codex-code-review-fix

Reviews code changes via Codex CLI and iteratively fixes confirmed findings until approval.

34|5|Updated Mar 8, 2026
One-click install
npx skills add https://github.com/AndreevED/claude-code-external-review --skill codex-code-review-fix-andreeved
Or copy as Structured Prompt for Agent
Please help me install this Agent Skill.
Skill: codex-code-review-fix
Source: https://github.com/AndreevED/claude-code-external-review/tree/main/skills/codex-code-review-fix
Command: npx skills add https://github.com/AndreevED/claude-code-external-review --skill codex-code-review-fix-andreeved

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill includes references (resource) components.

What problem does it solve? Code reviews by a single AI model can miss real bugs or produce false positives that waste developer time. This Skill sends your code changes to an independent external reviewer (Codex CLI in headless mode), verifies every finding against the actual codebase, and automatically fixes confirmed issues through iterative review cycles until the code is approved. ## Core Features & Use Cases - Cross-model code review: Sends git diffs (uncommitted changes, commits, ranges, or specific files) to Codex, which explores the codebase and returns structured JSON findings with severity, confidence scores, and evidence. - Verification before fixing: Every finding is checked against the real code and classified as CONFIRMED or FALSE POSITIVE, so only genuine issues get fixed by the 1c-code-writer agent. - Iterative review loop: After fixes, the code is re-reviewed with accumulated context of accepted and rejected findings, repeating until approval, iteration limit, or stalemate detection. - Use Case: After implementing a feature in a 1C:Enterprise project, run the skill on your uncommitted changes to get an independent review, automatic fixes for real problems, and a full audit log in .codex-code-review/review-log.md. ## Quick Start Ask the AI to review your uncommitted changes with the codex-code-review-fix skill, for example: run /codex-code-review-fix on my uncommitted changes with a maximum of 5 iterations.

Frequently Asked Questions about codex-code-review-fix

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I run an automated code review on uncommitted git changes?

Invoke the skill with the scope 'uncommitted' or 'unstaged', which maps to git diff HEAD. The skill collects changed files, sends the diff to Codex for review, verifies findings, and fixes confirmed issues iteratively until approval or the iteration limit.

How does cross-model AI code review reduce false positives?

Each finding from the external reviewer is verified against the actual codebase before any fix is applied. Findings are classified as CONFIRMED or FALSE POSITIVE, and rejected findings are sent back to the reviewer with detailed arguments, creating a dialogue between models.

What git scopes can I specify for code review?

Supported scopes include uncommitted changes, the last commit, the last N commits, explicit ranges like HEAD~3..HEAD, single or multiple commit hashes, and specific file paths. If no scope is given, the skill asks the user.

Does this code review skill require external tools installed?

Yes, it requires the codex CLI available in PATH, since reviews run through 'codex exec' in read-only headless mode. It also requires a git repository and uses the 1c-code-writer agent to apply fixes for confirmed findings.

What happens when the reviewer and the fixing model disagree?

The skill includes stalemate detection: if findings repeat across iterations with the same files and problems, the loop stops and the disagreement is shown to the user for a manual decision instead of wasting iterations.

Where are code review results and logs stored?

All artifacts are written to the .codex-code-review/ directory in the project root, including per-iteration prompts, the run log, optional change context, and review-log.md containing findings, verification decisions, and the final summary.