prepare

Pass

Audited by Gen Agent Trust Hub on Jul 1, 2026

Risk Level: SAFE
Full Analysis
  • [DATA_EXPOSURE_AND_EXFILTRATION]: The skill accesses local health files such as data/allergies.json and data/health-feeling-logs.json to provide the user with a medical summary. This behavior is the primary function of the skill and does not involve network exfiltration or unauthorized external access.
  • [INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from health logs into the agent's prompt context without explicit boundary markers.
  • Ingestion points: Reads health data from data/health-feeling-logs.json and data/allergies.json to generate summaries.
  • Boundary markers: Absent. The data from the files is interpolated into the summary output format.
  • Capability inventory: The skill possesses Read and Write permissions for local file operations.
  • Sanitization: There is no evidence of sanitization or filtering for the content within the health records before it is processed by the agent.
Audit Metadata
Risk Level
SAFE
Analyzed
Jul 1, 2026, 05:26 AM
Security Audit — agent-trust-hub — prepare