notebooklm

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

• Third-party content exposure detected (high risk: 0.75). Outsider free text can enter the LLM context via ask_question.py returning answer scraped from NotebookLM’s UI (page.query_selector_all(...).inner_text()), which is derived from the user-provided Notebook URL/notebook contents (outsider-authored document text). That scraped text is then printed and becomes the agent’s LLM-visible context.

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

• Potentially malicious external URL detected (high risk: 0.90). The skill programmatically opens and queries NotebookLM at runtime (e.g., https://notebooklm.google.com/ or https://notebooklm.google.com/notebook/...), reads the returned answers, and injects those responses into the agent output, so the external NotebookLM URL directly controls prompts/responses and is a required runtime dependency.