The Agent Skills Directory

[DYNAMIC_EXECUTION]: The example code in SKILL.md utilizes joblib.load('demand_forecast_model.pkl') to load a machine learning model. joblib uses Python's pickle module internally for serialization. Deserializing data from untrusted .pkl files can lead to arbitrary code execution if the file has been tampered with. While this is a standard practice in data science, it remains a security risk if model sources are not verified.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process untrusted external data, posing a vulnerability surface.
Ingestion points: Data enters the system via the orders list in RouteOptimizationAdvisor.generate_route_recommendations and suppliers_data in SupplierSelectionAdvisor.evaluate_suppliers within SKILL.md.
Boundary markers: The instructions do not define delimiters or specific warnings to ignore instructions embedded within the processed data.
Capability inventory: The skill includes logic for decision-making and result display (print statements), but does not currently implement direct file-write or network operations based on the input data.
Sanitization: There is no evidence of validation, escaping, or filtering of the external data fields (e.g., supplier_name, order_id) before they are processed or used in output generation.

prescriptive-analytics