Computer Vision Helper

Fail

Audited by Gen Agent Trust Hub on Feb 16, 2026

Risk Level: HIGH
Full Analysis
  • General Analysis (SAFE): The skill contains legitimate documentation and code templates for computer vision workflows using standard libraries like PyTorch and OpenAI. The logic follows established industry practices for model inference and training.
  • External Downloads (SAFE): The code snippets reference well-known, trusted Python packages (torch, clip, openai, ultralytics, segment-anything). These are standard tools in the AI research community and are considered trusted sources.
  • Indirect Prompt Injection (LOW): While the skill introduces templates for processing external images and text (Category 8), it does not create an autonomous vulnerability. The risk is inherent to vision-language models and is presented within the context of educational developer templates.
  • Automated Scan Alert (INFO): The scanner alert for 'box.co' is a false positive. The skill text contains legitimate computer vision terms such as 'bounding box' and 'box prompt' but does not contain the flagged URL or any malicious external links.
Recommendations
  • Contains 1 malicious URL(s) - DO NOT USE
Audit Metadata
Risk Level
HIGH
Analyzed
Feb 16, 2026, 09:13 AM