home / skills / jst-well-dan / skill-box / deep-reading
This skill guides deep reading by transforming articles from read to understood to mastered, delivering structured notes and actionable plans.
npx playbooks add skill jst-well-dan/skill-box --skill deep-readingReview the files below or copy the command above to add this skill to your agents.
---
name: deep-reading
description: Deep reading collaborative system using multi-layered AI agents to help transform articles from "read" to "understood" to "mastered", with actionable plans. Use when users need to deeply understand complex articles/papers, organize reading notes systematically, think critically about content, discover hidden logical issues and assumptions, or transform knowledge into action plans. Trigger keywords - deep reading, 深度阅读, critical thinking, 批判性思维, reading notes, 阅读笔记, article analysis, 文章分析, Socratic questioning, 苏格拉底提问, action plan, 行动计划
---
# Deep Reading - 深度阅读协作系统
深度阅读协作系统,通过多层次的 AI agents 协作,帮助你将文章从"读过"到"读懂"再到"读透",最终转化为可执行的行动计划。
## 使用场景
使用本 skill 当用户需要:
- 深度理解一篇复杂的文章、论文或长文
- 将阅读笔记系统化整理
- 批判性思考文章内容
- 发现隐藏的逻辑问题和假设
- 将知识转化为行动计划
## 核心功能
系统包含三层协作架构:
### 🔍 内化层(读薄) - 顺序执行
- **整理者**: 生成结构化的 Markdown 笔记
- **转述者**: 使用 SCQA + 费曼技巧通俗化解释
### 🎯 拓展层(读厚) - 并行执行
- **诊断引擎**: 批判性分析 + 逻辑剖析 (后台)
- **苏格拉底**: 引导性提问,不给答案 (前台)
### 🚀 产出层(行动)
- **规划师**: 将认知转化为具体的行动计划
## Response Pattern
当用户请求深度阅读分析时:
1. **接收输入**:
```
- 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. **确认配置**:
```
- 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. **执行处理流程**:
```
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. **整合输出**:
```
- 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. **可选: 交互式对话**:
```
如果用户请求,基于 socratic-questions.md 进行深度对话
```
## 详细文档
- **完整工作流程**: 参考 `references/workflow.md`
- **Agent 提示词**:
- `references/agents/organizer.md` - 内容整理专家
- `references/agents/explainer.md` - 费曼技巧转述者
- `references/agents/diagnosis.md` - 批判性诊断引擎
- `references/agents/socratic.md` - 苏格拉底式提问者
- `references/agents/planner.md` - 行动规划师
## 使用示例
```bash
# 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
```
## 输出文件
所有输出保存在 `outputs/[timestamp]/`:
- `organized-notes.md` - 结构化笔记
- `explained-notes.md` - 通俗化解释
- `diagnosis-report.json` - 诊断报告
- `socratic-questions.md` - 苏格拉底式问题
- `action-plan.md` - 行动计划
- `SUMMARY.md` - 汇总报告
## 依赖
- Claude Code CLI
- Task tool (用于调用 sub-agents)
- 文件系统访问权限
---
**提示**: 首次使用建议先查看 `examples/` 目录下的示例文件,了解输入格式。完整架构设计参见项目根目录的 `architecture-design.md`。
This skill is a deep-reading collaborative system that uses multi-layered AI agents to turn articles from "read" to "understood" to "mastered" and produce actionable plans. It combines structured note organization, plain-language explanations, critical diagnosis, Socratic questioning, and planning into a single automated workflow. The goal is to help users extract insight, spot hidden assumptions, and convert knowledge into concrete next steps.
The system runs three coordinated layers: an internalization layer that organizes and explains content, an expansion layer that runs critical diagnostics and generates Socratic questions, and an output layer that creates actionable plans. Agents operate in sequence and in parallel — organizer and explainer produce structured notes, diagnosis and Socratic agents run analysis in parallel, and a planner synthesizes final action items. Outputs are saved to a timestamped directory and summarized for quick review.
What inputs does the skill require?
Provide the article text or file path and optional draft notes. If missing, the system will prompt for content.
Can I run only part of the workflow?
Yes. Use flags to enable only the internalization layer, only the expansion layer, or to skip action planning.