benchmark

name	benchmark
description	Run evaluation benchmarks and report precision/recall
allowed-tools	Bash, Read
argument-hint	["--save"]

Benchmark Skill

Quick Start

1. Run auditor benchmark: python3 scripts/quality/benchmark_cairo_auditor.py --cases evals/cases/cairo_auditor_benchmark.jsonl --output /tmp/bench-auditor.md --min-precision 0.90 --min-recall 0.90
2. Run contract benchmark: python3 scripts/quality/benchmark_contract_skills.py --cases evals/cases/contract_skill_benchmark.jsonl --output /tmp/bench-contracts.md --min-precision 0.95 --min-recall 0.95
3. If $ARGUMENTS contains --save, append --save to both commands to copy scorecards into evals/scorecards/.
4. Report precision/recall and fail conditions per suite.
5. Follow the full workflow: Benchmark Workflow

When to Use

• You changed skills, references, datasets, or evaluation logic.
• You need comparable KPI numbers before/after a PR.

When NOT to Use

• You only need schema/structure validation (use validate skill instead).
• Benchmark inputs or dependencies are missing.