tooluniverse-target-research
Pass
Audited by Gen Agent Trust Hub on Oct 7, 2026
Risk Level: SAFE
Full Analysis
- [SAFE]: No security issues were identified during the analysis of the skill's instructions or implementation code.
- [INDIRECT_PROMPT_INJECTION]: The skill ingests data from external scientific sources like PubMed and Open Targets. It mitigates potential indirect prompt injection risks by requiring the agent to perform systematic evidence grading (T1-T4) and mandatory source attribution, which encourages the agent to treat external content as data rather than instructions.
- [DATA_EXFILTRATION]: Network operations are restricted to well-known scientific databases (e.g., UniProt, ChEMBL, PDB, GTEx). These communications are fundamental to the skill's primary research function and involve no unauthorized access to sensitive user data.
- [DYNAMIC_EXECUTION]: The skill employs dynamic tool invocation to manage retries and fallback chains for bioinformatics tools. This logic is confined to the agent's approved tool environment and is used to ensure high data completeness in research reports.
Audit Metadata