A universal AI skill that transforms any content into a complete, adaptive learning system.
Works with any AI agent: Claude Code, Codex, Gemini, OpenClaw, or any LLM that accepts system prompts.
Give it a topic, URL, PDF, or pasted text. It will:
- Build a knowledge base — actively collects multiple high-quality sources, filters by authority/recency/originality, and optionally uploads to NotebookLM to minimize hallucination
- Map the intellectual landscape — queries expert mental models, field-level debates, and generates diagnostic questions before teaching begins
- Generate a learning map — modules with weight, difficulty, dependencies, and consensus/debate annotations
- Teach each module using a proven triple structure:
- Concept — what it is, why it matters, real-world scenario anchor
- Anti-pattern — common mistake, why it's tempting, correct approach
- Practice scenario — realistic situation requiring judgment
- Calibrate your level — self-assessment + optional 3-minute quiz to verify actual depth
- Make It Easier on demand — analogies, historical context, simplified versions, prerequisite mini-lessons, first-person intuition, and formal boundaries
- Dissect concepts on demand — definition anchoring, common misconceptions, eight exploration dimensions, and compressed concept cards
- Diagnose errors — 3-step diagnosis on every wrong answer: what assumption was wrong, what key premise was missing, which type of misconception
- Track progress — persistent progress files across sessions, resume from any breakpoint
- Run spaced repetition — Ebbinghaus-curve review schedule (R1/R2/R3/R4) per module
- Award achievements — milestone celebrations after each module, randomized badge on knowledge base completion
| Pattern | What It Does |
|---|---|
| Multi-source knowledge base | 4 source types (authoritative / deep / multi-perspective / applied) with quality grading (A/B/C/D) before admission |
| Expert lens first | Before any teaching: maps expert mental models, field debates, and generates diagnostic questions |
| Adaptive depth | Teaching depth adjusts per module based on calibrated level, not just self-rating |
| Anti-pattern teaching | Each concept is taught, then stress-tested with the most tempting wrong approach |
| Make It Easier mode | 7 strategies (analogy / history / minimal version / prerequisite patch / 5yo / first-person intuition / formal boundary) for breaking through difficulty walls |
| Concept anatomy | Optional deep-dive mode: anchor the definition, expose misconceptions, slice through 8 dimensions, then compress into a reviewable card |
| 3-step error diagnosis | Wrong answers reveal: false assumption → missing premise → misconception type |
| Spaced repetition | 4-round review schedule baked into every module's completion |
| Session persistence | Full progress state saved; resume from exact breakpoint across sessions |
| Achievement system | Per-module milestones + 8 randomized completion badges with learning stats |
| Mode | Triggered by |
|---|---|
| Full deep dive | All modules in sequence, complete assessment loop |
| Quick review | faster command or time budget ≤ 20 min |
| Targeted | Specify module numbers at start |
| Exam sprint | Re-study after completion, focuses weak modules |
| Make It Easier | Any time — triggered by command or repeated wrong answers |
| Concept anatomy | concept <term>, dissect <term>, 概念解剖 <概念>, or repeated concept-confusion errors |
# Personal
mkdir -p ~/.claude/skills/learn-anything
cp SKILL.md ~/.claude/skills/learn-anything/
# Project-level (shared with team)
mkdir -p .claude/skills/learn-anything
cp SKILL.md .claude/skills/learn-anything/Invoke with:
/learn-anything Kubernetes networking
/learn-anything https://docs.example.com/guide
/learn-anything ~/papers/attention-is-all-you-need.pdf
/learn-anything concept entropy
/learn-anything resume
Copy the contents of SKILL.md (everything after the YAML frontmatter ---) into your agent's system prompt or custom instruction field. The skill uses no platform-specific syntax — pure Markdown that any LLM can follow.
NotebookLM is used as a low-hallucination knowledge base backend and for per-module audio generation.
Setup:
pip install notebooklm-mcp-cli
nlm loginWith NotebookLM:
- Sources are uploaded to a persistent notebook
- Expert-lens queries run against verified source content
- Per-module audio generated with custom focus prompts
- One audio per module by default (not one generic overview)
Without NotebookLM: All text features work fully. Audio is disabled. The skill auto-detects and degrades gracefully.
For better source collection during knowledge base construction, install omni-search-skill. The skill will detect and use it automatically.
| Command (EN / 中文) | Action / 动作 |
|---|---|
make it easier / 更简单点 |
Activate Make It Easier mode / 触发简化模式 |
skip / 跳过 |
Skip current concept or module / 跳过当前概念或模块 |
deeper / 展开 |
Go deeper on current topic / 当前概念深入讲解 |
concept <term> / 概念解剖 <概念> |
Dissect a concept and generate a compressed concept card / 解剖概念并生成压缩卡片 |
assess / 测试我 |
Jump to module assessment / 立即进入模块测评 |
pause / 暂停 |
Save progress and end session / 保存进度，结束学习 |
map / 进度 |
Show learning map with progress / 显示整体学习进度 |
review schedule / 复习计划 |
View spaced repetition schedule / 查看复习队列 |
export notes / 导出笔记 |
Export learning notes as Markdown / 导出学习笔记 |
export concepts / 导出概念卡片 |
Export all concept anatomy cards / 导出概念解剖卡片 |
export org / 导出 org |
Export notes or concept cards as org-mode / 导出 org-mode |
audio mode / 切换音频 |
Switch to audio mode (requires NotebookLM) / 切换音频模式 |
connections / 知识地图 |
Show cross-concept relationship map / 显示概念关联图谱 |
reset / 重置模块 |
Reset current module and restart / 重置当前模块 |
Natural language equivalents work too (e.g., "go deeper", "I don't get this", "show my progress" / "讲深一点"、"我听不懂"、"给我看看进度").
Progress is saved in {SKILL_DIR}/progress/:
progress/
index.md # Global index of all knowledge bases
{kb-slug}.md # Per-knowledge-base progress file
{kb-slug}-notes.md # Exported learning notes
{kb-slug}-concepts.md # Exported concept anatomy cards
standalone-concepts.md # Concept anatomy records without a knowledge base
Each progress file tracks: module status, self-rating vs actual score, calibration state, weak concepts, review schedule (R1–R4), study time, and session breakpoint.
MIT
通用 AI 学习系统。将任何内容转化为结构化、自适应的完整学习体验。
适用于任何 AI Agent：Claude Code、Codex、Gemini、OpenClaw，或任何接受系统提示词的 LLM。
输入一个主题、URL、PDF 或粘贴文本，它会：
- 构建知识库 — 主动采集多类高质量来源，按权威性/时效性/原创性分级过滤（A/B/C/D），可选上传到 NotebookLM 降低幻觉率
- 建立领域智识地图 — 在开始教学之前，先查询专家心智模型、领域核心争议，并生成诊断性问题集
- 生成学习地图 — 含权重、难度、前置依赖、共识/争议标注的模块化学习地图
- 用三重结构教学每个核心概念：
- 概念讲解 — 是什么 + 为什么重要 + 真实场景锚定
- 反模式 — 常见错误 + 为什么诱人 + 正确做法
- 场景练习 — 需要你做判断的现实情境
- 精准水平校准 — 用户自评 + 可选 3 分钟快速测验，以实际测验结果驱动教学深度
- Make It Easier 模式 — 类比 / 历史溯源 / 最小化版本 / 前置知识补课 / 5岁版解释 / 第一人称内观 / 形式化边界，随时可触发
- 概念解剖模式 — 定义定锚、常见误解、八维切面、内观和压缩卡片，专门攻克抽象/易混淆概念
- 三步错题诊断 — 每道错题：错误假设是什么 → 漏了什么关键前提 → 误区类型（概念混淆/前置缺口/场景判断）
- 进度持久化 — 完整进度跨会话保存，随时从断点恢复
- 间隔复习 — 基于艾宾浩斯曲线的 R1/R2/R3/R4 复习计划，内置于每个模块
- 成就激励 — 每个模块完成后的里程碑庆祝 + 知识库完成时的随机勋章（含详细学习统计）
| 机制 | 作用 |
|---|---|
| 多源知识库构建 | 4 类来源（权威基础/深度专业/多元视角/实践应用）+ 质量过滤，覆盖全面而非数量堆砌 |
| 专家视角优先 | 教学前先建立：专家心智模型 + 领域争议地图 + 诊断性问题集 |
| 自适应深度 | 教学深度由校准后的实际水平驱动，不完全依赖主观自评 |
| 反模式教学 | 每个概念：讲清楚什么是对的，也讲清楚什么是错的、为什么诱人 |
| Make It Easier | 7 种策略应对认知瓶颈，任何时候可触发 |
| 概念解剖 | 可选深挖模式：定锚定义、暴露误区、八维切开，再压缩成可复习卡片 |
| 三步错题诊断 | 不只给答案，找到思维缺口的根源 |
| 间隔复习 | 每个模块完成后自动计算 4 轮复习时间，防止遗忘 |
| 进度持久化 | 全状态保存，跨会话断点续学 |
| 成就体系 | 里程碑数据 + 8 种随机勋章样式（含学习用时/正确率/自评准确度等） |
# 个人使用
mkdir -p ~/.claude/skills/learn-anything
cp SKILL.md ~/.claude/skills/learn-anything/
# 项目级（团队共享）
mkdir -p .claude/skills/learn-anything
cp SKILL.md .claude/skills/learn-anything/调用方式：
/learn-anything 分布式系统共识算法
/learn-anything https://kubernetes.io/docs/concepts/
/learn-anything ~/papers/attention-is-all-you-need.pdf
/learn-anything 概念解剖 熵
/learn-anything resume
将 SKILL.md 中 YAML frontmatter（---）之后的全部内容复制到你的 agent 的系统提示词中。本 skill 不使用任何平台特有语法，纯 Markdown 指令，任何 LLM 均可遵循。
NotebookLM 作为低幻觉率的知识库后端，并支持按模块生成音频。
配置方法：
pip install notebooklm-mcp-cli
nlm login启用后：
- 源材料上传到持久化 notebook，AI 基于真实内容作答
- 领域智识地图查询在 notebook 上执行，幻觉率更低
- 按模块生成定向音频（每个模块一个，含专属 focus_prompt）
- 默认不生成整合音频，除非用户明确要求
未配置时： 所有文字功能完全正常使用，音频功能自动禁用，无报错。
知识库构建阶段的多源采集，可通过安装 omni-search-skill 增强。Skill 会自动检测并使用。
| 命令（英文 / 中文） | 动作 |
|---|---|
make it easier / 更简单点 |
触发 Make It Easier 简化模式 |
skip / 跳过 |
跳过当前概念或模块 |
deeper / 展开 |
当前概念深入讲解 |
concept <term> / 概念解剖 <概念> |
解剖概念并生成压缩卡片 |
assess / 测试我 |
立即进入当前模块测评 |
pause / 暂停 |
保存进度，结束本次 session |
map / 进度 |
显示整体学习进度 |
review schedule / 复习计划 |
查看复习队列和到期时间 |
export notes / 导出笔记 |
导出学习笔记为 Markdown |
export concepts / 导出概念卡片 |
导出概念解剖卡片 |
export org / 导出 org |
导出 org-mode 笔记或概念卡片 |
audio mode / 切换音频 |
切换到音频模式（需 NotebookLM） |
connections / 知识地图 |
显示已学概念关联图谱 |
reset / 重置模块 |
重置当前模块，从头学习 |
自然语言等效表达均可识别（如「讲深一点」「我听不懂」「给我看看进度」/ "go deeper", "I don't get this", "show my progress"）。
进度保存在 {SKILL_DIR}/progress/：
progress/
index.md # 全局知识库索引
{kb-slug}.md # 每个知识库的进度文件（含模块状态/得分/复习计划/断点）
{kb-slug}-notes.md # 导出的学习笔记
{kb-slug}-concepts.md # 导出的概念解剖卡片
standalone-concepts.md # 无知识库时的概念解剖记录
MIT