Each SKILL.md is a self-contained recipe an agent (or human) can follow end-to-end. cat the file to read it.
| Skill | When to use |
|---|---|
| run-benchmark | Launch a benchmark task locally against a user-provided inference server |
| check-inference | Probe a model inference WebSocket server and validate the response |
| convert-dataset | Convert robot-trajectory datasets between formats (agibot v1 → LeRobot v2.1) |
For the canonical CLI / configs / scoring contract, see ../../../.agent/geniesim_benchmark.md.
For the GitHub-facing intro, see ../README.md.