run-bench

Goal

Run benchmark tests to measure code performance and compare performance between implementations

Scope

Use for measuring function performance, comparing different implementations, detecting performance regressions, and optimizing code based on benchmark results

Execute

Pre-Run: always run /deep-review first — run-* skills must be reviewed/assessed before the main action, never skip; if findings are blockers, fix or report them before running (bench)

1. Setup Environment

Goal: Setup Environment

1. Run /run-install to install dependencies and benchmark libraries (e.g. vitest bench, tinybench)
2. Check that benchmark files exist in the project (e.g. *.bench.ts, bench/ directory)
3. Check the benchmark script in package.json (e.g. bun run bench)
4. If baseline results exist, record them for comparison

2. Run Benchmarks

Goal: Run Benchmarks

1. Run bun run bench or the script defined in package.json
2. To run only specific benchmarks, provide a pattern (e.g. bun run bench -- --grep "pattern")
3. Wait for the benchmarks to finish and record all results
4. Run /report to format the results

3. Analyze Results

Goal: Analyze Results

1. Review each benchmark's result and identify the slow and fast benchmarks
2. Run /deep-review per deep-review-codebase-everything/SKILL.md to analyze whether empirical growth matches theoretical complexity
3. Find benchmarks with high variance or outliers
4. Identify potential optimizations for slow benchmarks
5. If variance is high, re-run to confirm the results

4. Compare With Baseline

Goal: Compare With Baseline

1. Compare the results with the baseline if one exists
2. Identify regressions (slower) and improvements (faster)
3. Record results and action items for optimization
4. Run /report to summarize the results as a table

Rules

1. Benchmark Best Practices

• Each benchmark must be isolated
• Run many times for consistency
• Use realistic data representative of real use cases
• Minimize test framework overhead

2. Metrics To Collect

• ops/sec - operations per second
• avg time - average execution time
• min/max - min/max execution time
• samples - number of samples

3. Conditional Execution

• If the project has no benchmark files, skip the running steps
• If there are no baseline results, skip the comparison steps
• If a monorepo has multiple workspaces, run benchmarks one workspace at a time

Expected Outcome

• Benchmarks run successfully
• Metrics fully recorded (ops/sec, avg time, min/max, samples)
• Regressions and improvements identified
• Action items for optimization recorded