markitdown

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

• Third-party content exposure detected (high risk: 0.90). The skill accepts arbitrary http(s) URLs (see SKILL.md "YouTube URLs are detected..." and examples) and the wrapper script treats inputs starting with http/https as URLs and passes them to markitdown (scripts/markitdown.sh), meaning untrusted public web content (YouTube/public pages) is fetched, converted, and returned as Markdown for the agent to read and act on.

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

• Potentially malicious external URL detected (high risk: 0.80). The skill accepts arbitrary https:// URLs at runtime (for example: https://youtu.be/dQw4w9WgXcQ) and passes them to the markitdown CLI which fetches and converts the remote content and streams the resulting Markdown back into the agent context, enabling high-confidence remote prompt injection.