SuperAPI AI Directory

deep-doc-reader

Agents & models

View repository

Updated 10-16 00:04

About this skill

Long-Document Deep Reading and Q&A

Structured understanding and precise Q&A for long PDFs (annual reports, contracts, papers, manuals). Build documents into a hierarchical tree index via PageIndex MCP to achieve chapter-level localization and reasoning-based retrieval.

Use Cases

  • User says "Help me analyze the Q3 revenue data in this annual report"
  • User says "Where is the breach clause in this contract? What exactly does it say?"
  • User says "What does the experimental methods section of this 80-page paper say?"
  • User says "Help me find the section about installation steps in this manual"
  • User needs to repeatedly ask about different content in the same long document

When to Use This Skill (Instead of Other Tools)

判断逻辑:

Document-related requests ├── Where is the file? Find a certain file → use local-search ├── Simple summary/overview → use summarize ├── Edit PDF content → use PDF reading-type Skill ├── Create Word/Excel → use word-processor / excel-analyzer └── Long-document deep Q&A (the following scenarios) → use deep-doc-reader ✅ ├── PDF exceeds 20 pages ├── Need to precisely locate a specific chapter/page number ├── Need to repeatedly query different parts of the same document └── Need to understand the document's hierarchical structure


## 工作流程

### 第一步:上传文档建立索引

用户提供 PDF 文件后,通过 PageIndex MCP 工具上传并建立树索引:

User provides a PDF path or URL ↓ Call PageIndex MCP to upload the document ↓ PageIndex automatically builds a hierarchical tree index (table of contents → chapters → subsections) ↓ Return a document ID; subsequent queries use this ID


**注意**:
- 首次上传需要等待索引构建(大文档约 1-3 分钟)
- 索引构建完成后,后续查询都是即时的
- 免费额度:1000 页

### 第二步:基于树索引进行问答

User asks: "What is the Q3 revenue data?" ↓ PageIndex navigates via tree-structure reasoning: Document root node → Financial data chapter → Q3 quarterly report → Revenue data ↓ Return precise content + page number citation ↓ Organize the answer, attaching the source page number


### 第三步:多轮追问

同一文档支持多轮追问,无需重新上传:

User: "What is Q3 revenue?" → Answer + page number User: "How does it compare with Q2?" → Automatically navigate to the Q2 section User: "How does management explain this change?" → Locate the MD&A chapter


## 输出格式

### 标准回答格式

```markdown
## 回答

[具体回答内容]

### 来源

- 📄 第 87 页 - 第三季度财务摘要
- 📄 第 92-93 页 - 营收明细表

Structure Overview Format (when the user requests to understand the document structure)

## 文档结构

📑 XXX 年度报告(共 156 页)
├── 第一章:公司概况(p.1-15)
├── 第二章:经营情况讨论(p.16-45)
│   ├── 2.1 行业形势(p.16-22)
│   ├── 2.2 经营成果(p.23-38)
│   └── 2.3 现金流分析(p.39-45)
├── 第三章:财务报表(p.46-120)
│   ├── 3.1 资产负债表(p.46-52)
│   ├── 3.2 利润表(p.53-60)
│   └── ...
└── 附录(p.121-156)

Output Specifications

  • Answers must include page number citations (so the user can verify)
  • For data-type answers, directly quote the original numbers; do not make calculations
  • Divide long answers into paragraphs, labeling the source chapter for each paragraph
  • If the relevant content cannot be found in the document, clearly inform the user