* feat: Implement Claude orchestration with session management - Add CLAUDE_WEBHOOK_SECRET for webhook authentication - Fix Docker volume mounting for Claude credentials - Capture Claude's internal session ID from stream-json output - Update entrypoint script to support OUTPUT_FORMAT=stream-json - Fix environment variable naming (REPOSITORY -> REPO_FULL_NAME) - Enable parallel session execution with proper authentication - Successfully tested creating PRs via orchestrated sessions This enables the webhook to create and manage Claude Code sessions that can: - Clone repositories - Create feature branches - Implement code changes - Commit and push changes - Create pull requests All while capturing Claude's internal session ID for potential resumption. 🤖 Generated with [Claude Code](https://claude.ai/code) Co-Authored-By: Claude <noreply@anthropic.com> * fix: Update SessionManager tests for new implementation - Update test to expect docker volume create instead of docker create - Add unref() method to mock process objects to fix test environment error - Update spawn expectations to match new docker run implementation - Fix tests for both startSession and queueSession methods Tests now pass in CI environment. * feat: Add Claude API documentation and improve session validation - Add comprehensive Swagger/OpenAPI documentation for Claude webhook API - Add improved validation for session dependencies to handle edge cases - Add hackathon-specific Docker Compose configuration - Update SessionHandler to validate dependency UUIDs and filter invalid values - Update SessionManager to properly handle sessions without dependencies - Add API endpoint documentation with examples and schemas 🤖 Generated with [Claude Code](https://claude.ai/code) Co-Authored-By: Claude <noreply@anthropic.com> * test: Add comprehensive tests for SessionHandler dependency validation Add test coverage for dependency validation logic in SessionHandler: - Filter out invalid dependency values (empty strings, whitespace, "none") - Validate UUID format for dependencies - Handle mixed valid and invalid dependencies - Support empty dependency arrays - Handle arrays with only filtered values This improves test coverage from ~91% to ~97% for SessionHandler. 🤖 Generated with [Claude Code](https://claude.ai/code) Co-Authored-By: Claude <noreply@anthropic.com> * fix: Address PR #181 review comments - Remove unused docker-compose.hackathon.yml file - Extract UUID regex to constant for better maintainability - Document breaking changes in BREAKING_CHANGES.md - Add comprehensive examples to Swagger documentation 🤖 Generated with [Claude Code](https://claude.ai/code) Co-Authored-By: Claude <noreply@anthropic.com> --------- Co-authored-by: Claude <noreply@anthropic.com>
25 KiB
Claude Webhook API Documentation
Overview
The Claude Webhook API provides endpoints for creating and managing Claude Code sessions for automated code generation, analysis, and orchestration. This API is designed to enable parallel execution of multiple Claude instances for complex software engineering tasks.
API Design Philosophy
This API follows a simple, focused design:
- Single responsibility: Each session handles one specific task
- Orchestration via MCP/LLM agents: Complex workflows are managed by the calling agent, not the API
- Consistent response format: All responses follow the same structure for predictable parsing
Base Configuration
Base URL
POST https://your-domain.com/api/webhooks/claude
Authentication
All requests require Bearer token authentication:
Authorization: Bearer <CLAUDE_WEBHOOK_SECRET>
Content-Type: application/json
Response Format
All API responses follow this consistent structure:
{
"success": boolean,
"message": "string", // Human-readable status message
"data": object, // Response data (when success=true)
"error": "string" // Error description (when success=false)
}
Rate Limiting
- Currently not implemented (planned for future release)
- Recommended client-side rate limiting: 10 requests per minute
Endpoints
1. Create Session
Creates a new Claude Code session. Sessions can be configured with dependencies, metadata, and execution options.
Endpoint: POST /api/webhooks/claude
Type: session.create
Request Body
{
"type": "session.create",
"session": {
"type": "implementation | analysis | testing | review | coordination",
"project": {
"repository": "string", // Required: "owner/repo" format
"branch": "string", // Optional: target branch
"requirements": "string", // Required: task description
"context": "string" // Optional: additional context
},
"dependencies": ["string"], // Optional: array of session IDs to wait for
"metadata": { // Optional: custom metadata
"batchId": "string", // Group related sessions
"tags": ["string"], // Categorization tags
"priority": "string" // Priority level
}
},
"options": { // Optional: execution options
"autoStart": boolean, // Start when dependencies complete (default: false)
"timeout": number, // Custom timeout in seconds (default: 1800)
"notifyUrl": "string" // Webhook URL for completion notification
}
}
Parameters
| Parameter | Type | Required | Description |
|---|---|---|---|
type |
string | Yes | Must be "session.create" |
session |
object | Yes | Session configuration object |
session.type |
string | Yes | Type of session: implementation, analysis, testing, review, or coordination |
session.project |
object | Yes | Project information |
session.project.repository |
string | Yes | GitHub repository in "owner/repo" format |
session.project.branch |
string | No | Target branch name (defaults to main/master) |
session.project.requirements |
string | Yes | Clear description of what Claude should do |
session.project.context |
string | No | Additional context about the codebase or requirements |
session.dependencies |
string[] | No | Array of valid UUID session IDs that must complete before this session starts (filters out "none", empty strings) |
session.metadata |
object | No | Custom metadata for organizing sessions |
session.metadata.batchId |
string | No | User-provided ID for grouping related sessions |
session.metadata.tags |
string[] | No | Tags for categorization |
session.metadata.priority |
string | No | Priority level (high, medium, low) |
options |
object | No | Execution options |
options.autoStart |
boolean | No | Automatically start when dependencies complete (default: false) |
options.timeout |
number | No | Custom timeout in seconds (default: 1800 = 30 minutes) |
options.notifyUrl |
string | No | Webhook URL to call on completion/failure |
Response
{
"success": true,
"message": "Session created successfully",
"data": {
"session": {
"id": "550e8400-e29b-41d4-a716-446655440000",
"type": "implementation",
"status": "pending",
"project": {
"repository": "acme/webapp",
"branch": "feature/user-auth",
"requirements": "Implement JWT authentication middleware",
"context": "Use existing User model"
},
"dependencies": [],
"metadata": {
"batchId": "auth-feature-batch",
"tags": ["feature", "auth"],
"priority": "high"
},
"options": {
"autoStart": false,
"timeout": 1800,
"notifyUrl": null
},
"containerId": null,
"claudeSessionId": null,
"createdAt": "2024-01-06T10:00:00Z",
"startedAt": null,
"completedAt": null,
"output": null,
"error": null
}
}
}
Example
curl -X POST https://your-domain.com/api/webhooks/claude \
-H "Authorization: Bearer your-secret-token" \
-H "Content-Type: application/json" \
-d '{
"type": "session.create",
"session": {
"type": "implementation",
"project": {
"repository": "acme/webapp",
"branch": "feature/user-auth",
"requirements": "Implement JWT authentication middleware for Express.js",
"context": "Use existing User model and bcrypt for password hashing"
},
"dependencies": []
}
}'
2. Start Session
Starts a previously created session or queues it if dependencies aren't met.
Endpoint: POST /api/webhooks/claude
Type: session.start
Request Body
{
"type": "session.start",
"sessionId": "string"
}
Parameters
| Parameter | Type | Required | Description |
|---|---|---|---|
type |
string | Yes | Must be "session.start" |
sessionId |
string | Yes | UUID of the session to start |
Response
{
"success": true,
"message": "Session started successfully",
"data": {
"session": {
"id": "550e8400-e29b-41d4-a716-446655440000",
"status": "initializing", // or "running" if started immediately
"containerId": "docker-container-id",
"claudeSessionId": "claude-internal-session-id",
// ... full session object
}
}
}
For queued sessions (waiting on dependencies):
{
"success": true,
"message": "Session queued",
"data": {
"session": {
"id": "550e8400-e29b-41d4-a716-446655440000",
"status": "pending",
// ... full session object
},
"queueStatus": {
"waitingFor": ["dependency-session-id-1", "dependency-session-id-2"],
"estimatedStartTime": null
}
}
}
Example
curl -X POST https://your-domain.com/api/webhooks/claude \
-H "Authorization: Bearer your-secret-token" \
-H "Content-Type: application/json" \
-d '{
"type": "session.start",
"sessionId": "550e8400-e29b-41d4-a716-446655440000"
}'
3. Get Session Status
Retrieves the current status and details of a session.
Endpoint: POST /api/webhooks/claude
Type: session.get
Request Body
{
"type": "session.get",
"sessionId": "string"
}
Parameters
| Parameter | Type | Required | Description |
|---|---|---|---|
type |
string | Yes | Must be "session.get" |
sessionId |
string | Yes | UUID of the session to query |
Response
{
"success": true,
"message": "Session found",
"data": {
"session": {
"id": "550e8400-e29b-41d4-a716-446655440000",
"type": "implementation",
"status": "running",
"containerId": "docker-container-id",
"claudeSessionId": "claude-internal-session-id",
"project": {
"repository": "acme/webapp",
"branch": "feature/user-auth",
"requirements": "Implement JWT authentication middleware",
"context": "Use existing User model"
},
"dependencies": [],
"metadata": {},
"options": {},
"createdAt": "2024-01-06T10:00:00Z",
"startedAt": "2024-01-06T10:30:00Z",
"completedAt": null,
"output": null,
"error": null
}
}
}
Session Status Values
pending- Session created but not startedinitializing- Container is being createdrunning- Session is actively executingcompleted- Session finished successfullyfailed- Session encountered an errorcancelled- Session was manually cancelled
4. Get Session Output
Retrieves the output and artifacts from a completed session.
Endpoint: POST /api/webhooks/claude
Type: session.output
Request Body
{
"type": "session.output",
"sessionId": "string"
}
Parameters
| Parameter | Type | Required | Description |
|---|---|---|---|
type |
string | Yes | Must be "session.output" |
sessionId |
string | Yes | UUID of the session |
Response
{
"success": true,
"message": "Session output retrieved",
"data": {
"output": {
"logs": ["Container started", "Running Claude command...", "Task completed"],
"artifacts": [
{
"type": "file",
"path": "src/middleware/auth.js",
"content": "// JWT authentication middleware\n...",
"sha": "abc123...",
"url": "https://github.com/acme/webapp/blob/feature/user-auth/src/middleware/auth.js",
"metadata": {
"linesAdded": 150,
"linesRemoved": 0
}
}
],
"summary": "Implemented JWT authentication middleware with refresh token support",
"nextSteps": ["Add rate limiting", "Implement password reset flow"],
"executionTime": 180, // seconds
"resourceUsage": {
"cpuTime": 45.2,
"memoryPeak": "512MB"
}
}
}
}
Note: The current implementation returns a simplified structure. Full artifact details and metadata are planned for future releases.
5. List Sessions
Lists all sessions, optionally filtered by orchestration ID.
Endpoint: POST /api/webhooks/claude
Type: session.list
Request Body
{
"type": "session.list",
"orchestrationId": "string" // Optional
}
Parameters
| Parameter | Type | Required | Description |
|---|---|---|---|
type |
string | Yes | Must be "session.list" |
orchestrationId |
string | No | Filter sessions by orchestration ID |
Response
{
"success": true,
"message": "Sessions retrieved",
"data": {
"sessions": [
{
"id": "550e8400-e29b-41d4-a716-446655440000",
"type": "implementation",
"status": "completed",
"project": {
"repository": "acme/webapp",
"branch": "feature/user-auth",
"requirements": "Implement JWT authentication",
"context": null
},
"dependencies": [],
"metadata": {
"batchId": "auth-feature-batch",
"tags": ["feature", "auth"]
},
"createdAt": "2024-01-06T10:00:00Z",
"startedAt": "2024-01-06T10:30:00Z",
"completedAt": "2024-01-06T10:45:00Z",
"error": null
},
{
"id": "660e8400-e29b-41d4-a716-446655440001",
"type": "testing",
"status": "running",
"project": {
"repository": "acme/webapp",
"branch": "feature/user-auth",
"requirements": "Write tests for JWT middleware"
},
"dependencies": ["550e8400-e29b-41d4-a716-446655440000"],
"metadata": {
"batchId": "auth-feature-batch",
"tags": ["testing"]
},
"createdAt": "2024-01-06T10:46:00Z",
"startedAt": "2024-01-06T10:47:00Z",
"completedAt": null,
"error": null
}
]
}
}
Session Types
implementation
For implementing new features or functionality. Claude will:
- Analyze requirements
- Write production-ready code
- Follow existing patterns and conventions
- Create or modify files as needed
analysis
For analyzing existing code. Claude will:
- Review code structure and patterns
- Identify potential issues
- Suggest improvements
- Document findings
testing
For creating and running tests. Claude will:
- Write unit and integration tests
- Ensure code coverage
- Validate functionality
- Fix failing tests
review
For code review tasks. Claude will:
- Review pull requests
- Check for security issues
- Validate best practices
- Provide feedback
coordination
For orchestrating multiple sessions. Claude will:
- Break down complex tasks
- Create dependent sessions
- Monitor progress
- Coordinate results
Dependency Management
Sessions can depend on other sessions using the dependencies parameter:
{
"type": "session.create",
"session": {
"type": "testing",
"project": {
"repository": "acme/webapp",
"requirements": "Write tests for the JWT authentication middleware"
},
"dependencies": ["implementation-session-id"]
}
}
Dependency Behavior
- Sessions with dependencies won't start until all dependencies are
completed - If any dependency fails, the dependent session will be marked as
failed - Circular dependencies are detected and rejected
- Maximum dependency depth is 10 levels
Error Handling
Error Response Format
{
"success": false,
"error": "Error description"
}
Common Error Codes
400- Bad Request (invalid parameters)401- Unauthorized (invalid token)404- Not Found (session doesn't exist)409- Conflict (session already started)429- Too Many Requests (rate limit exceeded)500- Internal Server Error
Example Error Response
{
"success": false,
"error": "Session not found: 550e8400-e29b-41d4-a716-446655440000"
}
Best Practices
1. Clear Requirements
Provide detailed, actionable requirements:
{
"requirements": "Implement JWT authentication middleware with:\n- Access token (15min expiry)\n- Refresh token (7 days expiry)\n- Token blacklisting for logout\n- Rate limiting per user"
}
2. Use Dependencies Wisely
Chain related tasks:
analysis → implementation → testing → review
3. Provide Context
Include relevant context about your codebase:
{
"context": "We use Express.js with TypeScript, Prisma ORM, and follow REST API conventions. Authentication should integrate with existing User model."
}
4. Monitor Session Status
Poll session status every 5-10 seconds:
while [ "$status" != "completed" ]; do
status=$(curl -s -X POST ... | jq -r '.data.status')
sleep 5
done
5. Handle Failures Gracefully
Check session status and error messages:
if (response.data.status === 'failed') {
console.error('Session failed:', response.data.error);
// Implement retry logic or alternative approach
}
Integration Examples
Node.js/TypeScript
import axios from 'axios';
const CLAUDE_API_URL = 'https://your-domain.com/api/webhooks/claude';
const AUTH_TOKEN = process.env.CLAUDE_WEBHOOK_SECRET;
async function createAndRunSession() {
// Create session
const createResponse = await axios.post(
CLAUDE_API_URL,
{
type: 'session.create',
session: {
type: 'implementation',
project: {
repository: 'acme/webapp',
requirements: 'Implement user profile API endpoints',
context: 'Use existing auth middleware'
}
}
},
{
headers: {
'Authorization': `Bearer ${AUTH_TOKEN}`,
'Content-Type': 'application/json'
}
}
);
const sessionId = createResponse.data.data.sessionId;
// Start session
await axios.post(
CLAUDE_API_URL,
{
type: 'session.start',
sessionId
},
{
headers: {
'Authorization': `Bearer ${AUTH_TOKEN}`,
'Content-Type': 'application/json'
}
}
);
// Poll for completion
let status = 'running';
while (status === 'running' || status === 'initializing') {
await new Promise(resolve => setTimeout(resolve, 5000));
const statusResponse = await axios.post(
CLAUDE_API_URL,
{
type: 'session.get',
sessionId
},
{
headers: {
'Authorization': `Bearer ${AUTH_TOKEN}`,
'Content-Type': 'application/json'
}
}
);
status = statusResponse.data.data.status;
}
// Get output
if (status === 'completed') {
const outputResponse = await axios.post(
CLAUDE_API_URL,
{
type: 'session.output',
sessionId
},
{
headers: {
'Authorization': `Bearer ${AUTH_TOKEN}`,
'Content-Type': 'application/json'
}
}
);
console.log('Session completed:', outputResponse.data.data.summary);
console.log('Artifacts:', outputResponse.data.data.artifacts);
}
}
Python
import requests
import time
import os
CLAUDE_API_URL = 'https://your-domain.com/api/webhooks/claude'
AUTH_TOKEN = os.environ['CLAUDE_WEBHOOK_SECRET']
headers = {
'Authorization': f'Bearer {AUTH_TOKEN}',
'Content-Type': 'application/json'
}
# Create session
create_response = requests.post(
CLAUDE_API_URL,
json={
'type': 'session.create',
'session': {
'type': 'implementation',
'project': {
'repository': 'acme/webapp',
'requirements': 'Implement user profile API endpoints'
}
}
},
headers=headers
)
session_id = create_response.json()['data']['sessionId']
# Start session
requests.post(
CLAUDE_API_URL,
json={
'type': 'session.start',
'sessionId': session_id
},
headers=headers
)
# Poll for completion
status = 'running'
while status in ['running', 'initializing']:
time.sleep(5)
status_response = requests.post(
CLAUDE_API_URL,
json={
'type': 'session.get',
'sessionId': session_id
},
headers=headers
)
status = status_response.json()['data']['status']
# Get output
if status == 'completed':
output_response = requests.post(
CLAUDE_API_URL,
json={
'type': 'session.output',
'sessionId': session_id
},
headers=headers
)
output = output_response.json()['data']
print(f"Summary: {output['summary']}")
print(f"Artifacts: {output['artifacts']}")
LLM Agent Integration Guide
This section provides specific guidance for LLM agents (via MCP servers or other integrations) consuming this API.
Response Parsing
All responses follow a consistent structure, making them easy to parse:
interface ApiResponse<T> {
success: boolean;
message: string;
data?: T; // Present when success=true
error?: string; // Present when success=false
}
Session Orchestration Pattern
Since this API focuses on single-session creation, orchestration should be handled by the LLM agent:
# Example: LLM agent orchestrating a feature implementation
async def implement_feature(repo: str, feature_desc: str):
# 1. Create analysis session
analysis = await create_session(
type="analysis",
requirements=f"Analyze codebase for implementing: {feature_desc}"
)
# 2. Wait for analysis to complete
await wait_for_completion(analysis.id)
# 3. Create implementation session based on analysis
implementation = await create_session(
type="implementation",
requirements=f"Implement {feature_desc} based on analysis",
dependencies=[analysis.id]
)
# 4. Create testing session
testing = await create_session(
type="testing",
requirements=f"Write tests for {feature_desc}",
dependencies=[implementation.id],
options={"autoStart": true} # Auto-start when ready
)
return {
"analysis": analysis.id,
"implementation": implementation.id,
"testing": testing.id
}
Polling Best Practices
async function pollSession(sessionId, maxAttempts = 120) {
const pollInterval = 5000; // 5 seconds
let attempts = 0;
while (attempts < maxAttempts) {
const response = await getSession(sessionId);
const status = response.data.session.status;
if (['completed', 'failed', 'cancelled'].includes(status)) {
return response.data.session;
}
// Exponential backoff for long-running sessions
const delay = status === 'pending' ? pollInterval * 2 : pollInterval;
await sleep(delay);
attempts++;
}
throw new Error('Session polling timeout');
}
Batch Processing Pattern
Use metadata to group related sessions:
{
"type": "session.create",
"session": {
"type": "implementation",
"project": { ... },
"metadata": {
"batchId": "feature-xyz-batch",
"tags": ["feature", "priority-high"],
"priority": "high"
}
}
}
Then query all sessions in a batch:
{
"type": "session.list",
"orchestrationId": "feature-xyz-batch"
}
Error Handling
def handle_api_response(response):
if response.status_code == 429:
# Rate limited - implement exponential backoff
retry_after = int(response.headers.get('Retry-After', 60))
time.sleep(retry_after)
return retry_request()
data = response.json()
if not data['success']:
error = data.get('error', 'Unknown error')
if 'not found' in error:
# Handle missing session
pass
elif 'already started' in error:
# Session already running - just poll for status
pass
else:
raise ApiError(error)
return data['data']
Dependency Graph Building
class SessionGraph {
private sessions: Map<string, Session> = new Map();
addSession(session: Session) {
this.sessions.set(session.id, session);
}
getExecutionOrder(): string[] {
// Topological sort to determine execution order
const visited = new Set<string>();
const order: string[] = [];
const visit = (id: string) => {
if (visited.has(id)) return;
visited.add(id);
const session = this.sessions.get(id);
if (session?.dependencies) {
session.dependencies.forEach(dep => visit(dep));
}
order.push(id);
};
this.sessions.forEach((_, id) => visit(id));
return order;
}
}
Optimizing for Claude Code
When creating sessions for Claude Code:
-
Clear Requirements: Be specific and actionable
{ "requirements": "Implement REST API endpoint POST /api/users with:\n- Request validation (email, password)\n- Password hashing with bcrypt\n- Store in PostgreSQL users table\n- Return JWT token\n- Handle duplicate email error", "context": "Using Express.js, TypeScript, Prisma ORM. Follow existing auth patterns in src/middleware/auth.ts" } -
Provide Context: Reference existing code patterns
{ "context": "Follow patterns in src/controllers/. Use existing error handling middleware. See src/types/user.ts for User interface." } -
Use Session Types Effectively:
analysis- Before implementing, understand the codebaseimplementation- Write the actual codetesting- Ensure code works and has coveragereview- Final quality checkcoordination- For complex multi-part tasks
Performance Tips
- Parallel Sessions: Create independent sessions simultaneously
- Reuse Analysis: Cache analysis results for similar tasks
- Smart Dependencies: Only add dependencies when truly needed
- Batch Operations: Group related sessions with metadata
Troubleshooting
Session Stuck in "pending"
- Check if dependencies are completed
- Verify Docker daemon is running
- Check system resources (CPU, memory)
- Use
session.getto check dependency status
Authentication Errors
- Verify Bearer token matches CLAUDE_WEBHOOK_SECRET
- Ensure Authorization header is properly formatted
- Check token hasn't been rotated
Session Failures
- Review session output for error messages
- Check Docker container logs
- Verify repository access permissions
- Ensure Claude API credentials are valid
Timeout Issues
- Default timeout is 30 minutes per session
- For longer tasks, break into smaller sessions
- Use custom timeout in options:
{"timeout": 3600}
Changelog
v2.0.0 (2024-01-08)
- BREAKING: Removed orchestration endpoint (use session.create with type="coordination")
- BREAKING: Updated response structures (all data wrapped in
data.sessionordata.sessions) - Added enhanced session creation with metadata and options
- Added autoStart option for dependency-based execution
- Added timeout and notification options
- Improved dependency validation (filters invalid UUIDs)
v1.0.0 (2024-01-06)
- Initial release with session management
- Support for 5 session types
- Dependency management
- Orchestration capabilities