agentleFS
Sign inStart free
Smirnov-LabsSmirnov-Labs/claude-skills/skills/codex-review/SKILL.md

codex-review

Skill6 starsChanged 8 days agoMITView on GitHub

Use when you want an independent second opinion from Codex on a plan, a design spec or a code change, at a major checkpoint or before merging a non-trivial PR, when asked for a "codex review" or "second opinion", or when you are about to write a `codex exec` command or poll for Codex output by hand.

91 other projects use a version of this.

Adopted by

Most recently taken up by

Used with

Projects with it also use addskill, apply-firmware and 10 more.

Other files people use alongside this one.

Files found in the same public projects as a version of this one, more often than they turn up anywhere else. Each project counts once, large collections count for less, and this file's own repository is left out.
  • ryanportfolioSkill

    addskill

    Create, import, update, or install repo or personal skills: runtime ownership, resources, discovery checks.

    Also in 6 projects that use this

  • ryanportfolioSkill

    apply-firmware

    Bring current folder (plain folder, existing repo, old or partial firmware copy) up to full Harness Firmware in place. Mirror external repo → adopt-repo.

    Also in 6 projects that use this

  • +10 more from ryanportfolio

The file

We split this skill into the rules it gives an agent

6 rules. 0 are common, 3 are distinctive, and 3 are very specific to this project.

  • 0

    Common

    Most of the 79 edited versions we read keep these.

  • 3

    Distinctive

    This author's own choices. Worth a look.

  • 3

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Very specific:

    Run it with run_in_background: true.

    line 17
  2. Distinctive:

    Commit before a diff review.

    line 30
  3. Very specific:

    Read review.md. The exit status is a summary, not the review.

    line 52
  4. Distinctive:

    Check each Critical and Major finding against the code or the plan.

    line 53
  5. Distinctive:

    If you changed anything, commit and review the new HEAD.

    line 55
  6. Very specific:

    If Codex rejects the pin as "not supported", the script runs once more with the model from ~/.codex/config.toml and prints a note.

    line 70

“Common” is measured: kept by at least 6 in 10 of the 79 edited versions of this file we could read. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

Codex review

codex-review runs one read-only Codex review and prints a verdict. Hand-built codex exec lines kept failing in the same ways:

  • stderr went to /dev/null, so a crash read as an empty review;
  • the model pins went stale;
  • agents ran --help to find the flags again;
  • sleep loops waited for the output.

Run it

Run it with run_in_background: true. A high-effort review takes several minutes, and you are told when it exits. Don't poll it and don't wrap it in a loop.

To reviewCommand
An implementation plancodex-review plan docs/plans/feature.md
A design speccodex-review spec docs/specs/feature.md
This branch's commitscodex-review diff (origin/<default branch>...HEAD)
Other commitscodex-review diff main..HEAD (a single ref R means R..HEAD)
A PRcodex-review diff --pr 123, from a checkout of the PR branch
Your own prompt, as iscodex-review prompt prompt.md
  • --context FILE (repeatable) adds what the work must meet: the spec for a diff, the constraints for a plan.
  • --focus "..." says what to check hardest.
  • Commit before a diff review. Uncommitted changes are not in the diff.
  • codex-review --help has the rest (--out, --timeout, --model, --effort, --dry-run).
  • codex-review is on PATH where skillsync installed this skill. In a plugin install, call <this skill's base directory>/bin/codex-review by its full path, from the project directory.

Read the result

It prints the prompt, review and log paths, then a verdict:

verdict FINDINGS · critical 0 · major 1 · minor 3 · 6m12s
  [MAJOR] The merge runs before the MERGED check
ExitMeaning
0No Critical or Major findings (prompt mode: Codex answered)
1Critical or Major findings, or a review it could not parse
2No usable review: Codex failed, timed out, or said it could not review
3Usage error

Then:

  1. Read review.md. The exit status is a summary, not the review.
  2. Check each Critical and Major finding against the code or the plan. Codex can be wrong.
  3. Fix the findings you confirmed. List the Minor ones in your report.
  4. If you changed anything, commit and review the new HEAD.

On exit 2, read the log tail it prints. Don't report "Codex found nothing".

Model

The pin is gpt-6-astra, checked on 2026-11-05 with codex-cli 0.166. CODEX_REVIEW_MODEL or --model overrides it.

ModelUse
gpt-6-astraThe pin. Frontier model, 272K context.
gpt-6.1-sol500K context. Used on its own when the prompt passes 600,000 characters.
gpt-6.1-terra, gpt-6-sol, gpt-5.6-*Older or balanced models. Don't pin them.
gpt-6-luna, gpt-5.6-lunaFast and cheap. Too light for a review.

If Codex rejects the pin as "not supported", the script runs once more with the model from ~/.codex/config.toml and prints a note. Then update the pin: codex debug models | jq -r '.models[] | select(.visibility=="list") | .slug'.

Common mistakes

MistakeWhat happens
codex exec ... 2>/dev/nullA crash leaves an empty review, and the work goes on as if Codex agreed.
A sleep loop that waits for review.mdWasted turns. The background run tells you when it ends.
Exit 0 read as approvalExit 0 means no Critical or Major findings were parsed. Read the review.
Reviewing uncommitted workThe review names no SHA, and the diff misses the edits.

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

Check your setup, free