Jupyter Notebook

Overview

Use this skill to produce notebooks another engineer can execute and trust without hidden assumptions.

Scope Boundaries

• Use this skill when the task matches the trigger condition described in description.
• Do not use this skill when the primary task falls outside this skill's domain.

Shared References

• Notebook structure guidance:
  • references/notebook-structure.md
• Reproducibility and sanitization rules:
  • references/reproducibility-and-sanitization-rules.md

Templates And Assets

• Notebook run log template:
  • assets/notebook-run-log-template.md
• Notebook result summary template:
  • assets/notebook-summary-template.md
• Delivery checklist:
  • assets/notebook-delivery-checklist.md

Inputs To Gather

• Notebook purpose (exploration, debug, tutorial, verification).
• Runtime constraints (Python version, package policy, data access).
• Expected deliverable shape (single notebook or multi-notebook set).
• Sharing boundary (internal-only vs external audience).

Deliverables

• Executable notebook with deterministic order.
• Runtime/dependency assumptions and execution log.
• Decision-grade summary linked to output cells.
• Sanitized artifact for sharing when required.

Workflow

1. Define audience, decision question, and reproducibility constraints.
2. Create scaffold with scripts/new_notebook.py when useful.
3. Structure notebook sections per references/notebook-structure.md.
4. Execute from fresh kernel and record results in assets/notebook-run-log-template.md.
5. Summarize findings with assets/notebook-summary-template.md.
6. Validate shareability via assets/notebook-delivery-checklist.md.

Scripts

• Create experiment scaffold:
  • python3 scripts/new_notebook.py --kind experiment --title 'My Experiment' --out output/notebooks/my-experiment.ipynb
• Create tutorial scaffold:
  • python3 scripts/new_notebook.py --kind tutorial --title 'My Tutorial' --out output/notebooks/my-tutorial.ipynb

Quality Standard

• Notebook runs top-to-bottom from fresh kernel.
• Claims are tied to explicit output evidence.
• Runtime and data assumptions are reproducible.
• Shared outputs are sanitized for secrets and personal data.

Failure Conditions

• Stop when runtime/data preconditions cannot be specified precisely.
• Stop when repeated runs produce unstable outputs without explanation.
• Stop external sharing when sensitive output cannot be sanitized.