The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill instructs the user to configure connections to external Model Context Protocol (MCP) servers hosted at connect.patsnap.com. These are official service endpoints for the skill's author (Patsnap) and provide the necessary biological and patent data for its research functions.
[COMMAND_EXECUTION]: Provides a setup guide with claude mcp add commands. This is the standard, documented procedure for extending the agent's capabilities with specialized data services and requires explicit user action to execute.
[CREDENTIALS_UNSAFE]: Use of placeholder values (e.g., sk-xxxxxxxxxxxx and YOUR_API_KEY) is present in the documentation. This is a safe and standard practice for explaining how users should provide their own credentials during the setup process.
[INDIRECT_PROMPT_INJECTION]: The skill possesses a vulnerability surface common to research tools, as it ingests large volumes of external data from patent records, academic literature, and web search results. However, it implements a robust 'Search → Fetch' pattern that requires the agent to verify multiple IDs and retrieve full records, which significantly mitigates the effectiveness of basic injection attempts hidden in summaries or titles.

biomarker-investigation