The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The demonstration script (scripts/demo.py) specifies several standard Python dependencies including xgboost, scikit-learn, pandas, numpy, matplotlib, and marimo. These are well-established libraries for data science and reactive notebook development.
[INDIRECT_PROMPT_INJECTION]: The skill instructions and scripts involve processing external tabular data files (e.g., data/train.parquet). While this defines an ingestion point for untrusted data, the processing is performed using standard numerical libraries that do not interpret data as executable instructions.
Ingestion points: SKILL.md (reading parquet files via ibis).
Boundary markers: Not present, as is standard for tabular data processing.
Capability inventory: The skill performs local data processing, model training, and logging. It does not possess network exfiltration or arbitrary command execution capabilities.
Sanitization: Absent, which is expected for numerical feature engineering pipelines.

multilabel-classification