The Agent Skills Directory

[Dynamic Execution] (MEDIUM): File references/streamlit-advanced.md contains a code example using pickle.load(open("model.pkl", "rb")). Deserializing data using pickle is insecure and can lead to arbitrary code execution if the input file is malicious.
[Unverifiable Dependencies] (MEDIUM): Multiple files including SKILL.md, references/sharing-publishing.md, and references/notebook-testing.md recommend installing packages such as nbstripout, pre-commit, nbval, and voila using pip install without specifying versions or verifying hashes, which presents a supply chain risk.
[Indirect Prompt Injection] (LOW): The skill is designed to ingest and process external data (e.g., CSV, Parquet, and text in references/large-dataset-eda.md and references/text-features.md). While the impact is primarily limited to local analysis context, the lack of sanitization or boundary markers for untrusted data poses a minor risk if the agent makes decisions based on the content of processed data.
[Data Exposure] (SAFE): SKILL.md contains explicit security guidance and code examples (e.g., using os.environ.get) instructing users never to hardcode secrets like API keys.

data-science-notebooks