Codex
Claude Code plugin marketplace of my personal agent skills
npx -y skills add mwgg/skills --skill codexAssembled from the repository path, not quoted from the project. Check it against their README if it does not work.
2 things to look at
- 26 days oldThe repository was created 26 days ago. New is not bad, but a brand new repository carrying a familiar-sounding name is the shape a typosquat arrives in, and there has been no time for anyone else to find a problem with it.
- 1 stars1 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.
What its author says it does
Copied from the file, not written here
Get an independent second-opinion review from the Codex CLI on a plan or a round of code changes, returning a clear verdict. Use when the user asks to "run codex", "codex review", "codex check", "review this with codex", "second opinion", "review my plan", or wants an outside reviewer over the current changes — in any project, git or not. Two modes — plan review (before code, verdict SOLID/WEAKNESSES) and code review (after changes, verdict APPROVE/COMMENTS). Advisory, never a hard gate. Other skills (e.g. gsd) call this to add review checkpoints.
SKILL.md
5.7 KB, ~1.4k tokens by cl100k_base, as published. Nobody here has run it
codex — independent Codex review (plan + code)
A universal, on-demand wrapper around the codex CLI that gives you an independent
second opinion on your work. It is project-agnostic — it works in any directory,
git repo or not — and not tied to any particular workflow. Invoke it directly when
the user asks for a Codex review, or programmatically from another skill (e.g. gsd
runs it at its plan and code checkpoints).
Codex is advisory, never blocking. Every call is best-effort: if the codex CLI is
missing, out of quota, or times out, the script prints a note and exits 0 — you report
"Codex review skipped" and continue. Codex never gates completion. Treat its output as a
strong outside opinion: adopt what's valid, briefly note what you reject with a reason,
move on.
The driver is the codex.sh bundled with this skill — installed as a plugin it lives at
${CLAUDE_PLUGIN_ROOT}/codex.sh; copied as a plain skill folder, at <skill-dir>/codex.sh.
It has two modes, plan and code. The examples below write codex.sh for brevity —
substitute the bundled path. <root> = the project directory Codex is allowed to read.
Requirements: the codex CLI on PATH and available quota. The driver uses the
Codex-specific model ID gpt-5.6-sol by default. Do not shorten this to gpt-5.6: that is
a different, unsupported model ID for ChatGPT-account Codex auth. Override deliberately with
CODEX_REVIEW_MODEL=<model-id> when needed. The review is read-only and can take several
minutes; the diff handed to Codex is capped at ~120k chars, but Codex has read access to the
whole project and is told it may open full files.
Mode 1 — Plan review (before code is written)
Have Codex critique a design/plan document for weak points before implementation.
Verdict: VERDICT: SOLID or VERDICT: WEAKNESSES + a numbered list.
# Explicit plan file(s):
codex.sh plan <root> path/to/PLAN.md [more-plan-files...]
# Or, with no files given, auto-discover GSD plan artifacts under <root>/.gsd/
# (RESEARCH.md, REQUIREMENTS.md, ROADMAP.md, FORKS.md, S*-PLAN.md):
codex.sh plan <root>
Codex looks for: missing tasks/requirements, wrong or circular dependencies, tasks too coarse to verify, vague acceptance criteria, risky assumptions, scope gaps, reuse the plan ignored, and parallelization hazards.
Act on the verdict: SOLID → note it and proceed. WEAKNESSES → for each point decide
if it's a real gap; apply the valid ones to the plan, and briefly record anything you
deliberately reject with a reason. You may re-run once to confirm; don't loop — stop after
at most two rounds and log any residual. The goal is a stronger plan, not a perfect one.
Mode 2 — Code review (after a round of changes)
Have Codex review a diff against the instruction it was meant to satisfy.
Verdict: VERDICT: APPROVE or VERDICT: COMMENTS + a numbered list.
-
Assemble the changed-file list (newline-separated absolute paths). Any accurate source works:
# From git — changed tracked files AND new untracked ones, as absolute paths # (plain `git diff --name-only` alone would miss newly-created files): { git -C <root> diff --name-only; git -C <root> ls-files --others --exclude-standard; } \ | sed "s#^#<root>/#" | sort -u > /tmp/codex-files.txt # For an already-committed range (e.g. a whole feature branch — pair with CODEX_DIFF_BASE below): # git -C <root> diff --name-only <base>..HEAD | sed "s#^#<root>/#" > /tmp/codex-files.txt # …or an explicit list of the files you changed, one per line. -
Write the instruction the diff must satisfy (the user's request, a slice goal, the success criteria — the verdict is judged against it):
printf '%s\n' "<the instruction / goal for this round>" > /tmp/codex-instr.md -
Run the review:
codex.sh code <root> /tmp/codex-instr.md /tmp/codex-files.txt- Non-git projects work out of the box (new files are diffed against
/dev/null). - To review already-committed changes (e.g. a whole feature branch vs its base,
where
git diff HEADis empty), set the base ref:CODEX_DIFF_BASE=<base-commit-or-branch> \ codex.sh code <root> /tmp/codex-instr.md /tmp/codex-files.txt
- Non-git projects work out of the box (new files are diffed against
-
Act on the verdict:
APPROVE→ done.COMMENTS→ treat each numbered item: fix what's valid, note anything you disagree with and why, then optionally re-run. Codex reviews only code/doc content, not repo hygiene or version-control practices.
Notes
- To verify model access, run a tiny
codex exec --model gpt-5.6-solprobe and check both the startup banner'smodel:line and a successful response. CLI argument acceptance alone does not prove backend access. - Out-of-quota / timeout / exec errors print to stderr and yield no verdict — the same non-blocking behavior in both modes. Report "review skipped" and continue.
- For diff-based code review in a git repo, prefer running
git initearly on greenfield projects so tracked-file diffs are available; non-git still works via--no-index.
What ships with it: 4 files
11.8 KB alongside SKILL.md, 1 of them executable
.claude-plugin/
- plugin.json713 B
commands/
- codex.md1.1 KB