ecommerce-multimodal-recognize-image

Pass

Audited by Gen Agent Trust Hub on Sep 11, 2026

Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill is designed to analyze external images and user-provided requirements, which introduces a surface where malicious instructions embedded in images (e.g., via visual text) could attempt to influence agent behavior.
  • Ingestion points: The imageUrl and requirement parameters in SKILL.md and scripts/multimodal_recognize_image.py accept untrusted external data.
  • Boundary markers: The skill does not implement explicit delimiters or instructions to ignore instructions found within analyzed visual content.
  • Capability inventory: The skill utilizes urllib.request for network operations and os for managing local session data and caching.
  • Sanitization: Input requirements and the resulting multimodal analysis are processed without sanitization or strict schema validation.
  • [COMMAND_EXECUTION]: The skill executes Python scripts to perform its core image recognition and asset management tasks.
  • Evidence: scripts/multimodal_recognize_image.py and scripts/upload_image.py are used to communicate with the Nexscope API, handle local file reads for uploads, and manage a local cache directory in the project or home folder.
Audit Metadata
Risk Level
SAFE
Analyzed
Sep 11, 2026, 07:19 AM
Security Audit — agent-trust-hub — ecommerce-multimodal-recognize-image