Execute pilot experiments from Pilot Spec YAML and return raw metrics.
npx skills add https://github.com/skyllwt/AutoSci --skill exp-pilot-run-skyllwt
Please help me install this Agent Skill. Skill: /exp-pilot-run Source: https://github.com/skyllwt/AutoSci/tree/main/.claude/skills/exp-pilot-run Command: npx skills add https://github.com/skyllwt/AutoSci --skill exp-pilot-run-skyllwt
It helps you execute a short, diagnostic pilot experiment from a Pilot Spec YAML without committing changes to wiki pages or performing a final pass/fail judgment.
experiments/pilot/{slug}.yaml, generates the required pilot code, and runs a reduced experiment for quick failure detection.--env local (direct GPU run) and --env remote (SSH deployment via tools/remote.py and config/server.yaml)./exp-pilot-eval.experiments/pilot/code/{slug}/.Run the pilot for idea slug "my-idea" on your local GPU by telling the AI: exp-pilot-run my-idea --env local.
High-intent search queries and answers about installing and using this skill.
To run a pilot experiment from a YAML spec on a remote GPU, execute the command with the `--env remote` flag. This deploys the generated pilot code via SSH using `tools/remote.py` and `config/server.yaml`.
A pilot experiment is a short, diagnostic training or evaluation run designed to detect obvious implementation failures. It gathers raw metrics from a reduced run without making wiki modifications or final pass/fail judgments.
You can execute a short diagnostic model evaluation run locally by providing your idea slug and using the `--env local` flag. The system reads the Pilot Spec YAML, generates code under `experiments/pilot/code/`, and runs it on your direct local GPU.
Yes, pilot experiment execution requires manual code review before launch. A manual-inspection safety gate enforces explicit user confirmation of the generated code and configuration before the diagnostic run starts.
After running experiment orchestration, you get outputs ready for evaluation including generated pilot code, per-seed JSON results, and execution logs. These are saved directly to the `experiments/pilot/code/{slug}/` directory.
duany049Execute pilot experiments from YAML specs and collect raw JSON results.
duany049Generate pilot code from Pilot Spec YAML and run diagnostic experiments.
czh-ee-2023Deploy and execute machine learning experiments across local and remote environments.
TeoSlayerDeploy a four-agent ML training pipeline coordinating data preparation, training, evaluation, and serving.