phi-prompt-guard
Pass
Audited by Gen Agent Trust Hub on Sep 23, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill operates on untrusted data such as patient records and clinical documentation, which represents an attack surface for indirect prompt injection.\n
- Ingestion points: External data enters the context via user-pasted clinical text, images containing patient charts for OCR processing, and potential outputs from database clients or file system reads.\n
- Boundary markers: The skill employs a specific attestation token
[PHI-OK]and instructs the agent to evaluate the surrounding context for signals of real operational data to validate the attestation.\n - Capability inventory: The agent is capable of generating and potentially executing commands for various database engines (e.g.,
psql,mysql,bq) and reading clinical file exports.\n - Sanitization: The skill implements a redaction and refusal workflow, requiring the agent to generate commands for local user execution rather than processing sensitive data directly.
Audit Metadata