blog-notebooklm

Blog NotebookLM is a Claude Code skill that queries Google NotebookLM notebooks to return source-grounded, citation-backed answers from a user's own uploaded documents. It solves the problem of pulling verifiable, document-grounded research into a blog-writing or research workflow without Claude hallucinating: each question opens a headless browser session against NotebookLM, retrieves Gemini's source-grounded answer, and closes. The skill is careful to frame answers as source-grounded model output, not proof of truth, and instructs recording stable underlying source URLs and dates for citations rather than the private notebook URL.

Because NotebookLM has no public API, the skill drives a real browser. It manages a local notebook library (add, list, search, activate, remove) with metadata, handles one-time interactive Google authentication in a visible browser, and persists auth via a hybrid persistent-browser-profile plus cookie-injection approach that works around a Playwright bug. Chrome is installed automatically on first run via Patchright, and a run.py wrapper owns venv creation and dependency installation. Optional environment variables control browser visibility and enable stealth, human-like typing-speed simulation. It works standalone via /blog notebooklm or internally from blog-write and blog-researcher, falling back silently when not authenticated so it never blocks the writing workflow.

It targets bloggers and researchers who have curated document collections in NotebookLM and want to cite them reliably. Note the ~50 queries/day free-account rate limit and that the browser-automation, stealth-behavior, and cookie-injection mechanics carry some dual-use and account-terms considerations, since they emulate human interaction to keep automated sessions working against a service with no official API.