tooluniverse-target-research
Pass
Audited by Gen Agent Trust Hub on Apr 16, 2026
Risk Level: SAFEPROMPT_INJECTION
Full Analysis
- [SAFE]: The skill utilizes well-known biological and medical information services, including UniProt, Ensembl, Open Targets, PubMed, and Pharos, for its data gathering operations. These sources are recognized as established and trusted scientific entities.
- [PROMPT_INJECTION]: The skill's workflow creates a surface for indirect prompt injection by processing substantial amounts of data from external biological databases and scientific literature.
- Ingestion points: Untrusted data enters the context through multiple API tool calls (e.g., in
SKILL.mdviaPubMed_search_articlesandOpenTargetsendpoints). - Boundary markers: The instructions lack specific requirements for using delimiters or clear boundary markers to isolate external data from system instructions.
- Capability inventory: The agent is authorized to perform file-system operations (creating and updating reports) and execute further tool calls based on the ingested data.
- Sanitization: There are no explicit validation or sanitization steps defined for the content retrieved from external sources.
- [PROMPT_INJECTION]: The skill explicitly instructs the agent to suppress tool outputs and the research process from the user, requiring that only the final report be shown. This concealment practice reduces user oversight and could allow malicious instructions embedded in external data to be executed or influence the report content without detection.
Audit Metadata