blog-notebooklm

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

• Third-party content exposure detected (high risk: 0.90). The skill explicitly opens and scrapes NotebookLM pages provided by URLs (see scripts/ask_question.py which navigates to the given notebook_url and extracts responses, and SKILL.md's "discover " / "ask " workflows), thereby ingesting user- or third-party-hosted notebook content and using those answers to drive follow-up questions and synthesis — exposing the agent to untrusted, user-generated content that can influence behavior.

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

• Potentially malicious external URL detected (high risk: 0.70). The skill opens and scrapes NotebookLM pages at runtime (e.g., https://notebooklm.google.com/notebook/...) to obtain user-uploaded content and injects that content into its responses, meaning an external URL is fetched during runtime and directly controls the agent's output.