tooluniverse
Pass
Audited by Gen Agent Trust Hub on Sep 19, 2026
Risk Level: SAFEEXTERNAL_DOWNLOADSINDIRECT_PROMPT_INJECTION
Full Analysis
- [EXTERNAL_DOWNLOADS]: The skill provides instructions to install the tooluniverse Python package. The referenced resources, including the GitHub repository (mims-harvard/ToolUniverse) and documentation (hms.harvard.edu), are hosted on legitimate academic and research domains associated with Harvard Medical School.
- [INDIRECT_PROMPT_INJECTION]: The skill is designed to process data from over 600 external scientific databases and tools, which presents a surface for potential indirect prompt injection from those data sources.
- Ingestion points: Data enters the agent's context through the tu.run() method, which queries external scientific APIs and databases like OpenTargets, PubChem, and UniProt.
- Boundary markers: The instructions do not define explicit delimiters or markers to isolate external data from the agent's core instructions.
- Capability inventory: The skill includes network access to fetch research data and uses the ToolUniverse framework for dynamic scientific tool execution.
- Sanitization: There is no description of automated sanitization, filtering, or validation of the content retrieved from the external scientific databases.
Audit Metadata