gongshichen/luminacode/jupyter-notebook

Work with notebooks as structured JSON, not plain text. Preserve metadata unless there is a clear reason to change it, and keep edits scoped to the requested cells or narrative.

For new notebooks, choose the shape first:

• experiment: objective, setup, data loading, method, results, checks, conclusion
• tutorial: goal, prerequisites, step-by-step sections, runnable examples, recap

For existing notebooks:

• inspect cell order and metadata before editing
• keep code cells small and runnable
• avoid large noisy outputs unless the user needs them
• clear or summarize stale outputs when they would mislead readers
• execute top-to-bottom when the environment supports it

Prefer notebook-aware tools or JSON libraries for edits. If execution is not possible, say exactly what was not run and provide the command or kernel setup needed to validate locally.