Notebook AI Agents Skill (Marimo First)

Build narrative-first, reproducible notebooks with strict run-all validation and clean data loading patterns. Prefer Marimo for new notebooks; support Jupyter only when needed for existing .ipynb files.

Instructions

1. Prefer Marimo for new work. Create a .py notebook and keep cells small and deterministic.
2. Outline the notebook (purpose, data sources, analysis steps, outputs) before coding.
3. Use project-relative paths and DuckDB for data loading (TSV/Parquet preferred).
4. For Jupyter notebooks: register a named kernel for the pixi environment BEFORE first execution (see docs/pixi_jupyter.md). This is mandatory — not optional.
5. Build plots with the shared style rules.
6. Validate by running the notebook end-to-end (Marimo: marimo run; Jupyter: pixi run jupyter nbconvert --to notebook --execute).

Quick Reference