tooluniverse-spatial-omics-analysis
Pass
Audited by Gen Agent Trust Hub on Oct 12, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process external user data across 9 distinct analysis phases, creating a significant attack surface for indirect prompt injection.
- Ingestion points: User-supplied parameters including
svgs,tissue_type, anddisease_contextare ingested in Phase 0 and propagated throughout the workflow. - Boundary markers: The instructions lack explicit guidance on using boundary markers or specific "ignore embedded instructions" warnings when passing user data to external tool parameters.
- Capability inventory: The skill utilizes over 70 ToolUniverse tools for extensive database queries (e.g., STRING, Reactome, HPA, OpenTargets) and literature searches (PubMed, OpenAlex). These tools are used in sequence, where the output of one query may influence the logic of subsequent steps.
- Sanitization: There is no mention of input sanitization or validation for user-provided strings before they are interpolated into tool queries like
PubMed_search_articlesorOpenTargets_get_disease_id_description_by_name.
Audit Metadata