Cangjie Skill logo

Cangjie Skill

CommunityPopular
kangarooking
cangjie-skill

Distill a book, long-video transcript, podcast, course, or interview into a coherent set of executable skills. Use when the user asks to "拆书" / "蒸馏一本书" / "把 XX 书做成 skill" / "把这个视频/播客/课程蒸馏成 skill" / "turn a book or video into skills" — i.e. wants the frameworks, principles, and methodologies in long-form content extracted into atomic, reusable Claude skills that an agent can invoke in real-world situations. NOT for simple summarization, book reviews, or role-playing as the author (that is nuwa-skill's job).

Overview

Publisherkangarooking
Repositorycangjie-skill
Skill namecangjie-skill
Stars
10.2K
Forks
1.2K
Bundled files
249
LicenseMIT
Links
  • Markdown instructions

    A SKILL.md file the model loads on demand, so it only costs tokens when a request actually matches.

  • Works with any LLM

    AI skills are plain Markdown, not provider-specific code, so this works with GPT, Claude, Gemini, Grok, or a local model.

  • 249 bundled files

    Scripts, templates, and references the model can read while it works. Files are read-only and never executed.

  • Open source

    Published by kangarooking on GitHub. Read the source before you install it.

Installation

Install the Cangjie Skill AI skill in TypingMind to use it with any LLM, or drop it into another agent that reads SKILL.md.

1

Install in TypingMind

TypingMind installs a skill straight from its GitHub folder — it reads SKILL.md, bundles the resource files, and stores the result locally.

  1. Open the app and go to Plugins → Skills.
  2. Choose "Install from GitHub".
  3. Paste the skill folder URL below and confirm.
  4. Enable the skill in any chat where you want it available.
Plugins → Skills → Add skill → From GitHub URL, then paste the folder URL and press Continue.
2

Install in another agent

Any agent that reads the Agent Skills format can use this skill — copy the folder into that agent's skills directory.

Claude Code — .claude/skills
git clone --depth 1 https://github.com/kangarooking/cangjie-skill.git \
  .claude/skills/cangjie-skill
Restart Claude Code after copying so it picks up the new skill.

Use it in TypingMind

Enable Cangjie Skill in any TypingMind chat and the model takes it from there. Its name and description sit in the system prompt, and the moment a request matches, the model loads the full instructions itself — you never invoke it by hand, and it costs no tokens until it is actually used.

The model loads Cangjie Skill on its own as soon as a request matches it.

Works with any AI model

AI skills are plain Markdown instructions rather than provider-specific code, so Cangjie Skill is not tied to the model it was written for. Install it once in TypingMind and use it with GPT-5, Claude, Gemini, Grok, DeepSeek, Mistral, Llama, or a local model you run yourself — all on your own API keys.

  • Loaded only when it is needed

    The system prompt carries just the name and description. The instructions are fetched on the first matching request, so an idle skill costs nothing.

  • Switch models mid-chat

    Because the skill is instructions rather than code, changing model does not break it — the next model reads the same SKILL.md.

Skill instructions

This is the SKILL.md content the model loads. Read it before installing — a skill is instructions your model will follow.

cangjie-skill — 把一本书蒸馏成一组可执行 skills 的元 skill

使命

把一本书里沉淀的方法论,拆解成原子化、可被 agent 在真实场景下调用的能力,并按用户目的编译成合适数量的 skill,让读者真正用起来。

术语约定: 本文档及 methodology/extractors/ 中所有的"书",泛指一切被蒸馏的长内容 — 书籍、长视频转写、播客文字稿、课程、访谈、长文、资料集。

边界:

  • ✅ 做: 方法论 / 决策框架 / 操作流程 / 计算规则 / 排障 / 清单 / 原则的蒸馏,概念体系作为参考支撑
  • ❌ 不做: 书摘 / 读后感 / 作者人设角色扮演 (后者请用 nuwa-skill)

核心方法论: RIA-TV++(v2.5 Bundle 版)

一个五阶段 + 并行提取 + 三重验证 + 晋级门 + darwin 兼容测试的流水线。详见 methodology/00-overview.md

阶段 0:   Adler 整书理解        → BOOK_OVERVIEW.md
阶段 1:   5 个 agent 并行提取    → 候选方法论单元池
阶段 1.5: 三重验证筛选(知识验证) → 通过的单元 (用户轻确认)
阶段 1.6: 独立 Skill 晋级门(产品化验证)→ promoted / router 去向
阶段 2:   RIA++ 构造能力卡       → .cangjie/capabilities/cards/<slug>.md
阶段 3:   Zettelkasten 链接      → verified.yaml 的 also_read + GLOSSARY
阶段 4:   压力测试 (darwin 兼容)  → 评测用例 + 回炉淘汰
阶段 5:   编译与交付             → cangjie.py compile(single/pack)+ DIGEST.md + 安装

v2.5 关键变化(ADR-002): 阶段 2–4 不再直接把每个单元写成独立的最终 Skill 目录,而是产出一份 Capability Bundlebooks/<slug>/.cangjie/capabilities/verified.yaml + cards/*.md)。single 与 pack 都从这同一份 Bundle 由 scripts/cangjie.py compile 确定性编译,两者引用同一组稳定 capability_id。旧的 one-to-one 输出仍受支持(legacy-pack,见 docs/migrations/2026-08-25-v2.0-to-v2.1.md)。

何时调用此 skill

用户说类似:

  • "帮我拆《穷查理宝典》"
  • "把毛选蒸馏成 skill"
  • "把这个 B 站视频/播客/课程蒸馏成 skill"
  • "distill this book into skills: "
  • "我想把这本书的方法论做成可用的 skill"

输入要求

在开始前必须从用户处确认:

  1. 内容文本来源: PDF / EPUB / TXT / 字幕文件 / 转写稿路径, 或可访问的纯文本。不要在没有文本的情况下"凭记忆"蒸馏 — 宁可停下来问用户要。(视频/播客建议先用 video-downloader 类工具拿到转写文本)
  2. 内容元信息: 书籍是"书名 + 作者 + 出版年"; 视频/播客/课程是"标题 + 作者(UP 主/主播/讲者) + 发布时间"。用于目录命名和审计。
  3. 使用目的(决定输出模式推荐): 学习/查阅这本书 → 倾向 single;接入日常工作流、跨书组合 → 倾向 pack。不确定时按 single-first 原则先推荐 single。
  4. 是否首次试点: 如果用户是第一次用 cangjie-skill,建议先蒸馏 1 份内容验证流程再批量。

非书籍内容的字段映射: 章节类字段对视频填时间戳或分 P,对播客填集数,对课程填讲次 — 保证可追溯即可。

输出结构

books/<book-slug>/
├── PIPELINE_STATE.md          # 流水线状态: 当前阶段 + 进度 (断点续跑用)
├── BOOK_OVERVIEW.md           # 阶段 0 产出: 主旨/骨架/术语/批判
├── verified.md                # 阶段 1.5 产出: 通过三重验证的单元 + 判定理由
├── coverage-audit.md          # 原书关键任务 → 候选 → 判定 → 实际交付去向
├── references.md              # 有依据但不执行的参考内容及交付映射
├── needs-review.md            # 来源/条件/测试缺口;不编译为 active 能力
├── GLOSSARY.md                # 阶段 3 产出: 全书共享术语词典
├── DIGEST.md                  # 阶段 5 产出: 面向读者的精华长文
├── candidates/                # 阶段 1 产出: 原始候选池 (审计用)
├── rejected/                  # 阶段 1.5 淘汰的单元 + 原因 (审计用)
└── .cangjie/                  # v2.5 侧车层(编译事实源 + 运行记录)
    ├── capabilities/
    │   ├── verified.yaml      # Capability Bundle(唯一编译事实源, capability-bundle.schema.json)
    │   ├── cards/<slug>.md    # RIA 能力卡(R/I/A1/A2/E/B, 阶段 2 产出)
    │   ├── resources/         # 能力显式声明的 UTF-8 脚本、CSV/JSON/Markdown 模板(可选)
    │   ├── destinations.json  # 晋级/路由去向映射(阶段 1.6 产出)
    │   └── book/{overview.md,glossary.md}
    ├── runs/<run-id>/         # 每次编译/更新的决策报告与校验日志
    └── snapshots/             # 发布前快照(rollback 用)

最终交付物由 scripts/cangjie.py compile 从 Bundle 编译(--output single 得到 1 个入口 + 能力卡;--output pack 得到 1 个来源路由入口 + 少量晋级 Skill)。

执行流程 (严格按顺序)

断点续跑: 开始前先检查 books/<slug>/PIPELINE_STATE.md 是否存在。存在则读取并从记录的阶段续跑,不要从头重来。每完成一个阶段,更新该文件。

阶段 0 — 整书理解

  1. 读取用户提供的书本文本。大文件分块阅读。
  2. 执行 methodology/01-stage0-adler.md 中的 Adler 四步 (结构 / 解释 / 批判 / 应用)。
  3. templates/BOOK_OVERVIEW.md.template 填充,写入 books/<slug>/BOOK_OVERVIEW.md
  4. 把产出展示给用户确认:"骨架我理解对了吗?有没有你希望重点突出的方向?" 得到确认再进入阶段 1。

阶段 1 — 5 个 sub-agent 并行提取

并行 spawn 5 个 Task sub-agents(使用 Agent 工具,一次调用中发起 5 个):

sub-agent读取的 prompt上下文策略产出
框架提取器extractors/framework-extractor.md全量扫描(兼顾跨章节与单处完整机制)决策框架 / 思维模型 / 流程 / 排障
原则提取器extractors/principle-extractor.md全量扫描(含公式、表格、脚注)原则 / 清单 / 规则 / 计算口径
案例提取器extractors/case-extractor.md检索式取块(局部命中型)书中实例、转述案例、例题,标明类型
反例提取器extractors/counter-example-extractor.md检索式取块(局部命中型)书中警告的失败模式
术语提取器extractors/glossary-extractor.md检索式取块 + 脚本预筛关键概念词典

每个 sub-agent 独立判断、独立输出到 books/<slug>/candidates/<type>.md

  • 长文本: 超出单个 sub-agent 上下文的内容,按 methodology/02-stage1-parallel-extract.md 的分块策略处理;已建立内容索引(.cangjie/index/)时,检索式 extractor 按索引取相关块。
  • 覆盖率硬门: 同时对照全量扫描候选和阶段 0 的原书关键任务清单,不只检查旧版通过项。重要任务须有候选及明确去向;有未解释遗漏时补读/退回全量扫描,不虚报零遗漏。
  • 降级方案: 当前环境不支持并行 sub-agent 时,用同样 5 个 extractor prompt 串行执行,产出格式不变。

阶段 1.5 — 三重验证筛选(知识验证)

读取 methodology/03-stage1.5-triple-verify.md,对每个候选单元执行:

  • V1 来源充分性: 可定位原文足以支持限定范围内的做法吗?完整机制在一处讲透也可通过。
  • V2 可执行性: 用合法的新输入能完成任务并核验结果吗?不强求跨领域外推。
  • V3 任务增益: 能减少遗漏、统一口径或稳定交付吗?不要求作者独创。

verified / reference / needs_review / rejected 分流,记录依据及缺口;参考内容须映射到实际交付的 overview/glossary,不是所有未通过项都淘汰。更新 coverage-audit.md

用户轻确认 ★: 展示可执行候选、参考、待核查、淘汰四类及重要缺口,再进入阶段 1.6;确认不能替代来源验证。

阶段 1.6 — 独立 Skill 晋级门(产品化验证)

读取 methodology/03b-stage1.6-promotion-gate.md。知识验证通过 ≠ 值得成为独立 Skill。对每个单元评审五条判据(独立意图/独立契约/独立运行/独立复用/独立评测,前 3 条必须通过,后 2 条至少 1 条),把去向写入 Bundle 的 promotion.destination(promoted / router)。未晋级单元不淘汰,保留为来源路由入口内的能力卡。可发现入口软预算默认 8(含 1 个来源路由入口)。

阶段 2 — RIA++ 构造能力卡

对每个通过的单元,按 methodology/04-stage2-ria-plus.md 构造 R / I / A1 / A2 / E / B 六段能力卡:

  • 卡片正文写入 books/<slug>/.cangjie/capabilities/cards/<slug>.md不带 frontmatter,frontmatter 数据记入 Bundle);
  • 同时在 verified.yaml 登记该能力: 稳定 capability_id、intents、keywords、one_liner、importance(附依据)、frontmatter.description(A2 浓缩版)。
  • A1 区分原书案例与明确标记的合成演练;E 明确输入、输出、分支和完成标准。必要时通过 resources 显式声明配套文本资源,见阶段 2 说明。

阶段 3 — Zettelkasten 链接

methodology/05-stage3-zettelkasten.md:

  1. 找出能力之间的引用关系,写入 Bundle 中各能力的 also_read,并回填 A2 的"与相邻能力的区分"
  2. candidates/glossary.md 整理成 books/<slug>/GLOSSARY.md,并复制到 .cangjie/capabilities/book/glossary.md
  3. INDEX/路由表由编译器从 Bundle 生成,不再手写
  4. references.md 中需保留的内容并入 Bundle 的 book/overview.mdbook/glossary.md,在覆盖审计登记实际路径;不把待核查内容伪装为可执行方法

阶段 4 — 压力测试 (darwin 兼容)

methodology/06-stage4-pressure-test.md,晋级能力测触发,router 能力测可达;两类都要实际完成代表任务并核对输出,不能用“会调用”替代“做得对”。缺失输出计入评测分母,未完成不得声明通过。未过的回炉重做阶段 2。

阶段 5 — 编译与交付

methodology/07-stage5-deliver.md:

  1. 生成 books/<slug>/DIGEST.md — 面向读者的精华长文
  2. 运行 python3 scripts/cangjie.py compile --bundle books/<slug>/.cangjie/capabilities --out <目标目录> --output auto,把决策报告展示给用户轻确认(按推荐 / 改 single / 改 pack)
  3. 询问用户安装位置,把编译产物复制或 symlink 过去
  4. 告知用户: "已完成,可一键喂给 darwin-skill 自动进化"

质量红线 (违反则阻止输出)

  1. 每个能力必须通过全部三重验证
  2. 每张能力卡必须有完整的 R / I / A1 / A2 / E / B 六段;A1 无原书案例时如实标注,合成演练不冒充书中事实
  3. 原文引用 ≤150 字/段 (英文 ≤100 词/段)
  4. 每个 active 能力在 destinations.json 中恰好一个去向(promoted_to 或 served_by);未晋级能力必须可经来源路由入口到达
  5. 晋级 Skill 的 description 必须明确 trigger 条件,并与来源路由入口互有近邻负例
  6. 编译产物必须通过 scripts/validate_skill_pack.py(格式/相对引用/frontmatter 100%)
  7. 生成目录只读:检测到本地手改时不得静默覆盖(编译器强制三选一)
  8. 关键任务覆盖有可追溯去向;实际输出评测与来源核查分别记录。存在重要缺口只能交付明确标注范围的草稿,不能宣称完整通过

与 nuwa-skill / darwin-skill 的生态定位

  • nuwa-skill: 蒸馏人 (思维方式 / 表达 DNA)
  • cangjie-skill (本 skill): 蒸馏书 (方法论 / 框架 / 原则)
  • darwin-skill: 进化任意 skill

三者咬合: 本 skill 输出的评测用例遵循 darwin-skill 格式,以便产出的 skill 可直接接入 darwin 做自动进化。

调用惯例

  • 永远先试点 1 本 — 除非用户明确说"批量"
  • 阶段之间主动汇报进度 — 不要静默跑完再 dump 结果
  • 不凭记忆拆书 — 没文本就停下来问
  • 保留审计轨迹 — candidates/ 和 rejected/ 都要留
  • 随时可续跑 — 每完成一个阶段就更新 PIPELINE_STATE.md,中断后从状态文件恢复
  • 输出策略持久化 — update/repair 默认沿用原输出模式,不因新增材料静默改变产物形态

Bundled files

The model reads these on demand while the skill is loaded. They are exposed as readable files and are never executed.

and 140 more files.

Frequently asked questions

What does the Cangjie Skill AI skill do?

Distill a book, long-video transcript, podcast, course, or interview into a coherent set of executable skills. Use when the user asks to "拆书" / "蒸馏一本书" / "把 XX 书做成 skill" / "把这个视频/播客/课程蒸馏成 skill" / "turn a book or video into skills" — i.e. wants the frameworks, principles, and methodologies in long-form content extracted into atomic, reusable Claude skills that an agent can invoke in real-world situations. NOT for simple summarization, book reviews, or role-playing as the author (that is nuwa-skill's job).

Why use Cangjie Skill on TypingMind?

Because you install it once and use it with any model. Cangjie Skill is plain Markdown rather than provider-specific code, so the same skill runs on GPT-5, Claude, Gemini, Grok, or a local model — and you can switch model mid-chat without it breaking. TypingMind runs on your own API keys, so you pay providers directly instead of a per-seat subscription, and your skills and chats stay in your own storage.

How do I install Cangjie Skill in TypingMind?

Open Plugins → Skills → Install from GitHub in TypingMind and paste https://github.com/kangarooking/cangjie-skill/tree/main. TypingMind reads its SKILL.md and bundles its files and installs it as a skill you can enable per chat.

Which AI models can use Cangjie Skill?

Any model you connect in TypingMind. AI skills are plain Markdown instructions rather than provider-specific code, so GPT, Claude, Gemini, Grok, and local models can all load this skill when a request matches it.

How many AI models can I use with Cangjie Skill?

As many as you like. As long as a model supports skills, you can use Cangjie Skill with it — GPT, Claude, Gemini, Grok, DeepSeek, Mistral, Llama and more — all on TypingMind with your own API keys.

Is the Cangjie Skill AI skill free?

Yes. It is published on GitHub by kangarooking under the MIT license. You only pay your own AI provider for the tokens you use.

What are AI skills?

An AI skill is a reusable instruction bundle that teaches an AI model how to do one specific task. It follows the open Agent Skills format: a SKILL.md file with a name and description, plus any scripts, templates or reference files the model may need. The model reads the instructions only when your request matches the skill, so an installed skill costs nothing until it is used.

How are AI skills different from plugins or MCP servers?

A plugin or MCP server gives a model new tools to call — code that runs somewhere and returns a result. An AI skill gives the model knowledge and process instead: how to approach a task, which steps to follow, what good output looks like. Skills are plain Markdown, so they need no server, no API key and no runtime, and they work with any model.

View all

Set up your own AI workspace now

Get notified about new features and future giveaways by subscribing to our newsletter 👇