media-tools

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

• Third-party content exposure detected (high risk: 0.80). The skill sends images/prompts to external provider APIs and consumes their textual responses as part of its workflow (see scripts/read_image.py which builds params and calls the vision provider chain, and lib/providers/{openrouter,gemini,anthropic,dashscope}_vision.py which POST to third‑party endpoints and return provider-generated text), so untrusted third‑party outputs can directly influence the agent's next actions.