Does this skill work for specialized reviews like security or UX?

Yes, it includes specific prefixes and tracking IDs for security (SEC), architecture (ARC), tests (TST), UX (UXD), and several other specialized domains.

How does the skill ensure findings are actionable?

Every finding must include the specific file and line number, logical reasoning, concrete code evidence, and a specific suggestion on how to fix the issue.

How are severity levels (P1-P4) defined?

P1 indicates critical issues like security vulnerabilities or data loss, P2 is for broken features, P3 covers minor bugs, and P4 is for trivial or cosmetic suggestions.

What is the benefit of the confidence scoring system?

Confidence scoring filters out speculative concerns or 'hallucinations' by requiring a minimum 80% certainty score before an issue is reported to the user.

Code Review Methodology

Name: Code Review Methodology
Author: tenzir

bytenzir

•

Security & Testing

Standardizes code reviews by applying a dual-axis framework of severity ratings and confidence scores to identify high-impact issues.

The Reviewing Changes skill provides a rigorous framework for conducting code audits within Claude Code. It implements a specialized methodology that separates the impact of a finding (P1-P4 severity) from the certainty of the observation (0-100% confidence). By enforcing structured reporting—including unique issue IDs, mandatory code evidence, and specific fix suggestions—this skill ensures that AI-generated reviews remain actionable and objective. It is particularly useful for teams looking to automate security, architecture, and performance audits while minimizing noise through strict confidence thresholds.

Key Features

01Dual-axis scoring system (Severity vs. Confidence)

02Evidence-based reporting requirements for every finding

03Automated review directory management via pre-tool hooks

043 GitHub stars

05Persona-based prefixes for security, performance, and architecture

06Standardized markdown output with unique tracking IDs

Use Cases

01Standardizing bug reporting across multiple AI reviewer agents

02Automating high-confidence pull request reviews

03Filtering code suggestions to focus exclusively on critical path issues

What are Skills?·How to Install

Install with 🐟 Skill.Fish

npx skillfish add tenzir/claude-plugins reviewing-changes

For use in Claude.ai and ChatGPT

Download Skill