* feat: Implement Claude orchestration provider for parallel session management - Add ClaudeWebhookProvider implementing the webhook provider interface - Create orchestration system for running multiple Claude containers in parallel - Implement smart task decomposition to break complex projects into workstreams - Add session management with dependency tracking between sessions - Support multiple execution strategies (parallel, sequential, wait_for_core) - Create comprehensive test suite for all components - Add documentation for Claude orchestration API and usage This enables super-charged Claude capabilities for the MCP hackathon by allowing multiple Claude instances to work on different aspects of a project simultaneously, with intelligent coordination and result aggregation. 🤖 Generated with [Claude Code](https://claude.ai/code) Co-Authored-By: Claude <noreply@anthropic.com> * feat: Add session management endpoints for MCP integration - Add SessionHandler for individual session CRUD operations - Create endpoints: session.create, session.get, session.list, session.start, session.output - Fix Claude invocation in Docker containers using proper claude chat command - Add volume mounts for persistent storage across session lifecycle - Simplify OrchestrationHandler to create single coordination sessions - Update documentation with comprehensive MCP integration examples - Add comprehensive unit and integration tests for new endpoints - Support dependencies and automatic session queuing/starting This enables Claude Desktop to orchestrate multiple Claude Code sessions via MCP Server tools. 🤖 Generated with [Claude Code](https://claude.ai/code) Co-Authored-By: Claude <noreply@anthropic.com> * fix: Update ClaudeWebhookProvider validation for session endpoints - Make project fields optional for session management operations - Add validation for session.create requiring session field - Update tests to match new validation rules - Fix failing CI tests 🤖 Generated with [Claude Code](https://claude.ai/code) Co-Authored-By: Claude <noreply@anthropic.com> * fix: Use Promise.reject for validation errors in parsePayload - Convert synchronous throws to Promise.reject for async consistency - Fixes failing unit tests expecting rejected promises 🤖 Generated with [Claude Code](https://claude.ai/code) Co-Authored-By: Claude <noreply@anthropic.com> * fix: Mock SessionManager in integration tests to avoid Docker calls in CI - Add SessionManager mock to prevent Docker operations during tests - Fix claude-webhook.test.ts to use proper test setup and payload structure - Ensure all integration tests can run without Docker dependency - Fix payload structure to include 'data' wrapper 🤖 Generated with [Claude Code](https://claude.ai/code) Co-Authored-By: Claude <noreply@anthropic.com> * fix: Mock child_process to prevent Docker calls in CI tests - Mock execSync and spawn at child_process level to prevent any Docker commands - This ensures tests work in CI environment without Docker - Tests now pass both locally and in CI Docker build 🤖 Generated with [Claude Code](https://claude.ai/code) Co-Authored-By: Claude <noreply@anthropic.com> * fix: Address PR review comments and fix linter warnings - Move @types/uuid to devDependencies - Replace timestamp+Math.random with crypto.randomUUID() for better uniqueness - Extract magic number into EXTRA_SESSIONS_COUNT constant - Update determineStrategy return type to use literal union - Fix unnecessary optional chaining warnings - Handle undefined labels in GitHub transformers - Make TaskDecomposer.decompose synchronous - Add proper eslint-disable comments for intentional sync methods - Fix all TypeScript and formatting issues * fix: Mock SessionManager in integration tests to prevent Docker calls in CI - Add SessionManager mocks to claude-session.test.ts - Add SessionManager mocks to claude-webhook.test.ts - Prevents 500 errors when running tests in CI without Docker - All integration tests now pass without requiring Docker runtime * fix: Run only unit tests in Docker builds to avoid Docker-in-Docker issues - Change test stage to run 'npm run test:unit' instead of 'npm test' - Skips integration tests that require Docker runtime - Prevents CI failures in Docker container builds - Integration tests still run in regular CI workflow * fix: Use Dockerfile CMD for tests in Docker build CI - Remove explicit 'npm test' command from docker run - Let Docker use the CMD defined in Dockerfile (npm run test:unit) - This ensures consistency and runs only unit tests in Docker builds --------- Co-authored-by: Claude <noreply@anthropic.com>
14 KiB
Claude GitHub Webhook
🚀 Quick Start Guide | 💬 Discord | 📚 Documentation | 📖 Complete Setup | 🔐 Authentication
Deploy Claude Code as a fully autonomous GitHub bot. Create your own bot account, mention it in any issue or PR, and watch AI-powered development happen end-to-end. Claude can implement complete features, review code, merge PRs, wait for CI builds, and run for hours autonomously until tasks are completed. Production-ready microservice with container isolation, automated workflows, and intelligent project management.
What This Does
# In any GitHub issue or PR (using your configured bot account):
@YourBotName implement user authentication with OAuth
@YourBotName review this PR for security vulnerabilities
@YourBotName fix the failing CI tests and merge when ready
@YourBotName refactor the database layer for better performance
Claude autonomously handles complete development workflows. It analyzes your entire repository, implements features from scratch, conducts thorough code reviews, manages pull requests, monitors CI/CD pipelines, and responds to automated feedback - all without human intervention. No context switching. No manual oversight required. Just seamless autonomous development where you work.
🚀 Quick Start
Follow our 10-minute Quick Start Guide to get Claude responding to your GitHub issues using Cloudflare Tunnel - no domain or complex setup required!
# 1. Clone and configure
git clone https://github.com/claude-did-this/claude-hub.git
cd claude-hub
cp .env.quickstart .env
nano .env # Add your GitHub token and bot details
# 2. Authenticate Claude (uses your Claude.ai Max subscription)
./scripts/setup/setup-claude-interactive.sh
# 3. Start the service
docker compose up -d
# 4. Create a tunnel (see quickstart guide for details)
cloudflared tunnel --url http://localhost:3002
That's it! Your bot is ready to use. See the complete quickstart guide for detailed instructions and webhook setup.
Autonomous Workflow Capabilities
End-to-End Development 🚀
- Feature Implementation: From requirements to fully tested, production-ready code
- Code Review & Quality: Comprehensive analysis including security, performance, and best practices
- PR Lifecycle Management: Creates branches, commits changes, pushes code, and manages merge process
- CI/CD Monitoring: Actively waits for builds, analyzes test results, and fixes failures
- Automated Code Response: Responds to automated review comments and adapts based on feedback
Intelligent Task Management 🧠
- Multi-hour Operations: Continues working autonomously until complex tasks are 100% complete
- Dependency Resolution: Handles blockers, waits for external processes, and resumes work automatically
- Context Preservation: Maintains project state and progress across long-running operations
- Adaptive Problem Solving: Iterates on solutions based on test results and code review feedback
Key Features
Autonomous Development 🤖
- Complete Feature Implementation: Claude codes entire features from requirements to deployment
- Intelligent PR Management: Automatically creates, reviews, and merges pull requests
- CI/CD Integration: Waits for builds, responds to test failures, and handles automated workflows
- Long-running Tasks: Operates autonomously for hours until complex projects are completed
- Auto-labeling: New issues automatically tagged by content analysis
- Context-aware: Claude understands your entire repository structure and development patterns
- Stateless execution: Each request runs in isolated Docker containers
Claude Orchestration (NEW) 🎭
- Parallel Claude Sessions: Run multiple Claude containers concurrently for complex tasks
- Smart Task Decomposition: Automatically breaks down projects into parallel workstreams
- Dependency Management: Sessions wait for prerequisites before starting
- MCP Integration: Built for the MCP hackathon to showcase super-charged Claude capabilities
- See Claude Orchestration Documentation for details
Performance Architecture ⚡
- Parallel test execution with strategic runner distribution
- Conditional Docker builds (only when code changes)
- Repository caching for sub-second response times
- Advanced build profiling with timing metrics
Enterprise Security 🔒
- Webhook signature verification (HMAC-SHA256)
- AWS IAM role-based authentication
- Pre-commit credential scanning
- Container isolation with minimal permissions
- Fine-grained GitHub token scoping
Bot Account Setup
Current Setup: You need to create your own GitHub bot account:
- Create a dedicated GitHub account for your bot (e.g.,
MyProjectBot) - Generate a Personal Access Token with repository permissions
- Configure the bot username in your environment variables
- Add the bot account as a collaborator to your repositories
Future Release: We plan to release this as a GitHub App that provides a universal bot account, eliminating the need for individual bot setup while maintaining the same functionality for self-hosted instances.
Production Deployment
1. Environment Configuration
# Core settings
BOT_USERNAME=YourBotName # GitHub bot account username (create your own bot account)
GITHUB_WEBHOOK_SECRET=<generated> # Webhook validation
GITHUB_TOKEN=<fine-grained-pat> # Repository access (from your bot account)
# Claude Authentication - Choose ONE method:
# Option 1: Setup Container (Personal/Development)
# Use existing Claude Max subscription (5x or 20x plans)
# See docs/setup-container-guide.md for setup
# Option 2: Direct API Key (Production/Team)
ANTHROPIC_API_KEY=sk-ant-your-api-key
# Option 3: AWS Bedrock (Enterprise)
AWS_REGION=us-east-1
ANTHROPIC_MODEL=anthropic.claude-3-sonnet-20240229-v1:0
CLAUDE_CODE_USE_BEDROCK=1
# Security
AUTHORIZED_USERS=user1,user2,user3 # Allowed GitHub usernames
CLAUDE_API_AUTH_REQUIRED=1 # Enable API authentication
Authentication Methods
Setup Container (Personal/Development)
Use your existing Claude Max subscription for automation instead of pay-per-use API fees:
# 1. Run interactive authentication setup
./scripts/setup/setup-claude-interactive.sh
# 2. In container: authenticate with your subscription
claude --dangerously-skip-permissions # Follow authentication flow
exit # Save authentication
# 3. Use captured authentication
cp -r ${CLAUDE_HUB_DIR:-~/.claude-hub}/* ~/.claude/
Prerequisites: Claude Max subscription (5x or 20x plans). Claude Pro does not include Claude Code access.
Details: Setup Container Guide
Direct API Key (Production/Team)
ANTHROPIC_API_KEY=sk-ant-your-api-key-here
Best for: Production environments, team usage, guaranteed stability.
Details: Authentication Guide
AWS Bedrock (Enterprise)
AWS_REGION=us-east-1
ANTHROPIC_MODEL=anthropic.claude-3-sonnet-20240229-v1:0
CLAUDE_CODE_USE_BEDROCK=1
Best for: Enterprise deployments, AWS integration, compliance requirements.
Details: Authentication Guide
2. GitHub Webhook Setup
- Navigate to Repository → Settings → Webhooks
- Add webhook:
- Payload URL:
https://your-domain.com/api/webhooks/github - Content type:
application/json - Secret: Your
GITHUB_WEBHOOK_SECRET - Events: Select "Send me everything"
- Payload URL:
3. AWS Authentication Options
# Option 1: IAM Instance Profile (EC2)
# Automatically uses instance metadata
# Option 2: ECS Task Role
# Automatically uses container credentials
# Option 3: AWS Profile
./scripts/aws/setup-aws-profiles.sh
# Option 4: Static Credentials (not recommended)
AWS_ACCESS_KEY_ID=xxx
AWS_SECRET_ACCESS_KEY=xxx
Advanced Usage
Direct API Access
Integrate Claude without GitHub webhooks:
curl -X POST http://localhost:3002/api/claude \
-H "Content-Type: application/json" \
-d '{
"repoFullName": "owner/repo",
"command": "Analyze security vulnerabilities",
"authToken": "your-token",
"useContainer": true
}'
CLI Tool
# Basic usage
./cli/claude-webhook myrepo "Review the authentication flow"
# PR review
./cli/claude-webhook owner/repo "Review this PR" -p -b feature-branch
# Specific issue
./cli/claude-webhook myrepo "Fix this bug" -i 42
Container Execution Modes
Different operations use tailored security profiles for autonomous execution:
- Auto-tagging: Minimal permissions (Read + GitHub tools only)
- PR Reviews: Standard permissions (full tool access with automated merge capabilities)
- Feature Development: Full development permissions (code editing, testing, CI monitoring)
- Long-running Tasks: Extended container lifetime with checkpoint/resume functionality
- Custom Commands: Configurable via
--allowedToolsflag
Architecture Deep Dive
Autonomous Request Flow
GitHub Event → Webhook Endpoint → Signature Verification
↓ ↓
Container Spawn ← Command Parser ← Event Processor
↓
Claude Analysis → Feature Implementation → Testing & CI
↓ ↓ ↓
GitHub API ← Code Review ← PR Management ← Build Monitoring
↓
Autonomous Merge/Deploy → Task Completion
Autonomous Container Lifecycle
- Spawn: New Docker container per request with extended lifetime for long tasks
- Clone: Repository fetched (or cache hit) with full development setup
- Execute: Claude implements features, runs tests, monitors CI, handles feedback autonomously
- Iterate: Continuous development cycle until task completion
- Deploy: Results pushed, PRs merged, tasks marked complete
- Cleanup: Container destroyed after successful task completion
Security Layers
- Network: Webhook signature validation
- Authentication: GitHub user allowlist
- Authorization: Fine-grained token permissions
- Execution: Container isolation
- Tools: Operation-specific allowlists
Performance Tuning
Repository Caching
REPO_CACHE_DIR=/cache/repos
REPO_CACHE_MAX_AGE_MS=3600000 # 1 hour
Container Optimization
CONTAINER_LIFETIME_MS=7200000 # 2 hour timeout
CLAUDE_CONTAINER_IMAGE=claudecode:latest
CI/CD Pipeline
- Parallel Jest test execution
- Docker layer caching
- Conditional image builds
- Self-hosted runners for heavy operations
Monitoring & Debugging
Health Check
curl http://localhost:3002/health
Logs
docker compose logs -f webhook
Test Suite
npm test # All tests
npm run test:unit # Unit only
npm run test:integration # Integration only
npm run test:coverage # With coverage report
Debug Mode
DEBUG=claude:* npm run dev
Documentation
Deep Dive Guides
- Setup Container Authentication - Technical details for subscription-based auth
- Authentication Guide - All authentication methods and troubleshooting
- Complete Workflow - End-to-end technical guide
- Container Setup - Docker configuration details
- AWS Best Practices - IAM and credential management
- GitHub Integration - Webhook events and permissions
Reference
- Scripts Documentation - Utility scripts and commands
- Command Reference - Build and run commands
Contributing
Development Setup
# Install dependencies
npm install
# Setup pre-commit hooks
./scripts/setup/setup-precommit.sh
# Run in dev mode
npm run dev
Code Standards
- Node.js 20+ with async/await patterns
- Jest for testing with >80% coverage target
- ESLint + Prettier for code formatting
- Conventional commits for version management
Security Checklist
- No hardcoded credentials
- All inputs sanitized
- Webhook signatures verified
- Container permissions minimal
- Logs redact sensitive data
Troubleshooting
Common Issues
Webhook not responding
- Verify signature secret matches
- Check GitHub token permissions
- Confirm webhook URL is accessible
Claude timeouts
- Increase
CONTAINER_LIFETIME_MS - Check AWS Bedrock quotas
- Verify network connectivity
Permission denied
- Confirm user in
AUTHORIZED_USERS - Check GitHub token scopes
- Verify AWS IAM permissions
Support
- Report issues: GitHub Issues
- Detailed troubleshooting: Complete Workflow Guide
License
MIT - See the LICENSE file for details.
