tooluniverse-precision-medicine-stratification
Pass
Audited by Gen Agent Trust Hub on Apr 1, 2026
Risk Level: SAFEEXTERNAL_DOWNLOADS
Full Analysis
- [EXTERNAL_DOWNLOADS]: The skill interacts with approximately 13 external scientific databases and tools, including OpenTargets, ClinVar, Ensembl, PharmGKB, and PubMed. These are all established, reputable sources for medical and genomic data. These interactions are documented in 'TOOLS_REFERENCE.md' and 'test_skill.py'.
- [PROMPT_INJECTION]: The skill possesses an indirect prompt injection surface due to the ingestion of unstructured text from external sources like PubMed and OpenTargets. This is noted as a risk factor; however, the skill significantly mitigates this by requiring a rigid output format as defined in 'REPORT_TEMPLATE.md' and 'SCORING_REFERENCE.md'.
- Ingestion points: External API responses (PubMed, OpenTargets, CIViC) as seen in Phase 7 and Phase 8 of 'SKILL.md'.
- Boundary markers: Present in 'REPORT_TEMPLATE.md' which specifies a strict markdown structure for the final report.
- Capability inventory: The skill utilizes data retrieval and processing tools; no dangerous shell execution or dynamic evaluation of external content was found.
- Sanitization: The skill uses predefined scoring matrices in 'SCORING_REFERENCE.md' to quantify risk, which acts as a validation layer for the findings.
- [DATA_EXFILTRATION]: No unauthorized data exfiltration patterns were detected. While the skill transmits clinical and genomic parameters to external analysis services, this behavior is documented and strictly aligned with the primary purpose of providing patient risk stratification.
- [SAFE]: No hardcoded credentials, malicious persistence mechanisms, or privilege escalation patterns were found. Environment variable placeholders in '.env.template' follow security best practices. The provided 'test_skill.py' suite demonstrates legitimate testing of tool integrations.
Audit Metadata