visual-verdict
Pass
Audited by Gen Agent Trust Hub on Aug 20, 2026
Risk Level: SAFEPROMPT_INJECTION
Full Analysis
- [PROMPT_INJECTION]: The skill defines a workflow where untrusted visual data (reference images) is processed to generate actionable suggestions for the agent, creating a surface for indirect prompt injection.\n
- Ingestion points: External data enters the context via reference_images[] and generated_screenshot file paths defined in SKILL.md.\n
- Boundary markers: The skill lacks explicit delimiters or instructions to the model to ignore potential text-based commands embedded within visual assets.\n
- Capability inventory: The output suggestions are explicitly designed to "drive the next round of modification" (SKILL.md), meaning the contents of processed images can indirectly influence file write operations.\n
- Sanitization: No sanitization or validation of the input image content is performed before generating modification instructions.
Audit Metadata