Can I review specific files or commits?

Yes, you can provide specific file paths or commits as arguments. If no arguments are provided, the skill defaults to reviewing your most recent conversation history and file changes.

Does this skill automatically fix my code?

No, this is a report-only skill. It provides objective feedback and actionable recommendations for the developer to review and implement, ensuring you maintain full control over your codebase.

How does the multi-agent debate work in Reflexion Critique?

It spawns three specialized agents (Requirements Validator, Solution Architect, and Code Quality Reviewer) who analyze the work independently. It then synthesizes their findings to reach a consensus, highlighting areas where they agree or disagree.

Is Reflexion Critique compatible with other AI CLIs?

Yes, as part of the context-engineering-kit, it is designed to be compatible with Claude Code, OpenCode, Cursor, Antigravity, and Gemini CLI.

What is the benefit of the Chain-of-Verification (CoVe) process?

CoVe requires each judge agent to generate and answer verification questions about their own analysis before submitting it. This process significantly reduces errors and ensures the critique is based on evidence rather than assumptions.

Reflexion Critique - Claude Code Skill for Multi-Agent Review

Reflexion Critique is a sophisticated review coordinator that implements the Multi-Agent Debate pattern to provide high-fidelity feedback on completed work. By spawning three distinct agents—a Requirements Validator, a Solution Architect, and a Code Quality Reviewer—it performs a deep analysis of code changes, identifies architectural trade-offs, and suggests prioritized refactorings. This skill is ideal for developers seeking objective, structured feedback before merging code or finalizing complex features, ensuring that implementations are not only functional but also maintainable and aligned with original project goals.

주요 기능

01542 GitHub stars

02Structured Consensus Reporting: Generates a prioritized report with must-do, should-do, and could-do recommendations.

03Architectural Analysis: Evaluates design patterns, scalability, and explores alternative implementation approaches.

04Requirements Alignment: Automatically maps code changes against original project requests to identify gaps or scope creep.

05Chain-of-Verification (CoVe): Each judge validates their own critique to reduce hallucinations and ensure accuracy.

06Multi-Agent Debate: Orchestrates three specialized judges for independent, high-fidelity reviews.

사용 사례

01Pre-PR sanity checks to catch logic gaps and code smells before human review.

02Validating that a complex feature implementation meets all original technical and business requirements.

03Evaluating architectural trade-offs and identifying potential refactoring opportunities in a project.

What are Skills?·How to Install

Install with 🐟 Skill.Fish

npx skillfish add neolabhq/context-engineering-kit critique

For use in Claude.ai and ChatGPT

Download Skill