Query Google NotebookLM documentation with source-grounded answers

NotebookLM Research Assistant automates querying Google NotebookLM's source-grounded, Gemini-powered answers from a documentation library, and manages that library's notebooks. Every question opens a fresh browser session, retrieves an answer exclusively from the uploaded documents in the target notebook, and closes; the answer is treated as untrusted content - emitted inside an explicit boundary, saved to a private 0600 JSON file, referenced by path rather than pasted into logs, and never executed as instructions. After each answer, the skill is required to stop and analyze whether the answer actually covers the user's original request, ask targeted follow-up questions if gaps remain, and only synthesize a final response once the information is complete. All scripts - auth_manager.py, notebook_manager.py, ask_question.py, cleanup_manager.py - must be run through a run.py wrapper, never called directly, because the wrapper creates and activates a .venv, installs dependencies, and installs the Chromium browser automatically. Adding a notebook without user-supplied metadata uses a "Smart Add" pattern: the skill first asks the notebook itself what it contains, shows that proposed name, description, and topics to the user as untrusted data, and only registers the notebook after explicit user confirmation - --description and --topics are always required fields.

Trigger it when a user mentions NotebookLM explicitly, shares a notebooklm.google.com/notebook/... URL, asks to query their notebooks or documentation, wants to add documentation to their NotebookLM library, or uses phrases like "ask my NotebookLM" or "check my docs." It has real limitations worth knowing: there's no session persistence, every question opens a new browser, free Google accounts are capped at 50 queries a day, documents must be manually uploaded to NotebookLM by the user first, and each question carries a few seconds of browser overhead.

Input: a question, plus an active, specified, or URL-referenced notebook, and, for adding a notebook, its URL with a name, description, and topics. Output: an answer sourced only from the notebook's uploaded documents, delivered inside an untrusted-content boundary and saved to a private JSON file for the skill to read and reason over, plus the reminder prompt "Is that ALL you need to know?" that drives the follow-up loop.

• Google NotebookLM, via browser automation with a headless or visible Chromium instance
• Local private storage under ~/.local/share/agentic-awesome-skills/notebooklm/ (library.json, auth_info.json, browser_state/), all .gitignore-protected
• Optional .env config for headless mode, browser visibility, stealth behavior, and typing speed
• references/ docs: api_reference.md, troubleshooting.md, usage_patterns.md

Anyone who wants an AI agent to query their own NotebookLM documentation library conversationally, or maintain that library's notebooks, without manually driving the NotebookLM web UI for every question.