How does this skill handle conflicting advice from different models?

It identifies the specific point of contention and uses diagnostic tools to read the actual source code, determining which model is correct based on evidentiary proof rather than consensus.

Why is verification necessary if AI models are generally accurate?

LLMs often hallucinate specific details or fail to account for project-wide context; verification ensures that suggestions are grounded in the actual project reality rather than model probabilities.

Can I use this for non-coding tasks?

While optimized for codebase analysis, the principle of verifying external claims against a local source of truth can be adapted for any documentation or data-heavy project accessible to the file system.

Does it automatically apply fixes to my code?

No, it focuses on providing a verified assessment and a prioritized action item list, allowing developers to make informed decisions before manually or programmatically applying changes.

Multi-Model Meta-Analysis

Name: Multi-Model Meta-Analysis
Author: petekp

bypetekp

0•

Developer Tools

Synthesizes and verifies code analysis from multiple AI models by cross-referencing claims against the actual source code.

This skill provides a systematic framework for consolidating feedback from various LLMs like Claude, GPT-4, and Gemini into a single, reliable source of truth. It prevents AI hallucinations by extracting claims, deduplicating them, and rigorously verifying every assertion against the actual codebase using file-reading and search tools. By resolving model contradictions with concrete evidence, it delivers a comprehensive assessment that distinguishes between verified facts, refuted claims, and unique insights, ensuring that developer actions are based on reality rather than probabilistic guesses.

Key Features

010 GitHub stars

02Structured assessment reports categorized by severity, impact, and evidence

03Actionable checklist generation based on verified, high-impact findings

04Conflict resolution engine that uses source code as the ultimate arbiter

05Systematic code verification to confirm or refute AI-generated assertions

06Automated claim extraction and deduplication across multiple AI model outputs

Use Cases

01Consolidating security audit reports from multiple different LLMs

02Creating a unified source of truth for complex refactors based on various AI suggestions

03Resolving conflicting implementation advice when switching between AI assistants

What are Skills?·How to Install

Install with 🐟 Skill.Fish

npx skillfish add petekp/agent-skills multi-model-meta-analysis

For use in Claude.ai and ChatGPT

Download Skill