notebookLM

NotebookLM Skill

These are skills responsible for document generation.

Feature Details

Feature	Details
NotebookLM YAML	See references/notebooklm-yaml.md
Slide YAML	See references/notebooklm-slides.md

Execution Steps

1. Classify the user's request
2. Read the appropriate reference file from the above "Feature Details"
3. Generate according to its contents

---

🔧 PDF Page Range Reading (Claude Code 2.1.49+)

This feature is for efficiently handling large PDFs.

Read by specifying a page range

// ページ範囲指定で読み取り
Read({ file_path: "docs/spec.pdf", pages: "1-10" })

// 目次だけ確認
Read({ file_path: "docs/manual.pdf", pages: "1-3" })

// 特定のセクションのみ
Read({ file_path: "docs/api-reference.pdf", pages: "25-45" })

Recommended Approaches by Use Case

Case	Recommended Reading Method	Reason
PDFs over 100 pages	Table of contents (1-3) → only relevant chapters	Minimize token consumption
Specification review	Specify ranges by section	Read only the necessary parts in detail
API Documentation	Start from the endpoint list (table of contents)	Understand the overall structure before going into details
Academic papers	Abstract + conclusion → main text	Grasp the key points first
Technical manuals	Table of contents + troubleshooting chapter	Prioritize practical parts

Examples of Use When Generating NotebookLM YAML

大型PDF（300ページの技術仕様書）からYAMLを生成する場合：

1. **目次を読む**（1-5ページ）
   Read({ file_path: "spec.pdf", pages: "1-5" })
   → 章立てを把握

2. **各章の冒頭を読む**（各章の最初の2ページ）
   Read({ file_path: "spec.pdf", pages: "10-11" })  // 第1章
   Read({ file_path: "spec.pdf", pages: "45-46" })  // 第2章
   → 各章の概要を把握

3. **重要セクションを精読**
   Read({ file_path: "spec.pdf", pages: "78-95" })  // APIリファレンス
   → 詳細な内容を抽出

この方法で、300ページすべてを読むことなく効率的にYAMLを生成できます。

Best Practices

Principle	Description
Progressive reading	Read in the order: table of contents → overview → details
Only relevant pages	Specify only the pages needed for the task
Token saving	Reading all pages is a last resort
Structure understanding first	Grasp the overall picture from the table of contents before going into details

Comparison with Conventional Methods

Method	Token consumption	Processing time	Accuracy
Read all pages(300 pages)	~150,000	Long	High
Page range specification(necessary 30 pages)	~15,000	Short	High

→ 90% token reduction and shorter processing time are possible