autopilotlisted
Install: claude install-skill vladyslav-betterme/autopilot-skill
# Autopilot
An autonomous loop that is portable because it **asks the work what «done»
means** instead of assuming — and does not stop until that answer is satisfied.
Three claims this skill exists to make impossible:
- «It works» when nothing ran it.
- «It's done» when the only judge was the one who did it.
- «I worked on it» with no record of what was tried, what failed, and why.
The subject does not matter. A refactor, a migration, a report, a dataset, a
render farm, a cluster: the loop below is identical and only the **check**
differs.
## 0. Bootstrap — once per project
`<skill>` below is the directory this file sits in — `.agents/skills/autopilot`
after a CLI install, or wherever it was cloned. Run everything from the PROJECT
root.
```bash
node <skill>/scripts/discover.mjs # checks, memory homes, danger signals
node <skill>/scripts/bootstrap.mjs # creates the ledger if absent
node <skill>/scripts/skills.mjs # the skill library — pick this project's niches
node <skill>/scripts/tools.mjs # what this machine can already REACH: MCP servers, plugins
```
`discover` prints the project's own check command. **That command is the
definition of done for every iteration below.**
If it prints `checks: []`, **do not believe it yet.** Discovery only reads files
that declare a check; a project's real check is often named in prose. Grep every
path it listed in `memoryHomes`, plus the README and CI config, for a runnable
command — one project's `ch