jupyter-notebook

name	jupyter-notebook
description	Create Jupyter notebooks with correct JSON format.

IPYNB Notebook Skill

You are an expert at creating, editing, and manipulating Jupyter notebooks programmatically.

⚠️ CRITICAL: NEVER Write Raw Notebook JSON

DO NOT write .ipynb files by directly outputting JSON. LLMs frequently produce invalid JSON by turning \n escape sequences into literal backslash-n text, which breaks the file (all code appears on one line).

ALWAYS generate notebooks using a Python script that builds the notebook as a Python data structure and serializes it with json.dump(). This guarantees valid JSON output.

Creating Notebooks — The Only Correct Way

Step 1: Write a Python generator script

import json

notebook = {
    "nbformat": 4,
    "nbformat_minor": 0,
    "metadata": {
        "colab": {"provenance": []},
        "kernelspec": {"name": "python3", "display_name": "Python 3"},
        "language_info": {"name": "python"}
    },
    "cells": []
}

def md_cell(cell_id, source_lines):
    """Create a markdown cell. source_lines is a list of plain strings (no \\n needed)."""
    source = [line + "\n" for line in source_lines[:-1]] + [source_lines[-1]] if source_lines else []
    return {"cell_type": "markdown", "source": source, "metadata": {"id": cell_id}}

def code_cell(cell_id, source_lines):
    """Create a code cell. source_lines is a list of plain strings (no \\n needed)."""
    source = [line + "\n" for line in source_lines[:-1]] + [source_lines[-1]] if source_lines else []
    return {
        "cell_type": "code",
        "source": source,
        "metadata": {"id": cell_id},
        "execution_count": None,
        "outputs": []
    }

# --- Build cells ---

notebook["cells"].append(md_cell("intro", [
    "# My Notebook",
    "",
    "Description here."
]))

notebook["cells"].append(code_cell("imports", [
    "import torch",
    "import numpy as np",
    "",
    "print('Ready!')"
]))

# --- Write notebook ---

with open("notebook.ipynb", "w", encoding="utf-8") as f:
    json.dump(notebook, f, indent=2, ensure_ascii=False)

print("✓ Notebook created successfully")

Step 2: ALWAYS validate and auto-fix after generation

After generating any notebook, always run this validation and auto-fix script. This catches the most common LLM failure: writing literal \n text instead of actual newline characters.

import json, sys

path = "notebook.ipynb"

# --- Load ---
try:
    with open(path, "r") as f:
        nb = json.load(f)
except json.JSONDecodeError as e:
    print(f"✗ INVALID JSON: {e}", file=sys.stderr)
    sys.exit(1)

# --- Structural validation ---
assert "cells" in nb, "Missing 'cells' key"
assert nb.get("nbformat") == 4, "Bad nbformat"

fixed = False
for i, cell in enumerate(nb["cells"]):
    assert "cell_type" in cell, f"Cell {i}: missing cell_type"
    assert "source" in cell, f"Cell {i}: missing source"
    assert isinstance(cell["source"], list), f"Cell {i}: source must be a list"

    if cell["cell_type"] == "code":
        if "outputs" not in cell:
            cell["outputs"] = []
            fixed = 
            cell:
            cell[] = 
            fixed = 

    
    
    
    new_source = []
     line  cell[]:
         (line, ), 
           line:
            fixed = 
            line = line.replace(, )
        new_source.append(line)
    cell[] = new_source


 fixed:
     (path, , encoding=)  f:
        json.dump(nb, f, indent=, ensure_ascii=)
    ()
:
    ()

This validation + auto-fix step is MANDATORY. Never skip it.

Helper Functions Reference

Use these helper functions in every notebook generation script:

def md_cell(cell_id, lines):
    """Markdown cell from a list of plain strings (one per line)."""
    src = [l + "\n" for l in lines[:-1]] + [lines[-1]] if lines else []
    return {"cell_type": "markdown", "source": src, "metadata": {"id": cell_id}}

def code_cell(cell_id, lines):
    """Code cell from a list of plain strings (one per line)."""
    src = [l + "\n" for l in lines[:-1]] + [lines[-1]] if lines else []
    return {
        "cell_type": "code", "source": src,
        "metadata": {"id": cell_id}, "execution_count": None, "outputs": []
    }

Why plain strings? You pass ["import torch", "import numpy"] — the helper adds \n between lines programmatically. This eliminates the most common source of invalid JSON.

Editing Existing Notebooks

import json

with open('notebook.ipynb', 'r') as f:
    nb = json.load(f)

# Find and modify cell by ID
for cell in nb['cells']:
    if cell.get('metadata', {}).get('id') == 'target_id':
        cell['source'] = ["new line 1\n", "new line 2"]
        break

# Insert a cell at position
nb['cells'].insert(2, code_cell("new_cell", ["print('inserted')"]))

# Delete a cell by ID
nb['cells'] = [c for c in nb['cells'] if c.get('metadata', {}).get('id') != 'delete_me']

with open('notebook.ipynb', 'w') as f:
    json.dump(nb, f, indent=2, ensure_ascii=False)

Colab Form Fields

code_cell("config", [
    '#@title Configuration { display-mode: "form" }',
    '',
    'MODEL_NAME = "gpt2"  #@param {type:"string"}',
    'BATCH_SIZE = 32  #@param {type:"integer"}',
    'USE_GPU = True  #@param {type:"boolean"}',
    'MODE = "A"  #@param ["A", "B", "C"]',
])

Notebook Patterns

Setup Cell

code_cell("setup", [
    "#@title Setup",
    "!pip install -q package1 package2",
    "",
    "import package1",
    "import package2",
    "",
    "print('✓ Setup complete')",
])

Progress Display

code_cell("train_loop", [
    "from tqdm.notebook import tqdm",
    "",
    "for i in tqdm(range(100)):",
    "    # work",
    "    pass",
])

Quality Checklist

Before finalizing a notebook:

• Generated via Python script with json.dump() — never raw JSON
• Validation + auto-fix script ran successfully after generation
• All cells have unique IDs
• Markdown cells have proper headers and formatting
• Code cells are logically ordered
• Imports are at the top or in a setup cell
• Config values use Colab form fields where appropriate
• Error handling for common failures
• Clear output messages (✓ for success, ⚠️ for warnings)
• Section dividers between major parts