The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from external sources such as academic papers, patents, and news which could contain adversarial instructions targeting the agent context.
Ingestion points: External data is ingested through tools like ls_paper_fetch, ls_patent_fetch, ls_news_fetch, and ls_clinical_trial_fetch as defined in SKILL.md.
Boundary markers: None explicitly defined to delimit external content from instructions.
Capability inventory: The skill uses tools for clinical trial searching, drug database access, organization fetching, and academic literature retrieval.
Sanitization: Instructions include strict requirements for the agent to adhere to tool schemas and avoid inferring parameters, which provides a layer of validation.
[EXTERNAL_DOWNLOADS]: The setup instructions direct the user to connect to external Patsnap services using the Model Context Protocol (MCP).
Evidence: The setup guide instructs the execution of claude mcp add --transport http pharma_intelligence "https://connect.patsnap.com/096456/Logic-mcp?apiKey=...".
Context: The remote resources are hosted on the vendor's official domain (patsnap.com) and are necessary for the skill's primary functionality.

precision-oncology-zhcn