design-md

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

• Third-party content exposure detected (high risk: 0.75). The required workflow fetches outsider-authored free text/HTML at runtime via web_fetch/read_url_content from htmlCode.downloadUrl (and potentially screenshot-derived descriptions), which the LLM then parses and uses for synthesis.

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

• Potentially malicious external URL detected (high risk: 0.90). The skill explicitly instructs the agent at runtime to download and parse external content from the screen object's htmlCode.downloadUrl (and optionally screenshot.downloadUrl) via web_fetch/read_url_content, which would inject remote content into the agent's context and directly influence generated prompts/output.