What problem does it solve? Go performance work fails when developers draw conclusions from single benchmark runs, misread pprof profiles, or cannot tell a real regression from CI noise. This Skill provides the full measurement methodology — writing correct benchmarks, profiling hot paths, and proving improvements with statistical rigor. ## Core Features & Use Cases - Benchmark Authoring: Write benchmarks with b.Loop() (Go 1.24+), memory tracking via b.ReportAllocs(), custom metrics via b.ReportMetric(), and table-driven sub-benchmarks. - Profiling & Analysis: Capture CPU, memory, and execution trace profiles directly from go test, then interpret them with pprof (top -cum, list, peek), escape analysis (-gcflags="-m"), and SSA/assembly inspection. - Statistical Comparison & CI Gating: Compare runs with benchstat (p-values, confidence intervals, filters, projections) and set up CI regression detection with benchdiff, cob, or gobenchdata, including noisy-neighbor mitigation and self-hosted runner tuning. - Use Case: After optimizing a JSON parser, run both versions with -count=10, compare with benchstat, and paste the statistically significant result into the commit body so reviewers can verify the claimed 33% speedup. ## Quick Start Ask the agent to write a benchmark for your hot function, run it with -count=10 and -benchmem, then compare the before and after results with benchstat.
