jupyter

name	jupyter
description	Run Python data-science work through Jupyter — execute notebooks headlessly, run cells against a live kernel, and convert notebooks to scripts/HTML. Keyless, local. Use when asked to "run this notebook", "execute these cells", "analyze this dataset in Python", "convert notebook to".
allowed-tools	["@coder","Bash","@read","Write"]
triggers	["jupyter","run this notebook","execute notebook","run these cells","analyze dataset","convert notebook","rodar o notebook","executar notebook","analisar os dados"]

Jupyter / Data Science

Local, keyless notebook execution. Detect: command -v jupyter python3 / Get-Command jupyter, python -ErrorAction SilentlyContinue. Install: pipx install jupyter or pip install jupyterlab nbconvert ipykernel.

Execute a notebook headlessly

jupyter nbconvert --to notebook --execute --inplace analysis.ipynb
jupyter nbconvert --to html --execute report.ipynb        # produce a shareable HTML

Read results with @read on the output, or inspect the executed .ipynb cells.

Run ad-hoc Python (no notebook)

For quick analysis, write a .py and run it via @coder exec python3 script.py — manage deps in a venv: python3 -m venv .venv && . .venv/bin/activate && pip install pandas matplotlib.

Convert

jupyter nbconvert --to script notebook.ipynb     # → .py
jupyter nbconvert --to pdf notebook.ipynb         # needs LaTeX (see paper-writing skill)

Live kernel (interactive sessions)

For a persistent kernel across steps, start jupyter console or use jupyter run. Keep the kernel alive only as long as needed; one-shot --execute is simpler for batch work.

Rules

• Use a project venv; don't pollute the system Python.
• For plots, save to a file (plt.savefig) and report the path — terminals can't show inline plots.
• State which packages you installed; never assume pandas/numpy are present without checking.