Design, diagnose, and optimize software using user-facing performance budgets, systems analysis, representative measurement, hardware-aware algorithms, bounded allocation, numerical contracts, and SIMD. Use for slow code, latency, throughput, CPU or memory pressure, queues, capacity, performance reviews, new-system design, and regression prevention. Includes Rust and C++ guidance.
This skill should be used when the user asks to "add benchmarks", "set up criterion", "add a benchmark gate to CI", "track performance regressions", "add github-action-benchmark", "add benchmark floors", "publish a benchmark series to gh-pages", "count allocations in tests", "why is the benchmark job green but empty", "make the bench job faster", "my benchmarks are noisy in CI", "cargo bench in GitHub Actions", "criterion benchmarks", "benchmark regression", "performance gate", or is deciding whether a benchmark number proves a performance claim. Covers criterion harness setup, a github-action-benchmark regression gate with committed floors, allocation-count series, and measurement discipline.