---

name: large-document-reader description: | Intelligently splits long academic or technical documents into chapters, generates structured JSON summaries for each, and creates a file system with a global index. This enables efficient AI retrieval and analysis, perfectly solving context window limitations by enabling “overview via summaries, deep-dive on demand” workflows. version: 1.0.0 author: Document Assistant category: research tags: [document-processing, knowledge-management, summarization, ai-optimization] metadata: {}

Literature Structuring Expert

Automatically decompose long documents (papers, reports, books) into a structured, AI-friendly knowledge base. Splits by chapter, generates machine-readable summaries, and builds a navigable index to overcome context limits.

When to Use This Skill

Use this skill when the user:

• Has a document that is too long for the AI's context window.
• Needs to perform cross-chapter analysis or get a high-level overview of a long text.
• Wants to build a reusable, queryable knowledge base from a PDF, Markdown, or text file.
• Asks: "How can I get my AI to read this whole book/paper?"

Quick Reference

Situation	Action
User provides a long document	1. Analyze and split it into chapters. 2. Generate a JSON summary for each chapter. 3. Create a master index file.
User asks a high-level, cross-chapter question	Provide the content of the MASTER_INDEX.md file to the AI.
User asks a detailed, chapter-specific question	Provide the corresponding single file from the ./chapters/ directory to the AI.
Task completed	Present the generated file tree and MASTER_INDEX.md preview to the user.

Core Workflow

Phase 1: Intelligent Splitting

1. Analyze Input: Receive the long document text or file path.
2. Identify Structure: Automatically analyze the document to identify heading hierarchies (e.g., #, ##, 1., 1.1) to determine chapter boundaries. Prioritize user-specified splitting preferences.
3. Execute Split: Split the document into independent plain-text files by chapter.
   • Naming Convention: {sequence_number}_{chapter_title}.md (e.g., 01_Introduction.md).
   • Storage Location: All chapter files are saved in the ./chapters/ directory.

Phase 2: Summary Generation & Structuring

1. Generate Summary per Chapter: For each file in ./chapters/, generate a corresponding JSON summary file.
   • Structured Fields (JSON format):

     {
       "chapter_id": "Unique identifier matching the filename, e.g., 02_1",
       "chapter_title": "Chapter Title",
       "abstract": "Core summary of the chapter, 200-300 words.",
       "keywords": ["Keyword1", "Keyword2", "Keyword3"],
       "key_points": ["Key point one", "Key point two"],
       "related_sections": ["IDs of other chapters strongly related to this one"]
     }
     

   • Storage Location: JSON summary files are saved in the ./summaries/ directory (e.g., 01_Introduction.summary.json).

Phase 3: Create Global Index

1. Aggregate Information: Collect data from all JSON files in ./summaries/.
2. Generate Index: Create a global index file, MASTER_INDEX.md.
   • Content: Lists all chapters' IDs, titles, a short abstract preview, and keywords in a Markdown list or table.
   • Purpose: Provides a "bird's-eye view" for quick navigation and high-level Q&A.

Final Deliverables & File Structure

Upon completion, the following file tree is generated:

Project_Root/
├── chapters/           # 【Source Repository】Contains all split chapter texts (.md files)
│   ├── 01_Introduction.md
│   ├── 02_1_Experimental_Methods.md
│   └── ...
├── summaries/          # 【Summary Repository】Contains all structured JSON summaries
│   ├── 01_Introduction.summary.json
│   ├── 02_1_Experimental_Methods.summary.json
│   └── ...
└── MASTER_INDEX.md     # 【Global Navigation】Core document summary index

Usage Instructions for the User

For Global, Cross-Chapter Queries (e.g., “What is the paper's main thesis?”):

• Provide the content of the MASTER_INDEX.md file to the AI. This is token-efficient.

For Specific, In-Depth Queries Within a Chapter (e.g., “What were the parameters in the 'Methods' section?”):

• Provide the corresponding single chapter file from the chapters/ directory to the AI for full context.