deep-reading

Deep Reading - Deep Reading Collaboration System

Deep Reading Collaboration System, through multi-layered AI agents collaboration, helps you take an article from "read" to "understood" and then to "thoroughly comprehended", ultimately transforming it into an actionable plan.

Use Cases

Use this skill when the user needs to:

• Deeply understand a complex article, paper, or long-form text
• Systematically organize reading notes
• Think critically about the content of an article
• Discover hidden logical issues and assumptions
• Transform knowledge into an action plan

Core Features

The system includes a three-layer collaboration architecture:

🔍 Internalization Layer (Reading Thin) - Sequential Execution

• Organizer: Generates structured Markdown notes
• Paraphraser: Uses SCQA + Feynman Technique to explain in plain language

🎯 Expansion Layer (Reading Thick) - Parallel Execution

• Diagnostic Engine: Critical analysis + logical dissection (background)
• Socrates: Guiding questions, does not give answers (foreground)

🚀 Output Layer (Action)

• Planner: Transforms cognition into a concrete action plan

Response Pattern

When the user requests deep reading analysis:

1. Receive input:

   - 检查是否提供了文章路径和草稿笔记路径
   - 如果没有,询问用户或请求粘贴内容
   - 解析可选参数 (--internalize-only, --expand-only, --no-action)
   

2. Confirm configuration:

   - 确认要启用的层次 (默认全部启用)
   - 确认输出目录 (默认 outputs/[timestamp]/)
   - 显示即将执行的流程
   

3. Execute processing workflow:

   步骤 3.1: 内化层 (如果启用)
   - 调用 Task tool, subagent_type: "general-purpose"
     - Agent 1: 读取 references/agents/organizer.md
     - 输入: 原文 + 草稿
     - 输出: organized-notes.md
   
   - 调用 Task tool, subagent_type: "general-purpose"
     - Agent 2: 读取 references/agents/explainer.md
     - 输入: organized-notes.md
     - 输出: explained-notes.md
   
   步骤 3.2: 拓展层 (如果启用, 并行执行)
   - 并行调用两个 Task tools:
     - Agent 3: 读取 references/agents/diagnosis.md
       输入: organized-notes.md + explained-notes.md
       输出: diagnosis-report.json
   
     - Agent 4: 读取 references/agents/socratic.md
       输入: diagnosis-report.json
       输出: socratic-questions.md
   
   步骤 3.3: 产出层 (如果启用)
   - 调用 Task tool, subagent_type: "general-purpose"
     - Agent 5: 读取 references/agents/planner.md
     - 输入: 所有前序分析结果
     - 输出: action-plan.md
   

4. Integrate output:

   - 创建时间戳目录 (如 outputs/2026-01-21-14-20/)
   - 保存所有生成的文件
   - 生成汇总报告 SUMMARY.md
   - 向用户报告:
     ✓ 文件列表
     ✓ 输出目录路径
     ✓ 关键发现摘要
   

5. Optional: Interactive dialogue:

   如果用户请求,基于 socratic-questions.md 进行深度对话
   

Detailed Documentation

• Complete Workflow: Refer to references/workflow.md
• Agent Prompts:
  • references/agents/organizer.md - Content Organization Expert
  • references/agents/explainer.md - Feynman Technique Paraphraser
  • references/agents/diagnosis.md - Critical Diagnostic Engine
  • references/agents/socratic.md - Socratic Questioner
  • references/agents/planner.md - Action Planner

Usage Examples

# 完整流程
/deep-reading examples/sample-article.md examples/sample-draft.md

# 仅内化层
/deep-reading --internalize-only article.md draft.md

# 仅拓展层
/deep-reading --expand-only article.md draft.md

# 无行动计划
/deep-reading --no-action article.md draft.md

Output Files

All outputs are saved in outputs/[timestamp]/:

• organized-notes.md - Structured notes
• explained-notes.md - Plain-language explanation
• diagnosis-report.json - Diagnostic report
• socratic-questions.md - Socratic questions
• action-plan.md - Action plan
• SUMMARY.md - Summary report

Dependencies

• Claude Code CLI
• Task tool (used to invoke sub-agents)
• File system access permissions

---

Tip: For first-time use, it is recommended to first check the example files in the examples/ directory to understand the input format. For the complete architecture design, see the project root directory's architecture-design.md。