The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes and executes Jupyter notebooks (.ipynb), which are external data files. Maliciously crafted notebooks could contain instructions aimed at influencing the agent's behavior.
Ingestion points: Data enters the agent's context through commands like nb read, nb search, and nb execute in SKILL.md and its references.
Boundary markers: The nb tool utilizes structured sentinel lines (e.g., @@cell, @@output) to delimit notebook components in its output, facilitating parsing but not fully preventing the agent from following natural language instructions found within cell source code.
Capability inventory: The skill possesses capabilities for file modification (nb cell update), network interaction (nb connect), and code execution (nb execute).
Sanitization: The skill encourages using the AI-Optimized Markdown format to clearly separate metadata from source content, though it does not explicitly filter the content of markdown or code cells.
[COMMAND_EXECUTION]: The skill facilitates the execution of the nb CLI, which is designed to run arbitrary code within Jupyter notebooks. This is the primary purpose of the skill, but it grants the agent the ability to execute code on the local environment or remote kernels. The skill includes a 'Trust Boundary' section in SKILL.md instructing the agent to verify the binary path and version before initial use to mitigate the risk of interacting with a malicious executable.
[EXTERNAL_DOWNLOADS]: The skill provides instructions in references/best-practices.md and SKILL.md for installing external Python dependencies using %pip install or uv pip install within notebook cells. This is a standard Jupyter workflow that involves downloading and installing third-party packages at runtime.

notebook-cli