Golang Performance

A methodical performance optimization skill that enforces the "profile first, optimize second" discipline. It gives you a decision tree from pprof signals to concrete fixes: too many allocations? Memory patterns. CPU hot loop? Inlining and cache locality. The skill splits into architecture review mode (parallel sub-agents for memory, I/O, and algorithmic concerns) and optimize mode (iterative benchstat cycle, one change at a time). Includes deep-dive references for allocation reduction, GC tuning, struct alignment, and HTTP transport config. Honest take: the "no unsafe without 10% proof" rule and the mandate to paste benchstat output in commit messages will save you from cargo-cult optimizations. Requires go, benchstat, and works best when you already have profiling data pointing at a real bottleneck.