Autoreview

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/second-opinion paths that receive a generated diff prompt, not the full Codex review-mode runtime.

Use when:

• user asks for Codex review / autoreview / second-model review
• after non-trivial code edits, before final/commit/ship
• reviewing a local branch or PR branch after fixes

Contract

• Treat review output as advisory. Never blindly apply it.
• Verify every finding by reading the real code path and adjacent files.
• Read dependency docs/source/types when the finding depends on external behavior.
• Reject unrealistic edge cases, speculative risks, broad rewrites, and fixes that over-complicate the codebase.
• Prefer small fixes at the right ownership boundary; no refactor unless it clearly improves the bug class.
• Keep going until the selected review path returns no accepted/actionable findings.
• If a review-triggered fix changes code, rerun focused tests and rerun the review helper.
• Default to Codex review. If Codex is unavailable or exits with an error, the helper falls back to the first configured CLI from claude -p, pi -p, opencode run, droid exec, or copilot. Prefer Codex for final closeout because it uses native review mode; non-Codex reviewers use a Codex-inspired generated diff prompt. The helper runs nested Codex review in yolo/full-access mode by default; use --no-yolo only when intentionally testing sandbox behavior.
• Stop as soon as the review command/helper exits 0 with no accepted/actionable findings. Do not run an extra direct codex review just to get a nicer "clean" line, a second opinion, or clearer closeout wording.
• Treat the helper's successful exit plus absence of actionable findings as the clean review result, even if the underlying Codex CLI output is terse.
• If rejecting a finding as intentional/not worth fixing, add a brief inline code comment only when it explains a real invariant or ownership decision that future reviewers should know.
• Do not push just to review. Push only when the user requested push/ship/PR update.
• For OpenClaw maintainers, keep autoreview validation Crabbox/Testbox-aware when maintainer validation mode is enabled (OPENCLAW_TESTBOX=1 or AUTOREVIEW_OPENCLAW_MAINTAINER_VALIDATION=1). A review pass may inspect files and run cheap non-Node probes, but it must not start local pnpm, Vitest, tsgo, npm test, or node scripts/run-vitest.mjs from a Codex/worktree review unless the operator explicitly requested local proof. For runtime proof, use existing evidence or route through Crabbox/Testbox and report the id. Do not apply this rule to ordinary contributors who do not have maintainer Testbox access.

Pick Target

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; do not force --mode local / --uncommitted just because the helper docs mention dirty work first. A clean --uncommitted review only proves there is no local patch.

Branch/PR work:

git fetch origin
codex review --base origin/main

Do not pass any prompt with --base. Some Codex CLI versions reject both inline and stdin prompt forms, including the helper's codex review --base <ref> -, with --base <BRANCH> cannot be used with [PROMPT]. If the helper hits this error, run plain codex review --base <ref> and report that the helper prompt injection was skipped.

If an open PR exists, use its actual base:

base=$(gh pr view --json baseRefName --jq .baseRefName)
codex review --base "origin/$base"

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.

Parallel Closeout

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 "<focused test command>"

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.

Context Efficiency

Codex review is usually noisy. Default to a subagent filter when subagents are available. Ask it to run the review and return only:

• actionable findings it accepts
• findings it rejects, with one-line reason
• exact files/tests to rerun

Run inline only for tiny changes or when subagents are unavailable.

Helper

Bundled helper:

.agents/skills/autoreview/scripts/autoreview --help

The helper:

• chooses dirty --uncommitted first
• otherwise uses current PR base if gh pr view works
• otherwise uses origin/main for non-main branches
• auto-runs PNPM_CONFIG_PM_ON_FAIL=ignore PNPM_CONFIG_VERIFY_DEPS_BEFORE_RUN=false PNPM_CONFIG_OFFLINE=true pnpm run check in parallel when a repo has package.json, pnpm-lock.yaml, node_modules, and a check script; disable with AUTOREVIEW_AUTO_TESTS=0
• use --mode commit --commit <ref> for already-committed work, especially clean main after landing
• should be left in --mode auto or forced to --mode branch for PR/branch work; do not force --mode local after committing
• supports --reviewer codex|claude|pi|opencode|droid|copilot|auto; auto means Codex first
• supports --fallback-reviewer auto|claude|pi|opencode|droid|copilot|none; default is configured CLI fallback
• falls back only when Codex is unavailable or exits nonzero, not when Codex reports findings
• writes only to stdout unless --output or AUTOREVIEW_OUTPUT is set
• supports --dry-run, --parallel-tests, and commit refs
• runs nested review with --dangerously-bypass-approvals-and-sandbox --sandbox danger-full-access by default
• injects maintainer-only OpenClaw validation policy into native Codex review when OPENCLAW_TESTBOX=1 or AUTOREVIEW_OPENCLAW_MAINTAINER_VALIDATION=1, so local memory-heavy Node/Vitest checks are avoided in favor of Crabbox/Testbox proof
• branch mode may fail on Codex CLI versions that reject --base plus the helper's stdin prompt; on that exact parser error, rerun plain codex review --base <ref> instead of falling back to a non-Codex reviewer
• keeps accepting --full-access; use --no-yolo or AUTOREVIEW_YOLO=0 to opt out
• still accepts legacy CODEX_REVIEW_* env vars when the matching AUTOREVIEW_* var is unset
• prints autoreview clean: no accepted/actionable findings reported when the selected review command exits 0

Final Report

Include:

• review command used
• tests/proof run
• findings accepted/rejected, briefly why
• the clean review result from the final helper/review run, or why a remaining finding was consciously rejected

Do not run another Codex review solely to improve the final report wording. If the final helper run exited 0 and produced no accepted/actionable findings, report that exact run as clean.