markitdown

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

• Third-party content exposure detected (medium risk: 0.30). In scripts/batch_convert.py, the workflow directly ingests file-derived result.text_content (from outsider-authored files placed in an input directory) into the model during conversion, without any prior selection/allowlisting of specific content.

MEDIUM W012: Unverifiable external dependency detected (runtime URL that controls agent).

• Potentially malicious external URL detected (high risk: 0.90). The scripts call the OpenRouter API at runtime (base URL "https://openrouter.ai/api/v1"), and the responses (e.g., the review critique returned) are used to programmatically generate/improve subsequent prompts, so external content directly controls agent prompts.

MEDIUM W013: Attempt to modify system services in skill instructions.

• Attempt to modify system services in skill instructions detected (high risk: 1.00). The skill documentation explicitly instructs running a sudo command ("sudo apt-get install tesseract-ocr"), which asks for elevated privileges and would modify the host system state.