The Agent Skills Directory

[PROMPT_INJECTION]: The skill is subject to indirect prompt injection risks as it ingests and processes large volumes of data from external, public biomedical sources. Attackers could potentially plant malicious instructions within scientific abstracts or trial descriptions to influence the agent's subsequent reasoning or tool use. Ingestion points: Data is fetched from 14+ external biomedical databases via a unified API. Boundary markers: The instructions lack explicit delimitation or warnings to treat external tool outputs as potentially untrusted data. Capability inventory: The skill provides functionality for network operations using curl to interface with the research tools. Sanitization: There is no documented validation or sanitization of the content retrieved from external databases.
[DATA_EXFILTRATION]: The skill utilizes a unified endpoint hosted on a third-party domain (mcp.cloud.curiloo.com) to proxy requests to various biomedical databases. This architecture results in research queries being processed by an external service.
[EXTERNAL_DOWNLOADS]: The skill recommends the installation of a Python package (medical-mcps) from a non-standard source for local execution. While the package is associated with the stated author, its security posture has not been verified.

medical-research-toolkit