notebooklm-prep

name	notebooklm-prep
description	Prepare, format, and package vault research notes into NotebookLM-ready source bundles with automatic Audio Overview / Podcast transcript priming and source manifests.
user-invocable	true
argument-hint	[target-folder] [-o output-manifest.md]
allowed-tools	Bash, ReadFile, WriteFile
type	concept
title	SKILL
sources	["[[epistemic-contract]]"]
tags	["concept","okf"]

notebooklm-prep — NotebookLM Ingestion & Audio Overview Priming

Prepare, sanitize, and structure Obsidian vault notes into clean source packages optimized for Google NotebookLM batch upload, citation mapping, and Audio Overview (Podcast format) synthesis.

Workflow & Constraints

1. Clean Header & Frontmatter Injection: Every source file is verified to contain clean YAML frontmatter (title, type: concept, tags: [notebooklm-ingest]) without raw control characters or HTML noise.

2. Chunking & Size Check: NotebookLM accepts up to 500,000 words per source. Large transcripts are split cleanly into section-level source chunks (_part1.md, _part2.md).

3. Manifest Generation: Generates a master upload manifest (00-NOTEBOOKLM-INGEST-MANIFEST.md) linking all formatted source files, byte counts, and key research questions for the NotebookLM query box.

4. Audio Overview (Podcast) Prompt Generator: Generates a dedicated prompt file (NOTEBOOKLM-AUDIO-PROMPT.txt) tailored to elicit deep technical Audio Overviews focusing on:

   • Reasoning vs Fabrication ($R1\text{--}R10$) crossover thresholds ($\sim 54%$).
   • Model-vs-Model behavior differences (Opus 4.6 vs Gemini 3.1 Pro vs GPT-5.4).