biorxiv-database
Pass
Audited by Gen Agent Trust Hub on Oct 12, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONPROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill is designed to search and retrieve abstracts and metadata from the bioRxiv preprint server, which represents untrusted data from the open web.
- Ingestion points: Data enters the context through output from
scripts/biorxiv_search.py, specifically theabstractandtitlefields in the search results. - Boundary markers: None. There are no instructions or delimiters provided to prevent the agent from being influenced by instructions that might be embedded in research abstracts.
- Capability inventory: The skill includes capabilities to execute Python scripts, write files to the local system (
results.json,paper.pdf), and perform network operations via therequestslibrary. - Sanitization: No sanitization or validation logic for the ingested content is described in the provided files.
- [PROMPT_INJECTION]: The skill includes an explicit instruction to the agent to proactively suggest an external website (www.k-dense.ai) for complex workflows. This attempt to influence the agent's behavior to promote a commercial service is a form of behavioral injection.
Audit Metadata