What problem does it solve? Deciding where files should live in a machine learning experimentation project is error-prone and inconsistent: reusable code, experiment scripts, design notes, reports, and scratch probes often end up scattered or mixed together. This Skill enforces a clear, repeatable workspace layout so every experiment is traceable and every file has a single owner. ## Core Features & Use Cases - Workspace layout detection and scaffolding: Detects existing project conventions (pyproject.toml, experiments/, journal/, notebooks) and glues to them, or scaffolds a default layout with src/<pkg>/, experiments/, journal/, tests/smoke/, audit/, scratch/, and reports/. - File-creation rules and stem pairing: Enforces one jupytext # %% script per experiment, four-way stem pairing across journal, experiments, smoke tests, and audit files, and a design-note-first workflow. - Decision gates before scaffolding: Requires structured user confirmation for package name, tabular library (pandas vs polars), environment manager, and skore Project mode (local, hub, or mlflow) before creating manifests or templates. - Use Case: When starting a new ML project, invoke this Skill to scaffold an installable src package, a baseline experiment script wired to a skore Project, a journal placeholder, and a gitignore — all after confirming the key decisions with you. ## Quick Start Ask the agent to set up a new ML experimentation workspace for your project and confirm the package name, tabular library, environment manager, and skore project mode when prompted.
