holmes
Pass
Audited by Gen Agent Trust Hub on Sep 19, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes user-supplied images which represents a potential surface for indirect prompt injection if the image contains adversarial text.
- Ingestion points: User-uploaded photos of desks, rooms, or individuals as described in Step 1.
- Boundary markers: The skill contains explicit instructions to ignore and not transcribe private information found in the frame, such as passwords, addresses, and document contents. It also instructs the agent to stop if consent for a photo of a person is not clear.
- Capability inventory: The skill lacks network access, file-writing capabilities, or shell execution tools, limiting the potential impact of an injection.
- Sanitization: Instructions explicitly command the agent to exclude sensitive items like pill bottles or religious objects from the analysis.
Audit Metadata