clinical-decision-support
Pass
Audited by Gen Agent Trust Hub on Sep 15, 2026
Risk Level: SAFE
Full Analysis
- [SAFE]: The skill operates entirely locally, utilizing only the Python standard library. Its instructions and script implementations (scripts/_common.py) explicitly prohibit network access, the use of credentials, and external API calls, maintaining a closed execution environment.
- [SAFE]: The scripts implement a 'Data Gate' mechanism in scripts/_common.py that recursively scans input JSON for a comprehensive list of sensitive keys (e.g., social_security_number, medical_record_number, address) and rejects the input if identifiers are detected. Additionally, the skill enforces strict local path resolution, rejecting non-local URLs and symlinks to prevent unauthorized file access.
- [EXTERNAL_DOWNLOADS]: The skill provides instructions for the agent to retrieve updated citation metadata from arXiv (a well-known academic repository) if the skill contributes to a research publication. This reference follows the trusted scope rule and is consistent with the intended research-only use case.
- [INDIRECT_PROMPT_INJECTION]: The skill provides a surface for processing user-supplied JSON data through its validation scripts. It mitigates potential prompt injection risks by using structured schema validation, enforcing strict local-only read/write operations, and providing the agent with explicit refusal instructions for any requests involving patient-specific care or clinical diagnosis.
Audit Metadata