The Agent Skills Directory

[DYNAMIC_EXECUTION]: The skill's data caching strategy in references/data-access.md utilizes the pickle module. Deserializing data using pickle.load() is inherently unsafe because it allows for arbitrary code execution if an attacker replaces the local cache file with a malicious payload.- [INDIRECT_PROMPT_INJECTION]: The skill ingests external pharmaceutical data (XML format) from the DrugBank database as seen in references/drug-queries.md. It lacks explicit boundary markers or sanitization logic to prevent instructions potentially embedded in that data from overriding agent behavior during interpolation into prompts.- [COMMAND_EXECUTION]: The installation instructions in SKILL.md and references/data-access.md require the user to install multiple external Python packages, including drugbank-downloader and bioversions. While these are standard tools in the bioinformatics domain, they expand the agent's software supply chain risk.- [EXTERNAL_DOWNLOADS]: The skill is designed to download large datasets from go.drugbank.com. Although this is a well-known service for medical data, the process involves fetching and processing remote files at runtime which requires network and filesystem access.- [DATA_EXPOSURE]: The documentation in references/data-access.md provides guidance on managing DrugBank credentials via environment variables and configuration files (~/.config/drugbank.ini). This is noted as a safe practice for secret management compared to hardcoding.

drugbank-database