Notebooklm

NotebookLM CLI Wrapper (Python)

Required parameters

• python3 available.
• notebooklm-py installed (CLI binary: notebooklm).
• NotebookLM authenticated (login).

Quick start

• Wrapper script: scripts/notebooklm.py.
• Command form: python3 {baseDir}/scripts/notebooklm.py [args...].

python3 {baseDir}/scripts/notebooklm.py login
python3 {baseDir}/scripts/notebooklm.py list
python3 {baseDir}/scripts/notebooklm.py use 
python3 {baseDir}/scripts/notebooklm.py status
python3 {baseDir}/scripts/notebooklm.py ask "Summarize the key takeaways" --notebook 

Output guidance

• Prefer --json for machine-readable output where supported.
• Long-running waits are handled by native commands like:
• source wait
• artifact wait
• research wait

⚡ Sub-Agent Delegation (Anti-Blocking)

Problem

NotebookLM operations like source wait, artifact wait, research wait, generate slide-deck, and source add-research can take minutes to complete. Running them in the main session blocks the conversation.

Strategy

For any operation expected to take >30 seconds, delegate to a sub-agent via sessions_spawn:

1. Main session: Acknowledge the user's request, then spawn a sub-agent with a clear task description.
2. Sub-agent: Executes the long-running NotebookLM commands, waits for completion, and reports back.
3. Main session: Remains responsive. The sub-agent auto-announces completion.

Which operations to delegate

| Operation | Delegate? | Reason | |-----------|-----------|--------| | login, status, list, use, clear | ❌ No | Fast ( Commands to run (in order):

...

Use the CLI wrapper: python3 ~/.openclaw/skills/notebooklm-Invoke/scripts/notebooklm.py Prefer --json output where supported. If any step fails, report the error and stop. When complete, summarize what was accomplished and any output files created. mode: run label: notebooklm-

### Example: Generate slide deck

**User**: "帮我用 notebook X 生成一个 PPT"

**Main session response**:
> 好的，我派了一个后台任务去生成 PPT，完成后会通知你 ✧

**Spawn**:

sessions_spawn: task: | NotebookLM task: Generate a slide deck from notebook.

Steps:

1. python3 ~/.openclaw/skills/notebooklm-Invoke/scripts/notebooklm.py generate slide-deck "Create a comprehensive slide deck" --notebook
2. python3 ~/.openclaw/skills/notebooklm-Invoke/scripts/notebooklm.py artifact wait --notebook --timeout 600 --json
3. python3 ~/.openclaw/skills/notebooklm-Invoke/scripts/notebooklm.py download slide-deck ./output.pptx --notebook --latest --format pptx

Report: artifact details, file path, any errors. mode: run label: notebooklm-slide-deck

### Example: Add research source

**User**: "在 notebook Y 里加一个关于碳足迹的深度研究"

**Spawn**:

sessions_spawn: task: | NotebookLM task: Add deep research source.

Steps:

1. python3 ~/.openclaw/skills/notebooklm-Invoke/scripts/notebooklm.py source add-research "碳足迹最新研究进展" --mode deep --notebook
2. python3 ~/.openclaw/skills/notebooklm-Invoke/scripts/notebooklm.py research wait --notebook --timeout 600
3. python3 ~/.openclaw/skills/notebooklm-Invoke/scripts/notebooklm.py source list --notebook --json

Report: research status, new sources added, any errors. mode: run label: notebooklm-research

### Guidelines

- **Always tell the user** you're delegating to a background task before spawning.
- **Use `mode: run`** (one-shot) — no need for persistent sessions.
- **Use descriptive labels** like `notebooklm-slide-deck`, `notebooklm-research-carbon` for easy tracking.
- **Include all context in the task** — the sub-agent has no conversation history.
- **Error handling**: Instruct the sub-agent to report errors clearly so you can relay them.
- **File paths**: Use absolute paths for output files so the main session can find them.
- **Compound workflows**: Bundle related steps (add → wait → generate → wait → download) into a single sub-agent task rather than spawning multiple.

## PPT generation policy
- A single generated slide deck should target **at most 15 pages**.
- If user requirements exceed 15 pages, split into multiple decks (e.g., Part 1/2/3) and generate separately.
- After generation, provide downloadable **`.pptx`** output when possible:
  - `download slide-deck ... --format pptx`

## References
- `README.md` (installation, requirements, troubleshooting)
- `QUICKSTART_CN.md`（中文快速上手）
- `references/cli-commands.md`

## Assets
- None.

## Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

- **Author:** [tiangong-ai](https://github.com/tiangong-ai)
- **Source:** [tiangong-ai/agent-skills](https://github.com/tiangong-ai/agent-skills)
- **License:** MIT

Install and usage instructions live in the source repository linked above.