Code Review Skill

Overview

A systematic approach to code review that moves beyond "it looks good" to rigorous quality verification. This skill provides specific checklists and procedures for different review types.

When To Use

• Self-Review: Before submitting a PR or finishing /work
• Peer Review: When reviewing another agent's or human's code (/resolve_pr)
• Plan Review: When validating an implementation plan (/plan_review)

Instrumentation

# Log usage when using this skill
Call MCP `call_tool_logger_manager` { action: "logSkill", name: "code-review", outcome: "manual" }

What do you want to do?

1. Security Review (Auth, RLS, Input) → workflows/security-pass.md
2. Performance Review (Database, Re-renders) → workflows/performance-pass.md
3. Architecture Review (State, Data Flow) → workflows/architecture-pass.md
4. General Quality Check → checklists/pre-merge.md

Key Principles

• Review in Passes: Don't check everything at once. Do a security pass, then a performance pass, etc.
• Reference Patterns: Always check against docs/solutions/patterns/critical-patterns.md.
• Verify, Don't Guess: If you see a potential issue, verify it with a quick test or script.

Scientific Review Principles

Adapted from formal peer review standards to improve code review rigor:

1. Algorithmic Soundness Check: Avoid circular logic where state values mask deeper architectural flaws.
2. Control Variables Isolation: Ensure side-effects are heavily isolated and easily testable (simulating scientific controls).
3. Absolute Reproducibility: If a bug or edge-case is discussed in review, verify that the system has enough telemetry/logging to perfectly reproduce it.
4. Constructive Tone Constraint: Frame criticism objectively as an opportunity for improvement. Avoid dismissing implementations without offering actionable, pattern-compliant alternatives.