edisonlzy/deepin-topic

Deep-in Topic

Framework Context

See Framework Structure for the canonical Obsidian knowledge tree structure definition.

Configuration

See Configuration Check Flow before executing this skill.

---

Subskill: Deep-in Topic

Goal

Conduct deep, comprehensive research on a specified topic and produce authoritative knowledge-point Markdown files. Unlike the Divide Into Topic skill (which structures sub-topics), this skill goes deep into a single topic — producing rich, substantive content files ready to be "fruit nodes" in the knowledge tree.

Input

• Target topic name or path
• Vault root path (resolved from configuration obsidian_vault_path)
• Optional: specific knowledge points or angles the user wants to focus on

Execution Steps

Flow Overview

flowchart TD
    A[Start: identify target topic] --> B[Locate topic node via Retrieve Topic Skill]
    B --> C[Read topic README.md to understand scope]
    C --> D[Propose knowledge points to research]
    D --> E{User confirms knowledge points?}
    E -->|No / adjust| D
    E -->|Yes| F[Dispatch parallel research agents per knowledge point]
    F --> G[Synthesize research into draft MD files]
    G --> H[Apply Obsidian Markdown formatting via obsidian-markdown skill]
    H --> I[Update topic README.md with new file index]
    I --> J[✅ Output summary of created files]

---

Step 1: Locate Target Topic

Invoke the Retrieve Topic Node subskill using the user-specified topic name as the query.

• On success: obtain the confirmed topic directory path and read its README.md to understand the current scope, existing sub-topics, and what knowledge files already exist.
• If not found: inform the user and stop. Suggest creating the topic first with the Create Topic Skill.

---

Step 2: Propose Knowledge Points (Human-in-the-Loop)

Based on the topic's README.md and scope, propose a concrete list of knowledge points to research and produce as individual files.

Each knowledge point corresponds to one output .md file.

Present them to the user:

I'll research the following knowledge points for "[Topic]":

1. [Knowledge Point A] — e.g., "How the event loop works in JavaScript"
2. [Knowledge Point B] — e.g., "Comparison of JavaScript module systems (CJS vs ESM)"
3. [Knowledge Point C] — e.g., "Prototype chain and inheritance mechanisms"

Do you want to adjust, add, or remove any knowledge points before I start researching?

Iterate with the user until the list is explicitly confirmed. Do not proceed to Step 3 without confirmation.

---

Step 3: Parallel Deep Research

Once the knowledge point list is confirmed, dispatch multiple sub-agents in parallel — one per knowledge point — each using the web_search tool to conduct deep research.

Each research agent must:

1. Search broadly first: find authoritative sources (official docs, papers, reputable blogs, MDN, RFC specs, etc.)
2. Search deeply: follow up with targeted queries to cover edge cases, common pitfalls, and advanced nuances
3. Synthesize findings into a structured draft covering:
   • Core concepts and definitions
   • How it works (mechanisms, internals)
   • Practical usage and examples
   • Common pitfalls / gotchas
   • Advanced considerations
   • References and further reading

Research quality requirements:

• Comprehensive: cover both fundamentals and advanced aspects
• Authoritative: prefer official documentation, specifications, and well-known expert sources
• Deep: do not stop at surface-level explanations — dig into the "why" and internals

---

Step 4: Generate Knowledge Point Files

⚠️ 约束：如果需要新创建文件，只需要参考模版生成文件的内容结构即可，不要乱加任何未提及的内容。

For each researched knowledge point, produce a .md file in the target topic directory.

File naming: use sequential numbering prefix + kebab-case based on the knowledge point name. Format: {N}.{kebab-case-name}.md，N 为该 topic 目录下已有知识点文件数量 + 1（从 1 开始）。 Example: 1.event-loop.md, 2.module-systems.md, 3.prototype-chain.md

File structure: strictly follow resources/CONTENT-template.md. Key elements to fill in:

| Template 占位符 | 填写内容 | |----------------|---------| | title | 知识点标题 | | date | 当前日期（YYYY-MM-DD） | | tags | 继承父 topic 的 tags，追加知识点相关标签 | | aliases | 知识点的英文名或常见别称 | | cssclass | 所属 topic 名称 | | [!abstract] 目录 | 根据实际章节生成 wikilink 锚点目录 | | 一级 / 二级 / 子小节 | 按研究内容填充，保持模板层级结构 | | [!tip] / [!warning] / [!info] callout | 用于补充提示、常见错误、推荐做法 | | 对比表格 | 有多方案/多类型对比时使用，含 ✅/❌ 标记 | | [!success] 总结 | 用 1–4 条核心要点收尾 | | 相关资料 | 官方文档外链 + vault 内 wikilink |

---

Step 5: Obsidian Markdown Formatting

After all draft files are generated, invoke the /obsidian-markdown skill on each file to:

• 校验并补全 frontmatter（title、date、tags、aliases、cssclass）
• 将普通链接替换为 wikilink（[[...]]），关联 vault 内已有 topic
• 确保 callout 语法正确（> [!type]）
• 优化代码块语言标注、表格对齐、标题层级一致性

知识点文件格式优化参考 Optimize Topic Files 中的「4. Optimize Content Knowledge-point Files」规范。

---

Step 6: Update Topic README.md

After all files are created and formatted, update the target topic's README.md:

1. Read the current README.md
2. Add each new file under the appropriate section. If a ## Knowledge Files (or similar) section exists, append there. If not, create it:

## Knowledge Files

- [[1.event-loop.md]] — How the JavaScript event loop works
- [[2.module-systems.md]] — Comparison of CJS vs ESM module systems
- [[3.prototype-chain.md]] — Prototype chain and inheritance mechanisms

3. Use wikilink format [[filename]] for all entries.

---

Step 7: Output Summary

Output a clear summary of what was created:

✅ Deep research complete for topic: [Topic]

Files created:
  📄 1.event-loop.md
  📄 2.module-systems.md
  📄 3.prototype-chain.md

Location: [topic_path]/

README.md updated with links to all 3 new files.

---

Acceptance Criteria

• [ ] Target topic is located via Retrieve Topic subskill
• [ ] Knowledge points are proposed and confirmed by user before research begins
• [ ] Research is dispatched in parallel across all confirmed knowledge points
• [ ] Each knowledge point produces a well-structured .md file
• [ ] Research content is comprehensive, authoritative, and deep
• [ ] All files are formatted with Obsidian Markdown via /obsidian-markdown skill
• [ ] Topic README.md is updated with wikilinks to all new files
• [ ] Final summary is outputted

Helper Tools

• Web search: web_search tool (use multiple targeted queries per knowledge point)
• Formatting: /obsidian-markdown skill
• Directory listing: ls <topic_path>
• Parallel dispatch: invoke multiple sub-agents simultaneously for Step 3