NotebookLM Audio Generation (Manual Fallback)
Status: Manual fallback - Use when NotebookLM Enterprise API is unavailable.
When to Use This Skill
Use this skill when:
- NotebookLM Enterprise API is unavailable (no paid subscription)
- API automation fails and fallback is needed
- User explicitly requests manual workflow
Step 1: Generate the Prompt
CRITICAL: Always use the script. Never fabricate or modify the prompt.
bash1cd ~/src/research/podcast/tools 2python notebooklm_prompt.py ../episodes/EPISODE_PATH/ --copy
The script:
- Auto-detects episode title and series name from content_plan.md
- Verifies all 5 required files exist
- Outputs the correct prompt with proper branding
- Copies to clipboard with
--copyflag (macOS)
Required files (5 total):
episode-directory/
├── research/p1-brief.md # Research brief
├── research/p3-briefing.md # Master briefing
├── report.md # Narrative synthesis
├── sources.md # Validated sources
└── content_plan.md # Episode structure guide
Step 2: Show User the Script Output
Run the script and display its complete output to the user. The output includes:
- Episode and series info (auto-detected)
- File checklist with status (✓ or ✗ MISSING)
- The ready-to-paste prompt
- Settings reminder
- NotebookLM link
Example output:
============================================================
NOTEBOOKLM MANUAL AUDIO GENERATION
============================================================
Episode: Strategic Selection
Series: Algorithms for Life
Directory: ../episodes/algorithms-for-life/ep2-strategic-selection
📁 Files to Upload (5/5 ready):
✓ p1-brief.md
✓ report.md
✓ p3-briefing.md
✓ sources.md
✓ content_plan.md
============================================================
📋 NOTEBOOKLM PROMPT (copy-paste ready):
============================================================
Create a two-host podcast episode on: Strategic Selection from our Algorithms for Life series
...
============================================================
⚙️ Settings: Format: Deep Dive | Length: Long
🔗 Open: https://notebooklm.google.com/
✓ Prompt copied to clipboard!
Step 3: User Completes Manual Workflow
Guide user through these steps:
- Go to https://notebooklm.google.com/
- Create new notebook
- Upload all 5 source files (shown in the checklist)
- Click "Audio Overview" → "Customize"
- Paste the prompt (already on clipboard from
--copy) - Settings: Deep Dive format, Long length
- Generate and download audio (~10-15 minutes)
- Save audio file to episode directory
Step 4: Process Audio
After download, use the podcast-audio-processing skill:
- Convert to mp3 if needed
- Transcribe with local Whisper
- Generate chapter markers
- Embed chapters into mp3
Prompt Template Reference
The prompt is defined in podcast/tools/notebooklm_prompt.py (single source of truth).
Key elements:
- References content_plan.md for structure, hooks, key terms
- Brand intro: "Welcome to Yuda Me Research from our [Series] series by Valor Engels..."
- Brand outro: "research dot yuda dot me - that's Y-U-D-A dot M-E"
- Style: Define terms, cite specifics, distinguish correlation/causation
- Avoids: Undefined jargon, fabricated examples, over-hedging
DO NOT:
- Duplicate the template elsewhere
- Manually substitute placeholders
- Add episode-specific content arcs (content_plan.md handles this)
Troubleshooting
| Issue | Solution |
|---|---|
| Script shows missing files | Complete earlier phases first |
| Can't auto-detect title/series | Use --title and --series flags |
| Clipboard copy fails | Manually copy from terminal output |
| Audio too short | Check all 5 files uploaded, use Long setting |
