The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted metadata from the bioRxiv API, including paper titles and abstracts, which represents an indirect prompt injection attack surface.
Ingestion points: Data enters the agent's context through the search_by_date_range, search_by_keywords, and get_paper_details methods in scripts/biorxiv_search.py, which fetch content from api.biorxiv.org.
Boundary markers: The skill does not employ delimiters or explicit instructions to the agent to distinguish the retrieved preprint content from its own instructions.
Capability inventory: The skill has capabilities to write search results to local JSON files and download PDF documents to the local file system.
Sanitization: The script performs keyword filtering but does not implement sanitization or escaping of the retrieved text content before presenting it to the agent.
[EXTERNAL_DOWNLOADS]: Fetches preprint metadata and downloads full-text PDF files from the official bioRxiv infrastructure.
Evidence: The script scripts/biorxiv_search.py makes network requests to api.biorxiv.org for metadata and www.biorxiv.org for PDF downloads, which are the established official domains for the bioRxiv service.

alterlab-biorxiv