AgentSkillsCN

optimizing-code

在不改变行为的情况下提升代码性能。当代码无法满足延迟/吞吐量要求时使用。涵盖性能分析、缓存和算法优化。

SKILL.md
--- frontmatter
name: optimizing-code
description: Improve code performance without changing behavior. Use when code fails latency/throughput requirements. Covers profiling, caching, and algorithmic optimization.
allowed-tools: Read, Write, Edit, Bash, Glob, Grep

Optimizing Code

The Optimization Hat

When optimizing, you improve performance without changing behavior. Always measure before and after.

Golden Rules

  1. •Measure First: Never optimize without a benchmark
  2. •Profile Before Guessing: Find the actual bottleneck
  3. •Optimize the Right Thing: Focus on the critical path
  4. •Measure After: Verify the optimization worked

Workflows

  • • Benchmark: Establish baseline performance metrics
  • • Profile: Identify the actual bottleneck
  • • Hypothesize: What optimization will help?
  • • Implement: Make the change
  • • Measure: Verify improvement
  • • Document: Record the optimization and results

Common Optimizations

Algorithm Complexity

  • •Replace O(n²) with O(n log n) or O(n)
  • •Use appropriate data structures (Set for lookups, Map for key-value)

Caching

typescript
// Memoization
const cache = new Map<string, Result>();

function expensiveCalculation(input: string): Result {
  if (cache.has(input)) {
    return cache.get(input)!;
  }
  const result = /* expensive work */;
  cache.set(input, result);
  return result;
}

Database Queries

  • •Add indexes for frequently queried columns
  • •Avoid N+1 queries (use eager loading)
  • •Use pagination for large result sets

Memory

  • •Avoid creating unnecessary objects in loops
  • •Use streaming for large files
  • •Release references when done

Profiling Tools

bash
# Node.js
node --prof app.js
node --prof-process isolate-*.log

# Python
python -m cProfile -s cumtime script.py

# Go
go test -bench=. -cpuprofile=cpu.prof
go tool pprof cpu.prof

Anti-Patterns to Avoid

  • •Premature optimization (no benchmark)
  • •Micro-optimizations (negligible impact)
  • •Optimizing cold paths
  • •Sacrificing readability for minor gains