The Agent Skills Directory

[DYNAMIC_EXECUTION]: The skill provides patterns in notebook-organization.md for manual cell execution using exec() and syntax checking using compile() during the validation and debugging of notebooks. This capability allows the execution of arbitrary code extracted from .ipynb files, posing a risk if the notebook content originates from an untrusted source.
[COMMAND_EXECUTION]: Utility patterns in notebook-patterns.md utilize subprocess.run(shell=True) to execute shell commands for fetching genomic data from AWS S3 buckets. Although these are common patterns for projects like GenomeArk, the use of shell execution with variables derived from data files requires strict input validation to prevent potential command injection.
[INDIRECT_PROMPT_INJECTION]: The skill establishes a vulnerability surface by instructing the agent to ingest and process data from external CSV files and Jupyter Notebooks.
Ingestion points: pd.read_csv in notebook-patterns.md and json.load in notebook-organization.md.
Boundary markers: Not explicitly implemented in the provided code snippets.
Capability inventory: The skill utilizes the Bash and NotebookEdit tools, along with exec() and subprocess.run() calls for data management and validation.
Sanitization: No explicit sanitization or filtering logic is provided for external content before it is processed or used in analysis workflows.

jupyter-notebook-analysis