image-ocr

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

• Third-party content exposure detected (high risk: 1.00). The skill explicitly downloads and OCRs arbitrary user-supplied images (message.download_image() in integration_example.py and the "Auto Image-to-Text" OpenClaw integration in SKILL.md), then injects the extracted text into prompts sent to a model, so untrusted third-party content can directly influence agent behavior.