Code Review

Overview

Covers both sides of code review: requesting reviews (dispatching reviewer subagents) and receiving feedback (verifying before implementing suggestions).

Requesting Review

After completing implementation, dispatch two reviews in order:

1. Spec Compliance Review

Route to Codex CLI (read-only):

bash

${CLAUDE_PLUGIN_ROOT}/skills/plan-and-execute/codex-runner.sh \
  "Review code for spec compliance. Spec: [requirements]. Files: [paths]. Report: APPROVED or NEEDS CHANGES with specifics." \
  read-only /path/to/project

Fail-closed: if Codex returns exit 20, ask user for explicit approval before any Claude/Sonnet fallback.
Optional fallback only when CODEX_FAIL_CLOSED=0: Task(prompt, model:"sonnet", subagent_type:"superpower-router:code-reviewer")

2. Code Quality Review

Only after spec compliance passes. Route to Codex CLI (read-only):

bash

${CLAUDE_PLUGIN_ROOT}/skills/plan-and-execute/codex-runner.sh \
  "Review code quality. Files: [paths]. Check: naming, error handling, DRY, YAGNI, test coverage. Report: Strengths, Issues (Critical/Important/Minor), APPROVED or NEEDS CHANGES." \
  read-only /path/to/project

Review Loop

If reviewer finds issues → fix → re-review → repeat until approved.

Receiving Review Feedback

When YOU receive feedback, apply technical rigor:

Step	Action
1	Read feedback carefully — don't skim
2	Verify each suggestion is technically correct before implementing
3	If suggestion seems wrong, research it (WebSearch, docs) before disagreeing
4	If genuinely wrong, explain why with evidence — don't just agree
5	Implement valid suggestions, explain rejections

Red Flags

•Never agree performatively — verify first
•Never implement suggestions blindly without understanding why
•Never skip re-review after fixing issues
•Never start quality review before spec compliance passes

Reviewer Agent Prompt

See ./code-reviewer.md for the detailed reviewer agent prompt template.