address thread safety comment

- Resolve merge conflicts from merged SDK changes (PR #12103)
- Move sdk/ files from api/features/chat/sdk/ to copilot/sdk/
- Fix all imports to use backend.copilot.* paths
- Move new tools (bash_exec, sandbox, web_fetch, feature_requests,
  check_operation_status) to copilot/tools/ with updated imports
- Add append_and_save_message to model.py (adapted to chat_db() pattern)
- Wire SDK service into copilot executor processor with feature flag
- Add track_user_message to routes.py stream handler

.dockerignore

@@ -5,42 +5,13 @@
 !docs/
 
 # Platform - Libs
-!autogpt_platform/autogpt_libs/autogpt_libs/
-!autogpt_platform/autogpt_libs/pyproject.toml
-!autogpt_platform/autogpt_libs/poetry.lock
-!autogpt_platform/autogpt_libs/README.md
+!autogpt_platform/autogpt_libs/
 
 # Platform - Backend
-!autogpt_platform/backend/backend/
-!autogpt_platform/backend/test/e2e_test_data.py
-!autogpt_platform/backend/migrations/
-!autogpt_platform/backend/schema.prisma
-!autogpt_platform/backend/pyproject.toml
-!autogpt_platform/backend/poetry.lock
-!autogpt_platform/backend/README.md
-!autogpt_platform/backend/.env
-!autogpt_platform/backend/gen_prisma_types_stub.py
-
-# Platform - Market
-!autogpt_platform/market/market/
-!autogpt_platform/market/scripts.py
-!autogpt_platform/market/schema.prisma
-!autogpt_platform/market/pyproject.toml
-!autogpt_platform/market/poetry.lock
-!autogpt_platform/market/README.md
+!autogpt_platform/backend/
 
 # Platform - Frontend
-!autogpt_platform/frontend/src/
-!autogpt_platform/frontend/public/
-!autogpt_platform/frontend/scripts/
-!autogpt_platform/frontend/package.json
-!autogpt_platform/frontend/pnpm-lock.yaml
-!autogpt_platform/frontend/tsconfig.json
-!autogpt_platform/frontend/README.md
-## config
-!autogpt_platform/frontend/*.config.*
-!autogpt_platform/frontend/.env.*
-!autogpt_platform/frontend/.env
+!autogpt_platform/frontend/
 
 # Classic - AutoGPT
 !classic/original_autogpt/autogpt/
@@ -64,6 +35,38 @@
 # Classic - Frontend
 !classic/frontend/build/web/
 
-# Explicitly re-ignore some folders
-.*
-**/__pycache__
+# Explicitly re-ignore unwanted files from whitelisted directories
+# Note: These patterns MUST come after the whitelist rules to take effect
+
+# Hidden files and directories (but keep frontend .env files needed for build)
+**/.*
+!autogpt_platform/frontend/.env
+!autogpt_platform/frontend/.env.default
+!autogpt_platform/frontend/.env.production
+
+# Python artifacts
+**/__pycache__/
+**/*.pyc
+**/*.pyo
+**/.venv/
+**/.ruff_cache/
+**/.pytest_cache/
+**/.coverage
+**/htmlcov/
+
+# Node artifacts
+**/node_modules/
+**/.next/
+**/storybook-static/
+**/playwright-report/
+**/test-results/
+
+# Build artifacts
+**/dist/
+**/build/
+!autogpt_platform/frontend/src/**/build/
+**/target/
+
+# Logs and temp files
+**/*.log
+**/*.tmp

.github/workflows/claude-ci-failure-auto-fix.yml

@@ -40,6 +40,48 @@ jobs:
           git checkout -b "$BRANCH_NAME"
           echo "branch_name=$BRANCH_NAME" >> $GITHUB_OUTPUT
 
+      # Backend Python/Poetry setup (so Claude can run linting/tests)
+      - name: Set up Python
+        uses: actions/setup-python@v5
+        with:
+          python-version: "3.11"
+
+      - name: Set up Python dependency cache
+        uses: actions/cache@v5
+        with:
+          path: ~/.cache/pypoetry
+          key: poetry-${{ runner.os }}-${{ hashFiles('autogpt_platform/backend/poetry.lock') }}
+
+      - name: Install Poetry
+        run: |
+          cd autogpt_platform/backend
+          HEAD_POETRY_VERSION=$(python3 ../../.github/workflows/scripts/get_package_version_from_lockfile.py poetry)
+          curl -sSL https://install.python-poetry.org | POETRY_VERSION=$HEAD_POETRY_VERSION python3 -
+          echo "$HOME/.local/bin" >> $GITHUB_PATH
+
+      - name: Install Python dependencies
+        working-directory: autogpt_platform/backend
+        run: poetry install
+
+      - name: Generate Prisma Client
+        working-directory: autogpt_platform/backend
+        run: poetry run prisma generate && poetry run gen-prisma-stub
+
+      # Frontend Node.js/pnpm setup (so Claude can run linting/tests)
+      - name: Enable corepack
+        run: corepack enable
+
+      - name: Set up Node.js
+        uses: actions/setup-node@v6
+        with:
+          node-version: "22"
+          cache: "pnpm"
+          cache-dependency-path: autogpt_platform/frontend/pnpm-lock.yaml
+
+      - name: Install JavaScript dependencies
+        working-directory: autogpt_platform/frontend
+        run: pnpm install --frozen-lockfile
+
       - name: Get CI failure details
         id: failure_details
         uses: actions/github-script@v8

.github/workflows/claude-dependabot.yml

@@ -77,27 +77,15 @@ jobs:
         run: poetry run prisma generate && poetry run gen-prisma-stub
 
       # Frontend Node.js/pnpm setup (mirrors platform-frontend-ci.yml)
+      - name: Enable corepack
+        run: corepack enable
+
       - name: Set up Node.js
         uses: actions/setup-node@v6
         with:
           node-version: "22"
-
-      - name: Enable corepack
-        run: corepack enable
-
-      - name: Set pnpm store directory
-        run: |
-          pnpm config set store-dir ~/.pnpm-store
-          echo "PNPM_HOME=$HOME/.pnpm-store" >> $GITHUB_ENV
-
-      - name: Cache frontend dependencies
-        uses: actions/cache@v5
-        with:
-          path: ~/.pnpm-store
-          key: ${{ runner.os }}-pnpm-${{ hashFiles('autogpt_platform/frontend/pnpm-lock.yaml', 'autogpt_platform/frontend/package.json') }}
-          restore-keys: |
-            ${{ runner.os }}-pnpm-${{ hashFiles('autogpt_platform/frontend/pnpm-lock.yaml') }}
-            ${{ runner.os }}-pnpm-
+          cache: "pnpm"
+          cache-dependency-path: autogpt_platform/frontend/pnpm-lock.yaml
 
       - name: Install JavaScript dependencies
         working-directory: autogpt_platform/frontend

.github/workflows/claude.yml

@@ -93,27 +93,15 @@ jobs:
         run: poetry run prisma generate && poetry run gen-prisma-stub
 
       # Frontend Node.js/pnpm setup (mirrors platform-frontend-ci.yml)
+      - name: Enable corepack
+        run: corepack enable
+
       - name: Set up Node.js
         uses: actions/setup-node@v6
         with:
           node-version: "22"
-
-      - name: Enable corepack
-        run: corepack enable
-
-      - name: Set pnpm store directory
-        run: |
-          pnpm config set store-dir ~/.pnpm-store
-          echo "PNPM_HOME=$HOME/.pnpm-store" >> $GITHUB_ENV
-
-      - name: Cache frontend dependencies
-        uses: actions/cache@v5
-        with:
-          path: ~/.pnpm-store
-          key: ${{ runner.os }}-pnpm-${{ hashFiles('autogpt_platform/frontend/pnpm-lock.yaml', 'autogpt_platform/frontend/package.json') }}
-          restore-keys: |
-            ${{ runner.os }}-pnpm-${{ hashFiles('autogpt_platform/frontend/pnpm-lock.yaml') }}
-            ${{ runner.os }}-pnpm-
+          cache: "pnpm"
+          cache-dependency-path: autogpt_platform/frontend/pnpm-lock.yaml
 
       - name: Install JavaScript dependencies
         working-directory: autogpt_platform/frontend

.github/workflows/codeql.yml

@@ -62,7 +62,7 @@ jobs:
 
     # Initializes the CodeQL tools for scanning.
     - name: Initialize CodeQL
-      uses: github/codeql-action/init@v3
+      uses: github/codeql-action/init@v4
       with:
         languages: ${{ matrix.language }}
         build-mode: ${{ matrix.build-mode }}
@@ -93,6 +93,6 @@ jobs:
         exit 1
 
     - name: Perform CodeQL Analysis
-      uses: github/codeql-action/analyze@v3
+      uses: github/codeql-action/analyze@v4
       with:
         category: "/language:${{matrix.language}}"

.github/workflows/platform-frontend-ci.yml

@@ -26,7 +26,6 @@ jobs:
   setup:
     runs-on: ubuntu-latest
     outputs:
-      cache-key: ${{ steps.cache-key.outputs.key }}
       components-changed: ${{ steps.filter.outputs.components }}
 
     steps:
@@ -41,28 +40,17 @@ jobs:
             components:
               - 'autogpt_platform/frontend/src/components/**'
 
-      - name: Set up Node.js
-        uses: actions/setup-node@v6
-        with:
-          node-version: "22.18.0"
-
       - name: Enable corepack
         run: corepack enable
 
-      - name: Generate cache key
-        id: cache-key
-        run: echo "key=${{ runner.os }}-pnpm-${{ hashFiles('autogpt_platform/frontend/pnpm-lock.yaml', 'autogpt_platform/frontend/package.json') }}" >> $GITHUB_OUTPUT
-
-      - name: Cache dependencies
-        uses: actions/cache@v5
+      - name: Set up Node
+        uses: actions/setup-node@v6
         with:
-          path: ~/.pnpm-store
-          key: ${{ steps.cache-key.outputs.key }}
-          restore-keys: |
-            ${{ runner.os }}-pnpm-${{ hashFiles('autogpt_platform/frontend/pnpm-lock.yaml') }}
-            ${{ runner.os }}-pnpm-
+          node-version: "22.18.0"
+          cache: "pnpm"
+          cache-dependency-path: autogpt_platform/frontend/pnpm-lock.yaml
 
-      - name: Install dependencies
+      - name: Install dependencies to populate cache
         run: pnpm install --frozen-lockfile
 
   lint:
@@ -73,22 +61,15 @@ jobs:
       - name: Checkout repository
         uses: actions/checkout@v6
 
-      - name: Set up Node.js
-        uses: actions/setup-node@v6
-        with:
-          node-version: "22.18.0"
-
       - name: Enable corepack
         run: corepack enable
 
-      - name: Restore dependencies cache
-        uses: actions/cache@v5
+      - name: Set up Node
+        uses: actions/setup-node@v6
         with:
-          path: ~/.pnpm-store
-          key: ${{ needs.setup.outputs.cache-key }}
-          restore-keys: |
-            ${{ runner.os }}-pnpm-${{ hashFiles('autogpt_platform/frontend/pnpm-lock.yaml') }}
-            ${{ runner.os }}-pnpm-
+          node-version: "22.18.0"
+          cache: "pnpm"
+          cache-dependency-path: autogpt_platform/frontend/pnpm-lock.yaml
 
       - name: Install dependencies
         run: pnpm install --frozen-lockfile
@@ -111,22 +92,15 @@ jobs:
         with:
           fetch-depth: 0
 
-      - name: Set up Node.js
-        uses: actions/setup-node@v6
-        with:
-          node-version: "22.18.0"
-
       - name: Enable corepack
         run: corepack enable
 
-      - name: Restore dependencies cache
-        uses: actions/cache@v5
+      - name: Set up Node
+        uses: actions/setup-node@v6
         with:
-          path: ~/.pnpm-store
-          key: ${{ needs.setup.outputs.cache-key }}
-          restore-keys: |
-            ${{ runner.os }}-pnpm-${{ hashFiles('autogpt_platform/frontend/pnpm-lock.yaml') }}
-            ${{ runner.os }}-pnpm-
+          node-version: "22.18.0"
+          cache: "pnpm"
+          cache-dependency-path: autogpt_platform/frontend/pnpm-lock.yaml
 
       - name: Install dependencies
         run: pnpm install --frozen-lockfile
@@ -141,10 +115,8 @@ jobs:
           exitOnceUploaded: true
 
   e2e_test:
+    name: end-to-end tests
     runs-on: big-boi
-    needs: setup
-    strategy:
-      fail-fast: false
 
     steps:
       - name: Checkout repository
@@ -152,19 +124,11 @@ jobs:
         with:
           submodules: recursive
 
-      - name: Set up Node.js
-        uses: actions/setup-node@v6
-        with:
-          node-version: "22.18.0"
-
-      - name: Enable corepack
-        run: corepack enable
-
-      - name: Copy default supabase .env
+      - name: Set up Platform - Copy default supabase .env
         run: |
           cp ../.env.default ../.env
 
-      - name: Copy backend .env and set OpenAI API key
+      - name: Set up Platform - Copy backend .env and set OpenAI API key
         run: |
           cp ../backend/.env.default ../backend/.env
           echo "OPENAI_INTERNAL_API_KEY=${{ secrets.OPENAI_API_KEY }}" >> ../backend/.env
@@ -172,77 +136,125 @@ jobs:
           # Used by E2E test data script to generate embeddings for approved store agents
           OPENAI_API_KEY: ${{ secrets.OPENAI_API_KEY }}
 
-      - name: Set up Docker Buildx
+      - name: Set up Platform - Set up Docker Buildx
         uses: docker/setup-buildx-action@v3
+        with:
+          driver: docker-container
+          driver-opts: network=host
 
-      - name: Cache Docker layers
+      - name: Set up Platform - Expose GHA cache to docker buildx CLI
+        uses: crazy-max/ghaction-github-runtime@v3
+
+      - name: Set up Platform - Build Docker images (with cache)
+        working-directory: autogpt_platform
+        run: |
+          pip install pyyaml
+
+          # Resolve extends and generate a flat compose file that bake can understand
+          docker compose -f docker-compose.yml config > docker-compose.resolved.yml
+
+          # Add cache configuration to the resolved compose file
+          python ../.github/workflows/scripts/docker-ci-fix-compose-build-cache.py \
+            --source docker-compose.resolved.yml \
+            --cache-from "type=gha" \
+            --cache-to "type=gha,mode=max" \
+            --backend-hash "${{ hashFiles('autogpt_platform/backend/Dockerfile', 'autogpt_platform/backend/poetry.lock', 'autogpt_platform/backend/backend') }}" \
+            --frontend-hash "${{ hashFiles('autogpt_platform/frontend/Dockerfile', 'autogpt_platform/frontend/pnpm-lock.yaml', 'autogpt_platform/frontend/src') }}" \
+            --git-ref "${{ github.ref }}"
+
+          # Build with bake using the resolved compose file (now includes cache config)
+          docker buildx bake --allow=fs.read=.. -f docker-compose.resolved.yml --load
+        env:
+          NEXT_PUBLIC_PW_TEST: true
+
+      - name: Set up tests - Cache E2E test data
+        id: e2e-data-cache
         uses: actions/cache@v5
         with:
-          path: /tmp/.buildx-cache
-          key: ${{ runner.os }}-buildx-frontend-test-${{ hashFiles('autogpt_platform/docker-compose.yml', 'autogpt_platform/backend/Dockerfile', 'autogpt_platform/backend/pyproject.toml', 'autogpt_platform/backend/poetry.lock') }}
-          restore-keys: |
-            ${{ runner.os }}-buildx-frontend-test-
+          path: /tmp/e2e_test_data.sql
+          key: e2e-test-data-${{ hashFiles('autogpt_platform/backend/test/e2e_test_data.py', 'autogpt_platform/backend/migrations/**', '.github/workflows/platform-frontend-ci.yml') }}
 
-      - name: Run docker compose
+      - name: Set up Platform - Start Supabase DB + Auth
         run: |
-          NEXT_PUBLIC_PW_TEST=true docker compose -f ../docker-compose.yml up -d
+          docker compose -f ../docker-compose.resolved.yml up -d db auth --no-build
+          echo "Waiting for database to be ready..."
+          timeout 60 sh -c 'until docker compose -f ../docker-compose.resolved.yml exec -T db pg_isready -U postgres 2>/dev/null; do sleep 2; done'
+          echo "Waiting for auth service to be ready..."
+          timeout 60 sh -c 'until docker compose -f ../docker-compose.resolved.yml exec -T db psql -U postgres -d postgres -c "SELECT 1 FROM auth.users LIMIT 1" 2>/dev/null; do sleep 2; done' || echo "Auth schema check timeout, continuing..."
+
+      - name: Set up Platform - Run migrations
+        run: |
+          echo "Running migrations..."
+          docker compose -f ../docker-compose.resolved.yml run --rm migrate
+          echo "✅ Migrations completed"
         env:
-          DOCKER_BUILDKIT: 1
-          BUILDX_CACHE_FROM: type=local,src=/tmp/.buildx-cache
-          BUILDX_CACHE_TO: type=local,dest=/tmp/.buildx-cache-new,mode=max
+          NEXT_PUBLIC_PW_TEST: true
 
-      - name: Move cache
+      - name: Set up tests - Load cached E2E test data
+        if: steps.e2e-data-cache.outputs.cache-hit == 'true'
         run: |
-          rm -rf /tmp/.buildx-cache
-          if [ -d "/tmp/.buildx-cache-new" ]; then
-            mv /tmp/.buildx-cache-new /tmp/.buildx-cache
-          fi
+          echo "✅ Found cached E2E test data, restoring..."
+          {
+            echo "SET session_replication_role = 'replica';"
+            cat /tmp/e2e_test_data.sql
+            echo "SET session_replication_role = 'origin';"
+          } | docker compose -f ../docker-compose.resolved.yml exec -T db psql -U postgres -d postgres -b
+          # Refresh materialized views after restore
+          docker compose -f ../docker-compose.resolved.yml exec -T db \
+            psql -U postgres -d postgres -b -c "SET search_path TO platform; SELECT refresh_store_materialized_views();" || true
 
-      - name: Wait for services to be ready
+          echo "✅ E2E test data restored from cache"
+
+      - name: Set up Platform - Start (all other services)
         run: |
+          docker compose -f ../docker-compose.resolved.yml up -d --no-build
           echo "Waiting for rest_server to be ready..."
           timeout 60 sh -c 'until curl -f http://localhost:8006/health 2>/dev/null; do sleep 2; done' || echo "Rest server health check timeout, continuing..."
-          echo "Waiting for database to be ready..."
-          timeout 60 sh -c 'until docker compose -f ../docker-compose.yml exec -T db pg_isready -U postgres 2>/dev/null; do sleep 2; done' || echo "Database ready check timeout, continuing..."
+        env:
+          NEXT_PUBLIC_PW_TEST: true
 
-      - name: Create E2E test data
+      - name: Set up tests - Create E2E test data
+        if: steps.e2e-data-cache.outputs.cache-hit != 'true'
         run: |
           echo "Creating E2E test data..."
-          # First try to run the script from inside the container
-          if docker compose -f ../docker-compose.yml exec -T rest_server test -f /app/autogpt_platform/backend/test/e2e_test_data.py; then
-            echo "✅ Found e2e_test_data.py in container, running it..."
-            docker compose -f ../docker-compose.yml exec -T rest_server sh -c "cd /app/autogpt_platform && python backend/test/e2e_test_data.py" || {
-              echo "❌ E2E test data creation failed!"
-              docker compose -f ../docker-compose.yml logs --tail=50 rest_server
-              exit 1
-            }
-          else
-            echo "⚠️ e2e_test_data.py not found in container, copying and running..."
-            # Copy the script into the container and run it
-            docker cp ../backend/test/e2e_test_data.py $(docker compose -f ../docker-compose.yml ps -q rest_server):/tmp/e2e_test_data.py || {
-              echo "❌ Failed to copy script to container"
-              exit 1
-            }
-            docker compose -f ../docker-compose.yml exec -T rest_server sh -c "cd /app/autogpt_platform && python /tmp/e2e_test_data.py" || {
-              echo "❌ E2E test data creation failed!"
-              docker compose -f ../docker-compose.yml logs --tail=50 rest_server
-              exit 1
-            }
-          fi
+          docker cp ../backend/test/e2e_test_data.py $(docker compose -f ../docker-compose.resolved.yml ps -q rest_server):/tmp/e2e_test_data.py
+          docker compose -f ../docker-compose.resolved.yml exec -T rest_server sh -c "cd /app/autogpt_platform && python /tmp/e2e_test_data.py" || {
+            echo "❌ E2E test data creation failed!"
+            docker compose -f ../docker-compose.resolved.yml logs --tail=50 rest_server
+            exit 1
+          }
 
-      - name: Restore dependencies cache
-        uses: actions/cache@v5
+          # Dump auth.users + platform schema for cache (two separate dumps)
+          echo "Dumping database for cache..."
+          {
+            docker compose -f ../docker-compose.resolved.yml exec -T db \
+              pg_dump -U postgres --data-only --column-inserts \
+              --table='auth.users' postgres
+            docker compose -f ../docker-compose.resolved.yml exec -T db \
+              pg_dump -U postgres --data-only --column-inserts \
+              --schema=platform \
+              --exclude-table='platform._prisma_migrations' \
+              --exclude-table='platform.apscheduler_jobs' \
+              --exclude-table='platform.apscheduler_jobs_batched_notifications' \
+              postgres
+          } > /tmp/e2e_test_data.sql
+
+          echo "✅ Database dump created for caching ($(wc -l < /tmp/e2e_test_data.sql) lines)"
+
+      - name: Set up tests - Enable corepack
+        run: corepack enable
+
+      - name: Set up tests - Set up Node
+        uses: actions/setup-node@v6
         with:
-          path: ~/.pnpm-store
-          key: ${{ needs.setup.outputs.cache-key }}
-          restore-keys: |
-            ${{ runner.os }}-pnpm-${{ hashFiles('autogpt_platform/frontend/pnpm-lock.yaml') }}
-            ${{ runner.os }}-pnpm-
+          node-version: "22.18.0"
+          cache: "pnpm"
+          cache-dependency-path: autogpt_platform/frontend/pnpm-lock.yaml
 
-      - name: Install dependencies
+      - name: Set up tests - Install dependencies
         run: pnpm install --frozen-lockfile
 
-      - name: Install Browser 'chromium'
+      - name: Set up tests - Install browser 'chromium'
         run: pnpm playwright install --with-deps chromium
 
       - name: Run Playwright tests
@@ -269,7 +281,7 @@ jobs:
 
       - name: Print Final Docker Compose logs
         if: always()
-        run: docker compose -f ../docker-compose.yml logs
+        run: docker compose -f ../docker-compose.resolved.yml logs
 
   integration_test:
     runs-on: ubuntu-latest
@@ -281,22 +293,15 @@ jobs:
         with:
           submodules: recursive
 
-      - name: Set up Node.js
-        uses: actions/setup-node@v6
-        with:
-          node-version: "22.18.0"
-
       - name: Enable corepack
         run: corepack enable
 
-      - name: Restore dependencies cache
-        uses: actions/cache@v5
+      - name: Set up Node
+        uses: actions/setup-node@v6
         with:
-          path: ~/.pnpm-store
-          key: ${{ needs.setup.outputs.cache-key }}
-          restore-keys: |
-            ${{ runner.os }}-pnpm-${{ hashFiles('autogpt_platform/frontend/pnpm-lock.yaml') }}
-            ${{ runner.os }}-pnpm-
+          node-version: "22.18.0"
+          cache: "pnpm"
+          cache-dependency-path: autogpt_platform/frontend/pnpm-lock.yaml
 
       - name: Install dependencies
         run: pnpm install --frozen-lockfile

.github/workflows/pr-overlap-check.yml

@@ -0,0 +1,39 @@
+name: PR Overlap Detection
+
+on:
+  pull_request:
+    types: [opened, synchronize, reopened]
+    branches:
+      - dev
+      - master
+
+permissions:
+  contents: read
+  pull-requests: write
+
+jobs:
+  check-overlaps:
+    runs-on: ubuntu-latest
+    steps:
+      - name: Checkout repository
+        uses: actions/checkout@v4
+        with:
+          fetch-depth: 0  # Need full history for merge testing
+
+      - name: Set up Python
+        uses: actions/setup-python@v5
+        with:
+          python-version: '3.11'
+
+      - name: Configure git
+        run: |
+          git config user.email "github-actions[bot]@users.noreply.github.com"
+          git config user.name "github-actions[bot]"
+
+      - name: Run overlap detection
+        env:
+          GH_TOKEN: ${{ secrets.GITHUB_TOKEN }}
+        # Always succeed - this check informs contributors, it shouldn't block merging
+        continue-on-error: true
+        run: |
+          python .github/scripts/detect_overlaps.py ${{ github.event.pull_request.number }}

.github/workflows/scripts/docker-ci-fix-compose-build-cache.py

@@ -0,0 +1,195 @@
+#!/usr/bin/env python3
+"""
+Add cache configuration to a resolved docker-compose file for all services
+that have a build key, and ensure image names match what docker compose expects.
+"""
+
+import argparse
+
+import yaml
+
+
+DEFAULT_BRANCH = "dev"
+CACHE_BUILDS_FOR_COMPONENTS = ["backend", "frontend"]
+
+
+def main():
+    parser = argparse.ArgumentParser(
+        description="Add cache config to a resolved compose file"
+    )
+    parser.add_argument(
+        "--source",
+        required=True,
+        help="Source compose file to read (should be output of `docker compose config`)",
+    )
+    parser.add_argument(
+        "--cache-from",
+        default="type=gha",
+        help="Cache source configuration",
+    )
+    parser.add_argument(
+        "--cache-to",
+        default="type=gha,mode=max",
+        help="Cache destination configuration",
+    )
+    for component in CACHE_BUILDS_FOR_COMPONENTS:
+        parser.add_argument(
+            f"--{component}-hash",
+            default="",
+            help=f"Hash for {component} cache scope (e.g., from hashFiles())",
+        )
+    parser.add_argument(
+        "--git-ref",
+        default="",
+        help="Git ref for branch-based cache scope (e.g., refs/heads/master)",
+    )
+    args = parser.parse_args()
+
+    # Normalize git ref to a safe scope name (e.g., refs/heads/master -> master)
+    git_ref_scope = ""
+    if args.git_ref:
+        git_ref_scope = args.git_ref.replace("refs/heads/", "").replace("/", "-")
+
+    with open(args.source, "r") as f:
+        compose = yaml.safe_load(f)
+
+    # Get project name from compose file or default
+    project_name = compose.get("name", "autogpt_platform")
+
+    def get_image_name(dockerfile: str, target: str) -> str:
+        """Generate image name based on Dockerfile folder and build target."""
+        dockerfile_parts = dockerfile.replace("\\", "/").split("/")
+        if len(dockerfile_parts) >= 2:
+            folder_name = dockerfile_parts[-2]  # e.g., "backend" or "frontend"
+        else:
+            folder_name = "app"
+        return f"{project_name}-{folder_name}:{target}"
+
+    def get_build_key(dockerfile: str, target: str) -> str:
+        """Generate a unique key for a Dockerfile+target combination."""
+        return f"{dockerfile}:{target}"
+
+    def get_component(dockerfile: str) -> str | None:
+        """Get component name (frontend/backend) from dockerfile path."""
+        for component in CACHE_BUILDS_FOR_COMPONENTS:
+            if component in dockerfile:
+                return component
+        return None
+
+    # First pass: collect all services with build configs and identify duplicates
+    # Track which (dockerfile, target) combinations we've seen
+    build_key_to_first_service: dict[str, str] = {}
+    services_to_build: list[str] = []
+    services_to_dedupe: list[str] = []
+
+    for service_name, service_config in compose.get("services", {}).items():
+        if "build" not in service_config:
+            continue
+
+        build_config = service_config["build"]
+        dockerfile = build_config.get("dockerfile", "Dockerfile")
+        target = build_config.get("target", "default")
+        build_key = get_build_key(dockerfile, target)
+
+        if build_key not in build_key_to_first_service:
+            # First service with this build config - it will do the actual build
+            build_key_to_first_service[build_key] = service_name
+            services_to_build.append(service_name)
+        else:
+            # Duplicate - will just use the image from the first service
+            services_to_dedupe.append(service_name)
+
+    # Second pass: configure builds and deduplicate
+    modified_services = []
+    for service_name, service_config in compose.get("services", {}).items():
+        if "build" not in service_config:
+            continue
+
+        build_config = service_config["build"]
+        dockerfile = build_config.get("dockerfile", "Dockerfile")
+        target = build_config.get("target", "latest")
+        image_name = get_image_name(dockerfile, target)
+
+        # Set image name for all services (needed for both builders and deduped)
+        service_config["image"] = image_name
+
+        if service_name in services_to_dedupe:
+            # Remove build config - this service will use the pre-built image
+            del service_config["build"]
+            continue
+
+        # This service will do the actual build - add cache config
+        cache_from_list = []
+        cache_to_list = []
+
+        component = get_component(dockerfile)
+        if not component:
+            # Skip services that don't clearly match frontend/backend
+            continue
+
+        # Get the hash for this component
+        component_hash = getattr(args, f"{component}_hash")
+
+        # Scope format: platform-{component}-{target}-{hash|ref}
+        # Example: platform-backend-server-abc123
+
+        if "type=gha" in args.cache_from:
+            # 1. Primary: exact hash match (most specific)
+            if component_hash:
+                hash_scope = f"platform-{component}-{target}-{component_hash}"
+                cache_from_list.append(f"{args.cache_from},scope={hash_scope}")
+
+            # 2. Fallback: branch-based cache
+            if git_ref_scope:
+                ref_scope = f"platform-{component}-{target}-{git_ref_scope}"
+                cache_from_list.append(f"{args.cache_from},scope={ref_scope}")
+
+            # 3. Fallback: dev branch cache (for PRs/feature branches)
+            if git_ref_scope and git_ref_scope != DEFAULT_BRANCH:
+                master_scope = f"platform-{component}-{target}-{DEFAULT_BRANCH}"
+                cache_from_list.append(f"{args.cache_from},scope={master_scope}")
+
+        if "type=gha" in args.cache_to:
+            # Write to both hash-based and branch-based scopes
+            if component_hash:
+                hash_scope = f"platform-{component}-{target}-{component_hash}"
+                cache_to_list.append(f"{args.cache_to},scope={hash_scope}")
+
+            if git_ref_scope:
+                ref_scope = f"platform-{component}-{target}-{git_ref_scope}"
+                cache_to_list.append(f"{args.cache_to},scope={ref_scope}")
+
+        # Ensure we have at least one cache source/target
+        if not cache_from_list:
+            cache_from_list.append(args.cache_from)
+        if not cache_to_list:
+            cache_to_list.append(args.cache_to)
+
+        build_config["cache_from"] = cache_from_list
+        build_config["cache_to"] = cache_to_list
+        modified_services.append(service_name)
+
+    # Write back to the same file
+    with open(args.source, "w") as f:
+        yaml.dump(compose, f, default_flow_style=False, sort_keys=False)
+
+    print(f"Added cache config to {len(modified_services)} services in {args.source}:")
+    for svc in modified_services:
+        svc_config = compose["services"][svc]
+        build_cfg = svc_config.get("build", {})
+        cache_from_list = build_cfg.get("cache_from", ["none"])
+        cache_to_list = build_cfg.get("cache_to", ["none"])
+        print(f"  - {svc}")
+        print(f"      image: {svc_config.get('image', 'N/A')}")
+        print(f"      cache_from: {cache_from_list}")
+        print(f"      cache_to: {cache_to_list}")
+    if services_to_dedupe:
+        print(
+            f"Deduplicated {len(services_to_dedupe)} services (will use pre-built images):"
+        )
+        for svc in services_to_dedupe:
+            print(f"  - {svc} -> {compose['services'][svc].get('image', 'N/A')}")
+
+
+if __name__ == "__main__":
+    main()

autogpt_platform/CLAUDE.md

@@ -45,6 +45,11 @@ AutoGPT Platform is a monorepo containing:
 - Backend/Frontend services use YAML anchors for consistent configuration
 - Supabase services (`db/docker/docker-compose.yml`) follow the same pattern
 
+### Branching Strategy
+
+- **`dev`** is the main development branch. All PRs should target `dev`.
+- **`master`** is the production branch. Only used for production releases.
+
 ### Creating Pull Requests
 
 - Create the PR against the `dev` branch of the repository.

autogpt_platform/autogpt_libs/poetry.lock

@@ -448,61 +448,61 @@ toml = ["tomli ; python_full_version <= \"3.11.0a6\""]
 
 [[package]]
 name = "cryptography"
-version = "46.0.4"
+version = "46.0.5"
 description = "cryptography is a package which provides cryptographic recipes and primitives to Python developers."
 optional = false
 python-versions = "!=3.9.0,!=3.9.1,>=3.8"
 groups = ["main"]
 files = [
-    {file = "cryptography-46.0.4-cp311-abi3-macosx_10_9_universal2.whl", hash = "sha256:281526e865ed4166009e235afadf3a4c4cba6056f99336a99efba65336fd5485"},
-    {file = "cryptography-46.0.4-cp311-abi3-manylinux2014_aarch64.manylinux_2_17_aarch64.whl", hash = "sha256:5f14fba5bf6f4390d7ff8f086c566454bff0411f6d8aa7af79c88b6f9267aecc"},
-    {file = "cryptography-46.0.4-cp311-abi3-manylinux2014_x86_64.manylinux_2_17_x86_64.whl", hash = "sha256:47bcd19517e6389132f76e2d5303ded6cf3f78903da2158a671be8de024f4cd0"},
-    {file = "cryptography-46.0.4-cp311-abi3-manylinux_2_28_aarch64.whl", hash = "sha256:01df4f50f314fbe7009f54046e908d1754f19d0c6d3070df1e6268c5a4af09fa"},
-    {file = "cryptography-46.0.4-cp311-abi3-manylinux_2_28_ppc64le.whl", hash = "sha256:5aa3e463596b0087b3da0dbe2b2487e9fc261d25da85754e30e3b40637d61f81"},
-    {file = "cryptography-46.0.4-cp311-abi3-manylinux_2_28_x86_64.whl", hash = "sha256:0a9ad24359fee86f131836a9ac3bffc9329e956624a2d379b613f8f8abaf5255"},
-    {file = "cryptography-46.0.4-cp311-abi3-manylinux_2_31_armv7l.whl", hash = "sha256:dc1272e25ef673efe72f2096e92ae39dea1a1a450dd44918b15351f72c5a168e"},
-    {file = "cryptography-46.0.4-cp311-abi3-manylinux_2_34_aarch64.whl", hash = "sha256:de0f5f4ec8711ebc555f54735d4c673fc34b65c44283895f1a08c2b49d2fd99c"},
-    {file = "cryptography-46.0.4-cp311-abi3-manylinux_2_34_ppc64le.whl", hash = "sha256:eeeb2e33d8dbcccc34d64651f00a98cb41b2dc69cef866771a5717e6734dfa32"},
-    {file = "cryptography-46.0.4-cp311-abi3-manylinux_2_34_x86_64.whl", hash = "sha256:3d425eacbc9aceafd2cb429e42f4e5d5633c6f873f5e567077043ef1b9bbf616"},
-    {file = "cryptography-46.0.4-cp311-abi3-musllinux_1_2_aarch64.whl", hash = "sha256:91627ebf691d1ea3976a031b61fb7bac1ccd745afa03602275dda443e11c8de0"},
-    {file = "cryptography-46.0.4-cp311-abi3-musllinux_1_2_x86_64.whl", hash = "sha256:2d08bc22efd73e8854b0b7caff402d735b354862f1145d7be3b9c0f740fef6a0"},
-    {file = "cryptography-46.0.4-cp311-abi3-win32.whl", hash = "sha256:82a62483daf20b8134f6e92898da70d04d0ef9a75829d732ea1018678185f4f5"},
-    {file = "cryptography-46.0.4-cp311-abi3-win_amd64.whl", hash = "sha256:6225d3ebe26a55dbc8ead5ad1265c0403552a63336499564675b29eb3184c09b"},
-    {file = "cryptography-46.0.4-cp314-cp314t-macosx_10_9_universal2.whl", hash = "sha256:485e2b65d25ec0d901bca7bcae0f53b00133bf3173916d8e421f6fddde103908"},
-    {file = "cryptography-46.0.4-cp314-cp314t-manylinux2014_aarch64.manylinux_2_17_aarch64.whl", hash = "sha256:078e5f06bd2fa5aea5a324f2a09f914b1484f1d0c2a4d6a8a28c74e72f65f2da"},
-    {file = "cryptography-46.0.4-cp314-cp314t-manylinux2014_x86_64.manylinux_2_17_x86_64.whl", hash = "sha256:dce1e4f068f03008da7fa51cc7abc6ddc5e5de3e3d1550334eaf8393982a5829"},
-    {file = "cryptography-46.0.4-cp314-cp314t-manylinux_2_28_aarch64.whl", hash = "sha256:2067461c80271f422ee7bdbe79b9b4be54a5162e90345f86a23445a0cf3fd8a2"},
-    {file = "cryptography-46.0.4-cp314-cp314t-manylinux_2_28_ppc64le.whl", hash = "sha256:c92010b58a51196a5f41c3795190203ac52edfd5dc3ff99149b4659eba9d2085"},
-    {file = "cryptography-46.0.4-cp314-cp314t-manylinux_2_28_x86_64.whl", hash = "sha256:829c2b12bbc5428ab02d6b7f7e9bbfd53e33efd6672d21341f2177470171ad8b"},
-    {file = "cryptography-46.0.4-cp314-cp314t-manylinux_2_31_armv7l.whl", hash = "sha256:62217ba44bf81b30abaeda1488686a04a702a261e26f87db51ff61d9d3510abd"},
-    {file = "cryptography-46.0.4-cp314-cp314t-manylinux_2_34_aarch64.whl", hash = "sha256:9c2da296c8d3415b93e6053f5a728649a87a48ce084a9aaf51d6e46c87c7f2d2"},
-    {file = "cryptography-46.0.4-cp314-cp314t-manylinux_2_34_ppc64le.whl", hash = "sha256:9b34d8ba84454641a6bf4d6762d15847ecbd85c1316c0a7984e6e4e9f748ec2e"},
-    {file = "cryptography-46.0.4-cp314-cp314t-manylinux_2_34_x86_64.whl", hash = "sha256:df4a817fa7138dd0c96c8c8c20f04b8aaa1fac3bbf610913dcad8ea82e1bfd3f"},
-    {file = "cryptography-46.0.4-cp314-cp314t-musllinux_1_2_aarch64.whl", hash = "sha256:b1de0ebf7587f28f9190b9cb526e901bf448c9e6a99655d2b07fff60e8212a82"},
-    {file = "cryptography-46.0.4-cp314-cp314t-musllinux_1_2_x86_64.whl", hash = "sha256:9b4d17bc7bd7cdd98e3af40b441feaea4c68225e2eb2341026c84511ad246c0c"},
-    {file = "cryptography-46.0.4-cp314-cp314t-win32.whl", hash = "sha256:c411f16275b0dea722d76544a61d6421e2cc829ad76eec79280dbdc9ddf50061"},
-    {file = "cryptography-46.0.4-cp314-cp314t-win_amd64.whl", hash = "sha256:728fedc529efc1439eb6107b677f7f7558adab4553ef8669f0d02d42d7b959a7"},
-    {file = "cryptography-46.0.4-cp38-abi3-macosx_10_9_universal2.whl", hash = "sha256:a9556ba711f7c23f77b151d5798f3ac44a13455cc68db7697a1096e6d0563cab"},
-    {file = "cryptography-46.0.4-cp38-abi3-manylinux2014_aarch64.manylinux_2_17_aarch64.whl", hash = "sha256:8bf75b0259e87fa70bddc0b8b4078b76e7fd512fd9afae6c1193bcf440a4dbef"},
-    {file = "cryptography-46.0.4-cp38-abi3-manylinux2014_x86_64.manylinux_2_17_x86_64.whl", hash = "sha256:3c268a3490df22270955966ba236d6bc4a8f9b6e4ffddb78aac535f1a5ea471d"},
-    {file = "cryptography-46.0.4-cp38-abi3-manylinux_2_28_aarch64.whl", hash = "sha256:812815182f6a0c1d49a37893a303b44eaac827d7f0d582cecfc81b6427f22973"},
-    {file = "cryptography-46.0.4-cp38-abi3-manylinux_2_28_ppc64le.whl", hash = "sha256:a90e43e3ef65e6dcf969dfe3bb40cbf5aef0d523dff95bfa24256be172a845f4"},
-    {file = "cryptography-46.0.4-cp38-abi3-manylinux_2_28_x86_64.whl", hash = "sha256:a05177ff6296644ef2876fce50518dffb5bcdf903c85250974fc8bc85d54c0af"},
-    {file = "cryptography-46.0.4-cp38-abi3-manylinux_2_31_armv7l.whl", hash = "sha256:daa392191f626d50f1b136c9b4cf08af69ca8279d110ea24f5c2700054d2e263"},
-    {file = "cryptography-46.0.4-cp38-abi3-manylinux_2_34_aarch64.whl", hash = "sha256:e07ea39c5b048e085f15923511d8121e4a9dc45cee4e3b970ca4f0d338f23095"},
-    {file = "cryptography-46.0.4-cp38-abi3-manylinux_2_34_ppc64le.whl", hash = "sha256:d5a45ddc256f492ce42a4e35879c5e5528c09cd9ad12420828c972951d8e016b"},
-    {file = "cryptography-46.0.4-cp38-abi3-manylinux_2_34_x86_64.whl", hash = "sha256:6bb5157bf6a350e5b28aee23beb2d84ae6f5be390b2f8ee7ea179cda077e1019"},
-    {file = "cryptography-46.0.4-cp38-abi3-musllinux_1_2_aarch64.whl", hash = "sha256:dd5aba870a2c40f87a3af043e0dee7d9eb02d4aff88a797b48f2b43eff8c3ab4"},
-    {file = "cryptography-46.0.4-cp38-abi3-musllinux_1_2_x86_64.whl", hash = "sha256:93d8291da8d71024379ab2cb0b5c57915300155ad42e07f76bea6ad838d7e59b"},
-    {file = "cryptography-46.0.4-cp38-abi3-win32.whl", hash = "sha256:0563655cb3c6d05fb2afe693340bc050c30f9f34e15763361cf08e94749401fc"},
-    {file = "cryptography-46.0.4-cp38-abi3-win_amd64.whl", hash = "sha256:fa0900b9ef9c49728887d1576fd8d9e7e3ea872fa9b25ef9b64888adc434e976"},
-    {file = "cryptography-46.0.4-pp311-pypy311_pp73-macosx_11_0_arm64.whl", hash = "sha256:766330cce7416c92b5e90c3bb71b1b79521760cdcfc3a6a1a182d4c9fab23d2b"},
-    {file = "cryptography-46.0.4-pp311-pypy311_pp73-manylinux_2_28_aarch64.whl", hash = "sha256:c236a44acfb610e70f6b3e1c3ca20ff24459659231ef2f8c48e879e2d32b73da"},
-    {file = "cryptography-46.0.4-pp311-pypy311_pp73-manylinux_2_28_x86_64.whl", hash = "sha256:8a15fb869670efa8f83cbffbc8753c1abf236883225aed74cd179b720ac9ec80"},
-    {file = "cryptography-46.0.4-pp311-pypy311_pp73-manylinux_2_34_aarch64.whl", hash = "sha256:fdc3daab53b212472f1524d070735b2f0c214239df131903bae1d598016fa822"},
-    {file = "cryptography-46.0.4-pp311-pypy311_pp73-manylinux_2_34_x86_64.whl", hash = "sha256:44cc0675b27cadb71bdbb96099cca1fa051cd11d2ade09e5cd3a2edb929ed947"},
-    {file = "cryptography-46.0.4-pp311-pypy311_pp73-win_amd64.whl", hash = "sha256:be8c01a7d5a55f9a47d1888162b76c8f49d62b234d88f0ff91a9fbebe32ffbc3"},
-    {file = "cryptography-46.0.4.tar.gz", hash = "sha256:bfd019f60f8abc2ed1b9be4ddc21cfef059c841d86d710bb69909a688cbb8f59"},
+    {file = "cryptography-46.0.5-cp311-abi3-macosx_10_9_universal2.whl", hash = "sha256:351695ada9ea9618b3500b490ad54c739860883df6c1f555e088eaf25b1bbaad"},
+    {file = "cryptography-46.0.5-cp311-abi3-manylinux2014_aarch64.manylinux_2_17_aarch64.whl", hash = "sha256:c18ff11e86df2e28854939acde2d003f7984f721eba450b56a200ad90eeb0e6b"},
+    {file = "cryptography-46.0.5-cp311-abi3-manylinux2014_x86_64.manylinux_2_17_x86_64.whl", hash = "sha256:4d7e3d356b8cd4ea5aff04f129d5f66ebdc7b6f8eae802b93739ed520c47c79b"},
+    {file = "cryptography-46.0.5-cp311-abi3-manylinux_2_28_aarch64.whl", hash = "sha256:50bfb6925eff619c9c023b967d5b77a54e04256c4281b0e21336a130cd7fc263"},
+    {file = "cryptography-46.0.5-cp311-abi3-manylinux_2_28_ppc64le.whl", hash = "sha256:803812e111e75d1aa73690d2facc295eaefd4439be1023fefc4995eaea2af90d"},
+    {file = "cryptography-46.0.5-cp311-abi3-manylinux_2_28_x86_64.whl", hash = "sha256:3ee190460e2fbe447175cda91b88b84ae8322a104fc27766ad09428754a618ed"},
+    {file = "cryptography-46.0.5-cp311-abi3-manylinux_2_31_armv7l.whl", hash = "sha256:f145bba11b878005c496e93e257c1e88f154d278d2638e6450d17e0f31e558d2"},
+    {file = "cryptography-46.0.5-cp311-abi3-manylinux_2_34_aarch64.whl", hash = "sha256:e9251e3be159d1020c4030bd2e5f84d6a43fe54b6c19c12f51cde9542a2817b2"},
+    {file = "cryptography-46.0.5-cp311-abi3-manylinux_2_34_ppc64le.whl", hash = "sha256:47fb8a66058b80e509c47118ef8a75d14c455e81ac369050f20ba0d23e77fee0"},
+    {file = "cryptography-46.0.5-cp311-abi3-manylinux_2_34_x86_64.whl", hash = "sha256:4c3341037c136030cb46e4b1e17b7418ea4cbd9dd207e4a6f3b2b24e0d4ac731"},
+    {file = "cryptography-46.0.5-cp311-abi3-musllinux_1_2_aarch64.whl", hash = "sha256:890bcb4abd5a2d3f852196437129eb3667d62630333aacc13dfd470fad3aaa82"},
+    {file = "cryptography-46.0.5-cp311-abi3-musllinux_1_2_x86_64.whl", hash = "sha256:80a8d7bfdf38f87ca30a5391c0c9ce4ed2926918e017c29ddf643d0ed2778ea1"},
+    {file = "cryptography-46.0.5-cp311-abi3-win32.whl", hash = "sha256:60ee7e19e95104d4c03871d7d7dfb3d22ef8a9b9c6778c94e1c8fcc8365afd48"},
+    {file = "cryptography-46.0.5-cp311-abi3-win_amd64.whl", hash = "sha256:38946c54b16c885c72c4f59846be9743d699eee2b69b6988e0a00a01f46a61a4"},
+    {file = "cryptography-46.0.5-cp314-cp314t-macosx_10_9_universal2.whl", hash = "sha256:94a76daa32eb78d61339aff7952ea819b1734b46f73646a07decb40e5b3448e2"},
+    {file = "cryptography-46.0.5-cp314-cp314t-manylinux2014_aarch64.manylinux_2_17_aarch64.whl", hash = "sha256:5be7bf2fb40769e05739dd0046e7b26f9d4670badc7b032d6ce4db64dddc0678"},
+    {file = "cryptography-46.0.5-cp314-cp314t-manylinux2014_x86_64.manylinux_2_17_x86_64.whl", hash = "sha256:fe346b143ff9685e40192a4960938545c699054ba11d4f9029f94751e3f71d87"},
+    {file = "cryptography-46.0.5-cp314-cp314t-manylinux_2_28_aarch64.whl", hash = "sha256:c69fd885df7d089548a42d5ec05be26050ebcd2283d89b3d30676eb32ff87dee"},
+    {file = "cryptography-46.0.5-cp314-cp314t-manylinux_2_28_ppc64le.whl", hash = "sha256:8293f3dea7fc929ef7240796ba231413afa7b68ce38fd21da2995549f5961981"},
+    {file = "cryptography-46.0.5-cp314-cp314t-manylinux_2_28_x86_64.whl", hash = "sha256:1abfdb89b41c3be0365328a410baa9df3ff8a9110fb75e7b52e66803ddabc9a9"},
+    {file = "cryptography-46.0.5-cp314-cp314t-manylinux_2_31_armv7l.whl", hash = "sha256:d66e421495fdb797610a08f43b05269e0a5ea7f5e652a89bfd5a7d3c1dee3648"},
+    {file = "cryptography-46.0.5-cp314-cp314t-manylinux_2_34_aarch64.whl", hash = "sha256:4e817a8920bfbcff8940ecfd60f23d01836408242b30f1a708d93198393a80b4"},
+    {file = "cryptography-46.0.5-cp314-cp314t-manylinux_2_34_ppc64le.whl", hash = "sha256:68f68d13f2e1cb95163fa3b4db4bf9a159a418f5f6e7242564fc75fcae667fd0"},
+    {file = "cryptography-46.0.5-cp314-cp314t-manylinux_2_34_x86_64.whl", hash = "sha256:a3d1fae9863299076f05cb8a778c467578262fae09f9dc0ee9b12eb4268ce663"},
+    {file = "cryptography-46.0.5-cp314-cp314t-musllinux_1_2_aarch64.whl", hash = "sha256:c4143987a42a2397f2fc3b4d7e3a7d313fbe684f67ff443999e803dd75a76826"},
+    {file = "cryptography-46.0.5-cp314-cp314t-musllinux_1_2_x86_64.whl", hash = "sha256:7d731d4b107030987fd61a7f8ab512b25b53cef8f233a97379ede116f30eb67d"},
+    {file = "cryptography-46.0.5-cp314-cp314t-win32.whl", hash = "sha256:c3bcce8521d785d510b2aad26ae2c966092b7daa8f45dd8f44734a104dc0bc1a"},
+    {file = "cryptography-46.0.5-cp314-cp314t-win_amd64.whl", hash = "sha256:4d8ae8659ab18c65ced284993c2265910f6c9e650189d4e3f68445ef82a810e4"},
+    {file = "cryptography-46.0.5-cp38-abi3-macosx_10_9_universal2.whl", hash = "sha256:4108d4c09fbbf2789d0c926eb4152ae1760d5a2d97612b92d508d96c861e4d31"},
+    {file = "cryptography-46.0.5-cp38-abi3-manylinux2014_aarch64.manylinux_2_17_aarch64.whl", hash = "sha256:7d1f30a86d2757199cb2d56e48cce14deddf1f9c95f1ef1b64ee91ea43fe2e18"},
+    {file = "cryptography-46.0.5-cp38-abi3-manylinux2014_x86_64.manylinux_2_17_x86_64.whl", hash = "sha256:039917b0dc418bb9f6edce8a906572d69e74bd330b0b3fea4f79dab7f8ddd235"},
+    {file = "cryptography-46.0.5-cp38-abi3-manylinux_2_28_aarch64.whl", hash = "sha256:ba2a27ff02f48193fc4daeadf8ad2590516fa3d0adeeb34336b96f7fa64c1e3a"},
+    {file = "cryptography-46.0.5-cp38-abi3-manylinux_2_28_ppc64le.whl", hash = "sha256:61aa400dce22cb001a98014f647dc21cda08f7915ceb95df0c9eaf84b4b6af76"},
+    {file = "cryptography-46.0.5-cp38-abi3-manylinux_2_28_x86_64.whl", hash = "sha256:3ce58ba46e1bc2aac4f7d9290223cead56743fa6ab94a5d53292ffaac6a91614"},
+    {file = "cryptography-46.0.5-cp38-abi3-manylinux_2_31_armv7l.whl", hash = "sha256:420d0e909050490d04359e7fdb5ed7e667ca5c3c402b809ae2563d7e66a92229"},
+    {file = "cryptography-46.0.5-cp38-abi3-manylinux_2_34_aarch64.whl", hash = "sha256:582f5fcd2afa31622f317f80426a027f30dc792e9c80ffee87b993200ea115f1"},
+    {file = "cryptography-46.0.5-cp38-abi3-manylinux_2_34_ppc64le.whl", hash = "sha256:bfd56bb4b37ed4f330b82402f6f435845a5f5648edf1ad497da51a8452d5d62d"},
+    {file = "cryptography-46.0.5-cp38-abi3-manylinux_2_34_x86_64.whl", hash = "sha256:a3d507bb6a513ca96ba84443226af944b0f7f47dcc9a399d110cd6146481d24c"},
+    {file = "cryptography-46.0.5-cp38-abi3-musllinux_1_2_aarch64.whl", hash = "sha256:9f16fbdf4da055efb21c22d81b89f155f02ba420558db21288b3d0035bafd5f4"},
+    {file = "cryptography-46.0.5-cp38-abi3-musllinux_1_2_x86_64.whl", hash = "sha256:ced80795227d70549a411a4ab66e8ce307899fad2220ce5ab2f296e687eacde9"},
+    {file = "cryptography-46.0.5-cp38-abi3-win32.whl", hash = "sha256:02f547fce831f5096c9a567fd41bc12ca8f11df260959ecc7c3202555cc47a72"},
+    {file = "cryptography-46.0.5-cp38-abi3-win_amd64.whl", hash = "sha256:556e106ee01aa13484ce9b0239bca667be5004efb0aabbed28d353df86445595"},
+    {file = "cryptography-46.0.5-pp311-pypy311_pp73-macosx_11_0_arm64.whl", hash = "sha256:3b4995dc971c9fb83c25aa44cf45f02ba86f71ee600d81091c2f0cbae116b06c"},
+    {file = "cryptography-46.0.5-pp311-pypy311_pp73-manylinux_2_28_aarch64.whl", hash = "sha256:bc84e875994c3b445871ea7181d424588171efec3e185dced958dad9e001950a"},
+    {file = "cryptography-46.0.5-pp311-pypy311_pp73-manylinux_2_28_x86_64.whl", hash = "sha256:2ae6971afd6246710480e3f15824ed3029a60fc16991db250034efd0b9fb4356"},
+    {file = "cryptography-46.0.5-pp311-pypy311_pp73-manylinux_2_34_aarch64.whl", hash = "sha256:d861ee9e76ace6cf36a6a89b959ec08e7bc2493ee39d07ffe5acb23ef46d27da"},
+    {file = "cryptography-46.0.5-pp311-pypy311_pp73-manylinux_2_34_x86_64.whl", hash = "sha256:2b7a67c9cd56372f3249b39699f2ad479f6991e62ea15800973b956f4b73e257"},
+    {file = "cryptography-46.0.5-pp311-pypy311_pp73-win_amd64.whl", hash = "sha256:8456928655f856c6e1533ff59d5be76578a7157224dbd9ce6872f25055ab9ab7"},
+    {file = "cryptography-46.0.5.tar.gz", hash = "sha256:abace499247268e3757271b2f1e244b36b06f8515cf27c4d49468fc9eb16e93d"},
 ]
 
 [package.dependencies]
@@ -516,7 +516,7 @@ nox = ["nox[uv] (>=2024.4.15)"]
 pep8test = ["check-sdist", "click (>=8.0.1)", "mypy (>=1.14)", "ruff (>=0.11.11)"]
 sdist = ["build (>=1.0.0)"]
 ssh = ["bcrypt (>=3.1.5)"]
-test = ["certifi (>=2024)", "cryptography-vectors (==46.0.4)", "pretend (>=0.7)", "pytest (>=7.4.0)", "pytest-benchmark (>=4.0)", "pytest-cov (>=2.10.1)", "pytest-xdist (>=3.5.0)"]
+test = ["certifi (>=2024)", "cryptography-vectors (==46.0.5)", "pretend (>=0.7)", "pytest (>=7.4.0)", "pytest-benchmark (>=4.0)", "pytest-cov (>=2.10.1)", "pytest-xdist (>=3.5.0)"]
 test-randomorder = ["pytest-randomly"]
 
 [[package]]
@@ -570,24 +570,25 @@ tests = ["coverage", "coveralls", "dill", "mock", "nose"]
 
 [[package]]
 name = "fastapi"
-version = "0.128.0"
+version = "0.128.7"
 description = "FastAPI framework, high performance, easy to learn, fast to code, ready for production"
 optional = false
 python-versions = ">=3.9"
 groups = ["main"]
 files = [
-    {file = "fastapi-0.128.0-py3-none-any.whl", hash = "sha256:aebd93f9716ee3b4f4fcfe13ffb7cf308d99c9f3ab5622d8877441072561582d"},
-    {file = "fastapi-0.128.0.tar.gz", hash = "sha256:1cc179e1cef10a6be60ffe429f79b829dce99d8de32d7acb7e6c8dfdf7f2645a"},
+    {file = "fastapi-0.128.7-py3-none-any.whl", hash = "sha256:6bd9bd31cb7047465f2d3fa3ba3f33b0870b17d4eaf7cdb36d1576ab060ad662"},
+    {file = "fastapi-0.128.7.tar.gz", hash = "sha256:783c273416995486c155ad2c0e2b45905dedfaf20b9ef8d9f6a9124670639a24"},
 ]
 
 [package.dependencies]
 annotated-doc = ">=0.0.2"
 pydantic = ">=2.7.0"
-starlette = ">=0.40.0,<0.51.0"
+starlette = ">=0.40.0,<1.0.0"
 typing-extensions = ">=4.8.0"
+typing-inspection = ">=0.4.2"
 
 [package.extras]
-all = ["email-validator (>=2.0.0)", "fastapi-cli[standard] (>=0.0.8)", "httpx (>=0.23.0,<1.0.0)", "itsdangerous (>=1.1.0)", "jinja2 (>=3.1.5)", "orjson (>=3.2.1)", "pydantic-extra-types (>=2.0.0)", "pydantic-settings (>=2.0.0)", "python-multipart (>=0.0.18)", "pyyaml (>=5.3.1)", "ujson (>=4.0.1,!=4.0.2,!=4.1.0,!=4.2.0,!=4.3.0,!=5.0.0,!=5.1.0)", "uvicorn[standard] (>=0.12.0)"]
+all = ["email-validator (>=2.0.0)", "fastapi-cli[standard] (>=0.0.8)", "httpx (>=0.23.0,<1.0.0)", "itsdangerous (>=1.1.0)", "jinja2 (>=3.1.5)", "orjson (>=3.9.3)", "pydantic-extra-types (>=2.0.0)", "pydantic-settings (>=2.0.0)", "python-multipart (>=0.0.18)", "pyyaml (>=5.3.1)", "ujson (>=5.8.0)", "uvicorn[standard] (>=0.12.0)"]
 standard = ["email-validator (>=2.0.0)", "fastapi-cli[standard] (>=0.0.8)", "httpx (>=0.23.0,<1.0.0)", "jinja2 (>=3.1.5)", "pydantic-extra-types (>=2.0.0)", "pydantic-settings (>=2.0.0)", "python-multipart (>=0.0.18)", "uvicorn[standard] (>=0.12.0)"]
 standard-no-fastapi-cloud-cli = ["email-validator (>=2.0.0)", "fastapi-cli[standard-no-fastapi-cloud-cli] (>=0.0.8)", "httpx (>=0.23.0,<1.0.0)", "jinja2 (>=3.1.5)", "pydantic-extra-types (>=2.0.0)", "pydantic-settings (>=2.0.0)", "python-multipart (>=0.0.18)", "uvicorn[standard] (>=0.12.0)"]
 
@@ -1062,14 +1063,14 @@ urllib3 = ">=1.26.0,<3"
 
 [[package]]
 name = "launchdarkly-server-sdk"
-version = "9.14.1"
+version = "9.15.0"
 description = "LaunchDarkly SDK for Python"
 optional = false
-python-versions = ">=3.9"
+python-versions = ">=3.10"
 groups = ["main"]
 files = [
-    {file = "launchdarkly_server_sdk-9.14.1-py3-none-any.whl", hash = "sha256:a9e2bd9ecdef845cd631ae0d4334a1115e5b44257c42eb2349492be4bac7815c"},
-    {file = "launchdarkly_server_sdk-9.14.1.tar.gz", hash = "sha256:1df44baf0a0efa74d8c1dad7a00592b98bce7d19edded7f770da8dbc49922213"},
+    {file = "launchdarkly_server_sdk-9.15.0-py3-none-any.whl", hash = "sha256:c267e29bfa3fb5e2a06a208448ada6ed5557a2924979b8d79c970b45d227c668"},
+    {file = "launchdarkly_server_sdk-9.15.0.tar.gz", hash = "sha256:f31441b74bc1a69c381db57c33116509e407a2612628ad6dff0a7dbb39d5020b"},
 ]
 
 [package.dependencies]
@@ -1478,14 +1479,14 @@ testing = ["coverage", "pytest", "pytest-benchmark"]
 
 [[package]]
 name = "postgrest"
-version = "2.27.2"
+version = "2.28.0"
 description = "PostgREST client for Python. This library provides an ORM interface to PostgREST."
 optional = false
 python-versions = ">=3.9"
 groups = ["main"]
 files = [
-    {file = "postgrest-2.27.2-py3-none-any.whl", hash = "sha256:1666fef3de05ca097a314433dd5ae2f2d71c613cb7b233d0f468c4ffe37277da"},
-    {file = "postgrest-2.27.2.tar.gz", hash = "sha256:55407d530b5af3d64e883a71fec1f345d369958f723ce4a8ab0b7d169e313242"},
+    {file = "postgrest-2.28.0-py3-none-any.whl", hash = "sha256:7bca2f24dd1a1bf8a3d586c7482aba6cd41662da6733045fad585b63b7f7df75"},
+    {file = "postgrest-2.28.0.tar.gz", hash = "sha256:c36b38646d25ea4255321d3d924ce70f8d20ec7799cb42c1221d6a818d4f6515"},
 ]
 
 [package.dependencies]
@@ -2248,14 +2249,14 @@ cli = ["click (>=5.0)"]
 
 [[package]]
 name = "realtime"
-version = "2.27.2"
+version = "2.28.0"
 description = ""
 optional = false
 python-versions = ">=3.9"
 groups = ["main"]
 files = [
-    {file = "realtime-2.27.2-py3-none-any.whl", hash = "sha256:34a9cbb26a274e707e8fc9e3ee0a66de944beac0fe604dc336d1e985db2c830f"},
-    {file = "realtime-2.27.2.tar.gz", hash = "sha256:b960a90294d2cea1b3f1275ecb89204304728e08fff1c393cc1b3150739556b3"},
+    {file = "realtime-2.28.0-py3-none-any.whl", hash = "sha256:db1bd59bab9b1fcc9f9d3b1a073bed35bf4994d720e6751f10031a58d57a3836"},
+    {file = "realtime-2.28.0.tar.gz", hash = "sha256:d18cedcebd6a8f22fcd509bc767f639761eb218b7b2b6f14fc4205b6259b50fc"},
 ]
 
 [package.dependencies]
@@ -2436,14 +2437,14 @@ full = ["httpx (>=0.27.0,<0.29.0)", "itsdangerous", "jinja2", "python-multipart
 
 [[package]]
 name = "storage3"
-version = "2.27.2"
+version = "2.28.0"
 description = "Supabase Storage client for Python."
 optional = false
 python-versions = ">=3.9"
 groups = ["main"]
 files = [
-    {file = "storage3-2.27.2-py3-none-any.whl", hash = "sha256:e6f16e7a260729e7b1f46e9bf61746805a02e30f5e419ee1291007c432e3ec63"},
-    {file = "storage3-2.27.2.tar.gz", hash = "sha256:cb4807b7f86b4bb1272ac6fdd2f3cfd8ba577297046fa5f88557425200275af5"},
+    {file = "storage3-2.28.0-py3-none-any.whl", hash = "sha256:ecb50efd2ac71dabbdf97e99ad346eafa630c4c627a8e5a138ceb5fbbadae716"},
+    {file = "storage3-2.28.0.tar.gz", hash = "sha256:bc1d008aff67de7a0f2bd867baee7aadbcdb6f78f5a310b4f7a38e8c13c19865"},
 ]
 
 [package.dependencies]
@@ -2487,35 +2488,35 @@ python-dateutil = ">=2.6.0"
 
 [[package]]
 name = "supabase"
-version = "2.27.2"
+version = "2.28.0"
 description = "Supabase client for Python."
 optional = false
 python-versions = ">=3.9"
 groups = ["main"]
 files = [
-    {file = "supabase-2.27.2-py3-none-any.whl", hash = "sha256:d4dce00b3a418ee578017ec577c0e5be47a9a636355009c76f20ed2faa15bc54"},
-    {file = "supabase-2.27.2.tar.gz", hash = "sha256:2aed40e4f3454438822442a1e94a47be6694c2c70392e7ae99b51a226d4293f7"},
+    {file = "supabase-2.28.0-py3-none-any.whl", hash = "sha256:42776971c7d0ccca16034df1ab96a31c50228eb1eb19da4249ad2f756fc20272"},
+    {file = "supabase-2.28.0.tar.gz", hash = "sha256:aea299aaab2a2eed3c57e0be7fc035c6807214194cce795a3575add20268ece1"},
 ]
 
 [package.dependencies]
 httpx = ">=0.26,<0.29"
-postgrest = "2.27.2"
-realtime = "2.27.2"
-storage3 = "2.27.2"
-supabase-auth = "2.27.2"
-supabase-functions = "2.27.2"
+postgrest = "2.28.0"
+realtime = "2.28.0"
+storage3 = "2.28.0"
+supabase-auth = "2.28.0"
+supabase-functions = "2.28.0"
 yarl = ">=1.22.0"
 
 [[package]]
 name = "supabase-auth"
-version = "2.27.2"
+version = "2.28.0"
 description = "Python Client Library for Supabase Auth"
 optional = false
 python-versions = ">=3.9"
 groups = ["main"]
 files = [
-    {file = "supabase_auth-2.27.2-py3-none-any.whl", hash = "sha256:78ec25b11314d0a9527a7205f3b1c72560dccdc11b38392f80297ef98664ee91"},
-    {file = "supabase_auth-2.27.2.tar.gz", hash = "sha256:0f5bcc79b3677cb42e9d321f3c559070cfa40d6a29a67672cc8382fb7dc2fe97"},
+    {file = "supabase_auth-2.28.0-py3-none-any.whl", hash = "sha256:2ac85026cc285054c7fa6d41924f3a333e9ec298c013e5b5e1754039ba7caec9"},
+    {file = "supabase_auth-2.28.0.tar.gz", hash = "sha256:2bb8f18ff39934e44b28f10918db965659f3735cd6fbfcc022fe0b82dbf8233e"},
 ]
 
 [package.dependencies]
@@ -2525,14 +2526,14 @@ pyjwt = {version = ">=2.10.1", extras = ["crypto"]}
 
 [[package]]
 name = "supabase-functions"
-version = "2.27.2"
+version = "2.28.0"
 description = "Library for Supabase Functions"
 optional = false
 python-versions = ">=3.9"
 groups = ["main"]
 files = [
-    {file = "supabase_functions-2.27.2-py3-none-any.whl", hash = "sha256:db480efc669d0bca07605b9b6f167312af43121adcc842a111f79bea416ef754"},
-    {file = "supabase_functions-2.27.2.tar.gz", hash = "sha256:d0c8266207a94371cb3fd35ad3c7f025b78a97cf026861e04ccd35ac1775f80b"},
+    {file = "supabase_functions-2.28.0-py3-none-any.whl", hash = "sha256:30bf2d586f8df285faf0621bb5d5bb3ec3157234fc820553ca156f009475e4ae"},
+    {file = "supabase_functions-2.28.0.tar.gz", hash = "sha256:db3dddfc37aca5858819eb461130968473bd8c75bd284581013958526dac718b"},
 ]
 
 [package.dependencies]
@@ -2911,4 +2912,4 @@ type = ["pytest-mypy"]
 [metadata]
 lock-version = "2.1"
 python-versions = ">=3.10,<4.0"
-content-hash = "40eae94995dc0a388fa832ed4af9b6137f28d5b5ced3aaea70d5f91d4d9a179d"
+content-hash = "9619cae908ad38fa2c48016a58bcf4241f6f5793aa0e6cc140276e91c433cbbb"

autogpt_platform/autogpt_libs/pyproject.toml

@@ -11,14 +11,14 @@ python = ">=3.10,<4.0"
 colorama = "^0.4.6"
 cryptography = "^46.0"
 expiringdict = "^1.2.2"
-fastapi = "^0.128.0"
+fastapi = "^0.128.7"
 google-cloud-logging = "^3.13.0"
-launchdarkly-server-sdk = "^9.14.1"
+launchdarkly-server-sdk = "^9.15.0"
 pydantic = "^2.12.5"
 pydantic-settings = "^2.12.0"
 pyjwt = { version = "^2.11.0", extras = ["crypto"] }
 redis = "^6.2.0"
-supabase = "^2.27.2"
+supabase = "^2.28.0"
 uvicorn = "^0.40.0"
 
 [tool.poetry.group.dev.dependencies]

autogpt_platform/backend/.env.default

@@ -104,6 +104,12 @@ TWITTER_CLIENT_SECRET=
 # Make a new workspace for your OAuth APP -- trust me
 # https://linear.app/settings/api/applications/new
 # Callback URL: http://localhost:3000/auth/integrations/oauth_callback
+LINEAR_API_KEY=
+# Linear project and team IDs for the feature request tracker.
+# Find these in your Linear workspace URL: linear.app/<workspace>/project/<project-id>
+# and in team settings. Used by the chat copilot to file and search feature requests.
+LINEAR_FEATURE_REQUEST_PROJECT_ID=
+LINEAR_FEATURE_REQUEST_TEAM_ID=
 LINEAR_CLIENT_ID=
 LINEAR_CLIENT_SECRET=
 

autogpt_platform/backend/Dockerfile

@@ -1,3 +1,5 @@
+# ============================ DEPENDENCY BUILDER ============================ #
+
 FROM debian:13-slim AS builder
 
 # Set environment variables
@@ -51,7 +53,9 @@ COPY autogpt_platform/backend/backend/data/partial_types.py ./backend/data/parti
 COPY autogpt_platform/backend/gen_prisma_types_stub.py ./
 RUN poetry run prisma generate && poetry run gen-prisma-stub
 
-FROM debian:13-slim AS server_dependencies
+# ============================== BACKEND SERVER ============================== #
+
+FROM debian:13-slim AS server
 
 WORKDIR /app
 
@@ -62,16 +66,21 @@ ENV POETRY_HOME=/opt/poetry \
     DEBIAN_FRONTEND=noninteractive
 ENV PATH=/opt/poetry/bin:$PATH
 
-# Install Python, FFmpeg, and ImageMagick (required for video processing blocks)
-RUN apt-get update && apt-get install -y \
+# Install Python, FFmpeg, ImageMagick, and CLI tools for agent use.
+# bubblewrap provides OS-level sandbox (whitelist-only FS + no network)
+# for the bash_exec MCP tool.
+# Using --no-install-recommends saves ~650MB by skipping unnecessary deps like llvm, mesa, etc.
+RUN apt-get update && apt-get install -y --no-install-recommends \
     python3.13 \
     python3-pip \
     ffmpeg \
     imagemagick \
+    jq \
+    ripgrep \
+    tree \
+    bubblewrap \
     && rm -rf /var/lib/apt/lists/*
 
-# Copy only necessary files from builder
-COPY --from=builder /app /app
 COPY --from=builder /usr/local/lib/python3* /usr/local/lib/python3*
 COPY --from=builder /usr/local/bin/poetry /usr/local/bin/poetry
 # Copy Node.js installation for Prisma
@@ -81,30 +90,54 @@ COPY --from=builder /usr/bin/npm /usr/bin/npm
 COPY --from=builder /usr/bin/npx /usr/bin/npx
 COPY --from=builder /root/.cache/prisma-python/binaries /root/.cache/prisma-python/binaries
 
-ENV PATH="/app/autogpt_platform/backend/.venv/bin:$PATH"
-
-RUN mkdir -p /app/autogpt_platform/autogpt_libs
-RUN mkdir -p /app/autogpt_platform/backend
-
-COPY autogpt_platform/autogpt_libs /app/autogpt_platform/autogpt_libs
-
-COPY autogpt_platform/backend/poetry.lock autogpt_platform/backend/pyproject.toml /app/autogpt_platform/backend/
-
 WORKDIR /app/autogpt_platform/backend
 
-FROM server_dependencies AS migrate
+# Copy only the .venv from builder (not the entire /app directory)
+# The .venv includes the generated Prisma client
+COPY --from=builder /app/autogpt_platform/backend/.venv ./.venv
+ENV PATH="/app/autogpt_platform/backend/.venv/bin:$PATH"
 
-# Migration stage only needs schema and migrations - much lighter than full backend
-COPY autogpt_platform/backend/schema.prisma /app/autogpt_platform/backend/
-COPY autogpt_platform/backend/backend/data/partial_types.py /app/autogpt_platform/backend/backend/data/partial_types.py
-COPY autogpt_platform/backend/migrations /app/autogpt_platform/backend/migrations
+# Copy dependency files + autogpt_libs (path dependency)
+COPY autogpt_platform/autogpt_libs /app/autogpt_platform/autogpt_libs
+COPY autogpt_platform/backend/poetry.lock autogpt_platform/backend/pyproject.toml ./
 
-FROM server_dependencies AS server
-
-COPY autogpt_platform/backend /app/autogpt_platform/backend
+# Copy backend code + docs (for Copilot docs search)
+COPY autogpt_platform/backend ./
 COPY docs /app/docs
 RUN poetry install --no-ansi --only-root
 
 ENV PORT=8000
 
 CMD ["poetry", "run", "rest"]
+
+# =============================== DB MIGRATOR =============================== #
+
+# Lightweight migrate stage - only needs Prisma CLI, not full Python environment
+FROM debian:13-slim AS migrate
+
+WORKDIR /app/autogpt_platform/backend
+
+ENV DEBIAN_FRONTEND=noninteractive
+
+# Install only what's needed for prisma migrate: Node.js and minimal Python for prisma-python
+RUN apt-get update && apt-get install -y --no-install-recommends \
+    python3.13 \
+    python3-pip \
+    ca-certificates \
+    && rm -rf /var/lib/apt/lists/*
+
+# Copy Node.js from builder (needed for Prisma CLI)
+COPY --from=builder /usr/bin/node /usr/bin/node
+COPY --from=builder /usr/lib/node_modules /usr/lib/node_modules
+COPY --from=builder /usr/bin/npm /usr/bin/npm
+
+# Copy Prisma binaries
+COPY --from=builder /root/.cache/prisma-python/binaries /root/.cache/prisma-python/binaries
+
+# Install prisma-client-py directly (much smaller than copying full venv)
+RUN pip3 install prisma>=0.15.0 --break-system-packages
+
+COPY autogpt_platform/backend/schema.prisma ./
+COPY autogpt_platform/backend/backend/data/partial_types.py ./backend/data/partial_types.py
+COPY autogpt_platform/backend/gen_prisma_types_stub.py ./
+COPY autogpt_platform/backend/migrations ./migrations

autogpt_platform/backend/backend/api/conftest.py

@@ -1,4 +1,9 @@
-"""Common test fixtures for server tests."""
+"""Common test fixtures for server tests.
+
+Note: Common fixtures like test_user_id, admin_user_id, target_user_id,
+setup_test_user, and setup_admin_user are defined in the parent conftest.py
+(backend/conftest.py) and are available here automatically.
+"""
 
 import pytest
 from pytest_snapshot.plugin import Snapshot
@@ -11,54 +16,6 @@ def configured_snapshot(snapshot: Snapshot) -> Snapshot:
     return snapshot
 
 
-@pytest.fixture
-def test_user_id() -> str:
-    """Test user ID fixture."""
-    return "3e53486c-cf57-477e-ba2a-cb02dc828e1a"
-
-
-@pytest.fixture
-def admin_user_id() -> str:
-    """Admin user ID fixture."""
-    return "4e53486c-cf57-477e-ba2a-cb02dc828e1b"
-
-
-@pytest.fixture
-def target_user_id() -> str:
-    """Target user ID fixture."""
-    return "5e53486c-cf57-477e-ba2a-cb02dc828e1c"
-
-
-@pytest.fixture
-async def setup_test_user(test_user_id):
-    """Create test user in database before tests."""
-    from backend.data.user import get_or_create_user
-
-    # Create the test user in the database using JWT token format
-    user_data = {
-        "sub": test_user_id,
-        "email": "test@example.com",
-        "user_metadata": {"name": "Test User"},
-    }
-    await get_or_create_user(user_data)
-    return test_user_id
-
-
-@pytest.fixture
-async def setup_admin_user(admin_user_id):
-    """Create admin user in database before tests."""
-    from backend.data.user import get_or_create_user
-
-    # Create the admin user in the database using JWT token format
-    user_data = {
-        "sub": admin_user_id,
-        "email": "test-admin@example.com",
-        "user_metadata": {"name": "Test Admin"},
-    }
-    await get_or_create_user(user_data)
-    return admin_user_id
-
-
 @pytest.fixture
 def mock_jwt_user(test_user_id):
     """Provide mock JWT payload for regular user testing."""

autogpt_platform/backend/backend/api/external/v1/tools.py

@@ -15,9 +15,9 @@ from prisma.enums import APIKeyPermission
 from pydantic import BaseModel, Field
 
 from backend.api.external.middleware import require_permission
-from backend.api.features.chat.model import ChatSession
-from backend.api.features.chat.tools import find_agent_tool, run_agent_tool
-from backend.api.features.chat.tools.models import ToolResponseBase
+from backend.copilot.model import ChatSession
+from backend.copilot.tools import find_agent_tool, run_agent_tool
+from backend.copilot.tools.models import ToolResponseBase
 from backend.data.auth.base import APIAuthorizationInfo
 
 logger = logging.getLogger(__name__)

autogpt_platform/backend/backend/api/features/chat/routes.py

@@ -1,5 +1,6 @@
 """Chat API routes for chat session management and streaming via SSE."""
 
+import asyncio
 import logging
 import uuid as uuid_module
 from collections.abc import AsyncGenerator
@@ -10,20 +11,30 @@ from fastapi import APIRouter, Depends, Header, HTTPException, Query, Response,
 from fastapi.responses import StreamingResponse
 from pydantic import BaseModel
 
-from backend.util.exceptions import NotFoundError
-
-from . import service as chat_service
-from . import stream_registry
-from .completion_handler import process_operation_failure, process_operation_success
-from .config import ChatConfig
-from .model import ChatSession, create_chat_session, get_chat_session, get_user_sessions
-from .response_model import StreamFinish, StreamHeartbeat
-from .tools.models import (
+from backend.copilot import service as chat_service
+from backend.copilot import stream_registry
+from backend.copilot.completion_handler import (
+    process_operation_failure,
+    process_operation_success,
+)
+from backend.copilot.config import ChatConfig
+from backend.copilot.executor.utils import enqueue_copilot_task
+from backend.copilot.model import (
+    ChatMessage,
+    ChatSession,
+    append_and_save_message,
+    create_chat_session,
+    get_chat_session,
+    get_user_sessions,
+)
+from backend.copilot.response_model import StreamError, StreamFinish, StreamHeartbeat
+from backend.copilot.tools.models import (
     AgentDetailsResponse,
     AgentOutputResponse,
     AgentPreviewResponse,
     AgentSavedResponse,
     AgentsFoundResponse,
+    BlockDetailsResponse,
     BlockListResponse,
     BlockOutputResponse,
     ClarificationNeededResponse,
@@ -40,6 +51,8 @@ from .tools.models import (
     SetupRequirementsResponse,
     UnderstandingUpdatedResponse,
 )
+from backend.copilot.tracking import track_user_message
+from backend.util.exceptions import NotFoundError
 
 config = ChatConfig()
 
@@ -231,6 +244,10 @@ async def get_session(
     active_task, last_message_id = await stream_registry.get_active_task_for_session(
         session_id, user_id
     )
+    logger.info(
+        f"[GET_SESSION] session={session_id}, active_task={active_task is not None}, "
+        f"msg_count={len(messages)}, last_role={messages[-1].get('role') if messages else 'none'}"
+    )
     if active_task:
         # Filter out the in-progress assistant message from the session response.
         # The client will receive the complete assistant response through the SSE
@@ -300,10 +317,9 @@ async def stream_chat_post(
         f"user={user_id}, message_len={len(request.message)}",
         extra={"json_fields": log_meta},
     )
-
-    session = await _validate_and_get_session(session_id, user_id)
+    await _validate_and_get_session(session_id, user_id)
     logger.info(
-        f"[TIMING] session validated in {(time.perf_counter() - stream_start_time)*1000:.1f}ms",
+        f"[TIMING] session validated in {(time.perf_counter() - stream_start_time) * 1000:.1f}ms",
         extra={
             "json_fields": {
                 **log_meta,
@@ -312,6 +328,25 @@ async def stream_chat_post(
         },
     )
 
+    # Atomically append user message to session BEFORE creating task to avoid
+    # race condition where GET_SESSION sees task as "running" but message isn't
+    # saved yet.  append_and_save_message re-fetches inside a lock to prevent
+    # message loss from concurrent requests.
+    if request.message:
+        message = ChatMessage(
+            role="user" if request.is_user_message else "assistant",
+            content=request.message,
+        )
+        if request.is_user_message:
+            track_user_message(
+                user_id=user_id,
+                session_id=session_id,
+                message_length=len(request.message),
+            )
+        logger.info(f"[STREAM] Saving user message to session {session_id}")
+        await append_and_save_message(session_id, message)
+        logger.info(f"[STREAM] User message saved for session {session_id}")
+
     # Create a task in the stream registry for reconnection support
     task_id = str(uuid_module.uuid4())
     operation_id = str(uuid_module.uuid4())
@@ -327,7 +362,7 @@ async def stream_chat_post(
         operation_id=operation_id,
     )
     logger.info(
-        f"[TIMING] create_task completed in {(time.perf_counter() - task_create_start)*1000:.1f}ms",
+        f"[TIMING] create_task completed in {(time.perf_counter() - task_create_start) * 1000:.1f}ms",
         extra={
             "json_fields": {
                 **log_meta,
@@ -336,82 +371,19 @@ async def stream_chat_post(
         },
     )
 
-    # Background task that runs the AI generation independently of SSE connection
-    async def run_ai_generation():
-        import time as time_module
+    await enqueue_copilot_task(
+        task_id=task_id,
+        session_id=session_id,
+        user_id=user_id,
+        operation_id=operation_id,
+        message=request.message,
+        is_user_message=request.is_user_message,
+        context=request.context,
+    )
 
-        gen_start_time = time_module.perf_counter()
-        logger.info(
-            f"[TIMING] run_ai_generation STARTED, task={task_id}, session={session_id}, user={user_id}",
-            extra={"json_fields": log_meta},
-        )
-        first_chunk_time, ttfc = None, None
-        chunk_count = 0
-        try:
-            async for chunk in chat_service.stream_chat_completion(
-                session_id,
-                request.message,
-                is_user_message=request.is_user_message,
-                user_id=user_id,
-                session=session,  # Pass pre-fetched session to avoid double-fetch
-                context=request.context,
-                _task_id=task_id,  # Pass task_id so service emits start with taskId for reconnection
-            ):
-                chunk_count += 1
-                if first_chunk_time is None:
-                    first_chunk_time = time_module.perf_counter()
-                    ttfc = first_chunk_time - gen_start_time
-                    logger.info(
-                        f"[TIMING] FIRST AI CHUNK at {ttfc:.2f}s, type={type(chunk).__name__}",
-                        extra={
-                            "json_fields": {
-                                **log_meta,
-                                "chunk_type": type(chunk).__name__,
-                                "time_to_first_chunk_ms": ttfc * 1000,
-                            }
-                        },
-                    )
-                # Write to Redis (subscribers will receive via XREAD)
-                await stream_registry.publish_chunk(task_id, chunk)
-
-            gen_end_time = time_module.perf_counter()
-            total_time = (gen_end_time - gen_start_time) * 1000
-            logger.info(
-                f"[TIMING] run_ai_generation FINISHED in {total_time/1000:.1f}s; "
-                f"task={task_id}, session={session_id}, "
-                f"ttfc={ttfc or -1:.2f}s, n_chunks={chunk_count}",
-                extra={
-                    "json_fields": {
-                        **log_meta,
-                        "total_time_ms": total_time,
-                        "time_to_first_chunk_ms": (
-                            ttfc * 1000 if ttfc is not None else None
-                        ),
-                        "n_chunks": chunk_count,
-                    }
-                },
-            )
-            await stream_registry.mark_task_completed(task_id, "completed")
-        except Exception as e:
-            elapsed = time_module.perf_counter() - gen_start_time
-            logger.error(
-                f"[TIMING] run_ai_generation ERROR after {elapsed:.2f}s: {e}",
-                extra={
-                    "json_fields": {
-                        **log_meta,
-                        "elapsed_ms": elapsed * 1000,
-                        "error": str(e),
-                    }
-                },
-            )
-            await stream_registry.mark_task_completed(task_id, "failed")
-
-    # Start the AI generation in a background task
-    bg_task = asyncio.create_task(run_ai_generation())
-    await stream_registry.set_task_asyncio_task(task_id, bg_task)
     setup_time = (time.perf_counter() - stream_start_time) * 1000
     logger.info(
-        f"[TIMING] Background task started, setup={setup_time:.1f}ms",
+        f"[TIMING] Task enqueued to RabbitMQ, setup={setup_time:.1f}ms",
         extra={"json_fields": {**log_meta, "setup_time_ms": setup_time}},
     )
 
@@ -506,8 +478,14 @@ async def stream_chat_post(
                     "json_fields": {**log_meta, "elapsed_ms": elapsed, "error": str(e)}
                 },
             )
+            # Surface error to frontend so it doesn't appear stuck
+            yield StreamError(
+                errorText="An error occurred. Please try again.",
+                code="stream_error",
+            ).to_sse()
+            yield StreamFinish().to_sse()
         finally:
-            # Unsubscribe when client disconnects or stream ends to prevent resource leak
+            # Unsubscribe when client disconnects or stream ends
             if subscriber_queue is not None:
                 try:
                     await stream_registry.unsubscribe_from_task(
@@ -751,8 +729,6 @@ async def stream_task(
         )
 
     async def event_generator() -> AsyncGenerator[str, None]:
-        import asyncio
-
         heartbeat_interval = 15.0  # Send heartbeat every 15 seconds
         try:
             while True:
@@ -971,6 +947,7 @@ ToolResponseUnion = (
     | AgentSavedResponse
     | ClarificationNeededResponse
     | BlockListResponse
+    | BlockDetailsResponse
     | BlockOutputResponse
     | DocSearchResultsResponse
     | DocPageResponse

autogpt_platform/backend/backend/api/features/chat/service_test.py

@@ -1,82 +0,0 @@
-import logging
-from os import getenv
-
-import pytest
-
-from . import service as chat_service
-from .model import create_chat_session, get_chat_session, upsert_chat_session
-from .response_model import (
-    StreamError,
-    StreamFinish,
-    StreamTextDelta,
-    StreamToolOutputAvailable,
-)
-
-logger = logging.getLogger(__name__)
-
-
-@pytest.mark.asyncio(loop_scope="session")
-async def test_stream_chat_completion(setup_test_user, test_user_id):
-    """
-    Test the stream_chat_completion function.
-    """
-    api_key: str | None = getenv("OPEN_ROUTER_API_KEY")
-    if not api_key:
-        return pytest.skip("OPEN_ROUTER_API_KEY is not set, skipping test")
-
-    session = await create_chat_session(test_user_id)
-
-    has_errors = False
-    has_ended = False
-    assistant_message = ""
-    async for chunk in chat_service.stream_chat_completion(
-        session.session_id, "Hello, how are you?", user_id=session.user_id
-    ):
-        logger.info(chunk)
-        if isinstance(chunk, StreamError):
-            has_errors = True
-        if isinstance(chunk, StreamTextDelta):
-            assistant_message += chunk.delta
-        if isinstance(chunk, StreamFinish):
-            has_ended = True
-
-    assert has_ended, "Chat completion did not end"
-    assert not has_errors, "Error occurred while streaming chat completion"
-    assert assistant_message, "Assistant message is empty"
-
-
-@pytest.mark.asyncio(loop_scope="session")
-async def test_stream_chat_completion_with_tool_calls(setup_test_user, test_user_id):
-    """
-    Test the stream_chat_completion function.
-    """
-    api_key: str | None = getenv("OPEN_ROUTER_API_KEY")
-    if not api_key:
-        return pytest.skip("OPEN_ROUTER_API_KEY is not set, skipping test")
-
-    session = await create_chat_session(test_user_id)
-    session = await upsert_chat_session(session)
-
-    has_errors = False
-    has_ended = False
-    had_tool_calls = False
-    async for chunk in chat_service.stream_chat_completion(
-        session.session_id,
-        "Please find me an agent that can help me with my business. Use the query 'moneny printing agent'",
-        user_id=session.user_id,
-    ):
-        logger.info(chunk)
-        if isinstance(chunk, StreamError):
-            has_errors = True
-
-        if isinstance(chunk, StreamFinish):
-            has_ended = True
-        if isinstance(chunk, StreamToolOutputAvailable):
-            had_tool_calls = True
-
-    assert has_ended, "Chat completion did not end"
-    assert not has_errors, "Error occurred while streaming chat completion"
-    assert had_tool_calls, "Tool calls did not occur"
-    session = await get_chat_session(session.session_id)
-    assert session, "Session not found"
-    assert session.usage, "Usage is empty"

autogpt_platform/backend/backend/api/features/chat/tools/find_block_test.py

@@ -1,139 +0,0 @@
-"""Tests for block filtering in FindBlockTool."""
-
-from unittest.mock import AsyncMock, MagicMock, patch
-
-import pytest
-
-from backend.api.features.chat.tools.find_block import (
-    COPILOT_EXCLUDED_BLOCK_IDS,
-    COPILOT_EXCLUDED_BLOCK_TYPES,
-    FindBlockTool,
-)
-from backend.api.features.chat.tools.models import BlockListResponse
-from backend.blocks._base import BlockType
-
-from ._test_data import make_session
-
-_TEST_USER_ID = "test-user-find-block"
-
-
-def make_mock_block(
-    block_id: str, name: str, block_type: BlockType, disabled: bool = False
-):
-    """Create a mock block for testing."""
-    mock = MagicMock()
-    mock.id = block_id
-    mock.name = name
-    mock.description = f"{name} description"
-    mock.block_type = block_type
-    mock.disabled = disabled
-    mock.input_schema = MagicMock()
-    mock.input_schema.jsonschema.return_value = {"properties": {}, "required": []}
-    mock.input_schema.get_credentials_fields.return_value = {}
-    mock.output_schema = MagicMock()
-    mock.output_schema.jsonschema.return_value = {}
-    mock.categories = []
-    return mock
-
-
-class TestFindBlockFiltering:
-    """Tests for block filtering in FindBlockTool."""
-
-    def test_excluded_block_types_contains_expected_types(self):
-        """Verify COPILOT_EXCLUDED_BLOCK_TYPES contains all graph-only types."""
-        assert BlockType.INPUT in COPILOT_EXCLUDED_BLOCK_TYPES
-        assert BlockType.OUTPUT in COPILOT_EXCLUDED_BLOCK_TYPES
-        assert BlockType.WEBHOOK in COPILOT_EXCLUDED_BLOCK_TYPES
-        assert BlockType.WEBHOOK_MANUAL in COPILOT_EXCLUDED_BLOCK_TYPES
-        assert BlockType.NOTE in COPILOT_EXCLUDED_BLOCK_TYPES
-        assert BlockType.HUMAN_IN_THE_LOOP in COPILOT_EXCLUDED_BLOCK_TYPES
-        assert BlockType.AGENT in COPILOT_EXCLUDED_BLOCK_TYPES
-
-    def test_excluded_block_ids_contains_smart_decision_maker(self):
-        """Verify SmartDecisionMakerBlock is in COPILOT_EXCLUDED_BLOCK_IDS."""
-        assert "3b191d9f-356f-482d-8238-ba04b6d18381" in COPILOT_EXCLUDED_BLOCK_IDS
-
-    @pytest.mark.asyncio(loop_scope="session")
-    async def test_excluded_block_type_filtered_from_results(self):
-        """Verify blocks with excluded BlockTypes are filtered from search results."""
-        session = make_session(user_id=_TEST_USER_ID)
-
-        # Mock search returns an INPUT block (excluded) and a STANDARD block (included)
-        search_results = [
-            {"content_id": "input-block-id", "score": 0.9},
-            {"content_id": "standard-block-id", "score": 0.8},
-        ]
-
-        input_block = make_mock_block("input-block-id", "Input Block", BlockType.INPUT)
-        standard_block = make_mock_block(
-            "standard-block-id", "HTTP Request", BlockType.STANDARD
-        )
-
-        def mock_get_block(block_id):
-            return {
-                "input-block-id": input_block,
-                "standard-block-id": standard_block,
-            }.get(block_id)
-
-        with patch(
-            "backend.api.features.chat.tools.find_block.unified_hybrid_search",
-            new_callable=AsyncMock,
-            return_value=(search_results, 2),
-        ):
-            with patch(
-                "backend.api.features.chat.tools.find_block.get_block",
-                side_effect=mock_get_block,
-            ):
-                tool = FindBlockTool()
-                response = await tool._execute(
-                    user_id=_TEST_USER_ID, session=session, query="test"
-                )
-
-        # Should only return the standard block, not the INPUT block
-        assert isinstance(response, BlockListResponse)
-        assert len(response.blocks) == 1
-        assert response.blocks[0].id == "standard-block-id"
-
-    @pytest.mark.asyncio(loop_scope="session")
-    async def test_excluded_block_id_filtered_from_results(self):
-        """Verify SmartDecisionMakerBlock is filtered from search results."""
-        session = make_session(user_id=_TEST_USER_ID)
-
-        smart_decision_id = "3b191d9f-356f-482d-8238-ba04b6d18381"
-        search_results = [
-            {"content_id": smart_decision_id, "score": 0.9},
-            {"content_id": "normal-block-id", "score": 0.8},
-        ]
-
-        # SmartDecisionMakerBlock has STANDARD type but is excluded by ID
-        smart_block = make_mock_block(
-            smart_decision_id, "Smart Decision Maker", BlockType.STANDARD
-        )
-        normal_block = make_mock_block(
-            "normal-block-id", "Normal Block", BlockType.STANDARD
-        )
-
-        def mock_get_block(block_id):
-            return {
-                smart_decision_id: smart_block,
-                "normal-block-id": normal_block,
-            }.get(block_id)
-
-        with patch(
-            "backend.api.features.chat.tools.find_block.unified_hybrid_search",
-            new_callable=AsyncMock,
-            return_value=(search_results, 2),
-        ):
-            with patch(
-                "backend.api.features.chat.tools.find_block.get_block",
-                side_effect=mock_get_block,
-            ):
-                tool = FindBlockTool()
-                response = await tool._execute(
-                    user_id=_TEST_USER_ID, session=session, query="decision"
-                )
-
-        # Should only return normal block, not SmartDecisionMakerBlock
-        assert isinstance(response, BlockListResponse)
-        assert len(response.blocks) == 1
-        assert response.blocks[0].id == "normal-block-id"

autogpt_platform/backend/backend/api/features/chat/tools/run_block_test.py

@@ -1,106 +0,0 @@
-"""Tests for block execution guards in RunBlockTool."""
-
-from unittest.mock import MagicMock, patch
-
-import pytest
-
-from backend.api.features.chat.tools.models import ErrorResponse
-from backend.api.features.chat.tools.run_block import RunBlockTool
-from backend.blocks._base import BlockType
-
-from ._test_data import make_session
-
-_TEST_USER_ID = "test-user-run-block"
-
-
-def make_mock_block(
-    block_id: str, name: str, block_type: BlockType, disabled: bool = False
-):
-    """Create a mock block for testing."""
-    mock = MagicMock()
-    mock.id = block_id
-    mock.name = name
-    mock.block_type = block_type
-    mock.disabled = disabled
-    mock.input_schema = MagicMock()
-    mock.input_schema.jsonschema.return_value = {"properties": {}, "required": []}
-    mock.input_schema.get_credentials_fields_info.return_value = []
-    return mock
-
-
-class TestRunBlockFiltering:
-    """Tests for block execution guards in RunBlockTool."""
-
-    @pytest.mark.asyncio(loop_scope="session")
-    async def test_excluded_block_type_returns_error(self):
-        """Attempting to execute a block with excluded BlockType returns error."""
-        session = make_session(user_id=_TEST_USER_ID)
-
-        input_block = make_mock_block("input-block-id", "Input Block", BlockType.INPUT)
-
-        with patch(
-            "backend.api.features.chat.tools.run_block.get_block",
-            return_value=input_block,
-        ):
-            tool = RunBlockTool()
-            response = await tool._execute(
-                user_id=_TEST_USER_ID,
-                session=session,
-                block_id="input-block-id",
-                input_data={},
-            )
-
-        assert isinstance(response, ErrorResponse)
-        assert "cannot be run directly in CoPilot" in response.message
-        assert "designed for use within graphs only" in response.message
-
-    @pytest.mark.asyncio(loop_scope="session")
-    async def test_excluded_block_id_returns_error(self):
-        """Attempting to execute SmartDecisionMakerBlock returns error."""
-        session = make_session(user_id=_TEST_USER_ID)
-
-        smart_decision_id = "3b191d9f-356f-482d-8238-ba04b6d18381"
-        smart_block = make_mock_block(
-            smart_decision_id, "Smart Decision Maker", BlockType.STANDARD
-        )
-
-        with patch(
-            "backend.api.features.chat.tools.run_block.get_block",
-            return_value=smart_block,
-        ):
-            tool = RunBlockTool()
-            response = await tool._execute(
-                user_id=_TEST_USER_ID,
-                session=session,
-                block_id=smart_decision_id,
-                input_data={},
-            )
-
-        assert isinstance(response, ErrorResponse)
-        assert "cannot be run directly in CoPilot" in response.message
-
-    @pytest.mark.asyncio(loop_scope="session")
-    async def test_non_excluded_block_passes_guard(self):
-        """Non-excluded blocks pass the filtering guard (may fail later for other reasons)."""
-        session = make_session(user_id=_TEST_USER_ID)
-
-        standard_block = make_mock_block(
-            "standard-id", "HTTP Request", BlockType.STANDARD
-        )
-
-        with patch(
-            "backend.api.features.chat.tools.run_block.get_block",
-            return_value=standard_block,
-        ):
-            tool = RunBlockTool()
-            response = await tool._execute(
-                user_id=_TEST_USER_ID,
-                session=session,
-                block_id="standard-id",
-                input_data={},
-            )
-
-        # Should NOT be an ErrorResponse about CoPilot exclusion
-        # (may be other errors like missing credentials, but not the exclusion guard)
-        if isinstance(response, ErrorResponse):
-            assert "cannot be run directly in CoPilot" not in response.message

autogpt_platform/backend/backend/api/features/integrations/router.py

@@ -1,7 +1,7 @@
 import asyncio
 import logging
 from datetime import datetime, timedelta, timezone
-from typing import TYPE_CHECKING, Annotated, List, Literal
+from typing import TYPE_CHECKING, Annotated, Any, List, Literal
 
 from autogpt_libs.auth import get_user_id
 from fastapi import (
@@ -14,7 +14,7 @@ from fastapi import (
     Security,
     status,
 )
-from pydantic import BaseModel, Field, SecretStr
+from pydantic import BaseModel, Field, SecretStr, model_validator
 from starlette.status import HTTP_500_INTERNAL_SERVER_ERROR, HTTP_502_BAD_GATEWAY
 
 from backend.api.features.library.db import set_preset_webhook, update_preset
@@ -39,7 +39,11 @@ from backend.data.onboarding import OnboardingStep, complete_onboarding_step
 from backend.data.user import get_user_integrations
 from backend.executor.utils import add_graph_execution
 from backend.integrations.ayrshare import AyrshareClient, SocialPlatform
-from backend.integrations.creds_manager import IntegrationCredentialsManager
+from backend.integrations.credentials_store import provider_matches
+from backend.integrations.creds_manager import (
+    IntegrationCredentialsManager,
+    create_mcp_oauth_handler,
+)
 from backend.integrations.oauth import CREDENTIALS_BY_PROVIDER, HANDLERS_BY_NAME
 from backend.integrations.providers import ProviderName
 from backend.integrations.webhooks import get_webhook_manager
@@ -102,9 +106,37 @@ class CredentialsMetaResponse(BaseModel):
     scopes: list[str] | None
     username: str | None
     host: str | None = Field(
-        default=None, description="Host pattern for host-scoped credentials"
+        default=None,
+        description="Host pattern for host-scoped or MCP server URL for MCP credentials",
     )
 
+    @model_validator(mode="before")
+    @classmethod
+    def _normalize_provider(cls, data: Any) -> Any:
+        """Fix ``ProviderName.X`` format from Python 3.13 ``str(Enum)`` bug."""
+        if isinstance(data, dict):
+            prov = data.get("provider", "")
+            if isinstance(prov, str) and prov.startswith("ProviderName."):
+                member = prov.removeprefix("ProviderName.")
+                try:
+                    data = {**data, "provider": ProviderName[member].value}
+                except KeyError:
+                    pass
+        return data
+
+    @staticmethod
+    def get_host(cred: Credentials) -> str | None:
+        """Extract host from credential: HostScoped host or MCP server URL."""
+        if isinstance(cred, HostScopedCredentials):
+            return cred.host
+        if isinstance(cred, OAuth2Credentials) and cred.provider in (
+            ProviderName.MCP,
+            ProviderName.MCP.value,
+            "ProviderName.MCP",
+        ):
+            return (cred.metadata or {}).get("mcp_server_url")
+        return None
+
 
 @router.post("/{provider}/callback", summary="Exchange OAuth code for tokens")
 async def callback(
@@ -179,9 +211,7 @@ async def callback(
         title=credentials.title,
         scopes=credentials.scopes,
         username=credentials.username,
-        host=(
-            credentials.host if isinstance(credentials, HostScopedCredentials) else None
-        ),
+        host=(CredentialsMetaResponse.get_host(credentials)),
     )
 
 
@@ -199,7 +229,7 @@ async def list_credentials(
             title=cred.title,
             scopes=cred.scopes if isinstance(cred, OAuth2Credentials) else None,
             username=cred.username if isinstance(cred, OAuth2Credentials) else None,
-            host=cred.host if isinstance(cred, HostScopedCredentials) else None,
+            host=CredentialsMetaResponse.get_host(cred),
         )
         for cred in credentials
     ]
@@ -222,7 +252,7 @@ async def list_credentials_by_provider(
             title=cred.title,
             scopes=cred.scopes if isinstance(cred, OAuth2Credentials) else None,
             username=cred.username if isinstance(cred, OAuth2Credentials) else None,
-            host=cred.host if isinstance(cred, HostScopedCredentials) else None,
+            host=CredentialsMetaResponse.get_host(cred),
         )
         for cred in credentials
     ]
@@ -322,7 +352,11 @@ async def delete_credentials(
 
     tokens_revoked = None
     if isinstance(creds, OAuth2Credentials):
-        handler = _get_provider_oauth_handler(request, provider)
+        if provider_matches(provider.value, ProviderName.MCP.value):
+            # MCP uses dynamic per-server OAuth — create handler from metadata
+            handler = create_mcp_oauth_handler(creds)
+        else:
+            handler = _get_provider_oauth_handler(request, provider)
         tokens_revoked = await handler.revoke_tokens(creds)
 
     return CredentialsDeletionResponse(revoked=tokens_revoked)

autogpt_platform/backend/backend/api/features/library/db.py

@@ -7,7 +7,6 @@ import prisma.errors
 import prisma.models
 import prisma.types
 
-from backend.api.features.library.exceptions import FolderValidationError
 import backend.api.features.store.exceptions as store_exceptions
 import backend.api.features.store.image_gen as store_image_gen
 import backend.api.features.store.media as store_media
@@ -43,8 +42,6 @@ async def list_library_agents(
     page: int = 1,
     page_size: int = 50,
     include_executions: bool = False,
-    folder_id: Optional[str] = None,
-    include_root_only: bool = False,
 ) -> library_model.LibraryAgentResponse:
     """
     Retrieves a paginated list of LibraryAgent records for a given user.
@@ -55,8 +52,6 @@ async def list_library_agents(
         sort_by: Sorting field (createdAt, updatedAt, isFavorite, isCreatedByUser).
         page: Current page (1-indexed).
         page_size: Number of items per page.
-        folder_id: Filter by folder ID. If provided, only returns agents in this folder.
-        include_root_only: If True, only returns agents without a folder (root-level).
         include_executions: Whether to include execution data for status calculation.
             Defaults to False for performance (UI fetches status separately).
             Set to True when accurate status/metrics are needed (e.g., agent generator).
@@ -87,13 +82,6 @@ async def list_library_agents(
         "isArchived": False,
     }
 
-    # Apply folder filter (skip when searching — search spans all folders)
-    if folder_id is not None and not search_term:
-        where_clause["folderId"] = folder_id
-    elif include_root_only and not search_term:
-        where_clause["folderId"] = None
-
-    # Build search filter if applicable
     if search_term:
         where_clause["OR"] = [
             {
@@ -646,7 +634,6 @@ async def update_library_agent(
     is_archived: Optional[bool] = None,
     is_deleted: Optional[Literal[False]] = None,
     settings: Optional[GraphSettings] = None,
-    folder_id: Optional[str] = None,
 ) -> library_model.LibraryAgent:
     """
     Updates the specified LibraryAgent record.
@@ -659,7 +646,6 @@ async def update_library_agent(
         is_favorite: Whether this agent is marked as a favorite.
         is_archived: Whether this agent is archived.
         settings: User-specific settings for this library agent.
-        folder_id: Folder ID to move agent to (empty string "" for root, None to skip).
 
     Returns:
         The updated LibraryAgent.
@@ -687,10 +673,13 @@ async def update_library_agent(
             )
         update_fields["isDeleted"] = is_deleted
     if settings is not None:
-        update_fields["settings"] = SafeJson(settings.model_dump())
-    if folder_id is not None:
-        # Empty string means "move to root" (no folder)
-        update_fields["folderId"] = None if folder_id == "" else folder_id
+        existing_agent = await get_library_agent(id=library_agent_id, user_id=user_id)
+        current_settings_dict = (
+            existing_agent.settings.model_dump() if existing_agent.settings else {}
+        )
+        new_settings = settings.model_dump(exclude_unset=True)
+        merged_settings = {**current_settings_dict, **new_settings}
+        update_fields["settings"] = SafeJson(merged_settings)
 
     try:
         # If graph_version is provided, update to that specific version
@@ -929,813 +918,6 @@ async def add_store_agent_to_library(
         raise DatabaseError("Failed to add agent to library") from e
 
 
-##############################################
-############ Folder DB Functions #############
-##############################################
-
-MAX_FOLDER_DEPTH = 5
-
-
-async def list_folders(
-    user_id: str,
-    parent_id: Optional[str] = None,
-    include_counts: bool = True,
-) -> list[library_model.LibraryFolder]:
-    """
-    Lists folders for a user, optionally filtered by parent.
-
-    Args:
-        user_id: The ID of the user.
-        parent_id: If provided, only returns folders with this parent.
-                   If None, returns root-level folders.
-        include_counts: Whether to include agent and subfolder counts.
-
-    Returns:
-        A list of LibraryFolder objects.
-    """
-    logger.debug(f"Listing folders for user #{user_id}, parent_id={parent_id}")
-
-    try:
-        where_clause: prisma.types.LibraryFolderWhereInput = {
-            "userId": user_id,
-            "isDeleted": False,
-            "parentId": parent_id,
-        }
-
-        folders = await prisma.models.LibraryFolder.prisma().find_many(
-            where=where_clause,
-            order={"createdAt": "asc"},
-            include=(
-                {
-                    "LibraryAgents": {"where": {"isDeleted": False}},
-                    "Children": {"where": {"isDeleted": False}},
-                }
-                if include_counts
-                else None
-            ),
-        )
-
-        result = []
-        for folder in folders:
-            agent_count = len(folder.LibraryAgents) if folder.LibraryAgents else 0
-            subfolder_count = len(folder.Children) if folder.Children else 0
-            result.append(
-                library_model.LibraryFolder.from_db(
-                    folder,
-                    agent_count=agent_count,
-                    subfolder_count=subfolder_count,
-                )
-            )
-
-        return result
-
-    except prisma.errors.PrismaError as e:
-        logger.error(f"Database error listing folders: {e}")
-        raise DatabaseError("Failed to list folders") from e
-
-
-async def get_folder_tree(
-    user_id: str,
-) -> list[library_model.LibraryFolderTree]:
-    """
-    Gets the full folder tree for a user.
-
-    Args:
-        user_id: The ID of the user.
-
-    Returns:
-        A list of LibraryFolderTree objects (root folders with nested children).
-    """
-    logger.debug(f"Getting folder tree for user #{user_id}")
-
-    try:
-        # Fetch all folders for the user
-        all_folders = await prisma.models.LibraryFolder.prisma().find_many(
-            where={
-                "userId": user_id,
-                "isDeleted": False,
-            },
-            order={"createdAt": "asc"},
-            include={
-                "LibraryAgents": {"where": {"isDeleted": False}},
-                "Children": {"where": {"isDeleted": False}},
-            },
-        )
-
-        # Build a map of folder ID to folder data
-        folder_map: dict[str, library_model.LibraryFolderTree] = {}
-        for folder in all_folders:
-            agent_count = len(folder.LibraryAgents) if folder.LibraryAgents else 0
-            subfolder_count = len(folder.Children) if folder.Children else 0
-            folder_map[folder.id] = library_model.LibraryFolderTree(
-                **library_model.LibraryFolder.from_db(
-                    folder,
-                    agent_count=agent_count,
-                    subfolder_count=subfolder_count,
-                ).model_dump(),
-                children=[],
-            )
-
-        # Build the tree structure
-        root_folders: list[library_model.LibraryFolderTree] = []
-        for folder in all_folders:
-            tree_folder = folder_map[folder.id]
-            if folder.parentId and folder.parentId in folder_map:
-                folder_map[folder.parentId].children.append(tree_folder)
-            else:
-                root_folders.append(tree_folder)
-
-        return root_folders
-
-    except prisma.errors.PrismaError as e:
-        logger.error(f"Database error getting folder tree: {e}")
-        raise DatabaseError("Failed to get folder tree") from e
-
-
-async def get_folder(
-    folder_id: str,
-    user_id: str,
-) -> library_model.LibraryFolder:
-    """
-    Gets a single folder by ID.
-
-    Args:
-        folder_id: The ID of the folder.
-        user_id: The ID of the user (for ownership verification).
-
-    Returns:
-        The LibraryFolder object.
-
-    Raises:
-        NotFoundError: If the folder doesn't exist or doesn't belong to the user.
-    """
-    try:
-        folder = await prisma.models.LibraryFolder.prisma().find_first(
-            where={
-                "id": folder_id,
-                "userId": user_id,
-                "isDeleted": False,
-            },
-            include={
-                "LibraryAgents": {"where": {"isDeleted": False}},
-                "Children": {"where": {"isDeleted": False}},
-            },
-        )
-
-        if not folder:
-            raise NotFoundError(f"Folder #{folder_id} not found")
-
-        agent_count = len(folder.LibraryAgents) if folder.LibraryAgents else 0
-        subfolder_count = len(folder.Children) if folder.Children else 0
-
-        return library_model.LibraryFolder.from_db(
-            folder,
-            agent_count=agent_count,
-            subfolder_count=subfolder_count,
-        )
-
-    except prisma.errors.PrismaError as e:
-        logger.error(f"Database error getting folder: {e}")
-        raise DatabaseError("Failed to get folder") from e
-
-
-async def get_folder_depth(folder_id: str, user_id: str) -> int:
-    """
-    Calculate the depth of a folder in the hierarchy (root=0).
-
-    Args:
-        folder_id: The ID of the folder.
-        user_id: The ID of the user.
-
-    Returns:
-        The depth of the folder (0 for root-level folders).
-    """
-    depth = 0
-    current_id: str | None = folder_id
-
-    while current_id:
-        folder = await prisma.models.LibraryFolder.prisma().find_first(
-            where={
-                "id": current_id,
-                "userId": user_id,
-                "isDeleted": False,
-            }
-        )
-        if not folder:
-            break
-        if folder.parentId:
-            depth += 1
-            current_id = folder.parentId
-        else:
-            break
-
-    return depth
-
-
-async def is_descendant_of(
-    folder_id: str,
-    potential_ancestor_id: str,
-    user_id: str,
-) -> bool:
-    """
-    Check if folder_id is a descendant of potential_ancestor_id.
-
-    Args:
-        folder_id: The ID of the folder to check.
-        potential_ancestor_id: The ID of the potential ancestor.
-        user_id: The ID of the user.
-
-    Returns:
-        True if folder_id is a descendant of potential_ancestor_id.
-    """
-    current_id: str | None = folder_id
-
-    while current_id:
-        if current_id == potential_ancestor_id:
-            return True
-
-        folder = await prisma.models.LibraryFolder.prisma().find_first(
-            where={
-                "id": current_id,
-                "userId": user_id,
-                "isDeleted": False,
-            }
-        )
-        if not folder or not folder.parentId:
-            break
-        current_id = folder.parentId
-
-    return False
-
-
-async def validate_folder_operation(
-    folder_id: Optional[str],
-    target_parent_id: Optional[str],
-    user_id: str,
-    max_depth: int = MAX_FOLDER_DEPTH,
-) -> None:
-    """
-    Validate that a folder move/create operation is valid.
-
-    Args:
-        folder_id: The ID of the folder being moved (None for create).
-        target_parent_id: The target parent ID (None for root).
-        user_id: The ID of the user.
-        max_depth: Maximum allowed nesting depth.
-
-    Raises:
-        FolderValidationError: If the operation is invalid.
-    """
-    # Cannot move folder into itself
-    if folder_id and folder_id == target_parent_id:
-        raise FolderValidationError("Cannot move folder into itself")
-
-    # Check for circular reference
-    if folder_id and target_parent_id:
-        if await is_descendant_of(target_parent_id, folder_id, user_id):
-            raise FolderValidationError("Cannot move folder into its own descendant")
-
-    # Check depth limit
-    if target_parent_id:
-        parent_depth = await get_folder_depth(target_parent_id, user_id)
-        if parent_depth + 1 >= max_depth:
-            raise FolderValidationError(
-                f"Maximum folder nesting depth of {max_depth} exceeded"
-            )
-
-
-async def create_folder(
-    user_id: str,
-    name: str,
-    parent_id: Optional[str] = None,
-    icon: Optional[str] = None,
-    color: Optional[str] = None,
-) -> library_model.LibraryFolder:
-    """
-    Creates a new folder for the user.
-
-    Args:
-        user_id: The ID of the user.
-        name: The folder name.
-        parent_id: Optional parent folder ID.
-        icon: Optional icon identifier.
-        color: Optional hex color code.
-
-    Returns:
-        The created LibraryFolder.
-
-    Raises:
-        FolderValidationError: If validation fails.
-        DatabaseError: If there's a database error.
-    """
-    logger.debug(f"Creating folder '{name}' for user #{user_id}")
-
-    try:
-        # Validate operation
-        await validate_folder_operation(
-            folder_id=None,
-            target_parent_id=parent_id,
-            user_id=user_id,
-        )
-
-        # Verify parent exists if provided
-        if parent_id:
-            parent = await prisma.models.LibraryFolder.prisma().find_first(
-                where={
-                    "id": parent_id,
-                    "userId": user_id,
-                    "isDeleted": False,
-                }
-            )
-            if not parent:
-                raise NotFoundError(f"Parent folder #{parent_id} not found")
-
-        # Build data dict conditionally - don't include Parent key if no parent_id
-        create_data: dict = {
-            "name": name,
-            "User": {"connect": {"id": user_id}},
-        }
-        if icon is not None:
-            create_data["icon"] = icon
-        if color is not None:
-            create_data["color"] = color
-        if parent_id:
-            create_data["Parent"] = {"connect": {"id": parent_id}}
-
-        folder = await prisma.models.LibraryFolder.prisma().create(data=create_data)
-
-        return library_model.LibraryFolder.from_db(folder)
-
-    except prisma.errors.UniqueViolationError:
-        raise FolderValidationError(
-            "A folder with this name already exists in this location"
-        )
-    except prisma.errors.PrismaError as e:
-        logger.error(f"Database error creating folder: {e}")
-        raise DatabaseError("Failed to create folder") from e
-
-
-async def create_folder_with_unique_name(
-    user_id: str,
-    base_name: str,
-    parent_id: Optional[str] = None,
-    icon: Optional[str] = None,
-    color: Optional[str] = None,
-) -> library_model.LibraryFolder:
-    """
-    Creates a folder, appending (2), (3), etc. if name exists.
-
-    Args:
-        user_id: The ID of the user.
-        base_name: The base folder name.
-        parent_id: Optional parent folder ID.
-        icon: Optional icon identifier.
-        color: Optional hex color code.
-
-    Returns:
-        The created LibraryFolder.
-    """
-    name = base_name
-    suffix = 1
-
-    while True:
-        try:
-            return await create_folder(
-                user_id=user_id,
-                name=name,
-                parent_id=parent_id,
-                icon=icon,
-                color=color,
-            )
-        except FolderValidationError as e:
-            if "already exists" in str(e):
-                suffix += 1
-                name = f"{base_name} ({suffix})"
-            else:
-                raise
-
-
-async def update_folder(
-    folder_id: str,
-    user_id: str,
-    name: Optional[str] = None,
-    icon: Optional[str] = None,
-    color: Optional[str] = None,
-) -> library_model.LibraryFolder:
-    """
-    Updates a folder's properties.
-
-    Args:
-        folder_id: The ID of the folder to update.
-        user_id: The ID of the user.
-        name: New folder name.
-        icon: New icon identifier.
-        color: New hex color code.
-
-    Returns:
-        The updated LibraryFolder.
-
-    Raises:
-        NotFoundError: If the folder doesn't exist.
-        DatabaseError: If there's a database error.
-    """
-    logger.debug(f"Updating folder #{folder_id} for user #{user_id}")
-
-    try:
-        # Verify folder exists and belongs to user
-        existing = await prisma.models.LibraryFolder.prisma().find_first(
-            where={
-                "id": folder_id,
-                "userId": user_id,
-                "isDeleted": False,
-            }
-        )
-        if not existing:
-            raise NotFoundError(f"Folder #{folder_id} not found")
-
-        update_data: prisma.types.LibraryFolderUpdateInput = {}
-        if name is not None:
-            update_data["name"] = name
-        if icon is not None:
-            update_data["icon"] = icon
-        if color is not None:
-            update_data["color"] = color
-
-        if not update_data:
-            return await get_folder(folder_id, user_id)
-
-        folder = await prisma.models.LibraryFolder.prisma().update(
-            where={"id": folder_id},
-            data=update_data,
-            include={
-                "LibraryAgents": {"where": {"isDeleted": False}},
-                "Children": {"where": {"isDeleted": False}},
-            },
-        )
-
-        if not folder:
-            raise NotFoundError(f"Folder #{folder_id} not found")
-
-        agent_count = len(folder.LibraryAgents) if folder.LibraryAgents else 0
-        subfolder_count = len(folder.Children) if folder.Children else 0
-
-        return library_model.LibraryFolder.from_db(
-            folder,
-            agent_count=agent_count,
-            subfolder_count=subfolder_count,
-        )
-
-    except prisma.errors.UniqueViolationError:
-        raise FolderValidationError(
-            "A folder with this name already exists in this location"
-        )
-    except prisma.errors.PrismaError as e:
-        logger.error(f"Database error updating folder: {e}")
-        raise DatabaseError("Failed to update folder") from e
-
-
-async def move_folder(
-    folder_id: str,
-    user_id: str,
-    target_parent_id: Optional[str],
-) -> library_model.LibraryFolder:
-    """
-    Moves a folder to a new parent.
-
-    Args:
-        folder_id: The ID of the folder to move.
-        user_id: The ID of the user.
-        target_parent_id: The target parent ID (None for root).
-
-    Returns:
-        The moved LibraryFolder.
-
-    Raises:
-        FolderValidationError: If the move is invalid.
-        NotFoundError: If the folder doesn't exist.
-        DatabaseError: If there's a database error.
-    """
-    logger.debug(f"Moving folder #{folder_id} to parent #{target_parent_id}")
-
-    try:
-        # Validate operation
-        await validate_folder_operation(
-            folder_id=folder_id,
-            target_parent_id=target_parent_id,
-            user_id=user_id,
-        )
-
-        # Verify folder exists
-        existing = await prisma.models.LibraryFolder.prisma().find_first(
-            where={
-                "id": folder_id,
-                "userId": user_id,
-                "isDeleted": False,
-            }
-        )
-        if not existing:
-            raise NotFoundError(f"Folder #{folder_id} not found")
-
-        # Verify target parent exists if provided
-        if target_parent_id:
-            parent = await prisma.models.LibraryFolder.prisma().find_first(
-                where={
-                    "id": target_parent_id,
-                    "userId": user_id,
-                    "isDeleted": False,
-                }
-            )
-            if not parent:
-                raise NotFoundError(
-                    f"Target parent folder #{target_parent_id} not found"
-                )
-
-        folder = await prisma.models.LibraryFolder.prisma().update(
-            where={"id": folder_id},
-            data={
-                "parentId": target_parent_id,
-            },
-            include={
-                "LibraryAgents": {"where": {"isDeleted": False}},
-                "Children": {"where": {"isDeleted": False}},
-            },
-        )
-
-        if not folder:
-            raise NotFoundError(f"Folder #{folder_id} not found")
-
-        agent_count = len(folder.LibraryAgents) if folder.LibraryAgents else 0
-        subfolder_count = len(folder.Children) if folder.Children else 0
-
-        return library_model.LibraryFolder.from_db(
-            folder,
-            agent_count=agent_count,
-            subfolder_count=subfolder_count,
-        )
-
-    except prisma.errors.UniqueViolationError:
-        raise FolderValidationError(
-            "A folder with this name already exists in this location"
-        )
-    except prisma.errors.PrismaError as e:
-        logger.error(f"Database error moving folder: {e}")
-        raise DatabaseError("Failed to move folder") from e
-
-
-async def delete_folder(
-    folder_id: str,
-    user_id: str,
-    soft_delete: bool = True,
-) -> None:
-    """
-    Deletes a folder and all its contents (cascade).
-
-    Args:
-        folder_id: The ID of the folder to delete.
-        user_id: The ID of the user.
-        soft_delete: If True, soft-deletes; otherwise hard-deletes.
-
-    Raises:
-        NotFoundError: If the folder doesn't exist.
-        DatabaseError: If there's a database error.
-    """
-    logger.debug(f"Deleting folder #{folder_id} for user #{user_id}")
-
-    try:
-        # Verify folder exists
-        existing = await prisma.models.LibraryFolder.prisma().find_first(
-            where={
-                "id": folder_id,
-                "userId": user_id,
-                "isDeleted": False,
-            }
-        )
-        if not existing:
-            raise NotFoundError(f"Folder #{folder_id} not found")
-
-        # Collect all folder IDs (target + descendants) before the transaction
-        async with transaction() as tx:
-            descendant_ids = await _get_descendant_folder_ids(folder_id, user_id, tx)
-        all_folder_ids = [folder_id] + descendant_ids
-
-        if soft_delete:
-            # Clean up schedules/webhooks for each affected agent before
-            # soft-deleting, matching what delete_library_agent() does.
-            affected_agents = await prisma.models.LibraryAgent.prisma().find_many(
-                where={
-                    "folderId": {"in": all_folder_ids},
-                    "userId": user_id,
-                    "isDeleted": False,
-                },
-            )
-            for agent in affected_agents:
-                try:
-                    await _cleanup_schedules_for_graph(
-                        graph_id=agent.agentGraphId, user_id=user_id
-                    )
-                    await _cleanup_webhooks_for_graph(
-                        graph_id=agent.agentGraphId, user_id=user_id
-                    )
-                except Exception as e:
-                    logger.warning(
-                        f"Cleanup failed for agent {agent.id} "
-                        f"(graph {agent.agentGraphId}): {e}"
-                    )
-
-        async with transaction() as tx:
-            if soft_delete:
-                # Soft-delete all agents in these folders
-                await prisma.models.LibraryAgent.prisma(tx).update_many(
-                    where={
-                        "folderId": {"in": all_folder_ids},
-                        "userId": user_id,
-                    },
-                    data={"isDeleted": True},
-                )
-
-                # Soft-delete all folders
-                await prisma.models.LibraryFolder.prisma(tx).update_many(
-                    where={
-                        "id": {"in": all_folder_ids},
-                        "userId": user_id,
-                    },
-                    data={"isDeleted": True},
-                )
-            else:
-                # Move agents to root (or could hard-delete them)
-                await prisma.models.LibraryAgent.prisma(tx).update_many(
-                    where={
-                        "folderId": {"in": all_folder_ids},
-                        "userId": user_id,
-                    },
-                    data={"folderId": None},
-                )
-
-                # Hard-delete folders (children first due to FK constraints)
-                for fid in reversed(all_folder_ids):
-                    await prisma.models.LibraryFolder.prisma(tx).delete(
-                        where={"id": fid}
-                    )
-
-    except prisma.errors.PrismaError as e:
-        logger.error(f"Database error deleting folder: {e}")
-        raise DatabaseError("Failed to delete folder") from e
-
-
-async def _get_descendant_folder_ids(
-    folder_id: str,
-    user_id: str,
-    tx: Optional[prisma.Prisma] = None,
-) -> list[str]:
-    """
-    Recursively get all descendant folder IDs.
-
-    Args:
-        folder_id: The ID of the parent folder.
-        user_id: The ID of the user.
-        tx: Optional transaction.
-
-    Returns:
-        A list of descendant folder IDs.
-    """
-    prisma_client = prisma.models.LibraryFolder.prisma(tx)
-    children = await prisma_client.find_many(
-        where={
-            "parentId": folder_id,
-            "userId": user_id,
-            "isDeleted": False,
-        }
-    )
-
-    result: list[str] = []
-    for child in children:
-        result.append(child.id)
-        result.extend(await _get_descendant_folder_ids(child.id, user_id, tx))
-
-    return result
-
-
-async def move_agent_to_folder(
-    library_agent_id: str,
-    folder_id: Optional[str],
-    user_id: str,
-) -> library_model.LibraryAgent:
-    """
-    Moves a library agent to a folder.
-
-    Args:
-        library_agent_id: The ID of the library agent.
-        folder_id: The target folder ID (None for root).
-        user_id: The ID of the user.
-
-    Returns:
-        The updated LibraryAgent.
-
-    Raises:
-        NotFoundError: If the agent or folder doesn't exist.
-        DatabaseError: If there's a database error.
-    """
-    logger.debug(f"Moving agent #{library_agent_id} to folder #{folder_id}")
-
-    try:
-        # Verify agent exists
-        agent = await prisma.models.LibraryAgent.prisma().find_first(
-            where={
-                "id": library_agent_id,
-                "userId": user_id,
-                "isDeleted": False,
-            }
-        )
-        if not agent:
-            raise NotFoundError(f"Library agent #{library_agent_id} not found")
-
-        # Verify folder exists if provided
-        if folder_id:
-            folder = await prisma.models.LibraryFolder.prisma().find_first(
-                where={
-                    "id": folder_id,
-                    "userId": user_id,
-                    "isDeleted": False,
-                }
-            )
-            if not folder:
-                raise NotFoundError(f"Folder #{folder_id} not found")
-
-        await prisma.models.LibraryAgent.prisma().update(
-            where={"id": library_agent_id},
-            data={"folderId": folder_id},
-        )
-
-        return await get_library_agent(library_agent_id, user_id)
-
-    except prisma.errors.PrismaError as e:
-        logger.error(f"Database error moving agent to folder: {e}")
-        raise DatabaseError("Failed to move agent to folder") from e
-
-
-async def bulk_move_agents_to_folder(
-    agent_ids: list[str],
-    folder_id: Optional[str],
-    user_id: str,
-) -> list[library_model.LibraryAgent]:
-    """
-    Moves multiple library agents to a folder.
-
-    Args:
-        agent_ids: The IDs of the library agents.
-        folder_id: The target folder ID (None for root).
-        user_id: The ID of the user.
-
-    Returns:
-        The updated LibraryAgents.
-
-    Raises:
-        NotFoundError: If any agent or the folder doesn't exist.
-        DatabaseError: If there's a database error.
-    """
-    logger.debug(f"Bulk moving {len(agent_ids)} agents to folder #{folder_id}")
-
-    try:
-        # Verify folder exists if provided
-        if folder_id:
-            folder = await prisma.models.LibraryFolder.prisma().find_first(
-                where={
-                    "id": folder_id,
-                    "userId": user_id,
-                    "isDeleted": False,
-                }
-            )
-            if not folder:
-                raise NotFoundError(f"Folder #{folder_id} not found")
-
-        # Update all agents
-        await prisma.models.LibraryAgent.prisma().update_many(
-            where={
-                "id": {"in": agent_ids},
-                "userId": user_id,
-                "isDeleted": False,
-            },
-            data={"folderId": folder_id},
-        )
-
-        # Fetch and return updated agents
-        agents = await prisma.models.LibraryAgent.prisma().find_many(
-            where={
-                "id": {"in": agent_ids},
-                "userId": user_id,
-            },
-            include=library_agent_include(
-                user_id, include_nodes=False, include_executions=False
-            ),
-        )
-
-        return [library_model.LibraryAgent.from_db(agent) for agent in agents]
-
-    except prisma.errors.PrismaError as e:
-        logger.error(f"Database error bulk moving agents to folder: {e}")
-        raise DatabaseError("Failed to bulk move agents to folder") from e
-
-
 ##############################################
 ########### Presets DB Functions #############
 ##############################################

autogpt_platform/backend/backend/api/features/library/exceptions.py

@@ -1,4 +0,0 @@
-class FolderValidationError(Exception):
-    """Raised when folder operations fail validation."""
-
-    pass

autogpt_platform/backend/backend/api/features/library/model.py

@@ -26,95 +26,6 @@ class LibraryAgentStatus(str, Enum):
     ERROR = "ERROR"
 
 
-# === Folder Models ===
-
-
-class LibraryFolder(pydantic.BaseModel):
-    """Represents a folder for organizing library agents."""
-
-    id: str
-    user_id: str
-    name: str
-    icon: str | None = None
-    color: str | None = None
-    parent_id: str | None = None
-    created_at: datetime.datetime
-    updated_at: datetime.datetime
-    agent_count: int = 0  # Direct agents in folder
-    subfolder_count: int = 0  # Direct child folders
-
-    @staticmethod
-    def from_db(
-        folder: prisma.models.LibraryFolder,
-        agent_count: int = 0,
-        subfolder_count: int = 0,
-    ) -> "LibraryFolder":
-        """Factory method that constructs a LibraryFolder from a Prisma model."""
-        return LibraryFolder(
-            id=folder.id,
-            user_id=folder.userId,
-            name=folder.name,
-            icon=folder.icon,
-            color=folder.color,
-            parent_id=folder.parentId,
-            created_at=folder.createdAt,
-            updated_at=folder.updatedAt,
-            agent_count=agent_count,
-            subfolder_count=subfolder_count,
-        )
-
-
-class LibraryFolderTree(LibraryFolder):
-    """Folder with nested children for tree view."""
-
-    children: list["LibraryFolderTree"] = []
-
-
-class FolderCreateRequest(pydantic.BaseModel):
-    """Request model for creating a folder."""
-
-    name: str = pydantic.Field(..., min_length=1, max_length=100)
-    icon: str | None = None
-    color: str | None = pydantic.Field(
-        None, pattern=r"^#[0-9A-Fa-f]{6}$", description="Hex color code (#RRGGBB)"
-    )
-    parent_id: str | None = None
-
-
-class FolderUpdateRequest(pydantic.BaseModel):
-    """Request model for updating a folder."""
-
-    name: str | None = pydantic.Field(None, min_length=1, max_length=100)
-    icon: str | None = None
-    color: str | None = None
-
-
-class FolderMoveRequest(pydantic.BaseModel):
-    """Request model for moving a folder to a new parent."""
-
-    target_parent_id: str | None = None  # None = move to root
-
-
-class BulkMoveAgentsRequest(pydantic.BaseModel):
-    """Request model for moving multiple agents to a folder."""
-
-    agent_ids: list[str]
-    folder_id: str | None = None  # None = move to root
-
-
-class FolderListResponse(pydantic.BaseModel):
-    """Response schema for a list of folders."""
-
-    folders: list[LibraryFolder]
-    pagination: Pagination
-
-
-class FolderTreeResponse(pydantic.BaseModel):
-    """Response schema for folder tree structure."""
-
-    tree: list[LibraryFolderTree]
-
-
 class MarketplaceListingCreator(pydantic.BaseModel):
     """Creator information for a marketplace listing."""
 
@@ -209,9 +120,6 @@ class LibraryAgent(pydantic.BaseModel):
     can_access_graph: bool
     is_latest_version: bool
     is_favorite: bool
-    folder_id: str | None = None
-    folder_name: str | None = None  # Denormalized for display
-
     recommended_schedule_cron: str | None = None
     settings: GraphSettings = pydantic.Field(default_factory=GraphSettings)
     marketplace_listing: Optional["MarketplaceListing"] = None
@@ -320,10 +228,6 @@ class LibraryAgent(pydantic.BaseModel):
                 creator=creator_data,
             )
 
-        # Folder information
-        folder_id = agent.folderId
-        folder_name = agent.Folder.name if agent.Folder else None
-
         return LibraryAgent(
             id=agent.id,
             graph_id=agent.agentGraphId,
@@ -355,8 +259,6 @@ class LibraryAgent(pydantic.BaseModel):
             can_access_graph=can_access_graph,
             is_latest_version=is_latest_version,
             is_favorite=agent.isFavorite,
-            folder_id=folder_id,
-            folder_name=folder_name,
             recommended_schedule_cron=agent.AgentGraph.recommendedScheduleCron,
             settings=_parse_settings(agent.settings),
             marketplace_listing=marketplace_listing_data,
@@ -568,7 +470,3 @@ class LibraryAgentUpdateRequest(pydantic.BaseModel):
     settings: Optional[GraphSettings] = pydantic.Field(
         default=None, description="User-specific settings for this library agent"
     )
-    folder_id: Optional[str] = pydantic.Field(
-        default=None,
-        description="Folder ID to move agent to (empty string for root)",
-    )

autogpt_platform/backend/backend/api/features/library/routes/__init__.py

@@ -1,11 +1,9 @@
 import fastapi
 
 from .agents import router as agents_router
-from .folders import router as folders_router
 from .presets import router as presets_router
 
 router = fastapi.APIRouter()
 
 router.include_router(presets_router)
-router.include_router(folders_router)
 router.include_router(agents_router)

autogpt_platform/backend/backend/api/features/library/routes/agents.py

@@ -41,34 +41,17 @@ async def list_library_agents(
         ge=1,
         description="Number of agents per page (must be >= 1)",
     ),
-    folder_id: Optional[str] = Query(
-        None,
-        description="Filter by folder ID",
-    ),
-    include_root_only: bool = Query(
-        False,
-        description="Only return agents without a folder (root-level agents)",
-    ),
 ) -> library_model.LibraryAgentResponse:
     """
     Get all agents in the user's library (both created and saved).
     """
-    try:
-        return await library_db.list_library_agents(
-            user_id=user_id,
-            search_term=search_term,
-            sort_by=sort_by,
-            page=page,
-            page_size=page_size,
-            folder_id=folder_id,
-            include_root_only=include_root_only,
-        )
-    except Exception as e:
-        logger.error(f"Could not list library agents for user #{user_id}: {e}")
-        raise HTTPException(
-            status_code=status.HTTP_500_INTERNAL_SERVER_ERROR,
-            detail=str(e),
-        ) from e
+    return await library_db.list_library_agents(
+        user_id=user_id,
+        search_term=search_term,
+        sort_by=sort_by,
+        page=page,
+        page_size=page_size,
+    )
 
 
 @router.get(
@@ -177,38 +160,15 @@ async def update_library_agent(
     """
     Update the library agent with the given fields.
     """
-    try:
-        return await library_db.update_library_agent(
-            library_agent_id=library_agent_id,
-            user_id=user_id,
-            auto_update_version=payload.auto_update_version,
-            graph_version=payload.graph_version,
-            is_favorite=payload.is_favorite,
-            is_archived=payload.is_archived,
-            settings=payload.settings,
-            folder_id=payload.folder_id,
-        )
-    except NotFoundError as e:
-        raise HTTPException(
-            status_code=status.HTTP_404_NOT_FOUND,
-            detail=str(e),
-        ) from e
-    except DatabaseError as e:
-        logger.error(
-            f"Database error while updating library agent: {e}", exc_info=True
-        )
-        raise HTTPException(
-            status_code=status.HTTP_500_INTERNAL_SERVER_ERROR,
-            detail={"message": "Internal server error", "hint": "Contact support"},
-        ) from e
-    except Exception as e:
-        logger.error(
-            f"Unexpected error while updating library agent: {e}", exc_info=True
-        )
-        raise HTTPException(
-            status_code=status.HTTP_500_INTERNAL_SERVER_ERROR,
-            detail={"message": "Internal server error", "hint": "Contact support"},
-        ) from e
+    return await library_db.update_library_agent(
+        library_agent_id=library_agent_id,
+        user_id=user_id,
+        auto_update_version=payload.auto_update_version,
+        graph_version=payload.graph_version,
+        is_favorite=payload.is_favorite,
+        is_archived=payload.is_archived,
+        settings=payload.settings,
+    )
 
 
 @router.delete(

autogpt_platform/backend/backend/api/features/library/routes/folders.py

@@ -1,408 +0,0 @@
-import logging
-from typing import Optional
-
-import autogpt_libs.auth as autogpt_auth_lib
-from fastapi import APIRouter, HTTPException, Query, Security, status
-from fastapi.responses import Response
-
-from backend.util.exceptions import DatabaseError, NotFoundError
-
-from .. import db as library_db
-from .. import model as library_model
-
-logger = logging.getLogger(__name__)
-
-router = APIRouter(
-    prefix="/folders",
-    tags=["library", "folders", "private"],
-    dependencies=[Security(autogpt_auth_lib.requires_user)],
-)
-
-
-@router.get(
-    "",
-    summary="List Library Folders",
-    response_model=library_model.FolderListResponse,
-    responses={
-        200: {"description": "List of folders"},
-        500: {"description": "Server error"},
-    },
-)
-async def list_folders(
-    user_id: str = Security(autogpt_auth_lib.get_user_id),
-    parent_id: Optional[str] = Query(
-        None,
-        description="Filter by parent folder ID. If not provided, returns root-level folders.",
-    ),
-    include_counts: bool = Query(
-        True,
-        description="Include agent and subfolder counts",
-    ),
-) -> library_model.FolderListResponse:
-    """
-    List folders for the authenticated user.
-
-    Args:
-        user_id: ID of the authenticated user.
-        parent_id: Optional parent folder ID to filter by.
-        include_counts: Whether to include agent and subfolder counts.
-
-    Returns:
-        A FolderListResponse containing folders.
-    """
-    try:
-        folders = await library_db.list_folders(
-            user_id=user_id,
-            parent_id=parent_id,
-            include_counts=include_counts,
-        )
-        return library_model.FolderListResponse(
-            folders=folders,
-            pagination=library_model.Pagination(
-                total_items=len(folders),
-                total_pages=1,
-                current_page=1,
-                page_size=len(folders),
-            ),
-        )
-    except Exception as e:
-        logger.error(f"Could not list folders for user #{user_id}: {e}")
-        raise HTTPException(
-            status_code=status.HTTP_500_INTERNAL_SERVER_ERROR,
-            detail="Internal server error",
-        ) from e
-
-
-@router.get(
-    "/tree",
-    summary="Get Folder Tree",
-    response_model=library_model.FolderTreeResponse,
-    responses={
-        200: {"description": "Folder tree structure"},
-        500: {"description": "Server error"},
-    },
-)
-async def get_folder_tree(
-    user_id: str = Security(autogpt_auth_lib.get_user_id),
-) -> library_model.FolderTreeResponse:
-    """
-    Get the full folder tree for the authenticated user.
-
-    Args:
-        user_id: ID of the authenticated user.
-
-    Returns:
-        A FolderTreeResponse containing the nested folder structure.
-    """
-    try:
-        tree = await library_db.get_folder_tree(user_id=user_id)
-        return library_model.FolderTreeResponse(tree=tree)
-    except Exception as e:
-        logger.error(f"Could not get folder tree for user #{user_id}: {e}")
-        raise HTTPException(
-            status_code=status.HTTP_500_INTERNAL_SERVER_ERROR,
-            detail="Internal server error",
-        ) from e
-
-
-@router.get(
-    "/{folder_id}",
-    summary="Get Folder",
-    response_model=library_model.LibraryFolder,
-    responses={
-        200: {"description": "Folder details"},
-        404: {"description": "Folder not found"},
-        500: {"description": "Server error"},
-    },
-)
-async def get_folder(
-    folder_id: str,
-    user_id: str = Security(autogpt_auth_lib.get_user_id),
-) -> library_model.LibraryFolder:
-    """
-    Get a specific folder.
-
-    Args:
-        folder_id: ID of the folder to retrieve.
-        user_id: ID of the authenticated user.
-
-    Returns:
-        The requested LibraryFolder.
-    """
-    try:
-        return await library_db.get_folder(folder_id=folder_id, user_id=user_id)
-    except NotFoundError as e:
-        raise HTTPException(
-            status_code=status.HTTP_404_NOT_FOUND,
-            detail=str(e),
-        ) from e
-    except Exception as e:
-        logger.error(f"Could not get folder #{folder_id}: {e}")
-        raise HTTPException(
-            status_code=status.HTTP_500_INTERNAL_SERVER_ERROR,
-            detail="Internal server error",
-        ) from e
-
-
-@router.post(
-    "",
-    summary="Create Folder",
-    status_code=status.HTTP_201_CREATED,
-    response_model=library_model.LibraryFolder,
-    responses={
-        201: {"description": "Folder created successfully"},
-        400: {"description": "Validation error"},
-        404: {"description": "Parent folder not found"},
-        409: {"description": "Folder name conflict"},
-        500: {"description": "Server error"},
-    },
-)
-async def create_folder(
-    payload: library_model.FolderCreateRequest,
-    user_id: str = Security(autogpt_auth_lib.get_user_id),
-) -> library_model.LibraryFolder:
-    """
-    Create a new folder.
-
-    Args:
-        payload: The folder creation request.
-        user_id: ID of the authenticated user.
-
-    Returns:
-        The created LibraryFolder.
-    """
-    try:
-        return await library_db.create_folder(
-            user_id=user_id,
-            name=payload.name,
-            parent_id=payload.parent_id,
-            icon=payload.icon,
-            color=payload.color,
-        )
-    except library_db.FolderValidationError as e:
-        if "already exists" in str(e):
-            raise HTTPException(
-                status_code=status.HTTP_409_CONFLICT,
-                detail=str(e),
-            ) from e
-        raise HTTPException(
-            status_code=status.HTTP_400_BAD_REQUEST,
-            detail=str(e),
-        ) from e
-    except NotFoundError as e:
-        raise HTTPException(
-            status_code=status.HTTP_404_NOT_FOUND,
-            detail=str(e),
-        ) from e
-    except DatabaseError as e:
-        logger.error(f"Database error creating folder: {e}")
-        raise HTTPException(
-            status_code=status.HTTP_500_INTERNAL_SERVER_ERROR,
-            detail="Internal server error",
-        ) from e
-
-
-@router.patch(
-    "/{folder_id}",
-    summary="Update Folder",
-    response_model=library_model.LibraryFolder,
-    responses={
-        200: {"description": "Folder updated successfully"},
-        400: {"description": "Validation error"},
-        404: {"description": "Folder not found"},
-        409: {"description": "Folder name conflict"},
-        500: {"description": "Server error"},
-    },
-)
-async def update_folder(
-    folder_id: str,
-    payload: library_model.FolderUpdateRequest,
-    user_id: str = Security(autogpt_auth_lib.get_user_id),
-) -> library_model.LibraryFolder:
-    """
-    Update a folder's properties.
-
-    Args:
-        folder_id: ID of the folder to update.
-        payload: The folder update request.
-        user_id: ID of the authenticated user.
-
-    Returns:
-        The updated LibraryFolder.
-    """
-    try:
-        return await library_db.update_folder(
-            folder_id=folder_id,
-            user_id=user_id,
-            name=payload.name,
-            icon=payload.icon,
-            color=payload.color,
-        )
-    except library_db.FolderValidationError as e:
-        if "already exists" in str(e):
-            raise HTTPException(
-                status_code=status.HTTP_409_CONFLICT,
-                detail=str(e),
-            ) from e
-        raise HTTPException(
-            status_code=status.HTTP_400_BAD_REQUEST,
-            detail=str(e),
-        ) from e
-    except NotFoundError as e:
-        raise HTTPException(
-            status_code=status.HTTP_404_NOT_FOUND,
-            detail=str(e),
-        ) from e
-    except DatabaseError as e:
-        logger.error(f"Database error updating folder #{folder_id}: {e}")
-        raise HTTPException(
-            status_code=status.HTTP_500_INTERNAL_SERVER_ERROR,
-            detail="Internal server error",
-        ) from e
-
-
-@router.post(
-    "/{folder_id}/move",
-    summary="Move Folder",
-    response_model=library_model.LibraryFolder,
-    responses={
-        200: {"description": "Folder moved successfully"},
-        400: {"description": "Validation error (circular reference, depth exceeded)"},
-        404: {"description": "Folder or target parent not found"},
-        409: {"description": "Folder name conflict in target location"},
-        500: {"description": "Server error"},
-    },
-)
-async def move_folder(
-    folder_id: str,
-    payload: library_model.FolderMoveRequest,
-    user_id: str = Security(autogpt_auth_lib.get_user_id),
-) -> library_model.LibraryFolder:
-    """
-    Move a folder to a new parent.
-
-    Args:
-        folder_id: ID of the folder to move.
-        payload: The move request with target parent.
-        user_id: ID of the authenticated user.
-
-    Returns:
-        The moved LibraryFolder.
-    """
-    try:
-        return await library_db.move_folder(
-            folder_id=folder_id,
-            user_id=user_id,
-            target_parent_id=payload.target_parent_id,
-        )
-    except library_db.FolderValidationError as e:
-        if "already exists" in str(e):
-            raise HTTPException(
-                status_code=status.HTTP_409_CONFLICT,
-                detail=str(e),
-            ) from e
-        raise HTTPException(
-            status_code=status.HTTP_400_BAD_REQUEST,
-            detail=str(e),
-        ) from e
-    except NotFoundError as e:
-        raise HTTPException(
-            status_code=status.HTTP_404_NOT_FOUND,
-            detail=str(e),
-        ) from e
-    except DatabaseError as e:
-        logger.error(f"Database error moving folder #{folder_id}: {e}")
-        raise HTTPException(
-            status_code=status.HTTP_500_INTERNAL_SERVER_ERROR,
-            detail="Internal server error",
-        ) from e
-
-
-@router.delete(
-    "/{folder_id}",
-    summary="Delete Folder",
-    status_code=status.HTTP_204_NO_CONTENT,
-    responses={
-        204: {"description": "Folder deleted successfully"},
-        404: {"description": "Folder not found"},
-        500: {"description": "Server error"},
-    },
-)
-async def delete_folder(
-    folder_id: str,
-    user_id: str = Security(autogpt_auth_lib.get_user_id),
-) -> Response:
-    """
-    Soft-delete a folder and all its contents.
-
-    Args:
-        folder_id: ID of the folder to delete.
-        user_id: ID of the authenticated user.
-
-    Returns:
-        204 No Content if successful.
-    """
-    try:
-        await library_db.delete_folder(
-            folder_id=folder_id,
-            user_id=user_id,
-            soft_delete=True,
-        )
-        return Response(status_code=status.HTTP_204_NO_CONTENT)
-    except NotFoundError as e:
-        raise HTTPException(
-            status_code=status.HTTP_404_NOT_FOUND,
-            detail=str(e),
-        ) from e
-    except DatabaseError as e:
-        logger.error(f"Database error deleting folder #{folder_id}: {e}")
-        raise HTTPException(
-            status_code=status.HTTP_500_INTERNAL_SERVER_ERROR,
-            detail="Internal server error",
-        ) from e
-
-
-# === Bulk Agent Operations ===
-
-
-@router.post(
-    "/agents/bulk-move",
-    summary="Bulk Move Agents",
-    response_model=list[library_model.LibraryAgent],
-    responses={
-        200: {"description": "Agents moved successfully"},
-        404: {"description": "Folder not found"},
-        500: {"description": "Server error"},
-    },
-)
-async def bulk_move_agents(
-    payload: library_model.BulkMoveAgentsRequest,
-    user_id: str = Security(autogpt_auth_lib.get_user_id),
-) -> list[library_model.LibraryAgent]:
-    """
-    Move multiple agents to a folder.
-
-    Args:
-        payload: The bulk move request with agent IDs and target folder.
-        user_id: ID of the authenticated user.
-
-    Returns:
-        The updated LibraryAgents.
-    """
-    try:
-        return await library_db.bulk_move_agents_to_folder(
-            agent_ids=payload.agent_ids,
-            folder_id=payload.folder_id,
-            user_id=user_id,
-        )
-    except NotFoundError as e:
-        raise HTTPException(
-            status_code=status.HTTP_404_NOT_FOUND,
-            detail=str(e),
-        ) from e
-    except DatabaseError as e:
-        logger.error(f"Database error bulk moving agents: {e}")
-        raise HTTPException(
-            status_code=status.HTTP_500_INTERNAL_SERVER_ERROR,
-            detail="Internal server error",
-        ) from e

autogpt_platform/backend/backend/api/features/mcp/routes.py

@@ -0,0 +1,404 @@
+"""
+MCP (Model Context Protocol) API routes.
+
+Provides endpoints for MCP tool discovery and OAuth authentication so the
+frontend can list available tools on an MCP server before placing a block.
+"""
+
+import logging
+from typing import Annotated, Any
+from urllib.parse import urlparse
+
+import fastapi
+from autogpt_libs.auth import get_user_id
+from fastapi import Security
+from pydantic import BaseModel, Field
+
+from backend.api.features.integrations.router import CredentialsMetaResponse
+from backend.blocks.mcp.client import MCPClient, MCPClientError
+from backend.blocks.mcp.oauth import MCPOAuthHandler
+from backend.data.model import OAuth2Credentials
+from backend.integrations.creds_manager import IntegrationCredentialsManager
+from backend.integrations.providers import ProviderName
+from backend.util.request import HTTPClientError, Requests
+from backend.util.settings import Settings
+
+logger = logging.getLogger(__name__)
+
+settings = Settings()
+router = fastapi.APIRouter(tags=["mcp"])
+creds_manager = IntegrationCredentialsManager()
+
+
+# ====================== Tool Discovery ====================== #
+
+
+class DiscoverToolsRequest(BaseModel):
+    """Request to discover tools on an MCP server."""
+
+    server_url: str = Field(description="URL of the MCP server")
+    auth_token: str | None = Field(
+        default=None,
+        description="Optional Bearer token for authenticated MCP servers",
+    )
+
+
+class MCPToolResponse(BaseModel):
+    """A single MCP tool returned by discovery."""
+
+    name: str
+    description: str
+    input_schema: dict[str, Any]
+
+
+class DiscoverToolsResponse(BaseModel):
+    """Response containing the list of tools available on an MCP server."""
+
+    tools: list[MCPToolResponse]
+    server_name: str | None = None
+    protocol_version: str | None = None
+
+
+@router.post(
+    "/discover-tools",
+    summary="Discover available tools on an MCP server",
+    response_model=DiscoverToolsResponse,
+)
+async def discover_tools(
+    request: DiscoverToolsRequest,
+    user_id: Annotated[str, Security(get_user_id)],
+) -> DiscoverToolsResponse:
+    """
+    Connect to an MCP server and return its available tools.
+
+    If the user has a stored MCP credential for this server URL, it will be
+    used automatically — no need to pass an explicit auth token.
+    """
+    auth_token = request.auth_token
+
+    # Auto-use stored MCP credential when no explicit token is provided.
+    if not auth_token:
+        mcp_creds = await creds_manager.store.get_creds_by_provider(
+            user_id, ProviderName.MCP.value
+        )
+        # Find the freshest credential for this server URL
+        best_cred: OAuth2Credentials | None = None
+        for cred in mcp_creds:
+            if (
+                isinstance(cred, OAuth2Credentials)
+                and (cred.metadata or {}).get("mcp_server_url") == request.server_url
+            ):
+                if best_cred is None or (
+                    (cred.access_token_expires_at or 0)
+                    > (best_cred.access_token_expires_at or 0)
+                ):
+                    best_cred = cred
+        if best_cred:
+            # Refresh the token if expired before using it
+            best_cred = await creds_manager.refresh_if_needed(user_id, best_cred)
+            logger.info(
+                f"Using MCP credential {best_cred.id} for {request.server_url}, "
+                f"expires_at={best_cred.access_token_expires_at}"
+            )
+            auth_token = best_cred.access_token.get_secret_value()
+
+    client = MCPClient(request.server_url, auth_token=auth_token)
+
+    try:
+        init_result = await client.initialize()
+        tools = await client.list_tools()
+    except HTTPClientError as e:
+        if e.status_code in (401, 403):
+            raise fastapi.HTTPException(
+                status_code=401,
+                detail="This MCP server requires authentication. "
+                "Please provide a valid auth token.",
+            )
+        raise fastapi.HTTPException(status_code=502, detail=str(e))
+    except MCPClientError as e:
+        raise fastapi.HTTPException(status_code=502, detail=str(e))
+    except Exception as e:
+        raise fastapi.HTTPException(
+            status_code=502,
+            detail=f"Failed to connect to MCP server: {e}",
+        )
+
+    return DiscoverToolsResponse(
+        tools=[
+            MCPToolResponse(
+                name=t.name,
+                description=t.description,
+                input_schema=t.input_schema,
+            )
+            for t in tools
+        ],
+        server_name=(
+            init_result.get("serverInfo", {}).get("name")
+            or urlparse(request.server_url).hostname
+            or "MCP"
+        ),
+        protocol_version=init_result.get("protocolVersion"),
+    )
+
+
+# ======================== OAuth Flow ======================== #
+
+
+class MCPOAuthLoginRequest(BaseModel):
+    """Request to start an OAuth flow for an MCP server."""
+
+    server_url: str = Field(description="URL of the MCP server that requires OAuth")
+
+
+class MCPOAuthLoginResponse(BaseModel):
+    """Response with the OAuth login URL for the user to authenticate."""
+
+    login_url: str
+    state_token: str
+
+
+@router.post(
+    "/oauth/login",
+    summary="Initiate OAuth login for an MCP server",
+)
+async def mcp_oauth_login(
+    request: MCPOAuthLoginRequest,
+    user_id: Annotated[str, Security(get_user_id)],
+) -> MCPOAuthLoginResponse:
+    """
+    Discover OAuth metadata from the MCP server and return a login URL.
+
+    1. Discovers the protected-resource metadata (RFC 9728)
+    2. Fetches the authorization server metadata (RFC 8414)
+    3. Performs Dynamic Client Registration (RFC 7591) if available
+    4. Returns the authorization URL for the frontend to open in a popup
+    """
+    client = MCPClient(request.server_url)
+
+    # Step 1: Discover protected-resource metadata (RFC 9728)
+    protected_resource = await client.discover_auth()
+
+    metadata: dict[str, Any] | None = None
+
+    if protected_resource and protected_resource.get("authorization_servers"):
+        auth_server_url = protected_resource["authorization_servers"][0]
+        resource_url = protected_resource.get("resource", request.server_url)
+
+        # Step 2a: Discover auth-server metadata (RFC 8414)
+        metadata = await client.discover_auth_server_metadata(auth_server_url)
+    else:
+        # Fallback: Some MCP servers (e.g. Linear) are their own auth server
+        # and serve OAuth metadata directly without protected-resource metadata.
+        # Don't assume a resource_url — omitting it lets the auth server choose
+        # the correct audience for the token (RFC 8707 resource is optional).
+        resource_url = None
+        metadata = await client.discover_auth_server_metadata(request.server_url)
+
+    if (
+        not metadata
+        or "authorization_endpoint" not in metadata
+        or "token_endpoint" not in metadata
+    ):
+        raise fastapi.HTTPException(
+            status_code=400,
+            detail="This MCP server does not advertise OAuth support. "
+            "You may need to provide an auth token manually.",
+        )
+
+    authorize_url = metadata["authorization_endpoint"]
+    token_url = metadata["token_endpoint"]
+    registration_endpoint = metadata.get("registration_endpoint")
+    revoke_url = metadata.get("revocation_endpoint")
+
+    # Step 3: Dynamic Client Registration (RFC 7591) if available
+    frontend_base_url = settings.config.frontend_base_url
+    if not frontend_base_url:
+        raise fastapi.HTTPException(
+            status_code=500,
+            detail="Frontend base URL is not configured.",
+        )
+    redirect_uri = f"{frontend_base_url}/auth/integrations/mcp_callback"
+
+    client_id = ""
+    client_secret = ""
+    if registration_endpoint:
+        reg_result = await _register_mcp_client(
+            registration_endpoint, redirect_uri, request.server_url
+        )
+        if reg_result:
+            client_id = reg_result.get("client_id", "")
+            client_secret = reg_result.get("client_secret", "")
+
+    if not client_id:
+        client_id = "autogpt-platform"
+
+    # Step 4: Store state token with OAuth metadata for the callback
+    scopes = (protected_resource or {}).get("scopes_supported") or metadata.get(
+        "scopes_supported", []
+    )
+    state_token, code_challenge = await creds_manager.store.store_state_token(
+        user_id,
+        ProviderName.MCP.value,
+        scopes,
+        state_metadata={
+            "authorize_url": authorize_url,
+            "token_url": token_url,
+            "revoke_url": revoke_url,
+            "resource_url": resource_url,
+            "server_url": request.server_url,
+            "client_id": client_id,
+            "client_secret": client_secret,
+        },
+    )
+
+    # Step 5: Build and return the login URL
+    handler = MCPOAuthHandler(
+        client_id=client_id,
+        client_secret=client_secret,
+        redirect_uri=redirect_uri,
+        authorize_url=authorize_url,
+        token_url=token_url,
+        resource_url=resource_url,
+    )
+    login_url = handler.get_login_url(
+        scopes, state_token, code_challenge=code_challenge
+    )
+
+    return MCPOAuthLoginResponse(login_url=login_url, state_token=state_token)
+
+
+class MCPOAuthCallbackRequest(BaseModel):
+    """Request to exchange an OAuth code for tokens."""
+
+    code: str = Field(description="Authorization code from OAuth callback")
+    state_token: str = Field(description="State token for CSRF verification")
+
+
+class MCPOAuthCallbackResponse(BaseModel):
+    """Response after successfully storing OAuth credentials."""
+
+    credential_id: str
+
+
+@router.post(
+    "/oauth/callback",
+    summary="Exchange OAuth code for MCP tokens",
+)
+async def mcp_oauth_callback(
+    request: MCPOAuthCallbackRequest,
+    user_id: Annotated[str, Security(get_user_id)],
+) -> CredentialsMetaResponse:
+    """
+    Exchange the authorization code for tokens and store the credential.
+
+    The frontend calls this after receiving the OAuth code from the popup.
+    On success, subsequent ``/discover-tools`` calls for the same server URL
+    will automatically use the stored credential.
+    """
+    valid_state = await creds_manager.store.verify_state_token(
+        user_id, request.state_token, ProviderName.MCP.value
+    )
+    if not valid_state:
+        raise fastapi.HTTPException(
+            status_code=400,
+            detail="Invalid or expired state token.",
+        )
+
+    meta = valid_state.state_metadata
+    frontend_base_url = settings.config.frontend_base_url
+    if not frontend_base_url:
+        raise fastapi.HTTPException(
+            status_code=500,
+            detail="Frontend base URL is not configured.",
+        )
+    redirect_uri = f"{frontend_base_url}/auth/integrations/mcp_callback"
+
+    handler = MCPOAuthHandler(
+        client_id=meta["client_id"],
+        client_secret=meta.get("client_secret", ""),
+        redirect_uri=redirect_uri,
+        authorize_url=meta["authorize_url"],
+        token_url=meta["token_url"],
+        revoke_url=meta.get("revoke_url"),
+        resource_url=meta.get("resource_url"),
+    )
+
+    try:
+        credentials = await handler.exchange_code_for_tokens(
+            request.code, valid_state.scopes, valid_state.code_verifier
+        )
+    except Exception as e:
+        raise fastapi.HTTPException(
+            status_code=400,
+            detail=f"OAuth token exchange failed: {e}",
+        )
+
+    # Enrich credential metadata for future lookup and token refresh
+    if credentials.metadata is None:
+        credentials.metadata = {}
+    credentials.metadata["mcp_server_url"] = meta["server_url"]
+    credentials.metadata["mcp_client_id"] = meta["client_id"]
+    credentials.metadata["mcp_client_secret"] = meta.get("client_secret", "")
+    credentials.metadata["mcp_token_url"] = meta["token_url"]
+    credentials.metadata["mcp_resource_url"] = meta.get("resource_url", "")
+
+    hostname = urlparse(meta["server_url"]).hostname or meta["server_url"]
+    credentials.title = f"MCP: {hostname}"
+
+    # Remove old MCP credentials for the same server to prevent stale token buildup.
+    try:
+        old_creds = await creds_manager.store.get_creds_by_provider(
+            user_id, ProviderName.MCP.value
+        )
+        for old in old_creds:
+            if (
+                isinstance(old, OAuth2Credentials)
+                and (old.metadata or {}).get("mcp_server_url") == meta["server_url"]
+            ):
+                await creds_manager.store.delete_creds_by_id(user_id, old.id)
+                logger.info(
+                    f"Removed old MCP credential {old.id} for {meta['server_url']}"
+                )
+    except Exception:
+        logger.debug("Could not clean up old MCP credentials", exc_info=True)
+
+    await creds_manager.create(user_id, credentials)
+
+    return CredentialsMetaResponse(
+        id=credentials.id,
+        provider=credentials.provider,
+        type=credentials.type,
+        title=credentials.title,
+        scopes=credentials.scopes,
+        username=credentials.username,
+        host=credentials.metadata.get("mcp_server_url"),
+    )
+
+
+# ======================== Helpers ======================== #
+
+
+async def _register_mcp_client(
+    registration_endpoint: str,
+    redirect_uri: str,
+    server_url: str,
+) -> dict[str, Any] | None:
+    """Attempt Dynamic Client Registration (RFC 7591) with an MCP auth server."""
+    try:
+        response = await Requests(raise_for_status=True).post(
+            registration_endpoint,
+            json={
+                "client_name": "AutoGPT Platform",
+                "redirect_uris": [redirect_uri],
+                "grant_types": ["authorization_code"],
+                "response_types": ["code"],
+                "token_endpoint_auth_method": "client_secret_post",
+            },
+        )
+        data = response.json()
+        if isinstance(data, dict) and "client_id" in data:
+            return data
+        return None
+    except Exception as e:
+        logger.warning(f"Dynamic client registration failed for {server_url}: {e}")
+        return None

autogpt_platform/backend/backend/api/features/mcp/test_routes.py

@@ -0,0 +1,436 @@
+"""Tests for MCP API routes.
+
+Uses httpx.AsyncClient with ASGITransport instead of fastapi.testclient.TestClient
+to avoid creating blocking portals that can corrupt pytest-asyncio's session event loop.
+"""
+
+from unittest.mock import AsyncMock, patch
+
+import fastapi
+import httpx
+import pytest
+import pytest_asyncio
+from autogpt_libs.auth import get_user_id
+
+from backend.api.features.mcp.routes import router
+from backend.blocks.mcp.client import MCPClientError, MCPTool
+from backend.util.request import HTTPClientError
+
+app = fastapi.FastAPI()
+app.include_router(router)
+app.dependency_overrides[get_user_id] = lambda: "test-user-id"
+
+
+@pytest_asyncio.fixture(scope="module")
+async def client():
+    transport = httpx.ASGITransport(app=app)
+    async with httpx.AsyncClient(transport=transport, base_url="http://test") as c:
+        yield c
+
+
+class TestDiscoverTools:
+    @pytest.mark.asyncio(loop_scope="session")
+    async def test_discover_tools_success(self, client):
+        mock_tools = [
+            MCPTool(
+                name="get_weather",
+                description="Get weather for a city",
+                input_schema={
+                    "type": "object",
+                    "properties": {"city": {"type": "string"}},
+                    "required": ["city"],
+                },
+            ),
+            MCPTool(
+                name="add_numbers",
+                description="Add two numbers",
+                input_schema={
+                    "type": "object",
+                    "properties": {
+                        "a": {"type": "number"},
+                        "b": {"type": "number"},
+                    },
+                },
+            ),
+        ]
+
+        with (
+            patch("backend.api.features.mcp.routes.MCPClient") as MockClient,
+            patch("backend.api.features.mcp.routes.creds_manager") as mock_cm,
+        ):
+            mock_cm.store.get_creds_by_provider = AsyncMock(return_value=[])
+            instance = MockClient.return_value
+            instance.initialize = AsyncMock(
+                return_value={
+                    "protocolVersion": "2025-03-26",
+                    "serverInfo": {"name": "test-server"},
+                }
+            )
+            instance.list_tools = AsyncMock(return_value=mock_tools)
+
+            response = await client.post(
+                "/discover-tools",
+                json={"server_url": "https://mcp.example.com/mcp"},
+            )
+
+        assert response.status_code == 200
+        data = response.json()
+        assert len(data["tools"]) == 2
+        assert data["tools"][0]["name"] == "get_weather"
+        assert data["tools"][1]["name"] == "add_numbers"
+        assert data["server_name"] == "test-server"
+        assert data["protocol_version"] == "2025-03-26"
+
+    @pytest.mark.asyncio(loop_scope="session")
+    async def test_discover_tools_with_auth_token(self, client):
+        with patch("backend.api.features.mcp.routes.MCPClient") as MockClient:
+            instance = MockClient.return_value
+            instance.initialize = AsyncMock(
+                return_value={"serverInfo": {}, "protocolVersion": "2025-03-26"}
+            )
+            instance.list_tools = AsyncMock(return_value=[])
+
+            response = await client.post(
+                "/discover-tools",
+                json={
+                    "server_url": "https://mcp.example.com/mcp",
+                    "auth_token": "my-secret-token",
+                },
+            )
+
+        assert response.status_code == 200
+        MockClient.assert_called_once_with(
+            "https://mcp.example.com/mcp",
+            auth_token="my-secret-token",
+        )
+
+    @pytest.mark.asyncio(loop_scope="session")
+    async def test_discover_tools_auto_uses_stored_credential(self, client):
+        """When no explicit token is given, stored MCP credentials are used."""
+        from pydantic import SecretStr
+
+        from backend.data.model import OAuth2Credentials
+
+        stored_cred = OAuth2Credentials(
+            provider="mcp",
+            title="MCP: example.com",
+            access_token=SecretStr("stored-token-123"),
+            refresh_token=None,
+            access_token_expires_at=None,
+            refresh_token_expires_at=None,
+            scopes=[],
+            metadata={"mcp_server_url": "https://mcp.example.com/mcp"},
+        )
+
+        with (
+            patch("backend.api.features.mcp.routes.MCPClient") as MockClient,
+            patch("backend.api.features.mcp.routes.creds_manager") as mock_cm,
+        ):
+            mock_cm.store.get_creds_by_provider = AsyncMock(return_value=[stored_cred])
+            mock_cm.refresh_if_needed = AsyncMock(return_value=stored_cred)
+            instance = MockClient.return_value
+            instance.initialize = AsyncMock(
+                return_value={"serverInfo": {}, "protocolVersion": "2025-03-26"}
+            )
+            instance.list_tools = AsyncMock(return_value=[])
+
+            response = await client.post(
+                "/discover-tools",
+                json={"server_url": "https://mcp.example.com/mcp"},
+            )
+
+        assert response.status_code == 200
+        MockClient.assert_called_once_with(
+            "https://mcp.example.com/mcp",
+            auth_token="stored-token-123",
+        )
+
+    @pytest.mark.asyncio(loop_scope="session")
+    async def test_discover_tools_mcp_error(self, client):
+        with (
+            patch("backend.api.features.mcp.routes.MCPClient") as MockClient,
+            patch("backend.api.features.mcp.routes.creds_manager") as mock_cm,
+        ):
+            mock_cm.store.get_creds_by_provider = AsyncMock(return_value=[])
+            instance = MockClient.return_value
+            instance.initialize = AsyncMock(
+                side_effect=MCPClientError("Connection refused")
+            )
+
+            response = await client.post(
+                "/discover-tools",
+                json={"server_url": "https://bad-server.example.com/mcp"},
+            )
+
+        assert response.status_code == 502
+        assert "Connection refused" in response.json()["detail"]
+
+    @pytest.mark.asyncio(loop_scope="session")
+    async def test_discover_tools_generic_error(self, client):
+        with (
+            patch("backend.api.features.mcp.routes.MCPClient") as MockClient,
+            patch("backend.api.features.mcp.routes.creds_manager") as mock_cm,
+        ):
+            mock_cm.store.get_creds_by_provider = AsyncMock(return_value=[])
+            instance = MockClient.return_value
+            instance.initialize = AsyncMock(side_effect=Exception("Network timeout"))
+
+            response = await client.post(
+                "/discover-tools",
+                json={"server_url": "https://timeout.example.com/mcp"},
+            )
+
+        assert response.status_code == 502
+        assert "Failed to connect" in response.json()["detail"]
+
+    @pytest.mark.asyncio(loop_scope="session")
+    async def test_discover_tools_auth_required(self, client):
+        with (
+            patch("backend.api.features.mcp.routes.MCPClient") as MockClient,
+            patch("backend.api.features.mcp.routes.creds_manager") as mock_cm,
+        ):
+            mock_cm.store.get_creds_by_provider = AsyncMock(return_value=[])
+            instance = MockClient.return_value
+            instance.initialize = AsyncMock(
+                side_effect=HTTPClientError("HTTP 401 Error: Unauthorized", 401)
+            )
+
+            response = await client.post(
+                "/discover-tools",
+                json={"server_url": "https://auth-server.example.com/mcp"},
+            )
+
+        assert response.status_code == 401
+        assert "requires authentication" in response.json()["detail"]
+
+    @pytest.mark.asyncio(loop_scope="session")
+    async def test_discover_tools_forbidden(self, client):
+        with (
+            patch("backend.api.features.mcp.routes.MCPClient") as MockClient,
+            patch("backend.api.features.mcp.routes.creds_manager") as mock_cm,
+        ):
+            mock_cm.store.get_creds_by_provider = AsyncMock(return_value=[])
+            instance = MockClient.return_value
+            instance.initialize = AsyncMock(
+                side_effect=HTTPClientError("HTTP 403 Error: Forbidden", 403)
+            )
+
+            response = await client.post(
+                "/discover-tools",
+                json={"server_url": "https://auth-server.example.com/mcp"},
+            )
+
+        assert response.status_code == 401
+        assert "requires authentication" in response.json()["detail"]
+
+    @pytest.mark.asyncio(loop_scope="session")
+    async def test_discover_tools_missing_url(self, client):
+        response = await client.post("/discover-tools", json={})
+        assert response.status_code == 422
+
+
+class TestOAuthLogin:
+    @pytest.mark.asyncio(loop_scope="session")
+    async def test_oauth_login_success(self, client):
+        with (
+            patch("backend.api.features.mcp.routes.MCPClient") as MockClient,
+            patch("backend.api.features.mcp.routes.creds_manager") as mock_cm,
+            patch("backend.api.features.mcp.routes.settings") as mock_settings,
+            patch(
+                "backend.api.features.mcp.routes._register_mcp_client"
+            ) as mock_register,
+        ):
+            instance = MockClient.return_value
+            instance.discover_auth = AsyncMock(
+                return_value={
+                    "authorization_servers": ["https://auth.sentry.io"],
+                    "resource": "https://mcp.sentry.dev/mcp",
+                    "scopes_supported": ["openid"],
+                }
+            )
+            instance.discover_auth_server_metadata = AsyncMock(
+                return_value={
+                    "authorization_endpoint": "https://auth.sentry.io/authorize",
+                    "token_endpoint": "https://auth.sentry.io/token",
+                    "registration_endpoint": "https://auth.sentry.io/register",
+                }
+            )
+            mock_register.return_value = {
+                "client_id": "registered-client-id",
+                "client_secret": "registered-secret",
+            }
+            mock_cm.store.store_state_token = AsyncMock(
+                return_value=("state-token-123", "code-challenge-abc")
+            )
+            mock_settings.config.frontend_base_url = "http://localhost:3000"
+
+            response = await client.post(
+                "/oauth/login",
+                json={"server_url": "https://mcp.sentry.dev/mcp"},
+            )
+
+        assert response.status_code == 200
+        data = response.json()
+        assert "login_url" in data
+        assert data["state_token"] == "state-token-123"
+        assert "auth.sentry.io/authorize" in data["login_url"]
+        assert "registered-client-id" in data["login_url"]
+
+    @pytest.mark.asyncio(loop_scope="session")
+    async def test_oauth_login_no_oauth_support(self, client):
+        with patch("backend.api.features.mcp.routes.MCPClient") as MockClient:
+            instance = MockClient.return_value
+            instance.discover_auth = AsyncMock(return_value=None)
+            instance.discover_auth_server_metadata = AsyncMock(return_value=None)
+
+            response = await client.post(
+                "/oauth/login",
+                json={"server_url": "https://simple-server.example.com/mcp"},
+            )
+
+        assert response.status_code == 400
+        assert "does not advertise OAuth" in response.json()["detail"]
+
+    @pytest.mark.asyncio(loop_scope="session")
+    async def test_oauth_login_fallback_to_public_client(self, client):
+        """When DCR is unavailable, falls back to default public client ID."""
+        with (
+            patch("backend.api.features.mcp.routes.MCPClient") as MockClient,
+            patch("backend.api.features.mcp.routes.creds_manager") as mock_cm,
+            patch("backend.api.features.mcp.routes.settings") as mock_settings,
+        ):
+            instance = MockClient.return_value
+            instance.discover_auth = AsyncMock(
+                return_value={
+                    "authorization_servers": ["https://auth.example.com"],
+                    "resource": "https://mcp.example.com/mcp",
+                }
+            )
+            instance.discover_auth_server_metadata = AsyncMock(
+                return_value={
+                    "authorization_endpoint": "https://auth.example.com/authorize",
+                    "token_endpoint": "https://auth.example.com/token",
+                    # No registration_endpoint
+                }
+            )
+            mock_cm.store.store_state_token = AsyncMock(
+                return_value=("state-abc", "challenge-xyz")
+            )
+            mock_settings.config.frontend_base_url = "http://localhost:3000"
+
+            response = await client.post(
+                "/oauth/login",
+                json={"server_url": "https://mcp.example.com/mcp"},
+            )
+
+        assert response.status_code == 200
+        data = response.json()
+        assert "autogpt-platform" in data["login_url"]
+
+
+class TestOAuthCallback:
+    @pytest.mark.asyncio(loop_scope="session")
+    async def test_oauth_callback_success(self, client):
+        from pydantic import SecretStr
+
+        from backend.data.model import OAuth2Credentials
+
+        mock_creds = OAuth2Credentials(
+            provider="mcp",
+            title=None,
+            access_token=SecretStr("access-token-xyz"),
+            refresh_token=None,
+            access_token_expires_at=None,
+            refresh_token_expires_at=None,
+            scopes=[],
+            metadata={
+                "mcp_token_url": "https://auth.sentry.io/token",
+                "mcp_resource_url": "https://mcp.sentry.dev/mcp",
+            },
+        )
+
+        with (
+            patch("backend.api.features.mcp.routes.creds_manager") as mock_cm,
+            patch("backend.api.features.mcp.routes.settings") as mock_settings,
+            patch("backend.api.features.mcp.routes.MCPOAuthHandler") as MockHandler,
+        ):
+            mock_settings.config.frontend_base_url = "http://localhost:3000"
+
+            # Mock state verification
+            mock_state = AsyncMock()
+            mock_state.state_metadata = {
+                "authorize_url": "https://auth.sentry.io/authorize",
+                "token_url": "https://auth.sentry.io/token",
+                "client_id": "test-client-id",
+                "client_secret": "test-secret",
+                "server_url": "https://mcp.sentry.dev/mcp",
+            }
+            mock_state.scopes = ["openid"]
+            mock_state.code_verifier = "verifier-123"
+            mock_cm.store.verify_state_token = AsyncMock(return_value=mock_state)
+            mock_cm.create = AsyncMock()
+
+            handler_instance = MockHandler.return_value
+            handler_instance.exchange_code_for_tokens = AsyncMock(
+                return_value=mock_creds
+            )
+
+            # Mock old credential cleanup
+            mock_cm.store.get_creds_by_provider = AsyncMock(return_value=[])
+
+            response = await client.post(
+                "/oauth/callback",
+                json={"code": "auth-code-abc", "state_token": "state-token-123"},
+            )
+
+        assert response.status_code == 200
+        data = response.json()
+        assert "id" in data
+        assert data["provider"] == "mcp"
+        assert data["type"] == "oauth2"
+        mock_cm.create.assert_called_once()
+
+    @pytest.mark.asyncio(loop_scope="session")
+    async def test_oauth_callback_invalid_state(self, client):
+        with patch("backend.api.features.mcp.routes.creds_manager") as mock_cm:
+            mock_cm.store.verify_state_token = AsyncMock(return_value=None)
+
+            response = await client.post(
+                "/oauth/callback",
+                json={"code": "auth-code", "state_token": "bad-state"},
+            )
+
+        assert response.status_code == 400
+        assert "Invalid or expired" in response.json()["detail"]
+
+    @pytest.mark.asyncio(loop_scope="session")
+    async def test_oauth_callback_token_exchange_fails(self, client):
+        with (
+            patch("backend.api.features.mcp.routes.creds_manager") as mock_cm,
+            patch("backend.api.features.mcp.routes.settings") as mock_settings,
+            patch("backend.api.features.mcp.routes.MCPOAuthHandler") as MockHandler,
+        ):
+            mock_settings.config.frontend_base_url = "http://localhost:3000"
+            mock_state = AsyncMock()
+            mock_state.state_metadata = {
+                "authorize_url": "https://auth.example.com/authorize",
+                "token_url": "https://auth.example.com/token",
+                "client_id": "cid",
+                "server_url": "https://mcp.example.com/mcp",
+            }
+            mock_state.scopes = []
+            mock_state.code_verifier = "v"
+            mock_cm.store.verify_state_token = AsyncMock(return_value=mock_state)
+
+            handler_instance = MockHandler.return_value
+            handler_instance.exchange_code_for_tokens = AsyncMock(
+                side_effect=RuntimeError("Token exchange failed")
+            )
+
+            response = await client.post(
+                "/oauth/callback",
+                json={"code": "bad-code", "state_token": "state"},
+            )
+
+        assert response.status_code == 400
+        assert "token exchange failed" in response.json()["detail"].lower()

autogpt_platform/backend/backend/api/rest_api.py

@@ -26,6 +26,7 @@ import backend.api.features.executions.review.routes
 import backend.api.features.library.db
 import backend.api.features.library.model
 import backend.api.features.library.routes
+import backend.api.features.mcp.routes as mcp_routes
 import backend.api.features.oauth
 import backend.api.features.otto.routes
 import backend.api.features.postmark.postmark
@@ -40,11 +41,11 @@ import backend.data.user
 import backend.integrations.webhooks.utils
 import backend.util.service
 import backend.util.settings
-from backend.api.features.chat.completion_consumer import (
+from backend.blocks.llm import DEFAULT_LLM_MODEL
+from backend.copilot.completion_consumer import (
     start_completion_consumer,
     stop_completion_consumer,
 )
-from backend.blocks.llm import DEFAULT_LLM_MODEL
 from backend.data.model import Credentials
 from backend.integrations.providers import ProviderName
 from backend.monitoring.instrumentation import instrument_fastapi
@@ -343,6 +344,11 @@ app.include_router(
     tags=["workspace"],
     prefix="/api/workspace",
 )
+app.include_router(
+    mcp_routes.router,
+    tags=["v2", "mcp"],
+    prefix="/api/mcp",
+)
 app.include_router(
     backend.api.features.oauth.router,
     tags=["oauth"],

autogpt_platform/backend/backend/app.py

@@ -38,7 +38,9 @@ def main(**kwargs):
 
     from backend.api.rest_api import AgentServer
     from backend.api.ws_api import WebsocketServer
-    from backend.executor import DatabaseManager, ExecutionManager, Scheduler
+    from backend.copilot.executor.manager import CoPilotExecutor
+    from backend.data.db_manager import DatabaseManager
+    from backend.executor import ExecutionManager, Scheduler
     from backend.notifications import NotificationManager
 
     run_processes(
@@ -48,6 +50,7 @@ def main(**kwargs):
         WebsocketServer(),
         AgentServer(),
         ExecutionManager(),
+        CoPilotExecutor(),
         **kwargs,
     )
 

autogpt_platform/backend/backend/blocks/_base.py

@@ -64,6 +64,7 @@ class BlockType(Enum):
     AI = "AI"
     AYRSHARE = "Ayrshare"
     HUMAN_IN_THE_LOOP = "Human In The Loop"
+    MCP_TOOL = "MCP Tool"
 
 
 class BlockCategory(Enum):

autogpt_platform/backend/backend/blocks/basic.py

@@ -126,6 +126,7 @@ class PrintToConsoleBlock(Block):
             output_schema=PrintToConsoleBlock.Output,
             test_input={"text": "Hello, World!"},
             is_sensitive_action=True,
+            disabled=True,  # Disabled per Nick Tindle's request (OPEN-3000)
             test_output=[
                 ("output", "Hello, World!"),
                 ("status", "printed"),

autogpt_platform/backend/backend/blocks/mcp/block.py

@@ -0,0 +1,300 @@
+"""
+MCP (Model Context Protocol) Tool Block.
+
+A single dynamic block that can connect to any MCP server, discover available tools,
+and execute them. Works like AgentExecutorBlock — the user selects a tool from a
+dropdown and the input/output schema adapts dynamically.
+"""
+
+import json
+import logging
+from typing import Any, Literal
+
+from pydantic import SecretStr
+
+from backend.blocks._base import (
+    Block,
+    BlockCategory,
+    BlockSchemaInput,
+    BlockSchemaOutput,
+    BlockType,
+)
+from backend.blocks.mcp.client import MCPClient, MCPClientError
+from backend.data.block import BlockInput, BlockOutput
+from backend.data.model import (
+    CredentialsField,
+    CredentialsMetaInput,
+    OAuth2Credentials,
+    SchemaField,
+)
+from backend.integrations.providers import ProviderName
+from backend.util.json import validate_with_jsonschema
+
+logger = logging.getLogger(__name__)
+
+TEST_CREDENTIALS = OAuth2Credentials(
+    id="test-mcp-cred",
+    provider="mcp",
+    access_token=SecretStr("mock-mcp-token"),
+    refresh_token=SecretStr("mock-refresh"),
+    scopes=[],
+    title="Mock MCP credential",
+)
+TEST_CREDENTIALS_INPUT = {
+    "provider": TEST_CREDENTIALS.provider,
+    "id": TEST_CREDENTIALS.id,
+    "type": TEST_CREDENTIALS.type,
+    "title": TEST_CREDENTIALS.title,
+}
+
+
+MCPCredentials = CredentialsMetaInput[Literal[ProviderName.MCP], Literal["oauth2"]]
+
+
+class MCPToolBlock(Block):
+    """
+    A block that connects to an MCP server, lets the user pick a tool,
+    and executes it with dynamic input/output schema.
+
+    The flow:
+    1. User provides an MCP server URL (and optional credentials)
+    2. Frontend calls the backend to get tool list from that URL
+    3. User selects a tool from a dropdown (available_tools)
+    4. The block's input schema updates to reflect the selected tool's parameters
+    5. On execution, the block calls the MCP server to run the tool
+    """
+
+    class Input(BlockSchemaInput):
+        server_url: str = SchemaField(
+            description="URL of the MCP server (Streamable HTTP endpoint)",
+            placeholder="https://mcp.example.com/mcp",
+        )
+        credentials: MCPCredentials = CredentialsField(
+            discriminator="server_url",
+            description="MCP server OAuth credentials",
+            default={},
+        )
+        selected_tool: str = SchemaField(
+            description="The MCP tool to execute",
+            placeholder="Select a tool",
+            default="",
+        )
+        tool_input_schema: dict[str, Any] = SchemaField(
+            description="JSON Schema for the selected tool's input parameters. "
+            "Populated automatically when a tool is selected.",
+            default={},
+            hidden=True,
+        )
+
+        tool_arguments: dict[str, Any] = SchemaField(
+            description="Arguments to pass to the selected MCP tool. "
+            "The fields here are defined by the tool's input schema.",
+            default={},
+        )
+
+        @classmethod
+        def get_input_schema(cls, data: BlockInput) -> dict[str, Any]:
+            """Return the tool's input schema so the builder UI renders dynamic fields."""
+            return data.get("tool_input_schema", {})
+
+        @classmethod
+        def get_input_defaults(cls, data: BlockInput) -> BlockInput:
+            """Return the current tool_arguments as defaults for the dynamic fields."""
+            return data.get("tool_arguments", {})
+
+        @classmethod
+        def get_missing_input(cls, data: BlockInput) -> set[str]:
+            """Check which required tool arguments are missing."""
+            required_fields = cls.get_input_schema(data).get("required", [])
+            tool_arguments = data.get("tool_arguments", {})
+            return set(required_fields) - set(tool_arguments)
+
+        @classmethod
+        def get_mismatch_error(cls, data: BlockInput) -> str | None:
+            """Validate tool_arguments against the tool's input schema."""
+            tool_schema = cls.get_input_schema(data)
+            if not tool_schema:
+                return None
+            tool_arguments = data.get("tool_arguments", {})
+            return validate_with_jsonschema(tool_schema, tool_arguments)
+
+    class Output(BlockSchemaOutput):
+        result: Any = SchemaField(description="The result returned by the MCP tool")
+        error: str = SchemaField(description="Error message if the tool call failed")
+
+    def __init__(self):
+        super().__init__(
+            id="a0a4b1c2-d3e4-4f56-a7b8-c9d0e1f2a3b4",
+            description="Connect to any MCP server and execute its tools. "
+            "Provide a server URL, select a tool, and pass arguments dynamically.",
+            categories={BlockCategory.DEVELOPER_TOOLS},
+            input_schema=MCPToolBlock.Input,
+            output_schema=MCPToolBlock.Output,
+            block_type=BlockType.MCP_TOOL,
+            test_credentials=TEST_CREDENTIALS,
+            test_input={
+                "server_url": "https://mcp.example.com/mcp",
+                "credentials": TEST_CREDENTIALS_INPUT,
+                "selected_tool": "get_weather",
+                "tool_input_schema": {
+                    "type": "object",
+                    "properties": {"city": {"type": "string"}},
+                    "required": ["city"],
+                },
+                "tool_arguments": {"city": "London"},
+            },
+            test_output=[
+                (
+                    "result",
+                    {"weather": "sunny", "temperature": 20},
+                ),
+            ],
+            test_mock={
+                "_call_mcp_tool": lambda *a, **kw: {
+                    "weather": "sunny",
+                    "temperature": 20,
+                },
+            },
+        )
+
+    async def _call_mcp_tool(
+        self,
+        server_url: str,
+        tool_name: str,
+        arguments: dict[str, Any],
+        auth_token: str | None = None,
+    ) -> Any:
+        """Call a tool on the MCP server. Extracted for easy mocking in tests."""
+        client = MCPClient(server_url, auth_token=auth_token)
+        await client.initialize()
+        result = await client.call_tool(tool_name, arguments)
+
+        if result.is_error:
+            error_text = ""
+            for item in result.content:
+                if item.get("type") == "text":
+                    error_text += item.get("text", "")
+            raise MCPClientError(
+                f"MCP tool '{tool_name}' returned an error: "
+                f"{error_text or 'Unknown error'}"
+            )
+
+        # Extract text content from the result
+        output_parts = []
+        for item in result.content:
+            if item.get("type") == "text":
+                text = item.get("text", "")
+                # Try to parse as JSON for structured output
+                try:
+                    output_parts.append(json.loads(text))
+                except (json.JSONDecodeError, ValueError):
+                    output_parts.append(text)
+            elif item.get("type") == "image":
+                output_parts.append(
+                    {
+                        "type": "image",
+                        "data": item.get("data"),
+                        "mimeType": item.get("mimeType"),
+                    }
+                )
+            elif item.get("type") == "resource":
+                output_parts.append(item.get("resource", {}))
+
+        # If single result, unwrap
+        if len(output_parts) == 1:
+            return output_parts[0]
+        return output_parts if output_parts else None
+
+    @staticmethod
+    async def _auto_lookup_credential(
+        user_id: str, server_url: str
+    ) -> "OAuth2Credentials | None":
+        """Auto-lookup stored MCP credential for a server URL.
+
+        This is a fallback for nodes that don't have ``credentials`` explicitly
+        set (e.g. nodes created before the credential field was wired up).
+        """
+        from backend.integrations.creds_manager import IntegrationCredentialsManager
+        from backend.integrations.providers import ProviderName
+
+        try:
+            mgr = IntegrationCredentialsManager()
+            mcp_creds = await mgr.store.get_creds_by_provider(
+                user_id, ProviderName.MCP.value
+            )
+            best: OAuth2Credentials | None = None
+            for cred in mcp_creds:
+                if (
+                    isinstance(cred, OAuth2Credentials)
+                    and (cred.metadata or {}).get("mcp_server_url") == server_url
+                ):
+                    if best is None or (
+                        (cred.access_token_expires_at or 0)
+                        > (best.access_token_expires_at or 0)
+                    ):
+                        best = cred
+            if best:
+                best = await mgr.refresh_if_needed(user_id, best)
+                logger.info(
+                    "Auto-resolved MCP credential %s for %s", best.id, server_url
+                )
+            return best
+        except Exception:
+            logger.warning("Auto-lookup MCP credential failed", exc_info=True)
+            return None
+
+    async def run(
+        self,
+        input_data: Input,
+        *,
+        user_id: str,
+        credentials: OAuth2Credentials | None = None,
+        **kwargs,
+    ) -> BlockOutput:
+        if not input_data.server_url:
+            yield "error", "MCP server URL is required"
+            return
+
+        if not input_data.selected_tool:
+            yield "error", "No tool selected. Please select a tool from the dropdown."
+            return
+
+        # Validate required tool arguments before calling the server.
+        # The executor-level validation is bypassed for MCP blocks because
+        # get_input_defaults() flattens tool_arguments, stripping tool_input_schema
+        # from the validation context.
+        required = set(input_data.tool_input_schema.get("required", []))
+        if required:
+            missing = required - set(input_data.tool_arguments.keys())
+            if missing:
+                yield "error", (
+                    f"Missing required argument(s): {', '.join(sorted(missing))}. "
+                    f"Please fill in all required fields marked with * in the block form."
+                )
+                return
+
+        # If no credentials were injected by the executor (e.g. legacy nodes
+        # that don't have the credentials field set), try to auto-lookup
+        # the stored MCP credential for this server URL.
+        if credentials is None:
+            credentials = await self._auto_lookup_credential(
+                user_id, input_data.server_url
+            )
+
+        auth_token = (
+            credentials.access_token.get_secret_value() if credentials else None
+        )
+
+        try:
+            result = await self._call_mcp_tool(
+                server_url=input_data.server_url,
+                tool_name=input_data.selected_tool,
+                arguments=input_data.tool_arguments,
+                auth_token=auth_token,
+            )
+            yield "result", result
+        except MCPClientError as e:
+            yield "error", str(e)
+        except Exception as e:
+            logger.exception(f"MCP tool call failed: {e}")
+            yield "error", f"MCP tool call failed: {str(e)}"

autogpt_platform/backend/backend/blocks/mcp/client.py

@@ -0,0 +1,323 @@
+"""
+MCP (Model Context Protocol) HTTP client.
+
+Implements the MCP Streamable HTTP transport for listing tools and calling tools
+on remote MCP servers. Uses JSON-RPC 2.0 over HTTP POST.
+
+Handles both JSON and SSE (text/event-stream) response formats per the MCP spec.
+
+Reference: https://modelcontextprotocol.io/specification/2025-03-26/basic/transports
+"""
+
+import json
+import logging
+from dataclasses import dataclass, field
+from typing import Any
+
+from backend.util.request import Requests
+
+logger = logging.getLogger(__name__)
+
+
+@dataclass
+class MCPTool:
+    """Represents an MCP tool discovered from a server."""
+
+    name: str
+    description: str
+    input_schema: dict[str, Any]
+
+
+@dataclass
+class MCPCallResult:
+    """Result from calling an MCP tool."""
+
+    content: list[dict[str, Any]] = field(default_factory=list)
+    is_error: bool = False
+
+
+class MCPClientError(Exception):
+    """Raised when an MCP protocol error occurs."""
+
+    pass
+
+
+class MCPClient:
+    """
+    Async HTTP client for the MCP Streamable HTTP transport.
+
+    Communicates with MCP servers using JSON-RPC 2.0 over HTTP POST.
+    Supports optional Bearer token authentication.
+    """
+
+    def __init__(
+        self,
+        server_url: str,
+        auth_token: str | None = None,
+    ):
+        self.server_url = server_url.rstrip("/")
+        self.auth_token = auth_token
+        self._request_id = 0
+        self._session_id: str | None = None
+
+    def _next_id(self) -> int:
+        self._request_id += 1
+        return self._request_id
+
+    def _build_headers(self) -> dict[str, str]:
+        headers = {
+            "Content-Type": "application/json",
+            "Accept": "application/json, text/event-stream",
+        }
+        if self.auth_token:
+            headers["Authorization"] = f"Bearer {self.auth_token}"
+        if self._session_id:
+            headers["Mcp-Session-Id"] = self._session_id
+        return headers
+
+    def _build_jsonrpc_request(
+        self, method: str, params: dict[str, Any] | None = None
+    ) -> dict[str, Any]:
+        req: dict[str, Any] = {
+            "jsonrpc": "2.0",
+            "method": method,
+            "id": self._next_id(),
+        }
+        if params is not None:
+            req["params"] = params
+        return req
+
+    @staticmethod
+    def _parse_sse_response(text: str) -> dict[str, Any]:
+        """Parse an SSE (text/event-stream) response body into JSON-RPC data.
+
+        MCP servers may return responses as SSE with format:
+            event: message
+            data: {"jsonrpc":"2.0","result":{...},"id":1}
+
+        We extract the last `data:` line that contains a JSON-RPC response
+        (i.e. has an "id" field), which is the reply to our request.
+        """
+        last_data: dict[str, Any] | None = None
+        for line in text.splitlines():
+            stripped = line.strip()
+            if stripped.startswith("data:"):
+                payload = stripped[len("data:") :].strip()
+                if not payload:
+                    continue
+                try:
+                    parsed = json.loads(payload)
+                    # Only keep JSON-RPC responses (have "id"), skip notifications
+                    if isinstance(parsed, dict) and "id" in parsed:
+                        last_data = parsed
+                except (json.JSONDecodeError, ValueError):
+                    continue
+        if last_data is None:
+            raise MCPClientError("No JSON-RPC response found in SSE stream")
+        return last_data
+
+    async def _send_request(
+        self, method: str, params: dict[str, Any] | None = None
+    ) -> Any:
+        """Send a JSON-RPC request to the MCP server and return the result.
+
+        Handles both ``application/json`` and ``text/event-stream`` responses
+        as required by the MCP Streamable HTTP transport specification.
+        """
+        payload = self._build_jsonrpc_request(method, params)
+        headers = self._build_headers()
+
+        requests = Requests(
+            raise_for_status=True,
+            extra_headers=headers,
+        )
+        response = await requests.post(self.server_url, json=payload)
+
+        # Capture session ID from response (MCP Streamable HTTP transport)
+        session_id = response.headers.get("Mcp-Session-Id")
+        if session_id:
+            self._session_id = session_id
+
+        content_type = response.headers.get("content-type", "")
+        if "text/event-stream" in content_type:
+            body = self._parse_sse_response(response.text())
+        else:
+            try:
+                body = response.json()
+            except Exception as e:
+                raise MCPClientError(
+                    f"MCP server returned non-JSON response: {e}"
+                ) from e
+
+        if not isinstance(body, dict):
+            raise MCPClientError(
+                f"MCP server returned unexpected JSON type: {type(body).__name__}"
+            )
+
+        # Handle JSON-RPC error
+        if "error" in body:
+            error = body["error"]
+            if isinstance(error, dict):
+                raise MCPClientError(
+                    f"MCP server error [{error.get('code', '?')}]: "
+                    f"{error.get('message', 'Unknown error')}"
+                )
+            raise MCPClientError(f"MCP server error: {error}")
+
+        return body.get("result")
+
+    async def _send_notification(self, method: str) -> None:
+        """Send a JSON-RPC notification (no id, no response expected)."""
+        headers = self._build_headers()
+        notification = {"jsonrpc": "2.0", "method": method}
+        requests = Requests(
+            raise_for_status=False,
+            extra_headers=headers,
+        )
+        await requests.post(self.server_url, json=notification)
+
+    async def discover_auth(self) -> dict[str, Any] | None:
+        """Probe the MCP server's OAuth metadata (RFC 9728 / MCP spec).
+
+        Returns ``None`` if the server doesn't require auth, otherwise returns
+        a dict with:
+          - ``authorization_servers``: list of authorization server URLs
+          - ``resource``: the resource indicator URL (usually the MCP endpoint)
+          - ``scopes_supported``: optional list of supported scopes
+
+        The caller can then fetch the authorization server metadata to get
+        ``authorization_endpoint``, ``token_endpoint``, etc.
+        """
+        from urllib.parse import urlparse
+
+        parsed = urlparse(self.server_url)
+        base = f"{parsed.scheme}://{parsed.netloc}"
+
+        # Build candidates for protected-resource metadata (per RFC 9728)
+        path = parsed.path.rstrip("/")
+        candidates = []
+        if path and path != "/":
+            candidates.append(f"{base}/.well-known/oauth-protected-resource{path}")
+        candidates.append(f"{base}/.well-known/oauth-protected-resource")
+
+        requests = Requests(
+            raise_for_status=False,
+        )
+        for url in candidates:
+            try:
+                resp = await requests.get(url)
+                if resp.status == 200:
+                    data = resp.json()
+                    if isinstance(data, dict) and "authorization_servers" in data:
+                        return data
+            except Exception:
+                continue
+
+        return None
+
+    async def discover_auth_server_metadata(
+        self, auth_server_url: str
+    ) -> dict[str, Any] | None:
+        """Fetch the OAuth Authorization Server Metadata (RFC 8414).
+
+        Given an authorization server URL, returns a dict with:
+          - ``authorization_endpoint``
+          - ``token_endpoint``
+          - ``registration_endpoint`` (for dynamic client registration)
+          - ``scopes_supported``
+          - ``code_challenge_methods_supported``
+          - etc.
+        """
+        from urllib.parse import urlparse
+
+        parsed = urlparse(auth_server_url)
+        base = f"{parsed.scheme}://{parsed.netloc}"
+        path = parsed.path.rstrip("/")
+
+        # Try standard metadata endpoints (RFC 8414 and OpenID Connect)
+        candidates = []
+        if path and path != "/":
+            candidates.append(f"{base}/.well-known/oauth-authorization-server{path}")
+        candidates.append(f"{base}/.well-known/oauth-authorization-server")
+        candidates.append(f"{base}/.well-known/openid-configuration")
+
+        requests = Requests(
+            raise_for_status=False,
+        )
+        for url in candidates:
+            try:
+                resp = await requests.get(url)
+                if resp.status == 200:
+                    data = resp.json()
+                    if isinstance(data, dict) and "authorization_endpoint" in data:
+                        return data
+            except Exception:
+                continue
+
+        return None
+
+    async def initialize(self) -> dict[str, Any]:
+        """
+        Send the MCP initialize request.
+
+        This is required by the MCP protocol before any other requests.
+        Returns the server's capabilities.
+        """
+        result = await self._send_request(
+            "initialize",
+            {
+                "protocolVersion": "2025-03-26",
+                "capabilities": {},
+                "clientInfo": {"name": "AutoGPT-Platform", "version": "1.0.0"},
+            },
+        )
+        # Send initialized notification (no response expected)
+        await self._send_notification("notifications/initialized")
+
+        return result or {}
+
+    async def list_tools(self) -> list[MCPTool]:
+        """
+        Discover available tools from the MCP server.
+
+        Returns a list of MCPTool objects with name, description, and input schema.
+        """
+        result = await self._send_request("tools/list")
+        if not result or "tools" not in result:
+            return []
+
+        tools = []
+        for tool_data in result["tools"]:
+            tools.append(
+                MCPTool(
+                    name=tool_data.get("name", ""),
+                    description=tool_data.get("description", ""),
+                    input_schema=tool_data.get("inputSchema", {}),
+                )
+            )
+        return tools
+
+    async def call_tool(
+        self, tool_name: str, arguments: dict[str, Any]
+    ) -> MCPCallResult:
+        """
+        Call a tool on the MCP server.
+
+        Args:
+            tool_name: The name of the tool to call.
+            arguments: The arguments to pass to the tool.
+
+        Returns:
+            MCPCallResult with the tool's response content.
+        """
+        result = await self._send_request(
+            "tools/call",
+            {"name": tool_name, "arguments": arguments},
+        )
+        if not result:
+            return MCPCallResult(is_error=True)
+
+        return MCPCallResult(
+            content=result.get("content", []),
+            is_error=result.get("isError", False),
+        )

autogpt_platform/backend/backend/blocks/mcp/oauth.py

@@ -0,0 +1,204 @@
+"""
+MCP OAuth handler for MCP servers that use OAuth 2.1 authorization.
+
+Unlike other OAuth handlers (GitHub, Google, etc.) where endpoints are fixed,
+MCP servers have dynamic endpoints discovered via RFC 9728 / RFC 8414 metadata.
+This handler accepts those endpoints at construction time.
+"""
+
+import logging
+import time
+import urllib.parse
+from typing import ClassVar, Optional
+
+from pydantic import SecretStr
+
+from backend.data.model import OAuth2Credentials
+from backend.integrations.oauth.base import BaseOAuthHandler
+from backend.integrations.providers import ProviderName
+from backend.util.request import Requests
+
+logger = logging.getLogger(__name__)
+
+
+class MCPOAuthHandler(BaseOAuthHandler):
+    """
+    OAuth handler for MCP servers with dynamically-discovered endpoints.
+
+    Construction requires the authorization and token endpoint URLs,
+    which are obtained via MCP OAuth metadata discovery
+    (``MCPClient.discover_auth`` + ``discover_auth_server_metadata``).
+    """
+
+    PROVIDER_NAME: ClassVar[ProviderName | str] = ProviderName.MCP
+    DEFAULT_SCOPES: ClassVar[list[str]] = []
+
+    def __init__(
+        self,
+        client_id: str,
+        client_secret: str,
+        redirect_uri: str,
+        *,
+        authorize_url: str,
+        token_url: str,
+        revoke_url: str | None = None,
+        resource_url: str | None = None,
+    ):
+        self.client_id = client_id
+        self.client_secret = client_secret
+        self.redirect_uri = redirect_uri
+        self.authorize_url = authorize_url
+        self.token_url = token_url
+        self.revoke_url = revoke_url
+        self.resource_url = resource_url
+
+    def get_login_url(
+        self,
+        scopes: list[str],
+        state: str,
+        code_challenge: Optional[str],
+    ) -> str:
+        scopes = self.handle_default_scopes(scopes)
+
+        params: dict[str, str] = {
+            "response_type": "code",
+            "client_id": self.client_id,
+            "redirect_uri": self.redirect_uri,
+            "state": state,
+        }
+        if scopes:
+            params["scope"] = " ".join(scopes)
+        # PKCE (S256) — included when the caller provides a code_challenge
+        if code_challenge:
+            params["code_challenge"] = code_challenge
+            params["code_challenge_method"] = "S256"
+        # MCP spec requires resource indicator (RFC 8707)
+        if self.resource_url:
+            params["resource"] = self.resource_url
+
+        return f"{self.authorize_url}?{urllib.parse.urlencode(params)}"
+
+    async def exchange_code_for_tokens(
+        self,
+        code: str,
+        scopes: list[str],
+        code_verifier: Optional[str],
+    ) -> OAuth2Credentials:
+        data: dict[str, str] = {
+            "grant_type": "authorization_code",
+            "code": code,
+            "redirect_uri": self.redirect_uri,
+            "client_id": self.client_id,
+        }
+        if self.client_secret:
+            data["client_secret"] = self.client_secret
+        if code_verifier:
+            data["code_verifier"] = code_verifier
+        if self.resource_url:
+            data["resource"] = self.resource_url
+
+        response = await Requests(raise_for_status=True).post(
+            self.token_url,
+            data=data,
+            headers={"Content-Type": "application/x-www-form-urlencoded"},
+        )
+        tokens = response.json()
+
+        if "error" in tokens:
+            raise RuntimeError(
+                f"Token exchange failed: {tokens.get('error_description', tokens['error'])}"
+            )
+
+        if "access_token" not in tokens:
+            raise RuntimeError("OAuth token response missing 'access_token' field")
+
+        now = int(time.time())
+        expires_in = tokens.get("expires_in")
+
+        return OAuth2Credentials(
+            provider=self.PROVIDER_NAME,
+            title=None,
+            access_token=SecretStr(tokens["access_token"]),
+            refresh_token=(
+                SecretStr(tokens["refresh_token"])
+                if tokens.get("refresh_token")
+                else None
+            ),
+            access_token_expires_at=now + expires_in if expires_in else None,
+            refresh_token_expires_at=None,
+            scopes=scopes,
+            metadata={
+                "mcp_token_url": self.token_url,
+                "mcp_resource_url": self.resource_url,
+            },
+        )
+
+    async def _refresh_tokens(
+        self, credentials: OAuth2Credentials
+    ) -> OAuth2Credentials:
+        if not credentials.refresh_token:
+            raise ValueError("No refresh token available for MCP OAuth credentials")
+
+        data: dict[str, str] = {
+            "grant_type": "refresh_token",
+            "refresh_token": credentials.refresh_token.get_secret_value(),
+            "client_id": self.client_id,
+        }
+        if self.client_secret:
+            data["client_secret"] = self.client_secret
+        if self.resource_url:
+            data["resource"] = self.resource_url
+
+        response = await Requests(raise_for_status=True).post(
+            self.token_url,
+            data=data,
+            headers={"Content-Type": "application/x-www-form-urlencoded"},
+        )
+        tokens = response.json()
+
+        if "error" in tokens:
+            raise RuntimeError(
+                f"Token refresh failed: {tokens.get('error_description', tokens['error'])}"
+            )
+
+        if "access_token" not in tokens:
+            raise RuntimeError("OAuth refresh response missing 'access_token' field")
+
+        now = int(time.time())
+        expires_in = tokens.get("expires_in")
+
+        return OAuth2Credentials(
+            id=credentials.id,
+            provider=self.PROVIDER_NAME,
+            title=credentials.title,
+            access_token=SecretStr(tokens["access_token"]),
+            refresh_token=(
+                SecretStr(tokens["refresh_token"])
+                if tokens.get("refresh_token")
+                else credentials.refresh_token
+            ),
+            access_token_expires_at=now + expires_in if expires_in else None,
+            refresh_token_expires_at=credentials.refresh_token_expires_at,
+            scopes=credentials.scopes,
+            metadata=credentials.metadata,
+        )
+
+    async def revoke_tokens(self, credentials: OAuth2Credentials) -> bool:
+        if not self.revoke_url:
+            return False
+
+        try:
+            data = {
+                "token": credentials.access_token.get_secret_value(),
+                "token_type_hint": "access_token",
+                "client_id": self.client_id,
+            }
+            await Requests().post(
+                self.revoke_url,
+                data=data,
+                headers={"Content-Type": "application/x-www-form-urlencoded"},
+            )
+            return True
+        except Exception:
+            logger.warning("Failed to revoke MCP OAuth tokens", exc_info=True)
+            return False

autogpt_platform/backend/backend/blocks/mcp/test_e2e.py

@@ -0,0 +1,109 @@
+"""
+End-to-end tests against a real public MCP server.
+
+These tests hit the OpenAI docs MCP server (https://developers.openai.com/mcp)
+which is publicly accessible without authentication and returns SSE responses.
+
+Mark: These are tagged with ``@pytest.mark.e2e`` so they can be run/skipped
+independently of the rest of the test suite (they require network access).
+"""
+
+import json
+import os
+
+import pytest
+
+from backend.blocks.mcp.client import MCPClient
+
+# Public MCP server that requires no authentication
+OPENAI_DOCS_MCP_URL = "https://developers.openai.com/mcp"
+
+# Skip all tests in this module unless RUN_E2E env var is set
+pytestmark = pytest.mark.skipif(
+    not os.environ.get("RUN_E2E"), reason="set RUN_E2E=1 to run e2e tests"
+)
+
+
+class TestRealMCPServer:
+    """Tests against the live OpenAI docs MCP server."""
+
+    @pytest.mark.asyncio(loop_scope="session")
+    async def test_initialize(self):
+        """Verify we can complete the MCP handshake with a real server."""
+        client = MCPClient(OPENAI_DOCS_MCP_URL)
+        result = await client.initialize()
+
+        assert result["protocolVersion"] == "2025-03-26"
+        assert "serverInfo" in result
+        assert result["serverInfo"]["name"] == "openai-docs-mcp"
+        assert "tools" in result.get("capabilities", {})
+
+    @pytest.mark.asyncio(loop_scope="session")
+    async def test_list_tools(self):
+        """Verify we can discover tools from a real MCP server."""
+        client = MCPClient(OPENAI_DOCS_MCP_URL)
+        await client.initialize()
+        tools = await client.list_tools()
+
+        assert len(tools) >= 3  # server has at least 5 tools as of writing
+
+        tool_names = {t.name for t in tools}
+        # These tools are documented and should be stable
+        assert "search_openai_docs" in tool_names
+        assert "list_openai_docs" in tool_names
+        assert "fetch_openai_doc" in tool_names
+
+        # Verify schema structure
+        search_tool = next(t for t in tools if t.name == "search_openai_docs")
+        assert "query" in search_tool.input_schema.get("properties", {})
+        assert "query" in search_tool.input_schema.get("required", [])
+
+    @pytest.mark.asyncio(loop_scope="session")
+    async def test_call_tool_list_api_endpoints(self):
+        """Call the list_api_endpoints tool and verify we get real data."""
+        client = MCPClient(OPENAI_DOCS_MCP_URL)
+        await client.initialize()
+        result = await client.call_tool("list_api_endpoints", {})
+
+        assert not result.is_error
+        assert len(result.content) >= 1
+        assert result.content[0]["type"] == "text"
+
+        data = json.loads(result.content[0]["text"])
+        assert "paths" in data or "urls" in data
+        # The OpenAI API should have many endpoints
+        total = data.get("total", len(data.get("paths", [])))
+        assert total > 50
+
+    @pytest.mark.asyncio(loop_scope="session")
+    async def test_call_tool_search(self):
+        """Search for docs and verify we get results."""
+        client = MCPClient(OPENAI_DOCS_MCP_URL)
+        await client.initialize()
+        result = await client.call_tool(
+            "search_openai_docs", {"query": "chat completions", "limit": 3}
+        )
+
+        assert not result.is_error
+        assert len(result.content) >= 1
+
+    @pytest.mark.asyncio(loop_scope="session")
+    async def test_sse_response_handling(self):
+        """Verify the client correctly handles SSE responses from a real server.
+
+        This is the key test — our local test server returns JSON,
+        but real MCP servers typically return SSE. This proves the
+        SSE parsing works end-to-end.
+        """
+        client = MCPClient(OPENAI_DOCS_MCP_URL)
+        # initialize() internally calls _send_request which must parse SSE
+        result = await client.initialize()
+
+        # If we got here without error, SSE parsing works
+        assert isinstance(result, dict)
+        assert "protocolVersion" in result
+
+        # Also verify list_tools works (another SSE response)
+        tools = await client.list_tools()
+        assert len(tools) > 0
+        assert all(hasattr(t, "name") for t in tools)

autogpt_platform/backend/backend/blocks/mcp/test_integration.py

@@ -0,0 +1,389 @@
+"""
+Integration tests for MCP client and MCPToolBlock against a real HTTP server.
+
+These tests spin up a local MCP test server and run the full client/block flow
+against it — no mocking, real HTTP requests.
+"""
+
+import asyncio
+import json
+import threading
+from unittest.mock import patch
+
+import pytest
+from aiohttp import web
+from pydantic import SecretStr
+
+from backend.blocks.mcp.block import MCPToolBlock
+from backend.blocks.mcp.client import MCPClient
+from backend.blocks.mcp.test_server import create_test_mcp_app
+from backend.data.model import OAuth2Credentials
+
+MOCK_USER_ID = "test-user-integration"
+
+
+class _MCPTestServer:
+    """
+    Run an MCP test server in a background thread with its own event loop.
+    This avoids event loop conflicts with pytest-asyncio.
+    """
+
+    def __init__(self, auth_token: str | None = None):
+        self.auth_token = auth_token
+        self.url: str = ""
+        self._runner: web.AppRunner | None = None
+        self._loop: asyncio.AbstractEventLoop | None = None
+        self._thread: threading.Thread | None = None
+        self._started = threading.Event()
+
+    def _run(self):
+        self._loop = asyncio.new_event_loop()
+        asyncio.set_event_loop(self._loop)
+        self._loop.run_until_complete(self._start())
+        self._started.set()
+        self._loop.run_forever()
+
+    async def _start(self):
+        app = create_test_mcp_app(auth_token=self.auth_token)
+        self._runner = web.AppRunner(app)
+        await self._runner.setup()
+        site = web.TCPSite(self._runner, "127.0.0.1", 0)
+        await site.start()
+        port = site._server.sockets[0].getsockname()[1]  # type: ignore[union-attr]
+        self.url = f"http://127.0.0.1:{port}/mcp"
+
+    def start(self):
+        self._thread = threading.Thread(target=self._run, daemon=True)
+        self._thread.start()
+        if not self._started.wait(timeout=5):
+            raise RuntimeError("MCP test server failed to start within 5 seconds")
+        return self
+
+    def stop(self):
+        if self._loop and self._runner:
+            asyncio.run_coroutine_threadsafe(self._runner.cleanup(), self._loop).result(
+                timeout=5
+            )
+            self._loop.call_soon_threadsafe(self._loop.stop)
+        if self._thread:
+            self._thread.join(timeout=5)
+
+
+@pytest.fixture(scope="module")
+def mcp_server():
+    """Start a local MCP test server in a background thread."""
+    server = _MCPTestServer()
+    server.start()
+    yield server.url
+    server.stop()
+
+
+@pytest.fixture(scope="module")
+def mcp_server_with_auth():
+    """Start a local MCP test server with auth in a background thread."""
+    server = _MCPTestServer(auth_token="test-secret-token")
+    server.start()
+    yield server.url, "test-secret-token"
+    server.stop()
+
+
+@pytest.fixture(autouse=True)
+def _allow_localhost():
+    """
+    Allow 127.0.0.1 through SSRF protection for integration tests.
+
+    The Requests class blocks private IPs by default. We patch the Requests
+    constructor to always include 127.0.0.1 as a trusted origin so the local
+    test server is reachable.
+    """
+    from backend.util.request import Requests
+
+    original_init = Requests.__init__
+
+    def patched_init(self, *args, **kwargs):
+        trusted = list(kwargs.get("trusted_origins") or [])
+        trusted.append("http://127.0.0.1")
+        kwargs["trusted_origins"] = trusted
+        original_init(self, *args, **kwargs)
+
+    with patch.object(Requests, "__init__", patched_init):
+        yield
+
+
+def _make_client(url: str, auth_token: str | None = None) -> MCPClient:
+    """Create an MCPClient for integration tests."""
+    return MCPClient(url, auth_token=auth_token)
+
+
+# ── MCPClient integration tests ──────────────────────────────────────
+
+
+class TestMCPClientIntegration:
+    """Test MCPClient against a real local MCP server."""
+
+    @pytest.mark.asyncio(loop_scope="session")
+    async def test_initialize(self, mcp_server):
+        client = _make_client(mcp_server)
+        result = await client.initialize()
+
+        assert result["protocolVersion"] == "2025-03-26"
+        assert result["serverInfo"]["name"] == "test-mcp-server"
+        assert "tools" in result["capabilities"]
+
+    @pytest.mark.asyncio(loop_scope="session")
+    async def test_list_tools(self, mcp_server):
+        client = _make_client(mcp_server)
+        await client.initialize()
+        tools = await client.list_tools()
+
+        assert len(tools) == 3
+
+        tool_names = {t.name for t in tools}
+        assert tool_names == {"get_weather", "add_numbers", "echo"}
+
+        # Check get_weather schema
+        weather = next(t for t in tools if t.name == "get_weather")
+        assert weather.description == "Get current weather for a city"
+        assert "city" in weather.input_schema["properties"]
+        assert weather.input_schema["required"] == ["city"]
+
+        # Check add_numbers schema
+        add = next(t for t in tools if t.name == "add_numbers")
+        assert "a" in add.input_schema["properties"]
+        assert "b" in add.input_schema["properties"]
+
+    @pytest.mark.asyncio(loop_scope="session")
+    async def test_call_tool_get_weather(self, mcp_server):
+        client = _make_client(mcp_server)
+        await client.initialize()
+        result = await client.call_tool("get_weather", {"city": "London"})
+
+        assert not result.is_error
+        assert len(result.content) == 1
+        assert result.content[0]["type"] == "text"
+
+        data = json.loads(result.content[0]["text"])
+        assert data["city"] == "London"
+        assert data["temperature"] == 22
+        assert data["condition"] == "sunny"
+
+    @pytest.mark.asyncio(loop_scope="session")
+    async def test_call_tool_add_numbers(self, mcp_server):
+        client = _make_client(mcp_server)
+        await client.initialize()
+        result = await client.call_tool("add_numbers", {"a": 3, "b": 7})
+
+        assert not result.is_error
+        data = json.loads(result.content[0]["text"])
+        assert data["result"] == 10
+
+    @pytest.mark.asyncio(loop_scope="session")
+    async def test_call_tool_echo(self, mcp_server):
+        client = _make_client(mcp_server)
+        await client.initialize()
+        result = await client.call_tool("echo", {"message": "Hello MCP!"})
+
+        assert not result.is_error
+        assert result.content[0]["text"] == "Hello MCP!"
+
+    @pytest.mark.asyncio(loop_scope="session")
+    async def test_call_unknown_tool(self, mcp_server):
+        client = _make_client(mcp_server)
+        await client.initialize()
+        result = await client.call_tool("nonexistent_tool", {})
+
+        assert result.is_error
+        assert "Unknown tool" in result.content[0]["text"]
+
+    @pytest.mark.asyncio(loop_scope="session")
+    async def test_auth_success(self, mcp_server_with_auth):
+        url, token = mcp_server_with_auth
+        client = _make_client(url, auth_token=token)
+        result = await client.initialize()
+
+        assert result["protocolVersion"] == "2025-03-26"
+
+        tools = await client.list_tools()
+        assert len(tools) == 3
+
+    @pytest.mark.asyncio(loop_scope="session")
+    async def test_auth_failure(self, mcp_server_with_auth):
+        url, _ = mcp_server_with_auth
+        client = _make_client(url, auth_token="wrong-token")
+
+        with pytest.raises(Exception):
+            await client.initialize()
+
+    @pytest.mark.asyncio(loop_scope="session")
+    async def test_auth_missing(self, mcp_server_with_auth):
+        url, _ = mcp_server_with_auth
+        client = _make_client(url)
+
+        with pytest.raises(Exception):
+            await client.initialize()
+
+
+# ── MCPToolBlock integration tests ───────────────────────────────────
+
+
+class TestMCPToolBlockIntegration:
+    """Test MCPToolBlock end-to-end against a real local MCP server."""
+
+    @pytest.mark.asyncio(loop_scope="session")
+    async def test_full_flow_get_weather(self, mcp_server):
+        """Full flow: discover tools, select one, execute it."""
+        # Step 1: Discover tools (simulating what the frontend/API would do)
+        client = _make_client(mcp_server)
+        await client.initialize()
+        tools = await client.list_tools()
+        assert len(tools) == 3
+
+        # Step 2: User selects "get_weather" and we get its schema
+        weather_tool = next(t for t in tools if t.name == "get_weather")
+
+        # Step 3: Execute the block — no credentials (public server)
+        block = MCPToolBlock()
+        input_data = MCPToolBlock.Input(
+            server_url=mcp_server,
+            selected_tool="get_weather",
+            tool_input_schema=weather_tool.input_schema,
+            tool_arguments={"city": "Paris"},
+        )
+
+        outputs = []
+        async for name, data in block.run(input_data, user_id=MOCK_USER_ID):
+            outputs.append((name, data))
+
+        assert len(outputs) == 1
+        assert outputs[0][0] == "result"
+        result = outputs[0][1]
+        assert result["city"] == "Paris"
+        assert result["temperature"] == 22
+        assert result["condition"] == "sunny"
+
+    @pytest.mark.asyncio(loop_scope="session")
+    async def test_full_flow_add_numbers(self, mcp_server):
+        """Full flow for add_numbers tool."""
+        client = _make_client(mcp_server)
+        await client.initialize()
+        tools = await client.list_tools()
+        add_tool = next(t for t in tools if t.name == "add_numbers")
+
+        block = MCPToolBlock()
+        input_data = MCPToolBlock.Input(
+            server_url=mcp_server,
+            selected_tool="add_numbers",
+            tool_input_schema=add_tool.input_schema,
+            tool_arguments={"a": 42, "b": 58},
+        )
+
+        outputs = []
+        async for name, data in block.run(input_data, user_id=MOCK_USER_ID):
+            outputs.append((name, data))
+
+        assert len(outputs) == 1
+        assert outputs[0][0] == "result"
+        assert outputs[0][1]["result"] == 100
+
+    @pytest.mark.asyncio(loop_scope="session")
+    async def test_full_flow_echo_plain_text(self, mcp_server):
+        """Verify plain text (non-JSON) responses work."""
+        block = MCPToolBlock()
+        input_data = MCPToolBlock.Input(
+            server_url=mcp_server,
+            selected_tool="echo",
+            tool_input_schema={
+                "type": "object",
+                "properties": {"message": {"type": "string"}},
+                "required": ["message"],
+            },
+            tool_arguments={"message": "Hello from AutoGPT!"},
+        )
+
+        outputs = []
+        async for name, data in block.run(input_data, user_id=MOCK_USER_ID):
+            outputs.append((name, data))
+
+        assert len(outputs) == 1
+        assert outputs[0][0] == "result"
+        assert outputs[0][1] == "Hello from AutoGPT!"
+
+    @pytest.mark.asyncio(loop_scope="session")
+    async def test_full_flow_unknown_tool_yields_error(self, mcp_server):
+        """Calling an unknown tool should yield an error output."""
+        block = MCPToolBlock()
+        input_data = MCPToolBlock.Input(
+            server_url=mcp_server,
+            selected_tool="nonexistent_tool",
+            tool_arguments={},
+        )
+
+        outputs = []
+        async for name, data in block.run(input_data, user_id=MOCK_USER_ID):
+            outputs.append((name, data))
+
+        assert len(outputs) == 1
+        assert outputs[0][0] == "error"
+        assert "returned an error" in outputs[0][1]
+
+    @pytest.mark.asyncio(loop_scope="session")
+    async def test_full_flow_with_auth(self, mcp_server_with_auth):
+        """Full flow with authentication via credentials kwarg."""
+        url, token = mcp_server_with_auth
+
+        block = MCPToolBlock()
+        input_data = MCPToolBlock.Input(
+            server_url=url,
+            selected_tool="echo",
+            tool_input_schema={
+                "type": "object",
+                "properties": {"message": {"type": "string"}},
+                "required": ["message"],
+            },
+            tool_arguments={"message": "Authenticated!"},
+        )
+
+        # Pass credentials via the standard kwarg (as the executor would)
+        test_creds = OAuth2Credentials(
+            id="test-cred",
+            provider="mcp",
+            access_token=SecretStr(token),
+            refresh_token=SecretStr(""),
+            scopes=[],
+            title="Test MCP credential",
+        )
+
+        outputs = []
+        async for name, data in block.run(
+            input_data, user_id=MOCK_USER_ID, credentials=test_creds
+        ):
+            outputs.append((name, data))
+
+        assert len(outputs) == 1
+        assert outputs[0][0] == "result"
+        assert outputs[0][1] == "Authenticated!"
+
+    @pytest.mark.asyncio(loop_scope="session")
+    async def test_no_credentials_runs_without_auth(self, mcp_server):
+        """Block runs without auth when no credentials are provided."""
+        block = MCPToolBlock()
+        input_data = MCPToolBlock.Input(
+            server_url=mcp_server,
+            selected_tool="echo",
+            tool_input_schema={
+                "type": "object",
+                "properties": {"message": {"type": "string"}},
+                "required": ["message"],
+            },
+            tool_arguments={"message": "No auth needed"},
+        )
+
+        outputs = []
+        async for name, data in block.run(
+            input_data, user_id=MOCK_USER_ID, credentials=None
+        ):
+            outputs.append((name, data))
+
+        assert len(outputs) == 1
+        assert outputs[0][0] == "result"
+        assert outputs[0][1] == "No auth needed"

autogpt_platform/backend/backend/blocks/mcp/test_mcp.py

@@ -0,0 +1,619 @@
+"""
+Tests for MCP client and MCPToolBlock.
+"""
+
+import json
+from unittest.mock import AsyncMock, patch
+
+import pytest
+
+from backend.blocks.mcp.block import MCPToolBlock
+from backend.blocks.mcp.client import MCPCallResult, MCPClient, MCPClientError
+from backend.util.test import execute_block_test
+
+# ── SSE parsing unit tests ───────────────────────────────────────────
+
+
+class TestSSEParsing:
+    """Tests for SSE (text/event-stream) response parsing."""
+
+    def test_parse_sse_simple(self):
+        sse = (
+            "event: message\n"
+            'data: {"jsonrpc":"2.0","result":{"tools":[]},"id":1}\n'
+            "\n"
+        )
+        body = MCPClient._parse_sse_response(sse)
+        assert body["result"] == {"tools": []}
+        assert body["id"] == 1
+
+    def test_parse_sse_with_notifications(self):
+        """SSE streams can contain notifications (no id) before the response."""
+        sse = (
+            "event: message\n"
+            'data: {"jsonrpc":"2.0","method":"some/notification"}\n'
+            "\n"
+            "event: message\n"
+            'data: {"jsonrpc":"2.0","result":{"ok":true},"id":2}\n'
+            "\n"
+        )
+        body = MCPClient._parse_sse_response(sse)
+        assert body["result"] == {"ok": True}
+        assert body["id"] == 2
+
+    def test_parse_sse_error_response(self):
+        sse = (
+            "event: message\n"
+            'data: {"jsonrpc":"2.0","error":{"code":-32600,"message":"Bad Request"},"id":1}\n'
+        )
+        body = MCPClient._parse_sse_response(sse)
+        assert "error" in body
+        assert body["error"]["code"] == -32600
+
+    def test_parse_sse_no_data_raises(self):
+        with pytest.raises(MCPClientError, match="No JSON-RPC response found"):
+            MCPClient._parse_sse_response("event: message\n\n")
+
+    def test_parse_sse_empty_raises(self):
+        with pytest.raises(MCPClientError, match="No JSON-RPC response found"):
+            MCPClient._parse_sse_response("")
+
+    def test_parse_sse_ignores_non_data_lines(self):
+        sse = (
+            ": comment line\n"
+            "event: message\n"
+            "id: 123\n"
+            'data: {"jsonrpc":"2.0","result":"ok","id":1}\n'
+            "\n"
+        )
+        body = MCPClient._parse_sse_response(sse)
+        assert body["result"] == "ok"
+
+    def test_parse_sse_uses_last_response(self):
+        """If multiple responses exist, use the last one."""
+        sse = (
+            'data: {"jsonrpc":"2.0","result":"first","id":1}\n'
+            "\n"
+            'data: {"jsonrpc":"2.0","result":"second","id":2}\n'
+            "\n"
+        )
+        body = MCPClient._parse_sse_response(sse)
+        assert body["result"] == "second"
+
+
+# ── MCPClient unit tests ─────────────────────────────────────────────
+
+
+class TestMCPClient:
+    """Tests for the MCP HTTP client."""
+
+    def test_build_headers_without_auth(self):
+        client = MCPClient("https://mcp.example.com")
+        headers = client._build_headers()
+        assert "Authorization" not in headers
+        assert headers["Content-Type"] == "application/json"
+
+    def test_build_headers_with_auth(self):
+        client = MCPClient("https://mcp.example.com", auth_token="my-token")
+        headers = client._build_headers()
+        assert headers["Authorization"] == "Bearer my-token"
+
+    def test_build_jsonrpc_request(self):
+        client = MCPClient("https://mcp.example.com")
+        req = client._build_jsonrpc_request("tools/list")
+        assert req["jsonrpc"] == "2.0"
+        assert req["method"] == "tools/list"
+        assert "id" in req
+        assert "params" not in req
+
+    def test_build_jsonrpc_request_with_params(self):
+        client = MCPClient("https://mcp.example.com")
+        req = client._build_jsonrpc_request(
+            "tools/call", {"name": "test", "arguments": {"x": 1}}
+        )
+        assert req["params"] == {"name": "test", "arguments": {"x": 1}}
+
+    def test_request_id_increments(self):
+        client = MCPClient("https://mcp.example.com")
+        req1 = client._build_jsonrpc_request("tools/list")
+        req2 = client._build_jsonrpc_request("tools/list")
+        assert req2["id"] > req1["id"]
+
+    def test_server_url_trailing_slash_stripped(self):
+        client = MCPClient("https://mcp.example.com/mcp/")
+        assert client.server_url == "https://mcp.example.com/mcp"
+
+    @pytest.mark.asyncio(loop_scope="session")
+    async def test_send_request_success(self):
+        client = MCPClient("https://mcp.example.com")
+
+        mock_response = AsyncMock()
+        mock_response.json.return_value = {
+            "jsonrpc": "2.0",
+            "result": {"tools": []},
+            "id": 1,
+        }
+
+        with patch.object(client, "_send_request", return_value={"tools": []}):
+            result = await client._send_request("tools/list")
+            assert result == {"tools": []}
+
+    @pytest.mark.asyncio(loop_scope="session")
+    async def test_send_request_error(self):
+        client = MCPClient("https://mcp.example.com")
+
+        async def mock_send(*args, **kwargs):
+            raise MCPClientError("MCP server error [-32600]: Invalid Request")
+
+        with patch.object(client, "_send_request", side_effect=mock_send):
+            with pytest.raises(MCPClientError, match="Invalid Request"):
+                await client._send_request("tools/list")
+
+    @pytest.mark.asyncio(loop_scope="session")
+    async def test_list_tools(self):
+        client = MCPClient("https://mcp.example.com")
+
+        mock_result = {
+            "tools": [
+                {
+                    "name": "get_weather",
+                    "description": "Get current weather for a city",
+                    "inputSchema": {
+                        "type": "object",
+                        "properties": {"city": {"type": "string"}},
+                        "required": ["city"],
+                    },
+                },
+                {
+                    "name": "search",
+                    "description": "Search the web",
+                    "inputSchema": {
+                        "type": "object",
+                        "properties": {"query": {"type": "string"}},
+                        "required": ["query"],
+                    },
+                },
+            ]
+        }
+
+        with patch.object(client, "_send_request", return_value=mock_result):
+            tools = await client.list_tools()
+
+        assert len(tools) == 2
+        assert tools[0].name == "get_weather"
+        assert tools[0].description == "Get current weather for a city"
+        assert tools[0].input_schema["properties"]["city"]["type"] == "string"
+        assert tools[1].name == "search"
+
+    @pytest.mark.asyncio(loop_scope="session")
+    async def test_list_tools_empty(self):
+        client = MCPClient("https://mcp.example.com")
+
+        with patch.object(client, "_send_request", return_value={"tools": []}):
+            tools = await client.list_tools()
+
+        assert tools == []
+
+    @pytest.mark.asyncio(loop_scope="session")
+    async def test_list_tools_none_result(self):
+        client = MCPClient("https://mcp.example.com")
+
+        with patch.object(client, "_send_request", return_value=None):
+            tools = await client.list_tools()
+
+        assert tools == []
+
+    @pytest.mark.asyncio(loop_scope="session")
+    async def test_call_tool_success(self):
+        client = MCPClient("https://mcp.example.com")
+
+        mock_result = {
+            "content": [
+                {"type": "text", "text": json.dumps({"temp": 20, "city": "London"})}
+            ],
+            "isError": False,
+        }
+
+        with patch.object(client, "_send_request", return_value=mock_result):
+            result = await client.call_tool("get_weather", {"city": "London"})
+
+        assert not result.is_error
+        assert len(result.content) == 1
+        assert result.content[0]["type"] == "text"
+
+    @pytest.mark.asyncio(loop_scope="session")
+    async def test_call_tool_error(self):
+        client = MCPClient("https://mcp.example.com")
+
+        mock_result = {
+            "content": [{"type": "text", "text": "City not found"}],
+            "isError": True,
+        }
+
+        with patch.object(client, "_send_request", return_value=mock_result):
+            result = await client.call_tool("get_weather", {"city": "???"})
+
+        assert result.is_error
+
+    @pytest.mark.asyncio(loop_scope="session")
+    async def test_call_tool_none_result(self):
+        client = MCPClient("https://mcp.example.com")
+
+        with patch.object(client, "_send_request", return_value=None):
+            result = await client.call_tool("get_weather", {"city": "London"})
+
+        assert result.is_error
+
+    @pytest.mark.asyncio(loop_scope="session")
+    async def test_initialize(self):
+        client = MCPClient("https://mcp.example.com")
+
+        mock_result = {
+            "protocolVersion": "2025-03-26",
+            "capabilities": {"tools": {}},
+            "serverInfo": {"name": "test-server", "version": "1.0.0"},
+        }
+
+        with (
+            patch.object(client, "_send_request", return_value=mock_result) as mock_req,
+            patch.object(client, "_send_notification") as mock_notif,
+        ):
+            result = await client.initialize()
+
+        mock_req.assert_called_once()
+        mock_notif.assert_called_once_with("notifications/initialized")
+        assert result["protocolVersion"] == "2025-03-26"
+
+
+# ── MCPToolBlock unit tests ──────────────────────────────────────────
+
+MOCK_USER_ID = "test-user-123"
+
+
+class TestMCPToolBlock:
+    """Tests for the MCPToolBlock."""
+
+    def test_block_instantiation(self):
+        block = MCPToolBlock()
+        assert block.id == "a0a4b1c2-d3e4-4f56-a7b8-c9d0e1f2a3b4"
+        assert block.name == "MCPToolBlock"
+
+    def test_input_schema_has_required_fields(self):
+        block = MCPToolBlock()
+        schema = block.input_schema.jsonschema()
+        props = schema.get("properties", {})
+        assert "server_url" in props
+        assert "selected_tool" in props
+        assert "tool_arguments" in props
+        assert "credentials" in props
+
+    def test_output_schema(self):
+        block = MCPToolBlock()
+        schema = block.output_schema.jsonschema()
+        props = schema.get("properties", {})
+        assert "result" in props
+        assert "error" in props
+
+    def test_get_input_schema_with_tool_schema(self):
+        tool_schema = {
+            "type": "object",
+            "properties": {"query": {"type": "string"}},
+            "required": ["query"],
+        }
+        data = {"tool_input_schema": tool_schema}
+        result = MCPToolBlock.Input.get_input_schema(data)
+        assert result == tool_schema
+
+    def test_get_input_schema_without_tool_schema(self):
+        result = MCPToolBlock.Input.get_input_schema({})
+        assert result == {}
+
+    def test_get_input_defaults(self):
+        data = {"tool_arguments": {"city": "London"}}
+        result = MCPToolBlock.Input.get_input_defaults(data)
+        assert result == {"city": "London"}
+
+    def test_get_missing_input(self):
+        data = {
+            "tool_input_schema": {
+                "type": "object",
+                "properties": {
+                    "city": {"type": "string"},
+                    "units": {"type": "string"},
+                },
+                "required": ["city", "units"],
+            },
+            "tool_arguments": {"city": "London"},
+        }
+        missing = MCPToolBlock.Input.get_missing_input(data)
+        assert missing == {"units"}
+
+    def test_get_missing_input_all_present(self):
+        data = {
+            "tool_input_schema": {
+                "type": "object",
+                "properties": {"city": {"type": "string"}},
+                "required": ["city"],
+            },
+            "tool_arguments": {"city": "London"},
+        }
+        missing = MCPToolBlock.Input.get_missing_input(data)
+        assert missing == set()
+
+    @pytest.mark.asyncio(loop_scope="session")
+    async def test_run_with_mock(self):
+        """Test the block using the built-in test infrastructure."""
+        block = MCPToolBlock()
+        await execute_block_test(block)
+
+    @pytest.mark.asyncio(loop_scope="session")
+    async def test_run_missing_server_url(self):
+        block = MCPToolBlock()
+        input_data = MCPToolBlock.Input(
+            server_url="",
+            selected_tool="test",
+        )
+        outputs = []
+        async for name, data in block.run(input_data, user_id=MOCK_USER_ID):
+            outputs.append((name, data))
+        assert outputs == [("error", "MCP server URL is required")]
+
+    @pytest.mark.asyncio(loop_scope="session")
+    async def test_run_missing_tool(self):
+        block = MCPToolBlock()
+        input_data = MCPToolBlock.Input(
+            server_url="https://mcp.example.com/mcp",
+            selected_tool="",
+        )
+        outputs = []
+        async for name, data in block.run(input_data, user_id=MOCK_USER_ID):
+            outputs.append((name, data))
+        assert outputs == [
+            ("error", "No tool selected. Please select a tool from the dropdown.")
+        ]
+
+    @pytest.mark.asyncio(loop_scope="session")
+    async def test_run_success(self):
+        block = MCPToolBlock()
+        input_data = MCPToolBlock.Input(
+            server_url="https://mcp.example.com/mcp",
+            selected_tool="get_weather",
+            tool_input_schema={
+                "type": "object",
+                "properties": {"city": {"type": "string"}},
+            },
+            tool_arguments={"city": "London"},
+        )
+
+        async def mock_call(*args, **kwargs):
+            return {"temp": 20, "city": "London"}
+
+        block._call_mcp_tool = mock_call  # type: ignore
+
+        outputs = []
+        async for name, data in block.run(input_data, user_id=MOCK_USER_ID):
+            outputs.append((name, data))
+
+        assert len(outputs) == 1
+        assert outputs[0][0] == "result"
+        assert outputs[0][1] == {"temp": 20, "city": "London"}
+
+    @pytest.mark.asyncio(loop_scope="session")
+    async def test_run_mcp_error(self):
+        block = MCPToolBlock()
+        input_data = MCPToolBlock.Input(
+            server_url="https://mcp.example.com/mcp",
+            selected_tool="bad_tool",
+        )
+
+        async def mock_call(*args, **kwargs):
+            raise MCPClientError("Tool not found")
+
+        block._call_mcp_tool = mock_call  # type: ignore
+
+        outputs = []
+        async for name, data in block.run(input_data, user_id=MOCK_USER_ID):
+            outputs.append((name, data))
+
+        assert outputs[0][0] == "error"
+        assert "Tool not found" in outputs[0][1]
+
+    @pytest.mark.asyncio(loop_scope="session")
+    async def test_call_mcp_tool_parses_json_text(self):
+        block = MCPToolBlock()
+
+        mock_result = MCPCallResult(
+            content=[
+                {"type": "text", "text": '{"temp": 20}'},
+            ],
+            is_error=False,
+        )
+
+        async def mock_init(self):
+            return {}
+
+        async def mock_call(self, name, args):
+            return mock_result
+
+        with (
+            patch.object(MCPClient, "initialize", mock_init),
+            patch.object(MCPClient, "call_tool", mock_call),
+        ):
+            result = await block._call_mcp_tool(
+                "https://mcp.example.com", "test_tool", {}
+            )
+
+        assert result == {"temp": 20}
+
+    @pytest.mark.asyncio(loop_scope="session")
+    async def test_call_mcp_tool_plain_text(self):
+        block = MCPToolBlock()
+
+        mock_result = MCPCallResult(
+            content=[
+                {"type": "text", "text": "Hello, world!"},
+            ],
+            is_error=False,
+        )
+
+        async def mock_init(self):
+            return {}
+
+        async def mock_call(self, name, args):
+            return mock_result
+
+        with (
+            patch.object(MCPClient, "initialize", mock_init),
+            patch.object(MCPClient, "call_tool", mock_call),
+        ):
+            result = await block._call_mcp_tool(
+                "https://mcp.example.com", "test_tool", {}
+            )
+
+        assert result == "Hello, world!"
+
+    @pytest.mark.asyncio(loop_scope="session")
+    async def test_call_mcp_tool_multiple_content(self):
+        block = MCPToolBlock()
+
+        mock_result = MCPCallResult(
+            content=[
+                {"type": "text", "text": "Part 1"},
+                {"type": "text", "text": '{"part": 2}'},
+            ],
+            is_error=False,
+        )
+
+        async def mock_init(self):
+            return {}
+
+        async def mock_call(self, name, args):
+            return mock_result
+
+        with (
+            patch.object(MCPClient, "initialize", mock_init),
+            patch.object(MCPClient, "call_tool", mock_call),
+        ):
+            result = await block._call_mcp_tool(
+                "https://mcp.example.com", "test_tool", {}
+            )
+
+        assert result == ["Part 1", {"part": 2}]
+
+    @pytest.mark.asyncio(loop_scope="session")
+    async def test_call_mcp_tool_error_result(self):
+        block = MCPToolBlock()
+
+        mock_result = MCPCallResult(
+            content=[{"type": "text", "text": "Something went wrong"}],
+            is_error=True,
+        )
+
+        async def mock_init(self):
+            return {}
+
+        async def mock_call(self, name, args):
+            return mock_result
+
+        with (
+            patch.object(MCPClient, "initialize", mock_init),
+            patch.object(MCPClient, "call_tool", mock_call),
+        ):
+            with pytest.raises(MCPClientError, match="returned an error"):
+                await block._call_mcp_tool("https://mcp.example.com", "test_tool", {})
+
+    @pytest.mark.asyncio(loop_scope="session")
+    async def test_call_mcp_tool_image_content(self):
+        block = MCPToolBlock()
+
+        mock_result = MCPCallResult(
+            content=[
+                {
+                    "type": "image",
+                    "data": "base64data==",
+                    "mimeType": "image/png",
+                }
+            ],
+            is_error=False,
+        )
+
+        async def mock_init(self):
+            return {}
+
+        async def mock_call(self, name, args):
+            return mock_result
+
+        with (
+            patch.object(MCPClient, "initialize", mock_init),
+            patch.object(MCPClient, "call_tool", mock_call),
+        ):
+            result = await block._call_mcp_tool(
+                "https://mcp.example.com", "test_tool", {}
+            )
+
+        assert result == {
+            "type": "image",
+            "data": "base64data==",
+            "mimeType": "image/png",
+        }
+
+    @pytest.mark.asyncio(loop_scope="session")
+    async def test_run_with_credentials(self):
+        """Verify the block uses OAuth2Credentials and passes auth token."""
+        from pydantic import SecretStr
+
+        from backend.data.model import OAuth2Credentials
+
+        block = MCPToolBlock()
+        input_data = MCPToolBlock.Input(
+            server_url="https://mcp.example.com/mcp",
+            selected_tool="test_tool",
+        )
+
+        captured_tokens: list[str | None] = []
+
+        async def mock_call(server_url, tool_name, arguments, auth_token=None):
+            captured_tokens.append(auth_token)
+            return "ok"
+
+        block._call_mcp_tool = mock_call  # type: ignore
+
+        test_creds = OAuth2Credentials(
+            id="cred-123",
+            provider="mcp",
+            access_token=SecretStr("resolved-token"),
+            refresh_token=SecretStr(""),
+            scopes=[],
+            title="Test MCP credential",
+        )
+
+        async for _ in block.run(
+            input_data, user_id=MOCK_USER_ID, credentials=test_creds
+        ):
+            pass
+
+        assert captured_tokens == ["resolved-token"]
+
+    @pytest.mark.asyncio(loop_scope="session")
+    async def test_run_without_credentials(self):
+        """Verify the block works without credentials (public server)."""
+        block = MCPToolBlock()
+        input_data = MCPToolBlock.Input(
+            server_url="https://mcp.example.com/mcp",
+            selected_tool="test_tool",
+        )
+
+        captured_tokens: list[str | None] = []
+
+        async def mock_call(server_url, tool_name, arguments, auth_token=None):
+            captured_tokens.append(auth_token)
+            return "ok"
+
+        block._call_mcp_tool = mock_call  # type: ignore
+
+        outputs = []
+        async for name, data in block.run(input_data, user_id=MOCK_USER_ID):
+            outputs.append((name, data))
+
+        assert captured_tokens == [None]
+        assert outputs == [("result", "ok")]

autogpt_platform/backend/backend/blocks/mcp/test_oauth.py

@@ -0,0 +1,242 @@
+"""
+Tests for MCP OAuth handler.
+"""
+
+from unittest.mock import AsyncMock, MagicMock, patch
+
+import pytest
+from pydantic import SecretStr
+
+from backend.blocks.mcp.client import MCPClient
+from backend.blocks.mcp.oauth import MCPOAuthHandler
+from backend.data.model import OAuth2Credentials
+
+
+def _mock_response(json_data: dict, status: int = 200) -> MagicMock:
+    """Create a mock Response with synchronous json() (matching Requests.Response)."""
+    resp = MagicMock()
+    resp.status = status
+    resp.ok = 200 <= status < 300
+    resp.json.return_value = json_data
+    return resp
+
+
+class TestMCPOAuthHandler:
+    """Tests for the MCPOAuthHandler."""
+
+    def _make_handler(self, **overrides) -> MCPOAuthHandler:
+        defaults = {
+            "client_id": "test-client-id",
+            "client_secret": "test-client-secret",
+            "redirect_uri": "https://app.example.com/callback",
+            "authorize_url": "https://auth.example.com/authorize",
+            "token_url": "https://auth.example.com/token",
+        }
+        defaults.update(overrides)
+        return MCPOAuthHandler(**defaults)
+
+    def test_get_login_url_basic(self):
+        handler = self._make_handler()
+        url = handler.get_login_url(
+            scopes=["read", "write"],
+            state="random-state-token",
+            code_challenge="S256-challenge-value",
+        )
+
+        assert "https://auth.example.com/authorize?" in url
+        assert "response_type=code" in url
+        assert "client_id=test-client-id" in url
+        assert "state=random-state-token" in url
+        assert "code_challenge=S256-challenge-value" in url
+        assert "code_challenge_method=S256" in url
+        assert "scope=read+write" in url
+
+    def test_get_login_url_with_resource(self):
+        handler = self._make_handler(resource_url="https://mcp.example.com/mcp")
+        url = handler.get_login_url(
+            scopes=[], state="state", code_challenge="challenge"
+        )
+
+        assert "resource=https" in url
+
+    def test_get_login_url_without_pkce(self):
+        handler = self._make_handler()
+        url = handler.get_login_url(scopes=["read"], state="state", code_challenge=None)
+
+        assert "code_challenge" not in url
+        assert "code_challenge_method" not in url
+
+    @pytest.mark.asyncio(loop_scope="session")
+    async def test_exchange_code_for_tokens(self):
+        handler = self._make_handler()
+
+        resp = _mock_response(
+            {
+                "access_token": "new-access-token",
+                "refresh_token": "new-refresh-token",
+                "expires_in": 3600,
+                "token_type": "Bearer",
+            }
+        )
+
+        with patch("backend.blocks.mcp.oauth.Requests") as MockRequests:
+            instance = MockRequests.return_value
+            instance.post = AsyncMock(return_value=resp)
+
+            creds = await handler.exchange_code_for_tokens(
+                code="auth-code",
+                scopes=["read"],
+                code_verifier="pkce-verifier",
+            )
+
+        assert isinstance(creds, OAuth2Credentials)
+        assert creds.access_token.get_secret_value() == "new-access-token"
+        assert creds.refresh_token is not None
+        assert creds.refresh_token.get_secret_value() == "new-refresh-token"
+        assert creds.scopes == ["read"]
+        assert creds.access_token_expires_at is not None
+
+    @pytest.mark.asyncio(loop_scope="session")
+    async def test_refresh_tokens(self):
+        handler = self._make_handler()
+
+        existing_creds = OAuth2Credentials(
+            id="existing-id",
+            provider="mcp",
+            access_token=SecretStr("old-token"),
+            refresh_token=SecretStr("old-refresh"),
+            scopes=["read"],
+            title="test",
+        )
+
+        resp = _mock_response(
+            {
+                "access_token": "refreshed-token",
+                "refresh_token": "new-refresh",
+                "expires_in": 3600,
+            }
+        )
+
+        with patch("backend.blocks.mcp.oauth.Requests") as MockRequests:
+            instance = MockRequests.return_value
+            instance.post = AsyncMock(return_value=resp)
+
+            refreshed = await handler._refresh_tokens(existing_creds)
+
+        assert refreshed.id == "existing-id"
+        assert refreshed.access_token.get_secret_value() == "refreshed-token"
+        assert refreshed.refresh_token is not None
+        assert refreshed.refresh_token.get_secret_value() == "new-refresh"
+
+    @pytest.mark.asyncio(loop_scope="session")
+    async def test_refresh_tokens_no_refresh_token(self):
+        handler = self._make_handler()
+
+        creds = OAuth2Credentials(
+            provider="mcp",
+            access_token=SecretStr("token"),
+            scopes=["read"],
+            title="test",
+        )
+
+        with pytest.raises(ValueError, match="No refresh token"):
+            await handler._refresh_tokens(creds)
+
+    @pytest.mark.asyncio(loop_scope="session")
+    async def test_revoke_tokens_no_url(self):
+        handler = self._make_handler(revoke_url=None)
+
+        creds = OAuth2Credentials(
+            provider="mcp",
+            access_token=SecretStr("token"),
+            scopes=[],
+            title="test",
+        )
+
+        result = await handler.revoke_tokens(creds)
+        assert result is False
+
+    @pytest.mark.asyncio(loop_scope="session")
+    async def test_revoke_tokens_with_url(self):
+        handler = self._make_handler(revoke_url="https://auth.example.com/revoke")
+
+        creds = OAuth2Credentials(
+            provider="mcp",
+            access_token=SecretStr("token"),
+            scopes=[],
+            title="test",
+        )
+
+        resp = _mock_response({}, status=200)
+
+        with patch("backend.blocks.mcp.oauth.Requests") as MockRequests:
+            instance = MockRequests.return_value
+            instance.post = AsyncMock(return_value=resp)
+
+            result = await handler.revoke_tokens(creds)
+
+        assert result is True
+
+
+class TestMCPClientDiscovery:
+    """Tests for MCPClient OAuth metadata discovery."""
+
+    @pytest.mark.asyncio(loop_scope="session")
+    async def test_discover_auth_found(self):
+        client = MCPClient("https://mcp.example.com/mcp")
+
+        metadata = {
+            "authorization_servers": ["https://auth.example.com"],
+            "resource": "https://mcp.example.com/mcp",
+        }
+
+        resp = _mock_response(metadata, status=200)
+
+        with patch("backend.blocks.mcp.client.Requests") as MockRequests:
+            instance = MockRequests.return_value
+            instance.get = AsyncMock(return_value=resp)
+
+            result = await client.discover_auth()
+
+        assert result is not None
+        assert result["authorization_servers"] == ["https://auth.example.com"]
+
+    @pytest.mark.asyncio(loop_scope="session")
+    async def test_discover_auth_not_found(self):
+        client = MCPClient("https://mcp.example.com/mcp")
+
+        resp = _mock_response({}, status=404)
+
+        with patch("backend.blocks.mcp.client.Requests") as MockRequests:
+            instance = MockRequests.return_value
+            instance.get = AsyncMock(return_value=resp)
+
+            result = await client.discover_auth()
+
+        assert result is None
+
+    @pytest.mark.asyncio(loop_scope="session")
+    async def test_discover_auth_server_metadata(self):
+        client = MCPClient("https://mcp.example.com/mcp")
+
+        server_metadata = {
+            "issuer": "https://auth.example.com",
+            "authorization_endpoint": "https://auth.example.com/authorize",
+            "token_endpoint": "https://auth.example.com/token",
+            "registration_endpoint": "https://auth.example.com/register",
+            "code_challenge_methods_supported": ["S256"],
+        }
+
+        resp = _mock_response(server_metadata, status=200)
+
+        with patch("backend.blocks.mcp.client.Requests") as MockRequests:
+            instance = MockRequests.return_value
+            instance.get = AsyncMock(return_value=resp)
+
+            result = await client.discover_auth_server_metadata(
+                "https://auth.example.com"
+            )
+
+        assert result is not None
+        assert result["authorization_endpoint"] == "https://auth.example.com/authorize"
+        assert result["token_endpoint"] == "https://auth.example.com/token"

autogpt_platform/backend/backend/blocks/mcp/test_server.py

@@ -0,0 +1,162 @@
+"""
+Minimal MCP server for integration testing.
+
+Implements the MCP Streamable HTTP transport (JSON-RPC 2.0 over HTTP POST)
+with a few sample tools. Runs on localhost with a random available port.
+"""
+
+import json
+import logging
+
+from aiohttp import web
+
+logger = logging.getLogger(__name__)
+
+# Sample tools this test server exposes
+TEST_TOOLS = [
+    {
+        "name": "get_weather",
+        "description": "Get current weather for a city",
+        "inputSchema": {
+            "type": "object",
+            "properties": {
+                "city": {
+                    "type": "string",
+                    "description": "City name",
+                },
+            },
+            "required": ["city"],
+        },
+    },
+    {
+        "name": "add_numbers",
+        "description": "Add two numbers together",
+        "inputSchema": {
+            "type": "object",
+            "properties": {
+                "a": {"type": "number", "description": "First number"},
+                "b": {"type": "number", "description": "Second number"},
+            },
+            "required": ["a", "b"],
+        },
+    },
+    {
+        "name": "echo",
+        "description": "Echo back the input message",
+        "inputSchema": {
+            "type": "object",
+            "properties": {
+                "message": {"type": "string", "description": "Message to echo"},
+            },
+            "required": ["message"],
+        },
+    },
+]
+
+
+def _handle_initialize(params: dict) -> dict:
+    return {
+        "protocolVersion": "2025-03-26",
+        "capabilities": {"tools": {"listChanged": False}},
+        "serverInfo": {"name": "test-mcp-server", "version": "1.0.0"},
+    }
+
+
+def _handle_tools_list(params: dict) -> dict:
+    return {"tools": TEST_TOOLS}
+
+
+def _handle_tools_call(params: dict) -> dict:
+    tool_name = params.get("name", "")
+    arguments = params.get("arguments", {})
+
+    if tool_name == "get_weather":
+        city = arguments.get("city", "Unknown")
+        return {
+            "content": [
+                {
+                    "type": "text",
+                    "text": json.dumps(
+                        {"city": city, "temperature": 22, "condition": "sunny"}
+                    ),
+                }
+            ],
+        }
+
+    elif tool_name == "add_numbers":
+        a = arguments.get("a", 0)
+        b = arguments.get("b", 0)
+        return {
+            "content": [{"type": "text", "text": json.dumps({"result": a + b})}],
+        }
+
+    elif tool_name == "echo":
+        message = arguments.get("message", "")
+        return {
+            "content": [{"type": "text", "text": message}],
+        }
+
+    else:
+        return {
+            "content": [{"type": "text", "text": f"Unknown tool: {tool_name}"}],
+            "isError": True,
+        }
+
+
+HANDLERS = {
+    "initialize": _handle_initialize,
+    "tools/list": _handle_tools_list,
+    "tools/call": _handle_tools_call,
+}
+
+
+async def handle_mcp_request(request: web.Request) -> web.Response:
+    """Handle incoming MCP JSON-RPC 2.0 requests."""
+    # Check auth if configured
+    expected_token = request.app.get("auth_token")
+    if expected_token:
+        auth_header = request.headers.get("Authorization", "")
+        if auth_header != f"Bearer {expected_token}":
+            return web.json_response(
+                {
+                    "jsonrpc": "2.0",
+                    "error": {"code": -32001, "message": "Unauthorized"},
+                    "id": None,
+                },
+                status=401,
+            )
+
+    body = await request.json()
+
+    # Handle notifications (no id field) — just acknowledge
+    if "id" not in body:
+        return web.Response(status=202)
+
+    method = body.get("method", "")
+    params = body.get("params", {})
+    request_id = body.get("id")
+
+    handler = HANDLERS.get(method)
+    if not handler:
+        return web.json_response(
+            {
+                "jsonrpc": "2.0",
+                "error": {
+                    "code": -32601,
+                    "message": f"Method not found: {method}",
+                },
+                "id": request_id,
+            }
+        )
+
+    result = handler(params)
+    return web.json_response({"jsonrpc": "2.0", "result": result, "id": request_id})
+
+
+def create_test_mcp_app(auth_token: str | None = None) -> web.Application:
+    """Create an aiohttp app that acts as an MCP server."""
+    app = web.Application()
+    app.router.add_post("/mcp", handle_mcp_request)
+    if auth_token:
+        app["auth_token"] = auth_token
+    return app

autogpt_platform/backend/backend/conftest.py

@@ -1,6 +1,7 @@
 import logging
 import os
 
+import pytest
 import pytest_asyncio
 from dotenv import load_dotenv
 
@@ -27,6 +28,54 @@ async def server():
         yield server
 
 
+@pytest.fixture
+def test_user_id() -> str:
+    """Test user ID fixture."""
+    return "3e53486c-cf57-477e-ba2a-cb02dc828e1a"
+
+
+@pytest.fixture
+def admin_user_id() -> str:
+    """Admin user ID fixture."""
+    return "4e53486c-cf57-477e-ba2a-cb02dc828e1b"
+
+
+@pytest.fixture
+def target_user_id() -> str:
+    """Target user ID fixture."""
+    return "5e53486c-cf57-477e-ba2a-cb02dc828e1c"
+
+
+@pytest.fixture
+async def setup_test_user(test_user_id):
+    """Create test user in database before tests."""
+    from backend.data.user import get_or_create_user
+
+    # Create the test user in the database using JWT token format
+    user_data = {
+        "sub": test_user_id,
+        "email": "test@example.com",
+        "user_metadata": {"name": "Test User"},
+    }
+    await get_or_create_user(user_data)
+    return test_user_id
+
+
+@pytest.fixture
+async def setup_admin_user(admin_user_id):
+    """Create admin user in database before tests."""
+    from backend.data.user import get_or_create_user
+
+    # Create the admin user in the database using JWT token format
+    user_data = {
+        "sub": admin_user_id,
+        "email": "test-admin@example.com",
+        "user_metadata": {"name": "Test Admin"},
+    }
+    await get_or_create_user(user_data)
+    return admin_user_id
+
+
 @pytest_asyncio.fixture(scope="session", loop_scope="session", autouse=True)
 async def graph_cleanup(server):
     created_graph_ids = []

autogpt_platform/backend/backend/copilot/__init__.py

@@ -0,0 +1 @@
+

autogpt_platform/backend/backend/copilot/completion_consumer.py

@@ -37,12 +37,10 @@ stale pending messages from dead consumers.
 
 import asyncio
 import logging
-import os
 import uuid
 from typing import Any
 
 import orjson
-from prisma import Prisma
 from pydantic import BaseModel
 from redis.exceptions import ResponseError
 
@@ -69,8 +67,8 @@ class OperationCompleteMessage(BaseModel):
 class ChatCompletionConsumer:
     """Consumer for chat operation completion messages from Redis Streams.
 
-    This consumer initializes its own Prisma client in start() to ensure
-    database operations work correctly within this async context.
+    Database operations are handled through the chat_db() accessor, which
+    routes through DatabaseManager RPC when Prisma is not directly connected.
 
     Uses Redis consumer groups to allow multiple platform pods to consume
     messages reliably with automatic redelivery on failure.
@@ -79,7 +77,6 @@ class ChatCompletionConsumer:
     def __init__(self):
         self._consumer_task: asyncio.Task | None = None
         self._running = False
-        self._prisma: Prisma | None = None
         self._consumer_name = f"consumer-{uuid.uuid4().hex[:8]}"
 
     async def start(self) -> None:
@@ -115,15 +112,6 @@ class ChatCompletionConsumer:
             f"Chat completion consumer started (consumer: {self._consumer_name})"
         )
 
-    async def _ensure_prisma(self) -> Prisma:
-        """Lazily initialize Prisma client on first use."""
-        if self._prisma is None:
-            database_url = os.getenv("DATABASE_URL", "postgresql://localhost:5432")
-            self._prisma = Prisma(datasource={"url": database_url})
-            await self._prisma.connect()
-            logger.info("[COMPLETION] Consumer Prisma client connected (lazy init)")
-        return self._prisma
-
     async def stop(self) -> None:
         """Stop the completion consumer."""
         self._running = False
@@ -136,11 +124,6 @@ class ChatCompletionConsumer:
                 pass
             self._consumer_task = None
 
-        if self._prisma:
-            await self._prisma.disconnect()
-            self._prisma = None
-            logger.info("[COMPLETION] Consumer Prisma client disconnected")
-
         logger.info("Chat completion consumer stopped")
 
     async def _consume_messages(self) -> None:
@@ -252,7 +235,7 @@ class ChatCompletionConsumer:
             # XAUTOCLAIM after min_idle_time expires
 
     async def _handle_message(self, body: bytes) -> None:
-        """Handle a completion message using our own Prisma client."""
+        """Handle a completion message."""
         try:
             data = orjson.loads(body)
             message = OperationCompleteMessage(**data)
@@ -302,8 +285,7 @@ class ChatCompletionConsumer:
         message: OperationCompleteMessage,
     ) -> None:
         """Handle successful operation completion."""
-        prisma = await self._ensure_prisma()
-        await process_operation_success(task, message.result, prisma)
+        await process_operation_success(task, message.result)
 
     async def _handle_failure(
         self,
@@ -311,8 +293,7 @@ class ChatCompletionConsumer:
         message: OperationCompleteMessage,
     ) -> None:
         """Handle failed operation completion."""
-        prisma = await self._ensure_prisma()
-        await process_operation_failure(task, message.error, prisma)
+        await process_operation_failure(task, message.error)
 
 
 # Module-level consumer instance

autogpt_platform/backend/backend/copilot/completion_handler.py

@@ -9,7 +9,8 @@ import logging
 from typing import Any
 
 import orjson
-from prisma import Prisma
+
+from backend.data.db_accessors import chat_db
 
 from . import service as chat_service
 from . import stream_registry
@@ -72,48 +73,40 @@ async def _update_tool_message(
     session_id: str,
     tool_call_id: str,
     content: str,
-    prisma_client: Prisma | None,
 ) -> None:
-    """Update tool message in database.
+    """Update tool message in database using the chat_db accessor.
+
+    Routes through DatabaseManager RPC when Prisma is not directly
+    connected (e.g. in the CoPilot Executor microservice).
 
     Args:
         session_id: The session ID
         tool_call_id: The tool call ID to update
         content: The new content for the message
-        prisma_client: Optional Prisma client. If None, uses chat_service.
 
     Raises:
-        ToolMessageUpdateError: If the database update fails. The caller should
-            handle this to avoid marking the task as completed with inconsistent state.
+        ToolMessageUpdateError: If the database update fails.
     """
     try:
-        if prisma_client:
-            # Use provided Prisma client (for consumer with its own connection)
-            updated_count = await prisma_client.chatmessage.update_many(
-                where={
-                    "sessionId": session_id,
-                    "toolCallId": tool_call_id,
-                },
-                data={"content": content},
-            )
-            # Check if any rows were updated - 0 means message not found
-            if updated_count == 0:
-                raise ToolMessageUpdateError(
-                    f"No message found with tool_call_id={tool_call_id} in session {session_id}"
-                )
-        else:
-            # Use service function (for webhook endpoint)
-            await chat_service._update_pending_operation(
-                session_id=session_id,
-                tool_call_id=tool_call_id,
-                result=content,
+        updated = await chat_db().update_tool_message_content(
+            session_id=session_id,
+            tool_call_id=tool_call_id,
+            new_content=content,
+        )
+        if not updated:
+            raise ToolMessageUpdateError(
+                f"No message found with tool_call_id="
+                f"{tool_call_id} in session {session_id}"
             )
     except ToolMessageUpdateError:
         raise
     except Exception as e:
-        logger.error(f"[COMPLETION] Failed to update tool message: {e}", exc_info=True)
+        logger.error(
+            f"[COMPLETION] Failed to update tool message: {e}",
+            exc_info=True,
+        )
         raise ToolMessageUpdateError(
-            f"Failed to update tool message for tool_call_id={tool_call_id}: {e}"
+            f"Failed to update tool message for tool call #{tool_call_id}: {e}"
         ) from e
 
 
@@ -202,7 +195,6 @@ async def _save_agent_from_result(
 async def process_operation_success(
     task: stream_registry.ActiveTask,
     result: dict | str | None,
-    prisma_client: Prisma | None = None,
 ) -> None:
     """Handle successful operation completion.
 
@@ -212,12 +204,10 @@ async def process_operation_success(
     Args:
         task: The active task that completed
         result: The result data from the operation
-        prisma_client: Optional Prisma client for database operations.
-            If None, uses chat_service._update_pending_operation instead.
 
     Raises:
-        ToolMessageUpdateError: If the database update fails. The task will be
-            marked as failed instead of completed to avoid inconsistent state.
+        ToolMessageUpdateError: If the database update fails. The task
+            will be marked as failed instead of completed.
     """
     # For agent generation tools, save the agent to library
     if task.tool_name in AGENT_GENERATION_TOOLS and isinstance(result, dict):
@@ -250,7 +240,6 @@ async def process_operation_success(
             session_id=task.session_id,
             tool_call_id=task.tool_call_id,
             content=result_str,
-            prisma_client=prisma_client,
         )
     except ToolMessageUpdateError:
         # DB update failed - mark task as failed to avoid inconsistent state
@@ -293,18 +282,15 @@ async def process_operation_success(
 async def process_operation_failure(
     task: stream_registry.ActiveTask,
     error: str | None,
-    prisma_client: Prisma | None = None,
 ) -> None:
     """Handle failed operation completion.
 
-    Publishes the error to the stream registry, updates the database with
-    the error response, and marks the task as failed.
+    Publishes the error to the stream registry, updates the database
+    with the error response, and marks the task as failed.
 
     Args:
         task: The active task that failed
         error: The error message from the operation
-        prisma_client: Optional Prisma client for database operations.
-            If None, uses chat_service._update_pending_operation instead.
     """
     error_msg = error or "Operation failed"
 
@@ -325,7 +311,6 @@ async def process_operation_failure(
             session_id=task.session_id,
             tool_call_id=task.tool_call_id,
             content=error_response.model_dump_json(),
-            prisma_client=prisma_client,
         )
     except ToolMessageUpdateError:
         # DB update failed - log but continue with cleanup

autogpt_platform/backend/backend/copilot/config.py

@@ -27,12 +27,11 @@ class ChatConfig(BaseSettings):
     session_ttl: int = Field(default=43200, description="Session TTL in seconds")
 
     # Streaming Configuration
-    max_context_messages: int = Field(
-        default=50, ge=1, le=200, description="Maximum context messages"
-    )
-
     stream_timeout: int = Field(default=300, description="Stream timeout in seconds")
-    max_retries: int = Field(default=3, description="Maximum number of retries")
+    max_retries: int = Field(
+        default=3,
+        description="Max retries for fallback path (SDK handles retries internally)",
+    )
     max_agent_runs: int = Field(default=30, description="Maximum number of agent runs")
     max_agent_schedules: int = Field(
         default=30, description="Maximum number of agent schedules"
@@ -93,6 +92,31 @@ class ChatConfig(BaseSettings):
         description="Name of the prompt in Langfuse to fetch",
     )
 
+    # Claude Agent SDK Configuration
+    use_claude_agent_sdk: bool = Field(
+        default=True,
+        description="Use Claude Agent SDK for chat completions",
+    )
+    claude_agent_model: str | None = Field(
+        default=None,
+        description="Model for the Claude Agent SDK path. If None, derives from "
+        "the `model` field by stripping the OpenRouter provider prefix.",
+    )
+    claude_agent_max_buffer_size: int = Field(
+        default=10 * 1024 * 1024,  # 10MB (default SDK is 1MB)
+        description="Max buffer size in bytes for Claude Agent SDK JSON message parsing. "
+        "Increase if tool outputs exceed the limit.",
+    )
+    claude_agent_max_subtasks: int = Field(
+        default=10,
+        description="Max number of sub-agent Tasks the SDK can spawn per session.",
+    )
+    claude_agent_use_resume: bool = Field(
+        default=True,
+        description="Use --resume for multi-turn conversations instead of "
+        "history compression. Falls back to compression when unavailable.",
+    )
+
     # Extended thinking configuration for Claude models
     thinking_enabled: bool = Field(
         default=True,
@@ -138,6 +162,17 @@ class ChatConfig(BaseSettings):
             v = os.getenv("CHAT_INTERNAL_API_KEY")
         return v
 
+    @field_validator("use_claude_agent_sdk", mode="before")
+    @classmethod
+    def get_use_claude_agent_sdk(cls, v):
+        """Get use_claude_agent_sdk from environment if not provided."""
+        # Check environment variable - default to True if not set
+        env_val = os.getenv("CHAT_USE_CLAUDE_AGENT_SDK", "").lower()
+        if env_val:
+            return env_val in ("true", "1", "yes", "on")
+        # Default to True (SDK enabled by default)
+        return True if v is None else v
+
     # Prompt paths for different contexts
     PROMPT_PATHS: dict[str, str] = {
         "default": "prompts/chat_system.md",

autogpt_platform/backend/backend/copilot/db.py

@@ -14,29 +14,27 @@ from prisma.types import (
     ChatSessionWhereInput,
 )
 
-from backend.data.db import transaction
+from backend.data import db
 from backend.util.json import SafeJson
 
+from .model import ChatMessage, ChatSession
+
 logger = logging.getLogger(__name__)
 
 
-async def get_chat_session(session_id: str) -> PrismaChatSession | None:
+async def get_chat_session(session_id: str) -> ChatSession | None:
     """Get a chat session by ID from the database."""
     session = await PrismaChatSession.prisma().find_unique(
         where={"id": session_id},
-        include={"Messages": True},
+        include={"Messages": {"order_by": {"sequence": "asc"}}},
     )
-    if session and session.Messages:
-        # Sort messages by sequence in Python - Prisma Python client doesn't support
-        # order_by in include clauses (unlike Prisma JS), so we sort after fetching
-        session.Messages.sort(key=lambda m: m.sequence)
-    return session
+    return ChatSession.from_db(session) if session else None
 
 
 async def create_chat_session(
     session_id: str,
     user_id: str,
-) -> PrismaChatSession:
+) -> ChatSession:
     """Create a new chat session in the database."""
     data = ChatSessionCreateInput(
         id=session_id,
@@ -45,7 +43,8 @@ async def create_chat_session(
         successfulAgentRuns=SafeJson({}),
         successfulAgentSchedules=SafeJson({}),
     )
-    return await PrismaChatSession.prisma().create(data=data)
+    prisma_session = await PrismaChatSession.prisma().create(data=data)
+    return ChatSession.from_db(prisma_session)
 
 
 async def update_chat_session(
@@ -56,7 +55,7 @@ async def update_chat_session(
     total_prompt_tokens: int | None = None,
     total_completion_tokens: int | None = None,
     title: str | None = None,
-) -> PrismaChatSession | None:
+) -> ChatSession | None:
     """Update a chat session's metadata."""
     data: ChatSessionUpdateInput = {"updatedAt": datetime.now(UTC)}
 
@@ -76,12 +75,9 @@ async def update_chat_session(
     session = await PrismaChatSession.prisma().update(
         where={"id": session_id},
         data=data,
-        include={"Messages": True},
+        include={"Messages": {"order_by": {"sequence": "asc"}}},
     )
-    if session and session.Messages:
-        # Sort in Python - Prisma Python doesn't support order_by in include clauses
-        session.Messages.sort(key=lambda m: m.sequence)
-    return session
+    return ChatSession.from_db(session) if session else None
 
 
 async def add_chat_message(
@@ -94,7 +90,7 @@ async def add_chat_message(
     refusal: str | None = None,
     tool_calls: list[dict[str, Any]] | None = None,
     function_call: dict[str, Any] | None = None,
-) -> PrismaChatMessage:
+) -> ChatMessage:
     """Add a message to a chat session."""
     # Build input dict dynamically rather than using ChatMessageCreateInput directly
     # because Prisma's TypedDict validation rejects optional fields set to None.
@@ -129,14 +125,14 @@ async def add_chat_message(
         ),
         PrismaChatMessage.prisma().create(data=cast(ChatMessageCreateInput, data)),
     )
-    return message
+    return ChatMessage.from_db(message)
 
 
 async def add_chat_messages_batch(
     session_id: str,
     messages: list[dict[str, Any]],
     start_sequence: int,
-) -> list[PrismaChatMessage]:
+) -> list[ChatMessage]:
     """Add multiple messages to a chat session in a batch.
 
     Uses a transaction for atomicity - if any message creation fails,
@@ -147,7 +143,7 @@ async def add_chat_messages_batch(
 
     created_messages = []
 
-    async with transaction() as tx:
+    async with db.transaction() as tx:
         for i, msg in enumerate(messages):
             # Build input dict dynamically rather than using ChatMessageCreateInput
             # directly because Prisma's TypedDict validation rejects optional fields
@@ -187,21 +183,22 @@ async def add_chat_messages_batch(
             data={"updatedAt": datetime.now(UTC)},
         )
 
-    return created_messages
+    return [ChatMessage.from_db(m) for m in created_messages]
 
 
 async def get_user_chat_sessions(
     user_id: str,
     limit: int = 50,
     offset: int = 0,
-) -> list[PrismaChatSession]:
+) -> list[ChatSession]:
     """Get chat sessions for a user, ordered by most recent."""
-    return await PrismaChatSession.prisma().find_many(
+    prisma_sessions = await PrismaChatSession.prisma().find_many(
         where={"userId": user_id},
         order={"updatedAt": "desc"},
         take=limit,
         skip=offset,
     )
+    return [ChatSession.from_db(s) for s in prisma_sessions]
 
 
 async def get_user_session_count(user_id: str) -> int:

autogpt_platform/backend/backend/copilot/executor/__main__.py

@@ -0,0 +1,18 @@
+"""Entry point for running the CoPilot Executor service.
+
+Usage:
+    python -m backend.copilot.executor
+"""
+
+from backend.app import run_processes
+
+from .manager import CoPilotExecutor
+
+
+def main():
+    """Run the CoPilot Executor service."""
+    run_processes(CoPilotExecutor())
+
+
+if __name__ == "__main__":
+    main()

autogpt_platform/backend/backend/copilot/executor/manager.py

@@ -0,0 +1,508 @@
+"""CoPilot Executor Manager - main service for CoPilot task execution.
+
+This module contains the CoPilotExecutor class that consumes chat tasks from
+RabbitMQ and processes them using a thread pool, following the graph executor pattern.
+"""
+
+import logging
+import os
+import threading
+import time
+import uuid
+from concurrent.futures import Future, ThreadPoolExecutor
+
+from pika.adapters.blocking_connection import BlockingChannel
+from pika.exceptions import AMQPChannelError, AMQPConnectionError
+from pika.spec import Basic, BasicProperties
+from prometheus_client import Gauge, start_http_server
+
+from backend.data import redis_client as redis
+from backend.data.rabbitmq import SyncRabbitMQ
+from backend.executor.cluster_lock import ClusterLock
+from backend.util.decorator import error_logged
+from backend.util.logging import TruncatedLogger
+from backend.util.process import AppProcess
+from backend.util.retry import continuous_retry
+from backend.util.settings import Settings
+
+from .processor import execute_copilot_task, init_worker
+from .utils import (
+    COPILOT_CANCEL_QUEUE_NAME,
+    COPILOT_EXECUTION_QUEUE_NAME,
+    GRACEFUL_SHUTDOWN_TIMEOUT_SECONDS,
+    CancelCoPilotEvent,
+    CoPilotExecutionEntry,
+    create_copilot_queue_config,
+)
+
+logger = TruncatedLogger(logging.getLogger(__name__), prefix="[CoPilotExecutor]")
+settings = Settings()
+
+# Prometheus metrics
+active_tasks_gauge = Gauge(
+    "copilot_executor_active_tasks",
+    "Number of active CoPilot tasks",
+)
+pool_size_gauge = Gauge(
+    "copilot_executor_pool_size",
+    "Maximum number of CoPilot executor workers",
+)
+utilization_gauge = Gauge(
+    "copilot_executor_utilization_ratio",
+    "Ratio of active tasks to pool size",
+)
+
+
+class CoPilotExecutor(AppProcess):
+    """CoPilot Executor service for processing chat generation tasks.
+
+    This service consumes tasks from RabbitMQ, processes them using a thread pool,
+    and publishes results to Redis Streams. It follows the graph executor pattern
+    for reliable message handling and graceful shutdown.
+
+    Key features:
+    - RabbitMQ-based task distribution with manual acknowledgment
+    - Thread pool executor for concurrent task processing
+    - Cluster lock for duplicate prevention across pods
+    - Graceful shutdown with timeout for in-flight tasks
+    - FANOUT exchange for cancellation broadcast
+    """
+
+    def __init__(self):
+        super().__init__()
+        self.pool_size = settings.config.num_copilot_workers
+        self.active_tasks: dict[str, tuple[Future, threading.Event]] = {}
+        self.executor_id = str(uuid.uuid4())
+
+        self._executor = None
+        self._stop_consuming = None
+
+        self._cancel_thread = None
+        self._cancel_client = None
+        self._run_thread = None
+        self._run_client = None
+
+        self._task_locks: dict[str, ClusterLock] = {}
+        self._active_tasks_lock = threading.Lock()
+
+    # ============ Main Entry Points (AppProcess interface) ============ #
+
+    def run(self):
+        """Main service loop - consume from RabbitMQ."""
+        logger.info(f"Pod assigned executor_id: {self.executor_id}")
+        logger.info(f"Spawn max-{self.pool_size} workers...")
+
+        pool_size_gauge.set(self.pool_size)
+        self._update_metrics()
+        start_http_server(settings.config.copilot_executor_port)
+
+        self.cancel_thread.start()
+        self.run_thread.start()
+
+        while True:
+            time.sleep(1e5)
+
+    def cleanup(self):
+        """Graceful shutdown with active execution waiting."""
+        pid = os.getpid()
+        logger.info(f"[cleanup {pid}] Starting graceful shutdown...")
+
+        # Signal the consumer thread to stop
+        try:
+            self.stop_consuming.set()
+            run_channel = self.run_client.get_channel()
+            run_channel.connection.add_callback_threadsafe(
+                lambda: run_channel.stop_consuming()
+            )
+            logger.info(f"[cleanup {pid}] Consumer has been signaled to stop")
+        except Exception as e:
+            logger.error(f"[cleanup {pid}] Error stopping consumer: {e}")
+
+        # Wait for active executions to complete
+        if self.active_tasks:
+            logger.info(
+                f"[cleanup {pid}] Waiting for {len(self.active_tasks)} active tasks to complete (timeout: {GRACEFUL_SHUTDOWN_TIMEOUT_SECONDS}s)..."
+            )
+
+            start_time = time.monotonic()
+            last_refresh = start_time
+            lock_refresh_interval = settings.config.cluster_lock_timeout / 10
+
+            while (
+                self.active_tasks
+                and (time.monotonic() - start_time) < GRACEFUL_SHUTDOWN_TIMEOUT_SECONDS
+            ):
+                self._cleanup_completed_tasks()
+                if not self.active_tasks:
+                    break
+
+                # Refresh cluster locks periodically
+                current_time = time.monotonic()
+                if current_time - last_refresh >= lock_refresh_interval:
+                    for lock in list(self._task_locks.values()):
+                        try:
+                            lock.refresh()
+                        except Exception as e:
+                            logger.warning(
+                                f"[cleanup {pid}] Failed to refresh lock: {e}"
+                            )
+                    last_refresh = current_time
+
+                logger.info(
+                    f"[cleanup {pid}] {len(self.active_tasks)} tasks still active, waiting..."
+                )
+                time.sleep(10.0)
+
+        # Stop message consumers
+        if self._run_thread:
+            self._stop_message_consumers(
+                self._run_thread, self.run_client, "[cleanup][run]"
+            )
+        if self._cancel_thread:
+            self._stop_message_consumers(
+                self._cancel_thread, self.cancel_client, "[cleanup][cancel]"
+            )
+
+        # Shutdown executor
+        if self._executor:
+            logger.info(f"[cleanup {pid}] Shutting down executor...")
+            self._executor.shutdown(wait=False)
+
+        # Release any remaining locks
+        for task_id, lock in list(self._task_locks.items()):
+            try:
+                lock.release()
+                logger.info(f"[cleanup {pid}] Released lock for {task_id}")
+            except Exception as e:
+                logger.error(
+                    f"[cleanup {pid}] Failed to release lock for {task_id}: {e}"
+                )
+
+        logger.info(f"[cleanup {pid}] Graceful shutdown completed")
+
+    # ============ RabbitMQ Consumer Methods ============ #
+
+    @continuous_retry()
+    def _consume_cancel(self):
+        """Consume cancellation messages from FANOUT exchange."""
+        if self.stop_consuming.is_set() and not self.active_tasks:
+            logger.info("Stop reconnecting cancel consumer - service cleaned up")
+            return
+
+        if not self.cancel_client.is_ready:
+            self.cancel_client.disconnect()
+        self.cancel_client.connect()
+
+        # Check again after connect - shutdown may have been requested
+        if self.stop_consuming.is_set() and not self.active_tasks:
+            logger.info("Stop consuming requested during reconnect - disconnecting")
+            self.cancel_client.disconnect()
+            return
+
+        cancel_channel = self.cancel_client.get_channel()
+        cancel_channel.basic_consume(
+            queue=COPILOT_CANCEL_QUEUE_NAME,
+            on_message_callback=self._handle_cancel_message,
+            auto_ack=True,
+        )
+        logger.info("Starting cancel message consumer...")
+        cancel_channel.start_consuming()
+        if not self.stop_consuming.is_set() or self.active_tasks:
+            raise RuntimeError("Cancel message consumer stopped unexpectedly")
+        logger.info("Cancel message consumer stopped gracefully")
+
+    @continuous_retry()
+    def _consume_run(self):
+        """Consume run messages from DIRECT exchange."""
+        if self.stop_consuming.is_set():
+            logger.info("Stop reconnecting run consumer - service cleaned up")
+            return
+
+        if not self.run_client.is_ready:
+            self.run_client.disconnect()
+        self.run_client.connect()
+
+        # Check again after connect - shutdown may have been requested
+        if self.stop_consuming.is_set():
+            logger.info("Stop consuming requested during reconnect - disconnecting")
+            self.run_client.disconnect()
+            return
+
+        run_channel = self.run_client.get_channel()
+        run_channel.basic_qos(prefetch_count=self.pool_size)
+
+        run_channel.basic_consume(
+            queue=COPILOT_EXECUTION_QUEUE_NAME,
+            on_message_callback=self._handle_run_message,
+            auto_ack=False,
+            consumer_tag="copilot_execution_consumer",
+        )
+        logger.info("Starting to consume run messages...")
+        run_channel.start_consuming()
+        if not self.stop_consuming.is_set():
+            raise RuntimeError("Run message consumer stopped unexpectedly")
+        logger.info("Run message consumer stopped gracefully")
+
+    # ============ Message Handlers ============ #
+
+    @error_logged(swallow=True)
+    def _handle_cancel_message(
+        self,
+        _channel: BlockingChannel,
+        _method: Basic.Deliver,
+        _properties: BasicProperties,
+        body: bytes,
+    ):
+        """Handle cancel message from FANOUT exchange."""
+        request = CancelCoPilotEvent.model_validate_json(body)
+        task_id = request.task_id
+        if not task_id:
+            logger.warning("Cancel message missing 'task_id'")
+            return
+        if task_id not in self.active_tasks:
+            logger.debug(f"Cancel received for {task_id} but not active")
+            return
+
+        _, cancel_event = self.active_tasks[task_id]
+        logger.info(f"Received cancel for {task_id}")
+        if not cancel_event.is_set():
+            cancel_event.set()
+        else:
+            logger.debug(f"Cancel already set for {task_id}")
+
+    def _handle_run_message(
+        self,
+        _channel: BlockingChannel,
+        method: Basic.Deliver,
+        _properties: BasicProperties,
+        body: bytes,
+    ):
+        """Handle run message from DIRECT exchange."""
+        delivery_tag = method.delivery_tag
+        # Capture the channel used at message delivery time to ensure we ack
+        # on the correct channel. Delivery tags are channel-scoped and become
+        # invalid if the channel is recreated after reconnection.
+        delivery_channel = _channel
+
+        def ack_message(reject: bool, requeue: bool):
+            """Acknowledge or reject the message.
+
+            Uses the channel from the original message delivery. If the channel
+            is no longer open (e.g., after reconnection), logs a warning and
+            skips the ack - RabbitMQ will redeliver the message automatically.
+            """
+            try:
+                if not delivery_channel.is_open:
+                    logger.warning(
+                        f"Channel closed, cannot ack delivery_tag={delivery_tag}. "
+                        "Message will be redelivered by RabbitMQ."
+                    )
+                    return
+
+                if reject:
+                    delivery_channel.connection.add_callback_threadsafe(
+                        lambda: delivery_channel.basic_nack(
+                            delivery_tag, requeue=requeue
+                        )
+                    )
+                else:
+                    delivery_channel.connection.add_callback_threadsafe(
+                        lambda: delivery_channel.basic_ack(delivery_tag)
+                    )
+            except (AMQPChannelError, AMQPConnectionError) as e:
+                # Channel/connection errors indicate stale delivery tag - don't retry
+                logger.warning(
+                    f"Cannot ack delivery_tag={delivery_tag} due to channel/connection "
+                    f"error: {e}. Message will be redelivered by RabbitMQ."
+                )
+            except Exception as e:
+                # Other errors might be transient, but log and skip to avoid blocking
+                logger.error(
+                    f"Unexpected error acking delivery_tag={delivery_tag}: {e}"
+                )
+
+        # Check if we're shutting down
+        if self.stop_consuming.is_set():
+            logger.info("Rejecting new task during shutdown")
+            ack_message(reject=True, requeue=True)
+            return
+
+        # Check if we can accept more tasks
+        self._cleanup_completed_tasks()
+        if len(self.active_tasks) >= self.pool_size:
+            ack_message(reject=True, requeue=True)
+            return
+
+        try:
+            entry = CoPilotExecutionEntry.model_validate_json(body)
+        except Exception as e:
+            logger.error(f"Could not parse run message: {e}, body={body}")
+            ack_message(reject=True, requeue=False)
+            return
+
+        task_id = entry.task_id
+
+        # Check for local duplicate - task is already running on this executor
+        if task_id in self.active_tasks:
+            logger.warning(
+                f"Task {task_id} already running locally, rejecting duplicate"
+            )
+            ack_message(reject=True, requeue=False)
+            return
+
+        # Try to acquire cluster-wide lock
+        cluster_lock = ClusterLock(
+            redis=redis.get_redis(),
+            key=f"copilot:task:{task_id}:lock",
+            owner_id=self.executor_id,
+            timeout=settings.config.cluster_lock_timeout,
+        )
+        current_owner = cluster_lock.try_acquire()
+        if current_owner != self.executor_id:
+            if current_owner is not None:
+                logger.warning(f"Task {task_id} already running on pod {current_owner}")
+                ack_message(reject=True, requeue=False)
+            else:
+                logger.warning(
+                    f"Could not acquire lock for {task_id} - Redis unavailable"
+                )
+                ack_message(reject=True, requeue=True)
+            return
+
+        # Execute the task
+        try:
+            self._task_locks[task_id] = cluster_lock
+
+            logger.info(
+                f"Acquired cluster lock for {task_id}, executor_id={self.executor_id}"
+            )
+
+            cancel_event = threading.Event()
+            future = self.executor.submit(
+                execute_copilot_task, entry, cancel_event, cluster_lock
+            )
+            self.active_tasks[task_id] = (future, cancel_event)
+        except Exception as e:
+            logger.warning(f"Failed to setup execution for {task_id}: {e}")
+            cluster_lock.release()
+            if task_id in self._task_locks:
+                del self._task_locks[task_id]
+            ack_message(reject=True, requeue=True)
+            return
+
+        self._update_metrics()
+
+        def on_run_done(f: Future):
+            logger.info(f"Run completed for {task_id}")
+            try:
+                if exec_error := f.exception():
+                    logger.error(f"Execution for {task_id} failed: {exec_error}")
+                    # Don't requeue failed tasks - they've been marked as failed
+                    # in the stream registry. Requeuing would cause infinite retries
+                    # for deterministic failures.
+                    ack_message(reject=True, requeue=False)
+                else:
+                    ack_message(reject=False, requeue=False)
+            except BaseException as e:
+                logger.exception(f"Error in run completion callback: {e}")
+            finally:
+                # Release the cluster lock
+                if task_id in self._task_locks:
+                    logger.info(f"Releasing cluster lock for {task_id}")
+                    self._task_locks[task_id].release()
+                    del self._task_locks[task_id]
+                self._cleanup_completed_tasks()
+
+        future.add_done_callback(on_run_done)
+
+    # ============ Helper Methods ============ #
+
+    def _cleanup_completed_tasks(self) -> list[str]:
+        """Remove completed futures from active_tasks and update metrics."""
+        completed_tasks = []
+        with self._active_tasks_lock:
+            for task_id, (future, _) in list(self.active_tasks.items()):
+                if future.done():
+                    completed_tasks.append(task_id)
+                    self.active_tasks.pop(task_id, None)
+                    logger.info(f"Cleaned up completed task {task_id}")
+
+        self._update_metrics()
+        return completed_tasks
+
+    def _update_metrics(self):
+        """Update Prometheus metrics."""
+        active_count = len(self.active_tasks)
+        active_tasks_gauge.set(active_count)
+        if self.stop_consuming.is_set():
+            utilization_gauge.set(1.0)
+        else:
+            utilization_gauge.set(
+                active_count / self.pool_size if self.pool_size > 0 else 0
+            )
+
+    def _stop_message_consumers(
+        self, thread: threading.Thread, client: SyncRabbitMQ, prefix: str
+    ):
+        """Stop a message consumer thread."""
+        try:
+            channel = client.get_channel()
+            channel.connection.add_callback_threadsafe(lambda: channel.stop_consuming())
+
+            thread.join(timeout=300)
+            if thread.is_alive():
+                logger.error(
+                    f"{prefix} Thread did not finish in time, forcing disconnect"
+                )
+
+            client.disconnect()
+            logger.info(f"{prefix} Client disconnected")
+        except Exception as e:
+            logger.error(f"{prefix} Error disconnecting client: {e}")
+
+    # ============ Lazy-initialized Properties ============ #
+
+    @property
+    def cancel_thread(self) -> threading.Thread:
+        if self._cancel_thread is None:
+            self._cancel_thread = threading.Thread(
+                target=lambda: self._consume_cancel(),
+                daemon=True,
+            )
+        return self._cancel_thread
+
+    @property
+    def run_thread(self) -> threading.Thread:
+        if self._run_thread is None:
+            self._run_thread = threading.Thread(
+                target=lambda: self._consume_run(),
+                daemon=True,
+            )
+        return self._run_thread
+
+    @property
+    def stop_consuming(self) -> threading.Event:
+        if self._stop_consuming is None:
+            self._stop_consuming = threading.Event()
+        return self._stop_consuming
+
+    @property
+    def executor(self) -> ThreadPoolExecutor:
+        if self._executor is None:
+            self._executor = ThreadPoolExecutor(
+                max_workers=self.pool_size,
+                initializer=init_worker,
+            )
+        return self._executor
+
+    @property
+    def cancel_client(self) -> SyncRabbitMQ:
+        if self._cancel_client is None:
+            self._cancel_client = SyncRabbitMQ(create_copilot_queue_config())
+        return self._cancel_client
+
+    @property
+    def run_client(self) -> SyncRabbitMQ:
+        if self._run_client is None:
+            self._run_client = SyncRabbitMQ(create_copilot_queue_config())
+        return self._run_client

autogpt_platform/backend/backend/copilot/executor/processor.py

@@ -0,0 +1,253 @@
+"""CoPilot execution processor - per-worker execution logic.
+
+This module contains the processor class that handles CoPilot task execution
+in a thread-local context, following the graph executor pattern.
+"""
+
+import asyncio
+import logging
+import threading
+import time
+
+from backend.copilot import service as copilot_service
+from backend.copilot import stream_registry
+from backend.copilot.config import ChatConfig
+from backend.copilot.response_model import StreamError, StreamFinish, StreamFinishStep
+from backend.copilot.sdk import service as sdk_service
+from backend.executor.cluster_lock import ClusterLock
+from backend.util.decorator import error_logged
+from backend.util.feature_flag import Flag, is_feature_enabled
+from backend.util.logging import TruncatedLogger, configure_logging
+from backend.util.process import set_service_name
+from backend.util.retry import func_retry
+
+from .utils import CoPilotExecutionEntry, CoPilotLogMetadata
+
+logger = TruncatedLogger(logging.getLogger(__name__), prefix="[CoPilotExecutor]")
+
+
+# ============ Module Entry Points ============ #
+
+# Thread-local storage for processor instances
+_tls = threading.local()
+
+
+def execute_copilot_task(
+    entry: CoPilotExecutionEntry,
+    cancel: threading.Event,
+    cluster_lock: ClusterLock,
+):
+    """Execute a CoPilot task using the thread-local processor.
+
+    This function is the entry point called by the thread pool executor.
+
+    Args:
+        entry: The task payload
+        cancel: Threading event to signal cancellation
+        cluster_lock: Distributed lock for this execution
+    """
+    processor: CoPilotProcessor = _tls.processor
+    return processor.execute(entry, cancel, cluster_lock)
+
+
+def init_worker():
+    """Initialize the processor for the current worker thread.
+
+    This function is called by the thread pool executor when a new worker
+    thread is created. It ensures each worker has its own processor instance.
+    """
+    _tls.processor = CoPilotProcessor()
+    _tls.processor.on_executor_start()
+
+
+# ============ Processor Class ============ #
+
+
+class CoPilotProcessor:
+    """Per-worker execution logic for CoPilot tasks.
+
+    This class is instantiated once per worker thread and handles the execution
+    of CoPilot chat generation tasks. It maintains an async event loop for
+    running the async service code.
+
+    The execution flow:
+        1. CoPilot task is picked from RabbitMQ queue
+        2. Manager submits task to thread pool
+        3. Processor executes the task in its event loop
+        4. Results are published to Redis Streams
+    """
+
+    @func_retry
+    def on_executor_start(self):
+        """Initialize the processor when the worker thread starts.
+
+        This method is called once per worker thread to set up the async event
+        loop and initialize any required resources.
+
+        Database is accessed only through DatabaseManager, so we don't need to connect
+        to Prisma directly.
+        """
+        configure_logging()
+        set_service_name("CoPilotExecutor")
+        self.tid = threading.get_ident()
+        self.execution_loop = asyncio.new_event_loop()
+        self.execution_thread = threading.Thread(
+            target=self.execution_loop.run_forever, daemon=True
+        )
+        self.execution_thread.start()
+
+        logger.info(f"[CoPilotExecutor] Worker {self.tid} started")
+
+    @error_logged(swallow=False)
+    def execute(
+        self,
+        entry: CoPilotExecutionEntry,
+        cancel: threading.Event,
+        cluster_lock: ClusterLock,
+    ):
+        """Execute a CoPilot task.
+
+        This is the main entry point for task execution. It runs the async
+        execution logic in the worker's event loop and handles errors.
+
+        Args:
+            entry: The task payload containing session and message info
+            cancel: Threading event to signal cancellation
+            cluster_lock: Distributed lock to prevent duplicate execution
+        """
+        log = CoPilotLogMetadata(
+            logging.getLogger(__name__),
+            task_id=entry.task_id,
+            session_id=entry.session_id,
+            user_id=entry.user_id,
+        )
+        log.info("Starting execution")
+
+        start_time = time.monotonic()
+
+        try:
+            # Run the async execution in our event loop
+            future = asyncio.run_coroutine_threadsafe(
+                self._execute_async(entry, cancel, cluster_lock, log),
+                self.execution_loop,
+            )
+
+            # Wait for completion, checking cancel periodically
+            while not future.done():
+                try:
+                    future.result(timeout=1.0)
+                except asyncio.TimeoutError:
+                    if cancel.is_set():
+                        log.info("Cancellation requested")
+                        future.cancel()
+                        break
+                    # Refresh cluster lock to maintain ownership
+                    cluster_lock.refresh()
+
+            if not future.cancelled():
+                # Get result to propagate any exceptions
+                future.result()
+
+            elapsed = time.monotonic() - start_time
+            log.info(f"Execution completed in {elapsed:.2f}s")
+
+        except Exception as e:
+            elapsed = time.monotonic() - start_time
+            log.error(f"Execution failed after {elapsed:.2f}s: {e}")
+            # Note: _execute_async already marks the task as failed before re-raising,
+            # so we don't call _mark_task_failed here to avoid duplicate error events.
+            raise
+
+    async def _execute_async(
+        self,
+        entry: CoPilotExecutionEntry,
+        cancel: threading.Event,
+        cluster_lock: ClusterLock,
+        log: CoPilotLogMetadata,
+    ):
+        """Async execution logic for CoPilot task.
+
+        This method calls the existing stream_chat_completion service function
+        and publishes results to the stream registry.
+
+        Args:
+            entry: The task payload
+            cancel: Threading event to signal cancellation
+            cluster_lock: Distributed lock for refresh
+            log: Structured logger for this task
+        """
+        last_refresh = time.monotonic()
+        refresh_interval = 30.0  # Refresh lock every 30 seconds
+
+        try:
+            # Choose service based on LaunchDarkly flag
+            config = ChatConfig()
+            use_sdk = await is_feature_enabled(
+                Flag.COPILOT_SDK,
+                entry.user_id or "anonymous",
+                default=config.use_claude_agent_sdk,
+            )
+            stream_fn = (
+                sdk_service.stream_chat_completion_sdk
+                if use_sdk
+                else copilot_service.stream_chat_completion
+            )
+            log.info(f"Using {'SDK' if use_sdk else 'standard'} service")
+
+            # Stream chat completion and publish chunks to Redis
+            async for chunk in stream_fn(
+                session_id=entry.session_id,
+                message=entry.message if entry.message else None,
+                is_user_message=entry.is_user_message,
+                user_id=entry.user_id,
+                context=entry.context,
+            ):
+                # Check for cancellation
+                if cancel.is_set():
+                    log.info("Cancelled during streaming")
+                    await stream_registry.publish_chunk(
+                        entry.task_id, StreamError(errorText="Operation cancelled")
+                    )
+                    await stream_registry.publish_chunk(
+                        entry.task_id, StreamFinishStep()
+                    )
+                    await stream_registry.publish_chunk(entry.task_id, StreamFinish())
+                    await stream_registry.mark_task_completed(
+                        entry.task_id, status="failed"
+                    )
+                    return
+
+                # Refresh cluster lock periodically
+                current_time = time.monotonic()
+                if current_time - last_refresh >= refresh_interval:
+                    cluster_lock.refresh()
+                    last_refresh = current_time
+
+                # Publish chunk to stream registry
+                await stream_registry.publish_chunk(entry.task_id, chunk)
+
+            # Mark task as completed
+            await stream_registry.mark_task_completed(entry.task_id, status="completed")
+            log.info("Task completed successfully")
+
+        except asyncio.CancelledError:
+            log.info("Task cancelled")
+            await stream_registry.mark_task_completed(entry.task_id, status="failed")
+            raise
+
+        except Exception as e:
+            log.error(f"Task failed: {e}")
+            await self._mark_task_failed(entry.task_id, str(e))
+            raise
+
+    async def _mark_task_failed(self, task_id: str, error_message: str):
+        """Mark a task as failed and publish error to stream registry."""
+        try:
+            await stream_registry.publish_chunk(
+                task_id, StreamError(errorText=error_message)
+            )
+            await stream_registry.publish_chunk(task_id, StreamFinishStep())
+            await stream_registry.publish_chunk(task_id, StreamFinish())
+            await stream_registry.mark_task_completed(task_id, status="failed")
+        except Exception as e:
+            logger.error(f"Failed to mark task {task_id} as failed: {e}")

autogpt_platform/backend/backend/copilot/executor/utils.py

@@ -0,0 +1,207 @@
+"""RabbitMQ queue configuration for CoPilot executor.
+
+Defines two exchanges and queues following the graph executor pattern:
+- 'copilot_execution' (DIRECT) for chat generation tasks
+- 'copilot_cancel' (FANOUT) for cancellation requests
+"""
+
+import logging
+
+from pydantic import BaseModel
+
+from backend.data.rabbitmq import Exchange, ExchangeType, Queue, RabbitMQConfig
+from backend.util.logging import TruncatedLogger, is_structured_logging_enabled
+
+logger = logging.getLogger(__name__)
+
+
+# ============ Logging Helper ============ #
+
+
+class CoPilotLogMetadata(TruncatedLogger):
+    """Structured logging helper for CoPilot executor.
+
+    In cloud environments (structured logging enabled), uses a simple prefix
+    and passes metadata via json_fields. In local environments, uses a detailed
+    prefix with all metadata key-value pairs for easier debugging.
+
+    Args:
+        logger: The underlying logger instance
+        max_length: Maximum log message length before truncation
+        **kwargs: Metadata key-value pairs (e.g., task_id="abc", session_id="xyz")
+            These are added to json_fields in cloud mode, or to the prefix in local mode.
+    """
+
+    def __init__(
+        self,
+        logger: logging.Logger,
+        max_length: int = 1000,
+        **kwargs: str | None,
+    ):
+        # Filter out None values
+        metadata = {k: v for k, v in kwargs.items() if v is not None}
+        metadata["component"] = "CoPilotExecutor"
+
+        if is_structured_logging_enabled():
+            prefix = "[CoPilotExecutor]"
+        else:
+            # Build prefix from metadata key-value pairs
+            meta_parts = "|".join(
+                f"{k}:{v}" for k, v in metadata.items() if k != "component"
+            )
+            prefix = (
+                f"[CoPilotExecutor|{meta_parts}]" if meta_parts else "[CoPilotExecutor]"
+            )
+
+        super().__init__(
+            logger,
+            max_length=max_length,
+            prefix=prefix,
+            metadata=metadata,
+        )
+
+
+# ============ Exchange and Queue Configuration ============ #
+
+COPILOT_EXECUTION_EXCHANGE = Exchange(
+    name="copilot_execution",
+    type=ExchangeType.DIRECT,
+    durable=True,
+    auto_delete=False,
+)
+COPILOT_EXECUTION_QUEUE_NAME = "copilot_execution_queue"
+COPILOT_EXECUTION_ROUTING_KEY = "copilot.run"
+
+COPILOT_CANCEL_EXCHANGE = Exchange(
+    name="copilot_cancel",
+    type=ExchangeType.FANOUT,
+    durable=True,
+    auto_delete=False,
+)
+COPILOT_CANCEL_QUEUE_NAME = "copilot_cancel_queue"
+
+# CoPilot operations can include extended thinking and agent generation
+# which may take 30+ minutes to complete
+COPILOT_CONSUMER_TIMEOUT_SECONDS = 60 * 60  # 1 hour
+
+# Graceful shutdown timeout - allow in-flight operations to complete
+GRACEFUL_SHUTDOWN_TIMEOUT_SECONDS = 30 * 60  # 30 minutes
+
+
+def create_copilot_queue_config() -> RabbitMQConfig:
+    """Create RabbitMQ configuration for CoPilot executor.
+
+    Defines two exchanges and queues:
+    - 'copilot_execution' (DIRECT) for chat generation tasks
+    - 'copilot_cancel' (FANOUT) for cancellation requests
+
+    Returns:
+        RabbitMQConfig with exchanges and queues defined
+    """
+    run_queue = Queue(
+        name=COPILOT_EXECUTION_QUEUE_NAME,
+        exchange=COPILOT_EXECUTION_EXCHANGE,
+        routing_key=COPILOT_EXECUTION_ROUTING_KEY,
+        durable=True,
+        auto_delete=False,
+        arguments={
+            # Extended consumer timeout for long-running LLM operations
+            # Default 30-minute timeout is insufficient for extended thinking
+            # and agent generation which can take 30+ minutes
+            "x-consumer-timeout": COPILOT_CONSUMER_TIMEOUT_SECONDS
+            * 1000,
+        },
+    )
+    cancel_queue = Queue(
+        name=COPILOT_CANCEL_QUEUE_NAME,
+        exchange=COPILOT_CANCEL_EXCHANGE,
+        routing_key="",  # not used for FANOUT
+        durable=True,
+        auto_delete=False,
+    )
+    return RabbitMQConfig(
+        vhost="/",
+        exchanges=[COPILOT_EXECUTION_EXCHANGE, COPILOT_CANCEL_EXCHANGE],
+        queues=[run_queue, cancel_queue],
+    )
+
+
+# ============ Message Models ============ #
+
+
+class CoPilotExecutionEntry(BaseModel):
+    """Task payload for CoPilot AI generation.
+
+    This model represents a chat generation task to be processed by the executor.
+    """
+
+    task_id: str
+    """Unique identifier for this task (used for stream registry)"""
+
+    session_id: str
+    """Chat session ID"""
+
+    user_id: str | None
+    """User ID (may be None for anonymous users)"""
+
+    operation_id: str
+    """Operation ID for webhook callbacks and completion tracking"""
+
+    message: str
+    """User's message to process"""
+
+    is_user_message: bool = True
+    """Whether the message is from the user (vs system/assistant)"""
+
+    context: dict[str, str] | None = None
+    """Optional context for the message (e.g., {url: str, content: str})"""
+
+
+class CancelCoPilotEvent(BaseModel):
+    """Event to cancel a CoPilot operation."""
+
+    task_id: str
+    """Task ID to cancel"""
+
+
+# ============ Queue Publishing Helpers ============ #
+
+
+async def enqueue_copilot_task(
+    task_id: str,
+    session_id: str,
+    user_id: str | None,
+    operation_id: str,
+    message: str,
+    is_user_message: bool = True,
+    context: dict[str, str] | None = None,
+) -> None:
+    """Enqueue a CoPilot task for processing by the executor service.
+
+    Args:
+        task_id: Unique identifier for this task (used for stream registry)
+        session_id: Chat session ID
+        user_id: User ID (may be None for anonymous users)
+        operation_id: Operation ID for webhook callbacks and completion tracking
+        message: User's message to process
+        is_user_message: Whether the message is from the user (vs system/assistant)
+        context: Optional context for the message (e.g., {url: str, content: str})
+    """
+    from backend.util.clients import get_async_copilot_queue
+
+    entry = CoPilotExecutionEntry(
+        task_id=task_id,
+        session_id=session_id,
+        user_id=user_id,
+        operation_id=operation_id,
+        message=message,
+        is_user_message=is_user_message,
+        context=context,
+    )
+
+    queue_client = await get_async_copilot_queue()
+    await queue_client.publish_message(
+        routing_key=COPILOT_EXECUTION_ROUTING_KEY,
+        message=entry.model_dump_json(),
+        exchange=COPILOT_EXECUTION_EXCHANGE,
+    )

autogpt_platform/backend/backend/copilot/model.py

@@ -23,26 +23,17 @@ from prisma.models import ChatMessage as PrismaChatMessage
 from prisma.models import ChatSession as PrismaChatSession
 from pydantic import BaseModel
 
+from backend.data.db_accessors import chat_db
 from backend.data.redis_client import get_redis_async
 from backend.util import json
 from backend.util.exceptions import DatabaseError, RedisError
 
-from . import db as chat_db
 from .config import ChatConfig
 
 logger = logging.getLogger(__name__)
 config = ChatConfig()
 
 
-def _parse_json_field(value: str | dict | list | None, default: Any = None) -> Any:
-    """Parse a JSON field that may be stored as string or already parsed."""
-    if value is None:
-        return default
-    if isinstance(value, str):
-        return json.loads(value)
-    return value
-
-
 # Redis cache key prefix for chat sessions
 CHAT_SESSION_CACHE_PREFIX = "chat:session:"
 
@@ -52,28 +43,7 @@ def _get_session_cache_key(session_id: str) -> str:
     return f"{CHAT_SESSION_CACHE_PREFIX}{session_id}"
 
 
-# Session-level locks to prevent race conditions during concurrent upserts.
-# Uses WeakValueDictionary to automatically garbage collect locks when no longer referenced,
-# preventing unbounded memory growth while maintaining lock semantics for active sessions.
-# Invalidation: Locks are auto-removed by GC when no coroutine holds a reference (after
-# async with lock: completes). Explicit cleanup also occurs in delete_chat_session().
-_session_locks: WeakValueDictionary[str, asyncio.Lock] = WeakValueDictionary()
-_session_locks_mutex = asyncio.Lock()
-
-
-async def _get_session_lock(session_id: str) -> asyncio.Lock:
-    """Get or create a lock for a specific session to prevent concurrent upserts.
-
-    Uses WeakValueDictionary for automatic cleanup: locks are garbage collected
-    when no coroutine holds a reference to them, preventing memory leaks from
-    unbounded growth of session locks.
-    """
-    async with _session_locks_mutex:
-        lock = _session_locks.get(session_id)
-        if lock is None:
-            lock = asyncio.Lock()
-            _session_locks[session_id] = lock
-        return lock
+# ===================== Chat data models ===================== #
 
 
 class ChatMessage(BaseModel):
@@ -85,6 +55,19 @@ class ChatMessage(BaseModel):
     tool_calls: list[dict] | None = None
     function_call: dict | None = None
 
+    @staticmethod
+    def from_db(prisma_message: PrismaChatMessage) -> "ChatMessage":
+        """Convert a Prisma ChatMessage to a Pydantic ChatMessage."""
+        return ChatMessage(
+            role=prisma_message.role,
+            content=prisma_message.content,
+            name=prisma_message.name,
+            tool_call_id=prisma_message.toolCallId,
+            refusal=prisma_message.refusal,
+            tool_calls=_parse_json_field(prisma_message.toolCalls),
+            function_call=_parse_json_field(prisma_message.functionCall),
+        )
+
 
 class Usage(BaseModel):
     prompt_tokens: int
@@ -138,26 +121,8 @@ class ChatSession(BaseModel):
         )
 
     @staticmethod
-    def from_db(
-        prisma_session: PrismaChatSession,
-        prisma_messages: list[PrismaChatMessage] | None = None,
-    ) -> "ChatSession":
-        """Convert Prisma models to Pydantic ChatSession."""
-        messages = []
-        if prisma_messages:
-            for msg in prisma_messages:
-                messages.append(
-                    ChatMessage(
-                        role=msg.role,
-                        content=msg.content,
-                        name=msg.name,
-                        tool_call_id=msg.toolCallId,
-                        refusal=msg.refusal,
-                        tool_calls=_parse_json_field(msg.toolCalls),
-                        function_call=_parse_json_field(msg.functionCall),
-                    )
-                )
-
+    def from_db(prisma_session: PrismaChatSession) -> "ChatSession":
+        """Convert Prisma ChatSession to Pydantic ChatSession."""
         # Parse JSON fields from Prisma
         credentials = _parse_json_field(prisma_session.credentials, default={})
         successful_agent_runs = _parse_json_field(
@@ -183,7 +148,11 @@ class ChatSession(BaseModel):
             session_id=prisma_session.id,
             user_id=prisma_session.userId,
             title=prisma_session.title,
-            messages=messages,
+            messages=(
+                [ChatMessage.from_db(m) for m in prisma_session.Messages]
+                if prisma_session.Messages
+                else []
+            ),
             usage=usage,
             credentials=credentials,
             started_at=prisma_session.createdAt,
@@ -322,38 +291,26 @@ class ChatSession(BaseModel):
         return self._merge_consecutive_assistant_messages(messages)
 
 
-async def _get_session_from_cache(session_id: str) -> ChatSession | None:
-    """Get a chat session from Redis cache."""
-    redis_key = _get_session_cache_key(session_id)
-    async_redis = await get_redis_async()
-    raw_session: bytes | None = await async_redis.get(redis_key)
-
-    if raw_session is None:
-        return None
-
-    try:
-        session = ChatSession.model_validate_json(raw_session)
-        logger.info(
-            f"Loading session {session_id} from cache: "
-            f"message_count={len(session.messages)}, "
-            f"roles={[m.role for m in session.messages]}"
-        )
-        return session
-    except Exception as e:
-        logger.error(f"Failed to deserialize session {session_id}: {e}", exc_info=True)
-        raise RedisError(f"Corrupted session data for {session_id}") from e
+def _parse_json_field(value: str | dict | list | None, default: Any = None) -> Any:
+    """Parse a JSON field that may be stored as string or already parsed."""
+    if value is None:
+        return default
+    if isinstance(value, str):
+        return json.loads(value)
+    return value
 
 
-async def _cache_session(session: ChatSession) -> None:
-    """Cache a chat session in Redis."""
-    redis_key = _get_session_cache_key(session.session_id)
-    async_redis = await get_redis_async()
-    await async_redis.setex(redis_key, config.session_ttl, session.model_dump_json())
+# ================ Chat cache + DB operations ================ #
+
+# NOTE: Database calls are automatically routed through DatabaseManager if Prisma is not
+#       connected directly.
 
 
 async def cache_chat_session(session: ChatSession) -> None:
-    """Cache a chat session without persisting to the database."""
-    await _cache_session(session)
+    """Cache a chat session in Redis (without persisting to the database)."""
+    redis_key = _get_session_cache_key(session.session_id)
+    async_redis = await get_redis_async()
+    await async_redis.setex(redis_key, config.session_ttl, session.model_dump_json())
 
 
 async def invalidate_session_cache(session_id: str) -> None:
@@ -371,80 +328,6 @@ async def invalidate_session_cache(session_id: str) -> None:
         logger.warning(f"Failed to invalidate session cache for {session_id}: {e}")
 
 
-async def _get_session_from_db(session_id: str) -> ChatSession | None:
-    """Get a chat session from the database."""
-    prisma_session = await chat_db.get_chat_session(session_id)
-    if not prisma_session:
-        return None
-
-    messages = prisma_session.Messages
-    logger.info(
-        f"Loading session {session_id} from DB: "
-        f"has_messages={messages is not None}, "
-        f"message_count={len(messages) if messages else 0}, "
-        f"roles={[m.role for m in messages] if messages else []}"
-    )
-
-    return ChatSession.from_db(prisma_session, messages)
-
-
-async def _save_session_to_db(
-    session: ChatSession, existing_message_count: int
-) -> None:
-    """Save or update a chat session in the database."""
-    # Check if session exists in DB
-    existing = await chat_db.get_chat_session(session.session_id)
-
-    if not existing:
-        # Create new session
-        await chat_db.create_chat_session(
-            session_id=session.session_id,
-            user_id=session.user_id,
-        )
-        existing_message_count = 0
-
-    # Calculate total tokens from usage
-    total_prompt = sum(u.prompt_tokens for u in session.usage)
-    total_completion = sum(u.completion_tokens for u in session.usage)
-
-    # Update session metadata
-    await chat_db.update_chat_session(
-        session_id=session.session_id,
-        credentials=session.credentials,
-        successful_agent_runs=session.successful_agent_runs,
-        successful_agent_schedules=session.successful_agent_schedules,
-        total_prompt_tokens=total_prompt,
-        total_completion_tokens=total_completion,
-    )
-
-    # Add new messages (only those after existing count)
-    new_messages = session.messages[existing_message_count:]
-    if new_messages:
-        messages_data = []
-        for msg in new_messages:
-            messages_data.append(
-                {
-                    "role": msg.role,
-                    "content": msg.content,
-                    "name": msg.name,
-                    "tool_call_id": msg.tool_call_id,
-                    "refusal": msg.refusal,
-                    "tool_calls": msg.tool_calls,
-                    "function_call": msg.function_call,
-                }
-            )
-        logger.info(
-            f"Saving {len(new_messages)} new messages to DB for session {session.session_id}: "
-            f"roles={[m['role'] for m in messages_data]}, "
-            f"start_sequence={existing_message_count}"
-        )
-        await chat_db.add_chat_messages_batch(
-            session_id=session.session_id,
-            messages=messages_data,
-            start_sequence=existing_message_count,
-        )
-
-
 async def get_chat_session(
     session_id: str,
     user_id: str | None = None,
@@ -476,7 +359,7 @@ async def get_chat_session(
         logger.warning(f"Unexpected cache error for session {session_id}: {e}")
 
     # Fall back to database
-    logger.info(f"Session {session_id} not in cache, checking database")
+    logger.debug(f"Session {session_id} not in cache, checking database")
     session = await _get_session_from_db(session_id)
 
     if session is None:
@@ -492,7 +375,7 @@ async def get_chat_session(
 
     # Cache the session from DB
     try:
-        await _cache_session(session)
+        await cache_chat_session(session)
         logger.info(f"Cached session {session_id} from database")
     except Exception as e:
         logger.warning(f"Failed to cache session {session_id}: {e}")
@@ -500,6 +383,44 @@ async def get_chat_session(
     return session
 
 
+async def _get_session_from_cache(session_id: str) -> ChatSession | None:
+    """Get a chat session from Redis cache."""
+    redis_key = _get_session_cache_key(session_id)
+    async_redis = await get_redis_async()
+    raw_session: bytes | None = await async_redis.get(redis_key)
+
+    if raw_session is None:
+        return None
+
+    try:
+        session = ChatSession.model_validate_json(raw_session)
+        logger.info(
+            f"Loading session {session_id} from cache: "
+            f"message_count={len(session.messages)}, "
+            f"roles={[m.role for m in session.messages]}"
+        )
+        return session
+    except Exception as e:
+        logger.error(f"Failed to deserialize session {session_id}: {e}", exc_info=True)
+        raise RedisError(f"Corrupted session data for {session_id}") from e
+
+
+async def _get_session_from_db(session_id: str) -> ChatSession | None:
+    """Get a chat session from the database."""
+    session = await chat_db().get_chat_session(session_id)
+    if not session:
+        return None
+
+    logger.info(
+        f"Loaded session {session_id} from DB: "
+        f"has_messages={bool(session.messages)}, "
+        f"message_count={len(session.messages)}, "
+        f"roles={[m.role for m in session.messages]}"
+    )
+
+    return session
+
+
 async def upsert_chat_session(
     session: ChatSession,
 ) -> ChatSession:
@@ -520,7 +441,7 @@ async def upsert_chat_session(
 
     async with lock:
         # Get existing message count from DB for incremental saves
-        existing_message_count = await chat_db.get_chat_session_message_count(
+        existing_message_count = await chat_db().get_chat_session_message_count(
             session.session_id
         )
 
@@ -537,7 +458,7 @@ async def upsert_chat_session(
 
         # Save to cache (best-effort, even if DB failed)
         try:
-            await _cache_session(session)
+            await cache_chat_session(session)
         except Exception as e:
             # If DB succeeded but cache failed, raise cache error
             if db_error is None:
@@ -558,6 +479,99 @@ async def upsert_chat_session(
         return session
 
 
+async def _save_session_to_db(
+    session: ChatSession, existing_message_count: int
+) -> None:
+    """Save or update a chat session in the database."""
+    db = chat_db()
+
+    # Check if session exists in DB
+    existing = await db.get_chat_session(session.session_id)
+
+    if not existing:
+        # Create new session
+        await db.create_chat_session(
+            session_id=session.session_id,
+            user_id=session.user_id,
+        )
+        existing_message_count = 0
+
+    # Calculate total tokens from usage
+    total_prompt = sum(u.prompt_tokens for u in session.usage)
+    total_completion = sum(u.completion_tokens for u in session.usage)
+
+    # Update session metadata
+    await db.update_chat_session(
+        session_id=session.session_id,
+        credentials=session.credentials,
+        successful_agent_runs=session.successful_agent_runs,
+        successful_agent_schedules=session.successful_agent_schedules,
+        total_prompt_tokens=total_prompt,
+        total_completion_tokens=total_completion,
+    )
+
+    # Add new messages (only those after existing count)
+    new_messages = session.messages[existing_message_count:]
+    if new_messages:
+        messages_data = []
+        for msg in new_messages:
+            messages_data.append(
+                {
+                    "role": msg.role,
+                    "content": msg.content,
+                    "name": msg.name,
+                    "tool_call_id": msg.tool_call_id,
+                    "refusal": msg.refusal,
+                    "tool_calls": msg.tool_calls,
+                    "function_call": msg.function_call,
+                }
+            )
+        logger.info(
+            f"Saving {len(new_messages)} new messages to DB for session {session.session_id}: "
+            f"roles={[m['role'] for m in messages_data]}, "
+            f"start_sequence={existing_message_count}"
+        )
+        await db.add_chat_messages_batch(
+            session_id=session.session_id,
+            messages=messages_data,
+            start_sequence=existing_message_count,
+        )
+
+
+async def append_and_save_message(session_id: str, message: ChatMessage) -> ChatSession:
+    """Atomically append a message to a session and persist it.
+
+    Acquires the session lock, re-fetches the latest session state,
+    appends the message, and saves — preventing message loss when
+    concurrent requests modify the same session.
+    """
+    lock = await _get_session_lock(session_id)
+
+    async with lock:
+        session = await get_chat_session(session_id)
+        if session is None:
+            raise ValueError(f"Session {session_id} not found")
+
+        session.messages.append(message)
+        existing_message_count = await chat_db().get_chat_session_message_count(
+            session_id
+        )
+
+        try:
+            await _save_session_to_db(session, existing_message_count)
+        except Exception as e:
+            raise DatabaseError(
+                f"Failed to persist message to session {session_id}"
+            ) from e
+
+        try:
+            await cache_chat_session(session)
+        except Exception as e:
+            logger.warning(f"Cache write failed for session {session_id}: {e}")
+
+        return session
+
+
 async def create_chat_session(user_id: str) -> ChatSession:
     """Create a new chat session and persist it.
 
@@ -570,7 +584,7 @@ async def create_chat_session(user_id: str) -> ChatSession:
 
     # Create in database first - fail fast if this fails
     try:
-        await chat_db.create_chat_session(
+        await chat_db().create_chat_session(
             session_id=session.session_id,
             user_id=user_id,
         )
@@ -582,7 +596,7 @@ async def create_chat_session(user_id: str) -> ChatSession:
 
     # Cache the session (best-effort optimization, DB is source of truth)
     try:
-        await _cache_session(session)
+        await cache_chat_session(session)
     except Exception as e:
         logger.warning(f"Failed to cache new session {session.session_id}: {e}")
 
@@ -600,13 +614,9 @@ async def get_user_sessions(
         A tuple of (sessions, total_count) where total_count is the overall
         number of sessions for the user (not just the current page).
     """
-    prisma_sessions = await chat_db.get_user_chat_sessions(user_id, limit, offset)
-    total_count = await chat_db.get_user_session_count(user_id)
-
-    sessions = []
-    for prisma_session in prisma_sessions:
-        # Convert without messages for listing (lighter weight)
-        sessions.append(ChatSession.from_db(prisma_session, None))
+    db = chat_db()
+    sessions = await db.get_user_chat_sessions(user_id, limit, offset)
+    total_count = await db.get_user_session_count(user_id)
 
     return sessions, total_count
 
@@ -624,7 +634,7 @@ async def delete_chat_session(session_id: str, user_id: str | None = None) -> bo
     """
     # Delete from database first (with optional user_id validation)
     # This confirms ownership before invalidating cache
-    deleted = await chat_db.delete_chat_session(session_id, user_id)
+    deleted = await chat_db().delete_chat_session(session_id, user_id)
 
     if not deleted:
         return False
@@ -659,20 +669,52 @@ async def update_session_title(session_id: str, title: str) -> bool:
         True if updated successfully, False otherwise.
     """
     try:
-        result = await chat_db.update_chat_session(session_id=session_id, title=title)
+        result = await chat_db().update_chat_session(session_id=session_id, title=title)
         if result is None:
             logger.warning(f"Session {session_id} not found for title update")
             return False
 
-        # Invalidate cache so next fetch gets updated title
+        # Update title in cache if it exists (instead of invalidating).
+        # This prevents race conditions where cache invalidation causes
+        # the frontend to see stale DB data while streaming is still in progress.
         try:
-            redis_key = _get_session_cache_key(session_id)
-            async_redis = await get_redis_async()
-            await async_redis.delete(redis_key)
+            cached = await _get_session_from_cache(session_id)
+            if cached:
+                cached.title = title
+                await cache_chat_session(cached)
         except Exception as e:
-            logger.warning(f"Failed to invalidate cache for session {session_id}: {e}")
+            # Not critical - title will be correct on next full cache refresh
+            logger.warning(
+                f"Failed to update title in cache for session {session_id}: {e}"
+            )
 
         return True
     except Exception as e:
         logger.error(f"Failed to update title for session {session_id}: {e}")
         return False
+
+
+# ==================== Chat session locks ==================== #
+
+_session_locks: WeakValueDictionary[str, asyncio.Lock] = WeakValueDictionary()
+_session_locks_mutex = asyncio.Lock()
+
+
+async def _get_session_lock(session_id: str) -> asyncio.Lock:
+    """Get or create a lock for a specific session to prevent concurrent upserts.
+
+    This was originally added to solve the specific problem of race conditions between
+    the session title thread and the conversation thread, which always occurs on the
+    same instance as we prevent rapid request sends on the frontend.
+
+    Uses WeakValueDictionary for automatic cleanup: locks are garbage collected
+    when no coroutine holds a reference to them, preventing memory leaks from
+    unbounded growth of session locks. Explicit cleanup also occurs
+    in `delete_chat_session()`.
+    """
+    async with _session_locks_mutex:
+        lock = _session_locks.get(session_id)
+        if lock is None:
+            lock = asyncio.Lock()
+            _session_locks[session_id] = lock
+        return lock

autogpt_platform/backend/backend/copilot/sdk/__init__.py

@@ -0,0 +1,14 @@
+"""Claude Agent SDK integration for CoPilot.
+
+This module provides the integration layer between the Claude Agent SDK
+and the existing CoPilot tool system, enabling drop-in replacement of
+the current LLM orchestration with the battle-tested Claude Agent SDK.
+"""
+
+from .service import stream_chat_completion_sdk
+from .tool_adapter import create_copilot_mcp_server
+
+__all__ = [
+    "stream_chat_completion_sdk",
+    "create_copilot_mcp_server",
+]

autogpt_platform/backend/backend/copilot/sdk/response_adapter.py

@@ -0,0 +1,201 @@
+"""Response adapter for converting Claude Agent SDK messages to Vercel AI SDK format.
+
+This module provides the adapter layer that converts streaming messages from
+the Claude Agent SDK into the Vercel AI SDK UI Stream Protocol format that
+the frontend expects.
+"""
+
+import json
+import logging
+import uuid
+
+from claude_agent_sdk import (
+    AssistantMessage,
+    Message,
+    ResultMessage,
+    SystemMessage,
+    TextBlock,
+    ToolResultBlock,
+    ToolUseBlock,
+    UserMessage,
+)
+
+from backend.copilot.response_model import (
+    StreamBaseResponse,
+    StreamError,
+    StreamFinish,
+    StreamFinishStep,
+    StreamStart,
+    StreamStartStep,
+    StreamTextDelta,
+    StreamTextEnd,
+    StreamTextStart,
+    StreamToolInputAvailable,
+    StreamToolInputStart,
+    StreamToolOutputAvailable,
+)
+
+from .tool_adapter import MCP_TOOL_PREFIX, pop_pending_tool_output
+
+logger = logging.getLogger(__name__)
+
+
+class SDKResponseAdapter:
+    """Adapter for converting Claude Agent SDK messages to Vercel AI SDK format.
+
+    This class maintains state during a streaming session to properly track
+    text blocks, tool calls, and message lifecycle.
+    """
+
+    def __init__(self, message_id: str | None = None):
+        self.message_id = message_id or str(uuid.uuid4())
+        self.text_block_id = str(uuid.uuid4())
+        self.has_started_text = False
+        self.has_ended_text = False
+        self.current_tool_calls: dict[str, dict[str, str]] = {}
+        self.task_id: str | None = None
+        self.step_open = False
+
+    def set_task_id(self, task_id: str) -> None:
+        """Set the task ID for reconnection support."""
+        self.task_id = task_id
+
+    def convert_message(self, sdk_message: Message) -> list[StreamBaseResponse]:
+        """Convert a single SDK message to Vercel AI SDK format."""
+        responses: list[StreamBaseResponse] = []
+
+        if isinstance(sdk_message, SystemMessage):
+            if sdk_message.subtype == "init":
+                responses.append(
+                    StreamStart(messageId=self.message_id, taskId=self.task_id)
+                )
+                # Open the first step (matches non-SDK: StreamStart then StreamStartStep)
+                responses.append(StreamStartStep())
+                self.step_open = True
+
+        elif isinstance(sdk_message, AssistantMessage):
+            # After tool results, the SDK sends a new AssistantMessage for the
+            # next LLM turn. Open a new step if the previous one was closed.
+            if not self.step_open:
+                responses.append(StreamStartStep())
+                self.step_open = True
+
+            for block in sdk_message.content:
+                if isinstance(block, TextBlock):
+                    if block.text:
+                        self._ensure_text_started(responses)
+                        responses.append(
+                            StreamTextDelta(id=self.text_block_id, delta=block.text)
+                        )
+
+                elif isinstance(block, ToolUseBlock):
+                    self._end_text_if_open(responses)
+
+                    # Strip MCP prefix so frontend sees "find_block"
+                    # instead of "mcp__copilot__find_block".
+                    tool_name = block.name.removeprefix(MCP_TOOL_PREFIX)
+
+                    responses.append(
+                        StreamToolInputStart(toolCallId=block.id, toolName=tool_name)
+                    )
+                    responses.append(
+                        StreamToolInputAvailable(
+                            toolCallId=block.id,
+                            toolName=tool_name,
+                            input=block.input,
+                        )
+                    )
+                    self.current_tool_calls[block.id] = {"name": tool_name}
+
+        elif isinstance(sdk_message, UserMessage):
+            # UserMessage carries tool results back from tool execution.
+            content = sdk_message.content
+            blocks = content if isinstance(content, list) else []
+            for block in blocks:
+                if isinstance(block, ToolResultBlock) and block.tool_use_id:
+                    tool_info = self.current_tool_calls.get(block.tool_use_id, {})
+                    tool_name = tool_info.get("name", "unknown")
+
+                    # Prefer the stashed full output over the SDK's
+                    # (potentially truncated) ToolResultBlock content.
+                    # The SDK truncates large results, writing them to disk,
+                    # which breaks frontend widget parsing.
+                    output = pop_pending_tool_output(tool_name) or (
+                        _extract_tool_output(block.content)
+                    )
+
+                    responses.append(
+                        StreamToolOutputAvailable(
+                            toolCallId=block.tool_use_id,
+                            toolName=tool_name,
+                            output=output,
+                            success=not (block.is_error or False),
+                        )
+                    )
+
+            # Close the current step after tool results — the next
+            # AssistantMessage will open a new step for the continuation.
+            if self.step_open:
+                responses.append(StreamFinishStep())
+                self.step_open = False
+
+        elif isinstance(sdk_message, ResultMessage):
+            self._end_text_if_open(responses)
+            # Close the step before finishing.
+            if self.step_open:
+                responses.append(StreamFinishStep())
+                self.step_open = False
+
+            if sdk_message.subtype == "success":
+                responses.append(StreamFinish())
+            elif sdk_message.subtype in ("error", "error_during_execution"):
+                error_msg = getattr(sdk_message, "result", None) or "Unknown error"
+                responses.append(
+                    StreamError(errorText=str(error_msg), code="sdk_error")
+                )
+                responses.append(StreamFinish())
+            else:
+                logger.warning(
+                    f"Unexpected ResultMessage subtype: {sdk_message.subtype}"
+                )
+                responses.append(StreamFinish())
+
+        else:
+            logger.debug(f"Unhandled SDK message type: {type(sdk_message).__name__}")
+
+        return responses
+
+    def _ensure_text_started(self, responses: list[StreamBaseResponse]) -> None:
+        """Start (or restart) a text block if needed."""
+        if not self.has_started_text or self.has_ended_text:
+            if self.has_ended_text:
+                self.text_block_id = str(uuid.uuid4())
+                self.has_ended_text = False
+            responses.append(StreamTextStart(id=self.text_block_id))
+            self.has_started_text = True
+
+    def _end_text_if_open(self, responses: list[StreamBaseResponse]) -> None:
+        """End the current text block if one is open."""
+        if self.has_started_text and not self.has_ended_text:
+            responses.append(StreamTextEnd(id=self.text_block_id))
+            self.has_ended_text = True
+
+
+def _extract_tool_output(content: str | list[dict[str, str]] | None) -> str:
+    """Extract a string output from a ToolResultBlock's content field."""
+    if isinstance(content, str):
+        return content
+    if isinstance(content, list):
+        parts = [item.get("text", "") for item in content if item.get("type") == "text"]
+        if parts:
+            return "".join(parts)
+        try:
+            return json.dumps(content)
+        except (TypeError, ValueError):
+            return str(content)
+    if content is None:
+        return ""
+    try:
+        return json.dumps(content)
+    except (TypeError, ValueError):
+        return str(content)

autogpt_platform/backend/backend/copilot/sdk/response_adapter_test.py

@@ -0,0 +1,366 @@
+"""Unit tests for the SDK response adapter."""
+
+from claude_agent_sdk import (
+    AssistantMessage,
+    ResultMessage,
+    SystemMessage,
+    TextBlock,
+    ToolResultBlock,
+    ToolUseBlock,
+    UserMessage,
+)
+
+from backend.copilot.response_model import (
+    StreamBaseResponse,
+    StreamError,
+    StreamFinish,
+    StreamFinishStep,
+    StreamStart,
+    StreamStartStep,
+    StreamTextDelta,
+    StreamTextEnd,
+    StreamTextStart,
+    StreamToolInputAvailable,
+    StreamToolInputStart,
+    StreamToolOutputAvailable,
+)
+
+from .response_adapter import SDKResponseAdapter
+from .tool_adapter import MCP_TOOL_PREFIX
+
+
+def _adapter() -> SDKResponseAdapter:
+    a = SDKResponseAdapter(message_id="msg-1")
+    a.set_task_id("task-1")
+    return a
+
+
+# -- SystemMessage -----------------------------------------------------------
+
+
+def test_system_init_emits_start_and_step():
+    adapter = _adapter()
+    results = adapter.convert_message(SystemMessage(subtype="init", data={}))
+    assert len(results) == 2
+    assert isinstance(results[0], StreamStart)
+    assert results[0].messageId == "msg-1"
+    assert results[0].taskId == "task-1"
+    assert isinstance(results[1], StreamStartStep)
+
+
+def test_system_non_init_emits_nothing():
+    adapter = _adapter()
+    results = adapter.convert_message(SystemMessage(subtype="other", data={}))
+    assert results == []
+
+
+# -- AssistantMessage with TextBlock -----------------------------------------
+
+
+def test_text_block_emits_step_start_and_delta():
+    adapter = _adapter()
+    msg = AssistantMessage(content=[TextBlock(text="hello")], model="test")
+    results = adapter.convert_message(msg)
+    assert len(results) == 3
+    assert isinstance(results[0], StreamStartStep)
+    assert isinstance(results[1], StreamTextStart)
+    assert isinstance(results[2], StreamTextDelta)
+    assert results[2].delta == "hello"
+
+
+def test_empty_text_block_emits_only_step():
+    adapter = _adapter()
+    msg = AssistantMessage(content=[TextBlock(text="")], model="test")
+    results = adapter.convert_message(msg)
+    # Empty text skipped, but step still opens
+    assert len(results) == 1
+    assert isinstance(results[0], StreamStartStep)
+
+
+def test_multiple_text_deltas_reuse_block_id():
+    adapter = _adapter()
+    msg1 = AssistantMessage(content=[TextBlock(text="a")], model="test")
+    msg2 = AssistantMessage(content=[TextBlock(text="b")], model="test")
+    r1 = adapter.convert_message(msg1)
+    r2 = adapter.convert_message(msg2)
+    # First gets step+start+delta, second only delta (block & step already started)
+    assert len(r1) == 3
+    assert isinstance(r1[0], StreamStartStep)
+    assert isinstance(r1[1], StreamTextStart)
+    assert len(r2) == 1
+    assert isinstance(r2[0], StreamTextDelta)
+    assert r1[1].id == r2[0].id  # same block ID
+
+
+# -- AssistantMessage with ToolUseBlock --------------------------------------
+
+
+def test_tool_use_emits_input_start_and_available():
+    """Tool names arrive with MCP prefix and should be stripped for the frontend."""
+    adapter = _adapter()
+    msg = AssistantMessage(
+        content=[
+            ToolUseBlock(
+                id="tool-1",
+                name=f"{MCP_TOOL_PREFIX}find_agent",
+                input={"q": "x"},
+            )
+        ],
+        model="test",
+    )
+    results = adapter.convert_message(msg)
+    assert len(results) == 3
+    assert isinstance(results[0], StreamStartStep)
+    assert isinstance(results[1], StreamToolInputStart)
+    assert results[1].toolCallId == "tool-1"
+    assert results[1].toolName == "find_agent"  # prefix stripped
+    assert isinstance(results[2], StreamToolInputAvailable)
+    assert results[2].toolName == "find_agent"  # prefix stripped
+    assert results[2].input == {"q": "x"}
+
+
+def test_text_then_tool_ends_text_block():
+    adapter = _adapter()
+    text_msg = AssistantMessage(content=[TextBlock(text="thinking...")], model="test")
+    tool_msg = AssistantMessage(
+        content=[ToolUseBlock(id="t1", name=f"{MCP_TOOL_PREFIX}tool", input={})],
+        model="test",
+    )
+    adapter.convert_message(text_msg)  # opens step + text
+    results = adapter.convert_message(tool_msg)
+    # Step already open, so: TextEnd, ToolInputStart, ToolInputAvailable
+    assert len(results) == 3
+    assert isinstance(results[0], StreamTextEnd)
+    assert isinstance(results[1], StreamToolInputStart)
+
+
+# -- UserMessage with ToolResultBlock ----------------------------------------
+
+
+def test_tool_result_emits_output_and_finish_step():
+    adapter = _adapter()
+    # First register the tool call (opens step) — SDK sends prefixed name
+    tool_msg = AssistantMessage(
+        content=[ToolUseBlock(id="t1", name=f"{MCP_TOOL_PREFIX}find_agent", input={})],
+        model="test",
+    )
+    adapter.convert_message(tool_msg)
+
+    # Now send tool result
+    result_msg = UserMessage(
+        content=[ToolResultBlock(tool_use_id="t1", content="found 3 agents")]
+    )
+    results = adapter.convert_message(result_msg)
+    assert len(results) == 2
+    assert isinstance(results[0], StreamToolOutputAvailable)
+    assert results[0].toolCallId == "t1"
+    assert results[0].toolName == "find_agent"  # prefix stripped
+    assert results[0].output == "found 3 agents"
+    assert results[0].success is True
+    assert isinstance(results[1], StreamFinishStep)
+
+
+def test_tool_result_error():
+    adapter = _adapter()
+    adapter.convert_message(
+        AssistantMessage(
+            content=[
+                ToolUseBlock(id="t1", name=f"{MCP_TOOL_PREFIX}run_agent", input={})
+            ],
+            model="test",
+        )
+    )
+    result_msg = UserMessage(
+        content=[ToolResultBlock(tool_use_id="t1", content="timeout", is_error=True)]
+    )
+    results = adapter.convert_message(result_msg)
+    assert isinstance(results[0], StreamToolOutputAvailable)
+    assert results[0].success is False
+    assert isinstance(results[1], StreamFinishStep)
+
+
+def test_tool_result_list_content():
+    adapter = _adapter()
+    adapter.convert_message(
+        AssistantMessage(
+            content=[ToolUseBlock(id="t1", name=f"{MCP_TOOL_PREFIX}tool", input={})],
+            model="test",
+        )
+    )
+    result_msg = UserMessage(
+        content=[
+            ToolResultBlock(
+                tool_use_id="t1",
+                content=[
+                    {"type": "text", "text": "line1"},
+                    {"type": "text", "text": "line2"},
+                ],
+            )
+        ]
+    )
+    results = adapter.convert_message(result_msg)
+    assert isinstance(results[0], StreamToolOutputAvailable)
+    assert results[0].output == "line1line2"
+    assert isinstance(results[1], StreamFinishStep)
+
+
+def test_string_user_message_ignored():
+    """A plain string UserMessage (not tool results) produces no output."""
+    adapter = _adapter()
+    results = adapter.convert_message(UserMessage(content="hello"))
+    assert results == []
+
+
+# -- ResultMessage -----------------------------------------------------------
+
+
+def test_result_success_emits_finish_step_and_finish():
+    adapter = _adapter()
+    # Start some text first (opens step)
+    adapter.convert_message(
+        AssistantMessage(content=[TextBlock(text="done")], model="test")
+    )
+    msg = ResultMessage(
+        subtype="success",
+        duration_ms=100,
+        duration_api_ms=50,
+        is_error=False,
+        num_turns=1,
+        session_id="s1",
+    )
+    results = adapter.convert_message(msg)
+    # TextEnd + FinishStep + StreamFinish
+    assert len(results) == 3
+    assert isinstance(results[0], StreamTextEnd)
+    assert isinstance(results[1], StreamFinishStep)
+    assert isinstance(results[2], StreamFinish)
+
+
+def test_result_error_emits_error_and_finish():
+    adapter = _adapter()
+    msg = ResultMessage(
+        subtype="error",
+        duration_ms=100,
+        duration_api_ms=50,
+        is_error=True,
+        num_turns=0,
+        session_id="s1",
+        result="API rate limited",
+    )
+    results = adapter.convert_message(msg)
+    # No step was open, so no FinishStep — just Error + Finish
+    assert len(results) == 2
+    assert isinstance(results[0], StreamError)
+    assert "API rate limited" in results[0].errorText
+    assert isinstance(results[1], StreamFinish)
+
+
+# -- Text after tools (new block ID) ----------------------------------------
+
+
+def test_text_after_tool_gets_new_block_id():
+    adapter = _adapter()
+    # Text -> Tool -> ToolResult -> Text should get a new text block ID and step
+    adapter.convert_message(
+        AssistantMessage(content=[TextBlock(text="before")], model="test")
+    )
+    adapter.convert_message(
+        AssistantMessage(
+            content=[ToolUseBlock(id="t1", name=f"{MCP_TOOL_PREFIX}tool", input={})],
+            model="test",
+        )
+    )
+    # Send tool result (closes step)
+    adapter.convert_message(
+        UserMessage(content=[ToolResultBlock(tool_use_id="t1", content="ok")])
+    )
+    results = adapter.convert_message(
+        AssistantMessage(content=[TextBlock(text="after")], model="test")
+    )
+    # Should get StreamStartStep (new step) + StreamTextStart (new block) + StreamTextDelta
+    assert len(results) == 3
+    assert isinstance(results[0], StreamStartStep)
+    assert isinstance(results[1], StreamTextStart)
+    assert isinstance(results[2], StreamTextDelta)
+    assert results[2].delta == "after"
+
+
+# -- Full conversation flow --------------------------------------------------
+
+
+def test_full_conversation_flow():
+    """Simulate a complete conversation: init -> text -> tool -> result -> text -> finish."""
+    adapter = _adapter()
+    all_responses: list[StreamBaseResponse] = []
+
+    # 1. Init
+    all_responses.extend(
+        adapter.convert_message(SystemMessage(subtype="init", data={}))
+    )
+    # 2. Assistant text
+    all_responses.extend(
+        adapter.convert_message(
+            AssistantMessage(content=[TextBlock(text="Let me search")], model="test")
+        )
+    )
+    # 3. Tool use
+    all_responses.extend(
+        adapter.convert_message(
+            AssistantMessage(
+                content=[
+                    ToolUseBlock(
+                        id="t1",
+                        name=f"{MCP_TOOL_PREFIX}find_agent",
+                        input={"query": "email"},
+                    )
+                ],
+                model="test",
+            )
+        )
+    )
+    # 4. Tool result
+    all_responses.extend(
+        adapter.convert_message(
+            UserMessage(
+                content=[ToolResultBlock(tool_use_id="t1", content="Found 2 agents")]
+            )
+        )
+    )
+    # 5. More text
+    all_responses.extend(
+        adapter.convert_message(
+            AssistantMessage(content=[TextBlock(text="I found 2")], model="test")
+        )
+    )
+    # 6. Result
+    all_responses.extend(
+        adapter.convert_message(
+            ResultMessage(
+                subtype="success",
+                duration_ms=500,
+                duration_api_ms=400,
+                is_error=False,
+                num_turns=2,
+                session_id="s1",
+            )
+        )
+    )
+
+    types = [type(r).__name__ for r in all_responses]
+    assert types == [
+        "StreamStart",
+        "StreamStartStep",  # step 1: text + tool call
+        "StreamTextStart",
+        "StreamTextDelta",  # "Let me search"
+        "StreamTextEnd",  # closed before tool
+        "StreamToolInputStart",
+        "StreamToolInputAvailable",
+        "StreamToolOutputAvailable",  # tool result
+        "StreamFinishStep",  # step 1 closed after tool result
+        "StreamStartStep",  # step 2: continuation text
+        "StreamTextStart",  # new block after tool
+        "StreamTextDelta",  # "I found 2"
+        "StreamTextEnd",  # closed by result
+        "StreamFinishStep",  # step 2 closed
+        "StreamFinish",
+    ]

autogpt_platform/backend/backend/copilot/sdk/security_hooks.py

@@ -0,0 +1,335 @@
+"""Security hooks for Claude Agent SDK integration.
+
+This module provides security hooks that validate tool calls before execution,
+ensuring multi-user isolation and preventing unauthorized operations.
+"""
+
+import json
+import logging
+import os
+import re
+from collections.abc import Callable
+from typing import Any, cast
+
+from .tool_adapter import MCP_TOOL_PREFIX
+
+logger = logging.getLogger(__name__)
+
+# Tools that are blocked entirely (CLI/system access).
+# "Bash" (capital) is the SDK built-in — it's NOT in allowed_tools but blocked
+# here as defence-in-depth.  The agent uses mcp__copilot__bash_exec instead,
+# which has kernel-level network isolation (unshare --net).
+BLOCKED_TOOLS = {
+    "Bash",
+    "bash",
+    "shell",
+    "exec",
+    "terminal",
+    "command",
+}
+
+# Tools allowed only when their path argument stays within the SDK workspace.
+# The SDK uses these to handle oversized tool results (writes to tool-results/
+# files, then reads them back) and for workspace file operations.
+WORKSPACE_SCOPED_TOOLS = {"Read", "Write", "Edit", "Glob", "Grep"}
+
+# Dangerous patterns in tool inputs
+DANGEROUS_PATTERNS = [
+    r"sudo",
+    r"rm\s+-rf",
+    r"dd\s+if=",
+    r"/etc/passwd",
+    r"/etc/shadow",
+    r"chmod\s+777",
+    r"curl\s+.*\|.*sh",
+    r"wget\s+.*\|.*sh",
+    r"eval\s*\(",
+    r"exec\s*\(",
+    r"__import__",
+    r"os\.system",
+    r"subprocess",
+]
+
+
+def _deny(reason: str) -> dict[str, Any]:
+    """Return a hook denial response."""
+    return {
+        "hookSpecificOutput": {
+            "hookEventName": "PreToolUse",
+            "permissionDecision": "deny",
+            "permissionDecisionReason": reason,
+        }
+    }
+
+
+def _validate_workspace_path(
+    tool_name: str, tool_input: dict[str, Any], sdk_cwd: str | None
+) -> dict[str, Any]:
+    """Validate that a workspace-scoped tool only accesses allowed paths.
+
+    Allowed directories:
+    - The SDK working directory (``/tmp/copilot-<session>/``)
+    - The SDK tool-results directory (``~/.claude/projects/…/tool-results/``)
+    """
+    path = tool_input.get("file_path") or tool_input.get("path") or ""
+    if not path:
+        # Glob/Grep without a path default to cwd which is already sandboxed
+        return {}
+
+    # Resolve relative paths against sdk_cwd (the SDK sets cwd so the LLM
+    # naturally uses relative paths like "test.txt" instead of absolute ones).
+    # Tilde paths (~/) are home-dir references, not relative — expand first.
+    if path.startswith("~"):
+        resolved = os.path.realpath(os.path.expanduser(path))
+    elif not os.path.isabs(path) and sdk_cwd:
+        resolved = os.path.realpath(os.path.join(sdk_cwd, path))
+    else:
+        resolved = os.path.realpath(path)
+
+    # Allow access within the SDK working directory
+    if sdk_cwd:
+        norm_cwd = os.path.realpath(sdk_cwd)
+        if resolved.startswith(norm_cwd + os.sep) or resolved == norm_cwd:
+            return {}
+
+    # Allow access to ~/.claude/projects/*/tool-results/ (big tool results)
+    claude_dir = os.path.realpath(os.path.expanduser("~/.claude/projects"))
+    tool_results_seg = os.sep + "tool-results" + os.sep
+    if resolved.startswith(claude_dir + os.sep) and tool_results_seg in resolved:
+        return {}
+
+    logger.warning(
+        f"Blocked {tool_name} outside workspace: {path} (resolved={resolved})"
+    )
+    workspace_hint = f" Allowed workspace: {sdk_cwd}" if sdk_cwd else ""
+    return _deny(
+        f"[SECURITY] Tool '{tool_name}' can only access files within the workspace "
+        f"directory.{workspace_hint} "
+        "This is enforced by the platform and cannot be bypassed."
+    )
+
+
+def _validate_tool_access(
+    tool_name: str, tool_input: dict[str, Any], sdk_cwd: str | None = None
+) -> dict[str, Any]:
+    """Validate that a tool call is allowed.
+
+    Returns:
+        Empty dict to allow, or dict with hookSpecificOutput to deny
+    """
+    # Block forbidden tools
+    if tool_name in BLOCKED_TOOLS:
+        logger.warning(f"Blocked tool access attempt: {tool_name}")
+        return _deny(
+            f"[SECURITY] Tool '{tool_name}' is blocked for security. "
+            "This is enforced by the platform and cannot be bypassed. "
+            "Use the CoPilot-specific MCP tools instead."
+        )
+
+    # Workspace-scoped tools: allowed only within the SDK workspace directory
+    if tool_name in WORKSPACE_SCOPED_TOOLS:
+        return _validate_workspace_path(tool_name, tool_input, sdk_cwd)
+
+    # Check for dangerous patterns in tool input
+    # Use json.dumps for predictable format (str() produces Python repr)
+    input_str = json.dumps(tool_input) if tool_input else ""
+
+    for pattern in DANGEROUS_PATTERNS:
+        if re.search(pattern, input_str, re.IGNORECASE):
+            logger.warning(
+                f"Blocked dangerous pattern in tool input: {pattern} in {tool_name}"
+            )
+            return _deny(
+                "[SECURITY] Input contains a blocked pattern. "
+                "This is enforced by the platform and cannot be bypassed."
+            )
+
+    return {}
+
+
+def _validate_user_isolation(
+    tool_name: str, tool_input: dict[str, Any], user_id: str | None
+) -> dict[str, Any]:
+    """Validate that tool calls respect user isolation."""
+    # For workspace file tools, ensure path doesn't escape
+    if "workspace" in tool_name.lower():
+        path = tool_input.get("path", "") or tool_input.get("file_path", "")
+        if path:
+            # Check for path traversal
+            if ".." in path or path.startswith("/"):
+                logger.warning(
+                    f"Blocked path traversal attempt: {path} by user {user_id}"
+                )
+                return {
+                    "hookSpecificOutput": {
+                        "hookEventName": "PreToolUse",
+                        "permissionDecision": "deny",
+                        "permissionDecisionReason": "Path traversal not allowed",
+                    }
+                }
+
+    return {}
+
+
+def create_security_hooks(
+    user_id: str | None,
+    sdk_cwd: str | None = None,
+    max_subtasks: int = 3,
+    on_stop: Callable[[str, str], None] | None = None,
+) -> dict[str, Any]:
+    """Create the security hooks configuration for Claude Agent SDK.
+
+    Includes security validation and observability hooks:
+    - PreToolUse: Security validation before tool execution
+    - PostToolUse: Log successful tool executions
+    - PostToolUseFailure: Log and handle failed tool executions
+    - PreCompact: Log context compaction events (SDK handles compaction automatically)
+    - Stop: Capture transcript path for stateless resume (when *on_stop* is provided)
+
+    Args:
+        user_id: Current user ID for isolation validation
+        sdk_cwd: SDK working directory for workspace-scoped tool validation
+        max_subtasks: Maximum Task (sub-agent) spawns allowed per session
+        on_stop: Callback ``(transcript_path, sdk_session_id)`` invoked when
+            the SDK finishes processing — used to read the JSONL transcript
+            before the CLI process exits.
+
+    Returns:
+        Hooks configuration dict for ClaudeAgentOptions
+    """
+    try:
+        from claude_agent_sdk import HookMatcher
+        from claude_agent_sdk.types import HookContext, HookInput, SyncHookJSONOutput
+
+        # Per-session counter for Task sub-agent spawns
+        task_spawn_count = 0
+
+        async def pre_tool_use_hook(
+            input_data: HookInput,
+            tool_use_id: str | None,
+            context: HookContext,
+        ) -> SyncHookJSONOutput:
+            """Combined pre-tool-use validation hook."""
+            nonlocal task_spawn_count
+            _ = context  # unused but required by signature
+            tool_name = cast(str, input_data.get("tool_name", ""))
+            tool_input = cast(dict[str, Any], input_data.get("tool_input", {}))
+
+            # Rate-limit Task (sub-agent) spawns per session
+            if tool_name == "Task":
+                task_spawn_count += 1
+                if task_spawn_count > max_subtasks:
+                    logger.warning(
+                        f"[SDK] Task limit reached ({max_subtasks}), user={user_id}"
+                    )
+                    return cast(
+                        SyncHookJSONOutput,
+                        _deny(
+                            f"Maximum {max_subtasks} sub-tasks per session. "
+                            "Please continue in the main conversation."
+                        ),
+                    )
+
+            # Strip MCP prefix for consistent validation
+            is_copilot_tool = tool_name.startswith(MCP_TOOL_PREFIX)
+            clean_name = tool_name.removeprefix(MCP_TOOL_PREFIX)
+
+            # Only block non-CoPilot tools; our MCP-registered tools
+            # (including Read for oversized results) are already sandboxed.
+            if not is_copilot_tool:
+                result = _validate_tool_access(clean_name, tool_input, sdk_cwd)
+                if result:
+                    return cast(SyncHookJSONOutput, result)
+
+            # Validate user isolation
+            result = _validate_user_isolation(clean_name, tool_input, user_id)
+            if result:
+                return cast(SyncHookJSONOutput, result)
+
+            logger.debug(f"[SDK] Tool start: {tool_name}, user={user_id}")
+            return cast(SyncHookJSONOutput, {})
+
+        async def post_tool_use_hook(
+            input_data: HookInput,
+            tool_use_id: str | None,
+            context: HookContext,
+        ) -> SyncHookJSONOutput:
+            """Log successful tool executions for observability."""
+            _ = context
+            tool_name = cast(str, input_data.get("tool_name", ""))
+            logger.debug(f"[SDK] Tool success: {tool_name}, tool_use_id={tool_use_id}")
+            return cast(SyncHookJSONOutput, {})
+
+        async def post_tool_failure_hook(
+            input_data: HookInput,
+            tool_use_id: str | None,
+            context: HookContext,
+        ) -> SyncHookJSONOutput:
+            """Log failed tool executions for debugging."""
+            _ = context
+            tool_name = cast(str, input_data.get("tool_name", ""))
+            error = input_data.get("error", "Unknown error")
+            logger.warning(
+                f"[SDK] Tool failed: {tool_name}, error={error}, "
+                f"user={user_id}, tool_use_id={tool_use_id}"
+            )
+            return cast(SyncHookJSONOutput, {})
+
+        async def pre_compact_hook(
+            input_data: HookInput,
+            tool_use_id: str | None,
+            context: HookContext,
+        ) -> SyncHookJSONOutput:
+            """Log when SDK triggers context compaction.
+
+            The SDK automatically compacts conversation history when it grows too large.
+            This hook provides visibility into when compaction happens.
+            """
+            _ = context, tool_use_id
+            trigger = input_data.get("trigger", "auto")
+            logger.info(
+                f"[SDK] Context compaction triggered: {trigger}, user={user_id}"
+            )
+            return cast(SyncHookJSONOutput, {})
+
+        # --- Stop hook: capture transcript path for stateless resume ---
+        async def stop_hook(
+            input_data: HookInput,
+            tool_use_id: str | None,
+            context: HookContext,
+        ) -> SyncHookJSONOutput:
+            """Capture transcript path when SDK finishes processing.
+
+            The Stop hook fires while the CLI process is still alive, giving us
+            a reliable window to read the JSONL transcript before SIGTERM.
+            """
+            _ = context, tool_use_id
+            transcript_path = cast(str, input_data.get("transcript_path", ""))
+            sdk_session_id = cast(str, input_data.get("session_id", ""))
+
+            if transcript_path and on_stop:
+                logger.info(
+                    f"[SDK] Stop hook: transcript_path={transcript_path}, "
+                    f"sdk_session_id={sdk_session_id[:12]}..."
+                )
+                on_stop(transcript_path, sdk_session_id)
+
+            return cast(SyncHookJSONOutput, {})
+
+        hooks: dict[str, Any] = {
+            "PreToolUse": [HookMatcher(matcher="*", hooks=[pre_tool_use_hook])],
+            "PostToolUse": [HookMatcher(matcher="*", hooks=[post_tool_use_hook])],
+            "PostToolUseFailure": [
+                HookMatcher(matcher="*", hooks=[post_tool_failure_hook])
+            ],
+            "PreCompact": [HookMatcher(matcher="*", hooks=[pre_compact_hook])],
+        }
+
+        if on_stop is not None:
+            hooks["Stop"] = [HookMatcher(matcher=None, hooks=[stop_hook])]
+
+        return hooks
+    except ImportError:
+        # Fallback for when SDK isn't available - return empty hooks
+        logger.warning("claude-agent-sdk not available, security hooks disabled")
+        return {}

autogpt_platform/backend/backend/copilot/sdk/security_hooks_test.py

@@ -0,0 +1,190 @@
+"""Tests for SDK security hooks — workspace paths, tool access, and deny messages.
+
+These are pure unit tests with no external dependencies (no SDK, no DB, no server).
+They validate that the security hooks correctly block unauthorized paths,
+tool access, and dangerous input patterns.
+"""
+
+import os
+
+from .security_hooks import _validate_tool_access, _validate_user_isolation
+
+SDK_CWD = "/tmp/copilot-abc123"
+
+
+def _is_denied(result: dict) -> bool:
+    hook = result.get("hookSpecificOutput", {})
+    return hook.get("permissionDecision") == "deny"
+
+
+def _reason(result: dict) -> str:
+    return result.get("hookSpecificOutput", {}).get("permissionDecisionReason", "")
+
+
+# -- Blocked tools -----------------------------------------------------------
+
+
+def test_blocked_tools_denied():
+    for tool in ("bash", "shell", "exec", "terminal", "command"):
+        result = _validate_tool_access(tool, {})
+        assert _is_denied(result), f"{tool} should be blocked"
+
+
+def test_unknown_tool_allowed():
+    result = _validate_tool_access("SomeCustomTool", {})
+    assert result == {}
+
+
+# -- Workspace-scoped tools --------------------------------------------------
+
+
+def test_read_within_workspace_allowed():
+    result = _validate_tool_access(
+        "Read", {"file_path": f"{SDK_CWD}/file.txt"}, sdk_cwd=SDK_CWD
+    )
+    assert result == {}
+
+
+def test_write_within_workspace_allowed():
+    result = _validate_tool_access(
+        "Write", {"file_path": f"{SDK_CWD}/output.json"}, sdk_cwd=SDK_CWD
+    )
+    assert result == {}
+
+
+def test_edit_within_workspace_allowed():
+    result = _validate_tool_access(
+        "Edit", {"file_path": f"{SDK_CWD}/src/main.py"}, sdk_cwd=SDK_CWD
+    )
+    assert result == {}
+
+
+def test_glob_within_workspace_allowed():
+    result = _validate_tool_access("Glob", {"path": f"{SDK_CWD}/src"}, sdk_cwd=SDK_CWD)
+    assert result == {}
+
+
+def test_grep_within_workspace_allowed():
+    result = _validate_tool_access("Grep", {"path": f"{SDK_CWD}/src"}, sdk_cwd=SDK_CWD)
+    assert result == {}
+
+
+def test_read_outside_workspace_denied():
+    result = _validate_tool_access(
+        "Read", {"file_path": "/etc/passwd"}, sdk_cwd=SDK_CWD
+    )
+    assert _is_denied(result)
+
+
+def test_write_outside_workspace_denied():
+    result = _validate_tool_access(
+        "Write", {"file_path": "/home/user/secrets.txt"}, sdk_cwd=SDK_CWD
+    )
+    assert _is_denied(result)
+
+
+def test_traversal_attack_denied():
+    result = _validate_tool_access(
+        "Read",
+        {"file_path": f"{SDK_CWD}/../../etc/passwd"},
+        sdk_cwd=SDK_CWD,
+    )
+    assert _is_denied(result)
+
+
+def test_no_path_allowed():
+    """Glob/Grep without a path argument defaults to cwd — should pass."""
+    result = _validate_tool_access("Glob", {}, sdk_cwd=SDK_CWD)
+    assert result == {}
+
+
+def test_read_no_cwd_denies_absolute():
+    """If no sdk_cwd is set, absolute paths are denied."""
+    result = _validate_tool_access("Read", {"file_path": "/tmp/anything"})
+    assert _is_denied(result)
+
+
+# -- Tool-results directory --------------------------------------------------
+
+
+def test_read_tool_results_allowed():
+    home = os.path.expanduser("~")
+    path = f"{home}/.claude/projects/-tmp-copilot-abc123/tool-results/12345.txt"
+    result = _validate_tool_access("Read", {"file_path": path}, sdk_cwd=SDK_CWD)
+    assert result == {}
+
+
+def test_read_claude_projects_without_tool_results_denied():
+    home = os.path.expanduser("~")
+    path = f"{home}/.claude/projects/-tmp-copilot-abc123/settings.json"
+    result = _validate_tool_access("Read", {"file_path": path}, sdk_cwd=SDK_CWD)
+    assert _is_denied(result)
+
+
+# -- Built-in Bash is blocked (use bash_exec MCP tool instead) ---------------
+
+
+def test_bash_builtin_always_blocked():
+    """SDK built-in Bash is blocked — bash_exec MCP tool with bubblewrap is used instead."""
+    result = _validate_tool_access("Bash", {"command": "echo hello"}, sdk_cwd=SDK_CWD)
+    assert _is_denied(result)
+
+
+# -- Dangerous patterns ------------------------------------------------------
+
+
+def test_dangerous_pattern_blocked():
+    result = _validate_tool_access("SomeTool", {"cmd": "sudo rm -rf /"})
+    assert _is_denied(result)
+
+
+def test_subprocess_pattern_blocked():
+    result = _validate_tool_access("SomeTool", {"code": "subprocess.run(...)"})
+    assert _is_denied(result)
+
+
+# -- User isolation ----------------------------------------------------------
+
+
+def test_workspace_path_traversal_blocked():
+    result = _validate_user_isolation(
+        "workspace_read", {"path": "../../../etc/shadow"}, user_id="user-1"
+    )
+    assert _is_denied(result)
+
+
+def test_workspace_absolute_path_blocked():
+    result = _validate_user_isolation(
+        "workspace_read", {"path": "/etc/passwd"}, user_id="user-1"
+    )
+    assert _is_denied(result)
+
+
+def test_workspace_normal_path_allowed():
+    result = _validate_user_isolation(
+        "workspace_read", {"path": "src/main.py"}, user_id="user-1"
+    )
+    assert result == {}
+
+
+def test_non_workspace_tool_passes_isolation():
+    result = _validate_user_isolation(
+        "find_agent", {"query": "email"}, user_id="user-1"
+    )
+    assert result == {}
+
+
+# -- Deny message quality ----------------------------------------------------
+
+
+def test_blocked_tool_message_clarity():
+    """Deny messages must include [SECURITY] and 'cannot be bypassed'."""
+    reason = _reason(_validate_tool_access("bash", {}))
+    assert "[SECURITY]" in reason
+    assert "cannot be bypassed" in reason
+
+
+def test_bash_builtin_blocked_message_clarity():
+    reason = _reason(_validate_tool_access("Bash", {"command": "echo hello"}))
+    assert "[SECURITY]" in reason
+    assert "cannot be bypassed" in reason

autogpt_platform/backend/backend/copilot/sdk/service.py

@@ -0,0 +1,751 @@
+"""Claude Agent SDK service layer for CoPilot chat completions."""
+
+import asyncio
+import json
+import logging
+import os
+import uuid
+from collections.abc import AsyncGenerator
+from dataclasses import dataclass
+from typing import Any
+
+from backend.util.exceptions import NotFoundError
+
+from .. import stream_registry
+from ..config import ChatConfig
+from ..model import (
+    ChatMessage,
+    ChatSession,
+    get_chat_session,
+    update_session_title,
+    upsert_chat_session,
+)
+from ..response_model import (
+    StreamBaseResponse,
+    StreamError,
+    StreamFinish,
+    StreamStart,
+    StreamTextDelta,
+    StreamToolInputAvailable,
+    StreamToolOutputAvailable,
+)
+from ..service import (
+    _build_system_prompt,
+    _execute_long_running_tool_with_streaming,
+    _generate_session_title,
+)
+from ..tools.models import OperationPendingResponse, OperationStartedResponse
+from ..tools.sandbox import WORKSPACE_PREFIX, make_session_path
+from ..tracking import track_user_message
+from .response_adapter import SDKResponseAdapter
+from .security_hooks import create_security_hooks
+from .tool_adapter import (
+    COPILOT_TOOL_NAMES,
+    LongRunningCallback,
+    create_copilot_mcp_server,
+    set_execution_context,
+)
+from .transcript import (
+    download_transcript,
+    read_transcript_file,
+    upload_transcript,
+    validate_transcript,
+    write_transcript_to_tempfile,
+)
+
+logger = logging.getLogger(__name__)
+config = ChatConfig()
+
+# Set to hold background tasks to prevent garbage collection
+_background_tasks: set[asyncio.Task[Any]] = set()
+
+
+@dataclass
+class CapturedTranscript:
+    """Info captured by the SDK Stop hook for stateless --resume."""
+
+    path: str = ""
+    sdk_session_id: str = ""
+
+    @property
+    def available(self) -> bool:
+        return bool(self.path)
+
+
+_SDK_CWD_PREFIX = WORKSPACE_PREFIX
+
+# Appended to the system prompt to inform the agent about available tools.
+# The SDK built-in Bash is NOT available — use mcp__copilot__bash_exec instead,
+# which has kernel-level network isolation (unshare --net).
+_SDK_TOOL_SUPPLEMENT = """
+
+## Tool notes
+
+- The SDK built-in Bash tool is NOT available.  Use the `bash_exec` MCP tool
+  for shell commands — it runs in a network-isolated sandbox.
+- **Shared workspace**: The SDK Read/Write tools and `bash_exec` share the
+  same working directory. Files created by one are readable by the other.
+  These files are **ephemeral** — they exist only for the current session.
+- **Persistent storage**: Use `write_workspace_file` / `read_workspace_file`
+  for files that should persist across sessions (stored in cloud storage).
+- Long-running tools (create_agent, edit_agent, etc.) are handled
+  asynchronously.  You will receive an immediate response; the actual result
+  is delivered to the user via a background stream.
+"""
+
+
+def _build_long_running_callback(user_id: str | None) -> LongRunningCallback:
+    """Build a callback that delegates long-running tools to the non-SDK infrastructure.
+
+    Long-running tools (create_agent, edit_agent, etc.) are delegated to the
+    existing background infrastructure: stream_registry (Redis Streams),
+    database persistence, and SSE reconnection.  This means results survive
+    page refreshes / pod restarts, and the frontend shows the proper loading
+    widget with progress updates.
+
+    The returned callback matches the ``LongRunningCallback`` signature:
+    ``(tool_name, args, session) -> MCP response dict``.
+    """
+
+    async def _callback(
+        tool_name: str, args: dict[str, Any], session: ChatSession
+    ) -> dict[str, Any]:
+        operation_id = str(uuid.uuid4())
+        task_id = str(uuid.uuid4())
+        tool_call_id = f"sdk-{uuid.uuid4().hex[:12]}"
+        session_id = session.session_id
+
+        # --- Build user-friendly messages (matches non-SDK service) ---
+        if tool_name == "create_agent":
+            desc = args.get("description", "")
+            desc_preview = (desc[:100] + "...") if len(desc) > 100 else desc
+            pending_msg = (
+                f"Creating your agent: {desc_preview}"
+                if desc_preview
+                else "Creating agent... This may take a few minutes."
+            )
+            started_msg = (
+                "Agent creation started. You can close this tab - "
+                "check your library in a few minutes."
+            )
+        elif tool_name == "edit_agent":
+            changes = args.get("changes", "")
+            changes_preview = (changes[:100] + "...") if len(changes) > 100 else changes
+            pending_msg = (
+                f"Editing agent: {changes_preview}"
+                if changes_preview
+                else "Editing agent... This may take a few minutes."
+            )
+            started_msg = (
+                "Agent edit started. You can close this tab - "
+                "check your library in a few minutes."
+            )
+        else:
+            pending_msg = f"Running {tool_name}... This may take a few minutes."
+            started_msg = (
+                f"{tool_name} started. You can close this tab - "
+                "check back in a few minutes."
+            )
+
+        # --- Register task in Redis for SSE reconnection ---
+        await stream_registry.create_task(
+            task_id=task_id,
+            session_id=session_id,
+            user_id=user_id,
+            tool_call_id=tool_call_id,
+            tool_name=tool_name,
+            operation_id=operation_id,
+        )
+
+        # --- Save OperationPendingResponse to chat history ---
+        pending_message = ChatMessage(
+            role="tool",
+            content=OperationPendingResponse(
+                message=pending_msg,
+                operation_id=operation_id,
+                tool_name=tool_name,
+            ).model_dump_json(),
+            tool_call_id=tool_call_id,
+        )
+        session.messages.append(pending_message)
+        await upsert_chat_session(session)
+
+        # --- Spawn background task (reuses non-SDK infrastructure) ---
+        bg_task = asyncio.create_task(
+            _execute_long_running_tool_with_streaming(
+                tool_name=tool_name,
+                parameters=args,
+                tool_call_id=tool_call_id,
+                operation_id=operation_id,
+                task_id=task_id,
+                session_id=session_id,
+                user_id=user_id,
+            )
+        )
+        _background_tasks.add(bg_task)
+        bg_task.add_done_callback(_background_tasks.discard)
+        await stream_registry.set_task_asyncio_task(task_id, bg_task)
+
+        logger.info(
+            f"[SDK] Long-running tool {tool_name} delegated to background "
+            f"(operation_id={operation_id}, task_id={task_id})"
+        )
+
+        # --- Return OperationStartedResponse as MCP tool result ---
+        # This flows through SDK → response adapter → frontend, triggering
+        # the loading widget with SSE reconnection support.
+        started_json = OperationStartedResponse(
+            message=started_msg,
+            operation_id=operation_id,
+            tool_name=tool_name,
+            task_id=task_id,
+        ).model_dump_json()
+
+        return {
+            "content": [{"type": "text", "text": started_json}],
+            "isError": False,
+        }
+
+    return _callback
+
+
+def _resolve_sdk_model() -> str | None:
+    """Resolve the model name for the Claude Agent SDK CLI.
+
+    Uses ``config.claude_agent_model`` if set, otherwise derives from
+    ``config.model`` by stripping the OpenRouter provider prefix (e.g.,
+    ``"anthropic/claude-opus-4.6"`` → ``"claude-opus-4.6"``).
+    """
+    if config.claude_agent_model:
+        return config.claude_agent_model
+    model = config.model
+    if "/" in model:
+        return model.split("/", 1)[1]
+    return model
+
+
+def _build_sdk_env() -> dict[str, str]:
+    """Build env vars for the SDK CLI process.
+
+    Routes API calls through OpenRouter (or a custom base_url) using
+    the same ``config.api_key`` / ``config.base_url`` as the non-SDK path.
+    This gives per-call token and cost tracking on the OpenRouter dashboard.
+
+    Only overrides ``ANTHROPIC_API_KEY`` when a valid proxy URL and auth
+    token are both present — otherwise returns an empty dict so the SDK
+    falls back to its default credentials.
+    """
+    env: dict[str, str] = {}
+    if config.api_key and config.base_url:
+        # Strip /v1 suffix — SDK expects the base URL without a version path
+        base = config.base_url.rstrip("/")
+        if base.endswith("/v1"):
+            base = base[:-3]
+        if not base or not base.startswith("http"):
+            # Invalid base_url — don't override SDK defaults
+            return env
+        env["ANTHROPIC_BASE_URL"] = base
+        env["ANTHROPIC_AUTH_TOKEN"] = config.api_key
+        # Must be explicitly empty so the CLI uses AUTH_TOKEN instead
+        env["ANTHROPIC_API_KEY"] = ""
+    return env
+
+
+def _make_sdk_cwd(session_id: str) -> str:
+    """Create a safe, session-specific working directory path.
+
+    Delegates to :func:`~backend.copilot.tools.sandbox.make_session_path`
+    (single source of truth for path sanitization) and adds a defence-in-depth
+    assertion.
+    """
+    cwd = make_session_path(session_id)
+    # Defence-in-depth: normpath + startswith is a CodeQL-recognised sanitizer
+    cwd = os.path.normpath(cwd)
+    if not cwd.startswith(_SDK_CWD_PREFIX):
+        raise ValueError(f"SDK cwd escaped prefix: {cwd}")
+    return cwd
+
+
+def _cleanup_sdk_tool_results(cwd: str) -> None:
+    """Remove SDK tool-result files for a specific session working directory.
+
+    The SDK creates tool-result files under ~/.claude/projects/<encoded-cwd>/tool-results/.
+    We clean only the specific cwd's results to avoid race conditions between
+    concurrent sessions.
+
+    Security: cwd MUST be created by _make_sdk_cwd() which sanitizes session_id.
+    """
+    import shutil
+
+    # Validate cwd is under the expected prefix
+    normalized = os.path.normpath(cwd)
+    if not normalized.startswith(_SDK_CWD_PREFIX):
+        logger.warning(f"[SDK] Rejecting cleanup for path outside workspace: {cwd}")
+        return
+
+    # SDK encodes the cwd path by replacing '/' with '-'
+    encoded_cwd = normalized.replace("/", "-")
+
+    # Construct the project directory path (known-safe home expansion)
+    claude_projects = os.path.expanduser("~/.claude/projects")
+    project_dir = os.path.join(claude_projects, encoded_cwd)
+
+    # Security check 3: Validate project_dir is under ~/.claude/projects
+    project_dir = os.path.normpath(project_dir)
+    if not project_dir.startswith(claude_projects):
+        logger.warning(
+            f"[SDK] Rejecting cleanup for escaped project path: {project_dir}"
+        )
+        return
+
+    results_dir = os.path.join(project_dir, "tool-results")
+    if os.path.isdir(results_dir):
+        for filename in os.listdir(results_dir):
+            file_path = os.path.join(results_dir, filename)
+            try:
+                if os.path.isfile(file_path):
+                    os.remove(file_path)
+            except OSError:
+                pass
+
+    # Also clean up the temp cwd directory itself
+    try:
+        shutil.rmtree(normalized, ignore_errors=True)
+    except OSError:
+        pass
+
+
+async def _compress_conversation_history(
+    session: ChatSession,
+) -> list[ChatMessage]:
+    """Compress prior conversation messages if they exceed the token threshold.
+
+    Uses the shared compress_context() from prompt.py which supports:
+    - LLM summarization of old messages (keeps recent ones intact)
+    - Progressive content truncation as fallback
+    - Middle-out deletion as last resort
+
+    Returns the compressed prior messages (everything except the current message).
+    """
+    prior = session.messages[:-1]
+    if len(prior) < 2:
+        return prior
+
+    from backend.util.prompt import compress_context
+
+    # Convert ChatMessages to dicts for compress_context
+    messages_dict = []
+    for msg in prior:
+        msg_dict: dict[str, Any] = {"role": msg.role}
+        if msg.content:
+            msg_dict["content"] = msg.content
+        if msg.tool_calls:
+            msg_dict["tool_calls"] = msg.tool_calls
+        if msg.tool_call_id:
+            msg_dict["tool_call_id"] = msg.tool_call_id
+        messages_dict.append(msg_dict)
+
+    try:
+        import openai
+
+        async with openai.AsyncOpenAI(
+            api_key=config.api_key, base_url=config.base_url, timeout=30.0
+        ) as client:
+            result = await compress_context(
+                messages=messages_dict,
+                model=config.model,
+                client=client,
+            )
+    except Exception as e:
+        logger.warning(f"[SDK] Context compression with LLM failed: {e}")
+        # Fall back to truncation-only (no LLM summarization)
+        result = await compress_context(
+            messages=messages_dict,
+            model=config.model,
+            client=None,
+        )
+
+    if result.was_compacted:
+        logger.info(
+            f"[SDK] Context compacted: {result.original_token_count} -> "
+            f"{result.token_count} tokens "
+            f"({result.messages_summarized} summarized, "
+            f"{result.messages_dropped} dropped)"
+        )
+        # Convert compressed dicts back to ChatMessages
+        return [
+            ChatMessage(
+                role=m["role"],
+                content=m.get("content"),
+                tool_calls=m.get("tool_calls"),
+                tool_call_id=m.get("tool_call_id"),
+            )
+            for m in result.messages
+        ]
+
+    return prior
+
+
+def _format_conversation_context(messages: list[ChatMessage]) -> str | None:
+    """Format conversation messages into a context prefix for the user message.
+
+    Returns a string like:
+        <conversation_history>
+        User: hello
+        You responded: Hi! How can I help?
+        </conversation_history>
+
+    Returns None if there are no messages to format.
+    """
+    if not messages:
+        return None
+
+    lines: list[str] = []
+    for msg in messages:
+        if not msg.content:
+            continue
+        if msg.role == "user":
+            lines.append(f"User: {msg.content}")
+        elif msg.role == "assistant":
+            lines.append(f"You responded: {msg.content}")
+        # Skip tool messages — they're internal details
+
+    if not lines:
+        return None
+
+    return "<conversation_history>\n" + "\n".join(lines) + "\n</conversation_history>"
+
+
+async def stream_chat_completion_sdk(
+    session_id: str,
+    message: str | None = None,
+    tool_call_response: str | None = None,  # noqa: ARG001
+    is_user_message: bool = True,
+    user_id: str | None = None,
+    retry_count: int = 0,  # noqa: ARG001
+    session: ChatSession | None = None,
+    context: dict[str, str] | None = None,  # noqa: ARG001
+) -> AsyncGenerator[StreamBaseResponse, None]:
+    """Stream chat completion using Claude Agent SDK.
+
+    Drop-in replacement for stream_chat_completion with improved reliability.
+    """
+
+    if session is None:
+        session = await get_chat_session(session_id, user_id)
+
+    if not session:
+        raise NotFoundError(
+            f"Session {session_id} not found. Please create a new session first."
+        )
+
+    if message:
+        session.messages.append(
+            ChatMessage(
+                role="user" if is_user_message else "assistant", content=message
+            )
+        )
+        if is_user_message:
+            track_user_message(
+                user_id=user_id, session_id=session_id, message_length=len(message)
+            )
+
+    session = await upsert_chat_session(session)
+
+    # Generate title for new sessions (first user message)
+    if is_user_message and not session.title:
+        user_messages = [m for m in session.messages if m.role == "user"]
+        if len(user_messages) == 1:
+            first_message = user_messages[0].content or message or ""
+            if first_message:
+                task = asyncio.create_task(
+                    _update_title_async(session_id, first_message, user_id)
+                )
+                _background_tasks.add(task)
+                task.add_done_callback(_background_tasks.discard)
+
+    # Build system prompt (reuses non-SDK path with Langfuse support)
+    has_history = len(session.messages) > 1
+    system_prompt, _ = await _build_system_prompt(
+        user_id, has_conversation_history=has_history
+    )
+    system_prompt += _SDK_TOOL_SUPPLEMENT
+    message_id = str(uuid.uuid4())
+    task_id = str(uuid.uuid4())
+
+    yield StreamStart(messageId=message_id, taskId=task_id)
+
+    stream_completed = False
+    # Initialise sdk_cwd before the try so the finally can reference it
+    # even if _make_sdk_cwd raises (in that case it stays as "").
+    sdk_cwd = ""
+    use_resume = False
+
+    try:
+        # Use a session-specific temp dir to avoid cleanup race conditions
+        # between concurrent sessions.
+        sdk_cwd = _make_sdk_cwd(session_id)
+        os.makedirs(sdk_cwd, exist_ok=True)
+
+        set_execution_context(
+            user_id,
+            session,
+            long_running_callback=_build_long_running_callback(user_id),
+        )
+        try:
+            from claude_agent_sdk import ClaudeAgentOptions, ClaudeSDKClient
+
+            # Fail fast when no API credentials are available at all
+            sdk_env = _build_sdk_env()
+            if not sdk_env and not os.environ.get("ANTHROPIC_API_KEY"):
+                raise RuntimeError(
+                    "No API key configured. Set OPEN_ROUTER_API_KEY "
+                    "(or CHAT_API_KEY) for OpenRouter routing, "
+                    "or ANTHROPIC_API_KEY for direct Anthropic access."
+                )
+
+            mcp_server = create_copilot_mcp_server()
+
+            sdk_model = _resolve_sdk_model()
+
+            # --- Transcript capture via Stop hook ---
+            captured_transcript = CapturedTranscript()
+
+            def _on_stop(transcript_path: str, sdk_session_id: str) -> None:
+                captured_transcript.path = transcript_path
+                captured_transcript.sdk_session_id = sdk_session_id
+
+            security_hooks = create_security_hooks(
+                user_id,
+                sdk_cwd=sdk_cwd,
+                max_subtasks=config.claude_agent_max_subtasks,
+                on_stop=_on_stop if config.claude_agent_use_resume else None,
+            )
+
+            # --- Resume strategy: download transcript from bucket ---
+            resume_file: str | None = None
+            use_resume = False
+
+            if config.claude_agent_use_resume and user_id and len(session.messages) > 1:
+                transcript_content = await download_transcript(user_id, session_id)
+                if transcript_content and validate_transcript(transcript_content):
+                    resume_file = write_transcript_to_tempfile(
+                        transcript_content, session_id, sdk_cwd
+                    )
+                    if resume_file:
+                        use_resume = True
+                        logger.info(
+                            f"[SDK] Using --resume with transcript "
+                            f"({len(transcript_content)} bytes)"
+                        )
+
+            sdk_options_kwargs: dict[str, Any] = {
+                "system_prompt": system_prompt,
+                "mcp_servers": {"copilot": mcp_server},
+                "allowed_tools": COPILOT_TOOL_NAMES,
+                "disallowed_tools": ["Bash"],
+                "hooks": security_hooks,
+                "cwd": sdk_cwd,
+                "max_buffer_size": config.claude_agent_max_buffer_size,
+            }
+            if sdk_env:
+                sdk_options_kwargs["model"] = sdk_model
+                sdk_options_kwargs["env"] = sdk_env
+            if use_resume and resume_file:
+                sdk_options_kwargs["resume"] = resume_file
+
+            options = ClaudeAgentOptions(**sdk_options_kwargs)  # type: ignore[arg-type]
+
+            adapter = SDKResponseAdapter(message_id=message_id)
+            adapter.set_task_id(task_id)
+
+            async with ClaudeSDKClient(options=options) as client:
+                current_message = message or ""
+                if not current_message and session.messages:
+                    last_user = [m for m in session.messages if m.role == "user"]
+                    if last_user:
+                        current_message = last_user[-1].content or ""
+
+                if not current_message.strip():
+                    yield StreamError(
+                        errorText="Message cannot be empty.",
+                        code="empty_prompt",
+                    )
+                    yield StreamFinish()
+                    return
+
+                # Build query: with --resume the CLI already has full
+                # context, so we only send the new message.  Without
+                # resume, compress history into a context prefix.
+                query_message = current_message
+                if not use_resume and len(session.messages) > 1:
+                    logger.warning(
+                        f"[SDK] Using compression fallback for session "
+                        f"{session_id} ({len(session.messages)} messages) — "
+                        f"no transcript available for --resume"
+                    )
+                    compressed = await _compress_conversation_history(session)
+                    history_context = _format_conversation_context(compressed)
+                    if history_context:
+                        query_message = (
+                            f"{history_context}\n\n"
+                            f"Now, the user says:\n{current_message}"
+                        )
+
+                logger.info(
+                    f"[SDK] Sending query ({len(session.messages)} msgs in session)"
+                )
+                logger.debug(f"[SDK] Query preview: {current_message[:80]!r}")
+                await client.query(query_message, session_id=session_id)
+
+                assistant_response = ChatMessage(role="assistant", content="")
+                accumulated_tool_calls: list[dict[str, Any]] = []
+                has_appended_assistant = False
+                has_tool_results = False
+
+                async for sdk_msg in client.receive_messages():
+                    logger.debug(
+                        f"[SDK] Received: {type(sdk_msg).__name__} "
+                        f"{getattr(sdk_msg, 'subtype', '')}"
+                    )
+                    for response in adapter.convert_message(sdk_msg):
+                        if isinstance(response, StreamStart):
+                            continue
+
+                        yield response
+
+                        if isinstance(response, StreamTextDelta):
+                            delta = response.delta or ""
+                            # After tool results, start a new assistant
+                            # message for the post-tool text.
+                            if has_tool_results and has_appended_assistant:
+                                assistant_response = ChatMessage(
+                                    role="assistant", content=delta
+                                )
+                                accumulated_tool_calls = []
+                                has_appended_assistant = False
+                                has_tool_results = False
+                                session.messages.append(assistant_response)
+                                has_appended_assistant = True
+                            else:
+                                assistant_response.content = (
+                                    assistant_response.content or ""
+                                ) + delta
+                                if not has_appended_assistant:
+                                    session.messages.append(assistant_response)
+                                    has_appended_assistant = True
+
+                        elif isinstance(response, StreamToolInputAvailable):
+                            accumulated_tool_calls.append(
+                                {
+                                    "id": response.toolCallId,
+                                    "type": "function",
+                                    "function": {
+                                        "name": response.toolName,
+                                        "arguments": json.dumps(response.input or {}),
+                                    },
+                                }
+                            )
+                            assistant_response.tool_calls = accumulated_tool_calls
+                            if not has_appended_assistant:
+                                session.messages.append(assistant_response)
+                                has_appended_assistant = True
+
+                        elif isinstance(response, StreamToolOutputAvailable):
+                            session.messages.append(
+                                ChatMessage(
+                                    role="tool",
+                                    content=(
+                                        response.output
+                                        if isinstance(response.output, str)
+                                        else str(response.output)
+                                    ),
+                                    tool_call_id=response.toolCallId,
+                                )
+                            )
+                            has_tool_results = True
+
+                        elif isinstance(response, StreamFinish):
+                            stream_completed = True
+
+                    if stream_completed:
+                        break
+
+                if (
+                    assistant_response.content or assistant_response.tool_calls
+                ) and not has_appended_assistant:
+                    session.messages.append(assistant_response)
+
+                # --- Capture transcript while CLI is still alive ---
+                # Must happen INSIDE async with: close() sends SIGTERM
+                # which kills the CLI before it can flush the JSONL.
+                if (
+                    config.claude_agent_use_resume
+                    and user_id
+                    and captured_transcript.available
+                ):
+                    # Give CLI time to flush JSONL writes before we read
+                    await asyncio.sleep(0.5)
+                    raw_transcript = read_transcript_file(captured_transcript.path)
+                    if raw_transcript:
+                        task = asyncio.create_task(
+                            _upload_transcript_bg(user_id, session_id, raw_transcript)
+                        )
+                        _background_tasks.add(task)
+                        task.add_done_callback(_background_tasks.discard)
+                    else:
+                        logger.debug("[SDK] Stop hook fired but transcript not usable")
+
+        except ImportError:
+            raise RuntimeError(
+                "claude-agent-sdk is not installed. "
+                "Disable SDK mode (CHAT_USE_CLAUDE_AGENT_SDK=false) "
+                "to use the OpenAI-compatible fallback."
+            )
+
+        await upsert_chat_session(session)
+        logger.debug(
+            f"[SDK] Session {session_id} saved with {len(session.messages)} messages"
+        )
+        if not stream_completed:
+            yield StreamFinish()
+
+    except Exception as e:
+        logger.error(f"[SDK] Error: {e}", exc_info=True)
+        try:
+            await upsert_chat_session(session)
+        except Exception as save_err:
+            logger.error(f"[SDK] Failed to save session on error: {save_err}")
+        yield StreamError(
+            errorText="An error occurred. Please try again.",
+            code="sdk_error",
+        )
+        yield StreamFinish()
+    finally:
+        if sdk_cwd:
+            _cleanup_sdk_tool_results(sdk_cwd)
+
+
+async def _upload_transcript_bg(
+    user_id: str, session_id: str, raw_content: str
+) -> None:
+    """Background task to strip progress entries and upload transcript."""
+    try:
+        await upload_transcript(user_id, session_id, raw_content)
+    except Exception as e:
+        logger.error(f"[SDK] Failed to upload transcript for {session_id}: {e}")
+
+
+async def _update_title_async(
+    session_id: str, message: str, user_id: str | None = None
+) -> None:
+    """Background task to update session title."""
+    try:
+        title = await _generate_session_title(
+            message, user_id=user_id, session_id=session_id
+        )
+        if title:
+            await update_session_title(session_id, title)
+            logger.debug(f"[SDK] Generated title for {session_id}: {title}")
+    except Exception as e:
+        logger.warning(f"[SDK] Failed to update session title: {e}")

autogpt_platform/backend/backend/copilot/sdk/tool_adapter.py

@@ -0,0 +1,322 @@
+"""Tool adapter for wrapping existing CoPilot tools as Claude Agent SDK MCP tools.
+
+This module provides the adapter layer that converts existing BaseTool implementations
+into in-process MCP tools that can be used with the Claude Agent SDK.
+
+Long-running tools (``is_long_running=True``) are delegated to the non-SDK
+background infrastructure (stream_registry, Redis persistence, SSE reconnection)
+via a callback provided by the service layer.  This avoids wasteful SDK polling
+and makes results survive page refreshes.
+"""
+
+import itertools
+import json
+import logging
+import os
+import uuid
+from collections.abc import Awaitable, Callable
+from contextvars import ContextVar
+from typing import Any
+
+from backend.copilot.model import ChatSession
+from backend.copilot.tools import TOOL_REGISTRY
+from backend.copilot.tools.base import BaseTool
+
+logger = logging.getLogger(__name__)
+
+# Allowed base directory for the Read tool (SDK saves oversized tool results here).
+# Restricted to ~/.claude/projects/ and further validated to require "tool-results"
+# in the path — prevents reading settings, credentials, or other sensitive files.
+_SDK_PROJECTS_DIR = os.path.expanduser("~/.claude/projects/")
+
+# MCP server naming - the SDK prefixes tool names as "mcp__{server_name}__{tool}"
+MCP_SERVER_NAME = "copilot"
+MCP_TOOL_PREFIX = f"mcp__{MCP_SERVER_NAME}__"
+
+# Context variables to pass user/session info to tool execution
+_current_user_id: ContextVar[str | None] = ContextVar("current_user_id", default=None)
+_current_session: ContextVar[ChatSession | None] = ContextVar(
+    "current_session", default=None
+)
+# Stash for MCP tool outputs before the SDK potentially truncates them.
+# Keyed by tool_name → full output string. Consumed (popped) by the
+# response adapter when it builds StreamToolOutputAvailable.
+_pending_tool_outputs: ContextVar[dict[str, str]] = ContextVar(
+    "pending_tool_outputs", default=None  # type: ignore[arg-type]
+)
+
+# Callback type for delegating long-running tools to the non-SDK infrastructure.
+# Args: (tool_name, arguments, session) → MCP-formatted response dict.
+LongRunningCallback = Callable[
+    [str, dict[str, Any], ChatSession], Awaitable[dict[str, Any]]
+]
+
+# ContextVar so the service layer can inject the callback per-request.
+_long_running_callback: ContextVar[LongRunningCallback | None] = ContextVar(
+    "long_running_callback", default=None
+)
+
+
+def set_execution_context(
+    user_id: str | None,
+    session: ChatSession,
+    long_running_callback: LongRunningCallback | None = None,
+) -> None:
+    """Set the execution context for tool calls.
+
+    This must be called before streaming begins to ensure tools have access
+    to user_id and session information.
+
+    Args:
+        user_id: Current user's ID.
+        session: Current chat session.
+        long_running_callback: Optional callback to delegate long-running tools
+            to the non-SDK background infrastructure (stream_registry + Redis).
+    """
+    _current_user_id.set(user_id)
+    _current_session.set(session)
+    _pending_tool_outputs.set({})
+    _long_running_callback.set(long_running_callback)
+
+
+def get_execution_context() -> tuple[str | None, ChatSession | None]:
+    """Get the current execution context."""
+    return (
+        _current_user_id.get(),
+        _current_session.get(),
+    )
+
+
+def pop_pending_tool_output(tool_name: str) -> str | None:
+    """Pop and return the stashed full output for *tool_name*.
+
+    The SDK CLI may truncate large tool results (writing them to disk and
+    replacing the content with a file reference). This stash keeps the
+    original MCP output so the response adapter can forward it to the
+    frontend for proper widget rendering.
+
+    Returns ``None`` if nothing was stashed for *tool_name*.
+    """
+    pending = _pending_tool_outputs.get(None)
+    if pending is None:
+        return None
+    return pending.pop(tool_name, None)
+
+
+async def _execute_tool_sync(
+    base_tool: BaseTool,
+    user_id: str | None,
+    session: ChatSession,
+    args: dict[str, Any],
+) -> dict[str, Any]:
+    """Execute a tool synchronously and return MCP-formatted response."""
+    effective_id = f"sdk-{uuid.uuid4().hex[:12]}"
+    result = await base_tool.execute(
+        user_id=user_id,
+        session=session,
+        tool_call_id=effective_id,
+        **args,
+    )
+
+    text = (
+        result.output if isinstance(result.output, str) else json.dumps(result.output)
+    )
+
+    # Stash the full output before the SDK potentially truncates it.
+    pending = _pending_tool_outputs.get(None)
+    if pending is not None:
+        pending[base_tool.name] = text
+
+    return {
+        "content": [{"type": "text", "text": text}],
+        "isError": not result.success,
+    }
+
+
+def _mcp_error(message: str) -> dict[str, Any]:
+    return {
+        "content": [
+            {"type": "text", "text": json.dumps({"error": message, "type": "error"})}
+        ],
+        "isError": True,
+    }
+
+
+def create_tool_handler(base_tool: BaseTool):
+    """Create an async handler function for a BaseTool.
+
+    This wraps the existing BaseTool._execute method to be compatible
+    with the Claude Agent SDK MCP tool format.
+
+    Long-running tools (``is_long_running=True``) are delegated to the
+    non-SDK background infrastructure via a callback set in the execution
+    context.  The callback persists the operation in Redis (stream_registry)
+    so results survive page refreshes and pod restarts.
+    """
+
+    async def tool_handler(args: dict[str, Any]) -> dict[str, Any]:
+        """Execute the wrapped tool and return MCP-formatted response."""
+        user_id, session = get_execution_context()
+
+        if session is None:
+            return _mcp_error("No session context available")
+
+        # --- Long-running: delegate to non-SDK background infrastructure ---
+        if base_tool.is_long_running:
+            callback = _long_running_callback.get(None)
+            if callback:
+                try:
+                    return await callback(base_tool.name, args, session)
+                except Exception as e:
+                    logger.error(
+                        f"Long-running callback failed for {base_tool.name}: {e}",
+                        exc_info=True,
+                    )
+                    return _mcp_error(f"Failed to start {base_tool.name}: {e}")
+            # No callback — fall through to synchronous execution
+            logger.warning(
+                f"[SDK] No long-running callback for {base_tool.name}, "
+                f"executing synchronously (may block)"
+            )
+
+        # --- Normal (fast) tool: execute synchronously ---
+        try:
+            return await _execute_tool_sync(base_tool, user_id, session, args)
+        except Exception as e:
+            logger.error(f"Error executing tool {base_tool.name}: {e}", exc_info=True)
+            return _mcp_error(f"Failed to execute {base_tool.name}: {e}")
+
+    return tool_handler
+
+
+def _build_input_schema(base_tool: BaseTool) -> dict[str, Any]:
+    """Build a JSON Schema input schema for a tool."""
+    return {
+        "type": "object",
+        "properties": base_tool.parameters.get("properties", {}),
+        "required": base_tool.parameters.get("required", []),
+    }
+
+
+async def _read_file_handler(args: dict[str, Any]) -> dict[str, Any]:
+    """Read a file with optional offset/limit. Restricted to SDK working directory.
+
+    After reading, the file is deleted to prevent accumulation in long-running pods.
+    """
+    file_path = args.get("file_path", "")
+    offset = args.get("offset", 0)
+    limit = args.get("limit", 2000)
+
+    # Security: only allow reads under ~/.claude/projects/**/tool-results/
+    real_path = os.path.realpath(file_path)
+    if not real_path.startswith(_SDK_PROJECTS_DIR) or "tool-results" not in real_path:
+        return {
+            "content": [{"type": "text", "text": f"Access denied: {file_path}"}],
+            "isError": True,
+        }
+
+    try:
+        with open(real_path) as f:
+            selected = list(itertools.islice(f, offset, offset + limit))
+        content = "".join(selected)
+        # Cleanup happens in _cleanup_sdk_tool_results after session ends;
+        # don't delete here — the SDK may read in multiple chunks.
+        return {"content": [{"type": "text", "text": content}], "isError": False}
+    except FileNotFoundError:
+        return {
+            "content": [{"type": "text", "text": f"File not found: {file_path}"}],
+            "isError": True,
+        }
+    except Exception as e:
+        return {
+            "content": [{"type": "text", "text": f"Error reading file: {e}"}],
+            "isError": True,
+        }
+
+
+_READ_TOOL_NAME = "Read"
+_READ_TOOL_DESCRIPTION = (
+    "Read a file from the local filesystem. "
+    "Use offset and limit to read specific line ranges for large files."
+)
+_READ_TOOL_SCHEMA = {
+    "type": "object",
+    "properties": {
+        "file_path": {
+            "type": "string",
+            "description": "The absolute path to the file to read",
+        },
+        "offset": {
+            "type": "integer",
+            "description": "Line number to start reading from (0-indexed). Default: 0",
+        },
+        "limit": {
+            "type": "integer",
+            "description": "Number of lines to read. Default: 2000",
+        },
+    },
+    "required": ["file_path"],
+}
+
+
+# Create the MCP server configuration
+def create_copilot_mcp_server():
+    """Create an in-process MCP server configuration for CoPilot tools.
+
+    This can be passed to ClaudeAgentOptions.mcp_servers.
+
+    Note: The actual SDK MCP server creation depends on the claude-agent-sdk
+    package being available. This function returns the configuration that
+    can be used with the SDK.
+    """
+    try:
+        from claude_agent_sdk import create_sdk_mcp_server, tool
+
+        # Create decorated tool functions
+        sdk_tools = []
+
+        for tool_name, base_tool in TOOL_REGISTRY.items():
+            handler = create_tool_handler(base_tool)
+            decorated = tool(
+                tool_name,
+                base_tool.description,
+                _build_input_schema(base_tool),
+            )(handler)
+            sdk_tools.append(decorated)
+
+        # Add the Read tool so the SDK can read back oversized tool results
+        read_tool = tool(
+            _READ_TOOL_NAME,
+            _READ_TOOL_DESCRIPTION,
+            _READ_TOOL_SCHEMA,
+        )(_read_file_handler)
+        sdk_tools.append(read_tool)
+
+        server = create_sdk_mcp_server(
+            name=MCP_SERVER_NAME,
+            version="1.0.0",
+            tools=sdk_tools,
+        )
+
+        return server
+
+    except ImportError:
+        # Let ImportError propagate so service.py handles the fallback
+        raise
+
+
+# SDK built-in tools allowed within the workspace directory.
+# Security hooks validate that file paths stay within sdk_cwd.
+# Bash is NOT included — use the sandboxed MCP bash_exec tool instead,
+# which provides kernel-level network isolation via unshare --net.
+# Task allows spawning sub-agents (rate-limited by security hooks).
+_SDK_BUILTIN_TOOLS = ["Read", "Write", "Edit", "Glob", "Grep", "Task"]
+
+# List of tool names for allowed_tools configuration
+# Include MCP tools, the MCP Read tool for oversized results,
+# and SDK built-in file tools for workspace operations.
+COPILOT_TOOL_NAMES = [
+    *[f"{MCP_TOOL_PREFIX}{name}" for name in TOOL_REGISTRY.keys()],
+    f"{MCP_TOOL_PREFIX}{_READ_TOOL_NAME}",
+    *_SDK_BUILTIN_TOOLS,
+]

autogpt_platform/backend/backend/copilot/sdk/transcript.py

@@ -0,0 +1,356 @@
+"""JSONL transcript management for stateless multi-turn resume.
+
+The Claude Code CLI persists conversations as JSONL files (one JSON object per
+line).  When the SDK's ``Stop`` hook fires we read this file, strip bloat
+(progress entries, metadata), and upload the result to bucket storage.  On the
+next turn we download the transcript, write it to a temp file, and pass
+``--resume`` so the CLI can reconstruct the full conversation.
+
+Storage is handled via ``WorkspaceStorageBackend`` (GCS in prod, local
+filesystem for self-hosted) — no DB column needed.
+"""
+
+import json
+import logging
+import os
+import re
+
+logger = logging.getLogger(__name__)
+
+# UUIDs are hex + hyphens; strip everything else to prevent path injection.
+_SAFE_ID_RE = re.compile(r"[^0-9a-fA-F-]")
+
+# Entry types that can be safely removed from the transcript without breaking
+# the parentUuid conversation tree that ``--resume`` relies on.
+# - progress: UI progress ticks, no message content (avg 97KB for agent_progress)
+# - file-history-snapshot: undo tracking metadata
+# - queue-operation: internal queue bookkeeping
+# - summary: session summaries
+# - pr-link: PR link metadata
+STRIPPABLE_TYPES = frozenset(
+    {"progress", "file-history-snapshot", "queue-operation", "summary", "pr-link"}
+)
+
+# Workspace storage constants — deterministic path from session_id.
+TRANSCRIPT_STORAGE_PREFIX = "chat-transcripts"
+
+
+# ---------------------------------------------------------------------------
+# Progress stripping
+# ---------------------------------------------------------------------------
+
+
+def strip_progress_entries(content: str) -> str:
+    """Remove progress/metadata entries from a JSONL transcript.
+
+    Removes entries whose ``type`` is in ``STRIPPABLE_TYPES`` and reparents
+    any remaining child entries so the ``parentUuid`` chain stays intact.
+    Typically reduces transcript size by ~30%.
+    """
+    lines = content.strip().split("\n")
+
+    entries: list[dict] = []
+    for line in lines:
+        try:
+            entries.append(json.loads(line))
+        except json.JSONDecodeError:
+            # Keep unparseable lines as-is (safety)
+            entries.append({"_raw": line})
+
+    stripped_uuids: set[str] = set()
+    uuid_to_parent: dict[str, str] = {}
+    kept: list[dict] = []
+
+    for entry in entries:
+        if "_raw" in entry:
+            kept.append(entry)
+            continue
+        uid = entry.get("uuid", "")
+        parent = entry.get("parentUuid", "")
+        entry_type = entry.get("type", "")
+
+        if uid:
+            uuid_to_parent[uid] = parent
+
+        if entry_type in STRIPPABLE_TYPES:
+            if uid:
+                stripped_uuids.add(uid)
+        else:
+            kept.append(entry)
+
+    # Reparent: walk up chain through stripped entries to find surviving ancestor
+    for entry in kept:
+        if "_raw" in entry:
+            continue
+        parent = entry.get("parentUuid", "")
+        original_parent = parent
+        while parent in stripped_uuids:
+            parent = uuid_to_parent.get(parent, "")
+        if parent != original_parent:
+            entry["parentUuid"] = parent
+
+    result_lines: list[str] = []
+    for entry in kept:
+        if "_raw" in entry:
+            result_lines.append(entry["_raw"])
+        else:
+            result_lines.append(json.dumps(entry, separators=(",", ":")))
+
+    return "\n".join(result_lines) + "\n"
+
+
+# ---------------------------------------------------------------------------
+# Local file I/O (read from CLI's JSONL, write temp file for --resume)
+# ---------------------------------------------------------------------------
+
+
+def read_transcript_file(transcript_path: str) -> str | None:
+    """Read a JSONL transcript file from disk.
+
+    Returns the raw JSONL content, or ``None`` if the file is missing, empty,
+    or only contains metadata (≤2 lines with no conversation messages).
+    """
+    if not transcript_path or not os.path.isfile(transcript_path):
+        logger.debug(f"[Transcript] File not found: {transcript_path}")
+        return None
+
+    try:
+        with open(transcript_path) as f:
+            content = f.read()
+
+        if not content.strip():
+            logger.debug(f"[Transcript] Empty file: {transcript_path}")
+            return None
+
+        lines = content.strip().split("\n")
+        if len(lines) < 3:
+            # Raw files with ≤2 lines are metadata-only
+            # (queue-operation + file-history-snapshot, no conversation).
+            logger.debug(
+                f"[Transcript] Too few lines ({len(lines)}): {transcript_path}"
+            )
+            return None
+
+        # Quick structural validation — parse first and last lines.
+        json.loads(lines[0])
+        json.loads(lines[-1])
+
+        logger.info(
+            f"[Transcript] Read {len(lines)} lines, "
+            f"{len(content)} bytes from {transcript_path}"
+        )
+        return content
+
+    except (json.JSONDecodeError, OSError) as e:
+        logger.warning(f"[Transcript] Failed to read {transcript_path}: {e}")
+        return None
+
+
+def _sanitize_id(raw_id: str, max_len: int = 36) -> str:
+    """Sanitize an ID for safe use in file paths.
+
+    Session/user IDs are expected to be UUIDs (hex + hyphens).  Strip
+    everything else and truncate to *max_len* so the result cannot introduce
+    path separators or other special characters.
+    """
+    cleaned = _SAFE_ID_RE.sub("", raw_id or "")[:max_len]
+    return cleaned or "unknown"
+
+
+_SAFE_CWD_PREFIX = os.path.realpath("/tmp/copilot-")
+
+
+def write_transcript_to_tempfile(
+    transcript_content: str,
+    session_id: str,
+    cwd: str,
+) -> str | None:
+    """Write JSONL transcript to a temp file inside *cwd* for ``--resume``.
+
+    The file lives in the session working directory so it is cleaned up
+    automatically when the session ends.
+
+    Returns the absolute path to the file, or ``None`` on failure.
+    """
+    # Validate cwd is under the expected sandbox prefix (CodeQL sanitizer).
+    real_cwd = os.path.realpath(cwd)
+    if not real_cwd.startswith(_SAFE_CWD_PREFIX):
+        logger.warning(f"[Transcript] cwd outside sandbox: {cwd}")
+        return None
+
+    try:
+        os.makedirs(real_cwd, exist_ok=True)
+        safe_id = _sanitize_id(session_id, max_len=8)
+        jsonl_path = os.path.realpath(
+            os.path.join(real_cwd, f"transcript-{safe_id}.jsonl")
+        )
+        if not jsonl_path.startswith(real_cwd):
+            logger.warning(f"[Transcript] Path escaped cwd: {jsonl_path}")
+            return None
+
+        with open(jsonl_path, "w") as f:
+            f.write(transcript_content)
+
+        logger.info(f"[Transcript] Wrote resume file: {jsonl_path}")
+        return jsonl_path
+
+    except OSError as e:
+        logger.warning(f"[Transcript] Failed to write resume file: {e}")
+        return None
+
+
+def validate_transcript(content: str | None) -> bool:
+    """Check that a transcript has actual conversation messages.
+
+    A valid transcript for resume needs at least one user message and one
+    assistant message (not just queue-operation / file-history-snapshot
+    metadata).
+    """
+    if not content or not content.strip():
+        return False
+
+    lines = content.strip().split("\n")
+    if len(lines) < 2:
+        return False
+
+    has_user = False
+    has_assistant = False
+
+    for line in lines:
+        try:
+            entry = json.loads(line)
+            msg_type = entry.get("type")
+            if msg_type == "user":
+                has_user = True
+            elif msg_type == "assistant":
+                has_assistant = True
+        except json.JSONDecodeError:
+            return False
+
+    return has_user and has_assistant
+
+
+# ---------------------------------------------------------------------------
+# Bucket storage (GCS / local via WorkspaceStorageBackend)
+# ---------------------------------------------------------------------------
+
+
+def _storage_path_parts(user_id: str, session_id: str) -> tuple[str, str, str]:
+    """Return (workspace_id, file_id, filename) for a session's transcript.
+
+    Path structure: ``chat-transcripts/{user_id}/{session_id}.jsonl``
+    IDs are sanitized to hex+hyphen to prevent path traversal.
+    """
+    return (
+        TRANSCRIPT_STORAGE_PREFIX,
+        _sanitize_id(user_id),
+        f"{_sanitize_id(session_id)}.jsonl",
+    )
+
+
+def _build_storage_path(user_id: str, session_id: str, backend: object) -> str:
+    """Build the full storage path string that ``retrieve()`` expects.
+
+    ``store()`` returns a path like ``gcs://bucket/workspaces/...`` or
+    ``local://workspace_id/file_id/filename``.  Since we use deterministic
+    arguments we can reconstruct the same path for download/delete without
+    having stored the return value.
+    """
+    from backend.util.workspace_storage import GCSWorkspaceStorage
+
+    wid, fid, fname = _storage_path_parts(user_id, session_id)
+
+    if isinstance(backend, GCSWorkspaceStorage):
+        blob = f"workspaces/{wid}/{fid}/{fname}"
+        return f"gcs://{backend.bucket_name}/{blob}"
+    else:
+        # LocalWorkspaceStorage returns local://{relative_path}
+        return f"local://{wid}/{fid}/{fname}"
+
+
+async def upload_transcript(user_id: str, session_id: str, content: str) -> None:
+    """Strip progress entries and upload transcript to bucket storage.
+
+    Safety: only overwrites when the new (stripped) transcript is larger than
+    what is already stored.  Since JSONL is append-only, the latest transcript
+    is always the longest.  This prevents a slow/stale background task from
+    clobbering a newer upload from a concurrent turn.
+    """
+    from backend.util.workspace_storage import get_workspace_storage
+
+    stripped = strip_progress_entries(content)
+    if not validate_transcript(stripped):
+        logger.warning(
+            f"[Transcript] Skipping upload — stripped content is not a valid "
+            f"transcript for session {session_id}"
+        )
+        return
+
+    storage = await get_workspace_storage()
+    wid, fid, fname = _storage_path_parts(user_id, session_id)
+    encoded = stripped.encode("utf-8")
+    new_size = len(encoded)
+
+    # Check existing transcript size to avoid overwriting newer with older
+    path = _build_storage_path(user_id, session_id, storage)
+    try:
+        existing = await storage.retrieve(path)
+        if len(existing) >= new_size:
+            logger.info(
+                f"[Transcript] Skipping upload — existing transcript "
+                f"({len(existing)}B) >= new ({new_size}B) for session "
+                f"{session_id}"
+            )
+            return
+    except (FileNotFoundError, Exception):
+        pass  # No existing transcript or retrieval error — proceed with upload
+
+    await storage.store(
+        workspace_id=wid,
+        file_id=fid,
+        filename=fname,
+        content=encoded,
+    )
+    logger.info(
+        f"[Transcript] Uploaded {new_size} bytes "
+        f"(stripped from {len(content)}) for session {session_id}"
+    )
+
+
+async def download_transcript(user_id: str, session_id: str) -> str | None:
+    """Download transcript from bucket storage.
+
+    Returns the JSONL content string, or ``None`` if not found.
+    """
+    from backend.util.workspace_storage import get_workspace_storage
+
+    storage = await get_workspace_storage()
+    path = _build_storage_path(user_id, session_id, storage)
+
+    try:
+        data = await storage.retrieve(path)
+        content = data.decode("utf-8")
+        logger.info(
+            f"[Transcript] Downloaded {len(content)} bytes for session {session_id}"
+        )
+        return content
+    except FileNotFoundError:
+        logger.debug(f"[Transcript] No transcript in storage for {session_id}")
+        return None
+    except Exception as e:
+        logger.warning(f"[Transcript] Failed to download transcript: {e}")
+        return None
+
+
+async def delete_transcript(user_id: str, session_id: str) -> None:
+    """Delete transcript from bucket storage (e.g. after resume failure)."""
+    from backend.util.workspace_storage import get_workspace_storage
+
+    storage = await get_workspace_storage()
+    path = _build_storage_path(user_id, session_id, storage)
+
+    try:
+        await storage.delete(path)
+        logger.info(f"[Transcript] Deleted transcript for session {session_id}")
+    except Exception as e:
+        logger.warning(f"[Transcript] Failed to delete transcript: {e}")

autogpt_platform/backend/backend/copilot/sdk/transcript_test.py

@@ -0,0 +1,255 @@
+"""Unit tests for JSONL transcript management utilities."""
+
+import json
+import os
+
+from .transcript import (
+    STRIPPABLE_TYPES,
+    read_transcript_file,
+    strip_progress_entries,
+    validate_transcript,
+    write_transcript_to_tempfile,
+)
+
+
+def _make_jsonl(*entries: dict) -> str:
+    return "\n".join(json.dumps(e) for e in entries) + "\n"
+
+
+# --- Fixtures ---
+
+
+METADATA_LINE = {"type": "queue-operation", "subtype": "create"}
+FILE_HISTORY = {"type": "file-history-snapshot", "files": []}
+USER_MSG = {"type": "user", "uuid": "u1", "message": {"role": "user", "content": "hi"}}
+ASST_MSG = {
+    "type": "assistant",
+    "uuid": "a1",
+    "parentUuid": "u1",
+    "message": {"role": "assistant", "content": "hello"},
+}
+PROGRESS_ENTRY = {
+    "type": "progress",
+    "uuid": "p1",
+    "parentUuid": "u1",
+    "data": {"type": "bash_progress", "stdout": "running..."},
+}
+
+VALID_TRANSCRIPT = _make_jsonl(METADATA_LINE, FILE_HISTORY, USER_MSG, ASST_MSG)
+
+
+# --- read_transcript_file ---
+
+
+class TestReadTranscriptFile:
+    def test_returns_content_for_valid_file(self, tmp_path):
+        path = tmp_path / "session.jsonl"
+        path.write_text(VALID_TRANSCRIPT)
+        result = read_transcript_file(str(path))
+        assert result is not None
+        assert "user" in result
+
+    def test_returns_none_for_missing_file(self):
+        assert read_transcript_file("/nonexistent/path.jsonl") is None
+
+    def test_returns_none_for_empty_path(self):
+        assert read_transcript_file("") is None
+
+    def test_returns_none_for_empty_file(self, tmp_path):
+        path = tmp_path / "empty.jsonl"
+        path.write_text("")
+        assert read_transcript_file(str(path)) is None
+
+    def test_returns_none_for_metadata_only(self, tmp_path):
+        content = _make_jsonl(METADATA_LINE, FILE_HISTORY)
+        path = tmp_path / "meta.jsonl"
+        path.write_text(content)
+        assert read_transcript_file(str(path)) is None
+
+    def test_returns_none_for_invalid_json(self, tmp_path):
+        path = tmp_path / "bad.jsonl"
+        path.write_text("not json\n{}\n{}\n")
+        assert read_transcript_file(str(path)) is None
+
+    def test_no_size_limit(self, tmp_path):
+        """Large files are accepted — bucket storage has no size limit."""
+        big_content = {"type": "user", "uuid": "u9", "data": "x" * 1_000_000}
+        content = _make_jsonl(METADATA_LINE, FILE_HISTORY, big_content, ASST_MSG)
+        path = tmp_path / "big.jsonl"
+        path.write_text(content)
+        result = read_transcript_file(str(path))
+        assert result is not None
+
+
+# --- write_transcript_to_tempfile ---
+
+
+class TestWriteTranscriptToTempfile:
+    """Tests use /tmp/copilot-* paths to satisfy the sandbox prefix check."""
+
+    def test_writes_file_and_returns_path(self):
+        cwd = "/tmp/copilot-test-write"
+        try:
+            result = write_transcript_to_tempfile(
+                VALID_TRANSCRIPT, "sess-1234-abcd", cwd
+            )
+            assert result is not None
+            assert os.path.isfile(result)
+            assert result.endswith(".jsonl")
+            with open(result) as f:
+                assert f.read() == VALID_TRANSCRIPT
+        finally:
+            import shutil
+
+            shutil.rmtree(cwd, ignore_errors=True)
+
+    def test_creates_parent_directory(self):
+        cwd = "/tmp/copilot-test-mkdir"
+        try:
+            result = write_transcript_to_tempfile(VALID_TRANSCRIPT, "sess-1234", cwd)
+            assert result is not None
+            assert os.path.isdir(cwd)
+        finally:
+            import shutil
+
+            shutil.rmtree(cwd, ignore_errors=True)
+
+    def test_uses_session_id_prefix(self):
+        cwd = "/tmp/copilot-test-prefix"
+        try:
+            result = write_transcript_to_tempfile(
+                VALID_TRANSCRIPT, "abcdef12-rest", cwd
+            )
+            assert result is not None
+            assert "abcdef12" in os.path.basename(result)
+        finally:
+            import shutil
+
+            shutil.rmtree(cwd, ignore_errors=True)
+
+    def test_rejects_cwd_outside_sandbox(self, tmp_path):
+        cwd = str(tmp_path / "not-copilot")
+        result = write_transcript_to_tempfile(VALID_TRANSCRIPT, "sess-1234", cwd)
+        assert result is None
+
+
+# --- validate_transcript ---
+
+
+class TestValidateTranscript:
+    def test_valid_transcript(self):
+        assert validate_transcript(VALID_TRANSCRIPT) is True
+
+    def test_none_content(self):
+        assert validate_transcript(None) is False
+
+    def test_empty_content(self):
+        assert validate_transcript("") is False
+
+    def test_metadata_only(self):
+        content = _make_jsonl(METADATA_LINE, FILE_HISTORY)
+        assert validate_transcript(content) is False
+
+    def test_user_only_no_assistant(self):
+        content = _make_jsonl(METADATA_LINE, FILE_HISTORY, USER_MSG)
+        assert validate_transcript(content) is False
+
+    def test_assistant_only_no_user(self):
+        content = _make_jsonl(METADATA_LINE, FILE_HISTORY, ASST_MSG)
+        assert validate_transcript(content) is False
+
+    def test_invalid_json_returns_false(self):
+        assert validate_transcript("not json\n{}\n{}\n") is False
+
+
+# --- strip_progress_entries ---
+
+
+class TestStripProgressEntries:
+    def test_strips_all_strippable_types(self):
+        """All STRIPPABLE_TYPES are removed from the output."""
+        entries = [
+            USER_MSG,
+            {"type": "progress", "uuid": "p1", "parentUuid": "u1"},
+            {"type": "file-history-snapshot", "files": []},
+            {"type": "queue-operation", "subtype": "create"},
+            {"type": "summary", "text": "..."},
+            {"type": "pr-link", "url": "..."},
+            ASST_MSG,
+        ]
+        result = strip_progress_entries(_make_jsonl(*entries))
+        result_types = {json.loads(line)["type"] for line in result.strip().split("\n")}
+        assert result_types == {"user", "assistant"}
+        for stype in STRIPPABLE_TYPES:
+            assert stype not in result_types
+
+    def test_reparents_children_of_stripped_entries(self):
+        """An assistant message whose parent is a progress entry gets reparented."""
+        progress = {
+            "type": "progress",
+            "uuid": "p1",
+            "parentUuid": "u1",
+            "data": {"type": "bash_progress"},
+        }
+        asst = {
+            "type": "assistant",
+            "uuid": "a1",
+            "parentUuid": "p1",  # Points to progress
+            "message": {"role": "assistant", "content": "done"},
+        }
+        content = _make_jsonl(USER_MSG, progress, asst)
+        result = strip_progress_entries(content)
+        lines = [json.loads(line) for line in result.strip().split("\n")]
+
+        asst_entry = next(e for e in lines if e["type"] == "assistant")
+        # Should be reparented to u1 (the user message)
+        assert asst_entry["parentUuid"] == "u1"
+
+    def test_reparents_through_chain(self):
+        """Reparenting walks through multiple stripped entries."""
+        p1 = {"type": "progress", "uuid": "p1", "parentUuid": "u1"}
+        p2 = {"type": "progress", "uuid": "p2", "parentUuid": "p1"}
+        p3 = {"type": "progress", "uuid": "p3", "parentUuid": "p2"}
+        asst = {
+            "type": "assistant",
+            "uuid": "a1",
+            "parentUuid": "p3",  # 3 levels deep
+            "message": {"role": "assistant", "content": "done"},
+        }
+        content = _make_jsonl(USER_MSG, p1, p2, p3, asst)
+        result = strip_progress_entries(content)
+        lines = [json.loads(line) for line in result.strip().split("\n")]
+
+        asst_entry = next(e for e in lines if e["type"] == "assistant")
+        assert asst_entry["parentUuid"] == "u1"
+
+    def test_preserves_non_strippable_entries(self):
+        """User, assistant, and system entries are preserved."""
+        system = {"type": "system", "uuid": "s1", "message": "prompt"}
+        content = _make_jsonl(system, USER_MSG, ASST_MSG)
+        result = strip_progress_entries(content)
+        result_types = [json.loads(line)["type"] for line in result.strip().split("\n")]
+        assert result_types == ["system", "user", "assistant"]
+
+    def test_empty_input(self):
+        result = strip_progress_entries("")
+        # Should return just a newline (empty content stripped)
+        assert result.strip() == ""
+
+    def test_no_strippable_entries(self):
+        """When there's nothing to strip, output matches input structure."""
+        content = _make_jsonl(USER_MSG, ASST_MSG)
+        result = strip_progress_entries(content)
+        result_lines = result.strip().split("\n")
+        assert len(result_lines) == 2
+
+    def test_handles_entries_without_uuid(self):
+        """Entries without uuid field are handled gracefully."""
+        no_uuid = {"type": "queue-operation", "subtype": "create"}
+        content = _make_jsonl(no_uuid, USER_MSG, ASST_MSG)
+        result = strip_progress_entries(content)
+        result_types = [json.loads(line)["type"] for line in result.strip().split("\n")]
+        # queue-operation is strippable
+        assert "queue-operation" not in result_types
+        assert "user" in result_types
+        assert "assistant" in result_types

autogpt_platform/backend/backend/copilot/service.py

@@ -27,6 +27,7 @@ from openai.types.chat import (
     ChatCompletionToolParam,
 )
 
+from backend.data.db_accessors import chat_db
 from backend.data.redis_client import get_redis_async
 from backend.data.understanding import (
     format_understanding_for_prompt,
@@ -35,7 +36,6 @@ from backend.data.understanding import (
 from backend.util.exceptions import NotFoundError
 from backend.util.settings import AppEnvironment, Settings
 
-from . import db as chat_db
 from . import stream_registry
 from .config import ChatConfig
 from .model import (
@@ -245,12 +245,16 @@ async def _get_system_prompt_template(context: str) -> str:
     return DEFAULT_SYSTEM_PROMPT.format(users_information=context)
 
 
-async def _build_system_prompt(user_id: str | None) -> tuple[str, Any]:
+async def _build_system_prompt(
+    user_id: str | None, has_conversation_history: bool = False
+) -> tuple[str, Any]:
     """Build the full system prompt including business understanding if available.
 
     Args:
-        user_id: The user ID for fetching business understanding
-                     If "default" and this is the user's first session, will use "onboarding" instead.
+        user_id: The user ID for fetching business understanding.
+        has_conversation_history: Whether there's existing conversation history.
+            If True, we don't tell the model to greet/introduce (since they're
+            already in a conversation).
 
     Returns:
         Tuple of (compiled prompt string, business understanding object)
@@ -266,6 +270,8 @@ async def _build_system_prompt(user_id: str | None) -> tuple[str, Any]:
 
     if understanding:
         context = format_understanding_for_prompt(understanding)
+    elif has_conversation_history:
+        context = "No prior understanding saved yet. Continue the existing conversation naturally."
     else:
         context = "This is the first time you are meeting the user. Greet them and introduce them to the platform"
 
@@ -374,7 +380,6 @@ async def stream_chat_completion(
 
     Raises:
         NotFoundError: If session_id is invalid
-        ValueError: If max_context_messages is exceeded
 
     """
     completion_start = time.monotonic()
@@ -459,8 +464,9 @@ async def stream_chat_completion(
 
     # Generate title for new sessions on first user message (non-blocking)
     # Check: is_user_message, no title yet, and this is the first user message
-    if is_user_message and message and not session.title:
-        user_messages = [m for m in session.messages if m.role == "user"]
+    user_messages = [m for m in session.messages if m.role == "user"]
+    first_user_msg = message or (user_messages[0].content if user_messages else None)
+    if is_user_message and first_user_msg and not session.title:
         if len(user_messages) == 1:
             # First user message - generate title in background
             import asyncio
@@ -468,7 +474,7 @@ async def stream_chat_completion(
             # Capture only the values we need (not the session object) to avoid
             # stale data issues when the main flow modifies the session
             captured_session_id = session_id
-            captured_message = message
+            captured_message = first_user_msg
             captured_user_id = user_id
 
             async def _update_title():
@@ -1237,7 +1243,7 @@ async def _stream_chat_chunks(
 
                 total_time = (time_module.perf_counter() - stream_chunks_start) * 1000
                 logger.info(
-                    f"[TIMING] _stream_chat_chunks COMPLETED in {total_time/1000:.1f}s; "
+                    f"[TIMING] _stream_chat_chunks COMPLETED in {total_time / 1000:.1f}s; "
                     f"session={session.session_id}, user={session.user_id}",
                     extra={"json_fields": {**log_meta, "total_time_ms": total_time}},
                 )
@@ -1245,6 +1251,7 @@ async def _stream_chat_chunks(
                 return
             except Exception as e:
                 last_error = e
+
                 if _is_retryable_error(e) and retry_count < MAX_RETRIES:
                     retry_count += 1
                     # Calculate delay with exponential backoff
@@ -1260,12 +1267,27 @@ async def _stream_chat_chunks(
                     continue  # Retry the stream
                 else:
                     # Non-retryable error or max retries exceeded
-                    logger.error(
-                        f"Error in stream (not retrying): {e!s}",
-                        exc_info=True,
+                    _log_api_error(
+                        error=e,
+                        context="stream (not retrying)",
+                        session_id=session.session_id if session else None,
+                        message_count=len(messages) if messages else None,
+                        model=model,
+                        retry_count=retry_count,
                     )
                     error_code = None
                     error_text = str(e)
+
+                    error_details = _extract_api_error_details(e)
+                    if error_details.get("response_body"):
+                        body = error_details["response_body"]
+                        if isinstance(body, dict):
+                            err = body.get("error")
+                            if isinstance(err, dict) and err.get("message"):
+                                error_text = err["message"]
+                            elif body.get("message"):
+                                error_text = body["message"]
+
                     if _is_region_blocked_error(e):
                         error_code = "MODEL_NOT_AVAILABLE_REGION"
                         error_text = (
@@ -1282,9 +1304,13 @@ async def _stream_chat_chunks(
 
         # If we exit the retry loop without returning, it means we exhausted retries
         if last_error:
-            logger.error(
-                f"Max retries ({MAX_RETRIES}) exceeded. Last error: {last_error!s}",
-                exc_info=True,
+            _log_api_error(
+                error=last_error,
+                context=f"stream (max retries {MAX_RETRIES} exceeded)",
+                session_id=session.session_id if session else None,
+                message_count=len(messages) if messages else None,
+                model=model,
+                retry_count=MAX_RETRIES,
             )
             yield StreamError(errorText=f"Max retries exceeded: {last_error!s}")
             yield StreamFinish()
@@ -1744,7 +1770,7 @@ async def _update_pending_operation(
     This is called by background tasks when long-running operations complete.
     """
     # Update the message in database
-    updated = await chat_db.update_tool_message_content(
+    updated = await chat_db().update_tool_message_content(
         session_id=session_id,
         tool_call_id=tool_call_id,
         new_content=result,
@@ -1857,6 +1883,7 @@ async def _generate_llm_continuation(
                 break  # Success, exit retry loop
             except Exception as e:
                 last_error = e
+
                 if _is_retryable_error(e) and retry_count < MAX_RETRIES:
                     retry_count += 1
                     delay = min(
@@ -1870,17 +1897,25 @@ async def _generate_llm_continuation(
                     await asyncio.sleep(delay)
                     continue
                 else:
-                    # Non-retryable error - log and exit gracefully
-                    logger.error(
-                        f"Non-retryable error in LLM continuation: {e!s}",
-                        exc_info=True,
+                    # Non-retryable error - log details and exit gracefully
+                    _log_api_error(
+                        error=e,
+                        context="LLM continuation (not retrying)",
+                        session_id=session_id,
+                        message_count=len(messages) if messages else None,
+                        model=config.model,
+                        retry_count=retry_count,
                     )
                     return
 
         if last_error:
-            logger.error(
-                f"Max retries ({MAX_RETRIES}) exceeded for LLM continuation. "
-                f"Last error: {last_error!s}"
+            _log_api_error(
+                error=last_error,
+                context=f"LLM continuation (max retries {MAX_RETRIES} exceeded)",
+                session_id=session_id,
+                message_count=len(messages) if messages else None,
+                model=config.model,
+                retry_count=MAX_RETRIES,
             )
             return
 
@@ -1920,6 +1955,91 @@ async def _generate_llm_continuation(
         logger.error(f"Failed to generate LLM continuation: {e}", exc_info=True)
 
 
+def _log_api_error(
+    error: Exception,
+    context: str,
+    session_id: str | None = None,
+    message_count: int | None = None,
+    model: str | None = None,
+    retry_count: int = 0,
+) -> None:
+    """Log detailed API error information for debugging."""
+    details = _extract_api_error_details(error)
+    details["context"] = context
+    details["session_id"] = session_id
+    details["message_count"] = message_count
+    details["model"] = model
+    details["retry_count"] = retry_count
+
+    if isinstance(error, RateLimitError):
+        logger.warning(f"Rate limit error in {context}: {details}", exc_info=error)
+    elif isinstance(error, APIConnectionError):
+        logger.warning(f"API connection error in {context}: {details}", exc_info=error)
+    elif isinstance(error, APIStatusError) and error.status_code >= 500:
+        logger.error(f"API server error (5xx) in {context}: {details}", exc_info=error)
+    else:
+        logger.error(f"API error in {context}: {details}", exc_info=error)
+
+
+def _extract_api_error_details(error: Exception) -> dict[str, Any]:
+    """Extract detailed information from OpenAI/OpenRouter API errors."""
+    error_msg = str(error)
+    details: dict[str, Any] = {
+        "error_type": type(error).__name__,
+        "error_message": error_msg[:500] + "..." if len(error_msg) > 500 else error_msg,
+    }
+
+    if hasattr(error, "code"):
+        details["code"] = getattr(error, "code", None)
+    if hasattr(error, "param"):
+        details["param"] = getattr(error, "param", None)
+
+    if isinstance(error, APIStatusError):
+        details["status_code"] = error.status_code
+        details["request_id"] = getattr(error, "request_id", None)
+
+        if hasattr(error, "body") and error.body:
+            details["response_body"] = _sanitize_error_body(error.body)
+
+        if hasattr(error, "response") and error.response:
+            headers = error.response.headers
+            details["openrouter_provider"] = headers.get("x-openrouter-provider")
+            details["openrouter_model"] = headers.get("x-openrouter-model")
+            details["retry_after"] = headers.get("retry-after")
+            details["rate_limit_remaining"] = headers.get("x-ratelimit-remaining")
+
+    return details
+
+
+def _sanitize_error_body(
+    body: Any, max_length: int = 2000
+) -> dict[str, Any] | str | None:
+    """Extract only safe fields from error response body to avoid logging sensitive data."""
+    if not isinstance(body, dict):
+        # Non-dict bodies (e.g., HTML error pages) - return truncated string
+        if body is not None:
+            body_str = str(body)
+            if len(body_str) > max_length:
+                return body_str[:max_length] + "...[truncated]"
+            return body_str
+        return None
+
+    safe_fields = ("message", "type", "code", "param", "error")
+    sanitized: dict[str, Any] = {}
+
+    for field in safe_fields:
+        if field in body:
+            value = body[field]
+            if field == "error" and isinstance(value, dict):
+                sanitized[field] = _sanitize_error_body(value, max_length)
+            elif isinstance(value, str) and len(value) > max_length:
+                sanitized[field] = value[:max_length] + "...[truncated]"
+            else:
+                sanitized[field] = value
+
+    return sanitized if sanitized else None
+
+
 async def _generate_llm_continuation_with_streaming(
     session_id: str,
     user_id: str | None,

autogpt_platform/backend/backend/copilot/service_test.py

@@ -0,0 +1,178 @@
+import asyncio
+import logging
+from os import getenv
+
+import pytest
+
+from . import service as chat_service
+from .model import create_chat_session, get_chat_session, upsert_chat_session
+from .response_model import (
+    StreamError,
+    StreamFinish,
+    StreamTextDelta,
+    StreamToolOutputAvailable,
+)
+from .sdk import service as sdk_service
+from .sdk.transcript import download_transcript
+
+logger = logging.getLogger(__name__)
+
+
+@pytest.mark.asyncio(loop_scope="session")
+async def test_stream_chat_completion(setup_test_user, test_user_id):
+    """
+    Test the stream_chat_completion function.
+    """
+    api_key: str | None = getenv("OPEN_ROUTER_API_KEY")
+    if not api_key:
+        return pytest.skip("OPEN_ROUTER_API_KEY is not set, skipping test")
+
+    session = await create_chat_session(test_user_id)
+
+    has_errors = False
+    has_ended = False
+    assistant_message = ""
+    async for chunk in chat_service.stream_chat_completion(
+        session.session_id, "Hello, how are you?", user_id=session.user_id
+    ):
+        logger.info(chunk)
+        if isinstance(chunk, StreamError):
+            has_errors = True
+        if isinstance(chunk, StreamTextDelta):
+            assistant_message += chunk.delta
+        if isinstance(chunk, StreamFinish):
+            has_ended = True
+
+    assert has_ended, "Chat completion did not end"
+    assert not has_errors, "Error occurred while streaming chat completion"
+    assert assistant_message, "Assistant message is empty"
+
+
+@pytest.mark.asyncio(loop_scope="session")
+async def test_stream_chat_completion_with_tool_calls(setup_test_user, test_user_id):
+    """
+    Test the stream_chat_completion function.
+    """
+    api_key: str | None = getenv("OPEN_ROUTER_API_KEY")
+    if not api_key:
+        return pytest.skip("OPEN_ROUTER_API_KEY is not set, skipping test")
+
+    session = await create_chat_session(test_user_id)
+    session = await upsert_chat_session(session)
+
+    has_errors = False
+    has_ended = False
+    had_tool_calls = False
+    async for chunk in chat_service.stream_chat_completion(
+        session.session_id,
+        "Please find me an agent that can help me with my business. Use the query 'moneny printing agent'",
+        user_id=session.user_id,
+    ):
+        logger.info(chunk)
+        if isinstance(chunk, StreamError):
+            has_errors = True
+
+        if isinstance(chunk, StreamFinish):
+            has_ended = True
+        if isinstance(chunk, StreamToolOutputAvailable):
+            had_tool_calls = True
+
+    assert has_ended, "Chat completion did not end"
+    assert not has_errors, "Error occurred while streaming chat completion"
+    assert had_tool_calls, "Tool calls did not occur"
+    session = await get_chat_session(session.session_id)
+    assert session, "Session not found"
+    assert session.usage, "Usage is empty"
+
+
+@pytest.mark.asyncio(loop_scope="session")
+async def test_sdk_resume_multi_turn(setup_test_user, test_user_id):
+    """Test that the SDK --resume path captures and uses transcripts across turns.
+
+    Turn 1: Send a message containing a unique keyword.
+    Turn 2: Ask the model to recall that keyword — proving the transcript was
+    persisted and restored via --resume.
+    """
+    api_key: str | None = getenv("OPEN_ROUTER_API_KEY")
+    if not api_key:
+        return pytest.skip("OPEN_ROUTER_API_KEY is not set, skipping test")
+
+    from .config import ChatConfig
+
+    cfg = ChatConfig()
+    if not cfg.claude_agent_use_resume:
+        return pytest.skip("CLAUDE_AGENT_USE_RESUME is not enabled, skipping test")
+
+    session = await create_chat_session(test_user_id)
+    session = await upsert_chat_session(session)
+
+    # --- Turn 1: send a message with a unique keyword ---
+    keyword = "ZEPHYR42"
+    turn1_msg = (
+        f"Please remember this special keyword: {keyword}. "
+        "Just confirm you've noted it, keep your response brief."
+    )
+    turn1_text = ""
+    turn1_errors: list[str] = []
+    turn1_ended = False
+
+    async for chunk in sdk_service.stream_chat_completion_sdk(
+        session.session_id,
+        turn1_msg,
+        user_id=test_user_id,
+    ):
+        if isinstance(chunk, StreamTextDelta):
+            turn1_text += chunk.delta
+        elif isinstance(chunk, StreamError):
+            turn1_errors.append(chunk.errorText)
+        elif isinstance(chunk, StreamFinish):
+            turn1_ended = True
+
+    assert turn1_ended, "Turn 1 did not finish"
+    assert not turn1_errors, f"Turn 1 errors: {turn1_errors}"
+    assert turn1_text, "Turn 1 produced no text"
+
+    # Wait for background upload task to complete (retry up to 5s)
+    transcript = None
+    for _ in range(10):
+        await asyncio.sleep(0.5)
+        transcript = await download_transcript(test_user_id, session.session_id)
+        if transcript:
+            break
+    assert transcript, (
+        "Transcript was not uploaded to bucket after turn 1 — "
+        "Stop hook may not have fired or transcript was too small"
+    )
+    logger.info(f"Turn 1 transcript uploaded: {len(transcript)} bytes")
+
+    # Reload session for turn 2
+    session = await get_chat_session(session.session_id, test_user_id)
+    assert session, "Session not found after turn 1"
+
+    # --- Turn 2: ask model to recall the keyword ---
+    turn2_msg = "What was the special keyword I asked you to remember?"
+    turn2_text = ""
+    turn2_errors: list[str] = []
+    turn2_ended = False
+
+    async for chunk in sdk_service.stream_chat_completion_sdk(
+        session.session_id,
+        turn2_msg,
+        user_id=test_user_id,
+        session=session,
+    ):
+        if isinstance(chunk, StreamTextDelta):
+            turn2_text += chunk.delta
+        elif isinstance(chunk, StreamError):
+            turn2_errors.append(chunk.errorText)
+        elif isinstance(chunk, StreamFinish):
+            turn2_ended = True
+
+    assert turn2_ended, "Turn 2 did not finish"
+    assert not turn2_errors, f"Turn 2 errors: {turn2_errors}"
+    assert turn2_text, "Turn 2 produced no text"
+    assert keyword in turn2_text, (
+        f"Model did not recall keyword '{keyword}' in turn 2. "
+        f"Response: {turn2_text[:200]}"
+    )
+    logger.info(f"Turn 2 recalled keyword successfully: {turn2_text[:100]}")

autogpt_platform/backend/backend/copilot/stream_registry.py

@@ -814,6 +814,28 @@ async def get_active_task_for_session(
                 if task_user_id and user_id != task_user_id:
                     continue
 
+                # Auto-expire stale tasks that exceeded stream_timeout
+                created_at_str = meta.get("created_at", "")
+                if created_at_str:
+                    try:
+                        created_at = datetime.fromisoformat(created_at_str)
+                        age_seconds = (
+                            datetime.now(timezone.utc) - created_at
+                        ).total_seconds()
+                        if age_seconds > config.stream_timeout:
+                            logger.warning(
+                                f"[TASK_LOOKUP] Auto-expiring stale task {task_id[:8]}... "
+                                f"(age={age_seconds:.0f}s > timeout={config.stream_timeout}s)"
+                            )
+                            await mark_task_completed(task_id, "failed")
+                            continue
+                    except (ValueError, TypeError):
+                        pass
+
+                logger.info(
+                    f"[TASK_LOOKUP] Found running task {task_id[:8]}... for session {session_id[:8]}..."
+                )
+
                 # Get the last message ID from Redis Stream
                 stream_key = _get_task_stream_key(task_id)
                 last_id = "0-0"

autogpt_platform/backend/backend/copilot/tools/__init__.py

@@ -3,15 +3,18 @@ from typing import TYPE_CHECKING, Any
 
 from openai.types.chat import ChatCompletionToolParam
 
-from backend.api.features.chat.model import ChatSession
-from backend.api.features.chat.tracking import track_tool_called
+from backend.copilot.model import ChatSession
+from backend.copilot.tracking import track_tool_called
 
 from .add_understanding import AddUnderstandingTool
 from .agent_output import AgentOutputTool
 from .base import BaseTool
+from .bash_exec import BashExecTool
+from .check_operation_status import CheckOperationStatusTool
 from .create_agent import CreateAgentTool
 from .customize_agent import CustomizeAgentTool
 from .edit_agent import EditAgentTool
+from .feature_requests import CreateFeatureRequestTool, SearchFeatureRequestsTool
 from .find_agent import FindAgentTool
 from .find_block import FindBlockTool
 from .find_library_agent import FindLibraryAgentTool
@@ -19,6 +22,7 @@ from .get_doc_page import GetDocPageTool
 from .run_agent import RunAgentTool
 from .run_block import RunBlockTool
 from .search_docs import SearchDocsTool
+from .web_fetch import WebFetchTool
 from .workspace_files import (
     DeleteWorkspaceFileTool,
     ListWorkspaceFilesTool,
@@ -27,7 +31,7 @@ from .workspace_files import (
 )
 
 if TYPE_CHECKING:
-    from backend.api.features.chat.response_model import StreamToolOutputAvailable
+    from backend.copilot.response_model import StreamToolOutputAvailable
 
 logger = logging.getLogger(__name__)
 
@@ -43,8 +47,17 @@ TOOL_REGISTRY: dict[str, BaseTool] = {
     "run_agent": RunAgentTool(),
     "run_block": RunBlockTool(),
     "view_agent_output": AgentOutputTool(),
+    "check_operation_status": CheckOperationStatusTool(),
     "search_docs": SearchDocsTool(),
     "get_doc_page": GetDocPageTool(),
+    # Web fetch for safe URL retrieval
+    "web_fetch": WebFetchTool(),
+    # Sandboxed code execution (bubblewrap)
+    "bash_exec": BashExecTool(),
+    # Persistent workspace tools (cloud storage, survives across sessions)
+    # Feature request tools
+    "search_feature_requests": SearchFeatureRequestsTool(),
+    "create_feature_request": CreateFeatureRequestTool(),
     # Workspace tools for CoPilot file operations
     "list_workspace_files": ListWorkspaceFilesTool(),
     "read_workspace_file": ReadWorkspaceFileTool(),

autogpt_platform/backend/backend/copilot/tools/_test_data.py

@@ -6,11 +6,11 @@ import pytest
 from prisma.types import ProfileCreateInput
 from pydantic import SecretStr
 
-from backend.api.features.chat.model import ChatSession
 from backend.api.features.store import db as store_db
 from backend.blocks.firecrawl.scrape import FirecrawlScrapeBlock
 from backend.blocks.io import AgentInputBlock, AgentOutputBlock
 from backend.blocks.llm import AITextGeneratorBlock
+from backend.copilot.model import ChatSession
 from backend.data.db import prisma
 from backend.data.graph import Graph, Link, Node, create_graph
 from backend.data.model import APIKeyCredentials

autogpt_platform/backend/backend/copilot/tools/add_understanding.py

@@ -3,11 +3,9 @@
 import logging
 from typing import Any
 
-from backend.api.features.chat.model import ChatSession
-from backend.data.understanding import (
-    BusinessUnderstandingInput,
-    upsert_business_understanding,
-)
+from backend.copilot.model import ChatSession
+from backend.data.db_accessors import understanding_db
+from backend.data.understanding import BusinessUnderstandingInput
 
 from .base import BaseTool
 from .models import ErrorResponse, ToolResponseBase, UnderstandingUpdatedResponse
@@ -99,7 +97,9 @@ and automations for the user's specific needs."""
         ]
 
         # Upsert with merge
-        understanding = await upsert_business_understanding(user_id, input_data)
+        understanding = await understanding_db().upsert_business_understanding(
+            user_id, input_data
+        )
 
         # Build current understanding summary (filter out empty values)
         current_understanding = {

autogpt_platform/backend/backend/copilot/tools/agent_generator/core.py

@@ -5,9 +5,8 @@ import re
 import uuid
 from typing import Any, NotRequired, TypedDict
 
-from backend.api.features.library import db as library_db
-from backend.api.features.store import db as store_db
-from backend.data.graph import Graph, Link, Node, get_graph, get_store_listed_graphs
+from backend.data.db_accessors import graph_db, library_db, store_db
+from backend.data.graph import Graph, Link, Node
 from backend.util.exceptions import DatabaseError, NotFoundError
 
 from .service import (
@@ -145,8 +144,9 @@ async def get_library_agent_by_id(
     Returns:
         LibraryAgentSummary if found, None otherwise
     """
+    db = library_db()
     try:
-        agent = await library_db.get_library_agent_by_graph_id(user_id, agent_id)
+        agent = await db.get_library_agent_by_graph_id(user_id, agent_id)
         if agent:
             logger.debug(f"Found library agent by graph_id: {agent.name}")
             return LibraryAgentSummary(
@@ -163,7 +163,7 @@ async def get_library_agent_by_id(
         logger.debug(f"Could not fetch library agent by graph_id {agent_id}: {e}")
 
     try:
-        agent = await library_db.get_library_agent(agent_id, user_id)
+        agent = await db.get_library_agent(agent_id, user_id)
         if agent:
             logger.debug(f"Found library agent by library_id: {agent.name}")
             return LibraryAgentSummary(
@@ -215,7 +215,7 @@ async def get_library_agents_for_generation(
         List of LibraryAgentSummary with schemas and recent executions for sub-agent composition
     """
     try:
-        response = await library_db.list_library_agents(
+        response = await library_db().list_library_agents(
             user_id=user_id,
             search_term=search_query,
             page=1,
@@ -272,7 +272,7 @@ async def search_marketplace_agents_for_generation(
         List of LibraryAgentSummary with full input/output schemas
     """
     try:
-        response = await store_db.get_store_agents(
+        response = await store_db().get_store_agents(
             search_query=search_query,
             page=1,
             page_size=max_results,
@@ -286,7 +286,7 @@ async def search_marketplace_agents_for_generation(
             return []
 
         graph_ids = [agent.agent_graph_id for agent in agents_with_graphs]
-        graphs = await get_store_listed_graphs(*graph_ids)
+        graphs = await graph_db().get_store_listed_graphs(*graph_ids)
 
         results: list[LibraryAgentSummary] = []
         for agent in agents_with_graphs:
@@ -673,9 +673,10 @@ async def save_agent_to_library(
         Tuple of (created Graph, LibraryAgent)
     """
     graph = json_to_graph(agent_json)
+    db = library_db()
     if is_update:
-        return await library_db.update_graph_in_library(graph, user_id)
-    return await library_db.create_graph_in_library(graph, user_id)
+        return await db.update_graph_in_library(graph, user_id)
+    return await db.create_graph_in_library(graph, user_id)
 
 
 def graph_to_json(graph: Graph) -> dict[str, Any]:
@@ -735,12 +736,14 @@ async def get_agent_as_json(
     Returns:
         Agent as JSON dict or None if not found
     """
-    graph = await get_graph(agent_id, version=None, user_id=user_id)
+    db = graph_db()
+
+    graph = await db.get_graph(agent_id, version=None, user_id=user_id)
 
     if not graph and user_id:
         try:
-            library_agent = await library_db.get_library_agent(agent_id, user_id)
-            graph = await get_graph(
+            library_agent = await library_db().get_library_agent(agent_id, user_id)
+            graph = await db.get_graph(
                 library_agent.graph_id, version=None, user_id=user_id
             )
         except NotFoundError:

autogpt_platform/backend/backend/copilot/tools/agent_output.py

@@ -7,10 +7,9 @@ from typing import Any
 
 from pydantic import BaseModel, field_validator
 
-from backend.api.features.chat.model import ChatSession
-from backend.api.features.library import db as library_db
 from backend.api.features.library.model import LibraryAgent
-from backend.data import execution as execution_db
+from backend.copilot.model import ChatSession
+from backend.data.db_accessors import execution_db, library_db
 from backend.data.execution import ExecutionStatus, GraphExecution, GraphExecutionMeta
 
 from .base import BaseTool
@@ -165,10 +164,12 @@ class AgentOutputTool(BaseTool):
         Resolve agent from provided identifiers.
         Returns (library_agent, error_message).
         """
+        lib_db = library_db()
+
         # Priority 1: Exact library agent ID
         if library_agent_id:
             try:
-                agent = await library_db.get_library_agent(library_agent_id, user_id)
+                agent = await lib_db.get_library_agent(library_agent_id, user_id)
                 return agent, None
             except Exception as e:
                 logger.warning(f"Failed to get library agent by ID: {e}")
@@ -182,7 +183,7 @@ class AgentOutputTool(BaseTool):
                 return None, f"Agent '{store_slug}' not found in marketplace"
 
             # Find in user's library by graph_id
-            agent = await library_db.get_library_agent_by_graph_id(user_id, graph.id)
+            agent = await lib_db.get_library_agent_by_graph_id(user_id, graph.id)
             if not agent:
                 return (
                     None,
@@ -194,7 +195,7 @@ class AgentOutputTool(BaseTool):
         # Priority 3: Fuzzy name search in library
         if agent_name:
             try:
-                response = await library_db.list_library_agents(
+                response = await lib_db.list_library_agents(
                     user_id=user_id,
                     search_term=agent_name,
                     page_size=5,
@@ -228,9 +229,11 @@ class AgentOutputTool(BaseTool):
         Fetch execution(s) based on filters.
         Returns (single_execution, available_executions_meta, error_message).
         """
+        exec_db = execution_db()
+
         # If specific execution_id provided, fetch it directly
         if execution_id:
-            execution = await execution_db.get_graph_execution(
+            execution = await exec_db.get_graph_execution(
                 user_id=user_id,
                 execution_id=execution_id,
                 include_node_executions=False,
@@ -240,7 +243,7 @@ class AgentOutputTool(BaseTool):
             return execution, [], None
 
         # Get completed executions with time filters
-        executions = await execution_db.get_graph_executions(
+        executions = await exec_db.get_graph_executions(
             graph_id=graph_id,
             user_id=user_id,
             statuses=[ExecutionStatus.COMPLETED],
@@ -254,7 +257,7 @@ class AgentOutputTool(BaseTool):
 
         # If only one execution, fetch full details
         if len(executions) == 1:
-            full_execution = await execution_db.get_graph_execution(
+            full_execution = await exec_db.get_graph_execution(
                 user_id=user_id,
                 execution_id=executions[0].id,
                 include_node_executions=False,
@@ -262,7 +265,7 @@ class AgentOutputTool(BaseTool):
             return full_execution, [], None
 
         # Multiple executions - return latest with full details, plus list of available
-        full_execution = await execution_db.get_graph_execution(
+        full_execution = await exec_db.get_graph_execution(
             user_id=user_id,
             execution_id=executions[0].id,
             include_node_executions=False,
@@ -380,7 +383,7 @@ class AgentOutputTool(BaseTool):
             and not input_data.store_slug
         ):
             # Fetch execution directly to get graph_id
-            execution = await execution_db.get_graph_execution(
+            execution = await execution_db().get_graph_execution(
                 user_id=user_id,
                 execution_id=input_data.execution_id,
                 include_node_executions=False,
@@ -392,7 +395,7 @@ class AgentOutputTool(BaseTool):
                 )
 
             # Find library agent by graph_id
-            agent = await library_db.get_library_agent_by_graph_id(
+            agent = await library_db().get_library_agent_by_graph_id(
                 user_id, execution.graph_id
             )
             if not agent:

autogpt_platform/backend/backend/copilot/tools/agent_search.py

@@ -4,8 +4,7 @@ import logging
 import re
 from typing import Literal
 
-from backend.api.features.library import db as library_db
-from backend.api.features.store import db as store_db
+from backend.data.db_accessors import library_db, store_db
 from backend.util.exceptions import DatabaseError, NotFoundError
 
 from .models import (
@@ -45,8 +44,10 @@ async def _get_library_agent_by_id(user_id: str, agent_id: str) -> AgentInfo | N
     Returns:
         AgentInfo if found, None otherwise
     """
+    lib_db = library_db()
+
     try:
-        agent = await library_db.get_library_agent_by_graph_id(user_id, agent_id)
+        agent = await lib_db.get_library_agent_by_graph_id(user_id, agent_id)
         if agent:
             logger.debug(f"Found library agent by graph_id: {agent.name}")
             return AgentInfo(
@@ -71,7 +72,7 @@ async def _get_library_agent_by_id(user_id: str, agent_id: str) -> AgentInfo | N
         )
 
     try:
-        agent = await library_db.get_library_agent(agent_id, user_id)
+        agent = await lib_db.get_library_agent(agent_id, user_id)
         if agent:
             logger.debug(f"Found library agent by library_id: {agent.name}")
             return AgentInfo(
@@ -133,7 +134,7 @@ async def search_agents(
     try:
         if source == "marketplace":
             logger.info(f"Searching marketplace for: {query}")
-            results = await store_db.get_store_agents(search_query=query, page_size=5)
+            results = await store_db().get_store_agents(search_query=query, page_size=5)
             for agent in results.agents:
                 agents.append(
                     AgentInfo(
@@ -159,7 +160,7 @@ async def search_agents(
 
             if not agents:
                 logger.info(f"Searching user library for: {query}")
-                results = await library_db.list_library_agents(
+                results = await library_db().list_library_agents(
                     user_id=user_id,  # type: ignore[arg-type]
                     search_term=query,
                     page_size=10,

autogpt_platform/backend/backend/copilot/tools/base.py

@@ -5,8 +5,8 @@ from typing import Any
 
 from openai.types.chat import ChatCompletionToolParam
 
-from backend.api.features.chat.model import ChatSession
-from backend.api.features.chat.response_model import StreamToolOutputAvailable
+from backend.copilot.model import ChatSession
+from backend.copilot.response_model import StreamToolOutputAvailable
 
 from .models import ErrorResponse, NeedLoginResponse, ToolResponseBase
 

autogpt_platform/backend/backend/copilot/tools/bash_exec.py

@@ -0,0 +1,124 @@
+"""Bash execution tool — run shell commands in a bubblewrap sandbox.
+
+Full Bash scripting is allowed (loops, conditionals, pipes, functions, etc.).
+Safety comes from OS-level isolation (bubblewrap): only system dirs visible
+read-only, writable workspace only, clean env, no network.
+
+Requires bubblewrap (``bwrap``) — the tool is disabled when bwrap is not
+available (e.g. macOS development).
+"""
+
+import logging
+from typing import Any
+
+from backend.copilot.model import ChatSession
+
+from .base import BaseTool
+from .models import BashExecResponse, ErrorResponse, ToolResponseBase
+from .sandbox import get_workspace_dir, has_full_sandbox, run_sandboxed
+
+logger = logging.getLogger(__name__)
+
+
+class BashExecTool(BaseTool):
+    """Execute Bash commands in a bubblewrap sandbox."""
+
+    @property
+    def name(self) -> str:
+        return "bash_exec"
+
+    @property
+    def description(self) -> str:
+        if not has_full_sandbox():
+            return (
+                "Bash execution is DISABLED — bubblewrap sandbox is not "
+                "available on this platform. Do not call this tool."
+            )
+        return (
+            "Execute a Bash command or script in a bubblewrap sandbox. "
+            "Full Bash scripting is supported (loops, conditionals, pipes, "
+            "functions, etc.). "
+            "The sandbox shares the same working directory as the SDK Read/Write "
+            "tools — files created by either are accessible to both. "
+            "SECURITY: Only system directories (/usr, /bin, /lib, /etc) are "
+            "visible read-only, the per-session workspace is the only writable "
+            "path, environment variables are wiped (no secrets), all network "
+            "access is blocked at the kernel level, and resource limits are "
+            "enforced (max 64 processes, 512MB memory, 50MB file size). "
+            "Application code, configs, and other directories are NOT accessible. "
+            "To fetch web content, use the web_fetch tool instead. "
+            "Execution is killed after the timeout (default 30s, max 120s). "
+            "Returns stdout and stderr. "
+            "Useful for file manipulation, data processing with Unix tools "
+            "(grep, awk, sed, jq, etc.), and running shell scripts."
+        )
+
+    @property
+    def parameters(self) -> dict[str, Any]:
+        return {
+            "type": "object",
+            "properties": {
+                "command": {
+                    "type": "string",
+                    "description": "Bash command or script to execute.",
+                },
+                "timeout": {
+                    "type": "integer",
+                    "description": (
+                        "Max execution time in seconds (default 30, max 120)."
+                    ),
+                    "default": 30,
+                },
+            },
+            "required": ["command"],
+        }
+
+    @property
+    def requires_auth(self) -> bool:
+        return False
+
+    async def _execute(
+        self,
+        user_id: str | None,
+        session: ChatSession,
+        **kwargs: Any,
+    ) -> ToolResponseBase:
+        session_id = session.session_id if session else None
+
+        if not has_full_sandbox():
+            return ErrorResponse(
+                message="bash_exec requires bubblewrap sandbox (Linux only).",
+                error="sandbox_unavailable",
+                session_id=session_id,
+            )
+
+        command: str = (kwargs.get("command") or "").strip()
+        timeout: int = kwargs.get("timeout", 30)
+
+        if not command:
+            return ErrorResponse(
+                message="No command provided.",
+                error="empty_command",
+                session_id=session_id,
+            )
+
+        workspace = get_workspace_dir(session_id or "default")
+
+        stdout, stderr, exit_code, timed_out = await run_sandboxed(
+            command=["bash", "-c", command],
+            cwd=workspace,
+            timeout=timeout,
+        )
+
+        return BashExecResponse(
+            message=(
+                "Execution timed out"
+                if timed_out
+                else f"Command executed (exit {exit_code})"
+            ),
+            stdout=stdout,
+            stderr=stderr,
+            exit_code=exit_code,
+            timed_out=timed_out,
+            session_id=session_id,
+        )

autogpt_platform/backend/backend/copilot/tools/check_operation_status.py

@@ -0,0 +1,124 @@
+"""CheckOperationStatusTool — query the status of a long-running operation."""
+
+import logging
+from typing import Any
+
+from backend.copilot.model import ChatSession
+
+from .base import BaseTool
+from .models import ErrorResponse, ResponseType, ToolResponseBase
+
+logger = logging.getLogger(__name__)
+
+
+class OperationStatusResponse(ToolResponseBase):
+    """Response for check_operation_status tool."""
+
+    type: ResponseType = ResponseType.OPERATION_STATUS
+    task_id: str
+    operation_id: str
+    status: str  # "running", "completed", "failed"
+    tool_name: str | None = None
+    message: str = ""
+
+
+class CheckOperationStatusTool(BaseTool):
+    """Check the status of a long-running operation (create_agent, edit_agent, etc.).
+
+    The CoPilot uses this tool to report back to the user whether an
+    operation that was started earlier has completed, failed, or is still
+    running.
+    """
+
+    @property
+    def name(self) -> str:
+        return "check_operation_status"
+
+    @property
+    def description(self) -> str:
+        return (
+            "Check the current status of a long-running operation such as "
+            "create_agent or edit_agent. Accepts either an operation_id or "
+            "task_id from a previous operation_started response. "
+            "Returns the current status: running, completed, or failed."
+        )
+
+    @property
+    def parameters(self) -> dict[str, Any]:
+        return {
+            "type": "object",
+            "properties": {
+                "operation_id": {
+                    "type": "string",
+                    "description": (
+                        "The operation_id from an operation_started response."
+                    ),
+                },
+                "task_id": {
+                    "type": "string",
+                    "description": (
+                        "The task_id from an operation_started response. "
+                        "Used as fallback if operation_id is not provided."
+                    ),
+                },
+            },
+            "required": [],
+        }
+
+    @property
+    def requires_auth(self) -> bool:
+        return False
+
+    async def _execute(
+        self,
+        user_id: str | None,
+        session: ChatSession,
+        **kwargs,
+    ) -> ToolResponseBase:
+        from backend.copilot import stream_registry
+
+        operation_id = (kwargs.get("operation_id") or "").strip()
+        task_id = (kwargs.get("task_id") or "").strip()
+
+        if not operation_id and not task_id:
+            return ErrorResponse(
+                message="Please provide an operation_id or task_id.",
+                error="missing_parameter",
+            )
+
+        task = None
+        if operation_id:
+            task = await stream_registry.find_task_by_operation_id(operation_id)
+        if task is None and task_id:
+            task = await stream_registry.get_task(task_id)
+
+        if task is None:
+            # Task not in Redis — it may have already expired (TTL).
+            # Check conversation history for the result instead.
+            return ErrorResponse(
+                message=(
+                    "Operation not found — it may have already completed and "
+                    "expired from the status tracker. Check the conversation "
+                    "history for the result."
+                ),
+                error="not_found",
+            )
+
+        status_messages = {
+            "running": (
+                f"The {task.tool_name or 'operation'} is still running. "
+                "Please wait for it to complete."
+            ),
+            "completed": (
+                f"The {task.tool_name or 'operation'} has completed successfully."
+            ),
+            "failed": f"The {task.tool_name or 'operation'} has failed.",
+        }
+
+        return OperationStatusResponse(
+            task_id=task.task_id,
+            operation_id=task.operation_id,
+            status=task.status,
+            tool_name=task.tool_name,
+            message=status_messages.get(task.status, f"Status: {task.status}"),
+        )

autogpt_platform/backend/backend/copilot/tools/create_agent.py

@@ -3,7 +3,7 @@
 import logging
 from typing import Any
 
-from backend.api.features.chat.model import ChatSession
+from backend.copilot.model import ChatSession
 
 from .agent_generator import (
     AgentGeneratorNotConfiguredError,

autogpt_platform/backend/backend/copilot/tools/customize_agent.py

@@ -3,9 +3,9 @@
 import logging
 from typing import Any
 
-from backend.api.features.chat.model import ChatSession
-from backend.api.features.store import db as store_db
 from backend.api.features.store.exceptions import AgentNotFoundError
+from backend.copilot.model import ChatSession
+from backend.data.db_accessors import store_db as get_store_db
 
 from .agent_generator import (
     AgentGeneratorNotConfiguredError,
@@ -137,6 +137,8 @@ class CustomizeAgentTool(BaseTool):
 
         creator_username, agent_slug = parts
 
+        store_db = get_store_db()
+
         # Fetch the marketplace agent details
         try:
             agent_details = await store_db.get_store_agent_details(

autogpt_platform/backend/backend/copilot/tools/edit_agent.py

@@ -3,7 +3,7 @@
 import logging
 from typing import Any
 
-from backend.api.features.chat.model import ChatSession
+from backend.copilot.model import ChatSession
 
 from .agent_generator import (
     AgentGeneratorNotConfiguredError,

autogpt_platform/backend/backend/copilot/tools/feature_requests.py

@@ -0,0 +1,451 @@
+"""Feature request tools - search and create feature requests via Linear."""
+
+import logging
+from typing import Any
+
+from pydantic import SecretStr
+
+from backend.blocks.linear._api import LinearClient
+from backend.copilot.model import ChatSession
+from backend.data.db_accessors import user_db
+from backend.data.model import APIKeyCredentials
+from backend.util.settings import Settings
+
+from .base import BaseTool
+from .models import (
+    ErrorResponse,
+    FeatureRequestCreatedResponse,
+    FeatureRequestInfo,
+    FeatureRequestSearchResponse,
+    NoResultsResponse,
+    ToolResponseBase,
+)
+
+logger = logging.getLogger(__name__)
+
+MAX_SEARCH_RESULTS = 10
+
+# GraphQL queries/mutations
+SEARCH_ISSUES_QUERY = """
+query SearchFeatureRequests($term: String!, $filter: IssueFilter, $first: Int) {
+  searchIssues(term: $term, filter: $filter, first: $first) {
+    nodes {
+      id
+      identifier
+      title
+      description
+    }
+  }
+}
+"""
+
+CUSTOMER_UPSERT_MUTATION = """
+mutation CustomerUpsert($input: CustomerUpsertInput!) {
+  customerUpsert(input: $input) {
+    success
+    customer {
+      id
+      name
+      externalIds
+    }
+  }
+}
+"""
+
+ISSUE_CREATE_MUTATION = """
+mutation IssueCreate($input: IssueCreateInput!) {
+  issueCreate(input: $input) {
+    success
+    issue {
+      id
+      identifier
+      title
+      url
+    }
+  }
+}
+"""
+
+CUSTOMER_NEED_CREATE_MUTATION = """
+mutation CustomerNeedCreate($input: CustomerNeedCreateInput!) {
+  customerNeedCreate(input: $input) {
+    success
+    need {
+      id
+      body
+      customer {
+        id
+        name
+      }
+      issue {
+        id
+        identifier
+        title
+        url
+      }
+    }
+  }
+}
+"""
+
+
+_settings: Settings | None = None
+
+
+def _get_settings() -> Settings:
+    global _settings
+    if _settings is None:
+        _settings = Settings()
+    return _settings
+
+
+def _get_linear_config() -> tuple[LinearClient, str, str]:
+    """Return a configured Linear client, project ID, and team ID.
+
+    Raises RuntimeError if any required setting is missing.
+    """
+    secrets = _get_settings().secrets
+    if not secrets.linear_api_key:
+        raise RuntimeError("LINEAR_API_KEY is not configured")
+    if not secrets.linear_feature_request_project_id:
+        raise RuntimeError("LINEAR_FEATURE_REQUEST_PROJECT_ID is not configured")
+    if not secrets.linear_feature_request_team_id:
+        raise RuntimeError("LINEAR_FEATURE_REQUEST_TEAM_ID is not configured")
+
+    credentials = APIKeyCredentials(
+        id="system-linear",
+        provider="linear",
+        api_key=SecretStr(secrets.linear_api_key),
+        title="System Linear API Key",
+    )
+    client = LinearClient(credentials=credentials)
+    return (
+        client,
+        secrets.linear_feature_request_project_id,
+        secrets.linear_feature_request_team_id,
+    )
+
+
+class SearchFeatureRequestsTool(BaseTool):
+    """Tool for searching existing feature requests in Linear."""
+
+    @property
+    def name(self) -> str:
+        return "search_feature_requests"
+
+    @property
+    def description(self) -> str:
+        return (
+            "Search existing feature requests to check if a similar request "
+            "already exists before creating a new one. Returns matching feature "
+            "requests with their ID, title, and description."
+        )
+
+    @property
+    def parameters(self) -> dict[str, Any]:
+        return {
+            "type": "object",
+            "properties": {
+                "query": {
+                    "type": "string",
+                    "description": "Search term to find matching feature requests.",
+                },
+            },
+            "required": ["query"],
+        }
+
+    @property
+    def requires_auth(self) -> bool:
+        return True
+
+    async def _execute(
+        self,
+        user_id: str | None,
+        session: ChatSession,
+        **kwargs,
+    ) -> ToolResponseBase:
+        query = kwargs.get("query", "").strip()
+        session_id = session.session_id if session else None
+
+        if not query:
+            return ErrorResponse(
+                message="Please provide a search query.",
+                error="Missing query parameter",
+                session_id=session_id,
+            )
+
+        try:
+            client, project_id, _team_id = _get_linear_config()
+            data = await client.query(
+                SEARCH_ISSUES_QUERY,
+                {
+                    "term": query,
+                    "filter": {
+                        "project": {"id": {"eq": project_id}},
+                    },
+                    "first": MAX_SEARCH_RESULTS,
+                },
+            )
+
+            nodes = data.get("searchIssues", {}).get("nodes", [])
+
+            if not nodes:
+                return NoResultsResponse(
+                    message=f"No feature requests found matching '{query}'.",
+                    suggestions=[
+                        "Try different keywords",
+                        "Use broader search terms",
+                        "You can create a new feature request if none exists",
+                    ],
+                    session_id=session_id,
+                )
+
+            results = [
+                FeatureRequestInfo(
+                    id=node["id"],
+                    identifier=node["identifier"],
+                    title=node["title"],
+                    description=node.get("description"),
+                )
+                for node in nodes
+            ]
+
+            return FeatureRequestSearchResponse(
+                message=f"Found {len(results)} feature request(s) matching '{query}'.",
+                results=results,
+                count=len(results),
+                query=query,
+                session_id=session_id,
+            )
+        except Exception as e:
+            logger.exception("Failed to search feature requests")
+            return ErrorResponse(
+                message="Failed to search feature requests.",
+                error=str(e),
+                session_id=session_id,
+            )
+
+
+class CreateFeatureRequestTool(BaseTool):
+    """Tool for creating feature requests (or adding needs to existing ones)."""
+
+    @property
+    def name(self) -> str:
+        return "create_feature_request"
+
+    @property
+    def description(self) -> str:
+        return (
+            "Create a new feature request or add a customer need to an existing one. "
+            "Always search first with search_feature_requests to avoid duplicates. "
+            "If a matching request exists, pass its ID as existing_issue_id to add "
+            "the user's need to it instead of creating a duplicate."
+        )
+
+    @property
+    def parameters(self) -> dict[str, Any]:
+        return {
+            "type": "object",
+            "properties": {
+                "title": {
+                    "type": "string",
+                    "description": "Title for the feature request.",
+                },
+                "description": {
+                    "type": "string",
+                    "description": "Detailed description of what the user wants and why.",
+                },
+                "existing_issue_id": {
+                    "type": "string",
+                    "description": (
+                        "If adding a need to an existing feature request, "
+                        "provide its Linear issue ID (from search results). "
+                        "Omit to create a new feature request."
+                    ),
+                },
+            },
+            "required": ["title", "description"],
+        }
+
+    @property
+    def requires_auth(self) -> bool:
+        return True
+
+    async def _find_or_create_customer(
+        self, client: LinearClient, user_id: str, name: str
+    ) -> dict:
+        """Find existing customer by user_id or create a new one via upsert.
+
+        Args:
+            client: Linear API client.
+            user_id: Stable external ID used to deduplicate customers.
+            name: Human-readable display name (e.g. the user's email).
+        """
+        data = await client.mutate(
+            CUSTOMER_UPSERT_MUTATION,
+            {
+                "input": {
+                    "name": name,
+                    "externalId": user_id,
+                },
+            },
+        )
+        result = data.get("customerUpsert", {})
+        if not result.get("success"):
+            raise RuntimeError(f"Failed to upsert customer: {data}")
+        return result["customer"]
+
+    async def _execute(
+        self,
+        user_id: str | None,
+        session: ChatSession,
+        **kwargs,
+    ) -> ToolResponseBase:
+        title = kwargs.get("title", "").strip()
+        description = kwargs.get("description", "").strip()
+        existing_issue_id = kwargs.get("existing_issue_id")
+        session_id = session.session_id if session else None
+
+        if not title or not description:
+            return ErrorResponse(
+                message="Both title and description are required.",
+                error="Missing required parameters",
+                session_id=session_id,
+            )
+
+        if not user_id:
+            return ErrorResponse(
+                message="Authentication required to create feature requests.",
+                error="Missing user_id",
+                session_id=session_id,
+            )
+
+        try:
+            client, project_id, team_id = _get_linear_config()
+        except Exception as e:
+            logger.exception("Failed to initialize Linear client")
+            return ErrorResponse(
+                message="Failed to create feature request.",
+                error=str(e),
+                session_id=session_id,
+            )
+
+        # Resolve a human-readable name (email) for the Linear customer record.
+        # Fall back to user_id if the lookup fails or returns None.
+        try:
+            customer_display_name = (
+                await user_db().get_user_email_by_id(user_id) or user_id
+            )
+        except Exception:
+            customer_display_name = user_id
+
+        # Step 1: Find or create customer for this user
+        try:
+            customer = await self._find_or_create_customer(
+                client, user_id, customer_display_name
+            )
+            customer_id = customer["id"]
+            customer_name = customer["name"]
+        except Exception as e:
+            logger.exception("Failed to upsert customer in Linear")
+            return ErrorResponse(
+                message="Failed to create feature request.",
+                error=str(e),
+                session_id=session_id,
+            )
+
+        # Step 2: Create or reuse issue
+        issue_id: str | None = None
+        issue_identifier: str | None = None
+        if existing_issue_id:
+            # Add need to existing issue - we still need the issue details for response
+            is_new_issue = False
+            issue_id = existing_issue_id
+        else:
+            # Create new issue in the feature requests project
+            try:
+                data = await client.mutate(
+                    ISSUE_CREATE_MUTATION,
+                    {
+                        "input": {
+                            "title": title,
+                            "description": description,
+                            "teamId": team_id,
+                            "projectId": project_id,
+                        },
+                    },
+                )
+                result = data.get("issueCreate", {})
+                if not result.get("success"):
+                    return ErrorResponse(
+                        message="Failed to create feature request issue.",
+                        error=str(data),
+                        session_id=session_id,
+                    )
+                issue = result["issue"]
+                issue_id = issue["id"]
+                issue_identifier = issue.get("identifier")
+            except Exception as e:
+                logger.exception("Failed to create feature request issue")
+                return ErrorResponse(
+                    message="Failed to create feature request.",
+                    error=str(e),
+                    session_id=session_id,
+                )
+            is_new_issue = True
+
+        # Step 3: Create customer need on the issue
+        try:
+            data = await client.mutate(
+                CUSTOMER_NEED_CREATE_MUTATION,
+                {
+                    "input": {
+                        "customerId": customer_id,
+                        "issueId": issue_id,
+                        "body": description,
+                        "priority": 0,
+                    },
+                },
+            )
+            need_result = data.get("customerNeedCreate", {})
+            if not need_result.get("success"):
+                orphaned = (
+                    {"issue_id": issue_id, "issue_identifier": issue_identifier}
+                    if is_new_issue
+                    else None
+                )
+                return ErrorResponse(
+                    message="Failed to attach customer need to the feature request.",
+                    error=str(data),
+                    details=orphaned,
+                    session_id=session_id,
+                )
+            need = need_result["need"]
+            issue_info = need["issue"]
+        except Exception as e:
+            logger.exception("Failed to create customer need")
+            orphaned = (
+                {"issue_id": issue_id, "issue_identifier": issue_identifier}
+                if is_new_issue
+                else None
+            )
+            return ErrorResponse(
+                message="Failed to attach customer need to the feature request.",
+                error=str(e),
+                details=orphaned,
+                session_id=session_id,
+            )
+
+        return FeatureRequestCreatedResponse(
+            message=(
+                f"{'Created new feature request' if is_new_issue else 'Added your request to existing feature request'}: "
+                f"{issue_info['title']}."
+            ),
+            issue_id=issue_info["id"],
+            issue_identifier=issue_info["identifier"],
+            issue_title=issue_info["title"],
+            issue_url=issue_info.get("url", ""),
+            is_new_issue=is_new_issue,
+            customer_name=customer_name,
+            session_id=session_id,
+        )

autogpt_platform/backend/backend/copilot/tools/feature_requests_test.py

@@ -0,0 +1,612 @@
+"""Tests for SearchFeatureRequestsTool and CreateFeatureRequestTool."""
+
+from unittest.mock import AsyncMock, MagicMock, patch
+
+import pytest
+
+from ._test_data import make_session
+from .feature_requests import CreateFeatureRequestTool, SearchFeatureRequestsTool
+from .models import (
+    ErrorResponse,
+    FeatureRequestCreatedResponse,
+    FeatureRequestSearchResponse,
+    NoResultsResponse,
+)
+
+_TEST_USER_ID = "test-user-feature-requests"
+_TEST_USER_EMAIL = "testuser@example.com"
+
+
+# ---------------------------------------------------------------------------
+# Helpers
+# ---------------------------------------------------------------------------
+
+
+_FAKE_PROJECT_ID = "test-project-id"
+_FAKE_TEAM_ID = "test-team-id"
+
+
+def _mock_linear_config(*, query_return=None, mutate_return=None):
+    """Return a patched _get_linear_config that yields a mock LinearClient."""
+    client = AsyncMock()
+    if query_return is not None:
+        client.query.return_value = query_return
+    if mutate_return is not None:
+        client.mutate.return_value = mutate_return
+    return (
+        patch(
+            "backend.copilot.tools.feature_requests._get_linear_config",
+            return_value=(client, _FAKE_PROJECT_ID, _FAKE_TEAM_ID),
+        ),
+        client,
+    )
+
+
+def _search_response(nodes: list[dict]) -> dict:
+    return {"searchIssues": {"nodes": nodes}}
+
+
+def _customer_upsert_response(
+    customer_id: str = "cust-1", name: str = _TEST_USER_EMAIL, success: bool = True
+) -> dict:
+    return {
+        "customerUpsert": {
+            "success": success,
+            "customer": {"id": customer_id, "name": name, "externalIds": [name]},
+        }
+    }
+
+
+def _issue_create_response(
+    issue_id: str = "issue-1",
+    identifier: str = "FR-1",
+    title: str = "New Feature",
+    success: bool = True,
+) -> dict:
+    return {
+        "issueCreate": {
+            "success": success,
+            "issue": {
+                "id": issue_id,
+                "identifier": identifier,
+                "title": title,
+                "url": f"https://linear.app/issue/{identifier}",
+            },
+        }
+    }
+
+
+def _need_create_response(
+    need_id: str = "need-1",
+    issue_id: str = "issue-1",
+    identifier: str = "FR-1",
+    title: str = "New Feature",
+    success: bool = True,
+) -> dict:
+    return {
+        "customerNeedCreate": {
+            "success": success,
+            "need": {
+                "id": need_id,
+                "body": "description",
+                "customer": {"id": "cust-1", "name": _TEST_USER_EMAIL},
+                "issue": {
+                    "id": issue_id,
+                    "identifier": identifier,
+                    "title": title,
+                    "url": f"https://linear.app/issue/{identifier}",
+                },
+            },
+        }
+    }
+
+
+# ===========================================================================
+# SearchFeatureRequestsTool
+# ===========================================================================
+
+
+class TestSearchFeatureRequestsTool:
+    """Tests for SearchFeatureRequestsTool._execute."""
+
+    @pytest.mark.asyncio(loop_scope="session")
+    async def test_successful_search(self):
+        session = make_session(user_id=_TEST_USER_ID)
+        nodes = [
+            {
+                "id": "id-1",
+                "identifier": "FR-1",
+                "title": "Dark mode",
+                "description": "Add dark mode support",
+            },
+            {
+                "id": "id-2",
+                "identifier": "FR-2",
+                "title": "Dark theme",
+                "description": None,
+            },
+        ]
+        patcher, _ = _mock_linear_config(query_return=_search_response(nodes))
+        with patcher:
+            tool = SearchFeatureRequestsTool()
+            resp = await tool._execute(
+                user_id=_TEST_USER_ID, session=session, query="dark mode"
+            )
+
+        assert isinstance(resp, FeatureRequestSearchResponse)
+        assert resp.count == 2
+        assert resp.results[0].id == "id-1"
+        assert resp.results[1].identifier == "FR-2"
+        assert resp.query == "dark mode"
+
+    @pytest.mark.asyncio(loop_scope="session")
+    async def test_no_results(self):
+        session = make_session(user_id=_TEST_USER_ID)
+        patcher, _ = _mock_linear_config(query_return=_search_response([]))
+        with patcher:
+            tool = SearchFeatureRequestsTool()
+            resp = await tool._execute(
+                user_id=_TEST_USER_ID, session=session, query="nonexistent"
+            )
+
+        assert isinstance(resp, NoResultsResponse)
+        assert "nonexistent" in resp.message
+
+    @pytest.mark.asyncio(loop_scope="session")
+    async def test_empty_query_returns_error(self):
+        session = make_session(user_id=_TEST_USER_ID)
+        tool = SearchFeatureRequestsTool()
+        resp = await tool._execute(user_id=_TEST_USER_ID, session=session, query="   ")
+
+        assert isinstance(resp, ErrorResponse)
+        assert resp.error is not None
+        assert "query" in resp.error.lower()
+
+    @pytest.mark.asyncio(loop_scope="session")
+    async def test_missing_query_returns_error(self):
+        session = make_session(user_id=_TEST_USER_ID)
+        tool = SearchFeatureRequestsTool()
+        resp = await tool._execute(user_id=_TEST_USER_ID, session=session)
+
+        assert isinstance(resp, ErrorResponse)
+
+    @pytest.mark.asyncio(loop_scope="session")
+    async def test_api_failure(self):
+        session = make_session(user_id=_TEST_USER_ID)
+        patcher, client = _mock_linear_config()
+        client.query.side_effect = RuntimeError("Linear API down")
+        with patcher:
+            tool = SearchFeatureRequestsTool()
+            resp = await tool._execute(
+                user_id=_TEST_USER_ID, session=session, query="test"
+            )
+
+        assert isinstance(resp, ErrorResponse)
+        assert resp.error is not None
+        assert "Linear API down" in resp.error
+
+    @pytest.mark.asyncio(loop_scope="session")
+    async def test_malformed_node_returns_error(self):
+        """A node missing required keys should be caught by the try/except."""
+        session = make_session(user_id=_TEST_USER_ID)
+        # Node missing 'identifier' key
+        bad_nodes = [{"id": "id-1", "title": "Missing identifier"}]
+        patcher, _ = _mock_linear_config(query_return=_search_response(bad_nodes))
+        with patcher:
+            tool = SearchFeatureRequestsTool()
+            resp = await tool._execute(
+                user_id=_TEST_USER_ID, session=session, query="test"
+            )
+
+        assert isinstance(resp, ErrorResponse)
+
+    @pytest.mark.asyncio(loop_scope="session")
+    async def test_linear_client_init_failure(self):
+        session = make_session(user_id=_TEST_USER_ID)
+        with patch(
+            "backend.copilot.tools.feature_requests._get_linear_config",
+            side_effect=RuntimeError("No API key"),
+        ):
+            tool = SearchFeatureRequestsTool()
+            resp = await tool._execute(
+                user_id=_TEST_USER_ID, session=session, query="test"
+            )
+
+        assert isinstance(resp, ErrorResponse)
+        assert resp.error is not None
+        assert "No API key" in resp.error
+
+
+# ===========================================================================
+# CreateFeatureRequestTool
+# ===========================================================================
+
+
+class TestCreateFeatureRequestTool:
+    """Tests for CreateFeatureRequestTool._execute."""
+
+    @pytest.fixture(autouse=True)
+    def _patch_email_lookup(self):
+        mock_user_db = MagicMock()
+        mock_user_db.get_user_email_by_id = AsyncMock(return_value=_TEST_USER_EMAIL)
+        with patch(
+            "backend.copilot.tools.feature_requests.user_db",
+            return_value=mock_user_db,
+        ):
+            yield
+
+    # ---- Happy paths -------------------------------------------------------
+
+    @pytest.mark.asyncio(loop_scope="session")
+    async def test_create_new_issue(self):
+        """Full happy path: upsert customer -> create issue -> attach need."""
+        session = make_session(user_id=_TEST_USER_ID)
+
+        patcher, client = _mock_linear_config()
+        client.mutate.side_effect = [
+            _customer_upsert_response(),
+            _issue_create_response(),
+            _need_create_response(),
+        ]
+
+        with patcher:
+            tool = CreateFeatureRequestTool()
+            resp = await tool._execute(
+                user_id=_TEST_USER_ID,
+                session=session,
+                title="New Feature",
+                description="Please add this",
+            )
+
+        assert isinstance(resp, FeatureRequestCreatedResponse)
+        assert resp.is_new_issue is True
+        assert resp.issue_identifier == "FR-1"
+        assert resp.customer_name == _TEST_USER_EMAIL
+        assert client.mutate.call_count == 3
+
+    @pytest.mark.asyncio(loop_scope="session")
+    async def test_add_need_to_existing_issue(self):
+        """When existing_issue_id is provided, skip issue creation."""
+        session = make_session(user_id=_TEST_USER_ID)
+
+        patcher, client = _mock_linear_config()
+        client.mutate.side_effect = [
+            _customer_upsert_response(),
+            _need_create_response(issue_id="existing-1", identifier="FR-99"),
+        ]
+
+        with patcher:
+            tool = CreateFeatureRequestTool()
+            resp = await tool._execute(
+                user_id=_TEST_USER_ID,
+                session=session,
+                title="Existing Feature",
+                description="Me too",
+                existing_issue_id="existing-1",
+            )
+
+        assert isinstance(resp, FeatureRequestCreatedResponse)
+        assert resp.is_new_issue is False
+        assert resp.issue_id == "existing-1"
+        # Only 2 mutations: customer upsert + need create (no issue create)
+        assert client.mutate.call_count == 2
+
+    # ---- Validation errors -------------------------------------------------
+
+    @pytest.mark.asyncio(loop_scope="session")
+    async def test_missing_title(self):
+        session = make_session(user_id=_TEST_USER_ID)
+        tool = CreateFeatureRequestTool()
+        resp = await tool._execute(
+            user_id=_TEST_USER_ID,
+            session=session,
+            title="",
+            description="some desc",
+        )
+
+        assert isinstance(resp, ErrorResponse)
+        assert resp.error is not None
+        assert "required" in resp.error.lower()
+
+    @pytest.mark.asyncio(loop_scope="session")
+    async def test_missing_description(self):
+        session = make_session(user_id=_TEST_USER_ID)
+        tool = CreateFeatureRequestTool()
+        resp = await tool._execute(
+            user_id=_TEST_USER_ID,
+            session=session,
+            title="Some title",
+            description="",
+        )
+
+        assert isinstance(resp, ErrorResponse)
+        assert resp.error is not None
+        assert "required" in resp.error.lower()
+
+    @pytest.mark.asyncio(loop_scope="session")
+    async def test_missing_user_id(self):
+        session = make_session(user_id=_TEST_USER_ID)
+        tool = CreateFeatureRequestTool()
+        resp = await tool._execute(
+            user_id=None,
+            session=session,
+            title="Some title",
+            description="Some desc",
+        )
+
+        assert isinstance(resp, ErrorResponse)
+        assert resp.error is not None
+        assert "user_id" in resp.error.lower()
+
+    # ---- Linear client init failure ----------------------------------------
+
+    @pytest.mark.asyncio(loop_scope="session")
+    async def test_linear_client_init_failure(self):
+        session = make_session(user_id=_TEST_USER_ID)
+        with patch(
+            "backend.copilot.tools.feature_requests._get_linear_config",
+            side_effect=RuntimeError("No API key"),
+        ):
+            tool = CreateFeatureRequestTool()
+            resp = await tool._execute(
+                user_id=_TEST_USER_ID,
+                session=session,
+                title="Title",
+                description="Desc",
+            )
+
+        assert isinstance(resp, ErrorResponse)
+        assert resp.error is not None
+        assert "No API key" in resp.error
+
+    # ---- Customer upsert failures ------------------------------------------
+
+    @pytest.mark.asyncio(loop_scope="session")
+    async def test_customer_upsert_api_error(self):
+        session = make_session(user_id=_TEST_USER_ID)
+        patcher, client = _mock_linear_config()
+        client.mutate.side_effect = RuntimeError("Customer API error")
+
+        with patcher:
+            tool = CreateFeatureRequestTool()
+            resp = await tool._execute(
+                user_id=_TEST_USER_ID,
+                session=session,
+                title="Title",
+                description="Desc",
+            )
+
+        assert isinstance(resp, ErrorResponse)
+        assert resp.error is not None
+        assert "Customer API error" in resp.error
+
+    @pytest.mark.asyncio(loop_scope="session")
+    async def test_customer_upsert_not_success(self):
+        session = make_session(user_id=_TEST_USER_ID)
+        patcher, client = _mock_linear_config()
+        client.mutate.return_value = _customer_upsert_response(success=False)
+
+        with patcher:
+            tool = CreateFeatureRequestTool()
+            resp = await tool._execute(
+                user_id=_TEST_USER_ID,
+                session=session,
+                title="Title",
+                description="Desc",
+            )
+
+        assert isinstance(resp, ErrorResponse)
+
+    @pytest.mark.asyncio(loop_scope="session")
+    async def test_customer_malformed_response(self):
+        """Customer dict missing 'id' key should be caught."""
+        session = make_session(user_id=_TEST_USER_ID)
+        patcher, client = _mock_linear_config()
+        # success=True but customer has no 'id'
+        client.mutate.return_value = {
+            "customerUpsert": {
+                "success": True,
+                "customer": {"name": _TEST_USER_ID},
+            }
+        }
+
+        with patcher:
+            tool = CreateFeatureRequestTool()
+            resp = await tool._execute(
+                user_id=_TEST_USER_ID,
+                session=session,
+                title="Title",
+                description="Desc",
+            )
+
+        assert isinstance(resp, ErrorResponse)
+
+    # ---- Issue creation failures -------------------------------------------
+
+    @pytest.mark.asyncio(loop_scope="session")
+    async def test_issue_create_api_error(self):
+        session = make_session(user_id=_TEST_USER_ID)
+        patcher, client = _mock_linear_config()
+        client.mutate.side_effect = [
+            _customer_upsert_response(),
+            RuntimeError("Issue create failed"),
+        ]
+
+        with patcher:
+            tool = CreateFeatureRequestTool()
+            resp = await tool._execute(
+                user_id=_TEST_USER_ID,
+                session=session,
+                title="Title",
+                description="Desc",
+            )
+
+        assert isinstance(resp, ErrorResponse)
+        assert resp.error is not None
+        assert "Issue create failed" in resp.error
+
+    @pytest.mark.asyncio(loop_scope="session")
+    async def test_issue_create_not_success(self):
+        session = make_session(user_id=_TEST_USER_ID)
+        patcher, client = _mock_linear_config()
+        client.mutate.side_effect = [
+            _customer_upsert_response(),
+            _issue_create_response(success=False),
+        ]
+
+        with patcher:
+            tool = CreateFeatureRequestTool()
+            resp = await tool._execute(
+                user_id=_TEST_USER_ID,
+                session=session,
+                title="Title",
+                description="Desc",
+            )
+
+        assert isinstance(resp, ErrorResponse)
+        assert "Failed to create feature request issue" in resp.message
+
+    @pytest.mark.asyncio(loop_scope="session")
+    async def test_issue_create_malformed_response(self):
+        """issueCreate success=True but missing 'issue' key."""
+        session = make_session(user_id=_TEST_USER_ID)
+        patcher, client = _mock_linear_config()
+        client.mutate.side_effect = [
+            _customer_upsert_response(),
+            {"issueCreate": {"success": True}},  # no 'issue' key
+        ]
+
+        with patcher:
+            tool = CreateFeatureRequestTool()
+            resp = await tool._execute(
+                user_id=_TEST_USER_ID,
+                session=session,
+                title="Title",
+                description="Desc",
+            )
+
+        assert isinstance(resp, ErrorResponse)
+
+    # ---- Customer need attachment failures ---------------------------------
+
+    @pytest.mark.asyncio(loop_scope="session")
+    async def test_need_create_api_error_new_issue(self):
+        """Need creation fails after new issue was created -> orphaned issue info."""
+        session = make_session(user_id=_TEST_USER_ID)
+        patcher, client = _mock_linear_config()
+        client.mutate.side_effect = [
+            _customer_upsert_response(),
+            _issue_create_response(issue_id="orphan-1", identifier="FR-10"),
+            RuntimeError("Need attach failed"),
+        ]
+
+        with patcher:
+            tool = CreateFeatureRequestTool()
+            resp = await tool._execute(
+                user_id=_TEST_USER_ID,
+                session=session,
+                title="Title",
+                description="Desc",
+            )
+
+        assert isinstance(resp, ErrorResponse)
+        assert resp.error is not None
+        assert "Need attach failed" in resp.error
+        assert resp.details is not None
+        assert resp.details["issue_id"] == "orphan-1"
+        assert resp.details["issue_identifier"] == "FR-10"
+
+    @pytest.mark.asyncio(loop_scope="session")
+    async def test_need_create_api_error_existing_issue(self):
+        """Need creation fails on existing issue -> no orphaned info."""
+        session = make_session(user_id=_TEST_USER_ID)
+        patcher, client = _mock_linear_config()
+        client.mutate.side_effect = [
+            _customer_upsert_response(),
+            RuntimeError("Need attach failed"),
+        ]
+
+        with patcher:
+            tool = CreateFeatureRequestTool()
+            resp = await tool._execute(
+                user_id=_TEST_USER_ID,
+                session=session,
+                title="Title",
+                description="Desc",
+                existing_issue_id="existing-1",
+            )
+
+        assert isinstance(resp, ErrorResponse)
+        assert resp.details is None
+
+    @pytest.mark.asyncio(loop_scope="session")
+    async def test_need_create_not_success_includes_orphaned_info(self):
+        """customerNeedCreate returns success=False -> includes orphaned issue."""
+        session = make_session(user_id=_TEST_USER_ID)
+        patcher, client = _mock_linear_config()
+        client.mutate.side_effect = [
+            _customer_upsert_response(),
+            _issue_create_response(issue_id="orphan-2", identifier="FR-20"),
+            _need_create_response(success=False),
+        ]
+
+        with patcher:
+            tool = CreateFeatureRequestTool()
+            resp = await tool._execute(
+                user_id=_TEST_USER_ID,
+                session=session,
+                title="Title",
+                description="Desc",
+            )
+
+        assert isinstance(resp, ErrorResponse)
+        assert resp.details is not None
+        assert resp.details["issue_id"] == "orphan-2"
+        assert resp.details["issue_identifier"] == "FR-20"
+
+    @pytest.mark.asyncio(loop_scope="session")
+    async def test_need_create_not_success_existing_issue_no_details(self):
+        """customerNeedCreate fails on existing issue -> no orphaned info."""
+        session = make_session(user_id=_TEST_USER_ID)
+        patcher, client = _mock_linear_config()
+        client.mutate.side_effect = [
+            _customer_upsert_response(),
+            _need_create_response(success=False),
+        ]
+
+        with patcher:
+            tool = CreateFeatureRequestTool()
+            resp = await tool._execute(
+                user_id=_TEST_USER_ID,
+                session=session,
+                title="Title",
+                description="Desc",
+                existing_issue_id="existing-1",
+            )
+
+        assert isinstance(resp, ErrorResponse)
+        assert resp.details is None
+
+    @pytest.mark.asyncio(loop_scope="session")
+    async def test_need_create_malformed_response(self):
+        """need_result missing 'need' key after success=True."""
+        session = make_session(user_id=_TEST_USER_ID)
+        patcher, client = _mock_linear_config()
+        client.mutate.side_effect = [
+            _customer_upsert_response(),
+            _issue_create_response(),
+            {"customerNeedCreate": {"success": True}},  # no 'need' key
+        ]
+
+        with patcher:
+            tool = CreateFeatureRequestTool()
+            resp = await tool._execute(
+                user_id=_TEST_USER_ID,
+                session=session,
+                title="Title",
+                description="Desc",
+            )
+
+        assert isinstance(resp, ErrorResponse)
+        assert resp.details is not None
+        assert resp.details["issue_id"] == "issue-1"

autogpt_platform/backend/backend/copilot/tools/find_agent.py

@@ -2,7 +2,7 @@
 
 from typing import Any
 
-from backend.api.features.chat.model import ChatSession
+from backend.copilot.model import ChatSession
 
 from .agent_search import search_agents
 from .base import BaseTool

autogpt_platform/backend/backend/copilot/tools/find_block.py

@@ -3,18 +3,18 @@ from typing import Any
 
 from prisma.enums import ContentType
 
-from backend.api.features.chat.model import ChatSession
-from backend.api.features.chat.tools.base import BaseTool, ToolResponseBase
-from backend.api.features.chat.tools.models import (
+from backend.blocks import get_block
+from backend.blocks._base import BlockType
+from backend.copilot.model import ChatSession
+from backend.data.db_accessors import search
+
+from .base import BaseTool, ToolResponseBase
+from .models import (
     BlockInfoSummary,
-    BlockInputFieldInfo,
     BlockListResponse,
     ErrorResponse,
     NoResultsResponse,
 )
-from backend.api.features.store.hybrid_search import unified_hybrid_search
-from backend.blocks import get_block
-from backend.blocks._base import BlockType
 
 logger = logging.getLogger(__name__)
 
@@ -55,7 +55,8 @@ class FindBlockTool(BaseTool):
             "Blocks are reusable components that perform specific tasks like "
             "sending emails, making API calls, processing text, etc. "
             "IMPORTANT: Use this tool FIRST to get the block's 'id' before calling run_block. "
-            "The response includes each block's id, required_inputs, and input_schema."
+            "The response includes each block's id, name, and description. "
+            "Call run_block with the block's id **with no inputs** to see detailed inputs/outputs and execute it."
         )
 
     @property
@@ -107,7 +108,7 @@ class FindBlockTool(BaseTool):
 
         try:
             # Search for blocks using hybrid search
-            results, total = await unified_hybrid_search(
+            results, total = await search().unified_hybrid_search(
                 query=query,
                 content_types=[ContentType.BLOCK],
                 page=1,
@@ -124,7 +125,7 @@ class FindBlockTool(BaseTool):
                     session_id=session_id,
                 )
 
-            # Enrich results with full block information
+            # Enrich results with block information
             blocks: list[BlockInfoSummary] = []
             for result in results:
                 block_id = result["content_id"]
@@ -141,65 +142,12 @@ class FindBlockTool(BaseTool):
                 ):
                     continue
 
-                # Get input/output schemas
-                input_schema = {}
-                output_schema = {}
-                try:
-                    input_schema = block.input_schema.jsonschema()
-                except Exception as e:
-                    logger.debug(
-                        "Failed to generate input schema for block %s: %s",
-                        block_id,
-                        e,
-                    )
-                try:
-                    output_schema = block.output_schema.jsonschema()
-                except Exception as e:
-                    logger.debug(
-                        "Failed to generate output schema for block %s: %s",
-                        block_id,
-                        e,
-                    )
-
-                # Get categories from block instance
-                categories = []
-                if hasattr(block, "categories") and block.categories:
-                    categories = [cat.value for cat in block.categories]
-
-                # Extract required inputs for easier use
-                required_inputs: list[BlockInputFieldInfo] = []
-                if input_schema:
-                    properties = input_schema.get("properties", {})
-                    required_fields = set(input_schema.get("required", []))
-                    # Get credential field names to exclude from required inputs
-                    credentials_fields = set(
-                        block.input_schema.get_credentials_fields().keys()
-                    )
-
-                    for field_name, field_schema in properties.items():
-                        # Skip credential fields - they're handled separately
-                        if field_name in credentials_fields:
-                            continue
-
-                        required_inputs.append(
-                            BlockInputFieldInfo(
-                                name=field_name,
-                                type=field_schema.get("type", "string"),
-                                description=field_schema.get("description", ""),
-                                required=field_name in required_fields,
-                                default=field_schema.get("default"),
-                            )
-                        )
-
                 blocks.append(
                     BlockInfoSummary(
                         id=block_id,
                         name=block.name,
                         description=block.description or "",
-                        categories=categories,
-                        input_schema=input_schema,
-                        output_schema=output_schema,
-                        required_inputs=required_inputs,
+                        categories=[c.value for c in block.categories],
                     )
                 )
 
@@ -228,8 +176,7 @@ class FindBlockTool(BaseTool):
             return BlockListResponse(
                 message=(
                     f"Found {len(blocks)} block(s) matching '{query}'. "
-                    "To execute a block, use run_block with the block's 'id' field "
-                    "and provide 'input_data' matching the block's input_schema."
+                    "To see a block's inputs/outputs and execute it, use run_block with the block's 'id' - providing no inputs."
                 ),
                 blocks=blocks,
                 count=len(blocks),

autogpt_platform/backend/backend/copilot/tools/find_block_test.py

@@ -0,0 +1,398 @@
+"""Tests for block filtering in FindBlockTool."""
+
+from unittest.mock import AsyncMock, MagicMock, patch
+
+import pytest
+
+from backend.blocks._base import BlockType
+
+from ._test_data import make_session
+from .find_block import (
+    COPILOT_EXCLUDED_BLOCK_IDS,
+    COPILOT_EXCLUDED_BLOCK_TYPES,
+    FindBlockTool,
+)
+from .models import BlockListResponse
+
+_TEST_USER_ID = "test-user-find-block"
+
+
+def make_mock_block(
+    block_id: str,
+    name: str,
+    block_type: BlockType,
+    disabled: bool = False,
+    input_schema: dict | None = None,
+    output_schema: dict | None = None,
+    credentials_fields: dict | None = None,
+):
+    """Create a mock block for testing."""
+    mock = MagicMock()
+    mock.id = block_id
+    mock.name = name
+    mock.description = f"{name} description"
+    mock.block_type = block_type
+    mock.disabled = disabled
+    mock.input_schema = MagicMock()
+    mock.input_schema.jsonschema.return_value = input_schema or {
+        "properties": {},
+        "required": [],
+    }
+    mock.input_schema.get_credentials_fields.return_value = credentials_fields or {}
+    mock.output_schema = MagicMock()
+    mock.output_schema.jsonschema.return_value = output_schema or {}
+    mock.categories = []
+    return mock
+
+
+class TestFindBlockFiltering:
+    """Tests for block filtering in FindBlockTool."""
+
+    def test_excluded_block_types_contains_expected_types(self):
+        """Verify COPILOT_EXCLUDED_BLOCK_TYPES contains all graph-only types."""
+        assert BlockType.INPUT in COPILOT_EXCLUDED_BLOCK_TYPES
+        assert BlockType.OUTPUT in COPILOT_EXCLUDED_BLOCK_TYPES
+        assert BlockType.WEBHOOK in COPILOT_EXCLUDED_BLOCK_TYPES
+        assert BlockType.WEBHOOK_MANUAL in COPILOT_EXCLUDED_BLOCK_TYPES
+        assert BlockType.NOTE in COPILOT_EXCLUDED_BLOCK_TYPES
+        assert BlockType.HUMAN_IN_THE_LOOP in COPILOT_EXCLUDED_BLOCK_TYPES
+        assert BlockType.AGENT in COPILOT_EXCLUDED_BLOCK_TYPES
+
+    def test_excluded_block_ids_contains_smart_decision_maker(self):
+        """Verify SmartDecisionMakerBlock is in COPILOT_EXCLUDED_BLOCK_IDS."""
+        assert "3b191d9f-356f-482d-8238-ba04b6d18381" in COPILOT_EXCLUDED_BLOCK_IDS
+
+    @pytest.mark.asyncio(loop_scope="session")
+    async def test_excluded_block_type_filtered_from_results(self):
+        """Verify blocks with excluded BlockTypes are filtered from search results."""
+        session = make_session(user_id=_TEST_USER_ID)
+
+        # Mock search returns an INPUT block (excluded) and a STANDARD block (included)
+        search_results = [
+            {"content_id": "input-block-id", "score": 0.9},
+            {"content_id": "standard-block-id", "score": 0.8},
+        ]
+
+        input_block = make_mock_block("input-block-id", "Input Block", BlockType.INPUT)
+        standard_block = make_mock_block(
+            "standard-block-id", "HTTP Request", BlockType.STANDARD
+        )
+
+        def mock_get_block(block_id):
+            return {
+                "input-block-id": input_block,
+                "standard-block-id": standard_block,
+            }.get(block_id)
+
+        mock_search_db = MagicMock()
+        mock_search_db.unified_hybrid_search = AsyncMock(
+            return_value=(search_results, 2)
+        )
+
+        with patch(
+            "backend.copilot.tools.find_block.search",
+            return_value=mock_search_db,
+        ):
+            with patch(
+                "backend.copilot.tools.find_block.get_block",
+                side_effect=mock_get_block,
+            ):
+                tool = FindBlockTool()
+                response = await tool._execute(
+                    user_id=_TEST_USER_ID, session=session, query="test"
+                )
+
+        # Should only return the standard block, not the INPUT block
+        assert isinstance(response, BlockListResponse)
+        assert len(response.blocks) == 1
+        assert response.blocks[0].id == "standard-block-id"
+
+    @pytest.mark.asyncio(loop_scope="session")
+    async def test_excluded_block_id_filtered_from_results(self):
+        """Verify SmartDecisionMakerBlock is filtered from search results."""
+        session = make_session(user_id=_TEST_USER_ID)
+
+        smart_decision_id = "3b191d9f-356f-482d-8238-ba04b6d18381"
+        search_results = [
+            {"content_id": smart_decision_id, "score": 0.9},
+            {"content_id": "normal-block-id", "score": 0.8},
+        ]
+
+        # SmartDecisionMakerBlock has STANDARD type but is excluded by ID
+        smart_block = make_mock_block(
+            smart_decision_id, "Smart Decision Maker", BlockType.STANDARD
+        )
+        normal_block = make_mock_block(
+            "normal-block-id", "Normal Block", BlockType.STANDARD
+        )
+
+        def mock_get_block(block_id):
+            return {
+                smart_decision_id: smart_block,
+                "normal-block-id": normal_block,
+            }.get(block_id)
+
+        mock_search_db = MagicMock()
+        mock_search_db.unified_hybrid_search = AsyncMock(
+            return_value=(search_results, 2)
+        )
+
+        with patch(
+            "backend.copilot.tools.find_block.search",
+            return_value=mock_search_db,
+        ):
+            with patch(
+                "backend.copilot.tools.find_block.get_block",
+                side_effect=mock_get_block,
+            ):
+                tool = FindBlockTool()
+                response = await tool._execute(
+                    user_id=_TEST_USER_ID, session=session, query="decision"
+                )
+
+        # Should only return normal block, not SmartDecisionMakerBlock
+        assert isinstance(response, BlockListResponse)
+        assert len(response.blocks) == 1
+        assert response.blocks[0].id == "normal-block-id"
+
+    @pytest.mark.asyncio(loop_scope="session")
+    async def test_response_size_average_chars_per_block(self):
+        """Measure average chars per block in the serialized response."""
+        session = make_session(user_id=_TEST_USER_ID)
+
+        # Realistic block definitions modeled after real blocks
+        block_defs = [
+            {
+                "id": "http-block-id",
+                "name": "Send Web Request",
+                "input_schema": {
+                    "properties": {
+                        "url": {
+                            "type": "string",
+                            "description": "The URL to send the request to",
+                        },
+                        "method": {
+                            "type": "string",
+                            "description": "The HTTP method to use",
+                        },
+                        "headers": {
+                            "type": "object",
+                            "description": "Headers to include in the request",
+                        },
+                        "json_format": {
+                            "type": "boolean",
+                            "description": "If true, send the body as JSON",
+                        },
+                        "body": {
+                            "type": "object",
+                            "description": "Form/JSON body payload",
+                        },
+                        "credentials": {
+                            "type": "object",
+                            "description": "HTTP credentials",
+                        },
+                    },
+                    "required": ["url", "method"],
+                },
+                "output_schema": {
+                    "properties": {
+                        "response": {
+                            "type": "object",
+                            "description": "The response from the server",
+                        },
+                        "client_error": {
+                            "type": "object",
+                            "description": "Errors on 4xx status codes",
+                        },
+                        "server_error": {
+                            "type": "object",
+                            "description": "Errors on 5xx status codes",
+                        },
+                        "error": {
+                            "type": "string",
+                            "description": "Errors for all other exceptions",
+                        },
+                    },
+                },
+                "credentials_fields": {"credentials": True},
+            },
+            {
+                "id": "email-block-id",
+                "name": "Send Email",
+                "input_schema": {
+                    "properties": {
+                        "to_email": {
+                            "type": "string",
+                            "description": "Recipient email address",
+                        },
+                        "subject": {
+                            "type": "string",
+                            "description": "Subject of the email",
+                        },
+                        "body": {
+                            "type": "string",
+                            "description": "Body of the email",
+                        },
+                        "config": {
+                            "type": "object",
+                            "description": "SMTP Config",
+                        },
+                        "credentials": {
+                            "type": "object",
+                            "description": "SMTP credentials",
+                        },
+                    },
+                    "required": ["to_email", "subject", "body", "credentials"],
+                },
+                "output_schema": {
+                    "properties": {
+                        "status": {
+                            "type": "string",
+                            "description": "Status of the email sending operation",
+                        },
+                        "error": {
+                            "type": "string",
+                            "description": "Error message if sending failed",
+                        },
+                    },
+                },
+                "credentials_fields": {"credentials": True},
+            },
+            {
+                "id": "claude-code-block-id",
+                "name": "Claude Code",
+                "input_schema": {
+                    "properties": {
+                        "e2b_credentials": {
+                            "type": "object",
+                            "description": "API key for E2B platform",
+                        },
+                        "anthropic_credentials": {
+                            "type": "object",
+                            "description": "API key for Anthropic",
+                        },
+                        "prompt": {
+                            "type": "string",
+                            "description": "Task or instruction for Claude Code",
+                        },
+                        "timeout": {
+                            "type": "integer",
+                            "description": "Sandbox timeout in seconds",
+                        },
+                        "setup_commands": {
+                            "type": "array",
+                            "description": "Shell commands to run before execution",
+                        },
+                        "working_directory": {
+                            "type": "string",
+                            "description": "Working directory for Claude Code",
+                        },
+                        "session_id": {
+                            "type": "string",
+                            "description": "Session ID to resume a conversation",
+                        },
+                        "sandbox_id": {
+                            "type": "string",
+                            "description": "Sandbox ID to reconnect to",
+                        },
+                        "conversation_history": {
+                            "type": "string",
+                            "description": "Previous conversation history",
+                        },
+                        "dispose_sandbox": {
+                            "type": "boolean",
+                            "description": "Whether to dispose sandbox after execution",
+                        },
+                    },
+                    "required": [
+                        "e2b_credentials",
+                        "anthropic_credentials",
+                        "prompt",
+                    ],
+                },
+                "output_schema": {
+                    "properties": {
+                        "response": {
+                            "type": "string",
+                            "description": "Output from Claude Code execution",
+                        },
+                        "files": {
+                            "type": "array",
+                            "description": "Files created/modified by Claude Code",
+                        },
+                        "conversation_history": {
+                            "type": "string",
+                            "description": "Full conversation history",
+                        },
+                        "session_id": {
+                            "type": "string",
+                            "description": "Session ID for this conversation",
+                        },
+                        "sandbox_id": {
+                            "type": "string",
+                            "description": "ID of the sandbox instance",
+                        },
+                        "error": {
+                            "type": "string",
+                            "description": "Error message if execution failed",
+                        },
+                    },
+                },
+                "credentials_fields": {
+                    "e2b_credentials": True,
+                    "anthropic_credentials": True,
+                },
+            },
+        ]
+
+        search_results = [
+            {"content_id": d["id"], "score": 0.9 - i * 0.1}
+            for i, d in enumerate(block_defs)
+        ]
+        mock_blocks = {
+            d["id"]: make_mock_block(
+                block_id=d["id"],
+                name=d["name"],
+                block_type=BlockType.STANDARD,
+                input_schema=d["input_schema"],
+                output_schema=d["output_schema"],
+                credentials_fields=d["credentials_fields"],
+            )
+            for d in block_defs
+        }
+
+        mock_search_db = MagicMock()
+        mock_search_db.unified_hybrid_search = AsyncMock(
+            return_value=(search_results, len(search_results))
+        )
+
+        with patch(
+            "backend.copilot.tools.find_block.search",
+            return_value=mock_search_db,
+        ), patch(
+            "backend.copilot.tools.find_block.get_block",
+            side_effect=lambda bid: mock_blocks.get(bid),
+        ):
+            tool = FindBlockTool()
+            response = await tool._execute(
+                user_id=_TEST_USER_ID, session=session, query="test"
+            )
+
+        assert isinstance(response, BlockListResponse)
+        assert response.count == len(block_defs)
+
+        total_chars = len(response.model_dump_json())
+        avg_chars = total_chars // response.count
+
+        # Print for visibility in test output
+        print(f"\nTotal response size: {total_chars} chars")
+        print(f"Number of blocks: {response.count}")
+        print(f"Average chars per block: {avg_chars}")
+
+        # The old response was ~90K for 10 blocks (~9K per block).
+        # Previous optimization reduced it to ~1.5K per block (no raw JSON schemas).
+        # Now with only id/name/description, we expect ~300 chars per block.
+        assert avg_chars < 500, (
+            f"Average chars per block ({avg_chars}) exceeds 500. "
+            f"Total response: {total_chars} chars for {response.count} blocks."
+        )

autogpt_platform/backend/backend/copilot/tools/find_library_agent.py

@@ -2,7 +2,7 @@
 
 from typing import Any
 
-from backend.api.features.chat.model import ChatSession
+from backend.copilot.model import ChatSession
 
 from .agent_search import search_agents
 from .base import BaseTool

autogpt_platform/backend/backend/copilot/tools/get_doc_page.py

@@ -4,13 +4,10 @@ import logging
 from pathlib import Path
 from typing import Any
 
-from backend.api.features.chat.model import ChatSession
-from backend.api.features.chat.tools.base import BaseTool
-from backend.api.features.chat.tools.models import (
-    DocPageResponse,
-    ErrorResponse,
-    ToolResponseBase,
-)
+from backend.copilot.model import ChatSession
+
+from .base import BaseTool
+from .models import DocPageResponse, ErrorResponse, ToolResponseBase
 
 logger = logging.getLogger(__name__)
 

autogpt_platform/backend/backend/copilot/tools/models.py

@@ -25,6 +25,7 @@ class ResponseType(str, Enum):
     AGENT_SAVED = "agent_saved"
     CLARIFICATION_NEEDED = "clarification_needed"
     BLOCK_LIST = "block_list"
+    BLOCK_DETAILS = "block_details"
     BLOCK_OUTPUT = "block_output"
     DOC_SEARCH_RESULTS = "doc_search_results"
     DOC_PAGE = "doc_page"
@@ -40,6 +41,15 @@ class ResponseType(str, Enum):
     OPERATION_IN_PROGRESS = "operation_in_progress"
     # Input validation
     INPUT_VALIDATION_ERROR = "input_validation_error"
+    # Web fetch
+    WEB_FETCH = "web_fetch"
+    # Code execution
+    BASH_EXEC = "bash_exec"
+    # Operation status check
+    OPERATION_STATUS = "operation_status"
+    # Feature request types
+    FEATURE_REQUEST_SEARCH = "feature_request_search"
+    FEATURE_REQUEST_CREATED = "feature_request_created"
 
 
 # Base response model
@@ -335,11 +345,17 @@ class BlockInfoSummary(BaseModel):
     name: str
     description: str
     categories: list[str]
-    input_schema: dict[str, Any]
-    output_schema: dict[str, Any]
+    input_schema: dict[str, Any] = Field(
+        default_factory=dict,
+        description="Full JSON schema for block inputs",
+    )
+    output_schema: dict[str, Any] = Field(
+        default_factory=dict,
+        description="Full JSON schema for block outputs",
+    )
     required_inputs: list[BlockInputFieldInfo] = Field(
         default_factory=list,
-        description="List of required input fields for this block",
+        description="List of input fields for this block",
     )
 
 
@@ -352,10 +368,29 @@ class BlockListResponse(ToolResponseBase):
     query: str
     usage_hint: str = Field(
         default="To execute a block, call run_block with block_id set to the block's "
-        "'id' field and input_data containing the required fields from input_schema."
+        "'id' field and input_data containing the fields listed in required_inputs."
     )
 
 
+class BlockDetails(BaseModel):
+    """Detailed block information."""
+
+    id: str
+    name: str
+    description: str
+    inputs: dict[str, Any] = {}
+    outputs: dict[str, Any] = {}
+    credentials: list[CredentialsMetaInput] = []
+
+
+class BlockDetailsResponse(ToolResponseBase):
+    """Response for block details (first run_block attempt)."""
+
+    type: ResponseType = ResponseType.BLOCK_DETAILS
+    block: BlockDetails
+    user_authenticated: bool = False
+
+
 class BlockOutputResponse(ToolResponseBase):
     """Response for run_block tool."""
 
@@ -421,3 +456,55 @@ class AsyncProcessingResponse(ToolResponseBase):
     status: str = "accepted"  # Must be "accepted" for detection
     operation_id: str | None = None
     task_id: str | None = None
+
+
+class WebFetchResponse(ToolResponseBase):
+    """Response for web_fetch tool."""
+
+    type: ResponseType = ResponseType.WEB_FETCH
+    url: str
+    status_code: int
+    content_type: str
+    content: str
+    truncated: bool = False
+
+
+class BashExecResponse(ToolResponseBase):
+    """Response for bash_exec tool."""
+
+    type: ResponseType = ResponseType.BASH_EXEC
+    stdout: str
+    stderr: str
+    exit_code: int
+    timed_out: bool = False
+
+
+# Feature request models
+class FeatureRequestInfo(BaseModel):
+    """Information about a feature request issue."""
+
+    id: str
+    identifier: str
+    title: str
+    description: str | None = None
+
+
+class FeatureRequestSearchResponse(ToolResponseBase):
+    """Response for search_feature_requests tool."""
+
+    type: ResponseType = ResponseType.FEATURE_REQUEST_SEARCH
+    results: list[FeatureRequestInfo]
+    count: int
+    query: str
+
+
+class FeatureRequestCreatedResponse(ToolResponseBase):
+    """Response for create_feature_request tool."""
+
+    type: ResponseType = ResponseType.FEATURE_REQUEST_CREATED
+    issue_id: str
+    issue_identifier: str
+    issue_title: str
+    issue_url: str
+    is_new_issue: bool  # False if added to existing
+    customer_name: str

autogpt_platform/backend/backend/copilot/tools/run_agent.py

@@ -5,16 +5,12 @@ from typing import Any
 
 from pydantic import BaseModel, Field, field_validator
 
-from backend.api.features.chat.config import ChatConfig
-from backend.api.features.chat.model import ChatSession
-from backend.api.features.chat.tracking import (
-    track_agent_run_success,
-    track_agent_scheduled,
-)
-from backend.api.features.library import db as library_db
+from backend.copilot.config import ChatConfig
+from backend.copilot.model import ChatSession
+from backend.copilot.tracking import track_agent_run_success, track_agent_scheduled
+from backend.data.db_accessors import graph_db, library_db, user_db
 from backend.data.graph import GraphModel
 from backend.data.model import CredentialsMetaInput
-from backend.data.user import get_user_by_id
 from backend.executor import utils as execution_utils
 from backend.util.clients import get_scheduler_client
 from backend.util.exceptions import DatabaseError, NotFoundError
@@ -200,7 +196,7 @@ class RunAgentTool(BaseTool):
 
             # Priority: library_agent_id if provided
             if has_library_id:
-                library_agent = await library_db.get_library_agent(
+                library_agent = await library_db().get_library_agent(
                     params.library_agent_id, user_id
                 )
                 if not library_agent:
@@ -209,9 +205,7 @@ class RunAgentTool(BaseTool):
                         session_id=session_id,
                     )
                 # Get the graph from the library agent
-                from backend.data.graph import get_graph
-
-                graph = await get_graph(
+                graph = await graph_db().get_graph(
                     library_agent.graph_id,
                     library_agent.graph_version,
                     user_id=user_id,
@@ -522,7 +516,7 @@ class RunAgentTool(BaseTool):
         library_agent = await get_or_create_library_agent(graph, user_id)
 
         # Get user timezone
-        user = await get_user_by_id(user_id)
+        user = await user_db().get_user_by_id(user_id)
         user_timezone = get_user_timezone_or_utc(user.timezone if user else timezone)
 
         # Create schedule

autogpt_platform/backend/backend/copilot/tools/run_block.py

@@ -7,24 +7,24 @@ from typing import Any
 
 from pydantic_core import PydanticUndefined
 
-from backend.api.features.chat.model import ChatSession
-from backend.api.features.chat.tools.find_block import (
-    COPILOT_EXCLUDED_BLOCK_IDS,
-    COPILOT_EXCLUDED_BLOCK_TYPES,
-)
 from backend.blocks import get_block
 from backend.blocks._base import AnyBlockSchema
+from backend.copilot.model import ChatSession
+from backend.data.db_accessors import workspace_db
 from backend.data.execution import ExecutionContext
 from backend.data.model import CredentialsFieldInfo, CredentialsMetaInput
-from backend.data.workspace import get_or_create_workspace
 from backend.integrations.creds_manager import IntegrationCredentialsManager
 from backend.util.exceptions import BlockError
 
 from .base import BaseTool
+from .find_block import COPILOT_EXCLUDED_BLOCK_IDS, COPILOT_EXCLUDED_BLOCK_TYPES
 from .helpers import get_inputs_from_schema
 from .models import (
+    BlockDetails,
+    BlockDetailsResponse,
     BlockOutputResponse,
     ErrorResponse,
+    InputValidationErrorResponse,
     SetupInfo,
     SetupRequirementsResponse,
     ToolResponseBase,
@@ -51,8 +51,8 @@ class RunBlockTool(BaseTool):
             "Execute a specific block with the provided input data. "
             "IMPORTANT: You MUST call find_block first to get the block's 'id' - "
             "do NOT guess or make up block IDs. "
-            "Use the 'id' from find_block results and provide input_data "
-            "matching the block's required_inputs."
+            "On first attempt (without input_data), returns detailed schema showing "
+            "required inputs and outputs. Then call again with proper input_data to execute."
         )
 
     @property
@@ -67,11 +67,19 @@ class RunBlockTool(BaseTool):
                         "NEVER guess this - always get it from find_block first."
                     ),
                 },
+                "block_name": {
+                    "type": "string",
+                    "description": (
+                        "The block's human-readable name from find_block results. "
+                        "Used for display purposes in the UI."
+                    ),
+                },
                 "input_data": {
                     "type": "object",
                     "description": (
-                        "Input values for the block. Use the 'required_inputs' field "
-                        "from find_block to see what fields are needed."
+                        "Input values for the block. "
+                        "First call with empty {} to see the block's schema, "
+                        "then call again with proper values to execute."
                     ),
                 },
             },
@@ -156,6 +164,34 @@ class RunBlockTool(BaseTool):
             await self._resolve_block_credentials(user_id, block, input_data)
         )
 
+        # Get block schemas for details/validation
+        try:
+            input_schema: dict[str, Any] = block.input_schema.jsonschema()
+        except Exception as e:
+            logger.warning(
+                "Failed to generate input schema for block %s: %s",
+                block_id,
+                e,
+            )
+            return ErrorResponse(
+                message=f"Block '{block.name}' has an invalid input schema",
+                error=str(e),
+                session_id=session_id,
+            )
+        try:
+            output_schema: dict[str, Any] = block.output_schema.jsonschema()
+        except Exception as e:
+            logger.warning(
+                "Failed to generate output schema for block %s: %s",
+                block_id,
+                e,
+            )
+            return ErrorResponse(
+                message=f"Block '{block.name}' has an invalid output schema",
+                error=str(e),
+                session_id=session_id,
+            )
+
         if missing_credentials:
             # Return setup requirements response with missing credentials
             credentials_fields_info = block.input_schema.get_credentials_fields_info()
@@ -188,9 +224,56 @@ class RunBlockTool(BaseTool):
                 graph_version=None,
             )
 
+        # Check if this is a first attempt (required inputs missing)
+        # Return block details so user can see what inputs are needed
+        credentials_fields = set(block.input_schema.get_credentials_fields().keys())
+        required_keys = set(input_schema.get("required", []))
+        required_non_credential_keys = required_keys - credentials_fields
+        provided_input_keys = set(input_data.keys()) - credentials_fields
+
+        # Check for unknown input fields
+        valid_fields = (
+            set(input_schema.get("properties", {}).keys()) - credentials_fields
+        )
+        unrecognized_fields = provided_input_keys - valid_fields
+        if unrecognized_fields:
+            return InputValidationErrorResponse(
+                message=(
+                    f"Unknown input field(s) provided: {', '.join(sorted(unrecognized_fields))}. "
+                    f"Block was not executed. Please use the correct field names from the schema."
+                ),
+                session_id=session_id,
+                unrecognized_fields=sorted(unrecognized_fields),
+                inputs=input_schema,
+            )
+
+        # Show details when not all required non-credential inputs are provided
+        if not (required_non_credential_keys <= provided_input_keys):
+            # Get credentials info for the response
+            credentials_meta = []
+            for field_name, cred_meta in matched_credentials.items():
+                credentials_meta.append(cred_meta)
+
+            return BlockDetailsResponse(
+                message=(
+                    f"Block '{block.name}' details. "
+                    "Provide input_data matching the inputs schema to execute the block."
+                ),
+                session_id=session_id,
+                block=BlockDetails(
+                    id=block_id,
+                    name=block.name,
+                    description=block.description or "",
+                    inputs=input_schema,
+                    outputs=output_schema,
+                    credentials=credentials_meta,
+                ),
+                user_authenticated=True,
+            )
+
         try:
             # Get or create user's workspace for CoPilot file operations
-            workspace = await get_or_create_workspace(user_id)
+            workspace = await workspace_db().get_or_create_workspace(user_id)
 
             # Generate synthetic IDs for CoPilot context
             # Each chat session is treated as its own agent with one continuous run

autogpt_platform/backend/backend/copilot/tools/run_block_test.py

@@ -0,0 +1,366 @@
+"""Tests for block execution guards and input validation in RunBlockTool."""
+
+from unittest.mock import AsyncMock, MagicMock, patch
+
+import pytest
+
+from backend.blocks._base import BlockType
+
+from ._test_data import make_session
+from .models import (
+    BlockDetailsResponse,
+    BlockOutputResponse,
+    ErrorResponse,
+    InputValidationErrorResponse,
+)
+from .run_block import RunBlockTool
+
+_TEST_USER_ID = "test-user-run-block"
+
+
+def make_mock_block(
+    block_id: str, name: str, block_type: BlockType, disabled: bool = False
+):
+    """Create a mock block for testing."""
+    mock = MagicMock()
+    mock.id = block_id
+    mock.name = name
+    mock.block_type = block_type
+    mock.disabled = disabled
+    mock.input_schema = MagicMock()
+    mock.input_schema.jsonschema.return_value = {"properties": {}, "required": []}
+    mock.input_schema.get_credentials_fields_info.return_value = []
+    return mock
+
+
+def make_mock_block_with_schema(
+    block_id: str,
+    name: str,
+    input_properties: dict,
+    required_fields: list[str],
+    output_properties: dict | None = None,
+):
+    """Create a mock block with a defined input/output schema for validation tests."""
+    mock = MagicMock()
+    mock.id = block_id
+    mock.name = name
+    mock.block_type = BlockType.STANDARD
+    mock.disabled = False
+    mock.description = f"Test block: {name}"
+
+    input_schema = {
+        "properties": input_properties,
+        "required": required_fields,
+    }
+    mock.input_schema = MagicMock()
+    mock.input_schema.jsonschema.return_value = input_schema
+    mock.input_schema.get_credentials_fields_info.return_value = {}
+    mock.input_schema.get_credentials_fields.return_value = {}
+
+    output_schema = {
+        "properties": output_properties or {"result": {"type": "string"}},
+    }
+    mock.output_schema = MagicMock()
+    mock.output_schema.jsonschema.return_value = output_schema
+
+    return mock
+
+
+class TestRunBlockFiltering:
+    """Tests for block execution guards in RunBlockTool."""
+
+    @pytest.mark.asyncio(loop_scope="session")
+    async def test_excluded_block_type_returns_error(self):
+        """Attempting to execute a block with excluded BlockType returns error."""
+        session = make_session(user_id=_TEST_USER_ID)
+
+        input_block = make_mock_block("input-block-id", "Input Block", BlockType.INPUT)
+
+        with patch(
+            "backend.copilot.tools.run_block.get_block",
+            return_value=input_block,
+        ):
+            tool = RunBlockTool()
+            response = await tool._execute(
+                user_id=_TEST_USER_ID,
+                session=session,
+                block_id="input-block-id",
+                input_data={},
+            )
+
+        assert isinstance(response, ErrorResponse)
+        assert "cannot be run directly in CoPilot" in response.message
+        assert "designed for use within graphs only" in response.message
+
+    @pytest.mark.asyncio(loop_scope="session")
+    async def test_excluded_block_id_returns_error(self):
+        """Attempting to execute SmartDecisionMakerBlock returns error."""
+        session = make_session(user_id=_TEST_USER_ID)
+
+        smart_decision_id = "3b191d9f-356f-482d-8238-ba04b6d18381"
+        smart_block = make_mock_block(
+            smart_decision_id, "Smart Decision Maker", BlockType.STANDARD
+        )
+
+        with patch(
+            "backend.copilot.tools.run_block.get_block",
+            return_value=smart_block,
+        ):
+            tool = RunBlockTool()
+            response = await tool._execute(
+                user_id=_TEST_USER_ID,
+                session=session,
+                block_id=smart_decision_id,
+                input_data={},
+            )
+
+        assert isinstance(response, ErrorResponse)
+        assert "cannot be run directly in CoPilot" in response.message
+
+    @pytest.mark.asyncio(loop_scope="session")
+    async def test_non_excluded_block_passes_guard(self):
+        """Non-excluded blocks pass the filtering guard (may fail later for other reasons)."""
+        session = make_session(user_id=_TEST_USER_ID)
+
+        standard_block = make_mock_block(
+            "standard-id", "HTTP Request", BlockType.STANDARD
+        )
+
+        with patch(
+            "backend.copilot.tools.run_block.get_block",
+            return_value=standard_block,
+        ):
+            tool = RunBlockTool()
+            response = await tool._execute(
+                user_id=_TEST_USER_ID,
+                session=session,
+                block_id="standard-id",
+                input_data={},
+            )
+
+        # Should NOT be an ErrorResponse about CoPilot exclusion
+        # (may be other errors like missing credentials, but not the exclusion guard)
+        if isinstance(response, ErrorResponse):
+            assert "cannot be run directly in CoPilot" not in response.message
+
+
+class TestRunBlockInputValidation:
+    """Tests for input field validation in RunBlockTool.
+
+    run_block rejects unknown input field names with InputValidationErrorResponse,
+    preventing silent failures where incorrect keys would be ignored and the block
+    would execute with default values instead of the caller's intended values.
+    """
+
+    @pytest.mark.asyncio(loop_scope="session")
+    async def test_unknown_input_fields_are_rejected(self):
+        """run_block rejects unknown input fields instead of silently ignoring them.
+
+        Scenario: The AI Text Generator block has a field called 'model' (for LLM model
+        selection), but the LLM calling the tool guesses wrong and sends 'LLM_Model'
+        instead. The block should reject the request and return the valid schema.
+        """
+        session = make_session(user_id=_TEST_USER_ID)
+
+        mock_block = make_mock_block_with_schema(
+            block_id="ai-text-gen-id",
+            name="AI Text Generator",
+            input_properties={
+                "prompt": {"type": "string", "description": "The prompt to send"},
+                "model": {
+                    "type": "string",
+                    "description": "The LLM model to use",
+                    "default": "gpt-4o-mini",
+                },
+                "sys_prompt": {
+                    "type": "string",
+                    "description": "System prompt",
+                    "default": "",
+                },
+            },
+            required_fields=["prompt"],
+            output_properties={"response": {"type": "string"}},
+        )
+
+        with patch(
+            "backend.copilot.tools.run_block.get_block",
+            return_value=mock_block,
+        ):
+            tool = RunBlockTool()
+
+            # Provide 'prompt' (correct) but 'LLM_Model' instead of 'model' (wrong key)
+            response = await tool._execute(
+                user_id=_TEST_USER_ID,
+                session=session,
+                block_id="ai-text-gen-id",
+                input_data={
+                    "prompt": "Write a haiku about coding",
+                    "LLM_Model": "claude-opus-4-6",  # WRONG KEY - should be 'model'
+                },
+            )
+
+        assert isinstance(response, InputValidationErrorResponse)
+        assert "LLM_Model" in response.unrecognized_fields
+        assert "Block was not executed" in response.message
+        assert "inputs" in response.model_dump()  # valid schema included
+
+    @pytest.mark.asyncio(loop_scope="session")
+    async def test_multiple_wrong_keys_are_all_reported(self):
+        """All unrecognized field names are reported in a single error response."""
+        session = make_session(user_id=_TEST_USER_ID)
+
+        mock_block = make_mock_block_with_schema(
+            block_id="ai-text-gen-id",
+            name="AI Text Generator",
+            input_properties={
+                "prompt": {"type": "string"},
+                "model": {"type": "string", "default": "gpt-4o-mini"},
+                "sys_prompt": {"type": "string", "default": ""},
+                "retry": {"type": "integer", "default": 3},
+            },
+            required_fields=["prompt"],
+        )
+
+        with patch(
+            "backend.copilot.tools.run_block.get_block",
+            return_value=mock_block,
+        ):
+            tool = RunBlockTool()
+
+            response = await tool._execute(
+                user_id=_TEST_USER_ID,
+                session=session,
+                block_id="ai-text-gen-id",
+                input_data={
+                    "prompt": "Hello",  # correct
+                    "llm_model": "claude-opus-4-6",  # WRONG - should be 'model'
+                    "system_prompt": "Be helpful",  # WRONG - should be 'sys_prompt'
+                    "retries": 5,  # WRONG - should be 'retry'
+                },
+            )
+
+        assert isinstance(response, InputValidationErrorResponse)
+        assert set(response.unrecognized_fields) == {
+            "llm_model",
+            "system_prompt",
+            "retries",
+        }
+        assert "Block was not executed" in response.message
+
+    @pytest.mark.asyncio(loop_scope="session")
+    async def test_unknown_fields_rejected_even_with_missing_required(self):
+        """Unknown fields are caught before the missing-required-fields check."""
+        session = make_session(user_id=_TEST_USER_ID)
+
+        mock_block = make_mock_block_with_schema(
+            block_id="ai-text-gen-id",
+            name="AI Text Generator",
+            input_properties={
+                "prompt": {"type": "string"},
+                "model": {"type": "string", "default": "gpt-4o-mini"},
+            },
+            required_fields=["prompt"],
+        )
+
+        with patch(
+            "backend.copilot.tools.run_block.get_block",
+            return_value=mock_block,
+        ):
+            tool = RunBlockTool()
+
+            # 'prompt' is missing AND 'LLM_Model' is an unknown field
+            response = await tool._execute(
+                user_id=_TEST_USER_ID,
+                session=session,
+                block_id="ai-text-gen-id",
+                input_data={
+                    "LLM_Model": "claude-opus-4-6",  # wrong key, and 'prompt' is missing
+                },
+            )
+
+        # Unknown fields are caught first
+        assert isinstance(response, InputValidationErrorResponse)
+        assert "LLM_Model" in response.unrecognized_fields
+
+    @pytest.mark.asyncio(loop_scope="session")
+    async def test_correct_inputs_still_execute(self):
+        """Correct input field names pass validation and the block executes."""
+        session = make_session(user_id=_TEST_USER_ID)
+
+        mock_block = make_mock_block_with_schema(
+            block_id="ai-text-gen-id",
+            name="AI Text Generator",
+            input_properties={
+                "prompt": {"type": "string"},
+                "model": {"type": "string", "default": "gpt-4o-mini"},
+            },
+            required_fields=["prompt"],
+        )
+
+        async def mock_execute(input_data, **kwargs):
+            yield "response", "Generated text"
+
+        mock_block.execute = mock_execute
+
+        mock_workspace_db = MagicMock()
+        mock_workspace_db.get_or_create_workspace = AsyncMock(
+            return_value=MagicMock(id="test-workspace-id")
+        )
+
+        with (
+            patch(
+                "backend.copilot.tools.run_block.get_block",
+                return_value=mock_block,
+            ),
+            patch(
+                "backend.copilot.tools.run_block.workspace_db",
+                return_value=mock_workspace_db,
+            ),
+        ):
+            tool = RunBlockTool()
+
+            response = await tool._execute(
+                user_id=_TEST_USER_ID,
+                session=session,
+                block_id="ai-text-gen-id",
+                input_data={
+                    "prompt": "Write a haiku",
+                    "model": "gpt-4o-mini",  # correct field name
+                },
+            )
+
+        assert isinstance(response, BlockOutputResponse)
+        assert response.success is True
+
+    @pytest.mark.asyncio(loop_scope="session")
+    async def test_missing_required_fields_returns_details(self):
+        """Missing required fields returns BlockDetailsResponse with schema."""
+        session = make_session(user_id=_TEST_USER_ID)
+
+        mock_block = make_mock_block_with_schema(
+            block_id="ai-text-gen-id",
+            name="AI Text Generator",
+            input_properties={
+                "prompt": {"type": "string"},
+                "model": {"type": "string", "default": "gpt-4o-mini"},
+            },
+            required_fields=["prompt"],
+        )
+
+        with patch(
+            "backend.copilot.tools.run_block.get_block",
+            return_value=mock_block,
+        ):
+            tool = RunBlockTool()
+
+            # Only provide valid optional field, missing required 'prompt'
+            response = await tool._execute(
+                user_id=_TEST_USER_ID,
+                session=session,
+                block_id="ai-text-gen-id",
+                input_data={
+                    "model": "gpt-4o-mini",  # valid but optional
+                },
+            )
+
+        assert isinstance(response, BlockDetailsResponse)

autogpt_platform/backend/backend/copilot/tools/sandbox.py

@@ -0,0 +1,265 @@
+"""Sandbox execution utilities for code execution tools.
+
+Provides filesystem + network isolated command execution using **bubblewrap**
+(``bwrap``): whitelist-only filesystem (only system dirs visible read-only),
+writable workspace only, clean environment, network blocked.
+
+Tools that call :func:`run_sandboxed` must first check :func:`has_full_sandbox`
+and refuse to run if bubblewrap is not available.
+"""
+
+import asyncio
+import logging
+import os
+import platform
+import shutil
+
+logger = logging.getLogger(__name__)
+
+_DEFAULT_TIMEOUT = 30
+_MAX_TIMEOUT = 120
+
+
+# ---------------------------------------------------------------------------
+# Sandbox capability detection (cached at first call)
+# ---------------------------------------------------------------------------
+
+_BWRAP_AVAILABLE: bool | None = None
+
+
+def has_full_sandbox() -> bool:
+    """Return True if bubblewrap is available (filesystem + network isolation).
+
+    On non-Linux platforms (macOS), always returns False.
+    """
+    global _BWRAP_AVAILABLE
+    if _BWRAP_AVAILABLE is None:
+        _BWRAP_AVAILABLE = (
+            platform.system() == "Linux" and shutil.which("bwrap") is not None
+        )
+    return _BWRAP_AVAILABLE
+
+
+WORKSPACE_PREFIX = "/tmp/copilot-"
+
+
+def make_session_path(session_id: str) -> str:
+    """Build a sanitized, session-specific path under :data:`WORKSPACE_PREFIX`.
+
+    Shared by both the SDK working-directory setup and the sandbox tools so
+    they always resolve to the same directory for a given session.
+
+    Steps:
+        1. Strip all characters except ``[A-Za-z0-9-]``.
+        2. Construct ``/tmp/copilot-<safe_id>``.
+        3. Validate via ``os.path.normpath`` + ``startswith`` (CodeQL-recognised
+           sanitizer) to prevent path traversal.
+
+    Raises:
+        ValueError: If the resulting path escapes the prefix.
+    """
+    import re
+
+    safe_id = re.sub(r"[^A-Za-z0-9-]", "", session_id)
+    if not safe_id:
+        safe_id = "default"
+    path = os.path.normpath(f"{WORKSPACE_PREFIX}{safe_id}")
+    if not path.startswith(WORKSPACE_PREFIX):
+        raise ValueError(f"Session path escaped prefix: {path}")
+    return path
+
+
+def get_workspace_dir(session_id: str) -> str:
+    """Get or create the workspace directory for a session.
+
+    Uses :func:`make_session_path` — the same path the SDK uses — so that
+    bash_exec shares the workspace with the SDK file tools.
+    """
+    workspace = make_session_path(session_id)
+    os.makedirs(workspace, exist_ok=True)
+    return workspace
+
+
+# ---------------------------------------------------------------------------
+# Bubblewrap command builder
+# ---------------------------------------------------------------------------
+
+# System directories mounted read-only inside the sandbox.
+# ONLY these are visible — /app, /root, /home, /opt, /var etc. are NOT accessible.
+_SYSTEM_RO_BINDS = [
+    "/usr",  # binaries, libraries, Python interpreter
+    "/etc",  # system config: ld.so, locale, passwd, alternatives
+]
+
+# Compat paths: symlinks to /usr/* on modern Debian, real dirs on older systems.
+# On Debian 13 these are symlinks (e.g. /bin -> usr/bin).  bwrap --ro-bind
+# can't create a symlink target, so we detect and use --symlink instead.
+# /lib64 is critical: the ELF dynamic linker lives at /lib64/ld-linux-x86-64.so.2.
+_COMPAT_PATHS = [
+    ("/bin", "usr/bin"),  # -> /usr/bin on Debian 13
+    ("/sbin", "usr/sbin"),  # -> /usr/sbin on Debian 13
+    ("/lib", "usr/lib"),  # -> /usr/lib on Debian 13
+    ("/lib64", "usr/lib64"),  # 64-bit libraries / ELF interpreter
+]
+
+# Resource limits to prevent fork bombs, memory exhaustion, and disk abuse.
+# Applied via ulimit inside the sandbox before exec'ing the user command.
+_RESOURCE_LIMITS = (
+    "ulimit -u 64"  # max 64 processes  (prevents fork bombs)
+    " -v 524288"  # 512 MB virtual memory
+    " -f 51200"  # 50 MB max file size  (1024-byte blocks)
+    " -n 256"  # 256 open file descriptors
+    " 2>/dev/null"
+)
+
+
+def _build_bwrap_command(
+    command: list[str], cwd: str, env: dict[str, str]
+) -> list[str]:
+    """Build a bubblewrap command with strict filesystem + network isolation.
+
+    Security model:
+    - **Whitelist-only filesystem**: only system directories (``/usr``, ``/etc``,
+      ``/bin``, ``/lib``) are mounted read-only.  Application code (``/app``),
+      home directories, ``/var``, ``/opt``, etc. are NOT accessible at all.
+    - **Writable workspace only**: the per-session workspace is the sole
+      writable path.
+    - **Clean environment**: ``--clearenv`` wipes all inherited env vars.
+      Only the explicitly-passed safe env vars are set inside the sandbox.
+    - **Network isolation**: ``--unshare-net`` blocks all network access.
+    - **Resource limits**: ulimit caps on processes (64), memory (512MB),
+      file size (50MB), and open FDs (256) to prevent fork bombs and abuse.
+    - **New session**: prevents terminal control escape.
+    - **Die with parent**: prevents orphaned sandbox processes.
+    """
+    cmd = [
+        "bwrap",
+        # Create a new user namespace so bwrap can set up sandboxing
+        # inside unprivileged Docker containers (no CAP_SYS_ADMIN needed).
+        "--unshare-user",
+        # Wipe all inherited environment variables (API keys, secrets, etc.)
+        "--clearenv",
+    ]
+
+    # Set only the safe env vars inside the sandbox
+    for key, value in env.items():
+        cmd.extend(["--setenv", key, value])
+
+    # System directories: read-only
+    for path in _SYSTEM_RO_BINDS:
+        cmd.extend(["--ro-bind", path, path])
+
+    # Compat paths: use --symlink when host path is a symlink (Debian 13),
+    # --ro-bind when it's a real directory (older distros).
+    for path, symlink_target in _COMPAT_PATHS:
+        if os.path.islink(path):
+            cmd.extend(["--symlink", symlink_target, path])
+        elif os.path.exists(path):
+            cmd.extend(["--ro-bind", path, path])
+
+    # Wrap the user command with resource limits:
+    #   sh -c 'ulimit ...; exec "$@"' -- <original command>
+    # `exec "$@"` replaces the shell so there's no extra process overhead,
+    # and properly handles arguments with spaces.
+    limited_command = [
+        "sh",
+        "-c",
+        f'{_RESOURCE_LIMITS}; exec "$@"',
+        "--",
+        *command,
+    ]
+
+    cmd.extend(
+        [
+            # Fresh virtual filesystems
+            "--dev",
+            "/dev",
+            "--proc",
+            "/proc",
+            "--tmpfs",
+            "/tmp",
+            # Workspace bind AFTER --tmpfs /tmp so it's visible through the tmpfs.
+            # (workspace lives under /tmp/copilot-<session>)
+            "--bind",
+            cwd,
+            cwd,
+            # Isolation
+            "--unshare-net",
+            "--die-with-parent",
+            "--new-session",
+            "--chdir",
+            cwd,
+            "--",
+            *limited_command,
+        ]
+    )
+
+    return cmd
+
+
+# ---------------------------------------------------------------------------
+# Public API
+# ---------------------------------------------------------------------------
+
+
+async def run_sandboxed(
+    command: list[str],
+    cwd: str,
+    timeout: int = _DEFAULT_TIMEOUT,
+    env: dict[str, str] | None = None,
+) -> tuple[str, str, int, bool]:
+    """Run a command inside a bubblewrap sandbox.
+
+    Callers **must** check :func:`has_full_sandbox` before calling this
+    function.  If bubblewrap is not available, this function raises
+    :class:`RuntimeError` rather than running unsandboxed.
+
+    Returns:
+        (stdout, stderr, exit_code, timed_out)
+    """
+    if not has_full_sandbox():
+        raise RuntimeError(
+            "run_sandboxed() requires bubblewrap but bwrap is not available. "
+            "Callers must check has_full_sandbox() before calling this function."
+        )
+
+    timeout = min(max(timeout, 1), _MAX_TIMEOUT)
+
+    safe_env = {
+        "PATH": "/usr/local/bin:/usr/bin:/bin",
+        "HOME": cwd,
+        "TMPDIR": cwd,
+        "LANG": "en_US.UTF-8",
+        "PYTHONDONTWRITEBYTECODE": "1",
+        "PYTHONIOENCODING": "utf-8",
+    }
+    if env:
+        safe_env.update(env)
+
+    full_command = _build_bwrap_command(command, cwd, safe_env)
+
+    try:
+        proc = await asyncio.create_subprocess_exec(
+            *full_command,
+            stdout=asyncio.subprocess.PIPE,
+            stderr=asyncio.subprocess.PIPE,
+            cwd=cwd,
+            env=safe_env,
+        )
+
+        try:
+            stdout_bytes, stderr_bytes = await asyncio.wait_for(
+                proc.communicate(), timeout=timeout
+            )
+            stdout = stdout_bytes.decode("utf-8", errors="replace")
+            stderr = stderr_bytes.decode("utf-8", errors="replace")
+            return stdout, stderr, proc.returncode or 0, False
+        except asyncio.TimeoutError:
+            proc.kill()
+            await proc.communicate()
+            return "", f"Execution timed out after {timeout}s", -1, True
+
+    except RuntimeError:
+        raise
+    except Exception as e:
+        return "", f"Sandbox error: {e}", -1, False