minimal-run-and-audit

To capture execution evidence for reproducibility, run your short non-training commands through a script that executes them, records outputs, detects git status diffs, and generates a structured bundle with summaries, command logs, and patch documentation.

A reproducibility evidence bundle includes SUMMARY.md, COMMANDS.md, LOG.md, and PATCHES.md. These files document execution logs, command history, and patch notes generated when repository files are modified during the run.

Yes, you can generate patch notes from git status diffs during smoke tests. The skill tracks repository changes with git status snapshots to record patches when repository files are modified, including them in the repro_outputs bundle.

The best way to standardize inference run outputs for auditability is to execute commands through a normalization script that captures outputs, logs commands, and produces a reproducibility-ready package with structured markdown documentation.

The skill supports batch testing commands as indicated by its keyword coverage. It executes short non-training commands and normalizes their outputs into a standardized repro bundle, applicable to smoke tests, inference runs, or evaluation commands.