GitHub - shimo4228/codex-review: Cross-model code review — a read-only Claude Code skill wrapping the OpenAI Codex CLI for a decorrelated second opinion on a diff · GitHub
/" data-turbo-transient="true" />
Skip to content
Search or jump to...
Search code, repositories, users, issues, pull requests...
-->
Search
Clear
Search syntax tips
Provide feedback
--><br>We read every piece of feedback, and take your input very seriously.
Include my email address so I can be contacted
Cancel
Submit feedback
Saved searches
Use saved searches to filter your results more quickly
-->
Name
Query
To see all available qualifiers, see our documentation.
Cancel
Create saved search
Sign in
/;ref_cta:Sign up;ref_loc:header logged out"}"<br>Sign up
Appearance settings
Resetting focus
You signed in with another tab or window. Reload to refresh your session.<br>You signed out in another tab or window. Reload to refresh your session.<br>You switched accounts on another tab or window. Reload to refresh your session.
Dismiss alert
{{ message }}
shimo4228
codex-review
Public
Notifications<br>You must be signed in to change notification settings
Fork
Star
main
BranchesTags
Go to file
CodeOpen more actions menu
Folders and files<br>NameNameLast commit message<br>Last commit date<br>Latest commit
History<br>1 Commit<br>1 Commit
scripts
scripts
skills/codex-review
skills/codex-review
CHANGELOG.md
CHANGELOG.md
LICENSE
LICENSE
README.md
README.md
llms-full.txt
llms-full.txt
llms.txt
llms.txt
View all files
Repository files navigation
codex-review
An Agent Skill that adds one cross-model seam to a code review chain: a thin, read-only wrapper around the OpenAI Codex CLI (codex review) so a different model family reviews your diff and catches blind spots that an author and a same-model reviewer share.
Install
Claude Code
# Copy into your global skills directory<br>cp -r skills/codex-review ~/.claude/skills/codex-review
SkillsMP
/skills add shimo4228/codex-review
Requires the Codex CLI installed and authenticated (codex login or codex doctor) — the skill fails fast with a fallback message if it isn't.
How It Works
This is a decorrelation seam, not a throughput tool — use your agent's own<br>subagents/parallel workflows for throughput; reach for a second model only<br>where a different model's judgment adds something a same-model reviewer can't.
skills/codex-review/codex-review.sh $ARGUMENTS
Invocation<br>Scope
/codex-review<br>current branch vs auto-detected base (main/master/…) — PR-style
/codex-review --uncommitted<br>staged + unstaged + untracked — pre-commit
/codex-review --base<br>vs an explicit base branch
/codex-review --commit<br>a single commit
/codex-review -m<br>pick a Codex model (combine with any row)
/codex-review "focus on the auth changes"<br>prompt-driven review of the working tree
Scope and prompt are mutually exclusive (a codex-cli constraint, ≥ 0.142): a<br>scoped review (--uncommitted / --base / --commit, or the default) uses<br>Codex's built-in review instructions and takes no custom prompt; a bare prompt<br>drives a working-tree review with no scope flag. -m/--model may accompany<br>either.
Read-only by construction. The script uses codex review (never codex exec -p yolo) and only forwards the allowlisted flags above — any other flag<br>(a future --write, a -c config override) is rejected with exit 64. The<br>invariant is enforced in the script, not assumed of the Codex CLI.
After Running — fold, don't dump
Codex output is untrusted input to a review decision the calling agent owns ,<br>not a verdict to relay verbatim:
Verify each finding before acting on it — drop what you can disprove, keep<br>what you confirm.
Treat a confirmed critical finding as a stop signal, same as any other<br>reviewer in the chain.
Run it in parallel with same-model reviewers, then merge verdicts.
When It Triggers
Before commit on a non-trivial feature or fix, alongside your other reviewers
When you want a second opinion from a non-Claude model on a diff
High-stakes or error-prone changes where decorrelated review pays off
High-stakes prose diffs before publishing (README, paper, article) —<br>prompt-driven mode with writing-focused instructions; scoped modes run<br>Codex's code-review instructions, which fit prose poorly
Skip it for trivial edits, throwaway scripts, or when Codex isn't authenticated.
Failure Modes
exit 3 — codex CLI missing / not installed → continue with same-model reviewers only
exit 4 — not inside a git repository → cannot diff
Auth not configured → codex review errors; run codex login (or codex doctor)
Syncing from the harness
The canonical copy of this skill lives in the author's live Claude Code harness. This repository is a one-way publication mirror:
scripts/sync-from-local.sh --dry-run # report differences only<br>scripts/sync-from-local.sh # apply to working tree (never commits)
About this skill
This skill implements a single cross-model seam in a broader multi-agent<br>orchestration principle: same-model agents scale...