Review Coordinator Skill

Orchestrates comprehensive code reviews by spawning 4 specialized review agents in parallel:

1. Marcy Sutton & Léonie Watson - Inclusive design, semantic HTML, keyboard/screen reader accessibility
2. React Router Team - Progressive enhancement, nested routing, loaders/actions, web platform APIs
3. React Core Team - Declarative UI, functional design, state predictability
4. Anders Hejlsberg - Type safety, developer productivity, tooling ergonomics

Critical Workflow

When this skill is activated:

1. Gather context - Identify what code to review (files, diffs, specific components)

2. Spawn 4 agents in PARALLEL - Use a SINGLE message with 4 Task tool calls:

   • Agent 1: Accessibility perspective (using .claude/skills/review/references/a11y-reviewer.md)
   • Agent 2: React Router perspective (using .claude/skills/review/references/react-router-reviewer.md)
   • Agent 3: React Core perspective (using .claude/skills/review/references/react-reviewer.md)
   • Agent 4: Anders Hejlsberg perspective (using .claude/skills/review/references/anders-reviewer.md)

3. Collect results - Each agent returns independent review findings

4. Synthesize findings - Combine all 4 perspectives into a unified report:

   • Group issues by severity (critical, major, minor)
   • Highlight where perspectives agree (cross-cutting concerns)
   • Note perspective-specific insights
   • Prioritize actionable recommendations

5. Create daily note - After synthesizing the review, use the notes skill to save findings:

   • REQUIRED: Read the template reference at .claude/skills/notes/references/review-note-template.md
   • Use Skill("notes") to activate the notes skill
   • Follow the review note template structure from the reference
   • Store in docs/notes/reviews/[YYYY-MM-DD].md

Agent Prompts Template

Each spawned agent receives:

You are conducting a code review from the [PERSONA] perspective.

REQUIRED: First, read the persona reference file at .claude/skills/review/references/[reviewer-name].md to understand your review priorities, what to push back against, and your evaluation criteria.

Then review the following code:
[CODE/FILES/DIFF]

Apply your persona's specific priorities and provide:
1. Critical issues (blocking)
2. Major concerns (should fix)
3. Minor suggestions (nice to have)
4. What's done well

Format: Use clear headings, cite file:line numbers, provide concrete examples.

Output Format

# Multi-Perspective Code Review

## Executive Summary
[Brief overview of critical issues across all perspectives]

## Critical Issues
- **[Issue description]** - *Routing, TypeScript*
  - [Details with file:line references]
  - [Rationale from each reviewer]

## Major Issues
- **[Issue description]** - *TypeScript*
  - [Details with file:line references]
  - [Rationale]

## Minor Issues
- **[Issue description]** - *React*
  - [Details with file:line references]
  - [Rationale]

## What's Done Well
- **[Praise item]** - *Routing, TypeScript*
  - [What they liked and why]

## Detailed Reviewer Findings

### React Router Team
[Full review findings organized by severity]

### React Core Team
[Full review findings organized by severity]

### Anders Hejlsberg
[Full review findings organized by severity]

## Synthesis & Recommendations
[Unified action items prioritized by impact]

Best Practices

• Parallel execution: ALWAYS spawn all 4 agents in a single message (4 Task calls)
• Clear boundaries: Each agent reviews independently without cross-talk
• Concrete examples: Agents should cite specific file:line references
• Actionable output: Synthesize into clear next steps, not just criticism
• Balance: Celebrate good patterns alongside flagging issues
• Document findings: After completing review, prompt user about creating a daily note using the notes skill

Post-Review Workflow

After presenting the synthesized review to the user, offer to save the findings as a daily note:

Would you like me to save this review to your daily notes? I can create a structured note in docs/notes/reviews/[YYYY-MM-DD].md using the review note template.

If user agrees, use Skill("notes") and follow the review note template from .claude/skills/notes/references/review-note-template.md.

When NOT to Use This Skill

• For single-perspective reviews (use the review skill directly)
• For automated testing (not code review)
• For implementation tasks (this skill only reviews, doesn't write code)