taste

Review git diffs with Codex and validate findings via Claude.

68|10|Updated Jul 2, 2026

Search Tags:#git #quality assurance #code review #validation #codex #cross-model

Ready to run

npx skills add https://github.com/tomascupr/sous-chef --skill taste-tomascupr

💬 Prompt for ChatGPT / Claude Chat▼

Please help me install this Agent Skill.
Skill: taste
Source: https://github.com/tomascupr/sous-chef/tree/main/skills/taste
Command: npx skills add https://github.com/tomascupr/sous-chef --skill taste-tomascupr

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill includes references (resource) components.

What problem does it solve?

This skill addresses the blind spots inherent in single-model code reviews by orchestrating a multi-model validation process that filters out false positives and ensures only verified findings reach the user.

Core Features & Use Cases

Cross-Model Validation: Uses Codex for initial review and Claude for rigorous, read-only validation of every finding against the actual codebase.
Context-Aware Scoping: Supports working-tree, branch-based, and post-fire delta reviews to ensure findings are relevant to the specific changes made.
Use Case: Use this when you need a second opinion on a complex or risky pull request to catch edge-case bugs that a single model might miss during self-review.

Quick Start

Invoke the taste skill to perform a cross-model review of the current git diff and validate all findings before presentation.

Frequently Asked Questions about taste

High-intent search queries and answers about installing and using this skill.

FAQPage Schema

How does cross-model code review work for git diffs?▼

Cross-model code review uses Codex for an initial analysis of git diffs and Claude for a mandatory read-only validation pass. This multi-model process filters out false positives by verifying findings against the local repository state.

How do I validate code review findings against my local git repository?▼

You validate findings by running a read-only analysis that checks each code review issue against the actual codebase. This mandatory validation pass ensures only verified findings are presented.

Do I need Codex CLI configured to perform cross-model validation?▼

Yes, you need a configured Codex CLI environment and git access. These prerequisites allow the review process to independently verify findings against your local repository state.

What is the best way to review risky pull requests across multiple models?▼

The best way is using a cross-model validation process that applies context-aware scoping to working-tree, branch-based, or post-fire deltas. This catches edge-case bugs a single model might miss.

Can I scope code quality checks to specific branch-based changes?▼

Yes, context-aware scoping supports working-tree, branch-based, and post-fire delta reviews. This ensures the code quality findings are relevant to the specific changes made.

Why does single-model code review miss edge-case bugs?▼

Single-model code review has inherent blind spots that can miss edge-case bugs. Cross-model validation addresses this by filtering false positives through a rigorous secondary verification pass.