pleaseprompto/notebooklm-skill

Overview

This skill lets Claude Code query your Google NotebookLM notebooks and return source-grounded, citation-backed answers generated by Gemini. It automates browser interactions, manages a local library of notebooks, and maintains persistent authentication so queries come only from your uploaded documents. The workflow minimizes hallucinations by restricting responses to your knowledge base.

How this skill works

Each query opens a fresh browser session, asks NotebookLM for an answer that cites sources from your notebooks, and closes the session. A run.py wrapper handles virtual environment creation, dependency installation, and proper script execution. The skill manages notebook metadata, supports adding/searching/activating notebooks, and enforces a follow-up loop to ensure completeness before final user-facing synthesis.

When to use it

• You want citation-backed answers from documents you uploaded to NotebookLM.
• You mention NotebookLM or provide a NotebookLM URL.
• You need to add, list, or activate notebooks in the local library.
• You need a reproducible, document-only response to avoid hallucinations.
• You want to run automated queries while keeping authentication persistent.

Best practices

• Always invoke scripts through the run.py wrapper (python scripts/run.py [script]).
• Check authentication first (auth_manager.py status) and run setup if needed.
• Use Smart Add: query a notebook to discover its content before adding metadata.
• Provide a full description and topics when adding a notebook; never guess.
• Follow the built-in follow-up loop: verify answers and ask follow-ups until complete.

Example use cases

• Ask: "Summarize the security policies in my internal NotebookLM notebook" and get answers with document citations.
• Add a shared research notebook by first querying its contents, then adding it with name/description/topics.
• Search the library for a topic keyword and then query the active notebook for specifics.
• Run a complex multi-step query and let the follow-up loop iterate until all gaps are filled.