Build Notebook

Create, improve, or fix Jupyter notebooks iteratively.

Action: $ARGUMENTS

Actions

new - Create from scratch

Requires: --topic, --domain (ML, Probas, GenAI, GameTheory, SymbolicAI, Sudoku, Search), --level (intro, intermediate, advanced)

improve - Enhance existing notebook

Requires: notebook path, optional --quality (0-100, default 90), --focus (execution, pedagogy, content, structure)

fix - Correct execution errors

Requires: notebook path, optional --max-iter (default 5)

Iterative Workflow

while iteration < max_iterations and score < quality_target:
    1. Execute notebook (notebook-executor agent)
    2. Validate all aspects (notebook-validator agent)
    3. Plan corrective actions (prioritized by severity)
    4. Apply via appropriate sub-agent (enricher, cell-iterator, cleaner)
    5. Re-validate and measure improvement

Quality Scoring

Category	Weight	Criteria
Structure	15%	Valid JSON, metadata, cells
Syntax	15%	Valid code, markdown, LaTeX
Execution	30%	No errors, no timeouts
Pedagogy	25%	Progression, code/markdown ratio, explanations
Content	15%	Visualizations, formulas, conclusion

Sub-Agents Used

Agent	Role	Model
notebook-designer	Create initial structure	inherit
notebook-executor	Run the notebook	sonnet
notebook-validator	Evaluate quality	sonnet
notebook-enricher	Add pedagogical content	sonnet
notebook-cell-iterator	Fix specific cells	sonnet
notebook-cleaner	Clean structure	sonnet

Model Selection Strategy

• Use haiku for quick validation checks
• Use sonnet for enrichment and standard fixes
• Use inherit (or opus if available) for complex design and orchestration