Perf

Performance Audit

Response Time / Throughput

• Identify hot paths - what runs most frequently?
• Flag any blocking I/O on critical paths (network calls, disk reads, DB queries in request handlers)
• Check for N+1 query patterns
• Verify appropriate use of caching
• Check connection pooling for databases and HTTP clients

Memory

• No object allocations in tight loops or hot paths where avoidable
• Resources (connections, streams, file handles) properly closed
• No unbounded growth of collections, caches, or queues
• Check for closure/callback reference leaks
• Verify garbage collection pressure is manageable

CPU

• Algorithmic complexity appropriate? Flag O(n^2) or worse in hot paths
• No redundant computation (computing the same thing multiple times)
• Heavy computation offloaded from main/UI thread
• Regex compiled once, not per-invocation
• String operations efficient (no repeated concatenation in loops)

Concurrency

• Thread pool sizes appropriate for workload
• No contention on shared locks in hot paths
• Async operations used where I/O-bound (not CPU-bound)
• Connection pools sized correctly

Database

• Queries use indexes? Check EXPLAIN plans for common queries
• No full table scans on large tables
• Pagination for large result sets
• Batch operations where possible (not row-by-row)
• Connection pool not exhausted under load

Network

• Payloads sized appropriately? Compress large responses
• Connection reuse (keep-alive, HTTP/2)
• Timeouts configured (no infinite waits)
• Retry logic with backoff (no retry storms)

Utility Scripts

• scripts/find_hot_spots.sh [dir] - Find long functions (100+ lines), deeply nested loops, perf-related TODOs, and large files.

Output

[CRITICAL] file:line - description - impact - fix
[WARNING]  file:line - description - impact - fix
[OK]       area - looks good because...

End with: biggest risks, top 3 priorities, estimated impact of each fix.

Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: pvnarp
• Source: pvnarp/agent-skills
• License: MIT

Install and usage instructions live in the source repository linked above.