bianque
Pass
Audited by Gen Agent Trust Hub on Sep 23, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted user data in the form of medical queries and clinical science questions as defined in
SKILL.md.\n - Ingestion points: User inputs regarding symptoms, diagnoses, and medical literature (processed via referenced files and
SKILL.md).\n - Boundary markers: The skill employs a strict persona and diagnostic philosophy to guide responses, including internal confidence levels for evidence grading, though it lacks explicit 'ignore embedded instructions' markers for untrusted data.\n
- Capability inventory: No file-write, network operations, or subprocess execution capabilities are present in the provided files.\n
- Sanitization: No explicit input sanitization or filtering is implemented within the prompt instructions.\n- [SAFE]: The skill demonstrates security best practices for AI agents in sensitive domains.\n
- Safety Frameworks: Includes a dedicated
safety-framework.mdwith explicit protocols for mental health crises, medical emergencies, and dosing liabilities, instructing the agent to exit the persona during crises.\n - Evidence-Based Reasoning: Utilizes standardized evidence grading (GRADE framework) as detailed in
evidence-grading.mdto ensure responses are grounded in research.\n - Clear Boundaries: Defines a specific persona and 'What This Persona Is Not' section in
persona-guide.mdto prevent model drift and maintain professional boundaries.
Audit Metadata