The Agent Skills Directory

[DYNAMIC_EXECUTION]: The skill provides instructions and code examples for model persistence using libraries that perform unsafe deserialization.
Evidence: In references/model_evaluation.md, the skill demonstrates how to load models using joblib.load() and pickle.load() from local files.
Impact: These functions are susceptible to arbitrary code execution if the model files originate from an untrusted source or are tampered with.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process external data, creating a surface for indirect prompt injection attacks.
Ingestion points: Data is ingested via pd.read_csv('data.csv') in references/common_workflows.md and through various text feature extraction tools in references/preprocessing.md.
Capability inventory: The skill has the ability to write files (plot images) and execute Python scripts via the Bash tool.
Sanitization: No explicit logic is provided to sanitize or filter potential instructions embedded within the processed datasets.
Boundary markers: There are no explicit instructions or delimiters used to ensure the agent ignores non-data content (like instructions) within the input files.
[EXTERNAL_DOWNLOADS]: The skill recommends installing several third-party Python packages for extended functionality.
Evidence: Documentation suggests installing imbalanced-learn, category-encoders, and umap-learn via uv pip.
Context: While these are well-known packages in the machine learning ecosystem, they represent an expanded dependency surface beyond the core scikit-learn library.

scikit-learn