You describe what you want. The AI builds it. The workflow gets saved so you never rebuild it.
This is a GitHub template repository. Here's how to use it:
- Click the green "Use this template" button at the top of this page
- Select "Create a new repository"
- Name your repository and create it
- Clone your new repository locally
- Continue with setup below
Use this if you want to pull future updates from this template.
git clone <this-repo-url> my-project
cd my-project
rm -rf .git && git init # Start fresh git history# 1. Set up environment
cp .env.example .env
# Edit .env with your API keys (at minimum: ANTHROPIC_API_KEY)
# 2. Verify
python execution/sync_agent_files.py --check
# 3. Open in Claude Code
claude --dangerously-skip-permissions
# This flag allows Claude to run scripts without confirmation promptsThree layers:
- Directives - Plain English instructions (
directives/*.md) - Orchestration - AI makes decisions
- Execution - Python scripts do the work (
execution/*.py)
First time: "Build a workflow that scrapes competitor prices" -> AI researches, builds, saves
Every time after: "Scrape competitor prices" -> AI finds directive, runs script, done
your-project/
├── CLAUDE.md # AI instructions (source of truth)
├── AGENTS.md # Mirror for other AI tools
├── GEMINI.md # Mirror for Gemini
├── REFERENCE.md # Deep documentation
├── directives/ # Your workflows
├── execution/ # Your scripts
└── .env # API keys (gitignored)
- REFERENCE.md - Full documentation (architecture, error handling, advanced features)
- directives/csv_to_json.md - Example: Convert CSV files to JSON (no API needed)
- directives/weather_lookup.md - Example: Look up weather via API (shows dotenv pattern)
- directives/_TEMPLATE.md - Template for new workflows
- execution/_TEMPLATE.py - Template for new scripts
Based on Nick Saraev's DOE framework. Refined for clarity and self-improvement.
