skills

Overview

This skill manages the audit pipeline for machine learning experiments by maintaining one Jupytext percent Python file per experiment inside an audit/ directory. It executes these files using a bundled in-process IPython runner to extract metrics and check summaries, functioning as a read-only viewer over skore projects. It enforces strict separation between durable source audit files and ephemeral output digests, replacing ad-hoc scratch scripts with standardized audit execution.

Capabilities

• ▸In-process IPython cell execution and markdown digest generation
• ▸Read-only inspection of skore Projects and reports
• ▸Pre-flight checklist validation for experiment stems and environment prerequisites
• ▸Structured four-way pairing enforcement across journal notes, experiment scripts, smoke tests, and audit files

Best for

Extracting metrics and evaluation checks as the primary step after running an ML experiment, Generating human-readable narratives and digests from past experiment skore reports, Re-executing and re-auditing experiments when existing code or metrics are updated

Works with

Security considerations

Enforces a strict read-only security contract against skore projects and dataset directories, prohibiting any writes, evaluations, or state mutations during the audit phase.

Categories