// Use when ClawHub needs Codex review, autoreview, second-model review, or a final advisory review gate before commit, PR update, ship, or maintainer handoff.
Use for ClawHub staff moderation actions with the repo-local clawhub-mod tool: unhide skills, ban or unban users, change roles, and verify moderation state.
Use when reviewing, triaging, validating, or discussing ClawHub GitHub issues or pull requests, including author context, CI, UI proof, evidence, labels, close decisions, and maintainer handoff.
Use when ClawHub UI changes need visual proof, before/after comparison, new-feature screenshots, temporary Playwright scenarios, or Crabbox desktop recordings.
Use when ClawHub needs remote Linux validation, CI-parity checks, broad Bun gates, hosted-service checks, desktop/VNC inspection, or Crabbox lease cleanup.
| name | autoreview |
| description | Use when ClawHub needs Codex review, autoreview, second-model review, or a final advisory review gate before commit, PR update, ship, or maintainer handoff. |
Run Codex's built-in code review as a closeout check. This is code review
(codex review), not Guardian auto_review approval routing.
Codex native review mode performs best and is recommended. Non-Codex reviewers are fallback or second-opinion paths that receive a generated diff prompt, not the full Codex review-mode runtime.
Use when:
claude -p, pi -p, opencode run, droid exec, or
copilot.codex review just
to get a nicer clean line, a second opinion, or clearer closeout wording.Pick the smallest proof that matches the touched surface:
| Touched surface | Usual proof |
|---|---|
| Formatting/lint/static repo health | bun run ci:static |
| Unit-tested source behavior | focused bunx vitest run ..., then bun run ci:unit when PR-ready |
| Convex code | read convex/_generated/ai/guidelines.md first; run focused tests and the deploy/typecheck path that covers the change |
| Packages/CLI/mod tool | bun run ci:packages or the package-specific verify script |
| Runtime/build/package surface | bun run ci:types-build, bun run ci:e2e-http, or the matching broader gate |
| UI behavior | use clawhub-ui-proof with proof:ui; publish proof before final PR comments when needed |
| Linux/CI-parity validation | use crabbox, normally through the repo scripts |
For Convex query or schema work, apply the repo's Convex rules: prefer indexes
over .filter() scans, use cursor-based backfills for data shape changes, and
verify with the repo's Convex/typecheck path before claiming deploy safety.
Dirty local work:
codex review --uncommitted
Use this only when the patch is actually unstaged/staged/untracked in the
current checkout. For committed, pushed, or PR work, point Codex at the commit
or branch diff instead. A clean --uncommitted review only proves there is no
local patch.
Branch/PR work:
git fetch origin
codex review --base origin/main
If an open PR exists, use its actual base:
base=$(gh pr view --json baseRefName --jq .baseRefName)
codex review --base "origin/$base"
Do not pass a prompt with --base. Some Codex CLI versions reject
codex review --base <ref> - with --base <BRANCH> cannot be used with [PROMPT]. If that happens, rerun plain codex review --base <ref> and report
that prompt injection was skipped.
Committed single change:
codex review --commit HEAD
or with the helper:
.agents/skills/autoreview/scripts/autoreview --mode commit --commit HEAD
Use commit review for already-landed or already-pushed work on main.
Reviewing clean main against origin/main is usually an empty diff after
push. For a small stack, review each commit explicitly or review the branch
before merging with --base.
Format first if formatting can change line locations. Then it is OK to run tests and review in parallel:
.agents/skills/autoreview/scripts/autoreview --parallel-tests "bun run ci:static"
Tradeoff: tests may force code changes that stale the review. If tests or review lead to code edits, rerun the affected tests and rerun review until no accepted/actionable findings remain. Once that rerun exits cleanly, stop; do not spend another long review cycle on redundant confirmation.
Codex review is usually noisy. Default to a subagent filter when subagents are available. Ask it to run the review and return only:
Run inline only for tiny changes or when subagents are unavailable.
Bundled helper:
.agents/skills/autoreview/scripts/autoreview --help
The helper:
--uncommitted firstgh pr view worksorigin/main for non-main branchesbun run ci:static in parallel when the repo has package.json,
bun.lock, node_modules, and a ci:static script; disable with
AUTOREVIEW_AUTO_TESTS=0--mode commit --commit <ref> for already-committed work, especially
clean main after landing--mode auto or forced to --mode branch for PR/branch
work; do not force --mode local after committing--reviewer codex|claude|pi|opencode|droid|copilot|auto; auto
means Codex first--fallback-reviewer auto|claude|pi|opencode|droid|copilot|none--output or AUTOREVIEW_OUTPUT is set--dry-run, --parallel-tests, and commit refs--dangerously-bypass-approvals-and-sandbox --sandbox danger-full-access by default; use --no-yolo or AUTOREVIEW_YOLO=0 to opt
outautoreview clean: no accepted/actionable findings reported when the
selected review command exits 0 and no accepted/actionable findings are
reportedInclude:
Do not run another Codex review solely to improve final wording. If the final helper run exited 0 and produced no accepted/actionable findings, report that exact run as clean.
gh run view <run-id> --json jobs;
use PR rollup only for final mergeability.origin/main..HEAD; drop CI-fix commits already
upstream before pushing.