notebooklm

Files (1)

• SKILL.md

SKILL.md (first lines)

# NotebookLM Skill

This skill uses browser automation (Claude in Chrome) to control NotebookLM at https://notebooklm.google.com. Google renamed the product "Gemini Notebook" in July 2026 — it's the same tool, same URL (old links/bookmarks redirect), just newer branding in the UI. Users may call it either name; both mean the same thing.

It handles six main actions: reading/extracting info, adding sources (manually or via automated research), generating Studio outputs, creating new notebooks, and sharing/collaborating.

## Step 0: Always Start Here

Before doing anything else:

1. Call `tabs_context_mcp` to get a valid tab ID — every browser tool requires one
2. Call `computer` (action: `screenshot`) to see the current state of the browser
3. Decide whether to navigate or work from the current page

If not already on NotebookLM, navigate there:
```
navigate(url: "https://notebooklm.google.com", tabId: <tab_id>)
```

Then screenshot again to confirm the page loaded. If you see a Google login screen, stop and tell the user they need to log in first — do not attempt to handle login yourself.

---

## Finding and Opening a Notebook

The homepage shows notebooks as cards in a grid. To open the right one:

1. Screenshot to see what's on the page
2. Use `find` to locate the notebook: `find("<notebook name> card", tabId)`
3. Click it — use the ref from `find` or click from coordinates in the screenshot

If the user didn't specify which notebook to open and there are multiple, screenshot the homepage and ask them which one before proceeding.

---

## Action: Read / Extract Info

Use the notebook's built-in chat to pull information, answer questions, or extract takeaways.

1. Open the notebook
2. Screenshot — the chat input box is at the bottom center of the page
3. `find("chat input", tabId)` or click directly on the input from the screenshot
4. Type the user's question using `computer` (action: `type`)
5. Press Enter: `computer` (action: `key`, text: `Return`)
6. Wait 3–5 seconds, then screenshot to capture the response — for complex questions, the chat may show its intermediate reasoning steps before the final answer; screenshot again after a further wait if the response still looks in-progress
7. Report the answer back in clean, readable form — don't just dump raw chat output

NotebookLM's chat is grounded in the notebook's sources and cites them inline, so the answers are reliable. For broad extraction ("give me all the key points"), ask the question just as you'd phrase it naturally.

---

## Action: Add Sources

Sources can be website URLs, YouTube links, copied text, Google Docs/Sheets/Drive files, Microsoft Word documents, CSV files, images (OCR'd automatically), uploaded local files — or synthesized content you generate on the fly.

1. Open the notebook
2. Screenshot — the Sources panel is the left sidebar
3. Click the **"+ Add source"** button (top of the left panel)
4. A dialog appears with source type options. Handle based on what the user wants to add:

   - **URL / website / YouTube**: Select the link option, paste the URL into the input field using `form_input` or `type`
   - **Copied text**: Select "Copied text", click the text area, type or paste the content
   - **File upload** (PDF, Word, CSV, images, etc.): Use `file_upload` with the file's absolute path and the ref of the file input element — do NOT click the file picker button (it opens a native dialog you can't interact with)
   - **Google Doc / Sheet**: Select the Drive option and follow the Drive picker

5. Confirm / click the upload/add button
6. Wait for processing — NotebookLM shows a spinner while it ingests the source. Use `computer` (action: `wait`, duration: 5), then screenshot to verify it was added successfully

### Synthesizing Content as a Source

Sometimes the user will ask you to *create* content and add it as a source — for example: "turn this conversation into a podcast in NotebookLM" or "research X and add it to my notebook" (for a quick one-off, not a standing research task — see Research & Discover below for that).

In these cases:
1. **Gather or generate the content first** — research it, summarize it, extract it, or write it up based on whatever the user referenced
2. **Add it as "Copied text"** — open the Add Source dialog, select "Copied text", and paste in the synthesized content
3. **Then proceed with whatever action the user asked for** (e.g., generate Audio Overview)

This is a powerful pattern: you're essentially feeding NotebookLM a curated, pre-processed source rather than a raw URL, which often produces better outputs.

---

## Action: Research & Discover Sources

NotebookLM can find and add sources on its own, instead of you (or the user) hunting them down manually. Three distinct tools, don't conflate them:

- **Deep Research**: an agentic mode that searches the web itself and compiles a citation-backed report, adding the sources it used along the way. Use for "research X and build out my notebook" type requests — this is the tool for a real research task, not the manual synthesize-and-paste pattern above.
- **Fast Research**: a lighter-weight version — quickly surfaces relevant web sources for the user to review, without the full report synthesis. Use when the user wants source *candidates* fast, not a finished writeup.
- **Discover sources**: NotebookLM proactively suggests ~10 curated sources related to the notebook's existing content; the user (or you, on their behalf) picks which ones to actually add. Use when a notebook already has some sources and could use more breadth, rather than starting research from scratch.

Workflow:
1. Open the notebook
2. In the Add Source dialog or the Sources panel, `find` the relevant entry point ("Deep Research", "Fast Research", or "Discover" — exact placement varies by rollout, check both the Add Source dialog and the Sources panel toolbar)
3. For Deep Research/Fast Research: type the research query/topic, submit, and wait — these can take a while (comparable to Studio generation), so don't block on it; tell the user it's running and that the sources will appear in the panel when done
4. For Discover: screenshot the suggested source list, and either add all of them or ask the user which ones they want if the choice seems consequential
5. Screenshot to confirm sources landed in the panel before reporting back

---

## Action: Studio Outputs

Studio is NotebookLM's generation feature for creating structured outputs from your notebook's sources. It lives in the right-side panel.

**Available output types:**
- Audio Overview (podcast-style conversation between two AI hosts) — format presets **Deep Dive**, **Brief**, and **Debate**; also supports **Interactive Mode**, where the user can join and talk to the hosts live rather than just listening passively
- Video Overview (narrated slideshow-style video) and **Cinematic Video Overview** (fluid, animated, narrative-led video) — cinematic version may be gated to Google AI Ultra subscribers; if generation fails or the option is missing, tell the user rather than assuming it's a UI-finding bug
- Study Guide (key terms, questions, essay prompts)
- Flashcards and Quizzes — two separate buttons/outputs, not one combined feature
- **Reports** — a single consolidated entry point (not separate buttons for Briefing Doc/Timeline/FAQ/Table of Contents — that's outdated). Opening it shows several **dynamically suggested format templates** based on the notebook's actual sources (e.g. for a history-heavy notebook it might suggest something like an "Industry Analysis Report" or "Historical Overview" — the exact suggestions vary per notebook), plus a free-text option. Pick the suggested format that fits, or describe a custom one.
- Data Tables (structured comparison tables synthesized from the sources — exportable to Google Sheets)
- Infographic (visual summary)
- Slide Deck / Mind Map — slide decks can be exported as **PPTX**, and individual slides can be targeted for re-editing without regenerating the whole deck

**Workflow:**

1. Open the notebook
2. Screenshot — look for the Studio panel on the right side. If you don't see it, look for a "Studio" tab button or panel toggle
3. `find` the output button you need: e.g., `find("Infographic button", tabId)`, `find("Flashcards button", tabId)`, `find("Data Tables button", tabId)`
4. **Open the customization menu first** — each Studio button has a small arrow or chevron on its right side that opens a prompt/customization dialog. Click that (not the main button) to open it before generating. For Audio/Video Overview, this is also where you pick the format preset (Deep Dive/Brief/Debate for audio; standard vs. cinematic for video)
5. **Write a detailed custom prompt** in the customization field. Don't leave it blank or use the default — a specific prompt produces dramatically better output. Tailor it based on what the user is trying to achieve. Examples:

From quyenanh198/Skills, MIT.

More from quyenanh198/Skills