web-design-analyzer

Pass

Audited by Gen Agent Trust Hub on Jul 27, 2026

Risk Level: SAFEPROMPT_INJECTIONEXTERNAL_DOWNLOADS
Full Analysis
  • [PROMPT_INJECTION]: Indirect prompt injection vulnerability through image analysis.
  • Ingestion points: The script scripts/analyze_design.py ingests user-supplied image files through the --image parameter for processing by an LLM.
  • Boundary markers: Absent. The system prompt provided to the Vision API does not contain instructions to ignore or treat text found within the image as untrusted data.
  • Capability inventory: The skill is designed to generate "Coding Prompts" for use in external AI coding environments (e.g., Cursor, Claude Code) which typically have high-privilege access to local file systems and terminals.
  • Sanitization: Absent. The output from the Vision API is parsed and directly presented to the user without validation or escaping of potential instructions.
  • [EXTERNAL_DOWNLOADS]: Communicates with OpenAI's official API.
  • Evidence: The script scripts/analyze_design.py sends image data to https://api.openai.com/v1/chat/completions for processing. This is a well-known service and is necessary for the skill's primary function of visual analysis.
Audit Metadata
Risk Level
SAFE
Analyzed
Jul 27, 2026, 01:20 AM
Security Audit — agent-trust-hub — web-design-analyzer