Autoreview

Autoreview

Run this before telling the user a task is complete, a bug is fixed, or a feature works. Never report "done" without doing this first.

Universal checks (every repo, every task)

• git status -sb + read the full diff. Review it like a stranger's PR — not your own work.
• Reread the original request. Confirm the change does that, not something adjacent to it.
• Run the actual test suite / linter for this repo. Don't say "tests pass" without executing them.
• Check for stray debug output, commented-out code, leftover TODOs.
• Confirm no unrelated files changed.
• If the task touched config, secrets, auth, billing, or migrations — re-read that diff twice.

Project-specific checks

Look for a .autoreview.md file in the repo root.

• If found: treat its checks as mandatory, layered on top of the universal list above.
• If not found: run universal checks only, and say so in your report ("no project-specific

rules found — ran universal checks only").

Reporting format

Every completion report includes:

• Files changed (from git status)
• Test command run + result (paste real output, not a summary)
• Project-specific checks run (or note that none were found)
• Anything skipped, and why

Failure handling

If a check fails or you're not confident: say so explicitly and do not report the task as done. A partial fix reported honestly is worth more than a clean report that isn't true.

Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: EOEboh
• Source: EOEboh/agent-skills
• License: MIT

Install and usage instructions live in the source repository linked above.