tooluniverse-variant-interpretation
Pass
Audited by Gen Agent Trust Hub on Oct 12, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill's primary workflow involves ingesting and processing content from multiple external, third-party scientific databases and literature repositories, including ClinVar, PubMed, and preprint servers (BioRxiv/MedRxiv). These sources allow for community contributions or unvetted content, creating a potential surface for indirect prompt injection attacks.
- Ingestion points: Data is imported into the agent's context through tools such as
PubMed_search,BioRxiv_search_preprints,MedRxiv_search_preprints,clinvar_search, andmyvariant_query(which queries MyVariant.info). - Boundary markers: The skill lacks explicit instructions or systemic delimiters (such as XML tags or boundary markers) to prevent the agent from potentially following instructions that might be embedded within retrieved research abstracts or clinical reports.
- Capability inventory: The skill is primarily configured for data retrieval and interpretation. It utilizes specialized biological APIs and structural analysis tools (e.g., AlphaFold2) but does not display capabilities for arbitrary command execution or persistent file system modifications.
- Sanitization: There is no evidence of automated filtering, sanitization, or schema validation for the natural language content retrieved from external literature and clinical databases.
Audit Metadata