Documentation
README
conductor — autonomous doer→gate→judge loop
Iterate on one work item until its acceptance criteria are met and provable,
without human input mid-loop — this skill runs unattended (the outer loop
dispatches it headless via cron; see .ai/docs/outer-loop.md). Human review
happens exactly once, at the end: ship a gated PR — never merge.
Architecture context: .ai/docs/loop-system.md; the deterministic helpers
live in packages/harness (see its AGENTS.md).
Announce at start: "Using the conductor skill — autonomous loop on {work item}."
Autonomous decision protocol — no mid-loop questions
The loop never waits on a human. Concretely:
This is the opening of the README. Read the full README on GitHub.