The Agent Skills Directory

[COMMAND_EXECUTION]: The skill instructs the agent to use shell commands for discovering local development environments, including checks for conda environments (conda env list), Jupyter kernels (jupyter kernelspec list), and local virtual environments (ls -d .venv/). These operations are transparent and necessary for the stated purpose of notebook execution.
[DYNAMIC_EXECUTION]: The workflow involves executing Python code within Jupyter notebooks using jupyter execute. This is the primary function of the skill and is mitigated by instructions to ask the user for environment selection and to comment out potentially dangerous or sensitive setup cells before execution.
[EXTERNAL_DOWNLOADS]: The skill uses uvx to run jupyter-switch, a utility that may be fetched from a package registry at runtime. This is a standard method for executing specialized CLI tools in modern Python workflows.
[INDIRECT_PROMPT_INJECTION]: The skill processes user-supplied notebook and markdown content which could theoretically contain malicious instructions. The skill mitigates this surface by providing reference documents that define strict prompt templates using boundary markers (<context>, <question>) and by advising the agent to sanitize notebooks by commenting out dependency installations and credential assignments before automated runs.

jupyter-notebook-writing