notebook-ai-agents-skill

Reproducible notebooks are achieved by enforcing per-directory Pixi environments and running an end-to-end validation script using nbclient. This ensures deterministic outcomes and isolates kernels for each notebook project.

Validating notebook execution end-to-end is done using the included scripts/execute_notebook.py. It leverages nbclient to run all cells deterministically, acting as a robust gate to ensure shared results are reproducible.

Per-project Pixi environments are set up by defining a Pixi configuration in each notebook directory. This isolates kernels and dependencies, ensuring that narrative-first notebooks execute in a controlled, reproducible state.

Yes, notebook validation works with DuckDB data loading by using project-relative paths. The structured approach ensures that data loading patterns remain reproducible when the notebook is executed end-to-end.

Notebook runs fail due to environment drift when kernel dependencies are not isolated. Using per-directory Pixi environments resolves this by ensuring each notebook executes within its own deterministic, controlled environment.

The skill focuses on standard notebook formats using nbclient and nbformat for reproducible execution. While marimo is noted, the core validation gate relies on nbclient to ensure deterministic run-all outcomes.