draft.yaml
The main configuration file lives at the root of your repository. It controls the AI executor, verification commands, planner behavior, and more.
# AI Executor Configuration
executor_config:
executor_type: CLAUDE # CLAUDE, CURSOR_AGENT, or CURSOR
yolo_mode: false # Skip permission prompts (Claude CLI only)
yolo_allowlist: [] # Repo paths where YOLO mode is allowed
# Verification Pipeline
verify_config:
commands:
- "pytest tests/"
- "npm run lint"
# Planner Configuration
planner_config:
model: "claude-sonnet-4-20250514" # LLM model for planning
auto_execute: false # Auto-run tickets after planning
validate_tickets: false # Validate generated tickets with LLM
max_followups_per_ticket: 2 # Max follow-up tickets per blocked ticket
max_followups_per_tick: 3 # Max follow-ups per planner tick
# Cleanup Configuration
cleanup_config:
worktree_ttl_hours: 72 # Auto-clean worktrees older than this
Environment Variables
Backend (backend/.env)
Copy from the example file:
cp backend/.env.example backend/.env
| Variable | Description | Default |
|---|
APP_ENV | Environment mode | development |
DATABASE_URL | SQLite database path | sqlite+aiosqlite:///./draft.db |
FRONTEND_URL | Frontend URL for CORS | http://localhost:5173 |
GIT_REPO_PATH | Git repository root | Project root |
BASE_BRANCH | Base branch for worktrees | main (fallback: master) |
AUTH_SECRET_KEY | JWT signing secret | Dev default (change in production) |
Frontend (frontend/.env)
cp frontend/.env.example frontend/.env
| Variable | Description | Default |
|---|
VITE_BACKEND_URL | Backend API URL | http://localhost:8000 |
AI Provider Keys
Set the API key for your chosen LLM provider (used by the planner):
# Anthropic (recommended)
export ANTHROPIC_API_KEY=sk-ant-...
# OpenAI
export OPENAI_API_KEY=sk-...
# AWS Bedrock
export AWS_ACCESS_KEY_ID=...
export AWS_SECRET_ACCESS_KEY=...
export AWS_DEFAULT_REGION=us-east-1
AI executor CLIs (Claude Code, Cursor) use their own authentication. The API keys above are only for the planner’s LLM calls.
Executor Types
| Type | CLI Required | Mode | After Execution |
|---|
CLAUDE | claude | Headless | Auto-verifies |
CURSOR_AGENT | cursor | Headless | Auto-verifies |
CURSOR | cursor | Interactive | Needs human review |
YOLO Mode (Claude Only)
When enabled, Claude CLI runs with --dangerously-skip-permissions, allowing it to execute without confirmation prompts:
executor_config:
executor_type: CLAUDE
yolo_mode: true
yolo_allowlist:
- /path/to/trusted/repo
Only enable YOLO mode for repositories you trust. The AI agent will have unrestricted access to run commands in the worktree.
