The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill fetches preprint metadata and PDFs from bioRxiv's official API and website.\n
Evidence: Requests are made to api.biorxiv.org and www.biorxiv.org in scripts/biorxiv_search.py.\n
Note: These are well-known services for the life sciences community.\n- [PROMPT_INJECTION]: The skill processes untrusted text from external sources, making it susceptible to indirect prompt injection.\n
Ingestion points: Preprint titles and abstracts are retrieved via the BioRxivSearcher class in scripts/biorxiv_search.py.\n
Boundary markers: The skill does not use delimiters or warnings to isolate retrieved content from the agent's instructions.\n
Capability inventory: The skill can write search results and PDFs to the file system using the --output and --download-pdf flags across multiple Python methods.\n
Sanitization: No validation or sanitization of the retrieved text is performed before it is output to the user or agent context.\n- [COMMAND_EXECUTION]: The Python script writes data to file paths provided as command-line arguments without validation.\n
Evidence: In scripts/biorxiv_search.py, args.output and args.download_pdf are passed directly to the open() function for writing.\n
Impact: This could allow an agent to be manipulated into overwriting files if sensitive paths (e.g., shell profiles or configuration files) are provided by the model or user.

biorxiv-database