| name | notebook-orientation |
| description | Explain what a Jupyter notebook does by inspecting structure, summarizing workflow, and reading only key cells. Use when asked to walk through, summarize, or understand a notebook. |
| license | MIT |
| compatibility | opencode, claude-code, codex |
| metadata | {"audience":"data-scientists","workflow":"inspection"} |
What I do
I help you understand a Jupyter notebook without reading every cell. I use notebook-tools to:
- List the notebook structure with
list-cells
- Summarize the workflow with
summarize
- Read only the key cells that define the notebook's purpose
- Report sections, data flow, major outputs, and unknowns
When to use me
- "What does this notebook do?"
- "Walk me through this notebook"
- "Explain the structure of this notebook"
- "What are the main sections here?"
How I work
Step 1: Structure
notebook-tools list-cells --notebook <path> --pretty
This gives cell types, execution counts, and short summaries without pulling full content.
Step 2: Workflow Summary
notebook-tools summarize --notebook <path> --pretty
This returns a high-level workflow summary with major sections and key outputs.
Step 3: Targeted Reads
Only read cells that are essential to answer the question. Use:
notebook-tools read-cells --notebook <path> --index <N> --summary-mode compact --pretty
Or search for specific content:
notebook-tools search-cells --notebook <path> --query "<term>" --pretty
Step 4: Report
Report:
- Notebook mode:
file_only or live_session
- Major sections identified
- Data flow and key variables
- Major outputs
- Unknowns or risks
Rules
- Never read the entire notebook by default
- Always start with
list-cells before deeper reads
- Use
search-cells to find specific content before broad reads
- Report what was observed vs inferred
- In file-only mode, explicitly state that runtime validation is unavailable
Dependencies
Requires notebook-tools CLI installed. All commands need --notebook <absolute-path>.
