Can I run this on Windows?

Yes, it is fully compatible with Windows via WSL (Windows Subsystem for Linux) and can be triggered from Git Bash or MINGW.

What happens during the 'Triage' phase?

The triage phase cross-validates findings from all agents, classifies them by severity (Critical to Minor), and determines which issues can be automatically fixed by other Claude skills.

What models does Cross-Review use?

It uses Claude 3.5 Opus for the specialized agent teams and integrates with OpenAI's Codex CLI (configured for GPT-4 or higher) for secondary validation.

Do I need an OpenAI account to use this skill?

Yes, you must have the Codex CLI installed and authenticated with your OpenAI credentials to enable the secondary review layer.

How does it handle disagreements between Claude and Codex?

When the two reviewers disagree on a finding, the loop pauses and presents the conflicting perspectives to the user to make the final decision.

Cross-Review Loop (Claude x Codex)

Name: Cross-Review Loop (Claude x Codex)
Author: DmitriyYukhanov

byDmitriyYukhanov

•

协作工具

Automates deep code reviews by orchestrating parallel evaluations between Claude agent teams and OpenAI Codex to find, triage, and fix issues.

The Cross-Review skill implements an autonomous review-fix loop that leverages the combined strengths of Claude and OpenAI's Codex CLI. It orchestrates parallel review processes—spawning specialized Claude agent teams (Security, Performance, Test) alongside Codex—to perform independent assessments of code, architecture, or implementation plans. By cross-validating findings from multiple AI models, it eliminates blind spots, triages critical issues, and applies automated fixes through iterative cycles, ensuring high-quality, production-ready output with minimal manual intervention.

主要功能

01Autonomous Fix Loop: Iteratively detects, triages, and fixes issues until the codebase is clean or a human decision is required.

02Parallel Multi-Model Review: Simultaneously runs Claude agent teams and Codex CLI for independent perspectives.

03Specialized Reviewer Roles: Supports targeted agents for Security, Performance, Testing, and Architecture.

042 GitHub stars

05Artifact-Aware Analysis: Automatically detects and adjusts review prompts for source code, architecture designs, or implementation plans.

06Cross-Platform Compatibility: Seamlessly handles Windows (via WSL), macOS, and Linux environments.

使用场景

01Pre-PR validation to ensure code meets security and performance standards before a human review.

02Validating complex architectural designs against multiple AI perspectives to find hidden edge cases.

03Automated refactoring of legacy codebases by identifying and fixing technical debt in iterative rounds.

What are Skills?·How to Install

Install with 🐟 Skill.Fish

npx skillfish add dmitriyyukhanov/claude-plugins cross-review

For use in Claude.ai and ChatGPT

Download Skill