web-design-analyzer

MEDIUM W011: Third-party content exposure detected (indirect prompt injection risk).

• Third-party content exposure detected (high risk: 0.75). Outsider free text enters the agent’s LLM context via the user-provided uploaded webpage screenshot (an untrusted image) that scripts/analyze_design.py base64-encodes and sends to the OpenAI Vision prompt, where the model’s extracted content (including any embedded/typed text) is returned and then parsed for JSON/prompt output.