deep-domain-learning

name

deep-domain-learning

description

Structured workflow for deeply learning a new technical domain using AI-assisted research, inspired by Tw93's industrialized learning methodology. Use when the user wants to systematically study and master a new technical field, produce a high-quality technical article, or build a structured knowledge repository. Triggers include: learn a new domain, study a technical topic, research workflow, build knowledge base, write a technical deep-dive, industrialized learning, structured study plan, technical article writing, domain research methodology, "how to learn X deeply", "help me master Y", "research Z systematically", knowledge curation, material collection and organization.

Deep Domain Learning Skill

A structured, repeatable workflow for deeply learning a new technical domain and producing high-quality written output. Based on Tw93's "industrialized learning" methodology — treating the learning process like an engineering project with clear phases, tooling, and quality gates.

Core philosophy: Input alone is useless. Only output — what you write, explain, and teach — becomes truly yours.

When to Use

• The user wants to systematically learn and master a new technical domain
• The user wants to produce a high-quality technical article or deep-dive
• The user needs help organizing research materials into a structured knowledge base
• The user is preparing to write a technical explainer for a non-expert audience

Workflow Phases

Execute the following phases sequentially. Each phase has clear inputs, outputs, and quality criteria.

---

Phase 1: Material Collection (收集高质量资料)

Goal: Build a comprehensive, high-quality source library for the target domain.

Source types to collect (prioritized):

1. Seminal papers — key papers from the last 2-3 years in the domain
2. Vendor/org blog posts — official technical blogs from major players in the field
3. Expert articles — posts from recognized domain experts on X, personal blogs, etc.
4. University courses — recent courses from top universities (Stanford, MIT, etc.)
5. Hands-on code repos — canonical "build X from scratch" repositories

Actions:

• Help the user identify and list high-quality sources across all 5 categories
• Use web search and scraping tools to download content
• Convert all materials to Markdown format for uniform processing
• Clean and deduplicate content
• Organize into a structured directory:

research-repo/
├── papers/          # Key papers (converted to md)
├── blogs/           # Vendor and expert blog posts
├── courses/         # Course notes and materials
├── code/            # Reference implementations
├── raw/             # Original unconverted files
└── index.md         # Master index with categorization

Output: A structured repository with all materials indexed and categorized.

---

Phase 2: First-Pass Reading & Filtering (初筛阅读与过滤)

Goal: Build a rough mental model of the domain. Remove ~50% of low-value content.

Actions:

• For content the user understands: read through, mark as keep or discard
• For content the user doesn't understand: use AI to explain key concepts, translate if needed
• For code repos: run locally if possible, otherwise analyze the structure and architecture
• Tag each source with a quality rating (keep / maybe / discard)

AI assistance rules:

• Explain concepts in plain language when asked
• Translate technical content to the user's preferred language when needed
• Summarize long papers into key contributions and takeaways
• NEVER replace the user's own reading — only supplement understanding

Output: A filtered source library (~50% of original), with the user having a rough understanding of the domain landscape.

---

Phase 3: Outline Construction (构建文章大纲)

Goal: Create a detailed article outline that maps each section to source materials.

Key principles:

• Audience-first: Define the target reader's knowledge level before writing the outline
• The outline answers: "What do I want the reader to understand after reading this?"
• Each section in the outline should reference specific source materials
• Use Markdown for structure

Outline format:

# Article Title

## Target Audience
[Description of reader's expected background]

## Outline

### 1. Section Title
- Key points to cover
- Source: [reference to specific materials]
- Goal: [what the reader should understand after this section]

### 2. Section Title
...

Output: A complete outline in Markdown with section-to-source mapping.

---

Phase 4: Content Filling (填充内容 — 苦力活)

Goal: Write the full first draft by filling in each section of the outline.

Actions:

• Work through the outline section by section
• This is a review process — re-read source materials while writing
• Write in the user's own voice and understanding
• Include code examples, diagrams descriptions, and analogies where helpful
• Do not worry about length or verbosity at this stage — completeness is the priority

AI assistance rules:

• Help with specific factual questions or concept clarifications
• Assist with code examples if needed
• Do NOT ghostwrite entire sections — the user's understanding must drive the content

Output: A complete but verbose first draft.

---

Phase 5: AI-Assisted Editing (AI 辅助精简)

Goal: Trim redundancy, improve coherence, and fill knowledge gaps — without changing the author's voice.

Editing passes:

1. Redundancy removal: Remove repetitive content while preserving meaning and the author's tone
2. Coherence check: Identify and fix awkward transitions or logical gaps between sections
3. Gap analysis: Identify missing knowledge points that should be supplemented
4. Supplementary research: For identified gaps, research and help the user add missing content

AI assistance rules:

• Preserve the author's original meaning and writing style
• Suggest deletions and changes, but let the user approve
• When supplementing content, clearly mark what was added vs. original
• This phase often reveals new things the author missed — treat it as a learning opportunity

Output: A polished, coherent draft with gaps filled.

---

Phase 6: Self-Review & Publication (自测与发布)

Goal: Final quality gate — the author reads the article themselves (not AI).

Actions:

• The user reads the entire article from start to finish (at least 2 passes)
• First pass: fix factual errors, unclear explanations, and flow issues
• Second pass: polish language, check formatting, ensure reader experience is smooth
• This is like self-testing code — find and fix bugs before shipping

Critical rule: AI is a tool in this phase, NOT a replacement for the author's judgment. The author must read and evaluate the article with their own brain. Outsourcing this step to AI defeats the purpose of learning.

Output: A publication-ready article.

---

Usage Guidelines

When the user invokes this skill:

1. Ask which phase they're in — they may be starting fresh or picking up mid-process
2. Ask about the target domain — what they want to learn
3. Ask about the target audience — who will read the final output
4. Execute the relevant phase, following the rules and principles above
5. Track progress — maintain awareness of which phase is active and what's been completed

Anti-Patterns to Avoid

• Do NOT let AI replace the user's own thinking and reading
• Do NOT skip the filtering phase — quality over quantity
• Do NOT write the article for the user — assist, explain, organize, but the understanding must be theirs
• Do NOT use "3 minutes to understand X" shortcuts — deep learning requires deep engagement
• Do NOT skip the self-review phase — this is where real internalization happens