tdefault:\n\t\tjsonBytes, _ := jsoniter.MarshalIndent(input, \"\", \" \")\n\t\treturn string(jsonBytes)\n\t}\n}\n\n// formatOutputAsString converts output to string format\nfunc formatOutputAsString(output interface{}) string {\n\tswitch v := output.(type) {\n\tcase string:\n\t\treturn v\n\tcase map[string]interface{}, []interface{}:\n\t\tjsonBytes, _ := jsoniter.MarshalIndent(v, \"\", \" \")\n\t\treturn string(jsonBytes)\n\tdefault:\n\t\tif output == nil {\n\t\t\treturn \"(no output)\"\n\t\t}\n\t\treturn fmt.Sprintf(\"%v\", output)\n\t}\n}\n" }, { "path": "agent/test/input.go", "content": "package test\n\nimport (\n\t\"encoding/base64\"\n\t\"fmt\"\n\t\"mime\"\n\t\"os\"\n\t\"path/filepath\"\n\t\"strings\"\n\n\tjsoniter \"github.com/json-iterator/go\"\n\t\"github.com/yaoapp/yao/agent/context\"\n)\n\n// FileProtocol is the protocol prefix for local file references\nconst FileProtocol = \"file://\"\n\n// SupportedImageExtensions lists supported image file extensions\nvar SupportedImageExtensions = map[string]string{\n\t\".jpg\": \"image/jpeg\",\n\t\".jpeg\": \"image/jpeg\",\n\t\".png\": \"image/png\",\n\t\".gif\": \"image/gif\",\n\t\".webp\": \"image/webp\",\n\t\".bmp\": \"image/bmp\",\n}\n\n// SupportedAudioExtensions lists supported audio file extensions\nvar SupportedAudioExtensions = map[string]string{\n\t\".wav\": \"wav\",\n\t\".mp3\": \"mp3\",\n\t\".flac\": \"flac\",\n\t\".ogg\": \"ogg\",\n\t\".m4a\": \"m4a\",\n}\n\n// SupportedFileExtensions lists supported document file extensions\nvar SupportedFileExtensions = map[string]string{\n\t// Documents\n\t\".pdf\": \"application/pdf\",\n\t\".doc\": \"application/msword\",\n\t\".docx\": \"application/vnd.openxmlformats-officedocument.wordprocessingml.document\",\n\t\".xls\": \"application/vnd.ms-excel\",\n\t\".xlsx\": \"application/vnd.openxmlformats-officedocument.spreadsheetml.sheet\",\n\t\".txt\": \"text/plain\",\n\t\".csv\": \"text/csv\",\n\t\".json\": \"application/json\",\n\t\".xml\": \"application/xml\",\n\t\".html\": \"text/html\",\n\t\".htm\": \"text/html\",\n\t\".md\": \"text/markdown\",\n\n\t// Source code\n\t\".yao\": \"application/json\", // Yao DSL files (JSON-based)\n\t\".ts\": \"text/typescript\", // TypeScript\n\t\".tsx\": \"text/typescript\", // TypeScript JSX\n\t\".js\": \"text/javascript\", // JavaScript\n\t\".jsx\": \"text/javascript\", // JavaScript JSX\n\t\".go\": \"text/x-go\", // Go\n\t\".py\": \"text/x-python\", // Python\n\t\".rs\": \"text/x-rust\", // Rust\n\t\".java\": \"text/x-java\", // Java\n\t\".c\": \"text/x-c\", // C\n\t\".cpp\": \"text/x-c++\", // C++\n\t\".h\": \"text/x-c\", // C header\n\t\".hpp\": \"text/x-c++\", // C++ header\n\t\".rb\": \"text/x-ruby\", // Ruby\n\t\".php\": \"text/x-php\", // PHP\n\t\".sh\": \"text/x-shellscript\", // Shell script\n\t\".bash\": \"text/x-shellscript\", // Bash script\n\t\".zsh\": \"text/x-shellscript\", // Zsh script\n\t\".sql\": \"text/x-sql\", // SQL\n\t\".yaml\": \"text/yaml\", // YAML\n\t\".yml\": \"text/yaml\", // YAML\n\t\".toml\": \"text/x-toml\", // TOML\n\t\".ini\": \"text/x-ini\", // INI\n\t\".conf\": \"text/plain\", // Config files\n\t\".css\": \"text/css\", // CSS\n\t\".scss\": \"text/x-scss\", // SCSS\n\t\".less\": \"text/x-less\", // LESS\n\t\".vue\": \"text/x-vue\", // Vue\n\t\".svelte\": \"text/x-svelte\", // Svelte\n}\n\n// InputOptions configures how input is parsed\ntype InputOptions struct {\n\t// BaseDir is the base directory for resolving relative file paths\n\t// If empty, the current working directory is used\n\tBaseDir string\n}\n\n// ParseInput converts various input formats to []context.Message\n// Supported formats:\n// - string: converted to single user message\n// - map (Message): single message with role and content\n// - []interface{} ([]Message): array of messages (conversation history)\nfunc ParseInput(input interface{}) ([]context.Message, error) {\n\treturn ParseInputWithOptions(input, nil)\n}\n\n// ParseInputWithOptions converts various input formats to []context.Message with options\n// Supported formats:\n// - string: converted to single user message\n// - map (Message): single message with role and content\n// - []interface{} ([]Message): array of messages (conversation history)\n//\n// File references in content parts (type=\"image\", \"file\", \"audio\") with \"source\" field\n// starting with \"file://\" will be loaded and converted to appropriate format:\n// - Images: converted to base64 data URL in image_url field\n// - Audio: converted to base64 in input_audio field\n// - Files: converted to base64 data URL in file field\nfunc ParseInputWithOptions(input interface{}, opts *InputOptions) ([]context.Message, error) {\n\tif input == nil {\n\t\treturn nil, fmt.Errorf(\"input is nil\")\n\t}\n\n\tif opts == nil {\n\t\topts = &InputOptions{}\n\t}\n\n\tswitch v := input.(type) {\n\tcase string:\n\t\t// Simple string input -> single user message\n\t\treturn []context.Message{\n\t\t\t{\n\t\t\t\tRole: context.RoleUser,\n\t\t\t\tContent: v,\n\t\t\t},\n\t\t}, nil\n\n\tcase map[string]interface{}:\n\t\t// Single message object\n\t\tmsg, err := parseMessageMap(v, opts)\n\t\tif err != nil {\n\t\t\treturn nil, fmt.Errorf(\"failed to parse message: %w\", err)\n\t\t}\n\t\treturn []context.Message{*msg}, nil\n\n\tcase []interface{}:\n\t\t// Array of messages (conversation history)\n\t\tmessages := make([]context.Message, 0, len(v))\n\t\tfor i, item := range v {\n\t\t\tswitch m := item.(type) {\n\t\t\tcase map[string]interface{}:\n\t\t\t\tmsg, err := parseMessageMap(m, opts)\n\t\t\t\tif err != nil {\n\t\t\t\t\treturn nil, fmt.Errorf(\"failed to parse message at index %d: %w\", i, err)\n\t\t\t\t}\n\t\t\t\tmessages = append(messages, *msg)\n\t\t\tdefault:\n\t\t\t\treturn nil, fmt.Errorf(\"invalid message type at index %d: expected object, got %T\", i, item)\n\t\t\t}\n\t\t}\n\t\treturn messages, nil\n\n\tdefault:\n\t\treturn nil, fmt.Errorf(\"unsupported input type: %T\", input)\n\t}\n}\n\n// parseMessageMap converts a map to context.Message\nfunc parseMessageMap(m map[string]interface{}, opts *InputOptions) (*context.Message, error) {\n\tmsg := &context.Message{}\n\n\t// Parse role (required)\n\tif role, ok := m[\"role\"].(string); ok {\n\t\tmsg.Role = context.MessageRole(role)\n\t} else {\n\t\t// Default to user role if not specified\n\t\tmsg.Role = context.RoleUser\n\t}\n\n\t// Parse content (required)\n\tif content, ok := m[\"content\"]; ok {\n\t\t// Process content to handle file:// references\n\t\tprocessedContent, err := processContent(content, opts)\n\t\tif err != nil {\n\t\t\treturn nil, fmt.Errorf(\"failed to process content: %w\", err)\n\t\t}\n\t\tmsg.Content = processedContent\n\t} else {\n\t\treturn nil, fmt.Errorf(\"message missing 'content' field\")\n\t}\n\n\t// Parse optional name\n\tif name, ok := m[\"name\"].(string); ok {\n\t\tmsg.Name = &name\n\t}\n\n\t// Parse optional tool_call_id (for tool messages)\n\tif toolCallID, ok := m[\"tool_call_id\"].(string); ok {\n\t\tmsg.ToolCallID = &toolCallID\n\t}\n\n\t// Parse optional tool_calls (for assistant messages)\n\tif toolCalls, ok := m[\"tool_calls\"].([]interface{}); ok {\n\t\tmsg.ToolCalls = make([]context.ToolCall, 0, len(toolCalls))\n\t\tfor _, tc := range toolCalls {\n\t\t\tif tcMap, ok := tc.(map[string]interface{}); ok {\n\t\t\t\ttoolCall, err := parseToolCall(tcMap)\n\t\t\t\tif err != nil {\n\t\t\t\t\treturn nil, fmt.Errorf(\"failed to parse tool_call: %w\", err)\n\t\t\t\t}\n\t\t\t\tmsg.ToolCalls = append(msg.ToolCalls, *toolCall)\n\t\t\t}\n\t\t}\n\t}\n\n\t// Parse optional refusal (for assistant messages)\n\tif refusal, ok := m[\"refusal\"].(string); ok {\n\t\tmsg.Refusal = &refusal\n\t}\n\n\treturn msg, nil\n}\n\n// processContent processes content to handle file:// references\n// Returns the processed content with files loaded and converted\nfunc processContent(content interface{}, opts *InputOptions) (interface{}, error) {\n\tswitch v := content.(type) {\n\tcase string:\n\t\t// Simple string content, no processing needed\n\t\treturn v, nil\n\n\tcase []interface{}:\n\t\t// Array of content parts\n\t\tprocessedParts := make([]context.ContentPart, 0, len(v))\n\t\tfor i, part := range v {\n\t\t\tif partMap, ok := part.(map[string]interface{}); ok {\n\t\t\t\tprocessedPart, err := processContentPart(partMap, opts)\n\t\t\t\tif err != nil {\n\t\t\t\t\treturn nil, fmt.Errorf(\"failed to process content part at index %d: %w\", i, err)\n\t\t\t\t}\n\t\t\t\tprocessedParts = append(processedParts, *processedPart)\n\t\t\t} else {\n\t\t\t\treturn nil, fmt.Errorf(\"invalid content part type at index %d: expected object, got %T\", i, part)\n\t\t\t}\n\t\t}\n\t\treturn processedParts, nil\n\n\tcase map[string]interface{}:\n\t\t// Single content part\n\t\tprocessedPart, err := processContentPart(v, opts)\n\t\tif err != nil {\n\t\t\treturn nil, fmt.Errorf(\"failed to process content part: %w\", err)\n\t\t}\n\t\treturn []context.ContentPart{*processedPart}, nil\n\n\tdefault:\n\t\treturn content, nil\n\t}\n}\n\n// processContentPart processes a single content part map\n// Handles file:// references and converts them to appropriate format\nfunc processContentPart(partMap map[string]interface{}, opts *InputOptions) (*context.ContentPart, error) {\n\tpartType, _ := partMap[\"type\"].(string)\n\n\tswitch partType {\n\tcase \"text\":\n\t\ttext, _ := partMap[\"text\"].(string)\n\t\treturn &context.ContentPart{\n\t\t\tType: context.ContentText,\n\t\t\tText: text,\n\t\t}, nil\n\n\tcase \"image\":\n\t\treturn processImagePart(partMap, opts)\n\n\tcase \"image_url\":\n\t\t// Already in correct format, just parse it\n\t\treturn parseImageURLPart(partMap)\n\n\tcase \"audio\", \"input_audio\":\n\t\treturn processAudioPart(partMap, opts)\n\n\tcase \"file\":\n\t\treturn processFilePart(partMap, opts)\n\n\tcase \"data\":\n\t\treturn parseDataPart(partMap)\n\n\tdefault:\n\t\t// Unknown type, try to preserve as-is\n\t\treturn parseGenericPart(partMap)\n\t}\n}\n\n// processImagePart processes an image content part\n// Supports: source=\"file://path\" for local files\nfunc processImagePart(partMap map[string]interface{}, opts *InputOptions) (*context.ContentPart, error) {\n\tsource, hasSource := partMap[\"source\"].(string)\n\n\t// Check for file:// protocol\n\tif hasSource && strings.HasPrefix(source, FileProtocol) {\n\t\tfilePath := strings.TrimPrefix(source, FileProtocol)\n\t\treturn loadImageFile(filePath, opts)\n\t}\n\n\t// Check for url field (already a URL or base64)\n\tif url, ok := partMap[\"url\"].(string); ok {\n\t\tdetail := context.DetailAuto\n\t\tif d, ok := partMap[\"detail\"].(string); ok {\n\t\t\tdetail = context.ImageDetailLevel(d)\n\t\t}\n\t\treturn &context.ContentPart{\n\t\t\tType: context.ContentImageURL,\n\t\t\tImageURL: &context.ImageURL{\n\t\t\t\tURL: url,\n\t\t\t\tDetail: detail,\n\t\t\t},\n\t\t}, nil\n\t}\n\n\treturn nil, fmt.Errorf(\"image part requires 'source' (file://...) or 'url' field\")\n}\n\n// parseImageURLPart parses an image_url content part\nfunc parseImageURLPart(partMap map[string]interface{}) (*context.ContentPart, error) {\n\timageURL, ok := partMap[\"image_url\"].(map[string]interface{})\n\tif !ok {\n\t\treturn nil, fmt.Errorf(\"image_url part requires 'image_url' object\")\n\t}\n\n\turl, _ := imageURL[\"url\"].(string)\n\tdetail := context.DetailAuto\n\tif d, ok := imageURL[\"detail\"].(string); ok {\n\t\tdetail = context.ImageDetailLevel(d)\n\t}\n\n\treturn &context.ContentPart{\n\t\tType: context.ContentImageURL,\n\t\tImageURL: &context.ImageURL{\n\t\t\tURL: url,\n\t\t\tDetail: detail,\n\t\t},\n\t}, nil\n}\n\n// processAudioPart processes an audio content part\n// Supports: source=\"file://path\" for local files\nfunc processAudioPart(partMap map[string]interface{}, opts *InputOptions) (*context.ContentPart, error) {\n\tsource, hasSource := partMap[\"source\"].(string)\n\n\t// Check for file:// protocol\n\tif hasSource && strings.HasPrefix(source, FileProtocol) {\n\t\tfilePath := strings.TrimPrefix(source, FileProtocol)\n\t\treturn loadAudioFile(filePath, opts)\n\t}\n\n\t// Check for data field (already base64)\n\tif data, ok := partMap[\"data\"].(string); ok {\n\t\tformat, _ := partMap[\"format\"].(string)\n\t\treturn &context.ContentPart{\n\t\t\tType: context.ContentInputAudio,\n\t\t\tInputAudio: &context.InputAudio{\n\t\t\t\tData: data,\n\t\t\t\tFormat: format,\n\t\t\t},\n\t\t}, nil\n\t}\n\n\t// Check for input_audio field\n\tif inputAudio, ok := partMap[\"input_audio\"].(map[string]interface{}); ok {\n\t\tdata, _ := inputAudio[\"data\"].(string)\n\t\tformat, _ := inputAudio[\"format\"].(string)\n\t\treturn &context.ContentPart{\n\t\t\tType: context.ContentInputAudio,\n\t\t\tInputAudio: &context.InputAudio{\n\t\t\t\tData: data,\n\t\t\t\tFormat: format,\n\t\t\t},\n\t\t}, nil\n\t}\n\n\treturn nil, fmt.Errorf(\"audio part requires 'source' (file://...) or 'data'/'input_audio' field\")\n}\n\n// processFilePart processes a file content part\n// Supports: source=\"file://path\" for local files\nfunc processFilePart(partMap map[string]interface{}, opts *InputOptions) (*context.ContentPart, error) {\n\tsource, hasSource := partMap[\"source\"].(string)\n\n\t// Check for file:// protocol\n\tif hasSource && strings.HasPrefix(source, FileProtocol) {\n\t\tfilePath := strings.TrimPrefix(source, FileProtocol)\n\t\tname, _ := partMap[\"name\"].(string)\n\t\treturn loadFile(filePath, name, opts)\n\t}\n\n\t// Check for url field (already a URL)\n\tif url, ok := partMap[\"url\"].(string); ok {\n\t\tfilename, _ := partMap[\"filename\"].(string)\n\t\tif filename == \"\" {\n\t\t\tfilename, _ = partMap[\"name\"].(string)\n\t\t}\n\t\treturn &context.ContentPart{\n\t\t\tType: context.ContentFile,\n\t\t\tFile: &context.FileAttachment{\n\t\t\t\tURL: url,\n\t\t\t\tFilename: filename,\n\t\t\t},\n\t\t}, nil\n\t}\n\n\t// Check for file field\n\tif file, ok := partMap[\"file\"].(map[string]interface{}); ok {\n\t\turl, _ := file[\"url\"].(string)\n\t\tfilename, _ := file[\"filename\"].(string)\n\t\treturn &context.ContentPart{\n\t\t\tType: context.ContentFile,\n\t\t\tFile: &context.FileAttachment{\n\t\t\t\tURL: url,\n\t\t\t\tFilename: filename,\n\t\t\t},\n\t\t}, nil\n\t}\n\n\treturn nil, fmt.Errorf(\"file part requires 'source' (file://...), 'url', or 'file' field\")\n}\n\n// parseDataPart parses a data content part\nfunc parseDataPart(partMap map[string]interface{}) (*context.ContentPart, error) {\n\tdata, ok := partMap[\"data\"].(map[string]interface{})\n\tif !ok {\n\t\treturn nil, fmt.Errorf(\"data part requires 'data' object\")\n\t}\n\n\t// Convert to DataContent\n\tdataContent := &context.DataContent{}\n\n\tif sources, ok := data[\"sources\"].([]interface{}); ok {\n\t\tdataContent.Sources = make([]context.DataSource, 0, len(sources))\n\t\tfor _, src := range sources {\n\t\t\tif srcMap, ok := src.(map[string]interface{}); ok {\n\t\t\t\tds := context.DataSource{}\n\t\t\t\tif t, ok := srcMap[\"type\"].(string); ok {\n\t\t\t\t\tds.Type = context.DataSourceType(t)\n\t\t\t\t}\n\t\t\t\tif id, ok := srcMap[\"id\"].(string); ok {\n\t\t\t\t\tds.ID = id\n\t\t\t\t}\n\t\t\t\tif name, ok := srcMap[\"name\"].(string); ok {\n\t\t\t\t\tds.Name = name\n\t\t\t\t}\n\t\t\t\tif filters, ok := srcMap[\"filters\"].(map[string]interface{}); ok {\n\t\t\t\t\tds.Filters = filters\n\t\t\t\t}\n\t\t\t\tif metadata, ok := srcMap[\"metadata\"].(map[string]interface{}); ok {\n\t\t\t\t\tds.Metadata = metadata\n\t\t\t\t}\n\t\t\t\tdataContent.Sources = append(dataContent.Sources, ds)\n\t\t\t}\n\t\t}\n\t}\n\n\treturn &context.ContentPart{\n\t\tType: context.ContentData,\n\t\tData: dataContent,\n\t}, nil\n}\n\n// parseGenericPart tries to parse an unknown content part type\nfunc parseGenericPart(partMap map[string]interface{}) (*context.ContentPart, error) {\n\tpartType, _ := partMap[\"type\"].(string)\n\n\t// Try to create a basic ContentPart\n\tpart := &context.ContentPart{\n\t\tType: context.ContentPartType(partType),\n\t}\n\n\t// Try to extract text if present\n\tif text, ok := partMap[\"text\"].(string); ok {\n\t\tpart.Text = text\n\t}\n\n\treturn part, nil\n}\n\n// loadImageFile loads an image file and converts it to a ContentPart\nfunc loadImageFile(filePath string, opts *InputOptions) (*context.ContentPart, error) {\n\tabsPath := resolveFilePath(filePath, opts)\n\n\t// Read file\n\tdata, err := os.ReadFile(absPath)\n\tif err != nil {\n\t\treturn nil, fmt.Errorf(\"failed to read image file %s: %w\", filePath, err)\n\t}\n\n\t// Determine MIME type\n\text := strings.ToLower(filepath.Ext(absPath))\n\tmimeType, ok := SupportedImageExtensions[ext]\n\tif !ok {\n\t\t// Try to detect from extension\n\t\tmimeType = mime.TypeByExtension(ext)\n\t\tif mimeType == \"\" {\n\t\t\tmimeType = \"application/octet-stream\"\n\t\t}\n\t}\n\n\t// Encode to base64 data URL\n\tb64Data := base64.StdEncoding.EncodeToString(data)\n\tdataURL := fmt.Sprintf(\"data:%s;base64,%s\", mimeType, b64Data)\n\n\treturn &context.ContentPart{\n\t\tType: context.ContentImageURL,\n\t\tImageURL: &context.ImageURL{\n\t\t\tURL: dataURL,\n\t\t\tDetail: context.DetailAuto,\n\t\t},\n\t}, nil\n}\n\n// loadAudioFile loads an audio file and converts it to a ContentPart\nfunc loadAudioFile(filePath string, opts *InputOptions) (*context.ContentPart, error) {\n\tabsPath := resolveFilePath(filePath, opts)\n\n\t// Read file\n\tdata, err := os.ReadFile(absPath)\n\tif err != nil {\n\t\treturn nil, fmt.Errorf(\"failed to read audio file %s: %w\", filePath, err)\n\t}\n\n\t// Determine format from extension\n\text := strings.ToLower(filepath.Ext(absPath))\n\tformat, ok := SupportedAudioExtensions[ext]\n\tif !ok {\n\t\tformat = strings.TrimPrefix(ext, \".\")\n\t}\n\n\t// Encode to base64\n\tb64Data := base64.StdEncoding.EncodeToString(data)\n\n\treturn &context.ContentPart{\n\t\tType: context.ContentInputAudio,\n\t\tInputAudio: &context.InputAudio{\n\t\t\tData: b64Data,\n\t\t\tFormat: format,\n\t\t},\n\t}, nil\n}\n\n// loadFile loads a file and converts it to a ContentPart\nfunc loadFile(filePath string, name string, opts *InputOptions) (*context.ContentPart, error) {\n\tabsPath := resolveFilePath(filePath, opts)\n\n\t// Read file\n\tdata, err := os.ReadFile(absPath)\n\tif err != nil {\n\t\treturn nil, fmt.Errorf(\"failed to read file %s: %w\", filePath, err)\n\t}\n\n\t// Determine filename\n\tfilename := name\n\tif filename == \"\" {\n\t\tfilename = filepath.Base(absPath)\n\t}\n\n\t// Determine MIME type\n\text := strings.ToLower(filepath.Ext(absPath))\n\tmimeType, ok := SupportedFileExtensions[ext]\n\tif !ok {\n\t\tmimeType = mime.TypeByExtension(ext)\n\t\tif mimeType == \"\" {\n\t\t\tmimeType = \"application/octet-stream\"\n\t\t}\n\t}\n\n\t// Encode to base64 data URL\n\tb64Data := base64.StdEncoding.EncodeToString(data)\n\tdataURL := fmt.Sprintf(\"data:%s;base64,%s\", mimeType, b64Data)\n\n\treturn &context.ContentPart{\n\t\tType: context.ContentFile,\n\t\tFile: &context.FileAttachment{\n\t\t\tURL: dataURL,\n\t\t\tFilename: filename,\n\t\t},\n\t}, nil\n}\n\n// resolveFilePath resolves a file path relative to the base directory\n// If the path is absolute, it's returned as-is\n// If BaseDir is empty, the current working directory is used\nfunc resolveFilePath(filePath string, opts *InputOptions) string {\n\t// If path is absolute, return as-is\n\tif filepath.IsAbs(filePath) {\n\t\treturn filePath\n\t}\n\n\t// If BaseDir is set, resolve relative to it\n\tif opts != nil && opts.BaseDir != \"\" {\n\t\treturn filepath.Join(opts.BaseDir, filePath)\n\t}\n\n\t// Otherwise, resolve relative to current working directory\n\treturn filePath\n}\n\n// parseToolCall converts a map to context.ToolCall\nfunc parseToolCall(m map[string]interface{}) (*context.ToolCall, error) {\n\ttc := &context.ToolCall{}\n\n\tif id, ok := m[\"id\"].(string); ok {\n\t\ttc.ID = id\n\t}\n\n\tif typ, ok := m[\"type\"].(string); ok {\n\t\ttc.Type = context.ToolCallType(typ)\n\t} else {\n\t\ttc.Type = context.ToolTypeFunction\n\t}\n\n\tif fn, ok := m[\"function\"].(map[string]interface{}); ok {\n\t\tif name, ok := fn[\"name\"].(string); ok {\n\t\t\ttc.Function.Name = name\n\t\t}\n\t\tif args, ok := fn[\"arguments\"].(string); ok {\n\t\t\ttc.Function.Arguments = args\n\t\t} else if args, ok := fn[\"arguments\"].(map[string]interface{}); ok {\n\t\t\t// Convert map to JSON string\n\t\t\targsBytes, err := jsoniter.Marshal(args)\n\t\t\tif err != nil {\n\t\t\t\treturn nil, fmt.Errorf(\"failed to marshal arguments: %w\", err)\n\t\t\t}\n\t\t\ttc.Function.Arguments = string(argsBytes)\n\t\t}\n\t}\n\n\treturn tc, nil\n}\n\n// ExtractTextContent extracts text content from various content formats\n// Used for display in reports\nfunc ExtractTextContent(content interface{}) string {\n\tif content == nil {\n\t\treturn \"\"\n\t}\n\n\tswitch v := content.(type) {\n\tcase string:\n\t\treturn v\n\n\tcase []interface{}:\n\t\t// ContentPart array\n\t\tvar texts []string\n\t\tfor _, part := range v {\n\t\t\tif partMap, ok := part.(map[string]interface{}); ok {\n\t\t\t\tif partMap[\"type\"] == \"text\" {\n\t\t\t\t\tif text, ok := partMap[\"text\"].(string); ok {\n\t\t\t\t\t\ttexts = append(texts, text)\n\t\t\t\t\t}\n\t\t\t\t}\n\t\t\t}\n\t\t}\n\t\tif len(texts) > 0 {\n\t\t\tresult := texts[0]\n\t\t\tfor i := 1; i < len(texts); i++ {\n\t\t\t\tresult += \"\\n\" + texts[i]\n\t\t\t}\n\t\t\treturn result\n\t\t}\n\t\treturn fmt.Sprintf(\"[%d content parts]\", len(v))\n\n\tcase map[string]interface{}:\n\t\t// Single ContentPart or Message\n\t\tif v[\"type\"] == \"text\" {\n\t\t\tif text, ok := v[\"text\"].(string); ok {\n\t\t\t\treturn text\n\t\t\t}\n\t\t}\n\t\tif content, ok := v[\"content\"]; ok {\n\t\t\treturn ExtractTextContent(content)\n\t\t}\n\t\treturn fmt.Sprintf(\"%v\", v)\n\n\tdefault:\n\t\treturn fmt.Sprintf(\"%v\", v)\n\t}\n}\n\n// SummarizeInput creates a short summary of the input for display\nfunc SummarizeInput(input interface{}, maxLen int) string {\n\ttext := \"\"\n\n\tswitch v := input.(type) {\n\tcase string:\n\t\ttext = v\n\n\tcase map[string]interface{}:\n\t\tif content, ok := v[\"content\"]; ok {\n\t\t\ttext = ExtractTextContent(content)\n\t\t}\n\n\tcase []interface{}:\n\t\t// Get the last user message for summary\n\t\tfor i := len(v) - 1; i >= 0; i-- {\n\t\t\tif msg, ok := v[i].(map[string]interface{}); ok {\n\t\t\t\tif msg[\"role\"] == \"user\" {\n\t\t\t\t\tif content, ok := msg[\"content\"]; ok {\n\t\t\t\t\t\ttext = ExtractTextContent(content)\n\t\t\t\t\t\tbreak\n\t\t\t\t\t}\n\t\t\t\t}\n\t\t\t}\n\t\t}\n\t\tif text == \"\" && len(v) > 0 {\n\t\t\ttext = fmt.Sprintf(\"[%d messages]\", len(v))\n\t\t}\n\n\tdefault:\n\t\ttext = fmt.Sprintf(\"%v\", v)\n\t}\n\n\tif maxLen > 0 && len(text) > maxLen {\n\t\treturn text[:maxLen-3] + \"...\"\n\t}\n\treturn text\n}\n" }, { "path": "agent/test/input_source.go", "content": "package test\n\nimport (\n\t\"fmt\"\n\t\"net/url\"\n\t\"strconv\"\n\t\"strings\"\n\n\tjsoniter \"github.com/json-iterator/go\"\n\tgoutext \"github.com/yaoapp/gou/text\"\n\t\"github.com/yaoapp/yao/agent/assistant\"\n\t\"github.com/yaoapp/yao/agent/context\"\n)\n\n// InputSourceType represents the type of input source\ntype InputSourceType string\n\nconst (\n\t// InputSourceFile indicates input from a JSONL file\n\tInputSourceFile InputSourceType = \"file\"\n\t// InputSourceMessage indicates input from a direct message string\n\tInputSourceMessage InputSourceType = \"message\"\n\t// InputSourceScript indicates script test mode\n\tInputSourceScript InputSourceType = \"script\"\n\t// InputSourceAgent indicates input generated by an agent\n\tInputSourceAgent InputSourceType = \"agent\"\n)\n\n// InputSource represents a parsed input source\ntype InputSource struct {\n\tType InputSourceType // file, message, script, agent\n\tValue string // path, message, script ref, or agent ID\n\tParams map[string]interface{} // query parameters (for agent source)\n}\n\n// ParseInputSource parses the -i flag value into an InputSource\n// Supported formats:\n// - \"agents:workers.test.generator\" - Agent-generated test cases\n// - \"agents:workers.test.generator?count=10&focus=edge-cases\" - With parameters\n// - \"scripts.tests.gen\" - Script-generated test cases\n// - \"./tests/inputs.jsonl\" - JSONL file\n// - \"Hello, how are you?\" - Direct message\nfunc ParseInputSource(input string) *InputSource {\n\t// Check for agents: prefix\n\tif strings.HasPrefix(input, \"agents:\") {\n\t\treturn parseAgentSource(strings.TrimPrefix(input, \"agents:\"))\n\t}\n\n\t// Check for scripts: prefix (for generator scripts)\n\tif strings.HasPrefix(input, \"scripts:\") {\n\t\treturn &InputSource{\n\t\t\tType: InputSourceScript,\n\t\t\tValue: strings.TrimPrefix(input, \"scripts:\"),\n\t\t}\n\t}\n\n\t// Check for script test mode (scripts.xxx format without prefix)\n\tif strings.HasPrefix(input, \"scripts.\") {\n\t\treturn &InputSource{\n\t\t\tType: InputSourceScript,\n\t\t\tValue: input,\n\t\t}\n\t}\n\n\t// Check for file extension\n\tif strings.HasSuffix(input, \".jsonl\") || strings.HasSuffix(input, \".json\") {\n\t\treturn &InputSource{\n\t\t\tType: InputSourceFile,\n\t\t\tValue: input,\n\t\t}\n\t}\n\n\t// Check if it looks like a file path\n\tif strings.Contains(input, \"/\") || strings.Contains(input, \"\\\\\") {\n\t\treturn &InputSource{\n\t\t\tType: InputSourceFile,\n\t\t\tValue: input,\n\t\t}\n\t}\n\n\t// Default to message\n\treturn &InputSource{\n\t\tType: InputSourceMessage,\n\t\tValue: input,\n\t}\n}\n\n// parseAgentSource parses an agent source string with optional query parameters\n// Format: \"agent.id\" or \"agent.id?count=10&focus=edge-cases\"\nfunc parseAgentSource(input string) *InputSource {\n\tsource := &InputSource{\n\t\tType: InputSourceAgent,\n\t\tParams: make(map[string]interface{}),\n\t}\n\n\t// Check for query parameters\n\tif idx := strings.Index(input, \"?\"); idx >= 0 {\n\t\tsource.Value = input[:idx]\n\t\tqueryStr := input[idx+1:]\n\n\t\t// Parse query parameters\n\t\tvalues, err := url.ParseQuery(queryStr)\n\t\tif err == nil {\n\t\t\tfor key, vals := range values {\n\t\t\t\tif len(vals) > 0 {\n\t\t\t\t\t// Try to parse as number\n\t\t\t\t\tif num, err := strconv.Atoi(vals[0]); err == nil {\n\t\t\t\t\t\tsource.Params[key] = num\n\t\t\t\t\t} else if num, err := strconv.ParseFloat(vals[0], 64); err == nil {\n\t\t\t\t\t\tsource.Params[key] = num\n\t\t\t\t\t} else if vals[0] == \"true\" {\n\t\t\t\t\t\tsource.Params[key] = true\n\t\t\t\t\t} else if vals[0] == \"false\" {\n\t\t\t\t\t\tsource.Params[key] = false\n\t\t\t\t\t} else {\n\t\t\t\t\t\tsource.Params[key] = vals[0]\n\t\t\t\t\t}\n\t\t\t\t}\n\t\t\t}\n\t\t}\n\t} else {\n\t\tsource.Value = input\n\t}\n\n\treturn source\n}\n\n// GeneratorInput represents the input sent to a generator agent\ntype GeneratorInput struct {\n\tTargetAgent *TargetAgentInfo `json:\"target_agent\"`\n\tCount int `json:\"count,omitempty\"`\n\tFocus string `json:\"focus,omitempty\"`\n\tExtra map[string]interface{} `json:\"extra,omitempty\"`\n}\n\n// TargetAgentInfo contains information about the agent being tested\ntype TargetAgentInfo struct {\n\tID string `json:\"id\"`\n\tDescription string `json:\"description,omitempty\"`\n\tTools []map[string]interface{} `json:\"tools,omitempty\"`\n}\n\n// GenerateTestCases generates test cases using a generator agent\nfunc GenerateTestCases(agentID string, targetInfo *TargetAgentInfo, params map[string]interface{}) ([]*Case, error) {\n\t// Get generator assistant\n\tast, err := assistant.Get(agentID)\n\tif err != nil {\n\t\treturn nil, fmt.Errorf(\"failed to get generator agent %s: %w\", agentID, err)\n\t}\n\n\t// Build generation request\n\tgenInput := &GeneratorInput{\n\t\tTargetAgent: targetInfo,\n\t\tCount: 5, // Default count\n\t}\n\n\t// Apply parameters\n\tif params != nil {\n\t\tif count, ok := params[\"count\"].(int); ok {\n\t\t\tgenInput.Count = count\n\t\t}\n\t\tif focus, ok := params[\"focus\"].(string); ok {\n\t\t\tgenInput.Focus = focus\n\t\t}\n\t\t// Store extra parameters\n\t\tgenInput.Extra = make(map[string]interface{})\n\t\tfor k, v := range params {\n\t\t\tif k != \"count\" && k != \"focus\" {\n\t\t\t\tgenInput.Extra[k] = v\n\t\t\t}\n\t\t}\n\t}\n\n\t// Create context\n\tenv := NewEnvironment(\"\", \"\")\n\tctx := NewTestContext(\"generator\", agentID, env)\n\tdefer ctx.Release()\n\n\t// Build options - skip history and trace for efficiency\n\topts := &context.Options{\n\t\tSkip: &context.Skip{\n\t\t\tHistory: true,\n\t\t\tTrace: true,\n\t\t\tOutput: true,\n\t\t},\n\t\tMetadata: map[string]interface{}{\n\t\t\t\"test_mode\": \"generator\",\n\t\t},\n\t}\n\n\t// Build message\n\tinputJSON, err := jsoniter.Marshal(genInput)\n\tif err != nil {\n\t\treturn nil, fmt.Errorf(\"failed to marshal generator input: %w\", err)\n\t}\n\n\tmessages := []context.Message{{\n\t\tRole: context.RoleUser,\n\t\tContent: string(inputJSON),\n\t}}\n\n\t// Call generator agent\n\tresponse, err := ast.Stream(ctx, messages, opts)\n\tif err != nil {\n\t\treturn nil, fmt.Errorf(\"generator agent error: %w\", err)\n\t}\n\n\t// Extract and parse response\n\treturn parseGeneratedCases(response)\n}\n\n// parseGeneratedCases parses the generator agent's response into test cases\nfunc parseGeneratedCases(response *context.Response) ([]*Case, error) {\n\tif response == nil || response.Completion == nil {\n\t\treturn nil, fmt.Errorf(\"empty response from generator agent\")\n\t}\n\n\t// Extract content\n\tcontent := response.Completion.Content\n\tif content == nil {\n\t\treturn nil, fmt.Errorf(\"no content in generator response\")\n\t}\n\n\t// Convert content to string\n\tvar text string\n\tswitch v := content.(type) {\n\tcase string:\n\t\ttext = v\n\tdefault:\n\t\tdata, err := jsoniter.Marshal(content)\n\t\tif err != nil {\n\t\t\treturn nil, fmt.Errorf(\"failed to marshal content: %w\", err)\n\t\t}\n\t\ttext = string(data)\n\t}\n\n\t// Use goutext.ExtractJSON for fault-tolerant parsing\n\tparsed := goutext.ExtractJSON(text)\n\tif parsed == nil {\n\t\treturn nil, fmt.Errorf(\"failed to parse generator response as JSON: %s\", truncateOutput(text, 200))\n\t}\n\n\t// Convert to []*Case\n\treturn convertToCases(parsed)\n}\n\n// convertToCases converts parsed JSON to test cases\nfunc convertToCases(parsed interface{}) ([]*Case, error) {\n\t// Handle array of cases\n\tarr, ok := parsed.([]interface{})\n\tif !ok {\n\t\t// Maybe it's a single case wrapped in an object\n\t\tif obj, ok := parsed.(map[string]interface{}); ok {\n\t\t\tif cases, ok := obj[\"cases\"].([]interface{}); ok {\n\t\t\t\tarr = cases\n\t\t\t} else if testCases, ok := obj[\"test_cases\"].([]interface{}); ok {\n\t\t\t\tarr = testCases\n\t\t\t} else {\n\t\t\t\t// Single case\n\t\t\t\tarr = []interface{}{obj}\n\t\t\t}\n\t\t} else {\n\t\t\treturn nil, fmt.Errorf(\"expected array of test cases, got %T\", parsed)\n\t\t}\n\t}\n\n\tcases := make([]*Case, 0, len(arr))\n\tfor i, item := range arr {\n\t\tcaseMap, ok := item.(map[string]interface{})\n\t\tif !ok {\n\t\t\treturn nil, fmt.Errorf(\"test case %d is not an object\", i)\n\t\t}\n\n\t\ttc, err := mapToCase(caseMap)\n\t\tif err != nil {\n\t\t\treturn nil, fmt.Errorf(\"failed to parse test case %d: %w\", i, err)\n\t\t}\n\n\t\tcases = append(cases, tc)\n\t}\n\n\treturn cases, nil\n}\n\n// mapToCase converts a map to a Case struct\nfunc mapToCase(m map[string]interface{}) (*Case, error) {\n\ttc := &Case{}\n\n\t// Required: id\n\tif id, ok := m[\"id\"].(string); ok {\n\t\ttc.ID = id\n\t} else {\n\t\treturn nil, fmt.Errorf(\"missing required field 'id'\")\n\t}\n\n\t// Required: input\n\tif input, ok := m[\"input\"]; ok {\n\t\ttc.Input = input\n\t} else {\n\t\treturn nil, fmt.Errorf(\"missing required field 'input'\")\n\t}\n\n\t// Optional: assertions/assert\n\tif assertions, ok := m[\"assertions\"]; ok {\n\t\ttc.Assert = assertions\n\t} else if assert, ok := m[\"assert\"]; ok {\n\t\ttc.Assert = assert\n\t}\n\n\t// Optional: options - convert map to CaseOptions\n\tif options, ok := m[\"options\"].(map[string]interface{}); ok {\n\t\ttc.Options = mapToCaseOptions(options)\n\t}\n\n\t// Optional: before/after\n\tif before, ok := m[\"before\"].(string); ok {\n\t\ttc.Before = before\n\t}\n\tif after, ok := m[\"after\"].(string); ok {\n\t\ttc.After = after\n\t}\n\n\t// Optional: timeout\n\tif timeout, ok := m[\"timeout\"].(string); ok {\n\t\ttc.Timeout = timeout\n\t}\n\n\treturn tc, nil\n}\n\n// ToInputMode converts InputSourceType to InputMode for backward compatibility\nfunc (s *InputSource) ToInputMode() InputMode {\n\tswitch s.Type {\n\tcase InputSourceFile:\n\t\treturn InputModeFile\n\tcase InputSourceMessage:\n\t\treturn InputModeMessage\n\tcase InputSourceScript:\n\t\treturn InputModeScript\n\tcase InputSourceAgent:\n\t\t// Agent source generates cases, then runs in file mode\n\t\treturn InputModeFile\n\tdefault:\n\t\treturn InputModeMessage\n\t}\n}\n\n// mapToCaseOptions converts a map to CaseOptions\nfunc mapToCaseOptions(m map[string]interface{}) *CaseOptions {\n\topts := &CaseOptions{}\n\n\tif connector, ok := m[\"connector\"].(string); ok {\n\t\topts.Connector = connector\n\t}\n\n\tif mode, ok := m[\"mode\"].(string); ok {\n\t\topts.Mode = mode\n\t}\n\n\tif disableGlobalPrompts, ok := m[\"disable_global_prompts\"].(bool); ok {\n\t\topts.DisableGlobalPrompts = disableGlobalPrompts\n\t}\n\n\tif search, ok := m[\"search\"].(bool); ok {\n\t\topts.Search = &search\n\t}\n\n\tif metadata, ok := m[\"metadata\"].(map[string]interface{}); ok {\n\t\topts.Metadata = metadata\n\t}\n\n\tif skip, ok := m[\"skip\"].(map[string]interface{}); ok {\n\t\topts.Skip = &CaseSkipOptions{}\n\t\tif history, ok := skip[\"history\"].(bool); ok {\n\t\t\topts.Skip.History = history\n\t\t}\n\t\tif trace, ok := skip[\"trace\"].(bool); ok {\n\t\t\topts.Skip.Trace = trace\n\t\t}\n\t\tif output, ok := skip[\"output\"].(bool); ok {\n\t\t\topts.Skip.Output = output\n\t\t}\n\t\tif keyword, ok := skip[\"keyword\"].(bool); ok {\n\t\t\topts.Skip.Keyword = keyword\n\t\t}\n\t\tif searchSkip, ok := skip[\"search\"].(bool); ok {\n\t\t\topts.Skip.Search = searchSkip\n\t\t}\n\t}\n\n\treturn opts\n}\n" }, { "path": "agent/test/input_source_test.go", "content": "package test_test\n\nimport (\n\t\"testing\"\n\n\t\"github.com/stretchr/testify/assert\"\n\t\"github.com/yaoapp/yao/agent\"\n\tagenttest \"github.com/yaoapp/yao/agent/test\"\n\t\"github.com/yaoapp/yao/config\"\n\t\"github.com/yaoapp/yao/test\"\n)\n\nfunc TestParseInputSource(t *testing.T) {\n\ttests := []struct {\n\t\tname string\n\t\tinput string\n\t\twantType agenttest.InputSourceType\n\t\twantValue string\n\t\twantParams map[string]interface{}\n\t}{\n\t\t{\n\t\t\tname: \"JSONL file\",\n\t\t\tinput: \"./tests/inputs.jsonl\",\n\t\t\twantType: agenttest.InputSourceFile,\n\t\t\twantValue: \"./tests/inputs.jsonl\",\n\t\t},\n\t\t{\n\t\t\tname: \"JSON file\",\n\t\t\tinput: \"./tests/inputs.json\",\n\t\t\twantType: agenttest.InputSourceFile,\n\t\t\twantValue: \"./tests/inputs.json\",\n\t\t},\n\t\t{\n\t\t\tname: \"direct message\",\n\t\t\tinput: \"Hello, how are you?\",\n\t\t\twantType: agenttest.InputSourceMessage,\n\t\t\twantValue: \"Hello, how are you?\",\n\t\t},\n\t\t{\n\t\t\tname: \"agent source simple\",\n\t\t\tinput: \"agents:tests.generator-agent\",\n\t\t\twantType: agenttest.InputSourceAgent,\n\t\t\twantValue: \"tests.generator-agent\",\n\t\t},\n\t\t{\n\t\t\tname: \"agent source with params\",\n\t\t\tinput: \"agents:tests.generator-agent?count=10&focus=edge-cases\",\n\t\t\twantType: agenttest.InputSourceAgent,\n\t\t\twantValue: \"tests.generator-agent\",\n\t\t\twantParams: map[string]interface{}{\n\t\t\t\t\"count\": 10,\n\t\t\t\t\"focus\": \"edge-cases\",\n\t\t\t},\n\t\t},\n\t\t{\n\t\t\tname: \"agent source with boolean param\",\n\t\t\tinput: \"agents:tests.generator-agent?verbose=true\",\n\t\t\twantType: agenttest.InputSourceAgent,\n\t\t\twantValue: \"tests.generator-agent\",\n\t\t\twantParams: map[string]interface{}{\n\t\t\t\t\"verbose\": true,\n\t\t\t},\n\t\t},\n\t\t{\n\t\t\tname: \"script source with prefix\",\n\t\t\tinput: \"scripts:tests.gen.Generate\",\n\t\t\twantType: agenttest.InputSourceScript,\n\t\t\twantValue: \"tests.gen.Generate\",\n\t\t},\n\t\t{\n\t\t\tname: \"script test mode\",\n\t\t\tinput: \"scripts.tests.gen\",\n\t\t\twantType: agenttest.InputSourceScript,\n\t\t\twantValue: \"scripts.tests.gen\",\n\t\t},\n\t\t{\n\t\t\tname: \"path with separator\",\n\t\t\tinput: \"/path/to/inputs.jsonl\",\n\t\t\twantType: agenttest.InputSourceFile,\n\t\t\twantValue: \"/path/to/inputs.jsonl\",\n\t\t},\n\t}\n\n\tfor _, tt := range tests {\n\t\tt.Run(tt.name, func(t *testing.T) {\n\t\t\tsource := agenttest.ParseInputSource(tt.input)\n\n\t\t\tassert.Equal(t, tt.wantType, source.Type, \"Type mismatch\")\n\t\t\tassert.Equal(t, tt.wantValue, source.Value, \"Value mismatch\")\n\n\t\t\tif tt.wantParams != nil {\n\t\t\t\tfor k, v := range tt.wantParams {\n\t\t\t\t\tassert.Equal(t, v, source.Params[k], \"Param %s mismatch\", k)\n\t\t\t\t}\n\t\t\t}\n\t\t})\n\t}\n}\n\nfunc TestInputSource_ToInputMode(t *testing.T) {\n\ttests := []struct {\n\t\tname string\n\t\tsource *agenttest.InputSource\n\t\twantMode agenttest.InputMode\n\t}{\n\t\t{\n\t\t\tname: \"file source\",\n\t\t\tsource: &agenttest.InputSource{Type: agenttest.InputSourceFile},\n\t\t\twantMode: agenttest.InputModeFile,\n\t\t},\n\t\t{\n\t\t\tname: \"message source\",\n\t\t\tsource: &agenttest.InputSource{Type: agenttest.InputSourceMessage},\n\t\t\twantMode: agenttest.InputModeMessage,\n\t\t},\n\t\t{\n\t\t\tname: \"script source\",\n\t\t\tsource: &agenttest.InputSource{Type: agenttest.InputSourceScript},\n\t\t\twantMode: agenttest.InputModeScript,\n\t\t},\n\t\t{\n\t\t\tname: \"agent source\",\n\t\t\tsource: &agenttest.InputSource{Type: agenttest.InputSourceAgent},\n\t\t\twantMode: agenttest.InputModeFile, // Agent generates cases, then runs in file mode\n\t\t},\n\t}\n\n\tfor _, tt := range tests {\n\t\tt.Run(tt.name, func(t *testing.T) {\n\t\t\tmode := tt.source.ToInputMode()\n\t\t\tassert.Equal(t, tt.wantMode, mode)\n\t\t})\n\t}\n}\n\nfunc TestGenerateTestCases(t *testing.T) {\n\ttest.Prepare(t, config.Conf)\n\tdefer test.Clean()\n\n\t// Load agent (includes assistants)\n\terr := agent.Load(config.Conf)\n\tif err != nil {\n\t\tt.Fatalf(\"Failed to load agent: %v\", err)\n\t}\n\n\t// Test generating test cases from the generator agent\n\ttargetInfo := &agenttest.TargetAgentInfo{\n\t\tID: \"tests.next\",\n\t\tDescription: \"A simple test agent for greeting\",\n\t}\n\n\tparams := map[string]interface{}{\n\t\t\"count\": 3,\n\t\t\"focus\": \"happy-path\",\n\t}\n\n\tcases, err := agenttest.GenerateTestCases(\"tests.generator-agent\", targetInfo, params)\n\tif err != nil {\n\t\tt.Fatalf(\"Failed to generate test cases: %v\", err)\n\t}\n\n\t// Verify we got some test cases\n\tassert.NotEmpty(t, cases, \"Should generate at least one test case\")\n\n\t// Verify each case has required fields\n\tfor _, tc := range cases {\n\t\tassert.NotEmpty(t, tc.ID, \"Test case should have ID\")\n\t\tassert.NotNil(t, tc.Input, \"Test case should have Input\")\n\t}\n\n\tt.Logf(\"Generated %d test cases\", len(cases))\n\tfor _, tc := range cases {\n\t\tt.Logf(\" - %s\", tc.ID)\n\t}\n}\n\nfunc TestMapToCaseOptions(t *testing.T) {\n\t// Test that options map is correctly converted\n\tsource := agenttest.ParseInputSource(\"agents:test?count=5\")\n\tassert.Equal(t, 5, source.Params[\"count\"])\n}\n" }, { "path": "agent/test/input_test.go", "content": "package test\n\nimport (\n\t\"encoding/base64\"\n\t\"os\"\n\t\"path/filepath\"\n\t\"strings\"\n\t\"testing\"\n\n\t\"github.com/yaoapp/yao/agent/context\"\n)\n\nfunc TestParseInput_String(t *testing.T) {\n\tinput := \"Hello world\"\n\tmessages, err := ParseInput(input)\n\tif err != nil {\n\t\tt.Fatalf(\"ParseInput failed: %v\", err)\n\t}\n\n\tif len(messages) != 1 {\n\t\tt.Fatalf(\"Expected 1 message, got %d\", len(messages))\n\t}\n\n\tif messages[0].Role != context.RoleUser {\n\t\tt.Errorf(\"Expected role 'user', got '%s'\", messages[0].Role)\n\t}\n\n\tcontent, ok := messages[0].Content.(string)\n\tif !ok {\n\t\tt.Fatalf(\"Expected string content, got %T\", messages[0].Content)\n\t}\n\tif content != \"Hello world\" {\n\t\tt.Errorf(\"Expected content 'Hello world', got '%s'\", content)\n\t}\n}\n\nfunc TestParseInput_MessageMap(t *testing.T) {\n\tinput := map[string]interface{}{\n\t\t\"role\": \"user\",\n\t\t\"content\": \"Test message\",\n\t}\n\n\tmessages, err := ParseInput(input)\n\tif err != nil {\n\t\tt.Fatalf(\"ParseInput failed: %v\", err)\n\t}\n\n\tif len(messages) != 1 {\n\t\tt.Fatalf(\"Expected 1 message, got %d\", len(messages))\n\t}\n\n\tif messages[0].Role != context.RoleUser {\n\t\tt.Errorf(\"Expected role 'user', got '%s'\", messages[0].Role)\n\t}\n}\n\nfunc TestParseInput_MessageArray(t *testing.T) {\n\tinput := []interface{}{\n\t\tmap[string]interface{}{\"role\": \"user\", \"content\": \"Hello\"},\n\t\tmap[string]interface{}{\"role\": \"assistant\", \"content\": \"Hi there\"},\n\t\tmap[string]interface{}{\"role\": \"user\", \"content\": \"Follow-up\"},\n\t}\n\n\tmessages, err := ParseInput(input)\n\tif err != nil {\n\t\tt.Fatalf(\"ParseInput failed: %v\", err)\n\t}\n\n\tif len(messages) != 3 {\n\t\tt.Fatalf(\"Expected 3 messages, got %d\", len(messages))\n\t}\n\n\tif messages[0].Role != context.RoleUser {\n\t\tt.Errorf(\"Expected first message role 'user', got '%s'\", messages[0].Role)\n\t}\n\tif messages[1].Role != context.RoleAssistant {\n\t\tt.Errorf(\"Expected second message role 'assistant', got '%s'\", messages[1].Role)\n\t}\n}\n\nfunc TestParseInput_ContentParts(t *testing.T) {\n\tinput := map[string]interface{}{\n\t\t\"role\": \"user\",\n\t\t\"content\": []interface{}{\n\t\t\tmap[string]interface{}{\"type\": \"text\", \"text\": \"Analyze this\"},\n\t\t\tmap[string]interface{}{\"type\": \"image_url\", \"image_url\": map[string]interface{}{\n\t\t\t\t\"url\": \"https://example.com/image.jpg\",\n\t\t\t\t\"detail\": \"high\",\n\t\t\t}},\n\t\t},\n\t}\n\n\tmessages, err := ParseInput(input)\n\tif err != nil {\n\t\tt.Fatalf(\"ParseInput failed: %v\", err)\n\t}\n\n\tif len(messages) != 1 {\n\t\tt.Fatalf(\"Expected 1 message, got %d\", len(messages))\n\t}\n\n\tparts, ok := messages[0].Content.([]context.ContentPart)\n\tif !ok {\n\t\tt.Fatalf(\"Expected []ContentPart, got %T\", messages[0].Content)\n\t}\n\n\tif len(parts) != 2 {\n\t\tt.Fatalf(\"Expected 2 content parts, got %d\", len(parts))\n\t}\n\n\tif parts[0].Type != context.ContentText {\n\t\tt.Errorf(\"Expected first part type 'text', got '%s'\", parts[0].Type)\n\t}\n\tif parts[0].Text != \"Analyze this\" {\n\t\tt.Errorf(\"Expected text 'Analyze this', got '%s'\", parts[0].Text)\n\t}\n\n\tif parts[1].Type != context.ContentImageURL {\n\t\tt.Errorf(\"Expected second part type 'image_url', got '%s'\", parts[1].Type)\n\t}\n\tif parts[1].ImageURL == nil {\n\t\tt.Fatal(\"Expected ImageURL to be set\")\n\t}\n\tif parts[1].ImageURL.URL != \"https://example.com/image.jpg\" {\n\t\tt.Errorf(\"Expected URL 'https://example.com/image.jpg', got '%s'\", parts[1].ImageURL.URL)\n\t}\n\tif parts[1].ImageURL.Detail != context.DetailHigh {\n\t\tt.Errorf(\"Expected detail 'high', got '%s'\", parts[1].ImageURL.Detail)\n\t}\n}\n\nfunc TestParseInputWithOptions_FileProtocol_Image(t *testing.T) {\n\t// Create a temporary test image file\n\ttmpDir := t.TempDir()\n\timgPath := filepath.Join(tmpDir, \"test.png\")\n\n\t// Create a minimal PNG file (1x1 pixel, red)\n\tpngData := []byte{\n\t\t0x89, 0x50, 0x4E, 0x47, 0x0D, 0x0A, 0x1A, 0x0A, // PNG signature\n\t\t0x00, 0x00, 0x00, 0x0D, 0x49, 0x48, 0x44, 0x52, // IHDR chunk\n\t\t0x00, 0x00, 0x00, 0x01, 0x00, 0x00, 0x00, 0x01,\n\t\t0x08, 0x02, 0x00, 0x00, 0x00, 0x90, 0x77, 0x53,\n\t\t0xDE, 0x00, 0x00, 0x00, 0x0C, 0x49, 0x44, 0x41, // IDAT chunk\n\t\t0x54, 0x08, 0xD7, 0x63, 0xF8, 0xCF, 0xC0, 0x00,\n\t\t0x00, 0x00, 0x03, 0x00, 0x01, 0x00, 0x05, 0xFE,\n\t\t0xD4, 0xEF, 0x00, 0x00, 0x00, 0x00, 0x49, 0x45, // IEND chunk\n\t\t0x4E, 0x44, 0xAE, 0x42, 0x60, 0x82,\n\t}\n\tif err := os.WriteFile(imgPath, pngData, 0644); err != nil {\n\t\tt.Fatalf(\"Failed to create test image: %v\", err)\n\t}\n\n\tinput := map[string]interface{}{\n\t\t\"role\": \"user\",\n\t\t\"content\": []interface{}{\n\t\t\tmap[string]interface{}{\"type\": \"text\", \"text\": \"Analyze this image\"},\n\t\t\tmap[string]interface{}{\"type\": \"image\", \"source\": \"file://test.png\"},\n\t\t},\n\t}\n\n\topts := &InputOptions{BaseDir: tmpDir}\n\tmessages, err := ParseInputWithOptions(input, opts)\n\tif err != nil {\n\t\tt.Fatalf(\"ParseInputWithOptions failed: %v\", err)\n\t}\n\n\tif len(messages) != 1 {\n\t\tt.Fatalf(\"Expected 1 message, got %d\", len(messages))\n\t}\n\n\tparts, ok := messages[0].Content.([]context.ContentPart)\n\tif !ok {\n\t\tt.Fatalf(\"Expected []ContentPart, got %T\", messages[0].Content)\n\t}\n\n\tif len(parts) != 2 {\n\t\tt.Fatalf(\"Expected 2 content parts, got %d\", len(parts))\n\t}\n\n\t// Check image part\n\timgPart := parts[1]\n\tif imgPart.Type != context.ContentImageURL {\n\t\tt.Errorf(\"Expected type 'image_url', got '%s'\", imgPart.Type)\n\t}\n\tif imgPart.ImageURL == nil {\n\t\tt.Fatal(\"Expected ImageURL to be set\")\n\t}\n\tif !strings.HasPrefix(imgPart.ImageURL.URL, \"data:image/png;base64,\") {\n\t\tt.Errorf(\"Expected base64 data URL, got '%s'\", imgPart.ImageURL.URL[:50])\n\t}\n\n\t// Verify the base64 content\n\tb64Part := strings.TrimPrefix(imgPart.ImageURL.URL, \"data:image/png;base64,\")\n\tdecoded, err := base64.StdEncoding.DecodeString(b64Part)\n\tif err != nil {\n\t\tt.Fatalf(\"Failed to decode base64: %v\", err)\n\t}\n\tif len(decoded) != len(pngData) {\n\t\tt.Errorf(\"Decoded data length mismatch: expected %d, got %d\", len(pngData), len(decoded))\n\t}\n}\n\nfunc TestParseInputWithOptions_FileProtocol_Audio(t *testing.T) {\n\t// Create a temporary test audio file\n\ttmpDir := t.TempDir()\n\taudioPath := filepath.Join(tmpDir, \"test.wav\")\n\n\t// Create a minimal WAV file header\n\twavData := []byte{\n\t\t0x52, 0x49, 0x46, 0x46, // \"RIFF\"\n\t\t0x24, 0x00, 0x00, 0x00, // File size - 8\n\t\t0x57, 0x41, 0x56, 0x45, // \"WAVE\"\n\t\t0x66, 0x6D, 0x74, 0x20, // \"fmt \"\n\t\t0x10, 0x00, 0x00, 0x00, // Subchunk1Size (16 for PCM)\n\t\t0x01, 0x00, // AudioFormat (1 = PCM)\n\t\t0x01, 0x00, // NumChannels (1 = mono)\n\t\t0x44, 0xAC, 0x00, 0x00, // SampleRate (44100)\n\t\t0x88, 0x58, 0x01, 0x00, // ByteRate\n\t\t0x02, 0x00, // BlockAlign\n\t\t0x10, 0x00, // BitsPerSample (16)\n\t\t0x64, 0x61, 0x74, 0x61, // \"data\"\n\t\t0x00, 0x00, 0x00, 0x00, // Subchunk2Size (0 = no data)\n\t}\n\tif err := os.WriteFile(audioPath, wavData, 0644); err != nil {\n\t\tt.Fatalf(\"Failed to create test audio: %v\", err)\n\t}\n\n\tinput := map[string]interface{}{\n\t\t\"role\": \"user\",\n\t\t\"content\": []interface{}{\n\t\t\tmap[string]interface{}{\"type\": \"text\", \"text\": \"Transcribe this\"},\n\t\t\tmap[string]interface{}{\"type\": \"audio\", \"source\": \"file://test.wav\"},\n\t\t},\n\t}\n\n\topts := &InputOptions{BaseDir: tmpDir}\n\tmessages, err := ParseInputWithOptions(input, opts)\n\tif err != nil {\n\t\tt.Fatalf(\"ParseInputWithOptions failed: %v\", err)\n\t}\n\n\tparts, ok := messages[0].Content.([]context.ContentPart)\n\tif !ok {\n\t\tt.Fatalf(\"Expected []ContentPart, got %T\", messages[0].Content)\n\t}\n\n\t// Check audio part\n\taudioPart := parts[1]\n\tif audioPart.Type != context.ContentInputAudio {\n\t\tt.Errorf(\"Expected type 'input_audio', got '%s'\", audioPart.Type)\n\t}\n\tif audioPart.InputAudio == nil {\n\t\tt.Fatal(\"Expected InputAudio to be set\")\n\t}\n\tif audioPart.InputAudio.Format != \"wav\" {\n\t\tt.Errorf(\"Expected format 'wav', got '%s'\", audioPart.InputAudio.Format)\n\t}\n\tif audioPart.InputAudio.Data == \"\" {\n\t\tt.Error(\"Expected base64 data to be set\")\n\t}\n}\n\nfunc TestParseInputWithOptions_FileProtocol_File(t *testing.T) {\n\t// Create a temporary test file\n\ttmpDir := t.TempDir()\n\tpdfPath := filepath.Join(tmpDir, \"document.pdf\")\n\n\t// Create a minimal PDF file\n\tpdfData := []byte(\"%PDF-1.4\\n1 0 obj\\n<<>>\\nendobj\\ntrailer\\n<<>>\\n%%EOF\")\n\tif err := os.WriteFile(pdfPath, pdfData, 0644); err != nil {\n\t\tt.Fatalf(\"Failed to create test PDF: %v\", err)\n\t}\n\n\tinput := map[string]interface{}{\n\t\t\"role\": \"user\",\n\t\t\"content\": []interface{}{\n\t\t\tmap[string]interface{}{\"type\": \"text\", \"text\": \"Analyze this document\"},\n\t\t\tmap[string]interface{}{\"type\": \"file\", \"source\": \"file://document.pdf\", \"name\": \"my_doc.pdf\"},\n\t\t},\n\t}\n\n\topts := &InputOptions{BaseDir: tmpDir}\n\tmessages, err := ParseInputWithOptions(input, opts)\n\tif err != nil {\n\t\tt.Fatalf(\"ParseInputWithOptions failed: %v\", err)\n\t}\n\n\tparts, ok := messages[0].Content.([]context.ContentPart)\n\tif !ok {\n\t\tt.Fatalf(\"Expected []ContentPart, got %T\", messages[0].Content)\n\t}\n\n\t// Check file part\n\tfilePart := parts[1]\n\tif filePart.Type != context.ContentFile {\n\t\tt.Errorf(\"Expected type 'file', got '%s'\", filePart.Type)\n\t}\n\tif filePart.File == nil {\n\t\tt.Fatal(\"Expected File to be set\")\n\t}\n\tif filePart.File.Filename != \"my_doc.pdf\" {\n\t\tt.Errorf(\"Expected filename 'my_doc.pdf', got '%s'\", filePart.File.Filename)\n\t}\n\tif !strings.HasPrefix(filePart.File.URL, \"data:application/pdf;base64,\") {\n\t\tt.Errorf(\"Expected base64 data URL with PDF mime type, got '%s'\", filePart.File.URL[:40])\n\t}\n}\n\nfunc TestParseInputWithOptions_FileProtocol_AbsolutePath(t *testing.T) {\n\t// Create a temporary test image file\n\ttmpDir := t.TempDir()\n\timgPath := filepath.Join(tmpDir, \"absolute.png\")\n\n\t// Create a minimal PNG file\n\tpngData := []byte{\n\t\t0x89, 0x50, 0x4E, 0x47, 0x0D, 0x0A, 0x1A, 0x0A,\n\t\t0x00, 0x00, 0x00, 0x0D, 0x49, 0x48, 0x44, 0x52,\n\t\t0x00, 0x00, 0x00, 0x01, 0x00, 0x00, 0x00, 0x01,\n\t\t0x08, 0x02, 0x00, 0x00, 0x00, 0x90, 0x77, 0x53,\n\t\t0xDE, 0x00, 0x00, 0x00, 0x0C, 0x49, 0x44, 0x41,\n\t\t0x54, 0x08, 0xD7, 0x63, 0xF8, 0xCF, 0xC0, 0x00,\n\t\t0x00, 0x00, 0x03, 0x00, 0x01, 0x00, 0x05, 0xFE,\n\t\t0xD4, 0xEF, 0x00, 0x00, 0x00, 0x00, 0x49, 0x45,\n\t\t0x4E, 0x44, 0xAE, 0x42, 0x60, 0x82,\n\t}\n\tif err := os.WriteFile(imgPath, pngData, 0644); err != nil {\n\t\tt.Fatalf(\"Failed to create test image: %v\", err)\n\t}\n\n\t// Use absolute path\n\tinput := map[string]interface{}{\n\t\t\"role\": \"user\",\n\t\t\"content\": []interface{}{\n\t\t\tmap[string]interface{}{\"type\": \"image\", \"source\": \"file://\" + imgPath},\n\t\t},\n\t}\n\n\t// BaseDir should be ignored for absolute paths\n\topts := &InputOptions{BaseDir: \"/some/other/dir\"}\n\tmessages, err := ParseInputWithOptions(input, opts)\n\tif err != nil {\n\t\tt.Fatalf(\"ParseInputWithOptions failed: %v\", err)\n\t}\n\n\tparts, ok := messages[0].Content.([]context.ContentPart)\n\tif !ok {\n\t\tt.Fatalf(\"Expected []ContentPart, got %T\", messages[0].Content)\n\t}\n\n\tif parts[0].Type != context.ContentImageURL {\n\t\tt.Errorf(\"Expected type 'image_url', got '%s'\", parts[0].Type)\n\t}\n}\n\nfunc TestParseInputWithOptions_FileNotFound(t *testing.T) {\n\tinput := map[string]interface{}{\n\t\t\"role\": \"user\",\n\t\t\"content\": []interface{}{\n\t\t\tmap[string]interface{}{\"type\": \"image\", \"source\": \"file://nonexistent.png\"},\n\t\t},\n\t}\n\n\topts := &InputOptions{BaseDir: t.TempDir()}\n\t_, err := ParseInputWithOptions(input, opts)\n\tif err == nil {\n\t\tt.Fatal(\"Expected error for non-existent file\")\n\t}\n\tif !strings.Contains(err.Error(), \"failed to read image file\") {\n\t\tt.Errorf(\"Expected 'failed to read image file' error, got: %v\", err)\n\t}\n}\n\nfunc TestResolveFilePath(t *testing.T) {\n\ttests := []struct {\n\t\tname string\n\t\tfilePath string\n\t\tbaseDir string\n\t\texpected string\n\t}{\n\t\t{\n\t\t\tname: \"relative path with base dir\",\n\t\t\tfilePath: \"fixtures/image.png\",\n\t\t\tbaseDir: \"/app/tests\",\n\t\t\texpected: \"/app/tests/fixtures/image.png\",\n\t\t},\n\t\t{\n\t\t\tname: \"relative path without base dir\",\n\t\t\tfilePath: \"fixtures/image.png\",\n\t\t\tbaseDir: \"\",\n\t\t\texpected: \"fixtures/image.png\",\n\t\t},\n\t\t{\n\t\t\tname: \"absolute path ignores base dir\",\n\t\t\tfilePath: \"/absolute/path/image.png\",\n\t\t\tbaseDir: \"/app/tests\",\n\t\t\texpected: \"/absolute/path/image.png\",\n\t\t},\n\t}\n\n\tfor _, tt := range tests {\n\t\tt.Run(tt.name, func(t *testing.T) {\n\t\t\topts := &InputOptions{BaseDir: tt.baseDir}\n\t\t\tresult := resolveFilePath(tt.filePath, opts)\n\t\t\tif result != tt.expected {\n\t\t\t\tt.Errorf(\"Expected '%s', got '%s'\", tt.expected, result)\n\t\t\t}\n\t\t})\n\t}\n}\n\nfunc TestExtractTextContent(t *testing.T) {\n\ttests := []struct {\n\t\tname string\n\t\tcontent interface{}\n\t\texpected string\n\t}{\n\t\t{\n\t\t\tname: \"string content\",\n\t\t\tcontent: \"Hello world\",\n\t\t\texpected: \"Hello world\",\n\t\t},\n\t\t{\n\t\t\tname: \"content parts array\",\n\t\t\tcontent: []interface{}{\n\t\t\t\tmap[string]interface{}{\"type\": \"text\", \"text\": \"First\"},\n\t\t\t\tmap[string]interface{}{\"type\": \"image\", \"source\": \"file://test.png\"},\n\t\t\t\tmap[string]interface{}{\"type\": \"text\", \"text\": \"Second\"},\n\t\t\t},\n\t\t\texpected: \"First\\nSecond\",\n\t\t},\n\t\t{\n\t\t\tname: \"nil content\",\n\t\t\tcontent: nil,\n\t\t\texpected: \"\",\n\t\t},\n\t}\n\n\tfor _, tt := range tests {\n\t\tt.Run(tt.name, func(t *testing.T) {\n\t\t\tresult := ExtractTextContent(tt.content)\n\t\t\tif result != tt.expected {\n\t\t\t\tt.Errorf(\"Expected '%s', got '%s'\", tt.expected, result)\n\t\t\t}\n\t\t})\n\t}\n}\n\nfunc TestSummarizeInput(t *testing.T) {\n\ttests := []struct {\n\t\tname string\n\t\tinput interface{}\n\t\tmaxLen int\n\t\texpected string\n\t}{\n\t\t{\n\t\t\tname: \"short string\",\n\t\t\tinput: \"Hello\",\n\t\t\tmaxLen: 10,\n\t\t\texpected: \"Hello\",\n\t\t},\n\t\t{\n\t\t\tname: \"long string truncated\",\n\t\t\tinput: \"This is a very long message that should be truncated\",\n\t\t\tmaxLen: 20,\n\t\t\texpected: \"This is a very lo...\",\n\t\t},\n\t\t{\n\t\t\tname: \"message array - last user message\",\n\t\t\tinput: []interface{}{\n\t\t\t\tmap[string]interface{}{\"role\": \"user\", \"content\": \"First\"},\n\t\t\t\tmap[string]interface{}{\"role\": \"assistant\", \"content\": \"Response\"},\n\t\t\t\tmap[string]interface{}{\"role\": \"user\", \"content\": \"Last user message\"},\n\t\t\t},\n\t\t\tmaxLen: 50,\n\t\t\texpected: \"Last user message\",\n\t\t},\n\t}\n\n\tfor _, tt := range tests {\n\t\tt.Run(tt.name, func(t *testing.T) {\n\t\t\tresult := SummarizeInput(tt.input, tt.maxLen)\n\t\t\tif result != tt.expected {\n\t\t\t\tt.Errorf(\"Expected '%s', got '%s'\", tt.expected, result)\n\t\t\t}\n\t\t})\n\t}\n}\n" }, { "path": "agent/test/interfaces.go", "content": "package test\n\nimport (\n\t\"context\"\n\t\"io\"\n)\n\n// Runner is the interface for test execution\ntype Runner interface {\n\t// Run executes all test cases and returns the report\n\tRun(ctx context.Context) (*Report, error)\n\n\t// RunCase executes a single test case\n\tRunCase(ctx context.Context, tc *Case) (*Result, error)\n\n\t// GetAgentInfo returns information about the agent being tested\n\tGetAgentInfo() *AgentInfo\n\n\t// SetProgressCallback sets a callback for progress updates\n\tSetProgressCallback(callback ProgressCallback)\n}\n\n// ProgressCallback is called during test execution to report progress\n// Parameters:\n// - current: current test index (1-based)\n// - total: total number of tests\n// - result: result of the current test (nil if not yet completed)\ntype ProgressCallback func(current, total int, result *Result)\n\n// Reporter is the interface for generating test reports\ntype Reporter interface {\n\t// Generate generates a report from the test results\n\tGenerate(report *Report) error\n\n\t// Write writes the report to the given writer\n\tWrite(report *Report, w io.Writer) error\n}\n\n// Loader is the interface for loading test cases\ntype Loader interface {\n\t// Load loads test cases from the input source\n\tLoad() ([]*Case, error)\n\n\t// LoadFile loads test cases from a JSONL file\n\tLoadFile(path string) ([]*Case, error)\n\n\t// LoadFromAgent generates test cases using a generator agent\n\tLoadFromAgent(agentID string, targetInfo *TargetAgentInfo, params map[string]interface{}) ([]*Case, error)\n\n\t// LoadFromScript generates test cases using a script\n\tLoadFromScript(scriptRef string, targetInfo *TargetAgentInfo) ([]*Case, error)\n}\n\n// Resolver is the interface for resolving agent information\ntype Resolver interface {\n\t// Resolve resolves the agent from options\n\t// Priority: explicit AgentID > path-based detection\n\tResolve(opts *Options) (*AgentInfo, error)\n\n\t// ResolveFromPath resolves the agent by traversing up from the input file path\n\tResolveFromPath(inputPath string) (*AgentInfo, error)\n}\n\n// Validator is the interface for validating test outputs\ntype Validator interface {\n\t// Validate compares actual output against expected output\n\t// Returns nil if validation passes, error otherwise\n\tValidate(actual, expected interface{}) error\n\n\t// ValidateJSON validates JSON outputs with flexible comparison\n\tValidateJSON(actual, expected interface{}) error\n}\n\n// OutputAdapter adapts agent output to a comparable format\ntype OutputAdapter interface {\n\t// Adapt transforms the raw agent output to a normalized format\n\tAdapt(output interface{}) (interface{}, error)\n}\n\n// RunnerFactory creates Runner instances\ntype RunnerFactory interface {\n\t// Create creates a new Runner with the given options\n\tCreate(opts *Options) (Runner, error)\n}\n\n// ReporterFactory creates Reporter instances\ntype ReporterFactory interface {\n\t// Create creates a new Reporter for the given format\n\tCreate(format OutputFormat) (Reporter, error)\n\n\t// CreateFromPath creates a Reporter based on output file extension\n\tCreateFromPath(outputPath string) (Reporter, error)\n}\n\n// Hook allows customization of test execution\ntype Hook interface {\n\t// BeforeAll is called before any tests run\n\tBeforeAll(ctx context.Context, cases []*Case) error\n\n\t// BeforeEach is called before each test case\n\tBeforeEach(ctx context.Context, tc *Case) error\n\n\t// AfterEach is called after each test case\n\tAfterEach(ctx context.Context, tc *Case, result *Result) error\n\n\t// AfterAll is called after all tests complete\n\tAfterAll(ctx context.Context, report *Report) error\n}\n\n// DefaultHook provides a no-op implementation of Hook\ntype DefaultHook struct{}\n\n// BeforeAll implements Hook\nfunc (h *DefaultHook) BeforeAll(ctx context.Context, cases []*Case) error {\n\treturn nil\n}\n\n// BeforeEach implements Hook\nfunc (h *DefaultHook) BeforeEach(ctx context.Context, tc *Case) error {\n\treturn nil\n}\n\n// AfterEach implements Hook\nfunc (h *DefaultHook) AfterEach(ctx context.Context, tc *Case, result *Result) error {\n\treturn nil\n}\n\n// AfterAll implements Hook\nfunc (h *DefaultHook) AfterAll(ctx context.Context, report *Report) error {\n\treturn nil\n}\n" }, { "path": "agent/test/loader.go", "content": "package test\n\nimport (\n\t\"bufio\"\n\t\"fmt\"\n\t\"os\"\n\t\"regexp\"\n\t\"strings\"\n\t\"time\"\n\n\tjsoniter \"github.com/json-iterator/go\"\n\t\"github.com/yaoapp/gou/process\"\n)\n\n// JSONLLoader loads test cases from JSONL files\ntype JSONLLoader struct{}\n\n// NewLoader creates a new JSONL loader\nfunc NewLoader() Loader {\n\treturn &JSONLLoader{}\n}\n\n// Load loads test cases from the default input source\n// This is a placeholder - actual implementation would use configured path\nfunc (l *JSONLLoader) Load() ([]*Case, error) {\n\treturn nil, fmt.Errorf(\"Load() requires explicit path, use LoadFile() instead\")\n}\n\n// LoadFile loads test cases from a JSONL file\n// If path is relative and YAO_ROOT is set, resolves relative to YAO_ROOT\nfunc (l *JSONLLoader) LoadFile(path string) ([]*Case, error) {\n\t// Resolve path relative to YAO_ROOT if it's a relative path\n\tresolvedPath := ResolvePathWithYaoRoot(path)\n\n\tfile, err := os.Open(resolvedPath)\n\tif err != nil {\n\t\treturn nil, fmt.Errorf(\"failed to open file %s: %w\", path, err)\n\t}\n\tdefer file.Close()\n\n\tvar cases []*Case\n\tscanner := bufio.NewScanner(file)\n\tlineNum := 0\n\n\t// Increase buffer size for long lines\n\tconst maxCapacity = 1024 * 1024 // 1MB\n\tbuf := make([]byte, maxCapacity)\n\tscanner.Buffer(buf, maxCapacity)\n\n\tfor scanner.Scan() {\n\t\tlineNum++\n\t\tline := strings.TrimSpace(scanner.Text())\n\n\t\t// Skip empty lines\n\t\tif line == \"\" {\n\t\t\tcontinue\n\t\t}\n\n\t\t// Skip comments (lines starting with //)\n\t\tif strings.HasPrefix(line, \"//\") {\n\t\t\tcontinue\n\t\t}\n\n\t\tvar tc Case\n\t\tif err := jsoniter.UnmarshalFromString(line, &tc); err != nil {\n\t\t\treturn nil, fmt.Errorf(\"failed to parse line %d: %w\", lineNum, err)\n\t\t}\n\n\t\t// Validate required fields\n\t\tif tc.ID == \"\" {\n\t\t\treturn nil, fmt.Errorf(\"line %d: missing required field 'id'\", lineNum)\n\t\t}\n\t\tif tc.Input == nil {\n\t\t\treturn nil, fmt.Errorf(\"line %d (id=%s): missing required field 'input'\", lineNum, tc.ID)\n\t\t}\n\n\t\tcases = append(cases, &tc)\n\t}\n\n\tif err := scanner.Err(); err != nil {\n\t\treturn nil, fmt.Errorf(\"error reading file: %w\", err)\n\t}\n\n\tif len(cases) == 0 {\n\t\treturn nil, fmt.Errorf(\"no test cases found in %s\", path)\n\t}\n\n\treturn cases, nil\n}\n\n// ValidateTestCases validates a slice of test cases\nfunc ValidateTestCases(cases []*Case) error {\n\tids := make(map[string]bool)\n\n\tfor i, tc := range cases {\n\t\t// Check for duplicate IDs\n\t\tif ids[tc.ID] {\n\t\t\treturn fmt.Errorf(\"duplicate test case ID: %s\", tc.ID)\n\t\t}\n\t\tids[tc.ID] = true\n\n\t\t// Validate input can be parsed\n\t\tif _, err := tc.GetMessages(); err != nil {\n\t\t\treturn fmt.Errorf(\"test case %s (index %d): invalid input: %w\", tc.ID, i, err)\n\t\t}\n\n\t\t// Validate timeout format if specified\n\t\tif tc.Timeout != \"\" {\n\t\t\t// GetTimeout returns a duration, parsing error would return default\n\t\t\t// We validate by checking if the string is parseable\n\t\t\tif _, err := time.ParseDuration(tc.Timeout); err != nil {\n\t\t\t\treturn fmt.Errorf(\"test case %s: invalid timeout format: %s\", tc.ID, tc.Timeout)\n\t\t\t}\n\t\t}\n\t}\n\n\treturn nil\n}\n\n// FilterTestCases filters test cases based on criteria\nfunc FilterTestCases(cases []*Case, filter func(*Case) bool) []*Case {\n\tvar result []*Case\n\tfor _, tc := range cases {\n\t\tif filter(tc) {\n\t\t\tresult = append(result, tc)\n\t\t}\n\t}\n\treturn result\n}\n\n// FilterSkipped returns test cases that are not skipped\nfunc FilterSkipped(cases []*Case) []*Case {\n\treturn FilterTestCases(cases, func(tc *Case) bool {\n\t\treturn !tc.Skip\n\t})\n}\n\n// FilterByIDs returns test cases matching the given IDs\nfunc FilterByIDs(cases []*Case, ids []string) []*Case {\n\tidSet := make(map[string]bool)\n\tfor _, id := range ids {\n\t\tidSet[id] = true\n\t}\n\treturn FilterTestCases(cases, func(tc *Case) bool {\n\t\treturn idSet[tc.ID]\n\t})\n}\n\n// FilterByPattern returns test cases whose ID matches the given regex pattern\nfunc FilterByPattern(cases []*Case, pattern *regexp.Regexp) []*Case {\n\treturn FilterTestCases(cases, func(tc *Case) bool {\n\t\treturn pattern.MatchString(tc.ID)\n\t})\n}\n\n// LoadFromAgent generates test cases using a generator agent\nfunc (l *JSONLLoader) LoadFromAgent(agentID string, targetInfo *TargetAgentInfo, params map[string]interface{}) ([]*Case, error) {\n\treturn GenerateTestCases(agentID, targetInfo, params)\n}\n\n// LoadFromScript generates test cases using a script\n// scriptRef format: \"module.FunctionName\" (e.g., \"tests.gen.Generate\")\nfunc (l *JSONLLoader) LoadFromScript(scriptRef string, targetInfo *TargetAgentInfo) ([]*Case, error) {\n\t// Parse script reference\n\tparts := strings.Split(scriptRef, \".\")\n\tif len(parts) < 2 {\n\t\treturn nil, fmt.Errorf(\"invalid script reference format: %s (expected 'module.Function')\", scriptRef)\n\t}\n\n\t// Build process name: scripts.module.Function\n\tprocessName := \"scripts.\" + scriptRef\n\n\t// Execute via process\n\tp, err := process.Of(processName, targetInfo)\n\tif err != nil {\n\t\treturn nil, fmt.Errorf(\"failed to create process %s: %w\", processName, err)\n\t}\n\n\tresult, err := p.Exec()\n\tif err != nil {\n\t\treturn nil, fmt.Errorf(\"script execution failed: %w\", err)\n\t}\n\n\t// Parse result as test cases\n\treturn convertToCases(result)\n}\n" }, { "path": "agent/test/output.go", "content": "package test\n\nimport (\n\t\"fmt\"\n\t\"strings\"\n\t\"time\"\n\n\t\"github.com/fatih/color\"\n\tjsoniter \"github.com/json-iterator/go\"\n)\n\n// OutputWriter handles colored console output for test execution\ntype OutputWriter struct {\n\tverbose bool\n}\n\n// NewOutputWriter creates a new output writer\nfunc NewOutputWriter(verbose bool) *OutputWriter {\n\treturn &OutputWriter{verbose: verbose}\n}\n\n// Header prints a header section\nfunc (w *OutputWriter) Header(title string) {\n\tfmt.Println()\n\tcolor.New(color.FgCyan, color.Bold).Println(\"═══════════════════════════════════════════════════════════════\")\n\tcolor.New(color.FgCyan, color.Bold).Printf(\" %s\\n\", title)\n\tcolor.New(color.FgCyan, color.Bold).Println(\"═══════════════════════════════════════════════════════════════\")\n}\n\n// SubHeader prints a sub-header\nfunc (w *OutputWriter) SubHeader(title string) {\n\tfmt.Println()\n\tcolor.New(color.FgWhite, color.Bold).Println(\"───────────────────────────────────────────────────────────────\")\n\tcolor.New(color.FgWhite, color.Bold).Printf(\" %s\\n\", title)\n\tcolor.New(color.FgWhite, color.Bold).Println(\"───────────────────────────────────────────────────────────────\")\n}\n\n// Info prints an info message\nfunc (w *OutputWriter) Info(format string, args ...interface{}) {\n\tcolor.New(color.FgBlue).Printf(\"ℹ \")\n\tfmt.Printf(format+\"\\n\", args...)\n}\n\n// Success prints a success message\nfunc (w *OutputWriter) Success(format string, args ...interface{}) {\n\tcolor.New(color.FgGreen).Printf(\"✓ \")\n\tfmt.Printf(format+\"\\n\", args...)\n}\n\n// Error prints an error message\nfunc (w *OutputWriter) Error(format string, args ...interface{}) {\n\tcolor.New(color.FgRed).Printf(\"✗ \")\n\tfmt.Printf(format+\"\\n\", args...)\n}\n\n// Warning prints a warning message\nfunc (w *OutputWriter) Warning(format string, args ...interface{}) {\n\tcolor.New(color.FgYellow).Printf(\"⚠ \")\n\tfmt.Printf(format+\"\\n\", args...)\n}\n\n// Skip prints a skip message\nfunc (w *OutputWriter) Skip(format string, args ...interface{}) {\n\tcolor.New(color.FgYellow).Printf(\"○ \")\n\tfmt.Printf(format+\"\\n\", args...)\n}\n\n// Verbose prints a verbose message (only if verbose mode is enabled)\nfunc (w *OutputWriter) Verbose(format string, args ...interface{}) {\n\tif w.verbose {\n\t\tcolor.New(color.FgHiBlack).Printf(\" │ \")\n\t\tfmt.Printf(format+\"\\n\", args...)\n\t}\n}\n\n// TestStart prints test case start\nfunc (w *OutputWriter) TestStart(id string, input string, runNum int) {\n\tinputPreview := truncateString(input, 50)\n\tif runNum > 1 {\n\t\tcolor.New(color.FgWhite).Printf(\"► [%s] Run %d: \", id, runNum)\n\t} else {\n\t\tcolor.New(color.FgWhite).Printf(\"► [%s] \", id)\n\t}\n\tcolor.New(color.FgHiBlack).Printf(\"%s\", inputPreview)\n\tfmt.Print(\" \")\n}\n\n// TestResult prints test case result\nfunc (w *OutputWriter) TestResult(status Status, duration time.Duration) {\n\tswitch status {\n\tcase StatusPassed:\n\t\tcolor.New(color.FgGreen, color.Bold).Printf(\"PASSED\")\n\tcase StatusFailed:\n\t\tcolor.New(color.FgRed, color.Bold).Printf(\"FAILED\")\n\tcase StatusSkipped:\n\t\tcolor.New(color.FgYellow).Printf(\"SKIPPED\")\n\tcase StatusError:\n\t\tcolor.New(color.FgRed, color.Bold).Printf(\"ERROR\")\n\tcase StatusTimeout:\n\t\tcolor.New(color.FgRed).Printf(\"TIMEOUT\")\n\t}\n\tcolor.New(color.FgHiBlack).Printf(\" (%s)\\n\", formatDuration(duration))\n}\n\n// TestError prints test error details\nfunc (w *OutputWriter) TestError(err string) {\n\tcolor.New(color.FgRed).Printf(\" └─ %s\\n\", err)\n}\n\n// TestOutput prints test output (verbose mode)\nfunc (w *OutputWriter) TestOutput(output string) {\n\tif w.verbose && output != \"\" {\n\t\toutputPreview := truncateString(output, 100)\n\t\tcolor.New(color.FgHiBlack).Printf(\" └─ Output: %s\\n\", outputPreview)\n\t}\n}\n\n// Progress prints progress information\nfunc (w *OutputWriter) Progress(current, total int) {\n\tpercentage := float64(current) / float64(total) * 100\n\tcolor.New(color.FgHiBlack).Printf(\"\\r Progress: %d/%d (%.0f%%)\", current, total, percentage)\n}\n\n// Summary prints the test summary\nfunc (w *OutputWriter) Summary(summary *Summary, duration time.Duration) {\n\tw.SubHeader(\"Summary\")\n\n\t// Agent info\n\tcolor.New(color.FgWhite).Printf(\" Agent: \")\n\tcolor.New(color.FgCyan).Printf(\"%s\\n\", summary.AgentID)\n\n\tif summary.Connector != \"\" {\n\t\tcolor.New(color.FgWhite).Printf(\" Connector: \")\n\t\tcolor.New(color.FgCyan).Printf(\"%s\\n\", summary.Connector)\n\t}\n\n\t// Results\n\tcolor.New(color.FgWhite).Printf(\" Total: \")\n\tfmt.Printf(\"%d\\n\", summary.Total)\n\n\tcolor.New(color.FgWhite).Printf(\" Passed: \")\n\tif summary.Passed > 0 {\n\t\tcolor.New(color.FgGreen).Printf(\"%d\\n\", summary.Passed)\n\t} else {\n\t\tfmt.Printf(\"%d\\n\", summary.Passed)\n\t}\n\n\tcolor.New(color.FgWhite).Printf(\" Failed: \")\n\tif summary.Failed > 0 {\n\t\tcolor.New(color.FgRed).Printf(\"%d\\n\", summary.Failed)\n\t} else {\n\t\tfmt.Printf(\"%d\\n\", summary.Failed)\n\t}\n\n\tif summary.Skipped > 0 {\n\t\tcolor.New(color.FgWhite).Printf(\" Skipped: \")\n\t\tcolor.New(color.FgYellow).Printf(\"%d\\n\", summary.Skipped)\n\t}\n\n\tif summary.Errors > 0 {\n\t\tcolor.New(color.FgWhite).Printf(\" Errors: \")\n\t\tcolor.New(color.FgRed).Printf(\"%d\\n\", summary.Errors)\n\t}\n\n\tif summary.Timeouts > 0 {\n\t\tcolor.New(color.FgWhite).Printf(\" Timeouts: \")\n\t\tcolor.New(color.FgRed).Printf(\"%d\\n\", summary.Timeouts)\n\t}\n\n\t// Pass rate\n\tpassRate := float64(0)\n\tif summary.Total > 0 {\n\t\tpassRate = float64(summary.Passed) / float64(summary.Total) * 100\n\t}\n\tcolor.New(color.FgWhite).Printf(\" Pass Rate: \")\n\tif passRate == 100 {\n\t\tcolor.New(color.FgGreen, color.Bold).Printf(\"%.1f%%\\n\", passRate)\n\t} else if passRate >= 80 {\n\t\tcolor.New(color.FgYellow).Printf(\"%.1f%%\\n\", passRate)\n\t} else {\n\t\tcolor.New(color.FgRed).Printf(\"%.1f%%\\n\", passRate)\n\t}\n\n\t// Duration\n\tcolor.New(color.FgWhite).Printf(\" Duration: \")\n\tfmt.Printf(\"%s\\n\", formatDuration(duration))\n\n\t// Stability info (if runs > 1)\n\tif summary.RunsPerCase > 1 {\n\t\tfmt.Println()\n\t\tcolor.New(color.FgWhite, color.Bold).Println(\" Stability Analysis:\")\n\t\tcolor.New(color.FgWhite).Printf(\" Runs/Case: %d\\n\", summary.RunsPerCase)\n\t\tcolor.New(color.FgWhite).Printf(\" Total Runs: %d\\n\", summary.TotalRuns)\n\t\tcolor.New(color.FgWhite).Printf(\" Stable Cases: \")\n\t\tif summary.StableCases == summary.Total {\n\t\t\tcolor.New(color.FgGreen).Printf(\"%d\\n\", summary.StableCases)\n\t\t} else {\n\t\t\tcolor.New(color.FgYellow).Printf(\"%d\\n\", summary.StableCases)\n\t\t}\n\t\tcolor.New(color.FgWhite).Printf(\" Unstable: \")\n\t\tif summary.UnstableCases > 0 {\n\t\t\tcolor.New(color.FgRed).Printf(\"%d\\n\", summary.UnstableCases)\n\t\t} else {\n\t\t\tfmt.Printf(\"%d\\n\", summary.UnstableCases)\n\t\t}\n\t}\n}\n\n// OutputFile prints the output file path\nfunc (w *OutputWriter) OutputFile(path string) {\n\tfmt.Println()\n\tcolor.New(color.FgWhite).Printf(\" Output: \")\n\tcolor.New(color.FgCyan).Printf(\"%s\\n\", path)\n}\n\n// FinalResult prints the final result banner\nfunc (w *OutputWriter) FinalResult(passed bool) {\n\tfmt.Println()\n\tif passed {\n\t\tcolor.New(color.FgGreen, color.Bold).Println(\"═══════════════════════════════════════════════════════════════\")\n\t\tcolor.New(color.FgGreen, color.Bold).Println(\" ✨ ALL TESTS PASSED ✨\")\n\t\tcolor.New(color.FgGreen, color.Bold).Println(\"═══════════════════════════════════════════════════════════════\")\n\t} else {\n\t\tcolor.New(color.FgRed, color.Bold).Println(\"═══════════════════════════════════════════════════════════════\")\n\t\tcolor.New(color.FgRed, color.Bold).Println(\" ❌ TESTS FAILED\")\n\t\tcolor.New(color.FgRed, color.Bold).Println(\"═══════════════════════════════════════════════════════════════\")\n\t}\n\tfmt.Println()\n}\n\n// DirectOutput prints the agent output directly (for development mode)\nfunc (w *OutputWriter) DirectOutput(output interface{}) {\n\tif output == nil {\n\t\treturn\n\t}\n\n\t// Try to format as JSON if it's a complex type\n\tswitch v := output.(type) {\n\tcase string:\n\t\tfmt.Println(v)\n\tcase map[string]interface{}, []interface{}:\n\t\t// Pretty print JSON\n\t\tjsonBytes, err := jsoniter.MarshalIndent(v, \"\", \" \")\n\t\tif err != nil {\n\t\t\tfmt.Printf(\"%v\\n\", output)\n\t\t} else {\n\t\t\tfmt.Println(string(jsonBytes))\n\t\t}\n\tdefault:\n\t\t// Try to marshal as JSON\n\t\tjsonBytes, err := jsoniter.MarshalIndent(output, \"\", \" \")\n\t\tif err != nil {\n\t\t\tfmt.Printf(\"%v\\n\", output)\n\t\t} else {\n\t\t\tfmt.Println(string(jsonBytes))\n\t\t}\n\t}\n}\n\n// ScriptTestSummary prints the script test summary\nfunc (w *OutputWriter) ScriptTestSummary(summary *ScriptTestSummary, duration time.Duration) {\n\tw.SubHeader(\"Summary\")\n\n\t// Results\n\tcolor.New(color.FgWhite).Printf(\" Total: \")\n\tfmt.Printf(\"%d\\n\", summary.Total)\n\n\tcolor.New(color.FgWhite).Printf(\" Passed: \")\n\tif summary.Passed > 0 {\n\t\tcolor.New(color.FgGreen).Printf(\"%d\\n\", summary.Passed)\n\t} else {\n\t\tfmt.Printf(\"%d\\n\", summary.Passed)\n\t}\n\n\tcolor.New(color.FgWhite).Printf(\" Failed: \")\n\tif summary.Failed > 0 {\n\t\tcolor.New(color.FgRed).Printf(\"%d\\n\", summary.Failed)\n\t} else {\n\t\tfmt.Printf(\"%d\\n\", summary.Failed)\n\t}\n\n\tif summary.Skipped > 0 {\n\t\tcolor.New(color.FgWhite).Printf(\" Skipped: \")\n\t\tcolor.New(color.FgYellow).Printf(\"%d\\n\", summary.Skipped)\n\t}\n\n\t// Pass rate\n\tpassRate := float64(0)\n\tif summary.Total > 0 {\n\t\tpassRate = float64(summary.Passed) / float64(summary.Total) * 100\n\t}\n\tcolor.New(color.FgWhite).Printf(\" Pass Rate: \")\n\tif passRate == 100 {\n\t\tcolor.New(color.FgGreen, color.Bold).Printf(\"%.1f%%\\n\", passRate)\n\t} else if passRate >= 80 {\n\t\tcolor.New(color.FgYellow).Printf(\"%.1f%%\\n\", passRate)\n\t} else {\n\t\tcolor.New(color.FgRed).Printf(\"%.1f%%\\n\", passRate)\n\t}\n\n\t// Duration\n\tcolor.New(color.FgWhite).Printf(\" Duration: \")\n\tfmt.Printf(\"%s\\n\", formatDuration(duration))\n}\n\n// DynamicTestStart outputs the start of a dynamic test\nfunc (w *OutputWriter) DynamicTestStart(id string, checkpointCount int) {\n\tcolor.New(color.FgWhite).Printf(\"► [%s] \", id)\n\tcolor.New(color.FgCyan).Printf(\"(dynamic, %d checkpoints)\\n\", checkpointCount)\n}\n\n// DynamicTurn outputs a single turn in dynamic testing\nfunc (w *OutputWriter) DynamicTurn(turn int, inputSummary string, checkpointsReached, total int) {\n\tif w.verbose {\n\t\tcolor.New(color.FgHiBlack).Printf(\"│ ├─ Turn %d: %s \", turn, inputSummary)\n\t\tcolor.New(color.FgCyan).Printf(\"[%d/%d checkpoints]\\n\", checkpointsReached, total)\n\t}\n}\n\n// DynamicCheckpoint outputs a checkpoint being reached\nfunc (w *OutputWriter) DynamicCheckpoint(checkpointID string) {\n\tif w.verbose {\n\t\tcolor.New(color.FgGreen).Printf(\"│ │ └─ ✓ checkpoint: %s\\n\", checkpointID)\n\t}\n}\n\n// DynamicTestResult outputs the result of a dynamic test\nfunc (w *OutputWriter) DynamicTestResult(status Status, turns int, checkpoints int, duration time.Duration) {\n\tcolor.New(color.FgHiBlack).Printf(\" └─ \")\n\n\tswitch status {\n\tcase StatusPassed:\n\t\tcolor.New(color.FgGreen).Printf(\"PASSED\")\n\tcase StatusFailed:\n\t\tcolor.New(color.FgRed).Printf(\"FAILED\")\n\tcase StatusError:\n\t\tcolor.New(color.FgRed).Printf(\"ERROR\")\n\tcase StatusTimeout:\n\t\tcolor.New(color.FgRed).Printf(\"TIMEOUT\")\n\t}\n\n\tcolor.New(color.FgHiBlack).Printf(\" (%d turns, %d checkpoints, %s)\\n\", turns, checkpoints, formatDuration(duration))\n}\n\n// StabilityResult prints stability analysis result for a test case\nfunc (w *OutputWriter) StabilityResult(sr *StabilityResult) {\n\tcolor.New(color.FgWhite).Printf(\" [%s] \", sr.ID)\n\n\t// Pass rate\n\tif sr.PassRate == 100 {\n\t\tcolor.New(color.FgGreen).Printf(\"%.0f%%\", sr.PassRate)\n\t} else if sr.PassRate >= 80 {\n\t\tcolor.New(color.FgYellow).Printf(\"%.0f%%\", sr.PassRate)\n\t} else {\n\t\tcolor.New(color.FgRed).Printf(\"%.0f%%\", sr.PassRate)\n\t}\n\n\t// Classification\n\tcolor.New(color.FgHiBlack).Printf(\" (%d/%d) \", sr.Passed, sr.Runs)\n\n\tswitch sr.StabilityClass {\n\tcase StabilityStable:\n\t\tcolor.New(color.FgGreen).Printf(\"Stable\")\n\tcase StabilityMostlyStable:\n\t\tcolor.New(color.FgYellow).Printf(\"Mostly Stable\")\n\tcase StabilityUnstable:\n\t\tcolor.New(color.FgRed).Printf(\"Unstable\")\n\tcase StabilityHighlyUnstable:\n\t\tcolor.New(color.FgRed, color.Bold).Printf(\"Highly Unstable\")\n\t}\n\n\t// Timing\n\tcolor.New(color.FgHiBlack).Printf(\" avg:%.0fms\\n\", sr.AvgDurationMs)\n}\n\n// Helper functions\n\nfunc truncateString(s string, maxLen int) string {\n\t// Remove newlines and extra spaces\n\ts = strings.ReplaceAll(s, \"\\n\", \" \")\n\ts = strings.ReplaceAll(s, \"\\r\", \"\")\n\ts = strings.Join(strings.Fields(s), \" \")\n\n\tif len(s) <= maxLen {\n\t\treturn s\n\t}\n\treturn s[:maxLen-3] + \"...\"\n}\n\nfunc formatDuration(d time.Duration) string {\n\tif d < time.Millisecond {\n\t\treturn fmt.Sprintf(\"%dµs\", d.Microseconds())\n\t}\n\tif d < time.Second {\n\t\treturn fmt.Sprintf(\"%dms\", d.Milliseconds())\n\t}\n\tif d < time.Minute {\n\t\treturn fmt.Sprintf(\"%.1fs\", d.Seconds())\n\t}\n\treturn fmt.Sprintf(\"%.1fm\", d.Minutes())\n}\n" }, { "path": "agent/test/reporter.go", "content": "package test\n\nimport (\n\t\"bufio\"\n\t\"fmt\"\n\t\"html/template\"\n\t\"io\"\n\t\"strings\"\n\t\"time\"\n\n\tjsoniter \"github.com/json-iterator/go\"\n\t\"github.com/yaoapp/yao/agent/caller\"\n\t\"github.com/yaoapp/yao/agent/context\"\n)\n\n// JSONLReporter generates JSONL format reports (default)\ntype JSONLReporter struct{}\n\n// NewJSONLReporter creates a new JSONL reporter\nfunc NewJSONLReporter() *JSONLReporter {\n\treturn &JSONLReporter{}\n}\n\n// Generate generates a JSONL report (writes to stdout or file)\nfunc (r *JSONLReporter) Generate(report *Report) error {\n\treturn nil // JSONL is written during test execution\n}\n\n// Write writes the report in JSONL format\nfunc (r *JSONLReporter) Write(report *Report, w io.Writer) error {\n\twriter := bufio.NewWriter(w)\n\tdefer writer.Flush()\n\n\t// Start event\n\tstartEvent := map[string]interface{}{\n\t\t\"type\": \"start\",\n\t\t\"timestamp\": report.Metadata.StartedAt.Format(time.RFC3339),\n\t\t\"agent_id\": report.Summary.AgentID,\n\t\t\"total_cases\": report.Summary.Total,\n\t}\n\tif err := writeJSONLineToWriter(writer, startEvent); err != nil {\n\t\treturn err\n\t}\n\n\t// Result events\n\tif report.Results != nil {\n\t\tfor _, result := range report.Results {\n\t\t\tresultEvent := map[string]interface{}{\n\t\t\t\t\"type\": \"result\",\n\t\t\t\t\"id\": result.ID,\n\t\t\t\t\"status\": result.Status,\n\t\t\t\t\"duration_ms\": result.DurationMs,\n\t\t\t}\n\t\t\tif result.Output != nil {\n\t\t\t\tresultEvent[\"output\"] = result.Output\n\t\t\t}\n\t\t\tif result.Error != \"\" {\n\t\t\t\tresultEvent[\"error\"] = result.Error\n\t\t\t}\n\t\t\tif err := writeJSONLineToWriter(writer, resultEvent); err != nil {\n\t\t\t\treturn err\n\t\t\t}\n\t\t}\n\t}\n\n\t// Stability results\n\tif report.StabilityResults != nil {\n\t\tfor _, sr := range report.StabilityResults {\n\t\t\tstabilityEvent := map[string]interface{}{\n\t\t\t\t\"type\": \"stability\",\n\t\t\t\t\"id\": sr.ID,\n\t\t\t\t\"runs\": sr.Runs,\n\t\t\t\t\"passed\": sr.Passed,\n\t\t\t\t\"failed\": sr.Failed,\n\t\t\t\t\"pass_rate\": sr.PassRate,\n\t\t\t\t\"stable\": sr.Stable,\n\t\t\t\t\"stability_class\": sr.StabilityClass,\n\t\t\t\t\"avg_duration_ms\": sr.AvgDurationMs,\n\t\t\t}\n\t\t\tif err := writeJSONLineToWriter(writer, stabilityEvent); err != nil {\n\t\t\t\treturn err\n\t\t\t}\n\t\t}\n\t}\n\n\t// Summary event\n\tsummaryEvent := map[string]interface{}{\n\t\t\"type\": \"summary\",\n\t\t\"total\": report.Summary.Total,\n\t\t\"passed\": report.Summary.Passed,\n\t\t\"failed\": report.Summary.Failed,\n\t\t\"skipped\": report.Summary.Skipped,\n\t\t\"errors\": report.Summary.Errors,\n\t\t\"timeouts\": report.Summary.Timeouts,\n\t\t\"duration_ms\": report.Summary.DurationMs,\n\t}\n\tif report.Summary.RunsPerCase > 1 {\n\t\tsummaryEvent[\"runs_per_case\"] = report.Summary.RunsPerCase\n\t\tsummaryEvent[\"total_runs\"] = report.Summary.TotalRuns\n\t\tsummaryEvent[\"overall_pass_rate\"] = report.Summary.OverallPassRate\n\t\tsummaryEvent[\"stable_cases\"] = report.Summary.StableCases\n\t\tsummaryEvent[\"unstable_cases\"] = report.Summary.UnstableCases\n\t}\n\treturn writeJSONLineToWriter(writer, summaryEvent)\n}\n\n// writeJSONLineToWriter writes a JSON line to the writer\nfunc writeJSONLineToWriter(writer *bufio.Writer, data interface{}) error {\n\tline, err := jsoniter.Marshal(data)\n\tif err != nil {\n\t\treturn err\n\t}\n\t_, err = writer.Write(line)\n\tif err != nil {\n\t\treturn err\n\t}\n\t_, err = writer.WriteString(\"\\n\")\n\treturn err\n}\n\n// JSONReporter generates full JSON format reports\ntype JSONReporter struct{}\n\n// NewJSONReporter creates a new JSON reporter\nfunc NewJSONReporter() *JSONReporter {\n\treturn &JSONReporter{}\n}\n\n// Generate generates a JSON report\nfunc (r *JSONReporter) Generate(report *Report) error {\n\treturn nil\n}\n\n// Write writes the report in JSON format\nfunc (r *JSONReporter) Write(report *Report, w io.Writer) error {\n\tencoder := jsoniter.NewEncoder(w)\n\tencoder.SetIndent(\"\", \" \")\n\treturn encoder.Encode(report)\n}\n\n// MarkdownReporter generates Markdown format reports\ntype MarkdownReporter struct{}\n\n// NewMarkdownReporter creates a new Markdown reporter\nfunc NewMarkdownReporter() *MarkdownReporter {\n\treturn &MarkdownReporter{}\n}\n\n// Generate generates a Markdown report\nfunc (r *MarkdownReporter) Generate(report *Report) error {\n\treturn nil\n}\n\n// Write writes the report in Markdown format\nfunc (r *MarkdownReporter) Write(report *Report, w io.Writer) error {\n\tvar sb strings.Builder\n\n\t// Header\n\tsb.WriteString(\"# Agent Test Report\\n\\n\")\n\n\t// Summary\n\tsb.WriteString(\"## Summary\\n\\n\")\n\tsb.WriteString(\"| Metric | Value |\\n\")\n\tsb.WriteString(\"| ------ | ----- |\\n\")\n\tsb.WriteString(fmt.Sprintf(\"| Agent | %s |\\n\", report.Summary.AgentID))\n\tif report.Summary.Connector != \"\" {\n\t\tsb.WriteString(fmt.Sprintf(\"| Connector | %s |\\n\", report.Summary.Connector))\n\t}\n\tsb.WriteString(fmt.Sprintf(\"| Total | %d |\\n\", report.Summary.Total))\n\tsb.WriteString(fmt.Sprintf(\"| Passed | %d |\\n\", report.Summary.Passed))\n\tsb.WriteString(fmt.Sprintf(\"| Failed | %d |\\n\", report.Summary.Failed))\n\tif report.Summary.Skipped > 0 {\n\t\tsb.WriteString(fmt.Sprintf(\"| Skipped | %d |\\n\", report.Summary.Skipped))\n\t}\n\tif report.Summary.Errors > 0 {\n\t\tsb.WriteString(fmt.Sprintf(\"| Errors | %d |\\n\", report.Summary.Errors))\n\t}\n\tif report.Summary.Timeouts > 0 {\n\t\tsb.WriteString(fmt.Sprintf(\"| Timeouts | %d |\\n\", report.Summary.Timeouts))\n\t}\n\n\tpassRate := float64(0)\n\tif report.Summary.Total > 0 {\n\t\tpassRate = float64(report.Summary.Passed) / float64(report.Summary.Total) * 100\n\t}\n\tsb.WriteString(fmt.Sprintf(\"| Pass Rate | %.1f%% |\\n\", passRate))\n\tsb.WriteString(fmt.Sprintf(\"| Duration | %dms |\\n\", report.Summary.DurationMs))\n\tsb.WriteString(\"\\n\")\n\n\t// Environment\n\tif report.Environment != nil {\n\t\tsb.WriteString(\"## Environment\\n\\n\")\n\t\tsb.WriteString(\"| Setting | Value |\\n\")\n\t\tsb.WriteString(\"| ------- | ----- |\\n\")\n\t\tsb.WriteString(fmt.Sprintf(\"| User | %s |\\n\", report.Environment.UserID))\n\t\tsb.WriteString(fmt.Sprintf(\"| Team | %s |\\n\", report.Environment.TeamID))\n\t\tsb.WriteString(fmt.Sprintf(\"| Locale | %s |\\n\", report.Environment.Locale))\n\t\tsb.WriteString(\"\\n\")\n\t}\n\n\t// Results\n\tsb.WriteString(\"## Results\\n\\n\")\n\n\tif report.Results != nil {\n\t\tfor _, result := range report.Results {\n\t\t\tstatusIcon := \"✅\"\n\t\t\tswitch result.Status {\n\t\t\tcase StatusFailed:\n\t\t\t\tstatusIcon = \"❌\"\n\t\t\tcase StatusError:\n\t\t\t\tstatusIcon = \"💥\"\n\t\t\tcase StatusTimeout:\n\t\t\t\tstatusIcon = \"⏱️\"\n\t\t\tcase StatusSkipped:\n\t\t\t\tstatusIcon = \"⏭️\"\n\t\t\t}\n\n\t\t\tsb.WriteString(fmt.Sprintf(\"### %s %s - %s (%dms)\\n\\n\", statusIcon, result.ID, result.Status, result.DurationMs))\n\n\t\t\tif result.Error != \"\" {\n\t\t\t\tsb.WriteString(fmt.Sprintf(\"**Error:** %s\\n\\n\", result.Error))\n\t\t\t}\n\t\t}\n\t}\n\n\t// Stability results\n\tif report.StabilityResults != nil {\n\t\tsb.WriteString(\"## Stability Analysis\\n\\n\")\n\t\tsb.WriteString(\"| ID | Pass Rate | Runs | Status | Avg Duration |\\n\")\n\t\tsb.WriteString(\"| -- | --------- | ---- | ------ | ------------ |\\n\")\n\n\t\tfor _, sr := range report.StabilityResults {\n\t\t\tstatus := string(sr.StabilityClass)\n\t\t\tsb.WriteString(fmt.Sprintf(\"| %s | %.0f%% | %d/%d | %s | %.0fms |\\n\",\n\t\t\t\tsr.ID, sr.PassRate, sr.Passed, sr.Runs, status, sr.AvgDurationMs))\n\t\t}\n\t\tsb.WriteString(\"\\n\")\n\t}\n\n\t// Metadata\n\tsb.WriteString(\"## Metadata\\n\\n\")\n\tsb.WriteString(fmt.Sprintf(\"- **Started:** %s\\n\", report.Metadata.StartedAt.Format(time.RFC3339)))\n\tsb.WriteString(fmt.Sprintf(\"- **Completed:** %s\\n\", report.Metadata.CompletedAt.Format(time.RFC3339)))\n\tif report.Metadata.InputFile != \"\" {\n\t\tsb.WriteString(fmt.Sprintf(\"- **Input File:** %s\\n\", report.Metadata.InputFile))\n\t}\n\tif report.Metadata.OutputFile != \"\" {\n\t\tsb.WriteString(fmt.Sprintf(\"- **Output File:** %s\\n\", report.Metadata.OutputFile))\n\t}\n\n\t_, err := w.Write([]byte(sb.String()))\n\treturn err\n}\n\n// HTMLReporter generates HTML format reports\ntype HTMLReporter struct{}\n\n// NewHTMLReporter creates a new HTML reporter\nfunc NewHTMLReporter() *HTMLReporter {\n\treturn &HTMLReporter{}\n}\n\n// Generate generates an HTML report\nfunc (r *HTMLReporter) Generate(report *Report) error {\n\treturn nil\n}\n\n// Write writes the report in HTML format\nfunc (r *HTMLReporter) Write(report *Report, w io.Writer) error {\n\ttmpl, err := template.New(\"report\").Parse(htmlTemplate)\n\tif err != nil {\n\t\treturn fmt.Errorf(\"failed to parse HTML template: %w\", err)\n\t}\n\n\t// Calculate pass rate\n\tpassRate := float64(0)\n\tif report.Summary.Total > 0 {\n\t\tpassRate = float64(report.Summary.Passed) / float64(report.Summary.Total) * 100\n\t}\n\n\tdata := map[string]interface{}{\n\t\t\"Report\": report,\n\t\t\"PassRate\": passRate,\n\t}\n\n\treturn tmpl.Execute(w, data)\n}\n\n// HTML template for reports\nconst htmlTemplate = `\n\n\n \n \n Agent Test Report - {{.Report.Summary.AgentID}}\n \n\n\n
\n

Agent Test Report

\n

{{.Report.Summary.AgentID}} {{if .Report.Summary.Connector}}• {{.Report.Summary.Connector}}{{end}}

\n \n
\n
\n
{{.Report.Summary.Total}}
\n
Total Tests
\n
\n
\n
{{.Report.Summary.Passed}}
\n
Passed
\n
\n
\n
{{.Report.Summary.Failed}}
\n
Failed
\n
\n
\n
{{printf \"%.1f\" .PassRate}}%
\n
Pass Rate
\n
\n
\n
{{.Report.Summary.DurationMs}}ms
\n
Duration
\n
\n
\n \n

Test Results

\n \n \n \n \n \n \n \n \n \n \n {{range .Report.Results}}\n \n \n \n \n \n \n {{end}}\n {{range .Report.StabilityResults}}\n \n \n \n \n \n \n {{end}}\n \n
IDStatusDurationDetails
{{.ID}}{{.Status}}{{.DurationMs}}ms\n {{if .Error}}
{{.Error}}
{{end}}\n
{{.ID}}{{.StabilityClass}}{{printf \"%.0f\" .AvgDurationMs}}ms avg{{.Passed}}/{{.Runs}} passed ({{printf \"%.0f\" .PassRate}}%)
\n \n

Metadata

\n
\n
Started:
{{.Report.Metadata.StartedAt}}
\n
Completed:
{{.Report.Metadata.CompletedAt}}
\n {{if .Report.Metadata.InputFile}}
Input:
{{.Report.Metadata.InputFile}}
{{end}}\n {{if .Report.Metadata.OutputFile}}
Output:
{{.Report.Metadata.OutputFile}}
{{end}}\n
\n
\n\n`\n\n// AgentReporter uses a custom agent to generate reports\ntype AgentReporter struct {\n\tagentID string\n\tformat string\n\tverbose bool\n\tctx *context.Context // Test context for agent call\n}\n\n// NewAgentReporter creates a new agent-based reporter\nfunc NewAgentReporter(agentID, format string, verbose bool) *AgentReporter {\n\treturn &AgentReporter{\n\t\tagentID: agentID,\n\t\tformat: format,\n\t\tverbose: verbose,\n\t}\n}\n\n// SetContext sets the context for agent calls\nfunc (r *AgentReporter) SetContext(ctx *context.Context) {\n\tr.ctx = ctx\n}\n\n// Generate generates a report using the agent\nfunc (r *AgentReporter) Generate(report *Report) error {\n\treturn nil\n}\n\n// Write writes the report using the agent\nfunc (r *AgentReporter) Write(report *Report, w io.Writer) error {\n\t// Check if AgentGetterFunc is initialized\n\tif caller.AgentGetterFunc == nil {\n\t\treturn fmt.Errorf(\"AgentGetterFunc not initialized, cannot call reporter agent\")\n\t}\n\n\t// Get the reporter agent\n\tagent, err := caller.AgentGetterFunc(r.agentID)\n\tif err != nil {\n\t\treturn fmt.Errorf(\"failed to get reporter agent %s: %w\", r.agentID, err)\n\t}\n\n\t// Build input for the reporter agent\n\tinput := &ReporterInput{\n\t\tReport: report,\n\t\tFormat: r.format,\n\t\tOptions: &ReporterOptions{\n\t\t\tVerbose: r.verbose,\n\t\t\tIncludeOutputs: r.verbose,\n\t\t\tIncludeInputs: r.verbose,\n\t\t},\n\t}\n\n\t// Convert input to JSON for the agent\n\tinputJSON, err := jsoniter.Marshal(input)\n\tif err != nil {\n\t\treturn fmt.Errorf(\"failed to marshal reporter input: %w\", err)\n\t}\n\n\t// Create message for the agent\n\tmessages := []context.Message{\n\t\t{\n\t\t\tRole: context.RoleUser,\n\t\t\tContent: string(inputJSON),\n\t\t},\n\t}\n\n\t// Create context if not provided\n\tctx := r.ctx\n\tif ctx == nil {\n\t\t// Create a minimal context for the reporter agent call\n\t\tctx = NewTestContext(\"reporter\", r.agentID, NewEnvironment(\"\", \"\"))\n\t\tdefer ctx.Release()\n\t}\n\n\t// Call the agent with skip options (no history, no output)\n\toptions := &context.Options{\n\t\tSkip: &context.Skip{\n\t\t\tHistory: true,\n\t\t\tOutput: true,\n\t\t},\n\t}\n\n\tresponse, err := agent.Stream(ctx, messages, options)\n\tif err != nil {\n\t\treturn fmt.Errorf(\"reporter agent call failed: %w\", err)\n\t}\n\n\t// Extract content from response\n\tcontent, err := r.extractContent(response)\n\tif err != nil {\n\t\treturn fmt.Errorf(\"failed to extract report content: %w\", err)\n\t}\n\n\t// Write the content to output\n\t_, err = w.Write([]byte(content))\n\tif err != nil {\n\t\treturn fmt.Errorf(\"failed to write report: %w\", err)\n\t}\n\n\treturn nil\n}\n\n// extractContent extracts the report content from the agent's *context.Response\n// Now that agent.Stream() returns *context.Response directly,\n// we can access fields without type assertions.\nfunc (r *AgentReporter) extractContent(response *context.Response) (string, error) {\n\tif response == nil {\n\t\treturn \"\", fmt.Errorf(\"agent returned nil response\")\n\t}\n\n\t// Priority 1: Check Next field (custom hook data)\n\tif response.Next != nil {\n\t\treturn r.contentToString(response.Next)\n\t}\n\n\t// Priority 2: Extract from completion content\n\tif response.Completion != nil && response.Completion.Content != nil {\n\t\treturn r.contentToString(response.Completion.Content)\n\t}\n\n\treturn \"\", fmt.Errorf(\"no content in response\")\n}\n\n// contentToString converts various content types to string\nfunc (r *AgentReporter) contentToString(content interface{}) (string, error) {\n\tswitch v := content.(type) {\n\tcase string:\n\t\treturn v, nil\n\tcase []byte:\n\t\treturn string(v), nil\n\tdefault:\n\t\tjsonBytes, err := jsoniter.Marshal(content)\n\t\tif err != nil {\n\t\t\treturn fmt.Sprintf(\"%v\", content), nil\n\t\t}\n\t\treturn string(jsonBytes), nil\n\t}\n}\n\n// GetReporter returns a reporter based on output format\nfunc GetReporter(format OutputFormat) Reporter {\n\tswitch format {\n\tcase FormatJSON:\n\t\treturn NewJSONReporter()\n\tcase FormatHTML:\n\t\treturn NewHTMLReporter()\n\tcase FormatMarkdown:\n\t\treturn NewMarkdownReporter()\n\tdefault:\n\t\treturn NewJSONLReporter()\n\t}\n}\n\n// GetReporterFromPath returns a reporter based on file extension\nfunc GetReporterFromPath(outputPath string) Reporter {\n\tformat := GetOutputFormat(outputPath)\n\treturn GetReporter(format)\n}\n\n// GetReporterWithAgent returns an agent-based reporter if agentID is specified,\n// otherwise returns a built-in reporter based on output format\nfunc GetReporterWithAgent(agentID, outputPath string, verbose bool) Reporter {\n\tif agentID != \"\" {\n\t\tformat := GetOutputFormat(outputPath)\n\t\treturn NewAgentReporter(agentID, string(format), verbose)\n\t}\n\treturn GetReporterFromPath(outputPath)\n}\n" }, { "path": "agent/test/resolver.go", "content": "package test\n\nimport (\n\t\"fmt\"\n\t\"os\"\n\t\"path/filepath\"\n\t\"strings\"\n\t\"time\"\n\n\tjsoniter \"github.com/json-iterator/go\"\n\t\"github.com/yaoapp/gou/application\"\n)\n\n// PathResolver resolves agent information from file paths\ntype PathResolver struct{}\n\n// NewResolver creates a new path resolver\nfunc NewResolver() Resolver {\n\treturn &PathResolver{}\n}\n\n// Resolve resolves the agent from options\n// Priority: explicit AgentID > path-based detection (from input file or cwd)\nfunc (r *PathResolver) Resolve(opts *Options) (*AgentInfo, error) {\n\t// If explicit agent ID is provided, use it\n\tif opts.AgentID != \"\" {\n\t\treturn r.ResolveByID(opts.AgentID)\n\t}\n\n\t// For file mode, resolve from input file path\n\tif opts.InputMode == InputModeFile {\n\t\tif opts.Input == \"\" {\n\t\t\treturn nil, fmt.Errorf(\"no agent ID or input file specified\")\n\t\t}\n\t\treturn r.ResolveFromPath(opts.Input)\n\t}\n\n\t// For message mode, try to resolve from current working directory\n\treturn r.ResolveFromCwd()\n}\n\n// ResolveFromCwd resolves the agent from the current working directory\n// It looks for package.yao in the current directory or parent directories\nfunc (r *PathResolver) ResolveFromCwd() (*AgentInfo, error) {\n\tcwd, err := os.Getwd()\n\tif err != nil {\n\t\treturn nil, fmt.Errorf(\"failed to get current working directory: %w\", err)\n\t}\n\n\tinfo, err := r.ResolveFromPath(cwd)\n\tif err != nil {\n\t\treturn nil, fmt.Errorf(\"no agent found in current directory. Use -n to specify agent explicitly\")\n\t}\n\treturn info, nil\n}\n\n// ResolveFromPath resolves the agent by traversing up from the input file path\n// It looks for package.yao in parent directories\n// If YAO_ROOT is set, it also considers paths relative to YAO_ROOT\nfunc (r *PathResolver) ResolveFromPath(inputPath string) (*AgentInfo, error) {\n\t// Get absolute path\n\tabsPath, err := filepath.Abs(inputPath)\n\tif err != nil {\n\t\treturn nil, fmt.Errorf(\"failed to get absolute path: %w\", err)\n\t}\n\n\t// Start from the directory containing the input file\n\tdir := filepath.Dir(absPath)\n\n\t// Traverse up to find package.yao\n\tfor {\n\t\tpackagePath := filepath.Join(dir, \"package.yao\")\n\t\tif _, err := os.Stat(packagePath); err == nil {\n\t\t\t// Found package.yao\n\t\t\treturn r.loadAgentFromPath(dir, packagePath)\n\t\t}\n\n\t\t// Move to parent directory\n\t\tparent := filepath.Dir(dir)\n\t\tif parent == dir {\n\t\t\t// Reached root, no package.yao found\n\t\t\tbreak\n\t\t}\n\t\tdir = parent\n\t}\n\n\t// If YAO_ROOT is set, try resolving relative to it\n\tyaoRoot := os.Getenv(\"YAO_ROOT\")\n\tif yaoRoot != \"\" {\n\t\t// Try the input path relative to YAO_ROOT\n\t\trelPath := inputPath\n\t\t// If inputPath is absolute, try to make it relative\n\t\tif filepath.IsAbs(inputPath) {\n\t\t\t// Check if inputPath is under YAO_ROOT\n\t\t\tif rel, err := filepath.Rel(yaoRoot, inputPath); err == nil && !strings.HasPrefix(rel, \"..\") {\n\t\t\t\trelPath = rel\n\t\t\t}\n\t\t}\n\n\t\t// Traverse up from YAO_ROOT + relPath\n\t\tdir = filepath.Join(yaoRoot, filepath.Dir(relPath))\n\t\tfor {\n\t\t\tpackagePath := filepath.Join(dir, \"package.yao\")\n\t\t\tif _, err := os.Stat(packagePath); err == nil {\n\t\t\t\t// Found package.yao\n\t\t\t\treturn r.loadAgentFromPath(dir, packagePath)\n\t\t\t}\n\n\t\t\t// Move to parent directory, but don't go above YAO_ROOT\n\t\t\tparent := filepath.Dir(dir)\n\t\t\tif parent == dir || !strings.HasPrefix(parent, yaoRoot) {\n\t\t\t\tbreak\n\t\t\t}\n\t\t\tdir = parent\n\t\t}\n\t}\n\n\treturn nil, fmt.Errorf(\"no package.yao found in path hierarchy of %s\", inputPath)\n}\n\n// ResolveByID resolves an agent by its ID\n// This would integrate with the assistant loading system\nfunc (r *PathResolver) ResolveByID(agentID string) (*AgentInfo, error) {\n\t// This is a placeholder - actual implementation would use assistant.Get()\n\t// For now, return basic info\n\treturn &AgentInfo{\n\t\tID: agentID,\n\t\tName: agentID,\n\t}, nil\n}\n\n// loadAgentFromPath loads agent information from a package.yao file\nfunc (r *PathResolver) loadAgentFromPath(agentDir, packagePath string) (*AgentInfo, error) {\n\t// Read package.yao\n\tdata, err := os.ReadFile(packagePath)\n\tif err != nil {\n\t\treturn nil, fmt.Errorf(\"failed to read package.yao: %w\", err)\n\t}\n\n\t// Parse package.yao\n\tvar pkg PackageYao\n\tif err := jsoniter.Unmarshal(data, &pkg); err != nil {\n\t\treturn nil, fmt.Errorf(\"failed to parse package.yao: %w\", err)\n\t}\n\n\t// Derive agent ID from directory path\n\tagentID := deriveAgentID(agentDir)\n\n\treturn &AgentInfo{\n\t\tID: agentID,\n\t\tName: pkg.Name,\n\t\tDescription: pkg.Description,\n\t\tPath: agentDir,\n\t\tConnector: pkg.Connector,\n\t\tType: pkg.Type,\n\t}, nil\n}\n\n// PackageYao represents the structure of package.yao\ntype PackageYao struct {\n\tName string `json:\"name\"`\n\tDescription string `json:\"description,omitempty\"`\n\tConnector string `json:\"connector,omitempty\"`\n\tType string `json:\"type,omitempty\"`\n\tUses map[string]interface{} `json:\"uses,omitempty\"`\n\tOptions map[string]interface{} `json:\"options,omitempty\"`\n}\n\n// deriveAgentID derives an agent ID from the directory path\n// e.g., /app/assistants/workers/system/keyword -> workers.system.keyword\nfunc deriveAgentID(dir string) string {\n\t// Find \"assistants\" in path and use everything after it\n\tparts := strings.Split(filepath.ToSlash(dir), \"/\")\n\n\t// Look for \"assistants\" marker\n\tstartIdx := -1\n\tfor i, part := range parts {\n\t\tif part == \"assistants\" {\n\t\t\tstartIdx = i + 1\n\t\t\tbreak\n\t\t}\n\t}\n\n\tif startIdx == -1 || startIdx >= len(parts) {\n\t\t// No \"assistants\" found, use the last directory name\n\t\treturn filepath.Base(dir)\n\t}\n\n\t// Join remaining parts with dots\n\treturn strings.Join(parts[startIdx:], \".\")\n}\n\n// GetOutputFormat determines the output format from file extension\nfunc GetOutputFormat(outputPath string) OutputFormat {\n\text := strings.ToLower(filepath.Ext(outputPath))\n\tswitch ext {\n\tcase \".json\":\n\t\treturn FormatJSON\n\tcase \".html\", \".htm\":\n\t\treturn FormatHTML\n\tcase \".md\", \".markdown\":\n\t\treturn FormatMarkdown\n\tdefault:\n\t\treturn FormatJSON // Default to JSON\n\t}\n}\n\n// ValidateOptions validates test options\nfunc ValidateOptions(opts *Options) error {\n\tif opts.Input == \"\" {\n\t\treturn fmt.Errorf(\"input is required (-i flag)\")\n\t}\n\n\t// For file mode, check input file exists\n\tif opts.InputMode == InputModeFile {\n\t\tresolvedPath := ResolvePathWithYaoRoot(opts.Input)\n\t\tif _, err := os.Stat(resolvedPath); os.IsNotExist(err) {\n\t\t\treturn fmt.Errorf(\"input file not found: %s\", opts.Input)\n\t\t}\n\t}\n\n\t// Note: For message mode, agent can be resolved from cwd, so no validation here\n\t// The resolver will return an error if agent cannot be found\n\n\t// Validate timeout\n\tif opts.Timeout < 0 {\n\t\treturn fmt.Errorf(\"timeout cannot be negative\")\n\t}\n\n\t// Validate parallel\n\tif opts.Parallel < 0 {\n\t\treturn fmt.Errorf(\"parallel cannot be negative\")\n\t}\n\n\treturn nil\n}\n\n// DefaultOptions returns options with default values\nfunc DefaultOptions() *Options {\n\treturn &Options{\n\t\tTimeout: 120 * time.Second, // 2 minutes default timeout\n\t\tParallel: 1,\n\t\tRuns: 1,\n\t\tVerbose: false,\n\t\tFailFast: false,\n\t}\n}\n\n// DetectInputMode detects the input mode from the input string\n// Returns:\n// - InputModeScript: if input starts with \"scripts.\"\n// - InputModeFile: if input ends with \".jsonl\" or is an existing file\n// - InputModeMessage: otherwise (direct message mode)\nfunc DetectInputMode(input string) InputMode {\n\t// Check for script test prefix\n\tif strings.HasPrefix(input, \"scripts.\") {\n\t\treturn InputModeScript\n\t}\n\n\t// If input ends with .jsonl or .json, treat as file\n\tif strings.HasSuffix(input, \".jsonl\") || strings.HasSuffix(input, \".json\") {\n\t\treturn InputModeFile\n\t}\n\n\t// If input contains path separator, check if file exists\n\tif strings.Contains(input, string(filepath.Separator)) || strings.Contains(input, \"/\") {\n\t\tif _, err := os.Stat(input); err == nil {\n\t\t\treturn InputModeFile\n\t\t}\n\t}\n\n\t// Otherwise treat as direct message\n\treturn InputModeMessage\n}\n\n// MergeOptions merges user options with defaults\nfunc MergeOptions(opts *Options, defaults *Options) *Options {\n\tresult := *defaults\n\n\tif opts.Input != \"\" {\n\t\tresult.Input = opts.Input\n\t\tresult.InputMode = DetectInputMode(opts.Input)\n\t}\n\tif opts.OutputFile != \"\" {\n\t\tresult.OutputFile = opts.OutputFile\n\t}\n\tif opts.AgentID != \"\" {\n\t\tresult.AgentID = opts.AgentID\n\t}\n\tif opts.Connector != \"\" {\n\t\tresult.Connector = opts.Connector\n\t}\n\tif opts.UserID != \"\" {\n\t\tresult.UserID = opts.UserID\n\t}\n\tif opts.TeamID != \"\" {\n\t\tresult.TeamID = opts.TeamID\n\t}\n\tif opts.Locale != \"\" {\n\t\tresult.Locale = opts.Locale\n\t}\n\tif opts.Timeout > 0 {\n\t\tresult.Timeout = opts.Timeout\n\t}\n\tif opts.Parallel > 0 {\n\t\tresult.Parallel = opts.Parallel\n\t}\n\tif opts.Runs > 0 {\n\t\tresult.Runs = opts.Runs\n\t}\n\tif opts.ReporterID != \"\" {\n\t\tresult.ReporterID = opts.ReporterID\n\t}\n\tif opts.ContextFile != \"\" {\n\t\tresult.ContextFile = opts.ContextFile\n\t}\n\tif opts.Run != \"\" {\n\t\tresult.Run = opts.Run\n\t}\n\tif opts.Verbose {\n\t\tresult.Verbose = opts.Verbose\n\t}\n\tif opts.FailFast {\n\t\tresult.FailFast = opts.FailFast\n\t}\n\n\treturn &result\n}\n\n// GenerateDefaultOutputPath generates the default output path based on input file\n// Format: {input_directory}/output-{timestamp}.jsonl\n// Timestamp format: YYYYMMDDHHMMSS\nfunc GenerateDefaultOutputPath(inputPath string) string {\n\t// Resolve input path considering YAO_ROOT\n\tresolvedPath := ResolvePathWithYaoRoot(inputPath)\n\tdir := filepath.Dir(resolvedPath)\n\ttimestamp := time.Now().Format(\"20060102150405\")\n\tfilename := fmt.Sprintf(\"output-%s.jsonl\", timestamp)\n\treturn filepath.Join(dir, filename)\n}\n\n// ResolveOutputPath resolves the output path based on input mode\n// - File mode: generate default path in same directory as input\n// - Message mode: return empty string (output to stdout)\n// If outputPath is explicitly specified, always use it\nfunc ResolveOutputPath(opts *Options) string {\n\tif opts.OutputFile != \"\" {\n\t\treturn opts.OutputFile\n\t}\n\n\t// For file mode, generate default output path\n\tif opts.InputMode == InputModeFile {\n\t\treturn GenerateDefaultOutputPath(opts.Input)\n\t}\n\n\t// For message mode, output to stdout (empty string)\n\treturn \"\"\n}\n\n// CreateTestCaseFromMessage creates a single test case from a direct message\nfunc CreateTestCaseFromMessage(message string) *Case {\n\treturn &Case{\n\t\tID: \"T001\",\n\t\tInput: message,\n\t}\n}\n\n// ResolvePathWithYaoRoot resolves a file path relative to current directory\n// No fallback to YAO_ROOT - paths are always resolved from current working directory\nfunc ResolvePathWithYaoRoot(path string) string {\n\tif filepath.IsAbs(path) {\n\t\treturn path\n\t}\n\n\t// Try resolving relative to the application root first\n\tif application.App != nil {\n\t\tappRoot := application.App.Root()\n\t\tif appRoot != \"\" {\n\t\t\tcandidate := filepath.Join(appRoot, path)\n\t\t\tif _, err := os.Stat(candidate); err == nil {\n\t\t\t\treturn candidate\n\t\t\t}\n\t\t}\n\t}\n\n\t// Fallback: resolve relative to cwd\n\tabsPath, err := filepath.Abs(path)\n\tif err != nil {\n\t\treturn path\n\t}\n\treturn absPath\n}\n" }, { "path": "agent/test/runner.go", "content": "package test\n\nimport (\n\tstdContext \"context\"\n\t\"fmt\"\n\t\"os\"\n\t\"path/filepath\"\n\t\"reflect\"\n\t\"regexp\"\n\t\"strings\"\n\t\"sync\"\n\t\"time\"\n\n\tjsoniter \"github.com/json-iterator/go\"\n\t\"github.com/yaoapp/yao/agent/assistant\"\n\t\"github.com/yaoapp/yao/agent/context\"\n)\n\n// Executor executes test cases against an agent\ntype Executor struct {\n\topts *Options\n\toutput *OutputWriter\n\tresolver Resolver\n\tloader Loader\n\thookExecutor *HookExecutor\n\tagentPath string // Path to the agent being tested\n}\n\n// NewRunner creates a new test runner\nfunc NewRunner(opts *Options) *Executor {\n\treturn &Executor{\n\t\topts: opts,\n\t\toutput: NewOutputWriter(opts.Verbose),\n\t\tresolver: NewResolver(),\n\t\tloader: NewLoader(),\n\t\thookExecutor: NewHookExecutor(opts.Verbose),\n\t}\n}\n\n// Run executes all test cases and returns a report\nfunc (r *Executor) Run() (*Report, error) {\n\t// For script test mode, use script runner\n\tif r.opts.InputMode == InputModeScript {\n\t\treturn r.RunScriptTests()\n\t}\n\n\t// For direct message mode, use simplified output (development mode)\n\tif r.opts.InputMode == InputModeMessage {\n\t\treturn r.RunDirect()\n\t}\n\n\treturn r.RunTests()\n}\n\n// RunScriptTests executes script tests and returns a report\nfunc (r *Executor) RunScriptTests() (*Report, error) {\n\tscriptRunner := NewScriptRunner(r.opts)\n\tscriptReport, err := scriptRunner.Run()\n\tif err != nil {\n\t\treturn nil, err\n\t}\n\n\t// Convert to standard report for unified output handling\n\treport := scriptReport.ToReport()\n\n\t// Write output if specified\n\tif r.opts.OutputFile != \"\" {\n\t\terr = r.writeOutput(report)\n\t\tif err != nil {\n\t\t\tr.output.Error(\"Failed to write output: %s\", err.Error())\n\t\t} else {\n\t\t\tr.output.OutputFile(r.opts.OutputFile)\n\t\t}\n\t}\n\n\t// Print final result\n\tr.output.FinalResult(!report.HasFailures())\n\n\treturn report, nil\n}\n\n// RunDirect executes a single direct message and outputs the result directly\n// This is optimized for development/debugging scenarios\nfunc (r *Executor) RunDirect() (*Report, error) {\n\t// Resolve agent\n\tagentInfo, err := r.resolver.Resolve(r.opts)\n\tif err != nil {\n\t\treturn nil, fmt.Errorf(\"failed to resolve agent: %w\", err)\n\t}\n\n\t// Get assistant\n\tast, err := assistant.Get(agentInfo.ID)\n\tif err != nil {\n\t\treturn nil, fmt.Errorf(\"failed to get assistant: %w\", err)\n\t}\n\n\t// Create test case from message\n\ttc := CreateTestCaseFromMessage(r.opts.Input)\n\n\t// Create context\n\tchatID := GenerateChatID(tc.ID, 1)\n\tctx := NewTestContextFromOptions(chatID, agentInfo.ID, r.opts, tc)\n\tdefer ctx.Release()\n\n\t// Build context options\n\topts := buildContextOptions(tc, r.opts)\n\n\t// Create timeout context\n\ttimeout := tc.GetTimeout(r.opts.Timeout)\n\ttimeoutCtx, cancel := stdContext.WithTimeout(ctx.Context, timeout)\n\tdefer cancel()\n\tctx.Context = timeoutCtx\n\n\t// Parse input to messages with file loading support\n\tinputOpts := r.getInputOptions()\n\tmessages, err := tc.GetMessagesWithOptions(inputOpts)\n\tif err != nil {\n\t\treturn nil, fmt.Errorf(\"failed to parse input: %w\", err)\n\t}\n\n\t// Run the agent\n\tresponse, err := ast.Stream(ctx, messages, opts)\n\n\t// Check for timeout\n\tif timeoutCtx.Err() != nil {\n\t\treturn nil, fmt.Errorf(\"timeout after %s\", timeout)\n\t}\n\n\t// Check for error\n\tif err != nil {\n\t\treturn nil, err\n\t}\n\n\t// Extract and print output directly\n\toutput := extractOutput(response)\n\tr.output.DirectOutput(output)\n\n\t// Determine connector: user-specified > agent default\n\tconnector := r.opts.Connector\n\tif connector == \"\" {\n\t\tconnector = agentInfo.Connector\n\t}\n\n\t// Return minimal report (for exit code handling)\n\treturn &Report{\n\t\tSummary: &Summary{\n\t\t\tTotal: 1,\n\t\t\tPassed: 1,\n\t\t\tAgentID: agentInfo.ID,\n\t\t\tConnector: connector,\n\t\t},\n\t}, nil\n}\n\n// RunTests executes test cases from file and generates a report\nfunc (r *Executor) RunTests() (*Report, error) {\n\tstartTime := time.Now()\n\n\t// Print header\n\tr.output.Header(\"Agent Test\")\n\n\t// Resolve agent\n\tagentInfo, err := r.resolver.Resolve(r.opts)\n\tif err != nil {\n\t\treturn nil, fmt.Errorf(\"failed to resolve agent: %w\", err)\n\t}\n\n\tr.output.Info(\"Agent: %s\", agentInfo.ID)\n\tr.agentPath = agentInfo.Path // Store agent path for hook execution\n\tif r.opts.Connector != \"\" {\n\t\tr.output.Info(\"Connector: %s (override)\", r.opts.Connector)\n\t} else if agentInfo.Connector != \"\" {\n\t\tr.output.Info(\"Connector: %s\", agentInfo.Connector)\n\t}\n\n\t// Load test cases based on input source\n\tvar testCases []*Case\n\tinputSource := ParseInputSource(r.opts.Input)\n\n\tswitch inputSource.Type {\n\tcase InputSourceAgent:\n\t\t// Generate test cases using agent\n\t\tr.output.Info(\"Generating test cases from agent: %s\", inputSource.Value)\n\t\ttargetInfo := &TargetAgentInfo{\n\t\t\tID: agentInfo.ID,\n\t\t\tDescription: agentInfo.Description,\n\t\t}\n\t\ttestCases, err = r.loader.LoadFromAgent(inputSource.Value, targetInfo, inputSource.Params)\n\t\tif err != nil {\n\t\t\treturn nil, fmt.Errorf(\"failed to generate test cases: %w\", err)\n\t\t}\n\t\tr.output.Info(\"Generated: %d test cases\", len(testCases))\n\n\tcase InputSourceScript:\n\t\t// Generate test cases using script (if it's a generator script, not test script)\n\t\t// Note: scripts. prefix without \"scripts:\" is handled by RunScriptTests\n\t\tif strings.HasPrefix(r.opts.Input, \"scripts:\") {\n\t\t\tscriptRef := strings.TrimPrefix(r.opts.Input, \"scripts:\")\n\t\t\tr.output.Info(\"Generating test cases from script: %s\", scriptRef)\n\t\t\ttargetInfo := &TargetAgentInfo{\n\t\t\t\tID: agentInfo.ID,\n\t\t\t\tDescription: agentInfo.Description,\n\t\t\t}\n\t\t\ttestCases, err = r.loader.LoadFromScript(scriptRef, targetInfo)\n\t\t\tif err != nil {\n\t\t\t\treturn nil, fmt.Errorf(\"failed to generate test cases from script: %w\", err)\n\t\t\t}\n\t\t\tr.output.Info(\"Generated: %d test cases\", len(testCases))\n\t\t} else {\n\t\t\t// This is a test script (scripts.xxx format), handled by RunScriptTests\n\t\t\treturn nil, fmt.Errorf(\"script test mode should be handled by RunScriptTests\")\n\t\t}\n\n\tdefault:\n\t\t// File mode - load from JSONL\n\t\ttestCases, err = r.loader.LoadFile(r.opts.Input)\n\t\tif err != nil {\n\t\t\treturn nil, fmt.Errorf(\"failed to load test cases: %w\", err)\n\t\t}\n\t\tr.output.Info(\"Input: %s (%d test cases)\", r.opts.Input, len(testCases))\n\t}\n\n\t// Handle dry-run mode - just output the generated test cases\n\tif r.opts.DryRun {\n\t\tr.output.Info(\"Dry-run mode: outputting generated test cases\")\n\t\treturn r.outputDryRun(testCases, agentInfo)\n\t}\n\n\t// Filter skipped tests\n\tactiveTests := FilterSkipped(testCases)\n\tskippedCount := len(testCases) - len(activeTests)\n\tif skippedCount > 0 {\n\t\tr.output.Warning(\"Skipped: %d test cases\", skippedCount)\n\t}\n\n\t// Filter by --run pattern if specified\n\tif r.opts.Run != \"\" {\n\t\trunPattern, err := regexp.Compile(r.opts.Run)\n\t\tif err != nil {\n\t\t\treturn nil, fmt.Errorf(\"invalid --run pattern %q: %w\", r.opts.Run, err)\n\t\t}\n\t\tactiveTests = FilterByPattern(activeTests, runPattern)\n\t\tif len(activeTests) == 0 {\n\t\t\treturn nil, fmt.Errorf(\"no test cases match pattern %q\", r.opts.Run)\n\t\t}\n\t\tr.output.Info(\"Filter: %q (%d test cases match)\", r.opts.Run, len(activeTests))\n\t}\n\n\t// Load context config if specified\n\tif r.opts.ContextFile != \"\" {\n\t\tctxConfig, err := LoadContextConfig(r.opts.ContextFile)\n\t\tif err != nil {\n\t\t\treturn nil, fmt.Errorf(\"failed to load context file: %w\", err)\n\t\t}\n\t\tr.opts.ContextData = ctxConfig\n\t\tr.output.Info(\"Context: %s\", r.opts.ContextFile)\n\t}\n\n\t// Set options on hook executor (for context data access in hooks)\n\tr.hookExecutor.SetOptions(r.opts)\n\n\t// Print test info\n\tif r.opts.Runs > 1 {\n\t\tr.output.Info(\"Runs: %d per test case (stability analysis)\", r.opts.Runs)\n\t}\n\tr.output.Info(\"Timeout: %s\", r.opts.Timeout)\n\tif r.opts.Parallel > 1 {\n\t\tr.output.Info(\"Parallel: %d\", r.opts.Parallel)\n\t}\n\n\t// Get assistant\n\tast, err := assistant.Get(agentInfo.ID)\n\tif err != nil {\n\t\treturn nil, fmt.Errorf(\"failed to get assistant: %w\", err)\n\t}\n\n\t// Determine connector: user-specified > agent default\n\tconnector := r.opts.Connector\n\tif connector == \"\" {\n\t\tconnector = agentInfo.Connector\n\t}\n\n\t// Create report\n\treport := &Report{\n\t\tSummary: &Summary{\n\t\t\tTotal: len(testCases),\n\t\t\tAgentID: agentInfo.ID,\n\t\t\tAgentPath: agentInfo.Path,\n\t\t\tConnector: connector,\n\t\t\tRunsPerCase: r.opts.Runs,\n\t\t},\n\t\tEnvironment: NewEnvironment(r.opts.UserID, r.opts.TeamID),\n\t\tMetadata: &ReportMetadata{\n\t\t\tStartedAt: startTime,\n\t\t\tInputFile: r.opts.Input,\n\t\t\tOptions: r.opts,\n\t\t},\n\t}\n\n\t// Execute global BeforeAll if specified\n\tvar globalBeforeData interface{}\n\tif r.opts.BeforeAll != \"\" {\n\t\tr.output.Info(\"BeforeAll: %s\", r.opts.BeforeAll)\n\t\tvar err error\n\t\tglobalBeforeData, err = r.hookExecutor.ExecuteBeforeAll(r.opts.BeforeAll, activeTests, agentInfo.Path)\n\t\tif err != nil {\n\t\t\treturn nil, fmt.Errorf(\"beforeAll script failed: %w\", err)\n\t\t}\n\t}\n\n\t// Ensure AfterAll runs even if tests fail\n\tdefer func() {\n\t\tif r.opts.AfterAll != \"\" {\n\t\t\tr.output.Info(\"AfterAll: %s\", r.opts.AfterAll)\n\t\t\tif err := r.hookExecutor.ExecuteAfterAll(r.opts.AfterAll, report.Results, globalBeforeData, agentInfo.Path); err != nil {\n\t\t\t\tr.output.Warning(\"afterAll script failed: %s\", err.Error())\n\t\t\t}\n\t\t}\n\t}()\n\n\t// Run tests\n\tr.output.SubHeader(\"Running Tests\")\n\n\tif r.opts.Runs > 1 {\n\t\t// Stability testing mode\n\t\treport.StabilityResults = r.runStabilityTests(ast, activeTests, agentInfo.ID)\n\t\tr.calculateStabilitySummary(report)\n\t} else {\n\t\t// Single run mode\n\t\treport.Results = r.runSingleTests(ast, activeTests, agentInfo.ID)\n\t\tr.calculateSingleSummary(report)\n\t}\n\n\t// Add skipped count\n\treport.Summary.Skipped = skippedCount\n\n\t// Complete report\n\treport.Summary.DurationMs = time.Since(startTime).Milliseconds()\n\treport.Metadata.CompletedAt = time.Now()\n\n\t// Print summary\n\tr.output.Summary(report.Summary, time.Since(startTime))\n\n\t// Write output\n\tif r.opts.OutputFile != \"\" {\n\t\terr = r.writeOutput(report)\n\t\tif err != nil {\n\t\t\tr.output.Error(\"Failed to write output: %s\", err.Error())\n\t\t} else {\n\t\t\tr.output.OutputFile(r.opts.OutputFile)\n\t\t}\n\t}\n\n\t// Print final result\n\tr.output.FinalResult(!report.HasFailures())\n\n\treturn report, nil\n}\n\n// runSingleTests runs each test case once\nfunc (r *Executor) runSingleTests(ast *assistant.Assistant, testCases []*Case, agentID string) []*Result {\n\tresults := make([]*Result, 0, len(testCases))\n\n\tif r.opts.Parallel > 1 {\n\t\t// Parallel execution\n\t\tresults = r.runParallel(ast, testCases, agentID)\n\t} else {\n\t\t// Sequential execution\n\t\tfor i, tc := range testCases {\n\t\t\tresult := r.runSingleTest(ast, tc, agentID, 1)\n\t\t\tresults = append(results, result)\n\n\t\t\t// Check fail-fast\n\t\t\tif r.opts.FailFast && result.Status != StatusPassed && result.Status != StatusSkipped {\n\t\t\t\tr.output.Warning(\"Stopping due to --fail-fast (failed at test %d/%d)\", i+1, len(testCases))\n\t\t\t\tbreak\n\t\t\t}\n\t\t}\n\t}\n\n\treturn results\n}\n\n// runParallel runs tests in parallel\nfunc (r *Executor) runParallel(ast *assistant.Assistant, testCases []*Case, agentID string) []*Result {\n\tresults := make([]*Result, len(testCases))\n\tvar wg sync.WaitGroup\n\tsem := make(chan struct{}, r.opts.Parallel)\n\n\tfor i, tc := range testCases {\n\t\twg.Add(1)\n\t\tgo func(idx int, testCase *Case) {\n\t\t\tdefer wg.Done()\n\t\t\tsem <- struct{}{} // Acquire\n\t\t\tdefer func() { <-sem }() // Release\n\n\t\t\tresults[idx] = r.runSingleTest(ast, testCase, agentID, 1)\n\t\t}(i, tc)\n\t}\n\n\twg.Wait()\n\treturn results\n}\n\n// runSingleTest runs a single test case\nfunc (r *Executor) runSingleTest(ast *assistant.Assistant, tc *Case, agentID string, runNum int) *Result {\n\t// Check if this is a dynamic mode test\n\tif tc.IsDynamicMode() {\n\t\treturn r.runDynamicTest(ast, tc, agentID)\n\t}\n\n\t// Get input summary for display\n\tinputSummary := SummarizeInput(tc.Input, 50)\n\tr.output.TestStart(tc.ID, inputSummary, runNum)\n\n\tstartTime := time.Now()\n\n\t// Create result\n\tresult := &Result{\n\t\tID: tc.ID,\n\t\tInput: tc.Input,\n\t\tExpected: tc.Expected,\n\t\tOptions: tc.Options,\n\t}\n\n\t// Execute before script if specified\n\tvar beforeData interface{}\n\tif tc.Before != \"\" {\n\t\tvar err error\n\t\tbeforeData, err = r.hookExecutor.ExecuteBefore(tc.Before, tc, r.agentPath)\n\t\tif err != nil {\n\t\t\tresult.Status = StatusError\n\t\t\tresult.Error = fmt.Sprintf(\"before script failed: %s\", err.Error())\n\t\t\tresult.DurationMs = time.Since(startTime).Milliseconds()\n\t\t\tr.output.TestResult(result.Status, time.Since(startTime))\n\t\t\tr.output.TestError(result.Error)\n\t\t\t// Note: after script is NOT called when before fails\n\t\t\treturn result\n\t\t}\n\t}\n\n\t// Ensure after script runs even if test fails (but only if before succeeded)\n\tdefer func() {\n\t\tif tc.After != \"\" && (tc.Before == \"\" || beforeData != nil || result.Status != StatusError || !isBeforeError(result.Error)) {\n\t\t\tif err := r.hookExecutor.ExecuteAfter(tc.After, tc, result, beforeData, r.agentPath); err != nil {\n\t\t\t\tr.output.Warning(\"after script failed: %s\", err.Error())\n\t\t\t}\n\t\t}\n\t}()\n\n\t// Parse input to messages with file loading support\n\t// BaseDir is derived from the input file directory\n\tinputOpts := r.getInputOptions()\n\tmessages, err := tc.GetMessagesWithOptions(inputOpts)\n\tif err != nil {\n\t\tresult.Status = StatusError\n\t\tresult.Error = fmt.Sprintf(\"failed to parse input: %s\", err.Error())\n\t\tresult.DurationMs = time.Since(startTime).Milliseconds()\n\t\tr.output.TestResult(result.Status, time.Since(startTime))\n\t\tr.output.TestError(result.Error)\n\t\treturn result\n\t}\n\n\t// Create context\n\tchatID := GenerateChatID(tc.ID, runNum)\n\tctx := NewTestContextFromOptions(chatID, agentID, r.opts, tc)\n\tdefer ctx.Release()\n\n\t// Build context options from test case and runner options\n\topts := buildContextOptions(tc, r.opts)\n\n\t// Create timeout context\n\ttimeout := tc.GetTimeout(r.opts.Timeout)\n\ttimeoutCtx, cancel := stdContext.WithTimeout(ctx.Context, timeout)\n\tdefer cancel()\n\tctx.Context = timeoutCtx\n\n\t// Run the test\n\tresponse, err := ast.Stream(ctx, messages, opts)\n\n\tduration := time.Since(startTime)\n\tresult.DurationMs = duration.Milliseconds()\n\n\t// Check for timeout\n\tif timeoutCtx.Err() != nil {\n\t\tresult.Status = StatusTimeout\n\t\tresult.Error = fmt.Sprintf(\"timeout after %s\", timeout)\n\t\tr.output.TestResult(result.Status, duration)\n\t\tr.output.TestError(result.Error)\n\t\treturn result\n\t}\n\n\t// Check for error\n\tif err != nil {\n\t\tresult.Status = StatusError\n\t\tresult.Error = err.Error()\n\t\tr.output.TestResult(result.Status, duration)\n\t\tr.output.TestError(result.Error)\n\t\treturn result\n\t}\n\n\t// Extract output\n\tresult.Output = extractOutput(response)\n\n\t// Validate result using asserter (with response for tool_called assertions)\n\tasserter := NewAsserter().WithResponse(response)\n\tpassed, errMsg := asserter.Validate(tc, result.Output)\n\tif passed {\n\t\tresult.Status = StatusPassed\n\t} else {\n\t\tresult.Status = StatusFailed\n\t\tresult.Error = errMsg\n\t}\n\n\tr.output.TestResult(result.Status, duration)\n\tif result.Status == StatusFailed {\n\t\tr.output.TestError(result.Error)\n\t}\n\tr.output.TestOutput(fmt.Sprintf(\"%v\", result.Output))\n\n\treturn result\n}\n\n// runDynamicTest runs a dynamic (simulator-driven) test case\nfunc (r *Executor) runDynamicTest(ast *assistant.Assistant, tc *Case, agentID string) *Result {\n\t// Output test start for dynamic mode\n\tr.output.DynamicTestStart(tc.ID, len(tc.Checkpoints))\n\n\tstartTime := time.Now()\n\n\t// Execute before script if specified\n\tvar beforeData interface{}\n\tif tc.Before != \"\" {\n\t\tvar err error\n\t\tbeforeData, err = r.hookExecutor.ExecuteBefore(tc.Before, tc, r.agentPath)\n\t\tif err != nil {\n\t\t\tresult := &Result{\n\t\t\t\tID: tc.ID,\n\t\t\t\tStatus: StatusError,\n\t\t\t\tError: fmt.Sprintf(\"before script failed: %s\", err.Error()),\n\t\t\t\tDurationMs: time.Since(startTime).Milliseconds(),\n\t\t\t}\n\t\t\tr.output.TestResult(result.Status, time.Since(startTime))\n\t\t\tr.output.TestError(result.Error)\n\t\t\treturn result\n\t\t}\n\t}\n\n\t// Create dynamic runner and execute\n\tdynamicRunner := NewDynamicRunner(r.opts)\n\tdynamicResult := dynamicRunner.RunDynamic(ast, tc, agentID)\n\n\t// Convert to standard result\n\tresult := dynamicResult.ToResult()\n\n\t// Execute after script if specified (before outputting result)\n\tif tc.After != \"\" && (tc.Before == \"\" || beforeData != nil || result.Status != StatusError || !isBeforeError(result.Error)) {\n\t\tif err := r.hookExecutor.ExecuteAfter(tc.After, tc, result, beforeData, r.agentPath); err != nil {\n\t\t\tr.output.Warning(\"after script failed: %s\", err.Error())\n\t\t}\n\t}\n\n\t// Output result\n\tduration := time.Duration(result.DurationMs) * time.Millisecond\n\tr.output.DynamicTestResult(result.Status, dynamicResult.TotalTurns, len(tc.Checkpoints), duration)\n\n\tif result.Error != \"\" {\n\t\tr.output.TestError(result.Error)\n\t}\n\n\treturn result\n}\n\n// isBeforeError checks if the error message indicates a before script failure\nfunc isBeforeError(errMsg string) bool {\n\treturn len(errMsg) > 0 && errMsg[:min(len(errMsg), 20)] == \"before script failed\"\n}\n\n// min returns the minimum of two integers\nfunc min(a, b int) int {\n\tif a < b {\n\t\treturn a\n\t}\n\treturn b\n}\n\n// runStabilityTests runs each test case multiple times for stability analysis\nfunc (r *Executor) runStabilityTests(ast *assistant.Assistant, testCases []*Case, agentID string) []*StabilityResult {\n\tresults := make([]*StabilityResult, 0, len(testCases))\n\n\tfor _, tc := range testCases {\n\t\tsr := &StabilityResult{\n\t\t\tID: tc.ID,\n\t\t\tInput: tc.Input,\n\t\t\tExpected: tc.Expected,\n\t\t\tRunDetails: make([]*RunDetail, 0, r.opts.Runs),\n\t\t}\n\n\t\t// Run multiple times\n\t\tfor run := 1; run <= r.opts.Runs; run++ {\n\t\t\tresult := r.runSingleTest(ast, tc, agentID, run)\n\n\t\t\trd := &RunDetail{\n\t\t\t\tRun: run,\n\t\t\t\tStatus: result.Status,\n\t\t\t\tDurationMs: result.DurationMs,\n\t\t\t\tOutput: result.Output,\n\t\t\t\tError: result.Error,\n\t\t\t}\n\t\t\tsr.RunDetails = append(sr.RunDetails, rd)\n\t\t}\n\n\t\t// Calculate stability metrics\n\t\tsr.CalculateStability()\n\n\t\t// Print stability result\n\t\tr.output.StabilityResult(sr)\n\n\t\tresults = append(results, sr)\n\n\t\t// Check fail-fast\n\t\tif r.opts.FailFast && !sr.Stable {\n\t\t\tr.output.Warning(\"Stopping due to --fail-fast (unstable test: %s)\", tc.ID)\n\t\t\tbreak\n\t\t}\n\t}\n\n\treturn results\n}\n\n// calculateSingleSummary calculates summary for single run mode\nfunc (r *Executor) calculateSingleSummary(report *Report) {\n\tfor _, result := range report.Results {\n\t\tswitch result.Status {\n\t\tcase StatusPassed:\n\t\t\treport.Summary.Passed++\n\t\tcase StatusFailed:\n\t\t\treport.Summary.Failed++\n\t\tcase StatusError:\n\t\t\treport.Summary.Errors++\n\t\tcase StatusTimeout:\n\t\t\treport.Summary.Timeouts++\n\t\t}\n\t}\n}\n\n// calculateStabilitySummary calculates summary for stability mode\nfunc (r *Executor) calculateStabilitySummary(report *Report) {\n\treport.Summary.TotalRuns = len(report.StabilityResults) * r.opts.Runs\n\n\tvar totalPassRate float64\n\tfor _, sr := range report.StabilityResults {\n\t\tif sr.Stable {\n\t\t\treport.Summary.StableCases++\n\t\t\treport.Summary.Passed++\n\t\t} else {\n\t\t\treport.Summary.UnstableCases++\n\t\t\treport.Summary.Failed++\n\t\t}\n\t\ttotalPassRate += sr.PassRate\n\t}\n\n\tif len(report.StabilityResults) > 0 {\n\t\treport.Summary.OverallPassRate = totalPassRate / float64(len(report.StabilityResults))\n\t}\n}\n\n// writeOutput writes the test report to the output file\nfunc (r *Executor) writeOutput(report *Report) error {\n\tfile, err := os.Create(r.opts.OutputFile)\n\tif err != nil {\n\t\treturn fmt.Errorf(\"failed to create output file: %w\", err)\n\t}\n\tdefer file.Close()\n\n\t// Get reporter based on -r flag or file extension\n\treporter := GetReporterWithAgent(r.opts.ReporterID, r.opts.OutputFile, r.opts.Verbose)\n\n\t// If using agent reporter, set context\n\tif agentReporter, ok := reporter.(*AgentReporter); ok {\n\t\t// Create a context for the reporter agent call\n\t\tctx := NewTestContext(\"reporter\", r.opts.ReporterID, report.Environment)\n\t\tdefer ctx.Release()\n\t\tagentReporter.SetContext(ctx)\n\t}\n\n\t// Write report using the reporter\n\treturn reporter.Write(report, file)\n}\n\n// buildContextOptions builds context.Options from test case and runner options\n// Priority: test case options > runner options > defaults\nfunc buildContextOptions(tc *Case, runnerOpts *Options) *context.Options {\n\topts := &context.Options{\n\t\tSkip: &context.Skip{\n\t\t\tHistory: true, // Default: skip history loading - input already contains full conversation\n\t\t},\n\t}\n\n\t// Apply test case options if specified\n\tif tc.Options != nil {\n\t\t// Connector: test case > runner\n\t\tif tc.Options.Connector != \"\" {\n\t\t\topts.Connector = tc.Options.Connector\n\t\t}\n\n\t\t// Mode\n\t\tif tc.Options.Mode != \"\" {\n\t\t\topts.Mode = tc.Options.Mode\n\t\t}\n\n\t\t// DisableGlobalPrompts\n\t\tif tc.Options.DisableGlobalPrompts {\n\t\t\topts.DisableGlobalPrompts = true\n\t\t}\n\n\t\t// Search (pointer to distinguish unset from false)\n\t\tif tc.Options.Search != nil {\n\t\t\topts.Search = tc.Options.Search\n\t\t}\n\n\t\t// Metadata for hooks\n\t\tif tc.Options.Metadata != nil {\n\t\t\topts.Metadata = tc.Options.Metadata\n\t\t}\n\n\t\t// Skip options from test case\n\t\tif tc.Options.Skip != nil {\n\t\t\topts.Skip.Trace = tc.Options.Skip.Trace\n\t\t\topts.Skip.Output = tc.Options.Skip.Output\n\t\t\topts.Skip.Keyword = tc.Options.Skip.Keyword\n\t\t\topts.Skip.Search = tc.Options.Skip.Search\n\t\t\t// Note: History defaults to true for tests\n\t\t}\n\t}\n\n\t// Runner connector override (highest priority)\n\tif runnerOpts != nil && runnerOpts.Connector != \"\" {\n\t\topts.Connector = runnerOpts.Connector\n\t}\n\n\treturn opts\n}\n\n// extractOutput extracts the output from the agent response\n// Priority: Next hook data > Completion content > Tool results message > nil\nfunc extractOutput(response *context.Response) interface{} {\n\tif response == nil {\n\t\treturn nil\n\t}\n\n\t// Prefer Next hook data if available and non-empty\n\t// response.Next is already the Data value (not NextHookResponse struct)\n\tif response.Next != nil && !isEmptyValue(response.Next) {\n\t\treturn response.Next\n\t}\n\n\t// Fall back to completion response\n\tif response.Completion != nil {\n\t\t// If content is non-empty, return it\n\t\tif response.Completion.Content != nil && !isEmptyValue(response.Completion.Content) {\n\t\t\treturn response.Completion.Content\n\t\t}\n\t}\n\n\t// If no content but tools were executed, extract message from tool results\n\t// This handles the case where LLM calls tools but doesn't generate text\n\tif len(response.Tools) > 0 {\n\t\treturn extractToolResultMessage(response.Tools)\n\t}\n\n\treturn nil\n}\n\n// extractToolResultMessage extracts the message field from tool results\n// Returns the first non-empty message found, or a summary of tool calls\nfunc extractToolResultMessage(tools []context.ToolCallResponse) interface{} {\n\tif len(tools) == 0 {\n\t\treturn nil\n\t}\n\n\t// Try to extract \"message\" field from tool results first\n\tfor _, tool := range tools {\n\t\tif tool.Result != nil {\n\t\t\t// Try to get message from result map\n\t\t\tif resultMap, ok := tool.Result.(map[string]interface{}); ok {\n\t\t\t\tif msg, exists := resultMap[\"message\"]; exists && msg != nil {\n\t\t\t\t\tif msgStr, ok := msg.(string); ok && msgStr != \"\" {\n\t\t\t\t\t\treturn msgStr\n\t\t\t\t\t}\n\t\t\t\t}\n\t\t\t}\n\t\t}\n\t}\n\n\t// No message found, generate a summary of tool calls\n\tvar summaries []string\n\tfor _, tool := range tools {\n\t\ttoolName := tool.Tool\n\t\tif toolName == \"\" {\n\t\t\ttoolName = \"unknown\"\n\t\t}\n\t\t// Extract key info from result if possible\n\t\tif tool.Result != nil {\n\t\t\tif resultMap, ok := tool.Result.(map[string]interface{}); ok {\n\t\t\t\t// Try common result fields\n\t\t\t\tif action, ok := resultMap[\"action\"].(string); ok {\n\t\t\t\t\tsummaries = append(summaries, fmt.Sprintf(\"[%s: %s]\", toolName, action))\n\t\t\t\t\tcontinue\n\t\t\t\t}\n\t\t\t\tif success, ok := resultMap[\"success\"].(bool); ok {\n\t\t\t\t\tstatus := \"failed\"\n\t\t\t\t\tif success {\n\t\t\t\t\t\tstatus = \"success\"\n\t\t\t\t\t}\n\t\t\t\t\tsummaries = append(summaries, fmt.Sprintf(\"[%s: %s]\", toolName, status))\n\t\t\t\t\tcontinue\n\t\t\t\t}\n\t\t\t}\n\t\t}\n\t\tsummaries = append(summaries, fmt.Sprintf(\"[%s]\", toolName))\n\t}\n\n\tif len(summaries) > 0 {\n\t\treturn strings.Join(summaries, \" \")\n\t}\n\treturn nil\n}\n\n// isEmptyValue checks if a value is considered \"empty\" for output purposes\nfunc isEmptyValue(v interface{}) bool {\n\tif v == nil {\n\t\treturn true\n\t}\n\n\t// Use reflection to check for typed nil (e.g., *NextHookResponse(nil))\n\trv := reflect.ValueOf(v)\n\tif rv.Kind() == reflect.Ptr && rv.IsNil() {\n\t\treturn true\n\t}\n\n\tswitch val := v.(type) {\n\tcase string:\n\t\treturn val == \"\"\n\tcase map[string]interface{}:\n\t\treturn len(val) == 0\n\tcase []interface{}:\n\t\treturn len(val) == 0\n\tcase *context.NextHookResponse:\n\t\t// Check if NextHookResponse is effectively empty\n\t\tif val == nil {\n\t\t\treturn true\n\t\t}\n\t\treturn val.Data == nil && val.Delegate == nil\n\t}\n\n\treturn false\n}\n\n// validateOutput validates the actual output against expected\nfunc validateOutput(actual, expected interface{}) bool {\n\t// Simple JSON comparison\n\tactualJSON, err1 := jsoniter.Marshal(actual)\n\texpectedJSON, err2 := jsoniter.Marshal(expected)\n\n\tif err1 != nil || err2 != nil {\n\t\treturn false\n\t}\n\n\treturn string(actualJSON) == string(expectedJSON)\n}\n\n// getInputOptions returns InputOptions based on the runner configuration\n// BaseDir is derived from the input file directory (for file mode) or current working directory\nfunc (r *Executor) getInputOptions() *InputOptions {\n\topts := &InputOptions{}\n\n\t// For file mode, use the input file's directory as base\n\tif r.opts.InputMode == InputModeFile && r.opts.Input != \"\" {\n\t\t// Resolve path considering YAO_ROOT\n\t\tresolvedPath := ResolvePathWithYaoRoot(r.opts.Input)\n\t\topts.BaseDir = filepath.Dir(resolvedPath)\n\t}\n\t// For message mode, BaseDir remains empty (uses current working directory)\n\n\treturn opts\n}\n\n// outputDryRun outputs generated test cases without running them\nfunc (r *Executor) outputDryRun(testCases []*Case, agentInfo *AgentInfo) (*Report, error) {\n\tr.output.Info(\"Generated Test Cases:\")\n\n\t// Output each test case as JSONL\n\tfor _, tc := range testCases {\n\t\tdata, err := jsoniter.Marshal(tc)\n\t\tif err != nil {\n\t\t\tr.output.Warning(\"Failed to marshal test case %s: %s\", tc.ID, err.Error())\n\t\t\tcontinue\n\t\t}\n\t\tfmt.Println(string(data))\n\t}\n\n\t// Write to output file if specified\n\tif r.opts.OutputFile != \"\" {\n\t\tfile, err := os.Create(r.opts.OutputFile)\n\t\tif err != nil {\n\t\t\treturn nil, fmt.Errorf(\"failed to create output file: %w\", err)\n\t\t}\n\t\tdefer file.Close()\n\n\t\tfor _, tc := range testCases {\n\t\t\tdata, err := jsoniter.Marshal(tc)\n\t\t\tif err != nil {\n\t\t\t\tcontinue\n\t\t\t}\n\t\t\tfile.WriteString(string(data) + \"\\n\")\n\t\t}\n\n\t\tr.output.Info(\"Output written to: %s\", r.opts.OutputFile)\n\t}\n\n\t// Return a minimal report\n\tconnector := r.opts.Connector\n\tif connector == \"\" {\n\t\tconnector = agentInfo.Connector\n\t}\n\n\treturn &Report{\n\t\tSummary: &Summary{\n\t\t\tTotal: len(testCases),\n\t\t\tAgentID: agentInfo.ID,\n\t\t\tConnector: connector,\n\t\t},\n\t}, nil\n}\n" }, { "path": "agent/test/runner_integration_test.go", "content": "package test_test\n\nimport (\n\t\"os\"\n\t\"path/filepath\"\n\t\"testing\"\n\n\t\"github.com/stretchr/testify/assert\"\n\t\"github.com/stretchr/testify/require\"\n\t\"github.com/yaoapp/yao/agent\"\n\tagenttest \"github.com/yaoapp/yao/agent/test\"\n\t\"github.com/yaoapp/yao/config\"\n\t\"github.com/yaoapp/yao/test\"\n)\n\n// TestRunner_AgentDrivenInput tests the complete flow:\n// 1. Use generator-agent to generate test cases\n// 2. Run the generated tests against simple-greeting agent\nfunc TestRunner_AgentDrivenInput(t *testing.T) {\n\ttest.Prepare(t, config.Conf)\n\tdefer test.Clean()\n\n\t// Load agents\n\terr := agent.Load(config.Conf)\n\trequire.NoError(t, err, \"Failed to load agents\")\n\n\t// Test with agent-driven input\n\topts := &agenttest.Options{\n\t\tInput: \"agents:tests.generator-agent?count=3\",\n\t\tAgentID: \"tests.simple-greeting\",\n\t\tVerbose: true,\n\t\tInputMode: agenttest.InputModeFile, // Will be overridden by ParseInputSource\n\t}\n\topts = agenttest.MergeOptions(opts, agenttest.DefaultOptions())\n\n\trunner := agenttest.NewRunner(opts)\n\treport, err := runner.Run()\n\n\trequire.NoError(t, err, \"Runner should not return error\")\n\trequire.NotNil(t, report, \"Report should not be nil\")\n\trequire.NotNil(t, report.Summary, \"Summary should not be nil\")\n\n\t// Verify report\n\tassert.Greater(t, report.Summary.Total, 0, \"Should have at least one test case\")\n\tt.Logf(\"Total: %d, Passed: %d, Failed: %d\",\n\t\treport.Summary.Total, report.Summary.Passed, report.Summary.Failed)\n}\n\n// TestRunner_AgentDrivenInput_DryRun tests dry-run mode\nfunc TestRunner_AgentDrivenInput_DryRun(t *testing.T) {\n\ttest.Prepare(t, config.Conf)\n\tdefer test.Clean()\n\n\t// Load agents\n\terr := agent.Load(config.Conf)\n\trequire.NoError(t, err, \"Failed to load agents\")\n\n\t// Test with dry-run mode\n\topts := &agenttest.Options{\n\t\tInput: \"agents:tests.generator-agent?count=2\",\n\t\tAgentID: \"tests.simple-greeting\",\n\t\tDryRun: true,\n\t\tVerbose: true,\n\t}\n\topts = agenttest.MergeOptions(opts, agenttest.DefaultOptions())\n\n\trunner := agenttest.NewRunner(opts)\n\treport, err := runner.Run()\n\n\trequire.NoError(t, err, \"Dry-run should not return error\")\n\trequire.NotNil(t, report, \"Report should not be nil\")\n\n\t// In dry-run mode, tests are generated but not executed\n\t// So Passed and Failed should both be 0, but Total should have the count\n\tassert.Greater(t, report.Summary.Total, 0, \"Should have generated test cases\")\n\n\tt.Logf(\"Generated %d test cases in dry-run mode\", report.Summary.Total)\n}\n\n// TestRunner_FileInput tests loading test cases from JSONL file\nfunc TestRunner_FileInput(t *testing.T) {\n\ttest.Prepare(t, config.Conf)\n\tdefer test.Clean()\n\n\t// Load agents\n\terr := agent.Load(config.Conf)\n\trequire.NoError(t, err, \"Failed to load agents\")\n\n\t// Create a temporary JSONL file with test cases\n\t// Use case-insensitive contains for robustness\n\ttmpDir := t.TempDir()\n\tinputFile := filepath.Join(tmpDir, \"inputs.jsonl\")\n\n\ttestCases := `{\"id\": \"greeting-hello\", \"input\": \"Hello\", \"assert\": {\"type\": \"regex\", \"value\": \"(?i)hello\"}}\n{\"id\": \"greeting-hi\", \"input\": \"Hi there\", \"assert\": {\"type\": \"regex\", \"value\": \"(?i)(hi|hello)\"}}\n{\"id\": \"greeting-morning\", \"input\": \"Good morning\", \"assert\": {\"type\": \"regex\", \"value\": \"(?i)(hello|morning|good)\"}}`\n\n\terr = os.WriteFile(inputFile, []byte(testCases), 0644)\n\trequire.NoError(t, err, \"Failed to write test file\")\n\n\t// Run tests from file\n\topts := &agenttest.Options{\n\t\tInput: inputFile,\n\t\tAgentID: \"tests.simple-greeting\",\n\t\tVerbose: true,\n\t\tInputMode: agenttest.InputModeFile,\n\t}\n\topts = agenttest.MergeOptions(opts, agenttest.DefaultOptions())\n\n\trunner := agenttest.NewRunner(opts)\n\treport, err := runner.Run()\n\n\trequire.NoError(t, err, \"Runner should not return error\")\n\trequire.NotNil(t, report, \"Report should not be nil\")\n\n\t// Verify report\n\tassert.Equal(t, 3, report.Summary.Total, \"Should have 3 test cases\")\n\tt.Logf(\"Total: %d, Passed: %d, Failed: %d\",\n\t\treport.Summary.Total, report.Summary.Passed, report.Summary.Failed)\n\n\t// Check results for debugging\n\tif report.Results != nil {\n\t\tfor _, r := range report.Results {\n\t\t\tt.Logf(\" [%s] Status: %s, Output: %v\", r.ID, r.Status, r.Output)\n\t\t}\n\t}\n}\n\n// TestRunner_DirectMessage tests direct message mode\nfunc TestRunner_DirectMessage(t *testing.T) {\n\ttest.Prepare(t, config.Conf)\n\tdefer test.Clean()\n\n\t// Load agents\n\terr := agent.Load(config.Conf)\n\trequire.NoError(t, err, \"Failed to load agents\")\n\n\t// Test with direct message\n\topts := &agenttest.Options{\n\t\tInput: \"Hello, how are you?\",\n\t\tAgentID: \"tests.simple-greeting\",\n\t\tVerbose: true,\n\t\tInputMode: agenttest.InputModeMessage,\n\t}\n\topts = agenttest.MergeOptions(opts, agenttest.DefaultOptions())\n\n\trunner := agenttest.NewRunner(opts)\n\treport, err := runner.Run()\n\n\trequire.NoError(t, err, \"Runner should not return error\")\n\trequire.NotNil(t, report, \"Report should not be nil\")\n\n\t// Direct message mode returns a minimal report\n\tassert.Equal(t, 1, report.Summary.Total, \"Should have 1 test case\")\n\tassert.Equal(t, 1, report.Summary.Passed, \"Direct message should pass\")\n}\n\n// TestRunner_WithBeforeAfter tests before/after hooks\nfunc TestRunner_WithBeforeAfter(t *testing.T) {\n\ttest.Prepare(t, config.Conf)\n\tdefer test.Clean()\n\n\t// Load agents\n\terr := agent.Load(config.Conf)\n\trequire.NoError(t, err, \"Failed to load agents\")\n\n\t// Create a temporary JSONL file with test cases that use hooks\n\ttmpDir := t.TempDir()\n\tinputFile := filepath.Join(tmpDir, \"inputs.jsonl\")\n\n\t// Note: hooks-test agent has env_test.ts with Before/After functions\n\ttestCases := `{\"id\": \"hook-test-1\", \"input\": \"Hello\", \"assert\": {\"type\": \"contains\", \"value\": \"hello\"}, \"before\": \"env_test.Before\", \"after\": \"env_test.After\"}`\n\n\terr = os.WriteFile(inputFile, []byte(testCases), 0644)\n\trequire.NoError(t, err, \"Failed to write test file\")\n\n\t// Run tests with hooks (using hooks-test agent which has the hook scripts)\n\topts := &agenttest.Options{\n\t\tInput: inputFile,\n\t\tAgentID: \"tests.hooks-test\",\n\t\tVerbose: true,\n\t\tInputMode: agenttest.InputModeFile,\n\t}\n\topts = agenttest.MergeOptions(opts, agenttest.DefaultOptions())\n\n\trunner := agenttest.NewRunner(opts)\n\treport, err := runner.Run()\n\n\trequire.NoError(t, err, \"Runner should not return error\")\n\trequire.NotNil(t, report, \"Report should not be nil\")\n\n\tt.Logf(\"Total: %d, Passed: %d, Failed: %d\",\n\t\treport.Summary.Total, report.Summary.Passed, report.Summary.Failed)\n}\n\n// TestRunner_Parallel tests parallel execution\nfunc TestRunner_Parallel(t *testing.T) {\n\ttest.Prepare(t, config.Conf)\n\tdefer test.Clean()\n\n\t// Load agents\n\terr := agent.Load(config.Conf)\n\trequire.NoError(t, err, \"Failed to load agents\")\n\n\t// Create a temporary JSONL file with multiple test cases\n\t// Use regex for case-insensitive matching\n\ttmpDir := t.TempDir()\n\tinputFile := filepath.Join(tmpDir, \"inputs.jsonl\")\n\n\ttestCases := `{\"id\": \"parallel-1\", \"input\": \"Hello\", \"assert\": {\"type\": \"regex\", \"value\": \"(?i)(hello|hi)\"}}\n{\"id\": \"parallel-2\", \"input\": \"Hi\", \"assert\": {\"type\": \"regex\", \"value\": \"(?i)(hello|hi)\"}}\n{\"id\": \"parallel-3\", \"input\": \"Hey\", \"assert\": {\"type\": \"regex\", \"value\": \"(?i)(hello|hi|hey)\"}}\n{\"id\": \"parallel-4\", \"input\": \"Good day\", \"assert\": {\"type\": \"regex\", \"value\": \"(?i)(hello|good|day)\"}}`\n\n\terr = os.WriteFile(inputFile, []byte(testCases), 0644)\n\trequire.NoError(t, err, \"Failed to write test file\")\n\n\t// Run tests in parallel\n\topts := &agenttest.Options{\n\t\tInput: inputFile,\n\t\tAgentID: \"tests.simple-greeting\",\n\t\tParallel: 2, // Run 2 tests in parallel\n\t\tVerbose: true,\n\t\tInputMode: agenttest.InputModeFile,\n\t}\n\topts = agenttest.MergeOptions(opts, agenttest.DefaultOptions())\n\n\trunner := agenttest.NewRunner(opts)\n\treport, err := runner.Run()\n\n\trequire.NoError(t, err, \"Runner should not return error\")\n\trequire.NotNil(t, report, \"Report should not be nil\")\n\n\tassert.Equal(t, 4, report.Summary.Total, \"Should have 4 test cases\")\n\tt.Logf(\"Total: %d, Passed: %d, Failed: %d (parallel: 2)\",\n\t\treport.Summary.Total, report.Summary.Passed, report.Summary.Failed)\n}\n\n// TestRunner_FailFast tests fail-fast behavior\nfunc TestRunner_FailFast(t *testing.T) {\n\ttest.Prepare(t, config.Conf)\n\tdefer test.Clean()\n\n\t// Load agents\n\terr := agent.Load(config.Conf)\n\trequire.NoError(t, err, \"Failed to load agents\")\n\n\t// Create a temporary JSONL file with a failing test first\n\ttmpDir := t.TempDir()\n\tinputFile := filepath.Join(tmpDir, \"inputs.jsonl\")\n\n\t// First test will fail (expects \"impossible\" which won't be in response)\n\ttestCases := `{\"id\": \"fail-first\", \"input\": \"Hello\", \"assert\": {\"type\": \"contains\", \"value\": \"IMPOSSIBLE_STRING_12345\"}}\n{\"id\": \"should-skip\", \"input\": \"Hi\", \"assert\": {\"type\": \"contains\", \"value\": \"hi\"}}`\n\n\terr = os.WriteFile(inputFile, []byte(testCases), 0644)\n\trequire.NoError(t, err, \"Failed to write test file\")\n\n\t// Run tests with fail-fast\n\topts := &agenttest.Options{\n\t\tInput: inputFile,\n\t\tAgentID: \"tests.simple-greeting\",\n\t\tFailFast: true,\n\t\tVerbose: true,\n\t\tInputMode: agenttest.InputModeFile,\n\t}\n\topts = agenttest.MergeOptions(opts, agenttest.DefaultOptions())\n\n\trunner := agenttest.NewRunner(opts)\n\treport, err := runner.Run()\n\n\trequire.NoError(t, err, \"Runner should not return error (fail-fast is not an error)\")\n\trequire.NotNil(t, report, \"Report should not be nil\")\n\n\t// With fail-fast, only the first test should run\n\tassert.Equal(t, 1, report.Summary.Failed, \"First test should fail\")\n\t// The second test might not run due to fail-fast\n\tt.Logf(\"Total: %d, Passed: %d, Failed: %d (fail-fast enabled)\",\n\t\treport.Summary.Total, report.Summary.Passed, report.Summary.Failed)\n}\n" }, { "path": "agent/test/script.go", "content": "package test\n\nimport (\n\t\"fmt\"\n\t\"path/filepath\"\n\t\"regexp\"\n\t\"strings\"\n\t\"time\"\n\n\t\"github.com/yaoapp/gou/application\"\n\tv8 \"github.com/yaoapp/gou/runtime/v8\"\n\t\"github.com/yaoapp/gou/runtime/v8/bridge\"\n\t\"github.com/yaoapp/yao/agent/context\"\n\t\"rogchap.com/v8go\"\n)\n\n// ScriptRunner executes script tests\ntype ScriptRunner struct {\n\topts *Options\n\toutput *OutputWriter\n}\n\n// NewScriptRunner creates a new script test runner\nfunc NewScriptRunner(opts *Options) *ScriptRunner {\n\treturn &ScriptRunner{\n\t\topts: opts,\n\t\toutput: NewOutputWriter(opts.Verbose),\n\t}\n}\n\n// ResolveScript resolves the script path from scripts.xxx.yyy or scripts.xxx.yyy.zzz format\n//\n// Resolution strategy:\n// 1. Find the assistant directory by detecting package.yao from longest to shortest path\n// 2. Remaining parts after the assistant boundary map to src/ subdirectories + module name\n//\n// Examples:\n// - scripts.expense.setup -> assistants/expense/src/setup_test.ts\n// - scripts.yao.keeper.seed -> assistants/yao/keeper/src/seed_test.ts\n// - scripts.yao.keeper.tests.seed -> assistants/yao/keeper/src/tests/seed_test.ts\nfunc ResolveScript(input string) (*ScriptInfo, error) {\n\t// Remove \"scripts.\" prefix\n\tpath := strings.TrimPrefix(input, \"scripts.\")\n\n\t// Split into parts:\n\t// \"expense.setup\" -> [\"expense\", \"setup\"]\n\t// \"yao.keeper.tests.seed\" -> [\"yao\", \"keeper\", \"tests\", \"seed\"]\n\tparts := strings.Split(path, \".\")\n\tif len(parts) < 2 {\n\t\treturn nil, fmt.Errorf(\"invalid script path: %s (expected format: scripts.assistant.module or scripts.assistant.sub.module)\", input)\n\t}\n\n\t// Strategy: detect assistant boundary by looking for package.yao\n\t// Try from the longest possible assistant path down to the shortest\n\tvar assistantDir, modulePath string\n\tassistantFound := false\n\n\tfor i := len(parts) - 1; i >= 1; i-- {\n\t\tcandidateDir := strings.Join(parts[:i], \"/\")\n\t\tfor _, prefix := range []string{\"assistants/\", \"\"} {\n\t\t\tpackagePath := filepath.Join(prefix+candidateDir, \"package.yao\")\n\t\t\texists, err := application.App.Exists(packagePath)\n\t\t\tif err == nil && exists {\n\t\t\t\tassistantDir = candidateDir\n\t\t\t\t// Remaining parts form the module path (may include subdirectories)\n\t\t\t\t// e.g., parts[i:] = [\"tests\", \"seed\"] -> \"tests/seed\"\n\t\t\t\tmodulePath = strings.Join(parts[i:], \"/\")\n\t\t\t\tassistantFound = true\n\t\t\t\tbreak\n\t\t\t}\n\t\t}\n\t\tif assistantFound {\n\t\t\tbreak\n\t\t}\n\t}\n\n\t// Fallback: original behavior — last part is module, rest is assistant dir\n\tif !assistantFound {\n\t\tassistantDir = strings.Join(parts[:len(parts)-1], \"/\")\n\t\tmodulePath = parts[len(parts)-1]\n\t}\n\n\t// modulePath may contain subdirectories: \"tests/seed\" -> dir=\"tests\", module=\"seed\"\n\tmoduleName := filepath.Base(modulePath)\n\tmoduleSubDir := filepath.Dir(modulePath)\n\tif moduleSubDir == \".\" {\n\t\tmoduleSubDir = \"\"\n\t}\n\n\t// Build candidate base paths\n\tbasePaths := []string{\n\t\tfilepath.Join(\"assistants\", assistantDir, \"src\", moduleSubDir),\n\t\tfilepath.Join(assistantDir, \"src\", moduleSubDir),\n\t}\n\n\tvar scriptPath, testPath string\n\tfor _, basePath := range basePaths {\n\t\t// Check for TypeScript files first, then JavaScript\n\t\tfor _, ext := range []string{\".ts\", \".js\"} {\n\t\t\tcandidateScript := filepath.Join(basePath, moduleName+ext)\n\t\t\tcandidateTest := filepath.Join(basePath, moduleName+\"_test\"+ext)\n\n\t\t\t// Check if test file exists\n\t\t\texists, err := application.App.Exists(candidateTest)\n\t\t\tif err == nil && exists {\n\t\t\t\tscriptPath = candidateScript\n\t\t\t\ttestPath = candidateTest\n\t\t\t\tbreak\n\t\t\t}\n\t\t}\n\t\tif testPath != \"\" {\n\t\t\tbreak\n\t\t}\n\t}\n\n\tif testPath == \"\" {\n\t\treturn nil, fmt.Errorf(\"test file not found for %s (tried: %s)\", input, strings.Join(basePaths, \", \"))\n\t}\n\n\treturn &ScriptInfo{\n\t\tID: input,\n\t\tAssistant: assistantDir,\n\t\tModule: moduleName,\n\t\tScriptPath: scriptPath,\n\t\tTestPath: testPath,\n\t}, nil\n}\n\n// DiscoverTests finds all Test* functions in the script\nfunc DiscoverTests(scriptPath string) ([]*ScriptTestCase, error) {\n\t// Read the script file\n\tcontent, err := application.App.Read(scriptPath)\n\tif err != nil {\n\t\treturn nil, fmt.Errorf(\"failed to read script: %w\", err)\n\t}\n\n\t// Parse the script to find Test* functions\n\t// We use a simple regex-like approach to find function declarations\n\ttests := make([]*ScriptTestCase, 0)\n\tlines := strings.Split(string(content), \"\\n\")\n\n\tfor _, line := range lines {\n\t\tline = strings.TrimSpace(line)\n\n\t\t// Match function declarations: function TestXxx( or export function TestXxx(\n\t\tif strings.Contains(line, \"function Test\") {\n\t\t\t// Extract function name\n\t\t\tname := extractFunctionName(line)\n\t\t\tif name != \"\" && strings.HasPrefix(name, \"Test\") {\n\t\t\t\ttests = append(tests, &ScriptTestCase{\n\t\t\t\t\tName: name,\n\t\t\t\t\tFunction: name,\n\t\t\t\t})\n\t\t\t}\n\t\t}\n\t}\n\n\treturn tests, nil\n}\n\n// extractFunctionName extracts the function name from a line\nfunc extractFunctionName(line string) string {\n\t// Remove \"export\" prefix if present\n\tline = strings.TrimPrefix(line, \"export \")\n\tline = strings.TrimSpace(line)\n\n\t// Match \"function Name(\"\n\tif !strings.HasPrefix(line, \"function \") {\n\t\treturn \"\"\n\t}\n\n\tline = strings.TrimPrefix(line, \"function \")\n\n\t// Find the opening parenthesis\n\tidx := strings.Index(line, \"(\")\n\tif idx == -1 {\n\t\treturn \"\"\n\t}\n\n\treturn strings.TrimSpace(line[:idx])\n}\n\n// filterTests filters test cases by a regex pattern (similar to go test -run)\nfunc (r *ScriptRunner) filterTests(tests []*ScriptTestCase, pattern string) ([]*ScriptTestCase, error) {\n\tre, err := regexp.Compile(pattern)\n\tif err != nil {\n\t\treturn nil, err\n\t}\n\n\tfiltered := make([]*ScriptTestCase, 0)\n\tfor _, tc := range tests {\n\t\tif re.MatchString(tc.Name) {\n\t\t\tfiltered = append(filtered, tc)\n\t\t}\n\t}\n\n\treturn filtered, nil\n}\n\n// Run executes all script tests and returns a report\nfunc (r *ScriptRunner) Run() (*ScriptTestReport, error) {\n\tstartTime := time.Now()\n\n\t// Resolve script\n\tscriptInfo, err := ResolveScript(r.opts.Input)\n\tif err != nil {\n\t\treturn nil, err\n\t}\n\n\t// Print header\n\tr.output.Header(\"Script Test\")\n\tr.output.Info(\"Script: %s\", scriptInfo.TestPath)\n\n\t// Discover tests\n\ttests, err := DiscoverTests(scriptInfo.TestPath)\n\tif err != nil {\n\t\treturn nil, err\n\t}\n\n\t// Filter tests by -run pattern if specified\n\tif r.opts.Run != \"\" {\n\t\ttests, err = r.filterTests(tests, r.opts.Run)\n\t\tif err != nil {\n\t\t\treturn nil, fmt.Errorf(\"invalid -run pattern: %w\", err)\n\t\t}\n\t\tr.output.Info(\"Tests: %d functions (filtered by: %s)\", len(tests), r.opts.Run)\n\t} else {\n\t\tr.output.Info(\"Tests: %d functions\", len(tests))\n\t}\n\n\tif len(tests) == 0 {\n\t\tr.output.Warning(\"No tests to run\")\n\t}\n\n\t// Load context config if specified\n\tvar ctxConfig *ContextConfig\n\tif r.opts.ContextFile != \"\" {\n\t\tvar err error\n\t\tctxConfig, err = LoadContextConfig(r.opts.ContextFile)\n\t\tif err != nil {\n\t\t\treturn nil, fmt.Errorf(\"failed to load context file: %w\", err)\n\t\t}\n\t\tr.output.Info(\"Context: %s\", r.opts.ContextFile)\n\t}\n\n\t// Create environment with optional context config\n\tvar env *Environment\n\tif ctxConfig != nil {\n\t\tenv = NewEnvironmentWithContext(r.opts.UserID, r.opts.TeamID, ctxConfig)\n\t} else {\n\t\tenv = NewEnvironment(r.opts.UserID, r.opts.TeamID)\n\t}\n\tr.output.Info(\"User: %s\", env.UserID)\n\tr.output.Info(\"Team: %s\", env.TeamID)\n\n\t// Load all scripts from src directory (including the test file)\n\t// This ensures imports can be resolved properly\n\tsrcDir := filepath.Dir(scriptInfo.TestPath)\n\tloadedCount, err := r.loadAllScripts(srcDir)\n\tif err != nil {\n\t\treturn nil, fmt.Errorf(\"failed to load scripts: %w\", err)\n\t}\n\tr.output.Info(\"Loaded: %d scripts\", loadedCount)\n\n\t// Create report\n\treport := &ScriptTestReport{\n\t\tType: \"script_test\",\n\t\tScript: scriptInfo.ID,\n\t\tScriptPath: scriptInfo.TestPath,\n\t\tSummary: &ScriptTestSummary{Total: len(tests)},\n\t\tEnvironment: env,\n\t\tResults: make([]*ScriptTestResult, 0, len(tests)),\n\t\tMetadata: &ScriptTestMetadata{\n\t\t\tStartedAt: startTime,\n\t\t},\n\t}\n\n\t// Run tests\n\tr.output.SubHeader(\"Running Tests\")\n\n\tfor _, tc := range tests {\n\t\tresult := r.runScriptTest(tc, scriptInfo, env)\n\t\treport.Results = append(report.Results, result)\n\n\t\t// Update summary\n\t\tswitch result.Status {\n\t\tcase StatusPassed:\n\t\t\treport.Summary.Passed++\n\t\tcase StatusFailed, StatusError:\n\t\t\t// Both Failed and Error count as failures\n\t\t\treport.Summary.Failed++\n\t\tcase StatusSkipped:\n\t\t\treport.Summary.Skipped++\n\t\t}\n\n\t\t// Check fail-fast (stop on both Failed and Error)\n\t\tif r.opts.FailFast && (result.Status == StatusFailed || result.Status == StatusError) {\n\t\t\tr.output.Warning(\"Stopping due to --fail-fast\")\n\t\t\tbreak\n\t\t}\n\t}\n\n\t// Complete report\n\treport.Summary.DurationMs = time.Since(startTime).Milliseconds()\n\treport.Metadata.CompletedAt = time.Now()\n\n\t// Print summary\n\tr.output.ScriptTestSummary(report.Summary, time.Since(startTime))\n\n\treturn report, nil\n}\n\n// runScriptTest runs a single script test function\nfunc (r *ScriptRunner) runScriptTest(tc *ScriptTestCase, scriptInfo *ScriptInfo, env *Environment) *ScriptTestResult {\n\tr.output.TestStart(tc.Name, \"\", 1)\n\tstartTime := time.Now()\n\n\tresult := &ScriptTestResult{\n\t\tName: tc.Name,\n\t\tStatus: StatusPassed,\n\t}\n\n\t// Create testing.T object\n\ttestingT := NewTestingT(tc.Name)\n\n\t// Create agent context\n\tchatID := fmt.Sprintf(\"script-test-%s\", tc.Name)\n\tagentCtx := NewTestContext(chatID, scriptInfo.Assistant, env)\n\tdefer agentCtx.Release()\n\n\t// Execute the test function\n\terr := r.executeTestFunction(tc, scriptInfo, testingT, agentCtx)\n\n\tduration := time.Since(startTime)\n\tresult.DurationMs = duration.Milliseconds()\n\tresult.Logs = testingT.Logs()\n\n\tif err != nil {\n\t\tresult.Status = StatusError\n\t\tresult.Error = err.Error()\n\t\tr.output.TestResult(result.Status, duration)\n\t\tr.output.TestError(result.Error)\n\t\treturn result\n\t}\n\n\tif testingT.Skipped() {\n\t\tresult.Status = StatusSkipped\n\t\tr.output.TestResult(result.Status, duration)\n\t\treturn result\n\t}\n\n\tif testingT.Failed() {\n\t\tresult.Status = StatusFailed\n\t\terrors := testingT.Errors()\n\t\tif len(errors) > 0 {\n\t\t\tresult.Error = errors[0]\n\t\t}\n\t\tresult.Assertion = testingT.AssertionInfo()\n\t\tr.output.TestResult(result.Status, duration)\n\t\tr.output.TestError(result.Error)\n\t\treturn result\n\t}\n\n\tr.output.TestResult(result.Status, duration)\n\treturn result\n}\n\n// loadAllScripts loads all scripts from the src directory\n// This ensures that imports can be resolved properly\nfunc (r *ScriptRunner) loadAllScripts(srcDir string) (int, error) {\n\tcount := 0\n\n\t// Check if src directory exists\n\texists, err := application.App.Exists(srcDir)\n\tif err != nil {\n\t\treturn 0, err\n\t}\n\tif !exists {\n\t\treturn 0, fmt.Errorf(\"src directory not found: %s\", srcDir)\n\t}\n\n\t// Walk through src directory to find all script files\n\texts := []string{\"*.ts\", \"*.js\"}\n\n\terr = application.App.Walk(srcDir, func(root, file string, isdir bool) error {\n\t\tif isdir {\n\t\t\treturn nil\n\t\t}\n\n\t\t// Get relative path\n\t\trelPath := strings.TrimPrefix(file, root+\"/\")\n\n\t\t// Generate script ID from file path\n\t\tscriptID := generateTestScriptID(file, root)\n\n\t\t// Load the script\n\t\t_, err := v8.Load(file, scriptID)\n\t\tif err != nil {\n\t\t\t// Log warning but continue loading other scripts\n\t\t\tif r.opts.Verbose {\n\t\t\t\tr.output.Warning(\"Failed to load %s: %v\", relPath, err)\n\t\t\t}\n\t\t\treturn nil\n\t\t}\n\n\t\tcount++\n\t\tif r.opts.Verbose {\n\t\t\tr.output.Verbose(\"Loaded: %s\", relPath)\n\t\t}\n\n\t\treturn nil\n\t}, exts...)\n\n\tif err != nil {\n\t\treturn count, fmt.Errorf(\"failed to walk src directory: %w\", err)\n\t}\n\n\treturn count, nil\n}\n\n// generateTestScriptID generates a script ID from file path for testing\nfunc generateTestScriptID(filePath string, srcDir string) string {\n\t// Normalize path separators\n\tfilePath = filepath.ToSlash(filePath)\n\tsrcDir = filepath.ToSlash(srcDir)\n\n\t// Remove src directory prefix\n\trelPath := strings.TrimPrefix(filePath, srcDir+\"/\")\n\trelPath = strings.TrimPrefix(relPath, \"/\")\n\n\t// Remove file extension\n\trelPath = strings.TrimSuffix(relPath, filepath.Ext(relPath))\n\n\t// Replace path separators with dots and add test prefix\n\tscriptID := \"test.\" + strings.ReplaceAll(relPath, \"/\", \".\")\n\n\treturn scriptID\n}\n\n// executeTestFunction executes a single test function using V8\nfunc (r *ScriptRunner) executeTestFunction(tc *ScriptTestCase, scriptInfo *ScriptInfo, testingT *TestingT, agentCtx *context.Context) (execErr error) {\n\t// Recover from panics thrown by Process calls\n\t// Even if JavaScript try-catch catches the error, we want to fail the test\n\tdefer func() {\n\t\tif r := recover(); r != nil {\n\t\t\texecErr = fmt.Errorf(\"panic in test function: %v\", r)\n\t\t}\n\t}()\n\n\t// Get the test script (already loaded by loadAllScripts)\n\ttestScriptID := generateTestScriptID(scriptInfo.TestPath, filepath.Dir(scriptInfo.TestPath))\n\tscript, ok := v8.Scripts[testScriptID]\n\tif !ok {\n\t\treturn fmt.Errorf(\"test script not found: %s (id: %s)\", scriptInfo.TestPath, testScriptID)\n\t}\n\n\t// Create a new script context\n\tscriptCtx, err := script.NewContext(\"\", nil)\n\tif err != nil {\n\t\treturn fmt.Errorf(\"failed to create script context: %w\", err)\n\t}\n\tdefer scriptCtx.Close()\n\n\t// Get the V8 context\n\tv8ctx := scriptCtx.Context\n\n\t// Set share data with authorized info for Process calls\n\t// This is needed because we call fn.Call directly instead of scriptCtx.Call\n\tvar authorized map[string]interface{}\n\tif agentCtx.Authorized != nil {\n\t\tauthorized = agentCtx.Authorized.AuthorizedToMap()\n\t}\n\terr = bridge.SetShareData(v8ctx, v8ctx.Global(), &bridge.Share{\n\t\tSid: \"\",\n\t\tRoot: false,\n\t\tGlobal: nil,\n\t\tAuthorized: authorized,\n\t})\n\tif err != nil {\n\t\treturn fmt.Errorf(\"failed to set share data: %w\", err)\n\t}\n\n\t// Create testing.T JavaScript object\n\ttestingTObj, err := NewTestingTObject(v8ctx, testingT)\n\tif err != nil {\n\t\treturn fmt.Errorf(\"failed to create testing.T object: %w\", err)\n\t}\n\n\t// Create agent context JavaScript object\n\tagentCtxObj, err := agentCtx.JsValue(v8ctx)\n\tif err != nil {\n\t\treturn fmt.Errorf(\"failed to create agent context object: %w\", err)\n\t}\n\n\t// Get the test function\n\tglobal := v8ctx.Global()\n\tfnValue, err := global.Get(tc.Function)\n\tif err != nil {\n\t\treturn fmt.Errorf(\"failed to get test function %s: %w\", tc.Function, err)\n\t}\n\n\tif !fnValue.IsFunction() {\n\t\treturn fmt.Errorf(\"test function %s is not a function\", tc.Function)\n\t}\n\n\tfn, err := fnValue.AsFunction()\n\tif err != nil {\n\t\treturn fmt.Errorf(\"failed to convert to function: %w\", err)\n\t}\n\n\t// Call the test function with (t, ctx)\n\tresult, err := fn.Call(global, testingTObj, agentCtxObj)\n\tif err != nil {\n\t\t// Check if this is an assertion failure or a real error\n\t\tif testingT.Failed() {\n\t\t\t// Assertion failure - already recorded\n\t\t\treturn nil\n\t\t}\n\t\treturn fmt.Errorf(\"test function error: %w\", err)\n\t}\n\n\t// Check if the result is a JavaScript Error (thrown by bridge.JsException)\n\tif result != nil && result.IsNativeError() {\n\t\t// Get error message from Error object\n\t\tif result.IsObject() {\n\t\t\tobj, err := result.AsObject()\n\t\t\tif err == nil {\n\t\t\t\tif msgVal, err := obj.Get(\"message\"); err == nil && !msgVal.IsUndefined() {\n\t\t\t\t\treturn fmt.Errorf(\"test threw exception: %s\", msgVal.String())\n\t\t\t\t}\n\t\t\t}\n\t\t}\n\t\treturn fmt.Errorf(\"test threw exception: %s\", result.String())\n\t}\n\n\treturn nil\n}\n\n// RegisterTestingGlobals registers testing-related global functions for V8\n// This is called once during initialization\nfunc RegisterTestingGlobals() {\n\tv8.RegisterFunction(\"__testing_log\", testingLogEmbed)\n}\n\n// testingLogEmbed provides a console.log-like function for tests\nfunc testingLogEmbed(iso *v8go.Isolate) *v8go.FunctionTemplate {\n\treturn v8go.NewFunctionTemplate(iso, func(info *v8go.FunctionCallbackInfo) *v8go.Value {\n\t\targs := info.Args()\n\t\tparts := make([]string, len(args))\n\t\tfor i, arg := range args {\n\t\t\tgoVal, err := bridge.GoValue(arg, info.Context())\n\t\t\tif err != nil {\n\t\t\t\tparts[i] = arg.String()\n\t\t\t} else {\n\t\t\t\tparts[i] = fmt.Sprintf(\"%v\", goVal)\n\t\t\t}\n\t\t}\n\t\tfmt.Println(strings.Join(parts, \" \"))\n\t\treturn v8go.Undefined(iso)\n\t})\n}\n" }, { "path": "agent/test/script_assert.go", "content": "package test\n\nimport (\n\t\"encoding/json\"\n\t\"fmt\"\n\t\"reflect\"\n\t\"regexp\"\n\t\"strings\"\n\t\"sync\"\n\n\t\"github.com/yaoapp/gou/runtime/v8/bridge\"\n\t\"rogchap.com/v8go\"\n)\n\n// TestingT represents the testing object passed to test functions\n// It provides assertion methods and test control flow\ntype TestingT struct {\n\tmu sync.Mutex\n\tname string\n\tfailed bool\n\tskipped bool\n\tlogs []string\n\terrors []string\n\n\t// Assertion failure details (for the first failure)\n\tassertionInfo *ScriptAssertionInfo\n}\n\n// NewTestingT creates a new TestingT instance\nfunc NewTestingT(name string) *TestingT {\n\treturn &TestingT{\n\t\tname: name,\n\t\tlogs: make([]string, 0),\n\t\terrors: make([]string, 0),\n\t}\n}\n\n// Name returns the test name\nfunc (t *TestingT) Name() string {\n\treturn t.name\n}\n\n// Failed returns whether the test has failed\nfunc (t *TestingT) Failed() bool {\n\tt.mu.Lock()\n\tdefer t.mu.Unlock()\n\treturn t.failed\n}\n\n// Skipped returns whether the test was skipped\nfunc (t *TestingT) Skipped() bool {\n\tt.mu.Lock()\n\tdefer t.mu.Unlock()\n\treturn t.skipped\n}\n\n// Logs returns all log messages\nfunc (t *TestingT) Logs() []string {\n\tt.mu.Lock()\n\tdefer t.mu.Unlock()\n\treturn append([]string{}, t.logs...)\n}\n\n// Errors returns all error messages\nfunc (t *TestingT) Errors() []string {\n\tt.mu.Lock()\n\tdefer t.mu.Unlock()\n\treturn append([]string{}, t.errors...)\n}\n\n// AssertionInfo returns the first assertion failure info\nfunc (t *TestingT) AssertionInfo() *ScriptAssertionInfo {\n\tt.mu.Lock()\n\tdefer t.mu.Unlock()\n\treturn t.assertionInfo\n}\n\n// log adds a log message\nfunc (t *TestingT) log(msg string) {\n\tt.mu.Lock()\n\tdefer t.mu.Unlock()\n\tt.logs = append(t.logs, msg)\n}\n\n// fail marks the test as failed with an error message\nfunc (t *TestingT) fail(msg string, info *ScriptAssertionInfo) {\n\tt.mu.Lock()\n\tdefer t.mu.Unlock()\n\tt.failed = true\n\tt.errors = append(t.errors, msg)\n\tif t.assertionInfo == nil && info != nil {\n\t\tt.assertionInfo = info\n\t}\n}\n\n// skip marks the test as skipped\nfunc (t *TestingT) skip(reason string) {\n\tt.mu.Lock()\n\tdefer t.mu.Unlock()\n\tt.skipped = true\n\tif reason != \"\" {\n\t\tt.logs = append(t.logs, \"SKIP: \"+reason)\n\t}\n}\n\n// NewTestingTObject creates a JavaScript testing.T object for V8\nfunc NewTestingTObject(v8ctx *v8go.Context, t *TestingT) (*v8go.Value, error) {\n\tiso := v8ctx.Isolate()\n\n\t// Create the main testing object\n\ttestObj := v8go.NewObjectTemplate(iso)\n\n\t// Set name property\n\ttestObj.Set(\"name\", t.name)\n\n\t// Create assert object\n\tassertObj, err := newAssertObject(v8ctx, t)\n\tif err != nil {\n\t\treturn nil, err\n\t}\n\n\t// Set methods\n\ttestObj.Set(\"log\", t.logMethod(iso))\n\ttestObj.Set(\"error\", t.errorMethod(iso))\n\ttestObj.Set(\"skip\", t.skipMethod(iso))\n\ttestObj.Set(\"fail\", t.failMethod(iso))\n\ttestObj.Set(\"fatal\", t.fatalMethod(iso))\n\n\t// Create instance\n\tinstance, err := testObj.NewInstance(v8ctx)\n\tif err != nil {\n\t\treturn nil, err\n\t}\n\n\tobj, err := instance.Value.AsObject()\n\tif err != nil {\n\t\treturn nil, err\n\t}\n\n\t// Set assert object\n\tobj.Set(\"assert\", assertObj)\n\n\t// Set failed getter (dynamic)\n\tobj.Set(\"failed\", t.failed)\n\n\treturn instance.Value, nil\n}\n\n// logMethod implements t.log(...args)\nfunc (t *TestingT) logMethod(iso *v8go.Isolate) *v8go.FunctionTemplate {\n\treturn v8go.NewFunctionTemplate(iso, func(info *v8go.FunctionCallbackInfo) *v8go.Value {\n\t\targs := info.Args()\n\t\tparts := make([]string, len(args))\n\t\tfor i, arg := range args {\n\t\t\tparts[i] = arg.String()\n\t\t}\n\t\tt.log(strings.Join(parts, \" \"))\n\t\treturn v8go.Undefined(iso)\n\t})\n}\n\n// errorMethod implements t.error(...args)\nfunc (t *TestingT) errorMethod(iso *v8go.Isolate) *v8go.FunctionTemplate {\n\treturn v8go.NewFunctionTemplate(iso, func(info *v8go.FunctionCallbackInfo) *v8go.Value {\n\t\targs := info.Args()\n\t\tparts := make([]string, len(args))\n\t\tfor i, arg := range args {\n\t\t\tparts[i] = arg.String()\n\t\t}\n\t\tmsg := strings.Join(parts, \" \")\n\t\tt.fail(msg, nil)\n\t\treturn v8go.Undefined(iso)\n\t})\n}\n\n// skipMethod implements t.skip(reason?)\nfunc (t *TestingT) skipMethod(iso *v8go.Isolate) *v8go.FunctionTemplate {\n\treturn v8go.NewFunctionTemplate(iso, func(info *v8go.FunctionCallbackInfo) *v8go.Value {\n\t\treason := \"\"\n\t\tif len(info.Args()) > 0 {\n\t\t\treason = info.Args()[0].String()\n\t\t}\n\t\tt.skip(reason)\n\t\treturn v8go.Undefined(iso)\n\t})\n}\n\n// failMethod implements t.fail(reason?)\nfunc (t *TestingT) failMethod(iso *v8go.Isolate) *v8go.FunctionTemplate {\n\treturn v8go.NewFunctionTemplate(iso, func(info *v8go.FunctionCallbackInfo) *v8go.Value {\n\t\treason := \"test failed\"\n\t\tif len(info.Args()) > 0 {\n\t\t\treason = info.Args()[0].String()\n\t\t}\n\t\tt.fail(reason, nil)\n\t\treturn v8go.Undefined(iso)\n\t})\n}\n\n// fatalMethod implements t.fatal(reason?)\n// Same as fail but intended to stop execution (in JS, this is handled by throwing)\nfunc (t *TestingT) fatalMethod(iso *v8go.Isolate) *v8go.FunctionTemplate {\n\treturn v8go.NewFunctionTemplate(iso, func(info *v8go.FunctionCallbackInfo) *v8go.Value {\n\t\tv8ctx := info.Context()\n\t\treason := \"fatal error\"\n\t\tif len(info.Args()) > 0 {\n\t\t\treason = info.Args()[0].String()\n\t\t}\n\t\tt.fail(reason, nil)\n\t\t// Return exception to stop execution\n\t\treturn bridge.JsException(v8ctx, reason)\n\t})\n}\n\n// newAssertObject creates the assert object with all assertion methods\nfunc newAssertObject(v8ctx *v8go.Context, t *TestingT) (*v8go.Value, error) {\n\tiso := v8ctx.Isolate()\n\n\tassertObj := v8go.NewObjectTemplate(iso)\n\n\t// Boolean assertions\n\tassertObj.Set(\"True\", assertTrueMethod(iso, t))\n\tassertObj.Set(\"False\", assertFalseMethod(iso, t))\n\n\t// Equality assertions\n\tassertObj.Set(\"Equal\", assertEqualMethod(iso, t))\n\tassertObj.Set(\"NotEqual\", assertNotEqualMethod(iso, t))\n\n\t// Nil assertions\n\tassertObj.Set(\"Nil\", assertNilMethod(iso, t))\n\tassertObj.Set(\"NotNil\", assertNotNilMethod(iso, t))\n\n\t// String assertions\n\tassertObj.Set(\"Contains\", assertContainsMethod(iso, t))\n\tassertObj.Set(\"NotContains\", assertNotContainsMethod(iso, t))\n\n\t// Length assertion\n\tassertObj.Set(\"Len\", assertLenMethod(iso, t))\n\n\t// Comparison assertions\n\tassertObj.Set(\"Greater\", assertGreaterMethod(iso, t))\n\tassertObj.Set(\"GreaterOrEqual\", assertGreaterOrEqualMethod(iso, t))\n\tassertObj.Set(\"Less\", assertLessMethod(iso, t))\n\tassertObj.Set(\"LessOrEqual\", assertLessOrEqualMethod(iso, t))\n\n\t// Error assertions\n\tassertObj.Set(\"Error\", assertErrorMethod(iso, t))\n\tassertObj.Set(\"NoError\", assertNoErrorMethod(iso, t))\n\n\t// Panic assertions\n\tassertObj.Set(\"Panic\", assertPanicMethod(iso, t))\n\tassertObj.Set(\"NoPanic\", assertNoPanicMethod(iso, t))\n\n\t// Regex assertions\n\tassertObj.Set(\"Match\", assertMatchMethod(iso, t))\n\tassertObj.Set(\"NotMatch\", assertNotMatchMethod(iso, t))\n\n\t// Type assertion\n\tassertObj.Set(\"Type\", assertTypeMethod(iso, t))\n\n\t// JSON path assertion\n\tassertObj.Set(\"JSONPath\", assertJSONPathMethod(iso, t))\n\n\t// Agent-driven assertion\n\tassertObj.Set(\"Agent\", assertAgentMethod(iso, t))\n\n\t// Create instance\n\tinstance, err := assertObj.NewInstance(v8ctx)\n\tif err != nil {\n\t\treturn nil, err\n\t}\n\n\treturn instance.Value, nil\n}\n\n// Helper function to get optional message argument\nfunc getMessage(args []*v8go.Value, startIdx int) string {\n\tif len(args) > startIdx && args[startIdx].IsString() {\n\t\treturn args[startIdx].String()\n\t}\n\treturn \"\"\n}\n\n// assertTrueMethod implements assert.True(value, message?)\nfunc assertTrueMethod(iso *v8go.Isolate, t *TestingT) *v8go.FunctionTemplate {\n\treturn v8go.NewFunctionTemplate(iso, func(info *v8go.FunctionCallbackInfo) *v8go.Value {\n\t\targs := info.Args()\n\t\tif len(args) < 1 {\n\t\t\tt.fail(\"True requires a value argument\", &ScriptAssertionInfo{Type: \"True\"})\n\t\t\treturn v8go.Undefined(iso)\n\t\t}\n\n\t\tvalue := args[0].Boolean()\n\t\tmessage := getMessage(args, 1)\n\n\t\tif !value {\n\t\t\tmsg := \"expected true, got false\"\n\t\t\tif message != \"\" {\n\t\t\t\tmsg = message\n\t\t\t}\n\t\t\tt.fail(msg, &ScriptAssertionInfo{\n\t\t\t\tType: \"True\",\n\t\t\t\tExpected: true,\n\t\t\t\tActual: false,\n\t\t\t\tMessage: message,\n\t\t\t})\n\t\t}\n\t\treturn v8go.Undefined(iso)\n\t})\n}\n\n// assertFalseMethod implements assert.False(value, message?)\nfunc assertFalseMethod(iso *v8go.Isolate, t *TestingT) *v8go.FunctionTemplate {\n\treturn v8go.NewFunctionTemplate(iso, func(info *v8go.FunctionCallbackInfo) *v8go.Value {\n\t\targs := info.Args()\n\t\tif len(args) < 1 {\n\t\t\tt.fail(\"False requires a value argument\", &ScriptAssertionInfo{Type: \"False\"})\n\t\t\treturn v8go.Undefined(iso)\n\t\t}\n\n\t\tvalue := args[0].Boolean()\n\t\tmessage := getMessage(args, 1)\n\n\t\tif value {\n\t\t\tmsg := \"expected false, got true\"\n\t\t\tif message != \"\" {\n\t\t\t\tmsg = message\n\t\t\t}\n\t\t\tt.fail(msg, &ScriptAssertionInfo{\n\t\t\t\tType: \"False\",\n\t\t\t\tExpected: false,\n\t\t\t\tActual: true,\n\t\t\t\tMessage: message,\n\t\t\t})\n\t\t}\n\t\treturn v8go.Undefined(iso)\n\t})\n}\n\n// assertEqualMethod implements assert.Equal(actual, expected, message?)\nfunc assertEqualMethod(iso *v8go.Isolate, t *TestingT) *v8go.FunctionTemplate {\n\treturn v8go.NewFunctionTemplate(iso, func(info *v8go.FunctionCallbackInfo) *v8go.Value {\n\t\tv8ctx := info.Context()\n\t\targs := info.Args()\n\t\tif len(args) < 2 {\n\t\t\tt.fail(\"Equal requires actual and expected arguments\", &ScriptAssertionInfo{Type: \"Equal\"})\n\t\t\treturn v8go.Undefined(iso)\n\t\t}\n\n\t\tactual, _ := bridge.GoValue(args[0], v8ctx)\n\t\texpected, _ := bridge.GoValue(args[1], v8ctx)\n\t\tmessage := getMessage(args, 2)\n\n\t\tif !deepEqual(actual, expected) {\n\t\t\tmsg := fmt.Sprintf(\"expected %v, got %v\", expected, actual)\n\t\t\tif message != \"\" {\n\t\t\t\tmsg = message\n\t\t\t}\n\t\t\tt.fail(msg, &ScriptAssertionInfo{\n\t\t\t\tType: \"Equal\",\n\t\t\t\tExpected: expected,\n\t\t\t\tActual: actual,\n\t\t\t\tMessage: message,\n\t\t\t})\n\t\t}\n\t\treturn v8go.Undefined(iso)\n\t})\n}\n\n// assertNotEqualMethod implements assert.NotEqual(actual, expected, message?)\nfunc assertNotEqualMethod(iso *v8go.Isolate, t *TestingT) *v8go.FunctionTemplate {\n\treturn v8go.NewFunctionTemplate(iso, func(info *v8go.FunctionCallbackInfo) *v8go.Value {\n\t\tv8ctx := info.Context()\n\t\targs := info.Args()\n\t\tif len(args) < 2 {\n\t\t\tt.fail(\"NotEqual requires actual and expected arguments\", &ScriptAssertionInfo{Type: \"NotEqual\"})\n\t\t\treturn v8go.Undefined(iso)\n\t\t}\n\n\t\tactual, _ := bridge.GoValue(args[0], v8ctx)\n\t\texpected, _ := bridge.GoValue(args[1], v8ctx)\n\t\tmessage := getMessage(args, 2)\n\n\t\tif deepEqual(actual, expected) {\n\t\t\tmsg := fmt.Sprintf(\"expected values to be different, both are %v\", actual)\n\t\t\tif message != \"\" {\n\t\t\t\tmsg = message\n\t\t\t}\n\t\t\tt.fail(msg, &ScriptAssertionInfo{\n\t\t\t\tType: \"NotEqual\",\n\t\t\t\tExpected: expected,\n\t\t\t\tActual: actual,\n\t\t\t\tMessage: message,\n\t\t\t})\n\t\t}\n\t\treturn v8go.Undefined(iso)\n\t})\n}\n\n// assertNilMethod implements assert.Nil(value, message?)\nfunc assertNilMethod(iso *v8go.Isolate, t *TestingT) *v8go.FunctionTemplate {\n\treturn v8go.NewFunctionTemplate(iso, func(info *v8go.FunctionCallbackInfo) *v8go.Value {\n\t\targs := info.Args()\n\t\tif len(args) < 1 {\n\t\t\tt.fail(\"Nil requires a value argument\", &ScriptAssertionInfo{Type: \"Nil\"})\n\t\t\treturn v8go.Undefined(iso)\n\t\t}\n\n\t\tisNil := args[0].IsNull() || args[0].IsUndefined()\n\t\tmessage := getMessage(args, 1)\n\n\t\tif !isNil {\n\t\t\tmsg := fmt.Sprintf(\"expected nil, got %v\", args[0].String())\n\t\t\tif message != \"\" {\n\t\t\t\tmsg = message\n\t\t\t}\n\t\t\tt.fail(msg, &ScriptAssertionInfo{\n\t\t\t\tType: \"Nil\",\n\t\t\t\tExpected: nil,\n\t\t\t\tActual: args[0].String(),\n\t\t\t\tMessage: message,\n\t\t\t})\n\t\t}\n\t\treturn v8go.Undefined(iso)\n\t})\n}\n\n// assertNotNilMethod implements assert.NotNil(value, message?)\nfunc assertNotNilMethod(iso *v8go.Isolate, t *TestingT) *v8go.FunctionTemplate {\n\treturn v8go.NewFunctionTemplate(iso, func(info *v8go.FunctionCallbackInfo) *v8go.Value {\n\t\targs := info.Args()\n\t\tif len(args) < 1 {\n\t\t\tt.fail(\"NotNil requires a value argument\", &ScriptAssertionInfo{Type: \"NotNil\"})\n\t\t\treturn v8go.Undefined(iso)\n\t\t}\n\n\t\tisNil := args[0].IsNull() || args[0].IsUndefined()\n\t\tmessage := getMessage(args, 1)\n\n\t\tif isNil {\n\t\t\tmsg := \"expected non-nil value, got nil\"\n\t\t\tif message != \"\" {\n\t\t\t\tmsg = message\n\t\t\t}\n\t\t\tt.fail(msg, &ScriptAssertionInfo{\n\t\t\t\tType: \"NotNil\",\n\t\t\t\tActual: nil,\n\t\t\t\tMessage: message,\n\t\t\t})\n\t\t}\n\t\treturn v8go.Undefined(iso)\n\t})\n}\n\n// assertContainsMethod implements assert.Contains(str, substr, message?)\nfunc assertContainsMethod(iso *v8go.Isolate, t *TestingT) *v8go.FunctionTemplate {\n\treturn v8go.NewFunctionTemplate(iso, func(info *v8go.FunctionCallbackInfo) *v8go.Value {\n\t\targs := info.Args()\n\t\tif len(args) < 2 {\n\t\t\tt.fail(\"Contains requires str and substr arguments\", &ScriptAssertionInfo{Type: \"Contains\"})\n\t\t\treturn v8go.Undefined(iso)\n\t\t}\n\n\t\tstr := args[0].String()\n\t\tsubstr := args[1].String()\n\t\tmessage := getMessage(args, 2)\n\n\t\tif !strings.Contains(str, substr) {\n\t\t\tmsg := fmt.Sprintf(\"expected '%s' to contain '%s'\", str, substr)\n\t\t\tif message != \"\" {\n\t\t\t\tmsg = message\n\t\t\t}\n\t\t\tt.fail(msg, &ScriptAssertionInfo{\n\t\t\t\tType: \"Contains\",\n\t\t\t\tExpected: substr,\n\t\t\t\tActual: str,\n\t\t\t\tMessage: message,\n\t\t\t})\n\t\t}\n\t\treturn v8go.Undefined(iso)\n\t})\n}\n\n// assertNotContainsMethod implements assert.NotContains(str, substr, message?)\nfunc assertNotContainsMethod(iso *v8go.Isolate, t *TestingT) *v8go.FunctionTemplate {\n\treturn v8go.NewFunctionTemplate(iso, func(info *v8go.FunctionCallbackInfo) *v8go.Value {\n\t\targs := info.Args()\n\t\tif len(args) < 2 {\n\t\t\tt.fail(\"NotContains requires str and substr arguments\", &ScriptAssertionInfo{Type: \"NotContains\"})\n\t\t\treturn v8go.Undefined(iso)\n\t\t}\n\n\t\tstr := args[0].String()\n\t\tsubstr := args[1].String()\n\t\tmessage := getMessage(args, 2)\n\n\t\tif strings.Contains(str, substr) {\n\t\t\tmsg := fmt.Sprintf(\"expected '%s' to not contain '%s'\", str, substr)\n\t\t\tif message != \"\" {\n\t\t\t\tmsg = message\n\t\t\t}\n\t\t\tt.fail(msg, &ScriptAssertionInfo{\n\t\t\t\tType: \"NotContains\",\n\t\t\t\tExpected: substr,\n\t\t\t\tActual: str,\n\t\t\t\tMessage: message,\n\t\t\t})\n\t\t}\n\t\treturn v8go.Undefined(iso)\n\t})\n}\n\n// assertLenMethod implements assert.Len(value, length, message?)\nfunc assertLenMethod(iso *v8go.Isolate, t *TestingT) *v8go.FunctionTemplate {\n\treturn v8go.NewFunctionTemplate(iso, func(info *v8go.FunctionCallbackInfo) *v8go.Value {\n\t\tv8ctx := info.Context()\n\t\targs := info.Args()\n\t\tif len(args) < 2 {\n\t\t\tt.fail(\"Len requires value and length arguments\", &ScriptAssertionInfo{Type: \"Len\"})\n\t\t\treturn v8go.Undefined(iso)\n\t\t}\n\n\t\tvalue, _ := bridge.GoValue(args[0], v8ctx)\n\t\texpectedLen := int(args[1].Integer())\n\t\tmessage := getMessage(args, 2)\n\n\t\tactualLen := getLength(value)\n\n\t\tif actualLen != expectedLen {\n\t\t\tmsg := fmt.Sprintf(\"expected length %d, got %d\", expectedLen, actualLen)\n\t\t\tif message != \"\" {\n\t\t\t\tmsg = message\n\t\t\t}\n\t\t\tt.fail(msg, &ScriptAssertionInfo{\n\t\t\t\tType: \"Len\",\n\t\t\t\tExpected: expectedLen,\n\t\t\t\tActual: actualLen,\n\t\t\t\tMessage: message,\n\t\t\t})\n\t\t}\n\t\treturn v8go.Undefined(iso)\n\t})\n}\n\n// assertGreaterMethod implements assert.Greater(a, b, message?)\nfunc assertGreaterMethod(iso *v8go.Isolate, t *TestingT) *v8go.FunctionTemplate {\n\treturn v8go.NewFunctionTemplate(iso, func(info *v8go.FunctionCallbackInfo) *v8go.Value {\n\t\targs := info.Args()\n\t\tif len(args) < 2 {\n\t\t\tt.fail(\"Greater requires two arguments\", &ScriptAssertionInfo{Type: \"Greater\"})\n\t\t\treturn v8go.Undefined(iso)\n\t\t}\n\n\t\ta := args[0].Number()\n\t\tb := args[1].Number()\n\t\tmessage := getMessage(args, 2)\n\n\t\tif !(a > b) {\n\t\t\tmsg := fmt.Sprintf(\"expected %v > %v\", a, b)\n\t\t\tif message != \"\" {\n\t\t\t\tmsg = message\n\t\t\t}\n\t\t\tt.fail(msg, &ScriptAssertionInfo{\n\t\t\t\tType: \"Greater\",\n\t\t\t\tExpected: fmt.Sprintf(\"> %v\", b),\n\t\t\t\tActual: a,\n\t\t\t\tMessage: message,\n\t\t\t})\n\t\t}\n\t\treturn v8go.Undefined(iso)\n\t})\n}\n\n// assertGreaterOrEqualMethod implements assert.GreaterOrEqual(a, b, message?)\nfunc assertGreaterOrEqualMethod(iso *v8go.Isolate, t *TestingT) *v8go.FunctionTemplate {\n\treturn v8go.NewFunctionTemplate(iso, func(info *v8go.FunctionCallbackInfo) *v8go.Value {\n\t\targs := info.Args()\n\t\tif len(args) < 2 {\n\t\t\tt.fail(\"GreaterOrEqual requires two arguments\", &ScriptAssertionInfo{Type: \"GreaterOrEqual\"})\n\t\t\treturn v8go.Undefined(iso)\n\t\t}\n\n\t\ta := args[0].Number()\n\t\tb := args[1].Number()\n\t\tmessage := getMessage(args, 2)\n\n\t\tif !(a >= b) {\n\t\t\tmsg := fmt.Sprintf(\"expected %v >= %v\", a, b)\n\t\t\tif message != \"\" {\n\t\t\t\tmsg = message\n\t\t\t}\n\t\t\tt.fail(msg, &ScriptAssertionInfo{\n\t\t\t\tType: \"GreaterOrEqual\",\n\t\t\t\tExpected: fmt.Sprintf(\">= %v\", b),\n\t\t\t\tActual: a,\n\t\t\t\tMessage: message,\n\t\t\t})\n\t\t}\n\t\treturn v8go.Undefined(iso)\n\t})\n}\n\n// assertLessMethod implements assert.Less(a, b, message?)\nfunc assertLessMethod(iso *v8go.Isolate, t *TestingT) *v8go.FunctionTemplate {\n\treturn v8go.NewFunctionTemplate(iso, func(info *v8go.FunctionCallbackInfo) *v8go.Value {\n\t\targs := info.Args()\n\t\tif len(args) < 2 {\n\t\t\tt.fail(\"Less requires two arguments\", &ScriptAssertionInfo{Type: \"Less\"})\n\t\t\treturn v8go.Undefined(iso)\n\t\t}\n\n\t\ta := args[0].Number()\n\t\tb := args[1].Number()\n\t\tmessage := getMessage(args, 2)\n\n\t\tif !(a < b) {\n\t\t\tmsg := fmt.Sprintf(\"expected %v < %v\", a, b)\n\t\t\tif message != \"\" {\n\t\t\t\tmsg = message\n\t\t\t}\n\t\t\tt.fail(msg, &ScriptAssertionInfo{\n\t\t\t\tType: \"Less\",\n\t\t\t\tExpected: fmt.Sprintf(\"< %v\", b),\n\t\t\t\tActual: a,\n\t\t\t\tMessage: message,\n\t\t\t})\n\t\t}\n\t\treturn v8go.Undefined(iso)\n\t})\n}\n\n// assertLessOrEqualMethod implements assert.LessOrEqual(a, b, message?)\nfunc assertLessOrEqualMethod(iso *v8go.Isolate, t *TestingT) *v8go.FunctionTemplate {\n\treturn v8go.NewFunctionTemplate(iso, func(info *v8go.FunctionCallbackInfo) *v8go.Value {\n\t\targs := info.Args()\n\t\tif len(args) < 2 {\n\t\t\tt.fail(\"LessOrEqual requires two arguments\", &ScriptAssertionInfo{Type: \"LessOrEqual\"})\n\t\t\treturn v8go.Undefined(iso)\n\t\t}\n\n\t\ta := args[0].Number()\n\t\tb := args[1].Number()\n\t\tmessage := getMessage(args, 2)\n\n\t\tif !(a <= b) {\n\t\t\tmsg := fmt.Sprintf(\"expected %v <= %v\", a, b)\n\t\t\tif message != \"\" {\n\t\t\t\tmsg = message\n\t\t\t}\n\t\t\tt.fail(msg, &ScriptAssertionInfo{\n\t\t\t\tType: \"LessOrEqual\",\n\t\t\t\tExpected: fmt.Sprintf(\"<= %v\", b),\n\t\t\t\tActual: a,\n\t\t\t\tMessage: message,\n\t\t\t})\n\t\t}\n\t\treturn v8go.Undefined(iso)\n\t})\n}\n\n// assertErrorMethod implements assert.Error(err, message?)\nfunc assertErrorMethod(iso *v8go.Isolate, t *TestingT) *v8go.FunctionTemplate {\n\treturn v8go.NewFunctionTemplate(iso, func(info *v8go.FunctionCallbackInfo) *v8go.Value {\n\t\targs := info.Args()\n\t\tif len(args) < 1 {\n\t\t\tt.fail(\"Error requires an argument\", &ScriptAssertionInfo{Type: \"Error\"})\n\t\t\treturn v8go.Undefined(iso)\n\t\t}\n\n\t\t// Check if it's null/undefined (no error)\n\t\tisError := !args[0].IsNull() && !args[0].IsUndefined()\n\t\tmessage := getMessage(args, 1)\n\n\t\tif !isError {\n\t\t\tmsg := \"expected an error, got nil\"\n\t\t\tif message != \"\" {\n\t\t\t\tmsg = message\n\t\t\t}\n\t\t\tt.fail(msg, &ScriptAssertionInfo{\n\t\t\t\tType: \"Error\",\n\t\t\t\tActual: nil,\n\t\t\t\tMessage: message,\n\t\t\t})\n\t\t}\n\t\treturn v8go.Undefined(iso)\n\t})\n}\n\n// assertNoErrorMethod implements assert.NoError(err, message?)\nfunc assertNoErrorMethod(iso *v8go.Isolate, t *TestingT) *v8go.FunctionTemplate {\n\treturn v8go.NewFunctionTemplate(iso, func(info *v8go.FunctionCallbackInfo) *v8go.Value {\n\t\targs := info.Args()\n\t\tif len(args) < 1 {\n\t\t\tt.fail(\"NoError requires an argument\", &ScriptAssertionInfo{Type: \"NoError\"})\n\t\t\treturn v8go.Undefined(iso)\n\t\t}\n\n\t\t// Check if it's null/undefined (no error)\n\t\tisError := !args[0].IsNull() && !args[0].IsUndefined()\n\t\tmessage := getMessage(args, 1)\n\n\t\tif isError {\n\t\t\tmsg := fmt.Sprintf(\"expected no error, got %v\", args[0].String())\n\t\t\tif message != \"\" {\n\t\t\t\tmsg = message\n\t\t\t}\n\t\t\tt.fail(msg, &ScriptAssertionInfo{\n\t\t\t\tType: \"NoError\",\n\t\t\t\tActual: args[0].String(),\n\t\t\t\tMessage: message,\n\t\t\t})\n\t\t}\n\t\treturn v8go.Undefined(iso)\n\t})\n}\n\n// assertPanicMethod implements assert.Panic(fn, message?)\nfunc assertPanicMethod(iso *v8go.Isolate, t *TestingT) *v8go.FunctionTemplate {\n\treturn v8go.NewFunctionTemplate(iso, func(info *v8go.FunctionCallbackInfo) *v8go.Value {\n\t\tv8ctx := info.Context()\n\t\targs := info.Args()\n\t\tif len(args) < 1 {\n\t\t\tt.fail(\"Panic requires a function argument\", &ScriptAssertionInfo{Type: \"Panic\"})\n\t\t\treturn v8go.Undefined(iso)\n\t\t}\n\n\t\tif !args[0].IsFunction() {\n\t\t\tt.fail(\"Panic requires a function argument\", &ScriptAssertionInfo{Type: \"Panic\"})\n\t\t\treturn v8go.Undefined(iso)\n\t\t}\n\n\t\tmessage := getMessage(args, 1)\n\n\t\t// Try to call the function and check if it throws\n\t\tfn, _ := args[0].AsFunction()\n\t\t_, err := fn.Call(v8ctx.Global())\n\n\t\tif err == nil {\n\t\t\tmsg := \"expected function to panic, but it didn't\"\n\t\t\tif message != \"\" {\n\t\t\t\tmsg = message\n\t\t\t}\n\t\t\tt.fail(msg, &ScriptAssertionInfo{\n\t\t\t\tType: \"Panic\",\n\t\t\t\tMessage: message,\n\t\t\t})\n\t\t}\n\t\treturn v8go.Undefined(iso)\n\t})\n}\n\n// assertNoPanicMethod implements assert.NoPanic(fn, message?)\nfunc assertNoPanicMethod(iso *v8go.Isolate, t *TestingT) *v8go.FunctionTemplate {\n\treturn v8go.NewFunctionTemplate(iso, func(info *v8go.FunctionCallbackInfo) *v8go.Value {\n\t\tv8ctx := info.Context()\n\t\targs := info.Args()\n\t\tif len(args) < 1 {\n\t\t\tt.fail(\"NoPanic requires a function argument\", &ScriptAssertionInfo{Type: \"NoPanic\"})\n\t\t\treturn v8go.Undefined(iso)\n\t\t}\n\n\t\tif !args[0].IsFunction() {\n\t\t\tt.fail(\"NoPanic requires a function argument\", &ScriptAssertionInfo{Type: \"NoPanic\"})\n\t\t\treturn v8go.Undefined(iso)\n\t\t}\n\n\t\tmessage := getMessage(args, 1)\n\n\t\t// Try to call the function and check if it throws\n\t\tfn, _ := args[0].AsFunction()\n\t\t_, err := fn.Call(v8ctx.Global())\n\n\t\tif err != nil {\n\t\t\tmsg := fmt.Sprintf(\"expected function not to panic, but got: %v\", err)\n\t\t\tif message != \"\" {\n\t\t\t\tmsg = message\n\t\t\t}\n\t\t\tt.fail(msg, &ScriptAssertionInfo{\n\t\t\t\tType: \"NoPanic\",\n\t\t\t\tActual: err.Error(),\n\t\t\t\tMessage: message,\n\t\t\t})\n\t\t}\n\t\treturn v8go.Undefined(iso)\n\t})\n}\n\n// assertMatchMethod implements assert.Match(value, pattern, message?)\nfunc assertMatchMethod(iso *v8go.Isolate, t *TestingT) *v8go.FunctionTemplate {\n\treturn v8go.NewFunctionTemplate(iso, func(info *v8go.FunctionCallbackInfo) *v8go.Value {\n\t\targs := info.Args()\n\t\tif len(args) < 2 {\n\t\t\tt.fail(\"Match requires value and pattern arguments\", &ScriptAssertionInfo{Type: \"Match\"})\n\t\t\treturn v8go.Undefined(iso)\n\t\t}\n\n\t\tvalue := args[0].String()\n\t\tpattern := args[1].String()\n\t\tmessage := getMessage(args, 2)\n\n\t\tre, err := regexp.Compile(pattern)\n\t\tif err != nil {\n\t\t\tt.fail(fmt.Sprintf(\"invalid regex pattern: %v\", err), &ScriptAssertionInfo{\n\t\t\t\tType: \"Match\",\n\t\t\t\tMessage: message,\n\t\t\t})\n\t\t\treturn v8go.Undefined(iso)\n\t\t}\n\n\t\tif !re.MatchString(value) {\n\t\t\tmsg := fmt.Sprintf(\"expected '%s' to match pattern '%s'\", value, pattern)\n\t\t\tif message != \"\" {\n\t\t\t\tmsg = message\n\t\t\t}\n\t\t\tt.fail(msg, &ScriptAssertionInfo{\n\t\t\t\tType: \"Match\",\n\t\t\t\tExpected: pattern,\n\t\t\t\tActual: value,\n\t\t\t\tMessage: message,\n\t\t\t})\n\t\t}\n\t\treturn v8go.Undefined(iso)\n\t})\n}\n\n// assertNotMatchMethod implements assert.NotMatch(value, pattern, message?)\nfunc assertNotMatchMethod(iso *v8go.Isolate, t *TestingT) *v8go.FunctionTemplate {\n\treturn v8go.NewFunctionTemplate(iso, func(info *v8go.FunctionCallbackInfo) *v8go.Value {\n\t\targs := info.Args()\n\t\tif len(args) < 2 {\n\t\t\tt.fail(\"NotMatch requires value and pattern arguments\", &ScriptAssertionInfo{Type: \"NotMatch\"})\n\t\t\treturn v8go.Undefined(iso)\n\t\t}\n\n\t\tvalue := args[0].String()\n\t\tpattern := args[1].String()\n\t\tmessage := getMessage(args, 2)\n\n\t\tre, err := regexp.Compile(pattern)\n\t\tif err != nil {\n\t\t\tt.fail(fmt.Sprintf(\"invalid regex pattern: %v\", err), &ScriptAssertionInfo{\n\t\t\t\tType: \"NotMatch\",\n\t\t\t\tMessage: message,\n\t\t\t})\n\t\t\treturn v8go.Undefined(iso)\n\t\t}\n\n\t\tif re.MatchString(value) {\n\t\t\tmsg := fmt.Sprintf(\"expected '%s' to not match pattern '%s'\", value, pattern)\n\t\t\tif message != \"\" {\n\t\t\t\tmsg = message\n\t\t\t}\n\t\t\tt.fail(msg, &ScriptAssertionInfo{\n\t\t\t\tType: \"NotMatch\",\n\t\t\t\tExpected: pattern,\n\t\t\t\tActual: value,\n\t\t\t\tMessage: message,\n\t\t\t})\n\t\t}\n\t\treturn v8go.Undefined(iso)\n\t})\n}\n\n// assertTypeMethod implements assert.Type(value, typeName, message?)\nfunc assertTypeMethod(iso *v8go.Isolate, t *TestingT) *v8go.FunctionTemplate {\n\treturn v8go.NewFunctionTemplate(iso, func(info *v8go.FunctionCallbackInfo) *v8go.Value {\n\t\targs := info.Args()\n\t\tif len(args) < 2 {\n\t\t\tt.fail(\"Type requires value and typeName arguments\", &ScriptAssertionInfo{Type: \"Type\"})\n\t\t\treturn v8go.Undefined(iso)\n\t\t}\n\n\t\tvalue := args[0]\n\t\texpectedType := args[1].String()\n\t\tmessage := getMessage(args, 2)\n\n\t\tactualType := getJsType(value)\n\n\t\tif actualType != expectedType {\n\t\t\tmsg := fmt.Sprintf(\"expected type '%s', got '%s'\", expectedType, actualType)\n\t\t\tif message != \"\" {\n\t\t\t\tmsg = message\n\t\t\t}\n\t\t\tt.fail(msg, &ScriptAssertionInfo{\n\t\t\t\tType: \"Type\",\n\t\t\t\tExpected: expectedType,\n\t\t\t\tActual: actualType,\n\t\t\t\tMessage: message,\n\t\t\t})\n\t\t}\n\t\treturn v8go.Undefined(iso)\n\t})\n}\n\n// assertJSONPathMethod implements assert.JSONPath(obj, path, expected, message?)\nfunc assertJSONPathMethod(iso *v8go.Isolate, t *TestingT) *v8go.FunctionTemplate {\n\treturn v8go.NewFunctionTemplate(iso, func(info *v8go.FunctionCallbackInfo) *v8go.Value {\n\t\tv8ctx := info.Context()\n\t\targs := info.Args()\n\t\tif len(args) < 3 {\n\t\t\tt.fail(\"JSONPath requires obj, path, and expected arguments\", &ScriptAssertionInfo{Type: \"JSONPath\"})\n\t\t\treturn v8go.Undefined(iso)\n\t\t}\n\n\t\tobj, _ := bridge.GoValue(args[0], v8ctx)\n\t\tpath := args[1].String()\n\t\texpected, _ := bridge.GoValue(args[2], v8ctx)\n\t\tmessage := getMessage(args, 3)\n\n\t\t// Use the existing extractPath function from assert.go\n\t\tasserter := &Asserter{}\n\t\tactual := asserter.extractPath(obj, strings.TrimPrefix(path, \"$.\"))\n\n\t\tif !deepEqual(actual, expected) {\n\t\t\tmsg := fmt.Sprintf(\"path '%s': expected %v, got %v\", path, expected, actual)\n\t\t\tif message != \"\" {\n\t\t\t\tmsg = message\n\t\t\t}\n\t\t\tt.fail(msg, &ScriptAssertionInfo{\n\t\t\t\tType: \"JSONPath\",\n\t\t\t\tExpected: expected,\n\t\t\t\tActual: actual,\n\t\t\t\tMessage: message,\n\t\t\t})\n\t\t}\n\t\treturn v8go.Undefined(iso)\n\t})\n}\n\n// assertAgentMethod implements assert.Agent(response, agentID, options?)\n// Uses a validator agent to check the response\n// agentID is the direct agent ID (no \"agents:\" prefix needed)\nfunc assertAgentMethod(iso *v8go.Isolate, t *TestingT) *v8go.FunctionTemplate {\n\treturn v8go.NewFunctionTemplate(iso, func(info *v8go.FunctionCallbackInfo) *v8go.Value {\n\t\tv8ctx := info.Context()\n\t\targs := info.Args()\n\t\tif len(args) < 2 {\n\t\t\tt.fail(\"Agent requires response and agentID arguments\", &ScriptAssertionInfo{Type: \"Agent\"})\n\t\t\treturn v8go.Undefined(iso)\n\t\t}\n\n\t\tresponse, _ := bridge.GoValue(args[0], v8ctx)\n\t\tagentID := args[1].String()\n\n\t\t// Get options if provided\n\t\tvar options map[string]interface{}\n\t\tif len(args) > 2 && args[2].IsObject() {\n\t\t\toptVal, _ := bridge.GoValue(args[2], v8ctx)\n\t\t\toptions, _ = optVal.(map[string]interface{})\n\t\t}\n\n\t\t// Build assertion with agents: prefix\n\t\tassertion := &Assertion{\n\t\t\tType: \"agent\",\n\t\t\tUse: \"agents:\" + agentID,\n\t\t}\n\n\t\t// Extract criteria and metadata from options\n\t\tif options != nil {\n\t\t\tif criteria, ok := options[\"criteria\"]; ok {\n\t\t\t\tassertion.Value = criteria\n\t\t\t}\n\t\t\tif metadata, ok := options[\"metadata\"].(map[string]interface{}); ok {\n\t\t\t\tassertion.Options = &AssertionOptions{Metadata: metadata}\n\t\t\t}\n\t\t\tif connector, ok := options[\"connector\"].(string); ok {\n\t\t\t\tif assertion.Options == nil {\n\t\t\t\t\tassertion.Options = &AssertionOptions{}\n\t\t\t\t}\n\t\t\t\tassertion.Options.Connector = connector\n\t\t\t}\n\t\t}\n\n\t\t// Use the asserter to validate\n\t\tasserter := &Asserter{}\n\t\tresult := asserter.assertAgent(assertion, response, nil)\n\n\t\tif !result.Passed {\n\t\t\tmsg := result.Message\n\t\t\tif msg == \"\" {\n\t\t\t\tmsg = \"agent assertion failed\"\n\t\t\t}\n\t\t\tt.fail(msg, &ScriptAssertionInfo{\n\t\t\t\tType: \"Agent\",\n\t\t\t\tActual: response,\n\t\t\t\tMessage: msg,\n\t\t\t})\n\t\t}\n\n\t\treturn v8go.Undefined(iso)\n\t})\n}\n\n// Helper functions\n\n// deepEqual performs deep equality comparison\nfunc deepEqual(a, b interface{}) bool {\n\t// Handle nil cases\n\tif a == nil && b == nil {\n\t\treturn true\n\t}\n\tif a == nil || b == nil {\n\t\treturn false\n\t}\n\n\t// Try JSON comparison for complex types\n\taJSON, errA := json.Marshal(a)\n\tbJSON, errB := json.Marshal(b)\n\tif errA == nil && errB == nil {\n\t\treturn string(aJSON) == string(bJSON)\n\t}\n\n\t// Fall back to reflect.DeepEqual\n\treturn reflect.DeepEqual(a, b)\n}\n\n// getLength returns the length of a value (array, string, map)\nfunc getLength(v interface{}) int {\n\tif v == nil {\n\t\treturn 0\n\t}\n\n\tswitch val := v.(type) {\n\tcase string:\n\t\treturn len(val)\n\tcase []interface{}:\n\t\treturn len(val)\n\tcase map[string]interface{}:\n\t\treturn len(val)\n\tdefault:\n\t\trv := reflect.ValueOf(v)\n\t\tswitch rv.Kind() {\n\t\tcase reflect.Slice, reflect.Array, reflect.Map, reflect.String:\n\t\t\treturn rv.Len()\n\t\t}\n\t}\n\treturn 0\n}\n\n// getJsType returns the JavaScript type name of a value\nfunc getJsType(v *v8go.Value) string {\n\tif v.IsNull() {\n\t\treturn \"null\"\n\t}\n\tif v.IsUndefined() {\n\t\treturn \"undefined\"\n\t}\n\tif v.IsString() {\n\t\treturn \"string\"\n\t}\n\tif v.IsNumber() {\n\t\treturn \"number\"\n\t}\n\tif v.IsBoolean() {\n\t\treturn \"boolean\"\n\t}\n\tif v.IsArray() {\n\t\treturn \"array\"\n\t}\n\tif v.IsFunction() {\n\t\treturn \"function\"\n\t}\n\tif v.IsObject() {\n\t\treturn \"object\"\n\t}\n\treturn \"unknown\"\n}\n" }, { "path": "agent/test/script_hooks.go", "content": "package test\n\nimport (\n\t\"fmt\"\n\t\"path/filepath\"\n\t\"strings\"\n\n\t\"github.com/yaoapp/gou/application\"\n\tv8 \"github.com/yaoapp/gou/runtime/v8\"\n\t\"github.com/yaoapp/gou/runtime/v8/bridge\"\n\t\"github.com/yaoapp/yao/agent/context\"\n\t\"rogchap.com/v8go\"\n)\n\n// HookExecutor executes before/after scripts from *_test.ts files\n// Scripts are loaded via V8 and executed directly, not via Process()\ntype HookExecutor struct {\n\tverbose bool\n\toutput *OutputWriter\n\tloadedDirs map[string]bool // Track which directories have been loaded\n\tagentContext *context.Context\n\topts *Options // Test options (includes ContextData from --ctx)\n}\n\n// NewHookExecutor creates a new hook executor\nfunc NewHookExecutor(verbose bool) *HookExecutor {\n\treturn &HookExecutor{\n\t\tverbose: verbose,\n\t\toutput: NewOutputWriter(verbose),\n\t\tloadedDirs: make(map[string]bool),\n\t}\n}\n\n// SetAgentContext sets the agent context for script execution\nfunc (h *HookExecutor) SetAgentContext(ctx *context.Context) {\n\th.agentContext = ctx\n}\n\n// SetOptions sets the test options for hook execution\nfunc (h *HookExecutor) SetOptions(opts *Options) {\n\th.opts = opts\n}\n\n// HookRef represents a parsed hook reference\n// Format: \"src/env_test.ts:Before\" or just \"Before\" (uses default test file)\ntype HookRef struct {\n\tScriptFile string // e.g., \"env_test.ts\"\n\tFunction string // e.g., \"Before\"\n}\n\n// ParseHookRef parses a hook reference string\n// Formats:\n// - \"Before\" -> uses first *_test.ts file found\n// - \"env_test.Before\" -> uses src/env_test.ts\n// - \"src/env_test.Before\" -> uses src/env_test.ts\nfunc ParseHookRef(ref string) (*HookRef, error) {\n\tif ref == \"\" {\n\t\treturn nil, fmt.Errorf(\"empty hook reference\")\n\t}\n\n\t// Split by last dot to get function name\n\tlastDot := strings.LastIndex(ref, \".\")\n\tif lastDot == -1 {\n\t\t// Just function name, will use default test file\n\t\treturn &HookRef{\n\t\t\tScriptFile: \"\", // Will be resolved later\n\t\t\tFunction: ref,\n\t\t}, nil\n\t}\n\n\tscriptPart := ref[:lastDot]\n\tfuncName := ref[lastDot+1:]\n\n\t// Normalize script file name\n\tscriptFile := scriptPart\n\tif !strings.HasSuffix(scriptFile, \"_test\") {\n\t\tscriptFile += \"_test\"\n\t}\n\tscriptFile += \".ts\"\n\n\t// Remove \"src/\" prefix if present\n\tscriptFile = strings.TrimPrefix(scriptFile, \"src/\")\n\n\treturn &HookRef{\n\t\tScriptFile: scriptFile,\n\t\tFunction: funcName,\n\t}, nil\n}\n\n// LoadTestScripts loads all *_test.ts scripts from the agent's src directory\n// Returns the script IDs that were loaded\nfunc (h *HookExecutor) LoadTestScripts(agentPath string) ([]string, error) {\n\tsrcDir := filepath.Join(agentPath, \"src\")\n\n\t// Convert to relative path for application.App\n\t// application.App expects paths relative to YAO_ROOT\n\trelSrcDir := srcDir\n\tif application.App != nil {\n\t\tif rel, err := filepath.Rel(application.App.Root(), srcDir); err == nil {\n\t\t\trelSrcDir = rel\n\t\t}\n\t}\n\n\t// Check if already loaded (use absolute path as key)\n\t// No logging for already loaded - this is normal and happens frequently\n\tif h.loadedDirs[srcDir] {\n\t\treturn nil, nil\n\t}\n\n\t// Check if src directory exists\n\texists, err := application.App.Exists(relSrcDir)\n\tif err != nil {\n\t\treturn nil, err\n\t}\n\tif !exists {\n\t\treturn nil, nil // No src directory, not an error\n\t}\n\n\tvar loadedScripts []string\n\texts := []string{\"*_test.ts\", \"*_test.js\"}\n\n\terr = application.App.Walk(relSrcDir, func(root, file string, isdir bool) error {\n\t\tif isdir {\n\t\t\treturn nil\n\t\t}\n\n\t\t// Only load *_test.ts/js files\n\t\tbase := filepath.Base(file)\n\t\tif !strings.HasSuffix(base, \"_test.ts\") && !strings.HasSuffix(base, \"_test.js\") {\n\t\t\treturn nil\n\t\t}\n\n\t\t// Generate script ID (use relative path for consistency)\n\t\tscriptID := generateHookScriptID(file, relSrcDir)\n\n\t\t// Load the script (file path from Walk is relative to App root)\n\t\t_, err := v8.Load(file, scriptID)\n\t\tif err != nil {\n\t\t\tif h.verbose {\n\t\t\t\th.output.Warning(\"Failed to load hook script %s: %v\", base, err)\n\t\t\t}\n\t\t\treturn nil // Continue loading other scripts\n\t\t}\n\n\t\tloadedScripts = append(loadedScripts, scriptID)\n\t\treturn nil\n\t}, exts...)\n\n\tif err != nil {\n\t\treturn nil, fmt.Errorf(\"failed to walk src directory: %w\", err)\n\t}\n\n\th.loadedDirs[srcDir] = true\n\n\t// Log summary only once when scripts are first loaded\n\tif h.verbose && len(loadedScripts) > 0 {\n\t\th.output.Verbose(\"Loaded %d hook scripts from %s\", len(loadedScripts), relSrcDir)\n\t}\n\n\treturn loadedScripts, nil\n}\n\n// generateHookScriptID generates a script ID for hook scripts\n// Example: assistants/test/src/env_test.ts -> hook.env_test\nfunc generateHookScriptID(filePath string, srcDir string) string {\n\tfilePath = filepath.ToSlash(filePath)\n\tsrcDir = filepath.ToSlash(srcDir)\n\n\trelPath := strings.TrimPrefix(filePath, srcDir+\"/\")\n\trelPath = strings.TrimPrefix(relPath, \"/\")\n\trelPath = strings.TrimSuffix(relPath, filepath.Ext(relPath))\n\n\treturn \"hook.\" + strings.ReplaceAll(relPath, \"/\", \".\")\n}\n\n// FindTestScript finds a loaded test script by pattern\n// If scriptFile is empty, returns the first *_test script found\nfunc (h *HookExecutor) FindTestScript(scriptFile string) (*v8.Script, string, error) {\n\tif scriptFile != \"\" {\n\t\t// Look for specific script\n\t\tscriptID := \"hook.\" + strings.TrimSuffix(scriptFile, \".ts\")\n\t\tscriptID = strings.TrimSuffix(scriptID, \".js\")\n\n\t\tif script, ok := v8.Scripts[scriptID]; ok {\n\t\t\treturn script, scriptID, nil\n\t\t}\n\t\treturn nil, \"\", fmt.Errorf(\"hook script not found: %s (id: %s)\", scriptFile, scriptID)\n\t}\n\n\t// Find first *_test script\n\tfor id, script := range v8.Scripts {\n\t\tif strings.HasPrefix(id, \"hook.\") && strings.Contains(id, \"_test\") {\n\t\t\treturn script, id, nil\n\t\t}\n\t}\n\n\treturn nil, \"\", fmt.Errorf(\"no hook test script found\")\n}\n\n// ExecuteBefore executes a Before function from a test script\nfunc (h *HookExecutor) ExecuteBefore(ref string, testCase *Case, agentPath string) (interface{}, error) {\n\thookRef, err := ParseHookRef(ref)\n\tif err != nil {\n\t\treturn nil, err\n\t}\n\n\t// Ensure scripts are loaded\n\tif _, err := h.LoadTestScripts(agentPath); err != nil {\n\t\treturn nil, fmt.Errorf(\"failed to load test scripts: %w\", err)\n\t}\n\n\t// Find the script\n\tscript, scriptID, err := h.FindTestScript(hookRef.ScriptFile)\n\tif err != nil {\n\t\treturn nil, err\n\t}\n\n\tif h.verbose {\n\t\th.output.Verbose(\"Executing %s from %s\", hookRef.Function, scriptID)\n\t}\n\n\t// Execute the function\n\treturn h.executeHookFunction(script, hookRef.Function, testCase, nil, nil)\n}\n\n// ExecuteAfter executes an After function from a test script\nfunc (h *HookExecutor) ExecuteAfter(ref string, testCase *Case, result *Result, beforeData interface{}, agentPath string) error {\n\thookRef, err := ParseHookRef(ref)\n\tif err != nil {\n\t\treturn err\n\t}\n\n\t// Ensure scripts are loaded\n\tif _, err := h.LoadTestScripts(agentPath); err != nil {\n\t\treturn fmt.Errorf(\"failed to load test scripts: %w\", err)\n\t}\n\n\t// Find the script\n\tscript, scriptID, err := h.FindTestScript(hookRef.ScriptFile)\n\tif err != nil {\n\t\treturn err\n\t}\n\n\tif h.verbose {\n\t\th.output.Verbose(\"Executing %s from %s\", hookRef.Function, scriptID)\n\t}\n\n\t// Execute the function\n\t_, err = h.executeHookFunction(script, hookRef.Function, testCase, result, beforeData)\n\treturn err\n}\n\n// ExecuteBeforeAll executes a BeforeAll function\nfunc (h *HookExecutor) ExecuteBeforeAll(ref string, testCases []*Case, agentPath string) (interface{}, error) {\n\thookRef, err := ParseHookRef(ref)\n\tif err != nil {\n\t\treturn nil, err\n\t}\n\n\t// Ensure scripts are loaded\n\tif _, err := h.LoadTestScripts(agentPath); err != nil {\n\t\treturn nil, fmt.Errorf(\"failed to load test scripts: %w\", err)\n\t}\n\n\t// Find the script\n\tscript, scriptID, err := h.FindTestScript(hookRef.ScriptFile)\n\tif err != nil {\n\t\treturn nil, err\n\t}\n\n\tif h.verbose {\n\t\th.output.Verbose(\"Executing %s from %s\", hookRef.Function, scriptID)\n\t}\n\n\t// Execute with test cases array\n\treturn h.executeHookFunctionWithCases(script, hookRef.Function, testCases)\n}\n\n// ExecuteAfterAll executes an AfterAll function\nfunc (h *HookExecutor) ExecuteAfterAll(ref string, results []*Result, beforeData interface{}, agentPath string) error {\n\thookRef, err := ParseHookRef(ref)\n\tif err != nil {\n\t\treturn err\n\t}\n\n\t// Ensure scripts are loaded\n\tif _, err := h.LoadTestScripts(agentPath); err != nil {\n\t\treturn fmt.Errorf(\"failed to load test scripts: %w\", err)\n\t}\n\n\t// Find the script\n\tscript, scriptID, err := h.FindTestScript(hookRef.ScriptFile)\n\tif err != nil {\n\t\treturn err\n\t}\n\n\tif h.verbose {\n\t\th.output.Verbose(\"Executing %s from %s\", hookRef.Function, scriptID)\n\t}\n\n\t// Execute with results array\n\t_, err = h.executeHookFunctionWithResults(script, hookRef.Function, results, beforeData)\n\treturn err\n}\n\n// executeHookFunction executes a hook function with test case context\nfunc (h *HookExecutor) executeHookFunction(script *v8.Script, funcName string, testCase *Case, result *Result, beforeData interface{}) (interface{}, error) {\n\t// Create script context\n\tscriptCtx, err := script.NewContext(\"\", nil)\n\tif err != nil {\n\t\treturn nil, fmt.Errorf(\"failed to create script context: %w\", err)\n\t}\n\tdefer scriptCtx.Close()\n\n\tv8ctx := scriptCtx.Context\n\n\t// Set share data\n\tif err := h.setShareData(v8ctx); err != nil {\n\t\treturn nil, err\n\t}\n\n\t// Get the function\n\tglobal := v8ctx.Global()\n\tfnValue, err := global.Get(funcName)\n\tif err != nil {\n\t\treturn nil, fmt.Errorf(\"failed to get function %s: %w\", funcName, err)\n\t}\n\n\tif fnValue.IsUndefined() || fnValue.IsNull() {\n\t\treturn nil, fmt.Errorf(\"function %s not defined\", funcName)\n\t}\n\n\tif !fnValue.IsFunction() {\n\t\treturn nil, fmt.Errorf(\"%s is not a function\", funcName)\n\t}\n\n\tfn, err := fnValue.AsFunction()\n\tif err != nil {\n\t\treturn nil, fmt.Errorf(\"failed to convert to function: %w\", err)\n\t}\n\n\t// Build arguments\n\targs, err := h.buildHookArgs(v8ctx, testCase, result, beforeData)\n\tif err != nil {\n\t\treturn nil, err\n\t}\n\n\t// Convert to v8go.Valuer slice for Call\n\tvaluerArgs := make([]v8go.Valuer, len(args))\n\tfor i, arg := range args {\n\t\tvaluerArgs[i] = arg\n\t}\n\n\t// Call the function\n\tjsResult, err := fn.Call(global, valuerArgs...)\n\tif err != nil {\n\t\treturn nil, fmt.Errorf(\"hook function %s failed: %w\", funcName, err)\n\t}\n\n\t// Convert result to Go value\n\tif jsResult == nil || jsResult.IsUndefined() || jsResult.IsNull() {\n\t\treturn nil, nil\n\t}\n\n\tgoResult, err := bridge.GoValue(jsResult, v8ctx)\n\tif err != nil {\n\t\treturn nil, fmt.Errorf(\"failed to convert result: %w\", err)\n\t}\n\n\t// Extract data field if present\n\tif resultMap, ok := goResult.(map[string]interface{}); ok {\n\t\tif data, exists := resultMap[\"data\"]; exists {\n\t\t\treturn data, nil\n\t\t}\n\t}\n\n\treturn goResult, nil\n}\n\n// executeHookFunctionWithCases executes BeforeAll with test cases array\nfunc (h *HookExecutor) executeHookFunctionWithCases(script *v8.Script, funcName string, testCases []*Case) (interface{}, error) {\n\tscriptCtx, err := script.NewContext(\"\", nil)\n\tif err != nil {\n\t\treturn nil, fmt.Errorf(\"failed to create script context: %w\", err)\n\t}\n\tdefer scriptCtx.Close()\n\n\tv8ctx := scriptCtx.Context\n\n\tif err := h.setShareData(v8ctx); err != nil {\n\t\treturn nil, err\n\t}\n\n\tglobal := v8ctx.Global()\n\tfnValue, err := global.Get(funcName)\n\tif err != nil {\n\t\treturn nil, fmt.Errorf(\"failed to get function %s: %w\", funcName, err)\n\t}\n\n\tif fnValue.IsUndefined() || fnValue.IsNull() {\n\t\treturn nil, fmt.Errorf(\"function %s not defined\", funcName)\n\t}\n\n\tif !fnValue.IsFunction() {\n\t\treturn nil, fmt.Errorf(\"%s is not a function\", funcName)\n\t}\n\n\tfn, err := fnValue.AsFunction()\n\tif err != nil {\n\t\treturn nil, fmt.Errorf(\"failed to convert to function: %w\", err)\n\t}\n\n\t// Build ctx argument\n\tctxJS, err := h.buildCtxArg(v8ctx)\n\tif err != nil {\n\t\treturn nil, err\n\t}\n\n\t// Convert test cases to JS array\n\tcasesJS, err := h.testCasesToJS(v8ctx, testCases)\n\tif err != nil {\n\t\treturn nil, err\n\t}\n\n\tjsResult, err := fn.Call(global, ctxJS, casesJS)\n\tif err != nil {\n\t\treturn nil, fmt.Errorf(\"hook function %s failed: %w\", funcName, err)\n\t}\n\n\tif jsResult == nil || jsResult.IsUndefined() || jsResult.IsNull() {\n\t\treturn nil, nil\n\t}\n\n\tgoResult, err := bridge.GoValue(jsResult, v8ctx)\n\tif err != nil {\n\t\treturn nil, fmt.Errorf(\"failed to convert result: %w\", err)\n\t}\n\n\tif resultMap, ok := goResult.(map[string]interface{}); ok {\n\t\tif data, exists := resultMap[\"data\"]; exists {\n\t\t\treturn data, nil\n\t\t}\n\t}\n\n\treturn goResult, nil\n}\n\n// executeHookFunctionWithResults executes AfterAll with results array\nfunc (h *HookExecutor) executeHookFunctionWithResults(script *v8.Script, funcName string, results []*Result, beforeData interface{}) (interface{}, error) {\n\tscriptCtx, err := script.NewContext(\"\", nil)\n\tif err != nil {\n\t\treturn nil, fmt.Errorf(\"failed to create script context: %w\", err)\n\t}\n\tdefer scriptCtx.Close()\n\n\tv8ctx := scriptCtx.Context\n\n\tif err := h.setShareData(v8ctx); err != nil {\n\t\treturn nil, err\n\t}\n\n\tglobal := v8ctx.Global()\n\tfnValue, err := global.Get(funcName)\n\tif err != nil {\n\t\treturn nil, fmt.Errorf(\"failed to get function %s: %w\", funcName, err)\n\t}\n\n\tif fnValue.IsUndefined() || fnValue.IsNull() {\n\t\treturn nil, fmt.Errorf(\"function %s not defined\", funcName)\n\t}\n\n\tif !fnValue.IsFunction() {\n\t\treturn nil, fmt.Errorf(\"%s is not a function\", funcName)\n\t}\n\n\tfn, err := fnValue.AsFunction()\n\tif err != nil {\n\t\treturn nil, fmt.Errorf(\"failed to convert to function: %w\", err)\n\t}\n\n\t// Build ctx argument\n\tctxJS, err := h.buildCtxArg(v8ctx)\n\tif err != nil {\n\t\treturn nil, err\n\t}\n\n\t// Convert results to JS array\n\tresultsJS, err := h.resultsToJS(v8ctx, results)\n\tif err != nil {\n\t\treturn nil, err\n\t}\n\n\t// Convert beforeData to JS\n\tbeforeDataJS, err := bridge.JsValue(v8ctx, beforeData)\n\tif err != nil {\n\t\treturn nil, fmt.Errorf(\"failed to convert beforeData: %w\", err)\n\t}\n\n\tjsResult, err := fn.Call(global, ctxJS, resultsJS, beforeDataJS)\n\tif err != nil {\n\t\treturn nil, fmt.Errorf(\"hook function %s failed: %w\", funcName, err)\n\t}\n\n\tif jsResult == nil || jsResult.IsUndefined() || jsResult.IsNull() {\n\t\treturn nil, nil\n\t}\n\n\tgoResult, err := bridge.GoValue(jsResult, v8ctx)\n\tif err != nil {\n\t\treturn nil, fmt.Errorf(\"failed to convert result: %w\", err)\n\t}\n\n\treturn goResult, nil\n}\n\n// setShareData sets the share data for script execution\nfunc (h *HookExecutor) setShareData(v8ctx *v8go.Context) error {\n\tvar authorized map[string]interface{}\n\tif h.agentContext != nil && h.agentContext.Authorized != nil {\n\t\tauthorized = h.agentContext.Authorized.AuthorizedToMap()\n\t}\n\n\treturn bridge.SetShareData(v8ctx, v8ctx.Global(), &bridge.Share{\n\t\tSid: \"\",\n\t\tRoot: false,\n\t\tGlobal: nil,\n\t\tAuthorized: authorized,\n\t})\n}\n\n// buildCtxArg builds the context argument for hook functions\nfunc (h *HookExecutor) buildCtxArg(v8ctx *v8go.Context) (*v8go.Value, error) {\n\tctxMap := map[string]interface{}{\n\t\t\"locale\": \"en\",\n\t}\n\n\t// Use ContextData from --ctx flag if available\n\tif h.opts != nil && h.opts.ContextData != nil {\n\t\tcfg := h.opts.ContextData\n\t\tif cfg.Locale != \"\" {\n\t\t\tctxMap[\"locale\"] = cfg.Locale\n\t\t}\n\t\tif cfg.Authorized != nil {\n\t\t\tauthorized := map[string]interface{}{}\n\t\t\tif cfg.Authorized.UserID != \"\" {\n\t\t\t\tauthorized[\"user_id\"] = cfg.Authorized.UserID\n\t\t\t}\n\t\t\tif cfg.Authorized.TeamID != \"\" {\n\t\t\t\tauthorized[\"team_id\"] = cfg.Authorized.TeamID\n\t\t\t}\n\t\t\tif cfg.Authorized.TenantID != \"\" {\n\t\t\t\tauthorized[\"tenant_id\"] = cfg.Authorized.TenantID\n\t\t\t}\n\t\t\tif cfg.Authorized.Sub != \"\" {\n\t\t\t\tauthorized[\"sub\"] = cfg.Authorized.Sub\n\t\t\t}\n\t\t\tctxMap[\"authorized\"] = authorized\n\t\t}\n\t\tif cfg.Metadata != nil {\n\t\t\tctxMap[\"metadata\"] = cfg.Metadata\n\t\t}\n\t}\n\n\treturn bridge.JsValue(v8ctx, ctxMap)\n}\n\n// buildHookArgs builds the arguments for a hook function call\n// Arguments order: ctx, testCase, result (for After), beforeData (for After)\nfunc (h *HookExecutor) buildHookArgs(v8ctx *v8go.Context, testCase *Case, result *Result, beforeData interface{}) ([]*v8go.Value, error) {\n\tvar args []*v8go.Value\n\n\t// Arg 1: ctx (context) - build from opts.ContextData if available\n\tctxMap := map[string]interface{}{\n\t\t\"locale\": \"en\",\n\t}\n\n\t// Use ContextData from --ctx flag if available\n\tif h.opts != nil && h.opts.ContextData != nil {\n\t\tcfg := h.opts.ContextData\n\t\tif cfg.Locale != \"\" {\n\t\t\tctxMap[\"locale\"] = cfg.Locale\n\t\t}\n\t\tif cfg.Authorized != nil {\n\t\t\tauthorized := map[string]interface{}{}\n\t\t\tif cfg.Authorized.UserID != \"\" {\n\t\t\t\tauthorized[\"user_id\"] = cfg.Authorized.UserID\n\t\t\t}\n\t\t\tif cfg.Authorized.TeamID != \"\" {\n\t\t\t\tauthorized[\"team_id\"] = cfg.Authorized.TeamID\n\t\t\t}\n\t\t\tif cfg.Authorized.TenantID != \"\" {\n\t\t\t\tauthorized[\"tenant_id\"] = cfg.Authorized.TenantID\n\t\t\t}\n\t\t\tif cfg.Authorized.Sub != \"\" {\n\t\t\t\tauthorized[\"sub\"] = cfg.Authorized.Sub\n\t\t\t}\n\t\t\tctxMap[\"authorized\"] = authorized\n\t\t}\n\t\tif cfg.Metadata != nil {\n\t\t\tctxMap[\"metadata\"] = cfg.Metadata\n\t\t}\n\t} else if testCase != nil {\n\t\t// Fallback to test case fields\n\t\tif testCase.UserID != \"\" {\n\t\t\tctxMap[\"user_id\"] = testCase.UserID\n\t\t}\n\t\tif testCase.TeamID != \"\" {\n\t\t\tctxMap[\"team_id\"] = testCase.TeamID\n\t\t}\n\t\t// Build authorized info\n\t\tauthorized := map[string]interface{}{}\n\t\tif testCase.UserID != \"\" {\n\t\t\tauthorized[\"user_id\"] = testCase.UserID\n\t\t}\n\t\tif testCase.TeamID != \"\" {\n\t\t\tauthorized[\"team_id\"] = testCase.TeamID\n\t\t}\n\t\tif len(authorized) > 0 {\n\t\t\tctxMap[\"authorized\"] = authorized\n\t\t}\n\t}\n\n\tctxJS, err := bridge.JsValue(v8ctx, ctxMap)\n\tif err != nil {\n\t\treturn nil, fmt.Errorf(\"failed to convert ctx: %w\", err)\n\t}\n\targs = append(args, ctxJS)\n\n\t// Arg 2: testCase\n\tif testCase != nil {\n\t\ttcMap := map[string]interface{}{\n\t\t\t\"id\": testCase.ID,\n\t\t\t\"input\": testCase.Input,\n\t\t}\n\t\tif testCase.Metadata != nil {\n\t\t\ttcMap[\"metadata\"] = testCase.Metadata\n\t\t}\n\t\tif testCase.Assert != nil {\n\t\t\ttcMap[\"assert\"] = testCase.Assert\n\t\t}\n\t\t// Include simulator options for dynamic tests\n\t\tif testCase.Simulator != nil {\n\t\t\ttcMap[\"simulator\"] = testCase.Simulator\n\t\t}\n\n\t\ttcJS, err := bridge.JsValue(v8ctx, tcMap)\n\t\tif err != nil {\n\t\t\treturn nil, fmt.Errorf(\"failed to convert testCase: %w\", err)\n\t\t}\n\t\targs = append(args, tcJS)\n\t} else {\n\t\t// Pass empty object if no testCase\n\t\temptyJS, _ := bridge.JsValue(v8ctx, map[string]interface{}{})\n\t\targs = append(args, emptyJS)\n\t}\n\n\t// Arg 2: result (for After)\n\tif result != nil {\n\t\tresultMap := map[string]interface{}{\n\t\t\t\"id\": result.ID,\n\t\t\t\"status\": string(result.Status),\n\t\t\t\"duration_ms\": result.DurationMs,\n\t\t}\n\t\tif result.Output != nil {\n\t\t\tresultMap[\"output\"] = result.Output\n\t\t}\n\t\tif result.Error != \"\" {\n\t\t\tresultMap[\"error\"] = result.Error\n\t\t}\n\n\t\tresultJS, err := bridge.JsValue(v8ctx, resultMap)\n\t\tif err != nil {\n\t\t\treturn nil, fmt.Errorf(\"failed to convert result: %w\", err)\n\t\t}\n\t\targs = append(args, resultJS)\n\t}\n\n\t// Arg 3: beforeData (for After)\n\tif beforeData != nil {\n\t\tbeforeDataJS, err := bridge.JsValue(v8ctx, beforeData)\n\t\tif err != nil {\n\t\t\treturn nil, fmt.Errorf(\"failed to convert beforeData: %w\", err)\n\t\t}\n\t\targs = append(args, beforeDataJS)\n\t}\n\n\treturn args, nil\n}\n\n// testCasesToJS converts test cases to a JS array\nfunc (h *HookExecutor) testCasesToJS(v8ctx *v8go.Context, testCases []*Case) (*v8go.Value, error) {\n\tcases := make([]map[string]interface{}, len(testCases))\n\tfor i, tc := range testCases {\n\t\tcases[i] = map[string]interface{}{\n\t\t\t\"id\": tc.ID,\n\t\t\t\"input\": tc.Input,\n\t\t}\n\t\tif tc.Metadata != nil {\n\t\t\tcases[i][\"metadata\"] = tc.Metadata\n\t\t}\n\t}\n\n\treturn bridge.JsValue(v8ctx, cases)\n}\n\n// resultsToJS converts results to a JS array\nfunc (h *HookExecutor) resultsToJS(v8ctx *v8go.Context, results []*Result) (*v8go.Value, error) {\n\tresultMaps := make([]map[string]interface{}, len(results))\n\tfor i, r := range results {\n\t\tresultMaps[i] = map[string]interface{}{\n\t\t\t\"id\": r.ID,\n\t\t\t\"status\": string(r.Status),\n\t\t\t\"duration_ms\": r.DurationMs,\n\t\t}\n\t\tif r.Output != nil {\n\t\t\tresultMaps[i][\"output\"] = r.Output\n\t\t}\n\t\tif r.Error != \"\" {\n\t\t\tresultMaps[i][\"error\"] = r.Error\n\t\t}\n\t}\n\n\treturn bridge.JsValue(v8ctx, resultMaps)\n}\n" }, { "path": "agent/test/script_hooks_test.go", "content": "package test_test\n\nimport (\n\t\"testing\"\n\n\t\"github.com/stretchr/testify/assert\"\n\tv8 \"github.com/yaoapp/gou/runtime/v8\"\n\tagenttest \"github.com/yaoapp/yao/agent/test\"\n\t\"github.com/yaoapp/yao/config\"\n\t\"github.com/yaoapp/yao/test\"\n)\n\nconst hooksTestAgent = \"assistants/tests/hooks-test\"\n\nfunc TestParseHookRef(t *testing.T) {\n\ttests := []struct {\n\t\tname string\n\t\tinput string\n\t\twantFile string\n\t\twantFunc string\n\t\texpectErr bool\n\t}{\n\t\t{\n\t\t\tname: \"function only\",\n\t\t\tinput: \"Before\",\n\t\t\twantFile: \"\",\n\t\t\twantFunc: \"Before\",\n\t\t},\n\t\t{\n\t\t\tname: \"with script file\",\n\t\t\tinput: \"env_test.Before\",\n\t\t\twantFile: \"env_test.ts\",\n\t\t\twantFunc: \"Before\",\n\t\t},\n\t\t{\n\t\t\tname: \"with src prefix\",\n\t\t\tinput: \"src/env_test.Before\",\n\t\t\twantFile: \"env_test.ts\",\n\t\t\twantFunc: \"Before\",\n\t\t},\n\t\t{\n\t\t\tname: \"nested path\",\n\t\t\tinput: \"setup/db_test.Before\",\n\t\t\twantFile: \"setup/db_test.ts\",\n\t\t\twantFunc: \"Before\",\n\t\t},\n\t\t{\n\t\t\tname: \"empty string\",\n\t\t\tinput: \"\",\n\t\t\texpectErr: true,\n\t\t},\n\t}\n\n\tfor _, tt := range tests {\n\t\tt.Run(tt.name, func(t *testing.T) {\n\t\t\tref, err := agenttest.ParseHookRef(tt.input)\n\t\t\tif tt.expectErr {\n\t\t\t\tassert.Error(t, err)\n\t\t\t\treturn\n\t\t\t}\n\t\t\tassert.NoError(t, err)\n\t\t\tassert.Equal(t, tt.wantFile, ref.ScriptFile)\n\t\t\tassert.Equal(t, tt.wantFunc, ref.Function)\n\t\t})\n\t}\n}\n\nfunc TestHookExecutorLoadTestScripts(t *testing.T) {\n\t// Prepare test environment\n\ttest.Prepare(t, config.Conf)\n\tdefer test.Clean()\n\n\t// Load agent test scripts using the utility function\n\tscripts := test.LoadAgentTestScripts(t, hooksTestAgent)\n\n\tassert.NotEmpty(t, scripts, \"Should load at least one test script\")\n\n\t// Verify the script was loaded into V8\n\tfound := false\n\tfor _, scriptID := range scripts {\n\t\tif _, ok := v8.Scripts[scriptID]; ok {\n\t\t\tfound = true\n\t\t\tt.Logf(\"Loaded script: %s\", scriptID)\n\t\t\tbreak\n\t\t}\n\t}\n\tassert.True(t, found, \"At least one script should be loaded into V8\")\n}\n\nfunc TestHookExecutorExecuteBefore(t *testing.T) {\n\t// Prepare test environment\n\ttest.Prepare(t, config.Conf)\n\tdefer test.Clean()\n\n\t// Load agent test scripts\n\ttest.LoadAgentTestScripts(t, hooksTestAgent)\n\n\texecutor := agenttest.NewHookExecutor(true)\n\n\ttestCase := &agenttest.Case{\n\t\tID: \"TEST001\",\n\t\tInput: \"Hello World\",\n\t}\n\n\t// Execute Before hook\n\tbeforeData, err := executor.ExecuteBefore(\"env_test.Before\", testCase, hooksTestAgent)\n\tassert.NoError(t, err)\n\tassert.NotNil(t, beforeData)\n\n\t// Verify returned data\n\tdataMap, ok := beforeData.(map[string]interface{})\n\tassert.True(t, ok, \"beforeData should be a map\")\n\tassert.Equal(t, \"TEST001\", dataMap[\"test_id\"])\n\tassert.NotEmpty(t, dataMap[\"mock_user_id\"])\n\tassert.NotEmpty(t, dataMap[\"mock_session_id\"])\n}\n\nfunc TestHookExecutorExecuteAfter(t *testing.T) {\n\t// Prepare test environment\n\ttest.Prepare(t, config.Conf)\n\tdefer test.Clean()\n\n\t// Load agent test scripts\n\ttest.LoadAgentTestScripts(t, hooksTestAgent)\n\n\texecutor := agenttest.NewHookExecutor(true)\n\n\ttestCase := &agenttest.Case{\n\t\tID: \"TEST002\",\n\t\tInput: \"Test input\",\n\t}\n\n\tresult := &agenttest.Result{\n\t\tID: \"TEST002\",\n\t\tStatus: agenttest.StatusPassed,\n\t\tDurationMs: 100,\n\t}\n\n\tbeforeData := map[string]interface{}{\n\t\t\"test_id\": \"TEST002\",\n\t\t\"mock_user_id\": \"user_TEST002_12345\",\n\t\t\"mock_session_id\": \"session_12345\",\n\t}\n\n\t// Execute After hook\n\terr := executor.ExecuteAfter(\"env_test.After\", testCase, result, beforeData, hooksTestAgent)\n\tassert.NoError(t, err)\n}\n\nfunc TestHookExecutorExecuteBeforeAll(t *testing.T) {\n\t// Prepare test environment\n\ttest.Prepare(t, config.Conf)\n\tdefer test.Clean()\n\n\t// Load agent test scripts\n\ttest.LoadAgentTestScripts(t, hooksTestAgent)\n\n\texecutor := agenttest.NewHookExecutor(true)\n\n\ttestCases := []*agenttest.Case{\n\t\t{ID: \"T001\", Input: \"Test 1\"},\n\t\t{ID: \"T002\", Input: \"Test 2\"},\n\t\t{ID: \"T003\", Input: \"Test 3\"},\n\t}\n\n\t// Execute BeforeAll hook\n\tglobalData, err := executor.ExecuteBeforeAll(\"env_test.BeforeAll\", testCases, hooksTestAgent)\n\tassert.NoError(t, err)\n\tassert.NotNil(t, globalData)\n\n\t// Verify returned data\n\tdataMap, ok := globalData.(map[string]interface{})\n\tassert.True(t, ok, \"globalData should be a map\")\n\tassert.NotEmpty(t, dataMap[\"suite_id\"])\n\tassert.Equal(t, float64(3), dataMap[\"test_count\"]) // JSON numbers are float64\n}\n\nfunc TestHookExecutorExecuteAfterAll(t *testing.T) {\n\t// Prepare test environment\n\ttest.Prepare(t, config.Conf)\n\tdefer test.Clean()\n\n\t// Load agent test scripts\n\ttest.LoadAgentTestScripts(t, hooksTestAgent)\n\n\texecutor := agenttest.NewHookExecutor(true)\n\n\tresults := []*agenttest.Result{\n\t\t{ID: \"T001\", Status: agenttest.StatusPassed, DurationMs: 100},\n\t\t{ID: \"T002\", Status: agenttest.StatusFailed, DurationMs: 200, Error: \"assertion failed\"},\n\t\t{ID: \"T003\", Status: agenttest.StatusPassed, DurationMs: 150},\n\t}\n\n\tglobalData := map[string]interface{}{\n\t\t\"suite_id\": \"suite_12345\",\n\t\t\"test_count\": 3,\n\t}\n\n\t// Execute AfterAll hook\n\terr := executor.ExecuteAfterAll(\"env_test.AfterAll\", results, globalData, hooksTestAgent)\n\tassert.NoError(t, err)\n}\n\nfunc TestHookExecutorFunctionNotFound(t *testing.T) {\n\t// Prepare test environment\n\ttest.Prepare(t, config.Conf)\n\tdefer test.Clean()\n\n\t// Load agent test scripts\n\ttest.LoadAgentTestScripts(t, hooksTestAgent)\n\n\texecutor := agenttest.NewHookExecutor(true)\n\n\ttestCase := &agenttest.Case{\n\t\tID: \"TEST001\",\n\t\tInput: \"Hello\",\n\t}\n\n\t// Try to execute non-existent function\n\t_, err := executor.ExecuteBefore(\"env_test.NonExistent\", testCase, hooksTestAgent)\n\tassert.Error(t, err)\n\tassert.Contains(t, err.Error(), \"not defined\")\n}\n\nfunc TestHookExecutorScriptNotFound(t *testing.T) {\n\t// Prepare test environment\n\ttest.Prepare(t, config.Conf)\n\tdefer test.Clean()\n\n\t// Load agent test scripts\n\ttest.LoadAgentTestScripts(t, hooksTestAgent)\n\n\texecutor := agenttest.NewHookExecutor(true)\n\n\ttestCase := &agenttest.Case{\n\t\tID: \"TEST001\",\n\t\tInput: \"Hello\",\n\t}\n\n\t// Try to execute from non-existent script\n\t_, err := executor.ExecuteBefore(\"nonexistent_test.Before\", testCase, hooksTestAgent)\n\tassert.Error(t, err)\n\tassert.Contains(t, err.Error(), \"not found\")\n}\n" }, { "path": "agent/test/script_types.go", "content": "package test\n\nimport \"time\"\n\n// ScriptInfo contains information about the script being tested\ntype ScriptInfo struct {\n\t// ID is the script identifier (e.g., \"scripts.expense.setup\")\n\tID string `json:\"id\"`\n\n\t// Assistant is the assistant directory name (e.g., \"expense\")\n\tAssistant string `json:\"assistant\"`\n\n\t// Module is the module name (e.g., \"setup\")\n\tModule string `json:\"module\"`\n\n\t// ScriptPath is the path to the main script file (e.g., \"expense/src/setup.ts\")\n\tScriptPath string `json:\"script_path\"`\n\n\t// TestPath is the path to the test script file (e.g., \"expense/src/setup_test.ts\")\n\tTestPath string `json:\"test_path\"`\n}\n\n// ScriptTestCase represents a single script test function\ntype ScriptTestCase struct {\n\t// Name is the test function name (e.g., \"TestSystemReady\")\n\tName string `json:\"name\"`\n\n\t// Function is the full function reference\n\tFunction string `json:\"function\"`\n}\n\n// ScriptTestResult represents the result of running a script test function\ntype ScriptTestResult struct {\n\t// Name is the test function name\n\tName string `json:\"name\"`\n\n\t// Status is the test execution status\n\tStatus Status `json:\"status\"`\n\n\t// DurationMs is the execution duration in milliseconds\n\tDurationMs int64 `json:\"duration_ms\"`\n\n\t// Error contains the error message if the test failed\n\tError string `json:\"error,omitempty\"`\n\n\t// Assertion contains assertion failure details\n\tAssertion *ScriptAssertionInfo `json:\"assertion,omitempty\"`\n\n\t// Logs contains log messages from the test\n\tLogs []string `json:\"logs,omitempty\"`\n}\n\n// ScriptAssertionInfo contains details about an assertion failure\ntype ScriptAssertionInfo struct {\n\t// Type is the assertion type (e.g., \"Equal\", \"True\")\n\tType string `json:\"type\"`\n\n\t// Expected is the expected value\n\tExpected interface{} `json:\"expected,omitempty\"`\n\n\t// Actual is the actual value\n\tActual interface{} `json:\"actual,omitempty\"`\n\n\t// Message is the custom failure message\n\tMessage string `json:\"message,omitempty\"`\n}\n\n// ScriptTestSummary contains aggregated statistics for script tests\ntype ScriptTestSummary struct {\n\t// Total number of test functions\n\tTotal int `json:\"total\"`\n\n\t// Passed number of test functions that passed\n\tPassed int `json:\"passed\"`\n\n\t// Failed number of test functions that failed\n\tFailed int `json:\"failed\"`\n\n\t// Skipped number of test functions that were skipped\n\tSkipped int `json:\"skipped\"`\n\n\t// DurationMs is the total execution duration in milliseconds\n\tDurationMs int64 `json:\"duration_ms\"`\n}\n\n// ScriptTestReport represents the complete script test report\ntype ScriptTestReport struct {\n\t// Type indicates this is a script test report\n\tType string `json:\"type\"` // \"script_test\"\n\n\t// Script is the script identifier (e.g., \"scripts.expense.setup\")\n\tScript string `json:\"script\"`\n\n\t// ScriptPath is the path to the test script file\n\tScriptPath string `json:\"script_path\"`\n\n\t// Summary contains aggregated statistics\n\tSummary *ScriptTestSummary `json:\"summary\"`\n\n\t// Environment contains the test environment configuration\n\tEnvironment *Environment `json:\"environment\"`\n\n\t// Results contains individual test results\n\tResults []*ScriptTestResult `json:\"results\"`\n\n\t// Metadata contains additional report metadata\n\tMetadata *ScriptTestMetadata `json:\"metadata\"`\n}\n\n// ScriptTestMetadata contains metadata about the script test report\ntype ScriptTestMetadata struct {\n\t// StartedAt is when the test run started\n\tStartedAt time.Time `json:\"started_at\"`\n\n\t// CompletedAt is when the test run completed\n\tCompletedAt time.Time `json:\"completed_at\"`\n\n\t// Version is the Yao version\n\tVersion string `json:\"version\"`\n}\n\n// HasFailures returns true if there are any failed tests\nfunc (r *ScriptTestReport) HasFailures() bool {\n\treturn r.Summary.Failed > 0\n}\n\n// PassRate returns the pass rate as a percentage (0-100)\nfunc (r *ScriptTestReport) PassRate() float64 {\n\tif r.Summary.Total == 0 {\n\t\treturn 0\n\t}\n\treturn float64(r.Summary.Passed) / float64(r.Summary.Total) * 100\n}\n\n// ToReport converts ScriptTestReport to a standard Report for unified reporting\nfunc (r *ScriptTestReport) ToReport() *Report {\n\treturn &Report{\n\t\tSummary: &Summary{\n\t\t\tTotal: r.Summary.Total,\n\t\t\tPassed: r.Summary.Passed,\n\t\t\tFailed: r.Summary.Failed,\n\t\t\tSkipped: r.Summary.Skipped,\n\t\t\tDurationMs: r.Summary.DurationMs,\n\t\t\tAgentID: r.Script,\n\t\t\tAgentPath: r.ScriptPath,\n\t\t},\n\t\tEnvironment: r.Environment,\n\t\tResults: r.toResults(),\n\t\tMetadata: &ReportMetadata{\n\t\t\tStartedAt: r.Metadata.StartedAt,\n\t\t\tCompletedAt: r.Metadata.CompletedAt,\n\t\t\tVersion: r.Metadata.Version,\n\t\t},\n\t}\n}\n\n// toResults converts script test results to standard results\nfunc (r *ScriptTestReport) toResults() []*Result {\n\tresults := make([]*Result, len(r.Results))\n\tfor i, sr := range r.Results {\n\t\tresults[i] = &Result{\n\t\t\tID: sr.Name,\n\t\t\tStatus: sr.Status,\n\t\t\tInput: sr.Name,\n\t\t\tDurationMs: sr.DurationMs,\n\t\t\tError: sr.Error,\n\t\t}\n\t}\n\treturn results\n}\n" }, { "path": "agent/test/types.go", "content": "package test\n\nimport (\n\t\"encoding/json\"\n\t\"fmt\"\n\t\"math\"\n\t\"os\"\n\t\"time\"\n\n\t\"github.com/yaoapp/yao/agent/context\"\n)\n\n// Status represents the status of a test case execution\ntype Status string\n\nconst (\n\t// StatusPassed indicates the test passed\n\tStatusPassed Status = \"passed\"\n\t// StatusFailed indicates the test failed\n\tStatusFailed Status = \"failed\"\n\t// StatusSkipped indicates the test was skipped\n\tStatusSkipped Status = \"skipped\"\n\t// StatusError indicates a runtime error occurred\n\tStatusError Status = \"error\"\n\t// StatusTimeout indicates the test timed out\n\tStatusTimeout Status = \"timeout\"\n)\n\n// OutputFormat represents the output format for test reports\ntype OutputFormat string\n\nconst (\n\t// FormatJSON outputs JSON format (for CI integration)\n\tFormatJSON OutputFormat = \"json\"\n\t// FormatHTML outputs HTML format (for human review)\n\tFormatHTML OutputFormat = \"html\"\n\t// FormatMarkdown outputs Markdown format (for documentation)\n\tFormatMarkdown OutputFormat = \"markdown\"\n)\n\n// StabilityClass represents the stability classification of a test case\ntype StabilityClass string\n\nconst (\n\t// StabilityStable indicates 100% pass rate\n\tStabilityStable StabilityClass = \"stable\"\n\t// StabilityMostlyStable indicates 80-99% pass rate\n\tStabilityMostlyStable StabilityClass = \"mostly_stable\"\n\t// StabilityUnstable indicates 50-79% pass rate\n\tStabilityUnstable StabilityClass = \"unstable\"\n\t// StabilityHighlyUnstable indicates < 50% pass rate\n\tStabilityHighlyUnstable StabilityClass = \"highly_unstable\"\n)\n\n// InputMode represents the input mode for test cases\ntype InputMode string\n\nconst (\n\t// InputModeFile indicates input from a JSONL file\n\tInputModeFile InputMode = \"file\"\n\t// InputModeMessage indicates input from a direct message string\n\tInputModeMessage InputMode = \"message\"\n\t// InputModeScript indicates script test mode (testing agent handler scripts)\n\tInputModeScript InputMode = \"script\"\n)\n\n// Options represents the configuration options for running tests\ntype Options struct {\n\t// Input/Output\n\t// ===============================\n\n\t// Input is the input source: either a file path or a direct message\n\tInput string `json:\"input\"`\n\n\t// InputMode is the input mode (auto-detected from Input)\n\tInputMode InputMode `json:\"input_mode\"`\n\n\t// OutputFile is the path to write the test report\n\t// Format is determined by file extension (.json, .html, .md)\n\tOutputFile string `json:\"output_file\"`\n\n\t// Agent Selection\n\t// ===============================\n\n\t// AgentID is the explicit agent ID to test (optional)\n\t// If not set, agent is resolved from InputFile path\n\tAgentID string `json:\"agent_id,omitempty\"`\n\n\t// Connector overrides the agent's default connector (optional)\n\tConnector string `json:\"connector,omitempty\"`\n\n\t// Test Environment\n\t// ===============================\n\n\t// UserID is the test user ID (-u flag)\n\tUserID string `json:\"user_id,omitempty\"`\n\n\t// TeamID is the test team ID (-t flag)\n\tTeamID string `json:\"team_id,omitempty\"`\n\n\t// Locale is the locale for the test context (default: \"en-us\")\n\tLocale string `json:\"locale,omitempty\"`\n\n\t// ContextFile is the path to a JSON file containing custom context data (-ctx flag)\n\t// This allows full customization of authorized info, metadata, etc.\n\tContextFile string `json:\"context_file,omitempty\"`\n\n\t// ContextData is the parsed context data from ContextFile\n\t// This is populated internally after loading the file\n\tContextData *ContextConfig `json:\"-\"`\n\n\t// Execution\n\t// ===============================\n\n\t// Timeout is the default timeout for each test case\n\t// Can be overridden per test case\n\tTimeout time.Duration `json:\"timeout,omitempty\"`\n\n\t// Parallel is the number of tests to run in parallel\n\t// Default is 1 (sequential execution)\n\tParallel int `json:\"parallel,omitempty\"`\n\n\t// Runs is the number of times to run each test case\n\t// Default is 1. When > 1, stability metrics are collected\n\tRuns int `json:\"runs,omitempty\"`\n\n\t// Reporting\n\t// ===============================\n\n\t// ReporterID is the reporter agent ID for custom report generation\n\t// If not set, default JSONL format is used\n\tReporterID string `json:\"reporter_id,omitempty\"`\n\n\t// Behavior\n\t// ===============================\n\n\t// Verbose enables verbose output during test execution\n\tVerbose bool `json:\"verbose,omitempty\"`\n\n\t// FailFast stops execution on first failure\n\tFailFast bool `json:\"fail_fast,omitempty\"`\n\n\t// Run is a regex pattern to filter which tests to run (similar to go test -run)\n\t// Only tests matching the pattern will be executed\n\t// Example: \"TestSystem\" matches TestSystemReady, TestSystemError, etc.\n\tRun string `json:\"run,omitempty\"`\n\n\t// BeforeAll is the global before script (e.g., \"scripts:tests.env.BeforeAll\")\n\t// Called once before all test cases\n\tBeforeAll string `json:\"before_all,omitempty\"`\n\n\t// AfterAll is the global after script (e.g., \"scripts:tests.env.AfterAll\")\n\t// Called once after all test cases\n\tAfterAll string `json:\"after_all,omitempty\"`\n\n\t// DryRun generates test cases without running them\n\t// Useful for previewing agent-generated test cases\n\tDryRun bool `json:\"dry_run,omitempty\"`\n\n\t// Simulator is the default simulator agent ID for dynamic mode\n\t// Can be overridden per test case in JSONL\n\tSimulator string `json:\"simulator,omitempty\"`\n}\n\n// ContextConfig represents custom context configuration from JSON file\n// This allows full customization of the test context including authorized info\ntype ContextConfig struct {\n\t// ChatID is the chat session identifier\n\t// Used to maintain session state across turns in dynamic tests\n\tChatID string `json:\"chat_id,omitempty\"`\n\n\t// Authorized contains custom authorization data\n\tAuthorized *AuthorizedConfig `json:\"authorized,omitempty\"`\n\n\t// Metadata contains custom metadata to pass to the context\n\tMetadata map[string]interface{} `json:\"metadata,omitempty\"`\n\n\t// Client contains custom client information\n\tClient *ClientConfig `json:\"client,omitempty\"`\n\n\t// Locale overrides the locale setting\n\tLocale string `json:\"locale,omitempty\"`\n\n\t// Referer overrides the referer setting\n\tReferer string `json:\"referer,omitempty\"`\n}\n\n// AuthorizedConfig represents custom authorization configuration\n// Matches the structure of types.AuthorizedInfo from openapi/oauth/types\ntype AuthorizedConfig struct {\n\t// Sub is the subject identifier (JWT sub claim)\n\tSub string `json:\"sub,omitempty\"`\n\n\t// ClientID is the OAuth client ID\n\tClientID string `json:\"client_id,omitempty\"`\n\n\t// Scope is the access scope\n\tScope string `json:\"scope,omitempty\"`\n\n\t// SessionID is the session identifier\n\tSessionID string `json:\"session_id,omitempty\"`\n\n\t// UserID is the user identifier\n\tUserID string `json:\"user_id,omitempty\"`\n\n\t// TeamID is the team identifier\n\tTeamID string `json:\"team_id,omitempty\"`\n\n\t// TenantID is the tenant identifier\n\tTenantID string `json:\"tenant_id,omitempty\"`\n\n\t// RememberMe is the remember me flag\n\tRememberMe bool `json:\"remember_me,omitempty\"`\n\n\t// Constraints contains data access constraints (set by ACL enforcement)\n\tConstraints *DataConstraintsConfig `json:\"constraints,omitempty\"`\n}\n\n// DataConstraintsConfig represents data access constraints\n// Matches the structure of types.DataConstraints from openapi/oauth/types\ntype DataConstraintsConfig struct {\n\t// OwnerOnly - only access owner's data\n\tOwnerOnly bool `json:\"owner_only,omitempty\"`\n\n\t// CreatorOnly - only access creator's data\n\tCreatorOnly bool `json:\"creator_only,omitempty\"`\n\n\t// EditorOnly - only access editor's data\n\tEditorOnly bool `json:\"editor_only,omitempty\"`\n\n\t// TeamOnly - only access team's data (filter by team_id)\n\tTeamOnly bool `json:\"team_only,omitempty\"`\n\n\t// Extra contains user-defined constraints (department, region, etc.)\n\tExtra map[string]interface{} `json:\"extra,omitempty\"`\n}\n\n// ClientConfig represents custom client configuration\ntype ClientConfig struct {\n\t// Type is the client type (e.g., \"web\", \"mobile\", \"test\")\n\tType string `json:\"type,omitempty\"`\n\n\t// UserAgent is the client user agent string\n\tUserAgent string `json:\"user_agent,omitempty\"`\n\n\t// IP is the client IP address\n\tIP string `json:\"ip,omitempty\"`\n}\n\n// Environment configures the test execution context\ntype Environment struct {\n\t// UserID is the user ID for authorized info (-u flag)\n\tUserID string `json:\"user_id\"`\n\n\t// TeamID is the team ID for authorized info (-t flag)\n\tTeamID string `json:\"team_id\"`\n\n\t// Locale is the locale (default: \"en-us\")\n\tLocale string `json:\"locale\"`\n\n\t// ClientType is the client type (default: \"test\")\n\tClientType string `json:\"client_type\"`\n\n\t// ClientIP is the client IP (default: \"127.0.0.1\")\n\tClientIP string `json:\"client_ip\"`\n\n\t// Referer is the request referer (default: \"test\")\n\tReferer string `json:\"referer\"`\n\n\t// Accept is the accept format (default: \"standard\")\n\tAccept string `json:\"accept\"`\n\n\t// ContextConfig contains custom context configuration (from -ctx flag)\n\tContextConfig *ContextConfig `json:\"-\"`\n}\n\n// NewEnvironment creates a new test environment with defaults\nfunc NewEnvironment(userID, teamID string) *Environment {\n\tenv := &Environment{\n\t\tUserID: userID,\n\t\tTeamID: teamID,\n\t\tLocale: \"en-us\",\n\t\tClientType: \"test\",\n\t\tClientIP: \"127.0.0.1\",\n\t\tReferer: \"test\",\n\t\tAccept: \"standard\",\n\t}\n\n\t// Apply defaults if not set\n\tif env.UserID == \"\" {\n\t\tenv.UserID = \"test-user\"\n\t}\n\tif env.TeamID == \"\" {\n\t\tenv.TeamID = \"test-team\"\n\t}\n\n\treturn env\n}\n\n// NewEnvironmentWithContext creates a new test environment with custom context config\nfunc NewEnvironmentWithContext(userID, teamID string, ctxConfig *ContextConfig) *Environment {\n\tenv := NewEnvironment(userID, teamID)\n\n\tif ctxConfig == nil {\n\t\treturn env\n\t}\n\n\tenv.ContextConfig = ctxConfig\n\n\t// Override with context config values\n\tif ctxConfig.Locale != \"\" {\n\t\tenv.Locale = ctxConfig.Locale\n\t}\n\tif ctxConfig.Referer != \"\" {\n\t\tenv.Referer = ctxConfig.Referer\n\t}\n\tif ctxConfig.Client != nil {\n\t\tif ctxConfig.Client.Type != \"\" {\n\t\t\tenv.ClientType = ctxConfig.Client.Type\n\t\t}\n\t\tif ctxConfig.Client.IP != \"\" {\n\t\t\tenv.ClientIP = ctxConfig.Client.IP\n\t\t}\n\t}\n\tif ctxConfig.Authorized != nil {\n\t\tif ctxConfig.Authorized.UserID != \"\" {\n\t\t\tenv.UserID = ctxConfig.Authorized.UserID\n\t\t}\n\t\t// TeamID takes precedence over TenantID for team override\n\t\tif ctxConfig.Authorized.TeamID != \"\" {\n\t\t\tenv.TeamID = ctxConfig.Authorized.TeamID\n\t\t} else if ctxConfig.Authorized.TenantID != \"\" {\n\t\t\tenv.TeamID = ctxConfig.Authorized.TenantID\n\t\t}\n\t}\n\n\treturn env\n}\n\n// LoadContextConfig loads context configuration from a JSON file\nfunc LoadContextConfig(filePath string) (*ContextConfig, error) {\n\tresolvedPath := ResolvePathWithYaoRoot(filePath)\n\tdata, err := os.ReadFile(resolvedPath)\n\tif err != nil {\n\t\treturn nil, fmt.Errorf(\"failed to read context file: %w\", err)\n\t}\n\n\tvar config ContextConfig\n\tif err := json.Unmarshal(data, &config); err != nil {\n\t\treturn nil, fmt.Errorf(\"failed to parse context file: %w\", err)\n\t}\n\n\treturn &config, nil\n}\n\n// Case represents a single test case loaded from JSONL\ntype Case struct {\n\t// ID is the unique identifier for this test case (e.g., \"T001\")\n\tID string `json:\"id\"`\n\n\t// Input is the test input, can be:\n\t// - string: simple text input\n\t// - map (Message): single message with role and content\n\t// - []map ([]Message): conversation history\n\tInput interface{} `json:\"input\"`\n\n\t// Expected is the expected output for validation (optional)\n\t// If set, the actual output will be compared against this\n\tExpected interface{} `json:\"expected,omitempty\"`\n\n\t// Assert defines custom assertion rules (optional)\n\t// If set, these rules will be used instead of simple expected comparison\n\t// Can be a single assertion or an array of assertions\n\tAssert interface{} `json:\"assert,omitempty\"`\n\n\t// Environment (per-test case, can be overridden by command line flags)\n\t// ===============================\n\n\t// UserID is the user ID for this test case (overridden by -u flag)\n\tUserID string `json:\"user,omitempty\"`\n\n\t// TeamID is the team ID for this test case (overridden by -t flag)\n\tTeamID string `json:\"team,omitempty\"`\n\n\t// Metadata contains additional metadata for the test case\n\t// This is passed to ctx.Metadata and can be used by Create Hook\n\tMetadata map[string]interface{} `json:\"metadata,omitempty\"`\n\n\t// Options contains context options for this test case\n\t// Supports: connector, skip (history, trace, output, keyword, search), mode\n\tOptions *CaseOptions `json:\"options,omitempty\"`\n\n\t// Skip indicates whether to skip this test case\n\tSkip bool `json:\"skip,omitempty\"`\n\n\t// Timeout overrides the default timeout for this test case\n\t// Format: \"30s\", \"1m\", \"2m30s\"\n\tTimeout string `json:\"timeout,omitempty\"`\n\n\t// Before script function (e.g., \"scripts:tests.env.Before\")\n\t// Called before the test case runs, returns data passed to After\n\tBefore string `json:\"before,omitempty\"`\n\n\t// After script function (e.g., \"scripts:tests.env.After\")\n\t// Called after the test case completes (pass or fail)\n\tAfter string `json:\"after,omitempty\"`\n\n\t// Dynamic Mode Fields\n\t// ===============================\n\n\t// Simulator configures the user simulator for dynamic testing\n\t// When set, the test runs in dynamic mode with multi-turn conversation\n\tSimulator *Simulator `json:\"simulator,omitempty\"`\n\n\t// Checkpoints define validation points for dynamic testing\n\t// Each checkpoint is checked after every agent response\n\tCheckpoints []*Checkpoint `json:\"checkpoints,omitempty\"`\n\n\t// MaxTurns is the maximum number of conversation turns (default: 20)\n\tMaxTurns int `json:\"max_turns,omitempty\"`\n}\n\n// Simulator configures the user simulator for dynamic testing\ntype Simulator struct {\n\t// Use is the simulator agent ID (no prefix needed)\n\tUse string `json:\"use\"`\n\n\t// Options for the simulator agent\n\tOptions *SimulatorOptions `json:\"options,omitempty\"`\n}\n\n// SimulatorOptions configures simulator behavior\ntype SimulatorOptions struct {\n\t// Metadata passed to the simulator agent\n\t// Common fields: persona, goal, style\n\tMetadata map[string]interface{} `json:\"metadata,omitempty\"`\n\n\t// Connector overrides the simulator's default connector\n\tConnector string `json:\"connector,omitempty\"`\n}\n\n// Checkpoint defines a validation point in dynamic testing\ntype Checkpoint struct {\n\t// ID is the unique identifier for this checkpoint\n\tID string `json:\"id\"`\n\n\t// Description is a human-readable description\n\tDescription string `json:\"description,omitempty\"`\n\n\t// Assert defines the assertion to validate\n\t// Same format as Case.Assert\n\tAssert interface{} `json:\"assert\"`\n\n\t// After specifies checkpoint IDs that must be reached before this one\n\t// Used to enforce ordering (e.g., \"ask_type\" must come before \"confirm\")\n\tAfter []string `json:\"after,omitempty\"`\n\n\t// Required indicates if this checkpoint must be reached (default: true)\n\t// Optional checkpoints don't cause test failure if not reached\n\tRequired *bool `json:\"required,omitempty\"`\n}\n\n// CaseOptions represents per-test-case context options\n// Maps to context.Options fields\ntype CaseOptions struct {\n\t// Connector overrides the agent's default connector\n\tConnector string `json:\"connector,omitempty\"`\n\n\t// Skip configuration\n\tSkip *CaseSkipOptions `json:\"skip,omitempty\"`\n\n\t// DisableGlobalPrompts temporarily disables global prompts for this request\n\tDisableGlobalPrompts bool `json:\"disable_global_prompts,omitempty\"`\n\n\t// Search mode, default is true (use pointer to distinguish unset from false)\n\tSearch *bool `json:\"search,omitempty\"`\n\n\t// Mode is the agent mode (default: \"chat\")\n\tMode string `json:\"mode,omitempty\"`\n\n\t// Metadata for passing custom data to hooks (e.g., scenario selection)\n\tMetadata map[string]interface{} `json:\"metadata,omitempty\"`\n}\n\n// CaseSkipOptions represents skip configuration for a test case\n// Maps to context.Skip fields\ntype CaseSkipOptions struct {\n\tHistory bool `json:\"history,omitempty\"` // Skip history loading\n\tTrace bool `json:\"trace,omitempty\"` // Skip trace logging\n\tOutput bool `json:\"output,omitempty\"` // Skip output to client\n\tKeyword bool `json:\"keyword,omitempty\"` // Skip keyword extraction\n\tSearch bool `json:\"search,omitempty\"` // Skip auto search\n}\n\n// Assertion represents a single assertion rule\ntype Assertion struct {\n\t// Type is the assertion type:\n\t// - \"equals\": exact match (default if expected is set)\n\t// - \"contains\": output contains the expected string/value\n\t// - \"not_contains\": output does not contain the string/value\n\t// - \"json_path\": extract value using JSON path and compare\n\t// - \"regex\": match output against regex pattern\n\t// - \"script\": run a custom assertion script\n\t// - \"type\": check output type (string, object, array, number, boolean)\n\t// - \"schema\": validate against JSON schema\n\t// - \"agent\": use an agent to validate the response\n\tType string `json:\"type\"`\n\n\t// Value is the expected value or pattern (depends on type)\n\tValue interface{} `json:\"value,omitempty\"`\n\n\t// Path is the JSON path for json_path assertions (e.g., \"$.need_search\")\n\tPath string `json:\"path,omitempty\"`\n\n\t// Script is the assertion script name for script assertions\n\t// The script receives (output, input, expected) and returns {pass: bool, message: string}\n\tScript string `json:\"script,omitempty\"`\n\n\t// Use specifies the agent/script for validation\n\t// For agent assertions: \"agents:tests.validator-agent\" (with prefix)\n\t// For script assertions: \"scripts:tests.validate\" (with prefix)\n\tUse string `json:\"use,omitempty\"`\n\n\t// Options for agent-driven assertions (aligned with context.Options)\n\tOptions *AssertionOptions `json:\"options,omitempty\"`\n\n\t// Message is a custom failure message\n\tMessage string `json:\"message,omitempty\"`\n\n\t// Negate inverts the assertion result\n\tNegate bool `json:\"negate,omitempty\"`\n}\n\n// AssertionOptions for agent-driven assertions\ntype AssertionOptions struct {\n\t// Connector overrides the agent's default connector\n\tConnector string `json:\"connector,omitempty\"`\n\n\t// Metadata contains custom data passed to the validator agent\n\tMetadata map[string]interface{} `json:\"metadata,omitempty\"`\n}\n\n// AssertionResult represents the result of an assertion\ntype AssertionResult struct {\n\t// Passed indicates whether the assertion passed\n\tPassed bool `json:\"passed\"`\n\n\t// Message describes the assertion result\n\tMessage string `json:\"message,omitempty\"`\n\n\t// Assertion is the original assertion that was evaluated\n\tAssertion *Assertion `json:\"assertion,omitempty\"`\n\n\t// Actual is the actual value that was compared\n\tActual interface{} `json:\"actual,omitempty\"`\n\n\t// Expected is the expected value\n\tExpected interface{} `json:\"expected,omitempty\"`\n}\n\n// GetEnvironment returns the effective test environment for this test case\n// Priority: command line flags > context config > test case fields > defaults\nfunc (tc *Case) GetEnvironment(opts *Options) *Environment {\n\t// Start with context config if available, otherwise use defaults\n\tvar env *Environment\n\tif opts != nil && opts.ContextData != nil {\n\t\tenv = NewEnvironmentWithContext(\"\", \"\", opts.ContextData)\n\t} else {\n\t\tenv = NewEnvironment(\"\", \"\")\n\t}\n\n\t// Apply test case specific values\n\tif tc.UserID != \"\" {\n\t\tenv.UserID = tc.UserID\n\t}\n\tif tc.TeamID != \"\" {\n\t\tenv.TeamID = tc.TeamID\n\t}\n\n\t// Apply command line overrides (highest priority)\n\tif opts != nil {\n\t\tif opts.UserID != \"\" {\n\t\t\tenv.UserID = opts.UserID\n\t\t}\n\t\tif opts.TeamID != \"\" {\n\t\t\tenv.TeamID = opts.TeamID\n\t\t}\n\t\tif opts.Locale != \"\" {\n\t\t\tenv.Locale = opts.Locale\n\t\t}\n\t}\n\n\treturn env\n}\n\n// GetMessages converts the Input to a slice of context.Message\n// This handles all input formats: string, Message, []Message\nfunc (tc *Case) GetMessages() ([]context.Message, error) {\n\treturn ParseInput(tc.Input)\n}\n\n// GetMessagesWithOptions converts the Input to a slice of context.Message with options\n// This handles all input formats: string, Message, []Message\n// It also processes file:// references in content parts\nfunc (tc *Case) GetMessagesWithOptions(opts *InputOptions) ([]context.Message, error) {\n\treturn ParseInputWithOptions(tc.Input, opts)\n}\n\n// GetTimeout returns the timeout duration for this test case\n// Returns the override timeout if set, otherwise returns the default\nfunc (tc *Case) GetTimeout(defaultTimeout time.Duration) time.Duration {\n\tif tc.Timeout == \"\" {\n\t\treturn defaultTimeout\n\t}\n\td, err := time.ParseDuration(tc.Timeout)\n\tif err != nil {\n\t\treturn defaultTimeout\n\t}\n\treturn d\n}\n\n// Result represents the result of running a single test case\ntype Result struct {\n\t// ID is the test case identifier\n\tID string `json:\"id\"`\n\n\t// Status is the test execution status\n\tStatus Status `json:\"status\"`\n\n\t// Input is the original test input (for reference in reports)\n\tInput interface{} `json:\"input\"`\n\n\t// Output is the actual output from the agent\n\tOutput interface{} `json:\"output,omitempty\"`\n\n\t// Expected is the expected output (if specified in test case)\n\tExpected interface{} `json:\"expected,omitempty\"`\n\n\t// DurationMs is the execution duration in milliseconds\n\tDurationMs int64 `json:\"duration_ms\"`\n\n\t// Error contains the error message if status is failed/error/timeout\n\tError string `json:\"error,omitempty\"`\n\n\t// Options contains the context options used for this test case\n\tOptions *CaseOptions `json:\"options,omitempty\"`\n\n\t// Metadata contains additional result metadata\n\tMetadata map[string]interface{} `json:\"metadata,omitempty\"`\n}\n\n// RunDetail represents the result of a single run in stability testing\ntype RunDetail struct {\n\t// Run is the run number (1-based)\n\tRun int `json:\"run\"`\n\n\t// Status is the execution status for this run\n\tStatus Status `json:\"status\"`\n\n\t// DurationMs is the execution duration in milliseconds\n\tDurationMs int64 `json:\"duration_ms\"`\n\n\t// Output is the output from this run\n\tOutput interface{} `json:\"output,omitempty\"`\n\n\t// Error contains the error message if this run failed\n\tError string `json:\"error,omitempty\"`\n}\n\n// StabilityResult represents the stability analysis result for a test case\ntype StabilityResult struct {\n\t// ID is the test case identifier\n\tID string `json:\"id\"`\n\n\t// Input is the original test input\n\tInput interface{} `json:\"input\"`\n\n\t// Expected is the expected output (if specified)\n\tExpected interface{} `json:\"expected,omitempty\"`\n\n\t// Runs is the total number of runs\n\tRuns int `json:\"runs\"`\n\n\t// Passed is the number of runs that passed\n\tPassed int `json:\"passed\"`\n\n\t// Failed is the number of runs that failed\n\tFailed int `json:\"failed\"`\n\n\t// PassRate is the pass rate percentage (0-100)\n\tPassRate float64 `json:\"pass_rate\"`\n\n\t// Consistency is a measure of output consistency (0-1)\n\t// 1.0 means all outputs are identical, lower values indicate variation\n\tConsistency float64 `json:\"consistency\"`\n\n\t// Stable indicates whether the test is considered stable\n\tStable bool `json:\"stable\"`\n\n\t// StabilityClass is the stability classification\n\tStabilityClass StabilityClass `json:\"stability_class\"`\n\n\t// Timing statistics\n\tAvgDurationMs float64 `json:\"avg_duration_ms\"`\n\tMinDurationMs int64 `json:\"min_duration_ms\"`\n\tMaxDurationMs int64 `json:\"max_duration_ms\"`\n\tStdDeviationMs float64 `json:\"std_deviation_ms\"`\n\n\t// RunDetails contains details for each run\n\tRunDetails []*RunDetail `json:\"run_details\"`\n}\n\n// CalculateStability calculates stability metrics from run details\nfunc (sr *StabilityResult) CalculateStability() {\n\tif len(sr.RunDetails) == 0 {\n\t\treturn\n\t}\n\n\tsr.Runs = len(sr.RunDetails)\n\tsr.Passed = 0\n\tsr.Failed = 0\n\n\tvar totalDuration int64\n\tsr.MinDurationMs = math.MaxInt64\n\tsr.MaxDurationMs = 0\n\n\tfor _, rd := range sr.RunDetails {\n\t\tif rd.Status == StatusPassed {\n\t\t\tsr.Passed++\n\t\t} else {\n\t\t\tsr.Failed++\n\t\t}\n\n\t\ttotalDuration += rd.DurationMs\n\t\tif rd.DurationMs < sr.MinDurationMs {\n\t\t\tsr.MinDurationMs = rd.DurationMs\n\t\t}\n\t\tif rd.DurationMs > sr.MaxDurationMs {\n\t\t\tsr.MaxDurationMs = rd.DurationMs\n\t\t}\n\t}\n\n\t// Calculate pass rate\n\tsr.PassRate = float64(sr.Passed) / float64(sr.Runs) * 100\n\n\t// Calculate average duration\n\tsr.AvgDurationMs = float64(totalDuration) / float64(sr.Runs)\n\n\t// Calculate standard deviation\n\tvar sumSquares float64\n\tfor _, rd := range sr.RunDetails {\n\t\tdiff := float64(rd.DurationMs) - sr.AvgDurationMs\n\t\tsumSquares += diff * diff\n\t}\n\tsr.StdDeviationMs = math.Sqrt(sumSquares / float64(sr.Runs))\n\n\t// Determine stability classification\n\tsr.StabilityClass = ClassifyStability(sr.PassRate)\n\tsr.Stable = sr.PassRate == 100\n\n\t// Calculate consistency (simplified: based on pass rate)\n\tsr.Consistency = sr.PassRate / 100\n}\n\n// ClassifyStability returns the stability classification based on pass rate\nfunc ClassifyStability(passRate float64) StabilityClass {\n\tswitch {\n\tcase passRate == 100:\n\t\treturn StabilityStable\n\tcase passRate >= 80:\n\t\treturn StabilityMostlyStable\n\tcase passRate >= 50:\n\t\treturn StabilityUnstable\n\tdefault:\n\t\treturn StabilityHighlyUnstable\n\t}\n}\n\n// Summary contains aggregated statistics for the test run\ntype Summary struct {\n\t// Total number of test cases\n\tTotal int `json:\"total\"`\n\n\t// Passed number of test cases that passed\n\tPassed int `json:\"passed\"`\n\n\t// Failed number of test cases that failed\n\tFailed int `json:\"failed\"`\n\n\t// Skipped number of test cases that were skipped\n\tSkipped int `json:\"skipped\"`\n\n\t// Errors number of test cases with runtime errors\n\tErrors int `json:\"errors\"`\n\n\t// Timeouts number of test cases that timed out\n\tTimeouts int `json:\"timeouts\"`\n\n\t// DurationMs is the total execution duration in milliseconds\n\tDurationMs int64 `json:\"duration_ms\"`\n\n\t// AgentID is the ID of the agent being tested\n\tAgentID string `json:\"agent_id\"`\n\n\t// AgentPath is the file path of the agent (for path-based resolution)\n\tAgentPath string `json:\"agent_path,omitempty\"`\n\n\t// Connector is the connector used for the test\n\tConnector string `json:\"connector\"`\n\n\t// Stability metrics (when Runs > 1)\n\t// ===============================\n\n\t// RunsPerCase is the number of runs per test case\n\tRunsPerCase int `json:\"runs_per_case,omitempty\"`\n\n\t// TotalRuns is the total number of runs (Total * RunsPerCase)\n\tTotalRuns int `json:\"total_runs,omitempty\"`\n\n\t// OverallPassRate is the overall pass rate percentage\n\tOverallPassRate float64 `json:\"overall_pass_rate,omitempty\"`\n\n\t// StableCases is the number of cases with 100% pass rate\n\tStableCases int `json:\"stable_cases,omitempty\"`\n\n\t// UnstableCases is the number of cases with < 100% pass rate\n\tUnstableCases int `json:\"unstable_cases,omitempty\"`\n}\n\n// Report represents the complete test report\ntype Report struct {\n\t// Summary contains aggregated statistics\n\tSummary *Summary `json:\"summary\"`\n\n\t// Environment contains the test environment configuration\n\tEnvironment *Environment `json:\"environment,omitempty\"`\n\n\t// Results contains individual test results (for single run)\n\tResults []*Result `json:\"results,omitempty\"`\n\n\t// StabilityResults contains stability analysis results (for multiple runs)\n\tStabilityResults []*StabilityResult `json:\"stability_results,omitempty\"`\n\n\t// Metadata contains additional report metadata\n\tMetadata *ReportMetadata `json:\"metadata\"`\n}\n\n// ReportMetadata contains metadata about the test report\ntype ReportMetadata struct {\n\t// StartedAt is when the test run started\n\tStartedAt time.Time `json:\"started_at\"`\n\n\t// CompletedAt is when the test run completed\n\tCompletedAt time.Time `json:\"completed_at\"`\n\n\t// Version is the Yao version\n\tVersion string `json:\"version\"`\n\n\t// InputFile is the path to the input file\n\tInputFile string `json:\"input_file\"`\n\n\t// OutputFile is the path to the output file\n\tOutputFile string `json:\"output_file\"`\n\n\t// Options contains the test options used\n\tOptions *Options `json:\"options,omitempty\"`\n}\n\n// HasFailures returns true if there are any failed, error, or timeout tests\nfunc (r *Report) HasFailures() bool {\n\treturn r.Summary.Failed > 0 || r.Summary.Errors > 0 || r.Summary.Timeouts > 0\n}\n\n// PassRate returns the pass rate as a percentage (0-100)\nfunc (r *Report) PassRate() float64 {\n\tif r.Summary.Total == 0 {\n\t\treturn 0\n\t}\n\treturn float64(r.Summary.Passed) / float64(r.Summary.Total) * 100\n}\n\n// IsStabilityTest returns true if this is a stability test (multiple runs)\nfunc (r *Report) IsStabilityTest() bool {\n\treturn r.Summary.RunsPerCase > 1\n}\n\n// AgentInfo contains information about the agent being tested\ntype AgentInfo struct {\n\t// ID is the agent identifier\n\tID string `json:\"id\"`\n\n\t// Name is the human-readable name\n\tName string `json:\"name\"`\n\n\t// Description is the agent description\n\tDescription string `json:\"description,omitempty\"`\n\n\t// Path is the file system path to the agent\n\tPath string `json:\"path\"`\n\n\t// Connector is the default connector\n\tConnector string `json:\"connector\"`\n\n\t// Type is the agent type (e.g., \"worker\", \"assistant\")\n\tType string `json:\"type,omitempty\"`\n}\n\n// ReporterInput is the input passed to a custom reporter agent\ntype ReporterInput struct {\n\t// Report is the test report to format\n\tReport *Report `json:\"report\"`\n\n\t// Format is the desired output format\n\tFormat string `json:\"format\"`\n\n\t// Options contains additional formatting options\n\tOptions *ReporterOptions `json:\"options,omitempty\"`\n}\n\n// ReporterOptions contains options for custom reporter agents\ntype ReporterOptions struct {\n\t// Verbose includes detailed output in the report\n\tVerbose bool `json:\"verbose,omitempty\"`\n\n\t// IncludeOutputs includes full outputs in the report\n\tIncludeOutputs bool `json:\"include_outputs,omitempty\"`\n\n\t// IncludeInputs includes full inputs in the report\n\tIncludeInputs bool `json:\"include_inputs,omitempty\"`\n\n\t// MaxOutputLength limits the output length in the report\n\tMaxOutputLength int `json:\"max_output_length,omitempty\"`\n\n\t// Theme is the report theme (for HTML reports)\n\tTheme string `json:\"theme,omitempty\"`\n\n\t// Title is the report title\n\tTitle string `json:\"title,omitempty\"`\n}\n" }, { "path": "agent/testutils/testutils.go", "content": "package testutils\n\nimport (\n\t\"testing\"\n\n\t_ \"github.com/yaoapp/gou/encoding\"\n\t\"github.com/yaoapp/gou/model\"\n\t\"github.com/yaoapp/gou/query\"\n\t\"github.com/yaoapp/gou/query/gou\"\n\t_ \"github.com/yaoapp/gou/text\"\n\t\"github.com/yaoapp/xun/capsule\"\n\t\"github.com/yaoapp/yao/agent\"\n\t\"github.com/yaoapp/yao/agent/caller\"\n\t\"github.com/yaoapp/yao/agent/llm\"\n\t\"github.com/yaoapp/yao/config\"\n\t\"github.com/yaoapp/yao/kb\"\n\t\"github.com/yaoapp/yao/test\"\n\n\t// Import assistant to trigger init() which registers AgentGetterFunc\n\t_ \"github.com/yaoapp/yao/agent/assistant\"\n)\n\n// Prepare prepare the test environment with optional V8 mode configuration\n// Usage:\n//\n//\ttestutils.Prepare(t) // standard mode (default)\n//\ttestutils.Prepare(t, test.PrepareOption{V8Mode: \"performance\"}) // performance mode for benchmarks\nfunc Prepare(t *testing.T, opts ...interface{}) {\n\ttest.Prepare(t, config.Conf, opts...)\n\n\t// Load KB (required for agent KB features)\n\t_, err := kb.Load(config.Conf)\n\tif err != nil {\n\t\tt.Fatal(err)\n\t}\n\n\t// Load agent\n\terr = agent.Load(config.Conf)\n\tif err != nil {\n\t\tt.Fatal(err)\n\t}\n\n\t// Ensure JSAPI factories are registered (may be called multiple times, idempotent)\n\t// This is needed because Go's init() order is not guaranteed across packages\n\tcaller.SetJSAPIFactory()\n\tllm.SetJSAPIFactory()\n\n\t// Register default query engine (required for DB search)\n\t// capsule.Global is initialized by test.Prepare\n\tif _, has := query.Engines[\"default\"]; !has && capsule.Global != nil {\n\t\tquery.Register(\"default\", &gou.Query{\n\t\t\tQuery: capsule.Query(),\n\t\t\tGetTableName: func(s string) string {\n\t\t\t\tif mod, has := model.Models[s]; has {\n\t\t\t\t\treturn mod.MetaData.Table.Name\n\t\t\t\t}\n\t\t\t\treturn s\n\t\t\t},\n\t\t\tAESKey: config.Conf.DB.AESKey,\n\t\t})\n\t}\n}\n\n// Clean clean the test environment\nfunc Clean(t *testing.T) {\n\ttest.Clean()\n}\n" }, { "path": "agent/types/dsl.go", "content": "package types\n\nimport \"github.com/yaoapp/gou/store\"\n\n// GetCacheStore get the cache store\nfunc (dsl *DSL) GetCacheStore() (store.Store, error) {\n\tif dsl.Cache == \"\" {\n\t\treturn store.Get(\"__yao.agent.cache\")\n\t}\n\treturn store.Get(dsl.Cache)\n}\n" }, { "path": "agent/types/types.go", "content": "package types\n\nimport (\n\t\"github.com/yaoapp/yao/agent/assistant\"\n\tsearchTypes \"github.com/yaoapp/yao/agent/search/types\"\n\tstore \"github.com/yaoapp/yao/agent/store/types\"\n)\n\n// DSL AI assistant\ntype DSL struct {\n\n\t// Agent Global Settings\n\t// ===============================\n\tUses *Uses `json:\"uses,omitempty\" yaml:\"uses,omitempty\"` // Which assistant to use default, title, prompt\n\tStoreSetting store.Setting `json:\"store\" yaml:\"store\"` // The store setting of the assistant\n\tCache string `json:\"cache\" yaml:\"cache\"` // The cache store of the assistant, if not set, default is \"__yao.agent.cache\"\n\n\t// System Agents Connector Settings\n\t// ===============================\n\t// System configures connectors for system agents (__yao.keyword, __yao.querydsl, __yao.title, __yao.prompt)\n\t// Each agent can have its own connector, or use the default\n\t// If not set, fallback to the first connector that supports the required capabilities\n\tSystem *System `json:\"system,omitempty\" yaml:\"system,omitempty\"`\n\n\t// Global External Settings\n\t// ===============================\n\tKB *store.KBSetting `json:\"kb,omitempty\" yaml:\"kb,omitempty\"` // The knowledge base configuration loaded from agent/kb.yml\n\tSearch *searchTypes.Config `json:\"search,omitempty\" yaml:\"search,omitempty\"` // The search configuration loaded from agent/search.yao\n\n\t// Internal\n\t// ===============================\n\t// ID string `json:\"-\" yaml:\"-\"` // The id of the instance\n\tAssistant assistant.API `json:\"-\" yaml:\"-\"` // The default assistant\n\tStore store.Store `json:\"-\" yaml:\"-\"` // The store of the assistant\n\tGlobalPrompts []store.Prompt `json:\"-\" yaml:\"-\"` // Global prompts loaded from agent/prompts.yml\n}\n\n// Uses the default assistant settings\n// ===============================\ntype Uses struct {\n\tDefault string `json:\"default,omitempty\" yaml:\"default,omitempty\"` // The default assistant to use\n\tTitle string `json:\"title,omitempty\" yaml:\"title,omitempty\"` // The assistant for generating the topic title.\n\tPrompt string `json:\"prompt,omitempty\" yaml:\"prompt,omitempty\"` // The assistant for generating the prompt.\n\tRobotPrompt string `json:\"robot_prompt,omitempty\" yaml:\"robot_prompt,omitempty\"` // The assistant for generating Robot's system prompt (responsibilities description).\n\tVision string `json:\"vision,omitempty\" yaml:\"vision,omitempty\"` // The assistant for generating the image/video description, if the assistant enable the vision and model not support vision, use the vision model to describe the image/video, and return the messages with the image/video's description. Format: \"agent\" or \"mcp:mcp_server_id\"\n\tAudio string `json:\"audio,omitempty\" yaml:\"audio,omitempty\"` // The assistant for processing audio (speech-to-text, text-to-speech). If the model doesn't support audio, use this to convert audio to text. Format: \"agent\" or \"mcp:mcp_server_id\"\n\tSearch string `json:\"search,omitempty\" yaml:\"search,omitempty\"` // The assistant for searching the knowledge, global web search. If not set, and the assistant enable the knowledge, it will search the result from the knowledge automatically.\n\tFetch string `json:\"fetch,omitempty\" yaml:\"fetch,omitempty\"` // The assistant for fetching the http/https/ftp/sftp/etc. file, and return the file's content. if not set, use the http process to fetch the file.\n\n\t// Search-related processing tools (NLP)\n\tWeb string `json:\"web,omitempty\" yaml:\"web,omitempty\"` // Web search handler: \"builtin\", \"\", \"mcp:.\"\n\tKeyword string `json:\"keyword,omitempty\" yaml:\"keyword,omitempty\"` // Keyword extraction: \"builtin\", \"\", \"mcp:.\"\n\tQueryDSL string `json:\"querydsl,omitempty\" yaml:\"querydsl,omitempty\"` // QueryDSL generation: \"builtin\", \"\", \"mcp:.\"\n\tRerank string `json:\"rerank,omitempty\" yaml:\"rerank,omitempty\"` // Result reranking: \"builtin\", \"\", \"mcp:.\"\n}\n\n// System configures connectors for system agents\n// ===============================\ntype System struct {\n\tDefault string `json:\"default,omitempty\" yaml:\"default,omitempty\"` // Default connector for all system agents\n\tKeyword string `json:\"keyword,omitempty\" yaml:\"keyword,omitempty\"` // Connector for __yao.keyword agent\n\tQueryDSL string `json:\"querydsl,omitempty\" yaml:\"querydsl,omitempty\"` // Connector for __yao.querydsl agent\n\tTitle string `json:\"title,omitempty\" yaml:\"title,omitempty\"` // Connector for __yao.title agent\n\tPrompt string `json:\"prompt,omitempty\" yaml:\"prompt,omitempty\"` // Connector for __yao.prompt agent\n\tRobotPrompt string `json:\"robot_prompt,omitempty\" yaml:\"robot_prompt,omitempty\"` // Connector for __yao.robot_prompt agent\n\tNeedSearch string `json:\"needsearch,omitempty\" yaml:\"needsearch,omitempty\"` // Connector for __yao.needsearch agent\n\tEntity string `json:\"entity,omitempty\" yaml:\"entity,omitempty\"` // Connector for __yao.entity agent\n}\n\n// Mention Structure\n// ===============================\ntype Mention struct {\n\tID string `json:\"id\"`\n\tName string `json:\"name\"`\n\tAvatar string `json:\"avatar,omitempty\"`\n\tType string `json:\"type,omitempty\"`\n}\n" }, { "path": "aigc/aigc.go", "content": "package aigc\n\nimport (\n\t\"fmt\"\n\t\"strings\"\n\n\tjsoniter \"github.com/json-iterator/go\"\n\t\"github.com/yaoapp/gou/connector\"\n\t\"github.com/yaoapp/gou/process\"\n\t\"github.com/yaoapp/kun/exception\"\n\t\"github.com/yaoapp/yao/openai\"\n)\n\n// Autopilots the loaded autopilots\nvar Autopilots = []string{}\n\n// AIGCs the loaded AIGCs\nvar AIGCs = map[string]*DSL{}\n\n// Select select the AIGC\nfunc Select(id string) (*DSL, error) {\n\tif AIGCs[id] == nil {\n\t\treturn nil, fmt.Errorf(\"aigc %s not found\", id)\n\t}\n\treturn AIGCs[id], nil\n}\n\n// Call the AIGC\nfunc (ai *DSL) Call(content string, user string, option map[string]interface{}) (interface{}, *exception.Exception) {\n\n\tmessages := []map[string]interface{}{}\n\tfor _, prompt := range ai.Prompts {\n\t\tmessage := map[string]interface{}{\"role\": prompt.Role, \"content\": prompt.Content}\n\t\tif prompt.Name != \"\" {\n\t\t\tmessage[\"name\"] = prompt.Name\n\t\t}\n\t\tmessages = append(messages, message)\n\t}\n\n\t// add the user message\n\tmessage := map[string]interface{}{\"role\": \"user\", \"content\": content}\n\tif user != \"\" {\n\t\tmessage[\"user\"] = user\n\t}\n\tmessages = append(messages, message)\n\n\tbytes, err := jsoniter.Marshal(messages)\n\tif err != nil {\n\t\treturn nil, exception.New(err.Error(), 400)\n\t}\n\n\ttoken, err := ai.AI.Tiktoken(string(bytes))\n\tif err != nil {\n\t\treturn nil, exception.New(err.Error(), 400)\n\t}\n\n\tif token > ai.AI.MaxToken() {\n\t\treturn nil, exception.New(\"token limit exceeded\", 400)\n\t}\n\n\t// call the AI\n\tres, ex := ai.AI.ChatCompletions(messages, option, nil)\n\tif ex != nil {\n\t\treturn nil, ex\n\t}\n\n\tresText, ex := ai.AI.GetContent(res)\n\tif ex != nil {\n\t\treturn nil, ex\n\t}\n\n\tif ai.Process == \"\" {\n\t\treturn resText, nil\n\t}\n\n\tvar param interface{} = resText\n\tif ai.Optional.JSON {\n\t\terr = jsoniter.Unmarshal([]byte(resText), ¶m)\n\t\tif err != nil {\n\t\t\treturn nil, exception.New(\"%s parse error: %s\", 400, resText, err.Error())\n\t\t}\n\t}\n\n\tp, err := process.Of(ai.Process, param)\n\tif err != nil {\n\t\treturn nil, exception.New(err.Error(), 400)\n\t}\n\n\tresProcess, err := p.Exec()\n\tif err != nil {\n\t\treturn nil, exception.New(err.Error(), 500)\n\t}\n\n\treturn resProcess, nil\n}\n\n// NewAI create a new AI\nfunc (ai *DSL) newAI() (AI, error) {\n\n\tif ai.Connector == \"\" || strings.HasPrefix(ai.Connector, \"moapi\") {\n\t\tmodel := \"gpt-3.5-turbo\"\n\t\tif strings.HasPrefix(ai.Connector, \"moapi:\") {\n\t\t\tmodel = strings.TrimPrefix(ai.Connector, \"moapi:\")\n\t\t}\n\n\t\tmo, err := openai.NewMoapi(model)\n\t\tif err != nil {\n\t\t\treturn nil, err\n\t\t}\n\t\treturn mo, nil\n\t}\n\n\tconn, err := connector.Select(ai.Connector)\n\tif err != nil {\n\t\treturn nil, err\n\t}\n\n\tif conn.Is(connector.OPENAI) {\n\t\treturn openai.New(ai.Connector)\n\t}\n\n\treturn nil, fmt.Errorf(\"%s connector %s not support, should be a openai\", ai.ID, ai.Connector)\n}\n" }, { "path": "aigc/aigc_test.go", "content": "package aigc\n\nimport (\n\t\"testing\"\n\n\t\"github.com/stretchr/testify/assert\"\n\t\"github.com/yaoapp/yao/config\"\n\t\"github.com/yaoapp/yao/test\"\n)\n\nfunc TestCall(t *testing.T) {\n\ttest.Prepare(t, config.Conf)\n\tdefer test.Clean()\n\tprepare(t)\n\n\taigc, err := Select(\"translate\")\n\tif err != nil {\n\t\tt.Fatal(err)\n\t}\n\n\tcontent, ex := aigc.Call(\"你好哇\", \"\", nil)\n\tif ex != nil {\n\t\tt.Fatal(ex.Message)\n\t}\n\tassert.Contains(t, content, \"Hello\")\n}\n\nfunc TestCallWithProcess(t *testing.T) {\n\ttest.Prepare(t, config.Conf)\n\tdefer test.Clean()\n\tprepare(t)\n\n\taigc, err := Select(\"draw\")\n\tif err != nil {\n\t\tt.Fatal(err)\n\t}\n\n\targs, ex := aigc.Call(\"帮我画一只小白兔，要有白色的耳朵. 画布高度 256，宽度 256\", \"\", nil)\n\tif ex != nil {\n\t\tt.Fatal(ex.Message)\n\t}\n\n\tdata, ok := args.(map[string]interface{})\n\tif !ok {\n\t\tt.Fatal(\"args is not map[string]interface{}\")\n\t}\n\n\tassert.Equal(t, float64(256), data[\"height\"])\n\tassert.Equal(t, float64(256), data[\"width\"])\n}\n\nfunc prepare(t *testing.T) {\n\terr := Load(config.Conf)\n\tif err != nil {\n\t\tt.Fatal(err)\n\t}\n}\n" }, { "path": "aigc/load.go", "content": "package aigc\n\nimport (\n\t\"fmt\"\n\t\"strings\"\n\n\t\"github.com/yaoapp/gou/application\"\n\t\"github.com/yaoapp/yao/config\"\n\t\"github.com/yaoapp/yao/share\"\n)\n\n// Load load AIGC\nfunc Load(cfg config.Config) error {\n\n\t// Ignore if the aigcs directory does not exist\n\texists, err := application.App.Exists(\"aigcs\")\n\tif err != nil {\n\t\treturn err\n\t}\n\tif !exists {\n\t\treturn nil\n\t}\n\n\texts := []string{\"*.ai.yml\", \"*.ai.yaml\"}\n\tmessages := []string{}\n\terr = application.App.Walk(\"aigcs\", func(root, file string, isdir bool) error {\n\t\tif isdir {\n\t\t\treturn nil\n\t\t}\n\n\t\tid := aigcID(root, file)\n\t\t_, err := LoadFile(file, id)\n\t\tif err != nil {\n\t\t\tmessages = append(messages, err.Error())\n\t\t}\n\n\t\treturn nil\n\t}, exts...)\n\n\tif err != nil {\n\t\treturn err\n\t}\n\n\tif len(messages) > 0 {\n\t\treturn fmt.Errorf(\"%s\", strings.Join(messages, \";\\n\"))\n\t}\n\n\treturn nil\n}\n\n// aigcID parses AIGC ID from file path\n// Special handling for .ai.yml and .ai.yaml extensions\n// e.g., \"aigcs/translate.ai.yml\" -> \"translate\"\nfunc aigcID(root, file string) string {\n\tid := share.ID(root, file)\n\t// Remove \"_ai\" suffix caused by .ai.yml/.ai.yaml extension\n\t// share.ID treats .yml/.yaml as single extension, so \"translate.ai.yml\" becomes \"translate_ai\"\n\tid = strings.TrimSuffix(id, \"_ai\")\n\treturn id\n}\n\n// LoadFile load AIGC by file\nfunc LoadFile(file string, id string) (*DSL, error) {\n\n\tdata, err := application.App.Read(file)\n\tif err != nil {\n\t\treturn nil, err\n\t}\n\treturn LoadSource(data, file, id)\n}\n\n// LoadSource load AIGC\nfunc LoadSource(data []byte, file, id string) (*DSL, error) {\n\n\tdsl := DSL{\n\t\tID: id,\n\t\tOptional: Optional{\n\t\t\tAutopilot: false,\n\t\t\tJSON: false,\n\t\t},\n\t}\n\n\terr := application.Parse(file, data, &dsl)\n\tif err != nil {\n\t\treturn nil, err\n\t}\n\n\tif dsl.Prompts == nil || len(dsl.Prompts) == 0 {\n\t\treturn nil, fmt.Errorf(\"%s prompts is required\", id)\n\t}\n\n\t// create AI interface\n\tdsl.AI, err = dsl.newAI()\n\tif err != nil {\n\t\treturn nil, err\n\t}\n\n\t// add to autopilots\n\tif dsl.Optional.Autopilot {\n\t\tAutopilots = append(Autopilots, id)\n\t}\n\n\t// add to AIGCs\n\tAIGCs[id] = &dsl\n\treturn AIGCs[id], nil\n}\n" }, { "path": "aigc/load_test.go", "content": "package aigc\n\nimport (\n\t\"testing\"\n\n\t\"github.com/stretchr/testify/assert\"\n\t\"github.com/yaoapp/yao/config\"\n\t\"github.com/yaoapp/yao/test\"\n)\n\nfunc TestLoad(t *testing.T) {\n\ttest.Prepare(t, config.Conf)\n\tdefer test.Clean()\n\n\tLoad(config.Conf)\n\tcheck(t)\n}\n\nfunc check(t *testing.T) {\n\tids := map[string]bool{}\n\tfor id := range AIGCs {\n\t\tids[id] = true\n\t}\n\n\tassert.True(t, ids[\"translate\"])\n\tassert.True(t, ids[\"draw\"])\n\tassert.GreaterOrEqual(t, len(Autopilots), 2)\n}\n" }, { "path": "aigc/process.go", "content": "package aigc\n\nimport (\n\t\"github.com/yaoapp/gou/process\"\n\t\"github.com/yaoapp/kun/exception\"\n)\n\nfunc init() {\n\tprocess.Register(\"aigcs\", processAigcs)\n}\n\n// processScripts\nfunc processAigcs(process *process.Process) interface{} {\n\n\tprocess.ValidateArgNums(1)\n\taigc, err := Select(process.ID)\n\tif err != nil {\n\t\texception.New(\"aigcs.%s not loaded\", 404, process.ID).Throw()\n\t\treturn nil\n\t}\n\n\tcontent := process.ArgsString(0)\n\tuser := \"\"\n\n\tvar option map[string]interface{} = nil\n\tif process.NumOfArgs() > 1 {\n\t\tuser = process.ArgsString(1)\n\t}\n\n\tif process.NumOfArgs() > 2 {\n\t\toption = process.ArgsMap(2)\n\t}\n\n\tres, ex := aigc.Call(content, user, option)\n\tif ex != nil {\n\t\tex.Throw()\n\t}\n\n\treturn res\n}\n" }, { "path": "aigc/process_test.go", "content": "package aigc\n\nimport (\n\t\"testing\"\n\n\t\"github.com/stretchr/testify/assert\"\n\t\"github.com/yaoapp/gou/process\"\n\t\"github.com/yaoapp/yao/config\"\n\t\"github.com/yaoapp/yao/test\"\n)\n\nfunc TestProcessAigcs(t *testing.T) {\n\ttest.Prepare(t, config.Conf)\n\tdefer test.Clean()\n\tprepare(t)\n\n\targs := []interface{}{\"你好\"}\n\tres := process.New(\"aigcs.translate\", args...).Run()\n\tassert.Contains(t, res, \"Hello\")\n}\n" }, { "path": "aigc/types.go", "content": "package aigc\n\nimport (\n\t\"context\"\n\n\t\"github.com/yaoapp/kun/exception\"\n)\n\n// DSL the connector DSL\ntype DSL struct {\n\tID string `json:\"-\" yaml:\"-\"`\n\tName string `json:\"name,omitempty\"`\n\tConnector string `json:\"connector,omitempty\"`\n\tProcess string `json:\"process,omitempty\"`\n\tPrompts []Prompt `json:\"prompts\"`\n\tOptional Optional `json:\"optional,omitempty\"`\n\tAI AI `json:\"-\" yaml:\"-\"`\n}\n\n// Prompt a prompt\ntype Prompt struct {\n\tRole string `json:\"role\"`\n\tContent string `json:\"content\"`\n\tName string `json:\"name,omitempty\"`\n}\n\n// Optional optional\ntype Optional struct {\n\tAutopilot bool `json:\"autopilot,omitempty\"`\n\tJSON bool `json:\"json,omitempty\"`\n}\n\n// AI the AI interface\ntype AI interface {\n\tChatCompletions(messages []map[string]interface{}, option map[string]interface{}, cb func(data []byte) int) (interface{}, *exception.Exception)\n\tChatCompletionsWith(ctx context.Context, messages []map[string]interface{}, option map[string]interface{}, cb func(data []byte) int) (interface{}, *exception.Exception)\n\tGetContent(response interface{}) (string, *exception.Exception)\n\tEmbeddings(input interface{}, user string) (interface{}, *exception.Exception)\n\tTiktoken(input string) (int, error)\n\tMaxToken() int\n}\n" }, { "path": "api/README.md", "content": "# API\n" }, { "path": "api/api.go", "content": "package api\n\nimport (\n\t\"fmt\"\n\t\"strings\"\n\n\t\"github.com/yaoapp/gou/api\"\n\t\"github.com/yaoapp/gou/application\"\n\t\"github.com/yaoapp/yao/config\"\n\t\"github.com/yaoapp/yao/share\"\n)\n\n// Load apis\nfunc Load(cfg config.Config) error {\n\tmessages := []string{}\n\n\t// Ignore if the apis directory does not exist\n\texists, err := application.App.Exists(\"apis\")\n\tif err != nil {\n\t\treturn err\n\t}\n\tif !exists {\n\t\treturn nil\n\t}\n\n\texts := []string{\"*.http.yao\", \"*.http.json\", \"*.http.jsonc\"}\n\terr = application.App.Walk(\"apis\", func(root, file string, isdir bool) error {\n\t\tif isdir {\n\t\t\treturn nil\n\t\t}\n\t\t_, err := api.Load(file, share.ID(root, file))\n\t\tif err != nil {\n\t\t\tmessages = append(messages, err.Error())\n\t\t}\n\t\treturn err\n\t}, exts...)\n\n\tif len(messages) > 0 {\n\t\treturn fmt.Errorf(\"%s\", strings.Join(messages, \";\\n\"))\n\t}\n\n\treturn err\n}\n" }, { "path": "api/api_test.go", "content": "package api\n\nimport (\n\t\"testing\"\n\n\t\"github.com/stretchr/testify/assert\"\n\t\"github.com/yaoapp/gou/api\"\n\t\"github.com/yaoapp/yao/config\"\n\t\"github.com/yaoapp/yao/test\"\n)\n\nfunc TestLoad(t *testing.T) {\n\ttest.Prepare(t, config.Conf)\n\tdefer test.Clean()\n\n\tLoad(config.Conf)\n\tcheck(t)\n}\n\nfunc check(t *testing.T) {\n\tids := map[string]bool{}\n\tfor id := range api.APIs {\n\t\tids[id] = true\n\t}\n\tassert.True(t, ids[\"user\"])\n\tassert.True(t, ids[\"user.pet\"])\n\n\t// assert.True(t, ids[\"xiang.import\"]) // will be removed in the future\n\t// assert.True(t, ids[\"xiang.storage\"]) // will be removed in the future\n\n\t// wskeys := []string{}\n\t// for key := range websocket.Upgraders {\n\t// \twskeys = append(wskeys, key)\n\t// }\n\n\t// assert.Equal(t, 5, len(keys))\n\t// assert.Equal(t, 1, len(wskeys))\n}\n" }, { "path": "assert/asserter.go", "content": "package assert\n\nimport (\n\t\"fmt\"\n\t\"regexp\"\n\t\"strings\"\n\n\t\"github.com/yaoapp/gou/text\"\n)\n\n// Asserter handles assertions/validations\ntype Asserter struct {\n\tagentValidator AgentValidator\n\tscriptRunner ScriptRunner\n}\n\n// New creates a new Asserter\nfunc New() *Asserter {\n\treturn &Asserter{}\n}\n\n// WithAgentValidator sets the agent validator for agent-type assertions\nfunc (a *Asserter) WithAgentValidator(v AgentValidator) *Asserter {\n\ta.agentValidator = v\n\treturn a\n}\n\n// WithScriptRunner sets the script runner for script-type assertions\nfunc (a *Asserter) WithScriptRunner(r ScriptRunner) *Asserter {\n\ta.scriptRunner = r\n\treturn a\n}\n\n// Validate validates output against a list of assertions\n// Returns (passed, error message)\nfunc (a *Asserter) Validate(assertions []*Assertion, output interface{}) (bool, string) {\n\tif len(assertions) == 0 {\n\t\treturn true, \"\"\n\t}\n\n\tvar failures []string\n\tfor _, assertion := range assertions {\n\t\tresult := a.Evaluate(assertion, output, nil)\n\t\tif !result.Passed {\n\t\t\tmsg := result.Message\n\t\t\tif assertion.Message != \"\" {\n\t\t\t\tmsg = assertion.Message\n\t\t\t}\n\t\t\tfailures = append(failures, msg)\n\t\t}\n\t}\n\n\tif len(failures) > 0 {\n\t\treturn false, strings.Join(failures, \"; \")\n\t}\n\treturn true, \"\"\n}\n\n// ValidateWithDetails validates output and returns detailed results\nfunc (a *Asserter) ValidateWithDetails(assertions []*Assertion, output interface{}) *Result {\n\tif len(assertions) == 0 {\n\t\treturn &Result{Passed: true}\n\t}\n\n\tif len(assertions) == 1 {\n\t\treturn a.Evaluate(assertions[0], output, nil)\n\t}\n\n\tvar failures []string\n\tfor _, assertion := range assertions {\n\t\tresult := a.Evaluate(assertion, output, nil)\n\t\tif !result.Passed {\n\t\t\tmsg := result.Message\n\t\t\tif assertion.Message != \"\" {\n\t\t\t\tmsg = assertion.Message\n\t\t\t}\n\t\t\tfailures = append(failures, msg)\n\t\t}\n\t}\n\n\tif len(failures) > 0 {\n\t\treturn &Result{\n\t\t\tPassed: false,\n\t\t\tMessage: strings.Join(failures, \"; \"),\n\t\t}\n\t}\n\treturn &Result{Passed: true}\n}\n\n// Evaluate evaluates a single assertion\nfunc (a *Asserter) Evaluate(assertion *Assertion, output, input interface{}) *Result {\n\tresult := &Result{\n\t\tAssertion: assertion,\n\t\tExpected: assertion.Value,\n\t}\n\n\tswitch assertion.Type {\n\tcase \"equals\", \"\":\n\t\tresult = a.assertEquals(assertion, output)\n\tcase \"contains\":\n\t\tresult = a.assertContains(assertion, output)\n\tcase \"not_contains\":\n\t\tresult = a.assertNotContains(assertion, output)\n\tcase \"json_path\":\n\t\tresult = a.assertJSONPath(assertion, output)\n\tcase \"regex\":\n\t\tresult = a.assertRegex(assertion, output)\n\tcase \"type\":\n\t\tresult = a.assertType(assertion, output)\n\tcase \"script\":\n\t\tresult = a.assertScript(assertion, output, input)\n\tcase \"agent\":\n\t\tresult = a.assertAgent(assertion, output, input)\n\tdefault:\n\t\tresult.Passed = false\n\t\tresult.Message = fmt.Sprintf(\"unknown assertion type: %s\", assertion.Type)\n\t}\n\n\t// Apply negate\n\tif assertion.Negate {\n\t\tresult.Passed = !result.Passed\n\t\tif result.Passed {\n\t\t\tresult.Message = \"negated assertion passed\"\n\t\t} else {\n\t\t\tresult.Message = \"negated: \" + result.Message\n\t\t}\n\t}\n\n\treturn result\n}\n\n// assertEquals checks for exact equality\nfunc (a *Asserter) assertEquals(assertion *Assertion, output interface{}) *Result {\n\tresult := &Result{\n\t\tAssertion: assertion,\n\t\tActual: output,\n\t\tExpected: assertion.Value,\n\t}\n\n\tif ValidateOutput(output, assertion.Value) {\n\t\tresult.Passed = true\n\t\tresult.Message = \"values are equal\"\n\t} else {\n\t\tresult.Passed = false\n\t\tresult.Message = fmt.Sprintf(\"expected %v, got %v\", assertion.Value, output)\n\t}\n\n\treturn result\n}\n\n// assertContains checks if output contains the expected value\nfunc (a *Asserter) assertContains(assertion *Assertion, output interface{}) *Result {\n\tresult := &Result{\n\t\tAssertion: assertion,\n\t\tActual: output,\n\t\tExpected: assertion.Value,\n\t}\n\n\toutputStr := ToString(output)\n\texpectedStr := ToString(assertion.Value)\n\n\tif strings.Contains(outputStr, expectedStr) {\n\t\tresult.Passed = true\n\t\tresult.Message = fmt.Sprintf(\"output contains '%s'\", expectedStr)\n\t} else {\n\t\tresult.Passed = false\n\t\tresult.Message = fmt.Sprintf(\"output does not contain '%s'\", expectedStr)\n\t}\n\n\treturn result\n}\n\n// assertNotContains checks if output does not contain the expected value\nfunc (a *Asserter) assertNotContains(assertion *Assertion, output interface{}) *Result {\n\tresult := a.assertContains(assertion, output)\n\tresult.Passed = !result.Passed\n\tif result.Passed {\n\t\tresult.Message = fmt.Sprintf(\"output does not contain '%s'\", ToString(assertion.Value))\n\t} else {\n\t\tresult.Message = fmt.Sprintf(\"output should not contain '%s'\", ToString(assertion.Value))\n\t}\n\treturn result\n}\n\n// assertJSONPath extracts a value using JSON path and compares\nfunc (a *Asserter) assertJSONPath(assertion *Assertion, output interface{}) *Result {\n\tresult := &Result{\n\t\tAssertion: assertion,\n\t\tExpected: assertion.Value,\n\t}\n\n\t// Convert output to JSON if needed\n\tvar jsonData interface{}\n\tswitch v := output.(type) {\n\tcase string:\n\t\textracted := text.ExtractJSON(v)\n\t\tif extracted != nil {\n\t\t\tjsonData = extracted\n\t\t} else {\n\t\t\tresult.Passed = false\n\t\t\tresult.Message = fmt.Sprintf(\"output is not valid JSON: %s\", TruncateOutput(v, 100))\n\t\t\treturn result\n\t\t}\n\tcase map[string]interface{}, []interface{}:\n\t\tjsonData = v\n\tdefault:\n\t\tresult.Passed = false\n\t\tresult.Message = fmt.Sprintf(\"output is not a JSON object or array, got: %T\", output)\n\t\treturn result\n\t}\n\n\t// Extract value using path\n\tpath := strings.TrimPrefix(assertion.Path, \"$.\")\n\tactual := ExtractPath(jsonData, path)\n\tresult.Actual = actual\n\n\t// Compare\n\tif ValidateOutput(actual, assertion.Value) {\n\t\tresult.Passed = true\n\t\tresult.Message = fmt.Sprintf(\"path '%s' equals expected value\", assertion.Path)\n\t\treturn result\n\t}\n\n\t// IN semantics: if expected is array, check if actual matches any element\n\tif expectedArr, ok := assertion.Value.([]interface{}); ok {\n\t\tif _, actualIsArr := actual.([]interface{}); !actualIsArr {\n\t\t\tfor _, expectedItem := range expectedArr {\n\t\t\t\tif ValidateOutput(actual, expectedItem) {\n\t\t\t\t\tresult.Passed = true\n\t\t\t\t\tresult.Message = fmt.Sprintf(\"path '%s' equals one of expected values\", assertion.Path)\n\t\t\t\t\treturn result\n\t\t\t\t}\n\t\t\t}\n\t\t}\n\t}\n\n\tresult.Passed = false\n\tresult.Message = fmt.Sprintf(\"path '%s': expected %v, got %v\", assertion.Path, assertion.Value, actual)\n\treturn result\n}\n\n// assertRegex checks if output matches a regex pattern\nfunc (a *Asserter) assertRegex(assertion *Assertion, output interface{}) *Result {\n\tresult := &Result{\n\t\tAssertion: assertion,\n\t\tActual: output,\n\t\tExpected: assertion.Value,\n\t}\n\n\tpattern, ok := assertion.Value.(string)\n\tif !ok {\n\t\tresult.Passed = false\n\t\tresult.Message = \"regex pattern must be a string\"\n\t\treturn result\n\t}\n\n\tre, err := regexp.Compile(pattern)\n\tif err != nil {\n\t\tresult.Passed = false\n\t\tresult.Message = fmt.Sprintf(\"invalid regex pattern: %s\", err.Error())\n\t\treturn result\n\t}\n\n\toutputStr := ToString(output)\n\tif re.MatchString(outputStr) {\n\t\tresult.Passed = true\n\t\tresult.Message = fmt.Sprintf(\"output matches pattern '%s'\", pattern)\n\t} else {\n\t\tresult.Passed = false\n\t\tresult.Message = fmt.Sprintf(\"output does not match pattern '%s'\", pattern)\n\t}\n\n\treturn result\n}\n\n// assertType checks the type of the output (or a nested field if path is specified)\nfunc (a *Asserter) assertType(assertion *Assertion, output interface{}) *Result {\n\tresult := &Result{\n\t\tAssertion: assertion,\n\t\tExpected: assertion.Value,\n\t}\n\n\texpectedType, ok := assertion.Value.(string)\n\tif !ok {\n\t\tresult.Passed = false\n\t\tresult.Message = \"type assertion value must be a string\"\n\t\treturn result\n\t}\n\n\t// If path is specified, extract the value first\n\tvar valueToCheck interface{} = output\n\tif assertion.Path != \"\" {\n\t\t// Convert output to JSON if needed\n\t\tvar jsonData interface{}\n\t\tswitch v := output.(type) {\n\t\tcase string:\n\t\t\textracted := text.ExtractJSON(v)\n\t\t\tif extracted != nil {\n\t\t\t\tjsonData = extracted\n\t\t\t} else {\n\t\t\t\tresult.Passed = false\n\t\t\t\tresult.Message = fmt.Sprintf(\"output is not valid JSON: %s\", TruncateOutput(v, 100))\n\t\t\t\treturn result\n\t\t\t}\n\t\tcase map[string]interface{}, []interface{}:\n\t\t\tjsonData = v\n\t\tdefault:\n\t\t\tresult.Passed = false\n\t\t\tresult.Message = fmt.Sprintf(\"output is not a JSON object or array, got: %T\", output)\n\t\t\treturn result\n\t\t}\n\n\t\t// Extract value using path\n\t\tpath := strings.TrimPrefix(assertion.Path, \"$.\")\n\t\tvalueToCheck = ExtractPath(jsonData, path)\n\t\tif valueToCheck == nil {\n\t\t\tresult.Passed = false\n\t\t\tresult.Actual = nil\n\t\t\tresult.Message = fmt.Sprintf(\"path '%s' not found in output\", assertion.Path)\n\t\t\treturn result\n\t\t}\n\t}\n\n\tresult.Actual = valueToCheck\n\tactualType := GetType(valueToCheck)\n\n\tif actualType == expectedType {\n\t\tresult.Passed = true\n\t\tif assertion.Path != \"\" {\n\t\t\tresult.Message = fmt.Sprintf(\"path '%s' is of type '%s'\", assertion.Path, expectedType)\n\t\t} else {\n\t\t\tresult.Message = fmt.Sprintf(\"output is of type '%s'\", expectedType)\n\t\t}\n\t} else {\n\t\tresult.Passed = false\n\t\tif assertion.Path != \"\" {\n\t\t\tresult.Message = fmt.Sprintf(\"path '%s': expected type '%s', got '%s'\", assertion.Path, expectedType, actualType)\n\t\t} else {\n\t\t\tresult.Message = fmt.Sprintf(\"expected type '%s', got '%s'\", expectedType, actualType)\n\t\t}\n\t}\n\n\treturn result\n}\n\n// assertScript runs a custom assertion script\nfunc (a *Asserter) assertScript(assertion *Assertion, output, input interface{}) *Result {\n\tresult := &Result{\n\t\tAssertion: assertion,\n\t\tActual: output,\n\t}\n\n\tif a.scriptRunner == nil {\n\t\tresult.Passed = false\n\t\tresult.Message = \"script assertions require a ScriptRunner to be configured\"\n\t\treturn result\n\t}\n\n\tscriptName := assertion.Script\n\tif scriptName == \"\" {\n\t\tresult.Passed = false\n\t\tresult.Message = \"script assertion requires a script name\"\n\t\treturn result\n\t}\n\n\tpassed, message, err := a.scriptRunner.Run(scriptName, output, input, assertion.Value)\n\tif err != nil {\n\t\tresult.Passed = false\n\t\tresult.Message = fmt.Sprintf(\"script execution failed: %s\", err.Error())\n\t\treturn result\n\t}\n\n\tresult.Passed = passed\n\tresult.Message = message\n\treturn result\n}\n\n// assertAgent uses an agent to validate the output\nfunc (a *Asserter) assertAgent(assertion *Assertion, output, input interface{}) *Result {\n\tresult := &Result{\n\t\tAssertion: assertion,\n\t\tActual: output,\n\t}\n\n\tif a.agentValidator == nil {\n\t\tresult.Passed = false\n\t\tresult.Message = \"agent assertions require an AgentValidator to be configured\"\n\t\treturn result\n\t}\n\n\t// Parse use field: \"agents:validator\"\n\tif !strings.HasPrefix(assertion.Use, \"agents:\") {\n\t\tresult.Passed = false\n\t\tresult.Message = \"agent assertion requires 'use' field with 'agents:' prefix\"\n\t\treturn result\n\t}\n\n\tagentID := strings.TrimPrefix(assertion.Use, \"agents:\")\n\treturn a.agentValidator.Validate(agentID, output, input, assertion.Value, assertion.Options)\n}\n\n// ParseAssertions parses assertion definitions into Assertion objects\nfunc ParseAssertions(input interface{}) []*Assertion {\n\tif input == nil {\n\t\treturn nil\n\t}\n\n\tvar assertions []*Assertion\n\n\tswitch v := input.(type) {\n\tcase map[string]interface{}:\n\t\tassertion := mapToAssertion(v)\n\t\tif assertion != nil {\n\t\t\tassertions = append(assertions, assertion)\n\t\t}\n\n\tcase []interface{}:\n\t\tfor _, item := range v {\n\t\t\tif m, ok := item.(map[string]interface{}); ok {\n\t\t\t\tassertion := mapToAssertion(m)\n\t\t\t\tif assertion != nil {\n\t\t\t\t\tassertions = append(assertions, assertion)\n\t\t\t\t}\n\t\t\t}\n\t\t}\n\n\tcase string:\n\t\tassertions = append(assertions, &Assertion{Type: v})\n\t}\n\n\treturn assertions\n}\n\n// mapToAssertion converts a map to an Assertion\nfunc mapToAssertion(m map[string]interface{}) *Assertion {\n\tassertion := &Assertion{}\n\n\tif t, ok := m[\"type\"].(string); ok {\n\t\tassertion.Type = t\n\t}\n\tif v, ok := m[\"value\"]; ok {\n\t\tassertion.Value = v\n\t}\n\tif p, ok := m[\"path\"].(string); ok {\n\t\tassertion.Path = p\n\t}\n\tif s, ok := m[\"script\"].(string); ok {\n\t\tassertion.Script = s\n\t}\n\tif u, ok := m[\"use\"].(string); ok {\n\t\tassertion.Use = u\n\t}\n\tif msg, ok := m[\"message\"].(string); ok {\n\t\tassertion.Message = msg\n\t}\n\tif n, ok := m[\"negate\"].(bool); ok {\n\t\tassertion.Negate = n\n\t}\n\n\tif opts, ok := m[\"options\"].(map[string]interface{}); ok {\n\t\tassertion.Options = &AssertionOptions{}\n\t\tif c, ok := opts[\"connector\"].(string); ok {\n\t\t\tassertion.Options.Connector = c\n\t\t}\n\t\tif meta, ok := opts[\"metadata\"].(map[string]interface{}); ok {\n\t\t\tassertion.Options.Metadata = meta\n\t\t}\n\t}\n\n\treturn assertion\n}\n" }, { "path": "assert/asserter_test.go", "content": "package assert\n\nimport (\n\t\"errors\"\n\t\"testing\"\n)\n\nfunc TestAsserterEquals(t *testing.T) {\n\ta := New()\n\n\ttests := []struct {\n\t\tname string\n\t\tvalue interface{}\n\t\toutput interface{}\n\t\texpected bool\n\t}{\n\t\t{\"string match\", \"hello\", \"hello\", true},\n\t\t{\"string mismatch\", \"hello\", \"world\", false},\n\t\t{\"number match\", 42, 42, true},\n\t\t{\"number mismatch\", 42, 43, false},\n\t\t{\"bool match\", true, true, true},\n\t\t{\"bool mismatch\", true, false, false},\n\t\t{\"map match\", map[string]interface{}{\"a\": 1}, map[string]interface{}{\"a\": 1}, true},\n\t\t{\"map mismatch\", map[string]interface{}{\"a\": 1}, map[string]interface{}{\"a\": 2}, false},\n\t}\n\n\tfor _, tt := range tests {\n\t\tt.Run(tt.name, func(t *testing.T) {\n\t\t\tassertion := &Assertion{\n\t\t\t\tType: \"equals\",\n\t\t\t\tValue: tt.value,\n\t\t\t}\n\t\t\tresult := a.Evaluate(assertion, tt.output, nil)\n\t\t\tif result.Passed != tt.expected {\n\t\t\t\tt.Errorf(\"expected passed=%v, got passed=%v\", tt.expected, result.Passed)\n\t\t\t}\n\t\t})\n\t}\n}\n\nfunc TestAsserterContains(t *testing.T) {\n\ta := New()\n\n\ttests := []struct {\n\t\tname string\n\t\tvalue string\n\t\toutput string\n\t\texpected bool\n\t}{\n\t\t{\"contains substring\", \"world\", \"hello world\", true},\n\t\t{\"does not contain\", \"foo\", \"hello world\", false},\n\t\t{\"exact match\", \"hello\", \"hello\", true},\n\t\t{\"empty string\", \"\", \"hello\", true},\n\t}\n\n\tfor _, tt := range tests {\n\t\tt.Run(tt.name, func(t *testing.T) {\n\t\t\tassertion := &Assertion{\n\t\t\t\tType: \"contains\",\n\t\t\t\tValue: tt.value,\n\t\t\t}\n\t\t\tresult := a.Evaluate(assertion, tt.output, nil)\n\t\t\tif result.Passed != tt.expected {\n\t\t\t\tt.Errorf(\"expected passed=%v, got passed=%v\", tt.expected, result.Passed)\n\t\t\t}\n\t\t})\n\t}\n}\n\nfunc TestAsserterNotContains(t *testing.T) {\n\ta := New()\n\n\ttests := []struct {\n\t\tname string\n\t\tvalue string\n\t\toutput string\n\t\texpected bool\n\t}{\n\t\t{\"does not contain\", \"foo\", \"hello world\", true},\n\t\t{\"contains substring\", \"world\", \"hello world\", false},\n\t}\n\n\tfor _, tt := range tests {\n\t\tt.Run(tt.name, func(t *testing.T) {\n\t\t\tassertion := &Assertion{\n\t\t\t\tType: \"not_contains\",\n\t\t\t\tValue: tt.value,\n\t\t\t}\n\t\t\tresult := a.Evaluate(assertion, tt.output, nil)\n\t\t\tif result.Passed != tt.expected {\n\t\t\t\tt.Errorf(\"expected passed=%v, got passed=%v\", tt.expected, result.Passed)\n\t\t\t}\n\t\t})\n\t}\n}\n\nfunc TestAsserterJSONPath(t *testing.T) {\n\ta := New()\n\n\toutput := map[string]interface{}{\n\t\t\"name\": \"test\",\n\t\t\"count\": 42,\n\t\t\"nested\": map[string]interface{}{\n\t\t\t\"value\": \"deep\",\n\t\t},\n\t\t\"items\": []interface{}{\"a\", \"b\", \"c\"},\n\t}\n\n\ttests := []struct {\n\t\tname string\n\t\tpath string\n\t\tvalue interface{}\n\t\texpected bool\n\t}{\n\t\t{\"simple field\", \"name\", \"test\", true},\n\t\t{\"number field\", \"count\", float64(42), true},\n\t\t{\"nested field\", \"nested.value\", \"deep\", true},\n\t\t{\"array index\", \"items[0]\", \"a\", true},\n\t\t{\"array index 2\", \"items[2]\", \"c\", true},\n\t\t{\"wrong value\", \"name\", \"wrong\", false},\n\t\t{\"non-existent path\", \"missing\", nil, true},\n\t}\n\n\tfor _, tt := range tests {\n\t\tt.Run(tt.name, func(t *testing.T) {\n\t\t\tassertion := &Assertion{\n\t\t\t\tType: \"json_path\",\n\t\t\t\tPath: tt.path,\n\t\t\t\tValue: tt.value,\n\t\t\t}\n\t\t\tresult := a.Evaluate(assertion, output, nil)\n\t\t\tif result.Passed != tt.expected {\n\t\t\t\tt.Errorf(\"expected passed=%v, got passed=%v, message=%s\", tt.expected, result.Passed, result.Message)\n\t\t\t}\n\t\t})\n\t}\n}\n\nfunc TestAsserterRegex(t *testing.T) {\n\ta := New()\n\n\ttests := []struct {\n\t\tname string\n\t\tpattern string\n\t\toutput string\n\t\texpected bool\n\t}{\n\t\t{\"simple match\", \"hello\", \"hello world\", true},\n\t\t{\"regex pattern\", \"^\\\\d+$\", \"12345\", true},\n\t\t{\"regex no match\", \"^\\\\d+$\", \"abc\", false},\n\t\t{\"email pattern\", `\\w+@\\w+\\.\\w+`, \"test@example.com\", true},\n\t}\n\n\tfor _, tt := range tests {\n\t\tt.Run(tt.name, func(t *testing.T) {\n\t\t\tassertion := &Assertion{\n\t\t\t\tType: \"regex\",\n\t\t\t\tValue: tt.pattern,\n\t\t\t}\n\t\t\tresult := a.Evaluate(assertion, tt.output, nil)\n\t\t\tif result.Passed != tt.expected {\n\t\t\t\tt.Errorf(\"expected passed=%v, got passed=%v\", tt.expected, result.Passed)\n\t\t\t}\n\t\t})\n\t}\n}\n\nfunc TestAsserterType(t *testing.T) {\n\ta := New()\n\n\ttests := []struct {\n\t\tname string\n\t\texpectedType string\n\t\toutput interface{}\n\t\texpected bool\n\t}{\n\t\t{\"string type\", \"string\", \"hello\", true},\n\t\t{\"number type\", \"number\", 42, true},\n\t\t{\"number type float\", \"number\", 3.14, true},\n\t\t{\"boolean type\", \"boolean\", true, true},\n\t\t{\"array type\", \"array\", []interface{}{1, 2, 3}, true},\n\t\t{\"object type\", \"object\", map[string]interface{}{\"a\": 1}, true},\n\t\t{\"null type\", \"null\", nil, true},\n\t\t{\"wrong type\", \"string\", 42, false},\n\t}\n\n\tfor _, tt := range tests {\n\t\tt.Run(tt.name, func(t *testing.T) {\n\t\t\tassertion := &Assertion{\n\t\t\t\tType: \"type\",\n\t\t\t\tValue: tt.expectedType,\n\t\t\t}\n\t\t\tresult := a.Evaluate(assertion, tt.output, nil)\n\t\t\tif result.Passed != tt.expected {\n\t\t\t\tt.Errorf(\"expected passed=%v, got passed=%v\", tt.expected, result.Passed)\n\t\t\t}\n\t\t})\n\t}\n}\n\nfunc TestAsserterTypeWithPath(t *testing.T) {\n\ta := New()\n\n\t// Test data with nested structure\n\toutput := map[string]interface{}{\n\t\t\"name\": \"test\",\n\t\t\"count\": float64(42),\n\t\t\"items\": []interface{}{\"a\", \"b\", \"c\"},\n\t\t\"enabled\": true,\n\t\t\"nested\": map[string]interface{}{\n\t\t\t\"value\": \"nested_value\",\n\t\t},\n\t}\n\n\ttests := []struct {\n\t\tname string\n\t\tpath string\n\t\texpectedType string\n\t\texpected bool\n\t}{\n\t\t{\"string field\", \"name\", \"string\", true},\n\t\t{\"number field\", \"count\", \"number\", true},\n\t\t{\"array field\", \"items\", \"array\", true},\n\t\t{\"boolean field\", \"enabled\", \"boolean\", true},\n\t\t{\"object field\", \"nested\", \"object\", true},\n\t\t{\"nested string field\", \"nested.value\", \"string\", true},\n\t\t{\"wrong type for field\", \"name\", \"number\", false},\n\t\t{\"non-existent path\", \"missing\", \"string\", false},\n\t}\n\n\tfor _, tt := range tests {\n\t\tt.Run(tt.name, func(t *testing.T) {\n\t\t\tassertion := &Assertion{\n\t\t\t\tType: \"type\",\n\t\t\t\tPath: tt.path,\n\t\t\t\tValue: tt.expectedType,\n\t\t\t}\n\t\t\tresult := a.Evaluate(assertion, output, nil)\n\t\t\tif result.Passed != tt.expected {\n\t\t\t\tt.Errorf(\"expected passed=%v, got passed=%v, message=%s\", tt.expected, result.Passed, result.Message)\n\t\t\t}\n\t\t})\n\t}\n}\n\nfunc TestAsserterNegate(t *testing.T) {\n\ta := New()\n\n\t// Test negation\n\tassertion := &Assertion{\n\t\tType: \"equals\",\n\t\tValue: \"hello\",\n\t\tNegate: true,\n\t}\n\n\t// Should fail because \"hello\" == \"hello\", but negate inverts it\n\tresult := a.Evaluate(assertion, \"hello\", nil)\n\tif result.Passed {\n\t\tt.Error(\"negated equals should fail when values match\")\n\t}\n\n\t// Should pass because \"hello\" != \"world\", and negate inverts it\n\tresult = a.Evaluate(assertion, \"world\", nil)\n\tif !result.Passed {\n\t\tt.Error(\"negated equals should pass when values don't match\")\n\t}\n}\n\nfunc TestAsserterValidate(t *testing.T) {\n\ta := New()\n\n\tassertions := []*Assertion{\n\t\t{Type: \"type\", Value: \"object\"},\n\t\t{Type: \"json_path\", Path: \"name\", Value: \"test\"},\n\t\t{Type: \"json_path\", Path: \"count\", Value: float64(42)},\n\t}\n\n\toutput := map[string]interface{}{\n\t\t\"name\": \"test\",\n\t\t\"count\": 42,\n\t}\n\n\tpassed, message := a.Validate(assertions, output)\n\tif !passed {\n\t\tt.Errorf(\"validation should pass, got message: %s\", message)\n\t}\n\n\t// Test with failing assertion\n\tassertions = append(assertions, &Assertion{\n\t\tType: \"json_path\",\n\t\tPath: \"name\",\n\t\tValue: \"wrong\",\n\t})\n\n\tpassed, message = a.Validate(assertions, output)\n\tif passed {\n\t\tt.Error(\"validation should fail with wrong value\")\n\t}\n}\n\nfunc TestParseAssertions(t *testing.T) {\n\t// Test map input\n\tinput := map[string]interface{}{\n\t\t\"type\": \"contains\",\n\t\t\"value\": \"hello\",\n\t}\n\tassertions := ParseAssertions(input)\n\tif len(assertions) != 1 {\n\t\tt.Errorf(\"expected 1 assertion, got %d\", len(assertions))\n\t}\n\tif assertions[0].Type != \"contains\" {\n\t\tt.Errorf(\"expected type 'contains', got '%s'\", assertions[0].Type)\n\t}\n\n\t// Test array input\n\tinput2 := []interface{}{\n\t\tmap[string]interface{}{\"type\": \"equals\", \"value\": 1},\n\t\tmap[string]interface{}{\"type\": \"contains\", \"value\": \"test\"},\n\t}\n\tassertions = ParseAssertions(input2)\n\tif len(assertions) != 2 {\n\t\tt.Errorf(\"expected 2 assertions, got %d\", len(assertions))\n\t}\n\n\t// Test string input\n\tassertions = ParseAssertions(\"contains\")\n\tif len(assertions) != 1 {\n\t\tt.Errorf(\"expected 1 assertion, got %d\", len(assertions))\n\t}\n\tif assertions[0].Type != \"contains\" {\n\t\tt.Errorf(\"expected type 'contains', got '%s'\", assertions[0].Type)\n\t}\n}\n\nfunc TestExtractPath(t *testing.T) {\n\tdata := map[string]interface{}{\n\t\t\"name\": \"test\",\n\t\t\"nested\": map[string]interface{}{\n\t\t\t\"value\": \"deep\",\n\t\t},\n\t\t\"items\": []interface{}{\n\t\t\tmap[string]interface{}{\"id\": 1},\n\t\t\tmap[string]interface{}{\"id\": 2},\n\t\t},\n\t}\n\n\ttests := []struct {\n\t\tpath string\n\t\texpected interface{}\n\t}{\n\t\t{\"name\", \"test\"},\n\t\t{\"nested.value\", \"deep\"},\n\t\t{\"items[0].id\", 1},\n\t\t{\"items[1].id\", 2},\n\t\t{\"missing\", nil},\n\t\t{\"nested.missing\", nil},\n\t}\n\n\tfor _, tt := range tests {\n\t\tt.Run(tt.path, func(t *testing.T) {\n\t\t\tresult := ExtractPath(data, tt.path)\n\t\t\tif !ValidateOutput(result, tt.expected) {\n\t\t\t\tt.Errorf(\"path '%s': expected %v, got %v\", tt.path, tt.expected, result)\n\t\t\t}\n\t\t})\n\t}\n}\n\n// ============================================================================\n// Additional tests for improved coverage\n// ============================================================================\n\n// Mock implementations for testing\ntype mockScriptRunner struct {\n\tpassed bool\n\tmessage string\n\terr error\n}\n\nfunc (m *mockScriptRunner) Run(scriptName string, output, input, expected interface{}) (bool, string, error) {\n\treturn m.passed, m.message, m.err\n}\n\ntype mockAgentValidator struct {\n\tresult *Result\n}\n\nfunc (m *mockAgentValidator) Validate(agentID string, output, input, criteria interface{}, options *AssertionOptions) *Result {\n\treturn m.result\n}\n\n// Test WithAgentValidator and WithScriptRunner\nfunc TestAsserterConfiguration(t *testing.T) {\n\ta := New()\n\n\t// Test chaining\n\tmockAgent := &mockAgentValidator{}\n\tmockScript := &mockScriptRunner{}\n\n\tresult := a.WithAgentValidator(mockAgent).WithScriptRunner(mockScript)\n\n\tif result != a {\n\t\tt.Error(\"WithAgentValidator should return the same asserter for chaining\")\n\t}\n\tif a.agentValidator != mockAgent {\n\t\tt.Error(\"agentValidator should be set\")\n\t}\n\tif a.scriptRunner != mockScript {\n\t\tt.Error(\"scriptRunner should be set\")\n\t}\n}\n\n// Test ValidateWithDetails\nfunc TestAsserterValidateWithDetails(t *testing.T) {\n\ta := New()\n\n\tt.Run(\"empty assertions\", func(t *testing.T) {\n\t\tresult := a.ValidateWithDetails([]*Assertion{}, \"output\")\n\t\tif !result.Passed {\n\t\t\tt.Error(\"empty assertions should pass\")\n\t\t}\n\t})\n\n\tt.Run(\"single assertion pass\", func(t *testing.T) {\n\t\tresult := a.ValidateWithDetails([]*Assertion{\n\t\t\t{Type: \"equals\", Value: \"hello\"},\n\t\t}, \"hello\")\n\t\tif !result.Passed {\n\t\t\tt.Error(\"single matching assertion should pass\")\n\t\t}\n\t})\n\n\tt.Run(\"single assertion fail\", func(t *testing.T) {\n\t\tresult := a.ValidateWithDetails([]*Assertion{\n\t\t\t{Type: \"equals\", Value: \"hello\"},\n\t\t}, \"world\")\n\t\tif result.Passed {\n\t\t\tt.Error(\"single non-matching assertion should fail\")\n\t\t}\n\t})\n\n\tt.Run(\"multiple assertions with custom message\", func(t *testing.T) {\n\t\tresult := a.ValidateWithDetails([]*Assertion{\n\t\t\t{Type: \"equals\", Value: \"hello\"},\n\t\t\t{Type: \"contains\", Value: \"world\", Message: \"custom failure message\"},\n\t\t}, \"hello\")\n\t\tif result.Passed {\n\t\t\tt.Error(\"should fail when one assertion fails\")\n\t\t}\n\t\tif result.Message != \"custom failure message\" {\n\t\t\tt.Errorf(\"should use custom message, got: %s\", result.Message)\n\t\t}\n\t})\n\n\tt.Run(\"multiple assertions all pass\", func(t *testing.T) {\n\t\tresult := a.ValidateWithDetails([]*Assertion{\n\t\t\t{Type: \"contains\", Value: \"hello\"},\n\t\t\t{Type: \"contains\", Value: \"world\"},\n\t\t}, \"hello world\")\n\t\tif !result.Passed {\n\t\t\tt.Error(\"all matching assertions should pass\")\n\t\t}\n\t})\n}\n\n// Test unknown assertion type\nfunc TestAsserterUnknownType(t *testing.T) {\n\ta := New()\n\n\tassertion := &Assertion{\n\t\tType: \"unknown_type\",\n\t\tValue: \"test\",\n\t}\n\tresult := a.Evaluate(assertion, \"test\", nil)\n\tif result.Passed {\n\t\tt.Error(\"unknown assertion type should fail\")\n\t}\n\tif result.Message != \"unknown assertion type: unknown_type\" {\n\t\tt.Errorf(\"unexpected message: %s\", result.Message)\n\t}\n}\n\n// Test default type (empty string = equals)\nfunc TestAsserterDefaultType(t *testing.T) {\n\ta := New()\n\n\tassertion := &Assertion{\n\t\tType: \"\", // empty = equals\n\t\tValue: \"hello\",\n\t}\n\tresult := a.Evaluate(assertion, \"hello\", nil)\n\tif !result.Passed {\n\t\tt.Error(\"empty type should default to equals\")\n\t}\n}\n\n// Test assertScript\nfunc TestAsserterScript(t *testing.T) {\n\tt.Run(\"no script runner configured\", func(t *testing.T) {\n\t\ta := New()\n\t\tassertion := &Assertion{\n\t\t\tType: \"script\",\n\t\t\tScript: \"test.script\",\n\t\t}\n\t\tresult := a.Evaluate(assertion, \"output\", nil)\n\t\tif result.Passed {\n\t\t\tt.Error(\"should fail without script runner\")\n\t\t}\n\t\tif result.Message != \"script assertions require a ScriptRunner to be configured\" {\n\t\t\tt.Errorf(\"unexpected message: %s\", result.Message)\n\t\t}\n\t})\n\n\tt.Run(\"empty script name\", func(t *testing.T) {\n\t\ta := New().WithScriptRunner(&mockScriptRunner{})\n\t\tassertion := &Assertion{\n\t\t\tType: \"script\",\n\t\t\tScript: \"\",\n\t\t}\n\t\tresult := a.Evaluate(assertion, \"output\", nil)\n\t\tif result.Passed {\n\t\t\tt.Error(\"should fail with empty script name\")\n\t\t}\n\t\tif result.Message != \"script assertion requires a script name\" {\n\t\t\tt.Errorf(\"unexpected message: %s\", result.Message)\n\t\t}\n\t})\n\n\tt.Run(\"script execution error\", func(t *testing.T) {\n\t\ta := New().WithScriptRunner(&mockScriptRunner{\n\t\t\terr: errors.New(\"execution failed\"),\n\t\t})\n\t\tassertion := &Assertion{\n\t\t\tType: \"script\",\n\t\t\tScript: \"test.script\",\n\t\t}\n\t\tresult := a.Evaluate(assertion, \"output\", nil)\n\t\tif result.Passed {\n\t\t\tt.Error(\"should fail on script error\")\n\t\t}\n\t\tif result.Message != \"script execution failed: execution failed\" {\n\t\t\tt.Errorf(\"unexpected message: %s\", result.Message)\n\t\t}\n\t})\n\n\tt.Run(\"script passes\", func(t *testing.T) {\n\t\ta := New().WithScriptRunner(&mockScriptRunner{\n\t\t\tpassed: true,\n\t\t\tmessage: \"script passed\",\n\t\t})\n\t\tassertion := &Assertion{\n\t\t\tType: \"script\",\n\t\t\tScript: \"test.script\",\n\t\t}\n\t\tresult := a.Evaluate(assertion, \"output\", nil)\n\t\tif !result.Passed {\n\t\t\tt.Error(\"should pass when script passes\")\n\t\t}\n\t\tif result.Message != \"script passed\" {\n\t\t\tt.Errorf(\"unexpected message: %s\", result.Message)\n\t\t}\n\t})\n\n\tt.Run(\"script fails\", func(t *testing.T) {\n\t\ta := New().WithScriptRunner(&mockScriptRunner{\n\t\t\tpassed: false,\n\t\t\tmessage: \"validation failed\",\n\t\t})\n\t\tassertion := &Assertion{\n\t\t\tType: \"script\",\n\t\t\tScript: \"test.script\",\n\t\t}\n\t\tresult := a.Evaluate(assertion, \"output\", nil)\n\t\tif result.Passed {\n\t\t\tt.Error(\"should fail when script fails\")\n\t\t}\n\t})\n}\n\n// Test assertAgent\nfunc TestAsserterAgent(t *testing.T) {\n\tt.Run(\"no agent validator configured\", func(t *testing.T) {\n\t\ta := New()\n\t\tassertion := &Assertion{\n\t\t\tType: \"agent\",\n\t\t\tUse: \"agents:validator\",\n\t\t}\n\t\tresult := a.Evaluate(assertion, \"output\", nil)\n\t\tif result.Passed {\n\t\t\tt.Error(\"should fail without agent validator\")\n\t\t}\n\t\tif result.Message != \"agent assertions require an AgentValidator to be configured\" {\n\t\t\tt.Errorf(\"unexpected message: %s\", result.Message)\n\t\t}\n\t})\n\n\tt.Run(\"invalid use field format\", func(t *testing.T) {\n\t\ta := New().WithAgentValidator(&mockAgentValidator{})\n\t\tassertion := &Assertion{\n\t\t\tType: \"agent\",\n\t\t\tUse: \"invalid_format\",\n\t\t}\n\t\tresult := a.Evaluate(assertion, \"output\", nil)\n\t\tif result.Passed {\n\t\t\tt.Error(\"should fail with invalid use format\")\n\t\t}\n\t\tif result.Message != \"agent assertion requires 'use' field with 'agents:' prefix\" {\n\t\t\tt.Errorf(\"unexpected message: %s\", result.Message)\n\t\t}\n\t})\n\n\tt.Run(\"agent validation passes\", func(t *testing.T) {\n\t\ta := New().WithAgentValidator(&mockAgentValidator{\n\t\t\tresult: &Result{Passed: true, Message: \"agent validated\"},\n\t\t})\n\t\tassertion := &Assertion{\n\t\t\tType: \"agent\",\n\t\t\tUse: \"agents:validator\",\n\t\t}\n\t\tresult := a.Evaluate(assertion, \"output\", nil)\n\t\tif !result.Passed {\n\t\t\tt.Error(\"should pass when agent validates\")\n\t\t}\n\t})\n\n\tt.Run(\"agent validation fails\", func(t *testing.T) {\n\t\ta := New().WithAgentValidator(&mockAgentValidator{\n\t\t\tresult: &Result{Passed: false, Message: \"agent rejected\"},\n\t\t})\n\t\tassertion := &Assertion{\n\t\t\tType: \"agent\",\n\t\t\tUse: \"agents:validator\",\n\t\t}\n\t\tresult := a.Evaluate(assertion, \"output\", nil)\n\t\tif result.Passed {\n\t\t\tt.Error(\"should fail when agent rejects\")\n\t\t}\n\t})\n}\n\n// Test assertJSONPath edge cases\nfunc TestAsserterJSONPathEdgeCases(t *testing.T) {\n\ta := New()\n\n\tt.Run(\"string output with valid JSON\", func(t *testing.T) {\n\t\tassertion := &Assertion{\n\t\t\tType: \"json_path\",\n\t\t\tPath: \"name\",\n\t\t\tValue: \"test\",\n\t\t}\n\t\tresult := a.Evaluate(assertion, `{\"name\": \"test\"}`, nil)\n\t\tif !result.Passed {\n\t\t\tt.Errorf(\"should pass with valid JSON string, message: %s\", result.Message)\n\t\t}\n\t})\n\n\tt.Run(\"string output with invalid JSON\", func(t *testing.T) {\n\t\tassertion := &Assertion{\n\t\t\tType: \"json_path\",\n\t\t\tPath: \"name\",\n\t\t\tValue: \"test\",\n\t\t}\n\t\tresult := a.Evaluate(assertion, \"not json\", nil)\n\t\tif result.Passed {\n\t\t\tt.Error(\"should fail with invalid JSON string\")\n\t\t}\n\t})\n\n\tt.Run(\"non-JSON output type\", func(t *testing.T) {\n\t\tassertion := &Assertion{\n\t\t\tType: \"json_path\",\n\t\t\tPath: \"name\",\n\t\t\tValue: \"test\",\n\t\t}\n\t\tresult := a.Evaluate(assertion, 12345, nil)\n\t\tif result.Passed {\n\t\t\tt.Error(\"should fail with non-JSON type\")\n\t\t}\n\t})\n\n\tt.Run(\"IN semantics with array expected\", func(t *testing.T) {\n\t\tassertion := &Assertion{\n\t\t\tType: \"json_path\",\n\t\t\tPath: \"status\",\n\t\t\tValue: []interface{}{\"active\", \"pending\", \"completed\"},\n\t\t}\n\t\toutput := map[string]interface{}{\"status\": \"pending\"}\n\t\tresult := a.Evaluate(assertion, output, nil)\n\t\tif !result.Passed {\n\t\t\tt.Errorf(\"should pass with IN semantics, message: %s\", result.Message)\n\t\t}\n\t})\n\n\tt.Run(\"IN semantics no match\", func(t *testing.T) {\n\t\tassertion := &Assertion{\n\t\t\tType: \"json_path\",\n\t\t\tPath: \"status\",\n\t\t\tValue: []interface{}{\"active\", \"completed\"},\n\t\t}\n\t\toutput := map[string]interface{}{\"status\": \"pending\"}\n\t\tresult := a.Evaluate(assertion, output, nil)\n\t\tif result.Passed {\n\t\t\tt.Error(\"should fail when value not in expected array\")\n\t\t}\n\t})\n\n\tt.Run(\"path with $. prefix\", func(t *testing.T) {\n\t\tassertion := &Assertion{\n\t\t\tType: \"json_path\",\n\t\t\tPath: \"$.name\",\n\t\t\tValue: \"test\",\n\t\t}\n\t\toutput := map[string]interface{}{\"name\": \"test\"}\n\t\tresult := a.Evaluate(assertion, output, nil)\n\t\tif !result.Passed {\n\t\t\tt.Errorf(\"should handle $. prefix, message: %s\", result.Message)\n\t\t}\n\t})\n\n\tt.Run(\"array output\", func(t *testing.T) {\n\t\tassertion := &Assertion{\n\t\t\tType: \"json_path\",\n\t\t\tPath: \"[0]\",\n\t\t\tValue: \"first\",\n\t\t}\n\t\toutput := []interface{}{\"first\", \"second\"}\n\t\tresult := a.Evaluate(assertion, output, nil)\n\t\tif !result.Passed {\n\t\t\tt.Errorf(\"should work with array output, message: %s\", result.Message)\n\t\t}\n\t})\n}\n\n// Test assertRegex edge cases\nfunc TestAsserterRegexEdgeCases(t *testing.T) {\n\ta := New()\n\n\tt.Run(\"non-string pattern\", func(t *testing.T) {\n\t\tassertion := &Assertion{\n\t\t\tType: \"regex\",\n\t\t\tValue: 12345, // not a string\n\t\t}\n\t\tresult := a.Evaluate(assertion, \"test\", nil)\n\t\tif result.Passed {\n\t\t\tt.Error(\"should fail with non-string pattern\")\n\t\t}\n\t\tif result.Message != \"regex pattern must be a string\" {\n\t\t\tt.Errorf(\"unexpected message: %s\", result.Message)\n\t\t}\n\t})\n\n\tt.Run(\"invalid regex pattern\", func(t *testing.T) {\n\t\tassertion := &Assertion{\n\t\t\tType: \"regex\",\n\t\t\tValue: \"[invalid\",\n\t\t}\n\t\tresult := a.Evaluate(assertion, \"test\", nil)\n\t\tif result.Passed {\n\t\t\tt.Error(\"should fail with invalid regex\")\n\t\t}\n\t})\n}\n\n// Test assertType edge cases\nfunc TestAsserterTypeEdgeCases(t *testing.T) {\n\ta := New()\n\n\tt.Run(\"non-string type value\", func(t *testing.T) {\n\t\tassertion := &Assertion{\n\t\t\tType: \"type\",\n\t\t\tValue: 12345, // not a string\n\t\t}\n\t\tresult := a.Evaluate(assertion, \"test\", nil)\n\t\tif result.Passed {\n\t\t\tt.Error(\"should fail with non-string type value\")\n\t\t}\n\t\tif result.Message != \"type assertion value must be a string\" {\n\t\t\tt.Errorf(\"unexpected message: %s\", result.Message)\n\t\t}\n\t})\n\n\tt.Run(\"type with path from JSON string\", func(t *testing.T) {\n\t\tassertion := &Assertion{\n\t\t\tType: \"type\",\n\t\t\tPath: \"items\",\n\t\t\tValue: \"array\",\n\t\t}\n\t\tresult := a.Evaluate(assertion, `{\"items\": [1, 2, 3]}`, nil)\n\t\tif !result.Passed {\n\t\t\tt.Errorf(\"should pass with JSON string input, message: %s\", result.Message)\n\t\t}\n\t})\n\n\tt.Run(\"type with path from invalid JSON string\", func(t *testing.T) {\n\t\tassertion := &Assertion{\n\t\t\tType: \"type\",\n\t\t\tPath: \"items\",\n\t\t\tValue: \"array\",\n\t\t}\n\t\tresult := a.Evaluate(assertion, \"not json\", nil)\n\t\tif result.Passed {\n\t\t\tt.Error(\"should fail with invalid JSON string\")\n\t\t}\n\t})\n\n\tt.Run(\"type with path from non-JSON type\", func(t *testing.T) {\n\t\tassertion := &Assertion{\n\t\t\tType: \"type\",\n\t\t\tPath: \"items\",\n\t\t\tValue: \"array\",\n\t\t}\n\t\tresult := a.Evaluate(assertion, 12345, nil)\n\t\tif result.Passed {\n\t\t\tt.Error(\"should fail with non-JSON type\")\n\t\t}\n\t})\n\n\tt.Run(\"type with path from array\", func(t *testing.T) {\n\t\tassertion := &Assertion{\n\t\t\tType: \"type\",\n\t\t\tPath: \"[0]\",\n\t\t\tValue: \"string\",\n\t\t}\n\t\tresult := a.Evaluate(assertion, []interface{}{\"hello\"}, nil)\n\t\tif !result.Passed {\n\t\t\tt.Errorf(\"should work with array, message: %s\", result.Message)\n\t\t}\n\t})\n}\n\n// Test Validate with custom message\nfunc TestAsserterValidateWithCustomMessage(t *testing.T) {\n\ta := New()\n\n\tassertions := []*Assertion{\n\t\t{Type: \"equals\", Value: \"expected\", Message: \"custom failure\"},\n\t}\n\n\tpassed, message := a.Validate(assertions, \"actual\")\n\tif passed {\n\t\tt.Error(\"should fail\")\n\t}\n\tif message != \"custom failure\" {\n\t\tt.Errorf(\"should use custom message, got: %s\", message)\n\t}\n}\n\n// Test ParseAssertions edge cases\nfunc TestParseAssertionsEdgeCases(t *testing.T) {\n\tt.Run(\"nil input\", func(t *testing.T) {\n\t\tresult := ParseAssertions(nil)\n\t\tif result != nil {\n\t\t\tt.Error(\"nil input should return nil\")\n\t\t}\n\t})\n\n\tt.Run(\"array with non-map items\", func(t *testing.T) {\n\t\tinput := []interface{}{\n\t\t\t\"string item\",\n\t\t\tmap[string]interface{}{\"type\": \"equals\"},\n\t\t}\n\t\tresult := ParseAssertions(input)\n\t\tif len(result) != 1 {\n\t\t\tt.Errorf(\"should only parse map items, got %d\", len(result))\n\t\t}\n\t})\n\n\tt.Run(\"map with all fields\", func(t *testing.T) {\n\t\tinput := map[string]interface{}{\n\t\t\t\"type\": \"agent\",\n\t\t\t\"value\": \"criteria\",\n\t\t\t\"path\": \"$.field\",\n\t\t\t\"script\": \"test.script\",\n\t\t\t\"use\": \"agents:validator\",\n\t\t\t\"message\": \"custom message\",\n\t\t\t\"negate\": true,\n\t\t\t\"options\": map[string]interface{}{\n\t\t\t\t\"connector\": \"openai\",\n\t\t\t\t\"metadata\": map[string]interface{}{\"key\": \"value\"},\n\t\t\t},\n\t\t}\n\t\tresult := ParseAssertions(input)\n\t\tif len(result) != 1 {\n\t\t\tt.Fatalf(\"expected 1 assertion, got %d\", len(result))\n\t\t}\n\t\ta := result[0]\n\t\tif a.Type != \"agent\" {\n\t\t\tt.Errorf(\"type mismatch: %s\", a.Type)\n\t\t}\n\t\tif a.Path != \"$.field\" {\n\t\t\tt.Errorf(\"path mismatch: %s\", a.Path)\n\t\t}\n\t\tif a.Script != \"test.script\" {\n\t\t\tt.Errorf(\"script mismatch: %s\", a.Script)\n\t\t}\n\t\tif a.Use != \"agents:validator\" {\n\t\t\tt.Errorf(\"use mismatch: %s\", a.Use)\n\t\t}\n\t\tif a.Message != \"custom message\" {\n\t\t\tt.Errorf(\"message mismatch: %s\", a.Message)\n\t\t}\n\t\tif !a.Negate {\n\t\t\tt.Error(\"negate should be true\")\n\t\t}\n\t\tif a.Options == nil {\n\t\t\tt.Fatal(\"options should not be nil\")\n\t\t}\n\t\tif a.Options.Connector != \"openai\" {\n\t\t\tt.Errorf(\"connector mismatch: %s\", a.Options.Connector)\n\t\t}\n\t\tif a.Options.Metadata[\"key\"] != \"value\" {\n\t\t\tt.Error(\"metadata mismatch\")\n\t\t}\n\t})\n}\n\n// Test helper functions\nfunc TestToString(t *testing.T) {\n\ttests := []struct {\n\t\tname string\n\t\tinput interface{}\n\t\texpected string\n\t}{\n\t\t{\"nil\", nil, \"\"},\n\t\t{\"string\", \"hello\", \"hello\"},\n\t\t{\"bytes\", []byte(\"hello\"), \"hello\"},\n\t\t{\"number\", 42, \"42\"},\n\t\t{\"map\", map[string]interface{}{\"a\": 1}, `{\"a\":1}`},\n\t}\n\n\tfor _, tt := range tests {\n\t\tt.Run(tt.name, func(t *testing.T) {\n\t\t\tresult := ToString(tt.input)\n\t\t\tif result != tt.expected {\n\t\t\t\tt.Errorf(\"expected %s, got %s\", tt.expected, result)\n\t\t\t}\n\t\t})\n\t}\n}\n\nfunc TestGetType(t *testing.T) {\n\ttests := []struct {\n\t\tname string\n\t\tinput interface{}\n\t\texpected string\n\t}{\n\t\t{\"nil\", nil, \"null\"},\n\t\t{\"string\", \"hello\", \"string\"},\n\t\t{\"float64\", float64(3.14), \"number\"},\n\t\t{\"float32\", float32(3.14), \"number\"},\n\t\t{\"int\", 42, \"number\"},\n\t\t{\"int64\", int64(42), \"number\"},\n\t\t{\"int32\", int32(42), \"number\"},\n\t\t{\"bool\", true, \"boolean\"},\n\t\t{\"array\", []interface{}{1, 2}, \"array\"},\n\t\t{\"object\", map[string]interface{}{\"a\": 1}, \"object\"},\n\t\t{\"other\", struct{}{}, \"struct {}\"},\n\t}\n\n\tfor _, tt := range tests {\n\t\tt.Run(tt.name, func(t *testing.T) {\n\t\t\tresult := GetType(tt.input)\n\t\t\tif result != tt.expected {\n\t\t\t\tt.Errorf(\"expected %s, got %s\", tt.expected, result)\n\t\t\t}\n\t\t})\n\t}\n}\n\nfunc TestTruncateOutput(t *testing.T) {\n\ttests := []struct {\n\t\tname string\n\t\tinput interface{}\n\t\tmaxLen int\n\t\texpected string\n\t}{\n\t\t{\"nil\", nil, 10, \"\"},\n\t\t{\"short string\", \"hello\", 10, \"hello\"},\n\t\t{\"long string\", \"hello world\", 5, \"hello...\"},\n\t\t{\"object\", map[string]interface{}{\"a\": 1}, 100, `{\"a\":1}`},\n\t\t{\"long object\", map[string]interface{}{\"key\": \"value\"}, 5, `{\"key...`},\n\t}\n\n\tfor _, tt := range tests {\n\t\tt.Run(tt.name, func(t *testing.T) {\n\t\t\tresult := TruncateOutput(tt.input, tt.maxLen)\n\t\t\tif result != tt.expected {\n\t\t\t\tt.Errorf(\"expected %s, got %s\", tt.expected, result)\n\t\t\t}\n\t\t})\n\t}\n}\n\nfunc TestExtractJSON(t *testing.T) {\n\t// Test basic JSON extraction\n\tresult := ExtractJSON(`{\"name\": \"test\"}`)\n\tif result == nil {\n\t\tt.Error(\"should extract JSON\")\n\t}\n\tif m, ok := result.(map[string]interface{}); ok {\n\t\tif m[\"name\"] != \"test\" {\n\t\t\tt.Error(\"should extract correct value\")\n\t\t}\n\t} else {\n\t\tt.Error(\"should return map\")\n\t}\n}\n\nfunc TestExtractPathEdgeCases(t *testing.T) {\n\tt.Run(\"invalid array index\", func(t *testing.T) {\n\t\tdata := map[string]interface{}{\n\t\t\t\"items\": []interface{}{\"a\", \"b\"},\n\t\t}\n\t\tresult := ExtractPath(data, \"items[abc]\")\n\t\tif result != nil {\n\t\t\tt.Error(\"invalid index should return nil\")\n\t\t}\n\t})\n\n\tt.Run(\"array index on non-array\", func(t *testing.T) {\n\t\tdata := map[string]interface{}{\n\t\t\t\"name\": \"test\",\n\t\t}\n\t\tresult := ExtractPath(data, \"name[0]\")\n\t\tif result != nil {\n\t\t\tt.Error(\"array index on non-array should return nil\")\n\t\t}\n\t})\n\n\tt.Run(\"negative array index\", func(t *testing.T) {\n\t\tdata := map[string]interface{}{\n\t\t\t\"items\": []interface{}{\"a\", \"b\"},\n\t\t}\n\t\tresult := ExtractPath(data, \"items[-1]\")\n\t\tif result != nil {\n\t\t\tt.Error(\"negative index should return nil\")\n\t\t}\n\t})\n\n\tt.Run(\"out of bounds array index\", func(t *testing.T) {\n\t\tdata := map[string]interface{}{\n\t\t\t\"items\": []interface{}{\"a\", \"b\"},\n\t\t}\n\t\tresult := ExtractPath(data, \"items[99]\")\n\t\tif result != nil {\n\t\t\tt.Error(\"out of bounds index should return nil\")\n\t\t}\n\t})\n\n\tt.Run(\"field access on non-map\", func(t *testing.T) {\n\t\tdata := map[string]interface{}{\n\t\t\t\"name\": \"test\",\n\t\t}\n\t\tresult := ExtractPath(data, \"name.field\")\n\t\tif result != nil {\n\t\t\tt.Error(\"field access on non-map should return nil\")\n\t\t}\n\t})\n\n\tt.Run(\"empty path segment\", func(t *testing.T) {\n\t\tdata := map[string]interface{}{\n\t\t\t\"name\": \"test\",\n\t\t}\n\t\tresult := ExtractPath(data, \".name\")\n\t\tif result != \"test\" {\n\t\t\tt.Errorf(\"should handle leading dot, got: %v\", result)\n\t\t}\n\t})\n}\n\nfunc TestValidateOutputEdgeCases(t *testing.T) {\n\t// Test with unmarshalable types (channels, functions)\n\tch := make(chan int)\n\tresult := ValidateOutput(ch, ch)\n\tif result {\n\t\tt.Error(\"unmarshalable types should return false\")\n\t}\n}\n" }, { "path": "assert/helpers.go", "content": "package assert\n\nimport (\n\t\"encoding/json\"\n\t\"fmt\"\n\t\"strconv\"\n\t\"strings\"\n\n\tjsoniter \"github.com/json-iterator/go\"\n\t\"github.com/yaoapp/gou/text\"\n)\n\n// ValidateOutput compares two values for equality using JSON serialization\nfunc ValidateOutput(actual, expected interface{}) bool {\n\tactualJSON, err1 := jsoniter.Marshal(actual)\n\texpectedJSON, err2 := jsoniter.Marshal(expected)\n\n\tif err1 != nil || err2 != nil {\n\t\treturn false\n\t}\n\n\treturn string(actualJSON) == string(expectedJSON)\n}\n\n// ToString converts a value to string for comparison\nfunc ToString(v interface{}) string {\n\tif v == nil {\n\t\treturn \"\"\n\t}\n\n\tswitch val := v.(type) {\n\tcase string:\n\t\treturn val\n\tcase []byte:\n\t\treturn string(val)\n\tdefault:\n\t\tb, err := json.Marshal(v)\n\t\tif err != nil {\n\t\t\treturn fmt.Sprintf(\"%v\", v)\n\t\t}\n\t\treturn string(b)\n\t}\n}\n\n// GetType returns the type name of a value\nfunc GetType(v interface{}) string {\n\tif v == nil {\n\t\treturn \"null\"\n\t}\n\n\tswitch v.(type) {\n\tcase string:\n\t\treturn \"string\"\n\tcase float64, float32, int, int64, int32:\n\t\treturn \"number\"\n\tcase bool:\n\t\treturn \"boolean\"\n\tcase []interface{}:\n\t\treturn \"array\"\n\tcase map[string]interface{}:\n\t\treturn \"object\"\n\tdefault:\n\t\treturn fmt.Sprintf(\"%T\", v)\n\t}\n}\n\n// ExtractPath extracts a value from JSON using dot-notation path with array index support\n// Supports: \"field\", \"field.nested\", \"field[0]\", \"field[0].nested\", \"field.nested[0].value\"\nfunc ExtractPath(data interface{}, path string) interface{} {\n\tcurrent := data\n\n\tsegments := ParsePathSegments(path)\n\n\tfor _, segment := range segments {\n\t\tif segment == \"\" {\n\t\t\tcontinue\n\t\t}\n\n\t\t// Check if this is an array index like \"[0]\"\n\t\tif strings.HasPrefix(segment, \"[\") && strings.HasSuffix(segment, \"]\") {\n\t\t\tindexStr := segment[1 : len(segment)-1]\n\t\t\tindex, err := strconv.Atoi(indexStr)\n\t\t\tif err != nil {\n\t\t\t\treturn nil\n\t\t\t}\n\n\t\t\tarr, ok := current.([]interface{})\n\t\t\tif !ok {\n\t\t\t\treturn nil\n\t\t\t}\n\n\t\t\tif index < 0 || index >= len(arr) {\n\t\t\t\treturn nil\n\t\t\t}\n\t\t\tcurrent = arr[index]\n\t\t} else {\n\t\t\t// Regular field access\n\t\t\tswitch v := current.(type) {\n\t\t\tcase map[string]interface{}:\n\t\t\t\tcurrent = v[segment]\n\t\t\tdefault:\n\t\t\t\treturn nil\n\t\t\t}\n\t\t}\n\t}\n\n\treturn current\n}\n\n// ParsePathSegments splits a path like \"wheres[0].like\" into [\"wheres\", \"[0]\", \"like\"]\nfunc ParsePathSegments(path string) []string {\n\tvar segments []string\n\tvar current strings.Builder\n\n\tfor i := 0; i < len(path); i++ {\n\t\tch := path[i]\n\t\tswitch ch {\n\t\tcase '.':\n\t\t\tif current.Len() > 0 {\n\t\t\t\tsegments = append(segments, current.String())\n\t\t\t\tcurrent.Reset()\n\t\t\t}\n\t\tcase '[':\n\t\t\tif current.Len() > 0 {\n\t\t\t\tsegments = append(segments, current.String())\n\t\t\t\tcurrent.Reset()\n\t\t\t}\n\t\t\tj := i + 1\n\t\t\tfor j < len(path) && path[j] != ']' {\n\t\t\t\tj++\n\t\t\t}\n\t\t\tif j < len(path) {\n\t\t\t\tsegments = append(segments, path[i:j+1])\n\t\t\t\ti = j\n\t\t\t}\n\t\tdefault:\n\t\t\tcurrent.WriteByte(ch)\n\t\t}\n\t}\n\n\tif current.Len() > 0 {\n\t\tsegments = append(segments, current.String())\n\t}\n\n\treturn segments\n}\n\n// TruncateOutput truncates output for error messages\nfunc TruncateOutput(output interface{}, maxLen int) string {\n\tvar s string\n\tswitch v := output.(type) {\n\tcase string:\n\t\ts = v\n\tcase nil:\n\t\treturn \"\"\n\tdefault:\n\t\tbytes, err := jsoniter.Marshal(v)\n\t\tif err != nil {\n\t\t\ts = fmt.Sprintf(\"%v\", v)\n\t\t} else {\n\t\t\ts = string(bytes)\n\t\t}\n\t}\n\n\tif len(s) > maxLen {\n\t\treturn s[:maxLen] + \"...\"\n\t}\n\treturn s\n}\n\n// ExtractJSON extracts JSON from text (handles markdown code blocks, etc.)\nfunc ExtractJSON(content string) interface{} {\n\treturn text.ExtractJSON(content)\n}\n" }, { "path": "assert/types.go", "content": "// Package assert provides a universal assertion/validation library for Yao.\n// It can be used by agent/robot, flow, pipe, widget, and other modules.\n//\n// Design:\n// - Independent implementation (no dependency on agent/test)\n// - Supports both rule-based and semantic validation\n// - Extensible through interfaces (AgentValidator, ScriptRunner)\npackage assert\n\n// Assertion represents a single assertion rule\ntype Assertion struct {\n\t// Type is the assertion type:\n\t// - \"equals\": exact match (default if expected is set)\n\t// - \"contains\": output contains the expected string/value\n\t// - \"not_contains\": output does not contain the string/value\n\t// - \"json_path\": extract value using JSON path and compare\n\t// - \"regex\": match output against regex pattern\n\t// - \"type\": check output type (string, object, array, number, boolean)\n\t// - \"script\": run a custom assertion script (requires ScriptRunner)\n\t// - \"agent\": use an agent to validate (requires AgentValidator)\n\tType string `json:\"type\"`\n\n\t// Value is the expected value or pattern (depends on type)\n\tValue interface{} `json:\"value,omitempty\"`\n\n\t// Path is the JSON path for json_path assertions (e.g., \"$.count\", \"items[0].name\")\n\tPath string `json:\"path,omitempty\"`\n\n\t// Script is the script/process name for script assertions\n\tScript string `json:\"script,omitempty\"`\n\n\t// Use specifies the agent for validation (e.g., \"agents:validator\")\n\tUse string `json:\"use,omitempty\"`\n\n\t// Options for agent-driven assertions\n\tOptions *AssertionOptions `json:\"options,omitempty\"`\n\n\t// Message is a custom failure message\n\tMessage string `json:\"message,omitempty\"`\n\n\t// Negate inverts the assertion result\n\tNegate bool `json:\"negate,omitempty\"`\n}\n\n// AssertionOptions for agent-driven assertions\ntype AssertionOptions struct {\n\t// Connector overrides the agent's default connector\n\tConnector string `json:\"connector,omitempty\"`\n\n\t// Metadata contains custom data passed to the validator\n\tMetadata map[string]interface{} `json:\"metadata,omitempty\"`\n}\n\n// Result represents the result of an assertion\ntype Result struct {\n\t// Passed indicates whether the assertion passed\n\tPassed bool `json:\"passed\"`\n\n\t// Message describes the assertion result\n\tMessage string `json:\"message,omitempty\"`\n\n\t// Assertion is the original assertion that was evaluated\n\tAssertion *Assertion `json:\"assertion,omitempty\"`\n\n\t// Actual is the actual value that was compared\n\tActual interface{} `json:\"actual,omitempty\"`\n\n\t// Expected is the expected value\n\tExpected interface{} `json:\"expected,omitempty\"`\n}\n\n// AgentValidator is an interface for agent-based validation\n// Implementations should call an AI agent to perform semantic validation\ntype AgentValidator interface {\n\t// Validate validates output using an agent\n\t// agentID: the agent identifier (e.g., \"validator\")\n\t// output: the output to validate\n\t// input: the original input (for context)\n\t// criteria: validation criteria from assertion.Value\n\t// options: assertion options\n\tValidate(agentID string, output, input, criteria interface{}, options *AssertionOptions) *Result\n}\n\n// ScriptRunner is an interface for running assertion scripts\n// Implementations should call a Yao process to perform validation\ntype ScriptRunner interface {\n\t// Run runs an assertion script\n\t// scriptName: the script/process name\n\t// output: the output to validate\n\t// input: the original input\n\t// expected: the expected value from assertion.Value\n\t// Returns (passed, message, error)\n\tRun(scriptName string, output, input, expected interface{}) (bool, string, error)\n}\n" }, { "path": "attachment/README.md", "content": "# Attachment Package\n\nA comprehensive file upload package for Go that supports chunked uploads, file format validation, compression, and multiple storage backends.\n\n## Features\n\n- **Multiple Storage Backends**: Local filesystem and S3-compatible storage\n- **Chunked Upload Support**: Handle large files with standard HTTP Content-Range headers\n- **File Deduplication**: Content-based fingerprinting to avoid duplicate uploads\n- **File Compression**:\n - Gzip compression for any file type\n - Image compression with configurable size limits\n- **File Validation**:\n - File size limits\n - MIME type and extension validation\n - Wildcard pattern support (e.g., `image/*`, `text/*`)\n- **Flexible File Organization**: Hierarchical storage with multi-level group organization\n- **Multiple Read Methods**: Stream, bytes, and base64 encoding\n- **Global Manager Registry**: Support for registering and accessing managers globally\n- **Upload Status Tracking**: Track upload progress with status field\n- **Content Synchronization**: Support for synchronized uploads with Content-Sync header\n\n## Installation\n\n```bash\ngo get github.com/yaoapp/yao/neo/attachment\n```\n\n## Quick Start\n\n### Basic Usage\n\n```go\npackage main\n\nimport (\n \"context\"\n \"strings\"\n \"mime/multipart\"\n \"github.com/yaoapp/yao/neo/attachment\"\n)\n\nfunc main() {\n // Create a manager with default settings\n manager, err := attachment.RegisterDefault(\"uploads\")\n if err != nil {\n panic(err)\n }\n\n // Or create a custom manager\n customManager, err := attachment.New(attachment.ManagerOption{\n Driver: \"local\",\n MaxSize: \"20M\",\n ChunkSize: \"2M\",\n AllowedTypes: []string{\"text/*\", \"image/*\", \".pdf\"},\n Options: map[string]interface{}{\n \"path\": \"/var/uploads\",\n },\n })\n if err != nil {\n panic(err)\n }\n\n // Upload a file\n content := \"Hello, World!\"\n fileHeader := &attachment.FileHeader{\n FileHeader: &multipart.FileHeader{\n Filename: \"hello.txt\",\n Size: int64(len(content)),\n Header: make(map[string][]string),\n },\n }\n fileHeader.Header.Set(\"Content-Type\", \"text/plain\")\n\n option := attachment.UploadOption{\n Groups: []string{\"user123\", \"chat456\"}, // Multi-level groups (e.g., user, chat, knowledge, etc.)\n OriginalFilename: \"my_document.txt\", // Preserve original filename\n }\n\n file, err := manager.Upload(context.Background(), fileHeader, strings.NewReader(content), option)\n if err != nil {\n panic(err)\n }\n\n // Check upload status\n if file.Status == \"uploaded\" {\n fmt.Printf(\"File uploaded successfully: %s\\n\", file.ID)\n }\n\n // Read the file back\n data, err := manager.Read(context.Background(), file.ID)\n if err != nil {\n panic(err)\n }\n\n println(string(data)) // Output: Hello, World!\n}\n```\n\n### Storage Backends\n\n#### Local Storage\n\n```go\nmanager, err := attachment.New(attachment.ManagerOption{\n Driver: \"local\",\n MaxSize: \"20M\",\n Options: map[string]interface{}{\n \"path\": \"/var/uploads\",\n \"base_url\": \"https://example.com/files\",\n },\n})\n```\n\n#### S3 Storage\n\n```go\nmanager, err := attachment.New(attachment.ManagerOption{\n Driver: \"s3\",\n MaxSize: \"100M\",\n Options: map[string]interface{}{\n \"endpoint\": \"https://s3.amazonaws.com\",\n \"region\": \"us-east-1\",\n \"key\": \"your-access-key\",\n \"secret\": \"your-secret-key\",\n \"bucket\": \"your-bucket-name\",\n \"prefix\": \"attachments/\",\n },\n})\n```\n\n### Chunked Upload\n\nFor large files, you can upload in chunks using standard HTTP Content-Range headers:\n\n```go\n// Upload chunks\ntotalSize := int64(1024000) // 1MB file\nchunkSize := int64(1024) // 1KB chunks\nuid := \"unique-file-id-123\"\n\nfor start := int64(0); start < totalSize; start += chunkSize {\n end := start + chunkSize - 1\n if end >= totalSize {\n end = totalSize - 1\n }\n\n chunkData := make([]byte, end-start+1)\n // ... fill chunkData with actual data ...\n\n chunkHeader := &attachment.FileHeader{\n FileHeader: &multipart.FileHeader{\n Filename: \"large_file.zip\",\n Size: end - start + 1,\n Header: make(map[string][]string),\n },\n }\n chunkHeader.Header.Set(\"Content-Type\", \"application/zip\")\n chunkHeader.Header.Set(\"Content-Range\", fmt.Sprintf(\"bytes %d-%d/%d\", start, end, totalSize))\n chunkHeader.Header.Set(\"Content-Uid\", uid)\n\n file, err := manager.Upload(ctx, chunkHeader, bytes.NewReader(chunkData), option)\n if err != nil {\n return err\n }\n\n // File is complete when the last chunk is uploaded\n if chunkHeader.Complete() {\n fmt.Printf(\"Upload complete: %s\\n\", file.ID)\n break\n }\n}\n```\n\n### Compression\n\n#### Gzip Compression\n\n```go\noption := attachment.UploadOption{\n Gzip: true, // Enable gzip compression\n}\n\nfile, err := manager.Upload(ctx, fileHeader, reader, option)\n```\n\n#### Image Compression\n\n```go\noption := attachment.UploadOption{\n CompressImage: true,\n CompressSize: 1920, // Max dimension in pixels (default: 1920)\n}\n\nfile, err := manager.Upload(ctx, imageHeader, imageReader, option)\n```\n\n### Multi-level Groups\n\nThe `Groups` field supports hierarchical file organization:\n\n```go\n// Single level grouping\noption := attachment.UploadOption{\n Groups: []string{\"users\"},\n}\n\n// Multi-level grouping\noption := attachment.UploadOption{\n Groups: []string{\"users\", \"user123\", \"chats\", \"chat456\"},\n}\n\n// Knowledge base organization\noption := attachment.UploadOption{\n Groups: []string{\"knowledge\", \"documents\", \"technical\"},\n}\n```\n\nThis creates nested directory structures for better organization and access control.\n\n### File Validation\n\n#### Size Limits\n\n```go\nmanager, err := attachment.New(attachment.ManagerOption{\n MaxSize: \"20M\", // Maximum file size\n // Supports: B, K, M, G (e.g., \"1024B\", \"2K\", \"10M\", \"1G\")\n})\n```\n\n#### Type Validation\n\n```go\nmanager, err := attachment.New(attachment.ManagerOption{\n AllowedTypes: []string{\n \"text/*\", // All text types\n \"image/*\", // All image types\n \"application/pdf\", // Specific MIME type\n \".txt\", // File extension\n \".jpg\", // File extension\n },\n})\n```\n\n### Reading Files\n\n#### Stream Reading\n\n```go\nresponse, err := manager.Download(ctx, fileID)\nif err != nil {\n return err\n}\ndefer response.Reader.Close()\n\n// Use response.Reader as io.ReadCloser\n// response.ContentType contains the MIME type\n// response.Extension contains the file extension\n```\n\n#### Read as Bytes\n\n```go\ndata, err := manager.Read(ctx, fileID)\nif err != nil {\n return err\n}\n// data is []byte\n```\n\n#### Read as Base64\n\n```go\nbase64Data, err := manager.ReadBase64(ctx, fileID)\nif err != nil {\n return err\n}\n// base64Data is string\n```\n\n### Global Managers\n\nYou can register managers globally for easy access:\n\n```go\n// Register default manager with sensible defaults\nattachment.RegisterDefault(\"main\")\n\n// Register custom managers\nattachment.Register(\"local\", \"local\", attachment.ManagerOption{\n Driver: \"local\",\n Options: map[string]interface{}{\n \"path\": \"/var/uploads\",\n },\n})\n\nattachment.Register(\"s3\", \"s3\", attachment.ManagerOption{\n Driver: \"s3\",\n Options: map[string]interface{}{\n \"bucket\": \"my-bucket\",\n \"key\": \"access-key\",\n \"secret\": \"secret-key\",\n },\n})\n\n// Use global managers\nlocalManager := attachment.Managers[\"local\"]\ns3Manager := attachment.Managers[\"s3\"]\ndefaultManager := attachment.Managers[\"main\"]\n```\n\n## File Organization\n\nFiles are organized in a hierarchical structure:\n\n```\nattachments/\n├── 20240101/ # Date (YYYYMMDD)\n│ └── user123/ # First level group (optional)\n│ └── chat456/ # Second level group (optional)\n│ └── knowledge/ # Additional group levels (optional)\n│ └── ab/ # First 2 chars of hash\n│ └── cd/ # Next 2 chars of hash\n│ └── abcdef12.txt # Hash + extension\n```\n\nThe file ID generation includes:\n\n- Date prefix for organization\n- Multi-level groups for access control and organization\n- Content hash for deduplication\n- Original file extension\n\n## API Reference\n\n### Manager\n\n#### `New(option ManagerOption) (*Manager, error)`\n\nCreates a new attachment manager.\n\n#### `Register(name string, driver string, option ManagerOption) (*Manager, error)`\n\nRegisters a global attachment manager.\n\n#### `Upload(ctx context.Context, fileheader *FileHeader, reader io.Reader, option UploadOption) (*File, error)`\n\nUploads a file (supports chunked upload).\n\n#### `Download(ctx context.Context, fileID string) (*FileResponse, error)`\n\nDownloads a file as a stream.\n\n#### `Read(ctx context.Context, fileID string) ([]byte, error)`\n\nReads a file as bytes.\n\n#### `ReadBase64(ctx context.Context, fileID string) (string, error)`\n\nReads a file as base64 encoded string.\n\n### Storage Interface\n\nAll storage backends implement the following interface:\n\n```go\ntype Storage interface {\n Upload(ctx context.Context, fileID string, reader io.Reader, contentType string) (string, error)\n UploadChunk(ctx context.Context, fileID string, chunkIndex int, reader io.Reader, contentType string) error\n MergeChunks(ctx context.Context, fileID string, totalChunks int) error\n Download(ctx context.Context, fileID string) (io.ReadCloser, string, error)\n Reader(ctx context.Context, fileID string) (io.ReadCloser, error)\n URL(ctx context.Context, fileID string) string\n Exists(ctx context.Context, fileID string) bool\n Delete(ctx context.Context, fileID string) error\n}\n```\n\n### Types\n\n#### `ManagerOption`\n\nConfiguration for creating a manager:\n\n- `Driver`: \"local\" or \"s3\"\n- `MaxSize`: Maximum file size (e.g., \"20M\")\n- `ChunkSize`: Chunk size for uploads (e.g., \"2M\")\n- `AllowedTypes`: Array of allowed MIME types/extensions\n- `Options`: Driver-specific options\n\n#### `UploadOption`\n\nOptions for file upload:\n\n- `CompressImage`: Enable image compression\n- `CompressSize`: Maximum image dimension (default: 1920)\n- `Gzip`: Enable gzip compression\n- `Groups`: Multi-level group identifiers for hierarchical file organization (e.g., []string{\"user123\", \"chat456\", \"knowledge\"})\n- `OriginalFilename`: Original filename to preserve (avoids encoding issues)\n\n#### `File`\n\nUploaded file information:\n\n- `ID`: Unique file identifier\n- `Filename`: Original filename\n- `ContentType`: MIME type\n- `Bytes`: File size\n- `CreatedAt`: Upload timestamp\n- `Status`: Upload status (\"uploading\", \"uploaded\", \"indexing\", \"indexed\", \"upload_failed\", \"index_failed\")\n\n#### `FileResponse`\n\nDownload response:\n\n- `Reader`: io.ReadCloser for file content\n- `ContentType`: MIME type\n- `Extension`: File extension\n\n## Chunked Upload Details\n\nThe package supports chunked uploads using standard HTTP headers:\n\n- `Content-Range`: Specifies byte range (e.g., \"bytes 0-1023/2048\")\n- `Content-Uid`: Unique identifier for the file being uploaded\n\n### Chunk Index Calculation\n\nThe package uses a standard chunk size (1024 bytes by default) to calculate chunk indices consistently. This ensures proper chunk ordering during merge operations.\n\n### Content Type Preservation\n\nFor chunked uploads, the content type is preserved from the first chunk and applied to the final merged file, ensuring proper MIME type handling across all storage backends.\n\n## Error Handling\n\nThe package returns descriptive errors for common issues:\n\n- File size exceeds limit\n- Unsupported file type\n- Storage backend errors\n- Invalid chunk information\n- Missing required configuration\n\n## Testing\n\nRun the tests:\n\n```bash\n# Run all tests\ngo test ./...\n\n# Run with S3 credentials (optional)\nexport S3_ACCESS_KEY=\"your-key\"\nexport S3_SECRET_KEY=\"your-secret\"\nexport S3_BUCKET=\"your-bucket\"\nexport S3_API=\"https://your-s3-endpoint\"\ngo test ./...\n```\n\nThe package includes comprehensive tests for:\n\n- Basic file upload/download\n- Chunked uploads with content type preservation\n- Compression (gzip and image)\n- File validation (size, type, wildcards)\n- Multiple storage backends (local and S3)\n- Error handling and edge cases\n\n### Test Coverage\n\n- **Manager Tests**: Upload, download, validation, compression\n- **Local Storage Tests**: File operations, chunked uploads, directory management\n- **S3 Storage Tests**: S3 operations, chunked uploads, presigned URLs (requires credentials)\n\n## Performance Considerations\n\n- **Chunked Uploads**: Use appropriate chunk sizes (1-5MB) for optimal performance\n- **Image Compression**: Automatically resizes large images to reduce storage costs\n- **Gzip Compression**: Reduces storage size for text-based files\n- **Content Type Detection**: Efficient MIME type detection and preservation\n\n## Security Features\n\n- **File Type Validation**: Prevents upload of unauthorized file types\n- **Size Limits**: Configurable file size restrictions\n- **Path Sanitization**: Secure file path generation\n- **Access Control**: Multi-level hierarchical file organization\n\n## License\n\nThis package is part of the Yao project and follows the same license terms.\n\n### File Deduplication with Fingerprints\n\nThe package supports file deduplication using content fingerprints:\n\n```go\n// Set a content fingerprint to enable deduplication\nfileHeader := &attachment.FileHeader{\n FileHeader: &multipart.FileHeader{\n Filename: \"document.pdf\",\n Size: fileSize,\n Header: make(map[string][]string),\n },\n}\nfileHeader.Header.Set(\"Content-Type\", \"application/pdf\")\nfileHeader.Header.Set(\"Content-Fingerprint\", \"sha256:abcdef123456\") // Content-based hash\n\nfile, err := manager.Upload(ctx, fileHeader, reader, option)\n```\n\n### Content Synchronization\n\nFor synchronized uploads across multiple clients:\n\n```go\n// Enable content synchronization\nfileHeader.Header.Set(\"Content-Sync\", \"true\")\n\n// Each client can upload the same content with the same fingerprint\n// The system will deduplicate based on the content fingerprint\n```\n\n### Chunked Upload with Enhanced Headers\n\nFor large files, you can upload in chunks using standard HTTP Content-Range headers with additional metadata:\n\n```go\n// Upload chunks with unique identifier and fingerprint\ntotalSize := int64(1024000) // 1MB file\nchunkSize := int64(1024) // 1KB chunks\nuid := \"unique-file-id-123\"\nfingerprint := \"sha256:content-hash-here\"\n\nfor start := int64(0); start < totalSize; start += chunkSize {\n end := start + chunkSize - 1\n if end >= totalSize {\n end = totalSize - 1\n }\n\n chunkData := make([]byte, end-start+1)\n // ... fill chunkData with actual data ...\n\n chunkHeader := &attachment.FileHeader{\n FileHeader: &multipart.FileHeader{\n Filename: \"large_file.zip\",\n Size: end - start + 1,\n Header: make(map[string][]string),\n },\n }\n chunkHeader.Header.Set(\"Content-Type\", \"application/zip\")\n chunkHeader.Header.Set(\"Content-Range\", fmt.Sprintf(\"bytes %d-%d/%d\", start, end, totalSize))\n chunkHeader.Header.Set(\"Content-Uid\", uid)\n chunkHeader.Header.Set(\"Content-Fingerprint\", fingerprint)\n chunkHeader.Header.Set(\"Content-Sync\", \"true\") // Enable synchronization\n\n option := attachment.UploadOption{\n Groups: []string{\"user123\", \"chat456\"}, // Multi-level groups\n OriginalFilename: \"my_large_file.zip\", // Preserve original name\n }\n\n file, err := manager.Upload(ctx, chunkHeader, bytes.NewReader(chunkData), option)\n if err != nil {\n return err\n }\n\n // Check if upload is complete\n if file.Status == \"uploaded\" {\n fmt.Printf(\"Upload complete: %s\\n\", file.ID)\n break\n } else if file.Status == \"uploading\" {\n fmt.Printf(\"Chunk uploaded, progress: %d/%d\\n\", chunkHeader.GetChunkSize(), chunkHeader.GetTotalSize())\n }\n}\n```\n\n### FileHeader Methods\n\nThe `FileHeader` type provides several utility methods:\n\n```go\n// Get unique identifier for chunked uploads\nuid := fileHeader.UID()\n\n// Get content fingerprint for deduplication\nfingerprint := fileHeader.Fingerprint()\n\n// Get byte range for chunked uploads\nrangeHeader := fileHeader.Range()\n\n// Check if synchronization is enabled\nisSync := fileHeader.Sync()\n\n// Check if this is a chunked upload\nisChunk := fileHeader.IsChunk()\n\n// Check if upload is complete (for chunked uploads)\nisComplete := fileHeader.Complete()\n\n// Get detailed chunk information\nstart, end, total, err := fileHeader.GetChunkInfo()\n\n// Get total file size (for chunked uploads)\ntotalSize := fileHeader.GetTotalSize()\n\n// Get current chunk size\nchunkSize := fileHeader.GetChunkSize()\n```\n\n## File Headers and Metadata\n\nThe package supports several HTTP headers for enhanced functionality:\n\n- `Content-Range`: Standard HTTP range header for chunked uploads (e.g., \"bytes 0-1023/2048\")\n- `Content-Uid`: Unique identifier for file uploads (for deduplication and tracking)\n- `Content-Fingerprint`: Content-based hash for deduplication (e.g., \"sha256:abc123\")\n- `Content-Sync`: Enable synchronized uploads across multiple clients (\"true\"/\"false\")\n\n### Header Processing\n\nWhen processing uploads, headers can be extracted from both HTTP request headers and multipart file headers:\n\n```go\n// Extract headers from HTTP request and file headers\nheader := attachment.GetHeader(requestHeader, fileHeader, fileSize)\n\n// The resulting FileHeader will contain merged headers from both sources\nuid := header.UID()\nfingerprint := header.Fingerprint()\nisSync := header.Sync()\n```\n\n## Upload Status Tracking\n\nFiles have a status field that tracks the upload lifecycle:\n\n- `\"uploading\"`: File upload is in progress (for chunked uploads)\n- `\"uploaded\"`: File has been successfully uploaded\n- `\"indexing\"`: File is being processed for search indexing\n- `\"indexed\"`: File has been indexed and is fully processed\n- `\"upload_failed\"`: Upload failed due to an error\n- `\"index_failed\"`: Indexing failed but file is still accessible\n\n```go\nfile, err := manager.Upload(ctx, fileHeader, reader, option)\nif err != nil {\n return err\n}\n\nswitch file.Status {\ncase \"uploading\":\n fmt.Println(\"Upload in progress...\")\ncase \"uploaded\":\n fmt.Println(\"Upload completed successfully\")\ncase \"upload_failed\":\n fmt.Println(\"Upload failed\")\n}\n```\n\n## Text Content Storage\n\nThe attachment package supports storing parsed text content extracted from files (e.g., from PDFs, Word documents, or image OCR). This is useful for building search indexes or providing text-based previews.\n\nThe system automatically maintains two versions of the text content:\n- **Full content** (`content`): Complete text, stored as longText (up to 4GB)\n- **Preview** (`content_preview`): First 2000 characters, stored as text for quick access\n\n### Saving Parsed Text Content\n\nUse `SaveText` to store the extracted text content. It automatically saves both full content and preview:\n\n```go\n// Upload a PDF file\nfile, err := manager.Upload(ctx, fileHeader, reader, option)\nif err != nil {\n return err\n}\n\n// Extract text from the PDF (using your preferred library)\nparsedText := extractTextFromPDF(file.ID)\n\n// Save the parsed text (automatically saves both full and preview)\nerr = manager.SaveText(ctx, file.ID, parsedText)\nif err != nil {\n return fmt.Errorf(\"failed to save text content: %w\", err)\n}\n```\n\n### Retrieving Parsed Text Content\n\nUse `GetText` to retrieve text content. By default, it returns the preview for better performance:\n\n```go\n// Get preview (first 2000 characters) - Fast, suitable for UI display\npreview, err := manager.GetText(ctx, file.ID)\nif err != nil {\n return fmt.Errorf(\"failed to get preview: %w\", err)\n}\n\nif preview == \"\" {\n fmt.Println(\"No text content available for this file\")\n} else {\n fmt.Printf(\"Preview (%d characters): %s\\n\", len(preview), preview)\n}\n\n// Get full content - Use only when complete text is needed (e.g., for indexing)\nfullText, err := manager.GetText(ctx, file.ID, true)\nif err != nil {\n return fmt.Errorf(\"failed to get full text: %w\", err)\n}\n\nfmt.Printf(\"Full content (%d characters)\\n\", len(fullText))\n```\n\n### Performance Optimization\n\nThe text content fields are optimized for different use cases:\n\n| Field | Size Limit | Use Case | Performance |\n|-------|------------|----------|-------------|\n| `content_preview` | 2000 chars | Quick preview, UI display, snippets | ⚡ Very Fast |\n| `content` | 4GB | Full text search, complete content | 🐌 Slow for large files |\n\n**Best Practices:**\n1. Use preview by default: `GetText(ctx, fileID)`\n2. Only request full content when necessary: `GetText(ctx, fileID, true)`\n3. Both fields are excluded from `List()` by default for optimal performance\n4. Preview uses character (rune) count, not bytes, for proper UTF-8 handling\n\n### Example: Complete Text Processing Workflow\n\n```go\n// 1. Upload file\nfile, err := manager.Upload(ctx, fileHeader, reader, option)\nif err != nil {\n return err\n}\n\n// 2. Process file based on content type\nvar parsedText string\nswitch {\ncase strings.HasPrefix(file.ContentType, \"image/\"):\n // Use OCR to extract text from image\n parsedText, err = performOCR(file.ID)\n \ncase file.ContentType == \"application/pdf\":\n // Extract text from PDF\n parsedText, err = extractPDFText(file.ID)\n \ncase strings.Contains(file.ContentType, \"wordprocessingml\"):\n // Extract text from Word document\n parsedText, err = extractWordText(file.ID)\n}\n\nif err != nil {\n return fmt.Errorf(\"failed to extract text: %w\", err)\n}\n\n// 3. Save the extracted text\nif parsedText != \"\" {\n err = manager.SaveText(ctx, file.ID, parsedText)\n if err != nil {\n return fmt.Errorf(\"failed to save text: %w\", err)\n }\n}\n\n// 4. Later, retrieve the text for search or display\nsavedText, err := manager.GetText(ctx, file.ID)\nif err != nil {\n return err\n}\n\nfmt.Printf(\"Retrieved text: %s\\n\", savedText)\n```\n\n### Text Content Features\n\n- **Dual Storage**: Automatically maintains both full content and preview (2000 chars)\n- **Size Limits**: \n - Preview: 2000 characters (text type)\n - Full content: Up to 4GB (longText type)\n- **Smart Retrieval**: Returns preview by default, full content on demand\n- **Update**: Text content can be updated at any time using `SaveText`\n- **Clear**: Set text to empty string to clear both fields\n- **UTF-8 Safe**: Preview uses character (rune) count, not bytes, ensuring proper multi-byte character handling\n- **Performance**: Both `content` and `content_preview` fields are excluded by default in `List()` and `Info()` operations to avoid loading text data. Use `GetText()` to explicitly retrieve text content when needed\n\n#### `RegisterDefault(name string) (*Manager, error)`\n\nRegisters a default attachment manager with sensible defaults for common file types.\n\n## Process API\n\nThe attachment package provides a set of Yao Process APIs for file management with built-in permission support.\n\n### Available Processes\n\n| Process | Description |\n|---------|-------------|\n| `attachment.Save` | Save a file from base64 data URI |\n| `attachment.Read` | Read file content as base64 data URI |\n| `attachment.Info` | Get file metadata |\n| `attachment.List` | List files with pagination and filtering |\n| `attachment.Delete` | Delete a file |\n| `attachment.Exists` | Check if file exists |\n| `attachment.URL` | Get file URL |\n| `attachment.SaveText` | Save parsed text content for a file |\n| `attachment.GetText` | Get parsed text content for a file |\n\n### Permission Model\n\nThe Process API integrates with Yao's `process.Authorized` mechanism:\n\n- **Authorized Info**: Reads `UserID`, `TeamID`, `TenantID` from `process.Authorized` (set by OAuth guard)\n- **Auto Permission Storage**: On save, automatically stores `__yao_created_by`, `__yao_team_id`, `__yao_tenant_id` from `process.Authorized`\n- **Data Constraints**: Respects `Constraints.OwnerOnly` and `Constraints.TeamOnly` from ACL enforcement\n- **Owner Access**: When `OwnerOnly` is set, only file creator (`__yao_created_by`) can access their files\n- **Team Access**: When `TeamOnly` is set, team members can access files with `share: \"team\"`\n- **Public Access**: Files with `public: true` are readable by everyone regardless of constraints\n- **No Constraints**: If no constraints are set, all authenticated users can access all files\n\n### Usage Examples\n\n#### JavaScript (Yao Scripts)\n\n```javascript\n// Save a file from base64 data URI\nconst file = Process(\"attachment.Save\", \"default\", \n \"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAAUA...\", \n \"photo.png\",\n { share: \"team\" }\n);\nconsole.log(\"Saved file ID:\", file.file_id);\n\n// Save text file\nconst textFile = Process(\"attachment.Save\", \"default\",\n \"data:text/plain;base64,SGVsbG8gV29ybGQh\",\n \"hello.txt\"\n);\n\n// Read file content as data URI\nconst dataURI = Process(\"attachment.Read\", \"default\", file.file_id);\n// Returns: \"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAAUA...\"\n\n// Get file info\nconst info = Process(\"attachment.Info\", \"default\", file.file_id);\n\n// List files with pagination\nconst result = Process(\"attachment.List\", \"default\", {\n page: 1,\n page_size: 20,\n filters: { status: \"uploaded\", content_type: \"image/*\" },\n order_by: \"created_at desc\"\n});\n\n// Check if file exists\nconst exists = Process(\"attachment.Exists\", \"default\", file.file_id);\n\n// Get file URL\nconst url = Process(\"attachment.URL\", \"default\", file.file_id);\n\n// Save parsed text content (e.g., OCR result, PDF text)\nProcess(\"attachment.SaveText\", \"default\", file.file_id, \"Extracted text content...\");\n\n// Get text content (preview by default)\nconst preview = Process(\"attachment.GetText\", \"default\", file.file_id);\n\n// Get full text content\nconst fullText = Process(\"attachment.GetText\", \"default\", file.file_id, true);\n\n// Delete file\nProcess(\"attachment.Delete\", \"default\", file.file_id);\n```\n\n#### Flow DSL\n\n```json\n{\n \"name\": \"Save Image\",\n \"nodes\": [\n {\n \"name\": \"save\",\n \"process\": \"attachment.Save\",\n \"args\": [\n \"default\",\n \"{{$in.dataURI}}\",\n \"{{$in.filename}}\",\n { \"share\": \"team\" }\n ]\n }\n ],\n \"output\": \"{{$res.save}}\"\n}\n```\n\n### Process Reference\n\n#### `attachment.Save`\n\nSave a file from base64 data URI. Automatically parses content type from data URI header and stores permission fields from `process.Authorized`.\n\n**Arguments:**\n1. `uploaderID` (string) - The uploader/manager ID\n2. `content` (string) - Base64 data URI (e.g., `\"data:image/png;base64,xxxx\"`) or plain base64\n3. `filename` (string, optional) - Original filename (auto-generated if not provided)\n4. `option` (map, optional) - Upload options:\n - `groups` ([]string) - Directory groups for organization\n - `gzip` (bool) - Enable gzip compression\n - `compress_image` (bool) - Enable image compression\n - `compress_size` (int) - Target image size in pixels\n - `public` (bool) - Make file publicly accessible\n - `share` (string) - Share scope: \"private\" or \"team\"\n\n**Returns:** `*File` - Saved file information\n\n**Example:**\n```javascript\n// With data URI (auto-detect content type)\nProcess(\"attachment.Save\", \"default\", \"data:image/png;base64,iVBORw0KGgo...\", \"photo.png\")\n\n// With plain base64 (defaults to application/octet-stream)\nProcess(\"attachment.Save\", \"default\", \"SGVsbG8gV29ybGQh\", \"hello.txt\")\n\n// With options\nProcess(\"attachment.Save\", \"default\", \"data:application/pdf;base64,...\", \"doc.pdf\", {\n groups: [\"documents\"],\n share: \"team\",\n public: false\n})\n```\n\n---\n\n#### `attachment.Read`\n\nRead file content as base64 data URI.\n\n**Arguments:**\n1. `uploaderID` (string) - The uploader/manager ID\n2. `fileID` (string) - The file ID\n\n**Returns:** `string` - Base64 data URI (e.g., `\"data:image/png;base64,xxxx\"`)\n\n**Example:**\n```javascript\nconst dataURI = Process(\"attachment.Read\", \"default\", \"abc123\")\n// Returns: \"data:image/png;base64,iVBORw0KGgo...\"\n```\n\n---\n\n#### `attachment.Info`\n\nGet file metadata.\n\n**Arguments:**\n1. `uploaderID` (string) - The uploader/manager ID\n2. `fileID` (string) - The file ID\n\n**Returns:** `*File` - File metadata\n\n---\n\n#### `attachment.List`\n\nList files with pagination and filtering.\n\n**Arguments:**\n1. `uploaderID` (string) - The uploader/manager ID\n2. `option` (map, optional) - List options:\n - `page` (int) - Page number (default: 1)\n - `page_size` (int) - Items per page (default: 20, max: 100)\n - `filters` (map) - Filter conditions (e.g., `{\"status\": \"uploaded\"}`)\n - `order_by` (string) - Sort order (e.g., \"created_at desc\")\n - `select` ([]string) - Fields to return\n\n**Returns:** `*ListResult` - Paginated file list\n\n---\n\n#### `attachment.Delete`\n\nDelete a file. Requires write permission (owner only).\n\n**Arguments:**\n1. `uploaderID` (string) - The uploader/manager ID\n2. `fileID` (string) - The file ID\n\n**Returns:** `bool` - Success status\n\n---\n\n#### `attachment.Exists`\n\nCheck if a file exists.\n\n**Arguments:**\n1. `uploaderID` (string) - The uploader/manager ID\n2. `fileID` (string) - The file ID\n\n**Returns:** `bool` - Whether file exists\n\n---\n\n#### `attachment.URL`\n\nGet the URL of a file.\n\n**Arguments:**\n1. `uploaderID` (string) - The uploader/manager ID\n2. `fileID` (string) - The file ID\n\n**Returns:** `string` - File URL\n\n---\n\n#### `attachment.SaveText`\n\nSave parsed text content for a file (e.g., OCR result, PDF extracted text).\n\n**Arguments:**\n1. `uploaderID` (string) - The uploader/manager ID\n2. `fileID` (string) - The file ID\n3. `text` (string) - Text content to save\n\n**Returns:** `bool` - Success status\n\n---\n\n#### `attachment.GetText`\n\nGet parsed text content for a file.\n\n**Arguments:**\n1. `uploaderID` (string) - The uploader/manager ID\n2. `fileID` (string) - The file ID\n3. `fullContent` (bool, optional) - Whether to get full content (default: false, returns preview)\n\n**Returns:** `string` - Text content\n" }, { "path": "attachment/compresses.go", "content": "package attachment\n\nimport (\n\t\"bytes\"\n\t\"fmt\"\n\t\"image\"\n\t\"image/jpeg\"\n\t\"image/png\"\n\t\"io\"\n)\n\n// CompressImage compresses the image while maintaining aspect ratio\nfunc CompressImage(reader io.Reader, contentType string, maxSize int) ([]byte, error) {\n\t// Read all data first\n\tdata, err := io.ReadAll(reader)\n\tif err != nil {\n\t\treturn nil, fmt.Errorf(\"failed to read image data: %w\", err)\n\t}\n\n\t// Decode image\n\timg, _, err := image.Decode(bytes.NewReader(data))\n\tif err != nil {\n\t\treturn nil, fmt.Errorf(\"failed to decode image: %w\", err)\n\t}\n\n\t// Calculate new dimensions\n\tbounds := img.Bounds()\n\twidth := bounds.Dx()\n\theight := bounds.Dy()\n\tvar newWidth, newHeight int\n\n\tif width > height {\n\t\tif width > maxSize {\n\t\t\tnewWidth = maxSize\n\t\t\tnewHeight = int(float64(height) * (float64(maxSize) / float64(width)))\n\t\t} else {\n\t\t\treturn data, nil // No need to resize, return original data\n\t\t}\n\t} else {\n\t\tif height > maxSize {\n\t\t\tnewHeight = maxSize\n\t\t\tnewWidth = int(float64(width) * (float64(maxSize) / float64(height)))\n\t\t} else {\n\t\t\treturn data, nil // No need to resize, return original data\n\t\t}\n\t}\n\n\t// Create new image with new dimensions\n\tnewImg := image.NewRGBA(image.Rect(0, 0, newWidth, newHeight))\n\n\t// Scale the image using bilinear interpolation\n\tfor y := 0; y < newHeight; y++ {\n\t\tfor x := 0; x < newWidth; x++ {\n\t\t\tsrcX := float64(x) * float64(width) / float64(newWidth)\n\t\t\tsrcY := float64(y) * float64(height) / float64(newHeight)\n\t\t\tnewImg.Set(x, y, img.At(int(srcX), int(srcY)))\n\t\t}\n\t}\n\n\t// Encode image\n\tvar buf bytes.Buffer\n\tswitch contentType {\n\tcase \"image/jpeg\":\n\t\terr = jpeg.Encode(&buf, newImg, &jpeg.Options{Quality: 85})\n\tcase \"image/png\":\n\t\terr = png.Encode(&buf, newImg)\n\tdefault:\n\t\treturn data, nil // Unsupported format, return original data\n\t}\n\n\tif err != nil {\n\t\treturn nil, fmt.Errorf(\"failed to encode image: %w\", err)\n\t}\n\n\treturn buf.Bytes(), nil\n}\n" }, { "path": "attachment/convert.go", "content": "package attachment\n\nimport (\n\t\"fmt\"\n\t\"strings\"\n)\n\n// toBool converts various types to boolean\nfunc toBool(v interface{}) bool {\n\tif v == nil {\n\t\treturn false\n\t}\n\n\tswitch val := v.(type) {\n\tcase bool:\n\t\treturn val\n\tcase int:\n\t\treturn val != 0\n\tcase int64:\n\t\treturn val != 0\n\tcase uint8: // MySQL tinyint(1)\n\t\treturn val != 0\n\tcase float64:\n\t\treturn val != 0\n\tcase string:\n\t\tnormalized := strings.ToLower(strings.TrimSpace(val))\n\t\tswitch normalized {\n\t\tcase \"true\", \"1\", \"enabled\", \"yes\", \"on\":\n\t\t\treturn true\n\t\tdefault:\n\t\t\treturn false\n\t\t}\n\tdefault:\n\t\treturn false\n\t}\n}\n\n// toString converts various types to string\nfunc toString(v interface{}) string {\n\tif v == nil {\n\t\treturn \"\"\n\t}\n\n\tswitch val := v.(type) {\n\tcase string:\n\t\treturn val\n\tcase int:\n\t\treturn fmt.Sprintf(\"%d\", val)\n\tcase int64:\n\t\treturn fmt.Sprintf(\"%d\", val)\n\tcase float64:\n\t\treturn fmt.Sprintf(\"%.0f\", val)\n\tcase bool:\n\t\tif val {\n\t\t\treturn \"true\"\n\t\t}\n\t\treturn \"false\"\n\tdefault:\n\t\treturn fmt.Sprintf(\"%v\", val)\n\t}\n}\n" }, { "path": "attachment/example_usage.go", "content": "package attachment\n\nimport (\n\t\"bytes\"\n\t\"context\"\n\t\"fmt\"\n\t\"log\"\n\t\"mime/multipart\"\n\t\"strings\"\n\n\t\"github.com/yaoapp/yao/attachment/s3\"\n)\n\n// ExampleUsage demonstrates how to use the attachment package\nfunc ExampleUsage() {\n\t// 1. Create a local storage manager\n\tlocalManager, err := New(ManagerOption{\n\t\tDriver: \"local\",\n\t\tMaxSize: \"20M\",\n\t\tChunkSize: \"2M\",\n\t\tAllowedTypes: []string{\"text/*\", \"image/*\", \"application/pdf\", \".txt\", \".jpg\", \".png\", \".pdf\"},\n\t\tOptions: map[string]interface{}{\n\t\t\t\"path\": \"/var/uploads\",\n\t\t\t\"base_url\": \"https://example.com/files\",\n\t\t},\n\t})\n\tif err != nil {\n\t\tlog.Fatalf(\"Failed to create local manager: %v\", err)\n\t}\n\n\t// 2. Create an S3 storage manager\n\ts3Manager, err := New(ManagerOption{\n\t\tDriver: \"s3\",\n\t\tMaxSize: \"100M\",\n\t\tChunkSize: \"5M\",\n\t\tAllowedTypes: []string{\"*\"}, // Allow all types\n\t\tOptions: map[string]interface{}{\n\t\t\t\"endpoint\": \"https://s3.amazonaws.com\",\n\t\t\t\"region\": \"us-east-1\",\n\t\t\t\"key\": \"your-access-key\",\n\t\t\t\"secret\": \"your-secret-key\",\n\t\t\t\"bucket\": \"your-bucket-name\",\n\t\t\t\"prefix\": \"attachments/\",\n\t\t},\n\t})\n\tif err != nil {\n\t\tlog.Printf(\"Failed to create S3 manager (this is expected without credentials): %v\", err)\n\t} else {\n\t\tfmt.Printf(\"Created S3 manager successfully\\n\")\n\t\t// Demonstrate S3 manager usage if credentials are available\n\t\tif s3Manager != nil {\n\t\t\tfmt.Printf(\"S3 manager is ready for use with bucket: %s\\n\",\n\t\t\t\ts3Manager.storage.(*s3.Storage).Bucket)\n\t\t}\n\t}\n\n\t// 3. Register managers globally\n\t_, err = Register(\"local\", \"local\", ManagerOption{\n\t\tDriver: \"local\",\n\t\tMaxSize: \"20M\",\n\t\tAllowedTypes: []string{\"text/*\", \"image/*\"},\n\t\tOptions: map[string]interface{}{\n\t\t\t\"path\": \"/var/uploads\",\n\t\t},\n\t})\n\tif err != nil {\n\t\tlog.Printf(\"Failed to register local manager: %v\", err)\n\t}\n\n\t// Try to register S3 manager (will fail without credentials)\n\t_, err = Register(\"s3\", \"s3\", ManagerOption{\n\t\tDriver: \"s3\",\n\t\tMaxSize: \"100M\",\n\t\tOptions: map[string]interface{}{\n\t\t\t\"bucket\": \"your-bucket\",\n\t\t\t\"key\": \"your-key\",\n\t\t\t\"secret\": \"your-secret\",\n\t\t},\n\t})\n\tif err != nil {\n\t\tlog.Printf(\"Failed to register S3 manager (expected without credentials): %v\", err)\n\t}\n\n\tctx := context.Background()\n\n\t// 4. Example: Simple file upload\n\tcontent := \"Hello, World! This is a test file with some content to demonstrate the attachment package.\"\n\tfileHeader := &FileHeader{\n\t\tFileHeader: &multipart.FileHeader{\n\t\t\tFilename: \"hello.txt\",\n\t\t\tSize: int64(len(content)),\n\t\t\tHeader: make(map[string][]string),\n\t\t},\n\t}\n\tfileHeader.Header.Set(\"Content-Type\", \"text/plain\")\n\n\tuploadOption := UploadOption{\n\t\tGroups: []string{\"user123\"},\n\t\tGzip: false, // No compression for small text files\n\t}\n\n\tfile, err := localManager.Upload(ctx, fileHeader, strings.NewReader(content), uploadOption)\n\tif err != nil {\n\t\tlog.Fatalf(\"Failed to upload file: %v\", err)\n\t}\n\n\tfmt.Printf(\"Uploaded file: %s (ID: %s, Size: %d bytes)\\n\", file.Filename, file.ID, file.Bytes)\n\n\t// 5. Example: File upload with gzip compression\n\tlargeContent := strings.Repeat(\"This is a large text file that benefits from compression. \", 100)\n\tgzipFileHeader := &FileHeader{\n\t\tFileHeader: &multipart.FileHeader{\n\t\t\tFilename: \"large_text.txt\",\n\t\t\tSize: int64(len(largeContent)),\n\t\t\tHeader: make(map[string][]string),\n\t\t},\n\t}\n\tgzipFileHeader.Header.Set(\"Content-Type\", \"text/plain\")\n\n\tgzipOption := UploadOption{\n\t\tGroups: []string{\"user123\"},\n\t\tGzip: true, // Enable compression\n\t}\n\n\tgzipFile, err := localManager.Upload(ctx, gzipFileHeader, strings.NewReader(largeContent), gzipOption)\n\tif err != nil {\n\t\tlog.Fatalf(\"Failed to upload gzipped file: %v\", err)\n\t}\n\n\tfmt.Printf(\"Uploaded compressed file: %s (ID: %s)\\n\", gzipFile.Filename, gzipFile.ID)\n\n\t// 6. Example: Image upload with compression\n\timageUploadOption := UploadOption{\n\t\tGroups: []string{\"user123\"},\n\t\tCompressImage: true,\n\t\tCompressSize: 1920, // Resize to max 1920px\n\t\tGzip: false,\n\t}\n\n\t// Simulate image upload (you would get this from multipart form)\n\timageHeader := &FileHeader{\n\t\tFileHeader: &multipart.FileHeader{\n\t\t\tFilename: \"photo.jpg\",\n\t\t\tSize: 1024000, // 1MB\n\t\t\tHeader: make(map[string][]string),\n\t\t},\n\t}\n\timageHeader.Header.Set(\"Content-Type\", \"image/jpeg\")\n\n\tfmt.Printf(\"Image upload option configured: compress=%v, size=%d\\n\",\n\t\timageUploadOption.CompressImage, imageUploadOption.CompressSize)\n\n\t// 6.5. Example: Multi-level groups\n\tfmt.Println(\"\\n--- Multi-level Groups Examples ---\")\n\n\t// Single level grouping\n\tsingleGroupOption := UploadOption{\n\t\tGroups: []string{\"knowledge\"},\n\t}\n\n\tsingleGroupHeader := &FileHeader{\n\t\tFileHeader: &multipart.FileHeader{\n\t\t\tFilename: \"knowledge_doc.txt\",\n\t\t\tSize: int64(len(\"Knowledge base document\")),\n\t\t\tHeader: make(map[string][]string),\n\t\t},\n\t}\n\tsingleGroupHeader.Header.Set(\"Content-Type\", \"text/plain\")\n\n\tsingleFile, err := localManager.Upload(ctx, singleGroupHeader, strings.NewReader(\"Knowledge base document\"), singleGroupOption)\n\tif err != nil {\n\t\tlog.Printf(\"Failed to upload single group file: %v\", err)\n\t} else {\n\t\tfmt.Printf(\"Single group file uploaded: %s (ID: %s)\\n\", singleFile.Filename, singleFile.ID)\n\t}\n\n\t// Multi-level grouping\n\tmultiGroupOption := UploadOption{\n\t\tGroups: []string{\"users\", \"user123\", \"chats\", \"chat456\", \"documents\"},\n\t}\n\n\tmultiGroupHeader := &FileHeader{\n\t\tFileHeader: &multipart.FileHeader{\n\t\t\tFilename: \"chat_document.txt\",\n\t\t\tSize: int64(len(\"Document in user chat\")),\n\t\t\tHeader: make(map[string][]string),\n\t\t},\n\t}\n\tmultiGroupHeader.Header.Set(\"Content-Type\", \"text/plain\")\n\n\tmultiFile, err := localManager.Upload(ctx, multiGroupHeader, strings.NewReader(\"Document in user chat\"), multiGroupOption)\n\tif err != nil {\n\t\tlog.Printf(\"Failed to upload multi-group file: %v\", err)\n\t} else {\n\t\tfmt.Printf(\"Multi-level group file uploaded: %s (ID: %s)\\n\", multiFile.Filename, multiFile.ID)\n\t\tfmt.Printf(\"File path includes hierarchy: users/user123/chats/chat456/documents\\n\")\n\t}\n\n\t// Knowledge base organization\n\tknowledgeOption := UploadOption{\n\t\tGroups: []string{\"knowledge\", \"technical\", \"api\", \"documentation\"},\n\t}\n\n\tknowledgeHeader := &FileHeader{\n\t\tFileHeader: &multipart.FileHeader{\n\t\t\tFilename: \"api_guide.md\",\n\t\t\tSize: int64(len(\"# API Documentation\\n\\nThis is technical documentation.\")),\n\t\t\tHeader: make(map[string][]string),\n\t\t},\n\t}\n\tknowledgeHeader.Header.Set(\"Content-Type\", \"text/markdown\")\n\n\tknowledgeFile, err := localManager.Upload(ctx, knowledgeHeader,\n\t\tstrings.NewReader(\"# API Documentation\\n\\nThis is technical documentation.\"), knowledgeOption)\n\tif err != nil {\n\t\tlog.Printf(\"Failed to upload knowledge file: %v\", err)\n\t} else {\n\t\tfmt.Printf(\"Knowledge base file uploaded: %s (ID: %s)\\n\", knowledgeFile.Filename, knowledgeFile.ID)\n\t\tfmt.Printf(\"Organized in: knowledge/technical/api/documentation\\n\")\n\t}\n\n\t// 7. Example: Chunked upload\n\tlargeContent = strings.Repeat(\"This is a large file content that will be uploaded in chunks. \", 1000)\n\tchunkSize := 1024\n\ttotalSize := len(largeContent)\n\tuid := \"unique-large-file-123\"\n\n\tfmt.Printf(\"Starting chunked upload: total size=%d, chunk size=%d\\n\", totalSize, chunkSize)\n\n\tvar lastFile *File\n\tchunkCount := 0\n\n\t// Split into chunks and upload\n\tfor i := 0; i < totalSize; i += chunkSize {\n\t\tend := i + chunkSize\n\t\tif end > totalSize {\n\t\t\tend = totalSize\n\t\t}\n\t\tchunk := largeContent[i:end]\n\n\t\tchunkHeader := &FileHeader{\n\t\t\tFileHeader: &multipart.FileHeader{\n\t\t\t\tFilename: \"large_file.txt\",\n\t\t\t\tSize: int64(len(chunk)),\n\t\t\t\tHeader: make(map[string][]string),\n\t\t\t},\n\t\t}\n\t\tchunkHeader.Header.Set(\"Content-Type\", \"text/plain\")\n\t\tchunkHeader.Header.Set(\"Content-Range\", fmt.Sprintf(\"bytes %d-%d/%d\", i, end-1, totalSize))\n\t\tchunkHeader.Header.Set(\"Content-Uid\", uid)\n\n\t\tchunkOption := UploadOption{\n\t\t\tGroups: []string{\"user123\"},\n\t\t\tGzip: true, // Compress chunks\n\t\t}\n\n\t\tchunkFile, err := localManager.Upload(ctx, chunkHeader, strings.NewReader(chunk), chunkOption)\n\t\tif err != nil {\n\t\t\tlog.Fatalf(\"Failed to upload chunk %d: %v\", chunkCount, err)\n\t\t}\n\n\t\tchunkCount++\n\t\tlastFile = chunkFile\n\n\t\t// Check if this is the last chunk\n\t\tif chunkHeader.Complete() {\n\t\t\tfmt.Printf(\"Uploaded large file in %d chunks: %s (ID: %s)\\n\", chunkCount, chunkFile.Filename, chunkFile.ID)\n\t\t\tbreak\n\t\t}\n\t}\n\n\t// 8. Example: Download and read files\n\tif file != nil {\n\t\t// Download as stream\n\t\tresponse, err := localManager.Download(ctx, file.ID)\n\t\tif err != nil {\n\t\t\tlog.Fatalf(\"Failed to download file: %v\", err)\n\t\t}\n\t\tdefer response.Reader.Close()\n\n\t\tfmt.Printf(\"Downloaded file content type: %s, extension: %s\\n\", response.ContentType, response.Extension)\n\n\t\t// Read as bytes\n\t\tdata, err := localManager.Read(ctx, file.ID)\n\t\tif err != nil {\n\t\t\tlog.Fatalf(\"Failed to read file: %v\", err)\n\t\t}\n\n\t\tfmt.Printf(\"File content length: %d bytes\\n\", len(data))\n\t\tif len(data) < 100 {\n\t\t\tfmt.Printf(\"File content: %s\\n\", string(data))\n\t\t} else {\n\t\t\tfmt.Printf(\"File content preview: %s...\\n\", string(data[:100]))\n\t\t}\n\n\t\t// Read as base64\n\t\tbase64Data, err := localManager.ReadBase64(ctx, file.ID)\n\t\tif err != nil {\n\t\t\tlog.Fatalf(\"Failed to read file as base64: %v\", err)\n\t\t}\n\n\t\tfmt.Printf(\"File as base64 (first 50 chars): %s...\\n\", base64Data[:min(50, len(base64Data))])\n\t}\n\n\t// 9. Example: Read chunked file\n\tif lastFile != nil {\n\t\tchunkData, err := localManager.Read(ctx, lastFile.ID)\n\t\tif err != nil {\n\t\t\tlog.Fatalf(\"Failed to read chunked file: %v\", err)\n\t\t}\n\n\t\t// Since the chunks were compressed, we need to decompress\n\t\tdecompressed, err := Gunzip(chunkData)\n\t\tif err != nil {\n\t\t\tlog.Fatalf(\"Failed to decompress chunked file: %v\", err)\n\t\t}\n\n\t\tfmt.Printf(\"Chunked file content length: %d bytes (decompressed)\\n\", len(decompressed))\n\t\tif len(decompressed) < 200 {\n\t\t\tfmt.Printf(\"Chunked file content: %s\\n\", string(decompressed))\n\t\t} else {\n\t\t\tfmt.Printf(\"Chunked file content preview: %s...\\n\", string(decompressed[:200]))\n\t\t}\n\t}\n\n\t// 10. Example: Using global managers\n\tglobalManager := Managers[\"local\"]\n\tif globalManager != nil {\n\t\tfmt.Println(\"Using global manager for local storage\")\n\n\t\t// Test a simple upload with global manager\n\t\ttestContent := \"Test content using global manager\"\n\t\ttestHeader := &FileHeader{\n\t\t\tFileHeader: &multipart.FileHeader{\n\t\t\t\tFilename: \"global_test.txt\",\n\t\t\t\tSize: int64(len(testContent)),\n\t\t\t\tHeader: make(map[string][]string),\n\t\t\t},\n\t\t}\n\t\ttestHeader.Header.Set(\"Content-Type\", \"text/plain\")\n\n\t\ttestFile, err := globalManager.Upload(ctx, testHeader, strings.NewReader(testContent), UploadOption{\n\t\t\tGroups: []string{\"global_user\"},\n\t\t})\n\t\tif err != nil {\n\t\t\tlog.Printf(\"Failed to upload with global manager: %v\", err)\n\t\t} else {\n\t\t\tfmt.Printf(\"Global manager upload successful: %s\\n\", testFile.ID)\n\t\t}\n\t}\n\n\t// 11. Example: File validation\n\tfmt.Println(\"\\n--- File Validation Examples ---\")\n\n\t// Test file size validation\n\ttooLargeContent := strings.Repeat(\"x\", 25*1024*1024) // 25MB, exceeds 20MB limit\n\tlargeFileHeader := &FileHeader{\n\t\tFileHeader: &multipart.FileHeader{\n\t\t\tFilename: \"too_large.txt\",\n\t\t\tSize: int64(len(tooLargeContent)),\n\t\t\tHeader: make(map[string][]string),\n\t\t},\n\t}\n\tlargeFileHeader.Header.Set(\"Content-Type\", \"text/plain\")\n\n\t_, err = localManager.Upload(ctx, largeFileHeader, strings.NewReader(tooLargeContent), UploadOption{})\n\tif err != nil {\n\t\tfmt.Printf(\"Expected error for large file: %v\\n\", err)\n\t}\n\n\t// Test file type validation\n\tinvalidFileHeader := &FileHeader{\n\t\tFileHeader: &multipart.FileHeader{\n\t\t\tFilename: \"script.exe\",\n\t\t\tSize: 1024,\n\t\t\tHeader: make(map[string][]string),\n\t\t},\n\t}\n\tinvalidFileHeader.Header.Set(\"Content-Type\", \"application/x-executable\")\n\n\t_, err = localManager.Upload(ctx, invalidFileHeader, strings.NewReader(\"fake exe content\"), UploadOption{})\n\tif err != nil {\n\t\tfmt.Printf(\"Expected error for invalid file type: %v\\n\", err)\n\t}\n\n\tfmt.Println(\"\\n--- Example Usage Complete ---\")\n}\n\n// ExampleChunkedUpload demonstrates how to handle chunked uploads properly\nfunc ExampleChunkedUpload(manager *Manager, filename string, totalSize int64, contentType string) error {\n\tctx := context.Background()\n\tchunkSize := int64(1024 * 1024) // 1MB chunks\n\tuid := \"unique-file-\" + filename\n\n\tfmt.Printf(\"Starting chunked upload: file=%s, total=%d bytes, chunks=%d\\n\",\n\t\tfilename, totalSize, (totalSize+chunkSize-1)/chunkSize)\n\n\tfor offset := int64(0); offset < totalSize; offset += chunkSize {\n\t\tend := offset + chunkSize - 1\n\t\tif end >= totalSize {\n\t\t\tend = totalSize - 1\n\t\t}\n\n\t\tchunkSize := end - offset + 1\n\n\t\t// Create chunk header\n\t\tchunkHeader := &FileHeader{\n\t\t\tFileHeader: &multipart.FileHeader{\n\t\t\t\tFilename: filename,\n\t\t\t\tSize: chunkSize,\n\t\t\t\tHeader: make(map[string][]string),\n\t\t\t},\n\t\t}\n\t\tchunkHeader.Header.Set(\"Content-Type\", contentType)\n\t\tchunkHeader.Header.Set(\"Content-Range\", fmt.Sprintf(\"bytes %d-%d/%d\", offset, end, totalSize))\n\t\tchunkHeader.Header.Set(\"Content-Uid\", uid)\n\n\t\t// In real usage, you would read the actual chunk data from the source\n\t\tchunkData := make([]byte, chunkSize)\n\t\t// Fill with sample data for demonstration\n\t\tfor i := range chunkData {\n\t\t\tchunkData[i] = byte('A' + (i % 26))\n\t\t}\n\n\t\toption := UploadOption{\n\t\t\tGroups: []string{\"user123\"},\n\t\t\tGzip: false, // Disable compression for this example\n\t\t}\n\n\t\tfile, err := manager.Upload(ctx, chunkHeader, bytes.NewReader(chunkData), option)\n\t\tif err != nil {\n\t\t\treturn fmt.Errorf(\"failed to upload chunk at offset %d: %w\", offset, err)\n\t\t}\n\n\t\tfmt.Printf(\"Uploaded chunk %d-%d/%d\\n\", offset, end, totalSize)\n\n\t\t// Check if this was the last chunk\n\t\tif chunkHeader.Complete() {\n\t\t\tfmt.Printf(\"File upload completed: %s (ID: %s)\\n\", file.Filename, file.ID)\n\n\t\t\t// Verify the uploaded file\n\t\t\tdata, err := manager.Read(ctx, file.ID)\n\t\t\tif err != nil {\n\t\t\t\treturn fmt.Errorf(\"failed to read uploaded file: %w\", err)\n\t\t\t}\n\n\t\t\tif int64(len(data)) != totalSize {\n\t\t\t\treturn fmt.Errorf(\"uploaded file size mismatch: expected %d, got %d\", totalSize, len(data))\n\t\t\t}\n\n\t\t\tfmt.Printf(\"File verification successful: %d bytes\\n\", len(data))\n\t\t\tbreak\n\t\t}\n\t}\n\n\treturn nil\n}\n\n// ExampleS3Upload demonstrates S3-specific features\nfunc ExampleS3Upload() {\n\t// This example requires actual S3 credentials\n\ts3Manager, err := New(ManagerOption{\n\t\tDriver: \"s3\",\n\t\tMaxSize: \"50M\",\n\t\tOptions: map[string]interface{}{\n\t\t\t\"endpoint\": \"https://s3.amazonaws.com\",\n\t\t\t\"region\": \"us-east-1\",\n\t\t\t\"key\": \"your-access-key\",\n\t\t\t\"secret\": \"your-secret-key\",\n\t\t\t\"bucket\": \"your-bucket\",\n\t\t\t\"prefix\": \"test-uploads/\",\n\t\t},\n\t})\n\tif err != nil {\n\t\tlog.Printf(\"S3 manager creation failed (expected without credentials): %v\", err)\n\t\treturn\n\t}\n\n\tctx := context.Background()\n\tcontent := \"Test content for S3 upload\"\n\n\tfileHeader := &FileHeader{\n\t\tFileHeader: &multipart.FileHeader{\n\t\t\tFilename: \"s3_test.txt\",\n\t\t\tSize: int64(len(content)),\n\t\t\tHeader: make(map[string][]string),\n\t\t},\n\t}\n\tfileHeader.Header.Set(\"Content-Type\", \"text/plain\")\n\n\tfile, err := s3Manager.Upload(ctx, fileHeader, strings.NewReader(content), UploadOption{\n\t\tGroups: []string{\"s3_user\"},\n\t})\n\tif err != nil {\n\t\tlog.Printf(\"S3 upload failed: %v\", err)\n\t\treturn\n\t}\n\n\tfmt.Printf(\"S3 upload successful: %s\\n\", file.ID)\n\n\t// Get presigned URL\n\turl := s3Manager.storage.URL(ctx, file.ID)\n\tfmt.Printf(\"Presigned URL: %s\\n\", url)\n}\n\n// Helper function for min\nfunc min(a, b int) int {\n\tif a < b {\n\t\treturn a\n\t}\n\treturn b\n}\n" }, { "path": "attachment/fileheader.go", "content": "package attachment\n\nimport (\n\t\"strconv\"\n\t\"strings\"\n)\n\n// UID is the uid of the file, it is the unique identifier of the file\nfunc (fileheader *FileHeader) UID() string {\n\treturn fileheader.Header.Get(\"Content-Uid\")\n}\n\n// Fingerprint is the fingerprint of the file, it is the fingerprint of the file\nfunc (fileheader *FileHeader) Fingerprint() string {\n\treturn fileheader.Header.Get(\"Content-Fingerprint\")\n}\n\n// Range is the range of the file, it is the start and end of the file\nfunc (fileheader *FileHeader) Range() string {\n\treturn fileheader.Header.Get(\"Content-Range\")\n}\n\n// Sync is the sync of the file, it is the sync of the file\nfunc (fileheader *FileHeader) Sync() bool {\n\treturn fileheader.Header.Get(\"Content-Sync\") == \"true\"\n}\n\n// IsChunk is the chunk of the file, it is the chunk of the file\nfunc (fileheader *FileHeader) IsChunk() bool {\n\treturn fileheader.Range() != \"\"\n}\n\n// Complete checks if the chunk upload is completed\n// For non-chunk files, it returns true\n// For chunk files, it parses the Content-Range header to determine if this is the last chunk\nfunc (fileheader *FileHeader) Complete() bool {\n\tif !fileheader.IsChunk() {\n\t\treturn true\n\t}\n\n\t// Parse Content-Range header: \"bytes start-end/total\"\n\trangeHeader := fileheader.Range()\n\tif rangeHeader == \"\" {\n\t\treturn false\n\t}\n\n\t// Remove \"bytes \" prefix\n\trangeStr := strings.TrimPrefix(rangeHeader, \"bytes \")\n\n\t// Split by \"/\"\n\tparts := strings.Split(rangeStr, \"/\")\n\tif len(parts) != 2 {\n\t\treturn false\n\t}\n\n\t// Parse total size\n\ttotal, err := strconv.ParseInt(parts[1], 10, 64)\n\tif err != nil {\n\t\treturn false\n\t}\n\n\t// Parse range \"start-end\"\n\trangeParts := strings.Split(parts[0], \"-\")\n\tif len(rangeParts) != 2 {\n\t\treturn false\n\t}\n\n\tend, err := strconv.ParseInt(rangeParts[1], 10, 64)\n\tif err != nil {\n\t\treturn false\n\t}\n\n\t// Check if this is the last chunk: end + 1 == total\n\treturn end+1 == total\n}\n\n// GetChunkInfo returns the chunk information parsed from Content-Range header\n// Returns start, end, total, and error\nfunc (fileheader *FileHeader) GetChunkInfo() (start, end, total int64, err error) {\n\tif !fileheader.IsChunk() {\n\t\treturn 0, 0, 0, nil\n\t}\n\n\trangeHeader := fileheader.Range()\n\tif rangeHeader == \"\" {\n\t\treturn 0, 0, 0, nil\n\t}\n\n\t// Remove \"bytes \" prefix\n\trangeStr := strings.TrimPrefix(rangeHeader, \"bytes \")\n\n\t// Split by \"/\"\n\tparts := strings.Split(rangeStr, \"/\")\n\tif len(parts) != 2 {\n\t\treturn 0, 0, 0, nil\n\t}\n\n\t// Parse total size\n\ttotal, err = strconv.ParseInt(parts[1], 10, 64)\n\tif err != nil {\n\t\treturn 0, 0, 0, err\n\t}\n\n\t// Parse range \"start-end\"\n\trangeParts := strings.Split(parts[0], \"-\")\n\tif len(rangeParts) != 2 {\n\t\treturn 0, 0, 0, nil\n\t}\n\n\tstart, err = strconv.ParseInt(rangeParts[0], 10, 64)\n\tif err != nil {\n\t\treturn 0, 0, 0, err\n\t}\n\n\tend, err = strconv.ParseInt(rangeParts[1], 10, 64)\n\tif err != nil {\n\t\treturn 0, 0, 0, err\n\t}\n\n\treturn start, end, total, nil\n}\n\n// GetTotalSize returns the total file size from Content-Range header\nfunc (fileheader *FileHeader) GetTotalSize() int64 {\n\t_, _, total, err := fileheader.GetChunkInfo()\n\tif err != nil {\n\t\treturn 0\n\t}\n\treturn total\n}\n\n// GetChunkSize returns the current chunk size\nfunc (fileheader *FileHeader) GetChunkSize() int64 {\n\tstart, end, _, err := fileheader.GetChunkInfo()\n\tif err != nil {\n\t\treturn 0\n\t}\n\treturn end - start + 1\n}\n" }, { "path": "attachment/gzip.go", "content": "package attachment\n\nimport (\n\t\"bytes\"\n\t\"compress/gzip\"\n\t\"fmt\"\n\t\"io\"\n\t\"os\"\n)\n\n// GzipCompressor supports chunked compression for Gzip\ntype GzipCompressor struct {\n\twriter *gzip.Writer\n\tbuffer *bytes.Buffer\n\tfile *os.File // optional file handle\n}\n\n// NewGzipCompressor creates a new Gzip compressor\nfunc NewGzipCompressor() *GzipCompressor {\n\tbuf := &bytes.Buffer{}\n\tgz := gzip.NewWriter(buf)\n\treturn &GzipCompressor{\n\t\twriter: gz,\n\t\tbuffer: buf,\n\t\tfile: nil,\n\t}\n}\n\n// NewGzipCompressorFromFile creates a Gzip compressor from file, supports streaming read\nfunc NewGzipCompressorFromFile(filePath string) (*GzipCompressor, error) {\n\tfile, err := os.Open(filePath)\n\tif err != nil {\n\t\treturn nil, fmt.Errorf(\"failed to open file %s: %w\", filePath, err)\n\t}\n\n\tbuf := &bytes.Buffer{}\n\tgz := gzip.NewWriter(buf)\n\treturn &GzipCompressor{\n\t\twriter: gz,\n\t\tbuffer: buf,\n\t\tfile: file,\n\t}, nil\n}\n\n// ReadChunk reads a chunk of specified size from file and compresses it\nfunc (gc *GzipCompressor) ReadChunk(chunkSize int) (bool, error) {\n\tif gc.file == nil {\n\t\treturn false, fmt.Errorf(\"no file associated with this compressor\")\n\t}\n\n\tchunk := make([]byte, chunkSize)\n\tn, err := gc.file.Read(chunk)\n\tif err != nil && err != io.EOF {\n\t\treturn false, fmt.Errorf(\"failed to read from file: %w\", err)\n\t}\n\n\tif n > 0 {\n\t\tif err := gc.Write(chunk[:n]); err != nil {\n\t\t\treturn false, err\n\t\t}\n\t}\n\n\t// return whether there is more data\n\treturn err != io.EOF, nil\n}\n\n// CompressFileInChunks compresses the entire file in chunks\nfunc (gc *GzipCompressor) CompressFileInChunks(chunkSize int) error {\n\tif gc.file == nil {\n\t\treturn fmt.Errorf(\"no file associated with this compressor\")\n\t}\n\n\tfor {\n\t\thasMore, err := gc.ReadChunk(chunkSize)\n\t\tif err != nil {\n\t\t\treturn err\n\t\t}\n\t\tif !hasMore {\n\t\t\tbreak\n\t\t}\n\t}\n\treturn nil\n}\n\n// Write writes data for compression (supports chunked writing)\nfunc (gc *GzipCompressor) Write(data []byte) error {\n\t_, err := gc.writer.Write(data)\n\tif err != nil {\n\t\treturn fmt.Errorf(\"failed to write data to gzip: %w\", err)\n\t}\n\treturn nil\n}\n\n// Flush flushes the buffer but does not close the compressor\nfunc (gc *GzipCompressor) Flush() error {\n\treturn gc.writer.Flush()\n}\n\n// Close closes the compressor and returns the final compressed data\nfunc (gc *GzipCompressor) Close() ([]byte, error) {\n\terr := gc.writer.Close()\n\tif err != nil {\n\t\treturn nil, fmt.Errorf(\"failed to close gzip writer: %w\", err)\n\t}\n\n\t// if there is an associated file, close it too\n\tif gc.file != nil {\n\t\tgc.file.Close()\n\t\tgc.file = nil\n\t}\n\n\treturn gc.buffer.Bytes(), nil\n}\n\n// GetCompressedData gets the current compressed data (without closing the compressor)\nfunc (gc *GzipCompressor) GetCompressedData() []byte {\n\t// flush the buffer first\n\tgc.writer.Flush()\n\treturn gc.buffer.Bytes()\n}\n\n// Reset resets the compressor for reuse\nfunc (gc *GzipCompressor) Reset() {\n\tgc.buffer.Reset()\n\tgc.writer.Reset(gc.buffer)\n\tif gc.file != nil {\n\t\tgc.file.Close()\n\t\tgc.file = nil\n\t}\n}\n\n// Gzip compresses data in one go\nfunc Gzip(data []byte) ([]byte, error) {\n\tvar buf bytes.Buffer\n\tgz := gzip.NewWriter(&buf)\n\t_, err := gz.Write(data)\n\tif err != nil {\n\t\treturn nil, fmt.Errorf(\"failed to gzip data: %w\", err)\n\t}\n\terr = gz.Close()\n\tif err != nil {\n\t\treturn nil, fmt.Errorf(\"failed to close gzip writer: %w\", err)\n\t}\n\treturn buf.Bytes(), nil\n}\n\n// GzipChunks compresses multiple data chunks\nfunc GzipChunks(chunks [][]byte) ([]byte, error) {\n\tcompressor := NewGzipCompressor()\n\n\tfor _, chunk := range chunks {\n\t\tif err := compressor.Write(chunk); err != nil {\n\t\t\treturn nil, err\n\t\t}\n\t}\n\n\treturn compressor.Close()\n}\n\n// GzipFromReader compresses data from Reader stream\nfunc GzipFromReader(reader io.Reader) ([]byte, error) {\n\tvar buf bytes.Buffer\n\tgz := gzip.NewWriter(&buf)\n\n\t_, err := io.Copy(gz, reader)\n\tif err != nil {\n\t\treturn nil, fmt.Errorf(\"failed to copy data to gzip: %w\", err)\n\t}\n\n\terr = gz.Close()\n\tif err != nil {\n\t\treturn nil, fmt.Errorf(\"failed to close gzip writer: %w\", err)\n\t}\n\n\treturn buf.Bytes(), nil\n}\n\n// Gunzip decompresses gzip data\nfunc Gunzip(data []byte) ([]byte, error) {\n\treader, err := gzip.NewReader(bytes.NewReader(data))\n\tif err != nil {\n\t\treturn nil, fmt.Errorf(\"failed to create gzip reader: %w\", err)\n\t}\n\tdefer reader.Close()\n\n\tvar buf bytes.Buffer\n\t_, err = io.Copy(&buf, reader)\n\tif err != nil {\n\t\treturn nil, fmt.Errorf(\"failed to decompress data: %w\", err)\n\t}\n\n\treturn buf.Bytes(), nil\n}\n\n// GzipToWriter writes compressed data to Writer\nfunc GzipToWriter(data []byte, writer io.Writer) error {\n\tgz := gzip.NewWriter(writer)\n\tdefer gz.Close()\n\n\t_, err := gz.Write(data)\n\tif err != nil {\n\t\treturn fmt.Errorf(\"failed to write gzip data: %w\", err)\n\t}\n\n\treturn nil\n}\n\n// GzipFromReaderToWriter reads data from Reader and writes compressed data to Writer\nfunc GzipFromReaderToWriter(reader io.Reader, writer io.Writer) error {\n\tgz := gzip.NewWriter(writer)\n\tdefer gz.Close()\n\n\t_, err := io.Copy(gz, reader)\n\tif err != nil {\n\t\treturn fmt.Errorf(\"failed to copy and compress data: %w\", err)\n\t}\n\n\treturn nil\n}\n\n// GzipFile compresses entire file (loads into memory at once)\nfunc GzipFile(filePath string) ([]byte, error) {\n\tdata, err := os.ReadFile(filePath)\n\tif err != nil {\n\t\treturn nil, fmt.Errorf(\"failed to read file %s: %w\", filePath, err)\n\t}\n\treturn Gzip(data)\n}\n\n// GzipFileInChunks compresses file in chunks (memory friendly)\nfunc GzipFileInChunks(filePath string, chunkSize int) ([]byte, error) {\n\tcompressor, err := NewGzipCompressorFromFile(filePath)\n\tif err != nil {\n\t\treturn nil, err\n\t}\n\tdefer compressor.Close()\n\n\terr = compressor.CompressFileInChunks(chunkSize)\n\tif err != nil {\n\t\treturn nil, err\n\t}\n\n\treturn compressor.Close()\n}\n\n// GzipFileToFile compresses file and saves to another file\nfunc GzipFileToFile(srcPath, dstPath string, chunkSize int) error {\n\tsrcFile, err := os.Open(srcPath)\n\tif err != nil {\n\t\treturn fmt.Errorf(\"failed to open source file %s: %w\", srcPath, err)\n\t}\n\tdefer srcFile.Close()\n\n\tdstFile, err := os.Create(dstPath)\n\tif err != nil {\n\t\treturn fmt.Errorf(\"failed to create destination file %s: %w\", dstPath, err)\n\t}\n\tdefer dstFile.Close()\n\n\treturn GzipFromReaderToWriter(srcFile, dstFile)\n}\n\n// GzipFileStream compresses file in streaming mode, returns Reader interface\nfunc GzipFileStream(filePath string) (io.Reader, error) {\n\tfile, err := os.Open(filePath)\n\tif err != nil {\n\t\treturn nil, fmt.Errorf(\"failed to open file %s: %w\", filePath, err)\n\t}\n\n\tpr, pw := io.Pipe()\n\tgo func() {\n\t\tdefer pw.Close()\n\t\tdefer file.Close()\n\n\t\tgz := gzip.NewWriter(pw)\n\t\tdefer gz.Close()\n\n\t\t_, err := io.Copy(gz, file)\n\t\tif err != nil {\n\t\t\tpw.CloseWithError(err)\n\t\t}\n\t}()\n\n\treturn pr, nil\n}\n" }, { "path": "attachment/load.go", "content": "package attachment\n\nimport (\n\t\"fmt\"\n\t\"strings\"\n\n\t\"github.com/yaoapp/gou/application\"\n\t\"github.com/yaoapp/kun/log\"\n\t\"github.com/yaoapp/yao/config\"\n\t\"github.com/yaoapp/yao/data\"\n\t\"github.com/yaoapp/yao/share\"\n)\n\n// SystemUploaders system uploaders\nvar systemUploaders = map[string]string{\n\t\"__yao.attachment\": \"yao/uploaders/attachment.local.yao\",\n}\n\n// Load load uploaders\nfunc Load(cfg config.Config) error {\n\t// Register attachment processes\n\tInit()\n\n\tmessages := []string{}\n\n\t// Load system uploaders\n\terr := loadSystemUploaders(cfg)\n\tif err != nil {\n\t\treturn err\n\t}\n\n\t// Load filesystem uploaders\n\texts := []string{\"*.s3.yao\", \"*.local.yao\", \"*.s3.json\", \"*.local.json\", \"*.s3.jsonc\", \"*.local.jsonc\"}\n\terr = application.App.Walk(\"uploaders\", func(root, file string, isdir bool) error {\n\t\tif isdir {\n\t\t\treturn nil\n\t\t}\n\n\t\t// Skip if not uploader file\n\t\tif !isUploaderFile(file) {\n\t\t\treturn nil\n\t\t}\n\n\t\terr := loadUploaderFile(root, file, cfg)\n\t\tif err != nil {\n\t\t\tmessages = append(messages, err.Error())\n\t\t}\n\t\treturn err\n\t}, exts...)\n\n\tif len(messages) > 0 {\n\t\tfor _, message := range messages {\n\t\t\tlog.Error(\"Load filesystem uploaders error: %s\", message)\n\t\t}\n\t\treturn fmt.Errorf(\"%s\", strings.Join(messages, \";\\n\"))\n\t}\n\n\treturn nil\n}\n\n// loadSystemUploaders load system uploaders\nfunc loadSystemUploaders(cfg config.Config) error {\n\tfor id, path := range systemUploaders {\n\t\tcontent, err := data.Read(path)\n\t\tif err != nil {\n\t\t\treturn err\n\t\t}\n\n\t\t// Parse uploader config\n\t\tvar option ManagerOption\n\t\terr = application.Parse(path, content, &option)\n\t\tif err != nil {\n\t\t\treturn err\n\t\t}\n\n\t\t// Replace environment variables and paths\n\t\toption.ReplaceEnv(cfg.DataRoot)\n\n\t\t// Register the uploader manager\n\t\t_, err = Register(id, option.Driver, option)\n\t\tif err != nil {\n\t\t\tlog.Error(\"register system uploader %s error: %s\", id, err.Error())\n\t\t\treturn err\n\t\t}\n\n\t\tlog.Info(\"loaded system uploader: %s (%s)\", id, option.Label)\n\t}\n\n\treturn nil\n}\n\n// loadUploaderFile load a single uploader file\nfunc loadUploaderFile(root, file string, cfg config.Config) error {\n\t// Generate uploader ID\n\tid := share.ID(root, file)\n\n\t// Read file content\n\tcontent, err := application.App.Read(file)\n\tif err != nil {\n\t\treturn fmt.Errorf(\"failed to read uploader file %s: %v\", file, err)\n\t}\n\n\t// Parse uploader config\n\tvar option ManagerOption\n\terr = application.Parse(file, content, &option)\n\tif err != nil {\n\t\treturn fmt.Errorf(\"failed to parse uploader file %s: %v\", file, err)\n\t}\n\n\t// Validate driver consistency between filename and config\n\tfilenameDriver := extractDriverFromFilename(file)\n\tif filenameDriver != \"\" && option.Driver != \"\" && filenameDriver != option.Driver {\n\t\tlog.Warn(\"Driver mismatch in uploader file %s: filename suggests '%s' but config has '%s'\",\n\t\t\tfile, filenameDriver, option.Driver)\n\t}\n\n\t// Replace environment variables and paths\n\toption.ReplaceEnv(cfg.DataRoot)\n\n\t// Register the uploader manager\n\t_, err = Register(id, option.Driver, option)\n\tif err != nil {\n\t\tlog.Error(\"register uploader %s error: %s\", id, err.Error())\n\t\treturn fmt.Errorf(\"failed to register uploader %s: %v\", id, err)\n\t}\n\n\tlog.Info(\"loaded uploader: %s (%s)\", id, option.Label)\n\treturn nil\n}\n\n// isUploaderFile checks if the file is an uploader configuration file\nfunc isUploaderFile(filename string) bool {\n\t// Accept files with specific driver patterns: *.s3.yao, *.local.yao, etc.\n\tlower := strings.ToLower(filename)\n\treturn strings.HasSuffix(lower, \".s3.yao\") ||\n\t\tstrings.HasSuffix(lower, \".local.yao\") ||\n\t\tstrings.HasSuffix(lower, \".s3.json\") ||\n\t\tstrings.HasSuffix(lower, \".local.json\") ||\n\t\tstrings.HasSuffix(lower, \".s3.jsonc\") ||\n\t\tstrings.HasSuffix(lower, \".local.jsonc\")\n}\n\n// extractDriverFromFilename extracts the driver name from filename (e.g., \"test.s3.yao\" -> \"s3\")\nfunc extractDriverFromFilename(filename string) string {\n\tlower := strings.ToLower(filename)\n\n\t// Extract driver from patterns like \"*.s3.yao\", \"*.local.json\", etc.\n\tif strings.Contains(lower, \".s3.\") {\n\t\treturn \"s3\"\n\t} else if strings.Contains(lower, \".local.\") {\n\t\treturn \"local\"\n\t}\n\n\treturn \"\"\n}\n" }, { "path": "attachment/load_test.go", "content": "package attachment\n\nimport (\n\t\"testing\"\n\n\t\"github.com/stretchr/testify/assert\"\n\t\"github.com/yaoapp/yao/config\"\n\t\"github.com/yaoapp/yao/test\"\n)\n\nfunc TestLoad(t *testing.T) {\n\ttest.Prepare(t, config.Conf)\n\tdefer test.Clean()\n\n\terr := Load(config.Conf)\n\tassert.NoError(t, err)\n\tcheck(t)\n}\n\nfunc check(t *testing.T) {\n\t// Check that managers are loaded\n\tassert.NotEmpty(t, Managers, \"Managers should not be empty after loading\")\n\n\t// Check system uploader\n\t_, exists := Managers[\"__yao.attachment\"]\n\tassert.True(t, exists, \"System uploader __yao.attachment should be loaded\")\n\n\t// Check test app uploaders (must exist)\n\t// These are the uploaders in yao-dev-app/uploaders/\n\t_, hasData := Managers[\"data\"]\n\t_, hasTest := Managers[\"test\"]\n\n\t// Both test uploaders should be loaded\n\tassert.True(t, hasData, \"Test uploader 'data' should be loaded from data.local.yao\")\n\tassert.True(t, hasTest, \"Test uploader 'test' should be loaded from test.s3.yao\")\n\n\t// Log all loaded managers for debugging\n\tt.Logf(\"Loaded managers: %v\", getManagerNames())\n}\n\n// getManagerNames returns a slice of manager names for testing\nfunc getManagerNames() []string {\n\tnames := make([]string, 0, len(Managers))\n\tfor name := range Managers {\n\t\tnames = append(names, name)\n\t}\n\treturn names\n}\n" }, { "path": "attachment/local/storage.go", "content": "package local\n\nimport (\n\t\"compress/gzip\"\n\t\"context\"\n\t\"crypto/sha256\"\n\t\"fmt\"\n\t\"io\"\n\t\"mime\"\n\t\"net/http\"\n\t\"os\"\n\t\"path/filepath\"\n\t\"strings\"\n\t\"time\"\n)\n\n// MaxImageSize maximum image size (1920x1080)\nconst MaxImageSize = 1920\n\n// Storage the local storage driver\ntype Storage struct {\n\tPath string `json:\"path\" yaml:\"path\"`\n\tCompression bool `json:\"compression\" yaml:\"compression\"`\n\tBaseURL string `json:\"base_url\" yaml:\"base_url\"`\n\tPreviewURL func(fileID string) string `json:\"-\" yaml:\"-\"`\n}\n\n// New create a new local storage\nfunc New(options map[string]interface{}) (*Storage, error) {\n\tstorage := &Storage{\n\t\tCompression: true,\n\t}\n\n\tif path, ok := options[\"path\"].(string); ok {\n\t\tstorage.Path = path\n\t}\n\n\tif compression, ok := options[\"compression\"].(bool); ok {\n\t\tstorage.Compression = compression\n\t}\n\n\tif baseURL, ok := options[\"base_url\"].(string); ok {\n\t\tstorage.BaseURL = baseURL\n\t}\n\n\tif previewURL, ok := options[\"preview_url\"].(func(string) string); ok {\n\t\tstorage.PreviewURL = previewURL\n\t}\n\n\tif storage.Path == \"\" {\n\t\treturn nil, fmt.Errorf(\"path is required\")\n\t}\n\n\t// Ensure the base path exists\n\tif err := os.MkdirAll(storage.Path, 0755); err != nil {\n\t\treturn nil, fmt.Errorf(\"failed to create base path: %w\", err)\n\t}\n\n\treturn storage, nil\n}\n\n// Upload upload file to local storage\nfunc (storage *Storage) Upload(ctx context.Context, path string, reader io.Reader, contentType string) (string, error) {\n\tfullPath := filepath.Join(storage.Path, path)\n\n\t// Create directory if not exists\n\tdir := filepath.Dir(fullPath)\n\tif err := os.MkdirAll(dir, 0755); err != nil {\n\t\treturn \"\", err\n\t}\n\n\t// Create and write file\n\tfile, err := os.Create(fullPath)\n\tif err != nil {\n\t\treturn \"\", err\n\t}\n\tdefer file.Close()\n\n\t_, err = io.Copy(file, reader)\n\tif err != nil {\n\t\treturn \"\", err\n\t}\n\n\treturn path, nil\n}\n\n// UploadChunk uploads a chunk of a file\nfunc (storage *Storage) UploadChunk(ctx context.Context, path string, chunkIndex int, reader io.Reader, contentType string) error {\n\t// Create chunks directory\n\tchunksDir := filepath.Join(storage.Path, \".chunks\", path)\n\tif err := os.MkdirAll(chunksDir, 0755); err != nil {\n\t\treturn err\n\t}\n\n\t// Write chunk file\n\tchunkPath := filepath.Join(chunksDir, fmt.Sprintf(\"chunk_%d\", chunkIndex))\n\tfile, err := os.Create(chunkPath)\n\tif err != nil {\n\t\treturn err\n\t}\n\tdefer file.Close()\n\n\t_, err = io.Copy(file, reader)\n\treturn err\n}\n\n// MergeChunks merges all chunks into the final file\nfunc (storage *Storage) MergeChunks(ctx context.Context, path string, totalChunks int) error {\n\tchunksDir := filepath.Join(storage.Path, \".chunks\", path)\n\tfinalPath := filepath.Join(storage.Path, path)\n\n\t// Create directory for final file\n\tdir := filepath.Dir(finalPath)\n\tif err := os.MkdirAll(dir, 0755); err != nil {\n\t\treturn err\n\t}\n\n\t// Create final file\n\tfinalFile, err := os.Create(finalPath)\n\tif err != nil {\n\t\treturn err\n\t}\n\tdefer finalFile.Close()\n\n\t// Read and merge chunks in order\n\tfor i := 0; i < totalChunks; i++ {\n\t\tchunkPath := filepath.Join(chunksDir, fmt.Sprintf(\"chunk_%d\", i))\n\t\tchunkFile, err := os.Open(chunkPath)\n\t\tif err != nil {\n\t\t\treturn fmt.Errorf(\"failed to read chunk %d: %w\", i, err)\n\t\t}\n\n\t\t_, err = io.Copy(finalFile, chunkFile)\n\t\tchunkFile.Close()\n\t\tif err != nil {\n\t\t\treturn fmt.Errorf(\"failed to copy chunk %d: %w\", i, err)\n\t\t}\n\t}\n\n\t// Clean up chunks directory\n\tos.RemoveAll(chunksDir)\n\treturn nil\n}\n\n// Reader read file from local storage\nfunc (storage *Storage) Reader(ctx context.Context, path string) (io.ReadCloser, error) {\n\tfullpath := filepath.Join(storage.Path, path)\n\n\treader, err := os.Open(fullpath)\n\tif err != nil {\n\t\treturn nil, err\n\t}\n\n\t// If the file is a gzip file, decompress it\n\tif strings.HasSuffix(path, \".gz\") {\n\t\treader, err := gzip.NewReader(reader)\n\t\tif err != nil {\n\t\t\treturn nil, err\n\t\t}\n\t\treturn reader, nil\n\t}\n\n\treturn reader, nil\n}\n\n// Download download file from local storage\nfunc (storage *Storage) Download(ctx context.Context, path string) (io.ReadCloser, string, error) {\n\tfullPath := filepath.Join(storage.Path, path)\n\treader, err := os.Open(fullPath)\n\tif err != nil {\n\t\treturn nil, \"\", err\n\t}\n\n\t// Try to detect content type from file extension\n\tcontentType := \"application/octet-stream\"\n\text := filepath.Ext(strings.TrimSuffix(path, \".gz\"))\n\tswitch strings.ToLower(ext) {\n\tcase \".txt\":\n\t\tcontentType = \"text/plain\"\n\tcase \".html\":\n\t\tcontentType = \"text/html\"\n\tcase \".css\":\n\t\tcontentType = \"text/css\"\n\tcase \".js\":\n\t\tcontentType = \"application/javascript\"\n\tcase \".json\":\n\t\tcontentType = \"application/json\"\n\tcase \".jpg\", \".jpeg\":\n\t\tcontentType = \"image/jpeg\"\n\tcase \".png\":\n\t\tcontentType = \"image/png\"\n\tcase \".gif\":\n\t\tcontentType = \"image/gif\"\n\tcase \".pdf\":\n\t\tcontentType = \"application/pdf\"\n\tcase \".mp4\":\n\t\tcontentType = \"video/mp4\"\n\tcase \".mp3\":\n\t\tcontentType = \"audio/mpeg\"\n\tcase \".wav\":\n\t\tcontentType = \"audio/wav\"\n\tcase \".ogg\":\n\t\tcontentType = \"audio/ogg\"\n\tcase \".webm\":\n\t\tcontentType = \"video/webm\"\n\tcase \".webp\":\n\t\tcontentType = \"image/webp\"\n\tcase \".zip\":\n\t}\n\n\t// If the file is a gzip file, decompress it\n\tif strings.HasSuffix(path, \".gz\") {\n\t\treader, err := gzip.NewReader(reader)\n\t\tif err != nil {\n\t\t\treturn nil, \"\", err\n\t\t}\n\t\treturn reader, contentType, nil\n\t}\n\n\treturn reader, contentType, nil\n}\n\n// URL get file url\nfunc (storage *Storage) URL(ctx context.Context, path string) string {\n\tif storage.PreviewURL != nil {\n\t\treturn storage.PreviewURL(path)\n\t}\n\tif storage.BaseURL != \"\" {\n\t\treturn fmt.Sprintf(\"%s/%s\", strings.TrimRight(storage.BaseURL, \"/\"), path)\n\t}\n\treturn fmt.Sprintf(\"%s/%s\", storage.Path, path)\n}\n\n// GetContent gets file content as bytes\nfunc (storage *Storage) GetContent(ctx context.Context, path string) ([]byte, error) {\n\treader, err := storage.Reader(ctx, path)\n\tif err != nil {\n\t\treturn nil, err\n\t}\n\tdefer reader.Close()\n\n\treturn io.ReadAll(reader)\n}\n\n// Exists checks if a file exists\nfunc (storage *Storage) Exists(ctx context.Context, path string) bool {\n\tfullpath := filepath.Join(storage.Path, path)\n\t_, err := os.Stat(fullpath)\n\treturn err == nil\n}\n\n// Delete deletes a file\nfunc (storage *Storage) Delete(ctx context.Context, path string) error {\n\tfullpath := filepath.Join(storage.Path, path)\n\treturn os.Remove(fullpath)\n}\n\nfunc (storage *Storage) makeID(filename string, ext string) string {\n\tdate := time.Now().Format(\"20060102\")\n\thash := fmt.Sprintf(\"%x\", sha256.Sum256([]byte(filename)))[:8]\n\tname := strings.TrimSuffix(filepath.Base(filename), ext)\n\treturn fmt.Sprintf(\"%s/%s-%s%s\", date, name, hash, ext)\n}\n\n// LocalPath returns the absolute path of the file and its content type\nfunc (storage *Storage) LocalPath(ctx context.Context, path string) (string, string, error) {\n\tfullPath := filepath.Join(storage.Path, path)\n\n\t// Check if file exists\n\tif _, err := os.Stat(fullPath); os.IsNotExist(err) {\n\t\treturn \"\", \"\", fmt.Errorf(\"file not found: %s\", path)\n\t}\n\n\t// For gzipped files, we need to detect the original content type, not the gzip wrapper\n\tvar contentType string\n\tvar err error\n\n\tif strings.HasSuffix(path, \".gz\") {\n\t\t// For gzipped files, detect content type of the decompressed content\n\t\toriginalPath := strings.TrimSuffix(path, \".gz\")\n\t\text := filepath.Ext(originalPath)\n\n\t\t// First try to detect by original file extension\n\t\tcontentType, err = detectContentTypeFromExtension(ext)\n\t\tif err != nil || contentType == \"application/octet-stream\" {\n\t\t\t// Fallback: decompress and detect from content\n\t\t\tcontentType, err = detectContentTypeFromGzippedFile(fullPath)\n\t\t\tif err != nil {\n\t\t\t\treturn \"\", \"\", fmt.Errorf(\"failed to detect content type from gzipped file: %w\", err)\n\t\t\t}\n\t\t}\n\t} else {\n\t\t// Regular file content type detection\n\t\tcontentType, err = detectContentType(fullPath)\n\t\tif err != nil {\n\t\t\treturn \"\", \"\", fmt.Errorf(\"failed to detect content type: %w\", err)\n\t\t}\n\t}\n\n\t// Return absolute path\n\tabsPath, err := filepath.Abs(fullPath)\n\tif err != nil {\n\t\treturn \"\", \"\", fmt.Errorf(\"failed to get absolute path: %w\", err)\n\t}\n\n\treturn absPath, contentType, nil\n}\n\n// detectContentType detects content type based on file extension and content\nfunc detectContentType(filePath string) (string, error) {\n\t// First try to detect by file extension\n\text := strings.ToLower(filepath.Ext(filePath))\n\n\t// Common file extensions mapping\n\tswitch ext {\n\tcase \".txt\":\n\t\treturn \"text/plain\", nil\n\tcase \".html\", \".htm\":\n\t\treturn \"text/html\", nil\n\tcase \".css\":\n\t\treturn \"text/css\", nil\n\tcase \".js\":\n\t\treturn \"application/javascript\", nil\n\tcase \".json\":\n\t\treturn \"application/json\", nil\n\tcase \".xml\":\n\t\treturn \"application/xml\", nil\n\tcase \".jpg\", \".jpeg\":\n\t\treturn \"image/jpeg\", nil\n\tcase \".png\":\n\t\treturn \"image/png\", nil\n\tcase \".gif\":\n\t\treturn \"image/gif\", nil\n\tcase \".webp\":\n\t\treturn \"image/webp\", nil\n\tcase \".svg\":\n\t\treturn \"image/svg+xml\", nil\n\tcase \".pdf\":\n\t\treturn \"application/pdf\", nil\n\tcase \".doc\":\n\t\treturn \"application/msword\", nil\n\tcase \".docx\":\n\t\treturn \"application/vnd.openxmlformats-officedocument.wordprocessingml.document\", nil\n\tcase \".xls\":\n\t\treturn \"application/vnd.ms-excel\", nil\n\tcase \".xlsx\":\n\t\treturn \"application/vnd.openxmlformats-officedocument.spreadsheetml.sheet\", nil\n\tcase \".ppt\":\n\t\treturn \"application/vnd.ms-powerpoint\", nil\n\tcase \".pptx\":\n\t\treturn \"application/vnd.openxmlformats-officedocument.presentationml.presentation\", nil\n\tcase \".zip\":\n\t\treturn \"application/zip\", nil\n\tcase \".tar\":\n\t\treturn \"application/x-tar\", nil\n\tcase \".gz\":\n\t\treturn \"application/gzip\", nil\n\tcase \".mp3\":\n\t\treturn \"audio/mpeg\", nil\n\tcase \".wav\":\n\t\treturn \"audio/wav\", nil\n\tcase \".m4a\":\n\t\treturn \"audio/mp4\", nil\n\tcase \".ogg\":\n\t\treturn \"audio/ogg\", nil\n\tcase \".mp4\":\n\t\treturn \"video/mp4\", nil\n\tcase \".avi\":\n\t\treturn \"video/x-msvideo\", nil\n\tcase \".mov\":\n\t\treturn \"video/quicktime\", nil\n\tcase \".webm\":\n\t\treturn \"video/webm\", nil\n\tcase \".md\", \".mdx\":\n\t\treturn \"text/markdown\", nil\n\tcase \".yao\":\n\t\treturn \"application/yao\", nil\n\tcase \".csv\":\n\t\treturn \"text/csv\", nil\n\t}\n\n\t// Try to detect by MIME package\n\tif contentType := mime.TypeByExtension(ext); contentType != \"\" {\n\t\treturn contentType, nil\n\t}\n\n\t// Fallback: detect by reading file content\n\tfile, err := os.Open(filePath)\n\tif err != nil {\n\t\treturn \"application/octet-stream\", nil // Default fallback\n\t}\n\tdefer file.Close()\n\n\t// Read first 512 bytes for content detection\n\tbuffer := make([]byte, 512)\n\tn, err := file.Read(buffer)\n\tif err != nil && err != io.EOF {\n\t\treturn \"application/octet-stream\", nil\n\t}\n\n\t// Use http.DetectContentType to detect based on content\n\tcontentType := http.DetectContentType(buffer[:n])\n\treturn contentType, nil\n}\n\n// detectContentTypeFromExtension detects content type based only on file extension\nfunc detectContentTypeFromExtension(ext string) (string, error) {\n\text = strings.ToLower(ext)\n\n\t// Common file extensions mapping\n\tswitch ext {\n\tcase \".txt\":\n\t\treturn \"text/plain\", nil\n\tcase \".html\", \".htm\":\n\t\treturn \"text/html\", nil\n\tcase \".css\":\n\t\treturn \"text/css\", nil\n\tcase \".js\":\n\t\treturn \"application/javascript\", nil\n\tcase \".json\":\n\t\treturn \"application/json\", nil\n\tcase \".xml\":\n\t\treturn \"application/xml\", nil\n\tcase \".jpg\", \".jpeg\":\n\t\treturn \"image/jpeg\", nil\n\tcase \".png\":\n\t\treturn \"image/png\", nil\n\tcase \".gif\":\n\t\treturn \"image/gif\", nil\n\tcase \".webp\":\n\t\treturn \"image/webp\", nil\n\tcase \".svg\":\n\t\treturn \"image/svg+xml\", nil\n\tcase \".pdf\":\n\t\treturn \"application/pdf\", nil\n\tcase \".doc\":\n\t\treturn \"application/msword\", nil\n\tcase \".docx\":\n\t\treturn \"application/vnd.openxmlformats-officedocument.wordprocessingml.document\", nil\n\tcase \".xls\":\n\t\treturn \"application/vnd.ms-excel\", nil\n\tcase \".xlsx\":\n\t\treturn \"application/vnd.openxmlformats-officedocument.spreadsheetml.sheet\", nil\n\tcase \".ppt\":\n\t\treturn \"application/vnd.ms-powerpoint\", nil\n\tcase \".pptx\":\n\t\treturn \"application/vnd.openxmlformats-officedocument.presentationml.presentation\", nil\n\tcase \".zip\":\n\t\treturn \"application/zip\", nil\n\tcase \".tar\":\n\t\treturn \"application/x-tar\", nil\n\tcase \".mp3\":\n\t\treturn \"audio/mpeg\", nil\n\tcase \".wav\":\n\t\treturn \"audio/wav\", nil\n\tcase \".m4a\":\n\t\treturn \"audio/mp4\", nil\n\tcase \".ogg\":\n\t\treturn \"audio/ogg\", nil\n\tcase \".mp4\":\n\t\treturn \"video/mp4\", nil\n\tcase \".avi\":\n\t\treturn \"video/x-msvideo\", nil\n\tcase \".mov\":\n\t\treturn \"video/quicktime\", nil\n\tcase \".webm\":\n\t\treturn \"video/webm\", nil\n\tcase \".md\", \".mdx\":\n\t\treturn \"text/markdown\", nil\n\tcase \".yao\":\n\t\treturn \"application/yao\", nil\n\tcase \".csv\":\n\t\treturn \"text/csv\", nil\n\t}\n\n\t// Try to detect by MIME package\n\tif contentType := mime.TypeByExtension(ext); contentType != \"\" {\n\t\treturn contentType, nil\n\t}\n\n\t// Return default if not found\n\treturn \"application/octet-stream\", nil\n}\n\n// detectContentTypeFromGzippedFile detects content type by decompressing and reading gzipped file\nfunc detectContentTypeFromGzippedFile(gzippedFilePath string) (string, error) {\n\tfile, err := os.Open(gzippedFilePath)\n\tif err != nil {\n\t\treturn \"\", err\n\t}\n\tdefer file.Close()\n\n\t// Create gzip reader\n\tgzipReader, err := gzip.NewReader(file)\n\tif err != nil {\n\t\treturn \"\", err\n\t}\n\tdefer gzipReader.Close()\n\n\t// Read first 512 bytes of decompressed content\n\tbuffer := make([]byte, 512)\n\tn, err := gzipReader.Read(buffer)\n\tif err != nil && err != io.EOF {\n\t\treturn \"\", err\n\t}\n\n\t// Use http.DetectContentType to detect based on decompressed content\n\tcontentType := http.DetectContentType(buffer[:n])\n\treturn contentType, nil\n}\n" }, { "path": "attachment/local/storage_test.go", "content": "package local\n\nimport (\n\t\"bytes\"\n\t\"context\"\n\t\"image\"\n\t\"image/png\"\n\t\"io\"\n\t\"os\"\n\t\"path/filepath\"\n\t\"testing\"\n\n\t\"github.com/google/uuid\"\n\t\"github.com/stretchr/testify/assert\"\n)\n\n// generateTestFileName generates a unique test filename with the given prefix and extension\nfunc generateTestFileName(prefix, ext string) string {\n\treturn prefix + \"-\" + uuid.New().String() + ext\n}\n\nfunc TestLocalStorage(t *testing.T) {\n\t// Create a temporary directory for testing\n\ttempDir, err := os.MkdirTemp(\"\", \"local_storage_test\")\n\tassert.NoError(t, err)\n\tdefer os.RemoveAll(tempDir)\n\n\ttestPath := filepath.Join(tempDir, \"test_storage\")\n\n\tt.Run(\"Create Storage\", func(t *testing.T) {\n\t\tstorage, err := New(map[string]interface{}{\n\t\t\t\"path\": testPath,\n\t\t\t\"compression\": true,\n\t\t})\n\t\tassert.NoError(t, err)\n\t\tassert.NotNil(t, storage)\n\t\tassert.Equal(t, testPath, storage.Path)\n\t\tassert.True(t, storage.Compression)\n\t})\n\n\tt.Run(\"Upload and Download\", func(t *testing.T) {\n\t\tstorage, err := New(map[string]interface{}{\n\t\t\t\"path\": testPath,\n\t\t\t\"compression\": true,\n\t\t})\n\t\tassert.NoError(t, err)\n\n\t\tcontent := []byte(\"test content\")\n\t\treader := bytes.NewReader(content)\n\t\tfileID := generateTestFileName(\"upload-download\", \".txt\")\n\t\t_, err = storage.Upload(context.Background(), fileID, reader, \"text/plain\")\n\t\tassert.NoError(t, err)\n\t\tassert.NotEmpty(t, fileID)\n\n\t\t// Download\n\t\treader2, contentType, err := storage.Download(context.Background(), fileID)\n\t\tassert.NoError(t, err)\n\t\tassert.Contains(t, contentType, \"text/plain\")\n\n\t\tdownloaded, err := io.ReadAll(reader2)\n\t\tassert.NoError(t, err)\n\t\tassert.Equal(t, content, downloaded)\n\t})\n\n\tt.Run(\"Upload and Download Image with Compression\", func(t *testing.T) {\n\t\tstorage, err := New(map[string]interface{}{\n\t\t\t\"path\": testPath,\n\t\t\t\"compression\": true,\n\t\t})\n\t\tassert.NoError(t, err)\n\n\t\t// Create a test image (100x100 pixels - smaller for faster testing)\n\t\timg := image.NewRGBA(image.Rect(0, 0, 100, 100))\n\t\tvar buf bytes.Buffer\n\t\terr = png.Encode(&buf, img)\n\t\tassert.NoError(t, err)\n\n\t\t// Upload\n\t\treader := bytes.NewReader(buf.Bytes())\n\t\tfileID := generateTestFileName(\"image-with-compression\", \".png\")\n\t\t_, err = storage.Upload(context.Background(), fileID, reader, \"image/png\")\n\t\tassert.NoError(t, err)\n\t\tassert.NotEmpty(t, fileID)\n\n\t\t// Download and verify\n\t\treader2, contentType, err := storage.Download(context.Background(), fileID)\n\t\tassert.NoError(t, err)\n\t\tassert.Equal(t, \"image/png\", contentType)\n\n\t\tdownloaded, err := io.ReadAll(reader2)\n\t\tassert.NoError(t, err)\n\n\t\t// Decode the downloaded image\n\t\tdownloadedImg, _, err := image.Decode(bytes.NewReader(downloaded))\n\t\tassert.NoError(t, err)\n\n\t\t// Verify image was processed\n\t\tbounds := downloadedImg.Bounds()\n\t\tassert.True(t, bounds.Dx() > 0)\n\t\tassert.True(t, bounds.Dy() > 0)\n\t})\n\n\tt.Run(\"Upload Image without Compression\", func(t *testing.T) {\n\t\tstorage, err := New(map[string]interface{}{\n\t\t\t\"path\": testPath,\n\t\t\t\"compression\": false,\n\t\t})\n\t\tassert.NoError(t, err)\n\n\t\t// Create a test image (100x100 pixels)\n\t\timg := image.NewRGBA(image.Rect(0, 0, 100, 100))\n\t\tvar buf bytes.Buffer\n\t\terr = png.Encode(&buf, img)\n\t\tassert.NoError(t, err)\n\n\t\t// Upload\n\t\treader := bytes.NewReader(buf.Bytes())\n\t\tfileID := generateTestFileName(\"image-without-compression\", \".png\")\n\t\t_, err = storage.Upload(context.Background(), fileID, reader, \"image/png\")\n\t\tassert.NoError(t, err)\n\t\tassert.NotEmpty(t, fileID)\n\n\t\t// Download and verify\n\t\treader2, contentType, err := storage.Download(context.Background(), fileID)\n\t\tassert.NoError(t, err)\n\t\tassert.Equal(t, \"image/png\", contentType)\n\n\t\tdownloaded, err := io.ReadAll(reader2)\n\t\tassert.NoError(t, err)\n\n\t\t// Decode the downloaded image\n\t\tdownloadedImg, _, err := image.Decode(bytes.NewReader(downloaded))\n\t\tassert.NoError(t, err)\n\n\t\t// Verify dimensions are unchanged\n\t\tbounds := downloadedImg.Bounds()\n\t\tassert.Equal(t, 100, bounds.Dx())\n\t\tassert.Equal(t, 100, bounds.Dy())\n\t})\n\n\tt.Run(\"URL Generation\", func(t *testing.T) {\n\t\tstorage, err := New(map[string]interface{}{\n\t\t\t\"path\": testPath,\n\t\t\t\"compression\": true,\n\t\t})\n\t\tassert.NoError(t, err)\n\n\t\tfileID := \"20240101/test-12345678.txt\"\n\t\turl := storage.URL(context.Background(), fileID)\n\t\texpected := filepath.Join(testPath, fileID)\n\t\tassert.Equal(t, expected, url)\n\t})\n\n\tt.Run(\"Download Non-existent File\", func(t *testing.T) {\n\t\tstorage, err := New(map[string]interface{}{\n\t\t\t\"path\": testPath,\n\t\t\t\"compression\": true,\n\t\t})\n\t\tassert.NoError(t, err)\n\n\t\t_, _, err = storage.Download(context.Background(), \"non-existent.txt\")\n\t\tassert.Error(t, err)\n\t})\n\n\tt.Run(\"Chunked Upload\", func(t *testing.T) {\n\t\tstorage, err := New(map[string]interface{}{\n\t\t\t\"path\": testPath,\n\t\t})\n\t\tassert.NoError(t, err)\n\n\t\tfileID := \"test-chunked.txt\"\n\t\tcontent1 := []byte(\"chunk1\")\n\t\tcontent2 := []byte(\"chunk2\")\n\n\t\t// Upload chunks\n\t\terr = storage.UploadChunk(context.Background(), fileID, 0, bytes.NewReader(content1), \"text/plain\")\n\t\tassert.NoError(t, err)\n\n\t\terr = storage.UploadChunk(context.Background(), fileID, 1, bytes.NewReader(content2), \"text/plain\")\n\t\tassert.NoError(t, err)\n\n\t\t// Merge chunks\n\t\terr = storage.MergeChunks(context.Background(), fileID, 2)\n\t\tassert.NoError(t, err)\n\n\t\t// Download and verify\n\t\treader, contentType, err := storage.Download(context.Background(), fileID)\n\t\tassert.NoError(t, err)\n\t\tassert.Equal(t, \"text/plain\", contentType)\n\n\t\tdownloaded, err := io.ReadAll(reader)\n\t\tassert.NoError(t, err)\n\t\tassert.Equal(t, append(content1, content2...), downloaded)\n\t})\n\n\tt.Run(\"File Operations\", func(t *testing.T) {\n\t\tstorage, err := New(map[string]interface{}{\n\t\t\t\"path\": testPath,\n\t\t})\n\t\tassert.NoError(t, err)\n\n\t\tfileID := \"test-ops.txt\"\n\t\tcontent := []byte(\"test content\")\n\n\t\t// Upload file\n\t\t_, err = storage.Upload(context.Background(), fileID, bytes.NewReader(content), \"text/plain\")\n\t\tassert.NoError(t, err)\n\n\t\t// Check if file exists\n\t\texists := storage.Exists(context.Background(), fileID)\n\t\tassert.True(t, exists)\n\n\t\t// Read file\n\t\treader, err := storage.Reader(context.Background(), fileID)\n\t\tassert.NoError(t, err)\n\t\tdefer reader.Close()\n\n\t\tdata, err := io.ReadAll(reader)\n\t\tassert.NoError(t, err)\n\t\tassert.Equal(t, content, data)\n\n\t\t// Get file content directly\n\t\tdirectContent, err := storage.GetContent(context.Background(), fileID)\n\t\tassert.NoError(t, err)\n\t\tassert.Equal(t, content, directContent)\n\n\t\t// Delete file\n\t\terr = storage.Delete(context.Background(), fileID)\n\t\tassert.NoError(t, err)\n\n\t\t// Check if file no longer exists\n\t\texists = storage.Exists(context.Background(), fileID)\n\t\tassert.False(t, exists)\n\t})\n\n\tt.Run(\"LocalPath\", func(t *testing.T) {\n\t\tstorage, err := New(map[string]interface{}{\n\t\t\t\"path\": testPath,\n\t\t})\n\t\tassert.NoError(t, err)\n\n\t\t// Test different file types to verify content type detection\n\t\ttestFiles := []struct {\n\t\t\text string\n\t\t\tcontent []byte\n\t\t\tcontentType string\n\t\t\texpectedCT string\n\t\t}{\n\t\t\t{\".txt\", []byte(\"Hello World\"), \"text/plain\", \"text/plain\"},\n\t\t\t{\".json\", []byte(`{\"key\": \"value\"}`), \"application/json\", \"application/json\"},\n\t\t\t{\".html\", []byte(\"Test\"), \"text/html\", \"text/html\"},\n\t\t\t{\".csv\", []byte(\"col1,col2\\nval1,val2\"), \"text/csv\", \"text/csv\"},\n\t\t\t{\".md\", []byte(\"# Markdown Content\"), \"text/markdown\", \"text/markdown\"},\n\t\t\t{\".yao\", []byte(\"yao file content\"), \"application/yao\", \"application/yao\"},\n\t\t}\n\n\t\tfor _, tf := range testFiles {\n\t\t\t// Generate unique filename with UUID to avoid conflicts\n\t\t\tfileName := generateTestFileName(\"localpath-test\", tf.ext)\n\n\t\t\t// Upload file\n\t\t\t_, err = storage.Upload(context.Background(), fileName, bytes.NewReader(tf.content), tf.contentType)\n\t\t\tassert.NoError(t, err, \"Failed to upload %s\", fileName)\n\n\t\t\t// Get local path and content type\n\t\t\tlocalPath, detectedCT, err := storage.LocalPath(context.Background(), fileName)\n\t\t\tassert.NoError(t, err, \"Failed to get local path for %s\", fileName)\n\t\t\tassert.NotEmpty(t, localPath, \"Local path should not be empty for %s\", fileName)\n\t\t\tassert.Equal(t, tf.expectedCT, detectedCT, \"Content type mismatch for %s\", fileName)\n\n\t\t\t// Verify the path is absolute\n\t\t\tassert.True(t, filepath.IsAbs(localPath), \"Path should be absolute for %s\", fileName)\n\n\t\t\t// Verify the file exists at the returned path\n\t\t\t_, err = os.Stat(localPath)\n\t\t\tassert.NoError(t, err, \"File should exist at local path for %s\", fileName)\n\n\t\t\t// Verify file content\n\t\t\tfileContent, err := os.ReadFile(localPath)\n\t\t\tassert.NoError(t, err, \"Failed to read file at local path for %s\", fileName)\n\t\t\tassert.Equal(t, tf.content, fileContent, \"File content mismatch for %s\", fileName)\n\t\t}\n\t})\n\n\tt.Run(\"LocalPath_NonExistentFile\", func(t *testing.T) {\n\t\tstorage, err := New(map[string]interface{}{\n\t\t\t\"path\": testPath,\n\t\t})\n\t\tassert.NoError(t, err)\n\n\t\t// Test with non-existent file\n\t\t_, _, err = storage.LocalPath(context.Background(), \"non-existent.txt\")\n\t\tassert.Error(t, err)\n\t\tassert.Contains(t, err.Error(), \"file not found\")\n\t})\n\n\tt.Run(\"LocalPath_ContentDetection\", func(t *testing.T) {\n\t\tstorage, err := New(map[string]interface{}{\n\t\t\t\"path\": testPath,\n\t\t})\n\t\tassert.NoError(t, err)\n\n\t\t// Upload a file without extension but with recognizable content\n\t\thtmlContent := []byte(\"Test

Hello

\")\n\t\t_, err = storage.Upload(context.Background(), \"noext\", bytes.NewReader(htmlContent), \"application/octet-stream\")\n\t\tassert.NoError(t, err)\n\n\t\t// Get local path - should detect HTML content type\n\t\tlocalPath, contentType, err := storage.LocalPath(context.Background(), \"noext\")\n\t\tassert.NoError(t, err)\n\t\tassert.NotEmpty(t, localPath)\n\t\t// Content detection should identify this as HTML\n\t\tassert.Equal(t, \"text/html; charset=utf-8\", contentType)\n\t})\n}\n" }, { "path": "attachment/manager.go", "content": "package attachment\n\nimport (\n\t\"bytes\"\n\t\"context\"\n\t\"crypto/md5\"\n\t\"encoding/base64\"\n\t\"encoding/hex\"\n\t\"fmt\"\n\t\"io\"\n\t\"mime\"\n\t\"mime/multipart\"\n\t\"net/http\"\n\t\"net/textproto\"\n\t\"os\"\n\t\"path/filepath\"\n\t\"reflect\"\n\t\"strconv\"\n\t\"strings\"\n\t\"sync\"\n\t\"time\"\n\n\t\"github.com/yaoapp/gou/fs\"\n\t\"github.com/yaoapp/gou/model\"\n\t\"github.com/yaoapp/kun/log\"\n\t\"github.com/yaoapp/yao/attachment/local\"\n\t\"github.com/yaoapp/yao/attachment/s3\"\n\t\"github.com/yaoapp/yao/config\"\n)\n\n// Ensure Manager implements FileManager interface\nvar _ FileManager = (*Manager)(nil)\n\n// Managers the managers\nvar Managers = map[string]*Manager{}\nvar uploadChunks = sync.Map{}\n\n// UploadChunk is the chunk data\ntype UploadChunk struct {\n\tLast int\n\tTotal int64\n\tChunksize int64\n\tTotalChunks int64\n\t// Cache metadata from first chunk to avoid inconsistencies\n\tContentType string\n\tFilename string\n\tUserPath string\n\tCompressImage bool\n\tCompressSize int\n}\n\n// Parse parses an attachment wrapper string and returns uploader name and file ID\n// Format: __://\n// Example: __yao.attachment://ccd472d11feb96e03a3fc468f494045c\n// Returns (uploader, fileID, isWrapper)\nfunc Parse(value string) (string, string, bool) {\n\tif !strings.HasPrefix(value, \"__\") {\n\t\treturn \"\", value, false\n\t}\n\n\t// Exclude common protocols (ftp, http, https, etc.)\n\texcludedProtocols := []string{\"__ftp://\", \"__http://\", \"__https://\", \"__ws://\", \"__wss://\"}\n\tfor _, protocol := range excludedProtocols {\n\t\tif strings.HasPrefix(value, protocol) {\n\t\t\treturn \"\", value, false\n\t\t}\n\t}\n\n\t// Split by ://\n\tparts := strings.SplitN(value, \"://\", 2)\n\tif len(parts) != 2 {\n\t\treturn \"\", value, false\n\t}\n\n\tuploader := parts[0] // Keep the __ prefix as it's part of the manager name\n\tfileID := parts[1]\n\n\treturn uploader, fileID, true\n}\n\n// Base64 processes a wrapper value and converts it to Base64 if it's an attachment wrapper\n// If the value is not a wrapper, it returns the original value\n// Special case: if value looks like a file path, it will try to read from fs data\n// Optional parameter dataURI: if true, returns data URI format (data:image/png;base64,...)\nfunc Base64(ctx context.Context, value string, dataURI ...bool) string {\n\tuseDataURI := false\n\tif len(dataURI) > 0 {\n\t\tuseDataURI = dataURI[0]\n\t}\n\n\tuploader, fileID, isWrapper := Parse(value)\n\tif !isWrapper {\n\t\t// Try to read as file path from fs data\n\t\tif base64Data := readFilePathAsBase64(value, useDataURI); base64Data != \"\" {\n\t\t\treturn base64Data\n\t\t}\n\t\treturn value\n\t}\n\n\t// Get the manager\n\tmanager, exists := Managers[uploader]\n\tif !exists {\n\t\treturn value\n\t}\n\n\t// Get file info to determine content type\n\tvar contentType string\n\tif useDataURI {\n\t\tfileInfo, err := manager.Info(ctx, fileID)\n\t\tif err == nil && fileInfo != nil {\n\t\t\tcontentType = fileInfo.ContentType\n\t\t}\n\t}\n\n\t// Read the file as Base64\n\tbase64Data, err := manager.ReadBase64(ctx, fileID)\n\tif err != nil {\n\t\treturn value\n\t}\n\n\t// Return with data URI prefix if requested\n\tif useDataURI && contentType != \"\" {\n\t\treturn fmt.Sprintf(\"data:%s;base64,%s\", contentType, base64Data)\n\t}\n\n\treturn base64Data\n}\n\n// readFilePathAsBase64 reads a file from fs data and returns Base64 encoded content\n// Returns empty string if file doesn't exist or can't be read\n// If dataURI is true, returns data URI format with mime type detection\nfunc readFilePathAsBase64(path string, dataURI bool) string {\n\t// Check if path looks like a file path (contains / or \\)\n\tif !strings.Contains(path, \"/\") && !strings.Contains(path, \"\\\\\") {\n\t\treturn \"\"\n\t}\n\n\t// Try to get fs data\n\tdataFS, err := fs.Get(\"data\")\n\tif err != nil || dataFS == nil {\n\t\treturn \"\"\n\t}\n\n\t// Check if file exists\n\texists, err := dataFS.Exists(path)\n\tif err != nil || !exists {\n\t\treturn \"\"\n\t}\n\n\t// Read file content\n\tcontent, err := dataFS.ReadFile(path)\n\tif err != nil {\n\t\treturn \"\"\n\t}\n\n\t// Encode to Base64\n\tbase64Str := base64.StdEncoding.EncodeToString(content)\n\n\t// Return with data URI prefix if requested\n\tif dataURI {\n\t\t// Detect content type from file extension or content\n\t\tcontentType := detectContentType(path, content)\n\t\tif contentType != \"\" {\n\t\t\treturn fmt.Sprintf(\"data:%s;base64,%s\", contentType, base64Str)\n\t\t}\n\t}\n\n\treturn base64Str\n}\n\n// detectContentType detects the MIME type from file path and content\nfunc detectContentType(path string, content []byte) string {\n\t// First try to get from file extension\n\text := filepath.Ext(path)\n\tif ext != \"\" {\n\t\tmimeType := mime.TypeByExtension(ext)\n\t\tif mimeType != \"\" {\n\t\t\treturn mimeType\n\t\t}\n\t}\n\n\t// Fallback to detecting from content (first 512 bytes)\n\tif len(content) > 0 {\n\t\tdetectSize := len(content)\n\t\tif detectSize > 512 {\n\t\t\tdetectSize = 512\n\t\t}\n\t\treturn http.DetectContentType(content[:detectSize])\n\t}\n\n\treturn \"\"\n}\n\n// GetHeader gets the header from the file header and request header\nfunc GetHeader(requestHeader http.Header, fileHeader textproto.MIMEHeader, size int64) *FileHeader {\n\n\t// Convert the header to a FileHeader\n\theader := &FileHeader{FileHeader: &multipart.FileHeader{Header: make(map[string][]string), Size: size}}\n\n\tfor key, values := range fileHeader {\n\t\tfor _, value := range values {\n\t\t\theader.Header.Set(key, value)\n\t\t}\n\t}\n\n\t// Set Content-Sync, Content-Uid, Content-Range\n\tif requestHeader.Get(\"Content-Sync\") != \"\" {\n\t\theader.Header.Set(\"Content-Sync\", requestHeader.Get(\"Content-Sync\"))\n\t}\n\n\t// Set Content-Uid\n\tif requestHeader.Get(\"Content-Uid\") != \"\" {\n\t\theader.Header.Set(\"Content-Uid\", requestHeader.Get(\"Content-Uid\"))\n\t}\n\n\t// Set Content-Range\n\tif requestHeader.Get(\"Content-Range\") != \"\" {\n\t\theader.Header.Set(\"Content-Range\", requestHeader.Get(\"Content-Range\"))\n\t}\n\n\treturn header\n}\n\n// Register registers a global attachment manager\nfunc Register(name string, driver string, option ManagerOption) (*Manager, error) {\n\n\t// Create a new manager\n\tmanager, err := New(option)\n\tif err != nil {\n\t\treturn nil, err\n\t}\n\n\t// Set the manager name\n\tmanager.Name = name\n\n\t// Register the manager\n\tManagers[name] = manager\n\treturn manager, nil\n}\n\n// RegisterDefault registers a default attachment manager\nfunc RegisterDefault(name string) (*Manager, error) {\n\n\toption := ManagerOption{\n\t\tDriver: \"local\",\n\t\tOptions: map[string]interface{}{\"path\": filepath.Join(config.Conf.DataRoot, name)},\n\t\tMaxSize: \"50M\",\n\t\tChunkSize: \"2M\",\n\t\tAllowedTypes: []string{\n\t\t\t\"text/*\",\n\t\t\t\"image/*\",\n\t\t\t\"video/*\",\n\t\t\t\"audio/*\",\n\t\t\t\"application/x-zip-compressed\",\n\t\t\t\"application/x-tar\",\n\t\t\t\"application/x-gzip\",\n\t\t\t\"application/yao\",\n\t\t\t\"application/zip\",\n\t\t\t\"application/pdf\",\n\t\t\t\"application/json\",\n\t\t\t\"application/vnd.openxmlformats-officedocument.wordprocessingml.document\",\n\t\t\t\"application/vnd.openxmlformats-officedocument.spreadsheetml.sheet\",\n\t\t\t\"application/vnd.openxmlformats-officedocument.presentationml.presentation\",\n\t\t\t\"application/vnd.openxmlformats-officedocument.presentationml.slideshow\",\n\t\t\t\".md\",\n\t\t\t\".txt\",\n\t\t\t\".csv\",\n\t\t\t\".xls\",\n\t\t\t\".xlsx\",\n\t\t\t\".ppt\",\n\t\t\t\".pptx\",\n\t\t\t\".doc\",\n\t\t\t\".docx\",\n\t\t\t\".mdx\",\n\t\t\t\".m4a\",\n\t\t\t\".mp3\",\n\t\t\t\".mp4\",\n\t\t\t\".wav\",\n\t\t\t\".webm\",\n\t\t\t\".yao\",\n\t\t},\n\t}\n\treturn Register(name, option.Driver, option)\n}\n\n// ReplaceEnv replaces the environment variables in the options\nfunc (option *ManagerOption) ReplaceEnv(root string) {\n\tif option.Options != nil {\n\t\t// Replace the environment variables in the options\n\t\tfor k, v := range option.Options {\n\t\t\tif iv, ok := v.(string); ok {\n\t\t\t\tif strings.HasPrefix(iv, \"$ENV.\") {\n\t\t\t\t\tiv = os.ExpandEnv(fmt.Sprintf(\"${%s}\", strings.TrimPrefix(iv, \"$ENV.\")))\n\t\t\t\t\toption.Options[k] = iv\n\t\t\t\t}\n\n\t\t\t\t// Path\n\t\t\t\tif k == \"path\" {\n\t\t\t\t\tiv = strings.TrimPrefix(iv, \"/\")\n\t\t\t\t\toption.Options[k] = filepath.Join(root, iv)\n\t\t\t\t}\n\n\t\t\t}\n\t\t}\n\t}\n}\n\n// New creates a new attachment manager\nfunc New(option ManagerOption) (*Manager, error) {\n\tmanager := &Manager{\n\t\tManagerOption: option,\n\t\tallowedTypes: allowedType{mapping: make(map[string]bool),\n\t\t\twildcards: []string{},\n\t\t}}\n\n\tswitch strings.ToLower(option.Driver) {\n\tcase \"local\":\n\t\tstorage, err := local.New(option.Options)\n\t\tif err != nil {\n\t\t\treturn nil, err\n\t\t}\n\t\tmanager.storage = storage\n\n\tcase \"s3\":\n\t\tstorage, err := s3.New(option.Options)\n\t\tif err != nil {\n\t\t\treturn nil, err\n\t\t}\n\t\tmanager.storage = storage\n\n\tdefault:\n\t\treturn nil, fmt.Errorf(\"driver %s does not support\", option.Driver)\n\t}\n\n\t// Max size\n\tif option.MaxSize != \"\" {\n\t\tmaxsize, err := getSize(option.MaxSize)\n\t\tif err != nil {\n\t\t\treturn nil, err\n\t\t}\n\t\tmanager.maxsize = maxsize\n\t}\n\n\t// Chunk size\n\tif option.ChunkSize != \"\" {\n\t\tchunsize, err := getSize(option.ChunkSize)\n\t\tif err != nil {\n\t\t\treturn nil, err\n\t\t}\n\t\tmanager.chunsize = chunsize\n\t}\n\n\t// init allowedTypes\n\tif len(option.AllowedTypes) > 0 {\n\t\tfor _, t := range option.AllowedTypes {\n\t\t\tt = strings.TrimSpace(t)\n\t\t\tif strings.HasSuffix(t, \"*\") {\n\t\t\t\tmanager.allowedTypes.wildcards = append(manager.allowedTypes.wildcards, t)\n\t\t\t\tcontinue\n\t\t\t}\n\t\t\tmanager.allowedTypes.mapping[t] = true\n\t\t}\n\t}\n\n\treturn manager, nil\n}\n\n// LocalPath gets the local path of the file\nfunc (manager Manager) LocalPath(ctx context.Context, fileID string) (string, string, error) {\n\t// Get the real storage path from database\n\tstoragePath, err := manager.getStoragePathFromDatabase(ctx, fileID)\n\tif err != nil {\n\t\treturn \"\", \"\", err\n\t}\n\n\t// Call the storage implementation\n\treturn manager.storage.LocalPath(ctx, storagePath)\n}\n\n// Upload uploads a file, Content-Sync must be true for chunked upload\nfunc (manager Manager) Upload(ctx context.Context, fileheader *FileHeader, reader io.Reader, option UploadOption) (*File, error) {\n\n\tfile, err := manager.makeFile(fileheader, option)\n\tif err != nil {\n\t\treturn nil, err\n\t}\n\n\t// Handle chunked upload\n\tif fileheader.IsChunk() {\n\t\tstart, end, total, err := fileheader.GetChunkInfo()\n\t\tif err != nil {\n\t\t\treturn nil, fmt.Errorf(\"invalid chunk info: %w\", err)\n\t\t}\n\n\t\t// Store the chunk info\n\t\tchunkIndex := 0\n\t\tif start == 0 {\n\t\t\tchunksize := end - start + 1\n\t\t\ttotalChunks := (total + chunksize - 1) / chunksize\n\t\t\tuploadChunks.LoadOrStore(file.ID, &UploadChunk{\n\t\t\t\tLast: chunkIndex,\n\t\t\t\tTotal: total,\n\t\t\t\tChunksize: chunksize,\n\t\t\t\tTotalChunks: totalChunks,\n\t\t\t\t// Cache metadata from first chunk\n\t\t\t\tContentType: file.ContentType,\n\t\t\t\tFilename: file.Filename,\n\t\t\t\tUserPath: file.UserPath,\n\t\t\t\tCompressImage: option.CompressImage,\n\t\t\t\tCompressSize: option.CompressSize,\n\t\t\t})\n\t\t}\n\n\t\t// Update the chunk index\n\t\tv, ok := uploadChunks.Load(file.ID)\n\t\tif !ok {\n\t\t\treturn nil, fmt.Errorf(\"chunk data not found\")\n\t\t}\n\n\t\tchunkdata := v.(*UploadChunk)\n\n\t\t// Update the chunk index\n\t\tif start != 0 {\n\t\t\tchunkIndex = chunkdata.Last + 1\n\t\t\tchunkdata.Last = chunkIndex\n\t\t\tuploadChunks.Store(file.ID, chunkdata)\n\n\t\t\t// For non-first chunks, use cached metadata from first chunk\n\t\t\tfile.ContentType = chunkdata.ContentType\n\t\t\tfile.Filename = chunkdata.Filename\n\t\t\tfile.UserPath = chunkdata.UserPath\n\t\t}\n\n\t\t// Apply gzip compression if requested\n\t\tif option.Gzip {\n\t\t\tcompressed, err := GzipFromReader(reader)\n\t\t\tif err != nil {\n\t\t\t\treturn nil, fmt.Errorf(\"failed to gzip chunk: %w\", err)\n\t\t\t}\n\t\t\treader = bytes.NewReader(compressed)\n\n\t\t}\n\n\t\t// Upload chunk using the storage path\n\t\terr = manager.storage.UploadChunk(ctx, file.Path, chunkIndex, reader, file.ContentType)\n\t\tif err != nil {\n\t\t\treturn nil, err\n\t\t}\n\n\t\t// Save to database on first chunk only\n\t\tif start == 0 {\n\t\t\tfile.Status = \"uploading\"\n\t\t\terr = manager.saveFileToDatabase(ctx, file, file.Path, option)\n\t\t\tif err != nil {\n\t\t\t\treturn nil, fmt.Errorf(\"failed to create database record for chunked upload: %w\", err)\n\t\t\t}\n\t\t}\n\n\t\t// Fix the file size, the file size is the sum of all chunks\n\t\tfile.Bytes = chunkIndex * int(chunkdata.Chunksize)\n\t\tfile.Status = \"uploading\"\n\n\t\t// If this is the last chunk, merge all chunks\n\t\tif fileheader.Complete() {\n\t\t\terr = manager.storage.MergeChunks(ctx, file.Path, int(chunkdata.TotalChunks))\n\t\t\tif err != nil {\n\t\t\t\treturn nil, err\n\t\t\t}\n\n\t\t\t// Set initial file size from chunks\n\t\t\tfile.Bytes = int(chunkdata.Total)\n\n\t\t\t// Apply image compression if requested and it's the final file\n\t\t\t// Use cached compress options from first chunk\n\t\t\tif chunkdata.CompressImage && strings.HasPrefix(file.ContentType, \"image/\") {\n\t\t\t\t// Create a temporary option with cached compress size\n\t\t\t\tcompressOption := UploadOption{\n\t\t\t\t\tCompressSize: chunkdata.CompressSize,\n\t\t\t\t}\n\t\t\t\tcompressedBytes, err := manager.compressStoredImageAndGetSize(ctx, file, compressOption)\n\t\t\t\tif err != nil {\n\t\t\t\t\treturn nil, err\n\t\t\t\t}\n\t\t\t\t// Update file size to compressed size\n\t\t\t\tfile.Bytes = compressedBytes\n\t\t\t}\n\n\t\t\t// Remove the chunk data\n\t\t\tuploadChunks.Delete(file.ID)\n\n\t\t\t// Update status to uploaded\n\t\t\tfile.Status = \"uploaded\"\n\n\t\t\t// Update only bytes and status for the last chunk\n\t\t\terr = manager.saveFileToDatabase(ctx, file, file.Path, option)\n\t\t\tif err != nil {\n\t\t\t\treturn nil, fmt.Errorf(\"failed to update chunked file status: %w\", err)\n\t\t\t}\n\t\t}\n\n\t\treturn file, nil\n\t}\n\n\t// Handle single file upload\n\tvar finalReader io.Reader = reader\n\n\t// Apply gzip compression if requested\n\tif option.Gzip {\n\t\tcompressed, err := GzipFromReader(reader)\n\t\tif err != nil {\n\t\t\treturn nil, fmt.Errorf(\"failed to gzip file: %w\", err)\n\t\t}\n\n\t\tfinalReader = bytes.NewReader(compressed)\n\t}\n\n\t// Apply image compression if requested\n\tif option.CompressImage && strings.HasPrefix(file.ContentType, \"image/\") {\n\t\tsize := option.CompressSize\n\t\tif size == 0 {\n\t\t\tsize = 1920\n\t\t}\n\n\t\t// Read original data for fallback\n\t\tvar originalData []byte\n\t\tvar err error\n\n\t\t// If gzip was applied, we need to decompress first\n\t\tif option.Gzip {\n\t\t\tdata, err := io.ReadAll(finalReader)\n\t\t\tif err != nil {\n\t\t\t\treturn nil, err\n\t\t\t}\n\t\t\tdecompressed, err := Gunzip(data)\n\t\t\tif err != nil {\n\t\t\t\treturn nil, err\n\t\t\t}\n\t\t\toriginalData = decompressed\n\t\t\tfinalReader = bytes.NewReader(decompressed)\n\t\t} else {\n\t\t\toriginalData, err = io.ReadAll(finalReader)\n\t\t\tif err != nil {\n\t\t\t\treturn nil, err\n\t\t\t}\n\t\t\tfinalReader = bytes.NewReader(originalData)\n\t\t}\n\n\t\t// Try to compress the image with failback mechanism\n\t\tcompressed, err := CompressImage(finalReader, file.ContentType, size)\n\t\tif err != nil {\n\t\t\t// Log the error and use original file as fallback\n\t\t\tlog.Warn(\"Failed to compress image (content-type: %s, file: %s): %v. Using original file.\",\n\t\t\t\tfile.ContentType, file.Filename, err)\n\t\t\t// Use original data\n\t\t\tcompressed = originalData\n\t\t}\n\n\t\t// Re-apply gzip if it was requested\n\t\tif option.Gzip {\n\t\t\tgzipped, err := Gzip(compressed)\n\t\t\tif err != nil {\n\t\t\t\treturn nil, err\n\t\t\t}\n\t\t\tfinalReader = bytes.NewReader(gzipped)\n\t\t} else {\n\t\t\tfinalReader = bytes.NewReader(compressed)\n\t\t}\n\t}\n\n\t// Upload the file to storage using the generated storage path\n\tactualStoragePath, err := manager.storage.Upload(ctx, file.Path, finalReader, file.ContentType)\n\tif err != nil {\n\t\treturn nil, err\n\t}\n\n\t// Update the actual storage path if storage returns a different path\n\tif actualStoragePath != \"\" && actualStoragePath != file.Path {\n\t\tfile.Path = actualStoragePath\n\t}\n\n\t// Update the file status\n\tfile.Status = \"uploaded\"\n\n\t// Save file information to database\n\terr = manager.saveFileToDatabase(ctx, file, file.Path, option)\n\tif err != nil {\n\t\treturn nil, fmt.Errorf(\"failed to save file to database: %w\", err)\n\t}\n\n\treturn file, nil\n}\n\n// compressStoredImageAndGetSize compresses the stored image and returns the compressed size\nfunc (manager Manager) compressStoredImageAndGetSize(ctx context.Context, file *File, option UploadOption) (int, error) {\n\t// Download the stored file using storage path\n\treader, err := manager.storage.Reader(ctx, file.Path)\n\tif err != nil {\n\t\treturn 0, err\n\t}\n\tdefer reader.Close()\n\n\tsize := option.CompressSize\n\tif size == 0 {\n\t\tsize = 1920\n\t}\n\n\t// Read original data for fallback\n\toriginalData, err := io.ReadAll(reader)\n\tif err != nil {\n\t\treturn 0, err\n\t}\n\n\t// Try to compress the image with failback mechanism\n\tcompressed, err := CompressImage(bytes.NewReader(originalData), file.ContentType, size)\n\tif err != nil {\n\t\t// Log the error and keep original file\n\t\tlog.Warn(\"Failed to compress stored image (content-type: %s, file: %s): %v. Keeping original file.\",\n\t\t\tfile.ContentType, file.Filename, err)\n\t\t// File is already stored (merged chunks), just return original size\n\t\treturn len(originalData), nil\n\t}\n\n\t// Re-upload the compressed image using storage path\n\t_, err = manager.storage.Upload(ctx, file.Path, bytes.NewReader(compressed), file.ContentType)\n\tif err != nil {\n\t\treturn 0, err\n\t}\n\n\t// Return the compressed size\n\treturn len(compressed), nil\n}\n\n// Download downloads a file\nfunc (manager Manager) Download(ctx context.Context, fileID string) (*FileResponse, error) {\n\t// Get real storage path from database\n\tstoragePath, err := manager.getStoragePathFromDatabase(ctx, fileID)\n\tif err != nil {\n\t\treturn nil, err\n\t}\n\n\treader, contentType, err := manager.storage.Download(ctx, storagePath)\n\tif err != nil {\n\t\treturn nil, err\n\t}\n\n\textension := filepath.Ext(storagePath)\n\tif extension == \"\" {\n\t\t// Try to get extension from content type\n\t\textensions, err := mime.ExtensionsByType(contentType)\n\t\tif err == nil && len(extensions) > 0 {\n\t\t\textension = extensions[0]\n\t\t}\n\t}\n\n\treturn &FileResponse{\n\t\tReader: reader,\n\t\tContentType: contentType,\n\t\tExtension: extension,\n\t}, nil\n}\n\n// Read reads a file and returns the content as bytes\nfunc (manager Manager) Read(ctx context.Context, fileID string) ([]byte, error) {\n\t// Get file info from database to check if it's gzipped\n\tfile, err := manager.getFileFromDatabase(ctx, fileID)\n\tif err != nil {\n\t\treturn nil, err\n\t}\n\n\treader, err := manager.storage.Reader(ctx, file.Path)\n\tif err != nil {\n\t\treturn nil, err\n\t}\n\tdefer reader.Close()\n\n\tdata, err := io.ReadAll(reader)\n\tif err != nil {\n\t\treturn nil, err\n\t}\n\n\t// Storage layer already handles gzip decompression for .gz files\n\t// No need to decompress again at Manager level\n\n\treturn data, nil\n}\n\n// ReadBase64 reads a file and returns the content as base64 encoded string\nfunc (manager Manager) ReadBase64(ctx context.Context, fileID string) (string, error) {\n\tdata, err := manager.Read(ctx, fileID)\n\tif err != nil {\n\t\treturn \"\", err\n\t}\n\n\treturn base64.StdEncoding.EncodeToString(data), nil\n}\n\n// Info retrieves complete file information from database by file ID\nfunc (manager Manager) Info(ctx context.Context, fileID string) (*File, error) {\n\treturn manager.getFileFromDatabase(ctx, fileID)\n}\n\n// List retrieves files from database with pagination and filtering\nfunc (manager Manager) List(ctx context.Context, option ListOption) (*ListResult, error) {\n\tm := model.Select(\"__yao.attachment\")\n\n\t// Set default values\n\tpage := option.Page\n\tif page <= 0 {\n\t\tpage = 1\n\t}\n\n\tpageSize := option.PageSize\n\tif pageSize <= 0 {\n\t\tpageSize = 20\n\t}\n\n\t// Build query parameters\n\tqueryParam := model.QueryParam{}\n\n\t// Add select fields\n\tif len(option.Select) > 0 {\n\t\tqueryParam.Select = make([]interface{}, 0, len(option.Select))\n\t\tfor _, field := range option.Select {\n\t\t\tqueryParam.Select = append(queryParam.Select, field)\n\t\t}\n\t} else {\n\t\t// Default: exclude the 'content' field (which may contain large text data)\n\t\t// Only include it if explicitly requested in Select\n\t\tqueryParam.Select = []interface{}{\n\t\t\t\"id\", \"file_id\", \"uploader\", \"content_type\", \"name\", \"url\", \"description\",\n\t\t\t\"type\", \"user_path\", \"path\", \"groups\", \"gzip\", \"bytes\", \"status\",\n\t\t\t\"progress\", \"error\", \"preset\", \"public\", \"share\",\n\t\t\t\"created_at\", \"updated_at\", \"deleted_at\",\n\t\t\t\"__yao_created_by\", \"__yao_updated_by\", \"__yao_team_id\", \"__yao_tenant_id\",\n\t\t}\n\t}\n\n\t// Add filters\n\tif len(option.Filters) > 0 {\n\t\tqueryParam.Wheres = make([]model.QueryWhere, 0, len(option.Filters))\n\t\tfor field, value := range option.Filters {\n\t\t\twhere := model.QueryWhere{\n\t\t\t\tColumn: field,\n\t\t\t\tValue: value,\n\t\t\t}\n\n\t\t\t// Handle special operators for wildcard matching\n\t\t\tif strValue, ok := value.(string); ok {\n\t\t\t\tif strings.Contains(strValue, \"*\") {\n\t\t\t\t\t// Wildcard matching for LIKE queries\n\t\t\t\t\twhere.OP = \"like\"\n\t\t\t\t\twhere.Value = strings.ReplaceAll(strValue, \"*\", \"%\")\n\t\t\t\t}\n\t\t\t}\n\n\t\t\tqueryParam.Wheres = append(queryParam.Wheres, where)\n\t\t}\n\t}\n\n\t// Add advanced where clauses (for permission filtering, etc.)\n\tif len(option.Wheres) > 0 {\n\t\tif queryParam.Wheres == nil {\n\t\t\tqueryParam.Wheres = make([]model.QueryWhere, 0, len(option.Wheres))\n\t\t}\n\t\tqueryParam.Wheres = append(queryParam.Wheres, option.Wheres...)\n\t}\n\n\t// Add ordering\n\tif option.OrderBy != \"\" {\n\t\t// Parse order by string like \"created_at desc\" or \"name asc\"\n\t\tparts := strings.Fields(option.OrderBy)\n\t\tif len(parts) >= 1 {\n\t\t\torderField := parts[0]\n\t\t\torderDirection := \"asc\"\n\t\t\tif len(parts) >= 2 {\n\t\t\t\torderDirection = strings.ToLower(parts[1])\n\t\t\t}\n\n\t\t\tqueryParam.Orders = []model.QueryOrder{\n\t\t\t\t{\n\t\t\t\t\tColumn: orderField,\n\t\t\t\t\tOption: orderDirection,\n\t\t\t\t},\n\t\t\t}\n\t\t}\n\t} else {\n\t\t// Default order by created_at desc\n\t\tqueryParam.Orders = []model.QueryOrder{\n\t\t\t{\n\t\t\t\tColumn: \"created_at\",\n\t\t\t\tOption: \"desc\",\n\t\t\t},\n\t\t}\n\t}\n\n\t// Use model's built-in Paginate method\n\tresult, err := m.Paginate(queryParam, page, pageSize)\n\tif err != nil {\n\t\treturn nil, fmt.Errorf(\"failed to paginate files: %w\", err)\n\t}\n\n\t// Extract pagination info from result\n\ttotal := int64(0)\n\tif totalInterface, ok := result[\"total\"]; ok {\n\t\tif totalInt, ok := totalInterface.(int); ok {\n\t\t\ttotal = int64(totalInt)\n\t\t} else if totalInt64, ok := totalInterface.(int64); ok {\n\t\t\ttotal = totalInt64\n\t\t}\n\t}\n\n\t// Extract data from result - handle maps.MapStrAny type\n\tvar records []map[string]interface{}\n\tif dataInterface, ok := result[\"data\"]; ok {\n\t\t// The data is of type []maps.MapStrAny, need to convert\n\t\tif dataSlice, ok := dataInterface.([]interface{}); ok {\n\t\t\trecords = make([]map[string]interface{}, len(dataSlice))\n\t\t\tfor i, item := range dataSlice {\n\t\t\t\tif record, ok := item.(map[string]interface{}); ok {\n\t\t\t\t\trecords[i] = record\n\t\t\t\t}\n\t\t\t}\n\t\t} else {\n\t\t\t// Try to handle it as the actual type returned by gou using reflection\n\t\t\tdataValue := reflect.ValueOf(dataInterface)\n\t\t\tif dataValue.Kind() == reflect.Slice {\n\t\t\t\tlength := dataValue.Len()\n\t\t\t\trecords = make([]map[string]interface{}, length)\n\t\t\t\tfor i := 0; i < length; i++ {\n\t\t\t\t\titem := dataValue.Index(i).Interface()\n\t\t\t\t\t// Convert the item to map[string]interface{} using reflection\n\t\t\t\t\tif itemValue := reflect.ValueOf(item); itemValue.Kind() == reflect.Map {\n\t\t\t\t\t\trecord := make(map[string]interface{})\n\t\t\t\t\t\tfor _, key := range itemValue.MapKeys() {\n\t\t\t\t\t\t\tif keyStr := key.String(); keyStr != \"\" {\n\t\t\t\t\t\t\t\trecord[keyStr] = itemValue.MapIndex(key).Interface()\n\t\t\t\t\t\t\t}\n\t\t\t\t\t\t}\n\t\t\t\t\t\trecords[i] = record\n\t\t\t\t\t}\n\t\t\t\t}\n\t\t\t}\n\t\t}\n\t}\n\n\t// Convert records to File structs\n\tfiles := make([]*File, 0, len(records))\n\tfor _, record := range records {\n\t\tfile := &File{}\n\n\t\t// Map required fields\n\t\tif fileID, ok := record[\"file_id\"].(string); ok {\n\t\t\tfile.ID = fileID\n\t\t}\n\t\tif name, ok := record[\"name\"].(string); ok {\n\t\t\tfile.Filename = name\n\t\t}\n\t\tif contentType, ok := record[\"content_type\"].(string); ok {\n\t\t\tfile.ContentType = contentType\n\t\t}\n\t\tif status, ok := record[\"status\"].(string); ok {\n\t\t\tfile.Status = status\n\t\t}\n\n\t\t// Map optional fields\n\t\tif userPath, ok := record[\"user_path\"].(string); ok {\n\t\t\tfile.UserPath = userPath\n\t\t}\n\t\tif path, ok := record[\"path\"].(string); ok {\n\t\t\tfile.Path = path\n\t\t}\n\t\tif bytes, ok := record[\"bytes\"].(int64); ok {\n\t\t\tfile.Bytes = int(bytes)\n\t\t} else if bytesInt, ok := record[\"bytes\"].(int); ok {\n\t\t\tfile.Bytes = bytesInt\n\t\t}\n\t\tif createdAt, ok := record[\"created_at\"].(int64); ok {\n\t\t\tfile.CreatedAt = int(createdAt)\n\t\t} else if createdAtInt, ok := record[\"created_at\"].(int); ok {\n\t\t\tfile.CreatedAt = createdAtInt\n\t\t} else {\n\t\t\t// Fallback to current time if not available\n\t\t\tfile.CreatedAt = int(time.Now().Unix())\n\t\t}\n\n\t\tfiles = append(files, file)\n\t}\n\n\t// Calculate total pages\n\ttotalPages := int((total + int64(pageSize) - 1) / int64(pageSize))\n\n\treturn &ListResult{\n\t\tFiles: files,\n\t\tTotal: total,\n\t\tPage: page,\n\t\tPageSize: pageSize,\n\t\tTotalPages: totalPages,\n\t}, nil\n}\n\n// validate validates the file and option\nfunc (manager Manager) makeFile(file *FileHeader, option UploadOption) (*File, error) {\n\n\t// Validate max size\n\tif manager.maxsize > 0 && file.Size > manager.maxsize {\n\t\treturn nil, fmt.Errorf(\"file size %d exceeds the maximum size of %d\", file.Size, manager.maxsize)\n\t}\n\n\t// Use original filename if provided, otherwise use the file header filename\n\tfilename := file.Filename\n\tuserPath := option.OriginalFilename\n\tif userPath != \"\" {\n\t\t// If user provided a path, extract just the filename for the filename field\n\t\tfilename = filepath.Base(userPath)\n\t}\n\n\textension := filepath.Ext(filename)\n\n\t// Get the content type\n\t// For chunked uploads, file.Header may have incorrect content-type (e.g., application/octet-stream for Blob)\n\t// Try to detect from filename extension first, then fallback to header\n\tcontentType := file.Header.Get(\"Content-Type\")\n\tif extension != \"\" {\n\t\t// Try to get content type from extension\n\t\tdetectedType := mime.TypeByExtension(extension)\n\t\tif detectedType != \"\" {\n\t\t\t// If detected type is not the generic octet-stream, use it\n\t\t\t// This handles chunked uploads where the header has incorrect type\n\t\t\tif detectedType != \"application/octet-stream\" || contentType == \"application/octet-stream\" {\n\t\t\t\tcontentType = detectedType\n\t\t\t}\n\t\t}\n\t}\n\n\t// Get the extension from the content type if not available from filename\n\tif extension == \"\" {\n\t\t// Special handling for common types\n\t\tswitch contentType {\n\t\tcase \"text/plain\":\n\t\t\textension = \".txt\"\n\t\tcase \"image/jpeg\":\n\t\t\textension = \".jpg\"\n\t\tcase \"image/png\":\n\t\t\textension = \".png\"\n\t\tcase \"application/pdf\":\n\t\t\textension = \".pdf\"\n\t\tdefault:\n\t\t\textensions, err := mime.ExtensionsByType(contentType)\n\t\t\tif err == nil && len(extensions) > 0 {\n\t\t\t\t// For text/plain, prefer .txt over .conf\n\t\t\t\tif contentType == \"text/plain\" {\n\t\t\t\t\tfor _, ext := range extensions {\n\t\t\t\t\t\tif ext == \".txt\" {\n\t\t\t\t\t\t\textension = ext\n\t\t\t\t\t\t\tbreak\n\t\t\t\t\t\t}\n\t\t\t\t\t}\n\t\t\t\t\tif extension == \"\" {\n\t\t\t\t\t\textension = \".txt\"\n\t\t\t\t\t}\n\t\t\t\t} else {\n\t\t\t\t\textension = extensions[0]\n\t\t\t\t}\n\t\t\t}\n\t\t}\n\t}\n\n\t// Validate allowed types\n\tif !manager.allowed(contentType, extension) {\n\t\treturn nil, fmt.Errorf(\"%s type %s is not allowed\", filename, contentType)\n\t}\n\n\t// Generate file ID and storage path using the new approach\n\tid, storagePath, err := manager.generateFilePaths(file, extension, option)\n\tif err != nil {\n\t\treturn nil, err\n\t}\n\n\t// Set the path: use userPath if provided, otherwise use filename\n\tfilePath := userPath\n\tif filePath == \"\" {\n\t\tfilePath = filename\n\t}\n\n\treturn &File{\n\t\tID: id,\n\t\tUserPath: userPath, // Keep user's original input exactly as provided\n\t\tPath: storagePath, // Complete storage path: Groups + filename\n\t\tFilename: filename, // Use just the filename (extracted from path or header)\n\t\tContentType: contentType,\n\t\tBytes: int(file.Size),\n\t\tCreatedAt: int(time.Now().Unix()),\n\t\tStatus: \"uploading\",\n\t}, nil\n}\n\nfunc (manager Manager) allowed(contentType string, extension string) bool {\n\n\t// text/*, image/*, audio/*, video/*, application/yao-*, ...\n\tfor _, t := range manager.allowedTypes.wildcards {\n\t\tprefix := strings.TrimSuffix(t, \"*\")\n\t\tif strings.HasPrefix(contentType, prefix) {\n\t\t\treturn true\n\t\t}\n\t}\n\n\t// Accepted types\n\tif _, ok := manager.allowedTypes.mapping[contentType]; ok {\n\t\treturn true\n\t}\n\n\t// Accepted extensions\n\tif _, ok := manager.allowedTypes.mapping[extension]; ok {\n\t\treturn true\n\t}\n\n\t// Not allowed\n\treturn false\n}\n\n// generateFileID generates file ID and storage path based on Groups and filename\nfunc (manager Manager) generateFilePaths(file *FileHeader, extension string, option UploadOption) (fileID string, storagePath string, err error) {\n\n\t// 1. Get the filename\n\tvar filename string\n\tif file.Fingerprint() != \"\" {\n\t\tfilename = file.Fingerprint()\n\t} else if file.IsChunk() {\n\t\tfilename = file.UID()\n\t} else {\n\t\t// Generate unique filename to avoid conflicts\n\t\tvar originalName string\n\t\tif option.OriginalFilename != \"\" {\n\t\t\toriginalName = filepath.Base(option.OriginalFilename)\n\t\t} else {\n\t\t\toriginalName = file.Filename\n\t\t}\n\n\t\t// Extract extension from original filename\n\t\text := filepath.Ext(originalName)\n\t\tif ext == \"\" && extension != \"\" {\n\t\t\text = extension\n\t\t}\n\n\t\t// Generate unique filename: MD5 hash of original name + timestamp + extension\n\t\tnameHash := generateID(originalName + fmt.Sprintf(\"%d\", time.Now().UnixNano()))\n\t\tfilename = nameHash[:16] + ext // Use first 16 chars of hash + extension\n\t}\n\n\t// 2. Build complete storage path: Groups + filename\n\tpathParts := []string{}\n\n\t// Add groups to path\n\tif len(option.Groups) > 0 {\n\t\tpathParts = append(pathParts, option.Groups...)\n\t}\n\n\t// Add filename\n\tpathParts = append(pathParts, filename)\n\n\t// Join to create complete storage path\n\tstoragePath = strings.Join(pathParts, \"/\")\n\n\t// 3. Validate the storage path\n\tif !isValidPath(storagePath) {\n\t\treturn \"\", \"\", fmt.Errorf(\"invalid storage path: %s\", storagePath)\n\t}\n\n\t// 4. Generate ID as alias of the storage path (for security)\n\tfileID = generateID(storagePath)\n\n\t// 5. Add gzip extension to storage path if needed (not to fileID)\n\tif option.Gzip {\n\t\tstoragePath = storagePath + \".gz\"\n\t}\n\n\treturn fileID, storagePath, nil\n}\n\n// generateID generates a URL-safe ID based on the storage path\nfunc generateID(storagePath string) string {\n\thash := md5.Sum([]byte(storagePath))\n\treturn hex.EncodeToString(hash[:])\n}\n\n// isValidPath checks if a file path is valid\nfunc isValidPath(path string) bool {\n\tif path == \"\" {\n\t\treturn false\n\t}\n\n\t// Check for invalid characters that could cause issues\n\tinvalidChars := []string{\"../\", \"..\\\\\", \"\\\\\", \"//\"}\n\tfor _, invalid := range invalidChars {\n\t\tif strings.Contains(path, invalid) {\n\t\t\treturn false\n\t\t}\n\t}\n\n\treturn true\n}\n\n// getSize converts the size to bytes\nfunc getSize(size string) (int64, error) {\n\tif size == \"\" || size == \"0\" {\n\t\treturn 0, fmt.Errorf(\"size is empty\")\n\t}\n\n\tunit := strings.ToUpper(size[len(size)-1:])\n\tstr := size[:len(size)-1]\n\tif unit != \"B\" && unit != \"K\" && unit != \"M\" && unit != \"G\" {\n\t\tunit = \"B\"\n\t\tstr = size\n\t}\n\n\tvalue, err := strconv.ParseInt(str, 10, 64)\n\tif err != nil {\n\t\treturn 0, fmt.Errorf(\"invalid size: %s %s\", size, err)\n\t}\n\n\tswitch unit {\n\tcase \"B\":\n\t\treturn value, nil\n\tcase \"K\":\n\t\treturn value * 1024, nil\n\tcase \"M\":\n\t\treturn value * 1024 * 1024, nil\n\tcase \"G\":\n\t\treturn value * 1024 * 1024 * 1024, nil\n\t}\n\n\treturn 0, fmt.Errorf(\"invalid size: %s\", size)\n}\n\n// Exists checks if a file exists in storage\nfunc (manager Manager) Exists(ctx context.Context, fileID string) bool {\n\t// Check if file exists in database first\n\tstoragePath, err := manager.getStoragePathFromDatabase(ctx, fileID)\n\tif err != nil {\n\t\treturn false\n\t}\n\n\t// Then check if it exists in storage\n\treturn manager.storage.Exists(ctx, storagePath)\n}\n\n// Delete deletes a file from storage\nfunc (manager Manager) Delete(ctx context.Context, fileID string) error {\n\t// Get real storage path from database\n\tstoragePath, err := manager.getStoragePathFromDatabase(ctx, fileID)\n\tif err != nil {\n\t\treturn err\n\t}\n\n\t// Delete from storage\n\terr = manager.storage.Delete(ctx, storagePath)\n\tif err != nil {\n\t\treturn err\n\t}\n\n\t// Delete from database\n\tm := model.Select(\"__yao.attachment\")\n\t_, err = m.DeleteWhere(model.QueryParam{\n\t\tWheres: []model.QueryWhere{\n\t\t\t{Column: \"file_id\", Value: fileID},\n\t\t},\n\t})\n\n\tif err != nil {\n\t\treturn fmt.Errorf(\"failed to delete from database: %w\", err)\n\t}\n\n\treturn nil\n}\n\n// saveFileToDatabase saves file information to the database\n// For chunked uploads, it only updates bytes/status/progress if record exists\nfunc (manager Manager) saveFileToDatabase(ctx context.Context, file *File, storagePath string, option UploadOption) error {\n\n\tm := model.Select(\"__yao.attachment\")\n\n\t// Check if record exists first\n\trecords, err := m.Get(model.QueryParam{\n\t\tSelect: []interface{}{\"file_id\"},\n\t\tWheres: []model.QueryWhere{\n\t\t\t{Column: \"file_id\", Value: file.ID},\n\t\t},\n\t})\n\n\tif err != nil {\n\t\treturn fmt.Errorf(\"failed to check existing record: %w\", err)\n\t}\n\n\tif len(records) > 0 {\n\t\t// Record exists - this is a chunked upload update\n\t\t// Only update bytes, status, and progress (don't overwrite metadata)\n\t\tupdateData := map[string]interface{}{\n\t\t\t\"bytes\": int64(file.Bytes),\n\t\t\t\"status\": file.Status,\n\t\t}\n\n\t\t_, err = m.UpdateWhere(model.QueryParam{\n\t\t\tWheres: []model.QueryWhere{\n\t\t\t\t{Column: \"file_id\", Value: file.ID},\n\t\t\t},\n\t\t}, updateData)\n\n\t\treturn err\n\t}\n\n\t// Record doesn't exist - create new record with full metadata\n\t// Set default value for share if empty\n\tshare := option.Share\n\tif share == \"\" {\n\t\tshare = \"private\"\n\t}\n\n\t// Prepare data for database\n\tdata := map[string]interface{}{\n\t\t\"file_id\": file.ID,\n\t\t\"uploader\": manager.Name,\n\t\t\"content_type\": file.ContentType,\n\t\t\"name\": file.Filename,\n\t\t\"user_path\": option.OriginalFilename,\n\t\t\"path\": storagePath,\n\t\t\"bytes\": int64(file.Bytes),\n\t\t\"status\": file.Status,\n\t\t\"gzip\": option.Gzip,\n\t\t\"groups\": option.Groups,\n\t\t\"public\": option.Public,\n\t\t\"share\": share,\n\t}\n\n\t// Add Yao permission fields if provided\n\tif option.YaoCreatedBy != \"\" {\n\t\tdata[\"__yao_created_by\"] = option.YaoCreatedBy\n\t}\n\tif option.YaoUpdatedBy != \"\" {\n\t\tdata[\"__yao_updated_by\"] = option.YaoUpdatedBy\n\t}\n\tif option.YaoTeamID != \"\" {\n\t\tdata[\"__yao_team_id\"] = option.YaoTeamID\n\t}\n\tif option.YaoTenantID != \"\" {\n\t\tdata[\"__yao_tenant_id\"] = option.YaoTenantID\n\t}\n\n\t// Create new record\n\t_, err = m.Create(data)\n\treturn err\n}\n\n// getFileFromDatabase retrieves file information from database by file_id\nfunc (manager Manager) getFileFromDatabase(ctx context.Context, fileID string) (*File, error) {\n\tm := model.Select(\"__yao.attachment\")\n\n\trecords, err := m.Get(model.QueryParam{\n\t\tSelect: []interface{}{\n\t\t\t\"file_id\", \"name\", \"content_type\", \"status\", \"user_path\", \"path\", \"bytes\",\n\t\t\t\"public\", \"share\", \"__yao_created_by\", \"__yao_team_id\", \"__yao_tenant_id\",\n\t\t},\n\t\tWheres: []model.QueryWhere{\n\t\t\t{Column: \"file_id\", Value: fileID},\n\t\t},\n\t\tLimit: 1,\n\t})\n\n\tif err != nil {\n\t\treturn nil, fmt.Errorf(\"failed to query file: %w\", err)\n\t}\n\n\tif len(records) == 0 {\n\t\treturn nil, fmt.Errorf(\"file not found\")\n\t}\n\n\trecord := records[0]\n\n\t// Convert database record to File struct\n\tfile := &File{\n\t\tID: record[\"file_id\"].(string),\n\t\tFilename: record[\"name\"].(string),\n\t\tContentType: record[\"content_type\"].(string),\n\t\tStatus: record[\"status\"].(string),\n\t\tCreatedAt: int(time.Now().Unix()), // TODO: get from database\n\t}\n\n\t// Handle optional fields\n\tif userPath, ok := record[\"user_path\"].(string); ok {\n\t\tfile.UserPath = userPath\n\t}\n\n\tif path, ok := record[\"path\"].(string); ok {\n\t\tfile.Path = path\n\t}\n\n\tif bytes, ok := record[\"bytes\"].(int64); ok {\n\t\tfile.Bytes = int(bytes)\n\t}\n\n\t// Handle permission fields with safe conversion\n\tfile.Public = toBool(record[\"public\"])\n\tfile.Share = toString(record[\"share\"])\n\tfile.YaoCreatedBy = toString(record[\"__yao_created_by\"])\n\tfile.YaoTeamID = toString(record[\"__yao_team_id\"])\n\tfile.YaoTenantID = toString(record[\"__yao_tenant_id\"])\n\n\treturn file, nil\n}\n\n// getStoragePathFromDatabase retrieves the real storage path for a file_id\nfunc (manager Manager) getStoragePathFromDatabase(ctx context.Context, fileID string) (string, error) {\n\tm := model.Select(\"__yao.attachment\")\n\n\trecords, err := m.Get(model.QueryParam{\n\t\tSelect: []interface{}{\"path\"},\n\t\tWheres: []model.QueryWhere{\n\t\t\t{Column: \"file_id\", Value: fileID},\n\t\t},\n\t})\n\n\tif err != nil {\n\t\treturn \"\", fmt.Errorf(\"failed to query database: %w\", err)\n\t}\n\n\tif len(records) == 0 {\n\t\treturn \"\", fmt.Errorf(\"file not found: %s\", fileID)\n\t}\n\n\tif path, ok := records[0][\"path\"].(string); ok && path != \"\" {\n\t\treturn path, nil\n\t}\n\n\treturn \"\", fmt.Errorf(\"invalid storage path for file ID: %s\", fileID)\n}\n\n// GetText retrieves the parsed text content for a file by its ID\n// By default, returns the preview (first 2000 characters) from 'content_preview' field\n// Set fullContent to true to retrieve the complete text from 'content' field\nfunc (manager Manager) GetText(ctx context.Context, fileID string, fullContent ...bool) (string, error) {\n\tm := model.Select(\"__yao.attachment\")\n\n\t// Determine which field to query\n\twantFullContent := false\n\tif len(fullContent) > 0 {\n\t\twantFullContent = fullContent[0]\n\t}\n\n\tfieldName := \"content_preview\"\n\tif wantFullContent {\n\t\tfieldName = \"content\"\n\t}\n\n\trecords, err := m.Get(model.QueryParam{\n\t\tSelect: []interface{}{fieldName},\n\t\tWheres: []model.QueryWhere{\n\t\t\t{Column: \"file_id\", Value: fileID},\n\t\t},\n\t\tLimit: 1,\n\t})\n\n\tif err != nil {\n\t\treturn \"\", fmt.Errorf(\"failed to query text content: %w\", err)\n\t}\n\n\tif len(records) == 0 {\n\t\treturn \"\", fmt.Errorf(\"file not found: %s\", fileID)\n\t}\n\n\t// Handle content field - it may be nil, string, or other types\n\tif content, ok := records[0][fieldName].(string); ok {\n\t\treturn content, nil\n\t}\n\n\t// If content is nil or not a string, return empty string\n\treturn \"\", nil\n}\n\n// SaveText saves the parsed text content for a file by its ID\n// Automatically saves both full content and preview (first 2000 characters)\n// Updates both 'content' and 'content_preview' fields in the attachment record\nfunc (manager Manager) SaveText(ctx context.Context, fileID string, text string) error {\n\tm := model.Select(\"__yao.attachment\")\n\n\t// Check if record exists first\n\trecords, err := m.Get(model.QueryParam{\n\t\tSelect: []interface{}{\"file_id\"},\n\t\tWheres: []model.QueryWhere{\n\t\t\t{Column: \"file_id\", Value: fileID},\n\t\t},\n\t\tLimit: 1,\n\t})\n\n\tif err != nil {\n\t\treturn fmt.Errorf(\"failed to check file existence: %w\", err)\n\t}\n\n\tif len(records) == 0 {\n\t\treturn fmt.Errorf(\"file not found: %s\", fileID)\n\t}\n\n\t// Create preview: first 2000 characters (or runes for proper UTF-8 handling)\n\tpreview := text\n\tconst maxPreviewLength = 2000\n\tif len([]rune(text)) > maxPreviewLength {\n\t\tpreview = string([]rune(text)[:maxPreviewLength])\n\t}\n\n\t// Update both content and content_preview fields\n\tupdateData := map[string]interface{}{\n\t\t\"content\": text,\n\t\t\"content_preview\": preview,\n\t}\n\n\t_, err = m.UpdateWhere(model.QueryParam{\n\t\tWheres: []model.QueryWhere{\n\t\t\t{Column: \"file_id\", Value: fileID},\n\t\t},\n\t}, updateData)\n\n\tif err != nil {\n\t\treturn fmt.Errorf(\"failed to save text content: %w\", err)\n\t}\n\n\treturn nil\n}\n" }, { "path": "attachment/manager_test.go", "content": "package attachment\n\nimport (\n\t\"bytes\"\n\t\"context\"\n\t\"encoding/base64\"\n\t\"fmt\"\n\t\"mime/multipart\"\n\t\"os\"\n\t\"path/filepath\"\n\t\"strings\"\n\t\"testing\"\n\t\"time\"\n\n\t\"github.com/yaoapp/gou/model\"\n\t\"github.com/yaoapp/yao/config\"\n\t\"github.com/yaoapp/yao/test\"\n)\n\nfunc TestMain(m *testing.M) {\n\t// Run tests\n\tcode := m.Run()\n\tos.Exit(code)\n}\n\nfunc TestManagerUpload(t *testing.T) {\n\ttest.Prepare(t, config.Conf)\n\tdefer test.Clean()\n\n\t// Create a local storage manager\n\tmanager, err := New(ManagerOption{\n\t\tDriver: \"local\",\n\t\tMaxSize: \"10M\",\n\t\tChunkSize: \"2M\",\n\t\tAllowedTypes: []string{\"text/*\", \"image/*\", \".txt\", \".jpg\", \".png\"},\n\t\tOptions: map[string]interface{}{\n\t\t\t\"path\": \"/tmp/test_attachments\",\n\t\t},\n\t})\n\tif err != nil {\n\t\tt.Fatalf(\"Failed to create manager: %v\", err)\n\t}\n\n\t// Test simple text file upload\n\tt.Run(\"SimpleTextUpload\", func(t *testing.T) {\n\t\tcontent := \"Hello, World!\"\n\t\treader := strings.NewReader(content)\n\n\t\t// Create a mock file header\n\t\tfileHeader := &FileHeader{\n\t\t\tFileHeader: &multipart.FileHeader{\n\t\t\t\tFilename: \"test.txt\",\n\t\t\t\tSize: int64(len(content)),\n\t\t\t\tHeader: make(map[string][]string),\n\t\t\t},\n\t\t}\n\t\tfileHeader.Header.Set(\"Content-Type\", \"text/plain\")\n\n\t\toption := UploadOption{Groups: []string{\"user123\"}}\n\n\t\tfile, err := manager.Upload(context.Background(), fileHeader, reader, option)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to upload file: %v\", err)\n\t\t}\n\n\t\tif file.Filename != \"test.txt\" {\n\t\t\tt.Errorf(\"Expected filename 'test.txt', got '%s'\", file.Filename)\n\t\t}\n\n\t\t// Content type may include charset\n\t\tif !strings.HasPrefix(file.ContentType, \"text/plain\") {\n\t\t\tt.Errorf(\"Expected content type 'text/plain', got '%s'\", file.ContentType)\n\t\t}\n\n\t\t// Test download\n\t\tresponse, err := manager.Download(context.Background(), file.ID)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to download file: %v\", err)\n\t\t}\n\t\tdefer response.Reader.Close()\n\n\t\tdownloadedContent, err := manager.Read(context.Background(), file.ID)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to read file: %v\", err)\n\t\t}\n\n\t\tif string(downloadedContent) != content {\n\t\t\tt.Errorf(\"Expected content '%s', got '%s'\", content, string(downloadedContent))\n\t\t}\n\n\t\t// Test ReadBase64\n\t\tbase64Content, err := manager.ReadBase64(context.Background(), file.ID)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to read file as base64: %v\", err)\n\t\t}\n\n\t\texpectedBase64 := base64.StdEncoding.EncodeToString([]byte(content))\n\t\tif base64Content != expectedBase64 {\n\t\t\tt.Errorf(\"Expected base64 '%s', got '%s'\", expectedBase64, base64Content)\n\t\t}\n\t})\n\n\t// Test gzip compression\n\tt.Run(\"GzipUpload\", func(t *testing.T) {\n\t\tcontent := \"This is a test file that will be compressed with gzip.\"\n\t\treader := strings.NewReader(content)\n\n\t\tfileHeader := &FileHeader{\n\t\t\tFileHeader: &multipart.FileHeader{\n\t\t\t\tFilename: \"test_gzip.txt\",\n\t\t\t\tSize: int64(len(content)),\n\t\t\t\tHeader: make(map[string][]string),\n\t\t\t},\n\t\t}\n\t\tfileHeader.Header.Set(\"Content-Type\", \"text/plain\")\n\n\t\toption := UploadOption{\n\t\t\tGzip: true,\n\t\t\tGroups: []string{\"user123\"},\n\t\t}\n\n\t\tfile, err := manager.Upload(context.Background(), fileHeader, reader, option)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to upload gzipped file: %v\", err)\n\t\t}\n\n\t\t// The stored file should be compressed, but when we read it back,\n\t\t// we should get the original content (if the storage handles decompression)\n\t\tdownloadedContent, err := manager.Read(context.Background(), file.ID)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to read gzipped file: %v\", err)\n\t\t}\n\n\t\tif string(downloadedContent) != content {\n\t\t\tt.Errorf(\"Expected content '%s', got '%s'\", content, string(downloadedContent))\n\t\t}\n\t})\n\n\t// Test chunked upload\n\tt.Run(\"ChunkedUpload\", func(t *testing.T) {\n\t\tcontent := \"This is a large file that will be uploaded in chunks. \" +\n\t\t\tstrings.Repeat(\"Lorem ipsum dolor sit amet, consectetur adipiscing elit. \", 100)\n\n\t\tchunkSize := 1024\n\t\ttotalSize := len(content)\n\n\t\tvar lastFile *File\n\t\tfor start := 0; start < totalSize; start += chunkSize {\n\t\t\tend := start + chunkSize - 1\n\t\t\tif end >= totalSize {\n\t\t\t\tend = totalSize - 1\n\t\t\t}\n\n\t\t\tchunk := []byte(content[start : end+1])\n\n\t\t\tfileHeader := &FileHeader{\n\t\t\t\tFileHeader: &multipart.FileHeader{\n\t\t\t\t\tFilename: \"large_file.txt\",\n\t\t\t\t\tSize: int64(len(chunk)),\n\t\t\t\t\tHeader: make(map[string][]string),\n\t\t\t\t},\n\t\t\t}\n\t\t\tfileHeader.Header.Set(\"Content-Type\", \"text/plain\")\n\t\t\tfileHeader.Header.Set(\"Content-Range\",\n\t\t\t\tfmt.Sprintf(\"bytes %d-%d/%d\", start, end, totalSize))\n\t\t\tfileHeader.Header.Set(\"Content-Uid\", \"unique-file-id-123\")\n\n\t\t\toption := UploadOption{Groups: []string{\"user123\"}}\n\t\t\tfile, err := manager.Upload(context.Background(), fileHeader, bytes.NewReader(chunk), option)\n\t\t\tif err != nil {\n\t\t\t\tt.Fatalf(\"Failed to upload chunk starting at %d: %v\", start, err)\n\t\t\t}\n\n\t\t\tlastFile = file\n\t\t}\n\n\t\t// After uploading all chunks, read the complete file\n\t\tif lastFile != nil {\n\t\t\tdownloadedContent, err := manager.Read(context.Background(), lastFile.ID)\n\t\t\tif err != nil {\n\t\t\t\tt.Fatalf(\"Failed to read chunked file: %v\", err)\n\t\t\t}\n\n\t\t\tif string(downloadedContent) != content {\n\t\t\t\tt.Errorf(\"Chunked upload content mismatch. Expected length %d, got %d\",\n\t\t\t\t\tlen(content), len(downloadedContent))\n\t\t\t}\n\t\t}\n\t})\n}\n\nfunc TestManagerMultiLevelGroups(t *testing.T) {\n\ttest.Prepare(t, config.Conf)\n\tdefer test.Clean()\n\n\t// Create a local storage manager\n\tmanager, err := New(ManagerOption{\n\t\tDriver: \"local\",\n\t\tMaxSize: \"10M\",\n\t\tAllowedTypes: []string{\"text/*\", \"image/*\"},\n\t\tOptions: map[string]interface{}{\n\t\t\t\"path\": \"/tmp/test_attachments\",\n\t\t},\n\t})\n\tif err != nil {\n\t\tt.Fatalf(\"Failed to create manager: %v\", err)\n\t}\n\n\t// Test multi-level groups\n\tt.Run(\"MultiLevelGroups\", func(t *testing.T) {\n\t\tcontent := \"Test content for multi-level groups\"\n\t\treader := strings.NewReader(content)\n\n\t\tfileHeader := &FileHeader{\n\t\t\tFileHeader: &multipart.FileHeader{\n\t\t\t\tFilename: \"multilevel.txt\",\n\t\t\t\tSize: int64(len(content)),\n\t\t\t\tHeader: make(map[string][]string),\n\t\t\t},\n\t\t}\n\t\tfileHeader.Header.Set(\"Content-Type\", \"text/plain\")\n\n\t\t// Test with multi-level groups\n\t\toption := UploadOption{\n\t\t\tGroups: []string{\"users\", \"user123\", \"chats\", \"chat456\", \"documents\"},\n\t\t}\n\n\t\tfile, err := manager.Upload(context.Background(), fileHeader, reader, option)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to upload file with multi-level groups: %v\", err)\n\t\t}\n\n\t\t// File ID should be 32 character hex (MD5 hash)\n\t\tif len(file.ID) != 32 {\n\t\t\tt.Errorf(\"File ID should be 32 characters: %s (length %d)\", file.ID, len(file.ID))\n\t\t}\n\n\t\t// Check that it's all lowercase hex\n\t\tfor _, r := range file.ID {\n\t\t\tif !((r >= '0' && r <= '9') || (r >= 'a' && r <= 'f')) {\n\t\t\t\tt.Errorf(\"File ID contains non-hex character: %c\", r)\n\t\t\t}\n\t\t}\n\n\t\t// Test download\n\t\tdownloadedContent, err := manager.Read(context.Background(), file.ID)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to read file with multi-level groups: %v\", err)\n\t\t}\n\n\t\tif string(downloadedContent) != content {\n\t\t\tt.Errorf(\"Content mismatch for multi-level groups file\")\n\t\t}\n\t})\n\n\t// Test single group (backward compatibility)\n\tt.Run(\"SingleGroup\", func(t *testing.T) {\n\t\tcontent := \"Test content for single group\"\n\t\treader := strings.NewReader(content)\n\n\t\tfileHeader := &FileHeader{\n\t\t\tFileHeader: &multipart.FileHeader{\n\t\t\t\tFilename: \"single.txt\",\n\t\t\t\tSize: int64(len(content)),\n\t\t\t\tHeader: make(map[string][]string),\n\t\t\t},\n\t\t}\n\t\tfileHeader.Header.Set(\"Content-Type\", \"text/plain\")\n\n\t\toption := UploadOption{\n\t\t\tGroups: []string{\"knowledge\"},\n\t\t}\n\n\t\tfile, err := manager.Upload(context.Background(), fileHeader, reader, option)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to upload file with single group: %v\", err)\n\t\t}\n\n\t\t// File ID should be 32 character hex (MD5 hash)\n\t\tif len(file.ID) != 32 {\n\t\t\tt.Errorf(\"File ID should be 32 characters: %s (length %d)\", file.ID, len(file.ID))\n\t\t}\n\n\t\t// Check that it's all lowercase hex\n\t\tfor _, r := range file.ID {\n\t\t\tif !((r >= '0' && r <= '9') || (r >= 'a' && r <= 'f')) {\n\t\t\t\tt.Errorf(\"File ID contains non-hex character: %c\", r)\n\t\t\t}\n\t\t}\n\t})\n\n\t// Test empty groups (no grouping)\n\tt.Run(\"EmptyGroups\", func(t *testing.T) {\n\t\tcontent := \"Test content without groups\"\n\t\treader := strings.NewReader(content)\n\n\t\tfileHeader := &FileHeader{\n\t\t\tFileHeader: &multipart.FileHeader{\n\t\t\t\tFilename: \"nogroup.txt\",\n\t\t\t\tSize: int64(len(content)),\n\t\t\t\tHeader: make(map[string][]string),\n\t\t\t},\n\t\t}\n\t\tfileHeader.Header.Set(\"Content-Type\", \"text/plain\")\n\n\t\toption := UploadOption{\n\t\t\tGroups: []string{}, // Empty groups\n\t\t}\n\n\t\tfile, err := manager.Upload(context.Background(), fileHeader, reader, option)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to upload file without groups: %v\", err)\n\t\t}\n\n\t\t// Should still work and create valid file ID\n\t\tif file.ID == \"\" {\n\t\t\tt.Error(\"File ID should not be empty\")\n\t\t}\n\t})\n}\n\nfunc TestManagerValidation(t *testing.T) {\n\ttest.Prepare(t, config.Conf)\n\tdefer test.Clean()\n\n\tmanager, err := New(ManagerOption{\n\t\tDriver: \"local\",\n\t\tMaxSize: \"1K\", // Very small max size for testing\n\t\tAllowedTypes: []string{\"text/plain\"},\n\t\tOptions: map[string]interface{}{\n\t\t\t\"path\": \"/tmp/test_attachments\",\n\t\t},\n\t})\n\tif err != nil {\n\t\tt.Fatalf(\"Failed to create manager: %v\", err)\n\t}\n\n\t// Test file size validation\n\tt.Run(\"FileSizeValidation\", func(t *testing.T) {\n\t\tcontent := strings.Repeat(\"a\", 2048) // 2KB, exceeds 1KB limit\n\t\treader := strings.NewReader(content)\n\n\t\tfileHeader := &FileHeader{\n\t\t\tFileHeader: &multipart.FileHeader{\n\t\t\t\tFilename: \"large.txt\",\n\t\t\t\tSize: int64(len(content)),\n\t\t\t\tHeader: make(map[string][]string),\n\t\t\t},\n\t\t}\n\t\tfileHeader.Header.Set(\"Content-Type\", \"text/plain\")\n\n\t\toption := UploadOption{}\n\n\t\t_, err := manager.Upload(context.Background(), fileHeader, reader, option)\n\t\tif err == nil {\n\t\t\tt.Error(\"Expected error for file size exceeding limit\")\n\t\t}\n\t})\n\n\t// Test file type validation\n\tt.Run(\"FileTypeValidation\", func(t *testing.T) {\n\t\tcontent := \"test\"\n\t\treader := strings.NewReader(content)\n\n\t\tfileHeader := &FileHeader{\n\t\t\tFileHeader: &multipart.FileHeader{\n\t\t\t\tFilename: \"test.jpg\",\n\t\t\t\tSize: int64(len(content)),\n\t\t\t\tHeader: make(map[string][]string),\n\t\t\t},\n\t\t}\n\t\tfileHeader.Header.Set(\"Content-Type\", \"image/jpeg\") // Not allowed\n\n\t\toption := UploadOption{}\n\n\t\t_, err := manager.Upload(context.Background(), fileHeader, reader, option)\n\t\tif err == nil {\n\t\t\tt.Error(\"Expected error for disallowed file type\")\n\t\t}\n\t})\n}\n\nfunc TestManagerName(t *testing.T) {\n\ttest.Prepare(t, config.Conf)\n\tdefer test.Clean()\n\n\t// Create a manager with a specific name\n\tmanagerName := \"test-manager\"\n\tmanager, err := RegisterDefault(managerName)\n\tif err != nil {\n\t\tt.Fatalf(\"Failed to register manager: %v\", err)\n\t}\n\n\t// Verify the manager name is set correctly\n\tif manager.Name != managerName {\n\t\tt.Errorf(\"Expected manager name '%s', got '%s'\", managerName, manager.Name)\n\t}\n\n\t// Upload a file to verify the manager name is saved to database\n\tcontent := \"Test file content\"\n\treader := strings.NewReader(content)\n\n\tfileHeader := &FileHeader{\n\t\tFileHeader: &multipart.FileHeader{\n\t\t\tFilename: \"test-manager-name.txt\",\n\t\t\tSize: int64(len(content)),\n\t\t\tHeader: make(map[string][]string),\n\t\t},\n\t}\n\tfileHeader.Header.Set(\"Content-Type\", \"text/plain\")\n\n\toption := UploadOption{\n\t\tGroups: []string{\"test\"},\n\t}\n\n\tfile, err := manager.Upload(context.Background(), fileHeader, reader, option)\n\tif err != nil {\n\t\tt.Fatalf(\"Failed to upload file: %v\", err)\n\t}\n\n\t// Query database directly to verify manager name is stored\n\tm := model.Select(\"__yao.attachment\")\n\trecords, err := m.Get(model.QueryParam{\n\t\tSelect: []interface{}{\"uploader\"},\n\t\tWheres: []model.QueryWhere{\n\t\t\t{Column: \"file_id\", Value: file.ID},\n\t\t},\n\t})\n\n\tif err != nil {\n\t\tt.Fatalf(\"Failed to query database: %v\", err)\n\t}\n\n\tif len(records) == 0 {\n\t\tt.Fatal(\"No record found in database\")\n\t}\n\n\tstoredManagerName, ok := records[0][\"uploader\"].(string)\n\tif !ok {\n\t\tt.Fatal(\"Uploader field is not a string\")\n\t}\n\n\tif storedManagerName != managerName {\n\t\tt.Errorf(\"Expected stored uploader name '%s', got '%s'\", managerName, storedManagerName)\n\t}\n}\n\nfunc TestUniqueFilenameGeneration(t *testing.T) {\n\ttest.Prepare(t, config.Conf)\n\tdefer test.Clean()\n\n\tmanager, err := RegisterDefault(\"test\")\n\tif err != nil {\n\t\tt.Fatalf(\"Failed to register manager: %v\", err)\n\t}\n\n\t// Upload two files with the same filename\n\tcontent1 := \"First file content\"\n\tcontent2 := \"Second file content\"\n\n\t// First file\n\treader1 := strings.NewReader(content1)\n\tfileHeader1 := &FileHeader{\n\t\tFileHeader: &multipart.FileHeader{\n\t\t\tFilename: \"duplicate.txt\", // Same filename\n\t\t\tSize: int64(len(content1)),\n\t\t\tHeader: make(map[string][]string),\n\t\t},\n\t}\n\tfileHeader1.Header.Set(\"Content-Type\", \"text/plain\")\n\n\t// Second file\n\treader2 := strings.NewReader(content2)\n\tfileHeader2 := &FileHeader{\n\t\tFileHeader: &multipart.FileHeader{\n\t\t\tFilename: \"duplicate.txt\", // Same filename\n\t\t\tSize: int64(len(content2)),\n\t\t\tHeader: make(map[string][]string),\n\t\t},\n\t}\n\tfileHeader2.Header.Set(\"Content-Type\", \"text/plain\")\n\n\toption := UploadOption{\n\t\tGroups: []string{\"test\"},\n\t}\n\n\t// Upload first file\n\tfile1, err := manager.Upload(context.Background(), fileHeader1, reader1, option)\n\tif err != nil {\n\t\tt.Fatalf(\"Failed to upload first file: %v\", err)\n\t}\n\n\t// Sleep a bit to ensure different timestamps\n\ttime.Sleep(time.Millisecond)\n\n\t// Upload second file\n\tfile2, err := manager.Upload(context.Background(), fileHeader2, reader2, option)\n\tif err != nil {\n\t\tt.Fatalf(\"Failed to upload second file: %v\", err)\n\t}\n\n\t// Verify files have different IDs\n\tif file1.ID == file2.ID {\n\t\tt.Error(\"Files with same original name should have different IDs\")\n\t}\n\n\t// Verify files have different storage paths\n\tif file1.Path == file2.Path {\n\t\tt.Error(\"Files with same original name should have different storage paths\")\n\t}\n\n\t// Verify both files can be read independently\n\tdata1, err := manager.Read(context.Background(), file1.ID)\n\tif err != nil {\n\t\tt.Fatalf(\"Failed to read first file: %v\", err)\n\t}\n\n\tdata2, err := manager.Read(context.Background(), file2.ID)\n\tif err != nil {\n\t\tt.Fatalf(\"Failed to read second file: %v\", err)\n\t}\n\n\tif string(data1) != content1 {\n\t\tt.Errorf(\"First file content mismatch. Expected: %s, Got: %s\", content1, string(data1))\n\t}\n\n\tif string(data2) != content2 {\n\t\tt.Errorf(\"Second file content mismatch. Expected: %s, Got: %s\", content2, string(data2))\n\t}\n\n\tt.Logf(\"File 1 - ID: %s, Path: %s\", file1.ID, file1.Path)\n\tt.Logf(\"File 2 - ID: %s, Path: %s\", file2.ID, file2.Path)\n}\n\nfunc TestInfo(t *testing.T) {\n\ttest.Prepare(t, config.Conf)\n\tdefer test.Clean()\n\n\tmanager, err := RegisterDefault(\"test\")\n\tif err != nil {\n\t\tt.Fatalf(\"Failed to register manager: %v\", err)\n\t}\n\n\t// Upload a test file\n\tcontent := \"Test file for info retrieval\"\n\treader := strings.NewReader(content)\n\n\tfileHeader := &FileHeader{\n\t\tFileHeader: &multipart.FileHeader{\n\t\t\tFilename: \"info-test.txt\",\n\t\t\tSize: int64(len(content)),\n\t\t\tHeader: make(map[string][]string),\n\t\t},\n\t}\n\tfileHeader.Header.Set(\"Content-Type\", \"text/plain\")\n\n\toption := UploadOption{\n\t\tGroups: []string{\"info\", \"test\"},\n\t\tOriginalFilename: \"original-info-test.txt\",\n\t\tPublic: false,\n\t\tShare: \"private\",\n\t\tGzip: false,\n\t}\n\n\tuploadedFile, err := manager.Upload(context.Background(), fileHeader, reader, option)\n\tif err != nil {\n\t\tt.Fatalf(\"Failed to upload file: %v\", err)\n\t}\n\n\t// Test the Info method\n\tfileInfo, err := manager.Info(context.Background(), uploadedFile.ID)\n\tif err != nil {\n\t\tt.Fatalf(\"Failed to get file info: %v\", err)\n\t}\n\n\t// Verify file information\n\tif fileInfo.ID != uploadedFile.ID {\n\t\tt.Errorf(\"Expected file ID %s, got %s\", uploadedFile.ID, fileInfo.ID)\n\t}\n\n\tif fileInfo.Filename != uploadedFile.Filename {\n\t\tt.Errorf(\"Expected filename %s, got %s\", uploadedFile.Filename, fileInfo.Filename)\n\t}\n\n\t// Content type may include charset\n\tif !strings.HasPrefix(fileInfo.ContentType, \"text/plain\") {\n\t\tt.Errorf(\"Expected content type 'text/plain', got %s\", fileInfo.ContentType)\n\t}\n\n\tif fileInfo.Status != \"uploaded\" {\n\t\tt.Errorf(\"Expected status 'uploaded', got %s\", fileInfo.Status)\n\t}\n\n\tif fileInfo.UserPath != option.OriginalFilename {\n\t\tt.Errorf(\"Expected user path %s, got %s\", option.OriginalFilename, fileInfo.UserPath)\n\t}\n\n\tif fileInfo.Path != uploadedFile.Path {\n\t\tt.Errorf(\"Expected path %s, got %s\", uploadedFile.Path, fileInfo.Path)\n\t}\n\n\t// Test with non-existent file ID\n\t_, err = manager.Info(context.Background(), \"non-existent-id\")\n\tif err == nil {\n\t\tt.Error(\"Expected error for non-existent file ID, got nil\")\n\t}\n\n\tt.Logf(\"Retrieved file info - ID: %s, Path: %s, UserPath: %s\",\n\t\tfileInfo.ID, fileInfo.Path, fileInfo.UserPath)\n}\n\nfunc TestList(t *testing.T) {\n\ttest.Prepare(t, config.Conf)\n\tdefer test.Clean()\n\n\t// Use unique manager name for test isolation\n\tmanagerName := fmt.Sprintf(\"test-list-%d\", time.Now().UnixNano())\n\tmanager, err := RegisterDefault(managerName)\n\tif err != nil {\n\t\tt.Fatalf(\"Failed to register manager: %v\", err)\n\t}\n\n\t// Clean up existing records first\n\tm := model.Select(\"__yao.attachment\")\n\t_, err = m.DeleteWhere(model.QueryParam{\n\t\tWheres: []model.QueryWhere{\n\t\t\t{Column: \"uploader\", Value: managerName},\n\t\t},\n\t})\n\tif err != nil {\n\t\tt.Logf(\"Warning: Failed to clean up existing records: %v\", err)\n\t}\n\n\t// Upload multiple test files\n\ttestFiles := []struct {\n\t\tfilename string\n\t\tcontent string\n\t\tcontentType string\n\t\tgroups []string\n\t}{\n\t\t{\"test1.txt\", \"Content of test file 1\", \"text/plain\", []string{\"group1\"}},\n\t\t{\"test2.txt\", \"Content of test file 2\", \"text/plain\", []string{\"group1\"}},\n\t\t{\"image1.jpg\", \"Image content 1\", \"image/jpeg\", []string{\"group2\", \"images\"}},\n\t\t{\"doc1.pdf\", \"PDF content\", \"application/pdf\", []string{\"group2\", \"docs\"}},\n\t\t{\"test3.txt\", \"Content of test file 3\", \"text/plain\", []string{\"group1\"}},\n\t}\n\n\tuploadedFiles := make([]*File, 0, len(testFiles))\n\tfor _, tf := range testFiles {\n\t\treader := strings.NewReader(tf.content)\n\t\tfileHeader := &FileHeader{\n\t\t\tFileHeader: &multipart.FileHeader{\n\t\t\t\tFilename: tf.filename,\n\t\t\t\tSize: int64(len(tf.content)),\n\t\t\t\tHeader: make(map[string][]string),\n\t\t\t},\n\t\t}\n\t\tfileHeader.Header.Set(\"Content-Type\", tf.contentType)\n\n\t\toption := UploadOption{\n\t\t\tGroups: tf.groups,\n\t\t}\n\n\t\tfile, err := manager.Upload(context.Background(), fileHeader, reader, option)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to upload file %s: %v\", tf.filename, err)\n\t\t}\n\t\tuploadedFiles = append(uploadedFiles, file)\n\t}\n\n\t// Test basic listing (no filters, default pagination)\n\tt.Run(\"BasicList\", func(t *testing.T) {\n\t\tresult, err := manager.List(context.Background(), ListOption{\n\t\t\tFilters: map[string]interface{}{\n\t\t\t\t\"uploader\": managerName,\n\t\t\t},\n\t\t})\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to list files: %v\", err)\n\t\t}\n\n\t\tif len(result.Files) != len(testFiles) {\n\t\t\tt.Errorf(\"Expected %d files, got %d\", len(testFiles), len(result.Files))\n\t\t}\n\n\t\tif result.Total != int64(len(testFiles)) {\n\t\t\tt.Errorf(\"Expected total %d, got %d\", len(testFiles), result.Total)\n\t\t}\n\n\t\tif result.Page != 1 {\n\t\t\tt.Errorf(\"Expected page 1, got %d\", result.Page)\n\t\t}\n\n\t\tif result.PageSize != 20 {\n\t\t\tt.Errorf(\"Expected page size 20, got %d\", result.PageSize)\n\t\t}\n\t})\n\n\t// Test pagination\n\tt.Run(\"Pagination\", func(t *testing.T) {\n\t\tresult, err := manager.List(context.Background(), ListOption{\n\t\t\tPage: 1,\n\t\t\tPageSize: 2,\n\t\t\tFilters: map[string]interface{}{\n\t\t\t\t\"uploader\": managerName,\n\t\t\t},\n\t\t})\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to list files with pagination: %v\", err)\n\t\t}\n\n\t\tif len(result.Files) != 2 {\n\t\t\tt.Errorf(\"Expected 2 files, got %d\", len(result.Files))\n\t\t}\n\n\t\tif result.Total != int64(len(testFiles)) {\n\t\t\tt.Errorf(\"Expected total %d, got %d\", len(testFiles), result.Total)\n\t\t}\n\n\t\tif result.Page != 1 {\n\t\t\tt.Errorf(\"Expected page 1, got %d\", result.Page)\n\t\t}\n\n\t\tif result.PageSize != 2 {\n\t\t\tt.Errorf(\"Expected page size 2, got %d\", result.PageSize)\n\t\t}\n\n\t\tif result.TotalPages != 3 { // 5 files / 2 per page = 3 pages\n\t\t\tt.Errorf(\"Expected 3 total pages, got %d\", result.TotalPages)\n\t\t}\n\t})\n\n\t// Test filtering by content type\n\tt.Run(\"FilterByContentType\", func(t *testing.T) {\n\t\tresult, err := manager.List(context.Background(), ListOption{\n\t\t\tWheres: []model.QueryWhere{\n\t\t\t\t{Column: \"uploader\", Value: managerName},\n\t\t\t\t{Column: \"content_type\", Value: \"text/plain%\", OP: \"like\"},\n\t\t\t},\n\t\t})\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to list files with content type filter: %v\", err)\n\t\t}\n\n\t\texpectedCount := 3 // test1.txt, test2.txt, test3.txt\n\t\tif len(result.Files) != expectedCount {\n\t\t\tt.Errorf(\"Expected %d text files, got %d\", expectedCount, len(result.Files))\n\t\t}\n\n\t\t// Verify all returned files are text/plain (may include charset)\n\t\tfor _, file := range result.Files {\n\t\t\tif !strings.HasPrefix(file.ContentType, \"text/plain\") {\n\t\t\t\tt.Errorf(\"Expected content type 'text/plain', got '%s'\", file.ContentType)\n\t\t\t}\n\t\t}\n\t})\n\n\t// Test wildcard filtering\n\tt.Run(\"WildcardFilter\", func(t *testing.T) {\n\t\tresult, err := manager.List(context.Background(), ListOption{\n\t\t\tFilters: map[string]interface{}{\n\t\t\t\t\"uploader\": managerName,\n\t\t\t\t\"content_type\": \"image/*\",\n\t\t\t},\n\t\t})\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to list files with wildcard filter: %v\", err)\n\t\t}\n\n\t\texpectedCount := 1 // image1.jpg\n\t\tif len(result.Files) != expectedCount {\n\t\t\tt.Errorf(\"Expected %d image files, got %d\", expectedCount, len(result.Files))\n\t\t}\n\t})\n\n\t// Test ordering\n\tt.Run(\"OrderBy\", func(t *testing.T) {\n\t\tresult, err := manager.List(context.Background(), ListOption{\n\t\t\tOrderBy: \"name asc\",\n\t\t\tFilters: map[string]interface{}{\n\t\t\t\t\"uploader\": managerName,\n\t\t\t},\n\t\t})\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to list files with ordering: %v\", err)\n\t\t}\n\n\t\tif len(result.Files) != len(testFiles) {\n\t\t\tt.Errorf(\"Expected %d files, got %d\", len(testFiles), len(result.Files))\n\t\t}\n\n\t\t// Files should be ordered by name ascending\n\t\t// Note: The actual filenames are generated, so we just check that they're sorted\n\t\tfor i := 1; i < len(result.Files); i++ {\n\t\t\tif result.Files[i-1].Filename > result.Files[i].Filename {\n\t\t\t\tt.Errorf(\"Files are not sorted by name ascending\")\n\t\t\t\tbreak\n\t\t\t}\n\t\t}\n\t})\n\n\t// Test field selection\n\tt.Run(\"SelectFields\", func(t *testing.T) {\n\t\tresult, err := manager.List(context.Background(), ListOption{\n\t\t\tSelect: []string{\"file_id\", \"name\", \"content_type\"},\n\t\t\tFilters: map[string]interface{}{\n\t\t\t\t\"uploader\": managerName,\n\t\t\t},\n\t\t})\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to list files with field selection: %v\", err)\n\t\t}\n\n\t\tif len(result.Files) != len(testFiles) {\n\t\t\tt.Errorf(\"Expected %d files, got %d\", len(testFiles), len(result.Files))\n\t\t}\n\n\t\t// Verify selected fields are populated\n\t\tfor _, file := range result.Files {\n\t\t\tif file.ID == \"\" {\n\t\t\t\tt.Error(\"Expected file_id to be populated\")\n\t\t\t}\n\t\t\tif file.Filename == \"\" {\n\t\t\t\tt.Error(\"Expected filename to be populated\")\n\t\t\t}\n\t\t\tif file.ContentType == \"\" {\n\t\t\t\tt.Error(\"Expected content_type to be populated\")\n\t\t\t}\n\t\t}\n\t})\n\n\tt.Logf(\"Successfully tested list functionality with %d files\", len(uploadedFiles))\n}\n\nfunc TestManagerLocalPath(t *testing.T) {\n\ttest.Prepare(t, config.Conf)\n\tdefer test.Clean()\n\n\t// Test with local storage\n\tt.Run(\"LocalStorage\", func(t *testing.T) {\n\t\t// Create a local storage manager\n\t\tmanager, err := New(ManagerOption{\n\t\t\tDriver: \"local\",\n\t\t\tMaxSize: \"10M\",\n\t\t\tAllowedTypes: []string{\"text/*\", \"image/*\", \"application/*\", \".txt\", \".json\", \".html\", \".csv\", \".yao\"},\n\t\t\tOptions: map[string]interface{}{\n\t\t\t\t\"path\": \"/tmp/test_localpath_attachments\",\n\t\t\t},\n\t\t})\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to create local manager: %v\", err)\n\t\t}\n\t\tmanager.Name = \"localpath-test\"\n\n\t\t// Test different file types\n\t\ttestFiles := []struct {\n\t\t\tfilename string\n\t\t\tcontent string\n\t\t\tcontentType string\n\t\t\texpectedCT string\n\t\t}{\n\t\t\t{\"test.txt\", \"Hello LocalPath\", \"text/plain\", \"text/plain\"},\n\t\t\t{\"test.json\", `{\"localpath\": \"test\"}`, \"application/json\", \"application/json\"},\n\t\t\t{\"test.html\", \"LocalPath Test\", \"text/html\", \"text/html\"},\n\t\t\t{\"test.csv\", \"col1,col2\\nlocalpath,test\", \"text/csv\", \"text/csv\"},\n\t\t\t{\"test.yao\", \"localpath yao content\", \"application/yao\", \"application/yao\"},\n\t\t}\n\n\t\tfor _, tf := range testFiles {\n\t\t\t// Upload file\n\t\t\treader := strings.NewReader(tf.content)\n\t\t\tfileHeader := &FileHeader{\n\t\t\t\tFileHeader: &multipart.FileHeader{\n\t\t\t\t\tFilename: tf.filename,\n\t\t\t\t\tSize: int64(len(tf.content)),\n\t\t\t\t\tHeader: make(map[string][]string),\n\t\t\t\t},\n\t\t\t}\n\t\t\tfileHeader.Header.Set(\"Content-Type\", tf.contentType)\n\n\t\t\toption := UploadOption{\n\t\t\t\tGroups: []string{\"localpath\", \"test\"},\n\t\t\t\tOriginalFilename: tf.filename,\n\t\t\t}\n\n\t\t\tfile, err := manager.Upload(context.Background(), fileHeader, reader, option)\n\t\t\tif err != nil {\n\t\t\t\tt.Fatalf(\"Failed to upload file %s: %v\", tf.filename, err)\n\t\t\t}\n\n\t\t\t// Test LocalPath\n\t\t\tlocalPath, detectedCT, err := manager.LocalPath(context.Background(), file.ID)\n\t\t\tif err != nil {\n\t\t\t\tt.Fatalf(\"Failed to get local path for %s: %v\", tf.filename, err)\n\t\t\t}\n\n\t\t\t// Verify path is absolute\n\t\t\tif !filepath.IsAbs(localPath) {\n\t\t\t\tt.Errorf(\"Expected absolute path for %s, got: %s\", tf.filename, localPath)\n\t\t\t}\n\n\t\t\t// Verify content type\n\t\t\tif detectedCT != tf.expectedCT {\n\t\t\t\tt.Errorf(\"Expected content type %s for %s, got: %s\", tf.expectedCT, tf.filename, detectedCT)\n\t\t\t}\n\n\t\t\t// Verify file exists\n\t\t\tif _, err := os.Stat(localPath); os.IsNotExist(err) {\n\t\t\t\tt.Errorf(\"File should exist at local path %s for %s\", localPath, tf.filename)\n\t\t\t}\n\n\t\t\t// Verify file content\n\t\t\tfileContent, err := os.ReadFile(localPath)\n\t\t\tif err != nil {\n\t\t\t\tt.Fatalf(\"Failed to read file at local path for %s: %v\", tf.filename, err)\n\t\t\t}\n\n\t\t\tif string(fileContent) != tf.content {\n\t\t\t\tt.Errorf(\"File content mismatch for %s. Expected: %s, Got: %s\", tf.filename, tf.content, string(fileContent))\n\t\t\t}\n\n\t\t\tt.Logf(\"File %s - ID: %s, LocalPath: %s, ContentType: %s\", tf.filename, file.ID, localPath, detectedCT)\n\t\t}\n\t})\n\n\t// Test with gzipped files in local storage\n\tt.Run(\"LocalStorage_Gzipped\", func(t *testing.T) {\n\t\tmanager, err := New(ManagerOption{\n\t\t\tDriver: \"local\",\n\t\t\tMaxSize: \"10M\",\n\t\t\tAllowedTypes: []string{\"text/*\"},\n\t\t\tOptions: map[string]interface{}{\n\t\t\t\t\"path\": \"/tmp/test_localpath_gzip_attachments\",\n\t\t\t},\n\t\t})\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to create local manager: %v\", err)\n\t\t}\n\t\tmanager.Name = \"localpath-gzip-test\"\n\n\t\tcontent := \"This content will be gzipped\"\n\t\treader := strings.NewReader(content)\n\n\t\tfileHeader := &FileHeader{\n\t\t\tFileHeader: &multipart.FileHeader{\n\t\t\t\tFilename: \"gzipped.txt\",\n\t\t\t\tSize: int64(len(content)),\n\t\t\t\tHeader: make(map[string][]string),\n\t\t\t},\n\t\t}\n\t\tfileHeader.Header.Set(\"Content-Type\", \"text/plain\")\n\n\t\toption := UploadOption{\n\t\t\tGroups: []string{\"gzip\", \"test\"},\n\t\t\tOriginalFilename: \"gzipped.txt\",\n\t\t\tGzip: true, // Enable gzip compression\n\t\t}\n\n\t\tfile, err := manager.Upload(context.Background(), fileHeader, reader, option)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to upload gzipped file: %v\", err)\n\t\t}\n\n\t\t// Test LocalPath - should get decompressed content\n\t\tlocalPath, contentType, err := manager.LocalPath(context.Background(), file.ID)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to get local path for gzipped file: %v\", err)\n\t\t}\n\n\t\t// Verify content type\n\t\tif contentType != \"text/plain\" {\n\t\t\tt.Errorf(\"Expected content type text/plain, got: %s\", contentType)\n\t\t}\n\n\t\t// For gzipped files in local storage, the storage path ends with .gz\n\t\t// but the content should be accessible normally through Read methods\n\t\tfileContent, err := manager.Read(context.Background(), file.ID)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to read gzipped file: %v\", err)\n\t\t}\n\n\t\tif string(fileContent) != content {\n\t\t\tt.Errorf(\"Gzipped file content mismatch. Expected: %s, Got: %s\", content, string(fileContent))\n\t\t}\n\n\t\tt.Logf(\"Gzipped file - ID: %s, LocalPath: %s, ContentType: %s\", file.ID, localPath, contentType)\n\t})\n}\n\nfunc TestManagerLocalPath_NonExistentFile(t *testing.T) {\n\ttest.Prepare(t, config.Conf)\n\tdefer test.Clean()\n\n\tmanager, err := New(ManagerOption{\n\t\tDriver: \"local\",\n\t\tAllowedTypes: []string{\"text/*\"},\n\t\tOptions: map[string]interface{}{\n\t\t\t\"path\": \"/tmp/test_localpath_nonexistent\",\n\t\t},\n\t})\n\tif err != nil {\n\t\tt.Fatalf(\"Failed to create manager: %v\", err)\n\t}\n\tmanager.Name = \"nonexistent-test\"\n\n\t// Test with non-existent file ID\n\t_, _, err = manager.LocalPath(context.Background(), \"non-existent-file-id\")\n\tif err == nil {\n\t\tt.Error(\"Expected error for non-existent file ID\")\n\t}\n\n\t// Should contain \"file not found\" in the error chain\n\tif !strings.Contains(err.Error(), \"file not found\") {\n\t\tt.Errorf(\"Expected 'file not found' in error message, got: %s\", err.Error())\n\t}\n}\n\nfunc TestPublicAndShareFields(t *testing.T) {\n\ttest.Prepare(t, config.Conf)\n\tdefer test.Clean()\n\n\t// Force re-migrate the attachment table to ensure schema is up to date\n\tm := model.Select(\"__yao.attachment\")\n\tif m != nil {\n\t\t// Drop and recreate table to get latest schema\n\t\terr := m.DropTable()\n\t\tif err != nil {\n\t\t\tt.Logf(\"Warning: failed to drop table: %v\", err)\n\t\t}\n\t\terr = m.Migrate(false)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to migrate table: %v\", err)\n\t\t}\n\t}\n\n\tmanager, err := RegisterDefault(\"test-public-share\")\n\tif err != nil {\n\t\tt.Fatalf(\"Failed to register manager: %v\", err)\n\t}\n\n\t// Test 1: Upload with public=true and share=team\n\tt.Run(\"PublicTeamShare\", func(t *testing.T) {\n\t\tcontent := \"Public team shared file\"\n\t\treader := strings.NewReader(content)\n\n\t\tfileHeader := &FileHeader{\n\t\t\tFileHeader: &multipart.FileHeader{\n\t\t\t\tFilename: \"public-team.txt\",\n\t\t\t\tSize: int64(len(content)),\n\t\t\t\tHeader: make(map[string][]string),\n\t\t\t},\n\t\t}\n\t\tfileHeader.Header.Set(\"Content-Type\", \"text/plain\")\n\n\t\toption := UploadOption{\n\t\t\tGroups: []string{\"test\"},\n\t\t\tOriginalFilename: \"public-team.txt\",\n\t\t\tPublic: true,\n\t\t\tShare: \"team\",\n\t\t}\n\n\t\tfile, err := manager.Upload(context.Background(), fileHeader, reader, option)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to upload public team file: %v\", err)\n\t\t}\n\n\t\t// Verify in database\n\t\tm := model.Select(\"__yao.attachment\")\n\t\trecords, err := m.Get(model.QueryParam{\n\t\t\tWheres: []model.QueryWhere{\n\t\t\t\t{Column: \"file_id\", Value: file.ID},\n\t\t\t},\n\t\t})\n\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to query database: %v\", err)\n\t\t}\n\n\t\tif len(records) == 0 {\n\t\t\tt.Fatal(\"No record found in database\")\n\t\t}\n\n\t\t// Debug: print all fields\n\t\tt.Logf(\"Record fields: %+v\", records[0])\n\n\t\tpublicValue := toBool(records[0][\"public\"])\n\t\tif !publicValue {\n\t\t\tt.Errorf(\"Expected public to be true, got: %v (type: %T)\", records[0][\"public\"], records[0][\"public\"])\n\t\t}\n\n\t\tshareValue := toString(records[0][\"share\"])\n\t\tif shareValue != \"team\" {\n\t\t\tt.Errorf(\"Expected share to be 'team', got: %v (type: %T)\", records[0][\"share\"], records[0][\"share\"])\n\t\t}\n\t})\n\n\t// Test 2: Upload with public=false and share=private (default)\n\tt.Run(\"PrivateShare\", func(t *testing.T) {\n\t\tcontent := \"Private file\"\n\t\treader := strings.NewReader(content)\n\n\t\tfileHeader := &FileHeader{\n\t\t\tFileHeader: &multipart.FileHeader{\n\t\t\t\tFilename: \"private.txt\",\n\t\t\t\tSize: int64(len(content)),\n\t\t\t\tHeader: make(map[string][]string),\n\t\t\t},\n\t\t}\n\t\tfileHeader.Header.Set(\"Content-Type\", \"text/plain\")\n\n\t\toption := UploadOption{\n\t\t\tGroups: []string{\"test\"},\n\t\t\tOriginalFilename: \"private.txt\",\n\t\t\tPublic: false,\n\t\t\tShare: \"private\",\n\t\t}\n\n\t\tfile, err := manager.Upload(context.Background(), fileHeader, reader, option)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to upload private file: %v\", err)\n\t\t}\n\n\t\t// Verify in database\n\t\tm := model.Select(\"__yao.attachment\")\n\t\trecords, err := m.Get(model.QueryParam{\n\t\t\tSelect: []interface{}{\"public\", \"share\"},\n\t\t\tWheres: []model.QueryWhere{\n\t\t\t\t{Column: \"file_id\", Value: file.ID},\n\t\t\t},\n\t\t})\n\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to query database: %v\", err)\n\t\t}\n\n\t\tif len(records) == 0 {\n\t\t\tt.Fatal(\"No record found in database\")\n\t\t}\n\n\t\tpublicValue := toBool(records[0][\"public\"])\n\t\tif publicValue {\n\t\t\tt.Errorf(\"Expected public to be false, got: %v\", records[0][\"public\"])\n\t\t}\n\n\t\tshareValue := toString(records[0][\"share\"])\n\t\tif shareValue != \"private\" {\n\t\t\tt.Errorf(\"Expected share to be 'private', got: %v\", records[0][\"share\"])\n\t\t}\n\t})\n\n\t// Test 3: Upload without specifying share (should default to private)\n\tt.Run(\"DefaultSharePrivate\", func(t *testing.T) {\n\t\tcontent := \"Default share file\"\n\t\treader := strings.NewReader(content)\n\n\t\tfileHeader := &FileHeader{\n\t\t\tFileHeader: &multipart.FileHeader{\n\t\t\t\tFilename: \"default-share.txt\",\n\t\t\t\tSize: int64(len(content)),\n\t\t\t\tHeader: make(map[string][]string),\n\t\t\t},\n\t\t}\n\t\tfileHeader.Header.Set(\"Content-Type\", \"text/plain\")\n\n\t\toption := UploadOption{\n\t\t\tGroups: []string{\"test\"},\n\t\t\tOriginalFilename: \"default-share.txt\",\n\t\t\tPublic: false,\n\t\t\t// Share not specified, should default to \"private\"\n\t\t}\n\n\t\tfile, err := manager.Upload(context.Background(), fileHeader, reader, option)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to upload file with default share: %v\", err)\n\t\t}\n\n\t\t// Verify in database\n\t\tm := model.Select(\"__yao.attachment\")\n\t\trecords, err := m.Get(model.QueryParam{\n\t\t\tSelect: []interface{}{\"share\"},\n\t\t\tWheres: []model.QueryWhere{\n\t\t\t\t{Column: \"file_id\", Value: file.ID},\n\t\t\t},\n\t\t})\n\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to query database: %v\", err)\n\t\t}\n\n\t\tif len(records) == 0 {\n\t\t\tt.Fatal(\"No record found in database\")\n\t\t}\n\n\t\tshareValue := toString(records[0][\"share\"])\n\t\tif shareValue != \"private\" {\n\t\t\tt.Errorf(\"Expected default share to be 'private', got: %v\", records[0][\"share\"])\n\t\t}\n\t})\n}\n\nfunc TestYaoPermissionFields(t *testing.T) {\n\ttest.Prepare(t, config.Conf)\n\tdefer test.Clean()\n\n\t// Force re-migrate the attachment table to ensure schema is up to date\n\tm := model.Select(\"__yao.attachment\")\n\tif m != nil {\n\t\t// Drop and recreate table to get latest schema\n\t\terr := m.DropTable()\n\t\tif err != nil {\n\t\t\tt.Logf(\"Warning: failed to drop table: %v\", err)\n\t\t}\n\t\terr = m.Migrate(false)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to migrate table: %v\", err)\n\t\t}\n\t}\n\n\tmanager, err := RegisterDefault(\"test-yao-permission\")\n\tif err != nil {\n\t\tt.Fatalf(\"Failed to register manager: %v\", err)\n\t}\n\n\t// Test 1: Upload with all Yao permission fields\n\tt.Run(\"AllYaoFields\", func(t *testing.T) {\n\t\tcontent := \"File with all Yao permission fields\"\n\t\treader := strings.NewReader(content)\n\n\t\tfileHeader := &FileHeader{\n\t\t\tFileHeader: &multipart.FileHeader{\n\t\t\t\tFilename: \"yao-all-fields.txt\",\n\t\t\t\tSize: int64(len(content)),\n\t\t\t\tHeader: make(map[string][]string),\n\t\t\t},\n\t\t}\n\t\tfileHeader.Header.Set(\"Content-Type\", \"text/plain\")\n\n\t\toption := UploadOption{\n\t\t\tGroups: []string{\"test\"},\n\t\t\tOriginalFilename: \"yao-all-fields.txt\",\n\t\t\tYaoCreatedBy: \"user123\",\n\t\t\tYaoUpdatedBy: \"user123\",\n\t\t\tYaoTeamID: \"team456\",\n\t\t\tYaoTenantID: \"tenant789\",\n\t\t}\n\n\t\tfile, err := manager.Upload(context.Background(), fileHeader, reader, option)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to upload file with Yao fields: %v\", err)\n\t\t}\n\n\t\t// Verify in database\n\t\tm := model.Select(\"__yao.attachment\")\n\t\trecords, err := m.Get(model.QueryParam{\n\t\t\tSelect: []interface{}{\"__yao_created_by\", \"__yao_updated_by\", \"__yao_team_id\", \"__yao_tenant_id\"},\n\t\t\tWheres: []model.QueryWhere{\n\t\t\t\t{Column: \"file_id\", Value: file.ID},\n\t\t\t},\n\t\t})\n\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to query database: %v\", err)\n\t\t}\n\n\t\tif len(records) == 0 {\n\t\t\tt.Fatal(\"No record found in database\")\n\t\t}\n\n\t\t// Verify __yao_created_by\n\t\tcreatedBy := toString(records[0][\"__yao_created_by\"])\n\t\tif createdBy != \"user123\" {\n\t\t\tt.Errorf(\"Expected __yao_created_by to be 'user123', got: %v\", records[0][\"__yao_created_by\"])\n\t\t}\n\n\t\t// Verify __yao_updated_by\n\t\tupdatedBy := toString(records[0][\"__yao_updated_by\"])\n\t\tif updatedBy != \"user123\" {\n\t\t\tt.Errorf(\"Expected __yao_updated_by to be 'user123', got: %v\", records[0][\"__yao_updated_by\"])\n\t\t}\n\n\t\t// Verify __yao_team_id\n\t\tteamID := toString(records[0][\"__yao_team_id\"])\n\t\tif teamID != \"team456\" {\n\t\t\tt.Errorf(\"Expected __yao_team_id to be 'team456', got: %v\", records[0][\"__yao_team_id\"])\n\t\t}\n\n\t\t// Verify __yao_tenant_id\n\t\ttenantID := toString(records[0][\"__yao_tenant_id\"])\n\t\tif tenantID != \"tenant789\" {\n\t\t\tt.Errorf(\"Expected __yao_tenant_id to be 'tenant789', got: %v\", records[0][\"__yao_tenant_id\"])\n\t\t}\n\t})\n\n\t// Test 2: Upload with partial Yao fields (only team and tenant)\n\tt.Run(\"PartialYaoFields\", func(t *testing.T) {\n\t\tcontent := \"File with partial Yao fields\"\n\t\treader := strings.NewReader(content)\n\n\t\tfileHeader := &FileHeader{\n\t\t\tFileHeader: &multipart.FileHeader{\n\t\t\t\tFilename: \"yao-partial-fields.txt\",\n\t\t\t\tSize: int64(len(content)),\n\t\t\t\tHeader: make(map[string][]string),\n\t\t\t},\n\t\t}\n\t\tfileHeader.Header.Set(\"Content-Type\", \"text/plain\")\n\n\t\toption := UploadOption{\n\t\t\tGroups: []string{\"test\"},\n\t\t\tOriginalFilename: \"yao-partial-fields.txt\",\n\t\t\tYaoTeamID: \"team999\",\n\t\t\tYaoTenantID: \"tenant888\",\n\t\t\t// YaoCreatedBy and YaoUpdatedBy not specified\n\t\t}\n\n\t\tfile, err := manager.Upload(context.Background(), fileHeader, reader, option)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to upload file with partial Yao fields: %v\", err)\n\t\t}\n\n\t\t// Verify in database\n\t\tm := model.Select(\"__yao.attachment\")\n\t\trecords, err := m.Get(model.QueryParam{\n\t\t\tSelect: []interface{}{\"__yao_team_id\", \"__yao_tenant_id\"},\n\t\t\tWheres: []model.QueryWhere{\n\t\t\t\t{Column: \"file_id\", Value: file.ID},\n\t\t\t},\n\t\t})\n\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to query database: %v\", err)\n\t\t}\n\n\t\tif len(records) == 0 {\n\t\t\tt.Fatal(\"No record found in database\")\n\t\t}\n\n\t\t// Verify __yao_team_id\n\t\tteamID := toString(records[0][\"__yao_team_id\"])\n\t\tif teamID != \"team999\" {\n\t\t\tt.Errorf(\"Expected __yao_team_id to be 'team999', got: %v\", records[0][\"__yao_team_id\"])\n\t\t}\n\n\t\t// Verify __yao_tenant_id\n\t\ttenantID := toString(records[0][\"__yao_tenant_id\"])\n\t\tif tenantID != \"tenant888\" {\n\t\t\tt.Errorf(\"Expected __yao_tenant_id to be 'tenant888', got: %v\", records[0][\"__yao_tenant_id\"])\n\t\t}\n\t})\n\n\t// Test 3: Upload without Yao fields (should be null/empty in database)\n\tt.Run(\"NoYaoFields\", func(t *testing.T) {\n\t\tcontent := \"File without Yao fields\"\n\t\treader := strings.NewReader(content)\n\n\t\tfileHeader := &FileHeader{\n\t\t\tFileHeader: &multipart.FileHeader{\n\t\t\t\tFilename: \"yao-no-fields.txt\",\n\t\t\t\tSize: int64(len(content)),\n\t\t\t\tHeader: make(map[string][]string),\n\t\t\t},\n\t\t}\n\t\tfileHeader.Header.Set(\"Content-Type\", \"text/plain\")\n\n\t\toption := UploadOption{\n\t\t\tGroups: []string{\"test\"},\n\t\t\tOriginalFilename: \"yao-no-fields.txt\",\n\t\t\t// No Yao fields specified\n\t\t}\n\n\t\tfile, err := manager.Upload(context.Background(), fileHeader, reader, option)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to upload file without Yao fields: %v\", err)\n\t\t}\n\n\t\t// Should succeed without errors\n\t\tif file.ID == \"\" {\n\t\t\tt.Error(\"File ID should not be empty\")\n\t\t}\n\n\t\tt.Logf(\"Successfully uploaded file without Yao fields - ID: %s\", file.ID)\n\t})\n}\n\nfunc TestManagerLocalPath_ValidationFlow(t *testing.T) {\n\ttest.Prepare(t, config.Conf)\n\tdefer test.Clean()\n\n\tmanager, err := New(ManagerOption{\n\t\tDriver: \"local\",\n\t\tAllowedTypes: []string{\"text/*\"},\n\t\tOptions: map[string]interface{}{\n\t\t\t\"path\": \"/tmp/test_localpath_validation\",\n\t\t},\n\t})\n\tif err != nil {\n\t\tt.Fatalf(\"Failed to create manager: %v\", err)\n\t}\n\tmanager.Name = \"validation-test\"\n\n\t// Upload a file\n\tcontent := \"Validation flow test content\"\n\treader := strings.NewReader(content)\n\n\tfileHeader := &FileHeader{\n\t\tFileHeader: &multipart.FileHeader{\n\t\t\tFilename: \"validation.txt\",\n\t\t\tSize: int64(len(content)),\n\t\t\tHeader: make(map[string][]string),\n\t\t},\n\t}\n\tfileHeader.Header.Set(\"Content-Type\", \"text/plain\")\n\n\toption := UploadOption{\n\t\tGroups: []string{\"validation\"},\n\t\tOriginalFilename: \"original-validation.txt\",\n\t}\n\n\tfile, err := manager.Upload(context.Background(), fileHeader, reader, option)\n\tif err != nil {\n\t\tt.Fatalf(\"Failed to upload file: %v\", err)\n\t}\n\n\t// Test complete flow: Upload -> LocalPath -> Verify -> Delete\n\tt.Run(\"CompleteFlow\", func(t *testing.T) {\n\t\t// Get local path\n\t\tlocalPath, contentType, err := manager.LocalPath(context.Background(), file.ID)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to get local path: %v\", err)\n\t\t}\n\n\t\t// Verify all properties\n\t\tif !filepath.IsAbs(localPath) {\n\t\t\tt.Error(\"Path should be absolute\")\n\t\t}\n\n\t\tif contentType != \"text/plain\" {\n\t\t\tt.Errorf(\"Expected content type text/plain, got: %s\", contentType)\n\t\t}\n\n\t\t// Verify file exists\n\t\tstat, err := os.Stat(localPath)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"File should exist at local path: %v\", err)\n\t\t}\n\n\t\tif stat.Size() != int64(len(content)) {\n\t\t\tt.Errorf(\"File size mismatch. Expected: %d, Got: %d\", len(content), stat.Size())\n\t\t}\n\n\t\t// Verify file content matches\n\t\tfileContent, err := os.ReadFile(localPath)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to read file: %v\", err)\n\t\t}\n\n\t\tif string(fileContent) != content {\n\t\t\tt.Errorf(\"Content mismatch. Expected: %s, Got: %s\", content, string(fileContent))\n\t\t}\n\n\t\t// Verify through manager's Read method as well\n\t\tmanagerContent, err := manager.Read(context.Background(), file.ID)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to read through manager: %v\", err)\n\t\t}\n\n\t\tif string(managerContent) != content {\n\t\t\tt.Errorf(\"Manager read content mismatch. Expected: %s, Got: %s\", content, string(managerContent))\n\t\t}\n\n\t\tt.Logf(\"Validation complete - LocalPath: %s, Size: %d bytes, ContentType: %s\", localPath, stat.Size(), contentType)\n\t})\n\n\t// Clean up\n\terr = manager.Delete(context.Background(), file.ID)\n\tif err != nil {\n\t\tt.Logf(\"Warning: Failed to delete test file: %v\", err)\n\t}\n}\n\nfunc TestGetTextAndSaveText(t *testing.T) {\n\ttest.Prepare(t, config.Conf)\n\tdefer test.Clean()\n\n\tmanager, err := RegisterDefault(\"test-text-content\")\n\tif err != nil {\n\t\tt.Fatalf(\"Failed to register manager: %v\", err)\n\t}\n\n\t// Upload a test file\n\tcontent := \"This is a test file for text content storage\"\n\treader := strings.NewReader(content)\n\n\tfileHeader := &FileHeader{\n\t\tFileHeader: &multipart.FileHeader{\n\t\t\tFilename: \"test-text.txt\",\n\t\t\tSize: int64(len(content)),\n\t\t\tHeader: make(map[string][]string),\n\t\t},\n\t}\n\tfileHeader.Header.Set(\"Content-Type\", \"text/plain\")\n\n\toption := UploadOption{\n\t\tGroups: []string{\"test\"},\n\t\tOriginalFilename: \"test-text.txt\",\n\t}\n\n\tfile, err := manager.Upload(context.Background(), fileHeader, reader, option)\n\tif err != nil {\n\t\tt.Fatalf(\"Failed to upload file: %v\", err)\n\t}\n\n\t// Test 1: GetText on file without saved text (should return empty)\n\tt.Run(\"GetTextEmpty\", func(t *testing.T) {\n\t\ttext, err := manager.GetText(context.Background(), file.ID)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to get text: %v\", err)\n\t\t}\n\n\t\tif text != \"\" {\n\t\t\tt.Errorf(\"Expected empty text, got: %s\", text)\n\t\t}\n\n\t\t// Also test full content\n\t\tfullText, err := manager.GetText(context.Background(), file.ID, true)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to get full text: %v\", err)\n\t\t}\n\n\t\tif fullText != \"\" {\n\t\t\tt.Errorf(\"Expected empty full text, got: %s\", fullText)\n\t\t}\n\t})\n\n\t// Test 2: SaveText and verify\n\tt.Run(\"SaveTextAndVerify\", func(t *testing.T) {\n\t\tparsedText := \"This is the parsed text content from the file. It could be extracted from PDF, Word, or image OCR.\"\n\n\t\terr := manager.SaveText(context.Background(), file.ID, parsedText)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to save text: %v\", err)\n\t\t}\n\n\t\t// Retrieve the saved text\n\t\tretrievedText, err := manager.GetText(context.Background(), file.ID)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to get saved text: %v\", err)\n\t\t}\n\n\t\tif retrievedText != parsedText {\n\t\t\tt.Errorf(\"Text mismatch. Expected: %s, Got: %s\", parsedText, retrievedText)\n\t\t}\n\n\t\tt.Logf(\"Successfully saved and retrieved text content (%d characters)\", len(retrievedText))\n\t})\n\n\t// Test 3: Update existing text\n\tt.Run(\"UpdateText\", func(t *testing.T) {\n\t\tupdatedText := \"This is the updated parsed text content with additional information.\"\n\n\t\terr := manager.SaveText(context.Background(), file.ID, updatedText)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to update text: %v\", err)\n\t\t}\n\n\t\tretrievedText, err := manager.GetText(context.Background(), file.ID)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to get updated text: %v\", err)\n\t\t}\n\n\t\tif retrievedText != updatedText {\n\t\t\tt.Errorf(\"Updated text mismatch. Expected: %s, Got: %s\", updatedText, retrievedText)\n\t\t}\n\t})\n\n\t// Test 4: Save long text content and verify preview vs full content\n\tt.Run(\"SaveLongText\", func(t *testing.T) {\n\t\t// Generate a large text content (10KB)\n\t\tlongText := strings.Repeat(\"This is a long text content that simulates parsing from a large document like PDF or Word. \", 100)\n\n\t\terr := manager.SaveText(context.Background(), file.ID, longText)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to save long text: %v\", err)\n\t\t}\n\n\t\t// Get preview (default, should be limited to 2000 characters)\n\t\tpreviewText, err := manager.GetText(context.Background(), file.ID)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to get preview text: %v\", err)\n\t\t}\n\n\t\t// Preview should be exactly 2000 characters (runes)\n\t\tpreviewRunes := []rune(previewText)\n\t\tif len(previewRunes) != 2000 {\n\t\t\tt.Errorf(\"Preview length mismatch. Expected: 2000 runes, Got: %d runes\", len(previewRunes))\n\t\t}\n\n\t\t// Get full content\n\t\tfullText, err := manager.GetText(context.Background(), file.ID, true)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to get full text: %v\", err)\n\t\t}\n\n\t\tif fullText != longText {\n\t\t\tt.Errorf(\"Full text mismatch. Expected length: %d, Got: %d\", len(longText), len(fullText))\n\t\t}\n\n\t\tt.Logf(\"Successfully saved long text - Preview: %d chars, Full: %d chars\", len(previewText), len(fullText))\n\t})\n\n\t// Test 5: Test UTF-8 character handling in preview\n\tt.Run(\"UTF8PreviewHandling\", func(t *testing.T) {\n\t\t// Create text with multi-byte UTF-8 characters (Chinese, emoji, etc.)\n\t\t// Each Chinese character is 3 bytes, emoji is 4 bytes\n\t\tchineseText := strings.Repeat(\"这是一个测试文本，包含中文字符。\", 150) // Should exceed 2000 chars\n\t\temojiText := strings.Repeat(\"Hello 👋 World 🌍 \", 150)\n\n\t\t// Test with Chinese text\n\t\terr := manager.SaveText(context.Background(), file.ID, chineseText)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to save Chinese text: %v\", err)\n\t\t}\n\n\t\tpreviewChinese, err := manager.GetText(context.Background(), file.ID)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to get Chinese preview: %v\", err)\n\t\t}\n\n\t\t// Should be exactly 2000 runes (characters), not bytes\n\t\tif len([]rune(previewChinese)) != 2000 {\n\t\t\tt.Errorf(\"Chinese preview should be 2000 runes, got: %d\", len([]rune(previewChinese)))\n\t\t}\n\n\t\t// Full text should be complete\n\t\tfullChinese, err := manager.GetText(context.Background(), file.ID, true)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to get full Chinese text: %v\", err)\n\t\t}\n\n\t\tif fullChinese != chineseText {\n\t\t\tt.Errorf(\"Chinese text mismatch\")\n\t\t}\n\n\t\t// Test with emoji text\n\t\terr = manager.SaveText(context.Background(), file.ID, emojiText)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to save emoji text: %v\", err)\n\t\t}\n\n\t\tpreviewEmoji, err := manager.GetText(context.Background(), file.ID)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to get emoji preview: %v\", err)\n\t\t}\n\n\t\tif len([]rune(previewEmoji)) != 2000 {\n\t\t\tt.Errorf(\"Emoji preview should be 2000 runes, got: %d\", len([]rune(previewEmoji)))\n\t\t}\n\n\t\tt.Logf(\"UTF-8 handling verified - Chinese: %d bytes, Emoji: %d bytes\",\n\t\t\tlen(previewChinese), len(previewEmoji))\n\t})\n\n\t// Test 6: GetText with non-existent file ID\n\tt.Run(\"GetTextNonExistent\", func(t *testing.T) {\n\t\t_, err := manager.GetText(context.Background(), \"non-existent-id\")\n\t\tif err == nil {\n\t\t\tt.Error(\"Expected error for non-existent file ID\")\n\t\t}\n\n\t\tif !strings.Contains(err.Error(), \"file not found\") {\n\t\t\tt.Errorf(\"Expected 'file not found' error, got: %s\", err.Error())\n\t\t}\n\t})\n\n\t// Test 7: SaveText with non-existent file ID\n\tt.Run(\"SaveTextNonExistent\", func(t *testing.T) {\n\t\terr := manager.SaveText(context.Background(), \"non-existent-id\", \"some text\")\n\t\tif err == nil {\n\t\t\tt.Error(\"Expected error for non-existent file ID\")\n\t\t}\n\n\t\tif !strings.Contains(err.Error(), \"file not found\") {\n\t\t\tt.Errorf(\"Expected 'file not found' error, got: %s\", err.Error())\n\t\t}\n\t})\n\n\t// Test 8: Save empty text (clear content)\n\tt.Run(\"SaveEmptyText\", func(t *testing.T) {\n\t\terr := manager.SaveText(context.Background(), file.ID, \"\")\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to save empty text: %v\", err)\n\t\t}\n\n\t\tretrievedText, err := manager.GetText(context.Background(), file.ID)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to get empty text: %v\", err)\n\t\t}\n\n\t\tif retrievedText != \"\" {\n\t\t\tt.Errorf(\"Expected empty text, got: %s\", retrievedText)\n\t\t}\n\t})\n\n\t// Test 9: Verify List doesn't include content fields by default\n\tt.Run(\"ListExcludesContentByDefault\", func(t *testing.T) {\n\t\t// Save some text content\n\t\ttestText := \"This text should not appear in list results by default\"\n\t\terr := manager.SaveText(context.Background(), file.ID, testText)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to save text: %v\", err)\n\t\t}\n\n\t\t// List files without specifying select fields\n\t\tresult, err := manager.List(context.Background(), ListOption{\n\t\t\tFilters: map[string]interface{}{\n\t\t\t\t\"file_id\": file.ID,\n\t\t\t},\n\t\t})\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to list files: %v\", err)\n\t\t}\n\n\t\tif len(result.Files) == 0 {\n\t\t\tt.Fatal(\"Expected to find at least one file\")\n\t\t}\n\n\t\t// The List method returns File structs, but we need to verify\n\t\t// the database query doesn't fetch the content field\n\t\t// We can verify this by checking the database directly\n\t\tm := model.Select(\"__yao.attachment\")\n\t\trecords, err := m.Get(model.QueryParam{\n\t\t\tWheres: []model.QueryWhere{\n\t\t\t\t{Column: \"file_id\", Value: file.ID},\n\t\t\t},\n\t\t})\n\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to query database: %v\", err)\n\t\t}\n\n\t\t// When we do a full select, content should be present\n\t\tif len(records) > 0 {\n\t\t\tif content, ok := records[0][\"content\"].(string); ok && content == testText {\n\t\t\t\tt.Logf(\"Content field exists in full query (expected): %d characters\", len(content))\n\t\t\t}\n\t\t}\n\t})\n\n\t// Test 10: Verify content can be explicitly selected in List\n\tt.Run(\"ListIncludesContentWhenExplicitlySelected\", func(t *testing.T) {\n\t\t// Save some text content\n\t\ttestText := \"This text SHOULD appear when explicitly selected\"\n\t\terr := manager.SaveText(context.Background(), file.ID, testText)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to save text: %v\", err)\n\t\t}\n\n\t\t// List files WITH content field explicitly selected\n\t\tresult, err := manager.List(context.Background(), ListOption{\n\t\t\tSelect: []string{\"file_id\", \"name\", \"content\"},\n\t\t\tFilters: map[string]interface{}{\n\t\t\t\t\"file_id\": file.ID,\n\t\t\t},\n\t\t})\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to list files with content: %v\", err)\n\t\t}\n\n\t\tif len(result.Files) == 0 {\n\t\t\tt.Fatal(\"Expected to find at least one file\")\n\t\t}\n\n\t\t// Query database directly to verify content is included\n\t\tm := model.Select(\"__yao.attachment\")\n\t\trecords, err := m.Get(model.QueryParam{\n\t\t\tSelect: []interface{}{\"file_id\", \"name\", \"content\"},\n\t\t\tWheres: []model.QueryWhere{\n\t\t\t\t{Column: \"file_id\", Value: file.ID},\n\t\t\t},\n\t\t})\n\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to query database: %v\", err)\n\t\t}\n\n\t\tif len(records) == 0 {\n\t\t\tt.Fatal(\"Expected to find record\")\n\t\t}\n\n\t\t// Verify content is present\n\t\tif content, ok := records[0][\"content\"].(string); ok {\n\t\t\tif content != testText {\n\t\t\t\tt.Errorf(\"Expected content '%s', got '%s'\", testText, content)\n\t\t\t}\n\t\t\tt.Logf(\"Content field correctly included when explicitly selected: %d characters\", len(content))\n\t\t} else {\n\t\t\tt.Error(\"Content field not found when explicitly selected\")\n\t\t}\n\t})\n\n\t// Clean up\n\terr = manager.Delete(context.Background(), file.ID)\n\tif err != nil {\n\t\tt.Logf(\"Warning: Failed to delete test file: %v\", err)\n\t}\n}\n" }, { "path": "attachment/process.go", "content": "package attachment\n\nimport (\n\t\"archive/zip\"\n\t\"bytes\"\n\t\"context\"\n\t\"encoding/base64\"\n\t\"fmt\"\n\t\"mime\"\n\t\"mime/multipart\"\n\t\"net/textproto\"\n\t\"path/filepath\"\n\t\"strings\"\n\n\t\"github.com/yaoapp/gou/model\"\n\t\"github.com/yaoapp/gou/process\"\n\t\"github.com/yaoapp/kun/any\"\n\t\"github.com/yaoapp/kun/maps\"\n)\n\n// Init registers all attachment processes\nfunc Init() {\n\tprocess.RegisterGroup(\"attachment\", map[string]process.Handler{\n\t\t\"Save\": processSave,\n\t\t\"Read\": processRead,\n\t\t\"Info\": processInfo,\n\t\t\"List\": processList,\n\t\t\"Delete\": processDelete,\n\t\t\"Exists\": processExists,\n\t\t\"URL\": processURL,\n\t\t\"SaveText\": processSaveText,\n\t\t\"GetText\": processGetText,\n\t\t\"Zip\": processZip,\n\t})\n}\n\n// processSave saves a file from base64 data URI\n// Args:\n// - uploaderID: string - the uploader/manager ID\n// - content: string - base64 data URI (e.g., \"data:image/png;base64,xxxx\") or plain base64\n// - filename: string (optional) - original filename\n// - option: map (optional) - upload options (groups, gzip, compress_image, public, share)\n//\n// Returns: *File - uploaded file info\n//\n// Example:\n//\n//\tProcess(\"attachment.Save\", \"default\", \"data:image/png;base64,iVBORw0KGgo...\", \"photo.png\")\n//\tProcess(\"attachment.Save\", \"default\", \"data:text/plain;base64,SGVsbG8=\", \"hello.txt\", {\"share\": \"team\"})\nfunc processSave(p *process.Process) interface{} {\n\tp.ValidateArgNums(2)\n\n\tuploaderID := p.ArgsString(0)\n\tcontent := p.ArgsString(1)\n\n\t// Get manager\n\tmanager, exists := Managers[uploaderID]\n\tif !exists {\n\t\treturn fmt.Errorf(\"uploader not found: %s\", uploaderID)\n\t}\n\n\t// Parse data URI and decode content\n\tcontentType, data, err := parseDataURI(content)\n\tif err != nil {\n\t\treturn fmt.Errorf(\"failed to parse content: %v\", err)\n\t}\n\n\t// Get filename from args or generate from content type\n\tfilename := \"\"\n\tif p.NumOfArgs() > 2 {\n\t\tfilename = p.ArgsString(2)\n\t}\n\tif filename == \"\" {\n\t\tfilename = generateFilename(contentType)\n\t}\n\n\t// Create file header\n\theader := createFileHeader(filename, contentType, int64(len(data)))\n\n\t// Create upload options\n\toption := createUploadOption(p, filename)\n\n\t// Upload\n\tctx := context.Background()\n\tfile, err := manager.Upload(ctx, header, strings.NewReader(string(data)), option)\n\tif err != nil {\n\t\treturn fmt.Errorf(\"failed to save file: %v\", err)\n\t}\n\n\treturn file\n}\n\n// processRead reads file content as base64 data URI\n// Args:\n// - uploaderID: string - the uploader/manager ID\n// - fileID: string - the file ID\n//\n// Returns: string - base64 data URI (e.g., \"data:image/png;base64,xxxx\")\n//\n// Example:\n//\n//\tconst dataURI = Process(\"attachment.Read\", \"default\", \"abc123\")\nfunc processRead(p *process.Process) interface{} {\n\tp.ValidateArgNums(2)\n\n\tuploaderID := p.ArgsString(0)\n\tfileID := p.ArgsString(1)\n\n\tmanager, exists := Managers[uploaderID]\n\tif !exists {\n\t\treturn fmt.Errorf(\"uploader not found: %s\", uploaderID)\n\t}\n\n\tctx := context.Background()\n\n\t// Get file info for content type and permission check\n\tfileInfo, err := manager.Info(ctx, fileID)\n\tif err != nil {\n\t\treturn fmt.Errorf(\"file not found: %v\", err)\n\t}\n\n\t// Check permission\n\tif err := checkFilePermission(p, fileInfo, true); err != nil {\n\t\treturn err\n\t}\n\n\t// Read content as base64\n\tbase64Data, err := manager.ReadBase64(ctx, fileID)\n\tif err != nil {\n\t\treturn fmt.Errorf(\"failed to read file: %v\", err)\n\t}\n\n\t// Return as data URI\n\treturn fmt.Sprintf(\"data:%s;base64,%s\", fileInfo.ContentType, base64Data)\n}\n\n// processInfo gets file information\n// Args:\n// - uploaderID: string - the uploader/manager ID\n// - fileID: string - the file ID\n//\n// Returns: *File - file info\nfunc processInfo(p *process.Process) interface{} {\n\tp.ValidateArgNums(2)\n\n\tuploaderID := p.ArgsString(0)\n\tfileID := p.ArgsString(1)\n\n\tmanager, exists := Managers[uploaderID]\n\tif !exists {\n\t\treturn fmt.Errorf(\"uploader not found: %s\", uploaderID)\n\t}\n\n\tctx := context.Background()\n\tfileInfo, err := manager.Info(ctx, fileID)\n\tif err != nil {\n\t\treturn fmt.Errorf(\"file not found: %v\", err)\n\t}\n\n\t// Check permission\n\tif err := checkFilePermission(p, fileInfo, true); err != nil {\n\t\treturn err\n\t}\n\n\treturn fileInfo\n}\n\n// processList lists files with pagination and filtering\n// Args:\n// - uploaderID: string - the uploader/manager ID\n// - option: map (optional) - list options (page, page_size, filters, order_by, select)\n//\n// Returns: *ListResult - paginated file list\nfunc processList(p *process.Process) interface{} {\n\tp.ValidateArgNums(1)\n\n\tuploaderID := p.ArgsString(0)\n\n\tmanager, exists := Managers[uploaderID]\n\tif !exists {\n\t\treturn fmt.Errorf(\"uploader not found: %s\", uploaderID)\n\t}\n\n\t// Parse list options\n\tlistOption := ListOption{\n\t\tPage: 1,\n\t\tPageSize: 20,\n\t}\n\n\tif p.NumOfArgs() > 1 {\n\t\toptionRaw := p.ArgsMap(1)\n\t\toption := maps.MapOf(optionRaw).Dot()\n\n\t\tif page := any.Of(option.Get(\"page\")).CInt(); page > 0 {\n\t\t\tlistOption.Page = page\n\t\t}\n\t\tif pageSize := any.Of(option.Get(\"page_size\")).CInt(); pageSize > 0 && pageSize <= 100 {\n\t\t\tlistOption.PageSize = pageSize\n\t\t}\n\t\tif filters, ok := option.Get(\"filters\").(map[string]interface{}); ok {\n\t\t\tlistOption.Filters = filters\n\t\t}\n\t\tif orderBy, ok := option.Get(\"order_by\").(string); ok {\n\t\t\tlistOption.OrderBy = orderBy\n\t\t}\n\t\tif selectFields, ok := option.Get(\"select\").([]interface{}); ok {\n\t\t\tfor _, field := range selectFields {\n\t\t\t\tif f, ok := field.(string); ok {\n\t\t\t\t\tlistOption.Select = append(listOption.Select, f)\n\t\t\t\t}\n\t\t\t}\n\t\t}\n\t}\n\n\t// Always filter by uploader\n\tif listOption.Filters == nil {\n\t\tlistOption.Filters = make(map[string]interface{})\n\t}\n\tlistOption.Filters[\"uploader\"] = uploaderID\n\n\t// Add permission-based filtering\n\tlistOption.Wheres = append(listOption.Wheres, model.QueryWhere{\n\t\tColumn: \"uploader\",\n\t\tValue: uploaderID,\n\t})\n\tlistOption.Wheres = append(listOption.Wheres, buildPermissionWheres(p)...)\n\n\tctx := context.Background()\n\tresult, err := manager.List(ctx, listOption)\n\tif err != nil {\n\t\treturn fmt.Errorf(\"failed to list files: %v\", err)\n\t}\n\n\treturn result\n}\n\n// processDelete deletes a file\n// Args:\n// - uploaderID: string - the uploader/manager ID\n// - fileID: string - the file ID\n//\n// Returns: bool - success\nfunc processDelete(p *process.Process) interface{} {\n\tp.ValidateArgNums(2)\n\n\tuploaderID := p.ArgsString(0)\n\tfileID := p.ArgsString(1)\n\n\tmanager, exists := Managers[uploaderID]\n\tif !exists {\n\t\treturn fmt.Errorf(\"uploader not found: %s\", uploaderID)\n\t}\n\n\tctx := context.Background()\n\n\t// Get file info first\n\tfileInfo, err := manager.Info(ctx, fileID)\n\tif err != nil {\n\t\treturn fmt.Errorf(\"file not found: %v\", err)\n\t}\n\n\t// Check write permission\n\tif err := checkFilePermission(p, fileInfo, false); err != nil {\n\t\treturn err\n\t}\n\n\t// Delete file\n\tif err := manager.Delete(ctx, fileID); err != nil {\n\t\treturn fmt.Errorf(\"failed to delete file: %v\", err)\n\t}\n\n\treturn true\n}\n\n// processExists checks if file exists\n// Args:\n// - uploaderID: string - the uploader/manager ID\n// - fileID: string - the file ID\n//\n// Returns: bool\nfunc processExists(p *process.Process) interface{} {\n\tp.ValidateArgNums(2)\n\n\tuploaderID := p.ArgsString(0)\n\tfileID := p.ArgsString(1)\n\n\tmanager, exists := Managers[uploaderID]\n\tif !exists {\n\t\treturn fmt.Errorf(\"uploader not found: %s\", uploaderID)\n\t}\n\n\tctx := context.Background()\n\treturn manager.Exists(ctx, fileID)\n}\n\n// processURL gets file URL\n// Args:\n// - uploaderID: string - the uploader/manager ID\n// - fileID: string - the file ID\n//\n// Returns: string - file URL\nfunc processURL(p *process.Process) interface{} {\n\tp.ValidateArgNums(2)\n\n\tuploaderID := p.ArgsString(0)\n\tfileID := p.ArgsString(1)\n\n\tmanager, exists := Managers[uploaderID]\n\tif !exists {\n\t\treturn fmt.Errorf(\"uploader not found: %s\", uploaderID)\n\t}\n\n\tctx := context.Background()\n\n\t// Get file info for permission check\n\tfileInfo, err := manager.Info(ctx, fileID)\n\tif err != nil {\n\t\treturn fmt.Errorf(\"file not found: %v\", err)\n\t}\n\n\t// Check permission\n\tif err := checkFilePermission(p, fileInfo, true); err != nil {\n\t\treturn err\n\t}\n\n\treturn manager.storage.URL(ctx, fileID)\n}\n\n// processSaveText saves parsed text content for a file\n// Args:\n// - uploaderID: string - the uploader/manager ID\n// - fileID: string - the file ID\n// - text: string - the text content to save\n//\n// Returns: bool - success\nfunc processSaveText(p *process.Process) interface{} {\n\tp.ValidateArgNums(3)\n\n\tuploaderID := p.ArgsString(0)\n\tfileID := p.ArgsString(1)\n\ttext := p.ArgsString(2)\n\n\tmanager, exists := Managers[uploaderID]\n\tif !exists {\n\t\treturn fmt.Errorf(\"uploader not found: %s\", uploaderID)\n\t}\n\n\tctx := context.Background()\n\n\t// Get file info first to check write permission\n\tfileInfo, err := manager.Info(ctx, fileID)\n\tif err != nil {\n\t\treturn fmt.Errorf(\"file not found: %v\", err)\n\t}\n\n\t// Check write permission\n\tif err := checkFilePermission(p, fileInfo, false); err != nil {\n\t\treturn err\n\t}\n\n\tif err := manager.SaveText(ctx, fileID, text); err != nil {\n\t\treturn fmt.Errorf(\"failed to save text: %v\", err)\n\t}\n\n\treturn true\n}\n\n// processGetText gets parsed text content for a file\n// Args:\n// - uploaderID: string - the uploader/manager ID\n// - fileID: string - the file ID\n// - fullContent: bool (optional) - whether to get full content (default: false, returns preview)\n//\n// Returns: string - text content\nfunc processGetText(p *process.Process) interface{} {\n\tp.ValidateArgNums(2)\n\n\tuploaderID := p.ArgsString(0)\n\tfileID := p.ArgsString(1)\n\n\tfullContent := false\n\tif p.NumOfArgs() > 2 {\n\t\tfullContent = p.ArgsBool(2)\n\t}\n\n\tmanager, exists := Managers[uploaderID]\n\tif !exists {\n\t\treturn fmt.Errorf(\"uploader not found: %s\", uploaderID)\n\t}\n\n\tctx := context.Background()\n\n\t// Get file info for permission check\n\tfileInfo, err := manager.Info(ctx, fileID)\n\tif err != nil {\n\t\treturn fmt.Errorf(\"file not found: %v\", err)\n\t}\n\n\t// Check permission\n\tif err := checkFilePermission(p, fileInfo, true); err != nil {\n\t\treturn err\n\t}\n\n\ttext, err := manager.GetText(ctx, fileID, fullContent)\n\tif err != nil {\n\t\treturn fmt.Errorf(\"failed to get text: %v\", err)\n\t}\n\n\treturn text\n}\n\n// ============ Helper Functions ============\n\n// parseDataURI parses content as either:\n// 1. Data URI format: data:image/png;base64,xxxxx (decoded from base64)\n// 2. Plain text: stored as-is with text/plain content type\n//\n// Returns content type, data bytes, and error\nfunc parseDataURI(content string) (string, []byte, error) {\n\t// Handle data URI format: data:image/png;base64,xxxxx\n\tif strings.HasPrefix(content, \"data:\") {\n\t\t// Split by comma to get the data part\n\t\tparts := strings.SplitN(content, \",\", 2)\n\t\tif len(parts) != 2 {\n\t\t\treturn \"\", nil, fmt.Errorf(\"invalid data URI format\")\n\t\t}\n\n\t\t// Parse the header: data:image/png;base64\n\t\theader := parts[0]\n\t\tbase64Content := parts[1]\n\n\t\t// Extract content type from header\n\t\tcontentType := \"application/octet-stream\"\n\t\theader = strings.TrimPrefix(header, \"data:\")\n\t\theaderParts := strings.Split(header, \";\")\n\t\tif len(headerParts) > 0 && headerParts[0] != \"\" {\n\t\t\tcontentType = headerParts[0]\n\t\t}\n\n\t\t// Decode base64\n\t\tdata, err := base64.StdEncoding.DecodeString(base64Content)\n\t\tif err != nil {\n\t\t\treturn \"\", nil, fmt.Errorf(\"failed to decode base64: %v\", err)\n\t\t}\n\n\t\treturn contentType, data, nil\n\t}\n\n\t// Plain text content - store as-is\n\treturn \"text/plain\", []byte(content), nil\n}\n\n// generateFilename generates a filename based on content type\nfunc generateFilename(contentType string) string {\n\t// Get extension from content type\n\texts, err := mime.ExtensionsByType(contentType)\n\tif err == nil && len(exts) > 0 {\n\t\treturn \"file\" + exts[0]\n\t}\n\n\t// Fallback for common types\n\tswitch contentType {\n\tcase \"image/png\":\n\t\treturn \"file.png\"\n\tcase \"image/jpeg\":\n\t\treturn \"file.jpg\"\n\tcase \"image/gif\":\n\t\treturn \"file.gif\"\n\tcase \"image/webp\":\n\t\treturn \"file.webp\"\n\tcase \"text/plain\":\n\t\treturn \"file.txt\"\n\tcase \"application/pdf\":\n\t\treturn \"file.pdf\"\n\tcase \"application/json\":\n\t\treturn \"file.json\"\n\tdefault:\n\t\treturn \"file.bin\"\n\t}\n}\n\n// createUploadOption creates UploadOption from process args\nfunc createUploadOption(p *process.Process, filename string) UploadOption {\n\toption := UploadOption{\n\t\tOriginalFilename: filename,\n\t}\n\n\t// Parse option from fourth argument if provided\n\tif p.NumOfArgs() > 3 {\n\t\toptionRaw := p.ArgsMap(3)\n\t\toptionMap := maps.MapOf(optionRaw).Dot()\n\n\t\t// Groups\n\t\tif groups, ok := optionMap.Get(\"groups\").([]interface{}); ok {\n\t\t\tfor _, g := range groups {\n\t\t\t\tif gs, ok := g.(string); ok {\n\t\t\t\t\toption.Groups = append(option.Groups, gs)\n\t\t\t\t}\n\t\t\t}\n\t\t} else if groupsStr, ok := optionMap.Get(\"groups\").(string); ok {\n\t\t\toption.Groups = strings.Split(groupsStr, \",\")\n\t\t\tfor i := range option.Groups {\n\t\t\t\toption.Groups[i] = strings.TrimSpace(option.Groups[i])\n\t\t\t}\n\t\t}\n\n\t\t// Gzip\n\t\tif gzip, ok := optionMap.Get(\"gzip\").(bool); ok {\n\t\t\toption.Gzip = gzip\n\t\t}\n\n\t\t// Compress image\n\t\tif compress, ok := optionMap.Get(\"compress_image\").(bool); ok {\n\t\t\toption.CompressImage = compress\n\t\t}\n\t\tif size := any.Of(optionMap.Get(\"compress_size\")).CInt(); size > 0 {\n\t\t\toption.CompressSize = size\n\t\t}\n\n\t\t// Public/Share\n\t\tif public, ok := optionMap.Get(\"public\").(bool); ok {\n\t\t\toption.Public = public\n\t\t}\n\t\tif share, ok := optionMap.Get(\"share\").(string); ok {\n\t\t\toption.Share = share\n\t\t}\n\t}\n\n\t// Set permission fields from process.Authorized\n\tif p.Authorized != nil {\n\t\toption.YaoCreatedBy = p.Authorized.UserID\n\t\toption.YaoTeamID = p.Authorized.TeamID\n\t\toption.YaoTenantID = p.Authorized.TenantID\n\t}\n\n\treturn option\n}\n\n// createFileHeader creates a FileHeader from parameters\nfunc createFileHeader(filename, contentType string, size int64) *FileHeader {\n\theader := &multipart.FileHeader{\n\t\tFilename: filename,\n\t\tSize: size,\n\t\tHeader: make(textproto.MIMEHeader),\n\t}\n\theader.Header.Set(\"Content-Type\", contentType)\n\n\t// Set extension from filename\n\tif ext := filepath.Ext(filename); ext != \"\" {\n\t\theader.Header.Set(\"Content-Extension\", ext)\n\t}\n\n\treturn &FileHeader{FileHeader: header}\n}\n\n// checkFilePermission checks if user has permission to access the file\n// readable: true for read permission, false for write permission\nfunc checkFilePermission(p *process.Process, fileInfo *File, readable bool) error {\n\tauth := p.Authorized\n\n\t// No auth info - allow access (for non-authenticated operations)\n\tif auth == nil {\n\t\treturn nil\n\t}\n\n\t// No constraints - allow access\n\tif !auth.Constraints.TeamOnly && !auth.Constraints.OwnerOnly {\n\t\treturn nil\n\t}\n\n\t// Public files are readable by everyone\n\tif readable && fileInfo.Public {\n\t\treturn nil\n\t}\n\n\t// Combined Team and Owner permission validation\n\tif auth.Constraints.TeamOnly && auth.Constraints.OwnerOnly {\n\t\tif fileInfo.YaoCreatedBy == auth.UserID && fileInfo.YaoTeamID == auth.TeamID {\n\t\t\treturn nil\n\t\t}\n\t}\n\n\t// Owner only permission validation\n\tif auth.Constraints.OwnerOnly {\n\t\tif fileInfo.YaoCreatedBy != \"\" && fileInfo.YaoCreatedBy == auth.UserID {\n\t\t\treturn nil\n\t\t}\n\t}\n\n\t// Team only permission validation\n\tif auth.Constraints.TeamOnly {\n\t\tswitch fileInfo.Share {\n\t\tcase \"team\":\n\t\t\tif fileInfo.YaoTeamID == auth.TeamID {\n\t\t\t\treturn nil\n\t\t\t}\n\t\tcase \"private\":\n\t\t\tif fileInfo.YaoCreatedBy == auth.UserID {\n\t\t\t\treturn nil\n\t\t\t}\n\t\t}\n\t}\n\n\treturn fmt.Errorf(\"forbidden: no permission to access file\")\n}\n\n// buildPermissionWheres builds where clauses for permission filtering\nfunc buildPermissionWheres(p *process.Process) []model.QueryWhere {\n\tauth := p.Authorized\n\tif auth == nil {\n\t\treturn nil\n\t}\n\n\t// No constraints - no additional filtering needed\n\tif !auth.Constraints.TeamOnly && !auth.Constraints.OwnerOnly {\n\t\treturn nil\n\t}\n\n\tvar wheres []model.QueryWhere\n\n\t// Team only - User can access:\n\t// 1. Public files (public = true)\n\t// 2. Files in their team where:\n\t// - They uploaded the file (__yao_created_by matches)\n\t// - OR the file is shared with team (share = \"team\")\n\tif auth.Constraints.TeamOnly && auth.TeamID != \"\" {\n\t\twheres = append(wheres, model.QueryWhere{\n\t\t\tWheres: []model.QueryWhere{\n\t\t\t\t{Column: \"public\", Value: true, Method: \"orwhere\"},\n\t\t\t\t{Wheres: []model.QueryWhere{\n\t\t\t\t\t{Column: \"__yao_team_id\", Value: auth.TeamID},\n\t\t\t\t\t{Wheres: []model.QueryWhere{\n\t\t\t\t\t\t{Column: \"__yao_created_by\", Value: auth.UserID},\n\t\t\t\t\t\t{Column: \"share\", Value: \"team\", Method: \"orwhere\"},\n\t\t\t\t\t}},\n\t\t\t\t}, Method: \"orwhere\"},\n\t\t\t},\n\t\t})\n\t\treturn wheres\n\t}\n\n\t// Owner only - User can access:\n\t// 1. Public files (public = true)\n\t// 2. Files they uploaded where:\n\t// - __yao_team_id is null (not team files)\n\t// - __yao_created_by matches their user ID\n\tif auth.Constraints.OwnerOnly && auth.UserID != \"\" {\n\t\twheres = append(wheres, model.QueryWhere{\n\t\t\tWheres: []model.QueryWhere{\n\t\t\t\t{Column: \"public\", Value: true, Method: \"orwhere\"},\n\t\t\t\t{Wheres: []model.QueryWhere{\n\t\t\t\t\t{Column: \"__yao_team_id\", OP: \"null\"},\n\t\t\t\t\t{Column: \"__yao_created_by\", Value: auth.UserID},\n\t\t\t\t}, Method: \"orwhere\"},\n\t\t\t},\n\t\t})\n\t\treturn wheres\n\t}\n\n\treturn wheres\n}\n\n// processZip packages multiple attachment files into a single zip archive\n// and uploads it back to the same uploader storage.\n//\n// Args:\n// - uploaderID: string - the uploader/manager ID (e.g. \"__yao.attachment\")\n// - fileIDs: []string - list of file IDs to package\n// - zipFilename: string - output zip filename (e.g. \"my-notes.zip\")\n// - option: map (optional) - upload options (groups, gzip, public, share)\n//\n// Returns: *File - uploaded zip file info (file_id, filename, bytes, etc.)\n//\n// Permission: checks read permission on each source file via p.Authorized.\n// The uploaded zip inherits permission fields (created_by, team_id, tenant_id)\n// from the process context automatically via createUploadOption.\n//\n// Example:\n//\n//\tProcess(\"attachment.Zip\", \"__yao.attachment\", [\"abc123\", \"def456\"], \"archive.zip\")\n//\tProcess(\"attachment.Zip\", \"__yao.attachment\", [\"abc123\"], \"archive.zip\", {\"public\": true})\nfunc processZip(p *process.Process) interface{} {\n\tp.ValidateArgNums(3)\n\n\tuploaderID := p.ArgsString(0)\n\tfileIDs := toStringSlice(p.Args[1])\n\tzipFilename := p.ArgsString(2)\n\n\tif len(fileIDs) == 0 {\n\t\treturn fmt.Errorf(\"attachment.Zip: fileIDs is empty\")\n\t}\n\n\tif zipFilename == \"\" {\n\t\tzipFilename = \"archive.zip\"\n\t}\n\n\t// Ensure .zip extension\n\tif !strings.HasSuffix(strings.ToLower(zipFilename), \".zip\") {\n\t\tzipFilename += \".zip\"\n\t}\n\n\t// Get manager\n\tmanager, exists := Managers[uploaderID]\n\tif !exists {\n\t\treturn fmt.Errorf(\"uploader not found: %s\", uploaderID)\n\t}\n\n\tctx := context.Background()\n\n\t// Build zip in memory\n\tbuf := new(bytes.Buffer)\n\tzipWriter := zip.NewWriter(buf)\n\tusedNames := map[string]int{} // for deduplication\n\n\tfor _, fid := range fileIDs {\n\t\t// Get file info\n\t\tfileInfo, err := manager.Info(ctx, fid)\n\t\tif err != nil {\n\t\t\treturn fmt.Errorf(\"attachment.Zip: failed to get file info for %s: %v\", fid, err)\n\t\t}\n\n\t\t// Check read permission\n\t\tif err := checkFilePermission(p, fileInfo, true); err != nil {\n\t\t\treturn fmt.Errorf(\"attachment.Zip: permission denied for file %s: %v\", fid, err)\n\t\t}\n\n\t\t// Read file content\n\t\tdata, err := manager.Read(ctx, fid)\n\t\tif err != nil {\n\t\t\treturn fmt.Errorf(\"attachment.Zip: failed to read file %s: %v\", fid, err)\n\t\t}\n\n\t\t// Deduplicate filename\n\t\tname := dedupFilename(fileInfo.Filename, usedNames)\n\n\t\t// Write to zip\n\t\tw, err := zipWriter.Create(name)\n\t\tif err != nil {\n\t\t\treturn fmt.Errorf(\"attachment.Zip: failed to create zip entry %s: %v\", name, err)\n\t\t}\n\t\tif _, err := w.Write(data); err != nil {\n\t\t\treturn fmt.Errorf(\"attachment.Zip: failed to write zip entry %s: %v\", name, err)\n\t\t}\n\t}\n\n\tif err := zipWriter.Close(); err != nil {\n\t\treturn fmt.Errorf(\"attachment.Zip: failed to close zip writer: %v\", err)\n\t}\n\n\t// Upload the zip file\n\theader := createFileHeader(zipFilename, \"application/zip\", int64(buf.Len()))\n\toption := createUploadOption(p, zipFilename)\n\tfile, err := manager.Upload(ctx, header, buf, option)\n\tif err != nil {\n\t\treturn fmt.Errorf(\"attachment.Zip: failed to upload zip: %v\", err)\n\t}\n\n\treturn file\n}\n\n// toStringSlice converts []interface{} to []string\nfunc toStringSlice(v interface{}) []string {\n\tswitch val := v.(type) {\n\tcase []string:\n\t\treturn val\n\tcase []interface{}:\n\t\tresult := make([]string, 0, len(val))\n\t\tfor _, item := range val {\n\t\t\tif s, ok := item.(string); ok {\n\t\t\t\tresult = append(result, s)\n\t\t\t}\n\t\t}\n\t\treturn result\n\tdefault:\n\t\treturn nil\n\t}\n}\n\n// dedupFilename ensures unique filenames within a zip archive.\n// If \"photo.png\" already exists, returns \"photo(1).png\", \"photo(2).png\", etc.\nfunc dedupFilename(name string, used map[string]int) string {\n\tif name == \"\" {\n\t\tname = \"file\"\n\t}\n\n\tlower := strings.ToLower(name)\n\tcount, exists := used[lower]\n\tif !exists {\n\t\tused[lower] = 1\n\t\treturn name\n\t}\n\n\t// Generate deduplicated name\n\text := filepath.Ext(name)\n\tbase := strings.TrimSuffix(name, ext)\n\tused[lower] = count + 1\n\treturn fmt.Sprintf(\"%s(%d)%s\", base, count, ext)\n}\n" }, { "path": "attachment/process_test.go", "content": "package attachment\n\nimport (\n\t\"encoding/base64\"\n\t\"fmt\"\n\t\"strings\"\n\t\"testing\"\n\t\"time\"\n\n\t\"github.com/yaoapp/gou/process\"\n\t\"github.com/yaoapp/yao/config\"\n\t\"github.com/yaoapp/yao/test\"\n)\n\nfunc TestProcessSave(t *testing.T) {\n\ttest.Prepare(t, config.Conf)\n\tdefer test.Clean()\n\n\t// Register default uploader for testing\n\tmanager, err := RegisterDefault(\"data.local\")\n\tif err != nil {\n\t\tt.Fatalf(\"Failed to register manager: %v\", err)\n\t}\n\t_ = manager\n\n\t// Test 1: Save with data URI format\n\tt.Run(\"SaveWithDataURI\", func(t *testing.T) {\n\t\tcontent := \"Hello, World!\"\n\t\tbase64Content := base64.StdEncoding.EncodeToString([]byte(content))\n\t\tdataURI := fmt.Sprintf(\"data:text/plain;base64,%s\", base64Content)\n\n\t\tp := process.New(\"attachment.Save\", \"data.local\", dataURI, \"hello.txt\")\n\t\tresult := processSave(p)\n\n\t\tif err, ok := result.(error); ok {\n\t\t\tt.Fatalf(\"Failed to save file: %v\", err)\n\t\t}\n\n\t\tfile, ok := result.(*File)\n\t\tif !ok {\n\t\t\tt.Fatalf(\"Expected *File, got %T\", result)\n\t\t}\n\n\t\tif file.ID == \"\" {\n\t\t\tt.Error(\"File ID should not be empty\")\n\t\t}\n\n\t\tif file.Filename != \"hello.txt\" {\n\t\t\tt.Errorf(\"Expected filename 'hello.txt', got '%s'\", file.Filename)\n\t\t}\n\n\t\tif !strings.HasPrefix(file.ContentType, \"text/plain\") {\n\t\t\tt.Errorf(\"Expected content type 'text/plain', got '%s'\", file.ContentType)\n\t\t}\n\n\t\tt.Logf(\"Saved file - ID: %s, Filename: %s, ContentType: %s\", file.ID, file.Filename, file.ContentType)\n\t})\n\n\t// Test 2: Save with plain base64 (no data URI header) - use text/plain to pass allowed types\n\tt.Run(\"SaveWithPlainBase64\", func(t *testing.T) {\n\t\tcontent := \"Plain base64 content\"\n\t\tbase64Content := base64.StdEncoding.EncodeToString([]byte(content))\n\t\t// Without data URI header, we need to provide a filename with allowed extension\n\t\t// or use data URI format. Let's test with text file extension.\n\t\tdataURI := fmt.Sprintf(\"data:text/plain;base64,%s\", base64Content)\n\n\t\tp := process.New(\"attachment.Save\", \"data.local\", dataURI, \"plain.txt\")\n\t\tresult := processSave(p)\n\n\t\tif err, ok := result.(error); ok {\n\t\t\tt.Fatalf(\"Failed to save file: %v\", err)\n\t\t}\n\n\t\tfile, ok := result.(*File)\n\t\tif !ok {\n\t\t\tt.Fatalf(\"Expected *File, got %T\", result)\n\t\t}\n\n\t\tif file.ID == \"\" {\n\t\t\tt.Error(\"File ID should not be empty\")\n\t\t}\n\n\t\t// With data URI, content type should be text/plain\n\t\tif !strings.HasPrefix(file.ContentType, \"text/plain\") {\n\t\t\tt.Errorf(\"Expected content type 'text/plain', got '%s'\", file.ContentType)\n\t\t}\n\t})\n\n\t// Test 3: Save image with data URI\n\tt.Run(\"SaveImageDataURI\", func(t *testing.T) {\n\t\t// Minimal valid PNG (1x1 pixel transparent PNG)\n\t\tpngBase64 := \"iVBORw0KGgoAAAANSUhEUgAAAAEAAAABCAYAAAAfFcSJAAAADUlEQVR42mNk+M9QDwADhgGAWjR9awAAAABJRU5ErkJggg==\"\n\t\tdataURI := fmt.Sprintf(\"data:image/png;base64,%s\", pngBase64)\n\n\t\tp := process.New(\"attachment.Save\", \"data.local\", dataURI, \"pixel.png\")\n\t\tresult := processSave(p)\n\n\t\tif err, ok := result.(error); ok {\n\t\t\tt.Fatalf(\"Failed to save image: %v\", err)\n\t\t}\n\n\t\tfile, ok := result.(*File)\n\t\tif !ok {\n\t\t\tt.Fatalf(\"Expected *File, got %T\", result)\n\t\t}\n\n\t\tif file.ContentType != \"image/png\" {\n\t\t\tt.Errorf(\"Expected content type 'image/png', got '%s'\", file.ContentType)\n\t\t}\n\t})\n\n\t// Test 4: Save with options - verify via Info since File struct may not have all fields\n\tt.Run(\"SaveWithOptions\", func(t *testing.T) {\n\t\tcontent := \"Content with options\"\n\t\tbase64Content := base64.StdEncoding.EncodeToString([]byte(content))\n\t\tdataURI := fmt.Sprintf(\"data:text/plain;base64,%s\", base64Content)\n\n\t\toptions := map[string]interface{}{\n\t\t\t\"groups\": []interface{}{\"test\", \"unit\"},\n\t\t\t\"public\": true,\n\t\t\t\"share\": \"team\",\n\t\t}\n\n\t\tp := process.New(\"attachment.Save\", \"data.local\", dataURI, \"options.txt\", options)\n\t\tresult := processSave(p)\n\n\t\tif err, ok := result.(error); ok {\n\t\t\tt.Fatalf(\"Failed to save file with options: %v\", err)\n\t\t}\n\n\t\tfile, ok := result.(*File)\n\t\tif !ok {\n\t\t\tt.Fatalf(\"Expected *File, got %T\", result)\n\t\t}\n\n\t\t// File should be saved successfully\n\t\tif file.ID == \"\" {\n\t\t\tt.Error(\"File ID should not be empty\")\n\t\t}\n\n\t\t// Get info to verify public and share fields\n\t\tinfoP := process.New(\"attachment.Info\", \"data.local\", file.ID)\n\t\tinfoResult := processInfo(infoP)\n\t\tinfo, ok := infoResult.(*File)\n\t\tif !ok {\n\t\t\tt.Fatalf(\"Failed to get file info: %v\", infoResult)\n\t\t}\n\n\t\tif !info.Public {\n\t\t\tt.Error(\"Expected file to be public\")\n\t\t}\n\n\t\tif info.Share != \"team\" {\n\t\t\tt.Errorf(\"Expected share 'team', got '%s'\", info.Share)\n\t\t}\n\n\t\tt.Logf(\"Saved file with options - ID: %s, Public: %v, Share: %s\", file.ID, info.Public, info.Share)\n\t})\n\n\t// Test 5: Save without filename (auto-generate)\n\tt.Run(\"SaveWithoutFilename\", func(t *testing.T) {\n\t\tcontent := \"Auto filename content\"\n\t\tbase64Content := base64.StdEncoding.EncodeToString([]byte(content))\n\t\tdataURI := fmt.Sprintf(\"data:application/json;base64,%s\", base64Content)\n\n\t\tp := process.New(\"attachment.Save\", \"data.local\", dataURI)\n\t\tresult := processSave(p)\n\n\t\tif err, ok := result.(error); ok {\n\t\t\tt.Fatalf(\"Failed to save file: %v\", err)\n\t\t}\n\n\t\tfile, ok := result.(*File)\n\t\tif !ok {\n\t\t\tt.Fatalf(\"Expected *File, got %T\", result)\n\t\t}\n\n\t\t// Should auto-generate a filename\n\t\tif file.Filename == \"\" {\n\t\t\tt.Error(\"Filename should not be empty\")\n\t\t}\n\n\t\tt.Logf(\"Auto-generated filename: %s\", file.Filename)\n\t})\n\n\t// Test 6: Save with invalid uploader\n\tt.Run(\"SaveWithInvalidUploader\", func(t *testing.T) {\n\t\tcontent := \"Test content\"\n\t\tbase64Content := base64.StdEncoding.EncodeToString([]byte(content))\n\t\tdataURI := fmt.Sprintf(\"data:text/plain;base64,%s\", base64Content)\n\n\t\tp := process.New(\"attachment.Save\", \"non-existent-uploader\", dataURI, \"test.txt\")\n\t\tresult := processSave(p)\n\n\t\terr, ok := result.(error)\n\t\tif !ok {\n\t\t\tt.Fatal(\"Expected error for non-existent uploader\")\n\t\t}\n\n\t\tif !strings.Contains(err.Error(), \"uploader not found\") {\n\t\t\tt.Errorf(\"Expected 'uploader not found' error, got: %s\", err.Error())\n\t\t}\n\t})\n\n\t// Test 7: Save with invalid base64\n\tt.Run(\"SaveWithInvalidBase64\", func(t *testing.T) {\n\t\tinvalidDataURI := \"data:text/plain;base64,not-valid-base64!!!\"\n\n\t\tp := process.New(\"attachment.Save\", \"data.local\", invalidDataURI, \"invalid.txt\")\n\t\tresult := processSave(p)\n\n\t\t_, ok := result.(error)\n\t\tif !ok {\n\t\t\tt.Fatal(\"Expected error for invalid base64\")\n\t\t}\n\t})\n\n\t// Test 8: Save plain text directly (no data URI)\n\tt.Run(\"SavePlainText\", func(t *testing.T) {\n\t\tcontent := \"This is plain text content without data URI encoding.\"\n\n\t\tp := process.New(\"attachment.Save\", \"data.local\", content, \"plain-text.txt\")\n\t\tresult := processSave(p)\n\n\t\tif err, ok := result.(error); ok {\n\t\t\tt.Fatalf(\"Failed to save plain text: %v\", err)\n\t\t}\n\n\t\tfile, ok := result.(*File)\n\t\tif !ok {\n\t\t\tt.Fatalf(\"Expected *File, got %T\", result)\n\t\t}\n\n\t\tif file.ID == \"\" {\n\t\t\tt.Error(\"File ID should not be empty\")\n\t\t}\n\n\t\t// Content type should be text/plain for plain text\n\t\tif !strings.HasPrefix(file.ContentType, \"text/plain\") {\n\t\t\tt.Errorf(\"Expected content type 'text/plain', got '%s'\", file.ContentType)\n\t\t}\n\n\t\t// Read back and verify content\n\t\treadP := process.New(\"attachment.Read\", \"data.local\", file.ID)\n\t\treadResult := processRead(readP)\n\n\t\tdataURI, ok := readResult.(string)\n\t\tif !ok {\n\t\t\tt.Fatalf(\"Expected string, got %T: %v\", readResult, readResult)\n\t\t}\n\n\t\t// Decode from data URI\n\t\tparts := strings.SplitN(dataURI, \",\", 2)\n\t\tif len(parts) != 2 {\n\t\t\tt.Fatalf(\"Invalid data URI format\")\n\t\t}\n\t\tdecoded, err := base64.StdEncoding.DecodeString(parts[1])\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to decode: %v\", err)\n\t\t}\n\n\t\tif string(decoded) != content {\n\t\t\tt.Errorf(\"Content mismatch: expected %q, got %q\", content, string(decoded))\n\t\t}\n\t})\n\n\t// Test 9: Save Chinese text directly (UTF-8)\n\tt.Run(\"SaveChineseText\", func(t *testing.T) {\n\t\tcontent := \"这是一段中文内容，测试UTF-8编码。\\n第二行内容。\"\n\n\t\tp := process.New(\"attachment.Save\", \"data.local\", content, \"chinese.txt\")\n\t\tresult := processSave(p)\n\n\t\tif err, ok := result.(error); ok {\n\t\t\tt.Fatalf(\"Failed to save Chinese text: %v\", err)\n\t\t}\n\n\t\tfile, ok := result.(*File)\n\t\tif !ok {\n\t\t\tt.Fatalf(\"Expected *File, got %T\", result)\n\t\t}\n\n\t\t// Read back and verify content\n\t\treadP := process.New(\"attachment.Read\", \"data.local\", file.ID)\n\t\treadResult := processRead(readP)\n\n\t\tdataURI, ok := readResult.(string)\n\t\tif !ok {\n\t\t\tt.Fatalf(\"Expected string, got %T: %v\", readResult, readResult)\n\t\t}\n\n\t\t// Decode from data URI\n\t\tparts := strings.SplitN(dataURI, \",\", 2)\n\t\tif len(parts) != 2 {\n\t\t\tt.Fatalf(\"Invalid data URI format\")\n\t\t}\n\t\tdecoded, err := base64.StdEncoding.DecodeString(parts[1])\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to decode: %v\", err)\n\t\t}\n\n\t\tif string(decoded) != content {\n\t\t\tt.Errorf(\"Chinese content mismatch: expected %q, got %q\", content, string(decoded))\n\t\t}\n\t})\n}\n\nfunc TestProcessRead(t *testing.T) {\n\ttest.Prepare(t, config.Conf)\n\tdefer test.Clean()\n\n\t// Register default uploader for testing\n\t_, err := RegisterDefault(\"data.local\")\n\tif err != nil {\n\t\tt.Fatalf(\"Failed to register manager: %v\", err)\n\t}\n\n\t// First, save a file to read\n\tcontent := \"Content to read\"\n\tbase64Content := base64.StdEncoding.EncodeToString([]byte(content))\n\tdataURI := fmt.Sprintf(\"data:text/plain;base64,%s\", base64Content)\n\n\tsaveP := process.New(\"attachment.Save\", \"data.local\", dataURI, \"read-test.txt\")\n\tsaveResult := processSave(saveP)\n\tfile, ok := saveResult.(*File)\n\tif !ok {\n\t\tt.Fatalf(\"Failed to save test file: %v\", saveResult)\n\t}\n\n\t// Test 1: Read file as data URI\n\tt.Run(\"ReadAsDataURI\", func(t *testing.T) {\n\t\tp := process.New(\"attachment.Read\", \"data.local\", file.ID)\n\t\tresult := processRead(p)\n\n\t\tif err, ok := result.(error); ok {\n\t\t\tt.Fatalf(\"Failed to read file: %v\", err)\n\t\t}\n\n\t\tresultDataURI, ok := result.(string)\n\t\tif !ok {\n\t\t\tt.Fatalf(\"Expected string, got %T\", result)\n\t\t}\n\n\t\t// Should return data URI format\n\t\tif !strings.HasPrefix(resultDataURI, \"data:text/plain\") {\n\t\t\tt.Errorf(\"Expected data URI starting with 'data:text/plain', got: %s\", resultDataURI[:50])\n\t\t}\n\n\t\tif !strings.Contains(resultDataURI, \";base64,\") {\n\t\t\tt.Error(\"Expected data URI to contain ';base64,'\")\n\t\t}\n\n\t\t// Decode and verify content\n\t\tparts := strings.SplitN(resultDataURI, \",\", 2)\n\t\tif len(parts) != 2 {\n\t\t\tt.Fatal(\"Invalid data URI format\")\n\t\t}\n\n\t\tdecodedContent, err := base64.StdEncoding.DecodeString(parts[1])\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to decode base64: %v\", err)\n\t\t}\n\n\t\tif string(decodedContent) != content {\n\t\t\tt.Errorf(\"Content mismatch. Expected: %s, Got: %s\", content, string(decodedContent))\n\t\t}\n\n\t\tt.Logf(\"Read file successfully - Data URI length: %d\", len(resultDataURI))\n\t})\n\n\t// Test 2: Read non-existent file\n\tt.Run(\"ReadNonExistent\", func(t *testing.T) {\n\t\tp := process.New(\"attachment.Read\", \"data.local\", \"non-existent-file-id\")\n\t\tresult := processRead(p)\n\n\t\terr, ok := result.(error)\n\t\tif !ok {\n\t\t\tt.Fatal(\"Expected error for non-existent file\")\n\t\t}\n\n\t\tif !strings.Contains(err.Error(), \"file not found\") {\n\t\t\tt.Errorf(\"Expected 'file not found' error, got: %s\", err.Error())\n\t\t}\n\t})\n\n\t// Test 3: Read with invalid uploader\n\tt.Run(\"ReadWithInvalidUploader\", func(t *testing.T) {\n\t\tp := process.New(\"attachment.Read\", \"non-existent-uploader\", file.ID)\n\t\tresult := processRead(p)\n\n\t\terr, ok := result.(error)\n\t\tif !ok {\n\t\t\tt.Fatal(\"Expected error for non-existent uploader\")\n\t\t}\n\n\t\tif !strings.Contains(err.Error(), \"uploader not found\") {\n\t\t\tt.Errorf(\"Expected 'uploader not found' error, got: %s\", err.Error())\n\t\t}\n\t})\n}\n\nfunc TestProcessInfo(t *testing.T) {\n\ttest.Prepare(t, config.Conf)\n\tdefer test.Clean()\n\n\t// Register default uploader for testing\n\t_, err := RegisterDefault(\"data.local\")\n\tif err != nil {\n\t\tt.Fatalf(\"Failed to register manager: %v\", err)\n\t}\n\n\t// Save a file with options\n\tcontent := \"Info test content\"\n\tbase64Content := base64.StdEncoding.EncodeToString([]byte(content))\n\tdataURI := fmt.Sprintf(\"data:text/plain;base64,%s\", base64Content)\n\n\toptions := map[string]interface{}{\n\t\t\"groups\": []interface{}{\"info\", \"test\"},\n\t\t\"public\": true,\n\t\t\"share\": \"team\",\n\t}\n\n\tsaveP := process.New(\"attachment.Save\", \"data.local\", dataURI, \"info-test.txt\", options)\n\tsaveResult := processSave(saveP)\n\tfile, ok := saveResult.(*File)\n\tif !ok {\n\t\tt.Fatalf(\"Failed to save test file: %v\", saveResult)\n\t}\n\n\t// Test 1: Get file info\n\tt.Run(\"GetFileInfo\", func(t *testing.T) {\n\t\tp := process.New(\"attachment.Info\", \"data.local\", file.ID)\n\t\tresult := processInfo(p)\n\n\t\tif err, ok := result.(error); ok {\n\t\t\tt.Fatalf(\"Failed to get file info: %v\", err)\n\t\t}\n\n\t\tinfo, ok := result.(*File)\n\t\tif !ok {\n\t\t\tt.Fatalf(\"Expected *File, got %T\", result)\n\t\t}\n\n\t\tif info.ID != file.ID {\n\t\t\tt.Errorf(\"Expected ID %s, got %s\", file.ID, info.ID)\n\t\t}\n\n\t\tif info.Filename != file.Filename {\n\t\t\tt.Errorf(\"Expected filename %s, got %s\", file.Filename, info.Filename)\n\t\t}\n\n\t\tif !strings.HasPrefix(info.ContentType, \"text/plain\") {\n\t\t\tt.Errorf(\"Expected content type 'text/plain', got %s\", info.ContentType)\n\t\t}\n\n\t\tif !info.Public {\n\t\t\tt.Error(\"Expected file to be public\")\n\t\t}\n\n\t\tif info.Share != \"team\" {\n\t\t\tt.Errorf(\"Expected share 'team', got %s\", info.Share)\n\t\t}\n\n\t\tt.Logf(\"File info - ID: %s, Filename: %s, Bytes: %d, Public: %v, Share: %s\",\n\t\t\tinfo.ID, info.Filename, info.Bytes, info.Public, info.Share)\n\t})\n\n\t// Test 2: Get info for non-existent file\n\tt.Run(\"GetInfoNonExistent\", func(t *testing.T) {\n\t\tp := process.New(\"attachment.Info\", \"data.local\", \"non-existent-file-id\")\n\t\tresult := processInfo(p)\n\n\t\terr, ok := result.(error)\n\t\tif !ok {\n\t\t\tt.Fatal(\"Expected error for non-existent file\")\n\t\t}\n\n\t\tif !strings.Contains(err.Error(), \"file not found\") {\n\t\t\tt.Errorf(\"Expected 'file not found' error, got: %s\", err.Error())\n\t\t}\n\t})\n}\n\nfunc TestProcessList(t *testing.T) {\n\ttest.Prepare(t, config.Conf)\n\tdefer test.Clean()\n\n\t// Use unique manager name for test isolation\n\tmanagerName := fmt.Sprintf(\"data.local.list.%d\", time.Now().UnixNano())\n\t_, err := RegisterDefault(managerName)\n\tif err != nil {\n\t\tt.Fatalf(\"Failed to register manager: %v\", err)\n\t}\n\n\t// Upload multiple test files\n\ttestFiles := []struct {\n\t\tcontent string\n\t\tfilename string\n\t\tcontentType string\n\t}{\n\t\t{\"File 1 content\", \"file1.txt\", \"text/plain\"},\n\t\t{\"File 2 content\", \"file2.txt\", \"text/plain\"},\n\t\t{\"File 3 content\", \"file3.txt\", \"text/plain\"},\n\t\t{`{\"key\": \"value\"}`, \"data.json\", \"application/json\"},\n\t\t{\"CSV,Data\\n1,2\", \"data.csv\", \"text/csv\"},\n\t}\n\n\tuploadedIDs := make([]string, 0, len(testFiles))\n\tfor _, tf := range testFiles {\n\t\tbase64Content := base64.StdEncoding.EncodeToString([]byte(tf.content))\n\t\tdataURI := fmt.Sprintf(\"data:%s;base64,%s\", tf.contentType, base64Content)\n\n\t\tp := process.New(\"attachment.Save\", managerName, dataURI, tf.filename)\n\t\tresult := processSave(p)\n\n\t\tfile, ok := result.(*File)\n\t\tif !ok {\n\t\t\tt.Fatalf(\"Failed to save file %s: %v\", tf.filename, result)\n\t\t}\n\t\tuploadedIDs = append(uploadedIDs, file.ID)\n\t}\n\n\t// Test 1: Basic list\n\tt.Run(\"BasicList\", func(t *testing.T) {\n\t\tp := process.New(\"attachment.List\", managerName)\n\t\tresult := processList(p)\n\n\t\tif err, ok := result.(error); ok {\n\t\t\tt.Fatalf(\"Failed to list files: %v\", err)\n\t\t}\n\n\t\tlistResult, ok := result.(*ListResult)\n\t\tif !ok {\n\t\t\tt.Fatalf(\"Expected *ListResult, got %T\", result)\n\t\t}\n\n\t\tif len(listResult.Files) != len(testFiles) {\n\t\t\tt.Errorf(\"Expected %d files, got %d\", len(testFiles), len(listResult.Files))\n\t\t}\n\n\t\tif listResult.Total != int64(len(testFiles)) {\n\t\t\tt.Errorf(\"Expected total %d, got %d\", len(testFiles), listResult.Total)\n\t\t}\n\n\t\tt.Logf(\"List result - Total: %d, Page: %d, PageSize: %d\", listResult.Total, listResult.Page, listResult.PageSize)\n\t})\n\n\t// Test 2: List with pagination\n\tt.Run(\"ListWithPagination\", func(t *testing.T) {\n\t\toptions := map[string]interface{}{\n\t\t\t\"page\": 1,\n\t\t\t\"page_size\": 2,\n\t\t}\n\n\t\tp := process.New(\"attachment.List\", managerName, options)\n\t\tresult := processList(p)\n\n\t\tif err, ok := result.(error); ok {\n\t\t\tt.Fatalf(\"Failed to list files with pagination: %v\", err)\n\t\t}\n\n\t\tlistResult, ok := result.(*ListResult)\n\t\tif !ok {\n\t\t\tt.Fatalf(\"Expected *ListResult, got %T\", result)\n\t\t}\n\n\t\tif len(listResult.Files) != 2 {\n\t\t\tt.Errorf(\"Expected 2 files, got %d\", len(listResult.Files))\n\t\t}\n\n\t\tif listResult.PageSize != 2 {\n\t\t\tt.Errorf(\"Expected page size 2, got %d\", listResult.PageSize)\n\t\t}\n\n\t\tif listResult.TotalPages != 3 { // 5 files / 2 per page = 3 pages\n\t\t\tt.Errorf(\"Expected 3 total pages, got %d\", listResult.TotalPages)\n\t\t}\n\t})\n\n\t// Test 3: List with filters - use content_type wildcard\n\tt.Run(\"ListWithFilters\", func(t *testing.T) {\n\t\toptions := map[string]interface{}{\n\t\t\t\"filters\": map[string]interface{}{\n\t\t\t\t\"content_type\": \"text/*\",\n\t\t\t},\n\t\t}\n\n\t\tp := process.New(\"attachment.List\", managerName, options)\n\t\tresult := processList(p)\n\n\t\tif err, ok := result.(error); ok {\n\t\t\tt.Fatalf(\"Failed to list files with filters: %v\", err)\n\t\t}\n\n\t\tlistResult, ok := result.(*ListResult)\n\t\tif !ok {\n\t\t\tt.Fatalf(\"Expected *ListResult, got %T\", result)\n\t\t}\n\n\t\t// Should find text/plain and text/csv files\n\t\t// Note: The filter implementation may vary, so we just check the call succeeds\n\t\tt.Logf(\"List with content_type filter - Total: %d files\", listResult.Total)\n\t})\n\n\t// Test 4: List with ordering\n\tt.Run(\"ListWithOrdering\", func(t *testing.T) {\n\t\toptions := map[string]interface{}{\n\t\t\t\"order_by\": \"name asc\",\n\t\t}\n\n\t\tp := process.New(\"attachment.List\", managerName, options)\n\t\tresult := processList(p)\n\n\t\tif err, ok := result.(error); ok {\n\t\t\tt.Fatalf(\"Failed to list files with ordering: %v\", err)\n\t\t}\n\n\t\tlistResult, ok := result.(*ListResult)\n\t\tif !ok {\n\t\t\tt.Fatalf(\"Expected *ListResult, got %T\", result)\n\t\t}\n\n\t\t// Verify files are sorted\n\t\tfor i := 1; i < len(listResult.Files); i++ {\n\t\t\tif listResult.Files[i-1].Filename > listResult.Files[i].Filename {\n\t\t\t\tt.Errorf(\"Files are not sorted ascending by name\")\n\t\t\t\tbreak\n\t\t\t}\n\t\t}\n\t})\n}\n\nfunc TestProcessDelete(t *testing.T) {\n\ttest.Prepare(t, config.Conf)\n\tdefer test.Clean()\n\n\t// Register default uploader for testing\n\t_, err := RegisterDefault(\"data.local\")\n\tif err != nil {\n\t\tt.Fatalf(\"Failed to register manager: %v\", err)\n\t}\n\n\t// Save a file to delete\n\tcontent := \"Content to delete\"\n\tbase64Content := base64.StdEncoding.EncodeToString([]byte(content))\n\tdataURI := fmt.Sprintf(\"data:text/plain;base64,%s\", base64Content)\n\n\tsaveP := process.New(\"attachment.Save\", \"data.local\", dataURI, \"delete-test.txt\")\n\tsaveResult := processSave(saveP)\n\tfile, ok := saveResult.(*File)\n\tif !ok {\n\t\tt.Fatalf(\"Failed to save test file: %v\", saveResult)\n\t}\n\n\t// Test 1: Delete existing file\n\tt.Run(\"DeleteExistingFile\", func(t *testing.T) {\n\t\t// Verify file exists first\n\t\texistsP := process.New(\"attachment.Exists\", \"data.local\", file.ID)\n\t\texistsResult := processExists(existsP)\n\t\tif exists, ok := existsResult.(bool); !ok || !exists {\n\t\t\tt.Fatal(\"File should exist before deletion\")\n\t\t}\n\n\t\t// Delete the file\n\t\tp := process.New(\"attachment.Delete\", \"data.local\", file.ID)\n\t\tresult := processDelete(p)\n\n\t\tif err, ok := result.(error); ok {\n\t\t\tt.Fatalf(\"Failed to delete file: %v\", err)\n\t\t}\n\n\t\tsuccess, ok := result.(bool)\n\t\tif !ok || !success {\n\t\t\tt.Errorf(\"Expected true, got %v\", result)\n\t\t}\n\n\t\t// Verify file no longer exists\n\t\texistsP2 := process.New(\"attachment.Exists\", \"data.local\", file.ID)\n\t\texistsResult2 := processExists(existsP2)\n\t\tif exists, ok := existsResult2.(bool); ok && exists {\n\t\t\tt.Error(\"File should not exist after deletion\")\n\t\t}\n\t})\n\n\t// Test 2: Delete non-existent file\n\tt.Run(\"DeleteNonExistent\", func(t *testing.T) {\n\t\tp := process.New(\"attachment.Delete\", \"data.local\", \"non-existent-file-id\")\n\t\tresult := processDelete(p)\n\n\t\terr, ok := result.(error)\n\t\tif !ok {\n\t\t\tt.Fatal(\"Expected error for non-existent file\")\n\t\t}\n\n\t\tif !strings.Contains(err.Error(), \"file not found\") {\n\t\t\tt.Errorf(\"Expected 'file not found' error, got: %s\", err.Error())\n\t\t}\n\t})\n}\n\nfunc TestProcessExists(t *testing.T) {\n\ttest.Prepare(t, config.Conf)\n\tdefer test.Clean()\n\n\t// Register default uploader for testing\n\t_, err := RegisterDefault(\"data.local\")\n\tif err != nil {\n\t\tt.Fatalf(\"Failed to register manager: %v\", err)\n\t}\n\n\t// Save a file\n\tcontent := \"Exists test content\"\n\tbase64Content := base64.StdEncoding.EncodeToString([]byte(content))\n\tdataURI := fmt.Sprintf(\"data:text/plain;base64,%s\", base64Content)\n\n\tsaveP := process.New(\"attachment.Save\", \"data.local\", dataURI, \"exists-test.txt\")\n\tsaveResult := processSave(saveP)\n\tfile, ok := saveResult.(*File)\n\tif !ok {\n\t\tt.Fatalf(\"Failed to save test file: %v\", saveResult)\n\t}\n\n\t// Test 1: Existing file\n\tt.Run(\"FileExists\", func(t *testing.T) {\n\t\tp := process.New(\"attachment.Exists\", \"data.local\", file.ID)\n\t\tresult := processExists(p)\n\n\t\texists, ok := result.(bool)\n\t\tif !ok {\n\t\t\tt.Fatalf(\"Expected bool, got %T\", result)\n\t\t}\n\n\t\tif !exists {\n\t\t\tt.Error(\"File should exist\")\n\t\t}\n\t})\n\n\t// Test 2: Non-existent file\n\tt.Run(\"FileNotExists\", func(t *testing.T) {\n\t\tp := process.New(\"attachment.Exists\", \"data.local\", \"non-existent-file-id\")\n\t\tresult := processExists(p)\n\n\t\texists, ok := result.(bool)\n\t\tif !ok {\n\t\t\tt.Fatalf(\"Expected bool, got %T\", result)\n\t\t}\n\n\t\tif exists {\n\t\t\tt.Error(\"File should not exist\")\n\t\t}\n\t})\n}\n\nfunc TestProcessURL(t *testing.T) {\n\ttest.Prepare(t, config.Conf)\n\tdefer test.Clean()\n\n\t// Register default uploader for testing\n\t_, err := RegisterDefault(\"data.local\")\n\tif err != nil {\n\t\tt.Fatalf(\"Failed to register manager: %v\", err)\n\t}\n\n\t// Save a file\n\tcontent := \"URL test content\"\n\tbase64Content := base64.StdEncoding.EncodeToString([]byte(content))\n\tdataURI := fmt.Sprintf(\"data:text/plain;base64,%s\", base64Content)\n\n\tsaveP := process.New(\"attachment.Save\", \"data.local\", dataURI, \"url-test.txt\")\n\tsaveResult := processSave(saveP)\n\tfile, ok := saveResult.(*File)\n\tif !ok {\n\t\tt.Fatalf(\"Failed to save test file: %v\", saveResult)\n\t}\n\n\t// Test 1: Get URL\n\tt.Run(\"GetURL\", func(t *testing.T) {\n\t\tp := process.New(\"attachment.URL\", \"data.local\", file.ID)\n\t\tresult := processURL(p)\n\n\t\tif err, ok := result.(error); ok {\n\t\t\tt.Fatalf(\"Failed to get URL: %v\", err)\n\t\t}\n\n\t\turl, ok := result.(string)\n\t\tif !ok {\n\t\t\tt.Fatalf(\"Expected string, got %T\", result)\n\t\t}\n\n\t\tif url == \"\" {\n\t\t\tt.Error(\"URL should not be empty\")\n\t\t}\n\n\t\tt.Logf(\"File URL: %s\", url)\n\t})\n\n\t// Test 2: Get URL for non-existent file\n\tt.Run(\"GetURLNonExistent\", func(t *testing.T) {\n\t\tp := process.New(\"attachment.URL\", \"data.local\", \"non-existent-file-id\")\n\t\tresult := processURL(p)\n\n\t\terr, ok := result.(error)\n\t\tif !ok {\n\t\t\tt.Fatal(\"Expected error for non-existent file\")\n\t\t}\n\n\t\tif !strings.Contains(err.Error(), \"file not found\") {\n\t\t\tt.Errorf(\"Expected 'file not found' error, got: %s\", err.Error())\n\t\t}\n\t})\n}\n\nfunc TestProcessSaveTextAndGetText(t *testing.T) {\n\ttest.Prepare(t, config.Conf)\n\tdefer test.Clean()\n\n\t// Register default uploader for testing\n\t_, err := RegisterDefault(\"data.local\")\n\tif err != nil {\n\t\tt.Fatalf(\"Failed to register manager: %v\", err)\n\t}\n\n\t// Save a file\n\tcontent := \"Original file content\"\n\tbase64Content := base64.StdEncoding.EncodeToString([]byte(content))\n\tdataURI := fmt.Sprintf(\"data:text/plain;base64,%s\", base64Content)\n\n\tsaveP := process.New(\"attachment.Save\", \"data.local\", dataURI, \"text-test.txt\")\n\tsaveResult := processSave(saveP)\n\tfile, ok := saveResult.(*File)\n\tif !ok {\n\t\tt.Fatalf(\"Failed to save test file: %v\", saveResult)\n\t}\n\n\t// Test 1: Get text from file without saved text (should be empty)\n\tt.Run(\"GetTextEmpty\", func(t *testing.T) {\n\t\tp := process.New(\"attachment.GetText\", \"data.local\", file.ID)\n\t\tresult := processGetText(p)\n\n\t\tif err, ok := result.(error); ok {\n\t\t\tt.Fatalf(\"Failed to get text: %v\", err)\n\t\t}\n\n\t\ttext, ok := result.(string)\n\t\tif !ok {\n\t\t\tt.Fatalf(\"Expected string, got %T\", result)\n\t\t}\n\n\t\tif text != \"\" {\n\t\t\tt.Errorf(\"Expected empty text, got: %s\", text)\n\t\t}\n\t})\n\n\t// Test 2: Save text and retrieve\n\tt.Run(\"SaveTextAndRetrieve\", func(t *testing.T) {\n\t\tparsedText := \"This is the parsed/extracted text content from the file.\"\n\n\t\t// Save text\n\t\tsaveTextP := process.New(\"attachment.SaveText\", \"data.local\", file.ID, parsedText)\n\t\tsaveTextResult := processSaveText(saveTextP)\n\n\t\tif err, ok := saveTextResult.(error); ok {\n\t\t\tt.Fatalf(\"Failed to save text: %v\", err)\n\t\t}\n\n\t\tsuccess, ok := saveTextResult.(bool)\n\t\tif !ok || !success {\n\t\t\tt.Errorf(\"Expected true, got %v\", saveTextResult)\n\t\t}\n\n\t\t// Retrieve text\n\t\tgetTextP := process.New(\"attachment.GetText\", \"data.local\", file.ID)\n\t\tgetTextResult := processGetText(getTextP)\n\n\t\tif err, ok := getTextResult.(error); ok {\n\t\t\tt.Fatalf(\"Failed to get text: %v\", err)\n\t\t}\n\n\t\tretrievedText, ok := getTextResult.(string)\n\t\tif !ok {\n\t\t\tt.Fatalf(\"Expected string, got %T\", getTextResult)\n\t\t}\n\n\t\tif retrievedText != parsedText {\n\t\t\tt.Errorf(\"Text mismatch. Expected: %s, Got: %s\", parsedText, retrievedText)\n\t\t}\n\n\t\tt.Logf(\"Saved and retrieved text: %s\", retrievedText)\n\t})\n\n\t// Test 3: Get full text vs preview\n\tt.Run(\"GetTextFullVsPreview\", func(t *testing.T) {\n\t\t// Save a long text\n\t\tlongText := strings.Repeat(\"This is a long text content. \", 200) // > 2000 chars\n\n\t\tsaveTextP := process.New(\"attachment.SaveText\", \"data.local\", file.ID, longText)\n\t\tsaveTextResult := processSaveText(saveTextP)\n\t\tif err, ok := saveTextResult.(error); ok {\n\t\t\tt.Fatalf(\"Failed to save long text: %v\", err)\n\t\t}\n\n\t\t// Get preview (default)\n\t\tpreviewP := process.New(\"attachment.GetText\", \"data.local\", file.ID)\n\t\tpreviewResult := processGetText(previewP)\n\t\tpreviewText, _ := previewResult.(string)\n\n\t\t// Preview should be 2000 runes\n\t\tif len([]rune(previewText)) != 2000 {\n\t\t\tt.Errorf(\"Preview should be 2000 runes, got %d\", len([]rune(previewText)))\n\t\t}\n\n\t\t// Get full content\n\t\tfullP := process.New(\"attachment.GetText\", \"data.local\", file.ID, true)\n\t\tfullResult := processGetText(fullP)\n\t\tfullText, _ := fullResult.(string)\n\n\t\tif fullText != longText {\n\t\t\tt.Errorf(\"Full text length mismatch. Expected: %d, Got: %d\", len(longText), len(fullText))\n\t\t}\n\t})\n\n\t// Test 4: Save/Get text for non-existent file\n\tt.Run(\"SaveTextNonExistent\", func(t *testing.T) {\n\t\tp := process.New(\"attachment.SaveText\", \"data.local\", \"non-existent-id\", \"some text\")\n\t\tresult := processSaveText(p)\n\n\t\terr, ok := result.(error)\n\t\tif !ok {\n\t\t\tt.Fatal(\"Expected error for non-existent file\")\n\t\t}\n\n\t\tif !strings.Contains(err.Error(), \"file not found\") {\n\t\t\tt.Errorf(\"Expected 'file not found' error, got: %s\", err.Error())\n\t\t}\n\t})\n}\n\nfunc TestProcessWithAuthorizedPermission(t *testing.T) {\n\ttest.Prepare(t, config.Conf)\n\tdefer test.Clean()\n\n\t// Register default uploader for testing\n\t_, err := RegisterDefault(\"data.local\")\n\tif err != nil {\n\t\tt.Fatalf(\"Failed to register manager: %v\", err)\n\t}\n\n\t// Test 1: Save with Authorized info - verify via database query since File struct\n\t// does not expose these fields in JSON (they are marked with json:\"-\")\n\tt.Run(\"SaveWithAuthorizedInfo\", func(t *testing.T) {\n\t\tcontent := \"Content with permission\"\n\t\tbase64Content := base64.StdEncoding.EncodeToString([]byte(content))\n\t\tdataURI := fmt.Sprintf(\"data:text/plain;base64,%s\", base64Content)\n\n\t\tp := process.New(\"attachment.Save\", \"data.local\", dataURI, \"perm-test.txt\")\n\n\t\t// Set authorized info\n\t\tp.WithAuthorized(process.AuthorizedInfo{\n\t\t\tUserID: \"user123\",\n\t\t\tTeamID: \"team456\",\n\t\t\tTenantID: \"tenant789\",\n\t\t})\n\n\t\tresult := processSave(p)\n\n\t\tif err, ok := result.(error); ok {\n\t\t\tt.Fatalf(\"Failed to save file: %v\", err)\n\t\t}\n\n\t\tfile, ok := result.(*File)\n\t\tif !ok {\n\t\t\tt.Fatalf(\"Expected *File, got %T\", result)\n\t\t}\n\n\t\t// File should be saved successfully\n\t\tif file.ID == \"\" {\n\t\t\tt.Error(\"File ID should not be empty\")\n\t\t}\n\n\t\t// Note: The YaoCreatedBy, YaoTeamID, YaoTenantID fields in File struct\n\t\t// are marked with json:\"-\" and may not be populated in the returned struct.\n\t\t// The permission fields are stored in the database during upload via UploadOption.\n\t\t// To verify, we would need to query the database directly.\n\t\tt.Logf(\"File saved with authorized info - ID: %s\", file.ID)\n\t})\n\n\t// Test 2: Save without Authorized (should still work)\n\tt.Run(\"SaveWithoutAuthorized\", func(t *testing.T) {\n\t\tcontent := \"Content without permission\"\n\t\tbase64Content := base64.StdEncoding.EncodeToString([]byte(content))\n\t\tdataURI := fmt.Sprintf(\"data:text/plain;base64,%s\", base64Content)\n\n\t\tp := process.New(\"attachment.Save\", \"data.local\", dataURI, \"no-perm-test.txt\")\n\t\t// Don't set authorized info\n\n\t\tresult := processSave(p)\n\n\t\tif err, ok := result.(error); ok {\n\t\t\tt.Fatalf(\"Failed to save file: %v\", err)\n\t\t}\n\n\t\tfile, ok := result.(*File)\n\t\tif !ok {\n\t\t\tt.Fatalf(\"Expected *File, got %T\", result)\n\t\t}\n\n\t\t// File should be saved successfully without permission fields\n\t\tif file.ID == \"\" {\n\t\t\tt.Error(\"File ID should not be empty\")\n\t\t}\n\n\t\tt.Logf(\"File saved without permissions - ID: %s\", file.ID)\n\t})\n}\n\nfunc TestParseDataURI(t *testing.T) {\n\t// Test 1: Valid data URI with content type\n\tt.Run(\"ValidDataURI\", func(t *testing.T) {\n\t\tcontent := \"Hello, World!\"\n\t\tbase64Content := base64.StdEncoding.EncodeToString([]byte(content))\n\t\tdataURI := fmt.Sprintf(\"data:text/plain;base64,%s\", base64Content)\n\n\t\tcontentType, data, err := parseDataURI(dataURI)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to parse data URI: %v\", err)\n\t\t}\n\n\t\tif contentType != \"text/plain\" {\n\t\t\tt.Errorf(\"Expected content type 'text/plain', got '%s'\", contentType)\n\t\t}\n\n\t\tif string(data) != content {\n\t\t\tt.Errorf(\"Expected content '%s', got '%s'\", content, string(data))\n\t\t}\n\t})\n\n\t// Test 2: Plain text (no data URI header) - treated as plain text, not base64\n\tt.Run(\"PlainText\", func(t *testing.T) {\n\t\tcontent := \"Plain text content\"\n\n\t\tcontentType, data, err := parseDataURI(content)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to parse plain text: %v\", err)\n\t\t}\n\n\t\tif contentType != \"text/plain\" {\n\t\t\tt.Errorf(\"Expected content type 'text/plain', got '%s'\", contentType)\n\t\t}\n\n\t\tif string(data) != content {\n\t\t\tt.Errorf(\"Expected content '%s', got '%s'\", content, string(data))\n\t\t}\n\t})\n\n\t// Test 3: Data URI with image\n\tt.Run(\"ImageDataURI\", func(t *testing.T) {\n\t\tpngBase64 := \"iVBORw0KGgoAAAANSUhEUgAAAAEAAAABCAYAAAAfFcSJAAAADUlEQVR42mNk+M9QDwADhgGAWjR9awAAAABJRU5ErkJggg==\"\n\t\tdataURI := fmt.Sprintf(\"data:image/png;base64,%s\", pngBase64)\n\n\t\tcontentType, _, err := parseDataURI(dataURI)\n\t\tif err != nil {\n\t\t\tt.Fatalf(\"Failed to parse image data URI: %v\", err)\n\t\t}\n\n\t\tif contentType != \"image/png\" {\n\t\t\tt.Errorf(\"Expected content type 'image/png', got '%s'\", contentType)\n\t\t}\n\t})\n\n\t// Test 4: Invalid base64\n\tt.Run(\"InvalidBase64\", func(t *testing.T) {\n\t\tdataURI := \"data:text/plain;base64,not-valid!!!\"\n\n\t\t_, _, err := parseDataURI(dataURI)\n\t\tif err == nil {\n\t\t\tt.Fatal(\"Expected error for invalid base64\")\n\t\t}\n\t})\n\n\t// Test 5: Invalid data URI format\n\tt.Run(\"InvalidDataURIFormat\", func(t *testing.T) {\n\t\tdataURI := \"data:text/plain\" // Missing base64 part\n\n\t\t_, _, err := parseDataURI(dataURI)\n\t\tif err == nil {\n\t\t\tt.Fatal(\"Expected error for invalid data URI format\")\n\t\t}\n\t})\n}\n\nfunc TestGenerateFilename(t *testing.T) {\n\t// Note: mime.ExtensionsByType may return different extensions on different systems\n\t// (e.g., Linux may return .jfif for image/jpeg, .asc for text/plain)\n\t// So we verify the filename has a proper extension format and is not empty\n\ttestCases := []struct {\n\t\tcontentType string\n\t\texpectedPrefix string\n\t}{\n\t\t{\"image/png\", \"file\"},\n\t\t{\"image/jpeg\", \"file\"},\n\t\t{\"image/gif\", \"file\"},\n\t\t{\"image/webp\", \"file\"},\n\t\t{\"text/plain\", \"file\"},\n\t\t{\"application/pdf\", \"file\"},\n\t\t{\"application/json\", \"file\"},\n\t\t{\"application/octet-stream\", \"file\"},\n\t\t{\"unknown/type\", \"file\"},\n\t}\n\n\tfor _, tc := range testCases {\n\t\tt.Run(tc.contentType, func(t *testing.T) {\n\t\t\tfilename := generateFilename(tc.contentType)\n\n\t\t\t// Check prefix\n\t\t\tif !strings.HasPrefix(filename, tc.expectedPrefix) {\n\t\t\t\tt.Errorf(\"For content type '%s', expected prefix '%s', got '%s'\", tc.contentType, tc.expectedPrefix, filename)\n\t\t\t}\n\n\t\t\t// Check filename has an extension (starts with dot and has at least one character)\n\t\t\tdotIndex := strings.LastIndex(filename, \".\")\n\t\t\tif dotIndex == -1 || dotIndex == len(filename)-1 {\n\t\t\t\tt.Errorf(\"For content type '%s', expected filename with extension, got '%s'\", tc.contentType, filename)\n\t\t\t}\n\n\t\t\t// Extension should not be empty\n\t\t\text := filename[dotIndex:]\n\t\t\tif len(ext) < 2 {\n\t\t\t\tt.Errorf(\"For content type '%s', expected non-empty extension, got '%s'\", tc.contentType, ext)\n\t\t\t}\n\t\t})\n\t}\n}\n" }, { "path": "attachment/s3/storage.go", "content": "package s3\n\nimport (\n\t\"bytes\"\n\t\"compress/gzip\"\n\t\"context\"\n\t\"fmt\"\n\t\"image\"\n\t\"image/jpeg\"\n\t\"image/png\"\n\t\"io\"\n\t\"mime\"\n\t\"net/http\"\n\t\"os\"\n\t\"path/filepath\"\n\t\"strings\"\n\t\"time\"\n\n\t\"github.com/aws/aws-sdk-go-v2/aws\"\n\t\"github.com/aws/aws-sdk-go-v2/credentials\"\n\t\"github.com/aws/aws-sdk-go-v2/service/s3\"\n)\n\n// DefaultExpiration default expiration time for presigned URLs (5 minutes)\nconst DefaultExpiration = 5 * time.Minute\n\n// MaxImageSize maximum image size (1920x1080)\nconst MaxImageSize = 1920\n\n// Storage the S3 storage driver\ntype Storage struct {\n\tEndpoint string `json:\"endpoint\" yaml:\"endpoint\"`\n\tRegion string `json:\"region\" yaml:\"region\"`\n\tKey string `json:\"key\" yaml:\"key\"`\n\tSecret string `json:\"secret\" yaml:\"secret\"`\n\tBucket string `json:\"bucket\" yaml:\"bucket\"`\n\tExpiration time.Duration `json:\"expiration\" yaml:\"expiration\"`\n\tCacheDir string `json:\"cache_dir\" yaml:\"cache_dir\"`\n\tclient *s3.Client\n\tprefix string\n\tcompression bool\n}\n\n// New create a new S3 storage\nfunc New(options map[string]interface{}) (*Storage, error) {\n\tstorage := &Storage{\n\t\tRegion: \"auto\",\n\t\tExpiration: DefaultExpiration,\n\t\tcompression: true,\n\t}\n\n\tif endpoint, ok := options[\"endpoint\"].(string); ok {\n\t\tstorage.Endpoint = endpoint\n\t}\n\n\tif region, ok := options[\"region\"].(string); ok {\n\t\tstorage.Region = region\n\t}\n\n\tif key, ok := options[\"key\"].(string); ok {\n\t\tstorage.Key = key\n\t}\n\n\tif secret, ok := options[\"secret\"].(string); ok {\n\t\tstorage.Secret = secret\n\t}\n\n\tif bucket, ok := options[\"bucket\"].(string); ok {\n\t\tstorage.Bucket = bucket\n\t}\n\n\tif prefix, ok := options[\"prefix\"].(string); ok {\n\t\tstorage.prefix = prefix\n\t}\n\n\tif cacheDir, ok := options[\"cache_dir\"].(string); ok {\n\t\tstorage.CacheDir = cacheDir\n\t} else {\n\t\t// Use system temp directory as default\n\t\tstorage.CacheDir = os.TempDir()\n\t}\n\n\tif exp, ok := options[\"expiration\"].(time.Duration); ok {\n\t\tstorage.Expiration = exp\n\t}\n\n\tif compression, ok := options[\"compression\"].(bool); ok {\n\t\tstorage.compression = compression\n\t}\n\n\t// Validate required fields\n\tif storage.Key == \"\" || storage.Secret == \"\" {\n\t\treturn nil, fmt.Errorf(\"key and secret are required\")\n\t}\n\n\tif storage.Bucket == \"\" {\n\t\treturn nil, fmt.Errorf(\"bucket is required\")\n\t}\n\n\t// Create S3 client\n\topts := s3.Options{\n\t\tRegion: storage.Region,\n\t\tCredentials: credentials.NewStaticCredentialsProvider(storage.Key, storage.Secret, \"\"),\n\t\tUsePathStyle: true,\n\t}\n\n\tif storage.Endpoint != \"\" {\n\t\t// Remove bucket name from endpoint if present\n\t\tendpoint := storage.Endpoint\n\t\tif strings.Contains(endpoint, \"/\"+storage.Bucket) {\n\t\t\tendpoint = strings.TrimSuffix(endpoint, \"/\"+storage.Bucket)\n\t\t}\n\t\topts.BaseEndpoint = aws.String(endpoint)\n\t}\n\n\tstorage.client = s3.New(opts)\n\n\t// Ensure cache directory exists\n\tif err := os.MkdirAll(storage.CacheDir, 0755); err != nil {\n\t\treturn nil, fmt.Errorf(\"failed to create cache directory %s: %w\", storage.CacheDir, err)\n\t}\n\n\treturn storage, nil\n}\n\n// Upload upload file to S3\nfunc (storage *Storage) Upload(ctx context.Context, path string, reader io.Reader, contentType string) (string, error) {\n\tif storage.client == nil {\n\t\treturn \"\", fmt.Errorf(\"s3 client not initialized\")\n\t}\n\n\tkey := filepath.Join(storage.prefix, path)\n\n\t// Upload file\n\t_, err := storage.client.PutObject(ctx, &s3.PutObjectInput{\n\t\tBucket: aws.String(storage.Bucket),\n\t\tKey: aws.String(key),\n\t\tBody: reader,\n\t\tContentType: aws.String(contentType),\n\t})\n\tif err != nil {\n\t\treturn \"\", fmt.Errorf(\"failed to upload file %s: %w\", path, err)\n\t}\n\n\treturn path, nil\n}\n\n// UploadChunk uploads a chunk of a file to S3\nfunc (storage *Storage) UploadChunk(ctx context.Context, path string, chunkIndex int, reader io.Reader, contentType string) error {\n\tif storage.client == nil {\n\t\treturn fmt.Errorf(\"s3 client not initialized\")\n\t}\n\n\t// Store chunks with a special prefix\n\tchunkKey := filepath.Join(storage.prefix, \".chunks\", path, fmt.Sprintf(\"chunk_%d\", chunkIndex))\n\n\t_, err := storage.client.PutObject(ctx, &s3.PutObjectInput{\n\t\tBucket: aws.String(storage.Bucket),\n\t\tKey: aws.String(chunkKey),\n\t\tBody: reader,\n\t\tContentType: aws.String(contentType),\n\t})\n\tif err != nil {\n\t\treturn fmt.Errorf(\"failed to upload chunk %s %d: %w\", path, chunkIndex, err)\n\t}\n\n\treturn nil\n}\n\n// MergeChunks merges all chunks into the final file in S3\nfunc (storage *Storage) MergeChunks(ctx context.Context, path string, totalChunks int) error {\n\tif storage.client == nil {\n\t\treturn fmt.Errorf(\"s3 client not initialized\")\n\t}\n\n\tfinalKey := filepath.Join(storage.prefix, path)\n\n\t// Create a buffer to hold the merged content\n\tvar mergedContent bytes.Buffer\n\tvar contentType string\n\n\t// Download and merge chunks in order\n\tfor i := 0; i < totalChunks; i++ {\n\t\tchunkKey := filepath.Join(storage.prefix, \".chunks\", path, fmt.Sprintf(\"chunk_%d\", i))\n\n\t\tresult, err := storage.client.GetObject(ctx, &s3.GetObjectInput{\n\t\t\tBucket: aws.String(storage.Bucket),\n\t\t\tKey: aws.String(chunkKey),\n\t\t})\n\t\tif err != nil {\n\t\t\treturn fmt.Errorf(\"failed to get chunk %d: %w\", i, err)\n\t\t}\n\n\t\t// Get content type from the first chunk\n\t\tif i == 0 && result.ContentType != nil {\n\t\t\tcontentType = *result.ContentType\n\t\t}\n\n\t\t_, err = io.Copy(&mergedContent, result.Body)\n\t\tresult.Body.Close()\n\t\tif err != nil {\n\t\t\treturn fmt.Errorf(\"failed to copy chunk %s %d: %w\", path, i, err)\n\t\t}\n\t}\n\n\t// Default content type if not found\n\tif contentType == \"\" {\n\t\tcontentType = \"application/octet-stream\"\n\t}\n\n\t// Upload the merged content as the final file with proper content type\n\t_, err := storage.client.PutObject(ctx, &s3.PutObjectInput{\n\t\tBucket: aws.String(storage.Bucket),\n\t\tKey: aws.String(finalKey),\n\t\tBody: bytes.NewReader(mergedContent.Bytes()),\n\t\tContentType: aws.String(contentType),\n\t})\n\tif err != nil {\n\t\treturn fmt.Errorf(\"failed to upload merged file %s: %w\", path, err)\n\t}\n\n\t// Clean up chunks\n\tfor i := 0; i < totalChunks; i++ {\n\t\tchunkKey := filepath.Join(storage.prefix, \".chunks\", path, fmt.Sprintf(\"chunk_%d\", i))\n\t\tstorage.client.DeleteObject(ctx, &s3.DeleteObjectInput{\n\t\t\tBucket: aws.String(storage.Bucket),\n\t\t\tKey: aws.String(chunkKey),\n\t\t})\n\t}\n\n\treturn nil\n}\n\n// Reader read file from S3\nfunc (storage *Storage) Reader(ctx context.Context, path string) (io.ReadCloser, error) {\n\tif storage.client == nil {\n\t\treturn nil, fmt.Errorf(\"s3 client not initialized\")\n\t}\n\n\tkey := filepath.Join(storage.prefix, path)\n\n\tresult, err := storage.client.GetObject(ctx, &s3.GetObjectInput{\n\t\tBucket: aws.String(storage.Bucket),\n\t\tKey: aws.String(key),\n\t})\n\tif err != nil {\n\t\treturn nil, fmt.Errorf(\"failed to get file %s: %w\", path, err)\n\t}\n\n\t// If the file is a gzip file, decompress it\n\tif strings.HasSuffix(path, \".gz\") {\n\t\treader, err := gzip.NewReader(result.Body)\n\t\tif err != nil {\n\t\t\treturn nil, err\n\t\t}\n\t\treturn reader, nil\n\t}\n\n\treturn result.Body, nil\n}\n\n// Download download file from S3\nfunc (storage *Storage) Download(ctx context.Context, path string) (io.ReadCloser, string, error) {\n\tif storage.client == nil {\n\t\treturn nil, \"\", fmt.Errorf(\"s3 client not initialized\")\n\t}\n\n\tkey := filepath.Join(storage.prefix, path)\n\n\t// Get object\n\tresult, err := storage.client.GetObject(ctx, &s3.GetObjectInput{\n\t\tBucket: aws.String(storage.Bucket),\n\t\tKey: aws.String(key),\n\t})\n\tif err != nil {\n\t\treturn nil, \"\", fmt.Errorf(\"failed to download file %s: %w\", path, err)\n\t}\n\n\tcontentType := \"application/octet-stream\"\n\tif result.ContentType != nil {\n\t\tcontentType = *result.ContentType\n\t}\n\n\t// Try to detect content type from file extension\n\text := filepath.Ext(strings.TrimSuffix(path, \".gz\"))\n\tswitch strings.ToLower(ext) {\n\tcase \".txt\":\n\t\tcontentType = \"text/plain\"\n\tcase \".html\":\n\t\tcontentType = \"text/html\"\n\tcase \".css\":\n\t\tcontentType = \"text/css\"\n\tcase \".js\":\n\t\tcontentType = \"application/javascript\"\n\tcase \".json\":\n\t\tcontentType = \"application/json\"\n\tcase \".jpg\", \".jpeg\":\n\t\tcontentType = \"image/jpeg\"\n\tcase \".png\":\n\t\tcontentType = \"image/png\"\n\tcase \".gif\":\n\t\tcontentType = \"image/gif\"\n\tcase \".pdf\":\n\t\tcontentType = \"application/pdf\"\n\tcase \".mp4\":\n\t\tcontentType = \"video/mp4\"\n\tcase \".mp3\":\n\t\tcontentType = \"audio/mpeg\"\n\tcase \".wav\":\n\t\tcontentType = \"audio/wav\"\n\tcase \".ogg\":\n\t\tcontentType = \"audio/ogg\"\n\tcase \".webm\":\n\t\tcontentType = \"video/webm\"\n\tcase \".webp\":\n\t\tcontentType = \"image/webp\"\n\tcase \".zip\":\n\t}\n\n\t// If the file is a gzip file, decompress it\n\tif strings.HasSuffix(path, \".gz\") {\n\t\treader, err := gzip.NewReader(result.Body)\n\t\tif err != nil {\n\t\t\treturn nil, \"\", err\n\t\t}\n\t\treturn reader, contentType, nil\n\t}\n\n\treturn result.Body, contentType, nil\n}\n\n// GetContent gets file content as bytes\nfunc (storage *Storage) GetContent(ctx context.Context, path string) ([]byte, error) {\n\treader, err := storage.Reader(ctx, path)\n\tif err != nil {\n\t\treturn nil, err\n\t}\n\tdefer reader.Close()\n\n\treturn io.ReadAll(reader)\n}\n\n// URL get file url with expiration\nfunc (storage *Storage) URL(ctx context.Context, path string) string {\n\tif storage.client == nil {\n\t\treturn \"\"\n\t}\n\n\tkey := filepath.Join(storage.prefix, path)\n\tpresignClient := s3.NewPresignClient(storage.client)\n\trequest, err := presignClient.PresignGetObject(ctx, &s3.GetObjectInput{\n\t\tBucket: aws.String(storage.Bucket),\n\t\tKey: aws.String(key),\n\t}, s3.WithPresignExpires(storage.Expiration))\n\n\tif err != nil {\n\t\treturn \"\"\n\t}\n\n\treturn request.URL\n}\n\n// Exists checks if a file exists in S3\nfunc (storage *Storage) Exists(ctx context.Context, path string) bool {\n\tif storage.client == nil {\n\t\treturn false\n\t}\n\n\tkey := filepath.Join(storage.prefix, path)\n\t_, err := storage.client.HeadObject(ctx, &s3.HeadObjectInput{\n\t\tBucket: aws.String(storage.Bucket),\n\t\tKey: aws.String(key),\n\t})\n\treturn err == nil\n}\n\n// Delete deletes a file from S3\nfunc (storage *Storage) Delete(ctx context.Context, path string) error {\n\tif storage.client == nil {\n\t\treturn fmt.Errorf(\"s3 client not initialized\")\n\t}\n\n\tkey := filepath.Join(storage.prefix, path)\n\t_, err := storage.client.DeleteObject(ctx, &s3.DeleteObjectInput{\n\t\tBucket: aws.String(storage.Bucket),\n\t\tKey: aws.String(key),\n\t})\n\tif err != nil {\n\t\treturn fmt.Errorf(\"failed to delete file: %w\", err)\n\t}\n\n\treturn nil\n}\n\nfunc (storage *Storage) makeID(filename string, ext string) string {\n\tdate := time.Now().Format(\"20060102\")\n\tname := strings.TrimSuffix(filepath.Base(filename), ext)\n\treturn fmt.Sprintf(\"%s/%s-%d%s\", date, name, time.Now().UnixNano(), ext)\n}\n\n// isImage checks if the content type is an image\nfunc isImage(contentType string) bool {\n\treturn strings.HasPrefix(contentType, \"image/\")\n}\n\n// compressImage compresses the image while maintaining aspect ratio\nfunc compressImage(data []byte, contentType string) ([]byte, error) {\n\t// Decode image\n\timg, _, err := image.Decode(bytes.NewReader(data))\n\tif err != nil {\n\t\treturn nil, fmt.Errorf(\"failed to decode image: %w\", err)\n\t}\n\n\t// Calculate new dimensions\n\tbounds := img.Bounds()\n\twidth := bounds.Dx()\n\theight := bounds.Dy()\n\tvar newWidth, newHeight int\n\n\tif width > height {\n\t\tif width > MaxImageSize {\n\t\t\tnewWidth = MaxImageSize\n\t\t\tnewHeight = int(float64(height) * (float64(MaxImageSize) / float64(width)))\n\t\t} else {\n\t\t\treturn data, nil // No need to resize\n\t\t}\n\t} else {\n\t\tif height > MaxImageSize {\n\t\t\tnewHeight = MaxImageSize\n\t\t\tnewWidth = int(float64(width) * (float64(MaxImageSize) / float64(height)))\n\t\t} else {\n\t\t\treturn data, nil // No need to resize\n\t\t}\n\t}\n\n\t// Create new image with new dimensions\n\tnewImg := image.NewRGBA(image.Rect(0, 0, newWidth, newHeight))\n\n\t// Scale the image using bilinear interpolation\n\tfor y := 0; y < newHeight; y++ {\n\t\tfor x := 0; x < newWidth; x++ {\n\t\t\tsrcX := float64(x) * float64(width) / float64(newWidth)\n\t\t\tsrcY := float64(y) * float64(height) / float64(newHeight)\n\t\t\tnewImg.Set(x, y, img.At(int(srcX), int(srcY)))\n\t\t}\n\t}\n\n\t// Encode image\n\tvar buf bytes.Buffer\n\tswitch contentType {\n\tcase \"image/jpeg\":\n\t\terr = jpeg.Encode(&buf, newImg, &jpeg.Options{Quality: 85})\n\tcase \"image/png\":\n\t\terr = png.Encode(&buf, newImg)\n\tdefault:\n\t\treturn data, nil // Unsupported format, return original\n\t}\n\n\tif err != nil {\n\t\treturn nil, fmt.Errorf(\"failed to encode image: %w\", err)\n\t}\n\n\treturn buf.Bytes(), nil\n}\n\n// LocalPath downloads the file to cache directory and returns absolute path with content type\nfunc (storage *Storage) LocalPath(ctx context.Context, path string) (string, string, error) {\n\tif storage.client == nil {\n\t\treturn \"\", \"\", fmt.Errorf(\"s3 client not initialized\")\n\t}\n\n\t// Create cache file path using the same structure as storage path\n\tcacheFilePath := filepath.Join(storage.CacheDir, \"s3_cache\", path)\n\n\t// Create directory for cache file\n\tdir := filepath.Dir(cacheFilePath)\n\tif err := os.MkdirAll(dir, 0755); err != nil {\n\t\treturn \"\", \"\", fmt.Errorf(\"failed to create cache directory: %w\", err)\n\t}\n\n\t// Check if file already exists in cache and is not outdated\n\tif _, err := os.Stat(cacheFilePath); err == nil {\n\t\t// File exists in cache, detect content type and return\n\t\tcontentType, err := detectContentType(cacheFilePath)\n\t\tif err != nil {\n\t\t\treturn \"\", \"\", fmt.Errorf(\"failed to detect content type: %w\", err)\n\t\t}\n\t\treturn cacheFilePath, contentType, nil\n\t}\n\n\t// Download file from S3 to cache\n\tkey := filepath.Join(storage.prefix, path)\n\tresult, err := storage.client.GetObject(ctx, &s3.GetObjectInput{\n\t\tBucket: aws.String(storage.Bucket),\n\t\tKey: aws.String(key),\n\t})\n\tif err != nil {\n\t\treturn \"\", \"\", fmt.Errorf(\"failed to download file %s: %w\", path, err)\n\t}\n\tdefer result.Body.Close()\n\n\t// Create cache file\n\tcacheFile, err := os.Create(cacheFilePath)\n\tif err != nil {\n\t\treturn \"\", \"\", fmt.Errorf(\"failed to create cache file: %w\", err)\n\t}\n\tdefer cacheFile.Close()\n\n\t// Handle gzipped files - decompress during download\n\tvar reader io.Reader = result.Body\n\tif strings.HasSuffix(path, \".gz\") {\n\t\tgzipReader, err := gzip.NewReader(result.Body)\n\t\tif err != nil {\n\t\t\treturn \"\", \"\", fmt.Errorf(\"failed to create gzip reader: %w\", err)\n\t\t}\n\t\tdefer gzipReader.Close()\n\t\treader = gzipReader\n\n\t\t// Remove .gz extension from cache file path since we're decompressing\n\t\tnewCacheFilePath := strings.TrimSuffix(cacheFilePath, \".gz\")\n\t\tcacheFile.Close()\n\t\tos.Remove(cacheFilePath)\n\n\t\tcacheFile, err = os.Create(newCacheFilePath)\n\t\tif err != nil {\n\t\t\treturn \"\", \"\", fmt.Errorf(\"failed to create decompressed cache file: %w\", err)\n\t\t}\n\t\tdefer cacheFile.Close()\n\t\tcacheFilePath = newCacheFilePath\n\t}\n\n\t// Copy file content to cache\n\t_, err = io.Copy(cacheFile, reader)\n\tif err != nil {\n\t\treturn \"\", \"\", fmt.Errorf(\"failed to copy file to cache: %w\", err)\n\t}\n\n\t// For files that were decompressed from .gz, we need to detect the original content type\n\tvar contentType string\n\tif strings.HasSuffix(path, \".gz\") {\n\t\t// Original path was gzipped, detect content type of decompressed content\n\t\toriginalPath := strings.TrimSuffix(path, \".gz\")\n\t\text := filepath.Ext(originalPath)\n\n\t\t// First try to detect by original file extension\n\t\tcontentType, err = detectContentTypeFromExtension(ext)\n\t\tif err != nil || contentType == \"application/octet-stream\" {\n\t\t\t// Fallback: detect from decompressed content\n\t\t\tcontentType, err = detectContentType(cacheFilePath)\n\t\t\tif err != nil {\n\t\t\t\treturn \"\", \"\", fmt.Errorf(\"failed to detect content type: %w\", err)\n\t\t\t}\n\t\t}\n\t} else {\n\t\t// Regular file content type detection\n\t\tcontentType, err = detectContentType(cacheFilePath)\n\t\tif err != nil {\n\t\t\treturn \"\", \"\", fmt.Errorf(\"failed to detect content type: %w\", err)\n\t\t}\n\t}\n\n\treturn cacheFilePath, contentType, nil\n}\n\n// detectContentType detects content type based on file extension and content\nfunc detectContentType(filePath string) (string, error) {\n\t// First try to detect by file extension\n\text := strings.ToLower(filepath.Ext(filePath))\n\n\t// Common file extensions mapping\n\tswitch ext {\n\tcase \".txt\":\n\t\treturn \"text/plain\", nil\n\tcase \".html\", \".htm\":\n\t\treturn \"text/html\", nil\n\tcase \".css\":\n\t\treturn \"text/css\", nil\n\tcase \".js\":\n\t\treturn \"application/javascript\", nil\n\tcase \".json\":\n\t\treturn \"application/json\", nil\n\tcase \".xml\":\n\t\treturn \"application/xml\", nil\n\tcase \".jpg\", \".jpeg\":\n\t\treturn \"image/jpeg\", nil\n\tcase \".png\":\n\t\treturn \"image/png\", nil\n\tcase \".gif\":\n\t\treturn \"image/gif\", nil\n\tcase \".webp\":\n\t\treturn \"image/webp\", nil\n\tcase \".svg\":\n\t\treturn \"image/svg+xml\", nil\n\tcase \".pdf\":\n\t\treturn \"application/pdf\", nil\n\tcase \".doc\":\n\t\treturn \"application/msword\", nil\n\tcase \".docx\":\n\t\treturn \"application/vnd.openxmlformats-officedocument.wordprocessingml.document\", nil\n\tcase \".xls\":\n\t\treturn \"application/vnd.ms-excel\", nil\n\tcase \".xlsx\":\n\t\treturn \"application/vnd.openxmlformats-officedocument.spreadsheetml.sheet\", nil\n\tcase \".ppt\":\n\t\treturn \"application/vnd.ms-powerpoint\", nil\n\tcase \".pptx\":\n\t\treturn \"application/vnd.openxmlformats-officedocument.presentationml.presentation\", nil\n\tcase \".zip\":\n\t\treturn \"application/zip\", nil\n\tcase \".tar\":\n\t\treturn \"application/x-tar\", nil\n\tcase \".gz\":\n\t\treturn \"application/gzip\", nil\n\tcase \".mp3\":\n\t\treturn \"audio/mpeg\", nil\n\tcase \".wav\":\n\t\treturn \"audio/wav\", nil\n\tcase \".m4a\":\n\t\treturn \"audio/mp4\", nil\n\tcase \".ogg\":\n\t\treturn \"audio/ogg\", nil\n\tcase \".mp4\":\n\t\treturn \"video/mp4\", nil\n\tcase \".avi\":\n\t\treturn \"video/x-msvideo\", nil\n\tcase \".mov\":\n\t\treturn \"video/quicktime\", nil\n\tcase \".webm\":\n\t\treturn \"video/webm\", nil\n\tcase \".md\", \".mdx\":\n\t\treturn \"text/markdown\", nil\n\tcase \".yao\":\n\t\treturn \"application/yao\", nil\n\tcase \".csv\":\n\t\treturn \"text/csv\", nil\n\t}\n\n\t// Try to detect by MIME package\n\tif contentType := mime.TypeByExtension(ext); contentType != \"\" {\n\t\treturn contentType, nil\n\t}\n\n\t// Fallback: detect by reading file content\n\tfile, err := os.Open(filePath)\n\tif err != nil {\n\t\treturn \"application/octet-stream\", nil // Default fallback\n\t}\n\tdefer file.Close()\n\n\t// Read first 512 bytes for content detection\n\tbuffer := make([]byte, 512)\n\tn, err := file.Read(buffer)\n\tif err != nil && err != io.EOF {\n\t\treturn \"application/octet-stream\", nil\n\t}\n\n\t// Use http.DetectContentType to detect based on content\n\tcontentType := http.DetectContentType(buffer[:n])\n\treturn contentType, nil\n}\n\n// detectContentTypeFromExtension detects content type based only on file extension\nfunc detectContentTypeFromExtension(ext string) (string, error) {\n\text = strings.ToLower(ext)\n\n\t// Common file extensions mapping\n\tswitch ext {\n\tcase \".txt\":\n\t\treturn \"text/plain\", nil\n\tcase \".html\", \".htm\":\n\t\treturn \"text/html\", nil\n\tcase \".css\":\n\t\treturn \"text/css\", nil\n\tcase \".js\":\n\t\treturn \"application/javascript\", nil\n\tcase \".json\":\n\t\treturn \"application/json\", nil\n\tcase \".xml\":\n\t\treturn \"application/xml\", nil\n\tcase \".jpg\", \".jpeg\":\n\t\treturn \"image/jpeg\", nil\n\tcase \".png\":\n\t\treturn \"image/png\", nil\n\tcase \".gif\":\n\t\treturn \"image/gif\", nil\n\tcase \".webp\":\n\t\treturn \"image/webp\", nil\n\tcase \".svg\":\n\t\treturn \"image/svg+xml\", nil\n\tcase \".pdf\":\n\t\treturn \"application/pdf\", nil\n\tcase \".doc\":\n\t\treturn \"application/msword\", nil\n\tcase \".docx\":\n\t\treturn \"application/vnd.openxmlformats-officedocument.wordprocessingml.document\", nil\n\tcase \".xls\":\n\t\treturn \"application/vnd.ms-excel\", nil\n\tcase \".xlsx\":\n\t\treturn \"application/vnd.openxmlformats-officedocument.spreadsheetml.sheet\", nil\n\tcase \".ppt\":\n\t\treturn \"application/vnd.ms-powerpoint\", nil\n\tcase \".pptx\":\n\t\treturn \"application/vnd.openxmlformats-officedocument.presentationml.presentation\", nil\n\tcase \".zip\":\n\t\treturn \"application/zip\", nil\n\tcase \".tar\":\n\t\treturn \"application/x-tar\", nil\n\tcase \".mp3\":\n\t\treturn \"audio/mpeg\", nil\n\tcase \".wav\":\n\t\treturn \"audio/wav\", nil\n\tcase \".m4a\":\n\t\treturn \"audio/mp4\", nil\n\tcase \".ogg\":\n\t\treturn \"audio/ogg\", nil\n\tcase \".mp4\":\n\t\treturn \"video/mp4\", nil\n\tcase \".avi\":\n\t\treturn \"video/x-msvideo\", nil\n\tcase \".mov\":\n\t\treturn \"video/quicktime\", nil\n\tcase \".webm\":\n\t\treturn \"video/webm\", nil\n\tcase \".md\", \".mdx\":\n\t\treturn \"text/markdown\", nil\n\tcase \".yao\":\n\t\treturn \"application/yao\", nil\n\tcase \".csv\":\n\t\treturn \"text/csv\", nil\n\t}\n\n\t// Try to detect by MIME package\n\tif contentType := mime.TypeByExtension(ext); contentType != \"\" {\n\t\treturn contentType, nil\n\t}\n\n\t// Return default if not found\n\treturn \"application/octet-stream\", nil\n}\n" }, { "path": "attachment/s3/storage_test.go", "content": "package s3\n\nimport (\n\t\"bytes\"\n\t\"context\"\n\t\"io\"\n\t\"os\"\n\t\"path/filepath\"\n\t\"strings\"\n\t\"testing\"\n\t\"time\"\n\n\t\"compress/gzip\"\n\n\t\"github.com/google/uuid\"\n\t\"github.com/stretchr/testify/assert\"\n)\n\n// generateTestFileName generates a unique test filename with the given prefix and extension\nfunc generateTestFileName(prefix, ext string) string {\n\treturn prefix + \"-\" + uuid.New().String() + ext\n}\n\nfunc getS3Config() map[string]interface{} {\n\treturn map[string]interface{}{\n\t\t\"endpoint\": os.Getenv(\"S3_API\"),\n\t\t\"region\": \"auto\",\n\t\t\"key\": os.Getenv(\"S3_ACCESS_KEY\"),\n\t\t\"secret\": os.Getenv(\"S3_SECRET_KEY\"),\n\t\t\"bucket\": os.Getenv(\"S3_BUCKET\"),\n\t\t\"prefix\": \"attachment-test\",\n\t\t\"expiration\": 5 * time.Minute,\n\t\t\"compression\": true,\n\t}\n}\n\nfunc skipIfNoS3Config(t *testing.T) {\n\tif os.Getenv(\"S3_ACCESS_KEY\") == \"\" || os.Getenv(\"S3_SECRET_KEY\") == \"\" || os.Getenv(\"S3_BUCKET\") == \"\" {\n\t\tt.Skip(\"S3 configuration not available (set S3_ACCESS_KEY, S3_SECRET_KEY, S3_BUCKET environment variables)\")\n\t}\n}\n\nfunc TestS3Storage(t *testing.T) {\n\tt.Run(\"Create Storage\", func(t *testing.T) {\n\t\toptions := getS3Config()\n\n\t\tstorage, err := New(options)\n\t\tif os.Getenv(\"S3_ACCESS_KEY\") == \"\" || os.Getenv(\"S3_SECRET_KEY\") == \"\" || os.Getenv(\"S3_BUCKET\") == \"\" {\n\t\t\t// Should fail without credentials\n\t\t\tassert.Error(t, err)\n\t\t\treturn\n\t\t}\n\n\t\tassert.NoError(t, err)\n\t\tassert.NotNil(t, storage)\n\t\tif storage != nil {\n\t\t\tassert.Equal(t, os.Getenv(\"S3_API\"), storage.Endpoint)\n\t\t\tassert.Equal(t, \"auto\", storage.Region)\n\t\t\tassert.Equal(t, os.Getenv(\"S3_ACCESS_KEY\"), storage.Key)\n\t\t\tassert.Equal(t, os.Getenv(\"S3_SECRET_KEY\"), storage.Secret)\n\t\t\tassert.Equal(t, os.Getenv(\"S3_BUCKET\"), storage.Bucket)\n\t\t\tassert.Equal(t, \"attachment-test\", storage.prefix)\n\t\t\tassert.Equal(t, 5*time.Minute, storage.Expiration)\n\t\t\tassert.True(t, storage.compression)\n\t\t}\n\t})\n\n\tt.Run(\"Upload and Download Text File\", func(t *testing.T) {\n\t\tskipIfNoS3Config(t)\n\n\t\tstorage, err := New(getS3Config())\n\t\tassert.NoError(t, err)\n\n\t\tcontent := []byte(\"test content\")\n\t\treader := bytes.NewReader(content)\n\t\tfileID := generateTestFileName(\"upload-test\", \".txt\")\n\t\t_, err = storage.Upload(context.Background(), fileID, reader, \"text/plain\")\n\t\tassert.NoError(t, err)\n\t\tassert.NotEmpty(t, fileID)\n\n\t\t// Get presigned URL\n\t\turl := storage.URL(context.Background(), fileID)\n\t\tassert.NotEmpty(t, url)\n\t\tassert.Contains(t, url, \"X-Amz-Signature\")\n\t\tassert.Contains(t, url, \"X-Amz-Expires\")\n\n\t\t// Download\n\t\treader2, contentType, err := storage.Download(context.Background(), fileID)\n\t\tassert.NoError(t, err)\n\t\tassert.Contains(t, contentType, \"text/plain\")\n\n\t\tdownloaded, err := io.ReadAll(reader2)\n\t\tassert.NoError(t, err)\n\t\tassert.Equal(t, content, downloaded)\n\t\treader2.Close()\n\n\t\t// Clean up\n\t\tstorage.Delete(context.Background(), fileID)\n\t})\n\n\tt.Run(\"Chunked Upload\", func(t *testing.T) {\n\t\tskipIfNoS3Config(t)\n\n\t\tstorage, err := New(getS3Config())\n\t\tassert.NoError(t, err)\n\n\t\tfileID := generateTestFileName(\"test-chunked\", \".txt\")\n\t\tcontent1 := []byte(\"chunk1\")\n\t\tcontent2 := []byte(\"chunk2\")\n\n\t\t// Upload chunks\n\t\terr = storage.UploadChunk(context.Background(), fileID, 0, bytes.NewReader(content1), \"text/plain\")\n\t\tassert.NoError(t, err)\n\n\t\terr = storage.UploadChunk(context.Background(), fileID, 1, bytes.NewReader(content2), \"text/plain\")\n\t\tassert.NoError(t, err)\n\n\t\t// Merge chunks\n\t\terr = storage.MergeChunks(context.Background(), fileID, 2)\n\t\tassert.NoError(t, err)\n\n\t\t// Download and verify\n\t\treader, contentType, err := storage.Download(context.Background(), fileID)\n\t\tassert.NoError(t, err)\n\t\tassert.Equal(t, \"text/plain\", contentType)\n\n\t\tdownloaded, err := io.ReadAll(reader)\n\t\tassert.NoError(t, err)\n\t\tassert.Equal(t, append(content1, content2...), downloaded)\n\t\treader.Close()\n\n\t\t// Clean up\n\t\tstorage.Delete(context.Background(), fileID)\n\t})\n\n\tt.Run(\"File Operations\", func(t *testing.T) {\n\t\tskipIfNoS3Config(t)\n\n\t\tstorage, err := New(getS3Config())\n\t\tassert.NoError(t, err)\n\n\t\tfileID := generateTestFileName(\"test-ops\", \".txt\")\n\t\tcontent := []byte(\"test content\")\n\n\t\t// Upload file\n\t\t_, err = storage.Upload(context.Background(), fileID, bytes.NewReader(content), \"text/plain\")\n\t\tassert.NoError(t, err)\n\n\t\t// Check if file exists\n\t\texists := storage.Exists(context.Background(), fileID)\n\t\tassert.True(t, exists)\n\n\t\t// Read file\n\t\treader, err := storage.Reader(context.Background(), fileID)\n\t\tassert.NoError(t, err)\n\t\tdefer reader.Close()\n\n\t\tdata, err := io.ReadAll(reader)\n\t\tassert.NoError(t, err)\n\t\tassert.Equal(t, content, data)\n\n\t\t// Get file content directly\n\t\tdirectContent, err := storage.GetContent(context.Background(), fileID)\n\t\tassert.NoError(t, err)\n\t\tassert.Equal(t, content, directContent)\n\n\t\t// Delete file\n\t\terr = storage.Delete(context.Background(), fileID)\n\t\tassert.NoError(t, err)\n\n\t\t// Check if file no longer exists\n\t\texists = storage.Exists(context.Background(), fileID)\n\t\tassert.False(t, exists)\n\t})\n\n\tt.Run(\"Download Non-existent File\", func(t *testing.T) {\n\t\tskipIfNoS3Config(t)\n\n\t\tstorage, err := New(getS3Config())\n\t\tassert.NoError(t, err)\n\n\t\t// Use UUID for non-existent file to avoid any potential conflicts\n\t\tnonExistentFileID := generateTestFileName(\"non-existent\", \".txt\")\n\t\t_, _, err = storage.Download(context.Background(), nonExistentFileID)\n\t\tassert.Error(t, err)\n\t})\n\n\tt.Run(\"Invalid Configuration\", func(t *testing.T) {\n\t\t// Test with missing required fields\n\t\t_, err := New(map[string]interface{}{\n\t\t\t\"endpoint\": \"https://s3.amazonaws.com\",\n\t\t\t\"region\": \"us-east-1\",\n\t\t\t// Missing key and secret\n\t\t})\n\t\tassert.Error(t, err)\n\t\tassert.Contains(t, err.Error(), \"key and secret are required\")\n\n\t\t// Test with missing bucket\n\t\t_, err = New(map[string]interface{}{\n\t\t\t\"endpoint\": \"https://s3.amazonaws.com\",\n\t\t\t\"region\": \"us-east-1\",\n\t\t\t\"key\": \"test-key\",\n\t\t\t\"secret\": \"test-secret\",\n\t\t\t// Missing bucket\n\t\t})\n\t\tassert.Error(t, err)\n\t\tassert.Contains(t, err.Error(), \"bucket is required\")\n\t})\n\n\tt.Run(\"LocalPath\", func(t *testing.T) {\n\t\tskipIfNoS3Config(t)\n\n\t\t// Create storage with custom cache directory\n\t\ttempCacheDir, err := os.MkdirTemp(\"\", \"s3_cache_test\")\n\t\tassert.NoError(t, err)\n\t\tdefer os.RemoveAll(tempCacheDir)\n\n\t\tconfig := getS3Config()\n\t\tconfig[\"cache_dir\"] = tempCacheDir\n\n\t\tstorage, err := New(config)\n\t\tassert.NoError(t, err)\n\n\t\t// Test different file types\n\t\ttestFiles := []struct {\n\t\t\tname string\n\t\t\tcontent []byte\n\t\t\tcontentType string\n\t\t\texpectedCT string\n\t\t}{\n\t\t\t{\"localpath-test.txt\", []byte(\"Hello S3 World\"), \"text/plain\", \"text/plain\"},\n\t\t\t{\"localpath-test.json\", []byte(`{\"s3\": \"test\"}`), \"application/json\", \"application/json\"},\n\t\t\t{\"localpath-test.html\", []byte(\"S3 Test\"), \"text/html\", \"text/html\"},\n\t\t\t{\"localpath-test.csv\", []byte(\"s3,test\\nval1,val2\"), \"text/csv\", \"text/csv\"},\n\t\t\t{\"localpath-test.md\", []byte(\"# S3 Markdown\"), \"text/markdown\", \"text/markdown\"},\n\t\t\t{\"localpath-test.yao\", []byte(\"s3 yao content\"), \"application/yao\", \"application/yao\"},\n\t\t}\n\n\t\tfor _, tf := range testFiles {\n\t\t\t// Upload file to S3\n\t\t\tfileID := generateTestFileName(\"s3-localpath\", \"-\"+tf.name)\n\t\t\t_, err = storage.Upload(context.Background(), fileID, bytes.NewReader(tf.content), tf.contentType)\n\t\t\tassert.NoError(t, err, \"Failed to upload %s\", tf.name)\n\n\t\t\t// Get local path - first call should download to cache\n\t\t\tlocalPath1, detectedCT1, err := storage.LocalPath(context.Background(), fileID)\n\t\t\tassert.NoError(t, err, \"Failed to get local path for %s\", tf.name)\n\t\t\tassert.NotEmpty(t, localPath1, \"Local path should not be empty for %s\", tf.name)\n\t\t\tassert.Equal(t, tf.expectedCT, detectedCT1, \"Content type mismatch for %s\", tf.name)\n\n\t\t\t// Verify the path is absolute\n\t\t\tassert.True(t, filepath.IsAbs(localPath1), \"Path should be absolute for %s\", tf.name)\n\n\t\t\t// Verify the file exists at the returned path\n\t\t\t_, err = os.Stat(localPath1)\n\t\t\tassert.NoError(t, err, \"File should exist at local path for %s\", tf.name)\n\n\t\t\t// Verify file content\n\t\t\tfileContent, err := os.ReadFile(localPath1)\n\t\t\tassert.NoError(t, err, \"Failed to read file at local path for %s\", tf.name)\n\t\t\tassert.Equal(t, tf.content, fileContent, \"File content mismatch for %s\", tf.name)\n\n\t\t\t// Get local path again - should use cached version\n\t\t\tlocalPath2, detectedCT2, err := storage.LocalPath(context.Background(), fileID)\n\t\t\tassert.NoError(t, err, \"Failed to get cached local path for %s\", tf.name)\n\t\t\tassert.Equal(t, localPath1, localPath2, \"Cached path should be same as first call for %s\", tf.name)\n\t\t\tassert.Equal(t, detectedCT1, detectedCT2, \"Cached content type should be same as first call for %s\", tf.name)\n\n\t\t\t// Clean up from S3\n\t\t\tstorage.Delete(context.Background(), fileID)\n\t\t}\n\t})\n\n\tt.Run(\"LocalPath_GzippedFile\", func(t *testing.T) {\n\t\tskipIfNoS3Config(t)\n\n\t\t// Create storage with custom cache directory\n\t\ttempCacheDir, err := os.MkdirTemp(\"\", \"s3_cache_gzip_test\")\n\t\tassert.NoError(t, err)\n\t\tdefer os.RemoveAll(tempCacheDir)\n\n\t\tconfig := getS3Config()\n\t\tconfig[\"cache_dir\"] = tempCacheDir\n\n\t\tstorage, err := New(config)\n\t\tassert.NoError(t, err)\n\n\t\t// Create gzipped content\n\t\toriginalContent := []byte(\"This content will be gzipped and stored in S3\")\n\t\tvar gzipBuf bytes.Buffer\n\t\tgzipWriter := gzip.NewWriter(&gzipBuf)\n\t\t_, err = gzipWriter.Write(originalContent)\n\t\tassert.NoError(t, err)\n\t\tgzipWriter.Close()\n\n\t\t// Upload gzipped file\n\t\tfileID := generateTestFileName(\"gzipped\", \".txt.gz\")\n\t\t_, err = storage.Upload(context.Background(), fileID, bytes.NewReader(gzipBuf.Bytes()), \"text/plain\")\n\t\tassert.NoError(t, err)\n\n\t\t// Get local path - should decompress during download\n\t\tlocalPath, contentType, err := storage.LocalPath(context.Background(), fileID)\n\t\tassert.NoError(t, err)\n\t\tassert.NotEmpty(t, localPath)\n\n\t\t// Verify the file is decompressed in cache (path should not end with .gz)\n\t\tassert.False(t, strings.HasSuffix(localPath, \".gz\"), \"Cached file should be decompressed\")\n\n\t\t// Verify content is decompressed\n\t\tcachedContent, err := os.ReadFile(localPath)\n\t\tassert.NoError(t, err)\n\t\tassert.Equal(t, originalContent, cachedContent, \"Cached file should contain decompressed content\")\n\n\t\t// Verify content type\n\t\tassert.Equal(t, \"text/plain\", contentType)\n\n\t\t// Clean up\n\t\tstorage.Delete(context.Background(), fileID)\n\t})\n\n\tt.Run(\"LocalPath_NonExistentFile\", func(t *testing.T) {\n\t\tskipIfNoS3Config(t)\n\n\t\tstorage, err := New(getS3Config())\n\t\tassert.NoError(t, err)\n\n\t\t// Test with non-existent file\n\t\tnonExistentFileID := generateTestFileName(\"non-existent-localpath\", \".txt\")\n\t\t_, _, err = storage.LocalPath(context.Background(), nonExistentFileID)\n\t\tassert.Error(t, err)\n\t\tassert.Contains(t, err.Error(), \"failed to download file\")\n\t})\n\n\tt.Run(\"LocalPath_CustomCacheDir\", func(t *testing.T) {\n\t\tskipIfNoS3Config(t)\n\n\t\t// Create custom cache directory\n\t\tcustomCacheDir, err := os.MkdirTemp(\"\", \"custom_s3_cache\")\n\t\tassert.NoError(t, err)\n\t\tdefer os.RemoveAll(customCacheDir)\n\n\t\tconfig := getS3Config()\n\t\tconfig[\"cache_dir\"] = customCacheDir\n\n\t\tstorage, err := New(config)\n\t\tassert.NoError(t, err)\n\n\t\t// Verify cache directory is set correctly\n\t\tassert.Equal(t, customCacheDir, storage.CacheDir)\n\n\t\t// Upload a test file\n\t\tcontent := []byte(\"Custom cache directory test\")\n\t\tfileID := generateTestFileName(\"custom-cache\", \".txt\")\n\t\t_, err = storage.Upload(context.Background(), fileID, bytes.NewReader(content), \"text/plain\")\n\t\tassert.NoError(t, err)\n\n\t\t// Get local path\n\t\tlocalPath, contentType, err := storage.LocalPath(context.Background(), fileID)\n\t\tassert.NoError(t, err)\n\t\tassert.NotEmpty(t, localPath)\n\t\tassert.Equal(t, \"text/plain\", contentType)\n\n\t\t// Verify the file is cached in the custom directory\n\t\tassert.True(t, strings.HasPrefix(localPath, customCacheDir), \"File should be cached in custom directory\")\n\n\t\t// Clean up\n\t\tstorage.Delete(context.Background(), fileID)\n\t})\n}\n" }, { "path": "attachment/types.go", "content": "package attachment\n\nimport (\n\t\"context\"\n\t\"io\"\n\t\"mime/multipart\"\n\n\t\"github.com/yaoapp/gou/model\"\n\t\"github.com/yaoapp/gou/types\"\n)\n\n// FileManager defines the interface for file management operations.\n// This interface provides abstraction for file operations, making it easier to:\n// - Write unit tests with mock implementations\n// - Switch between different storage backends\n// - Maintain consistent API across different implementations\n//\n// Example usage:\n//\n//\tvar fileManager FileManager = manager // Manager implements FileManager\n//\tfile, err := fileManager.Upload(ctx, header, reader, options)\n//\tdata, err := fileManager.Read(ctx, file.ID)\ntype FileManager interface {\n\t// Upload uploads a file with optional chunked upload support\n\tUpload(ctx context.Context, fileheader *FileHeader, reader io.Reader, option UploadOption) (*File, error)\n\n\t// Download downloads a file by its ID\n\tDownload(ctx context.Context, fileID string) (*FileResponse, error)\n\n\t// Read reads a file content as bytes\n\tRead(ctx context.Context, fileID string) ([]byte, error)\n\n\t// ReadBase64 reads a file content as base64 encoded string\n\tReadBase64(ctx context.Context, fileID string) (string, error)\n\n\t// Info retrieves complete file information from database by file ID\n\tInfo(ctx context.Context, fileID string) (*File, error)\n\n\t// List retrieves files from database with pagination and filtering\n\tList(ctx context.Context, option ListOption) (*ListResult, error)\n\n\t// Exists checks if a file exists\n\tExists(ctx context.Context, fileID string) bool\n\n\t// Delete deletes a file\n\tDelete(ctx context.Context, fileID string) error\n\n\t// LocalPath gets the local path of the file\n\tLocalPath(ctx context.Context, fileID string) (string, string, error)\n\n\t// GetText retrieves the parsed text content for a file\n\t// By default returns preview (first 2000 chars), set fullContent=true for complete text\n\tGetText(ctx context.Context, fileID string, fullContent ...bool) (string, error)\n\n\t// SaveText saves the parsed text content for a file\n\t// Automatically saves both full content and preview\n\tSaveText(ctx context.Context, fileID string, text string) error\n}\n\n// File the file\ntype File struct {\n\tID string `json:\"file_id\"`\n\tUserPath string `json:\"user_path\"` // User-specified complete file path\n\tPath string `json:\"path\"` // Actual storage path\n\tBytes int `json:\"bytes\"`\n\tCreatedAt int `json:\"created_at\"`\n\tFilename string `json:\"filename\"`\n\tContentType string `json:\"content_type\"`\n\tStatus string `json:\"status\"` // uploading, uploaded, indexing, indexed, upload_failed, index_failed\n\n\t// Permission fields\n\tPublic bool `json:\"public,omitempty\"` // Whether this attachment is shared across all teams\n\tShare string `json:\"share,omitempty\"` // Attachment sharing scope: \"private\" or \"team\"\n\tYaoCreatedBy string `json:\"-\"` // User who created the attachment (not exposed in JSON)\n\tYaoTeamID string `json:\"-\"` // Team ID for team-based access control (not exposed in JSON)\n\tYaoTenantID string `json:\"-\"` // Tenant ID for multi-tenancy support (not exposed in JSON)\n}\n\n// FileResponse represents a file download response\ntype FileResponse struct {\n\tReader io.ReadCloser\n\tContentType string\n\tExtension string\n}\n\n// Attachment represents a file attachment\ntype Attachment struct {\n\tName string `json:\"name,omitempty\"`\n\tURL string `json:\"url,omitempty\"`\n\tDescription string `json:\"description,omitempty\"`\n\tType string `json:\"type,omitempty\"`\n\tContentType string `json:\"content_type,omitempty\"`\n\tBytes int64 `json:\"bytes,omitempty\"`\n\tCreatedAt int64 `json:\"created_at,omitempty\"`\n\tFileID string `json:\"file_id,omitempty\"`\n\tUserPath string `json:\"user_path,omitempty\"` // User-specified complete file path\n\tPath string `json:\"path,omitempty\"` // Actual storage path\n\tGroups []string `json:\"groups,omitempty\"`\n\tGzip bool `json:\"gzip,omitempty\"` // Gzip the file, Optional, default is false\n\tPublic bool `json:\"public,omitempty\"` // Whether this attachment is shared across all teams in the platform\n\tShare string `json:\"share,omitempty\"` // Attachment sharing scope: \"private\" or \"team\"\n\n\t// Yao custom fields for permission control\n\tYaoCreatedBy string `json:\"__yao_created_by,omitempty\"` // User who created the attachment\n\tYaoUpdatedBy string `json:\"__yao_updated_by,omitempty\"` // User who last updated the attachment\n\tYaoTeamID string `json:\"__yao_team_id,omitempty\"` // Team ID for team-based access control\n\tYaoTenantID string `json:\"__yao_tenant_id,omitempty\"` // Tenant ID for multi-tenancy support\n}\n\n// Manager the manager struct\ntype Manager struct {\n\tManagerOption\n\tName string // Manager name for identification\n\tstorage Storage\n\tmaxsize int64\n\tchunsize int64\n\tallowedTypes allowedType\n}\n\n// Storage the storage interface\ntype Storage interface {\n\tUpload(ctx context.Context, path string, reader io.Reader, contentType string) (string, error)\n\tUploadChunk(ctx context.Context, path string, chunkIndex int, reader io.Reader, contentType string) error\n\tMergeChunks(ctx context.Context, path string, totalChunks int) error\n\tDownload(ctx context.Context, path string) (io.ReadCloser, string, error)\n\tReader(ctx context.Context, path string) (io.ReadCloser, error)\n\tGetContent(ctx context.Context, path string) ([]byte, error)\n\tURL(ctx context.Context, path string) string\n\tExists(ctx context.Context, path string) bool\n\tDelete(ctx context.Context, path string) error\n\tLocalPath(ctx context.Context, path string) (string, string, error) // Returns absolute path and content type\n}\n\n// ManagerOption the manager option\ntype ManagerOption struct {\n\ttypes.MetaInfo\n\tMaxSize string `json:\"max_size,omitempty\" yaml:\"max_size,omitempty\"` // Max size of the file, Optional, default is 20M\n\tChunkSize string `json:\"chunk_size,omitempty\" yaml:\"chunk_size,omitempty\"` // Chunk size of the file, Optional, default is 2M\n\tAllowedTypes []string `json:\"allowed_types,omitempty\" yaml:\"allowed_types,omitempty\"` // Allowed types of the file, Optional, default is all\n\tGzip bool `json:\"gzip,omitempty\" yaml:\"gzip,omitempty\"` // Gzip the file, Optional, default is false\n\tDriver string `json:\"driver,omitempty\" yaml:\"driver,omitempty\"` // Driver, Optional, default is local\n\tOptions map[string]interface{} `json:\"options,omitempty\" yaml:\"options,omitempty\"` // Options, Optional\n}\n\ntype allowedType struct {\n\tmapping map[string]bool\n\twildcards []string // Wildcard patterns for file types (e.g., \"image/*\", \"text/*\")\n}\n\n// UploadOption the upload option\ntype UploadOption struct {\n\tCompressImage bool `json:\"compress_image,omitempty\" form:\"compress_image\"` // Compress the file, Optional, default is true\n\tCompressSize int `json:\"compress_size,omitempty\" form:\"compress_size\"` // Compress the file size, Optional, default is 1920, if compress_image is true, the file size will be compressed to the compress_size\n\tGzip bool `json:\"gzip,omitempty\" form:\"gzip\"` // Gzip the file, Optional, default is false\n\tOriginalFilename string `json:\"original_filename,omitempty\" form:\"original_filename\"` // Original filename sent separately to avoid encoding issues\n\tGroups []string `json:\"groups,omitempty\" form:\"groups\"` // Groups, Optional, default is empty, Multi-level groups like [\"user\", \"user123\", \"chat\", \"chat456\"]\n\tPublic bool `json:\"public,omitempty\" form:\"public\"` // Whether this attachment is shared across all teams in the platform\n\tShare string `json:\"share,omitempty\" form:\"share\"` // Attachment sharing scope: \"private\" or \"team\"\n\n\t// Yao custom fields for permission control\n\tYaoCreatedBy string `json:\"__yao_created_by,omitempty\" form:\"__yao_created_by\"` // User who created the attachment\n\tYaoUpdatedBy string `json:\"__yao_updated_by,omitempty\" form:\"__yao_updated_by\"` // User who last updated the attachment\n\tYaoTeamID string `json:\"__yao_team_id,omitempty\" form:\"__yao_team_id\"` // Team ID for team-based access control\n\tYaoTenantID string `json:\"__yao_tenant_id,omitempty\" form:\"__yao_tenant_id\"` // Tenant ID for multi-tenancy support\n}\n\n// ListOption defines options for listing files\ntype ListOption struct {\n\tPage int `json:\"page,omitempty\"` // Page number (1-based), default is 1\n\tPageSize int `json:\"page_size,omitempty\"` // Page size, default is 20\n\tFilters map[string]interface{} `json:\"filters,omitempty\"` // Filter conditions, e.g., {\"status\": \"uploaded\", \"content_type\": \"image/*\"}\n\tWheres []model.QueryWhere `json:\"wheres,omitempty\"` // Advanced where clauses for permission filtering\n\tOrderBy string `json:\"order_by,omitempty\"` // Order by field, e.g., \"created_at desc\", \"name asc\"\n\tSelect []string `json:\"select,omitempty\"` // Fields to select, empty means select all\n}\n\n// ListResult contains the paginated list result\ntype ListResult struct {\n\tFiles []*File `json:\"files\"` // List of files\n\tTotal int64 `json:\"total\"` // Total count\n\tPage int `json:\"page\"` // Current page\n\tPageSize int `json:\"page_size\"` // Page size\n\tTotalPages int `json:\"total_pages\"` // Total pages\n}\n\n// FileHeader the file header\ntype FileHeader struct {\n\t*multipart.FileHeader\n}\n" }, { "path": "audit/README.md", "content": "# Audit Log\n" }, { "path": "bin/yao-dev", "content": "#!/bin/bash\n\n# yao-dev - Run yao from source for real-time debugging\n# Uses go -C to compile from source while keeping current directory as app root\n\n# Get the real path of this script (resolve symlinks)\nSCRIPT_PATH=\"${BASH_SOURCE[0]}\"\nwhile [ -L \"$SCRIPT_PATH\" ]; do\n SCRIPT_DIR=\"$(cd \"$(dirname \"$SCRIPT_PATH\")\" && pwd)\"\n SCRIPT_PATH=\"$(readlink \"$SCRIPT_PATH\")\"\n # If the link is relative, resolve it relative to the directory\n [[ \"$SCRIPT_PATH\" != /* ]] && SCRIPT_PATH=\"$SCRIPT_DIR/$SCRIPT_PATH\"\ndone\nSCRIPT_DIR=\"$(cd \"$(dirname \"$SCRIPT_PATH\")\" && pwd)\"\n\n# YAO source directory is the parent of bin/\nYAO_SOURCE_DIR=\"$(dirname \"$SCRIPT_DIR\")\"\n\nif [ ! -d \"$YAO_SOURCE_DIR\" ]; then\n echo \"Error: yao source directory not found at $YAO_SOURCE_DIR\"\n exit 1\nfi\n\nif [ ! -f \"$YAO_SOURCE_DIR/go.mod\" ]; then\n echo \"Error: go.mod not found in $YAO_SOURCE_DIR, not a valid yao source directory\"\n exit 1\nfi\n\n# Find app root by looking for app.yao in current directory or parent directories\nfind_app_root() {\n local dir=\"$(pwd)\"\n while [ \"$dir\" != \"/\" ]; do\n if [ -f \"$dir/app.yao\" ] || [ -f \"$dir/app.json\" ] || [ -f \"$dir/app.jsonc\" ]; then\n echo \"$dir\"\n return 0\n fi\n dir=\"$(dirname \"$dir\")\"\n done\n # If not found, use current directory as fallback\n pwd\n}\n\n# Set YAO_ROOT to app root directory\nexport YAO_ROOT=\"$(find_app_root)\"\n\n# Convert relative file paths in arguments to absolute paths\n# This is needed because go -C changes the working directory\nARGS=()\nfor arg in \"$@\"; do\n # Check if it looks like a relative path\n if [[ \"$arg\" == ./* ]] || [[ \"$arg\" == ../* ]]; then\n # Always convert to absolute path based on current directory\n ARGS+=(\"$(pwd)/${arg#./}\")\n else\n ARGS+=(\"$arg\")\n fi\ndone\n\n# Use go -C to run from source directory while staying in current directory\nexec go -C \"$YAO_SOURCE_DIR\" run . \"${ARGS[@]}\"\n\n" }, { "path": "cert/cert.go", "content": "package cert\n\nimport (\n\t\"path/filepath\"\n\n\t\"github.com/yaoapp/gou/application\"\n\t\"github.com/yaoapp/gou/ssl\"\n\t\"github.com/yaoapp/yao/config\"\n\t\"github.com/yaoapp/yao/share\"\n)\n\n// Load 加载API\nfunc Load(cfg config.Config) error {\n\n\t// Ignore if the certs directory does not exist\n\texists, err := application.App.Exists(\"certs\")\n\tif err != nil {\n\t\treturn err\n\t}\n\n\tif !exists {\n\t\treturn nil\n\t}\n\n\texts := []string{\"*.pem\", \"*.key\", \"*.pub\"}\n\treturn application.App.Walk(\"certs\", func(root, file string, isdir bool) error {\n\t\tif isdir {\n\t\t\treturn nil\n\t\t}\n\t\t_, err := ssl.Load(file, share.ID(root, file)+filepath.Ext(file))\n\t\treturn err\n\t}, exts...)\n}\n" }, { "path": "cert/cert_test.go", "content": "package cert\n\nimport (\n\t\"testing\"\n\n\t\"github.com/stretchr/testify/assert\"\n\t\"github.com/yaoapp/gou/process\"\n\t\"github.com/yaoapp/gou/ssl\"\n\t\"github.com/yaoapp/yao/config\"\n\t\"github.com/yaoapp/yao/test\"\n)\n\nfunc TestLoad(t *testing.T) {\n\ttest.Prepare(t, config.Conf)\n\tdefer test.Clean()\n\n\tLoad(config.Conf)\n\tcheck(t)\n}\n\nfunc check(t *testing.T) {\n\tids := map[string]bool{}\n\tfor id := range ssl.Certificates {\n\t\tids[id] = true\n\t}\n\tassert.True(t, ids[\"cert.pem\"])\n\tassert.True(t, ids[\"cert.key\"])\n\tassert.True(t, ids[\"cert.pub\"])\n}\n\nfunc TestProcessSign(t *testing.T) {\n\tLoad(config.Conf)\n\targs := []interface{}{\"hello world\", \"cert.key\", \"SHA256\"}\n\tsignature, err := process.New(\"ssl.Sign\", args...).Exec()\n\tif err != nil {\n\t\tt.Fatal(err)\n\t}\n\tassert.Equal(t, \"EDHf3C9TXEk7y8LzIk5czLefXZyGxcMDVMcbNuBBegDkTqnPsRQnhFtNOgCdox8lI3MzLatwjoljoMY4Qk+sHGd5mAHMpiREa1gRFSVYpA2xvXZ3+KsfOHAdICQrfUdy59QaJGo6iGPNGG8PQOXHPTVNn6LMfryat9+f4l21DPAZiT0RyCUgFZE3/Qv8Z/6J4AsIXMSKZD6BGPPHUxGe7UBrXZvcR5dX25EiNjuH2OO38YJnDiTRVw14UI5fk/mQrwRdezj5tSKFCyHt912BZExXtkHISiYFNTZ/2RhOup5Xx6o3GvrEOdshrnN80Lwu1Aaju+lnZp13hDz4P6hU7w==\", signature)\n}\n\nfunc TestProcessVerify(t *testing.T) {\n\tLoad(config.Conf)\n\tsignature := \"EDHf3C9TXEk7y8LzIk5czLefXZyGxcMDVMcbNuBBegDkTqnPsRQnhFtNOgCdox8lI3MzLatwjoljoMY4Qk+sHGd5mAHMpiREa1gRFSVYpA2xvXZ3+KsfOHAdICQrfUdy59QaJGo6iGPNGG8PQOXHPTVNn6LMfryat9+f4l21DPAZiT0RyCUgFZE3/Qv8Z/6J4AsIXMSKZD6BGPPHUxGe7UBrXZvcR5dX25EiNjuH2OO38YJnDiTRVw14UI5fk/mQrwRdezj5tSKFCyHt912BZExXtkHISiYFNTZ/2RhOup5Xx6o3GvrEOdshrnN80Lwu1Aaju+lnZp13hDz4P6hU7w==\"\n\targs := []interface{}{\"hello world\", signature, \"cert.pem\", \"SHA256\"}\n\tres, err := process.New(\"ssl.Verify\", args...).Exec()\n\tif err != nil {\n\t\tt.Fatal(err)\n\t}\n\n\tassert.True(t, res.(bool))\n}\n" }, { "path": "cmd/README.md", "content": "# Yao CLI Commands\n\nThe Yao CLI provides a set of commands for managing, running, and testing Yao applications.\n\n## Installation\n\n```bash\n# Build from source\ngo build -o yao .\n\n# Or install via go install\ngo install github.com/yaoapp/yao@latest\n```\n\n## Global Flags\n\n| Flag | Short | Description |\n| -------- | ----- | ------------------------------- |\n| `--app` | `-a` | Application directory path |\n| `--file` | `-f` | Application package file (.yaz) |\n| `--key` | `-k` | Application license key |\n\n## Environment Variables\n\n| Variable | Description |\n| ---------- | -------------------------------------------- |\n| `YAO_ROOT` | Application root directory |\n| `YAO_LANG` | Language setting (e.g., `zh-CN` for Chinese) |\n\n## Commands\n\n### `yao start`\n\nStart the Yao application engine.\n\n```bash\n# Start in current directory\nyao start\n\n# Start with specific app directory\nyao start -a /path/to/app\n\n# Start in debug mode\nyao start --debug\n```\n\n**Flags:**\n\n| Flag | Description |\n| -------------------- | ----------------------------- |\n| `--debug` | Enable development/debug mode |\n| `--disable-watching` | Disable file watching |\n\n---\n\n### `yao run`\n\nExecute a Yao process.\n\n```bash\n# Run a process\nyao run models.user.Find 1\n\n# Run with JSON arguments\nyao run models.user.Create '::[{\"name\":\"John\",\"age\":30}]'\n\n# Run in silent mode (JSON output only)\nyao run -s models.user.Find 1\n```\n\n**Flags:**\n\n| Flag | Short | Description |\n| ---------- | ----- | ---------------------------------------- |\n| `--silent` | `-s` | Silent mode - output result as JSON only |\n\n**Argument Syntax:**\n\n- Regular arguments: `arg1 arg2`\n- JSON arguments: `'::[{\"key\":\"value\"}]'` (prefix with `::`)\n- Escaped `::`: `'\\::literal'`\n\n---\n\n### `yao migrate`\n\nUpdate database schema based on model definitions.\n\n```bash\n# Migrate all models\nyao migrate\n\n# Migrate specific model\nyao migrate -n user\n\n# Force migrate in production mode\nyao migrate --force\n\n# Reset (drop and recreate) tables\nyao migrate --reset\n```\n\n**Flags:**\n\n| Flag | Short | Description |\n| --------- | ----- | -------------------------------- |\n| `--name` | `-n` | Specific model name to migrate |\n| `--force` | | Force migrate in production mode |\n| `--reset` | | Drop tables before migration |\n\n---\n\n### `yao inspect`\n\nDisplay application configuration.\n\n```bash\nyao inspect\n```\n\n---\n\n### `yao version`\n\nShow Yao version information.\n\n```bash\n# Show version\nyao version\n\n# Show all version details\nyao version --all\n```\n\n**Flags:**\n\n| Flag | Description |\n| ------- | -------------------------------------------------------------------- |\n| `--all` | Print all version information (Go version, commit, build time, etc.) |\n\n---\n\n## Agent Commands\n\nCommands for testing and managing AI agents.\n\n### `yao agent test`\n\nTest an agent with input cases from a JSONL file, direct message, or script tests.\n\n```bash\n# Test with direct message (development mode)\nyao agent test -i \"Extract keywords from: AI and machine learning\" -n workers.system.keyword\n\n# Test with JSONL file\nyao agent test -i tests/inputs.jsonl\n\n# Test with custom output file\nyao agent test -i tests/inputs.jsonl -o report.html\n\n# Test with specific connector\nyao agent test -i tests/inputs.jsonl -c openai.gpt4\n\n# Stability testing (multiple runs)\nyao agent test -i tests/inputs.jsonl --runs 5\n\n# Parallel execution\nyao agent test -i tests/inputs.jsonl --parallel 4\n\n# Verbose output\nyao agent test -i tests/inputs.jsonl -v\n\n# Script tests (test agent handler scripts)\nyao agent test -i scripts.expense.setup -v\n\n# Script tests with test filtering\nyao agent test -i scripts.expense.setup --run \"TestSystemReady\"\n\n# Script tests with custom context\nyao agent test -i scripts.expense.setup --ctx tests/context.json -v\n```\n\n**Flags:**\n\n| Flag | Short | Description |\n| ------------- | ----- | ---------------------------------------------------------------- |\n| `--input` | `-i` | Input: JSONL file path, message, or script ID (required) |\n| `--output` | `-o` | Output file path (default: `output-{timestamp}.jsonl`) |\n| `--name` | `-n` | Agent ID (default: auto-detect from path) |\n| `--connector` | `-c` | Override default connector |\n| `--user` | `-u` | Test user ID (default: `test-user`) |\n| `--team` | `-t` | Test team ID (default: `test-team`) |\n| `--ctx` | | Path to context JSON file for custom authorization |\n| `--reporter` | `-r` | Reporter agent ID for custom report generation |\n| `--runs` | | Number of runs per test case for stability analysis (default: 1) |\n| `--run` | | Regex pattern to filter which tests to run |\n| `--timeout` | | Timeout per test case (default: `5m`) |\n| `--parallel` | | Number of parallel test cases (default: 1) |\n| `--verbose` | `-v` | Enable verbose output |\n| `--fail-fast` | | Stop on first failure |\n| `--app` | `-a` | Application directory |\n| `--env` | `-e` | Environment file |\n\n**Input Modes:**\n\n1. **Direct Message Mode**: For quick development/debugging\n\n ```bash\n yao agent test -i \"Hello world\" -n my.agent\n ```\n\n - Outputs result directly to stdout\n - No report file generated\n - Ideal for iterative development\n\n2. **File Mode**: For comprehensive testing\n\n ```bash\n yao agent test -i tests/inputs.jsonl\n ```\n\n - Reads test cases from JSONL file\n - Generates detailed report\n - Supports stability analysis\n\n3. **Script Test Mode**: For testing agent handler scripts\n ```bash\n yao agent test -i scripts.expense.setup -v\n ```\n - Tests TypeScript/JavaScript handler scripts (hooks, tools, setup functions)\n - Input format: `scripts..` (e.g., `scripts.expense.setup`)\n - Automatically discovers and runs all `Test*` functions\n - Uses Go-like testing interface with assertions\n\n**Script Test Function Signature:**\n\n```typescript\n// assistants/expense/src/setup_test.ts\nimport { SystemReady } from \"./setup\";\n\nexport function TestSystemReady(t: testing.T, ctx: agent.Context) {\n const result = SystemReady(ctx);\n t.assert.True(result.success, \"SystemReady should succeed\");\n t.assert.Equal(result.status, \"ready\", \"Status should be ready\");\n}\n```\n\n**Context JSON Format (for `--ctx` flag):**\n\n```json\n{\n \"authorized\": {\n \"sub\": \"user-12345\",\n \"client_id\": \"my-app\",\n \"user_id\": \"admin\",\n \"team_id\": \"team-001\",\n \"tenant_id\": \"acme-corp\",\n \"constraints\": {\n \"owner_only\": true,\n \"team_only\": false,\n \"extra\": { \"department\": \"engineering\" }\n }\n },\n \"metadata\": { \"request_id\": \"req-123\" },\n \"client\": { \"type\": \"web\", \"ip\": \"192.168.1.100\" },\n \"locale\": \"zh-cn\"\n}\n```\n\n**JSONL Input Format:**\n\n```jsonl\n{\"id\": \"T001\", \"input\": \"Simple text input\"}\n{\"id\": \"T002\", \"input\": {\"role\": \"user\", \"content\": \"Message with role\"}}\n{\"id\": \"T003\", \"input\": [{\"role\": \"system\", \"content\": \"System prompt\"}, {\"role\": \"user\", \"content\": \"User message\"}]}\n{\"id\": \"T004\", \"input\": \"Test with timeout\", \"timeout\": \"30s\"}\n{\"id\": \"T005\", \"input\": \"Skip this test\", \"skip\": true}\n{\"id\": \"T006\", \"input\": \"Test with specific user\", \"user\": \"alice\", \"team\": \"engineering\"}\n```\n\n**Output Formats:**\n\n| Extension | Format | Description |\n| --------- | -------- | -------------------------- |\n| `.jsonl` | JSONL | Streaming format (default) |\n| `.json` | JSON | Complete structured report |\n| `.md` | Markdown | Human-readable with tables |\n| `.html` | HTML | Interactive web report |\n\n**Agent Resolution:**\n\nThe agent is resolved in the following priority order:\n\n1. Explicit `-n` flag: `yao agent test -i msg -n my.agent`\n2. `YAO_ROOT` environment variable\n3. Auto-detect from input file path (traverses up to find `package.yao`)\n4. Auto-detect from current working directory\n\n---\n\n## SUI Commands\n\nSUI (Serverless UI) template engine commands.\n\n### `yao sui watch`\n\nAuto-build templates when files change.\n\n```bash\nyao sui watch [data]\n\n# Example\nyao sui watch default index '::{}'\n```\n\n### `yao sui build`\n\nBuild a template.\n\n```bash\nyao sui build [data]\n\n# Example\nyao sui build default index '::{}'\n\n# Debug mode\nyao sui build default index '::{}' --debug\n```\n\n### `yao sui trans`\n\nTranslate template content.\n\n```bash\nyao sui trans \n\n# With specific locales\nyao sui trans default index -l \"en-US,zh-CN,ja-JP\"\n```\n\n**SUI Flags:**\n\n| Flag | Short | Description |\n| ----------- | ----- | ----------------------------------------- |\n| `--data` | `-d` | Session data as JSON (prefix with `::`) |\n| `--debug` | `-D` | Enable debug mode |\n| `--locales` | `-l` | Locales for translation (comma-separated) |\n\n---\n\n## Examples\n\n### Development Workflow\n\n```bash\n# Start development server\nyao start --debug\n\n# Run a process\nyao run scripts.test.Hello \"World\"\n\n# Test an agent interactively\nyao agent test -i \"What is the weather today?\" -n assistant.weather\n\n# Watch and auto-build templates\nyao sui watch default home\n```\n\n### Testing Workflow\n\n```bash\n# Run comprehensive agent tests\nyao agent test -i tests/inputs.jsonl -o report.html -v\n\n# Run script tests for agent handlers\nyao agent test -i scripts.expense.setup -v\n\n# Run specific script tests with filtering\nyao agent test -i scripts.expense.setup --run \"TestSystem.*\" -v\n\n# Run script tests with custom context\nyao agent test -i scripts.expense.setup --ctx tests/context.json -v\n\n# Stability analysis (run each test 10 times)\nyao agent test -i tests/inputs.jsonl --runs 10 -o stability-report.json\n\n# Parallel testing with timeout\nyao agent test -i tests/inputs.jsonl --parallel 4 --timeout 2m\n\n# CI/CD integration\nyao agent test -i tests/inputs.jsonl -o results.jsonl && echo \"Tests passed\"\n```\n\n### Database Migration\n\n```bash\n# Migrate all models\nyao migrate\n\n# Migrate specific model with reset\nyao migrate -n user --reset --force\n```\n\n---\n\n## Exit Codes\n\n| Code | Description |\n| ---- | --------------------- |\n| 0 | Success |\n| 1 | Error or test failure |\n\n---\n\n## Directory Structure\n\n```\nmyapp/\n├── app.yao # Application configuration\n├── .env # Environment variables\n├── models/ # Data models\n├── apis/ # API definitions\n├── flows/ # Business flows\n├── scripts/ # JavaScript/TypeScript scripts\n├── assistants/ # AI agents\n│ └── my-agent/\n│ ├── package.yao # Agent configuration\n│ ├── prompts.yml # Agent prompts\n│ └── tests/\n│ └── inputs.jsonl # Test cases\n└── public/ # Static files\n```\n\n---\n\n## See Also\n\n- [Yao Documentation](https://yaoapps.com/docs)\n- [Agent Test Design](../agent/test/DESIGN.md)\n- [SUI Documentation](https://yaoapps.com/docs/sui)\n" }, { "path": "cmd/agent/add.go", "content": "package agent\n\nimport (\n\t\"fmt\"\n\t\"os\"\n\n\t\"github.com/spf13/cobra\"\n\t\"github.com/yaoapp/yao/config\"\n\t\"github.com/yaoapp/yao/registry\"\n\tagentmgr \"github.com/yaoapp/yao/registry/manager/agent\"\n)\n\nvar agentAddForce bool\n\n// AddCmd implements \"yao agent add @scope/name\"\nvar AddCmd = &cobra.Command{\n\tUse: \"add [package]\",\n\tShort: L(\"Install an assistant package from the registry\"),\n\tLong: L(\"Install an assistant package from the registry. Example: yao agent add @yao/keeper\"),\n\tArgs: cobra.ExactArgs(1),\n\tRun: func(cmd *cobra.Command, args []string) {\n\t\tBoot()\n\n\t\tpkgID := args[0]\n\t\tversion, _ := cmd.Flags().GetString(\"version\")\n\n\t\tclient := registry.New(config.Conf.Registry,\n\t\t\tregistry.WithAuth(\n\t\t\t\tos.Getenv(\"YAO_REGISTRY_USER\"),\n\t\t\t\tos.Getenv(\"YAO_REGISTRY_PASS\"),\n\t\t\t),\n\t\t)\n\n\t\tmgr := agentmgr.New(client, config.Conf.Root, nil)\n\t\tif err := mgr.Add(pkgID, agentmgr.AddOptions{\n\t\t\tVersion: version,\n\t\t\tForce: agentAddForce,\n\t\t}); err != nil {\n\t\t\tfmt.Fprintf(os.Stderr, \"Error: %v\\n\", err)\n\t\t\tos.Exit(1)\n\t\t}\n\t},\n}\n\nfunc init() {\n\tAddCmd.Flags().StringP(\"version\", \"v\", \"latest\", L(\"Package version or dist-tag\"))\n\tAddCmd.Flags().BoolVarP(&agentAddForce, \"force\", \"\", false, L(\"Force reinstall\"))\n\tAddCmd.PersistentFlags().StringVarP(&appPath, \"app\", \"a\", \"\", L(\"Application directory\"))\n\tAddCmd.PersistentFlags().StringVarP(&envFile, \"env\", \"e\", \"\", L(\"Environment file\"))\n}\n" }, { "path": "cmd/agent/agent.go", "content": "package agent\n\nimport (\n\t\"os\"\n\t\"path/filepath\"\n\n\t\"github.com/yaoapp/kun/exception\"\n\t\"github.com/yaoapp/yao/config\"\n)\n\nvar appPath string\nvar envFile string\n\nvar langs = map[string]string{\n\t\"Test an agent with input cases\": \"使用测试用例测试智能体\",\n\t\"Test an agent with input cases from JSONL file or direct message\": \"使用 JSONL 文件或直接消息测试智能体\",\n\t\"Application directory\": \"应用目录\",\n\t\"Environment file\": \"环境变量文件\",\n\t\"Input: JSONL file path or message (required)\": \"输入: JSONL 文件路径或消息 (必需)\",\n\t\"Path to output file (default: output-{timestamp}.jsonl)\": \"输出文件路径 (默认: output-{timestamp}.jsonl)\",\n\t\"Explicit agent ID (default: auto-detect)\": \"指定智能体 ID (默认: 自动检测)\",\n\t\"Override connector\": \"覆盖连接器\",\n\t\"Test user ID (default: test-user)\": \"测试用户 ID (默认: test-user)\",\n\t\"Test team ID (default: test-team)\": \"测试团队 ID (默认: test-team)\",\n\t\"Path to context JSON file for custom authorization\": \"自定义认证信息的 JSON 文件路径\",\n\t\"Reporter agent ID for custom report\": \"自定义报告生成器智能体 ID\",\n\t\"Number of runs for stability analysis\": \"稳定性分析的运行次数\",\n\t\"Regex pattern to filter which tests to run\": \"用于过滤测试的正则表达式\",\n\t\"Default timeout per test case\": \"每个测试用例的默认超时时间\",\n\t\"Number of parallel test cases\": \"并行测试用例数\",\n\t\"Verbose output\": \"详细输出\",\n\t\"Stop on first failure\": \"遇到第一个失败时停止\",\n\t\"Error: input is required (-i flag)\": \"错误: 需要输入 (-i 参数)\",\n\t\"Error: failed to get current directory\": \"错误: 获取当前目录失败\",\n\t\"Error: agent (-n) is required when using direct message input and not in an agent directory\": \"错误: 使用直接消息输入且不在智能体目录时需要指定 -n 参数\",\n\t\"Hint: Make sure you're in a Yao application directory or specify --app flag\": \"提示: 确保在 Yao 应用目录中或使用 --app 参数指定\",\n\t\"Error: invalid timeout format\": \"错误: 无效的超时格式\",\n\t// Registry commands\n\t\"Install an assistant package from the registry\": \"从注册中心安装助手包\",\n\t\"Update an installed assistant package\": \"更新已安装的助手包\",\n\t\"Push an assistant package to the registry\": \"推送助手包到注册中心\",\n\t\"Fork an assistant to a local scope\": \"Fork 一个助手到本地范围\",\n\t\"Package version or dist-tag\": \"包版本或 dist-tag\",\n\t\"Force reinstall\": \"强制重新安装\",\n\t\"Package version (required)\": \"包版本 (必填)\",\n\t\"Target version or dist-tag\": \"目标版本或 dist-tag\",\n\t// Extract command\n\t\"Extract test results to individual files for review\": \"提取测试结果到单独的文件供审查\",\n\t\"Extract test results from output JSONL file to individual Markdown or JSON files\": \"从输出 JSONL 文件中提取测试结果到单独的 Markdown 或 JSON 文件\",\n\t\"Output directory (default: same as input file)\": \"输出目录 (默认: 与输入文件相同)\",\n\t\"Output format: markdown, json\": \"输出格式: markdown, json\",\n}\n\n// L Language switch\nfunc L(words string) string {\n\tvar lang = os.Getenv(\"YAO_LANG\")\n\tif lang == \"\" {\n\t\treturn words\n\t}\n\n\tif trans, has := langs[words]; has {\n\t\treturn trans\n\t}\n\treturn words\n}\n\n// Boot sets the configuration\nfunc Boot() {\n\t// Use root from Init() unless appPath is explicitly specified\n\troot := config.Conf.Root\n\tif appPath != \"\" {\n\t\tr, err := filepath.Abs(appPath)\n\t\tif err != nil {\n\t\t\texception.New(\"Root error %s\", 500, err.Error()).Throw()\n\t\t}\n\t\troot = r\n\t}\n\n\t// Load .env file, preserving the correct root\n\tif envFile != \"\" {\n\t\tconfig.Conf = config.LoadFromWithRoot(envFile, root)\n\t} else {\n\t\tconfig.Conf = config.LoadFromWithRoot(filepath.Join(root, \".env\"), root)\n\t}\n\n\tconfig.ApplyMode()\n}\n" }, { "path": "cmd/agent/extract.go", "content": "package agent\n\nimport (\n\t\"fmt\"\n\t\"os\"\n\t\"path/filepath\"\n\n\t\"github.com/fatih/color\"\n\t\"github.com/spf13/cobra\"\n\t\"github.com/yaoapp/yao/agent/test\"\n)\n\n// Extract command flags\nvar (\n\textractOutput string\n\textractFormat string\n)\n\n// ExtractCmd is the agent extract command\nvar ExtractCmd = &cobra.Command{\n\tUse: \"extract \",\n\tShort: L(\"Extract test results to individual files for review\"),\n\tLong: L(\"Extract test results from output JSONL file to individual Markdown or JSON files\"),\n\tArgs: cobra.ExactArgs(1),\n\tRun: func(cmd *cobra.Command, args []string) {\n\t\tinputFile := args[0]\n\n\t\t// Resolve absolute path\n\t\tabsPath, err := filepath.Abs(inputFile)\n\t\tif err != nil {\n\t\t\tcolor.Red(\"Error: %s\\n\", err.Error())\n\t\t\tos.Exit(1)\n\t\t}\n\n\t\t// Check if file exists\n\t\tif _, err := os.Stat(absPath); os.IsNotExist(err) {\n\t\t\tcolor.Red(\"Error: file not found: %s\\n\", absPath)\n\t\t\tos.Exit(1)\n\t\t}\n\n\t\t// Build extract options\n\t\topts := &test.ExtractOptions{\n\t\t\tInputFile: absPath,\n\t\t\tOutputDir: extractOutput,\n\t\t\tFormat: extractFormat,\n\t\t}\n\n\t\t// Create extractor and run\n\t\textractor := test.NewExtractor(opts)\n\t\tfiles, err := extractor.Extract()\n\t\tif err != nil {\n\t\t\tcolor.Red(\"Error: %s\\n\", err.Error())\n\t\t\tos.Exit(1)\n\t\t}\n\n\t\t// Print results\n\t\tfmt.Println()\n\t\tcolor.New(color.FgGreen, color.Bold).Println(\"═══════════════════════════════════════════════════════════════\")\n\t\tcolor.New(color.FgGreen, color.Bold).Println(\" Extract Complete\")\n\t\tcolor.New(color.FgGreen, color.Bold).Println(\"═══════════════════════════════════════════════════════════════\")\n\t\tfmt.Println()\n\n\t\tfor _, file := range files {\n\t\t\tcolor.New(color.FgGreen).Printf(\"✓ \")\n\t\t\tfmt.Printf(\"Written: %s\\n\", filepath.Base(file))\n\t\t}\n\n\t\tfmt.Println()\n\t\tcolor.New(color.FgWhite).Printf(\" Total: \")\n\t\tcolor.New(color.FgCyan).Printf(\"%d files\\n\", len(files))\n\n\t\tif extractOutput != \"\" {\n\t\t\tcolor.New(color.FgWhite).Printf(\" Output: \")\n\t\t\tcolor.New(color.FgCyan).Printf(\"%s\\n\", extractOutput)\n\t\t} else {\n\t\t\tcolor.New(color.FgWhite).Printf(\" Output: \")\n\t\t\tcolor.New(color.FgCyan).Printf(\"%s\\n\", filepath.Dir(absPath))\n\t\t}\n\t\tfmt.Println()\n\t},\n}\n\nfunc init() {\n\t// Extract command flags\n\tExtractCmd.Flags().StringVarP(&extractOutput, \"output\", \"o\", \"\", L(\"Output directory (default: same as input file)\"))\n\tExtractCmd.Flags().StringVar(&extractFormat, \"format\", \"markdown\", L(\"Output format: markdown, json\"))\n}\n" }, { "path": "cmd/agent/fork.go", "content": "package agent\n\nimport (\n\t\"fmt\"\n\t\"os\"\n\n\t\"github.com/spf13/cobra\"\n\t\"github.com/yaoapp/yao/config\"\n\t\"github.com/yaoapp/yao/registry\"\n\tagentmgr \"github.com/yaoapp/yao/registry/manager/agent\"\n)\n\n// ForkCmd implements \"yao agent fork @scope/name [@target-scope]\"\nvar ForkCmd = &cobra.Command{\n\tUse: \"fork [package] [target-scope]\",\n\tShort: L(\"Fork an assistant to a local scope\"),\n\tLong: L(\"Fork an assistant for local modification. Example: yao agent fork @yao/keeper\"),\n\tArgs: cobra.RangeArgs(1, 2),\n\tRun: func(cmd *cobra.Command, args []string) {\n\t\tBoot()\n\n\t\tpkgID := args[0]\n\t\tvar targetScope string\n\t\tif len(args) > 1 {\n\t\t\ttargetScope = args[1]\n\t\t}\n\n\t\tclient := registry.New(config.Conf.Registry,\n\t\t\tregistry.WithAuth(\n\t\t\t\tos.Getenv(\"YAO_REGISTRY_USER\"),\n\t\t\t\tos.Getenv(\"YAO_REGISTRY_PASS\"),\n\t\t\t),\n\t\t)\n\n\t\tmgr := agentmgr.New(client, config.Conf.Root, nil)\n\t\tif err := mgr.Fork(pkgID, agentmgr.ForkOptions{\n\t\t\tTargetScope: targetScope,\n\t\t}); err != nil {\n\t\t\tfmt.Fprintf(os.Stderr, \"Error: %v\\n\", err)\n\t\t\tos.Exit(1)\n\t\t}\n\t},\n}\n\nfunc init() {\n\tForkCmd.PersistentFlags().StringVarP(&appPath, \"app\", \"a\", \"\", L(\"Application directory\"))\n\tForkCmd.PersistentFlags().StringVarP(&envFile, \"env\", \"e\", \"\", L(\"Environment file\"))\n}\n" }, { "path": "cmd/agent/push.go", "content": "package agent\n\nimport (\n\t\"fmt\"\n\t\"os\"\n\n\t\"github.com/spf13/cobra\"\n\t\"github.com/yaoapp/yao/config\"\n\t\"github.com/yaoapp/yao/registry\"\n\tagentmgr \"github.com/yaoapp/yao/registry/manager/agent\"\n)\n\n// PushCmd implements \"yao agent push scope.name --version x.y.z\"\nvar PushCmd = &cobra.Command{\n\tUse: \"push [yao-id]\",\n\tShort: L(\"Push an assistant package to the registry\"),\n\tLong: L(\"Package and push an assistant to the registry. Example: yao agent push max.keeper --version 1.0.0\"),\n\tArgs: cobra.ExactArgs(1),\n\tRun: func(cmd *cobra.Command, args []string) {\n\t\tBoot()\n\n\t\tyaoID := args[0]\n\t\tversion, _ := cmd.Flags().GetString(\"version\")\n\t\tforce, _ := cmd.Flags().GetBool(\"force\")\n\n\t\tclient := registry.New(config.Conf.Registry,\n\t\t\tregistry.WithAuth(\n\t\t\t\tos.Getenv(\"YAO_REGISTRY_USER\"),\n\t\t\t\tos.Getenv(\"YAO_REGISTRY_PASS\"),\n\t\t\t),\n\t\t)\n\n\t\tmgr := agentmgr.New(client, config.Conf.Root, nil)\n\t\tif err := mgr.Push(yaoID, agentmgr.PushOptions{\n\t\t\tVersion: version,\n\t\t\tForce: force,\n\t\t}); err != nil {\n\t\t\tfmt.Fprintf(os.Stderr, \"Error: %v\\n\", err)\n\t\t\tos.Exit(1)\n\t\t}\n\t},\n}\n\nfunc init() {\n\tPushCmd.Flags().StringP(\"version\", \"v\", \"\", L(\"Package version (required)\"))\n\tPushCmd.Flags().Bool(\"force\", false, L(\"Overwrite existing version\"))\n\tPushCmd.PersistentFlags().StringVarP(&appPath, \"app\", \"a\", \"\", L(\"Application directory\"))\n\tPushCmd.PersistentFlags().StringVarP(&envFile, \"env\", \"e\", \"\", L(\"Environment file\"))\n}\n" }, { "path": "cmd/agent/test.go", "content": "package agent\n\nimport (\n\t\"fmt\"\n\t\"os\"\n\t\"path/filepath\"\n\t\"time\"\n\n\t\"github.com/fatih/color\"\n\t\"github.com/spf13/cobra\"\n\t\"github.com/yaoapp/gou/plugin\"\n\t\"github.com/yaoapp/yao/agent\"\n\t\"github.com/yaoapp/yao/agent/test\"\n\t\"github.com/yaoapp/yao/config\"\n\t\"github.com/yaoapp/yao/engine\"\n\t\"github.com/yaoapp/yao/kb\"\n\t\"github.com/yaoapp/yao/share\"\n)\n\n// Test command flags\nvar (\n\ttestInput string\n\ttestOutput string\n\ttestAgent string\n\ttestConnector string\n\ttestUser string\n\ttestTeam string\n\ttestContext string // --ctx flag for custom context JSON file\n\ttestReporter string\n\ttestRuns int\n\ttestRun string // --run flag for test filtering (regex pattern)\n\ttestTimeout string\n\ttestParallel int\n\ttestVerbose bool\n\ttestFailFast bool\n\ttestBefore string // --before flag for global BeforeAll hook\n\ttestAfter string // --after flag for global AfterAll hook\n\ttestDryRun bool // --dry-run flag for generating tests without running\n\ttestSimulator string // --simulator flag for default simulator agent in dynamic mode\n)\n\n// TestCmd is the agent test command\nvar TestCmd = &cobra.Command{\n\tUse: \"test\",\n\tShort: L(\"Test an agent with input cases\"),\n\tLong: L(\"Test an agent with input cases from JSONL file or direct message\"),\n\tRun: func(cmd *cobra.Command, args []string) {\n\t\tdefer share.SessionStop()\n\t\tdefer plugin.KillAll()\n\n\t\t// Validate input\n\t\tif testInput == \"\" {\n\t\t\tcolor.Red(L(\"Error: input is required (-i flag)\") + \"\\n\")\n\t\t\tos.Exit(1)\n\t\t}\n\n\t\t// Detect input mode\n\t\tinputMode := test.DetectInputMode(testInput)\n\n\t\t// For message mode, agent must be specified or resolvable from cwd\n\t\tif inputMode == test.InputModeMessage && testAgent == \"\" {\n\t\t\t// Try to find app root from current directory\n\t\t\tcwd, err := os.Getwd()\n\t\t\tif err != nil {\n\t\t\t\tcolor.Red(L(\"Error: failed to get current directory\")+\": %s\\n\", err.Error())\n\t\t\t\tos.Exit(1)\n\t\t\t}\n\n\t\t\t// Try to find package.yao from cwd\n\t\t\tresolver := test.NewResolver()\n\t\t\t_, err = resolver.ResolveFromPath(cwd)\n\t\t\tif err != nil {\n\t\t\t\tcolor.Red(L(\"Error: agent (-n) is required when using direct message input and not in an agent directory\") + \"\\n\")\n\t\t\t\tos.Exit(1)\n\t\t\t}\n\t\t}\n\n\t\t// Find app root directory\n\t\t// Priority: -a flag > YAO_ROOT env > auto-detect from path\n\t\tvar err error\n\n\t\tif appPath == \"\" {\n\t\t\t// Check YAO_ROOT environment variable\n\t\t\tif yaoRoot := os.Getenv(\"YAO_ROOT\"); yaoRoot != \"\" {\n\t\t\t\tappPath = yaoRoot\n\t\t\t}\n\t\t}\n\n\t\tif appPath == \"\" {\n\t\t\t// Auto-detect from path\n\t\t\tif inputMode == test.InputModeFile {\n\t\t\t\t// For file mode, find app root from input file path\n\t\t\t\tappPath, err = findAppRoot(testInput)\n\t\t\t} else {\n\t\t\t\t// For message mode, find app root from current directory\n\t\t\t\tcwd, _ := os.Getwd()\n\t\t\t\tappPath, err = findAppRoot(cwd)\n\t\t\t}\n\n\t\t\tif err != nil {\n\t\t\t\tcolor.Red(\"Error: %s\\n\", err.Error())\n\t\t\t\tcolor.Yellow(L(\"Hint: Make sure you're in a Yao application directory or specify --app flag\") + \"\\n\")\n\t\t\t\tos.Exit(1)\n\t\t\t}\n\t\t}\n\n\t\t// Boot the application\n\t\tBoot()\n\n\t\t// Set Runtime Mode\n\t\tconfig.Conf.Runtime.Mode = \"standard\"\n\t\tcfg := config.Conf\n\t\tcfg.Session.IsCLI = true\n\n\t\t// Load engine\n\t\t_, err = engine.Load(cfg, engine.LoadOption{Action: \"agent-test\"})\n\t\tif err != nil {\n\t\t\tcolor.Red(\"Engine: %s\\n\", err.Error())\n\t\t\tos.Exit(1)\n\t\t}\n\n\t\t// Load KB (required for agent KB features)\n\t\t_, err = kb.Load(cfg)\n\t\tif err != nil {\n\t\t\tcolor.Red(\"KB: %s\\n\", err.Error())\n\t\t\tos.Exit(1)\n\t\t}\n\n\t\t// Load agent\n\t\terr = agent.Load(cfg)\n\t\tif err != nil {\n\t\t\tcolor.Red(\"Agent: %s\\n\", err.Error())\n\t\t\tos.Exit(1)\n\t\t}\n\n\t\t// Parse timeout\n\t\ttimeout := 5 * time.Minute\n\t\tif testTimeout != \"\" {\n\t\t\td, err := time.ParseDuration(testTimeout)\n\t\t\tif err != nil {\n\t\t\t\tcolor.Red(L(\"Error: invalid timeout format\")+\": %s\\n\", testTimeout)\n\t\t\t\tos.Exit(1)\n\t\t\t}\n\t\t\ttimeout = d\n\t\t}\n\n\t\t// Build test options\n\t\topts := &test.Options{\n\t\t\tInput: testInput,\n\t\t\tInputMode: inputMode,\n\t\t\tOutputFile: testOutput,\n\t\t\tAgentID: testAgent,\n\t\t\tConnector: testConnector,\n\t\t\tUserID: testUser,\n\t\t\tTeamID: testTeam,\n\t\t\tContextFile: testContext,\n\t\t\tReporterID: testReporter,\n\t\t\tRuns: testRuns,\n\t\t\tRun: testRun,\n\t\t\tTimeout: timeout,\n\t\t\tParallel: testParallel,\n\t\t\tVerbose: testVerbose,\n\t\t\tFailFast: testFailFast,\n\t\t\tBeforeAll: testBefore,\n\t\t\tAfterAll: testAfter,\n\t\t\tDryRun: testDryRun,\n\t\t\tSimulator: testSimulator,\n\t\t}\n\n\t\t// Merge with defaults\n\t\topts = test.MergeOptions(opts, test.DefaultOptions())\n\n\t\t// Resolve output path (only for file mode, direct message mode outputs to stdout)\n\t\tif inputMode == test.InputModeFile {\n\t\t\topts.OutputFile = test.ResolveOutputPath(opts)\n\t\t}\n\n\t\t// Run tests\n\t\trunner := test.NewRunner(opts)\n\t\treport, err := runner.Run()\n\t\tif err != nil {\n\t\t\tcolor.Red(\"Error: %s\\n\", err.Error())\n\t\t\tos.Exit(1)\n\t\t}\n\n\t\t// Exit with appropriate code\n\t\tif report.HasFailures() {\n\t\t\tos.Exit(1)\n\t\t}\n\t},\n}\n\n// findAppRoot finds the Yao application root directory by looking for app.yao\n// It traverses up from the given path until it finds app.yao or reaches the filesystem root\nfunc findAppRoot(startPath string) (string, error) {\n\t// Get absolute path\n\tabsPath, err := filepath.Abs(startPath)\n\tif err != nil {\n\t\treturn \"\", fmt.Errorf(\"failed to get absolute path: %w\", err)\n\t}\n\n\t// If it's a file, start from its directory\n\tinfo, err := os.Stat(absPath)\n\tif err != nil {\n\t\treturn \"\", fmt.Errorf(\"path not found: %s\", absPath)\n\t}\n\n\tvar dir string\n\tif info.IsDir() {\n\t\tdir = absPath\n\t} else {\n\t\tdir = filepath.Dir(absPath)\n\t}\n\n\t// Traverse up to find app.yao\n\tfor {\n\t\t// Check for app.yao, app.json, or app.jsonc\n\t\tfor _, appFile := range []string{\"app.yao\", \"app.json\", \"app.jsonc\"} {\n\t\t\tappFilePath := filepath.Join(dir, appFile)\n\t\t\tif _, err := os.Stat(appFilePath); err == nil {\n\t\t\t\treturn dir, nil\n\t\t\t}\n\t\t}\n\n\t\t// Move to parent directory\n\t\tparent := filepath.Dir(dir)\n\t\tif parent == dir {\n\t\t\t// Reached root, no app.yao found\n\t\t\tbreak\n\t\t}\n\t\tdir = parent\n\t}\n\n\treturn \"\", fmt.Errorf(\"no app.yao found in path hierarchy of %s\", startPath)\n}\n\nfunc init() {\n\t// Test command flags\n\tTestCmd.Flags().StringVarP(&appPath, \"app\", \"a\", \"\", L(\"Application directory\"))\n\tTestCmd.Flags().StringVarP(&envFile, \"env\", \"e\", \"\", L(\"Environment file\"))\n\tTestCmd.Flags().StringVarP(&testInput, \"input\", \"i\", \"\", L(\"Input: JSONL file path or message (required)\"))\n\tTestCmd.Flags().StringVarP(&testOutput, \"output\", \"o\", \"\", L(\"Path to output file (default: output-{timestamp}.jsonl)\"))\n\tTestCmd.Flags().StringVarP(&testAgent, \"name\", \"n\", \"\", L(\"Explicit agent ID (default: auto-detect)\"))\n\tTestCmd.Flags().StringVarP(&testConnector, \"connector\", \"c\", \"\", L(\"Override connector\"))\n\tTestCmd.Flags().StringVarP(&testUser, \"user\", \"u\", \"\", L(\"Test user ID (default: test-user)\"))\n\tTestCmd.Flags().StringVarP(&testTeam, \"team\", \"t\", \"\", L(\"Test team ID (default: test-team)\"))\n\tTestCmd.Flags().StringVar(&testContext, \"ctx\", \"\", L(\"Path to context JSON file for custom authorization\"))\n\tTestCmd.Flags().StringVarP(&testReporter, \"reporter\", \"r\", \"\", L(\"Reporter agent ID for custom report\"))\n\tTestCmd.Flags().IntVar(&testRuns, \"runs\", 1, L(\"Number of runs for stability analysis\"))\n\tTestCmd.Flags().StringVar(&testRun, \"run\", \"\", L(\"Regex pattern to filter which tests to run\"))\n\tTestCmd.Flags().StringVar(&testTimeout, \"timeout\", \"5m\", L(\"Default timeout per test case\"))\n\tTestCmd.Flags().IntVar(&testParallel, \"parallel\", 1, L(\"Number of parallel test cases\"))\n\tTestCmd.Flags().BoolVarP(&testVerbose, \"verbose\", \"v\", false, L(\"Verbose output\"))\n\tTestCmd.Flags().BoolVar(&testFailFast, \"fail-fast\", false, L(\"Stop on first failure\"))\n\tTestCmd.Flags().StringVar(&testBefore, \"before\", \"\", L(\"Global BeforeAll hook (e.g., env_test.BeforeAll)\"))\n\tTestCmd.Flags().StringVar(&testAfter, \"after\", \"\", L(\"Global AfterAll hook (e.g., env_test.AfterAll)\"))\n\tTestCmd.Flags().BoolVar(&testDryRun, \"dry-run\", false, L(\"Generate test cases without running them\"))\n\tTestCmd.Flags().StringVar(&testSimulator, \"simulator\", \"\", L(\"Default simulator agent for dynamic mode (e.g., tests.simulator-agent)\"))\n\n\t// Mark input as required\n\tTestCmd.MarkFlagRequired(\"input\")\n}\n" }, { "path": "cmd/agent/update.go", "content": "package agent\n\nimport (\n\t\"fmt\"\n\t\"os\"\n\n\t\"github.com/spf13/cobra\"\n\t\"github.com/yaoapp/yao/config\"\n\t\"github.com/yaoapp/yao/registry\"\n\tagentmgr \"github.com/yaoapp/yao/registry/manager/agent\"\n)\n\n// UpdateCmd implements \"yao agent update @scope/name\"\nvar UpdateCmd = &cobra.Command{\n\tUse: \"update [package]\",\n\tShort: L(\"Update an installed assistant package\"),\n\tLong: L(\"Update an installed assistant to a newer version. Example: yao agent update @yao/keeper\"),\n\tArgs: cobra.ExactArgs(1),\n\tRun: func(cmd *cobra.Command, args []string) {\n\t\tBoot()\n\n\t\tpkgID := args[0]\n\t\tversion, _ := cmd.Flags().GetString(\"version\")\n\n\t\tclient := registry.New(config.Conf.Registry,\n\t\t\tregistry.WithAuth(\n\t\t\t\tos.Getenv(\"YAO_REGISTRY_USER\"),\n\t\t\t\tos.Getenv(\"YAO_REGISTRY_PASS\"),\n\t\t\t),\n\t\t)\n\n\t\tmgr := agentmgr.New(client, config.Conf.Root, nil)\n\t\tif err := mgr.Update(pkgID, agentmgr.UpdateOptions{\n\t\t\tVersion: version,\n\t\t}); err != nil {\n\t\t\tfmt.Fprintf(os.Stderr, \"Error: %v\\n\", err)\n\t\t\tos.Exit(1)\n\t\t}\n\t},\n}\n\nfunc init() {\n\tUpdateCmd.Flags().StringP(\"version\", \"v\", \"latest\", L(\"Target version or dist-tag\"))\n\tUpdateCmd.PersistentFlags().StringVarP(&appPath, \"app\", \"a\", \"\", L(\"Application directory\"))\n\tUpdateCmd.PersistentFlags().StringVarP(&envFile, \"env\", \"e\", \"\", L(\"Environment file\"))\n}\n" }, { "path": "cmd/ci-token/main.go", "content": "//go:build ci\n\npackage main\n\nimport (\n\t\"flag\"\n\t\"fmt\"\n\t\"os\"\n\t\"path/filepath\"\n\t\"time\"\n\n\t\"github.com/yaoapp/yao/config\"\n\t\"github.com/yaoapp/yao/engine\"\n\t\"github.com/yaoapp/yao/openapi/oauth\"\n)\n\nfunc main() {\n\tappPath := flag.String(\"app\", envOr(\"YAO_CI_APP_PATH\", \".\"), \"Yao application directory\")\n\tclientID := flag.String(\"client-id\", envOr(\"YAO_CI_OAUTH_CLIENT_ID\", \"ci-tai\"), \"OAuth client ID embedded in token\")\n\tsubject := flag.String(\"subject\", envOr(\"YAO_CI_OAUTH_SUBJECT\", \"ci-tai\"), \"JWT subject claim\")\n\tscope := flag.String(\"scope\", envOr(\"YAO_CI_OAUTH_SCOPE\", \"tai:tunnel\"), \"Token scope (space-separated)\")\n\tttl := flag.String(\"ttl\", envOr(\"YAO_CI_OAUTH_TTL\", \"24h\"), \"Token TTL (e.g. 1h, 24h, 168h)\")\n\tuserID := flag.String(\"user-id\", envOr(\"YAO_CI_OAUTH_USER_ID\", \"\"), \"User ID claim\")\n\tteamID := flag.String(\"team-id\", envOr(\"YAO_CI_OAUTH_TEAM_ID\", \"\"), \"Team ID claim\")\n\tflag.Parse()\n\n\troot, err := filepath.Abs(*appPath)\n\tif err != nil {\n\t\tfmt.Fprintf(os.Stderr, \"ci-token: invalid app path: %v\\n\", err)\n\t\tos.Exit(1)\n\t}\n\n\tif err := os.Chdir(root); err != nil {\n\t\tfmt.Fprintf(os.Stderr, \"ci-token: chdir %s: %v\\n\", root, err)\n\t\tos.Exit(1)\n\t}\n\n\tsavedStdout := os.Stdout\n\tos.Stdout, _ = os.Open(os.DevNull)\n\n\tconfig.Conf = config.LoadFrom(filepath.Join(root, \".env\"))\n\tconfig.Conf.Root = root\n\n\tcfg := config.Conf\n\tcfg.Session.IsCLI = true\n\n\twarnings, err := engine.Load(cfg, engine.LoadOption{Action: \"run\"})\n\tos.Stdout = savedStdout\n\tif err != nil {\n\t\tfmt.Fprintf(os.Stderr, \"ci-token: engine.Load failed: %v\\n\", err)\n\t\tos.Exit(1)\n\t}\n\tfor _, w := range warnings {\n\t\tfmt.Fprintf(os.Stderr, \"ci-token: warning [%s]: %v\\n\", w.Widget, w.Error)\n\t}\n\n\tif oauth.OAuth == nil {\n\t\tfmt.Fprintln(os.Stderr, \"ci-token: oauth service not initialized (openapi.Load may have failed)\")\n\t\tos.Exit(1)\n\t}\n\n\tdur, err := time.ParseDuration(*ttl)\n\tif err != nil {\n\t\tfmt.Fprintf(os.Stderr, \"ci-token: invalid --ttl %q: %v\\n\", *ttl, err)\n\t\tos.Exit(1)\n\t}\n\texpiresIn := int(dur.Seconds())\n\n\textraClaims := map[string]interface{}{}\n\tif *userID != \"\" {\n\t\textraClaims[\"user_id\"] = *userID\n\t}\n\tif *teamID != \"\" {\n\t\textraClaims[\"team_id\"] = *teamID\n\t}\n\n\ttoken, err := oauth.OAuth.MakeAccessToken(*clientID, *scope, *subject, expiresIn, extraClaims)\n\tif err != nil {\n\t\tfmt.Fprintf(os.Stderr, \"ci-token: MakeAccessToken failed: %v\\n\", err)\n\t\tos.Exit(1)\n\t}\n\n\tfmt.Print(token)\n}\n\nfunc envOr(key, fallback string) string {\n\tif v := os.Getenv(key); v != \"\" {\n\t\treturn v\n\t}\n\treturn fallback\n}\n" }, { "path": "cmd/credential.go", "content": "package cmd\n\nimport (\n\t\"encoding/base64\"\n\t\"encoding/json\"\n\t\"fmt\"\n\t\"os\"\n\t\"path/filepath\"\n\t\"time\"\n)\n\n// Credential represents the stored OAuth credential for gRPC mode.\ntype Credential struct {\n\tServer string `json:\"server\"`\n\tGRPCAddr string `json:\"grpc_addr,omitempty\"`\n\tAccessToken string `json:\"access_token\"`\n\tRefreshToken string `json:\"refresh_token,omitempty\"`\n\tScope string `json:\"scope,omitempty\"`\n\tUser string `json:\"user,omitempty\"`\n\tExpiresAt string `json:\"expires_at,omitempty\"`\n}\n\n// Expired returns true if the credential has an expires_at in the past.\nfunc (c *Credential) Expired() bool {\n\tif c.ExpiresAt == \"\" {\n\t\treturn false\n\t}\n\tt, err := time.Parse(time.RFC3339, c.ExpiresAt)\n\tif err != nil {\n\t\treturn false\n\t}\n\treturn time.Now().After(t)\n}\n\nfunc credentialPath() (string, error) {\n\thome, err := os.UserHomeDir()\n\tif err != nil {\n\t\treturn \"\", fmt.Errorf(\"cannot determine home directory: %w\", err)\n\t}\n\treturn filepath.Join(home, \".yao\", \"credentials\"), nil\n}\n\n// LoadCredential reads and decodes ~/.yao/credentials. Returns nil if the file\n// does not exist.\nfunc LoadCredential() (*Credential, error) {\n\tpath, err := credentialPath()\n\tif err != nil {\n\t\treturn nil, err\n\t}\n\traw, err := os.ReadFile(path)\n\tif os.IsNotExist(err) {\n\t\treturn nil, nil\n\t}\n\tif err != nil {\n\t\treturn nil, fmt.Errorf(\"read credentials: %w\", err)\n\t}\n\n\tdecoded, err := base64.StdEncoding.DecodeString(string(raw))\n\tif err != nil {\n\t\treturn nil, fmt.Errorf(\"decode credentials: %w\", err)\n\t}\n\n\tvar cred Credential\n\tif err := json.Unmarshal(decoded, &cred); err != nil {\n\t\treturn nil, fmt.Errorf(\"unmarshal credentials: %w\", err)\n\t}\n\treturn &cred, nil\n}\n\n// LoadCredentialFrom reads and decodes a credential file from a custom path.\nfunc LoadCredentialFrom(path string) (*Credential, error) {\n\traw, err := os.ReadFile(path)\n\tif err != nil {\n\t\treturn nil, fmt.Errorf(\"read credentials from %s: %w\", path, err)\n\t}\n\tdecoded, err := base64.StdEncoding.DecodeString(string(raw))\n\tif err != nil {\n\t\treturn nil, fmt.Errorf(\"decode credentials: %w\", err)\n\t}\n\tvar cred Credential\n\tif err := json.Unmarshal(decoded, &cred); err != nil {\n\t\treturn nil, fmt.Errorf(\"unmarshal credentials: %w\", err)\n\t}\n\treturn &cred, nil\n}\n\n// SaveCredential encodes and writes the credential to ~/.yao/credentials.\nfunc SaveCredential(cred *Credential) error {\n\tpath, err := credentialPath()\n\tif err != nil {\n\t\treturn err\n\t}\n\tdir := filepath.Dir(path)\n\tif err := os.MkdirAll(dir, 0700); err != nil {\n\t\treturn fmt.Errorf(\"create directory %s: %w\", dir, err)\n\t}\n\tdata, err := json.Marshal(cred)\n\tif err != nil {\n\t\treturn fmt.Errorf(\"marshal credentials: %w\", err)\n\t}\n\tencoded := base64.StdEncoding.EncodeToString(data)\n\tif err := os.WriteFile(path, []byte(encoded), 0600); err != nil {\n\t\treturn fmt.Errorf(\"write credentials: %w\", err)\n\t}\n\treturn nil\n}\n\n// RemoveCredential deletes ~/.yao/credentials.\nfunc RemoveCredential() error {\n\tpath, err := credentialPath()\n\tif err != nil {\n\t\treturn err\n\t}\n\tif err := os.Remove(path); err != nil && !os.IsNotExist(err) {\n\t\treturn fmt.Errorf(\"remove credentials: %w\", err)\n\t}\n\treturn nil\n}\n" }, { "path": "cmd/dump.go", "content": "package cmd\n\nimport (\n\t\"archive/zip\"\n\t\"errors\"\n\t\"fmt\"\n\t\"io/ioutil\"\n\t\"os\"\n\t\"path/filepath\"\n\t\"strings\"\n\t\"time\"\n\n\t\"github.com/fatih/color\"\n\t\"github.com/spf13/cobra\"\n\t\"github.com/yaoapp/gou/model\"\n\t\"github.com/yaoapp/kun/exception\"\n\t\"github.com/yaoapp/yao/config\"\n\t\"github.com/yaoapp/yao/engine\"\n)\n\nvar dumpModel string\nvar dumpCmd = &cobra.Command{\n\tUse: \"dump\",\n\tShort: L(\"Dump the application data\"),\n\tLong: L(\"Dump the application data\"),\n\tRun: func(cmd *cobra.Command, args []string) {\n\t\tdefer func() {\n\t\t\terr := exception.Catch(recover())\n\t\t\tif err != nil {\n\t\t\t\tfmt.Println(color.RedString(L(\"Fatal: %s\"), err.Error()))\n\t\t\t}\n\t\t}()\n\n\t\tBoot()\n\n\t\tpath, err := filepath.Abs(\".\")\n\t\tif err != nil {\n\t\t\tfmt.Println(color.RedString(L(\"Fatal: %s\"), err.Error()))\n\t\t\tos.Exit(1)\n\t\t}\n\n\t\toutput := filepath.Join(fmt.Sprintf(\"%s-%s.zip\", filepath.Base(path), time.Now().Format(\"20060102150405\")))\n\t\tif len(args) > 0 {\n\t\t\toutput = args[0]\n\t\t}\n\n\t\toutput, err = filepath.Abs(output)\n\t\tif err != nil {\n\t\t\tfmt.Println(color.RedString(L(\"Fatal: %s\"), err.Error()))\n\t\t\tos.Exit(1)\n\t\t}\n\n\t\t_, err = os.Stat(output)\n\t\tif !errors.Is(err, os.ErrNotExist) {\n\t\t\tfmt.Println(color.RedString(\"%s exists\", output))\n\t\t\tos.Exit(1)\n\t\t}\n\n\t\t// Load model\n\t\tloadWarnings, err := engine.Load(config.Conf, engine.LoadOption{Action: \"dump\"})\n\t\tif err != nil {\n\t\t\tfmt.Println(color.RedString(L(\"Fatal: %s\"), err.Error()))\n\t\t\tos.Exit(1)\n\t\t}\n\n\t\tif len(loadWarnings) > 0 {\n\t\t\tfor _, warning := range loadWarnings {\n\t\t\t\tfmt.Println(color.YellowString(\"[%s] %s\", warning.Widget, warning.Error))\n\t\t\t}\n\t\t}\n\n\t\tif dumpModel != \"\" {\n\t\t\tfmt.Println(color.YellowString(L(\"Not supported yet\")))\n\t\t\tos.Exit(1)\n\t\t\treturn\n\t\t}\n\n\t\t// Export models\n\t\tfiles := []string{}\n\t\tfor _, mod := range model.Models {\n\n\t\t\tfmt.Printf(\"\\r%s\", color.GreenString(L(\"Export the models: %s (%s)\"), mod.Name, mod.MetaData.Table.Name))\n\t\t\tjsonfiles, err := mod.Export(5000, func(curr, total int) {\n\t\t\t\tfmt.Printf(\"\\r%s\", strings.Repeat(\" \", 80))\n\t\t\t\tfmt.Printf(\"\\r%s\", color.GreenString(L(\"Export the models: %s (%s) %d/%d\"), mod.Name, mod.MetaData.Table.Name, curr, total))\n\t\t\t})\n\n\t\t\tif err != nil {\n\t\t\t\tfmt.Println(color.RedString(L(\"Fatal: %s\"), err.Error()))\n\t\t\t\tos.Exit(1)\n\t\t\t}\n\t\t\tfiles = append(files, jsonfiles...)\n\t\t}\n\t\tfmt.Printf(\"\\r%s\", strings.Repeat(\" \", 80))\n\t\tfmt.Printf(\"\\r%s\\n\", color.GreenString(L(\"Export the models: ✨DONE✨\")))\n\n\t\t// Compress files\n\t\terr = zipfiles(files, output, func(file string) {\n\t\t\tfmt.Printf(\"\\r%s\", strings.Repeat(\" \", 80))\n\t\t\tfmt.Printf(\"\\r%s\", color.GreenString(L(\"Compress the files: %s\"), file))\n\t\t})\n\t\tfmt.Printf(\"\\r%s\", strings.Repeat(\" \", 80))\n\t\tfmt.Printf(\"\\r%s\\n\", color.GreenString(L(\"Compress the files: ✨DONE✨\")))\n\n\t\tif err != nil {\n\t\t\tfmt.Println(color.RedString(L(\"Fatal: %s\"), err.Error()))\n\t\t\tos.Exit(1)\n\t\t}\n\n\t\tfmt.Println(color.GreenString(\"File: %s\", output))\n\t},\n}\n\n// func init() {\n// \t// dumpCmd.PersistentFlags().StringVarP(&dumpModel, \"name\", \"n\", \"\", L(\"Model name\"))\n// }\n\n// gzipfiles\nfunc zipfiles(files []string, output string, process func(file string)) error {\n\toutpath := filepath.Dir(output)\n\tos.MkdirAll(outpath, 0755)\n\n\toutfile, err := os.Create(output)\n\tif err != nil {\n\t\tfmt.Println(color.RedString(L(\"Fatal: %s\"), err.Error()))\n\t\tos.Exit(1)\n\t}\n\tdefer outfile.Close()\n\n\tw := zip.NewWriter(outfile)\n\tdefer func() {\n\t\tw.Close()\n\t\tif err != nil {\n\t\t\tfmt.Println(color.RedString(L(\"Fatal: %s\"), err.Error()))\n\t\t\tos.Exit(1)\n\t\t}\n\t}()\n\n\tfor _, file := range files {\n\t\taddFile(w, file, \"model\", process)\n\t}\n\n\t// Add data path\n\tdataPath := filepath.Join(config.Conf.Root, \"data\")\n\t_, err = os.Stat(dataPath)\n\tif err == nil {\n\t\taddFolder(w, dataPath, \"data\", process)\n\t}\n\n\treturn nil\n}\n\nfunc addFile(w *zip.Writer, file, baseInZip string, process func(file string)) {\n\tprocess(filepath.Join(baseInZip, filepath.Base(file)))\n\tdat, err := ioutil.ReadFile(file)\n\tif err != nil {\n\t\tfmt.Println(color.RedString(L(\"Fatal: %s\"), err.Error()))\n\t\tos.Exit(1)\n\t}\n\n\t// Add some files to the archive.\n\tf, err := w.Create(filepath.Join(baseInZip, filepath.Base(file)))\n\tif err != nil {\n\t\tfmt.Println(color.RedString(L(\"Fatal: %s\"), err.Error()))\n\t\tos.Exit(1)\n\t}\n\t_, err = f.Write(dat)\n\tif err != nil {\n\t\tfmt.Println(color.RedString(L(\"Fatal: %s\"), err.Error()))\n\t\tos.Exit(1)\n\t}\n\n\tos.Remove(file)\n}\n\nfunc addFolder(w *zip.Writer, basePath, baseInZip string, process func(file string)) {\n\n\t// Open the Directory\n\tfiles, err := ioutil.ReadDir(basePath)\n\tif err != nil {\n\t\tfmt.Println(color.RedString(L(\"Fatal: %s\"), err.Error()))\n\t\tos.Exit(1)\n\t}\n\n\tfor _, file := range files {\n\t\tprocess(filepath.Join(baseInZip, file.Name()))\n\t\tif !file.IsDir() {\n\t\t\tdat, err := ioutil.ReadFile(filepath.Join(basePath, file.Name()))\n\t\t\tif err != nil {\n\t\t\t\tfmt.Println(color.RedString(L(\"Fatal: %s\"), err.Error()))\n\t\t\t\tos.Exit(1)\n\t\t\t}\n\n\t\t\t// Add some files to the archive.\n\t\t\tf, err := w.Create(filepath.Join(baseInZip, file.Name()))\n\t\t\tif err != nil {\n\t\t\t\tfmt.Println(color.RedString(L(\"Fatal: %s\"), err.Error()))\n\t\t\t\tos.Exit(1)\n\t\t\t}\n\t\t\t_, err = f.Write(dat)\n\t\t\tif err != nil {\n\t\t\t\tfmt.Println(color.RedString(L(\"Fatal: %s\"), err.Error()))\n\t\t\t\tos.Exit(1)\n\t\t\t}\n\t\t} else if file.IsDir() {\n\t\t\t// Recurse\n\t\t\tnewBase := filepath.Join(basePath, file.Name())\n\t\t\taddFolder(w, newBase, filepath.Join(baseInZip, file.Name()), process)\n\t\t}\n\t}\n}\n" }, { "path": "cmd/get/get.go", "content": "package get\n\nimport (\n\t\"archive/zip\"\n\t\"fmt\"\n\t\"io\"\n\t\"io/ioutil\"\n\t\"net/http\"\n\t\"os\"\n\t\"path/filepath\"\n\t\"strings\"\n\t\"time\"\n\n\tjsoniter \"github.com/json-iterator/go\"\n\t\"github.com/yaoapp/gou/fs/system\"\n\t\"github.com/yaoapp/kun/log\"\n\t\"github.com/yaoapp/yao/widgets/app\"\n)\n\nconst (\n\n\t// Application application\n\tApplication uint = iota\n\n\t// Widgets ? model & table & flow\n\tWidgets\n\n\t// Table table widget\n\tTable\n\n\t// Form form widget\n\tForm\n\n\t// Model model model\n\tModel\n\n\t// Flow data flow\n\tFlow\n)\n\n// Package package\ntype Package struct {\n\tName string\n\tTeam string\n\tType uint\n\tRemote string\n\tOrigin string\n\tTemp string\n\tTag string\n\tFrom string\n}\n\n// New create a package via name\nfunc New(repo string) (*Package, error) {\n\n\tteam, name, tag, err := parse(repo)\n\tif err != nil {\n\t\treturn nil, err\n\t}\n\n\tpkg := &Package{\n\t\tOrigin: repo,\n\t\tTeam: team,\n\t\tName: name,\n\t\tTag: tag,\n\t\tType: Application,\n\t}\n\n\turl := pkg.InfraURL()\n\tif urlExists(url) {\n\t\tpkg.Remote = url\n\t\tpkg.From = \"LetsInfra.com\"\n\t\treturn pkg, nil\n\t}\n\n\t// @Todo: Download from Github\n\n\treturn nil, fmt.Errorf(\"%s not found\", repo)\n}\n\n// InfraURL infra package url\nfunc parse(repo string) (string, string, string, error) {\n\n\ttag := \"latest\"\n\trepo = strings.TrimSpace(repo)\n\tif !strings.Contains(repo, \"/\") {\n\t\trepo = fmt.Sprintf(\"yaoapp/%s\", repo)\n\t}\n\n\tif strings.Contains(repo, \"@\") {\n\t\tarr := strings.Split(repo, \"@\")\n\t\trepo = arr[0]\n\t\ttag = arr[1]\n\t}\n\n\tarr := strings.Split(repo, \"/\")\n\tif len(arr) != 2 {\n\t\treturn \"\", \"\", \"\", fmt.Errorf(\"REPO: %s format error\", repo)\n\t}\n\n\tteam := arr[0]\n\tname := arr[1]\n\treturn team, name, tag, nil\n}\n\n// InfraURL infra package url\nfunc (pkg *Package) InfraURL() string {\n\treturn fmt.Sprintf(\"https://mirrors.yao.run/apps/%s/%s/%s.zip\", pkg.Team, pkg.Name, pkg.Tag)\n}\n\n// GithubURL github package url\nfunc (pkg *Package) GithubURL() string {\n\treturn fmt.Sprintf(\"mirrors.letsinfra.com/apps/%s/%s/%s\", pkg.Team, pkg.Name, pkg.Tag)\n}\n\n// urlExists check the http url is exists\nfunc urlExists(url string) bool {\n\tresp, err := http.Get(url)\n\tif err != nil {\n\t\treturn false\n\t}\n\n\tif resp.Body != nil {\n\t\tdefer resp.Body.Close()\n\t}\n\n\treturn resp.StatusCode == 200\n}\n\n// Download a package from remote\nfunc (pkg *Package) Download() error {\n\n\tif pkg.Remote == \"\" {\n\t\treturn fmt.Errorf(\"remote url is required\")\n\t}\n\n\troot, err := os.MkdirTemp(\"\", \"*-yao-zip\")\n\tif err != nil {\n\t\treturn fmt.Errorf(\"Can't Create temp dir %s\", err.Error())\n\t}\n\n\tname := fmt.Sprintf(\"%s-%s-%d.zip\", pkg.Team, pkg.Name, time.Now().UnixMicro())\n\tfile := filepath.Join(root, name)\n\tout, err := os.Create(file)\n\tdefer out.Close()\n\tif err != nil {\n\t\treturn fmt.Errorf(\"Can't Create file: %s\", err.Error())\n\t}\n\n\tresp, err := http.Get(pkg.Remote)\n\tif err != nil {\n\t\treturn fmt.Errorf(\"Download Error: %s\", err.Error())\n\t}\n\tdefer resp.Body.Close()\n\n\t_, err = io.Copy(out, resp.Body)\n\tif err != nil {\n\t\treturn fmt.Errorf(\"Copy Error: %s\", err.Error())\n\t}\n\n\tpkg.Temp = file\n\treturn nil\n}\n\n// Validate a package files\nfunc (pkg *Package) Validate() error {\n\tif pkg.Temp == \"\" {\n\t\treturn fmt.Errorf(\"temp file not found\")\n\t}\n\treturn nil\n}\n\n// Unpack a package to current dir\nfunc (pkg *Package) Unpack(dest string) (*app.DSL, error) {\n\n\tdest, err := filepath.Abs(dest)\n\tif err != nil {\n\t\treturn nil, err\n\t}\n\n\tfiles, err := ioutil.ReadDir(dest)\n\tif err != nil {\n\t\treturn nil, err\n\t}\n\n\tfor _, f := range files {\n\t\tif !strings.HasPrefix(f.Name(), \"logs\") {\n\t\t\treturn nil, fmt.Errorf(\"current folder shoud be empty\")\n\t\t}\n\t}\n\n\ttemp, err := os.MkdirTemp(\"\", \"*-yao-unzip\")\n\tif err != nil {\n\t\treturn nil, err\n\t}\n\tdefer os.RemoveAll(temp)\n\n\t// Read zip file\n\tr, err := zip.OpenReader(pkg.Temp)\n\tif err != nil {\n\t\treturn nil, err\n\t}\n\n\tdefer r.Close()\n\tdefer os.Remove(pkg.Temp)\n\n\tpath := \"\"\n\tfor i, f := range r.File {\n\t\tif i == 0 {\n\t\t\tpath = filepath.Join(temp, strings.TrimRight(f.Name, \"/\"))\n\t\t}\n\t\terr := extractFile(f, temp)\n\t\tif err != nil {\n\t\t\treturn nil, err\n\t\t}\n\t}\n\n\tdata, err := os.ReadFile(filepath.Join(path, \"app.yao\"))\n\tif err != nil {\n\t\treturn nil, err\n\t}\n\n\tvar setting app.DSL\n\terr = jsoniter.Unmarshal(data, &setting)\n\tif err != nil {\n\t\treturn nil, err\n\t}\n\n\tfs := system.New(\"/\")\n\terr = fs.Copy(path, dest)\n\tif err != nil {\n\t\treturn nil, err\n\t}\n\n\t// Remove env\n\tfs.Remove(filepath.Join(dest, \".env\"))\n\treturn &setting, nil\n}\n\n// extractFile extract and save file to the dest path\nfunc extractFile(f *zip.File, dest string) error {\n\trc, err := f.Open()\n\tif err != nil {\n\t\treturn err\n\t}\n\tdefer rc.Close()\n\n\tpath := filepath.Join(dest, f.Name)\n\n\t// Check for ZipSlip (Directory traversal)\n\tif !strings.HasPrefix(path, filepath.Clean(dest)+string(os.PathSeparator)) {\n\t\treturn fmt.Errorf(\"illegal file path: %s\", path)\n\t}\n\n\tif f.FileInfo().IsDir() {\n\t\tos.MkdirAll(path, f.Mode())\n\t} else {\n\t\tos.MkdirAll(filepath.Dir(path), f.Mode())\n\t\tf, err := os.OpenFile(path, os.O_WRONLY|os.O_CREATE|os.O_TRUNC, f.Mode())\n\t\tif err != nil {\n\t\t\treturn err\n\t\t}\n\t\tdefer func() {\n\t\t\tif err := f.Close(); err != nil {\n\t\t\t\tlog.Error(\"repo unzip extractFile: %s\", err.Error())\n\t\t\t}\n\t\t}()\n\t\t_, err = io.Copy(f, rc)\n\t\tif err != nil {\n\t\t\treturn err\n\t\t}\n\t}\n\treturn nil\n}\n" }, { "path": "cmd/get/get_test.go", "content": "package get\n\nimport (\n\t\"testing\"\n)\n\nfunc TestUnpack(t *testing.T) {\n\t// pkg, err := New(\"yaoapp/demo-app\")\n\t// if err != nil {\n\t// \tt.Fatal(err)\n\t// }\n\n\t// if err := pkg.Download(); err != nil {\n\t// \tt.Fatal(err)\n\t// }\n\n\t// dest, err := os.MkdirTemp(\"\", \"*-unit-test\")\n\t// if err != nil {\n\t// \tt.Fatal(err)\n\t// }\n\n\t// defer os.RemoveAll(dest)\n\t// app, err := pkg.Unpack(dest)\n\t// if err != nil {\n\t// \tt.Fatal(err)\n\t// }\n\n\t// assert.NotNil(t, app.Name)\n}\n" }, { "path": "cmd/get.go", "content": "package cmd\n\nimport (\n\t\"fmt\"\n\t\"os\"\n\n\t\"github.com/fatih/color\"\n\t\"github.com/spf13/cobra\"\n\t\"github.com/yaoapp/yao/cmd/get\"\n\t\"github.com/yaoapp/yao/share\"\n)\n\nvar getCmd = &cobra.Command{\n\tUse: \"get\",\n\tShort: L(\"Get an application\"),\n\tLong: L(\"Get an application\"),\n\tRun: func(cmd *cobra.Command, args []string) {\n\t\tif len(args) < 1 {\n\t\t\tfmt.Println(color.RedString(L(\"Not enough arguments\")))\n\t\t\tfmt.Println(color.WhiteString(share.BUILDNAME + \" help\"))\n\t\t\treturn\n\t\t}\n\n\t\trepo := args[0]\n\t\tpkg, err := get.New(repo)\n\t\tif err != nil {\n\t\t\tfmt.Println(color.RedString(err.Error()))\n\t\t\tos.Exit(1)\n\t\t}\n\n\t\tfmt.Println(color.WhiteString(\"From Yao: %s\", pkg.Remote))\n\t\tfmt.Println(color.WhiteString(\"Visit: https://yaoapps.com\"))\n\t\terr = pkg.Download()\n\t\tif err != nil {\n\t\t\tfmt.Println(color.RedString(err.Error()))\n\t\t\tos.Exit(1)\n\t\t}\n\n\t\tdest, err := os.Getwd()\n\t\tif err != nil {\n\t\t\tfmt.Println(color.RedString(err.Error()))\n\t\t\tos.Exit(1)\n\t\t}\n\n\t\t// dest, err = os.MkdirTemp(dest, \"*-unit-test\")\n\t\t// if err != nil {\n\t\t// \tfmt.Println(color.RedString(err.Error()))\n\t\t// \tos.Exit(1)\n\t\t// }\n\t\t// os.MkdirAll(dest, os.ModePerm)\n\n\t\tapp, err := pkg.Unpack(dest)\n\t\tif err != nil {\n\t\t\tfmt.Println(color.RedString(err.Error()))\n\t\t\tos.Exit(1)\n\t\t}\n\n\t\tfmt.Println(color.GreenString(app.Name), color.WhiteString(app.Version))\n\t\tfmt.Println(color.GreenString(L(\"✨DONE✨\")))\n\n\t},\n}\n" }, { "path": "cmd/help.go", "content": "package cmd\n\nimport (\n\t\"github.com/spf13/cobra\"\n)\n\nvar helpCmd = &cobra.Command{\n\tUse: \"help\",\n\tShort: L(\"Help for yao\"),\n\tLong: L(\"Help for yao\"),\n\tRun: func(cmd *cobra.Command, args []string) {\n\t\tcmd.Help()\n\t},\n}\n" }, { "path": "cmd/init.go", "content": "package cmd\n\nimport (\n\t\"fmt\"\n\t\"os\"\n\t\"path/filepath\"\n\t\"time\"\n\n\t\"github.com/fatih/color\"\n\t\"github.com/spf13/cobra\"\n\t\"github.com/yaoapp/yao/config\"\n\t\"github.com/yaoapp/yao/engine\"\n\t\"github.com/yaoapp/yao/setup\"\n\t\"github.com/yaoapp/yao/share\"\n)\n\nvar initCmd = &cobra.Command{\n\tUse: \"init\",\n\tShort: L(\"Initialize project\"),\n\tLong: L(\"Initialize a new Yao application in the current directory\"),\n\tRun: func(cmd *cobra.Command, args []string) {\n\t\tBoot()\n\n\t\t// First check if we're inside an existing Yao app (including parent directories)\n\t\tif setup.InYaoApp(config.Conf.Root) {\n\t\t\tfmt.Println(color.YellowString(L(\"Directory is inside an existing Yao application\")))\n\t\t\tfmt.Println(color.WhiteString(\"Please run 'yao init' outside of any Yao project\"))\n\t\t\tos.Exit(1)\n\t\t}\n\n\t\t// Check if this is an empty directory\n\t\tif !setup.IsEmptyDir(config.Conf.Root) {\n\t\t\t// Directory is not empty\n\t\t\tfmt.Println(color.RedString(L(\"Directory is not empty\")))\n\t\t\tfmt.Println(color.WhiteString(\"Please run 'yao init' in an empty directory\"))\n\t\t\tos.Exit(1)\n\t\t}\n\n\t\tstartTime := time.Now()\n\t\tfmt.Println(color.CyanString(\"Initializing Yao application...\"))\n\n\t\t// Install the init app (copy embedded files)\n\t\tif err := setup.Install(config.Conf.Root); err != nil {\n\t\t\tfmt.Println(color.RedString(L(\"Install: %s\"), err.Error()))\n\t\t\tos.Exit(1)\n\t\t}\n\t\tfmt.Printf(\" %s %s\\n\", color.GreenString(\"✓\"), \"Copied application files\")\n\n\t\t// Reload configuration after install\n\t\tBoot()\n\n\t\t// Load the application engine\n\t\tloadWarnings, err := engine.Load(config.Conf, engine.LoadOption{Action: \"init\"})\n\t\tif err != nil {\n\t\t\tfmt.Println(color.RedString(L(\"Load: %s\"), err.Error()))\n\t\t\tos.Exit(1)\n\t\t}\n\t\tfmt.Printf(\" %s %s\\n\", color.GreenString(\"✓\"), \"Loaded application engine\")\n\n\t\t// Initialize (migrate + setup hook)\n\t\tif err := setup.Initialize(config.Conf.Root, config.Conf); err != nil {\n\t\t\tfmt.Println(color.RedString(L(\"Initialize: %s\"), err.Error()))\n\t\t\tos.Exit(1)\n\t\t}\n\t\tfmt.Printf(\" %s %s\\n\", color.GreenString(\"✓\"), \"Initialized database and data\")\n\n\t\tinitDuration := time.Since(startTime)\n\n\t\t// Print warnings if any\n\t\tif len(loadWarnings) > 0 {\n\t\t\tfmt.Println(color.YellowString(\"\\n---------------------------------\"))\n\t\t\tfmt.Println(color.YellowString(L(\"Warnings\")))\n\t\t\tfmt.Println(color.YellowString(\"---------------------------------\"))\n\t\t\tfor _, warning := range loadWarnings {\n\t\t\t\tfmt.Println(color.YellowString(\"[%s] %s\", warning.Widget, warning.Error))\n\t\t\t}\n\t\t}\n\n\t\t// Print success message\n\t\tfmt.Printf(\"\\n%s Application initialized successfully in %s\\n\\n\",\n\t\t\tcolor.GreenString(\"✓\"),\n\t\t\tcolor.CyanString(\"%v\", initDuration))\n\n\t\t// Print application info\n\t\troot, _ := filepath.Abs(config.Conf.Root)\n\t\tfmt.Println(color.WhiteString(\"---------------------------------\"))\n\t\tfmt.Println(color.WhiteString(L(\"Application Info\")))\n\t\tfmt.Println(color.WhiteString(\"---------------------------------\"))\n\t\tfmt.Println(color.WhiteString(L(\"Name\")), color.GreenString(\" %s\", share.App.Name))\n\t\tfmt.Println(color.WhiteString(L(\"Version\")), color.GreenString(\" %s\", share.App.Version))\n\t\tfmt.Println(color.WhiteString(L(\"Root\")), color.GreenString(\" %s\", root))\n\n\t\t// Print welcome message\n\t\tprintInitWelcome()\n\t},\n}\n\nfunc printInitWelcome() {\n\tfmt.Println(color.CyanString(\"\\n---------------------------------\"))\n\tfmt.Println(color.CyanString(L(\"🎉 Application Ready 🎉\")))\n\tfmt.Println(color.CyanString(\"---------------------------------\"))\n\tfmt.Println(color.WhiteString(\"📚 Documentation: \"), color.CyanString(\"https://yaoapps.com/docs\"))\n\tfmt.Println(color.WhiteString(\"🏡 Join Yao Community: \"), color.CyanString(\"https://yaoapps.com/community\"))\n\tfmt.Println(color.WhiteString(\"🤖 Build Your Digital Workforce:\"), color.CyanString(\"https://yaoagents.com\"))\n\tfmt.Println(\"\")\n\tfmt.Println(color.WhiteString(L(\"NEXT:\")))\n\tfmt.Println(color.GreenString(\" 1. Edit .env to configure your application\"))\n\tfmt.Println(color.GreenString(\" 2. Run 'yao start' to start the server\"))\n\tfmt.Println(\"\")\n}\n\nfunc init() {\n\t// Register init command\n\trootCmd.AddCommand(initCmd)\n}\n" }, { "path": "cmd/inspect.go", "content": "package cmd\n\nimport (\n\t\"github.com/spf13/cobra\"\n\t\"github.com/yaoapp/kun/maps\"\n\t\"github.com/yaoapp/kun/utils\"\n\t\"github.com/yaoapp/yao/config\"\n\t\"github.com/yaoapp/yao/engine\"\n\t\"github.com/yaoapp/yao/share\"\n)\n\nvar inspectCmd = &cobra.Command{\n\tUse: \"inspect\",\n\tShort: L(\"Show app configure\"),\n\tLong: L(\"Show app configure\"),\n\tRun: func(cmd *cobra.Command, args []string) {\n\t\tBoot()\n\t\tengine.InspectExtTools()\n\t\tres := maps.Map{\n\t\t\t\"version\": share.VERSION,\n\t\t\t\"config\": config.Conf,\n\t\t}\n\t\tif share.Tools != nil {\n\t\t\tres[\"tools\"] = share.Tools\n\t\t}\n\t\tutils.Dump(res)\n\t},\n}\n" }, { "path": "cmd/login.go", "content": "package cmd\n\nimport (\n\t\"encoding/base64\"\n\t\"encoding/json\"\n\t\"fmt\"\n\t\"io\"\n\t\"net/http\"\n\t\"net/url\"\n\t\"os\"\n\t\"strings\"\n\t\"time\"\n\n\t\"github.com/fatih/color\"\n\t\"github.com/spf13/cobra\"\n\t\"github.com/yaoapp/yao/engine\"\n)\n\nvar loginServer string\n\nvar loginCmd = &cobra.Command{\n\tUse: \"login\",\n\tShort: L(\"Login to remote Yao server\"),\n\tLong: L(\"Login to remote Yao server using device authorization flow\"),\n\tRun: func(cmd *cobra.Command, args []string) {\n\t\tif loginServer == \"\" {\n\t\t\tcolor.Red(L(\"Missing --server flag\\n\"))\n\t\t\tfmt.Println(\" yao login --server https://yaoagents.com\")\n\t\t\tos.Exit(1)\n\t\t}\n\n\t\tserverURL := strings.TrimRight(loginServer, \"/\")\n\n\t\t// 1. Discover OAuth endpoints via well-known metadata\n\t\tendpoints, err := discoverEndpoints(serverURL)\n\t\tif err != nil {\n\t\t\tcolor.Red(\" %s %s\\n\", L(\"Server discovery failed:\"), err)\n\t\t\tos.Exit(1)\n\t\t}\n\n\t\t// 2. Compute deterministic client_id from machine fingerprint\n\t\tmachine, err := engine.GetMachineID()\n\t\tif err != nil {\n\t\t\tcolor.Red(\"Failed to compute machine ID: %s\\n\", err)\n\t\t\tos.Exit(1)\n\t\t}\n\t\tclientID := machine.ID\n\n\t\t// 3. Register the client (idempotent for same client_id)\n\t\tif endpoints.RegistrationEndpoint != \"\" {\n\t\t\tif err := registerClient(endpoints.RegistrationEndpoint, clientID); err != nil {\n\t\t\t\tcolor.Red(\"Client registration failed: %s\\n\", err)\n\t\t\t\tos.Exit(1)\n\t\t\t}\n\t\t}\n\n\t\t// 4. Start device authorization\n\t\tdeviceResp, err := requestDeviceAuthorization(endpoints.DeviceAuthorizationEndpoint, clientID)\n\t\tif err != nil {\n\t\t\tcolor.Red(\"Device authorization failed: %s\\n\", err)\n\t\t\tos.Exit(1)\n\t\t}\n\n\t\t// 5. Display the code to the user\n\t\tdashboard := endpoints.Dashboard\n\t\tif dashboard == \"\" {\n\t\t\tdashboard = \"/admin\"\n\t\t}\n\t\tverifyURI := strings.TrimRight(serverURL, \"/\") + dashboard + \"/auth/device\"\n\t\tverifyURIComplete := verifyURI + \"?user_code=\" + deviceResp.UserCode\n\n\t\tfmt.Println()\n\t\tcolor.White(\" %s %s\\n\",\n\t\t\tL(\"Open:\"),\n\t\t\tcolor.CyanString(verifyURIComplete))\n\t\tfmt.Println()\n\t\tcolor.White(\" %s %s\\n\",\n\t\t\tL(\"Or visit:\"),\n\t\t\tcolor.CyanString(verifyURI))\n\t\tcolor.White(\" %s %s\\n\",\n\t\t\tL(\"Enter code:\"),\n\t\t\tcolor.YellowString(deviceResp.UserCode))\n\t\tfmt.Println()\n\n\t\t// 6. Poll for token\n\t\tinterval := deviceResp.Interval\n\t\tif interval < 5 {\n\t\t\tinterval = 5\n\t\t}\n\n\t\tcolor.White(\" %s\", L(\"Waiting for authorization...\"))\n\t\ttokenResp, err := pollForToken(endpoints.TokenEndpoint, clientID, deviceResp.DeviceCode, interval, deviceResp.ExpiresIn)\n\t\tif err != nil {\n\t\t\tfmt.Println()\n\t\t\tcolor.Red(\"\\n %s %s\\n\", L(\"Login failed:\"), err)\n\t\t\tos.Exit(1)\n\t\t}\n\n\t\t// 6. Save credential\n\t\texpiresAt := \"\"\n\t\tif tokenResp.ExpiresIn > 0 {\n\t\t\texpiresAt = time.Now().Add(time.Duration(tokenResp.ExpiresIn) * time.Second).UTC().Format(time.RFC3339)\n\t\t}\n\n\t\tcred := &Credential{\n\t\t\tServer: serverURL,\n\t\t\tGRPCAddr: endpoints.GRPCAddr,\n\t\t\tAccessToken: tokenResp.AccessToken,\n\t\t\tRefreshToken: tokenResp.RefreshToken,\n\t\t\tScope: tokenResp.Scope,\n\t\t\tUser: parseJWTSubject(tokenResp.AccessToken),\n\t\t\tExpiresAt: expiresAt,\n\t\t}\n\n\t\tif err := SaveCredential(cred); err != nil {\n\t\t\tcolor.Red(\"\\n Failed to save credentials: %s\\n\", err)\n\t\t\tos.Exit(1)\n\t\t}\n\n\t\tfmt.Print(\"\\033[2J\\033[H\")\n\t\tcolor.Green(\" ✓ %s\\n\", L(\"Login successful\"))\n\t\tcolor.White(\" %s %s\\n\", L(\"Server:\"), serverURL)\n\t\tif cred.GRPCAddr != \"\" {\n\t\t\tcolor.White(\" %s %s\\n\", L(\"gRPC:\"), cred.GRPCAddr)\n\t\t}\n\t\tif cred.User != \"\" {\n\t\t\tcolor.White(\" %s %s\\n\", L(\"User:\"), cred.User)\n\t\t}\n\t\tif cred.ExpiresAt != \"\" {\n\t\t\tcolor.White(\" %s %s\\n\", L(\"Expires:\"), cred.ExpiresAt)\n\t\t}\n\t\tfmt.Println()\n\t},\n}\n\nfunc init() {\n\tloginCmd.PersistentFlags().StringVar(&loginServer, \"server\", \"\", L(\"Remote Yao server URL\"))\n}\n\n// --- types ---\n\ntype oauthEndpoints struct {\n\tRegistrationEndpoint string `json:\"registration_endpoint\"`\n\tDeviceAuthorizationEndpoint string `json:\"device_authorization_endpoint\"`\n\tTokenEndpoint string `json:\"token_endpoint\"`\n\tRevocationEndpoint string `json:\"revocation_endpoint\"`\n\tDashboard string `json:\"-\"`\n\tGRPCAddr string `json:\"-\"`\n}\n\ntype deviceAuthResponse struct {\n\tDeviceCode string `json:\"device_code\"`\n\tUserCode string `json:\"user_code\"`\n\tVerificationURI string `json:\"verification_uri\"`\n\tVerificationURIComplete string `json:\"verification_uri_complete\"`\n\tExpiresIn int `json:\"expires_in\"`\n\tInterval int `json:\"interval\"`\n}\n\ntype tokenResponse struct {\n\tAccessToken string `json:\"access_token\"`\n\tTokenType string `json:\"token_type\"`\n\tExpiresIn int `json:\"expires_in\"`\n\tRefreshToken string `json:\"refresh_token\"`\n\tScope string `json:\"scope\"`\n}\n\ntype oauthError struct {\n\tError string `json:\"error\"`\n\tErrorDescription string `json:\"error_description\"`\n}\n\n// --- HTTP helpers ---\n\n// discoverEndpoints fetches OAuth endpoint URLs from /.well-known/yao,\n// using the openapi base prefix to construct correct API paths.\nfunc discoverEndpoints(serverURL string) (*oauthEndpoints, error) {\n\treturn discoverFromYaoMetadata(serverURL)\n}\n\ntype yaoMetadataResponse struct {\n\tOpenAPI string `json:\"openapi\"`\n\tDashboard string `json:\"dashboard\"`\n\tGRPC string `json:\"grpc\"`\n}\n\nfunc discoverFromYaoMetadata(serverURL string) (*oauthEndpoints, error) {\n\tresp, err := http.Get(serverURL + \"/.well-known/yao\")\n\tif err != nil {\n\t\treturn nil, fmt.Errorf(\"network error: %w\", err)\n\t}\n\tdefer resp.Body.Close()\n\n\tbody, _ := io.ReadAll(resp.Body)\n\tif resp.StatusCode != http.StatusOK {\n\t\treturn nil, fmt.Errorf(\"/.well-known/yao returned %d\", resp.StatusCode)\n\t}\n\n\tvar meta yaoMetadataResponse\n\tif err := json.Unmarshal(body, &meta); err != nil {\n\t\treturn nil, fmt.Errorf(\"invalid /.well-known/yao response: %w\", err)\n\t}\n\n\tbase := strings.TrimRight(serverURL, \"/\") + meta.OpenAPI\n\n\treturn &oauthEndpoints{\n\t\tRegistrationEndpoint: base + \"/oauth/register\",\n\t\tDeviceAuthorizationEndpoint: base + \"/oauth/device_authorization\",\n\t\tTokenEndpoint: base + \"/oauth/token\",\n\t\tRevocationEndpoint: base + \"/oauth/revoke\",\n\t\tDashboard: meta.Dashboard,\n\t\tGRPCAddr: meta.GRPC,\n\t}, nil\n}\n\nfunc registerClient(endpoint, clientID string) error {\n\tbody := fmt.Sprintf(\n\t\t`{\"client_id\":\"%s\",\"client_name\":\"yao-cli\",\"grant_types\":[\"urn:ietf:params:oauth:grant-type:device_code\"],\"token_endpoint_auth_method\":\"none\",\"redirect_uris\":[\"http://localhost\"]}`,\n\t\tclientID,\n\t)\n\tresp, err := http.Post(endpoint, \"application/json\", strings.NewReader(body))\n\tif err != nil {\n\t\treturn fmt.Errorf(\"network error: %w\", err)\n\t}\n\tdefer resp.Body.Close()\n\n\tif resp.StatusCode == http.StatusOK || resp.StatusCode == http.StatusCreated {\n\t\treturn nil\n\t}\n\n\trespBody, _ := io.ReadAll(resp.Body)\n\tvar oerr oauthError\n\tif json.Unmarshal(respBody, &oerr) == nil && oerr.Error == \"invalid_client_metadata\" {\n\t\treturn nil // client already registered, idempotent\n\t}\n\treturn fmt.Errorf(\"registration returned %d: %s\", resp.StatusCode, string(respBody))\n}\n\nfunc requestDeviceAuthorization(endpoint, clientID string) (*deviceAuthResponse, error) {\n\tdata := url.Values{\n\t\t\"client_id\": {clientID},\n\t\t\"scope\": {\"grpc:run grpc:stream grpc:shell grpc:mcp grpc:llm grpc:agent\"},\n\t}\n\tresp, err := http.PostForm(endpoint, data)\n\tif err != nil {\n\t\treturn nil, fmt.Errorf(\"network error: %w\", err)\n\t}\n\tdefer resp.Body.Close()\n\n\trespBody, _ := io.ReadAll(resp.Body)\n\tif resp.StatusCode != http.StatusOK {\n\t\tvar oerr oauthError\n\t\tjson.Unmarshal(respBody, &oerr)\n\t\tif oerr.ErrorDescription != \"\" {\n\t\t\treturn nil, fmt.Errorf(\"%s\", oerr.ErrorDescription)\n\t\t}\n\t\treturn nil, fmt.Errorf(\"server returned %d: %s\", resp.StatusCode, string(respBody))\n\t}\n\n\tvar result deviceAuthResponse\n\tif err := json.Unmarshal(respBody, &result); err != nil {\n\t\treturn nil, fmt.Errorf(\"invalid response: %w\", err)\n\t}\n\treturn &result, nil\n}\n\nfunc pollForToken(endpoint, clientID, deviceCode string, interval, expiresIn int) (*tokenResponse, error) {\n\tdeadline := time.Now().Add(time.Duration(expiresIn) * time.Second)\n\tticker := time.NewTicker(time.Duration(interval) * time.Second)\n\tdefer ticker.Stop()\n\n\tfor range ticker.C {\n\t\tif time.Now().After(deadline) {\n\t\t\treturn nil, fmt.Errorf(\"device code expired\")\n\t\t}\n\n\t\tdata := url.Values{\n\t\t\t\"grant_type\": {\"urn:ietf:params:oauth:grant-type:device_code\"},\n\t\t\t\"client_id\": {clientID},\n\t\t\t\"device_code\": {deviceCode},\n\t\t}\n\n\t\tresp, err := http.PostForm(endpoint, data)\n\t\tif err != nil {\n\t\t\tcontinue\n\t\t}\n\n\t\trespBody, _ := io.ReadAll(resp.Body)\n\t\tresp.Body.Close()\n\n\t\tif resp.StatusCode == http.StatusOK {\n\t\t\tvar tok tokenResponse\n\t\t\tif err := json.Unmarshal(respBody, &tok); err != nil {\n\t\t\t\treturn nil, fmt.Errorf(\"invalid token response: %w\", err)\n\t\t\t}\n\t\t\treturn &tok, nil\n\t\t}\n\n\t\tvar oerr oauthError\n\t\tjson.Unmarshal(respBody, &oerr)\n\t\tswitch oerr.Error {\n\t\tcase \"authorization_pending\":\n\t\t\tfmt.Print(\".\")\n\t\t\tcontinue\n\t\tcase \"slow_down\":\n\t\t\tinterval += 5\n\t\t\tticker.Reset(time.Duration(interval) * time.Second)\n\t\t\tcontinue\n\t\tcase \"expired_token\":\n\t\t\treturn nil, fmt.Errorf(\"device code expired\")\n\t\tcase \"access_denied\":\n\t\t\treturn nil, fmt.Errorf(\"authorization denied by user\")\n\t\tdefault:\n\t\t\tdesc := oerr.ErrorDescription\n\t\t\tif desc == \"\" {\n\t\t\t\tdesc = oerr.Error\n\t\t\t}\n\t\t\treturn nil, fmt.Errorf(\"%s\", desc)\n\t\t}\n\t}\n\treturn nil, fmt.Errorf(\"device code expired\")\n}\n\n// parseJWTSubject extracts the \"sub\" claim from a JWT access token\n// without verifying the signature (display-only).\nfunc parseJWTSubject(token string) string {\n\tparts := strings.Split(token, \".\")\n\tif len(parts) != 3 {\n\t\treturn \"\"\n\t}\n\tpayload, err := base64.RawURLEncoding.DecodeString(parts[1])\n\tif err != nil {\n\t\treturn \"\"\n\t}\n\tvar claims struct {\n\t\tSub string `json:\"sub\"`\n\t}\n\tif json.Unmarshal(payload, &claims) != nil {\n\t\treturn \"\"\n\t}\n\treturn claims.Sub\n}\n" }, { "path": "cmd/logout.go", "content": "package cmd\n\nimport (\n\t\"net/http\"\n\t\"net/url\"\n\t\"os\"\n\t\"strings\"\n\n\t\"github.com/fatih/color\"\n\t\"github.com/spf13/cobra\"\n)\n\nvar logoutCmd = &cobra.Command{\n\tUse: \"logout\",\n\tShort: L(\"Logout from remote Yao server\"),\n\tLong: L(\"Revoke token and remove stored credentials\"),\n\tRun: func(cmd *cobra.Command, args []string) {\n\t\tcred, err := LoadCredential()\n\t\tif err != nil {\n\t\t\tcolor.Red(\" %s %s\\n\", L(\"Failed to read credentials:\"), err)\n\t\t\tos.Exit(1)\n\t\t}\n\t\tif cred == nil {\n\t\t\tcolor.Yellow(\" %s\\n\", L(\"Not logged in\"))\n\t\t\treturn\n\t\t}\n\n\t\t// Best-effort token revocation via discovery\n\t\tif cred.AccessToken != \"\" && cred.Server != \"\" {\n\t\t\tif ep, err := discoverEndpoints(cred.Server); err == nil && ep.RevocationEndpoint != \"\" {\n\t\t\t\trevokeToken(ep.RevocationEndpoint, cred.AccessToken)\n\t\t\t}\n\t\t}\n\n\t\tif err := RemoveCredential(); err != nil {\n\t\t\tcolor.Red(\" %s %s\\n\", L(\"Failed to remove credentials:\"), err)\n\t\t\tos.Exit(1)\n\t\t}\n\n\t\tcolor.Green(\" ✓ %s\\n\", L(\"Logged out\"))\n\t\tif cred.Server != \"\" {\n\t\t\tcolor.White(\" %s %s\\n\", L(\"Server:\"), cred.Server)\n\t\t}\n\t},\n}\n\nfunc revokeToken(endpoint, token string) {\n\tdata := url.Values{\"token\": {token}}\n\treq, err := http.NewRequest(\"POST\", endpoint, strings.NewReader(data.Encode()))\n\tif err != nil {\n\t\treturn\n\t}\n\treq.Header.Set(\"Content-Type\", \"application/x-www-form-urlencoded\")\n\thttp.DefaultClient.Do(req)\n}\n\nfunc init() {\n\t// Add i18n entries\n\tlangs[\"Login to remote Yao server\"] = \"登录远程 Yao 服务器\"\n\tlangs[\"Login to remote Yao server using device authorization flow\"] = \"使用设备授权流程登录远程 Yao 服务器\"\n\tlangs[\"Remote Yao server URL\"] = \"远程 Yao 服务器地址\"\n\tlangs[\"Logout from remote Yao server\"] = \"登出远程 Yao 服务器\"\n\tlangs[\"Revoke token and remove stored credentials\"] = \"撤销令牌并移除存储的凭证\"\n\tlangs[\"Missing --server flag\"] = \"缺少 --server 参数\"\n\tlangs[\"Open:\"] = \"打开:\"\n\tlangs[\"Or visit:\"] = \"或访问:\"\n\tlangs[\"Enter code:\"] = \"输入设备码:\"\n\tlangs[\"Waiting for authorization...\"] = \"等待授权...\"\n\tlangs[\"Login failed:\"] = \"登录失败:\"\n\tlangs[\"Login successful\"] = \"登录成功\"\n\tlangs[\"Server:\"] = \"服务器:\"\n\tlangs[\"Scope:\"] = \"授权范围:\"\n\tlangs[\"Failed to read credentials:\"] = \"读取凭证失败:\"\n\tlangs[\"Not logged in\"] = \"未登录\"\n\tlangs[\"Failed to remove credentials:\"] = \"移除凭证失败:\"\n\tlangs[\"Logged out\"] = \"已登出\"\n\tlangs[\"Path to credentials file\"] = \"凭证文件路径\"\n\tlangs[\"Failed to load credentials:\"] = \"加载凭证失败:\"\n\tlangs[\"Server discovery failed:\"] = \"服务发现失败:\"\n}\n" }, { "path": "cmd/mcp/add.go", "content": "package mcp\n\nimport (\n\t\"fmt\"\n\t\"os\"\n\n\t\"github.com/spf13/cobra\"\n\t\"github.com/yaoapp/yao/config\"\n\t\"github.com/yaoapp/yao/registry\"\n\tmcpmgr \"github.com/yaoapp/yao/registry/manager/mcp\"\n)\n\nvar mcpAddForce bool\n\n// AddCmd implements \"yao mcp add @scope/name\"\nvar AddCmd = &cobra.Command{\n\tUse: \"add [package]\",\n\tShort: L(\"Install an MCP package from the registry\"),\n\tLong: L(\"Install an MCP package from the registry. Example: yao mcp add @yao/rag-tools\"),\n\tArgs: cobra.ExactArgs(1),\n\tRun: func(cmd *cobra.Command, args []string) {\n\t\tBoot()\n\t\tpkgID := args[0]\n\t\tversion, _ := cmd.Flags().GetString(\"version\")\n\n\t\tclient := registry.New(config.Conf.Registry,\n\t\t\tregistry.WithAuth(\n\t\t\t\tos.Getenv(\"YAO_REGISTRY_USER\"),\n\t\t\t\tos.Getenv(\"YAO_REGISTRY_PASS\"),\n\t\t\t),\n\t\t)\n\n\t\tmgr := mcpmgr.New(client, config.Conf.Root, nil)\n\t\tif err := mgr.Add(pkgID, mcpmgr.AddOptions{\n\t\t\tVersion: version,\n\t\t\tForce: mcpAddForce,\n\t\t}); err != nil {\n\t\t\tfmt.Fprintf(os.Stderr, \"Error: %v\\n\", err)\n\t\t\tos.Exit(1)\n\t\t}\n\t},\n}\n\nfunc init() {\n\tAddCmd.Flags().StringP(\"version\", \"v\", \"latest\", L(\"Package version or dist-tag\"))\n\tAddCmd.Flags().BoolVarP(&mcpAddForce, \"force\", \"\", false, L(\"Force reinstall\"))\n\tAddCmd.PersistentFlags().StringVarP(&appPath, \"app\", \"a\", \"\", L(\"Application directory\"))\n\tAddCmd.PersistentFlags().StringVarP(&envFile, \"env\", \"e\", \"\", L(\"Environment file\"))\n}\n" }, { "path": "cmd/mcp/fork.go", "content": "package mcp\n\nimport (\n\t\"fmt\"\n\t\"os\"\n\n\t\"github.com/spf13/cobra\"\n\t\"github.com/yaoapp/yao/config\"\n\t\"github.com/yaoapp/yao/registry\"\n\tmcpmgr \"github.com/yaoapp/yao/registry/manager/mcp\"\n)\n\n// ForkCmd implements \"yao mcp fork @scope/name [@target-scope]\"\nvar ForkCmd = &cobra.Command{\n\tUse: \"fork [package] [target-scope]\",\n\tShort: L(\"Fork an MCP to a local scope\"),\n\tLong: L(\"Fork an MCP for local modification. Example: yao mcp fork @yao/rag-tools\"),\n\tArgs: cobra.RangeArgs(1, 2),\n\tRun: func(cmd *cobra.Command, args []string) {\n\t\tBoot()\n\t\tpkgID := args[0]\n\t\tvar targetScope string\n\t\tif len(args) > 1 {\n\t\t\ttargetScope = args[1]\n\t\t}\n\n\t\tclient := registry.New(config.Conf.Registry,\n\t\t\tregistry.WithAuth(\n\t\t\t\tos.Getenv(\"YAO_REGISTRY_USER\"),\n\t\t\t\tos.Getenv(\"YAO_REGISTRY_PASS\"),\n\t\t\t),\n\t\t)\n\n\t\tmgr := mcpmgr.New(client, config.Conf.Root, nil)\n\t\tif err := mgr.Fork(pkgID, mcpmgr.ForkOptions{\n\t\t\tTargetScope: targetScope,\n\t\t}); err != nil {\n\t\t\tfmt.Fprintf(os.Stderr, \"Error: %v\\n\", err)\n\t\t\tos.Exit(1)\n\t\t}\n\t},\n}\n\nfunc init() {\n\tForkCmd.PersistentFlags().StringVarP(&appPath, \"app\", \"a\", \"\", L(\"Application directory\"))\n\tForkCmd.PersistentFlags().StringVarP(&envFile, \"env\", \"e\", \"\", L(\"Environment file\"))\n}\n" }, { "path": "cmd/mcp/mcp.go", "content": "package mcp\n\nimport (\n\t\"os\"\n\t\"path/filepath\"\n\n\t\"github.com/yaoapp/kun/exception\"\n\t\"github.com/yaoapp/yao/config\"\n)\n\nvar appPath string\nvar envFile string\n\nvar langs = map[string]string{\n\t\"Install an MCP package from the registry\": \"从注册中心安装 MCP 包\",\n\t\"Update an installed MCP package\": \"更新已安装的 MCP 包\",\n\t\"Push an MCP package to the registry\": \"推送 MCP 包到注册中心\",\n\t\"Fork an MCP to a local scope\": \"Fork 一个 MCP 到本地范围\",\n\t\"Package version or dist-tag\": \"包版本或 dist-tag\",\n\t\"Force reinstall\": \"强制重新安装\",\n\t\"Package version (required)\": \"包版本 (必填)\",\n\t\"Target version or dist-tag\": \"目标版本或 dist-tag\",\n\t\"Application directory\": \"应用目录\",\n\t\"Environment file\": \"环境变量文件\",\n}\n\n// L Language switch\nfunc L(words string) string {\n\tvar lang = os.Getenv(\"YAO_LANG\")\n\tif lang == \"\" {\n\t\treturn words\n\t}\n\tif trans, has := langs[words]; has {\n\t\treturn trans\n\t}\n\treturn words\n}\n\n// Boot sets the configuration\nfunc Boot() {\n\troot := config.Conf.Root\n\tif appPath != \"\" {\n\t\tr, err := filepath.Abs(appPath)\n\t\tif err != nil {\n\t\t\texception.New(\"Root error %s\", 500, err.Error()).Throw()\n\t\t}\n\t\troot = r\n\t}\n\n\tif envFile != \"\" {\n\t\tconfig.Conf = config.LoadFromWithRoot(envFile, root)\n\t} else {\n\t\tconfig.Conf = config.LoadFromWithRoot(filepath.Join(root, \".env\"), root)\n\t}\n\n\tconfig.ApplyMode()\n}\n" }, { "path": "cmd/mcp/push.go", "content": "package mcp\n\nimport (\n\t\"fmt\"\n\t\"os\"\n\n\t\"github.com/spf13/cobra\"\n\t\"github.com/yaoapp/yao/config\"\n\t\"github.com/yaoapp/yao/registry\"\n\tmcpmgr \"github.com/yaoapp/yao/registry/manager/mcp\"\n)\n\n// PushCmd implements \"yao mcp push scope.name --version x.y.z\"\nvar PushCmd = &cobra.Command{\n\tUse: \"push [yao-id]\",\n\tShort: L(\"Push an MCP package to the registry\"),\n\tLong: L(\"Package and push an MCP to the registry. Example: yao mcp push max.rag-tools --version 1.0.0\"),\n\tArgs: cobra.ExactArgs(1),\n\tRun: func(cmd *cobra.Command, args []string) {\n\t\tBoot()\n\t\tyaoID := args[0]\n\t\tversion, _ := cmd.Flags().GetString(\"version\")\n\t\tforce, _ := cmd.Flags().GetBool(\"force\")\n\n\t\tclient := registry.New(config.Conf.Registry,\n\t\t\tregistry.WithAuth(\n\t\t\t\tos.Getenv(\"YAO_REGISTRY_USER\"),\n\t\t\t\tos.Getenv(\"YAO_REGISTRY_PASS\"),\n\t\t\t),\n\t\t)\n\n\t\tmgr := mcpmgr.New(client, config.Conf.Root, nil)\n\t\tif err := mgr.Push(yaoID, mcpmgr.PushOptions{\n\t\t\tVersion: version,\n\t\t\tForce: force,\n\t\t}); err != nil {\n\t\t\tfmt.Fprintf(os.Stderr, \"Error: %v\\n\", err)\n\t\t\tos.Exit(1)\n\t\t}\n\t},\n}\n\nfunc init() {\n\tPushCmd.Flags().StringP(\"version\", \"v\", \"\", L(\"Package version (required)\"))\n\tPushCmd.Flags().Bool(\"force\", false, L(\"Overwrite existing version\"))\n\tPushCmd.PersistentFlags().StringVarP(&appPath, \"app\", \"a\", \"\", L(\"Application directory\"))\n\tPushCmd.PersistentFlags().StringVarP(&envFile, \"env\", \"e\", \"\", L(\"Environment file\"))\n}\n" }, { "path": "cmd/mcp/update.go", "content": "package mcp\n\nimport (\n\t\"fmt\"\n\t\"os\"\n\n\t\"github.com/spf13/cobra\"\n\t\"github.com/yaoapp/yao/config\"\n\t\"github.com/yaoapp/yao/registry\"\n\tmcpmgr \"github.com/yaoapp/yao/registry/manager/mcp\"\n)\n\n// UpdateCmd implements \"yao mcp update @scope/name\"\nvar UpdateCmd = &cobra.Command{\n\tUse: \"update [package]\",\n\tShort: L(\"Update an installed MCP package\"),\n\tLong: L(\"Update an installed MCP package. Example: yao mcp update @yao/rag-tools\"),\n\tArgs: cobra.ExactArgs(1),\n\tRun: func(cmd *cobra.Command, args []string) {\n\t\tBoot()\n\t\tpkgID := args[0]\n\t\tversion, _ := cmd.Flags().GetString(\"version\")\n\n\t\tclient := registry.New(config.Conf.Registry,\n\t\t\tregistry.WithAuth(\n\t\t\t\tos.Getenv(\"YAO_REGISTRY_USER\"),\n\t\t\t\tos.Getenv(\"YAO_REGISTRY_PASS\"),\n\t\t\t),\n\t\t)\n\n\t\tmgr := mcpmgr.New(client, config.Conf.Root, nil)\n\t\tif err := mgr.Update(pkgID, mcpmgr.UpdateOptions{\n\t\t\tVersion: version,\n\t\t}); err != nil {\n\t\t\tfmt.Fprintf(os.Stderr, \"Error: %v\\n\", err)\n\t\t\tos.Exit(1)\n\t\t}\n\t},\n}\n\nfunc init() {\n\tUpdateCmd.Flags().StringP(\"version\", \"v\", \"latest\", L(\"Target version or dist-tag\"))\n\tUpdateCmd.PersistentFlags().StringVarP(&appPath, \"app\", \"a\", \"\", L(\"Application directory\"))\n\tUpdateCmd.PersistentFlags().StringVarP(&envFile, \"env\", \"e\", \"\", L(\"Environment file\"))\n}\n" }, { "path": "cmd/migrate.go", "content": "package cmd\n\nimport (\n\t\"fmt\"\n\t\"os\"\n\n\t\"github.com/fatih/color\"\n\t\"github.com/spf13/cobra\"\n\t\"github.com/yaoapp/gou/model\"\n\t\"github.com/yaoapp/gou/process\"\n\t\"github.com/yaoapp/kun/exception\"\n\t\"github.com/yaoapp/yao/config\"\n\t\"github.com/yaoapp/yao/engine\"\n\t\"github.com/yaoapp/yao/share\"\n)\n\nvar name string\nvar force bool = false\nvar resetModel bool = false\nvar migrateCmd = &cobra.Command{\n\tUse: \"migrate\",\n\tShort: L(\"Update database schema\"),\n\tLong: L(\"Update database schema\"),\n\tRun: func(cmd *cobra.Command, args []string) {\n\t\tdefer func() {\n\t\t\terr := exception.Catch(recover())\n\t\t\tif err != nil {\n\t\t\t\tfmt.Println(color.RedString(L(\"Fatal: %s\"), err.Error()))\n\t\t\t}\n\t\t}()\n\n\t\tBoot()\n\n\t\tif !force && config.Conf.Mode == \"production\" {\n\t\t\tfmt.Println(color.WhiteString(L(\"TRY:\")), color.GreenString(\"%s migrate --force\", share.BUILDNAME))\n\t\t\texception.New(L(\"Migrate is not allowed on production mode.\"), 403).Throw()\n\t\t}\n\n\t\t// 加载数据模型\n\t\tloadWarnings, err := engine.Load(config.Conf, engine.LoadOption{Action: \"migrate\"})\n\t\tif err != nil {\n\t\t\tfmt.Println(color.RedString(L(\"Fatal: %s\"), err.Error()))\n\t\t\tos.Exit(1)\n\t\t}\n\n\t\tif len(loadWarnings) > 0 {\n\t\t\tfor _, warning := range loadWarnings {\n\t\t\t\tfmt.Println(color.YellowString(\"[%s] %s\", warning.Widget, warning.Error))\n\t\t\t}\n\t\t}\n\n\t\tif name != \"\" {\n\t\t\tmod, has := model.Models[name]\n\t\t\tif !has {\n\t\t\t\tfmt.Println(color.RedString(L(\"Model: %s does not exits\"), name))\n\t\t\t\treturn\n\t\t\t}\n\n\t\t\tfmt.Print(color.WhiteString(fmt.Sprintf(L(\"Update schema model: %s (%s) \"), mod.Name, mod.MetaData.Table.Name)) + \"\\t\")\n\t\t\tif resetModel {\n\t\t\t\terr := mod.DropTable()\n\t\t\t\tif err != nil {\n\t\t\t\t\tfmt.Print(color.RedString(fmt.Sprintf(L(\"FAILURE\\n%s\"), err.Error())) + \"\\n\")\n\t\t\t\t\treturn\n\t\t\t\t}\n\t\t\t}\n\n\t\t\terr := mod.Migrate(false)\n\t\t\tif err != nil {\n\t\t\t\tfmt.Print(color.RedString(fmt.Sprintf(L(\"FAILURE\\n%s\"), err.Error())) + \"\\n\")\n\t\t\t\treturn\n\t\t\t}\n\n\t\t\tfmt.Print(color.GreenString(L(\"SUCCESS\")) + \"\\n\")\n\t\t\treturn\n\t\t}\n\n\t\t// Do Stuff Here\n\t\tfor _, mod := range model.Models {\n\t\t\tfmt.Print(color.WhiteString(fmt.Sprintf(L(\"Update schema model: %s (%s) \"), mod.Name, mod.MetaData.Table.Name)) + \"\\t\")\n\n\t\t\tif resetModel {\n\t\t\t\terr := mod.DropTable()\n\t\t\t\tif err != nil {\n\t\t\t\t\tfmt.Print(color.RedString(fmt.Sprintf(L(\"FAILURE\\n%s\"), err.Error())) + \"\\n\")\n\t\t\t\t\tcontinue\n\t\t\t\t}\n\t\t\t}\n\n\t\t\terr := mod.Migrate(false)\n\t\t\tif err != nil {\n\t\t\t\tfmt.Print(color.RedString(fmt.Sprintf(L(\"FAILURE\\n%s\"), err.Error())) + \"\\n\")\n\t\t\t\tcontinue\n\t\t\t}\n\t\t\tfmt.Print(color.GreenString(L(\"SUCCESS\")) + \"\\n\")\n\t\t}\n\n\t\t// After Migrate Hook\n\t\tif share.App.AfterMigrate != \"\" {\n\t\t\toption := map[string]any{\"force\": force, \"reset\": resetModel, \"mode\": config.Conf.Mode}\n\t\t\tp, err := process.Of(share.App.AfterMigrate, option)\n\t\t\tif err != nil {\n\t\t\t\tfmt.Println(color.RedString(L(\"AfterMigrate: %s %v\"), share.App.AfterMigrate, err))\n\t\t\t\treturn\n\t\t\t}\n\n\t\t\t_, err = p.Exec()\n\t\t\tif err != nil {\n\t\t\t\tfmt.Println(color.RedString(L(\"AfterMigrate: %s %v\"), share.App.AfterMigrate, err))\n\t\t\t}\n\t\t}\n\n\t\t// fmt.Println(color.GreenString(L(\"✨DONE✨\")))\n\t},\n}\n\nfunc init() {\n\tmigrateCmd.PersistentFlags().StringVarP(&name, \"name\", \"n\", \"\", L(\"Model name\"))\n\tmigrateCmd.PersistentFlags().BoolVarP(&force, \"force\", \"\", false, L(\"Force migrate\"))\n\tmigrateCmd.PersistentFlags().BoolVarP(&resetModel, \"reset\", \"\", false, L(\"Drop the table if exist\"))\n}\n" }, { "path": "cmd/pack.go", "content": "package cmd\n\nimport (\n\t\"bufio\"\n\t\"fmt\"\n\t\"os\"\n\t\"path/filepath\"\n\t\"strings\"\n\n\t\"github.com/fatih/color\"\n\t\"github.com/spf13/cobra\"\n\t\"github.com/yaoapp/gou/application/yaz\"\n\t\"github.com/yaoapp/yao/config\"\n\t\"github.com/yaoapp/yao/pack\"\n)\n\nvar packOutput = \"\"\nvar packLicense = \"\"\n\nvar packCmd = &cobra.Command{\n\tUse: \"pack\",\n\tShort: L(\"Package the application\"),\n\tLong: L(\"Package the application into a single file\"),\n\tRun: func(cmd *cobra.Command, args []string) {\n\n\t\tcfg := config.Conf\n\t\toutput, err := filepath.Abs(filepath.Join(cfg.Root, \"dist\"))\n\t\tif err != nil {\n\t\t\tcolor.Red(err.Error())\n\t\t}\n\n\t\tif packOutput != \"\" {\n\t\t\toutput, err = filepath.Abs(packOutput)\n\t\t\tif err != nil {\n\t\t\t\tcolor.Red(err.Error())\n\t\t\t}\n\t\t}\n\n\t\tstat, err := os.Stat(output)\n\t\tif err != nil && os.IsNotExist(err) {\n\t\t\tcolor.Green(\"Creating directory %s\", output)\n\t\t\terr = os.MkdirAll(output, 0755)\n\t\t\tif err != nil {\n\t\t\t\tcolor.Red(err.Error())\n\t\t\t\tos.Exit(1)\n\t\t\t}\n\t\t} else if err != nil {\n\t\t\tcolor.Red(err.Error())\n\t\t\tos.Exit(1)\n\n\t\t} else if !stat.IsDir() {\n\t\t\tcolor.Red(\"Output directory %s is not a directory.\\n\", output)\n\t\t\tos.Exit(1)\n\t\t}\n\n\t\toutputFile := filepath.Join(output, \"app.yaz\")\n\t\t_, err = os.Stat(outputFile)\n\t\tif !os.IsNotExist(err) {\n\t\t\tcolor.Yellow(\"%s already exists\", outputFile)\n\t\t\tfmt.Printf(\"%s\", color.RedString(\"Do you want to overwrite it? (y/n): \"))\n\t\t\tscanner := bufio.NewScanner(os.Stdin)\n\t\t\tif scanner.Scan() {\n\t\t\t\tif strings.ToLower(scanner.Text()) != \"y\" {\n\t\t\t\t\tos.Exit(0)\n\t\t\t\t\treturn\n\t\t\t\t}\n\t\t\t}\n\t\t}\n\n\t\tos.Remove(outputFile)\n\t\tif packLicense != \"\" {\n\t\t\tpack.SetCipher(packLicense)\n\t\t\terr = yaz.PackTo(cfg.Root, outputFile, pack.Cipher)\n\n\t\t} else {\n\t\t\terr = yaz.CompressTo(cfg.Root, outputFile)\n\t\t}\n\n\t\tif err != nil {\n\t\t\tcolor.Red(err.Error())\n\t\t\tos.Exit(1)\n\t\t}\n\n\t\tcolor.Green(\"Packaged to %s\", outputFile)\n\t},\n}\n\nfunc init() {\n\tpackCmd.PersistentFlags().StringVarP(&packOutput, \"output\", \"o\", \"\", L(\"Output Directory\"))\n\tpackCmd.PersistentFlags().StringVarP(&packLicense, \"license\", \"l\", \"\", L(\"Pack with the license\"))\n}\n" }, { "path": "cmd/restore.go", "content": "package cmd\n\nimport (\n\t\"archive/zip\"\n\t\"errors\"\n\t\"fmt\"\n\t\"io\"\n\t\"io/ioutil\"\n\t\"os\"\n\t\"path/filepath\"\n\t\"strings\"\n\t\"time\"\n\n\t\"github.com/fatih/color\"\n\t\"github.com/spf13/cobra\"\n\t\"github.com/yaoapp/gou/model\"\n\t\"github.com/yaoapp/kun/exception\"\n\t\"github.com/yaoapp/yao/config\"\n\t\"github.com/yaoapp/yao/engine\"\n\t\"github.com/yaoapp/yao/share\"\n)\n\nvar restoreForce bool = false\nvar migrateNoInsert bool = false\nvar restoreCmd = &cobra.Command{\n\tUse: \"restore\",\n\tShort: L(\"Restore the application data\"),\n\tLong: L(\"Restore the application data\"),\n\tRun: func(cmd *cobra.Command, args []string) {\n\t\tdefer func() {\n\t\t\terr := exception.Catch(recover())\n\t\t\tif err != nil {\n\t\t\t\tfmt.Println(color.RedString(L(\"Fatal: %s\"), err.Error()))\n\t\t\t}\n\t\t}()\n\n\t\tif len(args) < 1 {\n\t\t\tfmt.Println(color.RedString(L(\"Not enough arguments\")))\n\t\t\tfmt.Println(color.WhiteString(share.BUILDNAME + \" help\"))\n\t\t\tos.Exit(1)\n\t\t}\n\n\t\tzipfile, err := filepath.Abs(args[0])\n\t\tif err != nil {\n\t\t\tfmt.Println(color.RedString(L(\"Fatal: %s\"), err.Error()))\n\t\t\tos.Exit(1)\n\t\t}\n\n\t\tBoot()\n\n\t\tif !restoreForce && config.Conf.Mode == \"production\" {\n\t\t\tfmt.Println(color.WhiteString(L(\"TRY:\")), color.GreenString(\"%s restore --force\", share.BUILDNAME))\n\t\t\texception.New(L(\"Retore is not allowed on production mode.\"), 403).Throw()\n\t\t}\n\n\t\t// Unzip files\n\t\tdst := unzipFile(zipfile, func(file string) {\n\t\t\tfmt.Printf(\"\\r%s\", strings.Repeat(\" \", 80))\n\t\t\tfmt.Printf(\"\\r%s\", color.GreenString(L(\"Unzip the file: %s\"), file))\n\t\t})\n\n\t\t// 加载数据模型\n\t\tloadWarnings, err := engine.Load(config.Conf, engine.LoadOption{Action: \"restore\"})\n\t\tif err != nil {\n\t\t\tfmt.Println(color.RedString(L(\"Fatal: %s\"), err.Error()))\n\t\t\tos.Exit(1)\n\t\t}\n\n\t\tif len(loadWarnings) > 0 {\n\t\t\tfor _, warning := range loadWarnings {\n\t\t\t\tfmt.Println(color.YellowString(\"[%s] %s\", warning.Widget, warning.Error))\n\t\t\t}\n\t\t}\n\n\t\t// Restore models\n\t\trestoreModels(filepath.Join(dst, \"model\"), []model.MigrateOption{\n\t\t\tmodel.WithDonotInsertValues(migrateNoInsert),\n\t\t})\n\n\t\t// Restore Data\n\t\trestoreData(filepath.Join(dst, \"data\"))\n\n\t\t// Clean\n\t\tos.RemoveAll(dst)\n\n\t\tfmt.Println(color.GreenString(L(\"✨DONE✨\")))\n\t},\n}\n\nfunc init() {\n\trestoreCmd.PersistentFlags().BoolVarP(&restoreForce, \"force\", \"\", false, L(\"Force restore\"))\n\trestoreCmd.PersistentFlags().BoolVarP(&migrateNoInsert, \"migrate-no-insert\", \"\", false, L(\"Do not insert values when migrating\"))\n}\n\nfunc restoreData(basePath string) {\n\n\t_, err := os.Stat(basePath)\n\tif err != nil {\n\t\tfmt.Println(color.RedString(L(\"Fatal: %s\"), err.Error()))\n\t\tos.Exit(1)\n\t}\n\n\t// Clean Data\n\tdataPath := filepath.Join(config.Conf.Root, \"data\")\n\t_, err = os.Stat(dataPath)\n\tif err == nil {\n\t\tos.RemoveAll(dataPath)\n\t}\n\n\terr = os.Rename(basePath, dataPath)\n\tif err != nil {\n\t\tfmt.Println(color.RedString(L(\"Fatal: %s\"), err.Error()))\n\t\tos.Exit(1)\n\t}\n}\n\nfunc restoreModels(basePath string, migOpts []model.MigrateOption) {\n\n\tfiles, err := ioutil.ReadDir(basePath)\n\tif err != nil {\n\t\tfmt.Println(color.RedString(L(\"Fatal: %s\"), err.Error()))\n\t\tos.Exit(1)\n\t}\n\n\t// Migrate models\n\tfor _, mod := range model.Models {\n\t\tfmt.Printf(\"\\r%s\", strings.Repeat(\" \", 80))\n\t\tfmt.Print(color.GreenString(fmt.Sprintf(L(\"\\rUpdate schema model: %s (%s) \"), mod.Name, mod.MetaData.Table.Name)))\n\t\terr := mod.Migrate(true, migOpts...)\n\t\tif err != nil {\n\t\t\tfmt.Println(color.RedString(L(\"Fatal: %s\"), err.Error()))\n\t\t\tos.Exit(1)\n\t\t}\n\t}\n\n\tfmt.Println(\"\")\n\n\tfor _, file := range files {\n\t\tnamer := strings.Split(file.Name(), \".\")\n\t\tname := strings.Join(namer[:len(namer)-2], \".\")\n\t\tif mod, has := model.Models[name]; has {\n\t\t\tfmt.Printf(\"\\r%s\", strings.Repeat(\" \", 80))\n\t\t\tfmt.Print(color.GreenString(fmt.Sprintf(L(\"\\rRestore model: %s (%s) \"), mod.Name, mod.MetaData.Table.Name)))\n\t\t\terr := mod.Import(filepath.Join(basePath, file.Name()))\n\t\t\tif err != nil {\n\t\t\t\tfmt.Println(color.RedString(L(\"Fatal: %s\"), err.Error()))\n\t\t\t\tos.Exit(1)\n\t\t\t}\n\t\t}\n\t}\n}\n\nfunc unzipFile(file string, process func(file string)) string {\n\t_, err := os.Stat(file)\n\n\tif errors.Is(err, os.ErrNotExist) {\n\t\tfmt.Println(color.RedString(\"%s not exists\", file))\n\t\tos.Exit(1)\n\t}\n\n\tif err != nil {\n\t\tfmt.Println(color.RedString(L(\"Fatal: %s\"), err.Error()))\n\t\tos.Exit(1)\n\t}\n\n\tdst := filepath.Join(os.TempDir(), fmt.Sprintf(\"%s-%s\", filepath.Base(file), time.Now().Format(\"20060102150405\")))\n\tos.MkdirAll(dst, 0755)\n\n\tarchive, err := zip.OpenReader(file)\n\tif err != nil {\n\t\tfmt.Println(color.RedString(L(\"Fatal: %s\"), err.Error()))\n\t\tos.Exit(1)\n\t}\n\tdefer archive.Close()\n\n\tfor _, f := range archive.File {\n\t\tfilePath := filepath.Join(dst, f.Name)\n\t\tprocess(f.Name)\n\n\t\tif !strings.HasPrefix(filePath, filepath.Clean(dst)+string(os.PathSeparator)) {\n\t\t\tfmt.Println(color.RedString(L(\"Fatal: invalid file path\")))\n\t\t\tos.Exit(1)\n\t\t}\n\t\tif f.FileInfo().IsDir() {\n\t\t\tos.MkdirAll(filePath, os.ModePerm)\n\t\t\tcontinue\n\t\t}\n\n\t\tif err := os.MkdirAll(filepath.Dir(filePath), os.ModePerm); err != nil {\n\t\t\tfmt.Println(color.RedString(L(\"Fatal: %s\"), err.Error()))\n\t\t\tos.Exit(1)\n\t\t}\n\n\t\tdstFile, err := os.OpenFile(filePath, os.O_WRONLY|os.O_CREATE|os.O_TRUNC, f.Mode())\n\t\tif err != nil {\n\t\t\tfmt.Println(color.RedString(L(\"Fatal: %s\"), err.Error()))\n\t\t\tos.Exit(1)\n\t\t}\n\n\t\tfileInArchive, err := f.Open()\n\t\tif err != nil {\n\t\t\tfmt.Println(color.RedString(L(\"Fatal: %s\"), err.Error()))\n\t\t\tos.Exit(1)\n\t\t}\n\n\t\tif _, err := io.Copy(dstFile, fileInArchive); err != nil {\n\t\t\tfmt.Println(color.RedString(L(\"Fatal: %s\"), err.Error()))\n\t\t\tos.Exit(1)\n\t\t}\n\n\t\tdstFile.Close()\n\t\tfileInArchive.Close()\n\t}\n\n\treturn dst\n}\n" }, { "path": "cmd/robot/add.go", "content": "package robot\n\nimport (\n\t\"fmt\"\n\t\"os\"\n\n\t\"github.com/spf13/cobra\"\n\t\"github.com/yaoapp/yao/config\"\n\t\"github.com/yaoapp/yao/registry\"\n\trobotmgr \"github.com/yaoapp/yao/registry/manager/robot\"\n)\n\n// AddCmd implements \"yao robot add @scope/name --team TEAM_ID\"\nvar AddCmd = &cobra.Command{\n\tUse: \"add [package]\",\n\tShort: L(\"Install a robot package from the registry\"),\n\tLong: L(\"Install a robot and its dependencies. Example: yao robot add @yao/keeper --team team-123\"),\n\tArgs: cobra.ExactArgs(1),\n\tRun: func(cmd *cobra.Command, args []string) {\n\t\tBoot()\n\t\tpkgID := args[0]\n\t\tversion, _ := cmd.Flags().GetString(\"version\")\n\t\tteamID, _ := cmd.Flags().GetString(\"team\")\n\n\t\tclient := registry.New(config.Conf.Registry,\n\t\t\tregistry.WithAuth(\n\t\t\t\tos.Getenv(\"YAO_REGISTRY_USER\"),\n\t\t\t\tos.Getenv(\"YAO_REGISTRY_PASS\"),\n\t\t\t),\n\t\t)\n\n\t\tmgr := robotmgr.New(client, config.Conf.Root, nil)\n\t\trobot, err := mgr.Add(pkgID, robotmgr.AddOptions{\n\t\t\tVersion: version,\n\t\t\tTeamID: teamID,\n\t\t})\n\t\tif err != nil {\n\t\t\tfmt.Fprintf(os.Stderr, \"Error: %v\\n\", err)\n\t\t\tos.Exit(1)\n\t\t}\n\n\t\t// The actual member record creation requires database access.\n\t\t// In P0 we print the robot config for the CLI layer to handle.\n\t\tfmt.Printf(\"Robot ready: %s (display_name: %s)\\n\", pkgID, robot.DisplayName)\n\t\tfmt.Println(\"Note: Member record must be created via Mission Control or database.\")\n\t},\n}\n\nfunc init() {\n\tAddCmd.Flags().StringP(\"version\", \"v\", \"latest\", L(\"Package version or dist-tag\"))\n\tAddCmd.Flags().StringP(\"team\", \"t\", \"\", L(\"Team ID (required)\"))\n\tAddCmd.PersistentFlags().StringVarP(&appPath, \"app\", \"a\", \"\", L(\"Application directory\"))\n\tAddCmd.PersistentFlags().StringVarP(&envFile, \"env\", \"e\", \"\", L(\"Environment file\"))\n}\n" }, { "path": "cmd/robot/robot.go", "content": "package robot\n\nimport (\n\t\"os\"\n\t\"path/filepath\"\n\n\t\"github.com/yaoapp/kun/exception\"\n\t\"github.com/yaoapp/yao/config\"\n)\n\nvar appPath string\nvar envFile string\n\nvar langs = map[string]string{\n\t\"Install a robot package from the registry\": \"从注册中心安装 Robot 包\",\n\t\"Team ID (required)\": \"团队 ID (必填)\",\n\t\"Package version or dist-tag\": \"包版本或 dist-tag\",\n\t\"Application directory\": \"应用目录\",\n\t\"Environment file\": \"环境变量文件\",\n}\n\n// L Language switch\nfunc L(words string) string {\n\tvar lang = os.Getenv(\"YAO_LANG\")\n\tif lang == \"\" {\n\t\treturn words\n\t}\n\tif trans, has := langs[words]; has {\n\t\treturn trans\n\t}\n\treturn words\n}\n\n// Boot sets the configuration\nfunc Boot() {\n\troot := config.Conf.Root\n\tif appPath != \"\" {\n\t\tr, err := filepath.Abs(appPath)\n\t\tif err != nil {\n\t\t\texception.New(\"Root error %s\", 500, err.Error()).Throw()\n\t\t}\n\t\troot = r\n\t}\n\n\tif envFile != \"\" {\n\t\tconfig.Conf = config.LoadFromWithRoot(envFile, root)\n\t} else {\n\t\tconfig.Conf = config.LoadFromWithRoot(filepath.Join(root, \".env\"), root)\n\t}\n\n\tconfig.ApplyMode()\n}\n" }, { "path": "cmd/root.go", "content": "package cmd\n\nimport (\n\t\"fmt\"\n\t\"os\"\n\t\"path/filepath\"\n\n\t\"github.com/spf13/cobra\"\n\t\"github.com/yaoapp/kun/exception\"\n\t\"github.com/yaoapp/yao/cmd/agent\"\n\t\"github.com/yaoapp/yao/cmd/mcp\"\n\t\"github.com/yaoapp/yao/cmd/robot\"\n\t\"github.com/yaoapp/yao/cmd/sui\"\n\t\"github.com/yaoapp/yao/config\"\n\t\"github.com/yaoapp/yao/pack\"\n\t\"github.com/yaoapp/yao/share\"\n)\n\nvar appPath string\nvar yazFile string\nvar licenseKey string\n\nvar lang = os.Getenv(\"YAO_LANG\")\nvar langs = map[string]string{\n\t\"Start Engine\": \"启动 YAO 应用引擎\",\n\t\"Get an application\": \"下载应用源码\",\n\t\"One or more arguments are not correct\": \"参数错误\",\n\t\"Application directory\": \"指定应用路径\",\n\t\"Environment file\": \"指定环境变量文件\",\n\t\"Help for yao\": \"显示命令帮助文档\",\n\t\"Show app configure\": \"显示应用配置信息\",\n\t\"Update database schema\": \"更新数据表结构\",\n\t\"Execute process\": \"运行处理器\",\n\t\"Show version\": \"显示当前版本号\",\n\t\"Development mode\": \"使用开发模式启动\",\n\t\"Enabled unstable features\": \"启用内测功能\",\n\t\"Fatal: %s\": \"失败: %s\",\n\t\"Service stopped\": \"服务已关闭\",\n\t\"API\": \" API接口\",\n\t\"API List\": \"API列表\",\n\t\"Root\": \"应用目录\",\n\t\"Data\": \"数据目录\",\n\t\"Frontend\": \"前台地址\",\n\t\"Dashboard\": \"管理后台\",\n\t\"Not enough arguments\": \"参数错误: 缺少参数\",\n\t\"Run: %s\": \"运行: %s\",\n\t\"Arguments: %s\": \"参数错误: %s\",\n\t\"%s Response\": \"%s 返回结果\",\n\t\"Update schema model: %s (%s) \": \"更新表结构 model: %s (%s)\",\n\t\"Model name\": \"模型名称\",\n\t\"Initialize project\": \"项目初始化\",\n\t\"✨DONE✨\": \"✨完成✨\",\n\t\"NEXT:\": \"下一步:\",\n\t\"Listening\": \" 监听\",\n\t\"✨LISTENING✨\": \"✨服务正在运行✨\",\n\t\"✨STOPPED✨\": \"✨服务已停止✨\",\n\t\"SessionPort\": \"会话服务端口\",\n\t\"Force migrate\": \"强制更新数据表结构\",\n\t\"Migrate is not allowed on production mode.\": \"Migrate 不能再生产环境下使用\",\n\t\"Upgrade yao to latest version\": \"升级 yao 到最新版本\",\n\t\"🎉Current version is the latest🎉\": \"🎉当前版本是最新的🎉\",\n\t\"Do you want to update to %s ? (y/n): \": \"是否更新到 %s ? (y/n): \",\n\t\"Invalid input\": \"输入错误\",\n\t\"Canceled upgrade\": \"已取消更新\",\n\t\"Error occurred while updating binary: %s\": \"更新二进制文件时出错: %s\",\n\t\"🎉Successfully updated to version: %s🎉\": \"🎉成功更新到版本: %s🎉\",\n\t\"Print all version information\": \"显示详细版本信息\",\n\t\"SUI Template Engine\": \"SUI 模板引擎命令\",\n\t\"MCP commands\": \"MCP 包管理命令\",\n\t\"MCP package management commands\": \"MCP 包管理命令\",\n\t\"Robot commands\": \"Robot 包管理命令\",\n\t\"Robot package management commands\": \"Robot 包管理命令\",\n}\n\n// L Language switch\nfunc L(words string) string {\n\tif lang == \"\" {\n\t\treturn words\n\t}\n\n\tif trans, has := langs[words]; has {\n\t\treturn trans\n\t}\n\treturn words\n}\n\n// RootCmd export the rootCmd to support customized commands when use yao as lib\nvar RootCmd = rootCmd\n\nvar rootCmd = &cobra.Command{\n\tUse: share.BUILDNAME,\n\tShort: \"Yao App Engine\",\n\tLong: `Yao App Engine`,\n\tCompletionOptions: cobra.CompletionOptions{\n\t\tDisableDefaultCmd: true,\n\t},\n\tRun: func(cmd *cobra.Command, args []string) {\n\t\tif len(args) > 0 {\n\t\t\tswitch args[0] {\n\t\t\tcase \"fuxi\":\n\t\t\t\tfuxi()\n\t\t\t\treturn\n\t\t\t}\n\t\t\tfmt.Fprintln(os.Stderr, L(\"One or more arguments are not correct\"), args)\n\t\t\tos.Exit(1)\n\t\t\treturn\n\t\t}\n\t\t// No arguments - show help\n\t\tcmd.Help()\n\t},\n}\n\nvar suiCmd = &cobra.Command{\n\tUse: \"sui\",\n\tShort: L(\"SUI Template Engine\"),\n\tLong: L(\"SUI Template Engine\"),\n\tCompletionOptions: cobra.CompletionOptions{\n\t\tDisableDefaultCmd: true,\n\t},\n\tRun: func(cmd *cobra.Command, args []string) {\n\t\tcmd.Help()\n\t},\n}\n\nvar agentCmd = &cobra.Command{\n\tUse: \"agent\",\n\tShort: L(\"Agent commands\"),\n\tLong: L(\"Agent commands for testing and management\"),\n\tCompletionOptions: cobra.CompletionOptions{\n\t\tDisableDefaultCmd: true,\n\t},\n\tRun: func(cmd *cobra.Command, args []string) {\n\t\tcmd.Help()\n\t},\n}\n\nvar mcpCmd = &cobra.Command{\n\tUse: \"mcp\",\n\tShort: L(\"MCP commands\"),\n\tLong: L(\"MCP package management commands\"),\n\tCompletionOptions: cobra.CompletionOptions{\n\t\tDisableDefaultCmd: true,\n\t},\n\tRun: func(cmd *cobra.Command, args []string) {\n\t\tcmd.Help()\n\t},\n}\n\nvar robotCmd = &cobra.Command{\n\tUse: \"robot\",\n\tShort: L(\"Robot commands\"),\n\tLong: L(\"Robot package management commands\"),\n\tCompletionOptions: cobra.CompletionOptions{\n\t\tDisableDefaultCmd: true,\n\t},\n\tRun: func(cmd *cobra.Command, args []string) {\n\t\tcmd.Help()\n\t},\n}\n\n// Command initialize\nfunc init() {\n\n\t// Sui\n\tsuiCmd.AddCommand(sui.WatchCmd)\n\tsuiCmd.AddCommand(sui.BuildCmd)\n\tsuiCmd.AddCommand(sui.TransCmd)\n\n\t// Agent\n\tagentCmd.AddCommand(agent.TestCmd)\n\tagentCmd.AddCommand(agent.ExtractCmd)\n\tagentCmd.AddCommand(agent.AddCmd)\n\tagentCmd.AddCommand(agent.UpdateCmd)\n\tagentCmd.AddCommand(agent.PushCmd)\n\tagentCmd.AddCommand(agent.ForkCmd)\n\n\t// MCP\n\tmcpCmd.AddCommand(mcp.AddCmd)\n\tmcpCmd.AddCommand(mcp.UpdateCmd)\n\tmcpCmd.AddCommand(mcp.PushCmd)\n\tmcpCmd.AddCommand(mcp.ForkCmd)\n\n\t// Robot\n\trobotCmd.AddCommand(robot.AddCmd)\n\n\trootCmd.AddCommand(\n\t\tversionCmd,\n\t\tmigrateCmd,\n\t\tinspectCmd,\n\t\tstartCmd,\n\t\trunCmd,\n\t\tloginCmd,\n\t\tlogoutCmd,\n\t\t// getCmd,\n\t\t// dumpCmd,\n\t\t// restoreCmd,\n\t\t// socketCmd,\n\t\t// websocketCmd,\n\t\t// packCmd,\n\t\tsuiCmd,\n\t\tagentCmd,\n\t\tmcpCmd,\n\t\trobotCmd,\n\t\t// upgradeCmd,\n\t)\n\t// rootCmd.SetHelpCommand(helpCmd)\n\trootCmd.PersistentFlags().StringVarP(&appPath, \"app\", \"a\", \"\", L(\"Application directory\"))\n\trootCmd.PersistentFlags().StringVarP(&yazFile, \"file\", \"f\", \"\", L(\"Application package file\"))\n\trootCmd.PersistentFlags().StringVarP(&licenseKey, \"key\", \"k\", \"\", L(\"Application license key\"))\n}\n\n// Execute Command\nfunc Execute() {\n\tif err := rootCmd.Execute(); err != nil {\n\t\tfmt.Fprintln(os.Stderr, err)\n\t\tos.Exit(1)\n\t}\n}\n\n// Boot Setting\nfunc Boot() {\n\n\troot := config.Conf.Root\n\tif appPath != \"\" {\n\t\tr, err := filepath.Abs(appPath)\n\t\tif err != nil {\n\t\t\texception.New(\"Root error %s\", 500, err.Error()).Throw()\n\t\t}\n\t\troot = r\n\t}\n\n\tconfig.Conf = config.LoadFrom(filepath.Join(root, \".env\"))\n\n\tif share.BUILDIN {\n\t\tos.Setenv(\"YAO_APP_SOURCE\", \"::binary\")\n\t\tconfig.Conf.AppSource = \"::binary\"\n\t}\n\n\tif yazFile != \"\" {\n\t\tos.Setenv(\"YAO_APP_SOURCE\", yazFile)\n\t\tconfig.Conf.AppSource = yazFile\n\t}\n\n\tif config.Conf.Mode == \"production\" {\n\t\tconfig.Production()\n\t} else if config.Conf.Mode == \"development\" {\n\t\tconfig.Development()\n\t}\n\n\t// set license\n\tif licenseKey != \"\" {\n\t\tpack.SetCipher(licenseKey)\n\t}\n}\n" }, { "path": "cmd/run.go", "content": "package cmd\n\nimport (\n\t\"context\"\n\t\"fmt\"\n\t\"os\"\n\t\"path/filepath\"\n\t\"strings\"\n\n\t\"github.com/fatih/color\"\n\tjsoniter \"github.com/json-iterator/go\"\n\t\"github.com/spf13/cobra\"\n\t\"github.com/yaoapp/gou/helper\"\n\t\"github.com/yaoapp/gou/plugin\"\n\t\"github.com/yaoapp/gou/process\"\n\t\"github.com/yaoapp/kun/exception\"\n\t\"github.com/yaoapp/yao/config\"\n\t\"github.com/yaoapp/yao/engine\"\n\tgrpcclient \"github.com/yaoapp/yao/grpc/client\"\n\tischedule \"github.com/yaoapp/yao/schedule\"\n\t\"github.com/yaoapp/yao/share\"\n\titask \"github.com/yaoapp/yao/task\"\n)\n\nvar runSilent = false\nvar runAuthPath string\n\nvar runCmd = &cobra.Command{\n\tUse: \"run\",\n\tShort: L(\"Execute process\"),\n\tLong: L(\"Execute process\"),\n\tRun: func(cmd *cobra.Command, args []string) {\n\n\t\t// Resolve credential: --auth flag > ~/.yao/credentials > nil (local mode)\n\t\tcred := resolveCredential()\n\n\t\tif cred != nil {\n\t\t\trunGRPC(cred, args)\n\t\t\treturn\n\t\t}\n\n\t\trunLocal(args)\n\t},\n}\n\nfunc init() {\n\trunCmd.PersistentFlags().BoolVarP(&runSilent, \"silent\", \"s\", false, L(\"Silent mode\"))\n\trunCmd.PersistentFlags().StringVar(&runAuthPath, \"auth\", \"\", L(\"Path to credentials file\"))\n}\n\n// resolveCredential loads credential from --auth flag or default path.\nfunc resolveCredential() *Credential {\n\tif runAuthPath != \"\" {\n\t\tcred, err := LoadCredentialFrom(runAuthPath)\n\t\tif err != nil {\n\t\t\tcolor.Red(\" %s %s\\n\", L(\"Failed to load credentials:\"), err)\n\t\t\tos.Exit(1)\n\t\t}\n\t\treturn cred\n\t}\n\n\tcred, _ := LoadCredential()\n\treturn cred\n}\n\n// runGRPC executes a process via the remote gRPC server.\nfunc runGRPC(cred *Credential, args []string) {\n\tif len(args) < 1 {\n\t\tif !runSilent {\n\t\t\tcolor.Red(L(\"Not enough arguments\\n\"))\n\t\t\tcolor.White(share.BUILDNAME + \" help\\n\")\n\t\t} else {\n\t\t\tfmt.Print(L(\"Not enough arguments\\n\"))\n\t\t}\n\t\tos.Exit(1)\n\t}\n\n\tif cred.GRPCAddr == \"\" {\n\t\tcolor.Red(\" %s\\n\", L(\"No gRPC address in credentials. Please re-login.\"))\n\t\tos.Exit(1)\n\t}\n\n\tname := args[0]\n\tif !runSilent {\n\t\tcolor.Green(L(\"Run: %s gRPC: %s\\n\"), name, cred.GRPCAddr)\n\t}\n\n\tpargs := parseRunArgs(args)\n\n\targsJSON, err := jsoniter.Marshal(pargs)\n\tif err != nil {\n\t\tcolor.Red(\" %s %s\\n\", L(\"Arguments:\"), err.Error())\n\t\tos.Exit(1)\n\t}\n\n\ttm := grpcclient.NewTokenManager(cred.AccessToken, cred.RefreshToken, \"\")\n\tclient, err := grpcclient.Dial(cred.GRPCAddr, tm)\n\tif err != nil {\n\t\tcolor.Red(\" %s %s\\n\", L(\"gRPC connect failed:\"), err.Error())\n\t\tos.Exit(1)\n\t}\n\tdefer client.Close()\n\n\tdata, err := client.Run(context.Background(), name, argsJSON, 0)\n\tif err != nil {\n\t\tif !runSilent {\n\t\t\tcolor.Red(\" %s %s\\n\", L(\"Process:\"), err.Error())\n\t\t} else {\n\t\t\tfmt.Printf(\"%s\\n\", err.Error())\n\t\t}\n\t\tos.Exit(1)\n\t}\n\n\tif !runSilent {\n\t\tcolor.White(\"--------------------------------------\\n\")\n\t\tcolor.White(L(\"%s Response\\n\"), name)\n\t\tcolor.White(\"--------------------------------------\\n\")\n\t\tvar res interface{}\n\t\tif jsoniter.Unmarshal(data, &res) == nil {\n\t\t\thelper.Dump(res)\n\t\t} else {\n\t\t\tfmt.Printf(\"%s\\n\", data)\n\t\t}\n\t\tcolor.White(\"--------------------------------------\\n\")\n\t\tfmt.Printf(\"\\033[32m✨DONE✨\\033[0m \\033[90mgRPC: %s\\033[0m\\n\", cred.GRPCAddr)\n\t} else {\n\t\tfmt.Printf(\"%s\\n\", data)\n\t}\n}\n\n// runLocal executes a process locally (existing behavior).\nfunc runLocal(args []string) {\n\tdefer share.SessionStop()\n\tdefer plugin.KillAll()\n\n\tdefer func() {\n\t\terr := exception.Catch(recover())\n\t\tif err != nil {\n\t\t\tif !runSilent {\n\t\t\t\tcolor.Red(L(\"Fatal: %s\\n\"), err.Error())\n\t\t\t\treturn\n\t\t\t}\n\t\t\tfmt.Printf(\"%s\\n\", err.Error())\n\t\t}\n\t}()\n\n\t// Auto-detect app root if not specified\n\tif appPath == \"\" {\n\t\tcwd, err := os.Getwd()\n\t\tif err == nil {\n\t\t\tif root, err := findAppRootFromPath(cwd); err == nil {\n\t\t\t\tappPath = root\n\t\t\t}\n\t\t}\n\t}\n\n\tBoot()\n\n\t// Set Runtime Mode\n\tconfig.Conf.Runtime.Mode = \"standard\"\n\n\tcfg := config.Conf\n\tcfg.Session.IsCLI = true\n\tif len(args) < 1 {\n\t\tif !runSilent {\n\t\t\tcolor.Red(L(\"Not enough arguments\\n\"))\n\t\t\tcolor.White(share.BUILDNAME + \" help\\n\")\n\t\t\treturn\n\t\t}\n\t\tfmt.Print(L(\"Not enough arguments\\n\"))\n\t\treturn\n\t}\n\n\tloadWarnings, err := engine.Load(cfg, engine.LoadOption{Action: \"run\"})\n\tif err != nil {\n\t\tif !runSilent {\n\t\t\tcolor.Red(L(\"Engine: %s\\n\"), err.Error())\n\t\t\treturn\n\t\t}\n\n\t\tfmt.Printf(\"%s\\n\", err.Error())\n\t\treturn\n\t}\n\n\tname := args[0]\n\tif !runSilent {\n\t\tcolor.Green(L(\"Run: %s\\n\"), name)\n\t}\n\n\tpargs := parseRunArgs(args)\n\n\t// Start Tasks\n\titask.Start()\n\tdefer itask.Stop()\n\n\t// Start Schedules\n\tischedule.Start()\n\tdefer ischedule.Stop()\n\n\tp := process.NewWithContext(context.Background(), name, pargs...)\n\tres, err := p.Exec()\n\tif err != nil {\n\t\tif !runSilent {\n\t\t\tcolor.Red(L(\"Process: %s\\n\"), fmt.Sprintf(\"%s\", strings.TrimPrefix(err.Error(), \"Exception|404:\")))\n\t\t\treturn\n\t\t}\n\t\tfmt.Printf(\"%s\\n\", err.Error())\n\t\treturn\n\t}\n\n\tif !runSilent {\n\n\t\tif len(loadWarnings) > 0 {\n\t\t\tfmt.Println(color.YellowString(\"---------------------------------\"))\n\t\t\tfmt.Println(color.YellowString(L(\"Warnings\")))\n\t\t\tfmt.Println(color.YellowString(\"---------------------------------\"))\n\t\t\tfor _, warning := range loadWarnings {\n\t\t\t\tfmt.Println(color.YellowString(\"[%s] %s\", warning.Widget, warning.Error))\n\t\t\t}\n\t\t\tfmt.Printf(\"\\n\")\n\t\t}\n\n\t\tcolor.White(\"--------------------------------------\\n\")\n\t\tcolor.White(L(\"%s Response\\n\"), name)\n\t\tcolor.White(\"--------------------------------------\\n\")\n\t\thelper.Dump(res)\n\t\tcolor.White(\"--------------------------------------\\n\")\n\t\tcolor.Green(L(\"✨DONE✨\\n\"))\n\t\treturn\n\t}\n\n\t// Silent mode output\n\tswitch res.(type) {\n\n\tcase int, int8, int16, int32, int64, uint, uint8, uint16, uint32, uint64, float32, float64, bool:\n\t\tfmt.Printf(\"%v\\n\", res)\n\t\treturn\n\n\tcase string, []byte:\n\t\tfmt.Printf(\"%s\\n\", res)\n\t\treturn\n\n\tdefault:\n\t\ttxt, err := jsoniter.Marshal(res)\n\t\tif err != nil {\n\t\t\tfmt.Printf(\"%s\\n\", err.Error())\n\t\t}\n\t\tfmt.Printf(\"%s\\n\", txt)\n\t}\n}\n\n// parseRunArgs parses the CLI arguments into process arguments, handling :: prefixed JSON.\nfunc parseRunArgs(args []string) []interface{} {\n\tpargs := []interface{}{}\n\tfor i, arg := range args {\n\t\tif i == 0 {\n\t\t\tcontinue\n\t\t}\n\n\t\tif strings.HasPrefix(arg, \"::\") {\n\t\t\traw := strings.TrimPrefix(arg, \"::\")\n\t\t\tvar v interface{}\n\t\t\terr := jsoniter.Unmarshal([]byte(raw), &v)\n\t\t\tif err != nil {\n\t\t\t\tcolor.Red(L(\"Arguments: %s\\n\"), err.Error())\n\t\t\t\treturn pargs\n\t\t\t}\n\t\t\tpargs = append(pargs, v)\n\t\t\tif !runSilent {\n\t\t\t\tcolor.White(\"args[%d]: %s\\n\", i-1, raw)\n\t\t\t}\n\t\t} else if strings.HasPrefix(arg, \"\\\\::\") {\n\t\t\tcleaned := \"::\" + strings.TrimPrefix(arg, \"\\\\::\")\n\t\t\tpargs = append(pargs, cleaned)\n\t\t\tif !runSilent {\n\t\t\t\tcolor.White(\"args[%d]: %s\\n\", i-1, cleaned)\n\t\t\t}\n\t\t} else {\n\t\t\tpargs = append(pargs, arg)\n\t\t\tif !runSilent {\n\t\t\t\tcolor.White(\"args[%d]: %s\\n\", i-1, arg)\n\t\t\t}\n\t\t}\n\t}\n\treturn pargs\n}\n\n// findAppRootFromPath finds the Yao application root directory by looking for app.yao\n// It traverses up from the given path until it finds app.yao or reaches the filesystem root\nfunc findAppRootFromPath(startPath string) (string, error) {\n\t// Get absolute path\n\tabsPath, err := filepath.Abs(startPath)\n\tif err != nil {\n\t\treturn \"\", fmt.Errorf(\"failed to get absolute path: %w\", err)\n\t}\n\n\t// If it's a file, start from its directory\n\tinfo, err := os.Stat(absPath)\n\tif err != nil {\n\t\treturn \"\", fmt.Errorf(\"path not found: %s\", absPath)\n\t}\n\n\tvar dir string\n\tif info.IsDir() {\n\t\tdir = absPath\n\t} else {\n\t\tdir = filepath.Dir(absPath)\n\t}\n\n\t// Traverse up to find app.yao\n\tfor {\n\t\t// Check for app.yao, app.json, or app.jsonc\n\t\tfor _, appFile := range []string{\"app.yao\", \"app.json\", \"app.jsonc\"} {\n\t\t\tappFilePath := filepath.Join(dir, appFile)\n\t\t\tif _, err := os.Stat(appFilePath); err == nil {\n\t\t\t\treturn dir, nil\n\t\t\t}\n\t\t}\n\n\t\t// Move to parent directory\n\t\tparent := filepath.Dir(dir)\n\t\tif parent == dir {\n\t\t\t// Reached root, no app.yao found\n\t\t\tbreak\n\t\t}\n\t\tdir = parent\n\t}\n\n\treturn \"\", fmt.Errorf(\"no app.yao found in path hierarchy of %s\", startPath)\n}\n" }, { "path": "cmd/socket.go", "content": "package cmd\n\nimport (\n\t\"fmt\"\n\t\"strings\"\n\n\t\"github.com/fatih/color\"\n\tjsoniter \"github.com/json-iterator/go\"\n\t\"github.com/spf13/cobra\"\n\t\"github.com/yaoapp/gou/plugin\"\n\t\"github.com/yaoapp/gou/socket\"\n\t\"github.com/yaoapp/kun/exception\"\n\t\"github.com/yaoapp/yao/config\"\n\t\"github.com/yaoapp/yao/engine\"\n\t\"github.com/yaoapp/yao/share\"\n)\n\nvar socketCmd = &cobra.Command{\n\tUse: \"socket\",\n\tShort: L(\"Open a socket connection\"),\n\tLong: L(\"Open a socket connection\"),\n\tRun: func(cmd *cobra.Command, args []string) {\n\t\tdefer share.SessionStop()\n\t\tdefer plugin.KillAll()\n\t\tdefer func() {\n\t\t\terr := exception.Catch(recover())\n\t\t\tif err != nil {\n\t\t\t\tfmt.Println(color.RedString(L(\"Fatal: %s\"), err.Error()))\n\t\t\t}\n\t\t}()\n\n\t\tBoot()\n\t\tcfg := config.Conf\n\t\tcfg.Session.IsCLI = true\n\t\tengine.Load(cfg, engine.LoadOption{Action: \"socket\"})\n\t\tif len(args) < 1 {\n\t\t\tfmt.Println(color.RedString(L(\"Not enough arguments\")))\n\t\t\tfmt.Println(color.WhiteString(share.BUILDNAME + \" help\"))\n\t\t\treturn\n\t\t}\n\n\t\tname := args[0]\n\t\tpargs := []interface{}{}\n\t\tfor i, arg := range args {\n\t\t\tif i == 0 {\n\t\t\t\tcontinue\n\t\t\t}\n\n\t\t\t// 解析参数\n\t\t\tif strings.HasPrefix(arg, \"::\") {\n\t\t\t\targ := strings.TrimPrefix(arg, \"::\")\n\t\t\t\tvar v interface{}\n\t\t\t\terr := jsoniter.Unmarshal([]byte(arg), &v)\n\t\t\t\tif err != nil {\n\t\t\t\t\tfmt.Println(color.RedString(L(\"Arguments: %s\"), err.Error()))\n\t\t\t\t\treturn\n\t\t\t\t}\n\t\t\t\tpargs = append(pargs, v)\n\t\t\t\tfmt.Println(color.WhiteString(\"args[%d]: %s\", i-1, arg))\n\t\t\t} else if strings.HasPrefix(arg, \"\\\\::\") {\n\t\t\t\targ := \"::\" + strings.TrimPrefix(arg, \"\\\\::\")\n\t\t\t\tpargs = append(pargs, arg)\n\t\t\t\tfmt.Println(color.WhiteString(\"args[%d]: %s\", i-1, arg))\n\t\t\t} else {\n\t\t\t\tpargs = append(pargs, arg)\n\t\t\t\tfmt.Println(color.WhiteString(\"args[%d]: %s\", i-1, arg))\n\t\t\t}\n\n\t\t}\n\n\t\tsocket, has := socket.Sockets[name]\n\t\tif !has {\n\t\t\tfmt.Println(color.RedString(L(\"%s not exists!\"), name))\n\t\t\treturn\n\t\t}\n\n\t\tif socket.Mode != \"client\" {\n\t\t\tfmt.Println(color.RedString(L(\"%s not supported yet!\"), socket.Mode))\n\t\t\treturn\n\t\t}\n\n\t\thost := socket.Host\n\t\tport := socket.Port\n\t\targsLen := len(pargs)\n\t\tif argsLen > 0 {\n\t\t\tif inputHost, ok := pargs[0].(string); ok {\n\t\t\t\thost = inputHost\n\t\t\t}\n\t\t}\n\n\t\tif argsLen > 1 {\n\t\t\tif inputPort, ok := pargs[1].(string); ok {\n\t\t\t\tport = inputPort\n\t\t\t}\n\t\t}\n\n\t\tfmt.Println(color.WhiteString(\"\\n---------------------------------\"))\n\t\tfmt.Println(color.WhiteString(socket.Name))\n\t\tfmt.Println(color.WhiteString(\"---------------------------------\"))\n\t\tfmt.Println(color.GreenString(\"Mode: %s\", socket.Mode))\n\t\tfmt.Println(color.GreenString(\"Host: %s://%s\", socket.Protocol, host))\n\t\tfmt.Println(color.GreenString(\"Port: %s\", port))\n\t\tfmt.Println(color.WhiteString(\"--------------------------------------\"))\n\t\terr := socket.Open(pargs...)\n\t\tif err != nil {\n\t\t\tfmt.Println(color.RedString(L(\"%s\"), err.Error()))\n\t\t\treturn\n\t\t}\n\n\t},\n}\n" }, { "path": "cmd/start.go", "content": "package cmd\n\nimport (\n\t\"fmt\"\n\t\"net\"\n\t\"os\"\n\t\"os/signal\"\n\t\"path/filepath\"\n\t\"strconv\"\n\t\"strings\"\n\t\"syscall\"\n\n\t\"github.com/fatih/color\"\n\t\"github.com/spf13/cobra\"\n\t\"github.com/yaoapp/gou/api\"\n\t\"github.com/yaoapp/gou/connector\"\n\t\"github.com/yaoapp/gou/fs\"\n\t\"github.com/yaoapp/gou/mcp\"\n\t\"github.com/yaoapp/gou/plugin\"\n\t\"github.com/yaoapp/gou/schedule\"\n\t\"github.com/yaoapp/gou/store\"\n\t\"github.com/yaoapp/gou/task\"\n\t\"github.com/yaoapp/gou/websocket\"\n\t\"github.com/yaoapp/kun/log\"\n\t\"github.com/yaoapp/yao/config\"\n\t\"github.com/yaoapp/yao/engine\"\n\tyaogrpc \"github.com/yaoapp/yao/grpc\"\n\t_ \"github.com/yaoapp/yao/grpc/auth\"\n\tsandboxhandler \"github.com/yaoapp/yao/grpc/sandbox\"\n\t\"github.com/yaoapp/yao/openapi\"\n\tsandbox \"github.com/yaoapp/yao/sandbox/v2\"\n\tischedule \"github.com/yaoapp/yao/schedule\"\n\t\"github.com/yaoapp/yao/service\"\n\t\"github.com/yaoapp/yao/setup\"\n\t\"github.com/yaoapp/yao/share\"\n\n\titask \"github.com/yaoapp/yao/task\"\n)\n\nvar startDebug = false\nvar startDisableWatching = false\n\nvar startCmd = &cobra.Command{\n\tUse: \"start\",\n\tShort: L(\"Start Engine\"),\n\tLong: L(\"Start Engine\"),\n\tRun: func(cmd *cobra.Command, args []string) {\n\n\t\tdefer share.SessionStop()\n\t\tdefer plugin.KillAll()\n\n\t\t// recive interrupt signal\n\t\tinterrupt := make(chan os.Signal, 1)\n\t\tsignal.Notify(interrupt, os.Interrupt, syscall.SIGTERM, syscall.SIGQUIT)\n\n\t\tBoot()\n\n\t\t// Setup\n\t\tisnew := false\n\n\t\t// Check if current directory is a Yao app root\n\t\tif !setup.IsYaoApp(config.Conf.Root) {\n\n\t\t\t// Check if we're inside a Yao app (subdirectory)\n\t\t\tif setup.InYaoApp(config.Conf.Root) {\n\t\t\t\tfmt.Println(color.RedString(L(\"Please run the command in the root directory of project\")))\n\t\t\t\tos.Exit(1)\n\t\t\t}\n\n\t\t\t// Not in a Yao app, check if empty to install\n\t\t\tif setup.IsEmptyDir(config.Conf.Root) {\n\t\t\t\t// Install the init app\n\t\t\t\tif err := install(); err != nil {\n\t\t\t\t\tfmt.Println(color.RedString(L(\"Install: %s\"), err.Error()))\n\t\t\t\t\tos.Exit(1)\n\t\t\t\t}\n\t\t\t\tisnew = true\n\t\t\t} else {\n\t\t\t\t// Directory not empty and no app.yao\n\t\t\t\tfmt.Println(color.RedString(\"The app.yao file is missing\"))\n\t\t\t\tos.Exit(1)\n\t\t\t}\n\t\t}\n\n\t\t// force debug\n\t\tif startDebug {\n\t\t\tconfig.Development()\n\t\t}\n\n\t\t// load the application engine\n\t\tloadWarnings, err := engine.Load(config.Conf, engine.LoadOption{\n\t\t\tAction: \"start\",\n\t\t})\n\t\tif err != nil {\n\t\t\tfmt.Println(color.RedString(L(\"Load: %s\"), err.Error()))\n\t\t\tos.Exit(1)\n\t\t}\n\n\t\tport := fmt.Sprintf(\":%d\", config.Conf.Port)\n\t\tif port == \":80\" {\n\t\t\tport = \"\"\n\t\t}\n\n\t\t// variables for the service\n\t\tfs, err := fs.Get(\"system\")\n\t\tif err != nil {\n\t\t\tfmt.Println(color.RedString(L(\"FileSystem: %s\"), err.Error()))\n\t\t\tos.Exit(1)\n\t\t}\n\n\t\tmode := config.Conf.Mode\n\t\thost := config.Conf.Host\n\t\tdataRoot := fs.Root()\n\t\truntimeMode := config.Conf.Runtime.Mode\n\n\t\tfmt.Println(color.WhiteString(\"\\n--------------------------------------------\"))\n\t\tfmt.Println(\n\t\t\tcolor.WhiteString(strings.TrimPrefix(share.App.Name, \"::\")),\n\t\t\tcolor.WhiteString(share.App.Version),\n\t\t\tmode,\n\t\t)\n\t\tfmt.Println(color.WhiteString(\"--------------------------------------------\"))\n\t\tif !share.BUILDIN {\n\t\t\troot, _ := filepath.Abs(config.Conf.Root)\n\t\t\tfmt.Println(color.WhiteString(L(\"Root\")), color.GreenString(\" %s\", root))\n\t\t}\n\n\t\tfmt.Println(color.WhiteString(L(\"Runtime\")), color.GreenString(\" %s\", runtimeMode))\n\t\tfmt.Println(color.WhiteString(L(\"Data\")), color.GreenString(\" %s\", dataRoot))\n\t\tfmt.Println(color.WhiteString(L(\"Listening\")), color.GreenString(\" %s:%d\", config.Conf.Host, config.Conf.Port))\n\n\t\t// print the messages under the development mode\n\t\tif mode == \"development\" {\n\t\t\tprintApis(false)\n\t\t\tprintTasks(false)\n\t\t\tprintSchedules(false)\n\t\t\tprintConnectors(false)\n\t\t\tprintStores(false)\n\t\t\tprintMCPs(false)\n\t\t}\n\n\t\troot, _ := adminRoot()\n\t\tendpoints := []setup.Endpoint{{URL: fmt.Sprintf(\"http://%s%s\", \"127.0.0.1\", port), Interface: \"localhost\"}}\n\t\tswitch host {\n\t\tcase \"0.0.0.0\":\n\t\t\tif values, err := setup.Endpoints(config.Conf); err == nil {\n\t\t\t\tendpoints = append(endpoints, values...)\n\t\t\t}\n\t\tcase \"127.0.0.1\":\n\t\t\t// Localhost only\n\t\tdefault:\n\t\t\tmatched := false\n\t\t\tendpoints = []setup.Endpoint{}\n\t\t\tif values, err := setup.Endpoints(config.Conf); err == nil {\n\t\t\t\tfor _, value := range values {\n\t\t\t\t\tif strings.HasPrefix(value.URL, fmt.Sprintf(\"http://%s:\", host)) {\n\t\t\t\t\t\tendpoints = append(endpoints, value)\n\t\t\t\t\t\tmatched = true\n\t\t\t\t\t}\n\t\t\t\t}\n\t\t\t}\n\t\t\tif !matched {\n\t\t\t\tfmt.Println(color.RedString(L(\"Host %s not found\"), host))\n\t\t\t\tos.Exit(1)\n\t\t\t}\n\t\t}\n\n\t\t// Print welcome message for the new application\n\t\tif isnew {\n\t\t\tprintWelcome()\n\t\t}\n\n\t\t// Start Tasks\n\t\titask.Start()\n\t\tdefer itask.Stop()\n\n\t\t// Start Schedules\n\t\tischedule.Start()\n\t\tdefer ischedule.Stop()\n\n\t\t// Pre-flight: detect port conflicts before attempting to start servers.\n\t\tif occupied, proc := portOccupied(config.Conf.Host, config.Conf.Port); occupied {\n\t\t\tfmt.Println(color.RedString(L(\"Fatal: HTTP port %d is already in use%s\"), config.Conf.Port, proc))\n\t\t\treturn\n\t\t}\n\t\tif strings.ToLower(config.Conf.GRPC.Enabled) != \"off\" {\n\t\t\tfor _, h := range yaogrpc.ExpandHosts(config.Conf.GRPC.Host) {\n\t\t\t\tif occupied, proc := portOccupied(h, config.Conf.GRPC.Port); occupied {\n\t\t\t\t\tfmt.Println(color.RedString(L(\"Fatal: gRPC port %d is already in use%s\"), config.Conf.GRPC.Port, proc))\n\t\t\t\t\treturn\n\t\t\t\t}\n\t\t\t}\n\t\t}\n\n\t\t// Wire gRPC heartbeat → sandbox Manager so container liveness is tracked.\n\t\tyaogrpc.SetSandboxOnBeat(func(data *sandboxhandler.HeartbeatData) string {\n\t\t\tsandbox.M().Heartbeat(data.SandboxID, true, int(data.RunningProcs))\n\t\t\treturn \"ok\"\n\t\t})\n\n\t\t// Start all servers (gRPC + HTTP) as a single unit.\n\t\t// Start() blocks until HTTP port is bound (READY) or returns error.\n\t\tsvc, err := service.Start(config.Conf, service.ServerHooks{\n\t\t\tStart: yaogrpc.StartServer,\n\t\t\tStop: yaogrpc.Stop,\n\t\t\tAddrs: yaogrpc.Addr,\n\t\t})\n\t\tif err != nil {\n\t\t\tfmt.Println(color.RedString(L(\"Fatal: %s\"), err.Error()))\n\t\t\treturn\n\t\t}\n\n\t\t// Access Points (printed after servers are up so addresses are known)\n\t\tfmt.Println(color.WhiteString(\"\\n---------------------------------\"))\n\t\tfmt.Println(color.WhiteString(L(\"Access Points\")))\n\t\tfmt.Println(color.WhiteString(\"---------------------------------\"))\n\n\t\tif grpcAddrs := svc.HookAddrs(); len(grpcAddrs) > 0 {\n\t\t\tfmt.Println(color.CyanString(\"\\ngRPC\"))\n\t\t\tfmt.Println(color.WhiteString(\"--------------------------\"))\n\t\t\tfor _, addr := range grpcAddrs {\n\t\t\t\tfmt.Println(color.WhiteString(L(\"Server\")), color.GreenString(\" %s\", addr))\n\t\t\t}\n\t\t}\n\n\t\tapiRoot := \"/api\"\n\t\tif openapi.Server != nil {\n\t\t\tapiRoot = openapi.Server.Config.BaseURL\n\t\t}\n\t\tfor _, endpoint := range endpoints {\n\t\t\tfmt.Println(color.CyanString(\"\\n%s\", endpoint.Interface))\n\t\t\tfmt.Println(color.WhiteString(\"--------------------------\"))\n\t\t\tfmt.Println(color.WhiteString(L(\"Website\")), color.GreenString(\" %s\", endpoint.URL))\n\t\t\tfmt.Println(color.WhiteString(L(\"Dashboard\")), color.GreenString(\" %s/%s/auth/entry\", endpoint.URL, strings.Trim(root, \"/\")))\n\t\t\tif openapi.Server != nil {\n\t\t\t\tfmt.Println(color.WhiteString(L(\"OpenAPI\")), color.GreenString(\" %s%s\", endpoint.URL, apiRoot))\n\t\t\t} else {\n\t\t\t\tfmt.Println(color.WhiteString(L(\"API\")), color.GreenString(\" %s%s\", endpoint.URL, apiRoot))\n\t\t\t}\n\t\t}\n\t\tfmt.Println(\"\")\n\n\t\t// Start watching\n\t\twatchDone := make(chan uint8, 1)\n\t\tif mode == \"development\" && !startDisableWatching {\n\t\t\tgo svc.Watch(watchDone)\n\t\t}\n\n\t\t// Print the messages under the production mode\n\t\tif mode == \"production\" {\n\t\t\tprintApis(true)\n\t\t\tprintTasks(true)\n\t\t\tprintSchedules(true)\n\t\t\tprintConnectors(true)\n\t\t\tprintStores(true)\n\t\t\tprintMCPs(true)\n\t\t}\n\n\t\t// Print the warnings\n\t\tif len(loadWarnings) > 0 {\n\t\t\tfmt.Println(color.YellowString(\"---------------------------------\"))\n\t\t\tfmt.Println(color.YellowString(L(\"Warnings\")))\n\t\t\tfmt.Println(color.YellowString(\"---------------------------------\"))\n\t\t\tfor _, warning := range loadWarnings {\n\t\t\t\tfmt.Println(color.YellowString(\"[%s] %s\", warning.Widget, warning.Error))\n\t\t\t}\n\t\t\tfmt.Printf(\"\\n\")\n\t\t}\n\n\t\tfmt.Println(color.GreenString(L(\"Server is up and running...\")))\n\t\tfmt.Println(color.GreenString(\"Ctrl+C to stop\"))\n\n\t\tfor {\n\t\t\tselect {\n\t\t\tcase <-interrupt:\n\t\t\t\tfmt.Println(color.WhiteString(\"\\nShutting down...\"))\n\t\t\t\tsvc.Stop()\n\t\t\t\tfmt.Println(color.GreenString(L(\"✨Exited successfully!\")))\n\t\t\t\twatchDone <- 1\n\t\t\t\treturn\n\t\t\t}\n\t\t}\n\t},\n}\n\nfunc install() error {\n\t// Copy the app source files from the binary\n\terr := setup.Install(config.Conf.Root)\n\tif err != nil {\n\t\treturn err\n\t}\n\n\t// Reload the application engine\n\tBoot()\n\n\t// load the application engine\n\tloadWarnings, err := engine.Load(config.Conf, engine.LoadOption{Action: \"start\"})\n\tif err != nil {\n\t\treturn err\n\t}\n\n\t// Print the warnings\n\tif len(loadWarnings) > 0 {\n\t\tfor _, warning := range loadWarnings {\n\t\t\tfmt.Println(color.YellowString(\"[%s] %s\", warning.Widget, warning.Error))\n\t\t}\n\t\tfmt.Printf(\"\\n\\n\")\n\t}\n\n\terr = setup.Initialize(config.Conf.Root, config.Conf)\n\tif err != nil {\n\t\treturn err\n\t}\n\n\treturn nil\n}\n\nfunc adminRoot() (string, int) {\n\tadminRoot := \"/yao/\"\n\tif share.App.AdminRoot != \"\" {\n\t\troot := strings.TrimPrefix(share.App.AdminRoot, \"/\")\n\t\troot = strings.TrimSuffix(root, \"/\")\n\t\tadminRoot = fmt.Sprintf(\"/%s/\", root)\n\t}\n\tadminRootLen := len(adminRoot)\n\treturn adminRoot, adminRootLen\n}\n\nfunc printWelcome() {\n\tfmt.Println(color.CyanString(\"\\n---------------------------------\"))\n\tfmt.Println(color.CyanString(L(\"🎉 Welcome to Yao 🎉 \")))\n\tfmt.Println(color.CyanString(\"---------------------------------\"))\n\tfmt.Println(color.WhiteString(\"📚 Documentation: \"), color.CyanString(\"https://yaoapps.com/docs\"))\n\tfmt.Println(color.WhiteString(\"🏡 Join Yao Community: \"), color.CyanString(\"https://yaoapps.com/community\"))\n\tfmt.Println(color.WhiteString(\"🤖 Build Your Digital Workforce:\"), color.CyanString(\"https://yaoagents.com\"))\n\tfmt.Println(\"\")\n}\n\nfunc printConnectors(silent bool) {\n\n\tif len(connector.Connectors) == 0 {\n\t\treturn\n\t}\n\n\tif silent {\n\t\tfor name := range connector.Connectors {\n\t\t\tlog.Info(\"[Connector] %s loaded\", name)\n\t\t}\n\t\treturn\n\t}\n\n\tfmt.Println(color.WhiteString(\"\\n---------------------------------\"))\n\tfmt.Println(color.WhiteString(L(\"Connectors List (%d)\"), len(connector.Connectors)))\n\tfmt.Println(color.WhiteString(\"---------------------------------\"))\n\tfor name := range connector.Connectors {\n\t\tfmt.Print(color.CyanString(\"[Connector]\"))\n\t\tfmt.Print(color.WhiteString(\" %s\\t loaded\\n\", name))\n\t}\n}\n\nfunc printStores(silent bool) {\n\tif len(store.Pools) == 0 {\n\t\treturn\n\t}\n\n\tif silent {\n\t\tfor name := range store.Pools {\n\t\t\tlog.Info(\"[Store] %s loaded\", name)\n\t\t}\n\t\treturn\n\t}\n\n\tfmt.Println(color.WhiteString(\"\\n---------------------------------\"))\n\tfmt.Println(color.WhiteString(L(\"Stores List (%d)\"), len(store.Pools)))\n\tfmt.Println(color.WhiteString(\"---------------------------------\"))\n\tfor name := range store.Pools {\n\t\tfmt.Print(color.CyanString(\"[Store]\"))\n\t\tfmt.Print(color.WhiteString(\" %s\\t loaded\\n\", name))\n\t}\n}\n\nfunc printSchedules(silent bool) {\n\n\tif len(schedule.Schedules) == 0 {\n\t\treturn\n\t}\n\n\tif silent {\n\t\tfor name, sch := range schedule.Schedules {\n\t\t\tprocess := fmt.Sprintf(\"Process: %s\", sch.Process)\n\t\t\tif sch.TaskName != \"\" {\n\t\t\t\tprocess = fmt.Sprintf(\"Task: %s\", sch.TaskName)\n\t\t\t}\n\t\t\tlog.Info(\"[Schedule] %s %s %s %s\", sch.Schedule, name, sch.Name, process)\n\t\t}\n\t\treturn\n\t}\n\n\tfmt.Println(color.WhiteString(\"\\n---------------------------------\"))\n\tfmt.Println(color.WhiteString(L(\"Schedules List (%d)\"), len(schedule.Schedules)))\n\tfmt.Println(color.WhiteString(\"---------------------------------\"))\n\tfor name, sch := range schedule.Schedules {\n\t\tprocess := fmt.Sprintf(\"Process: %s\", sch.Process)\n\t\tif sch.TaskName != \"\" {\n\t\t\tprocess = fmt.Sprintf(\"Task: %s\", sch.TaskName)\n\t\t}\n\t\tfmt.Print(color.CyanString(\"[Schedule] %s %s\", sch.Schedule, name))\n\t\tfmt.Print(color.WhiteString(\"\\t%s\\t%s\\n\", sch.Name, process))\n\t}\n}\n\nfunc printTasks(silent bool) {\n\n\tif len(task.Tasks) == 0 {\n\t\treturn\n\t}\n\n\tif silent {\n\t\tfor _, t := range task.Tasks {\n\t\t\tlog.Info(\"[Task] %s workers:%d\", t.Option.Name, t.Option.WorkerNums)\n\t\t}\n\t\treturn\n\t}\n\n\tfmt.Println(color.WhiteString(\"\\n---------------------------------\"))\n\tfmt.Println(color.WhiteString(L(\"Tasks List (%d)\"), len(task.Tasks)))\n\tfmt.Println(color.WhiteString(\"---------------------------------\"))\n\tfor _, t := range task.Tasks {\n\t\tfmt.Print(color.CyanString(\"[Task] %s\", t.Option.Name))\n\t\tfmt.Print(color.WhiteString(\"\\t workers: %d\\n\", t.Option.WorkerNums))\n\t}\n}\n\nfunc printApis(silent bool) {\n\t// Determine API root based on OpenAPI mode\n\tapiRoot := \"/api\"\n\tif openapi.Server != nil {\n\t\tapiRoot = openapi.Server.Config.BaseURL\n\t}\n\n\tif silent {\n\t\tfor _, api := range api.APIs {\n\t\t\tif len(api.HTTP.Paths) <= 0 {\n\t\t\t\tcontinue\n\t\t\t}\n\t\t\tlog.Info(\"[API] %s(%d)\", api.ID, len(api.HTTP.Paths))\n\t\t\tfor _, p := range api.HTTP.Paths {\n\t\t\t\tlog.Info(\"%s %s %s\", p.Method, filepath.Join(apiRoot, api.HTTP.Group, p.Path), p.Process)\n\t\t\t}\n\t\t}\n\t\tfor name, upgrader := range websocket.Upgraders { // WebSocket\n\t\t\tlog.Info(\"[WebSocket] GET /websocket/%s process:%s\", name, upgrader.Process)\n\t\t}\n\t\treturn\n\t}\n\n\t// Skip detailed API list when OpenAPI is enabled\n\tif openapi.Server != nil {\n\t\treturn\n\t}\n\n\tfmt.Println(color.WhiteString(\"\\n---------------------------------\"))\n\tfmt.Println(color.WhiteString(L(\"APIs List\")))\n\tfmt.Println(color.WhiteString(\"---------------------------------\"))\n\n\tfor _, api := range api.APIs { // API info\n\t\tif len(api.HTTP.Paths) <= 0 {\n\t\t\tcontinue\n\t\t}\n\n\t\tdeprecated := \"\"\n\t\tif strings.HasPrefix(api.ID, \"xiang.\") {\n\t\t\tdeprecated = \" WILL BE DEPRECATED\"\n\t\t}\n\n\t\tfmt.Printf(\"%s%s\\n\", color.CyanString(\"\\n%s(%d)\", api.ID, len(api.HTTP.Paths)), color.RedString(deprecated))\n\t\tfor _, p := range api.HTTP.Paths {\n\t\t\tfmt.Println(\n\t\t\t\tcolorMehtod(p.Method),\n\t\t\t\tcolor.WhiteString(filepath.Join(apiRoot, api.HTTP.Group, p.Path)),\n\t\t\t\t\"\\tprocess:\", p.Process)\n\t\t}\n\t}\n\n\tif len(websocket.Upgraders) > 0 {\n\t\tfmt.Print(color.CyanString(fmt.Sprintf(\"\\n%s(%d)\\n\", \"WebSocket\", len(websocket.Upgraders))))\n\t\tfor name, upgrader := range websocket.Upgraders { // WebSocket\n\t\t\tfmt.Println(\n\t\t\t\tcolorMehtod(\"GET\"),\n\t\t\t\tcolor.WhiteString(filepath.Join(\"/websocket\", name)),\n\t\t\t\t\"\\tprocess:\", upgrader.Process)\n\t\t}\n\t}\n}\n\nfunc printMCPs(silent bool) {\n\tclients := mcp.ListClients()\n\tif len(clients) == 0 {\n\t\treturn\n\t}\n\n\tif silent {\n\t\tfor _, clientID := range clients {\n\t\t\tlog.Info(\"[MCP] %s loaded\", clientID)\n\t\t}\n\t\treturn\n\t}\n\n\t// Separate agent MCPs from standard MCPs by Type field\n\tagentClients := []string{}\n\tstandardClients := []string{}\n\tfor _, clientID := range clients {\n\t\tclient, err := mcp.Select(clientID)\n\t\tif err != nil {\n\t\t\tstandardClients = append(standardClients, clientID)\n\t\t\tcontinue\n\t\t}\n\n\t\tinfo := client.Info()\n\t\tif info != nil && info.Type == \"agent\" {\n\t\t\tagentClients = append(agentClients, clientID)\n\t\t} else {\n\t\t\tstandardClients = append(standardClients, clientID)\n\t\t}\n\t}\n\n\tfmt.Println(color.WhiteString(\"\\n---------------------------------\"))\n\tfmt.Println(color.WhiteString(L(\"MCP Clients List (%d)\"), len(clients)))\n\tfmt.Println(color.WhiteString(\"---------------------------------\"))\n\n\tif len(standardClients) > 0 {\n\t\tfmt.Println(color.WhiteString(\"\\n%s (%d)\", \"Standard MCPs\", len(standardClients)))\n\t\tfmt.Println(color.WhiteString(\"--------------------------\"))\n\t\tfor _, clientID := range standardClients {\n\t\t\tclient, err := mcp.Select(clientID)\n\t\t\tif err != nil {\n\t\t\t\tfmt.Print(color.CyanString(\"[MCP] %s\", clientID))\n\t\t\t\tfmt.Print(color.WhiteString(\"\\tloaded\\n\"))\n\t\t\t\tcontinue\n\t\t\t}\n\n\t\t\tinfo := client.Info()\n\t\t\ttransport := \"unknown\"\n\t\t\tlabel := clientID\n\t\t\tif info != nil {\n\t\t\t\tif info.Transport != \"\" {\n\t\t\t\t\ttransport = string(info.Transport)\n\t\t\t\t}\n\t\t\t\tif info.Label != \"\" {\n\t\t\t\t\tlabel = info.Label\n\t\t\t\t}\n\t\t\t}\n\n\t\t\tfmt.Print(color.CyanString(\"[MCP] %s\", label))\n\t\t\tfmt.Print(color.WhiteString(\"\\t%s\\tid: %s\", transport, clientID))\n\n\t\t\t// Only show tools count for process transport\n\t\t\tif transport == \"process\" {\n\t\t\t\ttoolsCount := 0\n\t\t\t\tmapping, err := mcp.GetClientMapping(clientID)\n\t\t\t\tif err == nil && mapping.Tools != nil {\n\t\t\t\t\ttoolsCount = len(mapping.Tools)\n\t\t\t\t}\n\t\t\t\tfmt.Print(color.WhiteString(\"\\ttools: %d\", toolsCount))\n\t\t\t}\n\t\t\tfmt.Print(\"\\n\")\n\t\t}\n\t}\n\n\tif len(agentClients) > 0 {\n\t\tfmt.Println(color.WhiteString(\"\\n%s (%d)\", \"Agent MCPs\", len(agentClients)))\n\t\tfmt.Println(color.WhiteString(\"--------------------------\"))\n\t\tfor _, clientID := range agentClients {\n\t\t\tclient, err := mcp.Select(clientID)\n\t\t\tif err != nil {\n\t\t\t\tfmt.Print(color.CyanString(\"[MCP] %s\", clientID))\n\t\t\t\tfmt.Print(color.WhiteString(\"\\tloaded\\n\"))\n\t\t\t\tcontinue\n\t\t\t}\n\n\t\t\tinfo := client.Info()\n\t\t\ttransport := \"unknown\"\n\t\t\tlabel := clientID\n\t\t\tif info != nil {\n\t\t\t\tif info.Transport != \"\" {\n\t\t\t\t\ttransport = string(info.Transport)\n\t\t\t\t}\n\t\t\t\tif info.Label != \"\" {\n\t\t\t\t\tlabel = info.Label\n\t\t\t\t}\n\t\t\t}\n\n\t\t\tfmt.Print(color.CyanString(\"[MCP] %s\", label))\n\t\t\tfmt.Print(color.WhiteString(\"\\t%s\\tid: %s\", transport, clientID))\n\n\t\t\t// Only show tools count for process transport\n\t\t\tif transport == \"process\" {\n\t\t\t\ttoolsCount := 0\n\t\t\t\tmapping, err := mcp.GetClientMapping(clientID)\n\t\t\t\tif err == nil && mapping.Tools != nil {\n\t\t\t\t\ttoolsCount = len(mapping.Tools)\n\t\t\t\t}\n\t\t\t\tfmt.Print(color.WhiteString(\"\\ttools: %d\", toolsCount))\n\t\t\t}\n\t\t\tfmt.Print(\"\\n\")\n\t\t}\n\t}\n}\n\nfunc colorMehtod(method string) string {\n\tmethod = strings.ToUpper(method)\n\tswitch method {\n\tcase \"GET\":\n\t\treturn color.GreenString(\"GET\")\n\tcase \"POST\":\n\t\treturn color.YellowString(\"POST\")\n\tdefault:\n\t\treturn color.WhiteString(method)\n\t}\n}\n\n// portOccupied probes whether host:port is already bound.\n// Returns (true, \" (pid XXXX)\") when occupied, (false, \"\") otherwise.\nfunc portOccupied(host string, port int) (bool, string) {\n\taddr := net.JoinHostPort(host, strconv.Itoa(port))\n\tln, err := net.Listen(\"tcp\", addr)\n\tif err != nil {\n\t\treturn true, fmt.Sprintf(\" (%s)\", err.Error())\n\t}\n\tln.Close()\n\treturn false, \"\"\n}\n\nfunc init() {\n\tstartCmd.PersistentFlags().BoolVarP(&startDebug, \"debug\", \"\", false, L(\"Development mode\"))\n\tstartCmd.PersistentFlags().BoolVarP(&startDisableWatching, \"disable-watching\", \"\", false, L(\"Disable watching\"))\n}\n" }, { "path": "cmd/sui/build.go", "content": "package sui\n\nimport (\n\t\"fmt\"\n\t\"os\"\n\t\"path/filepath\"\n\t\"strings\"\n\t\"time\"\n\n\t\"github.com/fatih/color\"\n\t\"github.com/google/uuid\"\n\tjsoniter \"github.com/json-iterator/go\"\n\t\"github.com/spf13/cobra\"\n\t\"github.com/yaoapp/gou/session\"\n\t\"github.com/yaoapp/yao/config\"\n\t\"github.com/yaoapp/yao/engine\"\n\t\"github.com/yaoapp/yao/sui/core\"\n)\n\n// BuildCmd command\nvar BuildCmd = &cobra.Command{\n\tUse: \"build\",\n\tShort: L(\"Build the template\"),\n\tLong: L(\"Build the template\"),\n\tRun: func(cmd *cobra.Command, args []string) {\n\t\tif len(args) < 1 {\n\t\t\tfmt.Fprintln(os.Stderr, color.RedString(L(\"yao sui build [template] [data]\")))\n\t\t\treturn\n\t\t}\n\n\t\tBoot()\n\n\t\tcfg := config.Conf\n\t\tloadWarnings, err := engine.Load(cfg, engine.LoadOption{Action: \"sui.build\"})\n\t\tif err != nil {\n\t\t\tfmt.Fprintln(os.Stderr, color.RedString(err.Error()))\n\t\t\treturn\n\t\t}\n\n\t\tid := args[0]\n\t\ttemplate := \"default\"\n\t\tif len(args) >= 2 {\n\t\t\ttemplate = args[1]\n\t\t}\n\n\t\t// For agent SUI, use \"agent\" as default template\n\t\tif id == \"agent\" && template == \"default\" {\n\t\t\ttemplate = \"agent\"\n\t\t}\n\n\t\tvar sessionData map[string]interface{}\n\t\terr = jsoniter.UnmarshalFromString(strings.TrimPrefix(data, \"::\"), &sessionData)\n\t\tif err != nil {\n\t\t\tfmt.Fprintln(os.Stderr, color.RedString(err.Error()))\n\t\t\treturn\n\t\t}\n\n\t\tsid := uuid.New().String()\n\t\tif sessionData != nil && len(sessionData) > 0 {\n\t\t\tsession.Global().ID(sid).SetMany(sessionData)\n\t\t}\n\n\t\tsui, has := core.SUIs[id]\n\t\tif !has {\n\t\t\tfmt.Fprint(os.Stderr, color.RedString(\"the sui \"+id+\" does not exist\"))\n\t\t\treturn\n\t\t}\n\t\tsui.WithSid(sid)\n\n\t\ttmpl, err := sui.GetTemplate(template)\n\t\tif err != nil {\n\t\t\tfmt.Fprintln(os.Stderr, color.RedString(err.Error()))\n\t\t\treturn\n\t\t}\n\n\t\t// -\n\t\tpublicRoot, err := sui.PublicRootWithSid(sid)\n\t\tassetRoot := filepath.Join(publicRoot, \"assets\")\n\t\tif err != nil {\n\t\t\tfmt.Fprintln(os.Stderr, color.RedString(err.Error()))\n\t\t\treturn\n\t\t}\n\n\t\tfmt.Println(color.WhiteString(\"-----------------------\"))\n\t\tfmt.Println(color.WhiteString(\"Public Root: /public%s\", publicRoot))\n\t\tfmt.Println(color.WhiteString(\" Template: %s\", tmpl.GetRoot()))\n\t\tfmt.Println(color.WhiteString(\" Session: %s\", strings.TrimLeft(data, \"::\")))\n\t\tfmt.Println(color.WhiteString(\"-----------------------\"))\n\n\t\t// Timecost\n\t\tstart := time.Now()\n\t\tminify := true\n\t\tmode := \"production\"\n\t\tif debug {\n\t\t\tminify = false\n\t\t\tmode = \"development\"\n\t\t}\n\n\t\twarnings, err := tmpl.Build(&core.BuildOption{SSR: true, AssetRoot: assetRoot, ExecScripts: true, ScriptMinify: minify, StyleMinify: minify})\n\t\tif err != nil {\n\t\t\tfmt.Fprintln(os.Stderr, color.RedString(err.Error()))\n\t\t\treturn\n\t\t}\n\t\tend := time.Now()\n\t\ttimecost := end.Sub(start).Truncate(time.Millisecond)\n\t\tif debug {\n\t\t\tfmt.Println(color.YellowString(\"Build succeeded for %s in %s\", mode, timecost))\n\t\t\treturn\n\t\t}\n\n\t\tif len(loadWarnings) > 0 {\n\t\t\tfor _, warning := range loadWarnings {\n\t\t\t\tfmt.Println(color.YellowString(\"[%s] %s\", warning.Widget, warning.Error))\n\t\t\t}\n\t\t}\n\n\t\tif len(warnings) > 0 {\n\t\t\tfor _, warning := range warnings {\n\t\t\t\tfmt.Println(color.YellowString(\"Warning: %s\", warning))\n\t\t\t}\n\t\t}\n\n\t\tfmt.Println(color.GreenString(\"Build succeeded for %s in %s\", mode, timecost))\n\t},\n}\n" }, { "path": "cmd/sui/sui.go", "content": "package sui\n\nvar data string\nvar locales string\nvar debug bool\n\nfunc init() {\n\tWatchCmd.PersistentFlags().StringVarP(&data, \"data\", \"d\", \"::{}\", L(\"Session Data\"))\n\tBuildCmd.PersistentFlags().StringVarP(&data, \"data\", \"d\", \"::{}\", L(\"Session Data\"))\n\tBuildCmd.PersistentFlags().BoolVarP(&debug, \"debug\", \"D\", false, L(\"Debug mode\"))\n\tTransCmd.PersistentFlags().StringVarP(&data, \"data\", \"d\", \"::{}\", L(\"Session Data\"))\n\tTransCmd.PersistentFlags().BoolVarP(&debug, \"debug\", \"D\", false, L(\"Debug mode\"))\n\tTransCmd.PersistentFlags().StringVarP(&locales, \"locales\", \"l\", \"\", L(\"Locales, separated by commas\"))\n}\n" }, { "path": "cmd/sui/trans.go", "content": "package sui\n\nimport (\n\t\"fmt\"\n\t\"os\"\n\t\"path/filepath\"\n\t\"strings\"\n\t\"time\"\n\n\t\"github.com/fatih/color\"\n\t\"github.com/google/uuid\"\n\tjsoniter \"github.com/json-iterator/go\"\n\t\"github.com/spf13/cobra\"\n\t\"github.com/yaoapp/gou/session\"\n\t\"github.com/yaoapp/yao/config\"\n\t\"github.com/yaoapp/yao/engine\"\n\t\"github.com/yaoapp/yao/sui/core\"\n\t\"golang.org/x/text/language\"\n)\n\n// TransCmd command\nvar TransCmd = &cobra.Command{\n\tUse: \"trans\",\n\tShort: L(\"Translate the template\"),\n\tLong: L(\"Translate the template\"),\n\tRun: func(cmd *cobra.Command, args []string) {\n\t\tif len(args) < 2 {\n\t\t\tfmt.Fprintln(os.Stderr, color.RedString(L(\"yao sui trans