ai-coding-agent-review

name	ai-coding-agent-review
description	Final sanity check on AI-generated code before merging. Run the checklist to catch common failure modes. Trigger on /ai-coding-agent-review, or before finalizing any significant AI-generated change.
user-invocable	true
allowed-tools	Read, Grep, Glob, Bash

AI Coding Agent Review

Pre-merge checklist for AI-generated code. Catches common failure modes before they land.

Checklist

Solves the actual problem?

• Does the code address the root cause or just the surface symptom?
• Would the fix survive a slightly different scenario?

Minimal?

• Could this be half the size and still work?
• Are there unused variables, dead code, commented-out blocks, or speculative abstractions?
• Does it add dependencies that aren't strictly needed?

Matches project patterns?

• Consistent naming, module structure, error handling, and import style.
• Uses existing helpers and patterns instead of inventing new ones.

Tests meaningful?

• Do tests actually verify the behavior change, or just execute code?
• Are there assertions on the right thing?
• Do they avoid testing implementation details?

Avoided test overfitting?

• Tests should break when behavior changes, not when internals change.
• No snapshot tests of large irrelevant data.
• No tests that pass for the wrong reason.

Avoided hardcoded hacks?

• No hardcoded paths, ports, secrets, environment names, or IPs.
• No magic numbers without named constants.

Avoided security leaks?

• No secrets or credentials in code, logs, or test output.
• No command injection, path traversal, or XSS vectors.
• Input validation at system boundaries.

Avoided broad refactors?

• Does it change only what's needed for the feature/fix?
• No drive-by reformatting, renaming, or restructuring.
• If refactoring is included, is it clearly separated?

Verified with smallest relevant command?

• Did the developer run the specific test or lint rule, not the entire suite?
• Is there evidence of verification in the output?

Limitations stated?

• Are edge cases, known gaps, or unverified configurations documented?
• Honest about what was NOT tested.

Verdict

Pass / Pass with nits / Fix issues / Rework

Experience Log

This skill learns from experience. Mechanism:

• Read .claude/experience/ai-coding-agent-review.md at skill start to benefit from past lessons.

• Write to .claude/experience/ai-coding-agent-review.md after use if you learned something new.

• Format: YYYY-MM-DD: <lesson> — one line per entry.

• Evolve: If the same issue repeats 3+ times, update this SKILL.md itself.

This skill improves over time. After each use, append lessons learned.

Record format: YYYY-MM-DD: <lesson> Read this at skill start to apply past lessons.