autopilot

A workflow for autonomous, test-driven development using Claude Code. Write a PRD, convert it to tasks, then let Claude implement everything using TDD while you sleep.

Inspired by the Ralph Wiggum technique, autopilot includes its own built-in loop mechanism - no external plugins required.

Curious where the project is headed and why it works the way it does? See VISION.md for the thesis, design principles, and roadmap.

Credits

This workflow gratefully builds on contributions from the community:

• Ralph Wiggum by Geoffrey Huntley — The autonomous loop approach
• ai-dev-tasks by Ryan Carson — PRD and task generation prompts
• Matt Pocock — Ralph workflow tips
• ralph-playbook by Clayton Farr — Additional workflow tips

Quick Start (5 minutes)

New project? Follow this flow:

Already have a task file?

**No autopilot.json yet?** /autopilot (and the tests/lint/entropy/analyze modes) now auto-initialize with detected defaults (init --force) and continue in the same session — you no longer have to run /autopilot init first. Review the generated autopilot.json afterward and adjust feedback loops if needed. Greenfield projects whose feedback-loop tooling is scaffolded by an early requirement are expected and supported; those loops stay enabled and activate once that requirement runs.

Need help? See Troubleshooting or run /autopilot --help.