# Code Review

> Evaluates code changes for correctness, style, and safety.

- Skill: `openkoi-ai/code-review` (Agent Skill)
- Install (CLI): `npx skillmds@latest add openkoi-ai/code-review`
- Raw SKILL.md: https://api.skillmd.com/api/skills/openkoi-ai/code-review/raw
- Safety review: pending (external: skill-scanner PASS, skillspector PASS)
- Works with: Claude Code, Claude.ai, OpenAI Codex
- Category: Coding & Dev Tools
- Author: openkoi-ai (https://skillmd.com/u/openkoi-ai)
- Updated: 2026-09-10
- Page: https://skillmd.com/skills/openkoi-ai/code-review

---


# Code Review Evaluator

Evaluate the code output against these criteria.

## Correctness (40%)

- Does the implementation match the task requirements?
- Are all specified behaviors implemented?
- Would this code produce correct results for normal inputs?
- Are there logic errors, off-by-one mistakes, or incorrect assumptions?
- Does the code handle the data types and structures correctly?
- If modifying existing code, does it preserve existing behavior where expected?

## Safety (25%)

- Are errors handled properly (no unwrap on user input, no silent failures)?
- Is user input validated and sanitized?
- Are there potential panics, integer overflows, or buffer issues?
- Are there resource leaks (unclosed files, connections, channels)?
- Are credentials, secrets, or sensitive data handled securely?
- Is there proper bounds checking on arrays, slices, and indices?
- Are concurrent access patterns safe (no data races, proper locking)?

## Style (15%)

- Is the code idiomatic for the language?
- Are variable and function names descriptive and consistent?
- Is the code DRY without being over-abstracted?
- Are functions at a reasonable size (not too long, not trivially small)?
- Is the code formatted consistently?
- Are comments present where logic is non-obvious?

## Completeness (20%)

- Are edge cases handled (empty inputs, None/null, max values, unicode)?
- Are tests included or updated (if applicable)?
- Is the change documented where needed (comments, docstrings, README)?
- Are error messages helpful and actionable?
- If adding a public API, is it documented?

## Severity Guide

- **Blocker**: Crashes, data loss, security vulnerability, incorrect behavior for normal inputs
- **Important**: Missing error handling, poor performance on common paths, missing tests for core logic
- **Suggestion**: Style nits, naming improvements, minor refactoring opportunities