该提示词引导AI为指定领域构建「从入门到精通」的三层结构化知识图谱，并采集黄金学习资源、定义通关标准、做三表绑定校验，供NotebookLM、Obsidian、RAG等系统使用。使用时替换领域与时间范围，适合知识管理与AI系统构建。

中文版提示词

你的任务是：针对指定领域，构建一套「从入门到精通」的【结构化、可索引、可评估、可淘汰】知识库索引，供 NotebookLM / Obsidian / RAG / Planning Agent 等AI系统直接使用。禁止输出叙事性内容、宣言式语言和背景长文。

研究领域：{XXXXXXXXXXXXXXX}；时间范围：{XXXXXXXXXXXXXXX}。

全局硬约束：1. 时效性：知识体系主干以时间范围内的方法、工具、实践为准，早于该范围的内容仅作背景或对比，不进主干、不作通关依据，被淘汰的知识点标注「Deprecated」；2. 复杂度过滤：屏蔽无法转化为可执行行为或决策的理论细节与前置学术门槛过高的内容；若知识点不能回答「做什么」「如何做」「什么情况下做或不做」之一则删除；3. 结构优先：先输出结构化列表，再输出最小必要说明，每条说明不超过2行，禁止连续长段落。

Task 1 全景知识图谱（强制三层树形结构）：Layer 1 一级节点（Learning Module，学习阶段或能力模块）；Layer 2 二级节点（Knowledge Category，学完即可独立完成一类任务）；Layer 3 三级节点（Atomic Skill，最小可学习可验证单元，用「动词+对象」描述）。每个三级节点生成唯一Knowledge ID，格式 M{模块编号}-K{分类编号}-A{原子编号}。

Task 2 黄金学习资源：仅收录一手、权威、可长期访问的资源，优先官方文档/课程/权威发布，严禁营销号与二手转述。输出格式：[资源名称] / [类型] / [对应Knowledge ID] / [解决的问题或补齐的能力] / [必读/选读/参考] / [URL]。

Task 3 通关标准：每条绑定Knowledge ID，必须可验证、可复现、可失败。格式：[Knowledge ID] / [能力描述] / [可验证的具体成果或Demo] / [失败判定标准]。

Task 4 三表绑定校验：输出Binding Summary表（Knowledge ID | 知识点名称 | 关联资源数 | 关联通关标准数），资源数或通关标准数小于1的标记×不合格。

输出质量门：自检是否存在超时间范围内容混入主干、不可执行/不可验证知识点、非严格三层结构、Knowledge ID无法反查资源与通关标准；任一项为否须修正后再输出。优先保证结构正确性、可执行性、可维护性，而非内容数量或语言表现。

英文版提示词

Your task: build a "beginner to mastery" knowledge base index for a given field that is structured, indexable, assessable and prunable, for direct use by AI systems such as NotebookLM / Obsidian / RAG / Planning Agent. Do not output narrative content, declarative language or long background text.

Research field: {XXXXXXXXXXXXXXX}; Time range: {XXXXXXXXXXXXXXX}.

Global constraints: 1. Timeliness: the knowledge backbone must follow methods, tools and practices within the time range; earlier content only as background or contrast, never in the backbone or as assessment criteria; obsolete knowledge must be marked "Deprecated"; 2. Complexity filter: remove theoretical details that cannot turn into executable actions or decisions, and content requiring heavy prerequisites; delete any knowledge point that cannot answer "what to do," "how to do it," or "when to do or not do it"; 3. Structure first: output structured lists before minimal necessary notes, each note no more than 2 lines, no long paragraphs.

Task 1 Knowledge Graph (strict three-layer tree): Layer 1 Learning Module (top-level stage or capability module); Layer 2 Knowledge Category (a category whose completion enables finishing one type of task); Layer 3 Atomic Skill (the smallest learnable, verifiable unit, described as "verb + object"). Each Layer 3 node gets a unique Knowledge ID in the format M{module}-K{category}-A{atomic}.

Task 2 Golden Repository: only first-hand, authoritative, long-term accessible resources; prefer official docs, official courses and authoritative publishers; no marketing accounts or second-hand retellings. Format: [resource name] / [type] / [Knowledge ID] / [problem solved or capability built] / [required/optional/reference] / [URL].

Task 3 Success Metrics: each bound to a Knowledge ID and must be verifiable, reproducible and fail-able. Format: [Knowledge ID] / [capability description] / [verifiable result or demo] / [failure criteria].

Task 4 Binding check: output a Binding Summary table (Knowledge ID | knowledge point | resource count | metric count); mark × failed if either count is below 1.

Quality gate: self-check for out-of-range content in the backbone, non-executable/unverifiable knowledge points, non-strict three-layer structure, and IDs that cannot trace back to resources and metrics; fix before output if any fails. Prioritize structural correctness, executability and maintainability over content volume or language polish.

🛠️ **适用 AI 工具**：Claude、ChatGPT、DeepSeek、Gemini、Kimi、通义千问