Solo Company Performance Architect

Problem

This skill addresses code that works but is slow and hard to diagnose. It focuses on performance rather than style: algorithmic complexity, blocking I/O, SQL queries, caching, and concurrency limits. You can provide a function, endpoint, SQL statement, or directory path, then receive a structured report with location, severity, fix code, and expected impact instead of vague advice.

How It Works

The workflow combines automated scanning and AI analysis. When file paths are available, it can run quick or deep scans for patterns such as db.query inside loops, fetch in repeated calls, synchronous file I/O, unbounded SELECT * FROM, and global caches without eviction. If only a code snippet is provided, it moves directly into AI review. It then applies L0, L1, or L2 depth: single functions get direct complexity and I/O checks; multi-function or refactoring tasks add boundary and database interaction review; architecture-level issues add concurrency, memory, and scalability analysis. Typical output includes:
- Anti-pattern detection: N+1 queries, blocking I/O, and memory-leak risks
- Fix code: runnable code changes rather than pseudocode
- Quantified impact: expected latency, QPS, or query-time changes
- Risks and validation: consistency or compatibility caveats, with EXPLAIN, cache hit-rate tests, or load-test script suggestions

Boundaries

It is useful for code-level performance review, SQL optimization guidance, cache strategy, and async task design across Python, Java, JavaScript, Go, Rust, and C#. It does not replace live load testing, production debugging, JVM or kernel tuning; SQL advice needs DDL and EXPLAIN, while architecture decisions need QPS, latency, and system topology. When data volume is small or performance is already sufficient, avoid over-optimization.