What happens if a plan fails the review?

If the verdict is 'NEEDS_FIXES', the skill presents specific findings to the user and suggests returning to the planning stage to address issues before any PR is created.

How does the multi-model review work?

The skill spawns three agents (Claude, Codex, and Gemini) to evaluate your plan.md from different perspectives—logic, architecture, and edge cases—then synthesizes their feedback into a single verdict.

Where are the review results saved?

All findings, consensus items, and the final verdict are saved to a dedicated plan-review.md file within your project's spec directory for full traceability.

Does this skill modify my source code?

No, the review stage is read-only for your plan. It only interacts with git to create worktrees and draft pull requests once a plan is officially approved for shipping.

Can I use this if I only have access to Claude?

Yes, the skill is designed with graceful degradation. If Codex or Gemini are unavailable, it will proceed with a Claude-only review based on available tools.

Review Plan

Name: Review Plan
Author: mildsalmon

bymildsalmon

0•

协作工具

Performs a multi-model consensus review of implementation plans to validate technical feasibility and automatically generate draft pull requests.

The Review Plan skill automates the critical bridge between architectural planning and implementation by leveraging Claude, Codex, and Gemini in parallel to audit technical specifications. It identifies logical inconsistencies, edge cases, and architectural flaws, synthesizing feedback into a unified report. If the plan meets quality standards, the skill automatically initializes git worktrees and creates draft pull requests, streamlining the transition from design to development while ensuring high-quality standards through multi-agent validation and consensus-based decision making.

主要功能

010 GitHub stars

02Automated Git workflow including worktree creation and Draft PR generation

03Triple-model consensus review utilizing Claude, Codex, and Gemini for diverse perspectives

04Graceful degradation to single-model review if external APIs are unavailable

05Automated synthesis of findings into structured, numbered CR-xxx reports

06Smart verdict system that determines if a feature is ready to SHIP or NEEDS_FIXES

使用场景

01Cross-referencing implementation plans against project requirements to ensure 100% coverage

02Automating the standard PR creation and branch setup workflow for engineering teams

03Validating complex feature specifications before starting the implementation phase

What are Skills?·How to Install

Install with 🐟 Skill.Fish

npx skillfish add mildsalmon/taco-harness review-plan

For use in Claude.ai and ChatGPT

主要功能

010 GitHub stars

02Automated Git workflow including worktree creation and Draft PR generation

03Triple-model consensus review utilizing Claude, Codex, and Gemini for diverse perspectives

04Graceful degradation to single-model review if external APIs are unavailable

05Automated synthesis of findings into structured, numbered CR-xxx reports

06Smart verdict system that determines if a feature is ready to SHIP or NEEDS_FIXES

使用场景

01Cross-referencing implementation plans against project requirements to ensure 100% coverage

02Automating the standard PR creation and branch setup workflow for engineering teams

03Validating complex feature specifications before starting the implementation phase