biorxiv-database
Pass
Audited by Gen Agent Trust Hub on Sep 21, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [SAFE]: The skill performs its stated research functions using a well-known academic service. Network operations are confined to official bioRxiv domains (api.biorxiv.org and www.biorxiv.org), and the scripts use standard Python libraries without dangerous dynamic execution or obfuscation.
- [INDIRECT_PROMPT_INJECTION]: The skill retrieves preprint titles and abstracts from an external API and presents them to the agent. While bioRxiv is a trusted academic source, the ingestion of this external content represents a theoretical surface for indirect prompt injection. 1. Ingestion points: Preprint metadata retrieved via the BioRxiv API in scripts/biorxiv_search.py. 2. Boundary markers: No specific delimiters are used to wrap the retrieved text. 3. Capability inventory: The skill has access to the run_terminal and write_file tools. 4. Sanitization: The retrieved academic text is not sanitized before being returned to the agent.
Audit Metadata