Automate iterative code reviews with Codex and Claude Code patches.
npx skills add https://github.com/shishi/dotfiles --skill codex-review-shishi
Please help me install this Agent Skill. Skill: codex-review Source: https://github.com/shishi/dotfiles/tree/main/claude/skills/codex-review Command: npx skills add https://github.com/shishi/dotfiles --skill codex-review-shishi
This Skill enables iterative, Codex-driven reviews where a read-only Codex review guides Claude Code fixes and performs a re-review until an ok: true state is reached. It is used after major milestones such as specs/PRD updates, and before commits, PRs, merges, or releases.
Use this skill to perform a Codex-guided review gate on a code patch; provide the diff and let Claude Code propose fixes, iterating until the result is ok: true.
High-intent search queries and answers about installing and using this skill.
Automated iterative code reviews use a read-only reviewer to propose fixes and re-evaluate patches until passing. This Skill runs Codex in read-only mode to guide Claude Code patch generation, looping until an ok: true state is reached before commits or PRs.
Yes, Claude Code can patch diffs identified by Codex during a review gate. The workflow uses Codex in read-only mode to review deterministic diff ranges, after which Claude Code generates the necessary fixes and triggers a re-review until the code passes.
A Codex-guided review loop is an automated process where read-only Codex evaluates code changes and Claude Code applies fixes. It iterates through review and patch cycles with deterministic diff ranges and a strict limit on iterations until achieving an ok: true status.
You should run an automated code review gate after major milestones like specs, PRD, or design updates, and immediately before commits, PRs, merges, or releases. This ensures code quality is verified before changes enter main branches.
Yes, this automated code review requires both Codex and Claude Code. Codex operates in read-only mode to evaluate the code and propose fixes, while Claude Code is required for generating the actual patches and re-evaluating the results.
The Codex and Claude Code review loop runs with a strict limit on iterations to prevent infinite cycles. It processes deterministic diff ranges and stops automatically once the read-only review returns an ok: true state or the iteration limit is reached.
rintiEvaluate uncommitted git diffs with Codex CLI review feedback.
alo-expFinalize Git branches by verifying tests, choosing merge strategies, and creating PRs.
ryosukesutoCreate pull requests with titles, bodies, and analysis from git diffs.
JSai23Detect and fix AI-generated code quality issues with before/after diffs.
