Performance Optimization - Data-Driven Frontend Optimization

🎯 Core Philosophy

"Premature optimization is the root of all evil" - Donald Knuth

Optimize after measuring. Make decisions based on data, not feelings.

What This Skill Provides

1. Web Vitals-Based Measurement - LCP, FID, CLS improvement techniques
2. React Optimization Patterns - Reducing re-renders, proper memoization usage
3. Bundle Optimization - Code splitting, Tree shaking, lazy loading
4. Measurement Tools - Chrome DevTools, Lighthouse, React DevTools

---

📚 Section-Based Content

This skill is organized into 3 specialized sections for efficient context usage:

🔍 Section 1: Web Vitals Optimization

File: references/web-vitals.md Tokens: ~500 Focus: Google's Core Web Vitals (LCP, FID, CLS)

Covers:

• Understanding Core Web Vitals metrics
• Improving LCP (Largest Contentful Paint)
• Reducing FID (First Input Delay)
• Preventing CLS (Cumulative Layout Shift)
• Chrome DevTools profiling

When to load: User mentions LCP, FID, CLS, Web Vitals, loading speed, layout shifts

---

⚛️ Section 2: React Optimization

File: references/react-optimization.md Tokens: ~800 Focus: React-specific performance patterns

Covers:

• React.memo for component memoization
• useMemo for computation caching
• useCallback for function memoization
• List virtualization
• State management optimization
• React DevTools Profiler

When to load: User mentions re-renders, React performance, useMemo, useCallback, React.memo

---

📦 Section 3: Bundle Optimization

File: references/bundle-optimization.md Tokens: ~600 Focus: Bundle size reduction and code splitting

Covers:

• Code splitting patterns
• Tree shaking techniques
• Image optimization
• Bundle size measurement
• Dynamic imports
• Lazy loading strategies

When to load: User mentions bundle size, code splitting, lazy loading, tree shaking

---

🔄 How Section Loading Works

Efficient Context Usage

// Before (Monolithic): Always load all 3000 tokens
User: "LCPが遅い"
→ Load entire skill.md (3000 tokens)

// After (Section-based): Load only relevant section
User: "LCPが遅い"
→ Load skill.md metadata (200 tokens)
→ Detect "LCP" keyword → Match to web-vitals section
→ Load references/web-vitals.md (500 tokens)
→ Total: 700 tokens (77% reduction)

Loading Strategy

1. Default: Load metadata + Core Philosophy (~200 tokens)
2. Keyword match: Load specific section (~500-800 tokens)
3. Multiple keywords: Load multiple sections if needed
4. Fallback: If no specific match, suggest relevant section

---

💡 Usage Examples

Example 1: LCP Optimization

User: "ページの読み込みが遅い。LCPを改善したい"

Claude loads:
✓ skill.md metadata (200 tokens)
✓ references/web-vitals.md (500 tokens)
Total: 700 tokens

Provides:
- LCP measurement techniques
- Priority loading strategies
- Preloading techniques
- Image optimization for LCP

Example 2: React Re-rendering Issues

User: "コンポーネントが不要に再レンダリングされる"

Claude loads:
✓ skill.md metadata (200 tokens)
✓ references/react-optimization.md (800 tokens)
Total: 1000 tokens

Provides:
- React.memo usage
- useMemo patterns
- React DevTools profiling
- State separation strategies

Example 3: Bundle Size Reduction

User: "バンドルサイズが大きすぎる"

Claude loads:
✓ skill.md metadata (200 tokens)
✓ references/bundle-optimization.md (600 tokens)
Total: 800 tokens

Provides:
- Code splitting techniques
- Tree shaking patterns
- Bundle analyzer usage
- Dynamic import strategies

---

🎯 Optimization Priorities

1. Measure First - Always use Lighthouse, Chrome DevTools before optimizing
2. User-Centric Metrics - Focus on Web Vitals (LCP, FID, CLS)
3. Progressive Enhancement - Start simple, optimize based on data
4. Avoid Premature Optimization - Only optimize problematic areas

---

✨ Key Takeaways

1. Measure First - Always measure before optimizing
2. User-Centric Metrics - Use Web Vitals as baseline
3. Section-based Learning - Load only relevant knowledge
4. Data-Driven Decisions - Optimize based on profiling data

---

Remember: "Make it work, make it right, make it fast - in that order"

For detailed information, see the specific section files listed above.