human-like

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

• Third-party content exposure detected (high risk: 1.00). The skill's SKILL.md explicitly instructs "If the user shares a URL, fetch it and analyze the actual content," so the agent will fetch and interpret arbitrary third-party URLs provided by users, which can contain untrusted instructions.

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

• Potentially malicious external URL detected (high risk: 1.00). The skill explicitly says "If the user shares a URL, fetch it and analyze the actual content," meaning arbitrary user-provided URLs (e.g., raw GitHub or pastebin links) are fetched at runtime and injected into the model context to control prompts and analysis.