specialist

Pass

Audited by Gen Agent Trust Hub on Jul 1, 2026

Risk Level: SAFE
Full Analysis
  • [SAFE]: The skill demonstrates safe operational patterns for medical data analysis. It explicitly defines 'Safety Redlines' to prevent the model from providing drug dosages, prescriptions, or clinical diagnoses, which is a security best practice for medical-domain agents.
  • [PROMPT_INJECTION]: The skill handles untrusted data as its primary function, which introduces a surface for indirect prompt injection. However, the risk is limited by the skill's narrow scope and safety instructions.
  • Ingestion points: Medical records and chronic disease tracker files are read from the data/ directory and interpolated into subagent prompts (SKILL.md, Step 4).
  • Boundary markers: The subagent template uses basic placeholders (e.g., {{读取相关的检查报告数据}}) but does not implement strict boundary delimiters or 'ignore' instructions for the interpolated data.
  • Capability inventory: The skill is limited to Read and Write operations for medical data management.
  • Sanitization: There is no explicit sanitization or filtering of the patient-provided medical data before it is processed by the specialist subagent.
Audit Metadata
Risk Level
SAFE
Analyzed
Jul 1, 2026, 05:26 AM
Security Audit — agent-trust-hub — specialist