AI agent skill · by @rileyhilliard

Optimizing Performance

Measure-first performance optimization that balances gains against complexity. Use when addressing slow code, profiling issues, or evaluating optimization trade-offs.

Archived 2026-10-23 · Source updated 2026-08-18 · License: MIT. Check the original repository for the latest version.

SKILL.md instructions

# Optimizing Performance

**Core principle:** Readable code that's "fast enough" beats complex code that's "optimal". Measure first.

**Focus area:** Use an explicitly named target if one was given. Otherwise, run `git diff` and focus on unstaged changes. If no unstaged changes exist, ask the user what to optimize.

## The Golden Rule

```
IF optimization reduces complexity AND improves performance → ALWAYS DO IT
IF optimization increases complexity → Only if 10x faster OR fixes critical UX (>16ms UI, >100ms input)
```

## Win-Win Optimizations (Always Do)

**Multiple loops → Single loop:**
```javascript
// ❌ Three passes
const ids = users.map(u => u.id);
const active = users.filter(u => u.active);

// ✅ One pass
const { ids, active } = users.reduce((acc, u) => {
  acc.ids.push(u.id);
  if (u.active) acc.active.push(u);
  return acc;
}, { ids: [], active: [] });
```

**Nested loops → Hash map (O(n²) → O(n)):**
```javascript
// ❌ O(n²)
const matched = orders.filter(o => users.some(u => u.id === o.userId));

// ✅ O(n)
const userIds = new Set(users.map(u => u.id));
const matched = orders.filter(o => userIds.has(o.userId));
```

## High-Value Optimizations

| Pattern | When | Fix |
|---------|------|-----|
| Virtualization | Lists >1000 items | react-window, tanstack-virtual |
| Memoization | >5ms calc OR unnecessary re-renders | `useMemo`, `React.memo` |
| Batching | Multiple state updates | Single setState, bulk INSERT |
| Lazy loading | Large dependencies | `import('./heavy-lib')` |

## Red Flags

- Optimizing without benchmark data
- Micro-optimizing <16ms code
- Adding complexity for minimal gain
- Optimizing infrequently-run code