tooluniverse-variant-to-mechanism
Pass
Audited by Gen Agent Trust Hub on Oct 7, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONEXTERNAL_DOWNLOADS
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill ingests user-provided genetic identifiers and traits which are then used to query external biological databases, creating a standard indirect prompt injection surface through remote data ingestion.
- Ingestion points: rsIDs, genomic coordinates, and trait descriptions provided by the user.
- Boundary markers: None defined for delimiting data returned from scientific APIs.
- Capability inventory: The skill uses network-enabled tools to fetch and process results from Ensembl, GTEx, and OpenTargets.
- Sanitization: The instructions do not specify sanitization or validation of the content retrieved from external biological repositories.
- [EXTERNAL_DOWNLOADS]: The skill initiates network requests to retrieve genomic data and functional annotations from external scientific providers.
- Evidence: Code snippets demonstrate fetching data from
rest.ensembl.org,gtexportal.org, andftp.ncbi.nlm.nih.gov. These domains are recognized as well-known and trusted academic and governmental services for scientific research.
Audit Metadata