Common Performance Engineering Agent Skill

Performance Engineering Standards

Priority: P0 (CRITICAL)

Workflow

• Baseline: Profile before changing anything — measure CPU, memory, and latency.
• Identify: Find top bottleneck (N+1 query, hot loop, memory leak).
• Fix: Apply targeted optimization from sections below.
• Verify: Re-profile to confirm improvement and check for regressions.

Resource Management

• Memory Efficiency:
• Avoid memory leaks: explicit cleanup of listeners, observers, and streams.
• Optimize data structures: Set for lookups, List for iteration.
• Lazy Initialization: Initialize expensive objects only when needed.
• CPU Optimization:
• Aim for O(1) or O(n); avoid O(n^2) in critical paths.
• Offload heavy computations to background threads or workers.
• Memoize pure, expensive functions.

See implementation examples for memoization and batching patterns.

Network & I/O

• Payload Reduction: Use efficient serialization (Protobuf, JSON minification) and compression (gzip/br).
• Batching: Group multiple small requests into single bulk operations.
• Caching: Implement multi-level caching (Memory -> Storage -> Network) with appropriate TTL and invalidation.
• Non-blocking I/O: Always use asynchronous operations for file system and network access.

UI/UX Performance

• Minimize Main Thread Work: Keep animations and interactions fluid by offloading to workers.
• Virtualization: Use lazy loading or virtualization for long lists/large datasets.
• Tree Shaking: Ensure build tools remove unused code and dependencies.

Monitoring & Testing

• Benchmarking: Write micro-benchmarks for performance-critical functions.
• SLIs/SLOs: Define Service Level Indicators (latency, throughput) and Objectives.
• Load Testing: Test system behavior under peak and stress conditions.

Anti-Patterns

• No premature optimization: Profile first, fix proven bottlenecks only.
• No N+1 queries: Always batch and paginate data-access operations.
• No synchronous I/O on main thread: Async all file/network access.

References

• Implementation Patterns — profiling patterns, benchmark setup

Canonical response anchors

When this skill applies, preserve the following domain terminology or equivalent concrete examples in the answer when relevant:

• lazy

• Additional task-grounded exact anchors: premature