The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests data from external scientific sources such as PubMed, UniProt, and NCBI Gene to perform research tasks.
Ingestion points: Data is retrieved through various tool calls like PubMed_search_articles, UniProt_get_entry_by_accession, and OpenTargets APIs (detailed in REFERENCE.md).
Boundary markers: The documentation does not specify explicit boundary markers for these external data streams.
Capability inventory: The SDK supports batch execution of tools, file writing via the FileSaveHook, and automated summarization of large tool outputs.
Sanitization: The skill relies on standard API handling and environment-based configuration for safety.
[EXTERNAL_DOWNLOADS]: The skill downloads model weights and specialized configurations required for embedding-based search and machine learning tools.
Source details: Resources are fetched from Harvard Medical School's infrastructure (zitniklab.hms.harvard.edu) and the project's official GitHub repository (github.com/mims-harvard/ToolUniverse).
Scale: Downloads can range from 200MB for the base package up to 5GB for full machine learning features.
[COMMAND_EXECUTION]: Provides instructions for standard software lifecycle management and server deployment.
Patterns detected: Includes standard pip installation commands and a tooluniverse-http-api CLI for deploying the SDK as a remote service.

tooluniverse-sdk