The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill recommends installing several well-known and trusted machine learning libraries, including scikit-learn, matplotlib, seaborn, pandas, and numpy, as well as domain-specific packages like imbalanced-learn and category-encoders from PyPI.
[REMOTE_CODE_EXECUTION]: Reference documentation in references/model_evaluation.md and references/quick_reference.md includes examples for saving and loading models using pickle and joblib. These functions are standard in the scikit-learn ecosystem but can lead to arbitrary code execution if the agent is instructed to load a malicious model file from an untrusted source.
[PROMPT_INJECTION]: The skill exhibits an indirect prompt injection surface as it is designed to ingest and analyze external datasets.
Ingestion points: Data enters the system via CSV files and dataset loaders in SKILL.md and scripts/classification_pipeline.py.
Boundary markers: No specific delimiters or instructions are used to distinguish untrusted data from processing commands.
Capability inventory: The skill has capabilities for file system writes and object deserialization.
Sanitization: There is no evidence of input validation or sanitization performed on the content of the data being analyzed.

scikit-learn