The Agent Skills Directory

Indirect Prompt Injection Surface: The skill transforms user-provided titles and descriptions into executable notebook cells. While utilizing templates provides a controlled structure, the conversion of natural language into code and markdown cells constitutes an inherent surface for indirect instructions.
Ingestion points: User-provided notebook titles and objectives are processed within SKILL.md and scripts/new_notebook.py.
Boundary markers: The skill employs predefined templates in the assets/ directory to maintain consistency, although it does not use explicit boundary markers for user-supplied strings.
Capability inventory: The skill performs file writing via scripts/new_notebook.py and supports command execution through uv run as described in the workflow.
Sanitization: A slugify function in scripts/new_notebook.py sanitizes titles for filenames, while the notebook's markdown content relies on the agent's internal generation logic.
External Package Installation: The documentation suggests the installation of well-known packages, such as jupyterlab and ipykernel, via the uv package manager. These are standard tools within the data science ecosystem.
Command Execution: The workflow requires executing a local Python helper script (new_notebook.py) using uv run to scaffold notebooks from templates. This involves the controlled execution of bundled scripts.

jupyter-notebook