review

Reviews git diffs with parallel audit agents, confidence scoring, and optional auto-fix.

6|Updated Mar 27, 2026
One-click install
npx skills add https://github.com/greglas75/zuvo --skill review-greglas75
Or copy as Structured Prompt for Agent
Please help me install this Agent Skill.
Skill: review
Source: https://github.com/greglas75/zuvo/tree/main/skills/review
Command: npx skills add https://github.com/greglas75/zuvo --skill review-greglas75

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve? Code review is inconsistent and easy to skip under time pressure. This Skill runs a structured, end-to-end review of uncommitted changes, staged diffs, or commit ranges, producing an evidence-backed tiered report (MUST-FIX / RECOMMENDED / NIT) and optionally applying fixes automatically. ## Core Features & Use Cases - Tiered Review Depth: Automatically scales review intensity (TIER 0 NANO through TIER 3 DEEP) based on diff size, file count, and risk signals like auth, payments, or DB migrations. - Parallel Audit Agents: Dispatches behavior, structure, and CQ (code quality gate) auditor sub-agents, then re-scores every finding with a confidence model before reporting. - Adversarial Cross-Model Validation: Runs cross-provider adversarial review to catch semantic bugs that mechanical scanners miss, with CRITICAL findings bypassing confidence thresholds. - Auto-Fix Modes: Default FIX-AUTO mode applies MUST-FIX and high-confidence RECOMMENDED fixes with verification; report-only, blocking-only, and batch queue modes are also supported. - Use Case: Before merging a feature branch touching payment logic, run the review to get a deployment risk score, a cited findings report, and verified automatic fixes for blocking issues. ## Quick Start Ask the assistant to review the current uncommitted changes and automatically fix any must-fix issues it finds.

Frequently Asked Questions about review

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

FAQPage Schema
How do I review uncommitted changes before committing?

Run the review skill with no arguments to audit all uncommitted changes via git diff against HEAD. It classifies the diff, selects a review tier, runs audit agents, and by default applies MUST-FIX fixes automatically.

How do I review a specific commit range or pull request?

Pass a range like HEAD~3 or abc123..def456 as the argument to review those commits. The skill validates the range, computes the diff stat, and scopes all analysis and findings to that range.

Can I get a review report without automatic code changes?

Yes, use the --report-only flag to audit and present findings without modifying any code. Other modes include blocking (MUST-FIX only), fix (all findings including NITs), and batch queue processing.

What languages and frameworks does the code review support?

It supports TypeScript, JavaScript, Python, PHP, Kotlin, Go, Rust, Ruby, and .NET, with framework-specific rules for Next.js, NestJS, React, Django, FastAPI, Yii2, Astro, and others. Stack detection loads at most two matching rule sets per review.

How does the review decide how deep to analyze a diff?

A tier system scales depth from TIER 0 (under 15 lines, no risk signals) to TIER 3 (over 500 lines or multiple risk signals). Risk signals include DB migrations, auth changes, API contract changes, and payment logic.

Why did my review report get rejected by the audit gate?

Every MUST-FIX and RECOMMENDED finding must cite a file path and line number that resolves in the current tree. Reports with missing or invalid citations fail verification and must be corrected before the review completes.