crew-work

What problem does it solve?

It solves the problem of implementing requested code changes without a reliable review-and-verification loop, which can lead to missed edge cases or untested regressions.

Core Features & Use Cases

• Round-based self-review loop: Performs up to N iterations (default 3) of implement → verify → review → fix to improve correctness.
• Focused planning context: Optionally uses .codex/crew/seed-latest.md as planning context when it is in plan mode.
• Safer workspace handling: Preserves unrelated user changes by checking git status --short --untracked-files=all and avoiding staging unrelated modifications.
• Verification discipline: Runs the narrowest meaningful verification first and broadens tests only when shared behavior changes.
• Change acceptance reporting: Writes an acceptance summary to .codex/crew/runs/<run-id>/accepted.md including verification results.

Quick Start

Ask the AI to run /crew-work (optionally with -r 2) to implement your requested change, verify it with the narrowest tests, and perform a self-review loop until the round limit is reached.