code-review

Reviews code changes across correctness, security, performance, and style layers with severity classification.

1|Updated Jul 23, 2026
One-click install
npx skills add https://github.com/sanjanb/my-agent-harness --skill code-review-sanjanb
Or copy as Structured Prompt for Agent
Please help me install this Agent Skill.
Skill: code-review
Source: https://github.com/sanjanb/my-agent-harness/tree/main/skills/code-review
Command: npx skills add https://github.com/sanjanb/my-agent-harness --skill code-review-sanjanb

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve? Code reviews often miss critical issues, produce vague feedback without file references, or drown reviewers in low-confidence noise. This Skill enforces a systematic review methodology so every change is checked consistently across four quality dimensions before merge. ## Core Features & Use Cases - Four-Layer Analysis: Systematically checks correctness, security, performance, and style/maintainability for every file in scope. - Severity Classification: Categorizes findings as Critical, Major, Minor, or Nitpick with clear action requirements for each level. - Confidence Thresholding: Only reports findings at 80% confidence or higher, explicitly flagging uncertain observations to reduce false positives. - Use Case: Before merging a pull request that touches authentication logic, run this review to catch hardcoded secrets, missing input validation, and N+1 query patterns, receiving a structured report with file:line references and an APPROVE or REQUEST_CHANGES verdict. ## Quick Start Review the changes in my current branch using the code-review methodology and report any critical or major issues with file and line references.

Frequently Asked Questions about code-review

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

FAQPage Schema
How do I do a systematic code review before merging?

Apply the four review layers in order: correctness, security, performance, then style and maintainability. Classify each finding by severity, include file:line references, and finish with an overall APPROVE, REQUEST_CHANGES, or NEEDS_DISCUSSION verdict.

What should a code review checklist include?

A thorough checklist covers logic errors and edge cases, hardcoded secrets and injection risks, N+1 queries and memory leaks, plus convention adherence and test coverage gaps. Each finding needs a severity level and a confidence assessment.

How do I reduce false positives in automated code review?

Only report findings at 80% confidence or higher, and explicitly label uncertain observations with their confidence percentage. Prefer false negatives over false positives to keep review noise low and maintain trust in the findings.

What severity levels should code review findings use?

Use four levels: Critical for security vulnerabilities and data loss, Major for bugs and performance issues, Minor for code smells and test gaps, and Nitpick for style preferences. Critical issues must block merge.

When should I not use this review methodology?

Avoid it for trivial changes where a full four-layer analysis adds no value, and never modify files during review since it is read-only. It also requires enough context about project conventions to evaluate the style layer accurately.