AI got superpowers through Skills and MCPs. What about humans?
In 2025, AI agents gained the ability to manipulate the real world — through Skills, MCP servers, and tool use. Claude can now run code, query databases, control browsers, and execute complex scientific workflows. ChatGPT can browse the web, write and run programs, analyze data. Gemini can see, hear, and interact with the physical world.
AI got its skill tree. But what about humans?
- A 35-year-old professional realizes their degree is becoming obsolete. How do they catch up?
- A 10-year-old will graduate into a world where AI does most knowledge work. What should they learn?
- A PhD student spends 5 years mastering a narrow field. Was that the right investment?
- A first-generation college student has no mentors. Who teaches them the unwritten rules?
These are not hypothetical questions. They are the questions of our time.
Here's what we know from the latest research:
AI tutoring dramatically improves learning, but only with pedagogical guardrails. A large-scale randomized controlled trial published in PNAS found that GPT-4-based tutoring improved high school math performance by 48–127%. Without structured pedagogical design, students became dependent on AI and showed reduced skill acquisition. Carefully designed guardrails (providing hints instead of answers) restored learning gains.
Spaced repetition and active recall are the most effective learning techniques known to science. A meta-analysis of 242 studies with 169,000+ participants confirmed that practice testing and distributed practice outperform all other methods, including re-reading, highlighting, and summarizing.
— Dunlosky et al. (2013), Psychological Science in the Public Interest; Donoghue & Hattie (2021)
Algorithmically optimized spaced repetition dramatically enhances human learning. A study published in PNAS analyzing data from 12 million Duolingo learners demonstrated that computationally optimized review schedules significantly outperform heuristic methods, enabling learners to memorize more material in less time.
The science is clear: AI + cognitive science = the most powerful learning combination in human history. The gap? AI agents have no built-in pedagogical structure. They know everything but teach nothing.
Human Skill Tree fills that gap.
A collection of 30+ ready-to-use AI agent skills that transform ChatGPT, Claude, Gemini, Copilot, DeepSeek, or any compatible AI into a structured, science-backed learning companion.
It's an upgrade to how AI interacts with humans who want to learn.
| ❌ Without Human Skill Tree | ✅ With Human Skill Tree |
|---|---|
|
You: Teach me calculus AI: Calculus is a branch of mathematics that studies continuous change. There are two main branches: differential calculus and integral calculus... (gives you a Wikipedia-style wall of text) |
You: Teach me calculus AI: Before we start, let me understand where you are:
(assesses your level, then builds from YOUR foundation) |
|
You: My code has a bug, help me fix it AI: The issue is on line 12. Change (gives you the answer directly — you learn nothing) |
You: My code has a bug, help me fix it AI: Let's debug this together. Before I look at the answer:
Try adding a (guides your debugging thinking — you learn the skill, not just the fix) |
| Traditional Learning | Raw AI Chat | 🌳 Human Skill Tree |
|---|---|---|
| One-size-fits-all | No structure | Adaptive + structured |
| Passive (read/listen) | Q&A only | Active recall, spaced repetition, Socratic dialogue |
| Single subject | Knows everything, teaches nothing | Curated paths from K-12 to career |
| No social skills training | Generic advice | Culturally-aware scenario simulation |
| Takes years | Instant but forgettable | Fast AND retainable |
The most meta question of our time: how do you learn about AI, using AI?
Layer 3: BUILD with AI → Prompt engineering, fine-tuning, RAG, agents, MCP development
Layer 2: WORK with AI → Using ChatGPT/Claude/Gemini/Copilot/DeepSeek as daily tools
Layer 1: THINK about AI → What AI is, how it works, what it can't do, ethics, society
Layer 1 — AI Fundamentals (everyone needs this):
- What is machine learning? What are LLMs? How do they actually work?
- What can AI do well? What can't it do? Where does it hallucinate?
- Ethics: bias, privacy, job displacement, deepfakes, alignment
- Critical thinking: evaluating AI outputs, detecting errors, verifying claims
Layer 2 — AI as a Daily Tool (professionals & students):
- Prompt engineering: how to get better outputs from any AI
- Using AI for writing, coding, research, analysis, creativity
- AI-assisted learning: using AI as a Socratic tutor (this project!)
- Workflow integration: AI + your existing tools and processes
Layer 3 — Building AI (developers & researchers):
- Machine learning fundamentals: supervised, unsupervised, reinforcement
- Deep learning: neural networks, transformers, attention mechanisms
- LLM application development: APIs, function calling, RAG, fine-tuning
- Agent development: Skills, MCP servers, tool use, multi-agent systems
- AI safety and alignment research
The paradox: The best way to learn AI is to use AI to learn about AI. Human Skill Tree provides the pedagogical structure that makes this self-referential loop actually work.
🌳 Human Skill Tree
│
├── 🧠 Phase 0: Learning How to Learn (Meta-Skill)
│ └── Spaced repetition, active recall, Feynman technique,
│ memory palace, mind mapping, Bloom's taxonomy, flow state
│
├── 📚 Phase 1: K-12 Foundation
│ ├── Mathematics (arithmetic → calculus)
│ ├── Sciences (physics, chemistry, biology, earth science)
│ ├── Languages (50+ languages, classical & modern)
│ ├── Humanities (history, geography, philosophy, civics)
│ └── Exam Systems (Gaokao, SAT/AP, A-Level, IB, CSAT, JEE...)
│
├── 🎓 Phase 2: University
│ ├── University Guide (major selection, course planning, transfer)
│ ├── STEM (CS, AI/ML, engineering, math, physics, chemistry, bio)
│ ├── Humanities & Social Sciences (law, politics, sociology, media)
│ ├── Business & Economics (finance, accounting, marketing, econ)
│ ├── Medical & Health (clinical, TCM, pharmacy, psychology)
│ └── Arts & Design (fine arts, music, film, architecture)
│
├── 🔬 Phase 3: Graduate & Research
│ ├── Research Methodology (qual, quant, mixed, causal inference)
│ ├── Academic Writing (papers, thesis, grants, LaTeX)
│ ├── Literature Review (systematic search, synthesis, gap analysis)
│ └── Data Analysis & Statistics (R, Python, SPSS, Stata, spatial)
│
├── 💼 Phase 4: Career
│ ├── Career Navigator (exploration, planning, transition)
│ ├── Interview Prep (behavioral, technical, case, whiteboard)
│ ├── Civil Service 公务员 (行测, 申论, 面试, 公文写作)
│ ├── Tech Career (system design, algorithms, AI/ML, PM, DevOps)
│ ├── Finance Career (CFA, modeling, valuation, risk)
│ └── Consulting Career (case prep, MECE, slide writing)
│
├── 🤝 Phase 5: Social Intelligence
│ ├── Chinese Social Intelligence 人情世故 (面子, 关系, 饭局, 酒桌)
│ ├── Cross-Cultural Skills (Hofstede, business etiquette)
│ ├── Emotional Intelligence (EQ, empathy, self-regulation)
│ ├── Negotiation & Persuasion (BATNA, Cialdini's 6 principles)
│ └── Communication (assertive, difficult conversations, public speaking)
│
└── 🌱 Phase 6: Self-Development
├── Financial Literacy (budgeting, investing, tax, insurance)
├── Critical Thinking (logic, fallacies, media literacy)
├── Health & Wellness (nutrition, exercise, sleep, mental health)
└── Creativity & Innovation (design thinking, lateral thinking)
800+ subjects across 30+ skills, covering 15 national education systems and 6 international curricula.
Every skill applies cognitive science principles — not just what to teach, but how human brains actually learn and remember:
| Principle | Mechanism | Evidence |
|---|---|---|
| 🔄 Spaced Repetition | Fights the forgetting curve with optimal review intervals | Meta-analysis: 242 studies, 169K+ participants (Donoghue & Hattie, 2021) |
| 🧪 Active Recall | Retrieval practice strengthens memory 10x vs re-reading | Roediger & Butler (2011), Journal of Memory and Language |
| 🎯 Desirable Difficulties | Short-term struggle → long-term retention | Bjork & Bjork (2011) |
| 🔀 Interleaving | Mixing topics builds discrimination ability | Pan et al. (2018), J. Exp. Psych: General |
| 🖼️ Dual Coding | Words + visuals = stronger encoding | Paivio (1991), Psychological Review |
| 🪞 Socratic Method | Questions > answers for deep understanding | Chi et al. (2001), Cognitive Science |
| 🧩 Chunking | Group information into meaningful units | Miller (1956), Psychological Review |
| 📊 Bloom's Taxonomy | 6 cognitive levels: remember → create | Anderson & Krathwohl (2001) |
# Clone the repository
git clone https://github.com/24kchengYe/human-skill-tree.git
# Copy all skills to your agent
cp -r human-skill-tree/skills/* ~/.claude/skills/# Meta-skill: learning methodology
cp -r human-skill-tree/skills/00-learning-how-to-learn ~/.claude/skills/
# K-12 math tutoring
cp -r human-skill-tree/skills/01-k12-mathematics ~/.claude/skills/
# Chinese social intelligence 人情世故
cp -r human-skill-tree/skills/05-social-intelligence ~/.claude/skills/| Tool | Skill Directory | Status |
|---|---|---|
| Claude Code | ~/.claude/skills/ |
✅ Primary |
| Cursor | ~/.cursor/skills/ |
✅ Supported |
| OpenAI Codex CLI | ~/.codex/skills/ |
✅ Supported |
| Gemini CLI | ~/.gemini/skills/ |
✅ Supported |
| GitHub Copilot | Custom config | 🔜 Planned |
| DeepSeek | Custom config | 🔜 Planned |
We build on top of existing tools. Here are the best complementary projects:
| Server | What It Does | Stars |
|---|---|---|
| DeepTutor | AI personalized learning assistant (HKU) | ⭐ 9,000+ |
| Anki MCP | Spaced repetition flashcard integration | ⭐ 154 |
| Canvas LMS MCP | 54 tools for learning management | Moderate |
| Wolfram Alpha MCP | Computational knowledge engine | Multiple |
| MandarinMCP | HSK Chinese vocabulary + spaced repetition | New |
| Skill | Install |
|---|---|
| education-tutor | npx skills add eddiebe147/claude-settings@education-tutor -g -y |
| learn-faster-kit | npx skills install hluaguo/learn-faster-kit |
| academic-research-writer | npx skills add endigo/claude-skills@academic-research-writer -g -y |
See docs/landscape.md for the full ecosystem survey (50+ tools).
| Region | Countries |
|---|---|
| East Asia | 🇨🇳 China (高考), 🇯🇵 Japan (共通テスト), 🇰🇷 Korea (수능), 🇸🇬 Singapore |
| Europe | 🇬🇧 UK (GCSE/A-Level), 🇩🇪 Germany (Abitur), 🇫🇷 France (Bac), 🇳🇱 🇨🇭 🇫🇮 |
| Americas | 🇺🇸 USA (SAT/AP), 🇨🇦 Canada |
| South Asia | 🇮🇳 India (JEE/NEET) |
| Middle East | 🇮🇱 Israel (Bagrut) |
| Oceania | 🇦🇺 Australia (ATAR) |
IB · Cambridge IGCSE/A-Level · Montessori · Waldorf/Steiner · Reggio Emilia · Classical Trivium
- Phase 1: Project structure, 3 flagship skills, ecosystem survey
- Phase 2: K-12 full coverage, university STEM tutor, exam systems
- Phase 3: Career skills, cross-cultural, EQ, MCP integrations
- Phase 4: Multi-language (日本語, 한국어, Español, Français), web visualization
See CONTRIBUTING.md for guidelines. We welcome:
- 🆕 New skills — Pick any uncovered subject
- 🌍 Translations — Help learners worldwide
- 📊 Research — Learning science insights with citations
- 🔗 Integrations — MCP servers, flashcard tools, LMS platforms
In RPGs, a skill tree is a branching structure where you unlock abilities by investing points. Real life works the same way — prerequisites matter, multiple paths exist, specialization is valid, and you can always respec.
No other AI skill project covers 人情世故 — the art of navigating human relationships. Yet this is arguably the most important skill set for success in life. AI can be a safe space to practice these skills through scenario simulation.
Most AI tutoring is "ask a question, get an answer." This creates the illusion of learning — you feel like you understand, but forget within days. Our skills are designed around how memory actually works: test before tell, space it out, make it hard, connect it, apply it.
If Human Skill Tree helps you learn anything, consider:
- ⭐ Star this repo — helps others discover it
- 🍕 Sponsor — fund continued development
- 🔀 Fork & contribute — help build the tree
- Bastani, H. et al. (2025). "Generative AI without guardrails can harm learning." PNAS, 122(26). Link
- Dunlosky, J. et al. (2013). "Improving Students' Learning With Effective Learning Techniques." Psychological Science in the Public Interest, 14(1), 4-58.
- Donoghue, G.M. & Hattie, J.A.C. (2021). "A Meta-Analysis of Ten Learning Techniques." Frontiers in Education.
- Pan, S.C. & Rickard, T.C. (2018). "Transfer of test-enhanced learning." J. Experimental Psychology: General, 147(11), 1641-1664.
- Brown, P.C., Roediger, H.L., & McDaniel, M.A. (2014). Make It Stick: The Science of Successful Learning. Harvard University Press.
- Tabibian, B. et al. (2019). "Enhancing human learning via spaced repetition optimization." PNAS, 116(10), 3988–3993. Link
- Bjork, R.A. & Bjork, E.L. (2011). "Making Things Hard on Yourself, But in a Good Way."
MIT — use it, modify it, share it. Knowledge should be free.
Built by humans, for humans, powered by AI.
"The only skill tree that matters is the one you actually climb."
2025 年，AI 智能体通过 Skill、MCP 和工具调用，获得了操控现实世界的能力。Claude 能运行代码、查询数据库、控制浏览器。ChatGPT 能联网搜索、写代码、分析数据。Gemini 能看、能听、能与物理世界交互。
AI 有了自己的技能树。人类呢？
- 一个 35 岁的职场人发现自己的学位正在过时。怎么赶上？
- 一个 2026 年出生的孩子，成年时 AI 已无处不在。该学什么？
- 一个博士生花 5 年钻研一个窄领域。这笔投资对吗？
- 一个农村来的大学生没有人脉、没有导师。谁来教他那些不成文的规则？
这些不是假设性的问题。这是我们这个时代最核心的问题。
最新研究告诉我们：
AI 辅导可大幅提升学习效果，但前提是有教学结构。 发表在 PNAS 上的大规模随机对照试验发现：GPT-4 辅导使高中数学成绩提升 48–127%，但缺乏教学设计时学生会依赖 AI，技能习得反而下降。精心设计的教学引导（给提示而非答案）才能恢复学习效果。
间隔重复和主动回忆是科学已知最有效的学习方法。 一项涵盖 242 项研究、169,000+ 参与者的 Meta 分析证实：练习测试和分散练习的效果超过所有其他方法。
科学结论很清晰：AI + 认知科学 = 人类历史上最强大的学习组合。 问题在于，AI 智能体没有内置的教学结构。它们什么都知道，但什么都不会教。
人类技能树填补这个缺口。
最 meta 的问题：如何用 AI 学习 AI？
第3层: 用 AI 构建 → 提示词工程、微调、RAG、Agent 开发、MCP 开发
第2层: 用 AI 工作 → 用 ChatGPT/Claude/Gemini/Copilot/DeepSeek 作为日常工具
第1层: 理解 AI → AI 是什么、怎么工作、不能做什么、伦理、社会影响
悖论：学 AI 最好的方式就是用 AI 学 AI。人类技能树提供了让这个自指循环真正有效的教学结构。
🌳 人类技能树
│
├── 🧠 第0层：学会学习（元技能）
│ └── 间隔重复、主动回忆、费曼学习法、记忆宫殿、思维导图
│
├── 📚 第1层：K-12 基础教育
│ ├── 数学 · 自然科学 · 语言（50+语种）· 人文社科
│ └── 考试体系（高考/SAT/AP/A-Level/IB/수능/JEE…）
│
├── 🎓 第2层：本科专业
│ ├── 理工科（CS/AI/工程/数理化生）· 人文社科 · 商科经济
│ ├── 医学健康 · 艺术设计 · 选专业/转专业指南
│ └── 🔥 AI 与机器学习（从入门到 Agent 开发）
│
├── 🔬 第3层：研究生与科研
│ └── 研究方法 · 学术写作 · 文献综述 · 数据分析
│
├── 💼 第4层：职业技能
│ ├── 职业导航 · 面试准备 · 科技/金融/咨询行业
│ └── 🇨🇳 公务员（行测/申论/面试/公文写作）
│
├── 🤝 第5层：社交智慧
│ ├── 🇨🇳 中国人情世故（面子/关系/饭局/酒桌/送礼）
│ ├── 跨文化沟通 · 情商 · 谈判与说服
│ └── 沟通技巧（果敢表达/困难对话/公共演讲）
│
└── 🌱 第6层：自我发展
└── 财商 · 批判性思维 · 健康管理 · 创造力
覆盖 800+ 学科，30+ 技能，15 个国家教育体系，6 种国际课程。
# 克隆
git clone https://github.com/24kchengYe/human-skill-tree.git
# 全部安装
cp -r human-skill-tree/skills/* ~/.claude/skills/
# 或按需安装
cp -r human-skill-tree/skills/00-learning-how-to-learn ~/.claude/skills/ # 学会学习
cp -r human-skill-tree/skills/01-k12-mathematics ~/.claude/skills/ # K-12 数学
cp -r human-skill-tree/skills/05-social-intelligence ~/.claude/skills/ # 人情世故兼容：Claude Code · Cursor · OpenAI Codex · Gemini CLI
- 🆕 添加新技能 — 选一个未覆盖的学科
- 🌍 翻译 — 日本語、한국어、Español、Français…
- 📊 研究 — 分享学习科学论文与洞察
- 🔗 集成 — 对接 MCP、闪卡、LMS
- ⭐ Star 这个仓库 — 让更多人发现它
- 🍕 赞助 — 支持持续开发
- 🔀 Fork & 贡献 — 一起建这棵树
MIT — 自由使用、修改、分享。知识应当自由。
由人类构建，为人类服务，AI 驱动。
"唯一重要的技能树，是你真正去攀爬的那棵。"