The Agent Skills Directory

[DYNAMIC_EXECUTION]: The skill uses joblib.load to load trained models and scalers from the local file system in the 'Model Serving' section.
Evidence: model = joblib.load('models/best_model.pkl') and scaler = joblib.load('models/scaler.pkl') in SKILL.md.
Risk: joblib uses pickle internally. Loading untrusted or modified .pkl files can result in arbitrary code execution on the host machine.
[INDIRECT_PROMPT_INJECTION]: The skill ingests data from external files (.csv, .parquet) which may contain untrusted content that could influence the agent's behavior.
Ingestion points: DataLoader.load_data(self.data_path) reads external files based on a provided path.
Boundary markers: The skill does not implement boundary markers or instructions to ignore embedded prompts in the ingested data.
Capability inventory: The skill has the capability to perform model training, log data to MLflow, and start a FastAPI web server.
Sanitization: No sanitization or validation of the input data content is performed beyond standard type checking during feature engineering.

machine-learning