Review changes with autoreview

Overview

Use autoreview to get an independent review of local changes or a committed branch.

Run a review

When independent review is requested, follow the shared skill setup and read the installed skill. From the repository root:

python3 "$HOME/.agents/skills/autoreview/scripts/autoreview" --mode local --model codex=gpt-6-astra

The helper requires Python 3.10+ and an authenticated reviewer CLI; image review also requires Pillow. Local mode includes untracked, staged, and unstaged changes. Use --dry-run to check preparation and --max-priority P2 when that scope is requested; the default threshold is P0.

For a committed branch, use --mode branch --base upstream/main only after verifying and fetching the intended upstream. For an existing or dependent PR, select its actual target branch. Pass --base explicitly: the default origin/main can point to a fork. Verify findings before applying them, keep reports outside the repository, and do not bypass isolation or treat an absent report as clean. See the installed skill for options and results.

Upstream provenance

openclaw/agent-skills owns the implementation, instructions, and tests. Enterprise keeps a Markdown entrypoint and the upstream MIT license.

Sync the skill

Contribute and validate shared changes upstream first. After active reviews finish, update the shared checkout. Copy-mode installations need reinstalling: python3 scripts/install-skills --mode copy --force autoreview. Review runs do not update code. If the entrypoint changes, copy upstream's skills/autoreview/references/repository-entrypoint.md to .agents/skills/autoreview/SKILL.md. Do not restore local helper or test copies.