visual-verdict

Pass

Audited by Gen Agent Trust Hub on Mar 24, 2026

Risk Level: SAFEPROMPT_INJECTION
Full Analysis
  • [PROMPT_INJECTION]: The skill is designed to analyze and interpret visual data from user-provided images, which creates an attack surface for indirect prompt injection if those images contain adversarial text instructions.\n
  • Ingestion points: Visual data enters the agent context through the reference_images and generated_screenshot arguments defined in SKILL.md.\n
  • Boundary markers: The skill uses XML-style tags for organization but does not provide specific instructions to the model to disregard or isolate text discovered within the input images.\n
  • Capability inventory: The skill itself does not define shell commands, network operations, or file system modifications, though it mentions external pixel diff tools as debug aids.\n
  • Sanitization: No sanitization or filtering is applied to the content extracted from the images by the vision system before it is used to generate the final verdict.
Audit Metadata
Risk Level
SAFE
Analyzed
Mar 24, 2026, 12:45 AM