| name | experiment-run |
| description | Use this skill when the task requires executing a local experiment command, capturing logs, recording structured results, and updating Catalyst4Sci experiment state. |
| category | experiment |
| recommended_for | ["run_experiment","baseline_experiment"] |
| tools | ["bash","read"] |
| risk_level | medium |
Experiment Run
Use this skill when the current task is to run a baseline experiment, execute a follow-up experiment, or reproduce a prior run.
Workflow
- Confirm the experiment objective, command, workspace, timeout, and expected metrics.
- Run the experiment through the project execution path instead of inventing an ad hoc shell workflow.
- Capture stdout, stderr, exit code, timeout status, and result artifact paths.
- Update the structured experiment record and checkpoint after execution.
- If the run fails, classify whether the failure is environmental, command-related, or idea-related before retrying.
Output
Return a concise execution summary that includes:
- What command ran
- Whether it succeeded, failed, or timed out
- Where logs and result artifacts were written
- Which metric or signal should influence the next action
References