/cs-notebooklm

disable-model-invocation: true
name: "cs-notebooklm"
description: "/cs-notebooklm — NotebookLM browser automation. Action-routing intake (Q1: read / add source / Studio output / create new) + per-action Q2-Q4 branching. Fire-and-notify for slow Studio ops. Mandatory custom prompts (defaults are mediocre). Requires browser automation environment — fails clean on web."

/cs-notebooklm — NotebookLM Browser Automation

**Command:** `/cs-notebooklm`

The `cs-notebooklm` persona controls Google NotebookLM via browser automation across 4 core actions.

Critical Prerequisite

**Requires browser automation environment.** Works in:

• Claude Code CLI with computer-use
• Claude Chrome Extension
• Playwright / Puppeteer with screenshot + click tools

Does NOT work in:

• Claude.ai web (no browser automation) — skill exits cleanly at Step 0

When to Run

• Want to ask your existing NotebookLM notebook a question (Action 1)
• Want to add a source (URL / text / file / Google Doc / YouTube) to a notebook (Action 2)
• Want to generate a Studio output (Audio Overview / Infographic / Slides / Study Guide / etc.) (Action 3)
• Want to create a new notebook from scratch (Action 4)

Action-Routing Intake (2-4 Forcing Questions)

| Q | Asks | Notes | |---|---|---| | Q1 | Action: read / add source / Studio output / create new | Forcing — refuses to start without commitment | | Q2 | Notebook name or URL (actions 1-3) OR title for new notebook (action 4) | Drives navigation | | Q3 | Action-specific parameter (question text / source type / Studio output type / initial sources) | Branches per Q1 | | Q4 | Studio custom prompt detail | Asked only if Q1=3 (Studio); mandatory |

Most invocations stop at Q3. Q4 only fires for Studio generation.

What You Get

Per action:

| Action | Result | |---|---| | Read/Extract | Clean response from notebook chat (not raw dump) | | Add Sources | Confirmation of ingestion (with screenshot) | | Studio Output | Confirmation that generation started + "NotebookLM will notify you when ready" — fire-and-notify | | Create New | New notebook URL + confirmation of initial sources added |

Studio Output Types

All 9 types supported:

• Audio Overview (5-10 min generation — fire-and-notify)
• Study Guide
• Briefing Doc
• Timeline
• FAQ
• Table of Contents
• Infographic
• Slides (slide deck)
• Mind Map

Mandatory Custom Prompts

Default Studio prompts produce mediocre output. The skill ALWAYS opens the customization menu and writes a detailed custom prompt before submitting.

Examples per output type:

| Output | Example custom prompt | |---|---| | Audio Overview | "Two-host conversation for a non-technical executive, 8-10 min, focus on business implications not technical depth" | | Infographic | "Decision-tree style, action-oriented, 6 panels max, monochrome navy" | | Study Guide | "Undergrad-level, definitions + 3 practice questions per concept" | | Slides | "12 slides max, 1-2 sentences per slide, presenter notes with examples per slide" |

Discipline

• **Step 0 environment check** — verify browser automation; fail fast if not
• **Screenshot-first** — every UI action preceded by screenshot
• **find()-before-click** — semantic finders over pixel coordinates
• **Never auto-handle login** — detect login wall, stop, tell user to log in manually
• **Studio custom prompts always** — open customization menu, write detailed prompt
• **Fire-and-notify for slow ops** — Studio generation doesn't block this session
• **Tool-agnostic language** — "browser automation tool", not "Claude Chrome Extension"

Trigger Phrases (auto-invoke without /cs:)

• "open NotebookLM"
• "check my [notebook name] notebook"
• "pull info from NotebookLM"
• "ask my notebook about X"
• "add [source] to NotebookLM"
• "create an infographic in NotebookLM"
• "use NotebookLM Studio"
• "generate a slide deck from my notebook"
• "what does my notebook say about X"
• Any variation involving NotebookLM

Workflow

# Step 0: environment check (silent if available; halt if not)

# Phase 0 intake (Q1 + Q2 minimum; Q3-Q4 branch per action)
python ${CLAUDE_PLUGIN_ROOT}/skills/notebooklm/scripts/action_router.py \
  --action read_extract --notebook "Q3 prep" --question "what are the latest trends?"

# Studio output flow includes custom prompt generation:
python ${CLAUDE_PLUGIN_ROOT}/skills/notebooklm/scripts/custom_prompt_template_generator.py \
  --output-type infographic --audience executive --length compact

# Async classification (for "should I wait or fire-and-notify?")
python ${CLAUDE_PLUGIN_ROOT}/skills/notebooklm/scripts/async_action_classifier.py --action audio_overview
# Returns: FIRE_AND_NOTIFY (5-10 min generation)

# Execute action via browser automation (screenshot → find → click → verify)
# Return clean summary

Stop Conditions

• Browser automation unavailable → halt at Step 0 with clear message
• Q1 action commitment refused → halt, re-ask
• Login wall detected → halt, ask user to log in manually
• Page layout changed unexpectedly → screenshot, ask user for guidance
• 3 consecutive UI find() failures → halt, alert user

Anti-Patterns Rejected

• Tool-specific names without abstraction (e.g., hardcoding "Claude Chrome Extension")
• Synchronous waiting on Studio generations (especially Audio Overview)
• Skipping screenshots between actions
• Using pixel coordinates when semantic find() is available
• Attempting to handle login flows automatically
• Generating Studio outputs without opening customization menu
• Using default Studio prompts (always write custom)

Related

• Agent: [`cs-notebooklm`](../agents/cs-notebooklm.md)
• Skill: [`notebooklm`](../skills/notebooklm/SKILL.md)
• Source spec: `megaprompts/03-notebooklm-megaprompt.md` (maintainer-local draft spec — gitignored, not in the public repo)
• Research-domain siblings (different shape): `/cs-pulse`, `/cs-litreview`, `/cs-grants`, `/cs-dossier`, `/cs-patent`, `/cs-syllabus`

---

**Version:** 1.0.0 **Source:** Pat