Code Review
Dispatch specialized review agents matched to the situation. Choose a mode or let context guide you.
Review Modes
Quick Review
When: Ad-hoc checks, quick sanity check, low-noise feedback Dispatch: quick-reviewer agent (confidence scoring, reports only >= 80) Output: High-confidence issues grouped by severity (Critical 90-100, Important 80-89)
Deep Review
When: After completing a feature, before merge, in plan-execution review loop Dispatch: deep-reviewer agent (full checklist with template placeholders) Placeholders: {WHAT_WAS_IMPLEMENTED}, {PLAN_OR_REQUIREMENTS}, {BASE_SHA}, {HEAD_SHA}, {DESCRIPTION} Output: Strengths, Issues (Critical/Important/Minor), Recommendations, Merge verdict
Specialist Review
When: Targeted analysis of a specific quality dimension Dispatch: One or more specialist agents:
| Agent | Focus |
|---|---|
| comment-analyzer | Comment accuracy, documentation quality, comment rot |
| test-analyzer | Test coverage quality, behavioral coverage gaps, test resilience |
| silent-failure-hunter | Error handling, silent failures, catch block specificity |
| type-design-analyzer | Type invariants, encapsulation, type usefulness |
Comprehensive Review
When: Before merging to main, finishing a development branch Dispatch: 4-pass structure (TDD compliance, diff-only quality, contextual quality, browser audit). See references/comprehensive-review-procedure.md for pass details and ordering.
How to Request
- Gather git context using
scripts/review-prep(auto-detects base branch, outputs BASE_SHA, HEAD_SHA, diff stats, changed files, commit log). Use--jsonfor structured output. - Read the reference file for your chosen agent(s)
- Spawn
general-purposeTask agent with the prompt + review-prep output - Act on feedback: fix Critical immediately, fix Important before proceeding
Workflow Integration
| Workflow | Review Mode |
|---|---|
| plan-execution (per-task) | Deep Review |
| plan-execution (batch) | Deep Review |
| Before merge to main | Comprehensive Review |
| Ad-hoc / when stuck | Quick Review |
| Specific concern | Specialist Review |
| GitHub PR review + posting | github-pr skill (automated PR review orchestrator) |
Simplification Mode
After review, optionally dispatch a simplification pass to improve clarity, consistency, and maintainability without changing functionality. Read references/simplifier.md for the full agent prompt, then spawn a general-purpose Task agent with that prompt targeting the reviewed files.
Handling Feedback
When receiving code review feedback (from humans or other agents), follow the structured 6-step response pattern: READ, UNDERSTAND, VERIFY, EVALUATE, RESPOND, IMPLEMENT. No performative agreement — verify before implementing, push back with technical reasoning when warranted. See references/handling-feedback.md for the full pattern, push-back criteria, YAGNI checks, and source-specific handling.
Red Flags
- Skipping review because "it's simple"
- Ignoring Critical issues
- Proceeding with unfixed Important issues
- Running Comprehensive passes all sequentially (1+2 are independent, parallelize)
- Not deduplicating between Pass 2 and Pass 3 in Comprehensive mode
Reference Files
| Agent | Reference |
|---|---|
| deep-reviewer | references/deep-reviewer.md |
| quick-reviewer | references/quick-reviewer.md |
| comment-analyzer | references/comment-analyzer.md |
| test-analyzer | references/test-analyzer.md |
| silent-failure-hunter | references/silent-failure-hunter.md |
| type-design-analyzer | references/type-design-analyzer.md |
| simplifier | references/simplifier.md |
| handling-feedback | references/handling-feedback.md |