codex-review_skill

This skill enables cross-agent code review with Codex, guiding setup, planning, execution, and state management to improve code quality.
  • Python

30

GitHub Stars

3

Bundled Files

2 months ago

Catalog Refreshed

4 months ago

First Indexed

Readme & install

Copy the install command, review bundled files from the catalogue, and read any extended description pulled from the listing source.

Installation

Preview and clipboard use veilstrat where the catalogue uses aiagentskills.

npx veilstrat add skill artwist-polyakov/polyakov-claude-skills --skill codex-review

  • .gitignore12 B
  • README.md7.3 KB
  • SKILL.md8.8 KB

Overview

This skill implements a cross-agent code review workflow where Claude coordinates and Codex performs the technical review. It provides shell scripts to initialize sessions, submit plans and code for review, manage review state, and handle iteration limits and escalations. The scripts live in a scripts/ directory adjacent to the skill files and operate on a .codex-review workspace inside the project.

How this skill works

You initialize or attach a Codex session, write a plan or describe code changes, and call the provided scripts to send that content to Codex for review. Codex reads the project files directly in the same directory, writes verdicts and notes into .codex-review, and returns status via exit codes. State and limits are managed with codex-state.sh, and the review script enforces rules like never calling codex exec directly.

When to use it

  • When you want a cross-agent workflow: Claude drives process, Codex reviews code.
  • Before implementing a feature: run a plan-mode review to get technical feedback.
  • After implementation: submit a code review describing what you changed and why.
  • When review iteration limits are reached and you need escalation options.
  • To maintain a persistent review session across multiple iterations.

Best practices

  • Store or set CODEX_SESSION_ID in .codex-review/config.env or use init to create a session.
  • Always include the full plan text when using plan mode; Codex cannot access files outside the project.
  • Describe clearly what you implemented and why when submitting code reviews; do not send git diffs.
  • Update phase with codex-state.sh set phase implementing before starting implementation.
  • Follow the ESCALATE procedure when iterations run out: summarize notes and ask the user how to proceed.

Example use cases

  • Plan review: draft an architecture/implementation plan and submit it for technical review before coding.
  • Feature delivery: implement JWT auth, then call the code review script describing files changed and design decisions.
  • Iteration control: increment or remove max-iter when Codex requests many small changes.
  • Session management: persist a session across a team’s review cycle by setting session_id in config.env.
  • Issue escalation: gather note files and present a concise summary to the user for decision when limit hits.

FAQ

Run the init command with a short task description or set CODEX_SESSION_ID in .codex-review/config.env.

What do exit codes mean?

0 means APPROVED or CHANGES_REQUESTED (check verdict); 1 is ERROR (check session_id); 2 is ESCALATE (follow escalate flow); 3 is NO_SESSION (create or set session).

Can I send git diffs to Codex?

No. Do not send diffs — Codex reads the repository directly. Describe what you changed and why, and Codex will inspect files.

Built by
VeilStrat
AI signals for GTM teams
© 2026 VeilStrat. All rights reserved.All systems operational