benchmark-cilisted

# Benchmark & CI Skill Determine mode from user intent, then execute. | User wants to... | Mode | Key phrases | |-------------------|------|-------------| | Run benchmarks and check regressions | **Runner** (below) | run benchmarks, benchmark check, check for regressions, update baseline, perf check | | Add benchmark regression gating to CI | **Gate** (below) | benchmark gate, ci benchmark, add benchmark to ci, gating benchmarks | | Audit GitHub Actions workflow YAML | **YAML Review** (below) | review github actions, audit ci yaml, check workflow versions, outdated actions | If ambiguous, ask which task they need. --- ## Mode: Runner Run benchmarks and compare against the stored baseline in `benchmarks/baselines.json`. ### Determine intent - "update baseline" or "save baseline" -> **Update Baseline** path - Otherwise -> **Run and Compare** path ### Run and Compare 1. Run the benchmark suite (BenchmarkDotNet, Release config) with an optional `--filter {ClassName}`, then diff the results against `benchmarks/baselines.json` 2. Interpret output — Status per benchmark: `OK`, `IMPROVED`, `REGRESSED`, `ALLOC_UP`, `NEW` 3. If `REGRESSED`/`ALLOC_UP`: report table with baseline/current/change, suggest investigating 4. If clean: report "All benchmarks clean. {N} improved, {M} stable." ### Update Baseline 1. Confirm intent (overwrites `benchmarks/baselines.json`) 2. Re-run the suite and write the current results to `benchmarks/baselines.json` 3. Stage: `git add benchmarks/