image-evaluation

Pass

Audited by Gen Agent Trust Hub on Mar 2, 2026

Risk Level: SAFEEXTERNAL_DOWNLOADSCOMMAND_EXECUTION
Full Analysis
  • [EXTERNAL_DOWNLOADS]: The evaluation script downloads pre-trained machine learning models (specifically CLIP and BLIP-2) from Hugging Face's official model hub using the standard transformers library. This is expected behavior for a tool performing automated image-text alignment evaluation.
  • [COMMAND_EXECUTION]: To generate reports, the skill requires executing a Python script (evaluate_image.py) and managing dependencies via pip. These operations are transparently documented and restricted to the local environment for the purpose of analyzing user-provided images.
  • [SAFE]: No signs of obfuscation, malicious prompt injection, or unauthorized data exfiltration were detected. The skill correctly implements HTML escaping for user-provided prompts in its reporting module to prevent cross-site scripting (XSS) in the generated HTML files.
Audit Metadata
Risk Level
SAFE
Analyzed
Mar 2, 2026, 07:29 AM