# Hyperfine Benchmarking

> Use this skill to benchmark shell commands reliably with warmup runs, statistical summaries, and exportable artifacts using hyperfine.

- Skill: `akillness/hyperfine-benchmarking` (Agent Skill)
- Install (CLI): `npx skillmds@latest add akillness/hyperfine-benchmarking`
- Raw SKILL.md: https://api.skillmd.com/api/skills/akillness/hyperfine-benchmarking/raw
- Safety review: pending (external: skill-scanner PASS, skillspector PASS)
- Works with: Claude Code, Claude.ai, OpenAI Codex
- Category: AI & ML
- Author: akillness (https://skillmd.com/u/akillness)
- Updated: 2026-09-08
- Page: https://skillmd.com/skills/akillness/hyperfine-benchmarking

---


# hyperfine-benchmarking

Benchmark CLI commands with reproducible methodology instead of one-off `time` output.

## When to use this skill
- Compare two or more command variants (`tool-a` vs `tool-b`)
- Validate performance impact before/after script changes
- Attach benchmark evidence to PRs

## Instructions
1. Confirm tool availability.
2. Keep input/workdir/environment stable across compared commands.
3. Run warmups and enough runs for stable variance.
4. Export JSON/markdown outputs for review.
5. Summarize relative speedup + risk notes.

## Examples

### Availability check
```bash
hyperfine --version
```

### Two-command comparison
```bash
hyperfine \
  --warmup 3 \
  --min-runs 10 \
  'cmd_a --with flags' \
  'cmd_b --with flags'
```

### Parameter sweep
```bash
hyperfine \
  --warmup 3 \
  --parameter-list mode fast,balanced,thorough \
  'mytool --mode {mode} input.txt'
```

### Export artifacts
```bash
hyperfine \
  --warmup 3 \
  --min-runs 10 \
  --export-json benchmark.json \
  --export-markdown benchmark.md \
  'cmd_a' 'cmd_b'
```

## Best practices
- Prefer relative speedup and confidence ranges over single-run claims.
- Do not compare commands with different semantics unless outputs are normalized.
- If variance is high, increase runs or reduce background noise before concluding.
- Record dataset/path and exact command strings in PR text.

## References
- hyperfine project: https://github.com/sharkdp/hyperfine
- Docs: https://github.com/sharkdp/hyperfine#readme