The Agent Skills Directory

[Dynamic Execution] (MEDIUM): The skill suggests using the pickle module for caching parsed drug data in references/data-access.md.
Evidence: The implementation example uses pickle.load(f) to restore data from a local file.
Risk: pickle is inherently insecure; deserializing a maliciously crafted pickle file can result in arbitrary code execution on the host system.
[Indirect Prompt Injection] (LOW): The skill processes large volumes of external data from the DrugBank database (XML/JSON format).
Ingestion points: DrugBank XML database files and REST API responses (referenced in SKILL.md and references/data-access.md).
Boundary markers: None identified in the prompt snippets; the agent is instructed to parse and analyze the raw content.
Capability inventory: Includes chemical property calculation, network analysis, and data manipulation using libraries like rdkit and pandas.
Sanitization: The documentation does not specify sanitization or validation of the external content before processing.
[Data Exposure & Exfiltration] (LOW): The skill performs network operations to go.drugbank.com to download databases and query APIs.
Evidence: requests.get(url, headers=headers) and download_drugbank() usage in references/data-access.md.
Risk: While necessary for the skill's purpose, communication with non-whitelisted domains is flagged as a potential exfiltration vector.

drugbank-database