Is there a way to save on token usage with this skill?

Yes, for lower-risk tasks, you can use 'Simplified Mode' to run only one specific reviewer (e.g., just Quality or just Architecture) instead of all four.

What does the 'Reality Reviewer' specifically check?

The Reality Reviewer verifies that all referenced methods, classes, and file paths actually exist in your codebase to prevent 'hallucinated' code references.

What is the Plan Reviewer skill?

It is a specialized capability for Claude Code that analyzes implementation plans using four specialized AI agents to ensure they are safe, accurate, and ready for execution.

When should I use the /review-plan command?

You should invoke it after generating an implementation plan but before starting the actual coding phase, particularly for high-complexity or high-risk features.

How are review results delivered?

The skill saves a detailed JSON report to your project directory and displays a prioritized human-readable summary with a clear 'Approved' or 'Changes Requested' verdict.

Plan Reviewer

Name: Plan Reviewer
Author: tachyon-beep

bytachyon-beep

•

协作工具

Validates implementation plans against codebase reality, architectural risks, and project conventions before code execution.

The Plan Review skill acts as a rigorous quality gate in the software development lifecycle, ensuring that proposed implementation plans are viable, secure, and aligned with project standards. By spawning four specialized reviewers—Reality, Architecture, Quality, and Systems—it performs a deep dive into symbol verification, dependency chains, and potential security vulnerabilities. This automated peer-review process prevents hallucinations, reduces technical debt, and provides a structured JSON report with a clear verdict, making it indispensable for high-stakes or complex coding tasks within Claude Code.

主要功能

01Detailed JSON and human-readable reports with prioritized issue scoring

02Architectural 'blast radius' analysis and one-way door detection

03Automated security scanning for SQL injection and hardcoded secrets

04Multi-agent parallel review focusing on Reality, Architecture, Quality, and Systems

0510 GitHub stars

06Symbol and file path verification against the actual live codebase

使用场景

01Ensuring new implementation plans strictly adhere to project conventions defined in CLAUDE.md

02Verifying complex feature plans before committing to multi-file code changes

03Automating security and architectural sanity checks on AI-generated implementation steps

What are Skills?·How to Install

Install with 🐟 Skill.Fish

npx skillfish add tachyon-beep/skillpacks plan-review

For use in Claude.ai and ChatGPT

Download Skill