SKILL: Multi-Angle Code Review

Purpose: Comprehensive plan review with gap detection, extended reviews, and GPT expert delegation Target: plan-reviewer Agent reviewing plans before execution

Quick Start

When to Use

• Review plan before execution (/01_confirm, /review)
• Detect gaps in external service integration
• Validate test plan completeness

Quick Reference

# Load plan → Detect type → Run reviews → Apply findings
/review .pilot/plan/pending/plan.md

In Scope: 8 mandatory reviews, extended reviews, gap detection, findings application

Out of Scope: Plan creation → spec-driven-workflow | Test execution → tdd | GPT delegation → @.claude/rules/delegator/orchestration.md

Execution Steps

Core Philosophy: Comprehensive (multi-angle) | Actionable (findings map to sections) | Severity-based (BLOCKING → Interactive Recovery)

Team Lead Role

Delegate Mode: Team Lead operates in coordinate-only mode (Shift+Tab)

Pattern: Spawn review-team → Monitor reviewers → Aggregate findings → Update plan

Native Context Isolation: Agent Teams provides automatic context isolation per teammate

Step 1: Load Plan

PLAN_PATH="${1:-$(find "$(pwd)/.pilot/plan/pending" "$(pwd)/.pilot/plan/in_progress" -name "*.md" -type f | head -1)}"
[ -f "$PLAN_PATH" ] || { echo "❌ No plan found"; exit 1; }

Step 2: Multi-Angle Review Team

Create review-team with specialized teammates:

Teammate 1: Test Coverage Reviewer

Spawn teammate "test-reviewer" with prompt:
  "You are a tester (see @.claude/agents/tester.md).
  Review test coverage for plan: $PLAN_PATH.
  Evaluate: SCs verifiable? test commands? coverage ≥80%?
  Discuss findings with other reviewers via Message.
  Output: PASS or FAIL with detailed findings."

Teammate 2: Type Safety & Quality Reviewer

Spawn teammate "quality-reviewer" with prompt:
  "You are a validator (see @.claude/agents/validator.md).
  Review type safety and code quality for plan: $PLAN_PATH.
  Evaluate: TypeScript types? lint config? quality (SRP/DRY/KISS)?
  Discuss findings with other reviewers via Message.
  Output: PASS or FAIL with detailed findings."

Teammate 3: Deep Code Reviewer

Spawn teammate "deep-reviewer" with prompt:
  "You are a code-reviewer (see @.claude/agents/code-reviewer.md).
  Perform deep code quality review for plan: $PLAN_PATH.
  Evaluate: architecture? function size (≤50)? file size (≤200)? nesting ≤3? edge cases?
  Discuss findings with other reviewers via Message.
  Output: PASS or FAIL with detailed findings."

Teammate 4: Security Reviewer

Spawn teammate "security-reviewer" with prompt:
  "You are a security-analyst (see @.claude/agents/security-analyst.md).
  Review security aspects for plan: $PLAN_PATH.
  Evaluate: input validation? auth/authz? secret management? OWASP Top 10?
  Discuss findings with other reviewers via Message.
  Output: PASS or FAIL with detailed findings."

Key Enhancement: Reviewers cross-reference and debate findings via direct messaging before Team Lead aggregates final report.

Step 3: Process Findings

echo "$findings" | grep -q "🛑.*BLOCKING" && { echo "🛑 BLOCKING"; return 1; }

Step 4: Update Plan

Review Workflow

Step 0: Extract SC count

sc_count=$(grep -c "^- \[.\] \*\*SC-" "$PLAN_PATH" || echo "0")

Step 1: Search "needs investigation/confirmation/review" keywords

Step 2: Type detection (code, config, docs, scenario, infra, db, ai)

Step 3: 8 mandatory reviews

1. Development Principles: SOLID, DRY, KISS, YAGNI
2. Project Structure: File locations, naming
3. Requirement Completeness: Explicit + implicit
4. Logic Errors: Order, dependencies, edge cases
5. Existing Code Reuse: Search utilities, patterns
6. Better Alternatives: Simpler/scalable approaches
7. Project Alignment: Type-check, API docs
8. Long-term Impact: Consequences, technical debt

Step 5: Extended reviews (type-activated)

• A: API Compatibility | B: Type Safety | C: Documentation | D: Test Coverage
• E: Migration | F: Deployment | G: Prompt Engineering | H: Scenarios

Step 6: Autonomous perspectives

• Security | Performance | UX | Maintainability | Concurrency | Error Recovery

Step 7: Gap detection (BLOCKING → Interactive Recovery)

• 9.1: External API | 9.2: Database | 9.3: Async | 9.4: File Ops
• 9.5: Env Vars | 9.6: Error Handling | 9.7: Test Plan (BLOCKING)

Step 9.5: Multi-angle review-team (5+ SCs)

# For complex plans (5+ SCs), add design-reviewer teammate if frontend changes detected
[ "$sc_count" -ge 5 ] && grep -qiE "component|UI|frontend|React|Vue" "$PLAN_PATH" && echo "🚀 Add design-reviewer teammate"

Step 10: GPT expert (5+ SCs or architecture/security/auth)

[ "$sc_count" -ge 5 ] || echo "$PLAN_PATH" | grep -qiE "architecture|security|auth" && echo "🤖 GPT"

Further Reading

Internal: @.claude/skills/review/REFERENCE.md - Detailed review criteria, gap detection, GPT delegation | @.claude/rules/delegator/orchestration.md - GPT expert delegation | @.claude/skills/parallel-subagents/SKILL.md - Multi-angle parallel review | @.claude/agents/code-reviewer.md - Code reviewer output format

External: Code Review by Jason Cohen | The Art of Readable Code