Does this skill use real external models or simulated ones?

This skill uses real CLI invocations for Gemini, Codex, and Cursor-Agent, providing genuine model diversity rather than simulated personas within a single session.

What happens if one of the external CLIs fails or times out?

The skill includes a monitoring strategy that polls for results; if a model fails, you can choose to proceed with partial consensus from the remaining models or retry the execution.

How does the consensus logic work?

It performs semantic matching across all feedback files to identify issues flagged by two or more models, highlighting them as priority consensus items in the final report.

Do I need to have the external CLIs installed locally?

Yes, this skill requires the Codex, Gemini, and Cursor-Agent binaries to be installed and authenticated on your system to perform the parallel reviews.

What is a 'Divergent Opinion' in the final report?

A divergence is flagged when different models explicitly suggest conflicting approaches to the same topic, signaling an architectural trade-off that requires human decision-making.

Multi-Model Spec Reviewer

Name: Multi-Model Spec Reviewer
Author: matthewod11-stack

bymatthewod11-stack

•

Collaboration Tools

Orchestrates parallel architectural reviews using real Gemini, Codex, and Cursor-Agent CLIs alongside native Claude sessions for diverse technical feedback.

This skill automates a sophisticated Multi-Agent Review Protocol by launching actual external AI CLIs in parallel with your active Claude session. By leveraging the unique strengths of different models—such as Gemini's breadth, Codex's implementation focus, and Cursor's structural insights—it provides a comprehensive 360-degree critique of your technical specifications. The skill automatically monitors background processes, handles timeouts, and synthesizes the results into a consolidated report that highlights consensus points, identifies divergent architectural opinions, and ensures no edge case is overlooked during the design phase.

Key Features

01Consolidated feedback reports with full source attribution and divergence flags

021 GitHub stars

03Automated consensus detection across four distinct AI models

04Specialized review lenses for security, architecture, and DX

05Real-time background process monitoring with sentinel file tracking

06Parallel execution of real AI CLIs (Gemini, Codex, Cursor-Agent)

Use Cases

01Cross-validating security implications and edge cases using multiple LLM providers

02Vetting high-stakes system architectures before writing any code

03Generating diverse peer-review perspectives for PRDs and technical design documents

What are Skills?·How to Install

Install with 🐟 Skill.Fish

npx skillfish add matthewod11-stack/claude-setup spec-review-multi

For use in Claude.ai and ChatGPT

Download Skill