The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill retrieves and processes untrusted metadata (titles, abstracts) from the bioRxiv API. This content is eventually presented to the agent and could contain instructions designed to manipulate its behavior. Ingestion points: scripts/biorxiv_search.py fetches paper data via api.biorxiv.org. Boundary markers: The skill instructions do not provide specific delimiters or warnings to ignore instructions within the retrieved metadata. Capability inventory: The skill includes network access to bioRxiv and the ability to write PDF files to the local file system (download_pdf method). Sanitization: No sanitization or filtering of the retrieved abstract or title text is performed before returning it to the agent.
[EXTERNAL_DOWNLOADS]: The skill downloads PDF documents from the official and well-known biorxiv.org domain.
[COMMAND_EXECUTION]: The skill is implemented as a Python CLI tool that is executed by the agent to perform searches, but it does not utilize dangerous functions like eval() or exec() for dynamic code execution.

biorxiv-database