marimo-notebook

Marimo Notebook Skill

Running & Editing

# Edit interactively in browser
uv run marimo edit <notebook.py>

# Run as web app (code hidden)
uv run marimo run <notebook.py>

# Run as plain script (non-interactive, for testing/CI)
uv run <notebook.py>

# Lint for common mistakes — always run before handing back to user
uvx marimo check <notebook.py>
uvx marimo check --fix <notebook.py>

When to use this skill

• Creating .py notebook files for interactive data exploration
• Building reactive dashboards/apps with marimo
• Converting Jupyter notebooks (.ipynb) to marimo

Ask clarifying questions if:

• User mentions "Cell.run()", "reusable cells", or "testing cells" (API use case, different patterns)
• User wants to "pass variables between cells" with parameters/returns (wrong mental model)

---

File Format

Marimo notebooks are .py files. Each cell is a function decorated with @app.cell. Marimo resolves dependencies via static analysis of the cell body — it reads what variables you assign and reference, then builds a DAG automatically.

When marimo saves, it auto-generates function parameters and return tuples as serialization metadata. You don't need to manage these yourself. Write def _(): for all cells. Marimo handles the rest.

# /// script
# requires-python = ">=3.12"
# dependencies = [
#     "marimo",
#     "pandas",
#     "numpy",
# ]
# ///

import marimo

__generated_with = "0.19.7"
app = marimo.App(width="medium")

@app.cell
def _():
    import marimo as mo
    import pandas as pd
    import numpy as np

@app.cell
def _():
    # df is automatically available to all other cells
    df = pd.read_parquet("data/input.parquet")

@app.cell
def _():
    # Reference df directly — marimo knows the dependency
    filtered = df[df["value"] > threshold.value]
    mo.md(f"**{len(filtered)} rows** after filtering")

if __name__ == "__main__":
    app.run()

PEP 723 inline dependencies (# /// script block) are added automatically by marimo edit --sandbox but should always be present. They enable --sandbox mode for isolated execution.