AgentSkillsCN

run-benchmark

运行基准测试套件，并如实汇报测试结果。

SKILL.md
--- frontmatter
name: run-benchmark
description: Run the benchmark suite and report results
user-invocable: true
allowed-tools: Bash, Read

Run Benchmark

Execute the benchmark suite and report results.

Steps

  1. •First verify tests pass with python -m pytest tests/ -v
  2. •If the user specified arguments, use those; otherwise run with defaults:
    code
    python -m app.run_benchmark --alpha 0.0 0.5 1.0
  3. •Read outputs/benchmark.csv and outputs/benchmark.md
  4. •Report results to the user covering:
    • •L2/L-infinity error for each solver
    • •Wall-clock time
    • •Best solver and why
    • •Differences from previous run (if any)